#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>
-#define I915_GEM_GPU_DOMAINS (~(I915_GEM_DOMAIN_CPU | I915_GEM_DOMAIN_GTT))
-
static void i915_gem_object_flush_gpu_write_domain(struct drm_gem_object *obj);
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);
unsigned alignment);
static void i915_gem_clear_fence_reg(struct drm_gem_object *obj);
static int i915_gem_evict_something(struct drm_device *dev, int min_size);
-static int i915_gem_evict_from_list(struct drm_device *dev,
- struct list_head *head);
+static int i915_gem_evict_from_inactive_list(struct drm_device *dev);
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);
args->size = roundup(args->size, PAGE_SIZE);
/* Allocate the new object */
- obj = drm_gem_object_alloc(dev, args->size);
+ obj = i915_gem_alloc_object(dev, args->size);
if (obj == NULL)
return -ENOMEM;
ret = drm_gem_handle_create(file_priv, obj, &handle);
- mutex_lock(&dev->struct_mutex);
- drm_gem_object_handle_unreference(obj);
- mutex_unlock(&dev->struct_mutex);
+ drm_gem_object_handle_unreference_unlocked(obj);
if (ret)
return ret;
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 = obj->driver_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;
struct drm_i915_gem_pread *args,
struct drm_file *file_priv)
{
- struct drm_i915_gem_object *obj_priv = obj->driver_private;
+ struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
ssize_t remain;
loff_t offset, page_base;
char __user *user_data;
mutex_lock(&dev->struct_mutex);
- ret = i915_gem_object_get_pages(obj);
+ ret = i915_gem_object_get_pages(obj, 0);
if (ret != 0)
goto fail_unlock;
if (ret != 0)
goto fail_put_pages;
- obj_priv = obj->driver_private;
+ obj_priv = to_intel_bo(obj);
offset = args->offset;
while (remain > 0) {
return ret;
}
-static inline gfp_t
-i915_gem_object_get_page_gfp_mask (struct drm_gem_object *obj)
-{
- return mapping_gfp_mask(obj->filp->f_path.dentry->d_inode->i_mapping);
-}
-
-static inline void
-i915_gem_object_set_page_gfp_mask (struct drm_gem_object *obj, gfp_t gfp)
-{
- mapping_set_gfp_mask(obj->filp->f_path.dentry->d_inode->i_mapping, gfp);
-}
-
static int
i915_gem_object_get_pages_or_evict(struct drm_gem_object *obj)
{
int ret;
- ret = i915_gem_object_get_pages(obj);
+ 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;
- gfp_t gfp;
ret = i915_gem_evict_something(dev, obj->size);
if (ret)
return ret;
- gfp = i915_gem_object_get_page_gfp_mask(obj);
- i915_gem_object_set_page_gfp_mask(obj, gfp & ~__GFP_NORETRY);
- ret = i915_gem_object_get_pages(obj);
- i915_gem_object_set_page_gfp_mask (obj, gfp);
+ ret = i915_gem_object_get_pages(obj, 0);
}
return ret;
struct drm_i915_gem_pread *args,
struct drm_file *file_priv)
{
- struct drm_i915_gem_object *obj_priv = obj->driver_private;
+ struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
struct mm_struct *mm = current->mm;
struct page **user_pages;
ssize_t remain;
if (ret != 0)
goto fail_put_pages;
- obj_priv = obj->driver_private;
+ obj_priv = to_intel_bo(obj);
offset = args->offset;
while (remain > 0) {
obj = drm_gem_object_lookup(dev, file_priv, args->handle);
if (obj == NULL)
return -EBADF;
- obj_priv = obj->driver_private;
+ obj_priv = to_intel_bo(obj);
/* Bounds check source.
*
*/
if (args->offset > obj->size || args->size > obj->size ||
args->offset + args->size > obj->size) {
- drm_gem_object_unreference(obj);
+ drm_gem_object_unreference_unlocked(obj);
return -EINVAL;
}
file_priv);
}
- drm_gem_object_unreference(obj);
+ drm_gem_object_unreference_unlocked(obj);
return ret;
}
struct drm_i915_gem_pwrite *args,
struct drm_file *file_priv)
{
- struct drm_i915_gem_object *obj_priv = obj->driver_private;
+ 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;
if (ret)
goto fail;
- obj_priv = obj->driver_private;
+ obj_priv = to_intel_bo(obj);
offset = obj_priv->gtt_offset + args->offset;
while (remain > 0) {
struct drm_i915_gem_pwrite *args,
struct drm_file *file_priv)
{
- struct drm_i915_gem_object *obj_priv = obj->driver_private;
+ 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;
if (ret)
goto out_unpin_object;
- obj_priv = obj->driver_private;
+ obj_priv = to_intel_bo(obj);
offset = obj_priv->gtt_offset + args->offset;
while (remain > 0) {
struct drm_i915_gem_pwrite *args,
struct drm_file *file_priv)
{
- struct drm_i915_gem_object *obj_priv = obj->driver_private;
+ struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
ssize_t remain;
loff_t offset, page_base;
char __user *user_data;
mutex_lock(&dev->struct_mutex);
- ret = i915_gem_object_get_pages(obj);
+ ret = i915_gem_object_get_pages(obj, 0);
if (ret != 0)
goto fail_unlock;
if (ret != 0)
goto fail_put_pages;
- obj_priv = obj->driver_private;
+ obj_priv = to_intel_bo(obj);
offset = args->offset;
obj_priv->dirty = 1;
struct drm_i915_gem_pwrite *args,
struct drm_file *file_priv)
{
- struct drm_i915_gem_object *obj_priv = obj->driver_private;
+ struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
struct mm_struct *mm = current->mm;
struct page **user_pages;
ssize_t remain;
if (ret != 0)
goto fail_put_pages;
- obj_priv = obj->driver_private;
+ obj_priv = to_intel_bo(obj);
offset = args->offset;
obj_priv->dirty = 1;
obj = drm_gem_object_lookup(dev, file_priv, args->handle);
if (obj == NULL)
return -EBADF;
- obj_priv = obj->driver_private;
+ obj_priv = to_intel_bo(obj);
/* Bounds check destination.
*
*/
if (args->offset > obj->size || args->size > obj->size ||
args->offset + args->size > obj->size) {
- drm_gem_object_unreference(obj);
+ drm_gem_object_unreference_unlocked(obj);
return -EINVAL;
}
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) {
+ 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,
DRM_INFO("pwrite failed %d\n", ret);
#endif
- drm_gem_object_unreference(obj);
+ drm_gem_object_unreference_unlocked(obj);
return ret;
}
obj = drm_gem_object_lookup(dev, file_priv, args->handle);
if (obj == NULL)
return -EBADF;
- obj_priv = obj->driver_private;
+ obj_priv = to_intel_bo(obj);
mutex_lock(&dev->struct_mutex);
* about to occur.
*/
if (obj_priv->fence_reg != I915_FENCE_REG_NONE) {
- list_move_tail(&obj_priv->fence_list,
+ struct drm_i915_fence_reg *reg =
+ &dev_priv->fence_regs[obj_priv->fence_reg];
+ list_move_tail(®->lru_list,
&dev_priv->mm.fence_list);
}
DRM_INFO("%s: sw_finish %d (%p %zd)\n",
__func__, args->handle, obj, obj->size);
#endif
- obj_priv = obj->driver_private;
+ obj_priv = to_intel_bo(obj);
/* Pinned buffers may be scanout, so flush the cache */
if (obj_priv->pin_count)
PROT_READ | PROT_WRITE, MAP_SHARED,
args->offset);
up_write(¤t->mm->mmap_sem);
- mutex_lock(&dev->struct_mutex);
- drm_gem_object_unreference(obj);
- mutex_unlock(&dev->struct_mutex);
+ drm_gem_object_unreference_unlocked(obj);
if (IS_ERR((void *)addr))
return addr;
struct drm_gem_object *obj = vma->vm_private_data;
struct drm_device *dev = obj->dev;
struct drm_i915_private *dev_priv = dev->dev_private;
- struct drm_i915_gem_object *obj_priv = obj->driver_private;
+ struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
pgoff_t page_offset;
unsigned long pfn;
int ret = 0;
mutex_lock(&dev->struct_mutex);
if (!obj_priv->gtt_space) {
ret = i915_gem_object_bind_to_gtt(obj, 0);
- if (ret) {
- mutex_unlock(&dev->struct_mutex);
- return VM_FAULT_SIGBUS;
- }
+ if (ret)
+ goto unlock;
+
list_add_tail(&obj_priv->list, &dev_priv->mm.inactive_list);
ret = i915_gem_object_set_to_gtt_domain(obj, write);
- if (ret) {
- mutex_unlock(&dev->struct_mutex);
- return VM_FAULT_SIGBUS;
- }
+ 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);
- if (ret) {
- mutex_unlock(&dev->struct_mutex);
- return VM_FAULT_SIGBUS;
- }
+ if (ret)
+ goto unlock;
}
pfn = ((dev->agp->base + obj_priv->gtt_offset) >> PAGE_SHIFT) +
/* 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;
- case -EFAULT:
- case -EINVAL:
- return VM_FAULT_SIGBUS;
default:
- return VM_FAULT_NOPAGE;
+ return VM_FAULT_SIGBUS;
}
}
{
struct drm_device *dev = obj->dev;
struct drm_gem_mm *mm = dev->mm_private;
- struct drm_i915_gem_object *obj_priv = obj->driver_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;
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;
}
* i915_gem_release_mmap - remove physical page mappings
* @obj: obj in question
*
- * Preserve the reservation of the mmaping with the DRM core code, but
+ * 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
i915_gem_release_mmap(struct drm_gem_object *obj)
{
struct drm_device *dev = obj->dev;
- struct drm_i915_gem_object *obj_priv = obj->driver_private;
+ struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
if (dev->dev_mapping)
unmap_mapping_range(dev->dev_mapping,
i915_gem_free_mmap_offset(struct drm_gem_object *obj)
{
struct drm_device *dev = obj->dev;
- struct drm_i915_gem_object *obj_priv = obj->driver_private;
+ struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
struct drm_gem_mm *mm = dev->mm_private;
struct drm_map_list *list;
i915_gem_get_gtt_alignment(struct drm_gem_object *obj)
{
struct drm_device *dev = obj->dev;
- struct drm_i915_gem_object *obj_priv = obj->driver_private;
+ struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
int start, i;
/*
mutex_lock(&dev->struct_mutex);
- obj_priv = obj->driver_private;
+ 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);
void
i915_gem_object_put_pages(struct drm_gem_object *obj)
{
- struct drm_i915_gem_object *obj_priv = obj->driver_private;
+ 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;
i915_gem_object_save_bit_17_swizzle(obj);
if (obj_priv->madv == I915_MADV_DONTNEED)
- obj_priv->dirty = 0;
+ obj_priv->dirty = 0;
for (i = 0; i < page_count; i++) {
- if (obj_priv->pages[i] == NULL)
- break;
-
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]);
+ mark_page_accessed(obj_priv->pages[i]);
page_cache_release(obj_priv->pages[i]);
}
}
static void
-i915_gem_object_move_to_active(struct drm_gem_object *obj, uint32_t seqno)
+i915_gem_object_move_to_active(struct drm_gem_object *obj, uint32_t seqno,
+ struct intel_ring_buffer *ring)
{
struct drm_device *dev = obj->dev;
drm_i915_private_t *dev_priv = dev->dev_private;
- struct drm_i915_gem_object *obj_priv = obj->driver_private;
+ struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
+ BUG_ON(ring == NULL);
+ obj_priv->ring = ring;
/* Add a reference if we're newly entering the active list. */
if (!obj_priv->active) {
}
/* Move from whatever list we were on to the tail of execution. */
spin_lock(&dev_priv->mm.active_list_lock);
- list_move_tail(&obj_priv->list,
- &dev_priv->mm.active_list);
+ list_move_tail(&obj_priv->list, &ring->active_list);
spin_unlock(&dev_priv->mm.active_list_lock);
obj_priv->last_rendering_seqno = seqno;
}
{
struct drm_device *dev = obj->dev;
drm_i915_private_t *dev_priv = dev->dev_private;
- struct drm_i915_gem_object *obj_priv = obj->driver_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;
+
+ inode = obj->filp->f_path.dentry->d_inode;
+ if (inode->i_op->truncate)
+ inode->i_op->truncate (inode);
+
+ 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 = obj->driver_private;
+ struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
i915_verify_inactive(dev, __FILE__, __LINE__);
if (obj_priv->pin_count != 0)
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__);
}
-/**
- * Creates a new sequence number, emitting a write of it to the status page
- * plus an interrupt, which will trigger i915_user_interrupt_handler.
- *
- * Must be called with struct_lock held.
- *
- * Returned sequence numbers are nonzero on success.
- */
-static uint32_t
+static void
+i915_gem_process_flushing_list(struct drm_device *dev,
+ uint32_t flush_domains, uint32_t seqno,
+ 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->write_domain &&
+ obj_priv->ring->ring_flag == ring->ring_flag) {
+ 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, seqno, 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(®->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,
- uint32_t flush_domains)
+ uint32_t flush_domains, struct intel_ring_buffer *ring)
{
drm_i915_private_t *dev_priv = dev->dev_private;
struct drm_i915_file_private *i915_file_priv = NULL;
struct drm_i915_gem_request *request;
uint32_t seqno;
int was_empty;
- RING_LOCALS;
if (file_priv != NULL)
i915_file_priv = file_priv->driver_priv;
if (request == NULL)
return 0;
- /* Grab the seqno we're going to make this request be, and bump the
- * next (skipping 0 so it can be the reserved no-seqno value).
- */
- seqno = dev_priv->mm.next_gem_seqno;
- dev_priv->mm.next_gem_seqno++;
- if (dev_priv->mm.next_gem_seqno == 0)
- dev_priv->mm.next_gem_seqno++;
-
- BEGIN_LP_RING(4);
- OUT_RING(MI_STORE_DWORD_INDEX);
- OUT_RING(I915_GEM_HWS_INDEX << MI_STORE_DWORD_INDEX_SHIFT);
- OUT_RING(seqno);
-
- OUT_RING(MI_USER_INTERRUPT);
- ADVANCE_LP_RING();
-
- DRM_DEBUG("%d\n", seqno);
+ seqno = ring->add_request(dev, ring, file_priv, flush_domains);
request->seqno = seqno;
+ request->ring = ring;
request->emitted_jiffies = jiffies;
- was_empty = list_empty(&dev_priv->mm.request_list);
- list_add_tail(&request->list, &dev_priv->mm.request_list);
+ 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);
/* Associate any objects on the flushing list matching the write
* domain we're flushing with our flush.
*/
- if (flush_domains != 0) {
- struct drm_i915_gem_object *obj_priv, *next;
-
- list_for_each_entry_safe(obj_priv, next,
- &dev_priv->mm.flushing_list, list) {
- struct drm_gem_object *obj = obj_priv->obj;
-
- if ((obj->write_domain & flush_domains) ==
- obj->write_domain) {
- obj->write_domain = 0;
- i915_gem_object_move_to_active(obj, seqno);
- }
- }
-
- }
+ if (flush_domains != 0)
+ i915_gem_process_flushing_list(dev, flush_domains, seqno, ring);
if (!dev_priv->mm.suspended) {
mod_timer(&dev_priv->hangcheck_timer, jiffies + DRM_I915_HANGCHECK_PERIOD);
* before signalling the CPU
*/
static uint32_t
-i915_retire_commands(struct drm_device *dev)
+i915_retire_commands(struct drm_device *dev, struct intel_ring_buffer *ring)
{
- drm_i915_private_t *dev_priv = dev->dev_private;
- uint32_t cmd = MI_FLUSH | MI_NO_WRITE_FLUSH;
uint32_t flush_domains = 0;
- RING_LOCALS;
/* The sampler always gets flushed on i965 (sigh) */
if (IS_I965G(dev))
flush_domains |= I915_GEM_DOMAIN_SAMPLER;
- BEGIN_LP_RING(2);
- OUT_RING(cmd);
- OUT_RING(0); /* noop */
- ADVANCE_LP_RING();
+
+ ring->flush(dev, ring,
+ I915_GEM_DOMAIN_COMMAND, flush_domains);
return flush_domains;
}
{
drm_i915_private_t *dev_priv = dev->dev_private;
+ trace_i915_gem_request_retire(dev, request->seqno);
+
/* Move any buffers on the active list that are no longer referenced
* by the ringbuffer to the flushing/inactive lists as appropriate.
*/
spin_lock(&dev_priv->mm.active_list_lock);
- while (!list_empty(&dev_priv->mm.active_list)) {
+ while (!list_empty(&request->ring->active_list)) {
struct drm_gem_object *obj;
struct drm_i915_gem_object *obj_priv;
- obj_priv = list_first_entry(&dev_priv->mm.active_list,
+ obj_priv = list_first_entry(&request->ring->active_list,
struct drm_i915_gem_object,
list);
- obj = obj_priv->obj;
+ obj = &obj_priv->base;
/* If the seqno being retired doesn't match the oldest in the
* list, then the oldest in the list must still be newer than
}
uint32_t
-i915_get_gem_seqno(struct drm_device *dev)
+i915_get_gem_seqno(struct drm_device *dev,
+ struct intel_ring_buffer *ring)
{
- drm_i915_private_t *dev_priv = dev->dev_private;
-
- return READ_HWSP(dev_priv, I915_GEM_HWS_INDEX);
+ return ring->get_gem_seqno(dev, ring);
}
/**
* This function clears the request list as sequence numbers are passed.
*/
void
-i915_gem_retire_requests(struct drm_device *dev)
+i915_gem_retire_requests(struct drm_device *dev,
+ struct intel_ring_buffer *ring)
{
drm_i915_private_t *dev_priv = dev->dev_private;
uint32_t seqno;
- if (!dev_priv->hw_status_page)
+ if (!ring->status_page.page_addr
+ || list_empty(&ring->request_list))
return;
- seqno = i915_get_gem_seqno(dev);
+ seqno = i915_get_gem_seqno(dev, ring);
- while (!list_empty(&dev_priv->mm.request_list)) {
+ while (!list_empty(&ring->request_list)) {
struct drm_i915_gem_request *request;
uint32_t retiring_seqno;
- request = list_first_entry(&dev_priv->mm.request_list,
+ request = list_first_entry(&ring->request_list,
struct drm_i915_gem_request,
list);
retiring_seqno = request->seqno;
} else
break;
}
+
+ 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
dev = dev_priv->dev;
mutex_lock(&dev->struct_mutex);
- i915_gem_retire_requests(dev);
+ i915_gem_retire_requests(dev, &dev_priv->render_ring);
+
+ if (HAS_BSD(dev))
+ i915_gem_retire_requests(dev, &dev_priv->bsd_ring);
+
if (!dev_priv->mm.suspended &&
- !list_empty(&dev_priv->mm.request_list))
+ (!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);
}
-/**
- * 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)
+int
+i915_do_wait_request(struct drm_device *dev, uint32_t seqno,
+ int interruptible, struct intel_ring_buffer *ring)
{
drm_i915_private_t *dev_priv = dev->dev_private;
u32 ier;
if (atomic_read(&dev_priv->mm.wedged))
return -EIO;
- if (!i915_seqno_passed(i915_get_gem_seqno(dev), seqno)) {
- if (IS_IGDNG(dev))
+ 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);
i915_driver_irq_postinstall(dev);
}
- dev_priv->mm.waiting_gem_seqno = seqno;
- i915_user_irq_get(dev);
- ret = wait_event_interruptible(dev_priv->irq_queue,
- i915_seqno_passed(i915_get_gem_seqno(dev),
- seqno) ||
- atomic_read(&dev_priv->mm.wedged));
- i915_user_irq_put(dev);
- dev_priv->mm.waiting_gem_seqno = 0;
+ 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)\n",
- __func__, ret, seqno, i915_get_gem_seqno(dev));
+ __func__, ret, seqno, ring->get_gem_seqno(dev, ring));
/* Directly dispatch request retiring. While we have the work queue
* to handle this, the waiter on a request often wants an associated
* a separate wait queue to handle that.
*/
if (ret == 0)
- i915_gem_retire_requests(dev);
+ i915_gem_retire_requests(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);
+}
+
static void
i915_gem_flush(struct drm_device *dev,
uint32_t invalidate_domains,
uint32_t flush_domains)
{
drm_i915_private_t *dev_priv = dev->dev_private;
- uint32_t cmd;
- RING_LOCALS;
-
-#if WATCH_EXEC
- DRM_INFO("%s: invalidate %08x flush %08x\n", __func__,
- invalidate_domains, flush_domains);
-#endif
-
if (flush_domains & I915_GEM_DOMAIN_CPU)
drm_agp_chipset_flush(dev);
+ dev_priv->render_ring.flush(dev, &dev_priv->render_ring,
+ invalidate_domains,
+ flush_domains);
+
+ if (HAS_BSD(dev))
+ dev_priv->bsd_ring.flush(dev, &dev_priv->bsd_ring,
+ invalidate_domains,
+ flush_domains);
+}
- if ((invalidate_domains | flush_domains) & I915_GEM_GPU_DOMAINS) {
- /*
- * read/write caches:
- *
- * I915_GEM_DOMAIN_RENDER is always invalidated, but is
- * only flushed if MI_NO_WRITE_FLUSH is unset. On 965, it is
- * also flushed at 2d versus 3d pipeline switches.
- *
- * read-only caches:
- *
- * I915_GEM_DOMAIN_SAMPLER is flushed on pre-965 if
- * MI_READ_FLUSH is set, and is always flushed on 965.
- *
- * I915_GEM_DOMAIN_COMMAND may not exist?
- *
- * I915_GEM_DOMAIN_INSTRUCTION, which exists on 965, is
- * invalidated when MI_EXE_FLUSH is set.
- *
- * I915_GEM_DOMAIN_VERTEX, which exists on 965, is
- * invalidated with every MI_FLUSH.
- *
- * TLBs:
- *
- * On 965, TLBs associated with I915_GEM_DOMAIN_COMMAND
- * and I915_GEM_DOMAIN_CPU in are invalidated at PTE write and
- * I915_GEM_DOMAIN_RENDER and I915_GEM_DOMAIN_SAMPLER
- * are flushed at any MI_FLUSH.
- */
-
- cmd = MI_FLUSH | MI_NO_WRITE_FLUSH;
- if ((invalidate_domains|flush_domains) &
- I915_GEM_DOMAIN_RENDER)
- cmd &= ~MI_NO_WRITE_FLUSH;
- if (!IS_I965G(dev)) {
- /*
- * On the 965, the sampler cache always gets flushed
- * and this bit is reserved.
- */
- if (invalidate_domains & I915_GEM_DOMAIN_SAMPLER)
- cmd |= MI_READ_FLUSH;
- }
- if (invalidate_domains & I915_GEM_DOMAIN_INSTRUCTION)
- cmd |= MI_EXE_FLUSH;
-
-#if WATCH_EXEC
- DRM_INFO("%s: queue flush %08x to ring\n", __func__, cmd);
-#endif
- BEGIN_LP_RING(2);
- OUT_RING(cmd);
- OUT_RING(0); /* noop */
- ADVANCE_LP_RING();
- }
+static void
+i915_gem_flush_ring(struct drm_device *dev,
+ uint32_t invalidate_domains,
+ uint32_t flush_domains,
+ struct intel_ring_buffer *ring)
+{
+ if (flush_domains & I915_GEM_DOMAIN_CPU)
+ drm_agp_chipset_flush(dev);
+ ring->flush(dev, ring,
+ invalidate_domains,
+ flush_domains);
}
/**
i915_gem_object_wait_rendering(struct drm_gem_object *obj)
{
struct drm_device *dev = obj->dev;
- struct drm_i915_gem_object *obj_priv = obj->driver_private;
+ struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
int ret;
/* This function only exists to support waiting for existing rendering,
DRM_INFO("%s: object %p wait for seqno %08x\n",
__func__, obj, obj_priv->last_rendering_seqno);
#endif
- ret = i915_wait_request(dev, obj_priv->last_rendering_seqno);
+ ret = i915_wait_request(dev,
+ obj_priv->last_rendering_seqno, obj_priv->ring);
if (ret != 0)
return ret;
}
i915_gem_object_unbind(struct drm_gem_object *obj)
{
struct drm_device *dev = obj->dev;
- struct drm_i915_gem_object *obj_priv = obj->driver_private;
+ drm_i915_private_t *dev_priv = dev->dev_private;
+ struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
int ret = 0;
#if WATCH_BUF
/* blow away mappings if mapped through GTT */
i915_gem_release_mmap(obj);
- if (obj_priv->fence_reg != I915_FENCE_REG_NONE)
- i915_gem_clear_fence_reg(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
BUG_ON(obj_priv->active);
+ /* 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);
}
i915_gem_object_put_pages(obj);
+ BUG_ON(obj_priv->pages_refcount);
if (obj_priv->gtt_space) {
atomic_dec(&dev->gtt_count);
}
/* Remove ourselves from the LRU list if present. */
+ spin_lock(&dev_priv->mm.active_list_lock);
if (!list_empty(&obj_priv->list))
list_del_init(&obj_priv->list);
+ spin_unlock(&dev_priv->mm.active_list_lock);
- return 0;
-}
+ if (i915_gem_object_is_purgeable(obj_priv))
+ i915_gem_object_truncate(obj);
-static inline int
-i915_gem_object_is_purgeable(struct drm_i915_gem_object *obj_priv)
-{
- return !obj_priv->dirty || obj_priv->madv == I915_MADV_DONTNEED;
+ trace_i915_gem_object_unbind(obj);
+
+ return 0;
}
static struct drm_gem_object *
/* Try to find the smallest clean object */
list_for_each_entry(obj_priv, &dev_priv->mm.inactive_list, list) {
- struct drm_gem_object *obj = obj_priv->obj;
+ struct drm_gem_object *obj = &obj_priv->base;
if (obj->size >= min_size) {
- if (i915_gem_object_is_purgeable(obj_priv) &&
+ if ((!obj_priv->dirty ||
+ i915_gem_object_is_purgeable(obj_priv)) &&
(!best || obj->size < best->size)) {
best = obj;
if (best->size == min_size)
}
static int
+i915_gpu_idle(struct drm_device *dev)
+{
+ drm_i915_private_t *dev_priv = dev->dev_private;
+ bool lists_empty;
+ uint32_t seqno1, seqno2;
+ int ret;
+
+ spin_lock(&dev_priv->mm.active_list_lock);
+ 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)));
+ spin_unlock(&dev_priv->mm.active_list_lock);
+
+ if (lists_empty)
+ return 0;
+
+ /* Flush everything onto the inactive list. */
+ i915_gem_flush(dev, I915_GEM_GPU_DOMAINS, I915_GEM_GPU_DOMAINS);
+ seqno1 = i915_add_request(dev, NULL, I915_GEM_GPU_DOMAINS,
+ &dev_priv->render_ring);
+ if (seqno1 == 0)
+ return -ENOMEM;
+ ret = i915_wait_request(dev, seqno1, &dev_priv->render_ring);
+
+ if (HAS_BSD(dev)) {
+ seqno2 = i915_add_request(dev, NULL, I915_GEM_GPU_DOMAINS,
+ &dev_priv->bsd_ring);
+ if (seqno2 == 0)
+ return -ENOMEM;
+
+ ret = i915_wait_request(dev, seqno2, &dev_priv->bsd_ring);
+ if (ret)
+ return ret;
+ }
+
+
+ return ret;
+}
+
+static int
i915_gem_evict_everything(struct drm_device *dev)
{
drm_i915_private_t *dev_priv = dev->dev_private;
- uint32_t seqno;
int ret;
bool lists_empty;
- DRM_INFO("GTT full, evicting everything: "
- "%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);
-
spin_lock(&dev_priv->mm.active_list_lock);
lists_empty = (list_empty(&dev_priv->mm.inactive_list) &&
list_empty(&dev_priv->mm.flushing_list) &&
- list_empty(&dev_priv->mm.active_list));
+ list_empty(&dev_priv->render_ring.active_list) &&
+ (!HAS_BSD(dev)
+ || list_empty(&dev_priv->bsd_ring.active_list)));
spin_unlock(&dev_priv->mm.active_list_lock);
- if (lists_empty) {
- DRM_ERROR("GTT full, but lists empty!\n");
+ if (lists_empty)
return -ENOSPC;
- }
/* Flush everything (on to the inactive lists) and evict */
- i915_gem_flush(dev, I915_GEM_GPU_DOMAINS, I915_GEM_GPU_DOMAINS);
- seqno = i915_add_request(dev, NULL, I915_GEM_GPU_DOMAINS);
- if (seqno == 0)
- return -ENOMEM;
-
- ret = i915_wait_request(dev, seqno);
+ ret = i915_gpu_idle(dev);
if (ret)
return ret;
- ret = i915_gem_evict_from_list(dev, &dev_priv->mm.inactive_list);
+ BUG_ON(!list_empty(&dev_priv->mm.flushing_list));
+
+ ret = i915_gem_evict_from_inactive_list(dev);
if (ret)
return ret;
spin_lock(&dev_priv->mm.active_list_lock);
lists_empty = (list_empty(&dev_priv->mm.inactive_list) &&
list_empty(&dev_priv->mm.flushing_list) &&
- list_empty(&dev_priv->mm.active_list));
+ list_empty(&dev_priv->render_ring.active_list) &&
+ (!HAS_BSD(dev)
+ || list_empty(&dev_priv->bsd_ring.active_list)));
spin_unlock(&dev_priv->mm.active_list_lock);
BUG_ON(!lists_empty);
{
drm_i915_private_t *dev_priv = dev->dev_private;
struct drm_gem_object *obj;
- int have_waited = 0;
int ret;
+ struct intel_ring_buffer *render_ring = &dev_priv->render_ring;
+ struct intel_ring_buffer *bsd_ring = &dev_priv->bsd_ring;
for (;;) {
- i915_gem_retire_requests(dev);
+ i915_gem_retire_requests(dev, render_ring);
+
+ if (HAS_BSD(dev))
+ i915_gem_retire_requests(dev, bsd_ring);
/* If there's an inactive buffer available now, grab it
* and be done.
#if WATCH_LRU
DRM_INFO("%s: evicting %p\n", __func__, obj);
#endif
- obj_priv = obj->driver_private;
+ obj_priv = to_intel_bo(obj);
BUG_ON(obj_priv->pin_count != 0);
BUG_ON(obj_priv->active);
return i915_gem_object_unbind(obj);
}
- if (have_waited)
- return 0;
-
/* If we didn't get anything, but the ring is still processing
* things, wait for the next to finish and hopefully leave us
* a buffer to evict.
*/
- if (!list_empty(&dev_priv->mm.request_list)) {
+ if (!list_empty(&render_ring->request_list)) {
+ struct drm_i915_gem_request *request;
+
+ request = list_first_entry(&render_ring->request_list,
+ struct drm_i915_gem_request,
+ list);
+
+ ret = i915_wait_request(dev,
+ request->seqno, request->ring);
+ if (ret)
+ return ret;
+
+ continue;
+ }
+
+ if (HAS_BSD(dev) && !list_empty(&bsd_ring->request_list)) {
struct drm_i915_gem_request *request;
- request = list_first_entry(&dev_priv->mm.request_list,
+ request = list_first_entry(&bsd_ring->request_list,
struct drm_i915_gem_request,
list);
- ret = i915_wait_request(dev, request->seqno);
+ ret = i915_wait_request(dev,
+ request->seqno, request->ring);
if (ret)
return ret;
- have_waited = 1;
continue;
}
*/
if (!list_empty(&dev_priv->mm.flushing_list)) {
struct drm_i915_gem_object *obj_priv;
- uint32_t seqno;
- obj_priv = list_first_entry(&dev_priv->mm.flushing_list,
- struct drm_i915_gem_object,
- list);
- obj = obj_priv->obj;
-
- i915_gem_flush(dev,
- obj->write_domain,
- obj->write_domain);
- seqno = i915_add_request(dev, NULL, obj->write_domain);
- if (seqno == 0)
- return -ENOMEM;
+ /* Find an object that we can immediately reuse */
+ list_for_each_entry(obj_priv, &dev_priv->mm.flushing_list, list) {
+ obj = &obj_priv->base;
+ if (obj->size >= min_size)
+ break;
- ret = i915_wait_request(dev, seqno);
- if (ret)
- return ret;
+ obj = NULL;
+ }
- have_waited = 1;
- continue;
+ if (obj != NULL) {
+ uint32_t seqno;
+
+ i915_gem_flush_ring(dev,
+ obj->write_domain,
+ obj->write_domain,
+ obj_priv->ring);
+ seqno = i915_add_request(dev, NULL,
+ obj->write_domain,
+ obj_priv->ring);
+ if (seqno == 0)
+ return -ENOMEM;
+ continue;
+ }
}
/* If we didn't do any of the above, there's no single buffer
* large enough to swap out for the new one, so just evict
* everything and start again. (This should be rare.)
*/
- if (!list_empty (&dev_priv->mm.inactive_list)) {
- DRM_INFO("GTT full, evicting inactive buffers\n");
- return i915_gem_evict_from_list(dev,
- &dev_priv->mm.inactive_list);
- } else
+ if (!list_empty (&dev_priv->mm.inactive_list))
+ return i915_gem_evict_from_inactive_list(dev);
+ else
return i915_gem_evict_everything(dev);
}
}
int
-i915_gem_object_get_pages(struct drm_gem_object *obj)
+i915_gem_object_get_pages(struct drm_gem_object *obj,
+ gfp_t gfpmask)
{
- struct drm_i915_gem_object *obj_priv = obj->driver_private;
+ 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;
- int ret;
+
+ BUG_ON(obj_priv->pages_refcount
+ == DRM_I915_GEM_OBJECT_MAX_PAGES_REFCOUNT);
if (obj_priv->pages_refcount++ != 0)
return 0;
BUG_ON(obj_priv->pages != NULL);
obj_priv->pages = drm_calloc_large(page_count, sizeof(struct page *));
if (obj_priv->pages == NULL) {
- DRM_ERROR("Failed to allocate page list\n");
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_mapping_page(mapping, i, NULL);
- if (IS_ERR(page)) {
- ret = PTR_ERR(page);
- i915_gem_object_put_pages(obj);
- return ret;
- }
+ page = read_cache_page_gfp(mapping, i,
+ mapping_gfp_mask (mapping) |
+ __GFP_COLD |
+ gfpmask);
+ if (IS_ERR(page))
+ goto err_pages;
+
obj_priv->pages[i] = page;
}
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 = obj->driver_private;
+ struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
int regnum = obj_priv->fence_reg;
uint64_t val;
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 = obj->driver_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;
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;
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 = obj->driver_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;
I915_WRITE(FENCE_REG_830_0 + (regnum * 4), val);
}
-/**
- * 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.
- */
+static int i915_find_fence_reg(struct drm_device *dev)
+{
+ 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);
+ 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)
{
struct drm_device *dev = obj->dev;
struct drm_i915_private *dev_priv = dev->dev_private;
- struct drm_i915_gem_object *obj_priv = obj->driver_private;
+ struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
struct drm_i915_fence_reg *reg = NULL;
- struct drm_i915_gem_object *old_obj_priv = NULL;
- int i, ret, avail;
+ int ret;
/* Just update our place in the LRU if our fence is getting used. */
if (obj_priv->fence_reg != I915_FENCE_REG_NONE) {
- list_move_tail(&obj_priv->fence_list, &dev_priv->mm.fence_list);
+ reg = &dev_priv->fence_regs[obj_priv->fence_reg];
+ list_move_tail(®->lru_list, &dev_priv->mm.fence_list);
return 0;
}
break;
}
- /* 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)
- break;
-
- old_obj_priv = reg->obj->driver_private;
- if (!old_obj_priv->pin_count)
- avail++;
- }
-
- /* None available, try to steal one or wait for a user to finish */
- if (i == dev_priv->num_fence_regs) {
- struct drm_gem_object *old_obj = NULL;
-
- if (avail == 0)
- return -ENOSPC;
-
- list_for_each_entry(old_obj_priv, &dev_priv->mm.fence_list,
- fence_list) {
- old_obj = old_obj_priv->obj;
-
- if (old_obj_priv->pin_count)
- continue;
-
- /* Take a reference, as otherwise the wait_rendering
- * below may cause the object to get freed out from
- * under us.
- */
- drm_gem_object_reference(old_obj);
-
- /* i915 uses fences for GPU access to tiled buffers */
- if (IS_I965G(dev) || !old_obj_priv->active)
- break;
-
- /* This brings the object to the head of the LRU if it
- * had been written to. The only way this should
- * result in us waiting longer than the expected
- * optimal amount of time is if there was a
- * fence-using buffer later that was read-only.
- */
- i915_gem_object_flush_gpu_write_domain(old_obj);
- ret = i915_gem_object_wait_rendering(old_obj);
- if (ret != 0) {
- drm_gem_object_unreference(old_obj);
- return ret;
- }
-
- break;
- }
-
- /*
- * Zap this virtual mapping so we can set up a fence again
- * for this object next time we need it.
- */
- i915_gem_release_mmap(old_obj);
-
- i = old_obj_priv->fence_reg;
- reg = &dev_priv->fence_regs[i];
-
- old_obj_priv->fence_reg = I915_FENCE_REG_NONE;
- list_del_init(&old_obj_priv->fence_list);
-
- drm_gem_object_unreference(old_obj);
- }
+ ret = i915_find_fence_reg(dev);
+ if (ret < 0)
+ return ret;
- obj_priv->fence_reg = i;
- list_add_tail(&obj_priv->fence_list, &dev_priv->mm.fence_list);
+ obj_priv->fence_reg = ret;
+ reg = &dev_priv->fence_regs[obj_priv->fence_reg];
+ list_add_tail(®->lru_list, &dev_priv->mm.fence_list);
reg->obj = obj;
- if (IS_I965G(dev))
+ if (IS_GEN6(dev))
+ sandybridge_write_fence_reg(reg);
+ else if (IS_I965G(dev))
i965_write_fence_reg(reg);
else if (IS_I9XX(dev))
i915_write_fence_reg(reg);
else
i830_write_fence_reg(reg);
+ trace_i915_gem_object_get_fence(obj, obj_priv->fence_reg,
+ obj_priv->tiling_mode);
+
return 0;
}
{
struct drm_device *dev = obj->dev;
drm_i915_private_t *dev_priv = dev->dev_private;
- struct drm_i915_gem_object *obj_priv = obj->driver_private;
-
- if (IS_I965G(dev))
+ 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];
+
+ if (IS_GEN6(dev)) {
+ I915_WRITE64(FENCE_REG_SANDYBRIDGE_0 +
+ (obj_priv->fence_reg * 8), 0);
+ } else if (IS_I965G(dev)) {
I915_WRITE64(FENCE_REG_965_0 + (obj_priv->fence_reg * 8), 0);
- else {
+ } else {
uint32_t fence_reg;
if (obj_priv->fence_reg < 8)
I915_WRITE(fence_reg, 0);
}
- dev_priv->fence_regs[obj_priv->fence_reg].obj = NULL;
+ reg->obj = NULL;
obj_priv->fence_reg = I915_FENCE_REG_NONE;
- list_del_init(&obj_priv->fence_list);
+ list_del_init(®->lru_list);
}
/**
i915_gem_object_put_fence_reg(struct drm_gem_object *obj)
{
struct drm_device *dev = obj->dev;
- struct drm_i915_gem_object *obj_priv = obj->driver_private;
+ struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
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.
int ret;
i915_gem_object_flush_gpu_write_domain(obj);
- i915_gem_object_flush_gtt_write_domain(obj);
ret = i915_gem_object_wait_rendering(obj);
if (ret != 0)
return ret;
}
+ i915_gem_object_flush_gtt_write_domain(obj);
i915_gem_clear_fence_reg (obj);
return 0;
{
struct drm_device *dev = obj->dev;
drm_i915_private_t *dev_priv = dev->dev_private;
- struct drm_i915_gem_object *obj_priv = obj->driver_private;
+ struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
struct drm_mm_node *free_space;
- bool retry_alloc = false;
+ gfp_t gfpmask = __GFP_NORETRY | __GFP_NOWARN;
int ret;
- if (dev_priv->mm.suspended)
- return -EBUSY;
-
- if (obj_priv->madv == I915_MADV_DONTNEED) {
+ if (obj_priv->madv != I915_MADV_WILLNEED) {
DRM_ERROR("Attempting to bind a purgeable object\n");
return -EINVAL;
}
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);
DRM_INFO("%s: GTT full, evicting something\n", __func__);
#endif
ret = i915_gem_evict_something(dev, obj->size);
- if (ret != 0) {
- if (ret != -ERESTARTSYS)
- DRM_ERROR("Failed to evict a buffer %d\n", ret);
+ if (ret)
return ret;
- }
+
goto search_free;
}
DRM_INFO("Binding object of size %zd at 0x%08x\n",
obj->size, obj_priv->gtt_offset);
#endif
- if (retry_alloc) {
- i915_gem_object_set_page_gfp_mask (obj,
- i915_gem_object_get_page_gfp_mask (obj) & ~__GFP_NORETRY);
- }
- ret = i915_gem_object_get_pages(obj);
- if (retry_alloc) {
- i915_gem_object_set_page_gfp_mask (obj,
- i915_gem_object_get_page_gfp_mask (obj) | __GFP_NORETRY);
- }
+ ret = i915_gem_object_get_pages(obj, gfpmask);
if (ret) {
drm_mm_put_block(obj_priv->gtt_space);
obj_priv->gtt_space = NULL;
/* first try to clear up some space from the GTT */
ret = i915_gem_evict_something(dev, obj->size);
if (ret) {
- if (ret != -ERESTARTSYS)
- DRM_ERROR("Failed to allocate space for backing pages %d\n", ret);
-
/* now try to shrink everyone else */
- if (! retry_alloc) {
- retry_alloc = true;
- goto search_free;
+ if (gfpmask) {
+ gfpmask = 0;
+ goto search_free;
}
return ret;
obj_priv->gtt_space = NULL;
ret = i915_gem_evict_something(dev, obj->size);
- if (ret) {
- if (ret != -ERESTARTSYS)
- DRM_ERROR("Failed to allocate space to bind AGP: %d\n", ret);
+ if (ret)
return ret;
- }
goto search_free;
}
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 = obj->driver_private;
+ 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
if (obj_priv->pages == NULL)
return;
+ trace_i915_gem_object_clflush(obj);
+
drm_clflush_pages(obj_priv->pages, obj->size / PAGE_SIZE);
}
i915_gem_object_flush_gpu_write_domain(struct drm_gem_object *obj)
{
struct drm_device *dev = obj->dev;
- uint32_t seqno;
+ uint32_t old_write_domain;
+ struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
if ((obj->write_domain & I915_GEM_GPU_DOMAINS) == 0)
return;
/* Queue the GPU write cache flushing we need. */
+ old_write_domain = obj->write_domain;
i915_gem_flush(dev, 0, obj->write_domain);
- seqno = i915_add_request(dev, NULL, obj->write_domain);
- obj->write_domain = 0;
- i915_gem_object_move_to_active(obj, seqno);
+ (void) i915_add_request(dev, NULL, obj->write_domain, obj_priv->ring);
+ BUG_ON(obj->write_domain);
+
+ trace_i915_gem_object_change_domain(obj,
+ obj->read_domains,
+ old_write_domain);
}
/** 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;
* 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. */
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);
+}
+
+void
+i915_gem_object_flush_write_domain(struct drm_gem_object *obj)
+{
+ switch (obj->write_domain) {
+ case I915_GEM_DOMAIN_GTT:
+ i915_gem_object_flush_gtt_write_domain(obj);
+ break;
+ case I915_GEM_DOMAIN_CPU:
+ i915_gem_object_flush_cpu_write_domain(obj);
+ break;
+ default:
+ i915_gem_object_flush_gpu_write_domain(obj);
+ break;
+ }
}
/**
int
i915_gem_object_set_to_gtt_domain(struct drm_gem_object *obj, int write)
{
- struct drm_i915_gem_object *obj_priv = obj->driver_private;
+ 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 (ret != 0)
return ret;
+ old_write_domain = obj->write_domain;
+ old_read_domains = obj->read_domains;
+
/* If we're writing through the GTT domain, then CPU and GPU caches
* will need to be invalidated at next use.
*/
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)
+{
+ struct drm_device *dev = obj->dev;
+ 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;
+
+ i915_gem_object_flush_gpu_write_domain(obj);
+
+ /* Wait on any GPU rendering and flushing to occur. */
+ 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,
+ 0,
+ obj_priv->ring);
+ if (ret != 0)
+ return ret;
+ }
+
+ i915_gem_object_flush_cpu_write_domain(obj);
+
+ 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;
+ 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;
}
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;
i915_gem_object_flush_gpu_write_domain(obj);
*/
i915_gem_object_set_to_full_cpu_read_domain(obj);
+ 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->write_domain = I915_GEM_DOMAIN_CPU;
}
+ trace_i915_gem_object_change_domain(obj,
+ old_read_domains,
+ old_write_domain);
+
return 0;
}
i915_gem_object_set_to_gpu_domain(struct drm_gem_object *obj)
{
struct drm_device *dev = obj->dev;
- struct drm_i915_gem_object *obj_priv = obj->driver_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);
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'
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);
}
/**
static void
i915_gem_object_set_to_full_cpu_read_domain(struct drm_gem_object *obj)
{
- struct drm_i915_gem_object *obj_priv = obj->driver_private;
+ struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
if (!obj_priv->page_cpu_valid)
return;
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 = obj->driver_private;
+ 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)
*/
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;
}
static int
i915_gem_object_pin_and_relocate(struct drm_gem_object *obj,
struct drm_file *file_priv,
- struct drm_i915_gem_exec_object *entry,
+ 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 = obj->driver_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);
+ if (ret != 0) {
+ i915_gem_object_unpin(obj);
+ return ret;
+ }
+ }
+
entry->offset = obj_priv->gtt_offset;
/* Apply the relocations, using the GTT aperture to avoid cache
i915_gem_object_unpin(obj);
return -EBADF;
}
- target_obj_priv = target_obj->driver_private;
+ target_obj_priv = to_intel_bo(target_obj);
#if WATCH_RELOC
DRM_INFO("%s: obj %p offset %08x target %d "
}
/* 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: "
return 0;
}
-/** Dispatch a batchbuffer to the ring
- */
-static int
-i915_dispatch_gem_execbuffer(struct drm_device *dev,
- struct drm_i915_gem_execbuffer *exec,
- struct drm_clip_rect *cliprects,
- uint64_t exec_offset)
-{
- drm_i915_private_t *dev_priv = dev->dev_private;
- int nbox = exec->num_cliprects;
- int i = 0, count;
- uint32_t exec_start, exec_len;
- RING_LOCALS;
-
- exec_start = (uint32_t) exec_offset + exec->batch_start_offset;
- exec_len = (uint32_t) exec->batch_len;
-
- count = nbox ? nbox : 1;
-
- for (i = 0; i < count; i++) {
- if (i < nbox) {
- int ret = i915_emit_box(dev, cliprects, i,
- exec->DR1, exec->DR4);
- if (ret)
- return ret;
- }
-
- if (IS_I830(dev) || IS_845G(dev)) {
- BEGIN_LP_RING(4);
- OUT_RING(MI_BATCH_BUFFER);
- OUT_RING(exec_start | MI_BATCH_NON_SECURE);
- OUT_RING(exec_start + exec_len - 4);
- OUT_RING(0);
- ADVANCE_LP_RING();
- } else {
- BEGIN_LP_RING(2);
- if (IS_I965G(dev)) {
- OUT_RING(MI_BATCH_BUFFER_START |
- (2 << 6) |
- MI_BATCH_NON_SECURE_I965);
- OUT_RING(exec_start);
- } else {
- OUT_RING(MI_BATCH_BUFFER_START |
- (2 << 6));
- OUT_RING(exec_start | MI_BATCH_NON_SECURE);
- }
- ADVANCE_LP_RING();
- }
- }
-
- /* XXX breadcrumb */
- return 0;
-}
-
/* Throttle our rendering by waiting until the ring has completed our requests
* emitted over 20 msec ago.
*
if (time_after_eq(request->emitted_jiffies, recent_enough))
break;
- ret = i915_wait_request(dev, request->seqno);
+ ret = i915_wait_request(dev, request->seqno, request->ring);
if (ret != 0)
break;
}
}
static int
-i915_gem_get_relocs_from_user(struct drm_i915_gem_exec_object *exec_list,
+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)
{
}
*relocs = drm_calloc_large(reloc_count, sizeof(**relocs));
- if (*relocs == NULL)
+ 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;
}
static int
-i915_gem_put_relocs_to_user(struct drm_i915_gem_exec_object *exec_list,
+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;
}
static int
-i915_gem_check_execbuffer (struct drm_i915_gem_execbuffer *exec,
+i915_gem_check_execbuffer (struct drm_i915_gem_execbuffer2 *exec,
uint64_t exec_offset)
{
uint32_t exec_start, exec_len;
return 0;
}
+static 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;
+}
+
int
-i915_gem_execbuffer(struct drm_device *dev, void *data,
- struct drm_file *file_priv)
+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_i915_gem_execbuffer *args = data;
- struct drm_i915_gem_exec_object *exec_list = NULL;
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;
- int ret, ret2, i, pinned = 0;
+ struct drm_i915_gem_relocation_entry *relocs = NULL;
+ int ret = 0, ret2, i, pinned = 0;
uint64_t exec_offset;
uint32_t seqno, flush_domains, reloc_index;
- int pin_tries;
+ 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;
}
- /* Copy in the exec list from userland */
- exec_list = drm_calloc_large(sizeof(*exec_list), args->buffer_count);
- object_list = drm_calloc_large(sizeof(*object_list), args->buffer_count);
- if (exec_list == NULL || object_list == NULL) {
- DRM_ERROR("Failed to allocate exec or object list "
- "for %d buffers\n",
+ 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;
}
- 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);
- goto pre_mutex_err;
- }
if (args->num_cliprects != 0) {
cliprects = kcalloc(args->num_cliprects, sizeof(*cliprects),
GFP_KERNEL);
- if (cliprects == NULL)
+ if (cliprects == NULL) {
+ ret = -ENOMEM;
goto pre_mutex_err;
+ }
ret = copy_from_user(cliprects,
(struct drm_clip_rect __user *)
i915_verify_inactive(dev, __FILE__, __LINE__);
if (atomic_read(&dev_priv->mm.wedged)) {
- DRM_ERROR("Execbuf while wedged\n");
mutex_unlock(&dev->struct_mutex);
ret = -EIO;
goto pre_mutex_err;
}
if (dev_priv->mm.suspended) {
- DRM_ERROR("Execbuf while VT-switched.\n");
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 = -EBADF;
goto err;
}
- obj_priv = object_list[i]->driver_private;
+ 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 = -EBADF;
goto err;
}
obj_priv->in_execbuffer = true;
+ flips += atomic_read(&obj_priv->pending_flip);
}
- /* Pin and relocate */
- for (pin_tries = 0; ; pin_tries++) {
+ 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;
if (ret != -ENOSPC || pin_tries >= 1) {
if (ret != -ERESTARTSYS) {
unsigned long long total_size = 0;
- for (i = 0; i < args->buffer_count; i++)
+ int num_fences = 0;
+ for (i = 0; i < args->buffer_count; i++) {
+ obj_priv = object_list[i]->driver_private;
+
total_size += object_list[i]->size;
- DRM_ERROR("Failed to pin buffer %d of %d, total %llu bytes: %d\n",
+ 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, ret);
+ total_size, num_fences,
+ ret);
DRM_ERROR("%d objects [%d pinned], "
"%d object bytes [%d pinned], "
"%d/%d gtt bytes\n",
i915_gem_flush(dev,
dev->invalidate_domains,
dev->flush_domains);
- if (dev->flush_domains)
+ if (dev->flush_domains & I915_GEM_GPU_DOMAINS) {
(void)i915_add_request(dev, file_priv,
- dev->flush_domains);
+ dev->flush_domains,
+ &dev_priv->render_ring);
+
+ if (HAS_BSD(dev))
+ (void)i915_add_request(dev, file_priv,
+ dev->flush_domains,
+ &dev_priv->bsd_ring);
+ }
}
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__);
#endif
/* Exec the batchbuffer */
- ret = i915_dispatch_gem_execbuffer(dev, args, cliprects, exec_offset);
+ 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
*/
- flush_domains = i915_retire_commands(dev);
+ flush_domains = i915_retire_commands(dev, ring);
i915_verify_inactive(dev, __FILE__, __LINE__);
* *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, flush_domains);
+ seqno = i915_add_request(dev, file_priv, flush_domains, ring);
BUG_ON(seqno == 0);
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, seqno);
+ i915_gem_object_move_to_active(obj, seqno, ring);
#if WATCH_LRU
DRM_INFO("%s: move to exec list %p\n", __func__, obj);
#endif
for (i = 0; i < args->buffer_count; i++) {
if (object_list[i]) {
- obj_priv = object_list[i]->driver_private;
+ 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);
+
+ 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 (!IS_I965G(dev))
+ 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,
}
}
- /* 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);
+ drm_free_large(exec_list);
+ drm_free_large(exec2_list);
+ return ret;
+}
- if (ret == 0)
- ret = ret2;
+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;
}
-pre_mutex_err:
- drm_free_large(object_list);
- drm_free_large(exec_list);
- kfree(cliprects);
+ 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;
}
i915_gem_object_pin(struct drm_gem_object *obj, uint32_t alignment)
{
struct drm_device *dev = obj->dev;
- struct drm_i915_gem_object *obj_priv = obj->driver_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) {
- ret = i915_gem_object_bind_to_gtt(obj, alignment);
- if (ret != 0) {
- if (ret != -EBUSY && ret != -ERESTARTSYS)
- DRM_ERROR("Failure to bind: %d\n", ret);
- return ret;
+
+ if (obj_priv->gtt_space != NULL) {
+ if (alignment == 0)
+ alignment = i915_gem_get_gtt_alignment(obj);
+ if (obj_priv->gtt_offset & (alignment - 1)) {
+ ret = i915_gem_object_unbind(obj);
+ if (ret)
+ return ret;
}
}
- /*
- * Pre-965 chips need a fence register set up in order to
- * properly handle tiled surfaces.
- */
- if (!IS_I965G(dev) && obj_priv->tiling_mode != I915_TILING_NONE) {
- ret = i915_gem_object_get_fence_reg(obj);
- if (ret != 0) {
- if (ret != -EBUSY && ret != -ERESTARTSYS)
- DRM_ERROR("Failure to install fence: %d\n",
- 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,
{
struct drm_device *dev = obj->dev;
drm_i915_private_t *dev_priv = dev->dev_private;
- struct drm_i915_gem_object *obj_priv = obj->driver_private;
+ struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
i915_verify_inactive(dev, __FILE__, __LINE__);
obj_priv->pin_count--;
mutex_unlock(&dev->struct_mutex);
return -EBADF;
}
- obj_priv = obj->driver_private;
+ obj_priv = to_intel_bo(obj);
- if (obj_priv->madv == I915_MADV_DONTNEED) {
- DRM_ERROR("Attempting to pin a I915_MADV_DONTNEED buffer\n");
+ 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;
return -EBADF;
}
- obj_priv = obj->driver_private;
+ 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);
struct drm_i915_gem_busy *args = data;
struct drm_gem_object *obj;
struct drm_i915_gem_object *obj_priv;
+ drm_i915_private_t *dev_priv = dev->dev_private;
obj = drm_gem_object_lookup(dev, file_priv, args->handle);
if (obj == NULL) {
* actually unmasked, and our working set ends up being larger than
* required.
*/
- i915_gem_retire_requests(dev);
+ i915_gem_retire_requests(dev, &dev_priv->render_ring);
+
+ if (HAS_BSD(dev))
+ i915_gem_retire_requests(dev, &dev_priv->bsd_ring);
- obj_priv = obj->driver_private;
+ obj_priv = to_intel_bo(obj);
/* Don't count being on the flushing list against the object being
* done. Otherwise, a buffer left on the flushing list but not getting
* flushed (because nobody's flushing that domain) won't ever return
}
mutex_lock(&dev->struct_mutex);
- obj_priv = obj->driver_private;
+ obj_priv = to_intel_bo(obj);
if (obj_priv->pin_count) {
drm_gem_object_unreference(obj);
return -EINVAL;
}
- obj_priv->madv = args->madv;
- args->retained = obj_priv->gtt_space != NULL;
+ 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;
}
-int i915_gem_init_object(struct drm_gem_object *obj)
+struct drm_gem_object * i915_gem_alloc_object(struct drm_device *dev,
+ size_t size)
{
- struct drm_i915_gem_object *obj_priv;
+ struct drm_i915_gem_object *obj;
- obj_priv = kzalloc(sizeof(*obj_priv), GFP_KERNEL);
- if (obj_priv == NULL)
- return -ENOMEM;
+ obj = kzalloc(sizeof(*obj), GFP_KERNEL);
+ if (obj == NULL)
+ return NULL;
- /*
- * We've just allocated pages from the kernel,
- * so they've just been written by the CPU with
- * zeros. They'll need to be clflushed before we
- * use them with the GPU.
- */
- obj->write_domain = I915_GEM_DOMAIN_CPU;
- obj->read_domains = I915_GEM_DOMAIN_CPU;
+ if (drm_gem_object_init(dev, &obj->base, size) != 0) {
+ kfree(obj);
+ return NULL;
+ }
- obj_priv->agp_type = AGP_USER_MEMORY;
+ obj->base.write_domain = I915_GEM_DOMAIN_CPU;
+ obj->base.read_domains = I915_GEM_DOMAIN_CPU;
- obj->driver_private = obj_priv;
- obj_priv->obj = obj;
- obj_priv->fence_reg = I915_FENCE_REG_NONE;
- INIT_LIST_HEAD(&obj_priv->list);
- INIT_LIST_HEAD(&obj_priv->fence_list);
- obj_priv->madv = I915_MADV_WILLNEED;
+ 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;
}
void i915_gem_free_object(struct drm_gem_object *obj)
{
struct drm_device *dev = obj->dev;
- struct drm_i915_gem_object *obj_priv = obj->driver_private;
+ 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->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->driver_private);
+ kfree(obj_priv);
}
-/** Unbinds all objects that are on the given buffer list. */
+/** Unbinds all inactive objects. */
static int
-i915_gem_evict_from_list(struct drm_device *dev, struct list_head *head)
+i915_gem_evict_from_inactive_list(struct drm_device *dev)
{
- struct drm_gem_object *obj;
- struct drm_i915_gem_object *obj_priv;
- int ret;
+ drm_i915_private_t *dev_priv = dev->dev_private;
- while (!list_empty(head)) {
- obj_priv = list_first_entry(head,
- struct drm_i915_gem_object,
- list);
- obj = obj_priv->obj;
+ while (!list_empty(&dev_priv->mm.inactive_list)) {
+ struct drm_gem_object *obj;
+ int ret;
- if (obj_priv->pin_count != 0) {
- DRM_ERROR("Pinned object in unbind list\n");
- mutex_unlock(&dev->struct_mutex);
- return -EINVAL;
- }
+ obj = &list_first_entry(&dev_priv->mm.inactive_list,
+ struct drm_i915_gem_object,
+ list)->base;
ret = i915_gem_object_unbind(obj);
if (ret != 0) {
- DRM_ERROR("Error unbinding object in LeaveVT: %d\n",
- ret);
- mutex_unlock(&dev->struct_mutex);
+ DRM_ERROR("Error unbinding object: %d\n", ret);
return ret;
}
}
-
return 0;
}
i915_gem_idle(struct drm_device *dev)
{
drm_i915_private_t *dev_priv = dev->dev_private;
- uint32_t seqno, cur_seqno, last_seqno;
- int stuck, ret;
+ int ret;
mutex_lock(&dev->struct_mutex);
- if (dev_priv->mm.suspended || dev_priv->ring.ring_obj == NULL) {
+ 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;
}
- /* Hack! Don't let anybody do execbuf while we don't control the chip.
- * We need to replace this with a semaphore, or something.
- */
- dev_priv->mm.suspended = 1;
- del_timer(&dev_priv->hangcheck_timer);
-
- /* Cancel the retire work handler, wait for it to finish if running
- */
- mutex_unlock(&dev->struct_mutex);
- cancel_delayed_work_sync(&dev_priv->mm.retire_work);
- mutex_lock(&dev->struct_mutex);
-
- i915_kernel_lost_context(dev);
-
- /* Flush the GPU along with all non-CPU write domains
- */
- i915_gem_flush(dev, I915_GEM_GPU_DOMAINS, I915_GEM_GPU_DOMAINS);
- seqno = i915_add_request(dev, NULL, I915_GEM_GPU_DOMAINS);
-
- if (seqno == 0) {
+ ret = i915_gpu_idle(dev);
+ if (ret) {
mutex_unlock(&dev->struct_mutex);
- return -ENOMEM;
+ return ret;
}
- dev_priv->mm.waiting_gem_seqno = seqno;
- last_seqno = 0;
- stuck = 0;
- for (;;) {
- cur_seqno = i915_get_gem_seqno(dev);
- if (i915_seqno_passed(cur_seqno, seqno))
- break;
- if (last_seqno == cur_seqno) {
- if (stuck++ > 100) {
- DRM_ERROR("hardware wedged\n");
- atomic_set(&dev_priv->mm.wedged, 1);
- DRM_WAKEUP(&dev_priv->irq_queue);
- break;
- }
+ /* Under UMS, be paranoid and evict. */
+ if (!drm_core_check_feature(dev, DRIVER_MODESET)) {
+ ret = i915_gem_evict_from_inactive_list(dev);
+ if (ret) {
+ mutex_unlock(&dev->struct_mutex);
+ return ret;
}
- msleep(10);
- last_seqno = cur_seqno;
- }
- dev_priv->mm.waiting_gem_seqno = 0;
-
- i915_gem_retire_requests(dev);
-
- spin_lock(&dev_priv->mm.active_list_lock);
- if (!atomic_read(&dev_priv->mm.wedged)) {
- /* Active and flushing should now be empty as we've
- * waited for a sequence higher than any pending execbuffer
- */
- WARN_ON(!list_empty(&dev_priv->mm.active_list));
- WARN_ON(!list_empty(&dev_priv->mm.flushing_list));
- /* Request should now be empty as we've also waited
- * for the last request in the list
- */
- WARN_ON(!list_empty(&dev_priv->mm.request_list));
}
- /* Empty the active and flushing lists to inactive. If there's
- * anything left at this point, it means that we're wedged and
- * nothing good's going to happen by leaving them there. So strip
- * the GPU domains and just stuff them onto inactive.
+ /* 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!
*/
- while (!list_empty(&dev_priv->mm.active_list)) {
- struct drm_i915_gem_object *obj_priv;
-
- obj_priv = list_first_entry(&dev_priv->mm.active_list,
- struct drm_i915_gem_object,
- list);
- obj_priv->obj->write_domain &= ~I915_GEM_GPU_DOMAINS;
- i915_gem_object_move_to_inactive(obj_priv->obj);
- }
- spin_unlock(&dev_priv->mm.active_list_lock);
-
- 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->obj->write_domain &= ~I915_GEM_GPU_DOMAINS;
- i915_gem_object_move_to_inactive(obj_priv->obj);
- }
-
-
- /* Move all inactive buffers out of the GTT. */
- ret = i915_gem_evict_from_list(dev, &dev_priv->mm.inactive_list);
- WARN_ON(!list_empty(&dev_priv->mm.inactive_list));
- if (ret) {
- mutex_unlock(&dev->struct_mutex);
- return ret;
- }
+ dev_priv->mm.suspended = 1;
+ del_timer(&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_hws(struct drm_device *dev)
+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;
- /* If we need a physical address for the status page, it's already
- * initialized at driver load time.
- */
- if (!I915_NEED_GFX_HWS(dev))
- return 0;
-
- obj = drm_gem_object_alloc(dev, 4096);
+ obj = i915_gem_alloc_object(dev, 4096);
if (obj == NULL) {
- DRM_ERROR("Failed to allocate status page\n");
- return -ENOMEM;
+ DRM_ERROR("Failed to allocate seqno page\n");
+ ret = -ENOMEM;
+ goto err;
}
- obj_priv = obj->driver_private;
+ obj_priv = to_intel_bo(obj);
obj_priv->agp_type = AGP_USER_CACHED_MEMORY;
ret = i915_gem_object_pin(obj, 4096);
- if (ret != 0) {
- drm_gem_object_unreference(obj);
- return ret;
- }
+ if (ret)
+ goto err_unref;
- dev_priv->status_gfx_addr = obj_priv->gtt_offset;
+ 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->hw_status_page = kmap(obj_priv->pages[0]);
- if (dev_priv->hw_status_page == NULL) {
- DRM_ERROR("Failed to map status page.\n");
- memset(&dev_priv->hws_map, 0, sizeof(dev_priv->hws_map));
- i915_gem_object_unpin(obj);
- drm_gem_object_unreference(obj);
- return -EINVAL;
- }
- dev_priv->hws_obj = obj;
- memset(dev_priv->hw_status_page, 0, PAGE_SIZE);
- I915_WRITE(HWS_PGA, dev_priv->status_gfx_addr);
- I915_READ(HWS_PGA); /* posting read */
- DRM_DEBUG("hws offset: 0x%08x\n", dev_priv->status_gfx_addr);
+ 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_hws(struct drm_device *dev)
+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;
- if (dev_priv->hws_obj == NULL)
- return;
-
- obj = dev_priv->hws_obj;
- obj_priv = obj->driver_private;
-
+ 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->hws_obj = NULL;
+ dev_priv->seqno_obj = NULL;
- memset(&dev_priv->hws_map, 0, sizeof(dev_priv->hws_map));
- dev_priv->hw_status_page = NULL;
-
- /* Write high address into HWS_PGA when disabling. */
- I915_WRITE(HWS_PGA, 0x1ffff000);
+ dev_priv->seqno_page = NULL;
}
int
i915_gem_init_ringbuffer(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;
- drm_i915_ring_buffer_t *ring = &dev_priv->ring;
int ret;
- u32 head;
- ret = i915_gem_init_hws(dev);
- if (ret != 0)
- return ret;
+ dev_priv->render_ring = render_ring;
- obj = drm_gem_object_alloc(dev, 128 * 1024);
- if (obj == NULL) {
- DRM_ERROR("Failed to allocate ringbuffer\n");
- i915_gem_cleanup_hws(dev);
- return -ENOMEM;
+ if (!I915_NEED_GFX_HWS(dev)) {
+ dev_priv->render_ring.status_page.page_addr
+ = dev_priv->status_page_dmah->vaddr;
+ memset(dev_priv->render_ring.status_page.page_addr,
+ 0, PAGE_SIZE);
}
- obj_priv = obj->driver_private;
- ret = i915_gem_object_pin(obj, 4096);
- if (ret != 0) {
- drm_gem_object_unreference(obj);
- i915_gem_cleanup_hws(dev);
- return ret;
+ if (HAS_PIPE_CONTROL(dev)) {
+ ret = i915_gem_init_pipe_control(dev);
+ if (ret)
+ return ret;
}
- /* Set up the kernel mapping for the ring. */
- ring->Size = obj->size;
-
- ring->map.offset = dev->agp->base + obj_priv->gtt_offset;
- ring->map.size = obj->size;
- ring->map.type = 0;
- ring->map.flags = 0;
- ring->map.mtrr = 0;
-
- drm_core_ioremap_wc(&ring->map, dev);
- if (ring->map.handle == NULL) {
- DRM_ERROR("Failed to map ringbuffer.\n");
- memset(&dev_priv->ring, 0, sizeof(dev_priv->ring));
- i915_gem_object_unpin(obj);
- drm_gem_object_unreference(obj);
- i915_gem_cleanup_hws(dev);
- return -EINVAL;
- }
- ring->ring_obj = obj;
- ring->virtual_start = ring->map.handle;
-
- /* Stop the ring if it's running. */
- I915_WRITE(PRB0_CTL, 0);
- I915_WRITE(PRB0_TAIL, 0);
- I915_WRITE(PRB0_HEAD, 0);
-
- /* Initialize the ring. */
- I915_WRITE(PRB0_START, obj_priv->gtt_offset);
- head = I915_READ(PRB0_HEAD) & HEAD_ADDR;
-
- /* G45 ring initialization fails to reset head to zero */
- if (head != 0) {
- DRM_ERROR("Ring head not reset to zero "
- "ctl %08x head %08x tail %08x start %08x\n",
- I915_READ(PRB0_CTL),
- I915_READ(PRB0_HEAD),
- I915_READ(PRB0_TAIL),
- I915_READ(PRB0_START));
- I915_WRITE(PRB0_HEAD, 0);
-
- DRM_ERROR("Ring head forced to zero "
- "ctl %08x head %08x tail %08x start %08x\n",
- I915_READ(PRB0_CTL),
- I915_READ(PRB0_HEAD),
- I915_READ(PRB0_TAIL),
- I915_READ(PRB0_START));
- }
-
- I915_WRITE(PRB0_CTL,
- ((obj->size - 4096) & RING_NR_PAGES) |
- RING_NO_REPORT |
- RING_VALID);
-
- head = I915_READ(PRB0_HEAD) & HEAD_ADDR;
-
- /* If the head is still not zero, the ring is dead */
- if (head != 0) {
- DRM_ERROR("Ring initialization failed "
- "ctl %08x head %08x tail %08x start %08x\n",
- I915_READ(PRB0_CTL),
- I915_READ(PRB0_HEAD),
- I915_READ(PRB0_TAIL),
- I915_READ(PRB0_START));
- return -EIO;
- }
+ ret = intel_init_ring_buffer(dev, &dev_priv->render_ring);
+ if (ret)
+ goto cleanup_pipe_control;
- /* Update our cache of the ring state */
- if (!drm_core_check_feature(dev, DRIVER_MODESET))
- i915_kernel_lost_context(dev);
- else {
- ring->head = I915_READ(PRB0_HEAD) & HEAD_ADDR;
- ring->tail = I915_READ(PRB0_TAIL) & TAIL_ADDR;
- ring->space = ring->head - (ring->tail + 8);
- if (ring->space < 0)
- ring->space += ring->Size;
+ if (HAS_BSD(dev)) {
+ dev_priv->bsd_ring = bsd_ring;
+ ret = intel_init_ring_buffer(dev, &dev_priv->bsd_ring);
+ if (ret)
+ goto cleanup_render_ring;
}
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
{
drm_i915_private_t *dev_priv = dev->dev_private;
- if (dev_priv->ring.ring_obj == NULL)
- return;
-
- drm_core_ioremapfree(&dev_priv->ring.map, dev);
-
- i915_gem_object_unpin(dev_priv->ring.ring_obj);
- drm_gem_object_unreference(dev_priv->ring.ring_obj);
- dev_priv->ring.ring_obj = NULL;
- memset(&dev_priv->ring, 0, sizeof(dev_priv->ring));
-
- i915_gem_cleanup_hws(dev);
+ 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
}
spin_lock(&dev_priv->mm.active_list_lock);
- BUG_ON(!list_empty(&dev_priv->mm.active_list));
+ BUG_ON(!list_empty(&dev_priv->render_ring.active_list));
+ BUG_ON(HAS_BSD(dev) && !list_empty(&dev_priv->bsd_ring.active_list));
spin_unlock(&dev_priv->mm.active_list_lock);
BUG_ON(!list_empty(&dev_priv->mm.flushing_list));
BUG_ON(!list_empty(&dev_priv->mm.inactive_list));
- BUG_ON(!list_empty(&dev_priv->mm.request_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);
drm_irq_install(dev);
i915_gem_leavevt_ioctl(struct drm_device *dev, void *data,
struct drm_file *file_priv)
{
- int ret;
-
if (drm_core_check_feature(dev, DRIVER_MODESET))
return 0;
- ret = i915_gem_idle(dev);
drm_irq_uninstall(dev);
-
- return ret;
+ return i915_gem_idle(dev);
}
void
drm_i915_private_t *dev_priv = dev->dev_private;
spin_lock_init(&dev_priv->mm.active_list_lock);
- INIT_LIST_HEAD(&dev_priv->mm.active_list);
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.request_list);
INIT_LIST_HEAD(&dev_priv->mm.fence_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);
- dev_priv->mm.next_gem_seqno = 1;
-
spin_lock(&shrink_list_lock);
list_add(&dev_priv->mm.shrink_list, &shrink_list);
spin_unlock(&shrink_list_lock);
/* Old X drivers will take 0-2 for front, back, depth buffers */
- dev_priv->fence_reg_start = 3;
+ if (!drm_core_check_feature(dev, DRIVER_MODESET))
+ dev_priv->fence_reg_start = 3;
if (IS_I965G(dev) || IS_I945G(dev) || IS_I945GM(dev) || IS_G33(dev))
dev_priv->num_fence_regs = 16;
for (i = 0; i < 8; i++)
I915_WRITE(FENCE_REG_945_8 + (i * 4), 0);
}
-
i915_gem_detect_bit_6_swizzle(dev);
+ init_waitqueue_head(&dev_priv->pending_flip_queue);
}
/*
phys_obj->id = id;
- phys_obj->handle = drm_pci_alloc(dev, size, 0, 0xffffffff);
+ phys_obj->handle = drm_pci_alloc(dev, size, 0);
if (!phys_obj->handle) {
ret = -ENOMEM;
goto kfree_obj;
int ret;
int page_count;
- obj_priv = obj->driver_private;
+ obj_priv = to_intel_bo(obj);
if (!obj_priv->phys_obj)
return;
- ret = i915_gem_object_get_pages(obj);
+ ret = i915_gem_object_get_pages(obj, 0);
if (ret)
goto out;
if (id > I915_MAX_PHYS_OBJECT)
return -EINVAL;
- obj_priv = obj->driver_private;
+ obj_priv = to_intel_bo(obj);
if (obj_priv->phys_obj) {
if (obj_priv->phys_obj->id == id)
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);
+ ret = i915_gem_object_get_pages(obj, 0);
if (ret) {
DRM_ERROR("failed to get page list\n");
goto out;
struct drm_i915_gem_pwrite *args,
struct drm_file *file_priv)
{
- struct drm_i915_gem_object *obj_priv = obj->driver_private;
+ 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("obj_addr %p, %lld\n", obj_addr, args->size);
+ 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;
mutex_unlock(&dev->struct_mutex);
}
-/* Immediately discard the backing storage */
-static void
-i915_gem_object_truncate(struct drm_gem_object *obj)
+static int
+i915_gpu_is_active(struct drm_device *dev)
{
- struct inode *inode;
+ drm_i915_private_t *dev_priv = dev->dev_private;
+ int lists_empty;
- inode = obj->filp->f_path.dentry->d_inode;
+ spin_lock(&dev_priv->mm.active_list_lock);
+ 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);
+ spin_unlock(&dev_priv->mm.active_list_lock);
- mutex_lock(&inode->i_mutex);
- truncate_inode_pages(inode->i_mapping, 0);
- mutex_unlock(&inode->i_mutex);
+ return !lists_empty;
}
static int
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) {
continue;
spin_unlock(&shrink_list_lock);
+ i915_gem_retire_requests(dev, &dev_priv->render_ring);
- i915_gem_retire_requests(dev);
+ if (HAS_BSD(dev))
+ i915_gem_retire_requests(dev, &dev_priv->bsd_ring);
list_for_each_entry_safe(obj_priv, next_obj,
&dev_priv->mm.inactive_list,
list) {
if (i915_gem_object_is_purgeable(obj_priv)) {
- struct drm_gem_object *obj = obj_priv->obj;
- i915_gem_object_unbind(obj);
- i915_gem_object_truncate(obj);
-
+ 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;
}
&dev_priv->mm.inactive_list,
list) {
if (nr_to_scan > 0) {
- struct drm_gem_object *obj = obj_priv->obj;
- i915_gem_object_unbind(obj);
- if (i915_gem_object_is_purgeable(obj_priv))
- i915_gem_object_truncate(obj);
-
+ i915_gem_object_unbind(&obj_priv->base);
nr_to_scan--;
} else
cnt++;
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)