This is the big pull request for amdgpu, the new driver for VI+ AMD
asics.  I currently supports Tonga, Iceland, and Carrizo and also
contains a Kconfig option to build support for CI parts for testing.

All major functionality is supported (displays, gfx, compute, dma,
video decode/encode, etc.).  Power management is working on Carrizo,
but is still being worked on for Tonga and Iceland.

* 'drm-next-4.2-amdgpu' of git://people.freedesktop.org/~agd5f/linux: (106 commits)
  drm/amdgpu: only support IBs in the buffer list (v2)
  drm/amdgpu: add vram_type and vram_bit_width for interface query (v2)
  drm/amdgpu: add ib_size/start_alignment interface query
  drm/amdgpu: add me/ce/pfp_feature_version interface query
  drm/amdgpu add ce_ram_size for interface query
  drm/amdgpu add max_memory_clock for interface query (v2)
  drm/amdgpu: add hdp flush for gfx8 compute ring
  drm/amdgpu: fix no hdp flush for compute ring
  drm/amdgpu: add HEVC/H.265 UVD support
  drm/amdgpu: stop loading firmware with pm.mutex locked
  drm/amdgpu: remove mclk_lock
  drm/amdgpu: fix description of vm_size module parameter (v2)
  drm/amdgpu: remove all sh mem register modification in vm flush
  drm/amdgpu: rename GEM_OP_SET_INITIAL_DOMAIN -> GEM_OP_SET_PLACEMENT
  drm/amdgpu: fence should be added to shared slot
  drm/amdgpu: sync fence of clear_invalids (v2)
  drm/amdgpu: max_pde_used usage should be under protect
  drm/amdgpu: fix bug of vm_bo_map (v2)
  drm/amdgpu: implement the allocation range (v3)
  drm/amdgpu: rename amdgpu_ip_funcs to amd_ip_funcs (v2)
  ...

This function could return a NULL pointer in case of handle not
present and in case of out of memory conditions however caller
function always returned EINVAL error hiding a possible ENOMEM.
This patch change the function to return the error instead to
be able to propagate the error instead of assuming EINVAL.
Signed-off-by: NFrediano Ziglio <fziglio@redhat.com>
Reviewed-by: NDave Airlie <airlied@redhat.com>
Signed-off-by: NDave Airlie <airlied@redhat.com>

In qxlhw_handle_to_bo we incremented counters twice, one time for release object
and one for reloc_info.
In the main function however reloc_info references was drop much earlier than
release so keeping the pointer only on release is safe and make cleaning
process easier.
Signed-off-by: NFrediano Ziglio <fziglio@redhat.com>
Reviewed-by: NDave Airlie <airlied@redhat.com>
Signed-off-by: NDave Airlie <airlied@redhat.com>

qxl_bo structure has two reference counters, one in the GEM object and
another in the TTM object. The GEM object keep a counter to the TTM object
so when GEM counter reached zero the TTM counter (using qxl_bo_unref) was
decremented. The qxl object is fully freed (both GEM and TTM part are cleaned)
when the TTM counter reach zero.
One issue was that surface idr structure has no owning on qxl_bo objects however
it contains a pointer to qxl_bo object. This caused some nasty race condition
for instance qxl_bo object was reaped even after counter was already zero.
This patch fix these races moving main counter (the one used by qxl_bo_(un)ref)
to GEM object which cleanup routine (qxl_gem_object_free) remove the idr pointer
(using qxl_surface_evict) when the counters are still valid.
Signed-off-by: NFrediano Ziglio <fziglio@redhat.com>
Reviewed-by: NDave Airlie <airlied@redhat.com>
Signed-off-by: NDave Airlie <airlied@redhat.com>

Enable format string checks for qxl_io_log and remove resulting warnings
which could lead to memory errors on different platform or just printing
wrong information.
Signed-off-by: NFrediano Ziglio <fziglio@redhat.com>
Reviewed-by: NDave Airlie <airlied@redhat.com>
Signed-off-by: NDave Airlie <airlied@redhat.com>

Free resources correctly if function fails
Signed-off-by: NFrediano Ziglio <fziglio@redhat.com>
Reviewed-by: NDave Airlie <airlied@redhat.com>
Signed-off-by: NDave Airlie <airlied@redhat.com>

This function return handle to allocated release object which is an int.
Signed-off-by: NFrediano Ziglio <fziglio@redhat.com>
Reviewed-by: NDave Airlie <airlied@redhat.com>
Signed-off-by: NDave Airlie <airlied@redhat.com>

Only EBUSY error was handled. This could cause code to believe
reserve was successful while it failed.
Signed-off-by: NFrediano Ziglio <fziglio@redhat.com>
Reviewed-by: NDave Airlie <airlied@redhat.com>
Signed-off-by: NDave Airlie <airlied@redhat.com>

Is we are not able to get source bo object from handle we free
destination bo object and call cleanup code however destination
object was already inserted in reloc_info array (num_relocs was
already incremented) so on cleanup we free destination again.
Signed-off-by: NFrediano Ziglio <fziglio@redhat.com>
Reviewed-by: NDave Airlie <airlied@redhat.com>
Signed-off-by: NDave Airlie <airlied@redhat.com>

reloc_info[i] is not still initialized in the print statement.
Signed-off-by: NFrediano Ziglio <fziglio@redhat.com>
Reviewed-by: NDave Airlie <airlied@redhat.com>
Signed-off-by: NDave Airlie <airlied@redhat.com>

If the function fails reference counter to the object is not decremented
causing leaks.
This is hard to spot as it happens only on very low memory situations.
Signed-off-by: NFrediano Ziglio <fziglio@redhat.com>
Cc: stable@vger.kernel.org
Reviewed-by: NDave Airlie <airlied@redhat.com>
Signed-off-by: NDave Airlie <airlied@redhat.com>

If objects are moved back from system memory to VRAM (and spice id
created again) memory is already initialized so we need to set flag
to not clear memory.
If you don't do it after a while using desktop many images turns to
black or transparents.
Signed-off-by: NFrediano Ziglio <fziglio@redhat.com>
Cc: stable@vger.kernel.org
Reviewed-by: NDave Airlie <airlied@redhat.com>
Signed-off-by: NDave Airlie <airlied@redhat.com>

amdgpu_cs_find_mapping doesn't work without all buffers being validated,
so the TTM validation must be done first.

v2: only use amdgpu_cs_find_mapping for UVD/VCE VM emulation
Signed-off-by: NMarek Olšák <marek.olsak@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>

Track the type of vram on the board and provide a query for it.
User mode drivers and tools want this information for determining
bandwidth information and form informational purposes.

v2: fix build when CI support is not enabled
Signed-off-by: NKen Wang <Qingqing.Wang@amd.com>
Reviewed-by: NJammy Zhou <Jammy.Zhou@amd.com>

Query the IB alignment requirements from the kernel rather
than hardcoding them in the user mode drivers.
Signed-off-by: NKen Wang <Qingqing.Wang@amd.com>
Reviewed-by: NJammy Zhou <Jammy.Zhou@amd.com>

Provide this information to usermode drivers.  We were previously
missing this info.
Signed-off-by: NKen Wang <Qingqing.Wang@amd.com>
Reviewed-by: NJammy Zhou <Jammy.Zhou@amd.com>

Add a query for the CE ram size.  User mode drivers
will want to use this to determine how much size
of the cache on the CE.
Signed-off-by: NKen Wang <Qingqing.Wang@amd.com>
Reviewd-by: NJammy Zhou <Jammy.Zhou@amd.com>

Add a query for the max memory clock.

v2: handle the dpm enabled case properly
Signed-off-by: NKen Wang <Qingqing.Wang@amd.com>
Reviewd-by: NJammy Zhou <Jammy.Zhou@amd.com>

We had forgotten to register the callback.
Signed-off-by: Nmonk.liu <monk.liu@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: Njammy zhou <jammy.zhou@amd.com>

No pfp on compute.
Signed-off-by: Nmonk.liu <monk.liu@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>

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

Loading firmware is a rather complicated process, in the end we add a
dependency between the pm mutex and the mm semaphore which results in
a harmless but annoying error message.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Reviewed-by: NJammy Zhou <Jammy.Zhou@amd.com>

Not needed any more.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>

default is 8GB, not 4GB.

v2: fix fallback setting when the user provides an invalid input
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Leave that at the values set during init.  No need to update
them repeatedly.
Signed-off-by: Nmonk.liu <monk.liu@amd.com>
Signed-off-by: NDavid Zhang <david1.zhang@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NJammy Zhou <jammy.zhou@amd.com>

Signed-off-by: NMarek Olšák <marek.olsak@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>

Signed-off-by: Nmonk.liu <monk.liu@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: Njammy zhou <jammy.zhou@amd.com>

bo_va may un-initialized, fix it.
Signed-off-by: Nmonk.liu <monk.liu@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NJammy Zhou <jammy.zhou@amd.com>

Need to take the lock when accessing this.
Signed-off-by: Nmonk.liu <monk.liu@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NJammy Zhou <jammy.zhou@amd.com>

call reservation_object_reserve_shared before amdgpu_bo_fence
Signed-off-by: Nmonk.liu <monk.liu@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NJammy Zhou <jammy.zhou@amd.com>

Pass a ttm_placement pointer to amdgpu_bo_create_restricted
add min_offset to amdgpu_bo_pin_restricted.  This makes it
easier to allocate memory with address restrictions.  With
this patch we can also enable 2-ended allocation again.

v2: fix rebase conflicts
v3: memset placements before using
Reviewed-by: NJammy Zhou <Jammy.Zhou@amd.com>
Signed-off-by: NChunming Zhou <david1.zhou@amd.com>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>

The structure is renamed and moved to amd_shared.h to make
the component independent.  This makes it easier to add
new components in the future.

v2: fix include path
Reviewed-by: NJammy Zhou <Jammy.Zhou@amd.com>
Signed-off-by: Nyanyang1 <young.yang@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Not needed any more.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Reviewed-by: NMonk Liu <monk.liu@amd.com>

It's causing issues with VMID handling and comparing the
fence value two times actually doesn't make handling faster.

Port of radeon commit "d6d5c5b8".
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Reviewed-by: NMonk Liu <monk.liu@amd.com>

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

Enable UVD dpm (dynamic power management) and powergating.  UVD dpm dynamically scales the UVD
clocks on demand.  Powergating turns off the power to the block when it's not in use.
Signed-off-by: NSonny Jiang <sonny.jiang@amd.com>
Acked-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>

VCE 3.0 has two indentical instances in the engine, they share
the same registers name in differrent memory block distinguished
by the grbm_gfx_index, we set to master instance after init, it
will dispatch task to slave instance. These two instances will
share the same firmware, but have their own stacks and heaps.

v2: add mutex for using grbm_gfx_index
Signed-off-by: NLeo Liu <leo.liu@amd.com>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>

Firmware required BO size changes in terms of ASIC family
Signed-off-by: NLeo Liu <leo.liu@amd.com>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>

Port of 77cb2fea to
amdgpu.
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Signed-off-by: NMarek Olšák <marek.olsak@amd.com>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Acked-by: NChristian König <christian.koenig@amd.com>