On 01/12/2017 01:01, Rodrigo Vivi wrote:
On Tue, Nov 21, 2017 at 06:18:45PM +0000, Tvrtko Ursulin wrote:
From: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx>
From: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
From: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx>
From: Dmitry Rogozhkin <dmitry.v.rogozhkin@xxxxxxxxx>
The first goal is to be able to measure GPU (and invidual ring) busyness
without having to poll registers from userspace. (Which not only incurs
holding the forcewake lock indefinitely, perturbing the system, but also
runs the risk of hanging the machine.) As an alternative we can use the
perf event counter interface to sample the ring registers periodically
and send those results to userspace.
Functionality we are exporting to userspace is via the existing perf PMU
API and can be exercised via the existing tools. For example:
perf stat -a -e i915/rcs0-busy/ -I 1000
Will print the render engine busynnes once per second. All the performance
counters can be enumerated (perf list) and have their unit of measure
correctly reported in sysfs.
[snip]
+static int i915_pmu_event_init(struct perf_event *event)
+{
+ struct drm_i915_private *i915 =
+ container_of(event->pmu, typeof(*i915), pmu.base);
+ int cpu, ret;
+
+ if (event->attr.type != event->pmu->type)
+ return -ENOENT;
+
+ /* unsupported modes and filters */
+ if (event->attr.sample_period) /* no sampling */
+ return -EINVAL;
+
+ if (has_branch_stack(event))
+ return -EOPNOTSUPP;
+
+ if (event->cpu < 0)
+ return -EINVAL;
+
+ cpu = cpumask_any_and(&i915_pmu_cpumask,
+ topology_sibling_cpumask(event->cpu));
+ if (cpu >= nr_cpu_ids)
+ return -ENODEV;
+
+ if (is_engine_event(event)) {
+ ret = engine_event_init(event);
+ } else {
+ ret = 0;
+ switch (event->attr.config) {
+ case I915_PMU_ACTUAL_FREQUENCY:
+ if (IS_VALLEYVIEW(i915) || IS_CHERRYVIEW(i915))
+ /* Requires a mutex for sampling! */
+ ret = -ENODEV;
Dummy random question...
Is the missing break here intentional?
on PMU_ACTUAL_FREQ we should return -ENODEV to gen < 6 || vlv || chv ?
I am not expert on pre-Gen6, but AFAICS yes, none of the frequency code
paths seem to be active on those platforms. Was there some alternative
way of managing frequencies back then, which would make some reporting
still possible? I can't spot anything - Chris?
Regards,
Tvrtko
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
