We've currently got 3 files doing open coded implementations of cpuid.
Each has its own set of workarounds and varying levels of how well
they're written and are generally hardcoded to specific cpuid functions.
If you try to build the latest gdb as a PIE on an i386 system, the build
will fail because one of them lacks PIC workarounds (wrt ebx).
Specifically, we have:
common/linux-btrace.c:
two copies of cpuid asm w/specific args, one has no workarounds
while the other implicitly does to avoid memcpy
go32-nat.c:
two copies of cpuid asm w/specific args, one has workarounds to
avoid memcpy
gdb/testsuite/gdb.arch/i386-cpuid.h:
one general cpuid asm w/many workarounds copied from older gcc
Fortunately, that last header there is pretty damn good -- it handles
lots of edge cases, the code is nice & tight (uses gcc asm operands
rather than manual movs), and is already almost a general library type
header. It's also the basis of what is now the public cpuid.h that is
shipped with gcc-4.3+.
So what I've done is pull that test header out and into gdb/common/
(not sure if there's a better place), synced to the version found in
gcc-4.8.0, put a wrapper API around it, and then cut over all the
existing call points to this new header.
Since the func already has support for "is cpuid supported on this proc",
it makes it trivial to push the i386/x86_64 ifdefs down into this wrapper
API too. Now it can be safely used for all targets and gcc will elide
the unused code for us.
I've verified the gdb.arch testsuite still passes, and this code compiles
for an armv7a host as well as x86_64. The go32-nat code has been left
ifdef-ed out until someone can test & verify the new stuff works (and if
it doesn't, figure out how to make the new code work).
URL: https://bugs.gentoo.org/467806
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
before using it.
(dw2_expand_symtabs_matching): Fix symbol kind validity check.
Move test of cu_index closer to use. Print complaint if cu_index
is bad.
We wrote a test case that tries every single 32bit opcode on the hardware
and compared it to the sim. There were a bunch of places in the sim where
we weren't strict enough (requiring certain parts of the opcode be set) so
we were treating a lot of invalid opcodes as valid ones. This sprinkles
out a lot additional checks in the dsp32alu class.
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
This patch fixes a cleanup leak in macho_symfile_read (symbol_table):
symbol_table = (asymbol **) xmalloc (storage_needed);
make_cleanup (xfree, symbol_table);
Unfortunately, fixing the leak alone triggers a crash which occurs
while loading the symbols from an executable:
% gdb
(gdb) file g_exe
[SIGSEGV]
The crash is caused by the fact that performing the cleanup
right after the call to macho_symtab_read, as currently done,
is too early.
Indeed, references to this symbol_table get saved in the oso_vector
global during the call to macho_symtab_read via calls to
macho_register_oso, and those references then get accessed
later on, when processing all the OSOs that got pushed (see
call to macho_symfile_read_all_oso).
This patch prevents this by using one single cleanup queue for
the entire function, rather than having additional separate
cleanup queues (Eg: for the handling of the minimal symbols),
thus preventing the premature free'ing of the minimal_symbols
array.
Secondly, this patch takes this opportunity for avoiding the use
of the oso_vector global, thus making it simpler to track its
lifetime.
gdb/ChangeLog:
* machoread.c (oso_vector): Delete this global.
(macho_register_oso): Add new parameter "oso_vector_ptr".
Use it instead of the "oso_vector" global.
(macho_symtab_read, macho_symfile_read_all_oso): Likewise.
(macho_symfile_read): Use a local oso_vector, to be free'ed
at the end of this function, in place of the old "oso_vector"
global. Update various function calls accordingly. Use one
single cleanup chain for the entire function.
This patch fixes a case of multiple calls freeing the same data
while free-ing objfiles that have child objfiles (separate debug
info, as is the case on Darwin targets).
Following the code, free_objfile_separate_debug iterates over
all child objfiles of the parent objfile, calling free_objfile:
for (child = objfile->separate_debug_objfile; child;)
{
struct objfile *next_child = child->separate_debug_objfile_link;
free_objfile (child);
child = next_child;
}
This causes, among other things, the free'ing of the child objfile's
private data:
/* Discard any data modules have associated with the objfile. The function
still may reference objfile->obfd. */
objfile_free_data (objfile);
This indirectly calls(back) dwarf2_per_objfile_free, which tries
to free the dwarf2read-specific data by using the dwarf2_per_objfile
global, eg:
for (ix = 0; ix < dwarf2_per_objfile->n_comp_units; ++ix)
Even if we were lucky enough the first time around that this global
actually corresponds to the objfile being destroyed, the global
will still have the same value at the second iteration, and thus
become dangling. Indeed, after dwarf2_per_objfile_free returns
eventually back to free_objfile, free_objfile then deallocates
its objfile_obstack, where the dwarf2_per_objfile is allocated.
Ironically, there should be no need to access that global at all,
here, since the data is passed as an argument of the callback.
And it looks like the dwo/dwp/[...]-handling code is in fact already
using that argument, rather than the global.
This patch thus fixes the problem by doing the same, replacing
all references to DWARF2_PER_OBJFILE by uses of DATA instead.
gdb/ChangeLog:
* dwarf2read.c (dwarf2_per_objfile): Replace uses of
DWARF2_PER_OBJFILE by uses of DATA instead.
This fixes PR cli/15603.
The bug here is that when a software watchpoint is being used, gdb
will stop responding to C-c. This is a regression caused by the
"catch signal" patch.
The problem is that software watchpoints always end up on the bpstat
list. However, this makes bpstat_explains_signal return
BPSTAT_SIGNAL_HIDE, causing infrun to think that the signal is not a
"random signal".
The fix is to change bpstat_explains_signal to handle this better. I
chose to do it in a "clean API" way, by passing the signal value to
bpstat_explains_signal and then adding an explains_signal method for
watchpoints, which handles the specifics.
Built and regtested on x86-64 Fedora 18.
New test case included.
* break-catch-sig.c (signal_catchpoint_explains_signal): Add 'sig'
argument.
* breakpoint.c (bpstat_explains_signal): Add 'sig' argument.
Special case signals other than GDB_SIGNAL_TRAP.
(explains_signal_watchpoint): New function.
(base_breakpoint_explains_signal): Add 'sig' argument.
(initialize_breakpoint_ops): Set 'explains_signal' method for
watchpoints.
* breakpoint.h (struct breakpoint_ops) <explains_signal>: Add
signal argument.
(bpstat_explains_signal): Likewise.
* infrun.c (handle_syscall_event, handle_inferior_event): Update.
* gdb.base/random-signal.c: New file.
* gdb.base/random-signal.exp: New file.
* config/tc-mips.c (mips_set_options): Replace separate "ase_*" fields
with a single "ase" bitmask.
(mips_opts): Update accordingly.
(file_ase, file_ase_explicit): New variables.
(file_ase_mips3d, file_ase_mdmx, file_ase_smartmips, file_ase_dsp)
(file_ase_dspr2, file_ase_eva, file_ase_mt, file_ase_virt): Delete.
(ISA_HAS_ROR): Adjust for mips_set_options change.
(is_opcode_valid): Take the base ase mask directly from mips_opts.
(mips_ip): Adjust for mips_set_options change.
(md_parse_option): Likewise. Update file_ase_explicit.
(mips_after_parse_args): Adjust for mips_set_options change.
Use bitmask operations to select the default ASEs. Set file_ase
rather than individual per-ASE variables.
(s_mipsset): Adjust for mips_set_options change.
(mips_elf_final_processing): Test file_ase rather than
file_ase_mdmx. Remove commented-out code.
* config/tc-mips.c (mips_cpu_info): Add an "ase" field.
(MIPS_CPU_ASE_SMARTMIPS, MIPS_CPU_ASE_DSP, MIPS_CPU_ASE_MT)
(MIPS_CPU_ASE_MIPS3D, MIPS_CPU_ASE_MDMX, MIPS_CPU_ASE_DSPR2)
(MIPS_CPU_ASE_MCU, MIPS_CPU_ASE_VIRT, MIPS_CPU_ASE_EVA): Delete.
(mips_after_parse_args): Use the new "ase" field to choose
the default ASEs.
(mips_cpu_info_table): Move ASEs from the "flags" field to the
"ase" field.
The skip test currently relies on the order of evaluation of
arguments which is not defined. Use the comma operator where
order is defined instead.
gdb/testsuite/ChangeLog:
2013-06-18 Will Newton <will.newton@linaro.org>
* gdb.base/skip.c: Use comma to evaluate results of foo()
and bar() before passing to baz().
* gdb.base/skip.c: baz() now takes one argument instead of
two.
PR symtab/15391 is a failure with the DW_OP_GNU_implicit_pointer
feature.
I tracked it down to a logic error in read_pieced_value. The code
truncates this_size_bits according to the type size and offset too
early -- it should do it after taking bits_to_skip into account.
This patch fixes the bug.
While testing this, I also tripped across a latent bug because
indirect_pieced_value does not sign-extend where needed. This patch
fixes this bug as well.
Finally, Pedro pointed out that a previous version implemented sign
extension incorrectly. This version introduces a new gdb_sign_extend
function for this. A couple of notes on this function:
* It has the gdb_ prefix to avoid clashes with various libraries that
felt free to avoid proper namespacing. There is a "sign_extend"
function in a Tile GX header, in an SOM-related BFD header (and in
sh64-tdep.c and as a macro in arm-wince-tdep.c, but those are
ours...)
* I looked at all the sign extensions in gdb and didn't see ones that
I felt comfortable converting to use this function; in large part
because I don't have a good way to test the conversion.
Built and regtested on x86-64 Fedora 18. New test cases included;
this required a minor addition to the DWARF assembler. Note that the
DWARF CU made by implptrpiece.exp uses a funny pointer size in order
to show the sign-extension bug on all platforms.
* dwarf2loc.c (read_pieced_value): Truncate this_size_bits
after taking bits_to_skip into account. Sign extend byte_offset.
* utils.h (gdb_sign_extend): Declare.
* utils.c (gdb_sign_extend): New function.
* gdb.dwarf2/implptrpiece.exp: New file.
* gdb.dwarf2/implptrconst.exp (d): New variable.
Print d.
* lib/dwarf2.exp (Dwarf::_location): Handle DW_OP_piece.
python-selftest.exp fails with an error when using the
native-gdbserver.exp board.
The bug is that the selftest code doesn't work in this situation. It
never has.
This patch fixes the problem by pushing the needed check into
do_self_tests. This helps prevent the problem in the future.
* lib/selftest-support.exp (do_self_tests): Reject remote or
non-native targets.
* gdb.gdb/complaints.exp: Remove check.
* gdb.gdb/observer.exp: Remove check.
* gdb.gdb/xfullpath.exp: Remove check.
* gdb.gdb/complaints.exp: Remove check.
Since these insns run in usermode, there should be no need to setup
RET[ENI] to safe values. They won't be dereferenced, and any insn
that returns via them are valid only in supervisor mode. Since this
is in the main exception code path, saving any insn at all is good
as it gets multiplied quickly (as in O(n^2) times).
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
* resolve.cc (Symbol::override_base): Don't override st_type
from plugin placeholder symbols.
(Symbol_table::resolve): Likewise.
(Symbol_table::should_override): Don't complain about TLS mismatch
if the TO symbol is a plugin placeholder.
* testsuite/Makefile.am (plugin_test_tls): New test.
* testsuite/Makefile.in: Regenerate.
* testsuite/plugin_test_tls.sh: New test script.
* testsuite/two_file_test_2_tls.cc: New test source.
* testsuite/two_file_test_tls.cc: New test source.
* elf64-aarch64.c (elf64_aarch64_final_link_relocate): Call
aarch64_resolve_relocation and bfd_elf_aarch64_put_addend to
handle the relocations of R_AARCH64_JUMP26, R_AARCH64_CALL26,
R_AARCH64_LD64_GOT_LO12_NC, R_AARCH64_ADR_GOT_PAGE and
R_AARCH64_GOT_LD_PREL19.
ld/testsuite/
* ld-aarch64/aarch64-elf.exp: Add 'ifunc-7c'.
* ld-aarch64/ifunc-7c.d: New test.
