#endif
#include <xf86drm.h>
+#include <fcntl.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <assert.h>
+#include <pthread.h>
#include <sys/ioctl.h>
#include <sys/mman.h>
+#include <sys/stat.h>
+#include <sys/types.h>
#include "errno.h"
+#include "libdrm_lists.h"
#include "intel_bufmgr.h"
#include "intel_bufmgr_priv.h"
+#include "intel_chipset.h"
#include "string.h"
#include "i915_drm.h"
fprintf(stderr, __VA_ARGS__); \
} while (0)
-typedef struct _dri_bo_gem dri_bo_gem;
+typedef struct _drm_intel_bo_gem drm_intel_bo_gem;
+
+struct drm_intel_gem_bo_bucket {
+ drmMMListHead head;
-struct dri_gem_bo_bucket {
- dri_bo_gem *head, **tail;
/**
* Limit on the number of entries in this bucket.
*
int num_entries;
};
-/* Arbitrarily chosen, 16 means that the maximum size we'll cache for reuse
- * is 1 << 16 pages, or 256MB.
+/* Only cache objects up to 64MB. Bigger than that, and the rounding of the
+ * size makes many operations fail that wouldn't otherwise.
*/
-#define INTEL_GEM_BO_BUCKETS 16
-typedef struct _dri_bufmgr_gem {
- dri_bufmgr bufmgr;
+#define DRM_INTEL_GEM_BO_BUCKETS 14
+typedef struct _drm_intel_bufmgr_gem {
+ drm_intel_bufmgr bufmgr;
int fd;
int max_relocs;
+ pthread_mutex_t lock;
+
struct drm_i915_gem_exec_object *exec_objects;
- dri_bo **exec_bos;
+ drm_intel_bo **exec_bos;
int exec_size;
int exec_count;
/** Array of lists of cached gem objects of power-of-two sizes */
- struct dri_gem_bo_bucket cache_bucket[INTEL_GEM_BO_BUCKETS];
-} dri_bufmgr_gem;
+ struct drm_intel_gem_bo_bucket cache_bucket[DRM_INTEL_GEM_BO_BUCKETS];
-struct _dri_bo_gem {
- dri_bo bo;
+ uint64_t gtt_size;
+ int available_fences;
+ int pci_device;
+} drm_intel_bufmgr_gem;
+
+struct _drm_intel_bo_gem {
+ drm_intel_bo bo;
int refcount;
/** Boolean whether the mmap ioctl has been called for this buffer yet. */
- int mapped;
uint32_t gem_handle;
const char *name;
*/
int swrast;
+ /**
+ * Current tiling mode
+ */
+ uint32_t tiling_mode;
+ uint32_t swizzle_mode;
+
/** Array passed to the DRM containing relocation information. */
struct drm_i915_gem_relocation_entry *relocs;
/** Array of bos corresponding to relocs[i].target_handle */
- dri_bo **reloc_target_bo;
+ drm_intel_bo **reloc_target_bo;
/** Number of entries in relocs */
int reloc_count;
- /** Mapped address for the buffer */
- void *virtual;
+ /** Mapped address for the buffer, saved across map/unmap cycles */
+ void *mem_virtual;
+ /** GTT virtual address for the buffer, saved across map/unmap cycles */
+ void *gtt_virtual;
- /** free list */
- dri_bo_gem *next;
+ /** BO cache list */
+ drmMMListHead head;
+
+ /**
+ * Boolean of whether this BO and its children have been included in
+ * the current drm_intel_bufmgr_check_aperture_space() total.
+ */
+ char included_in_check_aperture;
+
+ /**
+ * Boolean of whether this buffer has been used as a relocation
+ * target and had its size accounted for, and thus can't have any
+ * further relocations added to it.
+ */
+ char used_as_reloc_target;
+
+ /**
+ * Boolean of whether this buffer can be re-used
+ */
+ char reusable;
+
+ /**
+ * Size in bytes of this buffer and its relocation descendents.
+ *
+ * Used to avoid costly tree walking in drm_intel_bufmgr_check_aperture in
+ * the common case.
+ */
+ int reloc_tree_size;
+ /**
+ * Number of potential fence registers required by this buffer and its
+ * relocations.
+ */
+ int reloc_tree_fences;
};
+static void drm_intel_gem_bo_reference_locked(drm_intel_bo *bo);
+
+static unsigned int
+drm_intel_gem_estimate_batch_space(drm_intel_bo **bo_array, int count);
+
+static unsigned int
+drm_intel_gem_compute_batch_space(drm_intel_bo **bo_array, int count);
+
+static int
+drm_intel_gem_bo_get_tiling(drm_intel_bo *bo, uint32_t *tiling_mode,
+ uint32_t *swizzle_mode);
+
+static int
+drm_intel_gem_bo_set_tiling(drm_intel_bo *bo, uint32_t *tiling_mode,
+ uint32_t stride);
+
+static void
+drm_intel_gem_bo_unreference(drm_intel_bo *bo);
+
static int
logbase2(int n)
{
return log2;
}
-static struct dri_gem_bo_bucket *
-dri_gem_bo_bucket_for_size(dri_bufmgr_gem *bufmgr_gem, unsigned long size)
+static struct drm_intel_gem_bo_bucket *
+drm_intel_gem_bo_bucket_for_size(drm_intel_bufmgr_gem *bufmgr_gem,
+ unsigned long size)
{
int i;
/* We always allocate in units of pages */
i = ffs(size / 4096) - 1;
- if (i >= INTEL_GEM_BO_BUCKETS)
+ if (i >= DRM_INTEL_GEM_BO_BUCKETS)
return NULL;
return &bufmgr_gem->cache_bucket[i];
}
-static void dri_gem_dump_validation_list(dri_bufmgr_gem *bufmgr_gem)
+static void drm_intel_gem_dump_validation_list(drm_intel_bufmgr_gem *bufmgr_gem)
{
int i, j;
for (i = 0; i < bufmgr_gem->exec_count; i++) {
- dri_bo *bo = bufmgr_gem->exec_bos[i];
- dri_bo_gem *bo_gem = (dri_bo_gem *)bo;
+ drm_intel_bo *bo = bufmgr_gem->exec_bos[i];
+ drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *)bo;
if (bo_gem->relocs == NULL) {
DBG("%2d: %d (%s)\n", i, bo_gem->gem_handle, bo_gem->name);
}
for (j = 0; j < bo_gem->reloc_count; j++) {
- dri_bo *target_bo = bo_gem->reloc_target_bo[j];
- dri_bo_gem *target_gem = (dri_bo_gem *)target_bo;
+ drm_intel_bo *target_bo = bo_gem->reloc_target_bo[j];
+ drm_intel_bo_gem *target_gem = (drm_intel_bo_gem *)target_bo;
DBG("%2d: %d (%s)@0x%08llx -> %d (%s)@0x%08lx + 0x%08x\n",
i,
- bo_gem->gem_handle, bo_gem->name, bo_gem->relocs[j].offset,
+ bo_gem->gem_handle, bo_gem->name,
+ (unsigned long long)bo_gem->relocs[j].offset,
target_gem->gem_handle, target_gem->name, target_bo->offset,
bo_gem->relocs[j].delta);
}
* access flags.
*/
static void
-intel_add_validate_buffer(dri_bo *bo)
+drm_intel_add_validate_buffer(drm_intel_bo *bo)
{
- dri_bufmgr_gem *bufmgr_gem = (dri_bufmgr_gem *)bo->bufmgr;
- dri_bo_gem *bo_gem = (dri_bo_gem *)bo;
+ drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *)bo->bufmgr;
+ drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *)bo;
int index;
if (bo_gem->validate_index != -1)
bufmgr_gem->exec_objects[index].alignment = 0;
bufmgr_gem->exec_objects[index].offset = 0;
bufmgr_gem->exec_bos[index] = bo;
- dri_bo_reference(bo);
+ drm_intel_gem_bo_reference_locked(bo);
bufmgr_gem->exec_count++;
}
sizeof(uint32_t))
static int
-intel_setup_reloc_list(dri_bo *bo)
+drm_intel_setup_reloc_list(drm_intel_bo *bo)
{
- dri_bo_gem *bo_gem = (dri_bo_gem *)bo;
- dri_bufmgr_gem *bufmgr_gem = (dri_bufmgr_gem *)bo->bufmgr;
+ drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *)bo;
+ drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *)bo->bufmgr;
bo_gem->relocs = malloc(bufmgr_gem->max_relocs *
sizeof(struct drm_i915_gem_relocation_entry));
- bo_gem->reloc_target_bo = malloc(bufmgr_gem->max_relocs * sizeof(dri_bo *));
+ bo_gem->reloc_target_bo = malloc(bufmgr_gem->max_relocs *
+ sizeof(drm_intel_bo *));
return 0;
}
-static dri_bo *
-dri_gem_bo_alloc(dri_bufmgr *bufmgr, const char *name,
- unsigned long size, unsigned int alignment)
+static drm_intel_bo *
+drm_intel_gem_bo_alloc_internal(drm_intel_bufmgr *bufmgr, const char *name,
+ unsigned long size, unsigned int alignment,
+ int for_render)
{
- dri_bufmgr_gem *bufmgr_gem = (dri_bufmgr_gem *)bufmgr;
- dri_bo_gem *bo_gem;
+ drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *)bufmgr;
+ drm_intel_bo_gem *bo_gem;
unsigned int page_size = getpagesize();
int ret;
- struct dri_gem_bo_bucket *bucket;
+ struct drm_intel_gem_bo_bucket *bucket;
int alloc_from_cache = 0;
unsigned long bo_size;
bo_size = 1 << logbase2(size);
if (bo_size < page_size)
bo_size = page_size;
- bucket = dri_gem_bo_bucket_for_size(bufmgr_gem, bo_size);
+ bucket = drm_intel_gem_bo_bucket_for_size(bufmgr_gem, bo_size);
/* If we don't have caching at this size, don't actually round the
* allocation up.
bo_size = page_size;
}
+ pthread_mutex_lock(&bufmgr_gem->lock);
/* Get a buffer out of the cache if available */
if (bucket != NULL && bucket->num_entries > 0) {
struct drm_i915_gem_busy busy;
-
- bo_gem = bucket->head;
- busy.handle = bo_gem->gem_handle;
-
- ret = ioctl(bufmgr_gem->fd, DRM_IOCTL_I915_GEM_BUSY, &busy);
- alloc_from_cache = (ret == 0 && busy.busy == 0);
- if (alloc_from_cache) {
- bucket->head = bo_gem->next;
- if (bo_gem->next == NULL)
- bucket->tail = &bucket->head;
+ if (for_render) {
+ /* Allocate new render-target BOs from the tail (MRU)
+ * of the list, as it will likely be hot in the GPU cache
+ * and in the aperture for us.
+ */
+ bo_gem = DRMLISTENTRY(drm_intel_bo_gem, bucket->head.prev, head);
+ DRMLISTDEL(&bo_gem->head);
bucket->num_entries--;
+ alloc_from_cache = 1;
+ } else {
+ /* For non-render-target BOs (where we're probably going to map it
+ * first thing in order to fill it with data), check if the
+ * last BO in the cache is unbusy, and only reuse in that case.
+ * Otherwise, allocating a new buffer is probably faster than
+ * waiting for the GPU to finish.
+ */
+ bo_gem = DRMLISTENTRY(drm_intel_bo_gem, bucket->head.next, head);
+
+ memset(&busy, 0, sizeof(busy));
+ busy.handle = bo_gem->gem_handle;
+
+ ret = ioctl(bufmgr_gem->fd, DRM_IOCTL_I915_GEM_BUSY, &busy);
+ alloc_from_cache = (ret == 0 && busy.busy == 0);
+
+ if (alloc_from_cache) {
+ DRMLISTDEL(&bo_gem->head);
+ bucket->num_entries--;
+ }
}
}
+ pthread_mutex_unlock(&bufmgr_gem->lock);
if (!alloc_from_cache) {
struct drm_i915_gem_create create;
ret = ioctl(bufmgr_gem->fd, DRM_IOCTL_I915_GEM_CREATE, &create);
bo_gem->gem_handle = create.handle;
- //bo_gem->bo.handle = bo_gem->gem_handle;
+ bo_gem->bo.handle = bo_gem->gem_handle;
if (ret != 0) {
free(bo_gem);
return NULL;
bo_gem->name = name;
bo_gem->refcount = 1;
bo_gem->validate_index = -1;
+ bo_gem->reloc_tree_size = bo_gem->bo.size;
+ bo_gem->reloc_tree_fences = 0;
+ bo_gem->used_as_reloc_target = 0;
+ bo_gem->tiling_mode = I915_TILING_NONE;
+ bo_gem->swizzle_mode = I915_BIT_6_SWIZZLE_NONE;
+ bo_gem->reusable = 1;
DBG("bo_create: buf %d (%s) %ldb\n",
bo_gem->gem_handle, bo_gem->name, size);
return &bo_gem->bo;
}
+static drm_intel_bo *
+drm_intel_gem_bo_alloc_for_render(drm_intel_bufmgr *bufmgr, const char *name,
+ unsigned long size, unsigned int alignment)
+{
+ return drm_intel_gem_bo_alloc_internal(bufmgr, name, size, alignment, 1);
+}
+
+static drm_intel_bo *
+drm_intel_gem_bo_alloc(drm_intel_bufmgr *bufmgr, const char *name,
+ unsigned long size, unsigned int alignment)
+{
+ return drm_intel_gem_bo_alloc_internal(bufmgr, name, size, alignment, 0);
+}
+
/**
- * Returns a dri_bo wrapping the given buffer object handle.
+ * Returns a drm_intel_bo wrapping the given buffer object handle.
*
* This can be used when one application needs to pass a buffer object
* to another.
*/
-dri_bo *
-intel_bo_gem_create_from_name(dri_bufmgr *bufmgr, const char *name,
- unsigned int handle)
+drm_intel_bo *
+drm_intel_bo_gem_create_from_name(drm_intel_bufmgr *bufmgr, const char *name,
+ unsigned int handle)
{
- dri_bufmgr_gem *bufmgr_gem = (dri_bufmgr_gem *)bufmgr;
- dri_bo_gem *bo_gem;
+ drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *)bufmgr;
+ drm_intel_bo_gem *bo_gem;
int ret;
struct drm_gem_open open_arg;
+ struct drm_i915_gem_get_tiling get_tiling;
bo_gem = calloc(1, sizeof(*bo_gem));
if (!bo_gem)
bo_gem->validate_index = -1;
bo_gem->gem_handle = open_arg.handle;
bo_gem->global_name = handle;
+ bo_gem->reusable = 0;
+
+ memset(&get_tiling, 0, sizeof(get_tiling));
+ get_tiling.handle = bo_gem->gem_handle;
+ ret = ioctl(bufmgr_gem->fd, DRM_IOCTL_I915_GEM_GET_TILING, &get_tiling);
+ if (ret != 0) {
+ drm_intel_gem_bo_unreference(&bo_gem->bo);
+ return NULL;
+ }
+ bo_gem->tiling_mode = get_tiling.tiling_mode;
+ bo_gem->swizzle_mode = get_tiling.swizzle_mode;
+ if (bo_gem->tiling_mode == I915_TILING_NONE)
+ bo_gem->reloc_tree_fences = 0;
+ else
+ bo_gem->reloc_tree_fences = 1;
DBG("bo_create_from_handle: %d (%s)\n", handle, bo_gem->name);
}
static void
-dri_gem_bo_reference(dri_bo *bo)
+drm_intel_gem_bo_reference(drm_intel_bo *bo)
{
- dri_bo_gem *bo_gem = (dri_bo_gem *)bo;
+ drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *)bo->bufmgr;
+ drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *)bo;
+ assert(bo_gem->refcount > 0);
+ pthread_mutex_lock(&bufmgr_gem->lock);
bo_gem->refcount++;
+ pthread_mutex_unlock(&bufmgr_gem->lock);
}
static void
-dri_gem_bo_free(dri_bo *bo)
+drm_intel_gem_bo_reference_locked(drm_intel_bo *bo)
{
- dri_bufmgr_gem *bufmgr_gem = (dri_bufmgr_gem *)bo->bufmgr;
- dri_bo_gem *bo_gem = (dri_bo_gem *)bo;
+ drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *)bo;
+
+ assert(bo_gem->refcount > 0);
+ bo_gem->refcount++;
+}
+
+static void
+drm_intel_gem_bo_free(drm_intel_bo *bo)
+{
+ drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *)bo->bufmgr;
+ drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *)bo;
struct drm_gem_close close;
int ret;
- if (bo_gem->mapped)
- munmap (bo_gem->virtual, bo_gem->bo.size);
+ if (bo_gem->mem_virtual)
+ munmap (bo_gem->mem_virtual, bo_gem->bo.size);
+ if (bo_gem->gtt_virtual)
+ munmap (bo_gem->gtt_virtual, bo_gem->bo.size);
/* Close this object */
+ memset(&close, 0, sizeof(close));
close.handle = bo_gem->gem_handle;
ret = ioctl(bufmgr_gem->fd, DRM_IOCTL_GEM_CLOSE, &close);
if (ret != 0) {
}
static void
-dri_gem_bo_unreference(dri_bo *bo)
+drm_intel_gem_bo_unreference_locked(drm_intel_bo *bo)
{
- dri_bufmgr_gem *bufmgr_gem = (dri_bufmgr_gem *)bo->bufmgr;
- dri_bo_gem *bo_gem = (dri_bo_gem *)bo;
-
- if (!bo)
- return;
+ drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *)bo->bufmgr;
+ drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *)bo;
+ assert(bo_gem->refcount > 0);
if (--bo_gem->refcount == 0) {
- struct dri_gem_bo_bucket *bucket;
+ struct drm_intel_gem_bo_bucket *bucket;
+ uint32_t tiling_mode;
if (bo_gem->relocs != NULL) {
int i;
/* Unreference all the target buffers */
for (i = 0; i < bo_gem->reloc_count; i++)
- dri_bo_unreference(bo_gem->reloc_target_bo[i]);
+ drm_intel_gem_bo_unreference_locked(bo_gem->reloc_target_bo[i]);
free(bo_gem->reloc_target_bo);
free(bo_gem->relocs);
}
DBG("bo_unreference final: %d (%s)\n",
bo_gem->gem_handle, bo_gem->name);
- bucket = dri_gem_bo_bucket_for_size(bufmgr_gem, bo->size);
+ bucket = drm_intel_gem_bo_bucket_for_size(bufmgr_gem, bo->size);
/* Put the buffer into our internal cache for reuse if we can. */
- if (bucket != NULL &&
+ tiling_mode = I915_TILING_NONE;
+ if (bo_gem->reusable &&
+ bucket != NULL &&
(bucket->max_entries == -1 ||
(bucket->max_entries > 0 &&
- bucket->num_entries < bucket->max_entries)))
+ bucket->num_entries < bucket->max_entries)) &&
+ drm_intel_gem_bo_set_tiling(bo, &tiling_mode, 0) == 0)
{
- bo_gem->name = 0;
+ bo_gem->name = NULL;
bo_gem->validate_index = -1;
bo_gem->relocs = NULL;
bo_gem->reloc_target_bo = NULL;
bo_gem->reloc_count = 0;
- bo_gem->next = NULL;
- *bucket->tail = bo_gem;
- bucket->tail = &bo_gem->next;
+ DRMLISTADDTAIL(&bo_gem->head, &bucket->head);
bucket->num_entries++;
} else {
- dri_gem_bo_free(bo);
+ drm_intel_gem_bo_free(bo);
}
-
- return;
}
}
+static void
+drm_intel_gem_bo_unreference(drm_intel_bo *bo)
+{
+ drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *)bo->bufmgr;
+
+ pthread_mutex_lock(&bufmgr_gem->lock);
+ drm_intel_gem_bo_unreference_locked(bo);
+ pthread_mutex_unlock(&bufmgr_gem->lock);
+}
+
static int
-dri_gem_bo_map(dri_bo *bo, int write_enable)
+drm_intel_gem_bo_map(drm_intel_bo *bo, int write_enable)
{
- dri_bufmgr_gem *bufmgr_gem;
- dri_bo_gem *bo_gem = (dri_bo_gem *)bo;
+ drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *)bo->bufmgr;
+ drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *)bo;
struct drm_i915_gem_set_domain set_domain;
int ret;
- bufmgr_gem = (dri_bufmgr_gem *)bo->bufmgr;
+ pthread_mutex_lock(&bufmgr_gem->lock);
/* Allow recursive mapping. Mesa may recursively map buffers with
* nested display loops.
*/
- if (!bo_gem->mapped) {
-
- assert(bo->virtual == NULL);
-
+ if (!bo_gem->mem_virtual) {
+ struct drm_i915_gem_mmap mmap_arg;
+
DBG("bo_map: %d (%s)\n", bo_gem->gem_handle, bo_gem->name);
-
- if (bo_gem->virtual == NULL) {
- struct drm_i915_gem_mmap mmap_arg;
-
- memset(&mmap_arg, 0, sizeof(mmap_arg));
- mmap_arg.handle = bo_gem->gem_handle;
- mmap_arg.offset = 0;
- mmap_arg.size = bo->size;
- ret = ioctl(bufmgr_gem->fd, DRM_IOCTL_I915_GEM_MMAP, &mmap_arg);
- if (ret != 0) {
- fprintf(stderr, "%s:%d: Error mapping buffer %d (%s): %s .\n",
- __FILE__, __LINE__,
- bo_gem->gem_handle, bo_gem->name, strerror(errno));
- }
- bo_gem->virtual = (void *)(uintptr_t)mmap_arg.addr_ptr;
+
+ memset(&mmap_arg, 0, sizeof(mmap_arg));
+ mmap_arg.handle = bo_gem->gem_handle;
+ mmap_arg.offset = 0;
+ mmap_arg.size = bo->size;
+ ret = ioctl(bufmgr_gem->fd, DRM_IOCTL_I915_GEM_MMAP, &mmap_arg);
+ if (ret != 0) {
+ fprintf(stderr, "%s:%d: Error mapping buffer %d (%s): %s .\n",
+ __FILE__, __LINE__,
+ bo_gem->gem_handle, bo_gem->name, strerror(errno));
+ pthread_mutex_unlock(&bufmgr_gem->lock);
+ return ret;
}
- bo->virtual = bo_gem->virtual;
+ bo_gem->mem_virtual = (void *)(uintptr_t)mmap_arg.addr_ptr;
bo_gem->swrast = 0;
- bo_gem->mapped = 1;
- DBG("bo_map: %d (%s) -> %p\n", bo_gem->gem_handle, bo_gem->name, bo_gem->virtual);
}
+ DBG("bo_map: %d (%s) -> %p\n", bo_gem->gem_handle, bo_gem->name,
+ bo_gem->mem_virtual);
+ bo->virtual = bo_gem->mem_virtual;
- if (!bo_gem->swrast) {
+ if (bo_gem->global_name != 0 || !bo_gem->swrast) {
set_domain.handle = bo_gem->gem_handle;
set_domain.read_domains = I915_GEM_DOMAIN_CPU;
if (write_enable)
if (ret != 0) {
fprintf (stderr, "%s:%d: Error setting swrast %d: %s\n",
__FILE__, __LINE__, bo_gem->gem_handle, strerror (errno));
+ pthread_mutex_unlock(&bufmgr_gem->lock);
+ return ret;
}
bo_gem->swrast = 1;
}
+ pthread_mutex_unlock(&bufmgr_gem->lock);
+
return 0;
}
+int
+drm_intel_gem_bo_map_gtt(drm_intel_bo *bo)
+{
+ drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *)bo->bufmgr;
+ drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *)bo;
+ struct drm_i915_gem_set_domain set_domain;
+ int ret;
+
+ pthread_mutex_lock(&bufmgr_gem->lock);
+
+ /* Get a mapping of the buffer if we haven't before. */
+ if (bo_gem->gtt_virtual == NULL) {
+ struct drm_i915_gem_mmap_gtt mmap_arg;
+
+ DBG("bo_map_gtt: %d (%s)\n", bo_gem->gem_handle, bo_gem->name);
+
+ memset(&mmap_arg, 0, sizeof(mmap_arg));
+ mmap_arg.handle = bo_gem->gem_handle;
+
+ /* Get the fake offset back... */
+ ret = ioctl(bufmgr_gem->fd, DRM_IOCTL_I915_GEM_MMAP_GTT, &mmap_arg);
+ if (ret != 0) {
+ fprintf(stderr,
+ "%s:%d: Error preparing buffer map %d (%s): %s .\n",
+ __FILE__, __LINE__,
+ bo_gem->gem_handle, bo_gem->name,
+ strerror(errno));
+ pthread_mutex_unlock(&bufmgr_gem->lock);
+ return ret;
+ }
+
+ /* and mmap it */
+ bo_gem->gtt_virtual = mmap(0, bo->size, PROT_READ | PROT_WRITE,
+ MAP_SHARED, bufmgr_gem->fd,
+ mmap_arg.offset);
+ if (bo_gem->gtt_virtual == MAP_FAILED) {
+ fprintf(stderr,
+ "%s:%d: Error mapping buffer %d (%s): %s .\n",
+ __FILE__, __LINE__,
+ bo_gem->gem_handle, bo_gem->name,
+ strerror(errno));
+ pthread_mutex_unlock(&bufmgr_gem->lock);
+ return errno;
+ }
+ }
+
+ bo->virtual = bo_gem->gtt_virtual;
+
+ DBG("bo_map: %d (%s) -> %p\n", bo_gem->gem_handle, bo_gem->name,
+ bo_gem->gtt_virtual);
+
+ /* Now move it to the GTT domain so that the CPU caches are flushed */
+ set_domain.handle = bo_gem->gem_handle;
+ set_domain.read_domains = I915_GEM_DOMAIN_GTT;
+ set_domain.write_domain = I915_GEM_DOMAIN_GTT;
+ do {
+ ret = ioctl(bufmgr_gem->fd, DRM_IOCTL_I915_GEM_SET_DOMAIN,
+ &set_domain);
+ } while (ret == -1 && errno == EINTR);
+
+ if (ret != 0) {
+ fprintf (stderr, "%s:%d: Error setting domain %d: %s\n",
+ __FILE__, __LINE__, bo_gem->gem_handle, strerror (errno));
+ }
+
+ pthread_mutex_unlock(&bufmgr_gem->lock);
+
+ return 0;
+}
+
+int
+drm_intel_gem_bo_unmap_gtt(drm_intel_bo *bo)
+{
+ drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *)bo->bufmgr;
+ drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *)bo;
+ struct drm_i915_gem_sw_finish sw_finish;
+ int ret = 0;
+
+ if (bo == NULL)
+ return 0;
+
+ assert(bo_gem->gtt_virtual != NULL);
+
+ pthread_mutex_lock(&bufmgr_gem->lock);
+ bo->virtual = NULL;
+ pthread_mutex_unlock(&bufmgr_gem->lock);
+
+ return ret;
+}
+
static int
-dri_gem_bo_unmap(dri_bo *bo)
+drm_intel_gem_bo_unmap(drm_intel_bo *bo)
{
- dri_bufmgr_gem *bufmgr_gem = (dri_bufmgr_gem *)bo->bufmgr;
- dri_bo_gem *bo_gem = (dri_bo_gem *)bo;
+ drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *)bo->bufmgr;
+ drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *)bo;
struct drm_i915_gem_sw_finish sw_finish;
int ret;
if (bo == NULL)
return 0;
- assert(bo_gem->mapped);
+ assert(bo_gem->mem_virtual != NULL);
+ pthread_mutex_lock(&bufmgr_gem->lock);
if (bo_gem->swrast) {
sw_finish.handle = bo_gem->gem_handle;
do {
} while (ret == -1 && errno == EINTR);
bo_gem->swrast = 0;
}
+ bo->virtual = NULL;
+ pthread_mutex_unlock(&bufmgr_gem->lock);
return 0;
}
static int
-dri_gem_bo_subdata (dri_bo *bo, unsigned long offset,
- unsigned long size, const void *data)
+drm_intel_gem_bo_subdata (drm_intel_bo *bo, unsigned long offset,
+ unsigned long size, const void *data)
{
- dri_bufmgr_gem *bufmgr_gem = (dri_bufmgr_gem *)bo->bufmgr;
- dri_bo_gem *bo_gem = (dri_bo_gem *)bo;
+ drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *)bo->bufmgr;
+ drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *)bo;
struct drm_i915_gem_pwrite pwrite;
int ret;
}
static int
-dri_gem_bo_get_subdata (dri_bo *bo, unsigned long offset,
- unsigned long size, void *data)
+drm_intel_gem_get_pipe_from_crtc_id (drm_intel_bufmgr *bufmgr, int crtc_id)
{
- dri_bufmgr_gem *bufmgr_gem = (dri_bufmgr_gem *)bo->bufmgr;
- dri_bo_gem *bo_gem = (dri_bo_gem *)bo;
+ drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *) bufmgr;
+ struct drm_i915_get_pipe_from_crtc_id get_pipe_from_crtc_id;
+ int ret;
+
+ get_pipe_from_crtc_id.crtc_id = crtc_id;
+ ret = ioctl (bufmgr_gem->fd, DRM_IOCTL_I915_GET_PIPE_FROM_CRTC_ID,
+ &get_pipe_from_crtc_id);
+ if (ret != 0) {
+ /* We return -1 here to signal that we don't
+ * know which pipe is associated with this crtc.
+ * This lets the caller know that this information
+ * isn't available; using the wrong pipe for
+ * vblank waiting can cause the chipset to lock up
+ */
+ return -1;
+ }
+
+ return get_pipe_from_crtc_id.pipe;
+}
+
+static int
+drm_intel_gem_bo_get_subdata (drm_intel_bo *bo, unsigned long offset,
+ unsigned long size, void *data)
+{
+ drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *)bo->bufmgr;
+ drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *)bo;
struct drm_i915_gem_pread pread;
int ret;
return 0;
}
+/** Waits for all GPU rendering to the object to have completed. */
static void
-dri_gem_bo_wait_rendering(dri_bo *bo)
+drm_intel_gem_bo_wait_rendering(drm_intel_bo *bo)
{
- dri_bufmgr_gem *bufmgr_gem = (dri_bufmgr_gem *)bo->bufmgr;
- dri_bo_gem *bo_gem = (dri_bo_gem *)bo;
+ return drm_intel_gem_bo_start_gtt_access(bo, 0);
+}
+
+/**
+ * Sets the object to the GTT read and possibly write domain, used by the X
+ * 2D driver in the absence of kernel support to do drm_intel_gem_bo_map_gtt().
+ *
+ * In combination with drm_intel_gem_bo_pin() and manual fence management, we
+ * can do tiled pixmaps this way.
+ */
+void
+drm_intel_gem_bo_start_gtt_access(drm_intel_bo *bo, int write_enable)
+{
+ drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *)bo->bufmgr;
+ drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *)bo;
struct drm_i915_gem_set_domain set_domain;
int ret;
set_domain.handle = bo_gem->gem_handle;
set_domain.read_domains = I915_GEM_DOMAIN_GTT;
- set_domain.write_domain = 0;
- ret = ioctl (bufmgr_gem->fd, DRM_IOCTL_I915_GEM_SET_DOMAIN, &set_domain);
+ set_domain.write_domain = write_enable ? I915_GEM_DOMAIN_GTT : 0;
+ do {
+ ret = ioctl(bufmgr_gem->fd, DRM_IOCTL_I915_GEM_SET_DOMAIN, &set_domain);
+ } while (ret == -1 && errno == EINTR);
if (ret != 0) {
fprintf (stderr, "%s:%d: Error setting memory domains %d (%08x %08x): %s .\n",
__FILE__, __LINE__,
}
static void
-dri_bufmgr_gem_destroy(dri_bufmgr *bufmgr)
+drm_intel_bufmgr_gem_destroy(drm_intel_bufmgr *bufmgr)
{
- dri_bufmgr_gem *bufmgr_gem = (dri_bufmgr_gem *)bufmgr;
+ drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *)bufmgr;
int i;
free(bufmgr_gem->exec_objects);
free(bufmgr_gem->exec_bos);
+ pthread_mutex_destroy(&bufmgr_gem->lock);
+
/* Free any cached buffer objects we were going to reuse */
- for (i = 0; i < INTEL_GEM_BO_BUCKETS; i++) {
- struct dri_gem_bo_bucket *bucket = &bufmgr_gem->cache_bucket[i];
- dri_bo_gem *bo_gem;
-
- while ((bo_gem = bucket->head) != NULL) {
- bucket->head = bo_gem->next;
- if (bo_gem->next == NULL)
- bucket->tail = &bucket->head;
+ for (i = 0; i < DRM_INTEL_GEM_BO_BUCKETS; i++) {
+ struct drm_intel_gem_bo_bucket *bucket = &bufmgr_gem->cache_bucket[i];
+ drm_intel_bo_gem *bo_gem;
+
+ while (!DRMLISTEMPTY(&bucket->head)) {
+ bo_gem = DRMLISTENTRY(drm_intel_bo_gem, bucket->head.next, head);
+ DRMLISTDEL(&bo_gem->head);
bucket->num_entries--;
- dri_gem_bo_free(&bo_gem->bo);
+ drm_intel_gem_bo_free(&bo_gem->bo);
}
}
* last known offset in target_bo.
*/
static int
-dri_gem_bo_emit_reloc(dri_bo *bo, uint32_t read_domains, uint32_t write_domain,
- uint32_t delta, uint32_t offset, dri_bo *target_bo)
+drm_intel_gem_bo_emit_reloc(drm_intel_bo *bo, uint32_t offset,
+ drm_intel_bo *target_bo, uint32_t target_offset,
+ uint32_t read_domains, uint32_t write_domain)
{
- dri_bufmgr_gem *bufmgr_gem = (dri_bufmgr_gem *)bo->bufmgr;
- dri_bo_gem *bo_gem = (dri_bo_gem *)bo;
- dri_bo_gem *target_bo_gem = (dri_bo_gem *)target_bo;
+ drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *)bo->bufmgr;
+ drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *)bo;
+ drm_intel_bo_gem *target_bo_gem = (drm_intel_bo_gem *)target_bo;
+
+ pthread_mutex_lock(&bufmgr_gem->lock);
/* Create a new relocation list if needed */
if (bo_gem->relocs == NULL)
- intel_setup_reloc_list(bo);
+ drm_intel_setup_reloc_list(bo);
/* Check overflow */
assert(bo_gem->reloc_count < bufmgr_gem->max_relocs);
assert (offset <= bo->size - 4);
assert ((write_domain & (write_domain-1)) == 0);
+ /* Make sure that we're not adding a reloc to something whose size has
+ * already been accounted for.
+ */
+ assert(!bo_gem->used_as_reloc_target);
+ bo_gem->reloc_tree_size += target_bo_gem->reloc_tree_size;
+ bo_gem->reloc_tree_fences += target_bo_gem->reloc_tree_fences;
+
+ /* Flag the target to disallow further relocations in it. */
+ target_bo_gem->used_as_reloc_target = 1;
+
bo_gem->relocs[bo_gem->reloc_count].offset = offset;
- bo_gem->relocs[bo_gem->reloc_count].delta = delta;
+ bo_gem->relocs[bo_gem->reloc_count].delta = target_offset;
bo_gem->relocs[bo_gem->reloc_count].target_handle =
target_bo_gem->gem_handle;
bo_gem->relocs[bo_gem->reloc_count].read_domains = read_domains;
bo_gem->relocs[bo_gem->reloc_count].presumed_offset = target_bo->offset;
bo_gem->reloc_target_bo[bo_gem->reloc_count] = target_bo;
- dri_bo_reference(target_bo);
+ drm_intel_gem_bo_reference_locked(target_bo);
bo_gem->reloc_count++;
+
+ pthread_mutex_unlock(&bufmgr_gem->lock);
+
return 0;
}
* index values into the validation list.
*/
static void
-dri_gem_bo_process_reloc(dri_bo *bo)
+drm_intel_gem_bo_process_reloc(drm_intel_bo *bo)
{
- dri_bo_gem *bo_gem = (dri_bo_gem *)bo;
+ drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *)bo;
int i;
if (bo_gem->relocs == NULL)
return;
for (i = 0; i < bo_gem->reloc_count; i++) {
- dri_bo *target_bo = bo_gem->reloc_target_bo[i];
+ drm_intel_bo *target_bo = bo_gem->reloc_target_bo[i];
/* Continue walking the tree depth-first. */
- dri_gem_bo_process_reloc(target_bo);
+ drm_intel_gem_bo_process_reloc(target_bo);
/* Add the target to the validate list */
- intel_add_validate_buffer(target_bo);
+ drm_intel_add_validate_buffer(target_bo);
}
}
static void
-intel_update_buffer_offsets (dri_bufmgr_gem *bufmgr_gem)
+drm_intel_update_buffer_offsets (drm_intel_bufmgr_gem *bufmgr_gem)
{
int i;
for (i = 0; i < bufmgr_gem->exec_count; i++) {
- dri_bo *bo = bufmgr_gem->exec_bos[i];
- dri_bo_gem *bo_gem = (dri_bo_gem *)bo;
+ drm_intel_bo *bo = bufmgr_gem->exec_bos[i];
+ drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *)bo;
/* Update the buffer offset */
if (bufmgr_gem->exec_objects[i].offset != bo->offset) {
DBG("BO %d (%s) migrated: 0x%08lx -> 0x%08llx\n",
bo_gem->gem_handle, bo_gem->name, bo->offset,
- bufmgr_gem->exec_objects[i].offset);
+ (unsigned long long)bufmgr_gem->exec_objects[i].offset);
bo->offset = bufmgr_gem->exec_objects[i].offset;
}
}
}
static int
-dri_gem_bo_exec(dri_bo *bo, int used,
- drm_clip_rect_t *cliprects, int num_cliprects,
- int DR4)
+drm_intel_gem_bo_exec(drm_intel_bo *bo, int used,
+ drm_clip_rect_t *cliprects, int num_cliprects,
+ int DR4)
{
- dri_bufmgr_gem *bufmgr_gem = (dri_bufmgr_gem *)bo->bufmgr;
+ drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *)bo->bufmgr;
struct drm_i915_gem_execbuffer execbuf;
int ret, i;
+ pthread_mutex_lock(&bufmgr_gem->lock);
/* Update indices and set up the validate list. */
- dri_gem_bo_process_reloc(bo);
+ drm_intel_gem_bo_process_reloc(bo);
/* Add the batch buffer to the validation list. There are no relocations
* pointing to it.
*/
- intel_add_validate_buffer(bo);
+ drm_intel_add_validate_buffer(bo);
execbuf.buffers_ptr = (uintptr_t)bufmgr_gem->exec_objects;
execbuf.buffer_count = bufmgr_gem->exec_count;
do {
ret = ioctl(bufmgr_gem->fd, DRM_IOCTL_I915_GEM_EXECBUFFER, &execbuf);
- } while (ret == -EAGAIN);
-
- intel_update_buffer_offsets (bufmgr_gem);
+ } while (ret != 0 && errno == EAGAIN);
+
+ if (ret != 0 && errno == ENOMEM) {
+ fprintf(stderr, "Execbuffer fails to pin. Estimate: %u. Actual: %u. Available: %u\n",
+ drm_intel_gem_estimate_batch_space(bufmgr_gem->exec_bos,
+ bufmgr_gem->exec_count),
+ drm_intel_gem_compute_batch_space(bufmgr_gem->exec_bos,
+ bufmgr_gem->exec_count),
+ (unsigned int) bufmgr_gem->gtt_size);
+ }
+ drm_intel_update_buffer_offsets (bufmgr_gem);
if (bufmgr_gem->bufmgr.debug)
- dri_gem_dump_validation_list(bufmgr_gem);
+ drm_intel_gem_dump_validation_list(bufmgr_gem);
for (i = 0; i < bufmgr_gem->exec_count; i++) {
- dri_bo *bo = bufmgr_gem->exec_bos[i];
- dri_bo_gem *bo_gem = (dri_bo_gem *)bo;
+ drm_intel_bo *bo = bufmgr_gem->exec_bos[i];
+ drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *)bo;
/* Need to call swrast on next bo_map */
bo_gem->swrast = 0;
/* Disconnect the buffer from the validate list */
bo_gem->validate_index = -1;
- dri_bo_unreference(bo);
+ drm_intel_gem_bo_unreference_locked(bo);
bufmgr_gem->exec_bos[i] = NULL;
}
bufmgr_gem->exec_count = 0;
+ pthread_mutex_unlock(&bufmgr_gem->lock);
return 0;
}
static int
-dri_gem_bo_pin(dri_bo *bo, uint32_t alignment)
+drm_intel_gem_bo_pin(drm_intel_bo *bo, uint32_t alignment)
{
- dri_bufmgr_gem *bufmgr_gem = (dri_bufmgr_gem *)bo->bufmgr;
- dri_bo_gem *bo_gem = (dri_bo_gem *)bo;
+ drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *)bo->bufmgr;
+ drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *)bo;
struct drm_i915_gem_pin pin;
int ret;
+ memset(&pin, 0, sizeof(pin));
pin.handle = bo_gem->gem_handle;
pin.alignment = alignment;
- ret = ioctl(bufmgr_gem->fd, DRM_IOCTL_I915_GEM_PIN, &pin);
+ do {
+ ret = ioctl(bufmgr_gem->fd, DRM_IOCTL_I915_GEM_PIN, &pin);
+ } while (ret == -1 && errno == EINTR);
+
if (ret != 0)
return -errno;
}
static int
-dri_gem_bo_unpin(dri_bo *bo)
+drm_intel_gem_bo_unpin(drm_intel_bo *bo)
{
- dri_bufmgr_gem *bufmgr_gem = (dri_bufmgr_gem *)bo->bufmgr;
- dri_bo_gem *bo_gem = (dri_bo_gem *)bo;
+ drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *)bo->bufmgr;
+ drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *)bo;
struct drm_i915_gem_unpin unpin;
int ret;
+ memset(&unpin, 0, sizeof(unpin));
unpin.handle = bo_gem->gem_handle;
ret = ioctl(bufmgr_gem->fd, DRM_IOCTL_I915_GEM_UNPIN, &unpin);
}
static int
-dri_gem_bo_set_tiling(dri_bo *bo, uint32_t *tiling_mode)
+drm_intel_gem_bo_set_tiling(drm_intel_bo *bo, uint32_t *tiling_mode,
+ uint32_t stride)
{
- dri_bufmgr_gem *bufmgr_gem = (dri_bufmgr_gem *)bo->bufmgr;
- dri_bo_gem *bo_gem = (dri_bo_gem *)bo;
+ drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *)bo->bufmgr;
+ drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *)bo;
struct drm_i915_gem_set_tiling set_tiling;
int ret;
+ if (bo_gem->global_name == 0 && *tiling_mode == bo_gem->tiling_mode)
+ return 0;
+
+ /* If we're going from non-tiling to tiling, bump fence count */
+ if (bo_gem->tiling_mode == I915_TILING_NONE)
+ bo_gem->reloc_tree_fences++;
+
+ memset(&set_tiling, 0, sizeof(set_tiling));
set_tiling.handle = bo_gem->gem_handle;
set_tiling.tiling_mode = *tiling_mode;
+ set_tiling.stride = stride;
ret = ioctl(bufmgr_gem->fd, DRM_IOCTL_I915_GEM_SET_TILING, &set_tiling);
if (ret != 0) {
- *tiling_mode = I915_TILING_NONE;
+ *tiling_mode = bo_gem->tiling_mode;
return -errno;
}
+ bo_gem->tiling_mode = set_tiling.tiling_mode;
+ bo_gem->swizzle_mode = set_tiling.swizzle_mode;
+
+ /* If we're going from tiling to non-tiling, drop fence count */
+ if (bo_gem->tiling_mode == I915_TILING_NONE)
+ bo_gem->reloc_tree_fences--;
- *tiling_mode = set_tiling.tiling_mode;
+ *tiling_mode = bo_gem->tiling_mode;
return 0;
}
static int
-dri_gem_bo_flink(dri_bo *bo, uint32_t *name)
+drm_intel_gem_bo_get_tiling(drm_intel_bo *bo, uint32_t *tiling_mode,
+ uint32_t *swizzle_mode)
{
- dri_bufmgr_gem *bufmgr_gem = (dri_bufmgr_gem *)bo->bufmgr;
- dri_bo_gem *bo_gem = (dri_bo_gem *)bo;
+ drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *)bo;
+
+ *tiling_mode = bo_gem->tiling_mode;
+ *swizzle_mode = bo_gem->swizzle_mode;
+ return 0;
+}
+
+static int
+drm_intel_gem_bo_flink(drm_intel_bo *bo, uint32_t *name)
+{
+ drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *)bo->bufmgr;
+ drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *)bo;
struct drm_gem_flink flink;
int ret;
if (!bo_gem->global_name) {
+ memset(&flink, 0, sizeof(flink));
flink.handle = bo_gem->gem_handle;
ret = ioctl(bufmgr_gem->fd, DRM_IOCTL_GEM_FLINK, &flink);
if (ret != 0)
return -errno;
bo_gem->global_name = flink.name;
+ bo_gem->reusable = 0;
}
*name = bo_gem->global_name;
* in flight at once.
*/
void
-intel_bufmgr_gem_enable_reuse(dri_bufmgr *bufmgr)
+drm_intel_bufmgr_gem_enable_reuse(drm_intel_bufmgr *bufmgr)
{
- dri_bufmgr_gem *bufmgr_gem = (dri_bufmgr_gem *)bufmgr;
+ drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *)bufmgr;
int i;
- for (i = 0; i < INTEL_GEM_BO_BUCKETS; i++) {
+ for (i = 0; i < DRM_INTEL_GEM_BO_BUCKETS; i++) {
bufmgr_gem->cache_bucket[i].max_entries = -1;
}
}
-/*
+/**
+ * Return the additional aperture space required by the tree of buffer objects
+ * rooted at bo.
+ */
+static int
+drm_intel_gem_bo_get_aperture_space(drm_intel_bo *bo)
+{
+ drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *)bo;
+ int i;
+ int total = 0;
+
+ if (bo == NULL || bo_gem->included_in_check_aperture)
+ return 0;
+
+ total += bo->size;
+ bo_gem->included_in_check_aperture = 1;
+
+ for (i = 0; i < bo_gem->reloc_count; i++)
+ total += drm_intel_gem_bo_get_aperture_space(bo_gem->reloc_target_bo[i]);
+
+ return total;
+}
+
+/**
+ * Count the number of buffers in this list that need a fence reg
+ *
+ * If the count is greater than the number of available regs, we'll have
+ * to ask the caller to resubmit a batch with fewer tiled buffers.
+ *
+ * This function over-counts if the same buffer is used multiple times.
+ */
+static unsigned int
+drm_intel_gem_total_fences(drm_intel_bo **bo_array, int count)
+{
+ int i;
+ unsigned int total = 0;
+
+ for (i = 0; i < count; i++) {
+ drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *)bo_array[i];
+
+ if (bo_gem == NULL)
+ continue;
+
+ total += bo_gem->reloc_tree_fences;
+ }
+ return total;
+}
+
+/**
+ * Clear the flag set by drm_intel_gem_bo_get_aperture_space() so we're ready
+ * for the next drm_intel_bufmgr_check_aperture_space() call.
+ */
+static void
+drm_intel_gem_bo_clear_aperture_space_flag(drm_intel_bo *bo)
+{
+ drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *)bo;
+ int i;
+
+ if (bo == NULL || !bo_gem->included_in_check_aperture)
+ return;
+
+ bo_gem->included_in_check_aperture = 0;
+
+ for (i = 0; i < bo_gem->reloc_count; i++)
+ drm_intel_gem_bo_clear_aperture_space_flag(bo_gem->reloc_target_bo[i]);
+}
+
+/**
+ * Return a conservative estimate for the amount of aperture required
+ * for a collection of buffers. This may double-count some buffers.
+ */
+static unsigned int
+drm_intel_gem_estimate_batch_space(drm_intel_bo **bo_array, int count)
+{
+ int i;
+ unsigned int total = 0;
+
+ for (i = 0; i < count; i++) {
+ drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *)bo_array[i];
+ if (bo_gem != NULL)
+ total += bo_gem->reloc_tree_size;
+ }
+ return total;
+}
+
+/**
+ * Return the amount of aperture needed for a collection of buffers.
+ * This avoids double counting any buffers, at the cost of looking
+ * at every buffer in the set.
+ */
+static unsigned int
+drm_intel_gem_compute_batch_space(drm_intel_bo **bo_array, int count)
+{
+ int i;
+ unsigned int total = 0;
+
+ for (i = 0; i < count; i++) {
+ total += drm_intel_gem_bo_get_aperture_space(bo_array[i]);
+ /* For the first buffer object in the array, we get an accurate count
+ * back for its reloc_tree size (since nothing had been flagged as
+ * being counted yet). We can save that value out as a more
+ * conservative reloc_tree_size that avoids double-counting target
+ * buffers. Since the first buffer happens to usually be the batch
+ * buffer in our callers, this can pull us back from doing the tree
+ * walk on every new batch emit.
+ */
+ if (i == 0) {
+ drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *)bo_array[i];
+ bo_gem->reloc_tree_size = total;
+ }
+ }
+
+ for (i = 0; i < count; i++)
+ drm_intel_gem_bo_clear_aperture_space_flag(bo_array[i]);
+ return total;
+}
+
+/**
+ * Return -1 if the batchbuffer should be flushed before attempting to
+ * emit rendering referencing the buffers pointed to by bo_array.
+ *
+ * This is required because if we try to emit a batchbuffer with relocations
+ * to a tree of buffers that won't simultaneously fit in the aperture,
+ * the rendering will return an error at a point where the software is not
+ * prepared to recover from it.
*
+ * However, we also want to emit the batchbuffer significantly before we reach
+ * the limit, as a series of batchbuffers each of which references buffers
+ * covering almost all of the aperture means that at each emit we end up
+ * waiting to evict a buffer from the last rendering, and we get synchronous
+ * performance. By emitting smaller batchbuffers, we eat some CPU overhead to
+ * get better parallelism.
+ */
+static int
+drm_intel_gem_check_aperture_space(drm_intel_bo **bo_array, int count)
+{
+ drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *)bo_array[0]->bufmgr;
+ unsigned int total = 0;
+ unsigned int threshold = bufmgr_gem->gtt_size * 3 / 4;
+ int total_fences;
+
+ /* Check for fence reg constraints if necessary */
+ if (bufmgr_gem->available_fences) {
+ total_fences = drm_intel_gem_total_fences(bo_array, count);
+ if (total_fences > bufmgr_gem->available_fences)
+ return -1;
+ }
+
+ total = drm_intel_gem_estimate_batch_space(bo_array, count);
+
+ if (total > threshold)
+ total = drm_intel_gem_compute_batch_space(bo_array, count);
+
+ if (total > threshold) {
+ DBG("check_space: overflowed available aperture, %dkb vs %dkb\n",
+ total / 1024, (int)bufmgr_gem->gtt_size / 1024);
+ return -1;
+ } else {
+ DBG("drm_check_space: total %dkb vs bufgr %dkb\n", total / 1024 ,
+ (int)bufmgr_gem->gtt_size / 1024);
+ return 0;
+ }
+}
+
+/*
+ * Disable buffer reuse for objects which are shared with the kernel
+ * as scanout buffers
*/
static int
-dri_gem_check_aperture_space(dri_bo **bo_array, int count)
+drm_intel_gem_bo_disable_reuse(drm_intel_bo *bo)
{
+ drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *)bo;
+
+ bo_gem->reusable = 0;
return 0;
}
*
* \param fd File descriptor of the opened DRM device.
*/
-dri_bufmgr *
-intel_bufmgr_gem_init(int fd, int batch_size)
+drm_intel_bufmgr *
+drm_intel_bufmgr_gem_init(int fd, int batch_size)
{
- dri_bufmgr_gem *bufmgr_gem;
- int i;
+ drm_intel_bufmgr_gem *bufmgr_gem;
+ struct drm_i915_gem_get_aperture aperture;
+ drm_i915_getparam_t gp;
+ int ret, i;
bufmgr_gem = calloc(1, sizeof(*bufmgr_gem));
bufmgr_gem->fd = fd;
+ if (pthread_mutex_init(&bufmgr_gem->lock, NULL) != 0) {
+ free(bufmgr_gem);
+ return NULL;
+ }
+
+ ret = ioctl(bufmgr_gem->fd, DRM_IOCTL_I915_GEM_GET_APERTURE, &aperture);
+
+ if (ret == 0)
+ bufmgr_gem->gtt_size = aperture.aper_available_size;
+ else {
+ fprintf(stderr, "DRM_IOCTL_I915_GEM_APERTURE failed: %s\n",
+ strerror(errno));
+ bufmgr_gem->gtt_size = 128 * 1024 * 1024;
+ fprintf(stderr, "Assuming %dkB available aperture size.\n"
+ "May lead to reduced performance or incorrect rendering.\n",
+ (int)bufmgr_gem->gtt_size / 1024);
+ }
+
+ gp.param = I915_PARAM_CHIPSET_ID;
+ gp.value = &bufmgr_gem->pci_device;
+ ret = ioctl(bufmgr_gem->fd, DRM_IOCTL_I915_GETPARAM, &gp);
+ if (ret) {
+ fprintf(stderr, "get chip id failed: %d\n", ret);
+ fprintf(stderr, "param: %d, val: %d\n", gp.param, *gp.value);
+ }
+
+ if (!IS_I965G(bufmgr_gem)) {
+ gp.param = I915_PARAM_NUM_FENCES_AVAIL;
+ gp.value = &bufmgr_gem->available_fences;
+ ret = ioctl(bufmgr_gem->fd, DRM_IOCTL_I915_GETPARAM, &gp);
+ if (ret) {
+ fprintf(stderr, "get fences failed: %d\n", ret);
+ fprintf(stderr, "param: %d, val: %d\n", gp.param, *gp.value);
+ bufmgr_gem->available_fences = 0;
+ }
+ }
+
/* Let's go with one relocation per every 2 dwords (but round down a bit
* since a power of two will mean an extra page allocation for the reloc
* buffer).
*/
bufmgr_gem->max_relocs = batch_size / sizeof(uint32_t) / 2 - 2;
- bufmgr_gem->bufmgr.bo_alloc = dri_gem_bo_alloc;
- bufmgr_gem->bufmgr.bo_reference = dri_gem_bo_reference;
- bufmgr_gem->bufmgr.bo_unreference = dri_gem_bo_unreference;
- bufmgr_gem->bufmgr.bo_map = dri_gem_bo_map;
- bufmgr_gem->bufmgr.bo_unmap = dri_gem_bo_unmap;
- bufmgr_gem->bufmgr.bo_subdata = dri_gem_bo_subdata;
- bufmgr_gem->bufmgr.bo_get_subdata = dri_gem_bo_get_subdata;
- bufmgr_gem->bufmgr.bo_wait_rendering = dri_gem_bo_wait_rendering;
- bufmgr_gem->bufmgr.bo_emit_reloc = dri_gem_bo_emit_reloc;
- bufmgr_gem->bufmgr.bo_pin = dri_gem_bo_pin;
- bufmgr_gem->bufmgr.bo_unpin = dri_gem_bo_unpin;
- bufmgr_gem->bufmgr.bo_set_tiling = dri_gem_bo_set_tiling;
- bufmgr_gem->bufmgr.bo_flink = dri_gem_bo_flink;
- bufmgr_gem->bufmgr.bo_exec = dri_gem_bo_exec;
- bufmgr_gem->bufmgr.destroy = dri_bufmgr_gem_destroy;
+ bufmgr_gem->bufmgr.bo_alloc = drm_intel_gem_bo_alloc;
+ bufmgr_gem->bufmgr.bo_alloc_for_render = drm_intel_gem_bo_alloc_for_render;
+ bufmgr_gem->bufmgr.bo_reference = drm_intel_gem_bo_reference;
+ bufmgr_gem->bufmgr.bo_unreference = drm_intel_gem_bo_unreference;
+ bufmgr_gem->bufmgr.bo_map = drm_intel_gem_bo_map;
+ bufmgr_gem->bufmgr.bo_unmap = drm_intel_gem_bo_unmap;
+ bufmgr_gem->bufmgr.bo_subdata = drm_intel_gem_bo_subdata;
+ bufmgr_gem->bufmgr.bo_get_subdata = drm_intel_gem_bo_get_subdata;
+ bufmgr_gem->bufmgr.bo_wait_rendering = drm_intel_gem_bo_wait_rendering;
+ bufmgr_gem->bufmgr.bo_emit_reloc = drm_intel_gem_bo_emit_reloc;
+ bufmgr_gem->bufmgr.bo_pin = drm_intel_gem_bo_pin;
+ bufmgr_gem->bufmgr.bo_unpin = drm_intel_gem_bo_unpin;
+ bufmgr_gem->bufmgr.bo_get_tiling = drm_intel_gem_bo_get_tiling;
+ bufmgr_gem->bufmgr.bo_set_tiling = drm_intel_gem_bo_set_tiling;
+ bufmgr_gem->bufmgr.bo_flink = drm_intel_gem_bo_flink;
+ bufmgr_gem->bufmgr.bo_exec = drm_intel_gem_bo_exec;
+ bufmgr_gem->bufmgr.destroy = drm_intel_bufmgr_gem_destroy;
bufmgr_gem->bufmgr.debug = 0;
- bufmgr_gem->bufmgr.check_aperture_space = dri_gem_check_aperture_space;
+ bufmgr_gem->bufmgr.check_aperture_space = drm_intel_gem_check_aperture_space;
+ bufmgr_gem->bufmgr.bo_disable_reuse = drm_intel_gem_bo_disable_reuse;
+ bufmgr_gem->bufmgr.get_pipe_from_crtc_id = drm_intel_gem_get_pipe_from_crtc_id;
/* Initialize the linked lists for BO reuse cache. */
- for (i = 0; i < INTEL_GEM_BO_BUCKETS; i++)
- bufmgr_gem->cache_bucket[i].tail = &bufmgr_gem->cache_bucket[i].head;
+ for (i = 0; i < DRM_INTEL_GEM_BO_BUCKETS; i++)
+ DRMINITLISTHEAD(&bufmgr_gem->cache_bucket[i].head);
return &bufmgr_gem->bufmgr;
}