From nobody Sun Feb 8 11:26:06 2026 Received: from smtp.kernel.org (aws-us-west-2-korg-mail-1.web.codeaurora.org [10.30.226.201]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id 3722F15C158 for ; Mon, 7 Apr 2025 15:22:58 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; arc=none smtp.client-ip=10.30.226.201 ARC-Seal: i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1744039378; cv=none; b=GLziJNGDsEBhZBdWddtWjyp9J9JC2MEdGjXRVsDBk3368G27vbPlQRe5r34x48UyjVj/Zz4aDB1GEmiz1Yh3CmfMGC4M2WLwkVN+SAjl1j73uzrFIJ3P9GI/tM1mFvLJhvNBs8rTBI4KXpqpdBVJPVm3MAmyd0uvZb1LkZt1qnw= ARC-Message-Signature: i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1744039378; c=relaxed/simple; bh=3ddWVE+d0ZcwzSs4+cFUh4H/OJZsQa74J4Okb65lv9M=; h=From:To:Cc:Subject:Date:Message-ID:In-Reply-To:References: MIME-Version; b=sbCKb+NPON4cwSuhIhTN+FMgEszKR/u9gXOGPAEjq1dKhLMewCn+xwib0FKnKSCO4NImQApje1kr6I3+Uhqd1Qx7ULUnWbLBnXWhc44yx2aNb9sElCFuwnYuqk4lnq2FwwhDx8E/pV86Fk1o0SRnpkUQWXdPhN07AJex6aDZkrw= ARC-Authentication-Results: i=1; smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=kernel.org header.i=@kernel.org header.b=ijeqhBxb; arc=none smtp.client-ip=10.30.226.201 Authentication-Results: smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=kernel.org header.i=@kernel.org header.b="ijeqhBxb" Received: by smtp.kernel.org (Postfix) with ESMTPSA id DD5CEC4CEEC; Mon, 7 Apr 2025 15:22:54 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=kernel.org; s=k20201202; t=1744039378; bh=3ddWVE+d0ZcwzSs4+cFUh4H/OJZsQa74J4Okb65lv9M=; h=From:To:Cc:Subject:Date:In-Reply-To:References:From; b=ijeqhBxbtnRbqBykrf5cRYc1YRhk6oCP9jbac0KK+uuFQRrLdS4VN6YkATr7tZVTc V2T3e8KsJeSS/E8xnS596328lnGUjQyQOn0Nn7H/EVvFR8aoSNmsUIh9lvrSbSapGF DKY8PTa7+Dwvf/6GptS4YTdxy05LuzAAqOW+IdEP1kKTz1Qyk8EOn2XgVoux4g2YcB k/xOFgfiOCbX8+9GzmYJi6d6fLypTexgIov/tA/MGwUMKmQ+QjrSWT5t13r1cGPSWq AGuJGv6sGtP3EipRxSMpKt6YgaBw/GuDkATC0rQKDS2fuja3P4zEm/O/hLK4801a71 yyjztPY3xgaOw== From: Philipp Stanner To: Lyude Paul , Danilo Krummrich , David Airlie , Simona Vetter , Matthew Brost , Philipp Stanner , =?UTF-8?q?Christian=20K=C3=B6nig?= , Maarten Lankhorst , Maxime Ripard , Thomas Zimmermann , Tvrtko Ursulin Cc: dri-devel@lists.freedesktop.org, nouveau@lists.freedesktop.org, linux-kernel@vger.kernel.org, Danilo Krummrich Subject: [PATCH 2/5] drm/sched: Prevent teardown waitque from blocking too long Date: Mon, 7 Apr 2025 17:22:37 +0200 Message-ID: <20250407152239.34429-4-phasta@kernel.org> X-Mailer: git-send-email 2.48.1 In-Reply-To: <20250407152239.34429-2-phasta@kernel.org> References: <20250407152239.34429-2-phasta@kernel.org> Precedence: bulk X-Mailing-List: linux-kernel@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 Content-Transfer-Encoding: quoted-printable Content-Type: text/plain; charset="utf-8" The waitqueue that ensures that drm_sched_fini() blocks until the pending_list has become empty could theoretically cause that function to block for a very long time. That, ultimately, could block userspace procesess and prevent them from being killable through SIGKILL. When a driver calls drm_sched_fini(), it is safe to assume that all still pending jobs are not needed anymore anyways. Thus, they can be cancelled and thereby it can be ensured that drm_sched_fini() will return relatively quickly. Implement a new helper to stop all work items / submission except for the drm_sched_backend_ops.run_job(). Implement a driver callback, kill_fence_context(), that instructs the driver to kill the fence context associated with this scheduler, thereby causing all pending hardware fences to be signalled. Call those new routines in drm_sched_fini() and ensure backwards compatibility if the new callback is not implemented. Suggested-by: Danilo Krummrich Signed-off-by: Philipp Stanner --- drivers/gpu/drm/scheduler/sched_main.c | 47 +++++++++++++++++--------- include/drm/gpu_scheduler.h | 11 ++++++ 2 files changed, 42 insertions(+), 16 deletions(-) diff --git a/drivers/gpu/drm/scheduler/sched_main.c b/drivers/gpu/drm/sched= uler/sched_main.c index a6a4deeda72b..6b72278c4b72 100644 --- a/drivers/gpu/drm/scheduler/sched_main.c +++ b/drivers/gpu/drm/scheduler/sched_main.c @@ -1398,31 +1398,46 @@ drm_sched_no_jobs_pending(struct drm_gpu_scheduler = *sched) return empty; } =20 +/** + * drm_sched_cancel_jobs_and_wait - trigger freeing of all pending jobs + * @sched: scheduler instance + * + * Must only be called if &struct drm_sched_backend_ops.kill_fence_context= is + * implemented. + * + * Instructs the driver to kill the fence context associated with this sch= eduler, + * thereby signalling all pending fences. This, in turn, will trigger + * &struct drm_sched_backend_ops.free_job to be called for all pending job= s. + * The function then blocks until all pending jobs have been freed. + */ +static inline void +drm_sched_cancel_jobs_and_wait(struct drm_gpu_scheduler *sched) +{ + sched->ops->kill_fence_context(sched); + wait_event(sched->pending_list_waitque, drm_sched_no_jobs_pending(sched)); +} + /** * drm_sched_fini - Destroy a gpu scheduler * * @sched: scheduler instance * - * Tears down and cleans up the scheduler. - * - * Note that this function blocks until all the fences returned by - * &struct drm_sched_backend_ops.run_job have been signalled. + * Tears down and cleans up the scheduler. Might leak memory if + * &struct drm_sched_backend_ops.kill_fence_context is not implemented. */ void drm_sched_fini(struct drm_gpu_scheduler *sched) { struct drm_sched_entity *s_entity; int i; =20 - /* - * Jobs that have neither been scheduled or which have timed out are - * gone by now, but jobs that have been submitted through - * backend_ops.run_job() and have not yet terminated are still pending. - * - * Wait for the pending_list to become empty to avoid leaking those jobs. - */ - drm_sched_submission_and_timeout_stop(sched); - wait_event(sched->pending_list_waitque, drm_sched_no_jobs_pending(sched)); - drm_sched_free_stop(sched); + if (sched->ops->kill_fence_context) { + drm_sched_submission_and_timeout_stop(sched); + drm_sched_cancel_jobs_and_wait(sched); + drm_sched_free_stop(sched); + } else { + /* We're in "legacy free-mode" and ignore potential mem leaks */ + drm_sched_wqueue_stop(sched); + } =20 for (i =3D DRM_SCHED_PRIORITY_KERNEL; i < sched->num_rqs; i++) { struct drm_sched_rq *rq =3D sched->sched_rq[i]; @@ -1510,7 +1525,7 @@ bool drm_sched_wqueue_ready(struct drm_gpu_scheduler = *sched) EXPORT_SYMBOL(drm_sched_wqueue_ready); =20 /** - * drm_sched_wqueue_stop - stop scheduler submission + * drm_sched_wqueue_stop - stop scheduler submission and freeing * @sched: scheduler instance * * Stops the scheduler from pulling new jobs from entities. It also stops @@ -1526,7 +1541,7 @@ void drm_sched_wqueue_stop(struct drm_gpu_scheduler *= sched) EXPORT_SYMBOL(drm_sched_wqueue_stop); =20 /** - * drm_sched_wqueue_start - start scheduler submission + * drm_sched_wqueue_start - start scheduler submission and freeing * @sched: scheduler instance * * Restarts the scheduler after drm_sched_wqueue_stop() has stopped it. diff --git a/include/drm/gpu_scheduler.h b/include/drm/gpu_scheduler.h index badcf9ea4501..e13eb7e4c93a 100644 --- a/include/drm/gpu_scheduler.h +++ b/include/drm/gpu_scheduler.h @@ -509,6 +509,17 @@ struct drm_sched_backend_ops { * and it's time to clean it up. */ void (*free_job)(struct drm_sched_job *sched_job); + + /** + * @kill_fence_context: kill the fence context belonging to this scheduler + * + * Needed to cleanly tear the scheduler down in drm_sched_fini(). This + * callback will cause all hardware fences to be signalled by the driver, + * which, ultimately, ensures that all jobs get freed before teardown. + * + * This callback is optional, but it is highly recommended to implement i= t. + */ + void (*kill_fence_context)(struct drm_gpu_scheduler *sched); }; =20 /** --=20 2.48.1