binutils-gdb.git
15 months agox86: pack CPU flags in opcode table
Jan Beulich [Fri, 11 Aug 2023 08:04:49 +0000 (10:04 +0200)]
x86: pack CPU flags in opcode table

The table constantly growing in two dimensions (number of table entries
times number of ISA extension flags) doesn't scale very well. Use a more
compact representation: Only identifiers which need to combine with
other identifiers retain individual flag bits. All others are combined
into an enum, with a new helper added to transform the table entries
into the original i386_cpu_flags layout. This way the table in the final
binary shrinks by almost a third (the generated source code shrinks by
about half), and isn't likely to grow again in that dimension any time
soon.

While moving the 3DNow! fields, drop the stray inner 'a' from their
names.

15 months agowarn unused result for bfd IO functions
Alan Modra [Mon, 7 Aug 2023 11:29:08 +0000 (20:59 +0930)]
warn unused result for bfd IO functions

This patch fixes all the warnings I found in bfd, binutils and ld,
plus some bitrotted COFF_GO32 code that tried to allocate -168ul
bytes.  When the malloc fail was reported these testsuite fails
resulted:

i386-go32  +FAIL: go32 stub
i386-go32  +ERROR: tcl error sourcing /home/alan/src/binutils-gdb/ld/testsuite/ld-i386/i386.exp.
i386-go32  +ERROR: couldn't open "tmpdir/go32stub": no such file or directory
i386-go32  +FAIL: ld-scripts/sane1
i386-go32  +FAIL: ld-scripts/assign-loc
i386-go32  +FAIL: ld-scripts/pr18963

This does result in some warnings in gdb which are fixed in a followup
patch.

bfd/
* bfdio.c (bfd_read, bfd_write): Add ATTRIBUTE_WARN_UNUSED_RESULT.
(bfd_tell, bfd_stat, bfd_seek, bfd_mmap): Likewise.
* bfd-in2.h: Regenerate.
* coff-rs6000.c (xcoff_write_armap_big) Don't ignore bfd_write
return value.
(xcoff_generate_rtinit): Likewise.  Also free data_buffer and
string_table before returning.
* coff64-rs6000.c (xcoff64_generate_rtinit): Likewise.
* coff-stgo32.c (go32exe_check_format): Don't ignore bfd_seek
return value.
* coffcode.h (coff_apply_checksum): Don't ignore bfd_write return.
(coff_write_object_contents <COFF_GO32>): Likewise, and bfd_malloc.
Fix bitrotted code to look for first section with non-zero filepos.
* elf64-ia64-vms.c (elf64_vms_write_shdrs_and_ehdr): Don't ignore
bfd_seek or bfd_write return values.
* pef.c (bfd_pef_scan_section): Likewise.
(bfd_pef_read_header, bfd_pef_xlib_read_header): Likewise.
* vms-misc.c (_bfd_vms_output_end): Likewise.  Return status.
* vms.h (_bfd_vms_output_end): Update prototype.
* vms-alpha.c: Pass _bfd_vms_output_end status up call chains.
* wasm-module.c (wasm_compute_custom_section_file_position): Don't
ignore bfd_seek or bfd_write return values.
(wasm_compute_section_file_positions): Likewise.
* xsym.c (bfd_sym_scan): Don't ignore bfd_seek return value.
(bfd_sym_read_name_table): Likewise.
binutils/
* ar.c (print_contents, extract_file): Don't ignore bfd_seek
return value.
ld/
* pdb.c (create_section_contrib_substream): Don't ignore bfd_seek
return value.
(create_section_header_stream): Likewise.
* pe-dll.c (pe_get16, pe_get32): Add fail param to return results
from bfd_seek and bfd_read.
(pe_implied_import_dll): Handle these fails, and other bfd_seek
and bfd_read return values.

15 months agoRISC-V: Fix opcode entries of "vmsge{,u}.vx"
Tsukasa OI [Sun, 6 Aug 2023 01:37:05 +0000 (01:37 +0000)]
RISC-V: Fix opcode entries of "vmsge{,u}.vx"

Their check_func should be "match_never", not "match_opcode".  The reasons
this error did not cause any disassembler errors are:

1.  The problem will not reproduce if "no-aliases" is specified
    (because macro instructions are handled as aliases).
2.  If not, all affected compressed instructions or their aliases
    precede before "vmsge{,u}.vx" macro instructions.

However, it'll easily break if we reorder opcode entries.  This commit
fixes this issue before the *accident* occurs.

opcodes/ChangeLog:

* riscv-opc.c (riscv_opcodes): Make sure that we never match to
vmsge{,u}.vx instructions unless specified in the assembler.

15 months agoRISC-V: Remove support for non-existing 'Zve32d'
Tsukasa OI [Thu, 3 Aug 2023 01:01:23 +0000 (01:01 +0000)]
RISC-V: Remove support for non-existing 'Zve32d'

Since this "extension" does not exist (on the other hand, 'Zve64d' exists)
and it's not useful if we keep it (as other code portions just ignore
"zve32d"), this commit just removes it.

bfd/ChangeLog:

* elfxx-riscv.c (riscv_supported_std_z_ext): Remove 'Zve32d'
extension from the list.

15 months agoAutomatic date update in version.in
GDB Administrator [Fri, 11 Aug 2023 00:00:24 +0000 (00:00 +0000)]
Automatic date update in version.in

15 months ago[gdb/symtab] Fix off-by-one error in cooked_indexer::recurse
Tom de Vries [Thu, 10 Aug 2023 19:16:30 +0000 (21:16 +0200)]
[gdb/symtab] Fix off-by-one error in cooked_indexer::recurse

Test-case gdb.dwarf2/pr13961.exp contains:
...
 <1><25>: Abbrev Number: 8 (DW_TAG_class_type)
    <26>   DW_AT_specification: <0x2a>
 <1><2a>: Abbrev Number: 2 (DW_TAG_class_type)
    <2b>   DW_AT_name        : foo
    <2f>   DW_AT_byte_size   : 4
    <30>   DW_AT_decl_file   : 1
    <31>   DW_AT_decl_line   : 1
    <32>   DW_AT_sibling     : <0x44>
...

The DIE at 0x25 contains an intra-CU forward reference, and is deferred during
DIE indexing in the cooked_index, by adding it to m_deferred_entries.

The resulting cooked index entries are:
...
        [25] ((cooked_index_entry *) 0x333b5d0)
        name:       foo
        canonical:  foo
        qualified:  foo
        DWARF tag:  DW_TAG_class_type
        flags:      0x0 []
        DIE offset: 0x2a
        parent:     ((cooked_index_entry *) 0)

        [26] ((cooked_index_entry *) 0x333b630)
        name:       foo
        canonical:  foo
        qualified:  foo::foo
        DWARF tag:  DW_TAG_class_type
        flags:      0x0 []
        DIE offset: 0x25
        parent:     ((cooked_index_entry *) 0x333b5d0) [foo]
...

Notice that 0x2a is the parent of 0x25, and that this is why the qualified
name of 0x25 is "foo::foo", which is incorrect, it's supposed to be "foo".

The parent is set here in cooked_indexer::make_index:
...
  for (const auto &entry : m_deferred_entries)
    {
      void *obj = m_die_range_map.find (entry.spec_offset);
      cooked_index_entry *parent = static_cast<cooked_index_entry *> (obj);
      m_index_storage->add (entry.die_offset, entry.tag, entry.flags,
    entry.name, parent, m_per_cu);
    }
...
and AFAICT, we store in m_die_range_map the parent of the respective
spec_offset DIE (though that's not clear from the comment describing it).

So, the root cause of this is that when we lookup the parent for DIE 0x25, we get
m_die_range_map.find (0x2a) == 0x2a.

This is an off-by-one error, fixed in cooked_indexer::recurse by:
...
-      CORE_ADDR start = form_addr (parent_entry->die_offset,
+      CORE_ADDR start = form_addr (parent_entry->die_offset + 1,
...
which gives us:
...
    [12] ((cooked_index_entry *) 0x41e21f0)
    name:       foo
    canonical:  foo
    qualified:  foo
    DWARF tag:  DW_TAG_class_type
    flags:      0x0 []
    DIE offset: 0x25
    parent:     ((cooked_index_entry *) 0)

    [13] ((cooked_index_entry *) 0x41e2190)
    name:       foo
    canonical:  foo
    qualified:  foo
    DWARF tag:  DW_TAG_class_type
    flags:      0x0 []
    DIE offset: 0x2a
    parent:     ((cooked_index_entry *) 0)
...

Tested on x86_64-linux.

Approved-By: Tom Tromey <tom@tromey.com>
PR symtab/30739
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30739

15 months ago[gdb/symtab] Dump qualified name of cooked_index_entry
Tom de Vries [Thu, 10 Aug 2023 19:16:30 +0000 (21:16 +0200)]
[gdb/symtab] Dump qualified name of cooked_index_entry

When doing "maint print objfiles" for the exec of test-case
gdb.dwarf2/pr13961.exp, we get:
...
    [25] ((cooked_index_entry *) 0x37b25d0)
    name:       foo
    canonical:  foo
    DWARF tag:  DW_TAG_class_type
    flags:      0x0 []
    DIE offset: 0x2a
    parent:     ((cooked_index_entry *) 0)

    [26] ((cooked_index_entry *) 0x37b2630)
    name:       foo
    canonical:  foo
    DWARF tag:  DW_TAG_class_type
    flags:      0x0 []
    DIE offset: 0x25
    parent:     ((cooked_index_entry *) 0x37b25d0) [foo]
...

By following the parent links in the text, we can conclude that the qualified
name of DIE 0x25 is foo::foo (which is incorrect, that's PR symtab/30739).

But it's not evident, and also hard to verify in a test-case.

Add dumping of the qualified name, such that we have:
...
    [25] ((cooked_index_entry *) 0x333b5d0)
    name:       foo
    canonical:  foo
    qualified:  foo
    DWARF tag:  DW_TAG_class_type
    flags:      0x0 []
    DIE offset: 0x2a
    parent:     ((cooked_index_entry *) 0)

    [26] ((cooked_index_entry *) 0x333b630)
    name:       foo
    canonical:  foo
    qualified:  foo::foo
    DWARF tag:  DW_TAG_class_type
    flags:      0x0 []
    DIE offset: 0x25
    parent:     ((cooked_index_entry *) 0x333b5d0) [foo]
...

Tested on x86_64-linux.

Approved-By: Tom Tromey <tom@tromey.com>
15 months agoFix gdb.ada/O2_float_param.exp for PowerPC
Carl Love [Thu, 3 Aug 2023 17:55:51 +0000 (13:55 -0400)]
Fix gdb.ada/O2_float_param.exp for PowerPC

The frame command on Power pc prints the address in hex between the
#0 and in calle.increment.  For example

(gdb) frame
#0  0x0000000010010a88 in callee.increment (val=val@entry=99.0, msg=...)
    at /home/.../gdb/testsuite/gdb.ada/O2_float_param/callee.adb:19
19    procedure Increment (Val : in out Float; Msg: String) is

The printing of the address for the frame is done by function
print_frame in gdb/stack.c.  If SAL.IS_stmt is false for the frame,
function frame_show_address returns true and print_frame prints the
address.  Currently, SAL.IS is false on PowerPC and true on X86-64.

Update the set re string to accept the hex address if it exits.

Fixes two failures on PowerPC.

Patch tested on Power10 with no new regressions.

Approved-By: Tom Tromey <tom@tromey.com>
15 months agoChange py-thread-exited.exp to work with gdbserver
Tom Tromey [Mon, 31 Jul 2023 17:30:30 +0000 (11:30 -0600)]
Change py-thread-exited.exp to work with gdbserver

gdbserver does not notify gdb of new threads when they are created.
I'm not sure if this is documented anywhere, but it is mentioned on
this page:

https://sourceware.org/gdb/wiki/LocalRemoteFeatureParity

Search for "Finding new threads in the inferior".

This behavior is a bit unfortunate -- I would think that it would be
better to arrange for such notification if something on the gdb side
is interested.

Meanwhile, this patch fixes py-thread-exited.exp to work around this
problem.

Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30677

15 months agoPass unique_ptr to add_thread_with_info
Tom Tromey [Thu, 10 Mar 2022 23:56:57 +0000 (16:56 -0700)]
Pass unique_ptr to add_thread_with_info

This changes add_thread_with_info to accept a unique_ptr, making it
clear that it takes ownership of the passed-in pointer.

I can't test the AIX or Darwin changes, but I think they are
relatively obvious.

15 months agoaarch64: Enable Cortex-A520 CPU
Richard Ball [Thu, 10 Aug 2023 14:11:37 +0000 (15:11 +0100)]
aarch64: Enable Cortex-A520 CPU

This patch adds support for the Cortex-A520 CPU to gas.

No regressions on aarch64-none-elf.

gas/ChangeLog:

* NEWS: Update docs.
* config/tc-aarch64.c: Add Cortex-A520.
* doc/c-aarch64.texi: Update docs.

15 months ago[gdb/testsuite] Fix gdb.dwarf2/enqueued-cu-base-addr.exp with cc-with-gnu-debuglink
Tom de Vries [Thu, 10 Aug 2023 13:46:30 +0000 (15:46 +0200)]
[gdb/testsuite] Fix gdb.dwarf2/enqueued-cu-base-addr.exp with cc-with-gnu-debuglink

When running test-case gdb.dwarf2/enqueued-cu-base-addr.exp with target board
cc-with-gnu-debuglink, I run into:
...
(gdb) PASS: gdb.dwarf2/enqueued-cu-base-addr.exp: ptype foo
maint print symbols -objfile enqueued-cu-base-addr^M
(gdb) FAIL: gdb.dwarf2/enqueued-cu-base-addr.exp: CU addr found
...

The problem is that the CU we're trying to print is in objfile
enqueued-cu-base-addr.debug instead of enqueued-cu-base-addr.

Fix this by replacing "-objfile enqueued-cu-base-addr" with "-source cu2".

Tested on x86_64-linux.

15 months ago[gdb/testsuite] Improve failure mode in gdb.dwarf2/enqueued-cu-base-addr.exp
Tom de Vries [Thu, 10 Aug 2023 13:46:30 +0000 (15:46 +0200)]
[gdb/testsuite] Improve failure mode in gdb.dwarf2/enqueued-cu-base-addr.exp

I ran test-case gdb.dwarf2/enqueued-cu-base-addr.exp with target board
cc-with-debug-names, and ran into:
...
FAIL: gdb.dwarf2/enqueued-cu-base-addr.exp: ptype foo (GDB internal error)
FAIL: gdb.dwarf2/enqueued-cu-base-addr.exp: CU addr found
...

The first FAIL is a known issue, PR symtab/29572.

The following FAIL is a consequence of the first FAIL, so require for the
second test that the first test passes.

Tested on x86_64-linux, with target boards unix and cc-with-debug-names.

15 months ago[gdb/symtab] Fix assertion in write_debug_names
Tom de Vries [Thu, 10 Aug 2023 13:46:30 +0000 (15:46 +0200)]
[gdb/symtab] Fix assertion in write_debug_names

When running test-case gdb.dwarf2/pr13961.exp with target-board
cc-with-debug-names, I run into:
...
Running gdb.dwarf2/pr13961.exp ...
gdb compile failed, gdb/dwarf2/index-write.c:1305: internal-error: \
  write_debug_names: Assertion `counter == per_bfd->all_units.size ()' failed.
...

This is a regression since commit 542a33e348a ("Only use the per-BFD object to
 write a DWARF index"), which did:
...
-  gdb_assert (counter == per_objfile->per_bfd->all_comp_units.size ());
+  gdb_assert (counter == per_bfd->all_units.size ());
...

Fix this by reverting to using all_comp_units:
...
  gdb_assert (counter == per_bfd->all_comp_units.size ());
...

Tested on x86_64-linux, using target boards unix and cc-with-debug-names.

PR symtab/30741
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30741

15 months agoAutomatic date update in version.in
GDB Administrator [Thu, 10 Aug 2023 00:00:39 +0000 (00:00 +0000)]
Automatic date update in version.in

15 months agobpf: use w regs in 32-bit non-fetch atomic pseudo-c
David Faust [Wed, 9 Aug 2023 18:44:38 +0000 (11:44 -0700)]
bpf: use w regs in 32-bit non-fetch atomic pseudo-c

The 32-bit non-fetching atomic instructions treat the source register as
32-bits, which means in the pseudo-c syntax the "w" registers should be
used rather than the "r" registers.

opcodes/

* bpf-opc-c (bpf_opcodes): Use %sw for AAD32, AOR32, AAND32
and AXOR32 pseudo-c dialect asm templates.

gas/

* testsuite/gas/bpf/atomic-be-pseudoc.d: Use "w" for source reg
in non-fetching 32-bit atomic instructions.
* testsuite/gas/bpf/atomic-pseudoc.d: Likewise.
* testsuite/gas/bpf/atomic-pseudoc.s: Likewise.

15 months agogdb, breakpoint: add breakpoint location debugging logs
Mihails Strasuns [Tue, 3 Dec 2019 12:13:25 +0000 (13:13 +0100)]
gdb, breakpoint: add breakpoint location debugging logs

Add new commands:

  set debug breakpoint on|off
  show debug breakpoint

This patch introduces new debugging information that prints
breakpoint location insertion and removal flow.

The debug output looks like:
~~~
(gdb) set debug breakpoint on
(gdb) disassemble main
Dump of assembler code for function main:
   0x0000555555555129 <+0>: endbr64
   0x000055555555512d <+4>: push   %rbp
   0x000055555555512e <+5>: mov    %rsp,%rbp
=> 0x0000555555555131 <+8>: mov    $0x0,%eax
   0x0000555555555136 <+13>: pop    %rbp
   0x0000555555555137 <+14>: ret
End of assembler dump.
(gdb) break *0x0000555555555137
Breakpoint 2 at 0x555555555137: file main.c, line 4.
[breakpoint] update_global_location_list: insert_mode = UGLL_MAY_INSERT
(gdb) c
Continuing.
[breakpoint] update_global_location_list: insert_mode = UGLL_INSERT
[breakpoint] insert_bp_location: Breakpoint 2 (0x5565daddb1e0) at address 0x555555555137 in main at main.c:4
[breakpoint] insert_bp_location: Breakpoint -2 (0x5565dab51c10) at address 0x7ffff7fd37b5
[breakpoint] insert_bp_location: Breakpoint -5 (0x5565dab68f30) at address 0x7ffff7fe509e
[breakpoint] insert_bp_location: Breakpoint -7 (0x5565dab694f0) at address 0x7ffff7fe63f4
[breakpoint] remove_breakpoint_1: Breakpoint 2 (0x5565daddb1e0) at address 0x555555555137 in main at main.c:4 due to regular remove
[breakpoint] remove_breakpoint_1: Breakpoint -2 (0x5565dab51c10) at address 0x7ffff7fd37b5 due to regular remove
[breakpoint] remove_breakpoint_1: Breakpoint -5 (0x5565dab68f30) at address 0x7ffff7fe509e due to regular remove
[breakpoint] remove_breakpoint_1: Breakpoint -7 (0x5565dab694f0) at address 0x7ffff7fe63f4 due to regular remove

Breakpoint 2, 0x0000555555555137 in main () at main.c:4
4 }
~~~

Co-Authored-By: Christina Schimpe <christina.schimpe@intel.com>
15 months agoAutomatic date update in version.in
GDB Administrator [Wed, 9 Aug 2023 00:00:37 +0000 (00:00 +0000)]
Automatic date update in version.in

15 months agoRename bfd_bread and bfd_bwrite
Alan Modra [Mon, 7 Aug 2023 05:10:35 +0000 (14:40 +0930)]
Rename bfd_bread and bfd_bwrite

These were renamed from bfd_read and bfd_write back in 2001 when they
lost an unnecessary parameter.  Rename them back, and get rid of a few
casts that are only needed without prototyped functions (K&R C).

15 months agoAdd ld makefile dependencies
Alan Modra [Mon, 7 Aug 2023 05:07:41 +0000 (14:37 +0930)]
Add ld makefile dependencies

* Makefile.am (EXTRA_ld_new_SOURCES): Add pep-dll-aarch64.c
and pep-dll-x86_64.c.
* Makefile.in: Regenerate.

15 months agoPR30724, cygwin ld performance regression since 014a602b86
Alan Modra [Sun, 6 Aug 2023 22:58:55 +0000 (08:28 +0930)]
PR30724, cygwin ld performance regression since 014a602b86

According to the reporter of this bug the newlib fseek implementation
is likely slowed down by locking and fflush, only attempting to
optimise seeks when the file is opened read-only.  Thus when writing
the output we get a dramatic slowdown due to commit 014a602b86.

PR 30724
* bfd.c (enum bfd_last_io): New.
(struct bfd): Add last_io field.
* bfd-in2.h: Regenerate.
* bfd-io.c (bfd_bread, bfd_bwrite): Force seek if last_io is
opposite direction.
(bfd_seek): Reinstate optimisation for seek to same position.

15 months ago[gdb/build] Use move capture in gdb_demangle
Tom de Vries [Tue, 8 Aug 2023 22:58:29 +0000 (00:58 +0200)]
[gdb/build] Use move capture in gdb_demangle

Use move capture in gdb_demangle when compiling for c++14 or higher, to save a
std::string copy.

Tested on x86_64-linux.

Reported-By: Tom Tromey <tom@tromey.com>
Approved-By: Tom Tromey <tom@tromey.com>
15 months agold: Fix retain7a.d XFAIL/notarget entry for hppa
Sam James [Tue, 8 Aug 2023 01:19:57 +0000 (02:19 +0100)]
ld: Fix retain7a.d XFAIL/notarget entry for hppa

PR 30733
* ld/testsuite/ld-elf/retain7a.d: Fix XFAIL entry for hppa to match
  hppa{1.1,2.0}*, like hppa2.0-unknown-linux-gnu which Gentoo uses.

Signed-off-by: Sam James <sam@gentoo.org>
15 months agold: Fix relocatable.d XFAIL/notarget entry for hppa
Sam James [Tue, 8 Aug 2023 01:17:57 +0000 (02:17 +0100)]
ld: Fix relocatable.d XFAIL/notarget entry for hppa

PR 30734
* ld/testsuite/ld-elf/relocatable.d: Fix notarget entry for hppa to match
  hppa{1.1,2.0}*, like hppa2.0-unknown-linux-gnu which Gentoo uses.

Signed-off-by: Sam James <sam@gentoo.org>
15 months agoUpdate my name in maintainers file
Guinevere Larsen [Tue, 8 Aug 2023 08:47:32 +0000 (10:47 +0200)]
Update my name in maintainers file

15 months agoGuard against killing unrelated processes in amd64-disp-step.exp
Rainer Orth [Tue, 8 Aug 2023 08:13:58 +0000 (10:13 +0200)]
Guard against killing unrelated processes in amd64-disp-step.exp

When testing current gdb trunk on Solaris/amd64, the whole session was
reliably terminated by make check.  I could trace this to the following
entry in gdb.arch/amd64-disp-step/gdb.log:

FAIL: gdb.arch/amd64-disp-step.exp: add into rcx: send_signal=on: get
inferior pid
Executing on target: kill -ALRM -1    (timeout = 300)
builtin_spawn -ignore SIGHUP kill -ALRM -1

If $inferior_pid doesn't refer to a single process for some reason, this
kill would terminate either a process group or the whole session.

This patch avoids this by ensuring that the pid arg is positive.

Tested on amd64-pc-solaris2.11 and x86_64-pc-linux-gnu.

Co-Authored-By: Andrew Burgess <aburgess@redhat.com>
Approved-By: Tom Tromey <tom@tromey.com>
15 months ago[gdb/build] Fix build breaker with -std=c++11
Tom de Vries [Tue, 8 Aug 2023 08:08:04 +0000 (10:08 +0200)]
[gdb/build] Fix build breaker with -std=c++11

When building with -std=c++11 I run into:
...
gdb/dwarf2/cooked-index.c: In member function \
  ‘void cooked_index::start_writing_index(dwarf2_per_bfd*)’:
gdb/dwarf2/cooked-index.c:469:10: error: lambda capture initializers only \
  available with -std=c++14 or -std=gnu++14 [-Werror]
          ctx = std::move (ctx)] ()
          ^~~
...

Fix this by capturing a copy instead when using -std=c++11:
...
    = gdb::thread_pool::g_thread_pool->post_task ([this, per_bfd, ctx] ()
...

Tested by building with and without -stdc=++11 on x86_64-linux.

Reported-By: Tom Tromey <tom@tromey.com>
Reviewed-by: John Baldwin <jhb@FreeBSD.org>
15 months agoRISC-V: Update ratified 'Ztso' extension version
Tsukasa OI [Tue, 8 Aug 2023 04:30:41 +0000 (04:30 +0000)]
RISC-V: Update ratified 'Ztso' extension version

Because the 'Ztso' extension is now ratified, it has a version number of 1.0
(not 0.1).  This commit updates the number.

bfd/ChangeLog:

* elfxx-riscv.c (riscv_supported_std_z_ext): Update the version
number of the 'Ztso' extension since it's ratified.

15 months agoAutomatic date update in version.in
GDB Administrator [Tue, 8 Aug 2023 00:00:41 +0000 (00:00 +0000)]
Automatic date update in version.in

15 months agogprofng: 30700 tmpdir/gp-collect-app_F test fails
Vladimir Mezentsev [Thu, 3 Aug 2023 19:56:54 +0000 (12:56 -0700)]
gprofng: 30700 tmpdir/gp-collect-app_F test fails

gprofng/ChangeLog
2023-08-03  Vladimir Mezentsev  <vladimir.mezentsev@oracle.com>

PR gprofng/30700
* testsuite/gprofng.display/gp-collect-app_F.exp: Fix -name argument
for sub-experiment filtering.

15 months agoRISC-V: move comment describing rules for riscv_opcodes[]
Jan Beulich [Mon, 7 Aug 2023 09:56:50 +0000 (11:56 +0200)]
RISC-V: move comment describing rules for riscv_opcodes[]

It makes little sense to have this comment meanwhile over a hundred
lines ahead of the array. In fact until spotting the comment, I was
wondering why those pretty important aspects aren't spelled out
anywhere.

15 months agogdb/fortran: Align intrinsic/variable precedence
Richard Bunt [Mon, 7 Aug 2023 08:12:53 +0000 (09:12 +0100)]
gdb/fortran: Align intrinsic/variable precedence

Fortran allows variables and function to be named after language defined
intrinsics as they are not reserved keywords. For example, the abs maths
intrinsic can be hidden by a user declaring a variable called abs.

The behavior before this patch was to favour the intrinsic, which meant
that any variables named, for example "allocated", could not be
inspected by GDB.

This patch inverts this priority to bring GDB's behaviour closer to the
Fortran language, where the user defined symbol can hide the intrinsic.

Special care was need to prevent any C symbols from overriding either
Fortran intrinsics or user defined variables. This was observed to be
the case when GDB has access to symbols for abs from libm. This was
solved by only allowing symbols not marked with language_fortran to be
overridden.

In total this brings the order of precedence to the following (highest
first):

    1. User defined Fortran variable or function.
    2. Fortran intrinsic.
    3. Symbols from languages other than Fortran.

The sizeof intrinsic is now case insensitive. This is closer to the
Fortran language.  I believe this change is safe enough as it increases
the acceptance of the grammar, rather than restricts it. I.e. it should
not break any existing scripts which rely on it. Unless of course they
rely on SIZEOF being rejected.

GDB built with GCC 13.

No test suite regressions detected. Compilers: GCC, ACfL, Intel, Intel
LLVM, NVHPC; Platforms: x86_64, aarch64.

Existing tests in gdb.fortran cover the invocation of intrinsics
including: intrinsics.exp, shape.exp, rank.exp, lbound-ubound.exp.

Approved-By: Tom Tromey <tom@tromey.com>
15 months agoAutomatic date update in version.in
GDB Administrator [Mon, 7 Aug 2023 00:00:25 +0000 (00:00 +0000)]
Automatic date update in version.in

15 months agoAutomatic date update in version.in
GDB Administrator [Sun, 6 Aug 2023 00:00:29 +0000 (00:00 +0000)]
Automatic date update in version.in

15 months ago[gdb/symtab] Find main language without symtab expansion
Tom de Vries [Sat, 5 Aug 2023 15:57:13 +0000 (17:57 +0200)]
[gdb/symtab] Find main language without symtab expansion

When loading an executable using "file a.out", the language is set according
to a.out, which can involve looking up the language of symbol "main", which
will cause the symtab expansion for the containing CU.

Expansion of lto debug info can be slow, so in commit d3214198119 ("[gdb] Use
partial symbol table to find language for main") a feature was added to avoid
the symtab expansion.

This feature stopped working after commit 7f4307436fd ("Fix "start" for D,
Rust, etc").

[ The commit addresses problems related to command start, which requires finding
the main function:
- for language D, "main" was found instead of "D main", and
- for Rust, the correct function was found, but attributed the wrong name
  (not fully qualified). ]

Reimplement the feature by adding
cooked_index_functions::lookup_global_symbol_language.

Tested on x86_64-linux.

PR symtab/30661
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30661

15 months agoas: Fix typo in manual
David Carew [Fri, 4 Aug 2023 16:18:06 +0000 (09:18 -0700)]
as: Fix typo in manual

The -D flag should enable "debugging"

15 months agoAutomatic date update in version.in
GDB Administrator [Sat, 5 Aug 2023 00:00:30 +0000 (00:00 +0000)]
Automatic date update in version.in

15 months agoReindent recursive_dump_type
Tom Tromey [Fri, 4 Aug 2023 19:57:55 +0000 (13:57 -0600)]
Reindent recursive_dump_type

I noticed that a 'switch' in recursive_dump_type was incorrect
indented.  This patch fixes the problem.  Tested by rebuilding.

15 months agoConsolidate calls to bfd_set_cacheable
Tom Tromey [Fri, 4 Aug 2023 11:58:35 +0000 (05:58 -0600)]
Consolidate calls to bfd_set_cacheable

I noticed that some spots in gdb call bfd_set_cacheable after opening
a BFD.

The BFD file cache is a bit odd.  BFDs that are opened locally are
unconditionally registered with the cache, and their underlying file
descriptor will always be closed when bfd_cache_close_all is called.
However, only "cacheable" BFDs will be eligible for reopening when
needed -- and by default BFD decides that if a file descriptor is
passed in, then it should not be cacheable.  If a non-cacheable BFD's
file descriptor is closed, there is no offical way to reopen it.

gdb needs to call bfd_cache_close_all, because some systems cannot
start an executable when it has an open file descriptor referencing
it.

However, gdb also will sometimes passes an open file descriptor to the
various BFD open functions.  And, due to lazy DWARF reading, gdb may
also need to reopen these BFDs.

Rather than having all the callers figure out when exactly to set the
cacheable flag, I think it makes sense to consolidate this logic into
the gdb_bfd.c wrapper functions.  It is ok to do this because gdb
always passes a filename to these open functions, so reopening should
work ok.

Regression tested on x86-64 Fedora 38.

Reviewed-by: John Baldwin <jhb@FreeBSD.org>
15 months agoRemove extra '.' from error message
Tom Tromey [Fri, 4 Aug 2023 13:20:13 +0000 (07:20 -0600)]
Remove extra '.' from error message

A local gdb test failed with this error message:

 Remote communication error.  Target disconnected.: Arg list too long.

The ".:" seemed weird to me.  This patch removes the ".".

Reviewed-by: John Baldwin <jhb@FreeBSD.org>
15 months agoFix incorrect class name in free_objfile documentation
Tom Tromey [Fri, 4 Aug 2023 15:51:54 +0000 (09:51 -0600)]
Fix incorrect class name in free_objfile documentation

The documentation for the Python free_objfile event registry uses the
wrong class name.  This patch fixes it.  I'm checking this in as
obvious.

15 months agoFix potential infinite loop in bfd_cache_close_all.
Nick Clifton [Fri, 4 Aug 2023 13:19:28 +0000 (14:19 +0100)]
Fix potential infinite loop in bfd_cache_close_all.

  PR 15545 * cache.c (bfd_cache_close_all): Extend description to note that all files will be closed, even those that are not cacheable. Add code to prevent a possible infinite loop.

15 months ago[gdb/testsuite] Extend gdb.base/index-cache.exp further
Tom de Vries [Fri, 4 Aug 2023 13:05:57 +0000 (15:05 +0200)]
[gdb/testsuite] Extend gdb.base/index-cache.exp further

Add lookup of a non-existing symbol to test-case gdb.base/index-cache.exp.

This serves as regression test for PR symtab/30718.

PR symtab/30718
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30718

15 months ago[gdb/testsuite] Move "maint wait-for-index-cache" ALAP in gdb.base/index-cache.exp
Tom de Vries [Fri, 4 Aug 2023 13:05:57 +0000 (15:05 +0200)]
[gdb/testsuite] Move "maint wait-for-index-cache" ALAP in gdb.base/index-cache.exp

In test-case gdb.base/index-cache.exp proc run_test_with_flags contains:
...
clean_restart ${testfile}

# The tests generally want to check the cache, so make sure it
# has completed its work.
gdb_test_no_output "maintenance wait-for-index-cache"
...

This however hides data races between:
- index-cache writing (due to file $exec), and
- symbol lookups (due to subsequent ptype commands).

Fix this by:
- moving the "maintenance wait-for-index-cache" to proc check_cache_stats, and
- moving all calls to proc check_cache_stats ALAP.

Tested on x86_64-linux.

15 months ago[gdb/symtab] Fix data race on dwarf2_per_cu_data::{files_read,is_debug_types}
Tom de Vries [Fri, 4 Aug 2023 13:05:57 +0000 (15:05 +0200)]
[gdb/symtab] Fix data race on dwarf2_per_cu_data::{files_read,is_debug_types}

With gdb build with -fsanitize=thread, and the exec from test-case
gdb.base/index-cache.exp, I run into:
...
$ rm -f ~/.cache/gdb/*; \
  gdb -q -batch -iex "set index-cache enabled on" index-cache \
    -ex "print foobar"
  ...
WARNING: ThreadSanitizer: data race (pid=25018)
  Write of size 1 at 0x7b200000410d by main thread:
    #0 dw2_get_file_names_reader gdb/dwarf2/read.c:2033 (gdb+0x7ab023)
    #1 dw2_get_file_names gdb/dwarf2/read.c:2130 (gdb+0x7ab023)
    #2 dw_expand_symtabs_matching_file_matcher(dwarf2_per_objfile*, gdb::function_view<bool (char const*, bool)>) gdb/dwarf2/read.c:3105 (gdb+0x7ac6e9)
    #3 cooked_index_functions::expand_symtabs_matching(objfile*, gdb::function_view<bool (char const*, bool)>, lookup_name_info const*, gdb::function_view<bool (char const*)>, gdb::function_view<bool (compunit_symtab*)>, enum_flags<block_search_flag_values>, domain_enum, search_domain) gdb/dwarf2/read.c:16812 (gdb+0x7d040f)
    #4 objfile::map_symtabs_matching_filename(char const*, char const*, gdb::function_view<bool (symtab*)>) gdb/symfile-debug.c:219 (gdb+0xda5b6e)
    #5 iterate_over_symtabs(char const*, gdb::function_view<bool (symtab*)>) gdb/symtab.c:648 (gdb+0xdc441d)
    #6 lookup_symtab(char const*) gdb/symtab.c:662 (gdb+0xdc4522)
    #7 classify_name gdb/c-exp.y:3083 (gdb+0x61afec)
    #8 c_yylex gdb/c-exp.y:3251 (gdb+0x61dd13)
    #9 c_yyparse() build/gdb/c-exp.c.tmp:1988 (gdb+0x61f07e)
    #10 c_parse(parser_state*) gdb/c-exp.y:3417 (gdb+0x62d864)
    #11 language_defn::parser(parser_state*) const gdb/language.c:598 (gdb+0x977245)
    #12 parse_exp_in_context gdb/parse.c:414 (gdb+0xb10b1b)
    #13 parse_expression(char const*, innermost_block_tracker*, enum_flags<parser_flag>) gdb/parse.c:462 (gdb+0xb1112e)
    #14 process_print_command_args gdb/printcmd.c:1321 (gdb+0xb4bf8c)
    #15 print_command_1 gdb/printcmd.c:1335 (gdb+0xb4caaa)
    #16 print_command gdb/printcmd.c:1468 (gdb+0xb4cdda)
    #17 do_simple_func gdb/cli/cli-decode.c:95 (gdb+0x65b078)
    #18 cmd_func(cmd_list_element*, char const*, int) gdb/cli/cli-decode.c:2735 (gdb+0x65ed53)
    #19 execute_command(char const*, int) gdb/top.c:575 (gdb+0xe3a7ea)
    #20 catch_command_errors gdb/main.c:518 (gdb+0xa183fd)
    #21 execute_cmdargs gdb/main.c:617 (gdb+0xa185bf)
    #22 captured_main_1 gdb/main.c:1289 (gdb+0xa1aad8)
    #23 captured_main gdb/main.c:1310 (gdb+0xa1b9da)
    #24 gdb_main(captured_main_args*) gdb/main.c:1339 (gdb+0xa1b9da)
    #25 main gdb/gdb.c:39 (gdb+0x42506a)

  Previous read of size 1 at 0x7b200000410d by thread T2:
    #0 write_gdbindex gdb/dwarf2/index-write.c:1214 (gdb+0x75bb30)
    #1 write_dwarf_index(dwarf2_per_bfd*, char const*, char const*, char const*, dw_index_kind) gdb/dwarf2/index-write.c:1469 (gdb+0x75f803)
    #2 index_cache::store(dwarf2_per_bfd*, index_cache_store_context const&) gdb/dwarf2/index-cache.c:173 (gdb+0x755a36)
    #3 cooked_index::maybe_write_index(dwarf2_per_bfd*, index_cache_store_context const&) gdb/dwarf2/cooked-index.c:642 (gdb+0x71c96d)
    #4 operator() gdb/dwarf2/cooked-index.c:471 (gdb+0x71c96d)
    #5 _M_invoke /usr/include/c++/7/bits/std_function.h:316 (gdb+0x71c96d)
    #6 std::function<void ()>::operator()() const /usr/include/c++/7/bits/std_function.h:706 (gdb+0x72a57c)
    #7 void std::__invoke_impl<void, std::function<void ()>&>(std::__invoke_other, std::function<void ()>&) /usr/include/c++/7/bits/invoke.h:60 (gdb+0x72a5db)
    #8 std::__invoke_result<std::function<void ()>&>::type std::__invoke<std::function<void ()>&>(std::function<void ()>&) /usr/include/c++/7/bits/invoke.h:95 (gdb+0x72a5db)
    #9 std::__future_base::_Task_state<std::function<void ()>, std::allocator<int>, void ()>::_M_run()::{lambda()#1}::operator()() const /usr/include/c++/7/future:1421 (gdb+0x72a5db)
    #10 std::__future_base::_Task_setter<std::unique_ptr<std::__future_base::_Result<void>, std::__future_base::_Result_base::_Deleter>, std::__future_base::_Task_state<std::function<void ()>, std::allocator<int>, void ()>::_M_run()::{lambda()#1}, void>::operator()() const /usr/include/c++/7/future:1362 (gdb+0x72a5db)
    #11 std::_Function_handler<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> (), std::__future_base::_Task_setter<std::unique_ptr<std::__future_base::_Result<void>, std::__future_base::_Result_base::_Deleter>, std::__future_base::_Task_state<std::function<void ()>, std::allocator<int>, void ()>::_M_run()::{lambda()#1}, void> >::_M_invoke(std::_Any_data const&) /usr/include/c++/7/bits/std_function.h:302 (gdb+0x72a5db)
    #12 std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>::operator()() const /usr/include/c++/7/bits/std_function.h:706 (gdb+0x724954)
    #13 std::__future_base::_State_baseV2::_M_do_set(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*) /usr/include/c++/7/future:561 (gdb+0x724954)
    #14 void std::__invoke_impl<void, void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(std::__invoke_memfun_deref, void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&) /usr/include/c++/7/bits/invoke.h:73 (gdb+0x72434a)
    #15 std::__invoke_result<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>::type std::__invoke<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&) /usr/include/c++/7/bits/invoke.h:95 (gdb+0x72434a)
    #16 std::call_once<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(std::once_flag&, void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&)::{lambda()#1}::operator()() const /usr/include/c++/7/mutex:672 (gdb+0x72434a)
    #17 std::call_once<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(std::once_flag&, void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&)::{lambda()#2}::operator()() const /usr/include/c++/7/mutex:677 (gdb+0x72434a)
    #18 std::call_once<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(std::once_flag&, void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&)::{lambda()#2}::_FUN() /usr/include/c++/7/mutex:677 (gdb+0x72434a)
    #19 pthread_once <null> (libtsan.so.0+0x4457c)
    #20 __gthread_once /usr/include/c++/7/x86_64-suse-linux/bits/gthr-default.h:699 (gdb+0x72532b)
    #21 void std::call_once<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(std::once_flag&, void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&) /usr/include/c++/7/mutex:684 (gdb+0x72532b)
    #22 std::__future_base::_State_baseV2::_M_set_result(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>, bool) /usr/include/c++/7/future:401 (gdb+0x174570d)
    #23 std::__future_base::_Task_state<std::function<void ()>, std::allocator<int>, void ()>::_M_run() /usr/include/c++/7/future:1423 (gdb+0x174570d)
    #24 std::packaged_task<void ()>::operator()() /usr/include/c++/7/future:1556 (gdb+0x174570d)
    #25 gdb::thread_pool::thread_function() gdbsupport/thread-pool.cc:242 (gdb+0x174570d)
    #26 void std::__invoke_impl<void, void (gdb::thread_pool::*)(), gdb::thread_pool*>(std::__invoke_memfun_deref, void (gdb::thread_pool::*&&)(), gdb::thread_pool*&&) /usr/include/c++/7/bits/invoke.h:73 (gdb+0x17480c0)
    #27 std::__invoke_result<void (gdb::thread_pool::*)(), gdb::thread_pool*>::type std::__invoke<void (gdb::thread_pool::*)(), gdb::thread_pool*>(void (gdb::thread_pool::*&&)(), gdb::thread_pool*&&) /usr/include/c++/7/bits/invoke.h:95 (gdb+0x17480c0)
    #28 decltype (__invoke((_S_declval<0ul>)(), (_S_declval<1ul>)())) std::thread::_Invoker<std::tuple<void (gdb::thread_pool::*)(), gdb::thread_pool*> >::_M_invoke<0ul, 1ul>(std::_Index_tuple<0ul, 1ul>) /usr/include/c++/7/thread:234 (gdb+0x17480c0)
    #29 std::thread::_Invoker<std::tuple<void (gdb::thread_pool::*)(), gdb::thread_pool*> >::operator()() /usr/include/c++/7/thread:243 (gdb+0x17480c0)
    #30 std::thread::_State_impl<std::thread::_Invoker<std::tuple<void (gdb::thread_pool::*)(), gdb::thread_pool*> > >::_M_run() /usr/include/c++/7/thread:186 (gdb+0x17480c0)
    #31 <null> <null> (libstdc++.so.6+0xdcac2)
  ...
SUMMARY: ThreadSanitizer: data race gdb/dwarf2/read.c:2033 in dw2_get_file_names_reader
...

The race happens when issuing the "file $exec" command.

The race is between:
- a worker thread writing the index cache, and in the process reading
  dwarf2_per_cu_data::is_debug_type, and
- the main thread writing to dwarf2_per_cu_data::files_read.

The two bitfields dwarf2_per_cu_data::files_read and
dwarf2_per_cu_data::is_debug_type share the same bitfield container.

Fix this by making dwarf2_per_cu_data::files_read a packed<bool, 1>.

Tested on x86_64-linux.

PR symtab/30718
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30718

15 months ago[gdb/symtab] Fix data race on dwarf2_per_cu_data::{mark,is_debug_types}
Tom de Vries [Fri, 4 Aug 2023 13:05:57 +0000 (15:05 +0200)]
[gdb/symtab] Fix data race on dwarf2_per_cu_data::{mark,is_debug_types}

With gdb build with -fsanitize=thread, and the exec from test-case
gdb.base/index-cache.exp, I run into:
...
$ rm -f ~/.cache/gdb/*; \
  gdb -q -batch -iex "set index-cache enabled on" index-cache \
    -ex "print foobar"
  ...
WARNING: ThreadSanitizer: data race (pid=23970)
  Write of size 1 at 0x7b200000410d by main thread:
    #0 dw_expand_symtabs_matching_file_matcher(dwarf2_per_objfile*, gdb::function_view<bool (char const*, bool)>) gdb/dwarf2/read.c:3077 (gdb+0x7ac54e)
    #1 cooked_index_functions::expand_symtabs_matching(objfile*, gdb::function_view<bool (char const*, bool)>, lookup_name_info const*, gdb::function_view<bool (char const*)>, gdb::function_view<bool (compunit_symtab*)>, enum_flags<block_search_flag_values>, domain_enum, search_domain) gdb/dwarf2/read.c:16812 (gdb+0x7d039f)
    #2 objfile::map_symtabs_matching_filename(char const*, char const*, gdb::function_view<bool (symtab*)>) gdb/symfile-debug.c:219 (gdb+0xda5aee)
    #3 iterate_over_symtabs(char const*, gdb::function_view<bool (symtab*)>) gdb/symtab.c:648 (gdb+0xdc439d)
    #4 lookup_symtab(char const*) gdb/symtab.c:662 (gdb+0xdc44a2)
    #5 classify_name gdb/c-exp.y:3083 (gdb+0x61afec)
    #6 c_yylex gdb/c-exp.y:3251 (gdb+0x61dd13)
    #7 c_yyparse() build/gdb/c-exp.c.tmp:1988 (gdb+0x61f07e)
    #8 c_parse(parser_state*) gdb/c-exp.y:3417 (gdb+0x62d864)
    #9 language_defn::parser(parser_state*) const gdb/language.c:598 (gdb+0x9771c5)
    #10 parse_exp_in_context gdb/parse.c:414 (gdb+0xb10a9b)
    #11 parse_expression(char const*, innermost_block_tracker*, enum_flags<parser_flag>) gdb/parse.c:462 (gdb+0xb110ae)
    #12 process_print_command_args gdb/printcmd.c:1321 (gdb+0xb4bf0c)
    #13 print_command_1 gdb/printcmd.c:1335 (gdb+0xb4ca2a)
    #14 print_command gdb/printcmd.c:1468 (gdb+0xb4cd5a)
    #15 do_simple_func gdb/cli/cli-decode.c:95 (gdb+0x65b078)
    #16 cmd_func(cmd_list_element*, char const*, int) gdb/cli/cli-decode.c:2735 (gdb+0x65ed53)
    #17 execute_command(char const*, int) gdb/top.c:575 (gdb+0xe3a76a)
    #18 catch_command_errors gdb/main.c:518 (gdb+0xa1837d)
    #19 execute_cmdargs gdb/main.c:617 (gdb+0xa1853f)
    #20 captured_main_1 gdb/main.c:1289 (gdb+0xa1aa58)
    #21 captured_main gdb/main.c:1310 (gdb+0xa1b95a)
    #22 gdb_main(captured_main_args*) gdb/main.c:1339 (gdb+0xa1b95a)
    #23 main gdb/gdb.c:39 (gdb+0x42506a)

  Previous read of size 1 at 0x7b200000410d by thread T1:
    #0 write_gdbindex gdb/dwarf2/index-write.c:1214 (gdb+0x75bb30)
    #1 write_dwarf_index(dwarf2_per_bfd*, char const*, char const*, char const*, dw_index_kind) gdb/dwarf2/index-write.c:1469 (gdb+0x75f803)
    #2 index_cache::store(dwarf2_per_bfd*, index_cache_store_context const&) gdb/dwarf2/index-cache.c:173 (gdb+0x755a36)
    #3 cooked_index::maybe_write_index(dwarf2_per_bfd*, index_cache_store_context const&) gdb/dwarf2/cooked-index.c:642 (gdb+0x71c96d)
    #4 operator() gdb/dwarf2/cooked-index.c:471 (gdb+0x71c96d)
    #5 _M_invoke /usr/include/c++/7/bits/std_function.h:316 (gdb+0x71c96d)
    #6 std::function<void ()>::operator()() const /usr/include/c++/7/bits/std_function.h:706 (gdb+0x72a57c)
    #7 void std::__invoke_impl<void, std::function<void ()>&>(std::__invoke_other, std::function<void ()>&) /usr/include/c++/7/bits/invoke.h:60 (gdb+0x72a5db)
    #8 std::__invoke_result<std::function<void ()>&>::type std::__invoke<std::function<void ()>&>(std::function<void ()>&) /usr/include/c++/7/bits/invoke.h:95 (gdb+0x72a5db)
    #9 std::__future_base::_Task_state<std::function<void ()>, std::allocator<int>, void ()>::_M_run()::{lambda()#1}::operator()() const /usr/include/c++/7/future:1421 (gdb+0x72a5db)
    #10 std::__future_base::_Task_setter<std::unique_ptr<std::__future_base::_Result<void>, std::__future_base::_Result_base::_Deleter>, std::__future_base::_Task_state<std::function<void ()>, std::allocator<int>, void ()>::_M_run()::{lambda()#1}, void>::operator()() const /usr/include/c++/7/future:1362 (gdb+0x72a5db)
    #11 std::_Function_handler<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> (), std::__future_base::_Task_setter<std::unique_ptr<std::__future_base::_Result<void>, std::__future_base::_Result_base::_Deleter>, std::__future_base::_Task_state<std::function<void ()>, std::allocator<int>, void ()>::_M_run()::{lambda()#1}, void> >::_M_invoke(std::_Any_data const&) /usr/include/c++/7/bits/std_function.h:302 (gdb+0x72a5db)
    #12 std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>::operator()() const /usr/include/c++/7/bits/std_function.h:706 (gdb+0x724954)
    #13 std::__future_base::_State_baseV2::_M_do_set(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*) /usr/include/c++/7/future:561 (gdb+0x724954)
    #14 void std::__invoke_impl<void, void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(std::__invoke_memfun_deref, void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&) /usr/include/c++/7/bits/invoke.h:73 (gdb+0x72434a)
    #15 std::__invoke_result<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>::type std::__invoke<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&) /usr/include/c++/7/bits/invoke.h:95 (gdb+0x72434a)
    #16 std::call_once<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(std::once_flag&, void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&)::{lambda()#1}::operator()() const /usr/include/c++/7/mutex:672 (gdb+0x72434a)
    #17 std::call_once<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(std::once_flag&, void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&)::{lambda()#2}::operator()() const /usr/include/c++/7/mutex:677 (gdb+0x72434a)
    #18 std::call_once<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(std::once_flag&, void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&)::{lambda()#2}::_FUN() /usr/include/c++/7/mutex:677 (gdb+0x72434a)
    #19 pthread_once <null> (libtsan.so.0+0x4457c)
    #20 __gthread_once /usr/include/c++/7/x86_64-suse-linux/bits/gthr-default.h:699 (gdb+0x72532b)
    #21 void std::call_once<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(std::once_flag&, void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&) /usr/include/c++/7/mutex:684 (gdb+0x72532b)
    #22 std::__future_base::_State_baseV2::_M_set_result(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>, bool) /usr/include/c++/7/future:401 (gdb+0x174568d)
    #23 std::__future_base::_Task_state<std::function<void ()>, std::allocator<int>, void ()>::_M_run() /usr/include/c++/7/future:1423 (gdb+0x174568d)
    #24 std::packaged_task<void ()>::operator()() /usr/include/c++/7/future:1556 (gdb+0x174568d)
    #25 gdb::thread_pool::thread_function() gdbsupport/thread-pool.cc:242 (gdb+0x174568d)
    #26 void std::__invoke_impl<void, void (gdb::thread_pool::*)(), gdb::thread_pool*>(std::__invoke_memfun_deref, void (gdb::thread_pool::*&&)(), gdb::thread_pool*&&) /usr/include/c++/7/bits/invoke.h:73 (gdb+0x1748040)
    #27 std::__invoke_result<void (gdb::thread_pool::*)(), gdb::thread_pool*>::type std::__invoke<void (gdb::thread_pool::*)(), gdb::thread_pool*>(void (gdb::thread_pool::*&&)(), gdb::thread_pool*&&) /usr/include/c++/7/bits/invoke.h:95 (gdb+0x1748040)
    #28 decltype (__invoke((_S_declval<0ul>)(), (_S_declval<1ul>)())) std::thread::_Invoker<std::tuple<void (gdb::thread_pool::*)(), gdb::thread_pool*> >::_M_invoke<0ul, 1ul>(std::_Index_tuple<0ul, 1ul>) /usr/include/c++/7/thread:234 (gdb+0x1748040)
    #29 std::thread::_Invoker<std::tuple<void (gdb::thread_pool::*)(), gdb::thread_pool*> >::operator()() /usr/include/c++/7/thread:243 (gdb+0x1748040)
    #30 std::thread::_State_impl<std::thread::_Invoker<std::tuple<void (gdb::thread_pool::*)(), gdb::thread_pool*> > >::_M_run() /usr/include/c++/7/thread:186 (gdb+0x1748040)
    #31 <null> <null> (libstdc++.so.6+0xdcac2)
  ...
SUMMARY: ThreadSanitizer: data race gdb/dwarf2/read.c:3077 in dw_expand_symtabs_matching_file_matcher(dwarf2_per_objfile*, gdb::function_view<bool (char const*, bool)>)
...

The race happens when issuing the "file $exec" command.

The race is between:
- a worker thread writing the index cache, and in the process reading
  dwarf2_per_cu_data::is_debug_type, and
- the main thread writing to dwarf2_per_cu_data::mark.

The two bitfields dwarf2_per_cu_data::mark and
dwarf2_per_cu_data::is_debug_type share the same bitfield container.

Fix this by making dwarf2_per_cu_data::mark a packed<unsigned int, 1>.

Tested on x86_64-linux.

PR symtab/30718
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30718

15 months ago[gdb/testsuite] Extend gdb.base/index-cache.exp
Tom de Vries [Fri, 4 Aug 2023 13:02:43 +0000 (15:02 +0200)]
[gdb/testsuite] Extend gdb.base/index-cache.exp

The test-case gdb.base/index-cache.exp uses only one source file, which
contains main.

While doing "file $exec", in set_initial_language a symbol lookup of "main" is
done, causing the symtab containing main to be expanded.

Handling of main is special, and a future optimization may skip the lookup and
expansion.

Reliably exercise:
- the lookup of main, expanding the symtab containing main, by doing
  "ptype main", and
- the lookup of another symbol, expanding a symtab not containing main, by:
  - adding another source file containing function foo, and
  - doing "ptype foo".

This triggered a segfault with target board native-extended-gdbserver, filed
as PR symtab/30712, but that seems to be fixed by a previous commit in this
series.

Tested on x86_64-linux.

Approved-By: Tom Tromey <tom@tromey.com>
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30712

15 months ago[gdb/symtab] Fix data race on dwarf2_per_cu_data::{m_header_read_in,is_debug_type}
Tom de Vries [Fri, 4 Aug 2023 13:02:43 +0000 (15:02 +0200)]
[gdb/symtab] Fix data race on dwarf2_per_cu_data::{m_header_read_in,is_debug_type}

With gdb build with -fsanitize=thread and test-case gdb.base/index-cache.exp
and target board debug-types, I run into:
...
(gdb) file build/gdb/testsuite/outputs/gdb.base/index-cache/index-cache
Reading symbols from build/gdb/testsuite/outputs/gdb.base/index-cache/index-cache...
==================
WARNING: ThreadSanitizer: data race (pid=9654)
  Write of size 1 at 0x7b200000420d by main thread:
    #0 dwarf2_per_cu_data::get_header() const gdb/dwarf2/read.c:21513 (gdb+0x8d1eee)
    #1 dwarf2_per_cu_data::addr_size() const gdb/dwarf2/read.c:21524 (gdb+0x8d1f4e)
    #2 dwarf2_cu::addr_type() const gdb/dwarf2/cu.c:112 (gdb+0x806327)
    #3 set_die_type gdb/dwarf2/read.c:21932 (gdb+0x8d3870)
    #4 read_base_type gdb/dwarf2/read.c:15448 (gdb+0x8bcacb)
    #5 read_type_die_1 gdb/dwarf2/read.c:19832 (gdb+0x8cc0a5)
    #6 read_type_die gdb/dwarf2/read.c:19767 (gdb+0x8cbe6d)
    #7 lookup_die_type gdb/dwarf2/read.c:19739 (gdb+0x8cbdc7)
    #8 die_type gdb/dwarf2/read.c:19593 (gdb+0x8cb68a)
    #9 read_subroutine_type gdb/dwarf2/read.c:14648 (gdb+0x8b998e)
    #10 read_type_die_1 gdb/dwarf2/read.c:19792 (gdb+0x8cbf2f)
    #11 read_type_die gdb/dwarf2/read.c:19767 (gdb+0x8cbe6d)
    #12 read_func_scope gdb/dwarf2/read.c:10154 (gdb+0x8a4f36)
    #13 process_die gdb/dwarf2/read.c:6667 (gdb+0x898daa)
    #14 read_file_scope gdb/dwarf2/read.c:7682 (gdb+0x89bad8)
    #15 process_die gdb/dwarf2/read.c:6654 (gdb+0x898ced)
    #16 process_full_comp_unit gdb/dwarf2/read.c:6418 (gdb+0x8981de)
    #17 process_queue gdb/dwarf2/read.c:5690 (gdb+0x894433)
    #18 dw2_do_instantiate_symtab gdb/dwarf2/read.c:1770 (gdb+0x88623a)
    #19 dw2_instantiate_symtab gdb/dwarf2/read.c:1792 (gdb+0x886300)
    #20 dw2_expand_symtabs_matching_one(dwarf2_per_cu_data*, dwarf2_per_objfile*, gdb::function_view<bool (char const*, bool)>, gdb::function_view<bool (compunit_symtab*)>) gdb/dwarf2/read.c:3042 (gdb+0x88b1f1)
    #21 cooked_index_functions::expand_symtabs_matching(objfile*, gdb::function_view<bool (char const*, bool)>, lookup_name_info const*, gdb::function_view<bool (char const*)>, gdb::function_view<bool (compunit_symtab*)>, enum_flags<block_search_flag_values>, domain_enum, search_domain) gdb/dwarf2/read.c:16917 (gdb+0x8c228e)
    #22 objfile::lookup_symbol(block_enum, char const*, domain_enum) gdb/symfile-debug.c:288 (gdb+0xf39055)
    #23 lookup_symbol_via_quick_fns gdb/symtab.c:2385 (gdb+0xf66ab7)
    #24 lookup_symbol_in_objfile gdb/symtab.c:2516 (gdb+0xf6711b)
    #25 operator() gdb/symtab.c:2562 (gdb+0xf67272)
    #26 operator() gdb/../gdbsupport/function-view.h:305 (gdb+0xf776b1)
    #27 _FUN gdb/../gdbsupport/function-view.h:299 (gdb+0xf77708)
    #28 gdb::function_view<bool (objfile*)>::operator()(objfile*) const gdb/../gdbsupport/function-view.h:289 (gdb+0xc3fc97)
    #29 svr4_iterate_over_objfiles_in_search_order gdb/solib-svr4.c:3455 (gdb+0xecae47)
    #30 gdbarch_iterate_over_objfiles_in_search_order(gdbarch*, gdb::function_view<bool (objfile*)>, objfile*) gdb/gdbarch.c:5041 (gdb+0x537cad)
    #31 lookup_global_or_static_symbol gdb/symtab.c:2559 (gdb+0xf674fb)
    #32 lookup_global_symbol(char const*, block const*, domain_enum) gdb/symtab.c:2615 (gdb+0xf67780)
    #33 language_defn::lookup_symbol_nonlocal(char const*, block const*, domain_enum) const gdb/symtab.c:2447 (gdb+0xf66d6e)
    #34 lookup_symbol_aux gdb/symtab.c:2123 (gdb+0xf65cb3)
    #35 lookup_symbol_in_language(char const*, block const*, domain_enum, language, field_of_this_result*) gdb/symtab.c:1931 (gdb+0xf64dab)
    #36 set_initial_language() gdb/symfile.c:1708 (gdb+0xf43074)
    #37 symbol_file_add_main_1 gdb/symfile.c:1212 (gdb+0xf41608)
    #38 symbol_file_command(char const*, int) gdb/symfile.c:1681 (gdb+0xf42faf)
    #39 file_command gdb/exec.c:554 (gdb+0x94ff29)
    #40 do_simple_func gdb/cli/cli-decode.c:95 (gdb+0x6d9528)
    #41 cmd_func(cmd_list_element*, char const*, int) gdb/cli/cli-decode.c:2735 (gdb+0x6e0f69)
    #42 execute_command(char const*, int) gdb/top.c:575 (gdb+0xff379c)
    #43 command_handler(char const*) gdb/event-top.c:552 (gdb+0x94b5bc)
    #44 command_line_handler(std::unique_ptr<char, gdb::xfree_deleter<char> >&&) gdb/event-top.c:788 (gdb+0x94bc79)
    #45 tui_command_line_handler gdb/tui/tui-interp.c:104 (gdb+0x1034efc)
    #46 gdb_rl_callback_handler gdb/event-top.c:259 (gdb+0x94ab61)
    #47 rl_callback_read_char readline/readline/callback.c:290 (gdb+0x11be4ef)
    #48 gdb_rl_callback_read_char_wrapper_noexcept gdb/event-top.c:195 (gdb+0x94a960)
    #49 gdb_rl_callback_read_char_wrapper gdb/event-top.c:234 (gdb+0x94aa21)
    #50 stdin_event_handler gdb/ui.c:155 (gdb+0x10751a0)
    #51 handle_file_event gdbsupport/event-loop.cc:573 (gdb+0x1d95bac)
    #52 gdb_wait_for_event gdbsupport/event-loop.cc:694 (gdb+0x1d962e4)
    #53 gdb_do_one_event(int) gdbsupport/event-loop.cc:264 (gdb+0x1d946d0)
    #54 start_event_loop gdb/main.c:412 (gdb+0xb5ab52)
    #55 captured_command_loop gdb/main.c:476 (gdb+0xb5ad41)
    #56 captured_main gdb/main.c:1320 (gdb+0xb5cec1)
    #57 gdb_main(captured_main_args*) gdb/main.c:1339 (gdb+0xb5cf70)
    #58 main gdb/gdb.c:32 (gdb+0x416776)

  Previous read of size 1 at 0x7b200000420d by thread T11:
    #0 write_gdbindex gdb/dwarf2/index-write.c:1229 (gdb+0x831630)
    #1 write_dwarf_index(dwarf2_per_bfd*, char const*, char const*, char const*, dw_index_kind) gdb/dwarf2/index-write.c:1484 (gdb+0x832897)
    #2 index_cache::store(dwarf2_per_bfd*, index_cache_store_context const&) gdb/dwarf2/index-cache.c:173 (gdb+0x82db8d)
    #3 cooked_index::maybe_write_index(dwarf2_per_bfd*, index_cache_store_context const&) gdb/dwarf2/cooked-index.c:645 (gdb+0x7f1d49)
    #4 operator() gdb/dwarf2/cooked-index.c:474 (gdb+0x7f0f31)
    #5 _M_invoke /usr/include/c++/7/bits/std_function.h:316 (gdb+0x7f2a13)
    #6 std::function<void ()>::operator()() const /usr/include/c++/7/bits/std_function.h:706 (gdb+0x700952)
    #7 void std::__invoke_impl<void, std::function<void ()>&>(std::__invoke_other, std::function<void ()>&) /usr/include/c++/7/bits/invoke.h:60 (gdb+0x7381a0)
    #8 std::__invoke_result<std::function<void ()>&>::type std::__invoke<std::function<void ()>&>(std::function<void ()>&) /usr/include/c++/7/bits/invoke.h:95 (gdb+0x737e91)
    #9 std::__future_base::_Task_state<std::function<void ()>, std::allocator<int>, void ()>::_M_run()::{lambda()#1}::operator()() const /usr/include/c++/7/future:1421 (gdb+0x737b59)
    #10 std::__future_base::_Task_setter<std::unique_ptr<std::__future_base::_Result<void>, std::__future_base::_Result_base::_Deleter>, std::__future_base::_Task_state<std::function<void ()>, std::allocator<int>, void ()>::_M_run()::{lambda()#1}, void>::operator()() const /usr/include/c++/7/future:1362 (gdb+0x738660)
    #11 std::_Function_handler<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> (), std::__future_base::_Task_setter<std::unique_ptr<std::__future_base::_Result<void>, std::__future_base::_Result_base::_Deleter>, std::__future_base::_Task_state<std::function<void ()>, std::allocator<int>, void ()>::_M_run()::{lambda()#1}, void> >::_M_invoke(std::_Any_data const&) /usr/include/c++/7/bits/std_function.h:302 (gdb+0x73825c)
    #12 std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>::operator()() const /usr/include/c++/7/bits/std_function.h:706 (gdb+0x733623)
    #13 std::__future_base::_State_baseV2::_M_do_set(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*) /usr/include/c++/7/future:561 (gdb+0x732bdf)
    #14 void std::__invoke_impl<void, void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(std::__invoke_memfun_deref, void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&) /usr/include/c++/7/bits/invoke.h:73 (gdb+0x734c4f)
    #15 std::__invoke_result<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>::type std::__invoke<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&) /usr/include/c++/7/bits/invoke.h:95 (gdb+0x733bc5)
    #16 std::call_once<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(std::once_flag&, void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&)::{lambda()#1}::operator()() const /usr/include/c++/7/mutex:672 (gdb+0x73300d)
    #17 std::call_once<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(std::once_flag&, void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&)::{lambda()#2}::operator()() const /usr/include/c++/7/mutex:677 (gdb+0x7330b2)
    #18 std::call_once<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(std::once_flag&, void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&)::{lambda()#2}::_FUN() /usr/include/c++/7/mutex:677 (gdb+0x7330f2)
    #19 pthread_once <null> (libtsan.so.0+0x4457c)
    #20 __gthread_once /usr/include/c++/7/x86_64-suse-linux/bits/gthr-default.h:699 (gdb+0x72f5dd)
    #21 void std::call_once<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(std::once_flag&, void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&) /usr/include/c++/7/mutex:684 (gdb+0x733224)
    #22 std::__future_base::_State_baseV2::_M_set_result(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>, bool) /usr/include/c++/7/future:401 (gdb+0x732852)
    #23 std::__future_base::_Task_state<std::function<void ()>, std::allocator<int>, void ()>::_M_run() /usr/include/c++/7/future:1423 (gdb+0x737bef)
    #24 std::packaged_task<void ()>::operator()() /usr/include/c++/7/future:1556 (gdb+0x1dad25a)
    #25 gdb::thread_pool::thread_function() gdbsupport/thread-pool.cc:242 (gdb+0x1dacb7c)
    #26 void std::__invoke_impl<void, void (gdb::thread_pool::*)(), gdb::thread_pool*>(std::__invoke_memfun_deref, void (gdb::thread_pool::*&&)(), gdb::thread_pool*&&) /usr/include/c++/7/bits/invoke.h:73 (gdb+0x1dadc2b)
    #27 std::__invoke_result<void (gdb::thread_pool::*)(), gdb::thread_pool*>::type std::__invoke<void (gdb::thread_pool::*)(), gdb::thread_pool*>(void (gdb::thread_pool::*&&)(), gdb::thread_pool*&&) /usr/include/c++/7/bits/invoke.h:95 (gdb+0x1dad05c)
    #28 decltype (__invoke((_S_declval<0ul>)(), (_S_declval<1ul>)())) std::thread::_Invoker<std::tuple<void (gdb::thread_pool::*)(), gdb::thread_pool*> >::_M_invoke<0ul, 1ul>(std::_Index_tuple<0ul, 1ul>) /usr/include/c++/7/thread:234 (gdb+0x1db038e)
    #29 std::thread::_Invoker<std::tuple<void (gdb::thread_pool::*)(), gdb::thread_pool*> >::operator()() /usr/include/c++/7/thread:243 (gdb+0x1db0319)
    #30 std::thread::_State_impl<std::thread::_Invoker<std::tuple<void (gdb::thread_pool::*)(), gdb::thread_pool*> > >::_M_run() /usr/include/c++/7/thread:186 (gdb+0x1db02ce)
    #31 <null> <null> (libstdc++.so.6+0xdcac2)
  ...
SUMMARY: ThreadSanitizer: data race gdb/dwarf2/read.c:21513 in dwarf2_per_cu_data::get_header() const
...

The race happens when issuing the "file $exec" command.

The race is between:
- a worker thread writing the index cache, and in the process reading
   dwarf2_per_cu_data::is_debug_type, and
- the main thread writing to dwarf2_per_cu_data::m_header_read_in.

The two bitfields dwarf2_per_cu_data::m_header_read_in and
dwarf2_per_cu_data::is_debug_type share the same bitfield container.

Fix this by making dwarf2_per_cu_data::m_header_read_in a packed<bool, 1>.

Tested on x86_64-linux.

Approved-By: Tom Tromey <tom@tromey.com>
PR symtab/30392
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30392

15 months ago[gdb/symtab] Fix race on dwarf2_per_cu_data::{queued,is_debug_type}
Tom de Vries [Fri, 4 Aug 2023 13:02:43 +0000 (15:02 +0200)]
[gdb/symtab] Fix race on dwarf2_per_cu_data::{queued,is_debug_type}

With gdb build with -fsanitize=thread and test-case gdb.base/index-cache.exp I
run into:
...
(gdb) file build/gdb/testsuite/outputs/gdb.base/index-cache/index-cache
Reading symbols from build/gdb/testsuite/outputs/gdb.base/index-cache/index-cache...
==================
WARNING: ThreadSanitizer: data race (pid=24296)
  Write of size 1 at 0x7b200000420d by main thread:
    #0 queue_comp_unit gdb/dwarf2/read.c:5564 (gdb+0x8939ce)
    #1 dw2_do_instantiate_symtab gdb/dwarf2/read.c:1754 (gdb+0x885b96)
    #2 dw2_instantiate_symtab gdb/dwarf2/read.c:1792 (gdb+0x885d86)
    #3 dw2_expand_symtabs_matching_one(dwarf2_per_cu_data*, dwarf2_per_objfile*, gdb::function_view<bool (char const*, bool)>, gdb::function_view<bool (compunit_symtab*)>) gdb/dwarf2/read.c:3042 (gdb+0x88ac77)
    #4 cooked_index_functions::expand_symtabs_matching(objfile*, gdb::function_view<bool (char const*, bool)>, lookup_name_info const*, gdb::function_view<bool (char const*)>, gdb::function_view<bool (compunit_symtab*)>, enum_flags<block_search_flag_values>, domain_enum, search_domain) gdb/dwarf2/read.c:16915 (gdb+0x8c1c8a)
    #5 objfile::lookup_symbol(block_enum, char const*, domain_enum) gdb/symfile-debug.c:288 (gdb+0xf389a1)
    #6 lookup_symbol_via_quick_fns gdb/symtab.c:2385 (gdb+0xf66403)
    #7 lookup_symbol_in_objfile gdb/symtab.c:2516 (gdb+0xf66a67)
    #8 operator() gdb/symtab.c:2562 (gdb+0xf66bbe)
    #9 operator() gdb/../gdbsupport/function-view.h:305 (gdb+0xf76ffd)
    #10 _FUN gdb/../gdbsupport/function-view.h:299 (gdb+0xf77054)
    #11 gdb::function_view<bool (objfile*)>::operator()(objfile*) const gdb/../gdbsupport/function-view.h:289 (gdb+0xc3f5e3)
    #12 svr4_iterate_over_objfiles_in_search_order gdb/solib-svr4.c:3455 (gdb+0xeca793)
    #13 gdbarch_iterate_over_objfiles_in_search_order(gdbarch*, gdb::function_view<bool (objfile*)>, objfile*) gdb/gdbarch.c:5041 (gdb+0x537cad)
    #14 lookup_global_or_static_symbol gdb/symtab.c:2559 (gdb+0xf66e47)
    #15 lookup_global_symbol(char const*, block const*, domain_enum) gdb/symtab.c:2615 (gdb+0xf670cc)
    #16 language_defn::lookup_symbol_nonlocal(char const*, block const*, domain_enum) const gdb/symtab.c:2447 (gdb+0xf666ba)
    #17 lookup_symbol_aux gdb/symtab.c:2123 (gdb+0xf655ff)
    #18 lookup_symbol_in_language(char const*, block const*, domain_enum, language, field_of_this_result*) gdb/symtab.c:1931 (gdb+0xf646f7)
    #19 set_initial_language() gdb/symfile.c:1708 (gdb+0xf429c0)
    #20 symbol_file_add_main_1 gdb/symfile.c:1212 (gdb+0xf40f54)
    #21 symbol_file_command(char const*, int) gdb/symfile.c:1681 (gdb+0xf428fb)
    #22 file_command gdb/exec.c:554 (gdb+0x94f875)
    #23 do_simple_func gdb/cli/cli-decode.c:95 (gdb+0x6d9528)
    #24 cmd_func(cmd_list_element*, char const*, int) gdb/cli/cli-decode.c:2735 (gdb+0x6e0f69)
    #25 execute_command(char const*, int) gdb/top.c:575 (gdb+0xff3166)
    #26 command_handler(char const*) gdb/event-top.c:552 (gdb+0x94af08)
    #27 command_line_handler(std::unique_ptr<char, gdb::xfree_deleter<char> >&&) gdb/event-top.c:788 (gdb+0x94b5c5)
    #28 tui_command_line_handler gdb/tui/tui-interp.c:104 (gdb+0x10348c6)
    #29 gdb_rl_callback_handler gdb/event-top.c:259 (gdb+0x94a4ad)
    #30 rl_callback_read_char readline/readline/callback.c:290 (gdb+0x11bdf87)
    #31 gdb_rl_callback_read_char_wrapper_noexcept gdb/event-top.c:195 (gdb+0x94a2ac)
    #32 gdb_rl_callback_read_char_wrapper gdb/event-top.c:234 (gdb+0x94a36d)
    #33 stdin_event_handler gdb/ui.c:155 (gdb+0x1074b6a)
    #34 handle_file_event gdbsupport/event-loop.cc:573 (gdb+0x1d9502c)
    #35 gdb_wait_for_event gdbsupport/event-loop.cc:694 (gdb+0x1d95764)
    #36 gdb_do_one_event(int) gdbsupport/event-loop.cc:264 (gdb+0x1d93b50)
    #37 start_event_loop gdb/main.c:412 (gdb+0xb5a49e)
    #38 captured_command_loop gdb/main.c:476 (gdb+0xb5a68d)
    #39 captured_main gdb/main.c:1320 (gdb+0xb5c80d)
    #40 gdb_main(captured_main_args*) gdb/main.c:1339 (gdb+0xb5c8bc)
    #41 main gdb/gdb.c:32 (gdb+0x416776)

  Previous read of size 1 at 0x7b200000420d by thread T12:
    #0 write_gdbindex gdb/dwarf2/index-write.c:1229 (gdb+0x8310c8)
    #1 write_dwarf_index(dwarf2_per_bfd*, char const*, char const*, char const*, dw_index_kind) gdb/dwarf2/index-write.c:1484 (gdb+0x83232f)
    #2 index_cache::store(dwarf2_per_bfd*, index_cache_store_context*) gdb/dwarf2/index-cache.c:177 (gdb+0x82d62b)
    #3 cooked_index::maybe_write_index(dwarf2_per_bfd*) gdb/dwarf2/cooked-index.c:640 (gdb+0x7f1bf7)
    #4 operator() gdb/dwarf2/cooked-index.c:470 (gdb+0x7f0f40)
    #5 _M_invoke /usr/include/c++/7/bits/std_function.h:316 (gdb+0x7f2909)
    #6 std::function<void ()>::operator()() const /usr/include/c++/7/bits/std_function.h:706 (gdb+0x700952)
    #7 void std::__invoke_impl<void, std::function<void ()>&>(std::__invoke_other, std::function<void ()>&) /usr/include/c++/7/bits/invoke.h:60 (gdb+0x7381a0)
    #8 std::__invoke_result<std::function<void ()>&>::type std::__invoke<std::function<void ()>&>(std::function<void ()>&) /usr/include/c++/7/bits/invoke.h:95 (gdb+0x737e91)
    #9 std::__future_base::_Task_state<std::function<void ()>, std::allocator<int>, void ()>::_M_run()::{lambda()#1}::operator()() const /usr/include/c++/7/future:1421 (gdb+0x737b59)
    #10 std::__future_base::_Task_setter<std::unique_ptr<std::__future_base::_Result<void>, std::__future_base::_Result_base::_Deleter>, std::__future_base::_Task_state<std::function<void ()>, std::allocator<int>, void ()>::_M_run()::{lambda()#1}, void>::operator()() const /usr/include/c++/7/future:1362 (gdb+0x738660)
    #11 std::_Function_handler<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> (), std::__future_base::_Task_setter<std::unique_ptr<std::__future_base::_Result<void>, std::__future_base::_Result_base::_Deleter>, std::__future_base::_Task_state<std::function<void ()>, std::allocator<int>, void ()>::_M_run()::{lambda()#1}, void> >::_M_invoke(std::_Any_data const&) /usr/include/c++/7/bits/std_function.h:302 (gdb+0x73825c)
    #12 std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>::operator()() const /usr/include/c++/7/bits/std_function.h:706 (gdb+0x733623)
    #13 std::__future_base::_State_baseV2::_M_do_set(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*) /usr/include/c++/7/future:561 (gdb+0x732bdf)
    #14 void std::__invoke_impl<void, void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(std::__invoke_memfun_deref, void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&) /usr/include/c++/7/bits/invoke.h:73 (gdb+0x734c4f)
    #15 std::__invoke_result<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>::type std::__invoke<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&) /usr/include/c++/7/bits/invoke.h:95 (gdb+0x733bc5)
    #16 std::call_once<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(std::once_flag&, void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&)::{lambda()#1}::operator()() const /usr/include/c++/7/mutex:672 (gdb+0x73300d)
    #17 std::call_once<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(std::once_flag&, void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&)::{lambda()#2}::operator()() const /usr/include/c++/7/mutex:677 (gdb+0x7330b2)
    #18 std::call_once<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(std::once_flag&, void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&)::{lambda()#2}::_FUN() /usr/include/c++/7/mutex:677 (gdb+0x7330f2)
    #19 pthread_once <null> (libtsan.so.0+0x4457c)
    #20 __gthread_once /usr/include/c++/7/x86_64-suse-linux/bits/gthr-default.h:699 (gdb+0x72f5dd)
    #21 void std::call_once<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(std::once_flag&, void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&) /usr/include/c++/7/mutex:684 (gdb+0x733224)
    #22 std::__future_base::_State_baseV2::_M_set_result(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>, bool) /usr/include/c++/7/future:401 (gdb+0x732852)
    #23 std::__future_base::_Task_state<std::function<void ()>, std::allocator<int>, void ()>::_M_run() /usr/include/c++/7/future:1423 (gdb+0x737bef)
    #24 std::packaged_task<void ()>::operator()() /usr/include/c++/7/future:1556 (gdb+0x1dac6da)
    #25 gdb::thread_pool::thread_function() gdbsupport/thread-pool.cc:242 (gdb+0x1dabffc)
    #26 void std::__invoke_impl<void, void (gdb::thread_pool::*)(), gdb::thread_pool*>(std::__invoke_memfun_deref, void (gdb::thread_pool::*&&)(), gdb::thread_pool*&&) /usr/include/c++/7/bits/invoke.h:73 (gdb+0x1dad0ab)
    #27 std::__invoke_result<void (gdb::thread_pool::*)(), gdb::thread_pool*>::type std::__invoke<void (gdb::thread_pool::*)(), gdb::thread_pool*>(void (gdb::thread_pool::*&&)(), gdb::thread_pool*&&) /usr/include/c++/7/bits/invoke.h:95 (gdb+0x1dac4dc)
    #28 decltype (__invoke((_S_declval<0ul>)(), (_S_declval<1ul>)())) std::thread::_Invoker<std::tuple<void (gdb::thread_pool::*)(), gdb::thread_pool*> >::_M_invoke<0ul, 1ul>(std::_Index_tuple<0ul, 1ul>) /usr/include/c++/7/thread:234 (gdb+0x1daf80e)
    #29 std::thread::_Invoker<std::tuple<void (gdb::thread_pool::*)(), gdb::thread_pool*> >::operator()() /usr/include/c++/7/thread:243 (gdb+0x1daf799)
    #30 std::thread::_State_impl<std::thread::_Invoker<std::tuple<void (gdb::thread_pool::*)(), gdb::thread_pool*> > >::_M_run() /usr/include/c++/7/thread:186 (gdb+0x1daf74e)
    #31 <null> <null> (libstdc++.so.6+0xdcac2)
 ...
SUMMARY: ThreadSanitizer: data race gdb/dwarf2/read.c:5564 in queue_comp_unit
...

The race happens when issuing the "file $exec" command.

The race is between:
- a worker thread writing the index cache, and in the process reading
  dwarf2_per_cu_data::is_debug_type, and
- the main thread expanding the CU containing main, and in the process setting
  dwarf2_per_cu_data::queued.

The two bitfields dwarf2_per_cu_data::queue and
dwarf2_per_cu_data::is_debug_type share the same bitfield container.

Fix this by making dwarf2_per_cu_data::queued a packed<bool, 1>.

Tested on x86_64-linux.

Approved-By: Tom Tromey <tom@tromey.com>
PR symtab/30392
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30392

15 months ago[gdb/symtab] Fix data race on bfd::{cacheable,format}
Tom de Vries [Fri, 4 Aug 2023 13:02:43 +0000 (15:02 +0200)]
[gdb/symtab] Fix data race on bfd::{cacheable,format}

With gdb build with -fsanitize=thread and test-case gdb.base/index-cache.exp I
run into:
...
(gdb) file build/gdb/testsuite/outputs/gdb.base/index-cache/index-cache
Reading symbols from build/gdb/testsuite/outputs/gdb.base/index-cache/index-cache...
==================
WARNING: ThreadSanitizer: data race (pid=12261)
  Write of size 4 at 0x7b4400097d08 by main thread:
    #0 bfd_open_file bfd/cache.c:584 (gdb+0x148bb92)
    #1 bfd_cache_lookup_worker bfd/cache.c:261 (gdb+0x148b12a)
    #2 cache_bseek bfd/cache.c:289 (gdb+0x148b324)
    #3 bfd_seek bfd/bfdio.c:459 (gdb+0x1489c31)
    #4 _bfd_generic_get_section_contents bfd/libbfd.c:1069 (gdb+0x14977a4)
    #5 bfd_get_section_contents bfd/section.c:1606 (gdb+0x149cc7c)
    #6 gdb_bfd_scan_elf_dyntag(int, bfd*, unsigned long*, unsigned long*) gdb/solib.c:1601 (gdb+0xed8eca)
    #7 elf_locate_base gdb/solib-svr4.c:705 (gdb+0xec28ac)
    #8 svr4_iterate_over_objfiles_in_search_order gdb/solib-svr4.c:3430 (gdb+0xeca55d)
    #9 gdbarch_iterate_over_objfiles_in_search_order(gdbarch*, gdb::function_view<bool (objfile*)>, objfile*) gdb/gdbarch.c:5041 (gdb+0x537cad)
    #10 find_main_name gdb/symtab.c:6270 (gdb+0xf743a5)
    #11 main_language() gdb/symtab.c:6313 (gdb+0xf74499)
    #12 set_initial_language() gdb/symfile.c:1700 (gdb+0xf4285c)
    #13 symbol_file_add_main_1 gdb/symfile.c:1212 (gdb+0xf40e2a)
    #14 symbol_file_command(char const*, int) gdb/symfile.c:1681 (gdb+0xf427d1)
    #15 file_command gdb/exec.c:554 (gdb+0x94f74b)
    #16 do_simple_func gdb/cli/cli-decode.c:95 (gdb+0x6d9528)
    #17 cmd_func(cmd_list_element*, char const*, int) gdb/cli/cli-decode.c:2735 (gdb+0x6e0f69)
    #18 execute_command(char const*, int) gdb/top.c:575 (gdb+0xff303c)
    #19 command_handler(char const*) gdb/event-top.c:552 (gdb+0x94adde)
    #20 command_line_handler(std::unique_ptr<char, gdb::xfree_deleter<char> >&&) gdb/event-top.c:788 (gdb+0x94b49b)
    #21 tui_command_line_handler gdb/tui/tui-interp.c:104 (gdb+0x103479c)
    #22 gdb_rl_callback_handler gdb/event-top.c:259 (gdb+0x94a383)
    #23 rl_callback_read_char readline/readline/callback.c:290 (gdb+0x11bde5d)
    #24 gdb_rl_callback_read_char_wrapper_noexcept gdb/event-top.c:195 (gdb+0x94a182)
    #25 gdb_rl_callback_read_char_wrapper gdb/event-top.c:234 (gdb+0x94a243)
    #26 stdin_event_handler gdb/ui.c:155 (gdb+0x1074a40)
    #27 handle_file_event gdbsupport/event-loop.cc:573 (gdb+0x1d94f02)
    #28 gdb_wait_for_event gdbsupport/event-loop.cc:694 (gdb+0x1d9563a)
    #29 gdb_do_one_event(int) gdbsupport/event-loop.cc:264 (gdb+0x1d93a26)
    #30 start_event_loop gdb/main.c:412 (gdb+0xb5a374)
    #31 captured_command_loop gdb/main.c:476 (gdb+0xb5a563)
    #32 captured_main gdb/main.c:1320 (gdb+0xb5c6e3)
    #33 gdb_main(captured_main_args*) gdb/main.c:1339 (gdb+0xb5c792)
    #34 main gdb/gdb.c:32 (gdb+0x416776)

  Previous read of size 1 at 0x7b4400097d08 by thread T12:
    #0 bfd_check_format_matches bfd/format.c:323 (gdb+0x1492db4)
    #1 bfd_check_format bfd/format.c:94 (gdb+0x1492104)
    #2 build_id_bfd_get(bfd*) gdb/build-id.c:42 (gdb+0x6648f7)
    #3 index_cache::store(dwarf2_per_bfd*, index_cache_store_context*) gdb/dwarf2/index-cache.c:110 (gdb+0x82d205)
    #4 cooked_index::maybe_write_index(dwarf2_per_bfd*) gdb/dwarf2/cooked-index.c:640 (gdb+0x7f1bf1)
    #5 operator() gdb/dwarf2/cooked-index.c:470 (gdb+0x7f0f40)
    #6 _M_invoke /usr/include/c++/7/bits/std_function.h:316 (gdb+0x7f28f7)
    #7 std::function<void ()>::operator()() const /usr/include/c++/7/bits/std_function.h:706 (gdb+0x700952)
    #8 void std::__invoke_impl<void, std::function<void ()>&>(std::__invoke_other, std::function<void ()>&) /usr/include/c++/7/bits/invoke.h:60 (gdb+0x7381a0)
    #9 std::__invoke_result<std::function<void ()>&>::type std::__invoke<std::function<void ()>&>(std::function<void ()>&) /usr/include/c++/7/bits/invoke.h:95 (gdb+0x737e91)
    #10 std::__future_base::_Task_state<std::function<void ()>, std::allocator<int>, void ()>::_M_run()::{lambda()#1}::operator()() const /usr/include/c++/7/future:1421 (gdb+0x737b59)
    #11 std::__future_base::_Task_setter<std::unique_ptr<std::__future_base::_Result<void>, std::__future_base::_Result_base::_Deleter>, std::__future_base::_Task_state<std::function<void ()>, std::allocator<int>, void ()>::_M_run()::{lambda()#1}, void>::operator()() const /usr/include/c++/7/future:1362 (gdb+0x738660)
    #12 std::_Function_handler<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> (), std::__future_base::_Task_setter<std::unique_ptr<std::__future_base::_Result<void>, std::__future_base::_Result_base::_Deleter>, std::__future_base::_Task_state<std::function<void ()>, std::allocator<int>, void ()>::_M_run()::{lambda()#1}, void> >::_M_invoke(std::_Any_data const&) /usr/include/c++/7/bits/std_function.h:302 (gdb+0x73825c)
    #13 std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>::operator()() const /usr/include/c++/7/bits/std_function.h:706 (gdb+0x733623)
    #14 std::__future_base::_State_baseV2::_M_do_set(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*) /usr/include/c++/7/future:561 (gdb+0x732bdf)
    #15 void std::__invoke_impl<void, void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(std::__invoke_memfun_deref, void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&) /usr/include/c++/7/bits/invoke.h:73 (gdb+0x734c4f)
    #16 std::__invoke_result<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>::type std::__invoke<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&) /usr/include/c++/7/bits/invoke.h:95 (gdb+0x733bc5)
    #17 std::call_once<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(std::once_flag&, void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&)::{lambda()#1}::operator()() const /usr/include/c++/7/mutex:672 (gdb+0x73300d)
    #18 std::call_once<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(std::once_flag&, void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&)::{lambda()#2}::operator()() const /usr/include/c++/7/mutex:677 (gdb+0x7330b2)
    #19 std::call_once<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(std::once_flag&, void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&)::{lambda()#2}::_FUN() /usr/include/c++/7/mutex:677 (gdb+0x7330f2)
    #20 pthread_once <null> (libtsan.so.0+0x4457c)
    #21 __gthread_once /usr/include/c++/7/x86_64-suse-linux/bits/gthr-default.h:699 (gdb+0x72f5dd)
    #22 void std::call_once<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(std::once_flag&, void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&) /usr/include/c++/7/mutex:684 (gdb+0x733224)
    #23 std::__future_base::_State_baseV2::_M_set_result(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>, bool) /usr/include/c++/7/future:401 (gdb+0x732852)
    #24 std::__future_base::_Task_state<std::function<void ()>, std::allocator<int>, void ()>::_M_run() /usr/include/c++/7/future:1423 (gdb+0x737bef)
    #25 std::packaged_task<void ()>::operator()() /usr/include/c++/7/future:1556 (gdb+0x1dac5b0)
    #26 gdb::thread_pool::thread_function() gdbsupport/thread-pool.cc:242 (gdb+0x1dabed2)
    #27 void std::__invoke_impl<void, void (gdb::thread_pool::*)(), gdb::thread_pool*>(std::__invoke_memfun_deref, void (gdb::thread_pool::*&&)(), gdb::thread_pool*&&) /usr/include/c++/7/bits/invoke.h:73 (gdb+0x1dacf81)
    #28 std::__invoke_result<void (gdb::thread_pool::*)(), gdb::thread_pool*>::type std::__invoke<void (gdb::thread_pool::*)(), gdb::thread_pool*>(void (gdb::thread_pool::*&&)(), gdb::thread_pool*&&) /usr/include/c++/7/bits/invoke.h:95 (gdb+0x1dac3b2)
    #29 decltype (__invoke((_S_declval<0ul>)(), (_S_declval<1ul>)())) std::thread::_Invoker<std::tuple<void (gdb::thread_pool::*)(), gdb::thread_pool*> >::_M_invoke<0ul, 1ul>(std::_Index_tuple<0ul, 1ul>) /usr/include/c++/7/thread:234 (gdb+0x1daf6e4)
    #30 std::thread::_Invoker<std::tuple<void (gdb::thread_pool::*)(), gdb::thread_pool*> >::operator()() /usr/include/c++/7/thread:243 (gdb+0x1daf66f)
    #31 std::thread::_State_impl<std::thread::_Invoker<std::tuple<void (gdb::thread_pool::*)(), gdb::thread_pool*> > >::_M_run() /usr/include/c++/7/thread:186 (gdb+0x1daf624)
    #32 <null> <null> (libstdc++.so.6+0xdcac2)
  ...
SUMMARY: ThreadSanitizer: data race bfd/cache.c:584 in bfd_open_file
...

The race happens when issuing the "file $exec" command.

The race is between:
- a worker thread getting the build id while writing the index cache, and in
  the process reading bfd::format, and
- the main thread calling find_main_name, and in the process setting
  bfd::cacheable.

The two bitfields bfd::cacheable and bfd::format share the same bitfield
container.

Fix this by capturing the build id in the main thread, and using the captured
value in the worker thread.

Likewise for the dwz build id, which likely suffers from the same issue.

While we're at it, also move the creation of the cache directory to
the index_cache_store_context constructor, to:
- make sure there's no race between subsequent file commands, and
- issue any related warning or error messages during the file command.

Tested on x86_64-linux.

Approved-By: Tom Tromey <tom@tromey.com>
PR symtab/30392
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30392

15 months ago[gdb/symtab] Fix data race on index_cache::m_enabled
Tom de Vries [Fri, 4 Aug 2023 13:02:43 +0000 (15:02 +0200)]
[gdb/symtab] Fix data race on index_cache::m_enabled

With gdb build with -fsanitize=thread and test-case gdb.base/index-cache.exp I
run into:
...
(gdb) file build/gdb/testsuite/outputs/gdb.base/index-cache/index-cache
Reading symbols from build/gdb/testsuite/outputs/gdb.base/index-cache/index-cache...
(gdb) show index-cache enabled
The index cache is off.
(gdb) PASS: gdb.base/index-cache.exp: test_basic_stuff: index-cache is disabled by default
set index-cache enabled on
==================
WARNING: ThreadSanitizer: data race (pid=32248)
  Write of size 1 at 0x00000321f540 by main thread:
    #0 index_cache::enable() gdb/dwarf2/index-cache.c:76 (gdb+0x82cfdd)
    #1 set_index_cache_enabled_command gdb/dwarf2/index-cache.c:270 (gdb+0x82d9af)
    #2 bool setting::set<bool>(bool const&) gdb/command.h:353 (gdb+0x6fe5f2)
    #3 do_set_command(char const*, int, cmd_list_element*) gdb/cli/cli-setshow.c:414 (gdb+0x6fcd21)
    #4 execute_command(char const*, int) gdb/top.c:567 (gdb+0xff2e64)
    #5 command_handler(char const*) gdb/event-top.c:552 (gdb+0x94acc0)
    #6 command_line_handler(std::unique_ptr<char, gdb::xfree_deleter<char> >&&) gdb/event-top.c:788 (gdb+0x94b37d)
    #7 tui_command_line_handler gdb/tui/tui-interp.c:104 (gdb+0x103467e)
    #8 gdb_rl_callback_handler gdb/event-top.c:259 (gdb+0x94a265)
    #9 rl_callback_read_char readline/readline/callback.c:290 (gdb+0x11bdd3f)
    #10 gdb_rl_callback_read_char_wrapper_noexcept gdb/event-top.c:195 (gdb+0x94a064)
    #11 gdb_rl_callback_read_char_wrapper gdb/event-top.c:234 (gdb+0x94a125)
    #12 stdin_event_handler gdb/ui.c:155 (gdb+0x1074922)
    #13 handle_file_event gdbsupport/event-loop.cc:573 (gdb+0x1d94de4)
    #14 gdb_wait_for_event gdbsupport/event-loop.cc:694 (gdb+0x1d9551c)
    #15 gdb_do_one_event(int) gdbsupport/event-loop.cc:264 (gdb+0x1d93908)
    #16 start_event_loop gdb/main.c:412 (gdb+0xb5a256)
    #17 captured_command_loop gdb/main.c:476 (gdb+0xb5a445)
    #18 captured_main gdb/main.c:1320 (gdb+0xb5c5c5)
    #19 gdb_main(captured_main_args*) gdb/main.c:1339 (gdb+0xb5c674)
    #20 main gdb/gdb.c:32 (gdb+0x416776)

  Previous read of size 1 at 0x00000321f540 by thread T12:
    #0 index_cache::enabled() const gdb/dwarf2/index-cache.h:48 (gdb+0x82e1a6)
    #1 index_cache::store(dwarf2_per_bfd*) gdb/dwarf2/index-cache.c:94 (gdb+0x82d0bc)
    #2 cooked_index::maybe_write_index(dwarf2_per_bfd*) gdb/dwarf2/cooked-index.c:638 (gdb+0x7f1b97)
    #3 operator() gdb/dwarf2/cooked-index.c:468 (gdb+0x7f0f24)
    #4 _M_invoke /usr/include/c++/7/bits/std_function.h:316 (gdb+0x7f285b)
    #5 std::function<void ()>::operator()() const /usr/include/c++/7/bits/std_function.h:706 (gdb+0x700952)
    #6 void std::__invoke_impl<void, std::function<void ()>&>(std::__invoke_other, std::function<void ()>&) /usr/include/c++/7/bits/invoke.h:60 (gdb+0x7381a0)
    #7 std::__invoke_result<std::function<void ()>&>::type std::__invoke<std::function<void ()>&>(std::function<void ()>&) /usr/include/c++/7/bits/invoke.h:95 (gdb+0x737e91)
    #8 std::__future_base::_Task_state<std::function<void ()>, std::allocator<int>, void ()>::_M_run()::{lambda()#1}::operator()() const /usr/include/c++/7/future:1421 (gdb+0x737b59)
    #9 std::__future_base::_Task_setter<std::unique_ptr<std::__future_base::_Result<void>, std::__future_base::_Result_base::_Deleter>, std::__future_base::_Task_state<std::function<void ()>, std::allocator<int>, void ()>::_M_run()::{lambda()#1}, void>::operator()() const /usr/include/c++/7/future:1362 (gdb+0x738660)
    #10 std::_Function_handler<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> (), std::__future_base::_Task_setter<std::unique_ptr<std::__future_base::_Result<void>, std::__future_base::_Result_base::_Deleter>, std::__future_base::_Task_state<std::function<void ()>, std::allocator<int>, void ()>::_M_run()::{lambda()#1}, void> >::_M_invoke(std::_Any_data const&) /usr/include/c++/7/bits/std_function.h:302 (gdb+0x73825c)
    #11 std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>::operator()() const /usr/include/c++/7/bits/std_function.h:706 (gdb+0x733623)
    #12 std::__future_base::_State_baseV2::_M_do_set(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*) /usr/include/c++/7/future:561 (gdb+0x732bdf)
    #13 void std::__invoke_impl<void, void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(std::__invoke_memfun_deref, void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&) /usr/include/c++/7/bits/invoke.h:73 (gdb+0x734c4f)
    #14 std::__invoke_result<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>::type std::__invoke<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&) /usr/include/c++/7/bits/invoke.h:95 (gdb+0x733bc5)
    #15 std::call_once<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(std::once_flag&, void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&)::{lambda()#1}::operator()() const /usr/include/c++/7/mutex:672 (gdb+0x73300d)
    #16 std::call_once<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(std::once_flag&, void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&)::{lambda()#2}::operator()() const /usr/include/c++/7/mutex:677 (gdb+0x7330b2)
    #17 std::call_once<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(std::once_flag&, void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&)::{lambda()#2}::_FUN() /usr/include/c++/7/mutex:677 (gdb+0x7330f2)
    #18 pthread_once <null> (libtsan.so.0+0x4457c)
    #19 __gthread_once /usr/include/c++/7/x86_64-suse-linux/bits/gthr-default.h:699 (gdb+0x72f5dd)
    #20 void std::call_once<void (std::__future_base::_State_baseV2::*)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*>(std::once_flag&, void (std::__future_base::_State_baseV2::*&&)(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*, bool*), std::__future_base::_State_baseV2*&&, std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>*&&, bool*&&) /usr/include/c++/7/mutex:684 (gdb+0x733224)
    #21 std::__future_base::_State_baseV2::_M_set_result(std::function<std::unique_ptr<std::__future_base::_Result_base, std::__future_base::_Result_base::_Deleter> ()>, bool) /usr/include/c++/7/future:401 (gdb+0x732852)
    #22 std::__future_base::_Task_state<std::function<void ()>, std::allocator<int>, void ()>::_M_run() /usr/include/c++/7/future:1423 (gdb+0x737bef)
    #23 std::packaged_task<void ()>::operator()() /usr/include/c++/7/future:1556 (gdb+0x1dac492)
    #24 gdb::thread_pool::thread_function() gdbsupport/thread-pool.cc:242 (gdb+0x1dabdb4)
    #25 void std::__invoke_impl<void, void (gdb::thread_pool::*)(), gdb::thread_pool*>(std::__invoke_memfun_deref, void (gdb::thread_pool::*&&)(), gdb::thread_pool*&&) /usr/include/c++/7/bits/invoke.h:73 (gdb+0x1dace63)
    #26 std::__invoke_result<void (gdb::thread_pool::*)(), gdb::thread_pool*>::type std::__invoke<void (gdb::thread_pool::*)(), gdb::thread_pool*>(void (gdb::thread_pool::*&&)(), gdb::thread_pool*&&) /usr/include/c++/7/bits/invoke.h:95 (gdb+0x1dac294)
    #27 decltype (__invoke((_S_declval<0ul>)(), (_S_declval<1ul>)())) std::thread::_Invoker<std::tuple<void (gdb::thread_pool::*)(), gdb::thread_pool*> >::_M_invoke<0ul, 1ul>(std::_Index_tuple<0ul, 1ul>) /usr/include/c++/7/thread:234 (gdb+0x1daf5c6)
    #28 std::thread::_Invoker<std::tuple<void (gdb::thread_pool::*)(), gdb::thread_pool*> >::operator()() /usr/include/c++/7/thread:243 (gdb+0x1daf551)
    #29 std::thread::_State_impl<std::thread::_Invoker<std::tuple<void (gdb::thread_pool::*)(), gdb::thread_pool*> > >::_M_run() /usr/include/c++/7/thread:186 (gdb+0x1daf506)
    #30 <null> <null> (libstdc++.so.6+0xdcac2)

  Location is global 'global_index_cache' of size 48 at 0x00000321f520 (gdb+0x00000321f540)
  ...
SUMMARY: ThreadSanitizer: data race gdb/dwarf2/index-cache.c:76 in index_cache::enable()
...

The race happens when issuing a "file $exec" command followed by a
"set index-cache enabled on" command.

The race is between:
- a worker thread reading index_cache::m_enabled to determine whether an
  index-cache entry for $exec needs to be written
  (due to command "file $exec"), and
- the main thread setting index_cache::m_enabled
  (due to command "set index-cache enabled on").

Fix this by capturing the value of index_cache::m_enabled in the main thread,
and using the captured value in the worker thread.

Tested on x86_64-linux.

PR symtab/30392
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30392

15 months agoppc: sanity check writing relocs
Alan Modra [Fri, 4 Aug 2023 08:07:12 +0000 (17:37 +0930)]
ppc: sanity check writing relocs

Check for output buffer overruns.

* elf32-ppc.c (swap_reloc_out, count_and_swap_reloc_out): New
functions.  Use throughout file.
* elf64-ppc.c (swap_reloc_out, count_and_swap_reloc_out): Likewise.

15 months agoPR30697, ppc32 mix of local-dynamic and global-dynamic TLS
Alan Modra [Fri, 4 Aug 2023 05:39:53 +0000 (15:09 +0930)]
PR30697, ppc32 mix of local-dynamic and global-dynamic TLS

This fixes miscounting of dynamic relocations on GOT entries when
a) there are both local-dynamic and global-dynamic tls accesss for a
   given symbol, and
b) the symbol is global with non-default visibility, and
c) the __tls_get_addr calls aren't optimised away.

PR 30697
bfd/
* elf32-ppc.c (allocate_dynrelocs): Correct local-dynamic
reloc count.
ld/
* testsuite/ld-powerpc/tls32ldgd.d,
* testsuite/ld-powerpc/tls32ldgd.s: New test.
* testsuite/ld-powerpc/powerpc.exp: Run it.

15 months agogdb/testsuite: Disable gdb.compile when testing with clang
Bruno Larsen [Fri, 14 Jul 2023 08:06:03 +0000 (10:06 +0200)]
gdb/testsuite: Disable gdb.compile when testing with clang

Attempting to test the gdb.compile with clang as the compiler results in
over 300 unexpected errors, due to a segmentation fault and several
handshake failures. Since the whole feature is designed around a gcc
plugin, and even the gcc testing is shaky at best, this commit restricts
those tests to only running under gcc. If that gets fixed, this commit
can be reverted.

Approved-By: Tom Tromey <tom@tromey.com>
15 months agoAutomatic date update in version.in
GDB Administrator [Fri, 4 Aug 2023 00:00:29 +0000 (00:00 +0000)]
Automatic date update in version.in

15 months ago[gdb/symtab] Remove superfluous handling of Ada main in write_cooked_index
Tom de Vries [Thu, 3 Aug 2023 16:35:23 +0000 (18:35 +0200)]
[gdb/symtab] Remove superfluous handling of Ada main in write_cooked_index

I filed PR29179 about the following FAIL in test-case
gdb.ada/O2_float_param.exp with target board cc-with-gdb-index:
...
(gdb) break increment^M
Function "increment" not defined.^M
Make breakpoint pending on future shared library load? (y or [n]) n^M
(gdb) FAIL: gdb.ada/O2_float_param.exp: scenario=all: gdb_breakpoint: \
  set breakpoint at increment
...

The FAIL was a regression since commit 2cf349be0e3 ("Do not put linkage names
into .gdb_index").

Before that commit we had:
...
$ readelf -w foo > READELF
$ grep callee.*increment READELF
[1568] callee__increment: 5 [global, function]
[3115] callee.increment: 5 [global, function]
...
but after only:
...
$ grep callee.*increment READELF
[3115] callee.increment: 5 [global, function]
...

The regression was fixed by commit 67e83a0deef ("Fix regression in
c-linkage-name.exp with gdb index"), which got us again:
...
$ grep callee.*increment READELF
[1568] callee__increment: 5 [global, function]
[3115] callee.increment: 5 [global, function]
...

The commit however did not claim that particular PR.  A subsequent commit,
commit 5fea9794325 ("Improve Ada support in .gdb_index") did claim to fix it,
together with commit dd05fc7071a ("Change .gdb_index de-duplication
implementation").

The commit 5fea9794325 contained the following addition in write_cooked_index:
...
+      if (entry->per_cu->lang () == language_ada)
+ {
+   /* We want to ensure that the Ada main function's name
+      appears verbatim in the index.  However, this name will
+      be of the form "_ada_mumble", and will be rewritten by
+      ada_decode.  So, recognize it specially here and add it
+      to the index by hand.  */
+   if (entry->tag == DW_TAG_subprogram
+       && strcmp (main_for_ada, name) == 0)
+     {
+       /* Leave it alone.  */
+     }
+   else
+     {
+       /* In order for the index to work when read back into
+  gdb, it has to use the encoded name, with any
+  suffixes stripped.  */
+       std::string encoded = ada_encode (name, false);
+       name = obstack_strdup (&symtab->m_string_obstack,
+      encoded.c_str ());
+     }
+ }
...

The code contains some special handling related to the Ada main function, so
let's look at that one: foo.  Before commit 67e83a0deef we have:
...
$ grep foo.*function READELF
[3733] foo: 7 [global, function]
...
and after:
...
$ grep foo.*function READELF
[2738] _ada_foo: 7 [global, function]
[3733] foo: 7 [global, function]
...
so that looks identical to the callee.increment case.

At commit 5fea9794325, we have slightly different index numbers:
...
$ grep foo.*function READELF
[1658] foo: 7 [global, function]
[2738] _ada_foo: 7 [global, function]
...
but otherwise the same result.

If we disable the special handling of the Ada main function like so:
...
-   if (entry->tag == DW_TAG_subprogram
+   if (false && entry->tag == DW_TAG_subprogram
...
we still have the exact same result because:
...
(gdb) p main_for_ada
$1 = 0x352e6a0 "_ada_foo"
...
and ada_encode ("_ada_foo", false) == "_ada_foo".

The comment seems to be copied from debug_names::insert, which does indeed use
ada_decode, while the code in write_cooked_index uses ada_encode instead.

Remove the superfluous special handling of Ada main in write_cooked_index.

Tested on x86_64-linux, with target boards unix and cc-with-gdb-index.

Approved-By: Tom Tromey <tom@tromey.com>
15 months agoRemove f-string from DAP
Tom Tromey [Thu, 3 Aug 2023 16:25:18 +0000 (10:25 -0600)]
Remove f-string from DAP

One more f-string snuck into the DAP code, in breakpoint.py.  Most of
them were removed here:

    https://sourceware.org/pipermail/gdb-patches/2023-June/200023.html

but I think this one landed after that patch.

While DAP only supports Python 3.5 and later, f-strings were added in
3.6, so remove this.

Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30708

15 months agoUse frame.name() in FrameDecorator
Tom Tromey [Tue, 1 Aug 2023 19:57:19 +0000 (13:57 -0600)]
Use frame.name() in FrameDecorator

A co-worker pointed out that gdb's DAP implementation might return an
integer for the name of a stack frame, like:

    {"id": 1, "name": 93824992310799, ...}

This can be seen currently in the logs of the bt-nodebug.exp test
case.

What is happening is that FrameDecorator falls back on returning the
PC when the frame's function symbol cannot be found, relying on the
gdb core to look up the minsym and print its name.

This can actually yield the wrong answer sometimes, because it falls
into the get_frame_pc / get_frame_address_in_block problem -- if the
frame is at a call to a noreturn function, the PC in this case might
appear to be in the next function in memory.  For more on this, see:

    https://sourceware.org/bugzilla/show_bug.cgi?id=8416

and related bugs.

However, there's a different approach we can take: the code here can
simply use Frame.name.  This handles the PC problem correctly, and
gets us the information we need.

15 months agogdb: avoid double stop after failed breakpoint condition check
Andrew Burgess [Wed, 12 Jul 2023 20:56:50 +0000 (21:56 +0100)]
gdb: avoid double stop after failed breakpoint condition check

This commit replaces this earlier commit:

  commit 2e411b8c68eb2b035b31d5b00d940d4be1a0928b
  Date:   Fri Oct 14 14:53:15 2022 +0100

      gdb: don't always print breakpoint location after failed condition check

and is a result of feedback received here[1].

The original commit addressed a problem where, if a breakpoint
condition included an inferior function call, and if the inferior
function call failed, then GDB would announce the stop twice.  Here's
an example of GDB's output before the above commit that shows the
problem being addressed:

  (gdb) break foo if (some_func ())
  Breakpoint 1 at 0x40111e: file bpcond.c, line 11.
  (gdb) r
  Starting program: /tmp/bpcond

  Program received signal SIGSEGV, Segmentation fault.
  0x0000000000401116 in some_func () at bpcond.c:5
  5       return *p;
  Error in testing condition for breakpoint 1:
  The program being debugged stopped while in a function called from GDB.
  Evaluation of the expression containing the function
  (some_func) will be abandoned.
  When the function is done executing, GDB will silently stop.

  Breakpoint 1, 0x0000000000401116 in some_func () at bpcond.c:5
  5       return *p;
  (gdb)

The original commit addressed this issue in breakpoint.c, by spotting
that the $pc had changed while evaluating the breakpoint condition,
and inferring from this that GDB must have stopped elsewhere.

However, the way in which the original commit suppressed the second
stop announcement was to set bpstat::print to true -- this tells GDB
not to print the frame during the stop announcement, and for the CLI
this is fine, however, it was pointed out that for the MI this still
isn't really enough.  Below is an example from an MI session after the
above commit was applied, this shows the problem with the above
commit:

  -break-insert -c "cond_fail()" foo
  ^done,bkpt={number="1",type="breakpoint",disp="keep",enabled="y",addr="0x000000000040111e",func="foo",file="/tmp/mi-condbreak-fail.c",line="30",thread-groups=["i1"],cond="cond_fail()",times="0",original-location="foo"}
  (gdb)
  -exec-run
  =thread-group-started,id="i1",pid="2636270"
  =thread-created,id="1",group-id="i1"
  =library-loaded,id="/lib64/ld-linux-x86-64.so.2",target-name="/lib64/ld-linux-x86-64.so.2",host-name="/lib64/ld-linux-x86-64.so.2",symbols-loaded="0",thread-group="i1",ranges=[{from="0x00007ffff7fd3110",to="0x00007ffff7ff2bb4"}]
  ^running
  *running,thread-id="all"
  (gdb)
  =library-loaded,id="/lib64/libm.so.6",target-name="/lib64/libm.so.6",host-name="/lib64/libm.so.6",symbols-loaded="0",thread-group="i1",ranges=[{from="0x00007ffff7e59390",to="0x00007ffff7ef4f98"}]
  =library-loaded,id="/lib64/libc.so.6",target-name="/lib64/libc.so.6",host-name="/lib64/libc.so.6",symbols-loaded="0",thread-group="i1",ranges=[{from="0x00007ffff7ca66b0",to="0x00007ffff7df3c5f"}]
  ~"\nProgram"
  ~" received signal SIGSEGV, Segmentation fault.\n"
  ~"0x0000000000401116 in cond_fail () at /tmp/mi-condbreak-fail.c:24\n"
  ~"24\t  return *p;\t\t\t/* Crash here.  */\n"
  *stopped,reason="signal-received",signal-name="SIGSEGV",signal-meaning="Segmentation fault",frame={addr="0x0000000000401116",func="cond_fail",args=[],file="/tmp/mi-condbreak-fail.c",fullname="/tmp/mi-condbreak-fail.c",line="24",arch="i386:x86-64"},thread-id="1",stopped-threads="all",core="9"
  &"Error in testing condition for breakpoint 1:\n"
  &"The program being debugged was signaled while in a function called from GDB.\n"
  &"GDB remains in the frame where the signal was received.\n"
  &"To change this behavior use \"set unwindonsignal on\".\n"
  &"Evaluation of the expression containing the function\n"
  &"(cond_fail) will be abandoned.\n"
  &"When the function is done executing, GDB will silently stop.\n"
  =breakpoint-modified,bkpt={number="1",type="breakpoint",disp="keep",enabled="y",addr="0x000000000040111e",func="foo",file="/tmp/mi-condbreak-fail.c",fullname="/tmp/mi-condbreak-fail.c",line="30",thread-groups=["i1"],cond="cond_fail()",times="1",original-location="foo"}
  *stopped
  (gdb)

Notice that we still see two '*stopped' lines, the first includes the
full frame information, while the second has no frame information,
this is a result of bpstat::print having been set.  Ideally, the
second '*stopped' line should not be present.

By setting bpstat::print I was addressing the problem too late, this
flag really only changes how interp::on_normal_stop prints the stop
event, and interp::on_normal_stop is called (indirectly) from the
normal_stop function in infrun.c.  A better solution is to avoid
calling normal_stop at all for the stops which should not be reported
to the user, and this is what I do in this commit.

This commit has 3 parts:

  1. In breakpoint.c, revert the above commit,

  2. In fetch_inferior_event (infrun.c), capture the stop-id before
  calling handle_inferior_event.  If, after calling
  handle_inferior_event, the stop-id has changed, then this indicates
  that somewhere within handle_inferior_event, a stop was announced to
  the user.  If this is the case then GDB should not call normal_stop,
  and we should rely on whoever announced the stop to ensure that we
  are in a PROMPT_NEEDED state, which means the prompt will be
  displayed once fetch_inferior_event returns.  And,

  3. In infcall.c, do two things:

     (a) In run_inferior_call, after making the inferior call, ensure
     that either async_disable_stdin or async_enable_stdin is called
     to put the prompt state, and stdin handling into the correct
     state based on whether the inferior call completed successfully
     or not, and

     (b) In call_thread_fsm::should_stop, call async_enable_stdin
     rather than changing the prompt state directly.  This isn't
     strictly necessary, but helped me understand this code more.
     This async_enable_stdin call is only reached if normal_stop is
     not going to be called, and replaces the async_enable_stdin call
     that exists in normal_stop.  Though we could just adjust the
     prompt state if felt (to me) much easier to understand when I
     could see this call and the corresponding call in normal_stop.

With these changes in place now, when the inferior call (from the
breakpoint condition) fails, infcall.c leaves the prompt state as
PROMPT_NEEDED, and leaves stdin registered with the event loop.

Back in fetch_inferior_event GDB notices that the stop-id has changed
and so avoids calling normal_stop.

And on return from fetch_inferior_event GDB will display the prompt
and handle input from stdin.

As normal_stop is not called the MI problem is solved, and the test
added in the earlier mentioned commit still passes just fine, so the
CLI has not regressed.

[1] https://inbox.sourceware.org/gdb-patches/6fd4aa13-6003-2563-5841-e80d5a55d59e@palves.net/

15 months agoRemove PEI_HEADERS define
Tom Tromey [Sun, 30 Jul 2023 16:26:07 +0000 (10:26 -0600)]
Remove PEI_HEADERS define

I noticed a few files double-included libcoff.h, and digging deeper I
found that the PEI_HEADERS define is a sort of external include guard.

This patch adds include guards to the few files in include/coff that
were missing one, and then removes the PEI_HEADERS workaround and the
redundant includes.

I didn't see anything in these files that indicated that
double-inclusion would be useful, so it seems to me that this approach
is ok.

Tested by rebuilding with --enable-targets=all.

2023-08-02  Tom Tromey  <tromey@adacore.com>

* pei-x86_64.c (PEI_HEADERS): Do not define.
* pei-loongarch64.c (PEI_HEADERS): Do not define.
* pei-aarch64.c (PEI_HEADERS): Do not define.
* pe-x86_64.c (PEI_HEADERS): Do not define.
* pe-aarch64.c (PEI_HEADERS): Do not define.
* libpei.h (_LIBPEI_H): Add include guard.
* coff-x86_64.c (PEI_HEADERS): Do not check.
* coff-loongarch64.c (PEI_HEADERS): Do not check.
* coff-aarch64.c (PEI_HEADERS): Do not check.

include/ChangeLog
2023-08-02  Tom Tromey  <tromey@adacore.com>

* coff/x86_64.h (COFF_X86_64_H): Add include guard.
* coff/loongarch64.h (COFF_LOONGARCH64_H): Add include guard.
* coff/aarch64.h (COFF_AARCH64_H): Add include guard.

15 months agoreadelf sprintf optimisation
Alan Modra [Wed, 2 Aug 2023 01:57:27 +0000 (11:27 +0930)]
readelf sprintf optimisation

This replaces sprintf and strcat calls with stpcpy, and makes use of
sprintf return value rather than using strlen, for get_machine_flags.

decode_NDS32_machine_flags made use of snprintf, which is arguably the
"correct" way to do things if there can be a buffer overflow.  In this
case I don't think there can be, the buffer is 1k in size which is at
least 5 times more than needed.  What's more, snprintf returns the
count of chars that would be output given no buffer limit, which means
code like
  r += snprintf (buf + r, size - r, ...);
  r += snprintf (buf + r, size - r, ...);
is just wrong.  There needs to be a check on the return value in order
to prevent buf + r being out of bounds for the second snprintf call.

BTW, if you look closely you'll see the return value of the decode
functions is unused.  I admit to getting a little carried away with
writing "out = stpcpy (out, ...):" in each of the decode functions and
didn't notice that until get_machine_flags was trimmed down to a much
smaller size.  When I did notice, I decided it's not such a bad thing.

* readelf.c (decode_ARC_machine_flags, decode_ARM_machine_flags),
(decode_AVR_machine_flags, decode_NDS32_machine_flags),
(decode_AMDGPU_machine_flags): Use stpcpy and sprintf return
value.  Return end of string.
(decode_BLACKFIN_machine_flags, decode_FRV_machine_flags),
(decode_IA64_machine_flags, decode_LOONGARCH_machine_flags),
(decode_M68K_machine_flags, decode_MeP_machine_flags),
(decode_MIPS_machine_flags, decode_MSP430_machine_flags),
(decode_PARISC_machine_flags, decode_RISCV_machine_flags),
(decode_RL78_machine_flags, decode_RX_machine_flags),
(decode_SH_machine_flags, decode_SPARC_machine_flags),
(decode_V800_machine_flags, decode_V850_machine_flags),
(decode_Z80_machine_flags): New functions, split out from..
(get_machine_flags): ..here.  Similarly use stpcpy.

15 months agobinutils sprintf optimisation
Alan Modra [Thu, 3 Aug 2023 03:02:40 +0000 (12:32 +0930)]
binutils sprintf optimisation

Avoid the use of sprintf with a "%s" format string, replacing with
strcpy or stpcpy.  Use sprintf return value rather than a later
strlen.  Don't use strcat where we can keep track of the end of a
string output buffer.

* dlltool.c (look_for_prog): memcpy prefix and strcpy prog_name.
* dllwrap.c (look_for_prog): Likewise.
* resrc.c (look_for_default): Likewise.  Add quotes with memmove
rather than allocating another buffer.
* size.c (size_number): Use sprintf return value.
* stabs.c (parse_stab_argtypes): Likewise.
* windmc.c (write_bin): Likewes, and use stpcpy.
* wrstabs.c: Similarly throughout.

15 months agocris: sprintf optimisation
Alan Modra [Thu, 3 Aug 2023 02:56:46 +0000 (12:26 +0930)]
cris: sprintf optimisation

Since I was poking at cris-dis.c to avoid the sanitizer warning,
I figure I might as well make use of stpcpy and sprintf return value
in other places in this file.

* cris-dis.c (format_hex): Use sprintf return value.
(format_reg): Use stpcpy and sprintf return, avoiding strlen.
(format_sup_reg): Likewise.

15 months agoarm: sanitizer stringop-overflow
Alan Modra [Wed, 2 Aug 2023 23:30:05 +0000 (09:00 +0930)]
arm: sanitizer stringop-overflow

In function 'memset',
    inlined from 'create_unwind_entry' at /home/alan/src/binutils-gdb/gas/config/tc-arm.c:27881:3:
/usr/include/bits/string_fortified.h:59:10: error: '__builtin_memset' specified size between 2147483652 and 4294967295 exceeds maximum object size 2147483647 [-Werror=stringop-overflow=]
   59 |   return __builtin___memset_chk (__dest, __ch, __len,
      |          ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   60 |                                  __glibc_objsize0 (__dest));
      |                                  ~~~~~~~~~~~~~~~~~~~~~~~~~~

* config/tc-arm.c (create_unwind_entry): Return after bad size,
and bad opcode count.

15 months agoxtensa: sprintf sanitizer null destination pointer
Alan Modra [Wed, 2 Aug 2023 23:26:24 +0000 (08:56 +0930)]
xtensa: sprintf sanitizer null destination pointer

* config/tc-xtensa.c (xtensa_add_config_info): Use auto buffer
rather than malloc.  Use sprintf return value.

15 months agold: sprintf sanitizer null destination pointer
Alan Modra [Wed, 2 Aug 2023 23:10:12 +0000 (08:40 +0930)]
ld: sprintf sanitizer null destination pointer

* configure.ac (stpcpy): AC_CHECK_DECLS.
* sysdep.h (stpcpy): Add fallback declaraion.
* config.in: Regenerate.
* configure: Regenerate.
* emultempl/pe.em (open_dynamic_archive): Use
stpcpy rather than sprintf plus strlen.
* emultempl/pep.em (open_dynamic_archive): Likewise.
* emultempl/xtensaelf.em (elf_xtensa_before_allocation): Use
auto rather than malloc'd buffer.  Use sprintf count.
* ldelf.c (ldelf_search_needed): Use memcpy in place of sprintf.
* pe-dll.c (pe_process_import_defs): Use string already formed
for alias match rather than recreating.

15 months agogprof: sprintf sanitizer null destination pointer
Alan Modra [Wed, 2 Aug 2023 22:53:51 +0000 (08:23 +0930)]
gprof: sprintf sanitizer null destination pointer

* basic_blocks.c (annotate_with_count): Use output of sprintf
rather than strlen.

15 months agoresrc: sprintf sanitizer null destination pointer
Alan Modra [Wed, 2 Aug 2023 22:48:13 +0000 (08:18 +0930)]
resrc: sprintf sanitizer null destination pointer

* resrc.c (read_rc_file): Use stpcpy rather than sprintf
followed by strlen.  Tidy.

15 months agodlltool: sprintf sanitizer null destination pointer
Alan Modra [Wed, 2 Aug 2023 22:45:12 +0000 (08:15 +0930)]
dlltool: sprintf sanitizer null destination pointer

* dlltool.c (gen_lib_file): Avoid bogus sanitizer error.

15 months agowrstabs: sprintf sanitizer null destination pointer
Alan Modra [Wed, 2 Aug 2023 22:37:54 +0000 (08:07 +0930)]
wrstabs: sprintf sanitizer null destination pointer

gcc-2.12 seems to be ignoring __attribute__((__returns_nonnull__))
on xmalloc.

* wrstabs.c (stab_method_type): Use stpcpy rather than sprintf
or strcat.

15 months agocris: sprintf sanitizer null destination pointer
Alan Modra [Wed, 2 Aug 2023 22:29:47 +0000 (07:59 +0930)]
cris: sprintf sanitizer null destination pointer

Simplify the sprintf calls, and use sprintf return value.  Older code
in binutils avoided using the sprintf return count of chars printed,
because with some older C libraries it wasn't reliable.  Nowadays it
should be OK to use (and we already use the return value elsewhere).
sprintf can't return an error status of -1 here.

* cris-dis.c (format_dec): Avoid sanitizer warning.  Use sprintf
return value rather than calling strlen.

15 months agoobjdump, nm: sprintf sanitizer null destination pointer
Alan Modra [Wed, 2 Aug 2023 21:59:58 +0000 (07:29 +0930)]
objdump, nm: sprintf sanitizer null destination pointer

Seen on Ubuntu 23.04 x86_64-linux using gcc-12.2 and gcc-12.3 with
CFLAGS="-m32 -g -O2 -fsanitize=address,undefined".

  CC       objdump.o
In file included from /usr/include/stdio.h:906,
                 from /home/alan/src/binutils-gdb/binutils/sysdep.h:24,
                 from /home/alan/src/binutils-gdb/binutils/objdump.c:51:
In function 'sprintf',
    inlined from 'display_utf8' at /home/alan/src/binutils-gdb/binutils/objdump.c:621:14,
    inlined from 'sanitize_string.part.0' at /home/alan/src/binutils-gdb/binutils/objdump.c:742:11:
/usr/include/bits/stdio2.h:30:10: error: null destination pointer [-Werror=format-overflow=]
   30 |   return __builtin___sprintf_chk (__s, __USE_FORTIFY_LEVEL - 1,
      |          ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   31 |                                   __glibc_objsize (__s), __fmt,
      |                                   ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   32 |                                   __va_arg_pack ());
      |                                   ~~~~~~~~~~~~~~~~~
cc1: all warnings being treated as errors

The warning is bogus of course.  xmalloc is guaranteed to return
non-NULL, but apparently this isn't seen in display_utf6.  The same
doesn't happen with -m64, maybe due to inlining differences, I haven't
investigated fully.  Easily avoided as we hardly need to use sprintf
for a single char, or a two char string.

* objdump.c (display_utf8): Avoid bogus sprintf sanitizer warning.
Use hex ESC to switch back to default colour.
(sanitize_string): Comment.  Bump buffer size by one.  Fix overlong
line.
* nm.c (display_utf8, sanitize_string): As above.

15 months agogdb: fix possible nullptr dereference in a remote_debug_printf call
Andrew Burgess [Tue, 18 Jul 2023 12:52:20 +0000 (13:52 +0100)]
gdb: fix possible nullptr dereference in a remote_debug_printf call

While working on another patch I triggered a segfault from within the
function remote_target::discard_pending_stop_replies.  Turns out this
was caused by a cut&paste error introduced in this commit:

  commit df5ad102009c41ab4dfadbb8cfb8c8b2a02a4f78
  Date:   Wed Dec 1 09:40:03 2021 -0500

      gdb, gdbserver: detach fork child when detaching from fork parent

This commit adds a remote_debug_printf call that was copied from
earlier in the function, however, the new call wasn't updated to use
the appropriate local variable.  The local variable that it is using
might be nullptr, in which case we trigger undefined behaviour, and
could crash, which is what I was seeing.

Fixed by updating to use the correct local variable.

15 months ago Fix Wlto-type-mismatch in opcodes/ft32-dis.c
Tom de Vries [Thu, 3 Aug 2023 08:48:39 +0000 (09:48 +0100)]
 Fix Wlto-type-mismatch in opcodes/ft32-dis.c

15 months agoRISC-V: Add support for 'Zvfh' and 'Zvfhmin'
Tsukasa OI [Thu, 3 Aug 2023 05:35:53 +0000 (05:35 +0000)]
RISC-V: Add support for 'Zvfh' and 'Zvfhmin'

This commit adds support for recently ratified vector FP16 extensions:
'Zvfh' and 'Zvfhmin'.

This is based on:
<https://github.com/riscv/riscv-v-spec/blob/master/v-spec.adoc#zvfhmin-vector-extension-for-minimal-half-precision-floating-point>
<https://github.com/riscv/riscv-v-spec/blob/master/v-spec.adoc#zvfh-vector-extension-for-half-precision-floating-point>

Despite not having any new instructions, it will be necessary since those
extensions are already implemented in GCC.

Note that however, in this commit, following dependencies are implemented.

1.  'Zvfhmin' -> 'Zve32f'
2.  'Zvfh' -> 'Zvfhmin' (not 'Zvfh' -> 'Zve32f' as in the documentation)
3.  'Zvfh' -> 'Zfhmin'

This is because the instructions and configurations supported by the
'Zvfh' extension is a strict superset of the 'Zvfhmin' extension and
'Zvfh' -> 'Zve32f' dependency is indirectly derived from that fact.

bfd/ChangeLog:

* elfxx-riscv.c (riscv_implicit_subsets): Add implications
related to 'Zvfh' and 'Zvfhmin' extensions.
(riscv_supported_std_z_ext) Add 'Zvfh' and 'Zvfhmin' to the list.

15 months agoRISC-V: Imply 'Zicsr' from 'Zve32x'
Tsukasa OI [Wed, 2 Aug 2023 23:50:27 +0000 (23:50 +0000)]
RISC-V: Imply 'Zicsr' from 'Zve32x'

Further clarification is made so that 'Zve32x' implies 'Zicsr' (the same
implication is already implemented in LLVM).

See related issue (the author raised) on the vector specification:
<https://github.com/riscv/riscv-v-spec/issues/908>
and its resolution:
<https://github.com/riscv/riscv-v-spec/issues/909>

bfd/ChangeLog:

* elfxx-riscv.c (riscv_implicit_subsets): Add 'Zve32x' -> 'Zicsr'.

15 months agoAutomatic date update in version.in
GDB Administrator [Thu, 3 Aug 2023 00:00:40 +0000 (00:00 +0000)]
Automatic date update in version.in

15 months ago[gdb] Initialize main_thread_id earlier
Tom de Vries [Wed, 2 Aug 2023 21:17:35 +0000 (23:17 +0200)]
[gdb] Initialize main_thread_id earlier

I wrote a patch using is_main_thread (), and found it returning false in the
main thread due to main_thread_id not being initialized yet.

Initialization currently takes place in _initialize_run_on_main_thread, but
that's too late for earlier uses.

Fix this by initializing, either:
- when entering main, or
- on an earlier first use.

Tested on x86_64-linux.

Approved-By: Tom Tromey <tom@tromey.com>
15 months ago[gdb/dap] Disable DAP for python <= 3.5
Tom de Vries [Wed, 2 Aug 2023 21:14:58 +0000 (23:14 +0200)]
[gdb/dap] Disable DAP for python <= 3.5

DAP requires python module typing, which is supported starting python 3.5.

Make this formal by:
- disabling the dap interpreter for python version < 3.5
- returning 0 in allow_dap_tests for python version < 3.5

Approved-By: Tom Tromey <tom@tromey.com>
PR dap/30708
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30708

15 months agoAvoid failures in fixed_points.exp with older GCC
Tom Tromey [Tue, 1 Aug 2023 14:33:42 +0000 (08:33 -0600)]
Avoid failures in fixed_points.exp with older GCC

Tom de Vries pointed out that my recent change to fixed_points.exp
failed with older versions of GCC.  This patch fixes the problem by
skipping the new test in this situation.

15 months agoRevert "2.41 Release sources"
Sam James [Wed, 2 Aug 2023 11:06:23 +0000 (12:06 +0100)]
Revert "2.41 Release sources"

This reverts commit 675b9d612cc59446e84e2c6d89b45500cb603a8d.

See https://sourceware.org/pipermail/binutils/2023-August/128761.html.

15 months ago2.41 Release sources
Nick Clifton [Wed, 2 Aug 2023 08:23:36 +0000 (09:23 +0100)]
2.41 Release sources

15 months agogprofng: Fix build with 64bit file offset on 32bit machines
Khem Raj [Mon, 31 Jul 2023 22:31:40 +0000 (15:31 -0700)]
gprofng: Fix build with 64bit file offset on 32bit machines

gprofng/ChangeLog
2023-07-31  Khem Raj <raj.khem@gmail.com>

* libcollector/iotrace.c: Define open64, fgetpos64, and fsetpos64
  only when __USE_LARGEFILE64 and __USE_FILE_OFFSET64 are not
  defined.

15 months agoAutomatic date update in version.in
GDB Administrator [Wed, 2 Aug 2023 00:00:29 +0000 (00:00 +0000)]
Automatic date update in version.in

15 months agoDon't declare xmalloc and others in ldmisc.h
Alan Modra [Tue, 1 Aug 2023 22:24:47 +0000 (07:54 +0930)]
Don't declare xmalloc and others in ldmisc.h

* ldmisc.h (xmalloc, xrealloc, xexit, yyerror): Don't declare.
* emultempl/pdp11.em: Include libiberty.h.
* emultempl/ticoff.em: Likewise.
* emultempl/vms.em: Likewise.
* ldctor.c: Likewise.
* ldelfgen.c: Likewise.
* ldgram.y: Likewise.
(yyerror): Prototype and make static.

15 months agoDon't declare xmalloc or xrealloc in bucomm.h
Alan Modra [Tue, 1 Aug 2023 22:23:26 +0000 (07:53 +0930)]
Don't declare xmalloc or xrealloc in bucomm.h

It's better to include the proper header, which has declarations with
various attributes.  Commit 096aefc040 in 1994 introduced this wart.

* bucomm.h (xmalloc, xrealloc): Delete declaration.
* od-macho.c: Include libiberty.h.
* od-xcoff.c: Include libiberty.h.

15 months agoRegen ld/configure
Alan Modra [Tue, 1 Aug 2023 22:13:50 +0000 (07:43 +0930)]
Regen ld/configure

For commit 3d05c80b5dc4.

15 months agoImplement DAP "source" request
Tom Tromey [Thu, 27 Jul 2023 20:27:05 +0000 (14:27 -0600)]
Implement DAP "source" request

This implements the DAP "source" request.  I renamed the
"loadedSources" function from "sources" to "loaded_sources" to avoid
any confusion.  I also moved the loadedSources test to the new
sources.exp.

Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30691

15 months agoHandle Source in DAP breakpointLocations
Tom Tromey [Thu, 27 Jul 2023 20:21:13 +0000 (14:21 -0600)]
Handle Source in DAP breakpointLocations

This changes the DAP breakpointLocations request to accept a Source
and to decode it properly.

15 months agoIntroduce sourceReference handling in DAP
Tom Tromey [Thu, 27 Jul 2023 20:04:31 +0000 (14:04 -0600)]
Introduce sourceReference handling in DAP

This changes the gdb DAP implementation to emit a real
sourceReference, rather than emitting 0.  Sources are tracked in some
maps in sources.py, and a new helper function is introduced to compute
the "Source" object that can be sent to the client.

15 months agoDon't supply DAP 'path' for non-file shared libraries
Tom Tromey [Tue, 25 Jul 2023 16:55:14 +0000 (10:55 -0600)]
Don't supply DAP 'path' for non-file shared libraries

The DAP 'module' event may include a 'path' component.  I noticed that
this is supplied even when the module in question does not come from a
file.

This patch only emits this field when the objfile corresponds to a
real file.

No test case, because I wasn't sure how to write a portable one.
However, it's clear from gdb.log on Linux:

{"type": "event", "event": "module", "body": {"reason": "new", "module": {"id": "system-supplied DSO at 0x7ffff7fc4000", "name": "system-supplied DSO at 0x7ffff7fc4000"}}, "seq": 21}

Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30676

15 months agoImplement ValueFormat for DAP
Tom Tromey [Tue, 25 Jul 2023 16:13:52 +0000 (10:13 -0600)]
Implement ValueFormat for DAP

This patch implements ValueFormat for DAP.  Currently this only means
supporting "hex".

Note that StackFrameFormat is defined to have many more options, but
none are currently recognized.  It isn't entirely clear how these
should be handled.  I'll file a new gdb bug for this, and perhaps an
upstream DAP bug as well.

New in v2:
- I realized that the "hover" context was broken, and furthermore
  that we only had tests for "hover" failing, not for it succeeding.
  This version fixes the oversight and adds a test.

Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30469

15 months agoRespect supportsMemoryReferences in DAP
Tom Tromey [Thu, 27 Jul 2023 19:06:38 +0000 (13:06 -0600)]
Respect supportsMemoryReferences in DAP

I noticed that the support for memoryReference in the "variables"
output is gated on the client "supportsMemoryReferences" capability.

This patch implements this and makes some other changes to the DAP
memory reference code:

* Remove the memoryReference special case from _SetResult.
  Upstream DAP fixed this oversight in response to
  https://github.com/microsoft/debug-adapter-protocol/issues/414

* Don't use the address of a variable as its memoryReference -- only
  emit this for pointer types.  There's no spec support for the
  previous approach.

* Use strip_typedefs to handle typedefs of pointers.

15 months agoAdd DAP support for C++ exceptions
Tom Tromey [Tue, 25 Jul 2023 16:48:13 +0000 (10:48 -0600)]
Add DAP support for C++ exceptions

This adds DAP support for the various C++ exception-catching
operations.

Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30682

15 months agoImplement DAP 'terminated' event
Tom Tromey [Tue, 25 Jul 2023 15:16:35 +0000 (09:16 -0600)]
Implement DAP 'terminated' event

This implements the DAP 'terminated' event.  Vladimir Makaev noticed
that VSCode will not report the debug session as over unless this is
sent.

It's not completely clear when exactly this event ought to be sent.
Here I've done it when the inferior exits.

Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30681

15 months agoDo not send "new breakpoint" event when breakpoint is set
Tom Tromey [Mon, 24 Jul 2023 19:28:58 +0000 (13:28 -0600)]
Do not send "new breakpoint" event when breakpoint is set

When the DAP client sets a breakpoint, gdb currently sends a "new
breakpoint" event.  However, Vladimir Makaev discovered that this
causes VSCode to think there are two breakpoints.

This patch changes gdb to suppress the event in this case.

Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30678

15 months agoMove DAP breakpoint event code to breakpoint.py
Tom Tromey [Tue, 25 Jul 2023 13:04:45 +0000 (07:04 -0600)]
Move DAP breakpoint event code to breakpoint.py

A subsequent patch will add the ability to suppress breakpoint events
to DAP.  My first attempt at this ended up with recurse imports,
causing Python failures.  So, this patch moves all the DAP breakpoint
event code to breakpoint.py in preparation for the change.

I've renamed breakpoint_descriptor here as well, because it can now be
private to breakpoint.py.

15 months agoFull paths in DAP stackTrace responses
Tom Tromey [Mon, 24 Jul 2023 14:48:00 +0000 (08:48 -0600)]
Full paths in DAP stackTrace responses

Vladimir Makaev noticed that, in some cases, a DAP stackTrace response
would include a relative path name for the "path" component.

This patch changes the frame decorator code to add a new DAP-specific
decorator, and changes the DAP entry point to frame filters to use it.
This decorator prefers the symtab's full name, and does not fall back
to the solib's name.

I'm not entirely happy with this patch, because if a user frame filter
uses FrameDecorator, it may still do the wrong thing.  It would be
better to have frame filters return symtab-like objects instead, or to
have a separate method to return the full path to the source file.

I also tend to think that the solib fallback behavior of
FrameDecorator is a mistake.  If this is ever needed, it seems to me
that it should be a separate method.

Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30665

15 months agoAdd "cwd" parameter to DAP launch request
Tom Tromey [Mon, 24 Jul 2023 15:14:20 +0000 (09:14 -0600)]
Add "cwd" parameter to DAP launch request

This adds the "cwd" parameter to the DAP launch request.

This came up here:
    https://github.com/eclipse-cdt-cloud/cdt-gdb-adapter/issues/90
... and seemed like a good idea.

Reviewed-By: Eli Zaretskii <eliz@gnu.org>
15 months agoRefactor dap_launch
Tom Tromey [Mon, 24 Jul 2023 16:12:17 +0000 (10:12 -0600)]
Refactor dap_launch

This patch refactors dap_launch to make it more extensible and also
easier to use.