Re: [PATCH 25/25] drm/tests: probe-helper: Add .force_ctx() callback coverage

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential
issue to consider:
- [Medium] drm/tests: unhandled -EDEADLK injection in
  drm_probe_helper_contender_fn()
--

commit d28b5fe04c94cb62864f25a529c3f378922f36a4
Author: Cristian Ciocaltea <cristian.ciocaltea@collabora.com>

drm/tests: probe-helper: Add .force_ctx() callback coverage

This commit adds KUnit tests for the .force_ctx() connector helper
callback, including deadlock backoff handling using a contender kthread.

> diff --git a/drivers/gpu/drm/tests/drm_probe_helper_test.c b/drivers/gpu/drm/tests/drm_probe_helper_test.c
> index ebc0392585e72..5d7dc47255181 100644
> --- a/drivers/gpu/drm/tests/drm_probe_helper_test.c
> +++ b/drivers/gpu/drm/tests/drm_probe_helper_test.c
[ ... ]
> @@ -212,6 +225,289 @@ static struct kunit_suite drm_test_connector_helper_tv_get_modes_suite = {
>  
>  kunit_test_suite(drm_test_connector_helper_tv_get_modes_suite);
>  
> +/*
> + * Runs an acquire context older than the probe one, which first takes @lock,
> + * then waits to be told to also request the connection_mutex held by the
> + * probe context.  Being older, it wounds the probe context, which will get
> + * -EDEADLK as soon as it tries to take @lock.
> + */
> +struct drm_probe_helper_contender {
> +	struct drm_device *drm;
> +	struct drm_modeset_lock lock;
> +	struct completion locked;
> +	struct completion go;
> +	struct completion done;
> +	int ret;
> +};
> +
> +static int drm_probe_helper_contender_fn(void *data)
> +{
> +	struct drm_probe_helper_contender *contender = data;
> +	struct drm_modeset_acquire_ctx ctx;
> +
> +	drm_modeset_acquire_init(&ctx, 0);
> +
> +	contender->ret = drm_modeset_lock(&contender->lock, &ctx);
> +	complete(&contender->locked);
> +
> +	if (!contender->ret) {
> +		if (wait_for_completion_timeout(&contender->go, 5 * HZ))
> +			contender->ret =
> +				drm_modeset_lock(&contender->drm->mode_config.connection_mutex,
> +						 &ctx);
> +		else
> +			contender->ret = -ETIMEDOUT;
> +	}
> +
> +	drm_modeset_drop_locks(&ctx);

[Severity: Medium]
Can drm_modeset_lock(&contender->drm->mode_config.connection_mutex, &ctx)
return -EDEADLK when CONFIG_DEBUG_WW_MUTEX_SLOWPATH is enabled?

Acquiring connection_mutex is the second lock in ctx (ctx.ww_ctx.acquired
becomes 2), so ww_mutex_lock() runs ww_mutex_deadlock_injection().  When
(ctx.ww_ctx.stamp & 0xf) == 0, it injects -EDEADLK after the probe
context backs off and releases connection_mutex, setting ctx.contended.

Because drm_probe_helper_contender_fn() does not handle -EDEADLK,
drm_modeset_drop_locks(&ctx) hits WARN_ON(ctx->contended) and
drm_test_probe_helper_force_ctx_deadlock() fails KUNIT_ASSERT_EQ(test,
contender->ret, 0).

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261114-drm-conn-hdmi-cleanup-v1-0-20e8a33a2286@collabora.com?part=25