Re: [PATCH v2 05/12] drm/amdgpu/userq: Use drm_gem_objects_lookup in amdgpu_userq_wait_ioctl

[Tvrtko Ursulin]

On 23/02/2026 13:58, Khatri, Sunil wrote:

On 23-02-2026 06:11 pm, Tvrtko Ursulin wrote:
Use the existing helper instead of open coding it

Signed-off-by: Tvrtko Ursulin <tvrtko.ursu...@igalia.com>
---
  .../gpu/drm/amd/amdgpu/amdgpu_userq_fence.c   | 79 ++++++-------------
  1 file changed, 23 insertions(+), 56 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_userq_fence.c b/ 
drivers/gpu/drm/amd/amdgpu/amdgpu_userq_fence.c
index ee6f03f75b41..d779671bd0db 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_userq_fence.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_userq_fence.c
@@ -610,44 +610,29 @@ int amdgpu_userq_signal_ioctl(struct drm_device 
*dev, void *data,
  int amdgpu_userq_wait_ioctl(struct drm_device *dev, void *data,
                  struct drm_file *filp)
  {
-    u32 *syncobj_handles, *timeline_points, *timeline_handles, 
*bo_handles_read, *bo_handles_write;
-    u32 num_syncobj, num_read_bo_handles, num_write_bo_handles;
-    struct drm_amdgpu_userq_fence_info *fence_info = NULL;
      struct drm_amdgpu_userq_wait *wait_info = data;
+    const unsigned int num_write_bo_handles = wait_info- 
>num_bo_write_handles;
+    const unsigned int num_read_bo_handles = wait_info- 
>num_bo_read_handles;
+    struct drm_amdgpu_userq_fence_info *fence_info = NULL;
      struct amdgpu_fpriv *fpriv = filp->driver_priv;
      struct amdgpu_userq_mgr *userq_mgr = &fpriv->userq_mgr;
+    struct drm_gem_object **gobj_write, **gobj_read;
+    u32 *timeline_points, *timeline_handles;
      struct amdgpu_usermode_queue *waitq;
-    struct drm_gem_object **gobj_write;
-    struct drm_gem_object **gobj_read;
+    u32 *syncobj_handles, num_syncobj;
      struct dma_fence **fences = NULL;
      u16 num_points, num_fences = 0;
-    int r, i, rentry, wentry, cnt;
      struct drm_exec exec;
+    int r, i, cnt;
      if (!amdgpu_userq_enabled(dev))
          return -ENOTSUPP;
-    num_read_bo_handles = wait_info->num_bo_read_handles;
-    bo_handles_read = memdup_user(u64_to_user_ptr(wait_info- 
>bo_read_handles),
-                      size_mul(sizeof(u32), num_read_bo_handles));
-    if (IS_ERR(bo_handles_read))
-        return PTR_ERR(bo_handles_read);
-
-    num_write_bo_handles = wait_info->num_bo_write_handles;
-    bo_handles_write = memdup_user(u64_to_user_ptr(wait_info- 
>bo_write_handles),
-                       size_mul(sizeof(u32), num_write_bo_handles));
-    if (IS_ERR(bo_handles_write)) {
-        r = PTR_ERR(bo_handles_write);
-        goto free_bo_handles_read;
-    }
-
      num_syncobj = wait_info->num_syncobj_handles;
      syncobj_handles = memdup_user(u64_to_user_ptr(wait_info- 
>syncobj_handles),
                        size_mul(sizeof(u32), num_syncobj));
-    if (IS_ERR(syncobj_handles)) {
-        r = PTR_ERR(syncobj_handles);
-        goto free_bo_handles_write;
-    }
+    if (IS_ERR(syncobj_handles))
+        return PTR_ERR(syncobj_handles);
      num_points = wait_info->num_syncobj_timeline_handles;
      timeline_handles = memdup_user(u64_to_user_ptr(wait_info- 
>syncobj_timeline_handles),
@@ -664,33 +649,19 @@ int amdgpu_userq_wait_ioctl(struct drm_device 
*dev, void *data,
          goto free_timeline_handles;
      }
-    gobj_read = kmalloc_array(num_read_bo_handles, 
sizeof(*gobj_read), GFP_KERNEL);
-    if (!gobj_read) {
-        r = -ENOMEM;
+    r = drm_gem_objects_lookup(filp,
+                   u64_to_user_ptr(wait_info->bo_read_handles),
+                   num_read_bo_handles,
+                   &gobj_read);

If i am not wrong even using the helper function will not avoid an OOM 
situation in case num_read_bo_handles is a very big number? This is just 
using the helper to reorganize the code but not having additonal 
checks... and alos valid only on drm gem objects type...

Last part about only DRM GEM object type I did not understand. It is 
currently using drm_gem_object_lookup so that is the same.

Otherwise yes, although my understanding of the overall topic was you 
will apply the limits on top of the consolidation patches.

Regards,

Tvrtko

+    if (r)
          goto free_timeline_points;
-    }
-    for (rentry = 0; rentry < num_read_bo_handles; rentry++) {
-        gobj_read[rentry] = drm_gem_object_lookup(filp, 
bo_handles_read[rentry]);
-        if (!gobj_read[rentry]) {
-            r = -ENOENT;
-            goto put_gobj_read;
-        }
-    }
-
-    gobj_write = kmalloc_array(num_write_bo_handles, 
sizeof(*gobj_write), GFP_KERNEL);
-    if (!gobj_write) {
-        r = -ENOMEM;
+    r = drm_gem_objects_lookup(filp,
+                   u64_to_user_ptr(wait_info->bo_write_handles),
+                   num_write_bo_handles,
+                   &gobj_write);
+    if (r)
          goto put_gobj_read;
-    }
-
-    for (wentry = 0; wentry < num_write_bo_handles; wentry++) {
-        gobj_write[wentry] = drm_gem_object_lookup(filp, 
bo_handles_write[wentry]);
-        if (!gobj_write[wentry]) {
-            r = -ENOENT;
-            goto put_gobj_write;
-        }
-    }
      drm_exec_init(&exec, DRM_EXEC_INTERRUPTIBLE_WAIT,
                (num_read_bo_handles + num_write_bo_handles));
@@ -947,12 +918,12 @@ int amdgpu_userq_wait_ioctl(struct drm_device 
*dev, void *data,
  exec_fini:
      drm_exec_fini(&exec);
  put_gobj_write:
-    while (wentry-- > 0)
-        drm_gem_object_put(gobj_write[wentry]);
+    for (i = 0; i < num_write_bo_handles; i++)
+        drm_gem_object_put(gobj_write[i]);
      kfree(gobj_write);
  put_gobj_read:
-    while (rentry-- > 0)
-        drm_gem_object_put(gobj_read[rentry]);
+    for (i = 0; i < num_read_bo_handles; i++)
+        drm_gem_object_put(gobj_read[i]);;
      kfree(gobj_read);
  free_timeline_points:
      kfree(timeline_points);
@@ -960,10 +931,6 @@ int amdgpu_userq_wait_ioctl(struct drm_device 
*dev, void *data,
      kfree(timeline_handles);
  free_syncobj_handles:
      kfree(syncobj_handles);
-free_bo_handles_write:
-    kfree(bo_handles_write);
-free_bo_handles_read:
-    kfree(bo_handles_read);
      return r;
  }