5.19 fixes for msm-next
- Fix to add minimum ICC vote in the msm_mdss pm_resume path to address
bootup splats
- Fix to avoid dereferencing without checking in WB encoder
- Fix to avoid crash during suspend in DP driver by ensuring interrupt
mask bits are updated
- Remove unused code from dpu_encoder_virt_atomic_check()
- Fix to remove redundant init of dsc variable
Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Abhinav Kumar <quic_abhinavk@quicinc.com>
Link: https://patchwork.freedesktop.org/patch/msgid/927b201e-a734-a29d-b9fb-b9889e1f7795@quicinc.com
In commit a670ff578f ("drm/msm/dpu: always use mdp device to scale
bandwidth") we fully moved interconnect stuff to the DPU driver. This
had no change for sc7180 but _did_ have an impact for other SoCs. It
made them match the sc7180 scheme.
Unfortunately, the sc7180 scheme seems like it was a bit broken.
Specifically the interconnect needs to be on for more than just the
DPU driver's AXI bus. In the very least it also needs to be on for the
DSI driver's AXI bus. This can be seen fairly easily by doing this on
a ChromeOS sc7180-trogdor class device:
set_power_policy --ac_screen_dim_delay=5 --ac_screen_off_delay=10
sleep 10
cd /sys/bus/platform/devices/ae94000.dsi/power
echo on > control
When you do that, you'll get a warning splat in the logs about
"gcc_disp_hf_axi_clk status stuck at 'off'".
One could argue that perhaps what I have done above is "illegal" and
that it can't happen naturally in the system because in normal system
usage the DPU is pretty much always on when DSI is on. That being
said:
* In official ChromeOS builds (admittedly a 5.4 kernel with backports)
we have seen that splat at bootup.
* Even though we don't use "autosuspend" for these components, we
don't use the "put_sync" variants. Thus plausibly the DSI could stay
"runtime enabled" past when the DPU is enabled. Techncially we
shouldn't do that if the DPU's suspend ends up yanking our clock.
Let's change things such that the "bare minimum" request for the
interconnect happens in the mdss driver again. That means that all of
the children can assume that the interconnect is on at the minimum
bandwidth. We'll then let the DPU request the higher amount that it
wants.
It should be noted that this isn't as hacky of a solution as it might
initially appear. Specifically:
* Since MDSS and DPU individually get their own references to the
interconnect then the framework will actually handle aggregating
them. The two drivers are _not_ clobbering each other.
* When the Qualcomm interconnect driver aggregates it takes the max of
all the peaks. Thus having MDSS request a peak, as we're doing here,
won't actually change the total interconnect bandwidth (it won't be
added to the request for the DPU). This perhaps explains why the
"average" requested in MDSS was historically 0 since that one
_would_ be added in.
NOTE also that in the downstream ChromeOS 5.4 and 5.15 kernels, we're
also seeing some RPMH hangs that are addressed by this fix. These
hangs are showing up in the field and on _some_ devices with enough
stress testing of suspend/resume. Specifically right at suspend time
with a stack crawl that looks like this (from chromeos-5.15 tree):
rpmh_write_batch+0x19c/0x240
qcom_icc_bcm_voter_commit+0x210/0x420
qcom_icc_set+0x28/0x38
apply_constraints+0x70/0xa4
icc_set_bw+0x150/0x24c
dpu_runtime_resume+0x50/0x1c4
pm_generic_runtime_resume+0x30/0x44
__genpd_runtime_resume+0x68/0x7c
genpd_runtime_resume+0x12c/0x20c
__rpm_callback+0x98/0x138
rpm_callback+0x30/0x88
rpm_resume+0x370/0x4a0
__pm_runtime_resume+0x80/0xb0
dpu_kms_enable_commit+0x24/0x30
msm_atomic_commit_tail+0x12c/0x630
commit_tail+0xac/0x150
drm_atomic_helper_commit+0x114/0x11c
drm_atomic_commit+0x68/0x78
drm_atomic_helper_disable_all+0x158/0x1c8
drm_atomic_helper_suspend+0xc0/0x1c0
drm_mode_config_helper_suspend+0x2c/0x60
msm_pm_prepare+0x2c/0x40
pm_generic_prepare+0x30/0x44
genpd_prepare+0x80/0xd0
device_prepare+0x78/0x17c
dpm_prepare+0xb0/0x384
dpm_suspend_start+0x34/0xc0
We don't completely understand all the mechanisms in play, but the
hang seemed to come and go with random factors. It's not terribly
surprising that the hang is gone after this patch since the line of
code that was failing is no longer present in the kernel.
Fixes: a670ff578f ("drm/msm/dpu: always use mdp device to scale bandwidth")
Fixes: c33b7c0389 ("drm/msm/dpu: add support for clk and bw scaling for display")
Signed-off-by: Douglas Anderson <dianders@chromium.org>
Reviewed-by: Abhinav Kumar <quic_abhinavk@quicinc.com>
Tested-by: Jessica Zhang <quic_jesszhan@quicinc.com> # RB3 (sdm845) and
Reviewed-by: Stephen Boyd <swboyd@chromium.org>
Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@linaro.org>
Patchwork: https://patchwork.freedesktop.org/patch/487884/
Link: https://lore.kernel.org/r/20220531160059.v2.1.Ie7f6d4bf8cce28131da31a43354727e417cae98d@changeid
Signed-off-by: Abhinav Kumar <quic_abhinavk@quicinc.com>
dp_catalog_ctrl_reset() will software reset DP controller. But it will
not reset programmable registers to default value. DP driver still have
to clear mask bits to interrupt status registers to disable interrupts
after software reset of controller.
At current implementation, dp_ctrl_reset_irq_ctrl() will software reset dp
controller but did not call dp_catalog_ctrl_enable_irq(false) to clear hpd
related interrupt mask bits to disable hpd related interrupts due to it
mistakenly think hpd related interrupt mask bits will be cleared by software
reset of dp controller automatically. This mistake may cause system to crash
during suspending procedure due to unexpected irq fired and trigger event
thread to access dp controller registers with controller clocks are disabled.
This patch fixes system crash during suspending problem by removing "enable"
flag condition checking at dp_ctrl_reset_irq_ctrl() so that hpd related
interrupt mask bits are cleared to prevent unexpected from happening.
Changes in v2:
-- add more details commit text
Changes in v3:
-- add synchrons_irq()
-- add atomic_t suspended
Changes in v4:
-- correct Fixes's commit ID
-- remove synchrons_irq()
Changes in v5:
-- revise commit text
Changes in v6:
-- add event_lock to protect "suspended"
Changes in v7:
-- delete "suspended" flag
Fixes: 989ebe7bc4 ("drm/msm/dp: do not initialize phy until plugin interrupt received")
Signed-off-by: Kuogee Hsieh <quic_khsieh@quicinc.com>
Reviewed-by: Stephen Boyd <swboyd@chromium.org>
Patchwork: https://patchwork.freedesktop.org/patch/486591/
Link: https://lore.kernel.org/r/1652804494-19650-1-git-send-email-quic_khsieh@quicinc.com
Signed-off-by: Abhinav Kumar <quic_abhinavk@quicinc.com>
Convert fdinfo format to one documented in drm-usage-stats.rst.
It turned out that the existing implementation was actually completely
nonsense. The calculated percentages indeed represented the usage of the
engine, but with varying time slices.
So 10% usage for application A could mean something completely different
than 10% usage for application B.
Completely nuke that and just use the now standardized nanosecond
interface.
v2: drop the documentation change for now, nuke percentage calculation
v3: only account for each hw_ip, move the time_spend to the ctx mgr.
v4: move general ctx changes into separate patch, rework the fdinfo to
ctx_mgr interface so that all usages are calculated at once, drop
some unecessary and dangerous refcount dance.
v5: add one more comment how we calculate the time spend
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Shashank Sharma <shashank.sharma@amd.com>
Cc: Daniel Vetter <daniel@ffwll.ch>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Add a AMDGPU_GEM_CREATE_DISCARDABLE flag to note that the content of a BO
doesn't needs to be preserved during eviction.
KFD was already using a similar functionality for SVM BOs so replace the
internal flag with the new UAPI.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Spelling mistake (triple letters) in comment.
Detected with the help of Coccinelle.
Signed-off-by: Julia Lawall <Julia.Lawall@inria.fr>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Spelling mistakes (triple letters) in comments.
Detected with the help of Coccinelle.
Signed-off-by: Julia Lawall <Julia.Lawall@inria.fr>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Validate the VCN and SDMA instances against the driver
structure sizes to make sure we don't get into a
situation where the firmware reports more instances than
the driver supports.
Reviewed-by: Guchun Chen <guchun.chen@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
According to the KMS man page, there is a
"Coverage" alpha blend mode that assumes the
pixel color values have NOT been pre-multiplied
and will be done when the actual blending to
the background color values happens.
Previously, this mode hasn't been enabled
in our driver and it was assumed that all
normal overlay planes are pre-multiplied
by default.
When a 3rd party app is used to input a image
in a specific format, e.g. PNG, as a source
of a overlay plane to blend with the background
primary plane, the pixel color values are not
pre-multiplied. So by adding "Coverage" blend
mode, our driver will support those cases.
Issue fixed: Overlay plane alpha channel blending is incorrect
Issue tracker: https://gitlab.freedesktop.org/drm/amd/-/issues/1769
Reference:
https://dri.freedesktop.org/docs/drm/gpu/drm-kms.html#plane-composition-properties
Adding Coverage support also enables IGT
kms_plane_alpha_blend Coverage subtests:
1. coverage-7efc
2. coverage-vs-premult-vs-constant
Changes
1. Add DRM_MODE_BLEND_COVERAGE blend mode capability
2. Add "pre_multiplied_alpha" flag for Coverage case
3. Read the correct flag and set the DCN MPCC
pre_multiplied register bit (only on overlay plane)
Bug: https://gitlab.freedesktop.org/drm/amd/-/issues/1769
Signed-off-by: Sung Joon Kim <Sungjoon.Kim@amd.com>
Reviewed-by: Melissa Wen <mwen@igalia.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The > ARRAY_SIZE() should be >= ARRAY_SIZE() to prevent an out of bounds
access.
Fixes: e27c41d5b0 ("drm/amd/display: Support for DMUB HPD interrupt handling")
Reviewed-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Based on gfx10 with following changes:
- GPR_ALLOC.VGPR_SIZE field moved (and size corrected in gfx10)
- s_sendmsg_rtn_b64 replaces some s_sendmsg/s_getreg
- Buffer instructions no longer have direct-to-LDS modifier
Signed-off-by: Jay Cornwall <jay.cornwall@amd.com>
Reviewed-by: Laurent Morichetti <laurent.morichetti@amd.com>
Signed-off-by: Eric Huang <jinhuieric.huang@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
On aldebaran, when thermal throttling happens due to excessive GPU
temperature, the reason for throttling event is missed in warning
message. This patch fixes it.
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Yang Wang <kevinyang.wang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
To enable TMZ feature based on IP version needs adev->ip_version
populated but its empty. Move amdgpu_gmc_tmz_set to a place where
ip_version is populated.
Signed-off-by: Sunil Khatri <sunil.khatri@amd.com>
Reviewed-by: Alexander Deucher <Alexander.Deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
support umc/gfx/sdma ras on guest side
Changed from V1:
move sriov judgment in amdgpu_ras_interrupt_fatal_error_handler
Signed-off-by: Stanley.Yang <Stanley.Yang@amd.com>
Reviewed-by: Tao Zhou <tao.zhou1@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
There is no need to check if "clock_ranges' is non-NULL. It is checked
already on the line before.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
gpu_metrics_table is allocated in yellow_carp_init_smc_tables() but
not freed in yellow_carp_fini_smc_tables().
Signed-off-by: Gong Yuanjun <ruc_gongyuanjun@163.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
In radeon_fp_native_mode(), the return value of drm_mode_duplicate()
is assigned to mode, which will lead to a NULL pointer dereference
on failure of drm_mode_duplicate(). Add a check to avoid npd.
The failure status of drm_cvt_mode() on the other path is checked too.
Signed-off-by: Gong Yuanjun <ruc_gongyuanjun@163.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Make sure the queue is not longer active before programming
the kiq EOP registers.
Signed-off-by: Haohui Mai <ricetons@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Remove the accidental shifts on the values of RPTR_BLOCK_SIZE
in gfx_v8-v11. The bug essentially always programs the
corresponding fields to zero instead of the correct value.
The hardware clamps the min value to 5 so this resulted in a
value of 5 being programmed.
Signed-off-by: Haohui Mai <ricetons@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Let each context have a pointer to the ctx manager and properly
initialize the adev pointer inside the context manager.
Reduce the BUG_ON() in amdgpu_ctx_add_fence() into a WARN_ON() and
directly return the sequence number instead of writing into a parmeter.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Shashank Sharma <shashank.sharma@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
CPU hive assignment currently assumes when a GPU hive is connected_to_cpu,
there is only one hive in the system.
Only assign CPUs to the hive if they are explicitly directly connected to
the GPU hive to get rid of the need for this assumption.
It's more efficient to do this when querying IO links since other non-CRAT
info has to be filled in anyways. Also, stop re-assigning the
same CPU to the same GPU hive if it has already been done before.
Signed-off-by: Jonathan Kim <jonathan.kim@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Clean up redundant, copy-paste code blocks during the initialization of
the doorbells in mqd_init().
Signed-off-by: Haohui Mai <ricetons@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This version brings along the following:
- Improvements in link training fallback
- Adding individual edp hotplug support
- Fixes in DPIA HPD status, display clock change hang, etc.
- FPU isolation work for DCN30
Acked-by: Qingqing Zhuo <qingqing.zhuo@amd.com>
Signed-off-by: Aric Cyr <aric.cyr@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
