This name is more fitting, especially for the changes coming next to support multi-GPU systems with proper DMA mappings. Cleaned up the code and renamed some related functions and variables to improve readability.
Signed-off-by: Felix Kuehling Felix.Kuehling@amd.com --- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h | 8 +- .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 209 +++++++++--------- 2 files changed, 104 insertions(+), 113 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h index 14f68c028126..910c50956e16 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h @@ -38,10 +38,10 @@ extern uint64_t amdgpu_amdkfd_total_mem_size;
struct amdgpu_device;
-struct kfd_bo_va_list { - struct list_head bo_list; +struct kfd_mem_attachment { + struct list_head list; struct amdgpu_bo_va *bo_va; - void *kgd_dev; + struct amdgpu_device *adev; bool is_mapped; uint64_t va; uint64_t pte_flags; @@ -50,7 +50,7 @@ struct kfd_bo_va_list { struct kgd_mem { struct mutex lock; struct amdgpu_bo *bo; - struct list_head bo_va_list; + struct list_head attachments; /* protected by amdkfd_process_info.lock */ struct ttm_validate_buffer validate_list; struct ttm_validate_buffer resv_list; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c index 6622695a5eed..d021152314ad 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c @@ -75,16 +75,16 @@ static inline struct amdgpu_device *get_amdgpu_device(struct kgd_dev *kgd) return (struct amdgpu_device *)kgd; }
-static bool check_if_add_bo_to_vm(struct amdgpu_vm *avm, +static bool kfd_mem_is_attached(struct amdgpu_vm *avm, struct kgd_mem *mem) { - struct kfd_bo_va_list *entry; + struct kfd_mem_attachment *entry;
- list_for_each_entry(entry, &mem->bo_va_list, bo_list) + list_for_each_entry(entry, &mem->attachments, list) if (entry->bo_va->base.vm == avm) - return false; + return true;
- return true; + return false; }
/* Set memory usage limits. Current, limits are @@ -471,7 +471,7 @@ static uint64_t get_pte_flags(struct amdgpu_device *adev, struct kgd_mem *mem) return pte_flags; }
-/* add_bo_to_vm - Add a BO to a VM +/* kfd_mem_attach - Add a BO to a VM * * Everything that needs to bo done only once when a BO is first added * to a VM. It can later be mapped and unmapped many times without @@ -483,15 +483,14 @@ static uint64_t get_pte_flags(struct amdgpu_device *adev, struct kgd_mem *mem) * 4. Alloc page tables and directories if needed * 4a. Validate new page tables and directories */ -static int add_bo_to_vm(struct amdgpu_device *adev, struct kgd_mem *mem, +static int kfd_mem_attach(struct amdgpu_device *adev, struct kgd_mem *mem, struct amdgpu_vm *vm, bool is_aql, - struct kfd_bo_va_list **p_bo_va_entry) + struct kfd_mem_attachment **p_attachment) { int ret; - struct kfd_bo_va_list *bo_va_entry; + struct kfd_mem_attachment *attachment; struct amdgpu_bo *bo = mem->bo; uint64_t va = mem->va; - struct list_head *list_bo_va = &mem->bo_va_list; unsigned long bo_size = bo->tbo.base.size;
if (!va) { @@ -502,29 +501,29 @@ static int add_bo_to_vm(struct amdgpu_device *adev, struct kgd_mem *mem, if (is_aql) va += bo_size;
- bo_va_entry = kzalloc(sizeof(*bo_va_entry), GFP_KERNEL); - if (!bo_va_entry) + attachment = kzalloc(sizeof(*attachment), GFP_KERNEL); + if (!attachment) return -ENOMEM;
pr_debug("\t add VA 0x%llx - 0x%llx to vm %p\n", va, va + bo_size, vm);
/* Add BO to VM internal data structures*/ - bo_va_entry->bo_va = amdgpu_vm_bo_add(adev, vm, bo); - if (!bo_va_entry->bo_va) { + attachment->bo_va = amdgpu_vm_bo_add(adev, vm, bo); + if (!attachment->bo_va) { ret = -EINVAL; pr_err("Failed to add BO object to VM. ret == %d\n", ret); goto err_vmadd; }
- bo_va_entry->va = va; - bo_va_entry->pte_flags = get_pte_flags(adev, mem); - bo_va_entry->kgd_dev = (void *)adev; - list_add(&bo_va_entry->bo_list, list_bo_va); + attachment->va = va; + attachment->pte_flags = get_pte_flags(adev, mem); + attachment->adev = adev; + list_add(&attachment->list, &mem->attachments);
- if (p_bo_va_entry) - *p_bo_va_entry = bo_va_entry; + if (p_attachment) + *p_attachment = attachment;
/* Allocate validate page tables if needed */ ret = vm_validate_pt_pd_bos(vm); @@ -536,22 +535,20 @@ static int add_bo_to_vm(struct amdgpu_device *adev, struct kgd_mem *mem, return 0;
err_alloc_pts: - amdgpu_vm_bo_rmv(adev, bo_va_entry->bo_va); - list_del(&bo_va_entry->bo_list); + amdgpu_vm_bo_rmv(adev, attachment->bo_va); + list_del(&attachment->list); err_vmadd: - kfree(bo_va_entry); + kfree(attachment); return ret; }
-static void remove_bo_from_vm(struct amdgpu_device *adev, - struct kfd_bo_va_list *entry, unsigned long size) +static void kfd_mem_detach(struct kfd_mem_attachment *attachment) { - pr_debug("\t remove VA 0x%llx - 0x%llx in entry %p\n", - entry->va, - entry->va + size, entry); - amdgpu_vm_bo_rmv(adev, entry->bo_va); - list_del(&entry->bo_list); - kfree(entry); + pr_debug("\t remove VA 0x%llx in entry %p\n", + attachment->va, attachment); + amdgpu_vm_bo_rmv(attachment->adev, attachment->bo_va); + list_del(&attachment->list); + kfree(attachment); }
static void add_kgd_mem_to_kfd_bo_list(struct kgd_mem *mem, @@ -726,7 +723,7 @@ static int reserve_bo_and_cond_vms(struct kgd_mem *mem, struct bo_vm_reservation_context *ctx) { struct amdgpu_bo *bo = mem->bo; - struct kfd_bo_va_list *entry; + struct kfd_mem_attachment *entry; unsigned int i; int ret;
@@ -738,7 +735,7 @@ static int reserve_bo_and_cond_vms(struct kgd_mem *mem, INIT_LIST_HEAD(&ctx->list); INIT_LIST_HEAD(&ctx->duplicates);
- list_for_each_entry(entry, &mem->bo_va_list, bo_list) { + list_for_each_entry(entry, &mem->attachments, list) { if ((vm && vm != entry->bo_va->base.vm) || (entry->is_mapped != map_type && map_type != BO_VM_ALL)) @@ -760,7 +757,7 @@ static int reserve_bo_and_cond_vms(struct kgd_mem *mem, list_add(&ctx->kfd_bo.tv.head, &ctx->list);
i = 0; - list_for_each_entry(entry, &mem->bo_va_list, bo_list) { + list_for_each_entry(entry, &mem->attachments, list) { if ((vm && vm != entry->bo_va->base.vm) || (entry->is_mapped != map_type && map_type != BO_VM_ALL)) @@ -815,7 +812,7 @@ static int unreserve_bo_and_vms(struct bo_vm_reservation_context *ctx, }
static int unmap_bo_from_gpuvm(struct amdgpu_device *adev, - struct kfd_bo_va_list *entry, + struct kfd_mem_attachment *entry, struct amdgpu_sync *sync) { struct amdgpu_bo_va *bo_va = entry->bo_va; @@ -831,7 +828,7 @@ static int unmap_bo_from_gpuvm(struct amdgpu_device *adev, }
static int update_gpuvm_pte(struct amdgpu_device *adev, - struct kfd_bo_va_list *entry, + struct kfd_mem_attachment *entry, struct amdgpu_sync *sync) { int ret; @@ -848,7 +845,7 @@ static int update_gpuvm_pte(struct amdgpu_device *adev, }
static int map_bo_to_gpuvm(struct amdgpu_device *adev, - struct kfd_bo_va_list *entry, struct amdgpu_sync *sync, + struct kfd_mem_attachment *entry, struct amdgpu_sync *sync, bool no_update_pte) { int ret; @@ -1235,7 +1232,7 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu( ret = -ENOMEM; goto err; } - INIT_LIST_HEAD(&(*mem)->bo_va_list); + INIT_LIST_HEAD(&(*mem)->attachments); mutex_init(&(*mem)->lock); (*mem)->aql_queue = !!(flags & KFD_IOC_ALLOC_MEM_FLAGS_AQL_QUEUE_MEM);
@@ -1316,7 +1313,7 @@ int amdgpu_amdkfd_gpuvm_free_memory_of_gpu( { struct amdkfd_process_info *process_info = mem->process_info; unsigned long bo_size = mem->bo->tbo.base.size; - struct kfd_bo_va_list *entry, *tmp; + struct kfd_mem_attachment *entry, *tmp; struct bo_vm_reservation_context ctx; struct ttm_validate_buffer *bo_list_entry; unsigned int mapped_to_gpu_memory; @@ -1360,9 +1357,8 @@ int amdgpu_amdkfd_gpuvm_free_memory_of_gpu( mem->va + bo_size * (1 + mem->aql_queue));
/* Remove from VM internal data structures */ - list_for_each_entry_safe(entry, tmp, &mem->bo_va_list, bo_list) - remove_bo_from_vm((struct amdgpu_device *)entry->kgd_dev, - entry, bo_size); + list_for_each_entry_safe(entry, tmp, &mem->attachments, list) + kfd_mem_detach(entry);
ret = unreserve_bo_and_vms(&ctx, false, false);
@@ -1404,10 +1400,10 @@ int amdgpu_amdkfd_gpuvm_map_memory_to_gpu( int ret; struct amdgpu_bo *bo; uint32_t domain; - struct kfd_bo_va_list *entry; + struct kfd_mem_attachment *entry; struct bo_vm_reservation_context ctx; - struct kfd_bo_va_list *bo_va_entry = NULL; - struct kfd_bo_va_list *bo_va_entry_aql = NULL; + struct kfd_mem_attachment *attachment = NULL; + struct kfd_mem_attachment *attachment_aql = NULL; unsigned long bo_size; bool is_invalid_userptr = false;
@@ -1456,21 +1452,20 @@ int amdgpu_amdkfd_gpuvm_map_memory_to_gpu( bo->tbo.mem.mem_type == TTM_PL_SYSTEM) is_invalid_userptr = true;
- if (check_if_add_bo_to_vm(avm, mem)) { - ret = add_bo_to_vm(adev, mem, avm, false, - &bo_va_entry); + if (!kfd_mem_is_attached(avm, mem)) { + ret = kfd_mem_attach(adev, mem, avm, false, &attachment); if (ret) - goto add_bo_to_vm_failed; + goto attach_failed; if (mem->aql_queue) { - ret = add_bo_to_vm(adev, mem, avm, - true, &bo_va_entry_aql); + ret = kfd_mem_attach(adev, mem, avm, true, + &attachment_aql); if (ret) - goto add_bo_to_vm_failed_aql; + goto attach_failed_aql; } } else { ret = vm_validate_pt_pd_bos(avm); if (unlikely(ret)) - goto add_bo_to_vm_failed; + goto attach_failed; }
if (mem->mapped_to_gpu_memory == 0 && @@ -1486,30 +1481,30 @@ int amdgpu_amdkfd_gpuvm_map_memory_to_gpu( } }
- list_for_each_entry(entry, &mem->bo_va_list, bo_list) { - if (entry->bo_va->base.vm == vm && !entry->is_mapped) { - pr_debug("\t map VA 0x%llx - 0x%llx in entry %p\n", - entry->va, entry->va + bo_size, - entry); + list_for_each_entry(entry, &mem->attachments, list) { + if (entry->bo_va->base.vm != vm || entry->is_mapped) + continue;
- ret = map_bo_to_gpuvm(adev, entry, ctx.sync, - is_invalid_userptr); - if (ret) { - pr_err("Failed to map bo to gpuvm\n"); - goto map_bo_to_gpuvm_failed; - } + pr_debug("\t map VA 0x%llx - 0x%llx in entry %p\n", + entry->va, entry->va + bo_size, entry);
- ret = vm_update_pds(vm, ctx.sync); - if (ret) { - pr_err("Failed to update page directories\n"); - goto map_bo_to_gpuvm_failed; - } + ret = map_bo_to_gpuvm(adev, entry, ctx.sync, + is_invalid_userptr); + if (ret) { + pr_err("Failed to map bo to gpuvm\n"); + goto map_bo_to_gpuvm_failed; + }
- entry->is_mapped = true; - mem->mapped_to_gpu_memory++; - pr_debug("\t INC mapping count %d\n", - mem->mapped_to_gpu_memory); + ret = vm_update_pds(vm, ctx.sync); + if (ret) { + pr_err("Failed to update page directories\n"); + goto map_bo_to_gpuvm_failed; } + + entry->is_mapped = true; + mem->mapped_to_gpu_memory++; + pr_debug("\t INC mapping count %d\n", + mem->mapped_to_gpu_memory); }
if (!amdgpu_ttm_tt_get_usermm(bo->tbo.ttm) && !bo->tbo.pin_count) @@ -1521,12 +1516,12 @@ int amdgpu_amdkfd_gpuvm_map_memory_to_gpu( goto out;
map_bo_to_gpuvm_failed: - if (bo_va_entry_aql) - remove_bo_from_vm(adev, bo_va_entry_aql, bo_size); -add_bo_to_vm_failed_aql: - if (bo_va_entry) - remove_bo_from_vm(adev, bo_va_entry, bo_size); -add_bo_to_vm_failed: + if (attachment_aql) + kfd_mem_detach(attachment_aql); +attach_failed_aql: + if (attachment) + kfd_mem_detach(attachment); +attach_failed: unreserve_bo_and_vms(&ctx, false, false); out: mutex_unlock(&mem->process_info->lock); @@ -1541,7 +1536,7 @@ int amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu( struct amdkfd_process_info *process_info = ((struct amdgpu_vm *)vm)->process_info; unsigned long bo_size = mem->bo->tbo.base.size; - struct kfd_bo_va_list *entry; + struct kfd_mem_attachment *entry; struct bo_vm_reservation_context ctx; int ret;
@@ -1565,26 +1560,24 @@ int amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu( mem->va + bo_size * (1 + mem->aql_queue), vm);
- list_for_each_entry(entry, &mem->bo_va_list, bo_list) { - if (entry->bo_va->base.vm == vm && entry->is_mapped) { - pr_debug("\t unmap VA 0x%llx - 0x%llx from entry %p\n", - entry->va, - entry->va + bo_size, - entry); + list_for_each_entry(entry, &mem->attachments, list) { + if (entry->bo_va->base.vm != vm || !entry->is_mapped) + continue;
- ret = unmap_bo_from_gpuvm(adev, entry, ctx.sync); - if (ret == 0) { - entry->is_mapped = false; - } else { - pr_err("failed to unmap VA 0x%llx\n", - mem->va); - goto unreserve_out; - } + pr_debug("\t unmap VA 0x%llx - 0x%llx from entry %p\n", + entry->va, entry->va + bo_size, entry);
- mem->mapped_to_gpu_memory--; - pr_debug("\t DEC mapping count %d\n", - mem->mapped_to_gpu_memory); + ret = unmap_bo_from_gpuvm(adev, entry, ctx.sync); + if (ret == 0) { + entry->is_mapped = false; + } else { + pr_err("failed to unmap VA 0x%llx\n", mem->va); + goto unreserve_out; } + + mem->mapped_to_gpu_memory--; + pr_debug("\t DEC mapping count %d\n", + mem->mapped_to_gpu_memory); }
/* If BO is unmapped from all VMs, unfence it. It can be evicted if @@ -1726,7 +1719,7 @@ int amdgpu_amdkfd_gpuvm_import_dmabuf(struct kgd_dev *kgd, if (mmap_offset) *mmap_offset = amdgpu_bo_mmap_offset(bo);
- INIT_LIST_HEAD(&(*mem)->bo_va_list); + INIT_LIST_HEAD(&(*mem)->attachments); mutex_init(&(*mem)->lock);
(*mem)->alloc_flags = @@ -1923,7 +1916,7 @@ static int validate_invalid_user_pages(struct amdkfd_process_info *process_info) list_for_each_entry_safe(mem, tmp_mem, &process_info->userptr_inval_list, validate_list.head) { - struct kfd_bo_va_list *bo_va_entry; + struct kfd_mem_attachment *attachment;
bo = mem->bo;
@@ -1946,13 +1939,13 @@ static int validate_invalid_user_pages(struct amdkfd_process_info *process_info) * VM faults if the GPU tries to access the invalid * memory. */ - list_for_each_entry(bo_va_entry, &mem->bo_va_list, bo_list) { - if (!bo_va_entry->is_mapped) + list_for_each_entry(attachment, &mem->attachments, list) { + if (!attachment->is_mapped) continue;
ret = update_gpuvm_pte((struct amdgpu_device *) - bo_va_entry->kgd_dev, - bo_va_entry, &sync); + attachment->adev, + attachment, &sync); if (ret) { pr_err("%s: update PTE failed\n", __func__); /* make sure this gets validated again */ @@ -2133,7 +2126,7 @@ int amdgpu_amdkfd_gpuvm_restore_process_bos(void *info, struct dma_fence **ef)
struct amdgpu_bo *bo = mem->bo; uint32_t domain = mem->domain; - struct kfd_bo_va_list *bo_va_entry; + struct kfd_mem_attachment *attachment;
total_size += amdgpu_bo_size(bo);
@@ -2153,11 +2146,9 @@ int amdgpu_amdkfd_gpuvm_restore_process_bos(void *info, struct dma_fence **ef) pr_debug("Memory eviction: Sync BO fence failed. Try again\n"); goto validate_map_fail; } - list_for_each_entry(bo_va_entry, &mem->bo_va_list, - bo_list) { + list_for_each_entry(attachment, &mem->attachments, list) { ret = update_gpuvm_pte((struct amdgpu_device *) - bo_va_entry->kgd_dev, - bo_va_entry, + attachment->adev, attachment, &sync_obj); if (ret) { pr_debug("Memory eviction: update PTE failed. Try again\n"); @@ -2232,7 +2223,7 @@ int amdgpu_amdkfd_add_gws_to_process(void *info, void *gws, struct kgd_mem **mem return -ENOMEM;
mutex_init(&(*mem)->lock); - INIT_LIST_HEAD(&(*mem)->bo_va_list); + INIT_LIST_HEAD(&(*mem)->attachments); (*mem)->bo = amdgpu_bo_ref(gws_bo); (*mem)->domain = AMDGPU_GEM_DOMAIN_GWS; (*mem)->process_info = process_info;