into the mainline LLDB codebase. MCJIT introduces
API improvements and better architectural support.
This commit adds a new subsystem, the
ProcessDataAllocator, which is responsible for
performing static data allocations on behalf of the
IR transformer. MCJIT currently does not support
the relocations required to store the constant pool
in the same allocation as the function body, so we
allocate a heap region separately and redirect
static data references from the expression to that
heap region in a new IR modification pass.
This patch also fixes bugs in the IR
transformations that were exposed by the transition
to the MCJIT. Finally, the patch also pulls in a
more recent revision of LLVM so that the MCJIT is
available for use.
llvm-svn: 131923
of duplicated code from appearing all over LLDB:
lldb::addr_t
Process::ReadPointerFromMemory (lldb::addr_t vm_addr, Error &error);
bool
Process::WritePointerToMemory (lldb::addr_t vm_addr, lldb::addr_t ptr_value, Error &error);
size_t
Process::ReadScalarIntegerFromMemory (lldb::addr_t addr, uint32_t byte_size, bool is_signed, Scalar &scalar, Error &error);
size_t
Process::WriteScalarToMemory (lldb::addr_t vm_addr, const Scalar &scalar, uint32_t size, Error &error);
in lldb_private::Process the following functions were renamed:
From:
uint64_t
Process::ReadUnsignedInteger (lldb::addr_t load_addr,
size_t byte_size,
Error &error);
To:
uint64_t
Process::ReadUnsignedIntegerFromMemory (lldb::addr_t load_addr,
size_t byte_size,
uint64_t fail_value,
Error &error);
Cleaned up a lot of code that was manually doing what the above functions do
to use the functions listed above.
Added the ability to get a scalar value as a buffer that can be written down
to a process (byte swapping the Scalar value if needed):
uint32_t
Scalar::GetAsMemoryData (void *dst,
uint32_t dst_len,
lldb::ByteOrder dst_byte_order,
Error &error) const;
The "dst_len" can be smaller that the size of the scalar and the least
significant bytes will be written. "dst_len" can also be larger and the
most significant bytes will be padded with zeroes.
Centralized the code that adds or removes address bits for callable and opcode
addresses into lldb_private::Target:
lldb::addr_t
Target::GetCallableLoadAddress (lldb::addr_t load_addr, AddressClass addr_class) const;
lldb::addr_t
Target::GetOpcodeLoadAddress (lldb::addr_t load_addr, AddressClass addr_class) const;
All necessary lldb_private::Address functions now use the target versions so
changes should only need to happen in one place if anything needs updating.
Fixed up a lot of places that were calling :
addr_t
Address::GetLoadAddress(Target*);
to call the Address::GetCallableLoadAddress() or Address::GetOpcodeLoadAddress()
as needed. There were many places in the breakpoint code where things could
go wrong for ARM if these weren't used.
llvm-svn: 131878
thread plan. In order to get the return value, you can call:
void
ThreadPlanCallFunction::RequestReturnValue (lldb::ValueSP &return_value_sp);
This registers a shared pointer to a return value that will get filled in if
everything goes well. After the thread plan is run the return value will be
extracted for you.
Added an ifdef to be able to switch between the LLVM MCJIT and the standand JIT.
We currently have the standard JIT selected because we have some work to do to
get the MCJIT fuctioning properly.
Added the ability to call functions with 6 argument in the x86_64 ABI.
Added the ability for GDBRemoteCommunicationClient to detect if the allocate
and deallocate memory packets are supported and to not call allocate memory
("_M") or deallocate ("_m") if we find they aren't supported.
Modified the ProcessGDBRemote::DoAllocateMemory(...) and ProcessGDBRemote::DoDeallocateMemory(...)
to be able to deal with the allocate and deallocate memory packets not being
supported. If they are not supported, ProcessGDBRemote will switch to calling
"mmap" and "munmap" to allocate and deallocate memory instead using our
trivial function call support.
Modified the "void ProcessGDBRemote::DidLaunchOrAttach()" to correctly ignore
the qHostInfo triple information if any was specified in the target. Currently
if the target only specifies an architecture when creating the target:
(lldb) target create --arch i386 a.out
Then the vendor, os and environemnt will be adopted by the target.
If the target was created with any triple that specifies more than the arch:
(lldb) target create --arch i386-unknown-unknown a.out
Then the target will maintain its triple and not adopt any new values. This
can be used to help force bare board debugging where the dynamic loader for
static files will get used and users can then use "target modules load ..."
to set addressses for any files that are desired.
Added back some convenience functions to the lldb_private::RegisterContext class
for writing registers with unsigned values. Also made all RegisterContext
constructors explicit to make sure we know when an integer is being converted
to a RegisterValue.
llvm-svn: 131370
representing variables whose type must be inferred
from the way they are used. Functions without debug
information now return UnknownAnyTy and must be cast.
Variables with no debug information are not yet using
UnknownAnyTy; instead they are assumed to be void*.
Support for variables of unknown type is coming (and,
in fact, some relevant support functions are included
in this commit) but will take a bit of extra effort.
The testsuite has also been updated to reflect the new
requirement that the result of printf be cast, i.e.
expr (int) printf("Hello world!")
llvm-svn: 131263
variables be evaluated statically.
Also fixed a bug that caused the results of
statically-evaluated expressions to be materialized
improperly.
This bug also removes some duplicate code.
llvm-svn: 131042
an architecture into ArchSpec:
uint32_t
ArchSpec::GetMinimumOpcodeByteSize() const;
uint32_t
ArchSpec::GetMaximumOpcodeByteSize() const;
Added an AddressClass to the Instruction class in Disassembler.h.
This allows decoded instructions to know know if they are code,
code with alternate ISA (thumb), or even data which can be mixed
into code. The instruction does have an address, but it is a good
idea to cache this value so we don't have to look it up more than
once.
Fixed an issue in Opcode::SetOpcodeBytes() where the length wasn't
getting set.
Changed:
bool
SymbolContextList::AppendIfUnique (const SymbolContext& sc);
To:
bool
SymbolContextList::AppendIfUnique (const SymbolContext& sc,
bool merge_symbol_into_function);
This function was typically being used when looking up functions
and symbols. Now if you lookup a function, then find the symbol,
they can be merged into the same symbol context and not cause
multiple symbol contexts to appear in a symbol context list that
describes the same function.
Fixed the SymbolContext not equal operator which was causing mixed
mode disassembly to not work ("disassembler --mixed --name main").
Modified the disassembler classes to know about the fact we know,
for a given architecture, what the min and max opcode byte sizes
are. The InstructionList class was modified to return the max
opcode byte size for all of the instructions in its list.
These two fixes means when disassemble a list of instructions and dump
them and show the opcode bytes, we can format the output more
intelligently when showing opcode bytes. This affects any architectures
that have varying opcode byte sizes (x86_64 and i386). Knowing the max
opcode byte size also helps us to be able to disassemble N instructions
without having to re-read data if we didn't read enough bytes.
Added the ability to set the architecture for the disassemble command.
This means you can easily cross disassemble data for any supported
architecture. I also added the ability to specify "thumb" as an
architecture so that we can force disassembly into thumb mode when
needed. In GDB this was done using a hack of specifying an odd
address when disassembling. I don't want to repeat this hack in LLDB,
so the auto detection between ARM and thumb is failing, just specify
thumb when disassembling:
(lldb) disassemble --arch thumb --name main
You can also have data in say an x86_64 file executable and disassemble
data as any other supported architecture:
% lldb a.out
Current executable set to 'a.out' (x86_64).
(lldb) b main
(lldb) run
(lldb) disassemble --arch thumb --count 2 --start-address 0x0000000100001080 --bytes
0x100001080: 0xb580 push {r7, lr}
0x100001082: 0xaf00 add r7, sp, #0
Fixed Target::ReadMemory(...) to be able to deal with Address argument object
that isn't section offset. When an address object was supplied that was
out on the heap or stack, target read memory would fail. Disassembly uses
Target::ReadMemory(...), and the example above where we disassembler thumb
opcodes in an x86 binary was failing do to this bug.
llvm-svn: 128347
plugin by name on the command line for when there is more than one disassembler
plugin.
Taught the Opcode class to dump itself so that "disassembler -b" will dump
the bytes correctly for each opcode type. Modified all places that were passing
the opcode bytes buffer in so that the bytes could be displayed to just pass
in a bool that indicates if we should dump the opcode bytes since the opcode
now lives inside llvm_private::Instruction.
llvm-svn: 128290
public types and public enums. This was done to keep the SWIG stuff from
parsing all sorts of enums and types that weren't needed, and allows us to
abstract our API better.
llvm-svn: 128239
of Stephen Wilson's idea (thanks for the input Stephen!). What I ended up
doing was:
- Got rid of ArchSpec::CPU (which was a generic CPU enumeration that mimics
the contents of llvm::Triple::ArchType). We now rely upon the llvm::Triple
to give us the machine type from llvm::Triple::ArchType.
- There is a new ArchSpec::Core definition which further qualifies the CPU
core we are dealing with into a single enumeration. If you need support for
a new Core and want to debug it in LLDB, it must be added to this list. In
the future we can allow for dynamic core registration, but for now it is
hard coded.
- The ArchSpec can now be initialized with a llvm::Triple or with a C string
that represents the triple (it can just be an arch still like "i386").
- The ArchSpec can still initialize itself with a architecture type -- mach-o
with cpu type and subtype, or ELF with e_machine + e_flags -- and this will
then get translated into the internal llvm::Triple::ArchSpec + ArchSpec::Core.
The mach-o cpu type and subtype can be accessed using the getter functions:
uint32_t
ArchSpec::GetMachOCPUType () const;
uint32_t
ArchSpec::GetMachOCPUSubType () const;
But these functions are just converting out internal llvm::Triple::ArchSpec
+ ArchSpec::Core back into mach-o. Same goes for ELF.
All code has been updated to deal with the changes.
This should abstract us until later when the llvm::TargetSpec stuff gets
finalized and we can then adopt it.
llvm-svn: 126278
now, in addition to cpu type/subtype and architecture flavor, contains:
- byte order (big endian, little endian)
- address size in bytes
- llvm::Triple for true target triple support and for more powerful plug-in
selection.
llvm-svn: 125602
stuff soon when we get a fix for looking up the "OBJC_IVAR_$_Class.ivar"
style symbols into IRForTarget::ResolveExternals() next week.
llvm-svn: 123507
can avoid running the code in the target if the
expression's result is known and the expression
has no side effects.
Right now this feature is quite conservative in
its guess about side effects, and it only computes
integer results, but the machinery to make it more
sophisticated is there.
llvm-svn: 121952
values or persistent expression variables. Now if an expression consists of
a value that is a child of a variable, or of a persistent variable only, we
will create a value object for it and make a ValueObjectConstResult from it to
freeze the value (for program variables only, not persistent variables) and
avoid running JITed code. For everything else we still parse up and JIT code
and run it in the inferior.
There was also a lot of clean up in the expression code. I made the
ClangExpressionVariables be stored in collections of shared pointers instead
of in collections of objects. This will help stop a lot of copy constructors on
these large objects and also cleans up the code considerably. The persistent
clang expression variables were moved over to the Target to ensure they persist
across process executions.
Added the ability for lldb_private::Target objects to evaluate expressions.
We want to evaluate expressions at the target level in case we aren't running
yet, or we have just completed running. We still want to be able to access the
persistent expression variables between runs, and also evaluate constant
expressions.
Added extra logging to the dynamic loader plug-in for MacOSX. ModuleList objects
can now dump their contents with the UUID, arch and full paths being logged with
appropriate prefix values.
Thread hardened the Communication class a bit by making the connection auto_ptr
member into a shared pointer member and then making a local copy of the shared
pointer in each method that uses it to make sure another thread can't nuke the
connection object while it is being used by another thread.
Added a new file to the lldb/test/load_unload test that causes the test a.out file
to link to the libd.dylib file all the time. This will allow us to test using
the DYLD_LIBRARY_PATH environment variable after moving libd.dylib somewhere else.
llvm-svn: 121745
perform recursive type lookups, because these are not
required for full type fidelity. We also make the
SelectorTable last for the full lifetime of the Clang
compiler; this was the source of many bugs.
llvm-svn: 119835
when a function starts and ends, and also the
disassembly for anything that is a client of
ClangExpressionParser after it has been JIT
compiled.
llvm-svn: 118401
don't crash if we disable logging when some code already has a copy of the
logger. Prior to this fix, logs were handed out as pointers and if they were
held onto while a log got disabled, then it could cause a crash. Now all logs
are handed out as shared pointers so this problem shouldn't happen anymore.
We are also using our new shared pointers that put the shared pointer count
and the object into the same allocation for a tad better performance.
llvm-svn: 118319
Changed all of our synthesized "___clang" functions, types and variables
that get used in expressions over to have a prefix of "$_lldb". Now when we
do name lookups we can easily switch off of the first '$' character to know
if we should look through only our internal (when first char is '$') stuff,
or when we should look through program variables, functions and types.
Converted all of the clang expression code over to using "const ConstString&"
values for names instead of "const char *" since there were many places that
were converting the "const char *" names into ConstString names and them
throwing them away. We now avoid making a lot of ConstString conversions and
benefit from the quick comparisons in a few extra spots.
Converted a lot of code from LLVM coding conventions into LLDB coding
conventions.
llvm-svn: 116634
to using Clang to get type sizes. This fixes a bug
where the type size for a double[2] was being wrongly
reported as 8 instead of 16 bytes, causing problems
for IRForTarget.
Also improved logging so that the next bug in this
area will be easier to find.
llvm-svn: 115208
- Sema is now exported (and there was much rejoicing.)
- Storage classes are now centrally defined.
Also fixed some bugs that the new LLVM picked up.
llvm-svn: 114622
for C++ classes. Replaced it with a less hacky approach:
- If an expression is defined in the context of a
method of class A, then that expression is wrapped as
___clang_class::___clang_expr(void*) { ... }
instead of ___clang_expr(void*) { ... }.
- ___clang_class is resolved as the type of the target
of the "this" pointer in the method the expression
is defined in.
- When reporting the type of ___clang_class, a method
with the signature ___clang_expr(void*) is added to
that class, so that Clang doesn't complain about a
method being defined without a corresponding
declaration.
- Whenever the expression gets called, "this" gets
looked up, type-checked, and then passed in as the
first argument.
This required the following changes:
- The ABIs were changed to support passing of the "this"
pointer as part of trivial calls.
- ThreadPlanCallFunction and ClangFunction were changed
to support passing of an optional "this" pointer.
- ClangUserExpression was extended to perform the
wrapping described above.
- ClangASTSource was changed to revert the changes
required by the hack.
- ClangExpressionParser, IRForTarget, and
ClangExpressionDeclMap were changed to handle
different manglings of ___clang_expr flexibly. This
meant no longer searching for a function called
___clang_expr, but rather looking for a function whose
name *contains* ___clang_expr.
- ClangExpressionParser and ClangExpressionDeclMap now
remember whether "this" is required, and know how to
look it up as necessary.
A few inheritance bugs remain, and I'm trying to resolve
these. But it is now possible to use "this" as well as
refer implicitly to member variables, when in the proper
context.
llvm-svn: 114384
- If you put a semicolon at the end of an expression,
this no longer causes the expression parser to
error out. This was a two-part fix: first,
ClangExpressionDeclMap::Materialize now handles
an empty struct (such as when there is no return
value); second, ASTResultSynthesizer walks backward
from the end of the ASTs until it reaches something
that's not a NullStmt.
- ClangExpressionVariable now properly byte-swaps when
printing itself.
- ClangUtilityFunction now cleans up after itself when
it's done compiling itself.
- Utility functions can now use external functions just
like user expressions.
- If you end your expression with a statement that does
not return a value, the expression now runs correctly
anyway.
Also, added the beginnings of an Objective-C object
validator function, which is neither installed nor used
as yet.
llvm-svn: 113789
expressions. Values used by the expression are
checked by validation functions which cause the
program to crash if the values are unsafe.
Major changes:
- Added IRDynamicChecks.[ch], which contains the
core code related to this feature
- Modified CommandObjectExpression to install the
validator functions into the target process.
- Added an accessor to Process that gets/sets the
helper functions
llvm-svn: 112690
debugger to insert self-contained functions for use by
expressions (mainly for error-checking).
In order to support detecting whether a crash occurred
in one of these helpers -- currently our preferred way
of reporting that an error-check failed -- added a bit
of support for getting the extent of a JITted function
in addition to just its base.
llvm-svn: 112324
The goal is to separate the parser's data from the data
belonging to the parser's clients. This allows clients
to use the parser to obtain (for example) a JIT compiled
function or some DWARF code, and then discard the parser
state.
Previously, parser state was held in ClangExpression and
used liberally by ClangFunction, which inherited from
ClangExpression. The main effects of this refactoring
are:
- reducing ClangExpression to an abstract class that
declares methods that any client must expose to the
expression parser,
- moving the code specific to implementing the "expr"
command from ClangExpression and
CommandObjectExpression into ClangUserExpression,
a new class,
- moving the common parser interaction code from
ClangExpression into ClangExpressionParser, a new
class, and
- making ClangFunction rely only on
ClangExpressionParser and not depend on the
internal implementation of ClangExpression.
Side effects include:
- the compiler interaction code has been factored
out of ClangFunction and is now in an AST pass
(ASTStructExtractor),
- the header file for ClangFunction is now fully
documented,
- several bugs that only popped up when Clang was
deallocated (which never happened, since the
lifetime of the compiler was essentially infinite)
are now fixed, and
- the developer-only "call" command has been
disabled.
I have tested the expr command and the Objective-C
step-into code, which use ClangUserExpression and
ClangFunction, respectively, and verified that they
work. Please let me know if you encounter bugs or
poor documentation.
llvm-svn: 112249
Sean Callanan