From: Christian König ckoenig.leichtzumerken@gmail.com

Lockless container implementation similar to a dma_fence_array, but with only two elements per node and automatic garbage collection.

v2: properly document dma_fence_chain_for_each, add dma_fence_chain_find_seqno, drop prev reference during garbage collection if it's not a chain fence. v3: use head and iterator for dma_fence_chain_for_each v4: fix reference count in dma_fence_chain_enable_signaling v5: fix iteration when walking each chain node v6: add __rcu for member 'prev' of struct chain node v7: fix rcu warnings from kernel robot

Signed-off-by: Christian König christian.koenig@amd.com Cc: Lionel Landwerlin lionel.g.landwerlin@intel.com Reviewed-by: Lionel Landwerlin lionel.g.landwerlin@intel.com --- drivers/dma-buf/Makefile | 3 +- drivers/dma-buf/dma-fence-chain.c | 241 ++++++++++++++++++++++++++++++ include/linux/dma-fence-chain.h | 81 ++++++++++ 3 files changed, 324 insertions(+), 1 deletion(-) create mode 100644 drivers/dma-buf/dma-fence-chain.c create mode 100644 include/linux/dma-fence-chain.h

diff --git a/drivers/dma-buf/Makefile b/drivers/dma-buf/Makefile index 0913a6ccab5a..1f006e083eb9 100644 --- a/drivers/dma-buf/Makefile +++ b/drivers/dma-buf/Makefile @@ -1,4 +1,5 @@ -obj-y := dma-buf.o dma-fence.o dma-fence-array.o reservation.o seqno-fence.o +obj-y := dma-buf.o dma-fence.o dma-fence-array.o dma-fence-chain.o \ + reservation.o seqno-fence.o obj-$(CONFIG_SYNC_FILE) += sync_file.o obj-$(CONFIG_SW_SYNC) += sw_sync.o sync_debug.o obj-$(CONFIG_UDMABUF) += udmabuf.o diff --git a/drivers/dma-buf/dma-fence-chain.c b/drivers/dma-buf/dma-fence-chain.c new file mode 100644 index 000000000000..c729f98a7bd3 --- /dev/null +++ b/drivers/dma-buf/dma-fence-chain.c @@ -0,0 +1,241 @@ +/* + * fence-chain: chain fences together in a timeline + * + * Copyright (C) 2018 Advanced Micro Devices, Inc. + * Authors: + * Christian König christian.koenig@amd.com + * + * This program is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 as published by + * the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + */ + +#include <linux/dma-fence-chain.h> + +static bool dma_fence_chain_enable_signaling(struct dma_fence *fence); + +/** + * dma_fence_chain_get_prev - use RCU to get a reference to the previous fence + * @chain: chain node to get the previous node from + * + * Use dma_fence_get_rcu_safe to get a reference to the previous fence of the + * chain node. + */ +static struct dma_fence *dma_fence_chain_get_prev(struct dma_fence_chain *chain) +{ + struct dma_fence *prev; + + rcu_read_lock(); + prev = dma_fence_get_rcu_safe(&chain->prev); + rcu_read_unlock(); + return prev; +} + +/** + * dma_fence_chain_walk - chain walking function + * @fence: current chain node + * + * Walk the chain to the next node. Returns the next fence or NULL if we are at + * the end of the chain. Garbage collects chain nodes which are already + * signaled. + */ +struct dma_fence *dma_fence_chain_walk(struct dma_fence *fence) +{ + struct dma_fence_chain *chain, *prev_chain; + struct dma_fence *prev, *replacement, *tmp; + + chain = to_dma_fence_chain(fence); + if (!chain) { + dma_fence_put(fence); + return NULL; + } + + while ((prev = dma_fence_chain_get_prev(chain))) { + + prev_chain = to_dma_fence_chain(prev); + if (prev_chain) { + if (!dma_fence_is_signaled(prev_chain->fence)) + break; + + replacement = dma_fence_chain_get_prev(prev_chain); + } else { + if (!dma_fence_is_signaled(prev)) + break; + + replacement = NULL; + } + + tmp = cmpxchg((void **)&chain->prev, (void *)prev, (void *)replacement); + if (tmp == prev) + dma_fence_put(tmp); + else + dma_fence_put(replacement); + dma_fence_put(prev); + } + + dma_fence_put(fence); + return prev; +} +EXPORT_SYMBOL(dma_fence_chain_walk); + +/** + * dma_fence_chain_find_seqno - find fence chain node by seqno + * @pfence: pointer to the chain node where to start + * @seqno: the sequence number to search for + * + * Advance the fence pointer to the chain node which will signal this sequence + * number. If no sequence number is provided then this is a no-op. + * + * Returns EINVAL if the fence is not a chain node or the sequence number has + * not yet advanced far enough. + */ +int dma_fence_chain_find_seqno(struct dma_fence **pfence, uint64_t seqno) +{ + struct dma_fence_chain *chain; + + if (!seqno) + return 0; + + chain = to_dma_fence_chain(*pfence); + if (!chain || chain->base.seqno < seqno) + return -EINVAL; + + dma_fence_chain_for_each(*pfence, &chain->base) { + if ((*pfence)->context != chain->base.context || + to_dma_fence_chain(*pfence)->prev_seqno < seqno) + break; + } + dma_fence_put(&chain->base); + + return 0; +} +EXPORT_SYMBOL(dma_fence_chain_find_seqno); + +static const char *dma_fence_chain_get_driver_name(struct dma_fence *fence) +{ + return "dma_fence_chain"; +} + +static const char *dma_fence_chain_get_timeline_name(struct dma_fence *fence) +{ + return "unbound"; +} + +static void dma_fence_chain_irq_work(struct irq_work *work) +{ + struct dma_fence_chain *chain; + + chain = container_of(work, typeof(*chain), work); + + /* Try to rearm the callback */ + if (!dma_fence_chain_enable_signaling(&chain->base)) + /* Ok, we are done. No more unsignaled fences left */ + dma_fence_signal(&chain->base); + dma_fence_put(&chain->base); +} + +static void dma_fence_chain_cb(struct dma_fence *f, struct dma_fence_cb *cb) +{ + struct dma_fence_chain *chain; + + chain = container_of(cb, typeof(*chain), cb); + irq_work_queue(&chain->work); + dma_fence_put(f); +} + +static bool dma_fence_chain_enable_signaling(struct dma_fence *fence) +{ + struct dma_fence_chain *head = to_dma_fence_chain(fence); + + dma_fence_get(&head->base); + dma_fence_chain_for_each(fence, &head->base) { + struct dma_fence_chain *chain = to_dma_fence_chain(fence); + struct dma_fence *f = chain ? chain->fence : fence; + + dma_fence_get(f); + if (!dma_fence_add_callback(f, &head->cb, dma_fence_chain_cb)) { + dma_fence_put(fence); + return true; + } + dma_fence_put(f); + } + dma_fence_put(&head->base); + return false; +} + +static bool dma_fence_chain_signaled(struct dma_fence *fence) +{ + dma_fence_chain_for_each(fence, fence) { + struct dma_fence_chain *chain = to_dma_fence_chain(fence); + struct dma_fence *f = chain ? chain->fence : fence; + + if (!dma_fence_is_signaled(f)) { + dma_fence_put(fence); + return false; + } + } + + return true; +} + +static void dma_fence_chain_release(struct dma_fence *fence) +{ + struct dma_fence_chain *chain = to_dma_fence_chain(fence); + + dma_fence_put(rcu_dereference_protected(chain->prev, true)); + dma_fence_put(chain->fence); + dma_fence_free(fence); +} + +const struct dma_fence_ops dma_fence_chain_ops = { + .get_driver_name = dma_fence_chain_get_driver_name, + .get_timeline_name = dma_fence_chain_get_timeline_name, + .enable_signaling = dma_fence_chain_enable_signaling, + .signaled = dma_fence_chain_signaled, + .release = dma_fence_chain_release, +}; +EXPORT_SYMBOL(dma_fence_chain_ops); + +/** + * dma_fence_chain_init - initialize a fence chain + * @chain: the chain node to initialize + * @prev: the previous fence + * @fence: the current fence + * + * Initialize a new chain node and either start a new chain or add the node to + * the existing chain of the previous fence. + */ +void dma_fence_chain_init(struct dma_fence_chain *chain, + struct dma_fence *prev, + struct dma_fence *fence, + uint64_t seqno) +{ + struct dma_fence_chain *prev_chain = to_dma_fence_chain(prev); + uint64_t context; + + spin_lock_init(&chain->lock); + rcu_assign_pointer(chain->prev, prev); + chain->fence = fence; + chain->prev_seqno = 0; + init_irq_work(&chain->work, dma_fence_chain_irq_work); + + /* Try to reuse the context of the previous chain node. */ + if (prev_chain && __dma_fence_is_later(seqno, prev->seqno)) { + context = prev->context; + chain->prev_seqno = prev->seqno; + } else { + context = dma_fence_context_alloc(1); + /* Make sure that we always have a valid sequence number. */ + if (prev_chain) + seqno = max(prev->seqno, seqno); + } + + dma_fence_init(&chain->base, &dma_fence_chain_ops, + &chain->lock, context, seqno); +} +EXPORT_SYMBOL(dma_fence_chain_init); diff --git a/include/linux/dma-fence-chain.h b/include/linux/dma-fence-chain.h new file mode 100644 index 000000000000..934a442db8ac --- /dev/null +++ b/include/linux/dma-fence-chain.h @@ -0,0 +1,81 @@ +/* + * fence-chain: chain fences together in a timeline + * + * Copyright (C) 2018 Advanced Micro Devices, Inc. + * Authors: + * Christian König christian.koenig@amd.com + * + * This program is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 as published by + * the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + */ + +#ifndef __LINUX_DMA_FENCE_CHAIN_H +#define __LINUX_DMA_FENCE_CHAIN_H + +#include <linux/dma-fence.h> +#include <linux/irq_work.h> + +/** + * struct dma_fence_chain - fence to represent an node of a fence chain + * @base: fence base class + * @lock: spinlock for fence handling + * @prev: previous fence of the chain + * @prev_seqno: original previous seqno before garbage collection + * @fence: encapsulated fence + * @cb: callback structure for signaling + * @work: irq work item for signaling + */ +struct dma_fence_chain { + struct dma_fence base; + spinlock_t lock; + struct dma_fence __rcu *prev; + u64 prev_seqno; + struct dma_fence *fence; + struct dma_fence_cb cb; + struct irq_work work; +}; + +extern const struct dma_fence_ops dma_fence_chain_ops; + +/** + * to_dma_fence_chain - cast a fence to a dma_fence_chain + * @fence: fence to cast to a dma_fence_array + * + * Returns NULL if the fence is not a dma_fence_chain, + * or the dma_fence_chain otherwise. + */ +static inline struct dma_fence_chain * +to_dma_fence_chain(struct dma_fence *fence) +{ + if (!fence || fence->ops != &dma_fence_chain_ops) + return NULL; + + return container_of(fence, struct dma_fence_chain, base); +} + +/** + * dma_fence_chain_for_each - iterate over all fences in chain + * @iter: current fence + * @head: starting point + * + * Iterate over all fences in the chain. We keep a reference to the current + * fence while inside the loop which must be dropped when breaking out. + */ +#define dma_fence_chain_for_each(iter, head) \ + for (iter = dma_fence_get(head); iter; \ + iter = dma_fence_chain_walk(iter)) + +struct dma_fence *dma_fence_chain_walk(struct dma_fence *fence); +int dma_fence_chain_find_seqno(struct dma_fence **pfence, uint64_t seqno); +void dma_fence_chain_init(struct dma_fence_chain *chain, + struct dma_fence *prev, + struct dma_fence *fence, + uint64_t seqno); + +#endif /* __LINUX_DMA_FENCE_CHAIN_H */

points array is one-to-one match with syncobjs array. v2: add seperate ioctl for timeline point wait, otherwise break uapi. v3: userspace can specify two kinds waits:: a. Wait for time point to be completed. b. and wait for time point to become available v4: rebase v5: add comment for xxx_WAIT_AVAILABLE v6: rebase and rework on new container v7: drop _WAIT_COMPLETED, it is the default anyway v8: correctly handle garbage collected fences

Signed-off-by: Chunming Zhou david1.zhou@amd.com Signed-off-by: Christian König christian.koenig@amd.com Cc: Tobias Hector Tobias.Hector@amd.com Cc: Jason Ekstrand jason@jlekstrand.net Cc: Dave Airlie airlied@redhat.com Cc: Chris Wilson chris@chris-wilson.co.uk Cc: Lionel Landwerlin lionel.g.landwerlin@intel.com Reviewed-by: Lionel Landwerlin lionel.g.landwerlin@intel.com --- drivers/gpu/drm/drm_internal.h | 2 + drivers/gpu/drm/drm_ioctl.c | 2 + drivers/gpu/drm/drm_syncobj.c | 153 ++++++++++++++++++++++++++------- include/uapi/drm/drm.h | 15 ++++ 4 files changed, 143 insertions(+), 29 deletions(-)

diff --git a/drivers/gpu/drm/drm_internal.h b/drivers/gpu/drm/drm_internal.h index 251d67e04c2d..331ac6225b58 100644 --- a/drivers/gpu/drm/drm_internal.h +++ b/drivers/gpu/drm/drm_internal.h @@ -182,6 +182,8 @@ int drm_syncobj_fd_to_handle_ioctl(struct drm_device *dev, void *data, struct drm_file *file_private); int drm_syncobj_wait_ioctl(struct drm_device *dev, void *data, struct drm_file *file_private); +int drm_syncobj_timeline_wait_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_private); int drm_syncobj_reset_ioctl(struct drm_device *dev, void *data, struct drm_file *file_private); int drm_syncobj_signal_ioctl(struct drm_device *dev, void *data, diff --git a/drivers/gpu/drm/drm_ioctl.c b/drivers/gpu/drm/drm_ioctl.c index 687943df58e1..c984654646fa 100644 --- a/drivers/gpu/drm/drm_ioctl.c +++ b/drivers/gpu/drm/drm_ioctl.c @@ -688,6 +688,8 @@ static const struct drm_ioctl_desc drm_ioctls[] = { DRM_UNLOCKED|DRM_RENDER_ALLOW), DRM_IOCTL_DEF(DRM_IOCTL_SYNCOBJ_WAIT, drm_syncobj_wait_ioctl, DRM_UNLOCKED|DRM_RENDER_ALLOW), + DRM_IOCTL_DEF(DRM_IOCTL_SYNCOBJ_TIMELINE_WAIT, drm_syncobj_timeline_wait_ioctl, + DRM_UNLOCKED|DRM_RENDER_ALLOW), DRM_IOCTL_DEF(DRM_IOCTL_SYNCOBJ_RESET, drm_syncobj_reset_ioctl, DRM_UNLOCKED|DRM_RENDER_ALLOW), DRM_IOCTL_DEF(DRM_IOCTL_SYNCOBJ_SIGNAL, drm_syncobj_signal_ioctl, diff --git a/drivers/gpu/drm/drm_syncobj.c b/drivers/gpu/drm/drm_syncobj.c index dbe4a1c75fbc..b1262e92011c 100644 --- a/drivers/gpu/drm/drm_syncobj.c +++ b/drivers/gpu/drm/drm_syncobj.c @@ -61,6 +61,7 @@ struct syncobj_wait_entry { struct task_struct *task; struct dma_fence *fence; struct dma_fence_cb fence_cb; + u64 point; };

static void syncobj_wait_syncobj_func(struct drm_syncobj *syncobj, @@ -95,6 +96,8 @@ EXPORT_SYMBOL(drm_syncobj_find); static void drm_syncobj_fence_add_wait(struct drm_syncobj *syncobj, struct syncobj_wait_entry *wait) { + struct dma_fence *fence; + if (wait->fence) return;

@@ -103,11 +106,15 @@ static void drm_syncobj_fence_add_wait(struct drm_syncobj *syncobj, * have the lock, try one more time just to be sure we don't add a * callback when a fence has already been set. */ - if (syncobj->fence) - wait->fence = dma_fence_get( - rcu_dereference_protected(syncobj->fence, 1)); - else + fence = dma_fence_get(rcu_dereference_protected(syncobj->fence, 1)); + if (!fence || dma_fence_chain_find_seqno(&fence, wait->point)) { + dma_fence_put(fence); list_add_tail(&wait->node, &syncobj->cb_list); + } else if (!fence) { + wait->fence = dma_fence_get_stub(); + } else { + wait->fence = fence; + } spin_unlock(&syncobj->lock); }

@@ -150,10 +157,8 @@ void drm_syncobj_add_point(struct drm_syncobj *syncobj, dma_fence_chain_init(chain, prev, fence, point); rcu_assign_pointer(syncobj->fence, &chain->base);

- list_for_each_entry_safe(cur, tmp, &syncobj->cb_list, node) { - list_del_init(&cur->node); + list_for_each_entry_safe(cur, tmp, &syncobj->cb_list, node) syncobj_wait_syncobj_func(syncobj, cur); - } spin_unlock(&syncobj->lock);

/* Walk the chain once to trigger garbage collection */ @@ -185,10 +190,8 @@ void drm_syncobj_replace_fence(struct drm_syncobj *syncobj, rcu_assign_pointer(syncobj->fence, fence);

if (fence != old_fence) { - list_for_each_entry_safe(cur, tmp, &syncobj->cb_list, node) { - list_del_init(&cur->node); + list_for_each_entry_safe(cur, tmp, &syncobj->cb_list, node) syncobj_wait_syncobj_func(syncobj, cur); - } }

spin_unlock(&syncobj->lock); @@ -645,13 +648,27 @@ static void syncobj_wait_fence_func(struct dma_fence *fence, static void syncobj_wait_syncobj_func(struct drm_syncobj *syncobj, struct syncobj_wait_entry *wait) { + struct dma_fence *fence; + /* This happens inside the syncobj lock */ - wait->fence = dma_fence_get(rcu_dereference_protected(syncobj->fence, - lockdep_is_held(&syncobj->lock))); + fence = rcu_dereference_protected(syncobj->fence, + lockdep_is_held(&syncobj->lock)); + dma_fence_get(fence); + if (!fence || dma_fence_chain_find_seqno(&fence, wait->point)) { + dma_fence_put(fence); + return; + } else if (!fence) { + wait->fence = dma_fence_get_stub(); + } else { + wait->fence = fence; + } + wake_up_process(wait->task); + list_del_init(&wait->node); }

static signed long drm_syncobj_array_wait_timeout(struct drm_syncobj **syncobjs, + void __user *user_points, uint32_t count, uint32_t flags, signed long timeout, @@ -659,12 +676,27 @@ static signed long drm_syncobj_array_wait_timeout(struct drm_syncobj **syncobjs, { struct syncobj_wait_entry *entries; struct dma_fence *fence; + uint64_t *points; uint32_t signaled_count, i;

- entries = kcalloc(count, sizeof(*entries), GFP_KERNEL); - if (!entries) + points = kmalloc_array(count, sizeof(*points), GFP_KERNEL); + if (points == NULL) return -ENOMEM;

+ if (!user_points) { + memset(points, 0, count * sizeof(uint64_t)); + + } else if (copy_from_user(points, user_points, + sizeof(uint64_t) * count)) { + timeout = -EFAULT; + goto err_free_points; + } + + entries = kcalloc(count, sizeof(*entries), GFP_KERNEL); + if (!entries) { + timeout = -ENOMEM; + goto err_free_points; + } /* Walk the list of sync objects and initialize entries. We do * this up-front so that we can properly return -EINVAL if there is * a syncobj with a missing fence and then never have the chance of @@ -672,9 +704,13 @@ static signed long drm_syncobj_array_wait_timeout(struct drm_syncobj **syncobjs, */ signaled_count = 0; for (i = 0; i < count; ++i) { + struct dma_fence *fence; + entries[i].task = current; - entries[i].fence = drm_syncobj_fence_get(syncobjs[i]); - if (!entries[i].fence) { + entries[i].point = points[i]; + fence = drm_syncobj_fence_get(syncobjs[i]); + if (!fence || dma_fence_chain_find_seqno(&fence, points[i])) { + dma_fence_put(fence); if (flags & DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT) { continue; } else { @@ -683,7 +719,13 @@ static signed long drm_syncobj_array_wait_timeout(struct drm_syncobj **syncobjs, } }

- if (dma_fence_is_signaled(entries[i].fence)) { + if (fence) + entries[i].fence = fence; + else + entries[i].fence = dma_fence_get_stub(); + + if ((flags & DRM_SYNCOBJ_WAIT_FLAGS_WAIT_AVAILABLE) || + dma_fence_is_signaled(entries[i].fence)) { if (signaled_count == 0 && idx) *idx = i; signaled_count++; @@ -716,7 +758,8 @@ static signed long drm_syncobj_array_wait_timeout(struct drm_syncobj **syncobjs, if (!fence) continue;

- if (dma_fence_is_signaled(fence) || + if ((flags & DRM_SYNCOBJ_WAIT_FLAGS_WAIT_AVAILABLE) || + dma_fence_is_signaled(fence) || (!entries[i].fence_cb.func && dma_fence_add_callback(fence, &entries[i].fence_cb, @@ -761,6 +804,9 @@ static signed long drm_syncobj_array_wait_timeout(struct drm_syncobj **syncobjs, } kfree(entries);

+err_free_points: + kfree(points); + return timeout; }

@@ -800,19 +846,33 @@ EXPORT_SYMBOL(drm_timeout_abs_to_jiffies); static int drm_syncobj_array_wait(struct drm_device *dev, struct drm_file *file_private, struct drm_syncobj_wait *wait, - struct drm_syncobj **syncobjs) + struct drm_syncobj_timeline_wait *timeline_wait, + struct drm_syncobj **syncobjs, bool timeline) { - signed long timeout = drm_timeout_abs_to_jiffies(wait->timeout_nsec); + signed long timeout = 0; uint32_t first = ~0;

- timeout = drm_syncobj_array_wait_timeout(syncobjs, - wait->count_handles, - wait->flags, - timeout, &first); - if (timeout < 0) - return timeout; - - wait->first_signaled = first; + if (!timeline) { + timeout = drm_timeout_abs_to_jiffies(wait->timeout_nsec); + timeout = drm_syncobj_array_wait_timeout(syncobjs, + NULL, + wait->count_handles, + wait->flags, + timeout, &first); + if (timeout < 0) + return timeout; + wait->first_signaled = first; + } else { + timeout = drm_timeout_abs_to_jiffies(timeline_wait->timeout_nsec); + timeout = drm_syncobj_array_wait_timeout(syncobjs, + u64_to_user_ptr(timeline_wait->points), + timeline_wait->count_handles, + timeline_wait->flags, + timeout, &first); + if (timeout < 0) + return timeout; + timeline_wait->first_signaled = first; + } return 0; }

@@ -898,13 +958,48 @@ drm_syncobj_wait_ioctl(struct drm_device *dev, void *data, return ret;

ret = drm_syncobj_array_wait(dev, file_private, - args, syncobjs); + args, NULL, syncobjs, false);

drm_syncobj_array_free(syncobjs, args->count_handles);

return ret; }

+int +drm_syncobj_timeline_wait_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_private) +{ + struct drm_syncobj_timeline_wait *args = data; + struct drm_syncobj **syncobjs; + int ret = 0; + + if (!drm_core_check_feature(dev, DRIVER_SYNCOBJ)) + return -ENODEV; + + if (args->flags & ~(DRM_SYNCOBJ_WAIT_FLAGS_WAIT_ALL | + DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT | + DRM_SYNCOBJ_WAIT_FLAGS_WAIT_AVAILABLE)) + return -EINVAL; + + if (args->count_handles == 0) + return -EINVAL; + + ret = drm_syncobj_array_find(file_private, + u64_to_user_ptr(args->handles), + args->count_handles, + &syncobjs); + if (ret < 0) + return ret; + + ret = drm_syncobj_array_wait(dev, file_private, + NULL, args, syncobjs, true); + + drm_syncobj_array_free(syncobjs, args->count_handles); + + return ret; +} + + int drm_syncobj_reset_ioctl(struct drm_device *dev, void *data, struct drm_file *file_private) diff --git a/include/uapi/drm/drm.h b/include/uapi/drm/drm.h index 300f336633f2..44ebcdd9bd1d 100644 --- a/include/uapi/drm/drm.h +++ b/include/uapi/drm/drm.h @@ -737,6 +737,7 @@ struct drm_syncobj_handle {

#define DRM_SYNCOBJ_WAIT_FLAGS_WAIT_ALL (1 << 0) #define DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT (1 << 1) +#define DRM_SYNCOBJ_WAIT_FLAGS_WAIT_AVAILABLE (1 << 2) /* wait for time point to become available */ struct drm_syncobj_wait { __u64 handles; /* absolute timeout */ @@ -747,6 +748,19 @@ struct drm_syncobj_wait { __u32 pad; };

+struct drm_syncobj_timeline_wait { + __u64 handles; + /* wait on specific timeline point for every handles*/ + __u64 points; + /* absolute timeout */ + __s64 timeout_nsec; + __u32 count_handles; + __u32 flags; + __u32 first_signaled; /* only valid when not waiting all */ + __u32 pad; +}; + + struct drm_syncobj_array { __u64 handles; __u32 count_handles; @@ -909,6 +923,7 @@ extern "C" { #define DRM_IOCTL_MODE_GET_LEASE DRM_IOWR(0xC8, struct drm_mode_get_lease) #define DRM_IOCTL_MODE_REVOKE_LEASE DRM_IOWR(0xC9, struct drm_mode_revoke_lease)

+#define DRM_IOCTL_SYNCOBJ_TIMELINE_WAIT DRM_IOWR(0xCA, struct drm_syncobj_timeline_wait) /** * Device specific ioctls should only be in their respective headers * The device specific ioctl range is from 0x40 to 0x9f.

From: Christian König ckoenig.leichtzumerken@gmail.com

Implement finding the right timeline point in drm_syncobj_find_fence.

v2: return -EINVAL when the point is not submitted yet. v3: fix reference counting bug, add flags handling as well v4: add timeout for find fence

Signed-off-by: Christian König christian.koenig@amd.com Cc: Lionel Landwerlin lionel.g.landwerlin@intel.com Reviewed-by: Lionel Landwerlin lionel.g.landwerlin@intel.com --- drivers/gpu/drm/drm_syncobj.c | 50 ++++++++++++++++++++++++++++++++--- 1 file changed, 47 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/drm_syncobj.c b/drivers/gpu/drm/drm_syncobj.c index 3a1be7f79de1..6c273e73d920 100644 --- a/drivers/gpu/drm/drm_syncobj.c +++ b/drivers/gpu/drm/drm_syncobj.c @@ -214,6 +214,8 @@ static void drm_syncobj_assign_null_handle(struct drm_syncobj *syncobj) dma_fence_put(fence); }

+/* 5s default for wait submission */ +#define DRM_SYNCOBJ_WAIT_FOR_SUBMIT_TIMEOUT 5000000000ULL /** * drm_syncobj_find_fence - lookup and reference the fence in a sync object * @file_private: drm file private pointer @@ -234,16 +236,58 @@ int drm_syncobj_find_fence(struct drm_file *file_private, struct dma_fence **fence) { struct drm_syncobj *syncobj = drm_syncobj_find(file_private, handle); - int ret = 0; + struct syncobj_wait_entry wait; + u64 timeout = nsecs_to_jiffies64(DRM_SYNCOBJ_WAIT_FOR_SUBMIT_TIMEOUT); + int ret;

if (!syncobj) return -ENOENT;

*fence = drm_syncobj_fence_get(syncobj); - if (!*fence) { + drm_syncobj_put(syncobj); + + if (*fence) { + ret = dma_fence_chain_find_seqno(fence, point); + if (!ret) + return 0; + dma_fence_put(*fence); + } else { ret = -EINVAL; } - drm_syncobj_put(syncobj); + + if (!(flags & DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT)) + return ret; + + memset(&wait, 0, sizeof(wait)); + wait.task = current; + wait.point = point; + drm_syncobj_fence_add_wait(syncobj, &wait); + + do { + set_current_state(TASK_INTERRUPTIBLE); + if (wait.fence) { + ret = 0; + break; + } + if (timeout == 0) { + ret = -ETIME; + break; + } + + if (signal_pending(current)) { + ret = -ERESTARTSYS; + break; + } + + timeout = schedule_timeout(timeout); + } while (1); + + __set_current_state(TASK_RUNNING); + *fence = wait.fence; + + if (wait.node.next) + drm_syncobj_remove_wait(syncobj, &wait); + return ret; } EXPORT_SYMBOL(drm_syncobj_find_fence);

we need to import/export timeline point.

v2: unify to one transfer ioctl

Signed-off-by: Chunming Zhou david1.zhou@amd.com Cc: Lionel Landwerlin lionel.g.landwerlin@intel.com Reviewed-by: Lionel Landwerlin lionel.g.landwerlin@intel.com --- drivers/gpu/drm/drm_internal.h | 2 + drivers/gpu/drm/drm_ioctl.c | 2 + drivers/gpu/drm/drm_syncobj.c | 74 ++++++++++++++++++++++++++++++++++ include/uapi/drm/drm.h | 10 +++++ 4 files changed, 88 insertions(+)

diff --git a/drivers/gpu/drm/drm_internal.h b/drivers/gpu/drm/drm_internal.h index 695179bb88dc..dd11ae5f1eef 100644 --- a/drivers/gpu/drm/drm_internal.h +++ b/drivers/gpu/drm/drm_internal.h @@ -180,6 +180,8 @@ int drm_syncobj_handle_to_fd_ioctl(struct drm_device *dev, void *data, struct drm_file *file_private); int drm_syncobj_fd_to_handle_ioctl(struct drm_device *dev, void *data, struct drm_file *file_private); +int drm_syncobj_transfer_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_private); int drm_syncobj_wait_ioctl(struct drm_device *dev, void *data, struct drm_file *file_private); int drm_syncobj_timeline_wait_ioctl(struct drm_device *dev, void *data, diff --git a/drivers/gpu/drm/drm_ioctl.c b/drivers/gpu/drm/drm_ioctl.c index 7a534c184e52..92b3b7b2fd81 100644 --- a/drivers/gpu/drm/drm_ioctl.c +++ b/drivers/gpu/drm/drm_ioctl.c @@ -686,6 +686,8 @@ static const struct drm_ioctl_desc drm_ioctls[] = { DRM_UNLOCKED|DRM_RENDER_ALLOW), DRM_IOCTL_DEF(DRM_IOCTL_SYNCOBJ_FD_TO_HANDLE, drm_syncobj_fd_to_handle_ioctl, DRM_UNLOCKED|DRM_RENDER_ALLOW), + DRM_IOCTL_DEF(DRM_IOCTL_SYNCOBJ_TRANSFER, drm_syncobj_transfer_ioctl, + DRM_UNLOCKED|DRM_RENDER_ALLOW), DRM_IOCTL_DEF(DRM_IOCTL_SYNCOBJ_WAIT, drm_syncobj_wait_ioctl, DRM_UNLOCKED|DRM_RENDER_ALLOW), DRM_IOCTL_DEF(DRM_IOCTL_SYNCOBJ_TIMELINE_WAIT, drm_syncobj_timeline_wait_ioctl, diff --git a/drivers/gpu/drm/drm_syncobj.c b/drivers/gpu/drm/drm_syncobj.c index 6c273e73d920..63d5d2bf35c2 100644 --- a/drivers/gpu/drm/drm_syncobj.c +++ b/drivers/gpu/drm/drm_syncobj.c @@ -680,6 +680,80 @@ drm_syncobj_fd_to_handle_ioctl(struct drm_device *dev, void *data, &args->handle); }

+static int drm_syncobj_transfer_to_timeline(struct drm_file *file_private, + struct drm_syncobj_transfer *args) +{ + struct drm_syncobj *timeline_syncobj = NULL; + struct dma_fence *fence; + struct dma_fence_chain *chain; + int ret; + + timeline_syncobj = drm_syncobj_find(file_private, args->dst_handle); + if (!timeline_syncobj) { + return -ENOENT; + } + ret = drm_syncobj_find_fence(file_private, args->src_handle, + args->src_point, args->flags, + &fence); + if (ret) + goto err; + chain = kzalloc(sizeof(struct dma_fence_chain), GFP_KERNEL); + if (!chain) { + ret = -ENOMEM; + goto err1; + } + drm_syncobj_add_point(timeline_syncobj, chain, fence, args->dst_point); +err1: + dma_fence_put(fence); +err: + drm_syncobj_put(timeline_syncobj); + + return ret; +} + +static int +drm_syncobj_transfer_to_binary(struct drm_file *file_private, + struct drm_syncobj_transfer *args) +{ + struct drm_syncobj *binary_syncobj = NULL; + struct dma_fence *fence; + int ret; + + binary_syncobj = drm_syncobj_find(file_private, args->dst_handle); + if (!binary_syncobj) + return -ENOENT; + ret = drm_syncobj_find_fence(file_private, args->src_handle, + args->src_point, args->flags, &fence); + if (ret) + goto err; + drm_syncobj_replace_fence(binary_syncobj, fence); + dma_fence_put(fence); +err: + drm_syncobj_put(binary_syncobj); + + return ret; +} +int +drm_syncobj_transfer_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_private) +{ + struct drm_syncobj_transfer *args = data; + int ret; + + if (!drm_core_check_feature(dev, DRIVER_SYNCOBJ)) + return -ENODEV; + + if (args->pad) + return -EINVAL; + + if (args->dst_point) + ret = drm_syncobj_transfer_to_timeline(file_private, args); + else + ret = drm_syncobj_transfer_to_binary(file_private, args); + + return ret; +} + static void syncobj_wait_fence_func(struct dma_fence *fence, struct dma_fence_cb *cb) { diff --git a/include/uapi/drm/drm.h b/include/uapi/drm/drm.h index c62be0840ba5..e8d0d6b51875 100644 --- a/include/uapi/drm/drm.h +++ b/include/uapi/drm/drm.h @@ -735,6 +735,15 @@ struct drm_syncobj_handle { __u32 pad; };

+struct drm_syncobj_transfer { + __u32 src_handle; + __u32 dst_handle; + __u64 src_point; + __u64 dst_point; + __u32 flags; + __u32 pad; +}; + #define DRM_SYNCOBJ_WAIT_FLAGS_WAIT_ALL (1 << 0) #define DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT (1 << 1) #define DRM_SYNCOBJ_WAIT_FLAGS_WAIT_AVAILABLE (1 << 2) /* wait for time point to become available */ @@ -933,6 +942,7 @@ extern "C" {

#define DRM_IOCTL_SYNCOBJ_TIMELINE_WAIT DRM_IOWR(0xCA, struct drm_syncobj_timeline_wait) #define DRM_IOCTL_SYNCOBJ_QUERY DRM_IOWR(0xCB, struct drm_syncobj_timeline_array) +#define DRM_IOCTL_SYNCOBJ_TRANSFER DRM_IOWR(0xCC, struct drm_syncobj_transfer)

/** * Device specific ioctls should only be in their respective headers

Signed-off-by: Chunming Zhou david1.zhou@amd.com Reviewed-by: Lionel Landwerlin lionel.g.landwerlin@intel.com --- drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c index 8a0732088640..4d8db87048d3 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c @@ -74,9 +74,10 @@ * - 3.28.0 - Add AMDGPU_CHUNK_ID_SCHEDULED_DEPENDENCIES * - 3.29.0 - Add AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID * - 3.30.0 - Add AMDGPU_SCHED_OP_CONTEXT_PRIORITY_OVERRIDE. + * - 3.31.0 - Add syncobj timeline support to AMDGPU_CS. */ #define KMS_DRIVER_MAJOR 3 -#define KMS_DRIVER_MINOR 30 +#define KMS_DRIVER_MINOR 31 #define KMS_DRIVER_PATCHLEVEL 0

int amdgpu_vram_limit = 0;