The major drawback of commit 7e34f4e4aa ("drm/i915/gen8+: Add RC6 CTX
corruption WA") is that it disables RC6 while Skylake (and friends) is
active, and we do not consider the GPU idle until all outstanding
requests have been retired and the engine switched over to the kernel
context. If userspace is idle, this task falls onto our background idle
worker, which only runs roughly once a second, meaning that userspace has
to have been idle for a couple of seconds before we enable RC6 again.
Naturally, this causes us to consume considerably more energy than
before as powersaving is effectively disabled while a display server
(here's looking at you Xorg) is running.
As execlists will get a completion event as each context is completed,
we can use this interrupt to queue a retire worker bound to this engine
to cleanup idle timelines. We will then immediately notice the idle
engine (without userspace intervention or the aid of the background
retire worker) and start parking the GPU. Thus during light workloads,
we will do much more work to idle the GPU faster... Hopefully with
commensurate power saving!
v2: Watch context completions and only look at those local to the engine
when retiring to reduce the amount of excess work we perform.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=112315
References: 7e34f4e4aa ("drm/i915/gen8+: Add RC6 CTX corruption WA")
References: 2248a28384 ("drm/i915/gen8+: Add RC6 CTX corruption WA")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191125105858.1718307-3-chris@chris-wilson.co.uk
(cherry picked from commit 4f88f8747f)
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Since we want to do a lockless read of the current active request, and
that request is written to by process_csb also without serialisation, we
need to instruct gcc to take care in reading the pointer itself.
Otherwise, we have observed execlists_active() to report 0x40.
[ 2400.760381] igt/para-4098 1..s. 2376479300us : process_csb: rcs0 cs-irq head=3, tail=4
[ 2400.760826] igt/para-4098 1..s. 2376479303us : process_csb: rcs0 csb[4]: status=0x00000001:0x00000000
[ 2400.761271] igt/para-4098 1..s. 2376479306us : trace_ports: rcs0: promote { b9c59:2622, b9c55:2624 }
[ 2400.761726] igt/para-4097 0d... 2376479311us : __i915_schedule: rcs0: -2147483648->3, inflight:0000000000000040, rq:ffff888208c1e940
which is impossible!
The answer is that as we keep the existing execlists->active pointing
into the array as we copy over that array, the unserialised read may see
a partial pointer value.
Fixes: df40306902 ("drm/i915/execlists: Lift process_csb() out of the irq-off spinlock")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191125094318.1630806-1-chris@chris-wilson.co.uk
(cherry picked from commit 331bf90591)
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
The major drawback of commit 7e34f4e4aa ("drm/i915/gen8+: Add RC6 CTX
corruption WA") is that it disables RC6 while Skylake (and friends) is
active, and we do not consider the GPU idle until all outstanding
requests have been retired and the engine switched over to the kernel
context. If userspace is idle, this task falls onto our background idle
worker, which only runs roughly once a second, meaning that userspace has
to have been idle for a couple of seconds before we enable RC6 again.
Naturally, this causes us to consume considerably more energy than
before as powersaving is effectively disabled while a display server
(here's looking at you Xorg) is running.
As execlists will get a completion event as each context is completed,
we can use this interrupt to queue a retire worker bound to this engine
to cleanup idle timelines. We will then immediately notice the idle
engine (without userspace intervention or the aid of the background
retire worker) and start parking the GPU. Thus during light workloads,
we will do much more work to idle the GPU faster... Hopefully with
commensurate power saving!
v2: Watch context completions and only look at those local to the engine
when retiring to reduce the amount of excess work we perform.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=112315
References: 7e34f4e4aa ("drm/i915/gen8+: Add RC6 CTX corruption WA")
References: 2248a28384 ("drm/i915/gen8+: Add RC6 CTX corruption WA")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191125105858.1718307-3-chris@chris-wilson.co.uk
Since we want to do a lockless read of the current active request, and
that request is written to by process_csb also without serialisation, we
need to instruct gcc to take care in reading the pointer itself.
Otherwise, we have observed execlists_active() to report 0x40.
[ 2400.760381] igt/para-4098 1..s. 2376479300us : process_csb: rcs0 cs-irq head=3, tail=4
[ 2400.760826] igt/para-4098 1..s. 2376479303us : process_csb: rcs0 csb[4]: status=0x00000001:0x00000000
[ 2400.761271] igt/para-4098 1..s. 2376479306us : trace_ports: rcs0: promote { b9c59:2622, b9c55:2624 }
[ 2400.761726] igt/para-4097 0d... 2376479311us : __i915_schedule: rcs0: -2147483648->3, inflight:0000000000000040, rq:ffff888208c1e940
which is impossible!
The answer is that as we keep the existing execlists->active pointing
into the array as we copy over that array, the unserialised read may see
a partial pointer value.
Fixes: df40306902 ("drm/i915/execlists: Lift process_csb() out of the irq-off spinlock")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191125094318.1630806-1-chris@chris-wilson.co.uk
The gem_ctx_persistence/smoketest was detecting an odd coherency issue
inside the LRC context image; that the address of the ring buffer did
not match our associated struct intel_ring. As we set the address into
the context image when we pin the ring buffer into place before the
context is active, that leaves the question of where did it get
overwritten. Either the HW context save occurred after our pin which
would imply that our idle barriers are broken, or we overwrote the
context image ourselves. It is only in reset_active() where we dabble
inside the context image outside of a serialised path from schedule-out;
but we could equally perform the operation inside schedule-in which is
then fully serialised with the context pin -- and remains serialised by
the engine pulse with kill_context(). (The only downside, aside from
doing more work inside the engine->active.lock, was the plan to merge
all the reset paths into doing their context scrubbing on schedule-out
needs more thought.)
Fixes: d12acee84f ("drm/i915/execlists: Cancel banned contexts on schedule-out")
Testcase: igt/gem_ctx_persistence/smoketest
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191111133205.11590-3-chris@chris-wilson.co.uk
(cherry picked from commit 31b61f0ef9)
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
The gem_ctx_persistence/smoketest was detecting an odd coherency issue
inside the LRC context image; that the address of the ring buffer did
not match our associated struct intel_ring. As we set the address into
the context image when we pin the ring buffer into place before the
context is active, that leaves the question of where did it get
overwritten. Either the HW context save occurred after our pin which
would imply that our idle barriers are broken, or we overwrote the
context image ourselves. It is only in reset_active() where we dabble
inside the context image outside of a serialised path from schedule-out;
but we could equally perform the operation inside schedule-in which is
then fully serialised with the context pin -- and remains serialised by
the engine pulse with kill_context(). (The only downside, aside from
doing more work inside the engine->active.lock, was the plan to merge
all the reset paths into doing their context scrubbing on schedule-out
needs more thought.)
Fixes: d12acee84f ("drm/i915/execlists: Cancel banned contexts on schedule-out")
Testcase: igt/gem_ctx_persistence/smoketest
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191111133205.11590-3-chris@chris-wilson.co.uk
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. However, not all workloads necessarily benefit
from timeslicing and in the extreme some sysadmin may want to disable or
reduce the timeslicing granularity.
The timeslicing mechanism can be compiled out^W^W disabled (but should
DCE!) with
./scripts/config --set-val DRM_I915_TIMESLICE_DURATION 0
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191029091632.26281-1-chris@chris-wilson.co.uk
The request's timeline will only contain requests from this context, in
order of execution. Therefore, we can simply look back along this
timeline to find the currently executing request.
If we do find that the current context has completed its last request,
that does not imply that all requests are completed in the context, so
only advance the ring->head up to the end of the known completions!
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191028124125.25176-1-chris@chris-wilson.co.uk
On schedule-out (CS completion) of a banned context, scrub the context
image so that we do not replay the active payload. The intent is that we
skip banned payloads on request submission so that the timeline
advancement continues on in the background. However, if we are returning
to a preempted request, i915_request_skip() is ineffective and instead we
need to patch up the context image so that it continues from the start
of the next request.
v2: Fixup cancellation so that we only scrub the payload of the active
request and do not short-circuit the breadcrumbs (which might cause
other contexts to execute out of order).
v3: Grammar pass
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191023133108.21401-3-chris@chris-wilson.co.uk
If the preempted context takes too long to relinquish control, e.g. it
is stuck inside a shader with arbitration disabled, evict that context
with an engine reset. This ensures that preemptions are reasonably
responsive, providing a tighter QoS for the more important context at
the cost of flagging unresponsive contexts more frequently (i.e. instead
of using an ~10s hangcheck, we now evict at ~100ms). The challenge of
lies in picking a timeout that can be reasonably serviced by HW for
typical workloads, balancing the existing clients against the needs for
responsiveness.
Note that coupled with timeslicing, this will lead to rapid GPU "hang"
detection with multiple active contexts vying for GPU time.
The forced preemption mechanism can be compiled out with
./scripts/config --set-val DRM_I915_PREEMPT_TIMEOUT 0
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191023133108.21401-2-chris@chris-wilson.co.uk
The actual conditions are that we know the GPU is not accessing the
context, and we hold a pin on the context image to allow CPU access. We
used a fake lock on ce->pin_mutex so that we could try and use lockdep
to assert that access is serialised, but the various different
hardirq/softirq contexts where we need to *fake* holding the pin_mutex
are causing more trouble.
Still it would be nice if we did have a way to reassure ourselves that
the direct update to the context image is serialised with GPU execution.
In the meantime, stop lockdep complaining about false irq inversions.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=111923
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Acked-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191022122845.25038-1-chris@chris-wilson.co.uk
Normally, we try and skip submission if ELSP[1] is filled. However, we
may desire to enable timeslicing due to the queue priority, even if
ELSP[1] itself does not require timeslicing. That is the queue is equal
priority to ELSP[0] and higher priority then ELSP[1]. Previously, we
would wait until the context switch to preempt the current ELSP[1], but
with timeslicing, we want to preempt ELSP[0] and replace it with the
queue.
In writing the test case, it become quickly apparent that we were also
suppressing the tasklet during promotion and so failing to notice when
the queue started requiring timeslicing.
Fixes: 2229adc813 ("drm/i915/execlist: Trim immediate timeslice expiry")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191018072027.31948-1-chris@chris-wilson.co.uk
We perform timeslicing immediately upon receipt of a request that may be
put into the second ELSP slot. The idea behind this was that since we
didn't install the timer if the second ELSP slot was empty, we would not
have any idea of how long ELSP[0] had been running and so giving the
newcomer a chance on the GPU was fair. However, this causes us extra
busy work that we may be able to avoid if we wait a jiffie for the first
timeslice as normal.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191016100851.4979-1-chris@chris-wilson.co.uk
We rely on only the tasklet being allowed to call into process_csb(), so
assert that is locked when we do. As the tasklet uses a simple bitlock,
there is no strong lockdep checking so we must make do with a plain
assertion that the tasklet is running and assume that we are the
tasklet!
v2: Fixup intel_gt_sanitize() to prepare each engine for the reset so
that the locks are marked as held during the reset
v3: Check for existent function pointers for very early sanitisation.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191014121336.30137-1-chris@chris-wilson.co.uk
Since commit e2144503bf ("drm/i915: Prevent bonded requests from
overtaking each other on preemption") we have restricted requests to run
on their chosen engine across preemption events. We can take this
restriction into account to know that we will want to resubmit those
requests onto the same physical engine, and so can shortcircuit the
virtual engine selection process and keep the request on the same
engine during unwind.
References: e2144503bf ("drm/i915: Prevent bonded requests from overtaking each other on preemption")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Ramlingam C <ramalingam.c@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191013203012.25208-1-chris@chris-wilson.co.uk
We set out-of-bound parameters inside the i915_requests.flags field,
such as disabling preemption or marking the end-of-context. We should
not coalesce consecutive requests if they have differing instructions
as we only inspect the last active request in a context. Thus if we
allow a later request to be merged into the same execution context, it
will mask any of the earlier flags.
References: 2a98f4e65b ("drm/i915: add infrastructure to hold off preemption on a request")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191011190325.10979-9-chris@chris-wilson.co.uk
With the introduction of ctx->engines[] we allow multiple logical
contexts to be used on the same engine (e.g. with virtual engines).
According to bspec, aach logical context requires a unique tag in order
for context-switching to occur correctly between them. [Simple
experiments show that it is not so easy to trick the HW into performing
a lite-restore with matching logical IDs, though my memory from early
Broadwell experiments do suggest that it should be generating
lite-restores.]
We only need to keep a unique tag for the active lifetime of the
context, and for as long as we need to identify that context. The HW
uses the tag to determine if it should use a lite-restore (why not the
LRCA?) and passes the tag back for various status identifies. The only
status we need to track is for OA, so when using perf, we assign the
specific context a unique tag.
v2: Calculate required number of tags to fill ELSP.
Fixes: 976b55f0e1 ("drm/i915: Allow a context to define its set of engines")
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=111895
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Acked-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191004134015.13204-14-chris@chris-wilson.co.uk
If execlists's lite-restore is based on the common GEM context tag
rather than the per-intel_context LRCA, then a context switch between
two intel_contexts on the same engine derived from the same GEM context
will perform a lite-restore instead of a full context switch. We can
exploit this by poisoning the ringbuffer of the first context and trying
to trick a simple RING_TAIL update (i.e. lite-restore)
v2: Also check what happens if preempt ce[0] with ce[1] (both instances
on the same engine from the same parent context) [Tvrtko]
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191002183459.26614-1-chris@chris-wilson.co.uk
