maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/lib/ExecutionEngine/JITLink/MachO_x86_64.cpp

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

720 lines
25 KiB
C++
Raw Normal View History

[JITLink] Remove a lot of reduntant 'JITLink_' prefixes. NFC. llvm-svn: 358869
2019-04-22 11:03:09 +08:00
//===---- MachO_x86_64.cpp -JIT linker implementation for MachO/x86-64 ----===//
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
[JITLink] Remove a lot of reduntant 'JITLink_' prefixes. NFC. llvm-svn: 358869
2019-04-22 11:03:09 +08:00
// MachO/x86-64 jit-link implementation.
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
//
//===----------------------------------------------------------------------===//
[JITLink] Remove a lot of reduntant 'JITLink_' prefixes. NFC. llvm-svn: 358869
2019-04-22 11:03:09 +08:00
#include "llvm/ExecutionEngine/JITLink/MachO_x86_64.h"
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
[JITLink] Factor basic common GOT and stub creation code into its own class. llvm-svn: 358838
2019-04-21 11:14:42 +08:00
#include "BasicGOTAndStubsBuilder.h"
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
#include "MachOLinkGraphBuilder.h"
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
#define DEBUG_TYPE "jitlink"
using namespace llvm;
using namespace llvm::jitlink;
using namespace llvm::jitlink::MachO_x86_64_Edges;
namespace {
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
class MachOLinkGraphBuilder_x86_64 : public MachOLinkGraphBuilder {
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
public:
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
MachOLinkGraphBuilder_x86_64(const object::MachOObjectFile &Obj)
[JITLink] Refactor EH-frame handling to support eh-frames with existing relocs. Some targets (E.g. MachO/arm64) use relocations to fix some CFI record fields in the eh-frame section. When relocations are used the initial (pre-relocation) content of the eh-frame section can no longer be interpreted by following the eh-frame specification. This causes errors in the existing eh-frame parser. This patch moves eh-frame handling into two LinkGraph passes that are run after relocations have been parsed (but before they are applied). The first] pass breaks up blocks in the eh-frame section into per-CFI-record blocks, and the second parses blocks of (potentially multiple) CFI records and adds the appropriate edges to any CFI fields that do not have existing relocations. These passes can be run independently of one another. By handling eh-frame splitting/fixing with LinkGraph passes we can both re-use existing relocations for CFI record fields and avoid applying eh-frame fixups before parsing the section (which would complicate the linker and require extra temporary allocations of working memory).
2019-11-05 08:00:09 +08:00
: MachOLinkGraphBuilder(Obj) {}
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
private:
static Expected<MachOX86RelocationKind>
getRelocationKind(const MachO::relocation_info &RI) {
switch (RI.r_type) {
case MachO::X86_64_RELOC_UNSIGNED:
[JITLink] Add support for MachO/x86-64 UNSIGNED relocs with length=2. MachO/x86-64 UNSIGNED relocs are almost always 64-bit (length=3), but UNSIGNED relocs of length=2 are allowed if the target resides in the low 32-bits. This patch adds support for such relocations in JITLink (previously they would have triggered an unsupported relocation error). llvm-svn: 367764
2019-08-04 04:17:10 +08:00
if (!RI.r_pcrel) {
if (RI.r_length == 3)
return RI.r_extern ? Pointer64 : Pointer64Anon;
else if (RI.r_extern && RI.r_length == 2)
return Pointer32;
}
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
break;
case MachO::X86_64_RELOC_SIGNED:
if (RI.r_pcrel && RI.r_length == 2)
return RI.r_extern ? PCRel32 : PCRel32Anon;
break;
case MachO::X86_64_RELOC_BRANCH:
if (RI.r_pcrel && RI.r_extern && RI.r_length == 2)
return Branch32;
break;
case MachO::X86_64_RELOC_GOT_LOAD:
if (RI.r_pcrel && RI.r_extern && RI.r_length == 2)
return PCRel32GOTLoad;
break;
case MachO::X86_64_RELOC_GOT:
if (RI.r_pcrel && RI.r_extern && RI.r_length == 2)
return PCRel32GOT;
break;
case MachO::X86_64_RELOC_SUBTRACTOR:
// SUBTRACTOR must be non-pc-rel, extern, with length 2 or 3.
// Initially represent SUBTRACTOR relocations with 'Delta<W>'. They may
// be turned into NegDelta<W> by parsePairRelocation.
if (!RI.r_pcrel && RI.r_extern) {
if (RI.r_length == 2)
return Delta32;
else if (RI.r_length == 3)
return Delta64;
}
break;
case MachO::X86_64_RELOC_SIGNED_1:
if (RI.r_pcrel && RI.r_length == 2)
return RI.r_extern ? PCRel32Minus1 : PCRel32Minus1Anon;
break;
case MachO::X86_64_RELOC_SIGNED_2:
if (RI.r_pcrel && RI.r_length == 2)
return RI.r_extern ? PCRel32Minus2 : PCRel32Minus2Anon;
break;
case MachO::X86_64_RELOC_SIGNED_4:
if (RI.r_pcrel && RI.r_length == 2)
return RI.r_extern ? PCRel32Minus4 : PCRel32Minus4Anon;
break;
case MachO::X86_64_RELOC_TLV:
if (RI.r_pcrel && RI.r_extern && RI.r_length == 2)
return PCRel32TLV;
break;
}
[JITLink] Add more detail to MachO/x86-64 "unsupported relocation" errors. The extra information here will be helpful in diagnosing errors, like the ones currently occuring on the PPC big-endian bots. :) llvm-svn: 358826
2019-04-21 02:50:13 +08:00
return make_error<JITLinkError>(
[JITLink] Add yet more detail to MachO/x86-64 unsupported relocation errors. Knowing the address/symbolnum field values makes it easier to identify the unsupported relocation, and provides enough information for the full bit pattern of the relocation to be reconstructed. llvm-svn: 358833
2019-04-21 06:59:43 +08:00
"Unsupported x86-64 relocation: address=" +
formatv("{0:x8}", RI.r_address) +
", symbolnum=" + formatv("{0:x6}", RI.r_symbolnum) +
", kind=" + formatv("{0:x1}", RI.r_type) +
[JITLink] Add more detail to MachO/x86-64 "unsupported relocation" errors. The extra information here will be helpful in diagnosing errors, like the ones currently occuring on the PPC big-endian bots. :) llvm-svn: 358826
2019-04-21 02:50:13 +08:00
", pc_rel=" + (RI.r_pcrel ? "true" : "false") +
[JITLink] Fix error message formatting. llvm-svn: 367763
2019-08-04 04:17:08 +08:00
", extern=" + (RI.r_extern ? "true" : "false") +
[JITLink] Fix a bad formatv format string. llvm-svn: 358831
2019-04-21 06:06:12 +08:00
", length=" + formatv("{0:d}", RI.r_length));
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
}
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
using PairRelocInfo = std::tuple<MachOX86RelocationKind, Symbol *, uint64_t>;
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
// Parses paired SUBTRACTOR/UNSIGNED relocations and, on success,
// returns the edge kind and addend to be used.
Expected<PairRelocInfo>
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
parsePairRelocation(Block &BlockToFix, Edge::Kind SubtractorKind,
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
const MachO::relocation_info &SubRI,
JITTargetAddress FixupAddress, const char *FixupContent,
object::relocation_iterator &UnsignedRelItr,
object::relocation_iterator &RelEnd) {
using namespace support;
assert(((SubtractorKind == Delta32 && SubRI.r_length == 2) ||
(SubtractorKind == Delta64 && SubRI.r_length == 3)) &&
"Subtractor kind should match length");
assert(SubRI.r_extern && "SUBTRACTOR reloc symbol should be extern");
assert(!SubRI.r_pcrel && "SUBTRACTOR reloc should not be PCRel");
if (UnsignedRelItr == RelEnd)
return make_error<JITLinkError>("x86_64 SUBTRACTOR without paired "
"UNSIGNED relocation");
auto UnsignedRI = getRelocationInfo(UnsignedRelItr);
if (SubRI.r_address != UnsignedRI.r_address)
return make_error<JITLinkError>("x86_64 SUBTRACTOR and paired UNSIGNED "
"point to different addresses");
if (SubRI.r_length != UnsignedRI.r_length)
return make_error<JITLinkError>("length of x86_64 SUBTRACTOR and paired "
"UNSIGNED reloc must match");
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
Symbol *FromSymbol;
if (auto FromSymbolOrErr = findSymbolByIndex(SubRI.r_symbolnum))
FromSymbol = FromSymbolOrErr->GraphSymbol;
else
return FromSymbolOrErr.takeError();
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
// Read the current fixup value.
uint64_t FixupValue = 0;
if (SubRI.r_length == 3)
[JITLink] Fixed a signedness bug when processing X86_64_RELOC_SUBTRACTOR. Subtractor relocation addends are signed, so we need to read them via signed int pointers. Accidentally treating 32-bit addends as unsigned leads to out-of-range errors when we try to add very large (>INT32_MAX) bogus addends. llvm-svn: 360392
2019-05-10 07:17:41 +08:00
FixupValue = *(const little64_t *)FixupContent;
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
else
[JITLink] Fixed a signedness bug when processing X86_64_RELOC_SUBTRACTOR. Subtractor relocation addends are signed, so we need to read them via signed int pointers. Accidentally treating 32-bit addends as unsigned leads to out-of-range errors when we try to add very large (>INT32_MAX) bogus addends. llvm-svn: 360392
2019-05-10 07:17:41 +08:00
FixupValue = *(const little32_t *)FixupContent;
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
// Find 'ToSymbol' using symbol number or address, depending on whether the
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
// paired UNSIGNED relocation is extern.
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
Symbol *ToSymbol = nullptr;
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
if (UnsignedRI.r_extern) {
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
// Find target symbol by symbol index.
if (auto ToSymbolOrErr = findSymbolByIndex(UnsignedRI.r_symbolnum))
ToSymbol = ToSymbolOrErr->GraphSymbol;
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
else
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
return ToSymbolOrErr.takeError();
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
} else {
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
if (auto ToSymbolOrErr = findSymbolByAddress(FixupValue))
ToSymbol = &*ToSymbolOrErr;
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
else
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
return ToSymbolOrErr.takeError();
FixupValue -= ToSymbol->getAddress();
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
}
MachOX86RelocationKind DeltaKind;
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
Symbol *TargetSymbol;
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
uint64_t Addend;
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
if (&BlockToFix == &FromSymbol->getAddressable()) {
TargetSymbol = ToSymbol;
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
DeltaKind = (SubRI.r_length == 3) ? Delta64 : Delta32;
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
Addend = FixupValue + (FixupAddress - FromSymbol->getAddress());
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
// FIXME: handle extern 'from'.
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
} else if (&BlockToFix == &ToSymbol->getAddressable()) {
TargetSymbol = FromSymbol;
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
DeltaKind = (SubRI.r_length == 3) ? NegDelta64 : NegDelta32;
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
Addend = FixupValue - (FixupAddress - ToSymbol->getAddress());
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
} else {
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
// BlockToFix was neither FromSymbol nor ToSymbol.
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
return make_error<JITLinkError>("SUBTRACTOR relocation must fix up "
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
"either 'A' or 'B' (or a symbol in one "
"of their alt-entry chains)");
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
}
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
return PairRelocInfo(DeltaKind, TargetSymbol, Addend);
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
}
Error addRelocations() override {
using namespace support;
auto &Obj = getObject();
for (auto &S : Obj.sections()) {
JITTargetAddress SectionAddress = S.getAddress();
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
if (S.isVirtual()) {
if (S.relocation_begin() != S.relocation_end())
return make_error<JITLinkError>("Virtual section contains "
"relocations");
continue;
}
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
for (auto RelItr = S.relocation_begin(), RelEnd = S.relocation_end();
RelItr != RelEnd; ++RelItr) {
MachO::relocation_info RI = getRelocationInfo(RelItr);
// Sanity check the relocation kind.
auto Kind = getRelocationKind(RI);
if (!Kind)
return Kind.takeError();
// Find the address of the value to fix up.
JITTargetAddress FixupAddress = SectionAddress + (uint32_t)RI.r_address;
LLVM_DEBUG({
dbgs() << "Processing relocation at "
<< format("0x%016" PRIx64, FixupAddress) << "\n";
});
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
// Find the block that the fixup points to.
Block *BlockToFix = nullptr;
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
{
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
auto SymbolToFixOrErr = findSymbolByAddress(FixupAddress);
if (!SymbolToFixOrErr)
return SymbolToFixOrErr.takeError();
BlockToFix = &SymbolToFixOrErr->getBlock();
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
}
Fix MSVC "32-bit shift implicitly converted to 64 bits" warning. NFCI. llvm-svn: 358970
2019-04-23 19:16:16 +08:00
if (FixupAddress + static_cast<JITTargetAddress>(1ULL << RI.r_length) >
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
BlockToFix->getAddress() + BlockToFix->getContent().size())
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
return make_error<JITLinkError>(
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
"Relocation extends past end of fixup block");
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
// Get a pointer to the fixup content.
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
const char *FixupContent = BlockToFix->getContent().data() +
(FixupAddress - BlockToFix->getAddress());
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
// The target symbol and addend will be populated by the switch below.
Symbol *TargetSymbol = nullptr;
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
uint64_t Addend = 0;
switch (*Kind) {
case Branch32:
case PCRel32:
case PCRel32GOTLoad:
case PCRel32GOT:
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
if (auto TargetSymbolOrErr = findSymbolByIndex(RI.r_symbolnum))
TargetSymbol = TargetSymbolOrErr->GraphSymbol;
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
else
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
return TargetSymbolOrErr.takeError();
[JITLink] Make sure MachO/x86-64 handles 32-bit signed addends correctly. These need to be sign extended when loading into Edge addends.
2019-11-28 14:25:59 +08:00
Addend = *(const little32_t *)FixupContent;
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
break;
[JITLink] Add support for MachO/x86-64 UNSIGNED relocs with length=2. MachO/x86-64 UNSIGNED relocs are almost always 64-bit (length=3), but UNSIGNED relocs of length=2 are allowed if the target resides in the low 32-bits. This patch adds support for such relocations in JITLink (previously they would have triggered an unsupported relocation error). llvm-svn: 367764
2019-08-04 04:17:10 +08:00
case Pointer32:
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
if (auto TargetSymbolOrErr = findSymbolByIndex(RI.r_symbolnum))
TargetSymbol = TargetSymbolOrErr->GraphSymbol;
[JITLink] Add support for MachO/x86-64 UNSIGNED relocs with length=2. MachO/x86-64 UNSIGNED relocs are almost always 64-bit (length=3), but UNSIGNED relocs of length=2 are allowed if the target resides in the low 32-bits. This patch adds support for such relocations in JITLink (previously they would have triggered an unsupported relocation error). llvm-svn: 367764
2019-08-04 04:17:10 +08:00
else
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
return TargetSymbolOrErr.takeError();
[JITLink] Add support for MachO/x86-64 UNSIGNED relocs with length=2. MachO/x86-64 UNSIGNED relocs are almost always 64-bit (length=3), but UNSIGNED relocs of length=2 are allowed if the target resides in the low 32-bits. This patch adds support for such relocations in JITLink (previously they would have triggered an unsupported relocation error). llvm-svn: 367764
2019-08-04 04:17:10 +08:00
Addend = *(const ulittle32_t *)FixupContent;
break;
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
case Pointer64:
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
if (auto TargetSymbolOrErr = findSymbolByIndex(RI.r_symbolnum))
TargetSymbol = TargetSymbolOrErr->GraphSymbol;
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
else
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
return TargetSymbolOrErr.takeError();
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
Addend = *(const ulittle64_t *)FixupContent;
break;
case Pointer64Anon: {
JITTargetAddress TargetAddress = *(const ulittle64_t *)FixupContent;
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
if (auto TargetSymbolOrErr = findSymbolByAddress(TargetAddress))
TargetSymbol = &*TargetSymbolOrErr;
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
else
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
return TargetSymbolOrErr.takeError();
Addend = TargetAddress - TargetSymbol->getAddress();
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
break;
}
case PCRel32Minus1:
case PCRel32Minus2:
case PCRel32Minus4:
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
if (auto TargetSymbolOrErr = findSymbolByIndex(RI.r_symbolnum))
TargetSymbol = TargetSymbolOrErr->GraphSymbol;
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
else
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
return TargetSymbolOrErr.takeError();
[JITLink] Make sure MachO/x86-64 handles 32-bit signed addends correctly. These need to be sign extended when loading into Edge addends.
2019-11-28 14:25:59 +08:00
Addend = *(const little32_t *)FixupContent +
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
(1 << (*Kind - PCRel32Minus1));
break;
case PCRel32Anon: {
JITTargetAddress TargetAddress =
[JITLink] Make sure MachO/x86-64 handles 32-bit signed addends correctly. These need to be sign extended when loading into Edge addends.
2019-11-28 14:25:59 +08:00
FixupAddress + 4 + *(const little32_t *)FixupContent;
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
if (auto TargetSymbolOrErr = findSymbolByAddress(TargetAddress))
TargetSymbol = &*TargetSymbolOrErr;
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
else
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
return TargetSymbolOrErr.takeError();
Addend = TargetAddress - TargetSymbol->getAddress();
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
break;
}
case PCRel32Minus1Anon:
case PCRel32Minus2Anon:
case PCRel32Minus4Anon: {
[JITLink] Silence some MSVC implicit cast warnings. llvm-svn: 358824
2019-04-21 02:30:16 +08:00
JITTargetAddress Delta =
Fix MSVC "32-bit shift implicitly converted to 64 bits" warning. NFCI. llvm-svn: 358970
2019-04-23 19:16:16 +08:00
static_cast<JITTargetAddress>(1ULL << (*Kind - PCRel32Minus1Anon));
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
JITTargetAddress TargetAddress =
[JITLink] Make sure MachO/x86-64 handles 32-bit signed addends correctly. These need to be sign extended when loading into Edge addends.
2019-11-28 14:25:59 +08:00
FixupAddress + 4 + Delta + *(const little32_t *)FixupContent;
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
if (auto TargetSymbolOrErr = findSymbolByAddress(TargetAddress))
TargetSymbol = &*TargetSymbolOrErr;
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
else
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
return TargetSymbolOrErr.takeError();
Addend = TargetAddress - TargetSymbol->getAddress();
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
break;
}
case Delta32:
case Delta64: {
// We use Delta32/Delta64 to represent SUBTRACTOR relocations.
// parsePairRelocation handles the paired reloc, and returns the
// edge kind to be used (either Delta32/Delta64, or
// NegDelta32/NegDelta64, depending on the direction of the
// subtraction) along with the addend.
auto PairInfo =
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
parsePairRelocation(*BlockToFix, *Kind, RI, FixupAddress,
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
FixupContent, ++RelItr, RelEnd);
if (!PairInfo)
return PairInfo.takeError();
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
std::tie(*Kind, TargetSymbol, Addend) = *PairInfo;
assert(TargetSymbol && "No target symbol from parsePairRelocation?");
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
break;
}
default:
llvm_unreachable("Special relocation kind should not appear in "
"mach-o file");
}
LLVM_DEBUG({
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
Edge GE(*Kind, FixupAddress - BlockToFix->getAddress(), *TargetSymbol,
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
Addend);
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
printEdge(dbgs(), *BlockToFix, GE,
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
getMachOX86RelocationKindName(*Kind));
dbgs() << "\n";
});
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
BlockToFix->addEdge(*Kind, FixupAddress - BlockToFix->getAddress(),
*TargetSymbol, Addend);
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
}
}
return Error::success();
}
};
[JITLink] Factor basic common GOT and stub creation code into its own class. llvm-svn: 358838
2019-04-21 11:14:42 +08:00
class MachO_x86_64_GOTAndStubsBuilder
: public BasicGOTAndStubsBuilder<MachO_x86_64_GOTAndStubsBuilder> {
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
public:
[JITLink] Add a MachO x86-64 GOT and Stub bypass optimization. This optimization bypasses GOT loads and calls/branches through stubs when the ultimate target of the access/branch is found to be within range of the reference. Extra debugging output is also added to the generic JITLink algorithm and basic GOT and Stubs builder utility to aid debugging.
2020-02-24 14:53:48 +08:00
static const uint8_t NullGOTEntryContent[8];
static const uint8_t StubContent[6];
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
MachO_x86_64_GOTAndStubsBuilder(LinkGraph &G)
[JITLink] Factor basic common GOT and stub creation code into its own class. llvm-svn: 358838
2019-04-21 11:14:42 +08:00
: BasicGOTAndStubsBuilder<MachO_x86_64_GOTAndStubsBuilder>(G) {}
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
[JITLink] Factor basic common GOT and stub creation code into its own class. llvm-svn: 358838
2019-04-21 11:14:42 +08:00
bool isGOTEdge(Edge &E) const {
return E.getKind() == PCRel32GOT || E.getKind() == PCRel32GOTLoad;
}
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
Symbol &createGOTEntry(Symbol &Target) {
auto &GOTEntryBlock = G.createContentBlock(
getGOTSection(), getGOTEntryBlockContent(), 0, 8, 0);
GOTEntryBlock.addEdge(Pointer64, 0, Target, 0);
return G.addAnonymousSymbol(GOTEntryBlock, 0, 8, false, false);
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
}
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
void fixGOTEdge(Edge &E, Symbol &GOTEntry) {
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
assert((E.getKind() == PCRel32GOT || E.getKind() == PCRel32GOTLoad) &&
"Not a GOT edge?");
[JITLink] Add a MachO x86-64 GOT and Stub bypass optimization. This optimization bypasses GOT loads and calls/branches through stubs when the ultimate target of the access/branch is found to be within range of the reference. Extra debugging output is also added to the generic JITLink algorithm and basic GOT and Stubs builder utility to aid debugging.
2020-02-24 14:53:48 +08:00
// If this is a PCRel32GOT then change it to an ordinary PCRel32. If it is
// a PCRel32GOTLoad then leave it as-is for now. We will use the kind to
// check for GOT optimization opportunities in the
// optimizeMachO_x86_64_GOTAndStubs pass below.
if (E.getKind() == PCRel32GOT)
E.setKind(PCRel32);
[JITLink] Factor basic common GOT and stub creation code into its own class. llvm-svn: 358838
2019-04-21 11:14:42 +08:00
E.setTarget(GOTEntry);
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
// Leave the edge addend as-is.
}
[JITLink] Factor basic common GOT and stub creation code into its own class. llvm-svn: 358838
2019-04-21 11:14:42 +08:00
bool isExternalBranchEdge(Edge &E) {
return E.getKind() == Branch32 && !E.getTarget().isDefined();
}
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
Symbol &createStub(Symbol &Target) {
auto &StubContentBlock =
G.createContentBlock(getStubsSection(), getStubBlockContent(), 0, 1, 0);
[JITLink] Factor basic common GOT and stub creation code into its own class. llvm-svn: 358838
2019-04-21 11:14:42 +08:00
// Re-use GOT entries for stub targets.
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
auto &GOTEntrySymbol = getGOTEntrySymbol(Target);
StubContentBlock.addEdge(PCRel32, 2, GOTEntrySymbol, 0);
return G.addAnonymousSymbol(StubContentBlock, 0, 6, true, false);
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
}
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
void fixExternalBranchEdge(Edge &E, Symbol &Stub) {
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
assert(E.getKind() == Branch32 && "Not a Branch32 edge?");
assert(E.getAddend() == 0 && "Branch32 edge has non-zero addend?");
[JITLink] Add a MachO x86-64 GOT and Stub bypass optimization. This optimization bypasses GOT loads and calls/branches through stubs when the ultimate target of the access/branch is found to be within range of the reference. Extra debugging output is also added to the generic JITLink algorithm and basic GOT and Stubs builder utility to aid debugging.
2020-02-24 14:53:48 +08:00
// Set the edge kind to Branch32ToStub. We will use this to check for stub
// optimization opportunities in the optimizeMachO_x86_64_GOTAndStubs pass
// below.
E.setKind(Branch32ToStub);
[JITLink] Factor basic common GOT and stub creation code into its own class. llvm-svn: 358838
2019-04-21 11:14:42 +08:00
E.setTarget(Stub);
}
private:
Section &getGOTSection() {
if (!GOTSection)
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
GOTSection = &G.createSection("$__GOT", sys::Memory::MF_READ);
[JITLink] Factor basic common GOT and stub creation code into its own class. llvm-svn: 358838
2019-04-21 11:14:42 +08:00
return *GOTSection;
}
Section &getStubsSection() {
if (!StubsSection) {
auto StubsProt = static_cast<sys::Memory::ProtectionFlags>(
sys::Memory::MF_READ | sys::Memory::MF_EXEC);
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
StubsSection = &G.createSection("$__STUBS", StubsProt);
[JITLink] Factor basic common GOT and stub creation code into its own class. llvm-svn: 358838
2019-04-21 11:14:42 +08:00
}
return *StubsSection;
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
}
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
StringRef getGOTEntryBlockContent() {
return StringRef(reinterpret_cast<const char *>(NullGOTEntryContent),
sizeof(NullGOTEntryContent));
}
StringRef getStubBlockContent() {
return StringRef(reinterpret_cast<const char *>(StubContent),
sizeof(StubContent));
}
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
Section *GOTSection = nullptr;
Section *StubsSection = nullptr;
};
[JITLink] Factor basic common GOT and stub creation code into its own class. llvm-svn: 358838
2019-04-21 11:14:42 +08:00
const uint8_t MachO_x86_64_GOTAndStubsBuilder::NullGOTEntryContent[8] = {
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00};
[JITLink] Factor basic common GOT and stub creation code into its own class. llvm-svn: 358838
2019-04-21 11:14:42 +08:00
const uint8_t MachO_x86_64_GOTAndStubsBuilder::StubContent[6] = {
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
0xFF, 0x25, 0x00, 0x00, 0x00, 0x00};
} // namespace
Give helpers internal linkage. NFC.
2020-03-11 01:24:04 +08:00
static Error optimizeMachO_x86_64_GOTAndStubs(LinkGraph &G) {
[JITLink] Add a MachO x86-64 GOT and Stub bypass optimization. This optimization bypasses GOT loads and calls/branches through stubs when the ultimate target of the access/branch is found to be within range of the reference. Extra debugging output is also added to the generic JITLink algorithm and basic GOT and Stubs builder utility to aid debugging.
2020-02-24 14:53:48 +08:00
LLVM_DEBUG(dbgs() << "Optimizing GOT entries and stubs:\n");
for (auto *B : G.blocks())
for (auto &E : B->edges())
if (E.getKind() == PCRel32GOTLoad) {
assert(E.getOffset() >= 3 && "GOT edge occurs too early in block");
// Switch the edge kind to PCRel32: Whether we change the edge target
// or not this will be the desired kind.
E.setKind(PCRel32);
// Optimize GOT references.
auto &GOTBlock = E.getTarget().getBlock();
assert(GOTBlock.getSize() == G.getPointerSize() &&
"GOT entry block should be pointer sized");
assert(GOTBlock.edges_size() == 1 &&
"GOT entry should only have one outgoing edge");
auto &GOTTarget = GOTBlock.edges().begin()->getTarget();
JITTargetAddress EdgeAddr = B->getAddress() + E.getOffset();
JITTargetAddress TargetAddr = GOTTarget.getAddress();
// Check that this is a recognized MOV instruction.
// FIXME: Can we assume this?
constexpr uint8_t MOVQRIPRel[] = {0x48, 0x8b};
if (strncmp(B->getContent().data() + E.getOffset() - 3,
reinterpret_cast<const char *>(MOVQRIPRel), 2) != 0)
continue;
int64_t Displacement = TargetAddr - EdgeAddr + 4;
if (Displacement >= std::numeric_limits<int32_t>::min() &&
Displacement <= std::numeric_limits<int32_t>::max()) {
E.setTarget(GOTTarget);
auto *BlockData = reinterpret_cast<uint8_t *>(
const_cast<char *>(B->getContent().data()));
BlockData[E.getOffset() - 2] = 0x8d;
LLVM_DEBUG({
dbgs() << " Replaced GOT load wih LEA:\n ";
printEdge(dbgs(), *B, E,
getMachOX86RelocationKindName(E.getKind()));
dbgs() << "\n";
});
}
} else if (E.getKind() == Branch32ToStub) {
// Switch the edge kind to PCRel32: Whether we change the edge target
// or not this will be the desired kind.
E.setKind(Branch32);
auto &StubBlock = E.getTarget().getBlock();
assert(StubBlock.getSize() ==
sizeof(MachO_x86_64_GOTAndStubsBuilder::StubContent) &&
"Stub block should be stub sized");
assert(StubBlock.edges_size() == 1 &&
"Stub block should only have one outgoing edge");
auto &GOTBlock = StubBlock.edges().begin()->getTarget().getBlock();
assert(GOTBlock.getSize() == G.getPointerSize() &&
"GOT block should be pointer sized");
assert(GOTBlock.edges_size() == 1 &&
"GOT block should only have one outgoing edge");
auto &GOTTarget = GOTBlock.edges().begin()->getTarget();
JITTargetAddress EdgeAddr = B->getAddress() + E.getOffset();
JITTargetAddress TargetAddr = GOTTarget.getAddress();
int64_t Displacement = TargetAddr - EdgeAddr + 4;
if (Displacement >= std::numeric_limits<int32_t>::min() &&
Displacement <= std::numeric_limits<int32_t>::max()) {
E.setTarget(GOTTarget);
LLVM_DEBUG({
dbgs() << " Replaced stub branch with direct branch:\n ";
printEdge(dbgs(), *B, E,
getMachOX86RelocationKindName(E.getKind()));
dbgs() << "\n";
});
}
}
return Error::success();
}
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
namespace llvm {
namespace jitlink {
class MachOJITLinker_x86_64 : public JITLinker<MachOJITLinker_x86_64> {
friend class JITLinker<MachOJITLinker_x86_64>;
public:
MachOJITLinker_x86_64(std::unique_ptr<JITLinkContext> Ctx,
PassConfiguration PassConfig)
: JITLinker(std::move(Ctx), std::move(PassConfig)) {}
private:
StringRef getEdgeKindName(Edge::Kind R) const override {
return getMachOX86RelocationKindName(R);
}
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
Expected<std::unique_ptr<LinkGraph>>
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
buildGraph(MemoryBufferRef ObjBuffer) override {
auto MachOObj = object::ObjectFile::createMachOObjectFile(ObjBuffer);
if (!MachOObj)
return MachOObj.takeError();
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
return MachOLinkGraphBuilder_x86_64(**MachOObj).buildGraph();
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
}
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
static Error targetOutOfRangeError(const Block &B, const Edge &E) {
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
std::string ErrMsg;
{
raw_string_ostream ErrStream(ErrMsg);
[JITLink] Improve/fix some JITLink debugging output. Adds full edge details (rather than just edge targets) when out-of-range errors are generated. Also fixes a bug where debugging output accessed an invalidated DenseMap iterator by moving the debugging output above the invalidation point. llvm-svn: 360383
2019-05-10 06:03:57 +08:00
ErrStream << "Relocation target out of range: ";
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
printEdge(ErrStream, B, E, getMachOX86RelocationKindName(E.getKind()));
[JITLink] Improve/fix some JITLink debugging output. Adds full edge details (rather than just edge targets) when out-of-range errors are generated. Also fixes a bug where debugging output accessed an invalidated DenseMap iterator by moving the debugging output above the invalidation point. llvm-svn: 360383
2019-05-10 06:03:57 +08:00
ErrStream << "\n";
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
}
return make_error<JITLinkError>(std::move(ErrMsg));
}
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
Error applyFixup(Block &B, const Edge &E, char *BlockWorkingMem) const {
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
using namespace support;
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
char *FixupPtr = BlockWorkingMem + E.getOffset();
JITTargetAddress FixupAddress = B.getAddress() + E.getOffset();
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
switch (E.getKind()) {
case Branch32:
case PCRel32:
case PCRel32Anon: {
int64_t Value =
E.getTarget().getAddress() - (FixupAddress + 4) + E.getAddend();
if (Value < std::numeric_limits<int32_t>::min() ||
Value > std::numeric_limits<int32_t>::max())
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
return targetOutOfRangeError(B, E);
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
*(little32_t *)FixupPtr = Value;
break;
}
case Pointer64:
case Pointer64Anon: {
uint64_t Value = E.getTarget().getAddress() + E.getAddend();
*(ulittle64_t *)FixupPtr = Value;
break;
}
case PCRel32Minus1:
case PCRel32Minus2:
case PCRel32Minus4: {
int Delta = 4 + (1 << (E.getKind() - PCRel32Minus1));
int64_t Value =
E.getTarget().getAddress() - (FixupAddress + Delta) + E.getAddend();
if (Value < std::numeric_limits<int32_t>::min() ||
Value > std::numeric_limits<int32_t>::max())
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
return targetOutOfRangeError(B, E);
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
*(little32_t *)FixupPtr = Value;
break;
}
case PCRel32Minus1Anon:
case PCRel32Minus2Anon:
case PCRel32Minus4Anon: {
int Delta = 4 + (1 << (E.getKind() - PCRel32Minus1Anon));
int64_t Value =
E.getTarget().getAddress() - (FixupAddress + Delta) + E.getAddend();
if (Value < std::numeric_limits<int32_t>::min() ||
Value > std::numeric_limits<int32_t>::max())
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
return targetOutOfRangeError(B, E);
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
*(little32_t *)FixupPtr = Value;
break;
}
case Delta32:
case Delta64:
case NegDelta32:
case NegDelta64: {
int64_t Value;
if (E.getKind() == Delta32 || E.getKind() == Delta64)
Value = E.getTarget().getAddress() - FixupAddress + E.getAddend();
else
Value = FixupAddress - E.getTarget().getAddress() + E.getAddend();
if (E.getKind() == Delta32 || E.getKind() == NegDelta32) {
if (Value < std::numeric_limits<int32_t>::min() ||
Value > std::numeric_limits<int32_t>::max())
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
return targetOutOfRangeError(B, E);
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
*(little32_t *)FixupPtr = Value;
} else
*(little64_t *)FixupPtr = Value;
break;
}
[JITLink] Add support for MachO/x86-64 UNSIGNED relocs with length=2. MachO/x86-64 UNSIGNED relocs are almost always 64-bit (length=3), but UNSIGNED relocs of length=2 are allowed if the target resides in the low 32-bits. This patch adds support for such relocations in JITLink (previously they would have triggered an unsupported relocation error). llvm-svn: 367764
2019-08-04 04:17:10 +08:00
case Pointer32: {
uint64_t Value = E.getTarget().getAddress() + E.getAddend();
if (Value > std::numeric_limits<uint32_t>::max())
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
return targetOutOfRangeError(B, E);
[JITLink] Add support for MachO/x86-64 UNSIGNED relocs with length=2. MachO/x86-64 UNSIGNED relocs are almost always 64-bit (length=3), but UNSIGNED relocs of length=2 are allowed if the target resides in the low 32-bits. This patch adds support for such relocations in JITLink (previously they would have triggered an unsupported relocation error). llvm-svn: 367764
2019-08-04 04:17:10 +08:00
*(ulittle32_t *)FixupPtr = Value;
break;
}
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
default:
llvm_unreachable("Unrecognized edge kind");
}
return Error::success();
}
uint64_t NullValue = 0;
};
void jitLink_MachO_x86_64(std::unique_ptr<JITLinkContext> Ctx) {
PassConfiguration Config;
Triple TT("x86_64-apple-macosx");
if (Ctx->shouldAddDefaultTargetPasses(TT)) {
[JITLink] Refactor EH-frame handling to support eh-frames with existing relocs. Some targets (E.g. MachO/arm64) use relocations to fix some CFI record fields in the eh-frame section. When relocations are used the initial (pre-relocation) content of the eh-frame section can no longer be interpreted by following the eh-frame specification. This causes errors in the existing eh-frame parser. This patch moves eh-frame handling into two LinkGraph passes that are run after relocations have been parsed (but before they are applied). The first] pass breaks up blocks in the eh-frame section into per-CFI-record blocks, and the second parses blocks of (potentially multiple) CFI records and adds the appropriate edges to any CFI fields that do not have existing relocations. These passes can be run independently of one another. By handling eh-frame splitting/fixing with LinkGraph passes we can both re-use existing relocations for CFI record fields and avoid applying eh-frame fixups before parsing the section (which would complicate the linker and require extra temporary allocations of working memory).
2019-11-05 08:00:09 +08:00
// Add eh-frame passses.
Config.PrePrunePasses.push_back(EHFrameSplitter("__eh_frame"));
Config.PrePrunePasses.push_back(
EHFrameEdgeFixer("__eh_frame", NegDelta32, Delta64, Delta64));
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
// Add a mark-live pass.
if (auto MarkLive = Ctx->getMarkLivePass(TT))
Config.PrePrunePasses.push_back(std::move(MarkLive));
else
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
Config.PrePrunePasses.push_back(markAllSymbolsLive);
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
// Add an in-place GOT/Stubs pass.
[JITLink] Switch from an atom-based model to a "blocks and symbols" model. In the Atom model the symbols, content and relocations of a relocatable object file are represented as a graph of atoms, where each Atom represents a contiguous block of content with a single name (or no name at all if the content is anonymous), and where edges between Atoms represent relocations. If more than one symbol is associated with a contiguous block of content then the content is broken into multiple atoms and layout constraints (represented by edges) are introduced to ensure that the content remains effectively contiguous. These layout constraints must be kept in mind when examining the content associated with a symbol (it may be spread over multiple atoms) or when applying certain relocation types (e.g. MachO subtractors). This patch replaces the Atom model in JITLink with a blocks-and-symbols model. The blocks-and-symbols model represents relocatable object files as bipartite graphs, with one set of nodes representing contiguous content (Blocks) and another representing named or anonymous locations (Symbols) within a Block. Relocations are represented as edges from Blocks to Symbols. This scheme removes layout constraints (simplifying handling of MachO alt-entry symbols, and hopefully ELF sections at some point in the future) and simplifies some relocation logic. llvm-svn: 373689
2019-10-04 11:55:26 +08:00
Config.PostPrunePasses.push_back([](LinkGraph &G) -> Error {
[JITLink] Factor basic common GOT and stub creation code into its own class. llvm-svn: 358838
2019-04-21 11:14:42 +08:00
MachO_x86_64_GOTAndStubsBuilder(G).run();
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
return Error::success();
});
[JITLink] Add a MachO x86-64 GOT and Stub bypass optimization. This optimization bypasses GOT loads and calls/branches through stubs when the ultimate target of the access/branch is found to be within range of the reference. Extra debugging output is also added to the generic JITLink algorithm and basic GOT and Stubs builder utility to aid debugging.
2020-02-24 14:53:48 +08:00
// Add GOT/Stubs optimizer pass.
Config.PostAllocationPasses.push_back(optimizeMachO_x86_64_GOTAndStubs);
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
}
if (auto Err = Ctx->modifyPassConfig(TT, Config))
return Ctx->notifyFailed(std::move(Err));
// Construct a JITLinker and run the link function.
MachOJITLinker_x86_64::link(std::move(Ctx), std::move(Config));
}
StringRef getMachOX86RelocationKindName(Edge::Kind R) {
switch (R) {
case Branch32:
return "Branch32";
[JITLink] Add a MachO x86-64 GOT and Stub bypass optimization. This optimization bypasses GOT loads and calls/branches through stubs when the ultimate target of the access/branch is found to be within range of the reference. Extra debugging output is also added to the generic JITLink algorithm and basic GOT and Stubs builder utility to aid debugging.
2020-02-24 14:53:48 +08:00
case Branch32ToStub:
return "Branch32ToStub";
[JITLink] Add support for MachO/x86-64 UNSIGNED relocs with length=2. MachO/x86-64 UNSIGNED relocs are almost always 64-bit (length=3), but UNSIGNED relocs of length=2 are allowed if the target resides in the low 32-bits. This patch adds support for such relocations in JITLink (previously they would have triggered an unsupported relocation error). llvm-svn: 367764
2019-08-04 04:17:10 +08:00
case Pointer32:
return "Pointer32";
Initial implementation of JITLink - A replacement for RuntimeDyld. Summary: JITLink is a jit-linker that performs the same high-level task as RuntimeDyld: it parses relocatable object files and makes their contents runnable in a target process. JITLink aims to improve on RuntimeDyld in several ways: (1) A clear design intended to maximize code-sharing while minimizing coupling. RuntimeDyld has been developed in an ad-hoc fashion for a number of years and this had led to intermingling of code for multiple architectures (e.g. in RuntimeDyldELF::processRelocationRef) in a way that makes the code more difficult to read, reason about, extend. JITLink is designed to isolate format and architecture specific code, while still sharing generic code. (2) Support for native code models. RuntimeDyld required the use of large code models (where calls to external functions are made indirectly via registers) for many of platforms due to its restrictive model for stub generation (one "stub" per symbol). JITLink allows arbitrary mutation of the atom graph, allowing both GOT and PLT atoms to be added naturally. (3) Native support for asynchronous linking. JITLink uses asynchronous calls for symbol resolution and finalization: these callbacks are passed a continuation function that they must call to complete the linker's work. This allows for cleaner interoperation with the new concurrent ORC JIT APIs, while still being easily implementable in synchronous style if asynchrony is not needed. To maximise sharing, the design has a hierarchy of common code: (1) Generic atom-graph data structure and algorithms (e.g. dead stripping and | memory allocation) that are intended to be shared by all architectures. | + -- (2) Shared per-format code that utilizes (1), e.g. Generic MachO to | atom-graph parsing. | + -- (3) Architecture specific code that uses (1) and (2). E.g. JITLinkerMachO_x86_64, which adds x86-64 specific relocation support to (2) to build and patch up the atom graph. To support asynchronous symbol resolution and finalization, the callbacks for these operations take continuations as arguments: using JITLinkAsyncLookupContinuation = std::function<void(Expected<AsyncLookupResult> LR)>; using JITLinkAsyncLookupFunction = std::function<void(const DenseSet<StringRef> &Symbols, JITLinkAsyncLookupContinuation LookupContinuation)>; using FinalizeContinuation = std::function<void(Error)>; virtual void finalizeAsync(FinalizeContinuation OnFinalize); In addition to its headline features, JITLink also makes other improvements: - Dead stripping support: symbols that are not used (e.g. redundant ODR definitions) are discarded, and take up no memory in the target process (In contrast, RuntimeDyld supported pointer equality for weak definitions, but the redundant definitions stayed resident in memory). - Improved exception handling support. JITLink provides a much more extensive eh-frame parser than RuntimeDyld, and is able to correctly fix up many eh-frame sections that RuntimeDyld currently (silently) fails on. - More extensive validation and error handling throughout. This initial patch supports linking MachO/x86-64 only. Work on support for other architectures and formats will happen in-tree. Differential Revision: https://reviews.llvm.org/D58704 llvm-svn: 358818
2019-04-21 01:10:34 +08:00
case Pointer64:
return "Pointer64";
case Pointer64Anon:
return "Pointer64Anon";
case PCRel32:
return "PCRel32";
case PCRel32Minus1:
return "PCRel32Minus1";
case PCRel32Minus2:
return "PCRel32Minus2";
case PCRel32Minus4:
return "PCRel32Minus4";
case PCRel32Anon:
return "PCRel32Anon";
case PCRel32Minus1Anon:
return "PCRel32Minus1Anon";
case PCRel32Minus2Anon:
return "PCRel32Minus2Anon";
case PCRel32Minus4Anon:
return "PCRel32Minus4Anon";
case PCRel32GOTLoad:
return "PCRel32GOTLoad";
case PCRel32GOT:
return "PCRel32GOT";
case PCRel32TLV:
return "PCRel32TLV";
case Delta32:
return "Delta32";
case Delta64:
return "Delta64";
case NegDelta32:
return "NegDelta32";
case NegDelta64:
return "NegDelta64";
default:
return getGenericEdgeKindName(static_cast<Edge::Kind>(R));
}
}
} // end namespace jitlink
} // end namespace llvm