Currently, with a program built from these sources:
$ cat extern.c
void foo () {}
$ cat static.c
static void foo () {}
$ cat main.c
int main () { return 0; }
... if you set a breakpoint on "foo", like:
(gdb) break foo
.. when there's debug info, GDB creates a breakpoint with two
locations, one for each of the external and static functions.
But, when there's no debug info, GDB creates a breakpoint with a
single location, for the external foo. Vis:
$ gcc extern.c static.c main.c -o classify.nodebug
$ gcc extern.c static.c main.c -o classify.debug -g
$ gdb classify.nodebug
Reading symbols from classify.nodebug...
(No debugging symbols found in classify.nodebug)
(gdb) b foo
Breakpoint 1 at 0x40048b
(gdb) info breakpoints
Num Type Disp Enb Address What
1 breakpoint keep y 0x000000000040048b <foo+4>
(gdb)
$ gdb classify.debug
Reading symbols from classify.debug...
(gdb) b foo
Breakpoint 1 at 0x40048b: foo. (2 locations)
(gdb) info breakpoints
Num Type Disp Enb Address What
1 breakpoint keep y <MULTIPLE>
1.1 y 0x000000000040048b in foo at extern.c:1
1.2 y 0x0000000000400492 in foo at static.c:1
GDB drops the static function is search_minsyms_for_name, where at the
very end of that function we pick only the locations with highest
classification, according to classify_type.
The classify_type logic was introduced here:
https://sourceware.org/pipermail/gdb-patches/2011-December/087864.html
which said:
"Previously, linespec was trying to filter out minsyms as they were
seen. However, this isn't faithful to gdb's historical approach,
which is to priority-order minsyms; see lookup_minimal_symbol."
lookup_minimal_symbol's intro says, in the .c file:
/* Look through all the current minimal symbol tables and find the
first minimal symbol that matches NAME. If OBJF is non-NULL, limit
the search to that objfile. If SFILE is non-NULL, the only file-scope
symbols considered will be from that source file (global symbols are
still preferred). Returns a pointer to the minimal symbol that
matches, or NULL if no match is found.
Note: One instance where there may be duplicate minimal symbols with
the same name is when the symbol tables for a shared library and the
symbol tables for an executable contain global symbols with the same
names (the dynamic linker deals with the duplication).
It's also possible to have minimal symbols with different mangled
names, but identical demangled names. For example, the GNU C++ v3
ABI requires the generation of two (or perhaps three) copies of
constructor functions --- "in-charge", "not-in-charge", and
"allocate" copies; destructors may be duplicated as well.
Obviously, there must be distinct mangled names for each of these,
but the demangled names are all the same: S::S or S::~S. */
struct bound_minimal_symbol
lookup_minimal_symbol (const char *name, const char *sfile,
struct objfile *objf)
{
If you look inside this function, you'll see:
/* External symbols are best. */
...
/* File-local symbols are next best. */
...
/* Symbols for shared library trampolines are next best. */
...
While this logic is good when you're looking for the single "best"
symbol by name, I question it for linespecs, since we want to set
breakpoints in all the multiple locations that match. I see no point
in hidding static functions.
Now, for breakpoints, it does make sense to filter out PLT/trampoline
symbols if we find the actual global matching function symbol.
Otherwise, if we did no filtering (i.e., just removed the
classify_type logic), you would end up with e.g.:
(gdb) b printf
Breakpoint 1 at 0x413a60 (2 locations)
(top-gdb) info breakpoints
Num Type Disp Enb Address What
1 breakpoint keep y <MULTIPLE>
1.1 y 0x0000000000413a60 <printf@plt>
1.2 y 0x00007ffff4653640 in __printf at printf.c:28
instead of this (which is what we get currently) before the shared
library is loaded (a location set in the PLT):
(gdb) b printf
Breakpoint 1 at 0x413a60
Num Type Disp Enb Address What
1 breakpoint keep y 0x0000000000413a60 <printf@plt>
and this after the library is loaded (only one location, no breakpoint
in the PLT):
(gdb) b printf
Breakpoint 1 at 0x7ffff4653640: file printf.c, line 28.
(gdb) info breakpoints
Num Type Disp Enb Address What
1 breakpoint keep y 0x00007ffff4653640 in __printf at printf.c:28
This patch fixes the missing breakpoint locations issue by replacing
the classify_type logic in linespec.c with a different logic.
Instead, discard a trampoline symbol if we also found a
global/external symbol with the same name. The patch adds a couple of
testcases testing locations in external vs static functions vs
trampolines/PLTs.
We now get:
For the msym-bp.exp testcase (extern vs static), without debug info:
(gdb) info breakpoints
Num Type Disp Enb Address What
1 breakpoint keep y <MULTIPLE>
1.1 y 0x000000000040048b <foo+4> ### missing before patch
1.2 y 0x000000000040049d <foo+4>
For the msym-bp.exp testcase (extern vs static), with debug info:
(gdb) info breakpoints
Num Type Disp Enb Address What
1 breakpoint keep y <MULTIPLE>
1.1 y 0x000000000040048b in foo at src/gdb/testsuite/gdb.base/msym-bp.c:21
1.2 y 0x000000000040049d in foo at src/gdb/testsuite/gdb.base/msym-bp-2.c:21
For the msym-bp-shl.exp testcase (static vs plt), without debug info, before running to main:
(gdb) info breakpoints
Num Type Disp Enb Address What
1 breakpoint keep y <MULTIPLE>
1.1 y 0x00000000004004e0 <foo@plt> ### missing before patch
1.2 y 0x00000000004005db <foo+4>
For the msym-bp-shl.exp testcase (static vs plt), without debug info, after running to main:
(gdb) info breakpoints
Num Type Disp Enb Address What
1 breakpoint keep y <MULTIPLE>
1.1 y 0x00000000004005db <foo+4> ### missing before patch
1.2 y 0x00007ffff7bd65de <foo+4>
For the msym-bp-shl.exp testcase (static vs plt), with debug info, before running to main:
(gdb) info breakpoints
Num Type Disp Enb Address What
1 breakpoint keep y <MULTIPLE>
1.1 y 0x00000000004004e0 <foo@plt> ### missing before patch
1.2 y 0x00000000004005db in foo at src/gdb/testsuite/gdb.base/msym-bp-shl-main-2.c:21
For the msym-bp-shl.exp testcase (static vs plt), with debug info, after running to main:
(gdb) info breakpoints
Num Type Disp Enb Address What
1 breakpoint keep y <MULTIPLE>
1.1 y 0x00000000004005db in foo at src/gdb/testsuite/gdb.base/msym-bp-shl-main-2.c:21
1.2 y 0x00007ffff7bd65de in foo at src/gdb/testsuite/gdb.base/msym-bp-shl-lib.c:21
gdb/ChangeLog:
* linespec.c (classify_mtype, compare_msyms): Delete.
(search_minsyms_for_name): Remove classification logic. Instead
filter out trampoline symbols if we also found an external
function of the same name.
gdb/testsuite/ChangeLog:
* gdb.base/msym-bp-2.c: New.
* gdb.base/msym-bp-shl-lib.c: New file.
* gdb.base/msym-bp-shl-main-2.c: New file.
* gdb.base/msym-bp-shl-main.c: New file.
* gdb.base/msym-bp-shl.exp: New file.
* gdb.base/msym-bp.c: New file.
* gdb.base/msym-bp.exp: New file.
Now that the GDB 10 branch has been created, we can
bump the version number.
gdb/ChangeLog:
GDB 10 branch created (8087c3fa8b5d695e3e29e69d70d0b35ec902ac59):
* version.in: Bump version to 11.0.50.DATE-git.
gdb/testsuite/ChangeLog:
* gdb.base/default.exp: Change $_gdb_major to 11.
I noticed this while testing the GDB in the context of the upcoming
GDB 10 release branching, because part of the process involves setting
development to False, which in turn changes the default for including
unittest to false as well. As a result, without this patch, we get
compilation errors in infrun.c such as:
infrun.c:9219:5: error: `scoped_mock_context' was not declared in this scope
This patch fixes it by bracketing the unitttest in namespace selftest
with an #if GDB_SELF_TEST.
gdb/ChangeLog:
* infrun.c (namespace selftests): Only define #if GDB_SELF_TEST.
Tested on x86_64-linux, with and without self-tests.
The symbol string table in the .symtab section is optional and cosmetic.
The contents of the .symtab section have no impact on run-time execution.
The symbol names in the symbol string table help distinguish addresses at
different locations. Add a linker option, -z unique-symbol, to avoid
duplicated local symbol names in the symbol string table.
This feature was well received by the livepatch maintainers. It not only
solves the duplicated local symbol name problem, but also would allow
livepatch to more precisely locate duplicate symbols in general for
patching.
bfd/
PR ld/26391
* elflink.c (elf_final_link_info): Add local_hash_table.
(local_hash_entry): New.
(local_hash_newfunc): Likewise.
(elf_link_output_symstrtab): Append ".COUNT" to duplicated local
symbols.
(bfd_elf_final_link): Initialize and free local_hash_table for
"-z unique-symbol".
include/
PR ld/26391
* bfdlink.h (bfd_link_info): Add unique_symbol.
ld/
PR ld/26391
* NEWS: Mention "-z unique-symbol".
* emultempl/elf.em (gld${EMULATION_NAME}_handle_option): Handle
"-z unique-symbol" and "-z nounique-symbol".
* ld.texi: Document "-z unique-symbol" and "-z nounique-symbol".
* lexsup.c (elf_static_list_options): Add "-z unique-symbol" and
"-z nounique-symbol".
* testsuite/ld-elf/elf.exp: Add PR ld/26391 tests.
* testsuite/ld-elf/pr26391.nd: New file.
* testsuite/ld-elf/pr26391.out: Likewise.
* testsuite/ld-elf/pr26391a.c: Likewise.
* testsuite/ld-elf/pr26391b.c: Likewise.
* testsuite/ld-elf/pr26391c.c: Likewise.
* testsuite/ld-elf/pr26391d.c: Likewise.
This adds support for the bfloat16 datatype, which can be seen as a short
version of FP32, skipping the least significant 16 bits of the mantissa.
Since the datatype is currently only supported by the AVX512 registers,
the printing of bfloat16 values is only supported for xmm, ymm and zmm
registers.
gdb/ChangeLog:
2020-09-11 Moritz Riesterer <moritz.riesterer@intel.com>
Felix Willgerodt <Felix.Willgerodt@intel.com>
* gdbarch.sh: Added bfloat16 type.
* gdbarch.c: Regenerated.
* gdbarch.h: Regenerated.
* gdbtypes.c (floatformats_bfloat16): New struct.
(gdbtypes_post_init): Add builtin_bfloat16.
* gdbtypes.h (struct builtin_type) <builtin_bfloat16>: New member.
(floatformats_bfloat16): New struct.
* i386-tdep.c (i386_zmm_type): Add field "v32_bfloat16"
(i386_ymm_type): Add field "v16_bfloat16"
(i386_gdbarch_init): Add set_gdbarch_bfloat16_format.
* target-descriptions.c (make_gdb_type): Add case TDESC_TYPE_BFLOAT16.
* gdbsupport/tdesc.cc (tdesc_predefined_types): New member bfloat16.
* gdbsupport/tdesc.h (tdesc_type_kind): New member TDESC_TYPE_BFLOAT16.
* features/i386/64bit-avx512.xml: Add bfloat16 type.
* features/i386/64bit-avx512.c: Regenerated.
* features/i386/64bit-sse.xml: Add bfloat16 type.
* features/i386/64bit-sse.c: Regenerated.
gdb/testsuite/ChangeLog:
2020-09-11 Moritz Riesterer <moritz.riesterer@intel.com>
Felix Willgerodt <Felix.Willgerodt@intel.com>
* x86-avx512bf16.c: New file.
* x86-avx512bf16.exp: Likewise.
* lib/gdb.exp (skip_avx512bf16_tests): New function.
gdb/Changelog:
2020-07-02 Felix Willgerodt <felix.willgerodt@intel.com>
* i386-tdep.c (i386_zmm_type): Fix field names.
(i386_ymm_type): Fix field names.
When -g was used to generate DWARF gas would error out when a .debug_line
already exists. But when a .debug_info section already exists it would
simply skip generating one without warning or error. Do the same for
.debug_line. It is only an error when the user explicitly uses .loc
directives and also generates the .debug_line table itself.
The tests are unfortunately arch specific because the line table is only
generated when actual instructions have been emitted. Use i386 because
that is probably the most used architecture. Before this patch the new
dwarf-line-2 testcase would fail, with this patch it succeeds (and doesn't
try to add its own line table).
gas/ChangeLog:
* as.texi (-g): Explicitly mention when .debug_info and .debug_line
are generated for the DWARF format.
(Loc): Add that it is an error to both use a .loc directive and
generate a .debug_line yourself.
* dwarf2dbg.c (dwarf2_any_loc_directive_seen): New static variable.
(dwarf2_directive_loc): Set dwarf2_any_loc_directive_seen to TRUE.
(dwarf2_finish): Check dwarf2_any_loc_directive_seen before emitting
an error. Only create .debug_line if it is empty (or doesn't exist).
* testsuite/gas/i386/i386.exp: Add dwarf2-line-{1,2,3,4} when testing
an elf target.
* testsuite/gas/i386/dwarf2-line-{1,2,3,4}.{s,d,l}: New test files.
Define NT_X86_CET which is the proposed note for x86 CET state to support
Intel CET in Linux kernel. Double check it after Intel CET patches have
been merged into Linux kernel.
binutils/
* readelf.c (get_note_type): Support NT_X86_CET.
include/
* elf/common.h (NT_X86_CET): New.
PR 26595
* dwarf.c (load_separate_debug_info): Return NULL rather than
FALSE in error situations.
(load_separate_debug_file): Move code to load debug links to ...
(check_for_and_load_links): ... here. New function. Load
separate debug information pointed to by debuglink and
debugaltlink sections. Recursively scan newly loaded debug
information for more links and load them too.
The options for the "breakpoint condition-evaluation" setting are
"host", "target", and "auto". The help message mentions the option
"gdb" at one point instead of "host". Fix this typo. Also add a period.
gdb/ChangeLog:
2020-09-11 Tankut Baris Aktemur <tankut.baris.aktemur@intel.com>
* breakpoint.c: Fix typo in the help message of the
"set breakpoint condition-evaluation" command.
Remove a stale command that is apparently forgotten after a copy-paste
from 'gdb.base/break-main-file-remove-fail.exp'.
gdb/testsuite/ChangeLog:
2020-09-11 Tankut Baris Aktemur <tankut.baris.aktemur@intel.com>
* gdb.base/bp-cmds-execution-x-script.exp: Remove a stale comment.
The fix in 7e05773767820b441b23a16628b55c98cb1aef46 introduced a PLT
for conditional jumps when the target symbol is undefined. This is
incorrect because conditional branch relocations are not allowed to
clobber IP0/IP1 and hence, should not result in a dynamic relocation.
Revert that change and in its place, issue an error when the target
symbol is undefined.
bfd/
2020-09-10 Siddhesh Poyarekar <siddesh.poyarekar@arm.com>
* elfnn-aarch64.c (elfNN_aarch64_final_link_relocate): Revert
changes in 7e05773767820b441b23a16628b55c98cb1aef46. Set
error for undefined symbol in BFD_RELOC_AARCH64_BRANCH19 and
BFD_RELOC_AARCH64_TSTBR14 relocations.
ld/
2020-09-10 Siddhesh Poyarekar <siddesh.poyarekar@arm.com>
* testsuite/ld-aarch64/emit-relocs-560.d: Expect error instead
of valid output.
Prior to
commit 1e3b96fd6cf0c7d018083994ad951ccf92aba582
Author: Alan Modra <amodra@gmail.com>
Date: Fri Sep 4 13:54:21 2020 +0930
Allow plugin syms to mark as-needed shared libs needed
when removing unused IR symbol references, ld didn't add unnecessary
DT_NEEDED libraries which may lead to undefined symbol reference in a
--as-needed library when the symbol is defined in a prior --as-needed
library and there is no reference in relocatable inputs. This behavior
is desirable since it ensures that both lazy and non-lazy bindings work
the same way. The problem is with --as-needed libraries, which happens
with and without LTO. Now, the linker may add many unnecessary DT_NEEDED
libraries for IR inputs.
PR ld/26590
* testsuite/ld-elf/pr26590.err: New file.
* testsuite/ld-elf/pr26590a.c: Likewise.
* testsuite/ld-elf/pr26590b.c: Likewise.
* testsuite/ld-elf/pr26590c.c: Likewise.
* testsuite/ld-elf/pr26590d.c: Likewise.
* testsuite/ld-elf/shared.exp: Run ld/26590 tests.
gcc commit 387d0773f3 changed the D demangler, with the following
commit log:
libiberty: Add support for `in' and `in ref' storage classes.
The storage class `in' is now a first-class citizen with its own mangle
symbol, of which also permits `in ref'. Previously, `in' was an alias
to `const [scope]', which is a type constructor.
The mangle symbol repurposed for this is `I', which was originally used
by identifier types. However, while TypeIdentifier is part of the
grammar, it must be resolved to some other entity during the semantic
passes, and so shouldn't appear anywhere in the mangled name.
Old tests that are now no longer valid have been removed.
This patch makes the same changes to the gdb testsuite as were made to
the upstream gcc libiberty testsuite.
gdb/testsuite/
PR 26597
* gdb.dlang/demangle.exp: Update tests as per gcc commit 387d0773f3.
No functional change as the same functionality inlined in nbsd-nat.c
is offered in gdb/nat/netbsd-nat.c.
gdb/ChangeLog:
* nbsd-nat.c: Include "nat/netbsd-nat.h".
* (nbsd_nat_target::pid_to_exec_file)
(nbsd_nat_target::thread_alive, nbsd_nat_target::thread_name)
(nbsd_nat_target::post_startup_inferior)
(nbsd_nat_target::post_attach, nbsd_nat_target::xfer_partial)
(nbsd_add_threads): Switch local code to common gdb/nat functions.
* (nbsd_pid_to_cmdline): Call sysctl from the global namespace.
* (nbsd_thread_lister): Remove.
Do not free the last execd pathname as it will be used in
prepare_resume_reply(), after attaching a client side.
gdb/ChangeLog:
* fork-inferior.c (startup_inferior): Avoid double free.
Add generic function to enable debugger events in a process.
gdb/ChangeLog:
* netbsd-nat.h (netbsd_nat::enable_proc_events): Add.
* netbsd-nat.c: Include <sys/ptrace.h>.
* (netbsd_nat::enable_proc_events): Add.
* ppc-dis.c (ppc_symbol_is_valid): New function. Returns false
for hidden, local, no-type symbols.
(disassemble_init_powerpc): Point the symbol_is_valid field in the
info structure at the new function.
Add arches CK804, CK805 and CK800. CK800 is an special arch which
support all instructions for CSKYV2. Refine the cpu tables to
simplify adding a new cpu.
Co-Authored-By: Lifang Xia <lifang_xia@c-sky.com>
gas/
* config/tc-csky.c (struct csky_cpu_info): Add new members
isa_flag, features and ver.
(struct csky_cpu_feature): New.
(struct csky_cpu_version): New.
(CSKY_FEATURE_MAX): Define.
(CSKY_CPU_REVERISON_MAX): Define.
(FEATURE_DSP_EXT, FEATURE_DSP, FEATURE_MMU, FEATURE_VDSP,
FEATURE_FLOAT, FEATURE_TRUST, FEATURE_JAVA, FEATURE_SHIELD):
Define, each standard one collection of instructions.
(CSKY_FEATURES_DEF_NULL, CSKY_FEATURES_DEF_e,
CSKY_FEATURES_DEF_t, CSKY_FEATURES_DEF_f, CSKY_FEATURES_DEF_v,
CSKY_FEATURES_DEF_ef, CSKY_FEATURES_DEF_jt,
CSKY_FEATURES_DEF_efht, CSKY_FEATURES_DEF_efv,
CSKY_FEATURES_DEF_eft, CSKY_FEATURES_DEF_d,
CSKY_FEATURES_DEF_df, CSKY_FEATURES_DEF_ft,
CSKY_FEATURES_DEF_tv, CSKY_FEATURES_DEF_fv,
CSKY_FEATURES_DEF_dft, CSKY_FEATURES_DEF_dfv,
CSKY_FEATURES_DEF_ftv, CSKY_FEATURES_DEF_eftv): Define,
the features combination used by cpu.
(CSKY_CPU_REVERISON_r0p0, CSKY_CPU_REVERISON_r1p0,
CSKY_CPU_REVERISON_r2p0, CSKY_CPU_REVERISON_r3p0,
CSKY_CPU_REVERISON_RESERVED, CSKY_CPU_REVERISON_R3):
Define, version information used by cpu.
(csky_cpus): Refine, and add CK804, CK805 and CK800.
(parse_cpu): Refine.
(parse_arch): Refine.
(md_show_usage): Refine.
(md_begin): Refine.
include/
* opcode/csky.h (CSKY_ARCH_804): Define.
(CSKY_ARCH_805): Define.
(CSKY_ARCH_800): Define.
Some MIPS targets, for reasons I didn't analyse, use the larger common
symbol in a shared lib rather than a smaller common in an executable.
That doesn't seem unreasonable, so allow that to pass for pr26580-2.
bfin-elf complains about not supporting copy relocs, but it's quite
silly to want a copy reloc for common symbols, so leave the fail
there. mn10300-elf and score-elf both fail the test with "PHDR
segment not covered by LOAD segment". Other tests fail similarly so
one more doesn't hurt. The failure is a consequence of supporting
dynamic objects but setting EMBEDDED in ld scripts.
PR 26580
* testsuite/ld-elf/pr26580-2.sd: Accept undefined symbol.
Extend the test a little to archives, not that we expect this to
fail. Nor has the lto-18 test ever failed without -flto.
* testsuite/ld-plugin/lto-18b.c (select): Remove.
* testsuite/ld-plugin/lto-18c.c (select): Remove.
* testsuite/ld-plugin/lto.exp: Build archives for lto-18 too,
and run static versions of the test.
incldue * opcode/csky.h (CSKY_ISA_FLOAT_7E60): Use a long long type for
this value.
opcodes * csky-dis.c (csky_output_operand): Coerce the immediate values to
long before printing.
I believe ada-lang.c:remove_extra_symbols has a latent bug. This
function loops over a vector of symbols, removing duplicates according
to some criteria.
At the end of the loop it does:
if (remove_p)
syms->erase (syms->begin () + i);
i += 1;
However, this seems wrong to me -- when removing the i'th element,
this code still increments "i", thus skipping an element.
At first I thought this was a regression from the patches to remove
cleanups from ada-lang.c (this was when std::vector was introduced);
but I found that instead the bug seems to be older:
if (remove_p)
{
for (j = i + 1; j < nsyms; j += 1)
syms[j - 1] = syms[j];
nsyms -= 1;
}
i += 1;
No test, as I don't know how to write one.
As this is Ada-specific, and was already reviewed internally by Joel,
I am checking it in.
gdb/ChangeLog
2020-09-09 Tom Tromey <tromey@adacore.com>
* ada-lang.c (remove_extra_symbols): Do not increment when
removing an element
We don't support power10 on ppc32, mainly because some instructions
have 34-bit fields for which we don't have relocations on ppc32.
If you try to assemble typical code, you'll see errors saying
"reloc ... not supported by object file format". Also, on 32-bit
hosts with binutils configured without a 64-bit bfd, you'll see errors
saying "bignum invalid" when using large offsets. But let's not kill
output of prefix insns entirely on 32-bit hosts.
* config/tc-ppc.c (md_assemble): Emit prefix insn by parts when
valueT is smaller than 64 bits.
The option corresponds to GCC to control the float calling conversion,
and the value will be stored in .csky.attributes section.
Co-Authored-By: Lifang Xia <lifang_xia@c-sky.com>
gas/
* config/tc-csky.c (float_abi): New.
(md_longopts): Add mfloat-abi.
(struct sky_option_value_table): New.
(csky_float_abis): New, the possible values for -mfloat-abi.
(parse_float_abi): New funtion.
(md_show_usage): Show help information for -mfloat-abi.
(set_csky_attribute): Store float-abi value.
The associated check function doesn't do anything with its "data"
param, so nothing to adjust there.
PR 26578
* dwarf.c (parse_gnu_debugaltlink): Don't alloc build_id_data.
(load_separate_debug_files): Use a stack var for build_id_data.
Demonstrates a reason to use IR symbols when deciding an --as-needed
library should be loaded.
* testsuite/ld-plugin/lto-18a.c,
* testsuite/ld-plugin/lto-18b.c,
* testsuite/ld-plugin/lto-18c.c,
* testsuite/ld-plugin/lto-18d.c,
* testsuite/ld-plugin/lto-18.out: New test.
* testsuite/ld-plugin/lto.exp: Run it.
