[Intel-gfx] [PATCH 6/6] drm/i915/gem: Consolidate ctx->engines[] release
Chris Wilson
chris at chris-wilson.co.uk
Tue Feb 18 11:16:24 UTC 2020
Use the same engine_idle_release() routine for cleaning all old
ctx->engine[] state, closing any potential races with concurrent execbuf
submission.
Closes: https://gitlab.freedesktop.org/drm/intel/issues/1241
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
---
drivers/gpu/drm/i915/gem/i915_gem_context.c | 177 +++++++++++---------
1 file changed, 95 insertions(+), 82 deletions(-)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_context.c b/drivers/gpu/drm/i915/gem/i915_gem_context.c
index 3e82739bdbc0..e6d9dced6954 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_context.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_context.c
@@ -243,7 +243,6 @@ static void __free_engines(struct i915_gem_engines *e, unsigned int count)
if (!e->engines[count])
continue;
- RCU_INIT_POINTER(e->engines[count]->gem_context, NULL);
intel_context_put(e->engines[count]);
}
kfree(e);
@@ -305,7 +304,6 @@ static void i915_gem_context_free(struct i915_gem_context *ctx)
list_del(&ctx->link);
spin_unlock(&ctx->i915->gem.contexts.lock);
- free_engines(rcu_access_pointer(ctx->engines));
mutex_destroy(&ctx->engines_mutex);
if (ctx->timeline)
@@ -492,30 +490,107 @@ static void kill_engines(struct i915_gem_engines *engines)
static void kill_stale_engines(struct i915_gem_context *ctx)
{
struct i915_gem_engines *pos, *next;
- unsigned long flags;
- spin_lock_irqsave(&ctx->stale.lock, flags);
+ spin_lock_irq(&ctx->stale.lock);
+ GEM_BUG_ON(!i915_gem_context_is_closed(ctx));
list_for_each_entry_safe(pos, next, &ctx->stale.engines, link) {
- if (!i915_sw_fence_await(&pos->fence))
+ if (!i915_sw_fence_await(&pos->fence)) {
+ list_del_init(&pos->link);
continue;
+ }
- spin_unlock_irqrestore(&ctx->stale.lock, flags);
+ spin_unlock_irq(&ctx->stale.lock);
kill_engines(pos);
- spin_lock_irqsave(&ctx->stale.lock, flags);
+ spin_lock_irq(&ctx->stale.lock);
+ GEM_BUG_ON(i915_sw_fence_signaled(&pos->fence));
list_safe_reset_next(pos, next, link);
list_del_init(&pos->link); /* decouple from FENCE_COMPLETE */
i915_sw_fence_complete(&pos->fence);
}
- spin_unlock_irqrestore(&ctx->stale.lock, flags);
+ spin_unlock_irq(&ctx->stale.lock);
}
static void kill_context(struct i915_gem_context *ctx)
{
kill_stale_engines(ctx);
- kill_engines(__context_engines_static(ctx));
+}
+
+static int engines_notify(struct i915_sw_fence *fence,
+ enum i915_sw_fence_notify state)
+{
+ struct i915_gem_engines *engines =
+ container_of(fence, typeof(*engines), fence);
+
+ switch (state) {
+ case FENCE_COMPLETE:
+ if (!list_empty(&engines->link)) {
+ struct i915_gem_context *ctx = engines->ctx;
+ unsigned long flags;
+
+ spin_lock_irqsave(&ctx->stale.lock, flags);
+ list_del(&engines->link);
+ spin_unlock_irqrestore(&ctx->stale.lock, flags);
+ }
+ break;
+
+ case FENCE_FREE:
+ init_rcu_head(&engines->rcu);
+ call_rcu(&engines->rcu, free_engines_rcu);
+ break;
+ }
+
+ return NOTIFY_DONE;
+}
+
+static void engines_idle_release(struct i915_gem_engines *engines)
+{
+ struct i915_gem_engines_iter it;
+ struct intel_context *ce;
+
+ GEM_BUG_ON(!engines);
+ i915_sw_fence_init(&engines->fence, engines_notify);
+ INIT_LIST_HEAD(&engines->link);
+
+ for_each_gem_engine(ce, engines, it) {
+ int err = 0;
+
+ RCU_INIT_POINTER(ce->gem_context, NULL);
+
+ if (!ce->timeline) { /* XXX serialisation with execbuf? */
+ intel_context_set_banned(ce);
+ continue;
+ }
+
+ mutex_lock(&ce->timeline->mutex);
+ if (!list_empty(&ce->timeline->requests)) {
+ struct i915_request *rq;
+
+ rq = list_last_entry(&ce->timeline->requests,
+ typeof(*rq),
+ link);
+
+ err = i915_sw_fence_await_dma_fence(&engines->fence,
+ &rq->fence, 0,
+ GFP_KERNEL);
+ }
+ mutex_unlock(&ce->timeline->mutex);
+ if (err < 0)
+ goto kill;
+ }
+
+ spin_lock_irq(&engines->ctx->stale.lock);
+ if (!i915_gem_context_is_closed(engines->ctx))
+ list_add_tail(&engines->link, &engines->ctx->stale.engines);
+ spin_unlock_irq(&engines->ctx->stale.lock);
+
+kill:
+ if (list_empty(&engines->link)) /* raced, already closed */
+ kill_engines(engines);
+
+ i915_sw_fence_commit(&engines->fence);
}
static void set_closed_name(struct i915_gem_context *ctx)
@@ -559,6 +634,11 @@ static void context_close(struct i915_gem_context *ctx)
mutex_unlock(&ctx->mutex);
+ /* Flush any concurrent set_engines() */
+ mutex_lock(&ctx->engines_mutex);
+ engines_idle_release(rcu_replace_pointer(ctx->engines, NULL, 1));
+ mutex_unlock(&ctx->engines_mutex);
+
/*
* If the user has disabled hangchecking, we can not be sure that
* the batches will ever complete after the context is closed,
@@ -1562,77 +1642,6 @@ static const i915_user_extension_fn set_engines__extensions[] = {
[I915_CONTEXT_ENGINES_EXT_BOND] = set_engines__bond,
};
-static int engines_notify(struct i915_sw_fence *fence,
- enum i915_sw_fence_notify state)
-{
- struct i915_gem_engines *engines =
- container_of(fence, typeof(*engines), fence);
-
- switch (state) {
- case FENCE_COMPLETE:
- if (!list_empty(&engines->link)) {
- struct i915_gem_context *ctx = engines->ctx;
- unsigned long flags;
-
- spin_lock_irqsave(&ctx->stale.lock, flags);
- list_del(&engines->link);
- spin_unlock_irqrestore(&ctx->stale.lock, flags);
- }
- break;
-
- case FENCE_FREE:
- init_rcu_head(&engines->rcu);
- call_rcu(&engines->rcu, free_engines_rcu);
- break;
- }
-
- return NOTIFY_DONE;
-}
-
-static void engines_idle_release(struct i915_gem_engines *engines)
-{
- struct i915_gem_engines_iter it;
- struct intel_context *ce;
- unsigned long flags;
-
- GEM_BUG_ON(!engines);
- i915_sw_fence_init(&engines->fence, engines_notify);
-
- INIT_LIST_HEAD(&engines->link);
- spin_lock_irqsave(&engines->ctx->stale.lock, flags);
- if (!i915_gem_context_is_closed(engines->ctx))
- list_add(&engines->link, &engines->ctx->stale.engines);
- spin_unlock_irqrestore(&engines->ctx->stale.lock, flags);
- if (list_empty(&engines->link)) /* raced, already closed */
- goto kill;
-
- for_each_gem_engine(ce, engines, it) {
- struct dma_fence *fence;
- int err;
-
- if (!ce->timeline)
- continue;
-
- fence = i915_active_fence_get(&ce->timeline->last_request);
- if (!fence)
- continue;
-
- err = i915_sw_fence_await_dma_fence(&engines->fence,
- fence, 0,
- GFP_KERNEL);
-
- dma_fence_put(fence);
- if (err < 0)
- goto kill;
- }
- goto out;
-
-kill:
- kill_engines(engines);
-out:
- i915_sw_fence_commit(&engines->fence);
-}
-
static int
set_engines(struct i915_gem_context *ctx,
const struct drm_i915_gem_context_param *args)
@@ -1729,6 +1738,11 @@ set_engines(struct i915_gem_context *ctx,
replace:
mutex_lock(&ctx->engines_mutex);
+ if (i915_gem_context_is_closed(ctx)) {
+ mutex_unlock(&ctx->engines_mutex);
+ free_engines(set.engines);
+ return -ENOENT;
+ }
if (args->size)
i915_gem_context_set_user_engines(ctx);
else
@@ -2033,8 +2047,7 @@ static int clone_engines(struct i915_gem_context *dst,
i915_gem_context_unlock_engines(src);
/* Serialised by constructor */
- free_engines(__context_engines_static(dst));
- RCU_INIT_POINTER(dst->engines, clone);
+ engines_idle_release(rcu_replace_pointer(dst->engines, clone, 1));
if (user_engines)
i915_gem_context_set_user_engines(dst);
else
--
2.25.0
More information about the Intel-gfx
mailing list