git/linux-net
1
0
Fork 0
mirror of https://git.kernel.org/pub/scm/linux/kernel/git/netdev/net.git/ synced 2026-04-18 06:33:43 -04:00
Code Issues Packages Projects Releases Wiki Activity

drm/amdgpu: Add wait IOCTL timeline syncobj support

Browse Source 
Add user fence wait IOCTL timeline syncobj support.

v2:(Christian)
  - handle dma_fence_wait() return value.
  - shorten the variable name syncobj_timeline_points a bit.
  - move num_points up to avoid padding issues.

v3:(Christian)
  - Handle timeline drm_syncobj_find_fence() call error
    handling
  - Use dma_fence_unwrap_for_each() in timeline fence as
    there could be more than one fence.

v4:(Christian)
  - Drop the first num_fences since fence is always included in
    the dma_fence_unwrap_for_each() iteration, when fence != f
    then fence is most likely just a container.

v5: Added Alex RB to merge the kernel UAPI changes since he has
    already approved the amdgpu_drm.h changes.

Signed-off-by: Arunpravin Paneer Selvam <Arunpravin.PaneerSelvam@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This commit is contained in:
Arunpravin Paneer Selvam
2024-10-30 10:33:28 +05:30
committed by Alex Deucher
parent a292fdecd7
commit 15e30a6e47
2 changed files with 99 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

91
drivers/gpu/drm/amd/amdgpu/amdgpu_userq_fence.c
View File

@@ -24,6 +24,7 @@
#include <linux/kref.h>
#include <linux/slab.h>
#include <linux/dma-fence-unwrap.h>
#include <drm/drm_exec.h>
#include <drm/drm_syncobj.h>
@@ -474,11 +475,11 @@ free_syncobj_handles:
int amdgpu_userq_wait_ioctl(struct drm_device *dev, void *data,
struct drm_file *filp)
{
u32 *syncobj_handles, *timeline_points, *timeline_handles, *bo_handles;
u32 num_syncobj, num_bo_handles, num_points;
struct drm_amdgpu_userq_fence_info *fence_info = NULL;
struct drm_amdgpu_userq_wait *wait_info = data;
u32 *syncobj_handles, *bo_handles;
struct dma_fence **fences = NULL;
u32 num_syncobj, num_bo_handles;
struct drm_gem_object **gobj;
struct drm_exec exec;
int r, i, entry, cnt;
@@ -498,11 +499,26 @@ int amdgpu_userq_wait_ioctl(struct drm_device *dev, void *data,
goto free_bo_handles;
}
num_points = wait_info->num_points;
timeline_handles = memdup_user(u64_to_user_ptr(wait_info->syncobj_timeline_handles),
sizeof(u32) * num_points);
if (IS_ERR(timeline_handles)) {
r = PTR_ERR(timeline_handles);
goto free_syncobj_handles;
}
timeline_points = memdup_user(u64_to_user_ptr(wait_info->syncobj_timeline_points),
sizeof(u32) * num_points);
if (IS_ERR(timeline_points)) {
r = PTR_ERR(timeline_points);
goto free_timeline_handles;
}
/* Array of GEM object handles */
gobj = kmalloc_array(num_bo_handles, sizeof(*gobj), GFP_KERNEL);
if (!gobj) {
r = -ENOMEM;
goto free_syncobj_handles;
goto free_timeline_points;
}
for (entry = 0; entry < num_bo_handles; entry++) {
@@ -524,12 +540,34 @@ int amdgpu_userq_wait_ioctl(struct drm_device *dev, void *data,
}
if (!wait_info->num_fences) {
if (num_points) {
struct dma_fence_unwrap iter;
struct dma_fence *fence;
struct dma_fence *f;
for (i = 0; i < num_points; i++) {
r = drm_syncobj_find_fence(filp, timeline_handles[i],
timeline_points[i],
DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT,
&fence);
if (r)
goto exec_fini;
dma_fence_unwrap_for_each(f, &iter, fence)
num_fences++;
dma_fence_put(fence);
}
}
/* Count syncobj's fence */
for (i = 0; i < num_syncobj; i++) {
struct dma_fence *fence;
r = drm_syncobj_find_fence(filp, syncobj_handles[i],
0, 0, &fence);
0,
DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT,
&fence);
if (r)
goto exec_fini;
@@ -588,12 +626,41 @@ int amdgpu_userq_wait_ioctl(struct drm_device *dev, void *data,
}
}
if (num_points) {
struct dma_fence_unwrap iter;
struct dma_fence *fence;
struct dma_fence *f;
for (i = 0; i < num_points; i++) {
r = drm_syncobj_find_fence(filp, timeline_handles[i],
timeline_points[i],
DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT,
&fence);
if (r)
goto free_fences;
dma_fence_unwrap_for_each(f, &iter, fence) {
if (WARN_ON_ONCE(num_fences >= wait_info->num_fences)) {
r = -EINVAL;
goto free_fences;
}
dma_fence_get(f);
fences[num_fences++] = f;
}
dma_fence_put(fence);
}
}
/* Retrieve syncobj's fence */
for (i = 0; i < num_syncobj; i++) {
struct dma_fence *fence;
r = drm_syncobj_find_fence(filp, syncobj_handles[i],
0, 0, &fence);
0,
DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT,
&fence);
if (r)
goto free_fences;
@@ -616,9 +683,13 @@ int amdgpu_userq_wait_ioctl(struct drm_device *dev, void *data,
* Just waiting on other driver fences should
* be good for now
*/
dma_fence_wait(fences[i], false);
dma_fence_put(fences[i]);
r = dma_fence_wait(fences[i], true);
if (r) {
dma_fence_put(fences[i]);
goto free_fences;
}
dma_fence_put(fences[i]);
continue;
}
@@ -664,6 +735,8 @@ int amdgpu_userq_wait_ioctl(struct drm_device *dev, void *data,
drm_gem_object_put(gobj[i]);
kfree(gobj);
kfree(timeline_points);
kfree(timeline_handles);
kfree(syncobj_handles);
kfree(bo_handles);
@@ -681,6 +754,10 @@ put_gobj:
while (entry-- > 0)
drm_gem_object_put(gobj[entry]);
kfree(gobj);
free_timeline_points:
kfree(timeline_points);
free_timeline_handles:
kfree(timeline_handles);
free_syncobj_handles:
kfree(syncobj_handles);
free_bo_handles: