From 3c43177ffb54ea5be97505eb8e2690e99ac96bc9 Mon Sep 17 00:00:00 2001 From: Erik Kurzinger Date: Fri, 19 Jan 2024 08:32:06 -0800 Subject: drm/syncobj: call drm_syncobj_fence_add_wait when WAIT_AVAILABLE flag is set When waiting for a syncobj timeline point whose fence has not yet been submitted with the WAIT_FOR_SUBMIT flag, a callback is registered using drm_syncobj_fence_add_wait and the thread is put to sleep until the timeout expires. If the fence is submitted before then, drm_syncobj_add_point will wake up the sleeping thread immediately which will proceed to wait for the fence to be signaled. However, if the WAIT_AVAILABLE flag is used instead, drm_syncobj_fence_add_wait won't get called, meaning the waiting thread will always sleep for the full timeout duration, even if the fence gets submitted earlier. If it turns out that the fence *has* been submitted by the time it eventually wakes up, it will still indicate to userspace that the wait completed successfully (it won't return -ETIME), but it will have taken much longer than it should have. To fix this, we must call drm_syncobj_fence_add_wait if *either* the WAIT_FOR_SUBMIT flag or the WAIT_AVAILABLE flag is set. The only difference being that with WAIT_FOR_SUBMIT we will also wait for the fence to be signaled after it has been submitted while with WAIT_AVAILABLE we will return immediately. IGT test patch: https://lists.freedesktop.org/archives/igt-dev/2024-January/067537.html v1 -> v2: adjust lockdep_assert_none_held_once condition (cherry picked from commit 8c44ea81634a4a337df70a32621a5f3791be23df) Fixes: 01d6c3578379 ("drm/syncobj: add support for timeline point wait v8") Signed-off-by: Erik Kurzinger Signed-off-by: Simon Ser Reviewed-by: Daniel Vetter Reviewed-by: Simon Ser Link: https://patchwork.freedesktop.org/patch/msgid/20240119163208.3723457-1-ekurzinger@nvidia.com --- drivers/gpu/drm/drm_syncobj.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) (limited to 'drivers/gpu/drm/drm_syncobj.c') diff --git a/drivers/gpu/drm/drm_syncobj.c b/drivers/gpu/drm/drm_syncobj.c index 84101baeecc6..56b0677acfa3 100644 --- a/drivers/gpu/drm/drm_syncobj.c +++ b/drivers/gpu/drm/drm_syncobj.c @@ -1040,7 +1040,8 @@ static signed long drm_syncobj_array_wait_timeout(struct drm_syncobj **syncobjs, uint64_t *points; uint32_t signaled_count, i; - if (flags & DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT) + if (flags & (DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT | + DRM_SYNCOBJ_WAIT_FLAGS_WAIT_AVAILABLE)) lockdep_assert_none_held_once(); points = kmalloc_array(count, sizeof(*points), GFP_KERNEL); @@ -1109,7 +1110,8 @@ static signed long drm_syncobj_array_wait_timeout(struct drm_syncobj **syncobjs, * fallthough and try a 0 timeout wait! */ - if (flags & DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT) { + if (flags & (DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT | + DRM_SYNCOBJ_WAIT_FLAGS_WAIT_AVAILABLE)) { for (i = 0; i < count; ++i) drm_syncobj_fence_add_wait(syncobjs[i], &entries[i]); } -- cgit v1.2.3-70-g09d2 From 2aa6f5b0fd052e363bb9d4b547189f0bf6b3d6d3 Mon Sep 17 00:00:00 2001 From: Erik Kurzinger Date: Wed, 21 Feb 2024 10:44:28 -0800 Subject: drm/syncobj: handle NULL fence in syncobj_eventfd_entry_func During syncobj_eventfd_entry_func, dma_fence_chain_find_seqno may set the fence to NULL if the given seqno is signaled and a later seqno has already been submitted. In that case, the eventfd should be signaled immediately which currently does not happen. This is a similar issue to the one addressed by commit b19926d4f3a6 ("drm/syncobj: Deal with signalled fences in drm_syncobj_find_fence."). As a fix, if the return value of dma_fence_chain_find_seqno indicates success but it sets the fence to NULL, we will assign a stub fence to ensure the following code still signals the eventfd. v1 -> v2: assign a stub fence instead of signaling the eventfd Signed-off-by: Erik Kurzinger Fixes: c7a472297169 ("drm/syncobj: add IOCTL to register an eventfd") Signed-off-by: Simon Ser Link: https://patchwork.freedesktop.org/patch/msgid/20240221184527.37667-1-ekurzinger@nvidia.com --- drivers/gpu/drm/drm_syncobj.c | 13 ++++++++++++- 1 file changed, 12 insertions(+), 1 deletion(-) (limited to 'drivers/gpu/drm/drm_syncobj.c') diff --git a/drivers/gpu/drm/drm_syncobj.c b/drivers/gpu/drm/drm_syncobj.c index 56b0677acfa3..a6c19de46292 100644 --- a/drivers/gpu/drm/drm_syncobj.c +++ b/drivers/gpu/drm/drm_syncobj.c @@ -1418,10 +1418,21 @@ syncobj_eventfd_entry_func(struct drm_syncobj *syncobj, /* This happens inside the syncobj lock */ fence = dma_fence_get(rcu_dereference_protected(syncobj->fence, 1)); + if (!fence) + return; + ret = dma_fence_chain_find_seqno(&fence, entry->point); - if (ret != 0 || !fence) { + if (ret != 0) { + /* The given seqno has not been submitted yet. */ dma_fence_put(fence); return; + } else if (!fence) { + /* If dma_fence_chain_find_seqno returns 0 but sets the fence + * to NULL, it implies that the given seqno is signaled and a + * later seqno has already been submitted. Assign a stub fence + * so that the eventfd still gets signaled below. + */ + fence = dma_fence_get_stub(); } list_del_init(&entry->node); -- cgit v1.2.3-70-g09d2