| Message ID | 20191212140459.1307617-30-chris@chris-wilson.co.uk (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [01/33] drm/i915: Use EAGAIN for trylock failures | expand |
On Thu, Dec 12, 2019 at 02:04:56PM +0000, Chris Wilson wrote: > When we allow ourselves to sleep before a GPU reset after disabling > submission, even for a few milliseconds, gives an innocent context the > opportunity to clear the GPU before the reset occurs. However, how long > to sleep depends on the typical non-preemptible duration (a similar > problem to determining the ideal preempt-reset timeout or even the > heartbeat interval). As this seems of a hard policy decision, punt it to > userspace. > > The timeout can be adjusted using > > /sys/class/drm/card?/engine/*/stop_timeout_ms > > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> > Cc: Jon Bloomfield <jon.bloomfield@intel.com> Code looks good. Tested with other sysfs related patches in the series. The stop_timeout_ms file exists and can be modified. Reviewed-by: Steve Carbonari <steven.carbonari@intel.com> Tested-by: Steve Carbonari <steven.carbonari@intel.com > --- > drivers/gpu/drm/i915/Kconfig.profile | 3 ++ > drivers/gpu/drm/i915/gt/intel_engine_sysfs.c | 40 ++++++++++++++++++++ > 2 files changed, 43 insertions(+) > > diff --git a/drivers/gpu/drm/i915/Kconfig.profile b/drivers/gpu/drm/i915/Kconfig.profile > index 9ee3b59685b9..5f4ec3aec1d2 100644 > --- a/drivers/gpu/drm/i915/Kconfig.profile > +++ b/drivers/gpu/drm/i915/Kconfig.profile > @@ -63,6 +63,9 @@ config DRM_I915_STOP_TIMEOUT > that the reset itself may take longer and so be more disruptive to > interactive or low latency workloads. > > + This is adjustable via > + /sys/class/drm/card?/engine/*/stop_timeout_ms > + > config DRM_I915_TIMESLICE_DURATION > int "Scheduling quantum for userspace batches (ms, jiffy granularity)" > default 1 # milliseconds > diff --git a/drivers/gpu/drm/i915/gt/intel_engine_sysfs.c b/drivers/gpu/drm/i915/gt/intel_engine_sysfs.c > index 6d87529c64a7..2b65fed76435 100644 > --- a/drivers/gpu/drm/i915/gt/intel_engine_sysfs.c > +++ b/drivers/gpu/drm/i915/gt/intel_engine_sysfs.c > @@ -232,6 +232,45 @@ timeslice_show(struct kobject *kobj, struct kobj_attribute *attr, char *buf) > static struct kobj_attribute timeslice_duration_attr = > __ATTR(timeslice_duration_ms, 0644, timeslice_show, timeslice_store); > > +static ssize_t > +stop_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; > + > + /* > + * When we allow ourselves to sleep before a GPU reset after disabling > + * submission, even for a few milliseconds, gives an innocent context > + * the opportunity to clear the GPU before the reset occurs. However, > + * how long to sleep depends on the typical non-preemptible duration > + * (a similar problem to determining the ideal preempt-reset timeout > + * or even the heartbeat interval). > + */ > + > + err = kstrtoull(buf, 0, &duration); > + if (err) > + return err; > + > + if (duration > jiffies_to_msecs(MAX_SCHEDULE_TIMEOUT)) > + return -EINVAL; > + > + WRITE_ONCE(engine->props.stop_timeout_ms, duration); > + return count; > +} > + > +static ssize_t > +stop_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.stop_timeout_ms); > +} > + > +static struct kobj_attribute stop_timeout_attr = > +__ATTR(stop_timeout_ms, 0644, stop_show, stop_store); > + > static void kobj_engine_release(struct kobject *kobj) > { > kfree(kobj); > @@ -273,6 +312,7 @@ void intel_engines_add_sysfs(struct drm_i915_private *i915) > &caps_attr.attr, > &all_caps_attr.attr, > &max_spin_attr.attr, > + &stop_timeout_attr.attr, > NULL > }; >
diff --git a/drivers/gpu/drm/i915/Kconfig.profile b/drivers/gpu/drm/i915/Kconfig.profile index 9ee3b59685b9..5f4ec3aec1d2 100644 --- a/drivers/gpu/drm/i915/Kconfig.profile +++ b/drivers/gpu/drm/i915/Kconfig.profile @@ -63,6 +63,9 @@ config DRM_I915_STOP_TIMEOUT that the reset itself may take longer and so be more disruptive to interactive or low latency workloads. + This is adjustable via + /sys/class/drm/card?/engine/*/stop_timeout_ms + config DRM_I915_TIMESLICE_DURATION int "Scheduling quantum for userspace batches (ms, jiffy granularity)" default 1 # milliseconds diff --git a/drivers/gpu/drm/i915/gt/intel_engine_sysfs.c b/drivers/gpu/drm/i915/gt/intel_engine_sysfs.c index 6d87529c64a7..2b65fed76435 100644 --- a/drivers/gpu/drm/i915/gt/intel_engine_sysfs.c +++ b/drivers/gpu/drm/i915/gt/intel_engine_sysfs.c @@ -232,6 +232,45 @@ timeslice_show(struct kobject *kobj, struct kobj_attribute *attr, char *buf) static struct kobj_attribute timeslice_duration_attr = __ATTR(timeslice_duration_ms, 0644, timeslice_show, timeslice_store); +static ssize_t +stop_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; + + /* + * When we allow ourselves to sleep before a GPU reset after disabling + * submission, even for a few milliseconds, gives an innocent context + * the opportunity to clear the GPU before the reset occurs. However, + * how long to sleep depends on the typical non-preemptible duration + * (a similar problem to determining the ideal preempt-reset timeout + * or even the heartbeat interval). + */ + + err = kstrtoull(buf, 0, &duration); + if (err) + return err; + + if (duration > jiffies_to_msecs(MAX_SCHEDULE_TIMEOUT)) + return -EINVAL; + + WRITE_ONCE(engine->props.stop_timeout_ms, duration); + return count; +} + +static ssize_t +stop_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.stop_timeout_ms); +} + +static struct kobj_attribute stop_timeout_attr = +__ATTR(stop_timeout_ms, 0644, stop_show, stop_store); + static void kobj_engine_release(struct kobject *kobj) { kfree(kobj); @@ -273,6 +312,7 @@ void intel_engines_add_sysfs(struct drm_i915_private *i915) &caps_attr.attr, &all_caps_attr.attr, &max_spin_attr.attr, + &stop_timeout_attr.attr, NULL };
When we allow ourselves to sleep before a GPU reset after disabling submission, even for a few milliseconds, gives an innocent context the opportunity to clear the GPU before the reset occurs. However, how long to sleep depends on the typical non-preemptible duration (a similar problem to determining the ideal preempt-reset timeout or even the heartbeat interval). As this seems of a hard policy decision, punt it to userspace. The timeout can be adjusted using /sys/class/drm/card?/engine/*/stop_timeout_ms Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Cc: Jon Bloomfield <jon.bloomfield@intel.com> --- drivers/gpu/drm/i915/Kconfig.profile | 3 ++ drivers/gpu/drm/i915/gt/intel_engine_sysfs.c | 40 ++++++++++++++++++++ 2 files changed, 43 insertions(+)