SiFive has define as set of flexible instruction for extending vector
coprocessor, it able to encoding opcode like .insn but with predefined
format.
List of instructions:
sf.vc.x
sf.vc.i
sf.vc.vv
sf.vc.xv
sf.vc.iv
sf.vc.fv
sf.vc.vvv
sf.vc.xvv
sf.vc.ivv
sf.vc.fvv
sf.vc.vvw
sf.vc.xvw
sf.vc.ivw
sf.vc.fvw
sf.vc.v.x
sf.vc.v.i
sf.vc.v.vv
sf.vc.v.xv
sf.vc.v.iv
sf.vc.v.fv
sf.vc.v.vvv
sf.vc.v.xvv
sf.vc.v.ivv
sf.vc.v.fvv
sf.vc.v.vvw
sf.vc.v.xvw
sf.vc.v.ivw
sf.vc.v.fvw
Spec of Xsfvcp
https://www.sifive.com/document-file/sifive-vector-coprocessor-interface-vcix-software
Co-authored-by: Hau Hsu <hau.hsu@sifive.com>
Co-authored-by: Kito Cheng <kito.cheng@sifive.com>
Back then when the support for the RISC-V vector crypto extensions
was merged, the specification was frozen, but not ratified.
A frozen specification is allowed to change within tight bounds
before ratification and this has happend with the vector crypto
extensions.
The following changes were applied:
* A new extension Zvkb was defined, which is a strict subset of Zvbb.
* Zvkn and Zvks include now Zvkb instead of Zvbb.
This patch implements these changes between the frozen and the
ratified specification.
Note, that this technically an incompatible change of Zvkn and Zvks,
but I am not aware of any project that depends on the currently
implemented behaviour of Zvkn and Zvks. So this patch should be fine.
Reported-By: Jerry Shih <jerry.shih@sifive.com>
Reported-By: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Christoph Müllner <christoph.muellner@vrull.eu>
In gdb/configure the line:
...
$development || tentative_python_cflags="$tentative_python_cflags -DNDEBUG"
...
intends to ensure that -DNDEBUG is added to the python flags of a release
build.
However, when building gdb-14-branch we have:
...
configure:22024: checking compiler flags for python code
...
configure:22047: result: -fno-strict-aliasing -fwrapv
...
This is a regression since commit db6878ac55 ("Move sourcing of development.sh
to GDB_AC_COMMON"), which introduced a reference before assignment:
...
$development || tentative_python_cflags="$tentative_python_cflags -DNDEBUG"
...
. $srcdir/../bfd/development.sh
...
and consequently -DNDEBUG is never added.
[ This was not obvious to me, but apparently evaluating an empty or undefined
variable in this context is similar to using ':' or 'true', so the line is
evaluated as:
...
true || tentative_python_cflags="$tentative_python_cflags -DNDEBUG"
... ]
Fix this by moving GDB_AC_COMMON up in gdb/configure.ac, similar to how that
was done for gdbserver/configure.ac in commit db6878ac55.
[ Unfortunately, the move might introduce issues similar to the one we're
fixing, and I'm not sure how to check for this. Shellcheck doesn't detect
this type of problem. FWIW, I did run shellcheck (using arguments -xa, in the
src/gdb directory to make sure ../bfd/development.sh is taken into account)
before and after and observed that the number of lines/words/chars in the
shellcheck output is identical. ]
Build & tested on top of trunk.
Also build on top of gdb-14-branch, and observed this in gdb/config.log:
...
configure:25214: checking compiler flags for python code
...
configure:25237: result: -fno-strict-aliasing -fwrapv -DNDEBUG
...
Approved-By: Tom Tromey <tom@tromey.com>
PR build/31099
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=31099
On MIPSr6, the encoding of JR instruction has been chaned.
This patch can fix these failures for r6 default triples:
ST Microelectronics Loongson-2F workarounds of Jump Instruction issue
This behavior is used by downstream toolchain since 2014,
and has been in GCC since the same year.
We don't support mips64*-img* due to GCC doesn't support it,
and we believe that the multilib should be used for this case.
procfs.c doesn't currently compile on Solaris:
/vol/src/gnu/gdb/hg/master/dist/gdb/procfs.c: In member function ‘virtual int procfs_target::can_use_hw_breakpoint(bptype, int, int)’:
/vol/src/gnu/gdb/hg/master/dist/gdb/procfs.c:3017:9: error: ‘ptr_type’ was not declared in this scope; did you mean ‘var_types’?
3017 | type *ptr_type
| ^~~~~~~~
| var_types
This was caused by this patch:
commit 99d9c3b92c
Author: Simon Marchi <simon.marchi@efficios.com>
Date: Fri Sep 29 14:24:38 2023 -0400
gdb: remove target_gdbarch
Partially undoing it restores the build.
Tested on amd64-pc-solaris2.11.
This patch
commit bf4f40cc3195eb7b900bf5535cdba1ee51fdbb8e
Author: Jakub Jelinek <jakub@redhat.com>
Date: Tue Nov 28 13:14:05 2023 +0100
libiberty: Use x86 HW optimized sha1
broke Solaris/x86 bootstrap with the native as:
libtool: compile: /var/gcc/regression/master/11.4-gcc/build/./gcc/gccgo -B/var/gcc/regression/master/11.4-gcc/build/./gcc/ -B/vol/gcc/i386-pc-solaris2.11/bin/ -B/vol/gcc/i386-pc-solaris2.11/lib/ -isystem /vol/gcc/i386-pc-solaris2.11/include -isystem /vol/gcc/i386-pc-solaris2.11/sys-include -fchecking=1 -minline-all-stringops -O2 -g -I . -c -fgo-pkgpath=internal/goarch /vol/gcc/src/hg/master/local/libgo/go/internal/goarch/goarch.go zgoarch.go
ld.so.1: go1: fatal: /var/gcc/regression/master/11.4-gcc/build/gcc/go1: hardware capability (CA_SUNW_HW_2) unsupported: 0x4000000 [ SHA1 ]
gccgo: fatal error: Killed signal terminated program go1
As is already done in a couple of other similar cases, this patches
disables hwcaps support for libiberty.
Initially, this didn't work because config/hwcaps.m4 uses target_os, but
didn't ensure it is defined.
Tested on i386-pc-solaris2.11 with as and gas.
2023-11-29 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
config:
* hwcaps.m4 (GCC_CHECK_ASSEMBLER_HWCAP): Require
AC_CANONICAL_TARGET.
libiberty:
* configure.ac (GCC_CHECK_ASSEMBLER_HWCAP): Invoke.
* configure, aclocal.m4: Regenerate.
* Makefile.in (COMPILE.c): Add HWCAP_CFLAGS.
Currently objdump gets and updates the map state once per symbol. Updating the
state (partiularly riscv_parse_subset) is expensive and grows quadratically
since we iterate over all symbols. By deferring this until once we've found the
symbol of interest, we can reduce the time to dump a 4k insn file of .norvc and
.rvc insns from ~47 seconds to ~0.13 seconds.
opcodes/ChangeLog:
* riscv-dis.c (riscv_get_map_state): Remove state updating logic
and rename to riscv_is_valid_mapping_symbol.
(riscv_update_map_state): Add state updating logic to seperate function.
(riscv_search_mapping_symbol): Use new riscv_update_map_state.
(riscv_data_length): Ditto.
Signed-off-by: Patrick O'Neill <patrick@rivosinc.com>
This patch makes the BPF assembler to support double-slash line
comments, like the llvm BPF assembler does. At this point both
assemblers support the same commenting styles:
- Line comments preceded by # or //.
- Non-nestable block comments delimited by /* and */.
This patch also adds a couple of tests to make sure all the comment
styles work in both normal and pseudoc syntax. The manual is also
updated to mention double-slash line comments.
gdbsupport/scope-exit.h has a couple of comments about being able to
rely on copy elision in C++17. This patch makes the change.
Approved-By: Pedro Alves <pedro@palves.net>
gdbsupport/new-op.cc has a comment about relying on the C++-17 <new>
header. This patch implements the suggestion.
Approved-By: Pedro Alves <pedro@palves.net>
index-write.c has a comment indicating that C++17's try_emplace could
be used. This patch makes the change.
Approved-By: Pedro Alves <pedro@palves.net>
This changes the various gdb-related directories to use
-Wimplicit-fallthrough=5, meaning that only the fallthrough attribute
can be used in switches -- special 'fallthrough' comments will no
longer be usable.
Approved-By: Pedro Alves <pedro@palves.net>
This changes gdb to use the C++17 [[fallthrough]] attribute rather
than special comments.
This was mostly done by script, but I neglected a few spellings and so
also fixed it up by hand.
I suspect this fixes the bug mentioned below, by switching to a
standard approach that, presumably, clang supports.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=23159
Approved-By: John Baldwin <jhb@FreeBSD.org>
Approved-By: Luis Machado <luis.machado@arm.com>
Approved-By: Pedro Alves <pedro@palves.net>
This is a major update of gp-display-html. The option handling has been
modified to support the GNU style long option syntax. Compatibility with
the previous syntax has been preserved. If still used, a warning is issued.
Through the --nowarnings option, this can be suppressed.
In addition to this, various lay-out changes have been implemented. In
particular to reduce the number of lines that extend beyond column 79.
Several bugs have been fixed, for example in the handling of directory names.
gprofng/ChangeLog
2023-11-28 Ruud van der Pas <ruud.vanderpas@oracle.com>
* gp-display-html/gp-display-html.in: Change option syntax plus fixes.
This is a major update of all the man pages. Bugs 30679 and 30895 are
addressed. In addition to fixes for typos, the texts have been expanded
and clarified, and line lengths no longer extend beyond column 79. In
case of gp-display-html, the new option syntax is documented. The user
guide has a new section on the gprofng GUI.
gprofng/ChangeLog
2023-11-28 Ruud van der Pas <ruud.vanderpas@oracle.com>
PR 30679
PR 30895
* doc/gp-archive.texi: Expand the description of the options.
* doc/gp-collect-app.texi: Fix various typos and textual improvements.
* doc/gp-display-html.texi: Cover the new GNU long option syntax.
* doc/gp-display-src.texi: Fix various typos and textual improvements.
* doc/gp-display-text.texi: Fix typos fixed and textual improvements.
* doc/gp-macros.texi: Fix a bug in the vspace macro and add new macro.
* doc/gprofng.texi: Cover the GPROFNG_SYSCONFDIR environment variable.
* doc/gprofng_ug.texi: Fix various typos and textual improvements.
* doc/version.texi: Adapt the date and version number.
This commit makes the gdb.Command.complete methods more verbose when
it comes to error handling.
Previous to this commit if any commands implemented in Python
implemented the complete method, and if there were any errors
encountered when calling that complete method, then GDB would silently
hide the error and continue as if there were no completions.
This makes is difficult to debug any errors encountered when writing
completion methods, and encourages the idea that Python extensions can
be broken, and GDB will just silently work around them.
I don't think this is a good idea. GDB should encourage extensions to
be written correctly, and robustly, and one way in which GDB can (I
think) support this, is by pointing out when an extension goes wrong.
In this commit I've gone through the Python command completion code,
and added calls to gdbpy_print_stack() or gdbpy_print_stack_or_quit()
in places where we were either clearing the Python error, or, in some
cases, just not handling the error at all.
One thing I have not changed is in cmdpy_completer (py-cmd.c) where we
process the list of completions returned from the Command.complete
method; this routine includes a call to gdbpy_is_string to check a
possible completion is a string, if not the completion is ignored.
I was tempted to remove this check, attempt to complete each result to
a string, and display an error if the conversion fails. After all,
returning anything but a string is surely a mistake by the extension
author.
However, the docs clearly say that only strings within the returned
list will be considered as completions. Anything else is ignored. As
such, and to avoid (what I think is pretty unlikely) breakage of
existing code, I've retained the gdbpy_is_string check.
After the gdbpy_is_string check we call python_string_to_host_string,
if this call fails then I do now print the error, where before we
ignored the error. I think this is OK; if GDB thinks something is a
string, but still can't convert it to a string, then I think it's OK
to display the error in that case.
Another case which I was a little unsure about was in
cmdpy_completer_helper, and the call to PyObject_CallMethodObjArgs,
which is when we actually call Command.complete. Previously, if this
call resulted in an exception then we would ignore this and just
pretend there were no completions.
Of all the changes, this is possibly the one with the biggest
potential for breaking existing scripts, but also, is, I think, the
most useful change. If the user code is wrong in some way, such that
an exception is raised, then previously the user would have no obvious
feedback about this breakage. Now GDB will print the exception for
them, making it, I think, much easier to debug their extension. But,
if there is user code in the wild that relies on raising an exception
as a means to indicate there are no completions .... well, that code
is going to break after this commit. I think we can live with this
though, the exceptions means no completions thing was never documented
behaviour.
I also added a new error() call if the PyObject_CallMethodObjArgs call
raises an exception. This causes the completion mechanism within GDB
to stop. Within GDB the completion code is called twice, the first
time to compute the work break characters, and then a second time to
compute the actual completions.
If PyObject_CallMethodObjArgs raises an exception when computing the
word break character, and we print it by calling
gdbpy_print_stack_or_quit(), but then carry on as if
PyObject_CallMethodObjArgs had returns no completions, GDB will
call the Python completion code again, which results in another call
to PyObject_CallMethodObjArgs, which might raise the same exception
again. This results in the Python exception being printed twice.
By throwing a C++ exception after the failed
PyObject_CallMethodObjArgs call, the completion mechanism is aborted,
and no completions are offered. But importantly, the Python exception
is only printed once. I think this gives a much better user
experience.
I've added some tests to cover this case, as I think this is the most
likely case that a user will run into.
Approved-By: Tom Tromey <tom@tromey.com>
I spotted I made a small mistake in this commit:
commit aff250145a
Date: Fri Nov 24 12:04:36 2023 +0000
gdb: generate gdb-index identically regardless of work thread count
In this commit I added a new proc in testsuite/lib/gdb.exp called
gdb_get_worker_threads. This proc uses gdb_test_multiple with two
possible patterns. One pattern is anchored with '^', while the other
is missing the '^' which it could use.
This commit adds the missing '^'.
Fixes this issue, introduced by f9582a22db ("[gdb] Fix segfault in
for_each_block, part 1"):
CXX darwin-nat.o
/Users/smarchi/src/binutils-gdb/gdb/darwin-nat.c:1169:7: error: no matching function for call to 'breakpoint_inserted_here_p'
if (breakpoint_inserted_here_p (inf->aspace, pc))
^~~~~~~~~~~~~~~~~~~~~~~~~~
Change-Id: I3bb6be75b650319f0fa1dbdceb379b18531da96c
DAP specifies a "process" event that is sent when a process is started
or attached to. gdb was not emitting this (several DAP clients appear
to ignore it entirely), but it looked easy and harmless to implement.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30473
I noticed in gdb/tui/tui-stack.c a source-level micro-optimization where
strlen with a string literal argument:
...
strlen ("bla")
...
is replaced with sizeof:
...
sizeof ("bla") - 1
...
The benefit of this is that the optimization is also done at O0, but the
drawback is that it makes the expression harder to read.
Use const std::string to encapsulate the string literals, and use
std::string::size () instead.
I tried making the string names (PROC_PREFIX, LINE_PREFIX, PC_PREFIX and
SINGLE_KEY) lower-case, but that clashed with a pre-existing pc_prefix, so
I've left them upper-case.
Tested on x86_64-linux.
Tested-By: Alexandra Petlanova Hajkova <ahajkova@redhat.com>
The BPF assembler has been updated to follow the clang convention in
the interpretation of semicolons: they separate statements and
directives, and do not start line comments.
The BPF assembler in clang uses semi-colon (;) to separate statements,
not to be begin line comments. This patch adapts the GNU assembler
accordingly.
Testsuite and documentation updated accordingly.
2023-11-28 Jose E. Marchesi <jose.marchesi@oracle.com>
* config/tc-bpf.c: Semicolon does not start a comment, but
separates multiple commands on a single line.
* testsuite/gas/bpf/alu-pseudoc.s: Adapt test accordingly.
* testsuite/gas/bpf/spacing-pseudoc.s: Likewise.
* testsuite/gas/bpf/offset16-overflow.s: Likewise.
* testsuite/gas/bpf/jump-relax-jump.s: Likewise.
* testsuite/gas/bpf/jump-relax-ja.s: Likewise.
* testsuite/gas/bpf/imm32-overflow.s: Likewise.
* testsuite/gas/bpf/disp32-overflow.s: Likewise.
* testsuite/gas/bpf/disp16-overflow-relax.s: Likewise.
* testsuite/gas/bpf/disp16-overflow.s: Likewise.
* doc/c-bpf.texi (BPF Special Characters): Update.
The following patch attempts to use x86 SHA ISA if available to speed
up in my testing about 2.5x sha1 build-id processing (in my case on
AMD Ryzen 5 3600) while producing the same result.
I believe AArch64 has similar HW acceleration for SHA1, perhaps it
could be added similarly.
Note, seems lld uses BLAKE3 rather than md5/sha1. I think it would be
a bad idea to lie to users, if they choose --buildid=sha1, we should
be using SHA1, not some other checksum, but perhaps we could add some other
--buildid= styles and perhaps make one of the new the default.
Tested on x86_64-linux, both on Intel i9-7960X (which doesn't have
sha_ni ISA support) without/with the patch and on AMD Ryzen 5 3600
(which does have it) without/with the patch.
2023-11-28 Jakub Jelinek <jakub@redhat.com>
include/
* sha1.h (sha1_process_bytes_fn): New typedef.
(sha1_choose_process_bytes): Declare.
libiberty/
* configure.ac (HAVE_X86_SHA1_HW_SUPPORT): New check.
* sha1.c: If HAVE_X86_SHA1_HW_SUPPORT is defined, include x86intrin.h
and cpuid.h.
(sha1_hw_process_bytes, sha1_hw_process_block,
sha1_choose_process_bytes): New functions.
* config.in: Regenerated.
* configure: Regenerated.
ld/
* ldbuildid.c (generate_build_id): Use sha1_choose_process_bytes ()
instead of &sha1_process_bytes.
The make-check-all.sh script (gdb/testsuite/make-check-all.sh) is
great, it makes it super easy to run some test(s) using all the
available board files.
This commit aims to make this script even easier to access by adding a
check-all-boards target to the GDB Makefile. This new target checks
for (and requires) a number of environment variables, so the target
should be used like this:
make check-all-boards GDB_TARGET_USERNAME=remote-target \
GDB_HOST_USERNAME=remote-host \
TESTS="gdb.base/break.exp"
Where GDB_TARGET_USERNAME and GDB_HOST_USERNAME are the user names
that should be passed to the make-check-all.sh --target-user and
--host-user command line options respectively.
My personal intention is to set these variables in my environment, so
all I'll need to do is:
make check-all-boards TESTS="gdb.base/break.exp"
The make rule always passes --keep-results to the make-check-all.sh
script, as I find that the most useful. It's super frustrating to run
the tests and realise you forgot that option and the results have been
discarded.
I have been making more use of the make-check-all.sh script to run
tests against all boards.
But one thing is pretty annoying. When a test fails on some random
board, I have to run make-check-all.sh with --verbose and --dry-run in
order to see what RUNTESTFLAGS I should be using.
I always run with --keep-results on, so, in this commit, I propose
that, when --keep-results is on the 'make check' command will be
written out to a file within the stored results directory, like:
check-all/BOARD_NAME/make-check.sh
then, if I want to rerun a test, I can just:
sh check-all/BOARD_NAME/make-check.sh
and the test will be re-run for me.
Similar to the previous commit, this commit ensures that the dwarf-5
index files are generated identically as the number of worker-threads
changes.
Building the dwarf-5 index makes use of a closed hash table, the
bucket_hash local within debug_names::build(). Entries are added to
bucket_hash from m_name_to_value_set, which, in turn, is populated
by calls to debug_names::insert() in write_debug_names. The insert
calls are ordered based on the entries within the cooked_index, and
the ordering within cooked_index depends on the number of worker
threads that GDB is using.
My proposal is to sort each chain within the bucket_hash closed hash
table prior to using this to build the dwarf-5 index.
The buckets within bucket_hash will always have the same ordering (for
a given GDB build with a given executable), and by sorting the chains
within each bucket, we can be sure that GDB will see each entry in a
deterministic order.
I've extended the index creation test to cover this case.
Approved-By: Tom Tromey <tom@tromey.com>
It was observed that changing the number of worker threads that GDB
uses (maintenance set worker-threads NUM) would have an impact on the
layout of the generated gdb-index.
The cause seems to be how the CU are distributed between threads, and
then symbols that appear in multiple CU can be encountered earlier or
later depending on whether a particular CU moves between threads.
I certainly found this behaviour was reproducible when generating an
index for GDB itself, like:
gdb -q -nx -nh -batch \
-eiex 'maint set worker-threads NUM' \
-ex 'save gdb-index /tmp/'
And then setting different values for NUM will change the generated
index.
Now, the question is: does this matter?
I would like to suggest that yes, this does matter. At Red Hat we
generate a gdb-index as part of the build process, and we would
ideally like to have reproducible builds: for the same source,
compiled with the same tool-chain, we should get the exact same output
binary. And we do .... except for the index.
Now we could simply force GDB to only use a single worker thread when
we build the index, but, I don't think the idea of reproducible builds
is that strange, so I think we should ensure that our generated
indexes are always reproducible.
To achieve this, I propose that we add an extra step when building the
gdb-index file. After constructing the initial symbol hash table
contents, we will pull all the symbols out of the hash, sort them,
then re-insert them in sorted order. This will ensure that the
structure of the generated hash will remain consistent (given the same
set of symbols).
I've extended the existing index-file test to check that the generated
index doesn't change if we adjust the number of worker threads used.
Given that this test is already rather slow, I've only made one change
to the worker-thread count. Maybe this test should be changed to use
a smaller binary, which is quicker to load, and for which we could
then try many different worker thread counts.
Approved-By: Tom Tromey <tom@tromey.com>
Make static the functions add_index_entry, find_slot, and hash_expand,
member functions of the mapped_symtab class.
Fold an additional snippet of code from write_gdbindex into
mapped_symtab::minimize, this code relates to minimisation, so this
seems like a good home for it.
Make the n_elements, data, and m_string_obstack member variables of
mapped_symtab private. Provide a new obstack() member function to
provide access to the obstack when needed, and also add member
functions begin(), end(), cbegin(), and cend() so that the
mapped_symtab class can be treated like a contained and iterated
over.
I've also taken this opportunity to split out the logic for whether
the hash table (m_data) needs expanding, this is the new function
hash_needs_expanding. This will be useful in a later commit.
There should be no user visible changes after this commit.
Approved-By: Tom Tromey <tom@tromey.com>
I noticed in passing that out algorithm for generating the gdb-index
file is incorrect. When building the hash table in add_index_entry we
count every incoming entry rehash when the number of entries gets too
large. However, some of the incoming entries will be duplicates,
which don't actually result in new items being added to the hash
table.
As a result, we grow the gdb-index hash table far too often.
With an unmodified GDB, generating a gdb-index for GDB, I see a file
size of 90M, with a hash usage (in the generated index file) of just
2.6%.
With a patched GDB, generating a gdb-index for the _same_ GDB binary,
I now see a gdb-index file size of 30M, with a hash usage of 41.9%.
This is a 67% reduction in gdb-index file size.
Obviously, not every gdb-index file is going to see such big savings,
however, the larger a program, and the more symbols that are
duplicated between compilation units, the more GDB would over count,
and so, over-grow the index.
The gdb-index hash table we create has a minimum size of 1024, and
then we grow the hash when it is 75% full, doubling the hash table at
that time. Given this, then we expect that either:
a. The hash table is size 1024, and less than 75% full, or
b. The hash table is between 37.5% and 75% full.
I've include a test that checks some of these constraints -- I've not
bothered to check the upper limit, and over full hash table isn't
really a problem here, but if the fill percentage is less than 37.5%
then this indicates that we've done something wrong (obviously, I also
check for the 1024 minimum size).
Approved-By: Tom Tromey <tom@tromey.com>
Split out the code that makes a copy of the GDB executable ready for
self testing into a new proc. A later commit in this series wants to
load the GDB executable into GDB (for creating an on-disk debug
index), but doesn't need to make use of the full do_self_tests proc.
There should be no changes in what is tested after this commit.
Approved-By: Tom Tromey <tom@tromey.com>
Add proper support for option completion to the 'save gdb-index'
command. Update save_gdb_index_command function to make use of the
new option_def data structures for parsing the '-dwarf-5' option.
Approved-By: Tom Tromey <tom@tromey.com>
Add a call to gdb_tilde_expand in the save_gdb_index_command function,
this means that we can now do:
(gdb) save gdb-index ~/blah/
Previous this wouldn't work.
Approved-By: Tom Tromey <tom@tromey.com>
The previous commit describes PR gdb/30547, a segfault when running test-case
gdb.base/vfork-follow-parent.exp on powerpc64 (likewise on s390x).
The root cause for the segmentation fault is that linux_is_uclinux gives an
incorrect result: it returns true instead of false.
So, why does linux_is_uclinux:
...
int
linux_is_uclinux (void)
{
CORE_ADDR dummy;
return (target_auxv_search (AT_NULL, &dummy) > 0
&& target_auxv_search (AT_PAGESZ, &dummy) == 0);
...
return true?
This is because ppc_linux_target_wordsize returns 4 instead of 8, causing
ppc_linux_nat_target::auxv_parse to misinterpret the auxv vector.
So, why does ppc_linux_target_wordsize:
...
int
ppc_linux_target_wordsize (int tid)
{
int wordsize = 4;
/* Check for 64-bit inferior process. This is the case when the host is
64-bit, and in addition the top bit of the MSR register is set. */
long msr;
errno = 0;
msr = (long) ptrace (PTRACE_PEEKUSER, tid, PT_MSR * 8, 0);
if (errno == 0 && ppc64_64bit_inferior_p (msr))
wordsize = 8;
return wordsize;
}
...
return 4?
Specifically, we get this result because because tid == 0, so we get
errno == ESRCH.
The tid == 0 is caused by the switch_to_no_thread in
handle_vfork_child_exec_or_exit:
...
/* Switch to no-thread while running clone_program_space, so
that clone_program_space doesn't want to read the
selected frame of a dead process. */
scoped_restore_current_thread restore_thread;
switch_to_no_thread ();
inf->pspace = new program_space (maybe_new_address_space ());
...
but moving the maybe_new_address_space call to before that gives us the
same result. The tid is no longer 0, but we still get ESRCH because the
thread has exited.
Fix this in handle_vfork_child_exec_or_exit by doing the
maybe_new_address_space call in the context of the vfork parent.
Tested on top of trunk on x86_64-linux and ppc64le-linux.
Tested on top of gdb-14-branch on ppc64-linux.
Co-Authored-By: Simon Marchi <simon.marchi@polymtl.ca>
PR gdb/30547
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30547
When running test-case gdb.base/vfork-follow-parent.exp on powerpc64 (likewise
on s390x), I run into:
...
(gdb) PASS: gdb.base/vfork-follow-parent.exp: \
exec_file=vfork-follow-parent-exit: target-non-stop=on: non-stop=off: \
resolution_method=schedule-multiple: print unblock_parent = 1
continue^M
Continuing.^M
Reading symbols from vfork-follow-parent-exit...^M
^M
^M
Fatal signal: Segmentation fault^M
----- Backtrace -----^M
0x1027d3e7 gdb_internal_backtrace_1^M
src/gdb/bt-utils.c:122^M
0x1027d54f _Z22gdb_internal_backtracev^M
src/gdb/bt-utils.c:168^M
0x1057643f handle_fatal_signal^M
src/gdb/event-top.c:889^M
0x10576677 handle_sigsegv^M
src/gdb/event-top.c:962^M
0x3fffa7610477 ???^M
0x103f2144 for_each_block^M
src/gdb/dcache.c:199^M
0x103f235b _Z17dcache_invalidateP13dcache_struct^M
src/gdb/dcache.c:251^M
0x10bde8c7 _Z24target_dcache_invalidatev^M
src/gdb/target-dcache.c:50^M
...
or similar.
The root cause for the segmentation fault is that linux_is_uclinux gives an
incorrect result: it should always return false, given that we're running on a
regular linux system, but instead it returns first true, then false.
In more detail, the segmentation fault happens as follows:
- a program space with an address space is created
- a second program space is about to be created. maybe_new_address_space
is called, and because linux_is_uclinux returns true, maybe_new_address_space
returns false, and no new address space is created
- a second program space with the same address space is created
- a program space is deleted. Because linux_is_uclinux now returns false,
gdbarch_has_shared_address_space (current_inferior ()->arch ()) returns
false, and the address space is deleted
- when gdb uses the address space of the remaining program space, we run into
the segfault, because the address space is deleted.
Hardcoding linux_is_uclinux to false makes the test-case pass.
We leave addressing the root cause for the following commit in this series.
For now, prevent the segmentation fault by making the address space a refcounted
object.
This was already suggested here [1]:
...
A better solution might be to have the address spaces be reference counted
...
Tested on top of trunk on x86_64-linux and ppc64le-linux.
Tested on top of gdb-14-branch on ppc64-linux.
Co-Authored-By: Simon Marchi <simon.marchi@polymtl.ca>
PR gdb/30547
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30547
[1] https://sourceware.org/pipermail/gdb-patches/2023-October/202928.html
If a target provides a target description including registers from the
XSAVE extended region, but does not provide an XSAVE layout, use a
fallback XSAVE layout based on the included registers. This fallback
layout matches GDB's behavior in earlier releases which assumes the
layout from Intel CPUs.
This fallback layout is currently only used for remote targets since
native targets which support XSAVE provide an explicit layout derived
from CPUID.
PR gdb/30912
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30912
Approved-By: Simon Marchi <simon.marchi@efficios.com>
We have a target board cc-with-gdb-index that uses the gdb-add-index script to
add a .gdb_index index to an exec.
There is however an alternative way of adding a .gdb_index: the index-cache.
Add a new target board cc-with-index-cache.
This is not superfluous for two reasons:
- there is functionality that gdb-add-index doesn't support, but the
index-cache does: the index-cache can add an index to an exec with a
.gnu_debugaltlink (note that when using the cc-with-gdb-index board this
case is quietly ignored), and
- using the index-cache is excercised in only a few test-cases, and having
this target board extends the test coverage to the entire test suite. This
is for instance relevant because the index-cache is written by a worker
thread in the background, so we can check more thoroughly for data races
(see PR symtab/30837).
Tested on x86_64-linux.
Shell script changes checked with shellcheck.
Approved-By: Tom Tromey <tom@tromey.com>
