xref: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c (revision 546b928da0427b0d6c663cbb992bd7bfa9ac7971)

/*
 * 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 <glisse@freedesktop.org>
 *    Thomas Hellstrom <thomas-at-tungstengraphics-dot-com>
 *    Dave Airlie
 */

#include <linux/dma-mapping.h>
#include <linux/iommu.h>
#include <linux/pagemap.h>
#include <linux/sched/task.h>
#include <linux/sched/mm.h>
#include <linux/seq_file.h>
#include <linux/slab.h>
#include <linux/swap.h>
#include <linux/dma-buf.h>
#include <linux/sizes.h>
#include <linux/module.h>

#include <drm/drm_drv.h>
#include <drm/ttm/ttm_bo.h>
#include <drm/ttm/ttm_placement.h>
#include <drm/ttm/ttm_range_manager.h>
#include <drm/ttm/ttm_tt.h>

#include <drm/amdgpu_drm.h>

#include "amdgpu.h"
#include "amdgpu_object.h"
#include "amdgpu_trace.h"
#include "amdgpu_amdkfd.h"
#include "amdgpu_sdma.h"
#include "amdgpu_ras.h"
#include "amdgpu_hmm.h"
#include "amdgpu_atomfirmware.h"
#include "amdgpu_res_cursor.h"
#include "bif/bif_4_1_d.h"
#include "kfd_svm.h"

MODULE_IMPORT_NS("DMA_BUF");

#define AMDGPU_TTM_VRAM_MAX_DW_READ	((size_t)128)

static int amdgpu_ttm_backend_bind(struct ttm_device *bdev,
				   struct ttm_tt *ttm,
				   struct ttm_resource *bo_mem);
static void amdgpu_ttm_backend_unbind(struct ttm_device *bdev,
				      struct ttm_tt *ttm);

static int amdgpu_ttm_init_on_chip(struct amdgpu_device *adev,
				    unsigned int type,
				    uint64_t size_in_page)
{
	if (!size_in_page)
		return 0;

	return ttm_range_man_init(&adev->mman.bdev, type,
				  false, size_in_page);
}

/**
 * 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,
				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,
		.mem_type = TTM_PL_SYSTEM,
		.flags = 0
	};

	/* Don't handle scatter gather BOs */
	if (bo->type == ttm_bo_type_sg) {
		placement->num_placement = 0;
		return;
	}

	/* Object isn't an AMDGPU object so ignore */
	if (!amdgpu_bo_is_amdgpu_bo(bo)) {
		placement->placement = &placements;
		placement->num_placement = 1;
		return;
	}

	abo = ttm_to_amdgpu_bo(bo);
	if (abo->flags & AMDGPU_GEM_CREATE_DISCARDABLE) {
		placement->num_placement = 0;
		return;
	}

	switch (bo->resource->mem_type) {
	case AMDGPU_PL_GDS:
	case AMDGPU_PL_GWS:
	case AMDGPU_PL_OA:
	case AMDGPU_PL_DOORBELL:
	case AMDGPU_PL_MMIO_REMAP:
		placement->num_placement = 0;
		return;

	case TTM_PL_VRAM:
		if (!adev->mman.buffer_funcs_enabled) {
			/* Move to system memory */
			amdgpu_bo_placement_from_domain(abo, AMDGPU_GEM_DOMAIN_CPU);

		} else if (!amdgpu_gmc_vram_full_visible(&adev->gmc) &&
			   !(abo->flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED) &&
			   amdgpu_res_cpu_visible(adev, bo->resource)) {

			/* 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_bo_placement_from_domain(abo, AMDGPU_GEM_DOMAIN_VRAM |
							AMDGPU_GEM_DOMAIN_GTT |
							AMDGPU_GEM_DOMAIN_CPU);
			abo->placements[0].fpfn = adev->gmc.visible_vram_size >> PAGE_SHIFT;
			abo->placements[0].lpfn = 0;
			abo->placements[0].flags |= TTM_PL_FLAG_DESIRED;
		} else {
			/* Move to GTT memory */
			amdgpu_bo_placement_from_domain(abo, AMDGPU_GEM_DOMAIN_GTT |
							AMDGPU_GEM_DOMAIN_CPU);
		}
		break;
	case TTM_PL_TT:
	case AMDGPU_PL_PREEMPT:
	default:
		amdgpu_bo_placement_from_domain(abo, AMDGPU_GEM_DOMAIN_CPU);
		break;
	}
	*placement = abo->placement;
}

static struct dma_fence *
amdgpu_ttm_job_submit(struct amdgpu_device *adev, struct amdgpu_ttm_buffer_entity *entity,
		      struct amdgpu_job *job, u32 num_dw)
{
	struct amdgpu_ring *ring;

	ring = to_amdgpu_ring(adev->mman.buffer_funcs_scheds[0]);
	amdgpu_ring_pad_ib(ring, &job->ibs[0]);
	WARN_ON(job->ibs[0].length_dw > num_dw);

	lockdep_assert_held(&entity->lock);

	return amdgpu_job_submit(job);
}

/**
 * amdgpu_ttm_map_buffer - Map memory into the GART windows
 * @entity: entity to run the window setup job
 * @bo: buffer object to map
 * @mem: memory object to map
 * @mm_cur: range to map
 * @window: which GART window to use
 * @tmz: if we should setup a TMZ enabled mapping
 * @size: in number of bytes to map, out number of bytes mapped
 * @addr: resulting address inside the MC address space
 * @vm_needs_flush: out, set true if a GART window was programmed (VMID 0 flush
 *		    needed) or false for a direct address
 *
 * Setup one of the GART windows to access a specific piece of memory or return
 * the physical address for local memory.
 */
static int amdgpu_ttm_map_buffer(struct amdgpu_ttm_buffer_entity *entity,
				 struct ttm_buffer_object *bo,
				 struct ttm_resource *mem,
				 struct amdgpu_res_cursor *mm_cur,
				 unsigned int window,
				 bool tmz, uint64_t *size, uint64_t *addr,
				 bool *vm_needs_flush)
{
	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
	unsigned int offset, num_pages, num_dw, num_bytes;
	uint64_t src_addr, dst_addr;
	struct amdgpu_job *job;
	void *cpu_addr;
	uint64_t flags;
	int r;
	const u64 GTT_MAX_PAGES = (AMDGPU_GTT_MAX_TRANSFER_SIZE >> PAGE_SHIFT);

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

	if (WARN_ON(mem->mem_type == AMDGPU_PL_PREEMPT))
		return -EINVAL;

	/* Map only what can't be accessed directly */
	if (!tmz && mem->start != AMDGPU_BO_INVALID_OFFSET) {
		*addr = amdgpu_ttm_domain_start(adev, mem->mem_type) +
			mm_cur->start;
		*vm_needs_flush = false;
		return 0;
	}

	/* A GART window is programmed below, so its VMID 0 TLB needs a flush */
	*vm_needs_flush = true;

	/*
	 * If start begins at an offset inside the page, then adjust the size
	 * and addr accordingly
	 */
	offset = mm_cur->start & ~PAGE_MASK;

	num_pages = PFN_UP(*size + offset);
	num_pages = min_t(uint32_t, num_pages, GTT_MAX_PAGES);

	*size = min(*size, (uint64_t)num_pages * PAGE_SIZE - offset);

	*addr = amdgpu_compute_gart_address(&adev->gmc, entity, window);
	*addr += offset;

	num_dw = ALIGN(adev->mman.buffer_funcs->copy_num_dw, 8);
	num_bytes = num_pages * 8 * AMDGPU_GPU_PAGES_IN_CPU_PAGE;

	r = amdgpu_job_alloc_with_ib(adev, &entity->base,
				     AMDGPU_FENCE_OWNER_UNDEFINED,
				     num_dw * 4 + num_bytes,
				     AMDGPU_IB_POOL_DELAYED,
				     AMDGPU_KERNEL_JOB_ID_TTM_MAP_BUFFER,
				     &job);
	if (r)
		return r;

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

	dst_addr = amdgpu_bo_gpu_offset(adev->gart.bo);
	dst_addr += (entity->gart_window_offs[window] >> AMDGPU_GPU_PAGE_SHIFT) * 8;
	amdgpu_emit_copy_buffer(adev, &job->ibs[0], src_addr,
				dst_addr, num_bytes, 0);

	flags = amdgpu_ttm_tt_pte_flags(adev, bo->ttm, mem);
	if (tmz)
		flags |= AMDGPU_PTE_TMZ;

	cpu_addr = &job->ibs[0].ptr[num_dw];

	if (mem->mem_type == TTM_PL_TT) {
		dma_addr_t *dma_addr;

		dma_addr = &bo->ttm->dma_address[mm_cur->start >> PAGE_SHIFT];
		amdgpu_gart_map(adev, 0, num_pages, dma_addr, flags, cpu_addr);
	} else {
		u64 pa = mm_cur->start + adev->vm_manager.vram_base_offset;

		amdgpu_gart_map_vram_range(adev, pa, 0, num_pages, flags, cpu_addr);
	}

	dma_fence_put(amdgpu_ttm_job_submit(adev, entity, job, num_dw));
	return 0;
}

/**
 * amdgpu_ttm_copy_mem_to_mem - Helper function for copy
 * @adev: amdgpu device
 * @entity: entity to run the jobs
 * @src: buffer/address where to read from
 * @dst: buffer/address where to write to
 * @size: number of bytes to copy
 * @tmz: if a secure copy should be used
 * @resv: resv object to sync to
 * @f: Returns the last fence if multiple jobs are submitted.
 *
 * 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.
 *
 */
__attribute__((nonnull))
static int amdgpu_ttm_copy_mem_to_mem(struct amdgpu_device *adev,
				      struct amdgpu_ttm_buffer_entity *entity,
				      const struct amdgpu_copy_mem *src,
				      const struct amdgpu_copy_mem *dst,
				      uint64_t size, bool tmz,
				      struct dma_resv *resv,
				      struct dma_fence **f)
{
	struct amdgpu_res_cursor src_mm, dst_mm;
	struct dma_fence *fence = NULL;
	int r = 0;
	uint32_t copy_flags = 0;
	struct amdgpu_bo *abo_src, *abo_dst;

	if (!adev->mman.buffer_funcs_enabled) {
		dev_err(adev->dev,
			"Trying to move memory with ring turned off.\n");
		return -EINVAL;
	}

	amdgpu_res_first(src->mem, src->offset, size, &src_mm);
	amdgpu_res_first(dst->mem, dst->offset, size, &dst_mm);

	mutex_lock(&entity->lock);
	while (src_mm.remaining) {
		uint64_t from, to, cur_size, tiling_flags;
		uint32_t num_type, data_format, max_com, write_compress_disable;
		bool src_vm_flush, dst_vm_flush;
		struct dma_fence *next;

		/* Never copy more than 256MiB at once to avoid a timeout */
		cur_size = min3(src_mm.size, dst_mm.size, 256ULL << 20);

		/* Map src to window 0 and dst to window 1. */
		r = amdgpu_ttm_map_buffer(entity, src->bo, src->mem, &src_mm,
					  0, tmz, &cur_size, &from, &src_vm_flush);
		if (r)
			goto error;

		r = amdgpu_ttm_map_buffer(entity, dst->bo, dst->mem, &dst_mm,
					  1, tmz, &cur_size, &to, &dst_vm_flush);
		if (r)
			goto error;

		abo_src = ttm_to_amdgpu_bo(src->bo);
		abo_dst = ttm_to_amdgpu_bo(dst->bo);
		if (tmz)
			copy_flags |= AMDGPU_COPY_FLAGS_TMZ;
		if ((abo_src->flags & AMDGPU_GEM_CREATE_GFX12_DCC) &&
		    (abo_src->tbo.resource->mem_type == TTM_PL_VRAM))
			copy_flags |= AMDGPU_COPY_FLAGS_READ_DECOMPRESSED;
		if ((abo_dst->flags & AMDGPU_GEM_CREATE_GFX12_DCC) &&
		    (dst->mem->mem_type == TTM_PL_VRAM)) {
			copy_flags |= AMDGPU_COPY_FLAGS_WRITE_COMPRESSED;
			amdgpu_bo_get_tiling_flags(abo_dst, &tiling_flags);
			max_com = AMDGPU_TILING_GET(tiling_flags, GFX12_DCC_MAX_COMPRESSED_BLOCK);
			num_type = AMDGPU_TILING_GET(tiling_flags, GFX12_DCC_NUMBER_TYPE);
			data_format = AMDGPU_TILING_GET(tiling_flags, GFX12_DCC_DATA_FORMAT);
			write_compress_disable =
				AMDGPU_TILING_GET(tiling_flags, GFX12_DCC_WRITE_COMPRESS_DISABLE);
			copy_flags |= (AMDGPU_COPY_FLAGS_SET(MAX_COMPRESSED, max_com) |
				       AMDGPU_COPY_FLAGS_SET(NUMBER_TYPE, num_type) |
				       AMDGPU_COPY_FLAGS_SET(DATA_FORMAT, data_format) |
				       AMDGPU_COPY_FLAGS_SET(WRITE_COMPRESS_DISABLE,
							     write_compress_disable));
		}

		r = amdgpu_copy_buffer(adev, entity, from, to, cur_size, resv,
				       &next, src_vm_flush || dst_vm_flush, copy_flags);
		if (r)
			goto error;

		dma_fence_put(fence);
		fence = next;

		amdgpu_res_next(&src_mm, cur_size);
		amdgpu_res_next(&dst_mm, cur_size);
	}
error:
	mutex_unlock(&entity->lock);
	*f = fence;
	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,
			    bool evict,
			    struct ttm_resource *new_mem,
			    struct ttm_resource *old_mem)
{
	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
	struct amdgpu_bo *abo = ttm_to_amdgpu_bo(bo);
	struct amdgpu_ttm_buffer_entity *entity;
	struct amdgpu_copy_mem src, dst;
	struct dma_fence *fence = NULL;
	int r;
	u32 e;

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

	e = atomic_inc_return(&adev->mman.next_move_entity) %
			      adev->mman.num_move_entities;
	entity = &adev->mman.move_entities[e];

	r = amdgpu_ttm_copy_mem_to_mem(adev,
				       entity,
				       &src, &dst,
				       new_mem->size,
				       amdgpu_bo_encrypted(abo),
				       bo->base.resv, &fence);
	if (r)
		goto error;

	/* clear the space being freed */
	if (old_mem->mem_type == TTM_PL_VRAM &&
	    (abo->flags & AMDGPU_GEM_CREATE_VRAM_WIPE_ON_RELEASE)) {
		struct dma_fence *wipe_fence = NULL;
		r = amdgpu_ttm_clear_buffer(entity, abo, NULL, &wipe_fence,
					    false, AMDGPU_KERNEL_JOB_ID_MOVE_BLIT);
		if (r) {
			goto error;
		} else if (wipe_fence) {
			amdgpu_vram_mgr_set_cleared(bo->resource);
			dma_fence_put(fence);
			fence = wipe_fence;
		}
	}

	/* Always block for VM page tables before committing the new location */
	if (bo->type == ttm_bo_type_kernel)
		r = ttm_bo_move_accel_cleanup(bo, fence, true, false, new_mem);
	else
		r = ttm_bo_move_accel_cleanup(bo, fence, evict, true, new_mem);
	dma_fence_put(fence);
	return r;

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

/**
 * amdgpu_res_cpu_visible - Check that resource can be accessed by CPU
 * @adev: amdgpu device
 * @res: the resource to check
 *
 * Returns: true if the full resource is CPU visible, false otherwise.
 */
bool amdgpu_res_cpu_visible(struct amdgpu_device *adev,
			    struct ttm_resource *res)
{
	struct amdgpu_res_cursor cursor;

	if (!res)
		return false;

	if (res->mem_type == TTM_PL_SYSTEM || res->mem_type == TTM_PL_TT ||
	    res->mem_type == AMDGPU_PL_PREEMPT || res->mem_type == AMDGPU_PL_DOORBELL ||
	    res->mem_type == AMDGPU_PL_MMIO_REMAP)
		return true;

	if (res->mem_type != TTM_PL_VRAM)
		return false;

	amdgpu_res_first(res, 0, res->size, &cursor);
	while (cursor.remaining) {
		if ((cursor.start + cursor.size) > adev->gmc.visible_vram_size)
			return false;
		amdgpu_res_next(&cursor, cursor.size);
	}

	return true;
}

/*
 * amdgpu_res_copyable - Check that memory can be accessed by ttm_bo_move_memcpy
 *
 * Called by amdgpu_bo_move()
 */
static bool amdgpu_res_copyable(struct amdgpu_device *adev,
				struct ttm_resource *mem)
{
	if (!amdgpu_res_cpu_visible(adev, mem))
		return false;

	/* ttm_resource_ioremap only supports contiguous memory */
	if (mem->mem_type == TTM_PL_VRAM &&
	    !(mem->placement & TTM_PL_FLAG_CONTIGUOUS))
		return false;

	return true;
}

/*
 * 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,
			  struct ttm_operation_ctx *ctx,
			  struct ttm_resource *new_mem,
			  struct ttm_place *hop)
{
	struct amdgpu_device *adev;
	struct amdgpu_bo *abo;
	struct ttm_resource *old_mem = bo->resource;
	int r;

	if (new_mem->mem_type == TTM_PL_TT ||
	    new_mem->mem_type == AMDGPU_PL_PREEMPT) {
		if (old_mem && (old_mem->mem_type == TTM_PL_TT ||
				old_mem->mem_type == AMDGPU_PL_PREEMPT)) {
			r = ttm_bo_wait_ctx(bo, ctx);
			if (r)
				return r;

			amdgpu_ttm_backend_unbind(bo->bdev, bo->ttm);
		}

		r = amdgpu_ttm_backend_bind(bo->bdev, bo->ttm, new_mem);
		if (r)
			return r;
	}

	abo = ttm_to_amdgpu_bo(bo);
	adev = amdgpu_ttm_adev(bo->bdev);

	if (!old_mem || (old_mem->mem_type == TTM_PL_SYSTEM &&
			 bo->ttm == NULL)) {
		amdgpu_bo_move_notify(bo, evict, new_mem);
		ttm_bo_move_null(bo, new_mem);
		return 0;
	}
	if (old_mem->mem_type == TTM_PL_SYSTEM &&
	    (new_mem->mem_type == TTM_PL_TT ||
	     new_mem->mem_type == AMDGPU_PL_PREEMPT)) {
		amdgpu_bo_move_notify(bo, evict, new_mem);
		ttm_bo_move_null(bo, new_mem);
		return 0;
	}
	if ((old_mem->mem_type == TTM_PL_TT ||
	     old_mem->mem_type == AMDGPU_PL_PREEMPT) &&
	    new_mem->mem_type == TTM_PL_SYSTEM) {
		r = ttm_bo_wait_ctx(bo, ctx);
		if (r)
			return r;

		amdgpu_ttm_backend_unbind(bo->bdev, bo->ttm);
		amdgpu_bo_move_notify(bo, evict, new_mem);
		ttm_resource_free(bo, &bo->resource);
		ttm_bo_assign_mem(bo, new_mem);
		return 0;
	}
	if ((old_mem->mem_type == TTM_PL_TT ||
	     old_mem->mem_type == AMDGPU_PL_PREEMPT) &&
	    (new_mem->mem_type == TTM_PL_TT ||
	     new_mem->mem_type == AMDGPU_PL_PREEMPT)) {
		amdgpu_bo_move_notify(bo, evict, new_mem);
		ttm_resource_free(bo, &bo->resource);
		ttm_bo_assign_mem(bo, new_mem);
		return 0;
	}

	if (old_mem->mem_type == AMDGPU_PL_GDS ||
	    old_mem->mem_type == AMDGPU_PL_GWS ||
	    old_mem->mem_type == AMDGPU_PL_OA ||
	    old_mem->mem_type == AMDGPU_PL_DOORBELL ||
	    old_mem->mem_type == AMDGPU_PL_MMIO_REMAP ||
	    new_mem->mem_type == AMDGPU_PL_GDS ||
	    new_mem->mem_type == AMDGPU_PL_GWS ||
	    new_mem->mem_type == AMDGPU_PL_OA ||
	    new_mem->mem_type == AMDGPU_PL_DOORBELL ||
	    new_mem->mem_type == AMDGPU_PL_MMIO_REMAP) {
		/* Nothing to save here */
		amdgpu_bo_move_notify(bo, evict, new_mem);
		ttm_bo_move_null(bo, new_mem);
		return 0;
	}

	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;
	}

	if (adev->mman.buffer_funcs_enabled &&
	    ((old_mem->mem_type == TTM_PL_SYSTEM &&
	      new_mem->mem_type == TTM_PL_VRAM) ||
	     (old_mem->mem_type == TTM_PL_VRAM &&
	      new_mem->mem_type == TTM_PL_SYSTEM))) {
		hop->fpfn = 0;
		hop->lpfn = 0;
		hop->mem_type = TTM_PL_TT;
		hop->flags = TTM_PL_FLAG_TEMPORARY;
		return -EMULTIHOP;
	}

	amdgpu_bo_move_notify(bo, evict, new_mem);
	if (adev->mman.buffer_funcs_enabled)
		r = amdgpu_move_blit(bo, evict, new_mem, old_mem);
	else
		r = -ENODEV;

	if (r) {
		/* Check that all memory is CPU accessible */
		if (!amdgpu_res_copyable(adev, old_mem) ||
		    !amdgpu_res_copyable(adev, new_mem)) {
			pr_err("Move buffer fallback to memcpy unavailable\n");
			return r;
		}

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

	/* update statistics after the move */
	if (evict)
		atomic64_inc(&adev->num_evictions);
	atomic64_add(bo->base.size, &adev->num_bytes_moved);
	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_device *bdev,
				     struct ttm_resource *mem)
{
	struct amdgpu_device *adev = amdgpu_ttm_adev(bdev);

	switch (mem->mem_type) {
	case TTM_PL_SYSTEM:
		/* system memory */
		return 0;
	case TTM_PL_TT:
	case AMDGPU_PL_PREEMPT:
		break;
	case TTM_PL_VRAM:
		mem->bus.offset = mem->start << PAGE_SHIFT;

		if (adev->mman.aper_base_kaddr &&
		    mem->placement & TTM_PL_FLAG_CONTIGUOUS)
			mem->bus.addr = (u8 *)adev->mman.aper_base_kaddr +
					mem->bus.offset;

		mem->bus.offset += adev->gmc.aper_base;
		mem->bus.is_iomem = true;
		break;
	case AMDGPU_PL_DOORBELL:
		mem->bus.offset = mem->start << PAGE_SHIFT;
		mem->bus.offset += adev->doorbell.base;
		mem->bus.is_iomem = true;
		mem->bus.caching = ttm_uncached;
		break;
	case AMDGPU_PL_MMIO_REMAP:
		mem->bus.offset = mem->start << PAGE_SHIFT;
		mem->bus.offset += adev->rmmio_remap.bus_addr;
		mem->bus.is_iomem = true;
		mem->bus.caching = ttm_uncached;
		break;
	default:
		return -EINVAL;
	}
	return 0;
}

static unsigned long amdgpu_ttm_io_mem_pfn(struct ttm_buffer_object *bo,
					   unsigned long page_offset)
{
	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
	struct amdgpu_res_cursor cursor;

	amdgpu_res_first(bo->resource, (u64)page_offset << PAGE_SHIFT, 0,
			 &cursor);

	if (bo->resource->mem_type == AMDGPU_PL_DOORBELL)
		return ((uint64_t)(adev->doorbell.base + cursor.start)) >> PAGE_SHIFT;
	else if (bo->resource->mem_type == AMDGPU_PL_MMIO_REMAP)
		return ((uint64_t)(adev->rmmio_remap.bus_addr + cursor.start)) >> PAGE_SHIFT;

	return (adev->gmc.aper_base + cursor.start) >> PAGE_SHIFT;
}

/**
 * amdgpu_ttm_domain_start - Returns GPU start address
 * @adev: amdgpu device object
 * @type: type of the memory
 *
 * Returns:
 * GPU start address of a memory domain
 */

uint64_t amdgpu_ttm_domain_start(struct amdgpu_device *adev, uint32_t type)
{
	switch (type) {
	case TTM_PL_TT:
		return adev->gmc.gart_start;
	case TTM_PL_VRAM:
		return adev->gmc.vram_start;
	}

	return 0;
}

/*
 * TTM backend functions.
 */
struct amdgpu_ttm_tt {
	struct ttm_tt	ttm;
	struct drm_gem_object	*gobj;
	u64			offset;
	uint64_t		userptr;
	struct task_struct	*usertask;
	uint32_t		userflags;
	bool			bound;
	int32_t			pool_id;
};

#define ttm_to_amdgpu_ttm_tt(ptr)	container_of(ptr, struct amdgpu_ttm_tt, ttm)

#ifdef CONFIG_DRM_AMDGPU_USERPTR
/*
 * amdgpu_ttm_tt_get_user_pages - get device accessible pages that back user
 * memory and start HMM tracking CPU page table update
 *
 * Calling function must call amdgpu_ttm_tt_userptr_range_done() once and only
 * once afterwards to stop HMM tracking. Its the caller responsibility to ensure
 * that range is a valid memory and it is freed too.
 */
int amdgpu_ttm_tt_get_user_pages(struct amdgpu_bo *bo,
				 struct amdgpu_hmm_range *range)
{
	struct ttm_tt *ttm = bo->tbo.ttm;
	struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm);
	unsigned long start = gtt->userptr;
	struct vm_area_struct *vma;
	struct mm_struct *mm;
	bool readonly;
	int r = 0;

	mm = bo->notifier.mm;
	if (unlikely(!mm)) {
		DRM_DEBUG_DRIVER("BO is not registered?\n");
		return -EFAULT;
	}

	if (!mmget_not_zero(mm)) /* Happens during process shutdown */
		return -ESRCH;

	mmap_read_lock(mm);
	vma = vma_lookup(mm, start);
	if (unlikely(!vma)) {
		r = -EFAULT;
		goto out_unlock;
	}
	if (unlikely((gtt->userflags & AMDGPU_GEM_USERPTR_ANONONLY) &&
		vma->vm_file)) {
		r = -EPERM;
		goto out_unlock;
	}

	readonly = amdgpu_ttm_tt_is_readonly(ttm);
	r = amdgpu_hmm_range_get_pages(&bo->notifier, start, ttm->num_pages,
				       readonly, NULL, range);
out_unlock:
	mmap_read_unlock(mm);
	if (r)
		pr_debug("failed %d to get user pages 0x%lx\n", r, start);

	mmput(mm);

	return r;
}

#endif

/*
 * 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 amdgpu_hmm_range *range)
{
	unsigned long i;

	for (i = 0; i < ttm->num_pages; ++i)
		ttm->pages[i] = range ? hmm_pfn_to_page(range->hmm_range.hmm_pfns[i]) : NULL;
}

/*
 * 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_device *bdev,
				     struct ttm_tt *ttm)
{
	struct amdgpu_device *adev = amdgpu_ttm_adev(bdev);
	struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm);
	int write = !(gtt->userflags & AMDGPU_GEM_USERPTR_READONLY);
	enum dma_data_direction direction = write ?
		DMA_BIDIRECTIONAL : DMA_TO_DEVICE;
	int r;

	/* Allocate an SG array and squash pages into it */
	r = sg_alloc_table_from_pages(ttm->sg, ttm->pages, ttm->num_pages, 0,
				      (u64)ttm->num_pages << PAGE_SHIFT,
				      GFP_KERNEL);
	if (r)
		goto release_sg;

	/* Map SG to device */
	r = dma_map_sgtable(adev->dev, ttm->sg, direction, 0);
	if (r)
		goto release_sg_table;

	/* convert SG to linear array of pages and dma addresses */
	drm_prime_sg_to_dma_addr_array(ttm->sg, gtt->ttm.dma_address,
				       ttm->num_pages);

	return 0;

release_sg_table:
	sg_free_table(ttm->sg);
release_sg:
	kfree(ttm->sg);
	ttm->sg = NULL;
	return r;
}

/*
 * amdgpu_ttm_tt_unpin_userptr - Unpin and unmap userptr pages
 */
static void amdgpu_ttm_tt_unpin_userptr(struct ttm_device *bdev,
					struct ttm_tt *ttm)
{
	struct amdgpu_device *adev = amdgpu_ttm_adev(bdev);
	struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(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 || !ttm->sg->sgl)
		return;

	/* unmap the pages mapped to the device */
	dma_unmap_sgtable(adev->dev, ttm->sg, direction, 0);
	sg_free_table(ttm->sg);
}

/*
 * total_pages is constructed as MQD0+CtrlStack0 + MQD1+CtrlStack1 + ...
 * MQDn+CtrlStackn where n is the number of XCCs per partition.
 * pages_per_xcc is the size of one MQD+CtrlStack. The first page is MQD
 * and uses memory type default, UC. The rest of pages_per_xcc are
 * Ctrl stack and modify their memory type to NC.
 */
static void amdgpu_ttm_gart_bind_gfx9_mqd(struct amdgpu_device *adev,
				struct ttm_tt *ttm, uint64_t flags)
{
	struct amdgpu_ttm_tt *gtt = (void *)ttm;
	uint64_t total_pages = ttm->num_pages;
	int num_xcc = max(1U, adev->gfx.num_xcc_per_xcp);
	uint64_t page_idx, pages_per_xcc;
	int i;

	pages_per_xcc = total_pages;
	do_div(pages_per_xcc, num_xcc);

	for (i = 0, page_idx = 0; i < num_xcc; i++, page_idx += pages_per_xcc) {
		amdgpu_gart_map_gfx9_mqd(adev,
				gtt->offset + (page_idx << PAGE_SHIFT),
				pages_per_xcc, &gtt->ttm.dma_address[page_idx],
				flags);
	}
}

static void amdgpu_ttm_gart_bind(struct amdgpu_device *adev,
				 struct ttm_buffer_object *tbo,
				 uint64_t flags)
{
	struct amdgpu_bo *abo = ttm_to_amdgpu_bo(tbo);
	struct ttm_tt *ttm = tbo->ttm;
	struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm);

	if (amdgpu_bo_encrypted(abo))
		flags |= AMDGPU_PTE_TMZ;

	if (abo->flags & AMDGPU_GEM_CREATE_CP_MQD_GFX9) {
		amdgpu_ttm_gart_bind_gfx9_mqd(adev, ttm, flags);
	} else {
		amdgpu_gart_bind(adev, gtt->offset, ttm->num_pages,
				 gtt->ttm.dma_address, flags);
	}
	gtt->bound = true;
}

/*
 * 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_device *bdev,
				   struct ttm_tt *ttm,
				   struct ttm_resource *bo_mem)
{
	struct amdgpu_device *adev = amdgpu_ttm_adev(bdev);
	struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm);
	uint64_t flags;
	int r;

	if (!bo_mem)
		return -EINVAL;

	if (gtt->bound)
		return 0;

	if (gtt->userptr) {
		r = amdgpu_ttm_tt_pin_userptr(bdev, ttm);
		if (r) {
			dev_err(adev->dev, "failed to pin userptr\n");
			return r;
		}
	} else if (ttm->page_flags & TTM_TT_FLAG_EXTERNAL) {
		if (!ttm->sg) {
			struct dma_buf_attachment *attach;
			struct sg_table *sgt;

			attach = gtt->gobj->import_attach;
			sgt = dma_buf_map_attachment(attach, DMA_BIDIRECTIONAL);
			if (IS_ERR(sgt))
				return PTR_ERR(sgt);

			ttm->sg = sgt;
		}

		drm_prime_sg_to_dma_addr_array(ttm->sg, gtt->ttm.dma_address,
					       ttm->num_pages);
	}

	if (!ttm->num_pages) {
		WARN(1, "nothing to bind %u pages for mreg %p back %p!\n",
		     ttm->num_pages, bo_mem, ttm);
	}

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

	/* compute PTE flags relevant to this BO memory */
	flags = amdgpu_ttm_tt_pte_flags(adev, ttm, bo_mem);

	/* bind pages into GART page tables */
	gtt->offset = (u64)bo_mem->start << PAGE_SHIFT;
	amdgpu_gart_bind(adev, gtt->offset, ttm->num_pages,
			 gtt->ttm.dma_address, flags);
	gtt->bound = true;
	return 0;
}

/*
 * amdgpu_ttm_alloc_gart - Make sure buffer object is accessible either
 * through AGP or GART aperture.
 *
 * If bo is accessible through AGP aperture, then use AGP aperture
 * to access bo; otherwise allocate logical space in GART aperture
 * and map bo to GART aperture.
 */
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 = ttm_to_amdgpu_ttm_tt(bo->ttm);
	struct ttm_placement placement;
	struct ttm_place placements;
	struct ttm_resource *tmp;
	uint64_t addr, flags;
	int r;

	if (bo->resource->start != AMDGPU_BO_INVALID_OFFSET)
		return 0;

	addr = amdgpu_gmc_agp_addr(bo);
	if (addr != AMDGPU_BO_INVALID_OFFSET)
		return 0;

	/* allocate GART space */
	placement.num_placement = 1;
	placement.placement = &placements;
	placements.fpfn = 0;
	placements.lpfn = adev->gmc.gart_size >> PAGE_SHIFT;
	placements.mem_type = TTM_PL_TT;
	placements.flags = bo->resource->placement;

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

	/* compute PTE flags for this buffer object */
	flags = amdgpu_ttm_tt_pte_flags(adev, bo->ttm, tmp);

	/* Bind pages */
	gtt->offset = (u64)tmp->start << PAGE_SHIFT;
	amdgpu_ttm_gart_bind(adev, bo, flags);
	amdgpu_gart_invalidate_tlb(adev);
	ttm_resource_free(bo, &bo->resource);
	ttm_bo_assign_mem(bo, tmp);

	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.
 */
void amdgpu_ttm_recover_gart(struct ttm_buffer_object *tbo)
{
	struct amdgpu_device *adev = amdgpu_ttm_adev(tbo->bdev);
	uint64_t flags;

	if (!tbo->ttm)
		return;

	flags = amdgpu_ttm_tt_pte_flags(adev, tbo->ttm, tbo->resource);
	amdgpu_ttm_gart_bind(adev, tbo, flags);
}

/*
 * 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 void amdgpu_ttm_backend_unbind(struct ttm_device *bdev,
				      struct ttm_tt *ttm)
{
	struct amdgpu_device *adev = amdgpu_ttm_adev(bdev);
	struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm);

	/* if the pages have userptr pinning then clear that first */
	if (gtt->userptr) {
		amdgpu_ttm_tt_unpin_userptr(bdev, ttm);
	} else if (ttm->sg && drm_gem_is_imported(gtt->gobj)) {
		struct dma_buf_attachment *attach;

		attach = gtt->gobj->import_attach;
		dma_buf_unmap_attachment(attach, ttm->sg, DMA_BIDIRECTIONAL);
		ttm->sg = NULL;
	}

	if (!gtt->bound)
		return;

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

	/* unbind shouldn't be done for GDS/GWS/OA in ttm_bo_clean_mm */
	amdgpu_gart_unbind(adev, gtt->offset, ttm->num_pages);
	gtt->bound = false;
}

static void amdgpu_ttm_backend_destroy(struct ttm_device *bdev,
				       struct ttm_tt *ttm)
{
	struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm);

	if (gtt->usertask)
		put_task_struct(gtt->usertask);

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

/**
 * amdgpu_ttm_mmio_remap_alloc_sgt - build an sg_table for MMIO_REMAP I/O aperture
 * @adev: amdgpu device providing the remap BAR base (adev->rmmio_remap.bus_addr)
 * @res:  TTM resource of the BO to export; expected to live in AMDGPU_PL_MMIO_REMAP
 * @dev:  importing device to map for (typically @attach->dev in dma-buf paths)
 * @dir:  DMA data direction for the importer (passed to dma_map_resource())
 * @sgt:  output; on success, set to a newly allocated sg_table describing the I/O span
 *
 * The HDP flush page (AMDGPU_PL_MMIO_REMAP) is a fixed hardware I/O window in a PCI
 * BAR—there are no struct pages to back it. Importers still need a DMA address list,
 * so we synthesize a minimal sg_table and populate it from dma_map_resource(), not
 * from pages. Using the common amdgpu_res_cursor walker keeps the offset/size math
 * consistent with other TTM/manager users.
 *
 * - @res is assumed to be a small, contiguous I/O region (typically a single 4 KiB
 *   page) in AMDGPU_PL_MMIO_REMAP. Callers should validate placement before calling.
 * - The sg entry is created with sg_set_page(sg, NULL, …) to reflect I/O space.
 * - The mapping uses DMA_ATTR_SKIP_CPU_SYNC because this is MMIO, not cacheable RAM.
 * - Peer reachability / p2pdma policy checks must be done by the caller.
 *
 * Return:
 * * 0 on success, with *@sgt set to a valid table that must be freed via
 *   amdgpu_ttm_mmio_remap_free_sgt().
 * * -ENOMEM if allocation of the sg_table fails.
 * * -EIO if dma_map_resource() fails.
 *
 */
int amdgpu_ttm_mmio_remap_alloc_sgt(struct amdgpu_device *adev,
				    struct ttm_resource *res,
				    struct device *dev,
				    enum dma_data_direction dir,
				    struct sg_table **sgt)
{
	struct amdgpu_res_cursor cur;
	dma_addr_t dma;
	resource_size_t phys;
	struct scatterlist *sg;
	int r;

	/* Walk the resource once; MMIO_REMAP is expected to be contiguous+small. */
	amdgpu_res_first(res, 0, res->size, &cur);

	/* Translate byte offset in the remap window into a host physical BAR address. */
	phys = adev->rmmio_remap.bus_addr + cur.start;

	/* Build a single-entry sg_table mapped as I/O (no struct page backing). */
	*sgt = kzalloc_obj(**sgt);
	if (!*sgt)
		return -ENOMEM;
	r = sg_alloc_table(*sgt, 1, GFP_KERNEL);
	if (r) {
		kfree(*sgt);
		return r;
	}
	sg = (*sgt)->sgl;
	sg_set_page(sg, NULL, cur.size, 0);  /* WHY: I/O space → no pages */

	dma = dma_map_resource(dev, phys, cur.size, dir, DMA_ATTR_SKIP_CPU_SYNC);
	if (dma_mapping_error(dev, dma)) {
		sg_free_table(*sgt);
		kfree(*sgt);
		return -EIO;
	}
	sg_dma_address(sg) = dma;
	sg_dma_len(sg) = cur.size;
	return 0;
}

void amdgpu_ttm_mmio_remap_free_sgt(struct device *dev,
				    enum dma_data_direction dir,
				    struct sg_table *sgt)
{
	struct scatterlist *sg = sgt->sgl;

	dma_unmap_resource(dev, sg_dma_address(sg), sg_dma_len(sg),
			   dir, DMA_ATTR_SKIP_CPU_SYNC);
	sg_free_table(sgt);
	kfree(sgt);
}

/**
 * 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
 * @page_flags: Page flags to be added to the ttm_tt object
 *
 * Called by ttm_tt_create().
 */
static struct ttm_tt *amdgpu_ttm_tt_create(struct ttm_buffer_object *bo,
					   uint32_t page_flags)
{
	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
	struct amdgpu_bo *abo = ttm_to_amdgpu_bo(bo);
	struct amdgpu_ttm_tt *gtt;
	enum ttm_caching caching;

	gtt = kzalloc_obj(struct amdgpu_ttm_tt);
	if (!gtt)
		return NULL;

	gtt->gobj = &bo->base;
	if (adev->gmc.mem_partitions && abo->xcp_id >= 0)
		gtt->pool_id = KFD_XCP_MEM_ID(adev, abo->xcp_id);
	else
		gtt->pool_id = abo->xcp_id;

	if (abo->flags & AMDGPU_GEM_CREATE_CPU_GTT_USWC)
		caching = ttm_write_combined;
	else
		caching = ttm_cached;

	/* allocate space for the uninitialized page entries */
	if (ttm_sg_tt_init(&gtt->ttm, bo, page_flags, caching)) {
		kfree(gtt);
		return NULL;
	}
	return &gtt->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_device *bdev,
				  struct ttm_tt *ttm,
				  struct ttm_operation_ctx *ctx)
{
	struct amdgpu_device *adev = amdgpu_ttm_adev(bdev);
	struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm);
	struct ttm_pool *pool;
	pgoff_t i;
	int ret;

	/* user pages are bound by amdgpu_ttm_tt_pin_userptr() */
	if (gtt->userptr) {
		ttm->sg = kzalloc_obj(struct sg_table);
		if (!ttm->sg)
			return -ENOMEM;
		return 0;
	}

	if (ttm->page_flags & TTM_TT_FLAG_EXTERNAL)
		return 0;

	if (adev->mman.ttm_pools && gtt->pool_id >= 0)
		pool = &adev->mman.ttm_pools[gtt->pool_id];
	else
		pool = &adev->mman.bdev.pool;
	ret = ttm_pool_alloc(pool, ttm, ctx);
	if (ret)
		return ret;

	for (i = 0; i < ttm->num_pages; ++i)
		ttm->pages[i]->mapping = bdev->dev_mapping;

	return 0;
}

/*
 * 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_device *bdev,
				     struct ttm_tt *ttm)
{
	struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm);
	struct amdgpu_device *adev;
	struct ttm_pool *pool;
	pgoff_t i;

	amdgpu_ttm_backend_unbind(bdev, ttm);

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

	if (ttm->page_flags & TTM_TT_FLAG_EXTERNAL)
		return;

	for (i = 0; i < ttm->num_pages; ++i)
		ttm->pages[i]->mapping = NULL;

	adev = amdgpu_ttm_adev(bdev);

	if (adev->mman.ttm_pools && gtt->pool_id >= 0)
		pool = &adev->mman.ttm_pools[gtt->pool_id];
	else
		pool = &adev->mman.bdev.pool;

	return ttm_pool_free(pool, ttm);
}

/**
 * amdgpu_ttm_tt_get_userptr - Return the userptr GTT ttm_tt for the current
 * task
 *
 * @tbo: The ttm_buffer_object that contains the userptr
 * @user_addr:  The returned value
 */
int amdgpu_ttm_tt_get_userptr(const struct ttm_buffer_object *tbo,
			      uint64_t *user_addr)
{
	struct amdgpu_ttm_tt *gtt;

	if (!tbo->ttm)
		return -EINVAL;

	gtt = (void *)tbo->ttm;
	*user_addr = gtt->userptr;
	return 0;
}

/**
 * amdgpu_ttm_tt_set_userptr - Initialize userptr GTT ttm_tt for the current
 * task
 *
 * @bo: The ttm_buffer_object to bind this userptr to
 * @addr:  The address in the current tasks VM space to use
 * @flags: Requirements of userptr object.
 *
 * Called by amdgpu_gem_userptr_ioctl() and kfd_ioctl_alloc_memory_of_gpu() to
 * bind userptr pages to current task and by kfd_ioctl_acquire_vm() to
 * initialize GPU VM for a KFD process.
 */
int amdgpu_ttm_tt_set_userptr(struct ttm_buffer_object *bo,
			      uint64_t addr, uint32_t flags)
{
	struct amdgpu_ttm_tt *gtt;

	if (!bo->ttm) {
		/* TODO: We want a separate TTM object type for userptrs */
		bo->ttm = amdgpu_ttm_tt_create(bo, 0);
		if (bo->ttm == NULL)
			return -ENOMEM;
	}

	/* Set TTM_TT_FLAG_EXTERNAL before populate but after create. */
	bo->ttm->page_flags |= TTM_TT_FLAG_EXTERNAL;

	gtt = ttm_to_amdgpu_ttm_tt(bo->ttm);
	gtt->userptr = addr;
	gtt->userflags = flags;

	if (gtt->usertask)
		put_task_struct(gtt->usertask);
	gtt->usertask = current->group_leader;
	get_task_struct(gtt->usertask);

	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 amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm);

	if (gtt == NULL)
		return NULL;

	if (gtt->usertask == NULL)
		return NULL;

	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,
				  unsigned long end, unsigned long *userptr)
{
	struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm);
	unsigned long size;

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

	/* Return false if no part of the ttm_tt object lies within
	 * the range
	 */
	size = (unsigned long)gtt->ttm.num_pages * PAGE_SIZE;
	if (gtt->userptr > end || gtt->userptr + size <= start)
		return false;

	if (userptr)
		*userptr = gtt->userptr;
	return true;
}

/*
 * amdgpu_ttm_tt_is_userptr - Have the pages backing by userptr?
 */
bool amdgpu_ttm_tt_is_userptr(struct ttm_tt *ttm)
{
	struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm);

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

	return true;
}

/*
 * amdgpu_ttm_tt_is_readonly - Is the ttm_tt object read only?
 */
bool amdgpu_ttm_tt_is_readonly(struct ttm_tt *ttm)
{
	struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm);

	if (gtt == NULL)
		return false;

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

/**
 * amdgpu_ttm_tt_pde_flags - Compute PDE flags for ttm_tt object
 *
 * @ttm: The ttm_tt object to compute the flags for
 * @mem: The memory registry backing this ttm_tt object
 *
 * Figure out the flags to use for a VM PDE (Page Directory Entry).
 */
uint64_t amdgpu_ttm_tt_pde_flags(struct ttm_tt *ttm, struct ttm_resource *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 ||
		    mem->mem_type == AMDGPU_PL_DOORBELL ||
		    mem->mem_type == AMDGPU_PL_PREEMPT ||
		    mem->mem_type == AMDGPU_PL_MMIO_REMAP)) {
		flags |= AMDGPU_PTE_SYSTEM;

		if (ttm && ttm->caching == ttm_cached)
			flags |= AMDGPU_PTE_SNOOPED;
	}

	if (mem && mem->mem_type == TTM_PL_VRAM &&
			mem->bus.caching == ttm_cached)
		flags |= AMDGPU_PTE_SNOOPED;

	return flags;
}

/**
 * amdgpu_ttm_tt_pte_flags - Compute PTE flags for ttm_tt object
 *
 * @adev: amdgpu_device pointer
 * @ttm: The ttm_tt object to compute the flags for
 * @mem: The memory registry backing this ttm_tt object
 *
 * Figure out the flags to use for a VM PTE (Page Table Entry).
 */
uint64_t amdgpu_ttm_tt_pte_flags(struct amdgpu_device *adev, struct ttm_tt *ttm,
				 struct ttm_resource *mem)
{
	uint64_t flags = amdgpu_ttm_tt_pde_flags(ttm, mem);

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

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

	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,
					    const struct ttm_place *place)
{
	struct dma_resv_iter resv_cursor;
	struct amdgpu_bo *abo;
	struct dma_fence *f;

	if (!amdgpu_bo_is_amdgpu_bo(bo))
		return ttm_bo_eviction_valuable(bo, place);

	/* Swapout? */
	if (bo->resource->mem_type == TTM_PL_SYSTEM)
		return true;

	abo = ttm_to_amdgpu_bo(bo);
	if ((abo->flags & AMDGPU_GEM_CREATE_DISCARDABLE) &&
	    bo->destroy == &svm_range_bo_destroy) {
		/*
		 * SVM BOs are migrated to system memory synchronously in this
		 * TTM eviction context. The migration needs the owning
		 * process's mmap lock, but the normal lock order is
		 * mmap_lock -> BO reservation and the BO is already reserved
		 * here. svm_range_evict_svm_bo() only trylocks the mmap lock;
		 * if the eviction fails for any reason, we return false so TTM
		 * skips this BO instead of risking a deadlock.
		 */
		if (svm_range_evict_svm_bo(abo) < 0)
			return false;
	}

	if (bo->type == ttm_bo_type_kernel &&
	    !amdgpu_vm_evictable(ttm_to_amdgpu_bo(bo)))
		return false;

	/* If bo is a KFD BO, check if the bo belongs to the current process.
	 * If true, then return false as any KFD process needs all its BOs to
	 * be resident to run successfully
	 */
	dma_resv_for_each_fence(&resv_cursor, bo->base.resv,
				DMA_RESV_USAGE_BOOKKEEP, f) {
		if (amdkfd_fence_check_mm(f, current->mm) &&
		    !(place && (place->flags & TTM_PL_FLAG_CONTIGUOUS)))
			return false;
	}

	/* Preemptible BOs don't own system resources managed by the
	 * driver (pages, VRAM, GART space). They point to resources
	 * owned by someone else (e.g. pageable memory in user mode
	 * or a DMABuf). They are used in a preemptible context so we
	 * can guarantee no deadlocks and good QoS in case of MMU
	 * notifiers or DMABuf move notifiers from the resource owner.
	 */
	if (bo->resource->mem_type == AMDGPU_PL_PREEMPT)
		return false;

	if (bo->resource->mem_type == TTM_PL_TT &&
	    amdgpu_bo_encrypted(ttm_to_amdgpu_bo(bo)))
		return false;

	return ttm_bo_eviction_valuable(bo, place);
}

static void amdgpu_ttm_vram_mm_access(struct amdgpu_device *adev, loff_t pos,
				      void *buf, size_t size, bool write)
{
	while (size) {
		uint64_t aligned_pos = ALIGN_DOWN(pos, 4);
		uint64_t bytes = 4 - (pos & 0x3);
		uint32_t shift = (pos & 0x3) * 8;
		uint32_t mask = 0xffffffff << shift;
		uint32_t value = 0;

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

		if (mask != 0xffffffff) {
			amdgpu_device_mm_access(adev, aligned_pos, &value, 4, false);
			if (write) {
				value &= ~mask;
				value |= (*(uint32_t *)buf << shift) & mask;
				amdgpu_device_mm_access(adev, aligned_pos, &value, 4, true);
			} else {
				value = (value & mask) >> shift;
				memcpy(buf, &value, bytes);
			}
		} else {
			amdgpu_device_mm_access(adev, aligned_pos, buf, 4, write);
		}

		pos += bytes;
		buf += bytes;
		size -= bytes;
	}
}

static int amdgpu_ttm_access_memory_sdma(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 amdgpu_res_cursor src_mm;
	struct amdgpu_job *job;
	struct dma_fence *fence;
	uint64_t src_addr, dst_addr;
	unsigned int num_dw;
	int r, idx;

	if (len != PAGE_SIZE)
		return -EINVAL;

	if (!adev->mman.sdma_access_ptr)
		return -EACCES;

	if (!adev->mman.buffer_funcs_enabled || !drm_dev_enter(adev_to_drm(adev), &idx))
		return -ENODEV;

	if (write)
		memcpy(adev->mman.sdma_access_ptr, buf, len);

	num_dw = ALIGN(adev->mman.buffer_funcs->copy_num_dw, 8);
	r = amdgpu_job_alloc_with_ib(adev, &adev->mman.default_entity.base,
				     AMDGPU_FENCE_OWNER_UNDEFINED,
				     num_dw * 4, AMDGPU_IB_POOL_DELAYED,
				     AMDGPU_KERNEL_JOB_ID_TTM_ACCESS_MEMORY_SDMA,
				     &job);
	if (r)
		goto out;

	mutex_lock(&adev->mman.default_entity.lock);
	amdgpu_res_first(abo->tbo.resource, offset, len, &src_mm);
	src_addr = amdgpu_ttm_domain_start(adev, bo->resource->mem_type) +
		src_mm.start;
	dst_addr = amdgpu_bo_gpu_offset(adev->mman.sdma_access_bo);
	if (write)
		swap(src_addr, dst_addr);

	amdgpu_emit_copy_buffer(adev, &job->ibs[0], src_addr, dst_addr,
				PAGE_SIZE, 0);

	fence = amdgpu_ttm_job_submit(adev, &adev->mman.default_entity, job, num_dw);
	mutex_unlock(&adev->mman.default_entity.lock);

	if (!dma_fence_wait_timeout(fence, false, adev->sdma_timeout))
		r = -ETIMEDOUT;
	dma_fence_put(fence);

	if (!(r || write))
		memcpy(buf, adev->mman.sdma_access_ptr, len);
out:
	drm_dev_exit(idx);
	return r;
}

/**
 * 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,
				    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 amdgpu_res_cursor cursor;
	int ret = 0;

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

	if (amdgpu_device_has_timeouts_enabled(adev) &&
			!amdgpu_ttm_access_memory_sdma(bo, offset, buf, len, write))
		return len;

	amdgpu_res_first(bo->resource, offset, len, &cursor);
	while (cursor.remaining) {
		size_t count, size = cursor.size;
		loff_t pos = cursor.start;

		count = amdgpu_device_aper_access(adev, pos, buf, size, write);
		size -= count;
		if (size) {
			/* using MM to access rest vram and handle un-aligned address */
			pos += count;
			buf += count;
			amdgpu_ttm_vram_mm_access(adev, pos, buf, size, write);
		}

		ret += cursor.size;
		buf += cursor.size;
		amdgpu_res_next(&cursor, cursor.size);
	}

	return ret;
}

static void
amdgpu_bo_delete_mem_notify(struct ttm_buffer_object *bo)
{
	if (bo->resource && bo->resource->mem_type == TTM_PL_TT)
		amdgpu_gtt_mgr_mark_bo_teardown(bo);

	amdgpu_bo_move_notify(bo, false, NULL);
}

static struct ttm_device_funcs amdgpu_bo_driver = {
	.ttm_tt_create = &amdgpu_ttm_tt_create,
	.ttm_tt_populate = &amdgpu_ttm_tt_populate,
	.ttm_tt_unpopulate = &amdgpu_ttm_tt_unpopulate,
	.ttm_tt_destroy = &amdgpu_ttm_backend_destroy,
	.eviction_valuable = amdgpu_ttm_bo_eviction_valuable,
	.evict_flags = &amdgpu_evict_flags,
	.move = &amdgpu_bo_move,
	.delete_mem_notify = &amdgpu_bo_delete_mem_notify,
	.release_notify = &amdgpu_bo_release_notify,
	.io_mem_reserve = &amdgpu_ttm_io_mem_reserve,
	.io_mem_pfn = amdgpu_ttm_io_mem_pfn,
	.access_memory = &amdgpu_ttm_access_memory,
};

void amdgpu_ttm_init_vram_resv(struct amdgpu_device *adev,
				enum amdgpu_resv_region_id id,
				uint64_t offset, uint64_t size,
				bool needs_cpu_map)
{
	struct amdgpu_vram_resv *resv;

	if (id >= AMDGPU_RESV_MAX)
		return;

	resv = &adev->mman.resv_region[id];
	resv->offset = offset;
	resv->size = size;
	resv->needs_cpu_map = needs_cpu_map;
}

static void amdgpu_ttm_init_fw_resv_region(struct amdgpu_device *adev)
{
	uint32_t reserve_size = 0;

	if (!adev->discovery.reserve_tmr)
		return;

	/*
	 * Query reserved tmr size through atom firmwareinfo for Sienna_Cichlid and onwards for all
	 * the use cases (IP discovery/G6 memory training/profiling/diagnostic data.etc)
	 *
	 * Otherwise, fallback to legacy approach to check and reserve tmr block for ip
	 * discovery data and G6 memory training data respectively
	 */
	if (adev->bios)
		reserve_size =
			amdgpu_atomfirmware_get_fw_reserved_fb_size(adev);

	if (!adev->bios &&
	    (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 3) ||
	     amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 4) ||
	     amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 5, 0)))
		reserve_size = max(reserve_size, (uint32_t)280 << 20);
	else if (!adev->bios &&
		 amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(12, 1, 0)) {
		reserve_size = max(reserve_size, (uint32_t)150 << 20);
	} else if (!reserve_size)
		reserve_size = DISCOVERY_TMR_OFFSET;

	amdgpu_ttm_init_vram_resv(adev, AMDGPU_RESV_FW,
				  adev->gmc.real_vram_size - reserve_size,
				  reserve_size, false);
}

static void amdgpu_ttm_init_mem_train_resv_region(struct amdgpu_device *adev)
{
	uint64_t reserve_size;
	uint64_t offset;

	if (!adev->discovery.reserve_tmr)
		return;

	if (!adev->bios || amdgpu_sriov_vf(adev))
		return;

	if (!amdgpu_atomfirmware_mem_training_supported(adev))
		return;

	reserve_size = adev->mman.resv_region[AMDGPU_RESV_FW].size;
	offset = ALIGN((adev->gmc.mc_vram_size - reserve_size - SZ_1M), SZ_1M);
	amdgpu_ttm_init_vram_resv(adev, AMDGPU_RESV_MEM_TRAIN,
				  offset,
				  GDDR6_MEM_TRAINING_DATA_SIZE_IN_BYTES,
				  false);
}

static void amdgpu_ttm_init_vram_resv_regions(struct amdgpu_device *adev)
{
	uint64_t vram_size = adev->gmc.visible_vram_size;

	/* Initialize memory reservations 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.
	 */
	amdgpu_gmc_init_vga_resv_regions(adev);
	amdgpu_ttm_init_fw_resv_region(adev);
	amdgpu_ttm_init_mem_train_resv_region(adev);

	if (adev->mman.resv_region[AMDGPU_RESV_FW_VRAM_USAGE].size > vram_size)
		adev->mman.resv_region[AMDGPU_RESV_FW_VRAM_USAGE].size = 0;

	if (adev->mman.resv_region[AMDGPU_RESV_DRV_VRAM_USAGE].size > vram_size)
		adev->mman.resv_region[AMDGPU_RESV_DRV_VRAM_USAGE].size = 0;
}

int amdgpu_ttm_mark_vram_reserved(struct amdgpu_device *adev,
				  enum amdgpu_resv_region_id id)
{
	struct amdgpu_vram_resv *resv;
	int ret;

	if (id >= AMDGPU_RESV_MAX)
		return -EINVAL;

	resv = &adev->mman.resv_region[id];
	if (!resv->size)
		return 0;

	ret = amdgpu_bo_create_kernel_at(adev, resv->offset, resv->size,
					 &resv->bo,
					 resv->needs_cpu_map ? &resv->cpu_ptr : NULL);
	if (ret) {
		dev_err(adev->dev,
			"reserve vram failed: id=%d offset=0x%llx size=0x%llx ret=%d\n",
			id, resv->offset, resv->size, ret);
		memset(resv, 0, sizeof(*resv));
	}

	return ret;
}

void amdgpu_ttm_unmark_vram_reserved(struct amdgpu_device *adev,
				     enum amdgpu_resv_region_id id)
{
	struct amdgpu_vram_resv *resv;

	if (id >= AMDGPU_RESV_MAX)
		return;

	resv = &adev->mman.resv_region[id];
	if (!resv->bo)
		return;

	amdgpu_bo_free_kernel(&resv->bo, NULL,
			      resv->needs_cpu_map ? &resv->cpu_ptr : NULL);
	memset(resv, 0, sizeof(*resv));
}

/*
 * Reserve all regions with non-zero size. Regions whose info is not
 * yet available (e.g., fw extended region) may still be reserved
 * during runtime.
 */
static int amdgpu_ttm_alloc_vram_resv_regions(struct amdgpu_device *adev)
{
	int i, r;

	for (i = 0; i < AMDGPU_RESV_MAX; i++) {
		r = amdgpu_ttm_mark_vram_reserved(adev, i);
		if (r)
			return r;
	}

	return 0;
}

/*
 * Memoy training reservation functions
 */

/**
 * amdgpu_ttm_training_reserve_vram_fini - free memory training reserved vram
 *
 * @adev: amdgpu_device pointer
 *
 * free memory training reserved vram if it has been reserved.
 */
static int amdgpu_ttm_training_reserve_vram_fini(struct amdgpu_device *adev)
{
	struct psp_memory_training_context *ctx = &adev->psp.mem_train_ctx;

	ctx->init = PSP_MEM_TRAIN_NOT_SUPPORT;
	amdgpu_ttm_unmark_vram_reserved(adev, AMDGPU_RESV_MEM_TRAIN);

	return 0;
}

static void amdgpu_ttm_training_data_block_init(struct amdgpu_device *adev)
{
	struct psp_memory_training_context *ctx = &adev->psp.mem_train_ctx;
	struct amdgpu_vram_resv *resv =
			&adev->mman.resv_region[AMDGPU_RESV_MEM_TRAIN];

	memset(ctx, 0, sizeof(*ctx));

	ctx->c2p_train_data_offset = resv->offset;
	ctx->p2c_train_data_offset =
		(adev->gmc.mc_vram_size - GDDR6_MEM_TRAINING_OFFSET);
	ctx->train_data_size = resv->size;

	DRM_DEBUG("train_data_size:%llx,p2c_train_data_offset:%llx,c2p_train_data_offset:%llx.\n",
			ctx->train_data_size,
			ctx->p2c_train_data_offset,
			ctx->c2p_train_data_offset);
}

static int amdgpu_ttm_pools_init(struct amdgpu_device *adev)
{
	int i;

	if (!adev->gmc.is_app_apu || !adev->gmc.num_mem_partitions)
		return 0;

	adev->mman.ttm_pools = kzalloc_objs(*adev->mman.ttm_pools,
					    adev->gmc.num_mem_partitions);
	if (!adev->mman.ttm_pools)
		return -ENOMEM;

	for (i = 0; i < adev->gmc.num_mem_partitions; i++) {
		ttm_pool_init(&adev->mman.ttm_pools[i], adev->dev,
			      adev->gmc.mem_partitions[i].numa.node,
			      TTM_ALLOCATION_POOL_BENEFICIAL_ORDER(get_order(SZ_2M)));
	}
	return 0;
}

static void amdgpu_ttm_pools_fini(struct amdgpu_device *adev)
{
	int i;

	if (!adev->gmc.is_app_apu || !adev->mman.ttm_pools)
		return;

	for (i = 0; i < adev->gmc.num_mem_partitions; i++)
		ttm_pool_fini(&adev->mman.ttm_pools[i]);

	kfree(adev->mman.ttm_pools);
	adev->mman.ttm_pools = NULL;
}

/**
 * amdgpu_ttm_alloc_mmio_remap_bo - Allocate the singleton MMIO_REMAP BO
 * @adev: amdgpu device
 *
 * Allocates a global BO with backing AMDGPU_PL_MMIO_REMAP when the
 * hardware exposes a remap base (adev->rmmio_remap.bus_addr) and the host
 * PAGE_SIZE is <= AMDGPU_GPU_PAGE_SIZE (4K). The BO is created as a regular
 * GEM object (amdgpu_bo_create).
 *
 * Return:
 *  * 0 on success or intentional skip (feature not present/unsupported)
 *  * negative errno on allocation failure
 */
static int amdgpu_ttm_alloc_mmio_remap_bo(struct amdgpu_device *adev)
{
	struct ttm_operation_ctx ctx = { false, false };
	struct ttm_placement placement;
	struct ttm_buffer_object *tbo;
	struct ttm_place placements;
	struct amdgpu_bo_param bp;
	struct ttm_resource *tmp;
	int r;

	/* Skip if HW doesn't expose remap, or if PAGE_SIZE > AMDGPU_GPU_PAGE_SIZE (4K). */
	if (!adev->rmmio_remap.bus_addr || PAGE_SIZE > AMDGPU_GPU_PAGE_SIZE)
		return 0;

	/*
	 * Allocate a BO first and then move it to AMDGPU_PL_MMIO_REMAP.
	 * The initial TTM resource assigned by amdgpu_bo_create() is
	 * replaced below with a fixed MMIO_REMAP placement.
	 */
	memset(&bp, 0, sizeof(bp));
	bp.type        = ttm_bo_type_device;
	bp.size        = AMDGPU_GPU_PAGE_SIZE;
	bp.byte_align  = AMDGPU_GPU_PAGE_SIZE;
	bp.domain      = 0;
	bp.flags       = 0;
	bp.resv        = NULL;
	bp.bo_ptr_size = sizeof(struct amdgpu_bo);
	r = amdgpu_bo_create(adev, &bp, &adev->rmmio_remap.bo);
	if (r)
		return r;

	r = amdgpu_bo_reserve(adev->rmmio_remap.bo, true);
	if (r)
		goto err_unref;

	tbo = &adev->rmmio_remap.bo->tbo;

	/*
	 * MMIO_REMAP is a fixed I/O placement (AMDGPU_PL_MMIO_REMAP).
	 */
	placement.num_placement = 1;
	placement.placement = &placements;
	placements.fpfn = 0;
	placements.lpfn = 0;
	placements.mem_type = AMDGPU_PL_MMIO_REMAP;
	placements.flags = 0;
	/* Force the BO into the fixed MMIO_REMAP placement */
	r = ttm_bo_mem_space(tbo, &placement, &tmp, &ctx);
	if (unlikely(r))
		goto err_unlock;

	ttm_resource_free(tbo, &tbo->resource);
	ttm_bo_assign_mem(tbo, tmp);
	ttm_bo_pin(tbo);

	amdgpu_bo_unreserve(adev->rmmio_remap.bo);
	return 0;

err_unlock:
	amdgpu_bo_unreserve(adev->rmmio_remap.bo);

err_unref:
	amdgpu_bo_unref(&adev->rmmio_remap.bo);
	adev->rmmio_remap.bo = NULL;
	return r;
}

/**
 * amdgpu_ttm_free_mmio_remap_bo - Free the singleton MMIO_REMAP BO
 * @adev: amdgpu device
 *
 * Frees the kernel-owned MMIO_REMAP BO if it was allocated by
 * amdgpu_ttm_mmio_remap_bo_init().
 */
static void amdgpu_ttm_free_mmio_remap_bo(struct amdgpu_device *adev)
{
	if (!adev->rmmio_remap.bo)
		return;

	if (!amdgpu_bo_reserve(adev->rmmio_remap.bo, true)) {
		ttm_bo_unpin(&adev->rmmio_remap.bo->tbo);
		amdgpu_bo_unreserve(adev->rmmio_remap.bo);
	}

    /*
     * At this point we rely on normal DRM teardown ordering:
     * no new user ioctls can access the global MMIO_REMAP BO
     * once TTM teardown begins.
     */
	amdgpu_bo_unref(&adev->rmmio_remap.bo);
	adev->rmmio_remap.bo = NULL;
}

static int amdgpu_ttm_buffer_entity_init(struct amdgpu_gtt_mgr *mgr,
					 struct amdgpu_ttm_buffer_entity *entity,
					 enum drm_sched_priority prio,
					 struct drm_gpu_scheduler **scheds,
					 int num_schedulers,
					 u32 num_gart_windows)
{
	int i, r, num_pages;
	const u64 GTT_MAX_PAGES = (AMDGPU_GTT_MAX_TRANSFER_SIZE >> PAGE_SHIFT);

	r = drm_sched_entity_init(&entity->base, prio, scheds, num_schedulers, NULL);
	if (r)
		return r;

	mutex_init(&entity->lock);

	if (ARRAY_SIZE(entity->gart_window_offs) < num_gart_windows)
		return -EINVAL;
	if (num_gart_windows == 0)
		return 0;

	num_pages = num_gart_windows * GTT_MAX_PAGES;
	r = amdgpu_gtt_mgr_alloc_entries(mgr, &entity->gart_node, num_pages,
					 DRM_MM_INSERT_BEST);
	if (r) {
		drm_sched_entity_destroy(&entity->base);
		return r;
	}

	for (i = 0; i < num_gart_windows; i++) {
		entity->gart_window_offs[i] =
			amdgpu_gtt_node_to_byte_offset(&entity->gart_node) +
				i * GTT_MAX_PAGES * PAGE_SIZE;
	}

	return 0;
}

static void amdgpu_ttm_buffer_entity_fini(struct amdgpu_gtt_mgr *mgr,
					  struct amdgpu_ttm_buffer_entity *entity)
{
	amdgpu_gtt_mgr_free_entries(mgr, &entity->gart_node);
	drm_sched_entity_destroy(&entity->base);
}

/*
 * 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)
{
	uint64_t gtt_size;
	int r;

	dma_set_max_seg_size(adev->dev, UINT_MAX);
	/* No others user of address space so set it to 0 */
	r = ttm_device_init(&adev->mman.bdev, &amdgpu_bo_driver, adev->dev,
			       adev_to_drm(adev)->anon_inode->i_mapping,
			       adev_to_drm(adev)->vma_offset_manager,
			       (adev->need_swiotlb ?
				TTM_ALLOCATION_POOL_USE_DMA_ALLOC : 0) |
			       (dma_addressing_limited(adev->dev) ?
				TTM_ALLOCATION_POOL_USE_DMA32 : 0) |
			       TTM_ALLOCATION_POOL_BENEFICIAL_ORDER(get_order(SZ_2M)));
	if (r) {
		dev_err(adev->dev,
			"failed initializing buffer object driver(%d).\n", r);
		return r;
	}

	r = amdgpu_ttm_pools_init(adev);
	if (r) {
		dev_err(adev->dev, "failed to init ttm pools(%d).\n", r);
		return r;
	}
	adev->mman.initialized = true;

	if (!adev->gmc.is_app_apu) {
		/* Initialize VRAM pool with all of VRAM divided into pages */
		r = amdgpu_vram_mgr_init(adev);
		if (r) {
			dev_err(adev->dev, "Failed initializing VRAM heap.\n");
			return r;
		}
	}

	/* Change the size here instead of the init above so only lpfn is affected */
	amdgpu_ttm_disable_buffer_funcs(adev);
#ifdef CONFIG_64BIT
#ifdef CONFIG_X86
	if (adev->gmc.xgmi.connected_to_cpu)
		adev->mman.aper_base_kaddr = ioremap_cache(adev->gmc.aper_base,
				adev->gmc.visible_vram_size);
	else if (adev->gmc.is_app_apu)
		DRM_DEBUG_DRIVER(
			"No need to ioremap when real vram size is 0\n");
	else
#endif
		adev->mman.aper_base_kaddr = ioremap_wc(adev->gmc.aper_base,
				adev->gmc.visible_vram_size);
#endif

	amdgpu_ttm_init_vram_resv_regions(adev);

	r = amdgpu_ttm_alloc_vram_resv_regions(adev);
	if (r)
		return r;

	if (adev->mman.resv_region[AMDGPU_RESV_MEM_TRAIN].size) {
		struct psp_memory_training_context *ctx =
					&adev->psp.mem_train_ctx;

		amdgpu_ttm_training_data_block_init(adev);
		ctx->init = PSP_MEM_TRAIN_RESERVE_SUCCESS;
	}

	dev_info(adev->dev, " %uM of VRAM memory ready\n",
		 (unsigned int)(adev->gmc.real_vram_size / (1024 * 1024)));

	/* Compute GTT size, either based on TTM limit
	 * or whatever the user passed on module init.
	 */
	gtt_size = ttm_tt_pages_limit() << PAGE_SHIFT;
	if (amdgpu_gtt_size != -1) {
		uint64_t configured_size = (uint64_t)amdgpu_gtt_size << 20;

		drm_warn(&adev->ddev,
			"Configuring gttsize via module parameter is deprecated, please use ttm.pages_limit\n");
		if (gtt_size != configured_size)
			drm_warn(&adev->ddev,
				"GTT size has been set as %llu but TTM size has been set as %llu, this is unusual\n",
				configured_size, gtt_size);

		gtt_size = configured_size;
	}

	/* Cap GTT so that it does not exceed total physical RAM. */
	if (adev->flags & AMD_IS_APU) {
		u64 phys_ram = (u64)totalram_pages() << PAGE_SHIFT;

		if (gtt_size > phys_ram) {
			gtt_size = phys_ram;
			dev_info(adev->dev,
				 "Capping GTT to %uM to not exceed available system memory\n",
				 (unsigned int)(gtt_size / (1024 * 1024)));
		}
	}

	/* Initialize GTT memory pool */
	r = amdgpu_gtt_mgr_init(adev, gtt_size);
	if (r) {
		dev_err(adev->dev, "Failed initializing GTT heap.\n");
		return r;
	}
	dev_info(adev->dev, " %uM of GTT memory ready.\n",
		 (unsigned int)(gtt_size / (1024 * 1024)));

	if (adev->flags & AMD_IS_APU) {
		if (adev->gmc.real_vram_size < gtt_size)
			adev->apu_prefer_gtt = true;
	}

	/* Initialize doorbell pool on PCI BAR */
	r = amdgpu_ttm_init_on_chip(adev, AMDGPU_PL_DOORBELL, adev->doorbell.size / PAGE_SIZE);
	if (r) {
		dev_err(adev->dev, "Failed initializing doorbell heap.\n");
		return r;
	}

	/* Create a doorbell page for kernel usages */
	r = amdgpu_doorbell_create_kernel_doorbells(adev);
	if (r) {
		dev_err(adev->dev, "Failed to initialize kernel doorbells.\n");
		return r;
	}

	/* Initialize MMIO-remap pool (single page 4K) */
	r = amdgpu_ttm_init_on_chip(adev, AMDGPU_PL_MMIO_REMAP, 1);
	if (r) {
		dev_err(adev->dev, "Failed initializing MMIO-remap heap.\n");
		return r;
	}

	/* Allocate the singleton MMIO_REMAP BO if supported */
	r = amdgpu_ttm_alloc_mmio_remap_bo(adev);
	if (r)
		return r;

	/* Initialize preemptible memory pool */
	r = amdgpu_preempt_mgr_init(adev);
	if (r) {
		dev_err(adev->dev, "Failed initializing PREEMPT heap.\n");
		return r;
	}

	/* Initialize various on-chip memory pools */
	r = amdgpu_ttm_init_on_chip(adev, AMDGPU_PL_GDS, adev->gds.gds_size);
	if (r) {
		dev_err(adev->dev, "Failed initializing GDS heap.\n");
		return r;
	}

	r = amdgpu_ttm_init_on_chip(adev, AMDGPU_PL_GWS, adev->gds.gws_size);
	if (r) {
		dev_err(adev->dev, "Failed initializing gws heap.\n");
		return r;
	}

	r = amdgpu_ttm_init_on_chip(adev, AMDGPU_PL_OA, adev->gds.oa_size);
	if (r) {
		dev_err(adev->dev, "Failed initializing oa heap.\n");
		return r;
	}
	if (amdgpu_bo_create_kernel(adev, PAGE_SIZE, PAGE_SIZE,
				AMDGPU_GEM_DOMAIN_GTT,
				&adev->mman.sdma_access_bo, NULL,
				&adev->mman.sdma_access_ptr))
		drm_warn(adev_to_drm(adev),
				"Debug VRAM access will use slowpath MM access\n");

	return 0;
}

/*
 * amdgpu_ttm_fini - De-initialize the TTM memory pools
 */
void amdgpu_ttm_fini(struct amdgpu_device *adev)
{
	if (!adev->mman.initialized)
		return;

	amdgpu_ttm_pools_fini(adev);

	amdgpu_ttm_training_reserve_vram_fini(adev);
	/* return the stolen vga memory back to VRAM */
	if (!adev->gmc.is_app_apu) {
		amdgpu_ttm_unmark_vram_reserved(adev, AMDGPU_RESV_STOLEN_VGA);
		amdgpu_ttm_unmark_vram_reserved(adev, AMDGPU_RESV_STOLEN_EXTENDED);
		/* return the FW reserved memory back to VRAM */
		amdgpu_ttm_unmark_vram_reserved(adev, AMDGPU_RESV_FW);
		amdgpu_ttm_unmark_vram_reserved(adev, AMDGPU_RESV_FW_EXTEND);
		amdgpu_ttm_unmark_vram_reserved(adev, AMDGPU_RESV_STOLEN_RESERVED);
	}
	amdgpu_bo_free_kernel(&adev->mman.sdma_access_bo, NULL,
					&adev->mman.sdma_access_ptr);

	amdgpu_ttm_free_mmio_remap_bo(adev);
	amdgpu_ttm_unmark_vram_reserved(adev, AMDGPU_RESV_FW_VRAM_USAGE);
	amdgpu_ttm_unmark_vram_reserved(adev, AMDGPU_RESV_DRV_VRAM_USAGE);

	if (adev->mman.aper_base_kaddr) {
		iounmap(adev->mman.aper_base_kaddr);
		adev->mman.aper_base_kaddr = NULL;
	}

	if (!adev->gmc.is_app_apu)
		amdgpu_vram_mgr_fini(adev);
	amdgpu_gtt_mgr_fini(adev);
	amdgpu_preempt_mgr_fini(adev);
	amdgpu_doorbell_fini(adev);

	ttm_range_man_fini(&adev->mman.bdev, AMDGPU_PL_GDS);
	ttm_range_man_fini(&adev->mman.bdev, AMDGPU_PL_GWS);
	ttm_range_man_fini(&adev->mman.bdev, AMDGPU_PL_OA);
	ttm_range_man_fini(&adev->mman.bdev, AMDGPU_PL_DOORBELL);
	ttm_range_man_fini(&adev->mman.bdev, AMDGPU_PL_MMIO_REMAP);
	ttm_device_fini(&adev->mman.bdev);
	adev->mman.initialized = false;
	dev_info(adev->dev, " ttm finalized\n");
}

/**
 * amdgpu_ttm_enable_buffer_funcs - enable use of buffer functions
 *
 * @adev: amdgpu_device pointer
 *
 * Enable use of buffer functions during suspend/resume. This should
 * only be called at bootup or when userspace isn't running.
 */
void amdgpu_ttm_enable_buffer_funcs(struct amdgpu_device *adev)
{
	struct ttm_resource_manager *man = ttm_manager_type(&adev->mman.bdev, TTM_PL_VRAM);
	u32 num_clear_entities, num_move_entities;
	int r, i, j;

	if (!adev->mman.initialized || amdgpu_in_reset(adev) ||
	    adev->mman.buffer_funcs_enabled || adev->gmc.is_app_apu)
		return;

	if (!adev->mman.num_buffer_funcs_scheds) {
		dev_warn(adev->dev, "Not enabling DMA transfers for in kernel use");
		return;
	}

	/* default_entity doesn't need multiple schedulers so pass only 1. */
	r = amdgpu_ttm_buffer_entity_init(&adev->mman.gtt_mgr,
						&adev->mman.default_entity,
						DRM_SCHED_PRIORITY_KERNEL,
						adev->mman.buffer_funcs_scheds, 1, 0);
	if (r < 0) {
		dev_err(adev->dev,
			"Failed setting up TTM entity (%d)\n", r);
		return;
	}

	num_clear_entities = MIN(adev->mman.num_buffer_funcs_scheds, TTM_NUM_MOVE_FENCES);
	num_move_entities = MIN(adev->mman.num_buffer_funcs_scheds, TTM_NUM_MOVE_FENCES);

	adev->mman.clear_entities = kzalloc_objs(struct amdgpu_ttm_buffer_entity,
						 num_clear_entities);
	atomic_set(&adev->mman.next_clear_entity, 0);
	if (!adev->mman.clear_entities)
		goto error_free_default_entity;

	adev->mman.num_clear_entities = num_clear_entities;

	for (i = 0; i < num_clear_entities; i++) {
		r = amdgpu_ttm_buffer_entity_init(
			&adev->mman.gtt_mgr,
			&adev->mman.clear_entities[i],
			DRM_SCHED_PRIORITY_KERNEL,
			adev->mman.buffer_funcs_scheds,
			adev->mman.num_buffer_funcs_scheds, 1);

		if (r < 0) {
			for (j = 0; j < i; j++)
				amdgpu_ttm_buffer_entity_fini(
					&adev->mman.gtt_mgr, &adev->mman.clear_entities[j]);
			adev->mman.num_clear_entities = 0;
			kfree(adev->mman.clear_entities);
			goto error_free_default_entity;
		}
	}

	adev->mman.num_move_entities = num_move_entities;
	atomic_set(&adev->mman.next_move_entity, 0);
	for (i = 0; i < num_move_entities; i++) {
		r = amdgpu_ttm_buffer_entity_init(
			&adev->mman.gtt_mgr,
			&adev->mman.move_entities[i],
			DRM_SCHED_PRIORITY_KERNEL,
			adev->mman.buffer_funcs_scheds,
			adev->mman.num_buffer_funcs_scheds, 2);

		if (r < 0) {
			for (j = 0; j < i; j++)
				amdgpu_ttm_buffer_entity_fini(
					&adev->mman.gtt_mgr,
					&adev->mman.move_entities[j]);
			adev->mman.num_move_entities = 0;
			goto error_free_clear_entities;
		}
	}

	/* this just adjusts TTM size idea, which sets lpfn to the correct value */
	man->size = adev->gmc.real_vram_size;
	adev->mman.buffer_funcs_enabled = true;

	return;

error_free_clear_entities:
	for (i = 0; i < adev->mman.num_clear_entities; i++)
		amdgpu_ttm_buffer_entity_fini(&adev->mman.gtt_mgr,
					      &adev->mman.clear_entities[i]);
	kfree(adev->mman.clear_entities);
	adev->mman.clear_entities = NULL;
	adev->mman.num_clear_entities = 0;
error_free_default_entity:
	amdgpu_ttm_buffer_entity_fini(&adev->mman.gtt_mgr,
				      &adev->mman.default_entity);
}

/**
 * amdgpu_ttm_disable_buffer_funcs - disable use of buffer functions
 *
 * @adev: amdgpu_device pointer
 */
void amdgpu_ttm_disable_buffer_funcs(struct amdgpu_device *adev)
{
	struct ttm_resource_manager *man =
		ttm_manager_type(&adev->mman.bdev, TTM_PL_VRAM);
	int i;

	if (!adev->mman.buffer_funcs_enabled || amdgpu_in_reset(adev))
		return;

	amdgpu_ttm_buffer_entity_fini(&adev->mman.gtt_mgr,
				      &adev->mman.default_entity);
	for (i = 0; i < adev->mman.num_move_entities; i++)
		amdgpu_ttm_buffer_entity_fini(&adev->mman.gtt_mgr,
					      &adev->mman.move_entities[i]);
	for (i = 0; i < adev->mman.num_clear_entities; i++)
		amdgpu_ttm_buffer_entity_fini(&adev->mman.gtt_mgr,
					      &adev->mman.clear_entities[i]);
	/* Drop all the old fences since re-creating the scheduler entities
	 * will allocate new contexts.
	 */
	ttm_resource_manager_cleanup(man);

	kfree(adev->mman.clear_entities);
	adev->mman.clear_entities = NULL;
	adev->mman.num_clear_entities = 0;
	adev->mman.num_move_entities = 0;

	man->size = adev->gmc.visible_vram_size;
	adev->mman.buffer_funcs_enabled = false;
}

static int amdgpu_ttm_prepare_job(struct amdgpu_device *adev,
				  struct amdgpu_ttm_buffer_entity *entity,
				  unsigned int num_dw,
				  struct dma_resv *resv,
				  bool vm_needs_flush,
				  struct amdgpu_job **job,
				  u64 k_job_id)
{
	enum amdgpu_ib_pool_type pool = AMDGPU_IB_POOL_DELAYED;
	int r;
	r = amdgpu_job_alloc_with_ib(adev, &entity->base,
				     AMDGPU_FENCE_OWNER_UNDEFINED,
				     num_dw * 4, pool, k_job_id, job);
	if (r)
		return r;

	if (vm_needs_flush) {
		(*job)->vm_pd_addr = amdgpu_gmc_pd_addr(adev->gmc.pdb0_bo ?
							adev->gmc.pdb0_bo :
							adev->gart.bo);
		(*job)->vm_needs_flush = true;
	}
	if (!resv)
		return 0;

	return drm_sched_job_add_resv_dependencies(&(*job)->base, resv,
						   DMA_RESV_USAGE_BOOKKEEP);
}

static int amdgpu_calc_bytes_per_packet(u32 max_bytes_per_packet,
					u32 byte_count)
{
	/* Byte count is dword-aligned and fits a single packet */
	if (!(byte_count & 0x3) && byte_count <= max_bytes_per_packet)
		return max_bytes_per_packet;

	/*
	 * Align down maximum byte count to 256 bytes so that
	 * the copy optimally uses all memory channels and
	 * also to ensure that SDMA can use its dword mode, which
	 * is faster.
	 *
	 * This assumes that the starting addresses of BOs are always
	 * dword aligned, which should be the case for every copy
	 * operation in the kernel, because the kernel always copies
	 * pages.
	 */
	return ALIGN_DOWN(max_bytes_per_packet, SZ_256);
}

int amdgpu_copy_buffer(struct amdgpu_device *adev,
		       struct amdgpu_ttm_buffer_entity *entity,
		       uint64_t src_offset,
		       uint64_t dst_offset, uint32_t byte_count,
		       struct dma_resv *resv,
		       struct dma_fence **fence,
		       bool vm_needs_flush, uint32_t copy_flags)
{
	unsigned int num_loops, num_dw;
	struct amdgpu_ring *ring;
	struct amdgpu_job *job;
	uint32_t max_bytes;
	unsigned int i;
	int r;

	ring = to_amdgpu_ring(adev->mman.buffer_funcs_scheds[0]);

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

	max_bytes = amdgpu_calc_bytes_per_packet(adev->mman.buffer_funcs->copy_max_bytes,
						 byte_count);
	num_loops = DIV_ROUND_UP(byte_count, max_bytes);
	num_dw = ALIGN(num_loops * adev->mman.buffer_funcs->copy_num_dw, 8);
	r = amdgpu_ttm_prepare_job(adev, entity, num_dw,
				   resv, vm_needs_flush, &job,
				   AMDGPU_KERNEL_JOB_ID_TTM_COPY_BUFFER);
	if (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, copy_flags);
		src_offset += cur_size_in_bytes;
		dst_offset += cur_size_in_bytes;
		byte_count -= cur_size_in_bytes;
	}

	*fence = amdgpu_ttm_job_submit(adev, entity, job, num_dw);

	return 0;

error_free:
	amdgpu_job_free(job);
	dev_err(adev->dev, "Error scheduling IBs (%d)\n", r);
	return r;
}

static int amdgpu_ttm_fill_mem(struct amdgpu_device *adev,
			       struct amdgpu_ttm_buffer_entity *entity,
			       uint32_t src_data,
			       uint64_t dst_addr, uint32_t byte_count,
			       struct dma_resv *resv,
			       struct dma_fence **fence,
			       bool vm_needs_flush,
			       u64 k_job_id)
{
	unsigned int num_loops, num_dw;
	struct amdgpu_job *job;
	uint32_t max_bytes;
	unsigned int i;
	int r;

	max_bytes = amdgpu_calc_bytes_per_packet(adev->mman.buffer_funcs->fill_max_bytes,
						 byte_count);
	num_loops = DIV_ROUND_UP_ULL(byte_count, max_bytes);
	num_dw = ALIGN(num_loops * adev->mman.buffer_funcs->fill_num_dw, 8);
	r = amdgpu_ttm_prepare_job(adev, entity, num_dw, resv,
				   vm_needs_flush, &job, k_job_id);
	if (r)
		return r;

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

		amdgpu_emit_fill_buffer(adev, &job->ibs[0], src_data, dst_addr,
					cur_size);

		dst_addr += cur_size;
		byte_count -= cur_size;
	}

	*fence = amdgpu_ttm_job_submit(adev, entity, job, num_dw);
	return 0;
}

/**
 * amdgpu_ttm_clear_buffer - fill a buffer with 0
 * @entity: entity to use
 * @bo: the bo to fill
 * @resv: fences contained in this reservation will be used as dependencies.
 * @out_fence: the fence from the last clear will be stored here. It might be
 *             NULL if no job was run.
 * @consider_clear_status: true if region reported as cleared by amdgpu_res_cleared()
 *                         are skipped.
 * @k_job_id: trace id
 *
 */
int amdgpu_ttm_clear_buffer(struct amdgpu_ttm_buffer_entity *entity,
			    struct amdgpu_bo *bo,
			    struct dma_resv *resv,
			    struct dma_fence **out_fence,
			    bool consider_clear_status,
			    u64 k_job_id)
{
	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
	struct dma_fence *fence = NULL;
	struct amdgpu_res_cursor dst;
	bool vm_needs_flush = false;
	int r;

	if (!entity)
		return -EINVAL;

	amdgpu_res_first(bo->tbo.resource, 0, amdgpu_bo_size(bo), &dst);

	mutex_lock(&entity->lock);
	while (dst.remaining) {
		struct dma_fence *next;
		uint64_t cur_size, to;

		if (consider_clear_status && amdgpu_res_cleared(&dst)) {
			amdgpu_res_next(&dst, dst.size);
			continue;
		}

		/* Never fill more than 256MiB at once to avoid timeouts */
		cur_size = min(dst.size, 256ULL << 20);

		r = amdgpu_ttm_map_buffer(entity, &bo->tbo, bo->tbo.resource, &dst,
					  0, false, &cur_size, &to, &vm_needs_flush);
		if (r)
			goto error;

		r = amdgpu_ttm_fill_mem(adev, entity,
					0, to, cur_size, resv,
					&next, vm_needs_flush, k_job_id);
		if (r)
			goto error;

		dma_fence_put(fence);
		fence = next;

		amdgpu_res_next(&dst, cur_size);
	}
error:
	mutex_unlock(&entity->lock);
	*out_fence = fence;
	return r;
}

struct amdgpu_ttm_buffer_entity *
amdgpu_ttm_next_clear_entity(struct amdgpu_device *adev)
{
	struct amdgpu_mman *mman = &adev->mman;
	u32 i;

	if (mman->num_clear_entities == 0)
		return NULL;

	i = atomic_inc_return(&mman->next_clear_entity) %
			      mman->num_clear_entities;
	return &mman->clear_entities[i];
}

/**
 * amdgpu_ttm_evict_resources - evict memory buffers
 * @adev: amdgpu device object
 * @mem_type: evicted BO's memory type
 *
 * Evicts all @mem_type buffers on the lru list of the memory type.
 *
 * Returns:
 * 0 for success or a negative error code on failure.
 */
int amdgpu_ttm_evict_resources(struct amdgpu_device *adev, int mem_type)
{
	struct ttm_resource_manager *man;

	switch (mem_type) {
	case TTM_PL_VRAM:
	case TTM_PL_TT:
	case AMDGPU_PL_GWS:
	case AMDGPU_PL_GDS:
	case AMDGPU_PL_OA:
		man = ttm_manager_type(&adev->mman.bdev, mem_type);
		break;
	default:
		dev_err(adev->dev, "Trying to evict invalid memory type\n");
		return -EINVAL;
	}

	return ttm_resource_manager_evict_all(&adev->mman.bdev, man);
}

void amdgpu_sdma_set_buffer_funcs_scheds(struct amdgpu_device *adev,
					 const struct amdgpu_buffer_funcs *buffer_funcs)
{
	struct drm_gpu_scheduler *sched;
	struct amdgpu_vmhub *hub;
	int i, n;

	adev->mman.buffer_funcs = buffer_funcs;

	for (i = 0, n = 0; i < adev->sdma.num_instances; i++) {
		if (adev->sdma.has_page_queue)
			sched = &adev->sdma.instance[i].page.sched;
		else
			sched = &adev->sdma.instance[i].ring.sched;

		if (!sched->ready)
			continue;

		adev->mman.buffer_funcs_scheds[n++] = sched;
	}

	if (n == 0) {
		adev->mman.num_buffer_funcs_scheds = 0;
		drm_warn(&adev->ddev, "No working sdma ring available\n");
		return;
	}

	hub = &adev->vmhub[AMDGPU_GFXHUB(0)];

	/*
	 * Allow using multiple SDMA schedulers only on GPUs where
	 * we are allowed to do concurrent VM flushes.
	 * This consideration is necessary because all GART windows
	 * are mapped in VMID 0 (the kernel VMID) so each buffer
	 * entity would flush VMID 0 concurrently.
	 *
	 * Also consider the SDMA invalidation workaround on
	 * Navi 1x GPUs, which also prevents us from using
	 * multiple SDMA engines on VMID 0 at the same time.
	 */
	adev->mman.num_buffer_funcs_scheds =
		(adev->vm_manager.concurrent_flush &&
		 !hub->sdma_invalidation_workaround) ? n : 1;
}

#if defined(CONFIG_DEBUG_FS)

static int amdgpu_ttm_page_pool_show(struct seq_file *m, void *unused)
{
	struct amdgpu_device *adev = m->private;

	return ttm_pool_debugfs(&adev->mman.bdev.pool, m);
}

DEFINE_SHOW_ATTRIBUTE(amdgpu_ttm_page_pool);

/*
 * 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,
				    size_t size, loff_t *pos)
{
	struct amdgpu_device *adev = file_inode(f)->i_private;
	ssize_t result = 0;

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

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

	size = min(size, (size_t)(adev->gmc.mc_vram_size - *pos));
	while (size) {
		size_t bytes = min(size, AMDGPU_TTM_VRAM_MAX_DW_READ * 4);
		uint32_t value[AMDGPU_TTM_VRAM_MAX_DW_READ];

		amdgpu_device_vram_access(adev, *pos, value, bytes, false);
		if (copy_to_user(buf, value, bytes))
			return -EFAULT;

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

	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,
				    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->gmc.mc_vram_size)
		return -ENXIO;

	while (size) {
		uint32_t value;

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

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

		amdgpu_device_mm_access(adev, *pos, &value, 4, true);

		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,
};

/*
 * 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,
				 size_t size, loff_t *pos)
{
	struct amdgpu_device *adev = file_inode(f)->i_private;
	struct iommu_domain *dom;
	ssize_t result = 0;
	int r;

	/* retrieve the IOMMU domain if any for this device */
	dom = iommu_get_domain_for_dev(adev->dev);

	while (size) {
		phys_addr_t addr = *pos & PAGE_MASK;
		loff_t off = *pos & ~PAGE_MASK;
		size_t bytes = PAGE_SIZE - off;
		unsigned long pfn;
		struct page *p;
		void *ptr;

		bytes = min(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;

		pfn = addr >> PAGE_SHIFT;
		if (!pfn_valid(pfn))
			return -EPERM;

		p = pfn_to_page(pfn);
		if (p->mapping != adev->mman.bdev.dev_mapping)
			return -EPERM;

		ptr = kmap_local_page(p);
		r = copy_to_user(buf, ptr + off, bytes);
		kunmap_local(ptr);
		if (r)
			return -EFAULT;

		size -= bytes;
		*pos += bytes;
		result += bytes;
	}

	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,
				 size_t size, loff_t *pos)
{
	struct amdgpu_device *adev = file_inode(f)->i_private;
	struct iommu_domain *dom;
	ssize_t result = 0;
	int r;

	dom = iommu_get_domain_for_dev(adev->dev);

	while (size) {
		phys_addr_t addr = *pos & PAGE_MASK;
		loff_t off = *pos & ~PAGE_MASK;
		size_t bytes = PAGE_SIZE - off;
		unsigned long pfn;
		struct page *p;
		void *ptr;

		bytes = min(bytes, size);

		addr = dom ? iommu_iova_to_phys(dom, addr) : addr;

		pfn = addr >> PAGE_SHIFT;
		if (!pfn_valid(pfn))
			return -EPERM;

		p = pfn_to_page(pfn);
		if (p->mapping != adev->mman.bdev.dev_mapping)
			return -EPERM;

		ptr = kmap_local_page(p);
		r = copy_from_user(ptr + off, buf, bytes);
		kunmap_local(ptr);
		if (r)
			return -EFAULT;

		size -= bytes;
		*pos += bytes;
		result += bytes;
	}

	return result;
}

static const struct file_operations amdgpu_ttm_iomem_fops = {
	.owner = THIS_MODULE,
	.read = amdgpu_iomem_read,
	.write = amdgpu_iomem_write,
	.llseek = default_llseek
};

#endif

void amdgpu_ttm_debugfs_init(struct amdgpu_device *adev)
{
#if defined(CONFIG_DEBUG_FS)
	struct drm_minor *minor = adev_to_drm(adev)->primary;
	struct dentry *root = minor->debugfs_root;

	debugfs_create_file_size("amdgpu_vram", 0444, root, adev,
				 &amdgpu_ttm_vram_fops, adev->gmc.mc_vram_size);
	debugfs_create_file("amdgpu_iomem", 0444, root, adev,
			    &amdgpu_ttm_iomem_fops);
	debugfs_create_file("ttm_page_pool", 0444, root, adev,
			    &amdgpu_ttm_page_pool_fops);
	ttm_resource_manager_create_debugfs(ttm_manager_type(&adev->mman.bdev,
							     TTM_PL_VRAM),
					    root, "amdgpu_vram_mm");
	ttm_resource_manager_create_debugfs(ttm_manager_type(&adev->mman.bdev,
							     TTM_PL_TT),
					    root, "amdgpu_gtt_mm");
	ttm_resource_manager_create_debugfs(ttm_manager_type(&adev->mman.bdev,
							     AMDGPU_PL_GDS),
					    root, "amdgpu_gds_mm");
	ttm_resource_manager_create_debugfs(ttm_manager_type(&adev->mman.bdev,
							     AMDGPU_PL_GWS),
					    root, "amdgpu_gws_mm");
	ttm_resource_manager_create_debugfs(ttm_manager_type(&adev->mman.bdev,
							     AMDGPU_PL_OA),
					    root, "amdgpu_oa_mm");

#endif
}