GitHub-Mirror/Star64_linux
2
0
Fork 0
mirror of https://github.com/Fishwaldo/Star64_linux.git synced 2025-06-20 05:31:15 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

drm/amd/amdgpu: Code comments for the amdgpu_ttm.c driver. (v2)

Browse source 
NFC just comments.

(v2):  Updated based on feedback from Alex Deucher.

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This commit is contained in:
Tom St Denis 2018-05-09 14:22:29 -04:00 • committed by Alex Deucher
parent fa19a6e9d0
commit 50da51744f
1 changed files with 341 additions and 7 deletions
Download patch file Download diff file Expand all files Collapse all files

348
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
View file

@ -63,16 +63,44 @@ static void amdgpu_ttm_debugfs_fini(struct amdgpu_device *adev);
/* /*
* Global memory. * Global memory.
*/ */
/**
* amdgpu_ttm_mem_global_init - Initialize and acquire reference to
* memory object
*
* @ref: Object for initialization.
*
* This is called by drm_global_item_ref() when an object is being
* initialized.
*/
static int amdgpu_ttm_mem_global_init(struct drm_global_reference *ref) static int amdgpu_ttm_mem_global_init(struct drm_global_reference *ref)
{ {
return ttm_mem_global_init(ref->object); return ttm_mem_global_init(ref->object);
} }
/**
* amdgpu_ttm_mem_global_release - Drop reference to a memory object
*
* @ref: Object being removed
*
* This is called by drm_global_item_unref() when an object is being
* released.
*/
static void amdgpu_ttm_mem_global_release(struct drm_global_reference *ref) static void amdgpu_ttm_mem_global_release(struct drm_global_reference *ref)
{ {
ttm_mem_global_release(ref->object); ttm_mem_global_release(ref->object);
} }
/**
* amdgpu_ttm_global_init - Initialize global TTM memory reference
* structures.
*
* @adev: AMDGPU device for which the global structures need to be
* registered.
*
* This is called as part of the AMDGPU ttm init from amdgpu_ttm_init()
* during bring up.
*/
static int amdgpu_ttm_global_init(struct amdgpu_device *adev) static int amdgpu_ttm_global_init(struct amdgpu_device *adev)
{ {
struct drm_global_reference *global_ref; struct drm_global_reference *global_ref;
@ -80,7 +108,9 @@ static int amdgpu_ttm_global_init(struct amdgpu_device *adev)
struct drm_sched_rq *rq; struct drm_sched_rq *rq;
int r; int r;
/* ensure reference is false in case init fails */
adev->mman.mem_global_referenced = false; adev->mman.mem_global_referenced = false;
global_ref = &adev->mman.mem_global_ref; global_ref = &adev->mman.mem_global_ref;
global_ref->global_type = DRM_GLOBAL_TTM_MEM; global_ref->global_type = DRM_GLOBAL_TTM_MEM;
global_ref->size = sizeof(struct ttm_mem_global); global_ref->size = sizeof(struct ttm_mem_global);
@ -146,6 +176,18 @@ static int amdgpu_invalidate_caches(struct ttm_bo_device *bdev, uint32_t flags)
return 0; return 0;
} }
/**
* amdgpu_init_mem_type - Initialize a memory manager for a specific
* type of memory request.
*
* @bdev: The TTM BO device object (contains a reference to
* amdgpu_device)
* @type: The type of memory requested
* @man:
*
* This is called by ttm_bo_init_mm() when a buffer object is being
* initialized.
*/
static int amdgpu_init_mem_type(struct ttm_bo_device *bdev, uint32_t type, static int amdgpu_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
struct ttm_mem_type_manager *man) struct ttm_mem_type_manager *man)
{ {
@ -161,6 +203,7 @@ static int amdgpu_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
man->default_caching = TTM_PL_FLAG_CACHED; man->default_caching = TTM_PL_FLAG_CACHED;
break; break;
case TTM_PL_TT: case TTM_PL_TT:
/* GTT memory */
man->func = &amdgpu_gtt_mgr_func; man->func = &amdgpu_gtt_mgr_func;
man->gpu_offset = adev->gmc.gart_start; man->gpu_offset = adev->gmc.gart_start;
man->available_caching = TTM_PL_MASK_CACHING; man->available_caching = TTM_PL_MASK_CACHING;
@ -193,6 +236,14 @@ static int amdgpu_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
return 0; return 0;
} }
/**
* amdgpu_evict_flags - Compute placement flags
*
* @bo: The buffer object to evict
* @placement: Possible destination(s) for evicted BO
*
* Fill in placement data when ttm_bo_evict() is called
*/
static void amdgpu_evict_flags(struct ttm_buffer_object *bo, static void amdgpu_evict_flags(struct ttm_buffer_object *bo,
struct ttm_placement *placement) struct ttm_placement *placement)
{ {
@ -204,12 +255,14 @@ static void amdgpu_evict_flags(struct ttm_buffer_object *bo,
.flags = TTM_PL_MASK_CACHING | TTM_PL_FLAG_SYSTEM .flags = TTM_PL_MASK_CACHING | TTM_PL_FLAG_SYSTEM
}; };
/* Don't handle scatter gather BOs */
if (bo->type == ttm_bo_type_sg) { if (bo->type == ttm_bo_type_sg) {
placement->num_placement = 0; placement->num_placement = 0;
placement->num_busy_placement = 0; placement->num_busy_placement = 0;
return; return;
} }
/* Object isn't an AMDGPU object so ignore */
if (!amdgpu_ttm_bo_is_amdgpu_bo(bo)) { if (!amdgpu_ttm_bo_is_amdgpu_bo(bo)) {
placement->placement = &placements; placement->placement = &placements;
placement->busy_placement = &placements; placement->busy_placement = &placements;
@ -217,10 +270,12 @@ static void amdgpu_evict_flags(struct ttm_buffer_object *bo,
placement->num_busy_placement = 1; placement->num_busy_placement = 1;
return; return;
} }
abo = ttm_to_amdgpu_bo(bo); abo = ttm_to_amdgpu_bo(bo);
switch (bo->mem.mem_type) { switch (bo->mem.mem_type) {
case TTM_PL_VRAM: case TTM_PL_VRAM:
if (!adev->mman.buffer_funcs_enabled) { if (!adev->mman.buffer_funcs_enabled) {
/* Move to system memory */
amdgpu_ttm_placement_from_domain(abo, AMDGPU_GEM_DOMAIN_CPU); amdgpu_ttm_placement_from_domain(abo, AMDGPU_GEM_DOMAIN_CPU);
} else if (adev->gmc.visible_vram_size < adev->gmc.real_vram_size && } else if (adev->gmc.visible_vram_size < adev->gmc.real_vram_size &&
!(abo->flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED) && !(abo->flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED) &&
@ -238,6 +293,7 @@ static void amdgpu_evict_flags(struct ttm_buffer_object *bo,
abo->placement.busy_placement = &abo->placements[1]; abo->placement.busy_placement = &abo->placements[1];
abo->placement.num_busy_placement = 1; abo->placement.num_busy_placement = 1;
} else { } else {
/* Move to GTT memory */
amdgpu_ttm_placement_from_domain(abo, AMDGPU_GEM_DOMAIN_GTT); amdgpu_ttm_placement_from_domain(abo, AMDGPU_GEM_DOMAIN_GTT);
} }
break; break;
@ -248,6 +304,15 @@ static void amdgpu_evict_flags(struct ttm_buffer_object *bo,
*placement = abo->placement; *placement = abo->placement;
} }
/**
* amdgpu_verify_access - Verify access for a mmap call
*
* @bo: The buffer object to map
* @filp: The file pointer from the process performing the mmap
*
* This is called by ttm_bo_mmap() to verify whether a process
* has the right to mmap a BO to their process space.
*/
static int amdgpu_verify_access(struct ttm_buffer_object *bo, struct file *filp) static int amdgpu_verify_access(struct ttm_buffer_object *bo, struct file *filp)
{ {
struct amdgpu_bo *abo = ttm_to_amdgpu_bo(bo); struct amdgpu_bo *abo = ttm_to_amdgpu_bo(bo);
@ -265,6 +330,15 @@ static int amdgpu_verify_access(struct ttm_buffer_object *bo, struct file *filp)
filp->private_data); filp->private_data);
} }
/**
* amdgpu_move_null - Register memory for a buffer object
*
* @bo: The bo to assign the memory to
* @new_mem: The memory to be assigned.
*
* Assign the memory from new_mem to the memory of the buffer object
* bo.
*/
static void amdgpu_move_null(struct ttm_buffer_object *bo, static void amdgpu_move_null(struct ttm_buffer_object *bo,
struct ttm_mem_reg *new_mem) struct ttm_mem_reg *new_mem)
{ {
@ -275,6 +349,10 @@ static void amdgpu_move_null(struct ttm_buffer_object *bo,
new_mem->mm_node = NULL; new_mem->mm_node = NULL;
} }
/**
* amdgpu_mm_node_addr - Compute the GPU relative offset of a GTT
* buffer.
*/
static uint64_t amdgpu_mm_node_addr(struct ttm_buffer_object *bo, static uint64_t amdgpu_mm_node_addr(struct ttm_buffer_object *bo,
struct drm_mm_node *mm_node, struct drm_mm_node *mm_node,
struct ttm_mem_reg *mem) struct ttm_mem_reg *mem)
@ -289,9 +367,10 @@ static uint64_t amdgpu_mm_node_addr(struct ttm_buffer_object *bo,
} }
/** /**
* amdgpu_find_mm_node - Helper function finds the drm_mm_node * amdgpu_find_mm_node - Helper function finds the drm_mm_node
* corresponding to @offset. It also modifies the offset to be * corresponding to @offset. It also modifies
* within the drm_mm_node returned * the offset to be within the drm_mm_node
* returned
*/ */
static struct drm_mm_node *amdgpu_find_mm_node(struct ttm_mem_reg *mem, static struct drm_mm_node *amdgpu_find_mm_node(struct ttm_mem_reg *mem,
unsigned long *offset) unsigned long *offset)
@ -430,7 +509,12 @@ error:
return r; return r;
} }
/**
* amdgpu_move_blit - Copy an entire buffer to another buffer
*
* This is a helper called by amdgpu_bo_move() and
* amdgpu_move_vram_ram() to help move buffers to and from VRAM.
*/
static int amdgpu_move_blit(struct ttm_buffer_object *bo, static int amdgpu_move_blit(struct ttm_buffer_object *bo,
bool evict, bool no_wait_gpu, bool evict, bool no_wait_gpu,
struct ttm_mem_reg *new_mem, struct ttm_mem_reg *new_mem,
@ -465,6 +549,11 @@ error:
return r; return r;
} }
/**
* amdgpu_move_vram_ram - Copy VRAM buffer to RAM buffer
*
* Called by amdgpu_bo_move().
*/
static int amdgpu_move_vram_ram(struct ttm_buffer_object *bo, bool evict, static int amdgpu_move_vram_ram(struct ttm_buffer_object *bo, bool evict,
struct ttm_operation_ctx *ctx, struct ttm_operation_ctx *ctx,
struct ttm_mem_reg *new_mem) struct ttm_mem_reg *new_mem)
@ -477,6 +566,8 @@ static int amdgpu_move_vram_ram(struct ttm_buffer_object *bo, bool evict,
int r; int r;
adev = amdgpu_ttm_adev(bo->bdev); adev = amdgpu_ttm_adev(bo->bdev);
/* create space/pages for new_mem in GTT space */
tmp_mem = *new_mem; tmp_mem = *new_mem;
tmp_mem.mm_node = NULL; tmp_mem.mm_node = NULL;
placement.num_placement = 1; placement.num_placement = 1;
@ -491,25 +582,36 @@ static int amdgpu_move_vram_ram(struct ttm_buffer_object *bo, bool evict,
return r; return r;
} }
/* set caching flags */
r = ttm_tt_set_placement_caching(bo->ttm, tmp_mem.placement); r = ttm_tt_set_placement_caching(bo->ttm, tmp_mem.placement);
if (unlikely(r)) { if (unlikely(r)) {
goto out_cleanup; goto out_cleanup;
} }
/* Bind the memory to the GTT space */
r = ttm_tt_bind(bo->ttm, &tmp_mem, ctx); r = ttm_tt_bind(bo->ttm, &tmp_mem, ctx);
if (unlikely(r)) { if (unlikely(r)) {
goto out_cleanup; goto out_cleanup;
} }
/* blit VRAM to GTT */
r = amdgpu_move_blit(bo, true, ctx->no_wait_gpu, &tmp_mem, old_mem); r = amdgpu_move_blit(bo, true, ctx->no_wait_gpu, &tmp_mem, old_mem);
if (unlikely(r)) { if (unlikely(r)) {
goto out_cleanup; goto out_cleanup;
} }
/* move BO (in tmp_mem) to new_mem */
r = ttm_bo_move_ttm(bo, ctx, new_mem); r = ttm_bo_move_ttm(bo, ctx, new_mem);
out_cleanup: out_cleanup:
ttm_bo_mem_put(bo, &tmp_mem); ttm_bo_mem_put(bo, &tmp_mem);
return r; return r;
} }
/**
* amdgpu_move_ram_vram - Copy buffer from RAM to VRAM
*
* Called by amdgpu_bo_move().
*/
static int amdgpu_move_ram_vram(struct ttm_buffer_object *bo, bool evict, static int amdgpu_move_ram_vram(struct ttm_buffer_object *bo, bool evict,
struct ttm_operation_ctx *ctx, struct ttm_operation_ctx *ctx,
struct ttm_mem_reg *new_mem) struct ttm_mem_reg *new_mem)
@ -522,6 +624,8 @@ static int amdgpu_move_ram_vram(struct ttm_buffer_object *bo, bool evict,
int r; int r;
adev = amdgpu_ttm_adev(bo->bdev); adev = amdgpu_ttm_adev(bo->bdev);
/* make space in GTT for old_mem buffer */
tmp_mem = *new_mem; tmp_mem = *new_mem;
tmp_mem.mm_node = NULL; tmp_mem.mm_node = NULL;
placement.num_placement = 1; placement.num_placement = 1;
@ -535,10 +639,14 @@ static int amdgpu_move_ram_vram(struct ttm_buffer_object *bo, bool evict,
if (unlikely(r)) { if (unlikely(r)) {
return r; return r;
} }
/* move/bind old memory to GTT space */
r = ttm_bo_move_ttm(bo, ctx, &tmp_mem); r = ttm_bo_move_ttm(bo, ctx, &tmp_mem);
if (unlikely(r)) { if (unlikely(r)) {
goto out_cleanup; goto out_cleanup;
} }
/* copy to VRAM */
r = amdgpu_move_blit(bo, true, ctx->no_wait_gpu, new_mem, old_mem); r = amdgpu_move_blit(bo, true, ctx->no_wait_gpu, new_mem, old_mem);
if (unlikely(r)) { if (unlikely(r)) {
goto out_cleanup; goto out_cleanup;
@ -548,6 +656,11 @@ out_cleanup:
return r; return r;
} }
/**
* amdgpu_bo_move - Move a buffer object to a new memory location
*
* Called by ttm_bo_handle_move_mem()
*/
static int amdgpu_bo_move(struct ttm_buffer_object *bo, bool evict, static int amdgpu_bo_move(struct ttm_buffer_object *bo, bool evict,
struct ttm_operation_ctx *ctx, struct ttm_operation_ctx *ctx,
struct ttm_mem_reg *new_mem) struct ttm_mem_reg *new_mem)
@ -613,6 +726,11 @@ memcpy:
return 0; return 0;
} }
/**
* amdgpu_ttm_io_mem_reserve - Reserve a block of memory during a fault
*
* Called by ttm_mem_io_reserve() ultimately via ttm_bo_vm_fault()
*/
static int amdgpu_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem) static int amdgpu_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
{ {
struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type]; struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type];
@ -690,6 +808,14 @@ struct amdgpu_ttm_tt {
uint32_t last_set_pages; uint32_t last_set_pages;
}; };
/**
* amdgpu_ttm_tt_get_user_pages - Pin pages of memory pointed to
* by a USERPTR pointer to memory
*
* Called by amdgpu_gem_userptr_ioctl() and amdgpu_cs_parser_bos().
* This provides a wrapper around the get_user_pages() call to provide
* device accessible pages that back user memory.
*/
int amdgpu_ttm_tt_get_user_pages(struct ttm_tt *ttm, struct page **pages) int amdgpu_ttm_tt_get_user_pages(struct ttm_tt *ttm, struct page **pages)
{ {
struct amdgpu_ttm_tt *gtt = (void *)ttm; struct amdgpu_ttm_tt *gtt = (void *)ttm;
@ -719,6 +845,7 @@ int amdgpu_ttm_tt_get_user_pages(struct ttm_tt *ttm, struct page **pages)
} }
} }
/* loop enough times using contiguous pages of memory */
do { do {
unsigned num_pages = ttm->num_pages - pinned; unsigned num_pages = ttm->num_pages - pinned;
uint64_t userptr = gtt->userptr + pinned * PAGE_SIZE; uint64_t userptr = gtt->userptr + pinned * PAGE_SIZE;
@ -757,6 +884,14 @@ release_pages:
return r; return r;
} }
/**
* amdgpu_ttm_tt_set_user_pages - Copy pages in, putting old pages
* as necessary.
*
* Called by amdgpu_cs_list_validate(). This creates the page list
* that backs user memory and will ultimately be mapped into the device
* address space.
*/
void amdgpu_ttm_tt_set_user_pages(struct ttm_tt *ttm, struct page **pages) void amdgpu_ttm_tt_set_user_pages(struct ttm_tt *ttm, struct page **pages)
{ {
struct amdgpu_ttm_tt *gtt = (void *)ttm; struct amdgpu_ttm_tt *gtt = (void *)ttm;
@ -771,6 +906,11 @@ void amdgpu_ttm_tt_set_user_pages(struct ttm_tt *ttm, struct page **pages)
} }
} }
/**
* amdgpu_ttm_tt_mark_user_page - Mark pages as dirty
*
* Called while unpinning userptr pages
*/
void amdgpu_ttm_tt_mark_user_pages(struct ttm_tt *ttm) void amdgpu_ttm_tt_mark_user_pages(struct ttm_tt *ttm)
{ {
struct amdgpu_ttm_tt *gtt = (void *)ttm; struct amdgpu_ttm_tt *gtt = (void *)ttm;
@ -789,7 +929,12 @@ void amdgpu_ttm_tt_mark_user_pages(struct ttm_tt *ttm)
} }
} }
/* prepare the sg table with the user pages */ /**
* amdgpu_ttm_tt_pin_userptr - prepare the sg table with the
* user pages
*
* Called by amdgpu_ttm_backend_bind()
**/
static int amdgpu_ttm_tt_pin_userptr(struct ttm_tt *ttm) static int amdgpu_ttm_tt_pin_userptr(struct ttm_tt *ttm)
{ {
struct amdgpu_device *adev = amdgpu_ttm_adev(ttm->bdev); struct amdgpu_device *adev = amdgpu_ttm_adev(ttm->bdev);
@ -801,17 +946,20 @@ static int amdgpu_ttm_tt_pin_userptr(struct ttm_tt *ttm)
enum dma_data_direction direction = write ? enum dma_data_direction direction = write ?
DMA_BIDIRECTIONAL : DMA_TO_DEVICE; DMA_BIDIRECTIONAL : DMA_TO_DEVICE;
/* Allocate an SG array and squash pages into it */
r = sg_alloc_table_from_pages(ttm->sg, ttm->pages, ttm->num_pages, 0, r = sg_alloc_table_from_pages(ttm->sg, ttm->pages, ttm->num_pages, 0,
ttm->num_pages << PAGE_SHIFT, ttm->num_pages << PAGE_SHIFT,
GFP_KERNEL); GFP_KERNEL);
if (r) if (r)
goto release_sg; goto release_sg;
/* Map SG to device */
r = -ENOMEM; r = -ENOMEM;
nents = dma_map_sg(adev->dev, ttm->sg->sgl, ttm->sg->nents, direction); nents = dma_map_sg(adev->dev, ttm->sg->sgl, ttm->sg->nents, direction);
if (nents != ttm->sg->nents) if (nents != ttm->sg->nents)
goto release_sg; goto release_sg;
/* convert SG to linear array of pages and dma addresses */
drm_prime_sg_to_page_addr_arrays(ttm->sg, ttm->pages, drm_prime_sg_to_page_addr_arrays(ttm->sg, ttm->pages,
gtt->ttm.dma_address, ttm->num_pages); gtt->ttm.dma_address, ttm->num_pages);
@ -822,6 +970,9 @@ release_sg:
return r; return r;
} }
/**
* amdgpu_ttm_tt_unpin_userptr - Unpin and unmap userptr pages
*/
static void amdgpu_ttm_tt_unpin_userptr(struct ttm_tt *ttm) static void amdgpu_ttm_tt_unpin_userptr(struct ttm_tt *ttm)
{ {
struct amdgpu_device *adev = amdgpu_ttm_adev(ttm->bdev); struct amdgpu_device *adev = amdgpu_ttm_adev(ttm->bdev);
@ -835,9 +986,10 @@ static void amdgpu_ttm_tt_unpin_userptr(struct ttm_tt *ttm)
if (!ttm->sg->sgl) if (!ttm->sg->sgl)
return; return;
/* free the sg table and pages again */ /* unmap the pages mapped to the device */
dma_unmap_sg(adev->dev, ttm->sg->sgl, ttm->sg->nents, direction); dma_unmap_sg(adev->dev, ttm->sg->sgl, ttm->sg->nents, direction);
/* mark the pages as dirty */
amdgpu_ttm_tt_mark_user_pages(ttm); amdgpu_ttm_tt_mark_user_pages(ttm);
sg_free_table(ttm->sg); sg_free_table(ttm->sg);
@ -882,6 +1034,12 @@ gart_bind_fail:
return r; return r;
} }
/**
* amdgpu_ttm_backend_bind - Bind GTT memory
*
* Called by ttm_tt_bind() on behalf of ttm_bo_handle_move_mem().
* This handles binding GTT memory to the device address space.
*/
static int amdgpu_ttm_backend_bind(struct ttm_tt *ttm, static int amdgpu_ttm_backend_bind(struct ttm_tt *ttm,
struct ttm_mem_reg *bo_mem) struct ttm_mem_reg *bo_mem)
{ {
@ -912,7 +1070,10 @@ static int amdgpu_ttm_backend_bind(struct ttm_tt *ttm,
return 0; return 0;
} }
/* compute PTE flags relevant to this BO memory */
flags = amdgpu_ttm_tt_pte_flags(adev, ttm, bo_mem); flags = amdgpu_ttm_tt_pte_flags(adev, ttm, bo_mem);
/* bind pages into GART page tables */
gtt->offset = (u64)bo_mem->start << PAGE_SHIFT; gtt->offset = (u64)bo_mem->start << PAGE_SHIFT;
r = amdgpu_gart_bind(adev, gtt->offset, ttm->num_pages, r = amdgpu_gart_bind(adev, gtt->offset, ttm->num_pages,
ttm->pages, gtt->ttm.dma_address, flags); ttm->pages, gtt->ttm.dma_address, flags);
@ -923,6 +1084,9 @@ static int amdgpu_ttm_backend_bind(struct ttm_tt *ttm,
return r; return r;
} }
/**
* amdgpu_ttm_alloc_gart - Allocate GART memory for buffer object
*/
int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo) int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo)
{ {
struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev); struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
@ -938,6 +1102,7 @@ int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo)
amdgpu_gtt_mgr_has_gart_addr(&bo->mem)) amdgpu_gtt_mgr_has_gart_addr(&bo->mem))
return 0; return 0;
/* allocate GTT space */
tmp = bo->mem; tmp = bo->mem;
tmp.mm_node = NULL; tmp.mm_node = NULL;
placement.num_placement = 1; placement.num_placement = 1;
@ -953,7 +1118,10 @@ int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo)
if (unlikely(r)) if (unlikely(r))
return r; return r;
/* compute PTE flags for this buffer object */
flags = amdgpu_ttm_tt_pte_flags(adev, bo->ttm, &tmp); flags = amdgpu_ttm_tt_pte_flags(adev, bo->ttm, &tmp);
/* Bind pages */
gtt->offset = (u64)tmp.start << PAGE_SHIFT; gtt->offset = (u64)tmp.start << PAGE_SHIFT;
r = amdgpu_ttm_gart_bind(adev, bo, flags); r = amdgpu_ttm_gart_bind(adev, bo, flags);
if (unlikely(r)) { if (unlikely(r)) {
@ -969,6 +1137,12 @@ int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo)
return 0; return 0;
} }
/**
* amdgpu_ttm_recover_gart - Rebind GTT pages
*
* Called by amdgpu_gtt_mgr_recover() from amdgpu_device_reset() to
* rebind GTT pages during a GPU reset.
*/
int amdgpu_ttm_recover_gart(struct ttm_buffer_object *tbo) int amdgpu_ttm_recover_gart(struct ttm_buffer_object *tbo)
{ {
struct amdgpu_device *adev = amdgpu_ttm_adev(tbo->bdev); struct amdgpu_device *adev = amdgpu_ttm_adev(tbo->bdev);
@ -984,12 +1158,19 @@ int amdgpu_ttm_recover_gart(struct ttm_buffer_object *tbo)
return r; return r;
} }
/**
* amdgpu_ttm_backend_unbind - Unbind GTT mapped pages
*
* Called by ttm_tt_unbind() on behalf of ttm_bo_move_ttm() and
* ttm_tt_destroy().
*/
static int amdgpu_ttm_backend_unbind(struct ttm_tt *ttm) static int amdgpu_ttm_backend_unbind(struct ttm_tt *ttm)
{ {
struct amdgpu_device *adev = amdgpu_ttm_adev(ttm->bdev); struct amdgpu_device *adev = amdgpu_ttm_adev(ttm->bdev);
struct amdgpu_ttm_tt *gtt = (void *)ttm; struct amdgpu_ttm_tt *gtt = (void *)ttm;
int r; int r;
/* if the pages have userptr pinning then clear that first */
if (gtt->userptr) if (gtt->userptr)
amdgpu_ttm_tt_unpin_userptr(ttm); amdgpu_ttm_tt_unpin_userptr(ttm);
@ -1021,6 +1202,13 @@ static struct ttm_backend_func amdgpu_backend_func = {
.destroy = &amdgpu_ttm_backend_destroy, .destroy = &amdgpu_ttm_backend_destroy,
}; };
/**
* amdgpu_ttm_tt_create - Create a ttm_tt object for a given BO
*
* @bo: The buffer object to create a GTT ttm_tt object around
*
* Called by ttm_tt_create().
*/
static struct ttm_tt *amdgpu_ttm_tt_create(struct ttm_buffer_object *bo, static struct ttm_tt *amdgpu_ttm_tt_create(struct ttm_buffer_object *bo,
uint32_t page_flags) uint32_t page_flags)
{ {
@ -1034,6 +1222,8 @@ static struct ttm_tt *amdgpu_ttm_tt_create(struct ttm_buffer_object *bo,
return NULL; return NULL;
} }
gtt->ttm.ttm.func = &amdgpu_backend_func; gtt->ttm.ttm.func = &amdgpu_backend_func;
/* allocate space for the uninitialized page entries */
if (ttm_sg_tt_init(&gtt->ttm, bo, page_flags)) { if (ttm_sg_tt_init(&gtt->ttm, bo, page_flags)) {
kfree(gtt); kfree(gtt);
return NULL; return NULL;
@ -1041,6 +1231,12 @@ static struct ttm_tt *amdgpu_ttm_tt_create(struct ttm_buffer_object *bo,
return &gtt->ttm.ttm; return &gtt->ttm.ttm;
} }
/**
* amdgpu_ttm_tt_populate - Map GTT pages visible to the device
*
* Map the pages of a ttm_tt object to an address space visible
* to the underlying device.
*/
static int amdgpu_ttm_tt_populate(struct ttm_tt *ttm, static int amdgpu_ttm_tt_populate(struct ttm_tt *ttm,
struct ttm_operation_ctx *ctx) struct ttm_operation_ctx *ctx)
{ {
@ -1048,6 +1244,7 @@ static int amdgpu_ttm_tt_populate(struct ttm_tt *ttm,
struct amdgpu_ttm_tt *gtt = (void *)ttm; struct amdgpu_ttm_tt *gtt = (void *)ttm;
bool slave = !!(ttm->page_flags & TTM_PAGE_FLAG_SG); bool slave = !!(ttm->page_flags & TTM_PAGE_FLAG_SG);
/* user pages are bound by amdgpu_ttm_tt_pin_userptr() */
if (gtt && gtt->userptr) { if (gtt && gtt->userptr) {
ttm->sg = kzalloc(sizeof(struct sg_table), GFP_KERNEL); ttm->sg = kzalloc(sizeof(struct sg_table), GFP_KERNEL);
if (!ttm->sg) if (!ttm->sg)
@ -1072,9 +1269,17 @@ static int amdgpu_ttm_tt_populate(struct ttm_tt *ttm,
} }
#endif #endif
/* fall back to generic helper to populate the page array
* and map them to the device */
return ttm_populate_and_map_pages(adev->dev, &gtt->ttm, ctx); return ttm_populate_and_map_pages(adev->dev, &gtt->ttm, ctx);
} }
/**
* amdgpu_ttm_tt_unpopulate - unmap GTT pages and unpopulate page arrays
*
* Unmaps pages of a ttm_tt object from the device address space and
* unpopulates the page array backing it.
*/
static void amdgpu_ttm_tt_unpopulate(struct ttm_tt *ttm) static void amdgpu_ttm_tt_unpopulate(struct ttm_tt *ttm)
{ {
struct amdgpu_device *adev; struct amdgpu_device *adev;
@ -1100,9 +1305,21 @@ static void amdgpu_ttm_tt_unpopulate(struct ttm_tt *ttm)
} }
#endif #endif
/* fall back to generic helper to unmap and unpopulate array */
ttm_unmap_and_unpopulate_pages(adev->dev, &gtt->ttm); ttm_unmap_and_unpopulate_pages(adev->dev, &gtt->ttm);
} }
/**
* amdgpu_ttm_tt_set_userptr - Initialize userptr GTT ttm_tt
* for the current task
*
* @ttm: The ttm_tt object to bind this userptr object to
* @addr: The address in the current tasks VM space to use
* @flags: Requirements of userptr object.
*
* Called by amdgpu_gem_userptr_ioctl() to bind userptr pages
* to current task
*/
int amdgpu_ttm_tt_set_userptr(struct ttm_tt *ttm, uint64_t addr, int amdgpu_ttm_tt_set_userptr(struct ttm_tt *ttm, uint64_t addr,
uint32_t flags) uint32_t flags)
{ {
@ -1127,6 +1344,9 @@ int amdgpu_ttm_tt_set_userptr(struct ttm_tt *ttm, uint64_t addr,
return 0; return 0;
} }
/**
* amdgpu_ttm_tt_get_usermm - Return memory manager for ttm_tt object
*/
struct mm_struct *amdgpu_ttm_tt_get_usermm(struct ttm_tt *ttm) struct mm_struct *amdgpu_ttm_tt_get_usermm(struct ttm_tt *ttm)
{ {
struct amdgpu_ttm_tt *gtt = (void *)ttm; struct amdgpu_ttm_tt *gtt = (void *)ttm;
@ -1140,6 +1360,12 @@ struct mm_struct *amdgpu_ttm_tt_get_usermm(struct ttm_tt *ttm)
return gtt->usertask->mm; return gtt->usertask->mm;
} }
/**
* amdgpu_ttm_tt_affect_userptr - Determine if a ttm_tt object lays
* inside an address range for the
* current task.
*
*/
bool amdgpu_ttm_tt_affect_userptr(struct ttm_tt *ttm, unsigned long start, bool amdgpu_ttm_tt_affect_userptr(struct ttm_tt *ttm, unsigned long start,
unsigned long end) unsigned long end)
{ {
@ -1150,10 +1376,16 @@ bool amdgpu_ttm_tt_affect_userptr(struct ttm_tt *ttm, unsigned long start,
if (gtt == NULL || !gtt->userptr) if (gtt == NULL || !gtt->userptr)
return false; return false;
/* Return false if no part of the ttm_tt object lies within
* the range
*/
size = (unsigned long)gtt->ttm.ttm.num_pages * PAGE_SIZE; size = (unsigned long)gtt->ttm.ttm.num_pages * PAGE_SIZE;
if (gtt->userptr > end || gtt->userptr + size <= start) if (gtt->userptr > end || gtt->userptr + size <= start)
return false; return false;
/* Search the lists of tasks that hold this mapping and see
* if current is one of them. If it is return false.
*/
spin_lock(&gtt->guptasklock); spin_lock(&gtt->guptasklock);
list_for_each_entry(entry, &gtt->guptasks, list) { list_for_each_entry(entry, &gtt->guptasks, list) {
if (entry->task == current) { if (entry->task == current) {
@ -1168,6 +1400,10 @@ bool amdgpu_ttm_tt_affect_userptr(struct ttm_tt *ttm, unsigned long start,
return true; return true;
} }
/**
* amdgpu_ttm_tt_userptr_invalidated - Has the ttm_tt object been
* invalidated?
*/
bool amdgpu_ttm_tt_userptr_invalidated(struct ttm_tt *ttm, bool amdgpu_ttm_tt_userptr_invalidated(struct ttm_tt *ttm,
int *last_invalidated) int *last_invalidated)
{ {
@ -1178,6 +1414,12 @@ bool amdgpu_ttm_tt_userptr_invalidated(struct ttm_tt *ttm,
return prev_invalidated != *last_invalidated; return prev_invalidated != *last_invalidated;
} }
/**
* amdgpu_ttm_tt_userptr_needs_pages - Have the pages backing this
* ttm_tt object been invalidated
* since the last time they've
* been set?
*/
bool amdgpu_ttm_tt_userptr_needs_pages(struct ttm_tt *ttm) bool amdgpu_ttm_tt_userptr_needs_pages(struct ttm_tt *ttm)
{ {
struct amdgpu_ttm_tt *gtt = (void *)ttm; struct amdgpu_ttm_tt *gtt = (void *)ttm;
@ -1188,6 +1430,9 @@ bool amdgpu_ttm_tt_userptr_needs_pages(struct ttm_tt *ttm)
return atomic_read(&gtt->mmu_invalidations) != gtt->last_set_pages; return atomic_read(&gtt->mmu_invalidations) != gtt->last_set_pages;
} }
/**
* amdgpu_ttm_tt_is_readonly - Is the ttm_tt object read only?
*/
bool amdgpu_ttm_tt_is_readonly(struct ttm_tt *ttm) bool amdgpu_ttm_tt_is_readonly(struct ttm_tt *ttm)
{ {
struct amdgpu_ttm_tt *gtt = (void *)ttm; struct amdgpu_ttm_tt *gtt = (void *)ttm;
@ -1198,6 +1443,12 @@ bool amdgpu_ttm_tt_is_readonly(struct ttm_tt *ttm)
return !!(gtt->userflags & AMDGPU_GEM_USERPTR_READONLY); return !!(gtt->userflags & AMDGPU_GEM_USERPTR_READONLY);
} }
/**
* amdgpu_ttm_tt_pte_flags - Compute PTE flags for ttm_tt object
*
* @ttm: The ttm_tt object to compute the flags for
* @mem: The memory registry backing this ttm_tt object
*/
uint64_t amdgpu_ttm_tt_pte_flags(struct amdgpu_device *adev, struct ttm_tt *ttm, uint64_t amdgpu_ttm_tt_pte_flags(struct amdgpu_device *adev, struct ttm_tt *ttm,
struct ttm_mem_reg *mem) struct ttm_mem_reg *mem)
{ {
@ -1222,6 +1473,16 @@ uint64_t amdgpu_ttm_tt_pte_flags(struct amdgpu_device *adev, struct ttm_tt *ttm,
return flags; return flags;
} }
/**
* amdgpu_ttm_bo_eviction_valuable - Check to see if we can evict
* a buffer object.
*
* Return true if eviction is sensible. Called by
* ttm_mem_evict_first() on behalf of ttm_bo_mem_force_space()
* which tries to evict buffer objects until it can find space
* for a new object and by ttm_bo_force_list_clean() which is
* used to clean out a memory space.
*/
static bool amdgpu_ttm_bo_eviction_valuable(struct ttm_buffer_object *bo, static bool amdgpu_ttm_bo_eviction_valuable(struct ttm_buffer_object *bo,
const struct ttm_place *place) const struct ttm_place *place)
{ {
@ -1268,6 +1529,19 @@ static bool amdgpu_ttm_bo_eviction_valuable(struct ttm_buffer_object *bo,
return ttm_bo_eviction_valuable(bo, place); return ttm_bo_eviction_valuable(bo, place);
} }
/**
* amdgpu_ttm_access_memory - Read or Write memory that backs a
* buffer object.
*
* @bo: The buffer object to read/write
* @offset: Offset into buffer object
* @buf: Secondary buffer to write/read from
* @len: Length in bytes of access
* @write: true if writing
*
* This is used to access VRAM that backs a buffer object via MMIO
* access for debugging purposes.
*/
static int amdgpu_ttm_access_memory(struct ttm_buffer_object *bo, static int amdgpu_ttm_access_memory(struct ttm_buffer_object *bo,
unsigned long offset, unsigned long offset,
void *buf, int len, int write) void *buf, int len, int write)
@ -1444,13 +1718,22 @@ error_create:
adev->fw_vram_usage.reserved_bo = NULL; adev->fw_vram_usage.reserved_bo = NULL;
return r; return r;
} }
/**
* amdgpu_ttm_init - Init the memory management (ttm) as well as
* various gtt/vram related fields.
*
* This initializes all of the memory space pools that the TTM layer
* will need such as the GTT space (system memory mapped to the device),
* VRAM (on-board memory), and on-chip memories (GDS, GWS, OA) which
* can be mapped per VMID.
*/
int amdgpu_ttm_init(struct amdgpu_device *adev) int amdgpu_ttm_init(struct amdgpu_device *adev)
{ {
uint64_t gtt_size; uint64_t gtt_size;
int r; int r;
u64 vis_vram_limit; u64 vis_vram_limit;
/* initialize global references for vram/gtt */
r = amdgpu_ttm_global_init(adev); r = amdgpu_ttm_global_init(adev);
if (r) { if (r) {
return r; return r;
@ -1471,6 +1754,7 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
/* We opt to avoid OOM on system pages allocations */ /* We opt to avoid OOM on system pages allocations */
adev->mman.bdev.no_retry = true; adev->mman.bdev.no_retry = true;
/* Initialize VRAM pool with all of VRAM divided into pages */
r = ttm_bo_init_mm(&adev->mman.bdev, TTM_PL_VRAM, r = ttm_bo_init_mm(&adev->mman.bdev, TTM_PL_VRAM,
adev->gmc.real_vram_size >> PAGE_SHIFT); adev->gmc.real_vram_size >> PAGE_SHIFT);
if (r) { if (r) {
@ -1500,6 +1784,10 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
return r; return r;
} }
/* allocate memory as required for VGA
* This is used for VGA emulation and pre-OS scanout buffers to
* avoid display artifacts while transitioning between pre-OS
* and driver. */
if (adev->gmc.stolen_size) { if (adev->gmc.stolen_size) {
r = amdgpu_bo_create_kernel(adev, adev->gmc.stolen_size, PAGE_SIZE, r = amdgpu_bo_create_kernel(adev, adev->gmc.stolen_size, PAGE_SIZE,
AMDGPU_GEM_DOMAIN_VRAM, AMDGPU_GEM_DOMAIN_VRAM,
@ -1511,6 +1799,8 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
DRM_INFO("amdgpu: %uM of VRAM memory ready\n", DRM_INFO("amdgpu: %uM of VRAM memory ready\n",
(unsigned) (adev->gmc.real_vram_size / (1024 * 1024))); (unsigned) (adev->gmc.real_vram_size / (1024 * 1024)));
/* Compute GTT size, either bsaed on 3/4th the size of RAM size
* or whatever the user passed on module init */
if (amdgpu_gtt_size == -1) { if (amdgpu_gtt_size == -1) {
struct sysinfo si; struct sysinfo si;
@ -1521,6 +1811,8 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
} }
else else
gtt_size = (uint64_t)amdgpu_gtt_size << 20; gtt_size = (uint64_t)amdgpu_gtt_size << 20;
/* Initialize GTT memory pool */
r = ttm_bo_init_mm(&adev->mman.bdev, TTM_PL_TT, gtt_size >> PAGE_SHIFT); r = ttm_bo_init_mm(&adev->mman.bdev, TTM_PL_TT, gtt_size >> PAGE_SHIFT);
if (r) { if (r) {
DRM_ERROR("Failed initializing GTT heap.\n"); DRM_ERROR("Failed initializing GTT heap.\n");
@ -1529,6 +1821,7 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
DRM_INFO("amdgpu: %uM of GTT memory ready.\n", DRM_INFO("amdgpu: %uM of GTT memory ready.\n",
(unsigned)(gtt_size / (1024 * 1024))); (unsigned)(gtt_size / (1024 * 1024)));
/* Initialize various on-chip memory pools */
adev->gds.mem.total_size = adev->gds.mem.total_size << AMDGPU_GDS_SHIFT; adev->gds.mem.total_size = adev->gds.mem.total_size << AMDGPU_GDS_SHIFT;
adev->gds.mem.gfx_partition_size = adev->gds.mem.gfx_partition_size << AMDGPU_GDS_SHIFT; adev->gds.mem.gfx_partition_size = adev->gds.mem.gfx_partition_size << AMDGPU_GDS_SHIFT;
adev->gds.mem.cs_partition_size = adev->gds.mem.cs_partition_size << AMDGPU_GDS_SHIFT; adev->gds.mem.cs_partition_size = adev->gds.mem.cs_partition_size << AMDGPU_GDS_SHIFT;
@ -1568,6 +1861,7 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
} }
} }
/* Register debugfs entries for amdgpu_ttm */
r = amdgpu_ttm_debugfs_init(adev); r = amdgpu_ttm_debugfs_init(adev);
if (r) { if (r) {
DRM_ERROR("Failed to init debugfs\n"); DRM_ERROR("Failed to init debugfs\n");
@ -1576,11 +1870,19 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
return 0; return 0;
} }
/**
* amdgpu_ttm_late_init - Handle any late initialization for
* amdgpu_ttm
*/
void amdgpu_ttm_late_init(struct amdgpu_device *adev) void amdgpu_ttm_late_init(struct amdgpu_device *adev)
{ {
/* return the VGA stolen memory (if any) back to VRAM */
amdgpu_bo_free_kernel(&adev->stolen_vga_memory, NULL, NULL); amdgpu_bo_free_kernel(&adev->stolen_vga_memory, NULL, NULL);
} }
/**
* amdgpu_ttm_fini - De-initialize the TTM memory pools
*/
void amdgpu_ttm_fini(struct amdgpu_device *adev) void amdgpu_ttm_fini(struct amdgpu_device *adev)
{ {
if (!adev->mman.initialized) if (!adev->mman.initialized)
@ -1908,6 +2210,11 @@ static const struct drm_info_list amdgpu_ttm_debugfs_list[] = {
#endif #endif
}; };
/**
* amdgpu_ttm_vram_read - Linear read access to VRAM
*
* Accesses VRAM via MMIO for debugging purposes.
*/
static ssize_t amdgpu_ttm_vram_read(struct file *f, char __user *buf, static ssize_t amdgpu_ttm_vram_read(struct file *f, char __user *buf,
size_t size, loff_t *pos) size_t size, loff_t *pos)
{ {
@ -1947,6 +2254,11 @@ static ssize_t amdgpu_ttm_vram_read(struct file *f, char __user *buf,
return result; return result;
} }
/**
* amdgpu_ttm_vram_write - Linear write access to VRAM
*
* Accesses VRAM via MMIO for debugging purposes.
*/
static ssize_t amdgpu_ttm_vram_write(struct file *f, const char __user *buf, static ssize_t amdgpu_ttm_vram_write(struct file *f, const char __user *buf,
size_t size, loff_t *pos) size_t size, loff_t *pos)
{ {
@ -1995,6 +2307,9 @@ static const struct file_operations amdgpu_ttm_vram_fops = {
#ifdef CONFIG_DRM_AMDGPU_GART_DEBUGFS #ifdef CONFIG_DRM_AMDGPU_GART_DEBUGFS
/**
* amdgpu_ttm_gtt_read - Linear read access to GTT memory
*/
static ssize_t amdgpu_ttm_gtt_read(struct file *f, char __user *buf, static ssize_t amdgpu_ttm_gtt_read(struct file *f, char __user *buf,
size_t size, loff_t *pos) size_t size, loff_t *pos)
{ {
@ -2042,6 +2357,13 @@ static const struct file_operations amdgpu_ttm_gtt_fops = {
#endif #endif
/**
* amdgpu_iomem_read - Virtual read access to GPU mapped memory
*
* This function is used to read memory that has been mapped to the
* GPU and the known addresses are not physical addresses but instead
* bus addresses (e.g., what you'd put in an IB or ring buffer).
*/
static ssize_t amdgpu_iomem_read(struct file *f, char __user *buf, static ssize_t amdgpu_iomem_read(struct file *f, char __user *buf,
size_t size, loff_t *pos) size_t size, loff_t *pos)
{ {
@ -2050,6 +2372,7 @@ static ssize_t amdgpu_iomem_read(struct file *f, char __user *buf,
ssize_t result = 0; ssize_t result = 0;
int r; int r;
/* retrieve the IOMMU domain if any for this device */
dom = iommu_get_domain_for_dev(adev->dev); dom = iommu_get_domain_for_dev(adev->dev);
while (size) { while (size) {
@ -2062,6 +2385,10 @@ static ssize_t amdgpu_iomem_read(struct file *f, char __user *buf,
bytes = bytes < size ? bytes : size; bytes = bytes < size ? bytes : size;
/* Translate the bus address to a physical address. If
* the domain is NULL it means there is no IOMMU active
* and the address translation is the identity
*/
addr = dom ? iommu_iova_to_phys(dom, addr) : addr; addr = dom ? iommu_iova_to_phys(dom, addr) : addr;
pfn = addr >> PAGE_SHIFT; pfn = addr >> PAGE_SHIFT;
@ -2086,6 +2413,13 @@ static ssize_t amdgpu_iomem_read(struct file *f, char __user *buf,
return result; return result;
} }
/**
* amdgpu_iomem_write - Virtual write access to GPU mapped memory
*
* This function is used to write memory that has been mapped to the
* GPU and the known addresses are not physical addresses but instead
* bus addresses (e.g., what you'd put in an IB or ring buffer).
*/
static ssize_t amdgpu_iomem_write(struct file *f, const char __user *buf, static ssize_t amdgpu_iomem_write(struct file *f, const char __user *buf,
size_t size, loff_t *pos) size_t size, loff_t *pos)
{ {