* of the Software.
*/
-#define _FILE_OFFSET_BITS 64
#include "radeon_drm_cs.h"
#include "util/u_hash_table.h"
#include "util/u_memory.h"
-#include "util/u_simple_list.h"
-#include "util/u_double_list.h"
+#include "util/simple_list.h"
#include "os/os_thread.h"
#include "os/os_mman.h"
+#include "os/os_time.h"
#include "state_tracker/drm_driver.h"
#include <sys/ioctl.h>
#include <xf86drm.h>
#include <errno.h>
+#include <fcntl.h>
+#include <stdio.h>
-/*
- * this are copy from radeon_drm, once an updated libdrm is released
- * we should bump configure.ac requirement for it and remove the following
- * field
- */
-#define RADEON_BO_FLAGS_MACRO_TILE 1
-#define RADEON_BO_FLAGS_MICRO_TILE 2
-#define RADEON_BO_FLAGS_MICRO_TILE_SQUARE 0x20
-
-#ifndef DRM_RADEON_GEM_WAIT
-#define DRM_RADEON_GEM_WAIT 0x2b
-
-#define RADEON_GEM_NO_WAIT 0x1
-#define RADEON_GEM_USAGE_READ 0x2
-#define RADEON_GEM_USAGE_WRITE 0x4
-
-struct drm_radeon_gem_wait {
- uint32_t handle;
- uint32_t flags; /* one of RADEON_GEM_* */
-};
-
-#endif
-
-#ifndef RADEON_VA_MAP
-
-#define RADEON_VA_MAP 1
-#define RADEON_VA_UNMAP 2
-
-#define RADEON_VA_RESULT_OK 0
-#define RADEON_VA_RESULT_ERROR 1
-#define RADEON_VA_RESULT_VA_EXIST 2
-
-#define RADEON_VM_PAGE_VALID (1 << 0)
-#define RADEON_VM_PAGE_READABLE (1 << 1)
-#define RADEON_VM_PAGE_WRITEABLE (1 << 2)
-#define RADEON_VM_PAGE_SYSTEM (1 << 3)
-#define RADEON_VM_PAGE_SNOOPED (1 << 4)
-
-struct drm_radeon_gem_va {
- uint32_t handle;
- uint32_t operation;
- uint32_t vm_id;
- uint32_t flags;
- uint64_t offset;
-};
-
-#define DRM_RADEON_GEM_VA 0x2b
-#endif
-
-
-
-extern const struct pb_vtbl radeon_bo_vtbl;
-
-
-static INLINE struct radeon_bo *radeon_bo(struct pb_buffer *bo)
+static inline struct radeon_bo *radeon_bo(struct pb_buffer *bo)
{
- assert(bo->vtbl == &radeon_bo_vtbl);
return (struct radeon_bo *)bo;
}
uint64_t size;
};
-struct radeon_bomgr {
- /* Base class. */
- struct pb_manager base;
-
- /* Winsys. */
- struct radeon_drm_winsys *rws;
-
- /* List of buffer handles and its mutex. */
- struct util_hash_table *bo_handles;
- pipe_mutex bo_handles_mutex;
- pipe_mutex bo_va_mutex;
+static bool radeon_bo_is_busy(struct radeon_bo *bo)
+{
+ struct drm_radeon_gem_busy args = {0};
- /* is virtual address supported */
- bool va;
- unsigned va_offset;
- struct list_head va_holes;
-};
+ args.handle = bo->handle;
+ return drmCommandWriteRead(bo->rws->fd, DRM_RADEON_GEM_BUSY,
+ &args, sizeof(args)) != 0;
+}
-static INLINE struct radeon_bomgr *radeon_bomgr(struct pb_manager *mgr)
+static void radeon_bo_wait_idle(struct radeon_bo *bo)
{
- return (struct radeon_bomgr *)mgr;
+ struct drm_radeon_gem_wait_idle args = {0};
+
+ args.handle = bo->handle;
+ while (drmCommandWrite(bo->rws->fd, DRM_RADEON_GEM_WAIT_IDLE,
+ &args, sizeof(args)) == -EBUSY);
}
-static struct radeon_bo *get_radeon_bo(struct pb_buffer *_buf)
+static bool radeon_bo_wait(struct pb_buffer *_buf, uint64_t timeout,
+ enum radeon_bo_usage usage)
{
- struct radeon_bo *bo = NULL;
+ struct radeon_bo *bo = radeon_bo(_buf);
+ int64_t abs_timeout;
- if (_buf->vtbl == &radeon_bo_vtbl) {
- bo = radeon_bo(_buf);
- } else {
- struct pb_buffer *base_buf;
- pb_size offset;
- pb_get_base_buffer(_buf, &base_buf, &offset);
+ /* No timeout. Just query. */
+ if (timeout == 0)
+ return !bo->num_active_ioctls && !radeon_bo_is_busy(bo);
+
+ abs_timeout = os_time_get_absolute_timeout(timeout);
- if (base_buf->vtbl == &radeon_bo_vtbl)
- bo = radeon_bo(base_buf);
+ /* Wait if any ioctl is being submitted with this buffer. */
+ if (!os_wait_until_zero_abs_timeout(&bo->num_active_ioctls, abs_timeout))
+ return false;
+
+ /* Infinite timeout. */
+ if (abs_timeout == PIPE_TIMEOUT_INFINITE) {
+ radeon_bo_wait_idle(bo);
+ return true;
}
- return bo;
+ /* Other timeouts need to be emulated with a loop. */
+ while (radeon_bo_is_busy(bo)) {
+ if (os_time_get_nano() >= abs_timeout)
+ return false;
+ os_time_sleep(10);
+ }
+
+ return true;
}
-static void radeon_bo_wait(struct pb_buffer *_buf, enum radeon_bo_usage usage)
+static enum radeon_bo_domain get_valid_domain(enum radeon_bo_domain domain)
{
- struct radeon_bo *bo = get_radeon_bo(_buf);
+ /* Zero domains the driver doesn't understand. */
+ domain &= RADEON_DOMAIN_VRAM_GTT;
- while (p_atomic_read(&bo->num_active_ioctls)) {
- sched_yield();
- }
+ /* If no domain is set, we must set something... */
+ if (!domain)
+ domain = RADEON_DOMAIN_VRAM_GTT;
- /* XXX use this when it's ready */
- /*if (bo->rws->info.drm_minor >= 12) {
- struct drm_radeon_gem_wait args = {};
- args.handle = bo->handle;
- args.flags = usage;
- while (drmCommandWriteRead(bo->rws->fd, DRM_RADEON_GEM_WAIT,
- &args, sizeof(args)) == -EBUSY);
- } else*/ {
- struct drm_radeon_gem_wait_idle args;
- memset(&args, 0, sizeof(args));
- args.handle = bo->handle;
- while (drmCommandWriteRead(bo->rws->fd, DRM_RADEON_GEM_WAIT_IDLE,
- &args, sizeof(args)) == -EBUSY);
- }
+ return domain;
}
-static boolean radeon_bo_is_busy(struct pb_buffer *_buf,
- enum radeon_bo_usage usage)
+static enum radeon_bo_domain radeon_bo_get_initial_domain(
+ struct pb_buffer *buf)
{
- struct radeon_bo *bo = get_radeon_bo(_buf);
+ struct radeon_bo *bo = (struct radeon_bo*)buf;
+ struct drm_radeon_gem_op args;
- if (p_atomic_read(&bo->num_active_ioctls)) {
- return TRUE;
- }
+ if (bo->rws->info.drm_minor < 38)
+ return RADEON_DOMAIN_VRAM_GTT;
- /* XXX use this when it's ready */
- /*if (bo->rws->info.drm_minor >= 12) {
- struct drm_radeon_gem_wait args = {};
- args.handle = bo->handle;
- args.flags = usage | RADEON_GEM_NO_WAIT;
- return drmCommandWriteRead(bo->rws->fd, DRM_RADEON_GEM_WAIT,
- &args, sizeof(args)) != 0;
- } else*/ {
- struct drm_radeon_gem_busy args;
- memset(&args, 0, sizeof(args));
- args.handle = bo->handle;
- return drmCommandWriteRead(bo->rws->fd, DRM_RADEON_GEM_BUSY,
- &args, sizeof(args)) != 0;
- }
+ memset(&args, 0, sizeof(args));
+ args.handle = bo->handle;
+ args.op = RADEON_GEM_OP_GET_INITIAL_DOMAIN;
+
+ drmCommandWriteRead(bo->rws->fd, DRM_RADEON_GEM_OP,
+ &args, sizeof(args));
+
+ /* GEM domains and winsys domains are defined the same. */
+ return get_valid_domain(args.value);
}
-static uint64_t radeon_bomgr_find_va(struct radeon_bomgr *mgr, uint64_t size)
+static uint64_t radeon_bomgr_find_va(struct radeon_drm_winsys *rws,
+ uint64_t size, uint64_t alignment)
{
struct radeon_bo_va_hole *hole, *n;
- uint64_t offset = 0;
+ uint64_t offset = 0, waste = 0;
+
+ /* All VM address space holes will implicitly start aligned to the
+ * size alignment, so we don't need to sanitize the alignment here
+ */
+ size = align(size, rws->size_align);
- pipe_mutex_lock(mgr->bo_va_mutex);
+ pipe_mutex_lock(rws->bo_va_mutex);
/* first look for a hole */
- LIST_FOR_EACH_ENTRY_SAFE(hole, n, &mgr->va_holes, list) {
- if (hole->size == size) {
+ LIST_FOR_EACH_ENTRY_SAFE(hole, n, &rws->va_holes, list) {
+ offset = hole->offset;
+ waste = offset % alignment;
+ waste = waste ? alignment - waste : 0;
+ offset += waste;
+ if (offset >= (hole->offset + hole->size)) {
+ continue;
+ }
+ if (!waste && hole->size == size) {
offset = hole->offset;
list_del(&hole->list);
FREE(hole);
- pipe_mutex_unlock(mgr->bo_va_mutex);
+ pipe_mutex_unlock(rws->bo_va_mutex);
return offset;
}
- if (hole->size > size) {
- offset = hole->offset;
- hole->size -= size;
- hole->offset += size;
- pipe_mutex_unlock(mgr->bo_va_mutex);
+ if ((hole->size - waste) > size) {
+ if (waste) {
+ n = CALLOC_STRUCT(radeon_bo_va_hole);
+ n->size = waste;
+ n->offset = hole->offset;
+ list_add(&n->list, &hole->list);
+ }
+ hole->size -= (size + waste);
+ hole->offset += size + waste;
+ pipe_mutex_unlock(rws->bo_va_mutex);
+ return offset;
+ }
+ if ((hole->size - waste) == size) {
+ hole->size = waste;
+ pipe_mutex_unlock(rws->bo_va_mutex);
return offset;
}
}
- offset = mgr->va_offset;
- mgr->va_offset += size;
- pipe_mutex_unlock(mgr->bo_va_mutex);
+ offset = rws->va_offset;
+ waste = offset % alignment;
+ waste = waste ? alignment - waste : 0;
+ if (waste) {
+ n = CALLOC_STRUCT(radeon_bo_va_hole);
+ n->size = waste;
+ n->offset = offset;
+ list_add(&n->list, &rws->va_holes);
+ }
+ offset += waste;
+ rws->va_offset += size + waste;
+ pipe_mutex_unlock(rws->bo_va_mutex);
return offset;
}
-static void radeon_bomgr_force_va(struct radeon_bomgr *mgr, uint64_t va, uint64_t size)
+static void radeon_bomgr_free_va(struct radeon_drm_winsys *rws,
+ uint64_t va, uint64_t size)
{
- pipe_mutex_lock(mgr->bo_va_mutex);
- if (va >= mgr->va_offset) {
- if (va > mgr->va_offset) {
- struct radeon_bo_va_hole *hole;
- hole = CALLOC_STRUCT(radeon_bo_va_hole);
- if (hole) {
- hole->size = va - mgr->va_offset;
- hole->offset = mgr->va_offset;
- list_add(&hole->list, &mgr->va_holes);
+ struct radeon_bo_va_hole *hole;
+
+ size = align(size, rws->size_align);
+
+ pipe_mutex_lock(rws->bo_va_mutex);
+ if ((va + size) == rws->va_offset) {
+ rws->va_offset = va;
+ /* Delete uppermost hole if it reaches the new top */
+ if (!LIST_IS_EMPTY(&rws->va_holes)) {
+ hole = container_of(rws->va_holes.next, hole, list);
+ if ((hole->offset + hole->size) == va) {
+ rws->va_offset = hole->offset;
+ list_del(&hole->list);
+ FREE(hole);
}
}
- mgr->va_offset = va + size;
} else {
- struct radeon_bo_va_hole *hole, *n;
- uint64_t stmp, etmp;
+ struct radeon_bo_va_hole *next;
- /* free all holes that fall into the range
- * NOTE that we might lose virtual address space
- */
- LIST_FOR_EACH_ENTRY_SAFE(hole, n, &mgr->va_holes, list) {
- stmp = hole->offset;
- etmp = stmp + hole->size;
- if (va >= stmp && va < etmp) {
- list_del(&hole->list);
- FREE(hole);
+ hole = container_of(&rws->va_holes, hole, list);
+ LIST_FOR_EACH_ENTRY(next, &rws->va_holes, list) {
+ if (next->offset < va)
+ break;
+ hole = next;
+ }
+
+ if (&hole->list != &rws->va_holes) {
+ /* Grow upper hole if it's adjacent */
+ if (hole->offset == (va + size)) {
+ hole->offset = va;
+ hole->size += size;
+ /* Merge lower hole if it's adjacent */
+ if (next != hole && &next->list != &rws->va_holes &&
+ (next->offset + next->size) == va) {
+ next->size += hole->size;
+ list_del(&hole->list);
+ FREE(hole);
+ }
+ goto out;
}
}
- }
- pipe_mutex_unlock(mgr->bo_va_mutex);
-}
-static void radeon_bomgr_free_va(struct radeon_bomgr *mgr, uint64_t va, uint64_t size)
-{
- pipe_mutex_lock(mgr->bo_va_mutex);
- if ((va + size) == mgr->va_offset) {
- mgr->va_offset = va;
- } else {
- struct radeon_bo_va_hole *hole;
+ /* Grow lower hole if it's adjacent */
+ if (next != hole && &next->list != &rws->va_holes &&
+ (next->offset + next->size) == va) {
+ next->size += size;
+ goto out;
+ }
/* FIXME on allocation failure we just lose virtual address space
* maybe print a warning
*/
- hole = CALLOC_STRUCT(radeon_bo_va_hole);
- if (hole) {
- hole->size = size;
- hole->offset = va;
- list_add(&hole->list, &mgr->va_holes);
+ next = CALLOC_STRUCT(radeon_bo_va_hole);
+ if (next) {
+ next->size = size;
+ next->offset = va;
+ list_add(&next->list, &hole->list);
}
}
- pipe_mutex_unlock(mgr->bo_va_mutex);
+out:
+ pipe_mutex_unlock(rws->bo_va_mutex);
}
-static void radeon_bo_destroy(struct pb_buffer *_buf)
+void radeon_bo_destroy(struct pb_buffer *_buf)
{
struct radeon_bo *bo = radeon_bo(_buf);
- struct radeon_bomgr *mgr = bo->mgr;
+ struct radeon_drm_winsys *rws = bo->rws;
struct drm_gem_close args;
memset(&args, 0, sizeof(args));
- if (bo->name) {
- pipe_mutex_lock(bo->mgr->bo_handles_mutex);
- util_hash_table_remove(bo->mgr->bo_handles,
- (void*)(uintptr_t)bo->name);
- pipe_mutex_unlock(bo->mgr->bo_handles_mutex);
+ pipe_mutex_lock(rws->bo_handles_mutex);
+ util_hash_table_remove(rws->bo_handles, (void*)(uintptr_t)bo->handle);
+ if (bo->flink_name) {
+ util_hash_table_remove(rws->bo_names,
+ (void*)(uintptr_t)bo->flink_name);
}
+ pipe_mutex_unlock(rws->bo_handles_mutex);
if (bo->ptr)
os_munmap(bo->ptr, bo->base.size);
- if (mgr->va) {
- radeon_bomgr_free_va(mgr, bo->va, bo->va_size);
+ if (rws->info.r600_virtual_address) {
+ if (rws->va_unmap_working) {
+ struct drm_radeon_gem_va va;
+
+ va.handle = bo->handle;
+ va.vm_id = 0;
+ va.operation = RADEON_VA_UNMAP;
+ va.flags = RADEON_VM_PAGE_READABLE |
+ RADEON_VM_PAGE_WRITEABLE |
+ RADEON_VM_PAGE_SNOOPED;
+ va.offset = bo->va;
+
+ if (drmCommandWriteRead(rws->fd, DRM_RADEON_GEM_VA, &va,
+ sizeof(va)) != 0 &&
+ va.operation == RADEON_VA_RESULT_ERROR) {
+ fprintf(stderr, "radeon: Failed to deallocate virtual address for buffer:\n");
+ fprintf(stderr, "radeon: size : %d bytes\n", bo->base.size);
+ fprintf(stderr, "radeon: va : 0x%016llx\n", (unsigned long long)bo->va);
+ }
+ }
+
+ radeon_bomgr_free_va(rws, bo->va, bo->base.size);
}
/* Close object. */
args.handle = bo->handle;
- drmIoctl(bo->rws->fd, DRM_IOCTL_GEM_CLOSE, &args);
+ drmIoctl(rws->fd, DRM_IOCTL_GEM_CLOSE, &args);
+
pipe_mutex_destroy(bo->map_mutex);
+
+ if (bo->initial_domain & RADEON_DOMAIN_VRAM)
+ rws->allocated_vram -= align(bo->base.size, rws->size_align);
+ else if (bo->initial_domain & RADEON_DOMAIN_GTT)
+ rws->allocated_gtt -= align(bo->base.size, rws->size_align);
FREE(bo);
}
-static unsigned get_pb_usage_from_transfer_flags(enum pipe_transfer_usage usage)
+static void radeon_bo_destroy_or_cache(struct pb_buffer *_buf)
+{
+ struct radeon_bo *bo = radeon_bo(_buf);
+
+ if (bo->use_reusable_pool)
+ pb_cache_add_buffer(&bo->cache_entry);
+ else
+ radeon_bo_destroy(_buf);
+}
+
+void *radeon_bo_do_map(struct radeon_bo *bo)
{
- unsigned res = 0;
+ struct drm_radeon_gem_mmap args = {0};
+ void *ptr;
- if (usage & PIPE_TRANSFER_WRITE)
- res |= PB_USAGE_CPU_WRITE;
+ /* If the buffer is created from user memory, return the user pointer. */
+ if (bo->user_ptr)
+ return bo->user_ptr;
- if (usage & PIPE_TRANSFER_DONTBLOCK)
- res |= PB_USAGE_DONTBLOCK;
+ /* Map the buffer. */
+ pipe_mutex_lock(bo->map_mutex);
+ /* Return the pointer if it's already mapped. */
+ if (bo->ptr) {
+ bo->map_count++;
+ pipe_mutex_unlock(bo->map_mutex);
+ return bo->ptr;
+ }
+ args.handle = bo->handle;
+ args.offset = 0;
+ args.size = (uint64_t)bo->base.size;
+ if (drmCommandWriteRead(bo->rws->fd,
+ DRM_RADEON_GEM_MMAP,
+ &args,
+ sizeof(args))) {
+ pipe_mutex_unlock(bo->map_mutex);
+ fprintf(stderr, "radeon: gem_mmap failed: %p 0x%08X\n",
+ bo, bo->handle);
+ return NULL;
+ }
- if (usage & PIPE_TRANSFER_UNSYNCHRONIZED)
- res |= PB_USAGE_UNSYNCHRONIZED;
+ ptr = os_mmap(0, args.size, PROT_READ|PROT_WRITE, MAP_SHARED,
+ bo->rws->fd, args.addr_ptr);
+ if (ptr == MAP_FAILED) {
+ /* Clear the cache and try again. */
+ pb_cache_release_all_buffers(&bo->rws->bo_cache);
+
+ ptr = os_mmap(0, args.size, PROT_READ|PROT_WRITE, MAP_SHARED,
+ bo->rws->fd, args.addr_ptr);
+ if (ptr == MAP_FAILED) {
+ pipe_mutex_unlock(bo->map_mutex);
+ fprintf(stderr, "radeon: mmap failed, errno: %i\n", errno);
+ return NULL;
+ }
+ }
+ bo->ptr = ptr;
+ bo->map_count = 1;
+ pipe_mutex_unlock(bo->map_mutex);
- return res;
+ return bo->ptr;
}
-static void *radeon_bo_map_internal(struct pb_buffer *_buf,
- unsigned flags, void *flush_ctx)
+static void *radeon_bo_map(struct pb_buffer *buf,
+ struct radeon_winsys_cs *rcs,
+ enum pipe_transfer_usage usage)
{
- struct radeon_bo *bo = radeon_bo(_buf);
- struct radeon_drm_cs *cs = flush_ctx;
- struct drm_radeon_gem_mmap args;
- void *ptr;
-
- memset(&args, 0, sizeof(args));
+ struct radeon_bo *bo = (struct radeon_bo*)buf;
+ struct radeon_drm_cs *cs = (struct radeon_drm_cs*)rcs;
/* If it's not unsynchronized bo_map, flush CS if needed and then wait. */
- if (!(flags & PB_USAGE_UNSYNCHRONIZED)) {
+ if (!(usage & PIPE_TRANSFER_UNSYNCHRONIZED)) {
/* DONTBLOCK doesn't make sense with UNSYNCHRONIZED. */
- if (flags & PB_USAGE_DONTBLOCK) {
- if (!(flags & PB_USAGE_CPU_WRITE)) {
+ if (usage & PIPE_TRANSFER_DONTBLOCK) {
+ if (!(usage & PIPE_TRANSFER_WRITE)) {
/* Mapping for read.
*
* Since we are mapping for read, we don't need to wait
* (neither one is changing it).
*
* Only check whether the buffer is being used for write. */
- if (radeon_bo_is_referenced_by_cs_for_write(cs, bo)) {
- cs->flush_cs(cs->flush_data, RADEON_FLUSH_ASYNC);
+ if (cs && radeon_bo_is_referenced_by_cs_for_write(cs, bo)) {
+ cs->flush_cs(cs->flush_data, RADEON_FLUSH_ASYNC, NULL);
return NULL;
}
- if (radeon_bo_is_busy((struct pb_buffer*)bo,
- RADEON_USAGE_WRITE)) {
+ if (!radeon_bo_wait((struct pb_buffer*)bo, 0,
+ RADEON_USAGE_WRITE)) {
return NULL;
}
} else {
- if (radeon_bo_is_referenced_by_cs(cs, bo)) {
- cs->flush_cs(cs->flush_data, RADEON_FLUSH_ASYNC);
+ if (cs && radeon_bo_is_referenced_by_cs(cs, bo)) {
+ cs->flush_cs(cs->flush_data, RADEON_FLUSH_ASYNC, NULL);
return NULL;
}
- if (radeon_bo_is_busy((struct pb_buffer*)bo,
- RADEON_USAGE_READWRITE)) {
+ if (!radeon_bo_wait((struct pb_buffer*)bo, 0,
+ RADEON_USAGE_READWRITE)) {
return NULL;
}
}
} else {
- if (!(flags & PB_USAGE_CPU_WRITE)) {
+ uint64_t time = os_time_get_nano();
+
+ if (!(usage & PIPE_TRANSFER_WRITE)) {
/* Mapping for read.
*
* Since we are mapping for read, we don't need to wait
* (neither one is changing it).
*
* Only check whether the buffer is being used for write. */
- if (radeon_bo_is_referenced_by_cs_for_write(cs, bo)) {
- cs->flush_cs(cs->flush_data, 0);
+ if (cs && radeon_bo_is_referenced_by_cs_for_write(cs, bo)) {
+ cs->flush_cs(cs->flush_data, 0, NULL);
}
- radeon_bo_wait((struct pb_buffer*)bo,
+ radeon_bo_wait((struct pb_buffer*)bo, PIPE_TIMEOUT_INFINITE,
RADEON_USAGE_WRITE);
} else {
/* Mapping for write. */
- if (radeon_bo_is_referenced_by_cs(cs, bo)) {
- cs->flush_cs(cs->flush_data, 0);
- } else {
- /* Try to avoid busy-waiting in radeon_bo_wait. */
- if (p_atomic_read(&bo->num_active_ioctls))
- radeon_drm_cs_sync_flush(cs);
+ if (cs) {
+ if (radeon_bo_is_referenced_by_cs(cs, bo)) {
+ cs->flush_cs(cs->flush_data, 0, NULL);
+ } else {
+ /* Try to avoid busy-waiting in radeon_bo_wait. */
+ if (p_atomic_read(&bo->num_active_ioctls))
+ radeon_drm_cs_sync_flush(rcs);
+ }
}
- radeon_bo_wait((struct pb_buffer*)bo, RADEON_USAGE_READWRITE);
+ radeon_bo_wait((struct pb_buffer*)bo, PIPE_TIMEOUT_INFINITE,
+ RADEON_USAGE_READWRITE);
}
+
+ bo->rws->buffer_wait_time += os_time_get_nano() - time;
}
}
- /* Return the pointer if it's already mapped. */
- if (bo->ptr)
- return bo->ptr;
+ return radeon_bo_do_map(bo);
+}
+
+static void radeon_bo_unmap(struct pb_buffer *_buf)
+{
+ struct radeon_bo *bo = (struct radeon_bo*)_buf;
+
+ if (bo->user_ptr)
+ return;
- /* Map the buffer. */
pipe_mutex_lock(bo->map_mutex);
- /* Return the pointer if it's already mapped (in case of a race). */
- if (bo->ptr) {
+ if (!bo->ptr) {
pipe_mutex_unlock(bo->map_mutex);
- return bo->ptr;
- }
- args.handle = bo->handle;
- args.offset = 0;
- args.size = (uint64_t)bo->base.size;
- if (drmCommandWriteRead(bo->rws->fd,
- DRM_RADEON_GEM_MMAP,
- &args,
- sizeof(args))) {
- pipe_mutex_unlock(bo->map_mutex);
- fprintf(stderr, "radeon: gem_mmap failed: %p 0x%08X\n",
- bo, bo->handle);
- return NULL;
+ return; /* it's not been mapped */
}
- ptr = os_mmap(0, args.size, PROT_READ|PROT_WRITE, MAP_SHARED,
- bo->rws->fd, args.addr_ptr);
- if (ptr == MAP_FAILED) {
+ assert(bo->map_count);
+ if (--bo->map_count) {
pipe_mutex_unlock(bo->map_mutex);
- fprintf(stderr, "radeon: mmap failed, errno: %i\n", errno);
- return NULL;
+ return; /* it's been mapped multiple times */
}
- bo->ptr = ptr;
- pipe_mutex_unlock(bo->map_mutex);
- return bo->ptr;
+ os_munmap(bo->ptr, bo->base.size);
+ bo->ptr = NULL;
+ pipe_mutex_unlock(bo->map_mutex);
}
-static void radeon_bo_unmap_internal(struct pb_buffer *_buf)
-{
- /* NOP */
-}
+static const struct pb_vtbl radeon_bo_vtbl = {
+ radeon_bo_destroy_or_cache
+ /* other functions are never called */
+};
-static void radeon_bo_get_base_buffer(struct pb_buffer *buf,
- struct pb_buffer **base_buf,
- unsigned *offset)
-{
- *base_buf = buf;
- *offset = 0;
-}
+#ifndef RADEON_GEM_GTT_WC
+#define RADEON_GEM_GTT_WC (1 << 2)
+#endif
+#ifndef RADEON_GEM_CPU_ACCESS
+/* BO is expected to be accessed by the CPU */
+#define RADEON_GEM_CPU_ACCESS (1 << 3)
+#endif
+#ifndef RADEON_GEM_NO_CPU_ACCESS
+/* CPU access is not expected to work for this BO */
+#define RADEON_GEM_NO_CPU_ACCESS (1 << 4)
+#endif
-static enum pipe_error radeon_bo_validate(struct pb_buffer *_buf,
- struct pb_validate *vl,
+static struct radeon_bo *radeon_create_bo(struct radeon_drm_winsys *rws,
+ unsigned size, unsigned alignment,
+ unsigned usage,
+ unsigned initial_domains,
unsigned flags)
{
- /* Always pinned */
- return PIPE_OK;
-}
-
-static void radeon_bo_fence(struct pb_buffer *buf,
- struct pipe_fence_handle *fence)
-{
-}
-
-const struct pb_vtbl radeon_bo_vtbl = {
- radeon_bo_destroy,
- radeon_bo_map_internal,
- radeon_bo_unmap_internal,
- radeon_bo_validate,
- radeon_bo_fence,
- radeon_bo_get_base_buffer,
-};
-
-static struct pb_buffer *radeon_bomgr_create_bo(struct pb_manager *_mgr,
- pb_size size,
- const struct pb_desc *desc)
-{
- struct radeon_bomgr *mgr = radeon_bomgr(_mgr);
- struct radeon_drm_winsys *rws = mgr->rws;
struct radeon_bo *bo;
struct drm_radeon_gem_create args;
- struct radeon_bo_desc *rdesc = (struct radeon_bo_desc*)desc;
int r;
memset(&args, 0, sizeof(args));
- assert(rdesc->initial_domains);
- assert((rdesc->initial_domains &
+ assert(initial_domains);
+ assert((initial_domains &
~(RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM)) == 0);
args.size = size;
- args.alignment = desc->alignment;
- args.initial_domain = rdesc->initial_domains;
+ args.alignment = alignment;
+ args.initial_domain = initial_domains;
+ args.flags = 0;
+
+ if (flags & RADEON_FLAG_GTT_WC)
+ args.flags |= RADEON_GEM_GTT_WC;
+ if (flags & RADEON_FLAG_CPU_ACCESS)
+ args.flags |= RADEON_GEM_CPU_ACCESS;
+ if (flags & RADEON_FLAG_NO_CPU_ACCESS)
+ args.flags |= RADEON_GEM_NO_CPU_ACCESS;
if (drmCommandWriteRead(rws->fd, DRM_RADEON_GEM_CREATE,
&args, sizeof(args))) {
fprintf(stderr, "radeon: Failed to allocate a buffer:\n");
fprintf(stderr, "radeon: size : %d bytes\n", size);
- fprintf(stderr, "radeon: alignment : %d bytes\n", desc->alignment);
+ fprintf(stderr, "radeon: alignment : %d bytes\n", alignment);
fprintf(stderr, "radeon: domains : %d\n", args.initial_domain);
+ fprintf(stderr, "radeon: flags : %d\n", args.flags);
return NULL;
}
return NULL;
pipe_reference_init(&bo->base.reference, 1);
- bo->base.alignment = desc->alignment;
- bo->base.usage = desc->usage;
+ bo->base.alignment = alignment;
+ bo->base.usage = usage;
bo->base.size = size;
bo->base.vtbl = &radeon_bo_vtbl;
- bo->mgr = mgr;
- bo->rws = mgr->rws;
+ bo->rws = rws;
bo->handle = args.handle;
bo->va = 0;
+ bo->initial_domain = initial_domains;
pipe_mutex_init(bo->map_mutex);
+ pb_cache_init_entry(&rws->bo_cache, &bo->cache_entry, &bo->base);
- if (mgr->va) {
+ if (rws->info.r600_virtual_address) {
struct drm_radeon_gem_va va;
- bo->va_size = align(size, 4096);
- bo->va = radeon_bomgr_find_va(mgr, bo->va_size);
+ bo->va = radeon_bomgr_find_va(rws, size, alignment);
va.handle = bo->handle;
va.vm_id = 0;
va.offset = bo->va;
r = drmCommandWriteRead(rws->fd, DRM_RADEON_GEM_VA, &va, sizeof(va));
if (r && va.operation == RADEON_VA_RESULT_ERROR) {
- fprintf(stderr, "radeon: Failed to allocate a buffer:\n");
+ fprintf(stderr, "radeon: Failed to allocate virtual address for buffer:\n");
fprintf(stderr, "radeon: size : %d bytes\n", size);
- fprintf(stderr, "radeon: alignment : %d bytes\n", desc->alignment);
+ fprintf(stderr, "radeon: alignment : %d bytes\n", alignment);
fprintf(stderr, "radeon: domains : %d\n", args.initial_domain);
+ fprintf(stderr, "radeon: va : 0x%016llx\n", (unsigned long long)bo->va);
radeon_bo_destroy(&bo->base);
return NULL;
}
+ pipe_mutex_lock(rws->bo_handles_mutex);
if (va.operation == RADEON_VA_RESULT_VA_EXIST) {
- radeon_bomgr_free_va(mgr, bo->va, bo->va_size);
- bo->va = va.offset;
- radeon_bomgr_force_va(mgr, bo->va, bo->va_size);
+ struct pb_buffer *b = &bo->base;
+ struct radeon_bo *old_bo =
+ util_hash_table_get(rws->bo_vas, (void*)(uintptr_t)va.offset);
+
+ pipe_mutex_unlock(rws->bo_handles_mutex);
+ pb_reference(&b, &old_bo->base);
+ return radeon_bo(b);
}
+
+ util_hash_table_set(rws->bo_vas, (void*)(uintptr_t)bo->va, bo);
+ pipe_mutex_unlock(rws->bo_handles_mutex);
}
- return &bo->base;
-}
+ if (initial_domains & RADEON_DOMAIN_VRAM)
+ rws->allocated_vram += align(size, rws->size_align);
+ else if (initial_domains & RADEON_DOMAIN_GTT)
+ rws->allocated_gtt += align(size, rws->size_align);
-static void radeon_bomgr_flush(struct pb_manager *mgr)
-{
- /* NOP */
+ return bo;
}
-/* This is for the cache bufmgr. */
-static boolean radeon_bomgr_is_buffer_busy(struct pb_manager *_mgr,
- struct pb_buffer *_buf)
+bool radeon_bo_can_reclaim(struct pb_buffer *_buf)
{
struct radeon_bo *bo = radeon_bo(_buf);
- if (radeon_bo_is_referenced_by_any_cs(bo)) {
- return TRUE;
- }
-
- if (radeon_bo_is_busy((struct pb_buffer*)bo, RADEON_USAGE_READWRITE)) {
- return TRUE;
- }
-
- return FALSE;
-}
-
-static void radeon_bomgr_destroy(struct pb_manager *_mgr)
-{
- struct radeon_bomgr *mgr = radeon_bomgr(_mgr);
- util_hash_table_destroy(mgr->bo_handles);
- pipe_mutex_destroy(mgr->bo_handles_mutex);
- pipe_mutex_destroy(mgr->bo_va_mutex);
- FREE(mgr);
-}
-
-#define PTR_TO_UINT(x) ((unsigned)((intptr_t)(x)))
+ if (radeon_bo_is_referenced_by_any_cs(bo))
+ return false;
-static unsigned handle_hash(void *key)
-{
- return PTR_TO_UINT(key);
-}
-
-static int handle_compare(void *key1, void *key2)
-{
- return PTR_TO_UINT(key1) != PTR_TO_UINT(key2);
-}
-
-struct pb_manager *radeon_bomgr_create(struct radeon_drm_winsys *rws)
-{
- struct radeon_bomgr *mgr;
-
- mgr = CALLOC_STRUCT(radeon_bomgr);
- if (!mgr)
- return NULL;
-
- mgr->base.destroy = radeon_bomgr_destroy;
- mgr->base.create_buffer = radeon_bomgr_create_bo;
- mgr->base.flush = radeon_bomgr_flush;
- mgr->base.is_buffer_busy = radeon_bomgr_is_buffer_busy;
-
- mgr->rws = rws;
- mgr->bo_handles = util_hash_table_create(handle_hash, handle_compare);
- pipe_mutex_init(mgr->bo_handles_mutex);
- pipe_mutex_init(mgr->bo_va_mutex);
-
- mgr->va = rws->info.r600_virtual_address;
- mgr->va_offset = rws->info.r600_va_start;
- list_inithead(&mgr->va_holes);
-
- return &mgr->base;
-}
-
-static void *radeon_bo_map(struct pb_buffer *buf,
- struct radeon_winsys_cs *cs,
- enum pipe_transfer_usage usage)
-{
- return pb_map(buf, get_pb_usage_from_transfer_flags(usage), cs);
+ return radeon_bo_wait(_buf, 0, RADEON_USAGE_READWRITE);
}
static unsigned eg_tile_split(unsigned tile_split)
return tile_split;
}
+static unsigned eg_tile_split_rev(unsigned eg_tile_split)
+{
+ switch (eg_tile_split) {
+ case 64: return 0;
+ case 128: return 1;
+ case 256: return 2;
+ case 512: return 3;
+ default:
+ case 1024: return 4;
+ case 2048: return 5;
+ case 4096: return 6;
+ }
+}
+
static void radeon_bo_get_tiling(struct pb_buffer *_buf,
enum radeon_bo_layout *microtiled,
enum radeon_bo_layout *macrotiled,
unsigned *bankw, unsigned *bankh,
unsigned *tile_split,
unsigned *stencil_tile_split,
- unsigned *mtilea)
+ unsigned *mtilea,
+ bool *scanout)
{
- struct radeon_bo *bo = get_radeon_bo(_buf);
+ struct radeon_bo *bo = radeon_bo(_buf);
struct drm_radeon_gem_set_tiling args;
memset(&args, 0, sizeof(args));
*microtiled = RADEON_LAYOUT_LINEAR;
*macrotiled = RADEON_LAYOUT_LINEAR;
- if (args.tiling_flags & RADEON_BO_FLAGS_MICRO_TILE)
+ if (args.tiling_flags & RADEON_TILING_MICRO)
*microtiled = RADEON_LAYOUT_TILED;
+ else if (args.tiling_flags & RADEON_TILING_MICRO_SQUARE)
+ *microtiled = RADEON_LAYOUT_SQUARETILED;
- if (args.tiling_flags & RADEON_BO_FLAGS_MACRO_TILE)
+ if (args.tiling_flags & RADEON_TILING_MACRO)
*macrotiled = RADEON_LAYOUT_TILED;
if (bankw && tile_split && stencil_tile_split && mtilea && tile_split) {
*bankw = (args.tiling_flags >> RADEON_TILING_EG_BANKW_SHIFT) & RADEON_TILING_EG_BANKW_MASK;
*mtilea = (args.tiling_flags >> RADEON_TILING_EG_MACRO_TILE_ASPECT_SHIFT) & RADEON_TILING_EG_MACRO_TILE_ASPECT_MASK;
*tile_split = eg_tile_split(*tile_split);
}
+ if (scanout)
+ *scanout = bo->rws->gen >= DRV_SI && !(args.tiling_flags & RADEON_TILING_R600_NO_SCANOUT);
}
static void radeon_bo_set_tiling(struct pb_buffer *_buf,
struct radeon_winsys_cs *rcs,
enum radeon_bo_layout microtiled,
enum radeon_bo_layout macrotiled,
- uint32_t pitch)
+ unsigned pipe_config,
+ unsigned bankw, unsigned bankh,
+ unsigned tile_split,
+ unsigned stencil_tile_split,
+ unsigned mtilea, unsigned num_banks,
+ uint32_t pitch,
+ bool scanout)
{
- struct radeon_bo *bo = get_radeon_bo(_buf);
+ struct radeon_bo *bo = radeon_bo(_buf);
struct radeon_drm_cs *cs = radeon_drm_cs(rcs);
struct drm_radeon_gem_set_tiling args;
/* Tiling determines how DRM treats the buffer data.
* We must flush CS when changing it if the buffer is referenced. */
if (cs && radeon_bo_is_referenced_by_cs(cs, bo)) {
- cs->flush_cs(cs->flush_data, 0);
+ cs->flush_cs(cs->flush_data, 0, NULL);
}
- while (p_atomic_read(&bo->num_active_ioctls)) {
- sched_yield();
- }
+ os_wait_until_zero(&bo->num_active_ioctls, PIPE_TIMEOUT_INFINITE);
if (microtiled == RADEON_LAYOUT_TILED)
- args.tiling_flags |= RADEON_BO_FLAGS_MICRO_TILE;
+ args.tiling_flags |= RADEON_TILING_MICRO;
else if (microtiled == RADEON_LAYOUT_SQUARETILED)
- args.tiling_flags |= RADEON_BO_FLAGS_MICRO_TILE_SQUARE;
+ args.tiling_flags |= RADEON_TILING_MICRO_SQUARE;
if (macrotiled == RADEON_LAYOUT_TILED)
- args.tiling_flags |= RADEON_BO_FLAGS_MACRO_TILE;
+ args.tiling_flags |= RADEON_TILING_MACRO;
+
+ args.tiling_flags |= (bankw & RADEON_TILING_EG_BANKW_MASK) <<
+ RADEON_TILING_EG_BANKW_SHIFT;
+ args.tiling_flags |= (bankh & RADEON_TILING_EG_BANKH_MASK) <<
+ RADEON_TILING_EG_BANKH_SHIFT;
+ if (tile_split) {
+ args.tiling_flags |= (eg_tile_split_rev(tile_split) &
+ RADEON_TILING_EG_TILE_SPLIT_MASK) <<
+ RADEON_TILING_EG_TILE_SPLIT_SHIFT;
+ }
+ args.tiling_flags |= (stencil_tile_split &
+ RADEON_TILING_EG_STENCIL_TILE_SPLIT_MASK) <<
+ RADEON_TILING_EG_STENCIL_TILE_SPLIT_SHIFT;
+ args.tiling_flags |= (mtilea & RADEON_TILING_EG_MACRO_TILE_ASPECT_MASK) <<
+ RADEON_TILING_EG_MACRO_TILE_ASPECT_SHIFT;
+
+ if (bo->rws->gen >= DRV_SI && !scanout)
+ args.tiling_flags |= RADEON_TILING_R600_NO_SCANOUT;
args.handle = bo->handle;
args.pitch = pitch;
sizeof(args));
}
-static struct radeon_winsys_cs_handle *radeon_drm_get_cs_handle(
- struct pb_buffer *_buf)
-{
- /* return radeon_bo. */
- return (struct radeon_winsys_cs_handle*)get_radeon_bo(_buf);
-}
-
static struct pb_buffer *
radeon_winsys_bo_create(struct radeon_winsys *rws,
unsigned size,
unsigned alignment,
- unsigned bind,
- enum radeon_bo_domain domain)
+ boolean use_reusable_pool,
+ enum radeon_bo_domain domain,
+ enum radeon_bo_flag flags)
{
struct radeon_drm_winsys *ws = radeon_drm_winsys(rws);
- struct radeon_bo_desc desc;
- struct pb_manager *provider;
- struct pb_buffer *buffer;
+ struct radeon_bo *bo;
+ unsigned usage = 0;
+
+ /* Align size to page size. This is the minimum alignment for normal
+ * BOs. Aligning this here helps the cached bufmgr. Especially small BOs,
+ * like constant/uniform buffers, can benefit from better and more reuse.
+ */
+ size = align(size, ws->size_align);
+
+ /* Only set one usage bit each for domains and flags, or the cache manager
+ * might consider different sets of domains / flags compatible
+ */
+ if (domain == RADEON_DOMAIN_VRAM_GTT)
+ usage = 1 << 2;
+ else
+ usage = domain >> 1;
+ assert(flags < sizeof(usage) * 8 - 3);
+ usage |= 1 << (flags + 3);
+
+ if (use_reusable_pool) {
+ bo = radeon_bo(pb_cache_reclaim_buffer(&ws->bo_cache, size, alignment, usage));
+ if (bo)
+ return &bo->base;
+ }
- memset(&desc, 0, sizeof(desc));
- desc.base.alignment = alignment;
+ bo = radeon_create_bo(ws, size, alignment, usage, domain, flags);
+ if (!bo) {
+ /* Clear the cache and try again. */
+ pb_cache_release_all_buffers(&ws->bo_cache);
+ bo = radeon_create_bo(ws, size, alignment, usage, domain, flags);
+ if (!bo)
+ return NULL;
+ }
- /* Additional criteria for the cache manager. */
- desc.base.usage = domain;
- desc.initial_domains = domain;
+ bo->use_reusable_pool = use_reusable_pool;
- /* Assign a buffer manager. */
- if (bind & (PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER |
- PIPE_BIND_CONSTANT_BUFFER | PIPE_BIND_CUSTOM))
- provider = ws->cman;
- else
- provider = ws->kman;
+ pipe_mutex_lock(ws->bo_handles_mutex);
+ util_hash_table_set(ws->bo_handles, (void*)(uintptr_t)bo->handle, bo);
+ pipe_mutex_unlock(ws->bo_handles_mutex);
+
+ return &bo->base;
+}
- buffer = provider->create_buffer(provider, size, &desc.base);
- if (!buffer)
+static struct pb_buffer *radeon_winsys_bo_from_ptr(struct radeon_winsys *rws,
+ void *pointer, unsigned size)
+{
+ struct radeon_drm_winsys *ws = radeon_drm_winsys(rws);
+ struct drm_radeon_gem_userptr args;
+ struct radeon_bo *bo;
+ int r;
+
+ bo = CALLOC_STRUCT(radeon_bo);
+ if (!bo)
+ return NULL;
+
+ memset(&args, 0, sizeof(args));
+ args.addr = (uintptr_t)pointer;
+ args.size = align(size, sysconf(_SC_PAGE_SIZE));
+ args.flags = RADEON_GEM_USERPTR_ANONONLY |
+ RADEON_GEM_USERPTR_VALIDATE |
+ RADEON_GEM_USERPTR_REGISTER;
+ if (drmCommandWriteRead(ws->fd, DRM_RADEON_GEM_USERPTR,
+ &args, sizeof(args))) {
+ FREE(bo);
return NULL;
+ }
+
+ pipe_mutex_lock(ws->bo_handles_mutex);
+
+ /* Initialize it. */
+ pipe_reference_init(&bo->base.reference, 1);
+ bo->handle = args.handle;
+ bo->base.alignment = 0;
+ bo->base.usage = PB_USAGE_GPU_WRITE | PB_USAGE_GPU_READ;
+ bo->base.size = size;
+ bo->base.vtbl = &radeon_bo_vtbl;
+ bo->rws = ws;
+ bo->user_ptr = pointer;
+ bo->va = 0;
+ bo->initial_domain = RADEON_DOMAIN_GTT;
+ pipe_mutex_init(bo->map_mutex);
+
+ util_hash_table_set(ws->bo_handles, (void*)(uintptr_t)bo->handle, bo);
+
+ pipe_mutex_unlock(ws->bo_handles_mutex);
+
+ if (ws->info.r600_virtual_address) {
+ struct drm_radeon_gem_va va;
+
+ bo->va = radeon_bomgr_find_va(ws, bo->base.size, 1 << 20);
+
+ va.handle = bo->handle;
+ va.operation = RADEON_VA_MAP;
+ va.vm_id = 0;
+ va.offset = bo->va;
+ va.flags = RADEON_VM_PAGE_READABLE |
+ RADEON_VM_PAGE_WRITEABLE |
+ RADEON_VM_PAGE_SNOOPED;
+ va.offset = bo->va;
+ r = drmCommandWriteRead(ws->fd, DRM_RADEON_GEM_VA, &va, sizeof(va));
+ if (r && va.operation == RADEON_VA_RESULT_ERROR) {
+ fprintf(stderr, "radeon: Failed to assign virtual address space\n");
+ radeon_bo_destroy(&bo->base);
+ return NULL;
+ }
+ pipe_mutex_lock(ws->bo_handles_mutex);
+ if (va.operation == RADEON_VA_RESULT_VA_EXIST) {
+ struct pb_buffer *b = &bo->base;
+ struct radeon_bo *old_bo =
+ util_hash_table_get(ws->bo_vas, (void*)(uintptr_t)va.offset);
+
+ pipe_mutex_unlock(ws->bo_handles_mutex);
+ pb_reference(&b, &old_bo->base);
+ return b;
+ }
+
+ util_hash_table_set(ws->bo_vas, (void*)(uintptr_t)bo->va, bo);
+ pipe_mutex_unlock(ws->bo_handles_mutex);
+ }
- return (struct pb_buffer*)buffer;
+ ws->allocated_gtt += align(bo->base.size, ws->size_align);
+
+ return (struct pb_buffer*)bo;
}
static struct pb_buffer *radeon_winsys_bo_from_handle(struct radeon_winsys *rws,
{
struct radeon_drm_winsys *ws = radeon_drm_winsys(rws);
struct radeon_bo *bo;
- struct radeon_bomgr *mgr = radeon_bomgr(ws->kman);
- struct drm_gem_open open_arg = {};
int r;
-
- memset(&open_arg, 0, sizeof(open_arg));
+ unsigned handle;
+ uint64_t size = 0;
/* We must maintain a list of pairs <handle, bo>, so that we always return
* the same BO for one particular handle. If we didn't do that and created
* we would hit a deadlock in the kernel.
*
* The list of pairs is guarded by a mutex, of course. */
- pipe_mutex_lock(mgr->bo_handles_mutex);
+ pipe_mutex_lock(ws->bo_handles_mutex);
+
+ if (whandle->type == DRM_API_HANDLE_TYPE_SHARED) {
+ /* First check if there already is an existing bo for the handle. */
+ bo = util_hash_table_get(ws->bo_names, (void*)(uintptr_t)whandle->handle);
+ } else if (whandle->type == DRM_API_HANDLE_TYPE_FD) {
+ /* We must first get the GEM handle, as fds are unreliable keys */
+ r = drmPrimeFDToHandle(ws->fd, whandle->handle, &handle);
+ if (r)
+ goto fail;
+ bo = util_hash_table_get(ws->bo_handles, (void*)(uintptr_t)handle);
+ } else {
+ /* Unknown handle type */
+ goto fail;
+ }
- /* First check if there already is an existing bo for the handle. */
- bo = util_hash_table_get(mgr->bo_handles, (void*)(uintptr_t)whandle->handle);
if (bo) {
/* Increase the refcount. */
struct pb_buffer *b = NULL;
goto fail;
}
- /* Open the BO. */
- open_arg.name = whandle->handle;
- if (drmIoctl(ws->fd, DRM_IOCTL_GEM_OPEN, &open_arg)) {
- FREE(bo);
- goto fail;
+ if (whandle->type == DRM_API_HANDLE_TYPE_SHARED) {
+ struct drm_gem_open open_arg = {};
+ memset(&open_arg, 0, sizeof(open_arg));
+ /* Open the BO. */
+ open_arg.name = whandle->handle;
+ if (drmIoctl(ws->fd, DRM_IOCTL_GEM_OPEN, &open_arg)) {
+ FREE(bo);
+ goto fail;
+ }
+ handle = open_arg.handle;
+ size = open_arg.size;
+ bo->flink_name = whandle->handle;
+ } else if (whandle->type == DRM_API_HANDLE_TYPE_FD) {
+ size = lseek(whandle->handle, 0, SEEK_END);
+ /*
+ * Could check errno to determine whether the kernel is new enough, but
+ * it doesn't really matter why this failed, just that it failed.
+ */
+ if (size == (off_t)-1) {
+ FREE(bo);
+ goto fail;
+ }
+ lseek(whandle->handle, 0, SEEK_SET);
}
- bo->handle = open_arg.handle;
- bo->name = whandle->handle;
+
+ bo->handle = handle;
/* Initialize it. */
pipe_reference_init(&bo->base.reference, 1);
bo->base.alignment = 0;
bo->base.usage = PB_USAGE_GPU_WRITE | PB_USAGE_GPU_READ;
- bo->base.size = open_arg.size;
+ bo->base.size = (unsigned) size;
bo->base.vtbl = &radeon_bo_vtbl;
- bo->mgr = mgr;
- bo->rws = mgr->rws;
+ bo->rws = ws;
bo->va = 0;
pipe_mutex_init(bo->map_mutex);
- util_hash_table_set(mgr->bo_handles, (void*)(uintptr_t)whandle->handle, bo);
+ if (bo->flink_name)
+ util_hash_table_set(ws->bo_names, (void*)(uintptr_t)bo->flink_name, bo);
+
+ util_hash_table_set(ws->bo_handles, (void*)(uintptr_t)bo->handle, bo);
done:
- pipe_mutex_unlock(mgr->bo_handles_mutex);
+ pipe_mutex_unlock(ws->bo_handles_mutex);
if (stride)
*stride = whandle->stride;
- if (mgr->va) {
+ if (ws->info.r600_virtual_address && !bo->va) {
struct drm_radeon_gem_va va;
- bo->va_size = ((bo->base.size + 4095) & ~4095);
- bo->va = radeon_bomgr_find_va(mgr, bo->va_size);
+ bo->va = radeon_bomgr_find_va(ws, bo->base.size, 1 << 20);
va.handle = bo->handle;
va.operation = RADEON_VA_MAP;
radeon_bo_destroy(&bo->base);
return NULL;
}
+ pipe_mutex_lock(ws->bo_handles_mutex);
if (va.operation == RADEON_VA_RESULT_VA_EXIST) {
- radeon_bomgr_free_va(mgr, bo->va, bo->va_size);
- bo->va = va.offset;
- radeon_bomgr_force_va(mgr, bo->va, bo->va_size);
+ struct pb_buffer *b = &bo->base;
+ struct radeon_bo *old_bo =
+ util_hash_table_get(ws->bo_vas, (void*)(uintptr_t)va.offset);
+
+ pipe_mutex_unlock(ws->bo_handles_mutex);
+ pb_reference(&b, &old_bo->base);
+ return b;
}
+
+ util_hash_table_set(ws->bo_vas, (void*)(uintptr_t)bo->va, bo);
+ pipe_mutex_unlock(ws->bo_handles_mutex);
}
+ bo->initial_domain = radeon_bo_get_initial_domain((void*)bo);
+
+ if (bo->initial_domain & RADEON_DOMAIN_VRAM)
+ ws->allocated_vram += align(bo->base.size, ws->size_align);
+ else if (bo->initial_domain & RADEON_DOMAIN_GTT)
+ ws->allocated_gtt += align(bo->base.size, ws->size_align);
+
return (struct pb_buffer*)bo;
fail:
- pipe_mutex_unlock(mgr->bo_handles_mutex);
+ pipe_mutex_unlock(ws->bo_handles_mutex);
return NULL;
}
struct winsys_handle *whandle)
{
struct drm_gem_flink flink;
- struct radeon_bo *bo = get_radeon_bo(buffer);
+ struct radeon_bo *bo = radeon_bo(buffer);
+ struct radeon_drm_winsys *ws = bo->rws;
memset(&flink, 0, sizeof(flink));
+ bo->use_reusable_pool = false;
+
if (whandle->type == DRM_API_HANDLE_TYPE_SHARED) {
- if (!bo->flinked) {
+ if (!bo->flink_name) {
flink.handle = bo->handle;
- if (ioctl(bo->rws->fd, DRM_IOCTL_GEM_FLINK, &flink)) {
+ if (ioctl(ws->fd, DRM_IOCTL_GEM_FLINK, &flink)) {
return FALSE;
}
- bo->flinked = TRUE;
- bo->flink = flink.name;
+ bo->flink_name = flink.name;
+
+ pipe_mutex_lock(ws->bo_handles_mutex);
+ util_hash_table_set(ws->bo_names, (void*)(uintptr_t)bo->flink_name, bo);
+ pipe_mutex_unlock(ws->bo_handles_mutex);
}
- whandle->handle = bo->flink;
+ whandle->handle = bo->flink_name;
} else if (whandle->type == DRM_API_HANDLE_TYPE_KMS) {
whandle->handle = bo->handle;
+ } else if (whandle->type == DRM_API_HANDLE_TYPE_FD) {
+ if (drmPrimeHandleToFD(ws->fd, bo->handle, DRM_CLOEXEC, (int*)&whandle->handle))
+ return FALSE;
}
whandle->stride = stride;
return TRUE;
}
-static uint64_t radeon_winsys_bo_va(struct pb_buffer *buffer)
+static uint64_t radeon_winsys_bo_va(struct pb_buffer *buf)
{
- struct radeon_bo *bo = get_radeon_bo(buffer);
-
- return bo->va;
+ return ((struct radeon_bo*)buf)->va;
}
-void radeon_bomgr_init_functions(struct radeon_drm_winsys *ws)
+void radeon_drm_bo_init_functions(struct radeon_drm_winsys *ws)
{
- ws->base.buffer_get_cs_handle = radeon_drm_get_cs_handle;
ws->base.buffer_set_tiling = radeon_bo_set_tiling;
ws->base.buffer_get_tiling = radeon_bo_get_tiling;
ws->base.buffer_map = radeon_bo_map;
- ws->base.buffer_unmap = pb_unmap;
+ ws->base.buffer_unmap = radeon_bo_unmap;
ws->base.buffer_wait = radeon_bo_wait;
- ws->base.buffer_is_busy = radeon_bo_is_busy;
ws->base.buffer_create = radeon_winsys_bo_create;
ws->base.buffer_from_handle = radeon_winsys_bo_from_handle;
+ ws->base.buffer_from_ptr = radeon_winsys_bo_from_ptr;
ws->base.buffer_get_handle = radeon_winsys_bo_get_handle;
ws->base.buffer_get_virtual_address = radeon_winsys_bo_va;
+ ws->base.buffer_get_initial_domain = radeon_bo_get_initial_domain;
}
projects / mesa.git / blobdiff