Execlists uses a scheduling quantum (a timeslice) to alternate execution
between ready-to-run contexts of equal priority. This ensures that all
users (though only if they of equal importance) have the opportunity to
run and prevents livelocks where contexts may have implicit ordering due
to userspace semaphores.
The timeslicing mechanism can be compiled out with
./scripts/config --set-val DRM_I915_TIMESLICE_DURATION 0
The timeslice duration can be adjusted per-engine using,
/sys/class/drm/card?/engine/*/timeslice_duration_ms
Signed-off-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
Cc: Joonas Lahtinen <joonas.lahtinen@xxxxxxxxxxxxxxx>
---
drivers/gpu/drm/i915/Kconfig.profile | 18 ++++++
drivers/gpu/drm/i915/gt/intel_engine.h | 9 +++
drivers/gpu/drm/i915/gt/intel_engine_cs.c | 3 +
drivers/gpu/drm/i915/gt/intel_engine_sysfs.c | 46 +++++++++++++++
drivers/gpu/drm/i915/gt/intel_engine_types.h | 4 ++
drivers/gpu/drm/i915/gt/intel_lrc.c | 59 ++++++++++++++++----
drivers/gpu/drm/i915/gt/selftest_lrc.c | 13 ++++-
7 files changed, 138 insertions(+), 14 deletions(-)
diff --git a/drivers/gpu/drm/i915/Kconfig.profile b/drivers/gpu/drm/i915/Kconfig.profile
index 48df8889a88a..b8df80bc0b47 100644
--- a/drivers/gpu/drm/i915/Kconfig.profile
+++ b/drivers/gpu/drm/i915/Kconfig.profile
@@ -25,3 +25,21 @@ config DRM_I915_SPIN_REQUEST
May be 0 to disable the initial spin. In practice, we estimate
the cost of enabling the interrupt (if currently disabled) to be
a few microseconds.
+
+config DRM_I915_TIMESLICE_DURATION
+ int "Scheduling quantum for userspace batches (ms, jiffy granularity)"
+ default 1 # milliseconds
+ help
+ When two user batches of equal priority are executing, we will
+ alternate execution of each batch to ensure forward progress of
+ all users. This is necessary in some cases where there may be
+ an implicit dependency between those batches that requires
+ concurrent execution in order for them to proceed, e.g. they
+ interact with each other via userspace semaphores. Each context
+ is scheduled for execution for the timeslice duration, before
+ switching to the next context.
+
+ This is adjustable via
+ /sys/class/drm/card?/engine/*/timeslice_duration_ms
+
+ May be 0 to disable timeslicing.
diff --git a/drivers/gpu/drm/i915/gt/intel_engine.h b/drivers/gpu/drm/i915/gt/intel_engine.h
index 93ea367fe624..393481374f96 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine.h
+++ b/drivers/gpu/drm/i915/gt/intel_engine.h
@@ -525,4 +525,13 @@ void intel_engine_init_active(struct intel_engine_cs *engine,
#define ENGINE_MOCK 1
#define ENGINE_VIRTUAL 2
+static inline bool
+intel_engine_has_timeslices(const struct intel_engine_cs *engine)
+{
+ if (!CONFIG_DRM_I915_TIMESLICE_DURATION)
+ return 0;
+
+ return intel_engine_has_semaphores(engine);
+}
+
#endif /* _INTEL_RINGBUFFER_H_ */
diff --git a/drivers/gpu/drm/i915/gt/intel_engine_cs.c b/drivers/gpu/drm/i915/gt/intel_engine_cs.c
index 051734c9b733..b93dbed67fe7 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine_cs.c
+++ b/drivers/gpu/drm/i915/gt/intel_engine_cs.c
@@ -308,6 +308,9 @@ static int intel_engine_setup(struct intel_gt *gt, enum intel_engine_id id)
engine->instance = info->instance;
__sprint_engine_name(engine);
+ engine->props.timeslice_duration_ms =
+ CONFIG_DRM_I915_TIMESLICE_DURATION;
+
/*
* To be overridden by the backend on setup. However to facilitate
* cleanup on error during setup, we always provide the destroy vfunc.
diff --git a/drivers/gpu/drm/i915/gt/intel_engine_sysfs.c b/drivers/gpu/drm/i915/gt/intel_engine_sysfs.c
index 557ef1e2f84a..55ae81769a8e 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine_sysfs.c
+++ b/drivers/gpu/drm/i915/gt/intel_engine_sysfs.c
@@ -142,6 +142,48 @@ all_caps_show(struct kobject *kobj, struct kobj_attribute *attr, char *buf)
static struct kobj_attribute all_caps_attr =
__ATTR(known_capabilities, 0444, all_caps_show, NULL);
+static ssize_t
+timeslice_show(struct kobject *kobj, struct kobj_attribute *attr, char *buf)
+{
+ struct intel_engine_cs *engine = kobj_to_engine(kobj);
+
+ return sprintf(buf, "%lu\n", engine->props.timeslice_duration_ms);
+}
+
+static ssize_t
+timeslice_store(struct kobject *kobj, struct kobj_attribute *attr,
+ const char *buf, size_t count)
+{
+ struct intel_engine_cs *engine = kobj_to_engine(kobj);
+ unsigned long long duration;
+ int err;
+
+ err = kstrtoull(buf, 0, &duration);
+ if (err)
+ return err;
+
+ if (duration > jiffies_to_msecs(MAX_SCHEDULE_TIMEOUT))
+ return -EINVAL;
+
+ WRITE_ONCE(engine->props.timeslice_duration_ms, duration);
+
+ if (execlists_active(&engine->execlists)) {
+ struct timer_list *t = &engine->execlists.timer;
+
+ if (!duration) {
+ cancel_timer(t);
+ } else {
+ duration = msecs_to_jiffies_timeout(duration);
+ mod_timer(t, jiffies + duration);
+ }
+ }
+
+ return count;
+}
+
+static struct kobj_attribute timeslice_duration_attr =
+__ATTR(timeslice_duration_ms, 0644, timeslice_show, timeslice_store);
+
static void kobj_engine_release(struct kobject *kobj)
{
kfree(kobj);
@@ -203,6 +245,10 @@ void intel_engines_add_sysfs(struct drm_i915_private *i915)
if (sysfs_create_files(kobj, files))
goto err_object;
+ if (intel_engine_has_timeslices(engine) &&
+ sysfs_create_file(kobj, ×lice_duration_attr.attr))
+ goto err_engine;
+
if (0) {
err_object:
kobject_put(kobj);
diff --git a/drivers/gpu/drm/i915/gt/intel_engine_types.h b/drivers/gpu/drm/i915/gt/intel_engine_types.h
index 3451be034caf..89a9616e8539 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine_types.h
+++ b/drivers/gpu/drm/i915/gt/intel_engine_types.h
@@ -542,6 +542,10 @@ struct intel_engine_cs {
*/
ktime_t total;
} stats;
+
+ struct {
+ unsigned long timeslice_duration_ms;
+ } props;
};
static inline bool
diff --git a/drivers/gpu/drm/i915/gt/intel_lrc.c b/drivers/gpu/drm/i915/gt/intel_lrc.c
index f9f3e985bb79..30142ebce8fe 100644
--- a/drivers/gpu/drm/i915/gt/intel_lrc.c
+++ b/drivers/gpu/drm/i915/gt/intel_lrc.c
@@ -252,6 +252,26 @@ static inline u32 intel_hws_preempt_address(struct intel_engine_cs *engine)
I915_GEM_HWS_PREEMPT_ADDR);
}
+static void __set_timer(struct timer_list *t, unsigned long timeout)
+{
+ if (!timeout) {
+ cancel_timer(t);
+ return;
+ }
+
+ timeout = msecs_to_jiffies_timeout(timeout);
+
+ /*
+ * Paranoia to make sure the compiler computes the timeout before
+ * loading 'jiffies' as jiffies is volatile and may be updated in
+ * the background by a timer tick. All to reduce the complexity
+ * of the addition and reduce the risk of losing a jiffie.
+ */
+ barrier();
+
+ mod_timer(t, jiffies + timeout);
+}
+
static inline void
ring_set_paused(const struct intel_engine_cs *engine, int state)
{
@@ -1335,7 +1355,7 @@ need_timeslice(struct intel_engine_cs *engine, const struct i915_request *rq)
{
int hint;
- if (!intel_engine_has_semaphores(engine))
+ if (!intel_engine_has_timeslices(engine))
return false;
if (list_is_last(&rq->sched.link, &engine->active.requests))
@@ -1356,15 +1376,32 @@ switch_prio(struct intel_engine_cs *engine, const struct i915_request *rq)
return rq_prio(list_next_entry(rq, sched.link));
}
-static bool
-enable_timeslice(const struct intel_engine_execlists *execlists)
+static inline unsigned long
+timeslice(const struct intel_engine_cs *engine)
+{
+ return READ_ONCE(engine->props.timeslice_duration_ms);
+}
+
+static unsigned long
+active_timeslice(const struct intel_engine_cs *engine)
{
- const struct i915_request *rq = *execlists->active;
+ const struct i915_request *rq = *engine->execlists.active;
if (i915_request_completed(rq))
- return false;
+ return 0;
- return execlists->switch_priority_hint >= effective_prio(rq);
+ if (engine->execlists.switch_priority_hint < effective_prio(rq))
+ return 0;
+
+ return timeslice(engine);
+}
+
+static void set_timeslice(struct intel_engine_cs *engine)
+{
+ if (!intel_engine_has_timeslices(engine))
+ return;
+
+ __set_timer(&engine->execlists.timer, active_timeslice(engine));
}
static void record_preemption(struct intel_engine_execlists *execlists)
@@ -1511,8 +1548,9 @@ static void execlists_dequeue(struct intel_engine_cs *engine)
*/
if (!execlists->timer.expires &&
need_timeslice(engine, last))
- mod_timer(&execlists->timer,
- jiffies + 1);
+ __set_timer(&execlists->timer,
+ timeslice(engine));
+
return;
}
@@ -1934,10 +1972,7 @@ static void process_csb(struct intel_engine_cs *engine)
execlists_num_ports(execlists) *
sizeof(*execlists->pending));
- if (enable_timeslice(execlists))
- mod_timer(&execlists->timer, jiffies + 1);
- else
- cancel_timer(&execlists->timer);
+ set_timeslice(engine);
WRITE_ONCE(execlists->pending[0], NULL);
} else {
diff --git a/drivers/gpu/drm/i915/gt/selftest_lrc.c b/drivers/gpu/drm/i915/gt/selftest_lrc.c
index 2c9c75166570..04864d7588a2 100644
--- a/drivers/gpu/drm/i915/gt/selftest_lrc.c
+++ b/drivers/gpu/drm/i915/gt/selftest_lrc.c
@@ -439,6 +439,8 @@ static int live_timeslice_preempt(void *arg)
* need to preempt the current task and replace it with another
* ready task.
*/
+ if (!CONFIG_DRM_I915_TIMESLICE_DURATION)
+ return 0;
obj = i915_gem_object_create_internal(gt->i915, PAGE_SIZE);
if (IS_ERR(obj))
@@ -513,6 +515,11 @@ static void wait_for_submit(struct intel_engine_cs *engine,
} while (!i915_request_is_active(rq));
}
+static long timeslice_threshold(const struct intel_engine_cs *engine)
+{
+ return 2 * msecs_to_jiffies_timeout(timeslice(engine)) + 1;
+}
+
static int live_timeslice_queue(void *arg)
{
struct intel_gt *gt = arg;
@@ -530,6 +537,8 @@ static int live_timeslice_queue(void *arg)
* ELSP[1] is already occupied, so must rely on timeslicing to
* eject ELSP[0] in favour of the queue.)
*/
+ if (!CONFIG_DRM_I915_TIMESLICE_DURATION)
+ return 0;
obj = i915_gem_object_create_internal(gt->i915, PAGE_SIZE);
if (IS_ERR(obj))
@@ -607,8 +616,8 @@ static int live_timeslice_queue(void *arg)
err = -EINVAL;
}
- /* Timeslice every jiffie, so within 2 we should signal */
- if (i915_request_wait(rq, 0, 3) < 0) {
+ /* Timeslice every jiffy, so within 2 we should signal */
+ if (i915_request_wait(rq, 0, timeslice_threshold(engine)) < 0) {
struct drm_printer p =
drm_info_printer(gt->i915->drm.dev);
--
2.24.0.rc0
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
