i915_gem.c

/*
 * Copyright © 2008 Intel Corporation
 *
 * 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, sublicense,
 * 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 above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * 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 NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS 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.
 *
 * Authors:
 *    Eric Anholt <eric@anholt.net>
 *
 */

#include "drmP.h"
#include "drm.h"
#include "i915_drm.h"
#include "i915_drv.h"
#include "i915_trace.h"
#include "intel_drv.h"
#include <linux/slab.h>
#include <linux/swap.h>
#include <linux/pci.h>
#include <linux/intel-gtt.h>

static uint32_t i915_gem_get_gtt_alignment(struct drm_gem_object *obj);

static int i915_gem_object_flush_gpu_write_domain(struct drm_gem_object *obj,
						  bool pipelined);
static void i915_gem_object_flush_gtt_write_domain(struct drm_gem_object *obj);
static void i915_gem_object_flush_cpu_write_domain(struct drm_gem_object *obj);
static int i915_gem_object_set_to_cpu_domain(struct drm_gem_object *obj,
					     int write);
static int i915_gem_object_set_cpu_read_domain_range(struct drm_gem_object *obj,
						     uint64_t offset,
						     uint64_t size);
static void i915_gem_object_set_to_full_cpu_read_domain(struct drm_gem_object *obj);
static int i915_gem_object_wait_rendering(struct drm_gem_object *obj,
					  bool interruptible);
static int i915_gem_object_bind_to_gtt(struct drm_gem_object *obj,
					   unsigned alignment);
static void i915_gem_clear_fence_reg(struct drm_gem_object *obj);
static int i915_gem_phys_pwrite(struct drm_device *dev, struct drm_gem_object *obj,
				struct drm_i915_gem_pwrite *args,
				struct drm_file *file_priv);
static void i915_gem_free_object_tail(struct drm_gem_object *obj);

static LIST_HEAD(shrink_list);
static DEFINE_SPINLOCK(shrink_list_lock);

static inline bool
i915_gem_object_is_inactive(struct drm_i915_gem_object *obj_priv)
{
	return obj_priv->gtt_space &&
		!obj_priv->active &&
		obj_priv->pin_count == 0;
}

int i915_gem_do_init(struct drm_device *dev, unsigned long start,
		     unsigned long end)
{
	drm_i915_private_t *dev_priv = dev->dev_private;

	if (start >= end ||
	    (start & (PAGE_SIZE - 1)) != 0 ||
	    (end & (PAGE_SIZE - 1)) != 0) {
		return -EINVAL;
	}

	drm_mm_init(&dev_priv->mm.gtt_space, start,
		    end - start);

	dev->gtt_total = (uint32_t) (end - start);

	return 0;
}

int
i915_gem_init_ioctl(struct drm_device *dev, void *data,
		    struct drm_file *file_priv)
{
	struct drm_i915_gem_init *args = data;
	int ret;

	mutex_lock(&dev->struct_mutex);
	ret = i915_gem_do_init(dev, args->gtt_start, args->gtt_end);
	mutex_unlock(&dev->struct_mutex);

	return ret;
}

int
i915_gem_get_aperture_ioctl(struct drm_device *dev, void *data,
			    struct drm_file *file_priv)
{
	struct drm_i915_gem_get_aperture *args = data;

	if (!(dev->driver->driver_features & DRIVER_GEM))
		return -ENODEV;

	args->aper_size = dev->gtt_total;
	args->aper_available_size = (args->aper_size -
				     atomic_read(&dev->pin_memory));

	return 0;
}


/**
 * Creates a new mm object and returns a handle to it.
 */
int
i915_gem_create_ioctl(struct drm_device *dev, void *data,
		      struct drm_file *file_priv)
{
	struct drm_i915_gem_create *args = data;
	struct drm_gem_object *obj;
	int ret;
	u32 handle;

	args->size = roundup(args->size, PAGE_SIZE);

	/* Allocate the new object */
	obj = i915_gem_alloc_object(dev, args->size);
	if (obj == NULL)
		return -ENOMEM;

	ret = drm_gem_handle_create(file_priv, obj, &handle);
	if (ret) {
		drm_gem_object_unreference_unlocked(obj);
		return ret;
	}

	/* Sink the floating reference from kref_init(handlecount) */
	drm_gem_object_handle_unreference_unlocked(obj);

	args->handle = handle;
	return 0;
}

static inline int
fast_shmem_read(struct page **pages,
		loff_t page_base, int page_offset,
		char __user *data,
		int length)
{
	char __iomem *vaddr;
	int unwritten;

	vaddr = kmap_atomic(pages[page_base >> PAGE_SHIFT], KM_USER0);
	if (vaddr == NULL)
		return -ENOMEM;
	unwritten = __copy_to_user_inatomic(data, vaddr + page_offset, length);
	kunmap_atomic(vaddr, KM_USER0);

	if (unwritten)
		return -EFAULT;

	return 0;
}

static int i915_gem_object_needs_bit17_swizzle(struct drm_gem_object *obj)
{
	drm_i915_private_t *dev_priv = obj->dev->dev_private;
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);

	return dev_priv->mm.bit_6_swizzle_x == I915_BIT_6_SWIZZLE_9_10_17 &&
		obj_priv->tiling_mode != I915_TILING_NONE;
}

static inline void
slow_shmem_copy(struct page *dst_page,
		int dst_offset,
		struct page *src_page,
		int src_offset,
		int length)
{
	char *dst_vaddr, *src_vaddr;

	dst_vaddr = kmap(dst_page);
	src_vaddr = kmap(src_page);

	memcpy(dst_vaddr + dst_offset, src_vaddr + src_offset, length);

	kunmap(src_page);
	kunmap(dst_page);
}

static inline void
slow_shmem_bit17_copy(struct page *gpu_page,
		      int gpu_offset,
		      struct page *cpu_page,
		      int cpu_offset,
		      int length,
		      int is_read)
{
	char *gpu_vaddr, *cpu_vaddr;

	/* Use the unswizzled path if this page isn't affected. */
	if ((page_to_phys(gpu_page) & (1 << 17)) == 0) {
		if (is_read)
			return slow_shmem_copy(cpu_page, cpu_offset,
					       gpu_page, gpu_offset, length);
		else
			return slow_shmem_copy(gpu_page, gpu_offset,
					       cpu_page, cpu_offset, length);
	}

	gpu_vaddr = kmap(gpu_page);
	cpu_vaddr = kmap(cpu_page);

	/* Copy the data, XORing A6 with A17 (1). The user already knows he's
	 * XORing with the other bits (A9 for Y, A9 and A10 for X)
	 */
	while (length > 0) {
		int cacheline_end = ALIGN(gpu_offset + 1, 64);
		int this_length = min(cacheline_end - gpu_offset, length);
		int swizzled_gpu_offset = gpu_offset ^ 64;

		if (is_read) {
			memcpy(cpu_vaddr + cpu_offset,
			       gpu_vaddr + swizzled_gpu_offset,
			       this_length);
		} else {
			memcpy(gpu_vaddr + swizzled_gpu_offset,
			       cpu_vaddr + cpu_offset,
			       this_length);
		}
		cpu_offset += this_length;
		gpu_offset += this_length;
		length -= this_length;
	}

	kunmap(cpu_page);
	kunmap(gpu_page);
}

/**
 * This is the fast shmem pread path, which attempts to copy_from_user directly
 * from the backing pages of the object to the user's address space.  On a
 * fault, it fails so we can fall back to i915_gem_shmem_pwrite_slow().
 */
static int
i915_gem_shmem_pread_fast(struct drm_device *dev, struct drm_gem_object *obj,
			  struct drm_i915_gem_pread *args,
			  struct drm_file *file_priv)
{
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
	ssize_t remain;
	loff_t offset, page_base;
	char __user *user_data;
	int page_offset, page_length;
	int ret;

	user_data = (char __user *) (uintptr_t) args->data_ptr;
	remain = args->size;

	mutex_lock(&dev->struct_mutex);

	ret = i915_gem_object_get_pages(obj, 0);
	if (ret != 0)
		goto fail_unlock;

	ret = i915_gem_object_set_cpu_read_domain_range(obj, args->offset,
							args->size);
	if (ret != 0)
		goto fail_put_pages;

	obj_priv = to_intel_bo(obj);
	offset = args->offset;

	while (remain > 0) {
		/* Operation in this page
		 *
		 * page_base = page offset within aperture
		 * page_offset = offset within page
		 * page_length = bytes to copy for this page
		 */
		page_base = (offset & ~(PAGE_SIZE-1));
		page_offset = offset & (PAGE_SIZE-1);
		page_length = remain;
		if ((page_offset + remain) > PAGE_SIZE)
			page_length = PAGE_SIZE - page_offset;

		ret = fast_shmem_read(obj_priv->pages,
				      page_base, page_offset,
				      user_data, page_length);
		if (ret)
			goto fail_put_pages;

		remain -= page_length;
		user_data += page_length;
		offset += page_length;
	}

fail_put_pages:
	i915_gem_object_put_pages(obj);
fail_unlock:
	mutex_unlock(&dev->struct_mutex);

	return ret;
}

static int
i915_gem_object_get_pages_or_evict(struct drm_gem_object *obj)
{
	int ret;

	ret = i915_gem_object_get_pages(obj, __GFP_NORETRY | __GFP_NOWARN);

	/* If we've insufficient memory to map in the pages, attempt
	 * to make some space by throwing out some old buffers.
	 */
	if (ret == -ENOMEM) {
		struct drm_device *dev = obj->dev;

		ret = i915_gem_evict_something(dev, obj->size,
					       i915_gem_get_gtt_alignment(obj));
		if (ret)
			return ret;

		ret = i915_gem_object_get_pages(obj, 0);
	}

	return ret;
}

/**
 * This is the fallback shmem pread path, which allocates temporary storage
 * in kernel space to copy_to_user into outside of the struct_mutex, so we
 * can copy out of the object's backing pages while holding the struct mutex
 * and not take page faults.
 */
static int
i915_gem_shmem_pread_slow(struct drm_device *dev, struct drm_gem_object *obj,
			  struct drm_i915_gem_pread *args,
			  struct drm_file *file_priv)
{
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
	struct mm_struct *mm = current->mm;
	struct page **user_pages;
	ssize_t remain;
	loff_t offset, pinned_pages, i;
	loff_t first_data_page, last_data_page, num_pages;
	int shmem_page_index, shmem_page_offset;
	int data_page_index,  data_page_offset;
	int page_length;
	int ret;
	uint64_t data_ptr = args->data_ptr;
	int do_bit17_swizzling;

	remain = args->size;

	/* Pin the user pages containing the data.  We can't fault while
	 * holding the struct mutex, yet we want to hold it while
	 * dereferencing the user data.
	 */
	first_data_page = data_ptr / PAGE_SIZE;
	last_data_page = (data_ptr + args->size - 1) / PAGE_SIZE;
	num_pages = last_data_page - first_data_page + 1;

	user_pages = drm_calloc_large(num_pages, sizeof(struct page *));
	if (user_pages == NULL)
		return -ENOMEM;

	down_read(&mm->mmap_sem);
	pinned_pages = get_user_pages(current, mm, (uintptr_t)args->data_ptr,
				      num_pages, 1, 0, user_pages, NULL);
	up_read(&mm->mmap_sem);
	if (pinned_pages < num_pages) {
		ret = -EFAULT;
		goto fail_put_user_pages;
	}

	do_bit17_swizzling = i915_gem_object_needs_bit17_swizzle(obj);

	mutex_lock(&dev->struct_mutex);

	ret = i915_gem_object_get_pages_or_evict(obj);
	if (ret)
		goto fail_unlock;

	ret = i915_gem_object_set_cpu_read_domain_range(obj, args->offset,
							args->size);
	if (ret != 0)
		goto fail_put_pages;

	obj_priv = to_intel_bo(obj);
	offset = args->offset;

	while (remain > 0) {
		/* Operation in this page
		 *
		 * shmem_page_index = page number within shmem file
		 * shmem_page_offset = offset within page in shmem file
		 * data_page_index = page number in get_user_pages return
		 * data_page_offset = offset with data_page_index page.
		 * page_length = bytes to copy for this page
		 */
		shmem_page_index = offset / PAGE_SIZE;
		shmem_page_offset = offset & ~PAGE_MASK;
		data_page_index = data_ptr / PAGE_SIZE - first_data_page;
		data_page_offset = data_ptr & ~PAGE_MASK;

		page_length = remain;
		if ((shmem_page_offset + page_length) > PAGE_SIZE)
			page_length = PAGE_SIZE - shmem_page_offset;
		if ((data_page_offset + page_length) > PAGE_SIZE)
			page_length = PAGE_SIZE - data_page_offset;

		if (do_bit17_swizzling) {
			slow_shmem_bit17_copy(obj_priv->pages[shmem_page_index],
					      shmem_page_offset,
					      user_pages[data_page_index],
					      data_page_offset,
					      page_length,
					      1);
		} else {
			slow_shmem_copy(user_pages[data_page_index],
					data_page_offset,
					obj_priv->pages[shmem_page_index],
					shmem_page_offset,
					page_length);
		}

		remain -= page_length;
		data_ptr += page_length;
		offset += page_length;
	}

fail_put_pages:
	i915_gem_object_put_pages(obj);
fail_unlock:
	mutex_unlock(&dev->struct_mutex);
fail_put_user_pages:
	for (i = 0; i < pinned_pages; i++) {
		SetPageDirty(user_pages[i]);
		page_cache_release(user_pages[i]);
	}
	drm_free_large(user_pages);

	return ret;
}

/**
 * Reads data from the object referenced by handle.
 *
 * On error, the contents of *data are undefined.
 */
int
i915_gem_pread_ioctl(struct drm_device *dev, void *data,
		     struct drm_file *file_priv)
{
	struct drm_i915_gem_pread *args = data;
	struct drm_gem_object *obj;
	struct drm_i915_gem_object *obj_priv;
	int ret;

	obj = drm_gem_object_lookup(dev, file_priv, args->handle);
	if (obj == NULL)
		return -ENOENT;
	obj_priv = to_intel_bo(obj);

	/* Bounds check source.
	 *
	 * XXX: This could use review for overflow issues...
	 */
	if (args->offset > obj->size || args->size > obj->size ||
	    args->offset + args->size > obj->size) {
		drm_gem_object_unreference_unlocked(obj);
		return -EINVAL;
	}

	if (i915_gem_object_needs_bit17_swizzle(obj)) {
		ret = i915_gem_shmem_pread_slow(dev, obj, args, file_priv);
	} else {
		ret = i915_gem_shmem_pread_fast(dev, obj, args, file_priv);
		if (ret != 0)
			ret = i915_gem_shmem_pread_slow(dev, obj, args,
							file_priv);
	}

	drm_gem_object_unreference_unlocked(obj);

	return ret;
}

/* This is the fast write path which cannot handle
 * page faults in the source data
 */

static inline int
fast_user_write(struct io_mapping *mapping,
		loff_t page_base, int page_offset,
		char __user *user_data,
		int length)
{
	char *vaddr_atomic;
	unsigned long unwritten;

	vaddr_atomic = io_mapping_map_atomic_wc(mapping, page_base, KM_USER0);
	unwritten = __copy_from_user_inatomic_nocache(vaddr_atomic + page_offset,
						      user_data, length);
	io_mapping_unmap_atomic(vaddr_atomic, KM_USER0);
	if (unwritten)
		return -EFAULT;
	return 0;
}

/* Here's the write path which can sleep for
 * page faults
 */

static inline void
slow_kernel_write(struct io_mapping *mapping,
		  loff_t gtt_base, int gtt_offset,
		  struct page *user_page, int user_offset,
		  int length)
{
	char __iomem *dst_vaddr;
	char *src_vaddr;

	dst_vaddr = io_mapping_map_wc(mapping, gtt_base);
	src_vaddr = kmap(user_page);

	memcpy_toio(dst_vaddr + gtt_offset,
		    src_vaddr + user_offset,
		    length);

	kunmap(user_page);
	io_mapping_unmap(dst_vaddr);
}

static inline int
fast_shmem_write(struct page **pages,
		 loff_t page_base, int page_offset,
		 char __user *data,
		 int length)
{
	char __iomem *vaddr;
	unsigned long unwritten;

	vaddr = kmap_atomic(pages[page_base >> PAGE_SHIFT], KM_USER0);
	if (vaddr == NULL)
		return -ENOMEM;
	unwritten = __copy_from_user_inatomic(vaddr + page_offset, data, length);
	kunmap_atomic(vaddr, KM_USER0);

	if (unwritten)
		return -EFAULT;
	return 0;
}

/**
 * This is the fast pwrite path, where we copy the data directly from the
 * user into the GTT, uncached.
 */
static int
i915_gem_gtt_pwrite_fast(struct drm_device *dev, struct drm_gem_object *obj,
			 struct drm_i915_gem_pwrite *args,
			 struct drm_file *file_priv)
{
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
	drm_i915_private_t *dev_priv = dev->dev_private;
	ssize_t remain;
	loff_t offset, page_base;
	char __user *user_data;
	int page_offset, page_length;
	int ret;

	user_data = (char __user *) (uintptr_t) args->data_ptr;
	remain = args->size;
	if (!access_ok(VERIFY_READ, user_data, remain))
		return -EFAULT;


	mutex_lock(&dev->struct_mutex);
	ret = i915_gem_object_pin(obj, 0);
	if (ret) {
		mutex_unlock(&dev->struct_mutex);
		return ret;
	}
	ret = i915_gem_object_set_to_gtt_domain(obj, 1);
	if (ret)
		goto fail;

	obj_priv = to_intel_bo(obj);
	offset = obj_priv->gtt_offset + args->offset;

	while (remain > 0) {
		/* Operation in this page
		 *
		 * page_base = page offset within aperture
		 * page_offset = offset within page
		 * page_length = bytes to copy for this page
		 */
		page_base = (offset & ~(PAGE_SIZE-1));
		page_offset = offset & (PAGE_SIZE-1);
		page_length = remain;
		if ((page_offset + remain) > PAGE_SIZE)
			page_length = PAGE_SIZE - page_offset;

		ret = fast_user_write (dev_priv->mm.gtt_mapping, page_base,
				       page_offset, user_data, page_length);

		/* If we get a fault while copying data, then (presumably) our
		 * source page isn't available.  Return the error and we'll
		 * retry in the slow path.
		 */
		if (ret)
			goto fail;

		remain -= page_length;
		user_data += page_length;
		offset += page_length;
	}

fail:
	i915_gem_object_unpin(obj);
	mutex_unlock(&dev->struct_mutex);

	return ret;
}

/**
 * This is the fallback GTT pwrite path, which uses get_user_pages to pin
 * the memory and maps it using kmap_atomic for copying.
 *
 * This code resulted in x11perf -rgb10text consuming about 10% more CPU
 * than using i915_gem_gtt_pwrite_fast on a G45 (32-bit).
 */
static int
i915_gem_gtt_pwrite_slow(struct drm_device *dev, struct drm_gem_object *obj,
			 struct drm_i915_gem_pwrite *args,
			 struct drm_file *file_priv)
{
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
	drm_i915_private_t *dev_priv = dev->dev_private;
	ssize_t remain;
	loff_t gtt_page_base, offset;
	loff_t first_data_page, last_data_page, num_pages;
	loff_t pinned_pages, i;
	struct page **user_pages;
	struct mm_struct *mm = current->mm;
	int gtt_page_offset, data_page_offset, data_page_index, page_length;
	int ret;
	uint64_t data_ptr = args->data_ptr;

	remain = args->size;

	/* Pin the user pages containing the data.  We can't fault while
	 * holding the struct mutex, and all of the pwrite implementations
	 * want to hold it while dereferencing the user data.
	 */
	first_data_page = data_ptr / PAGE_SIZE;
	last_data_page = (data_ptr + args->size - 1) / PAGE_SIZE;
	num_pages = last_data_page - first_data_page + 1;

	user_pages = drm_calloc_large(num_pages, sizeof(struct page *));
	if (user_pages == NULL)
		return -ENOMEM;

	down_read(&mm->mmap_sem);
	pinned_pages = get_user_pages(current, mm, (uintptr_t)args->data_ptr,
				      num_pages, 0, 0, user_pages, NULL);
	up_read(&mm->mmap_sem);
	if (pinned_pages < num_pages) {
		ret = -EFAULT;
		goto out_unpin_pages;
	}

	mutex_lock(&dev->struct_mutex);
	ret = i915_gem_object_pin(obj, 0);
	if (ret)
		goto out_unlock;

	ret = i915_gem_object_set_to_gtt_domain(obj, 1);
	if (ret)
		goto out_unpin_object;

	obj_priv = to_intel_bo(obj);
	offset = obj_priv->gtt_offset + args->offset;

	while (remain > 0) {
		/* Operation in this page
		 *
		 * gtt_page_base = page offset within aperture
		 * gtt_page_offset = offset within page in aperture
		 * data_page_index = page number in get_user_pages return
		 * data_page_offset = offset with data_page_index page.
		 * page_length = bytes to copy for this page
		 */
		gtt_page_base = offset & PAGE_MASK;
		gtt_page_offset = offset & ~PAGE_MASK;
		data_page_index = data_ptr / PAGE_SIZE - first_data_page;
		data_page_offset = data_ptr & ~PAGE_MASK;

		page_length = remain;
		if ((gtt_page_offset + page_length) > PAGE_SIZE)
			page_length = PAGE_SIZE - gtt_page_offset;
		if ((data_page_offset + page_length) > PAGE_SIZE)
			page_length = PAGE_SIZE - data_page_offset;

		slow_kernel_write(dev_priv->mm.gtt_mapping,
				  gtt_page_base, gtt_page_offset,
				  user_pages[data_page_index],
				  data_page_offset,
				  page_length);

		remain -= page_length;
		offset += page_length;
		data_ptr += page_length;
	}

out_unpin_object:
	i915_gem_object_unpin(obj);
out_unlock:
	mutex_unlock(&dev->struct_mutex);
out_unpin_pages:
	for (i = 0; i < pinned_pages; i++)
		page_cache_release(user_pages[i]);
	drm_free_large(user_pages);

	return ret;
}

/**
 * This is the fast shmem pwrite path, which attempts to directly
 * copy_from_user into the kmapped pages backing the object.
 */
static int
i915_gem_shmem_pwrite_fast(struct drm_device *dev, struct drm_gem_object *obj,
			   struct drm_i915_gem_pwrite *args,
			   struct drm_file *file_priv)
{
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
	ssize_t remain;
	loff_t offset, page_base;
	char __user *user_data;
	int page_offset, page_length;
	int ret;

	user_data = (char __user *) (uintptr_t) args->data_ptr;
	remain = args->size;

	mutex_lock(&dev->struct_mutex);

	ret = i915_gem_object_get_pages(obj, 0);
	if (ret != 0)
		goto fail_unlock;

	ret = i915_gem_object_set_to_cpu_domain(obj, 1);
	if (ret != 0)
		goto fail_put_pages;

	obj_priv = to_intel_bo(obj);
	offset = args->offset;
	obj_priv->dirty = 1;

	while (remain > 0) {
		/* Operation in this page
		 *
		 * page_base = page offset within aperture
		 * page_offset = offset within page
		 * page_length = bytes to copy for this page
		 */
		page_base = (offset & ~(PAGE_SIZE-1));
		page_offset = offset & (PAGE_SIZE-1);
		page_length = remain;
		if ((page_offset + remain) > PAGE_SIZE)
			page_length = PAGE_SIZE - page_offset;

		ret = fast_shmem_write(obj_priv->pages,
				       page_base, page_offset,
				       user_data, page_length);
		if (ret)
			goto fail_put_pages;

		remain -= page_length;
		user_data += page_length;
		offset += page_length;
	}

fail_put_pages:
	i915_gem_object_put_pages(obj);
fail_unlock:
	mutex_unlock(&dev->struct_mutex);

	return ret;
}

/**
 * This is the fallback shmem pwrite path, which uses get_user_pages to pin
 * the memory and maps it using kmap_atomic for copying.
 *
 * This avoids taking mmap_sem for faulting on the user's address while the
 * struct_mutex is held.
 */
static int
i915_gem_shmem_pwrite_slow(struct drm_device *dev, struct drm_gem_object *obj,
			   struct drm_i915_gem_pwrite *args,
			   struct drm_file *file_priv)
{
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
	struct mm_struct *mm = current->mm;
	struct page **user_pages;
	ssize_t remain;
	loff_t offset, pinned_pages, i;
	loff_t first_data_page, last_data_page, num_pages;
	int shmem_page_index, shmem_page_offset;
	int data_page_index,  data_page_offset;
	int page_length;
	int ret;
	uint64_t data_ptr = args->data_ptr;
	int do_bit17_swizzling;

	remain = args->size;

	/* Pin the user pages containing the data.  We can't fault while
	 * holding the struct mutex, and all of the pwrite implementations
	 * want to hold it while dereferencing the user data.
	 */
	first_data_page = data_ptr / PAGE_SIZE;
	last_data_page = (data_ptr + args->size - 1) / PAGE_SIZE;
	num_pages = last_data_page - first_data_page + 1;

	user_pages = drm_calloc_large(num_pages, sizeof(struct page *));
	if (user_pages == NULL)
		return -ENOMEM;

	down_read(&mm->mmap_sem);
	pinned_pages = get_user_pages(current, mm, (uintptr_t)args->data_ptr,
				      num_pages, 0, 0, user_pages, NULL);
	up_read(&mm->mmap_sem);
	if (pinned_pages < num_pages) {
		ret = -EFAULT;
		goto fail_put_user_pages;
	}

	do_bit17_swizzling = i915_gem_object_needs_bit17_swizzle(obj);

	mutex_lock(&dev->struct_mutex);

	ret = i915_gem_object_get_pages_or_evict(obj);
	if (ret)
		goto fail_unlock;

	ret = i915_gem_object_set_to_cpu_domain(obj, 1);
	if (ret != 0)
		goto fail_put_pages;

	obj_priv = to_intel_bo(obj);
	offset = args->offset;
	obj_priv->dirty = 1;

	while (remain > 0) {
		/* Operation in this page
		 *
		 * shmem_page_index = page number within shmem file
		 * shmem_page_offset = offset within page in shmem file
		 * data_page_index = page number in get_user_pages return
		 * data_page_offset = offset with data_page_index page.
		 * page_length = bytes to copy for this page
		 */
		shmem_page_index = offset / PAGE_SIZE;
		shmem_page_offset = offset & ~PAGE_MASK;
		data_page_index = data_ptr / PAGE_SIZE - first_data_page;
		data_page_offset = data_ptr & ~PAGE_MASK;

		page_length = remain;
		if ((shmem_page_offset + page_length) > PAGE_SIZE)
			page_length = PAGE_SIZE - shmem_page_offset;
		if ((data_page_offset + page_length) > PAGE_SIZE)
			page_length = PAGE_SIZE - data_page_offset;

		if (do_bit17_swizzling) {
			slow_shmem_bit17_copy(obj_priv->pages[shmem_page_index],
					      shmem_page_offset,
					      user_pages[data_page_index],
					      data_page_offset,
					      page_length,
					      0);
		} else {
			slow_shmem_copy(obj_priv->pages[shmem_page_index],
					shmem_page_offset,
					user_pages[data_page_index],
					data_page_offset,
					page_length);
		}

		remain -= page_length;
		data_ptr += page_length;
		offset += page_length;
	}

fail_put_pages:
	i915_gem_object_put_pages(obj);
fail_unlock:
	mutex_unlock(&dev->struct_mutex);
fail_put_user_pages:
	for (i = 0; i < pinned_pages; i++)
		page_cache_release(user_pages[i]);
	drm_free_large(user_pages);

	return ret;
}

/**
 * Writes data to the object referenced by handle.
 *
 * On error, the contents of the buffer that were to be modified are undefined.
 */
int
i915_gem_pwrite_ioctl(struct drm_device *dev, void *data,
		      struct drm_file *file_priv)
{
	struct drm_i915_gem_pwrite *args = data;
	struct drm_gem_object *obj;
	struct drm_i915_gem_object *obj_priv;
	int ret = 0;

	obj = drm_gem_object_lookup(dev, file_priv, args->handle);
	if (obj == NULL)
		return -ENOENT;
	obj_priv = to_intel_bo(obj);

	/* Bounds check destination.
	 *
	 * XXX: This could use review for overflow issues...
	 */
	if (args->offset > obj->size || args->size > obj->size ||
	    args->offset + args->size > obj->size) {
		drm_gem_object_unreference_unlocked(obj);
		return -EINVAL;
	}

	/* We can only do the GTT pwrite on untiled buffers, as otherwise
	 * it would end up going through the fenced access, and we'll get
	 * different detiling behavior between reading and writing.
	 * pread/pwrite currently are reading and writing from the CPU
	 * perspective, requiring manual detiling by the client.
	 */
	if (obj_priv->phys_obj)
		ret = i915_gem_phys_pwrite(dev, obj, args, file_priv);
	else if (obj_priv->tiling_mode == I915_TILING_NONE &&
		 dev->gtt_total != 0 &&
		 obj->write_domain != I915_GEM_DOMAIN_CPU) {
		ret = i915_gem_gtt_pwrite_fast(dev, obj, args, file_priv);
		if (ret == -EFAULT) {
			ret = i915_gem_gtt_pwrite_slow(dev, obj, args,
						       file_priv);
		}
	} else if (i915_gem_object_needs_bit17_swizzle(obj)) {
		ret = i915_gem_shmem_pwrite_slow(dev, obj, args, file_priv);
	} else {
		ret = i915_gem_shmem_pwrite_fast(dev, obj, args, file_priv);
		if (ret == -EFAULT) {
			ret = i915_gem_shmem_pwrite_slow(dev, obj, args,
							 file_priv);
		}
	}

#if WATCH_PWRITE
	if (ret)
		DRM_INFO("pwrite failed %d\n", ret);
#endif

	drm_gem_object_unreference_unlocked(obj);

	return ret;
}

/**
 * Called when user space prepares to use an object with the CPU, either
 * through the mmap ioctl's mapping or a GTT mapping.
 */
int
i915_gem_set_domain_ioctl(struct drm_device *dev, void *data,
			  struct drm_file *file_priv)
{
	struct drm_i915_private *dev_priv = dev->dev_private;
	struct drm_i915_gem_set_domain *args = data;
	struct drm_gem_object *obj;
	struct drm_i915_gem_object *obj_priv;
	uint32_t read_domains = args->read_domains;
	uint32_t write_domain = args->write_domain;
	int ret;

	if (!(dev->driver->driver_features & DRIVER_GEM))
		return -ENODEV;

	/* Only handle setting domains to types used by the CPU. */
	if (write_domain & I915_GEM_GPU_DOMAINS)
		return -EINVAL;

	if (read_domains & I915_GEM_GPU_DOMAINS)
		return -EINVAL;

	/* Having something in the write domain implies it's in the read
	 * domain, and only that read domain.  Enforce that in the request.
	 */
	if (write_domain != 0 && read_domains != write_domain)
		return -EINVAL;

	obj = drm_gem_object_lookup(dev, file_priv, args->handle);
	if (obj == NULL)
		return -ENOENT;
	obj_priv = to_intel_bo(obj);

	mutex_lock(&dev->struct_mutex);

	intel_mark_busy(dev, obj);

#if WATCH_BUF
	DRM_INFO("set_domain_ioctl %p(%zd), %08x %08x\n",
		 obj, obj->size, read_domains, write_domain);
#endif
	if (read_domains & I915_GEM_DOMAIN_GTT) {
		ret = i915_gem_object_set_to_gtt_domain(obj, write_domain != 0);

		/* Update the LRU on the fence for the CPU access that's
		 * about to occur.
		 */
		if (obj_priv->fence_reg != I915_FENCE_REG_NONE) {
			struct drm_i915_fence_reg *reg =
				&dev_priv->fence_regs[obj_priv->fence_reg];
			list_move_tail(&reg->lru_list,
				       &dev_priv->mm.fence_list);
		}

		/* Silently promote "you're not bound, there was nothing to do"
		 * to success, since the client was just asking us to
		 * make sure everything was done.
		 */
		if (ret == -EINVAL)
			ret = 0;
	} else {
		ret = i915_gem_object_set_to_cpu_domain(obj, write_domain != 0);
	}

	/* Maintain LRU order of "inactive" objects */
	if (ret == 0 && i915_gem_object_is_inactive(obj_priv))
		list_move_tail(&obj_priv->list, &dev_priv->mm.inactive_list);

	drm_gem_object_unreference(obj);
	mutex_unlock(&dev->struct_mutex);
	return ret;
}

/**
 * Called when user space has done writes to this buffer
 */
int
i915_gem_sw_finish_ioctl(struct drm_device *dev, void *data,
		      struct drm_file *file_priv)
{
	struct drm_i915_gem_sw_finish *args = data;
	struct drm_gem_object *obj;
	struct drm_i915_gem_object *obj_priv;
	int ret = 0;

	if (!(dev->driver->driver_features & DRIVER_GEM))
		return -ENODEV;

	mutex_lock(&dev->struct_mutex);
	obj = drm_gem_object_lookup(dev, file_priv, args->handle);
	if (obj == NULL) {
		mutex_unlock(&dev->struct_mutex);
		return -ENOENT;
	}

#if WATCH_BUF
	DRM_INFO("%s: sw_finish %d (%p %zd)\n",
		 __func__, args->handle, obj, obj->size);
#endif
	obj_priv = to_intel_bo(obj);

	/* Pinned buffers may be scanout, so flush the cache */
	if (obj_priv->pin_count)
		i915_gem_object_flush_cpu_write_domain(obj);

	drm_gem_object_unreference(obj);
	mutex_unlock(&dev->struct_mutex);
	return ret;
}

/**
 * Maps the contents of an object, returning the address it is mapped
 * into.
 *
 * While the mapping holds a reference on the contents of the object, it doesn't
 * imply a ref on the object itself.
 */
int
i915_gem_mmap_ioctl(struct drm_device *dev, void *data,
		   struct drm_file *file_priv)
{
	struct drm_i915_gem_mmap *args = data;
	struct drm_gem_object *obj;
	loff_t offset;
	unsigned long addr;

	if (!(dev->driver->driver_features & DRIVER_GEM))
		return -ENODEV;

	obj = drm_gem_object_lookup(dev, file_priv, args->handle);
	if (obj == NULL)
		return -ENOENT;

	offset = args->offset;

	down_write(&current->mm->mmap_sem);
	addr = do_mmap(obj->filp, 0, args->size,
		       PROT_READ | PROT_WRITE, MAP_SHARED,
		       args->offset);
	up_write(&current->mm->mmap_sem);
	drm_gem_object_unreference_unlocked(obj);
	if (IS_ERR((void *)addr))
		return addr;

	args->addr_ptr = (uint64_t) addr;

	return 0;
}

/**
 * i915_gem_fault - fault a page into the GTT
 * vma: VMA in question
 * vmf: fault info
 *
 * The fault handler is set up by drm_gem_mmap() when a object is GTT mapped
 * from userspace.  The fault handler takes care of binding the object to
 * the GTT (if needed), allocating and programming a fence register (again,
 * only if needed based on whether the old reg is still valid or the object
 * is tiled) and inserting a new PTE into the faulting process.
 *
 * Note that the faulting process may involve evicting existing objects
 * from the GTT and/or fence registers to make room.  So performance may
 * suffer if the GTT working set is large or there are few fence registers
 * left.
 */
int i915_gem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
{
	struct drm_gem_object *obj = vma->vm_private_data;
	struct drm_device *dev = obj->dev;
	drm_i915_private_t *dev_priv = dev->dev_private;
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
	pgoff_t page_offset;
	unsigned long pfn;
	int ret = 0;
	bool write = !!(vmf->flags & FAULT_FLAG_WRITE);

	/* We don't use vmf->pgoff since that has the fake offset */
	page_offset = ((unsigned long)vmf->virtual_address - vma->vm_start) >>
		PAGE_SHIFT;

	/* Now bind it into the GTT if needed */
	mutex_lock(&dev->struct_mutex);
	if (!obj_priv->gtt_space) {
		ret = i915_gem_object_bind_to_gtt(obj, 0);
		if (ret)
			goto unlock;

		ret = i915_gem_object_set_to_gtt_domain(obj, write);
		if (ret)
			goto unlock;
	}

	/* Need a new fence register? */
	if (obj_priv->tiling_mode != I915_TILING_NONE) {
		ret = i915_gem_object_get_fence_reg(obj, true);
		if (ret)
			goto unlock;
	}

	if (i915_gem_object_is_inactive(obj_priv))
		list_move_tail(&obj_priv->list, &dev_priv->mm.inactive_list);

	pfn = ((dev->agp->base + obj_priv->gtt_offset) >> PAGE_SHIFT) +
		page_offset;

	/* Finally, remap it using the new GTT offset */
	ret = vm_insert_pfn(vma, (unsigned long)vmf->virtual_address, pfn);
unlock:
	mutex_unlock(&dev->struct_mutex);

	switch (ret) {
	case 0:
	case -ERESTARTSYS:
		return VM_FAULT_NOPAGE;
	case -ENOMEM:
	case -EAGAIN:
		return VM_FAULT_OOM;
	default:
		return VM_FAULT_SIGBUS;
	}
}

/**
 * i915_gem_create_mmap_offset - create a fake mmap offset for an object
 * @obj: obj in question
 *
 * GEM memory mapping works by handing back to userspace a fake mmap offset
 * it can use in a subsequent mmap(2) call.  The DRM core code then looks
 * up the object based on the offset and sets up the various memory mapping
 * structures.
 *
 * This routine allocates and attaches a fake offset for @obj.
 */
static int
i915_gem_create_mmap_offset(struct drm_gem_object *obj)
{
	struct drm_device *dev = obj->dev;
	struct drm_gem_mm *mm = dev->mm_private;
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
	struct drm_map_list *list;
	struct drm_local_map *map;
	int ret = 0;

	/* Set the object up for mmap'ing */
	list = &obj->map_list;
	list->map = kzalloc(sizeof(struct drm_map_list), GFP_KERNEL);
	if (!list->map)
		return -ENOMEM;

	map = list->map;
	map->type = _DRM_GEM;
	map->size = obj->size;
	map->handle = obj;

	/* Get a DRM GEM mmap offset allocated... */
	list->file_offset_node = drm_mm_search_free(&mm->offset_manager,
						    obj->size / PAGE_SIZE, 0, 0);
	if (!list->file_offset_node) {
		DRM_ERROR("failed to allocate offset for bo %d\n", obj->name);
		ret = -ENOMEM;
		goto out_free_list;
	}

	list->file_offset_node = drm_mm_get_block(list->file_offset_node,
						  obj->size / PAGE_SIZE, 0);
	if (!list->file_offset_node) {
		ret = -ENOMEM;
		goto out_free_list;
	}

	list->hash.key = list->file_offset_node->start;
	if (drm_ht_insert_item(&mm->offset_hash, &list->hash)) {
		DRM_ERROR("failed to add to map hash\n");
		ret = -ENOMEM;
		goto out_free_mm;
	}

	/* By now we should be all set, any drm_mmap request on the offset
	 * below will get to our mmap & fault handler */
	obj_priv->mmap_offset = ((uint64_t) list->hash.key) << PAGE_SHIFT;

	return 0;

out_free_mm:
	drm_mm_put_block(list->file_offset_node);
out_free_list:
	kfree(list->map);

	return ret;
}

/**
 * i915_gem_release_mmap - remove physical page mappings
 * @obj: obj in question
 *
 * Preserve the reservation of the mmapping with the DRM core code, but
 * relinquish ownership of the pages back to the system.
 *
 * It is vital that we remove the page mapping if we have mapped a tiled
 * object through the GTT and then lose the fence register due to
 * resource pressure. Similarly if the object has been moved out of the
 * aperture, than pages mapped into userspace must be revoked. Removing the
 * mapping will then trigger a page fault on the next user access, allowing
 * fixup by i915_gem_fault().
 */
void
i915_gem_release_mmap(struct drm_gem_object *obj)
{
	struct drm_device *dev = obj->dev;
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);

	if (dev->dev_mapping)
		unmap_mapping_range(dev->dev_mapping,
				    obj_priv->mmap_offset, obj->size, 1);
}

static void
i915_gem_free_mmap_offset(struct drm_gem_object *obj)
{
	struct drm_device *dev = obj->dev;
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
	struct drm_gem_mm *mm = dev->mm_private;
	struct drm_map_list *list;

	list = &obj->map_list;
	drm_ht_remove_item(&mm->offset_hash, &list->hash);

	if (list->file_offset_node) {
		drm_mm_put_block(list->file_offset_node);
		list->file_offset_node = NULL;
	}

	if (list->map) {
		kfree(list->map);
		list->map = NULL;
	}

	obj_priv->mmap_offset = 0;
}

/**
 * i915_gem_get_gtt_alignment - return required GTT alignment for an object
 * @obj: object to check
 *
 * Return the required GTT alignment for an object, taking into account
 * potential fence register mapping if needed.
 */
static uint32_t
i915_gem_get_gtt_alignment(struct drm_gem_object *obj)
{
	struct drm_device *dev = obj->dev;
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
	int start, i;

	/*
	 * Minimum alignment is 4k (GTT page size), but might be greater
	 * if a fence register is needed for the object.
	 */
	if (INTEL_INFO(dev)->gen >= 4 || obj_priv->tiling_mode == I915_TILING_NONE)
		return 4096;

	/*
	 * Previous chips need to be aligned to the size of the smallest
	 * fence register that can contain the object.
	 */
	if (INTEL_INFO(dev)->gen == 3)
		start = 1024*1024;
	else
		start = 512*1024;

	for (i = start; i < obj->size; i <<= 1)
		;

	return i;
}

/**
 * i915_gem_mmap_gtt_ioctl - prepare an object for GTT mmap'ing
 * @dev: DRM device
 * @data: GTT mapping ioctl data
 * @file_priv: GEM object info
 *
 * Simply returns the fake offset to userspace so it can mmap it.
 * The mmap call will end up in drm_gem_mmap(), which will set things
 * up so we can get faults in the handler above.
 *
 * The fault handler will take care of binding the object into the GTT
 * (since it may have been evicted to make room for something), allocating
 * a fence register, and mapping the appropriate aperture address into
 * userspace.
 */
int
i915_gem_mmap_gtt_ioctl(struct drm_device *dev, void *data,
			struct drm_file *file_priv)
{
	struct drm_i915_gem_mmap_gtt *args = data;
	struct drm_gem_object *obj;
	struct drm_i915_gem_object *obj_priv;
	int ret;

	if (!(dev->driver->driver_features & DRIVER_GEM))
		return -ENODEV;

	obj = drm_gem_object_lookup(dev, file_priv, args->handle);
	if (obj == NULL)
		return -ENOENT;

	mutex_lock(&dev->struct_mutex);

	obj_priv = to_intel_bo(obj);

	if (obj_priv->madv != I915_MADV_WILLNEED) {
		DRM_ERROR("Attempting to mmap a purgeable buffer\n");
		drm_gem_object_unreference(obj);
		mutex_unlock(&dev->struct_mutex);
		return -EINVAL;
	}


	if (!obj_priv->mmap_offset) {
		ret = i915_gem_create_mmap_offset(obj);
		if (ret) {
			drm_gem_object_unreference(obj);
			mutex_unlock(&dev->struct_mutex);
			return ret;
		}
	}

	args->offset = obj_priv->mmap_offset;

	/*
	 * Pull it into the GTT so that we have a page list (makes the
	 * initial fault faster and any subsequent flushing possible).
	 */
	if (!obj_priv->agp_mem) {
		ret = i915_gem_object_bind_to_gtt(obj, 0);
		if (ret) {
			drm_gem_object_unreference(obj);
			mutex_unlock(&dev->struct_mutex);
			return ret;
		}
	}

	drm_gem_object_unreference(obj);
	mutex_unlock(&dev->struct_mutex);

	return 0;
}

void
i915_gem_object_put_pages(struct drm_gem_object *obj)
{
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
	int page_count = obj->size / PAGE_SIZE;
	int i;

	BUG_ON(obj_priv->pages_refcount == 0);
	BUG_ON(obj_priv->madv == __I915_MADV_PURGED);

	if (--obj_priv->pages_refcount != 0)
		return;

	if (obj_priv->tiling_mode != I915_TILING_NONE)
		i915_gem_object_save_bit_17_swizzle(obj);

	if (obj_priv->madv == I915_MADV_DONTNEED)
		obj_priv->dirty = 0;

	for (i = 0; i < page_count; i++) {
		if (obj_priv->dirty)
			set_page_dirty(obj_priv->pages[i]);

		if (obj_priv->madv == I915_MADV_WILLNEED)
			mark_page_accessed(obj_priv->pages[i]);

		page_cache_release(obj_priv->pages[i]);
	}
	obj_priv->dirty = 0;

	drm_free_large(obj_priv->pages);
	obj_priv->pages = NULL;
}

static uint32_t
i915_gem_next_request_seqno(struct drm_device *dev,
			    struct intel_ring_buffer *ring)
{
	drm_i915_private_t *dev_priv = dev->dev_private;

	ring->outstanding_lazy_request = true;

	return dev_priv->next_seqno;
}

static void
i915_gem_object_move_to_active(struct drm_gem_object *obj,
			       struct intel_ring_buffer *ring)
{
	struct drm_device *dev = obj->dev;
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
	uint32_t seqno = i915_gem_next_request_seqno(dev, ring);

	BUG_ON(ring == NULL);
	obj_priv->ring = ring;

	/* Add a reference if we're newly entering the active list. */
	if (!obj_priv->active) {
		drm_gem_object_reference(obj);
		obj_priv->active = 1;
	}

	/* Move from whatever list we were on to the tail of execution. */
	list_move_tail(&obj_priv->list, &ring->active_list);
	obj_priv->last_rendering_seqno = seqno;
}

static void
i915_gem_object_move_to_flushing(struct drm_gem_object *obj)
{
	struct drm_device *dev = obj->dev;
	drm_i915_private_t *dev_priv = dev->dev_private;
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);

	BUG_ON(!obj_priv->active);
	list_move_tail(&obj_priv->list, &dev_priv->mm.flushing_list);
	obj_priv->last_rendering_seqno = 0;
}

/* Immediately discard the backing storage */
static void
i915_gem_object_truncate(struct drm_gem_object *obj)
{
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
	struct inode *inode;

	/* Our goal here is to return as much of the memory as
	 * is possible back to the system as we are called from OOM.
	 * To do this we must instruct the shmfs to drop all of its
	 * backing pages, *now*. Here we mirror the actions taken
	 * when by shmem_delete_inode() to release the backing store.
	 */
	inode = obj->filp->f_path.dentry->d_inode;
	truncate_inode_pages(inode->i_mapping, 0);
	if (inode->i_op->truncate_range)
		inode->i_op->truncate_range(inode, 0, (loff_t)-1);

	obj_priv->madv = __I915_MADV_PURGED;
}

static inline int
i915_gem_object_is_purgeable(struct drm_i915_gem_object *obj_priv)
{
	return obj_priv->madv == I915_MADV_DONTNEED;
}

static void
i915_gem_object_move_to_inactive(struct drm_gem_object *obj)
{
	struct drm_device *dev = obj->dev;
	drm_i915_private_t *dev_priv = dev->dev_private;
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);

	i915_verify_inactive(dev, __FILE__, __LINE__);
	if (obj_priv->pin_count != 0)
		list_move_tail(&obj_priv->list, &dev_priv->mm.pinned_list);
	else
		list_move_tail(&obj_priv->list, &dev_priv->mm.inactive_list);

	BUG_ON(!list_empty(&obj_priv->gpu_write_list));

	obj_priv->last_rendering_seqno = 0;
	obj_priv->ring = NULL;
	if (obj_priv->active) {
		obj_priv->active = 0;
		drm_gem_object_unreference(obj);
	}
	i915_verify_inactive(dev, __FILE__, __LINE__);
}

static void
i915_gem_process_flushing_list(struct drm_device *dev,
			       uint32_t flush_domains,
			       struct intel_ring_buffer *ring)
{
	drm_i915_private_t *dev_priv = dev->dev_private;
	struct drm_i915_gem_object *obj_priv, *next;

	list_for_each_entry_safe(obj_priv, next,
				 &dev_priv->mm.gpu_write_list,
				 gpu_write_list) {
		struct drm_gem_object *obj = &obj_priv->base;

		if (obj->write_domain & flush_domains &&
		    obj_priv->ring == ring) {
			uint32_t old_write_domain = obj->write_domain;

			obj->write_domain = 0;
			list_del_init(&obj_priv->gpu_write_list);
			i915_gem_object_move_to_active(obj, ring);

			/* update the fence lru list */
			if (obj_priv->fence_reg != I915_FENCE_REG_NONE) {
				struct drm_i915_fence_reg *reg =
					&dev_priv->fence_regs[obj_priv->fence_reg];
				list_move_tail(&reg->lru_list,
						&dev_priv->mm.fence_list);
			}

			trace_i915_gem_object_change_domain(obj,
							    obj->read_domains,
							    old_write_domain);
		}
	}
}

uint32_t
i915_add_request(struct drm_device *dev,
		 struct drm_file *file_priv,
		 struct drm_i915_gem_request *request,
		 struct intel_ring_buffer *ring)
{
	drm_i915_private_t *dev_priv = dev->dev_private;
	struct drm_i915_file_private *i915_file_priv = NULL;
	uint32_t seqno;
	int was_empty;

	if (file_priv != NULL)
		i915_file_priv = file_priv->driver_priv;

	if (request == NULL) {
		request = kzalloc(sizeof(*request), GFP_KERNEL);
		if (request == NULL)
			return 0;
	}

	seqno = ring->add_request(dev, ring, file_priv, 0);

	request->seqno = seqno;
	request->ring = ring;
	request->emitted_jiffies = jiffies;
	was_empty = list_empty(&ring->request_list);
	list_add_tail(&request->list, &ring->request_list);

	if (i915_file_priv) {
		list_add_tail(&request->client_list,
			      &i915_file_priv->mm.request_list);
	} else {
		INIT_LIST_HEAD(&request->client_list);
	}

	if (!dev_priv->mm.suspended) {
		mod_timer(&dev_priv->hangcheck_timer,
			  jiffies + msecs_to_jiffies(DRM_I915_HANGCHECK_PERIOD));
		if (was_empty)
			queue_delayed_work(dev_priv->wq,
					   &dev_priv->mm.retire_work, HZ);
	}
	return seqno;
}

/**
 * Command execution barrier
 *
 * Ensures that all commands in the ring are finished
 * before signalling the CPU
 */
static void
i915_retire_commands(struct drm_device *dev, struct intel_ring_buffer *ring)
{
	uint32_t flush_domains = 0;

	/* The sampler always gets flushed on i965 (sigh) */
	if (INTEL_INFO(dev)->gen >= 4)
		flush_domains |= I915_GEM_DOMAIN_SAMPLER;

	ring->flush(dev, ring,
			I915_GEM_DOMAIN_COMMAND, flush_domains);
}

/**
 * Returns true if seq1 is later than seq2.
 */
bool
i915_seqno_passed(uint32_t seq1, uint32_t seq2)
{
	return (int32_t)(seq1 - seq2) >= 0;
}

uint32_t
i915_get_gem_seqno(struct drm_device *dev,
		   struct intel_ring_buffer *ring)
{
	return ring->get_gem_seqno(dev, ring);
}

void i915_gem_reset_flushing_list(struct drm_device *dev)
{
	struct drm_i915_private *dev_priv = dev->dev_private;

	while (!list_empty(&dev_priv->mm.flushing_list)) {
		struct drm_i915_gem_object *obj_priv;

		obj_priv = list_first_entry(&dev_priv->mm.flushing_list,
					    struct drm_i915_gem_object,
					    list);

		obj_priv->base.write_domain = 0;
		i915_gem_object_move_to_inactive(&obj_priv->base);
	}
}

void i915_gem_reset_inactive_gpu_domains(struct drm_device *dev)
{
	struct drm_i915_private *dev_priv = dev->dev_private;
	struct drm_i915_gem_object *obj_priv;

	list_for_each_entry(obj_priv,
			    &dev_priv->mm.inactive_list,
			    list)
	{
		obj_priv->base.read_domains &= ~I915_GEM_GPU_DOMAINS;
	}
}

/**
 * This function clears the request list as sequence numbers are passed.
 */
static void
i915_gem_retire_requests_ring(struct drm_device *dev,
			      struct intel_ring_buffer *ring)
{
	drm_i915_private_t *dev_priv = dev->dev_private;
	uint32_t seqno;
	bool wedged;

	if (!ring->status_page.page_addr ||
	    list_empty(&ring->request_list))
		return;

	seqno = i915_get_gem_seqno(dev, ring);
	wedged = atomic_read(&dev_priv->mm.wedged);

	while (!list_empty(&ring->request_list)) {
		struct drm_i915_gem_request *request;

		request = list_first_entry(&ring->request_list,
					   struct drm_i915_gem_request,
					   list);

		if (!wedged && !i915_seqno_passed(seqno, request->seqno))
			break;

		trace_i915_gem_request_retire(dev, request->seqno);

		list_del(&request->list);
		list_del(&request->client_list);
		kfree(request);
	}

	/* Move any buffers on the active list that are no longer referenced
	 * by the ringbuffer to the flushing/inactive lists as appropriate.
	 */
	while (!list_empty(&ring->active_list)) {
		struct drm_gem_object *obj;
		struct drm_i915_gem_object *obj_priv;

		obj_priv = list_first_entry(&ring->active_list,
					    struct drm_i915_gem_object,
					    list);

		if (!wedged &&
		    !i915_seqno_passed(seqno, obj_priv->last_rendering_seqno))
			break;

		obj = &obj_priv->base;

#if WATCH_LRU
		DRM_INFO("%s: retire %d moves to inactive list %p\n",
			 __func__, request->seqno, obj);
#endif

		if (obj->write_domain != 0)
			i915_gem_object_move_to_flushing(obj);
		else
			i915_gem_object_move_to_inactive(obj);
	}

	if (unlikely (dev_priv->trace_irq_seqno &&
		      i915_seqno_passed(dev_priv->trace_irq_seqno, seqno))) {
		ring->user_irq_put(dev, ring);
		dev_priv->trace_irq_seqno = 0;
	}
}

void
i915_gem_retire_requests(struct drm_device *dev)
{
	drm_i915_private_t *dev_priv = dev->dev_private;

	if (!list_empty(&dev_priv->mm.deferred_free_list)) {
	    struct drm_i915_gem_object *obj_priv, *tmp;

	    /* We must be careful that during unbind() we do not
	     * accidentally infinitely recurse into retire requests.
	     * Currently:
	     *   retire -> free -> unbind -> wait -> retire_ring
	     */
	    list_for_each_entry_safe(obj_priv, tmp,
				     &dev_priv->mm.deferred_free_list,
				     list)
		    i915_gem_free_object_tail(&obj_priv->base);
	}

	i915_gem_retire_requests_ring(dev, &dev_priv->render_ring);
	if (HAS_BSD(dev))
		i915_gem_retire_requests_ring(dev, &dev_priv->bsd_ring);
}

static void
i915_gem_retire_work_handler(struct work_struct *work)
{
	drm_i915_private_t *dev_priv;
	struct drm_device *dev;

	dev_priv = container_of(work, drm_i915_private_t,
				mm.retire_work.work);
	dev = dev_priv->dev;

	mutex_lock(&dev->struct_mutex);
	i915_gem_retire_requests(dev);

	if (!dev_priv->mm.suspended &&
		(!list_empty(&dev_priv->render_ring.request_list) ||
			(HAS_BSD(dev) &&
			 !list_empty(&dev_priv->bsd_ring.request_list))))
		queue_delayed_work(dev_priv->wq, &dev_priv->mm.retire_work, HZ);
	mutex_unlock(&dev->struct_mutex);
}

int
i915_do_wait_request(struct drm_device *dev, uint32_t seqno,
		     bool interruptible, struct intel_ring_buffer *ring)
{
	drm_i915_private_t *dev_priv = dev->dev_private;
	u32 ier;
	int ret = 0;

	BUG_ON(seqno == 0);

	if (seqno == dev_priv->next_seqno) {
		seqno = i915_add_request(dev, NULL, NULL, ring);
		if (seqno == 0)
			return -ENOMEM;
	}

	if (atomic_read(&dev_priv->mm.wedged))
		return -EIO;

	if (!i915_seqno_passed(ring->get_gem_seqno(dev, ring), seqno)) {
		if (HAS_PCH_SPLIT(dev))
			ier = I915_READ(DEIER) | I915_READ(GTIER);
		else
			ier = I915_READ(IER);
		if (!ier) {
			DRM_ERROR("something (likely vbetool) disabled "
				  "interrupts, re-enabling\n");
			i915_driver_irq_preinstall(dev);
			i915_driver_irq_postinstall(dev);
		}

		trace_i915_gem_request_wait_begin(dev, seqno);

		ring->waiting_gem_seqno = seqno;
		ring->user_irq_get(dev, ring);
		if (interruptible)
			ret = wait_event_interruptible(ring->irq_queue,
				i915_seqno_passed(
					ring->get_gem_seqno(dev, ring), seqno)
				|| atomic_read(&dev_priv->mm.wedged));
		else
			wait_event(ring->irq_queue,
				i915_seqno_passed(
					ring->get_gem_seqno(dev, ring), seqno)
				|| atomic_read(&dev_priv->mm.wedged));

		ring->user_irq_put(dev, ring);
		ring->waiting_gem_seqno = 0;

		trace_i915_gem_request_wait_end(dev, seqno);
	}
	if (atomic_read(&dev_priv->mm.wedged))
		ret = -EIO;

	if (ret && ret != -ERESTARTSYS)
		DRM_ERROR("%s returns %d (awaiting %d at %d, next %d)\n",
			  __func__, ret, seqno, ring->get_gem_seqno(dev, ring),
			  dev_priv->next_seqno);

	/* Directly dispatch request retiring.  While we have the work queue
	 * to handle this, the waiter on a request often wants an associated
	 * buffer to have made it to the inactive list, and we would need
	 * a separate wait queue to handle that.
	 */
	if (ret == 0)
		i915_gem_retire_requests_ring(dev, ring);

	return ret;
}

/**
 * Waits for a sequence number to be signaled, and cleans up the
 * request and object lists appropriately for that event.
 */
static int
i915_wait_request(struct drm_device *dev, uint32_t seqno,
		struct intel_ring_buffer *ring)
{
	return i915_do_wait_request(dev, seqno, 1, ring);
}

void
i915_gem_flush_ring(struct drm_device *dev,
		    struct drm_file *file_priv,
		    struct intel_ring_buffer *ring,
		    uint32_t invalidate_domains,
		    uint32_t flush_domains)
{
	ring->flush(dev, ring, invalidate_domains, flush_domains);
	i915_gem_process_flushing_list(dev, flush_domains, ring);

	if (ring->outstanding_lazy_request) {
		(void)i915_add_request(dev, file_priv, NULL, ring);
		ring->outstanding_lazy_request = false;
	}
}

static void
i915_gem_flush(struct drm_device *dev,
	       struct drm_file *file_priv,
	       uint32_t invalidate_domains,
	       uint32_t flush_domains,
	       uint32_t flush_rings)
{
	drm_i915_private_t *dev_priv = dev->dev_private;

	if (flush_domains & I915_GEM_DOMAIN_CPU)
		drm_agp_chipset_flush(dev);

	if ((flush_domains | invalidate_domains) & I915_GEM_GPU_DOMAINS) {
		if (flush_rings & RING_RENDER)
			i915_gem_flush_ring(dev, file_priv,
					    &dev_priv->render_ring,
					    invalidate_domains, flush_domains);
		if (flush_rings & RING_BSD)
			i915_gem_flush_ring(dev, file_priv,
					    &dev_priv->bsd_ring,
					    invalidate_domains, flush_domains);
	}
}

/**
 * Ensures that all rendering to the object has completed and the object is
 * safe to unbind from the GTT or access from the CPU.
 */
static int
i915_gem_object_wait_rendering(struct drm_gem_object *obj,
			       bool interruptible)
{
	struct drm_device *dev = obj->dev;
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
	int ret;

	/* This function only exists to support waiting for existing rendering,
	 * not for emitting required flushes.
	 */
	BUG_ON((obj->write_domain & I915_GEM_GPU_DOMAINS) != 0);

	/* If there is rendering queued on the buffer being evicted, wait for
	 * it.
	 */
	if (obj_priv->active) {
#if WATCH_BUF
		DRM_INFO("%s: object %p wait for seqno %08x\n",
			  __func__, obj, obj_priv->last_rendering_seqno);
#endif
		ret = i915_do_wait_request(dev,
					   obj_priv->last_rendering_seqno,
					   interruptible,
					   obj_priv->ring);
		if (ret)
			return ret;
	}

	return 0;
}

/**
 * Unbinds an object from the GTT aperture.
 */
int
i915_gem_object_unbind(struct drm_gem_object *obj)
{
	struct drm_device *dev = obj->dev;
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
	int ret = 0;

#if WATCH_BUF
	DRM_INFO("%s:%d %p\n", __func__, __LINE__, obj);
	DRM_INFO("gtt_space %p\n", obj_priv->gtt_space);
#endif
	if (obj_priv->gtt_space == NULL)
		return 0;

	if (obj_priv->pin_count != 0) {
		DRM_ERROR("Attempting to unbind pinned buffer\n");
		return -EINVAL;
	}

	/* blow away mappings if mapped through GTT */
	i915_gem_release_mmap(obj);

	/* Move the object to the CPU domain to ensure that
	 * any possible CPU writes while it's not in the GTT
	 * are flushed when we go to remap it. This will
	 * also ensure that all pending GPU writes are finished
	 * before we unbind.
	 */
	ret = i915_gem_object_set_to_cpu_domain(obj, 1);
	if (ret == -ERESTARTSYS)
		return ret;
	/* Continue on if we fail due to EIO, the GPU is hung so we
	 * should be safe and we need to cleanup or else we might
	 * cause memory corruption through use-after-free.
	 */

	/* release the fence reg _after_ flushing */
	if (obj_priv->fence_reg != I915_FENCE_REG_NONE)
		i915_gem_clear_fence_reg(obj);

	if (obj_priv->agp_mem != NULL) {
		drm_unbind_agp(obj_priv->agp_mem);
		drm_free_agp(obj_priv->agp_mem, obj->size / PAGE_SIZE);
		obj_priv->agp_mem = NULL;
	}

	i915_gem_object_put_pages(obj);
	BUG_ON(obj_priv->pages_refcount);

	if (obj_priv->gtt_space) {
		atomic_dec(&dev->gtt_count);
		atomic_sub(obj->size, &dev->gtt_memory);

		drm_mm_put_block(obj_priv->gtt_space);
		obj_priv->gtt_space = NULL;
	}

	list_del_init(&obj_priv->list);

	if (i915_gem_object_is_purgeable(obj_priv))
		i915_gem_object_truncate(obj);

	trace_i915_gem_object_unbind(obj);

	return ret;
}

int
i915_gpu_idle(struct drm_device *dev)
{
	drm_i915_private_t *dev_priv = dev->dev_private;
	bool lists_empty;
	u32 seqno;
	int ret;

	lists_empty = (list_empty(&dev_priv->mm.flushing_list) &&
		       list_empty(&dev_priv->render_ring.active_list) &&
		       (!HAS_BSD(dev) ||
			list_empty(&dev_priv->bsd_ring.active_list)));
	if (lists_empty)
		return 0;

	/* Flush everything onto the inactive list. */
	seqno = i915_gem_next_request_seqno(dev, &dev_priv->render_ring);
	i915_gem_flush_ring(dev, NULL, &dev_priv->render_ring,
			    I915_GEM_GPU_DOMAINS, I915_GEM_GPU_DOMAINS);
	ret = i915_wait_request(dev, seqno, &dev_priv->render_ring);
	if (ret)
		return ret;

	if (HAS_BSD(dev)) {
		seqno = i915_gem_next_request_seqno(dev, &dev_priv->render_ring);
		i915_gem_flush_ring(dev, NULL, &dev_priv->bsd_ring,
				    I915_GEM_GPU_DOMAINS, I915_GEM_GPU_DOMAINS);
		ret = i915_wait_request(dev, seqno, &dev_priv->bsd_ring);
		if (ret)
			return ret;
	}

	return 0;
}

int
i915_gem_object_get_pages(struct drm_gem_object *obj,
			  gfp_t gfpmask)
{
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
	int page_count, i;
	struct address_space *mapping;
	struct inode *inode;
	struct page *page;

	BUG_ON(obj_priv->pages_refcount
			== DRM_I915_GEM_OBJECT_MAX_PAGES_REFCOUNT);

	if (obj_priv->pages_refcount++ != 0)
		return 0;

	/* Get the list of pages out of our struct file.  They'll be pinned
	 * at this point until we release them.
	 */
	page_count = obj->size / PAGE_SIZE;
	BUG_ON(obj_priv->pages != NULL);
	obj_priv->pages = drm_calloc_large(page_count, sizeof(struct page *));
	if (obj_priv->pages == NULL) {
		obj_priv->pages_refcount--;
		return -ENOMEM;
	}

	inode = obj->filp->f_path.dentry->d_inode;
	mapping = inode->i_mapping;
	for (i = 0; i < page_count; i++) {
		page = read_cache_page_gfp(mapping, i,
					   GFP_HIGHUSER |
					   __GFP_COLD |
					   __GFP_RECLAIMABLE |
					   gfpmask);
		if (IS_ERR(page))
			goto err_pages;

		obj_priv->pages[i] = page;
	}

	if (obj_priv->tiling_mode != I915_TILING_NONE)
		i915_gem_object_do_bit_17_swizzle(obj);

	return 0;

err_pages:
	while (i--)
		page_cache_release(obj_priv->pages[i]);

	drm_free_large(obj_priv->pages);
	obj_priv->pages = NULL;
	obj_priv->pages_refcount--;
	return PTR_ERR(page);
}

static void sandybridge_write_fence_reg(struct drm_i915_fence_reg *reg)
{
	struct drm_gem_object *obj = reg->obj;
	struct drm_device *dev = obj->dev;
	drm_i915_private_t *dev_priv = dev->dev_private;
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
	int regnum = obj_priv->fence_reg;
	uint64_t val;

	val = (uint64_t)((obj_priv->gtt_offset + obj->size - 4096) &
		    0xfffff000) << 32;
	val |= obj_priv->gtt_offset & 0xfffff000;
	val |= (uint64_t)((obj_priv->stride / 128) - 1) <<
		SANDYBRIDGE_FENCE_PITCH_SHIFT;

	if (obj_priv->tiling_mode == I915_TILING_Y)
		val |= 1 << I965_FENCE_TILING_Y_SHIFT;
	val |= I965_FENCE_REG_VALID;

	I915_WRITE64(FENCE_REG_SANDYBRIDGE_0 + (regnum * 8), val);
}

static void i965_write_fence_reg(struct drm_i915_fence_reg *reg)
{
	struct drm_gem_object *obj = reg->obj;
	struct drm_device *dev = obj->dev;
	drm_i915_private_t *dev_priv = dev->dev_private;
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
	int regnum = obj_priv->fence_reg;
	uint64_t val;

	val = (uint64_t)((obj_priv->gtt_offset + obj->size - 4096) &
		    0xfffff000) << 32;
	val |= obj_priv->gtt_offset & 0xfffff000;
	val |= ((obj_priv->stride / 128) - 1) << I965_FENCE_PITCH_SHIFT;
	if (obj_priv->tiling_mode == I915_TILING_Y)
		val |= 1 << I965_FENCE_TILING_Y_SHIFT;
	val |= I965_FENCE_REG_VALID;

	I915_WRITE64(FENCE_REG_965_0 + (regnum * 8), val);
}

static void i915_write_fence_reg(struct drm_i915_fence_reg *reg)
{
	struct drm_gem_object *obj = reg->obj;
	struct drm_device *dev = obj->dev;
	drm_i915_private_t *dev_priv = dev->dev_private;
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
	int regnum = obj_priv->fence_reg;
	int tile_width;
	uint32_t fence_reg, val;
	uint32_t pitch_val;

	if ((obj_priv->gtt_offset & ~I915_FENCE_START_MASK) ||
	    (obj_priv->gtt_offset & (obj->size - 1))) {
		WARN(1, "%s: object 0x%08x not 1M or size (0x%zx) aligned\n",
		     __func__, obj_priv->gtt_offset, obj->size);
		return;
	}

	if (obj_priv->tiling_mode == I915_TILING_Y &&
	    HAS_128_BYTE_Y_TILING(dev))
		tile_width = 128;
	else
		tile_width = 512;

	/* Note: pitch better be a power of two tile widths */
	pitch_val = obj_priv->stride / tile_width;
	pitch_val = ffs(pitch_val) - 1;

	if (obj_priv->tiling_mode == I915_TILING_Y &&
	    HAS_128_BYTE_Y_TILING(dev))
		WARN_ON(pitch_val > I830_FENCE_MAX_PITCH_VAL);
	else
		WARN_ON(pitch_val > I915_FENCE_MAX_PITCH_VAL);

	val = obj_priv->gtt_offset;
	if (obj_priv->tiling_mode == I915_TILING_Y)
		val |= 1 << I830_FENCE_TILING_Y_SHIFT;
	val |= I915_FENCE_SIZE_BITS(obj->size);
	val |= pitch_val << I830_FENCE_PITCH_SHIFT;
	val |= I830_FENCE_REG_VALID;

	if (regnum < 8)
		fence_reg = FENCE_REG_830_0 + (regnum * 4);
	else
		fence_reg = FENCE_REG_945_8 + ((regnum - 8) * 4);
	I915_WRITE(fence_reg, val);
}

static void i830_write_fence_reg(struct drm_i915_fence_reg *reg)
{
	struct drm_gem_object *obj = reg->obj;
	struct drm_device *dev = obj->dev;
	drm_i915_private_t *dev_priv = dev->dev_private;
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
	int regnum = obj_priv->fence_reg;
	uint32_t val;
	uint32_t pitch_val;
	uint32_t fence_size_bits;

	if ((obj_priv->gtt_offset & ~I830_FENCE_START_MASK) ||
	    (obj_priv->gtt_offset & (obj->size - 1))) {
		WARN(1, "%s: object 0x%08x not 512K or size aligned\n",
		     __func__, obj_priv->gtt_offset);
		return;
	}

	pitch_val = obj_priv->stride / 128;
	pitch_val = ffs(pitch_val) - 1;
	WARN_ON(pitch_val > I830_FENCE_MAX_PITCH_VAL);

	val = obj_priv->gtt_offset;
	if (obj_priv->tiling_mode == I915_TILING_Y)
		val |= 1 << I830_FENCE_TILING_Y_SHIFT;
	fence_size_bits = I830_FENCE_SIZE_BITS(obj->size);
	WARN_ON(fence_size_bits & ~0x00000f00);
	val |= fence_size_bits;
	val |= pitch_val << I830_FENCE_PITCH_SHIFT;
	val |= I830_FENCE_REG_VALID;

	I915_WRITE(FENCE_REG_830_0 + (regnum * 4), val);
}

static int i915_find_fence_reg(struct drm_device *dev,
			       bool interruptible)
{
	struct drm_i915_fence_reg *reg = NULL;
	struct drm_i915_gem_object *obj_priv = NULL;
	struct drm_i915_private *dev_priv = dev->dev_private;
	struct drm_gem_object *obj = NULL;
	int i, avail, ret;

	/* First try to find a free reg */
	avail = 0;
	for (i = dev_priv->fence_reg_start; i < dev_priv->num_fence_regs; i++) {
		reg = &dev_priv->fence_regs[i];
		if (!reg->obj)
			return i;

		obj_priv = to_intel_bo(reg->obj);
		if (!obj_priv->pin_count)
		    avail++;
	}

	if (avail == 0)
		return -ENOSPC;

	/* None available, try to steal one or wait for a user to finish */
	i = I915_FENCE_REG_NONE;
	list_for_each_entry(reg, &dev_priv->mm.fence_list,
			    lru_list) {
		obj = reg->obj;
		obj_priv = to_intel_bo(obj);

		if (obj_priv->pin_count)
			continue;

		/* found one! */
		i = obj_priv->fence_reg;
		break;
	}

	BUG_ON(i == I915_FENCE_REG_NONE);

	/* We only have a reference on obj from the active list. put_fence_reg
	 * might drop that one, causing a use-after-free in it. So hold a
	 * private reference to obj like the other callers of put_fence_reg
	 * (set_tiling ioctl) do. */
	drm_gem_object_reference(obj);
	ret = i915_gem_object_put_fence_reg(obj, interruptible);
	drm_gem_object_unreference(obj);
	if (ret != 0)
		return ret;

	return i;
}

/**
 * i915_gem_object_get_fence_reg - set up a fence reg for an object
 * @obj: object to map through a fence reg
 *
 * When mapping objects through the GTT, userspace wants to be able to write
 * to them without having to worry about swizzling if the object is tiled.
 *
 * This function walks the fence regs looking for a free one for @obj,
 * stealing one if it can't find any.
 *
 * It then sets up the reg based on the object's properties: address, pitch
 * and tiling format.
 */
int
i915_gem_object_get_fence_reg(struct drm_gem_object *obj,
			      bool interruptible)
{
	struct drm_device *dev = obj->dev;
	struct drm_i915_private *dev_priv = dev->dev_private;
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
	struct drm_i915_fence_reg *reg = NULL;
	int ret;

	/* Just update our place in the LRU if our fence is getting used. */
	if (obj_priv->fence_reg != I915_FENCE_REG_NONE) {
		reg = &dev_priv->fence_regs[obj_priv->fence_reg];
		list_move_tail(&reg->lru_list, &dev_priv->mm.fence_list);
		return 0;
	}

	switch (obj_priv->tiling_mode) {
	case I915_TILING_NONE:
		WARN(1, "allocating a fence for non-tiled object?\n");
		break;
	case I915_TILING_X:
		if (!obj_priv->stride)
			return -EINVAL;
		WARN((obj_priv->stride & (512 - 1)),
		     "object 0x%08x is X tiled but has non-512B pitch\n",
		     obj_priv->gtt_offset);
		break;
	case I915_TILING_Y:
		if (!obj_priv->stride)
			return -EINVAL;
		WARN((obj_priv->stride & (128 - 1)),
		     "object 0x%08x is Y tiled but has non-128B pitch\n",
		     obj_priv->gtt_offset);
		break;
	}

	ret = i915_find_fence_reg(dev, interruptible);
	if (ret < 0)
		return ret;

	obj_priv->fence_reg = ret;
	reg = &dev_priv->fence_regs[obj_priv->fence_reg];
	list_add_tail(&reg->lru_list, &dev_priv->mm.fence_list);

	reg->obj = obj;

	switch (INTEL_INFO(dev)->gen) {
	case 6:
		sandybridge_write_fence_reg(reg);
		break;
	case 5:
	case 4:
		i965_write_fence_reg(reg);
		break;
	case 3:
		i915_write_fence_reg(reg);
		break;
	case 2:
		i830_write_fence_reg(reg);
		break;
	}

	trace_i915_gem_object_get_fence(obj, obj_priv->fence_reg,
			obj_priv->tiling_mode);

	return 0;
}

/**
 * i915_gem_clear_fence_reg - clear out fence register info
 * @obj: object to clear
 *
 * Zeroes out the fence register itself and clears out the associated
 * data structures in dev_priv and obj_priv.
 */
static void
i915_gem_clear_fence_reg(struct drm_gem_object *obj)
{
	struct drm_device *dev = obj->dev;
	drm_i915_private_t *dev_priv = dev->dev_private;
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
	struct drm_i915_fence_reg *reg =
		&dev_priv->fence_regs[obj_priv->fence_reg];
	uint32_t fence_reg;

	switch (INTEL_INFO(dev)->gen) {
	case 6:
		I915_WRITE64(FENCE_REG_SANDYBRIDGE_0 +
			     (obj_priv->fence_reg * 8), 0);
		break;
	case 5:
	case 4:
		I915_WRITE64(FENCE_REG_965_0 + (obj_priv->fence_reg * 8), 0);
		break;
	case 3:
		if (obj_priv->fence_reg > 8)
			fence_reg = FENCE_REG_945_8 + (obj_priv->fence_reg - 8) * 4;
		else
	case 2:
			fence_reg = FENCE_REG_830_0 + obj_priv->fence_reg * 4;

		I915_WRITE(fence_reg, 0);
		break;
	}

	reg->obj = NULL;
	obj_priv->fence_reg = I915_FENCE_REG_NONE;
	list_del_init(&reg->lru_list);
}

/**
 * i915_gem_object_put_fence_reg - waits on outstanding fenced access
 * to the buffer to finish, and then resets the fence register.
 * @obj: tiled object holding a fence register.
 * @bool: whether the wait upon the fence is interruptible
 *
 * Zeroes out the fence register itself and clears out the associated
 * data structures in dev_priv and obj_priv.
 */
int
i915_gem_object_put_fence_reg(struct drm_gem_object *obj,
			      bool interruptible)
{
	struct drm_device *dev = obj->dev;
	struct drm_i915_private *dev_priv = dev->dev_private;
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
	struct drm_i915_fence_reg *reg;

	if (obj_priv->fence_reg == I915_FENCE_REG_NONE)
		return 0;

	/* If we've changed tiling, GTT-mappings of the object
	 * need to re-fault to ensure that the correct fence register
	 * setup is in place.
	 */
	i915_gem_release_mmap(obj);

	/* On the i915, GPU access to tiled buffers is via a fence,
	 * therefore we must wait for any outstanding access to complete
	 * before clearing the fence.
	 */
	reg = &dev_priv->fence_regs[obj_priv->fence_reg];
	if (reg->gpu) {
		int ret;

		ret = i915_gem_object_flush_gpu_write_domain(obj, true);
		if (ret)
			return ret;

		ret = i915_gem_object_wait_rendering(obj, interruptible);
		if (ret)
			return ret;

		reg->gpu = false;
	}

	i915_gem_object_flush_gtt_write_domain(obj);
	i915_gem_clear_fence_reg(obj);

	return 0;
}

/**
 * Finds free space in the GTT aperture and binds the object there.
 */
static int
i915_gem_object_bind_to_gtt(struct drm_gem_object *obj, unsigned alignment)
{
	struct drm_device *dev = obj->dev;
	drm_i915_private_t *dev_priv = dev->dev_private;
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
	struct drm_mm_node *free_space;
	gfp_t gfpmask =  __GFP_NORETRY | __GFP_NOWARN;
	int ret;

	if (obj_priv->madv != I915_MADV_WILLNEED) {
		DRM_ERROR("Attempting to bind a purgeable object\n");
		return -EINVAL;
	}

	if (alignment == 0)
		alignment = i915_gem_get_gtt_alignment(obj);
	if (alignment & (i915_gem_get_gtt_alignment(obj) - 1)) {
		DRM_ERROR("Invalid object alignment requested %u\n", alignment);
		return -EINVAL;
	}

	/* If the object is bigger than the entire aperture, reject it early
	 * before evicting everything in a vain attempt to find space.
	 */
	if (obj->size > dev->gtt_total) {
		DRM_ERROR("Attempting to bind an object larger than the aperture\n");
		return -E2BIG;
	}

 search_free:
	free_space = drm_mm_search_free(&dev_priv->mm.gtt_space,
					obj->size, alignment, 0);
	if (free_space != NULL) {
		obj_priv->gtt_space = drm_mm_get_block(free_space, obj->size,
						       alignment);
		if (obj_priv->gtt_space != NULL)
			obj_priv->gtt_offset = obj_priv->gtt_space->start;
	}
	if (obj_priv->gtt_space == NULL) {
		/* If the gtt is empty and we're still having trouble
		 * fitting our object in, we're out of memory.
		 */
#if WATCH_LRU
		DRM_INFO("%s: GTT full, evicting something\n", __func__);
#endif
		ret = i915_gem_evict_something(dev, obj->size, alignment);
		if (ret)
			return ret;

		goto search_free;
	}

#if WATCH_BUF
	DRM_INFO("Binding object of size %zd at 0x%08x\n",
		 obj->size, obj_priv->gtt_offset);
#endif
	ret = i915_gem_object_get_pages(obj, gfpmask);
	if (ret) {
		drm_mm_put_block(obj_priv->gtt_space);
		obj_priv->gtt_space = NULL;

		if (ret == -ENOMEM) {
			/* first try to clear up some space from the GTT */
			ret = i915_gem_evict_something(dev, obj->size,
						       alignment);
			if (ret) {
				/* now try to shrink everyone else */
				if (gfpmask) {
					gfpmask = 0;
					goto search_free;
				}

				return ret;
			}

			goto search_free;
		}

		return ret;
	}

	/* Create an AGP memory structure pointing at our pages, and bind it
	 * into the GTT.
	 */
	obj_priv->agp_mem = drm_agp_bind_pages(dev,
					       obj_priv->pages,
					       obj->size >> PAGE_SHIFT,
					       obj_priv->gtt_offset,
					       obj_priv->agp_type);
	if (obj_priv->agp_mem == NULL) {
		i915_gem_object_put_pages(obj);
		drm_mm_put_block(obj_priv->gtt_space);
		obj_priv->gtt_space = NULL;

		ret = i915_gem_evict_something(dev, obj->size, alignment);
		if (ret)
			return ret;

		goto search_free;
	}
	atomic_inc(&dev->gtt_count);
	atomic_add(obj->size, &dev->gtt_memory);

	/* keep track of bounds object by adding it to the inactive list */
	list_add_tail(&obj_priv->list, &dev_priv->mm.inactive_list);

	/* Assert that the object is not currently in any GPU domain. As it
	 * wasn't in the GTT, there shouldn't be any way it could have been in
	 * a GPU cache
	 */
	BUG_ON(obj->read_domains & I915_GEM_GPU_DOMAINS);
	BUG_ON(obj->write_domain & I915_GEM_GPU_DOMAINS);

	trace_i915_gem_object_bind(obj, obj_priv->gtt_offset);

	return 0;
}

void
i915_gem_clflush_object(struct drm_gem_object *obj)
{
	struct drm_i915_gem_object	*obj_priv = to_intel_bo(obj);

	/* If we don't have a page list set up, then we're not pinned
	 * to GPU, and we can ignore the cache flush because it'll happen
	 * again at bind time.
	 */
	if (obj_priv->pages == NULL)
		return;

	trace_i915_gem_object_clflush(obj);

	drm_clflush_pages(obj_priv->pages, obj->size / PAGE_SIZE);
}

/** Flushes any GPU write domain for the object if it's dirty. */
static int
i915_gem_object_flush_gpu_write_domain(struct drm_gem_object *obj,
				       bool pipelined)
{
	struct drm_device *dev = obj->dev;
	uint32_t old_write_domain;

	if ((obj->write_domain & I915_GEM_GPU_DOMAINS) == 0)
		return 0;

	/* Queue the GPU write cache flushing we need. */
	old_write_domain = obj->write_domain;
	i915_gem_flush_ring(dev, NULL,
			    to_intel_bo(obj)->ring,
			    0, obj->write_domain);
	BUG_ON(obj->write_domain);

	trace_i915_gem_object_change_domain(obj,
					    obj->read_domains,
					    old_write_domain);

	if (pipelined)
		return 0;

	return i915_gem_object_wait_rendering(obj, true);
}

/** Flushes the GTT write domain for the object if it's dirty. */
static void
i915_gem_object_flush_gtt_write_domain(struct drm_gem_object *obj)
{
	uint32_t old_write_domain;

	if (obj->write_domain != I915_GEM_DOMAIN_GTT)
		return;

	/* No actual flushing is required for the GTT write domain.   Writes
	 * to it immediately go to main memory as far as we know, so there's
	 * no chipset flush.  It also doesn't land in render cache.
	 */
	old_write_domain = obj->write_domain;
	obj->write_domain = 0;

	trace_i915_gem_object_change_domain(obj,
					    obj->read_domains,
					    old_write_domain);
}

/** Flushes the CPU write domain for the object if it's dirty. */
static void
i915_gem_object_flush_cpu_write_domain(struct drm_gem_object *obj)
{
	struct drm_device *dev = obj->dev;
	uint32_t old_write_domain;

	if (obj->write_domain != I915_GEM_DOMAIN_CPU)
		return;

	i915_gem_clflush_object(obj);
	drm_agp_chipset_flush(dev);
	old_write_domain = obj->write_domain;
	obj->write_domain = 0;

	trace_i915_gem_object_change_domain(obj,
					    obj->read_domains,
					    old_write_domain);
}

/**
 * Moves a single object to the GTT read, and possibly write domain.
 *
 * This function returns when the move is complete, including waiting on
 * flushes to occur.
 */
int
i915_gem_object_set_to_gtt_domain(struct drm_gem_object *obj, int write)
{
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
	uint32_t old_write_domain, old_read_domains;
	int ret;

	/* Not valid to be called on unbound objects. */
	if (obj_priv->gtt_space == NULL)
		return -EINVAL;

	ret = i915_gem_object_flush_gpu_write_domain(obj, false);
	if (ret != 0)
		return ret;

	i915_gem_object_flush_cpu_write_domain(obj);

	if (write) {
		ret = i915_gem_object_wait_rendering(obj, true);
		if (ret)
			return ret;
	}

	old_write_domain = obj->write_domain;
	old_read_domains = obj->read_domains;

	/* It should now be out of any other write domains, and we can update
	 * the domain values for our changes.
	 */
	BUG_ON((obj->write_domain & ~I915_GEM_DOMAIN_GTT) != 0);
	obj->read_domains |= I915_GEM_DOMAIN_GTT;
	if (write) {
		obj->read_domains = I915_GEM_DOMAIN_GTT;
		obj->write_domain = I915_GEM_DOMAIN_GTT;
		obj_priv->dirty = 1;
	}

	trace_i915_gem_object_change_domain(obj,
					    old_read_domains,
					    old_write_domain);

	return 0;
}

/*
 * Prepare buffer for display plane. Use uninterruptible for possible flush
 * wait, as in modesetting process we're not supposed to be interrupted.
 */
int
i915_gem_object_set_to_display_plane(struct drm_gem_object *obj,
				     bool pipelined)
{
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
	uint32_t old_read_domains;
	int ret;

	/* Not valid to be called on unbound objects. */
	if (obj_priv->gtt_space == NULL)
		return -EINVAL;

	ret = i915_gem_object_flush_gpu_write_domain(obj, pipelined);
	if (ret)
		return ret;

	i915_gem_object_flush_cpu_write_domain(obj);

	old_read_domains = obj->read_domains;
	obj->read_domains |= I915_GEM_DOMAIN_GTT;

	trace_i915_gem_object_change_domain(obj,
					    old_read_domains,
					    obj->write_domain);

	return 0;
}

/**
 * Moves a single object to the CPU read, and possibly write domain.
 *
 * This function returns when the move is complete, including waiting on
 * flushes to occur.
 */
static int
i915_gem_object_set_to_cpu_domain(struct drm_gem_object *obj, int write)
{
	uint32_t old_write_domain, old_read_domains;
	int ret;

	ret = i915_gem_object_flush_gpu_write_domain(obj, false);
	if (ret != 0)
		return ret;

	i915_gem_object_flush_gtt_write_domain(obj);

	/* If we have a partially-valid cache of the object in the CPU,
	 * finish invalidating it and free the per-page flags.
	 */
	i915_gem_object_set_to_full_cpu_read_domain(obj);

	if (write) {
		ret = i915_gem_object_wait_rendering(obj, true);
		if (ret)
			return ret;
	}

	old_write_domain = obj->write_domain;
	old_read_domains = obj->read_domains;

	/* Flush the CPU cache if it's still invalid. */
	if ((obj->read_domains & I915_GEM_DOMAIN_CPU) == 0) {
		i915_gem_clflush_object(obj);

		obj->read_domains |= I915_GEM_DOMAIN_CPU;
	}

	/* It should now be out of any other write domains, and we can update
	 * the domain values for our changes.
	 */
	BUG_ON((obj->write_domain & ~I915_GEM_DOMAIN_CPU) != 0);

	/* If we're writing through the CPU, then the GPU read domains will
	 * need to be invalidated at next use.
	 */
	if (write) {
		obj->read_domains = I915_GEM_DOMAIN_CPU;
		obj->write_domain = I915_GEM_DOMAIN_CPU;
	}

	trace_i915_gem_object_change_domain(obj,
					    old_read_domains,
					    old_write_domain);

	return 0;
}

/*
 * Set the next domain for the specified object. This
 * may not actually perform the necessary flushing/invaliding though,
 * as that may want to be batched with other set_domain operations
 *
 * This is (we hope) the only really tricky part of gem. The goal
 * is fairly simple -- track which caches hold bits of the object
 * and make sure they remain coherent. A few concrete examples may
 * help to explain how it works. For shorthand, we use the notation
 * (read_domains, write_domain), e.g. (CPU, CPU) to indicate the
 * a pair of read and write domain masks.
 *
 * Case 1: the batch buffer
 *
 *	1. Allocated
 *	2. Written by CPU
 *	3. Mapped to GTT
 *	4. Read by GPU
 *	5. Unmapped from GTT
 *	6. Freed
 *
 *	Let's take these a step at a time
 *
 *	1. Allocated
 *		Pages allocated from the kernel may still have
 *		cache contents, so we set them to (CPU, CPU) always.
 *	2. Written by CPU (using pwrite)
 *		The pwrite function calls set_domain (CPU, CPU) and
 *		this function does nothing (as nothing changes)
 *	3. Mapped by GTT
 *		This function asserts that the object is not
 *		currently in any GPU-based read or write domains
 *	4. Read by GPU
 *		i915_gem_execbuffer calls set_domain (COMMAND, 0).
 *		As write_domain is zero, this function adds in the
 *		current read domains (CPU+COMMAND, 0).
 *		flush_domains is set to CPU.
 *		invalidate_domains is set to COMMAND
 *		clflush is run to get data out of the CPU caches
 *		then i915_dev_set_domain calls i915_gem_flush to
 *		emit an MI_FLUSH and drm_agp_chipset_flush
 *	5. Unmapped from GTT
 *		i915_gem_object_unbind calls set_domain (CPU, CPU)
 *		flush_domains and invalidate_domains end up both zero
 *		so no flushing/invalidating happens
 *	6. Freed
 *		yay, done
 *
 * Case 2: The shared render buffer
 *
 *	1. Allocated
 *	2. Mapped to GTT
 *	3. Read/written by GPU
 *	4. set_domain to (CPU,CPU)
 *	5. Read/written by CPU
 *	6. Read/written by GPU
 *
 *	1. Allocated
 *		Same as last example, (CPU, CPU)
 *	2. Mapped to GTT
 *		Nothing changes (assertions find that it is not in the GPU)
 *	3. Read/written by GPU
 *		execbuffer calls set_domain (RENDER, RENDER)
 *		flush_domains gets CPU
 *		invalidate_domains gets GPU
 *		clflush (obj)
 *		MI_FLUSH and drm_agp_chipset_flush
 *	4. set_domain (CPU, CPU)
 *		flush_domains gets GPU
 *		invalidate_domains gets CPU
 *		wait_rendering (obj) to make sure all drawing is complete.
 *		This will include an MI_FLUSH to get the data from GPU
 *		to memory
 *		clflush (obj) to invalidate the CPU cache
 *		Another MI_FLUSH in i915_gem_flush (eliminate this somehow?)
 *	5. Read/written by CPU
 *		cache lines are loaded and dirtied
 *	6. Read written by GPU
 *		Same as last GPU access
 *
 * Case 3: The constant buffer
 *
 *	1. Allocated
 *	2. Written by CPU
 *	3. Read by GPU
 *	4. Updated (written) by CPU again
 *	5. Read by GPU
 *
 *	1. Allocated
 *		(CPU, CPU)
 *	2. Written by CPU
 *		(CPU, CPU)
 *	3. Read by GPU
 *		(CPU+RENDER, 0)
 *		flush_domains = CPU
 *		invalidate_domains = RENDER
 *		clflush (obj)
 *		MI_FLUSH
 *		drm_agp_chipset_flush
 *	4. Updated (written) by CPU again
 *		(CPU, CPU)
 *		flush_domains = 0 (no previous write domain)
 *		invalidate_domains = 0 (no new read domains)
 *	5. Read by GPU
 *		(CPU+RENDER, 0)
 *		flush_domains = CPU
 *		invalidate_domains = RENDER
 *		clflush (obj)
 *		MI_FLUSH
 *		drm_agp_chipset_flush
 */
static void
i915_gem_object_set_to_gpu_domain(struct drm_gem_object *obj)
{
	struct drm_device		*dev = obj->dev;
	struct drm_i915_private		*dev_priv = dev->dev_private;
	struct drm_i915_gem_object	*obj_priv = to_intel_bo(obj);
	uint32_t			invalidate_domains = 0;
	uint32_t			flush_domains = 0;
	uint32_t			old_read_domains;

	BUG_ON(obj->pending_read_domains & I915_GEM_DOMAIN_CPU);
	BUG_ON(obj->pending_write_domain == I915_GEM_DOMAIN_CPU);

	intel_mark_busy(dev, obj);

#if WATCH_BUF
	DRM_INFO("%s: object %p read %08x -> %08x write %08x -> %08x\n",
		 __func__, obj,
		 obj->read_domains, obj->pending_read_domains,
		 obj->write_domain, obj->pending_write_domain);
#endif
	/*
	 * If the object isn't moving to a new write domain,
	 * let the object stay in multiple read domains
	 */
	if (obj->pending_write_domain == 0)
		obj->pending_read_domains |= obj->read_domains;
	else
		obj_priv->dirty = 1;

	/*
	 * Flush the current write domain if
	 * the new read domains don't match. Invalidate
	 * any read domains which differ from the old
	 * write domain
	 */
	if (obj->write_domain &&
	    obj->write_domain != obj->pending_read_domains) {
		flush_domains |= obj->write_domain;
		invalidate_domains |=
			obj->pending_read_domains & ~obj->write_domain;
	}
	/*
	 * Invalidate any read caches which may have
	 * stale data. That is, any new read domains.
	 */
	invalidate_domains |= obj->pending_read_domains & ~obj->read_domains;
	if ((flush_domains | invalidate_domains) & I915_GEM_DOMAIN_CPU) {
#if WATCH_BUF
		DRM_INFO("%s: CPU domain flush %08x invalidate %08x\n",
			 __func__, flush_domains, invalidate_domains);
#endif
		i915_gem_clflush_object(obj);
	}

	old_read_domains = obj->read_domains;

	/* The actual obj->write_domain will be updated with
	 * pending_write_domain after we emit the accumulated flush for all
	 * of our domain changes in execbuffers (which clears objects'
	 * write_domains).  So if we have a current write domain that we
	 * aren't changing, set pending_write_domain to that.
	 */
	if (flush_domains == 0 && obj->pending_write_domain == 0)
		obj->pending_write_domain = obj->write_domain;
	obj->read_domains = obj->pending_read_domains;

	dev->invalidate_domains |= invalidate_domains;
	dev->flush_domains |= flush_domains;
	if (obj_priv->ring)
		dev_priv->mm.flush_rings |= obj_priv->ring->id;
#if WATCH_BUF
	DRM_INFO("%s: read %08x write %08x invalidate %08x flush %08x\n",
		 __func__,
		 obj->read_domains, obj->write_domain,
		 dev->invalidate_domains, dev->flush_domains);
#endif

	trace_i915_gem_object_change_domain(obj,
					    old_read_domains,
					    obj->write_domain);
}

/**
 * Moves the object from a partially CPU read to a full one.
 *
 * Note that this only resolves i915_gem_object_set_cpu_read_domain_range(),
 * and doesn't handle transitioning from !(read_domains & I915_GEM_DOMAIN_CPU).
 */
static void
i915_gem_object_set_to_full_cpu_read_domain(struct drm_gem_object *obj)
{
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);

	if (!obj_priv->page_cpu_valid)
		return;

	/* If we're partially in the CPU read domain, finish moving it in.
	 */
	if (obj->read_domains & I915_GEM_DOMAIN_CPU) {
		int i;

		for (i = 0; i <= (obj->size - 1) / PAGE_SIZE; i++) {
			if (obj_priv->page_cpu_valid[i])
				continue;
			drm_clflush_pages(obj_priv->pages + i, 1);
		}
	}

	/* Free the page_cpu_valid mappings which are now stale, whether
	 * or not we've got I915_GEM_DOMAIN_CPU.
	 */
	kfree(obj_priv->page_cpu_valid);
	obj_priv->page_cpu_valid = NULL;
}

/**
 * Set the CPU read domain on a range of the object.
 *
 * The object ends up with I915_GEM_DOMAIN_CPU in its read flags although it's
 * not entirely valid.  The page_cpu_valid member of the object flags which
 * pages have been flushed, and will be respected by
 * i915_gem_object_set_to_cpu_domain() if it's called on to get a valid mapping
 * of the whole object.
 *
 * This function returns when the move is complete, including waiting on
 * flushes to occur.
 */
static int
i915_gem_object_set_cpu_read_domain_range(struct drm_gem_object *obj,
					  uint64_t offset, uint64_t size)
{
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
	uint32_t old_read_domains;
	int i, ret;

	if (offset == 0 && size == obj->size)
		return i915_gem_object_set_to_cpu_domain(obj, 0);

	ret = i915_gem_object_flush_gpu_write_domain(obj, false);
	if (ret != 0)
		return ret;
	i915_gem_object_flush_gtt_write_domain(obj);

	/* If we're already fully in the CPU read domain, we're done. */
	if (obj_priv->page_cpu_valid == NULL &&
	    (obj->read_domains & I915_GEM_DOMAIN_CPU) != 0)
		return 0;

	/* Otherwise, create/clear the per-page CPU read domain flag if we're
	 * newly adding I915_GEM_DOMAIN_CPU
	 */
	if (obj_priv->page_cpu_valid == NULL) {
		obj_priv->page_cpu_valid = kzalloc(obj->size / PAGE_SIZE,
						   GFP_KERNEL);
		if (obj_priv->page_cpu_valid == NULL)
			return -ENOMEM;
	} else if ((obj->read_domains & I915_GEM_DOMAIN_CPU) == 0)
		memset(obj_priv->page_cpu_valid, 0, obj->size / PAGE_SIZE);

	/* Flush the cache on any pages that are still invalid from the CPU's
	 * perspective.
	 */
	for (i = offset / PAGE_SIZE; i <= (offset + size - 1) / PAGE_SIZE;
	     i++) {
		if (obj_priv->page_cpu_valid[i])
			continue;

		drm_clflush_pages(obj_priv->pages + i, 1);

		obj_priv->page_cpu_valid[i] = 1;
	}

	/* It should now be out of any other write domains, and we can update
	 * the domain values for our changes.
	 */
	BUG_ON((obj->write_domain & ~I915_GEM_DOMAIN_CPU) != 0);

	old_read_domains = obj->read_domains;
	obj->read_domains |= I915_GEM_DOMAIN_CPU;

	trace_i915_gem_object_change_domain(obj,
					    old_read_domains,
					    obj->write_domain);

	return 0;
}

/**
 * Pin an object to the GTT and evaluate the relocations landing in it.
 */
static int
i915_gem_object_pin_and_relocate(struct drm_gem_object *obj,
				 struct drm_file *file_priv,
				 struct drm_i915_gem_exec_object2 *entry,
				 struct drm_i915_gem_relocation_entry *relocs)
{
	struct drm_device *dev = obj->dev;
	drm_i915_private_t *dev_priv = dev->dev_private;
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
	int i, ret;
	void __iomem *reloc_page;
	bool need_fence;

	need_fence = entry->flags & EXEC_OBJECT_NEEDS_FENCE &&
	             obj_priv->tiling_mode != I915_TILING_NONE;

	/* Check fence reg constraints and rebind if necessary */
	if (need_fence &&
	    !i915_gem_object_fence_offset_ok(obj,
					     obj_priv->tiling_mode)) {
		ret = i915_gem_object_unbind(obj);
		if (ret)
			return ret;
	}

	/* Choose the GTT offset for our buffer and put it there. */
	ret = i915_gem_object_pin(obj, (uint32_t) entry->alignment);
	if (ret)
		return ret;

	/*
	 * Pre-965 chips need a fence register set up in order to
	 * properly handle blits to/from tiled surfaces.
	 */
	if (need_fence) {
		ret = i915_gem_object_get_fence_reg(obj, true);
		if (ret != 0) {
			i915_gem_object_unpin(obj);
			return ret;
		}

		dev_priv->fence_regs[obj_priv->fence_reg].gpu = true;
	}

	entry->offset = obj_priv->gtt_offset;

	/* Apply the relocations, using the GTT aperture to avoid cache
	 * flushing requirements.
	 */
	for (i = 0; i < entry->relocation_count; i++) {
		struct drm_i915_gem_relocation_entry *reloc= &relocs[i];
		struct drm_gem_object *target_obj;
		struct drm_i915_gem_object *target_obj_priv;
		uint32_t reloc_val, reloc_offset;
		uint32_t __iomem *reloc_entry;

		target_obj = drm_gem_object_lookup(obj->dev, file_priv,
						   reloc->target_handle);
		if (target_obj == NULL) {
			i915_gem_object_unpin(obj);
			return -ENOENT;
		}
		target_obj_priv = to_intel_bo(target_obj);

#if WATCH_RELOC
		DRM_INFO("%s: obj %p offset %08x target %d "
			 "read %08x write %08x gtt %08x "
			 "presumed %08x delta %08x\n",
			 __func__,
			 obj,
			 (int) reloc->offset,
			 (int) reloc->target_handle,
			 (int) reloc->read_domains,
			 (int) reloc->write_domain,
			 (int) target_obj_priv->gtt_offset,
			 (int) reloc->presumed_offset,
			 reloc->delta);
#endif

		/* The target buffer should have appeared before us in the
		 * exec_object list, so it should have a GTT space bound by now.
		 */
		if (target_obj_priv->gtt_space == NULL) {
			DRM_ERROR("No GTT space found for object %d\n",
				  reloc->target_handle);
			drm_gem_object_unreference(target_obj);
			i915_gem_object_unpin(obj);
			return -EINVAL;
		}

		/* Validate that the target is in a valid r/w GPU domain */
		if (reloc->write_domain & (reloc->write_domain - 1)) {
			DRM_ERROR("reloc with multiple write domains: "
				  "obj %p target %d offset %d "
				  "read %08x write %08x",
				  obj, reloc->target_handle,
				  (int) reloc->offset,
				  reloc->read_domains,
				  reloc->write_domain);
			return -EINVAL;
		}
		if (reloc->write_domain & I915_GEM_DOMAIN_CPU ||
		    reloc->read_domains & I915_GEM_DOMAIN_CPU) {
			DRM_ERROR("reloc with read/write CPU domains: "
				  "obj %p target %d offset %d "
				  "read %08x write %08x",
				  obj, reloc->target_handle,
				  (int) reloc->offset,
				  reloc->read_domains,
				  reloc->write_domain);
			drm_gem_object_unreference(target_obj);
			i915_gem_object_unpin(obj);
			return -EINVAL;
		}
		if (reloc->write_domain && target_obj->pending_write_domain &&
		    reloc->write_domain != target_obj->pending_write_domain) {
			DRM_ERROR("Write domain conflict: "
				  "obj %p target %d offset %d "
				  "new %08x old %08x\n",
				  obj, reloc->target_handle,
				  (int) reloc->offset,
				  reloc->write_domain,
				  target_obj->pending_write_domain);
			drm_gem_object_unreference(target_obj);
			i915_gem_object_unpin(obj);
			return -EINVAL;
		}

		target_obj->pending_read_domains |= reloc->read_domains;
		target_obj->pending_write_domain |= reloc->write_domain;

		/* If the relocation already has the right value in it, no
		 * more work needs to be done.
		 */
		if (target_obj_priv->gtt_offset == reloc->presumed_offset) {
			drm_gem_object_unreference(target_obj);
			continue;
		}

		/* Check that the relocation address is valid... */
		if (reloc->offset > obj->size - 4) {
			DRM_ERROR("Relocation beyond object bounds: "
				  "obj %p target %d offset %d size %d.\n",
				  obj, reloc->target_handle,
				  (int) reloc->offset, (int) obj->size);
			drm_gem_object_unreference(target_obj);
			i915_gem_object_unpin(obj);
			return -EINVAL;
		}
		if (reloc->offset & 3) {
			DRM_ERROR("Relocation not 4-byte aligned: "
				  "obj %p target %d offset %d.\n",
				  obj, reloc->target_handle,
				  (int) reloc->offset);
			drm_gem_object_unreference(target_obj);
			i915_gem_object_unpin(obj);
			return -EINVAL;
		}

		/* and points to somewhere within the target object. */
		if (reloc->delta >= target_obj->size) {
			DRM_ERROR("Relocation beyond target object bounds: "
				  "obj %p target %d delta %d size %d.\n",
				  obj, reloc->target_handle,
				  (int) reloc->delta, (int) target_obj->size);
			drm_gem_object_unreference(target_obj);
			i915_gem_object_unpin(obj);
			return -EINVAL;
		}

		ret = i915_gem_object_set_to_gtt_domain(obj, 1);
		if (ret != 0) {
			drm_gem_object_unreference(target_obj);
			i915_gem_object_unpin(obj);
			return -EINVAL;
		}

		/* Map the page containing the relocation we're going to
		 * perform.
		 */
		reloc_offset = obj_priv->gtt_offset + reloc->offset;
		reloc_page = io_mapping_map_atomic_wc(dev_priv->mm.gtt_mapping,
						      (reloc_offset &
						       ~(PAGE_SIZE - 1)),
						      KM_USER0);
		reloc_entry = (uint32_t __iomem *)(reloc_page +
						   (reloc_offset & (PAGE_SIZE - 1)));
		reloc_val = target_obj_priv->gtt_offset + reloc->delta;

#if WATCH_BUF
		DRM_INFO("Applied relocation: %p@0x%08x %08x -> %08x\n",
			  obj, (unsigned int) reloc->offset,
			  readl(reloc_entry), reloc_val);
#endif
		writel(reloc_val, reloc_entry);
		io_mapping_unmap_atomic(reloc_page, KM_USER0);

		/* The updated presumed offset for this entry will be
		 * copied back out to the user.
		 */
		reloc->presumed_offset = target_obj_priv->gtt_offset;

		drm_gem_object_unreference(target_obj);
	}

#if WATCH_BUF
	if (0)
		i915_gem_dump_object(obj, 128, __func__, ~0);
#endif
	return 0;
}

/* Throttle our rendering by waiting until the ring has completed our requests
 * emitted over 20 msec ago.
 *
 * Note that if we were to use the current jiffies each time around the loop,
 * we wouldn't escape the function with any frames outstanding if the time to
 * render a frame was over 20ms.
 *
 * This should get us reasonable parallelism between CPU and GPU but also
 * relatively low latency when blocking on a particular request to finish.
 */
static int
i915_gem_ring_throttle(struct drm_device *dev, struct drm_file *file_priv)
{
	struct drm_i915_file_private *i915_file_priv = file_priv->driver_priv;
	int ret = 0;
	unsigned long recent_enough = jiffies - msecs_to_jiffies(20);

	mutex_lock(&dev->struct_mutex);
	while (!list_empty(&i915_file_priv->mm.request_list)) {
		struct drm_i915_gem_request *request;

		request = list_first_entry(&i915_file_priv->mm.request_list,
					   struct drm_i915_gem_request,
					   client_list);

		if (time_after_eq(request->emitted_jiffies, recent_enough))
			break;

		ret = i915_wait_request(dev, request->seqno, request->ring);
		if (ret != 0)
			break;
	}
	mutex_unlock(&dev->struct_mutex);

	return ret;
}

static int
i915_gem_get_relocs_from_user(struct drm_i915_gem_exec_object2 *exec_list,
			      uint32_t buffer_count,
			      struct drm_i915_gem_relocation_entry **relocs)
{
	uint32_t reloc_count = 0, reloc_index = 0, i;
	int ret;

	*relocs = NULL;
	for (i = 0; i < buffer_count; i++) {
		if (reloc_count + exec_list[i].relocation_count < reloc_count)
			return -EINVAL;
		reloc_count += exec_list[i].relocation_count;
	}

	*relocs = drm_calloc_large(reloc_count, sizeof(**relocs));
	if (*relocs == NULL) {
		DRM_ERROR("failed to alloc relocs, count %d\n", reloc_count);
		return -ENOMEM;
	}

	for (i = 0; i < buffer_count; i++) {
		struct drm_i915_gem_relocation_entry __user *user_relocs;

		user_relocs = (void __user *)(uintptr_t)exec_list[i].relocs_ptr;

		ret = copy_from_user(&(*relocs)[reloc_index],
				     user_relocs,
				     exec_list[i].relocation_count *
				     sizeof(**relocs));
		if (ret != 0) {
			drm_free_large(*relocs);
			*relocs = NULL;
			return -EFAULT;
		}

		reloc_index += exec_list[i].relocation_count;
	}

	return 0;
}

static int
i915_gem_put_relocs_to_user(struct drm_i915_gem_exec_object2 *exec_list,
			    uint32_t buffer_count,
			    struct drm_i915_gem_relocation_entry *relocs)
{
	uint32_t reloc_count = 0, i;
	int ret = 0;

	if (relocs == NULL)
	    return 0;

	for (i = 0; i < buffer_count; i++) {
		struct drm_i915_gem_relocation_entry __user *user_relocs;
		int unwritten;

		user_relocs = (void __user *)(uintptr_t)exec_list[i].relocs_ptr;

		unwritten = copy_to_user(user_relocs,
					 &relocs[reloc_count],
					 exec_list[i].relocation_count *
					 sizeof(*relocs));

		if (unwritten) {
			ret = -EFAULT;
			goto err;
		}

		reloc_count += exec_list[i].relocation_count;
	}

err:
	drm_free_large(relocs);

	return ret;
}

static int
i915_gem_check_execbuffer (struct drm_i915_gem_execbuffer2 *exec,
			   uint64_t exec_offset)
{
	uint32_t exec_start, exec_len;

	exec_start = (uint32_t) exec_offset + exec->batch_start_offset;
	exec_len = (uint32_t) exec->batch_len;

	if ((exec_start | exec_len) & 0x7)
		return -EINVAL;

	if (!exec_start)
		return -EINVAL;

	return 0;
}

int
i915_gem_wait_for_pending_flip(struct drm_device *dev,
			       struct drm_gem_object **object_list,
			       int count)
{
	drm_i915_private_t *dev_priv = dev->dev_private;
	struct drm_i915_gem_object *obj_priv;
	DEFINE_WAIT(wait);
	int i, ret = 0;

	for (;;) {
		prepare_to_wait(&dev_priv->pending_flip_queue,
				&wait, TASK_INTERRUPTIBLE);
		for (i = 0; i < count; i++) {
			obj_priv = to_intel_bo(object_list[i]);
			if (atomic_read(&obj_priv->pending_flip) > 0)
				break;
		}
		if (i == count)
			break;

		if (!signal_pending(current)) {
			mutex_unlock(&dev->struct_mutex);
			schedule();
			mutex_lock(&dev->struct_mutex);
			continue;
		}
		ret = -ERESTARTSYS;
		break;
	}
	finish_wait(&dev_priv->pending_flip_queue, &wait);

	return ret;
}

static int
i915_gem_do_execbuffer(struct drm_device *dev, void *data,
		       struct drm_file *file_priv,
		       struct drm_i915_gem_execbuffer2 *args,
		       struct drm_i915_gem_exec_object2 *exec_list)
{
	drm_i915_private_t *dev_priv = dev->dev_private;
	struct drm_gem_object **object_list = NULL;
	struct drm_gem_object *batch_obj;
	struct drm_i915_gem_object *obj_priv;
	struct drm_clip_rect *cliprects = NULL;
	struct drm_i915_gem_relocation_entry *relocs = NULL;
	struct drm_i915_gem_request *request = NULL;
	int ret = 0, ret2, i, pinned = 0;
	uint64_t exec_offset;
	uint32_t seqno, reloc_index;
	int pin_tries, flips;

	struct intel_ring_buffer *ring = NULL;

#if WATCH_EXEC
	DRM_INFO("buffers_ptr %d buffer_count %d len %08x\n",
		  (int) args->buffers_ptr, args->buffer_count, args->batch_len);
#endif
	if (args->flags & I915_EXEC_BSD) {
		if (!HAS_BSD(dev)) {
			DRM_ERROR("execbuf with wrong flag\n");
			return -EINVAL;
		}
		ring = &dev_priv->bsd_ring;
	} else {
		ring = &dev_priv->render_ring;
	}

	if (args->buffer_count < 1) {
		DRM_ERROR("execbuf with %d buffers\n", args->buffer_count);
		return -EINVAL;
	}
	object_list = drm_malloc_ab(sizeof(*object_list), args->buffer_count);
	if (object_list == NULL) {
		DRM_ERROR("Failed to allocate object list for %d buffers\n",
			  args->buffer_count);
		ret = -ENOMEM;
		goto pre_mutex_err;
	}

	if (args->num_cliprects != 0) {
		cliprects = kcalloc(args->num_cliprects, sizeof(*cliprects),
				    GFP_KERNEL);
		if (cliprects == NULL) {
			ret = -ENOMEM;
			goto pre_mutex_err;
		}

		ret = copy_from_user(cliprects,
				     (struct drm_clip_rect __user *)
				     (uintptr_t) args->cliprects_ptr,
				     sizeof(*cliprects) * args->num_cliprects);
		if (ret != 0) {
			DRM_ERROR("copy %d cliprects failed: %d\n",
				  args->num_cliprects, ret);
			ret = -EFAULT;
			goto pre_mutex_err;
		}
	}

	request = kzalloc(sizeof(*request), GFP_KERNEL);
	if (request == NULL) {
		ret = -ENOMEM;
		goto pre_mutex_err;
	}

	ret = i915_gem_get_relocs_from_user(exec_list, args->buffer_count,
					    &relocs);
	if (ret != 0)
		goto pre_mutex_err;

	mutex_lock(&dev->struct_mutex);

	i915_verify_inactive(dev, __FILE__, __LINE__);

	if (atomic_read(&dev_priv->mm.wedged)) {
		mutex_unlock(&dev->struct_mutex);
		ret = -EIO;
		goto pre_mutex_err;
	}

	if (dev_priv->mm.suspended) {
		mutex_unlock(&dev->struct_mutex);
		ret = -EBUSY;
		goto pre_mutex_err;
	}

	/* Look up object handles */
	flips = 0;
	for (i = 0; i < args->buffer_count; i++) {
		object_list[i] = drm_gem_object_lookup(dev, file_priv,
						       exec_list[i].handle);
		if (object_list[i] == NULL) {
			DRM_ERROR("Invalid object handle %d at index %d\n",
				   exec_list[i].handle, i);
			/* prevent error path from reading uninitialized data */
			args->buffer_count = i + 1;
			ret = -ENOENT;
			goto err;
		}

		obj_priv = to_intel_bo(object_list[i]);
		if (obj_priv->in_execbuffer) {
			DRM_ERROR("Object %p appears more than once in object list\n",
				   object_list[i]);
			/* prevent error path from reading uninitialized data */
			args->buffer_count = i + 1;
			ret = -EINVAL;
			goto err;
		}
		obj_priv->in_execbuffer = true;
		flips += atomic_read(&obj_priv->pending_flip);
	}

	if (flips > 0) {
		ret = i915_gem_wait_for_pending_flip(dev, object_list,
						     args->buffer_count);
		if (ret)
			goto err;
	}

	/* Pin and relocate */
	for (pin_tries = 0; ; pin_tries++) {
		ret = 0;
		reloc_index = 0;

		for (i = 0; i < args->buffer_count; i++) {
			object_list[i]->pending_read_domains = 0;
			object_list[i]->pending_write_domain = 0;
			ret = i915_gem_object_pin_and_relocate(object_list[i],
							       file_priv,
							       &exec_list[i],
							       &relocs[reloc_index]);
			if (ret)
				break;
			pinned = i + 1;
			reloc_index += exec_list[i].relocation_count;
		}
		/* success */
		if (ret == 0)
			break;

		/* error other than GTT full, or we've already tried again */
		if (ret != -ENOSPC || pin_tries >= 1) {
			if (ret != -ERESTARTSYS) {
				unsigned long long total_size = 0;
				int num_fences = 0;
				for (i = 0; i < args->buffer_count; i++) {
					obj_priv = to_intel_bo(object_list[i]);

					total_size += object_list[i]->size;
					num_fences +=
						exec_list[i].flags & EXEC_OBJECT_NEEDS_FENCE &&
						obj_priv->tiling_mode != I915_TILING_NONE;
				}
				DRM_ERROR("Failed to pin buffer %d of %d, total %llu bytes, %d fences: %d\n",
					  pinned+1, args->buffer_count,
					  total_size, num_fences,
					  ret);
				DRM_ERROR("%d objects [%d pinned], "
					  "%d object bytes [%d pinned], "
					  "%d/%d gtt bytes\n",
					  atomic_read(&dev->object_count),
					  atomic_read(&dev->pin_count),
					  atomic_read(&dev->object_memory),
					  atomic_read(&dev->pin_memory),
					  atomic_read(&dev->gtt_memory),
					  dev->gtt_total);
			}
			goto err;
		}

		/* unpin all of our buffers */
		for (i = 0; i < pinned; i++)
			i915_gem_object_unpin(object_list[i]);
		pinned = 0;

		/* evict everyone we can from the aperture */
		ret = i915_gem_evict_everything(dev);
		if (ret && ret != -ENOSPC)
			goto err;
	}

	/* Set the pending read domains for the batch buffer to COMMAND */
	batch_obj = object_list[args->buffer_count-1];
	if (batch_obj->pending_write_domain) {
		DRM_ERROR("Attempting to use self-modifying batch buffer\n");
		ret = -EINVAL;
		goto err;
	}
	batch_obj->pending_read_domains |= I915_GEM_DOMAIN_COMMAND;

	/* Sanity check the batch buffer, prior to moving objects */
	exec_offset = exec_list[args->buffer_count - 1].offset;
	ret = i915_gem_check_execbuffer (args, exec_offset);
	if (ret != 0) {
		DRM_ERROR("execbuf with invalid offset/length\n");
		goto err;
	}

	i915_verify_inactive(dev, __FILE__, __LINE__);

	/* Zero the global flush/invalidate flags. These
	 * will be modified as new domains are computed
	 * for each object
	 */
	dev->invalidate_domains = 0;
	dev->flush_domains = 0;
	dev_priv->mm.flush_rings = 0;

	for (i = 0; i < args->buffer_count; i++) {
		struct drm_gem_object *obj = object_list[i];

		/* Compute new gpu domains and update invalidate/flush */
		i915_gem_object_set_to_gpu_domain(obj);
	}

	i915_verify_inactive(dev, __FILE__, __LINE__);

	if (dev->invalidate_domains | dev->flush_domains) {
#if WATCH_EXEC
		DRM_INFO("%s: invalidate_domains %08x flush_domains %08x\n",
			  __func__,
			 dev->invalidate_domains,
			 dev->flush_domains);
#endif
		i915_gem_flush(dev, file_priv,
			       dev->invalidate_domains,
			       dev->flush_domains,
			       dev_priv->mm.flush_rings);
	}

	for (i = 0; i < args->buffer_count; i++) {
		struct drm_gem_object *obj = object_list[i];
		struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
		uint32_t old_write_domain = obj->write_domain;

		obj->write_domain = obj->pending_write_domain;
		if (obj->write_domain)
			list_move_tail(&obj_priv->gpu_write_list,
				       &dev_priv->mm.gpu_write_list);
		else
			list_del_init(&obj_priv->gpu_write_list);

		trace_i915_gem_object_change_domain(obj,
						    obj->read_domains,
						    old_write_domain);
	}

	i915_verify_inactive(dev, __FILE__, __LINE__);

#if WATCH_COHERENCY
	for (i = 0; i < args->buffer_count; i++) {
		i915_gem_object_check_coherency(object_list[i],
						exec_list[i].handle);
	}
#endif

#if WATCH_EXEC
	i915_gem_dump_object(batch_obj,
			      args->batch_len,
			      __func__,
			      ~0);
#endif

	/* Exec the batchbuffer */
	ret = ring->dispatch_gem_execbuffer(dev, ring, args,
			cliprects, exec_offset);
	if (ret) {
		DRM_ERROR("dispatch failed %d\n", ret);
		goto err;
	}

	/*
	 * Ensure that the commands in the batch buffer are
	 * finished before the interrupt fires
	 */
	i915_retire_commands(dev, ring);

	i915_verify_inactive(dev, __FILE__, __LINE__);

	for (i = 0; i < args->buffer_count; i++) {
		struct drm_gem_object *obj = object_list[i];
		obj_priv = to_intel_bo(obj);

		i915_gem_object_move_to_active(obj, ring);
#if WATCH_LRU
		DRM_INFO("%s: move to exec list %p\n", __func__, obj);
#endif
	}

	/*
	 * Get a seqno representing the execution of the current buffer,
	 * which we can wait on.  We would like to mitigate these interrupts,
	 * likely by only creating seqnos occasionally (so that we have
	 * *some* interrupts representing completion of buffers that we can
	 * wait on when trying to clear up gtt space).
	 */
	seqno = i915_add_request(dev, file_priv, request, ring);
	request = NULL;

#if WATCH_LRU
	i915_dump_lru(dev, __func__);
#endif

	i915_verify_inactive(dev, __FILE__, __LINE__);

err:
	for (i = 0; i < pinned; i++)
		i915_gem_object_unpin(object_list[i]);

	for (i = 0; i < args->buffer_count; i++) {
		if (object_list[i]) {
			obj_priv = to_intel_bo(object_list[i]);
			obj_priv->in_execbuffer = false;
		}
		drm_gem_object_unreference(object_list[i]);
	}

	mutex_unlock(&dev->struct_mutex);

pre_mutex_err:
	/* Copy the updated relocations out regardless of current error
	 * state.  Failure to update the relocs would mean that the next
	 * time userland calls execbuf, it would do so with presumed offset
	 * state that didn't match the actual object state.
	 */
	ret2 = i915_gem_put_relocs_to_user(exec_list, args->buffer_count,
					   relocs);
	if (ret2 != 0) {
		DRM_ERROR("Failed to copy relocations back out: %d\n", ret2);

		if (ret == 0)
			ret = ret2;
	}

	drm_free_large(object_list);
	kfree(cliprects);
	kfree(request);

	return ret;
}

/*
 * Legacy execbuffer just creates an exec2 list from the original exec object
 * list array and passes it to the real function.
 */
int
i915_gem_execbuffer(struct drm_device *dev, void *data,
		    struct drm_file *file_priv)
{
	struct drm_i915_gem_execbuffer *args = data;
	struct drm_i915_gem_execbuffer2 exec2;
	struct drm_i915_gem_exec_object *exec_list = NULL;
	struct drm_i915_gem_exec_object2 *exec2_list = NULL;
	int ret, i;

#if WATCH_EXEC
	DRM_INFO("buffers_ptr %d buffer_count %d len %08x\n",
		  (int) args->buffers_ptr, args->buffer_count, args->batch_len);
#endif

	if (args->buffer_count < 1) {
		DRM_ERROR("execbuf with %d buffers\n", args->buffer_count);
		return -EINVAL;
	}

	/* Copy in the exec list from userland */
	exec_list = drm_malloc_ab(sizeof(*exec_list), args->buffer_count);
	exec2_list = drm_malloc_ab(sizeof(*exec2_list), args->buffer_count);
	if (exec_list == NULL || exec2_list == NULL) {
		DRM_ERROR("Failed to allocate exec list for %d buffers\n",
			  args->buffer_count);
		drm_free_large(exec_list);
		drm_free_large(exec2_list);
		return -ENOMEM;
	}
	ret = copy_from_user(exec_list,
			     (struct drm_i915_relocation_entry __user *)
			     (uintptr_t) args->buffers_ptr,
			     sizeof(*exec_list) * args->buffer_count);
	if (ret != 0) {
		DRM_ERROR("copy %d exec entries failed %d\n",
			  args->buffer_count, ret);
		drm_free_large(exec_list);
		drm_free_large(exec2_list);
		return -EFAULT;
	}

	for (i = 0; i < args->buffer_count; i++) {
		exec2_list[i].handle = exec_list[i].handle;
		exec2_list[i].relocation_count = exec_list[i].relocation_count;
		exec2_list[i].relocs_ptr = exec_list[i].relocs_ptr;
		exec2_list[i].alignment = exec_list[i].alignment;
		exec2_list[i].offset = exec_list[i].offset;
		if (INTEL_INFO(dev)->gen < 4)
			exec2_list[i].flags = EXEC_OBJECT_NEEDS_FENCE;
		else
			exec2_list[i].flags = 0;
	}

	exec2.buffers_ptr = args->buffers_ptr;
	exec2.buffer_count = args->buffer_count;
	exec2.batch_start_offset = args->batch_start_offset;
	exec2.batch_len = args->batch_len;
	exec2.DR1 = args->DR1;
	exec2.DR4 = args->DR4;
	exec2.num_cliprects = args->num_cliprects;
	exec2.cliprects_ptr = args->cliprects_ptr;
	exec2.flags = I915_EXEC_RENDER;

	ret = i915_gem_do_execbuffer(dev, data, file_priv, &exec2, exec2_list);
	if (!ret) {
		/* Copy the new buffer offsets back to the user's exec list. */
		for (i = 0; i < args->buffer_count; i++)
			exec_list[i].offset = exec2_list[i].offset;
		/* ... and back out to userspace */
		ret = copy_to_user((struct drm_i915_relocation_entry __user *)
				   (uintptr_t) args->buffers_ptr,
				   exec_list,
				   sizeof(*exec_list) * args->buffer_count);
		if (ret) {
			ret = -EFAULT;
			DRM_ERROR("failed to copy %d exec entries "
				  "back to user (%d)\n",
				  args->buffer_count, ret);
		}
	}

	drm_free_large(exec_list);
	drm_free_large(exec2_list);
	return ret;
}

int
i915_gem_execbuffer2(struct drm_device *dev, void *data,
		     struct drm_file *file_priv)
{
	struct drm_i915_gem_execbuffer2 *args = data;
	struct drm_i915_gem_exec_object2 *exec2_list = NULL;
	int ret;

#if WATCH_EXEC
	DRM_INFO("buffers_ptr %d buffer_count %d len %08x\n",
		  (int) args->buffers_ptr, args->buffer_count, args->batch_len);
#endif

	if (args->buffer_count < 1) {
		DRM_ERROR("execbuf2 with %d buffers\n", args->buffer_count);
		return -EINVAL;
	}

	exec2_list = drm_malloc_ab(sizeof(*exec2_list), args->buffer_count);
	if (exec2_list == NULL) {
		DRM_ERROR("Failed to allocate exec list for %d buffers\n",
			  args->buffer_count);
		return -ENOMEM;
	}
	ret = copy_from_user(exec2_list,
			     (struct drm_i915_relocation_entry __user *)
			     (uintptr_t) args->buffers_ptr,
			     sizeof(*exec2_list) * args->buffer_count);
	if (ret != 0) {
		DRM_ERROR("copy %d exec entries failed %d\n",
			  args->buffer_count, ret);
		drm_free_large(exec2_list);
		return -EFAULT;
	}

	ret = i915_gem_do_execbuffer(dev, data, file_priv, args, exec2_list);
	if (!ret) {
		/* Copy the new buffer offsets back to the user's exec list. */
		ret = copy_to_user((struct drm_i915_relocation_entry __user *)
				   (uintptr_t) args->buffers_ptr,
				   exec2_list,
				   sizeof(*exec2_list) * args->buffer_count);
		if (ret) {
			ret = -EFAULT;
			DRM_ERROR("failed to copy %d exec entries "
				  "back to user (%d)\n",
				  args->buffer_count, ret);
		}
	}

	drm_free_large(exec2_list);
	return ret;
}

int
i915_gem_object_pin(struct drm_gem_object *obj, uint32_t alignment)
{
	struct drm_device *dev = obj->dev;
	struct drm_i915_private *dev_priv = dev->dev_private;
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
	int ret;

	BUG_ON(obj_priv->pin_count == DRM_I915_GEM_OBJECT_MAX_PIN_COUNT);

	i915_verify_inactive(dev, __FILE__, __LINE__);

	if (obj_priv->gtt_space != NULL) {
		if (alignment == 0)
			alignment = i915_gem_get_gtt_alignment(obj);
		if (obj_priv->gtt_offset & (alignment - 1)) {
			WARN(obj_priv->pin_count,
			     "bo is already pinned with incorrect alignment:"
			     " offset=%x, req.alignment=%x\n",
			     obj_priv->gtt_offset, alignment);
			ret = i915_gem_object_unbind(obj);
			if (ret)
				return ret;
		}
	}

	if (obj_priv->gtt_space == NULL) {
		ret = i915_gem_object_bind_to_gtt(obj, alignment);
		if (ret)
			return ret;
	}

	obj_priv->pin_count++;

	/* If the object is not active and not pending a flush,
	 * remove it from the inactive list
	 */
	if (obj_priv->pin_count == 1) {
		atomic_inc(&dev->pin_count);
		atomic_add(obj->size, &dev->pin_memory);
		if (!obj_priv->active)
			list_move_tail(&obj_priv->list,
				       &dev_priv->mm.pinned_list);
	}
	i915_verify_inactive(dev, __FILE__, __LINE__);

	return 0;
}

void
i915_gem_object_unpin(struct drm_gem_object *obj)
{
	struct drm_device *dev = obj->dev;
	drm_i915_private_t *dev_priv = dev->dev_private;
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);

	i915_verify_inactive(dev, __FILE__, __LINE__);
	obj_priv->pin_count--;
	BUG_ON(obj_priv->pin_count < 0);
	BUG_ON(obj_priv->gtt_space == NULL);

	/* If the object is no longer pinned, and is
	 * neither active nor being flushed, then stick it on
	 * the inactive list
	 */
	if (obj_priv->pin_count == 0) {
		if (!obj_priv->active)
			list_move_tail(&obj_priv->list,
				       &dev_priv->mm.inactive_list);
		atomic_dec(&dev->pin_count);
		atomic_sub(obj->size, &dev->pin_memory);
	}
	i915_verify_inactive(dev, __FILE__, __LINE__);
}

int
i915_gem_pin_ioctl(struct drm_device *dev, void *data,
		   struct drm_file *file_priv)
{
	struct drm_i915_gem_pin *args = data;
	struct drm_gem_object *obj;
	struct drm_i915_gem_object *obj_priv;
	int ret;

	mutex_lock(&dev->struct_mutex);

	obj = drm_gem_object_lookup(dev, file_priv, args->handle);
	if (obj == NULL) {
		DRM_ERROR("Bad handle in i915_gem_pin_ioctl(): %d\n",
			  args->handle);
		mutex_unlock(&dev->struct_mutex);
		return -ENOENT;
	}
	obj_priv = to_intel_bo(obj);

	if (obj_priv->madv != I915_MADV_WILLNEED) {
		DRM_ERROR("Attempting to pin a purgeable buffer\n");
		drm_gem_object_unreference(obj);
		mutex_unlock(&dev->struct_mutex);
		return -EINVAL;
	}

	if (obj_priv->pin_filp != NULL && obj_priv->pin_filp != file_priv) {
		DRM_ERROR("Already pinned in i915_gem_pin_ioctl(): %d\n",
			  args->handle);
		drm_gem_object_unreference(obj);
		mutex_unlock(&dev->struct_mutex);
		return -EINVAL;
	}

	obj_priv->user_pin_count++;
	obj_priv->pin_filp = file_priv;
	if (obj_priv->user_pin_count == 1) {
		ret = i915_gem_object_pin(obj, args->alignment);
		if (ret != 0) {
			drm_gem_object_unreference(obj);
			mutex_unlock(&dev->struct_mutex);
			return ret;
		}
	}

	/* XXX - flush the CPU caches for pinned objects
	 * as the X server doesn't manage domains yet
	 */
	i915_gem_object_flush_cpu_write_domain(obj);
	args->offset = obj_priv->gtt_offset;
	drm_gem_object_unreference(obj);
	mutex_unlock(&dev->struct_mutex);

	return 0;
}

int
i915_gem_unpin_ioctl(struct drm_device *dev, void *data,
		     struct drm_file *file_priv)
{
	struct drm_i915_gem_pin *args = data;
	struct drm_gem_object *obj;
	struct drm_i915_gem_object *obj_priv;

	mutex_lock(&dev->struct_mutex);

	obj = drm_gem_object_lookup(dev, file_priv, args->handle);
	if (obj == NULL) {
		DRM_ERROR("Bad handle in i915_gem_unpin_ioctl(): %d\n",
			  args->handle);
		mutex_unlock(&dev->struct_mutex);
		return -ENOENT;
	}

	obj_priv = to_intel_bo(obj);
	if (obj_priv->pin_filp != file_priv) {
		DRM_ERROR("Not pinned by caller in i915_gem_pin_ioctl(): %d\n",
			  args->handle);
		drm_gem_object_unreference(obj);
		mutex_unlock(&dev->struct_mutex);
		return -EINVAL;
	}
	obj_priv->user_pin_count--;
	if (obj_priv->user_pin_count == 0) {
		obj_priv->pin_filp = NULL;
		i915_gem_object_unpin(obj);
	}

	drm_gem_object_unreference(obj);
	mutex_unlock(&dev->struct_mutex);
	return 0;
}

int
i915_gem_busy_ioctl(struct drm_device *dev, void *data,
		    struct drm_file *file_priv)
{
	struct drm_i915_gem_busy *args = data;
	struct drm_gem_object *obj;
	struct drm_i915_gem_object *obj_priv;

	obj = drm_gem_object_lookup(dev, file_priv, args->handle);
	if (obj == NULL) {
		DRM_ERROR("Bad handle in i915_gem_busy_ioctl(): %d\n",
			  args->handle);
		return -ENOENT;
	}

	mutex_lock(&dev->struct_mutex);

	/* Count all active objects as busy, even if they are currently not used
	 * by the gpu. Users of this interface expect objects to eventually
	 * become non-busy without any further actions, therefore emit any
	 * necessary flushes here.
	 */
	obj_priv = to_intel_bo(obj);
	args->busy = obj_priv->active;
	if (args->busy) {
		/* Unconditionally flush objects, even when the gpu still uses this
		 * object. Userspace calling this function indicates that it wants to
		 * use this buffer rather sooner than later, so issuing the required
		 * flush earlier is beneficial.
		 */
		if (obj->write_domain & I915_GEM_GPU_DOMAINS)
			i915_gem_flush_ring(dev, file_priv,
					    obj_priv->ring,
					    0, obj->write_domain);

		/* Update the active list for the hardware's current position.
		 * Otherwise this only updates on a delayed timer or when irqs
		 * are actually unmasked, and our working set ends up being
		 * larger than required.
		 */
		i915_gem_retire_requests_ring(dev, obj_priv->ring);

		args->busy = obj_priv->active;
	}

	drm_gem_object_unreference(obj);
	mutex_unlock(&dev->struct_mutex);
	return 0;
}

int
i915_gem_throttle_ioctl(struct drm_device *dev, void *data,
			struct drm_file *file_priv)
{
    return i915_gem_ring_throttle(dev, file_priv);
}

int
i915_gem_madvise_ioctl(struct drm_device *dev, void *data,
		       struct drm_file *file_priv)
{
	struct drm_i915_gem_madvise *args = data;
	struct drm_gem_object *obj;
	struct drm_i915_gem_object *obj_priv;

	switch (args->madv) {
	case I915_MADV_DONTNEED:
	case I915_MADV_WILLNEED:
	    break;
	default:
	    return -EINVAL;
	}

	obj = drm_gem_object_lookup(dev, file_priv, args->handle);
	if (obj == NULL) {
		DRM_ERROR("Bad handle in i915_gem_madvise_ioctl(): %d\n",
			  args->handle);
		return -ENOENT;
	}

	mutex_lock(&dev->struct_mutex);
	obj_priv = to_intel_bo(obj);

	if (obj_priv->pin_count) {
		drm_gem_object_unreference(obj);
		mutex_unlock(&dev->struct_mutex);

		DRM_ERROR("Attempted i915_gem_madvise_ioctl() on a pinned object\n");
		return -EINVAL;
	}

	if (obj_priv->madv != __I915_MADV_PURGED)
		obj_priv->madv = args->madv;

	/* if the object is no longer bound, discard its backing storage */
	if (i915_gem_object_is_purgeable(obj_priv) &&
	    obj_priv->gtt_space == NULL)
		i915_gem_object_truncate(obj);

	args->retained = obj_priv->madv != __I915_MADV_PURGED;

	drm_gem_object_unreference(obj);
	mutex_unlock(&dev->struct_mutex);

	return 0;
}

struct drm_gem_object * i915_gem_alloc_object(struct drm_device *dev,
					      size_t size)
{
	struct drm_i915_gem_object *obj;

	obj = kzalloc(sizeof(*obj), GFP_KERNEL);
	if (obj == NULL)
		return NULL;

	if (drm_gem_object_init(dev, &obj->base, size) != 0) {
		kfree(obj);
		return NULL;
	}

	obj->base.write_domain = I915_GEM_DOMAIN_CPU;
	obj->base.read_domains = I915_GEM_DOMAIN_CPU;

	obj->agp_type = AGP_USER_MEMORY;
	obj->base.driver_private = NULL;
	obj->fence_reg = I915_FENCE_REG_NONE;
	INIT_LIST_HEAD(&obj->list);
	INIT_LIST_HEAD(&obj->gpu_write_list);
	obj->madv = I915_MADV_WILLNEED;

	trace_i915_gem_object_create(&obj->base);

	return &obj->base;
}

int i915_gem_init_object(struct drm_gem_object *obj)
{
	BUG();

	return 0;
}

static void i915_gem_free_object_tail(struct drm_gem_object *obj)
{
	struct drm_device *dev = obj->dev;
	drm_i915_private_t *dev_priv = dev->dev_private;
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
	int ret;

	ret = i915_gem_object_unbind(obj);
	if (ret == -ERESTARTSYS) {
		list_move(&obj_priv->list,
			  &dev_priv->mm.deferred_free_list);
		return;
	}

	if (obj_priv->mmap_offset)
		i915_gem_free_mmap_offset(obj);

	drm_gem_object_release(obj);

	kfree(obj_priv->page_cpu_valid);
	kfree(obj_priv->bit_17);
	kfree(obj_priv);
}

void i915_gem_free_object(struct drm_gem_object *obj)
{
	struct drm_device *dev = obj->dev;
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);

	trace_i915_gem_object_destroy(obj);

	while (obj_priv->pin_count > 0)
		i915_gem_object_unpin(obj);

	if (obj_priv->phys_obj)
		i915_gem_detach_phys_object(dev, obj);

	i915_gem_free_object_tail(obj);
}

int
i915_gem_idle(struct drm_device *dev)
{
	drm_i915_private_t *dev_priv = dev->dev_private;
	int ret;

	mutex_lock(&dev->struct_mutex);

	if (dev_priv->mm.suspended ||
			(dev_priv->render_ring.gem_object == NULL) ||
			(HAS_BSD(dev) &&
			 dev_priv->bsd_ring.gem_object == NULL)) {
		mutex_unlock(&dev->struct_mutex);
		return 0;
	}

	ret = i915_gpu_idle(dev);
	if (ret) {
		mutex_unlock(&dev->struct_mutex);
		return ret;
	}

	/* Under UMS, be paranoid and evict. */
	if (!drm_core_check_feature(dev, DRIVER_MODESET)) {
		ret = i915_gem_evict_inactive(dev);
		if (ret) {
			mutex_unlock(&dev->struct_mutex);
			return ret;
		}
	}

	/* Hack!  Don't let anybody do execbuf while we don't control the chip.
	 * We need to replace this with a semaphore, or something.
	 * And not confound mm.suspended!
	 */
	dev_priv->mm.suspended = 1;
	del_timer_sync(&dev_priv->hangcheck_timer);

	i915_kernel_lost_context(dev);
	i915_gem_cleanup_ringbuffer(dev);

	mutex_unlock(&dev->struct_mutex);

	/* Cancel the retire work handler, which should be idle now. */
	cancel_delayed_work_sync(&dev_priv->mm.retire_work);

	return 0;
}

/*
 * 965+ support PIPE_CONTROL commands, which provide finer grained control
 * over cache flushing.
 */
static int
i915_gem_init_pipe_control(struct drm_device *dev)
{
	drm_i915_private_t *dev_priv = dev->dev_private;
	struct drm_gem_object *obj;
	struct drm_i915_gem_object *obj_priv;
	int ret;

	obj = i915_gem_alloc_object(dev, 4096);
	if (obj == NULL) {
		DRM_ERROR("Failed to allocate seqno page\n");
		ret = -ENOMEM;
		goto err;
	}
	obj_priv = to_intel_bo(obj);
	obj_priv->agp_type = AGP_USER_CACHED_MEMORY;

	ret = i915_gem_object_pin(obj, 4096);
	if (ret)
		goto err_unref;

	dev_priv->seqno_gfx_addr = obj_priv->gtt_offset;
	dev_priv->seqno_page =  kmap(obj_priv->pages[0]);
	if (dev_priv->seqno_page == NULL)
		goto err_unpin;

	dev_priv->seqno_obj = obj;
	memset(dev_priv->seqno_page, 0, PAGE_SIZE);

	return 0;

err_unpin:
	i915_gem_object_unpin(obj);
err_unref:
	drm_gem_object_unreference(obj);
err:
	return ret;
}


static void
i915_gem_cleanup_pipe_control(struct drm_device *dev)
{
	drm_i915_private_t *dev_priv = dev->dev_private;
	struct drm_gem_object *obj;
	struct drm_i915_gem_object *obj_priv;

	obj = dev_priv->seqno_obj;
	obj_priv = to_intel_bo(obj);
	kunmap(obj_priv->pages[0]);
	i915_gem_object_unpin(obj);
	drm_gem_object_unreference(obj);
	dev_priv->seqno_obj = NULL;

	dev_priv->seqno_page = NULL;
}

int
i915_gem_init_ringbuffer(struct drm_device *dev)
{
	drm_i915_private_t *dev_priv = dev->dev_private;
	int ret;

	if (HAS_PIPE_CONTROL(dev)) {
		ret = i915_gem_init_pipe_control(dev);
		if (ret)
			return ret;
	}

	ret = intel_init_render_ring_buffer(dev);
	if (ret)
		goto cleanup_pipe_control;

	if (HAS_BSD(dev)) {
		ret = intel_init_bsd_ring_buffer(dev);
		if (ret)
			goto cleanup_render_ring;
	}

	dev_priv->next_seqno = 1;

	return 0;

cleanup_render_ring:
	intel_cleanup_ring_buffer(dev, &dev_priv->render_ring);
cleanup_pipe_control:
	if (HAS_PIPE_CONTROL(dev))
		i915_gem_cleanup_pipe_control(dev);
	return ret;
}

void
i915_gem_cleanup_ringbuffer(struct drm_device *dev)
{
	drm_i915_private_t *dev_priv = dev->dev_private;

	intel_cleanup_ring_buffer(dev, &dev_priv->render_ring);
	if (HAS_BSD(dev))
		intel_cleanup_ring_buffer(dev, &dev_priv->bsd_ring);
	if (HAS_PIPE_CONTROL(dev))
		i915_gem_cleanup_pipe_control(dev);
}

int
i915_gem_entervt_ioctl(struct drm_device *dev, void *data,
		       struct drm_file *file_priv)
{
	drm_i915_private_t *dev_priv = dev->dev_private;
	int ret;

	if (drm_core_check_feature(dev, DRIVER_MODESET))
		return 0;

	if (atomic_read(&dev_priv->mm.wedged)) {
		DRM_ERROR("Reenabling wedged hardware, good luck\n");
		atomic_set(&dev_priv->mm.wedged, 0);
	}

	mutex_lock(&dev->struct_mutex);
	dev_priv->mm.suspended = 0;

	ret = i915_gem_init_ringbuffer(dev);
	if (ret != 0) {
		mutex_unlock(&dev->struct_mutex);
		return ret;
	}

	BUG_ON(!list_empty(&dev_priv->render_ring.active_list));
	BUG_ON(HAS_BSD(dev) && !list_empty(&dev_priv->bsd_ring.active_list));
	BUG_ON(!list_empty(&dev_priv->mm.flushing_list));
	BUG_ON(!list_empty(&dev_priv->mm.inactive_list));
	BUG_ON(!list_empty(&dev_priv->render_ring.request_list));
	BUG_ON(HAS_BSD(dev) && !list_empty(&dev_priv->bsd_ring.request_list));
	mutex_unlock(&dev->struct_mutex);

	ret = drm_irq_install(dev);
	if (ret)
		goto cleanup_ringbuffer;

	return 0;

cleanup_ringbuffer:
	mutex_lock(&dev->struct_mutex);
	i915_gem_cleanup_ringbuffer(dev);
	dev_priv->mm.suspended = 1;
	mutex_unlock(&dev->struct_mutex);

	return ret;
}

int
i915_gem_leavevt_ioctl(struct drm_device *dev, void *data,
		       struct drm_file *file_priv)
{
	if (drm_core_check_feature(dev, DRIVER_MODESET))
		return 0;

	drm_irq_uninstall(dev);
	return i915_gem_idle(dev);
}

void
i915_gem_lastclose(struct drm_device *dev)
{
	int ret;

	if (drm_core_check_feature(dev, DRIVER_MODESET))
		return;

	ret = i915_gem_idle(dev);
	if (ret)
		DRM_ERROR("failed to idle hardware: %d\n", ret);
}

void
i915_gem_load(struct drm_device *dev)
{
	int i;
	drm_i915_private_t *dev_priv = dev->dev_private;

	INIT_LIST_HEAD(&dev_priv->mm.flushing_list);
	INIT_LIST_HEAD(&dev_priv->mm.gpu_write_list);
	INIT_LIST_HEAD(&dev_priv->mm.inactive_list);
	INIT_LIST_HEAD(&dev_priv->mm.pinned_list);
	INIT_LIST_HEAD(&dev_priv->mm.fence_list);
	INIT_LIST_HEAD(&dev_priv->mm.deferred_free_list);
	INIT_LIST_HEAD(&dev_priv->render_ring.active_list);
	INIT_LIST_HEAD(&dev_priv->render_ring.request_list);
	if (HAS_BSD(dev)) {
		INIT_LIST_HEAD(&dev_priv->bsd_ring.active_list);
		INIT_LIST_HEAD(&dev_priv->bsd_ring.request_list);
	}
	for (i = 0; i < 16; i++)
		INIT_LIST_HEAD(&dev_priv->fence_regs[i].lru_list);
	INIT_DELAYED_WORK(&dev_priv->mm.retire_work,
			  i915_gem_retire_work_handler);
	spin_lock(&shrink_list_lock);
	list_add(&dev_priv->mm.shrink_list, &shrink_list);
	spin_unlock(&shrink_list_lock);

	/* On GEN3 we really need to make sure the ARB C3 LP bit is set */
	if (IS_GEN3(dev)) {
		u32 tmp = I915_READ(MI_ARB_STATE);
		if (!(tmp & MI_ARB_C3_LP_WRITE_ENABLE)) {
			/* arb state is a masked write, so set bit + bit in mask */
			tmp = MI_ARB_C3_LP_WRITE_ENABLE | (MI_ARB_C3_LP_WRITE_ENABLE << MI_ARB_MASK_SHIFT);
			I915_WRITE(MI_ARB_STATE, tmp);
		}
	}

	/* Old X drivers will take 0-2 for front, back, depth buffers */
	if (!drm_core_check_feature(dev, DRIVER_MODESET))
		dev_priv->fence_reg_start = 3;

	if (INTEL_INFO(dev)->gen >= 4 || IS_I945G(dev) || IS_I945GM(dev) || IS_G33(dev))
		dev_priv->num_fence_regs = 16;
	else
		dev_priv->num_fence_regs = 8;

	/* Initialize fence registers to zero */
	switch (INTEL_INFO(dev)->gen) {
	case 6:
		for (i = 0; i < 16; i++)
			I915_WRITE64(FENCE_REG_SANDYBRIDGE_0 + (i * 8), 0);
		break;
	case 5:
	case 4:
		for (i = 0; i < 16; i++)
			I915_WRITE64(FENCE_REG_965_0 + (i * 8), 0);
		break;
	case 3:
		if (IS_I945G(dev) || IS_I945GM(dev) || IS_G33(dev))
			for (i = 0; i < 8; i++)
				I915_WRITE(FENCE_REG_945_8 + (i * 4), 0);
	case 2:
		for (i = 0; i < 8; i++)
			I915_WRITE(FENCE_REG_830_0 + (i * 4), 0);
		break;
	}
	i915_gem_detect_bit_6_swizzle(dev);
	init_waitqueue_head(&dev_priv->pending_flip_queue);
}

/*
 * Create a physically contiguous memory object for this object
 * e.g. for cursor + overlay regs
 */
static int i915_gem_init_phys_object(struct drm_device *dev,
				     int id, int size, int align)
{
	drm_i915_private_t *dev_priv = dev->dev_private;
	struct drm_i915_gem_phys_object *phys_obj;
	int ret;

	if (dev_priv->mm.phys_objs[id - 1] || !size)
		return 0;

	phys_obj = kzalloc(sizeof(struct drm_i915_gem_phys_object), GFP_KERNEL);
	if (!phys_obj)
		return -ENOMEM;

	phys_obj->id = id;

	phys_obj->handle = drm_pci_alloc(dev, size, align);
	if (!phys_obj->handle) {
		ret = -ENOMEM;
		goto kfree_obj;
	}
#ifdef CONFIG_X86
	set_memory_wc((unsigned long)phys_obj->handle->vaddr, phys_obj->handle->size / PAGE_SIZE);
#endif

	dev_priv->mm.phys_objs[id - 1] = phys_obj;

	return 0;
kfree_obj:
	kfree(phys_obj);
	return ret;
}

static void i915_gem_free_phys_object(struct drm_device *dev, int id)
{
	drm_i915_private_t *dev_priv = dev->dev_private;
	struct drm_i915_gem_phys_object *phys_obj;

	if (!dev_priv->mm.phys_objs[id - 1])
		return;

	phys_obj = dev_priv->mm.phys_objs[id - 1];
	if (phys_obj->cur_obj) {
		i915_gem_detach_phys_object(dev, phys_obj->cur_obj);
	}

#ifdef CONFIG_X86
	set_memory_wb((unsigned long)phys_obj->handle->vaddr, phys_obj->handle->size / PAGE_SIZE);
#endif
	drm_pci_free(dev, phys_obj->handle);
	kfree(phys_obj);
	dev_priv->mm.phys_objs[id - 1] = NULL;
}

void i915_gem_free_all_phys_object(struct drm_device *dev)
{
	int i;

	for (i = I915_GEM_PHYS_CURSOR_0; i <= I915_MAX_PHYS_OBJECT; i++)
		i915_gem_free_phys_object(dev, i);
}

void i915_gem_detach_phys_object(struct drm_device *dev,
				 struct drm_gem_object *obj)
{
	struct drm_i915_gem_object *obj_priv;
	int i;
	int ret;
	int page_count;

	obj_priv = to_intel_bo(obj);
	if (!obj_priv->phys_obj)
		return;

	ret = i915_gem_object_get_pages(obj, 0);
	if (ret)
		goto out;

	page_count = obj->size / PAGE_SIZE;

	for (i = 0; i < page_count; i++) {
		char *dst = kmap_atomic(obj_priv->pages[i], KM_USER0);
		char *src = obj_priv->phys_obj->handle->vaddr + (i * PAGE_SIZE);

		memcpy(dst, src, PAGE_SIZE);
		kunmap_atomic(dst, KM_USER0);
	}
	drm_clflush_pages(obj_priv->pages, page_count);
	drm_agp_chipset_flush(dev);

	i915_gem_object_put_pages(obj);
out:
	obj_priv->phys_obj->cur_obj = NULL;
	obj_priv->phys_obj = NULL;
}

int
i915_gem_attach_phys_object(struct drm_device *dev,
			    struct drm_gem_object *obj,
			    int id,
			    int align)
{
	drm_i915_private_t *dev_priv = dev->dev_private;
	struct drm_i915_gem_object *obj_priv;
	int ret = 0;
	int page_count;
	int i;

	if (id > I915_MAX_PHYS_OBJECT)
		return -EINVAL;

	obj_priv = to_intel_bo(obj);

	if (obj_priv->phys_obj) {
		if (obj_priv->phys_obj->id == id)
			return 0;
		i915_gem_detach_phys_object(dev, obj);
	}

	/* create a new object */
	if (!dev_priv->mm.phys_objs[id - 1]) {
		ret = i915_gem_init_phys_object(dev, id,
						obj->size, align);
		if (ret) {
			DRM_ERROR("failed to init phys object %d size: %zu\n", id, obj->size);
			goto out;
		}
	}

	/* bind to the object */
	obj_priv->phys_obj = dev_priv->mm.phys_objs[id - 1];
	obj_priv->phys_obj->cur_obj = obj;

	ret = i915_gem_object_get_pages(obj, 0);
	if (ret) {
		DRM_ERROR("failed to get page list\n");
		goto out;
	}

	page_count = obj->size / PAGE_SIZE;

	for (i = 0; i < page_count; i++) {
		char *src = kmap_atomic(obj_priv->pages[i], KM_USER0);
		char *dst = obj_priv->phys_obj->handle->vaddr + (i * PAGE_SIZE);

		memcpy(dst, src, PAGE_SIZE);
		kunmap_atomic(src, KM_USER0);
	}

	i915_gem_object_put_pages(obj);

	return 0;
out:
	return ret;
}

static int
i915_gem_phys_pwrite(struct drm_device *dev, struct drm_gem_object *obj,
		     struct drm_i915_gem_pwrite *args,
		     struct drm_file *file_priv)
{
	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
	void *obj_addr;
	int ret;
	char __user *user_data;

	user_data = (char __user *) (uintptr_t) args->data_ptr;
	obj_addr = obj_priv->phys_obj->handle->vaddr + args->offset;

	DRM_DEBUG_DRIVER("obj_addr %p, %lld\n", obj_addr, args->size);
	ret = copy_from_user(obj_addr, user_data, args->size);
	if (ret)
		return -EFAULT;

	drm_agp_chipset_flush(dev);
	return 0;
}

void i915_gem_release(struct drm_device * dev, struct drm_file *file_priv)
{
	struct drm_i915_file_private *i915_file_priv = file_priv->driver_priv;

	/* Clean up our request list when the client is going away, so that
	 * later retire_requests won't dereference our soon-to-be-gone
	 * file_priv.
	 */
	mutex_lock(&dev->struct_mutex);
	while (!list_empty(&i915_file_priv->mm.request_list))
		list_del_init(i915_file_priv->mm.request_list.next);
	mutex_unlock(&dev->struct_mutex);
}

static int
i915_gpu_is_active(struct drm_device *dev)
{
	drm_i915_private_t *dev_priv = dev->dev_private;
	int lists_empty;

	lists_empty = list_empty(&dev_priv->mm.flushing_list) &&
		      list_empty(&dev_priv->render_ring.active_list);
	if (HAS_BSD(dev))
		lists_empty &= list_empty(&dev_priv->bsd_ring.active_list);

	return !lists_empty;
}

static int
i915_gem_shrink(struct shrinker *shrink, int nr_to_scan, gfp_t gfp_mask)
{
	drm_i915_private_t *dev_priv, *next_dev;
	struct drm_i915_gem_object *obj_priv, *next_obj;
	int cnt = 0;
	int would_deadlock = 1;

	/* "fast-path" to count number of available objects */
	if (nr_to_scan == 0) {
		spin_lock(&shrink_list_lock);
		list_for_each_entry(dev_priv, &shrink_list, mm.shrink_list) {
			struct drm_device *dev = dev_priv->dev;

			if (mutex_trylock(&dev->struct_mutex)) {
				list_for_each_entry(obj_priv,
						    &dev_priv->mm.inactive_list,
						    list)
					cnt++;
				mutex_unlock(&dev->struct_mutex);
			}
		}
		spin_unlock(&shrink_list_lock);

		return (cnt / 100) * sysctl_vfs_cache_pressure;
	}

	spin_lock(&shrink_list_lock);

rescan:
	/* first scan for clean buffers */
	list_for_each_entry_safe(dev_priv, next_dev,
				 &shrink_list, mm.shrink_list) {
		struct drm_device *dev = dev_priv->dev;

		if (! mutex_trylock(&dev->struct_mutex))
			continue;

		spin_unlock(&shrink_list_lock);
		i915_gem_retire_requests(dev);

		list_for_each_entry_safe(obj_priv, next_obj,
					 &dev_priv->mm.inactive_list,
					 list) {
			if (i915_gem_object_is_purgeable(obj_priv)) {
				i915_gem_object_unbind(&obj_priv->base);
				if (--nr_to_scan <= 0)
					break;
			}
		}

		spin_lock(&shrink_list_lock);
		mutex_unlock(&dev->struct_mutex);

		would_deadlock = 0;

		if (nr_to_scan <= 0)
			break;
	}

	/* second pass, evict/count anything still on the inactive list */
	list_for_each_entry_safe(dev_priv, next_dev,
				 &shrink_list, mm.shrink_list) {
		struct drm_device *dev = dev_priv->dev;

		if (! mutex_trylock(&dev->struct_mutex))
			continue;

		spin_unlock(&shrink_list_lock);

		list_for_each_entry_safe(obj_priv, next_obj,
					 &dev_priv->mm.inactive_list,
					 list) {
			if (nr_to_scan > 0) {
				i915_gem_object_unbind(&obj_priv->base);
				nr_to_scan--;
			} else
				cnt++;
		}

		spin_lock(&shrink_list_lock);
		mutex_unlock(&dev->struct_mutex);

		would_deadlock = 0;
	}

	if (nr_to_scan) {
		int active = 0;

		/*
		 * We are desperate for pages, so as a last resort, wait
		 * for the GPU to finish and discard whatever we can.
		 * This has a dramatic impact to reduce the number of
		 * OOM-killer events whilst running the GPU aggressively.
		 */
		list_for_each_entry(dev_priv, &shrink_list, mm.shrink_list) {
			struct drm_device *dev = dev_priv->dev;

			if (!mutex_trylock(&dev->struct_mutex))
				continue;

			spin_unlock(&shrink_list_lock);

			if (i915_gpu_is_active(dev)) {
				i915_gpu_idle(dev);
				active++;
			}

			spin_lock(&shrink_list_lock);
			mutex_unlock(&dev->struct_mutex);
		}

		if (active)
			goto rescan;
	}

	spin_unlock(&shrink_list_lock);

	if (would_deadlock)
		return -1;
	else if (cnt > 0)
		return (cnt / 100) * sysctl_vfs_cache_pressure;
	else
		return 0;
}

static struct shrinker shrinker = {
	.shrink = i915_gem_shrink,
	.seeks = DEFAULT_SEEKS,
};

__init void
i915_gem_shrinker_init(void)
{
    register_shrinker(&shrinker);
}

__exit void
i915_gem_shrinker_exit(void)
{
    unregister_shrinker(&shrinker);
}