Make PrologEpilogInserter save/restore all callee saved registers in
functions which call __builtin_unwind_init()
__builtin_unwind_init() is an undocumented gcc intrinsic which has
this effect, and is used in libgcc_eh.
Goes part of the way toward fixing PR8541.
This obsoletes the ugly hack to libgcc's unwind code from r245272, and
should also work for other arches, so revert the hack too.
Initial support for the AMD amdfam10 chipsets has been available in the
gcc43 branch under GPLv2. AMD and some linux distributions (OpenSUSE) did
a backport of the amdfam10 support and made it available.
This is a revised subset of the support initially brought in in r236962
and later reverted. The collateral efects seem to have disappeared but
it is still recommended to set the CPUTYPE with caution.
Reviewed by: jkim (ages ago)
MFC after: 3 weeks
The issue solved in r250392 actually originated upstream and was
fixed in upstream gcc43 branch in 2006/10/24 (under GPLv2).
Register it in the appropriate ChangeLog for reference.
MFC after: 3 days
_Unwind_ForcedUnwind in libgcc_s takes as one of it's parameters a stop
function to tell it when to stop unwinding. One of the stop function's
parameters is a _Unwind_Exception_Class. On most architectures this is an
int64_t, however on ARM EABI the gcc developers have made this a char array
with 8 items. While both of these take the same space they are passed into
the stop function differently, an int64_t is passed in in registers r2 and
r3, while the char[8] is passed in as a pointer to the first item in
register r2.
Because libthr expects the value to be an int64_t we would get incorrect
results when it passes a function that take an int64_t but libgcc passes in
a pointer to a char array including crashing.
The fix is to update libgcc_s to make it pass an int64_t to the stop
function and to libstdc++ as it expects _Unwind_Exception_Class to be an
array.
ARM EABI support is disabled by default and can be enabled by setting
WITH_ARM_EABI when building, however only the kernel-toolchain target will
work with this flag until the rest of the support is added.
runtime on amd64, when it is compiled by clang. Some versions of clang
don't save and restore all callee registers, if a __builtin_eh_return()
intrinsic is used in a function. This is particularly bad on amd64.
Until the problem gets fixed by upstream, use an asm statement to force
clang to assume the registers in question are clobbered, when invoking
__builtin_eh_return(), so it will emit code to save and restore them.
This should fix the crashes reported on -current with some C++ programs,
particularly those that throw exceptions over multiple function
boundaries.
Reported by: stefanf
MFC after: 3 days
There is a bug in gcc (GCC/35998) where dwarf reports
sizes of unsigned -1 (0xffffffff).
On NetBSD this generated a faulty CTF entry which then
caused a segfault in ctfmerge. The issue was worked
around in NetBSD's Dtrace but since the issue originated
in gcc, it seems reasonable to fix it here.
Upstream gcc has been slow to react to this issue and
the author that submitted the patch is not interested
in licensing the change to us, so I did an independent
workaround for the issue.
MFC after: 1 week
built with clang. When these are defined the lists are defined similar to:
asm(".section .ctors");
STATIC func_ptr __CTOR_LIST__[1] = { (func_ptr) (-1) };
asm(".section .dtors");
STATIC func_ptr __DTOR_LIST__[1] = { (func_ptr) (-1) };
The problem is clang will move the two arrays out of the .ctors and .dtors
sections causing these sections to contain a single null address. By not
defining these macros we use the version of the code that places the arrays
is their sections by using __attribute__((section(".ctors"))) and similar
for .dtors.
Submitted by: Daisuke Aoyama <aoyama AT peach.ne.jp>
share/mk/sys.mk instead.
This is part of a medium term project to permit deterministic builds of
FreeBSD.
Submitted by: Erik Cederstrand <erik@cederstrand.dk>
Reviewed by: imp, toolchain@
Approved by: cperciva
MFC after: 2 weeks
r238211:
Support TARGET_ARCH=armv6 and TARGET_ARCH=armv6eb
This adds a new TARGET_ARCH for building on ARM
processors that support the ARMv6K multiprocessor
extensions. In particular, these processors have
better support for TLS and mutex operations.
This mostly touches a lot of Makefiles to extend
existing patterns for inferring CPUARCH from ARCH.
It also configures:
* GCC to default to arm1176jz-s
* GCC to predefine __FreeBSD_ARCH_armv6__
* gas to default to ARM_ARCH_V6K
* uname -p to return 'armv6'
* make so that MACHINE_ARCH defaults to 'armv6'
It also changes a number of headers to use
the compiler __ARM_ARCH_XXX__ macros to configure
processor-specific support routines.
Submitted by: Tim Kientzle <kientzle@freebsd.org>
desired, one can turn off the generation of post-ELF standard dtags by
overriding it with --disable-new-dtags after the default switch.
Immediate effect of the change is that -rpath path is now stored both
in DT_RPATH and DT_RUNPATH tags, which is the right way to provide
rpath for dynamic linker supporting DT_RUNPATH per specification.
Reviewed by: kan
MFC after: 1 month
The GCC4.3 branch contains some optimization fixes
that were not considered regressions and therefore
were never backported. We are bringing a couple of
them that are under GPLv2 since they were made
before the license switch upstream.
While here, add the GCC revision numbers in the
log.
Discussed with: jkim
MFC after: 1 week
The patches are unexpectedly causing gcc to fail while
building ports/graphics/ImageMagick even when the cpu
flags are not used.
Reported by: Andreas Tobler
Initial support for the AMD barcelona chipsets has been available in the
gcc43 branch under GPLv2 but was not included when the Core 2 support
was brought to the system gcc.
AMD and some linux distributions (OpenSUSE) did a backport of the amdfam10
support and made them available. Unfortunately this is still experimental
and while it can improve performance, enabling the CPUTYPE may break some
C++ ports (like clang).
Special care was taken to make sure that the patches predate the GPLv3
switch upstream.
Tested by: Vladimir Kushnir
Reviewed by: mm
Approved by: jhb (mentor)
MFC after: 2 weeks
the GCC 4.1 branch and are available under GPLv2.
2007-11-07 Eric Botcazou <ebotcazou@libertysurf.fr>
PR rtl-optimization/33822
* rtl.h (REG_OFFSET): Fix comment.
* var-tracking.c (INT_MEM_OFFSET): New macro.
(var_mem_set): Use it.
(var_mem_delete_and_set): Likewise.
(var_mem_delete): Likewise.
(vt_get_decl_and_offset): Likewise.
(offset_valid_for_tracked_p): New predicate.
(count_uses): Do not track locations with invalid offsets.
(add_uses): Likewise.
(add_stores): Likewise.
http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=129972
2007-11-16 Richard Guenther <rguenther@suse.de>
PR middle-end/34030
* fold-const.c (fold_binary): Use correct types for folding
1 << X & Y to Y >> X & 1.
http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=130242
2008-01-14 Eric Botcazou <ebotcazou@adacore.com>
PR rtl-optimization/31944
* cse.c (remove_pseudo_from_table): New function.
(merge_equiv_classes): Use above function to remove pseudo-registers.
(invalidate): Likewise
http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=131524
2008-01-24 Kaveh R. Ghazi <ghazi@caip.rutgers.edu>
Backport:
2007-11-07 Kenneth Zadeck <zadeck@naturalbridge.com>
PR middle-end/33826
* ipa-pure-const (static_execute): Added code to keep recursive
functions from being marked as pure or const.
* ipa-utils (searchc): Fixed comment.
http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=131807
2008-02-01 Kaveh R. Ghazi <ghazi@caip.rutgers.edu>
Backport:
2007-08-02 Nathan Froyd <froydnj@codesourcery.com>
PR middle-end/25445
* varasm.c (default_binds_local_p_1): Consult flag_whole_program
if we are compiling with -fPIC.
http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=132061
2008-02-04 Richard Guenther <rguenther@suse.de>
PR middle-end/33631
* expr.c (count_type_elements): Give for unions instead of
guessing.
http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=132101
2008-02-14 Alan Modra <amodra@bigpond.net.au>
PR target/34393
* config/rs6000/rs6000.md (restore_stack_block): Force operands[1]
to a reg.
http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=132309
2008-03-25 Richard Guenther <rguenther@suse.de>
Backport from mainline:
2008-02-12 Richard Guenther <rguenther@suse.de>
PR middle-end/35163
* fold-const.c (fold_widened_comparison): Use get_unwidened in
value-preserving mode. Disallow final truncation.
http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=133509
2008-11-30 Eric Botcazou <ebotcazou@adacore.com>
PR target/38287
* config/sparc/sparc.md (divsi3 expander): Remove constraints.
(divsi3_sp32): Add new alternative with 'K' for operand #2.
(cmp_sdiv_cc_set): Factor common string.
(udivsi3_sp32): Add new alternative with 'K' for operand #2.
Add TARGET_V9 case.
(cmp_udiv_cc_set): Factor common string.
http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=142298
Reviewed by: mm
Approved by: jhb (mentor)
MFC after: 1 week
The fix is similar to the one applied in GCC-4.3 in
GCCSVN-r117929 under the GPLv2.
Submitted by: Andrey Simonenko
Reviewed by: mm
Approved by: jhb (mentor)
MFC after: 3 days
but used by some Linux boot loaders). This option prints
out the directory holding the include files needed by
a freestanding program. The default implementation of
this doesn't work on FreeBSD because of the different
include file layout. But it's easy to implement:
just return /usr/include (or the cross-compiling equivalent).
Reviewed by: kan
MFC after: 1 week
Backported from the gcc-4_3-branch, revision 118001,
under the GPLv2.
This issue was also fixed in Apple's gcc.
PR: 157025
Reviewed by: mm
Approved by: jhb (mentor)
MFC: 2 weeks
declares the proper size of a function. Without this macro recent GNU as will
complain about with:
'Error: .size expression for main does not evaluate to a constant.'
Up to now we produce this:
.L.main:
....
.size main, .-main
With the macro defined the output is this:
.L.main:
....
.size main,.-.L.main
This affects only the 64-bit compiler.
Tested with world and kernel on both, 32 and 64-bit powerpc.
as anonymous namespaces are local to the current translation.
GCC PR: c++/33094
Reviewed by: uqs
Approved by: re (kib)
Obtained from: gcc (branches/redhat/gcc-4_1-branch, rev. 129554, GPLv2)
MFC after: 1 week
- Implement -Wvariable-decl.
- Port -Wtrampolines support from gcc3.
(all three also via OpenBSD)
PR: gnu/127136, gnu/157019
Submitted by: Henning Petersen, Pedro Giffuni
MFC after: 6 weeks
These options are supported in this shape in all newer GCC versions.
PR: gnu/155308
Obtained from: gcc 4.3 (rev. 118090, 118973, 120846; GPLv2)
MFC after: 2 weeks
and k8-sse3 cpu-types for -march=/-mtune= gcc options.
These new cpu-types include the SSE3 instruction set that is supported
by all newer AMD Athlon 64 and Opteron processors.
All three cpu-types are supported by clang and all gcc versions
starting with 4.3 SVN rev 124339 (at that time GPLv2 licensed).
PR: gnu/154906
Discussed with: kib, kan, dim
Obtained from: gcc 4.3 (r124339, GPLv2 licensed)
MFC after: 2 weeks
Mark gcc-provided asm files as not requiring executable stack.
It seems that non-FreeBSD ABIs for powerpc64 claim stack non-executable.
Due to this, rs6000_elf_end_indicate_exec_stack() only emit the note for
32 bit target. I decided not to change FreeBSD ABI and patch
emit the notes for both variants.
Reviewed and tested by: nwhitehorn
actually work, linking of libgcc_s.so.1 on ia64 will fail with:
unwind-ia64.So(.text+0x1762): In function `_Unwind_FindEnclosingFunction':
: undefined reference to `_Unwind_FindTableEntry'
unwind-ia64.So(.text+0x1d82): In function `uw_frame_state_for':
: undefined reference to `_Unwind_FindTableEntry'
/usr/bin/ld: libgcc_s.so.1: hidden symbol `_Unwind_FindTableEntry' isn't defined
Repair this by not hiding the _Unwind_FindTableEntry symbol; on FreeBSD,
it is in libc, not in libgcc.
Silence from: current@
stack unwinding, instead of naively trying to check the instruction at
that location. This fixes signal handling in threaded applications after
recent changes regarding unwinding in libthr. While here, clean up our
MD_FROB_UPDATE_CONTEXT() implementation a little.
o force_to_data
o __CTOR_LIST__
o __do_global_dtors_aux_fini_array_entry
o __frame_dummy_init_array_entry
This is necessary because Clang is smart enough to optimize out these
variables was they were marked as __unused__ (and they are also static).
The end result was programs segfaulting because these symbols weren't
present.
This follows the upstream revision 159228 and the author of that
revision (Jan Hubicka <jh at suse.de>) allowed us to backport this to our
GPLv2 GCC.
Discussed with: core
which are apparently "heinous" GNU extensions, so clang can
compile this without using the -fheinous-gnu-extensions option.
Results in *no* binary change, neither with clang, nor with gcc.
Submitted by: Dimitry Andric <dimitry at andric.com>
a variety of bugs in binutils related to handling of 64-bit PPC ELF,
provides a GCC configuration for 64-bit PowerPC on FreeBSD, and
associated build systems tweaks.
Obtained from: projects/ppc64
o) Add TARGET_ABI to the MIPS toolchain build process. This sets the default
ABI to one of o32, n32 or n64. If it is not set, o32 is assumed as that is
the current default.
o) Set the default GCC cpu type to any specified TARGET_CPUTYPE. This is
necessary to have a working "cc" if e.g. mips64 is specified, as binutils
will refuse to link objects using different ISAs in some cases.
o) Add support for n32 and n64 ABIs to binutils and GCC.
o) Add additional required libgcc2 stubs for n32 and n64.
o) Add support for the "mips64r2" architecture to GCC. Add the "octeon"
o) When static linking, wrap default libraries in --start-group and
--end-group. This is required for static linking to work on n64 with the
interdependencies between libraries there. This is what other OSes that
support n64 seem to do, as well.
o) Fix our GCC spec to define __mips64 for 64-bit targets, not __mips64__, the
former being what libgcc, etc., check and the latter seemingly being a
misspelling of a hand merge from a Linux spec.
o) When no TARGET_CPUTYPE is specified at build time, make GCC take the default
ISA from the ABI. Our old defaults were too liberal and assumed that 64-bit
ABIs should default to the MIPS64 ISA and that 32-bit ABIs should default to
the MIPS32 ISA, when we are supporting or will support some systems based on
earlier 32-bit and 64-bit ISAs, most notably MIPS-III.
o) Merge a new opcode file (and support code) from a later version of binutils
and add flags and code necessary to support Octeon-specific instructions.
This should also make merging opcodes for other modern architectures easier.
Reviewed by: imp
the linker spec. Provide the ability to have a default ABI that's
different than o32 (again, for all 4).
Submitted by: C. Jayachandran (JC) with tweaks for o64/o32 by me
r194519 | gonzo | 2009-06-19 17:28:26 -0600 (Fri, 19 Jun 2009) | 3 lines
- set -mabicalls and -msoft-float as a default in order to
simplify building ports
Scrt1.o instead of crt1.o, since the later is built as non-PIC.
Separate i386-elf crt1.c into the pure assembler part and C code,
supplying all data extracted by assembler stub as explicit parameters [1].
Hide and localize _start1 symbol used as an interface between asm and
C code.
In collaboration with: kan
Inspired by: PR i386/127387 [1]
Prodded and tested by: rdivacky [1]
MFC after: 3 weeks
on "i386". Doing it in the compiler is deemed to be less fragile then
attempting to provide a default -march setting via bsd.cpu.mk. FreeBSD
itself has not supported plain 386 CPUs since 5.x.
Suggested by: kan
Requested by: rdivacky
MFC after: 1 month
of just blowing up. A very similar change to this exists which is
GPLv3 licensed, this is my own change.
This problem was triggered by running the Boost regression tests.
See also: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31899
Reviewed by: luigi
Approved by: re (kib)
compiled with stack protector.
Use libssp_nonshared library to pull __stack_chk_fail_local symbol into
each library that needs it instead of pulling it from libc. GCC
generates local calls to this function which result in absolute
relocations put into position-independent code segment, making dynamic
loader do extra work every time given shared library is being relocated
and making affected text pages non-shareable.
Reviewed by: kib
Approved by: re (kib)
inline function support. This should fix instances where gcc
spuriously reports the following error:
error: nested function 'foo' declared but never defined
original intent, but the functionality wasn't implemented until after
gcc 4.2 was released. However, if you compiled a program that would
behave differently before and after this change, gcc 4.2 would have
warned you; hence, everything currently in the base system is
unaffected by this change. This patch also adds additional warnings
about certain inline function-related bogosity, e.g., using a
static non-const local variable in an inline function.
These changes were merged from a snapshot of gcc mainline from March
2007, prior to the GPLv3 switch. I then ran the regression test suite
from a more recent gcc snapshot and fixed the important bugs it found.
I also squelched the following warning unless -pedantic is specified:
foo is static but used in inline function bar which is not static
This is consistent with LLVM's behavior, but not consistent with gcc 4.3.
Reviewed by: arch@
operating systems. Previously, gcc would inhibit the generation of
fsqrt, fsin, and several other floating point instructions, for the
benefit of the old in-kernel math emulator, which was removed over 5
years ago.
for FreeBSD as well.
This is the fix for __gthread_active_p() returning false positives
which was committed as rev. 1.1.1.8.2.1 to RELENG_7 but now looped
back to the vendor branch via the GCC repository and relicensed to
be GPLv2 by me.
Thanks go to gerald@ for getting the fix approved upstream and for
committing to the GCC repository.
PR: 119289
Approved by: core
understand that non-local variables can never be collected, and when
it collected the static variable for mexitcount_libfunc, gcc aborted
on the next use of this variable.
This quick fix is to reinitialize the variable on every use and depend
on garbage collection recovering the small amount of memory wasted by
this, and not worry by the small amount of time wasted by this. It
would be better to initialize the variable together with most of the
other libfuncs in optabs.c and depend on whatever magic is there to
prevent its collection, but we initialize it here to avoid taking at
least 2 more files off the vendor branch.
* freebsd-spec.h (FBSD_TARGET_OS_CPP_BUILTINS):
Use builtin_define_with_int_value() instead of
adding a new check for every new major FreeBSD version.
Motivated by: simon
Discussed with: obrien, kan
This allows FreeBSD/PPC to build and run out of stock CVS sources. This
also takes the file off of the vendor branch.
Submitted by: kan, grehan
Approved by: re, kan
is still mostly the GCC 2.95 one, with faint attempts by me over the years
to document a few new switches. Thus, it is dubious to claim documentation
of specific modern GCC versions.
Noted by: delphij
