Currently the GPU MMU is attached in the adreno_gpu code but as
more and more of the GPU initialization moves to the generic
GPU path we have a need to map and use GPU memory earlier and
earlier.  There isn't any reason to defer attaching the MMU
until later so attach it right after the address space is
created so it can be used immediately.
Signed-off-by: NJordan Crouse <jcrouse@codeaurora.org>
Signed-off-by: NRob Clark <robdclark@gmail.com>

Buffer object specific resources like pages, domains, sg list
need not be protected with struct_mutex. They can be protected
with a buffer object level lock. This simplifies locking and
makes it easier to avoid potential recursive locking scenarios
for SVM involving mmap_sem and struct_mutex. This also removes
unnecessary serialization when creating buffer objects, and also
between buffer object creation and GPU command submission.
Signed-off-by: NSushmita Susheelendra <ssusheel@codeaurora.org>
[robclark: squash in handling new locking for shrinker]
Signed-off-by: NRob Clark <robdclark@gmail.com>

Now that the msm_gem supports an arbitrary number of vma's, we no longer
need to assign an id (index) to each address space.  So rip out the
associated code.
Signed-off-by: NRob Clark <robdclark@gmail.com>

No functional change, that will come later.  But this will make it
easier to deal with dynamically created address spaces (ie. per-
process pagetables for gpu).
Signed-off-by: NRob Clark <robdclark@gmail.com>

Most, but not all, paths where calling the with struct_mutex held.  The
fast-path in msm_gem_get_iova() (plus some sub-code-paths that only run
the first time) was masking this issue.

So lets just always hold struct_mutex for hw_init().  And sprinkle some
WARN_ON()'s and might_lock() to avoid this sort of problem in the
future.
Signed-off-by: NRob Clark <robdclark@gmail.com>

The amount of information that we need to pass into msm_gpu_init()
is steadily increasing, so add a new struct to stabilize the function
call and make it easier to add new configuration down the line.
Signed-off-by: NJordan Crouse <jcrouse@codeaurora.org>
Signed-off-by: NRob Clark <robdclark@gmail.com>

Otherwise if someone was using old bindings with "core_clk" instead of
"core" as the clock name, we'd never find it and gpu would be stuck at
27MHz (or whatever it's slowest rate is).

Fixes: 98db803f ("msm/drm: gpu: Dynamically locate the clocks from the device tree")
Signed-off-by: NRob Clark <robdclark@gmail.com>

Instead of using a fixed list of clock names use the clock-names
list in the device tree to discover and get the list of clocks
that we need.
Signed-off-by: NJordan Crouse <jcrouse@codeaurora.org>
Signed-off-by: NRob Clark <robdclark@gmail.com>

Some A3XX and A4XX GPU targets required that the GPU clock be
programmed to a non zero value when it was disabled so
27Mhz was chosen as the "invalid" frequency.

Even though newer targets do not have the same clock restrictions
we still write 27Mhz on clock disable and expect the clock subsystem
to round down to zero.

For unknown reasons even though the slow clock speed is always
27Mhz and it isn't actually a functional level the legacy device tree
frequency tables always defined it and then did gymnastics to work
around it.

Instead of playing the same silly games just hard code the "slow" clock
speed in the code as 27MHz and save ourselves a bit of infrastructure.
Signed-off-by: NJordan Crouse <jcrouse@codeaurora.org>
Signed-off-by: NRob Clark <robdclark@gmail.com>

We should be detaching the MMU before destroying the address
space. To do this cleanly, the detach has to happen in
adreno_gpu_cleanup() because it needs access to structs
in adreno_gpu.c.  Plus it is better symmetry to have
the attach and detach at the same code level.
Signed-off-by: NJordan Crouse <jcrouse@codeaurora.org>
Signed-off-by: NRob Clark <robdclark@gmail.com>

We need to use pm-runtime properly when IOMMU is using device_link() to
control it's own clocks.
Signed-off-by: NRob Clark <robdclark@gmail.com>

We should be detaching the MMU before destroying the address
space. To do this cleanly, the detach has to happen in
adreno_gpu_cleanup() because it needs access to structs
in adreno_gpu.c.  Plus it is better symmetry to have
the attach and detach at the same code level.
Signed-off-by: NJordan Crouse <jcrouse@codeaurora.org>
Signed-off-by: NRob Clark <robdclark@gmail.com>

Suggested by Rob Herring.  We still support the old names for
compatibility with downstream android dt files.

Cc: Rob Herring <robh@kernel.org>
Signed-off-by: NRob Clark <robdclark@gmail.com>
Reviewed-by: NEric Anholt <eric@anholt.net>
Acked-by: NRob Herring <robh@kernel.org>

Add support for the A5XX family of Adreno GPUs.
Signed-off-by: NJordan Crouse <jcrouse@codeaurora.org>
Signed-off-by: NRob Clark <robdclark@gmail.com>

The adreno code inherited a silly workaround from downstream
from the bad old days before decent clock control. grp_clk[0]
(named 'src_clk') doesn't actually exist - it was used as a proxy
for whatever the core clock actually was (usually 'core_clk').

All targets should be able to correctly request 'core_clk' and
get the right thing back so zap the anachronism and directly
use grp_clk[0] to control the clock rate.
Signed-off-by: NJordan Crouse <jcrouse@codeaurora.org>
Signed-off-by: NRob Clark <robdclark@gmail.com>

For a5xx the gpu is 64b so we need to change iova to 64b everywhere.  On
the display side, iova is still 32b so it can ignore the upper bits.
(Although all the armv8 devices have an iommu that can map 64b pa to 32b
iova.)
Signed-off-by: NRob Clark <robdclark@gmail.com>

We can have various combinations of 64b and 32b address space, ie. 64b
CPU but 32b display and gpu, or 64b CPU and GPU but 32b display.  So
best to decouple the device iova's from mmap offset.
Signed-off-by: NRob Clark <robdclark@gmail.com>

I plan to usurp the short name of struct fence for a core kernel struct,
and so I need to rename the specialised fence/timeline for DMA
operations to make room.

A consensus was reached in
https://lists.freedesktop.org/archives/dri-devel/2016-July/113083.html
that making clear this fence applies to DMA operations was a good thing.
Since then the patch has grown a bit as usage increases, so hopefully it
remains a good thing!

(v2...: rebase, rerun spatch)
v3: Compile on msm, spotted a manual fixup that I broke.
v4: Try again for msm, sorry Daniel

coccinelle script:
@@

@@
- struct fence
+ struct dma_fence
@@

@@
- struct fence_ops
+ struct dma_fence_ops
@@

@@
- struct fence_cb
+ struct dma_fence_cb
@@

@@
- struct fence_array
+ struct dma_fence_array
@@

@@
- enum fence_flag_bits
+ enum dma_fence_flag_bits
@@

@@
(
- fence_init
+ dma_fence_init
|
- fence_release
+ dma_fence_release
|
- fence_free
+ dma_fence_free
|
- fence_get
+ dma_fence_get
|
- fence_get_rcu
+ dma_fence_get_rcu
|
- fence_put
+ dma_fence_put
|
- fence_signal
+ dma_fence_signal
|
- fence_signal_locked
+ dma_fence_signal_locked
|
- fence_default_wait
+ dma_fence_default_wait
|
- fence_add_callback
+ dma_fence_add_callback
|
- fence_remove_callback
+ dma_fence_remove_callback
|
- fence_enable_sw_signaling
+ dma_fence_enable_sw_signaling
|
- fence_is_signaled_locked
+ dma_fence_is_signaled_locked
|
- fence_is_signaled
+ dma_fence_is_signaled
|
- fence_is_later
+ dma_fence_is_later
|
- fence_later
+ dma_fence_later
|
- fence_wait_timeout
+ dma_fence_wait_timeout
|
- fence_wait_any_timeout
+ dma_fence_wait_any_timeout
|
- fence_wait
+ dma_fence_wait
|
- fence_context_alloc
+ dma_fence_context_alloc
|
- fence_array_create
+ dma_fence_array_create
|
- to_fence_array
+ to_dma_fence_array
|
- fence_is_array
+ dma_fence_is_array
|
- trace_fence_emit
+ trace_dma_fence_emit
|
- FENCE_TRACE
+ DMA_FENCE_TRACE
|
- FENCE_WARN
+ DMA_FENCE_WARN
|
- FENCE_ERR
+ DMA_FENCE_ERR
)
 (
 ...
 )
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NGustavo Padovan <gustavo.padovan@collabora.co.uk>
Acked-by: NSumit Semwal <sumit.semwal@linaro.org>
Acked-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/20161025120045.28839-1-chris@chris-wilson.co.uk

Prep work for next patch.
Signed-off-by: NRob Clark <robdclark@gmail.com>

Track the pid per submit, so we can print the name of the task which
submitted the batch that caused the gpu to hang.
Signed-off-by: NRob Clark <robdclark@gmail.com>

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

At this point, there is nothing left to fail.  And submit already has a
fence assigned and is added to the submit_list.  Any problems from here
on out are asynchronous (ie. hangcheck/recovery).
Signed-off-by: NRob Clark <robdclark@gmail.com>

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

Better encapsulate the per-timeline stuff into fence-context.  For now
there is just a single fence-context, but eventually we'll also have one
per-CRTC to enable fully explicit fencing.
Signed-off-by: NRob Clark <robdclark@gmail.com>

Since we already track the array of bo's in the submit object, just
unconditionally take and drop ref's per submit (rather than only taking
ref's if bo is not already active).  This simplifies later patches.
Signed-off-by: NRob Clark <robdclark@gmail.com>

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

msm_iommu_new() can fail and this change makes sure that we
detect the failure and free the allocated domain before going
any further.
Signed-off-by: NStephane Viau <sviau@codeaurora.org>
Signed-off-by: NRob Clark <robdclark@gmail.com>

Track the list of in-flight submits.  If the gpu hangs, retire up to an
including the offending submit, and then re-submit the remainder.  This
way, for concurrently running piglit tests (for example), one failing
test doesn't cause unrelated tests to fail simply because it's submit
was queued up after one that triggered a hang.
Signed-off-by: NRob Clark <robdclark@gmail.com>

As found in apq8016 (used in DragonBoard 410c) and msm8916.

Note that numerically a306 is actually 307 (since a305c already claimed
306).  Nice and confusing.
Signed-off-by: NRob Clark <robdclark@gmail.com>

A few spots in the driver have support for downstream android
CONFIG_MSM_BUS_SCALING.  This is mainly to simplify backporting the
driver for various devices which do not have sufficient upstream
kernel support.  But the intentionally dead code seems to cause
some confusion.  Rename the #define to make this more clear.
Signed-off-by: NRob Clark <robdclark@gmail.com>

Somewhere along the way, the firmware loader sprouted another lock
dependency, resulting in possible deadlock scenario:

 &dev->struct_mutex --> &sb->s_type->i_mutex_key#2 --> &mm->mmap_sem

which is problematic vs things like gem mmap.

So introduce a separate mutex to synchronize gpu init.
Signed-off-by: NRob Clark <robdclark@gmail.com>

Downstream kernel IOMMU had a non-standard way of dealing with multiple
devices and multiple ports/contexts.  We don't need that on upstream
kernel, so rip out the crazy.

Note that we have to move the pinning of the ringbuffer to after the
IOMMU is attached.  No idea how that managed to work properly on the
downstream kernel.

For now, I am leaving the IOMMU port name stuff in place, to simplify
things for folks trying to backport latest drm/msm to device kernels.
Once we no longer have to care about pre-DT kernels, we can drop this
and instead backport upstream IOMMU driver.
Signed-off-by: NRob Clark <robdclark@gmail.com>

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

To ease debugging, add debugfs file which can be cat/tail'd to log
submits, along with fence #.  If GPU hangs, you can look at 'gpu'
debugfs file to find last completed fence and current register state,
and compare with logged rd file to narrow down the DRAW_INDX which
triggered the GPU hang.
Signed-off-by: NRob Clark <robdclark@gmail.com>

Shut down the clks when the gpu has nothing to do.  A short inactivity
timer is used to provide a low pass filter for power transitions.
Signed-off-by: NRob Clark <robdclark@gmail.com>

Because we use a list_head in the bo to track it's position in a submit,
we need to serialize at a higher layer.  Otherwise there are problems
when multiple contexts are SUBMIT'ing in parallel cmdstreams referencing
a shared bo.
Signed-off-by: NRob Clark <robdclark@gmail.com>

Add a VRAM carveout that is used for systems which do not have an IOMMU.

The VRAM carveout uses CMA.  The arch code must setup a CMA pool for the
device (preferrably in highmem.. a 256m-512m VRAM pool in lowmem is not
cool).  The user can configure the VRAM pool size using msm.vram module
param.

Technically, the abstraction of IOMMU behind msm_mmu is not strictly
needed, but it simplifies the GEM code a bit, and will be useful later
when I add support for a2xx devices with GPUMMU, so I decided to keep
this part.

It appears to be possible to configure the GPU to restrict access to
addresses within the VRAM pool, but this is not done yet.  So for now
the GPU will refuse to load if there is no sort of mmu.  Once address
based limits are supported and tested to confirm that we aren't giving
the GPU access to arbitrary memory, this restriction can be lifted
Signed-off-by: NRob Clark <robdclark@gmail.com>

This got a bit broken with original patches when re-arranging things to
move dependencies on mach-msm inside #ifndef OF.
Signed-off-by: NRob Clark <robdclark@gmail.com>

Re-arrange things a bit so that we can get work requested after a bo
fence passes, like pageflip, done before retiring bo's.  Without any
sort of bo cache in userspace, some games can trigger hundred's of
transient bo's, which can cause retire to take a long time (5-10ms).
Obviously we want a bo cache.. but this cleanup will make things a
bit easier for atomic as well and makes things a bit cleaner.
Signed-off-by: NRob Clark <robdclark@gmail.com>
Acked-by: NDavid Brown <davidb@codeaurora.org>

The dereference to 'pdata' should be moved below the NULL test.
Signed-off-by: NWei Yongjun <yongjun_wei@trendmicro.com.cn>