drm/msm/mdp5: Add a CAP for Source Split

Some of the newer MDP5 versions support Source Split of SSPPs. It is a
feature that allows us to route the output of a hwpipe to 2 Layer
Mixers. This is required to achieve the following use cases:

- Dual DSI: For high res DSI panels (such as 2560x1600 etc), a single
  DSI interface doesn't have the bandwidth to drive the required pixel
  clock. We use 2 DSI interfaces to drive the left and right halves
  of the panel (i.e, 1280x1600 each). The MDP5 pipeline here would look
  like:

         LM0 -- DSPP0 -- INTF1 -- DSI1
        /
hwpipe--
        \
         LM1 -- DSPP1 -- INTF2 -- DSI2

  A single hwpipe is used to scan out the left and right halves to DSI1
  and DSI2 respectively. In order to do this, we need to configure the
  2 Layer Mixers in Source Split mode.

- HDMI 4K: In order to support resolutions with width higher than the
  max width supported by a hwpipe, we club 2 hwpipes together:

hwpipe1 --- LM0 -- DSPP0
       -   -             \
         -                -- 3D Mux -- INTF0 -- HDMI
       -   -             /
hwpipe2 --- LM1 -- DSPP1

  hwpipe1 is staged on the 'left' Layer Mixer, and hwpipe2 is staged on
  the 'right' Layer Mixer. An additional block called the '3D Mux' is
  used to merge the output of the 2 DSPPs to a single interface.
  In this use case, it is possible that a 4K surface is downscaled and
  placed completely within one of the halves. In order to support such
  scenarios (and keep the programming simple), Layer Mixers with Source
  Split can be assigned 2 hw pipes per stage. While scanning out, the HW
  takes care of fetching the pixels fom the correct pipe.

Add a MDP cap to tell whether the HW supports source split or not.
Add a MDP LM cap that tells whether a LM instance can operate in
source split mode (and generate the 'left' part of the display
output).

Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>

drm/msm/mdp5: Remove mixer/intf pointers from mdp5_ctl

These are a part of CRTC state, it doesn't feel nice to leave them
hanging in mdp5_ctl struct. Pass mdp5_pipeline pointer instead
wherever it is needed.

We still have some params in mdp5_ctl like start_mask etc which
are derivative of atomic state, and should be rolled back if
a commit fails, but it doesn't seem to cause much trouble.

Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>

drm/msm/mdp5: Start using parameters from CRTC state

In the last few commits, we've been adding params to mdp5_crtc_state, and
assigning them in the atomic_check() funcs. Now it's time to actually
start using them.

Remove the duplicated params from the mdp5_crtc struct, and start using
them in the mdp5_crtc code. The majority of the references to these params
is in code that executes after the atomic swap has occurred, so it's okay
to use crtc->state in them. There are a couple of legacy LM cursor ops that
may not use the updated state, but (I think) it's okay to live with that.

Now that we dynamically allocate a mixer to the CRTC, we can also remove
the static assignment to it in mdp5_crtc_init, and also drop the code that
skipped init-ing WB bound mixers (those will now be rejected by
mdp5_mixer_assign()).

Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>

drm/msm/mdp5: Add more stuff to CRTC state

Things like vblank/err irq masks, mode of operation (command mode or not)
are derivative of the interface and mixer state. Therefore, they need to
be a part of the CRTC state too.

Add them to mdp5_crtc_state, and assign them in the CRTC's atomic_check()
func, so that it can be rolled back to a clean state.

Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>

drm/msm/mdp5: Assign INTF and CTL in encoder's atomic_check()

The INTF and CTL used in a display pipeline are going to be maintained as
a part of the CRTC state (i.e, in mdp5_crtc_state).

These entities, however, are currently statically assigned to drm_encoders
(i.e. mdp5_encoder). Since these aren't directly visible to the CRTC, we
assign them to the CRTC state in the encoder's atomic_check() op.

With this approach, we assign portions of CRTC state in two different
places: the layer mixer in CRTC's atomic_check(), and the INTF and CTL
pieces in the encoder's atomic_check() op.

We'd have more options here if the drm core maintained encoder state too,
but the current approach of clubbing everything in CRTC's state works just
fine.

Unlike hwpipes and mixers, we don't need to keep a track of INTF/CTL
assignments in the global atomic state. This is because they're currently
not sharable resources. For example, INTF0 and CTL0 will always be assigned
to one drm_encoder. This can change later when we implement writeback and
want a CRTC to use a CTL for a while, and then release it for others to use
it. Or, when a drm_encoder can switch between using a single INTF vs
2 INTFs.

Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>

drm/msm/mdp5: Prepare for dynamic assignment of mixers

Add the stuff needed to allow dynamically assigning a mixer to a CRTC.

Since mixers are a resource that can be shared across multiple CRTCs, we
need to maintain a 'hwmixer_to_crtc' map in the global atomic state,
acquire the mdp5_kms.state_lock modeset lock and so on.

The mixer is assigned in the CRTC's atomic_check() func, a failure will
result in the new state being cleanly rolled back.

The mixer assignment itself is straightforward, and almost identical to
what we do for hwpipes. We don't need to grab the old hwmixer_to_crtc
state like we do in hwpipes since we don't need to compare anything
with the old state at the moment.

The only LM capability we care about at the moment is whether the mixer
instance can be used to display stuff (i.e, connect to an INTF
downstream).

Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>

drm/msm/mdp5: subclass CRTC state

Subclass drm_crtc_state so that we can maintain additional state for
our CRTCs.

Add mdp5_pipeline and mdp5_ctl pointers in the subclassed state.
mdp5_pipeline is a grouping of the HW entities that forms the downstream
pipeline for a particular CRTC. It currently contains pointers to
mdp5_interface and mdp5_hw_mixer tied to this CRTC. Later, we will
have 2 hwmixers in this struct. (We could also have 2 intfs if we want
to support dual DSI with Source Split enabled. Implementing that feature
isn't planned at the moment).

The mdp5_pipeline state isn't used at the moment. For now, we just
introduce mdp5_crtc_state and the crtc funcs needed to manage the
subclassed state.

Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>

drm/msm/mdp5: Remove the pipeline stuff in mdp5_ctl

The mdp5_ctl has an 'op_mode' struct which contains info on
the downstream pipeline.

Grouping these params together in a struct doesn't serve much
purpose in the code. Maybe there was a plan to expand this
further that never happened.

Remove the op_mode struct, and place its members directly in
mdp5_ctl. This will help avoid confusion later when I introduce
my own verion of a mdp5 pipeline :)

Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>

drm/msm/mdp5: Clean up interface assignment

mdp5_interface struct contains data corresponding to a INTF
instance in MDP5 hardware. This sturct is memcpy'd to the
mdp5_encoder struct, and then later to the mdp5_ctl struct.

Instead of copying around interface data, create mdp5_interface
instances in mdp5_init, like how it's done currently done for
pipes and layer mixers. Pass around the interface pointers to
mdp5_encoder and mdp5_ctl. This simplifies the code, and allows
us to decouple encoders from INTFs in the future if needed.

Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>

drm/msm/mdp5: Simplify LM <-> PP mapping

PingPong ID for a Layer Mixer is already contained in
mdp5_hw_mixer.

This avoids the need to retrieve PP ID using macros

Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>

drm/msm/mdp5: Start using mdp5_hw_mixer

Use the mdp5_hw_mixer struct in the mdp5_crtc and mdp5_ctl instead of
using the LM index.

Like before, the Layer Mixers are assigned statically to the CRTCs.
The hwmixer(s) will later be dynamically assigned to CRTCs.

For now, ignore the hwmixers that can only do WB.

Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>

drm/msm/mdp5: Add structs for hw Layer Mixers

Create a struct to represent MDP5 Layer Mixer instances. This will
eventually allow us to detach CRTCs from the Layer Mixers, and
generally clean things up a bit.

This is very similar to how hwpipes were previously abstracted away
from drm planes.

Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>

drm/msm/mdp5: describe LM instances in mdp5_cfg

The number of Layer Mixers and the downstream blocks (DSPPs and PPs)
connected to each LM can vary with different MDP5 revisions. These
parameters are also static.

Keep the per instance LM data in mdp5_cfg. This will avoid the need
to have macros which identify PP id or DSPP id the LM is connected
to. We don't configure DSPPs at the moment, but keeping the DSPP
instance # here might come handy later.

Also add a 'caps' field that identifies features supported by a
LM instance. Introduce the caps MDP_LM_CAP_DISPLAY and MDP_LM_CAP_WB
that identify whether a LM instance can be used for display or
writeback.

Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>

drm/msm/mdp5: Bring back pipe_lock to mdp5_plane struct

We'd previously moved the pipe_lock spinlock to the hwpipe struct. Bring
it back to mdp5_plane. We will need this because an mdp5_plane in the
future could comprise of 2 hw pipes. It makes more sense to have a single
lock to protect the registers for the hw pipes used by a plane, rather
than trying to take individual locks per hwpipe when committing a
configuration.

Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>

drm/msm/hdmi: redefinitions of macros not required

4 macros already defined in hdmi.h,
which is not required to redefine in hdmi_audio.c

Signed-off-by: Vinay Simha BN <simhavcs@gmail.com>
Signed-off-by: Rob Clark <robdclark@gmail.com>

drm/msm/mdp5: Update SSPP_MAX value

'SSPP_MAX + 1' is the max number of hwpipes that can be present on a
MDP5 platform. Recently, 2 new cursor hwpipes were added, which
caused overflows in arrays that used SSPP_MAX to represent the number
of elements. Update the SSPP_MAX value to incorporate the extra
hwpipes.

Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>

drm/msm/dsi: Fix bug in dsi_mgr_phy_enable

A recent commit introduces a bug in dsi_mgr_phy_enable. In the non
dual DSI mode, we reset the mdsi (master DSI) PHY. This isn't right
since master and slave DSI exist only in dual DSI mode. For the normal
mode of operation, we should simply reset the PHY of the DSI device
(i.e. msm_dsi) corresponding to the current bridge.

Usage of the wrong DSI pointer also resulted in a static checker
warning. That too is resolved with this fix.

Fixes: b62aa70a98c5 (drm/msm/dsi: Move PHY operations out of host)

Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Archit Taneja <architt@codeaurora.org>
Reviewed-by: Rob Clark <robdclark@gmail.com>
Signed-off-by: Rob Clark <robdclark@gmail.com>

drm/msm: Don't allow zero sized buffer objects

Zero sized buffer objects tend to make various bits of the GEM
infrastructure complain:

 WARNING: CPU: 1 PID: 2323 at drivers/gpu/drm/drm_mm.c:389 drm_mm_insert_node_generic+0x258/0x2f0
 Modules linked in:

 CPU: 1 PID: 2323 Comm: drm-api-test Tainted: G        W 4.9.0-rc4-00906-g693af44 #213
 Hardware name: Qualcomm Technologies, Inc. DB820c (DT)
 task: ffff8000d7353400 task.stack: ffff8000d7720000
 PC is at drm_mm_insert_node_generic+0x258/0x2f0
 LR is at drm_vma_offset_add+0x4c/0x70

Zero sized buffers serve no appreciable value to the user so disallow
them at create time.

Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>

drm/msm: Support 64 bit iova in RD_CMDSTREAM_ADDR

Output the upper 32 bits of a 64 bit iova in the RD_CMDSTREAM_ADDR
section while maintaining backwards compatibility for tools that
only understand 32 bit iovas.

Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>

drm/msm: Pass interrupt status to a5xx_rbbm_err_irq()

The interrupt status was being cleared before processing the handlers.
a5xx_rbbm_err_irq() was checking the interrupt status again, which would
likely turn out bad because the interrupt status would be 0 (or at least
different). Pass the original status to the function instead.

Also, skip clearing RBBM_AHB_ERROR from the interrupt status. The interrupt
will keep firing until the error source is cleared.  Skip the clear to
avoid a storm until the error is cleared in a5xx_rbbm_err_irq().

Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>

drm/msm: Don't increase priv->num_aspaces until we know that it fits

priv->num_aspaces is increased and then checked to see if it still fits
in the priv->aspace array.  If it doesn't, we warn and exit but
priv->num_aspaces remains incremented.

Don't incremement the count until we know that it fits in the array.

Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>

drm/msm: Fix wrong pointer check in a5xx_destroy

Instead of checking for a5xx_gpu->gpmu_iova during destroy we
accidently check a5xx_gpu->gpmu_bo.

Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>

drm/msm: Simplify vblank event delivery

The core takes care of handling the send_event vs. close() issues, we
can remove that driver code.

Cc: Rob Clark <robdclark@gmail.com>
Cc: linux-arm-msm@vger.kernel.org
Cc: freedreno@lists.freedesktop.org
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Signed-off-by: Rob Clark <robdclark@gmail.com>

drm/msm: switch to postclose

I didn't spot anything that would require ordering here (well not
anywhere else either), and I'm trying to unify at least modern drivers
on one close hook.

Cc: Rob Clark <robdclark@gmail.com>
Cc: linux-arm-msm@vger.kernel.org
Cc: freedreno@lists.freedesktop.org
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Signed-off-by: Rob Clark <robdclark@gmail.com>

drm/msm: adreno: fix build error without debugfs

The newly added a5xx support fails to build when debugfs is diabled:

drivers/gpu/drm/msm/adreno/a5xx_gpu.c:849:4: error: 'struct msm_gpu_funcs' has no member named 'show'
drivers/gpu/drm/msm/adreno/a5xx_gpu.c:849:11: error: 'a5xx_show' undeclared here (not in a function); did you mean 'a5xx_irq'?

This adds a missing #ifdef.

Fixes: b5f103ab98c7 ("drm/msm: gpu: Add A5XX target support")
Cc: stable@vger.kernel.org
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Rob Clark <robdclark@gmail.com>

drm/msm: move submit fence wait out of struct_mutex

Probably a symptom of needing finer grained locking, but if we wait on
the incoming fence-fd (which could come from a different context) while
holding struct_mutex, that blocks retire_worker so gpu fences cannot get
signalled.

This causes a problem if userspace manages to get more than a frame
ahead, leaving the atomic-commit worker blocked waiting on fences that
cannot be signaled because submit is blocked waiting for a fence
signalled from vblank (after the atomic commit which is blocked).

If we start having multiple fence ctxs for the gpu, submit_fence_sync()
would probably need to move outside of struct_mutex as well.

Signed-off-by: Rob Clark <robdclark@gmail.com>

drm/msm: pm runtime support for iommu

In particular, attach() and unmap() need pm-runtime get/put to ensure
iommu clks are enabled.

Signed-off-by: Rob Clark <robdclark@gmail.com>

drm/msm: convert to iommu_map_sg

Significantly simplifies things.  Also iommu_unmap() can unmap an entire
iova range.

(If backporting to downstream kernel you might need to revert this.  Or
at least double check older iommu implementation.)

Signed-off-by: Rob Clark <robdclark@gmail.com>

drm/msm/adreno: reset ringbuffer in hw_init

We need to do this also in resume path when we need to re-hw_init().

Signed-off-by: Rob Clark <robdclark@gmail.com>

drm/msm/gpu: use pm-runtime

We need to use pm-runtime properly when IOMMU is using device_link() to
control it's own clocks.

Signed-off-by: Rob Clark <robdclark@gmail.com>

drm/msm/gpu: move suspend/resume into debugfs->show

Each of the per-generation callbacks was doing this.  Lets just simplify
and move it into toplevel show() fxn.

Signed-off-by: Rob Clark <robdclark@gmail.com>

Merge branch 'drm-next-4.12' of git://people.freedesktop.org/~agd5f/linux into drm-next

A few more things for 4.12:
- ttm and amdgpu support for non-contiguous vram CPU mappings
- lots of bug fixes and cleanups for vega10
- misc bug fixes and code cleanups

[airlied: fix do_div error on 32-bit arm, not sure it's 100% correct]

* 'drm-next-4.12' of git://people.freedesktop.org/~agd5f/linux: (58 commits)
  drm/amdgpu: use uintptr_t instead of unsigned long to store pointer
  drm/amdgpu: Avoid using signed integer to store pointer value
  drm/amdgpu:invoke new implemented AI MB func
  drm/amdgpu/vega10:timeout set to equal with VI
  drm/amdgpu:implement the reset MB func for vega10
  drm/amdgpu:fix typo for mxgpu_ai
  drm/amdgpu:no need to involv HDP in KIQ
  drm/amdgpu:add PSP block only load_type=PSP (v2)
  drm/amdgpu/smu9: update to latest driver interface
  drm/amd/amdgpu: cleanup gfx_v9_0_gpu_init()
  drm/amd/amdgpu: cleanup gfx_v9_0_rlc_reset()
  drm/amd/amdgpu: cleanup gfx_v9_0_rlc_start()
  drm/amd/amdgpu: simplify gfx_v9_0_cp_gfx_enable()
  drm/amd/amdgpu: cleanup gfx_v9_0_kiq_init_register()
  drm/amd/amdgpu: Drop gfx_v9_0_print_status()
  drm/amd/amdgpu: cleanup gfx_v9_0_set_gfx_eop_interrupt_state()
  drm/amd/amdgpu: cleanup gfx_v9_0_set_priv_reg_fault_state()
  drm/amd/amdgpu: cleanup gfx_v9_0_set_priv_inst_fault_state()
  drm/amd/amdgpu: cleanup gfx_v9_0_init_queue()
  drm/amdgpu: Move function amdgpu_has_atpx near other similar functions
  ...

Merge branch 'etnaviv/next' of https://git.pengutronix.de/git/lst/linux into drm-next

Highlights:
- Cooling device support from Russell, to allow GPU throttling on system
thermal overload.
- Explicit fencing support from Philipp, implemented in a similar way to
drm/msm.

* 'etnaviv/next' of https://git.pengutronix.de/git/lst/linux:
  drm/etnaviv: submit support for out-fences
  drm/etnaviv: return GPU fence through the submit structure
  drm/etnaviv: submit support for in-fences
  drm/etnaviv: add etnaviv cooling device
  drm/etnaviv: switch to postclose
  drm/etnaviv: add lockdep assert to fence allocation

Merge tag 'imx-drm-next-2017-04-04' of git://git.pengutronix.de/git/pza/linux into drm-next

imx-drm module/dependency changes

- The PRE/PRG drivers added an unwanted DRM dependency to the ipu-v3 driver.
  Remove the dependency by conditionally disabling PRE/PRG support depending
  on CONFIG_DRM.
- Merge the imx-ipuv3-crtc module into the imxdrm module. There is no reason
  anymore for a separation between core drm driver and crtc/plane drivers,
  especially since commit eb8c88808c83 ("drm/imx: add deferred plane
  disabling"), which added a dependency on imx-ipuv3-crtc to the imxdrm
  module.

* tag 'imx-drm-next-2017-04-04' of git://git.pengutronix.de/git/pza/linux:
  drm/imx: merge imx-drm-core and ipuv3-crtc in one module
  gpu: ipu-v3: don't depend on DRM being enabled

Merge branch 'linux-4.12' of git://github.com/skeggsb/linux into drm-next

A bit more for 4.12:
- GP10B support
- GP107 acceleration support

* 'linux-4.12' of git://github.com/skeggsb/linux: (23 commits)
  drm/nouveau/gpio: enable interrupts on cards with 32 gpio lines
  drm/nouveau/gr/gp107: initial support
  drm/nouveau/core: recognise GP10B chipset
  drm/nouveau/platform: support for probing GP10B
  drm/nouveau/platform: make VDD regulator optional
  drm/nouveau/gr: support for GP10B
  drm/nouveau/ibus: add GP10B support
  drm/nouveau/mc: add GP10B support
  drm/nouveau/fb: add GP10B support
  drm/nouveau/fifo: add GP10B support
  drm/nouveau/msgqueue: support for GP10B PMU firmware
  drm/nouveau/secboot: add GP10B support
  drm/nouveau/secboot/gm20b: specify MC base address as argument
  drm/nouveau/secboot: start LS firmware in post-run hook
  drm/nouveau/secboot: let LS post_run hooks return error
  drm/nouveau/secboot: pass instance to LS firmware loaders
  drm/nouveau/secboot: allow to boot multiple falcons
  drm/nouveau/imem/gk20a: Turn instmem lock into mutex
  drm/nouveau: initial support (display-only) for GP107
  drm/nouveau/kms/nv50: fix double dma_fence_put() when destroying plane state
  ...

drm/amdgpu: use uintptr_t instead of unsigned long to store pointer

Signed-off-by: Alex Xie <AlexBin.Xie@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Avoid using signed integer to store pointer value

Signed-off-by: Alex Xie <AlexBin.Xie@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu:invoke new implemented AI MB func

Implement the sr-iov mailbox for soc15 asics.

Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/vega10:timeout set to equal with VI

Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu:implement the reset MB func for vega10

they are lack in the bringup stage, we need them for GPU reset
feature.

Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu:fix typo for mxgpu_ai

Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu:no need to involv HDP in KIQ

1,KIQ won't touch VRAM so no need to involv HDP flush/invalidate at all.
2,According to CP hw designer KIQ better not use any PM4 package lead to wait behave.

Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu:add PSP block only load_type=PSP (v2)

SRIOV currently only can load ucode directly, and PSP
block is not supported by VF temporarily.

will remove this restrict and use PSP load all ucode
even for SRIOV later

v2: squash in check against module parameter

Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/smu9: update to latest driver interface

Signed-off-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/amdgpu: cleanup gfx_v9_0_gpu_init()

Use new WREG32_FIELD15 macro

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/amdgpu: cleanup gfx_v9_0_rlc_reset()

Use new WREG32_FIELD15 macro

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/amdgpu: cleanup gfx_v9_0_rlc_start()

Use new WREG32_FIELD15 macro

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/amdgpu: simplify gfx_v9_0_cp_gfx_enable()

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/amdgpu: cleanup gfx_v9_0_kiq_init_register()

Use new WREG32_FIELD macro

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/amdgpu: Drop gfx_v9_0_print_status()

It's not used in gfx 6/7/8 so drop it from gfx 9 as well.

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/amdgpu: cleanup gfx_v9_0_set_gfx_eop_interrupt_state()

Use new WREG32_FIELD15 macro.

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/amdgpu: cleanup gfx_v9_0_set_priv_reg_fault_state()

Use new WREG32_FIELD15 macro.

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/amdgpu: cleanup gfx_v9_0_set_priv_inst_fault_state()

Use new WREG32_FIELD15 macro.

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/amdgpu: cleanup gfx_v9_0_init_queue()

Introduce WREG32_FIELD15 macro for SOC15 architectures.

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Move function amdgpu_has_atpx near other similar functions

Signed-off-by: Alex Xie <AlexBin.Xie@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: fix over allocating of IRQ sources

We need an array of pointers to IRQ sources, not an array of sources.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/amdgpu: Clean up psp reload_quirk()

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/amdgpu: Fix psp_v3_1 compare sram

Had the wrong sense in the loop

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: cleanup get_invalidate_req v2

The two hubs are just instances of the same hardware,
so the register bits are identical.

v2: keep the function pointer

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: fix vm size and block size for VMPT (v5)

Set reasonable defaults per family.

v2: set both of them in gmc
v3: move vm size and block size in vm manager
v4: squash in warning fix from Alex Xie
v5: squash in min() warning fix

Signed-off-by: Junwei Zhang <Jerry.Zhang@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/amdgpu: Fix srbm_indexing in init/inactive hqd code

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/amdgpu: Clean up gfx_v8_0_mqd_init()

Clean up a toggle with ?:.

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/amdgpu: de-numberify HQD_ACTIVE check.

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/amdgpu: clean up gfx_v8_0_kiq_init_register()

Swap read/write pattern for WREG32_FIELD()

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/amdgpu: Clean up gfx_v8_0_inactive_hqd()

Swap read/write pattern for WREG32_FIELD()

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/amdgpu: Clean up gfx_v8_0_kiq_set_interrupt_state()

Use new WREG32_FIELD_OFFSET() to clean up code.

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/gfx8: KIQ is also disabled when MEC is disabled

Set the ready flag to reflect this.

Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: cleanup VMHUB bit definitions v2

The two hubs are just instances of the same hardware,
so the register bits are identical.

v2: only remove get_vm_protection_bits for now

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: handle CPU access for split VRAM buffers (v2)

This avoids merging them together on page fault.

v2: squash in 64-bit division fix

Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/nouveau/gpio: enable interrupts on cards with 32 gpio lines

The code attempts to enable them, but hits an undefined behaviour by
shifting by the entire register's width:

    int lines = 32;
    u32 mask = (1 << lines) - 1;    // 00000000 on x86
    u32 mask = (1 << lines) - 1;    // ffffffff on arm (32)
    u32 mask = (1 << lines) - 1;    // 00000000 on arm64
    u32 mask = (1ULL << lines) - 1; // ffffffff everywhere

Signed-off-by: Adam Borowski <kilobyte@angband.pl>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/gr/gp107: initial support

Forked from GP106 implementation.

Differences:
- 1 PPC/GPC
- Slightly different grctx magics

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/core: recognise GP10B chipset

Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/platform: support for probing GP10B

Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/platform: make VDD regulator optional

GP10B's power is managed by generic PM domains, so it does not require a
VDD regulator. Add this option into the chip function structure.

Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/gr: support for GP10B

GR is similar to GP100, with a few unavailable registers.

Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/ibus: add GP10B support

GP10B requires a specific initialization sequence due to the absence of
devinit.

Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/mc: add GP10B support

GP10B's MC is compatible with GP100's, but engines need to be explicitly
put out of ELPG during init.

Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/fb: add GP10B support

GP10B's FB is largely compatible with the GP100 implementation.

Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/fifo: add GP10B support

GP10B's FIFO is similar to GP100's, but only allows 512 channels.

Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/msgqueue: support for GP10B PMU firmware

The GP10B firmware is very close to GM20B's. The only difference is that
it supports booting multiple falcons. In order to avoid having too much
functions and structures shared, implement its support in the same
source file as GM20B firmware.

Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/secboot: add GP10B support

GP10B's secboot is largely similar to GM20B's. Only differences are MC
base address and the fact that GPCCS is also securely managed.

Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/secboot/gm20b: specify MC base address as argument

Allow the MC base address to be specified as an argument for the WPR
region reading function. GP10B uses a different address layout as GM20B,
so this is necessary. Also export the function to be used by GP10B.

Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/secboot: start LS firmware in post-run hook

The LS firmware post-run hook is the right place to start said LS
firmware. Moving it here also allows to remove special handling in the
ACR code.

Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/secboot: let LS post_run hooks return error

A LS post-run hook can meet an error meaning the failure of secure boot.
Make sure this can be reported.

Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/secboot: pass instance to LS firmware loaders

Having access to the secboot instance loading a LS firmware can be
useful to LS firmware handlers. At least more useful than just having an
out-of-context subdev pointer.

GP10B's firmware will also need to know the WPR address, which can be
obtained from the secboot instance.

Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/secboot: allow to boot multiple falcons

Change the secboot and msgqueue interfaces to take a mask of falcons to
reset instead of a single falcon. The GP10B firmware interface requires
FECS and GPCCS to be booted in a single firmware command.

For firmwares that only support single falcon boot, it is trivial to
loop over the mask and boot each falcons individually.

Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/imem/gk20a: Turn instmem lock into mutex

The gk20a implementation of instance memory uses vmap()/vunmap() to map
memory regions into the kernel's virtual address space. These functions
may sleep, so protecting them by a spin lock is not safe. This triggers
a warning if the DEBUG_ATOMIC_SLEEP Kconfig option is enabled. Fix this
by using a mutex instead.

Signed-off-by: Thierry Reding <treding@nvidia.com>
Reviewed-by: Alexandre Courbot <acourbot@nvidia.com>
Tested-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau: initial support (display-only) for GP107

Forked from GP106 implementation.

Split out from commit enabling secboot/gr support so that it can be
added to earlier kernels.

Cc: stable@vger.kernel.org [4.10+]
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/kms/nv50: fix double dma_fence_put() when destroying plane state

When the atomic support was added to nouveau, the DRM core did not do this.

However, later in the same merge window, a commit (drm/fence: add in-fences
support) was merged that added it, leading to use-after-frees of the fence
object.

Cc: stable@vger.kernel.org [4.10+]
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/kms/nv50: fix setting of HeadSetRasterVertBlankDmi method

Cc: stable@vger.kernel.org [4.10+]
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/mmu/nv4a: use nv04 mmu rather than the nv44 one

The NV4A (aka NV44A) is an oddity in the family. It only comes in AGP
and PCI varieties, rather than a core PCIE chip with a bridge for
AGP/PCI as necessary. As a result, it appears that the MMU is also
non-functional. For AGP cards, the vast majority of the NV4A lineup,
this worked out since we force AGP cards to use the nv04 mmu. However
for PCI variants, this did not work.

Switching to the NV04 MMU makes it work like a charm. Thanks to mwk for
the suggestion. This should be a no-op for NV4A AGP boards, as they were
using it already.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=70388
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: stable@vger.kernel.org
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/mpeg: mthd returns true on success now

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Fixes: 590801c1a3 ("drm/nouveau/mpeg: remove dependence on namedb/engctx lookup")
Cc: stable@vger.kernel.org # v4.3+
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/amdgpu: use TTM_PL_FLAG_CONTIGUOUS v2

Implement AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS using TTM_PL_FLAG_CONTIGUOUS
instead of a placement limit. That allows us to better handle CPU
accessible placements.

v2: prevent virtual BO start address from overflowing

Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: move adjust_mc_addr into amdgpu_gart_funcs

We should probably rename amdgpu_gart_funcs sooner or later.

Reviewed-by: Junwei Zhang <Jerry.Zhang@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: cleanup logic in amdgpu_vm_flush

Remove some of the extra checks where they don't hurt us.

Reviewed-by: Junwei Zhang <Jerry.Zhang@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: cleanup coding style in amdgpu_vm_flush

Abort early if there is nothing todo and correctly indent the "if"s.

Reviewed-by: Junwei Zhang <Jerry.Zhang@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: coding style of amdgpu_vm_is_gpu_reset

The name is a bit confusing and the extra "? true : false" is superflous.

Additional to that remove setting the reset counter directly after checking it.

Reviewed-by: Junwei Zhang <Jerry.Zhang@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: remove VMID first tracking

Not used any more.

Reviewed-by: Junwei Zhang <Jerry.Zhang@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: move VM related defines into amdgpu_vm.h

Try to clean up amdgpu.h.

Reviewed-by: Junwei Zhang <Jerry.Zhang@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: fix VMHUB order to match the hardware

Match our defines with what the hw uses.

Reviewed-by: Junwei Zhang <Jerry.Zhang@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>