pointed out that within gdb.base/maint.exp, some regexps within a
gdb_test_multiple were failing to match a complete line, while later
regexps within the gdb_test_multiple made use of the '^' anchor, and
so assumed that earlier lines had been completely matched and removed
from expect's buffer.
When testing with READ1 set this assumption was failing.
Fix this by extending the offending patterns with a trailing '\r\n'.
Some older kernels cannot handle the newly generated R_LARCH_32/64_PCREL,
so the assembler generates R_LARCH_ADD32/64+R_LARCH_SUB32/64 by default,
and use the assembler option mthin-add-sub to generate R_LARCH_32/64_PCREL
as much as possible.
The Option of mthin-add-sub does not affect the generation of R_LARCH_32_PCREL
relocation in .eh_frame.
Tom de Vries [Sat, 7 Oct 2023 08:33:29 +0000 (10:33 +0200)]
[gdb/testsuite] Fix gdb.arch/i386-signal.exp on x86_64
On x86_64-linux, with test-case gdb.arch/i386-signal.exp I run into:
...
builtin_spawn -ignore SIGHUP gcc -fno-stack-protector i386-signal.c \
-fdiagnostics-color=never -fno-pie -g -no-pie -lm -o i386-signal^M
/tmp/cc2xydTG.s: Assembler messages:^M
/tmp/cc2xydTG.s:50: Error: operand size mismatch for `push'^M
compiler exited with status 1
output is:
/tmp/cc2xydTG.s: Assembler messages:^M
/tmp/cc2xydTG.s:50: Error: operand size mismatch for `push'^M
gdb compile failed, /tmp/cc2xydTG.s: Assembler messages:
/tmp/cc2xydTG.s:50: Error: operand size mismatch for `push'
UNTESTED: gdb.arch/i386-signal.exp: failed to compile
...
This is with gas 2.41, it compiles without problems with gas 2.40. Some more
strict checking was added in commit 5cc007751cd ("x86: further adjust
extend-to-32bit-address conditions"). This may or may not be a gas regression
( https://sourceware.org/pipermail/binutils/2023-October/129818.html ).
The offending bit is:
...
" push $sigframe\n"
...
which refers to a function:
...
" .globl sigframe\n"
"sigframe:\n"
...
The test-case passes with target board unix/-m32.
Make the test-case work by using pushq instead of push for the
is_amd64_regs_target case.
Tested on x86_64-linux, with target boards:
- unix/-m64 (is_amd64_regs_target == 1), and
- unix/-m32 (is_amd64_regs_target == 0),
Ilya Leoshkevich [Wed, 21 Jun 2023 23:03:04 +0000 (01:03 +0200)]
gdb: support rseq auxvs
Linux kernel commit commit 317c8194e6ae ("rseq: Introduce feature size
and alignment ELF auxiliary vector entries") introduced two new auxvs:
AT_RSEQ_FEATURE_SIZE and AT_RSEQ_ALIGN. Support them in GDB. This
fixes auxv.exp on kernels >= v6.3.
Change-Id: I8966c4d5c73eb7b45de6d410a9b28a6628edad2e
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30540 Approved-By: Tom Tromey <tom@tromey.com>
process-dies-while-detaching.exp: Exit early if GDB misses sync breakpoint
I'm seeing a lot of variability in the failures of
gdb.threads/process-dies-while-detaching.exp on aarch64-linux. On this
platform, a problem yet to be investigated causes GDB to miss the _exit
breakpoint. What happens next is random because after missing that
breakpoint, GDB is out of sync with the inferior. This causes the tests
following that point in the testcase to fail in a random way.
In this scenario it's better to exit the testcase early to avoid random
results in the testsuite.
We are relying on gdb_continue_to_breakpoint to return the result of
gdb_test_multiple. This is already the case because in Tcl the return
value of a function is the return value of the last command it runs. But
change gdb_continue_to_breakpoint to explicitly return this value, to make
it clear this is the intended behaviour.
Tested on aarch64-linux.
Tested-By: Guinevere Larsen <blarsen@redhat.com> Approved-By: Andrew Burgess <aburgess@redhat.com>
Neal Frager [Thu, 5 Oct 2023 12:51:03 +0000 (13:51 +0100)]
opcodes: microblaze: Add new bit-field instructions
This patches adds new bsefi and bsifi instructions.
BSEFI- The instruction shall extract a bit field from a
register and place it right-adjusted in the destination register.
The other bits in the destination register shall be set to zero.
BSIFI- The instruction shall insert a right-adjusted bit field
from a register at another position in the destination register.
The rest of the bits in the destination register shall be unchanged.
Further documentation of these instructions can be found here:
https://docs.xilinx.com/v/u/en-US/ug984-vivado-microblaze-ref
This patch has been tested for years of AMD Xilinx Yocto
releases as part of the following patch set:
Andrew Burgess [Fri, 29 Sep 2023 13:26:01 +0000 (14:26 +0100)]
gdb/NEWS: reorder some entries in the NEWS file
I spotted two entries in the NEWS file that I believe are in the wrong
place, these are:
- An entry about MI v1 being deprecated, this feels like it should
be the first entry under the 'MI changes' heading, and
- An entry for the $_shell convenience function which is currently
under the 'New commands' heading (sort of), when I think this
should be listed in the general news section.
The risc-v, loongarch, and csky gdbserver builds were broken. A use
of target_desc::expedite_regs (for each architecture) was not updated
to take account of the type change.
I've tested that this fixes the risc-v build. I haven't tested the
other architectures, but they should be fine.
Andrew Burgess [Wed, 27 Sep 2023 17:22:52 +0000 (18:22 +0100)]
gdb/testsuite: cleanup in gdb.base/args.exp
The last few commits resolved the KFAILs in gdb.base/args.exp. With
those out of the way we can clean up this test script a little.
In this commit I have:
- Stopped passing 'nowarnings' flag when building the source file.
I see no reason why this source should issue a warning,
- Moved setup of GDBFLAGS into args_test proc, callers that passed a
newline needed a small tweak, and also the matching code needs
updating for newline handling, but I think this is nicer, the
argument lists are now given just once,
- Updated comment on args_test,
- Updated other comments.
There should be no change in what is tested after this commit.
which also includes this fix as part of a larger set of changes. I'm
giving a Co-Authored-By credit to the author of that original series.
I believe this smaller fix brings some benefits on its own, though the
original series does offer additional improvements. Once this is
merged I'll take a look at rebasing and resubmitting the original series.
Co-Authored-By: Michael Weghorn <m.weghorn@posteo.de> Approved-By: Tom Tromey <tom@tromey.com>
Andrew Burgess [Wed, 27 Sep 2023 16:54:15 +0000 (17:54 +0100)]
gdbserver: handle newlines in inferior arguments
Similarly to how single quotes were mishandled, which was fixed two
commits ago, this commit fixes handling of newlines in arguments
passed to gdbserver.
We already had a test that covered this, gdb.base/args.exp, which,
when run with the native-extended-gdbserver board contained several
KFAIL covering this situation.
In this commit I remove the unnecessary, attempt to quote incoming
newlines within arguments, and do some minimal cleanup of the related
code. There is additional cleanup that can be done, but I'm leaving
that for the next commit.
Then I've removed the KFAIL from the test case, and performed some
minimal cleanup there too.
After this commit the gdb.base/args.exp is 100% passing with the
native-extended-gdbserver board file.
which also includes this fix as part of a larger set of changes. I'm
giving a Co-Authored-By credit to the author of that original series.
I believe this smaller fix brings some benefits on its own, though the
original series does offer additional improvements. Once this is
merged I'll take a look at rebasing and resubmitting the original series.
Andrew Burgess [Wed, 27 Sep 2023 16:18:01 +0000 (17:18 +0100)]
gdbserver: fix handling of trailing empty argument
When I posted the previous patch for review Andreas Schwab pointed out
that passing a trailing empty argument also doesn't work.
The fix for this is in the same area of code as the previous patch,
but is sufficiently different that I felt it deserved a patch of its
own.
I noticed that passing arguments containing single quotes to gdbserver
didn't work correctly:
gdb -ex 'set sysroot' --args /tmp/show-args
Reading symbols from /tmp/show-args...
(gdb) target extended-remote | gdbserver --once --multi - /tmp/show-args
Remote debugging using | gdbserver --once --multi - /tmp/show-args
stdin/stdout redirected
Process /tmp/show-args created; pid = 176054
Remote debugging using stdio
Reading symbols from /lib64/ld-linux-x86-64.so.2...
(No debugging symbols found in /lib64/ld-linux-x86-64.so.2)
0x00007ffff7fd3110 in _start () from /lib64/ld-linux-x86-64.so.2
(gdb) set args abc ""
(gdb) run
The program being debugged has been started already.
Start it from the beginning? (y or n) y
Starting program: /tmp/show-args \'
stdin/stdout redirected
Process /tmp/show-args created; pid = 176088
2 args are:
/tmp/show-args
abc
Done.
[Inferior 1 (process 176088) exited normally]
(gdb) target native
Done. Use the "run" command to start a process.
(gdb) run
Starting program: /tmp/show-args \'
2 args are:
/tmp/show-args
abc
The 'shows-args' program used here just prints the arguments passed to
the inferior.
Notice that when starting the inferior using the extended-remote
target there is only a single argument 'abc', while when using the
native target there is a second argument, the blank line, representing
the empty argument.
The problem here is that the vRun packet coming from GDB looks like
this (I've removing the trailing checksum):
$vRun;PROGRAM_NAME;616263;
If we compare this to a packet with only a single argument and no
trailing empty argument:
$vRun;PROGRAM_NAME;616263
Notice the lack of the trailing ';' character here.
The problem is that gdbserver processes this string in a loop. At
each point we maintain a pointer to the character just after a ';',
and then we process everything up to either the next ';' character, or
to the end of the string.
We break out of this loop when the character we start with (in that
loop iteration) is the null-character. This means in the trailing
empty argument case, we abort the loop before doing anything with the
empty argument.
In this commit I've updated the loop, we now break out using a 'break'
statement at the end of the loop if the (sub-)string we just processed
was empty, with this change we now notice the trailing empty
argument.
Andrew Burgess [Tue, 26 Sep 2023 16:32:24 +0000 (17:32 +0100)]
gdbserver: fix handling of single quote arguments
I noticed that passing arguments containing single quotes to gdbserver
didn't work correctly:
gdb -ex 'set sysroot' --args /tmp/show-args
Reading symbols from /tmp/show-args...
(gdb) target extended-remote | gdbserver --once --multi - /tmp/show-args
Remote debugging using | gdbserver --once --multi - /tmp/show-args
stdin/stdout redirected
Process /tmp/show-args created; pid = 176054
Remote debugging using stdio
Reading symbols from /lib64/ld-linux-x86-64.so.2...
(No debugging symbols found in /lib64/ld-linux-x86-64.so.2)
0x00007ffff7fd3110 in _start () from /lib64/ld-linux-x86-64.so.2
(gdb) set args \'
(gdb) r
The program being debugged has been started already.
Start it from the beginning? (y or n) y
Starting program: /tmp/show-args \'
stdin/stdout redirected
Process /tmp/show-args created; pid = 176088
2 args are:
/tmp/show-args
\'
Done.
[Inferior 1 (process 176088) exited normally]
(gdb) target native
Done. Use the "run" command to start a process.
(gdb) run
Starting program: /tmp/show-args \'
2 args are:
/tmp/show-args
'
Done.
[Inferior 1 (process 176095) exited normally]
(gdb) q
The 'shows-args' program used here just prints the arguments passed to
the inferior.
Notice that when starting the inferior using the extended-remote
target the second argument is "\'", while when running using native
target the argument is "'". The second of these is correct, the \'
used with the "set args" command is just to show GDB that the single
quote is not opening an argument string.
It turns out that the extra backslash is injected on the gdbserver
side when gdbserver processes the arguments that GDB passes it, the
code that does this was added as part of this much larger commit:
In this commit I propose removing the specific code that adds what I
believe is a stray backslash. I've extended an existing test to cover
this case, and I now see identical behaviour when using an
extended-remote target as with the native target.
This partially fixes PR gdb/27989, though there are still some issues
with newline handling which I'll address in a later commit.
which also includes this fix as part of a larger set of changes. I'm
giving a Co-Authored-By credit to the author of that original series.
I believe this smaller fix brings some benefits on its own, though the
original series does offer additional improvements. Once this is
merged I'll take a look at rebasing and resubmitting the original series.
If you want to install GDB in a custom prefix, have it look for debug info
in that prefix but also in the distro's default location (typically,
/usr/lib/debug) and run the GDB testsuite before doing "make install", you
have a bit of a problem:
Configuring GDB with '--prefix=$PREFIX' sets the GDB 'debug-file-directory'
parameter to $PREFIX/lib/debug. Unfortunately this precludes GDB from
looking for distro-installed debug info in /usr/lib/debug. For regular GDB
use you could set debug-file-directory to $PREFIX:/usr/lib/debug in
$PREFIX/etc/gdbinit so that GDB will look in both places, but if you want
to run the testsuite then that doesn't help because in that case GDB runs
with the '-nx' option.
There's the configure option '--with-separate-debug-dir' to set the default
value for 'debug-file-directory', but it accepts only one directory and not
a list. I considered modifying it to accept a list, but it's not obvious
how to do that because its value is also used by BFD, as well as processed
for "relocatability".
I thought it was simpler to add a new option to specify a list of
additional directories that will be appended to the debug-file-directory
setting.
Reviewed-By: Eli Zaretskii <eliz@gnu.org> Approved-By: Tom Tromey <tom@tromey.com>
Tom de Vries [Thu, 5 Oct 2023 21:22:11 +0000 (23:22 +0200)]
[gdb/go] Handle v3 go_0 mangled prefix
With gcc-10 we have:
...
(gdb) break package2.Foo^M
Breakpoint 2 at 0x402563: file package2.go, line 5.^M
(gdb) PASS: gdb.go/package.exp: setting breakpoint 1
...
but with gcc-11:
...
gdb) break package2.Foo^M
Function "package2.Foo" not defined.^M
Make breakpoint pending on future shared library load? (y or [n]) n^M
(gdb) FAIL: gdb.go/package.exp: gdb_breakpoint: set breakpoint at package2.Foo
...
In the gcc-10 case, though the exec contains dwarf, it's not used to set the
breakpoint (which is an independent problem, filed as PR go/30941), instead
the minimal symbol information is used.
The minimal symbol information changed between gcc-10 and gcc-11:
...
$ nm a.out.10 | grep Foo 000000000040370d T go.package2.Foo 0000000000404e50 R go.package2.Foo..f
$ nm a.out.11 | grep Foo 0000000000403857 T go_0package2.Foo 0000000000405030 R go_0package2.Foo..f
...
A new v3 mangling scheme was used. The mangling schemes define a separator
character and mangling character:
- for v2, dot is used both as separator character and mangling character, and
- for v3, dot is used as separator character and underscore as mangling
character.
For more details, see [1] and [2].
In v3, "_0" demangles to ".". [ See gcc commit a01dda3c23b ("compiler, libgo:
change mangling scheme"), function Special_char_code::Special_char_code. ]
Handle the new go_0 prefix in unpack_mangled_go_symbol, which fixes the
test-case.
Note that this doesn't fix this regression:
...
$ gccgo-10 package2.go -c -g0
$ gccgo-10 package1.go package2.o -g0
$ gdb -q -batch a.out -ex "break go.package2.Foo"
Breakpoint 1 at 0x40370d
$ gccgo-11 package2.go -c -g0
$ gccgo-11 package1.go package2.o -g0
$ gdb -q -batch a.out -ex "break go.package2.Foo"
Function "go.package2.Foo" not defined.
...
With gcc-10, we set a breakpoint on the mangled minimal symbol. That
one has simply changed for gcc-11, so it's equivalent to using:
...
$ gdb -q -batch a.out -ex "break go_0package2.Foo"
Breakpoint 1 at 0x403857
...
which does work.
Tested on x86_64-linux:
- openSUSE Leap 15.4, using gccgo-7,
- openSUSE Tumbleweed, using gccgo-13.
Simon Marchi [Wed, 4 Oct 2023 02:20:21 +0000 (22:20 -0400)]
gdb: add all_objfiles_removed observer
The new_objfile observer is currently used to indicate both when a new
objfile is added to program space (when passed non-nullptr) and when all
objfiles of a program space were just removed (when passed nullptr).
I think this is confusing (and Andrew apparently thinks so too [1]).
Add a new "all_objfiles_removed" observer to remove the second role from
"new_objfile".
Some existing users of new_objfile do nothing if the passed objfile is
nullptr. For them, we can simply drop the nullptr check. For others,
add a new all_objfiles_removed callback, and refactor things a bit to
keep the existing behavior as much as possible.
Some callbacks relied on current_program_space, and following
the refactoring now use either objfile->pspace or the pspace passed to
all_objfiles_removed. I think this should be relatively safe, and in
general a step in the right direction.
On the notify side, I found only one call site to change from
new_objfile to all_objfiles_removed, in clear_symtab_users. It is not
entirely clear to me that this is entirely correct. clear_symtab_users
appears to be called in spots that don't remove all objfiles
(functions finish_new_objfile, remove_symbol_file_command, reread_symbols,
do_module_cleanups). But I think that this patch at least makes the
current code clearer.
Simon Marchi [Wed, 4 Oct 2023 02:20:18 +0000 (22:20 -0400)]
gdb: add program_space parameter to emit_clear_objfiles_event
Add program_space space parameters to emit_clear_objfiles_event and
create_clear_objfiles_event_object, making the reference to
current_program_space bubble up a bit.
Change-Id: I5fde2071712781e5d45971fa0ab34d85d3a49a71 Approved-By: Tom Tromey <tom@tromey.com>
Simon Marchi [Wed, 4 Oct 2023 02:20:17 +0000 (22:20 -0400)]
gdb: add program_space parameters to some functions in symtab.c
Add some program_space parameters to functions related to getting and
setting the main name, making the references to current_program_space
bubble up a bit. find_main_name calls ada_main_name, which implicitly
relies on the current program space, so I didn't add a parameter to that
function.
Change-Id: I9996955e8ae56832bbd461964d978e700e6feaf4 Approved-By: Tom Tromey <tom@tromey.com>
Andrew Burgess [Tue, 3 Oct 2023 16:20:14 +0000 (17:20 +0100)]
gdb: fix auxv cache clearing from new_objfile observer
It was pointed out on the mailing list that a recently added
test (gdb.python/py-progspace-events.exp) was failing when run with
the native-extended-gdbserver board. This test was added with this
commit:
gdb: add Python events for program space addition and removal
It turns out though that the test is failing due to a existing bug
in GDB, the new test just exposes the problem. Additionally, the
failure really doesn't even rely on the new functionality added in the
above commit. I reduced the test to a simple set of steps that
reproduced the failure and tested against GDB 13, and the test passes;
so the bug was introduced since then. In fact, the bug was introduced
with this commit:
gdb: remove final user of the executable_changed observer
This commit changed how the per-inferior auxv data cache is managed,
specifically, when the cache is cleared, and it is this that leads to
the failure.
This bug is interesting because it exposes a number of issues with
GDB, I'll explain all of the problems I see, though ultimately, I only
propose fixing one problem in this commit, which is enough to resolve
the crash we are currently seeing.
The crash that we are seeing manifests like this:
...
[Inferior 2 (process 3970384) exited normally]
+inferior 1
[Switching to inferior 1 [process 3970383] (/tmp/build/gdb/testsuite/outputs/gdb.python/py-progspace-events/py-progspace-events)]
[Switching to thread 1.1 (Thread 3970383.3970383)]
#0 breakpt () at /tmp/build/gdb/testsuite/../../../src/gdb/testsuite/gdb.python/py-progspace-events.c:28
28 { /* Nothing. */ }
(gdb) step
+step
terminate called after throwing an instance of 'gdb_exception_error'
Fatal signal: Aborted
... etc ...
What's happening is that GDB attempts to refill the auxv cache as a
result of the gdbarch_has_shared_address_space call in
program_space::~program_space, the backtrace looks like this:
#0 0x00007fb4f419a9a5 in raise () from /lib64/libpthread.so.0
#1 0x00000000008b635d in handle_fatal_signal (sig=6) at ../../src/gdb/event-top.c:912
#2 <signal handler called>
#3 0x00007fb4f38e3625 in raise () from /lib64/libc.so.6
#4 0x00007fb4f38cc8d9 in abort () from /lib64/libc.so.6
#5 0x00007fb4f3c70756 in __gnu_cxx::__verbose_terminate_handler() [clone .cold] () from /lib64/libstdc++.so.6
#6 0x00007fb4f3c7c6dc in __cxxabiv1::__terminate(void (*)()) () from /lib64/libstdc++.so.6
#7 0x00007fb4f3c7b6e9 in __cxa_call_terminate () from /lib64/libstdc++.so.6
#8 0x00007fb4f3c7c094 in __gxx_personality_v0 () from /lib64/libstdc++.so.6
#9 0x00007fb4f3a80c63 in _Unwind_RaiseException_Phase2 () from /lib64/libgcc_s.so.1
#10 0x00007fb4f3a8154e in _Unwind_Resume () from /lib64/libgcc_s.so.1
#11 0x0000000000e8832d in target_read_alloc_1<unsigned char> (ops=0x408a3a0, object=TARGET_OBJECT_AUXV, annex=0x0) at ../../src/gdb/target.c:2266
#12 0x0000000000e73dea in target_read_alloc (ops=0x408a3a0, object=TARGET_OBJECT_AUXV, annex=0x0) at ../../src/gdb/target.c:2315
#13 0x000000000058248c in target_read_auxv_raw (ops=0x408a3a0) at ../../src/gdb/auxv.c:379
#14 0x000000000058243d in target_read_auxv () at ../../src/gdb/auxv.c:368
#15 0x000000000058255c in target_auxv_search (match=0x0, valp=0x7ffdee17c598) at ../../src/gdb/auxv.c:415
#16 0x0000000000a464bb in linux_is_uclinux () at ../../src/gdb/linux-tdep.c:433
#17 0x0000000000a464f6 in linux_has_shared_address_space (gdbarch=0x409a2d0) at ../../src/gdb/linux-tdep.c:440
#18 0x0000000000510eae in gdbarch_has_shared_address_space (gdbarch=0x409a2d0) at ../../src/gdb/gdbarch.c:4889
#19 0x0000000000bc7558 in program_space::~program_space (this=0x4544aa0, __in_chrg=<optimized out>) at ../../src/gdb/progspace.c:124
#20 0x00000000009b245d in delete_inferior (inf=0x47b3de0) at ../../src/gdb/inferior.c:290
#21 0x00000000009b2c10 in prune_inferiors () at ../../src/gdb/inferior.c:480
#22 0x00000000009c5e3e in fetch_inferior_event () at ../../src/gdb/infrun.c:4558
#23 0x000000000099b4dc in inferior_event_handler (event_type=INF_REG_EVENT) at ../../src/gdb/inf-loop.c:42
#24 0x0000000000cbc64f in remote_async_serial_handler (scb=0x4090a30, context=0x408a6b0) at ../../src/gdb/remote.c:14859
#25 0x0000000000d83d3a in run_async_handler_and_reschedule (scb=0x4090a30) at ../../src/gdb/ser-base.c:138
#26 0x0000000000d83e1f in fd_event (error=0, context=0x4090a30) at ../../src/gdb/ser-base.c:189
So this is problem #1, if we throw an exception while deleting a
program_space then this is not caught, and is going to crash GDB.
Problem #2 becomes evident when we ask why GDB is throwing an error in
this case; the error is thrown because the remote target, operating in
non-async mode, can't read the auxv data while an inferior is running
and GDB is waiting for a stop reply. The problem here then, is why
does GDB get into a position where it tries to interact with the
remote target in this way, at this time? The problem is caused by the
prune_inferiors call which can be seen in the above backtrace.
In prune_inferiors we check if the inferior is deletable, and if it
is, we delete it. The problem is, I think, we should also check if
the target is currently in a state that would allow us to delete the
inferior. We don't currently have such a check available, we'd need
to add one, but for the remote target, this would return false if the
remote is in async mode and the remote is currently waiting for a stop
reply. With this change in place GDB would defer deleting the
inferior until the remote target has stopped, at which point GDB would
be able to refill the auxv cache successfully.
And then, problem #3 becomes evident when we ask why GDB is needing to
refill the auxv cache now when it didn't need to for GDB 13. This is
where the second commit mentioned above (a2827364e2bf) comes in.
Prior to this commit, the auxv cache was cleared by the
executable_changed observer, while after that commit the auxv cache
was cleared by the new_objfile observer -- but only when the
new_objfile observer is used in the special mode that actually means
that all objfiles have been unloaded (I know, the overloading of the
new_objfile observer is horrible, and unnecessary, but it's not really
important for this bug).
The difference arises because the new_objfile observer is triggered
from clear_symtab_users, which in turn is called from
program_space::~program_space. The new_objfile observer for auxv does
this:
That is, when all the objfiles are unloaded, we clear the auxv cache
for the current inferior.
The problem is, then when we look at the prune_inferiors ->
delete_inferior -> ~program_space path, we see that the current
inferior is not going to be an inferior that exists within the
program_space being deleted; delete_inferior removes the deleted
inferior from the global inferior list, and then only deletes the
program_space if program_space::empty() returns true, which is only
the case if the current inferior isn't within the program_space to
delete, and no other inferior exists within that program_space
either.
What this means is that when the new_objfile observer is called we
can't rely on the current inferior having any relationship with the
program space in which the objfiles were removed. This was an error
in the commit a2827364e2bf, the only thing we can rely on is the
current program space. As a result of this mistake, after commit a2827364e2bf, GDB was sometimes clearing the auxv cache for a random
inferior. In the native target case this was harmless as we can
always refill the cache when needed, but in the remote target case, if
we need to refill the cache when the remote target is executing, then
we get the crash we observed.
And additionally, if we think about this a little more, we see that
commit a2827364e2bf made another mistake. When all the objfiles are
removed, they are removed from a program_space, a program_space might
contain multiple inferiors, so surely, we should clear the auxv cache
for all of the matching inferiors?
Given these two insights, that the current_inferior is not relevant,
only the current_program_space, and that we should be clearing the
cache for all inferiors in the current_program_space, we can update
auxv_new_objfile_observer to:
if (objfile == nullptr)
{
for (inferior *inf : all_inferiors ())
{
if (inf->pspace == current_program_space)
invalidate_auxv_cache_inf (inf);
}
}
With this change we now correctly clear the auxv cache for the correct
inferiors, and GDB no longer needs to refill the cache at an
inconvenient time, this avoids the crash we were seeing.
And finally, we reach problem #4. Inspired by the observation that
using the current_inferior from within the ~program_space function was
not correct, I added some debug to see if current_inferior() was
called anywhere else (below ~program_space), and the answer is yes,
it's called a often. Mostly the culprit is GDB doing:
current_inferior ()->top_target ()-> ....
But I think all of these calls are most likely doing the wrong thing,
and only work because the top target in all these cases is shared
between all inferiors, e.g. it's the native target, or the remote
target for all inferiors. But if we had a truly multi-connection
setup, then we might start to see odd behaviour.
Problem #1 I'm just ignoring for now, I guess at some point we might
run into this again, and then we'd need to solve this. But in this
case I wasn't sure what a "good" solution would look like. We need
the auxv data in order to implement the linux_is_uclinux() function.
If we can't get the auxv data then what should we do, assume yes, or
assume no? The right answer would probably be to propagate the error
back up the stack, but then we reach ~program_space, and throwing
exceptions from a destructor is problematic, so we'd need to catch and
deal at this point. The linux_is_uclinux() call is made from within
gdbarch_has_shared_address_space(), which is used like:
if (!gdbarch_has_shared_address_space (target_gdbarch ()))
delete this->aspace;
So, we would have to choose; delete the address space or not. If we
delete it on error, then we might delete an address space that is
shared within another program space. If we don't delete the address
space, then we might leak it. Neither choice is great.
A better solution might be to have the address spaces be reference
counted, then we could remove the gdbarch_has_shared_address_space
call completely, and just rely on the reference count to auto-delete
the address space when appropriate.
The solution for problem #2 I already hinted at above, we should have
a new target_can_delete_inferiors() call, which should be called from
prune_inferiors, this would prevent GDB from trying to delete
inferiors when a (remote) target is in a state where we know it can't
delete the inferior. Deleting an inferior often (always?) requires
sending packets to the remote, and if the remote is waiting for a stop
reply then this will never work, so the pruning should be deferred in
this case.
The solution for problem #3 is included in this commit.
And, for problem #4, I'm not sure what the right solution is. Maybe
delete_inferior should ensure the inferior to be deleted is in place
when ~program_space is called? But that seems a little weird, as the
current inferior would, in theory, still be using the current
program_space...
Anyway, after this commit, the gdb.python/py-progspace-events.exp test
now passes when run with the native-extended-remote board.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30935 Approved-By: Simon Marchi <simon.marchi@efficios.com>
Change-Id: I41f0e6e2d7ecc1e5e55ec170f37acd4052f46eaf
When running from a distribution tarball, bison should not be necessary.
The generated files (QLParser.tab.cc, QLParser.tab.hh) should be distributed.
configure.ac should not abort if bison is missing.
configure.ac should remove temporary files (dummy.c, Simple.class).
bison must be run once to create QLParser.tab.cc and QLParser.tab.hh.
gprofng/ChangeLog
2023-10-03 Vladimir Mezentsev <vladimir.mezentsev@oracle.com>
PR gprofng/30894
* configure.ac: Don't abort if bison is missing. Remove temporary files.
* src/Makefile.am: Distribute QLParser.tab.cc and QLParser.tab.hh.
* Run bison once to create QLParser.tab.cc and QLParser.tab.hh.
* configure: Rebuild.
* src/Makefile.in: Rebuild.
Andrew Burgess [Thu, 21 Sep 2023 15:35:30 +0000 (16:35 +0100)]
gdb: fix reread_symbols when an objfile has target: prefix
When using a remote target, it is possible to tell GDB that the
executable to be debugged is located on the remote machine, like this:
(gdb) target extended-remote :54321
... snip ...
(gdb) file target:/tmp/hello.x
Reading /tmp/hello.x from remote target...
warning: File transfers from remote targets can be slow. Use "set sysroot" to access files locally instead.
Reading /tmp/hello.x from remote target...
Reading symbols from target:/tmp/hello.x...
(gdb)
So far so good. However, when we try to start the inferior we run
into a small problem:
(gdb) set remote exec-file /tmp/hello.x
(gdb) start
`target:/tmp/hello.x' has disappeared; keeping its symbols.
Temporary breakpoint 1 at 0x401198: file /tmp/hello.c, line 18.
Starting program: target:/tmp/hello.x
... snip ...
Temporary breakpoint 1, main () at /tmp/hello.c:18
18 printf ("Hello World\n");
(gdb)
Notice this line:
`target:/tmp/hello.x' has disappeared; keeping its symbols.
That's wrong, the executable hasn't been removed, GDB just doesn't
know how to check if the remote file has changed, and so falls back to
assuming that the file has been removed.
In this commit I update reread_symbols to use bfd_stat instead of
a direct stat call, this adds support for target: files, and fixes the
problem.
However, that patch never got merged, and seemed to get stuck
discussing issues around gnulib stat vs system stat as used by BFD.
I didn't 100% understand the issues discussed in that thread, however,
I think the problem with the previous thread related to the changes in
gdb_bfd.c, rather than to the change in symfile.c. As such, I think
this change might be acceptable, my reasoning is:
- the objfile::mtime field is set by a call to bfd_get_mtime (see
objfiles.c), which calls bfd_stat under the hood. This will end
up using the system stat,
- In symfile.c we currently call stat directly, which will call the
gnulib stat, which, if I understand the above thread correctly,
might give a different result to the system stat in some cases,
- By switching to using bfd_stat in symfile.c we should now be
consistently calling the system stat.
There is another issue that came up during testing that this commit
fixes. Consider this GDB session:
$ gdb -q
(gdb) target extended-remote | ./gdbserver/gdbserver --multi --once -
Remote debugging using | ./gdbserver/gdbserver --multi --once -
Remote debugging using stdio
(gdb) file /tmp/hello.x
Reading symbols from /tmp/hello.x...
(gdb) set remote exec-file /tmp/hello.x
(gdb) start
... snip ...
(gdb) load
`system-supplied DSO at 0x7ffff7fcf000' has disappeared; keeping its symbols.
Loading section .interp, size 0x1c lma 0x4002a8
... snip ...
Start address 0x0000000000401050, load size 2004
Transfer rate: 326 KB/sec, 87 bytes/write.
Notice this line:
`system-supplied DSO at 0x7ffff7fcf000' has disappeared; keeping its symbols.
We actually see the same output, for the same reasons, when using a
native target, like this:
$ gdb -q
(gdb) file /tmp/hello.x
Reading symbols from /tmp/hello.x...
(gdb) start
... snip ...
(gdb) load
`system-supplied DSO at 0x7ffff7fcf000' has disappeared; keeping its symbols.
You can't do that when your target is `native'
(gdb)
In both cases this line appears because load_command (symfile.c) calls
reread_symbols, and reread_symbols loops over every currently loaded
objfile and tries to check if the file has changed on disk by calling
stat.
However, the `system-supplied DSO at 0x7ffff7fcf000' is an in-memory
BFD, the filename for this BFD is literally the string
'system-supplied DSO at 0x7ffff7fcf000'.
Before this commit GDB would try to use the system 'stat' call to stat
the file `system-supplied DSO at 0x7ffff7fcf000', which obviously
fails; there's no file with that name (usually). As a consequence of
the stat failing GDB prints the ' .... has disappeared ...' line.
Initially, all this commit did was switch from using 'stat' to using
'bfd_stat'. Calling bfd_stat on an in-memory BFD works just fine,
however, BFD just fills the 'struct stat' buffer with zeros (except
for the file size), see memory_bstat in bfd/bfdio.c.
However, there is a bit of a weirdness about in-memory BFDs. When
they are initially created the libbfd caches an mtime within the bfd
object, this is done in bfd_from_remote_memory (elfcode.h), the cached
mtime is the time at which the in-memory BFD is created.
What this means is that when GDB creates the in-memory BFD, and we
call bfd_get_mtime(), the value returned, which GDB caches within
objfile::mtime is the creation time of the in-memory BFD. But, when
this patch changes to use bfd_stat() we now get back 0, and so we
believe that the in-memory BFD has changed. This is a change in
behaviour.
To avoid this change in behaviour, in this commit, I propose that we
always skip in-memory BFDs in reread_symbols. This preserves the
behaviour from before this commit -- mostly.
As I'm not specifically checking for, and then skipping, in-memory
BFDs, we no longer see this line:
`system-supplied DSO at 0x7ffff7fcf000' has disappeared; keeping its symbols.
Which I think is an improvement.
Co-Authored-By: Tom Tromey <tromey@adacore.com> Approved-By: Tom Tromey <tom@tromey.com>
Andrew Burgess [Sun, 24 Sep 2023 11:37:40 +0000 (12:37 +0100)]
gdb: remove print_sys_errmsg
This started with me running into this comment in symfile.c:
/* FIXME, should use print_sys_errmsg but it's not filtered. */
gdb_printf (_("`%ps' has disappeared; keeping its symbols.\n"),
styled_string (file_name_style.style (), filename));
In this particular case I think I disagree with the comment; I think
the output should be a warning rather than just a message printed to
gdb_stdout, I think when the executable, or some other objfile that is
currently being debugged, disappears from disk, this is likely an
unexpected situation, and worth warning the user about.
So, in theory, I could just call print_sys_errmsg and remove the
comment, but that would mean loosing the filename styling in the
output... so in the end I remove the comment and updated the code to
call warning.
But that got me looking at print_sys_errmsg and how it's used.
Currently the function takes a string and an errno, and prints, to
stderr, the string followed by the result of calling strerror on the
errno.
In some places the string passed to print_sys_errmsg is just a
filename, and this is used when something goes wrong. In these cases,
I think calling warning rather than gdb_printf to gdb_stderr, would be
better, and in fact, in a couple of places we manually print a
"warning" prefix, and then call print_sys_errmsg. And so, for these
users I have added a new function warning_filename_and_errno, which
takes a filename, which is printed with styling, and an errno, which
is passed through strerror and the resulting string printed. This new
function calls warning to print its output. I then updated some of
the print_sys_errmsg users to use this new function.
Some other users of print_sys_errmsg are also emitting what is clearly
a warning, however, the string being passed in is more than just a
filename, so the new warning_filename_and_errno function can't be
used, it would style the whole string. For these users I have
switched to calling warning directly, this allows me to style the
warning message correctly.
Finally, in inflow.c there is one last call to print_sys_errmsg, in
this case I just inlined the definition of print_sys_errmsg. This is
a really weird case, as after printing this message GDB just does a
hard exit. This is pretty old code, dating back to the initial GDB
import, I guess it should be updated to call error() maybe, but I'm
reluctant to make this change as part of this commit, just in case
there's some reason why we can't throw an error at this point.
With that done there are now no users of print_sys_errmsg, and so the
old function can be removed.
While I was doing all of the above I added some additional filename
styling in soure.c, this is in an else block where the if contained
the print_sys_errmsg call, so these felt related.
And finally, while I was updating the uses of print_sys_errmsg in
procfs.c, I noticed that we used a static errmsg buffer to format some
error strings. As the above changes got rid of one of the users of
errmsg I also removed the other two users, and the static buffer.
There were a couple of tests that depended on the existing output
message format that needed updating. In one case we gained an extra
'warning: ' prefix, and in the other 'Warning: ' becomes 'warning: ',
I think in both cases the new output is an improvement.
Andrew Burgess [Sun, 24 Sep 2023 12:37:06 +0000 (13:37 +0100)]
gdb: remove use of a static buffer for building error strings
I noticed in procfs.c that we use a static buffer for building error
strings when we could easily use std::string and string_printf to
achieve the same result, this commit does that.
I ran into this while performing a further refactor/cleanup that will
be presented in a later patch in this series, and thought this was
worth splitting out into its own patch.
As far as I can tell, only Solaris uses procfs.c, so I did a test
build on a Solaris machine, and I don't believe that I've broken
anything.
There should be no user visible changes after this commit.
Andrew Burgess [Sat, 23 Sep 2023 13:29:40 +0000 (14:29 +0100)]
gdb: use archive name in warning when appropriate
While working on some other patch I noticed that in reread_symbols
there is a diagnostic message that can be printed, and in some cases
we might use the wrong filename in the message.
The code in question is checking to see if an objfile has changed on
disk, we do this by stat-ing the on disk file and checking the mtime.
If this file has been removed from disk then we print a message that
the file has been removed, however, if the objfile is within an
archive then we stat the archive itself, but then warn that the
component within the archive has disappeared. I think it makes more
sense to say that the archive has disappeared.
Check library name rather than member name when rereading symbols.
Though this just makes the code to stat the archive unconditional, the
code in question existed before this commit.
However, the above commit doesn't include any tests, and seems to
indicate that the problem being addressed was seen on Darwin. I'm not
sure how to setup a test where GDB is using an objfile from within an
archive, and so there's no tests for this commit...
... but if someone can let me know how I can setup a suitable test,
please let me know and I'll try to get something working.
Guinevere Larsen [Fri, 11 Aug 2023 09:58:11 +0000 (11:58 +0200)]
gdb/testsuite: XFAIL some gdb.base/fileio.exp
Some gdb.base/fileio.exp tests expect the inferior to not have write
access to some files. If the test is being run as root, this is never
possible. This commit adds a way to identify if the user is root and
xfails the tests that expect no write access.
The linker will generate warnings if it is creating an executable
stack or a segment with all three read, write and execute permissions.
These settings are not appropriate for all targets including
MicroBlaze.
Signed-off-by: Neal Frager <neal.frager@amd.com> Signed-off-by: Michael J. Eager <eager@eagercon.com>
Luis Machado [Fri, 16 Jun 2023 14:02:39 +0000 (15:02 +0100)]
sme2: Extend SME tests to include SME2
Reusing the SME tests, this patch introduces additional tests to exercise
reading/writing ZT0, availability of the register set, signal context reading
for ZT0 and also core file generation.
Luis Machado [Wed, 14 Jun 2023 08:28:35 +0000 (09:28 +0100)]
sme2: Core file support for ZT register set
This patch adds support for ZT register dumps/reads for core files. The
ZT register is available when the SME2 feature is advertised as available
by the Linux Kernel.
Unlike the enablement for SME1 and the ZA register, support for SME2 is rather
simple given the fixed size of the ZT0 register.
Luis Machado [Tue, 4 Apr 2023 16:28:51 +0000 (17:28 +0100)]
sme2: Enable SME2 support in gdbserver
This patch teaches gdbserver about the SME2 and the ZT0 register.
Since most of the code used by gdbserver for SME2 is shared with gdb, this
is a rather small patch that reuses most of the code put in place for native
AArch64 Linux.
Luis Machado [Tue, 4 Apr 2023 16:20:46 +0000 (17:20 +0100)]
sme2: Enable SME2 for AArch64 gdb on Linux
SME2 defines a new 512-bit register named ZT0, and it is only available
if SME is also supported. The ZT0 state is valid only if the SVCR ZA bit
is enabled. Otherwise its contents are empty (0).
The target description is dynamic and gets generated at runtime based on the
availability of the feature.
Luis Machado [Mon, 10 Oct 2022 12:31:44 +0000 (13:31 +0100)]
sme: Add SVE/SME testcases
Add 5 SVE/SME tests to exercise all the new features like reading/writing
registers, pseudo-registers, signal frames and core files.
- Sanity check for SME: Gives a brief smoke test to make sure the most basic
of features are working correctly.
- ZA unavailability tests: Validates the behavior/content of the ZA register
is correct when no payload is available. It also exercises changing the
vector lengths.
- ZA availability tests: These tests exercise reading/writing to all the
possible ZA pseudo-registers, and validates the state is correct.
- Core file tests: Validates that core file reading and writing works
correctly and that all state dumped/loaded is sane. This is exercised for
both Linux Kernel core files and gcore core files.
- Signal frame tests: Validates the correct restoration of SME/SVE/FPSIMD
values across signal frames.
Since some of these tests are very lengthy and take a little while to run
(under QEMU at the moment), I decided to parallelize them into smaller
chunks so we can throw some more CPU power at them so they run faster.
I'd still like to add a few more tests to give the testsuite more coverage
in the areas of SME/SVE. Hopefully in the near future that will happen.
Just a reminder that these SME tests are currently unsupported when gdb is
connected to a remote target. That's because the RSP doesn't support
communicating changes in vector lenghts mid-execution, so gdb will always
get wrong state from the remote target.
Luis Machado [Thu, 7 Sep 2023 15:20:15 +0000 (16:20 +0100)]
corefile/bug: Add hook to control the use of target description notes from corefiles
Due to the nature of the AArch64 SVE/SME extensions in GDB, each thread
can potentially have distinct target descriptions/gdbarches.
When loading a gcore-generated core file, at the moment GDB gives priority
to the target description dumped to NT_GDB_TDESC. Though technically correct
for most targets, it doesn't work correctly for AArch64 with SVE or SME
support.
The correct approach for AArch64/Linux is to either have per-thread target
description notes in the corefiles or to rely on the
gdbarch_core_read_description hook, so it can figure out the proper target
description for a given thread based on the various available register notes.
The former, although more correct, doesn't address the case of existing gdb's
that only output a single target description note.
This patch goes for the latter, and adds a new gdbarch hook to conditionalize
the use of the corefile target description note. The hook is called
use_target_description_from_corefile_notes.
The hook defaults to returning true, meaning targets will use the corefile
target description note. AArch64 Linux overrides the hook to return false
when it detects any of the SVE or SME register notes in the corefile.
Otherwise it should be fine for AArch64 Linux to use the corefile target
description note.
When we support per-thread target description notes, then we can augment
the AArch64 Linux hook to rely on those notes.
Regression-tested on aarch64-linux Ubuntu 22.04/20.04.
Approved-By: Simon Marchi <simon.marchi@efficios.com> Reviewed-by: Thiago Jung Bauermann <thiago.bauermann@linaro.org>
Luis Machado [Tue, 31 Jan 2023 17:17:09 +0000 (17:17 +0000)]
corefile/bug: Use thread-specific gdbarch when dumping register state to core files
When we have a core file generated by gdb (via the gcore command), gdb dumps
the target description to a note. During loading of that core file, gdb will
first try to load that saved target description.
This works fine for almost all architectures. But AArch64 has a few
dynamically-generated target descriptions/gdbarch depending on the vector
length that was in use at the time the core file was generated.
The target description gdb dumps to the core file note is the one generated
at the time of attachment/startup. If, for example, the SVE vector length
changed during execution, this would not reflect on the core file, as gdb
would still dump the initial target description.
Another issue is that the gdbarch potentially doesn't match the thread's
real gdbarch, and so things like the register cache may have different formats
and sizes.
To address this, fetch the thread's architecture before dumping its register
state. That way we will always use the correct target description/gdbarch.
Approved-By: Simon Marchi <simon.marchi@efficios.com> Approved-By: Tom Tromey <tom@tromey.com> Reviewed-by: Thiago Jung Bauermann <thiago.bauermann@linaro.org>
Luis Machado [Mon, 11 Sep 2023 09:39:48 +0000 (10:39 +0100)]
Get rid of linux-core-thread-data
This struct type seems to have been used in the past as a callback
parameter. Now it seems that case is no longer true, so we can simplify
things by passing the individual parameters linux_core_thread_data
encapsulates directly to the functions.
This is just a cleanup before the next change.
Approved-By: Simon Marchi <simon.marchi@efficios.com>
Luis Machado [Thu, 16 Mar 2023 12:14:18 +0000 (12:14 +0000)]
sme: Support TPIDR2 signal frame context
The Linux Kernel defines a separate context for the TPIDR2 register in a
signal frame. Handle this additional context in gdb so this register
gets restored properly when unwinding through signal frames.
The TPIDR2 register is closely related to SME, and is available when SME
support is reported.
This is tested by testcases that are available in a later patch in the series.
Regressions-tested on aarch64-linux Ubuntu 22.04/20.04.
Luis Machado [Thu, 2 Feb 2023 14:00:58 +0000 (14:00 +0000)]
sme: Fixup sigframe gdbarch when vg/svg changes
With SME, where you have two different vector lengths (vl and svl), it may be
the case that the current frame has a set of vector lengths (A) but the signal
context has a distinct set of vector lengths (B).
In this case, we may run into a situation where GDB attempts to use a gdbarch
created for set A, but it is really dealing with a frame that was using set
B.
This is problematic, specially with SME, because now we have a different
number of pseudo-registers and types that gets cached on creation of each
gdbarch variation.
For AArch64 we really need to be able to use the correct gdbarch for each
frame, and I noticed the signal frame (tramp-frame) doesn't have a settable
prev_arch field. So it ends up using the default frame_unwind_arch function
and eventually calling get_frame_arch (next_frame). That means the previous
frame will always have the same gdbarch as the current frame.
This patch first refactors the AArch64/Linux signal context code, simplifying
it and making it reusable for our purposes of calculating the previous frame's
gdbarch.
I introduced a struct that holds information that we have found in the signal
context, and with which we can make various decisions.
Finally, a small change to tramp-frame.c and tramp-frame.h to expose a
prev_arch hook that the architecture can set.
With this new field, AArch64/Linux can implement a hook that looks at the
signal context and infers the gdbarch for the previous frame.
Regression-tested on aarch64-linux Ubuntu 22.04/20.04.
Approved-By: Simon Marchi <simon.marchi@efficios.com> Reviewed-by: Thiago Jung Bauermann <thiago.bauermann@linaro.org>
Luis Machado [Mon, 20 Feb 2023 09:56:19 +0000 (09:56 +0000)]
sve: Fix signal frame z/v register restore
While doing some SME work, I ran into the situation where the Z register
contents restored from a signal frame are incorrect if the signal frame
only contains fpsimd state and no sve state.
This happens because we only restore the v register values in that case,
and don't do anything for the z registers.
Fix this by initializing the z registers to 0 and then copying over the
overlapping part of the v registers to the z registers.
While at it, refactor the code a bit to simplify it and make it smaller.
Regression-tested on aarch64-linux Ubuntu 22.04/20.04.
Luis Machado [Tue, 15 Nov 2022 09:07:09 +0000 (09:07 +0000)]
sme: Add support for SME
Enable SME support in gdbserver by adjusting the usual fields. There is
not much to this patch because the code is either in gdb or it is shared
between gdbserver and gdb. One exception is the bump to gdbserver's
PBUFSIZ from 18432 to 131104.
Since the ZA register can be quite big (256 * 256 bytes), the g/G remote
packet will also become quite big
From gdbserver/tdesc.cc:init_target_desc, I estimated the new size should
be at least (2 * 256 * 256 + 32), which yields 131104.
It is also unlikely we will find a process starting up with SVL set to 256.
Ideally we'd adjust the packet size dynamically based on what we need, but
for now this should do.
Please note we have the same limitation for SME that we have for SVE, and
that is the fact gdbserver cannot communicate vector length changes to gdb
via the remote protocol.
Thiago is working on this improvement, which hopefully will be able to be
adapted to SME in an easy way.
Luis Machado [Mon, 11 Sep 2023 11:42:00 +0000 (12:42 +0100)]
Convert tdesc's expedite_regs to a string vector
Right now the list of expedited registers is stored as an array of char *,
with a nullptr element at the end to signal its last element.
Convert expedite_regs to a std::vector of std::string so it is easier to
manage the elements and the storage is handled automatically.
Eventually we might want to convert all the target functions so they pass a
std::vector of std::string as well. Or maybe expose an interface that target can
use to add expedited registers on-by-one depending on the target description
discovery needs, as opposed to just a static list of char *.
Reviewed-by: Thiago Jung Bauermann <thiago.bauermann@linaro.org> Approved-By: Tom Tromey <tom@tromey.com>
Luis Machado [Tue, 7 Feb 2023 09:36:23 +0000 (09:36 +0000)]
sme: Enable SME registers and pseudo-registers
The SME (Scalable Matrix Extension) [1] exposes a new matrix register ZA with
variable sizes. It also exposes a new mode called streaming mode.
Similarly to SVE, the ZA register size is dictated by a vector length, but the
SME vector length is called streaming vetor length. The total size for
ZA in a given moment is svl x svl.
In streaming mode, the SVE registers have their sizes based on svl rather than
the regular vector length (vl).
The feature detection is controlled by the HWCAP2_SME bit, but actual support
should be validated by attempting a ptrace call for one of the new register
sets: NT_ARM_ZA and NT_ARM_SSVE.
Due to its large size, the ZA register is exposed as a vector of bytes, but we
introduce a number of pseudo-registers that gives various different views
into the ZA contents. These can be arranged in a couple categories: tiles and
tile slices.
Tiles are matrices the same size or smaller than ZA. Tile slices are vectors
which map to ZA's rows/columns in different ways.
A new dynamic target description is provided containing the ZA register, the SVG
register and the SVCR register. The size of ZA, like the SVE vector registers,
is based on the vector length register SVG (VG for SVE).
Luis Machado [Thu, 30 Mar 2023 14:16:53 +0000 (15:16 +0100)]
sve: Fix return command when using V registers in a SVE-enabled target
In a target without SVE support, the V registers have a size of 16 bytes,
otherwise they may have a size bigger than 16 bytes (depending on the current
vector length for the Z registers, as they overlap the V registers).
In aarch64-tdep.c:aarch64_store_return_value, the code is laid
out in a way that allocates the buffer with the size of the register, but
only updates the amount of bytes for the particular type we're returning.
This may cause a situation where we have a register size of 32 bytes but
are returning a floating point value of 8 bytes. The temporary buffer
will therefore have 32 bytes, but we'll only update 8 bytes of it.
When we write the entire register back, it will have potentially 24 bytes
of garbage in it.
Fix this by first reading the original contents of the register and then
overriding only the bytes that we need for the return value.
Luis Machado [Tue, 7 Feb 2023 10:08:23 +0000 (10:08 +0000)]
refactor: Simplify SVE interface to read/write registers
This is a patch in preparation to upcoming patches enabling SME support. It
attempts to simplify the gdb/gdbserver shared interface used to read/write
SVE registers.
Where the current code makes use of unique_ptr, allocating a new buffer by
hand and passing a buffer around, this patch makes that code use
gdb::byte_vector and passes a reference to this byte vector to the functions,
allowing the functions to have ready access to the size of the buffer.
It also shares a bit more code between gdb and gdbserver, in particular around
handling of ptrace get/set requests for SVE.
I think gdbserver could be refactored to handle register reads/writes more
like gdb's native layer as opposed to letting the generic linux-low layer do
the ptrace calls. This is not very flexible and assumes one size for the
responses. If you have something like NT_ARM_SVE, where you can have either
FPSIMD or SVE contents, it doesn't work that well.
I didn't want to change that interface right now as it is a bit too much work
and touches all the targets, some of which I can't easily test.
Hence the reason why the buffer the generic linux-now passes down to
linux-aarch64-low is unused or ignored.
No user-visible changes should happen as part of this refactor other than a
slightly reworded warning message.
While doing the refactor, I also noticed what seems to be a mistake in checking
if the register cache contains active (non-zero) SVE data.
For instance, the original code did something like this in
aarch64_sve_regs_copy_from_reg_buf:
has_sve_state |= reg_buf->raw_compare (AARCH64_SVE_Z0_REGNUM + i
reg, sizeof (__int128_t));
"reg" is a zeroed-out buffer that we compare the Z register contents
past the first 128 bits. The problem here is that raw_compare returns
1 if the contents compare the same, which means has_sve_state will be
true. But if we compared the Z register contents to 0, it means we
*do not* have SVE state, and therefore has_sve_state should be false.
The consequence of this mistake is that we convert the initial
FPSIMD-formatted data we get from ptrace for the NT_ARM_SVE register
set to a SVE-formatted one.
In the end, this doesn't cause user-visible differences because the
values of both the Z and V registers will still be the same. But the
logic is not correct.
I used the opportunity to fix this, and it gets tested later on by
the additional SME tests.
I do plan on submitting some SVE-specific tests to make sure we have
a bit more coverage in GDB's testsuite.
Regression-tested on aarch64-linux Ubuntu 22.04/20.04.
Luis Machado [Fri, 17 Mar 2023 21:23:59 +0000 (21:23 +0000)]
Fix register fetch/store order for native AArch64 Linux
I noticed we don't handle register reads/writes in the best way for native
AArch64 Linux. Some other registers are fetched/stored even if upper level
code told us to fetch a particular register number.
Fix this by being more strict about which registers we touch when
reading/writing them in the native AArch64 Linux layer.
There should be no user-visible changes due to this patch.
Regression-tested on aarch64-linux Ubuntu 22.04/20.04.
This patch moves instances of system register definitions, represented
by the SYSREG macro, out of their original place in `aarch64-opc.c'
and into a dedicated .def file, `aarch64-sys-regs.def'.
System register entries in this new file are ordered alphabetically by
name. This choice is made to enable the use of fast search algorithms
such as binary search when validating register names.
The SYSREG macro, defined as SYSREG (name, encoding, flags, features)
is kept as is and used in the def file, but all other SR_* macros
which previously served as indirections to SYSREG are removed.
This patch adds a mechanism for system register name alias detection
to register-matching mechanisms.
A new `F_REG_ALIAS' flag is added to the set of register flags and
used to label which entries in aarch64_sys_regs[] correspond to
aliases (and thus which CPENC values are non-unique in this array).
Where this is used is, for example, in `aarch64_print_operand' where,
in the case of system register decoding, the aarch64_sys_regs[] array
is iterated through until a match in CPENC value is made and the first
match accepted. If insufficient care is given in the ordering of
system registers in this array, the alias is encountered before the
"real" register and used incorrectly as the register name in the
disassembled output.
With this flag and the new `aarch64_sys_reg_alias_p' test, search
candidates corresponding to aliases can be conveniently skipped over.
One concrete example of where this is useful is with the
`trcextinselr0' system register. It was initially placed in the
system register list before `trcextinselr', in contrast to a more
natural alphabetical order.
Andrew Burgess [Mon, 25 Sep 2023 09:32:14 +0000 (10:32 +0100)]
gdb/corefile: write NT_GDB_TDESC based on signalled thread
When creating a core file from within GDB we include a NT_GDB_TDESC
that includes the target description of the architecture in use.
For architectures with dynamic architectures (e.g. AArch64 with
sve/sme) the original architecture, calculated from the original
target description, might not match the per-thread architecture.
In the general case, where each thread has a different architecture,
then we really need a separate NT_GDB_TDESC for each thread, however,
there's currently no way to read in multiple NT_GDB_TDESC.
This commit is a step towards per-thread NT_GDB_TDESC. In this commit
I have updated the function that writes the NT_GDB_TDESC to accept a
gdbarch (rather than calling target_gdbarch() to find a gdbarch), and
I now pass in the gdbarch of the signalled thread.
In many cases (though NOT all) targets with dynamic architectures
really only use a single architecture, even when there are multiple
threads, so in the common case, this should ensure that GDB emits an
architecture that is more likely to be correct.
Additional work will be needed in order to support corefiles with
truly per-thread architectures, but that will need to be done in the
future.
YunQiang Su [Tue, 3 Oct 2023 11:41:02 +0000 (12:41 +0100)]
MIPS: Fix `readelf -S bintest' test for n64 targets
Add a 64-bit traditional MIPS dump variant for the `readelf -S bintest'
test from binutils-all/readelf.exp, using a filename suffix according to
the rules set there, removing:
FAIL: readelf -S bintest
regressions with `mips64-linux-gnuabi64', `mips64el-linux-gnuabi64',
`mips64-openbsd', and `mips64el-openbsd' targets, which default to the
n64 ABI and consequently produce a section layout that is different from
what the generic dump pattern covers.
Co-Authored-By: Maciej W. Rozycki <macro@orcam.me.uk>
binutils/
* testsuite/binutils-all/readelf.s-64-tmips: New test variant.
Tom Tromey [Thu, 21 Sep 2023 14:49:44 +0000 (08:49 -0600)]
Add regression test for instructionReference change
Yesterday I pushed a patch to fix a small oversight in the DAP code
that caused an instructionReference to be an array instead of a
string.
This patch adds a test case for that regression. This required
exposing the TON form of the response -- something I mentioned might
be necessary when this code was changed to return just the Tcl form.
I tested this by backing out yesterday's bug fix and verifying that a
failure is seen.
Tom Tromey [Fri, 15 Sep 2023 14:59:09 +0000 (08:59 -0600)]
Clean up intermediate values in val_print_packed_array_elements
Following on Tom de Vries' work in the other array-printers, this
patch changes val_print_packed_array_elements to also avoid allocating
too many values when printing an Ada packed array.
Aaron Merey [Tue, 15 Aug 2023 16:17:48 +0000 (12:17 -0400)]
gdb/debuginfod: Add debuginfod_section_query
Add new function debuginfod_section_query. This function queries
debuginfod servers for an individual ELF/DWARF section associated with
a given build-id.
Tom de Vries [Mon, 2 Oct 2023 17:52:48 +0000 (19:52 +0200)]
[gdb/testsuite] Handle older gcc in gdb.ada/import.exp
When running test-case gdb.ada/import.exp with gcc 7, most test fail:
...
FAIL: gdb.ada/import.exp: print imported_var_ada
FAIL: gdb.ada/import.exp: print local_imported_var
FAIL: gdb.ada/import.exp: print pkg.imported_var_ada
FAIL: gdb.ada/import.exp: print pkg.exported_var_ada
FAIL: gdb.ada/import.exp: print exported_var_ada
FAIL: gdb.ada/import.exp: gdb_breakpoint: set breakpoint at pkg.imported_func_ada
FAIL: gdb.ada/import.exp: gdb_breakpoint: set breakpoint at imported_func_ada
FAIL: gdb.ada/import.exp: gdb_breakpoint: set breakpoint at local_imported_func
...
When running with gcc 8 or 9, only 2 tests fail:
...
FAIL: gdb.ada/import.exp: gdb_breakpoint: set breakpoint at pkg.imported_func_ada
FAIL: gdb.ada/import.exp: gdb_breakpoint: set breakpoint at imported_func_ada
...
The test-case passes fully with gcc 10, 11, 12 and 13.
Debug info for pragma import seems to not have been supported before gcc 8, so
require that version.
The two FAILs with gcc 8 and 9 seem to be due to problems in debug info. Add
an xfail for these.
Tom de Vries [Mon, 2 Oct 2023 17:48:21 +0000 (19:48 +0200)]
[gdb/testsuite] Add KFAIL for PR ada/30908
With gcc 13.2.1, I run into a cluster of fails:
...
FAIL: gdb.ada/str_binop_equal.exp: print my_str = "ABCD"
FAIL: gdb.ada/widewide.exp: print my_wws = " helo"
FAIL: gdb.ada/widewide.exp: print my_ws = "wide"
...
The problem is that the debug info contains information about function
ada.strings.maps."=", and gdb uses it to implement the comparison.
The function is supposed to compare two char sets, not strings, so gdb
shouldn't use it. This is PR ada/30908.
I don't see the same problem with gcc 7.5.0, because the exec doesn't contain
the debug info for the function, because the corresponding object is not
linked in. Adter adding "with Ada.Strings.Maps; use Ada.Strings.Maps;" to
gdb.ada/widewide/foo.adb I run into the same problem with gcc 7.5.0.
Add KFAILs for the PR.
Tested on x86_64-linux:
- openSUSE Leap 15.4 (using gcc 7.5.0), and
- openSUSE Tumbleweed (using gcc 13.2.1).
Approved-By: Tom Tromey <tom@tromey.com>
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30908
Andrew Burgess [Tue, 19 Sep 2023 10:45:36 +0000 (11:45 +0100)]
gdb: add Python events for program space addition and removal
Initially I just wanted a Python event for when GDB removes a program
space, I'm writing a Python extension that caches information for each
program space, and need to know when I should discard entries for a
particular program space.
But, it seemed easy enough to also add an event for when GDB adds a
new program space, so I went ahead and added both new events.
Of course, we don't currently have an observable for program space
addition or removal, so I first needed to add these. After that it's
pretty simple to add two new Python events and have these trigger.
The two new event registries are:
events.new_progspace
events.free_progspace
These emit NewProgspaceEvent and FreeProgspaceEvent objects
respectively, each of these new event types has a 'progspace'
attribute that contains the relevant gdb.Progspace object.
There's a couple of things to be mindful of.
First, it is not possible to catch the NewProgspaceEvent for the very
first program space, the one that is created when GDB first starts, as
this program space is created before any Python scripts are sourced.
In order to allow this event to be caught we would need to defer
creating the first program space, and as a consequence the first
inferior, until some later time. But, existing scripts could easily
depend on there being an initial inferior, so I really don't think we
should change that -- and so, we end up with the consequence that we
can't catch the event for the first program space.
The second, I think minor, issue, is that GDB doesn't clean up its
program spaces upon exit -- or at least, they are not cleaned up
before Python is shut down. As a result, any program spaces in use at
the time GDB exits don't generate a FreeProgspaceEvent. I'm not
particularly worried about this for my use case, I'm using the event
to ensure that a cache doesn't hold stale entries within a single GDB
session. It's also easy enough to add a Python at-exit callback which
can do any final cleanup if needed.
Finally, when testing, I did hit a slightly weird issue with some of
the remote boards (e.g. remote-stdio-gdbserver). As a consequence of
this issue I see some output like this in the gdb.log:
(gdb) PASS: gdb.python/py-progspace-events.exp: inferior 1
step
FreeProgspaceEvent: <gdb.Progspace object at 0x7fb7e1d19c10>
warning: cannot close "target:/lib64/libm.so.6": Cannot execute this command while the target is running.
Use the "interrupt" command to stop the target
and then try again.
warning: cannot close "target:/lib64/libc.so.6": Cannot execute this command while the target is running.
Use the "interrupt" command to stop the target
and then try again.
warning: cannot close "target:/lib64/ld-linux-x86-64.so.2": Cannot execute this command while the target is running.
Use the "interrupt" command to stop the target
and then try again.
do_parent_stuff () at py-progspace-events.c:41
41 ++global_var;
(gdb) PASS: gdb.python/py-progspace-events.exp: step
The 'FreeProgspaceEvent ...' line is expected, that's my test Python
extension logging the event. What isn't expected are all the blocks
like:
warning: cannot close "target:/lib64/libm.so.6": Cannot execute this command while the target is running.
Use the "interrupt" command to stop the target
and then try again.
It turns out that this has nothing to do with my changes, this is just
a consequence of reading files over the remote protocol. The test
forks a child process which GDB stays attached too. When the child
exits, GDB cleans up by calling prune_inferiors, which in turn can
result in GDB trying to close some files that are open because of the
inferior being deleted.
If the prune_inferiors call occurs when the remote target is
running (and in non-async mode) then GDB will try to send a fileio
packet while the remote target is waiting for a stop reply, and the
remote target will throw an error, see remote_target::putpkt_binary in
remote.c for details.
I'm going to look at fixing this, but, as I said, this is nothing to
do with this change, I just mention it because I ended up needing to
account for these warning messages in one of my tests, and it all
looks a bit weird.
Approved-By: Tom Tromey <tom@tromey.com> Reviewed-By: Eli Zaretskii <eliz@gnu.org>
projects / thirdparty / binutils-gdb.git / log
