/************************************************************************** * * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, TX., USA * All Rights Reserved. * Copyright 2009 VMware, Inc., Palo Alto, CA., USA * 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 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 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. * **************************************************************************/ /* * Authors: Thomas Hellstrom */ #ifdef HAVE_CONFIG_H #include "config.h" #endif #include #include #include #include #include #include #include "wsbm_pool.h" #include "wsbm_fencemgr.h" #include "wsbm_manager.h" #include "wsbm_mm.h" #include "wsbm_priv.h" /* * Malloced memory must be aligned to 16 bytes, since that's what * the DMA bitblt requires. */ #define WSBM_USER_ALIGN_ADD 16 #define WSBM_USER_ALIGN_SYSMEM(_val) \ ((void *)(((unsigned long) (_val) + 15) & ~15)) struct _WsbmUserBuffer { struct _WsbmBufStorage buf; struct _WsbmKernelBuf kBuf; /* Protected by the pool mutex */ struct _WsbmListHead lru; struct _WsbmListHead delayed; /* Protected by the buffer mutex */ unsigned long size; unsigned long alignment; struct _WsbmCond event; uint32_t proposedPlacement; uint32_t newFenceType; void *map; void *sysmem; int unFenced; struct _WsbmFenceObject *fence; struct _WsbmMMNode *node; struct _WsbmAtomic writers; }; struct _WsbmUserPool { /* * Constant after initialization. */ struct _WsbmBufferPool pool; unsigned long agpOffset; unsigned long agpMap; unsigned long agpSize; unsigned long vramOffset; unsigned long vramMap; unsigned long vramSize; struct _WsbmMutex mutex; struct _WsbmListHead delayed; struct _WsbmListHead vramLRU; struct _WsbmListHead agpLRU; struct _WsbmMM vramMM; struct _WsbmMM agpMM; uint32_t(*fenceTypes) (uint64_t); }; static inline struct _WsbmUserPool * userPool(struct _WsbmUserBuffer *buf) { return containerOf(buf->buf.pool, struct _WsbmUserPool, pool); } static inline struct _WsbmUserBuffer * userBuf(struct _WsbmBufStorage *buf) { return containerOf(buf, struct _WsbmUserBuffer, buf); } static void waitIdleLocked(struct _WsbmBufStorage *buf, int lazy) { struct _WsbmUserBuffer *vBuf = userBuf(buf); while (vBuf->unFenced || vBuf->fence != NULL) { if (vBuf->unFenced) WSBM_COND_WAIT(&vBuf->event, &buf->mutex); if (vBuf->fence != NULL) { if (!wsbmFenceSignaled(vBuf->fence, vBuf->kBuf.fence_type_mask)) { struct _WsbmFenceObject *fence = wsbmFenceReference(vBuf->fence); WSBM_MUTEX_UNLOCK(&buf->mutex); (void)wsbmFenceFinish(fence, vBuf->kBuf.fence_type_mask, lazy); WSBM_MUTEX_LOCK(&buf->mutex); if (vBuf->fence == fence) wsbmFenceUnreference(&vBuf->fence); wsbmFenceUnreference(&fence); } else { wsbmFenceUnreference(&vBuf->fence); } } } } static int pool_waitIdle(struct _WsbmBufStorage *buf, int lazy) { WSBM_MUTEX_UNLOCK(&buf->mutex); waitIdleLocked(buf, lazy); WSBM_MUTEX_UNLOCK(&buf->mutex); return 0; } static int evict_lru(struct _WsbmListHead *lru) { struct _WsbmUserBuffer *vBuf; struct _WsbmUserPool *p; struct _WsbmListHead *list = lru->next; int err; if (list == lru) { return -ENOMEM; } vBuf = WSBMLISTENTRY(list, struct _WsbmUserBuffer, lru); p = userPool(vBuf); WSBM_MUTEX_UNLOCK(&p->mutex); WSBM_MUTEX_LOCK(&vBuf->buf.mutex); WSBM_MUTEX_LOCK(&p->mutex); vBuf->sysmem = malloc(vBuf->size + WSBM_USER_ALIGN_ADD); if (!vBuf->sysmem) { err = -ENOMEM; goto out_unlock; } (void)wsbmFenceFinish(vBuf->fence, vBuf->kBuf.fence_type_mask, 0); wsbmFenceUnreference(&vBuf->fence); memcpy(WSBM_USER_ALIGN_SYSMEM(vBuf->sysmem), vBuf->map, vBuf->size); WSBMLISTDELINIT(&vBuf->lru); vBuf->kBuf.placement = WSBM_PL_FLAG_SYSTEM; vBuf->map = WSBM_USER_ALIGN_SYSMEM(vBuf->sysmem); /* * FIXME: Free memory. */ err = 0; out_unlock: WSBM_MUTEX_UNLOCK(&vBuf->buf.mutex); return err; } static struct _WsbmBufStorage * pool_create(struct _WsbmBufferPool *pool, unsigned long size, uint32_t placement, unsigned alignment) { struct _WsbmUserPool *p = containerOf(pool, struct _WsbmUserPool, pool); struct _WsbmUserBuffer *vBuf = calloc(1, sizeof(*vBuf)); if (!vBuf) return NULL; wsbmBufStorageInit(&vBuf->buf, pool); vBuf->sysmem = NULL; vBuf->proposedPlacement = placement; vBuf->size = size; vBuf->alignment = alignment; WSBMINITLISTHEAD(&vBuf->lru); WSBMINITLISTHEAD(&vBuf->delayed); WSBM_MUTEX_LOCK(&p->mutex); if (placement & WSBM_PL_FLAG_TT) { vBuf->node = wsbmMMSearchFree(&p->agpMM, size, alignment, 1); if (vBuf->node) vBuf->node = wsbmMMGetBlock(vBuf->node, size, alignment); if (vBuf->node) { vBuf->kBuf.placement = WSBM_PL_FLAG_TT; vBuf->kBuf.gpuOffset = p->agpOffset + vBuf->node->start; vBuf->map = (void *)(p->agpMap + vBuf->node->start); WSBMLISTADDTAIL(&vBuf->lru, &p->agpLRU); goto have_mem; } } if (placement & WSBM_PL_FLAG_VRAM) { vBuf->node = wsbmMMSearchFree(&p->vramMM, size, alignment, 1); if (vBuf->node) vBuf->node = wsbmMMGetBlock(vBuf->node, size, alignment); if (vBuf->node) { vBuf->kBuf.placement = WSBM_PL_FLAG_VRAM; vBuf->kBuf.gpuOffset = p->vramOffset + vBuf->node->start; vBuf->map = (void *)(p->vramMap + vBuf->node->start); WSBMLISTADDTAIL(&vBuf->lru, &p->vramLRU); goto have_mem; } } if ((placement & WSBM_PL_FLAG_NO_EVICT) && !(placement & WSBM_PL_FLAG_SYSTEM)) { WSBM_MUTEX_UNLOCK(&p->mutex); goto out_err; } vBuf->sysmem = malloc(size + WSBM_USER_ALIGN_ADD); vBuf->kBuf.placement = WSBM_PL_FLAG_SYSTEM; vBuf->map = WSBM_USER_ALIGN_SYSMEM(vBuf->sysmem); have_mem: WSBM_MUTEX_UNLOCK(&p->mutex); if (vBuf->sysmem != NULL || (!(vBuf->kBuf.placement & WSBM_PL_FLAG_SYSTEM))) return &vBuf->buf; out_err: free(vBuf); return NULL; } static int pool_validate(struct _WsbmBufStorage *buf, uint64_t set_flags, uint64_t clr_flags) { struct _WsbmUserBuffer *vBuf = userBuf(buf); struct _WsbmUserPool *p = userPool(vBuf); int err = -ENOMEM; WSBM_MUTEX_LOCK(&buf->mutex); while (wsbmAtomicRead(&vBuf->writers) != 0) WSBM_COND_WAIT(&vBuf->event, &buf->mutex); vBuf->unFenced = 1; WSBM_MUTEX_LOCK(&p->mutex); WSBMLISTDELINIT(&vBuf->lru); vBuf->proposedPlacement = (vBuf->proposedPlacement | set_flags) & ~clr_flags; if ((vBuf->proposedPlacement & vBuf->kBuf.placement & WSBM_PL_MASK_MEM) == vBuf->kBuf.placement) { err = 0; goto have_mem; } /* * We're moving to another memory region, so evict first and we'll * do a sw copy to the other region. */ if (!(vBuf->kBuf.placement & WSBM_PL_FLAG_SYSTEM)) { struct _WsbmListHead tmpLRU; WSBMINITLISTHEAD(&tmpLRU); WSBMLISTADDTAIL(&tmpLRU, &vBuf->lru); err = evict_lru(&tmpLRU); if (err) goto have_mem; } if (vBuf->proposedPlacement & WSBM_PL_FLAG_TT) { do { vBuf->node = wsbmMMSearchFree(&p->agpMM, vBuf->size, vBuf->alignment, 1); if (vBuf->node) vBuf->node = wsbmMMGetBlock(vBuf->node, vBuf->size, vBuf->alignment); if (vBuf->node) { vBuf->kBuf.placement = WSBM_PL_FLAG_TT; vBuf->kBuf.gpuOffset = p->agpOffset + vBuf->node->start; vBuf->map = (void *)(p->agpMap + vBuf->node->start); memcpy(vBuf->map, WSBM_USER_ALIGN_SYSMEM(vBuf->sysmem), vBuf->size); free(vBuf->sysmem); goto have_mem; } } while (evict_lru(&p->agpLRU) == 0); } if (vBuf->proposedPlacement & WSBM_PL_FLAG_VRAM) { do { vBuf->node = wsbmMMSearchFree(&p->vramMM, vBuf->size, vBuf->alignment, 1); if (vBuf->node) vBuf->node = wsbmMMGetBlock(vBuf->node, vBuf->size, vBuf->alignment); if (!err && vBuf->node) { vBuf->kBuf.placement = WSBM_PL_FLAG_VRAM; vBuf->kBuf.gpuOffset = p->vramOffset + vBuf->node->start; vBuf->map = (void *)(p->vramMap + vBuf->node->start); memcpy(vBuf->map, WSBM_USER_ALIGN_SYSMEM(vBuf->sysmem), vBuf->size); free(vBuf->sysmem); goto have_mem; } } while (evict_lru(&p->vramLRU) == 0); } if (vBuf->proposedPlacement & WSBM_PL_FLAG_SYSTEM) goto have_mem; err = -ENOMEM; have_mem: vBuf->newFenceType = p->fenceTypes(set_flags); WSBM_MUTEX_UNLOCK(&p->mutex); WSBM_MUTEX_UNLOCK(&buf->mutex); return err; } static int pool_setStatus(struct _WsbmBufStorage *buf, uint32_t set_placement, uint32_t clr_placement) { struct _WsbmUserBuffer *vBuf = userBuf(buf); int ret; ret = pool_validate(buf, set_placement, clr_placement); vBuf->unFenced = 0; return ret; } void release_delayed_buffers(struct _WsbmUserPool *p) { struct _WsbmUserBuffer *vBuf; struct _WsbmListHead *list, *next; WSBM_MUTEX_LOCK(&p->mutex); /* * We don't need to take the buffer mutexes in this loop, since * the only other user is the evict_lru function, which has the * pool mutex held when accessing the buffer fence member. */ WSBMLISTFOREACHSAFE(list, next, &p->delayed) { vBuf = WSBMLISTENTRY(list, struct _WsbmUserBuffer, delayed); if (!vBuf->fence || wsbmFenceSignaled(vBuf->fence, vBuf->kBuf.fence_type_mask)) { if (vBuf->fence) wsbmFenceUnreference(&vBuf->fence); WSBMLISTDEL(&vBuf->delayed); WSBMLISTDEL(&vBuf->lru); if ((vBuf->kBuf.placement & WSBM_PL_FLAG_SYSTEM) == 0) wsbmMMPutBlock(vBuf->node); else free(vBuf->sysmem); free(vBuf); } else break; } WSBM_MUTEX_UNLOCK(&p->mutex); } static void pool_destroy(struct _WsbmBufStorage **buf) { struct _WsbmUserBuffer *vBuf = userBuf(*buf); struct _WsbmUserPool *p = userPool(vBuf); *buf = NULL; WSBM_MUTEX_LOCK(&vBuf->buf.mutex); if ((vBuf->fence && !wsbmFenceSignaled(vBuf->fence, vBuf->kBuf.fence_type_mask))) { WSBM_MUTEX_LOCK(&p->mutex); WSBMLISTADDTAIL(&vBuf->delayed, &p->delayed); WSBM_MUTEX_UNLOCK(&p->mutex); WSBM_MUTEX_UNLOCK(&vBuf->buf.mutex); return; } if (vBuf->fence) wsbmFenceUnreference(&vBuf->fence); WSBM_MUTEX_LOCK(&p->mutex); WSBMLISTDEL(&vBuf->lru); WSBM_MUTEX_UNLOCK(&p->mutex); if (!(vBuf->kBuf.placement & WSBM_PL_FLAG_SYSTEM)) wsbmMMPutBlock(vBuf->node); else free(vBuf->sysmem); free(vBuf); return; } static int pool_map(struct _WsbmBufStorage *buf, unsigned mode __attribute__ ((unused)), void **virtual) { struct _WsbmUserBuffer *vBuf = userBuf(buf); *virtual = vBuf->map; return 0; } static void pool_unmap(struct _WsbmBufStorage *buf __attribute__ ((unused))) { ; } static void pool_releaseFromCpu(struct _WsbmBufStorage *buf, unsigned mode __attribute__ ((unused))) { struct _WsbmUserBuffer *vBuf = userBuf(buf); if (wsbmAtomicDecZero(&vBuf->writers)) WSBM_COND_BROADCAST(&vBuf->event); } static int pool_syncForCpu(struct _WsbmBufStorage *buf, unsigned mode) { struct _WsbmUserBuffer *vBuf = userBuf(buf); int ret = 0; WSBM_MUTEX_LOCK(&buf->mutex); if ((mode & WSBM_SYNCCPU_DONT_BLOCK)) { if (vBuf->unFenced) { ret = -EBUSY; goto out_unlock; } ret = 0; if ((vBuf->fence == NULL) || wsbmFenceSignaled(vBuf->fence, vBuf->kBuf.fence_type_mask)) { wsbmFenceUnreference(&vBuf->fence); wsbmAtomicInc(&vBuf->writers); } else ret = -EBUSY; goto out_unlock; } waitIdleLocked(buf, 0); wsbmAtomicInc(&vBuf->writers); out_unlock: WSBM_MUTEX_UNLOCK(&buf->mutex); return ret; } static unsigned long pool_offset(struct _WsbmBufStorage *buf) { return userBuf(buf)->kBuf.gpuOffset; } static unsigned long pool_poolOffset(struct _WsbmBufStorage *buf __attribute__ ((unused))) { return 0UL; } static unsigned long pool_size(struct _WsbmBufStorage *buf) { return userBuf(buf)->size; } static void pool_fence(struct _WsbmBufStorage *buf, struct _WsbmFenceObject *fence) { struct _WsbmUserBuffer *vBuf = userBuf(buf); struct _WsbmUserPool *p = userPool(vBuf); WSBM_MUTEX_LOCK(&buf->mutex); if (vBuf->fence) wsbmFenceUnreference(&vBuf->fence); vBuf->fence = wsbmFenceReference(fence); vBuf->unFenced = 0; vBuf->kBuf.fence_type_mask = vBuf->newFenceType; WSBM_COND_BROADCAST(&vBuf->event); WSBM_MUTEX_LOCK(&p->mutex); if (vBuf->kBuf.placement & WSBM_PL_FLAG_VRAM) WSBMLISTADDTAIL(&vBuf->lru, &p->vramLRU); else if (vBuf->kBuf.placement & WSBM_PL_FLAG_TT) WSBMLISTADDTAIL(&vBuf->lru, &p->agpLRU); WSBM_MUTEX_UNLOCK(&p->mutex); WSBM_MUTEX_UNLOCK(&buf->mutex); } static void pool_unvalidate(struct _WsbmBufStorage *buf) { struct _WsbmUserBuffer *vBuf = userBuf(buf); struct _WsbmUserPool *p = userPool(vBuf); WSBM_MUTEX_LOCK(&buf->mutex); if (!vBuf->unFenced) goto out_unlock; vBuf->unFenced = 0; WSBM_COND_BROADCAST(&vBuf->event); WSBM_MUTEX_LOCK(&p->mutex); if (vBuf->kBuf.placement & WSBM_PL_FLAG_VRAM) WSBMLISTADDTAIL(&vBuf->lru, &p->vramLRU); else if (vBuf->kBuf.placement & WSBM_PL_FLAG_TT) WSBMLISTADDTAIL(&vBuf->lru, &p->agpLRU); WSBM_MUTEX_UNLOCK(&p->mutex); out_unlock: WSBM_MUTEX_UNLOCK(&buf->mutex); } static struct _WsbmKernelBuf * pool_kernel(struct _WsbmBufStorage *buf) { struct _WsbmUserBuffer *vBuf = userBuf(buf); return &vBuf->kBuf; } static void pool_takedown(struct _WsbmBufferPool *pool) { struct _WsbmUserPool *p = containerOf(pool, struct _WsbmUserPool, pool); int empty; do { release_delayed_buffers(p); WSBM_MUTEX_LOCK(&p->mutex); empty = (p->delayed.next == &p->delayed); WSBM_MUTEX_UNLOCK(&p->mutex); if (!empty) usleep(1000); } while (!empty); WSBM_MUTEX_LOCK(&p->mutex); while (evict_lru(&p->vramLRU) == 0) ; while (evict_lru(&p->agpLRU) == 0) ; WSBM_MUTEX_UNLOCK(&p->mutex); wsbmMMtakedown(&p->agpMM); wsbmMMtakedown(&p->vramMM); free(p); } void wsbmUserPoolClean(struct _WsbmBufferPool *pool, int cleanVram, int cleanAgp) { struct _WsbmUserPool *p = containerOf(pool, struct _WsbmUserPool, pool); WSBM_MUTEX_LOCK(&p->mutex); if (cleanVram) while (evict_lru(&p->vramLRU) == 0) ; if (cleanAgp) while (evict_lru(&p->agpLRU) == 0) ; WSBM_MUTEX_UNLOCK(&p->mutex); } struct _WsbmBufferPool * wsbmUserPoolInit(void *vramAddr, unsigned long vramStart, unsigned long vramSize, void *agpAddr, unsigned long agpStart, unsigned long agpSize, uint32_t(*fenceTypes) (uint64_t set_flags)) { struct _WsbmBufferPool *pool; struct _WsbmUserPool *uPool; int ret; uPool = calloc(1, sizeof(*uPool)); if (!uPool) goto out_err0; ret = WSBM_MUTEX_INIT(&uPool->mutex); if (ret) goto out_err0; ret = wsbmMMinit(&uPool->vramMM, 0, vramSize); if (ret) goto out_err1; ret = wsbmMMinit(&uPool->agpMM, 0, agpSize); if (ret) goto out_err2; WSBMINITLISTHEAD(&uPool->delayed); WSBMINITLISTHEAD(&uPool->vramLRU); WSBMINITLISTHEAD(&uPool->agpLRU); uPool->agpOffset = agpStart; uPool->agpMap = (unsigned long)agpAddr; uPool->vramOffset = vramStart; uPool->vramMap = (unsigned long)vramAddr; uPool->fenceTypes = fenceTypes; pool = &uPool->pool; pool->map = &pool_map; pool->unmap = &pool_unmap; pool->destroy = &pool_destroy; pool->offset = &pool_offset; pool->poolOffset = &pool_poolOffset; pool->size = &pool_size; pool->create = &pool_create; pool->fence = &pool_fence; pool->unvalidate = &pool_unvalidate; pool->kernel = &pool_kernel; pool->validate = &pool_validate; pool->waitIdle = &pool_waitIdle; pool->takeDown = &pool_takedown; pool->setStatus = &pool_setStatus; pool->syncforcpu = &pool_syncForCpu; pool->releasefromcpu = &pool_releaseFromCpu; return pool; out_err2: wsbmMMtakedown(&uPool->vramMM); out_err1: WSBM_MUTEX_FREE(&uPool->mutex); out_err0: free(uPool); return NULL; }