drm/radeon/radeon_ttm.c

/*	$NetBSD: radeon_ttm.c,v 1.10 2018/08/27 04:58:36 riastradh Exp $	*/

/*
 * 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 (at) freedesktop.org>
 *    Thomas Hellstrom <thomas-at-tungstengraphics-dot-com>
 *    Dave Airlie
 */
#include <sys/cdefs.h>
__KERNEL_RCSID(0, "$NetBSD: radeon_ttm.c,v 1.10 2018/08/27 04:58:36 riastradh Exp $");

#include <ttm/ttm_bo_api.h>
#include <ttm/ttm_bo_driver.h>
#include <ttm/ttm_placement.h>
#include <ttm/ttm_module.h>
#include <ttm/ttm_page_alloc.h>
#include <drm/drmP.h>
#include <drm/radeon_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 "radeon_reg.h"
#include "radeon.h"

#ifdef __NetBSD__
#include <uvm/uvm_extern.h>
#include <uvm/uvm_fault.h>
#include <uvm/uvm_param.h>
#include <drm/bus_dma_hacks.h>
#endif

#ifdef _LP64
#define DRM_FILE_PAGE_OFFSET (0x100000000ULL >> PAGE_SHIFT)
#else
#define DRM_FILE_PAGE_OFFSET (0xa0000000UL >> PAGE_SHIFT)
#endif

static int radeon_ttm_debugfs_init(struct radeon_device *rdev);
static void radeon_ttm_debugfs_fini(struct radeon_device *rdev);

static struct radeon_device *radeon_get_rdev(struct ttm_bo_device *bdev)
{
	struct radeon_mman *mman;
	struct radeon_device *rdev;

	mman = container_of(bdev, struct radeon_mman, bdev);
	rdev = container_of(mman, struct radeon_device, mman);
	return rdev;
}


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

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

static int radeon_ttm_global_init(struct radeon_device *rdev)
{
	struct drm_global_reference *global_ref;
	int r;

	rdev->mman.mem_global_referenced = false;
	global_ref = &rdev->mman.mem_global_ref;
	global_ref->global_type = DRM_GLOBAL_TTM_MEM;
	global_ref->size = sizeof(struct ttm_mem_global);
	global_ref->init = &radeon_ttm_mem_global_init;
	global_ref->release = &radeon_ttm_mem_global_release;
	r = drm_global_item_ref(global_ref);
	if (r != 0) {
		DRM_ERROR("Failed setting up TTM memory accounting "
			  "subsystem.\n");
		return r;
	}

	rdev->mman.bo_global_ref.mem_glob =
		rdev->mman.mem_global_ref.object;
	global_ref = &rdev->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 != 0) {
		DRM_ERROR("Failed setting up TTM BO subsystem.\n");
		drm_global_item_unref(&rdev->mman.mem_global_ref);
		return r;
	}

	rdev->mman.mem_global_referenced = true;
	return 0;
}

static void radeon_ttm_global_fini(struct radeon_device *rdev)
{
	if (rdev->mman.mem_global_referenced) {
		drm_global_item_unref(&rdev->mman.bo_global_ref.ref);
		drm_global_item_unref(&rdev->mman.mem_global_ref);
		rdev->mman.mem_global_referenced = false;
	}
}

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

static int radeon_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
				struct ttm_mem_type_manager *man)
{
	struct radeon_device *rdev;

	rdev = radeon_get_rdev(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 = &ttm_bo_manager_func;
		man->gpu_offset = rdev->mc.gtt_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;
#if IS_ENABLED(CONFIG_AGP)
		if (rdev->flags & RADEON_IS_AGP) {
			if (!rdev->ddev->agp) {
				DRM_ERROR("AGP is not enabled for memory type %u\n",
					  (unsigned)type);
				return -EINVAL;
			}
			if (!rdev->ddev->agp->cant_use_aperture)
				man->flags = TTM_MEMTYPE_FLAG_MAPPABLE;
			man->available_caching = TTM_PL_FLAG_UNCACHED |
						 TTM_PL_FLAG_WC;
			man->default_caching = TTM_PL_FLAG_WC;
		}
#endif
		break;
	case TTM_PL_VRAM:
		/* "On-card" video ram */
		man->func = &ttm_bo_manager_func;
		man->gpu_offset = rdev->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;
	default:
		DRM_ERROR("Unsupported memory type %u\n", (unsigned)type);
		return -EINVAL;
	}
	return 0;
}

static void radeon_evict_flags(struct ttm_buffer_object *bo,
				struct ttm_placement *placement)
{
	static struct ttm_place placements = {
		.fpfn = 0,
		.lpfn = 0,
		.flags = TTM_PL_MASK_CACHING | TTM_PL_FLAG_SYSTEM
	};

	struct radeon_bo *rbo;

	if (!radeon_ttm_bo_is_radeon_bo(bo)) {
		placement->placement = &placements;
		placement->busy_placement = &placements;
		placement->num_placement = 1;
		placement->num_busy_placement = 1;
		return;
	}
	rbo = container_of(bo, struct radeon_bo, tbo);
	switch (bo->mem.mem_type) {
	case TTM_PL_VRAM:
		if (rbo->rdev->ring[radeon_copy_ring_index(rbo->rdev)].ready == false)
			radeon_ttm_placement_from_domain(rbo, RADEON_GEM_DOMAIN_CPU);
		else if (rbo->rdev->mc.visible_vram_size < rbo->rdev->mc.real_vram_size &&
			 bo->mem.start < (rbo->rdev->mc.visible_vram_size >> PAGE_SHIFT)) {
			unsigned fpfn = rbo->rdev->mc.visible_vram_size >> PAGE_SHIFT;
			int i;

			/* 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
			 */
			radeon_ttm_placement_from_domain(rbo, RADEON_GEM_DOMAIN_VRAM |
							 RADEON_GEM_DOMAIN_GTT);
			rbo->placement.num_busy_placement = 0;
			for (i = 0; i < rbo->placement.num_placement; i++) {
				if (rbo->placements[i].flags & TTM_PL_FLAG_VRAM) {
					if (rbo->placements[i].fpfn < fpfn)
						rbo->placements[i].fpfn = fpfn;
				} else {
					rbo->placement.busy_placement =
						&rbo->placements[i];
					rbo->placement.num_busy_placement = 1;
				}
			}
		} else
			radeon_ttm_placement_from_domain(rbo, RADEON_GEM_DOMAIN_GTT);
		break;
	case TTM_PL_TT:
	default:
		radeon_ttm_placement_from_domain(rbo, RADEON_GEM_DOMAIN_CPU);
	}
	*placement = rbo->placement;
}

static int radeon_verify_access(struct ttm_buffer_object *bo, struct file *filp)
{
	struct radeon_bo *rbo = container_of(bo, struct radeon_bo, tbo);

	if (radeon_ttm_tt_has_userptr(bo->ttm))
		return -EPERM;
	return drm_vma_node_verify_access(&rbo->gem_base.vma_node, filp);
}

static void radeon_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 int radeon_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 radeon_device *rdev;
	uint64_t old_start, new_start;
	struct radeon_fence *fence;
	unsigned num_pages;
	int r, ridx;

	rdev = radeon_get_rdev(bo->bdev);
	ridx = radeon_copy_ring_index(rdev);
	old_start = (u64)old_mem->start << PAGE_SHIFT;
	new_start = (u64)new_mem->start << PAGE_SHIFT;

	switch (old_mem->mem_type) {
	case TTM_PL_VRAM:
		old_start += rdev->mc.vram_start;
		break;
	case TTM_PL_TT:
		old_start += rdev->mc.gtt_start;
		break;
	default:
		DRM_ERROR("Unknown placement %d\n", old_mem->mem_type);
		return -EINVAL;
	}
	switch (new_mem->mem_type) {
	case TTM_PL_VRAM:
		new_start += rdev->mc.vram_start;
		break;
	case TTM_PL_TT:
		new_start += rdev->mc.gtt_start;
		break;
	default:
		DRM_ERROR("Unknown placement %d\n", old_mem->mem_type);
		return -EINVAL;
	}
	if (!rdev->ring[ridx].ready) {
		DRM_ERROR("Trying to move memory with ring turned off.\n");
		return -EINVAL;
	}

	BUILD_BUG_ON((PAGE_SIZE % RADEON_GPU_PAGE_SIZE) != 0);

	num_pages = new_mem->num_pages * (PAGE_SIZE / RADEON_GPU_PAGE_SIZE);
	fence = radeon_copy(rdev, old_start, new_start, num_pages, bo->resv);
	if (IS_ERR(fence))
		return PTR_ERR(fence);

	r = ttm_bo_move_accel_cleanup(bo, &fence->base,
				      evict, no_wait_gpu, new_mem);
	radeon_fence_unref(&fence);
	return r;
}

static int radeon_move_vram_ram(struct ttm_buffer_object *bo,
				bool evict, bool interruptible,
				bool no_wait_gpu,
				struct ttm_mem_reg *new_mem)
{
	struct radeon_device *rdev __unused;
	struct ttm_mem_reg *old_mem = &bo->mem;
	struct ttm_mem_reg tmp_mem;
	struct ttm_place placements;
	struct ttm_placement placement;
	int r;

	rdev = radeon_get_rdev(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,
			     interruptible, no_wait_gpu);
	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);
	if (unlikely(r)) {
		goto out_cleanup;
	}
	r = radeon_move_blit(bo, true, no_wait_gpu, &tmp_mem, old_mem);
	if (unlikely(r)) {
		goto out_cleanup;
	}
	r = ttm_bo_move_ttm(bo, true, no_wait_gpu, new_mem);
out_cleanup:
	ttm_bo_mem_put(bo, &tmp_mem);
	return r;
}

static int radeon_move_ram_vram(struct ttm_buffer_object *bo,
				bool evict, bool interruptible,
				bool no_wait_gpu,
				struct ttm_mem_reg *new_mem)
{
	struct radeon_device *rdev __unused;
	struct ttm_mem_reg *old_mem = &bo->mem;
	struct ttm_mem_reg tmp_mem;
	struct ttm_placement placement;
	struct ttm_place placements;
	int r;

	rdev = radeon_get_rdev(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,
			     interruptible, no_wait_gpu);
	if (unlikely(r)) {
		return r;
	}
	r = ttm_bo_move_ttm(bo, true, no_wait_gpu, &tmp_mem);
	if (unlikely(r)) {
		goto out_cleanup;
	}
	r = radeon_move_blit(bo, true, 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 radeon_bo_move(struct ttm_buffer_object *bo,
			bool evict, bool interruptible,
			bool no_wait_gpu,
			struct ttm_mem_reg *new_mem)
{
	struct radeon_device *rdev;
	struct ttm_mem_reg *old_mem = &bo->mem;
	int r;

	rdev = radeon_get_rdev(bo->bdev);
	if (old_mem->mem_type == TTM_PL_SYSTEM && bo->ttm == NULL) {
		radeon_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 */
		radeon_move_null(bo, new_mem);
		return 0;
	}
	if (!rdev->ring[radeon_copy_ring_index(rdev)].ready ||
	    rdev->asic->copy.copy == NULL) {
		/* use memcpy */
		goto memcpy;
	}

	if (old_mem->mem_type == TTM_PL_VRAM &&
	    new_mem->mem_type == TTM_PL_SYSTEM) {
		r = radeon_move_vram_ram(bo, evict, interruptible,
					no_wait_gpu, new_mem);
	} else if (old_mem->mem_type == TTM_PL_SYSTEM &&
		   new_mem->mem_type == TTM_PL_VRAM) {
		r = radeon_move_ram_vram(bo, evict, interruptible,
					    no_wait_gpu, new_mem);
	} else {
		r = radeon_move_blit(bo, evict, no_wait_gpu, new_mem, old_mem);
	}

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

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

static int radeon_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 radeon_device *rdev = radeon_get_rdev(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:
#if IS_ENABLED(CONFIG_AGP)
		if (rdev->flags & RADEON_IS_AGP) {
			/* RADEON_IS_AGP is set only if AGP is active */
			mem->bus.offset = mem->start << PAGE_SHIFT;
			mem->bus.base = rdev->mc.agp_base;
			mem->bus.is_iomem = !rdev->ddev->agp->cant_use_aperture;
			KASSERTMSG((mem->bus.base & (PAGE_SIZE - 1)) == 0,
			    "agp aperture is not page-aligned: %lx",
			    mem->bus.base);
			KASSERT((mem->bus.offset & (PAGE_SIZE - 1)) == 0);
		}
#endif
		break;
	case TTM_PL_VRAM:
		mem->bus.offset = mem->start << PAGE_SHIFT;
		/* check if it's visible */
		if ((mem->bus.offset + mem->bus.size) > rdev->mc.visible_vram_size)
			return -EINVAL;
		mem->bus.base = rdev->mc.aper_base;
		mem->bus.is_iomem = true;
#ifdef __alpha__
		/*
		 * Alpha: use bus.addr to hold the ioremap() return,
		 * so we can modify bus.base below.
		 */
		if (mem->placement & TTM_PL_FLAG_WC)
			mem->bus.addr =
				ioremap_wc(mem->bus.base + mem->bus.offset,
					   mem->bus.size);
		else
			mem->bus.addr =
				ioremap_nocache(mem->bus.base + mem->bus.offset,
						mem->bus.size);

		/*
		 * Alpha: Use just the bus offset plus
		 * the hose/domain memory base for bus.base.
		 * It then can be used to build PTEs for VRAM
		 * access, as done in ttm_bo_vm_fault().
		 */
		mem->bus.base = (mem->bus.base & 0x0ffffffffUL) +
			rdev->ddev->hose->dense_mem_base;
#endif
		KASSERTMSG((mem->bus.base & (PAGE_SIZE - 1)) == 0,
		    "mc aperture is not page-aligned: %lx",
		    mem->bus.base);
		KASSERT((mem->bus.offset & (PAGE_SIZE - 1)) == 0);
		break;
	default:
		return -EINVAL;
	}
	return 0;
}

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

/*
 * TTM backend functions.
 */
struct radeon_ttm_tt {
	struct ttm_dma_tt		ttm;
	struct radeon_device		*rdev;
	u64				offset;

	uint64_t			userptr;
	struct mm_struct		*usermm;
	uint32_t			userflags;
};

/* prepare the sg table with the user pages */
static int radeon_ttm_tt_pin_userptr(struct ttm_tt *ttm)
{
	struct radeon_device *rdev = radeon_get_rdev(ttm->bdev);
	struct radeon_ttm_tt *gtt = (void *)ttm;
	unsigned pinned = 0, nents;
	int r;

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

	if (current->mm != gtt->usermm)
		return -EPERM;

	if (gtt->userflags & RADEON_GEM_USERPTR_ANONONLY) {
		/* check that we only pin down 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)
			return -EPERM;
	}

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

		r = get_user_pages(current, current->mm, userptr, num_pages,
				   write, 0, pages, NULL);
		if (r < 0)
			goto release_pages;

		pinned += r;

	} while (pinned < ttm->num_pages);

	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(rdev->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);

release_pages:
	release_pages(ttm->pages, pinned, 0);
	return r;
}

static void radeon_ttm_tt_unpin_userptr(struct ttm_tt *ttm)
{
	struct radeon_device *rdev = radeon_get_rdev(ttm->bdev);
	struct radeon_ttm_tt *gtt = (void *)ttm;
	struct sg_page_iter sg_iter;

	int write = !(gtt->userflags & RADEON_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(rdev->dev, ttm->sg->sgl, ttm->sg->nents, direction);

	for_each_sg_page(ttm->sg->sgl, &sg_iter, ttm->sg->nents, 0) {
		struct page *page = sg_page_iter_page(&sg_iter);
		if (!(gtt->userflags & RADEON_GEM_USERPTR_READONLY))
			set_page_dirty(page);

		mark_page_accessed(page);
		page_cache_release(page);
	}

	sg_free_table(ttm->sg);
}

static int radeon_ttm_backend_bind(struct ttm_tt *ttm,
				   struct ttm_mem_reg *bo_mem)
{
	struct radeon_ttm_tt *gtt = (void*)ttm;
	uint32_t flags = RADEON_GART_PAGE_VALID | RADEON_GART_PAGE_READ |
		RADEON_GART_PAGE_WRITE;
	int r;

	if (gtt->userptr) {
		radeon_ttm_tt_pin_userptr(ttm);
		flags &= ~RADEON_GART_PAGE_WRITE;
	}

	gtt->offset = (unsigned long)(bo_mem->start << PAGE_SHIFT);
	if (!ttm->num_pages) {
		WARN(1, "nothing to bind %lu pages for mreg %p back %p!\n",
		     ttm->num_pages, bo_mem, ttm);
	}
	if (ttm->caching_state == tt_cached)
		flags |= RADEON_GART_PAGE_SNOOP;
	r = radeon_gart_bind(gtt->rdev, gtt->offset, ttm->num_pages,
			     ttm->pages, gtt->ttm.dma_address, flags);
	if (r) {
		DRM_ERROR("failed to bind %lu pages at 0x%08X\n",
			  ttm->num_pages, (unsigned)gtt->offset);
		return r;
	}
	return 0;
}

static int radeon_ttm_backend_unbind(struct ttm_tt *ttm)
{
	struct radeon_ttm_tt *gtt = (void *)ttm;

	radeon_gart_unbind(gtt->rdev, gtt->offset, ttm->num_pages);

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

	return 0;
}

static void radeon_ttm_backend_destroy(struct ttm_tt *ttm)
{
	struct radeon_ttm_tt *gtt = (void *)ttm;

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

static struct ttm_backend_func radeon_backend_func = {
	.bind = &radeon_ttm_backend_bind,
	.unbind = &radeon_ttm_backend_unbind,
	.destroy = &radeon_ttm_backend_destroy,
};

static struct ttm_tt *radeon_ttm_tt_create(struct ttm_bo_device *bdev,
				    unsigned long size, uint32_t page_flags,
				    struct page *dummy_read_page)
{
	struct radeon_device *rdev;
	struct radeon_ttm_tt *gtt;

	rdev = radeon_get_rdev(bdev);
#if IS_ENABLED(CONFIG_AGP)
	if (rdev->flags & RADEON_IS_AGP) {
		return ttm_agp_tt_create(bdev, rdev->ddev->agp->bridge,
					 size, page_flags, dummy_read_page);
	}
#endif

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

static struct radeon_ttm_tt *radeon_ttm_tt_to_gtt(struct ttm_tt *ttm)
{
	if (!ttm || ttm->func != &radeon_backend_func)
		return NULL;
	return (struct radeon_ttm_tt *)ttm;
}

static int radeon_ttm_tt_populate(struct ttm_tt *ttm)
{
	struct radeon_ttm_tt *gtt = radeon_ttm_tt_to_gtt(ttm);
	struct radeon_device *rdev;
#ifndef __NetBSD__
	unsigned i;
	int r;
#endif
	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) {
#ifdef __NetBSD__		/* XXX drm prime */
		return -EINVAL;
#else
		drm_prime_sg_to_page_addr_arrays(ttm->sg, ttm->pages,
						 gtt->ttm.dma_address, ttm->num_pages);
		ttm->state = tt_unbound;
		return 0;
#endif
	}

	rdev = radeon_get_rdev(ttm->bdev);
#if IS_ENABLED(CONFIG_AGP)
	if (rdev->flags & RADEON_IS_AGP) {
		return ttm_agp_tt_populate(ttm);
	}
#endif

#ifdef __NetBSD__
	/* XXX errno NetBSD->Linux */
	return ttm_bus_dma_populate(&gtt->ttm);
#else

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

	r = ttm_pool_populate(ttm);
	if (r) {
		return r;
	}

	for (i = 0; i < ttm->num_pages; i++) {
		gtt->ttm.dma_address[i] = pci_map_page(rdev->pdev, ttm->pages[i],
						       0, PAGE_SIZE,
						       PCI_DMA_BIDIRECTIONAL);
		if (pci_dma_mapping_error(rdev->pdev, gtt->ttm.dma_address[i])) {
			while (i--) {
				pci_unmap_page(rdev->pdev, gtt->ttm.dma_address[i],
					       PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
				gtt->ttm.dma_address[i] = 0;
			}
			ttm_pool_unpopulate(ttm);
			return -EFAULT;
		}
	}
	return 0;
#endif
}

static void radeon_ttm_tt_unpopulate(struct ttm_tt *ttm)
{
	struct radeon_device *rdev;
	struct radeon_ttm_tt *gtt = radeon_ttm_tt_to_gtt(ttm);
#ifndef __NetBSD__
	unsigned i;
#endif
	bool slave = !!(ttm->page_flags & TTM_PAGE_FLAG_SG);

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

	if (slave)
		return;

	rdev = radeon_get_rdev(ttm->bdev);
#if IS_ENABLED(CONFIG_AGP)
	if (rdev->flags & RADEON_IS_AGP) {
		ttm_agp_tt_unpopulate(ttm);
		return;
	}
#endif

#ifdef __NetBSD__
	ttm_bus_dma_unpopulate(&gtt->ttm);
	return;
#else

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

	for (i = 0; i < ttm->num_pages; i++) {
		if (gtt->ttm.dma_address[i]) {
			pci_unmap_page(rdev->pdev, gtt->ttm.dma_address[i],
				       PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
		}
	}

	ttm_pool_unpopulate(ttm);
#endif
}

#ifdef __NetBSD__
static void radeon_ttm_tt_swapout(struct ttm_tt *ttm)
{
	struct radeon_ttm_tt *gtt = container_of(ttm, struct radeon_ttm_tt,
	    ttm.ttm);
	struct ttm_dma_tt *ttm_dma = &gtt->ttm;

	ttm_bus_dma_swapout(ttm_dma);
}

static int	radeon_ttm_fault(struct uvm_faultinfo *, vaddr_t,
		    struct vm_page **, int, int, vm_prot_t, int);

static const struct uvm_pagerops radeon_uvm_ops = {
	.pgo_reference = &ttm_bo_uvm_reference,
	.pgo_detach = &ttm_bo_uvm_detach,
	.pgo_fault = &radeon_ttm_fault,
};
#endif

int radeon_ttm_tt_set_userptr(struct ttm_tt *ttm, uint64_t addr,
			      uint32_t flags)
{
	struct radeon_ttm_tt *gtt = radeon_ttm_tt_to_gtt(ttm);

	if (gtt == NULL)
		return -EINVAL;

	gtt->userptr = addr;
	gtt->usermm = current->mm;
	gtt->userflags = flags;
	return 0;
}

bool radeon_ttm_tt_has_userptr(struct ttm_tt *ttm)
{
	struct radeon_ttm_tt *gtt = radeon_ttm_tt_to_gtt(ttm);

	if (gtt == NULL)
		return false;

	return !!gtt->userptr;
}

bool radeon_ttm_tt_is_readonly(struct ttm_tt *ttm)
{
	struct radeon_ttm_tt *gtt = radeon_ttm_tt_to_gtt(ttm);

	if (gtt == NULL)
		return false;

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

static struct ttm_bo_driver radeon_bo_driver = {
	.ttm_tt_create = &radeon_ttm_tt_create,
	.ttm_tt_populate = &radeon_ttm_tt_populate,
	.ttm_tt_unpopulate = &radeon_ttm_tt_unpopulate,
#ifdef __NetBSD__
	.ttm_tt_swapout = &radeon_ttm_tt_swapout,
	.ttm_uvm_ops = &radeon_uvm_ops,
#endif
	.invalidate_caches = &radeon_invalidate_caches,
	.init_mem_type = &radeon_init_mem_type,
	.evict_flags = &radeon_evict_flags,
	.move = &radeon_bo_move,
	.verify_access = &radeon_verify_access,
	.move_notify = &radeon_bo_move_notify,
	.fault_reserve_notify = &radeon_bo_fault_reserve_notify,
	.io_mem_reserve = &radeon_ttm_io_mem_reserve,
	.io_mem_free = &radeon_ttm_io_mem_free,
};

int radeon_ttm_init(struct radeon_device *rdev)
{
	int r;

	r = radeon_ttm_global_init(rdev);
	if (r) {
		return r;
	}
	/* No others user of address space so set it to 0 */
	r = ttm_bo_device_init(&rdev->mman.bdev,
			       rdev->mman.bo_global_ref.ref.object,
			       &radeon_bo_driver,
#ifdef __NetBSD__
			       rdev->ddev->bst,
			       rdev->ddev->dmat,
#else
			       rdev->ddev->anon_inode->i_mapping,
#endif
			       DRM_FILE_PAGE_OFFSET,
			       rdev->need_dma32);
	if (r) {
		DRM_ERROR("failed initializing buffer object driver(%d).\n", r);
		return r;
	}
	rdev->mman.initialized = true;
	r = ttm_bo_init_mm(&rdev->mman.bdev, TTM_PL_VRAM,
				rdev->mc.real_vram_size >> PAGE_SHIFT);
	if (r) {
		DRM_ERROR("Failed initializing VRAM heap.\n");
		return r;
	}
	/* Change the size here instead of the init above so only lpfn is affected */
	radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);

	r = radeon_bo_create(rdev, 256 * 1024, PAGE_SIZE, true,
			     RADEON_GEM_DOMAIN_VRAM, 0, NULL,
			     NULL, &rdev->stollen_vga_memory);
	if (r) {
		return r;
	}
	r = radeon_bo_reserve(rdev->stollen_vga_memory, false);
	if (r)
		return r;
	r = radeon_bo_pin(rdev->stollen_vga_memory, RADEON_GEM_DOMAIN_VRAM, NULL);
	radeon_bo_unreserve(rdev->stollen_vga_memory);
	if (r) {
		radeon_bo_unref(&rdev->stollen_vga_memory);
		return r;
	}
	DRM_INFO("radeon: %uM of VRAM memory ready\n",
		 (unsigned) (rdev->mc.real_vram_size / (1024 * 1024)));
	r = ttm_bo_init_mm(&rdev->mman.bdev, TTM_PL_TT,
				rdev->mc.gtt_size >> PAGE_SHIFT);
	if (r) {
		DRM_ERROR("Failed initializing GTT heap.\n");
		return r;
	}
	DRM_INFO("radeon: %uM of GTT memory ready.\n",
		 (unsigned)(rdev->mc.gtt_size / (1024 * 1024)));

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

void radeon_ttm_fini(struct radeon_device *rdev)
{
	int r;

	if (!rdev->mman.initialized)
		return;
	radeon_ttm_debugfs_fini(rdev);
	if (rdev->stollen_vga_memory) {
		r = radeon_bo_reserve(rdev->stollen_vga_memory, false);
		if (r == 0) {
			radeon_bo_unpin(rdev->stollen_vga_memory);
			radeon_bo_unreserve(rdev->stollen_vga_memory);
		}
		radeon_bo_unref(&rdev->stollen_vga_memory);
	}
	ttm_bo_clean_mm(&rdev->mman.bdev, TTM_PL_VRAM);
	ttm_bo_clean_mm(&rdev->mman.bdev, TTM_PL_TT);
	ttm_bo_device_release(&rdev->mman.bdev);
	radeon_gart_fini(rdev);
	radeon_ttm_global_fini(rdev);
	rdev->mman.initialized = false;
	DRM_INFO("radeon: ttm finalized\n");
}

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

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

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

#ifdef __NetBSD__

static int
radeon_ttm_fault(struct uvm_faultinfo *ufi, vaddr_t vaddr,
    struct vm_page **pps, int npages, int centeridx, vm_prot_t access_type,
    int flags)
{
	struct uvm_object *const uobj = ufi->entry->object.uvm_obj;
	struct ttm_buffer_object *const bo = container_of(uobj,
	    struct ttm_buffer_object, uvmobj);
	struct radeon_device *const rdev = radeon_get_rdev(bo->bdev);
	int error;

	KASSERT(rdev != NULL);
	down_read(&rdev->pm.mclk_lock);
	error = ttm_bo_uvm_fault(ufi, vaddr, pps, npages, centeridx,
	    access_type, flags);
	up_read(&rdev->pm.mclk_lock);

	return error;
}

int
radeon_mmap_object(struct drm_device *dev, off_t offset, size_t size,
    vm_prot_t prot, struct uvm_object **uobjp, voff_t *uoffsetp,
    struct file *file)
{
	struct radeon_device *rdev = dev->dev_private;

	KASSERT(0 == (offset & (PAGE_SIZE - 1)));

	if (__predict_false(rdev == NULL))	/* XXX How?? */
		return -EINVAL;

	if (__predict_false((offset >> PAGE_SHIFT) < DRM_FILE_PAGE_OFFSET))
		return drm_mmap_object(dev, offset, size, prot, uobjp,
		    uoffsetp, file);
	else
		return ttm_bo_mmap_object(&rdev->mman.bdev, offset, size, prot,
		    uobjp, uoffsetp, file);
}

#else

static struct vm_operations_struct radeon_ttm_vm_ops;
static const struct vm_operations_struct *ttm_vm_ops = NULL;

static int radeon_ttm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
{
	struct ttm_buffer_object *bo;
	struct radeon_device *rdev;
	int r;

	bo = (struct ttm_buffer_object *)vma->vm_private_data;
	if (bo == NULL) {
		return VM_FAULT_NOPAGE;
	}
	rdev = radeon_get_rdev(bo->bdev);
	down_read(&rdev->pm.mclk_lock);
	r = ttm_vm_ops->fault(vma, vmf);
	up_read(&rdev->pm.mclk_lock);
	return r;
}

int radeon_mmap(struct file *filp, struct vm_area_struct *vma)
{
	struct drm_file *file_priv;
	struct radeon_device *rdev;
	int r;

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

	file_priv = filp->private_data;
	rdev = file_priv->minor->dev->dev_private;
	if (rdev == NULL) {
		return -EINVAL;
	}
	r = ttm_bo_mmap(filp, vma, &rdev->mman.bdev);
	if (unlikely(r != 0)) {
		return r;
	}
	if (unlikely(ttm_vm_ops == NULL)) {
		ttm_vm_ops = vma->vm_ops;
		radeon_ttm_vm_ops = *ttm_vm_ops;
		radeon_ttm_vm_ops.fault = &radeon_ttm_fault;
	}
	vma->vm_ops = &radeon_ttm_vm_ops;
	return 0;
}

#endif	/* __NetBSD__ */

#if defined(CONFIG_DEBUG_FS)

static int radeon_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 radeon_device *rdev = dev->dev_private;
	struct drm_mm *mm = (struct drm_mm *)rdev->mman.bdev.man[ttm_pl].priv;
	int ret;
	struct ttm_bo_global *glob = rdev->mman.bdev.glob;

	spin_lock(&glob->lru_lock);
	ret = drm_mm_dump_table(m, mm);
	spin_unlock(&glob->lru_lock);
	return ret;
}

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

static struct drm_info_list radeon_ttm_debugfs_list[] = {
	{"radeon_vram_mm", radeon_mm_dump_table, 0, &ttm_pl_vram},
	{"radeon_gtt_mm", radeon_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 int radeon_ttm_vram_open(struct inode *inode, struct file *filep)
{
	struct radeon_device *rdev = inode->i_private;
	i_size_write(inode, rdev->mc.mc_vram_size);
	filep->private_data = inode->i_private;
	return 0;
}

static ssize_t radeon_ttm_vram_read(struct file *f, char __user *buf,
				    size_t size, loff_t *pos)
{
	struct radeon_device *rdev = f->private_data;
	ssize_t result = 0;
	int r;

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

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

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

		spin_lock_irqsave(&rdev->mmio_idx_lock, flags);
		WREG32(RADEON_MM_INDEX, ((uint32_t)*pos) | 0x80000000);
		if (rdev->family >= CHIP_CEDAR)
			WREG32(EVERGREEN_MM_INDEX_HI, *pos >> 31);
		value = RREG32(RADEON_MM_DATA);
		spin_unlock_irqrestore(&rdev->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 const struct file_operations radeon_ttm_vram_fops = {
	.owner = THIS_MODULE,
	.open = radeon_ttm_vram_open,
	.read = radeon_ttm_vram_read,
	.llseek = default_llseek
};

static int radeon_ttm_gtt_open(struct inode *inode, struct file *filep)
{
	struct radeon_device *rdev = inode->i_private;
	i_size_write(inode, rdev->mc.gtt_size);
	filep->private_data = inode->i_private;
	return 0;
}

static ssize_t radeon_ttm_gtt_read(struct file *f, char __user *buf,
				   size_t size, loff_t *pos)
{
	struct radeon_device *rdev = f->private_data;
	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 >= rdev->gart.num_cpu_pages)
			return result;

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

			r = copy_to_user(buf, ptr, cur_size);
			kunmap(rdev->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 radeon_ttm_gtt_fops = {
	.owner = THIS_MODULE,
	.open = radeon_ttm_gtt_open,
	.read = radeon_ttm_gtt_read,
	.llseek = default_llseek
};

#endif

static int radeon_ttm_debugfs_init(struct radeon_device *rdev)
{
#if defined(CONFIG_DEBUG_FS)
	unsigned count;

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

	ent = debugfs_create_file("radeon_vram", S_IFREG | S_IRUGO, root,
				  rdev, &radeon_ttm_vram_fops);
	if (IS_ERR(ent))
		return PTR_ERR(ent);
	rdev->mman.vram = ent;

	ent = debugfs_create_file("radeon_gtt", S_IFREG | S_IRUGO, root,
				  rdev, &radeon_ttm_gtt_fops);
	if (IS_ERR(ent))
		return PTR_ERR(ent);
	rdev->mman.gtt = ent;

	count = ARRAY_SIZE(radeon_ttm_debugfs_list);

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

	return radeon_debugfs_add_files(rdev, radeon_ttm_debugfs_list, count);
#else

	return 0;
#endif
}

static void radeon_ttm_debugfs_fini(struct radeon_device *rdev)
{
#if defined(CONFIG_DEBUG_FS)

	debugfs_remove(rdev->mman.vram);
	rdev->mman.vram = NULL;

	debugfs_remove(rdev->mman.gtt);
	rdev->mman.gtt = NULL;
#endif
}