linux/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h, branch v4.20

drm/amdgpu: use a single linked list for amdgpu_vm_bo_base

2018-09-13T20:14:12Z

Instead of the double linked list. Gets the size of amdgpu_vm_pt down to 64 bytes again. We could even reduce it down to 32 bytes, but that would require some rather extreme hacks. Signed-off-by: Christian König Reviewed-by: Chunming Zhou Acked-by: Felix Kuehling Signed-off-by: Alex Deucher

drm/amdgpu: Move fault hash table to amdgpu vm

2018-09-12T21:28:53Z

In stead of share one fault hash table per device, make it per vm. This can avoid inter-process lock issue when fault hash table is full. Change-Id: I5d1281b7c41eddc8e26113e010516557588d3708 Signed-off-by: Oak Zeng Suggested-by: Christian Konig Suggested-by: Felix Kuehling Reviewed-by: Christian Konig Signed-off-by: Alex Deucher

drm/amdgpu: correctly sign extend 48bit addresses v3

2018-09-11T03:41:24Z

Correct sign extend the GMC addresses to 48bit. v2: sign extending turned out easier than thought. v3: clean up the defines and move them into amdgpu_gmc.h as well Signed-off-by: Christian König Reviewed-by: Junwei Zhang Signed-off-by: Alex Deucher

drm/amdgpu: separate per VM BOs from normal in the moved state

2018-09-11T03:40:16Z

Allows us to avoid taking the spinlock in more places. Signed-off-by: Christian König Reviewed-by: Junwei Zhang Signed-off-by: Alex Deucher

drm/amdkfd: Release an acquired process vm

2018-08-29T17:35:00Z

For compute vm acquired from amdgpu, vm.pasid is managed by kfd. Decouple pasid from such vm on process destroy to avoid duplicate pasid release. Signed-off-by: Oak Zeng Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher

drm/amdgpu: Set pasid for compute vm (v2)

2018-08-29T17:34:49Z

To make a amdgpu vm to a compute vm, the old pasid will be freed and replaced with a pasid managed by kfd. Kfd can't reuse original pasid allocated by amdgpu because kfd uses different pasid policy with amdgpu. For example, all graphic devices share one same pasid in a process. v2: rebase (Alex) Signed-off-by: Oak Zeng Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher

drm/amdgpu: remove extra root PD alignment

2018-08-27T20:12:23Z

Just another leftover from radeon. Signed-off-by: Christian König Reviewed-by: Alex Deucher Reviewed-by: Junwei Zhang Acked-by: Huang Rui Signed-off-by: Alex Deucher

drm/amdgpu: Adjust the VM size based on system memory size v2

2018-08-27T20:10:42Z

Set the VM size based on system memory size between the ASIC-specific limits given by min_vm_size and max_bits. GFXv9 GPUs will keep their default VM size of 256TB (48 bit). Only older GPUs will adjust VM size depending on system memory size. This makes more VM space available for ROCm applications on GFXv8 GPUs that want to map all available VRAM and system memory in their SVM address space. v2: * Clarify comment * Round up memory size before >> 30 * Round up automatic vm_size to power of two Signed-off-by: Felix Kuehling Acked-by: Junwei Zhang Reviewed-by: Huang Rui Reviewed-by: Christian König Signed-off-by: Alex Deucher

drm/amdgpu: use bulk moves for efficient VM LRU handling (v6)

2018-08-27T16:11:22Z

I continue to work for bulk moving that based on the proposal by Christian. Background: amdgpu driver will move all PD/PT and PerVM BOs into idle list. Then move all of them on the end of LRU list one by one. Thus, that cause so many BOs moved to the end of the LRU, and impact performance seriously. Then Christian provided a workaround to not move PD/PT BOs on LRU with below patch: Commit 0bbf32026cf5ba41e9922b30e26e1bed1ecd38ae ("drm/amdgpu: band aid validating VM PTs") However, the final solution should bulk move all PD/PT and PerVM BOs on the LRU instead of one by one. Whenever amdgpu_vm_validate_pt_bos() is called and we have BOs which need to be validated we move all BOs together to the end of the LRU without dropping the lock for the LRU. While doing so we note the beginning and end of this block in the LRU list. Now when amdgpu_vm_validate_pt_bos() is called and we don't have anything to do, we don't move every BO one by one, but instead cut the LRU list into pieces so that we bulk move everything to the end in just one operation. Test data: +--------------+-----------------+-----------+---------------------------------------+ | |The Talos |Clpeak(OCL)|BusSpeedReadback(OCL) | | |Principle(Vulkan)| | | +------------------------------------------------------------------------------------+ | | | |0.319 ms(1k) 0.314 ms(2K) 0.308 ms(4K) | | Original | 147.7 FPS | 76.86 us |0.307 ms(8K) 0.310 ms(16K) | +------------------------------------------------------------------------------------+ | Orignial + WA| | |0.254 ms(1K) 0.241 ms(2K) | |(don't move | 162.1 FPS | 42.15 us |0.230 ms(4K) 0.223 ms(8K) 0.204 ms(16K)| |PT BOs on LRU)| | | | +------------------------------------------------------------------------------------+ | Bulk move | 163.1 FPS | 40.52 us |0.244 ms(1K) 0.252 ms(2K) 0.213 ms(4K) | | | | |0.214 ms(8K) 0.225 ms(16K) | +--------------+-----------------+-----------+---------------------------------------+ After test them with above three benchmarks include vulkan and opencl. We can see the visible improvement than original, and even better than original with workaround. v2: move all BOs include idle, relocated, and moved list to the end of LRU and put them together. v3: remove unused parameter and use list_for_each_entry instead of the one with save entry. v4: move the amdgpu_vm_move_to_lru_tail after command submission, at that time, all bo will be back on idle list. v5: remove amdgpu_vm_move_to_lru_tail_by_list(), use bulk_moveable instread of validated, and move ttm_bo_bulk_move_lru_tail() also into amdgpu_vm_move_to_lru_tail(). v6: clean up and fix return value. Signed-off-by: Christian König Signed-off-by: Huang Rui Tested-by: Mike Lothian Tested-by: Dieter Nützel Acked-by: Chunming Zhou Reviewed-by: Junwei Zhang Signed-off-by: Alex Deucher

drm/amdgpu: use new scheduler load balancing for VMs

2018-08-27T16:10:45Z

Instead of the fixed round robin use let the scheduler balance the load of page table updates. Signed-off-by: Christian König Reviewed-by: Chunming Zhou Signed-off-by: Alex Deucher