Before this commit when we want userspace to use the acpi_video backlight
device we register both the GPU's native backlight device and acpi_video's
firmware acpi_video# backlight device. This relies on userspace preferring
firmware type backlight devices over native ones.
Registering 2 backlight devices for a single display really is
undesirable, don't register the GPU's native backlight device when
another backlight device should be used.
Changes in v2:
- Use drm_info(drm_dev, ...) for log messages
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Host Turbo operates at efficient frequency when GT is not idle unless
the user or workload has forced it to a higher level. Replicate the same
behavior in SLPC by allowing the algorithm to use efficient frequency.
We had disabled it during boot due to concerns that it might break
kernel ABI for min frequency. However, this is not the case since
SLPC will still abide by the (min,max) range limits.
With this change, min freq will be at efficient frequency level at init
instead of fused min (RPn). If user chooses to reduce min freq below the
efficient freq, we will turn off usage of efficient frequency and honor
the user request. When a higher value is written, it will get toggled
back again.
The patch also corrects the register which needs to be read for obtaining
the correct efficient frequency for Gen9+.
We see much better perf numbers with benchmarks like glmark2 with
efficient frequency usage enabled as expected.
v2: Address review comments (Rodrigo)
v3: with efficient frequency being dynamic, it is possible that the req
frequency may go beyond max freq. This will cause SLPC selftests to fail.
Add a FIXME there to start the test with [RPn, RP0] instead and restore
it afterwards.
BugLink: https://gitlab.freedesktop.org/drm/intel/-/issues/5468
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Vinay Belgaumkar <vinay.belgaumkar@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220820010832.15350-1-vinay.belgaumkar@intel.com
Now that we've finally gotten rid of the non-atomic MST users leftover in
the kernel, we can finally get rid of all of the legacy payload code we
have and move as much as possible into the MST atomic state structs. The
main purpose of this is to make the MST code a lot less confusing to work
on, as there's a lot of duplicated logic that doesn't really need to be
here. As well, this should make introducing features like fallback link
retraining and DSC support far easier.
Since the old payload code was pretty gnarly and there's a Lot of changes
here, I expect this might be a bit difficult to review. So to make things
as easy as possible for reviewers, I'll sum up how both the old and new
code worked here (it took me a while to figure this out too!).
The old MST code basically worked by maintaining two different payload
tables - proposed_vcpis, and payloads. proposed_vcpis would hold the
modified payload we wanted to push to the topology, while payloads held the
payload table that was currently programmed in hardware. Modifications to
proposed_vcpis would be handled through drm_dp_allocate_vcpi(),
drm_dp_mst_deallocate_vcpi(), and drm_dp_mst_reset_vcpi_slots(). Then, they
would be pushed via drm_dp_mst_update_payload_step1() and
drm_dp_mst_update_payload_step2().
Furthermore, it's important to note how adding and removing VC payloads
actually worked with drm_dp_mst_update_payload_step1(). When a VC payload
is removed from the VC table, all VC payloads which come after the removed
VC payload's slots must have their time slots shifted towards the start of
the table. The old code handles this by looping through the entire payload
table and recomputing the start slot for every payload in the topology from
scratch. While very much overkill, this ends up doing the right thing
because we always order the VCPIs for payloads from first to last starting
timeslot.
It's important to also note that drm_dp_mst_update_payload_step2() isn't
actually limited to updating a single payload - the driver can use it to
queue up multiple payload changes so that as many of them can be sent as
possible before waiting for the ACT. This is -technically- not against
spec, but as Wayne Lin has pointed out it's not consistently implemented
correctly in hubs - so it might as well be.
drm_dp_mst_update_payload_step2() is pretty self explanatory and basically
the same between the old and new code, save for the fact we don't have a
second step for deleting payloads anymore -and thus rename it to
drm_dp_mst_add_payload_step2().
The new payload code stores all of the current payload info within the MST
atomic state and computes as much of the state as possible ahead of time.
This has the one exception of the starting timeslots for payloads, which
can't be determined at atomic check time since the starting time slots will
vary depending on what order CRTCs are enabled in the atomic state - which
varies from driver to driver. These are still stored in the atomic MST
state, but are only copied from the old MST state during atomic commit
time. Likewise, this is when new start slots are determined.
Adding/removing payloads now works much more closely to how things are
described in the spec. When we delete a payload, we loop through the
current list of payloads and update the start slots for any payloads whose
time slots came after the payload we just deleted. Determining the starting
time slots for new payloads being added is done by simply keeping track of
where the end of the VC table is in
drm_dp_mst_topology_mgr->next_start_slot. Additionally, it's worth noting
that we no longer have a single update_payload() function. Instead, we now
have drm_dp_mst_add_payload_step1|2() and drm_dp_mst_remove_payload(). As
such, it's now left it up to the driver to figure out when to add or remove
payloads. The driver already knows when it's disabling/enabling CRTCs, so
it also already knows when payloads should be added or removed.
Changes since v1:
* Refactor around all of the completely dead code changes that are
happening in amdgpu for some reason when they really shouldn't even be
there in the first place… :\
* Remove mention of sending one ACT per series of payload updates. As Wayne
Lin pointed out, there are apparently hubs on the market that don't work
correctly with this scheme and require a separate ACT per payload update.
* Fix accidental drop of mst_mgr.lock - Wayne Lin
* Remove mentions of allowing multiple ACT updates per payload change,
mention that this is a result of vendors not consistently supporting this
part of the spec and requiring a unique ACT for each payload change.
* Get rid of reference to drm_dp_mst_port in DC - turns out I just got
myself confused by DC and we don't actually need this.
Changes since v2:
* Get rid of fix for not sending payload deallocations if ddps=0 and just
go back to wayne's fix
Signed-off-by: Lyude Paul <lyude@redhat.com>
Cc: Wayne Lin <Wayne.Lin@amd.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Fangzhi Zuo <Jerry.Zuo@amd.com>
Cc: Jani Nikula <jani.nikula@intel.com>
Cc: Imre Deak <imre.deak@intel.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Sean Paul <sean@poorly.run>
Acked-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220817193847.557945-18-lyude@redhat.com
There's another kind of situation where we could potentially race with
nonblocking modesets and MST, especially if we were to only use the locking
provided by atomic modesetting:
* Display 1 begins as enabled on DP-1 in SST mode
* Display 1 switches to MST mode, exposes one sink in MST mode
* Userspace does non-blocking modeset to disable the SST display
* Userspace does non-blocking modeset to enable the MST display with a
different CRTC, but the SST display hasn't been fully taken down yet
* Execution order between the last two commits isn't guaranteed since they
share no drm resources
We can fix this however, by ensuring that we always pull in the atomic
topology state whenever a connector capable of driving an MST display
performs its atomic check - and then tracking CRTC commits happening on the
SST connector in the MST topology state. So, let's add some simple helpers
for doing that and hook them up in various drivers.
v2:
* Use intel_dp_mst_source_support() to check for MST support in i915, fixes
CI failures
Signed-off-by: Lyude Paul <lyude@redhat.com>
Cc: Wayne Lin <Wayne.Lin@amd.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Fangzhi Zuo <Jerry.Zuo@amd.com>
Cc: Jani Nikula <jani.nikula@intel.com>
Cc: Imre Deak <imre.deak@intel.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Sean Paul <sean@poorly.run>
Acked-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220817193847.557945-14-lyude@redhat.com
As Daniel Vetter pointed out, if we only use the atomic modesetting locks
with MST it's technically possible for a driver with non-blocking modesets
to race when it comes to MST displays - as we make the mistake of not doing
our own CRTC commit tracking in the topology_state object.
This could potentially cause problems if something like this happens:
* User starts non-blocking commit to disable CRTC-1 on MST topology 1
* User starts non-blocking commit to enable CRTC-2 on MST topology 1
There's no guarantee here that the commit for disabling CRTC-2 will only
occur after CRTC-1 has finished, since neither commit shares a CRTC - only
the private modesetting object for MST. Keep in mind this likely isn't a
problem for blocking modesets, only non-blocking.
So, begin fixing this by keeping track of which CRTCs on a topology have
changed by keeping track of which CRTCs we release or allocate timeslots
on. As well, add some helpers for:
* Setting up the drm_crtc_commit structs in the ->commit_setup hook
* Waiting for any CRTC dependencies from the previous topology state
v2:
* Use drm_dp_mst_atomic_setup_commit() directly - Jani
Signed-off-by: Lyude Paul <lyude@redhat.com>
Cc: Wayne Lin <Wayne.Lin@amd.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Fangzhi Zuo <Jerry.Zuo@amd.com>
Cc: Jani Nikula <jani.nikula@intel.com>
Cc: Imre Deak <imre.deak@intel.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Sean Paul <sean@poorly.run>
Acked-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220817193847.557945-9-lyude@redhat.com
VCPI is only sort of the correct term here, originally the majority of this
code simply referred to timeslots vaguely as "slots" - and since I started
working on it and adding atomic functionality, the name "VCPI slots" has
been used to represent time slots.
Now that we actually have consistent access to the DisplayPort spec thanks
to VESA, I now know this isn't actually the proper term - as the
specification refers to these as time slots.
Since we're trying to make this code as easy to figure out as possible,
let's take this opportunity to correct this nomenclature and call them by
their proper name - timeslots. Likewise, we rename various functions
appropriately, along with replacing references in the kernel documentation
and various debugging messages.
It's important to note that this patch series leaves the legacy MST code
untouched for the most part, which is fine since we'll be removing it soon
anyhow. There should be no functional changes in this series.
v2:
* Add note that Wayne Lin from AMD suggested regarding slots being between
the source DP Tx and the immediate downstream DP Rx
Signed-off-by: Lyude Paul <lyude@redhat.com>
Cc: Wayne Lin <Wayne.Lin@amd.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Fangzhi Zuo <Jerry.Zuo@amd.com>
Cc: Jani Nikula <jani.nikula@intel.com>
Cc: Imre Deak <imre.deak@intel.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Sean Paul <sean@poorly.run>
Acked-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220817193847.557945-5-lyude@redhat.com
Commit 20f85ef89d ("drm/i915/backlight: use unique backlight device
names") added support for multiple backlight devices on dual panel
systems, but did so with error handling on -EEXIST from
backlight_device_register(). Unfortunately, that triggered a warning in
dmesg all the way down from sysfs_add_file_mode_ns() and
sysfs_warn_dup().
Instead of optimistically always attempting to register with the default
name ("intel_backlight", which we have to retain for backward
compatibility), check if a backlight device with the name exists first,
and, if so, use the card and connector based name.
v2: reworked on top of the patch commit 20f85ef89d
("drm/i915/backlight: use unique backlight device names")
v3: fixed the ref count leak(Jani N)
Fixes: 20f85ef89d ("drm/i915/backlight: use unique backlight device names")
Signed-off-by: Arun R Murthy <arun.r.murthy@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220808035750.3111046-1-arun.r.murthy@intel.com
MTL introduces a second FBC engine. The two FBC engines can
operate entirely independently, FBC A serving pipe A and
FBC B serving pipe B.
The one place where things might go a bit wrong is the CFB
allocation from stolen. We might have to consider some change
to the allocation strategy to have a better chance of both
engines being able to allocate its CFB. Maybe FBC A should
allocate bottom up and FBC B top down, or something? For the
moment the allocation strategy is DRM_MM_INSERT_BEST for both.
Cc: Mika Kahola <mika.kahola@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220817122624.213889-1-jani.nikula@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
If the GuC CTs are full and we need to stall the request submission
while waiting for space, we save the stalled request and where the stall
occurred; when the CTs have space again we pick up the request submission
from where we left off.
If a full GT reset occurs, the state of all contexts is cleared and all
non-guilty requests are unsubmitted, therefore we need to restart the
stalled request submission from scratch. To make sure that we do so,
clear the saved request after a reset.
Fixes note: the patch that introduced the bug is in 5.15, but no
officially supported platform had GuC submission enabled by default
in that kernel, so the backport to that particular version (and only
that one) can potentially be skipped.
Fixes: 925dc1cf58 ("drm/i915/guc: Implement GuC submission tasklet")
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: John Harrison <john.c.harrison@intel.com>
Cc: <stable@vger.kernel.org> # v5.15+
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220811210812.3239621-1-daniele.ceraolospurio@intel.com
