Commit c6291d749a introduced a number of errors, found by clang.
elf.c:456:7: error: variable 'alloc_ext_size' is used uninitialized whenever 'if' condition is true [-Werror,-Wsometimes-uninitialized]
if (_bfd_mul_overflow (symcount, extsym_size, &amt))
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
elf.c:464:7: error: variable 'alloc_extshndx_size' is used uninitialized whenever 'if' condition is true [-Werror,-Wsometimes-uninitialized]
if (bfd_seek (ibfd, pos, SEEK_SET) != 0
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
elflink.c:2837:11: error: variable 'alloc1_size' is used uninitialized whenever 'if' condition is true [-Werror,-Wsometimes-uninitialized]
if (internal_relocs == NULL)
^~~~~~~~~~~~~~~~~~~~~~~
elflink.c:12595:16: error: variable 'ext_size' set but not used [-Werror,-Wunused-but-set-variable]
size_t ext_size = 0;
* elf.c (bfd_elf_get_elf_syms): Fix use of uninitialised variables.
* elflink.c (_bfd_elf_link_info_read_relocs): Likewise.
(bfd_elf_final_link): Fix set but not used warning.
If mmap is used without sanity checking, then we'll get a SIGBUS if
an access is done to the mmap'd memory corresponding to a page past
end of file.
* aoutx.h (aout_get_external_symbols): Check that mmap regions
are within file contents. Catch stringsize overflow.
(some_aout_object_p): Don't clear already zeroed fields. Tidy.
* pdp11.c: As for aoutx.h. Copy some fixes too.
To copy input section contents, add _bfd_elf_link_mmap_section_contents
and _bfd_elf_link_munmap_section_contents to mmap in the input sections.
* elf-bfd.h (_bfd_elf_link_mmap_section_contents): New.
(_bfd_elf_link_munmap_section_contents): Likewise.
* elf.c (elf_mmap_section_contents): New.
(_bfd_elf_mmap_section_contents): Use it.
(_bfd_elf_link_mmap_section_contents): New.
(_bfd_elf_link_munmap_section_contents): Likewise.
* elflink.c (elf_link_input_bfd): Call
_bfd_elf_link_mmap_section_contents instead of
bfd_get_full_section_contents. Call
_bfd_elf_link_munmap_section_contents to munmap the section
contents.
(bfd_elf_final_link): When mmap is used, initialize
max_contents_size to _bfd_minimum_mmap_size and increase it
for compressed or linker created sections or sections whose
rawsize != size.
When --no-keep-memory is used, the symbol table and relocation info for
eh_frame are freed after they are retrieved for each text section in the
input object. If an input object has many text sections, the same data
is retrieved and freed many times which can take a very long time.
Update _bfd_elf_gc_mark to keep the symbol table and relocation info for
eh_frame to avoid it. Data to link the 3.5GB clang executable in LLVM
17 debug build on Linux/x86-64 with 32GB RAM is:
before after improvement
user 86.31 86.44 -0.2%
system 8.77 8.63 1.6%
total 95.58 96.81 -1.3%
maximum set(GB) 13.1 13.1 0%
page faults 3024752 3028699 -1.3%
and data to link the 275M cc1plus executable in GCC 14 stage 1 build is:
user 5.49 5.46 -0.5%
system 0.73 0.73 0%
total 6.26 6.25 0.3%
maximum set(MB) 964 964 0%
page faults 235173 235796 -0.3%
The memory usage impact is minimum and the link time of the Rust binary
in
https://sourceware.org/bugzilla/show_bug.cgi?id=31466
is reduced from 500+ seconds to 1.44 seconds, a 300x speedup.
PR ld/31466
* elflink.c (init_reloc_cookie): Add a bool argument, keep_memory,
for keeping memory. Always keep memory if keep_memory is true.
(init_reloc_cookie_rels): Likewise
(init_reloc_cookie_for_section): Add a bool argument for keeping
memory and pass it to init_reloc_cookie and
init_reloc_cookie_rels.
(_bfd_elf_gc_mark_reloc): Pass false to _bfd_elf_gc_mark.
(_bfd_elf_gc_mark): Pass true to init_reloc_cookie_for_section
for the eh_frame section. Pass false to
init_reloc_cookie_for_section for other sections.
(_bfd_elf_gc_mark_extra_sections): Add Add a bool argument for
keeping memory and pass it to _bfd_elf_gc_mark.
(bfd_elf_parse_eh_frame_entries): Pass false to init_reloc_cookie
and init_reloc_cookie_rels.
(bfd_elf_gc_sections): Pass false to init_reloc_cookie_for_section
and _bfd_elf_gc_mark.
(bfd_elf_discard_info): Pass false to
init_reloc_cookie_for_section and init_reloc_cookie.
During a "-j 8" LLVM 17 debug build on a machine with 32GB RAM and 16GB
swap, ld was killed by kernel because of out of memory:
[79437.949336] oom-kill:constraint=CONSTRAINT_NONE,nodemask=(null),cpuset=/,mems_allowed=0,global_oom,task_memcg=/user.slice/user-1000.slice/session-9.scope,task=ld,pid=797431,uid=1000
[79437.949349] Out of memory: Killed process 797431 (ld) total-vm:9219600kB, anon-rss:6558156kB, file-rss:1792kB, shmem-rss:0kB, UID:1000 pgtables:17552kB oom_score_adj:0
Don't cache symbol nor relocation tables if they are mapped in. Data to
link the 3.5GB clang executable in LLVM 17 debug build on Linux/x86-64
with 32GB RAM is:
stdio mmap improvement
user 86.73 87.02 -0.3%
system 9.55 9.21 3.6%
total 100.40 97.66 0.7%
maximum set(GB) 17.34 13.14 24%
page faults 4047667 3042877 25%
and data to link the 275M cc1plus executable in GCC 14 stage 1 build is:
user 5.41 5.44 -0.5%
system 0.80 0.76 5%
total 6.25 6.26 -0.2%
maximum set(MB) 1323 968 27%
page faults 323451 236371 27%
These improve the overall system performance for parallel build by
reducing memory usage and page faults.
Also rename _bfd_link_keep_memory to _bfd_elf_link_keep_memory. Since
the --no-keep-memory linker option causes:
https://sourceware.org/bugzilla/show_bug.cgi?id=31458
this is opt-in by each backend.
bfd/
* elf32-i386.c (elf_i386_scan_relocs): Remove
_bfd_link_keep_memory.
* elf64-x86-64.c (elf_x86_64_scan_relocs): Likewise.
* elflink.c (_bfd_elf_link_keep_memory): New.
(_bfd_elf_link_iterate_on_relocs): Replace _bfd_link_keep_memory
with _bfd_elf_link_keep_memory.
(elf_link_add_object_symbols): Likewise.
(init_reloc_cookie): Likewise.
(init_reloc_cookie_rels): Likewise.
* libbfd-in.h (_bfd_link_keep_memory): Removed.
* linker.c (_bfd_link_keep_memory): Likewise.
* libbfd.h: Regenerated.
Add _bfd_mmap_read_temporary to mmap in symbol tables and relocations
whose sizes >= 4 * page size. For the final link, allocate an external
relocation buffer of 4 * page size to avoid using mmap and munmap on
smaller relocation sections. Since _bfd_mmap_read_temporary allocates
buffer as needed, its callers don't need to.
When mmap is used to map in all ELF sections, data to link the 3.5GB
clang executable in LLVM 17 debug build on Linux/x86-64 with 32GB RAM
is:
stdio mmap improvement
user 84.79 85.27 -0.5%
system 10.95 9.09 17%
total 97.91 94.90 3%
page faults 4837944 4033778 17%
and data to link the 275M cc1plus executable in GCC 14 stage 1 build
is:
user 5.31 5.33 -0.4%
system 0.86 0.76 12%
total 6.19 6.13 1%
page faults 361273 322491 11%
* elf.c (bfd_elf_get_elf_syms): Don't allocate buffer for external
symbol table. Replace bfd_read with _bfd_mmap_read_temporary.
* elflink.c (elf_link_read_relocs_from_section): Add 2 arguments
to return mmap memory address and size.
(_bfd_elf_link_info_read_relocs): Don't allocate buffer for
external relocation information. Replace bfd_read with
_bfd_mmap_read_temporary.
(bfd_elf_final_link): Cache external relocations up to
_bfd_minimum_mmap_size bytes when mmap is used.
* libbfd.c (_bfd_mmap_read_temporary): New.
* libbfd-in.h (_bfd_mmap_read_temporary): Likewise.
* libbfd.h: Regenerated.
There are many linker input files in LLVM debug build with huge string
sections. All these string sections can be treated as read-only. But
linker copies all of them into memory which consumes huge amount of
memory and slows down linker significantly.
Add _bfd_mmap_readonly_persistent and _bfd_mmap_readonly_temporary to
mmap in reado-only sections with size >= 4 * page size.
NB: All string sections in valid ELF inputs must be null terminated.
There is no need to terminate it again and string sections are mmapped
as read-only.
* bfd.c (bfd_mmapped_entry): New.
(bfd_mmapped): Likewise.
(bfd): Add mmapped.
* bfdwin.c (bfd_get_file_window): Use _bfd_pagesize.
* cache.c (cache_bmmap): Remove pagesize_m1 and use pagesize_m1
instead.
* elf.c (bfd_elf_get_str_section): Call
_bfd_mmap_readonly_persistent instead of _bfd_alloc_and_read.
Don't terminate the string section again.
(get_hash_table_data): Call _bfd_mmap_readonly_temporary and
_bfd_munmap_readonly_temporary instead of _bfd_malloc_and_read
and free.
(_bfd_elf_get_dynamic_symbols): Call _bfd_mmap_readonly_persistent
instead of _bfd_alloc_and_read. Don't terminate the string
section again. Call _bfd_mmap_readonly_temporary and
_bfd_munmap_readonly_temporary instead of _bfd_malloc_and_read
and free.
(_bfd_elf_slurp_version_tables): Call _bfd_mmap_readonly_temporary
and _bfd_munmap_readonly_temporary instead of _bfd_malloc_and_read
and free.
* elflink.c (bfd_elf_link_record_dynamic_symbol): Use bfd_malloc
to get the unversioned symbol.
* libbfd-in.h (_bfd_pagesize): New.
(_bfd_pagesize_m1): Likewise.
(_bfd_minimum_mmap_size): Likewise.
(_bfd_mmap_readonly_persistent): Likewise.
(_bfd_mmap_readonly_temporary): Likewise.
(_bfd_munmap_readonly_temporary): Likewise.
* libbfd.c
(bfd_allocate_mmapped_page): New.
(_bfd_mmap_readonly_temporary): Likewise.
(_bfd_munmap_readonly_temporary): Likewise.
(_bfd_mmap_readonly_persistent): Likewise.
(_bfd_pagesize): Likewise.
(_bfd_pagesize_m1): Likewise.
(_bfd_minimum_mmap_size): Likewise.
(bfd_init_pagesize): Likewise.
* lynx-core.c (lynx_core_file_p): Use _bfd_pagesize.
* opncls.c (_bfd_delete_bfd): Munmap tracked mmapped memories.
* sysdep.h (MAP_ANONYMOUS): New. Define if undefined.
* bfd-in2.h: Regenerated.
* libbfd.h: Likewise.
The XML response to the "qXfer:threads:read" packet may include
a "handle" attribute. The attribute is mentioned in the document
but not shown in the sample XML structure. Add it.
Reviewed-By: Eli Zaretskii <eliz@gnu.org>
In a previous review, I noticed that some code in gdb/compile/compile.c
could use c++17's `std::filesystem::remove_all` instead of using some
`system ("rm -rf ...");`.
This patch implements this.
Note that I use the noexcept overload of std::filesystem::remove_all and
explicitly check for an error code. This means that this code called
during the cleanup procedure cannot throw, and does not risk preventing
other cleanup functions to be called.
Tested on x86_64-linux.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=31420
Change-Id: If5668bf3e15e66c020e5c3b4fa999f861690e4cf
Approved-By: Tom Tromey <tom@tromey.com>
I recent change (e9b738dfbd "Avoid race when reading dwz file") moved
the call to dwarf2_read_dwz_file from dwarf2_initialize_objfile to
dwarf2_has_info.
Before that patch, dwarf2_initialize_objfile was only called when
dwarf2_has_info returned true, and since that patch it is always called.
When reading a file that has no debug info (.debug_info/.debug_abbrev
sections), but has a .gnu_debugaltlink section, GDB’s behavior is
different. I can observe this when loading
/lib/x86_64-linux-gnu/libtinfo.so on Ubuntu 22.04 (or while debugging
any program dynamically loading this library).
Before e9b738dfbd, we had:
$ ./gdb/gdb -data-directory ./gdb/data-directory -q /lib/x86_64-linux-gnu/libtinfo.so
Reading symbols from /lib/x86_64-linux-gnu/libtinfo.so...
(No debugging symbols found in /lib/x86_64-linux-gnu/libtinfo.so)
(gdb)
while after we have:
$ ./gdb/gdb -data-directory ./gdb/data-directory -q /lib/x86_64-linux-gnu/libtinfo.so
Reading symbols from /lib/x86_64-linux-gnu/libtinfo.so...
warning: could not find '.gnu_debugaltlink' file for /usr/lib/x86_64-linux-gnu/libtinfo.so.6.3
(No debugging symbols found in /lib/x86_64-linux-gnu/libtinfo.so)
(gdb)
This patch restores the previous behavior of only trying to load the
DWZ file for objfiles when the main part of the debuginfo is present
(i.e. when dwarf2_has_info returns true). We still make sure that
dwarf2_read_dwz_file is called at most once per objfile.
A consequence of this change is that the per_bfd->dwz_file optional
object can now remain empty (instead of containing a nullptr), so also
this patch also adjusts dwarf2_get_dwz_file to account for this
possibility. This effectively reverts the changes to
dwarf2_get_dwz_file done by e9b738dfbd.
Regression tested on x86_64-linux-gnu Ubuntu 22.04.
Approved-By: Tom Tromey <tom@tromey.com>
Make sure that field actually matches the specified operands. Don't
follow existing F_PSEUDO checking in using assertions, though. Print
meaningful error messages, thus - while not having a line number
available - at least providing some indication of where things are
wrong.
Fix SVE2.1's extq accordingly, but don't extend the testsuite there:
There are further issues with its operands (SVE_Zm_imm4 doesn't look to
be correct to use there, as that describes an indexed vector register,
while here a separate vector register and immediate operand are to be
specified).
Like for WRUSSQ it's not needed here. The legacy insn had gained it in
the course of zapping Rex64, but that attribute wasn't needed here
either. The APX insn then simply gained it by copy-and-paste, I suppose.
Investigating GDB PR d/31580 showed that the libiberty demangler
doesn't automatically demangle D mangled names. However, I think it
should -- like C++ and Rust (new-style), D mangled names are readily
distinguished by the leading "_D", and so the likelihood of confusion
is low. The other non-"auto" cases in this code are Ada (where the
encoded form could more easily be confused by ordinary programs) and
Java (which is long gone, but which also shared the C++ mangling and
thus was just an output style preference).
This patch also fixed another GDB bug, though of course that part
won't apply to the GCC repository.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=31580
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30276
libiberty
* cplus-dem.c (cplus_demangle): Try the D demangler with
"auto" format.
* testsuite/d-demangle-expected: Add --format=auto test.
The recent change to how unsized Rust values are printed included a
small regression from past behavior. Previously, a slice's type would
be printed, like:
(gdb) print slice
$80 = &[i32] [3]
The patch changed this to just
(gdb) print slice
$80 = [3]
This patch restores the previous behavior.
Reviewed-By: Simon Marchi <simon.marchi@efficios.com>
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30330
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=31517
While examining the Ada parser globals with 'nm', I noticed that the
lexer's "attributes" array should be const. This change moves it into
read-only storage.
This patch moves the "temp_parse_space" global into ada_parse_state.
It is also renamed to remove the redundant "parse". Finally, it is
changed to an auto_obstack to avoid the need for any manual
management.
This patch introduces the ada_parse_state class and the ada_parser
global. It also changes find_completion_bounds to be a method of this
new type.
Note that find_completion_bounds never used its parameter; and because
it is generally fine to use the 'pstate' global throughout the parser,
this patch removes the parameter entirely.
Ada 2022 includes iterated assignment for array initialization. This
patch implements a subset of this for gdb. In particular, only arrays
with integer index types really work -- currently there's no decent
way to get the index type in EVAL_AVOID_SIDE_EFFECTS mode during
parsing. Fixing this probably requires the Ada parser to take a
somewhat more sophisticated approach to type resolution; and while
this would help fix another bug in this area, this patch is already
useful without it.
This patch is a refactoring to add a new aggregate_assigner type.
This type is passed to Ada aggregate assignment operations in place of
passing a number of separate arguments. This new approach makes it
simpler to change some aspects of aggregate assignment behavior.
This patch prepares gdb for isort: it adds a couple of isort marker
comments where needed, and it adds an isort clause to setup.cfg.
Approved-By: Simon Marchi <simon.marchi@efficios.com>
flake8 warns about a bare "except". The docs point out that this will
also catch KeyboardInterrupt and SystemExit exceptions, which is
normally undesirable. Using "except Exception" catches everything
reasonable, so this patch makes this change.
Approved-By: Simon Marchi <simon.marchi@efficios.com>
flake8 warns about some identifiers in __init__.py, because it does
not realize these come from the star-imported _gdb module. This patch
suppresses these warnings.
styling.py has a long try/except surrounding most of the body. flake8
warns about the final bare "except". However, this except is really
only there to catch the situation where the host doesn't have Pygments
installed. This patch changes this to only catch ImportError.
Approved-By: Simon Marchi <simon.marchi@efficios.com>
flake8 warns about the "from _gdb.disassembler import *" line in
disassembler.py, and a similar line from __init__.py. These line are
needed to re-export names from the corresponding C++ module, so this
patch applies the appropriate "noqa" flags.
Approved-By: Simon Marchi <simon.marchi@efficios.com>
flake8 complains about a bare "except" in disassembler.py. In this
case, the code purports to guard against some kind of user error
involving data structure corruption. I think it's better here to just
let the error occur -- py-disasm.c will show a stack trace in this
case.
Approved-By: Simon Marchi <simon.marchi@efficios.com>
flake8 warns about dap/__init__.py because it has a number of unused
imports. Most of these are intentional: the import is done to ensure
that the a DAP request is registered with the server object.
This patch applies a "noqa" comment to these imports, and also removes
one import that is truly unnecessary.
Approved-By: Simon Marchi <simon.marchi@efficios.com>
Commit 032d23a6 ("Fix stray KeyboardInterrupt after cancel")
introduced some errors into dap/server.py. A function is called but
not imported, and the wrong variable name is used. This patch
corrects both errors.
Approved-By: Simon Marchi <simon.marchi@efficios.com>
I re-ran flake8 today and was puzzled to see W503 warnings.
Eventually I found out that the setup.cfg config overrides .flake8.
This patch merges the two and removes .flake8, to avoid future
confusion.
Approved-By: Simon Marchi <simon.marchi@efficios.com>
On fedora rawhide, when running test-case gdb.base/ctf-ptype.exp, I get:
...
gdb compile failed, ctf-ptype.c: In function 'main':
ctf-ptype.c:242:29: error: implicit declaration of function 'malloc' \
[-Wimplicit-function-declaration]
242 | v_char_pointer = (char *) malloc (1);
| ^~~~~~
ctf-ptype.c:1:1: note: include '<stdlib.h>' or provide a declaration of 'malloc'
+++ |+#include <stdlib.h>
1 | /* This test program is part of GDB, the GNU debugger.
...
Fix this by adding the missing include.
Tested on aarch64-linux.
In an aarch32-linux chroot on an aarch64-linux system, I run into:
...
(gdb) print x^M
$1 = 9223372036854775807^M
(gdb) FAIL: gdb.ada/verylong.exp: print x
...
A passing version on aarch64-linux looks like:
...
(gdb) print x^M
$1 = 170141183460469231731687303715884105727^M
(gdb) PASS: gdb.ada/verylong.exp: print x
...
The difference is caused by the size of the type Long_Long_Long_Integer, which
is:
- a 128-bit signed on 64-bit targets, and
- a 64-bit signed on 32-bit target.
Fix this by detecting the size of the Long_Long_Long_Integer type, and
handling it.
Tested on aarch64-linux and aarch32-linux.
Approved-By: Tom Tromey <tom@tromey.com>
PR testsuite/31574
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=31574
[1] https://gcc.gnu.org/onlinedocs/gnat_rm/Implementation-Defined-Characteristics.html
After starting TUI like this with a hello world a.out:
...
$ gdb -q a.out -ex start -ex "tui enable"
...
we get:
...
┌─hello.c──────────────────────────────┐
│ 5 { │
│ 6 printf ("hello\n"); │
│ 7 │
│ 8 return 0; │
│ 9 } │
│ │
└──────────────────────────────────────┘
...
This is a regression since commit ee1e9bbb51 ("[gdb/tui] Fix displaying main
after resizing"), before which we had instead:
...
┌─hello.c──────────────────────────────┐
│ 4 main (void) │
│ 5 { │
│ > 6 [7m printf ("hello\n");[0m │
│ 7 │
│ 8 return 0; │
│ 9 } │
└──────────────────────────────────────┘
...
In other words, the problems are:
- the active line (source line 6) is no longer highlighted, and
- the active line is not vertically centered (screen line 2 out 6 instead of
screen line 3 out of 6).
Fix these problems respectively by:
- in tui_enable, instead of "tui_show_frame_info (0)" using
'tui_show_frame_info (deprecated_safe_get_selected_frame ())", and
- in tui_source_window_base::rerender, adding centering functionality.
Tested on aarch64-linux.
Co-Authored-By: Tom Tromey <tom@tromey.com>
Approved-By: Tom Tromey <tom@tromey.com>
PR tui/31522
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=31522
