diff options
Diffstat (limited to 'src/gallium/winsys/radeon')
| -rw-r--r-- | src/gallium/winsys/radeon/drm/Makefile | 6 | ||||
| -rw-r--r-- | src/gallium/winsys/radeon/drm/SConscript | 11 | ||||
| -rw-r--r-- | src/gallium/winsys/radeon/drm/radeon_drm_bo.c | 602 | ||||
| -rw-r--r-- | src/gallium/winsys/radeon/drm/radeon_drm_bo.h (renamed from src/gallium/winsys/radeon/drm/radeon_drm_buffer.h) | 54 | ||||
| -rw-r--r-- | src/gallium/winsys/radeon/drm/radeon_drm_buffer.c | 564 | ||||
| -rw-r--r-- | src/gallium/winsys/radeon/drm/radeon_drm_common.c | 129 | ||||
| -rw-r--r-- | src/gallium/winsys/radeon/drm/radeon_drm_cs.c | 446 | ||||
| -rw-r--r-- | src/gallium/winsys/radeon/drm/radeon_drm_cs.h | 102 | ||||
| -rw-r--r-- | src/gallium/winsys/radeon/drm/radeon_r300.c | 266 | ||||
| -rw-r--r-- | src/gallium/winsys/radeon/drm/radeon_winsys.h | 40 |
10 files changed, 1283 insertions, 937 deletions
diff --git a/src/gallium/winsys/radeon/drm/Makefile b/src/gallium/winsys/radeon/drm/Makefile index aa73edde34..e63ae6f500 100644 --- a/src/gallium/winsys/radeon/drm/Makefile +++ b/src/gallium/winsys/radeon/drm/Makefile @@ -5,9 +5,9 @@ include $(TOP)/configs/current LIBNAME = radeonwinsys C_SOURCES = \ - radeon_drm_buffer.c \ - radeon_drm_common.c \ - radeon_r300.c + radeon_drm_bo.c \ + radeon_drm_cs.c \ + radeon_drm_common.c LIBRARY_INCLUDES = -I$(TOP)/src/gallium/drivers/r300 \ $(shell pkg-config libdrm --cflags-only-I) diff --git a/src/gallium/winsys/radeon/drm/SConscript b/src/gallium/winsys/radeon/drm/SConscript index 2dbf61a7ba..b16e03556d 100644 --- a/src/gallium/winsys/radeon/drm/SConscript +++ b/src/gallium/winsys/radeon/drm/SConscript @@ -3,12 +3,17 @@ Import('*') env = env.Clone() radeon_sources = [ - 'radeon_drm_buffer.c', + 'radeon_drm_bo.c', + 'radeon_drm_cs.c', 'radeon_drm_common.c', - 'radeon_r300.c', ] -env.ParseConfig('pkg-config --cflags libdrm_radeon') +try: + env.ParseConfig('pkg-config --cflags libdrm_radeon') +except: + print 'warning: not building r300g' + Return() + env.Append(CPPPATH = '#/src/gallium/drivers/r300') radeonwinsys = env.ConvenienceLibrary( diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_bo.c b/src/gallium/winsys/radeon/drm/radeon_drm_bo.c new file mode 100644 index 0000000000..9eb833454d --- /dev/null +++ b/src/gallium/winsys/radeon/drm/radeon_drm_bo.c @@ -0,0 +1,602 @@ +/* + * Copyright © 2011 Marek Olšák <maraeo@gmail.com> + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES + * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS + * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * 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 "os/os_thread.h" + +#include "state_tracker/drm_driver.h" + +#include <sys/ioctl.h> +#include <sys/mman.h> +#include <xf86drm.h> +#include <errno.h> + +#define RADEON_BO_FLAGS_MACRO_TILE 1 +#define RADEON_BO_FLAGS_MICRO_TILE 2 +#define RADEON_BO_FLAGS_MICRO_TILE_SQUARE 0x20 + +extern const struct pb_vtbl radeon_bo_vtbl; + + +static INLINE struct radeon_bo *radeon_bo(struct pb_buffer *bo) +{ + assert(bo->vtbl == &radeon_bo_vtbl); + return (struct radeon_bo *)bo; +} + +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; +}; + +static INLINE struct radeon_bomgr *radeon_bomgr(struct pb_manager *mgr) +{ + return (struct radeon_bomgr *)mgr; +} + +static struct radeon_bo *get_radeon_bo(struct pb_buffer *_buf) +{ + struct radeon_bo *bo = NULL; + + 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); + + if (base_buf->vtbl == &radeon_bo_vtbl) + bo = radeon_bo(base_buf); + } + + return bo; +} + +void radeon_bo_unref(struct radeon_bo *bo) +{ + struct drm_gem_close args = {}; + + if (!p_atomic_dec_zero(&bo->ref_count)) + return; + + 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); + } + + if (bo->ptr) + munmap(bo->ptr, bo->size); + + /* Close object. */ + args.handle = bo->handle; + drmIoctl(bo->rws->fd, DRM_IOCTL_GEM_CLOSE, &args); + pipe_mutex_destroy(bo->map_mutex); + FREE(bo); +} + +static void radeon_bo_wait(struct r300_winsys_bo *_buf) +{ + struct radeon_bo *bo = get_radeon_bo(pb_buffer(_buf)); + struct drm_radeon_gem_wait_idle args = {}; + + while (p_atomic_read(&bo->num_active_ioctls)) { + sched_yield(); + } + + args.handle = bo->handle; + while (drmCommandWriteRead(bo->rws->fd, DRM_RADEON_GEM_WAIT_IDLE, + &args, sizeof(args)) == -EBUSY); +} + +static boolean radeon_bo_is_busy(struct r300_winsys_bo *_buf) +{ + struct radeon_bo *bo = get_radeon_bo(pb_buffer(_buf)); + struct drm_radeon_gem_busy args = {}; + + if (p_atomic_read(&bo->num_active_ioctls)) { + return TRUE; + } + + args.handle = bo->handle; + return drmCommandWriteRead(bo->rws->fd, DRM_RADEON_GEM_BUSY, + &args, sizeof(args)) != 0; +} + +static void radeon_bo_destroy(struct pb_buffer *_buf) +{ + struct radeon_bo *bo = radeon_bo(_buf); + + radeon_bo_unref(bo); +} + +static unsigned get_pb_usage_from_transfer_flags(enum pipe_transfer_usage usage) +{ + unsigned res = 0; + + if (usage & PIPE_TRANSFER_DONTBLOCK) + res |= PB_USAGE_DONTBLOCK; + + if (usage & PIPE_TRANSFER_UNSYNCHRONIZED) + res |= PB_USAGE_UNSYNCHRONIZED; + + return res; +} + +static void *radeon_bo_map_internal(struct pb_buffer *_buf, + unsigned flags, void *flush_ctx) +{ + struct radeon_bo *bo = radeon_bo(_buf); + struct radeon_drm_cs *cs = flush_ctx; + struct drm_radeon_gem_mmap args = {}; + void *ptr; + + /* If it's not unsynchronized bo_map, flush CS if needed and then wait. */ + if (!(flags & PB_USAGE_UNSYNCHRONIZED)) { + /* DONTBLOCK doesn't make sense with UNSYNCHRONIZED. */ + if (flags & PB_USAGE_DONTBLOCK) { + if (radeon_bo_is_referenced_by_cs(cs, bo)) { + cs->flush_cs(cs->flush_data, R300_FLUSH_ASYNC); + return NULL; + } + + if (radeon_bo_is_busy((struct r300_winsys_bo*)bo)) { + return NULL; + } + } else { + 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); + } + + radeon_bo_wait((struct r300_winsys_bo*)bo); + } + } + + /* Return the pointer if it's already mapped. */ + if (bo->ptr) + return bo->ptr; + + /* Map the buffer. */ + pipe_mutex_lock(bo->map_mutex); + args.handle = bo->handle; + args.offset = 0; + args.size = (uint64_t)bo->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; + } + + ptr = 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; + pipe_mutex_unlock(bo->map_mutex); + + return bo->ptr; +} + +static void radeon_bo_unmap_internal(struct pb_buffer *_buf) +{ + /* NOP */ +} + +static void radeon_bo_get_base_buffer(struct pb_buffer *buf, + struct pb_buffer **base_buf, + unsigned *offset) +{ + *base_buf = buf; + *offset = 0; +} + +static enum pipe_error radeon_bo_validate(struct pb_buffer *_buf, + struct pb_validate *vl, + 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 = {}; + + args.size = size; + args.alignment = desc->alignment; + args.initial_domain = + (desc->usage & RADEON_PB_USAGE_DOMAIN_GTT ? + RADEON_GEM_DOMAIN_GTT : 0) | + (desc->usage & RADEON_PB_USAGE_DOMAIN_VRAM ? + RADEON_GEM_DOMAIN_VRAM : 0); + + if (drmCommandWriteRead(rws->fd, DRM_RADEON_GEM_CREATE, + &args, sizeof(args))) { + fprintf(stderr, "Failed to allocate :\n"); + fprintf(stderr, " size : %d bytes\n", size); + fprintf(stderr, " alignment : %d bytes\n", desc->alignment); + fprintf(stderr, " domains : %d\n", args.initial_domain); + return NULL; + } + + bo = CALLOC_STRUCT(radeon_bo); + if (!bo) + return NULL; + + pipe_reference_init(&bo->base.base.reference, 1); + bo->base.base.alignment = desc->alignment; + bo->base.base.usage = desc->usage; + bo->base.base.size = size; + bo->base.vtbl = &radeon_bo_vtbl; + bo->mgr = mgr; + bo->rws = mgr->rws; + bo->handle = args.handle; + bo->size = size; + pipe_mutex_init(bo->map_mutex); + + radeon_bo_ref(bo); + return &bo->base; +} + +static void radeon_bomgr_flush(struct pb_manager *mgr) +{ + /* NOP */ +} + +/* This is for the cache bufmgr. */ +static boolean radeon_bomgr_is_buffer_busy(struct pb_manager *_mgr, + 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 r300_winsys_bo*)bo)) { + 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); + FREE(mgr); +} + +#define PTR_TO_UINT(x) ((unsigned)((intptr_t)(x))) + +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); + return &mgr->base; +} + +static void *radeon_bo_map(struct r300_winsys_bo *buf, + struct r300_winsys_cs *cs, + enum pipe_transfer_usage usage) +{ + struct pb_buffer *_buf = pb_buffer(buf); + + return pb_map(_buf, get_pb_usage_from_transfer_flags(usage), cs); +} + +static void radeon_bo_get_tiling(struct r300_winsys_bo *_buf, + enum r300_buffer_tiling *microtiled, + enum r300_buffer_tiling *macrotiled) +{ + struct radeon_bo *bo = get_radeon_bo(pb_buffer(_buf)); + struct drm_radeon_gem_set_tiling args = {}; + + args.handle = bo->handle; + + drmCommandWriteRead(bo->rws->fd, + DRM_RADEON_GEM_GET_TILING, + &args, + sizeof(args)); + + *microtiled = R300_BUFFER_LINEAR; + *macrotiled = R300_BUFFER_LINEAR; + if (args.tiling_flags & RADEON_BO_FLAGS_MICRO_TILE) + *microtiled = R300_BUFFER_TILED; + + if (args.tiling_flags & RADEON_BO_FLAGS_MACRO_TILE) + *macrotiled = R300_BUFFER_TILED; +} + +static void radeon_bo_set_tiling(struct r300_winsys_bo *_buf, + struct r300_winsys_cs *rcs, + enum r300_buffer_tiling microtiled, + enum r300_buffer_tiling macrotiled, + uint32_t pitch) +{ + struct radeon_bo *bo = get_radeon_bo(pb_buffer(_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); + } + + while (p_atomic_read(&bo->num_active_ioctls)) { + sched_yield(); + } + + if (microtiled == R300_BUFFER_TILED) + args.tiling_flags |= RADEON_BO_FLAGS_MICRO_TILE; + else if (microtiled == R300_BUFFER_SQUARETILED) + args.tiling_flags |= RADEON_BO_FLAGS_MICRO_TILE_SQUARE; + + if (macrotiled == R300_BUFFER_TILED) + args.tiling_flags |= RADEON_BO_FLAGS_MACRO_TILE; + + args.handle = bo->handle; + args.pitch = pitch; + + drmCommandWriteRead(bo->rws->fd, + DRM_RADEON_GEM_SET_TILING, + &args, + sizeof(args)); +} + +static struct r300_winsys_cs_handle *radeon_drm_get_cs_handle( + struct r300_winsys_bo *_buf) +{ + /* return radeon_bo. */ + return (struct r300_winsys_cs_handle*) + get_radeon_bo(pb_buffer(_buf)); +} + +static unsigned get_pb_usage_from_create_flags(unsigned bind, unsigned usage, + enum r300_buffer_domain domain) +{ + unsigned res = 0; + + if (bind & (PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER)) + res |= RADEON_PB_USAGE_CACHE; + + if (domain & R300_DOMAIN_GTT) + res |= RADEON_PB_USAGE_DOMAIN_GTT; + + if (domain & R300_DOMAIN_VRAM) + res |= RADEON_PB_USAGE_DOMAIN_VRAM; + + return res; +} + +static struct r300_winsys_bo * +radeon_winsys_bo_create(struct r300_winsys_screen *rws, + unsigned size, + unsigned alignment, + unsigned bind, + unsigned usage, + enum r300_buffer_domain domain) +{ + struct radeon_drm_winsys *ws = radeon_drm_winsys(rws); + struct pb_desc desc; + struct pb_manager *provider; + struct pb_buffer *buffer; + + memset(&desc, 0, sizeof(desc)); + desc.alignment = alignment; + desc.usage = get_pb_usage_from_create_flags(bind, usage, domain); + + /* Assign a buffer manager. */ + if (bind & (PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER)) + provider = ws->cman; + else + provider = ws->kman; + + buffer = provider->create_buffer(provider, size, &desc); + if (!buffer) + return NULL; + + return (struct r300_winsys_bo*)buffer; +} + +static struct r300_winsys_bo *radeon_winsys_bo_from_handle(struct r300_winsys_screen *rws, + struct winsys_handle *whandle, + unsigned *stride, + unsigned *size) +{ + 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 = {}; + + /* 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 + * more than one BO for the same handle and then relocated them in a CS, + * 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); + + /* 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; + pb_reference(&b, &bo->base); + goto done; + } + + /* There isn't, create a new one. */ + bo = CALLOC_STRUCT(radeon_bo); + if (!bo) { + goto fail; + } + + /* Open the BO. */ + open_arg.name = whandle->handle; + if (drmIoctl(ws->fd, DRM_IOCTL_GEM_OPEN, &open_arg)) { + FREE(bo); + goto fail; + } + bo->handle = open_arg.handle; + bo->size = open_arg.size; + bo->name = whandle->handle; + radeon_bo_ref(bo); + + /* Initialize it. */ + pipe_reference_init(&bo->base.base.reference, 1); + bo->base.base.alignment = 0; + bo->base.base.usage = PB_USAGE_GPU_WRITE | PB_USAGE_GPU_READ; + bo->base.base.size = bo->size; + bo->base.vtbl = &radeon_bo_vtbl; + bo->mgr = mgr; + bo->rws = mgr->rws; + pipe_mutex_init(bo->map_mutex); + + util_hash_table_set(mgr->bo_handles, (void*)(uintptr_t)whandle->handle, bo); + +done: + pipe_mutex_unlock(mgr->bo_handles_mutex); + + if (stride) + *stride = whandle->stride; + if (size) + *size = bo->base.base.size; + + return (struct r300_winsys_bo*)bo; + +fail: + pipe_mutex_unlock(mgr->bo_handles_mutex); + return NULL; +} + +static boolean radeon_winsys_bo_get_handle(struct r300_winsys_bo *buffer, + unsigned stride, + struct winsys_handle *whandle) +{ + struct drm_gem_flink flink = {}; + struct radeon_bo *bo = get_radeon_bo(pb_buffer(buffer)); + + if (whandle->type == DRM_API_HANDLE_TYPE_SHARED) { + if (!bo->flinked) { + flink.handle = bo->handle; + + if (ioctl(bo->rws->fd, DRM_IOCTL_GEM_FLINK, &flink)) { + return FALSE; + } + + bo->flinked = TRUE; + bo->flink = flink.name; + } + whandle->handle = bo->flink; + } else if (whandle->type == DRM_API_HANDLE_TYPE_KMS) { + whandle->handle = bo->handle; + } + + whandle->stride = stride; + return TRUE; +} + +void radeon_bomgr_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_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_get_handle = radeon_winsys_bo_get_handle; +} diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_buffer.h b/src/gallium/winsys/radeon/drm/radeon_drm_bo.h index 494abdc0b4..a26866b7e7 100644 --- a/src/gallium/winsys/radeon/drm/radeon_drm_buffer.h +++ b/src/gallium/winsys/radeon/drm/radeon_drm_bo.h @@ -1,5 +1,6 @@ /* * Copyright © 2008 Jérôme Glisse + * Copyright © 2011 Marek Olšák <maraeo@gmail.com> * All Rights Reserved. * * Permission is hereby granted, free of charge, to any person obtaining @@ -26,28 +27,61 @@ /* * Authors: * Jérôme Glisse <glisse@freedesktop.org> + * Marek Olšák <maraeo@gmail.com> */ #ifndef RADEON_DRM_BUFFER_H #define RADEON_DRM_BUFFER_H #include "radeon_winsys.h" +#include "pipebuffer/pb_bufmgr.h" +#include "os/os_thread.h" -#define RADEON_PB_USAGE_VERTEX (1 << 28) +#define RADEON_PB_USAGE_CACHE (1 << 28) #define RADEON_PB_USAGE_DOMAIN_GTT (1 << 29) #define RADEON_PB_USAGE_DOMAIN_VRAM (1 << 30) +struct radeon_bomgr; + +struct radeon_bo { + struct pb_buffer base; + struct radeon_bomgr *mgr; + struct radeon_drm_winsys *rws; + + void *ptr; + pipe_mutex map_mutex; + + uint32_t size; + uint32_t handle; + uint32_t name; + + int ref_count; + + /* how many command streams is this bo referenced in? */ + int num_cs_references; + + /* how many command streams, which are being emitted in a separate + * thread, is this bo referenced in? */ + int num_active_ioctls; + + boolean flinked; + uint32_t flink; +}; + +struct pb_manager *radeon_bomgr_create(struct radeon_drm_winsys *rws); +void radeon_bomgr_init_functions(struct radeon_drm_winsys *ws); + +void radeon_bo_unref(struct radeon_bo *buf); + + +static INLINE void radeon_bo_ref(struct radeon_bo *bo) +{ + p_atomic_inc(&bo->ref_count); +} + static INLINE struct pb_buffer * -radeon_pb_buffer(struct r300_winsys_buffer *buffer) +pb_buffer(struct r300_winsys_bo *buffer) { return (struct pb_buffer *)buffer; } -struct pb_manager *radeon_drm_bufmgr_create(struct radeon_drm_winsys *rws); -struct pb_buffer *radeon_drm_bufmgr_create_buffer_from_handle(struct pb_manager *_mgr, - uint32_t handle); -void radeon_drm_bufmgr_flush_maps(struct pb_manager *_mgr); -boolean radeon_drm_bufmgr_get_handle(struct pb_buffer *_buf, - struct winsys_handle *whandle); -void radeon_drm_bufmgr_init_functions(struct radeon_drm_winsys *ws); - #endif diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_buffer.c b/src/gallium/winsys/radeon/drm/radeon_drm_buffer.c deleted file mode 100644 index dc4b51747f..0000000000 --- a/src/gallium/winsys/radeon/drm/radeon_drm_buffer.c +++ /dev/null @@ -1,564 +0,0 @@ -#include "radeon_cs_gem.h" -#include "radeon_drm_buffer.h" - -#include "util/u_hash_table.h" -#include "util/u_memory.h" -#include "util/u_simple_list.h" -#include "pipebuffer/pb_bufmgr.h" -#include "os/os_thread.h" - -#include "state_tracker/drm_driver.h" - -#include <radeon_drm.h> -#include <radeon_bo_gem.h> -#include <sys/ioctl.h> - -struct radeon_drm_bufmgr; - -struct radeon_drm_buffer { - struct pb_buffer base; - struct radeon_drm_bufmgr *mgr; - - struct radeon_bo *bo; - - boolean flinked; - uint32_t flink; - - struct radeon_drm_buffer *next, *prev; -}; - -extern const struct pb_vtbl radeon_drm_buffer_vtbl; - - -static INLINE struct radeon_drm_buffer * -radeon_drm_buffer(struct pb_buffer *buf) -{ - assert(buf); - assert(buf->vtbl == &radeon_drm_buffer_vtbl); - return (struct radeon_drm_buffer *)buf; -} - -struct radeon_drm_bufmgr { - /* Base class. */ - struct pb_manager base; - - /* Winsys. */ - struct radeon_drm_winsys *rws; - - /* List of mapped buffers and its mutex. */ - struct radeon_drm_buffer buffer_map_list; - pipe_mutex buffer_map_list_mutex; - - /* List of buffer handles and its mutex. */ - struct util_hash_table *buffer_handles; - pipe_mutex buffer_handles_mutex; -}; - -static INLINE struct radeon_drm_bufmgr * -radeon_drm_bufmgr(struct pb_manager *mgr) -{ - assert(mgr); - return (struct radeon_drm_bufmgr *)mgr; -} - -static void -radeon_drm_buffer_destroy(struct pb_buffer *_buf) -{ - struct radeon_drm_buffer *buf = radeon_drm_buffer(_buf); - int name; - - if (buf->bo->ptr != NULL) { - pipe_mutex_lock(buf->mgr->buffer_map_list_mutex); - /* Now test it again inside the mutex. */ - if (buf->bo->ptr != NULL) { - remove_from_list(buf); - radeon_bo_unmap(buf->bo); - buf->bo->ptr = NULL; - } - pipe_mutex_unlock(buf->mgr->buffer_map_list_mutex); - } - name = radeon_gem_name_bo(buf->bo); - if (name) { - pipe_mutex_lock(buf->mgr->buffer_handles_mutex); - util_hash_table_remove(buf->mgr->buffer_handles, - (void*)(uintptr_t)name); - pipe_mutex_unlock(buf->mgr->buffer_handles_mutex); - } - radeon_bo_unref(buf->bo); - - FREE(buf); -} - -static unsigned get_pb_usage_from_transfer_flags(enum pipe_transfer_usage usage) -{ - unsigned res = 0; - - if (usage & PIPE_TRANSFER_READ) - res |= PB_USAGE_CPU_READ; - - if (usage & PIPE_TRANSFER_WRITE) - res |= PB_USAGE_CPU_WRITE; - - if (usage & PIPE_TRANSFER_DONTBLOCK) - res |= PB_USAGE_DONTBLOCK; - - if (usage & PIPE_TRANSFER_UNSYNCHRONIZED) - res |= PB_USAGE_UNSYNCHRONIZED; - - return res; -} - -static void * -radeon_drm_buffer_map_internal(struct pb_buffer *_buf, - unsigned flags, void *flush_ctx) -{ - struct radeon_drm_buffer *buf = radeon_drm_buffer(_buf); - struct radeon_drm_cs *cs = flush_ctx; - int write = 0; - - /* Note how we use radeon_bo_is_referenced_by_cs here. There are - * basically two places this map function can be called from: - * - pb_map - * - create_buffer (in the buffer reuse case) - * - * Since pb managers are per-winsys managers, not per-context managers, - * and we shouldn't reuse buffers if they are in-use in any context, - * we simply ask: is this buffer referenced by *any* CS? - * - * The problem with buffer_create is that it comes from pipe_screen, - * so we have no CS to look at, though luckily the following code - * is sufficient to tell whether the buffer is in use. */ - if (flags & PB_USAGE_DONTBLOCK) { - if (_buf->base.usage & RADEON_PB_USAGE_VERTEX) - if (radeon_bo_is_referenced_by_cs(buf->bo, NULL)) - return NULL; - } - - if (buf->bo->ptr != NULL) { - pipe_mutex_lock(buf->mgr->buffer_map_list_mutex); - /* Now test ptr again inside the mutex. We might have gotten a race - * during the first test. */ - if (buf->bo->ptr != NULL) { - remove_from_list(buf); - } - pipe_mutex_unlock(buf->mgr->buffer_map_list_mutex); - return buf->bo->ptr; - } - - if (flags & PB_USAGE_DONTBLOCK) { - uint32_t domain; - if (radeon_bo_is_busy(buf->bo, &domain)) - return NULL; - } - - /* If we don't have any CS and the buffer is referenced, - * we cannot flush. */ - assert(cs || !radeon_bo_is_referenced_by_cs(buf->bo, NULL)); - - if (cs && radeon_bo_is_referenced_by_cs(buf->bo, cs->cs)) { - cs->flush_cs(cs->flush_data); - } - - if (flags & PB_USAGE_CPU_WRITE) { - write = 1; - } - - if (radeon_bo_map(buf->bo, write)) { - return NULL; - } - - pipe_mutex_lock(buf->mgr->buffer_map_list_mutex); - remove_from_list(buf); - pipe_mutex_unlock(buf->mgr->buffer_map_list_mutex); - return buf->bo->ptr; -} - -static void -radeon_drm_buffer_unmap_internal(struct pb_buffer *_buf) -{ - struct radeon_drm_buffer *buf = radeon_drm_buffer(_buf); - pipe_mutex_lock(buf->mgr->buffer_map_list_mutex); - if (is_empty_list(buf)) { /* = is not inserted... */ - insert_at_tail(&buf->mgr->buffer_map_list, buf); - } - pipe_mutex_unlock(buf->mgr->buffer_map_list_mutex); -} - -static void -radeon_drm_buffer_get_base_buffer(struct pb_buffer *buf, - struct pb_buffer **base_buf, - unsigned *offset) -{ - *base_buf = buf; - *offset = 0; -} - - -static enum pipe_error -radeon_drm_buffer_validate(struct pb_buffer *_buf, - struct pb_validate *vl, - unsigned flags) -{ - /* Always pinned */ - return PIPE_OK; -} - -static void -radeon_drm_buffer_fence(struct pb_buffer *buf, - struct pipe_fence_handle *fence) -{ -} - -const struct pb_vtbl radeon_drm_buffer_vtbl = { - radeon_drm_buffer_destroy, - radeon_drm_buffer_map_internal, - radeon_drm_buffer_unmap_internal, - radeon_drm_buffer_validate, - radeon_drm_buffer_fence, - radeon_drm_buffer_get_base_buffer, -}; - -static struct pb_buffer * -radeon_drm_bufmgr_create_buffer_from_handle_unsafe(struct pb_manager *_mgr, - uint32_t handle) -{ - struct radeon_drm_bufmgr *mgr = radeon_drm_bufmgr(_mgr); - struct radeon_drm_winsys *rws = mgr->rws; - struct radeon_drm_buffer *buf; - struct radeon_bo *bo; - - buf = util_hash_table_get(mgr->buffer_handles, (void*)(uintptr_t)handle); - - if (buf) { - struct pb_buffer *b = NULL; - pb_reference(&b, &buf->base); - return b; - } - - bo = radeon_bo_open(rws->bom, handle, 0, - 0, 0, 0); - if (bo == NULL) - return NULL; - - buf = CALLOC_STRUCT(radeon_drm_buffer); - if (!buf) { - radeon_bo_unref(bo); - return NULL; - } - - make_empty_list(buf); - - pipe_reference_init(&buf->base.base.reference, 1); - buf->base.base.alignment = 0; - buf->base.base.usage = PB_USAGE_GPU_WRITE | PB_USAGE_GPU_READ; - buf->base.base.size = bo->size; - buf->base.vtbl = &radeon_drm_buffer_vtbl; - buf->mgr = mgr; - - buf->bo = bo; - - util_hash_table_set(mgr->buffer_handles, (void*)(uintptr_t)handle, buf); - - return &buf->base; -} - -struct pb_buffer * -radeon_drm_bufmgr_create_buffer_from_handle(struct pb_manager *_mgr, - uint32_t handle) -{ - struct radeon_drm_bufmgr *mgr = radeon_drm_bufmgr(_mgr); - struct pb_buffer *pb; - - pipe_mutex_lock(mgr->buffer_handles_mutex); - pb = radeon_drm_bufmgr_create_buffer_from_handle_unsafe(_mgr, handle); - pipe_mutex_unlock(mgr->buffer_handles_mutex); - - return pb; -} - -static struct pb_buffer * -radeon_drm_bufmgr_create_buffer(struct pb_manager *_mgr, - pb_size size, - const struct pb_desc *desc) -{ - struct radeon_drm_bufmgr *mgr = radeon_drm_bufmgr(_mgr); - struct radeon_drm_winsys *rws = mgr->rws; - struct radeon_drm_buffer *buf; - uint32_t domain; - - buf = CALLOC_STRUCT(radeon_drm_buffer); - if (!buf) - goto error1; - - pipe_reference_init(&buf->base.base.reference, 1); - buf->base.base.alignment = desc->alignment; - buf->base.base.usage = desc->usage; - buf->base.base.size = size; - buf->base.vtbl = &radeon_drm_buffer_vtbl; - buf->mgr = mgr; - - make_empty_list(buf); - - domain = - (desc->usage & RADEON_PB_USAGE_DOMAIN_GTT ? RADEON_GEM_DOMAIN_GTT : 0) | - (desc->usage & RADEON_PB_USAGE_DOMAIN_VRAM ? RADEON_GEM_DOMAIN_VRAM : 0); - - buf->bo = radeon_bo_open(rws->bom, 0, size, - desc->alignment, domain, 0); - if (buf->bo == NULL) - goto error2; - - return &buf->base; - - error2: - FREE(buf); - error1: - return NULL; -} - -static void -radeon_drm_bufmgr_flush(struct pb_manager *mgr) -{ - /* NOP */ -} - -static void -radeon_drm_bufmgr_destroy(struct pb_manager *_mgr) -{ - struct radeon_drm_bufmgr *mgr = radeon_drm_bufmgr(_mgr); - util_hash_table_destroy(mgr->buffer_handles); - pipe_mutex_destroy(mgr->buffer_map_list_mutex); - pipe_mutex_destroy(mgr->buffer_handles_mutex); - FREE(mgr); -} - -static unsigned handle_hash(void *key) -{ - return (unsigned)key; -} - -static int handle_compare(void *key1, void *key2) -{ - return !((int)key1 == (int)key2); -} - -struct pb_manager * -radeon_drm_bufmgr_create(struct radeon_drm_winsys *rws) -{ - struct radeon_drm_bufmgr *mgr; - - mgr = CALLOC_STRUCT(radeon_drm_bufmgr); - if (!mgr) - return NULL; - - mgr->base.destroy = radeon_drm_bufmgr_destroy; - mgr->base.create_buffer = radeon_drm_bufmgr_create_buffer; - mgr->base.flush = radeon_drm_bufmgr_flush; - - mgr->rws = rws; - make_empty_list(&mgr->buffer_map_list); - mgr->buffer_handles = util_hash_table_create(handle_hash, handle_compare); - pipe_mutex_init(mgr->buffer_map_list_mutex); - pipe_mutex_init(mgr->buffer_handles_mutex); - return &mgr->base; -} - -static struct radeon_drm_buffer *get_drm_buffer(struct pb_buffer *_buf) -{ - struct radeon_drm_buffer *buf = NULL; - - if (_buf->vtbl == &radeon_drm_buffer_vtbl) { - buf = radeon_drm_buffer(_buf); - } else { - struct pb_buffer *base_buf; - pb_size offset; - pb_get_base_buffer(_buf, &base_buf, &offset); - - if (base_buf->vtbl == &radeon_drm_buffer_vtbl) - buf = radeon_drm_buffer(base_buf); - } - - return buf; -} - -static void *radeon_drm_buffer_map(struct r300_winsys_screen *ws, - struct r300_winsys_buffer *buf, - struct r300_winsys_cs *cs, - enum pipe_transfer_usage usage) -{ - struct pb_buffer *_buf = radeon_pb_buffer(buf); - - return pb_map(_buf, get_pb_usage_from_transfer_flags(usage), radeon_drm_cs(cs)); -} - -static void radeon_drm_buffer_unmap(struct r300_winsys_screen *ws, - struct r300_winsys_buffer *buf) -{ - struct pb_buffer *_buf = radeon_pb_buffer(buf); - - pb_unmap(_buf); -} - -boolean radeon_drm_bufmgr_get_handle(struct pb_buffer *_buf, - struct winsys_handle *whandle) -{ - struct drm_gem_flink flink; - struct radeon_drm_buffer *buf = get_drm_buffer(_buf); - - if (whandle->type == DRM_API_HANDLE_TYPE_SHARED) { - if (!buf->flinked) { - flink.handle = buf->bo->handle; - - if (ioctl(buf->mgr->rws->fd, DRM_IOCTL_GEM_FLINK, &flink)) { - return FALSE; - } - - buf->flinked = TRUE; - buf->flink = flink.name; - } - whandle->handle = buf->flink; - } else if (whandle->type == DRM_API_HANDLE_TYPE_KMS) { - whandle->handle = buf->bo->handle; - } - return TRUE; -} - -static void radeon_drm_buffer_get_tiling(struct r300_winsys_screen *ws, - struct r300_winsys_buffer *_buf, - enum r300_buffer_tiling *microtiled, - enum r300_buffer_tiling *macrotiled) -{ - struct radeon_drm_buffer *buf = get_drm_buffer(radeon_pb_buffer(_buf)); - uint32_t flags = 0, pitch; - - radeon_bo_get_tiling(buf->bo, &flags, &pitch); - - *microtiled = R300_BUFFER_LINEAR; - *macrotiled = R300_BUFFER_LINEAR; - if (flags & RADEON_BO_FLAGS_MICRO_TILE) - *microtiled = R300_BUFFER_TILED; - - if (flags & RADEON_BO_FLAGS_MACRO_TILE) - *macrotiled = R300_BUFFER_TILED; -} - -static void radeon_drm_buffer_set_tiling(struct r300_winsys_screen *ws, - struct r300_winsys_buffer *_buf, - enum r300_buffer_tiling microtiled, - enum r300_buffer_tiling macrotiled, - uint32_t pitch) -{ - struct radeon_drm_buffer *buf = get_drm_buffer(radeon_pb_buffer(_buf)); - uint32_t flags = 0; - if (microtiled == R300_BUFFER_TILED) - flags |= RADEON_BO_FLAGS_MICRO_TILE; -/* XXX Remove this ifdef when libdrm version 2.4.19 becomes mandatory. */ -#ifdef RADEON_BO_FLAGS_MICRO_TILE_SQUARE - else if (microtiled == R300_BUFFER_SQUARETILED) - flags |= RADEON_BO_FLAGS_MICRO_TILE_SQUARE; -#endif - if (macrotiled == R300_BUFFER_TILED) - flags |= RADEON_BO_FLAGS_MACRO_TILE; - - radeon_bo_set_tiling(buf->bo, flags, pitch); -} - -static void radeon_drm_bufmgr_add_buffer(struct r300_winsys_cs *rcs, - struct r300_winsys_cs_buffer *_buf, - enum r300_buffer_domain rd, - enum r300_buffer_domain wd) -{ - struct radeon_drm_cs *cs = radeon_drm_cs(rcs); - struct radeon_bo *bo = (struct radeon_bo*)_buf; - - radeon_cs_space_add_persistent_bo(cs->cs, bo, rd, wd); -} - -static void radeon_drm_bufmgr_write_reloc(struct r300_winsys_cs *rcs, - struct r300_winsys_cs_buffer *_buf, - enum r300_buffer_domain rd, - enum r300_buffer_domain wd) -{ - struct radeon_drm_cs *cs = radeon_drm_cs(rcs); - struct radeon_bo *bo = (struct radeon_bo*)_buf; - int retval; - - cs->cs->cdw = cs->base.cdw; - retval = radeon_cs_write_reloc(cs->cs, bo, rd, wd, 0); - cs->base.cdw = cs->cs->cdw; - if (retval) { - fprintf(stderr, "radeon: Relocation of %p (%d, %d, %d) failed!\n", - bo, rd, wd, 0); - } -} - -static struct r300_winsys_cs_buffer *radeon_drm_get_cs_handle( - struct r300_winsys_screen *rws, - struct r300_winsys_buffer *_buf) -{ - /* return pure radeon_bo. */ - return (struct r300_winsys_cs_buffer*) - get_drm_buffer(radeon_pb_buffer(_buf))->bo; -} - -static boolean radeon_drm_is_buffer_referenced(struct r300_winsys_cs *rcs, - struct r300_winsys_cs_buffer *_buf, - enum r300_reference_domain domain) -{ - struct radeon_drm_cs *cs = radeon_drm_cs(rcs); - struct radeon_bo *bo = (struct radeon_bo*)_buf; - uint32_t tmp; - - if (domain & R300_REF_CS) { - if (radeon_bo_is_referenced_by_cs(bo, cs->cs)) { - return TRUE; - } - } - - if (domain & R300_REF_HW) { - if (radeon_bo_is_busy(bo, &tmp)) { - return TRUE; - } - } - - return FALSE; -} - -void radeon_drm_bufmgr_flush_maps(struct pb_manager *_mgr) -{ - struct radeon_drm_bufmgr *mgr = radeon_drm_bufmgr(_mgr); - struct radeon_drm_buffer *rpb, *t_rpb; - - pipe_mutex_lock(mgr->buffer_map_list_mutex); - - foreach_s(rpb, t_rpb, &mgr->buffer_map_list) { - radeon_bo_unmap(rpb->bo); - rpb->bo->ptr = NULL; - remove_from_list(rpb); - } - - make_empty_list(&mgr->buffer_map_list); - - pipe_mutex_unlock(mgr->buffer_map_list_mutex); -} - -static void radeon_drm_buffer_wait(struct r300_winsys_screen *ws, - struct r300_winsys_buffer *_buf) -{ - struct radeon_drm_buffer *buf = get_drm_buffer(radeon_pb_buffer(_buf)); - - radeon_bo_wait(buf->bo); -} - -void radeon_drm_bufmgr_init_functions(struct radeon_drm_winsys *ws) -{ - ws->base.buffer_get_cs_handle = radeon_drm_get_cs_handle; - ws->base.buffer_set_tiling = radeon_drm_buffer_set_tiling; - ws->base.buffer_get_tiling = radeon_drm_buffer_get_tiling; - ws->base.buffer_map = radeon_drm_buffer_map; - ws->base.buffer_unmap = radeon_drm_buffer_unmap; - ws->base.buffer_wait = radeon_drm_buffer_wait; - ws->base.cs_is_buffer_referenced = radeon_drm_is_buffer_referenced; - ws->base.cs_add_buffer = radeon_drm_bufmgr_add_buffer; - ws->base.cs_write_reloc = radeon_drm_bufmgr_write_reloc; -} diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_common.c b/src/gallium/winsys/radeon/drm/radeon_drm_common.c index 6bc6244115..72c2ff1112 100644 --- a/src/gallium/winsys/radeon/drm/radeon_drm_common.c +++ b/src/gallium/winsys/radeon/drm/radeon_drm_common.c @@ -1,5 +1,6 @@ /* * Copyright © 2009 Corbin Simpson + * Copyright © 2011 Marek Olšák <maraeo@gmail.com> * All Rights Reserved. * * Permission is hereby granted, free of charge, to any person obtaining @@ -27,39 +28,35 @@ * Authors: * Corbin Simpson <MostAwesomeDude@gmail.com> * Joakim Sindholt <opensource@zhasha.com> + * Marek Olšák <maraeo@gmail.com> */ #include "radeon_winsys.h" -#include "radeon_drm_buffer.h" +#include "radeon_drm_bo.h" +#include "radeon_drm_cs.h" #include "radeon_drm_public.h" #include "pipebuffer/pb_bufmgr.h" #include "util/u_memory.h" -#include "state_tracker/drm_driver.h" - -#include <radeon_drm.h> -#include <radeon_bo_gem.h> -#include <radeon_cs_gem.h> #include <xf86drm.h> #include <stdio.h> - -/* Enable/disable Hyper-Z access. Return TRUE on success. */ -static boolean radeon_set_hyperz_access(int fd, boolean enable) -{ #ifndef RADEON_INFO_WANT_HYPERZ #define RADEON_INFO_WANT_HYPERZ 7 #endif +#ifndef RADEON_INFO_WANT_CMASK +#define RADEON_INFO_WANT_CMASK 8 +#endif +/* Enable/disable feature access. Return TRUE on success. */ +static boolean radeon_set_fd_access(int fd, unsigned request, boolean enable) +{ struct drm_radeon_info info = {0}; unsigned value = enable ? 1 : 0; - if (!debug_get_bool_option("RADEON_HYPERZ", FALSE)) - return FALSE; - info.value = (unsigned long)&value; - info.request = RADEON_INFO_WANT_HYPERZ; + info.request = request; if (drmCommandWriteRead(fd, DRM_RADEON_INFO, &info, sizeof(info)) != 0) return FALSE; @@ -99,28 +96,21 @@ static void do_ioctls(struct radeon_drm_winsys *winsys) * we don't actually use the info for anything yet. */ version = drmGetVersion(winsys->fd); - if (version->version_major != 2) { + if (version->version_major != 2 || + version->version_minor < 3) { fprintf(stderr, "%s: DRM version is %d.%d.%d but this driver is " - "only compatible with 2.x.x\n", __FUNCTION__, - version->version_major, version->version_minor, + "only compatible with 2.3.x (kernel 2.6.34) and later.\n", + __FUNCTION__, + version->version_major, + version->version_minor, version->version_patchlevel); drmFreeVersion(version); exit(1); } -/* XXX Remove this ifdef when libdrm version 2.4.19 becomes mandatory. */ -#ifdef RADEON_BO_FLAGS_MICRO_TILE_SQUARE - // Supported since 2.1.0. - winsys->squaretiling = version->version_major > 2 || - version->version_minor >= 1; -#endif - - winsys->drm_2_3_0 = version->version_major > 2 || - version->version_minor >= 3; - - winsys->drm_2_6_0 = version->version_major > 2 || - (version->version_major == 2 && - version->version_minor >= 6); + winsys->drm_major = version->version_major; + winsys->drm_minor = version->version_minor; + winsys->drm_patchlevel = version->version_patchlevel; info.request = RADEON_INFO_DEVICE_ID; retval = drmCommandWriteRead(winsys->fd, DRM_RADEON_INFO, &info, sizeof(info)); @@ -149,7 +139,15 @@ static void do_ioctls(struct radeon_drm_winsys *winsys) } winsys->z_pipes = target; - winsys->hyperz = radeon_set_hyperz_access(winsys->fd, TRUE); + if (debug_get_bool_option("RADEON_HYPERZ", FALSE)) { + winsys->hyperz = radeon_set_fd_access(winsys->fd, + RADEON_INFO_WANT_HYPERZ, TRUE); + } + + if (debug_get_bool_option("RADEON_CMASK", FALSE)) { + winsys->aacompress = radeon_set_fd_access(winsys->fd, + RADEON_INFO_WANT_CMASK, TRUE); + } retval = drmCommandWriteRead(winsys->fd, DRM_RADEON_GEM_INFO, &gem_info, sizeof(gem_info)); @@ -161,18 +159,9 @@ static void do_ioctls(struct radeon_drm_winsys *winsys) winsys->gart_size = gem_info.gart_size; winsys->vram_size = gem_info.vram_size; - debug_printf("radeon: Successfully grabbed chipset info from kernel!\n" - "radeon: DRM version: %d.%d.%d ID: 0x%04x GB: %d Z: %d\n" - "radeon: GART size: %d MB VRAM size: %d MB\n" - "radeon: HyperZ: %s\n", - version->version_major, version->version_minor, - version->version_patchlevel, winsys->pci_id, - winsys->gb_pipes, winsys->z_pipes, - winsys->gart_size / 1024 / 1024, - winsys->vram_size / 1024 / 1024, - winsys->hyperz ? "YES" : "NO"); - drmFreeVersion(version); + + winsys->num_cpus = sysconf(_SC_NPROCESSORS_ONLN); } static void radeon_winsys_destroy(struct r300_winsys_screen *rws) @@ -181,12 +170,43 @@ static void radeon_winsys_destroy(struct r300_winsys_screen *rws) ws->cman->destroy(ws->cman); ws->kman->destroy(ws->kman); - - radeon_bo_manager_gem_dtor(ws->bom); - radeon_cs_manager_gem_dtor(ws->csm); FREE(rws); } +static uint32_t radeon_get_value(struct r300_winsys_screen *rws, + enum r300_value_id id) +{ + struct radeon_drm_winsys *ws = (struct radeon_drm_winsys *)rws; + + switch(id) { + case R300_VID_PCI_ID: + return ws->pci_id; + case R300_VID_GB_PIPES: + return ws->gb_pipes; + case R300_VID_Z_PIPES: + return ws->z_pipes; + case R300_VID_GART_SIZE: + return ws->gart_size; + case R300_VID_VRAM_SIZE: + return ws->vram_size; + case R300_VID_DRM_MAJOR: + return ws->drm_major; + case R300_VID_DRM_MINOR: + return ws->drm_minor; + case R300_VID_DRM_PATCHLEVEL: + return ws->drm_patchlevel; + case R300_VID_DRM_2_6_0: + return ws->drm_major*100 + ws->drm_minor >= 206; + case R300_VID_DRM_2_8_0: + return ws->drm_major*100 + ws->drm_minor >= 208; + case R300_CAN_HYPERZ: + return ws->hyperz; + case R300_CAN_AACOMPRESS: + return ws->aacompress; + } + return 0; +} + struct r300_winsys_screen *r300_drm_winsys_screen_create(int fd) { struct radeon_drm_winsys *ws = CALLOC_STRUCT(radeon_drm_winsys); @@ -202,13 +222,7 @@ struct r300_winsys_screen *r300_drm_winsys_screen_create(int fd) } /* Create managers. */ - ws->bom = radeon_bo_manager_gem_ctor(fd); - if (!ws->bom) - goto fail; - ws->csm = radeon_cs_manager_gem_ctor(fd); - if (!ws->csm) - goto fail; - ws->kman = radeon_drm_bufmgr_create(ws); + ws->kman = radeon_bomgr_create(ws); if (!ws->kman) goto fail; ws->cman = pb_cache_manager_create(ws->kman, 1000000); @@ -217,23 +231,18 @@ struct r300_winsys_screen *r300_drm_winsys_screen_create(int fd) /* Set functions. */ ws->base.destroy = radeon_winsys_destroy; + ws->base.get_value = radeon_get_value; - radeon_drm_bufmgr_init_functions(ws); - radeon_winsys_init_functions(ws); + radeon_bomgr_init_functions(ws); + radeon_drm_cs_init_functions(ws); return &ws->base; fail: - if (ws->bom) - radeon_bo_manager_gem_dtor(ws->bom); - if (ws->csm) - radeon_cs_manager_gem_dtor(ws->csm); - if (ws->cman) ws->cman->destroy(ws->cman); if (ws->kman) ws->kman->destroy(ws->kman); - FREE(ws); return NULL; } diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_cs.c b/src/gallium/winsys/radeon/drm/radeon_drm_cs.c new file mode 100644 index 0000000000..951791a172 --- /dev/null +++ b/src/gallium/winsys/radeon/drm/radeon_drm_cs.c @@ -0,0 +1,446 @@ +/* + * Copyright © 2008 Jérôme Glisse + * Copyright © 2010 Marek Olšák <maraeo@gmail.com> + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES + * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS + * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + */ +/* + * Authors: + * Marek Olšák <maraeo@gmail.com> + * + * Based on work from libdrm_radeon by: + * Aapo Tahkola <aet@rasterburn.org> + * Nicolai Haehnle <prefect_@gmx.net> + * Jérôme Glisse <glisse@freedesktop.org> + */ + +/* + This file replaces libdrm's radeon_cs_gem with our own implemention. + It's optimized specifically for r300g, but r600g could use it as well. + Reloc writes and space checking are faster and simpler than their + counterparts in libdrm (the time complexity of all the functions + is O(1) in nearly all scenarios, thanks to hashing). + + It works like this: + + cs_add_reloc(cs, buf, read_domain, write_domain) adds a new relocation and + also adds the size of 'buf' to the used_gart and used_vram winsys variables + based on the domains, which are simply or'd for the accounting purposes. + The adding is skipped if the reloc is already present in the list, but it + accounts any newly-referenced domains. + + cs_validate is then called, which just checks: + used_vram/gart < vram/gart_size * 0.8 + The 0.8 number allows for some memory fragmentation. If the validation + fails, the pipe driver flushes CS and tries do the validation again, + i.e. it validates only that one operation. If it fails again, it drops + the operation on the floor and prints some nasty message to stderr. + (done in the pipe driver) + + cs_write_reloc(cs, buf) just writes a reloc that has been added using + cs_add_reloc. The read_domain and write_domain parameters have been removed, + because we already specify them in cs_add_reloc. +*/ + +#include "radeon_drm_cs.h" + +#include "util/u_memory.h" + +#include <stdio.h> +#include <stdlib.h> +#include <stdint.h> +#include <xf86drm.h> + +#define RELOC_DWORDS (sizeof(struct drm_radeon_cs_reloc) / sizeof(uint32_t)) + +static boolean radeon_init_cs_context(struct radeon_cs_context *csc, int fd) +{ + csc->fd = fd; + csc->nrelocs = 512; + csc->relocs_bo = (struct radeon_bo**) + CALLOC(1, csc->nrelocs * sizeof(struct radeon_bo*)); + if (!csc->relocs_bo) { + return FALSE; + } + + csc->relocs = (struct drm_radeon_cs_reloc*) + CALLOC(1, csc->nrelocs * sizeof(struct drm_radeon_cs_reloc)); + if (!csc->relocs) { + FREE(csc->relocs_bo); + return FALSE; + } + + csc->chunks[0].chunk_id = RADEON_CHUNK_ID_IB; + csc->chunks[0].length_dw = 0; + csc->chunks[0].chunk_data = (uint64_t)(uintptr_t)csc->buf; + csc->chunks[1].chunk_id = RADEON_CHUNK_ID_RELOCS; + csc->chunks[1].length_dw = 0; + csc->chunks[1].chunk_data = (uint64_t)(uintptr_t)csc->relocs; + + csc->chunk_array[0] = (uint64_t)(uintptr_t)&csc->chunks[0]; + csc->chunk_array[1] = (uint64_t)(uintptr_t)&csc->chunks[1]; + + csc->cs.num_chunks = 2; + csc->cs.chunks = (uint64_t)(uintptr_t)csc->chunk_array; + return TRUE; +} + +static void radeon_cs_context_cleanup(struct radeon_cs_context *csc) +{ + unsigned i; + + for (i = 0; i < csc->crelocs; i++) { + p_atomic_dec(&csc->relocs_bo[i]->num_cs_references); + radeon_bo_unref(csc->relocs_bo[i]); + csc->relocs_bo[i] = NULL; + } + + csc->crelocs = 0; + csc->chunks[0].length_dw = 0; + csc->chunks[1].length_dw = 0; + csc->used_gart = 0; + csc->used_vram = 0; + memset(csc->is_handle_added, 0, sizeof(csc->is_handle_added)); +} + +static void radeon_destroy_cs_context(struct radeon_cs_context *csc) +{ + radeon_cs_context_cleanup(csc); + FREE(csc->relocs_bo); + FREE(csc->relocs); +} + +static struct r300_winsys_cs *radeon_drm_cs_create(struct r300_winsys_screen *rws) +{ + struct radeon_drm_winsys *ws = radeon_drm_winsys(rws); + struct radeon_drm_cs *cs; + + cs = CALLOC_STRUCT(radeon_drm_cs); + if (!cs) { + return NULL; + } + + cs->ws = ws; + + if (!radeon_init_cs_context(&cs->csc1, cs->ws->fd)) { + FREE(cs); + return NULL; + } + if (!radeon_init_cs_context(&cs->csc2, cs->ws->fd)) { + radeon_destroy_cs_context(&cs->csc1); + FREE(cs); + return NULL; + } + + /* Set the first command buffer as current. */ + cs->csc = &cs->csc1; + cs->cst = &cs->csc2; + cs->base.buf = cs->csc->buf; + + p_atomic_inc(&ws->num_cs); + return &cs->base; +} + +#define OUT_CS(cs, value) (cs)->buf[(cs)->cdw++] = (value) + +static INLINE void update_domains(struct drm_radeon_cs_reloc *reloc, + enum r300_buffer_domain rd, + enum r300_buffer_domain wd, + enum r300_buffer_domain *added_domains) +{ + *added_domains = (rd | wd) & ~(reloc->read_domains | reloc->write_domain); + + if (reloc->read_domains & wd) { + reloc->read_domains = rd; + reloc->write_domain = wd; + } else if (rd & reloc->write_domain) { + reloc->read_domains = rd; + reloc->write_domain |= wd; + } else { + reloc->read_domains |= rd; + reloc->write_domain |= wd; + } +} + +int radeon_get_reloc(struct radeon_cs_context *csc, struct radeon_bo *bo) +{ + struct drm_radeon_cs_reloc *reloc; + unsigned i; + unsigned hash = bo->handle & (sizeof(csc->is_handle_added)-1); + + if (csc->is_handle_added[hash]) { + reloc = csc->relocs_hashlist[hash]; + if (reloc->handle == bo->handle) { + return csc->reloc_indices_hashlist[hash]; + } + + /* Hash collision, look for the BO in the list of relocs linearly. */ + for (i = csc->crelocs; i != 0;) { + --i; + reloc = &csc->relocs[i]; + if (reloc->handle == bo->handle) { + /* Put this reloc in the hash list. + * This will prevent additional hash collisions if there are + * several subsequent get_reloc calls of the same buffer. + * + * Example: Assuming buffers A,B,C collide in the hash list, + * the following sequence of relocs: + * AAAAAAAAAAABBBBBBBBBBBBBBCCCCCCCC + * will collide here: ^ and here: ^, + * meaning that we should get very few collisions in the end. */ + csc->relocs_hashlist[hash] = reloc; + csc->reloc_indices_hashlist[hash] = i; + /*printf("write_reloc collision, hash: %i, handle: %i\n", hash, bo->handle);*/ + return i; + } + } + } + + return -1; +} + +static void radeon_add_reloc(struct radeon_cs_context *csc, + struct radeon_bo *bo, + enum r300_buffer_domain rd, + enum r300_buffer_domain wd, + enum r300_buffer_domain *added_domains) +{ + struct drm_radeon_cs_reloc *reloc; + unsigned i; + unsigned hash = bo->handle & (sizeof(csc->is_handle_added)-1); + + if (csc->is_handle_added[hash]) { + reloc = csc->relocs_hashlist[hash]; + if (reloc->handle == bo->handle) { + update_domains(reloc, rd, wd, added_domains); + return; + } + + /* Hash collision, look for the BO in the list of relocs linearly. */ + for (i = csc->crelocs; i != 0;) { + --i; + reloc = &csc->relocs[i]; + if (reloc->handle == bo->handle) { + update_domains(reloc, rd, wd, added_domains); + + csc->relocs_hashlist[hash] = reloc; + csc->reloc_indices_hashlist[hash] = i; + /*printf("write_reloc collision, hash: %i, handle: %i\n", hash, bo->handle);*/ + return; + } + } + } + + /* New relocation, check if the backing array is large enough. */ + if (csc->crelocs >= csc->nrelocs) { + uint32_t size; + csc->nrelocs += 10; + + size = csc->nrelocs * sizeof(struct radeon_bo*); + csc->relocs_bo = (struct radeon_bo**)realloc(csc->relocs_bo, size); + + size = csc->nrelocs * sizeof(struct drm_radeon_cs_reloc); + csc->relocs = (struct drm_radeon_cs_reloc*)realloc(csc->relocs, size); + + csc->chunks[1].chunk_data = (uint64_t)(uintptr_t)csc->relocs; + } + + /* Initialize the new relocation. */ + radeon_bo_ref(bo); + p_atomic_inc(&bo->num_cs_references); + csc->relocs_bo[csc->crelocs] = bo; + reloc = &csc->relocs[csc->crelocs]; + reloc->handle = bo->handle; + reloc->read_domains = rd; + reloc->write_domain = wd; + reloc->flags = 0; + + csc->is_handle_added[hash] = TRUE; + csc->relocs_hashlist[hash] = reloc; + csc->reloc_indices_hashlist[hash] = csc->crelocs; + + csc->chunks[1].length_dw += RELOC_DWORDS; + csc->crelocs++; + + *added_domains = rd | wd; +} + +static void radeon_drm_cs_add_reloc(struct r300_winsys_cs *rcs, + struct r300_winsys_cs_handle *buf, + enum r300_buffer_domain rd, + enum r300_buffer_domain wd) +{ + struct radeon_drm_cs *cs = radeon_drm_cs(rcs); + struct radeon_bo *bo = (struct radeon_bo*)buf; + enum r300_buffer_domain added_domains; + + radeon_add_reloc(cs->csc, bo, rd, wd, &added_domains); + + if (!added_domains) + return; + + if (added_domains & R300_DOMAIN_GTT) + cs->csc->used_gart += bo->size; + if (added_domains & R300_DOMAIN_VRAM) + cs->csc->used_vram += bo->size; +} + +static boolean radeon_drm_cs_validate(struct r300_winsys_cs *rcs) +{ + struct radeon_drm_cs *cs = radeon_drm_cs(rcs); + + return cs->csc->used_gart < cs->ws->gart_size * 0.8 && + cs->csc->used_vram < cs->ws->vram_size * 0.8; +} + +static void radeon_drm_cs_write_reloc(struct r300_winsys_cs *rcs, + struct r300_winsys_cs_handle *buf) +{ + struct radeon_drm_cs *cs = radeon_drm_cs(rcs); + struct radeon_bo *bo = (struct radeon_bo*)buf; + + unsigned index = radeon_get_reloc(cs->csc, bo); + + if (index == -1) { + fprintf(stderr, "r300: Cannot get a relocation in %s.\n", __func__); + return; + } + + OUT_CS(&cs->base, 0xc0001000); + OUT_CS(&cs->base, index * RELOC_DWORDS); +} + +static PIPE_THREAD_ROUTINE(radeon_drm_cs_emit_ioctl, param) +{ + struct radeon_cs_context *csc = (struct radeon_cs_context*)param; + unsigned i; + + if (drmCommandWriteRead(csc->fd, DRM_RADEON_CS, + &csc->cs, sizeof(struct drm_radeon_cs))) { + if (debug_get_bool_option("RADEON_DUMP_CS", FALSE)) { + unsigned i; + + fprintf(stderr, "radeon: The kernel rejected CS, dumping...\n"); + for (i = 0; i < csc->chunks[0].length_dw; i++) { + fprintf(stderr, "0x%08X\n", csc->buf[i]); + } + } else { + fprintf(stderr, "radeon: The kernel rejected CS, " + "see dmesg for more information.\n"); + } + } + + for (i = 0; i < csc->crelocs; i++) + p_atomic_dec(&csc->relocs_bo[i]->num_active_ioctls); + return NULL; +} + +void radeon_drm_cs_sync_flush(struct radeon_drm_cs *cs) +{ + /* Wait for any pending ioctl to complete. */ + if (cs->thread) { + pipe_thread_wait(cs->thread); + cs->thread = 0; + } +} + +DEBUG_GET_ONCE_BOOL_OPTION(thread, "RADEON_THREAD", TRUE) + +static void radeon_drm_cs_flush(struct r300_winsys_cs *rcs, unsigned flags) +{ + struct radeon_drm_cs *cs = radeon_drm_cs(rcs); + struct radeon_cs_context *tmp; + + radeon_drm_cs_sync_flush(cs); + + /* If the CS is not empty, emit it in a newly-spawned thread. */ + if (cs->base.cdw) { + unsigned i, crelocs = cs->csc->crelocs; + + cs->csc->chunks[0].length_dw = cs->base.cdw; + + for (i = 0; i < crelocs; i++) + p_atomic_inc(&cs->csc->relocs_bo[i]->num_active_ioctls); + + if (cs->ws->num_cpus > 1 && debug_get_option_thread() && + (flags & R300_FLUSH_ASYNC)) { + cs->thread = pipe_thread_create(radeon_drm_cs_emit_ioctl, cs->csc); + assert(cs->thread); + } else { + radeon_drm_cs_emit_ioctl(cs->csc); + } + } + + /* Flip command streams. */ + tmp = cs->csc; + cs->csc = cs->cst; + cs->cst = tmp; + + /* Prepare a new CS. */ + radeon_cs_context_cleanup(cs->csc); + + cs->base.buf = cs->csc->buf; + cs->base.cdw = 0; +} + +static void radeon_drm_cs_destroy(struct r300_winsys_cs *rcs) +{ + struct radeon_drm_cs *cs = radeon_drm_cs(rcs); + radeon_drm_cs_sync_flush(cs); + radeon_cs_context_cleanup(&cs->csc1); + radeon_cs_context_cleanup(&cs->csc2); + p_atomic_dec(&cs->ws->num_cs); + radeon_destroy_cs_context(&cs->csc1); + radeon_destroy_cs_context(&cs->csc2); + FREE(cs); +} + +static void radeon_drm_cs_set_flush(struct r300_winsys_cs *rcs, + void (*flush)(void *ctx, unsigned flags), + void *user) +{ + struct radeon_drm_cs *cs = radeon_drm_cs(rcs); + cs->flush_cs = flush; + cs->flush_data = user; +} + +static boolean radeon_bo_is_referenced(struct r300_winsys_cs *rcs, + struct r300_winsys_cs_handle *_buf) +{ + struct radeon_drm_cs *cs = radeon_drm_cs(rcs); + struct radeon_bo *bo = (struct radeon_bo*)_buf; + + return radeon_bo_is_referenced_by_cs(cs, bo); +} + +void radeon_drm_cs_init_functions(struct radeon_drm_winsys *ws) +{ + ws->base.cs_create = radeon_drm_cs_create; + ws->base.cs_destroy = radeon_drm_cs_destroy; + ws->base.cs_add_reloc = radeon_drm_cs_add_reloc; + ws->base.cs_validate = radeon_drm_cs_validate; + ws->base.cs_write_reloc = radeon_drm_cs_write_reloc; + ws->base.cs_flush = radeon_drm_cs_flush; + ws->base.cs_set_flush = radeon_drm_cs_set_flush; + ws->base.cs_is_buffer_referenced = radeon_bo_is_referenced; +} diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_cs.h b/src/gallium/winsys/radeon/drm/radeon_drm_cs.h new file mode 100644 index 0000000000..dfaa161c31 --- /dev/null +++ b/src/gallium/winsys/radeon/drm/radeon_drm_cs.h @@ -0,0 +1,102 @@ +/* + * Copyright © 2011 Marek Olšák <maraeo@gmail.com> + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES + * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS + * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + */ + +#ifndef RADEON_DRM_CS_H +#define RADEON_DRM_CS_H + +#include "radeon_drm_bo.h" +#include <radeon_drm.h> + +struct radeon_cs_context { + uint32_t buf[R300_MAX_CMDBUF_DWORDS]; + + int fd; + struct drm_radeon_cs cs; + struct drm_radeon_cs_chunk chunks[2]; + uint64_t chunk_array[2]; + + /* Relocs. */ + unsigned nrelocs; + unsigned crelocs; + struct radeon_bo **relocs_bo; + struct drm_radeon_cs_reloc *relocs; + + /* 0 = BO not added, 1 = BO added */ + char is_handle_added[256]; + struct drm_radeon_cs_reloc *relocs_hashlist[256]; + unsigned reloc_indices_hashlist[256]; + + unsigned used_vram; + unsigned used_gart; +}; + +struct radeon_drm_cs { + struct r300_winsys_cs base; + + /* We flip between these two CS. While one is being consumed + * by the kernel in another thread, the other one is being filled + * by the pipe driver. */ + struct radeon_cs_context csc1; + struct radeon_cs_context csc2; + /* The currently-used CS. */ + struct radeon_cs_context *csc; + /* The CS being currently-owned by the other thread. */ + struct radeon_cs_context *cst; + + /* The winsys. */ + struct radeon_drm_winsys *ws; + + /* Flush CS. */ + void (*flush_cs)(void *ctx, unsigned flags); + void *flush_data; + + pipe_thread thread; +}; + +int radeon_get_reloc(struct radeon_cs_context *csc, struct radeon_bo *bo); + +static INLINE struct radeon_drm_cs * +radeon_drm_cs(struct r300_winsys_cs *base) +{ + return (struct radeon_drm_cs*)base; +} + +static INLINE boolean radeon_bo_is_referenced_by_cs(struct radeon_drm_cs *cs, + struct radeon_bo *bo) +{ + return bo->num_cs_references == bo->rws->num_cs || + (bo->num_cs_references && radeon_get_reloc(cs->csc, bo) != -1); +} + +static INLINE boolean radeon_bo_is_referenced_by_any_cs(struct radeon_bo *bo) +{ + return bo->num_cs_references; +} + +void radeon_drm_cs_sync_flush(struct radeon_drm_cs *cs); +void radeon_drm_cs_init_functions(struct radeon_drm_winsys *ws); + +#endif diff --git a/src/gallium/winsys/radeon/drm/radeon_r300.c b/src/gallium/winsys/radeon/drm/radeon_r300.c deleted file mode 100644 index 9f59b3de46..0000000000 --- a/src/gallium/winsys/radeon/drm/radeon_r300.c +++ /dev/null @@ -1,266 +0,0 @@ -/* - * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com> - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * on the rights to use, copy, modify, merge, publish, distribute, sub - * license, and/or sell copies of the Software, and to permit persons to whom - * the Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice (including the next - * paragraph) shall be included in all copies or substantial portions of the - * Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL - * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, - * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR - * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE - * USE OR OTHER DEALINGS IN THE SOFTWARE. */ - -#include "radeon_drm_buffer.h" - -#include "util/u_memory.h" -#include "pipebuffer/pb_bufmgr.h" - -#include "radeon_cs_gem.h" -#include "state_tracker/drm_driver.h" - -static unsigned get_pb_usage_from_create_flags(unsigned bind, unsigned usage, - enum r300_buffer_domain domain) -{ - unsigned res = 0; - - if (bind & (PIPE_BIND_DEPTH_STENCIL | PIPE_BIND_RENDER_TARGET | - PIPE_BIND_DISPLAY_TARGET | PIPE_BIND_SCANOUT)) - res |= PB_USAGE_GPU_WRITE; - - if (bind & PIPE_BIND_SAMPLER_VIEW) - res |= PB_USAGE_GPU_READ | PB_USAGE_GPU_WRITE; - - if (bind & (PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER)) - res |= PB_USAGE_GPU_READ; - - if (bind & PIPE_BIND_TRANSFER_WRITE) - res |= PB_USAGE_CPU_WRITE; - - if (bind & PIPE_BIND_TRANSFER_READ) - res |= PB_USAGE_CPU_READ; - - /* Is usage of any use for us? Probably not. */ - - /* Now add driver-specific usage flags. */ - if (bind & (PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER)) - res |= RADEON_PB_USAGE_VERTEX; - - if (domain & R300_DOMAIN_GTT) - res |= RADEON_PB_USAGE_DOMAIN_GTT; - - if (domain & R300_DOMAIN_VRAM) - res |= RADEON_PB_USAGE_DOMAIN_VRAM; - - return res; -} - -static struct r300_winsys_buffer * -radeon_r300_winsys_buffer_create(struct r300_winsys_screen *rws, - unsigned size, - unsigned alignment, - unsigned bind, - unsigned usage, - enum r300_buffer_domain domain) -{ - struct radeon_drm_winsys *ws = radeon_drm_winsys(rws); - struct pb_desc desc; - struct pb_manager *provider; - struct pb_buffer *buffer; - - memset(&desc, 0, sizeof(desc)); - desc.alignment = alignment; - desc.usage = get_pb_usage_from_create_flags(bind, usage, domain); - - /* Assign a buffer manager. */ - if (bind & (PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER)) - provider = ws->cman; - else - provider = ws->kman; - - buffer = provider->create_buffer(provider, size, &desc); - if (!buffer) - return NULL; - - return (struct r300_winsys_buffer*)buffer; -} - -static void radeon_r300_winsys_buffer_reference(struct r300_winsys_screen *rws, - struct r300_winsys_buffer **pdst, - struct r300_winsys_buffer *src) -{ - struct pb_buffer *_src = radeon_pb_buffer(src); - struct pb_buffer *_dst = radeon_pb_buffer(*pdst); - - pb_reference(&_dst, _src); - - *pdst = (struct r300_winsys_buffer*)_dst; -} - -static struct r300_winsys_buffer *radeon_r300_winsys_buffer_from_handle(struct r300_winsys_screen *rws, - struct winsys_handle *whandle, - unsigned *stride, - unsigned *size) -{ - struct radeon_drm_winsys *ws = radeon_drm_winsys(rws); - struct pb_buffer *_buf; - - _buf = radeon_drm_bufmgr_create_buffer_from_handle(ws->kman, whandle->handle); - - if (stride) - *stride = whandle->stride; - if (size) - *size = _buf->base.size; - - return (struct r300_winsys_buffer*)_buf; -} - -static boolean radeon_r300_winsys_buffer_get_handle(struct r300_winsys_screen *rws, - struct r300_winsys_buffer *buffer, - unsigned stride, - struct winsys_handle *whandle) -{ - struct pb_buffer *_buf = radeon_pb_buffer(buffer); - whandle->stride = stride; - return radeon_drm_bufmgr_get_handle(_buf, whandle); -} - -static void radeon_r300_winsys_cs_set_flush(struct r300_winsys_cs *rcs, - void (*flush)(void *), - void *user) -{ - struct radeon_drm_cs *cs = radeon_drm_cs(rcs); - cs->flush_cs = flush; - cs->flush_data = user; - radeon_cs_space_set_flush(cs->cs, flush, user); -} - -static boolean radeon_r300_winsys_cs_validate(struct r300_winsys_cs *rcs) -{ - struct radeon_drm_cs *cs = radeon_drm_cs(rcs); - - return radeon_cs_space_check(cs->cs) >= 0; -} - -static void radeon_r300_winsys_cs_reset_buffers(struct r300_winsys_cs *rcs) -{ - struct radeon_drm_cs *cs = radeon_drm_cs(rcs); - radeon_cs_space_reset_bos(cs->cs); -} - -static void radeon_r300_winsys_cs_flush(struct r300_winsys_cs *rcs) -{ - struct radeon_drm_cs *cs = radeon_drm_cs(rcs); - int retval; - - /* Don't flush a zero-sized CS. */ - if (!cs->base.cdw) { - return; - } - - cs->cs->cdw = cs->base.cdw; - - radeon_drm_bufmgr_flush_maps(cs->ws->kman); - - /* Emit the CS. */ - retval = radeon_cs_emit(cs->cs); - if (retval) { - if (debug_get_bool_option("RADEON_DUMP_CS", FALSE)) { - fprintf(stderr, "radeon: The kernel rejected CS, dumping...\n"); - radeon_cs_print(cs->cs, stderr); - } else { - fprintf(stderr, "radeon: The kernel rejected CS, " - "see dmesg for more information.\n"); - } - } - - /* Reset CS. - * Someday, when we care about performance, we should really find a way - * to rotate between two or three CS objects so that the GPU can be - * spinning through one CS while another one is being filled. */ - radeon_cs_erase(cs->cs); - - cs->base.buf = cs->cs->packets; - cs->base.cdw = cs->cs->cdw; -} - -static uint32_t radeon_get_value(struct r300_winsys_screen *rws, - enum r300_value_id id) -{ - struct radeon_drm_winsys *ws = (struct radeon_drm_winsys *)rws; - - switch(id) { - case R300_VID_PCI_ID: - return ws->pci_id; - case R300_VID_GB_PIPES: - return ws->gb_pipes; - case R300_VID_Z_PIPES: - return ws->z_pipes; - case R300_VID_SQUARE_TILING_SUPPORT: - return ws->squaretiling; - case R300_VID_DRM_2_3_0: - return ws->drm_2_3_0; - case R300_VID_DRM_2_6_0: - return ws->drm_2_6_0; - case R300_CAN_HYPERZ: - return ws->hyperz; - } - return 0; -} - -static struct r300_winsys_cs *radeon_r300_winsys_cs_create(struct r300_winsys_screen *rws) -{ - struct radeon_drm_winsys *ws = radeon_drm_winsys(rws); - struct radeon_drm_cs *cs = CALLOC_STRUCT(radeon_drm_cs); - - if (!cs) - return NULL; - - /* Size limit on IBs is 64 kibibytes. */ - cs->cs = radeon_cs_create(ws->csm, 1024 * 64 / 4); - if (!cs->cs) { - FREE(cs); - return NULL; - } - - radeon_cs_set_limit(cs->cs, - RADEON_GEM_DOMAIN_GTT, ws->gart_size); - radeon_cs_set_limit(cs->cs, - RADEON_GEM_DOMAIN_VRAM, ws->vram_size); - - cs->ws = ws; - cs->base.buf = cs->cs->packets; - cs->base.cdw = cs->cs->cdw; - return &cs->base; -} - -static void radeon_r300_winsys_cs_destroy(struct r300_winsys_cs *rcs) -{ - struct radeon_drm_cs *cs = radeon_drm_cs(rcs); - radeon_cs_destroy(cs->cs); - FREE(cs); -} - -void radeon_winsys_init_functions(struct radeon_drm_winsys *ws) -{ - ws->base.get_value = radeon_get_value; - ws->base.buffer_create = radeon_r300_winsys_buffer_create; - ws->base.buffer_reference = radeon_r300_winsys_buffer_reference; - ws->base.buffer_from_handle = radeon_r300_winsys_buffer_from_handle; - ws->base.buffer_get_handle = radeon_r300_winsys_buffer_get_handle; - ws->base.cs_create = radeon_r300_winsys_cs_create; - ws->base.cs_destroy = radeon_r300_winsys_cs_destroy; - ws->base.cs_validate = radeon_r300_winsys_cs_validate; - ws->base.cs_flush = radeon_r300_winsys_cs_flush; - ws->base.cs_reset_buffers = radeon_r300_winsys_cs_reset_buffers; - ws->base.cs_set_flush = radeon_r300_winsys_cs_set_flush; -} diff --git a/src/gallium/winsys/radeon/drm/radeon_winsys.h b/src/gallium/winsys/radeon/drm/radeon_winsys.h index 81da1a25e0..9ecbb07457 100644 --- a/src/gallium/winsys/radeon/drm/radeon_winsys.h +++ b/src/gallium/winsys/radeon/drm/radeon_winsys.h @@ -36,8 +36,8 @@ struct radeon_drm_winsys { struct r300_winsys_screen base; int fd; /* DRM file descriptor */ + int num_cs; /* The number of command streams created. */ - struct radeon_bo_manager *bom; /* Radeon BO manager. */ struct pb_manager *kman; struct pb_manager *cman; @@ -46,44 +46,22 @@ struct radeon_drm_winsys { uint32_t z_pipes; /* Z pipe count (rv530 only) */ uint32_t gart_size; /* GART size. */ uint32_t vram_size; /* VRAM size. */ - boolean squaretiling; /* Square tiling support. */ - /* DRM 2.3.0 (R500 VAP regs, MSPOS regs, fixed tex3D size checking) */ - boolean drm_2_3_0; - /* DRM 2.6.0 (Hyper-Z, GB_Z_PEQ_CONFIG allowed on rv350->r4xx) */ - boolean drm_2_6_0; - /* Hyper-Z user */ - boolean hyperz; - - /* Radeon CS manager. */ - struct radeon_cs_manager *csm; -}; - -struct radeon_drm_cs { - struct r300_winsys_cs base; - - /* The winsys. */ - struct radeon_drm_winsys *ws; + uint32_t num_cpus; /* Number of CPUs. */ - /* The libdrm command stream. */ - struct radeon_cs *cs; + unsigned drm_major; + unsigned drm_minor; + unsigned drm_patchlevel; - /* Flush CS. */ - void (*flush_cs)(void *); - void *flush_data; + /* Hyper-Z user */ + boolean hyperz; + /* AA compression (CMask) */ + boolean aacompress; }; -static INLINE struct radeon_drm_cs * -radeon_drm_cs(struct r300_winsys_cs *base) -{ - return (struct radeon_drm_cs*)base; -} - static INLINE struct radeon_drm_winsys * radeon_drm_winsys(struct r300_winsys_screen *base) { return (struct radeon_drm_winsys*)base; } -void radeon_winsys_init_functions(struct radeon_drm_winsys *ws); - #endif |