to #25447038

commit 6ceeb144b1d6952a36afa6c29718beac575f2a3f upstream.

When a page tables needs to be evicted the VM code should
decide if that is possible or not.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NFelix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: Nmfkang <mfkang@linux.alibaba.com>
Reviewed-by: Nluanshi <zhangliguang@linux.alibaba.com>

to #25447038

commit 1bd4e4ca7bb8f681ff4e2b05c97ce975ccd781d6 upstream.

Otherwise we won't be able to cleanly handle page faults.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NChunming Zhou <david1.zhou@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: Nmfkang <mfkang@linux.alibaba.com>
Reviewed-by: Nluanshi <zhangliguang@linux.alibaba.com>

This patch cleans up spaces and align the text to refine the comment for
amdgpu_ttm.
Signed-off-by: NHuang Rui <ray.huang@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

pass the evict flag instead of hard code
Signed-off-by: NJunwei Zhang <Jerry.Zhang@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

entity has a scheduler field and we don't need the sched argument
in any of the functions where entity is provided.
Signed-off-by: NNayan Deshmukh <nayan26deshmukh@gmail.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NEric Anholt <eric@anholt.net>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Just rename functions, no functional change.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Problem:
amdgpu_ttm_set_buffer_funcs_status destroys adev->mman.entity on suspend
without releasing adev->mman.bdev.man[TTM_PL_VRAM].move fence
so on resume the new drm_sched_entity.fence_context causes
the warning against the old fence context which is different.

Fix:
When destroying sched_entity in amdgpu_ttm_set_buffer_funcs_status
release  man->move and set the pointer to NULL.
Signed-off-by: NAndrey Grodzovsky <andrey.grodzovsky@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Make sure that we properly initialize at least the sched member.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NJunwei Zhang <Jerry.Zhang@amd.com>
Acked-by: NChunming Zhou <david1.zhou@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

We know the ring through the entity anyway.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NJunwei Zhang <Jerry.Zhang@amd.com>
Acked-by: NChunming Zhou <david1.zhou@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

replace run queue by a list of run queues and remove the
sched arg as that is part of run queue itself
Signed-off-by: NNayan Deshmukh <nayan26deshmukh@gmail.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Acked-by: NEric Anholt <eric@anholt.net>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

We are initializing the entity before the scheduler is actually
initialized.

This can lead to all kind of problem, but especially NULL pointer deref
because of Nayan's scheduler work.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Acked-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

It could be got by amdgpu_bo_gpu_offset() if need
Signed-off-by: NJunwei Zhang <Jerry.Zhang@amd.com>
Reviewed-by: NMichel Dänzer <michel.daenzer@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Everything in the flush code path (i.e. waiting for SW queue
to become empty) names with *_flush()
and everything in the release code path names *_fini()

This patch also effect the amdgpu and etnaviv drivers which
use those functions.

v2:
Also pplay the change to vd3.
Signed-off-by: NAndrey Grodzovsky <andrey.grodzovsky@amd.com>
Suggested-by: NChristian König <christian.koenig@amd.com>
Acked-by: NLucas Stach <l.stach@pengutronix.de>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Move all instnaces of this check into a function in amdgpu_gmc.h
Rename the original function to a more proper name.

v2:
Add more places to cleanup.
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAndrey Grodzovsky <andrey.grodzovsky@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

NFC just comments.

(v2):  Updated based on feedback from Alex Deucher.
Signed-off-by: NTom St Denis <tom.stdenis@amd.com>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

this patch also effect the amdgpu and etnaviv drivers which
use the function drm_sched_entity_init
Signed-off-by: NNayan Deshmukh <nayan26deshmukh@gmail.com>
Suggested-by: NChristian König <christian.koenig@amd.com>
Acked-by: NLucas Stach <l.stach@pengutronix.de>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

This change prepares for a workaround in amdkfd for a GFX9 HW bug. It
requires the control stack memory of compute queues, which is allocated
from the second page of MQD gart BOs, to have mtype NC, rather than
the default UC.
Signed-off-by: NYong Zhao <yong.zhao@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

After that, we can easily add new parameter when need.

v2:
a) rebase.
b) Initialize struct amdgpu_bo_param, future new
member could only be used in some one case, but all member
should have its own initial value.
Signed-off-by: NChunming Zhou <david1.zhou@amd.com>
Reviewed-by: Huang Rui <ray.huang@amd.com> (v1)
Reviewed-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NJunwei Zhang <Jerry.Zhang@amd.com>
Cc: christian.koenig@amd.com
Cc: Felix.Kuehling@amd.com
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Reserved VRAM is used to avoid overriding pre OS FB.
Once our display stack takes over we don't need the reserved
VRAM anymore.

v2:
Remove comment, we know actually why we need to reserve the stolen VRAM.
Fix return type for amdgpu_ttm_late_init.
v3:
Return 0 in amdgpu_bo_late_init, rebase on changes to previous patch
v4: rebase
v5:
For GMC9 reserve always just 9M and keep the stolem memory around
until GART table curruption on S3 resume is resolved.
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NAndrey Grodzovsky <andrey.grodzovsky@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Steal 9 MB for vga emulation and fb if vga is enabled, otherwise,
steal enough to cover the current display size as set by the vbios.

If no memory is used (e.g., secondary or headless card), skip
stolen memory reserve.

v2: skip reservation if vram is limited, address Christian's comments
v3: squash in fix from Harry

Reviewed-and-Tested-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com> (v2)
Signed-off-by: NAlex Deucher <alexdeucher@gmail.com>

The detection if a BO was placed in CPU visible VRAM was incorrect.

Fix it and merge it with the correct detection in amdgpu_ttm.c
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NMichel Dänzer <michel.daenzer@amd.com>
Reviewed-by: NChunming Zhou <david1.zhou@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

This commit allows amdgpu_ttm_tt_get_user_pages to work in a worker
thread rather than regular process context. This will be used when
KFD userptr BOs are restored after an MMU-notifier eviction.

v2: Manage task reference with get_task_struct/put_task_struct
Signed-off-by: NFelix Kuehling <Felix.Kuehling@amd.com>
Acked-by: NOded Gabbay <oded.gabbay@gmail.com>
Acked-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NOded Gabbay <oded.gabbay@gmail.com>

The offset inside the page wasn't included in the copy call meaning
the start of the page was being read/written instead.
Reported-by: NJay Cornwall <Jay.Cornwall@amd.com>
Signed-off-by: NTom St Denis <tom.stdenis@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Drop the "kernel" and sg parameter and give the BO type to create
explicit to amdgpu_bo_create instead of figuring it out from the
parameters.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NRoger He <Hongbo.He@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Instead of moving this to the SYSTEM domain just drop the backing store
and let the resulting allocation be freed.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Acked-by: NRoger He <Hongbo.He@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Instead of calculating the size in bytes just to recalculate the number
of pages from it pass the BO directly to the function.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NRoger He <Hongbo.He@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

We don't need the page array for prime shared BOs, stop allocating it.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NRoger He <Hongbo.He@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

This allows access to pages allocated through the driver with optional
IOMMU mapping.

v2: Fix number of bytes copied and add write method
v3: drop check for kmap return
Original-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NTom St Denis <tom.stdenis@amd.com>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

We can use ttm->bdev instead.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NAlex Deucher <alexdeucher@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

The ring status can change during GPU reset, but we still need to be
able to schedule TTM buffer moves in the meantime.

Otherwise we can ran into problems because of aborted move/fill
operations during GPU resets.

v2: still check if ring is available during direct submit.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Acked-by: NChunming zhou <david1.zhou@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

When we reset the GPU we also disable/enable the SDMA, but we don't want
to change TTM idea of the VRAM size in the middle of that.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Acked-by: NChunming zhou <david1.zhou@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Instead of setting the active VRAM size directly provide a the info if
we can use the buffer functions or not.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Acked-by: NChunming zhou <david1.zhou@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

When using CPU to update page table, we need to kmap all the PDs/PTs after
they are allocated and that requires a TLB shot down on each CPU, which is
quite heavy.

Instead, we map the whole visible VRAM to a kernel address at once. Pages
can be obtained from the offset.

v2: move the mapping base from gmc to amdgpu_mman structure, and the
    implementation in amdgpu_ttm_* functions
Signed-off-by: NAmber Lin <Amber.Lin@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Only used by the AGP backend and there it can be easily accessed using
ttm->bdev->glob.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NMichel Dänzer <michel.daenzer@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

The subsystem should check that, not the driver.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NRoger He <Hongbo.He@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

This reverts commit 7bdc53f9 and commit
330df03b.

Neither are needed any more.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NFelix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

This reverts commit 2046d46d.

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

Avoid OOM on syatem pages allocations.

v2:
Remove modeprobe parameter, make this behaviour the only option.

v3:
Move setting no_retry flag into amdgpu_ttm_init.
Signed-off-by: NAndrey Grodzovsky <andrey.grodzovsky@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NFelix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

And rename it to amdgpu_gmc as well.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NSamuel Li <Samuel.Li@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

get the max io mapping address of system memory to see if it is over
our card accessing range.
v2: move checking later
Signed-off-by: NChunming Zhou <david1.zhou@amd.com>
Reviewed-by: NMonk Liu <monk.liu@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180209024410.1469-2-david1.zhou@amd.com