Re: [PATCH v3 8/8] gdb/testsuite: Add test to exercise multi-threaded AArch64 SVE inferiors

[Luis Machado <luis.machado@arm.com>]

Just a nit.

On 1/30/23 04:45, Thiago Jung Bauermann via Gdb-patches wrote:
> This testcase exercises two scenarios with a multi-threaded inferior, one
> proposed by Luis and another one proposed by Simon.
> 
> In the first scenario, one of the inferior threads changes its vector
> length and then hits a breakpoint.  In the second one, the main thread hits
> a breakpoint and then GDB switches to another thread.
> ---
>   gdb/testsuite/gdb.arch/aarch64-sve-threads.c  | 125 ++++++++++++++++++
>   .../gdb.arch/aarch64-sve-threads.exp          |  70 ++++++++++
>   2 files changed, 195 insertions(+)
>   create mode 100644 gdb/testsuite/gdb.arch/aarch64-sve-threads.c
>   create mode 100644 gdb/testsuite/gdb.arch/aarch64-sve-threads.exp
> 
> diff --git a/gdb/testsuite/gdb.arch/aarch64-sve-threads.c b/gdb/testsuite/gdb.arch/aarch64-sve-threads.c
> new file mode 100644
> index 000000000000..7fad77008da3
> --- /dev/null
> +++ b/gdb/testsuite/gdb.arch/aarch64-sve-threads.c
> @@ -0,0 +1,125 @@
> +/* This testcase is part of GDB, the GNU debugger.
> +
> +   Copyright 2023 Free Software Foundation, Inc.
> +
> +   This program is free software; you can redistribute it and/or modify
> +   it under the terms of the GNU General Public License as published by
> +   the Free Software Foundation; either version 3 of the License, or
> +   (at your option) any later version.
> +
> +   This program is distributed in the hope that it will be useful,
> +   but WITHOUT ANY WARRANTY; without even the implied warranty of
> +   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
> +   GNU General Public License for more details.
> +
> +   You should have received a copy of the GNU General Public License
> +   along with this program.  If not, see <http://www.gnu.org/licenses/>.  */
> +
> +/* Exercise AArch64's Scalable Vector Extension in a multi-threaded program.  */
> +
> +#include <stdbool.h>
> +#include <stdio.h>
> +#include <pthread.h>
> +#include <semaphore.h>
> +#include <sys/prctl.h>
> +#include <unistd.h>
> +
> +/* For one of the tests, the thread needs to sleep after setting the vector
> +   length.  This variable is set by GDB.  */
> +volatile bool should_sleep = false;
> +
> +/* Used to signal to the main thread that the additional thread's vector length
> +   was changed.  */
> +sem_t vl_changed;
> +
> +/* Start routine for the additional thread.  Sets a new vector length, sleeps if
> +   requested then restores the original vector length.  */
> +
> +static void *
> +thread_function (void *unused)
> +{
> +  unsigned int vl;
> +  int rc;
> +
> +  rc = prctl (PR_SVE_GET_VL, 0, 0, 0, 0);
> +  if (rc < 0)
> +    {
> +      perror ("FAILED to PR_SVE_GET_VL");
> +      sem_post (&vl_changed);
> +      return NULL;
> +    }
> +
> +  vl = rc & PR_SVE_VL_LEN_MASK;
> +
> +  /* Decrease vector length by 16 bytes.  */
> +  vl -= 16;

Is there an initial vector length set before we decrement it by 16? I'm wondering about a case where we already start with 16, and then we'd decrement the vector length to an invalid value.

> +
> +  rc = prctl (PR_SVE_SET_VL, vl, 0, 0, 0, 0);
> +  if (rc < 0)
> +    {
> +      perror ("FAILED to PR_SVE_SET_VL");
> +      sem_post (&vl_changed);
> +      return NULL;
> +    }
> +
> +  /* Let the main thread continue.  */
> +  rc = sem_post (&vl_changed);
> +  if (rc != 0)
> +    {
> +      perror ("sem_post");
> +      return NULL;
> +    }
> +
> +  if (should_sleep)
> +    sleep (10);
> +
> +  /* Restore original vector length.  */
> +  vl += 16; /* break here 1 */
> +
> +  rc = prctl (PR_SVE_SET_VL, vl, 0, 0, 0, 0);
> +  if (rc < 0)
> +    {
> +      perror ("FAILED to PR_SVE_SET_VL");
> +      return NULL;
> +    }
> +
> +  return NULL; /* break here 2 */
> +}
> +
> +int
> +main (int argc, char **argv)
> +{
> +  pthread_t thread;
> +  int rc;
> +
> +  rc = sem_init (&vl_changed, 0, 0);
> +  if (rc != 0)
> +    {
> +      perror ("sem_init");
> +      return 1;
> +    }
> +
> +  rc = pthread_create (&thread, NULL, thread_function, NULL);
> +  if (rc != 0)
> +    {
> +      perror ("pthread_create");
> +      return 1;
> +    }
> +
> +  /* Wait until the additional thread changes it's vector length.  */
> +  rc = sem_wait (&vl_changed);
> +  if (rc != 0)
> +    {
> +      perror ("sem_wait");
> +      return 1;
> +    }
> +
> +  rc = pthread_join (thread, NULL); /* break here 3 */
> +  if (rc != 0)
> +    {
> +      perror ("pthread_join");
> +      return 1;
> +    }
> +
> +  return 0;
> +}
> diff --git a/gdb/testsuite/gdb.arch/aarch64-sve-threads.exp b/gdb/testsuite/gdb.arch/aarch64-sve-threads.exp
> new file mode 100644
> index 000000000000..48197650e1de
> --- /dev/null
> +++ b/gdb/testsuite/gdb.arch/aarch64-sve-threads.exp
> @@ -0,0 +1,70 @@
> +# Copyright 2023 Free Software Foundation, Inc.
> +
> +# This program is free software; you can redistribute it and/or modify
> +# it under the terms of the GNU General Public License as published by
> +# the Free Software Foundation; either version 3 of the License, or
> +# (at your option) any later version.
> +#
> +# This program is distributed in the hope that it will be useful,
> +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
> +# GNU General Public License for more details.
> +#
> +# You should have received a copy of the GNU General Public License
> +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
> +
> +# Test a multi-threaded binary that uses SVE and changes the SVE vector length
> +# in the additional thread.
> +
> +if {[skip_aarch64_sve_tests]} {
> +    verbose "Skipping ${gdb_test_file_name}."
> +    return
> +}
> +
> +standard_testfile
> +if {[prepare_for_testing "failed to prepare" ${testfile} ${srcfile} \
> +	 {debug pthreads}] == -1} {
> +    return -1
> +}
> +
> +if ![runto_main] {
> +    return -1
> +}
> +
> +# Stop after the additional thread has changed its vector length.
> +gdb_breakpoint [gdb_get_line_number "break here 1"]
> +gdb_continue_to_breakpoint "break here 1"
> +
> +# If GDB and gdbserver don't agree on the thread's vector length, this command
> +# will fail.
> +gdb_test "print \$z0" " = {q = {u = {.*}}}" "print z0 register"
> +

Out of curiosity, what is the failure mode when gdbserver doesn't do the right thing.

> +# Stop after the additional thread has restored its original vector length.
> +gdb_breakpoint [gdb_get_line_number "break here 2"]
> +gdb_continue_to_breakpoint "break here 2"
> +
> +# Test that going back to the original vector length doesn't confuse GDB or
> +# gdbserver.
> +gdb_test "print \$z0" " = {q = {u = {.*}}}" "print z0 register again"
> +
> +# Restart GDB to test a scenario where GDB switches to a thread that changed its
> +# vector length but hasn't hit any breakpoints yet.
> +clean_restart ${binfile}
> +
> +if ![runto_main] {
> +    return -1
> +}
> +
> +# Make the thread sleep after changing its vector length.
> +gdb_test_no_output -nopass "set var should_sleep = 1" "make thread sleep"
> +
> +# Stop after the additional thread has been created.
> +gdb_breakpoint [gdb_get_line_number "break here 3"]
> +gdb_continue_to_breakpoint "break here 3"
> +
> +# The regexp accounts for two lines of output after the "Switching to thread" message.
> +gdb_test_lines "thread 2" "switch to another thread" \
> +    {\[Switching to thread 2 \(.*\)\]\r\n#0  [[:print:]]+}
> +
> +# Make sure everything is still fine.
> +gdb_test "print \$z0" " = {q = {u = {.*}}}" "print z0 register in thread 2