In exynos7_decon_resume, When it fails, we must use clk_disable_unprepare()
to free resource that have been used.
Fixes: 6f83d20838 ("drm/exynos: use DRM_DEV_ERROR to print out error
message")
Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Jian Zhang <zhangjian210@huawei.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
Emails to Joonyoung Shim bounce ("550 5.1.1 Recipient address rejected:
User unknown"), so remove him from maintainers of DT bindings (display,
phy).
Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
Emails to Joonyoung Shim bounce ("550 5.1.1 Recipient address rejected:
User unknown"), so move him to credits file.
Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
UAPI Changes:
- Expose per tile media freq factor in sysfs (Ashutosh Dixit, Dale B Stimson)
- Document memory residency and Flat-CCS capability of obj (Ramalingam C)
- Disable GETPARAM lookups of I915_PARAM_[SUB]SLICE_MASK on Xe_HP+ (Matt Roper)
Cross-subsystem Changes:
- Rename intel-gtt symbols (Lucas De Marchi)
Core Changes:
Driver Changes:
- Support programming the EU priority in the GuC descriptor (DG2) (Matthew Brost)
- DG2 HuC loading support (Daniele Ceraolo Spurio)
- Fix build error without CONFIG_PM (YueHaibing)
- Enable THP on Icelake and beyond (Tvrtko Ursulin)
- Only setup private tmpfs mount when needed and fix logging (Tvrtko Ursulin)
- Make __guc_reset_context aware of guilty engines (Umesh Nerlige Ramappa)
- DG2 small bar memory probing fixes (Nirmoy Das)
- Remove unnecessary GuC err capture noise (Alan Previn)
- Fix i915_gem_object_ggtt_pin_ww regression on old platforms (Maarten Lankhorst)
- Fix undefined behavior in GuC backend due to shift overflowing the constant (Borislav Petkov)
- New DG2 workarounds (Swathi Dhanavanthri, Anshuman Gupta)
- Report no hwconfig support on ADL-N (Balasubramani Vivekanandan)
- Fix error_state_read ptr + offset use (Alan Previn)
- Expose per tile media freq factor in sysfs (Ashutosh Dixit, Dale B Stimson)
- Fix memory leaks in per-gt sysfs (Ashutosh Dixit)
- Fix dma_resv fence handling in multi-batch execbuf (Nirmoy Das)
- Add extra registers to GPU error dump on Gen11+ (Stuart Summers)
- More PVC+DG2 workarounds (Matt Roper)
- Improve user experience and driver robustness under SIGINT or similar (Tvrtko Ursulin)
- Don't show engine classes not present (Tvrtko Ursulin)
- Improve on suspend / resume time with VT-d enabled (Thomas Hellström)
- Add missing else (katrinzhou)
- Don't leak lmem mapping in vma_evict (Juha-Pekka Heikkila)
- Add smem fallback allocation for dpt (Juha-Pekka Heikkila)
- Tweak the ordering in cpu_write_needs_clflush (Matthew Auld)
- Do not access rq->engine without a reference (Niranjana Vishwanathapura)
- Revert "drm/i915: Hold reference to intel_context over life of i915_request" (Niranjana Vishwanathapura)
- Don't update engine busyness stats too frequently (Alan Previn)
- Add additional steps for Wa_22011802037 for execlist backend (Umesh Nerlige Ramappa)
- Fix a lockdep warning at error capture (Nirmoy Das)
- Ponte Vecchio prep work and new blitter engines (Matt Roper, John Harrison, Lucas De Marchi)
- Read correct RP_STATE_CAP register (PVC) (Matt Roper)
- Define MOCS table for PVC (Ayaz A Siddiqui)
- Driver refactor and support Ponte Vecchio forcewake handling (Matt Roper)
- Remove additional 3D flags from PIPE_CONTROL (Ponte Vecchio) (Stuart Summers)
- XEHPSDV and PVC do not use HuC (Daniele Ceraolo Spurio)
- Extract stepping information from PCI revid (Ponte Vecchio) (Matt Roper)
- Add initial PVC workarounds (Stuart Summers)
- SSEU handling driver refactor and Ponte Vecchio support (Matt Roper)
- GuC depriv applies to PVC (Matt Roper)
- Add register steering (Ponte Vecchio) (Matt Roper)
- Add recommended MMIO setting (Ponte Vecchio) (Matt Roper)
- Move multicast register handling to a dedicated file (Matt Roper)
- Cleanup interface for MCR operations (Matt Roper)
- Extend i915_vma_pin_iomap() (CQ Tang)
- Re-do the intel-gtt split (Lucas De Marchi)
- Correct duplicated/misplaced GT register definitions (Matt Roper)
- Prefer "XEHP_" prefix for registers (Matt Roper)
- Don't use DRM_DEBUG_WARN_ON for unexpected l3bank/mslice config (Tvrtko Ursulin)
- Don't use DRM_DEBUG_WARN_ON for ring unexpectedly not idle (Tvrtko Ursulin)
- Make drop_pages() return bool (Lucas De Marchi)
- Fix CFI violation with show_dynamic_id() (Nathan Chancellor)
- Use i915_probe_error instead of drm_error in GuC code (Vinay Belgaumkar)
- Fix use of static in macro mismatch (Andi Shyti)
- Update tiled blits selftest (Bommu Krishnaiah)
- Future-proof platform checks (Matt Roper)
- Only include what's needed (Jani Nikula)
- remove accidental static from a local variable (Jani Nikula)
- Add global forcewake request to drpc (Vinay Belgaumkar)
- Fix spelling typo in comment (pengfuyuan)
- Increase timeout for live_parallel_switch selftest (Akeem G Abodunrin)
- Use non-blocking H2G for waitboost (Vinay Belgaumkar)
Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/YrwtLM081SQUG1Dc@tursulin-desk
Like was done for read, provide the equivalent for write. Even if
current users are not in the hot path, this should future-proof it.
v2:
- Remove default from _Generic() - callers wanting to write more
than u64 should use iosys_map_memcpy_to()
- Add WRITE_ONCE() cases dereferencing the pointer when using system
memory
v3:
- Fix precedence issue when casting inside WRITE_ONCE(). By not using ()
around vaddr__ the offset was not part of the cast, but rather added
to it, producing a wrong address
- Remove compiletime_assert() as WRITE_ONCE() already contains it
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Reviewed-by: Christian König <christian.koenig@amd.com> # v1
Reviewed-by: Thomas Zimmermann <tzimmermann@suse.de>
Link: https://patchwork.freedesktop.org/patch/msgid/20220628191016.3899428-2-lucas.demarchi@intel.com
Instead of always falling back to memcpy_fromio() for any size, prefer
using read{b,w,l}(). When reading struct members it's common to read
individual integer variables individually. Going through memcpy_fromio()
for each of them poses a high penalty.
Employ a similar trick as __seqprop() by using _Generic() to generate
only the specific call based on a type-compatible variable.
For a pariticular i915 workload producing GPU context switches,
__get_engine_usage_record() is particularly hot since the engine usage
is read from device local memory with dgfx, possibly multiple times
since it's racy. Test execution time for this test shows a ~12.5%
improvement with DG2:
Before:
nrepeats = 1000; min = 7.63243e+06; max = 1.01817e+07;
median = 9.52548e+06; var = 526149;
After:
nrepeats = 1000; min = 7.03402e+06; max = 8.8832e+06;
median = 8.33955e+06; var = 333113;
Other things attempted that didn't prove very useful:
1) Change the _Generic() on x86 to just dereference the memory address
2) Change __get_engine_usage_record() to do just 1 read per loop,
comparing with the previous value read
3) Change __get_engine_usage_record() to access the fields directly as it
was before the conversion to iosys-map
(3) did gave a small improvement (~3%), but doesn't seem to scale well
to other similar cases in the driver.
Additional test by Chris Wilson using gem_create from igt with some
changes to track object creation time. This happens to accidentally
stress this code path:
Pre iosys_map conversion of engine busyness:
lmem0: Creating 262144 4KiB objects took 59274.2ms
Unpatched:
lmem0: Creating 262144 4KiB objects took 108830.2ms
With readl (this patch):
lmem0: Creating 262144 4KiB objects took 61348.6ms
s/readl/READ_ONCE/
lmem0: Creating 262144 4KiB objects took 61333.2ms
So we do take a little bit more time than before the conversion, but
that is due to other factors: bringing the READ_ONCE back would be as
good as just doing this conversion.
v2:
- Remove default from _Generic() - callers wanting to read more
than u64 should use iosys_map_memcpy_from()
- Add READ_ONCE() cases dereferencing the pointer when using system
memory
v3:
- Fix precedence issue when casting inside READ_ONCE(). By not using ()
around vaddr__ the offset was not part of the cast, but rather added
to it, producing a wrong address
- Remove compiletime_assert() as READ_ONCE() already contains it
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Christian König <christian.koenig@amd.com> # v1
Reviewed-by: Thomas Zimmermann <tzimmermann@suse.de>
Link: https://patchwork.freedesktop.org/patch/msgid/20220628191016.3899428-1-lucas.demarchi@intel.com
With the change to 2 pixels/clock, the pixel doubling in the PV
results in doubling each pair of pixels, ie ABABCDCD instead of
AABBCCDD.
Move the pixel doubling to the HDMI block, however this means
that DBLCLK modes now fall foul of requiring even values for
all the horizontal timing parameters.
As both 480i and 576i fail this, attempt to fix up DBLCLK modes
that have odd timings values.
Fixes: 8323989140 ("drm/vc4: hdmi: Support the BCM2711 HDMI controllers")
Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
Link: https://lore.kernel.org/r/20220613144800.326124-34-maxime@cerno.tech
Signed-off-by: Maxime Ripard <maxime@cerno.tech>
Whenever the maximum BPC is changed, vc4_hdmi_encoder_compute_config()
might pick up a different BPC or format depending on the display
capabilities.
That change will have a number of side effects, including the clock
rates and whether the scrambling is enabled.
However, only drm_crtc_state.connectors_changed will be set to true,
since that properly only affects the connector.
This means that while drm_atomic_crtc_needs_modeset() will return true,
and thus drm_atomic_helper_commit_modeset_enables() will call our
encoder atomic_enable() hook, mode_changed will be false.
So crtc_set_mode() will not call our encoder .atomic_mode_set() hook. We
use this hook in vc4 to set the vc4_hdmi_connector_state.output_bpc (and
output_format), and will then reuse the value in .atomic_enable() to select
whether or not scrambling should be enabled.
However, since our clock rate is pre-computed during .atomic_check(), we
end up with the clocks properly configured, but the scrambling disabled,
leading to a blank screen.
Let's set mode_changed to true in our HDMI driver to force the update of
output_bpc, and thus prevent the issue entirely.
Fixes: ba8c0faebb ("drm/vc4: hdmi: Enable 10/12 bpc output")
Signed-off-by: Dom Cobley <popcornmix@gmail.com>
Link: https://lore.kernel.org/r/20220613144800.326124-32-maxime@cerno.tech
Signed-off-by: Maxime Ripard <maxime@cerno.tech>
Increase the number of post-sync blanking lines on odd fields instead of
decreasing it on even fields. This makes the total number of lines
properly match the modelines.
Additionally fix the value of PV_VCONTROL_ODD_DELAY, which did not take
pixels_per_clock into account, causing some displays to invert the
fields when driven by bcm2711.
Fixes: 682e62c454 ("drm/vc4: Fix support for interlaced modes on HDMI.")
Signed-off-by: Mateusz Kwiatkowski <kfyatek+publicgit@gmail.com>
Link: https://lore.kernel.org/r/20220613144800.326124-31-maxime@cerno.tech
Signed-off-by: Maxime Ripard <maxime@cerno.tech>
The current HDMI driver, in vc4_hdmi_audio_can_stream() checks whether
the display output is enabled.
This has been there in one form or the other since the introduction of
the audio support in the VC4 HDMI driver in commit bb7d785688
("drm/vc4: Add HDMI audio support"), but no justification for this check
is in the commit message, or in the discussions around the patches.
One can only assume this was done to prevent a user from playing audio
on the ALSA soundcard when the monitor doesn't support it.
However, this is causing some issues. Indeed, Kodi, for example, was
hitting some errors if it was streaming audio during a modeset. With the
theory above, it does make sense, but the display and audio threads are
typically completely different processes with no opportunity to
synchronise which makes it hard to workaround.
Removing that check also doesn't seem to cause any trouble, so let's
just remove it.
Signed-off-by: Dom Cobley <popcornmix@gmail.com>
Link: https://lore.kernel.org/r/20220613144800.326124-25-maxime@cerno.tech
Signed-off-by: Maxime Ripard <maxime@cerno.tech>
The BCM2835-37 found in the RaspberryPi 0 to 3 have a power domain
attached to the HDMI block, handled in Linux through runtime_pm.
That power domain is shared with the VEC block, so even if we put our
runtime_pm reference in the HDMI driver it would keep being on. If the
VEC is disabled though, the power domain would be disabled and we would
lose any initialization done in our bind implementation.
That initialization involves calling the reset function and initializing
the CEC registers.
Let's move the initialization to our runtime_resume implementation so
that we initialize everything properly if we ever need to.
Fixes: c86b412143 ("drm/vc4: hdmi: Move the HSM clock enable to runtime_pm")
Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
Link: https://lore.kernel.org/r/20220613144800.326124-24-maxime@cerno.tech
Signed-off-by: Maxime Ripard <maxime@cerno.tech>
If the controller isn't clocked or its domain powered up, the register
accesses will either stall the CPU or return garbage, respectively.
Thus, we had a warning in our register access function to complain when
that kind of risky accesses were performed.
In order to check the runtime_pm power state, we were using
pm_runtime_active(), but it turns out that it will become active only
once the runtime_resume hook has been executed.
This prevents us from doing any WARN-free register access in our
runtime_resume() implementation, while this is valid.
Let's switch to pm_runtime_status_suspended() instead.
Fixes: 14e193b956 ("drm/vc4: hdmi: Warn if we access the controller while disabled")
Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
Link: https://lore.kernel.org/r/20220613144800.326124-23-maxime@cerno.tech
Signed-off-by: Maxime Ripard <maxime@cerno.tech>
The HDMI block can repeat pixels for double clocked modes,
and the firmware is now configuring the block to do this as
the PV is doing it incorrectly when at 2pixels/clock.
If the kernel doesn't reset it then we end up with strange
modes.
Reset MISC_CONTROL.
Fixes: 8323989140 ("drm/vc4: hdmi: Support the BCM2711 HDMI controllers")
Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
Link: https://lore.kernel.org/r/20220613144800.326124-22-maxime@cerno.tech
Signed-off-by: Maxime Ripard <maxime@cerno.tech>
The dmas property is used to hold the dmaengine channel used for audio
output.
Older device trees were missing that property, so if it's not there we
disable the audio output entirely.
However, some overlays have set an empty value to that property, mostly
to workaround the fact that overlays cannot remove a property. Let's add
a test for that case and if it's empty, let's disable it as well.
Cc: <stable@vger.kernel.org>
Signed-off-by: Phil Elwell <phil@raspberrypi.org>
Link: https://lore.kernel.org/r/20220613144800.326124-18-maxime@cerno.tech
Signed-off-by: Maxime Ripard <maxime@cerno.tech>
The divider calculations tried to find the divider just faster than the
clock requested. However if it required a divider of 7 then the for loop
aborted without handling the "error" case, and could end up with a clock
lower than requested.
The integer divider from parent PLL to DSI clock is also capable of
going up to /255, not just /7 that the driver was trying. This allows
for slower link frequencies on the DSI bus where the resolution permits.
Correct the loop so that we always have a clock greater than requested,
and covering the whole range of dividers.
Fixes: 86c1b9eff3 ("drm/vc4: Adjust modes in DSI to work around the integer PLL divider.")
Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
Link: https://lore.kernel.org/r/20220613144800.326124-13-maxime@cerno.tech
Signed-off-by: Maxime Ripard <maxime@cerno.tech>
There is little harm in ignoring fractional coordinates
(they just get truncated).
Without this:
modetest -M vc4 -F tiles,gradient -s 32:1920x1080-60 -P89@74:1920x1080*.1.1@XR24
is rejected. We have the same issue in Kodi when trying to
use zoom options on video.
Note: even if all coordinates are fully integer. e.g.
src:[0,0,1920,1080] dest:[-10,-10,1940,1100]
it will still get rejected as drm_atomic_helper_check_plane_state
uses drm_rect_clip_scaled which transforms this to fractional src coords
Fixes: 21af94cf1a ("drm/vc4: Add support for scaling of display planes.")
Signed-off-by: Dom Cobley <popcornmix@gmail.com>
Link: https://lore.kernel.org/r/20220613144800.326124-5-maxime@cerno.tech
Signed-off-by: Maxime Ripard <maxime@cerno.tech>
The core clock computation takes into account both the load due to the
input (ie, planes) and its output (ie, encoders).
However, while the input load needs to consider all the planes, and thus
sum all of their associated loads, the output happens mostly in
parallel.
Therefore, we need to consider only the maximum of all the output loads,
and not the sum like we were doing. This resulted in a clock rate way
too high which could be discarded for being too high by the clock
framework.
Since recent changes, the clock framework will even downright reject it,
leading to a core clock being too low for its current needs.
Fixes: 16e101051f ("drm/vc4: Increase the core clock based on HVS load")
Signed-off-by: Maxime Ripard <maxime@cerno.tech>
Reviewed-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
Link: https://lore.kernel.org/r/20220613144800.326124-4-maxime@cerno.tech
vc4_drv isn't necessarily under the /soc node in DT as it is a
virtual device, but it is the one that does the allocations.
The DMA addresses are consumed by primarily the HVS or V3D, and
those require VideoCore cache alias address mapping, and so will be
under /soc.
During probe find the a suitable device node for HVS or V3D,
and adopt the DMA configuration of that node.
Cc: <stable@vger.kernel.org>
Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
Link: https://lore.kernel.org/r/20220613144800.326124-2-maxime@cerno.tech
Signed-off-by: Maxime Ripard <maxime@cerno.tech>
SLPC min/max frequency updates require H2G calls. We are seeing
timeouts when GuC channel is backed up and it is unable to respond
in a timely fashion causing warnings and affecting CI.
This is seen when waitboosting happens during a stress test.
this patch updates the waitboost path to use a non-blocking
H2G call instead, which returns as soon as the message is
successfully transmitted.
v2: Use drm_notice to report any errors that might occur while
sending the waitboost H2G request (Tvrtko)
v3: Add drm_notice inside force_min_freq (Ashutosh)
Cc: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: Vinay Belgaumkar <vinay.belgaumkar@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220623003225.23301-1-vinay.belgaumkar@intel.com
For execlists backend, current implementation of Wa_22011802037 is to
stop the CS before doing a reset of the engine. This WA was further
extended to wait for any pending MI FORCE WAKEUPs before issuing a
reset. Add the extended steps in the execlist path of reset.
In addition, extend the WA to gen11.
v2: (Tvrtko)
- Clarify comments, commit message, fix typos
- Use IS_GRAPHICS_VER for gen 11/12 checks
v3: (Daneile)
- Drop changes to intel_ring_submission since WA does not apply to it
- Log an error if MSG IDLE is not defined for an engine
Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Fixes: f6aa0d713c ("drm/i915: Add Wa_22011802037 force cs halt")
Acked-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220621192105.2100585-1-umesh.nerlige.ramappa@intel.com
Using two different types of workoads, it was observed that
guc_update_engine_gt_clks was being called too frequently and/or
causing a CPU-to-lmem bandwidth hit over PCIE. Details on
the workloads and numbers are in the notes below.
Background: At the moment, guc_update_engine_gt_clks can be invoked
via one of 3 ways. #1 and #2 are infrequent under normal operating
conditions:
1.When a predefined "ping_delay" timer expires so that GuC-
busyness can sample the GTPM clock counter to ensure it
doesn't miss a wrap-around of the 32-bits of the HW counter.
(The ping_delay is calculated based on 1/8th the time taken
for the counter go from 0x0 to 0xffffffff based on the
GT frequency. This comes to about once every 28 seconds at a
GT frequency of 19.2Mhz).
2.In preparation for a gt reset.
3.In response to __gt_park events (as the gt power management
puts the gt into a lower power state when there is no work
being done).
Root-cause: For both the workloads described farther below, it was
observed that when user space calls IOCTLs that unparks the
gt momentarily and repeats such calls many times in quick succession,
it triggers calling guc_update_engine_gt_clks as many times. However,
the primary purpose of guc_update_engine_gt_clks is to ensure we don't
miss the wraparound while the counter is ticking. Thus, the solution
is to ensure we skip that check if gt_park is calling this function
earlier than necessary.
Solution: Snapshot jiffies when we do actually update the busyness
stats. Then get the new jiffies every time intel_guc_busyness_park
is called and bail if we are being called too soon. Use half of the
ping_delay as a safe threshold.
NOTE1: Workload1: IGTs' gem_create was modified to create a file handle,
allocate memory with sizes that range from a min of 4K to the max supported
(in power of two step-sizes). Its maps, modifies and reads back the
memory. Allocations and modification is repeated until total memory
allocation reaches the max. Then the file handle is closed. With this
workload, guc_update_engine_gt_clks was called over 188 thousand times
in the span of 15 seconds while this test ran three times. With this patch,
the number of calls reduced to 14.
NOTE2: Workload2: 30 transcode sessions are created in quick succession.
While these sessions are created, pcm-iio tool was used to measure I/O
read operation bandwidth consumption sampled at 100 milisecond intervals
over the course of 20 seconds. The total bandwidth consumed over 20 seconds
without this patch was measured at average at 311KBps per sample. With this
patch, the number went down to about 175Kbps which is about a 43% savings.
Signed-off-by: Alan Previn <alan.previn.teres.alexis@intel.com>
Reviewed-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Acked-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220623023157.211650-2-alan.previn.teres.alexis@intel.com
