All of lore.kernel.org
 help / color / mirror / Atom feed
From: Zhu Lingshan <lingshan.zhu@amd.com>
To: <Alexander.Deucher@amd.com>, <Christian.Koenig@amd.com>,
	<felix.kuehling@amd.com>
Cc: <Ray.Huang@amd.com>, <amd-gfx@lists.freedesktop.org>,
	Zhu Lingshan <lingshan.zhu@amd.com>
Subject: [PATCH 06/10] drm/amdgpu: hold userq kref in MES reset
Date: Fri, 28 Aug 2026 17:53:45 +0800	[thread overview]
Message-ID: <20260828095349.9797-7-lingshan.zhu@amd.com> (raw)
In-Reply-To: <20260828095349.9797-1-lingshan.zhu@amd.com>

Current mes_userq_reset_queue() walks the
user queue doorbell xarray without holding its
spin lock and does not take a queue->kref when
if finds a queue. So the queue could be
unexpectedly deconstructed and casuing an
use-after-free problem.

This commit fixes this problem by using
the helper amdgpu_lookup_queue_by_doorbell()
to properly acquire the spin lock of the xarray,
and hold its kref during processing the queue.

Signed-off-by: Zhu Lingshan <lingshan.zhu@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/mes_userqueue.c | 46 +++++++++++-----------
 1 file changed, 23 insertions(+), 23 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/mes_userqueue.c b/drivers/gpu/drm/amd/amdgpu/mes_userqueue.c
index 7f334f718cd8..3de71615cde0 100644
--- a/drivers/gpu/drm/amd/amdgpu/mes_userqueue.c
+++ b/drivers/gpu/drm/amd/amdgpu/mes_userqueue.c
@@ -244,30 +244,30 @@ int mes_userq_reset_queue(struct amdgpu_device *adev,
 {
 	struct amdgpu_usermode_queue *uq;
 	bool use_mmio = adev->gfx.mec.use_mmio_for_reset;
-	unsigned long uq_id;
-	int r;
+	int r = 0;
 
-	xa_for_each(&adev->userq_doorbell_xa, uq_id, uq) {
-		if (uq->queue_type == queue_type) {
-			if (uq == guilty_uq)
-				continue;
-			if (uq->doorbell_index == db) {
-				uq->state = AMDGPU_USERQ_STATE_HUNG;
-				if (use_mmio)
-					r = amdgpu_mes_reset_queue_mmio(adev, queue_type, 0, 1, pipe, queue, 0);
-				else
-					r = amdgpu_mes_reset_user_queue(adev, queue_type, db, 0);
-				if (r)
-					return r;
-				r = mes_userq_unmap(uq);
-				if (r)
-					return r;
-				amdgpu_userq_fence_driver_force_completion(uq);
-				break;
-			}
-		}
-	}
-	return 0;
+	uq = amdgpu_lookup_queue_by_doorbell(&adev->userq_doorbell_xa, db);
+	if (!uq)
+		return 0;
+
+	if (uq == guilty_uq || uq->queue_type != queue_type)
+		goto put_queue;
+
+	uq->state = AMDGPU_USERQ_STATE_HUNG;
+	if (use_mmio)
+		r = amdgpu_mes_reset_queue_mmio(adev, queue_type, 0, 1, pipe, queue, 0);
+	else
+		r = amdgpu_mes_reset_user_queue(adev, queue_type, db, 0);
+	if (r)
+		goto put_queue;
+
+	r = mes_userq_unmap(uq);
+	if (!r)
+		amdgpu_userq_fence_driver_force_completion(uq);
+
+put_queue:
+	amdgpu_userq_put(uq);
+	return r;
 }
 
 static int mes_userq_create_ctx_space(struct amdgpu_userq_mgr *uq_mgr,
-- 
2.53.0


  parent reply	other threads:[~2026-08-28  9:53 UTC|newest]

Thread overview: 18+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-08-28  9:53 [PATCH 00/10] drm/amdgpu: secure userq lifecycle by its kref Zhu Lingshan
2026-08-28  9:53 ` [PATCH 01/10] drm/amdgpu: introduce amdgpu_lookup_queue_by_doorbell Zhu Lingshan
2026-08-28 13:08   ` Christian König
2026-08-28 15:59     ` Zhu, Lingshan
2026-08-28  9:53 ` [PATCH 02/10] drm/amdgpu: keep the userq manager alive as long as its queues Zhu Lingshan
2026-08-28 13:09   ` Christian König
2026-08-28 15:59     ` Zhu, Lingshan
2026-08-28 16:26       ` Christian König
2026-08-28  9:53 ` [PATCH 03/10] drm/amdgpu/gfx11: hold userq refs in private fault worker Zhu Lingshan
2026-08-28 13:11   ` Christian König
2026-08-28 15:59     ` Zhu, Lingshan
2026-08-28  9:53 ` [PATCH 04/10] drm/amdgpu/gfx12: " Zhu Lingshan
2026-08-28  9:53 ` [PATCH 05/10] drm/amdgpu: implement asynchronous userq destruction routine Zhu Lingshan
2026-08-28  9:53 ` Zhu Lingshan [this message]
2026-08-28  9:53 ` [PATCH 07/10] drm/amdgpu: hold userq kref during isolation scheduling Zhu Lingshan
2026-08-28  9:53 ` [PATCH 08/10] drm/amdgpu: hold userq kref during suspend and resume Zhu Lingshan
2026-08-28  9:53 ` [PATCH 09/10] drm/amdgpu: free userq by kref_put when fails to create Zhu Lingshan
2026-08-28  9:53 ` [PATCH 10/10] drm/amdgpu: take queue kref in userq_create to avoid UAF Zhu Lingshan

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=20260828095349.9797-7-lingshan.zhu@amd.com \
    --to=lingshan.zhu@amd.com \
    --cc=Alexander.Deucher@amd.com \
    --cc=Christian.Koenig@amd.com \
    --cc=Ray.Huang@amd.com \
    --cc=amd-gfx@lists.freedesktop.org \
    --cc=felix.kuehling@amd.com \
    /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.