AMD-GFX Archive on lore.kernel.org
 help / color / mirror / Atom feed
From: "David (Ming Qiang) Wu" <David.Wu3@amd.com>
To: <amd-gfx@lists.freedesktop.org>, <Christian.Koenig@amd.com>,
	<alexander.deucher@amd.com>
Cc: <leo.liu@amd.com>, <Boyuan.Zhang@amd.com>, <David.Wu3@amd.com>
Subject: [PATCH 09/14] drm/amdgpu/userq: rework on amdgpu_userq_create_wptr_mapping
Date: Tue, 10 Feb 2026 16:47:24 -0500	[thread overview]
Message-ID: <20260210214729.80964-10-David.Wu3@amd.com> (raw)
In-Reply-To: <20260210214729.80964-1-David.Wu3@amd.com>

v3 - lock both VM and WPTR BO (Christian)
v2 - get a reference (amdgpu_bo_ref(wptr_obj->obj)) before
     amdgpu_bo_unreserve() to avoid use-after-free issue

remove amdgpu_userq_map_gtt_bo_to_gart() and move its calls
into amdgpu_userq_create_wptr_mapping() to eliminate duplicated
calls.

Signed-off-by: David (Ming Qiang) Wu <David.Wu3@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c | 89 ++++++++---------------
 1 file changed, 32 insertions(+), 57 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c
index ebb0d8a9967f..59e593b3bae7 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c
@@ -1573,34 +1573,6 @@ int amdgpu_userq_post_reset(struct amdgpu_device *adev, bool vram_lost)
 	return r;
 }
 
-static int
-amdgpu_userq_map_gtt_bo_to_gart(struct amdgpu_device *adev, struct amdgpu_bo *bo)
-{
-	int ret;
-
-	ret = amdgpu_bo_reserve(bo, true);
-	if (ret) {
-		dev_err(adev->dev, "Failed to reserve bo. ret %d\n", ret);
-		goto err_reserve_bo_failed;
-	}
-
-	ret = amdgpu_ttm_alloc_gart(&bo->tbo);
-	if (ret) {
-		dev_err(adev->dev, "Failed to bind bo to GART. ret %d\n", ret);
-		goto err_map_bo_gart_failed;
-	}
-
-	amdgpu_bo_unreserve(bo);
-	bo = amdgpu_bo_ref(bo);
-
-	return 0;
-
-err_map_bo_gart_failed:
-	amdgpu_bo_unreserve(bo);
-err_reserve_bo_failed:
-	return ret;
-}
-
 int amdgpu_userq_create_wptr_mapping(struct amdgpu_userq_mgr *uq_mgr,
 			      struct amdgpu_usermode_queue *queue,
 			      uint64_t wptr)
@@ -1609,53 +1581,56 @@ int amdgpu_userq_create_wptr_mapping(struct amdgpu_userq_mgr *uq_mgr,
 	struct amdgpu_bo_va_mapping *wptr_mapping;
 	struct amdgpu_vm *wptr_vm;
 	struct amdgpu_userq_obj *wptr_obj = &queue->wptr_obj;
+	struct drm_exec exec;
 	int ret;
 
+	wptr &= AMDGPU_GMC_HOLE_MASK;
 	wptr_vm = queue->vm;
-	ret = amdgpu_bo_reserve(wptr_vm->root.bo, false);
-	if (ret)
-		return ret;
+	drm_exec_init(&exec, DRM_EXEC_IGNORE_DUPLICATES, 0);
+	drm_exec_until_all_locked(&exec) {
+		ret = amdgpu_vm_lock_pd(wptr_vm, &exec, 0);
+		drm_exec_retry_on_contention(&exec);
+		if (unlikely(ret))
+			goto out_unlock;
 
-	wptr &= AMDGPU_GMC_HOLE_MASK;
-	wptr_mapping = amdgpu_vm_bo_lookup_mapping(wptr_vm, wptr >> PAGE_SHIFT);
-	amdgpu_bo_unreserve(wptr_vm->root.bo);
-	if (!wptr_mapping) {
-		dev_err(adev->dev, "Failed to lookup wptr bo\n");
-		return -EINVAL;
+		wptr_mapping = amdgpu_vm_bo_lookup_mapping(wptr_vm, wptr >> PAGE_SHIFT);
+		if (!wptr_mapping) {
+			dev_err(adev->dev, "Failed to lookup wptr bo\n");
+			goto out_unlock;
+		}
+
+		wptr_obj->obj = wptr_mapping->bo_va->base.bo;
+		ret = drm_exec_lock_obj(&exec, &wptr_obj->obj->tbo.base);
+		drm_exec_retry_on_contention(&exec);
+		if (unlikely(ret))
+			goto out_unlock;
 	}
 
-	wptr_obj->obj = wptr_mapping->bo_va->base.bo;
+	/* Now both VM and WPTR BO are locked */
 	if (wptr_obj->obj->tbo.base.size > PAGE_SIZE) {
 		dev_err(adev->dev, "Requested GART mapping for wptr bo larger than one page\n");
-		return -EINVAL;
-	}
-
-	ret = amdgpu_userq_map_gtt_bo_to_gart(adev, wptr_obj->obj);
-	if (ret) {
-		dev_err(adev->dev, "Failed to map wptr bo to GART\n");
-		return ret;
-	}
-
-	ret = amdgpu_bo_reserve(wptr_obj->obj, true);
-	if (ret) {
-		dev_err(adev->dev, "Failed to reserve wptr bo\n");
-		return ret;
+		ret = -EINVAL;
+		goto out_unlock;
 	}
 
 	/* TODO use eviction fence instead of pinning. */
 	ret = amdgpu_bo_pin(wptr_obj->obj, AMDGPU_GEM_DOMAIN_GTT);
 	if (ret) {
 		drm_file_err(uq_mgr->file, "[Usermode queues] Failed to pin wptr bo\n");
-		goto unresv_bo;
+		goto out_unlock;
 	}
 
+	ret = amdgpu_ttm_alloc_gart(&wptr_obj->obj->tbo);
+	if (ret) {
+		dev_err(adev->dev, "Failed to bind bo to GART. ret %d\n", ret);
+		amdgpu_bo_unpin(wptr_obj->obj);
+		goto out_unlock;
+	}
 	queue->wptr_obj.gpu_addr = amdgpu_bo_gpu_offset(wptr_obj->obj);
-	amdgpu_bo_unreserve(wptr_obj->obj);
-
-	return 0;
+	amdgpu_bo_ref(wptr_obj->obj);
 
-unresv_bo:
-	amdgpu_bo_unreserve(wptr_obj->obj);
+out_unlock:
+	drm_exec_fini(&exec);
 	return ret;
 
 }
-- 
2.43.0


  parent reply	other threads:[~2026-02-10 21:48 UTC|newest]

Thread overview: 22+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-02-10 21:47 [PATCH 00/14] user queue support for VCN 4.0.5 David (Ming Qiang) Wu
2026-02-10 21:47 ` [PATCH 01/14] amdgpu: add global aggregated doorbell bo David (Ming Qiang) Wu
2026-02-10 21:47 ` [PATCH 02/14] drm/amdgpu: add AMDGPU_GEM_GLOBAL_AGGREGATED_DOORBELL David (Ming Qiang) Wu
2026-02-10 21:47 ` [PATCH 03/14] drm/amdgpu/userq: add doorbell size for VCN and VPE David (Ming Qiang) Wu
2026-02-10 21:47 ` [PATCH 04/14] amdgpu/umsch: Update UMSCH interface and mqd structure David (Ming Qiang) Wu
2026-02-10 21:47 ` [PATCH 05/14] drm/amdgpu: use amdgpu_user_queue instead of amdgpu_umsch_mm David (Ming Qiang) Wu
2026-02-10 21:47 ` [PATCH 06/14] drm/amdgpu/vcn: changes when kernel queue is disabled David (Ming Qiang) Wu
2026-02-10 21:47 ` [PATCH 07/14] amdgpu/umsch: Add VCN IP init to umsch driver David (Ming Qiang) Wu
2026-02-10 21:47 ` [PATCH 08/14] drm/amdgpu/userq: change mes_userq_create_wptr_mapping() to be common David (Ming Qiang) Wu
2026-02-10 21:47 ` David (Ming Qiang) Wu [this message]
2026-04-24 13:37   ` [PATCH 09/14] drm/amdgpu/userq: rework on amdgpu_userq_create_wptr_mapping Alex Deucher
2026-04-27  8:44     ` Zhang, Jesse(Jie)
2026-02-10 21:47 ` [PATCH 10/14] drm/amdgpu/umsch: user queue support for vcn David (Ming Qiang) Wu
2026-02-10 21:47 ` [PATCH 11/14] drm/amdgpu: add AMDGPU_INFO_DOORBELL David (Ming Qiang) Wu
2026-02-10 21:47 ` [PATCH 12/14] drm/amdgpu/vcn: handle interrupt received from fw David (Ming Qiang) Wu
2026-02-10 21:47 ` [PATCH 13/14] drm/amdgpu/umsch: userq suspend and resume context David (Ming Qiang) Wu
2026-02-10 21:47 ` [PATCH 14/14] drm/amdgpu/vcn: handle the suspend context interrupt David (Ming Qiang) Wu
2026-02-18 15:37 ` [PATCH 00/14] user queue support for VCN 4.0.5 Saleemkhan
2026-02-18 18:13   ` Wu, David
2026-02-18 23:59     ` Saleemkhan
2026-02-19 18:00 ` David Wu
2026-04-17 15:22   ` Wu, David

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=20260210214729.80964-10-David.Wu3@amd.com \
    --to=david.wu3@amd.com \
    --cc=Boyuan.Zhang@amd.com \
    --cc=Christian.Koenig@amd.com \
    --cc=alexander.deucher@amd.com \
    --cc=amd-gfx@lists.freedesktop.org \
    --cc=leo.liu@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 a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox