Should help avoid fragmentation of vram due to CPU access
requirements.
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
For kernel driver BOs, be explicit about whether we need
vram access up front. This avoids unecessary migrations and
avoids using visible vram for buffers were it's not needed.
v2: line wrap fixes
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Customize the insert_nop func for SDMA rings, and use burst NOP for
ring/IB submissions in other places as well
Signed-off-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
The insert_nop function is added to amdgpu_ring_funcs structure as
well as the default implementation
Signed-off-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
This function is added to map the ring to sdma instance
Signed-off-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
The burst NOP is supported for SDMA when feature_version is >= 20.
Signed-off-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
This is added to support the burst NOP
v2: squash the typo fix
Signed-off-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Instead of the array which is used for ID management.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
No copy_(to|from)_user while BO is reserved.
v2: handle default path as well
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
That isn't used any more.
v2: rebase
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
sync_resv is to handle both amdgpu_fence and sched_fence.
Signed-off-by: Chunming Zhou <david1.zhou@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Christian K?nig <christian.koenig@amd.com>
AUX addresses are 20 bits long. Send out the entire address instead of
just the low 16 bits.
Port of:
drm/radeon/atom: Send out the full AUX address
to amdgpu
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
Saving the current UVD state on suspend and restoring it on resume
just doesn't work reliable. Just close cleanup all sessions on suspend.
Ported from radeon commit "12e49feadff6d7b7ebbe852b36943a71524d8d34".
v2: rebased
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com> (v1)
This way the scheduler doesn't wait in it's work thread any more.
v2: fix race conditions
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
The context needs to finish before everything else.
Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
This aids handling buffers moves with the scheduler.
Signed-off-by: Chunming Zhou <david1.zhou@amd.com>
Reviewed-by: Christian K?nig <christian.koenig@amd.com>
The fence in the array may be skipped if wait_all is false,
thus the related callback is not initialized with list head.
So removing this kind callback will cause NULL pointer reference.
Signed-off-by: Junwei Zhang <Jerry.Zhang@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
when eviction is happening, if don't handle
dependency, then the fence could be dead off.
Signed-off-by: Chunming Zhou <david1.zhou@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Christian K?nig <christian.koenig@amd.com>
Entity don't live as long as scheduler fences.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Calling schedule() is probably the worse things we can do.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Chunming Zhou <david1.zhou@amd.com>
Reviewed-by: Christian K?nig <christian.koenig@amd.com>
Simply not used any more. Only keep 32bit atomic for fence sequence numbering.
v2: trivial rebase
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com> (v1)
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com> (v1)
Reviewed-by: Chunming Zhou <david1.zhou@amd.com> (v1)
Rename the function and update the related code with this modified function.
Add the new parameter of bool wait_all.
If wait_all is true, it will return when all fences are signaled or timeout.
If wait_all is false, it will return when any fence is signaled or timeout.
Signed-off-by: Junwei Zhang <Jerry.Zhang@amd.com>
Reviewed-by: Monk Liu <monk.liu@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Use pci_alloc_consistent rather than kzalloc since we
need 256 byte aligned memory for the ring buffer.
v2: fix copy paste typo in free function noticed
by Jammy.
bug:
https://bugs.freedesktop.org/show_bug.cgi?id=91749
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
This is a port of:
DRM - radeon: Don't link train DisplayPort on HPD until we get the dpcd
to amdgpu.
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
Stop double freeing the the BO list by pulling the content
of amdgpu_cs_parser_prepare_job() into the IOCTL function again.
v2: better commit message
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com> (v1)
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
The problem now is that we don't necessarily call amdgpu_ib_get()
in some error paths and so work with uninitialized data.
Better require that the memory is already zeroed.
v2: better commit message
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com> (v1)
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Used by mesa, etc. for profiling.
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Pagetables can be moved and therefore the page directory update can be necessary
for the current cs even if none of the the bo's are moved. In that scenario
there is no fence between the sdma0 and gfx ring, so we add one.
v2 (chk): rebased
Signed-off-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Rework run queue implementation, especially remove the odd list handling.
v2: cleanup the code only, no algorithem change.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
fix the bug that there is duplicated bo_update_mapping issued
Signed-off-by: monk.liu <monk.liu@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
bo_list_clone() will take a lot of time when bo_list hold too much
elements, like above 7000
Signed-off-by: Monk.Liu <monk.liu@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Reviewed-by: Jammy Zhou <jammy.zhou@amd.com>
It's not validated yet and causes more harm than good.
Avoids spurious resets.
Reviewed-by: Leo Liu <leo.liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
remaining timeout returned by amdgpu_fence_wait_any can be larger than
max int value, thus the truncated 32 bit value in r ends up being
negative while its original long value is positive.
Signed-off-by: monk.liu <monk.liu@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Jammy Zhou <jammy.zhou@amd.com>
fix fence is released when pass to **fence sometimes.
add reference for it.
Signed-off-by: Chunming Zhou <david1.zhou@amd.com>
Reviewed-by: Christian K?nig <christian.koenig@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Tested-and-Reviewed-by: Leo Liu <leo.liu@amd.com>
Not used any more.
v2: remove amd_sched_emit as well.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Similar to radeon, except that amdgpu doesn't even use struct_mutex to
protect anything like the shared z buffer (sane gpu architecture,
yay!). And the code already grabs the globa adev->ring_lock, so this
code can't race with itself. Which makes struct_mutex completely
redundnant. Remove it.
Cc: Alex Deucher <alexander.deucher@amd.com>
Cc: "Christian König" <christian.koenig@amd.com>
Reviewed-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
It really doesn't protect anything which doesn't have other locks
already. Also this is run from driver unload code so not much need for
locks anyway.
Same changes as for radeon really.
Cc: Alex Deucher <alexander.deucher@amd.com>
Cc: "Christian König" <christian.koenig@amd.com>
Reviewed-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The workaround simply doesn't work because VM mappings
are controlled by userspace not the kernel.
Additional to that this is just a performance problem
which happens if you have holes in your VM mapping.
v2: adjust virtual addr alignment as well.
v3: fix trivial warning
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Monk Liu <monk.liu@amd.com> (v1)
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com> (v2)
Looks like that somehow got missed while during porting the radeon changes.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
It was just a wrapper for fence_wait anyway.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
The common kernel function does the same thing.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
scheduler fence is based on kernel fence framework.
v2: squash in Christian's build fix
Signed-off-by: Chunming Zhou <david1.zhou@amd.com>
Reviewed-by: Christian K?nig <christian.koenig@amd.com>
v2: rebased
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com> (v1)
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Avoiding a couple of casts.
v2: rename c_entity to entity as well
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Cleanup the kernel context handling.
v2: rebased
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com> (v1)
Id's are for the IOCTL ABI only.
v2: remove tgid as well
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
We didn't initialized the mutex in the cloned bo list resulting in nice
warnings from lockdep. Also fixes error handling in this function.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
This way can avoid interrupt lost, and can process sched job exactly.
Signed-off-by: Chunming Zhou <david1.zhou@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
This function is used to get the next queued sequence number
Signed-off-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
This function is to update last_emitted_v_seq and wake up the waiters.
It should be called by driver in the run_job backend function
Signed-off-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
It is clean to update last_queued_v_seq in the scheduler module
Signed-off-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Fix the code alignment, etc.
v2: rebase the code
Signed-off-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
every sbumission should be able to get a fence.
Signed-off-by: Chunming Zhou <david1.zhou@amd.com>
Reviewed-by: Christian K?nig <christian.koenig@amd.com>
Reviewed-by: Jammy Zhou <jammy.zhou@amd.com>
It is theoretically possible that a swapped out BO gets the
same GTT address, but different backing pages while being swapped in.
Instead just use another VA state to note updated areas.
Ported from not upstream yet radeon commit with the same name.
v2: fix some bugs in the original implementation found in the radeon code.
v3: squash in VCE/UVD fix
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
thus unnecessary wake up could be avoid between rings
v2:
move wait_queue_head to fence_drv from ring
Signed-off-by: monk.liu <monk.liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
origninal method will sleep/schedule at the granurarity of HZ/2 and
based on seq signal method, the new implement is based on kernel fance
interface, no unnecessary schedule at all
v2: replace logic of original amdgpu_fence_wait_any
Signed-off-by: monk.liu <monk.liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
