Implement the same optimization committed under r271004 on non-EHABI,
non-SJLJ unwinder as well.
Change-Id: I7f80ed91a75d1e778b50ba87cf8fb68658a083c7
llvm-svn: 272680
The whole file is guarded with #if _LIBUNWIND_ARM_EHABI, and then in the
middle we have these two blocks, which render them pretty unused. An
artefact of a refactoring it seems.
NFC.
llvm-svn: 271737
Summary:
This patch changes the libunwind CMake so that it adds certain target flags like '-m32' or '--gcc-toolchain' before including config-ix.cmake.
Since these flags can affect things like check_library_exists([...]) they needed to be added before the tests are performed.
Additionally this patch adds LIBUNWIND_BUILD_32_BITS which defaults to LLVM_BUILD_32_BITS.
This patch fixes:
https://llvm.org/bugs/show_bug.cgi?id=27950https://llvm.org/bugs/show_bug.cgi?id=27959
Reviewers: jroelofs, danalbert, bcraig, rmaprath, compnerd
Subscribers: cfe-commits
Differential Revision: http://reviews.llvm.org/D20889
llvm-svn: 271458
unwind_phase1 and unwind_phase2 allocate their own copies of unw_cursor_t buffers
on the stack. This can blow-up stack usage of the unwinder depending on how these
two functions get inlined into _Unwind_RaiseException. Clang seems to inline
unwind_phase1 into _Unwind_RaiseException but not unwind_phase2, thus creating
two unw_cursor_t buffers on the stack.
One way to work-around this problem is to mark both unwind_phase1 and
unwind_phase2 as noinline. This patch takes the less compiler-dependent approach
and explicitly allocate a unw_cursor_t buffer and pass that into unwind_phase1
and unwind_phase2 functions.
A follow-up patch will replicate this behavior for the non-EHABI and non-SJLJ
implementations.
Reviewers: jroelofs, bcraig.
Differential revision: http://reviews.llvm.org/D20320
llvm-svn: 271004
Cross unwinding requires larger sizes for unw_context_t and unw_cursor_t
buffers (large enough to hold the VRS of any architecture). This is not
desirable for the most common situation where libunwind is used for native
unwinding only.
This patch makes native unwinding the default build configuration. This
will start testing the tigher (compile-time) bounds of unw_context_t
and unw_cursor_t buffers for each architecture.
Change-Id: Ic61c476a75603ca4812959c233cfe56f83dc0b00
llvm-svn: 270972
r270692 seems to have broken gcc builds of libunwind. This is because
statements like:
static_assert(check_fit<Registers_or1k, unw_context_t>::does_fit,
"or1k registers do not fit into unw_context_t");
Do not work when static_assert is a macro taking two parameters, the
extra comma separating the template parameters confuses the pre-processor.
The fix is to change those statements to:
static_assert((check_fit<Registers_or1k, unw_context_t>::does_fit),
"or1k registers do not fit into unw_context_t");
Also fixed a gcc warning about a trivial un-intended narrowing.
Differential revision: http://reviews.llvm.org/D20119
llvm-svn: 270925
Currently libunwind is built to support cross-unwinding [1] by default, which
requires the buffers unw_context_t and unw_cursor_t to be large enough to hold
the vritual register set (VRS) of any supported architecture. This is not
desirable for some platforms where the stack usage of the unwinder needs
to be kept to a minimum (e.g. bare-metal targets). The current patch introduces
a native-only (-DLIBUNWIND_ENABLE_CROSS_UNWINDING=OFF) unwinder variant that
adopts strict sizes for the buffers unw_context_t and unw_cursor_t depending
on the target architecture.
[1] http://www.nongnu.org/libunwind/man/libunwind(3).html#section_4
Change-Id: I380fff9a56c16a0fc520e3b1d8454a34b4a48373
llvm-svn: 270692
This unifies the definition of _LIBUNWIND_BUILD_SJLJ_APIS. It also further
generalises the definition to allow building these APIs on non-Apple targets
when `-fsjlj-excceptions` is used. The header is now clean of macros which
expand to defined checks.
llvm-svn: 267509
Availablity.h is not used within config.h. The locations which use the
availability infrastructure already include the necessary header(s). NFC.
llvm-svn: 267365
Rather than use the `__assert_rtn` on libSystem based targets and a local
`assert_rtn` function on others, expand the function definition into a macro
which will perform the writing to stderr and then abort. This unifies the
definition and behaviour across targets.
Ensure that we flush stderr prior to aborting.
llvm-svn: 267364
The macros were defined identically across both cases. Unify the definitions to
have a single definition for _LIBUWNIND_{HIDDEN,EXPORT} and _LIBUNWIND_LOG.
NFC.
llvm-svn: 267169
Remove the use of undefined behaviour in the c preprocessor by always defining
the value according to the state that was being checked. NFC.
llvm-svn: 266927
Unify the definition of the _LIBUNWIND_SUPPORT_FRAME_APIS macro. This is in
preparation to remove another instance of -Wexpansion-to-defined. NFC.
llvm-svn: 266926
Remove the use of undefined behaviour in the c preprocessor by always defining
the value according to the state that was being checked. NFC.
llvm-svn: 266916
This follows the pattern in the Apple clause duplicating a tuple of definitions.
However, it will define them to a value rather than a defined check to remove
the `-Wexpansion-to-defined` warning (which may be treated as an error).
This also opens the door to unifying the two code paths into one.
NFC.
llvm-svn: 266915
Join the two paths for this macro. At the end of the day, the difference was
that MIPS and ARM on Apple have different behaviour. This is a setup change to
remove an instance of -Wexpansion-to-defined. NFC.
llvm-svn: 266913
Use x29 and x30 for fp and lr respectively.
This does not change the code generation with integrated asm
but using x30 and x29 helps compile the code with gnu as. Currently gas
fails to assemble this code with errors as below.
Error: operand X should be an integer register.
Newer versions of binutils should be fixed, but enough exists in the wild
to make this change harmless and worthy.
Patch by Khem Raj.
llvm-svn: 260595
This patch allows to use libunwind on bare-metal systems that do not
include malloc/free by conditionally turning off nonessential
functionality that requires these functions.
The disabled functionality includes:
* the .cfi_remember_state and .cfi_restore_state instructions;
* the DWARF FDE cache.
The .cfi_{remember,restore}_state instructions don't seem to be used
by contemporary compilers. None of the LLVM backends emit it.
The DWARF FDE cache is bypassed if _LIBUNWIND_NO_HEAP is defined.
Specifically, entries are never added to it, so the search begins
and ends at the statically allocated, empty initial cache.
Such heap-less libunwind on a bare metal system is successfully used
in the ARTIQ project[1], and it is my hope that it will be useful
elsewhere.
[1]: http://m-labs.hk/artiq
Differential Revision: http://reviews.llvm.org/D11897
llvm-svn: 252452
Summary:
Currently, libunwind doesn't support MIPS. However, with this patch
we do allow the library to build, and we warn the user about the lack of
support for MIPS. Also, the dummy unw_getcontext() implementation for MIPS just
traps on function entry in order to avoid any confusion with silent/weird
failures at runtime.
This allows us to test an LLVM-based toolchain without the dependency on a
GCC toolchain. Of course, C++ exception handling and other things that depend
on stack unwinding will not work until we add a proper implementation of the
stub functions.
Reviewers: compnerd, logan
Subscribers: dsanders, llvm-commits
Differential Revision: http://reviews.llvm.org/D13160
llvm-svn: 248673
This patch makes no assumptions on ABI past the ABI defined in
the OpenRISC 1000 spec except that the DWARF register numbers will
be 0-31 for registers r0-r31, which is true for both gcc and
clang at the moment.
llvm-svn: 246413
Use the canonical __aarch64__ predefined macro for 64-bit ARM. Apple-
specific cases are left as __arm64__. Also add an #error for unsupported
architectures to catch this sort of case in the future.
Differential Revision: http://reviews.llvm.org/D12005
llvm-svn: 244893
The register save routine in libunwind was using write-back addressing
mode to r0 for thumb, when that was not only different from the ARM
version and more importantly the register restore, but also saving the
wrong address.
Patch by Manuel Freiberger.
Fixes PR24331.
llvm-svn: 244237
To build libc++abi without libunwind, we should make sure that all
function calls to _Unwind_{Get,Set}{GR,IP}() are inlined as function
calls to _Unwind_VRS_{Get,Set}(). Otherwise, libc++abi.so will fail to
link since libgcc does not provide these symbol at all.
This commit fixes the problem by providing both the inlined version and
exported version.
llvm-svn: 243073
unw_getcontext() should return UNW_ESUCCESS on success. Therefore, the
assembly for AArch64 is incorrect because "ldr x0, #0" is a PC-relative
load instead of an immediate value load.
llvm-svn: 240648
This commit fixes the unw_step() for ARM EHABI. However, this commit
also changes the implementation details for ARM EHABI.
The first change is that the personality function should call
__gnu_unwind_frame() for default (or de facto) frame unwinding based on
the ARM-defined unwind opcode. The function __gnu_unwind_frame() will
in turn calls unw_step() which actually unwinds the frame.
The second change is that the implementation _Unwind_Backtrace() should
no longer calls unw_step() to unwind the frame; since according to ARM
EHABI, the personality function should unwind the frame for us.
Special thanks to Anton for helpful suggestion on the initial version of
this patch.
llvm-svn: 238560
Include algorithm early as otherwise you get a number of particularly unhelpful
messages about failed static assertions. This fixes compilation on Linux with
gcc.
llvm-svn: 237002
This just copies the default clang-format from the LLVM project. Many
developers use clang-format to ensure that the code is appropriately formatted,
and this commit should also test the git-svn bridge.
llvm-svn: 236767
We currently only include <link.h> on CloudABI and Linux. We can enable
it on FreeBSD as well, as it also supports the dl_iterate_phdr()
function that's provided by <link.h>.
FreeBSD, however, does not provide the ElfW() macro. Instead, the
host-specific ELF datastructures are named just Elf_XXX in addition to
the host-independent Elf32_XXX and Elf64_XXX types.
Differential Revision: http://reviews.llvm.org/D8169
Approved by: emaste
llvm-svn: 236154
Asiri Rathnayake