kernel / drivers / gpu / drm / amd / amdgpu / amdgpu_ttm.c

/*
 * Copyright 2009 Jerome Glisse.
 * All Rights Reserved.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the
 * "Software"), to deal in the Software without restriction, including
 * without limitation the rights to use, copy, modify, merge, publish,
 * distribute, sub license, and/or sell copies of the Software, and to
 * permit persons to whom the Software is furnished to do so, subject to
 * the following conditions:
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
 * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
 * USE OR OTHER DEALINGS IN THE SOFTWARE.
 *
 * The above copyright notice and this permission notice (including the
 * next paragraph) shall be included in all copies or substantial portions
 * of the Software.
 *
 */
/*
 * Authors:
 *    Jerome Glisse <[e-mail]>
 *    Thomas Hellstrom <thomas-at-tungstengraphics-dot-com>
 *    Dave Airlie
 */
#include <drm/ttm/ttm_bo_api.h>
#include <drm/ttm/ttm_bo_driver.h>
#include <drm/ttm/ttm_placement.h>
#include <drm/ttm/ttm_module.h>
#include <drm/ttm/ttm_page_alloc.h>
#include <drm/drmP.h>
#include <drm/amdgpu_drm.h>
#include <linux/seq_file.h>
#include <linux/slab.h>
#include <linux/swiotlb.h>
#include <linux/swap.h>
#include <linux/pagemap.h>
#include <linux/debugfs.h>
#include <linux/iommu.h>
#include "amdgpu.h"
#include "amdgpu_object.h"
#include "amdgpu_trace.h"
#include "bif/bif_4_1_d.h"

#define DRM_FILE_PAGE_OFFSET (0x100000000ULL >> PAGE_SHIFT)

static int amdgpu_map_buffer(struct ttm_buffer_object *bo,
			     struct ttm_mem_reg *mem, unsigned num_pages,
			     uint64_t offset, unsigned window,
			     struct amdgpu_ring *ring,
			     uint64_t *addr);

static int amdgpu_ttm_debugfs_init(struct amdgpu_device *adev);
static void amdgpu_ttm_debugfs_fini(struct amdgpu_device *adev);

/*
 * Global memory.
 */
static int amdgpu_ttm_mem_global_init(struct drm_global_reference *ref)
{
	return ttm_mem_global_init(ref->object);
}

static void amdgpu_ttm_mem_global_release(struct drm_global_reference *ref)
{
	ttm_mem_global_release(ref->object);
}

static int amdgpu_ttm_global_init(struct amdgpu_device *adev)
{
	struct drm_global_reference *global_ref;
	struct amdgpu_ring *ring;
	struct drm_sched_rq *rq;
	int r;

	adev->mman.mem_global_referenced = false;
	global_ref = &adev->mman.mem_global_ref;
	global_ref->global_type = DRM_GLOBAL_TTM_MEM;
	global_ref->size = sizeof(struct ttm_mem_global);
	global_ref->init = &amdgpu_ttm_mem_global_init;
	global_ref->release = &amdgpu_ttm_mem_global_release;
	r = drm_global_item_ref(global_ref);
	if (r) {
		DRM_ERROR("Failed setting up TTM memory accounting "
			  "subsystem.\n");
		goto error_mem;
	}

	adev->mman.bo_global_ref.mem_glob =
		adev->mman.mem_global_ref.object;
	global_ref = &adev->mman.bo_global_ref.ref;
	global_ref->global_type = DRM_GLOBAL_TTM_BO;
	global_ref->size = sizeof(struct ttm_bo_global);
	global_ref->init = &ttm_bo_global_init;
	global_ref->release = &ttm_bo_global_release;
	r = drm_global_item_ref(global_ref);
	if (r) {
		DRM_ERROR("Failed setting up TTM BO subsystem.\n");
		goto error_bo;
	}

	mutex_init(&adev->mman.gtt_window_lock);

	ring = adev->mman.buffer_funcs_ring;
	rq = &ring->sched.sched_rq[DRM_SCHED_PRIORITY_KERNEL];
	r = drm_sched_entity_init(&ring->sched, &adev->mman.entity,
				  rq, amdgpu_sched_jobs, NULL);
	if (r) {
		DRM_ERROR("Failed setting up TTM BO move run queue.\n");
		goto error_entity;
	}

	adev->mman.mem_global_referenced = true;

	return 0;

error_entity:
	drm_global_item_unref(&adev->mman.bo_global_ref.ref);
error_bo:
	drm_global_item_unref(&adev->mman.mem_global_ref);
error_mem:
	return r;
}

static void amdgpu_ttm_global_fini(struct amdgpu_device *adev)
{
	if (adev->mman.mem_global_referenced) {
		drm_sched_entity_fini(adev->mman.entity.sched,
				      &adev->mman.entity);
		mutex_destroy(&adev->mman.gtt_window_lock);
		drm_global_item_unref(&adev->mman.bo_global_ref.ref);
		drm_global_item_unref(&adev->mman.mem_global_ref);
		adev->mman.mem_global_referenced = false;
	}
}

static int amdgpu_invalidate_caches(struct ttm_bo_device *bdev, uint32_t flags)
{
	return 0;
}

static int amdgpu_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
				struct ttm_mem_type_manager *man)
{
	struct amdgpu_device *adev;

	adev = amdgpu_ttm_adev(bdev);

	switch (type) {
	case TTM_PL_SYSTEM:
		/* System memory */
		man->flags = TTM_MEMTYPE_FLAG_MAPPABLE;
		man->available_caching = TTM_PL_MASK_CACHING;
		man->default_caching = TTM_PL_FLAG_CACHED;
		break;
	case TTM_PL_TT:
		man->func = &amdgpu_gtt_mgr_func;
		man->gpu_offset = adev->mc.gart_start;
		man->available_caching = TTM_PL_MASK_CACHING;
		man->default_caching = TTM_PL_FLAG_CACHED;
		man->flags = TTM_MEMTYPE_FLAG_MAPPABLE | TTM_MEMTYPE_FLAG_CMA;
		break;
	case TTM_PL_VRAM:
		/* "On-card" video ram */
		man->func = &amdgpu_vram_mgr_func;
		man->gpu_offset = adev->mc.vram_start;
		man->flags = TTM_MEMTYPE_FLAG_FIXED |
			     TTM_MEMTYPE_FLAG_MAPPABLE;
		man->available_caching = TTM_PL_FLAG_UNCACHED | TTM_PL_FLAG_WC;
		man->default_caching = TTM_PL_FLAG_WC;
		break;
	case AMDGPU_PL_GDS:
	case AMDGPU_PL_GWS:
	case AMDGPU_PL_OA:
		/* On-chip GDS memory*/
		man->func = &ttm_bo_manager_func;
		man->gpu_offset = 0;
		man->flags = TTM_MEMTYPE_FLAG_FIXED | TTM_MEMTYPE_FLAG_CMA;
		man->available_caching = TTM_PL_FLAG_UNCACHED;
		man->default_caching = TTM_PL_FLAG_UNCACHED;
		break;
	default:
		DRM_ERROR("Unsupported memory type %u\n", (unsigned)type);
		return -EINVAL;
	}
	return 0;
}

static void amdgpu_evict_flags(struct ttm_buffer_object *bo,
				struct ttm_placement *placement)
{
	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
	struct amdgpu_bo *abo;
	static const struct ttm_place placements = {
		.fpfn = 0,
		.lpfn = 0,
		.flags = TTM_PL_MASK_CACHING | TTM_PL_FLAG_SYSTEM
	};

	if (!amdgpu_ttm_bo_is_amdgpu_bo(bo)) {
		placement->placement = &placements;
		placement->busy_placement = &placements;
		placement->num_placement = 1;
		placement->num_busy_placement = 1;
		return;
	}
	abo = ttm_to_amdgpu_bo(bo);
	switch (bo->mem.mem_type) {
	case TTM_PL_VRAM:
		if (adev->mman.buffer_funcs &&
		    adev->mman.buffer_funcs_ring &&
		    adev->mman.buffer_funcs_ring->ready == false) {
			amdgpu_ttm_placement_from_domain(abo, AMDGPU_GEM_DOMAIN_CPU);
		} else if (adev->mc.visible_vram_size < adev->mc.real_vram_size &&
			   !(abo->flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED)) {
			unsigned fpfn = adev->mc.visible_vram_size >> PAGE_SHIFT;
			struct drm_mm_node *node = bo->mem.mm_node;
			unsigned long pages_left;

			for (pages_left = bo->mem.num_pages;
			     pages_left;
			     pages_left -= node->size, node++) {
				if (node->start < fpfn)
					break;
			}

			if (!pages_left)
				goto gtt;

			/* Try evicting to the CPU inaccessible part of VRAM
			 * first, but only set GTT as busy placement, so this
			 * BO will be evicted to GTT rather than causing other
			 * BOs to be evicted from VRAM
			 */
			amdgpu_ttm_placement_from_domain(abo, AMDGPU_GEM_DOMAIN_VRAM |
							 AMDGPU_GEM_DOMAIN_GTT);
			abo->placements[0].fpfn = fpfn;
			abo->placements[0].lpfn = 0;
			abo->placement.busy_placement = &abo->placements[1];
			abo->placement.num_busy_placement = 1;
		} else {
gtt:
			amdgpu_ttm_placement_from_domain(abo, AMDGPU_GEM_DOMAIN_GTT);
		}
		break;
	case TTM_PL_TT:
	default:
		amdgpu_ttm_placement_from_domain(abo, AMDGPU_GEM_DOMAIN_CPU);
	}
	*placement = abo->placement;
}

static int amdgpu_verify_access(struct ttm_buffer_object *bo, struct file *filp)
{
	struct amdgpu_bo *abo = ttm_to_amdgpu_bo(bo);

	if (amdgpu_ttm_tt_get_usermm(bo->ttm))
		return -EPERM;
	return drm_vma_node_verify_access(&abo->gem_base.vma_node,
					  filp->private_data);
}

static void amdgpu_move_null(struct ttm_buffer_object *bo,
			     struct ttm_mem_reg *new_mem)
{
	struct ttm_mem_reg *old_mem = &bo->mem;

	BUG_ON(old_mem->mm_node != NULL);
	*old_mem = *new_mem;
	new_mem->mm_node = NULL;
}

static uint64_t amdgpu_mm_node_addr(struct ttm_buffer_object *bo,
				    struct drm_mm_node *mm_node,
				    struct ttm_mem_reg *mem)
{
	uint64_t addr = 0;

	if (mem->mem_type != TTM_PL_TT || amdgpu_gtt_mgr_has_gart_addr(mem)) {
		addr = mm_node->start << PAGE_SHIFT;
		addr += bo->bdev->man[mem->mem_type].gpu_offset;
	}
	return addr;
}

/**
 * amdgpu_find_mm_node - Helper function finds the drm_mm_node
 *  corresponding to @offset. It also modifies the offset to be
 *  within the drm_mm_node returned
 */
static struct drm_mm_node *amdgpu_find_mm_node(struct ttm_mem_reg *mem,
					       unsigned long *offset)
{
	struct drm_mm_node *mm_node = mem->mm_node;

	while (*offset >= (mm_node->size << PAGE_SHIFT)) {
		*offset -= (mm_node->size << PAGE_SHIFT);
		++mm_node;
	}
	return mm_node;
}

/**
 * amdgpu_copy_ttm_mem_to_mem - Helper function for copy
 *
 * The function copies @size bytes from {src->mem + src->offset} to
 * {dst->mem + dst->offset}. src->bo and dst->bo could be same BO for a
 * move and different for a BO to BO copy.
 *
 * @f: Returns the last fence if multiple jobs are submitted.
 */
int amdgpu_ttm_copy_mem_to_mem(struct amdgpu_device *adev,
			       struct amdgpu_copy_mem *src,
			       struct amdgpu_copy_mem *dst,
			       uint64_t size,
			       struct reservation_object *resv,
			       struct dma_fence **f)
{
	struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring;
	struct drm_mm_node *src_mm, *dst_mm;
	uint64_t src_node_start, dst_node_start, src_node_size,
		 dst_node_size, src_page_offset, dst_page_offset;
	struct dma_fence *fence = NULL;
	int r = 0;
	const uint64_t GTT_MAX_BYTES = (AMDGPU_GTT_MAX_TRANSFER_SIZE *
					AMDGPU_GPU_PAGE_SIZE);

	if (!ring->ready) {
		DRM_ERROR("Trying to move memory with ring turned off.\n");
		return -EINVAL;
	}

	src_mm = amdgpu_find_mm_node(src->mem, &src->offset);
	src_node_start = amdgpu_mm_node_addr(src->bo, src_mm, src->mem) +
					     src->offset;
	src_node_size = (src_mm->size << PAGE_SHIFT) - src->offset;
	src_page_offset = src_node_start & (PAGE_SIZE - 1);

	dst_mm = amdgpu_find_mm_node(dst->mem, &dst->offset);
	dst_node_start = amdgpu_mm_node_addr(dst->bo, dst_mm, dst->mem) +
					     dst->offset;
	dst_node_size = (dst_mm->size << PAGE_SHIFT) - dst->offset;
	dst_page_offset = dst_node_start & (PAGE_SIZE - 1);

	mutex_lock(&adev->mman.gtt_window_lock);

	while (size) {
		unsigned long cur_size;
		uint64_t from = src_node_start, to = dst_node_start;
		struct dma_fence *next;

		/* Copy size cannot exceed GTT_MAX_BYTES. So if src or dst
		 * begins at an offset, then adjust the size accordingly
		 */
		cur_size = min3(min(src_node_size, dst_node_size), size,
				GTT_MAX_BYTES);
		if (cur_size + src_page_offset > GTT_MAX_BYTES ||
		    cur_size + dst_page_offset > GTT_MAX_BYTES)
			cur_size -= max(src_page_offset, dst_page_offset);

		/* Map only what needs to be accessed. Map src to window 0 and
		 * dst to window 1
		 */
		if (src->mem->mem_type == TTM_PL_TT &&
		    !amdgpu_gtt_mgr_has_gart_addr(src->mem)) {
			r = amdgpu_map_buffer(src->bo, src->mem,
					PFN_UP(cur_size + src_page_offset),
					src_node_start, 0, ring,
					&from);
			if (r)
				goto error;
			/* Adjust the offset because amdgpu_map_buffer returns
			 * start of mapped page
			 */
			from += src_page_offset;
		}

		if (dst->mem->mem_type == TTM_PL_TT &&
		    !amdgpu_gtt_mgr_has_gart_addr(dst->mem)) {
			r = amdgpu_map_buffer(dst->bo, dst->mem,
					PFN_UP(cur_size + dst_page_offset),
					dst_node_start, 1, ring,
					&to);
			if (r)
				goto error;
			to += dst_page_offset;
		}

		r = amdgpu_copy_buffer(ring, from, to, cur_size,
				       resv, &next, false, true);
		if (r)
			goto error;

		dma_fence_put(fence);
		fence = next;

		size -= cur_size;
		if (!size)
			break;

		src_node_size -= cur_size;
		if (!src_node_size) {
			src_node_start = amdgpu_mm_node_addr(src->bo, ++src_mm,
							     src->mem);
			src_node_size = (src_mm->size << PAGE_SHIFT);
		} else {
			src_node_start += cur_size;
			src_page_offset = src_node_start & (PAGE_SIZE - 1);
		}
		dst_node_size -= cur_size;
		if (!dst_node_size) {
			dst_node_start = amdgpu_mm_node_addr(dst->bo, ++dst_mm,
							     dst->mem);
			dst_node_size = (dst_mm->size << PAGE_SHIFT);
		} else {
			dst_node_start += cur_size;
			dst_page_offset = dst_node_start & (PAGE_SIZE - 1);
		}
	}
error:
	mutex_unlock(&adev->mman.gtt_window_lock);
	if (f)
		*f = dma_fence_get(fence);
	dma_fence_put(fence);
	return r;
}


static int amdgpu_move_blit(struct ttm_buffer_object *bo,
			    bool evict, bool no_wait_gpu,
			    struct ttm_mem_reg *new_mem,
			    struct ttm_mem_reg *old_mem)
{
	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
	struct amdgpu_copy_mem src, dst;
	struct dma_fence *fence = NULL;
	int r;

	src.bo = bo;
	dst.bo = bo;
	src.mem = old_mem;
	dst.mem = new_mem;
	src.offset = 0;
	dst.offset = 0;

	r = amdgpu_ttm_copy_mem_to_mem(adev, &src, &dst,
				       new_mem->num_pages << PAGE_SHIFT,
				       bo->resv, &fence);
	if (r)
		goto error;

	r = ttm_bo_pipeline_move(bo, fence, evict, new_mem);
	dma_fence_put(fence);
	return r;

error:
	if (fence)
		dma_fence_wait(fence, false);
	dma_fence_put(fence);
	return r;
}

static int amdgpu_move_vram_ram(struct ttm_buffer_object *bo, bool evict,
				struct ttm_operation_ctx *ctx,
				struct ttm_mem_reg *new_mem)
{
	struct amdgpu_device *adev;
	struct ttm_mem_reg *old_mem = &bo->mem;
	struct ttm_mem_reg tmp_mem;
	struct ttm_place placements;
	struct ttm_placement placement;
	int r;

	adev = amdgpu_ttm_adev(bo->bdev);
	tmp_mem = *new_mem;
	tmp_mem.mm_node = NULL;
	placement.num_placement = 1;
	placement.placement = &placements;
	placement.num_busy_placement = 1;
	placement.busy_placement = &placements;
	placements.fpfn = 0;
	placements.lpfn = 0;
	placements.flags = TTM_PL_MASK_CACHING | TTM_PL_FLAG_TT;
	r = ttm_bo_mem_space(bo, &placement, &tmp_mem, ctx);
	if (unlikely(r)) {
		return r;
	}

	r = ttm_tt_set_placement_caching(bo->ttm, tmp_mem.placement);
	if (unlikely(r)) {
		goto out_cleanup;
	}

	r = ttm_tt_bind(bo->ttm, &tmp_mem, ctx);
	if (unlikely(r)) {
		goto out_cleanup;
	}
	r = amdgpu_move_blit(bo, true, ctx->no_wait_gpu, &tmp_mem, old_mem);
	if (unlikely(r)) {
		goto out_cleanup;
	}
	r = ttm_bo_move_ttm(bo, ctx, new_mem);
out_cleanup:
	ttm_bo_mem_put(bo, &tmp_mem);
	return r;
}

static int amdgpu_move_ram_vram(struct ttm_buffer_object *bo, bool evict,
				struct ttm_operation_ctx *ctx,
				struct ttm_mem_reg *new_mem)
{
	struct amdgpu_device *adev;
	struct ttm_mem_reg *old_mem = &bo->mem;
	struct ttm_mem_reg tmp_mem;
	struct ttm_placement placement;
	struct ttm_place placements;
	int r;

	adev = amdgpu_ttm_adev(bo->bdev);
	tmp_mem = *new_mem;
	tmp_mem.mm_node = NULL;
	placement.num_placement = 1;
	placement.placement = &placements;
	placement.num_busy_placement = 1;
	placement.busy_placement = &placements;
	placements.fpfn = 0;
	placements.lpfn = 0;
	placements.flags = TTM_PL_MASK_CACHING | TTM_PL_FLAG_TT;
	r = ttm_bo_mem_space(bo, &placement, &tmp_mem, ctx);
	if (unlikely(r)) {
		return r;
	}
	r = ttm_bo_move_ttm(bo, ctx, &tmp_mem);
	if (unlikely(r)) {
		goto out_cleanup;
	}
	r = amdgpu_move_blit(bo, true, ctx->no_wait_gpu, new_mem, old_mem);
	if (unlikely(r)) {
		goto out_cleanup;
	}
out_cleanup:
	ttm_bo_mem_put(bo, &tmp_mem);
	return r;
}

static int amdgpu_bo_move(struct ttm_buffer_object *bo, bool evict,
			  struct ttm_operation_ctx *ctx,
			  struct ttm_mem_reg *new_mem)
{
	struct amdgpu_device *adev;
	struct amdgpu_bo *abo;
	struct ttm_mem_reg *old_mem = &bo->mem;
	int r;

	/* Can't move a pinned BO */
	abo = ttm_to_amdgpu_bo(bo);
	if (WARN_ON_ONCE(abo->pin_count > 0))
		return -EINVAL;

	adev = amdgpu_ttm_adev(bo->bdev);

	if (old_mem->mem_type == TTM_PL_SYSTEM && bo->ttm == NULL) {
		amdgpu_move_null(bo, new_mem);
		return 0;
	}
	if ((old_mem->mem_type == TTM_PL_TT &&
	     new_mem->mem_type == TTM_PL_SYSTEM) ||
	    (old_mem->mem_type == TTM_PL_SYSTEM &&
	     new_mem->mem_type == TTM_PL_TT)) {
		/* bind is enough */
		amdgpu_move_null(bo, new_mem);
		return 0;
	}
	if (adev->mman.buffer_funcs == NULL ||
	    adev->mman.buffer_funcs_ring == NULL ||
	    !adev->mman.buffer_funcs_ring->ready) {
		/* use memcpy */
		goto memcpy;
	}

	if (old_mem->mem_type == TTM_PL_VRAM &&
	    new_mem->mem_type == TTM_PL_SYSTEM) {
		r = amdgpu_move_vram_ram(bo, evict, ctx, new_mem);
	} else if (old_mem->mem_type == TTM_PL_SYSTEM &&
		   new_mem->mem_type == TTM_PL_VRAM) {
		r = amdgpu_move_ram_vram(bo, evict, ctx, new_mem);
	} else {
		r = amdgpu_move_blit(bo, evict, ctx->no_wait_gpu,
				     new_mem, old_mem);
	}

	if (r) {
memcpy:
		r = ttm_bo_move_memcpy(bo, ctx, new_mem);
		if (r) {
			return r;
		}
	}

	if (bo->type == ttm_bo_type_device &&
	    new_mem->mem_type == TTM_PL_VRAM &&
	    old_mem->mem_type != TTM_PL_VRAM) {
		/* amdgpu_bo_fault_reserve_notify will re-set this if the CPU
		 * accesses the BO after it's moved.
		 */
		abo->flags &= ~AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
	}

	/* update statistics */
	atomic64_add((u64)bo->num_pages << PAGE_SHIFT, &adev->num_bytes_moved);
	return 0;
}

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 amdgpu_device *adev = amdgpu_ttm_adev(bdev);

	mem->bus.addr = NULL;
	mem->bus.offset = 0;
	mem->bus.size = mem->num_pages << PAGE_SHIFT;
	mem->bus.base = 0;
	mem->bus.is_iomem = false;
	if (!(man->flags & TTM_MEMTYPE_FLAG_MAPPABLE))
		return -EINVAL;
	switch (mem->mem_type) {
	case TTM_PL_SYSTEM:
		/* system memory */
		return 0;
	case TTM_PL_TT:
		break;
	case TTM_PL_VRAM:
		mem->bus.offset = mem->start << PAGE_SHIFT;
		/* check if it's visible */
		if ((mem->bus.offset + mem->bus.size) > adev->mc.visible_vram_size)
			return -EINVAL;
		mem->bus.base = adev->mc.aper_base;
		mem->bus.is_iomem = true;
		break;
	default:
		return -EINVAL;
	}
	return 0;
}

static void amdgpu_ttm_io_mem_free(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
{
}

static unsigned long amdgpu_ttm_io_mem_pfn(struct ttm_buffer_object *bo,
					   unsigned long page_offset)
{
	struct drm_mm_node *mm;
	unsigned long offset = (page_offset << PAGE_SHIFT);

	mm = amdgpu_find_mm_node(&bo->mem, &offset);
	return (bo->mem.bus.base >> PAGE_SHIFT) + mm->start +
		(offset >> PAGE_SHIFT);
}

/*
 * TTM backend functions.
 */
struct amdgpu_ttm_gup_task_list {
	struct list_head	list;
	struct task_struct	*task;
};

struct amdgpu_ttm_tt {
	struct ttm_dma_tt	ttm;
	struct amdgpu_device	*adev;
	u64			offset;
	uint64_t		userptr;
	struct mm_struct	*usermm;
	uint32_t		userflags;
	spinlock_t              guptasklock;
	struct list_head        guptasks;
	atomic_t		mmu_invalidations;
	uint32_t		last_set_pages;
};

int amdgpu_ttm_tt_get_user_pages(struct ttm_tt *ttm, struct page **pages)
{
	struct amdgpu_ttm_tt *gtt = (void *)ttm;
	unsigned int flags = 0;
	unsigned pinned = 0;
	int r;

	if (!(gtt->userflags & AMDGPU_GEM_USERPTR_READONLY))
		flags |= FOLL_WRITE;

	down_read(&current->mm->mmap_sem);

	if (gtt->userflags & AMDGPU_GEM_USERPTR_ANONONLY) {
		/* check that we only use anonymous memory
		   to prevent problems with writeback */
		unsigned long end = gtt->userptr + ttm->num_pages * PAGE_SIZE;
		struct vm_area_struct *vma;

		vma = find_vma(gtt->usermm, gtt->userptr);
		if (!vma || vma->vm_file || vma->vm_end < end) {
			up_read(&current->mm->mmap_sem);
			return -EPERM;
		}
	}

	do {
		unsigned num_pages = ttm->num_pages - pinned;
		uint64_t userptr = gtt->userptr + pinned * PAGE_SIZE;
		struct page **p = pages + pinned;
		struct amdgpu_ttm_gup_task_list guptask;

		guptask.task = current;
		spin_lock(&gtt->guptasklock);
		list_add(&guptask.list, &gtt->guptasks);
		spin_unlock(&gtt->guptasklock);

		r = get_user_pages(userptr, num_pages, flags, p, NULL);

		spin_lock(&gtt->guptasklock);
		list_del(&guptask.list);
		spin_unlock(&gtt->guptasklock);

		if (r < 0)
			goto release_pages;

		pinned += r;

	} while (pinned < ttm->num_pages);

	up_read(&current->mm->mmap_sem);
	return 0;

release_pages:
	release_pages(pages, pinned);
	up_read(&current->mm->mmap_sem);
	return r;
}

void amdgpu_ttm_tt_set_user_pages(struct ttm_tt *ttm, struct page **pages)
{
	struct amdgpu_ttm_tt *gtt = (void *)ttm;
	unsigned i;

	gtt->last_set_pages = atomic_read(&gtt->mmu_invalidations);
	for (i = 0; i < ttm->num_pages; ++i) {
		if (ttm->pages[i])
			put_page(ttm->pages[i]);

		ttm->pages[i] = pages ? pages[i] : NULL;
	}
}

void amdgpu_ttm_tt_mark_user_pages(struct ttm_tt *ttm)
{
	struct amdgpu_ttm_tt *gtt = (void *)ttm;
	unsigned i;

	for (i = 0; i < ttm->num_pages; ++i) {
		struct page *page = ttm->pages[i];

		if (!page)
			continue;

		if (!(gtt->userflags & AMDGPU_GEM_USERPTR_READONLY))
			set_page_dirty(page);

		mark_page_accessed(page);
	}
}

/* prepare the sg table with the user pages */
static int amdgpu_ttm_tt_pin_userptr(struct ttm_tt *ttm)
{
	struct amdgpu_device *adev = amdgpu_ttm_adev(ttm->bdev);
	struct amdgpu_ttm_tt *gtt = (void *)ttm;
	unsigned nents;
	int r;

	int write = !(gtt->userflags & AMDGPU_GEM_USERPTR_READONLY);
	enum dma_data_direction direction = write ?
		DMA_BIDIRECTIONAL : DMA_TO_DEVICE;

	r = sg_alloc_table_from_pages(ttm->sg, ttm->pages, ttm->num_pages, 0,
				      ttm->num_pages << PAGE_SHIFT,
				      GFP_KERNEL);
	if (r)
		goto release_sg;

	r = -ENOMEM;
	nents = dma_map_sg(adev->dev, ttm->sg->sgl, ttm->sg->nents, direction);
	if (nents != ttm->sg->nents)
		goto release_sg;

	drm_prime_sg_to_page_addr_arrays(ttm->sg, ttm->pages,
					 gtt->ttm.dma_address, ttm->num_pages);

	return 0;

release_sg:
	kfree(ttm->sg);
	return r;
}

static void amdgpu_ttm_tt_unpin_userptr(struct ttm_tt *ttm)
{
	struct amdgpu_device *adev = amdgpu_ttm_adev(ttm->bdev);
	struct amdgpu_ttm_tt *gtt = (void *)ttm;

	int write = !(gtt->userflags & AMDGPU_GEM_USERPTR_READONLY);
	enum dma_data_direction direction = write ?
		DMA_BIDIRECTIONAL : DMA_TO_DEVICE;

	/* double check that we don't free the table twice */
	if (!ttm->sg->sgl)
		return;

	/* free the sg table and pages again */
	dma_unmap_sg(adev->dev, ttm->sg->sgl, ttm->sg->nents, direction);

	amdgpu_ttm_tt_mark_user_pages(ttm);

	sg_free_table(ttm->sg);
}

static int amdgpu_ttm_backend_bind(struct ttm_tt *ttm,
				   struct ttm_mem_reg *bo_mem)
{
	struct amdgpu_ttm_tt *gtt = (void*)ttm;
	uint64_t flags;
	int r = 0;

	if (gtt->userptr) {
		r = amdgpu_ttm_tt_pin_userptr(ttm);
		if (r) {
			DRM_ERROR("failed to pin userptr\n");
			return r;
		}
	}
	if (!ttm->num_pages) {
		WARN(1, "nothing to bind %lu pages for mreg %p back %p!\n",
		     ttm->num_pages, bo_mem, ttm);
	}

	if (bo_mem->mem_type == AMDGPU_PL_GDS ||
	    bo_mem->mem_type == AMDGPU_PL_GWS ||
	    bo_mem->mem_type == AMDGPU_PL_OA)
		return -EINVAL;

	if (!amdgpu_gtt_mgr_has_gart_addr(bo_mem)) {
		gtt->offset = AMDGPU_BO_INVALID_OFFSET;
		return 0;
	}

	flags = amdgpu_ttm_tt_pte_flags(gtt->adev, ttm, bo_mem);
	gtt->offset = (u64)bo_mem->start << PAGE_SHIFT;
	r = amdgpu_gart_bind(gtt->adev, gtt->offset, ttm->num_pages,
		ttm->pages, gtt->ttm.dma_address, flags);

	if (r)
		DRM_ERROR("failed to bind %lu pages at 0x%08llX\n",
			  ttm->num_pages, gtt->offset);
	return r;
}

int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo)
{
	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
	struct ttm_operation_ctx ctx = { false, false };
	struct amdgpu_ttm_tt *gtt = (void*)bo->ttm;
	struct ttm_mem_reg tmp;
	struct ttm_placement placement;
	struct ttm_place placements;
	uint64_t flags;
	int r;

	if (bo->mem.mem_type != TTM_PL_TT ||
	    amdgpu_gtt_mgr_has_gart_addr(&bo->mem))
		return 0;

	tmp = bo->mem;
	tmp.mm_node = NULL;
	placement.num_placement = 1;
	placement.placement = &placements;
	placement.num_busy_placement = 1;
	placement.busy_placement = &placements;
	placements.fpfn = 0;
	placements.lpfn = adev->mc.gart_size >> PAGE_SHIFT;
	placements.flags = (bo->mem.placement & ~TTM_PL_MASK_MEM) |
		TTM_PL_FLAG_TT;

	r = ttm_bo_mem_space(bo, &placement, &tmp, &ctx);
	if (unlikely(r))
		return r;

	flags = amdgpu_ttm_tt_pte_flags(adev, bo->ttm, &tmp);
	gtt->offset = (u64)tmp.start << PAGE_SHIFT;
	r = amdgpu_gart_bind(adev, gtt->offset, bo->ttm->num_pages,
			     bo->ttm->pages, gtt->ttm.dma_address, flags);
	if (unlikely(r)) {
		ttm_bo_mem_put(bo, &tmp);
		return r;
	}

	ttm_bo_mem_put(bo, &bo->mem);
	bo->mem = tmp;
	bo->offset = (bo->mem.start << PAGE_SHIFT) +
		bo->bdev->man[bo->mem.mem_type].gpu_offset;

	return 0;
}

int amdgpu_ttm_recover_gart(struct ttm_buffer_object *tbo)
{
	struct amdgpu_device *adev = amdgpu_ttm_adev(tbo->bdev);
	struct amdgpu_ttm_tt *gtt = (void *)tbo->ttm;
	uint64_t flags;
	int r;

	if (!gtt)
		return 0;

	flags = amdgpu_ttm_tt_pte_flags(adev, &gtt->ttm.ttm, &tbo->mem);
	r = amdgpu_gart_bind(adev, gtt->offset, gtt->ttm.ttm.num_pages,
			     gtt->ttm.ttm.pages, gtt->ttm.dma_address, flags);
	if (r)
		DRM_ERROR("failed to bind %lu pages at 0x%08llX\n",
			  gtt->ttm.ttm.num_pages, gtt->offset);
	return r;
}

static int amdgpu_ttm_backend_unbind(struct ttm_tt *ttm)
{
	struct amdgpu_ttm_tt *gtt = (void *)ttm;
	int r;

	if (gtt->userptr)
		amdgpu_ttm_tt_unpin_userptr(ttm);

	if (gtt->offset == AMDGPU_BO_INVALID_OFFSET)
		return 0;

	/* unbind shouldn't be done for GDS/GWS/OA in ttm_bo_clean_mm */
	r = amdgpu_gart_unbind(gtt->adev, gtt->offset, ttm->num_pages);
	if (r)
		DRM_ERROR("failed to unbind %lu pages at 0x%08llX\n",
			  gtt->ttm.ttm.num_pages, gtt->offset);
	return r;
}

static void amdgpu_ttm_backend_destroy(struct ttm_tt *ttm)
{
	struct amdgpu_ttm_tt *gtt = (void *)ttm;

	ttm_dma_tt_fini(&gtt->ttm);
	kfree(gtt);
}

static struct ttm_backend_func amdgpu_backend_func = {
	.bind = &amdgpu_ttm_backend_bind,
	.unbind = &amdgpu_ttm_backend_unbind,
	.destroy = &amdgpu_ttm_backend_destroy,
};

static struct ttm_tt *amdgpu_ttm_tt_create(struct ttm_bo_device *bdev,
				    unsigned long size, uint32_t page_flags,
				    struct page *dummy_read_page)
{
	struct amdgpu_device *adev;
	struct amdgpu_ttm_tt *gtt;

	adev = amdgpu_ttm_adev(bdev);

	gtt = kzalloc(sizeof(struct amdgpu_ttm_tt), GFP_KERNEL);
	if (gtt == NULL) {
		return NULL;
	}
	gtt->ttm.ttm.func = &amdgpu_backend_func;
	gtt->adev = adev;
	if (ttm_dma_tt_init(&gtt->ttm, bdev, size, page_flags, dummy_read_page)) {
		kfree(gtt);
		return NULL;
	}
	return &gtt->ttm.ttm;
}

static int amdgpu_ttm_tt_populate(struct ttm_tt *ttm,
			struct ttm_operation_ctx *ctx)
{
	struct amdgpu_device *adev = amdgpu_ttm_adev(ttm->bdev);
	struct amdgpu_ttm_tt *gtt = (void *)ttm;
	bool slave = !!(ttm->page_flags & TTM_PAGE_FLAG_SG);

	if (ttm->state != tt_unpopulated)
		return 0;

	if (gtt && gtt->userptr) {
		ttm->sg = kzalloc(sizeof(struct sg_table), GFP_KERNEL);
		if (!ttm->sg)
			return -ENOMEM;

		ttm->page_flags |= TTM_PAGE_FLAG_SG;
		ttm->state = tt_unbound;
		return 0;
	}

	if (slave && ttm->sg) {
		drm_prime_sg_to_page_addr_arrays(ttm->sg, ttm->pages,
						 gtt->ttm.dma_address, ttm->num_pages);
		ttm->state = tt_unbound;
		return 0;
	}

#ifdef CONFIG_SWIOTLB
	if (swiotlb_nr_tbl()) {
		return ttm_dma_populate(&gtt->ttm, adev->dev, ctx);
	}
#endif

	return ttm_populate_and_map_pages(adev->dev, &gtt->ttm, ctx);
}

static void amdgpu_ttm_tt_unpopulate(struct ttm_tt *ttm)
{
	struct amdgpu_device *adev;
	struct amdgpu_ttm_tt *gtt = (void *)ttm;
	bool slave = !!(ttm->page_flags & TTM_PAGE_FLAG_SG);

	if (gtt && gtt->userptr) {
		amdgpu_ttm_tt_set_user_pages(ttm, NULL);
		kfree(ttm->sg);
		ttm->page_flags &= ~TTM_PAGE_FLAG_SG;
		return;
	}

	if (slave)
		return;

	adev = amdgpu_ttm_adev(ttm->bdev);

#ifdef CONFIG_SWIOTLB
	if (swiotlb_nr_tbl()) {
		ttm_dma_unpopulate(&gtt->ttm, adev->dev);
		return;
	}
#endif

	ttm_unmap_and_unpopulate_pages(adev->dev, &gtt->ttm);
}

int amdgpu_ttm_tt_set_userptr(struct ttm_tt *ttm, uint64_t addr,
			      uint32_t flags)
{
	struct amdgpu_ttm_tt *gtt = (void *)ttm;

	if (gtt == NULL)
		return -EINVAL;

	gtt->userptr = addr;
	gtt->usermm = current->mm;
	gtt->userflags = flags;
	spin_lock_init(&gtt->guptasklock);
	INIT_LIST_HEAD(&gtt->guptasks);
	atomic_set(&gtt->mmu_invalidations, 0);
	gtt->last_set_pages = 0;

	return 0;
}

struct mm_struct *amdgpu_ttm_tt_get_usermm(struct ttm_tt *ttm)
{
	struct amdgpu_ttm_tt *gtt = (void *)ttm;

	if (gtt == NULL)
		return NULL;

	return gtt->usermm;
}

bool amdgpu_ttm_tt_affect_userptr(struct ttm_tt *ttm, unsigned long start,
				  unsigned long end)
{
	struct amdgpu_ttm_tt *gtt = (void *)ttm;
	struct amdgpu_ttm_gup_task_list *entry;
	unsigned long size;

	if (gtt == NULL || !gtt->userptr)
		return false;

	size = (unsigned long)gtt->ttm.ttm.num_pages * PAGE_SIZE;
	if (gtt->userptr > end || gtt->userptr + size <= start)
		return false;

	spin_lock(&gtt->guptasklock);
	list_for_each_entry(entry, &gtt->guptasks, list) {
		if (entry->task == current) {
			spin_unlock(&gtt->guptasklock);
			return false;
		}
	}
	spin_unlock(&gtt->guptasklock);

	atomic_inc(&gtt->mmu_invalidations);

	return true;
}

bool amdgpu_ttm_tt_userptr_invalidated(struct ttm_tt *ttm,
				       int *last_invalidated)
{
	struct amdgpu_ttm_tt *gtt = (void *)ttm;
	int prev_invalidated = *last_invalidated;

	*last_invalidated = atomic_read(&gtt->mmu_invalidations);
	return prev_invalidated != *last_invalidated;
}

bool amdgpu_ttm_tt_userptr_needs_pages(struct ttm_tt *ttm)
{
	struct amdgpu_ttm_tt *gtt = (void *)ttm;

	if (gtt == NULL || !gtt->userptr)
		return false;

	return atomic_read(&gtt->mmu_invalidations) != gtt->last_set_pages;
}

bool amdgpu_ttm_tt_is_readonly(struct ttm_tt *ttm)
{
	struct amdgpu_ttm_tt *gtt = (void *)ttm;

	if (gtt == NULL)
		return false;

	return !!(gtt->userflags & AMDGPU_GEM_USERPTR_READONLY);
}

uint64_t amdgpu_ttm_tt_pte_flags(struct amdgpu_device *adev, struct ttm_tt *ttm,
				 struct ttm_mem_reg *mem)
{
	uint64_t flags = 0;

	if (mem && mem->mem_type != TTM_PL_SYSTEM)
		flags |= AMDGPU_PTE_VALID;

	if (mem && mem->mem_type == TTM_PL_TT) {
		flags |= AMDGPU_PTE_SYSTEM;

		if (ttm->caching_state == tt_cached)
			flags |= AMDGPU_PTE_SNOOPED;
	}

	flags |= adev->gart.gart_pte_flags;
	flags |= AMDGPU_PTE_READABLE;

	if (!amdgpu_ttm_tt_is_readonly(ttm))
		flags |= AMDGPU_PTE_WRITEABLE;

	return flags;
}

static bool amdgpu_ttm_bo_eviction_valuable(struct ttm_buffer_object *bo,
					    const struct ttm_place *place)
{
	unsigned long num_pages = bo->mem.num_pages;
	struct drm_mm_node *node = bo->mem.mm_node;

	switch (bo->mem.mem_type) {
	case TTM_PL_TT:
		return true;

	case TTM_PL_VRAM:
		/* Check each drm MM node individually */
		while (num_pages) {
			if (place->fpfn < (node->start + node->size) &&
			    !(place->lpfn && place->lpfn <= node->start))
				return true;

			num_pages -= node->size;
			++node;
		}
		return false;

	default:
		break;
	}

	return ttm_bo_eviction_valuable(bo, place);
}

static int amdgpu_ttm_access_memory(struct ttm_buffer_object *bo,
				    unsigned long offset,
				    void *buf, int len, int write)
{
	struct amdgpu_bo *abo = ttm_to_amdgpu_bo(bo);
	struct amdgpu_device *adev = amdgpu_ttm_adev(abo->tbo.bdev);
	struct drm_mm_node *nodes;
	uint32_t value = 0;
	int ret = 0;
	uint64_t pos;
	unsigned long flags;

	if (bo->mem.mem_type != TTM_PL_VRAM)
		return -EIO;

	nodes = amdgpu_find_mm_node(&abo->tbo.mem, &offset);
	pos = (nodes->start << PAGE_SHIFT) + offset;

	while (len && pos < adev->mc.mc_vram_size) {
		uint64_t aligned_pos = pos & ~(uint64_t)3;
		uint32_t bytes = 4 - (pos & 3);
		uint32_t shift = (pos & 3) * 8;
		uint32_t mask = 0xffffffff << shift;

		if (len < bytes) {
			mask &= 0xffffffff >> (bytes - len) * 8;
			bytes = len;
		}

		spin_lock_irqsave(&adev->mmio_idx_lock, flags);
		WREG32_NO_KIQ(mmMM_INDEX, ((uint32_t)aligned_pos) | 0x80000000);
		WREG32_NO_KIQ(mmMM_INDEX_HI, aligned_pos >> 31);
		if (!write || mask != 0xffffffff)
			value = RREG32_NO_KIQ(mmMM_DATA);
		if (write) {
			value &= ~mask;
			value |= (*(uint32_t *)buf << shift) & mask;
			WREG32_NO_KIQ(mmMM_DATA, value);
		}
		spin_unlock_irqrestore(&adev->mmio_idx_lock, flags);
		if (!write) {
			value = (value & mask) >> shift;
			memcpy(buf, &value, bytes);
		}

		ret += bytes;
		buf = (uint8_t *)buf + bytes;
		pos += bytes;
		len -= bytes;
		if (pos >= (nodes->start + nodes->size) << PAGE_SHIFT) {
			++nodes;
			pos = (nodes->start << PAGE_SHIFT);
		}
	}

	return ret;
}

static struct ttm_bo_driver amdgpu_bo_driver = {
	.ttm_tt_create = &amdgpu_ttm_tt_create,
	.ttm_tt_populate = &amdgpu_ttm_tt_populate,
	.ttm_tt_unpopulate = &amdgpu_ttm_tt_unpopulate,
	.invalidate_caches = &amdgpu_invalidate_caches,
	.init_mem_type = &amdgpu_init_mem_type,
	.eviction_valuable = amdgpu_ttm_bo_eviction_valuable,
	.evict_flags = &amdgpu_evict_flags,
	.move = &amdgpu_bo_move,
	.verify_access = &amdgpu_verify_access,
	.move_notify = &amdgpu_bo_move_notify,
	.fault_reserve_notify = &amdgpu_bo_fault_reserve_notify,
	.io_mem_reserve = &amdgpu_ttm_io_mem_reserve,
	.io_mem_free = &amdgpu_ttm_io_mem_free,
	.io_mem_pfn = amdgpu_ttm_io_mem_pfn,
	.access_memory = &amdgpu_ttm_access_memory
};

/*
 * Firmware Reservation functions
 */
/**
 * amdgpu_ttm_fw_reserve_vram_fini - free fw reserved vram
 *
 * @adev: amdgpu_device pointer
 *
 * free fw reserved vram if it has been reserved.
 */
static void amdgpu_ttm_fw_reserve_vram_fini(struct amdgpu_device *adev)
{
	amdgpu_bo_free_kernel(&adev->fw_vram_usage.reserved_bo,
		NULL, &adev->fw_vram_usage.va);
}

/**
 * amdgpu_ttm_fw_reserve_vram_init - create bo vram reservation from fw
 *
 * @adev: amdgpu_device pointer
 *
 * create bo vram reservation from fw.
 */
static int amdgpu_ttm_fw_reserve_vram_init(struct amdgpu_device *adev)
{
	struct ttm_operation_ctx ctx = { false, false };
	int r = 0;
	int i;
	u64 vram_size = adev->mc.visible_vram_size;
	u64 offset = adev->fw_vram_usage.start_offset;
	u64 size = adev->fw_vram_usage.size;
	struct amdgpu_bo *bo;

	adev->fw_vram_usage.va = NULL;
	adev->fw_vram_usage.reserved_bo = NULL;

	if (adev->fw_vram_usage.size > 0 &&
		adev->fw_vram_usage.size <= vram_size) {

		r = amdgpu_bo_create(adev, adev->fw_vram_usage.size,
			PAGE_SIZE, true, AMDGPU_GEM_DOMAIN_VRAM,
			AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
			AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS, NULL, NULL, 0,
			&adev->fw_vram_usage.reserved_bo);
		if (r)
			goto error_create;

		r = amdgpu_bo_reserve(adev->fw_vram_usage.reserved_bo, false);
		if (r)
			goto error_reserve;

		/* remove the original mem node and create a new one at the
		 * request position
		 */
		bo = adev->fw_vram_usage.reserved_bo;
		offset = ALIGN(offset, PAGE_SIZE);
		for (i = 0; i < bo->placement.num_placement; ++i) {
			bo->placements[i].fpfn = offset >> PAGE_SHIFT;
			bo->placements[i].lpfn = (offset + size) >> PAGE_SHIFT;
		}

		ttm_bo_mem_put(&bo->tbo, &bo->tbo.mem);
		r = ttm_bo_mem_space(&bo->tbo, &bo->placement,
				     &bo->tbo.mem, &ctx);
		if (r)
			goto error_pin;

		r = amdgpu_bo_pin_restricted(adev->fw_vram_usage.reserved_bo,
			AMDGPU_GEM_DOMAIN_VRAM,
			adev->fw_vram_usage.start_offset,
			(adev->fw_vram_usage.start_offset +
			adev->fw_vram_usage.size), NULL);
		if (r)
			goto error_pin;
		r = amdgpu_bo_kmap(adev->fw_vram_usage.reserved_bo,
			&adev->fw_vram_usage.va);
		if (r)
			goto error_kmap;

		amdgpu_bo_unreserve(adev->fw_vram_usage.reserved_bo);
	}
	return r;

error_kmap:
	amdgpu_bo_unpin(adev->fw_vram_usage.reserved_bo);
error_pin:
	amdgpu_bo_unreserve(adev->fw_vram_usage.reserved_bo);
error_reserve:
	amdgpu_bo_unref(&adev->fw_vram_usage.reserved_bo);
error_create:
	adev->fw_vram_usage.va = NULL;
	adev->fw_vram_usage.reserved_bo = NULL;
	return r;
}

int amdgpu_ttm_init(struct amdgpu_device *adev)
{
	uint64_t gtt_size;
	int r;
	u64 vis_vram_limit;

	r = amdgpu_ttm_global_init(adev);
	if (r) {
		return r;
	}
	/* No others user of address space so set it to 0 */
	r = ttm_bo_device_init(&adev->mman.bdev,
			       adev->mman.bo_global_ref.ref.object,
			       &amdgpu_bo_driver,
			       adev->ddev->anon_inode->i_mapping,
			       DRM_FILE_PAGE_OFFSET,
			       adev->need_dma32);
	if (r) {
		DRM_ERROR("failed initializing buffer object driver(%d).\n", r);
		return r;
	}
	adev->mman.initialized = true;
	r = ttm_bo_init_mm(&adev->mman.bdev, TTM_PL_VRAM,
				adev->mc.real_vram_size >> PAGE_SHIFT);
	if (r) {
		DRM_ERROR("Failed initializing VRAM heap.\n");
		return r;
	}

	/* Reduce size of CPU-visible VRAM if requested */
	vis_vram_limit = (u64)amdgpu_vis_vram_limit * 1024 * 1024;
	if (amdgpu_vis_vram_limit > 0 &&
	    vis_vram_limit <= adev->mc.visible_vram_size)
		adev->mc.visible_vram_size = vis_vram_limit;

	/* Change the size here instead of the init above so only lpfn is affected */
	amdgpu_ttm_set_active_vram_size(adev, adev->mc.visible_vram_size);

	/*
	 *The reserved vram for firmware must be pinned to the specified
	 *place on the VRAM, so reserve it early.
	 */
	r = amdgpu_ttm_fw_reserve_vram_init(adev);
	if (r) {
		return r;
	}

	r = amdgpu_bo_create_kernel(adev, adev->mc.stolen_size, PAGE_SIZE,
				    AMDGPU_GEM_DOMAIN_VRAM,
				    &adev->stolen_vga_memory,
				    NULL, NULL);
	if (r)
		return r;
	DRM_INFO("amdgpu: %uM of VRAM memory ready\n",
		 (unsigned) (adev->mc.real_vram_size / (1024 * 1024)));

	if (amdgpu_gtt_size == -1) {
		struct sysinfo si;

		si_meminfo(&si);
		gtt_size = min(max((AMDGPU_DEFAULT_GTT_SIZE_MB << 20),
			       adev->mc.mc_vram_size),
			       ((uint64_t)si.totalram * si.mem_unit * 3/4));
	}
	else
		gtt_size = (uint64_t)amdgpu_gtt_size << 20;
	r = ttm_bo_init_mm(&adev->mman.bdev, TTM_PL_TT, gtt_size >> PAGE_SHIFT);
	if (r) {
		DRM_ERROR("Failed initializing GTT heap.\n");
		return r;
	}
	DRM_INFO("amdgpu: %uM of GTT memory ready.\n",
		 (unsigned)(gtt_size / (1024 * 1024)));

	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.cs_partition_size = adev->gds.mem.cs_partition_size << AMDGPU_GDS_SHIFT;
	adev->gds.gws.total_size = adev->gds.gws.total_size << AMDGPU_GWS_SHIFT;
	adev->gds.gws.gfx_partition_size = adev->gds.gws.gfx_partition_size << AMDGPU_GWS_SHIFT;
	adev->gds.gws.cs_partition_size = adev->gds.gws.cs_partition_size << AMDGPU_GWS_SHIFT;
	adev->gds.oa.total_size = adev->gds.oa.total_size << AMDGPU_OA_SHIFT;
	adev->gds.oa.gfx_partition_size = adev->gds.oa.gfx_partition_size << AMDGPU_OA_SHIFT;
	adev->gds.oa.cs_partition_size = adev->gds.oa.cs_partition_size << AMDGPU_OA_SHIFT;
	/* GDS Memory */
	if (adev->gds.mem.total_size) {
		r = ttm_bo_init_mm(&adev->mman.bdev, AMDGPU_PL_GDS,
				   adev->gds.mem.total_size >> PAGE_SHIFT);
		if (r) {
			DRM_ERROR("Failed initializing GDS heap.\n");
			return r;
		}
	}

	/* GWS */
	if (adev->gds.gws.total_size) {
		r = ttm_bo_init_mm(&adev->mman.bdev, AMDGPU_PL_GWS,
				   adev->gds.gws.total_size >> PAGE_SHIFT);
		if (r) {
			DRM_ERROR("Failed initializing gws heap.\n");
			return r;
		}
	}

	/* OA */
	if (adev->gds.oa.total_size) {
		r = ttm_bo_init_mm(&adev->mman.bdev, AMDGPU_PL_OA,
				   adev->gds.oa.total_size >> PAGE_SHIFT);
		if (r) {
			DRM_ERROR("Failed initializing oa heap.\n");
			return r;
		}
	}

	r = amdgpu_ttm_debugfs_init(adev);
	if (r) {
		DRM_ERROR("Failed to init debugfs\n");
		return r;
	}
	return 0;
}

void amdgpu_ttm_fini(struct amdgpu_device *adev)
{
	if (!adev->mman.initialized)
		return;

	amdgpu_ttm_debugfs_fini(adev);
	amdgpu_bo_free_kernel(&adev->stolen_vga_memory, NULL, NULL);
	amdgpu_ttm_fw_reserve_vram_fini(adev);

	ttm_bo_clean_mm(&adev->mman.bdev, TTM_PL_VRAM);
	ttm_bo_clean_mm(&adev->mman.bdev, TTM_PL_TT);
	if (adev->gds.mem.total_size)
		ttm_bo_clean_mm(&adev->mman.bdev, AMDGPU_PL_GDS);
	if (adev->gds.gws.total_size)
		ttm_bo_clean_mm(&adev->mman.bdev, AMDGPU_PL_GWS);
	if (adev->gds.oa.total_size)
		ttm_bo_clean_mm(&adev->mman.bdev, AMDGPU_PL_OA);
	ttm_bo_device_release(&adev->mman.bdev);
	amdgpu_ttm_global_fini(adev);
	adev->mman.initialized = false;
	DRM_INFO("amdgpu: ttm finalized\n");
}

/* this should only be called at bootup or when userspace
 * isn't running */
void amdgpu_ttm_set_active_vram_size(struct amdgpu_device *adev, u64 size)
{
	struct ttm_mem_type_manager *man;

	if (!adev->mman.initialized)
		return;

	man = &adev->mman.bdev.man[TTM_PL_VRAM];
	/* this just adjusts TTM size idea, which sets lpfn to the correct value */
	man->size = size >> PAGE_SHIFT;
}

int amdgpu_mmap(struct file *filp, struct vm_area_struct *vma)
{
	struct drm_file *file_priv;
	struct amdgpu_device *adev;

	if (unlikely(vma->vm_pgoff < DRM_FILE_PAGE_OFFSET))
		return -EINVAL;

	file_priv = filp->private_data;
	adev = file_priv->minor->dev->dev_private;
	if (adev == NULL)
		return -EINVAL;

	return ttm_bo_mmap(filp, vma, &adev->mman.bdev);
}

static int amdgpu_map_buffer(struct ttm_buffer_object *bo,
			     struct ttm_mem_reg *mem, unsigned num_pages,
			     uint64_t offset, unsigned window,
			     struct amdgpu_ring *ring,
			     uint64_t *addr)
{
	struct amdgpu_ttm_tt *gtt = (void *)bo->ttm;
	struct amdgpu_device *adev = ring->adev;
	struct ttm_tt *ttm = bo->ttm;
	struct amdgpu_job *job;
	unsigned num_dw, num_bytes;
	dma_addr_t *dma_address;
	struct dma_fence *fence;
	uint64_t src_addr, dst_addr;
	uint64_t flags;
	int r;

	BUG_ON(adev->mman.buffer_funcs->copy_max_bytes <
	       AMDGPU_GTT_MAX_TRANSFER_SIZE * 8);

	*addr = adev->mc.gart_start;
	*addr += (u64)window * AMDGPU_GTT_MAX_TRANSFER_SIZE *
		AMDGPU_GPU_PAGE_SIZE;

	num_dw = adev->mman.buffer_funcs->copy_num_dw;
	while (num_dw & 0x7)
		num_dw++;

	num_bytes = num_pages * 8;

	r = amdgpu_job_alloc_with_ib(adev, num_dw * 4 + num_bytes, &job);
	if (r)
		return r;

	src_addr = num_dw * 4;
	src_addr += job->ibs[0].gpu_addr;

	dst_addr = adev->gart.table_addr;
	dst_addr += window * AMDGPU_GTT_MAX_TRANSFER_SIZE * 8;
	amdgpu_emit_copy_buffer(adev, &job->ibs[0], src_addr,
				dst_addr, num_bytes);

	amdgpu_ring_pad_ib(ring, &job->ibs[0]);
	WARN_ON(job->ibs[0].length_dw > num_dw);

	dma_address = &gtt->ttm.dma_address[offset >> PAGE_SHIFT];
	flags = amdgpu_ttm_tt_pte_flags(adev, ttm, mem);
	r = amdgpu_gart_map(adev, 0, num_pages, dma_address, flags,
			    &job->ibs[0].ptr[num_dw]);
	if (r)
		goto error_free;

	r = amdgpu_job_submit(job, ring, &adev->mman.entity,
			      AMDGPU_FENCE_OWNER_UNDEFINED, &fence);
	if (r)
		goto error_free;

	dma_fence_put(fence);

	return r;

error_free:
	amdgpu_job_free(job);
	return r;
}

int amdgpu_copy_buffer(struct amdgpu_ring *ring, uint64_t src_offset,
		       uint64_t dst_offset, uint32_t byte_count,
		       struct reservation_object *resv,
		       struct dma_fence **fence, bool direct_submit,
		       bool vm_needs_flush)
{
	struct amdgpu_device *adev = ring->adev;
	struct amdgpu_job *job;

	uint32_t max_bytes;
	unsigned num_loops, num_dw;
	unsigned i;
	int r;

	max_bytes = adev->mman.buffer_funcs->copy_max_bytes;
	num_loops = DIV_ROUND_UP(byte_count, max_bytes);
	num_dw = num_loops * adev->mman.buffer_funcs->copy_num_dw;

	/* for IB padding */
	while (num_dw & 0x7)
		num_dw++;

	r = amdgpu_job_alloc_with_ib(adev, num_dw * 4, &job);
	if (r)
		return r;

	job->vm_needs_flush = vm_needs_flush;
	if (resv) {
		r = amdgpu_sync_resv(adev, &job->sync, resv,
				     AMDGPU_FENCE_OWNER_UNDEFINED,
				     false);
		if (r) {
			DRM_ERROR("sync failed (%d).\n", r);
			goto error_free;
		}
	}

	for (i = 0; i < num_loops; i++) {
		uint32_t cur_size_in_bytes = min(byte_count, max_bytes);

		amdgpu_emit_copy_buffer(adev, &job->ibs[0], src_offset,
					dst_offset, cur_size_in_bytes);

		src_offset += cur_size_in_bytes;
		dst_offset += cur_size_in_bytes;
		byte_count -= cur_size_in_bytes;
	}

	amdgpu_ring_pad_ib(ring, &job->ibs[0]);
	WARN_ON(job->ibs[0].length_dw > num_dw);
	if (direct_submit) {
		r = amdgpu_ib_schedule(ring, job->num_ibs, job->ibs,
				       NULL, fence);
		job->fence = dma_fence_get(*fence);
		if (r)
			DRM_ERROR("Error scheduling IBs (%d)\n", r);
		amdgpu_job_free(job);
	} else {
		r = amdgpu_job_submit(job, ring, &adev->mman.entity,
				      AMDGPU_FENCE_OWNER_UNDEFINED, fence);
		if (r)
			goto error_free;
	}

	return r;

error_free:
	amdgpu_job_free(job);
	return r;
}

int amdgpu_fill_buffer(struct amdgpu_bo *bo,
		       uint64_t src_data,
		       struct reservation_object *resv,
		       struct dma_fence **fence)
{
	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
	uint32_t max_bytes = 8 *
			adev->vm_manager.vm_pte_funcs->set_max_nums_pte_pde;
	struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring;

	struct drm_mm_node *mm_node;
	unsigned long num_pages;
	unsigned int num_loops, num_dw;

	struct amdgpu_job *job;
	int r;

	if (!ring->ready) {
		DRM_ERROR("Trying to clear memory with ring turned off.\n");
		return -EINVAL;
	}

	if (bo->tbo.mem.mem_type == TTM_PL_TT) {
		r = amdgpu_ttm_alloc_gart(&bo->tbo);
		if (r)
			return r;
	}

	num_pages = bo->tbo.num_pages;
	mm_node = bo->tbo.mem.mm_node;
	num_loops = 0;
	while (num_pages) {
		uint32_t byte_count = mm_node->size << PAGE_SHIFT;

		num_loops += DIV_ROUND_UP(byte_count, max_bytes);
		num_pages -= mm_node->size;
		++mm_node;
	}

	/* num of dwords for each SDMA_OP_PTEPDE cmd */
	num_dw = num_loops * adev->vm_manager.vm_pte_funcs->set_pte_pde_num_dw;

	/* for IB padding */
	num_dw += 64;

	r = amdgpu_job_alloc_with_ib(adev, num_dw * 4, &job);
	if (r)
		return r;

	if (resv) {
		r = amdgpu_sync_resv(adev, &job->sync, resv,
				     AMDGPU_FENCE_OWNER_UNDEFINED, false);
		if (r) {
			DRM_ERROR("sync failed (%d).\n", r);
			goto error_free;
		}
	}

	num_pages = bo->tbo.num_pages;
	mm_node = bo->tbo.mem.mm_node;

	while (num_pages) {
		uint32_t byte_count = mm_node->size << PAGE_SHIFT;
		uint64_t dst_addr;

		WARN_ONCE(byte_count & 0x7, "size should be a multiple of 8");

		dst_addr = amdgpu_mm_node_addr(&bo->tbo, mm_node, &bo->tbo.mem);
		while (byte_count) {
			uint32_t cur_size_in_bytes = min(byte_count, max_bytes);

			amdgpu_vm_set_pte_pde(adev, &job->ibs[0],
					dst_addr, 0,
					cur_size_in_bytes >> 3, 0,
					src_data);

			dst_addr += cur_size_in_bytes;
			byte_count -= cur_size_in_bytes;
		}

		num_pages -= mm_node->size;
		++mm_node;
	}

	amdgpu_ring_pad_ib(ring, &job->ibs[0]);
	WARN_ON(job->ibs[0].length_dw > num_dw);
	r = amdgpu_job_submit(job, ring, &adev->mman.entity,
			      AMDGPU_FENCE_OWNER_UNDEFINED, fence);
	if (r)
		goto error_free;

	return 0;

error_free:
	amdgpu_job_free(job);
	return r;
}

#if defined(CONFIG_DEBUG_FS)

static int amdgpu_mm_dump_table(struct seq_file *m, void *data)
{
	struct drm_info_node *node = (struct drm_info_node *)m->private;
	unsigned ttm_pl = *(int *)node->info_ent->data;
	struct drm_device *dev = node->minor->dev;
	struct amdgpu_device *adev = dev->dev_private;
	struct ttm_mem_type_manager *man = &adev->mman.bdev.man[ttm_pl];
	struct drm_printer p = drm_seq_file_printer(m);

	man->func->debug(man, &p);
	return 0;
}

static int ttm_pl_vram = TTM_PL_VRAM;
static int ttm_pl_tt = TTM_PL_TT;

static const struct drm_info_list amdgpu_ttm_debugfs_list[] = {
	{"amdgpu_vram_mm", amdgpu_mm_dump_table, 0, &ttm_pl_vram},
	{"amdgpu_gtt_mm", amdgpu_mm_dump_table, 0, &ttm_pl_tt},
	{"ttm_page_pool", ttm_page_alloc_debugfs, 0, NULL},
#ifdef CONFIG_SWIOTLB
	{"ttm_dma_page_pool", ttm_dma_page_alloc_debugfs, 0, NULL}
#endif
};

static ssize_t amdgpu_ttm_vram_read(struct file *f, char __user *buf,
				    size_t size, loff_t *pos)
{
	struct amdgpu_device *adev = file_inode(f)->i_private;
	ssize_t result = 0;
	int r;

	if (size & 0x3 || *pos & 0x3)
		return -EINVAL;

	if (*pos >= adev->mc.mc_vram_size)
		return -ENXIO;

	while (size) {
		unsigned long flags;
		uint32_t value;

		if (*pos >= adev->mc.mc_vram_size)
			return result;

		spin_lock_irqsave(&adev->mmio_idx_lock, flags);
		WREG32_NO_KIQ(mmMM_INDEX, ((uint32_t)*pos) | 0x80000000);
		WREG32_NO_KIQ(mmMM_INDEX_HI, *pos >> 31);
		value = RREG32_NO_KIQ(mmMM_DATA);
		spin_unlock_irqrestore(&adev->mmio_idx_lock, flags);

		r = put_user(value, (uint32_t *)buf);
		if (r)
			return r;

		result += 4;
		buf += 4;
		*pos += 4;
		size -= 4;
	}

	return result;
}

static ssize_t amdgpu_ttm_vram_write(struct file *f, const char __user *buf,
				    size_t size, loff_t *pos)
{
	struct amdgpu_device *adev = file_inode(f)->i_private;
	ssize_t result = 0;
	int r;

	if (size & 0x3 || *pos & 0x3)
		return -EINVAL;

	if (*pos >= adev->mc.mc_vram_size)
		return -ENXIO;

	while (size) {
		unsigned long flags;
		uint32_t value;

		if (*pos >= adev->mc.mc_vram_size)
			return result;

		r = get_user(value, (uint32_t *)buf);
		if (r)
			return r;

		spin_lock_irqsave(&adev->mmio_idx_lock, flags);
		WREG32_NO_KIQ(mmMM_INDEX, ((uint32_t)*pos) | 0x80000000);
		WREG32_NO_KIQ(mmMM_INDEX_HI, *pos >> 31);
		WREG32_NO_KIQ(mmMM_DATA, value);
		spin_unlock_irqrestore(&adev->mmio_idx_lock, flags);

		result += 4;
		buf += 4;
		*pos += 4;
		size -= 4;
	}

	return result;
}

static const struct file_operations amdgpu_ttm_vram_fops = {
	.owner = THIS_MODULE,
	.read = amdgpu_ttm_vram_read,
	.write = amdgpu_ttm_vram_write,
	.llseek = default_llseek,
};

#ifdef CONFIG_DRM_AMDGPU_GART_DEBUGFS

static ssize_t amdgpu_ttm_gtt_read(struct file *f, char __user *buf,
				   size_t size, loff_t *pos)
{
	struct amdgpu_device *adev = file_inode(f)->i_private;
	ssize_t result = 0;
	int r;

	while (size) {
		loff_t p = *pos / PAGE_SIZE;
		unsigned off = *pos & ~PAGE_MASK;
		size_t cur_size = min_t(size_t, size, PAGE_SIZE - off);
		struct page *page;
		void *ptr;

		if (p >= adev->gart.num_cpu_pages)
			return result;

		page = adev->gart.pages[p];
		if (page) {
			ptr = kmap(page);
			ptr += off;

			r = copy_to_user(buf, ptr, cur_size);
			kunmap(adev->gart.pages[p]);
		} else
			r = clear_user(buf, cur_size);

		if (r)
			return -EFAULT;

		result += cur_size;
		buf += cur_size;
		*pos += cur_size;
		size -= cur_size;
	}

	return result;
}

static const struct file_operations amdgpu_ttm_gtt_fops = {
	.owner = THIS_MODULE,
	.read = amdgpu_ttm_gtt_read,
	.llseek = default_llseek
};

#endif

static ssize_t amdgpu_iova_to_phys_read(struct file *f, char __user *buf,
				   size_t size, loff_t *pos)
{
	struct amdgpu_device *adev = file_inode(f)->i_private;
	int r;
	uint64_t phys;
	struct iommu_domain *dom;

	// always return 8 bytes
	if (size != 8)
		return -EINVAL;

	// only accept page addresses
	if (*pos & 0xFFF)
		return -EINVAL;

	dom = iommu_get_domain_for_dev(adev->dev);
	if (dom)
		phys = iommu_iova_to_phys(dom, *pos);
	else
		phys = *pos;

	r = copy_to_user(buf, &phys, 8);
	if (r)
		return -EFAULT;

	return 8;
}

static const struct file_operations amdgpu_ttm_iova_fops = {
	.owner = THIS_MODULE,
	.read = amdgpu_iova_to_phys_read,
	.llseek = default_llseek
};

static const struct {
	char *name;
	const struct file_operations *fops;
	int domain;
} ttm_debugfs_entries[] = {
	{ "amdgpu_vram", &amdgpu_ttm_vram_fops, TTM_PL_VRAM },
#ifdef CONFIG_DRM_AMDGPU_GART_DEBUGFS
	{ "amdgpu_gtt", &amdgpu_ttm_gtt_fops, TTM_PL_TT },
#endif
	{ "amdgpu_iova", &amdgpu_ttm_iova_fops, TTM_PL_SYSTEM },
};

#endif

static int amdgpu_ttm_debugfs_init(struct amdgpu_device *adev)
{
#if defined(CONFIG_DEBUG_FS)
	unsigned count;

	struct drm_minor *minor = adev->ddev->primary;
	struct dentry *ent, *root = minor->debugfs_root;

	for (count = 0; count < ARRAY_SIZE(ttm_debugfs_entries); count++) {
		ent = debugfs_create_file(
				ttm_debugfs_entries[count].name,
				S_IFREG | S_IRUGO, root,
				adev,
				ttm_debugfs_entries[count].fops);
		if (IS_ERR(ent))
			return PTR_ERR(ent);
		if (ttm_debugfs_entries[count].domain == TTM_PL_VRAM)
			i_size_write(ent->d_inode, adev->mc.mc_vram_size);
		else if (ttm_debugfs_entries[count].domain == TTM_PL_TT)
			i_size_write(ent->d_inode, adev->mc.gart_size);
		adev->mman.debugfs_entries[count] = ent;
	}

	count = ARRAY_SIZE(amdgpu_ttm_debugfs_list);

#ifdef CONFIG_SWIOTLB
	if (!swiotlb_nr_tbl())
		--count;
#endif

	return amdgpu_debugfs_add_files(adev, amdgpu_ttm_debugfs_list, count);
#else
	return 0;
#endif
}

static void amdgpu_ttm_debugfs_fini(struct amdgpu_device *adev)
{
#if defined(CONFIG_DEBUG_FS)
	unsigned i;

	for (i = 0; i < ARRAY_SIZE(ttm_debugfs_entries); i++)
		debugfs_remove(adev->mman.debugfs_entries[i]);
#endif
}