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

[Andrew Burgess <aburgess@redhat.com>]

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.

> +{
> +  /* 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.

> +
> +  /* 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?

> +	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?

> +	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.

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