Having it as a bitfield causes extra work, and this is not memory-sensitive.
Furthermore, once https://sourceware.org/ml/gdb-patches/2019-10/msg00812.html
lands, the bitfield won't even save any memory at all.
gdb/ChangeLog:
2019-10-22 Christian Biesinger <cbiesinger@google.com>
* symtab.c (struct demangled_name_entry) <language>: Change from
bitfield to regular variable.
Change-Id: I4ea31d1cfcbe0f09a09bd058cd304862308dc388
I accidentally pushed the wrong version of the patch for commit
7bb4305982 (where the review
comments were not fixed), and I did a bad conflict resolution
for ccb1ba6229 leading to a
compile error when libxxhash is available. This fixes both
issues.
gdb/ChangeLog:
2019-10-22 Christian Biesinger <cbiesinger@google.com>
* symtab.c (struct demangled_name_entry): Add a constructor.
(free_demangled_name_entry): New function to call the destructor
for demangled_name_entry.
(create_demangled_names_hash): Pass free_demangled_name_entry to
htab_create_alloc.
(symbol_set_names): Call placement new for demangled_name_entry.
* utils.c: No longer include xxhash.h here, now that fast_hash
is inlined in the header.
* utils.h: Instead, include it here.
Change-Id: If776099d39a65a12733d42efcb859feca1b07a39
XXHash is faster than htab_hash_string:
------------------------------------------------------------
Benchmark Time CPU Iterations
------------------------------------------------------------
BM_xxh3 11 ns 11 ns 65887249
BM_xxh32 19 ns 19 ns 36511877
BM_xxh64 16 ns 16 ns 42964585
BM_hash_string 182 ns 182 ns 3853125
BM_iterative_hash 77 ns 77 ns 9087638
Unfortunately, XXH3 is still experimental (see
https://github.com/Cyan4973/xxHash#user-content-new-experimental-hash-algorithm)
However, regular XXH64 is still a lot faster than
htab_hash_string per my benchmark above. I used the
following string for the benchmark:
static constexpr char str[] = "_ZZZL13make_gdb_typeP7gdbarchP10tdesc_typeEN16gdb_type_creator19make_gdb_type_flagsEPK22tdesc_type_with_fieldsE19__PRETTY_FUNCTION__";
htab_hash_string is currently 4.35% + 7.98% (rehashing) of gdb
startup when attaching to Chrome's content_shell.
An additional 5.21% is spent in msymbol_hash, which does not use
this hash function. Unfortunately, since it has to lowercase the
string, it can't use this hash function.
BM_msymbol_hash 52 ns 52 ns 13281495
It may be worth investigating if strlen+XXHash is still faster than
htab_hash_string, which would make it easier to use in more places.
Debian ships xxhash as libxxhash{0,-dev}. Fedora ships it as xxhash-devel.
gdb/ChangeLog:
2019-10-22 Christian Biesinger <cbiesinger@google.com>
* Makefile.in: Link with libxxhash.
* config.in: Regenerate.
* configure: Regenerate.
* configure.ac: Search for libxxhash.
* utils.c (fast_hash): Use xxhash if present.
Change-Id: Icab218388b9f829522ed3977f04301ae6d4fc4ca
Also updates a caller in symtab.c. For now this just calls htab_hash_string
but the next patch will change it to xxhash, if available.
gdb/ChangeLog:
2019-10-22 Christian Biesinger <cbiesinger@google.com>
* utils.h (fast_hash): New function.
* symtab.c (hash_demangled_name_entry): Call new function
fast_hash.
Change-Id: I77cac0d9aa78fc65316a2af449f52edcae72dc9b
This should be a bit faster (because we can compare the size first),
but it is also a dependency for the next patch.
(3.47% of gdb startup time is spent in eq_demangled_name_entry when
attaching to Chrome's content_shell binary)
gdb/ChangeLog:
2019-10-22 Christian Biesinger <cbiesinger@google.com>
* symtab.c (struct demangled_name_entry): Change type of mangled
to gdb::string_view. Also adds a constructor that takes the
mangled name.
(hash_demangled_name_entry): Update.
(eq_demangled_name_entry): Update.
(free_demangled_name_entry): New function to call the destructor
now that this is not a POD anymore.
(create_demangled_names_hash): Pass free_demangled_name_entry to
htab_create_alloc.
(symbol_set_names): Update.
Change-Id: I24711ae2bcaa9e79ca89a6f8fda385d400419175
* rx-dis.c (get_size_name): New function. Provides safe
access to name array.
(get_opsize_name): Likewise.
(print_insn_rx): Use the accessor functions.
At least, not in the GOT. R_PPC64_RELATIVE is fine for powerpc64 in
the .opd section.
PR 22269
* testsuite/ld-elf/pr22269-1.rd: Look for GOT section NONE and
RELATIVE relocs.
* testsuite/ld-elf/shared.exp (pr22269-1): Give test a better
name. Use -z nocombreloc.
* Fix handling of file and directory indexes in line tables; in DWARF 5 the
indexes are zero-based. Make file_names field private to abstract this detail
from the clients. Introduce file_names, is_valid_file_index and
file_names_size methods. Reflect these changes in clients.
* Handle DW_FORM_data16 in read_formatted_entries; it is used to record MD5
of the file entries in DWARF 5.
* Fix a bug in line header parsing that calculates the length of the header
incorrectly. (Seemingly this manifests itself only in DWARF 5).
Tested with CC=/usr/bin/gcc (version 8.3.0) against master branch (also with
-gsplit-dwarf and -gdwarf-4 flags) and there was no increase in the set of
tests that fails. (gdb still cannot debug a 'hello world' program with DWARF 5,
so for the time being, this is all we care about).
This is part of an effort to support DWARF 5 in gdb.
gdb/ChangeLog:
* dwarf2read.c (dir_index): Change type.
(file_name_index): Likewise.
(line_header::include_dir_at): Change comment and implementation on
whether it is DWARF 5.
(line_header::is_valid_file_index): New function.
(line_header::file_name_at): Change comment and implementation on
whether it is DWARF 5.
(line_header::file_names): Change to private field renamed as
m_file_names and introduce a new accessor method.
(line_header::file_names_size): New method.
(line_header::include_dirs): Change to private field and rename as
m_include_dirs.
(dw2_get_file_names_reader): Define local var at a smaller scope and
reflect API change.
(dwarf2_cu::setup_type_unit_groups): Reflect API change.
(process_structure_scope): Likewise.
(line_header::add_include_dir): Change message and reflect renaming.
(line_header::add_file_name): Likewise.
(read_formatted_entries): Handle DW_FORM_data16.
(dwarf_decode_line_header): Fix line header length calculation.
(psymtab_include_file_name): Change comment and API.
(lnp_state_machine::m_file): Update comment and reflect type change.
(lnp_state_machine::record_line): Reflect type change.
(dwarf_decode_lines): Reflect API change.
(file_file_name): Likewise.
(file_full_name): Likewise.
After the switch to use std::sort, if GDB is compiled with the
-D_GLIBCXX_DEBUG=1 flag then we see an error when using sort_cmp (in
objfiles.c) to sort obj_section objects.
The problem is that std::sort checks that the condition !(a < a)
holds, and currently this is not true. GDB's sort_cmp is really
designed to sort lists in which no obj_section repeats, however, there
is some code in place to try and ensure we get a stable sort order if
there is a bug in GDB, unfortunately this code fails the above check.
By reordering some of the checks inside sort_cmp, it is pretty easy to
ensure that the !(a < a) condition holds.
I've not bothered to make this condition check optimal, like I said
this code is only in place to ensure that we get stable results if GDB
goes wrong, so I've made the smallest change needed to get the correct
behaviour.
After this commit I see no regressions when running GDB compiled with
-D_GLIBCXX_DEBUG=1.
gdb/ChangeLog:
* objfiles.c (sort_cmp): Ensure that !(a < a) holds true.
Change-Id: I4b1e3e1640865104c0896cbb6c3fdbbc04d9645b
I happened to notice that the tui_exec_info_content typedef is unused.
This patch removes it. Tested by rebuilding.
gdb/ChangeLog
2019-10-21 Tom Tromey <tom@tromey.com>
* tui/tui-winsource.h (tui_exec_info_content): Remove typedef.
Change-Id: I768edc482366e830eb4528c799686bb27518cdcb
The dg-extract-results scripts have been updated in the gcc
repository. This commit copies the updated versions of the scripts in
to the binutils-gdb repository.
There are two changes, these are:
1. Improved detection of timeout lines, though I suspect this only
applies to gcc results, and
2. Detection of KPASS results, this is of interest to gdb, where
these results would not be included in the final .sum file.
A grep over binutils-gdb shows the dg-extract-results is not used by
ld, gas, or binutils, however I tested these anyway and saw no changes
in the final .sum files (tested on x86-64 GNU/Linux).
On gdb when running tests in parallel dg-extract-results is used, and
the final .sum file now includes the KPASS results.
contrib/ChangeLog:
* dg-extract-results.py: Update from gcc repo.
* dg-extract-results.sh: Likewise.
Change-Id: I54abd07f4e8f5cf88a6db74519674f6939860157
My earlier patch -- commit c5adaa192 ("Fix creation of stamp-h by
gdb's configure script") -- broke the creation of nm.h. In
particular, configure removes nm.h, so if you touch configure and
rebuild, nothing will re-create the link, breaking the build.
This patch fixes the bug, and also updates configure.ac to use
AC_CONFIG_LINKS, rather than the obsolete AC_LINK_FILES.
Finally, I noticed that gcore is in generated_files in the
Makefile.in. I think this is incorrect, as generated_files is only
needed for files that can be the target of a #include. So, this patch
removes it.
gdb/ChangeLog
2019-10-21 Tom Tromey <tom@tromey.com>
* configure.ac (nm.h): Conditionally create nm.h link. Subst
NM_H. Use AC_CONFIG_LINKS.
* configure: Rebuild.
* Makefile.in (NM_H): New variable.
(generated_files): Add NM_H. Remove gcore.
(nm.h, stamp-nmh): New targets.
Change-Id: I8dd539785d52455e85389425e4bb996c8a127a0e
As mentioned in commit 745ff14e6e "[gdb/tdep] Fix 'Unexpected register class'
assert in amd64_push_arguments", of the 12 KFAILs added there, 3 are KPASSing
with g++ 4.8.5.
The KPASSes are due to:
- gdb incorrectly expecting the second half of the result of function
rtn_str_struct_02_01 in register %rdx.
- rtn_str_struct_02_01 using %rdx as a temporary, thereby accidentally setting
it to the expected value.
Reduce the chance of hiding errors due accidental register settings by
compiling the test-case with -O2.
This fixes the KPASSes when applied on top of commit 745ff14e6e.
Tested on x86_64-linux.
Tested with g++ 4.8.5, 7.4.1, 8.3.1, 9.2.1.
gdb/testsuite/ChangeLog:
2019-10-21 Tom de Vries <tdevries@suse.de>
* gdb.base/infcall-nested-structs.c: Add
__attribute__((noinline,noclone)) to all functions.
(call_all): Add missing variable initialization. Simplify return value.
(breakpt): Increment volatile variable, to prevent call from being
optimized out.
* gdb.base/infcall-nested-structs.exp: Compile with -O2.
Change-Id: Ic027e1c957fecd6686345639db99f5eaee3cdf05
This patch extends "ar P" to allow creation of normal (as distinct
from thin) archives with full path names.
PR 452
PR 25104
bfd/
* archive.c (normalize): Return file unchanged when
BFD_ARCHIVE_FULL_PATH.
(_bfd_construct_extended_name_table): Pass abfd, the output
bfd, to normalize.
(_bfd_archive_bsd44_construct_extended_name_table): Likewise.
* bfd.c (struct bfd): Make flags a full flagword.
(BFD_ARCHIVE_FULL_PATH): Define.
* bfd-in2.h: Regenerate.
binutils/
* ar.c (write_archive): Set BFD_ARCHIVE_FULL_PATH.
* doc/binutils.texi (extract from archive): Mention
restrictions when extracting from archives with full paths.
(ar P): Update to current P support.
(ar -X32_64): Fix spelling.
I noticed an obsolete comment just before unlink_objfile, and then I
noticed that both unlink_objfile and put_objfile_before could be
static. This patch makes these changes, and also moves unlink_objfile
earlier, so that a forward declaration is not needed.
Tested by rebuilding.
gdb/ChangeLog
2019-10-20 Tom Tromey <tom@tromey.com>
* objfiles.h (unlink_objfile, put_objfile_before): Don't declare.
* objfiles.c (unlink_objfile): Move earlier. Now static. Remove
obsolete comment.
(put_objfile_before): Now static.
Change-Id: I1b5927a60fd1cc59bfc9c6761f61652a01ef13e0
This change increases the default alignment for the .plt section
from 4 bytes to 8 bytes. When function descriptors are 8-byte
aligned, they can be updated atomically on 32-bit hppa. This
helps with ordering issues on SMP machines. It also ensures
that descriptors reside on the same cache line. This reduces
the probability of a double TLB miss in a call.
2019-10-20 John David Anglin <danglin@gcc.gnu.org>
* elf32-hppa.c (elf32_hppa_size_dynamic_sections): Provide 8-byte
minimum alignment for .plt section.
This commit updates the import stubs to leave the pointer to the
function descriptor in register %r22. This provides a backup
mechanism for _dl_runtime_resolve to fixup descriptors during
lazy binding.
bfd/ChangeLog
2019-10-19 John David Anglin <danglin@gcc.gnu.org>
* elf32-hppa.c: Revise import stub sequences.
(LONG_BRANCH_STUB_SIZE): Define.
(LONG_BRANCH_SHARED_STUB_SIZE): Define.
(IMPORT_STUB_SIZE): Define.
(IMPORT_SHARED_STUB_SIZE): Define.
(EXPORT_STUB_SIZE): Define.
(plt_stub): Revise to not use register %r22.
(LDO_R1_R22): Define.
(LDW_R22_R21): Define.
(LDW_R22_R19): Define.
(hppa_build_one_stub): Update stub generation and use new defines.
(hppa_size_one_stub): Likewise.
This commit fixes two simple typos, one in gdb/symfile.c and the other
in gdb/i386-darwin-tdep.c. s/wether/whether/.
2019-10-19 Sergio Durigan Junior <sergiodj@redhat.com>
* symfile.c (init_entry_point_info): Fix typo.
* i386-darwin-tdep.c (darwin_dwarf_signal_frame_p): Fix typo.
Change-Id: I1fbb39c32009c61c862b6bd56ce12f24a9edb2c4
Two patches from Nelson Chu.
It is better to use the linker's callback functions to handle the link time
error when relocating. The unresolved relocation error can be regarded as
an unsupported relocation. To make user easier to understand different errors,
we need to extend the current error message format of the callback function
since the format is fixed.
bfd/
* elfnn-riscv.c (riscv_elf_relocate_section): Use asprintf to extend
the error message if needed, and then store the result into the
`msg_buf`. Finally, remember to free the unused `msg_buf`. All error
message for the dangerous relocation should be set before we call the
callback function. If we miss the error message since linker runs out
of memory, we should set the default error message for the error.
ld/
* testsuite/ld-riscv-elf/lib-nopic-01a.s: Create the shared library
lib-nopic-01a.so, it will be linked with lib-nopic-01b.s.
* testsuite/ld-riscv-elf/lib-nopic-01b.s: Add new test for the
unresolved relocation. Link the non-pic code into a shared library
may cause the error.
* testsuite/ld-riscv-elf/lib-nopic-01b.d: Likewise.
* testsuite/ld-riscv-elf/ld-riscv-elf.exp: Run the new test only when
the shared library is supported.
R_RISCV_CALL, R_RISCV_JAL and R_RISCV_RVC_JUMP are pc-relative relocation.
For now, we do not allow the object with these relocation links into a shared
library since the referenced symbols may be loaded to the places that too far
from the pc. We can improve the error message for these unsupported relocation
to notice user that they should recompile their code with `fPIC`.
bfd/
* elfnn-riscv.c (riscv_elf_relocate_section): Report the error message
that user should recompile their code with `fPIC` when linking non-pic
code into shared library.
ld/
* testsuite/ld-riscv-elf/lib-nopic-01b.d: Update the error message.
Change-Id: Ib3347a0a6fa1c2b20a9647c314d5bec2c322ff04
I happened to notice that "make" would always print:
CONFIG_HEADERS=config.h:config.in \
CONFIG_COMMANDS="default depdir" \
CONFIG_FILES= \
CONFIG_LINKS= \
/bin/sh config.status
config.status: creating config.h
config.status: config.h is unchanged
on every rebuild. This seems to have changed due to an autoconf
upgrade at some point in the past. In the autoconf gdb uses now, it
works to use AC_CONFIG_HEADERS and then create the stamp file via the
"commands" argument.
This patch also fixes up Makefile.in to use the new-style
config.status invocation. It's no longer necessary to pass the output
file names via environment variables.
gdb/ChangeLog
2019-10-17 Tom Tromey <tromey@adacore.com>
* configure: Rebuild.
* configure.ac: Use AC_CONFIG_HEADERS. Create stamp-h there, not
in AC_CONFIG_FILES invocation.
* Makefile.in (Makefile, data-directory/Makefile, stamp-h): Use
new-style config.status invocation.
gdb/gdbserver/ChangeLog
2019-10-17 Tom Tromey <tromey@adacore.com>
* configure: Rebuild.
* configure.ac: Use AC_CONFIG_HEADERS. Create stamp-h there, not
in AC_CONFIG_FILES invocation.
* Makefile.in (stamp-h, Makefile): Use new-style config.status
invocation.
Change-Id: Ia0530d1c5b9756812d29ddb8dc1062326155e61e
We're forgetting to include the libctf directory when creating the
snapshot. This commit changes src-release.sh and adds it to
GDB_SUPPORT_DIRS.
2019-10-17 Sergio Durigan Junior <sergiodj@redhat.com>
* src-release.sh (GDB_SUPPORT_DIRS): Add libctf.
Change-Id: Iada82e5c9e4f8d5a0a5e467d2a628f1fba66002e
On openSUSE Leap 15.1, I get:
...
FAIL: gdb.fortran/module.exp: info variables -n
...
because the info variables command prints info also for init.c:
...
File init.c:^M
24: const int _IO_stdin_used;^M
...
while the regexps in the test-case only expect info for module.f90.
Fix this by extending the regexps.
Tested on x86_64-linux.
gdb/testsuite/ChangeLog:
2019-10-17 Tom de Vries <tdevries@suse.de>
* gdb.fortran/module.exp: Allow info variables to print info for files
other than module.f90.
Change-Id: I401d8018b121fc7343f6bc8b671900349462457f
As of today we have just the following oddities left
./gnulib/update-gnulib.sh:ver=`autoconf --version 2>&1 | head -1 | sed 's/.*) //'`
./gnulib/update-gnulib.sh:ver=`automake --version 2>&1 | head -1 | sed 's/.*) //'`
./gnulib/update-gnulib.sh:ver=`aclocal --version 2>&1 | grep -v "called too early to check prototype" | head -1 | sed 's/.*) //'`
./src-release.sh: head -1 $tool/version.in
./contrib/dg-extract-results.sh:tail -2 $FIRST_SUM | $GREP '^#' > /dev/null || tail -2 $FIRST_SUM
gnulib and contrib (from gcc) are outside of binutils control, so with
this patch I'm going to declare this 15 year old bug fixed.
PR 29
* src-release.sh (getver): Replace "head -1" with "head -n 1".
I noticed that objfile::original_name could be a "const char *" rather
than a plain "char *". This patch implements this change. Tested by
rebuilding.
gdb/ChangeLog
2019-10-16 Tom Tromey <tom@tromey.com>
* objfiles.h (struct objfile) <original_name>: Now const.
Recent work from Tom Tromey to better handle variables with associated
copy relocations has fixed a Fortran issue where module variables
wouldn't show up in the output of 'info variables'.
This commit adds a test for this functionality to ensure it doesn't
get broken in the future.
gdb/testsuite/ChangeLog:
* gdb.fortran/module.exp: Extend with 'info variables' test.
Change-Id: I7306b1d0a9a72947fd48ad7a03f49df774d6573b
The board file cc-with-tweaks is used as the core for lots of other
board files, for example cc-with-gdb-index and cc-with-debug-names.
This commit extends cc-with-tweaks so that it will wrap the Fortran
compiler, allowing for more test coverage.
I tested all of the board files that make use of cc-with-tweaks
running the gdb.fortran/*.exp test set, and in some cases I did see
extra failures. The "standard" results are:
=== gdb Summary ===
# of expected passes 953
# of known failures 2
With board file 'cc-with-dwz-m':
=== gdb Summary ===
# of expected passes 903
# of unexpected failures 1
# of known failures 2
# of untested testcases 4
With board file 'dwarf4-gdb-index':
=== gdb Summary ===
# of expected passes 950
# of unexpected failures 3
# of known failures 2
With board file 'fission-dwp':
=== gdb Summary ===
# of expected passes 949
# of unexpected failures 4
# of known failures 2
Despite these extra failure I don't think this should prevent this
change going in as these failures presumably already exist in GDB.
gdb/testsuite/ChangeLog:
* boards/cc-with-tweaks.exp: Setup F90_FOR_TARGET and
F77_FOR_TARGET.
Change-Id: I06d412f94d0e119ad652dd6c20829f6705a54622
Saving the signal state is very slow (this patch is a 14% speedup). The
reason we need this code is because signal handler will leave the
signal blocked when we longjmp out of it. But in this case we can
just manually unblock the signal instead of taking the unconditional
perf hit.
gdb/ChangeLog:
2019-10-16 Christian Biesinger <cbiesinger@google.com>
* gdbsupport/gdb_setjmp.h (SIGSETJMP): Allow passing in the value to
pass on to sigsetjmp's second argument.
* cp-support.c (gdb_demangle): Unblock SIGSEGV if we caught a crash.
Change-Id: Ib3010966050c64b4cc8b47d8cb45871652b0b3ea
This is another fuzzer bug, gdb/23567. This time, the fuzzer has
specifically altered the size of .debug_str:
$ eu-readelf -S objdump
Section Headers:
[Nr] Name Type Addr Off Size ES Flags Lk Inf Al
[31] .debug_str PROGBITS 0000000000000000 0057116d ffffffffffffffff 1 MS 0 0 1
When this file is loaded into GDB, the DWARF reader crashes attempting
to access the string table (or it may just store a bunch of nonsense):
[gdb-8.3-6-fc30]
$ gdb -nx -q objdump
BFD: warning: /path/to/objdump has a corrupt section with a size (ffffffffffffffff) larger than the file size
Reading symbols from /path/to/objdump...
Segmentation fault (core dumped)
Nick has already committed a BFD patch to issue the warning seen above.
[gdb master 6acc1a0b]
$ gdb -BFD: warning: /path/to/objdump has a corrupt section with a size (ffffffffffffffff) larger than the file size
Reading symbols from /path/to/objdump...
(gdb) inf func
All defined functions:
File ./../include/dwarf2.def:
186: const
8 *>(.:
;'@�B);
747: const
8 *�(.:
;'@�B);
701: const
8 *�D �
(.:
;'@�B);
71: const
8 *(.:
;'@�B);
/* and more gibberish */
Consider read_indirect_string_at_offset_from:
static const char *
read_indirect_string_at_offset_from (struct objfile *objfile,
bfd *abfd, LONGEST str_offset,
struct dwarf2_section_info *sect,
const char *form_name,
const char *sect_name)
{
dwarf2_read_section (objfile, sect);
if (sect->buffer == NULL)
error (_("%s used without %s section [in module %s]"),
form_name, sect_name, bfd_get_filename (abfd));
if (str_offset >= sect->size)
error (_("%s pointing outside of %s section [in module %s]"),
form_name, sect_name, bfd_get_filename (abfd));
gdb_assert (HOST_CHAR_BIT == 8);
if (sect->buffer[str_offset] == '\0')
return NULL;
return (const char *) (sect->buffer + str_offset);
}
With sect_size being ginormous, the code attempts to access
sect->buffer[GINORMOUS], and depending on the layout of memory,
GDB either stores a bunch of gibberish strings or crashes.
This is an attempt to mitigate this by implementing a similar approach
used by BFD. In our case, we simply reject the section with the invalid
length:
$ ./gdb -nx -q objdump
BFD: warning: /path/to/objdump has a corrupt section with a size (ffffffffffffffff) larger than the file size
Reading symbols from /path/to/objdump...
warning: Discarding section .debug_str which has a section size (ffffffffffffffff) larger than the file size [in module /path/to/objdump]
DW_FORM_strp used without .debug_str section [in module /path/to/objdump]
(No debugging symbols found in /path/to/objdump)
(gdb)
Unfortunately, I have not found a way to regression test this, since it
requires poking ELF section headers.
gdb/ChangeLog:
2019-10-16 Keith Seitz <keiths@redhat.com>
PR gdb/23567
* dwarf2read.c (dwarf2_per_objfile::locate_sections): Discard
sections whose size is greater than the file size.
Change-Id: I896ac3b4eb2207c54e8e05c16beab3051d9b4b2f
This adds initial compile command support to the RISC-V port. This fixes
about 228 testsuite failures on a riscv64-linux machine. We need to get
the triplet right which is normally riscv64 or riscv32 instead of the
default riscv. Also, we need to get the compiler options right, since we
don't accept the default -m64 and -mcmodel=large options, so we need to
construct -march and -mabi options which are correct for the target. We
currently don't have info about all extensions used by the target, so this
may need to be adjusted later. For now, I'm assuming that we have all
extensions required by the linux platform spec.
gdb/
* riscv-tdep.c (riscv_gcc_target_options): New.
(riscv_gnu_triplet_regexp): New.
(riscv_gdbarch_init): Call set_gdbarch_gcc_triplet_options and
set_gdbarch_gnu_triplet_regexp.
Change-Id: I315ce8de7789ddf7bdd3b532f917519464941294
xml-builtin.c only has character arrays and no dependencies, so this
creates a simple header file for that purpose so that gdbserver
can include that instead of re-declaring xml_builtin.
Despite the name, feature_to_c.sh is already specific to xml_builtins
(it hardcodes the variable name), so making it always output the
include for xml-builtin.h seems fine.
gdb/ChangeLog:
2019-10-16 Christian Biesinger <cbiesinger@google.com>
* Makefile.in: Add xml-builtin.h.
* features/feature_to_c.sh: Add an include for xml-builtin.h
to ensure that the compiler checks that the types match.
* xml-builtin.h: New file.
* xml-support.c (fetch_xml_builtin): Add missing const.
* xml-support.h: Remove declaration of xml_builtins.
gdb/gdbserver/ChangeLog:
2019-10-16 Christian Biesinger <cbiesinger@google.com>
* server.c: Include xml-builtin.h.
(get_xml_features): Don't declare xml_builtins here.
Change-Id: I806ef0851c43ead90b545a11794e41f5e5178436
When building binutils with mingw-w64, I get the following errors:
make[4]: Entering directory '/home/simark/build/binutils-gdb-mingw/binutils'
/bin/sh ./libtool --tag=CC --mode=link ccache x86_64-w64-mingw32-gcc -W -Wall -Wstrict-prototypes -Wmissing-prototypes -Wshadow -Wstack-usage=262144 -Wno-format -Werror -I/home/simark/src/binutils-gdb/binutils/../zlib -g3 -O0 -D__USE_MINGW_ACCESS -Wl,--stack,12582912 -o objdump.exe objdump.o dwarf.o prdbg.o rddbg.o debug.o stabs.o rdcoff.o bucomm.o version.o filemode.o elfcomm.o ../opcodes/libopcodes.la ../libctf/libctf.la ../bfd/libbfd.la ../libiberty/libiberty.a -lintl
libtool: link: ccache x86_64-w64-mingw32-gcc -W -Wall -Wstrict-prototypes -Wmissing-prototypes -Wshadow -Wstack-usage=262144 -Wno-format -Werror -I/home/simark/src/binutils-gdb/binutils/../zlib -g3 -O0 -D__USE_MINGW_ACCESS -Wl,--stack -Wl,12582912 -o .libs/objdump.exe objdump.o dwarf.o prdbg.o rddbg.o debug.o stabs.o rdcoff.o bucomm.o version.o filemode.o elfcomm.o ../opcodes/.libs/libopcodes.a ../libctf/.libs/libctf.a -L/home/simark/build/binutils-gdb-mingw/zlib ../bfd/.libs/libbfd.a -lz ../libiberty/libiberty.a -lintl
/usr/lib/gcc/x86_64-w64-mingw32/9.2.0/../../../../x86_64-w64-mingw32/bin/ld: ../libctf/.libs/libctf.a(ctf-open.o): in function `flip_header':
/home/simark/src/binutils-gdb/libctf/ctf-open.c:964: undefined reference to `bswap_16'
/usr/lib/gcc/x86_64-w64-mingw32/9.2.0/../../../../x86_64-w64-mingw32/bin/ld: /home/simark/src/binutils-gdb/libctf/ctf-open.c:967: undefined reference to `bswap_32'
/usr/lib/gcc/x86_64-w64-mingw32/9.2.0/../../../../x86_64-w64-mingw32/bin/ld: /home/simark/src/binutils-gdb/libctf/ctf-open.c:968: undefined reference to `bswap_32'
/usr/lib/gcc/x86_64-w64-mingw32/9.2.0/../../../../x86_64-w64-mingw32/bin/ld: /home/simark/src/binutils-gdb/libctf/ctf-open.c:969: undefined reference to `bswap_32'
/usr/lib/gcc/x86_64-w64-mingw32/9.2.0/../../../../x86_64-w64-mingw32/bin/ld: /home/simark/src/binutils-gdb/libctf/ctf-open.c:970: undefined reference to `bswap_32'
/usr/lib/gcc/x86_64-w64-mingw32/9.2.0/../../../../x86_64-w64-mingw32/bin/ld: /home/simark/src/binutils-gdb/libctf/ctf-open.c:971: undefined reference to `bswap_32'
/usr/lib/gcc/x86_64-w64-mingw32/9.2.0/../../../../x86_64-w64-mingw32/bin/ld: ../libctf/.libs/libctf.a(ctf-open.o):/home/simark/src/binutils-gdb/libctf/ctf-open.c:972: more undefined references to `bswap_32' follow
/usr/lib/gcc/x86_64-w64-mingw32/9.2.0/../../../../x86_64-w64-mingw32/bin/ld: ../libctf/.libs/libctf.a(ctf-open.o): in function `flip_types':
/home/simark/src/binutils-gdb/libctf/ctf-open.c:1112: undefined reference to `bswap_16'
/usr/lib/gcc/x86_64-w64-mingw32/9.2.0/../../../../x86_64-w64-mingw32/bin/ld: /home/simark/src/binutils-gdb/libctf/ctf-open.c:1113: undefined reference to `bswap_16'
/usr/lib/gcc/x86_64-w64-mingw32/9.2.0/../../../../x86_64-w64-mingw32/bin/ld: /home/simark/src/binutils-gdb/libctf/ctf-open.c:1132: undefined reference to `bswap_32'
/usr/lib/gcc/x86_64-w64-mingw32/9.2.0/../../../../x86_64-w64-mingw32/bin/ld: /home/simark/src/binutils-gdb/libctf/ctf-open.c:1133: undefined reference to `bswap_32'
/usr/lib/gcc/x86_64-w64-mingw32/9.2.0/../../../../x86_64-w64-mingw32/bin/ld: /home/simark/src/binutils-gdb/libctf/ctf-open.c:1134: undefined reference to `bswap_32'
/usr/lib/gcc/x86_64-w64-mingw32/9.2.0/../../../../x86_64-w64-mingw32/bin/ld: /home/simark/src/binutils-gdb/libctf/ctf-open.c:1135: undefined reference to `bswap_32'
/usr/lib/gcc/x86_64-w64-mingw32/9.2.0/../../../../x86_64-w64-mingw32/bin/ld: /home/simark/src/binutils-gdb/libctf/ctf-open.c:1144: undefined reference to `bswap_32'
/usr/lib/gcc/x86_64-w64-mingw32/9.2.0/../../../../x86_64-w64-mingw32/bin/ld: ../libctf/.libs/libctf.a(ctf-open.o):/home/simark/src/binutils-gdb/libctf/ctf-open.c:1145: more undefined references to `bswap_32' follow
/usr/lib/gcc/x86_64-w64-mingw32/9.2.0/../../../../x86_64-w64-mingw32/bin/ld: ../libctf/.libs/libctf.a(ctf-open.o): in function `ctf_bufopen_internal':
/home/simark/src/binutils-gdb/libctf/ctf-open.c:1342: undefined reference to `bswap_16'
/usr/lib/gcc/x86_64-w64-mingw32/9.2.0/../../../../x86_64-w64-mingw32/bin/ld: ../libctf/.libs/libctf.a(ctf-open-bfd.o): in function `ctf_fdopen':
/home/simark/src/binutils-gdb/libctf/ctf-open-bfd.c:268: undefined reference to `bswap_16'
Apparently [1], if we have a function with `inline` but not `static`,
there should be a compilation unit defining the symbol too.
Alternatively, making those functions `static` fixes that.
[1] https://stackoverflow.com/questions/16245521/c99-inline-function-in-c-file/16254679#16254679
libctf/ChangeLog:
* swap.h (bswap_16, bswap_32, bswap_64): Make static.
Change-Id: I8fd12aedf6c90f9b7418af948e5e0bae0c32eead
We currently have 12 KFAILS in gdb.base/infcall-nested-structs.exp for
PR tdep/25096.
A minimal version of the failure looks like this. Consider test.c:
...
struct s { int c; struct { int a; float b; } s1; };
struct s ref = { 0, { 'a', 'b' } };
int __attribute__((noinline,noclone)) check (struct s arg)
{ return arg.s1.a == 'a' && arg.s1.b == 'b' && arg.c == 0; }
int main (void)
{ return check (ref); }
...
When calling 'check (ref)' from main, we have '1' as expected:
...
$ g++ test.c -g ; ./a.out ; echo $?
1
...
But when calling 'check (ref)' from the gdb prompt, we get '0':
...
$ gdb a.out -batch -ex start -ex "p check (ref)"
Temporary breakpoint 1 at 0x400518: file test.c, line 8.
Temporary breakpoint 1, main () at test.c:8
8 { return check (ref); }
$1 = 0
...
The layout of struct s is this:
- the field c occupies 4 bytes at offset 0,
- the s1.a field occupies 4 bytes at offset 4, and
- the s1.b field occupies 4 bytes at offset 8.
When compiling at -O2, we can see from the disassembly of main:
...
4003f0: 48 8b 3d 31 0c 20 00 mov 0x200c31(%rip),%rdi \
# 601028 <ref>
4003f7: f3 0f 10 05 31 0c 20 movss 0x200c31(%rip),%xmm0 \
# 601030 <ref+0x8>
4003fe: 00
4003ff: e9 ec 00 00 00 jmpq 4004f0 <_Z5check1s>
...
that check is called with fields c and s1.a passed in %rdi, and s1.b passed
in %xmm0.
However, the classification in theclass (a variable representing the first and
second eightbytes, to put it in SYSV X86_64 psABI terms) in
amd64_push_arguments is incorrect:
...
(gdb) p theclass
$1 = {AMD64_INTEGER, AMD64_INTEGER}
...
and therefore the struct is passed using %rdi and %rsi instead of using %rdi
and %xmm0, which explains the failure.
The reason that we're misclassifying the argument in amd64_classify_aggregate
has to do with how nested struct are handled.
Rather than using fields c and s1.a for the first eightbyte, and using field
s1.b for the second eightbyte, instead field c is used for the first
eightbyte, and fields s1.a and s1.b are classified together in an intermediate
eightbyte, which is then used to merge with both the first and second
eightbyte.
Fix this by factoring out a new function amd64_classify_aggregate_field, and
letting it recursively handle fields of nested structs.
Tested on x86_64-linux.
Tested with g++ 4.8.5, 7.4.1, 8.3.1, 9.2.1.
Tested with clang++ 5.0.2 (which requires removing additional_flags=-Wno-psabi
and adding additional_flags=-Wno-deprecated).
gdb/ChangeLog:
2019-10-16 Tom de Vries <tdevries@suse.de>
PR tdep/25096
* amd64-tdep.c (amd64_classify_aggregate_field): Factor out of ...
(amd64_classify_aggregate): ... here.
(amd64_classify_aggregate_field): Handled fiels of nested structs
recursively.
gdb/testsuite/ChangeLog:
2019-10-16 Tom de Vries <tdevries@suse.de>
PR tdep/25096
* gdb.base/infcall-nested-structs.exp: Remove PR25096 KFAILs.
Change-Id: Id55c74755f0a431ce31223acc86865718ae0c123
Atm, when executing gdb.base/infcall-nested-structs.exp on x86_64-linux, we get:
...
FAIL: gdb.base/infcall-nested-structs.exp: l=c++: types-tc-tf: \
p/d check_arg_struct_02_01 (ref_val_struct_02_01)
FAIL: gdb.base/infcall-nested-structs.exp: l=c++: types-ts-tf: \
p/d check_arg_struct_02_01 (ref_val_struct_02_01)
FAIL: gdb.base/infcall-nested-structs.exp: l=c++: types-ti-tf: \
p/d check_arg_struct_02_01 (ref_val_struct_02_01)
=== gdb Summary ===
nr of expected passes 9255
nr of unexpected failures 3
nr of expected failures 142
...
The 3 FAILs are reported as PR tdep/25096.
The 142 XFAILs are for a gdb assertion failure, reported in PR tdep/24104,
which should have been KFAILs since there's a problem in gdb rather than in
the environment.
A minimal version of the assertion failure looks like this. Consider test.c:
...
struct s { struct { } es1; long f; };
struct s ref = { {}, 'f' };
int __attribute__((noinline,noclone)) check (struct s arg)
{ return arg.f == 'f'; }
int main (void)
{ return check (ref); }
...
When calling 'check (ref)' from main, we have '1' as expected:
...
$ g++ test3.c -g && ( ./a.out; echo $? )
1
...
But when calling 'check (ref)' from the gdb prompt, we get:
...
$ gdb a.out -batch -ex start -ex "p check (ref)"
Temporary breakpoint 1 at 0x4004f7: file test.c, line 8.
Temporary breakpoint 1, main () at test.c:8
8 { return check (ref); }
src/gdb/amd64-tdep.c:982: internal-error: \
CORE_ADDR amd64_push_arguments(regcache*, int, value**, CORE_ADDR, \
function_call_return_method): \
Assertion `!"Unexpected register class."' failed.
...
The assert happens in this loop in amd64_push_arguments:
...
for (j = 0; len > 0; j++, len -= 8)
{
int regnum = -1;
int offset = 0;
switch (theclass[j])
{
case AMD64_INTEGER:
regnum = integer_regnum[integer_reg++];
break;
case AMD64_SSE:
regnum = sse_regnum[sse_reg++];
break;
case AMD64_SSEUP:
gdb_assert (sse_reg > 0);
regnum = sse_regnum[sse_reg - 1];
offset = 8;
break;
default:
gdb_assert (!"Unexpected register class.");
}
...
}
...
when processing theclass[0], which is AMD64_NO_CLASS:
...
(gdb) p theclass
$1 = {AMD64_NO_CLASS, AMD64_INTEGER}
...
The layout of struct s is that the empty field es1 occupies one byte (due to
c++) at offset 0, and the long field f occupies 8 bytes at offset 8.
When compiling at -O2, we can see from the disassembly of main:
...
4003f0: 48 8b 3d 41 0c 20 00 mov 0x200c41(%rip),%rdi \
# 601038 <ref+0x8>
4003f7: e9 e4 00 00 00 jmpq 4004e0 <_Z5check1s>
4003fc: 0f 1f 40 00 nopl 0x0(%rax)
...
that check is called with field f passed in %rdi, meaning that the
classification in theclass is correct, it's just not supported in the loop in
amd64_push_arguments mentioned above.
Fix the assert by implementing support for 'AMD64_NO_CLASS' in that loop.
This exposes 9 more FAILs of the PR tdep/25096 type, so mark all 12 of them as
KFAIL.
Tested on x86_64-linux.
Tested with g++ 4.8.5, 7.4.1, 8.3.1, 9.2.1. With 4.8.5, 3 of the 12 KFAILs
are KPASSing.
Tested with clang++ 5.0.2 (which requires removing additional_flags=-Wno-psabi
and adding additional_flags=-Wno-deprecated).
gdb/ChangeLog:
2019-10-16 Tom de Vries <tdevries@suse.de>
PR tdep/24104
* amd64-tdep.c (amd64_push_arguments): Handle AMD64_NO_CLASS in loop
that handles 'theclass'.
gdb/testsuite/ChangeLog:
2019-10-16 Tom de Vries <tdevries@suse.de>
PR tdep/24104
* gdb.base/infcall-nested-structs.exp: Remove XFAIL for PR tdep/24104.
Add KFAIL for PR tdep/25096.
Change-Id: I8b66345bbf5c00209ca75b1209fd4d60b36e9ede
With g++-4.8, I see:
...
(gdb) PASS: gdb.cp/local-static.exp: c++: print free_inline_func(void)
print 'S::method()'::S_M_s_var_int^M
No symbol "S_M_s_var_int" in specified context.^M
(gdb) FAIL: gdb.cp/local-static.exp: c++: print 'S::method()'::S_M_s_var_int
...
The variable is declared like this (showing pruned .ii):
...
void S::method ()
{
static int S_M_s_var_int = 4;
}
...
But the DWARF generated for the variable is encapsulated in an unnamed lexical
block:
...
<1><121>: Abbrev Number: 5 (DW_TAG_structure_type)
<122> DW_AT_name : S
...
<2><14f>: Abbrev Number: 6 (DW_TAG_subprogram)
...
<150> DW_AT_name : (indirect string, offset: 0x599): method
<156> DW_AT_linkage_name: (indirect string, offset: 0x517): \
_ZN1S6methodEv /* demangled: dS::method() */
...
<1><3f8>: Abbrev Number: 21 (DW_TAG_subprogram)
<3f9> DW_AT_specification: <0x14f>
...
<3fe> DW_AT_low_pc : 0x4004fc
<406> DW_AT_high_pc : 0x2c /* 0x400528 */
...
<2><418>: Abbrev Number: 17 (DW_TAG_formal_parameter)
<419> DW_AT_name : (indirect string, offset: 0x68a): this
...
<2><424>: Abbrev Number: 18 (DW_TAG_lexical_block)
<425> DW_AT_low_pc : 0x400508
<42d> DW_AT_high_pc : 0x1e /* 0x400526 */
<3><435>: Abbrev Number: 22 (DW_TAG_variable)
<436> DW_AT_name : (indirect string, offset: 0x29d): S_M_s_var_int
...
which has the effect that the variable is not addressable unless the program
counter is in the range of the lexical block.
This is caused by gcc PR debug/55541, which was fixed in gcc 5.
Mark in total 225 FAILs as XFAIL.
Tested on x86_64-linux.
gdb/testsuite/ChangeLog:
2019-10-16 Tom de Vries <tdevries@suse.de>
PR testsuite/25059
* gdb.cp/local-static.exp (do_test): Add xfails for gcc PR debug/55541.
Change-Id: Ibe86707eecffc79f1bb474d7928ea7d0c39a00a2
On openSUSE Leap 15.1 (as well as on Fedora-x86_64-m64 buildbot) I see:
...
FAIL: gdb.base/jit-reader.exp: with jit-reader: after mangling: current frame: info registers
...
The problem is that r10 is printed signed:
...
r10 0xffffffffffffffb0 -80^M
...
but the regexp expects a signed value:
...
"r10 $hex +$decimal" \
...
Fix this by allowing signed values.
Tested on x86_64-linux.
gdb/testsuite/ChangeLog:
2019-10-16 Tom de Vries <tdevries@suse.de>
* gdb.base/jit-reader.exp: Allow non-pointer registers to be printed
as signed.
Change-Id: Ie494d24fad7a9af7ac6bfaf731c4aa04f1333830
This implements padding of orphan executable sections for PowerPC.
Of course, the simple implementation of bfd_arch_ppc_nop_fill and
removing the NOP definition didn't work, with powerpc64 hitting a
testsuite failure linking to S-records. That's because the srec
target is BFD_ENDIAN_UNKNOWN so the test of bfd_big_endian (abfd) in
default_data_link_order therefore returned false, resulting in a
little-endian nop pattern. The rest of the patch fixes that problem
by adding a new field to bfd_link_info that can be used to determine
actual endianness on targets like srec.
PR 13616
include/
* bfdlink.h (struct bfd_link_info <big_endian>): New field.
bfd/
* cpu-powerpc.c (bfd_arch_ppc_nop_fill): New function, use it
for all ppc arch info.
* linker.c (default_data_link_order): Pass info->big_endian to
arch_info->fill function.
ld/
* emulparams/elf64lppc.sh (NOP): Don't define.
* emulparams/elf64ppc.sh (NOP): Don't define.
* ldwrite.c (build_link_order): Use link_info.big_endian. Move
code determining endian to use for data_statement to..
* ldemul.c (after_open_default): ..here. Set link_info.big_endian.
opcodes * rx-dis.c (get_register_name): New function. Provides safe
access to name array.
(get_condition_name, get_flag_name, get_double_register_name)
(get_double_register_high_name, get_double_register_low_name)
(get_double_control_register_name, get_double_condition_name):
Likewise.
(print_insn_rx): Use the accessor functions.
Some of the comments in this file are ancient and no longer reflect
reality. This patch removes those comments, and also the description
of ld options emitted to script files. While describing what an
option does in the script file might help reinforce what the option
does, it's unnecessary and makes for overlong lines. Also, some of
the descriptions did not mention all the options.
* genscripts.sh: Correct comments. Remove outdated comment block
saying "Generate 5 or 6 script files..". Remove description of
ld options from comment emitted to script files, and order options
as per comment block in genscripts.sh saying which scripts are
generated.
