[PATCH 42/42] drm/i915/execlists: Minimalistic timeslicing
Chris Wilson
chris at chris-wilson.co.uk
Thu Apr 25 05:54:13 UTC 2019
If we have multiple contexts of equal priority pending execution,
activate a timer to demote the currently executing context in favour of
the next in the queue when that timeslice expires. This enforces
fairness between contexts (so long as they allow preemption -- forced
preemption, in the future, will kick those who do not obey) and allows
us to avoid userspace blocking forward progress with e.g. unbounded
MI_SEMAPHORE_WAIT.
For the starting point here, we use the jiffie as our timeslice so that
we should be reasonably efficient wrt frequent CPU wakeups.
Testcase: igt/gem_exec_scheduler/semaphore-resolve
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
---
drivers/gpu/drm/i915/gt/intel_engine_types.h | 6 +
drivers/gpu/drm/i915/gt/intel_lrc.c | 134 ++++++++++++++++---
drivers/gpu/drm/i915/gt/selftest_lrc.c | 3 +
drivers/gpu/drm/i915/i915_scheduler.c | 1 +
drivers/gpu/drm/i915/i915_scheduler_types.h | 1 +
5 files changed, 128 insertions(+), 17 deletions(-)
diff --git a/drivers/gpu/drm/i915/gt/intel_engine_types.h b/drivers/gpu/drm/i915/gt/intel_engine_types.h
index 1a3c8e2bf93e..25aa9edeb057 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine_types.h
+++ b/drivers/gpu/drm/i915/gt/intel_engine_types.h
@@ -12,6 +12,7 @@
#include <linux/kref.h>
#include <linux/list.h>
#include <linux/llist.h>
+#include <linux/timer.h>
#include <linux/types.h>
#include "i915_gem.h"
@@ -137,6 +138,11 @@ struct intel_engine_execlists {
*/
struct tasklet_struct tasklet;
+ /**
+ * @timer: kick the current context if its timeslice expires
+ */
+ struct timer_list timer;
+
/**
* @default_priolist: priority list for I915_PRIORITY_NORMAL
*/
diff --git a/drivers/gpu/drm/i915/gt/intel_lrc.c b/drivers/gpu/drm/i915/gt/intel_lrc.c
index 67b4e62bf952..d436662b6f89 100644
--- a/drivers/gpu/drm/i915/gt/intel_lrc.c
+++ b/drivers/gpu/drm/i915/gt/intel_lrc.c
@@ -833,6 +833,80 @@ last_active(const struct intel_engine_execlists *execlists)
return *last;
}
+static void
+defer_request(struct i915_request * const rq, struct list_head * const pl)
+{
+ struct i915_dependency *p;
+
+ /*
+ * We want to move the interrupted request to the back of
+ * the round-robin list (i.e. its priority level), but
+ * in doing so, we must then move all requests that were in
+ * flight and were waiting for the interrupted request to
+ * after it again.
+ */
+ list_move_tail(&rq->sched.link, pl);
+
+ list_for_each_entry(p, &rq->sched.waiters_list, wait_link) {
+ struct i915_request *w =
+ container_of(p->waiter, typeof(*w), sched);
+
+ if (!i915_sw_fence_done(&w->submit))
+ continue;
+ /* Leave semaphores spinning on the other engines */
+ if (w->engine != rq->engine)
+ continue;
+
+ /* No waiter should start before the active request completed */
+ GEM_BUG_ON(i915_request_started(w));
+
+ GEM_BUG_ON(rq_prio(w) > rq_prio(rq));
+ if (rq_prio(w) < rq_prio(rq))
+ continue;
+
+ /*
+ * This should be very shallow as it is limited by the
+ * number of requests that can fit in a ring (<64) and
+ * the number of contexts that can be in flight.
+ */
+ defer_request(w, pl);
+ }
+}
+
+static void defer_active(struct intel_engine_cs *engine)
+{
+ struct i915_request *rq;
+
+ rq = __unwind_incomplete_requests(engine);
+ if (rq) {
+ defer_request(rq,
+ i915_sched_lookup_priolist(engine, rq_prio(rq)));
+ }
+}
+
+static bool
+need_timeslice(struct intel_engine_cs *engine, const struct i915_request *rq)
+{
+ int hint;
+
+ if (list_is_last(&rq->sched.link, &engine->active.requests))
+ return false;
+
+ hint = max(rq_prio(list_next_entry(rq, sched.link)),
+ queue_prio(&engine->execlists));
+ hint |= __NO_PREEMPTION;
+
+ return hint >= rq_prio(rq);
+}
+
+static bool
+enable_timeslice(struct intel_engine_cs *engine)
+{
+ struct i915_request *last = last_active(&engine->execlists);
+
+ return last && need_timeslice(engine, last);
+}
+
static bool execlists_dequeue(struct intel_engine_cs *engine)
{
struct intel_engine_execlists * const execlists = &engine->execlists;
@@ -926,6 +1000,27 @@ static bool execlists_dequeue(struct intel_engine_cs *engine)
*/
last->hw_context->lrc_desc |= CTX_DESC_FORCE_RESTORE;
last = NULL;
+ } else if (need_timeslice(engine, last) &&
+ !timer_pending(&engine->execlists.timer)) {
+ GEM_TRACE("%s: expired last=%llx:%lld, prio=%d, hint=%d\n",
+ engine->name,
+ last->fence.context,
+ last->fence.seqno,
+ last->sched.attr.priority,
+ execlists->queue_priority_hint);
+
+ ring_suspend(engine) = 1;
+ defer_active(engine);
+
+ /*
+ * Unlike for preemption, if we rewind and continue
+ * executing the same context as previously active,
+ * the order of execution will remain the same and
+ * the tail will only advance. We do not need to
+ * force a full context restore, as a lite-restore
+ * is sufficient to resample the monotonic TAIL.
+ */
+ last = NULL;
} else {
/*
* Otherwise if we already have a request pending
@@ -1247,6 +1342,9 @@ static void process_csb(struct intel_engine_cs *engine)
sizeof(*execlists->pending));
execlists->pending[0] = NULL;
+ if (enable_timeslice(engine))
+ mod_timer(&execlists->timer, jiffies + 1);
+
if (!inject_preempt_hang(execlists))
ring_suspend(engine) = 0;
} else if (status & GEN8_CTX_STATUS_PREEMPTED) {
@@ -1323,6 +1421,15 @@ static void execlists_submission_tasklet(unsigned long data)
spin_unlock_irqrestore(&engine->active.lock, flags);
}
+static void execlists_submission_timer(struct timer_list *timer)
+{
+ struct intel_engine_cs *engine =
+ from_timer(engine, timer, execlists.timer);
+
+ /* Kick the tasklet for some interrupt coalescing and reset handling */
+ tasklet_hi_schedule(&engine->execlists.tasklet);
+}
+
static void queue_request(struct intel_engine_cs *engine,
struct i915_sched_node *node,
int prio)
@@ -2556,27 +2663,19 @@ static int gen8_init_rcs_context(struct i915_request *rq)
return i915_gem_render_state_emit(rq);
}
+static void execlists_park(struct intel_engine_cs *engine)
+{
+ del_timer_sync(&engine->execlists.timer);
+ tasklet_kill(&engine->execlists.tasklet);
+}
+
/**
* intel_logical_ring_cleanup() - deallocate the Engine Command Streamer
* @engine: Engine Command Streamer.
*/
void intel_logical_ring_cleanup(struct intel_engine_cs *engine)
{
- struct drm_i915_private *dev_priv;
-
- /*
- * Tasklet cannot be active at this point due intel_mark_active/idle
- * so this is just for documentation.
- */
- if (WARN_ON(test_bit(TASKLET_STATE_SCHED,
- &engine->execlists.tasklet.state)))
- tasklet_kill(&engine->execlists.tasklet);
-
- dev_priv = engine->i915;
-
- if (engine->buffer) {
- WARN_ON((ENGINE_READ(engine, RING_MI_MODE) & MODE_IDLE) == 0);
- }
+ struct drm_i915_private *i915 = engine->i915;
if (engine->cleanup)
engine->cleanup(engine);
@@ -2586,7 +2685,7 @@ void intel_logical_ring_cleanup(struct intel_engine_cs *engine)
lrc_destroy_wa_ctx(engine);
engine->i915 = NULL;
- dev_priv->engine[engine->id] = NULL;
+ i915->engine[engine->id] = NULL;
kfree(engine);
}
@@ -2601,7 +2700,7 @@ void intel_execlists_set_default_submission(struct intel_engine_cs *engine)
engine->reset.reset = execlists_reset;
engine->reset.finish = execlists_reset_finish;
- engine->park = NULL;
+ engine->park = execlists_park;
engine->unpark = NULL;
engine->flags |= I915_ENGINE_SUPPORTS_STATS;
@@ -2675,6 +2774,7 @@ int intel_execlists_submission_setup(struct intel_engine_cs *engine)
tasklet_init(&engine->execlists.tasklet,
execlists_submission_tasklet, (unsigned long)engine);
+ timer_setup(&engine->execlists.timer, execlists_submission_timer, 0);
logical_ring_default_vfuncs(engine);
logical_ring_default_irqs(engine);
diff --git a/drivers/gpu/drm/i915/gt/selftest_lrc.c b/drivers/gpu/drm/i915/gt/selftest_lrc.c
index 3e9bbcb45fa2..19459b950c1b 100644
--- a/drivers/gpu/drm/i915/gt/selftest_lrc.c
+++ b/drivers/gpu/drm/i915/gt/selftest_lrc.c
@@ -396,6 +396,9 @@ static int live_late_preempt(void *arg)
if (!ctx_lo)
goto err_ctx_hi;
+ /* Make sure ctx_lo stays before ctx_hi until we trigger preemption. */
+ ctx_lo->sched.priority = I915_USER_PRIORITY(1);
+
for_each_engine(engine, i915, id) {
struct igt_live_test t;
struct i915_request *rq;
diff --git a/drivers/gpu/drm/i915/i915_scheduler.c b/drivers/gpu/drm/i915/i915_scheduler.c
index 601aae909491..7d0c14a5e687 100644
--- a/drivers/gpu/drm/i915/i915_scheduler.c
+++ b/drivers/gpu/drm/i915/i915_scheduler.c
@@ -71,6 +71,7 @@ bool __i915_sched_node_add_dependency(struct i915_sched_node *node,
list_add(&dep->wait_link, &signal->waiters_list);
list_add(&dep->signal_link, &node->signalers_list);
dep->signaler = signal;
+ dep->waiter = node;
dep->flags = flags;
/* Keep track of whether anyone on this chain has a semaphore */
diff --git a/drivers/gpu/drm/i915/i915_scheduler_types.h b/drivers/gpu/drm/i915/i915_scheduler_types.h
index 166a457884b2..21fb9cd81fcb 100644
--- a/drivers/gpu/drm/i915/i915_scheduler_types.h
+++ b/drivers/gpu/drm/i915/i915_scheduler_types.h
@@ -62,6 +62,7 @@ struct i915_sched_node {
struct i915_dependency {
struct i915_sched_node *signaler;
+ struct i915_sched_node *waiter;
struct list_head signal_link;
struct list_head wait_link;
struct list_head dfs_link;
--
2.20.1
More information about the Intel-gfx-trybot
mailing list