diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 1132 |
1 files changed, 119 insertions, 1013 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c index 5277c10d901d..f9479e23de18 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -89,6 +89,21 @@ struct amdgpu_prt_cb { }; /** + * struct amdgpu_vm_tlb_seq_cb - Helper to increment the TLB flush sequence + */ +struct amdgpu_vm_tlb_seq_cb { + /** + * @vm: pointer to the amdgpu_vm structure to set the fence sequence on + */ + struct amdgpu_vm *vm; + + /** + * @cb: callback + */ + struct dma_fence_cb cb; +}; + +/** * amdgpu_vm_set_pasid - manage pasid and vm ptr mapping * * @adev: amdgpu_device pointer @@ -155,108 +170,6 @@ static inline void amdgpu_vm_eviction_unlock(struct amdgpu_vm *vm) } /** - * amdgpu_vm_level_shift - return the addr shift for each level - * - * @adev: amdgpu_device pointer - * @level: VMPT level - * - * Returns: - * The number of bits the pfn needs to be right shifted for a level. - */ -static unsigned amdgpu_vm_level_shift(struct amdgpu_device *adev, - unsigned level) -{ - switch (level) { - case AMDGPU_VM_PDB2: - case AMDGPU_VM_PDB1: - case AMDGPU_VM_PDB0: - return 9 * (AMDGPU_VM_PDB0 - level) + - adev->vm_manager.block_size; - case AMDGPU_VM_PTB: - return 0; - default: - return ~0; - } -} - -/** - * amdgpu_vm_num_entries - return the number of entries in a PD/PT - * - * @adev: amdgpu_device pointer - * @level: VMPT level - * - * Returns: - * The number of entries in a page directory or page table. - */ -static unsigned amdgpu_vm_num_entries(struct amdgpu_device *adev, - unsigned level) -{ - unsigned shift = amdgpu_vm_level_shift(adev, - adev->vm_manager.root_level); - - if (level == adev->vm_manager.root_level) - /* For the root directory */ - return round_up(adev->vm_manager.max_pfn, 1ULL << shift) - >> shift; - else if (level != AMDGPU_VM_PTB) - /* Everything in between */ - return 512; - else - /* For the page tables on the leaves */ - return AMDGPU_VM_PTE_COUNT(adev); -} - -/** - * amdgpu_vm_num_ats_entries - return the number of ATS entries in the root PD - * - * @adev: amdgpu_device pointer - * - * Returns: - * The number of entries in the root page directory which needs the ATS setting. - */ -static unsigned amdgpu_vm_num_ats_entries(struct amdgpu_device *adev) -{ - unsigned shift; - - shift = amdgpu_vm_level_shift(adev, adev->vm_manager.root_level); - return AMDGPU_GMC_HOLE_START >> (shift + AMDGPU_GPU_PAGE_SHIFT); -} - -/** - * amdgpu_vm_entries_mask - the mask to get the entry number of a PD/PT - * - * @adev: amdgpu_device pointer - * @level: VMPT level - * - * Returns: - * The mask to extract the entry number of a PD/PT from an address. - */ -static uint32_t amdgpu_vm_entries_mask(struct amdgpu_device *adev, - unsigned int level) -{ - if (level <= adev->vm_manager.root_level) - return 0xffffffff; - else if (level != AMDGPU_VM_PTB) - return 0x1ff; - else - return AMDGPU_VM_PTE_COUNT(adev) - 1; -} - -/** - * amdgpu_vm_bo_size - returns the size of the BOs in bytes - * - * @adev: amdgpu_device pointer - * @level: VMPT level - * - * Returns: - * The size of the BO for a page directory or page table in bytes. - */ -static unsigned amdgpu_vm_bo_size(struct amdgpu_device *adev, unsigned level) -{ - return AMDGPU_GPU_PAGE_ALIGN(amdgpu_vm_num_entries(adev, level) * 8); -} - -/** * amdgpu_vm_bo_evicted - vm_bo is evicted * * @vm_bo: vm_bo which is evicted @@ -358,9 +271,8 @@ static void amdgpu_vm_bo_done(struct amdgpu_vm_bo_base *vm_bo) * Initialize a bo_va_base structure and add it to the appropriate lists * */ -static void amdgpu_vm_bo_base_init(struct amdgpu_vm_bo_base *base, - struct amdgpu_vm *vm, - struct amdgpu_bo *bo) +void amdgpu_vm_bo_base_init(struct amdgpu_vm_bo_base *base, + struct amdgpu_vm *vm, struct amdgpu_bo *bo) { base->vm = vm; base->bo = bo; @@ -396,228 +308,6 @@ static void amdgpu_vm_bo_base_init(struct amdgpu_vm_bo_base *base, } /** - * amdgpu_vm_pt_parent - get the parent page directory - * - * @pt: child page table - * - * Helper to get the parent entry for the child page table. NULL if we are at - * the root page directory. - */ -static struct amdgpu_vm_bo_base *amdgpu_vm_pt_parent(struct amdgpu_vm_bo_base *pt) -{ - struct amdgpu_bo *parent = pt->bo->parent; - - if (!parent) - return NULL; - - return parent->vm_bo; -} - -/* - * amdgpu_vm_pt_cursor - state for for_each_amdgpu_vm_pt - */ -struct amdgpu_vm_pt_cursor { - uint64_t pfn; - struct amdgpu_vm_bo_base *parent; - struct amdgpu_vm_bo_base *entry; - unsigned level; -}; - -/** - * amdgpu_vm_pt_start - start PD/PT walk - * - * @adev: amdgpu_device pointer - * @vm: amdgpu_vm structure - * @start: start address of the walk - * @cursor: state to initialize - * - * Initialize a amdgpu_vm_pt_cursor to start a walk. - */ -static void amdgpu_vm_pt_start(struct amdgpu_device *adev, - struct amdgpu_vm *vm, uint64_t start, - struct amdgpu_vm_pt_cursor *cursor) -{ - cursor->pfn = start; - cursor->parent = NULL; - cursor->entry = &vm->root; - cursor->level = adev->vm_manager.root_level; -} - -/** - * amdgpu_vm_pt_descendant - go to child node - * - * @adev: amdgpu_device pointer - * @cursor: current state - * - * Walk to the child node of the current node. - * Returns: - * True if the walk was possible, false otherwise. - */ -static bool amdgpu_vm_pt_descendant(struct amdgpu_device *adev, - struct amdgpu_vm_pt_cursor *cursor) -{ - unsigned mask, shift, idx; - - if ((cursor->level == AMDGPU_VM_PTB) || !cursor->entry || - !cursor->entry->bo) - return false; - - mask = amdgpu_vm_entries_mask(adev, cursor->level); - shift = amdgpu_vm_level_shift(adev, cursor->level); - - ++cursor->level; - idx = (cursor->pfn >> shift) & mask; - cursor->parent = cursor->entry; - cursor->entry = &to_amdgpu_bo_vm(cursor->entry->bo)->entries[idx]; - return true; -} - -/** - * amdgpu_vm_pt_sibling - go to sibling node - * - * @adev: amdgpu_device pointer - * @cursor: current state - * - * Walk to the sibling node of the current node. - * Returns: - * True if the walk was possible, false otherwise. - */ -static bool amdgpu_vm_pt_sibling(struct amdgpu_device *adev, - struct amdgpu_vm_pt_cursor *cursor) -{ - unsigned shift, num_entries; - - /* Root doesn't have a sibling */ - if (!cursor->parent) - return false; - - /* Go to our parents and see if we got a sibling */ - shift = amdgpu_vm_level_shift(adev, cursor->level - 1); - num_entries = amdgpu_vm_num_entries(adev, cursor->level - 1); - - if (cursor->entry == &to_amdgpu_bo_vm(cursor->parent->bo)->entries[num_entries - 1]) - return false; - - cursor->pfn += 1ULL << shift; - cursor->pfn &= ~((1ULL << shift) - 1); - ++cursor->entry; - return true; -} - -/** - * amdgpu_vm_pt_ancestor - go to parent node - * - * @cursor: current state - * - * Walk to the parent node of the current node. - * Returns: - * True if the walk was possible, false otherwise. - */ -static bool amdgpu_vm_pt_ancestor(struct amdgpu_vm_pt_cursor *cursor) -{ - if (!cursor->parent) - return false; - - --cursor->level; - cursor->entry = cursor->parent; - cursor->parent = amdgpu_vm_pt_parent(cursor->parent); - return true; -} - -/** - * amdgpu_vm_pt_next - get next PD/PT in hieratchy - * - * @adev: amdgpu_device pointer - * @cursor: current state - * - * Walk the PD/PT tree to the next node. - */ -static void amdgpu_vm_pt_next(struct amdgpu_device *adev, - struct amdgpu_vm_pt_cursor *cursor) -{ - /* First try a newborn child */ - if (amdgpu_vm_pt_descendant(adev, cursor)) - return; - - /* If that didn't worked try to find a sibling */ - while (!amdgpu_vm_pt_sibling(adev, cursor)) { - /* No sibling, go to our parents and grandparents */ - if (!amdgpu_vm_pt_ancestor(cursor)) { - cursor->pfn = ~0ll; - return; - } - } -} - -/** - * amdgpu_vm_pt_first_dfs - start a deep first search - * - * @adev: amdgpu_device structure - * @vm: amdgpu_vm structure - * @start: optional cursor to start with - * @cursor: state to initialize - * - * Starts a deep first traversal of the PD/PT tree. - */ -static void amdgpu_vm_pt_first_dfs(struct amdgpu_device *adev, - struct amdgpu_vm *vm, - struct amdgpu_vm_pt_cursor *start, - struct amdgpu_vm_pt_cursor *cursor) -{ - if (start) - *cursor = *start; - else - amdgpu_vm_pt_start(adev, vm, 0, cursor); - while (amdgpu_vm_pt_descendant(adev, cursor)); -} - -/** - * amdgpu_vm_pt_continue_dfs - check if the deep first search should continue - * - * @start: starting point for the search - * @entry: current entry - * - * Returns: - * True when the search should continue, false otherwise. - */ -static bool amdgpu_vm_pt_continue_dfs(struct amdgpu_vm_pt_cursor *start, - struct amdgpu_vm_bo_base *entry) -{ - return entry && (!start || entry != start->entry); -} - -/** - * amdgpu_vm_pt_next_dfs - get the next node for a deep first search - * - * @adev: amdgpu_device structure - * @cursor: current state - * - * Move the cursor to the next node in a deep first search. - */ -static void amdgpu_vm_pt_next_dfs(struct amdgpu_device *adev, - struct amdgpu_vm_pt_cursor *cursor) -{ - if (!cursor->entry) - return; - - if (!cursor->parent) - cursor->entry = NULL; - else if (amdgpu_vm_pt_sibling(adev, cursor)) - while (amdgpu_vm_pt_descendant(adev, cursor)); - else - amdgpu_vm_pt_ancestor(cursor); -} - -/* - * for_each_amdgpu_vm_pt_dfs_safe - safe deep first search of all PDs/PTs - */ -#define for_each_amdgpu_vm_pt_dfs_safe(adev, vm, start, cursor, entry) \ - for (amdgpu_vm_pt_first_dfs((adev), (vm), (start), &(cursor)), \ - (entry) = (cursor).entry, amdgpu_vm_pt_next_dfs((adev), &(cursor));\ - amdgpu_vm_pt_continue_dfs((start), (entry)); \ - (entry) = (cursor).entry, amdgpu_vm_pt_next_dfs((adev), &(cursor))) - -/** * amdgpu_vm_get_pd_bo - add the VM PD to a validation list * * @vm: vm providing the BOs @@ -726,316 +416,6 @@ bool amdgpu_vm_ready(struct amdgpu_vm *vm) } /** - * amdgpu_vm_clear_bo - initially clear the PDs/PTs - * - * @adev: amdgpu_device pointer - * @vm: VM to clear BO from - * @vmbo: BO to clear - * @immediate: use an immediate update - * - * Root PD needs to be reserved when calling this. - * - * Returns: - * 0 on success, errno otherwise. - */ -static int amdgpu_vm_clear_bo(struct amdgpu_device *adev, - struct amdgpu_vm *vm, - struct amdgpu_bo_vm *vmbo, - bool immediate) -{ - struct ttm_operation_ctx ctx = { true, false }; - unsigned level = adev->vm_manager.root_level; - struct amdgpu_vm_update_params params; - struct amdgpu_bo *ancestor = &vmbo->bo; - struct amdgpu_bo *bo = &vmbo->bo; - unsigned entries, ats_entries; - uint64_t addr; - int r, idx; - - /* Figure out our place in the hierarchy */ - if (ancestor->parent) { - ++level; - while (ancestor->parent->parent) { - ++level; - ancestor = ancestor->parent; - } - } - - entries = amdgpu_bo_size(bo) / 8; - if (!vm->pte_support_ats) { - ats_entries = 0; - - } else if (!bo->parent) { - ats_entries = amdgpu_vm_num_ats_entries(adev); - ats_entries = min(ats_entries, entries); - entries -= ats_entries; - - } else { - struct amdgpu_vm_bo_base *pt; - - pt = ancestor->vm_bo; - ats_entries = amdgpu_vm_num_ats_entries(adev); - if ((pt - to_amdgpu_bo_vm(vm->root.bo)->entries) >= ats_entries) { - ats_entries = 0; - } else { - ats_entries = entries; - entries = 0; - } - } - - r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx); - if (r) - return r; - - if (vmbo->shadow) { - struct amdgpu_bo *shadow = vmbo->shadow; - - r = ttm_bo_validate(&shadow->tbo, &shadow->placement, &ctx); - if (r) - return r; - } - - if (!drm_dev_enter(adev_to_drm(adev), &idx)) - return -ENODEV; - - r = vm->update_funcs->map_table(vmbo); - if (r) - goto exit; - - memset(¶ms, 0, sizeof(params)); - params.adev = adev; - params.vm = vm; - params.immediate = immediate; - - r = vm->update_funcs->prepare(¶ms, NULL, AMDGPU_SYNC_EXPLICIT); - if (r) - goto exit; - - addr = 0; - if (ats_entries) { - uint64_t value = 0, flags; - - flags = AMDGPU_PTE_DEFAULT_ATC; - if (level != AMDGPU_VM_PTB) { - /* Handle leaf PDEs as PTEs */ - flags |= AMDGPU_PDE_PTE; - amdgpu_gmc_get_vm_pde(adev, level, &value, &flags); - } - - r = vm->update_funcs->update(¶ms, vmbo, addr, 0, ats_entries, - value, flags); - if (r) - goto exit; - - addr += ats_entries * 8; - } - - if (entries) { - uint64_t value = 0, flags = 0; - - if (adev->asic_type >= CHIP_VEGA10) { - if (level != AMDGPU_VM_PTB) { - /* Handle leaf PDEs as PTEs */ - flags |= AMDGPU_PDE_PTE; - amdgpu_gmc_get_vm_pde(adev, level, - &value, &flags); - } else { - /* Workaround for fault priority problem on GMC9 */ - flags = AMDGPU_PTE_EXECUTABLE; - } - } - - r = vm->update_funcs->update(¶ms, vmbo, addr, 0, entries, - value, flags); - if (r) - goto exit; - } - - r = vm->update_funcs->commit(¶ms, NULL); -exit: - drm_dev_exit(idx); - return r; -} - -/** - * amdgpu_vm_pt_create - create bo for PD/PT - * - * @adev: amdgpu_device pointer - * @vm: requesting vm - * @level: the page table level - * @immediate: use a immediate update - * @vmbo: pointer to the buffer object pointer - */ -static int amdgpu_vm_pt_create(struct amdgpu_device *adev, - struct amdgpu_vm *vm, - int level, bool immediate, - struct amdgpu_bo_vm **vmbo) -{ - struct amdgpu_bo_param bp; - struct amdgpu_bo *bo; - struct dma_resv *resv; - unsigned int num_entries; - int r; - - memset(&bp, 0, sizeof(bp)); - - bp.size = amdgpu_vm_bo_size(adev, level); - bp.byte_align = AMDGPU_GPU_PAGE_SIZE; - bp.domain = AMDGPU_GEM_DOMAIN_VRAM; - bp.domain = amdgpu_bo_get_preferred_domain(adev, bp.domain); - bp.flags = AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS | - AMDGPU_GEM_CREATE_CPU_GTT_USWC; - - if (level < AMDGPU_VM_PTB) - num_entries = amdgpu_vm_num_entries(adev, level); - else - num_entries = 0; - - bp.bo_ptr_size = struct_size((*vmbo), entries, num_entries); - - if (vm->use_cpu_for_update) - bp.flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED; - - bp.type = ttm_bo_type_kernel; - bp.no_wait_gpu = immediate; - if (vm->root.bo) - bp.resv = vm->root.bo->tbo.base.resv; - - r = amdgpu_bo_create_vm(adev, &bp, vmbo); - if (r) - return r; - - bo = &(*vmbo)->bo; - if (vm->is_compute_context || (adev->flags & AMD_IS_APU)) { - (*vmbo)->shadow = NULL; - return 0; - } - - if (!bp.resv) - WARN_ON(dma_resv_lock(bo->tbo.base.resv, - NULL)); - resv = bp.resv; - memset(&bp, 0, sizeof(bp)); - bp.size = amdgpu_vm_bo_size(adev, level); - bp.domain = AMDGPU_GEM_DOMAIN_GTT; - bp.flags = AMDGPU_GEM_CREATE_CPU_GTT_USWC; - bp.type = ttm_bo_type_kernel; - bp.resv = bo->tbo.base.resv; - bp.bo_ptr_size = sizeof(struct amdgpu_bo); - - r = amdgpu_bo_create(adev, &bp, &(*vmbo)->shadow); - - if (!resv) - dma_resv_unlock(bo->tbo.base.resv); - - if (r) { - amdgpu_bo_unref(&bo); - return r; - } - - (*vmbo)->shadow->parent = amdgpu_bo_ref(bo); - amdgpu_bo_add_to_shadow_list(*vmbo); - - return 0; -} - -/** - * amdgpu_vm_alloc_pts - Allocate a specific page table - * - * @adev: amdgpu_device pointer - * @vm: VM to allocate page tables for - * @cursor: Which page table to allocate - * @immediate: use an immediate update - * - * Make sure a specific page table or directory is allocated. - * - * Returns: - * 1 if page table needed to be allocated, 0 if page table was already - * allocated, negative errno if an error occurred. - */ -static int amdgpu_vm_alloc_pts(struct amdgpu_device *adev, - struct amdgpu_vm *vm, - struct amdgpu_vm_pt_cursor *cursor, - bool immediate) -{ - struct amdgpu_vm_bo_base *entry = cursor->entry; - struct amdgpu_bo *pt_bo; - struct amdgpu_bo_vm *pt; - int r; - - if (entry->bo) - return 0; - - r = amdgpu_vm_pt_create(adev, vm, cursor->level, immediate, &pt); - if (r) - return r; - - /* Keep a reference to the root directory to avoid - * freeing them up in the wrong order. - */ - pt_bo = &pt->bo; - pt_bo->parent = amdgpu_bo_ref(cursor->parent->bo); - amdgpu_vm_bo_base_init(entry, vm, pt_bo); - r = amdgpu_vm_clear_bo(adev, vm, pt, immediate); - if (r) - goto error_free_pt; - - return 0; - -error_free_pt: - amdgpu_bo_unref(&pt->shadow); - amdgpu_bo_unref(&pt_bo); - return r; -} - -/** - * amdgpu_vm_free_table - fre one PD/PT - * - * @entry: PDE to free - */ -static void amdgpu_vm_free_table(struct amdgpu_vm_bo_base *entry) -{ - struct amdgpu_bo *shadow; - - if (!entry->bo) - return; - - shadow = amdgpu_bo_shadowed(entry->bo); - if (shadow) { - ttm_bo_set_bulk_move(&shadow->tbo, NULL); - amdgpu_bo_unref(&shadow); - } - - ttm_bo_set_bulk_move(&entry->bo->tbo, NULL); - entry->bo->vm_bo = NULL; - list_del(&entry->vm_status); - amdgpu_bo_unref(&entry->bo); -} - -/** - * amdgpu_vm_free_pts - free PD/PT levels - * - * @adev: amdgpu device structure - * @vm: amdgpu vm structure - * @start: optional cursor where to start freeing PDs/PTs - * - * Free the page directory or page table level and all sub levels. - */ -static void amdgpu_vm_free_pts(struct amdgpu_device *adev, - struct amdgpu_vm *vm, - struct amdgpu_vm_pt_cursor *start) -{ - struct amdgpu_vm_pt_cursor cursor; - struct amdgpu_vm_bo_base *entry; - - for_each_amdgpu_vm_pt_dfs_safe(adev, vm, start, cursor, entry) - amdgpu_vm_free_table(entry); - - if (start) - amdgpu_vm_free_table(start->entry); -} - -/** * amdgpu_vm_check_compute_bug - check whether asic has compute vm bug * * @adev: amdgpu_device pointer @@ -1282,53 +662,6 @@ uint64_t amdgpu_vm_map_gart(const dma_addr_t *pages_addr, uint64_t addr) } /** - * amdgpu_vm_update_pde - update a single level in the hierarchy - * - * @params: parameters for the update - * @vm: requested vm - * @entry: entry to update - * - * Makes sure the requested entry in parent is up to date. - */ -static int amdgpu_vm_update_pde(struct amdgpu_vm_update_params *params, - struct amdgpu_vm *vm, - struct amdgpu_vm_bo_base *entry) -{ - struct amdgpu_vm_bo_base *parent = amdgpu_vm_pt_parent(entry); - struct amdgpu_bo *bo = parent->bo, *pbo; - uint64_t pde, pt, flags; - unsigned level; - - for (level = 0, pbo = bo->parent; pbo; ++level) - pbo = pbo->parent; - - level += params->adev->vm_manager.root_level; - amdgpu_gmc_get_pde_for_bo(entry->bo, level, &pt, &flags); - pde = (entry - to_amdgpu_bo_vm(parent->bo)->entries) * 8; - return vm->update_funcs->update(params, to_amdgpu_bo_vm(bo), pde, pt, - 1, 0, flags); -} - -/** - * amdgpu_vm_invalidate_pds - mark all PDs as invalid - * - * @adev: amdgpu_device pointer - * @vm: related vm - * - * Mark all PD level as invalid after an error. - */ -static void amdgpu_vm_invalidate_pds(struct amdgpu_device *adev, - struct amdgpu_vm *vm) -{ - struct amdgpu_vm_pt_cursor cursor; - struct amdgpu_vm_bo_base *entry; - - for_each_amdgpu_vm_pt_dfs_safe(adev, vm, NULL, cursor, entry) - if (entry->bo && !entry->moved) - amdgpu_vm_bo_relocated(entry); -} - -/** * amdgpu_vm_update_pdes - make sure that all directories are valid * * @adev: amdgpu_device pointer @@ -1344,6 +677,7 @@ int amdgpu_vm_update_pdes(struct amdgpu_device *adev, struct amdgpu_vm *vm, bool immediate) { struct amdgpu_vm_update_params params; + struct amdgpu_vm_bo_base *entry; int r, idx; if (list_empty(&vm->relocated)) @@ -1359,17 +693,10 @@ int amdgpu_vm_update_pdes(struct amdgpu_device *adev, r = vm->update_funcs->prepare(¶ms, NULL, AMDGPU_SYNC_EXPLICIT); if (r) - goto exit; - - while (!list_empty(&vm->relocated)) { - struct amdgpu_vm_bo_base *entry; - - entry = list_first_entry(&vm->relocated, - struct amdgpu_vm_bo_base, - vm_status); - amdgpu_vm_bo_idle(entry); + goto error; - r = amdgpu_vm_update_pde(¶ms, vm, entry); + list_for_each_entry(entry, &vm->relocated, vm_status) { + r = amdgpu_vm_pde_update(¶ms, entry); if (r) goto error; } @@ -1377,297 +704,68 @@ int amdgpu_vm_update_pdes(struct amdgpu_device *adev, r = vm->update_funcs->commit(¶ms, &vm->last_update); if (r) goto error; - drm_dev_exit(idx); - return 0; + + while (!list_empty(&vm->relocated)) { + entry = list_first_entry(&vm->relocated, + struct amdgpu_vm_bo_base, + vm_status); + amdgpu_vm_bo_idle(entry); + } error: - amdgpu_vm_invalidate_pds(adev, vm); -exit: drm_dev_exit(idx); return r; } -/* - * amdgpu_vm_update_flags - figure out flags for PTE updates - * - * Make sure to set the right flags for the PTEs at the desired level. - */ -static void amdgpu_vm_update_flags(struct amdgpu_vm_update_params *params, - struct amdgpu_bo_vm *pt, unsigned int level, - uint64_t pe, uint64_t addr, - unsigned int count, uint32_t incr, - uint64_t flags) - -{ - if (level != AMDGPU_VM_PTB) { - flags |= AMDGPU_PDE_PTE; - amdgpu_gmc_get_vm_pde(params->adev, level, &addr, &flags); - - } else if (params->adev->asic_type >= CHIP_VEGA10 && - !(flags & AMDGPU_PTE_VALID) && - !(flags & AMDGPU_PTE_PRT)) { - - /* Workaround for fault priority problem on GMC9 */ - flags |= AMDGPU_PTE_EXECUTABLE; - } - - params->vm->update_funcs->update(params, pt, pe, addr, count, incr, - flags); -} - /** - * amdgpu_vm_fragment - get fragment for PTEs + * amdgpu_vm_tlb_seq_cb - make sure to increment tlb sequence + * @fence: unused + * @cb: the callback structure * - * @params: see amdgpu_vm_update_params definition - * @start: first PTE to handle - * @end: last PTE to handle - * @flags: hw mapping flags - * @frag: resulting fragment size - * @frag_end: end of this fragment - * - * Returns the first possible fragment for the start and end address. - */ -static void amdgpu_vm_fragment(struct amdgpu_vm_update_params *params, - uint64_t start, uint64_t end, uint64_t flags, - unsigned int *frag, uint64_t *frag_end) -{ - /** - * The MC L1 TLB supports variable sized pages, based on a fragment - * field in the PTE. When this field is set to a non-zero value, page - * granularity is increased from 4KB to (1 << (12 + frag)). The PTE - * flags are considered valid for all PTEs within the fragment range - * and corresponding mappings are assumed to be physically contiguous. - * - * The L1 TLB can store a single PTE for the whole fragment, - * significantly increasing the space available for translation - * caching. This leads to large improvements in throughput when the - * TLB is under pressure. - * - * The L2 TLB distributes small and large fragments into two - * asymmetric partitions. The large fragment cache is significantly - * larger. Thus, we try to use large fragments wherever possible. - * Userspace can support this by aligning virtual base address and - * allocation size to the fragment size. - * - * Starting with Vega10 the fragment size only controls the L1. The L2 - * is now directly feed with small/huge/giant pages from the walker. - */ - unsigned max_frag; - - if (params->adev->asic_type < CHIP_VEGA10) - max_frag = params->adev->vm_manager.fragment_size; - else - max_frag = 31; - - /* system pages are non continuously */ - if (params->pages_addr) { - *frag = 0; - *frag_end = end; - return; - } - - /* This intentionally wraps around if no bit is set */ - *frag = min((unsigned)ffs(start) - 1, (unsigned)fls64(end - start) - 1); - if (*frag >= max_frag) { - *frag = max_frag; - *frag_end = end & ~((1ULL << max_frag) - 1); - } else { - *frag_end = start + (1 << *frag); - } -} - -/** - * amdgpu_vm_update_ptes - make sure that page tables are valid - * - * @params: see amdgpu_vm_update_params definition - * @start: start of GPU address range - * @end: end of GPU address range - * @dst: destination address to map to, the next dst inside the function - * @flags: mapping flags - * - * Update the page tables in the range @start - @end. - * - * Returns: - * 0 for success, -EINVAL for failure. + * Increments the tlb sequence to make sure that future CS execute a VM flush. */ -static int amdgpu_vm_update_ptes(struct amdgpu_vm_update_params *params, - uint64_t start, uint64_t end, - uint64_t dst, uint64_t flags) +static void amdgpu_vm_tlb_seq_cb(struct dma_fence *fence, + struct dma_fence_cb *cb) { - struct amdgpu_device *adev = params->adev; - struct amdgpu_vm_pt_cursor cursor; - uint64_t frag_start = start, frag_end; - unsigned int frag; - int r; - - /* figure out the initial fragment */ - amdgpu_vm_fragment(params, frag_start, end, flags, &frag, &frag_end); - - /* walk over the address space and update the PTs */ - amdgpu_vm_pt_start(adev, params->vm, start, &cursor); - while (cursor.pfn < end) { - unsigned shift, parent_shift, mask; - uint64_t incr, entry_end, pe_start; - struct amdgpu_bo *pt; - - if (!params->unlocked) { - /* make sure that the page tables covering the - * address range are actually allocated - */ - r = amdgpu_vm_alloc_pts(params->adev, params->vm, - &cursor, params->immediate); - if (r) - return r; - } - - shift = amdgpu_vm_level_shift(adev, cursor.level); - parent_shift = amdgpu_vm_level_shift(adev, cursor.level - 1); - if (params->unlocked) { - /* Unlocked updates are only allowed on the leaves */ - if (amdgpu_vm_pt_descendant(adev, &cursor)) - continue; - } else if (adev->asic_type < CHIP_VEGA10 && - (flags & AMDGPU_PTE_VALID)) { - /* No huge page support before GMC v9 */ - if (cursor.level != AMDGPU_VM_PTB) { - if (!amdgpu_vm_pt_descendant(adev, &cursor)) - return -ENOENT; - continue; - } - } else if (frag < shift) { - /* We can't use this level when the fragment size is - * smaller than the address shift. Go to the next - * child entry and try again. - */ - if (amdgpu_vm_pt_descendant(adev, &cursor)) - continue; - } else if (frag >= parent_shift) { - /* If the fragment size is even larger than the parent - * shift we should go up one level and check it again. - */ - if (!amdgpu_vm_pt_ancestor(&cursor)) - return -EINVAL; - continue; - } + struct amdgpu_vm_tlb_seq_cb *tlb_cb; - pt = cursor.entry->bo; - if (!pt) { - /* We need all PDs and PTs for mapping something, */ - if (flags & AMDGPU_PTE_VALID) - return -ENOENT; - - /* but unmapping something can happen at a higher - * level. - */ - if (!amdgpu_vm_pt_ancestor(&cursor)) - return -EINVAL; - - pt = cursor.entry->bo; - shift = parent_shift; - frag_end = max(frag_end, ALIGN(frag_start + 1, - 1ULL << shift)); - } - - /* Looks good so far, calculate parameters for the update */ - incr = (uint64_t)AMDGPU_GPU_PAGE_SIZE << shift; - mask = amdgpu_vm_entries_mask(adev, cursor.level); - pe_start = ((cursor.pfn >> shift) & mask) * 8; - entry_end = ((uint64_t)mask + 1) << shift; - entry_end += cursor.pfn & ~(entry_end - 1); - entry_end = min(entry_end, end); - - do { - struct amdgpu_vm *vm = params->vm; - uint64_t upd_end = min(entry_end, frag_end); - unsigned nptes = (upd_end - frag_start) >> shift; - uint64_t upd_flags = flags | AMDGPU_PTE_FRAG(frag); - - /* This can happen when we set higher level PDs to - * silent to stop fault floods. - */ - nptes = max(nptes, 1u); - - trace_amdgpu_vm_update_ptes(params, frag_start, upd_end, - min(nptes, 32u), dst, incr, upd_flags, - vm->task_info.pid, - vm->immediate.fence_context); - amdgpu_vm_update_flags(params, to_amdgpu_bo_vm(pt), - cursor.level, pe_start, dst, - nptes, incr, upd_flags); - - pe_start += nptes * 8; - dst += nptes * incr; - - frag_start = upd_end; - if (frag_start >= frag_end) { - /* figure out the next fragment */ - amdgpu_vm_fragment(params, frag_start, end, - flags, &frag, &frag_end); - if (frag < shift) - break; - } - } while (frag_start < entry_end); - - if (amdgpu_vm_pt_descendant(adev, &cursor)) { - /* Free all child entries. - * Update the tables with the flags and addresses and free up subsequent - * tables in the case of huge pages or freed up areas. - * This is the maximum you can free, because all other page tables are not - * completely covered by the range and so potentially still in use. - */ - while (cursor.pfn < frag_start) { - /* Make sure previous mapping is freed */ - if (cursor.entry->bo) { - params->table_freed = true; - amdgpu_vm_free_pts(adev, params->vm, &cursor); - } - amdgpu_vm_pt_next(adev, &cursor); - } - - } else if (frag >= shift) { - /* or just move on to the next on the same level. */ - amdgpu_vm_pt_next(adev, &cursor); - } - } - - return 0; + tlb_cb = container_of(cb, typeof(*tlb_cb), cb); + atomic64_inc(&tlb_cb->vm->tlb_seq); + kfree(tlb_cb); } /** - * amdgpu_vm_bo_update_mapping - update a mapping in the vm page table + * amdgpu_vm_update_range - update a range in the vm page table * - * @adev: amdgpu_device pointer of the VM - * @bo_adev: amdgpu_device pointer of the mapped BO - * @vm: requested vm + * @adev: amdgpu_device pointer to use for commands + * @vm: the VM to update the range * @immediate: immediate submission in a page fault * @unlocked: unlocked invalidation during MM callback + * @flush_tlb: trigger tlb invalidation after update completed * @resv: fences we need to sync to * @start: start of mapped range * @last: last mapped entry * @flags: flags for the entries * @offset: offset into nodes and pages_addr + * @vram_base: base for vram mappings * @res: ttm_resource to map * @pages_addr: DMA addresses to use for mapping * @fence: optional resulting fence - * @table_freed: return true if page table is freed * * Fill in the page table entries between @start and @last. * * Returns: - * 0 for success, -EINVAL for failure. + * 0 for success, negative erro code for failure. */ -int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev, - struct amdgpu_device *bo_adev, - struct amdgpu_vm *vm, bool immediate, - bool unlocked, struct dma_resv *resv, - uint64_t start, uint64_t last, - uint64_t flags, uint64_t offset, - struct ttm_resource *res, - dma_addr_t *pages_addr, - struct dma_fence **fence, - bool *table_freed) +int amdgpu_vm_update_range(struct amdgpu_device *adev, struct amdgpu_vm *vm, + bool immediate, bool unlocked, bool flush_tlb, + struct dma_resv *resv, uint64_t start, uint64_t last, + uint64_t flags, uint64_t offset, uint64_t vram_base, + struct ttm_resource *res, dma_addr_t *pages_addr, + struct dma_fence **fence) { struct amdgpu_vm_update_params params; + struct amdgpu_vm_tlb_seq_cb *tlb_cb; struct amdgpu_res_cursor cursor; enum amdgpu_sync_mode sync_mode; int r, idx; @@ -1675,6 +773,18 @@ int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev, if (!drm_dev_enter(adev_to_drm(adev), &idx)) return -ENODEV; + tlb_cb = kmalloc(sizeof(*tlb_cb), GFP_KERNEL); + if (!tlb_cb) { + r = -ENOMEM; + goto error_unlock; + } + + /* Vega20+XGMI where PTEs get inadvertently cached in L2 texture cache, + * heavy-weight flush TLB unconditionally. + */ + flush_tlb |= adev->gmc.xgmi.num_physical_nodes && + adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 0); + memset(¶ms, 0, sizeof(params)); params.adev = adev; params.vm = vm; @@ -1693,7 +803,7 @@ int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev, amdgpu_vm_eviction_lock(vm); if (vm->evicting) { r = -EBUSY; - goto error_unlock; + goto error_free; } if (!unlocked && !dma_fence_is_signaled(vm->last_unlocked)) { @@ -1706,7 +816,7 @@ int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev, r = vm->update_funcs->prepare(¶ms, resv, sync_mode); if (r) - goto error_unlock; + goto error_free; amdgpu_res_first(pages_addr ? NULL : res, offset, (last - start + 1) * AMDGPU_GPU_PAGE_SIZE, &cursor); @@ -1746,16 +856,15 @@ int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev, } } else if (flags & (AMDGPU_PTE_VALID | AMDGPU_PTE_PRT)) { - addr = bo_adev->vm_manager.vram_base_offset + - cursor.start; + addr = vram_base + cursor.start; } else { addr = 0; } tmp = start + num_entries; - r = amdgpu_vm_update_ptes(¶ms, start, tmp, addr, flags); + r = amdgpu_vm_ptes_update(¶ms, start, tmp, addr, flags); if (r) - goto error_unlock; + goto error_free; amdgpu_res_next(&cursor, num_entries * AMDGPU_GPU_PAGE_SIZE); start = tmp; @@ -1763,8 +872,21 @@ int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev, r = vm->update_funcs->commit(¶ms, fence); - if (table_freed) - *table_freed = *table_freed || params.table_freed; + if (flush_tlb || params.table_freed) { + tlb_cb->vm = vm; + if (fence && *fence && + !dma_fence_add_callback(*fence, &tlb_cb->cb, + amdgpu_vm_tlb_seq_cb)) { + dma_fence_put(vm->last_tlb_flush); + vm->last_tlb_flush = dma_fence_get(*fence); + } else { + amdgpu_vm_tlb_seq_cb(NULL, &tlb_cb->cb); + } + tlb_cb = NULL; + } + +error_free: + kfree(tlb_cb); error_unlock: amdgpu_vm_eviction_unlock(vm); @@ -1822,7 +944,6 @@ void amdgpu_vm_get_memory(struct amdgpu_vm *vm, uint64_t *vram_mem, * @adev: amdgpu_device pointer * @bo_va: requested BO and VM object * @clear: if true clear the entries - * @table_freed: return true if page table is freed * * Fill in the page table entries for @bo_va. * @@ -1830,7 +951,7 @@ void amdgpu_vm_get_memory(struct amdgpu_vm *vm, uint64_t *vram_mem, * 0 for success, -EINVAL for failure. */ int amdgpu_vm_bo_update(struct amdgpu_device *adev, struct amdgpu_bo_va *bo_va, - bool clear, bool *table_freed) + bool clear) { struct amdgpu_bo *bo = bo_va->base.bo; struct amdgpu_vm *vm = bo_va->base.vm; @@ -1838,9 +959,10 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev, struct amdgpu_bo_va *bo_va, dma_addr_t *pages_addr = NULL; struct ttm_resource *mem; struct dma_fence **last_update; + bool flush_tlb = clear; struct dma_resv *resv; + uint64_t vram_base; uint64_t flags; - struct amdgpu_device *bo_adev = adev; int r; if (clear || !bo) { @@ -1865,14 +987,18 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev, struct amdgpu_bo_va *bo_va, } if (bo) { + struct amdgpu_device *bo_adev; + flags = amdgpu_ttm_tt_pte_flags(adev, bo->tbo.ttm, mem); if (amdgpu_bo_encrypted(bo)) flags |= AMDGPU_PTE_TMZ; bo_adev = amdgpu_ttm_adev(bo->tbo.bdev); + vram_base = bo_adev->vm_manager.vram_base_offset; } else { flags = 0x0; + vram_base = 0; } if (clear || (bo && bo->tbo.base.resv == @@ -1882,7 +1008,7 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev, struct amdgpu_bo_va *bo_va, last_update = &bo_va->last_pt_update; if (!clear && bo_va->base.moved) { - bo_va->base.moved = false; + flush_tlb = true; list_splice_init(&bo_va->valids, &bo_va->invalids); } else if (bo_va->cleared != clear) { @@ -1905,11 +1031,11 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev, struct amdgpu_bo_va *bo_va, trace_amdgpu_vm_bo_update(mapping); - r = amdgpu_vm_bo_update_mapping(adev, bo_adev, vm, false, false, - resv, mapping->start, - mapping->last, update_flags, - mapping->offset, mem, - pages_addr, last_update, table_freed); + r = amdgpu_vm_update_range(adev, vm, false, false, flush_tlb, + resv, mapping->start, mapping->last, + update_flags, mapping->offset, + vram_base, mem, pages_addr, + last_update); if (r) return r; } @@ -1932,6 +1058,7 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev, struct amdgpu_bo_va *bo_va, list_splice_init(&bo_va->invalids, &bo_va->valids); bo_va->cleared = clear; + bo_va->base.moved = false; if (trace_amdgpu_vm_bo_mapping_enabled()) { list_for_each_entry(mapping, &bo_va->valids, list) @@ -2100,10 +1227,10 @@ int amdgpu_vm_clear_freed(struct amdgpu_device *adev, mapping->start < AMDGPU_GMC_HOLE_START) init_pte_value = AMDGPU_PTE_DEFAULT_ATC; - r = amdgpu_vm_bo_update_mapping(adev, adev, vm, false, false, - resv, mapping->start, - mapping->last, init_pte_value, - 0, NULL, NULL, &f, NULL); + r = amdgpu_vm_update_range(adev, vm, false, false, true, resv, + mapping->start, mapping->last, + init_pte_value, 0, 0, NULL, NULL, + &f); amdgpu_vm_free_mapping(adev, vm, mapping, f); if (r) { dma_fence_put(f); @@ -2145,7 +1272,7 @@ int amdgpu_vm_handle_moved(struct amdgpu_device *adev, list_for_each_entry_safe(bo_va, tmp, &vm->moved, base.vm_status) { /* Per VM BOs never need to bo cleared in the page tables */ - r = amdgpu_vm_bo_update(adev, bo_va, false, NULL); + r = amdgpu_vm_bo_update(adev, bo_va, false); if (r) return r; } @@ -2164,7 +1291,7 @@ int amdgpu_vm_handle_moved(struct amdgpu_device *adev, else clear = true; - r = amdgpu_vm_bo_update(adev, bo_va, clear, NULL); + r = amdgpu_vm_bo_update(adev, bo_va, clear); if (r) return r; @@ -2914,6 +2041,7 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm) vm->update_funcs = &amdgpu_vm_sdma_funcs; vm->last_update = NULL; vm->last_unlocked = dma_fence_get_stub(); + vm->last_tlb_flush = dma_fence_get_stub(); mutex_init(&vm->eviction_lock); vm->evicting = false; @@ -2933,7 +2061,7 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm) amdgpu_vm_bo_base_init(&vm->root, vm, root_bo); - r = amdgpu_vm_clear_bo(adev, vm, root, false); + r = amdgpu_vm_pt_clear(adev, vm, root, false); if (r) goto error_unreserve; @@ -2952,6 +2080,7 @@ error_free_root: vm->root.bo = NULL; error_free_delayed: + dma_fence_put(vm->last_tlb_flush); dma_fence_put(vm->last_unlocked); drm_sched_entity_destroy(&vm->delayed); @@ -2962,34 +2091,6 @@ error_free_immediate: } /** - * amdgpu_vm_check_clean_reserved - check if a VM is clean - * - * @adev: amdgpu_device pointer - * @vm: the VM to check - * - * check all entries of the root PD, if any subsequent PDs are allocated, - * it means there are page table creating and filling, and is no a clean - * VM - * - * Returns: - * 0 if this VM is clean - */ -static int amdgpu_vm_check_clean_reserved(struct amdgpu_device *adev, - struct amdgpu_vm *vm) -{ - enum amdgpu_vm_level root = adev->vm_manager.root_level; - unsigned int entries = amdgpu_vm_num_entries(adev, root); - unsigned int i = 0; - - for (i = 0; i < entries; i++) { - if (to_amdgpu_bo_vm(vm->root.bo)->entries[i].bo) - return -EINVAL; - } - - return 0; -} - -/** * amdgpu_vm_make_compute - Turn a GFX VM into a compute VM * * @adev: amdgpu_device pointer @@ -3018,17 +2119,17 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, struct amdgpu_vm *vm) return r; /* Sanity checks */ - r = amdgpu_vm_check_clean_reserved(adev, vm); - if (r) + if (!amdgpu_vm_pt_is_root_clean(adev, vm)) { + r = -EINVAL; goto unreserve_bo; + } /* Check if PD needs to be reinitialized and do it before * changing any other state, in case it fails. */ if (pte_support_ats != vm->pte_support_ats) { vm->pte_support_ats = pte_support_ats; - r = amdgpu_vm_clear_bo(adev, vm, - to_amdgpu_bo_vm(vm->root.bo), + r = amdgpu_vm_pt_clear(adev, vm, to_amdgpu_bo_vm(vm->root.bo), false); if (r) goto unreserve_bo; @@ -3096,6 +2197,7 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm) struct amdgpu_bo_va_mapping *mapping, *tmp; bool prt_fini_needed = !!adev->gmc.gmc_funcs->set_prt; struct amdgpu_bo *root; + unsigned long flags; int i; amdgpu_amdkfd_gpuvm_destroy_cb(adev, vm); @@ -3105,6 +2207,11 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm) amdgpu_vm_set_pasid(adev, vm, 0); dma_fence_wait(vm->last_unlocked, false); dma_fence_put(vm->last_unlocked); + dma_fence_wait(vm->last_tlb_flush, false); + /* Make sure that all fence callbacks have completed */ + spin_lock_irqsave(vm->last_tlb_flush->lock, flags); + spin_unlock_irqrestore(vm->last_tlb_flush->lock, flags); + dma_fence_put(vm->last_tlb_flush); list_for_each_entry_safe(mapping, tmp, &vm->freed, list) { if (mapping->flags & AMDGPU_PTE_PRT && prt_fini_needed) { @@ -3116,7 +2223,7 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm) amdgpu_vm_free_mapping(adev, vm, mapping, NULL); } - amdgpu_vm_free_pts(adev, vm, NULL); + amdgpu_vm_pt_free_root(adev, vm); amdgpu_bo_unreserve(root); amdgpu_bo_unref(&root); WARN_ON(vm->root.bo); @@ -3376,9 +2483,8 @@ bool amdgpu_vm_handle_fault(struct amdgpu_device *adev, u32 pasid, goto error_unlock; } - r = amdgpu_vm_bo_update_mapping(adev, adev, vm, true, false, NULL, addr, - addr, flags, value, NULL, NULL, NULL, - NULL); + r = amdgpu_vm_update_range(adev, vm, true, false, false, NULL, addr, + addr, flags, value, 0, NULL, NULL, NULL); if (r) goto error_unlock; |