Use the dGPU path instead. There were a lot of platform
issues with IOMMU in general on these chips due to windows
not enabling IOMMU at the time. The dGPU path has been
used for a long time with newer APUs and works fine. This
also paves the way to simplify the driver significantly.
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Tested-by: Mike Lothian <mike@fireburn.co.uk>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Use the dGPU path instead. There were a lot of platform
issues with IOMMU in general on these chips due to windows
not enabling IOMMU at the time. The dGPU path has been
used for a long time with newer APUs and works fine. This
also paves the way to simplify the driver significantly.
v2: use the dGPU queue manager functions
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Tested-by: Mike Lothian <mike@fireburn.co.uk>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This is only required for SR-IOV world switches, but it
adds additional latency leading to reduced performance in
some benchmarks. Disable for now on bare metal.
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The existing OD interface cannot support the growing demand for more
OD features. We are in the transition to a new OD mechanism. So,
disable the SMU13 OD feature support temporarily. And this should be
reverted when the new OD mechanism online.
Signed-off-by: Evan Quan <evan.quan@amd.com>
Reviewed-by: Guchun Chen <guchun.chen@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
On PSP v13.x ASICs, boot loader will set only the MSB to 1 and clear the
least significant bits for any command submission. Hence match against
the exact register value, otherwise a register value of all 0xFFs also
could falsely indicate that boot loader is ready. Also, from PSP v13.0.6
and newer, bits[7:0] will be used to indicate command error status.
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Currently we use the drm_dp_dpcd_read_caps() helper in the DRM side of
nouveau in order to read the DPCD of a DP connector, which makes sure we do
the right thing and also check for extended DPCD caps. However, it turns
out we're not currently doing this on the nvkm side since we don't have
access to the drm_dp_aux structure there - which means that the DRM side of
the driver and the NVKM side can end up with different DPCD capabilities
for the same connector.
Ideally in order to fix this, we just want to use the
drm_dp_read_dpcd_caps() helper in nouveau. That's not currently possible
though, and is going to depend on having a bunch of the DP code moved out
of nvkm and into the DRM side of things as part of the GSP enablement work.
Until then however, let's workaround this problem by porting a copy of
drm_dp_read_dpcd_caps() into NVKM - which should fix this issue.
Signed-off-by: Lyude Paul <lyude@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Link: https://gitlab.freedesktop.org/drm/nouveau/-/issues/211
Link: https://patchwork.freedesktop.org/patch/msgid/20230728225858.350581-1-lyude@redhat.com
(cherry picked from commit cc4adf3a73 in drm-misc-next)
Cc: <stable@vger.kernel.org> # 6.3+
Signed-off-by: Karol Herbst <kherbst@redhat.com>
We have a lurking bug where Fragment Shader Helper Invocations can't load
from memory. But this is actually required in OpenGL and is causing random
hangs or failures in random shaders.
It is unknown how widespread this issue is, but shaders hitting this can
end up with infinite loops.
We enable those only on all Kepler and newer GPUs where we use our own
Firmware.
Nvidia's firmware provides a way to set a kernelspace controlled list of
mmio registers in the gr space from push buffers via MME macros.
v2: drop code for gm200 and newer.
Cc: Ben Skeggs <bskeggs@redhat.com>
Cc: David Airlie <airlied@gmail.com>
Cc: nouveau@lists.freedesktop.org
Cc: stable@vger.kernel.org # 4.19+
Signed-off-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230622152017.2512101-1-kherbst@redhat.com
On system resume, the driver might call it6505_poweron directly if the
runtime PM hasn't been enabled. In such case, pm_runtime_get_if_in_use
will always return 0 because dev->power.runtime_status stays at
RPM_SUSPENDED, and the IRQ will never be handled.
Use it6505->powered from the driver struct fixes this because it always
gets updated when it6505_poweron is called.
Fixes: 5eb9a43140 ("drm/bridge: it6505: Guard bridge power in IRQ handler")
Signed-off-by: Pin-yen Lin <treapking@chromium.org>
Reviewed-by: Neil Armstrong <neil.armstrong@linaro.org>
Signed-off-by: Neil Armstrong <neil.armstrong@linaro.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20230727100131.2338127-1-treapking@chromium.org
Infinite waits for completion of GPU activity have been observed in CI,
mostly inside __i915_active_wait(), triggered by igt@gem_barrier_race or
igt@perf@stress-open-close. Root cause analysis, based of ftrace dumps
generated with a lot of extra trace_printk() calls added to the code,
revealed loops of request dependencies being accidentally built,
preventing the requests from being processed, each waiting for completion
of another one's activity.
After we substitute a new request for a last active one tracked on a
timeline, we set up a dependency of our new request to wait on completion
of current activity of that previous one. While doing that, we must take
care of keeping the old request still in memory until we use its
attributes for setting up that await dependency, or we can happen to set
up the await dependency on an unrelated request that already reuses the
memory previously allocated to the old one, already released. Combined
with perf adding consecutive kernel context remote requests to different
user context timelines, unresolvable loops of await dependencies can be
built, leading do infinite waits.
We obtain a pointer to the previous request to wait upon when we
substitute it with a pointer to our new request in an active tracker,
e.g. in intel_timeline.last_request. In some processing paths we protect
that old request from being freed before we use it by getting a reference
to it under RCU protection, but in others, e.g. __i915_request_commit()
-> __i915_request_add_to_timeline() -> __i915_request_ensure_ordering(),
we don't. But anyway, since the requests' memory is SLAB_FAILSAFE_BY_RCU,
that RCU protection is not sufficient against reuse of memory.
We could protect i915_request's memory from being prematurely reused by
calling its release function via call_rcu() and using rcu_read_lock()
consequently, as proposed in v1. However, that approach leads to
significant (up to 10 times) increase of SLAB utilization by i915_request
SLAB cache. Another potential approach is to take a reference to the
previous active fence.
When updating an active fence tracker, we first lock the new fence,
substitute a pointer of the current active fence with the new one, then we
lock the substituted fence. With this approach, there is a time window
after the substitution and before the lock when the request can be
concurrently released by an interrupt handler and its memory reused, then
we may happen to lock and return a new, unrelated request.
Always get a reference to the current active fence first, before
replacing it with a new one. Having it protected from premature release
and reuse, lock it and then replace with the new one but only if not
yet signalled via a potential concurrent interrupt nor replaced with
another one by a potential concurrent thread, otherwise retry, starting
from getting a reference to the new current one. Adjust users to not
get a reference to the previous active fence themselves and always put the
reference got by __i915_active_fence_set() when no longer needed.
v3: Fix lockdep splat reports and other issues caused by incorrect use of
try_cmpxchg() (use (cmpxchg() != prev) instead)
v2: Protect request's memory by getting a reference to it in favor of
delegating its release to call_rcu() (Chris)
Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/8211
Fixes: df9f85d858 ("drm/i915: Serialise i915_active_fence_set() with itself")
Suggested-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Janusz Krzysztofik <janusz.krzysztofik@linux.intel.com>
Cc: <stable@vger.kernel.org> # v5.6+
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Signed-off-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230720093543.832147-2-janusz.krzysztofik@linux.intel.com
(cherry picked from commit 946e047a3d)
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Fixes for v6.5-rc4
Display:
+ Fix to correct the UBWC programming for decoder version 4.3 seen
on SM8550
+ Add the missing flush and fetch bits for DMA4 and DMA5 SSPPs.
+ Fix to drop the unused dpu_core_perf_data_bus_id enum from the code
+ Drop the unused dsi_phy_14nm_17mA_regulators from QCM 2290 DSI cfg.
GPU:
+ Fix warn splat for newer devices without revn
+ Remove name/revn for a690.. we shouldn't be populating these for
newer devices, for consistency, but it slipped through review
+ Fix a6xx gpu snapshot BINDLESS_DATA size (was listed in bytes
instead of dwords, causing AHB faults on a6xx gen4/a660-family)
+ Disallow submit with fence id 0
Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Rob Clark <robdclark@gmail.com>
Link: https://patchwork.freedesktop.org/patch/msgid/CAF6AEGs9MwCSfiyv8i7yWAsJKYEzCDyzaTx=ujX80Y23rZd9RA@mail.gmail.com
A fence id of zero is expected to be invalid, and is not removed from
the fence_idr table. If userspace is requesting to specify the fence
id with the FENCE_SN_IN flag, we need to reject a zero fence id value.
Fixes: 17154addc5 ("drm/msm: Add MSM_SUBMIT_FENCE_SN_IN")
Signed-off-by: Rob Clark <robdclark@chromium.org>
Patchwork: https://patchwork.freedesktop.org/patch/549180/
On GFX v9.4.3, compute queue MQD is populated using the values in HQD
persistent state register. Hence don't clear the values on module
unload, instead restore it to the default reset value so that MQD is
initialized correctly during next module load. In particular, preload
flag needs to be set on compute queue MQD, otherwise it could cause
uninitialized values being used at device reset state resulting in EDC.
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: Asad Kamal <asad.kamal@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This error path needs to unlock the "aconnector->handle_mst_msg_ready"
mutex before returning.
Fixes: 4f6d9e38c4 ("drm/amd/display: Add polling method to handle MST reply packet")
Signed-off-by: Dan Carpenter <dan.carpenter@linaro.org>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Description]
It is not valid to set the WDIVIDER value to 0, so do not
re-write to DISPCLK_WDIVIDER if the current value is 0
(i.e., it is at it's initial value and we have not made any
requests to change DISPCLK yet).
Reviewed-by: Saaem Rizvi <syedsaaem.rizvi@amd.com>
Acked-by: Alex Hung <alex.hung@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>
Report current GFX clock also from average clock value as the original
CurrClock data is not valid/accurate any more as per FW team
Signed-off-by: Jane Jian <Jane.Jian@amd.com>
Reviewed-by: Lijo Lazar <lijo.lazar@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
If the second call to amdgpu_bo_create_kernel() fails, the memory
allocated from the first call should be cleared. If the third call
fails, the memory from the second call should be cleared.
Fixes: b95b539168 ("drm/amdgpu/psp: move PSP memory alloc from hw_init to sw_init")
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
Reviewed-by: Lijo Lazar <lijo.lazar@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
An instance of for_each_inst() was not changed to match its new
behaviour and is causing a loop.
v2: remove tmp_mask variable
Fixes: b579ea632f ("drm/amdgpu: Modify for_each_inst macro")
Signed-off-by: Victor Lu <victorchengchi.lu@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
