Add a delay, configurable via debugfs (default 34ms), to disable
scheduling of a context after the pin count goes to zero. Disable
scheduling is a costly operation as it requires synchronizing with
the GuC. So the idea is that a delay allows the user to resubmit
something before doing this operation. This delay is only done if
the context isn't closed and less than a given threshold
(default is 3/4) of the guc_ids are in use.
Alan Previn: Matt Brost first introduced this patch back in Oct 2021.
However no real world workload with measured performance impact was
available to prove the intended results. Today, this series is being
republished in response to a real world workload that benefited greatly
from it along with measured performance improvement.
Workload description: 36 containers were created on a DG2 device where
each container was performing a combination of 720p 3d game rendering
and 30fps video encoding. The workload density was configured in a way
that guaranteed each container to ALWAYS be able to render and
encode no less than 30fps with a predefined maximum render + encode
latency time. That means the totality of all 36 containers and their
workloads were not saturating the engines to their max (in order to
maintain just enough headrooom to meet the min fps and max latencies
of incoming container submissions).
Problem statement: It was observed that the CPU core processing the i915
soft IRQ work was experiencing severe load. Using tracelogs and an
instrumentation patch to count specific i915 IRQ events, it was confirmed
that the majority of the CPU cycles were caused by the
gen11_other_irq_handler() -> guc_irq_handler() code path. The vast
majority of the cycles was determined to be processing a specific G2H
IRQ: i.e. INTEL_GUC_ACTION_SCHED_CONTEXT_MODE_DONE. These IRQs are sent
by GuC in response to i915 KMD sending H2G requests:
INTEL_GUC_ACTION_SCHED_CONTEXT_MODE_SET. Those H2G requests are sent
whenever a context goes idle so that we can unpin the context from GuC.
The high CPU utilization % symptom was limiting density scaling.
Root Cause Analysis: Because the incoming execution buffers were spread
across 36 different containers (each with multiple contexts) but the
system in totality was NOT saturated to the max, it was assumed that each
context was constantly idling between submissions. This was causing
a thrashing of unpinning contexts from GuC at one moment, followed quickly
by repinning them due to incoming workload the very next moment. These
event-pairs were being triggered across multiple contexts per container,
across all containers at the rate of > 30 times per sec per context.
Metrics: When running this workload without this patch, we measured an
average of ~69K INTEL_GUC_ACTION_SCHED_CONTEXT_MODE_DONE events every 10
seconds or ~10 million times over ~25+ mins. With this patch, the count
reduced to ~480 every 10 seconds or about ~28K over ~10 mins. The
improvement observed is ~99% for the average counts per 10 seconds.
Design awareness: Selftest impact.
As temporary WA disable this feature for the selftests. Selftests are
very timing sensitive and any change in timing can cause failure. A
follow up patch will fixup the selftests to understand this delay.
Design awareness: Race between guc_request_alloc and guc_context_close.
If a context close is issued while there is a request submission in
flight and a delayed schedule disable is pending, guc_context_close
and guc_request_alloc will race to cancel the delayed disable.
To close the race, make sure that guc_request_alloc waits for
guc_context_close to finish running before checking any state.
Design awareness: GT Reset event.
If a gt reset is triggered, as preparation steps, add an additional step
to ensure all contexts that have a pending delay-disable-schedule task
be flushed of it. Move them directly into the closed state after cancelling
the worker. This is okay because the existing flow flushes all
yet-to-arrive G2H's dropping them anyway.
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Alan Previn <alan.previn.teres.alexis@intel.com>
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221006225121.826257-2-alan.previn.teres.alexis@intel.com
Add a delay, configurable via debugfs (default 34ms), to disable
scheduling of a context after the pin count goes to zero. Disable
scheduling is a costly operation as it requires synchronizing with
the GuC. So the idea is that a delay allows the user to resubmit
something before doing this operation. This delay is only done if
the context isn't closed and less than a given threshold
(default is 3/4) of the guc_ids are in use.
As temporary WA disable this feature for the selftests. Selftests are
very timing sensitive and any change in timing can cause failure. A
follow up patch will fixup the selftests to understand this delay.
Alan Previn: Matt Brost first introduced this series back in Oct 2021.
However no real world workload with measured performance impact was
available to prove the intended results. Today, this series is being
republished in response to a real world workload that benefited greatly
from it along with measured performance improvement.
Workload description: 36 containers were created on a DG2 device where
each container was performing a combination of 720p 3d game rendering
and 30fps video encoding. The workload density was configured in a way
that guaranteed each container to ALWAYS be able to render and
encode no less than 30fps with a predefined maximum render + encode
latency time. That means the totality of all 36 containers and their
workloads were not saturating the engines to their max (in order to
maintain just enough headrooom to meet the min fps and max latencies
of incoming container submissions).
Problem statement: It was observed that the CPU core processing the i915
soft IRQ work was experiencing severe load. Using tracelogs and an
instrumentation patch to count specific i915 IRQ events, it was confirmed
that the majority of the CPU cycles were caused by the
gen11_other_irq_handler() -> guc_irq_handler() code path. The vast
majority of the cycles was determined to be processing a specific G2H
IRQ: i.e. INTEL_GUC_ACTION_SCHED_CONTEXT_MODE_DONE. These IRQs are sent
by GuC in response to i915 KMD sending H2G requests:
INTEL_GUC_ACTION_SCHED_CONTEXT_MODE_SET. Those H2G requests are sent
whenever a context goes idle so that we can unpin the context from GuC.
The high CPU utilization % symptom was limiting density scaling.
Root Cause Analysis: Because the incoming execution buffers were spread
across 36 different containers (each with multiple contexts) but the
system in totality was NOT saturated to the max, it was assumed that each
context was constantly idling between submissions. This was causing
a thrashing of unpinning contexts from GuC at one moment, followed quickly
by repinning them due to incoming workload the very next moment. These
event-pairs were being triggered across multiple contexts per container,
across all containers at the rate of > 30 times per sec per context.
Metrics: When running this workload without this patch, we measured an
average of ~69K INTEL_GUC_ACTION_SCHED_CONTEXT_MODE_DONE events every 10
seconds or ~10 million times over ~25+ mins. With this patch, the count
reduced to ~480 every 10 seconds or about ~28K over ~10 mins. The
improvement observed is ~99% for the average counts per 10 seconds.
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Alan Previn <alan.previn.teres.alexis@intel.com>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220817020511.2180747-3-alan.previn.teres.alexis@intel.com
The older arches did not convert MI_STORE_DATA_IMM to using the GTT, but
left them writing to a physical address. The notes suggest that the
primary reason would be so that the writes were cache coherent, as the
CPU cache uses physical tagging. As such we did not implement the
legacy variant of MI_STORE_DATA_IMM and so left all the relocations
synchronous -- but with a small function to convert from the vma address
into the physical address, we can implement asynchronous relocs on these
older arches, fixing up a few tests that require them.
In order to be able to test the legacy paths, refactor the gpu
relocations so that we can hook them up to a selftest.
v2: Use an array of offsets not enum labels for the selftest
v3: Refactor the common igt_hexdump()
Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/757
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/20200504140629.28240-1-chris@chris-wilson.co.uk
In bringup on simulated HW even rudimentary tests are slow, and so many
may fail that we want to be able to filter out the noise to focus on the
specific problem. Even just the tests groups provided for igt is not
specific enough, and we would like to isolate one particular subtest
(and probably subsubtests!). For simplicity, allow the user to provide a
command line parameter such as
i915.st_filter=i915_timeline_mock_selftests/igt_sync
to restrict ourselves to only running on subtest. The exact name to use
is given during a normal run, highlighted as an error if it failed,
debug otherwise. The test group is optional, and then all subtests are
compared for an exact match with the filter (most subtests have unique
names). The filter can be negated, e.g. i915.st_filter=!igt_sync and
then all tests but those that match will be run. More than one match can
be supplied separated by a comma, e.g.
i915.st_filter=igt_vma_create,igt_vma_pin1
to only run those specified, or
i915.st_filter=!igt_vma_create,!igt_vma_pin1
to run all but those named. Mixing a blacklist and whitelist will only
execute those subtests matching the whitelist so long as they are
previously excluded in the blacklist.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190129185452.20989-1-chris@chris-wilson.co.uk
It is possible whilst allocating the page-directory tree for a ppgtt
bind that the shrinker may run and reap unused parts of the tree. If the
shrinker happens to remove a chunk of the tree that the
allocate_va_range has already processed, we may then try to insert into
the dangling tree. This test uses the fault-injection framework to force
the shrinker to be invoked before we allocate new pages, i.e. new chunks
of the PD tree.
References: https://bugs.freedesktop.org/show_bug.cgi?id=99295
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Some pieces of code are independent of hardware but are very tricky to
exercise through the normal userspace ABI or via debugfs hooks. Being
able to create mock unit tests and execute them through CI is vital.
Start by adding a central point where we can execute unit tests and
a parameter to enable them. This is disabled by default as the
expectation is that these tests will occasionally explode.
To facilitate integration with igt, any parameter beginning with
i915.igt__ is interpreted as a subtest executable independently via
igt/drv_selftest.
Two classes of selftests are recognised: mock unit tests and integration
tests. Mock unit tests are run as soon as the module is loaded, before
the device is probed. At that point there is no driver instantiated and
all hw interactions must be "mocked". This is very useful for writing
universal tests to exercise code not typically run on a broad range of
architectures. Alternatively, you can hook into the live selftests and
run when the device has been instantiated - hw interactions are real.
v2: Add a macro for compiling conditional code for mock objects inside
real objects.
v3: Differentiate between mock unit tests and late integration test.
v4: List the tests in natural order, use igt to sort after modparam.
v5: s/late/live/
v6: s/unsigned long/unsigned int/
v7: Use igt_ prefixes for long helpers.
v8: Deobfuscate macros overriding functions, stop using -I$(src)
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170213171558.20942-1-chris@chris-wilson.co.uk
