This is a note to let you know that I've just added the patch titled dma-buf/sw-sync: Use an rbtree to sort fences in the timeline to the 4.9-stable tree which can be found at: http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=summary The filename of the patch is: dma-buf-sw-sync-use-an-rbtree-to-sort-fences-in-the-timeline.patch and it can be found in the queue-4.9 subdirectory. If you, or anyone else, feels it should not be added to the stable tree, please let <stable@xxxxxxxxxxxxxxx> know about it. >From f1e8c67123cf171e2b0357e885e426328b241d7d Mon Sep 17 00:00:00 2001 From: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx> Date: Thu, 29 Jun 2017 22:12:53 +0100 Subject: dma-buf/sw-sync: Use an rbtree to sort fences in the timeline From: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx> commit f1e8c67123cf171e2b0357e885e426328b241d7d upstream. Reduce the list iteration when incrementing the timeline by storing the fences in increasing order. v2: Prevent spinlock recursion on free during create v3: Fixup rebase conflict inside comments that escaped the compiler. Signed-off-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx> Cc: Sumit Semwal <sumit.semwal@xxxxxxxxxx> Cc: Sean Paul <seanpaul@xxxxxxxxxxxx> Cc: Gustavo Padovan <gustavo@xxxxxxxxxxx> Reviewed-by: Sean Paul <seanpaul@xxxxxxxxxxxx> Signed-off-by: Gustavo Padovan <gustavo.padovan@xxxxxxxxxxxxx> Link: http://patchwork.freedesktop.org/patch/msgid/20170629211253.22766-1-chris@xxxxxxxxxxxxxxxxxx [s/dma_fence/fence/g - gregkh] Cc: Jisheng Zhang <Jisheng.Zhang@xxxxxxxxxxxxx> Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx> --- drivers/dma-buf/sw_sync.c | 49 +++++++++++++++++++++++++++++++++++++------ drivers/dma-buf/sync_debug.h | 5 ++++ 2 files changed, 48 insertions(+), 6 deletions(-) --- a/drivers/dma-buf/sw_sync.c +++ b/drivers/dma-buf/sw_sync.c @@ -96,6 +96,7 @@ struct sync_timeline *sync_timeline_crea obj->context = fence_context_alloc(1); strlcpy(obj->name, name, sizeof(obj->name)); + obj->pt_tree = RB_ROOT; INIT_LIST_HEAD(&obj->pt_list); spin_lock_init(&obj->lock); @@ -142,9 +143,13 @@ static void sync_timeline_signal(struct obj->value += inc; - list_for_each_entry_safe(pt, next, &obj->pt_list, link) - if (fence_is_signaled_locked(&pt->base)) - list_del_init(&pt->link); + list_for_each_entry_safe(pt, next, &obj->pt_list, link) { + if (!fence_is_signaled_locked(&pt->base)) + break; + + list_del_init(&pt->link); + rb_erase(&pt->node, &obj->pt_tree); + } spin_unlock_irq(&obj->lock); } @@ -174,8 +179,38 @@ static struct sync_pt *sync_pt_create(st INIT_LIST_HEAD(&pt->link); spin_lock_irq(&obj->lock); - if (!fence_is_signaled_locked(&pt->base)) - list_add_tail(&pt->link, &obj->pt_list); + if (!fence_is_signaled_locked(&pt->base)) { + struct rb_node **p = &obj->pt_tree.rb_node; + struct rb_node *parent = NULL; + + while (*p) { + struct sync_pt *other; + int cmp; + + parent = *p; + other = rb_entry(parent, typeof(*pt), node); + cmp = value - other->base.seqno; + if (cmp > 0) { + p = &parent->rb_right; + } else if (cmp < 0) { + p = &parent->rb_left; + } else { + if (fence_get_rcu(&other->base)) { + fence_put(&pt->base); + pt = other; + goto unlock; + } + p = &parent->rb_left; + } + } + rb_link_node(&pt->node, parent, p); + rb_insert_color(&pt->node, &obj->pt_tree); + + parent = rb_next(&pt->node); + list_add_tail(&pt->link, + parent ? &rb_entry(parent, typeof(*pt), node)->link : &obj->pt_list); + } +unlock: spin_unlock_irq(&obj->lock); return pt; @@ -202,8 +237,10 @@ static void timeline_fence_release(struc unsigned long flags; spin_lock_irqsave(fence->lock, flags); - if (!list_empty(&pt->link)) + if (!list_empty(&pt->link)) { list_del(&pt->link); + rb_erase(&pt->node, &parent->pt_tree); + } spin_unlock_irqrestore(fence->lock, flags); } --- a/drivers/dma-buf/sync_debug.h +++ b/drivers/dma-buf/sync_debug.h @@ -14,6 +14,7 @@ #define _LINUX_SYNC_H #include <linux/list.h> +#include <linux/rbtree.h> #include <linux/spinlock.h> #include <linux/fence.h> @@ -25,6 +26,7 @@ * @kref: reference count on fence. * @name: name of the sync_timeline. Useful for debugging * @lock: lock protecting @pt_list and @value + * @pt_tree: rbtree of active (unsignaled/errored) sync_pts * @pt_list: list of active (unsignaled/errored) sync_pts * @sync_timeline_list: membership in global sync_timeline_list */ @@ -36,6 +38,7 @@ struct sync_timeline { u64 context; int value; + struct rb_root pt_tree; struct list_head pt_list; spinlock_t lock; @@ -51,10 +54,12 @@ static inline struct sync_timeline *fenc * struct sync_pt - sync_pt object * @base: base fence object * @link: link on the sync timeline's list + * @node: node in the sync timeline's tree */ struct sync_pt { struct fence base; struct list_head link; + struct rb_node node; }; #ifdef CONFIG_SW_SYNC Patches currently in stable-queue which might be from chris@xxxxxxxxxxxxxxxxxx are queue-4.9/dma-buf-sw-sync-fix-the-is-signaled-test-to-handle-u32-wraparound.patch queue-4.9/dma-fence-clear-fence-status-during-dma_fence_init.patch queue-4.9/dma-buf-sw-sync-fix-locking-around-sync_timeline-lists.patch queue-4.9/dma-fence-wrap-querying-the-fence-status.patch queue-4.9/dma-buf-sw_sync-clean-up-list-before-signaling-the-fence.patch queue-4.9/dma-buf-sw-sync-reduce-irqsave-irqrestore-from-known-context.patch queue-4.9/dma-buf-sw_sync-move-timeline_fence_ops-around.patch queue-4.9/dma-buf-sw-sync-prevent-user-overflow-on-timeline-advance.patch queue-4.9/dma-fence-introduce-drm_fence_set_error-helper.patch queue-4.9/dma-buf-sw_sync-force-signal-all-unsignaled-fences-on-dying-timeline.patch queue-4.9/dma-buf-sw-sync-use-an-rbtree-to-sort-fences-in-the-timeline.patch queue-4.9/dma-buf-sw-sync-sync_pt-is-private-and-of-fixed-size.patch queue-4.9/dma-buf-dma-fence-extract-__dma_fence_is_later.patch