A8x is the next generation of Adreno GPUs, featuring a significant
hardware design change. A major update to the design is the introduction
of Slice architecture. Slices are sort of mini-GPUs within the GPU which
are more independent in processing Graphics and compute workloads. Also,
in addition to the BV and BR pipe we saw in A7x, CP has more concurrency
with additional pipes.
From a software interface perspective, these changes have a significant
impact on the KMD side. First, the GPU register space has been extensively
reorganized. Second, to avoid a register space explosion caused by the
new slice architecture and additional pipes, many registers are now
virtualized, instead of duplicated as in A7x. KMD must configure an
aperture register with the appropriate slice and pipe ID before accessing
these virtualized registers.
Signed-off-by: Akhil P Oommen <akhilpo@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/689019/
Message-ID: <20251118-kaana-gpu-support-v4-14-86eeb8e93fb6@oss.qualcomm.com>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
If there is a flood of faults then the MMU can become saturated while it
waits for the kernel to process the first fault and resume it, so that
the GMU becomes blocked. This is mainly a problem when the kernel reads
the state of the GPU for a devcoredump, because this takes a while. If
we timeout waiting for the GMU, check if this has happened and retry
after we're finished.
Signed-off-by: Connor Abbott <cwabbott0@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/664685/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Add support for Adreno 612 GPU found in SM6150/QCS615 chipsets.
A612 falls under ADRENO_6XX_GEN1 family and is a cut down version
of A615 GPU.
A612 has a new IP called Reduced Graphics Management Unit or RGMU
which is a small state machine which helps to toggle GX GDSC
(connected to CX rail) to implement IFPC feature. It doesn't support
any other features of a full fledged GMU like clock control, resource
voting to rpmh etc. So we need linux clock driver support like other
gmu-wrapper implementations to control gpu core clock and gpu GX gdsc.
This patch skips RGMU core initialization and act more like a
gmu-wrapper case.
Signed-off-by: Jie Zhang <quic_jiezh@quicinc.com>
Signed-off-by: Akhil P Oommen <akhilpo@oss.qualcomm.com>
Reviewed-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/686212/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
When things go wrong, the GPU is capable of quickly generating millions
of faulting translation requests per second. When that happens, in the
stall-on-fault model each access will stall until it wins the race to
signal the fault and then the RESUME register is written. This slows
processing page faults to a crawl as the GPU can generate faults much
faster than the CPU can acknowledge them. It also means that all
available resources in the SMMU are saturated waiting for the stalled
transactions, so that other transactions such as transactions generated
by the GMU, which shares translation resources with the GPU, cannot
proceed. This causes a GMU watchdog timeout, which leads to a failed
reset because GX cannot collapse when there is a transaction pending and
a permanently hung GPU.
On older platforms with qcom,smmu-v2, it seems that when one transaction
is stalled subsequent faulting transactions are terminated, which avoids
this problem, but the MMU-500 follows the spec here.
To work around these problems, disable stall-on-fault as soon as we get a
page fault until a cooldown period after pagefaults stop. This allows
the GMU some guaranteed time to continue working. We only use
stall-on-fault to halt the GPU while we collect a devcoredump and we
always terminate the transaction afterward, so it's fine to miss some
subsequent page faults. We also keep it disabled so long as the current
devcoredump hasn't been deleted, because in that case we likely won't
capture another one if there's a fault.
After this commit HFI messages still occasionally time out, because the
crashdump handler doesn't run fast enough to let the GMU resume, but the
driver seems to recover from it. This will probably go away after the
HFI timeout is increased.
Signed-off-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Rob Clark <robdclark@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/654891/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Really the only purpose of this was to limit the address space size to
4GB to avoid 32b rollover problems in 64b pointer math in older sqe fw.
So replace the address_space_size with a quirk limiting the address
space to 4GB. In all other cases, use the SMMU input address size (IAS)
to determine the address space size.
v2: Properly account for vm_start
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/649467/
The Adreno GPU Management Unit (GMU) can also scale the DDR Bandwidth
along the Frequency and Power Domain level, until now we left the OPP
core scale the OPP bandwidth via the interconnect path.
In order to enable bandwidth voting via the GPU Management
Unit (GMU), when an opp is set by devfreq we also look for
the corresponding bandwidth index in the previously generated
bw_table and pass this value along the frequency index to the GMU.
The GMU also takes another vote called AB which is a 16bit quantized
value of the floor bandwidth against the maximum supported bandwidth.
The AB is calculated with a default 25% of the bandwidth like the
downstream implementation too inform the GMU firmware the minimal
quantity of bandwidth we require for this OPP. Only pass the AB
vote starting from A750 GPUs.
Since we now vote for all resources via the GMU, setting the OPP
is no more needed, so we can completely skip calling
dev_pm_opp_set_opp() in this situation.
Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@linaro.org>
Reviewed-by: Akhil P Oommen <quic_akhilpo@quicinc.com>
Signed-off-by: Neil Armstrong <neil.armstrong@linaro.org>
Reviewed-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/629397/
Signed-off-by: Rob Clark <robdclark@chromium.org>
According to downstream we should be setting RBBM_NC_MODE_CNTL to a
non-default value on a663 and a680, we don't support a663 and on a680
we're leaving it at the wrong (suboptimal) value. Just set it on all
GPUs. Similarly, plumb through level2_swizzling_dis which will be
necessary on a663.
ubwc_mode is expanded and renamed to ubwc_swizzle to match the name on
the display side. Similarly macrotile_mode should match the display
side.
Signed-off-by: Connor Abbott <cwabbott0@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/607397/
Signed-off-by: Rob Clark <robdclark@chromium.org>
On all Qualcomm platforms with a7xx GPUs, qcom_scm provides a method to
initialize cx_mem. Copy this from downstream (minus BCL which we
currently don't support). On a750, this includes a new "fuse" register
which can be used by qcom_scm to fuse off certain features like
raytracing in software. The fuse is default off, and is initialized by
calling the method. Afterwards we have to read it to find out which
features were enabled.
Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@linaro.org>
Signed-off-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Konrad Dybcio <konrad.dybcio@linaro.org>
Patchwork: https://patchwork.freedesktop.org/patch/592042/
Signed-off-by: Rob Clark <robdclark@chromium.org>
The A702 is a weird mix of 600 and 700 series.. Perhaps even a
testing ground for some A7xx features with good ol' A6xx silicon.
It's basically A610 that's been beefed up with some new registers
and hw features (like APRIV!), that was then cut back in size,
memory bus and some other ways.
Add support for it, tested with QCM2290 / RB1.
Signed-off-by: Konrad Dybcio <konrad.dybcio@linaro.org>
Patchwork: https://patchwork.freedesktop.org/patch/579752/
Signed-off-by: Rob Clark <robdclark@chromium.org>
Add support for the A750 GPU found on the SM8650 platform
Unlike the the very close A740 GPU on the SM8550 SoC, the A750 GPU
doesn't have an HWCFG block but a separate register set.
The A750 GPU info are added under the adreno_is_a750() macro and
the ADRENO_7XX_GEN3 family id.
Reviewed-by: Konrad Dybcio <konrad.dybcio@linaro.org>
Signed-off-by: Neil Armstrong <neil.armstrong@linaro.org>
Patchwork: https://patchwork.freedesktop.org/patch/578693/
Signed-off-by: Rob Clark <robdclark@chromium.org>
Add support for the Adreno 305B GPU that is found in MSM8226(v2) SoC.
Previously this was mistakenly claimed to be supported but using wrong
a configuration.
In MSM8226v1 there's also a A305B but with chipid 0x03000510 which
should work with the same configuration but due to lack of hardware for
testing this is not added.
Reviewed-by: Konrad Dybcio <konrad.dybcio@linaro.org>
Signed-off-by: Luca Weiss <luca@z3ntu.xyz>
Reviewed-by: David Heidelberg <david@ixit.cz>
Patchwork: https://patchwork.freedesktop.org/patch/575274/
Signed-off-by: Rob Clark <robdclark@chromium.org>
A7xx GPUs are - from kernel's POV anyway - basically another generation
of A6xx. They build upon the A650/A660_family advancements, skipping some
writes (presumably more values are preset correctly on reset), adding
some new ones and changing others.
One notable difference is the introduction of a second shadow, called BV.
To handle this with the current code, allocate it right after the current
RPTR shadow.
BV handling and .submit are mostly based on Jonathan Marek's work.
All A7xx GPUs are assumed to have a GMU.
A702 is not an A7xx-class GPU, it's a weird forked A610.
Tested-by: Neil Armstrong <neil.armstrong@linaro.org> # on SM8550-QRD
Tested-by: Dmitry Baryshkov <dmitry.baryshkov@linaro.org> # sm8450
Signed-off-by: Konrad Dybcio <konrad.dybcio@linaro.org>
Patchwork: https://patchwork.freedesktop.org/patch/559285/
Signed-off-by: Rob Clark <robdclark@chromium.org>
Since the revision becomes an opaque identifier with future GPUs, move
away from treating different ranges of bits as having a given meaning.
This means that we need to explicitly list different patch revisions in
the device table.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@linaro.org>
Patchwork: https://patchwork.freedesktop.org/patch/549782/
