[v2] drm/syncobj: ensure progress for syncobj queries

[PATCH v2] drm/syncobj: ensure progress for syncobj queries

Posted by Chia-I Wu 1 year, 3 months ago

Userspace might poll a syncobj with the query ioctl.  Call
dma_fence_enable_sw_signaling to ensure dma_fence_is_signaled returns
true in finite time.

Fixes: 27b575a9aa2f ("drm/syncobj: add timeline payload query ioctl v6")
Signed-off-by: Chia-I Wu <olvaffe@gmail.com>

---

v2: add Signed-off-by and Fixes tags
---
 drivers/gpu/drm/drm_syncobj.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/drivers/gpu/drm/drm_syncobj.c b/drivers/gpu/drm/drm_syncobj.c
index 4fcfc0b9b386c..58c5593c897a2 100644
--- a/drivers/gpu/drm/drm_syncobj.c
+++ b/drivers/gpu/drm/drm_syncobj.c
@@ -1689,6 +1689,9 @@ int drm_syncobj_query_ioctl(struct drm_device *dev, void *data,
 			    DRM_SYNCOBJ_QUERY_FLAGS_LAST_SUBMITTED) {
 				point = fence->seqno;
 			} else {
+				/* ensure forward progress */
+				dma_fence_enable_sw_signaling(fence);
+
 				dma_fence_chain_for_each(iter, fence) {
 					if (iter->context != fence->context) {
 						dma_fence_put(iter);
-- 
2.47.0.105.g07ac214952-goog

Re: [PATCH v2] drm/syncobj: ensure progress for syncobj queries

Posted by Christian König 1 year, 3 months ago

Am 22.10.24 um 18:18 schrieb Chia-I Wu:
> Userspace might poll a syncobj with the query ioctl.  Call
> dma_fence_enable_sw_signaling to ensure dma_fence_is_signaled returns
> true in finite time.

Wait a second, just querying the fence status is absolutely not 
guaranteed to return true in finite time. That is well documented on the 
dma_fence() object.

When you want to poll on signaling from userspace you really need to 
call poll or the wait IOCTL with a zero timeout. That will also return 
immediately but should enable signaling while doing that.

So just querying the status should absolutely *not* enable signaling. 
That's an intentional separation.

Regards,
Christian.

>
> Fixes: 27b575a9aa2f ("drm/syncobj: add timeline payload query ioctl v6")
> Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
>
> ---
>
> v2: add Signed-off-by and Fixes tags
> ---
>   drivers/gpu/drm/drm_syncobj.c | 3 +++
>   1 file changed, 3 insertions(+)
>
> diff --git a/drivers/gpu/drm/drm_syncobj.c b/drivers/gpu/drm/drm_syncobj.c
> index 4fcfc0b9b386c..58c5593c897a2 100644
> --- a/drivers/gpu/drm/drm_syncobj.c
> +++ b/drivers/gpu/drm/drm_syncobj.c
> @@ -1689,6 +1689,9 @@ int drm_syncobj_query_ioctl(struct drm_device *dev, void *data,
>   			    DRM_SYNCOBJ_QUERY_FLAGS_LAST_SUBMITTED) {
>   				point = fence->seqno;
>   			} else {
> +				/* ensure forward progress */
> +				dma_fence_enable_sw_signaling(fence);
> +
>   				dma_fence_chain_for_each(iter, fence) {
>   					if (iter->context != fence->context) {
>   						dma_fence_put(iter);

Re: [PATCH v2] drm/syncobj: ensure progress for syncobj queries

Posted by Chia-I Wu 1 year, 3 months ago

On Tue, Oct 22, 2024 at 9:53 AM Christian König
<christian.koenig@amd.com> wrote:
>
> Am 22.10.24 um 18:18 schrieb Chia-I Wu:
> > Userspace might poll a syncobj with the query ioctl.  Call
> > dma_fence_enable_sw_signaling to ensure dma_fence_is_signaled returns
> > true in finite time.
>
> Wait a second, just querying the fence status is absolutely not
> guaranteed to return true in finite time. That is well documented on the
> dma_fence() object.
>
> When you want to poll on signaling from userspace you really need to
> call poll or the wait IOCTL with a zero timeout. That will also return
> immediately but should enable signaling while doing that.
>
> So just querying the status should absolutely *not* enable signaling.
> That's an intentional separation.
I think it depends on what semantics DRM_IOCTL_SYNCOBJ_QUERY should have.

If DRM_IOCTL_SYNCOBJ_QUERY is mainly for vulkan timeline semaphores,
it is a bit heavy if userspace has to do a
DRM_IOCTL_SYNCOBJ_TIMELINE_WAIT before a query.


>
> Regards,
> Christian.
>
> >
> > Fixes: 27b575a9aa2f ("drm/syncobj: add timeline payload query ioctl v6")
> > Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
> >
> > ---
> >
> > v2: add Signed-off-by and Fixes tags
> > ---
> >   drivers/gpu/drm/drm_syncobj.c | 3 +++
> >   1 file changed, 3 insertions(+)
> >
> > diff --git a/drivers/gpu/drm/drm_syncobj.c b/drivers/gpu/drm/drm_syncobj.c
> > index 4fcfc0b9b386c..58c5593c897a2 100644
> > --- a/drivers/gpu/drm/drm_syncobj.c
> > +++ b/drivers/gpu/drm/drm_syncobj.c
> > @@ -1689,6 +1689,9 @@ int drm_syncobj_query_ioctl(struct drm_device *dev, void *data,
> >                           DRM_SYNCOBJ_QUERY_FLAGS_LAST_SUBMITTED) {
> >                               point = fence->seqno;
> >                       } else {
> > +                             /* ensure forward progress */
> > +                             dma_fence_enable_sw_signaling(fence);
> > +
> >                               dma_fence_chain_for_each(iter, fence) {
> >                                       if (iter->context != fence->context) {
> >                                               dma_fence_put(iter);
>

Re: [PATCH v2] drm/syncobj: ensure progress for syncobj queries

Posted by Chia-I Wu 1 year, 3 months ago

On Tue, Oct 22, 2024 at 10:24 AM Chia-I Wu <olvaffe@gmail.com> wrote:
>
> On Tue, Oct 22, 2024 at 9:53 AM Christian König
> <christian.koenig@amd.com> wrote:
> >
> > Am 22.10.24 um 18:18 schrieb Chia-I Wu:
> > > Userspace might poll a syncobj with the query ioctl.  Call
> > > dma_fence_enable_sw_signaling to ensure dma_fence_is_signaled returns
> > > true in finite time.
> >
> > Wait a second, just querying the fence status is absolutely not
> > guaranteed to return true in finite time. That is well documented on the
> > dma_fence() object.
> >
> > When you want to poll on signaling from userspace you really need to
> > call poll or the wait IOCTL with a zero timeout. That will also return
> > immediately but should enable signaling while doing that.
> >
> > So just querying the status should absolutely *not* enable signaling.
> > That's an intentional separation.
> I think it depends on what semantics DRM_IOCTL_SYNCOBJ_QUERY should have.
>
> If DRM_IOCTL_SYNCOBJ_QUERY is mainly for vulkan timeline semaphores,
> it is a bit heavy if userspace has to do a
> DRM_IOCTL_SYNCOBJ_TIMELINE_WAIT before a query.

I filed a Mesa issue,
https://gitlab.freedesktop.org/mesa/mesa/-/issues/12094, and Faith
suggested a kernel-side fix as well.  Should we reconsider this?