Re: [PATCH] gdb/testsuite: add test for memory requirements of gcore

[Guinevere Larsen <guinevere@redhat.com>]

On 2/24/25 3:21 PM, Andrew Burgess wrote:
> Guinevere Larsen <guinevere@redhat.com> writes:
>
>> For a long time, Fedora has been carrying an out-of-tree patch with a
>> similar test to the one proposed in this patch, that ensures that the
>> memory requirements don't grow with the inferior's memory. It's been
>> so long that the context for why this test exists has been lost, but
>> it looked like it could be interesting for upstream.
>>
>> The test runs twice, once with the inferior allocating 4Mb of memory,
>> and the other allocating 64Mb. My plan was to find the rate at which
>> things increase based on inferior size, and have that tested to ensure
>> we're not growing that requirement accidentally, but my testing
>> actually showed memory requirements going down as the inferior increases,
>> so instead I just hardcoded that we need less than 2Mb for the command,
>> and it can be tweaked later if necessary.
>> ---
>>   gdb/testsuite/gdb.base/gcore-memory-usage.c   | 37 +++++++
>>   gdb/testsuite/gdb.base/gcore-memory-usage.exp | 97 +++++++++++++++++++
>>   2 files changed, 134 insertions(+)
>>   create mode 100644 gdb/testsuite/gdb.base/gcore-memory-usage.c
>>   create mode 100644 gdb/testsuite/gdb.base/gcore-memory-usage.exp
>>
>> diff --git a/gdb/testsuite/gdb.base/gcore-memory-usage.c b/gdb/testsuite/gdb.base/gcore-memory-usage.c
>> new file mode 100644
>> index 00000000000..514fdb905cf
>> --- /dev/null
>> +++ b/gdb/testsuite/gdb.base/gcore-memory-usage.c
>> @@ -0,0 +1,37 @@
>> +/* This testcase is part of GDB, the GNU debugger.
>> +
>> +   Copyright 2025 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>
>> +
>> +int main(int argc, char **argv)
> This should use GNU style.  So 'main' on a new line.
oops, sorry, habits...
>
>> +{
>> +  /* Use argv to calculate how many megabytes to allocate.
>> +     Do this to see how much gcore memory usage increases
>> +     based on inferior dynamic memory.  */
>> +  int megs = atoi (argv[1]);
>> +  char *p = malloc (megs * 1024 * 1024);
>> +
>> +  /* Wait a long time so GDB can attach to this.  */
>> +  sleep (10);
> Ideally we'd not use a hard coded sleep like this.  Instead, you could
> have the test set an alarm with some large delay (common for tests that
> can spin), then have the test enter a spin loop:
>
>    while (some_flag_that_is_true)
>      sleep (1);
>
> Then GDB can terminate the test by setting some_flag_that_is_true to
> false if it wants, or you can just stick with the existing approach of
> sending SIGKILL, which is fine too.
>
> Adding an alarm call ensures the test will self terminate eventually, if
> DejaGNU doesn't clean it up correctly for some reason.
Oh, I see. This makes sense, I'll go with it then
>
>> +
>> +  /* Let's be nice citizens :).  */
>> +  free (p);
>> +  return 0;
>> +}
>> diff --git a/gdb/testsuite/gdb.base/gcore-memory-usage.exp b/gdb/testsuite/gdb.base/gcore-memory-usage.exp
>> new file mode 100644
>> index 00000000000..cdb6d73c11b
>> --- /dev/null
>> +++ b/gdb/testsuite/gdb.base/gcore-memory-usage.exp
>> @@ -0,0 +1,97 @@
>> +# Copyright (C) 2025 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 that gcore doesn't end up using excessive memory.
>> +
>> +require {istarget "*-*-linux*"}
>> +
>> +standard_testfile
>> +
>> +if {[build_executable "failed to prepare" $testfile $srcfile {debug}] == -1} {
>> +    return -1
>> +}
>> +
>> +# Read the proc_pid_status page, to find how much memory the given
>> +# PID is using.  This is meant to be used to find the
>> +# memory usage for the GDB in this test.
>> +# Returns memory usage in Kb, or -1 if it couldn't be found.
>> +proc get_mem_usage {pid prefix} {
>> +    set fd [open "/proc/$pid/status"]
>> +    set memory -1
>> +    while {[gets $fd line] != -1} {
>> +	if {[regexp {VmSize:\s*([0-9]+) kB} $line full mem]} {
>> +	    set memory $mem
>> +	    break
>> +	}
>> +    }
>> +    close $fd
>> +
>> +    gdb_assert {$memory != -1} "$prefix: Managed to read the memory usage"
>> +
>> +    return $memory
>> +}
>> +
>> +# This proc restarts GDB, runs the inferior with the desired
>> +# amount of memory, then checks how much memory is necessary
>> +# to run the gcore command.  It will return -1 if the gcore
>> +# command fails, 0 otherwise.
>> +proc run_test {megs} {
>> +    with_test_prefix "$megs Mb" {
>> +	clean_restart $::testfile
>> +
>> +	set bin [standard_output_file $::testfile]
> Isn't bin the same as binfile, which is setup by standard_testfile?  Or
> am I missing something?
I think I forgot about the existence of binfile. This was copied from 
the upstream test...
>
>> +	set corefile [standard_output_file "${::testfile}.core"]
>> +	set inferior_pid [eval exec $bin $megs &]
> If there was no argument passing here then I'd say you should be using
> spawn_wait_for_attach.  But that doesn't support argument passing...
>
> ... however, if you read that proc (and its helper proc) you'll see some
> comments that suggest using eval/exec like you do are not the right
> choice.
>
> So maybe we should either extend (somehow) spawn_wait_for_attach to
> allow argument passing, or write something like spawn_wait_for_attach
> that handles arguments?

If I found the correct documentation page, I think it shouldn't be hard 
to extend spawn_wait_for_attach to handle arguments. And using optional 
arguments, it should be a  very minor patch, so I'll see what I can cook up.

Thanks for the pointer, this definitely sounds like a better solution.

>
>> +	set gdb_pid [exp_pid -i [board_info host fileid]]
>> +
>> +	# wait for memory allocation to finish.
> Capitalise comments please.
>
>> +	sleep 1
> Ideally we'd not rely on sleeps like this.  Better would be to attach,
> and then ensure the inferior has reached some known point before sending
> the gcore command.
Right, I see. This makes sense if the inferior can spin, so I'll switch 
to this.

-- 
Cheers,
Guinevere Larsen
She/Her/Hers

>
> Thanks,
> Andrew
>
>> +
>> +	# Get the important info.
>> +	gdb_test "attach $inferior_pid" "Attaching to.*"
>> +	set mem_before [get_mem_usage $gdb_pid before]
>> +	if {![gdb_gcore_cmd $corefile "create the corefile"]} {
>> +	    return -1
>> +	}
>> +	set mem_after [get_mem_usage $gdb_pid after]
>> +
>> +	# Do the main part of the test: How much is the memory
>> +	# usage of GDB going to grow after using the gcore command.
>> +	set diff_k [expr $mem_after - $mem_before]
>> +	set diff [expr $diff_k/1024]
>> +	verbose -log "The gcore command used $diff Mb ($diff_k Kb)"
>> +	# The original plan was to compare to a multiple of MEGS
>> +	# but since the requirements don't seem to go up as the
>> +	# inferior allocated more memory, we instead just hardcode
>> +	# 2 megs, since sometimes 1 is used.
>> +	gdb_assert {$diff < 2} \
>> +	    "gdb did not use as much memory as the inferior"
>> +
>> +	# Kill the inferior so we don't have to wait until the process
>> +	# finishes.
>> +	gdb_test "signal SIGKILL" ".*The program no longer exists."
>> +    }
>> +    return 0
>> +}
>> +
>> +# If we couldn't create the first corefile, there's no point
>> +# in running the second part of the test.
>> +if {[run_test 4] != 0} {
>> +    return
>> +}
>> +# Surprisingly enough, the larger inferior doesn't seem to use
>> +# any extra memory, it usually uses less memory.  Which is good,
>> +# it means our memory requirements aren't growing with the inferior.
>> +run_test 64
>> -- 
>> 2.48.1