Linaro-mm-sig

linaro-mm-sig@lists.linaro.org

10 participants
3232 discussions

Re: [PATCH 14/19] drm/amdgpu: independence for the amdkfd_fence! v4

by Christian König

On 12/11/25 16:08, Philipp Stanner wrote: > On Thu, 2025-12-11 at 13:16 +0100, Christian König wrote: >> This allows amdkfd_fences to outlive the amdgpu module. >> >> v2: implement Felix suggestion to lock the fence while signaling it. >> v3: fix typos >> v4: fix return code in signal_eviction_fence >> >> Signed-off-by: Christian König <christian.koenig(a)amd.com> >> --- >> drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h | 7 +++ >> .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_fence.c | 44 +++++++++---------- >> drivers/gpu/drm/amd/amdkfd/kfd_process.c | 2 +- >> drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 4 +- >> 4 files changed, 31 insertions(+), 26 deletions(-) >> >> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h >> index 8bdfcde2029b..2f2b277cfaed 100644 >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h >> @@ -196,6 +196,7 @@ int kfd_debugfs_kfd_mem_limits(struct seq_file *m, void *data); >> #endif >> #if IS_ENABLED(CONFIG_HSA_AMD) >> bool amdkfd_fence_check_mm(struct dma_fence *f, struct mm_struct *mm); >> +bool amdkfd_fence_signal(struct dma_fence *f); >> struct amdgpu_amdkfd_fence *to_amdgpu_amdkfd_fence(struct dma_fence *f); >> void amdgpu_amdkfd_remove_all_eviction_fences(struct amdgpu_bo *bo); >> int amdgpu_amdkfd_evict_userptr(struct mmu_interval_notifier *mni, >> @@ -210,6 +211,12 @@ bool amdkfd_fence_check_mm(struct dma_fence *f, struct mm_struct *mm) >> return false; >> } >> >> +static inline >> +bool amdkfd_fence_signal(struct dma_fence *f) >> +{ >> + return false; >> +} > > Huh? What's that? > > That function seems to be just a NOP. It's return code is used nowhere, > is it? It's the dummy which is used when CONFIG_HSA_AMD isn't enabled. Not sure if it's actually used or not, but we have dummies for all functions declared in this file. > >> + >> static inline >> struct amdgpu_amdkfd_fence *to_amdgpu_amdkfd_fence(struct dma_fence *f) >> { >> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_fence.c >> index 09c919f72b6c..9cd413e325f0 100644 >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_fence.c >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_fence.c >> @@ -127,29 +127,9 @@ static bool amdkfd_fence_enable_signaling(struct dma_fence *f) >> if (!svm_range_schedule_evict_svm_bo(fence)) >> return true; >> } >> - return false; >> -} >> - >> -/** >> - * amdkfd_fence_release - callback that fence can be freed >> - * >> - * @f: dma_fence >> - * >> - * This function is called when the reference count becomes zero. >> - * Drops the mm_struct reference and RCU schedules freeing up the fence. >> - */ >> -static void amdkfd_fence_release(struct dma_fence *f) >> -{ >> - struct amdgpu_amdkfd_fence *fence = to_amdgpu_amdkfd_fence(f); >> - >> - /* Unconditionally signal the fence. The process is getting >> - * terminated. >> - */ >> - if (WARN_ON(!fence)) >> - return; /* Not an amdgpu_amdkfd_fence */ >> - >> mmdrop(fence->mm); >> - kfree_rcu(f, rcu); >> + fence->mm = NULL; >> + return false; >> } >> >> /** >> @@ -174,9 +154,27 @@ bool amdkfd_fence_check_mm(struct dma_fence *f, struct mm_struct *mm) >> return false; >> } >> >> +bool amdkfd_fence_signal(struct dma_fence *f) >> +{ >> + struct amdgpu_amdkfd_fence *fence = to_amdgpu_amdkfd_fence(f); >> + unsigned long flags; >> + bool was_signaled; >> + >> + dma_fence_lock_irqsave(f, flags); >> + if (fence->mm) { >> + mmdrop(fence->mm); >> + fence->mm = NULL; >> + } >> + was_signaled = dma_fence_is_signaled_locked(f); >> + if (!was_signaled) >> + dma_fence_signal_locked(f); >> + dma_fence_unlock_irqrestore(f, flags); >> + >> + return was_signaled; >> +} >> + >> static const struct dma_fence_ops amdkfd_fence_ops = { >> .get_driver_name = amdkfd_fence_get_driver_name, >> .get_timeline_name = amdkfd_fence_get_timeline_name, >> .enable_signaling = amdkfd_fence_enable_signaling, >> - .release = amdkfd_fence_release, >> }; >> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c >> index bb252ec43733..2cf39e3d3fae 100644 >> --- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c >> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c >> @@ -1173,7 +1173,7 @@ static void kfd_process_wq_release(struct work_struct *work) >> synchronize_rcu(); >> ef = rcu_access_pointer(p->ef); >> if (ef) >> - dma_fence_signal(ef); >> + amdkfd_fence_signal(ef); >> >> kfd_process_remove_sysfs(p); >> kfd_debugfs_remove_process(p); >> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c >> index 97c2270f278f..0e94f3a976b1 100644 >> --- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c >> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c >> @@ -428,7 +428,7 @@ static void svm_range_bo_release(struct kref *kref) >> >> if (!dma_fence_is_signaled(&svm_bo->eviction_fence->base)) >> /* We're not in the eviction worker. Signal the fence. */ >> - dma_fence_signal(&svm_bo->eviction_fence->base); >> + amdkfd_fence_signal(&svm_bo->eviction_fence->base); >> dma_fence_put(&svm_bo->eviction_fence->base); >> amdgpu_bo_unref(&svm_bo->bo); >> kfree(svm_bo); >> @@ -3628,7 +3628,7 @@ static void svm_range_evict_svm_bo_worker(struct work_struct *work) >> mmap_read_unlock(mm); >> mmput(mm); >> >> - dma_fence_signal(&svm_bo->eviction_fence->base); >> + amdkfd_fence_signal(&svm_bo->eviction_fence->base); > > > And why do you do those changes and why doesn't the commit message > explain it? > > You stop signalling those fences, after all. Hui? I don't stop signaling the fences. I just delegate signaling into a separate helper function which does some extra cleanup before signaling the fence. Regards, Christian. > > > P. > >> >> /* This is the last reference to svm_bo, after svm_range_vram_node_free >> * has been called in svm_migrate_vram_to_ram >

2 days, 10 hours

Re: [PATCH 09/19] drm/sched: use inline locks for the drm-sched-fence

by Christian König

On 12/11/25 16:13, Tvrtko Ursulin wrote: > > On 11/12/2025 13:16, Christian König wrote: >> Using the inline lock is now the recommended way for dma_fence implementations. >> >> So use this approach for the scheduler fences as well just in case if >> anybody uses this as blueprint for its own implementation. >> >> Also saves about 4 bytes for the external spinlock. >> >> Signed-off-by: Christian König <christian.koenig(a)amd.com> >> --- >> drivers/gpu/drm/scheduler/sched_fence.c | 7 +++---- >> include/drm/gpu_scheduler.h | 4 ---- >> 2 files changed, 3 insertions(+), 8 deletions(-) >> >> diff --git a/drivers/gpu/drm/scheduler/sched_fence.c b/drivers/gpu/drm/scheduler/sched_fence.c >> index 08ccbde8b2f5..47471b9e43f9 100644 >> --- a/drivers/gpu/drm/scheduler/sched_fence.c >> +++ b/drivers/gpu/drm/scheduler/sched_fence.c >> @@ -161,7 +161,7 @@ static void drm_sched_fence_set_deadline_finished(struct dma_fence *f, >> /* If we already have an earlier deadline, keep it: */ >> if (test_bit(DRM_SCHED_FENCE_FLAG_HAS_DEADLINE_BIT, &f->flags) && >> ktime_before(fence->deadline, deadline)) { >> - spin_unlock_irqrestore(&fence->lock, flags); >> + dma_fence_unlock_irqrestore(f, flags); > > Rebase error I guess. Pull into the locking helpers patch. No that is actually completely intentional here. Previously we had a separate lock which protected both the DMA-fences as well as the deadline state. Now we turn that upside down by dropping the separate lock and protecting the deadline state with the dma_fence lock instead. Regards, Christian. > > Regards, > > Tvrtko > >> return; >> } >> @@ -217,7 +217,6 @@ struct drm_sched_fence *drm_sched_fence_alloc(struct drm_sched_entity *entity, >> fence->owner = owner; >> fence->drm_client_id = drm_client_id; >> - spin_lock_init(&fence->lock); >> return fence; >> } >> @@ -230,9 +229,9 @@ void drm_sched_fence_init(struct drm_sched_fence *fence, >> fence->sched = entity->rq->sched; >> seq = atomic_inc_return(&entity->fence_seq); >> dma_fence_init(&fence->scheduled, &drm_sched_fence_ops_scheduled, >> - &fence->lock, entity->fence_context, seq); >> + NULL, entity->fence_context, seq); >> dma_fence_init(&fence->finished, &drm_sched_fence_ops_finished, >> - &fence->lock, entity->fence_context + 1, seq); >> + NULL, entity->fence_context + 1, seq); >> } >> module_init(drm_sched_fence_slab_init); >> diff --git a/include/drm/gpu_scheduler.h b/include/drm/gpu_scheduler.h >> index fb88301b3c45..b77f24a783e3 100644 >> --- a/include/drm/gpu_scheduler.h >> +++ b/include/drm/gpu_scheduler.h >> @@ -297,10 +297,6 @@ struct drm_sched_fence { >> * belongs to. >> */ >> struct drm_gpu_scheduler *sched; >> - /** >> - * @lock: the lock used by the scheduled and the finished fences. >> - */ >> - spinlock_t lock; >> /** >> * @owner: job owner for debugging >> */ >

2 days, 11 hours

Re: [PATCH 11/19] drm/amdgpu: independence for the amdgpu_fence! v2

by Christian König

On 12/11/25 16:53, Tvrtko Ursulin wrote: > > On 11/12/2025 13:16, Christian König wrote: >> This allows amdgpu_fences to outlive the amdgpu module. >> >> v2: use dma_fence_get_rcu_safe to be NULL safe here. >> >> Signed-off-by: Christian König <christian.koenig(a)amd.com> >> --- >> drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c | 63 +++++++---------------- >> drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h | 1 - >> 2 files changed, 20 insertions(+), 44 deletions(-) >> >> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c >> index c7843e336310..c636347801c1 100644 >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c >> @@ -112,8 +112,7 @@ int amdgpu_fence_emit(struct amdgpu_ring *ring, struct amdgpu_fence *af, >> af->ring = ring; >> seq = ++ring->fence_drv.sync_seq; >> - dma_fence_init(fence, &amdgpu_fence_ops, >> - &ring->fence_drv.lock, >> + dma_fence_init(fence, &amdgpu_fence_ops, NULL, >> adev->fence_context + ring->idx, seq); >> amdgpu_ring_emit_fence(ring, ring->fence_drv.gpu_addr, >> @@ -467,7 +466,6 @@ int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring) >> timer_setup(&ring->fence_drv.fallback_timer, amdgpu_fence_fallback, 0); >> ring->fence_drv.num_fences_mask = ring->num_hw_submission * 2 - 1; >> - spin_lock_init(&ring->fence_drv.lock); >> ring->fence_drv.fences = kcalloc(ring->num_hw_submission * 2, sizeof(void *), >> GFP_KERNEL); >> @@ -654,16 +652,20 @@ void amdgpu_fence_driver_set_error(struct amdgpu_ring *ring, int error) >> struct amdgpu_fence_driver *drv = &ring->fence_drv; >> unsigned long flags; >> - spin_lock_irqsave(&drv->lock, flags); >> + rcu_read_lock(); >> for (unsigned int i = 0; i <= drv->num_fences_mask; ++i) { >> struct dma_fence *fence; >> - fence = rcu_dereference_protected(drv->fences[i], >> - lockdep_is_held(&drv->lock)); >> - if (fence && !dma_fence_is_signaled_locked(fence)) >> + fence = dma_fence_get_rcu(drv->fences[i]); > > dma_fence_get_rcu is not safe against passing a NULL fence in, while the existing code makes it look like drv->fence[] slot can contain NULL at this point? > > amdgpu_fence_process() is the place which can NULL the slots? Irq context? Why is that safe with no reference held from clearing the slot to operating on the fence? The slots are never cleared. It can only be that they are NULL when the driver is loaded. I've switched over to dma_fence_get_rcu_safe() where appropriated. Regards, Christian. > >> + if (!fence) >> + continue; >> + >> + dma_fence_lock_irqsave(fence, flags); >> + if (!dma_fence_is_signaled_locked(fence)) >> dma_fence_set_error(fence, error); >> + dma_fence_unlock_irqrestore(fence, flags); >> } >> - spin_unlock_irqrestore(&drv->lock, flags); >> + rcu_read_unlock(); >> } >> /** >> @@ -714,16 +716,19 @@ void amdgpu_fence_driver_guilty_force_completion(struct amdgpu_fence *af) >> seq = ring->fence_drv.sync_seq & ring->fence_drv.num_fences_mask; >> /* mark all fences from the guilty context with an error */ >> - spin_lock_irqsave(&ring->fence_drv.lock, flags); >> + rcu_read_lock(); >> do { >> last_seq++; >> last_seq &= ring->fence_drv.num_fences_mask; >> ptr = &ring->fence_drv.fences[last_seq]; >> - rcu_read_lock(); >> - unprocessed = rcu_dereference(*ptr); >> + unprocessed = dma_fence_get_rcu_safe(ptr); > > Similar concern like the above. > > Regards, > > Tvrtko >> + >> + if (!unprocessed) >> + continue; >> - if (unprocessed && !dma_fence_is_signaled_locked(unprocessed)) { >> + dma_fence_lock_irqsave(unprocessed, flags); >> + if (dma_fence_is_signaled_locked(unprocessed)) { >> fence = container_of(unprocessed, struct amdgpu_fence, base); >> if (fence == af) >> @@ -731,9 +736,10 @@ void amdgpu_fence_driver_guilty_force_completion(struct amdgpu_fence *af) >> else if (fence->context == af->context) >> dma_fence_set_error(&fence->base, -ECANCELED); >> } >> - rcu_read_unlock(); >> + dma_fence_unlock_irqrestore(unprocessed, flags); >> + dma_fence_put(unprocessed); >> } while (last_seq != seq); >> - spin_unlock_irqrestore(&ring->fence_drv.lock, flags); >> + rcu_read_unlock(); >> /* signal the guilty fence */ >> amdgpu_fence_write(ring, (u32)af->base.seqno); >> amdgpu_fence_process(ring); >> @@ -823,39 +829,10 @@ static bool amdgpu_fence_enable_signaling(struct dma_fence *f) >> return true; >> } >> -/** >> - * amdgpu_fence_free - free up the fence memory >> - * >> - * @rcu: RCU callback head >> - * >> - * Free up the fence memory after the RCU grace period. >> - */ >> -static void amdgpu_fence_free(struct rcu_head *rcu) >> -{ >> - struct dma_fence *f = container_of(rcu, struct dma_fence, rcu); >> - >> - /* free fence_slab if it's separated fence*/ >> - kfree(to_amdgpu_fence(f)); >> -} >> - >> -/** >> - * amdgpu_fence_release - callback that fence can be freed >> - * >> - * @f: fence >> - * >> - * This function is called when the reference count becomes zero. >> - * It just RCU schedules freeing up the fence. >> - */ >> -static void amdgpu_fence_release(struct dma_fence *f) >> -{ >> - call_rcu(&f->rcu, amdgpu_fence_free); >> -} >> - >> static const struct dma_fence_ops amdgpu_fence_ops = { >> .get_driver_name = amdgpu_fence_get_driver_name, >> .get_timeline_name = amdgpu_fence_get_timeline_name, >> .enable_signaling = amdgpu_fence_enable_signaling, >> - .release = amdgpu_fence_release, >> }; >> /* >> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h >> index 7a27c6c4bb44..9cbf63454004 100644 >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h >> @@ -125,7 +125,6 @@ struct amdgpu_fence_driver { >> unsigned irq_type; >> struct timer_list fallback_timer; >> unsigned num_fences_mask; >> - spinlock_t lock; >> struct dma_fence **fences; >> }; >> >

2 days, 11 hours

Re: [PATCH 15/19] drm/amdgpu: independence for the amdgpu_userq_fence!

by Christian König

On 12/11/25 17:12, Tvrtko Ursulin wrote: > > On 11/12/2025 13:16, Christian König wrote: >> This allows amdgpu_userq_fences to outlive the amdgpu module. >> >> Signed-off-by: Christian König <christian.koenig(a)amd.com> >> --- >> drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 13 +---- >> .../gpu/drm/amd/amdgpu/amdgpu_userq_fence.c | 54 ++++--------------- >> .../gpu/drm/amd/amdgpu/amdgpu_userq_fence.h | 8 --- >> 3 files changed, 11 insertions(+), 64 deletions(-) >> >> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c >> index 2dfbddcef9ab..f206297aae8b 100644 >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c >> @@ -3155,11 +3155,7 @@ static int __init amdgpu_init(void) >> r = amdgpu_sync_init(); >> if (r) >> - goto error_sync; >> - >> - r = amdgpu_userq_fence_slab_init(); >> - if (r) >> - goto error_fence; >> + return r; >> DRM_INFO("amdgpu kernel modesetting enabled.\n"); >> amdgpu_register_atpx_handler(); >> @@ -3176,12 +3172,6 @@ static int __init amdgpu_init(void) >> /* let modprobe override vga console setting */ >> return pci_register_driver(&amdgpu_kms_pci_driver); >> - >> -error_fence: >> - amdgpu_sync_fini(); >> - >> -error_sync: >> - return r; >> } >> static void __exit amdgpu_exit(void) >> @@ -3191,7 +3181,6 @@ static void __exit amdgpu_exit(void) >> amdgpu_unregister_atpx_handler(); >> amdgpu_acpi_release(); >> amdgpu_sync_fini(); >> - amdgpu_userq_fence_slab_fini(); >> mmu_notifier_synchronize(); >> amdgpu_xcp_drv_release(); >> } >> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_userq_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_userq_fence.c >> index eba9fb359047..bb19f72770b0 100644 >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_userq_fence.c >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_userq_fence.c >> @@ -33,26 +33,6 @@ >> #include "amdgpu_userq_fence.h" >> static const struct dma_fence_ops amdgpu_userq_fence_ops; >> -static struct kmem_cache *amdgpu_userq_fence_slab; >> - >> -int amdgpu_userq_fence_slab_init(void) >> -{ >> - amdgpu_userq_fence_slab = kmem_cache_create("amdgpu_userq_fence", >> - sizeof(struct amdgpu_userq_fence), >> - 0, >> - SLAB_HWCACHE_ALIGN, >> - NULL); >> - if (!amdgpu_userq_fence_slab) >> - return -ENOMEM; >> - >> - return 0; >> -} >> - >> -void amdgpu_userq_fence_slab_fini(void) >> -{ >> - rcu_barrier(); > > What was this rcu_barrier() for? Cargo culted or more to it? All dma_fences are RCU protected. When they are backed by a kmem_cache you need to make sure to wait for an RCU grace period to pass before destroying the kmem_cache. Since the dma_fence framework now uses kfree_rcu that shouldn't be problematic any more. >> - kmem_cache_destroy(amdgpu_userq_fence_slab); >> -} >> static inline struct amdgpu_userq_fence *to_amdgpu_userq_fence(struct dma_fence *f) >> { >> @@ -227,7 +207,7 @@ void amdgpu_userq_fence_driver_put(struct amdgpu_userq_fence_driver *fence_drv) >> static int amdgpu_userq_fence_alloc(struct amdgpu_userq_fence **userq_fence) >> { >> - *userq_fence = kmem_cache_alloc(amdgpu_userq_fence_slab, GFP_ATOMIC); >> + *userq_fence = kmalloc(sizeof(**userq_fence), GFP_ATOMIC); > This GFP_ATOMIC is suboptimal for sure being on the ioctl path. It is outside of the scope for this patch, but once my userq cleanup patches get reviewed next on my list was to try and understand this. >> return *userq_fence ? 0 : -ENOMEM; >> } >> @@ -243,12 +223,11 @@ static int amdgpu_userq_fence_create(struct amdgpu_usermode_queue *userq, >> if (!fence_drv) >> return -EINVAL; >> - spin_lock_init(&userq_fence->lock); >> INIT_LIST_HEAD(&userq_fence->link); >> fence = &userq_fence->base; >> userq_fence->fence_drv = fence_drv; >> - dma_fence_init64(fence, &amdgpu_userq_fence_ops, &userq_fence->lock, >> + dma_fence_init64(fence, &amdgpu_userq_fence_ops, NULL, >> fence_drv->context, seq); >> amdgpu_userq_fence_driver_get(fence_drv); >> @@ -318,35 +297,22 @@ static bool amdgpu_userq_fence_signaled(struct dma_fence *f) >> rptr = amdgpu_userq_fence_read(fence_drv); >> wptr = fence->base.seqno; >> - if (rptr >= wptr) >> + if (rptr >= wptr) { >> + amdgpu_userq_fence_driver_put(fence->fence_drv); > > fence_drv is in a local already. > >> + fence->fence_drv = NULL; > > amdgpu_userq_fence_get_timeline_name could now oops somehow? > >> + >> + kvfree(fence->fence_drv_array); >> + fence->fence_drv_array = NULL; > > Not sure if this is safe either. amdgpu_userq_fence_driver_process() drops its reference before it unlinks the fence from the list. Can someone external trigger the fence_is_signaled check, before the interrupt processing kicks in, which will clear fence_drv_array, and so amdgpu_userq_fence_driver_process() would oops? Oh, good question. I need to double check that. Thanks, Christian. > > Regards, > > Tvrtko > >> return true; >> + } >> return false; >> } >> -static void amdgpu_userq_fence_free(struct rcu_head *rcu) >> -{ >> - struct dma_fence *fence = container_of(rcu, struct dma_fence, rcu); >> - struct amdgpu_userq_fence *userq_fence = to_amdgpu_userq_fence(fence); >> - struct amdgpu_userq_fence_driver *fence_drv = userq_fence->fence_drv; >> - >> - /* Release the fence driver reference */ >> - amdgpu_userq_fence_driver_put(fence_drv); >> - >> - kvfree(userq_fence->fence_drv_array); >> - kmem_cache_free(amdgpu_userq_fence_slab, userq_fence); >> -} >> - >> -static void amdgpu_userq_fence_release(struct dma_fence *f) >> -{ >> - call_rcu(&f->rcu, amdgpu_userq_fence_free); >> -} >> - >> static const struct dma_fence_ops amdgpu_userq_fence_ops = { >> .get_driver_name = amdgpu_userq_fence_get_driver_name, >> .get_timeline_name = amdgpu_userq_fence_get_timeline_name, >> .signaled = amdgpu_userq_fence_signaled, >> - .release = amdgpu_userq_fence_release, >> }; >> /** >> @@ -560,7 +526,7 @@ int amdgpu_userq_signal_ioctl(struct drm_device *dev, void *data, >> r = amdgpu_userq_fence_create(queue, userq_fence, wptr, &fence); >> if (r) { >> mutex_unlock(&userq_mgr->userq_mutex); >> - kmem_cache_free(amdgpu_userq_fence_slab, userq_fence); >> + kfree(userq_fence); >> goto put_gobj_write; >> } >> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_userq_fence.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_userq_fence.h >> index d76add2afc77..6f04782f3ea9 100644 >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_userq_fence.h >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_userq_fence.h >> @@ -31,11 +31,6 @@ >> struct amdgpu_userq_fence { >> struct dma_fence base; >> - /* >> - * This lock is necessary to synchronize the >> - * userqueue dma fence operations. >> - */ >> - spinlock_t lock; >> struct list_head link; >> unsigned long fence_drv_array_count; >> struct amdgpu_userq_fence_driver *fence_drv; >> @@ -58,9 +53,6 @@ struct amdgpu_userq_fence_driver { >> char timeline_name[TASK_COMM_LEN]; >> }; >> -int amdgpu_userq_fence_slab_init(void); >> -void amdgpu_userq_fence_slab_fini(void); >> - >> void amdgpu_userq_fence_driver_get(struct amdgpu_userq_fence_driver *fence_drv); >> void amdgpu_userq_fence_driver_put(struct amdgpu_userq_fence_driver *fence_drv); >> int amdgpu_userq_fence_driver_alloc(struct amdgpu_device *adev, >

2 days, 11 hours

Re: [PATCH v2] drm: Fix object leak in DRM_IOCTL_GEM_CHANGE_HANDLE

by Christian König

On 12/12/25 14:20, Karol Wachowski wrote: > Add missing drm_gem_object_put() call when drm_gem_object_lookup() > successfully returns an object. This fixes a GEM object reference > leak that can prevent driver modules from unloading when using > prime buffers. > > Fixes: 53096728b891 ("drm: Add DRM prime interface to reassign GEM handle") > Signed-off-by: Karol Wachowski <karol.wachowski(a)linux.intel.com> > --- > Changes between v1 and v2: > - move setting ret value under if branch as suggested in review > - add Cc: stable 6.18+ Oh don't CC the stable list on the review mail directly, just add "CC: stable(a)vger.kernel.org # 6.18+" to the tags. Greg is going to complain about that :( With that done Reviewed-by: Christian König <christian.koenig(a)amd.com> and please push to drm-misc-fixes. If you don't have commit rights for drm-misc-fixes please ping me and I'm going to push that. Thanks, Christian. > --- > drivers/gpu/drm/drm_gem.c | 8 ++++++-- > 1 file changed, 6 insertions(+), 2 deletions(-) > > diff --git a/drivers/gpu/drm/drm_gem.c b/drivers/gpu/drm/drm_gem.c > index ca1956608261..bcc08a6aebf8 100644 > --- a/drivers/gpu/drm/drm_gem.c > +++ b/drivers/gpu/drm/drm_gem.c > @@ -1010,8 +1010,10 @@ int drm_gem_change_handle_ioctl(struct drm_device *dev, void *data, > if (!obj) > return -ENOENT; > > - if (args->handle == args->new_handle) > - return 0; > + if (args->handle == args->new_handle) { > + ret = 0; > + goto out; > + } > > mutex_lock(&file_priv->prime.lock); > > @@ -1043,6 +1045,8 @@ int drm_gem_change_handle_ioctl(struct drm_device *dev, void *data, > > out_unlock: > mutex_unlock(&file_priv->prime.lock); > +out: > + drm_gem_object_put(obj); > > return ret; > }

2 days, 13 hours

Re: [PATCH] drm: Fix object leak in DRM_IOCTL_GEM_CHANGE_HANDLE

by Christian König

On 12/12/25 14:02, Karol Wachowski wrote: > Add missing drm_gem_object_put() call when drm_gem_object_lookup() > successfully returns an object. This fixes a GEM object reference > leak that can prevent driver modules from unloading when using > prime buffers. Good catch. > Fixes: 53096728b891 ("drm: Add DRM prime interface to reassign GEM handle") > Signed-off-by: Karol Wachowski <karol.wachowski(a)linux.intel.com> CC: stable 6.18? > --- > drivers/gpu/drm/drm_gem.c | 6 ++++-- > 1 file changed, 4 insertions(+), 2 deletions(-) > > diff --git a/drivers/gpu/drm/drm_gem.c b/drivers/gpu/drm/drm_gem.c > index ca1956608261..e150bc1ce65a 100644 > --- a/drivers/gpu/drm/drm_gem.c > +++ b/drivers/gpu/drm/drm_gem.c > @@ -1001,7 +1001,7 @@ int drm_gem_change_handle_ioctl(struct drm_device *dev, void *data, > { > struct drm_gem_change_handle *args = data; > struct drm_gem_object *obj; > - int ret; > + int ret = 0; Please set ret explicitly in the if branch below. Always initializing return values is usually considered bad coding style. Apart from that looks good to me. Thanks, Christian. > > if (!drm_core_check_feature(dev, DRIVER_GEM)) > return -EOPNOTSUPP; > @@ -1011,7 +1011,7 @@ int drm_gem_change_handle_ioctl(struct drm_device *dev, void *data, > return -ENOENT; > > if (args->handle == args->new_handle) > - return 0; > + goto out; > > mutex_lock(&file_priv->prime.lock); > > @@ -1043,6 +1043,8 @@ int drm_gem_change_handle_ioctl(struct drm_device *dev, void *data, > > out_unlock: > mutex_unlock(&file_priv->prime.lock); > +out: > + drm_gem_object_put(obj); > > return ret; > }

2 days, 13 hours

Re: [PATCH 03/19] dma-buf: inline spinlock for fence protection v3

by Christian König

On 12/11/25 15:35, Tvrtko Ursulin wrote: > > Hi, > > On 11/12/2025 13:16, Christian König wrote: >> Implement per-fence spinlocks, allowing implementations to not give an >> external spinlock to protect the fence internal statei. Instead a spinlock >> embedded into the fence structure itself is used in this case. >> >> Shared spinlocks have the problem that implementations need to guarantee >> that the lock live at least as long all fences referencing them. >> >> Using a per-fence spinlock allows completely decoupling spinlock producer >> and consumer life times, simplifying the handling in most use cases. >> >> v2: improve naming, coverage and function documentation >> v3: fix one additional locking in the selftests >> >> Signed-off-by: Christian König <christian.koenig(a)amd.com> >> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin(a)igalia.com> > > I don't think I gave r-b on this one. Not just yet at least. Maybe you have missed the comments I had in the previous two rounds? I will repeat them below. I was already wondering why you gave comments and an rb but though that the comments might just be optional. Going to remove that and see on the comments below. >> @@ -365,7 +364,7 @@ void dma_fence_signal_timestamp_locked(struct dma_fence *fence, >> struct dma_fence_cb *cur, *tmp; >> struct list_head cb_list; >> - lockdep_assert_held(fence->lock); >> + lockdep_assert_held(dma_fence_spinlock(fence)); >> if (unlikely(test_and_set_bit(DMA_FENCE_FLAG_SIGNALED_BIT, >> &fence->flags))) >> @@ -412,9 +411,9 @@ void dma_fence_signal_timestamp(struct dma_fence *fence, ktime_t timestamp) >> if (WARN_ON(!fence)) >> return; >> - spin_lock_irqsave(fence->lock, flags); >> + dma_fence_lock_irqsave(fence, flags); > > For the locking wrappers I think it would be better to introduce them in a purely mechanical patch preceding this one. That is, just add the wrappers and nothing else. That doesn't fully work for all cases, but I will separate it out a bit more. >> static inline uint64_t amdgpu_vm_tlb_seq(struct amdgpu_vm *vm) >> { >> + struct dma_fence *fence; >> unsigned long flags; >> - spinlock_t *lock; >> /* >> * Workaround to stop racing between the fence signaling and handling >> - * the cb. The lock is static after initially setting it up, just make >> - * sure that the dma_fence structure isn't freed up. >> + * the cb. >> */ >> rcu_read_lock(); >> - lock = vm->last_tlb_flush->lock; >> + fence = dma_fence_get_rcu(vm->last_tlb_flush); > > Why does this belong here? If taking a reference fixes some race it needs to be a separate patch. If it doesn't then this patch shouldn't be adding it. The code previously assumed that the lock is global and can't go away while the function is called. When we start to use an inline lock that assumption is not true any more. But you're right that can be a separate patch. >> @@ -362,6 +368,38 @@ dma_fence_get_rcu_safe(struct dma_fence __rcu **fencep) >> } while (1); >> } >> +/** >> + * dma_fence_spinlock - return pointer to the spinlock protecting the fence >> + * @fence: the fence to get the lock from >> + * >> + * Return either the pointer to the embedded or the external spin lock. >> + */ >> +static inline spinlock_t *dma_fence_spinlock(struct dma_fence *fence) >> +{ >> + return test_bit(DMA_FENCE_FLAG_INLINE_LOCK_BIT, &fence->flags) ? >> + &fence->inline_lock : fence->extern_lock; > > Is sprinkling of conditionals better than growing the struct? Probably yes, since branch misses are cheaper than cache misses. Unless the code grows significantly on some hot path and we get instruction cache misses instead. Who knows. But let say in the commit message we considered it and decided on this solution due xyz. Sure. > > On a quick grep there is one arch where this grows the struct past a cache line anyway, but as it is PA-RISC I guess no one cares. Lets mention that in the commit message as well. Interesting, I was aware of the problems on Sparc regarding spinlocks but that PA-RISC also has something more complicated then an int is news to me. Anyway I agree it doesn't really matter. Regards, Christian. > > Regards, > > Tvrtko >> +} >> + >> +/** >> + * dma_fence_lock_irqsave - irqsave lock the fence >> + * @fence: the fence to lock >> + * @flags: where to store the CPU flags. >> + * >> + * Lock the fence, preventing it from changing to the signaled state. >> + */ >> +#define dma_fence_lock_irqsave(fence, flags) \ >> + spin_lock_irqsave(dma_fence_spinlock(fence), flags) >> + >> +/** >> + * dma_fence_unlock_irqrestore - unlock the fence and irqrestore >> + * @fence: the fence to unlock >> + * @flags the CPU flags to restore >> + * >> + * Unlock the fence, allowing it to change it's state to signaled again. >> + */ >> +#define dma_fence_unlock_irqrestore(fence, flags) \ >> + spin_unlock_irqrestore(dma_fence_spinlock(fence), flags) >> + >> #ifdef CONFIG_LOCKDEP >> bool dma_fence_begin_signalling(void); >> void dma_fence_end_signalling(bool cookie); >

2 days, 17 hours

Re: Independence for dma_fences! v4

by Christian König

On 12/11/25 13:33, Philipp Stanner wrote: > On Thu, 2025-12-11 at 13:16 +0100, Christian König wrote: >> Hi everyone, >> >> dma_fences have ever lived under the tyranny dictated by the module >> lifetime of their issuer, leading to crashes should anybody still holding >> a reference to a dma_fence when the module of the issuer was unloaded. >> >> The basic problem is that when buffer are shared between drivers >> dma_fence objects can leak into external drivers and stay there even >> after they are signaled. The dma_resv object for example only lazy releases >> dma_fences. >> >> So what happens is that when the module who originally created the dma_fence >> unloads the dma_fence_ops function table becomes unavailable as well and so >> any attempt to release the fence crashes the system. >> >> Previously various approaches have been discussed, including changing the >> locking semantics of the dma_fence callbacks (by me) as well as using the >> drm scheduler as intermediate layer (by Sima) to disconnect dma_fences >> from their actual users, but none of them are actually solving all problems. >> >> Tvrtko did some really nice prerequisite work by protecting the returned >> strings of the dma_fence_ops by RCU. This way dma_fence creators where >> able to just wait for an RCU grace period after fence signaling before >> they could be save to free those data structures. >> >> Now this patch set here goes a step further and protects the whole >> dma_fence_ops structure by RCU, so that after the fence signals the >> pointer to the dma_fence_ops is set to NULL when there is no wait nor >> release callback given. All functionality which use the dma_fence_ops >> reference are put inside an RCU critical section, except for the >> deprecated issuer specific wait and of course the optional release >> callback. >> >> Additional to the RCU changes the lock protecting the dma_fence state >> previously had to be allocated external. This set here now changes the >> functionality to make that external lock optional and allows dma_fences >> to use an inline lock and be self contained. >> >> v4: >> >> Rebases the whole set on upstream changes, especially the cleanup >> from Philip in patch "drm/amdgpu: independence for the amdkfd_fence!". >> >> Adding two patches which brings the DMA-fence self tests up to date. >> The first selftest changes removes the mock_wait and so actually starts >> testing the default behavior instead of some hacky implementation in the >> test. This one should probably go upstream independent of this set. >> The second drops the mock_fence as well and tests the new RCU and inline >> spinlock functionality. >> >> Especially the first patch still needs a Reviewed-by, apart from that I >> think I've addressed all review comments. >> >> The plan is to push the core DMA-buf changes to drm-misc-next and then the >> driver specific changes through the driver channels as approprite. > > This does not apply to drm-misc-next (unless I'm screwing up badly). > > Where can I apply it? I'd like to test the drm_sched changes before > this gets merged. drm-tip from a few days ago, otherwise the xe changes won't work. Regards, Christian. > > P. > >> >> Please review and comment, >> Christian. >> >> >

2 days, 17 hours

Re: [PATCH] dma-buf: system_heap: account for system heap allocation in memcg

by T.J. Mercier

On Fri, Dec 12, 2025 at 4:31 AM Eric Chanudet <echanude(a)redhat.com> wrote: > > The system dma-buf heap lets userspace allocate buffers from the page > allocator. However, these allocations are not accounted for in memcg, > allowing processes to escape limits that may be configured. > > Pass the __GFP_ACCOUNT for our allocations to account them into memcg. We had a discussion just last night in the MM track at LPC about how shared memory accounted in memcg is pretty broken. Without a way to identify (and possibly transfer) ownership of a shared buffer, this makes the accounting of shared memory, and zombie memcg problems worse. :\ > > Userspace components using the system heap can be constrained with, e.g: > systemd-run --user --scope -p MemoryMax=10M ... > > Signed-off-by: Eric Chanudet <echanude(a)redhat.com> > --- > drivers/dma-buf/heaps/system_heap.c | 4 ++-- > 1 file changed, 2 insertions(+), 2 deletions(-) > > diff --git a/drivers/dma-buf/heaps/system_heap.c b/drivers/dma-buf/heaps/system_heap.c > index 4c782fe33fd4..c91fcdff4b77 100644 > --- a/drivers/dma-buf/heaps/system_heap.c > +++ b/drivers/dma-buf/heaps/system_heap.c > @@ -38,10 +38,10 @@ struct dma_heap_attachment { > bool mapped; > }; > > -#define LOW_ORDER_GFP (GFP_HIGHUSER | __GFP_ZERO) > +#define LOW_ORDER_GFP (GFP_HIGHUSER | __GFP_ZERO | __GFP_ACCOUNT) > #define HIGH_ORDER_GFP (((GFP_HIGHUSER | __GFP_ZERO | __GFP_NOWARN \ > | __GFP_NORETRY) & ~__GFP_RECLAIM) \ > - | __GFP_COMP) > + | __GFP_COMP | __GFP_ACCOUNT) > static gfp_t order_flags[] = {HIGH_ORDER_GFP, HIGH_ORDER_GFP, LOW_ORDER_GFP}; > /* > * The selection of the orders used for allocation (1MB, 64K, 4K) is designed > -- > 2.52.0 >

3 days, 3 hours

Independence for dma_fences! v4

by Christian König

Hi everyone, dma_fences have ever lived under the tyranny dictated by the module lifetime of their issuer, leading to crashes should anybody still holding a reference to a dma_fence when the module of the issuer was unloaded. The basic problem is that when buffer are shared between drivers dma_fence objects can leak into external drivers and stay there even after they are signaled. The dma_resv object for example only lazy releases dma_fences. So what happens is that when the module who originally created the dma_fence unloads the dma_fence_ops function table becomes unavailable as well and so any attempt to release the fence crashes the system. Previously various approaches have been discussed, including changing the locking semantics of the dma_fence callbacks (by me) as well as using the drm scheduler as intermediate layer (by Sima) to disconnect dma_fences from their actual users, but none of them are actually solving all problems. Tvrtko did some really nice prerequisite work by protecting the returned strings of the dma_fence_ops by RCU. This way dma_fence creators where able to just wait for an RCU grace period after fence signaling before they could be save to free those data structures. Now this patch set here goes a step further and protects the whole dma_fence_ops structure by RCU, so that after the fence signals the pointer to the dma_fence_ops is set to NULL when there is no wait nor release callback given. All functionality which use the dma_fence_ops reference are put inside an RCU critical section, except for the deprecated issuer specific wait and of course the optional release callback. Additional to the RCU changes the lock protecting the dma_fence state previously had to be allocated external. This set here now changes the functionality to make that external lock optional and allows dma_fences to use an inline lock and be self contained. v4: Rebases the whole set on upstream changes, especially the cleanup from Philip in patch "drm/amdgpu: independence for the amdkfd_fence!". Adding two patches which brings the DMA-fence self tests up to date. The first selftest changes removes the mock_wait and so actually starts testing the default behavior instead of some hacky implementation in the test. This one should probably go upstream independent of this set. The second drops the mock_fence as well and tests the new RCU and inline spinlock functionality. Especially the first patch still needs a Reviewed-by, apart from that I think I've addressed all review comments. The plan is to push the core DMA-buf changes to drm-misc-next and then the driver specific changes through the driver channels as approprite. Please review and comment, Christian.

3 days, 14 hours

Jump to page:

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

Linaro-mm-sig