The real problem is that libcollector doesn't interpose thread_create@GLIBC_2.34
We interpose a lot of libC functions (dlopen, fork, pthread_create, etc.).
Some of these functions have versions. For example, dlopen@GLIBC_2.34,
dlopen@GLIBC_2.17, dlopen@GLIBC_2.2.5, etc.
We have to interpose each of the functions because we don't know
which version of libC will be used during profiling.
Historically, we have used three versions of scripts (mapfile.aarch64-Linux,
mapfile.amd64-Linux, mapfile.intel-Linux).
Three are not needed. One is enough
The fixes below include:
- merged all version symbols into one version script.
- added new version symbols which are defined in latest versions of libC.
- removed unused defines and duplicated code.
- added the DCL_FUNC_VER macro to define the version symbols.
Tested on x86_64 and aarch64 (OL8/OL9). No regression.
gprofng/ChangeLog
2023-03-23 Vladimir Mezentsev <vladimir.mezentsev@oracle.com>
PR gprofng/30089
* libcollector/Makefile.am: Use libgprofng.ver instead of mapfile.*
* libcollector/configure.ac: Delete GPROFNG_VARIANT.
* src/collector_module.h: Move the SYMVER_ATTRIBUTE macro to collector.h
* libcollector/collector.h: Add macros (SYMVER_ATTRIBUTE, DCL_FUNC_VER).
Remove unused defines.
* libcollector/dispatcher.c: Interpose functions from libC.
Clean up the old code.
* libcollector/iotrace.c: Likewise.
* libcollector/libcol_util.c: Likewise.
* libcollector/linetrace.c: Likewise.
* libcollector/mmaptrace.c: Likewise.
* libcollector/synctrace.c: Likewise.
* libcollector/libgprofng.ver: New file.
* libcollector/Makefile.in: Rebuild.
* libcollector/configure: Rebuild.
* libcollector/mapfile.aarch64-Linux: Removed.
* libcollector/mapfile.amd64-Linux: Removed.
* libcollector/mapfile.intel-Linux: Removed.
* libcollector/mapfile.sparc-Linux: Removed.
* libcollector/mapfile.sparcv9-Linux: Removed.
I noticed that some debug log output printing an lwp's pending status
wasn't considering lp->waitstatus. This fixes it, by introducing a
new pending_status_str function.
Also fix the comment in gdb/linux-nat.h describing
lwp_info::waitstatus and details the description of lwp_info::status
while at it.
Change-Id: I66e5c7a363d30a925b093b195d72925ce5b6b980
Approved-By: Andrew Burgess <aburgess@redhat.com>
This commit tweaks displaced_step_finish & friends to pass down a
target_waitstatus instead of a gdb_signal. This is needed because a
patch later in the step-over-{thread-exit,clone] series will want to
make displaced_step_buffers::finish handle
TARGET_WAITKIND_THREAD_EXITED. It also helps with the
TARGET_WAITKIND_THREAD_CLONED patch later in that same series.
It's also a bit more logical this way, as we don't have to pass down
signals when the thread didn't actually stop for a signal. So we can
also think of it as a clean up.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=27338
Change-Id: I4c5d338647b028071bc498c4e47063795a2db4c0
Approved-By: Andrew Burgess <aburgess@redhat.com>
Fix test-case gdb.stabs/weird.exp for remote host by not using an absolute
destfile argument to gdb_remote_download, which doesn't work well with remotedir.
Fix test-case gdb.gdb/unittest.exp for remote host, by:
- disabling the completion tests if readline is not used, and
- not using with_gdb_cwd $dir for remote host (because it does
not support changing to ".").
Tested on x86_64-linux.
In do_self_tests we try to find out the location of the gdb to debug, which
will then be copied and renamed to xgdb.
In principle, the host board specifies the location of GDB, on host.
With remote host, we could upload that gdb from host to build/target, but we
would miss the data directory (which is listed as the reason to skip
do_self_tests for remote target).
We could fix that by instead taking the gdb from build instead, but that
wouldn't work with installed testing.
It seems easier to just skip this on remote host.
It could be made to work for the "[is_remote host] && [is_remote target]
&& host == target" scenario (see board local-remote-host-native.exp), but
that doesn't seem worth the effort.
Tested on x86_64-linux.
A user here at AdaCore noticed that, when debugging a certain program,
a stack frame reported line 34358, where it should have been line
99894.
After debugging a bit, I discovered:
(top) p (99894 & ~65536)
$60 = 34358
That line, symbol::line is too narrow.
This patch widens the member and changes all the uses that currently
use the narrower type.
Approved-By: Simon Marchi <simon.marchi@efficios.com>
This changes gdb to use scalar arithmetic for expression evaluation.
I suspect this patch is not truly complete, as there may be code paths
that still don't correctly handle 128-bit integers. However, many
things do work now.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30190
Both value_equal and value_less use value_as_long to check a
presumably boolean result of calling value_binop. However,
value_binop in this case actually returns an int as wide as its
arguments, and this approach can then fail for integers wider than
LONGEST. Instead, rewrite this in a form that works for any size
integer.
binop_promote currently only handles integer sizes up to
builtin_long_long. However, this may not handle 128-bit types.
Simplify this code, unify the C and non-C (but not OpenCL, as I don't
know how to test this) cases, and handle 128-bit integers as well.
This still doesn't exactly follow C or C++ rules. This could be
implemented, but if so, I think it makes more sense as a C-specific
expression node.
Currently, gdb_mpz::safe_export will always make a copy of *this.
However, this copy isn't always needed. This patch makes this code
slightly more efficient, by avoiding the copy when possible.
This adds many operator overloads and other useful methods to gdb_mpz.
This is preparation for using this class for scalar arithmetic in gdb
expression evaluation.
Hannes pointed out that cooked_index_shard::do_finalize never
populates the seen_names hash table. This patch adds the necessary
store. This reduces memory use a little for "gdb gdb":
(before) Space used: 28909568 (+0 for this command)
(after) Space used: 28884992 (+0 for this command)
What this means, btw, is that in gdb there are not many symbols that
are both mentioned in many CUs and that also require name
canonicalization. It's possible this would differ in other programs.
Declare test-case gdb.dwarf2/gdb-add-index-symlink.exp unsupported for remote
host, because the current implementation of gdb_ensure_index doesn't support it.
Tested on x86_64-linux.
Fix test-case gdb.dwarf2/gdb-index.exp on remote host using
gdb_remote_download and host_standard_output_file.
Also declare the test-case unsupported with readnow.
Tested on x86_64-linux.
For test-case gdb.base/index-cache.exp and remote host, this:
...
lassign [remote_exec host sh "-c \"rm $cache_dir/*.gdb-index\""] ret
...
gives us:
...
Executing on host: sh -c rm /tmp/tmp.m3L7m2AVkL/*.gdb-index (timeout = 300)
builtin_spawn -ignore SIGHUP sh -c rm /tmp/tmp.m3L7m2AVkL/*.gdb-index^M
rm: missing operand^M
Try 'rm --help' for more information.^M
FAIL: gdb.dwarf2/per-bfd-sharing.exp: couldn't remove files in temporary cache dir
...
Fix this using quote_for_host. Likewise in gdb.dwarf2/per-bfd-sharing.exp.
Tested on x86_64-linux.
A few test-cases in gdb.dwarf2 use something like:
...
additional_flags=\"-DFOO=BAR + 10\"
...
which doesn't work on remote host.
Fix this by introducing a new proc quote_for_host that also works for remote
host, such that we have:
...
additional_flags=[quote_for_host -DFOO=BAR + 10]
...
Tested on x86_64-linux.
Proc have_index is mostly used with $binfile, which gives problems
for remote host.
Fix this by using "file tail" on the proc argument.
Tested on x86_64-linux.
For test-case gdb.dlang/dlang-start.exp, I run into:
...
UNSUPPORTED: gdb.dlang/dlang-start.exp: require failed: can_compile d
...
My distro has no support for gdc, but I'd like to have the test-case
running and passing, so let's rewrite the test-case using dwarf assembly
and add it alongside (rather than replacing it, because it's good to use
actual compiler output if we have it available).
My distro does have a package providing dmd, so let's mimic that debug info in
the dwarf assembly. This gives us:
...
(gdb) start ^M
Temporary breakpoint 1 at 0x4004ab^M
Starting program: dlang-start-2 ^M
^M
Temporary breakpoint 1, 0x00000000004004ab in _Dmain ()^M
...
The "_Dmain ()" should probably be "D main", I've filed PR30276 about that.
Also add a "show language" to check that we automatically set the language
correctly to D.
Note that the dwarf assembly also describes main, otherwise the test-case
doesn't function as regression test for commit 47fe57c928 ("Fix "start" for
D, Rust, etc").
Tested on x86_64-linux.
It's better not to drill down into u.auxent but instead use a pointer
to the combined_entry_type. That way the fix_scnlen field is
available, and no one looking at the codes needs to wonder whether
coffsymbol (symbol_get_bfdsym (sym))->native[i + 1] is the same
auxent.
* config/tc-ppc.c (ppc_frob_symbol): Tidy XCOFF auxent access.
(ppc_adjust_symtab): Likewise.
I'm fairly certain the table_end checks are redundant now. This
patch reverts commit 334d4ced42.
* coffgen.c (coff_pointerize_aux): Remove table_end parameter.
(coff_get_normalized_symtab): Adjust to suit.
long is a poor choice of type to store 32-bit values read from
objects files by H_GET_32. H_GET_32 doesn't sign extend so tests like
that in gdb/coffread.c for "negative" values won't work if long is
larger than 32 bits. If long is 32-bit then code needs to be careful
to not accidentally index negative array elements. (I'd rather see a
segfault on an unmapped 4G array index than silently reading bogus
data.) long is also a poor choice for x_sect.s_scnlen, which might
have 64-bit values. It's better to use unsigned exact width types to
avoid surprises.
I decided to change the field names too, which makes most of this
patch simply renaming. Besides that there are a few places where
casts are no longer needed, and where printf format strings or tests
need adjusting.
include/
* coff/internal.h (union internal_auxent): Use unsigned stdint
types. Rename l fields to u32 and u64 as appropriate.
bfd/
* coff-bfd.c,
* coff-rs6000.c,
* coff64-rs6000.c,
* coffcode.h,
* coffgen.c,
* cofflink.c,
* coffswap.h,
* peXXigen.c,
* xcofflink.c: Adjust to suit internal_auxent changes.
binutils/
* rdcoff.c: Adjust to suit internal_auxent changes.
gas/
* config/obj-coff.h,
* config/tc-ppc.c: Adjust to suit internal_auxent changes.
gdb/
* coffread.c,
* xcoffread.c: Adjust to suit internal_auxent changes.
ld/
* pe-dll.c: Adjust to suit internal_auxent changes.
u.auxent.x_sym.x_tagndx is a union. The p field is only valid when
fix_tag is set. This patch fixes code in coffgrok.c that accessed the
field without first checking fix_tag, and removes a whole lot of code
validating bogus pointers to prevent segfaults (which no longer
happen, I checked the referenced PR 17512 testcases). The patch also
documents this in the fix_tag comment, makes is_sym a bitfield, and
sorts the selecter fields a little.
bfd/
* coffcode.h (combined_entry_type): Make is_sym a bitfield.
Sort and comment on union selectors.
* libcoff.h: Regenerate.
binutils/
* coffgrok.c (do_type): Make aux a combined_entry_type. Test
fix_tag before accessing u.auxent.x_sym.x_tagndx.p. Remove
now unnecessary pointer bounds checking.
When given two or more DW_AT_call_file for a given function we
currently leak the concat memory.
* dwarf2.c (scan_unit_for_symbols): Don't leak on duplicate
DW_AT_call_file.
PR 30187
* options.h (class General_options): Add enable-linker-version.
* layout.cc (Layout::create_gold_note): If linker-version is enabled put the version string into the .comment section.
On openSUSE Leap 15.4, I get:
...
Running gdb.dlang/dlang-start.exp ...
gdb compile failed, default_target_compile: Can't find gdc.
UNTESTED: gdb.dlang/dlang-start.exp: failed to prepare
...
Fix this by:
- introducing a new proc can_compile, and
- requiring "can_compile d" in the test-case,
such that I have instead:
...
Running gdb.dlang/dlang-start.exp ...
UNSUPPORTED: gdb.dlang/dlang-start.exp: require failed: can_compile d
...
Tested on x86_64-linux, on openSUSE Leap 15.4 and Fedora 37.
While trying to use gdb_can_simple_compile with a d program, I ran into:
...
/data/vries/gdb/f37/build/gdb/testsuite/temp/105856/can_compile_d-105856.d: \
error: module 'can_compile_d-105856' has non-identifier characters in \
filename, use module declaration instead
...
The d compiler has a problem with the filename can_compile_d-105856.d, which
contains the pid. The pid is added by gdb_simple_compile:
...
set obj [standard_temp_file $name-[pid].$postfix]
...
but it's unnecessary because standard_temp_file already uses the pid.
Fix this by removing "[pid]" in all calls to standard_temp_file.
Tested on x86_64-linux.
Simon pointed out that with gdb.dap/*.exp and target board
native-gdbserver, we run into problems.
I see for each test-case:
...
+++ run
Traceback (most recent call last):
File "startup.py", line 146, in exec_and_log
output = gdb.execute(cmd, from_tty=True, to_string=True)
gdb.error: Don't know how to run. Try "help target".
...
Likewise with target board native-extended-gdbserver.
Fix this by:
- adding a new proc allow_dap_tests,
- using it in all the gdb.dap tests, and
- bailing out if GDBFLAGS/INTERNAL_GDBFLAGS contains
"set auto-connect-native-target off".
Tested on x86_64-linux.
Reported-By: Simon Marchi <simon.marchi@efficios.com>
Approved-By: Tom Tromey <tom@tromey.com>
The type-allocation patches introduced a small regression that was
picked up by the AdaCore internal test suite. Previously, the name of
a range type was preserved by resolve_dynamic_range, but now it is
not. This patch changes this code to preserve the name.
Reviewed-By: Bruno Larsen <blarsen@redhat.com>
The evaluate command supports a "context" parameter which tells the
adapter the context in which an evaluation occurs. One of the
supported values is "repl", which we took to mean evaluation of a gdb
command. That is what this patch implements.
Note that some gdb commands probably will not work correctly with the
rest of the protocol. For example if the user types "continue",
confusion may result.
This patch requires the earlier patch to fix up scopes in DAP.
Since commit 6d263fe46e ("Avoid bad breakpoints with --gc-sections"), there
was a silent regression on openSUSE Leap 15.4 for test-case
gdb.cp/m-static.exp, from:
...
(gdb) info variable everywhere^M
All variables matching regular expression "everywhere":^M
^M
File /home/vries/tmp.local-remote-host-native/m-static.h:^M
8: const int gnu_obj_4::everywhere;^M
(gdb)
...
to:
...
(gdb) info variable everywhere^M
All variables matching regular expression "everywhere":^M
^M
File /data/vries/gdb/src/gdb/testsuite/gdb.cp/m-static.h:^M
8: const int gnu_obj_4::everywhere;^M
^M
File /data/vries/gdb/src/gdb/testsuite/gdb.cp/m-static1.cc:^M
8: const int gnu_obj_4::everywhere;^M
(gdb)
...
Another regression was found due to that commit, and it was fixed in commit
99d679e7b3 ("[gdb/symtab] Fix "file index out of range" complaint") by
limiting the scope of the fix in the original commit.
Fix this regression by yet further limiting the scope of that fix, making sure
that this bit in dwarf_decode_lines is executed again for m-static1.cc:
...
/* Make sure a symtab is created for every file, even files
which contain only variables (i.e. no code with associated
line numbers). */
...
Tested on x86_64-linux.
PR symtab/30265
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30265
We were failing to add the offsets of the containing struct/union
in this case, leading to all offsets being relative to the unnamed
struct/union itself.
libctf/
PR libctf/30264
* ctf-types.c (ctf_member_info): Add the offset of the unnamed
member of the current struct as necessary.
* testsuite/libctf-lookup/unnamed-field-info*: New test.
ctf_dedup's intern() function does not return a dynamically allocated
string, so I just spent ten minutes auditing for obvious memory leaks
that couldn't actually happen. Update the comment to note what it
actually returns (a pointer into an atoms table: i.e. possibly not
a new string, and not so easily leakable).
libctf/
* ctf-dedup.c (intern): Update comment.
GCC 11+ complains that sym is uninitialized in ctf_symbol_next. It
isn't, but it's not quite smart enough to figure that out (it requires
domain-specific knowledge of the state of the ctf_next_t iterator
over multiple calls).
libctf/
* ctf-lookup.c (ctf_symbol_next): Initialize sym to a suitable
value for returning if never reset during the function.
