With GuC handling scheduling, i915 is not aware of the time that a
context is scheduled in and out of the engine. Since i915 pmu relies on
this info to provide engine busyness to the user, GuC shares this info
with i915 for all engines using shared memory. For each engine, this
info contains:
- total busyness: total time that the context was running (total)
- id: id of the running context (id)
- start timestamp: timestamp when the context started running (start)
At the time (now) of sampling the engine busyness, if the id is valid
(!= ~0), and start is non-zero, then the context is considered to be
active and the engine busyness is calculated using the below equation
engine busyness = total + (now - start)
All times are obtained from the gt clock base. For inactive contexts,
engine busyness is just equal to the total.
The start and total values provided by GuC are 32 bits and wrap around
in a few minutes. Since perf pmu provides busyness as 64 bit
monotonically increasing values, there is a need for this implementation
to account for overflows and extend the time to 64 bits before returning
busyness to the user. In order to do that, a worker runs periodically at
frequency = 1/8th the time it takes for the timestamp to wrap. As an
example, that would be once in 27 seconds for a gt clock frequency of
19.2 MHz.
Note:
There might be an over-accounting of busyness due to the fact that GuC
may be updating the total and start values while kmd is reading them.
(i.e kmd may read the updated total and the stale start). In such a
case, user may see higher busyness value followed by smaller ones which
would eventually catch up to the higher value.
v2: (Tvrtko)
- Include details in commit message
- Move intel engine busyness function into execlist code
- Use union inside engine->stats
- Use natural type for ping delay jiffies
- Drop active_work condition checks
- Use for_each_engine if iterating all engines
- Drop seq locking, use spinlock at GuC level to update engine stats
- Document worker specific details
v3: (Tvrtko/Umesh)
- Demarcate GuC and execlist stat objects with comments
- Document known over-accounting issue in commit
- Provide a consistent view of GuC state
- Add hooks to gt park/unpark for GuC busyness
- Stop/start worker in gt park/unpark path
- Drop inline
- Move spinlock and worker inits to GuC initialization
- Drop helpers that are called only once
v4: (Tvrtko/Matt/Umesh)
- Drop addressed opens from commit message
- Get runtime pm in ping, remove from the park path
- Use cancel_delayed_work_sync in disable_submission path
- Update stats during reset prepare
- Skip ping if reset in progress
- Explicitly name execlists and GuC stats objects
- Since disable_submission is called from many places, move resetting
stats to intel_guc_submission_reset_prepare
v5: (Tvrtko)
- Add a trylock helper that does not sleep and synchronize PMU event
callbacks and worker with gt reset
v6: (CI BAT failures)
- DUTs using execlist submission failed to boot since __gt_unpark is
called during i915 load. This ends up calling the GuC busyness unpark
hook and results in kick-starting an uninitialized worker. Let
park/unpark hooks check if GuC submission has been initialized.
- drop cant_sleep() from trylock helper since rcu_read_lock takes care
of that.
v7: (CI) Fix igt@i915_selftest@live@gt_engines
- For GuC mode of submission the engine busyness is derived from gt time
domain. Use gt time elapsed as reference in the selftest.
- Increase busyness calculation to 10ms duration to ensure batch runs
longer and falls within the busyness tolerances in selftest.
v8:
- Use ktime_get in selftest as before
- intel_reset_trylock_no_wait results in a lockdep splat that is not
trivial to fix since the PMU callback runs in irq context and the
reset paths are tightly knit into the driver. The test that uncovers
this is igt@perf_pmu@faulting-read. Drop intel_reset_trylock_no_wait,
instead use the reset_count to synchronize with gt reset during pmu
callback. For the ping, continue to use intel_reset_trylock since ping
is not run in irq context.
- GuC PM timestamp does not tick when GuC is idle. This can potentially
result in wrong busyness values when a context is active on the
engine, but GuC is idle. Use the RING TIMESTAMP as GPU timestamp to
process the GuC busyness stats. This works since both GuC timestamp and
RING timestamp are synced with the same clock.
- The busyness stats may get updated after the batch starts running.
This delay causes the busyness reported for 100us duration to fall
below 95% in the selftest. The only option at this time is to wait for
GuC busyness to change from idle to active before we sample busyness
over a 100us period.
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Acked-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20211027004821.66097-2-umesh.nerlige.ramappa@intel.com
Try to make bigjoiner pipes less special.
The main things here are that each pipe now does full
clock computation/readout with its own shared_dpll reference.
Also every pipe's cpu_transcoder always points correctly
at the master transcoder.
Due to the above changes state readout is now complete
and all the related hacks can go away. The actual modeset
sequence code is still a mess, but I think in order to clean
that up properly we're probably going to have to redesign
the modeset logic to treat transcoders vs. pipes separately.
That is going to require significant amounts of work.
Reviewed-by: Manasi Navare <manasi.d.navare@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20211022103304.24164-9-ville.syrjala@linux.intel.com
Let's disable planes on all pipes affected by the modeset before
we start doing the actual modeset. This means we have less
random planes enabled during the modeset, and it also mirrors
what we already do when enabling pipes on skl+ since we enable
planes on all pipes as the very last step. As a bonus we also
nuke a bunch og bigjoiner special casing.
I've occasionally pondered about going even furher here and
doing the pre_plane_update() stuff for all pipes first, then
actually disabling the planes, and finally running the rest
of the modeset sequence. This would potentially allow
parallelizing all the extra vblank waits across multiple pipes,
and would make the plane disable even more atomic. But let's
go one step a time here.
Cc: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Manasi Navare <manasi.d.navare@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20211022103304.24164-4-ville.syrjala@linux.intel.com
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Disabling planes in the middle of the modeset seuqnece does not make
sense since userspace can anyway disable planes before the modeset
even starts. So when the modeset seuqence starts the set of enabled
planes is entirely arbitrary. Trying to sprinkle the plane disabling
into the modeset sequence just means more randomness and potential
for hard to reproduce bugs.
So it makes most sense to just disable all planes first so that the
rest of the modeset sequence remains identical regardless of which
planes happen to be enabled by userspace at the time.
This reverts commit 84030adb9e.
Cc: José Roberto de Souza <jose.souza@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20211022103304.24164-3-ville.syrjala@linux.intel.com
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
PSR2 apparently requires some planes to be enabled for some
silly reason, and so we are now trying to turn PSR off before
planes go off. Except during a full modeset that is handled
less clearly through reorganization of the modeset sequence.
That is not great as it makes the code mode complex, and
prevents us from doing nice things such as just turning off
all the planes at the very start of the modeset. So let's
move the PSR pre_plane_update() thing to a spot where it
will handle both full modesets and everything else.
Cc: José Roberto de Souza <jose.souza@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20211022103304.24164-2-ville.syrjala@linux.intel.com
Reviewed-by: Jouni Högander <jouni.hogander@intel.com>
Atm until the DPCD for a connector is read the max link rate and lane
count params are invalid. If the connector is modeset, in
intel_dp_compute_config(), intel_dp_common_len_rate_limit(max_link_rate)
will return 0, leading to a intel_dp->common_rates[-1] access.
Fix the above by making sure the max link params are always valid.
The above access leads to an undefined behaviour by definition, though
not causing a user visible problem to my best knowledge, see the previous
patch why. Nevertheless it is an undefined behaviour and it triggers a
BUG() in CONFIG_UBSAN builds, hence CC:stable.
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Acked-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20211018094154.1407705-4-imre.deak@intel.com
(cherry picked from commit 9ad87de473)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Atm, there are no sink rate values set for DP (vs. eDP) sinks until the
DPCD capabilities are successfully read from the sink. During this time
intel_dp->num_common_rates is 0 which can lead to a
intel_dp->common_rates[-1] (*)
access, which is an undefined behaviour, in the following cases:
- In intel_dp_sync_state(), if the encoder is enabled without a sink
connected to the encoder's connector (BIOS enabled a monitor, but the
user unplugged the monitor until the driver loaded).
- In intel_dp_sync_state() if the encoder is enabled with a sink
connected, but for some reason the DPCD read has failed.
- In intel_dp_compute_link_config() if modesetting a connector without
a sink connected on it.
- In intel_dp_compute_link_config() if modesetting a connector with a
a sink connected on it, but before probing the connector first.
To avoid the (*) access in all the above cases, make sure that the sink
rate table - and hence the common rate table - is always valid, by
setting a default minimum sink rate when registering the connector
before anything could use it.
I also considered setting all the DP link rates by default, so that
modesetting with higher resolution modes also succeeds in the last two
cases above. However in case a sink is not connected that would stop
working after the first modeset, due to the LT fallback logic. So this
would need more work, beyond the scope of this fix.
As I mentioned in the previous patch, I don't think the issue this patch
fixes is user visible, however it is an undefined behaviour by
definition and triggers a BUG() in CONFIG_UBSAN builds, hence CC:stable.
v2: Clear the default sink rates, before initializing these for eDP.
Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/4297
References: https://gitlab.freedesktop.org/drm/intel/-/issues/4298
Suggested-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Acked-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20211018143417.1452632-1-imre.deak@intel.com
(cherry picked from commit 3f61ef9777)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Alderlake-P was getting 'max time under evasion' messages when PSR2
is enabled, this is due PIPE_SCANLINE/PIPEDSL returning 0 over a
period of time longer than VBLANK_EVASION_TIME_US.
For PSR1 we had the same issue so intel_psr_wait_for_idle() was
implemented to wait for PSR1 to get into idle state but nothing was
done for PSR2.
For PSR2 we can't only wait for idle state as PSR2 tends to keep
into sleep state(ready to send selective updates).
Waiting for any state below deep sleep proved to be effective in
avoiding the evasion messages and also not wasted a lot of time.
v2:
- dropping the additional wait_for loops, only the _wait_for_atomic()
is necessary
- waiting for states below EDP_PSR2_STATUS_STATE_DEEP_SLEEP
v3:
- dropping intel_wait_for_condition_atomic() function
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Gwan-gyeong Mun <gwan-gyeong.mun@intel.com>
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Gwan-gyeong Mun <gwan-gyeong.mun@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20211005231851.67698-1-jose.souza@intel.com
Reading out the DP encoders' DPCD during booting or resume is only
required for enabled encoders: such encoders may be modesetted during
the initial commit and the link training this involves depends on an
initialized DPCD. For DDI encoders reading out the DPCD is skipped, do
the same on pre-DDI platforms.
Atm, the first DPCD readout without a sink connected - which is a likely
scneario if the encoder is disabled - leaves intel_dp->num_common_rates
at 0, which resulted in
intel_dp_sync_state()->intel_dp_max_common_rate()
in a
intel_dp->common_rates[-1]
access. This by definition results in an undefined behaviour, though to
my best knowledge in all HW/compiler configurations it actually results
in accessing the array item type value preceding the array. In this
case the preceding value happens to be intel_dp->num_common_rates,
which is 0, so this issue - by luck - didn't cause a user visible
problem.
Nevertheless it's still an undefined behaviour and in CONFIG_UBSAN
builds leads to a kernel BUG() (which revealed this problem for us),
hence CC:stable.
A related problem in case the encoder is enabled but the sink is not
connected or the DPCD readout fails is fixed by the next patch.
v2: Amend the commit message describing the root cause of the
CONFIG_UBSAN BUG().
Fixes: a532cde31d ("drm/i915/tc: Fix TypeC port init/resume time sanitization")
References: https://gitlab.freedesktop.org/drm/intel/-/issues/4297
Reported-and-tested-by: Mat Jonczyk <mat.jonczyk@o2.pl>
Cc: Mat Jonczyk <mat.jonczyk@o2.pl>
Cc: José Roberto de Souza <jose.souza@intel.com>
Cc: Jani Nikula <jani.nikula@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Acked-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20211018094154.1407705-2-imre.deak@intel.com
(cherry picked from commit 4ec5ffc341)
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
8b/10b encoding format requires to reserve the first slot for
recording metadata. Real data transmission starts from the second slot,
with a total of available 63 slots available.
In 128b/132b encoding format, metadata is transmitted separately
in LLCP packet before MTP. Real data transmission starts from
the first slot, with a total of 64 slots available.
v2:
* Move total/start slots to mst_state, and copy it to mst_mgr in
atomic_check
v3:
* Only keep the slot info on the mst_state
* add a start_slot parameter to the payload function, to facilitate non
atomic drivers (this is a temporary workaround and should be removed when
we are moving out the non atomic driver helpers)
v4:
*fixed typo and formatting
v5: (no functional changes)
* Fixed formatting in drm_dp_mst_update_slots()
* Reference mst_state instead of mst_state->mgr for debugging info
Signed-off-by: Bhawanpreet Lakha <Bhawanpreet.Lakha@amd.com>
Signed-off-by: Fangzhi Zuo <Jerry.Zuo@amd.com>
[v5 nitpicks]
Reviewed-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Lyude Paul <lyude@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20211025223825.301703-3-lyude@redhat.com
If we want to return from for_each_intel_connector_iter(), one
way is calling drm_connector_list_iter_end() before returning
to avoid memleak. The other way is just breaking from the bracket
and then returning after the outside drm_connector_list_iter_end().
Obviously, the second way makes code smaller and more clear.
Apply it to the function intel_dp_mst_atomic_master_trans_check().
Signed-off-by: He Ying <heying24@huawei.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20211022022243.138860-1-heying24@huawei.com
perf_parallel_engines is micro benchmark to test i915 request
scheduling. The test creates a thread per physical engine and submits
NOP requests and waits the requests to complete in a loop. In execlists
mode this works perfectly fine as powerful CPU has enough cores to feed
each engine and process the CSBs. With GuC submission the uC gets
overwhelmed as all threads feed into a single CTB channel and the GuC
gets bombarded with CSBs as contexts are immediately switched in and out
on the engines due to the zero runtime of the requests. When the GuC is
overwhelmed scheduling of contexts is unfair due to the nature of the
GuC scheduling algorithm. This behavior is understood and deemed
acceptable as this micro benchmark isn't close to real world use case.
Increasing the timeout of wait period for requests to complete. This
makes the test understand that is ok for contexts to get starved in this
scenario.
A future patch / cleanup may just delete these micro benchmark tests as
they basically mean nothing. We care about real workloads not made up
ones.
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20211011175704.28509-1-matthew.brost@intel.com
We currently just evict lmem objects to system memory when under memory
pressure. For this case we might lack the usual object mm.pages, which
effectively hides the pages from the i915-gem shrinker, until we
actually "attach" the TT to the object, or in the case of lmem-only
objects it just gets migrated back to lmem when touched again.
For all cases we can just adjust the i915 shrinker LRU each time we also
adjust the TTM LRU. The two cases we care about are:
1) When something is moved by TTM, including when initially populating
an object. Importantly this covers the case where TTM moves something from
lmem <-> smem, outside of the normal get_pages() interface, which
should still ensure the shmem pages underneath are reclaimable.
2) When calling into i915_gem_object_unlock(). The unlock should
ensure the object is removed from the shinker LRU, if it was indeed
swapped out, or just purged, when the shrinker drops the object lock.
v2(Thomas):
- Handle managing the shrinker LRU in adjust_lru, where it is always
safe to touch the object.
v3(Thomas):
- Pretty much a re-write. This time piggy back off the shrink_pin
stuff, which actually seems to fit quite well for what we want here.
v4(Thomas):
- Just use a simple boolean for tracking ttm_shrinkable.
v5:
- Ensure we call adjust_lru when faulting the object, to ensure the
pages are visible to the shrinker, if needed.
- Add back the adjust_lru when in i915_ttm_move (Thomas)
v6(Reported-by: kernel test robot <lkp@intel.com>):
- Remove unused i915_tt
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com> #v4
Link: https://patchwork.freedesktop.org/patch/msgid/20211018091055.1998191-6-matthew.auld@intel.com
