#include "i915_drv.h"
+static void intel_breadcrumbs_hangcheck(unsigned long data)
+{
+ struct intel_engine_cs *engine = (struct intel_engine_cs *)data;
+ struct intel_breadcrumbs *b = &engine->breadcrumbs;
+
+ if (!b->irq_enabled)
+ return;
+
+ if (time_before(jiffies, b->timeout)) {
+ mod_timer(&b->hangcheck, b->timeout);
+ return;
+ }
+
+ DRM_DEBUG("Hangcheck timer elapsed... %s idle\n", engine->name);
+ set_bit(engine->id, &engine->i915->gpu_error.missed_irq_rings);
+ mod_timer(&engine->breadcrumbs.fake_irq, jiffies + 1);
+
+ /* Ensure that even if the GPU hangs, we get woken up.
+ *
+ * However, note that if no one is waiting, we never notice
+ * a gpu hang. Eventually, we will have to wait for a resource
+ * held by the GPU and so trigger a hangcheck. In the most
+ * pathological case, this will be upon memory starvation! To
+ * prevent this, we also queue the hangcheck from the retire
+ * worker.
+ */
+ i915_queue_hangcheck(engine->i915);
+}
+
+static unsigned long wait_timeout(void)
+{
+ return round_jiffies_up(jiffies + DRM_I915_HANGCHECK_JIFFIES);
+}
+
static void intel_breadcrumbs_fake_irq(unsigned long data)
{
struct intel_engine_cs *engine = (struct intel_engine_cs *)data;
* every jiffie in order to kick the oldest waiter to do the
* coherent seqno check.
*/
- rcu_read_lock();
if (intel_engine_wakeup(engine))
mod_timer(&engine->breadcrumbs.fake_irq, jiffies + 1);
- rcu_read_unlock();
}
static void irq_enable(struct intel_engine_cs *engine)
}
if (!b->irq_enabled ||
- test_bit(engine->id, &i915->gpu_error.missed_irq_rings))
+ test_bit(engine->id, &i915->gpu_error.missed_irq_rings)) {
mod_timer(&b->fake_irq, jiffies + 1);
-
- /* Ensure that even if the GPU hangs, we get woken up.
- *
- * However, note that if no one is waiting, we never notice
- * a gpu hang. Eventually, we will have to wait for a resource
- * held by the GPU and so trigger a hangcheck. In the most
- * pathological case, this will be upon memory starvation!
- */
- i915_queue_hangcheck(i915);
+ } else {
+ /* Ensure we never sleep indefinitely */
+ GEM_BUG_ON(!time_after(b->timeout, jiffies));
+ mod_timer(&b->hangcheck, b->timeout);
+ }
}
static void __intel_breadcrumbs_disable_irq(struct intel_breadcrumbs *b)
}
rb_link_node(&wait->node, parent, p);
rb_insert_color(&wait->node, &b->waiters);
- GEM_BUG_ON(!first && !b->irq_seqno_bh);
+ GEM_BUG_ON(!first && !rcu_access_pointer(b->irq_seqno_bh));
if (completed) {
struct rb_node *next = rb_next(completed);
GEM_BUG_ON(!next && !first);
if (next && next != &wait->node) {
GEM_BUG_ON(first);
+ b->timeout = wait_timeout();
b->first_wait = to_wait(next);
- smp_store_mb(b->irq_seqno_bh, b->first_wait->tsk);
+ rcu_assign_pointer(b->irq_seqno_bh, b->first_wait->tsk);
/* As there is a delay between reading the current
* seqno, processing the completed tasks and selecting
* the next waiter, we may have missed the interrupt
if (first) {
GEM_BUG_ON(rb_first(&b->waiters) != &wait->node);
+ b->timeout = wait_timeout();
b->first_wait = wait;
- smp_store_mb(b->irq_seqno_bh, wait->tsk);
+ rcu_assign_pointer(b->irq_seqno_bh, wait->tsk);
/* After assigning ourselves as the new bottom-half, we must
* perform a cursory check to prevent a missed interrupt.
* Either we miss the interrupt whilst programming the hardware,
*/
__intel_breadcrumbs_enable_irq(b);
}
- GEM_BUG_ON(!b->irq_seqno_bh);
+ GEM_BUG_ON(!rcu_access_pointer(b->irq_seqno_bh));
GEM_BUG_ON(!b->first_wait);
GEM_BUG_ON(rb_first(&b->waiters) != &b->first_wait->node);
return first;
}
-void intel_engine_enable_fake_irq(struct intel_engine_cs *engine)
-{
- mod_timer(&engine->breadcrumbs.fake_irq, jiffies + 1);
-}
-
static inline bool chain_wakeup(struct rb_node *rb, int priority)
{
return rb && to_wait(rb)->tsk->prio <= priority;
const int priority = wakeup_priority(b, wait->tsk);
struct rb_node *next;
- GEM_BUG_ON(b->irq_seqno_bh != wait->tsk);
+ GEM_BUG_ON(rcu_access_pointer(b->irq_seqno_bh) != wait->tsk);
/* We are the current bottom-half. Find the next candidate,
* the first waiter in the queue on the remaining oldest
* the interrupt, or if we have to handle an
* exception rather than a seqno completion.
*/
+ b->timeout = wait_timeout();
b->first_wait = to_wait(next);
- smp_store_mb(b->irq_seqno_bh, b->first_wait->tsk);
+ rcu_assign_pointer(b->irq_seqno_bh, b->first_wait->tsk);
if (b->first_wait->seqno != wait->seqno)
__intel_breadcrumbs_enable_irq(b);
- wake_up_process(b->irq_seqno_bh);
+ wake_up_process(b->first_wait->tsk);
} else {
b->first_wait = NULL;
- WRITE_ONCE(b->irq_seqno_bh, NULL);
+ rcu_assign_pointer(b->irq_seqno_bh, NULL);
__intel_breadcrumbs_disable_irq(b);
}
} else {
GEM_BUG_ON(b->first_wait == wait);
GEM_BUG_ON(rb_first(&b->waiters) !=
(b->first_wait ? &b->first_wait->node : NULL));
- GEM_BUG_ON(!b->irq_seqno_bh ^ RB_EMPTY_ROOT(&b->waiters));
+ GEM_BUG_ON(!rcu_access_pointer(b->irq_seqno_bh) ^ RB_EMPTY_ROOT(&b->waiters));
spin_unlock(&b->lock);
}
intel_engine_remove_wait(engine,
&request->signaling.wait);
+ local_bh_disable();
+ fence_signal(&request->fence);
+ local_bh_enable(); /* kick start the tasklets */
+
/* Find the next oldest signal. Note that as we have
* not been holding the lock, another client may
* have installed an even older signal than the one
rb_erase(&request->signaling.node, &b->signals);
spin_unlock(&b->lock);
- i915_gem_request_unreference(request);
+ i915_gem_request_put(request);
} else {
if (kthread_should_stop())
break;
struct rb_node *parent, **p;
bool first, wakeup;
- if (unlikely(READ_ONCE(request->signaling.wait.tsk)))
- return;
-
- spin_lock(&b->lock);
- if (unlikely(request->signaling.wait.tsk)) {
- wakeup = false;
- goto unlock;
- }
+ /* locked by fence_enable_sw_signaling() */
+ assert_spin_locked(&request->lock);
request->signaling.wait.tsk = b->signaler;
- request->signaling.wait.seqno = request->seqno;
- i915_gem_request_reference(request);
+ request->signaling.wait.seqno = request->fence.seqno;
+ i915_gem_request_get(request);
+
+ spin_lock(&b->lock);
/* First add ourselves into the list of waiters, but register our
* bottom-half as the signaller thread. As per usual, only the oldest
p = &b->signals.rb_node;
while (*p) {
parent = *p;
- if (i915_seqno_passed(request->seqno,
- to_signaler(parent)->seqno)) {
+ if (i915_seqno_passed(request->fence.seqno,
+ to_signaler(parent)->fence.seqno)) {
p = &parent->rb_right;
first = false;
} else {
if (first)
smp_store_mb(b->first_signal, request);
-unlock:
spin_unlock(&b->lock);
if (wakeup)
setup_timer(&b->fake_irq,
intel_breadcrumbs_fake_irq,
(unsigned long)engine);
+ setup_timer(&b->hangcheck,
+ intel_breadcrumbs_hangcheck,
+ (unsigned long)engine);
/* Spawn a thread to provide a common bottom-half for all signals.
* As this is an asynchronous interface we cannot steal the current
return 0;
}
+static void cancel_fake_irq(struct intel_engine_cs *engine)
+{
+ struct intel_breadcrumbs *b = &engine->breadcrumbs;
+
+ del_timer_sync(&b->hangcheck);
+ del_timer_sync(&b->fake_irq);
+ clear_bit(engine->id, &engine->i915->gpu_error.missed_irq_rings);
+}
+
+void intel_engine_reset_breadcrumbs(struct intel_engine_cs *engine)
+{
+ struct intel_breadcrumbs *b = &engine->breadcrumbs;
+
+ cancel_fake_irq(engine);
+ spin_lock(&b->lock);
+
+ __intel_breadcrumbs_disable_irq(b);
+ if (intel_engine_has_waiter(engine)) {
+ b->timeout = wait_timeout();
+ __intel_breadcrumbs_enable_irq(b);
+ if (READ_ONCE(b->irq_posted))
+ wake_up_process(b->first_wait->tsk);
+ } else {
+ /* sanitize the IMR and unmask any auxiliary interrupts */
+ irq_disable(engine);
+ }
+
+ spin_unlock(&b->lock);
+}
+
void intel_engine_fini_breadcrumbs(struct intel_engine_cs *engine)
{
struct intel_breadcrumbs *b = &engine->breadcrumbs;
if (!IS_ERR_OR_NULL(b->signaler))
kthread_stop(b->signaler);
- del_timer_sync(&b->fake_irq);
+ cancel_fake_irq(engine);
}
unsigned int intel_kick_waiters(struct drm_i915_private *i915)
* RCU lock, i.e. as we call wake_up_process() we must be holding the
* rcu_read_lock().
*/
- rcu_read_lock();
for_each_engine(engine, i915)
if (unlikely(intel_engine_wakeup(engine)))
mask |= intel_engine_flag(engine);
- rcu_read_unlock();
return mask;
}
projects / cascardo / linux.git / blobdiff