soc: qcom: use no-UBWC config for MSM8956/76

Both MSM8956 and MSM8976 have MDSS 1.11 which doesn't support UBWC
(although they also have Adreno 510, which might support UBWC). Disable
UBWC support for those platforms.

Fixes: 1924272b9ce1 ("soc: qcom: Add UBWC config provider")
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Reviewed-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/668503/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

soc: qcom: add configuration for MSM8929

MSM8929 is similar to MSM8939, it doesn't support UBWC. Provide no-UBWC
config for the platform.

Fixes: 197713d0cf01 ("soc: qcom: ubwc: provide no-UBWC configuration")
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Reviewed-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/668502/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

soc: qcom: ubwc: add more missing platforms

Add UBWC configuration for SDA660 (modem-less variant of SDM660), SDM450
(similar to MSM8953), SDM632 (similar to MSM8953) and SM7325 (similar to
SC7280).

Fixes: 1924272b9ce1 ("soc: qcom: Add UBWC config provider")
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Reviewed-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/668501/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

soc: qcom: ubwc: use no-uwbc config for MSM8917

MSM8917 has MDSS 1.15 and Adreno 308, neither of which support UBWC.
Change UBWC configuration to point out that UBWC is not supported on
this platform.

Fixes: 1924272b9ce1 ("soc: qcom: Add UBWC config provider")
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Reviewed-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/668500/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

drm/msm/dpu: Add a null ptr check for dpu_encoder_needs_modeset

The drm_atomic_get_new_connector_state() can return NULL if the
connector is not part of the atomic state. Add a check to prevent
a NULL pointer dereference.

This follows the same pattern used in dpu_encoder_update_topology()
within the same file, which checks for NULL before using conn_state.

Signed-off-by: Chenyuan Yang <chenyuan0y@gmail.com>
Fixes: 1ce69c265a53 ("drm/msm/dpu: move resource allocation to CRTC")
Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/665188/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

dt-bindings: display/msm: qcom,mdp5: drop lut clock

None of MDP5 platforms have a LUT clock on the display-controller, it
was added by the mistake. Drop it, fixing DT warnings on MSM8976 /
MSM8956 platforms. Technically it's an ABI break, but no other platforms
are affected.

Fixes: 385c8ac763b3 ("dt-bindings: display/msm: convert MDP5 schema to YAML format")
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Acked-by: Rob Herring (Arm) <robh@kernel.org>
Patchwork: https://patchwork.freedesktop.org/patch/667822/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

drm/msm: skip re-emitting IBs for unusable VMs

When a VM is marked as an usuable we disallow new submissions from it,
however submissions that where already scheduled on the ring would still
be re-sent.

Since this can lead to further hangs, avoid emitting the actual IBs.

Fixes: 6a4d287a1ae6 ("drm/msm: Mark VM as unusable on GPU hangs")
Signed-off-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Akhil P Oommen <akhilpo@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/668314/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

drm/msm: Don't use %pK through printk

In the past %pK was preferable to %p as it would not leak raw pointer
values into the kernel log.
Since commit ad67b74d2469 ("printk: hash addresses printed with %p")
the regular %p has been improved to avoid this issue.
Furthermore, restricted pointers ("%pK") were never meant to be used
through printk(). They can still unintentionally leak raw pointers or
acquire sleeping locks in atomic contexts.

Switch to the regular pointer formatting which is safer and
easier to reason about.

Signed-off-by: Thomas Weißschuh <thomas.weissschuh@linutronix.de>
Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/667895/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

soc: qcom: ubwc: Add missing UBWC config for SM7225

SM7225 is a variation of SM6350, and also needs an entry in the table.

Fixes: 1924272b9ce1 ("soc: qcom: Add UBWC config provider")
Signed-off-by: Luca Weiss <luca.weiss@fairphone.com>
Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Acked-by: Bjorn Andersson <andersson@kernel.org>
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>

soc: qcom: ubwc: provide no-UBWC configuration

After the commit 45a2974157d2 ("drm/msm: Use the central UBWC config
database") the MDSS driver errors out if UBWC database didn't provide it
with the UBWC configuration. Make UBWC database return zero data for
MSM8916 / APQ8016, MSM8974 / APQ8074, MSM8226 and MSM8939.

Fixes: 1924272b9ce1 ("soc: qcom: Add UBWC config provider")
Acked-by: Bjorn Andersson <andersson@kernel.org>
Reviewed-by: Rob Clark <robin.clark@oss.qualcomm.com>
Reviewed-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/665313/
Link: https://lore.kernel.org/r/20250723-ubwc-no-ubwc-v3-1-81bdb75685bf@oss.qualcomm.com

drm/msm/dpu: Initialize crtc_state to NULL in dpu_plane_virtual_atomic_check()

After a recent change in clang to expose uninitialized warnings from
const variables and pointers [1], there is a warning around crtc_state
in dpu_plane_virtual_atomic_check():

  drivers/gpu/drm/msm/disp/dpu1/dpu_plane.c:1145:6: error: variable 'crtc_state' is used uninitialized whenever 'if' condition is false [-Werror,-Wsometimes-uninitialized]
   1145 |         if (plane_state->crtc)
        |             ^~~~~~~~~~~~~~~~~
  drivers/gpu/drm/msm/disp/dpu1/dpu_plane.c:1149:58: note: uninitialized use occurs here
   1149 |         ret = dpu_plane_atomic_check_nosspp(plane, plane_state, crtc_state);
        |                                                                 ^~~~~~~~~~
  drivers/gpu/drm/msm/disp/dpu1/dpu_plane.c:1145:2: note: remove the 'if' if its condition is always true
   1145 |         if (plane_state->crtc)
        |         ^~~~~~~~~~~~~~~~~~~~~~
   1146 |                 crtc_state = drm_atomic_get_new_crtc_state(state,
  drivers/gpu/drm/msm/disp/dpu1/dpu_plane.c:1139:35: note: initialize the variable 'crtc_state' to silence this warning
   1139 |         struct drm_crtc_state *crtc_state;
        |                                          ^
        |                                           = NULL

Initialize crtc_state to NULL like other places in the driver do, so
that it is consistently initialized.

Cc: stable@vger.kernel.org
Closes: https://github.com/ClangBuiltLinux/linux/issues/2106
Fixes: 774bcfb73176 ("drm/msm/dpu: add support for virtual planes")
Link: https://github.com/llvm/llvm-project/commit/2464313eef01c5b1edf0eccf57a32cdee01472c7
Signed-off-by: Nathan Chancellor <nathan@kernel.org>
Reviewed-by: Jessica Zhang <jessica.zhang@oss.qualcomm.com>
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>

drm/msm: update the high bitfield of certain DSI registers

Currently, the high bitfield of certain DSI registers
do not align with the configuration of the SWI registers
description. This can lead to wrong programming these DSI
registers, for example for 4k resloution where H_TOTAL is
taking 13 bits but software is programming only 12 bits
because of the incorrect bitmask for H_TOTAL bitfeild,
this is causing DSI FIFO errors. To resolve this issue,
increase the high bitfield of the DSI registers from 12 bits
to 16 bits in dsi.xml to match the SWI register configuration.

Signed-off-by: Ayushi Makhija <quic_amakhija@quicinc.com>
Fixes: 4f52f5e63b62 ("drm/msm: import XML display registers database")
Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/666229/
Link: https://lore.kernel.org/r/20250730123938.1038640-1-quic_amakhija@quicinc.com
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>

drm/msm/dpu: correct dpu_plane_virtual_atomic_check()

Fix c&p error in dpu_plane_virtual_atomic_check(), compare CRTC width
too, in addition to CRTC height.

Fixes: 8c62a31607f6 ("drm/msm/dpu: allow using two SSPP blocks for a single plane")
Reported-by: kernel test robot <lkp@intel.com>
Closes: https://lore.kernel.org/oe-kbuild-all/202507150432.U0cALR6W-lkp@intel.com/
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Reviewed-by: Jessica Zhang <jessica.zhang@oss.qualcomm.com>
Reviewed-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/664170/
Link: https://lore.kernel.org/r/20250715-msm-fix-virt-atomic-check-v1-1-9bab02c9f952@oss.qualcomm.com

drm/msm/kms: move snapshot init earlier in KMS init

Various parts of the display driver can be triggering the display
snapshot (including the IOMMU fault handlers). Move the call to
msm_disp_snapshot_init() before KMS initialization, otherwise it is
possible to ocassionally trigger the kernel fault during init:

  __lock_acquire+0x44/0x2798 (P)
  lock_acquire+0x114/0x25c
  _raw_spin_lock_irqsave+0x6c/0x90
  kthread_queue_work+0x2c/0xac
  msm_disp_snapshot_state+0x2c/0x4c
  msm_kms_fault_handler+0x2c/0x74
  msm_disp_fault_handler+0x30/0x48
  report_iommu_fault+0x54/0x128
  arm_smmu_context_fault+0x74/0x184
  __handle_irq_event_percpu+0xa4/0x24c
  handle_irq_event_percpu+0x20/0x5c
  handle_irq_event+0x48/0x84
  handle_fasteoi_irq+0xcc/0x170
  generic_handle_domain_irq+0x48/0x70
  gic_handle_irq+0x54/0x11c
  call_on_irq_stack+0x3c/0x50
  do_interrupt_handler+0x54/0x78
  el1_interrupt+0x3c/0x5c
  el1h_64_irq_handler+0x20/0x30
  el1h_64_irq+0x6c/0x70
  _raw_spin_unlock_irqrestore+0x44/0x68 (P)
  klist_next+0xc4/0x124
  bus_for_each_drv+0x9c/0xe8
  __device_attach+0xfc/0x190
  device_initial_probe+0x1c/0x2c
  bus_probe_device+0x44/0xa0
  device_add+0x204/0x3e4
  platform_device_add+0x170/0x244
  platform_device_register_full+0x130/0x138
  drm_connector_hdmi_audio_init+0xc0/0x108
  drm_bridge_connector_init+0x318/0x394
  msm_dsi_manager_connector_init+0xac/0xdc
  msm_dsi_modeset_init+0x78/0xc0
  _dpu_kms_drm_obj_init+0x198/0x75c
  dpu_kms_hw_init+0x2f8/0x494
  msm_drm_kms_init+0xb0/0x230
  msm_drm_init+0x218/0x250
  msm_drm_bind+0x3c/0x4c
  try_to_bring_up_aggregate_device+0x208/0x2a4
  __component_add+0xa8/0x188
  component_add+0x1c/0x2c
  dsi_dev_attach+0x24/0x34
  dsi_host_attach+0x68/0xa0
  devm_mipi_dsi_attach+0x40/0xcc
  lt9611_attach_dsi+0x94/0x118
  lt9611_probe+0x368/0x3c8
  i2c_device_probe+0x2d0/0x3d8
  really_probe+0x130/0x354
  __driver_probe_device+0xac/0x110
  driver_probe_device+0x44/0x110
  __device_attach_driver+0xb0/0x138
  bus_for_each_drv+0x90/0xe8
  __device_attach+0xfc/0x190
  device_initial_probe+0x1c/0x2c
  bus_probe_device+0x44/0xa0
  deferred_probe_work_func+0xac/0x110
  process_one_work+0x20c/0x51c
  process_scheduled_works+0x58/0x88
  worker_thread+0x1ec/0x304
  kthread+0x194/0x1d4
  ret_from_fork+0x10/0x20

Reported-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Fixes: 98659487b845 ("drm/msm: add support to take dpu snapshot")
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Reviewed-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/664149/
Link: https://lore.kernel.org/r/20250715-msm-move-snapshot-init-v1-1-f39c396192ab@oss.qualcomm.com

drm/msm/dsi: Fix 14nm DSI PHY PLL Lock issue

To configure and enable the DSI PHY PLL clocks, the MDSS AHB clock must
be active for MMIO operations. Typically, this AHB clock is enabled as
part of the DSI PHY interface enabling (dsi_phy_enable_resource).

However, since these PLL clocks are registered as clock entities, they
can be enabled independently of the DSI PHY interface, leading to
enabling failures and subsequent warnings:

```
msm_dsi_phy 5e94400.phy: [drm:dsi_pll_14nm_vco_prepare] *ERROR* DSI PLL lock failed
------------[ cut here ]------------
dsi0pllbyte already disabled
WARNING: CPU: 3 PID: 1 at drivers/clk/clk.c:1194 clk_core_disable+0xa4/0xac
CPU: 3 UID: 0 PID: 1 Comm: swapper/0 Tainted:
Tainted: [W]=WARN
Hardware name: Qualcomm Technologies, Inc. Robotics RB1 (DT)
pstate: 600000c5 (nZCv daIF -PAN -UAO -TCO -DIT -SSBS BTYPE=--)
[...]
```

This issue is particularly prevalent at boot time during the disabling of
unused clocks (clk_disable_unused()) which includes enabling the parent
clock(s) when CLK_OPS_PARENT_ENABLE flag is set (this is the case for the
14nm DSI PHY PLL consumers).

To resolve this issue, we move the AHB clock as a PM dependency of the DSI
PHY device (via pm_clk). Since the DSI PHY device is the parent of the PLL
clocks, this resolves the PLL/AHB dependency. Now the AHB clock is enabled
prior the PLL clk_prepare callback, as part of the runtime-resume chain.

We also eliminate dsi_phy_[enable|disable]_resource functions, which are
superseded by runtime PM.

Note that it breaks compatibility with kernels before 6.0, as we do not
support anymore the legacy `iface_clk` name.

Signed-off-by: Loic Poulain <loic.poulain@oss.qualcomm.com>
Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/663239/
Link: https://lore.kernel.org/r/20250709140836.124143-1-loic.poulain@oss.qualcomm.com
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>

drm/msm: Fix objtool warning in submit_lock_objects()

Split the vmbind case into a separate helper function
submit_lock_objects_vmbind() to fix objtool warning:

  drivers/gpu/drm/msm/msm.o: warning: objtool: submit_lock_objects+0x451:
  sibling call from callable instruction with modified stack frame

The drm_exec_until_all_locked() macro uses computed gotos internally
for its retry loop. Having return statements inside this macro, or
immediately after it in certain code paths, confuses objtool's static
analysis of stack frames, causing it to incorrectly flag tail call
optimizations.

Fixes: 92395af63a99 ("drm/msm: Add VM_BIND submitqueue")
Signed-off-by: Sasha Levin <sashal@kernel.org>
Patchwork: https://patchwork.freedesktop.org/patch/667539/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

drm/msm: Handle in-place remaps

Detect and handle the special case of a MAP op simply updating the vma
flags of an existing vma, and skip the pgtable updates.  This allows
turnip to set the MSM_VMA_DUMP flag on an existing mapping without
requiring additional synchronization against commands running on the
GPU.

Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Connor Abbott <cwabbott0@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/667238/

drm/msm: Fix a few comments

Fix a couple comments which had become (partially) obsolete or incorrect
with the gpuvm conversion.

Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/667237/

drm/msm: Fix a7xx TPL1 cluster snapshot

Later gens have both a PIPE_BR and PIPE_NONE section.  The snapshot tool
seems to expect this for x1-85 as well.  I guess this was just a bug in
downstream kgsl, which went unnoticed?

Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/666662/

drm/msm: Fix debugbus snapshot

We weren't setting the # of captured debugbus blocks.

Reported-by: Connor Abbott <cwabbott0@gmail.com>
Suggested-by: Connor Abbott <cwabbott0@gmail.com>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/666660/

drm/msm: Fix a7xx debugbus read

The bitfield positions changed in a7xx.

v2: Don't open-code the bitfield building
v3: Also fix cx_debugbus

Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/666659/

drm/msm: Constify snapshot tables

A bit of divergence from the downstream driver from which these headers
were imported.  But no need for these tables not to be const.

Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/666656/

drm/msm: Fix order of selector programming in cluster snapshot

Program the selector _after_ selecting the aperture.  This aligns with
the downstream driver, and fixes a case where we were failing to capture
ctx0 regs (and presumably what we thought were ctx1 regs were actually
ctx0).

Suggested-by: Akhil P Oommen <akhilpo@oss.qualcomm.com>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/666655/

drm/msm: Fix section names and sizes

The section names randomly appended _DATA or _ADDR in many cases, and/or
didn't match the reg names.  Fix them so crashdec can properly resolve
the section names back to reg names.

Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/666654/

drm/msm: Add missing "location"s to devcoredump

This is needed to properly interpret some of the sections.

v2: Fix missing \n

Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/666651/

drm/msm: Fix dereference of pointer minor before null check

Currently the pointer minor is being dereferenced before it is null
checked, leading to a potential null pointer dereference issue. Fix this
by dereferencing the pointer only after it has been null checked. Also
Replace minor->dev with dev.

Fixes: 4f89cf40d01e ("drm/msm: bail out late_init_minor() if it is not a GPU device")
Signed-off-by: Colin Ian King <colin.i.king@gmail.com>
Reviewed-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/666259/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

drm/msm: Defer fd_install in VM_BIND ioctl

Avoid fd_install() until there are no more potential error paths, to
avoid put_unused_fd() after the fd is made visible to userspace.

Fixes: 2e6a8a1fe2b2 ("drm/msm: Add VM_BIND ioctl")
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/665365/

drm/msm: Defer fd_install in SUBMIT ioctl

Avoid fd_install() until there are no more potential error paths, to
avoid put_unused_fd() after the fd is made visible to userspace.

Fixes: 68dc6c2d5eec ("drm/msm: Fix submit error-path leaks")
Reported-by: Dan Carpenter <dan.carpenter@linaro.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/665363/

drm/msm: Fix submit error path cleanup

submit_unpin_objects() should come before we unlock the objects.  This
fixes the splat:

   WARNING: CPU: 2 PID: 2171 at drivers/gpu/drm/msm/msm_gem.h:395 msm_gem_unpin_locked+0x8c/0xd8 [msm]
   Modules linked in: uinput snd_seq_dummy snd_hrtimer aes_ce_ccm snd_soc_wsa884x regmap_sdw q6prm_clocks q6apm_lpass_dais q6apm_dai snd_q6dsp_common q6prm snd_q6apm qcom_pd_mapper cdc_mbim cdc_wdm cdc_ncm r8153_ecm cdc_ether usbnet sunrpc nls_ascii nls_cp437 vfat fat snd_soc_x1e80100 snd_soc_lpass_rx_macro snd_soc_lpass_tx_macro snd_soc_lpass_va_macro snd_soc_lpass_wsa_macro snd_soc_qcom_common soundwire_qcom snd_soc_lpass_macro_common snd_soc_hdmi_codec snd_soc_qcom_sdw ext4 snd_soc_core snd_compress soundwire_bus snd_pcm_dmaengine snd_seq mbcache jbd2 snd_seq_device snd_pcm pm8941_pwrkey snd_timer r8152 qcom_spmi_temp_alarm industrialio snd lenovo_yoga_slim7x ath12k mii arm_smccc_trng soundcore rng_core evdev loop panel_samsung_atna33xc20 msm ubwc_config drm_client_lib drm_gpuvm drm_exec gpu_sched drm_display_helper pmic_glink_altmode aux_hpd_bridge ucsi_glink qcom_battmgr phy_qcom_qmp_combo ps883x cec aux_bridge drm_dp_aux_bus i2c_hid_of aes_ce_blk drm_kms_helper aes_ce_cipher i2c_hid qcom_q6v5_pas
    ghash_ce qcom_pil_info drm sha1_ce qcom_common phy_snps_eusb2 qcom_geni_serial qcom_q6v5 qcom_sysmon pinctrl_sm8550_lpass_lpi lpasscc_sc8280xp sbsa_gwdt mdt_loader gpio_keys pmic_glink i2c_dev efivarfs autofs4
   CPU: 2 UID: 1000 PID: 2171 Comm: gnome-shell Not tainted 6.16.0-rc4-debug+ #25 PREEMPT(voluntary)
   Hardware name: LENOVO 83ED/LNVNB161216, BIOS NHCN53WW 08/02/2024
   pstate: 61400005 (nZCv daif +PAN -UAO -TCO +DIT -SSBS BTYPE=--)
   pc : msm_gem_unpin_locked+0x8c/0xd8 [msm]
   lr : msm_gem_unpin_locked+0x88/0xd8 [msm]
   sp : ffff80009c963820
   x29: ffff80009c963820 x28: ffff80009c9639f8 x27: ffff00080552a830
   x26: 0000000000000000 x25: ffff0009d5655800 x24: 0000000000000000
   x23: 0000000000000000 x22: 0000000000000000 x21: 0000000000000000
   x20: ffff000831db5480 x19: ffff000816e74400 x18: 0000000000000000
   x17: 0000000000000000 x16: ffffc1396afdd720 x15: 0000000000000000
   x14: 0000000000000000 x13: 0000000000000000 x12: ffff0008c065bc00
   x11: ffff0008c065c000 x10: 0000000000000000 x9 : ffffc13945b19074
   x8 : 0000000000000000 x7 : 0000000000000209 x6 : 0000000000000002
   x5 : 0000000000019d01 x4 : ffff0008ba8db080 x3 : 000000000004093f
   x2 : ffff3ed5e727f000 x1 : 0000000000000000 x0 : 0000000000000000
   Call trace:
    msm_gem_unpin_locked+0x8c/0xd8 [msm] (P)
    msm_ioctl_gem_submit+0x32c/0x1760 [msm]
    drm_ioctl_kernel+0xc8/0x138 [drm]
    drm_ioctl+0x2c8/0x618 [drm]
    __arm64_sys_ioctl+0xac/0x108
    invoke_syscall.constprop.0+0x64/0xe8
    el0_svc_common.constprop.0+0x40/0xe8
    do_el0_svc+0x24/0x38
    el0_svc+0x54/0x1d8
    el0t_64_sync_handler+0x10c/0x138
    el0t_64_sync+0x19c/0x1a0
   irq event stamp: 2185036
   hardirqs last  enabled at (2185035): [<ffffc1396afeef9c>] _raw_spin_unlock_irqrestore+0x74/0x80
   hardirqs last disabled at (2185036): [<ffffc1396afd8164>] el1_dbg+0x24/0x90
   softirqs last  enabled at (2184778): [<ffffc13969675e44>] fpsimd_restore_current_state+0x3c/0x328
   softirqs last disabled at (2184776): [<ffffc13969675e14>] fpsimd_restore_current_state+0xc/0x328
   ---[ end trace 0000000000000000 ]---

Fixes: 111fdd2198e6 ("drm/msm: drm_gpuvm conversion")
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/665357/

drm/msm: Fix refcnt underflow in error path

If we hit an error path in GEM obj creation before msm_gem_new_handle()
updates obj->resv to point to the gpuvm resv object, then obj->resv
still points to &obj->_resv.  In this case we don't want to decrement
the refcount of the object being freed (since the refcnt is already
zero).  This fixes the following splat:

   ------------[ cut here ]------------
   refcount_t: underflow; use-after-free.
   WARNING: CPU: 9 PID: 7013 at lib/refcount.c:28 refcount_warn_saturate+0xf4/0x148
   Modules linked in: uinput snd_seq_dummy snd_hrtimer aes_ce_ccm snd_soc_wsa884x regmap_sdw q6prm_clocks q6apm_lpass_da>
    qcom_pil_info i2c_hid drm_kms_helper qcom_common qcom_q6v5 phy_snps_eusb2 qcom_geni_serial drm qcom_sysmon pinctrl_s>
   CPU: 9 UID: 1000 PID: 7013 Comm: deqp-vk Not tainted 6.16.0-rc4-debug+ #25 PREEMPT(voluntary)
   Hardware name: LENOVO 83ED/LNVNB161216, BIOS NHCN53WW 08/02/2024
   pstate: 61400005 (nZCv daif +PAN -UAO -TCO +DIT -SSBS BTYPE=--)
   pc : refcount_warn_saturate+0xf4/0x148
   lr : refcount_warn_saturate+0xf4/0x148
   sp : ffff8000a2073920
   x29: ffff8000a2073920 x28: 0000000000000010 x27: 0000000000000010
   x26: 0000000000000042 x25: ffff000810e09800 x24: 0000000000000010
   x23: ffff8000a2073b94 x22: ffff000ddb22de00 x21: ffff000ddb22dc00
   x20: ffff000ddb22ddf8 x19: ffff0008024934e0 x18: 000000000000000a
   x17: 0000000000000000 x16: ffff9f8c67d77340 x15: 0000000000000000
   x14: 00000000ffffffff x13: 2e656572662d7265 x12: 7466612d65737520
   x11: 3b776f6c66726564 x10: 00000000ffff7fff x9 : ffff9f8c67506c70
   x8 : ffff9f8c69fa26f0 x7 : 00000000000bffe8 x6 : c0000000ffff7fff
   x5 : ffff000f53e14548 x4 : ffff6082ea2b2000 x3 : ffff0008b86ab080
   x2 : 0000000000000000 x1 : 0000000000000000 x0 : ffff0008b86ab080
   Call trace:
    refcount_warn_saturate+0xf4/0x148 (P)
    msm_gem_free_object+0x248/0x260 [msm]
    drm_gem_object_free+0x24/0x40 [drm]
    msm_gem_new+0x1c4/0x1e0 [msm]
    msm_gem_new_handle+0x3c/0x1a0 [msm]
    msm_ioctl_gem_new+0x38/0x70 [msm]
    drm_ioctl_kernel+0xc8/0x138 [drm]
    drm_ioctl+0x2c8/0x618 [drm]
    __arm64_sys_ioctl+0xac/0x108
    invoke_syscall.constprop.0+0x64/0xe8
    el0_svc_common.constprop.0+0x40/0xe8
    do_el0_svc+0x24/0x38
    el0_svc+0x54/0x1d8
    el0t_64_sync_handler+0x10c/0x138
    el0t_64_sync+0x19c/0x1a0
   irq event stamp: 3698694
   hardirqs last  enabled at (3698693): [<ffff9f8c675021dc>] __up_console_sem+0x74/0x90
   hardirqs last disabled at (3698694): [<ffff9f8c68ce8164>] el1_dbg+0x24/0x90
   softirqs last  enabled at (3697578): [<ffff9f8c6744ec5c>] handle_softirqs+0x454/0x4b0
   softirqs last disabled at (3697567): [<ffff9f8c67360244>] __do_softirq+0x1c/0x28
   ---[ end trace 0000000000000000 ]---

Fixes: b58e12a66e47 ("drm/msm: Add _NO_SHARE flag")
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/665355/

drm/msm: Update global fault counter when faulty process has already ended

The global fault counter is no longer used since commit 12578c075f89
("drm/msm/gpu: Skip retired submits in recover worker"). However, it's
still needed, as we need to handle cases where a GPU fault occurs after
the faulting process has already ended.

Hence, increment the global fault counter when the submitting process
had already ended. This way, the number of faults returned by
MSM_PARAM_FAULTS will stay consistent.

While here, s/unusuable/unusable.

Fixes: 12578c075f89 ("drm/msm/gpu: Skip retired submits in recover worker")
Signed-off-by: Maíra Canal <mcanal@igalia.com>
Patchwork: https://patchwork.freedesktop.org/patch/664853/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

drm/msm: Fix pagetables setup/teardown serialization

An atomic counter is not sufficient, as one task could still be in the
process of tearing things down while another task increments the counter
back up to one and begins setup again.  The race condition existed since
commit b145c6e65eb0 ("drm/msm: Add support to create a local pagetable")
but got bigger in commit dbbde63c9e9d ("drm/msm: Add PRR support").

Fixes: dbbde63c9e9d ("drm/msm: Add PRR support")
Fixes: b145c6e65eb0 ("drm/msm: Add support to create a local pagetable")
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/664433/

drm/msm: Fix build with KMS disabled

When commit 98290b0a7d60 ("drm/msm: make it possible to disable
KMS-related code.") was rebased on top of commit 3bebfd53af0f ("drm/msm:
Defer VMA unmap for fb unpins"), the additional use of msm_kms was
overlooked, resulting in a build break when KMS is disabled.  Add some
additional ifdef to fix that.

Reported-by: Arnd Bergmann <arnd@arndb.de>
Fixes: 98290b0a7d60 ("drm/msm: make it possible to disable KMS-related code.")
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Jessica Zhang <jessica.zhang@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/663240/

drm/gpuvm: Wrap drm_gpuvm_sm_map_exec_lock() expected usage in literal code block

Stephen Rothwell reports multiple indentation warnings when merging
drm-msm tree:

Documentation/gpu/drm-mm:506: ./drivers/gpu/drm/drm_gpuvm.c:2445: ERROR: Unexpected indentation. [docutils]
Documentation/gpu/drm-mm:506: ./drivers/gpu/drm/drm_gpuvm.c:2447: WARNING: Block quote ends without a blank line; unexpected unindent. [docutils]
Documentation/gpu/drm-mm:506: ./drivers/gpu/drm/drm_gpuvm.c:2451: WARNING: Definition list ends without a blank line; unexpected unindent. [docutils]
Documentation/gpu/drm-mm:506: ./drivers/gpu/drm/drm_gpuvm.c:2452: WARNING: Definition list ends without a blank line; unexpected unindent. [docutils]
Documentation/gpu/drm-mm:506: ./drivers/gpu/drm/drm_gpuvm.c:2456: ERROR: Unexpected indentation. [docutils]
Documentation/gpu/drm-mm:506: ./drivers/gpu/drm/drm_gpuvm.c:2457: WARNING: Definition list ends without a blank line; unexpected unindent. [docutils]
Documentation/gpu/drm-mm:506: ./drivers/gpu/drm/drm_gpuvm.c:2458: WARNING: Definition list ends without a blank line; unexpected unindent. [docutils]
Documentation/gpu/drm-mm:506: ./drivers/gpu/drm/drm_gpuvm.c:2459: WARNING: Definition list ends without a blank line; unexpected unindent. [docutils]

Fix these by wrapping drm_gpuvm_sm_map_exec_lock() expected usage
example in literal code block.

Fixes: 471920ce25d5 ("drm/gpuvm: Add locking helpers")
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Closes: https://lore.kernel.org/linux-next/20250708192038.6b0fd31d@canb.auug.org.au/
Signed-off-by: Bagas Sanjaya <bagasdotme@gmail.com>
Acked-by: Randy Dunlap <rdunlap@infradead.org>
Tested-by: Randy Dunlap <rdunlap@infradead.org>
Acked-by: Danilo Krummrich <dakr@kernel.org>
Patchwork: https://patchwork.freedesktop.org/patch/663121/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

drm/msm: Small function param doc fix

Reported-by: kernel test robot <lkp@intel.com>
Closes: https://lore.kernel.org/oe-kbuild-all/202507032334.9SCwc952-lkp@intel.com/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

drm/msm: Take the ioctls away from the KMS-only driver

KMS-only drivers should only allocate dumb buffers.  The driver custom
ioctls are only meant for the usermode gpu driver (mesa), and not for
general consumption, so they don't make sense for standalone KMS
drivers.

Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/662598/

drm/msm: Clean up split driver features

Avoid the possibility of missing features between the split and unified
drm driver cases by defining DRIVER_FEATURES_GPU / KMS and using those
in the drm_driver initializations.

Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/662595/

drm/msm: enable separate binding of GPU and display devices

There are cases when we want to have separate DRM devices for GPU and
display pipelines.
One example is development, when it is beneficial to be able to bind the
GPU driver separately, without the display pipeline (and without the
hacks adding "amd,imageon" to the compatible string).
Another example is some of Qualcomm platforms, which have two MDSS
units, but only one GPU. With current approach it is next to impossible
to support this usecase properly, while separate binding allows users to
have three DRM devices: two for MDSS units and a single headless GPU.

Add kernel param msm.separate_gpu_kms, which if set to true forces
creation of separate display and GPU DRM devices. Mesa supports this
setup by using the kmsro wrapper.

The param is disabled by default, in order to be able to test userspace
for the compatibility issues. Simple clients are able to handle this
setup automatically.

Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/662590/
[Rob: renamed the modparam to separate_gpu_kms, and add missing
      DRIVER_GEM_GPUVA]
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

drm/msm: rework binding of Imageon GPUs

Currently the msm driver creates an extra interim platform device for
Imageon GPUs. This is not ideal, as the device doesn't have
corresponding OF node. If the headless mode is used for newer GPUs, then
the msm_use_mmu() function can not detect corresponding IOMMU devices.
Also the DRM device (although it's headless) is created with modesetting
flags being set.

To solve all these issues, rework the way the Imageon devices are bound.
Remove the interim device, don't register a component and instead use a
cut-down version of the normal functions to probe or remove the driver.

Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/662584/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

drm/msm: rearrange symbol selection

Move symbol selection to be more fine grained: select DP helpers only if
DP driver is also enabled, move KMS and display helpers to the newly
introduced DRM_MSM_KMS.

Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/662589/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

drm/msm: bail out late_init_minor() if it is not a GPU device

Both perf and hangrd make sense only for GPU devices. Bail out if we are
registering a KMS-only device.

Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/662583/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

drm/msm: make it possible to disable KMS-related code.

If the Adreno device is used in a headless mode, there is no need to
build all KMS components. Build corresponding parts conditionally, only
selecting them if modeset support is actually required.

Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/662581/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

drm/msm: move KMS driver data to msm_kms

Data for HDMI, DSI and DP blocks only makes sense for the KMS parts of
the driver. Move corresponding data pointers from struct msm_drm_private
to struct msm_kms.

Suggested-by: Abhinav Kumar <quic_abhinavk@quicinc.com>
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/662580/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

drm/msm: get rid of msm_drm_private::num_crtcs

Drop superfluous msm_drm_private::num_crtcs in favour of using
drm_mode_config::num_crtc or MAX_CRCS as appropriate.

Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/662578/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

drm/msm/mdp4: get rid of mdp4_crtc.id

There is no reason to store CRTC id, it's a part of the drm_crtc. Drop
this member and use drm_crtc.name for the warning message.

Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/662576/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

drm/msm: move helper calls to msm_kms.c

Extract two more KMS-related codepieces to msm_kms.c, removing last
pieces of KMS code from msm_drv.c.

Reviewed-by: Abhinav Kumar <quic_abhinavk@quicinc.com>
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/662574/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

drm/msm: move wq handling to KMS code

The global workqueue is only used for vblanks inside KMS code. Move
allocation / flushing / deallcation of it to msm_kms.c

Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/662573/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

drm/msm/adreno: Switch to the common UBWC config struct

Now that Adreno specifics are out of the way, use the common config
(but leave the HBB hardcoding in place until that is wired up on the
other side).

Acked-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/660985/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

soc: qcom: ubwc: Fill in UBWC swizzle cfg for platforms that lack one

The UBWC 1.0 case is easy - it must be all 3 enabled.
UBWC2.0 and 3.x require that level1 is removed, follow suit.

Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/660983/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

soc: qcom: ubwc: Add #defines for UBWC swizzle bits

Make the values a bit more meaningful.

This commit is intentionally cross-subsystem to ease review, as the
patchset is intended to be merged together, with a maintainer
consensus.

Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/660981/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

soc: qcom: ubwc: Fix SM6125's ubwc_swizzle value

The value of 7 (a.k.a. GENMASK(2, 0), a.k.a. disabling levels 1-3 of
swizzling) is what we want on this platform (and others with a UBWC
1.0 encoder).

Fix it to make mesa happy (the hardware doesn't care about the 2 higher
bits, as they weren't consumed on this platform).

Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/660980/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

drm/msm/a6xx: Simplify min_acc_len calculation

It's only necessary for some lower end parts.
Also rename it to min_acc_len_64b to denote that if set, the minimum
access length is 64 bits, 32b otherwise.

Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/660977/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

drm/msm/a6xx: Resolve the meaning of rgb565_predicator

It's supposed to be on when the UBWC encoder version is >= 4.0.
Drop the per-GPU assignments.

Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/660975/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

drm/msm/a6xx: Replace '2' with BIT(1) in level2_swizzling_dis calc

ubwc_swizzle is a bitmask. Check for a bit to make it more obvious.

Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/660973/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

drm/msm/a6xx: Resolve the meaning of UBWC_MODE

This bit is set iff the UBWC version is 1.0. That notably does not
include QCM2290's "no UBWC".

This commit is intentionally cross-subsystem to ease review, as the
patchset is intended to be merged together, with a maintainer
consensus.

Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/660971/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

drm/msm/a6xx: Simplify uavflagprd_inv detection

Instead of setting it on a gpu-per-gpu basis, converge it to the
intended "is A650 family or A7xx".

Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/660969/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

drm/msm/a6xx: Resolve the meaning of AMSBC

The bit must be set to 1 if the UBWC encoder version is >= 3.0, drop it
as a separate field.

Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/660967/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

drm/msm/a6xx: Get a handle to the common UBWC config

Start the great despaghettification by getting a pointer to the common
UBWC configuration, which houses e.g. UBWC versions that we need to
make decisions.

Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/660965/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

drm/msm: Use the central UBWC config database

As discussed a lot in the past, the UBWC config must be coherent across
a number of IP blocks (currently display and GPU, but it also may/will
concern camera/video as the drivers evolve).

So far, we've been trying to keep the values reasonable in each of the
two drivers separately, but it really make sense to do so centrally,
especially given certain fields (e.g. HBB) may need to be gathered
dynamically.

To reduce room for error, move to fetching the config from a central
source, so that the data programmed into the hardware is consistent
across all multimedia blocks that request it.

Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/660963/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

drm/msm: Offset MDSS HBB value by 13

The Adreno part of the driver exposes this value to userspace, and the
SMEM data source also presents a x+13 value. Keep things coherent and
make the value uniform across them.

Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/660961/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

soc: qcom: Add UBWC config provider

Add a file that will serve as a single source of truth for UBWC
configuration data for various multimedia blocks.

Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/660959/
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>

drm/msm: Update register xml

Sync register xml from mesa commit eb3e0b7164a3 ("freedreno/a6xx: Split
descriptors out into their own file").

Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Acked-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Patchwork: https://patchwork.freedesktop.org/patch/662470/

drm/msm: Add VM_BIND throttling

A large number of (unsorted or separate) small (<2MB) mappings can cause
a lot of, probably unnecessary, prealloc pages.  Ie. a single 4k page
size mapping will pre-allocate 3 pages (for levels 2-4) for the
pagetable.  Which can chew up a large amount of unneeded memory.  So add
a mechanism to put an upper bound on the # of pre-alloc pages.

Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661529/

drm/msm: Defer VMA unmap for fb unpins

With the conversion to drm_gpuvm, we lost the lazy VMA cleanup, which
means that fb cleanup/unpin when pageflipping to new scanout buffers
immediately unmaps the scanout buffer.  This is costly (with tlbinv,
it can be 4-6ms for a 1080p scanout buffer, and more for higher
resolutions)!

To avoid this, introduce a vma_ref, which is incremented whenever
userspace has a GEM handle or dma-buf fd.  When unpinning if the
vm is the kms->vm we defer tearing down the VMA until the vma_ref
drops to zero.  If the buffer is still part of a flip-chain then
userspace will be holding some sort of reference to the BO, either
via a GEM handle and/or dma-buf fd.  So this avoids unmapping the VMA
when there is a strong possibility that it will be needed again.

Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661538/

drm/msm: Bump UAPI version

Bump version to signal to userspace that VM_BIND is supported.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661535/

drm/msm: use trylock for debugfs

This resolves a potential deadlock vs msm_gem_vm_close().  Otherwise for
_NO_SHARE buffers msm_gem_describe() could be trying to acquire the
shared vm resv, while already holding priv->obj_lock.  But _vm_close()
might drop the last reference to a GEM obj while already holding the vm
resv, and msm_gem_free_object() needs to grab priv->obj_lock, a locking
inversion.

OTOH this is only for debugfs and it isn't critical if we undercount by
skipping a locked obj.  So just use trylock() and move along if we can't
get the lock.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661525/

drm/msm: Add mmu prealloc tracepoint

So we can monitor how many pages are getting preallocated vs how many
get used.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661521/

drm/msm: Add VMA unmap reason

Make the VM log a bit more useful by providing a reason for the unmap
(ie. closing VM vs evict/purge, etc)

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661527/

drm/msm: Add VM logging for VM_BIND updates

When userspace opts in to VM_BIND, the submit no longer holds references
keeping the VMA alive.  This makes it difficult to distinguish between
UMD/KMD/app bugs.  So add a debug option for logging the most recent VM
updates and capturing these in GPU devcoredumps.

The submitqueue id is also captured, a value of zero means the operation
did not go via a submitqueue (ie. comes from msm_gem_vm_close() tearing
down the remaining mappings when the device file is closed.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661518/

drm/msm: Add VM_BIND ioctl

Add a VM_BIND ioctl for binding/unbinding buffers into a VM.  This is
only supported if userspace has opted in to MSM_PARAM_EN_VM_BIND.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661524/

drm/msm: Split out map/unmap ops

With async VM_BIND, the actual pgtable updates are deferred.
Synchronously, a list of map/unmap ops will be generated, but the
actual pgtable changes are deferred.  To support that, split out
op handlers and change the existing non-VM_BIND paths to use them.

Note in particular, the vma itself may already be destroyed/freed
by the time an UNMAP op runs (or even a MAP op if there is a later
queued UNMAP).  For this reason, the op handlers cannot reference
the vma pointer.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661516/

drm/msm: Support pgtable preallocation

Introduce a mechanism to count the worst case # of pages required in a
VM_BIND op.

Note that previously we would have had to somehow account for
allocations in unmap, when splitting a block.  This behavior was removed
in commit 33729a5fc0ca ("iommu/io-pgtable-arm: Remove split on unmap
behavior)"

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661515/

drm/msm: Support IO_PGTABLE_QUIRK_NO_WARN_ON

With user managed VMs and multiple queues, it is in theory possible to
trigger map/unmap errors.  These will (in a later patch) mark the VM as
unusable.  But we want to tell the io-pgtable helpers not to spam the
log.  In addition, in the unmap path, we don't want to bail early from
the unmap, to ensure we don't leave some dangling pages mapped.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661520/

drm/msm: Add VM_BIND submitqueue

This submitqueue type isn't tied to a hw ringbuffer, but instead
executes on the CPU for performing async VM_BIND ops.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661517/

drm/msm: Use DMA_RESV_USAGE_BOOKKEEP/KERNEL

Any place we wait for a BO to become idle, we should use BOOKKEEP usage,
to ensure that it waits for _any_ activity.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661506/

drm/msm: Extract out syncobj helpers

We'll be re-using these for the VM_BIND ioctl.

Also, rename a few things in the uapi header to reflect that syncobj use
is not specific to the submit ioctl.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661512/

drm/msm: rd dumping support for sparse

As with devcoredump, we need to iterate the VMAs to figure out what to
dump.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661510/

drm/msm: Crashdump support for sparse

In this case, we need to iterate the VMAs looking for ones with
MSM_VMA_DUMP flag.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661504/

drm/msm: rd dumping prep for sparse mappings

Similar to the previous commit, add support for dumping partial
mappings.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661514/

drm/msm: Crashdump prep for sparse mappings

In this case, userspace could request dumping partial GEM obj mappings.
Also drop use of should_dump() helper, which really only makes sense in
the old submit->bos[] table world.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661496/

drm/msm: Add _NO_SHARE flag

Buffers that are not shared between contexts can share a single resv
object.  This way drm_gpuvm will not track them as external objects, and
submit-time validating overhead will be O(1) for all N non-shared BOs,
instead of O(n).

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661497/

drm/msm: Mark VM as unusable on GPU hangs

If userspace has opted-in to VM_BIND, then GPU hangs and VM_BIND errors
will mark the VM as unusable.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661499/

drm/msm: Add opt-in for VM_BIND

Add a SET_PARAM for userspace to request to manage to the VM itself,
instead of getting a kernel managed VM.

In order to transition to a userspace managed VM, this param must be set
before any mappings are created.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661494/

drm/msm: Lazily create context VM

In the next commit, a way for userspace to opt-in to userspace managed
VM is added.  For this to work, we need to defer creation of the VM
until it is needed.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661490/

drm/msm: Drop queued submits on lastclose()

If we haven't written the submit into the ringbuffer yet, then drop it.
The submit still retires through the normal path, to preserve fence
signalling order, but we can skip the IB's to userspace cmdstream.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661489/

drm/msm: Rename msm_gem_vma_purge() -> _unmap()

This is a more descriptive name.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661487/

drm/msm: Add PRR support

Add PRR (Partial Resident Region) is a bypass address which make GPU
writes go to /dev/null and reads return zero.  This is used to implement
vulkan sparse residency.

To support PRR/NULL mappings, we allocate a page to reserve a physical
address which we know will not be used as part of a GEM object, and
configure the SMMU to use this address for PRR/NULL mappings.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661486/

drm/msm: Add mmu support for non-zero offset

Only needs to be supported for iopgtables mmu, the other cases are
either only used for kernel managed mappings (where offset is always
zero) or devices which do not support sparse bindings.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661501/

drm/msm: Split out helper to get iommu prot flags

We'll re-use this in the vm_bind path.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661484/

drm/msm: Use drm_gpuvm types more

Most of the driver code doesn't need to reach in to msm specific fields,
so just use the drm_gpuvm/drm_gpuva types directly.  This should
hopefully improve commonality with other drivers and make the code
easier to understand.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661483/

drm/msm: Convert vm locking

Convert to using the gpuvm's r_obj for serializing access to the VM.
This way we can use the drm_exec helper for dealing with deadlock
detection and backoff.

This will let us deal with upcoming locking order conflicts with the
VM_BIND implmentation (ie. in some scenarious we need to acquire the obj
lock first, for ex. to iterate all the VMs an obj is bound in, and in
other scenarious we need to acquire the VM lock first).

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661478/

drm/msm: drm_gpuvm conversion

Now that we've realigned deletion and allocation, switch over to using
drm_gpuvm/drm_gpuva.  This allows us to support multiple VMAs per BO per
VM, to allow mapping different parts of a single BO at different virtual
addresses, which is a key requirement for sparse/VM_BIND.

This prepares us for using drm_gpuvm to translate a batch of MAP/
MAP_NULL/UNMAP operations from userspace into a sequence of map/remap/
unmap steps for updating the page tables.

Since, unlike our prior vm/vma setup, with drm_gpuvm the vm_bo holds a
reference to the GEM object.  To prevent reference loops causing us to
leak all GEM objects, we implicitly tear down the mapping when the GEM
handle is close or when the obj is unpinned.  Which means the submit
needs to also hold a reference to the vm_bo, to prevent the VMA from
being torn down while the submit is in-flight.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661479/

drm/msm: Refcount framebuffer pins

We were already keeping a refcount of # of prepares (pins), to clear the
iova array.  Use that to avoid unpinning the iova until the last cleanup
(unpin).  This way, when msm_gem_unpin_iova() actually tears down the
mapping, we won't have problems if the fb is being scanned out on
another display (for example).

Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661477/

drm/msm: Stop passing vm to msm_framebuffer

The fb only deals with kms->vm, so make that explicit.  This will start
letting us refcount the # of times the fb is pinned, so we can only
unpin the vma after last user of the fb is done.  Having a single
reference count really only works if there is only a single vm.

Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661476/

drm/msm: Don't close VMAs on purge

Previously we'd also tear down the VMA, making the address space
available again.  But with drm_gpuvm conversion, this would require
holding the locks of all VMs the GEM object is mapped in.  Which is
problematic for the shrinker.

Instead just let the VMA hang around until the GEM object is freed.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661472/

drm/msm: Collapse vma close and delete

This fits better drm_gpuvm/drm_gpuva.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661470/

drm/msm: Collapse vma allocation and initialization

Now that we've dropped vram carveout support, we can collapse vma
allocation and initialization.  This better matches how things work
with drm_gpuvm.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661471/

drm/msm: Remove vram carveout support

It is standing in the way of drm_gpuvm / VM_BIND support.  Not to
mention frequently broken and rarely tested.  And I think only needed
for a 10yr old not quite upstream SoC (msm8974).

Maybe we can add support back in later, but I'm doubtful.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661467/

drm/msm: Rename msm_gem_address_space -> msm_gem_vm

Re-aligning naming to better match drm_gpuvm terminology will make
things less confusing at the end of the drm_gpuvm conversion.

This is just rename churn, no functional change.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661466/

drm/msm: Improve msm_context comments

Just some tidying up.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661461/