[Intel-gfx] [PATCH] drm/i915: Unshare the idle-barrier from other kernel requests
Chris Wilson
chris at chris-wilson.co.uk
Wed Jul 24 12:50:34 UTC 2019
Under some circumstances (see intel_context_prepare_remote_request), we
may use a request along a kernel context to modify the logical state of
another. To keep the target context in place while the request executes,
we take an active reference on it using the kernel timeline. This is the
same timeline as we use for the idle-barrier, and so we end up reusing
the same active node. Except that the idle barrier is special and cannot
be reused in this manner! Give the idle-barrier a reserved timeline
index (0) so that it will always be unique (give or take we may issue
multiple idle barriers across multiple engines).
Reported-by: Lionel Landwerlin <lionel.g.landwerlin at intel.com>
Fixes: ce476c80b8bf ("drm/i915: Keep contexts pinned until after the next kernel context switch")
Fixes: a9877da2d629 ("drm/i915/oa: Reconfigure contexts on the fly")
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
Cc: Lionel Landwerlin <lionel.g.landwerlin at intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
---
Replace recursive attempt to acquire the i915_active.mutex with a
lockdep annotation.
---
drivers/gpu/drm/i915/i915_active.c | 58 +++++++++++++++++++++---------
1 file changed, 42 insertions(+), 16 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_active.c b/drivers/gpu/drm/i915/i915_active.c
index 13f304a29fc8..99bebd77f364 100644
--- a/drivers/gpu/drm/i915/i915_active.c
+++ b/drivers/gpu/drm/i915/i915_active.c
@@ -184,6 +184,7 @@ active_instance(struct i915_active *ref, u64 idx)
ref->cache = node;
mutex_unlock(&ref->mutex);
+ BUILD_BUG_ON(offsetof(typeof(*node), base));
return &node->base;
}
@@ -212,6 +213,8 @@ int i915_active_ref(struct i915_active *ref,
struct i915_active_request *active;
int err;
+ GEM_BUG_ON(!timeline); /* reserved for idle-barrier */
+
/* Prevent reaping in case we malloc/wait while building the tree */
err = i915_active_acquire(ref);
if (err)
@@ -342,6 +345,26 @@ void i915_active_fini(struct i915_active *ref)
}
#endif
+static struct active_node *idle_barrier(struct i915_active *ref)
+{
+ struct active_node *node;
+ struct rb_node *rb;
+
+ lockdep_assert_held(&ref->mutex);
+ rb = rb_first(&ref->tree);
+ if (!rb)
+ return NULL;
+
+ node = rb_entry(rb, typeof(*node), node);
+ if (node->timeline || i915_active_request_isset(&node->base))
+ return NULL;
+
+ GEM_BUG_ON(!list_empty(&node->base.link));
+ rb_erase(rb, &ref->tree);
+
+ return node;
+}
+
int i915_active_acquire_preallocate_barrier(struct i915_active *ref,
struct intel_engine_cs *engine)
{
@@ -352,22 +375,29 @@ int i915_active_acquire_preallocate_barrier(struct i915_active *ref,
GEM_BUG_ON(!engine->mask);
for_each_engine_masked(engine, i915, engine->mask, tmp) {
- struct intel_context *kctx = engine->kernel_context;
struct active_node *node;
- node = kmem_cache_alloc(global.slab_cache, GFP_KERNEL);
- if (unlikely(!node)) {
- err = -ENOMEM;
- goto unwind;
+ node = idle_barrier(ref);
+ if (!node) {
+ node = kmem_cache_alloc(global.slab_cache,
+ GFP_KERNEL |
+ __GFP_RETRY_MAYFAIL |
+ __GFP_NOWARN);
+ if (unlikely(!node)) {
+ err = -ENOMEM;
+ goto unwind;
+ }
+
+ node->ref = ref;
+ node->timeline = 0;
+ node->base.retire = node_retire;
}
- i915_active_request_init(&node->base,
- (void *)engine, node_retire);
- node->timeline = kctx->ring->timeline->fence_context;
- node->ref = ref;
+ intel_engine_pm_get(engine);
+
+ RCU_INIT_POINTER(node->base.request, (void *)engine);
atomic_inc(&ref->count);
- intel_engine_pm_get(engine);
llist_add((struct llist_node *)&node->base.link,
&ref->barriers);
}
@@ -402,6 +432,7 @@ void i915_active_acquire_barrier(struct i915_active *ref)
node = container_of((struct list_head *)pos,
typeof(*node), base.link);
+ GEM_BUG_ON(node->timeline);
engine = (void *)rcu_access_pointer(node->base.request);
RCU_INIT_POINTER(node->base.request, ERR_PTR(-EAGAIN));
@@ -410,12 +441,7 @@ void i915_active_acquire_barrier(struct i915_active *ref)
p = &ref->tree.rb_node;
while (*p) {
parent = *p;
- if (rb_entry(parent,
- struct active_node,
- node)->timeline < node->timeline)
- p = &parent->rb_right;
- else
- p = &parent->rb_left;
+ p = &parent->rb_left;
}
rb_link_node(&node->node, parent, p);
rb_insert_color(&node->node, &ref->tree);
--
2.22.0
More information about the Intel-gfx
mailing list