As the conversion between idle-barrier and full i915_active_fence is already serialised by explicit memory barriers, we can reduce the spinlock in i915_active_acquire_preallocate_barrier() for finding an idle-barrier to reuse to an RCU read lock to ensure the fence remains valid, only taking the spinlock for the update of the rbtree itself.
Signed-off-by: Chris Wilson <[email protected]> Reviewed-by: Thomas Hellström <[email protected]> --- drivers/gpu/drm/i915/i915_active.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) diff --git a/drivers/gpu/drm/i915/i915_active.c b/drivers/gpu/drm/i915/i915_active.c index 7b51045c8461..3ee5e02c3e17 100644 --- a/drivers/gpu/drm/i915/i915_active.c +++ b/drivers/gpu/drm/i915/i915_active.c @@ -807,7 +807,6 @@ static struct active_node *reuse_idle_barrier(struct i915_active *ref, u64 idx) if (RB_EMPTY_ROOT(&ref->tree)) return NULL; - spin_lock_irq(&ref->tree_lock); GEM_BUG_ON(i915_active_is_idle(ref)); /* @@ -833,9 +832,9 @@ static struct active_node *reuse_idle_barrier(struct i915_active *ref, u64 idx) prev = p; if (node->timeline < idx) - p = p->rb_right; + p = READ_ONCE(p->rb_right); else - p = p->rb_left; + p = READ_ONCE(p->rb_left); } /* @@ -872,11 +871,10 @@ static struct active_node *reuse_idle_barrier(struct i915_active *ref, u64 idx) goto match; } - spin_unlock_irq(&ref->tree_lock); - return NULL; match: + spin_lock_irq(&ref->tree_lock); rb_erase(p, &ref->tree); /* Hide from waits and sibling allocations */ if (p == &ref->cache->node) WRITE_ONCE(ref->cache, NULL); @@ -911,7 +909,9 @@ int i915_active_acquire_preallocate_barrier(struct i915_active *ref, struct llist_node *prev = first; struct active_node *node; + rcu_read_lock(); node = reuse_idle_barrier(ref, idx); + rcu_read_unlock(); if (!node) { node = kmem_cache_alloc(global.slab_cache, GFP_KERNEL); if (!node) { -- 2.20.1 _______________________________________________ Intel-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/intel-gfx
