korg/linux
0
0
Fork 0
mirror of https://mirrors.bfsu.edu.cn/git/linux.git synced 2024-12-24 03:24:55 +08:00
Code Issues Packages Projects Releases Wiki Activity

drm/amdgpu: revert "drm/amdgpu: use AMDGPU_GEM_CREATE_VRAM_CLEARED for VM PD/PTs" v2

Browse Source 
Using the standard clear turned out to be to inflexible.

First of all it is executed on the system queue, together with buffer
moves instead on the per VM queue.

And second we need to fill in the page tables with more than just zero.

We keep the new functionality of initializing the PDEs/PTEs with ATC
routing entries intact.

v2: update commit message.

Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This commit is contained in:
Christian König 2018-01-24 17:19:04 +01:00 committed by Alex Deucher
parent 4383736340
commit 13307f7e1d
1 changed files with 92 additions and 24 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

116
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
View File

@ -257,6 +257,74 @@ bool amdgpu_vm_ready(struct amdgpu_vm *vm)
return ready;
}
/**
* amdgpu_vm_clear_bo - initially clear the PDs/PTs
*
* @adev: amdgpu_device pointer
* @bo: BO to clear
* @level: level this BO is at
*
* Root PD needs to be reserved when calling this.
*/
static int amdgpu_vm_clear_bo(struct amdgpu_device *adev,
struct amdgpu_vm *vm,
struct amdgpu_bo *bo,
unsigned level)
{
struct ttm_operation_ctx ctx = { true, false };
struct dma_fence *fence = NULL;
uint64_t addr, init_value;
struct amdgpu_ring *ring;
struct amdgpu_job *job;
unsigned entries;
int r;
if (vm->pte_support_ats) {
init_value = AMDGPU_PTE_DEFAULT_ATC;
if (level != AMDGPU_VM_PTB)
init_value |= AMDGPU_PDE_PTE;
} else {
init_value = 0;
}
ring = container_of(vm->entity.sched, struct amdgpu_ring, sched);
r = reservation_object_reserve_shared(bo->tbo.resv);
if (r)
return r;
r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
if (r)
goto error;
addr = amdgpu_bo_gpu_offset(bo);
entries = amdgpu_bo_size(bo) / 8;
r = amdgpu_job_alloc_with_ib(adev, 64, &job);
if (r)
goto error;
amdgpu_vm_set_pte_pde(adev, &job->ibs[0], addr, 0,
entries, 0, init_value);
amdgpu_ring_pad_ib(ring, &job->ibs[0]);
WARN_ON(job->ibs[0].length_dw > 64);
r = amdgpu_job_submit(job, ring, &vm->entity,
AMDGPU_FENCE_OWNER_UNDEFINED, &fence);
if (r)
goto error_free;
amdgpu_bo_fence(bo, fence, true);
dma_fence_put(fence);
return 0;
error_free:
amdgpu_job_free(job);
error:
return r;
}
/**
* amdgpu_vm_alloc_levels - allocate the PD/PT levels
*
@ -275,9 +343,8 @@ static int amdgpu_vm_alloc_levels(struct amdgpu_device *adev,
{
unsigned shift = amdgpu_vm_level_shift(adev, level);
unsigned pt_idx, from, to;
int r;
u64 flags;
uint64_t init_value = 0;
int r;
if (!parent->entries) {
unsigned num_entries = amdgpu_vm_num_entries(adev, level);
@ -300,21 +367,13 @@ static int amdgpu_vm_alloc_levels(struct amdgpu_device *adev,
saddr = saddr & ((1 << shift) - 1);
eaddr = eaddr & ((1 << shift) - 1);
flags = AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS |
AMDGPU_GEM_CREATE_VRAM_CLEARED;
flags = AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
if (vm->use_cpu_for_update)
flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
else
flags |= (AMDGPU_GEM_CREATE_NO_CPU_ACCESS |
AMDGPU_GEM_CREATE_SHADOW);
if (vm->pte_support_ats) {
init_value = AMDGPU_PTE_DEFAULT_ATC;
if (level != AMDGPU_VM_PTB)
init_value |= AMDGPU_PDE_PTE;
}
/* walk over the address space and allocate the page tables */
for (pt_idx = from; pt_idx <= to; ++pt_idx) {
struct reservation_object *resv = vm->root.base.bo->tbo.resv;
@ -325,12 +384,17 @@ static int amdgpu_vm_alloc_levels(struct amdgpu_device *adev,
r = amdgpu_bo_create(adev,
amdgpu_vm_bo_size(adev, level),
AMDGPU_GPU_PAGE_SIZE, true,
AMDGPU_GEM_DOMAIN_VRAM,
flags,
NULL, resv, init_value, &pt);
AMDGPU_GEM_DOMAIN_VRAM, flags,
NULL, resv, 0, &pt);
if (r)
return r;
r = amdgpu_vm_clear_bo(adev, vm, pt, level);
if (r) {
amdgpu_bo_unref(&pt);
return r;
}
if (vm->use_cpu_for_update) {
r = amdgpu_bo_kmap(pt, NULL);
if (r) {
@ -2241,11 +2305,11 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
{
const unsigned align = min(AMDGPU_VM_PTB_ALIGN_SIZE,
AMDGPU_VM_PTE_COUNT(adev) * 8);
uint64_t init_pde_value = 0, flags;
unsigned ring_instance;
struct amdgpu_ring *ring;
struct drm_sched_rq *rq;
unsigned long size;
uint64_t flags;
int r, i;
vm->va = RB_ROOT_CACHED;
@ -2274,23 +2338,19 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
vm->use_cpu_for_update = !!(adev->vm_manager.vm_update_mode &
AMDGPU_VM_USE_CPU_FOR_COMPUTE);
if (adev->asic_type == CHIP_RAVEN) {
if (adev->asic_type == CHIP_RAVEN)
vm->pte_support_ats = true;
init_pde_value = AMDGPU_PTE_DEFAULT_ATC
| AMDGPU_PDE_PTE;
}
} else
} else {
vm->use_cpu_for_update = !!(adev->vm_manager.vm_update_mode &
AMDGPU_VM_USE_CPU_FOR_GFX);
}
DRM_DEBUG_DRIVER("VM update mode is %s\n",
vm->use_cpu_for_update ? "CPU" : "SDMA");
WARN_ONCE((vm->use_cpu_for_update & !amdgpu_vm_is_large_bar(adev)),
"CPU update of VM recommended only for large BAR system\n");
vm->last_update = NULL;
flags = AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS |
AMDGPU_GEM_CREATE_VRAM_CLEARED;
flags = AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
if (vm->use_cpu_for_update)
flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
else
@ -2299,7 +2359,7 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
size = amdgpu_vm_bo_size(adev, adev->vm_manager.root_level);
r = amdgpu_bo_create(adev, size, align, true, AMDGPU_GEM_DOMAIN_VRAM,
flags, NULL, NULL, init_pde_value,
flags, NULL, NULL, 0,
&vm->root.base.bo);
if (r)
goto error_free_sched_entity;
@ -2308,6 +2368,11 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
if (r)
goto error_free_root;
r = amdgpu_vm_clear_bo(adev, vm, vm->root.base.bo,
adev->vm_manager.root_level);
if (r)
goto error_unreserve;
vm->root.base.vm = vm;
list_add_tail(&vm->root.base.bo_list, &vm->root.base.bo->va);
list_add_tail(&vm->root.base.vm_status, &vm->evicted);
@ -2331,6 +2396,9 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
return 0;
error_unreserve:
amdgpu_bo_unreserve(vm->root.base.bo);
error_free_root:
amdgpu_bo_unref(&vm->root.base.bo->shadow);
amdgpu_bo_unref(&vm->root.base.bo);