Move reset_context out of gpu recover function to make it configurable
for different reset purpose.
For the reset way of call gpu_recovery sysfs, force to use full reset
method. Otherwise, try soft reset by default if the related ASIC
supportted, if soft reset failed, will use full reset.
Signed-off-by: Likun Gao <Likun.Gao@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Support SDMA soft reset for SDMA v6.
V3: use ib test to check soft reset.
V4: squash in unused variable fix (Alex)
Signed-off-by: Likun Gao <Likun.Gao@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
Status flags definition is reduced to read
less bytes in SCDC transaction for status update.
[How]
Reduce definition of reserved bytes from 3 to 1
for status update.
Reviewed-by: Charlene Liu <Charlene.Liu@amd.com>
Acked-by: Solomon Chiu <solomon.chiu@amd.com>
Signed-off-by: Chris Park <chris.park@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
On the GC 10.3.7 platform the initial MEC release version #3 can support
atomic operation,so need correct and set its MEC atomic support version to #3.
Signed-off-by: Prike Liang <Prike.Liang@amd.com>
Reviewed-by: Aaron Liu <aaron.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[why]
Ideally link capability should be independent from the link
configuration that we decide to use in enable link. Otherwise if link
capability is changed after validation has completed, we could end up
enabling a link configuration with invalid configuration. This would
lead to over link bandwidth subscription or in the extreme case
causes us to enable HPO link to a DIO stream.
[how]
Add a new struct in pipe ctx called link config. This structure will
contain link configuration to enable a link. It will be populated
during map pool resources after we validate link bandwidth. Remove
the reference of verified link cap during enable link process and
use link config in pipe ctx instead.
Reviewed-by: George Shen <George.Shen@amd.com>
Acked-by: Solomon Chiu <solomon.chiu@amd.com>
Signed-off-by: Wenjing Liu <wenjing.liu@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[why]
First MST sideband message returns AUX_RET_ERROR_HPD_DISCON
on certain intel platform. Aux transaction considered failure
if HPD unexpected pulled low. The actual aux transaction success
in such case, hence do not return error.
[how]
Not returning error when AUX_RET_ERROR_HPD_DISCON detected
on the first sideband message.
v2: squash in additional DMI entries
v3: squash in static fix
Signed-off-by: Fangzhi Zuo <Jerry.Zuo@amd.com>
Acked-by: Solomon Chiu <solomon.chiu@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Didn't really know what this buffer was when initially implemented,
but these days we do, so move it somewhere more appropriate.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Doesn't fix any known issue, but noticed fifo being initialised in
logs in response to mmu allocation.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Fixes resume from hibernate failing on (at least) TU102, where cursor
channel init failed due to being performed before the core channel.
Not solid idea why suspend-to-ram worked, but, presumably HW being in
an entirely clean state has something to do with it.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Userspace never ended up using this to be clever about dealing with
channel death, and it won't be, not like this anyway.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Next for v5.20
GPU:
- a619 support
- Fix for unclocked GMU register access
- Devcore dump enhancements
Core:
- client utilization via fdinfo support
- fix fence rollover issue
- gem: Lockdep false-positive warning fix
- gem: Switch to pfn mappings
DPU:
- constification of HW catalog
- support for using encoder as CRC source
- WB support on sc7180
- WB resolution fixes
DP:
- dropped custom bulk clock implementation
- made dp_bridge_mode_valid() return MODE_CLOCK_HIGH where applicable
- fix link retraining on resolution change
MDP5:
- MSM8953 perf data
HDMI:
- YAML'ification of schema
- dropped obsolete GPIO support
- misc cleanups
Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Rob Clark <robdclark@gmail.com>
Link: https://patchwork.freedesktop.org/patch/msgid/CAF6AEGtuqswBGPw-kCYzJvckK2RR1XTeUEgaXwVG_mvpbv3gPA@mail.gmail.com
One impact of commit 047a1b877e ("dma-buf & drm/amdgpu: remove
dma_resv workaround") is that it stores many, many more fences. Whereas
adding an exclusive fence used to remove the shared fence list, that
list is now preserved and the write fences included into the list. Not
just a single write fence, but now a write/read fence per context. That
causes us to have to track more fences than before (albeit half of those
are redundant), and we trigger more interrupts for multi-engine
workloads.
As part of reducing the impact from handling more signaling, we observe
we only need to kick the signal worker after adding a fence iff we have
good cause to believe that there is work to be done in processing the
fence i.e. we either need to enable the interrupt or the request is
already complete but we don't know if we saw the interrupt and so need
to check signaling.
References: 047a1b877e ("dma-buf & drm/amdgpu: remove dma_resv workaround")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Karolina Drobnik <karolina.drobnik@intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/d7b953c7a4ba747c8196a164e2f8c5aef468d048.1657289332.git.karolina.drobnik@intel.com
We employ a "waitboost" heuristic to detect when userspace is stalled
waiting for results from earlier execution. Under latency sensitive work
mixed between the gpu/cpu, the GPU is typically under-utilised and so
RPS sees that low utilisation as a reason to downclock the frequency,
causing longer stalls and lower throughput. The user left waiting for
the results is not impressed.
On applying commit 047a1b877e ("dma-buf & drm/amdgpu: remove dma_resv
workaround") it was observed that deinterlacing h264 on Haswell
performance dropped by 2-5x. The reason being that the natural workload
was not intense enough to trigger RPS (using HW evaluation intervals) to
upclock, and so it was depending on waitboosting for the throughput.
Commit 047a1b877e ("dma-buf & drm/amdgpu: remove dma_resv workaround")
changes the composition of dma-resv from keeping a single write fence +
multiple read fences, to a single array of multiple write and read
fences (a maximum of one pair of write/read fences per context). The
iteration order was also changed implicitly from all-read fences then
the single write fence, to a mix of write fences followed by read
fences. It is that ordering change that belied the fragility of
waitboosting.
Currently, a waitboost is inspected at the point of waiting on an
outstanding fence. If the GPU is backlogged such that we haven't yet
stated the request we need to wait on, we force the GPU to upclock until
the completion of that request. By changing the order in which we waited
upon requests, we ended up waiting on those requests in sequence and as
such we saw that each request was already started and so not a suitable
candidate for waitboosting.
Instead of asking whether to boost each fence in turn, we can look at
whether boosting is required for the dma-resv ensemble prior to waiting
on any fence, making the heuristic more robust to the order in which
fences are stored in the dma-resv.
Reported-by: Thomas Voegtle <tv@lio96.de>
Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/6284
Fixes: 047a1b877e ("dma-buf & drm/amdgpu: remove dma_resv workaround")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Karolina Drobnik <karolina.drobnik@intel.com>
Tested-by: Thomas Voegtle <tv@lio96.de>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Acked-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/07e05518d9f6620d20cc1101ec1849203fe973f9.1657289332.git.karolina.drobnik@intel.com
[Why]
When playing NV12 1080p MPO video, it is pipe splitting so
we see two pipes in fullscreen and four pipes in windowed
mode. Pipe split is happening because we are setting
MaximumMPCCombine = 1
[How]
Algorithm for MaximumMPCCombine has extra conditions we do
not need. Use DCN31 algorithm instead
Signed-off-by: Samson Tam <Samson.Tam@amd.com>
Reviewed-by: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
Black screen encountered when disabling Freesync through OSD on some
displays.
[How]
Set the should_disable flag when new top pipe has no plane state to
ensure that pipes get cleaned up.
Reviewed-by: Chris Park <Chris.Park@amd.com>
Acked-by: Solomon Chiu <solomon.chiu@amd.com>
Signed-off-by: Ilya Bakoulin <Ilya.Bakoulin@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
When unplug one sst monitor from a mst hub and plug in the same
port with another sst monitor, we don't read the corresponding
edid. That's because we detect there is already an edid stored in
aconnector->edid which is a stale one.
[How]
Clean up aconnector->edid when unplug mst connector.
Reviewed-by: Hersen Wu <hersen.wu@amd.com>
Acked-by: Solomon Chiu <solomon.chiu@amd.com>
Signed-off-by: Wayne Lin <Wayne.Lin@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
In single display configuration, windowed MPO does not work
with ODM combine.
[How]
For ODM + MPO window on one half of ODM, only 3 pipes should
be allocated and scaling parameters adjusted to handle this case.
Otherwise, we use 4 pipes.
Move copy_surface_update_to_plane() before dc_add_plane_to_context()
so that it gets the updated rect information when setting up
the pipes.
Add dc_check_boundary_crossing_for_windowed_mpo_with_odm() to force
a full update when we cross a boundary requiring us to reconfigure
the number of pipes between 3 and 4 pipes.
Set config.enable_windowed_mpo_odm to true when we have the
debug.enable_single_display_2to1_odm_policy set to true.
Don't fail validating ODM with windowed MPO if
config.enable_windowed_mpo_odm is true.
Reviewed-by: Aric Cyr <Aric.Cyr@amd.com>
Acked-by: Solomon Chiu <solomon.chiu@amd.com>
Signed-off-by: Samson Tam <Samson.Tam@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Description]
Exit SubVP if MPO is in use since SubVP + MPO together is not supported.
- Don't add SubVP at validation time if we see MPO is in use
Issues fixed in the SubVP / MPO transition:
1. Enable phantom pipes in post unlock function to prevent underflow
when an active pipe is being transitioned to be a phantom pipe (VTG
updates take place right away). Also must wait for VUPDATE of the main
pipe to complete first
2. Don't wait for MPCC idle when transitioning a phantom pipe to an
actual pipe. MPCC_STATUS is never asserted due to OTG being off for
phantom pipes
3. When transitioning an active pipe to phantom, program DET right away
(same as disabling the pipe) or the DET update will only take when
the phantom pipe is enabled which can cause DET allocation errors.
4. For K1/K2 programming of phantom pipes, use same settings as the
main pipe. Also don't program K1 / K2 = 0xF ever since the field is only
1 / 2 bits wide.
Reviewed-by: Jun Lei <Jun.Lei@amd.com>
Acked-by: Solomon Chiu <solomon.chiu@amd.com>
Signed-off-by: Alvin Lee <Alvin.Lee2@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
For GMC 10 parts which support scatter/gather display (display
from system memory), we should allocate a larger gart size
to better handler larger displays. This mirrors what we already
do for GMC 9 parts.
v2: fix typo (Alex)
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Need reserve buffers before unmap mes ctx bo va.
v2: fix removal of dma_resv_excl_fence() (Alex)
v3: fix dma_resv_usage (Alex)
Signed-off-by: Jack Xiao <Jack.Xiao@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
