blob: c26d36cc4b313ac4d03ade4167739d7fd7be6e9c [file] [log] [blame]
/*
* 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 <drm/drmP.h>
#include <drm/drm_vma_manager.h>
#include <drm/i915_drm.h>
#include "i915_drv.h"
#include "i915_trace.h"
#include "intel_drv.h"
#include <linux/oom.h>
#include <linux/shmem_fs.h>
#include <linux/slab.h>
#include <linux/swap.h>
#include <linux/pci.h>
#include <linux/dma-buf.h>
static void i915_gem_object_flush_gtt_write_domain(struct drm_i915_gem_object *obj);
static void i915_gem_object_flush_cpu_write_domain(struct drm_i915_gem_object *obj);
static __must_check int
i915_gem_object_wait_rendering(struct drm_i915_gem_object *obj,
bool readonly);
static void
i915_gem_object_retire(struct drm_i915_gem_object *obj);
static void i915_gem_write_fence(struct drm_device *dev, int reg,
struct drm_i915_gem_object *obj);
static void i915_gem_object_update_fence(struct drm_i915_gem_object *obj,
struct drm_i915_fence_reg *fence,
bool enable);
static unsigned long i915_gem_shrinker_count(struct shrinker *shrinker,
struct shrink_control *sc);
static unsigned long i915_gem_shrinker_scan(struct shrinker *shrinker,
struct shrink_control *sc);
static int i915_gem_shrinker_oom(struct notifier_block *nb,
unsigned long event,
void *ptr);
static unsigned long i915_gem_shrink_all(struct drm_i915_private *dev_priv);
static bool cpu_cache_is_coherent(struct drm_device *dev,
enum i915_cache_level level)
{
return HAS_LLC(dev) || level != I915_CACHE_NONE;
}
static bool cpu_write_needs_clflush(struct drm_i915_gem_object *obj)
{
if (!cpu_cache_is_coherent(obj->base.dev, obj->cache_level))
return true;
return obj->pin_display;
}
static inline void i915_gem_object_fence_lost(struct drm_i915_gem_object *obj)
{
if (obj->tiling_mode)
i915_gem_release_mmap(obj);
/* As we do not have an associated fence register, we will force
* a tiling change if we ever need to acquire one.
*/
obj->fence_dirty = false;
obj->fence_reg = I915_FENCE_REG_NONE;
}
/* some bookkeeping */
static void i915_gem_info_add_obj(struct drm_i915_private *dev_priv,
size_t size)
{
spin_lock(&dev_priv->mm.object_stat_lock);
dev_priv->mm.object_count++;
dev_priv->mm.object_memory += size;
spin_unlock(&dev_priv->mm.object_stat_lock);
}
static void i915_gem_info_remove_obj(struct drm_i915_private *dev_priv,
size_t size)
{
spin_lock(&dev_priv->mm.object_stat_lock);
dev_priv->mm.object_count--;
dev_priv->mm.object_memory -= size;
spin_unlock(&dev_priv->mm.object_stat_lock);
}
static int
i915_gem_wait_for_error(struct i915_gpu_error *error)
{
int ret;
#define EXIT_COND (!i915_reset_in_progress(error) || \
i915_terminally_wedged(error))
if (EXIT_COND)
return 0;
/*
* Only wait 10 seconds for the gpu reset to complete to avoid hanging
* userspace. If it takes that long something really bad is going on and
* we should simply try to bail out and fail as gracefully as possible.
*/
ret = wait_event_interruptible_timeout(error->reset_queue,
EXIT_COND,
10*HZ);
if (ret == 0) {
DRM_ERROR("Timed out waiting for the gpu reset to complete\n");
return -EIO;
} else if (ret < 0) {
return ret;
}
#undef EXIT_COND
return 0;
}
int i915_mutex_lock_interruptible(struct drm_device *dev)
{
struct drm_i915_private *dev_priv = dev->dev_private;
int ret;
ret = i915_gem_wait_for_error(&dev_priv->gpu_error);
if (ret)
return ret;
ret = mutex_lock_interruptible(&dev->struct_mutex);
if (ret)
return ret;
WARN_ON(i915_verify_lists(dev));
return 0;
}
int
i915_gem_get_aperture_ioctl(struct drm_device *dev, void *data,
struct drm_file *file)
{
struct drm_i915_private *dev_priv = dev->dev_private;
struct drm_i915_gem_get_aperture *args = data;
struct drm_i915_gem_object *obj;
size_t pinned;
pinned = 0;
mutex_lock(&dev->struct_mutex);
list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list)
if (i915_gem_obj_is_pinned(obj))
pinned += i915_gem_obj_ggtt_size(obj);
mutex_unlock(&dev->struct_mutex);
args->aper_size = dev_priv->gtt.base.total;
args->aper_available_size = args->aper_size - pinned;
return 0;
}
static int
i915_gem_object_get_pages_phys(struct drm_i915_gem_object *obj)
{
struct address_space *mapping = file_inode(obj->base.filp)->i_mapping;
char *vaddr = obj->phys_handle->vaddr;
struct sg_table *st;
struct scatterlist *sg;
int i;
if (WARN_ON(i915_gem_object_needs_bit17_swizzle(obj)))
return -EINVAL;
for (i = 0; i < obj->base.size / PAGE_SIZE; i++) {
struct page *page;
char *src;
page = shmem_read_mapping_page(mapping, i);
if (IS_ERR(page))
return PTR_ERR(page);
src = kmap_atomic(page);
memcpy(vaddr, src, PAGE_SIZE);
drm_clflush_virt_range(vaddr, PAGE_SIZE);
kunmap_atomic(src);
page_cache_release(page);
vaddr += PAGE_SIZE;
}
i915_gem_chipset_flush(obj->base.dev);
st = kmalloc(sizeof(*st), GFP_KERNEL);
if (st == NULL)
return -ENOMEM;
if (sg_alloc_table(st, 1, GFP_KERNEL)) {
kfree(st);
return -ENOMEM;
}
sg = st->sgl;
sg->offset = 0;
sg->length = obj->base.size;
sg_dma_address(sg) = obj->phys_handle->busaddr;
sg_dma_len(sg) = obj->base.size;
obj->pages = st;
obj->has_dma_mapping = true;
return 0;
}
static void
i915_gem_object_put_pages_phys(struct drm_i915_gem_object *obj)
{
int ret;
BUG_ON(obj->madv == __I915_MADV_PURGED);
ret = i915_gem_object_set_to_cpu_domain(obj, true);
if (ret) {
/* In the event of a disaster, abandon all caches and
* hope for the best.
*/
WARN_ON(ret != -EIO);
obj->base.read_domains = obj->base.write_domain = I915_GEM_DOMAIN_CPU;
}
if (obj->madv == I915_MADV_DONTNEED)
obj->dirty = 0;
if (obj->dirty) {
struct address_space *mapping = file_inode(obj->base.filp)->i_mapping;
char *vaddr = obj->phys_handle->vaddr;
int i;
for (i = 0; i < obj->base.size / PAGE_SIZE; i++) {
struct page *page;
char *dst;
page = shmem_read_mapping_page(mapping, i);
if (IS_ERR(page))
continue;
dst = kmap_atomic(page);
drm_clflush_virt_range(vaddr, PAGE_SIZE);
memcpy(dst, vaddr, PAGE_SIZE);
kunmap_atomic(dst);
set_page_dirty(page);
if (obj->madv == I915_MADV_WILLNEED)
mark_page_accessed(page);
page_cache_release(page);
vaddr += PAGE_SIZE;
}
obj->dirty = 0;
}
sg_free_table(obj->pages);
kfree(obj->pages);
obj->has_dma_mapping = false;
}
static void
i915_gem_object_release_phys(struct drm_i915_gem_object *obj)
{
drm_pci_free(obj->base.dev, obj->phys_handle);
}
static const struct drm_i915_gem_object_ops i915_gem_phys_ops = {
.get_pages = i915_gem_object_get_pages_phys,
.put_pages = i915_gem_object_put_pages_phys,
.release = i915_gem_object_release_phys,
};
static int
drop_pages(struct drm_i915_gem_object *obj)
{
struct i915_vma *vma, *next;
int ret;
drm_gem_object_reference(&obj->base);
list_for_each_entry_safe(vma, next, &obj->vma_list, vma_link)
if (i915_vma_unbind(vma))
break;
ret = i915_gem_object_put_pages(obj);
drm_gem_object_unreference(&obj->base);
return ret;
}
int
i915_gem_object_attach_phys(struct drm_i915_gem_object *obj,
int align)
{
drm_dma_handle_t *phys;
int ret;
if (obj->phys_handle) {
if ((unsigned long)obj->phys_handle->vaddr & (align -1))
return -EBUSY;
return 0;
}
if (obj->madv != I915_MADV_WILLNEED)
return -EFAULT;
if (obj->base.filp == NULL)
return -EINVAL;
ret = drop_pages(obj);
if (ret)
return ret;
/* create a new object */
phys = drm_pci_alloc(obj->base.dev, obj->base.size, align);
if (!phys)
return -ENOMEM;
obj->phys_handle = phys;
obj->ops = &i915_gem_phys_ops;
return i915_gem_object_get_pages(obj);
}
static int
i915_gem_phys_pwrite(struct drm_i915_gem_object *obj,
struct drm_i915_gem_pwrite *args,
struct drm_file *file_priv)
{
struct drm_device *dev = obj->base.dev;
void *vaddr = obj->phys_handle->vaddr + args->offset;
char __user *user_data = to_user_ptr(args->data_ptr);
int ret;
/* We manually control the domain here and pretend that it
* remains coherent i.e. in the GTT domain, like shmem_pwrite.
*/
ret = i915_gem_object_wait_rendering(obj, false);
if (ret)
return ret;
if (__copy_from_user_inatomic_nocache(vaddr, user_data, args->size)) {
unsigned long unwritten;
/* The physical object once assigned is fixed for the lifetime
* of the obj, so we can safely drop the lock and continue
* to access vaddr.
*/
mutex_unlock(&dev->struct_mutex);
unwritten = copy_from_user(vaddr, user_data, args->size);
mutex_lock(&dev->struct_mutex);
if (unwritten)
return -EFAULT;
}
drm_clflush_virt_range(vaddr, args->size);
i915_gem_chipset_flush(dev);
return 0;
}
void *i915_gem_object_alloc(struct drm_device *dev)
{
struct drm_i915_private *dev_priv = dev->dev_private;
return kmem_cache_zalloc(dev_priv->slab, GFP_KERNEL);
}
void i915_gem_object_free(struct drm_i915_gem_object *obj)
{
struct drm_i915_private *dev_priv = obj->base.dev->dev_private;
kmem_cache_free(dev_priv->slab, obj);
}
static int
i915_gem_create(struct drm_file *file,
struct drm_device *dev,
uint64_t size,
uint32_t *handle_p)
{
struct drm_i915_gem_object *obj;
int ret;
u32 handle;
size = roundup(size, PAGE_SIZE);
if (size == 0)
return -EINVAL;
/* Allocate the new object */
obj = i915_gem_alloc_object(dev, size);
if (obj == NULL)
return -ENOMEM;
ret = drm_gem_handle_create(file, &obj->base, &handle);
/* drop reference from allocate - handle holds it now */
drm_gem_object_unreference_unlocked(&obj->base);
if (ret)
return ret;
*handle_p = handle;
return 0;
}
int
i915_gem_dumb_create(struct drm_file *file,
struct drm_device *dev,
struct drm_mode_create_dumb *args)
{
/* have to work out size/pitch and return them */
args->pitch = ALIGN(args->width * DIV_ROUND_UP(args->bpp, 8), 64);
args->size = args->pitch * args->height;
return i915_gem_create(file, dev,
args->size, &args->handle);
}
/**
* 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)
{
struct drm_i915_gem_create *args = data;
return i915_gem_create(file, dev,
args->size, &args->handle);
}
static inline int
__copy_to_user_swizzled(char __user *cpu_vaddr,
const char *gpu_vaddr, int gpu_offset,
int length)
{
int ret, cpu_offset = 0;
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;
ret = __copy_to_user(cpu_vaddr + cpu_offset,
gpu_vaddr + swizzled_gpu_offset,
this_length);
if (ret)
return ret + length;
cpu_offset += this_length;
gpu_offset += this_length;
length -= this_length;
}
return 0;
}
static inline int
__copy_from_user_swizzled(char *gpu_vaddr, int gpu_offset,
const char __user *cpu_vaddr,
int length)
{
int ret, cpu_offset = 0;
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;
ret = __copy_from_user(gpu_vaddr + swizzled_gpu_offset,
cpu_vaddr + cpu_offset,
this_length);
if (ret)
return ret + length;
cpu_offset += this_length;
gpu_offset += this_length;
length -= this_length;
}
return 0;
}
/*
* Pins the specified object's pages and synchronizes the object with
* GPU accesses. Sets needs_clflush to non-zero if the caller should
* flush the object from the CPU cache.
*/
int i915_gem_obj_prepare_shmem_read(struct drm_i915_gem_object *obj,
int *needs_clflush)
{
int ret;
*needs_clflush = 0;
if (!obj->base.filp)
return -EINVAL;
if (!(obj->base.read_domains & I915_GEM_DOMAIN_CPU)) {
/* If we're not in the cpu read domain, set ourself into the gtt
* read domain and manually flush cachelines (if required). This
* optimizes for the case when the gpu will dirty the data
* anyway again before the next pread happens. */
*needs_clflush = !cpu_cache_is_coherent(obj->base.dev,
obj->cache_level);
ret = i915_gem_object_wait_rendering(obj, true);
if (ret)
return ret;
i915_gem_object_retire(obj);
}
ret = i915_gem_object_get_pages(obj);
if (ret)
return ret;
i915_gem_object_pin_pages(obj);
return ret;
}
/* Per-page copy function for the shmem pread fastpath.
* Flushes invalid cachelines before reading the target if
* needs_clflush is set. */
static int
shmem_pread_fast(struct page *page, int shmem_page_offset, int page_length,
char __user *user_data,
bool page_do_bit17_swizzling, bool needs_clflush)
{
char *vaddr;
int ret;
if (unlikely(page_do_bit17_swizzling))
return -EINVAL;
vaddr = kmap_atomic(page);
if (needs_clflush)
drm_clflush_virt_range(vaddr + shmem_page_offset,
page_length);
ret = __copy_to_user_inatomic(user_data,
vaddr + shmem_page_offset,
page_length);
kunmap_atomic(vaddr);
return ret ? -EFAULT : 0;
}
static void
shmem_clflush_swizzled_range(char *addr, unsigned long length,
bool swizzled)
{
if (unlikely(swizzled)) {
unsigned long start = (unsigned long) addr;
unsigned long end = (unsigned long) addr + length;
/* For swizzling simply ensure that we always flush both
* channels. Lame, but simple and it works. Swizzled
* pwrite/pread is far from a hotpath - current userspace
* doesn't use it at all. */
start = round_down(start, 128);
end = round_up(end, 128);
drm_clflush_virt_range((void *)start, end - start);
} else {
drm_clflush_virt_range(addr, length);
}
}
/* Only difference to the fast-path function is that this can handle bit17
* and uses non-atomic copy and kmap functions. */
static int
shmem_pread_slow(struct page *page, int shmem_page_offset, int page_length,
char __user *user_data,
bool page_do_bit17_swizzling, bool needs_clflush)
{
char *vaddr;
int ret;
vaddr = kmap(page);
if (needs_clflush)
shmem_clflush_swizzled_range(vaddr + shmem_page_offset,
page_length,
page_do_bit17_swizzling);
if (page_do_bit17_swizzling)
ret = __copy_to_user_swizzled(user_data,
vaddr, shmem_page_offset,
page_length);
else
ret = __copy_to_user(user_data,
vaddr + shmem_page_offset,
page_length);
kunmap(page);
return ret ? - EFAULT : 0;
}
static int
i915_gem_shmem_pread(struct drm_device *dev,
struct drm_i915_gem_object *obj,
struct drm_i915_gem_pread *args,
struct drm_file *file)
{
char __user *user_data;
ssize_t remain;
loff_t offset;
int shmem_page_offset, page_length, ret = 0;
int obj_do_bit17_swizzling, page_do_bit17_swizzling;
int prefaulted = 0;
int needs_clflush = 0;
struct sg_page_iter sg_iter;
user_data = to_user_ptr(args->data_ptr);
remain = args->size;
obj_do_bit17_swizzling = i915_gem_object_needs_bit17_swizzle(obj);
ret = i915_gem_obj_prepare_shmem_read(obj, &needs_clflush);
if (ret)
return ret;
offset = args->offset;
for_each_sg_page(obj->pages->sgl, &sg_iter, obj->pages->nents,
offset >> PAGE_SHIFT) {
struct page *page = sg_page_iter_page(&sg_iter);
if (remain <= 0)
break;
/* Operation in this page
*
* shmem_page_offset = offset within page in shmem file
* page_length = bytes to copy for this page
*/
shmem_page_offset = offset_in_page(offset);
page_length = remain;
if ((shmem_page_offset + page_length) > PAGE_SIZE)
page_length = PAGE_SIZE - shmem_page_offset;
page_do_bit17_swizzling = obj_do_bit17_swizzling &&
(page_to_phys(page) & (1 << 17)) != 0;
ret = shmem_pread_fast(page, shmem_page_offset, page_length,
user_data, page_do_bit17_swizzling,
needs_clflush);
if (ret == 0)
goto next_page;
mutex_unlock(&dev->struct_mutex);
if (likely(!i915.prefault_disable) && !prefaulted) {
ret = fault_in_multipages_writeable(user_data, remain);
/* Userspace is tricking us, but we've already clobbered
* its pages with the prefault and promised to write the
* data up to the first fault. Hence ignore any errors
* and just continue. */
(void)ret;
prefaulted = 1;
}
ret = shmem_pread_slow(page, shmem_page_offset, page_length,
user_data, page_do_bit17_swizzling,
needs_clflush);
mutex_lock(&dev->struct_mutex);
if (ret)
goto out;
next_page:
remain -= page_length;
user_data += page_length;
offset += page_length;
}
out:
i915_gem_object_unpin_pages(obj);
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)
{
struct drm_i915_gem_pread *args = data;
struct drm_i915_gem_object *obj;
int ret = 0;
if (args->size == 0)
return 0;
if (!access_ok(VERIFY_WRITE,
to_user_ptr(args->data_ptr),
args->size))
return -EFAULT;
ret = i915_mutex_lock_interruptible(dev);
if (ret)
return ret;
obj = to_intel_bo(drm_gem_object_lookup(dev, file, args->handle));
if (&obj->base == NULL) {
ret = -ENOENT;
goto unlock;
}
/* Bounds check source. */
if (args->offset > obj->base.size ||
args->size > obj->base.size - args->offset) {
ret = -EINVAL;
goto out;
}
/* prime objects have no backing filp to GEM pread/pwrite
* pages from.
*/
if (!obj->base.filp) {
ret = -EINVAL;
goto out;
}
trace_i915_gem_object_pread(obj, args->offset, args->size);
ret = i915_gem_shmem_pread(dev, obj, args, file);
out:
drm_gem_object_unreference(&obj->base);
unlock:
mutex_unlock(&dev->struct_mutex);
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)
{
void __iomem *vaddr_atomic;
void *vaddr;
unsigned long unwritten;
vaddr_atomic = io_mapping_map_atomic_wc(mapping, page_base);
/* We can use the cpu mem copy function because this is X86. */
vaddr = (void __force*)vaddr_atomic + page_offset;
unwritten = __copy_from_user_inatomic_nocache(vaddr,
user_data, length);
io_mapping_unmap_atomic(vaddr_atomic);
return unwritten;
}
/**
* 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_i915_gem_object *obj,
struct drm_i915_gem_pwrite *args,
struct drm_file *file)
{
struct drm_i915_private *dev_priv = dev->dev_private;
ssize_t remain;
loff_t offset, page_base;
char __user *user_data;
int page_offset, page_length, ret;
ret = i915_gem_obj_ggtt_pin(obj, 0, PIN_MAPPABLE | PIN_NONBLOCK);
if (ret)
goto out;
ret = i915_gem_object_set_to_gtt_domain(obj, true);
if (ret)
goto out_unpin;
ret = i915_gem_object_put_fence(obj);
if (ret)
goto out_unpin;
user_data = to_user_ptr(args->data_ptr);
remain = args->size;
offset = i915_gem_obj_ggtt_offset(obj) + 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_MASK;
page_offset = offset_in_page(offset);
page_length = remain;
if ((page_offset + remain) > PAGE_SIZE)
page_length = PAGE_SIZE - page_offset;
/* 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 (fast_user_write(dev_priv->gtt.mappable, page_base,
page_offset, user_data, page_length)) {
ret = -EFAULT;
goto out_unpin;
}
remain -= page_length;
user_data += page_length;
offset += page_length;
}
out_unpin:
i915_gem_object_ggtt_unpin(obj);
out:
return ret;
}
/* Per-page copy function for the shmem pwrite fastpath.
* Flushes invalid cachelines before writing to the target if
* needs_clflush_before is set and flushes out any written cachelines after
* writing if needs_clflush is set. */
static int
shmem_pwrite_fast(struct page *page, int shmem_page_offset, int page_length,
char __user *user_data,
bool page_do_bit17_swizzling,
bool needs_clflush_before,
bool needs_clflush_after)
{
char *vaddr;
int ret;
if (unlikely(page_do_bit17_swizzling))
return -EINVAL;
vaddr = kmap_atomic(page);
if (needs_clflush_before)
drm_clflush_virt_range(vaddr + shmem_page_offset,
page_length);
ret = __copy_from_user_inatomic(vaddr + shmem_page_offset,
user_data, page_length);
if (needs_clflush_after)
drm_clflush_virt_range(vaddr + shmem_page_offset,
page_length);
kunmap_atomic(vaddr);
return ret ? -EFAULT : 0;
}
/* Only difference to the fast-path function is that this can handle bit17
* and uses non-atomic copy and kmap functions. */
static int
shmem_pwrite_slow(struct page *page, int shmem_page_offset, int page_length,
char __user *user_data,
bool page_do_bit17_swizzling,
bool needs_clflush_before,
bool needs_clflush_after)
{
char *vaddr;
int ret;
vaddr = kmap(page);
if (unlikely(needs_clflush_before || page_do_bit17_swizzling))
shmem_clflush_swizzled_range(vaddr + shmem_page_offset,
page_length,
page_do_bit17_swizzling);
if (page_do_bit17_swizzling)
ret = __copy_from_user_swizzled(vaddr, shmem_page_offset,
user_data,
page_length);
else
ret = __copy_from_user(vaddr + shmem_page_offset,
user_data,
page_length);
if (needs_clflush_after)
shmem_clflush_swizzled_range(vaddr + shmem_page_offset,
page_length,
page_do_bit17_swizzling);
kunmap(page);
return ret ? -EFAULT : 0;
}
static int
i915_gem_shmem_pwrite(struct drm_device *dev,
struct drm_i915_gem_object *obj,
struct drm_i915_gem_pwrite *args,
struct drm_file *file)
{
ssize_t remain;
loff_t offset;
char __user *user_data;
int shmem_page_offset, page_length, ret = 0;
int obj_do_bit17_swizzling, page_do_bit17_swizzling;
int hit_slowpath = 0;
int needs_clflush_after = 0;
int needs_clflush_before = 0;
struct sg_page_iter sg_iter;
user_data = to_user_ptr(args->data_ptr);
remain = args->size;
obj_do_bit17_swizzling = i915_gem_object_needs_bit17_swizzle(obj);
if (obj->base.write_domain != I915_GEM_DOMAIN_CPU) {
/* If we're not in the cpu write domain, set ourself into the gtt
* write domain and manually flush cachelines (if required). This
* optimizes for the case when the gpu will use the data
* right away and we therefore have to clflush anyway. */
needs_clflush_after = cpu_write_needs_clflush(obj);
ret = i915_gem_object_wait_rendering(obj, false);
if (ret)
return ret;
i915_gem_object_retire(obj);
}
/* Same trick applies to invalidate partially written cachelines read
* before writing. */
if ((obj->base.read_domains & I915_GEM_DOMAIN_CPU) == 0)
needs_clflush_before =
!cpu_cache_is_coherent(dev, obj->cache_level);
ret = i915_gem_object_get_pages(obj);
if (ret)
return ret;
i915_gem_object_pin_pages(obj);
offset = args->offset;
obj->dirty = 1;
for_each_sg_page(obj->pages->sgl, &sg_iter, obj->pages->nents,
offset >> PAGE_SHIFT) {
struct page *page = sg_page_iter_page(&sg_iter);
int partial_cacheline_write;
if (remain <= 0)
break;
/* Operation in this page
*
* shmem_page_offset = offset within page in shmem file
* page_length = bytes to copy for this page
*/
shmem_page_offset = offset_in_page(offset);
page_length = remain;
if ((shmem_page_offset + page_length) > PAGE_SIZE)
page_length = PAGE_SIZE - shmem_page_offset;
/* If we don't overwrite a cacheline completely we need to be
* careful to have up-to-date data by first clflushing. Don't
* overcomplicate things and flush the entire patch. */
partial_cacheline_write = needs_clflush_before &&
((shmem_page_offset | page_length)
& (boot_cpu_data.x86_clflush_size - 1));
page_do_bit17_swizzling = obj_do_bit17_swizzling &&
(page_to_phys(page) & (1 << 17)) != 0;
ret = shmem_pwrite_fast(page, shmem_page_offset, page_length,
user_data, page_do_bit17_swizzling,
partial_cacheline_write,
needs_clflush_after);
if (ret == 0)
goto next_page;
hit_slowpath = 1;
mutex_unlock(&dev->struct_mutex);
ret = shmem_pwrite_slow(page, shmem_page_offset, page_length,
user_data, page_do_bit17_swizzling,
partial_cacheline_write,
needs_clflush_after);
mutex_lock(&dev->struct_mutex);
if (ret)
goto out;
next_page:
remain -= page_length;
user_data += page_length;
offset += page_length;
}
out:
i915_gem_object_unpin_pages(obj);
if (hit_slowpath) {
/*
* Fixup: Flush cpu caches in case we didn't flush the dirty
* cachelines in-line while writing and the object moved
* out of the cpu write domain while we've dropped the lock.
*/
if (!needs_clflush_after &&
obj->base.write_domain != I915_GEM_DOMAIN_CPU) {
if (i915_gem_clflush_object(obj, obj->pin_display))
i915_gem_chipset_flush(dev);
}
}
if (needs_clflush_after)
i915_gem_chipset_flush(dev);
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)
{
struct drm_i915_private *dev_priv = dev->dev_private;
struct drm_i915_gem_pwrite *args = data;
struct drm_i915_gem_object *obj;
int ret;
if (args->size == 0)
return 0;
if (!access_ok(VERIFY_READ,
to_user_ptr(args->data_ptr),
args->size))
return -EFAULT;
if (likely(!i915.prefault_disable)) {
ret = fault_in_multipages_readable(to_user_ptr(args->data_ptr),
args->size);
if (ret)
return -EFAULT;
}
intel_runtime_pm_get(dev_priv);
ret = i915_mutex_lock_interruptible(dev);
if (ret)
goto put_rpm;
obj = to_intel_bo(drm_gem_object_lookup(dev, file, args->handle));
if (&obj->base == NULL) {
ret = -ENOENT;
goto unlock;
}
/* Bounds check destination. */
if (args->offset > obj->base.size ||
args->size > obj->base.size - args->offset) {
ret = -EINVAL;
goto out;
}
/* prime objects have no backing filp to GEM pread/pwrite
* pages from.
*/
if (!obj->base.filp) {
ret = -EINVAL;
goto out;
}
trace_i915_gem_object_pwrite(obj, args->offset, args->size);
ret = -EFAULT;
/* 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->tiling_mode == I915_TILING_NONE &&
obj->base.write_domain != I915_GEM_DOMAIN_CPU &&
cpu_write_needs_clflush(obj)) {
ret = i915_gem_gtt_pwrite_fast(dev, obj, args, file);
/* Note that the gtt paths might fail with non-page-backed user
* pointers (e.g. gtt mappings when moving data between
* textures). Fallback to the shmem path in that case. */
}
if (ret == -EFAULT || ret == -ENOSPC) {
if (obj->phys_handle)
ret = i915_gem_phys_pwrite(obj, args, file);
else
ret = i915_gem_shmem_pwrite(dev, obj, args, file);
}
out:
drm_gem_object_unreference(&obj->base);
unlock:
mutex_unlock(&dev->struct_mutex);
put_rpm:
intel_runtime_pm_put(dev_priv);
return ret;
}
int
i915_gem_check_wedge(struct i915_gpu_error *error,
bool interruptible)
{
if (i915_reset_in_progress(error)) {
/* Non-interruptible callers can't handle -EAGAIN, hence return
* -EIO unconditionally for these. */
if (!interruptible)
return -EIO;
/* Recovery complete, but the reset failed ... */
if (i915_terminally_wedged(error))
return -EIO;
/*
* Check if GPU Reset is in progress - we need intel_ring_begin
* to work properly to reinit the hw state while the gpu is
* still marked as reset-in-progress. Handle this with a flag.
*/
if (!error->reload_in_reset)
return -EAGAIN;
}
return 0;
}
/*
* Compare arbitrary request against outstanding lazy request. Emit on match.
*/
int
i915_gem_check_olr(struct drm_i915_gem_request *req)
{
int ret;
WARN_ON(!mutex_is_locked(&req->ring->dev->struct_mutex));
ret = 0;
if (req == req->ring->outstanding_lazy_request)
ret = i915_add_request(req->ring);
return ret;
}
static void fake_irq(unsigned long data)
{
wake_up_process((struct task_struct *)data);
}
static bool missed_irq(struct drm_i915_private *dev_priv,
struct intel_engine_cs *ring)
{
return test_bit(ring->id, &dev_priv->gpu_error.missed_irq_rings);
}
static bool can_wait_boost(struct drm_i915_file_private *file_priv)
{
if (file_priv == NULL)
return true;
return !atomic_xchg(&file_priv->rps_wait_boost, true);
}
/**
* __i915_wait_request - wait until execution of request has finished
* @req: duh!
* @reset_counter: reset sequence associated with the given request
* @interruptible: do an interruptible wait (normally yes)
* @timeout: in - how long to wait (NULL forever); out - how much time remaining
*
* Note: It is of utmost importance that the passed in seqno and reset_counter
* values have been read by the caller in an smp safe manner. Where read-side
* locks are involved, it is sufficient to read the reset_counter before
* unlocking the lock that protects the seqno. For lockless tricks, the
* reset_counter _must_ be read before, and an appropriate smp_rmb must be
* inserted.
*
* Returns 0 if the request was found within the alloted time. Else returns the
* errno with remaining time filled in timeout argument.
*/
int __i915_wait_request(struct drm_i915_gem_request *req,
unsigned reset_counter,
bool interruptible,
s64 *timeout,
struct drm_i915_file_private *file_priv)
{
struct intel_engine_cs *ring = i915_gem_request_get_ring(req);
struct drm_device *dev = ring->dev;
struct drm_i915_private *dev_priv = dev->dev_private;
const bool irq_test_in_progress =
ACCESS_ONCE(dev_priv->gpu_error.test_irq_rings) & intel_ring_flag(ring);
DEFINE_WAIT(wait);
unsigned long timeout_expire;
s64 before, now;
int ret;
WARN(!intel_irqs_enabled(dev_priv), "IRQs disabled");
if (i915_gem_request_completed(req, true))
return 0;
timeout_expire = timeout ?
jiffies + nsecs_to_jiffies_timeout((u64)*timeout) : 0;
if (INTEL_INFO(dev)->gen >= 6 && ring->id == RCS && can_wait_boost(file_priv)) {
gen6_rps_boost(dev_priv);
if (file_priv)
mod_delayed_work(dev_priv->wq,
&file_priv->mm.idle_work,
msecs_to_jiffies(100));
}
if (!irq_test_in_progress && WARN_ON(!ring->irq_get(ring)))
return -ENODEV;
/* Record current time in case interrupted by signal, or wedged */
trace_i915_gem_request_wait_begin(req);
before = ktime_get_raw_ns();
for (;;) {
struct timer_list timer;
prepare_to_wait(&ring->irq_queue, &wait,
interruptible ? TASK_INTERRUPTIBLE : TASK_UNINTERRUPTIBLE);
/* We need to check whether any gpu reset happened in between
* the caller grabbing the seqno and now ... */
if (reset_counter != atomic_read(&dev_priv->gpu_error.reset_counter)) {
/* ... but upgrade the -EAGAIN to an -EIO if the gpu
* is truely gone. */
ret = i915_gem_check_wedge(&dev_priv->gpu_error, interruptible);
if (ret == 0)
ret = -EAGAIN;
break;
}
if (i915_gem_request_completed(req, false)) {
ret = 0;
break;
}
if (interruptible && signal_pending(current)) {
ret = -ERESTARTSYS;
break;
}
if (timeout && time_after_eq(jiffies, timeout_expire)) {
ret = -ETIME;
break;
}
timer.function = NULL;
if (timeout || missed_irq(dev_priv, ring)) {
unsigned long expire;
setup_timer_on_stack(&timer, fake_irq, (unsigned long)current);
expire = missed_irq(dev_priv, ring) ? jiffies + 1 : timeout_expire;
mod_timer(&timer, expire);
}
io_schedule();
if (timer.function) {
del_singleshot_timer_sync(&timer);
destroy_timer_on_stack(&timer);
}
}
now = ktime_get_raw_ns();
trace_i915_gem_request_wait_end(req);
if (!irq_test_in_progress)
ring->irq_put(ring);
finish_wait(&ring->irq_queue, &wait);
if (timeout) {
s64 tres = *timeout - (now - before);
*timeout = tres < 0 ? 0 : tres;
/*
* Apparently ktime isn't accurate enough and occasionally has a
* bit of mismatch in the jiffies<->nsecs<->ktime loop. So patch
* things up to make the test happy. We allow up to 1 jiffy.
*
* This is a regrssion from the timespec->ktime conversion.
*/
if (ret == -ETIME && *timeout < jiffies_to_usecs(1)*1000)
*timeout = 0;
}
return ret;
}
/**
* Waits for a request to be signaled, and cleans up the
* request and object lists appropriately for that event.
*/
int
i915_wait_request(struct drm_i915_gem_request *req)
{
struct drm_device *dev;
struct drm_i915_private *dev_priv;
bool interruptible;
unsigned reset_counter;
int ret;
BUG_ON(req == NULL);
dev = req->ring->dev;
dev_priv = dev->dev_private;
interruptible = dev_priv->mm.interruptible;
BUG_ON(!mutex_is_locked(&dev->struct_mutex));
ret = i915_gem_check_wedge(&dev_priv->gpu_error, interruptible);
if (ret)
return ret;
ret = i915_gem_check_olr(req);
if (ret)
return ret;
reset_counter = atomic_read(&dev_priv->gpu_error.reset_counter);
i915_gem_request_reference(req);
ret = __i915_wait_request(req, reset_counter,
interruptible, NULL, NULL);
i915_gem_request_unreference(req);
return ret;
}
static int
i915_gem_object_wait_rendering__tail(struct drm_i915_gem_object *obj)
{
if (!obj->active)
return 0;
/* Manually manage the write flush as we may have not yet
* retired the buffer.
*
* Note that the last_write_req is always the earlier of
* the two (read/write) requests, so if we haved successfully waited,
* we know we have passed the last write.
*/
i915_gem_request_assign(&obj->last_write_req, NULL);
return 0;
}
/**
* 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 __must_check int
i915_gem_object_wait_rendering(struct drm_i915_gem_object *obj,
bool readonly)
{
struct drm_i915_gem_request *req;
int ret;
req = readonly ? obj->last_write_req : obj->last_read_req;
if (!req)
return 0;
ret = i915_wait_request(req);
if (ret)
return ret;
return i915_gem_object_wait_rendering__tail(obj);
}
/* A nonblocking variant of the above wait. This is a highly dangerous routine
* as the object state may change during this call.
*/
static __must_check int
i915_gem_object_wait_rendering__nonblocking(struct drm_i915_gem_object *obj,
struct drm_i915_file_private *file_priv,
bool readonly)
{
struct drm_i915_gem_request *req;
struct drm_device *dev = obj->base.dev;
struct drm_i915_private *dev_priv = dev->dev_private;
unsigned reset_counter;
int ret;
BUG_ON(!mutex_is_locked(&dev->struct_mutex));
BUG_ON(!dev_priv->mm.interruptible);
req = readonly ? obj->last_write_req : obj->last_read_req;
if (!req)
return 0;
ret = i915_gem_check_wedge(&dev_priv->gpu_error, true);
if (ret)
return ret;
ret = i915_gem_check_olr(req);
if (ret)
return ret;
reset_counter = atomic_read(&dev_priv->gpu_error.reset_counter);
i915_gem_request_reference(req);
mutex_unlock(&dev->struct_mutex);
ret = __i915_wait_request(req, reset_counter, true, NULL, file_priv);
mutex_lock(&dev->struct_mutex);
i915_gem_request_unreference(req);
if (ret)
return ret;
return i915_gem_object_wait_rendering__tail(obj);
}
/**
* 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)
{
struct drm_i915_gem_set_domain *args = data;
struct drm_i915_gem_object *obj;
uint32_t read_domains = args->read_domains;
uint32_t write_domain = args->write_domain;
int ret;
/* 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;
ret = i915_mutex_lock_interruptible(dev);
if (ret)
return ret;
obj = to_intel_bo(drm_gem_object_lookup(dev, file, args->handle));
if (&obj->base == NULL) {
ret = -ENOENT;
goto unlock;
}
/* Try to flush the object off the GPU without holding the lock.
* We will repeat the flush holding the lock in the normal manner
* to catch cases where we are gazumped.
*/
ret = i915_gem_object_wait_rendering__nonblocking(obj,
file->driver_priv,
!write_domain);
if (ret)
goto unref;
if (read_domains & I915_GEM_DOMAIN_GTT)
ret = i915_gem_object_set_to_gtt_domain(obj, write_domain != 0);
else
ret = i915_gem_object_set_to_cpu_domain(obj, write_domain != 0);
unref:
drm_gem_object_unreference(&obj->base);
unlock:
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)
{
struct drm_i915_gem_sw_finish *args = data;
struct drm_i915_gem_object *obj;
int ret = 0;
ret = i915_mutex_lock_interruptible(dev);
if (ret)
return ret;
obj = to_intel_bo(drm_gem_object_lookup(dev, file, args->handle));
if (&obj->base == NULL) {
ret = -ENOENT;
goto unlock;
}
/* Pinned buffers may be scanout, so flush the cache */
if (obj->pin_display)
i915_gem_object_flush_cpu_write_domain(obj);
drm_gem_object_unreference(&obj->base);
unlock:
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.
*
* IMPORTANT:
*
* DRM driver writers who look a this function as an example for how to do GEM
* mmap support, please don't implement mmap support like here. The modern way
* to implement DRM mmap support is with an mmap offset ioctl (like
* i915_gem_mmap_gtt) and then using the mmap syscall on the DRM fd directly.
* That way debug tooling like valgrind will understand what's going on, hiding
* the mmap call in a driver private ioctl will break that. The i915 driver only
* does cpu mmaps this way because we didn't know better.
*/
int
i915_gem_mmap_ioctl(struct drm_device *dev, void *data,
struct drm_file *file)
{
struct drm_i915_gem_mmap *args = data;
struct drm_gem_object *obj;
unsigned long addr;
if (args->flags & ~(I915_MMAP_WC))
return -EINVAL;
if (args->flags & I915_MMAP_WC && !cpu_has_pat)
return -ENODEV;
obj = drm_gem_object_lookup(dev, file, args->handle);
if (obj == NULL)
return -ENOENT;
/* prime objects have no backing filp to GEM mmap
* pages from.
*/
if (!obj->filp) {
drm_gem_object_unreference_unlocked(obj);
return -EINVAL;
}
addr = vm_mmap(obj->filp, 0, args->size,
PROT_READ | PROT_WRITE, MAP_SHARED,
args->offset);
if (args->flags & I915_MMAP_WC) {
struct mm_struct *mm = current->mm;
struct vm_area_struct *vma;
down_write(&mm->mmap_sem);
vma = find_vma(mm, addr);
if (vma)
vma->vm_page_prot =
pgprot_writecombine(vm_get_page_prot(vma->vm_flags));
else
addr = -ENOMEM;
up_write(&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_i915_gem_object *obj = to_intel_bo(vma->vm_private_data);
struct drm_device *dev = obj->base.dev;
struct drm_i915_private *dev_priv = dev->dev_private;
pgoff_t page_offset;
unsigned long pfn;
int ret = 0;
bool write = !!(vmf->flags & FAULT_FLAG_WRITE);
intel_runtime_pm_get(dev_priv);
/* We don't use vmf->pgoff since that has the fake offset */
page_offset = ((unsigned long)vmf->virtual_address - vma->vm_start) >>
PAGE_SHIFT;
ret = i915_mutex_lock_interruptible(dev);
if (ret)
goto out;
trace_i915_gem_object_fault(obj, page_offset, true, write);
/* Try to flush the object off the GPU first without holding the lock.
* Upon reacquiring the lock, we will perform our sanity checks and then
* repeat the flush holding the lock in the normal manner to catch cases
* where we are gazumped.
*/
ret = i915_gem_object_wait_rendering__nonblocking(obj, NULL, !write);
if (ret)
goto unlock;
/* Access to snoopable pages through the GTT is incoherent. */
if (obj->cache_level != I915_CACHE_NONE && !HAS_LLC(dev)) {
ret = -EFAULT;
goto unlock;
}
/* Now bind it into the GTT if needed */
ret = i915_gem_obj_ggtt_pin(obj, 0, PIN_MAPPABLE);
if (ret)
goto unlock;
ret = i915_gem_object_set_to_gtt_domain(obj, write);
if (ret)
goto unpin;
ret = i915_gem_object_get_fence(obj);
if (ret)
goto unpin;
/* Finally, remap it using the new GTT offset */
pfn = dev_priv->gtt.mappable_base + i915_gem_obj_ggtt_offset(obj);
pfn >>= PAGE_SHIFT;
if (!obj->fault_mappable) {
unsigned long size = min_t(unsigned long,
vma->vm_end - vma->vm_start,
obj->base.size);
int i;
for (i = 0; i < size >> PAGE_SHIFT; i++) {
ret = vm_insert_pfn(vma,
(unsigned long)vma->vm_start + i * PAGE_SIZE,
pfn + i);
if (ret)
break;
}
obj->fault_mappable = true;
} else
ret = vm_insert_pfn(vma,
(unsigned long)vmf->virtual_address,
pfn + page_offset);
unpin:
i915_gem_object_ggtt_unpin(obj);
unlock:
mutex_unlock(&dev->struct_mutex);
out:
switch (ret) {
case -EIO:
/*
* We eat errors when the gpu is terminally wedged to avoid
* userspace unduly crashing (gl has no provisions for mmaps to
* fail). But any other -EIO isn't ours (e.g. swap in failure)
* and so needs to be reported.
*/
if (!i915_terminally_wedged(&dev_priv->gpu_error)) {
ret = VM_FAULT_SIGBUS;
break;
}
case -EAGAIN:
/*
* EAGAIN means the gpu is hung and we'll wait for the error
* handler to reset everything when re-faulting in
* i915_mutex_lock_interruptible.
*/
case 0:
case -ERESTARTSYS:
case -EINTR:
case -EBUSY:
/*
* EBUSY is ok: this just means that another thread
* already did the job.
*/
ret = VM_FAULT_NOPAGE;
break;
case -ENOMEM:
ret = VM_FAULT_OOM;
break;
case -ENOSPC:
case -EFAULT:
ret = VM_FAULT_SIGBUS;
break;
default:
WARN_ONCE(ret, "unhandled error in i915_gem_fault: %i\n", ret);
ret = VM_FAULT_SIGBUS;
break;
}
intel_runtime_pm_put(dev_priv);
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_i915_gem_object *obj)
{
if (!obj->fault_mappable)
return;
drm_vma_node_unmap(&obj->base.vma_node,
obj->base.dev->anon_inode->i_mapping);
obj->fault_mappable = false;
}
void
i915_gem_release_all_mmaps(struct drm_i915_private *dev_priv)
{
struct drm_i915_gem_object *obj;
list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list)
i915_gem_release_mmap(obj);
}
uint32_t
i915_gem_get_gtt_size(struct drm_device *dev, uint32_t size, int tiling_mode)
{
uint32_t gtt_size;
if (INTEL_INFO(dev)->gen >= 4 ||
tiling_mode == I915_TILING_NONE)
return size;
/* Previous chips need a power-of-two fence region when tiling */
if (INTEL_INFO(dev)->gen == 3)
gtt_size = 1024*1024;
else
gtt_size = 512*1024;
while (gtt_size < size)
gtt_size <<= 1;
return gtt_size;
}
/**
* 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.
*/
uint32_t
i915_gem_get_gtt_alignment(struct drm_device *dev, uint32_t size,
int tiling_mode, bool fenced)
{
/*
* 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 || (!fenced && IS_G33(dev)) ||
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.
*/
return i915_gem_get_gtt_size(dev, size, tiling_mode);
}
static int i915_gem_object_create_mmap_offset(struct drm_i915_gem_object *obj)
{
struct drm_i915_private *dev_priv = obj->base.dev->dev_private;
int ret;
if (drm_vma_node_has_offset(&obj->base.vma_node))
return 0;
dev_priv->mm.shrinker_no_lock_stealing = true;
ret = drm_gem_create_mmap_offset(&obj->base);
if (ret != -ENOSPC)
goto out;
/* Badly fragmented mmap space? The only way we can recover
* space is by destroying unwanted objects. We can't randomly release
* mmap_offsets as userspace expects them to be persistent for the
* lifetime of the objects. The closest we can is to release the
* offsets on purgeable objects by truncating it and marking it purged,
* which prevents userspace from ever using that object again.
*/
i915_gem_shrink(dev_priv,
obj->base.size >> PAGE_SHIFT,
I915_SHRINK_BOUND |
I915_SHRINK_UNBOUND |
I915_SHRINK_PURGEABLE);
ret = drm_gem_create_mmap_offset(&obj->base);
if (ret != -ENOSPC)
goto out;
i915_gem_shrink_all(dev_priv);
ret = drm_gem_create_mmap_offset(&obj->base);
out:
dev_priv->mm.shrinker_no_lock_stealing = false;
return ret;
}
static void i915_gem_object_free_mmap_offset(struct drm_i915_gem_object *obj)
{
drm_gem_free_mmap_offset(&obj->base);
}
int
i915_gem_mmap_gtt(struct drm_file *file,
struct drm_device *dev,
uint32_t handle,
uint64_t *offset)
{
struct drm_i915_private *dev_priv = dev->dev_private;
struct drm_i915_gem_object *obj;
int ret;
ret = i915_mutex_lock_interruptible(dev);
if (ret)
return ret;
obj = to_intel_bo(drm_gem_object_lookup(dev, file, handle));
if (&obj->base == NULL) {
ret = -ENOENT;
goto unlock;
}
if (obj->base.size > dev_priv->gtt.mappable_end) {
ret = -E2BIG;
goto out;
}
if (obj->madv != I915_MADV_WILLNEED) {
DRM_DEBUG("Attempting to mmap a purgeable buffer\n");
ret = -EFAULT;
goto out;
}
ret = i915_gem_object_create_mmap_offset(obj);
if (ret)
goto out;
*offset = drm_vma_node_offset_addr(&obj->base.vma_node);
out:
drm_gem_object_unreference(&obj->base);
unlock:
mutex_unlock(&dev->struct_mutex);
return ret;
}
/**
* i915_gem_mmap_gtt_ioctl - prepare an object for GTT mmap'ing
* @dev: DRM device
* @data: GTT mapping ioctl data
* @file: 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)
{
struct drm_i915_gem_mmap_gtt *args = data;
return i915_gem_mmap_gtt(file, dev, args->handle, &args->offset);
}
static inline int
i915_gem_object_is_purgeable(struct drm_i915_gem_object *obj)
{
return obj->madv == I915_MADV_DONTNEED;
}
/* Immediately discard the backing storage */
static void
i915_gem_object_truncate(struct drm_i915_gem_object *obj)
{
i915_gem_object_free_mmap_offset(obj);
if (obj->base.filp == NULL)
return;
/* 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*.
*/
shmem_truncate_range(file_inode(obj->base.filp), 0, (loff_t)-1);
obj->madv = __I915_MADV_PURGED;
}
/* Try to discard unwanted pages */
static void
i915_gem_object_invalidate(struct drm_i915_gem_object *obj)
{
struct address_space *mapping;
switch (obj->madv) {
case I915_MADV_DONTNEED:
i915_gem_object_truncate(obj);
case __I915_MADV_PURGED:
return;
}
if (obj->base.filp == NULL)
return;
mapping = file_inode(obj->base.filp)->i_mapping,
invalidate_mapping_pages(mapping, 0, (loff_t)-1);
}
static void
i915_gem_object_put_pages_gtt(struct drm_i915_gem_object *obj)
{
struct sg_page_iter sg_iter;
int ret;
BUG_ON(obj->madv == __I915_MADV_PURGED);
ret = i915_gem_object_set_to_cpu_domain(obj, true);
if (ret) {
/* In the event of a disaster, abandon all caches and
* hope for the best.
*/
WARN_ON(ret != -EIO);
i915_gem_clflush_object(obj, true);
obj->base.read_domains = obj->base.write_domain = I915_GEM_DOMAIN_CPU;
}
if (i915_gem_object_needs_bit17_swizzle(obj))
i915_gem_object_save_bit_17_swizzle(obj);
if (obj->madv == I915_MADV_DONTNEED)
obj->dirty = 0;
for_each_sg_page(obj->pages->sgl, &sg_iter, obj->pages->nents, 0) {
struct page *page = sg_page_iter_page(&sg_iter);
if (obj->dirty)
set_page_dirty(page);
if (obj->madv == I915_MADV_WILLNEED)
mark_page_accessed(page);
page_cache_release(page);
}
obj->dirty = 0;
sg_free_table(obj->pages);
kfree(obj->pages);
}
int
i915_gem_object_put_pages(struct drm_i915_gem_object *obj)
{
const struct drm_i915_gem_object_ops *ops = obj->ops;
if (obj->pages == NULL)
return 0;
if (obj->pages_pin_count)
return -EBUSY;
BUG_ON(i915_gem_obj_bound_any(obj));
/* ->put_pages might need to allocate memory for the bit17 swizzle
* array, hence protect them from being reaped by removing them from gtt
* lists early. */
list_del(&obj->global_list);
ops->put_pages(obj);
obj->pages = NULL;
i915_gem_object_invalidate(obj);
return 0;
}
unsigned long
i915_gem_shrink(struct drm_i915_private *dev_priv,
long target, unsigned flags)
{
const struct {
struct list_head *list;
unsigned int bit;
} phases[] = {
{ &dev_priv->mm.unbound_list, I915_SHRINK_UNBOUND },
{ &dev_priv->mm.bound_list, I915_SHRINK_BOUND },
{ NULL, 0 },
}, *phase;
unsigned long count = 0;
/*
* As we may completely rewrite the (un)bound list whilst unbinding
* (due to retiring requests) we have to strictly process only
* one element of the list at the time, and recheck the list
* on every iteration.
*
* In particular, we must hold a reference whilst removing the
* object as we may end up waiting for and/or retiring the objects.
* This might release the final reference (held by the active list)
* and result in the object being freed from under us. This is
* similar to the precautions the eviction code must take whilst
* removing objects.
*
* Also note that although these lists do not hold a reference to
* the object we can safely grab one here: The final object
* unreferencing and the bound_list are both protected by the
* dev->struct_mutex and so we won't ever be able to observe an
* object on the bound_list with a reference count equals 0.
*/
for (phase = phases; phase->list; phase++) {
struct list_head still_in_list;
if ((flags & phase->bit) == 0)
continue;
INIT_LIST_HEAD(&still_in_list);
while (count < target && !list_empty(phase->list)) {
struct drm_i915_gem_object *obj;
struct i915_vma *vma, *v;
obj = list_first_entry(phase->list,
typeof(*obj), global_list);
list_move_tail(&obj->global_list, &still_in_list);
if (flags & I915_SHRINK_PURGEABLE &&
!i915_gem_object_is_purgeable(obj))
continue;
drm_gem_object_reference(&obj->base);
/* For the unbound phase, this should be a no-op! */
list_for_each_entry_safe(vma, v,
&obj->vma_list, vma_link)
if (i915_vma_unbind(vma))
break;
if (i915_gem_object_put_pages(obj) == 0)
count += obj->base.size >> PAGE_SHIFT;
drm_gem_object_unreference(&obj->base);
}
list_splice(&still_in_list, phase->list);
}
return count;
}
static unsigned long
i915_gem_shrink_all(struct drm_i915_private *dev_priv)
{
i915_gem_evict_everything(dev_priv->dev);
return i915_gem_shrink(dev_priv, LONG_MAX,
I915_SHRINK_BOUND | I915_SHRINK_UNBOUND);
}
static int
i915_gem_object_get_pages_gtt(struct drm_i915_gem_object *obj)
{
struct drm_i915_private *dev_priv = obj->base.dev->dev_private;
int page_count, i;
struct address_space *mapping;
struct sg_table *st;
struct scatterlist *sg;
struct sg_page_iter sg_iter;
struct page *page;
unsigned long last_pfn = 0; /* suppress gcc warning */
gfp_t gfp;
/* 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->base.read_domains & I915_GEM_GPU_DOMAINS);
BUG_ON(obj->base.write_domain & I915_GEM_GPU_DOMAINS);
st = kmalloc(sizeof(*st), GFP_KERNEL);
if (st == NULL)
return -ENOMEM;
page_count = obj->base.size / PAGE_SIZE;
if (sg_alloc_table(st, page_count, GFP_KERNEL)) {
kfree(st);
return -ENOMEM;
}
/* Get the list of pages out of our struct file. They'll be pinned
* at this point until we release them.
*
* Fail silently without starting the shrinker
*/
mapping = file_inode(obj->base.filp)->i_mapping;
gfp = mapping_gfp_mask(mapping);
gfp |= __GFP_NORETRY | __GFP_NOWARN | __GFP_NO_KSWAPD;
gfp &= ~(__GFP_IO | __GFP_WAIT);
sg = st->sgl;
st->nents = 0;
for (i = 0; i < page_count; i++) {
page = shmem_read_mapping_page_gfp(mapping, i, gfp);
if (IS_ERR(page)) {
i915_gem_shrink(dev_priv,
page_count,
I915_SHRINK_BOUND |
I915_SHRINK_UNBOUND |
I915_SHRINK_PURGEABLE);
page = shmem_read_mapping_page_gfp(mapping, i, gfp);
}
if (IS_ERR(page)) {
/* We've tried hard to allocate the memory by reaping
* our own buffer, now let the real VM do its job and
* go down in flames if truly OOM.
*/
i915_gem_shrink_all(dev_priv);
page = shmem_read_mapping_page(mapping, i);
if (IS_ERR(page))
goto err_pages;
}
#ifdef CONFIG_SWIOTLB
if (swiotlb_nr_tbl()) {
st->nents++;
sg_set_page(sg, page, PAGE_SIZE, 0);
sg = sg_next(sg);
continue;
}
#endif
if (!i || page_to_pfn(page) != last_pfn + 1) {
if (i)
sg = sg_next(sg);
st->nents++;
sg_set_page(sg, page, PAGE_SIZE, 0);
} else {
sg->length += PAGE_SIZE;
}
last_pfn = page_to_pfn(page);
/* Check that the i965g/gm workaround works. */
WARN_ON((gfp & __GFP_DMA32) && (last_pfn >= 0x00100000UL));
}
#ifdef CONFIG_SWIOTLB
if (!swiotlb_nr_tbl())
#endif
sg_mark_end(sg);
obj->pages = st;
if (i915_gem_object_needs_bit17_swizzle(obj))
i915_gem_object_do_bit_17_swizzle(obj);
if (obj->tiling_mode != I915_TILING_NONE &&
dev_priv->quirks & QUIRK_PIN_SWIZZLED_PAGES)
i915_gem_object_pin_pages(obj);
return 0;
err_pages:
sg_mark_end(sg);
for_each_sg_page(st->sgl, &sg_iter, st->nents, 0)
page_cache_release(sg_page_iter_page(&sg_iter));
sg_free_table(st);
kfree(st);
/* shmemfs first checks if there is enough memory to allocate the page
* and reports ENOSPC should there be insufficient, along with the usual
* ENOMEM for a genuine allocation failure.
*
* We use ENOSPC in our driver to mean that we have run out of aperture
* space and so want to translate the error from shmemfs back to our
* usual understanding of ENOMEM.
*/
if (PTR_ERR(page) == -ENOSPC)
return -ENOMEM;
else
return PTR_ERR(page);
}
/* Ensure that the associated pages are gathered from the backing storage
* and pinned into our object. i915_gem_object_get_pages() may be called
* multiple times before they are released by a single call to
* i915_gem_object_put_pages() - once the pages are no longer referenced
* either as a result of memory pressure (reaping pages under the shrinker)
* or as the object is itself released.
*/
int
i915_gem_object_get_pages(struct drm_i915_gem_object *obj)
{
struct drm_i915_private *dev_priv = obj->base.dev->dev_private;
const struct drm_i915_gem_object_ops *ops = obj->ops;
int ret;
if (obj->pages)
return 0;
if (obj->madv != I915_MADV_WILLNEED) {
DRM_DEBUG("Attempting to obtain a purgeable object\n");
return -EFAULT;
}
BUG_ON(obj->pages_pin_count);
ret = ops->get_pages(obj);
if (ret)
return ret;
list_add_tail(&obj->global_list, &dev_priv->mm.unbound_list);
return 0;
}
static void
i915_gem_object_move_to_active(struct drm_i915_gem_object *obj,
struct intel_engine_cs *ring)
{
struct drm_i915_gem_request *req;
struct intel_engine_cs *old_ring;
BUG_ON(ring == NULL);
req = intel_ring_get_request(ring);
old_ring = i915_gem_request_get_ring(obj->last_read_req);
if (old_ring != ring && obj->last_write_req) {
/* Keep the request relative to the current ring */
i915_gem_request_assign(&obj->last_write_req, req);
}
/* Add a reference if we're newly entering the active list. */
if (!obj->active) {
drm_gem_object_reference(&obj->base);
obj->active = 1;
}
list_move_tail(&obj->ring_list, &ring->active_list);
i915_gem_request_assign(&obj->last_read_req, req);
}
void i915_vma_move_to_active(struct i915_vma *vma,
struct intel_engine_cs *ring)
{
list_move_tail(&vma->mm_list, &vma->vm->active_list);
return i915_gem_object_move_to_active(vma->obj, ring);
}
static void
i915_gem_object_move_to_inactive(struct drm_i915_gem_object *obj)
{
struct i915_vma *vma;
BUG_ON(obj->base.write_domain & ~I915_GEM_GPU_DOMAINS);
BUG_ON(!obj->active);
list_for_each_entry(vma, &obj->vma_list, vma_link) {
if (!list_empty(&vma->mm_list))
list_move_tail(&vma->mm_list, &vma->vm->inactive_list);
}
intel_fb_obj_flush(obj, true);
list_del_init(&obj->ring_list);
i915_gem_request_assign(&obj->last_read_req, NULL);
i915_gem_request_assign(&obj->last_write_req, NULL);
obj->base.write_domain = 0;
i915_gem_request_assign(&obj->last_fenced_req, NULL);
obj->active = 0;
drm_gem_object_unreference(&obj->base);
WARN_ON(i915_verify_lists(dev));
}
static void
i915_gem_object_retire(struct drm_i915_gem_object *obj)
{
if (obj->last_read_req == NULL)
return;
if (i915_gem_request_completed(obj->last_read_req, true))
i915_gem_object_move_to_inactive(obj);
}
static int
i915_gem_init_seqno(struct drm_device *dev, u32 seqno)
{
struct drm_i915_private *dev_priv = dev->dev_private;
struct intel_engine_cs *ring;
int ret, i, j;
/* Carefully retire all requests without writing to the rings */
for_each_ring(ring, dev_priv, i) {
ret = intel_ring_idle(ring);
if (ret)
return ret;
}
i915_gem_retire_requests(dev);
/* Finally reset hw state */
for_each_ring(ring, dev_priv, i) {
intel_ring_init_seqno(ring, seqno);
for (j = 0; j < ARRAY_SIZE(ring->semaphore.sync_seqno); j++)
ring->semaphore.sync_seqno[j] = 0;
}
return 0;
}
int i915_gem_set_seqno(struct drm_device *dev, u32 seqno)
{
struct drm_i915_private *dev_priv = dev->dev_private;
int ret;
if (seqno == 0)
return -EINVAL;
/* HWS page needs to be set less than what we
* will inject to ring
*/
ret = i915_gem_init_seqno(dev, seqno - 1);
if (ret)
return ret;
/* Carefully set the last_seqno value so that wrap
* detection still works
*/
dev_priv->next_seqno = seqno;
dev_priv->last_seqno = seqno - 1;
if (dev_priv->last_seqno == 0)
dev_priv->last_seqno--;
return 0;
}
int
i915_gem_get_seqno(struct drm_device *dev, u32 *seqno)
{
struct drm_i915_private *dev_priv = dev->dev_private;
/* reserve 0 for non-seqno */
if (dev_priv->next_seqno == 0) {
int ret = i915_gem_init_seqno(dev, 0);
if (ret)
return ret;
dev_priv->next_seqno = 1;
}
*seqno = dev_priv->last_seqno = dev_priv->next_seqno++;
return 0;
}
int __i915_add_request(struct intel_engine_cs *ring,
struct drm_file *file,
struct drm_i915_gem_object *obj)
{
struct drm_i915_private *dev_priv = ring->dev->dev_private;
struct drm_i915_gem_request *request;
struct intel_ringbuffer *ringbuf;
u32 request_start;
int ret;
request = ring->outstanding_lazy_request;
if (WARN_ON(request == NULL))
return -ENOMEM;
if (i915.enable_execlists) {
ringbuf = request->ctx->engine[ring->id].ringbuf;
} else
ringbuf = ring->buffer;
request_start = intel_ring_get_tail(ringbuf);
/*
* Emit any outstanding flushes - execbuf can fail to emit the flush
* after having emitted the batchbuffer command. Hence we need to fix
* things up similar to emitting the lazy request. The difference here
* is that the flush _must_ happen before the next request, no matter
* what.
*/
if (i915.enable_execlists) {
ret = logical_ring_flush_all_caches(ringbuf, request->ctx);
if (ret)
return ret;
} else {
ret = intel_ring_flush_all_caches(ring);
if (ret)
return ret;
}
/* Record the position of the start of the request so that
* should we detect the updated seqno part-way through the
* GPU processing the request, we never over-estimate the
* position of the head.
*/
request->postfix = intel_ring_get_tail(ringbuf);
if (i915.enable_execlists) {
ret = ring->emit_request(ringbuf, request);
if (ret)
return ret;
} else {
ret = ring->add_request(ring);
if (ret)
return ret;
}
request->head = request_start;
request->tail = intel_ring_get_tail(ringbuf);
/* Whilst this request exists, batch_obj will be on the
* active_list, and so will hold the active reference. Only when this
* request is retired will the the batch_obj be moved onto the
* inactive_list and lose its active reference. Hence we do not need
* to explicitly hold another reference here.
*/
request->batch_obj = obj;
if (!i915.enable_execlists) {
/* Hold a reference to the current context so that we can inspect
* it later in case a hangcheck error event fires.
*/
request->ctx = ring->last_context;
if (request->ctx)
i915_gem_context_reference(request->ctx);
}
request->emitted_jiffies = jiffies;
list_add_tail(&request->list, &ring->request_list);
request->file_priv = NULL;
if (file) {
struct drm_i915_file_private *file_priv = file->driver_priv;
spin_lock(&file_priv->mm.lock);
request->file_priv = file_priv;
list_add_tail(&request->client_list,
&file_priv->mm.request_list);
spin_unlock(&file_priv->mm.lock);
}
trace_i915_gem_request_add(request);
ring->outstanding_lazy_request = NULL;
i915_queue_hangcheck(ring->dev);
cancel_delayed_work_sync(&dev_priv->mm.idle_work);
queue_delayed_work(dev_priv->wq,
&dev_priv->mm.retire_work,
round_jiffies_up_relative(HZ));
intel_mark_busy(dev_priv->dev);
return 0;
}
static inline void
i915_gem_request_remove_from_client(struct drm_i915_gem_request *request)
{
struct drm_i915_file_private *file_priv = request->file_priv;
if (!file_priv)
return;
spin_lock(&file_priv->mm.lock);
list_del(&request->client_list);
request->file_priv = NULL;
spin_unlock(&file_priv->mm.lock);
}
static bool i915_context_is_banned(struct drm_i915_private *dev_priv,
const struct intel_context *ctx)
{
unsigned long elapsed;
elapsed = get_seconds() - ctx->hang_stats.guilty_ts;
if (ctx->hang_stats.banned)
return true;
if (ctx->hang_stats.ban_period_seconds &&
elapsed <= ctx->hang_stats.ban_period_seconds) {
if (!i915_gem_context_is_default(ctx)) {
DRM_DEBUG("context hanging too fast, banning!\n");
return true;
} else if (i915_stop_ring_allow_ban(dev_priv)) {
if (i915_stop_ring_allow_warn(dev_priv))
DRM_ERROR("gpu hanging too fast, banning!\n");
return true;
}
}
return false;
}
static void i915_set_reset_status(struct drm_i915_private *dev_priv,
struct intel_context *ctx,
const bool guilty)
{
struct i915_ctx_hang_stats *hs;
if (WARN_ON(!ctx))
return;
hs = &ctx->hang_stats;
if (guilty) {
hs->banned = i915_context_is_banned(dev_priv, ctx);
hs->batch_active++;
hs->guilty_ts = get_seconds();
} else {
hs->batch_pending++;
}
}
static void i915_gem_free_request(struct drm_i915_gem_request *request)
{
list_del(&request->list);
i915_gem_request_remove_from_client(request);
i915_gem_request_unreference(request);
}
void i915_gem_request_free(struct kref *req_ref)
{
struct drm_i915_gem_request *req = container_of(req_ref,
typeof(*req), ref);
struct intel_context *ctx = req->ctx;
if (ctx) {
if (i915.enable_execlists) {
struct intel_engine_cs *ring = req->ring;
if (ctx != ring->default_context)
intel_lr_context_unpin(ring, ctx);
}
i915_gem_context_unreference(ctx);
}
kfree(req);
}
struct drm_i915_gem_request *
i915_gem_find_active_request(struct intel_engine_cs *ring)
{
struct drm_i915_gem_request *request;
list_for_each_entry(request, &ring->request_list, list) {
if (i915_gem_request_completed(request, false))
continue;
return request;
}
return NULL;
}
static void i915_gem_reset_ring_status(struct drm_i915_private *dev_priv,
struct intel_engine_cs *ring)
{
struct drm_i915_gem_request *request;
bool ring_hung;
request = i915_gem_find_active_request(ring);
if (request == NULL)
return;
ring_hung = ring->hangcheck.score >= HANGCHECK_SCORE_RING_HUNG;
i915_set_reset_status(dev_priv, request->ctx, ring_hung);
list_for_each_entry_continue(request, &ring->request_list, list)
i915_set_reset_status(dev_priv, request->ctx, false);
}
static void i915_gem_reset_ring_cleanup(struct drm_i915_private *dev_priv,
struct intel_engine_cs *ring)
{
while (!list_empty(&ring->active_list)) {
struct drm_i915_gem_object *obj;
obj = list_first_entry(&ring->active_list,
struct drm_i915_gem_object,
ring_list);
i915_gem_object_move_to_inactive(obj);
}
/*
* Clear the execlists queue up before freeing the requests, as those
* are the ones that keep the context and ringbuffer backing objects
* pinned in place.
*/
while (!list_empty(&ring->execlist_queue)) {
struct drm_i915_gem_request *submit_req;
submit_req = list_first_entry(&ring->execlist_queue,
struct drm_i915_gem_request,
execlist_link);
list_del(&submit_req->execlist_link);
intel_runtime_pm_put(dev_priv);
if (submit_req->ctx != ring->default_context)
intel_lr_context_unpin(ring, submit_req->ctx);
i915_gem_context_unreference(submit_req->ctx);
kfree(submit_req);
}
/*
* We must free the requests after all the corresponding objects have
* been moved off active lists. Which is the same order as the normal
* retire_requests function does. This is important if object hold
* implicit references on things like e.g. ppgtt address spaces through
* the request.
*/
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);
i915_gem_free_request(request);
}
/* This may not have been flushed before the reset, so clean it now */
i915_gem_request_assign(&ring->outstanding_lazy_request, NULL);
}
void i915_gem_restore_fences(struct drm_device *dev)
{
struct drm_i915_private *dev_priv = dev->dev_private;
int i;
for (i = 0; i < dev_priv->num_fence_regs; i++) {
struct drm_i915_fence_reg *reg = &dev_priv->fence_regs[i];
/*
* Commit delayed tiling changes if we have an object still
* attached to the fence, otherwise just clear the fence.
*/
if (reg->obj) {
i915_gem_object_update_fence(reg->obj, reg,
reg->obj->tiling_mode);
} else {
i915_gem_write_fence(dev, i, NULL);
}
}
}
void i915_gem_reset(struct drm_device *dev)
{
struct drm_i915_private *dev_priv = dev->dev_private;
struct intel_engine_cs *ring;
int i;
/*
* Before we free the objects from the requests, we need to inspect
* them for finding the guilty party. As the requests only borrow
* their reference to the objects, the inspection must be done first.
*/
for_each_ring(ring, dev_priv, i)
i915_gem_reset_ring_status(dev_priv, ring);
for_each_ring(ring, dev_priv, i)
i915_gem_reset_ring_cleanup(dev_priv, ring);
i915_gem_context_reset(dev);
i915_gem_restore_fences(dev);
}
/**
* This function clears the request list as sequence numbers are passed.
*/
void
i915_gem_retire_requests_ring(struct intel_engine_cs *ring)
{
if (list_empty(&ring->request_list))
return;
WARN_ON(i915_verify_lists(ring->dev));
/* Move any buffers on the active list that are no longer referenced
* by the ringbuffer to the flushing/inactive lists as appropriate,
* before we free the context associated with the requests.
*/
while (!list_empty(&ring->active_list)) {
struct drm_i915_gem_object *obj;
obj = list_first_entry(&ring->active_list,
struct drm_i915_gem_object,
ring_list);
if (!i915_gem_request_completed(obj->last_read_req, true))
break;
i915_gem_object_move_to_inactive(obj);
}
while (!list_empty(&ring->request_list)) {
struct drm_i915_gem_request *request;
struct intel_ringbuffer *ringbuf;
request = list_first_entry(&ring->request_list,
struct drm_i915_gem_request,
list);
if (!i915_gem_request_completed(request, true))
break;
trace_i915_gem_request_retire(request);
/* This is one of the few common intersection points
* between legacy ringbuffer submission and execlists:
* we need to tell them apart in order to find the correct
* ringbuffer to which the request belongs to.
*/
if (i915.enable_execlists) {
struct intel_context *ctx = request->ctx;
ringbuf = ctx->engine[ring->id].ringbuf;
} else
ringbuf = ring->buffer;
/* We know the GPU must have read the request to have
* sent us the seqno + interrupt, so use the position
* of tail of the request to update the last known position
* of the GPU head.
*/
ringbuf->last_retired_head = request->postfix;
i915_gem_free_request(request);
}
if (unlikely(ring->trace_irq_req &&
i915_gem_request_completed(ring->trace_irq_req, true))) {
ring->irq_put(ring);
i915_gem_request_assign(&ring->trace_irq_req, NULL);
}
WARN_ON(i915_verify_lists(ring->dev));
}
bool
i915_gem_retire_requests(struct drm_device *dev)
{
struct drm_i915_private *dev_priv = dev->dev_private;
struct intel_engine_cs *ring;
bool idle = true;
int i;
for_each_ring(ring, dev_priv, i) {
i915_gem_retire_requests_ring(ring);
idle &= list_empty(&ring->request_list);
if (i915.enable_execlists) {
unsigned long flags;
spin_lock_irqsave(&ring->execlist_lock, flags);
idle &= list_empty(&ring->execlist_queue);
spin_unlock_irqrestore(&ring->execlist_lock, flags);
intel_execlists_retire_requests(ring);
}
}
if (idle)
mod_delayed_work(dev_priv->wq,
&dev_priv->mm.idle_work,
msecs_to_jiffies(100));
return idle;
}
static void
i915_gem_retire_work_handler(struct work_struct *work)
{
struct drm_i915_private *dev_priv =
container_of(work, typeof(*dev_priv), mm.retire_work.work);
struct drm_device *dev = dev_priv->dev;
bool idle;
/* Come back later if the device is busy... */
idle = false;
if (mutex_trylock(&dev->struct_mutex)) {
idle = i915_gem_retire_requests(dev);
mutex_unlock(&dev->struct_mutex);
}
if (!idle)
queue_delayed_work(dev_priv->wq, &dev_priv->mm.retire_work,
round_jiffies_up_relative(HZ));
}
static void
i915_gem_idle_work_handler(struct work_struct *work)
{
struct drm_i915_private *dev_priv =
container_of(work, typeof(*dev_priv), mm.idle_work.work);
intel_mark_idle(dev_priv->dev);
}
/**
* Ensures that an object will eventually get non-busy by flushing any required
* write domains, emitting any outstanding lazy request and retiring and
* completed requests.
*/
static int
i915_gem_object_flush_active(struct drm_i915_gem_object *obj)
{
struct intel_engine_cs *ring;
int ret;
if (obj->active) {
ring = i915_gem_request_get_ring(obj->last_read_req);
ret = i915_gem_check_olr(obj->last_read_req);
if (ret)
return ret;
i915_gem_retire_requests_ring(ring);
}
return 0;
}
/**
* i915_gem_wait_ioctl - implements DRM_IOCTL_I915_GEM_WAIT
* @DRM_IOCTL_ARGS: standard ioctl arguments
*
* Returns 0 if successful, else an error is returned with the remaining time in
* the timeout parameter.
* -ETIME: object is still busy after timeout
* -ERESTARTSYS: signal interrupted the wait
* -ENONENT: object doesn't exist
* Also possible, but rare:
* -EAGAIN: GPU wedged
* -ENOMEM: damn
* -ENODEV: Internal IRQ fail
* -E?: The add request failed
*
* The wait ioctl with a timeout of 0 reimplements the busy ioctl. With any
* non-zero timeout parameter the wait ioctl will wait for the given number of
* nanoseconds on an object becoming unbusy. Since the wait itself does so
* without holding struct_mutex the object may become re-busied before this
* function completes. A similar but shorter * race condition exists in the busy
* ioctl
*/
int
i915_gem_wait_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
{
struct drm_i915_private *dev_priv = dev->dev_private;
struct drm_i915_gem_wait *args = data;