Skip to contentSkip to conversation

drm/syncobj: Fix sync syncobj issue

PATCH In review jie1zhan 4 participants 6 messages Jul 7, 2022 – Jul 13, 2022

Subsystems (2)
lore
Share & tools
lore

jie1zhan

author
#0 lore raw

PATCH v2 1/1 drm/syncobj: Fix sync syncobj issue

enable signaling after flatten dma_fence_chains on transfer

---
 drivers/gpu/drm/drm_syncobj.c | 1 +
 1 file changed, 1 insertion(+)
drivers/gpu/drm/drm_syncobj.c +1 -0
index 7e48dcd1bee4..0d9d3577325f 100644
@@ -920,6 +920,7 @@ static int drm_syncobj_transfer_to_timeline(struct drm_file *file_private,
 	if (ret)
 		goto err_free_fence;
 
+	dma_fence_enable_sw_signaling(fence);
 	chain = dma_fence_chain_alloc();
 	if (!chain) {
 		ret = -ENOMEM;
Signature · 1 line
2.25.1
#1 lore raw

Ping to the Intel guys here. Especially Lucas/Nirmoy/Lionel.

IIRC you stumbled over that problem as well, have you found any solution?

Regards,
Christian.

Am 07.07.22 um 12:29 schrieb jie1zhan:

Quoted 19 lines

enable signaling after flatten dma_fence_chains on transfer

---
  drivers/gpu/drm/drm_syncobj.c | 1 +
  1 file changed, 1 insertion(+)
drivers/gpu/drm/drm_syncobj.c +1 -0
index 7e48dcd1bee4..0d9d3577325f 100644
@@ -920,6 +920,7 @@ static int drm_syncobj_transfer_to_timeline(struct drm_file *file_private,
  	if (ret)
  		goto err_free_fence;
 
+	dma_fence_enable_sw_signaling(fence);
  	chain = dma_fence_chain_alloc();
  	if (!chain) {
  		ret = -ENOMEM;
#2 lore raw

I'll let Lucas comment. I've only looked a little at it.
From what I remember just enabling sw_signaling was enough to fix the issue.

-Lionel

Quoted 28 lines · Christian König

Ping to the Intel guys here. Especially Lucas/Nirmoy/Lionel.

IIRC you stumbled over that problem as well, have you found any solution?

Regards,
Christian.

Am 07.07.22 um 12:29 schrieb jie1zhan:

Quoted 19 lines

enable signaling after flatten dma_fence_chains on transfer

---
  drivers/gpu/drm/drm_syncobj.c | 1 +

  1 file changed, 1 insertion(+)
drivers/gpu/drm/drm_syncobj.c +0 -0
index 7e48dcd1bee4..0d9d3577325f 100644
@@ -920,6 +920,7 @@ static int 
 drm_syncobj_transfer_to_timeline(struct drm_file *file_private,
       if (ret)
           goto err_free_fence;
   +    dma_fence_enable_sw_signaling(fence);
       chain = dma_fence_chain_alloc();
       if (!chain) {

          ret = -ENOMEM;

#3 lore raw

Yeah, adding dma_fence_enable_sw_signaling() is the right thing to do.

The question is where to add that? Usually right before the fence is returned from the object or queried from userspace would probably be the right place.

Regards,
Christian.

Am 12.07.22 um 16:22 schrieb Lionel Landwerlin:

Quoted 35 lines

I'll let Lucas comment. I've only looked a little at it.
From what I remember just enabling sw_signaling was enough to fix the issue.

-Lionel

Quoted 28 lines · Christian König

Ping to the Intel guys here. Especially Lucas/Nirmoy/Lionel.

IIRC you stumbled over that problem as well, have you found any solution?

Regards,
Christian.

Am 07.07.22 um 12:29 schrieb jie1zhan:

Quoted 19 lines

enable signaling after flatten dma_fence_chains on transfer

---
  drivers/gpu/drm/drm_syncobj.c | 1 +

  1 file changed, 1 insertion(+)
drivers/gpu/drm/drm_syncobj.c +0 -0
index 7e48dcd1bee4..0d9d3577325f 100644
@@ -920,6 +920,7 @@ static int 
 drm_syncobj_transfer_to_timeline(struct drm_file *file_private,
       if (ret)
           goto err_free_fence;
   +    dma_fence_enable_sw_signaling(fence);
       chain = dma_fence_chain_alloc();
       if (!chain) {

          ret = -ENOMEM;

#4 lore raw
In reply to Christian König

Hi Christian,

Quoted 3 lines · Christian König

Ping to the Intel guys here. Especially Lucas/Nirmoy/Lionel.

IIRC you stumbled over that problem as well, have you found any solution?

I might be wrong but  I think you are talking about igt@syncobj_timeline@transfer-timeline-point testcase which seems to be

green in CI now https://intel-gfx-ci.01.org/tree/drm-tip/igt@[email protected]

Lucas found out that the issues got fixed after ec8d985ff26f ("drm: use dma_fence_unwrap_merge() in drm_syncobj")

Regards,

Nirmoy

Quoted 25 lines

Regards,
Christian.

Am 07.07.22 um 12:29 schrieb jie1zhan:

Quoted 19 lines

enable signaling after flatten dma_fence_chains on transfer

---
  drivers/gpu/drm/drm_syncobj.c | 1 +

  1 file changed, 1 insertion(+)
drivers/gpu/drm/drm_syncobj.c +0 -0
index 7e48dcd1bee4..0d9d3577325f 100644
@@ -920,6 +920,7 @@ static int 
 drm_syncobj_transfer_to_timeline(struct drm_file *file_private,
       if (ret)
           goto err_free_fence;
   +    dma_fence_enable_sw_signaling(fence);
       chain = dma_fence_chain_alloc();
       if (!chain) {

          ret = -ENOMEM;

#5 lore raw

Am 13.07.22 um 10:42 schrieb Das, Nirmoy:

Quoted 5 lines

Hi Christian,

Quoted 2 lines · Christian König

Ping to the Intel guys here. Especially Lucas/Nirmoy/Lionel.

Quoted 7 lines
Quoted 1 line

IIRC you stumbled over that problem as well, have you found any solution?

I might be wrong but  I think you are talking about igt@syncobj_timeline@transfer-timeline-point testcase which seems to be

green in CI now https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fintel-gfx-ci.01.org%2Ftree%2Fdrm-tip%2Figt%40syncobj_timeline%40transfer-timeline-point.html&data=05%7C01%7Cchristian.koenig%40amd.com%7C722fc33842734ef5c05108da64abac60%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637932985747614383%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=q8JygSVZfGA8cEZn%2BdxVXX79pkpXRjZTS8kBQ6Lq%2Bmw%3D&reserved=0

Lucas found out that the issues got fixed after ec8d985ff26f ("drm: use dma_fence_unwrap_merge() in drm_syncobj")

Yeah, but that's just coincident. The original bug that we fail to enable signaling in the syncobj is still there.

No that this is any major problem, but it would still be nice to have that fixed.

Regards,
Christian.

Quoted 31 lines

Regards,

Nirmoy

Quoted 25 lines

Regards,
Christian.

Am 07.07.22 um 12:29 schrieb jie1zhan:

Quoted 19 lines

enable signaling after flatten dma_fence_chains on transfer

---
  drivers/gpu/drm/drm_syncobj.c | 1 +

  1 file changed, 1 insertion(+)
drivers/gpu/drm/drm_syncobj.c +0 -0
index 7e48dcd1bee4..0d9d3577325f 100644
@@ -920,6 +920,7 @@ static int 
 drm_syncobj_transfer_to_timeline(struct drm_file *file_private,
       if (ret)
           goto err_free_fence;
   +    dma_fence_enable_sw_signaling(fence);
       chain = dma_fence_chain_alloc();
       if (!chain) {

          ret = -ENOMEM;