Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

For support EGL_ANDROID_native_fence_sync.

Submit accept dependency from other tasks. Dependency
can be lima native fence or sync file fd. Explicit
sync can be enabled by the submit flags.

Signed-off-by: Qiang Yu <yuq825@gmail.com>

No need to wait for each buffer object idle
if we fini ctx mgr before free them.

va unmap case also don't need to wait because
UMD is responsible for it.

Signed-off-by: Qiang Yu <yuq825@gmail.com>

This reverts commit 45c3d213.

lima driver need preclose to wait all task in the context
created within closing file to finish before free all the
buffer object. Otherwise pending tesk may fail and get
noisy MMU fault message.

Move this wait to each buffer object free function can
achieve the same result but some buffer object is shared
with other file context, but we only want to wait the
closing file context's tasks. So the implementation is
not that straight forword compared to the preclose one.

Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

Reg definition may change from time to time as
this is a reverse engineering project.

Signed-off-by: Qiang Yu <yuq825@gmail.com>

lima_sched_context_add_fence may call dma_fence_wait
and go to sleep.

Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

On mali400, GP and PP pipe may access the same
l2 cache at the same time.

Signed-off-by: Qiang Yu <yuq825@gmail.com>

Prepare for full mali450 support.

Signed-off-by: Qiang Yu <yuq825@gmail.com>

Alban Browaeys authored Apr 12, 2018 and Qiang Yu committed May 08, 2018

Try to preallocate the memory with idr_prealloc(GFP_KERNEL)
and change the allocation flags under spin lock.

Based on commit 2c8468dc
("net: sched: don't use GFP_KERNEL under spin lock")

Signed-off-by: Alban Browaeys <alban.browaeys@gmail.com>

Andrei Paulau authored Apr 08, 2018 and Qiang Yu committed May 08, 2018

Meson mali GPU operate in high clock frequency, need
this value be high to work in a stable state.

Signed-off-by: Andrei Paulau <7134956@gmail.com>

This error is due to assigned long return value
to int which may cause overflow in 64bit arch.

Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

When job error, task->fence is un-init, so job free should
not call dma_fence_put on it.

Now we just move fence alloc into job run, so task->fence
is NULL when job error, and job free dma_fence_put has no
effect on it.

Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

Meson mali GPU operate in high clock frequency, need
this value be high to work in a stable state.

Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

Vasily Khoruzhick authored Nov 23, 2017 and Qiang Yu committed May 08, 2018

All users of inteval tree are required to use rb_root_cached
since f808c13f. Switch to rb_root_cached to fix lima compilation
with 4.15

Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

Heiko Stuebner authored Mar 12, 2018 and Qiang Yu committed May 08, 2018

Without that, at least on arm64 te build fails with
../drivers/gpu/drm/lima/lima_gem.c: In function ‘lima_gem_create_handle’:
../drivers/gpu/drm/lima/lima_gem.c:92:2: error: implicit declaration of function ‘mapping_set_gfp_mask’; did you mean ‘pci_set_dma_mask’? [-Werror=implicit-function-declaration]
  mapping_set_gfp_mask(bo->gem.filp->f_mapping, mask);
  ^~~~~~~~~~~~~~~~~~~~

Signed-off-by: Heiko Stuebner <heiko@sntech.de>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

This field indicates how many tasks before current
submitted task is done which is used for mesa to
free finished task resources.

Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

PP do error handling when all processors done.
Add missing MMU resume.

Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>