[PATCH 12/13] drm/i915: Move vma binding to obj, instead of relying on i915_active
Maarten Lankhorst
maarten.lankhorst at linux.intel.com
Tue Jul 27 08:27:17 UTC 2021
Instead of relying on the i915_active exclusive fence, use obj->moving
fence. This will also be used for later support for async gpu
binding/clearing, which does allow us to keep it all in a single place.
As i915_active kept the vma alive, just in case of a race add an extra
ref to i915_vma in the worker.
i915_active.excl is now unused, so remove all related infrastructure as
well.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst at linux.intel.com>
---
drivers/gpu/drm/i915/gt/intel_ggtt.c | 24 ++++++++--
drivers/gpu/drm/i915/i915_active.c | 47 --------------------
drivers/gpu/drm/i915/i915_active.h | 19 --------
drivers/gpu/drm/i915/i915_active_types.h | 3 --
drivers/gpu/drm/i915/i915_vma.c | 32 ++++++++-----
drivers/gpu/drm/i915/selftests/i915_active.c | 1 -
6 files changed, 40 insertions(+), 86 deletions(-)
diff --git a/drivers/gpu/drm/i915/gt/intel_ggtt.c b/drivers/gpu/drm/i915/gt/intel_ggtt.c
index b3eae0722250..ee485a7be32c 100644
--- a/drivers/gpu/drm/i915/gt/intel_ggtt.c
+++ b/drivers/gpu/drm/i915/gt/intel_ggtt.c
@@ -118,6 +118,7 @@ void i915_ggtt_suspend(struct i915_ggtt *ggtt)
struct i915_vma *vma, *vn;
int open;
+retry:
mutex_lock(&ggtt->vm.mutex);
/* Skip rewriting PTE on VMA unbind. */
@@ -125,15 +126,30 @@ void i915_ggtt_suspend(struct i915_ggtt *ggtt)
list_for_each_entry_safe(vma, vn, &ggtt->vm.bound_list, vm_link) {
GEM_BUG_ON(!drm_mm_node_allocated(&vma->node));
- i915_vma_wait_for_bind(vma);
- if (i915_vma_is_pinned(vma))
- continue;
+ i915_vma_get(vma);
+
+ /* unlikely to race when GPU is idle, so no worry about slowpath.. */
+ if (!i915_gem_object_trylock(vma->obj)) {
+ atomic_set(&ggtt->vm.open, open);
+ mutex_unlock(&ggtt->vm.mutex);
+
+ i915_gem_object_lock(vma->obj, NULL);
+ i915_vma_wait_for_bind(vma);
+ i915_gem_object_unlock(vma->obj);
+
+ i915_vma_put(vma);
+ goto retry;
+ }
+
+ i915_vma_wait_for_bind(vma);
- if (!i915_vma_is_bound(vma, I915_VMA_GLOBAL_BIND)) {
+ if (!i915_vma_is_pinned(vma) && !i915_vma_is_bound(vma, I915_VMA_GLOBAL_BIND)) {
__i915_vma_evict(vma);
drm_mm_remove_node(&vma->node);
}
+ i915_gem_object_unlock(vma->obj);
+ i915_vma_put(vma);
}
ggtt->vm.clear_range(&ggtt->vm, 0, ggtt->vm.total);
diff --git a/drivers/gpu/drm/i915/i915_active.c b/drivers/gpu/drm/i915/i915_active.c
index eb73ed5e4854..bb2faed0f04d 100644
--- a/drivers/gpu/drm/i915/i915_active.c
+++ b/drivers/gpu/drm/i915/i915_active.c
@@ -140,7 +140,6 @@ __active_retire(struct i915_active *ref)
if (!atomic_dec_and_lock_irqsave(&ref->count, &ref->tree_lock, flags))
return;
- GEM_BUG_ON(rcu_access_pointer(ref->excl.fence));
debug_active_deactivate(ref);
/* Even if we have not used the cache, we may still have a barrier */
@@ -227,13 +226,6 @@ node_retire(struct dma_fence *fence, struct dma_fence_cb *cb)
active_retire(container_of(cb, struct active_node, base.cb)->ref);
}
-static void
-excl_retire(struct dma_fence *fence, struct dma_fence_cb *cb)
-{
- if (active_fence_cb(fence, cb))
- active_retire(container_of(cb, struct i915_active, excl.cb));
-}
-
static struct active_node *__active_lookup(struct i915_active *ref, u64 idx)
{
struct active_node *it;
@@ -360,7 +352,6 @@ void __i915_active_init(struct i915_active *ref,
init_llist_head(&ref->preallocated_barriers);
atomic_set(&ref->count, 0);
__mutex_init(&ref->mutex, "i915_active", mkey);
- __i915_active_fence_init(&ref->excl, NULL, excl_retire);
INIT_WORK(&ref->work, active_work);
#if IS_ENABLED(CONFIG_LOCKDEP)
lockdep_init_map(&ref->work.lockdep_map, "i915_active.work", wkey, 0);
@@ -459,36 +450,6 @@ int i915_active_add_request(struct i915_active *ref, struct i915_request *rq)
return err;
}
-static struct dma_fence *
-__i915_active_set_fence(struct i915_active *ref,
- struct i915_active_fence *active,
- struct dma_fence *fence)
-{
- struct dma_fence *prev;
-
- if (replace_barrier(ref, active)) {
- RCU_INIT_POINTER(active->fence, fence);
- return NULL;
- }
-
- rcu_read_lock();
- prev = __i915_active_fence_set(active, fence);
- if (prev)
- prev = dma_fence_get_rcu(prev);
- else
- __i915_active_acquire(ref);
- rcu_read_unlock();
-
- return prev;
-}
-
-struct dma_fence *
-i915_active_set_exclusive(struct i915_active *ref, struct dma_fence *f)
-{
- /* We expect the caller to manage the exclusive timeline ordering */
- return __i915_active_set_fence(ref, &ref->excl, f);
-}
-
bool i915_active_acquire_if_busy(struct i915_active *ref)
{
debug_active_assert(ref);
@@ -589,7 +550,6 @@ static int flush_lazy_signals(struct i915_active *ref)
struct active_node *it, *n;
int err = 0;
- enable_signaling(&ref->excl);
rbtree_postorder_for_each_entry_safe(it, n, &ref->tree, node) {
err = flush_barrier(it); /* unconnected idle barrier? */
if (err)
@@ -701,13 +661,6 @@ static int await_active(struct i915_active *ref,
if (!i915_active_acquire_if_busy(ref))
return 0;
- if (flags & I915_ACTIVE_AWAIT_EXCL &&
- rcu_access_pointer(ref->excl.fence)) {
- err = __await_active(&ref->excl, fn, arg);
- if (err)
- goto out;
- }
-
if (flags & I915_ACTIVE_AWAIT_ACTIVE) {
struct active_node *it, *n;
diff --git a/drivers/gpu/drm/i915/i915_active.h b/drivers/gpu/drm/i915/i915_active.h
index 2ac875888872..c6ab5fe8da0c 100644
--- a/drivers/gpu/drm/i915/i915_active.h
+++ b/drivers/gpu/drm/i915/i915_active.h
@@ -166,9 +166,6 @@ void __i915_active_init(struct i915_active *ref,
int i915_active_add_request(struct i915_active *ref, struct i915_request *rq);
-struct dma_fence *
-i915_active_set_exclusive(struct i915_active *ref, struct dma_fence *f);
-
int __i915_active_wait(struct i915_active *ref, int state);
static inline int i915_active_wait(struct i915_active *ref)
{
@@ -181,7 +178,6 @@ int i915_sw_fence_await_active(struct i915_sw_fence *fence,
int i915_request_await_active(struct i915_request *rq,
struct i915_active *ref,
unsigned int flags);
-#define I915_ACTIVE_AWAIT_EXCL BIT(0)
#define I915_ACTIVE_AWAIT_ACTIVE BIT(1)
#define I915_ACTIVE_AWAIT_BARRIER BIT(2)
@@ -217,19 +213,4 @@ struct i915_active *i915_active_create(void);
struct i915_active *i915_active_get(struct i915_active *ref);
void i915_active_put(struct i915_active *ref);
-static inline int __i915_request_await_exclusive(struct i915_request *rq,
- struct i915_active *active)
-{
- struct dma_fence *fence;
- int err = 0;
-
- fence = i915_active_fence_get(&active->excl);
- if (fence) {
- err = i915_request_await_dma_fence(rq, fence);
- dma_fence_put(fence);
- }
-
- return err;
-}
-
#endif /* _I915_ACTIVE_H_ */
diff --git a/drivers/gpu/drm/i915/i915_active_types.h b/drivers/gpu/drm/i915/i915_active_types.h
index c149f348a972..c3109602cf23 100644
--- a/drivers/gpu/drm/i915/i915_active_types.h
+++ b/drivers/gpu/drm/i915/i915_active_types.h
@@ -32,9 +32,6 @@ struct i915_active {
struct active_node *cache;
struct rb_root tree;
- /* Preallocated "exclusive" node */
- struct i915_active_fence excl;
-
unsigned long flags;
#define I915_ACTIVE_RETIRE_SLEEPS BIT(0)
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index f370c3581038..5350defac2a2 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -317,6 +317,8 @@ static void __vma_release(struct dma_fence_work *work)
}
i915_vm_free_pt_stash(vw->vm, &vw->stash);
+ if (work->vma)
+ __i915_vma_put(work->vma);
i915_vm_put(vw->vm);
}
@@ -342,18 +344,14 @@ struct i915_vma_work *i915_vma_work(void)
int i915_vma_wait_for_bind(struct i915_vma *vma)
{
+ struct dma_fence *fence = i915_gem_object_get_moving_fence(vma->obj);
int err = 0;
- if (rcu_access_pointer(vma->active.excl.fence)) {
- struct dma_fence *fence;
+ assert_vma_held(vma);
- rcu_read_lock();
- fence = dma_fence_get_rcu_safe(&vma->active.excl.fence);
- rcu_read_unlock();
- if (fence) {
- err = dma_fence_wait(fence, MAX_SCHEDULE_TIMEOUT);
- dma_fence_put(fence);
- }
+ if (fence) {
+ err = dma_fence_wait(fence, true);
+ dma_fence_put(fence);
}
return err;
@@ -403,9 +401,10 @@ int i915_vma_bind(struct i915_vma *vma,
trace_i915_vma_bind(vma, bind_flags);
if (work && bind_flags & vma->vm->bind_async_flags) {
+ struct drm_i915_gem_object *obj = vma->obj;
struct dma_fence *prev;
- work->vma = vma;
+ work->vma = __i915_vma_get(vma);
work->cache_level = cache_level;
work->flags = bind_flags;
@@ -418,13 +417,14 @@ int i915_vma_bind(struct i915_vma *vma,
* part of the obj->resv->excl_fence as it only affects
* execution and not content or object's backing store lifetime.
*/
- prev = i915_active_set_exclusive(&vma->active, &work->base.dma);
+ prev = i915_gem_object_get_moving_fence(obj);
if (prev) {
__i915_sw_fence_await_dma_fence(&work->base.chain,
prev,
&work->cb);
dma_fence_put(prev);
}
+ i915_gem_object_set_moving_fence(obj, &work->base.dma);
work->base.dma.error = 0; /* enable the queue_work() */
@@ -1512,7 +1512,15 @@ void i915_vma_revoke_mmap(struct i915_vma *vma)
static int
__i915_request_await_bind(struct i915_request *rq, struct i915_vma *vma)
{
- return __i915_request_await_exclusive(rq, &vma->active);
+ struct dma_fence *fence = i915_gem_object_get_moving_fence(vma->obj);
+ int ret;
+
+ if (!fence)
+ return 0;
+
+ ret = i915_request_await_dma_fence(rq, fence);
+ dma_fence_put(fence);
+ return ret;
}
static int __i915_vma_move_to_active(struct i915_vma *vma, struct i915_request *rq)
diff --git a/drivers/gpu/drm/i915/selftests/i915_active.c b/drivers/gpu/drm/i915/selftests/i915_active.c
index 61bf4560d8af..ddaf389b35e0 100644
--- a/drivers/gpu/drm/i915/selftests/i915_active.c
+++ b/drivers/gpu/drm/i915/selftests/i915_active.c
@@ -335,7 +335,6 @@ void i915_active_unlock_wait(struct i915_active *ref)
/* Wait for all active callbacks */
rcu_read_lock();
- active_flush(ref, &ref->excl);
rbtree_postorder_for_each_entry_safe(it, n, &ref->tree, node)
active_flush(ref, &it->base);
rcu_read_unlock();
--
2.31.0
More information about the Intel-gfx-trybot
mailing list