From: sashiko-bot@kernel.org
To: "Albert Esteve" <aesteve@redhat.com>
Cc: dri-devel@lists.freedesktop.org, tj@kernel.org,
hannes@cmpxchg.org, mkoutny@suse.com
Subject: Re: [PATCH v7 1/3] selftests: cgroup: Add dmem_selftest helper module
Date: Tue, 17 Nov 2026 13:46:01 +0000 [thread overview]
Message-ID: <sashiko-outbox-218900@kernel.org> (raw)
In-Reply-To: <20261117-kunit_cgroups-v7-1-5bcbfe17f323@redhat.com>
Thank you for your contribution! Sashiko AI review found 3 potential
issues to consider:
- [Medium] selftests: cgroup: default KDIR in test_modules/Makefile
breaks build
- [Medium] selftests: cgroup: race on selftest_region in
dmem_selftest_exit()
- [Medium] selftests: cgroup: ERR_PTR stored in selftest_region in
dmem_selftest_init()
--
commit ff1dfd0d8169df4ccb406a2e9530169608e94814
Author: Albert Esteve <aesteve@redhat.com>
selftests: cgroup: Add dmem_selftest helper module
This commit adds an out-of-tree dmem_selftest kernel module under
tools/testing/selftests/cgroup/test_modules/ that registers a synthetic
dmem cgroup region and exposes alloc and free module parameters for
selftests.
> diff --git a/tools/testing/selftests/cgroup/test_modules/Makefile b/tools/testing/selftests/cgroup/test_modules/Makefile
> new file mode 100644
> index 0000000000000..dd8a17313cec2
> --- /dev/null
> +++ b/tools/testing/selftests/cgroup/test_modules/Makefile
> @@ -0,0 +1,17 @@
> +# SPDX-License-Identifier: GPL-2.0
> +TESTMODS_DIR := $(realpath $(dir $(abspath $(lastword $(MAKEFILE_LIST)))))
> +KDIR ?= /lib/modules/$(shell uname -r)/build
> +
> +obj-m += dmem_selftest.o
> +
> +# Ensure that KDIR exists, otherwise skip the compilation
> +modules:
> +ifneq ("$(wildcard $(KDIR))", "")
> + $(Q)$(MAKE) -C $(KDIR) modules KBUILD_EXTMOD=$(TESTMODS_DIR)
> +endif
[Severity: Medium]
Does defaulting KDIR to /lib/modules/$(shell uname -r)/build break the
cgroup selftest build when the host's running kernel headers are older
than the target kernel tree or when cross-compiling?
Because tools/testing/selftests/cgroup/Makefile sets TEST_GEN_MODS_DIR,
the default all target invokes this Makefile unconditionally. When
/lib/modules/$(uname -r)/build exists for an older host kernel that lacks
linux/cgroup_dmem.h or struct dmem_cgroup_init, Kbuild compiles
dmem_selftest.c against the host kernel headers instead of the tree under
test and fails the build. In addition, checking $(wildcard $(KDIR))
without checking $(wildcard $(KDIR)/Module.symvers) does not verify that
the kernel tree has been built.
Defaulting KDIR to $(if $(O),$(O),$(realpath ../../../../..)) and
checking $(wildcard $(KDIR)/Module.symvers), as in
tools/testing/selftests/sched_ext/test_modules/Makefile, would build
against the target kernel tree when available.
[ ... ]
> diff --git a/tools/testing/selftests/cgroup/test_modules/dmem_selftest.c b/tools/testing/selftests/cgroup/test_modules/dmem_selftest.c
> new file mode 100644
> index 0000000000000..a763a156f13f3
> --- /dev/null
> +++ b/tools/testing/selftests/cgroup/test_modules/dmem_selftest.c
> @@ -0,0 +1,146 @@
[ ... ]
> +static struct dmem_cgroup_region *selftest_region;
> +static struct dmem_cgroup_pool_state *alloc_pool;
> +static u64 alloc_size;
> +static DEFINE_MUTEX(alloc_lock);
> +
> +static int param_set_alloc(const char *val, const struct kernel_param *kp)
> +{
> + struct dmem_cgroup_pool_state *pool = NULL, *limit = NULL;
> + u64 size;
> + int ret;
> +
> + if (!selftest_region)
> + return -ENODEV;
> +
> + ret = kstrtou64(val, 0, &size);
> + if (ret)
> + return ret;
> + if (!size || size > DM_SELFTEST_REGION_SIZE)
> + return -EINVAL;
> +
> + mutex_lock(&alloc_lock);
> + if (alloc_pool) {
> + mutex_unlock(&alloc_lock);
> + return -EBUSY;
> + }
> +
> + ret = dmem_cgroup_try_charge(selftest_region, size, &pool, &limit);
> + if (ret == -EAGAIN && limit)
> + dmem_cgroup_pool_state_put(limit);
> + if (ret) {
> + mutex_unlock(&alloc_lock);
> + return ret;
> + }
> +
> + alloc_pool = pool;
> + alloc_size = size;
> + mutex_unlock(&alloc_lock);
> + return 0;
> +}
[ ... ]
> +static int __init dmem_selftest_init(void)
> +{
> + static const struct dmem_cgroup_init init = {
> + .size = DM_SELFTEST_REGION_SIZE,
> + };
> +
> + selftest_region = dmem_cgroup_register_region(&init, DM_SELFTEST_REGION_NAME);
> + if (IS_ERR(selftest_region))
> + return PTR_ERR(selftest_region);
[Severity: Medium]
If dmem_cgroup_register_region() returns an error pointer, can a
concurrent write to the alloc parameter dereference that ERR_PTR in
dmem_cgroup_try_charge()?
Module parameter sysfs attributes are created by mod_sysfs_setup() before
dmem_selftest_init() runs and remain accessible until do_init_module()
tears them down on error. Storing the return value directly into
selftest_region leaves a non-NULL ERR_PTR(-ENOMEM) visible to
param_set_alloc(), which only checks !selftest_region before passing it to
dmem_cgroup_try_charge() -> get_cg_pool_unlocked(). Checking the return
value in a local variable before assigning selftest_region avoids exposing
the error pointer.
[ ... ]
> +static void __exit dmem_selftest_exit(void)
> +{
> + mutex_lock(&alloc_lock);
> + if (alloc_pool) {
> + dmem_cgroup_uncharge(alloc_pool, alloc_size);
> + alloc_pool = NULL;
> + alloc_size = 0;
> + }
> + mutex_unlock(&alloc_lock);
> +
> + if (selftest_region) {
> + dmem_cgroup_unregister_region(selftest_region);
> + selftest_region = NULL;
> + }
> + pr_info("unloaded.\n");
> +}
[Severity: Medium]
Can param_set_alloc() race with dmem_selftest_exit() over selftest_region
and alloc_pool?
In delete_module(), dmem_selftest_exit() runs before free_module() removes
the module parameter sysfs files, and param_attr_store() does not hold a
module reference:
CPU0 CPU1
param_set_alloc() dmem_selftest_exit()
if (!selftest_region)
mutex_lock(&alloc_lock)
...
mutex_unlock(&alloc_lock)
dmem_cgroup_unregister_region(...)
selftest_region = NULL
mutex_lock(&alloc_lock)
dmem_cgroup_try_charge(selftest_region, ...)
get_cg_pool_unlocked(NULL, ...) <-- NULL pointer dereference
If CPU0 instead acquires alloc_lock right after CPU1 unlocks it and before
dmem_cgroup_unregister_region() marks the region unregistered,
dmem_cgroup_try_charge() succeeds and sets alloc_pool after
dmem_selftest_exit() has already finished uncharging alloc_pool, leaking
the charge, pool state, and cgroup CSS reference.
Checking selftest_region under alloc_lock in param_set_alloc() and
detaching selftest_region under the same lock in dmem_selftest_exit()
before unregistering the region would close both races.
--
Sashiko AI review · https://sashiko.dev/#/patchset/20261117-kunit_cgroups-v7-0-5bcbfe17f323@redhat.com?part=1
next prev parent reply other threads:[~2026-11-17 13:46 UTC|newest]
Thread overview: 6+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-11-17 13:27 [PATCH v7 0/3] cgroup: dmem: add selftest helper and coverage Albert Esteve
2026-11-17 13:27 ` [PATCH v7 1/3] selftests: cgroup: Add dmem_selftest helper module Albert Esteve
2026-11-17 13:46 ` sashiko-bot [this message]
2026-11-17 13:27 ` [PATCH v7 2/3] selftests: cgroup: Add KB() helper Albert Esteve
2026-11-17 13:27 ` [PATCH v7 3/3] selftests: cgroup: Add dmem selftest coverage Albert Esteve
2026-11-17 13:33 ` sashiko-bot
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=sashiko-outbox-218900@kernel.org \
--to=sashiko-bot@kernel.org \
--cc=aesteve@redhat.com \
--cc=dri-devel@lists.freedesktop.org \
--cc=hannes@cmpxchg.org \
--cc=mkoutny@suse.com \
--cc=sashiko-reviews@lists.linux.dev \
--cc=tj@kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.