static int semaphore_passed(struct intel_engine_cs *engine)
{
struct drm_i915_private *dev_priv = engine->i915;
struct intel_engine_cs *signaller;
u32 seqno;
engine->hangcheck.deadlock++;
signaller = semaphore_waits_for(engine, &seqno);
if (signaller == NULL)
return -1;
if (IS_ERR(signaller))
return 0;
/* Prevent pathological recursion due to driver bugs */
if (signaller->hangcheck.deadlock >= I915_NUM_ENGINES)
return -1;
if (i915_seqno_passed(intel_engine_get_seqno(signaller), seqno))
return 1;
/* cursory check for an unkickable deadlock */
if (I915_READ_CTL(signaller) & RING_WAIT_SEMAPHORE &&
semaphore_passed(signaller) < 0)
return -1;
return 0;
}
static int igt_wakeup_thread(void *arg)
{
struct igt_wakeup *w = arg;
struct intel_wait wait;
while (wait_for_ready(w)) {
GEM_BUG_ON(kthread_should_stop());
intel_wait_init_for_seqno(&wait, w->seqno);
intel_engine_add_wait(w->engine, &wait);
for (;;) {
set_current_state(TASK_UNINTERRUPTIBLE);
if (i915_seqno_passed(intel_engine_get_seqno(w->engine),
w->seqno))
break;
if (test_bit(STOP, &w->flags)) /* emergency escape */
break;
schedule();
}
intel_engine_remove_wait(w->engine, &wait);
__set_current_state(TASK_RUNNING);
}
return 0;
}
static int __igt_reset_engine(struct drm_i915_private *i915, bool active)
{
struct intel_engine_cs *engine;
enum intel_engine_id id;
struct hang h;
int err = 0;
/* Check that we can issue an engine reset on an idle engine (no-op) */
if (!intel_has_reset_engine(i915))
return 0;
if (active) {
mutex_lock(&i915->drm.struct_mutex);
err = hang_init(&h, i915);
mutex_unlock(&i915->drm.struct_mutex);
if (err)
return err;
}
for_each_engine(engine, i915, id) {
unsigned int reset_count, reset_engine_count;
IGT_TIMEOUT(end_time);
if (active && !intel_engine_can_store_dword(engine))
continue;
if (!wait_for_idle(engine)) {
pr_err("%s failed to idle before reset\n",
engine->name);
err = -EIO;
break;
}
reset_count = i915_reset_count(&i915->gpu_error);
reset_engine_count = i915_reset_engine_count(&i915->gpu_error,
engine);
set_bit(I915_RESET_ENGINE + id, &i915->gpu_error.flags);
do {
u32 seqno = intel_engine_get_seqno(engine);
if (active) {
struct i915_request *rq;
mutex_lock(&i915->drm.struct_mutex);
rq = hang_create_request(&h, engine);
if (IS_ERR(rq)) {
err = PTR_ERR(rq);
mutex_unlock(&i915->drm.struct_mutex);
break;
}
i915_request_get(rq);
i915_request_add(rq);
mutex_unlock(&i915->drm.struct_mutex);
if (!wait_until_running(&h, rq)) {
struct drm_printer p = drm_info_printer(i915->drm.dev);
pr_err("%s: Failed to start request %x, at %x\n",
__func__, rq->fence.seqno, hws_seqno(&h, rq));
intel_engine_dump(engine, &p,
"%s\n", engine->name);
i915_request_put(rq);
err = -EIO;
break;
}
GEM_BUG_ON(!rq->global_seqno);
seqno = rq->global_seqno - 1;
i915_request_put(rq);
}
err = i915_reset_engine(engine, NULL);
if (err) {
pr_err("i915_reset_engine failed\n");
break;
}
if (i915_reset_count(&i915->gpu_error) != reset_count) {
pr_err("Full GPU reset recorded! (engine reset expected)\n");
err = -EINVAL;
break;
}
reset_engine_count += active;
if (i915_reset_engine_count(&i915->gpu_error, engine) !=
reset_engine_count) {
pr_err("%s engine reset %srecorded!\n",
engine->name, active ? "not " : "");
err = -EINVAL;
break;
}
if (!wait_for_idle(engine)) {
struct drm_printer p =
drm_info_printer(i915->drm.dev);
pr_err("%s failed to idle after reset\n",
engine->name);
intel_engine_dump(engine, &p,
"%s\n", engine->name);
err = -EIO;
break;
}
} while (time_before(jiffies, end_time));
clear_bit(I915_RESET_ENGINE + id, &i915->gpu_error.flags);
if (err)
break;
err = igt_flush_test(i915, 0);
if (err)
break;
}
static int igt_wakeup(void *arg)
{
I915_RND_STATE(prng);
struct intel_engine_cs *engine = arg;
struct igt_wakeup *waiters;
DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
const int count = 4096;
const u32 max_seqno = count / 4;
atomic_t ready, set, done;
int err = -ENOMEM;
int n, step;
mock_engine_reset(engine);
waiters = kvmalloc_array(count, sizeof(*waiters), GFP_KERNEL);
if (!waiters)
goto out_engines;
/* Create a large number of threads, each waiting on a random seqno.
* Multiple waiters will be waiting for the same seqno.
*/
atomic_set(&ready, count);
for (n = 0; n < count; n++) {
waiters[n].wq = &wq;
waiters[n].ready = &ready;
waiters[n].set = &set;
waiters[n].done = &done;
waiters[n].engine = engine;
waiters[n].flags = BIT(IDLE);
waiters[n].tsk = kthread_run(igt_wakeup_thread, &waiters[n],
"i915/igt:%d", n);
if (IS_ERR(waiters[n].tsk))
goto out_waiters;
get_task_struct(waiters[n].tsk);
}
for (step = 1; step <= max_seqno; step <<= 1) {
u32 seqno;
/* The waiter threads start paused as we assign them a random
* seqno and reset the engine. Once the engine is reset,
* we signal that the threads may begin their wait upon their
* seqno.
*/
for (n = 0; n < count; n++) {
GEM_BUG_ON(!test_bit(IDLE, &waiters[n].flags));
waiters[n].seqno =
1 + prandom_u32_state(&prng) % max_seqno;
}
mock_seqno_advance(engine, 0);
igt_wake_all_sync(&ready, &set, &done, &wq, count);
/* Simulate the GPU doing chunks of work, with one or more
* seqno appearing to finish at the same time. A random number
* of threads will be waiting upon the update and hopefully be
* woken.
*/
for (seqno = 1; seqno <= max_seqno + step; seqno += step) {
usleep_range(50, 500);
mock_seqno_advance(engine, seqno);
}
GEM_BUG_ON(intel_engine_get_seqno(engine) < 1 + max_seqno);
/* With the seqno now beyond any of the waiting threads, they
* should all be woken, see that they are complete and signal
* that they are ready for the next test. We wait until all
* threads are complete and waiting for us (i.e. not a seqno).
*/
if (!wait_var_event_timeout(&done,
!atomic_read(&done), 10 * HZ)) {
pr_err("Timed out waiting for %d remaining waiters\n",
atomic_read(&done));
err = -ETIMEDOUT;
break;
}
err = check_rbtree_empty(engine);
if (err)
break;
}
out_waiters:
for (n = 0; n < count; n++) {
if (IS_ERR(waiters[n].tsk))
break;
set_bit(STOP, &waiters[n].flags);
}
mock_seqno_advance(engine, INT_MAX); /* wakeup any broken waiters */
igt_wake_all_sync(&ready, &set, &done, &wq, n);
for (n = 0; n < count; n++) {
if (IS_ERR(waiters[n].tsk))
break;
kthread_stop(waiters[n].tsk);
put_task_struct(waiters[n].tsk);
}
kvfree(waiters);
out_engines:
mock_engine_flush(engine);
return err;
}