Re: [PATCH 3/3] Add test for fetching TLS from core file

["Maciej W. Rozycki" <macro at mips dot com>]

On Tue, 17 Oct 2017, Djordje Todorovic wrote:

> diff --git a/gdb/testsuite/gdb.threads/tls-core.c b/gdb/testsuite/gdb.threads/tls-core.c
> new file mode 100644
> index 0000000..e41c584
> --- /dev/null
> +++ b/gdb/testsuite/gdb.threads/tls-core.c
> @@ -0,0 +1,44 @@
> +/* This test is part of GDB, the GNU debugger.
> +
> +   Copyright 2017 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/>.  */
> +
> +#include <stdio.h>
> +#include <stdlib.h>
> +#include <unistd.h>
> +#include <pthread.h>
> +
> +#define NUMBER_OF_THREADS 5
> +
> +int __thread foo = 0xdeadbeef;
> +
> +void *
> +thread (void *in) /* BREAK */
> +{
> +  int *tmp = (int *) in;
> +  int value = *tmp;
> +  foo += *tmp;
> +  while (1)
> +    sleep (10);
> +}
> +
> +int
> +main (void)
> +{
> +  pthread_t threads[NUMBER_OF_THREADS];
> +  int i;
> +  for (i = 0; i < NUMBER_OF_THREADS; i++)
> +    pthread_create (&threads[i], NULL, thread, &i);

 Missing `return' statement here.

 Also `while (1) sleep (10);' perhaps, so that you don't cause a race with 
the whole process terminating (I think `pthread_join (&threads[0], NULL);' 
would do too; I'm sure there are other ways as well to prevent from 
running to exit(2)).  Overall please don't assume any particular thread 
scheduling characteristics of the underlying OS, and especially on SMP 
systems effects may be sometimes surprising.

> +}
> \ No newline at end of file

 Missing new line here.

> diff --git a/gdb/testsuite/gdb.threads/tls-core.exp b/gdb/testsuite/gdb.threads/tls-core.exp
> new file mode 100644
> index 0000000..dc291ab
> --- /dev/null
> +++ b/gdb/testsuite/gdb.threads/tls-core.exp
> @@ -0,0 +1,60 @@
> +# Copyright 2017 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/>.
> +
> +standard_testfile
> +
> +if {[gdb_compile_pthreads "${srcdir}/${subdir}/${srcfile}" "${binfile}" \
> +	 executable { debug }] != "" } {

 Formatting: space after a tab here.

> +    return -1
> +}
> +
> +
> +clean_restart ${binfile}
> +
> +#
> +# Set breakpoint at line 28.
> +#
> +gdb_breakpoint [gdb_get_line_number "BREAK"]
> +gdb_test "run" "Starting program: .*"

 The `run' command isn't supported with all targets and we do not wan't to 
cause a test failure unrelated to the matter being verified.  Please use 
`runto_main' then, i.e.:

clean_restart ${binfile}

runto_main

Sorry to miss that before.

 Also since you set a breakpoint at a function (which wasn't evident 
before) you should be able to simplify processing and just use:

runto thread

instead of going through the dance to set and catch the breakpoint 
manually.

> +
> +#
> +# Generate corefile.
> +#
> +set corefile [standard_output_file gcore.test]
> +set core_supported [gdb_gcore_cmd "$corefile" "save a corefile"]
> +if {!$core_supported} {
> +    return -1
> +}
> +
> +gdb_exit
> +
> +#
> +# Restart gdb and load generated corefile.
> +#
> +gdb_start
> +gdb_reinitialize_dir $srcdir/$subdir
> +gdb_load ${binfile}
> +
> +set core_loaded [gdb_core_cmd "$corefile" "load generated corefile"]
> +if { $core_loaded == -1 } {

 I think this has to be `$core_loaded != 1' or suchlike as I doubt a core 
file with no registers recorded will be usable for the actual test being 
made here.  I missed this peculiarity before, sorry.

> +    # No use proceeding from here.
> +    return

 I missed this on the previous iteration, sorry: why no return result here 
(i.e. an empty string being returned) vs -1 elsewhere?

 I don't think we have a clear rule for return results defined and 
moreover the caller actually ignores it.  Regardless, we set the result, 
differently, across the test suite.

 The value of -1 seems to be used as an indicator of a fatal test 
processing failure, such as with the first `return' statement above.  And 
in the case of a failure `gdb_compile_pthreads' will have already called 
`unsupported', so there's nothing else for us to do here.

 The second `return' statement is neither a fatal test processing failure 
nor a test failure.  It merely marks the `gcore' feature not being 
supported and consequently the inability to continue with the test case.  
In this case `gdb_gcore_cmd' will have already called `unsupported' as 
expected.  So I think it should return 0 or nothing (i.e. an empty 
string).

 The third `return' statement actually marks a test failure, because GDB 
is supposed to be able to load a core file in all cases, but not a fatal 
test processing failure.  Again in this case `gdb_core_cmd' will have 
already called `fail' as expected.  So again I think it should return 0 or 
nothing.

 I'd like to hear from a general maintainer about this though.

 Pedro, would you be able to comment on my observations and, ultimately 
(once the details have been sorted out), approve this change (as 
technically it's outside my area of competence anyway)?

> +}
> +
> +gdb_test "p/x foo" \
> +                    "\\$\[0-9]+ = 0xdeadbeef" \
> +                    "Printing thread-local storage variable."

 Formatting: use tabs rather than spaces to indent; one per line will do.

 Please wait a couple of days to give Pedro an opportunity to chime in and 
then resubmit the whole series with the details above and 1/3 updated.

  Maciej