diff options
Diffstat (limited to 'drivers/gpu/drm/ttm')
| -rw-r--r-- | drivers/gpu/drm/ttm/Makefile | 14 | ||||
| -rw-r--r-- | drivers/gpu/drm/ttm/ttm_agp_backend.c | 151 | ||||
| -rw-r--r-- | drivers/gpu/drm/ttm/ttm_bo.c | 1941 | ||||
| -rw-r--r-- | drivers/gpu/drm/ttm/ttm_bo_manager.c | 157 | ||||
| -rw-r--r-- | drivers/gpu/drm/ttm/ttm_bo_util.c | 702 | ||||
| -rw-r--r-- | drivers/gpu/drm/ttm/ttm_bo_vm.c | 465 | ||||
| -rw-r--r-- | drivers/gpu/drm/ttm/ttm_execbuf_util.c | 242 | ||||
| -rw-r--r-- | drivers/gpu/drm/ttm/ttm_lock.c | 310 | ||||
| -rw-r--r-- | drivers/gpu/drm/ttm/ttm_memory.c | 601 | ||||
| -rw-r--r-- | drivers/gpu/drm/ttm/ttm_module.c | 102 | ||||
| -rw-r--r-- | drivers/gpu/drm/ttm/ttm_object.c | 454 | ||||
| -rw-r--r-- | drivers/gpu/drm/ttm/ttm_page_alloc.c | 919 | ||||
| -rw-r--r-- | drivers/gpu/drm/ttm/ttm_page_alloc_dma.c | 1131 | ||||
| -rw-r--r-- | drivers/gpu/drm/ttm/ttm_tt.c | 377 |
14 files changed, 7566 insertions, 0 deletions
diff --git a/drivers/gpu/drm/ttm/Makefile b/drivers/gpu/drm/ttm/Makefile new file mode 100644 index 0000000..2914e5c --- /dev/null +++ b/drivers/gpu/drm/ttm/Makefile @@ -0,0 +1,14 @@ +# +# Makefile for the drm device driver. This driver provides support for the + +ccflags-y := -Iinclude/drm +ttm-y := ttm_agp_backend.o ttm_memory.o ttm_tt.o ttm_bo.o \ + ttm_bo_util.o ttm_bo_vm.o ttm_module.o \ + ttm_object.o ttm_lock.o ttm_execbuf_util.o ttm_page_alloc.o \ + ttm_bo_manager.o + +ifeq ($(CONFIG_SWIOTLB),y) +ttm-y += ttm_page_alloc_dma.o +endif + +obj-$(CPTCFG_DRM_TTM) += ttm.o diff --git a/drivers/gpu/drm/ttm/ttm_agp_backend.c b/drivers/gpu/drm/ttm/ttm_agp_backend.c new file mode 100644 index 0000000..3302f99 --- /dev/null +++ b/drivers/gpu/drm/ttm/ttm_agp_backend.c @@ -0,0 +1,151 @@ +/************************************************************************** + * + * Copyright (c) 2006-2009 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 <thellstrom-at-vmware-dot-com> + * Keith Packard. + */ + +#define pr_fmt(fmt) "[TTM] " fmt + +#include <drm/ttm/ttm_module.h> +#include <drm/ttm/ttm_bo_driver.h> +#include <drm/ttm/ttm_page_alloc.h> +#ifdef TTM_HAS_AGP +#include <drm/ttm/ttm_placement.h> +#include <linux/agp_backend.h> +#include <linux/module.h> +#include <linux/slab.h> +#include <linux/io.h> +#include <asm/agp.h> + +struct ttm_agp_backend { + struct ttm_tt ttm; + struct agp_memory *mem; + struct agp_bridge_data *bridge; +}; + +static int ttm_agp_bind(struct ttm_tt *ttm, struct ttm_mem_reg *bo_mem) +{ + struct ttm_agp_backend *agp_be = container_of(ttm, struct ttm_agp_backend, ttm); + struct drm_mm_node *node = bo_mem->mm_node; + struct agp_memory *mem; + int ret, cached = (bo_mem->placement & TTM_PL_FLAG_CACHED); + unsigned i; + + mem = agp_allocate_memory(agp_be->bridge, ttm->num_pages, AGP_USER_MEMORY); + if (unlikely(mem == NULL)) + return -ENOMEM; + + mem->page_count = 0; + for (i = 0; i < ttm->num_pages; i++) { + struct page *page = ttm->pages[i]; + + if (!page) + page = ttm->dummy_read_page; + + mem->pages[mem->page_count++] = page; + } + agp_be->mem = mem; + + mem->is_flushed = 1; + mem->type = (cached) ? AGP_USER_CACHED_MEMORY : AGP_USER_MEMORY; + + ret = agp_bind_memory(mem, node->start); + if (ret) + pr_err("AGP Bind memory failed\n"); + + return ret; +} + +static int ttm_agp_unbind(struct ttm_tt *ttm) +{ + struct ttm_agp_backend *agp_be = container_of(ttm, struct ttm_agp_backend, ttm); + + if (agp_be->mem) { + if (agp_be->mem->is_bound) + return agp_unbind_memory(agp_be->mem); + agp_free_memory(agp_be->mem); + agp_be->mem = NULL; + } + return 0; +} + +static void ttm_agp_destroy(struct ttm_tt *ttm) +{ + struct ttm_agp_backend *agp_be = container_of(ttm, struct ttm_agp_backend, ttm); + + if (agp_be->mem) + ttm_agp_unbind(ttm); + ttm_tt_fini(ttm); + kfree(agp_be); +} + +static struct ttm_backend_func ttm_agp_func = { + .bind = ttm_agp_bind, + .unbind = ttm_agp_unbind, + .destroy = ttm_agp_destroy, +}; + +struct ttm_tt *ttm_agp_tt_create(struct ttm_bo_device *bdev, + struct agp_bridge_data *bridge, + unsigned long size, uint32_t page_flags, + struct page *dummy_read_page) +{ + struct ttm_agp_backend *agp_be; + + agp_be = kmalloc(sizeof(*agp_be), GFP_KERNEL); + if (!agp_be) + return NULL; + + agp_be->mem = NULL; + agp_be->bridge = bridge; + agp_be->ttm.func = &ttm_agp_func; + + if (ttm_tt_init(&agp_be->ttm, bdev, size, page_flags, dummy_read_page)) { + return NULL; + } + + return &agp_be->ttm; +} +EXPORT_SYMBOL(ttm_agp_tt_create); + +int ttm_agp_tt_populate(struct ttm_tt *ttm) +{ + if (ttm->state != tt_unpopulated) + return 0; + + return ttm_pool_populate(ttm); +} +EXPORT_SYMBOL(ttm_agp_tt_populate); + +void ttm_agp_tt_unpopulate(struct ttm_tt *ttm) +{ + ttm_pool_unpopulate(ttm); +} +EXPORT_SYMBOL(ttm_agp_tt_unpopulate); + +#endif diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c new file mode 100644 index 0000000..9b07b7d --- /dev/null +++ b/drivers/gpu/drm/ttm/ttm_bo.c @@ -0,0 +1,1941 @@ +/************************************************************************** + * + * Copyright (c) 2006-2009 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 <thellstrom-at-vmware-dot-com> + */ + +#define pr_fmt(fmt) "[TTM] " fmt + +#include <drm/ttm/ttm_module.h> +#include <drm/ttm/ttm_bo_driver.h> +#include <drm/ttm/ttm_placement.h> +#include <linux/jiffies.h> +#include <linux/slab.h> +#include <linux/sched.h> +#include <linux/mm.h> +#include <linux/file.h> +#include <linux/module.h> +#include <linux/atomic.h> + +#define TTM_ASSERT_LOCKED(param) +#define TTM_DEBUG(fmt, arg...) +#define TTM_BO_HASH_ORDER 13 + +static int ttm_bo_setup_vm(struct ttm_buffer_object *bo); +static int ttm_bo_swapout(struct ttm_mem_shrink *shrink); +static void ttm_bo_global_kobj_release(struct kobject *kobj); + +static struct attribute ttm_bo_count = { + .name = "bo_count", + .mode = S_IRUGO +}; + +static inline int ttm_mem_type_from_flags(uint32_t flags, uint32_t *mem_type) +{ + int i; + + for (i = 0; i <= TTM_PL_PRIV5; i++) + if (flags & (1 << i)) { + *mem_type = i; + return 0; + } + return -EINVAL; +} + +static void ttm_mem_type_debug(struct ttm_bo_device *bdev, int mem_type) +{ + struct ttm_mem_type_manager *man = &bdev->man[mem_type]; + + pr_err(" has_type: %d\n", man->has_type); + pr_err(" use_type: %d\n", man->use_type); + pr_err(" flags: 0x%08X\n", man->flags); + pr_err(" gpu_offset: 0x%08lX\n", man->gpu_offset); + pr_err(" size: %llu\n", man->size); + pr_err(" available_caching: 0x%08X\n", man->available_caching); + pr_err(" default_caching: 0x%08X\n", man->default_caching); + if (mem_type != TTM_PL_SYSTEM) + (*man->func->debug)(man, TTM_PFX); +} + +static void ttm_bo_mem_space_debug(struct ttm_buffer_object *bo, + struct ttm_placement *placement) +{ + int i, ret, mem_type; + + pr_err("No space for %p (%lu pages, %luK, %luM)\n", + bo, bo->mem.num_pages, bo->mem.size >> 10, + bo->mem.size >> 20); + for (i = 0; i < placement->num_placement; i++) { + ret = ttm_mem_type_from_flags(placement->placement[i], + &mem_type); + if (ret) + return; + pr_err(" placement[%d]=0x%08X (%d)\n", + i, placement->placement[i], mem_type); + ttm_mem_type_debug(bo->bdev, mem_type); + } +} + +static ssize_t ttm_bo_global_show(struct kobject *kobj, + struct attribute *attr, + char *buffer) +{ + struct ttm_bo_global *glob = + container_of(kobj, struct ttm_bo_global, kobj); + + return snprintf(buffer, PAGE_SIZE, "%lu\n", + (unsigned long) atomic_read(&glob->bo_count)); +} + +static struct attribute *ttm_bo_global_attrs[] = { + &ttm_bo_count, + NULL +}; + +static const struct sysfs_ops ttm_bo_global_ops = { + .show = &ttm_bo_global_show +}; + +static struct kobj_type ttm_bo_glob_kobj_type = { + .release = &ttm_bo_global_kobj_release, + .sysfs_ops = &ttm_bo_global_ops, + .default_attrs = ttm_bo_global_attrs +}; + + +static inline uint32_t ttm_bo_type_flags(unsigned type) +{ + return 1 << (type); +} + +static void ttm_bo_release_list(struct kref *list_kref) +{ + struct ttm_buffer_object *bo = + container_of(list_kref, struct ttm_buffer_object, list_kref); + struct ttm_bo_device *bdev = bo->bdev; + size_t acc_size = bo->acc_size; + + BUG_ON(atomic_read(&bo->list_kref.refcount)); + BUG_ON(atomic_read(&bo->kref.refcount)); + BUG_ON(atomic_read(&bo->cpu_writers)); + BUG_ON(bo->sync_obj != NULL); + BUG_ON(bo->mem.mm_node != NULL); + BUG_ON(!list_empty(&bo->lru)); + BUG_ON(!list_empty(&bo->ddestroy)); + + if (bo->ttm) + ttm_tt_destroy(bo->ttm); + atomic_dec(&bo->glob->bo_count); + if (bo->destroy) + bo->destroy(bo); + else { + kfree(bo); + } + ttm_mem_global_free(bdev->glob->mem_glob, acc_size); +} + +static int ttm_bo_wait_unreserved(struct ttm_buffer_object *bo, + bool interruptible) +{ + if (interruptible) { + return wait_event_interruptible(bo->event_queue, + !ttm_bo_is_reserved(bo)); + } else { + wait_event(bo->event_queue, !ttm_bo_is_reserved(bo)); + return 0; + } +} + +void ttm_bo_add_to_lru(struct ttm_buffer_object *bo) +{ + struct ttm_bo_device *bdev = bo->bdev; + struct ttm_mem_type_manager *man; + + BUG_ON(!ttm_bo_is_reserved(bo)); + + if (!(bo->mem.placement & TTM_PL_FLAG_NO_EVICT)) { + + BUG_ON(!list_empty(&bo->lru)); + + man = &bdev->man[bo->mem.mem_type]; + list_add_tail(&bo->lru, &man->lru); + kref_get(&bo->list_kref); + + if (bo->ttm != NULL) { + list_add_tail(&bo->swap, &bo->glob->swap_lru); + kref_get(&bo->list_kref); + } + } +} + +int ttm_bo_del_from_lru(struct ttm_buffer_object *bo) +{ + int put_count = 0; + + if (!list_empty(&bo->swap)) { + list_del_init(&bo->swap); + ++put_count; + } + if (!list_empty(&bo->lru)) { + list_del_init(&bo->lru); + ++put_count; + } + + /* + * TODO: Add a driver hook to delete from + * driver-specific LRU's here. + */ + + return put_count; +} + +int ttm_bo_reserve_nolru(struct ttm_buffer_object *bo, + bool interruptible, + bool no_wait, bool use_sequence, uint32_t sequence) +{ + int ret; + + while (unlikely(atomic_xchg(&bo->reserved, 1) != 0)) { + /** + * Deadlock avoidance for multi-bo reserving. + */ + if (use_sequence && bo->seq_valid) { + /** + * We've already reserved this one. + */ + if (unlikely(sequence == bo->val_seq)) + return -EDEADLK; + /** + * Already reserved by a thread that will not back + * off for us. We need to back off. + */ + if (unlikely(sequence - bo->val_seq < (1 << 31))) + return -EAGAIN; + } + + if (no_wait) + return -EBUSY; + + ret = ttm_bo_wait_unreserved(bo, interruptible); + + if (unlikely(ret)) + return ret; + } + + if (use_sequence) { + bool wake_up = false; + /** + * Wake up waiters that may need to recheck for deadlock, + * if we decreased the sequence number. + */ + if (unlikely((bo->val_seq - sequence < (1 << 31)) + || !bo->seq_valid)) + wake_up = true; + + /* + * In the worst case with memory ordering these values can be + * seen in the wrong order. However since we call wake_up_all + * in that case, this will hopefully not pose a problem, + * and the worst case would only cause someone to accidentally + * hit -EAGAIN in ttm_bo_reserve when they see old value of + * val_seq. However this would only happen if seq_valid was + * written before val_seq was, and just means some slightly + * increased cpu usage + */ + bo->val_seq = sequence; + bo->seq_valid = true; + if (wake_up) + wake_up_all(&bo->event_queue); + } else { + bo->seq_valid = false; + } + + return 0; +} +EXPORT_SYMBOL(ttm_bo_reserve); + +static void ttm_bo_ref_bug(struct kref *list_kref) +{ + BUG(); +} + +void ttm_bo_list_ref_sub(struct ttm_buffer_object *bo, int count, + bool never_free) +{ + kref_sub(&bo->list_kref, count, + (never_free) ? ttm_bo_ref_bug : ttm_bo_release_list); +} + +int ttm_bo_reserve(struct ttm_buffer_object *bo, + bool interruptible, + bool no_wait, bool use_sequence, uint32_t sequence) +{ + struct ttm_bo_global *glob = bo->glob; + int put_count = 0; + int ret; + + ret = ttm_bo_reserve_nolru(bo, interruptible, no_wait, use_sequence, + sequence); + if (likely(ret == 0)) { + spin_lock(&glob->lru_lock); + put_count = ttm_bo_del_from_lru(bo); + spin_unlock(&glob->lru_lock); + ttm_bo_list_ref_sub(bo, put_count, true); + } + + return ret; +} + +int ttm_bo_reserve_slowpath_nolru(struct ttm_buffer_object *bo, + bool interruptible, uint32_t sequence) +{ + bool wake_up = false; + int ret; + + while (unlikely(atomic_xchg(&bo->reserved, 1) != 0)) { + WARN_ON(bo->seq_valid && sequence == bo->val_seq); + + ret = ttm_bo_wait_unreserved(bo, interruptible); + + if (unlikely(ret)) + return ret; + } + + if ((bo->val_seq - sequence < (1 << 31)) || !bo->seq_valid) + wake_up = true; + + /** + * Wake up waiters that may need to recheck for deadlock, + * if we decreased the sequence number. + */ + bo->val_seq = sequence; + bo->seq_valid = true; + if (wake_up) + wake_up_all(&bo->event_queue); + + return 0; +} + +int ttm_bo_reserve_slowpath(struct ttm_buffer_object *bo, + bool interruptible, uint32_t sequence) +{ + struct ttm_bo_global *glob = bo->glob; + int put_count, ret; + + ret = ttm_bo_reserve_slowpath_nolru(bo, interruptible, sequence); + if (likely(!ret)) { + spin_lock(&glob->lru_lock); + put_count = ttm_bo_del_from_lru(bo); + spin_unlock(&glob->lru_lock); + ttm_bo_list_ref_sub(bo, put_count, true); + } + return ret; +} +EXPORT_SYMBOL(ttm_bo_reserve_slowpath); + +void ttm_bo_unreserve_locked(struct ttm_buffer_object *bo) +{ + ttm_bo_add_to_lru(bo); + atomic_set(&bo->reserved, 0); + wake_up_all(&bo->event_queue); +} + +void ttm_bo_unreserve(struct ttm_buffer_object *bo) +{ + struct ttm_bo_global *glob = bo->glob; + + spin_lock(&glob->lru_lock); + ttm_bo_unreserve_locked(bo); + spin_unlock(&glob->lru_lock); +} +EXPORT_SYMBOL(ttm_bo_unreserve); + +/* + * Call bo->mutex locked. + */ +static int ttm_bo_add_ttm(struct ttm_buffer_object *bo, bool zero_alloc) +{ + struct ttm_bo_device *bdev = bo->bdev; + struct ttm_bo_global *glob = bo->glob; + int ret = 0; + uint32_t page_flags = 0; + + TTM_ASSERT_LOCKED(&bo->mutex); + bo->ttm = NULL; + + if (bdev->need_dma32) + page_flags |= TTM_PAGE_FLAG_DMA32; + + switch (bo->type) { + case ttm_bo_type_device: + if (zero_alloc) + page_flags |= TTM_PAGE_FLAG_ZERO_ALLOC; + case ttm_bo_type_kernel: + bo->ttm = bdev->driver->ttm_tt_create(bdev, bo->num_pages << PAGE_SHIFT, + page_flags, glob->dummy_read_page); + if (unlikely(bo->ttm == NULL)) + ret = -ENOMEM; + break; + case ttm_bo_type_sg: + bo->ttm = bdev->driver->ttm_tt_create(bdev, bo->num_pages << PAGE_SHIFT, + page_flags | TTM_PAGE_FLAG_SG, + glob->dummy_read_page); + if (unlikely(bo->ttm == NULL)) { + ret = -ENOMEM; + break; + } + bo->ttm->sg = bo->sg; + break; + default: + pr_err("Illegal buffer object type\n"); + ret = -EINVAL; + break; + } + + return ret; +} + +static int ttm_bo_handle_move_mem(struct ttm_buffer_object *bo, + struct ttm_mem_reg *mem, + bool evict, bool interruptible, + bool no_wait_gpu) +{ + struct ttm_bo_device *bdev = bo->bdev; + bool old_is_pci = ttm_mem_reg_is_pci(bdev, &bo->mem); + bool new_is_pci = ttm_mem_reg_is_pci(bdev, mem); + struct ttm_mem_type_manager *old_man = &bdev->man[bo->mem.mem_type]; + struct ttm_mem_type_manager *new_man = &bdev->man[mem->mem_type]; + int ret = 0; + + if (old_is_pci || new_is_pci || + ((mem->placement & bo->mem.placement & TTM_PL_MASK_CACHING) == 0)) { + ret = ttm_mem_io_lock(old_man, true); + if (unlikely(ret != 0)) + goto out_err; + ttm_bo_unmap_virtual_locked(bo); + ttm_mem_io_unlock(old_man); + } + + /* + * Create and bind a ttm if required. + */ + + if (!(new_man->flags & TTM_MEMTYPE_FLAG_FIXED)) { + if (bo->ttm == NULL) { + bool zero = !(old_man->flags & TTM_MEMTYPE_FLAG_FIXED); + ret = ttm_bo_add_ttm(bo, zero); + if (ret) + goto out_err; + } + + ret = ttm_tt_set_placement_caching(bo->ttm, mem->placement); + if (ret) + goto out_err; + + if (mem->mem_type != TTM_PL_SYSTEM) { + ret = ttm_tt_bind(bo->ttm, mem); + if (ret) + goto out_err; + } + + if (bo->mem.mem_type == TTM_PL_SYSTEM) { + if (bdev->driver->move_notify) + bdev->driver->move_notify(bo, mem); + bo->mem = *mem; + mem->mm_node = NULL; + goto moved; + } + } + + if (bdev->driver->move_notify) + bdev->driver->move_notify(bo, mem); + + if (!(old_man->flags & TTM_MEMTYPE_FLAG_FIXED) && + !(new_man->flags & TTM_MEMTYPE_FLAG_FIXED)) + ret = ttm_bo_move_ttm(bo, evict, no_wait_gpu, mem); + else if (bdev->driver->move) + ret = bdev->driver->move(bo, evict, interruptible, + no_wait_gpu, mem); + else + ret = ttm_bo_move_memcpy(bo, evict, no_wait_gpu, mem); + + if (ret) { + if (bdev->driver->move_notify) { + struct ttm_mem_reg tmp_mem = *mem; + *mem = bo->mem; + bo->mem = tmp_mem; + bdev->driver->move_notify(bo, mem); + bo->mem = *mem; + *mem = tmp_mem; + } + + goto out_err; + } + +moved: + if (bo->evicted) { + ret = bdev->driver->invalidate_caches(bdev, bo->mem.placement); + if (ret) + pr_err("Can not flush read caches\n"); + bo->evicted = false; + } + + if (bo->mem.mm_node) { + bo->offset = (bo->mem.start << PAGE_SHIFT) + + bdev->man[bo->mem.mem_type].gpu_offset; + bo->cur_placement = bo->mem.placement; + } else + bo->offset = 0; + + return 0; + +out_err: + new_man = &bdev->man[bo->mem.mem_type]; + if ((new_man->flags & TTM_MEMTYPE_FLAG_FIXED) && bo->ttm) { + ttm_tt_unbind(bo->ttm); + ttm_tt_destroy(bo->ttm); + bo->ttm = NULL; + } + + return ret; +} + +/** + * Call bo::reserved. + * Will release GPU memory type usage on destruction. + * This is the place to put in driver specific hooks to release + * driver private resources. + * Will release the bo::reserved lock. + */ + +static void ttm_bo_cleanup_memtype_use(struct ttm_buffer_object *bo) +{ + if (bo->bdev->driver->move_notify) + bo->bdev->driver->move_notify(bo, NULL); + + if (bo->ttm) { + ttm_tt_unbind(bo->ttm); + ttm_tt_destroy(bo->ttm); + bo->ttm = NULL; + } + ttm_bo_mem_put(bo, &bo->mem); + + atomic_set(&bo->reserved, 0); + wake_up_all(&bo->event_queue); + + /* + * Since the final reference to this bo may not be dropped by + * the current task we have to put a memory barrier here to make + * sure the changes done in this function are always visible. + * + * This function only needs protection against the final kref_put. + */ + smp_mb__before_atomic_dec(); +} + +static void ttm_bo_cleanup_refs_or_queue(struct ttm_buffer_object *bo) +{ + struct ttm_bo_device *bdev = bo->bdev; + struct ttm_bo_global *glob = bo->glob; + struct ttm_bo_driver *driver = bdev->driver; + void *sync_obj = NULL; + int put_count; + int ret; + + spin_lock(&glob->lru_lock); + ret = ttm_bo_reserve_nolru(bo, false, true, false, 0); + + spin_lock(&bdev->fence_lock); + (void) ttm_bo_wait(bo, false, false, true); + if (!ret && !bo->sync_obj) { + spin_unlock(&bdev->fence_lock); + put_count = ttm_bo_del_from_lru(bo); + + spin_unlock(&glob->lru_lock); + ttm_bo_cleanup_memtype_use(bo); + + ttm_bo_list_ref_sub(bo, put_count, true); + + return; + } + if (bo->sync_obj) + sync_obj = driver->sync_obj_ref(bo->sync_obj); + spin_unlock(&bdev->fence_lock); + + if (!ret) { + atomic_set(&bo->reserved, 0); + wake_up_all(&bo->event_queue); + } + + kref_get(&bo->list_kref); + list_add_tail(&bo->ddestroy, &bdev->ddestroy); + spin_unlock(&glob->lru_lock); + + if (sync_obj) { + driver->sync_obj_flush(sync_obj); + driver->sync_obj_unref(&sync_obj); + } + schedule_delayed_work(&bdev->wq, + ((HZ / 100) < 1) ? 1 : HZ / 100); +} + +/** + * function ttm_bo_cleanup_refs_and_unlock + * If bo idle, remove from delayed- and lru lists, and unref. + * If not idle, do nothing. + * + * Must be called with lru_lock and reservation held, this function + * will drop both before returning. + * + * @interruptible Any sleeps should occur interruptibly. + * @no_wait_gpu Never wait for gpu. Return -EBUSY instead. + */ + +static int ttm_bo_cleanup_refs_and_unlock(struct ttm_buffer_object *bo, + bool interruptible, + bool no_wait_gpu) +{ + struct ttm_bo_device *bdev = bo->bdev; + struct ttm_bo_driver *driver = bdev->driver; + struct ttm_bo_global *glob = bo->glob; + int put_count; + int ret; + + spin_lock(&bdev->fence_lock); + ret = ttm_bo_wait(bo, false, false, true); + + if (ret && !no_wait_gpu) { + void *sync_obj; + + /* + * Take a reference to the fence and unreserve, + * at this point the buffer should be dead, so + * no new sync objects can be attached. + */ + sync_obj = driver->sync_obj_ref(bo->sync_obj); + spin_unlock(&bdev->fence_lock); + + atomic_set(&bo->reserved, 0); + wake_up_all(&bo->event_queue); + spin_unlock(&glob->lru_lock); + + ret = driver->sync_obj_wait(sync_obj, false, interruptible); + driver->sync_obj_unref(&sync_obj); + if (ret) + return ret; + + /* + * remove sync_obj with ttm_bo_wait, the wait should be + * finished, and no new wait object should have been added. + */ + spin_lock(&bdev->fence_lock); + ret = ttm_bo_wait(bo, false, false, true); + WARN_ON(ret); + spin_unlock(&bdev->fence_lock); + if (ret) + return ret; + + spin_lock(&glob->lru_lock); + ret = ttm_bo_reserve_nolru(bo, false, true, false, 0); + + /* + * We raced, and lost, someone else holds the reservation now, + * and is probably busy in ttm_bo_cleanup_memtype_use. + * + * Even if it's not the case, because we finished waiting any + * delayed destruction would succeed, so just return success + * here. + */ + if (ret) { + spin_unlock(&glob->lru_lock); + return 0; + } + } else + spin_unlock(&bdev->fence_lock); + + if (ret || unlikely(list_empty(&bo->ddestroy))) { + atomic_set(&bo->reserved, 0); + wake_up_all(&bo->event_queue); + spin_unlock(&glob->lru_lock); + return ret; + } + + put_count = ttm_bo_del_from_lru(bo); + list_del_init(&bo->ddestroy); + ++put_count; + + spin_unlock(&glob->lru_lock); + ttm_bo_cleanup_memtype_use(bo); + + ttm_bo_list_ref_sub(bo, put_count, true); + + return 0; +} + +/** + * Traverse the delayed list, and call ttm_bo_cleanup_refs on all + * encountered buffers. + */ + +static int ttm_bo_delayed_delete(struct ttm_bo_device *bdev, bool remove_all) +{ + struct ttm_bo_global *glob = bdev->glob; + struct ttm_buffer_object *entry = NULL; + int ret = 0; + + spin_lock(&glob->lru_lock); + if (list_empty(&bdev->ddestroy)) + goto out_unlock; + + entry = list_first_entry(&bdev->ddestroy, + struct ttm_buffer_object, ddestroy); + kref_get(&entry->list_kref); + + for (;;) { + struct ttm_buffer_object *nentry = NULL; + + if (entry->ddestroy.next != &bdev->ddestroy) { + nentry = list_first_entry(&entry->ddestroy, + struct ttm_buffer_object, ddestroy); + kref_get(&nentry->list_kref); + } + + ret = ttm_bo_reserve_nolru(entry, false, true, false, 0); + if (remove_all && ret) { + spin_unlock(&glob->lru_lock); + ret = ttm_bo_reserve_nolru(entry, false, false, + false, 0); + spin_lock(&glob->lru_lock); + } + + if (!ret) + ret = ttm_bo_cleanup_refs_and_unlock(entry, false, + !remove_all); + else + spin_unlock(&glob->lru_lock); + + kref_put(&entry->list_kref, ttm_bo_release_list); + entry = nentry; + + if (ret || !entry) + goto out; + + spin_lock(&glob->lru_lock); + if (list_empty(&entry->ddestroy)) + break; + } + +out_unlock: + spin_unlock(&glob->lru_lock); +out: + if (entry) + kref_put(&entry->list_kref, ttm_bo_release_list); + return ret; +} + +static void ttm_bo_delayed_workqueue(struct work_struct *work) +{ + struct ttm_bo_device *bdev = + container_of(work, struct ttm_bo_device, wq.work); + + if (ttm_bo_delayed_delete(bdev, false)) { + schedule_delayed_work(&bdev->wq, + ((HZ / 100) < 1) ? 1 : HZ / 100); + } +} + +static void ttm_bo_release(struct kref *kref) +{ + struct ttm_buffer_object *bo = + container_of(kref, struct ttm_buffer_object, kref); + struct ttm_bo_device *bdev = bo->bdev; + struct ttm_mem_type_manager *man = &bdev->man[bo->mem.mem_type]; + + write_lock(&bdev->vm_lock); + if (likely(bo->vm_node != NULL)) { + rb_erase(&bo->vm_rb, &bdev->addr_space_rb); + drm_mm_put_block(bo->vm_node); + bo->vm_node = NULL; + } + write_unlock(&bdev->vm_lock); + ttm_mem_io_lock(man, false); + ttm_mem_io_free_vm(bo); + ttm_mem_io_unlock(man); + ttm_bo_cleanup_refs_or_queue(bo); + kref_put(&bo->list_kref, ttm_bo_release_list); +} + +void ttm_bo_unref(struct ttm_buffer_object **p_bo) +{ + struct ttm_buffer_object *bo = *p_bo; + + *p_bo = NULL; + kref_put(&bo->kref, ttm_bo_release); +} +EXPORT_SYMBOL(ttm_bo_unref); + +int ttm_bo_lock_delayed_workqueue(struct ttm_bo_device *bdev) +{ + return cancel_delayed_work_sync(&bdev->wq); +} +EXPORT_SYMBOL(ttm_bo_lock_delayed_workqueue); + +void ttm_bo_unlock_delayed_workqueue(struct ttm_bo_device *bdev, int resched) +{ + if (resched) + schedule_delayed_work(&bdev->wq, + ((HZ / 100) < 1) ? 1 : HZ / 100); +} +EXPORT_SYMBOL(ttm_bo_unlock_delayed_workqueue); + +static int ttm_bo_evict(struct ttm_buffer_object *bo, bool interruptible, + bool no_wait_gpu) +{ + struct ttm_bo_device *bdev = bo->bdev; + struct ttm_mem_reg evict_mem; + struct ttm_placement placement; + int ret = 0; + + spin_lock(&bdev->fence_lock); + ret = ttm_bo_wait(bo, false, interruptible, no_wait_gpu); + spin_unlock(&bdev->fence_lock); + + if (unlikely(ret != 0)) { + if (ret != -ERESTARTSYS) { + pr_err("Failed to expire sync object before buffer eviction\n"); + } + goto out; + } + + BUG_ON(!ttm_bo_is_reserved(bo)); + + evict_mem = bo->mem; + evict_mem.mm_node = NULL; + evict_mem.bus.io_reserved_vm = false; + evict_mem.bus.io_reserved_count = 0; + + placement.fpfn = 0; + placement.lpfn = 0; + placement.num_placement = 0; + placement.num_busy_placement = 0; + bdev->driver->evict_flags(bo, &placement); + ret = ttm_bo_mem_space(bo, &placement, &evict_mem, interruptible, + no_wait_gpu); + if (ret) { + if (ret != -ERESTARTSYS) { + pr_err("Failed to find memory space for buffer 0x%p eviction\n", + bo); + ttm_bo_mem_space_debug(bo, &placement); + } + goto out; + } + + ret = ttm_bo_handle_move_mem(bo, &evict_mem, true, interruptible, + no_wait_gpu); + if (ret) { + if (ret != -ERESTARTSYS) + pr_err("Buffer eviction failed\n"); + ttm_bo_mem_put(bo, &evict_mem); + goto out; + } + bo->evicted = true; +out: + return ret; +} + +static int ttm_mem_evict_first(struct ttm_bo_device *bdev, + uint32_t mem_type, + bool interruptible, + bool no_wait_gpu) +{ + struct ttm_bo_global *glob = bdev->glob; + struct ttm_mem_type_manager *man = &bdev->man[mem_type]; + struct ttm_buffer_object *bo; + int ret = -EBUSY, put_count; + + spin_lock(&glob->lru_lock); + list_for_each_entry(bo, &man->lru, lru) { + ret = ttm_bo_reserve_nolru(bo, false, true, false, 0); + if (!ret) + break; + } + + if (ret) { + spin_unlock(&glob->lru_lock); + return ret; + } + + kref_get(&bo->list_kref); + + if (!list_empty(&bo->ddestroy)) { + ret = ttm_bo_cleanup_refs_and_unlock(bo, interruptible, + no_wait_gpu); + kref_put(&bo->list_kref, ttm_bo_release_list); + return ret; + } + + put_count = ttm_bo_del_from_lru(bo); + spin_unlock(&glob->lru_lock); + + BUG_ON(ret != 0); + + ttm_bo_list_ref_sub(bo, put_count, true); + + ret = ttm_bo_evict(bo, interruptible, no_wait_gpu); + ttm_bo_unreserve(bo); + + kref_put(&bo->list_kref, ttm_bo_release_list); + return ret; +} + +void ttm_bo_mem_put(struct ttm_buffer_object *bo, struct ttm_mem_reg *mem) +{ + struct ttm_mem_type_manager *man = &bo->bdev->man[mem->mem_type]; + + if (mem->mm_node) + (*man->func->put_node)(man, mem); +} +EXPORT_SYMBOL(ttm_bo_mem_put); + +/** + * Repeatedly evict memory from the LRU for @mem_type until we create enough + * space, or we've evicted everything and there isn't enough space. + */ +static int ttm_bo_mem_force_space(struct ttm_buffer_object *bo, + uint32_t mem_type, + struct ttm_placement *placement, + struct ttm_mem_reg *mem, + bool interruptible, + bool no_wait_gpu) +{ + struct ttm_bo_device *bdev = bo->bdev; + struct ttm_mem_type_manager *man = &bdev->man[mem_type]; + int ret; + + do { + ret = (*man->func->get_node)(man, bo, placement, mem); + if (unlikely(ret != 0)) + return ret; + if (mem->mm_node) + break; + ret = ttm_mem_evict_first(bdev, mem_type, + interruptible, no_wait_gpu); + if (unlikely(ret != 0)) + return ret; + } while (1); + if (mem->mm_node == NULL) + return -ENOMEM; + mem->mem_type = mem_type; + return 0; +} + +static uint32_t ttm_bo_select_caching(struct ttm_mem_type_manager *man, + uint32_t cur_placement, + uint32_t proposed_placement) +{ + uint32_t caching = proposed_placement & TTM_PL_MASK_CACHING; + uint32_t result = proposed_placement & ~TTM_PL_MASK_CACHING; + + /** + * Keep current caching if possible. + */ + + if ((cur_placement & caching) != 0) + result |= (cur_placement & caching); + else if ((man->default_caching & caching) != 0) + result |= man->default_caching; + else if ((TTM_PL_FLAG_CACHED & caching) != 0) + result |= TTM_PL_FLAG_CACHED; + else if ((TTM_PL_FLAG_WC & caching) != 0) + result |= TTM_PL_FLAG_WC; + else if ((TTM_PL_FLAG_UNCACHED & caching) != 0) + result |= TTM_PL_FLAG_UNCACHED; + + return result; +} + +static bool ttm_bo_mt_compatible(struct ttm_mem_type_manager *man, + uint32_t mem_type, + uint32_t proposed_placement, + uint32_t *masked_placement) +{ + uint32_t cur_flags = ttm_bo_type_flags(mem_type); + + if ((cur_flags & proposed_placement & TTM_PL_MASK_MEM) == 0) + return false; + + if ((proposed_placement & man->available_caching) == 0) + return false; + + cur_flags |= (proposed_placement & man->available_caching); + + *masked_placement = cur_flags; + return true; +} + +/** + * Creates space for memory region @mem according to its type. + * + * This function first searches for free space in compatible memory types in + * the priority order defined by the driver. If free space isn't found, then + * ttm_bo_mem_force_space is attempted in priority order to evict and find + * space. + */ +int ttm_bo_mem_space(struct ttm_buffer_object *bo, + struct ttm_placement *placement, + struct ttm_mem_reg *mem, + bool interruptible, + bool no_wait_gpu) +{ + struct ttm_bo_device *bdev = bo->bdev; + struct ttm_mem_type_manager *man; + uint32_t mem_type = TTM_PL_SYSTEM; + uint32_t cur_flags = 0; + bool type_found = false; + bool type_ok = false; + bool has_erestartsys = false; + int i, ret; + + mem->mm_node = NULL; + for (i = 0; i < placement->num_placement; ++i) { + ret = ttm_mem_type_from_flags(placement->placement[i], + &mem_type); + if (ret) + return ret; + man = &bdev->man[mem_type]; + + type_ok = ttm_bo_mt_compatible(man, + mem_type, + placement->placement[i], + &cur_flags); + + if (!type_ok) + continue; + + cur_flags = ttm_bo_select_caching(man, bo->mem.placement, + cur_flags); + /* + * Use the access and other non-mapping-related flag bits from + * the memory placement flags to the current flags + */ + ttm_flag_masked(&cur_flags, placement->placement[i], + ~TTM_PL_MASK_MEMTYPE); + + if (mem_type == TTM_PL_SYSTEM) + break; + + if (man->has_type && man->use_type) { + type_found = true; + ret = (*man->func->get_node)(man, bo, placement, mem); + if (unlikely(ret)) + return ret; + } + if (mem->mm_node) + break; + } + + if ((type_ok && (mem_type == TTM_PL_SYSTEM)) || mem->mm_node) { + mem->mem_type = mem_type; + mem->placement = cur_flags; + return 0; + } + + if (!type_found) + return -EINVAL; + + for (i = 0; i < placement->num_busy_placement; ++i) { + ret = ttm_mem_type_from_flags(placement->busy_placement[i], + &mem_type); + if (ret) + return ret; + man = &bdev->man[mem_type]; + if (!man->has_type) + continue; + if (!ttm_bo_mt_compatible(man, + mem_type, + placement->busy_placement[i], + &cur_flags)) + continue; + + cur_flags = ttm_bo_select_caching(man, bo->mem.placement, + cur_flags); + /* + * Use the access and other non-mapping-related flag bits from + * the memory placement flags to the current flags + */ + ttm_flag_masked(&cur_flags, placement->busy_placement[i], + ~TTM_PL_MASK_MEMTYPE); + + + if (mem_type == TTM_PL_SYSTEM) { + mem->mem_type = mem_type; + mem->placement = cur_flags; + mem->mm_node = NULL; + return 0; + } + + ret = ttm_bo_mem_force_space(bo, mem_type, placement, mem, + interruptible, no_wait_gpu); + if (ret == 0 && mem->mm_node) { + mem->placement = cur_flags; + return 0; + } + if (ret == -ERESTARTSYS) + has_erestartsys = true; + } + ret = (has_erestartsys) ? -ERESTARTSYS : -ENOMEM; + return ret; +} +EXPORT_SYMBOL(ttm_bo_mem_space); + +int ttm_bo_move_buffer(struct ttm_buffer_object *bo, + struct ttm_placement *placement, + bool interruptible, + bool no_wait_gpu) +{ + int ret = 0; + struct ttm_mem_reg mem; + struct ttm_bo_device *bdev = bo->bdev; + + BUG_ON(!ttm_bo_is_reserved(bo)); + + /* + * FIXME: It's possible to pipeline buffer moves. + * Have the driver move function wait for idle when necessary, + * instead of doing it here. + */ + spin_lock(&bdev->fence_lock); + ret = ttm_bo_wait(bo, false, interruptible, no_wait_gpu); + spin_unlock(&bdev->fence_lock); + if (ret) + return ret; + mem.num_pages = bo->num_pages; + mem.size = mem.num_pages << PAGE_SHIFT; + mem.page_alignment = bo->mem.page_alignment; + mem.bus.io_reserved_vm = false; + mem.bus.io_reserved_count = 0; + /* + * Determine where to move the buffer. + */ + ret = ttm_bo_mem_space(bo, placement, &mem, + interruptible, no_wait_gpu); + if (ret) + goto out_unlock; + ret = ttm_bo_handle_move_mem(bo, &mem, false, + interruptible, no_wait_gpu); +out_unlock: + if (ret && mem.mm_node) + ttm_bo_mem_put(bo, &mem); + return ret; +} + +static int ttm_bo_mem_compat(struct ttm_placement *placement, + struct ttm_mem_reg *mem) +{ + int i; + + if (mem->mm_node && placement->lpfn != 0 && + (mem->start < placement->fpfn || + mem->start + mem->num_pages > placement->lpfn)) + return -1; + + for (i = 0; i < placement->num_placement; i++) { + if ((placement->placement[i] & mem->placement & + TTM_PL_MASK_CACHING) && + (placement->placement[i] & mem->placement & + TTM_PL_MASK_MEM)) + return i; + } + return -1; +} + +int ttm_bo_validate(struct ttm_buffer_object *bo, + struct ttm_placement *placement, + bool interruptible, + bool no_wait_gpu) +{ + int ret; + + BUG_ON(!ttm_bo_is_reserved(bo)); + /* Check that range is valid */ + if (placement->lpfn || placement->fpfn) + if (placement->fpfn > placement->lpfn || + (placement->lpfn - placement->fpfn) < bo->num_pages) + return -EINVAL; + /* + * Check whether we need to move buffer. + */ + ret = ttm_bo_mem_compat(placement, &bo->mem); + if (ret < 0) { + ret = ttm_bo_move_buffer(bo, placement, interruptible, + no_wait_gpu); + if (ret) + return ret; + } else { + /* + * Use the access and other non-mapping-related flag bits from + * the compatible memory placement flags to the active flags + */ + ttm_flag_masked(&bo->mem.placement, placement->placement[ret], + ~TTM_PL_MASK_MEMTYPE); + } + /* + * We might need to add a TTM. + */ + if (bo->mem.mem_type == TTM_PL_SYSTEM && bo->ttm == NULL) { + ret = ttm_bo_add_ttm(bo, true); + if (ret) + return ret; + } + return 0; +} +EXPORT_SYMBOL(ttm_bo_validate); + +int ttm_bo_check_placement(struct ttm_buffer_object *bo, + struct ttm_placement *placement) +{ + BUG_ON((placement->fpfn || placement->lpfn) && + (bo->mem.num_pages > (placement->lpfn - placement->fpfn))); + + return 0; +} + +int ttm_bo_init(struct ttm_bo_device *bdev, + struct ttm_buffer_object *bo, + unsigned long size, + enum ttm_bo_type type, + struct ttm_placement *placement, + uint32_t page_alignment, + bool interruptible, + struct file *persistent_swap_storage, + size_t acc_size, + struct sg_table *sg, + void (*destroy) (struct ttm_buffer_object *)) +{ + int ret = 0; + unsigned long num_pages; + struct ttm_mem_global *mem_glob = bdev->glob->mem_glob; + + ret = ttm_mem_global_alloc(mem_glob, acc_size, false, false); + if (ret) { + pr_err("Out of kernel memory\n"); + if (destroy) + (*destroy)(bo); + else + kfree(bo); + return -ENOMEM; + } + + num_pages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; + if (num_pages == 0) { + pr_err("Illegal buffer object size\n"); + if (destroy) + (*destroy)(bo); + else + kfree(bo); + ttm_mem_global_free(mem_glob, acc_size); + return -EINVAL; + } + bo->destroy = destroy; + + kref_init(&bo->kref); + kref_init(&bo->list_kref); + atomic_set(&bo->cpu_writers, 0); + atomic_set(&bo->reserved, 1); + init_waitqueue_head(&bo->event_queue); + INIT_LIST_HEAD(&bo->lru); + INIT_LIST_HEAD(&bo->ddestroy); + INIT_LIST_HEAD(&bo->swap); + INIT_LIST_HEAD(&bo->io_reserve_lru); + bo->bdev = bdev; + bo->glob = bdev->glob; + bo->type = type; + bo->num_pages = num_pages; + bo->mem.size = num_pages << PAGE_SHIFT; + bo->mem.mem_type = TTM_PL_SYSTEM; + bo->mem.num_pages = bo->num_pages; + bo->mem.mm_node = NULL; + bo->mem.page_alignment = page_alignment; + bo->mem.bus.io_reserved_vm = false; + bo->mem.bus.io_reserved_count = 0; + bo->priv_flags = 0; + bo->mem.placement = (TTM_PL_FLAG_SYSTEM | TTM_PL_FLAG_CACHED); + bo->seq_valid = false; + bo->persistent_swap_storage = persistent_swap_storage; + bo->acc_size = acc_size; + bo->sg = sg; + atomic_inc(&bo->glob->bo_count); + + ret = ttm_bo_check_placement(bo, placement); + if (unlikely(ret != 0)) + goto out_err; + + /* + * For ttm_bo_type_device buffers, allocate + * address space from the device. + */ + if (bo->type == ttm_bo_type_device || + bo->type == ttm_bo_type_sg) { + ret = ttm_bo_setup_vm(bo); + if (ret) + goto out_err; + } + + ret = ttm_bo_validate(bo, placement, interruptible, false); + if (ret) + goto out_err; + + ttm_bo_unreserve(bo); + return 0; + +out_err: + ttm_bo_unreserve(bo); + ttm_bo_unref(&bo); + + return ret; +} +EXPORT_SYMBOL(ttm_bo_init); + +size_t ttm_bo_acc_size(struct ttm_bo_device *bdev, + unsigned long bo_size, + unsigned struct_size) +{ + unsigned npages = (PAGE_ALIGN(bo_size)) >> PAGE_SHIFT; + size_t size = 0; + + size += ttm_round_pot(struct_size); + size += PAGE_ALIGN(npages * sizeof(void *)); + size += ttm_round_pot(sizeof(struct ttm_tt)); + return size; +} +EXPORT_SYMBOL(ttm_bo_acc_size); + +size_t ttm_bo_dma_acc_size(struct ttm_bo_device *bdev, + unsigned long bo_size, + unsigned struct_size) +{ + unsigned npages = (PAGE_ALIGN(bo_size)) >> PAGE_SHIFT; + size_t size = 0; + + size += ttm_round_pot(struct_size); + size += PAGE_ALIGN(npages * sizeof(void *)); + size += PAGE_ALIGN(npages * sizeof(dma_addr_t)); + size += ttm_round_pot(sizeof(struct ttm_dma_tt)); + return size; +} +EXPORT_SYMBOL(ttm_bo_dma_acc_size); + +int ttm_bo_create(struct ttm_bo_device *bdev, + unsigned long size, + enum ttm_bo_type type, + struct ttm_placement *placement, + uint32_t page_alignment, + bool interruptible, + struct file *persistent_swap_storage, + struct ttm_buffer_object **p_bo) +{ + struct ttm_buffer_object *bo; + size_t acc_size; + int ret; + + bo = kzalloc(sizeof(*bo), GFP_KERNEL); + if (unlikely(bo == NULL)) + return -ENOMEM; + + acc_size = ttm_bo_acc_size(bdev, size, sizeof(struct ttm_buffer_object)); + ret = ttm_bo_init(bdev, bo, size, type, placement, page_alignment, + interruptible, persistent_swap_storage, acc_size, + NULL, NULL); + if (likely(ret == 0)) + *p_bo = bo; + + return ret; +} +EXPORT_SYMBOL(ttm_bo_create); + +static int ttm_bo_force_list_clean(struct ttm_bo_device *bdev, + unsigned mem_type, bool allow_errors) +{ + struct ttm_mem_type_manager *man = &bdev->man[mem_type]; + struct ttm_bo_global *glob = bdev->glob; + int ret; + + /* + * Can't use standard list traversal since we're unlocking. + */ + + spin_lock(&glob->lru_lock); + while (!list_empty(&man->lru)) { + spin_unlock(&glob->lru_lock); + ret = ttm_mem_evict_first(bdev, mem_type, false, false); + if (ret) { + if (allow_errors) { + return ret; + } else { + pr_err("Cleanup eviction failed\n"); + } + } + spin_lock(&glob->lru_lock); + } + spin_unlock(&glob->lru_lock); + return 0; +} + +int ttm_bo_clean_mm(struct ttm_bo_device *bdev, unsigned mem_type) +{ + struct ttm_mem_type_manager *man; + int ret = -EINVAL; + + if (mem_type >= TTM_NUM_MEM_TYPES) { + pr_err("Illegal memory type %d\n", mem_type); + return ret; + } + man = &bdev->man[mem_type]; + + if (!man->has_type) { + pr_err("Trying to take down uninitialized memory manager type %u\n", + mem_type); + return ret; + } + + man->use_type = false; + man->has_type = false; + + ret = 0; + if (mem_type > 0) { + ttm_bo_force_list_clean(bdev, mem_type, false); + + ret = (*man->func->takedown)(man); + } + + return ret; +} +EXPORT_SYMBOL(ttm_bo_clean_mm); + +int ttm_bo_evict_mm(struct ttm_bo_device *bdev, unsigned mem_type) +{ + struct ttm_mem_type_manager *man = &bdev->man[mem_type]; + + if (mem_type == 0 || mem_type >= TTM_NUM_MEM_TYPES) { + pr_err("Illegal memory manager memory type %u\n", mem_type); + return -EINVAL; + } + + if (!man->has_type) { + pr_err("Memory type %u has not been initialized\n", mem_type); + return 0; + } + + return ttm_bo_force_list_clean(bdev, mem_type, true); +} +EXPORT_SYMBOL(ttm_bo_evict_mm); + +int ttm_bo_init_mm(struct ttm_bo_device *bdev, unsigned type, + unsigned long p_size) +{ + int ret = -EINVAL; + struct ttm_mem_type_manager *man; + + BUG_ON(type >= TTM_NUM_MEM_TYPES); + man = &bdev->man[type]; + BUG_ON(man->has_type); + man->io_reserve_fastpath = true; + man->use_io_reserve_lru = false; + mutex_init(&man->io_reserve_mutex); + INIT_LIST_HEAD(&man->io_reserve_lru); + + ret = bdev->driver->init_mem_type(bdev, type, man); + if (ret) + return ret; + man->bdev = bdev; + + ret = 0; + if (type != TTM_PL_SYSTEM) { + ret = (*man->func->init)(man, p_size); + if (ret) + return ret; + } + man->has_type = true; + man->use_type = true; + man->size = p_size; + + INIT_LIST_HEAD(&man->lru); + + return 0; +} +EXPORT_SYMBOL(ttm_bo_init_mm); + +static void ttm_bo_global_kobj_release(struct kobject *kobj) +{ + struct ttm_bo_global *glob = + container_of(kobj, struct ttm_bo_global, kobj); + + ttm_mem_unregister_shrink(glob->mem_glob, &glob->shrink); + __free_page(glob->dummy_read_page); + kfree(glob); +} + +void ttm_bo_global_release(struct drm_global_reference *ref) +{ + struct ttm_bo_global *glob = ref->object; + + kobject_del(&glob->kobj); + kobject_put(&glob->kobj); +} +EXPORT_SYMBOL(ttm_bo_global_release); + +int ttm_bo_global_init(struct drm_global_reference *ref) +{ + struct ttm_bo_global_ref *bo_ref = + container_of(ref, struct ttm_bo_global_ref, ref); + struct ttm_bo_global *glob = ref->object; + int ret; + + mutex_init(&glob->device_list_mutex); + spin_lock_init(&glob->lru_lock); + glob->mem_glob = bo_ref->mem_glob; + glob->dummy_read_page = alloc_page(__GFP_ZERO | GFP_DMA32); + + if (unlikely(glob->dummy_read_page == NULL)) { + ret = -ENOMEM; + goto out_no_drp; + } + + INIT_LIST_HEAD(&glob->swap_lru); + INIT_LIST_HEAD(&glob->device_list); + + ttm_mem_init_shrink(&glob->shrink, ttm_bo_swapout); + ret = ttm_mem_register_shrink(glob->mem_glob, &glob->shrink); + if (unlikely(ret != 0)) { + pr_err("Could not register buffer object swapout\n"); + goto out_no_shrink; + } + + atomic_set(&glob->bo_count, 0); + + ret = kobject_init_and_add( + &glob->kobj, &ttm_bo_glob_kobj_type, ttm_get_kobj(), "buffer_objects"); + if (unlikely(ret != 0)) + kobject_put(&glob->kobj); + return ret; +out_no_shrink: + __free_page(glob->dummy_read_page); +out_no_drp: + kfree(glob); + return ret; +} +EXPORT_SYMBOL(ttm_bo_global_init); + + +int ttm_bo_device_release(struct ttm_bo_device *bdev) +{ + int ret = 0; + unsigned i = TTM_NUM_MEM_TYPES; + struct ttm_mem_type_manager *man; + struct ttm_bo_global *glob = bdev->glob; + + while (i--) { + man = &bdev->man[i]; + if (man->has_type) { + man->use_type = false; + if ((i != TTM_PL_SYSTEM) && ttm_bo_clean_mm(bdev, i)) { + ret = -EBUSY; + pr_err("DRM memory manager type %d is not clean\n", + i); + } + man->has_type = false; + } + } + + mutex_lock(&glob->device_list_mutex); + list_del(&bdev->device_list); + mutex_unlock(&glob->device_list_mutex); + + cancel_delayed_work_sync(&bdev->wq); + + while (ttm_bo_delayed_delete(bdev, true)) + ; + + spin_lock(&glob->lru_lock); + if (list_empty(&bdev->ddestroy)) + TTM_DEBUG("Delayed destroy list was clean\n"); + + if (list_empty(&bdev->man[0].lru)) + TTM_DEBUG("Swap list was clean\n"); + spin_unlock(&glob->lru_lock); + + BUG_ON(!drm_mm_clean(&bdev->addr_space_mm)); + write_lock(&bdev->vm_lock); + drm_mm_takedown(&bdev->addr_space_mm); + write_unlock(&bdev->vm_lock); + + return ret; +} +EXPORT_SYMBOL(ttm_bo_device_release); + +int ttm_bo_device_init(struct ttm_bo_device *bdev, + struct ttm_bo_global *glob, + struct ttm_bo_driver *driver, + uint64_t file_page_offset, + bool need_dma32) +{ + int ret = -EINVAL; + + rwlock_init(&bdev->vm_lock); + bdev->driver = driver; + + memset(bdev->man, 0, sizeof(bdev->man)); + + /* + * Initialize the system memory buffer type. + * Other types need to be driver / IOCTL initialized. + */ + ret = ttm_bo_init_mm(bdev, TTM_PL_SYSTEM, 0); + if (unlikely(ret != 0)) + goto out_no_sys; + + bdev->addr_space_rb = RB_ROOT; + ret = drm_mm_init(&bdev->addr_space_mm, file_page_offset, 0x10000000); + if (unlikely(ret != 0)) + goto out_no_addr_mm; + + INIT_DELAYED_WORK(&bdev->wq, ttm_bo_delayed_workqueue); + INIT_LIST_HEAD(&bdev->ddestroy); + bdev->dev_mapping = NULL; + bdev->glob = glob; + bdev->need_dma32 = need_dma32; + bdev->val_seq = 0; + spin_lock_init(&bdev->fence_lock); + mutex_lock(&glob->device_list_mutex); + list_add_tail(&bdev->device_list, &glob->device_list); + mutex_unlock(&glob->device_list_mutex); + + return 0; +out_no_addr_mm: + ttm_bo_clean_mm(bdev, 0); +out_no_sys: + return ret; +} +EXPORT_SYMBOL(ttm_bo_device_init); + +/* + * buffer object vm functions. + */ + +bool ttm_mem_reg_is_pci(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem) +{ + struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type]; + + if (!(man->flags & TTM_MEMTYPE_FLAG_FIXED)) { + if (mem->mem_type == TTM_PL_SYSTEM) + return false; + + if (man->flags & TTM_MEMTYPE_FLAG_CMA) + return false; + + if (mem->placement & TTM_PL_FLAG_CACHED) + return false; + } + return true; +} + +void ttm_bo_unmap_virtual_locked(struct ttm_buffer_object *bo) +{ + struct ttm_bo_device *bdev = bo->bdev; + loff_t offset = (loff_t) bo->addr_space_offset; + loff_t holelen = ((loff_t) bo->mem.num_pages) << PAGE_SHIFT; + + if (!bdev->dev_mapping) + return; + unmap_mapping_range(bdev->dev_mapping, offset, holelen, 1); + ttm_mem_io_free_vm(bo); +} + +void ttm_bo_unmap_virtual(struct ttm_buffer_object *bo) +{ + struct ttm_bo_device *bdev = bo->bdev; + struct ttm_mem_type_manager *man = &bdev->man[bo->mem.mem_type]; + + ttm_mem_io_lock(man, false); + ttm_bo_unmap_virtual_locked(bo); + ttm_mem_io_unlock(man); +} + + +EXPORT_SYMBOL(ttm_bo_unmap_virtual); + +static void ttm_bo_vm_insert_rb(struct ttm_buffer_object *bo) +{ + struct ttm_bo_device *bdev = bo->bdev; + struct rb_node **cur = &bdev->addr_space_rb.rb_node; + struct rb_node *parent = NULL; + struct ttm_buffer_object *cur_bo; + unsigned long offset = bo->vm_node->start; + unsigned long cur_offset; + + while (*cur) { + parent = *cur; + cur_bo = rb_entry(parent, struct ttm_buffer_object, vm_rb); + cur_offset = cur_bo->vm_node->start; + if (offset < cur_offset) + cur = &parent->rb_left; + else if (offset > cur_offset) + cur = &parent->rb_right; + else + BUG(); + } + + rb_link_node(&bo->vm_rb, parent, cur); + rb_insert_color(&bo->vm_rb, &bdev->addr_space_rb); +} + +/** + * ttm_bo_setup_vm: + * + * @bo: the buffer to allocate address space for + * + * Allocate address space in the drm device so that applications + * can mmap the buffer and access the contents. This only + * applies to ttm_bo_type_device objects as others are not + * placed in the drm device address space. + */ + +static int ttm_bo_setup_vm(struct ttm_buffer_object *bo) +{ + struct ttm_bo_device *bdev = bo->bdev; + int ret; + +retry_pre_get: + ret = drm_mm_pre_get(&bdev->addr_space_mm); + if (unlikely(ret != 0)) + return ret; + + write_lock(&bdev->vm_lock); + bo->vm_node = drm_mm_search_free(&bdev->addr_space_mm, + bo->mem.num_pages, 0, 0); + + if (unlikely(bo->vm_node == NULL)) { + ret = -ENOMEM; + goto out_unlock; + } + + bo->vm_node = drm_mm_get_block_atomic(bo->vm_node, + bo->mem.num_pages, 0); + + if (unlikely(bo->vm_node == NULL)) { + write_unlock(&bdev->vm_lock); + goto retry_pre_get; + } + + ttm_bo_vm_insert_rb(bo); + write_unlock(&bdev->vm_lock); + bo->addr_space_offset = ((uint64_t) bo->vm_node->start) << PAGE_SHIFT; + + return 0; +out_unlock: + write_unlock(&bdev->vm_lock); + return ret; +} + +int ttm_bo_wait(struct ttm_buffer_object *bo, + bool lazy, bool interruptible, bool no_wait) +{ + struct ttm_bo_driver *driver = bo->bdev->driver; + struct ttm_bo_device *bdev = bo->bdev; + void *sync_obj; + int ret = 0; + + if (likely(bo->sync_obj == NULL)) + return 0; + + while (bo->sync_obj) { + + if (driver->sync_obj_signaled(bo->sync_obj)) { + void *tmp_obj = bo->sync_obj; + bo->sync_obj = NULL; + clear_bit(TTM_BO_PRIV_FLAG_MOVING, &bo->priv_flags); + spin_unlock(&bdev->fence_lock); + driver->sync_obj_unref(&tmp_obj); + spin_lock(&bdev->fence_lock); + continue; + } + + if (no_wait) + return -EBUSY; + + sync_obj = driver->sync_obj_ref(bo->sync_obj); + spin_unlock(&bdev->fence_lock); + ret = driver->sync_obj_wait(sync_obj, + lazy, interruptible); + if (unlikely(ret != 0)) { + driver->sync_obj_unref(&sync_obj); + spin_lock(&bdev->fence_lock); + return ret; + } + spin_lock(&bdev->fence_lock); + if (likely(bo->sync_obj == sync_obj)) { + void *tmp_obj = bo->sync_obj; + bo->sync_obj = NULL; + clear_bit(TTM_BO_PRIV_FLAG_MOVING, + &bo->priv_flags); + spin_unlock(&bdev->fence_lock); + driver->sync_obj_unref(&sync_obj); + driver->sync_obj_unref(&tmp_obj); + spin_lock(&bdev->fence_lock); + } else { + spin_unlock(&bdev->fence_lock); + driver->sync_obj_unref(&sync_obj); + spin_lock(&bdev->fence_lock); + } + } + return 0; +} +EXPORT_SYMBOL(ttm_bo_wait); + +int ttm_bo_synccpu_write_grab(struct ttm_buffer_object *bo, bool no_wait) +{ + struct ttm_bo_device *bdev = bo->bdev; + int ret = 0; + + /* + * Using ttm_bo_reserve makes sure the lru lists are updated. + */ + + ret = ttm_bo_reserve(bo, true, no_wait, false, 0); + if (unlikely(ret != 0)) + return ret; + spin_lock(&bdev->fence_lock); + ret = ttm_bo_wait(bo, false, true, no_wait); + spin_unlock(&bdev->fence_lock); + if (likely(ret == 0)) + atomic_inc(&bo->cpu_writers); + ttm_bo_unreserve(bo); + return ret; +} +EXPORT_SYMBOL(ttm_bo_synccpu_write_grab); + +void ttm_bo_synccpu_write_release(struct ttm_buffer_object *bo) +{ + atomic_dec(&bo->cpu_writers); +} +EXPORT_SYMBOL(ttm_bo_synccpu_write_release); + +/** + * A buffer object shrink method that tries to swap out the first + * buffer object on the bo_global::swap_lru list. + */ + +static int ttm_bo_swapout(struct ttm_mem_shrink *shrink) +{ + struct ttm_bo_global *glob = + container_of(shrink, struct ttm_bo_global, shrink); + struct ttm_buffer_object *bo; + int ret = -EBUSY; + int put_count; + uint32_t swap_placement = (TTM_PL_FLAG_CACHED | TTM_PL_FLAG_SYSTEM); + + spin_lock(&glob->lru_lock); + list_for_each_entry(bo, &glob->swap_lru, swap) { + ret = ttm_bo_reserve_nolru(bo, false, true, false, 0); + if (!ret) + break; + } + + if (ret) { + spin_unlock(&glob->lru_lock); + return ret; + } + + kref_get(&bo->list_kref); + + if (!list_empty(&bo->ddestroy)) { + ret = ttm_bo_cleanup_refs_and_unlock(bo, false, false); + kref_put(&bo->list_kref, ttm_bo_release_list); + return ret; + } + + put_count = ttm_bo_del_from_lru(bo); + spin_unlock(&glob->lru_lock); + + ttm_bo_list_ref_sub(bo, put_count, true); + + /** + * Wait for GPU, then move to system cached. + */ + + spin_lock(&bo->bdev->fence_lock); + ret = ttm_bo_wait(bo, false, false, false); + spin_unlock(&bo->bdev->fence_lock); + + if (unlikely(ret != 0)) + goto out; + + if ((bo->mem.placement & swap_placement) != swap_placement) { + struct ttm_mem_reg evict_mem; + + evict_mem = bo->mem; + evict_mem.mm_node = NULL; + evict_mem.placement = TTM_PL_FLAG_SYSTEM | TTM_PL_FLAG_CACHED; + evict_mem.mem_type = TTM_PL_SYSTEM; + + ret = ttm_bo_handle_move_mem(bo, &evict_mem, true, + false, false); + if (unlikely(ret != 0)) + goto out; + } + + ttm_bo_unmap_virtual(bo); + + /** + * Swap out. Buffer will be swapped in again as soon as + * anyone tries to access a ttm page. + */ + + if (bo->bdev->driver->swap_notify) + bo->bdev->driver->swap_notify(bo); + + ret = ttm_tt_swapout(bo->ttm, bo->persistent_swap_storage); +out: + + /** + * + * Unreserve without putting on LRU to avoid swapping out an + * already swapped buffer. + */ + + atomic_set(&bo->reserved, 0); + wake_up_all(&bo->event_queue); + kref_put(&bo->list_kref, ttm_bo_release_list); + return ret; +} + +void ttm_bo_swapout_all(struct ttm_bo_device *bdev) +{ + while (ttm_bo_swapout(&bdev->glob->shrink) == 0) + ; +} +EXPORT_SYMBOL(ttm_bo_swapout_all); diff --git a/drivers/gpu/drm/ttm/ttm_bo_manager.c b/drivers/gpu/drm/ttm/ttm_bo_manager.c new file mode 100644 index 0000000..9212494 --- /dev/null +++ b/drivers/gpu/drm/ttm/ttm_bo_manager.c @@ -0,0 +1,157 @@ +/************************************************************************** + * + * Copyright (c) 2007-2010 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 <thellstrom-at-vmware-dot-com> + */ + +#include <drm/ttm/ttm_module.h> +#include <drm/ttm/ttm_bo_driver.h> +#include <drm/ttm/ttm_placement.h> +#include <drm/drm_mm.h> +#include <linux/slab.h> +#include <linux/spinlock.h> +#include <linux/module.h> + +/** + * Currently we use a spinlock for the lock, but a mutex *may* be + * more appropriate to reduce scheduling latency if the range manager + * ends up with very fragmented allocation patterns. + */ + +struct ttm_range_manager { + struct drm_mm mm; + spinlock_t lock; +}; + +static int ttm_bo_man_get_node(struct ttm_mem_type_manager *man, + struct ttm_buffer_object *bo, + struct ttm_placement *placement, + struct ttm_mem_reg *mem) +{ + struct ttm_range_manager *rman = (struct ttm_range_manager *) man->priv; + struct drm_mm *mm = &rman->mm; + struct drm_mm_node *node = NULL; + unsigned long lpfn; + int ret; + + lpfn = placement->lpfn; + if (!lpfn) + lpfn = man->size; + do { + ret = drm_mm_pre_get(mm); + if (unlikely(ret)) + return ret; + + spin_lock(&rman->lock); + node = drm_mm_search_free_in_range(mm, + mem->num_pages, mem->page_alignment, + placement->fpfn, lpfn, 1); + if (unlikely(node == NULL)) { + spin_unlock(&rman->lock); + return 0; + } + node = drm_mm_get_block_atomic_range(node, mem->num_pages, + mem->page_alignment, + placement->fpfn, + lpfn); + spin_unlock(&rman->lock); + } while (node == NULL); + + mem->mm_node = node; + mem->start = node->start; + return 0; +} + +static void ttm_bo_man_put_node(struct ttm_mem_type_manager *man, + struct ttm_mem_reg *mem) +{ + struct ttm_range_manager *rman = (struct ttm_range_manager *) man->priv; + + if (mem->mm_node) { + spin_lock(&rman->lock); + drm_mm_put_block(mem->mm_node); + spin_unlock(&rman->lock); + mem->mm_node = NULL; + } +} + +static int ttm_bo_man_init(struct ttm_mem_type_manager *man, + unsigned long p_size) +{ + struct ttm_range_manager *rman; + int ret; + + rman = kzalloc(sizeof(*rman), GFP_KERNEL); + if (!rman) + return -ENOMEM; + + ret = drm_mm_init(&rman->mm, 0, p_size); + if (ret) { + kfree(rman); + return ret; + } + + spin_lock_init(&rman->lock); + man->priv = rman; + return 0; +} + +static int ttm_bo_man_takedown(struct ttm_mem_type_manager *man) +{ + struct ttm_range_manager *rman = (struct ttm_range_manager *) man->priv; + struct drm_mm *mm = &rman->mm; + + spin_lock(&rman->lock); + if (drm_mm_clean(mm)) { + drm_mm_takedown(mm); + spin_unlock(&rman->lock); + kfree(rman); + man->priv = NULL; + return 0; + } + spin_unlock(&rman->lock); + return -EBUSY; +} + +static void ttm_bo_man_debug(struct ttm_mem_type_manager *man, + const char *prefix) +{ + struct ttm_range_manager *rman = (struct ttm_range_manager *) man->priv; + + spin_lock(&rman->lock); + drm_mm_debug_table(&rman->mm, prefix); + spin_unlock(&rman->lock); +} + +const struct ttm_mem_type_manager_func ttm_bo_manager_func = { + ttm_bo_man_init, + ttm_bo_man_takedown, + ttm_bo_man_get_node, + ttm_bo_man_put_node, + ttm_bo_man_debug +}; +EXPORT_SYMBOL(ttm_bo_manager_func); diff --git a/drivers/gpu/drm/ttm/ttm_bo_util.c b/drivers/gpu/drm/ttm/ttm_bo_util.c new file mode 100644 index 0000000..af89458 --- /dev/null +++ b/drivers/gpu/drm/ttm/ttm_bo_util.c @@ -0,0 +1,702 @@ +/************************************************************************** + * + * Copyright (c) 2007-2009 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 <thellstrom-at-vmware-dot-com> + */ + +#include <drm/ttm/ttm_bo_driver.h> +#include <drm/ttm/ttm_placement.h> +#include <linux/io.h> +#include <linux/highmem.h> +#include <linux/wait.h> +#include <linux/slab.h> +#include <linux/vmalloc.h> +#include <linux/module.h> + +void ttm_bo_free_old_node(struct ttm_buffer_object *bo) +{ + ttm_bo_mem_put(bo, &bo->mem); +} + +int ttm_bo_move_ttm(struct ttm_buffer_object *bo, + bool evict, + bool no_wait_gpu, struct ttm_mem_reg *new_mem) +{ + struct ttm_tt *ttm = bo->ttm; + struct ttm_mem_reg *old_mem = &bo->mem; + int ret; + + if (old_mem->mem_type != TTM_PL_SYSTEM) { + ttm_tt_unbind(ttm); + ttm_bo_free_old_node(bo); + ttm_flag_masked(&old_mem->placement, TTM_PL_FLAG_SYSTEM, + TTM_PL_MASK_MEM); + old_mem->mem_type = TTM_PL_SYSTEM; + } + + ret = ttm_tt_set_placement_caching(ttm, new_mem->placement); + if (unlikely(ret != 0)) + return ret; + + if (new_mem->mem_type != TTM_PL_SYSTEM) { + ret = ttm_tt_bind(ttm, new_mem); + if (unlikely(ret != 0)) + return ret; + } + + *old_mem = *new_mem; + new_mem->mm_node = NULL; + + return 0; +} +EXPORT_SYMBOL(ttm_bo_move_ttm); + +int ttm_mem_io_lock(struct ttm_mem_type_manager *man, bool interruptible) +{ + if (likely(man->io_reserve_fastpath)) + return 0; + + if (interruptible) + return mutex_lock_interruptible(&man->io_reserve_mutex); + + mutex_lock(&man->io_reserve_mutex); + return 0; +} +EXPORT_SYMBOL(ttm_mem_io_lock); + +void ttm_mem_io_unlock(struct ttm_mem_type_manager *man) +{ + if (likely(man->io_reserve_fastpath)) + return; + + mutex_unlock(&man->io_reserve_mutex); +} +EXPORT_SYMBOL(ttm_mem_io_unlock); + +static int ttm_mem_io_evict(struct ttm_mem_type_manager *man) +{ + struct ttm_buffer_object *bo; + + if (!man->use_io_reserve_lru || list_empty(&man->io_reserve_lru)) + return -EAGAIN; + + bo = list_first_entry(&man->io_reserve_lru, + struct ttm_buffer_object, + io_reserve_lru); + list_del_init(&bo->io_reserve_lru); + ttm_bo_unmap_virtual_locked(bo); + + return 0; +} + + +int ttm_mem_io_reserve(struct ttm_bo_device *bdev, + struct ttm_mem_reg *mem) +{ + struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type]; + int ret = 0; + + if (!bdev->driver->io_mem_reserve) + return 0; + if (likely(man->io_reserve_fastpath)) + return bdev->driver->io_mem_reserve(bdev, mem); + + if (bdev->driver->io_mem_reserve && + mem->bus.io_reserved_count++ == 0) { +retry: + ret = bdev->driver->io_mem_reserve(bdev, mem); + if (ret == -EAGAIN) { + ret = ttm_mem_io_evict(man); + if (ret == 0) + goto retry; + } + } + return ret; +} +EXPORT_SYMBOL(ttm_mem_io_reserve); + +void ttm_mem_io_free(struct ttm_bo_device *bdev, + struct ttm_mem_reg *mem) +{ + struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type]; + + if (likely(man->io_reserve_fastpath)) + return; + + if (bdev->driver->io_mem_reserve && + --mem->bus.io_reserved_count == 0 && + bdev->driver->io_mem_free) + bdev->driver->io_mem_free(bdev, mem); + +} +EXPORT_SYMBOL(ttm_mem_io_free); + +int ttm_mem_io_reserve_vm(struct ttm_buffer_object *bo) +{ + struct ttm_mem_reg *mem = &bo->mem; + int ret; + + if (!mem->bus.io_reserved_vm) { + struct ttm_mem_type_manager *man = + &bo->bdev->man[mem->mem_type]; + + ret = ttm_mem_io_reserve(bo->bdev, mem); + if (unlikely(ret != 0)) + return ret; + mem->bus.io_reserved_vm = true; + if (man->use_io_reserve_lru) + list_add_tail(&bo->io_reserve_lru, + &man->io_reserve_lru); + } + return 0; +} + +void ttm_mem_io_free_vm(struct ttm_buffer_object *bo) +{ + struct ttm_mem_reg *mem = &bo->mem; + + if (mem->bus.io_reserved_vm) { + mem->bus.io_reserved_vm = false; + list_del_init(&bo->io_reserve_lru); + ttm_mem_io_free(bo->bdev, mem); + } +} + +int ttm_mem_reg_ioremap(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem, + void **virtual) +{ + struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type]; + int ret; + void *addr; + + *virtual = NULL; + (void) ttm_mem_io_lock(man, false); + ret = ttm_mem_io_reserve(bdev, mem); + ttm_mem_io_unlock(man); + if (ret || !mem->bus.is_iomem) + return ret; + + if (mem->bus.addr) { + addr = mem->bus.addr; + } else { + if (mem->placement & TTM_PL_FLAG_WC) + addr = ioremap_wc(mem->bus.base + mem->bus.offset, mem->bus.size); + else + addr = ioremap_nocache(mem->bus.base + mem->bus.offset, mem->bus.size); + if (!addr) { + (void) ttm_mem_io_lock(man, false); + ttm_mem_io_free(bdev, mem); + ttm_mem_io_unlock(man); + return -ENOMEM; + } + } + *virtual = addr; + return 0; +} + +void ttm_mem_reg_iounmap(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem, + void *virtual) +{ + struct ttm_mem_type_manager *man; + + man = &bdev->man[mem->mem_type]; + + if (virtual && mem->bus.addr == NULL) + iounmap(virtual); + (void) ttm_mem_io_lock(man, false); + ttm_mem_io_free(bdev, mem); + ttm_mem_io_unlock(man); +} + +static int ttm_copy_io_page(void *dst, void *src, unsigned long page) +{ + uint32_t *dstP = + (uint32_t *) ((unsigned long)dst + (page << PAGE_SHIFT)); + uint32_t *srcP = + (uint32_t *) ((unsigned long)src + (page << PAGE_SHIFT)); + + int i; + for (i = 0; i < PAGE_SIZE / sizeof(uint32_t); ++i) + iowrite32(ioread32(srcP++), dstP++); + return 0; +} + +static int ttm_copy_io_ttm_page(struct ttm_tt *ttm, void *src, + unsigned long page, + pgprot_t prot) +{ + struct page *d = ttm->pages[page]; + void *dst; + + if (!d) + return -ENOMEM; + + src = (void *)((unsigned long)src + (page << PAGE_SHIFT)); + +#ifdef CONFIG_X86 + dst = kmap_atomic_prot(d, prot); +#else + if (pgprot_val(prot) != pgprot_val(PAGE_KERNEL)) + dst = vmap(&d, 1, 0, prot); + else + dst = kmap(d); +#endif + if (!dst) + return -ENOMEM; + + memcpy_fromio(dst, src, PAGE_SIZE); + +#ifdef CONFIG_X86 + kunmap_atomic(dst); +#else + if (pgprot_val(prot) != pgprot_val(PAGE_KERNEL)) + vunmap(dst); + else + kunmap(d); +#endif + + return 0; +} + +static int ttm_copy_ttm_io_page(struct ttm_tt *ttm, void *dst, + unsigned long page, + pgprot_t prot) +{ + struct page *s = ttm->pages[page]; + void *src; + + if (!s) + return -ENOMEM; + + dst = (void *)((unsigned long)dst + (page << PAGE_SHIFT)); +#ifdef CONFIG_X86 + src = kmap_atomic_prot(s, prot); +#else + if (pgprot_val(prot) != pgprot_val(PAGE_KERNEL)) + src = vmap(&s, 1, 0, prot); + else + src = kmap(s); +#endif + if (!src) + return -ENOMEM; + + memcpy_toio(dst, src, PAGE_SIZE); + +#ifdef CONFIG_X86 + kunmap_atomic(src); +#else + if (pgprot_val(prot) != pgprot_val(PAGE_KERNEL)) + vunmap(src); + else + kunmap(s); +#endif + + return 0; +} + +int ttm_bo_move_memcpy(struct ttm_buffer_object *bo, + bool evict, bool no_wait_gpu, + struct ttm_mem_reg *new_mem) +{ + struct ttm_bo_device *bdev = bo->bdev; + struct ttm_mem_type_manager *man = &bdev->man[new_mem->mem_type]; + struct ttm_tt *ttm = bo->ttm; + struct ttm_mem_reg *old_mem = &bo->mem; + struct ttm_mem_reg old_copy = *old_mem; + void *old_iomap; + void *new_iomap; + int ret; + unsigned long i; + unsigned long page; + unsigned long add = 0; + int dir; + + ret = ttm_mem_reg_ioremap(bdev, old_mem, &old_iomap); + if (ret) + return ret; + ret = ttm_mem_reg_ioremap(bdev, new_mem, &new_iomap); + if (ret) + goto out; + + if (old_iomap == NULL && new_iomap == NULL) + goto out2; + if (old_iomap == NULL && ttm == NULL) + goto out2; + + if (ttm->state == tt_unpopulated) { + ret = ttm->bdev->driver->ttm_tt_populate(ttm); + if (ret) { + /* if we fail here don't nuke the mm node + * as the bo still owns it */ + old_copy.mm_node = NULL; + goto out1; + } + } + + add = 0; + dir = 1; + + if ((old_mem->mem_type == new_mem->mem_type) && + (new_mem->start < old_mem->start + old_mem->size)) { + dir = -1; + add = new_mem->num_pages - 1; + } + + for (i = 0; i < new_mem->num_pages; ++i) { + page = i * dir + add; + if (old_iomap == NULL) { + pgprot_t prot = ttm_io_prot(old_mem->placement, + PAGE_KERNEL); + ret = ttm_copy_ttm_io_page(ttm, new_iomap, page, + prot); + } else if (new_iomap == NULL) { + pgprot_t prot = ttm_io_prot(new_mem->placement, + PAGE_KERNEL); + ret = ttm_copy_io_ttm_page(ttm, old_iomap, page, + prot); + } else + ret = ttm_copy_io_page(new_iomap, old_iomap, page); + if (ret) { + /* failing here, means keep old copy as-is */ + old_copy.mm_node = NULL; + goto out1; + } + } + mb(); +out2: + old_copy = *old_mem; + *old_mem = *new_mem; + new_mem->mm_node = NULL; + + if ((man->flags & TTM_MEMTYPE_FLAG_FIXED) && (ttm != NULL)) { + ttm_tt_unbind(ttm); + ttm_tt_destroy(ttm); + bo->ttm = NULL; + } + +out1: + ttm_mem_reg_iounmap(bdev, old_mem, new_iomap); +out: + ttm_mem_reg_iounmap(bdev, &old_copy, old_iomap); + ttm_bo_mem_put(bo, &old_copy); + return ret; +} +EXPORT_SYMBOL(ttm_bo_move_memcpy); + +static void ttm_transfered_destroy(struct ttm_buffer_object *bo) +{ + kfree(bo); +} + +/** + * ttm_buffer_object_transfer + * + * @bo: A pointer to a struct ttm_buffer_object. + * @new_obj: A pointer to a pointer to a newly created ttm_buffer_object, + * holding the data of @bo with the old placement. + * + * This is a utility function that may be called after an accelerated move + * has been scheduled. A new buffer object is created as a placeholder for + * the old data while it's being copied. When that buffer object is idle, + * it can be destroyed, releasing the space of the old placement. + * Returns: + * !0: Failure. + */ + +static int ttm_buffer_object_transfer(struct ttm_buffer_object *bo, + struct ttm_buffer_object **new_obj) +{ + struct ttm_buffer_object *fbo; + struct ttm_bo_device *bdev = bo->bdev; + struct ttm_bo_driver *driver = bdev->driver; + + fbo = kmalloc(sizeof(*fbo), GFP_KERNEL); + if (!fbo) + return -ENOMEM; + + *fbo = *bo; + + /** + * Fix up members that we shouldn't copy directly: + * TODO: Explicit member copy would probably be better here. + */ + + init_waitqueue_head(&fbo->event_queue); + INIT_LIST_HEAD(&fbo->ddestroy); + INIT_LIST_HEAD(&fbo->lru); + INIT_LIST_HEAD(&fbo->swap); + INIT_LIST_HEAD(&fbo->io_reserve_lru); + fbo->vm_node = NULL; + atomic_set(&fbo->cpu_writers, 0); + + spin_lock(&bdev->fence_lock); + if (bo->sync_obj) + fbo->sync_obj = driver->sync_obj_ref(bo->sync_obj); + else + fbo->sync_obj = NULL; + spin_unlock(&bdev->fence_lock); + kref_init(&fbo->list_kref); + kref_init(&fbo->kref); + fbo->destroy = &ttm_transfered_destroy; + fbo->acc_size = 0; + + *new_obj = fbo; + return 0; +} + +pgprot_t ttm_io_prot(uint32_t caching_flags, pgprot_t tmp) +{ +#if defined(__i386__) || defined(__x86_64__) + if (caching_flags & TTM_PL_FLAG_WC) + tmp = pgprot_writecombine(tmp); + else if (boot_cpu_data.x86 > 3) + tmp = pgprot_noncached(tmp); + +#elif defined(__powerpc__) + if (!(caching_flags & TTM_PL_FLAG_CACHED)) { + pgprot_val(tmp) |= _PAGE_NO_CACHE; + if (caching_flags & TTM_PL_FLAG_UNCACHED) + pgprot_val(tmp) |= _PAGE_GUARDED; + } +#endif +#if defined(__ia64__) + if (caching_flags & TTM_PL_FLAG_WC) + tmp = pgprot_writecombine(tmp); + else + tmp = pgprot_noncached(tmp); +#endif +#if defined(__sparc__) || defined(__mips__) + if (!(caching_flags & TTM_PL_FLAG_CACHED)) + tmp = pgprot_noncached(tmp); +#endif + return tmp; +} +EXPORT_SYMBOL(ttm_io_prot); + +static int ttm_bo_ioremap(struct ttm_buffer_object *bo, + unsigned long offset, + unsigned long size, + struct ttm_bo_kmap_obj *map) +{ + struct ttm_mem_reg *mem = &bo->mem; + + if (bo->mem.bus.addr) { + map->bo_kmap_type = ttm_bo_map_premapped; + map->virtual = (void *)(((u8 *)bo->mem.bus.addr) + offset); + } else { + map->bo_kmap_type = ttm_bo_map_iomap; + if (mem->placement & TTM_PL_FLAG_WC) + map->virtual = ioremap_wc(bo->mem.bus.base + bo->mem.bus.offset + offset, + size); + else + map->virtual = ioremap_nocache(bo->mem.bus.base + bo->mem.bus.offset + offset, + size); + } + return (!map->virtual) ? -ENOMEM : 0; +} + +static int ttm_bo_kmap_ttm(struct ttm_buffer_object *bo, + unsigned long start_page, + unsigned long num_pages, + struct ttm_bo_kmap_obj *map) +{ + struct ttm_mem_reg *mem = &bo->mem; pgprot_t prot; + struct ttm_tt *ttm = bo->ttm; + int ret; + + BUG_ON(!ttm); + + if (ttm->state == tt_unpopulated) { + ret = ttm->bdev->driver->ttm_tt_populate(ttm); + if (ret) + return ret; + } + + if (num_pages == 1 && (mem->placement & TTM_PL_FLAG_CACHED)) { + /* + * We're mapping a single page, and the desired + * page protection is consistent with the bo. + */ + + map->bo_kmap_type = ttm_bo_map_kmap; + map->page = ttm->pages[start_page]; + map->virtual = kmap(map->page); + } else { + /* + * We need to use vmap to get the desired page protection + * or to make the buffer object look contiguous. + */ + prot = (mem->placement & TTM_PL_FLAG_CACHED) ? + PAGE_KERNEL : + ttm_io_prot(mem->placement, PAGE_KERNEL); + map->bo_kmap_type = ttm_bo_map_vmap; + map->virtual = vmap(ttm->pages + start_page, num_pages, + 0, prot); + } + return (!map->virtual) ? -ENOMEM : 0; +} + +int ttm_bo_kmap(struct ttm_buffer_object *bo, + unsigned long start_page, unsigned long num_pages, + struct ttm_bo_kmap_obj *map) +{ + struct ttm_mem_type_manager *man = + &bo->bdev->man[bo->mem.mem_type]; + unsigned long offset, size; + int ret; + + BUG_ON(!list_empty(&bo->swap)); + map->virtual = NULL; + map->bo = bo; + if (num_pages > bo->num_pages) + return -EINVAL; + if (start_page > bo->num_pages) + return -EINVAL; +#if 0 + if (num_pages > 1 && !DRM_SUSER(DRM_CURPROC)) + return -EPERM; +#endif + (void) ttm_mem_io_lock(man, false); + ret = ttm_mem_io_reserve(bo->bdev, &bo->mem); + ttm_mem_io_unlock(man); + if (ret) + return ret; + if (!bo->mem.bus.is_iomem) { + return ttm_bo_kmap_ttm(bo, start_page, num_pages, map); + } else { + offset = start_page << PAGE_SHIFT; + size = num_pages << PAGE_SHIFT; + return ttm_bo_ioremap(bo, offset, size, map); + } +} +EXPORT_SYMBOL(ttm_bo_kmap); + +void ttm_bo_kunmap(struct ttm_bo_kmap_obj *map) +{ + struct ttm_buffer_object *bo = map->bo; + struct ttm_mem_type_manager *man = + &bo->bdev->man[bo->mem.mem_type]; + + if (!map->virtual) + return; + switch (map->bo_kmap_type) { + case ttm_bo_map_iomap: + iounmap(map->virtual); + break; + case ttm_bo_map_vmap: + vunmap(map->virtual); + break; + case ttm_bo_map_kmap: + kunmap(map->page); + break; + case ttm_bo_map_premapped: + break; + default: + BUG(); + } + (void) ttm_mem_io_lock(man, false); + ttm_mem_io_free(map->bo->bdev, &map->bo->mem); + ttm_mem_io_unlock(man); + map->virtual = NULL; + map->page = NULL; +} +EXPORT_SYMBOL(ttm_bo_kunmap); + +int ttm_bo_move_accel_cleanup(struct ttm_buffer_object *bo, + void *sync_obj, + bool evict, + bool no_wait_gpu, + struct ttm_mem_reg *new_mem) +{ + struct ttm_bo_device *bdev = bo->bdev; + struct ttm_bo_driver *driver = bdev->driver; + struct ttm_mem_type_manager *man = &bdev->man[new_mem->mem_type]; + struct ttm_mem_reg *old_mem = &bo->mem; + int ret; + struct ttm_buffer_object *ghost_obj; + void *tmp_obj = NULL; + + spin_lock(&bdev->fence_lock); + if (bo->sync_obj) { + tmp_obj = bo->sync_obj; + bo->sync_obj = NULL; + } + bo->sync_obj = driver->sync_obj_ref(sync_obj); + if (evict) { + ret = ttm_bo_wait(bo, false, false, false); + spin_unlock(&bdev->fence_lock); + if (tmp_obj) + driver->sync_obj_unref(&tmp_obj); + if (ret) + return ret; + + if ((man->flags & TTM_MEMTYPE_FLAG_FIXED) && + (bo->ttm != NULL)) { + ttm_tt_unbind(bo->ttm); + ttm_tt_destroy(bo->ttm); + bo->ttm = NULL; + } + ttm_bo_free_old_node(bo); + } else { + /** + * This should help pipeline ordinary buffer moves. + * + * Hang old buffer memory on a new buffer object, + * and leave it to be released when the GPU + * operation has completed. + */ + + set_bit(TTM_BO_PRIV_FLAG_MOVING, &bo->priv_flags); + spin_unlock(&bdev->fence_lock); + if (tmp_obj) + driver->sync_obj_unref(&tmp_obj); + + ret = ttm_buffer_object_transfer(bo, &ghost_obj); + if (ret) + return ret; + + /** + * If we're not moving to fixed memory, the TTM object + * needs to stay alive. Otherwhise hang it on the ghost + * bo to be unbound and destroyed. + */ + + if (!(man->flags & TTM_MEMTYPE_FLAG_FIXED)) + ghost_obj->ttm = NULL; + else + bo->ttm = NULL; + + ttm_bo_unreserve(ghost_obj); + ttm_bo_unref(&ghost_obj); + } + + *old_mem = *new_mem; + new_mem->mm_node = NULL; + + return 0; +} +EXPORT_SYMBOL(ttm_bo_move_accel_cleanup); diff --git a/drivers/gpu/drm/ttm/ttm_bo_vm.c b/drivers/gpu/drm/ttm/ttm_bo_vm.c new file mode 100644 index 0000000..3df9f16 --- /dev/null +++ b/drivers/gpu/drm/ttm/ttm_bo_vm.c @@ -0,0 +1,465 @@ +/************************************************************************** + * + * Copyright (c) 2006-2009 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 <thellstrom-at-vmware-dot-com> + */ + +#define pr_fmt(fmt) "[TTM] " fmt + +#include <ttm/ttm_module.h> +#include <ttm/ttm_bo_driver.h> +#include <ttm/ttm_placement.h> +#include <linux/mm.h> +#include <linux/rbtree.h> +#include <linux/module.h> +#include <linux/uaccess.h> + +#define TTM_BO_VM_NUM_PREFAULT 16 + +static struct ttm_buffer_object *ttm_bo_vm_lookup_rb(struct ttm_bo_device *bdev, + unsigned long page_start, + unsigned long num_pages) +{ + struct rb_node *cur = bdev->addr_space_rb.rb_node; + unsigned long cur_offset; + struct ttm_buffer_object *bo; + struct ttm_buffer_object *best_bo = NULL; + + while (likely(cur != NULL)) { + bo = rb_entry(cur, struct ttm_buffer_object, vm_rb); + cur_offset = bo->vm_node->start; + if (page_start >= cur_offset) { + cur = cur->rb_right; + best_bo = bo; + if (page_start == cur_offset) + break; + } else + cur = cur->rb_left; + } + + if (unlikely(best_bo == NULL)) + return NULL; + + if (unlikely((best_bo->vm_node->start + best_bo->num_pages) < + (page_start + num_pages))) + return NULL; + + return best_bo; +} + +static int ttm_bo_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf) +{ + struct ttm_buffer_object *bo = (struct ttm_buffer_object *) + vma->vm_private_data; + struct ttm_bo_device *bdev = bo->bdev; + unsigned long page_offset; + unsigned long page_last; + unsigned long pfn; + struct ttm_tt *ttm = NULL; + struct page *page; + int ret; + int i; + unsigned long address = (unsigned long)vmf->virtual_address; + int retval = VM_FAULT_NOPAGE; + struct ttm_mem_type_manager *man = + &bdev->man[bo->mem.mem_type]; + + /* + * Work around locking order reversal in fault / nopfn + * between mmap_sem and bo_reserve: Perform a trylock operation + * for reserve, and if it fails, retry the fault after scheduling. + */ + + ret = ttm_bo_reserve(bo, true, true, false, 0); + if (unlikely(ret != 0)) { + if (ret == -EBUSY) + set_need_resched(); + return VM_FAULT_NOPAGE; + } + + if (bdev->driver->fault_reserve_notify) { + ret = bdev->driver->fault_reserve_notify(bo); + switch (ret) { + case 0: + break; + case -EBUSY: + set_need_resched(); + case -ERESTARTSYS: + retval = VM_FAULT_NOPAGE; + goto out_unlock; + default: + retval = VM_FAULT_SIGBUS; + goto out_unlock; + } + } + + /* + * Wait for buffer data in transit, due to a pipelined + * move. + */ + + spin_lock(&bdev->fence_lock); + if (test_bit(TTM_BO_PRIV_FLAG_MOVING, &bo->priv_flags)) { + ret = ttm_bo_wait(bo, false, true, false); + spin_unlock(&bdev->fence_lock); + if (unlikely(ret != 0)) { + retval = (ret != -ERESTARTSYS) ? + VM_FAULT_SIGBUS : VM_FAULT_NOPAGE; + goto out_unlock; + } + } else + spin_unlock(&bdev->fence_lock); + + ret = ttm_mem_io_lock(man, true); + if (unlikely(ret != 0)) { + retval = VM_FAULT_NOPAGE; + goto out_unlock; + } + ret = ttm_mem_io_reserve_vm(bo); + if (unlikely(ret != 0)) { + retval = VM_FAULT_SIGBUS; + goto out_io_unlock; + } + + page_offset = ((address - vma->vm_start) >> PAGE_SHIFT) + + bo->vm_node->start - vma->vm_pgoff; + page_last = vma_pages(vma) + + bo->vm_node->start - vma->vm_pgoff; + + if (unlikely(page_offset >= bo->num_pages)) { + retval = VM_FAULT_SIGBUS; + goto out_io_unlock; + } + + /* + * Strictly, we're not allowed to modify vma->vm_page_prot here, + * since the mmap_sem is only held in read mode. However, we + * modify only the caching bits of vma->vm_page_prot and + * consider those bits protected by + * the bo->mutex, as we should be the only writers. + * There shouldn't really be any readers of these bits except + * within vm_insert_mixed()? fork? + * + * TODO: Add a list of vmas to the bo, and change the + * vma->vm_page_prot when the object changes caching policy, with + * the correct locks held. + */ + if (bo->mem.bus.is_iomem) { + vma->vm_page_prot = ttm_io_prot(bo->mem.placement, + vma->vm_page_prot); + } else { + ttm = bo->ttm; + vma->vm_page_prot = (bo->mem.placement & TTM_PL_FLAG_CACHED) ? + vm_get_page_prot(vma->vm_flags) : + ttm_io_prot(bo->mem.placement, vma->vm_page_prot); + + /* Allocate all page at once, most common usage */ + if (ttm->bdev->driver->ttm_tt_populate(ttm)) { + retval = VM_FAULT_OOM; + goto out_io_unlock; + } + } + + /* + * Speculatively prefault a number of pages. Only error on + * first page. + */ + for (i = 0; i < TTM_BO_VM_NUM_PREFAULT; ++i) { + if (bo->mem.bus.is_iomem) + pfn = ((bo->mem.bus.base + bo->mem.bus.offset) >> PAGE_SHIFT) + page_offset; + else { + page = ttm->pages[page_offset]; + if (unlikely(!page && i == 0)) { + retval = VM_FAULT_OOM; + goto out_io_unlock; + } else if (unlikely(!page)) { + break; + } + pfn = page_to_pfn(page); + } + + ret = vm_insert_mixed(vma, address, pfn); + /* + * Somebody beat us to this PTE or prefaulting to + * an already populated PTE, or prefaulting error. + */ + + if (unlikely((ret == -EBUSY) || (ret != 0 && i > 0))) + break; + else if (unlikely(ret != 0)) { + retval = + (ret == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS; + goto out_io_unlock; + } + + address += PAGE_SIZE; + if (unlikely(++page_offset >= page_last)) + break; + } +out_io_unlock: + ttm_mem_io_unlock(man); +out_unlock: + ttm_bo_unreserve(bo); + return retval; +} + +static void ttm_bo_vm_open(struct vm_area_struct *vma) +{ + struct ttm_buffer_object *bo = + (struct ttm_buffer_object *)vma->vm_private_data; + + (void)ttm_bo_reference(bo); +} + +static void ttm_bo_vm_close(struct vm_area_struct *vma) +{ + struct ttm_buffer_object *bo = (struct ttm_buffer_object *)vma->vm_private_data; + + ttm_bo_unref(&bo); + vma->vm_private_data = NULL; +} + +static const struct vm_operations_struct ttm_bo_vm_ops = { + .fault = ttm_bo_vm_fault, + .open = ttm_bo_vm_open, + .close = ttm_bo_vm_close +}; + +int ttm_bo_mmap(struct file *filp, struct vm_area_struct *vma, + struct ttm_bo_device *bdev) +{ + struct ttm_bo_driver *driver; + struct ttm_buffer_object *bo; + int ret; + + read_lock(&bdev->vm_lock); + bo = ttm_bo_vm_lookup_rb(bdev, vma->vm_pgoff, + vma_pages(vma)); + if (likely(bo != NULL) && !kref_get_unless_zero(&bo->kref)) + bo = NULL; + read_unlock(&bdev->vm_lock); + + if (unlikely(bo == NULL)) { + pr_err("Could not find buffer object to map\n"); + return -EINVAL; + } + + driver = bo->bdev->driver; + if (unlikely(!driver->verify_access)) { + ret = -EPERM; + goto out_unref; + } + ret = driver->verify_access(bo, filp); + if (unlikely(ret != 0)) + goto out_unref; + + vma->vm_ops = &ttm_bo_vm_ops; + + /* + * Note: We're transferring the bo reference to + * vma->vm_private_data here. + */ + + vma->vm_private_data = bo; + vma->vm_flags |= VM_IO | VM_MIXEDMAP | VM_DONTEXPAND | VM_DONTDUMP; + return 0; +out_unref: + ttm_bo_unref(&bo); + return ret; +} +EXPORT_SYMBOL(ttm_bo_mmap); + +int ttm_fbdev_mmap(struct vm_area_struct *vma, struct ttm_buffer_object *bo) +{ + if (vma->vm_pgoff != 0) + return -EACCES; + + vma->vm_ops = &ttm_bo_vm_ops; + vma->vm_private_data = ttm_bo_reference(bo); + vma->vm_flags |= VM_IO | VM_MIXEDMAP | VM_DONTEXPAND; + return 0; +} +EXPORT_SYMBOL(ttm_fbdev_mmap); + + +ssize_t ttm_bo_io(struct ttm_bo_device *bdev, struct file *filp, + const char __user *wbuf, char __user *rbuf, size_t count, + loff_t *f_pos, bool write) +{ + struct ttm_buffer_object *bo; + struct ttm_bo_driver *driver; + struct ttm_bo_kmap_obj map; + unsigned long dev_offset = (*f_pos >> PAGE_SHIFT); + unsigned long kmap_offset; + unsigned long kmap_end; + unsigned long kmap_num; + size_t io_size; + unsigned int page_offset; + char *virtual; + int ret; + bool no_wait = false; + bool dummy; + + read_lock(&bdev->vm_lock); + bo = ttm_bo_vm_lookup_rb(bdev, dev_offset, 1); + if (likely(bo != NULL)) + ttm_bo_reference(bo); + read_unlock(&bdev->vm_lock); + + if (unlikely(bo == NULL)) + return -EFAULT; + + driver = bo->bdev->driver; + if (unlikely(!driver->verify_access)) { + ret = -EPERM; + goto out_unref; + } + + ret = driver->verify_access(bo, filp); + if (unlikely(ret != 0)) + goto out_unref; + + kmap_offset = dev_offset - bo->vm_node->start; + if (unlikely(kmap_offset >= bo->num_pages)) { + ret = -EFBIG; + goto out_unref; + } + + page_offset = *f_pos & ~PAGE_MASK; + io_size = bo->num_pages - kmap_offset; + io_size = (io_size << PAGE_SHIFT) - page_offset; + if (count < io_size) + io_size = count; + + kmap_end = (*f_pos + count - 1) >> PAGE_SHIFT; + kmap_num = kmap_end - kmap_offset + 1; + + ret = ttm_bo_reserve(bo, true, no_wait, false, 0); + + switch (ret) { + case 0: + break; + case -EBUSY: + ret = -EAGAIN; + goto out_unref; + default: + goto out_unref; + } + + ret = ttm_bo_kmap(bo, kmap_offset, kmap_num, &map); + if (unlikely(ret != 0)) { + ttm_bo_unreserve(bo); + goto out_unref; + } + + virtual = ttm_kmap_obj_virtual(&map, &dummy); + virtual += page_offset; + + if (write) + ret = copy_from_user(virtual, wbuf, io_size); + else + ret = copy_to_user(rbuf, virtual, io_size); + + ttm_bo_kunmap(&map); + ttm_bo_unreserve(bo); + ttm_bo_unref(&bo); + + if (unlikely(ret != 0)) + return -EFBIG; + + *f_pos += io_size; + + return io_size; +out_unref: + ttm_bo_unref(&bo); + return ret; +} + +ssize_t ttm_bo_fbdev_io(struct ttm_buffer_object *bo, const char __user *wbuf, + char __user *rbuf, size_t count, loff_t *f_pos, + bool write) +{ + struct ttm_bo_kmap_obj map; + unsigned long kmap_offset; + unsigned long kmap_end; + unsigned long kmap_num; + size_t io_size; + unsigned int page_offset; + char *virtual; + int ret; + bool no_wait = false; + bool dummy; + + kmap_offset = (*f_pos >> PAGE_SHIFT); + if (unlikely(kmap_offset >= bo->num_pages)) + return -EFBIG; + + page_offset = *f_pos & ~PAGE_MASK; + io_size = bo->num_pages - kmap_offset; + io_size = (io_size << PAGE_SHIFT) - page_offset; + if (count < io_size) + io_size = count; + + kmap_end = (*f_pos + count - 1) >> PAGE_SHIFT; + kmap_num = kmap_end - kmap_offset + 1; + + ret = ttm_bo_reserve(bo, true, no_wait, false, 0); + + switch (ret) { + case 0: + break; + case -EBUSY: + return -EAGAIN; + default: + return ret; + } + + ret = ttm_bo_kmap(bo, kmap_offset, kmap_num, &map); + if (unlikely(ret != 0)) { + ttm_bo_unreserve(bo); + return ret; + } + + virtual = ttm_kmap_obj_virtual(&map, &dummy); + virtual += page_offset; + + if (write) + ret = copy_from_user(virtual, wbuf, io_size); + else + ret = copy_to_user(rbuf, virtual, io_size); + + ttm_bo_kunmap(&map); + ttm_bo_unreserve(bo); + ttm_bo_unref(&bo); + + if (unlikely(ret != 0)) + return ret; + + *f_pos += io_size; + + return io_size; +} diff --git a/drivers/gpu/drm/ttm/ttm_execbuf_util.c b/drivers/gpu/drm/ttm/ttm_execbuf_util.c new file mode 100644 index 0000000..7b90def --- /dev/null +++ b/drivers/gpu/drm/ttm/ttm_execbuf_util.c @@ -0,0 +1,242 @@ +/************************************************************************** + * + * Copyright (c) 2006-2009 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. + * + **************************************************************************/ + +#include <drm/ttm/ttm_execbuf_util.h> +#include <drm/ttm/ttm_bo_driver.h> +#include <drm/ttm/ttm_placement.h> +#include <linux/wait.h> +#include <linux/sched.h> +#include <linux/module.h> + +static void ttm_eu_backoff_reservation_locked(struct list_head *list) +{ + struct ttm_validate_buffer *entry; + + list_for_each_entry(entry, list, head) { + struct ttm_buffer_object *bo = entry->bo; + if (!entry->reserved) + continue; + + if (entry->removed) { + ttm_bo_add_to_lru(bo); + entry->removed = false; + + } + entry->reserved = false; + atomic_set(&bo->reserved, 0); + wake_up_all(&bo->event_queue); + } +} + +static void ttm_eu_del_from_lru_locked(struct list_head *list) +{ + struct ttm_validate_buffer *entry; + + list_for_each_entry(entry, list, head) { + struct ttm_buffer_object *bo = entry->bo; + if (!entry->reserved) + continue; + + if (!entry->removed) { + entry->put_count = ttm_bo_del_from_lru(bo); + entry->removed = true; + } + } +} + +static void ttm_eu_list_ref_sub(struct list_head *list) +{ + struct ttm_validate_buffer *entry; + + list_for_each_entry(entry, list, head) { + struct ttm_buffer_object *bo = entry->bo; + + if (entry->put_count) { + ttm_bo_list_ref_sub(bo, entry->put_count, true); + entry->put_count = 0; + } + } +} + +void ttm_eu_backoff_reservation(struct list_head *list) +{ + struct ttm_validate_buffer *entry; + struct ttm_bo_global *glob; + + if (list_empty(list)) + return; + + entry = list_first_entry(list, struct ttm_validate_buffer, head); + glob = entry->bo->glob; + spin_lock(&glob->lru_lock); + ttm_eu_backoff_reservation_locked(list); + spin_unlock(&glob->lru_lock); +} +EXPORT_SYMBOL(ttm_eu_backoff_reservation); + +/* + * Reserve buffers for validation. + * + * If a buffer in the list is marked for CPU access, we back off and + * wait for that buffer to become free for GPU access. + * + * If a buffer is reserved for another validation, the validator with + * the highest validation sequence backs off and waits for that buffer + * to become unreserved. This prevents deadlocks when validating multiple + * buffers in different orders. + */ + +int ttm_eu_reserve_buffers(struct list_head *list) +{ + struct ttm_bo_global *glob; + struct ttm_validate_buffer *entry; + int ret; + uint32_t val_seq; + + if (list_empty(list)) + return 0; + + list_for_each_entry(entry, list, head) { + entry->reserved = false; + entry->put_count = 0; + entry->removed = false; + } + + entry = list_first_entry(list, struct ttm_validate_buffer, head); + glob = entry->bo->glob; + + spin_lock(&glob->lru_lock); + val_seq = entry->bo->bdev->val_seq++; + +retry: + list_for_each_entry(entry, list, head) { + struct ttm_buffer_object *bo = entry->bo; + + /* already slowpath reserved? */ + if (entry->reserved) + continue; + + ret = ttm_bo_reserve_nolru(bo, true, true, true, val_seq); + switch (ret) { + case 0: + break; + case -EBUSY: + ttm_eu_del_from_lru_locked(list); + spin_unlock(&glob->lru_lock); + ret = ttm_bo_reserve_nolru(bo, true, false, + true, val_seq); + spin_lock(&glob->lru_lock); + if (!ret) + break; + + if (unlikely(ret != -EAGAIN)) + goto err; + + /* fallthrough */ + case -EAGAIN: + ttm_eu_backoff_reservation_locked(list); + + /* + * temporarily increase sequence number every retry, + * to prevent us from seeing our old reservation + * sequence when someone else reserved the buffer, + * but hasn't updated the seq_valid/seqno members yet. + */ + val_seq = entry->bo->bdev->val_seq++; + + spin_unlock(&glob->lru_lock); + ttm_eu_list_ref_sub(list); + ret = ttm_bo_reserve_slowpath_nolru(bo, true, val_seq); + if (unlikely(ret != 0)) + return ret; + spin_lock(&glob->lru_lock); + entry->reserved = true; + if (unlikely(atomic_read(&bo->cpu_writers) > 0)) { + ret = -EBUSY; + goto err; + } + goto retry; + default: + goto err; + } + + entry->reserved = true; + if (unlikely(atomic_read(&bo->cpu_writers) > 0)) { + ret = -EBUSY; + goto err; + } + } + + ttm_eu_del_from_lru_locked(list); + spin_unlock(&glob->lru_lock); + ttm_eu_list_ref_sub(list); + + return 0; + +err: + ttm_eu_backoff_reservation_locked(list); + spin_unlock(&glob->lru_lock); + ttm_eu_list_ref_sub(list); + return ret; +} +EXPORT_SYMBOL(ttm_eu_reserve_buffers); + +void ttm_eu_fence_buffer_objects(struct list_head *list, void *sync_obj) +{ + struct ttm_validate_buffer *entry; + struct ttm_buffer_object *bo; + struct ttm_bo_global *glob; + struct ttm_bo_device *bdev; + struct ttm_bo_driver *driver; + + if (list_empty(list)) + return; + + bo = list_first_entry(list, struct ttm_validate_buffer, head)->bo; + bdev = bo->bdev; + driver = bdev->driver; + glob = bo->glob; + + spin_lock(&glob->lru_lock); + spin_lock(&bdev->fence_lock); + + list_for_each_entry(entry, list, head) { + bo = entry->bo; + entry->old_sync_obj = bo->sync_obj; + bo->sync_obj = driver->sync_obj_ref(sync_obj); + ttm_bo_unreserve_locked(bo); + entry->reserved = false; + } + spin_unlock(&bdev->fence_lock); + spin_unlock(&glob->lru_lock); + + list_for_each_entry(entry, list, head) { + if (entry->old_sync_obj) + driver->sync_obj_unref(&entry->old_sync_obj); + } +} +EXPORT_SYMBOL(ttm_eu_fence_buffer_objects); diff --git a/drivers/gpu/drm/ttm/ttm_lock.c b/drivers/gpu/drm/ttm/ttm_lock.c new file mode 100644 index 0000000..3daa9a3 --- /dev/null +++ b/drivers/gpu/drm/ttm/ttm_lock.c @@ -0,0 +1,310 @@ +/************************************************************************** + * + * Copyright (c) 2007-2009 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 <thellstrom-at-vmware-dot-com> + */ + +#include <drm/ttm/ttm_lock.h> +#include <drm/ttm/ttm_module.h> +#include <linux/atomic.h> +#include <linux/errno.h> +#include <linux/wait.h> +#include <linux/sched.h> +#include <linux/module.h> + +#define TTM_WRITE_LOCK_PENDING (1 << 0) +#define TTM_VT_LOCK_PENDING (1 << 1) +#define TTM_SUSPEND_LOCK_PENDING (1 << 2) +#define TTM_VT_LOCK (1 << 3) +#define TTM_SUSPEND_LOCK (1 << 4) + +void ttm_lock_init(struct ttm_lock *lock) +{ + spin_lock_init(&lock->lock); + init_waitqueue_head(&lock->queue); + lock->rw = 0; + lock->flags = 0; + lock->kill_takers = false; + lock->signal = SIGKILL; +} +EXPORT_SYMBOL(ttm_lock_init); + +void ttm_read_unlock(struct ttm_lock *lock) +{ + spin_lock(&lock->lock); + if (--lock->rw == 0) + wake_up_all(&lock->queue); + spin_unlock(&lock->lock); +} +EXPORT_SYMBOL(ttm_read_unlock); + +static bool __ttm_read_lock(struct ttm_lock *lock) +{ + bool locked = false; + + spin_lock(&lock->lock); + if (unlikely(lock->kill_takers)) { + send_sig(lock->signal, current, 0); + spin_unlock(&lock->lock); + return false; + } + if (lock->rw >= 0 && lock->flags == 0) { + ++lock->rw; + locked = true; + } + spin_unlock(&lock->lock); + return locked; +} + +int ttm_read_lock(struct ttm_lock *lock, bool interruptible) +{ + int ret = 0; + + if (interruptible) + ret = wait_event_interruptible(lock->queue, + __ttm_read_lock(lock)); + else + wait_event(lock->queue, __ttm_read_lock(lock)); + return ret; +} +EXPORT_SYMBOL(ttm_read_lock); + +static bool __ttm_read_trylock(struct ttm_lock *lock, bool *locked) +{ + bool block = true; + + *locked = false; + + spin_lock(&lock->lock); + if (unlikely(lock->kill_takers)) { + send_sig(lock->signal, current, 0); + spin_unlock(&lock->lock); + return false; + } + if (lock->rw >= 0 && lock->flags == 0) { + ++lock->rw; + block = false; + *locked = true; + } else if (lock->flags == 0) { + block = false; + } + spin_unlock(&lock->lock); + + return !block; +} + +int ttm_read_trylock(struct ttm_lock *lock, bool interruptible) +{ + int ret = 0; + bool locked; + + if (interruptible) + ret = wait_event_interruptible + (lock->queue, __ttm_read_trylock(lock, &locked)); + else + wait_event(lock->queue, __ttm_read_trylock(lock, &locked)); + + if (unlikely(ret != 0)) { + BUG_ON(locked); + return ret; + } + + return (locked) ? 0 : -EBUSY; +} + +void ttm_write_unlock(struct ttm_lock *lock) +{ + spin_lock(&lock->lock); + lock->rw = 0; + wake_up_all(&lock->queue); + spin_unlock(&lock->lock); +} +EXPORT_SYMBOL(ttm_write_unlock); + +static bool __ttm_write_lock(struct ttm_lock *lock) +{ + bool locked = false; + + spin_lock(&lock->lock); + if (unlikely(lock->kill_takers)) { + send_sig(lock->signal, current, 0); + spin_unlock(&lock->lock); + return false; + } + if (lock->rw == 0 && ((lock->flags & ~TTM_WRITE_LOCK_PENDING) == 0)) { + lock->rw = -1; + lock->flags &= ~TTM_WRITE_LOCK_PENDING; + locked = true; + } else { + lock->flags |= TTM_WRITE_LOCK_PENDING; + } + spin_unlock(&lock->lock); + return locked; +} + +int ttm_write_lock(struct ttm_lock *lock, bool interruptible) +{ + int ret = 0; + + if (interruptible) { + ret = wait_event_interruptible(lock->queue, + __ttm_write_lock(lock)); + if (unlikely(ret != 0)) { + spin_lock(&lock->lock); + lock->flags &= ~TTM_WRITE_LOCK_PENDING; + wake_up_all(&lock->queue); + spin_unlock(&lock->lock); + } + } else + wait_event(lock->queue, __ttm_read_lock(lock)); + + return ret; +} +EXPORT_SYMBOL(ttm_write_lock); + +void ttm_write_lock_downgrade(struct ttm_lock *lock) +{ + spin_lock(&lock->lock); + lock->rw = 1; + wake_up_all(&lock->queue); + spin_unlock(&lock->lock); +} + +static int __ttm_vt_unlock(struct ttm_lock *lock) +{ + int ret = 0; + + spin_lock(&lock->lock); + if (unlikely(!(lock->flags & TTM_VT_LOCK))) + ret = -EINVAL; + lock->flags &= ~TTM_VT_LOCK; + wake_up_all(&lock->queue); + spin_unlock(&lock->lock); + + return ret; +} + +static void ttm_vt_lock_remove(struct ttm_base_object **p_base) +{ + struct ttm_base_object *base = *p_base; + struct ttm_lock *lock = container_of(base, struct ttm_lock, base); + int ret; + + *p_base = NULL; + ret = __ttm_vt_unlock(lock); + BUG_ON(ret != 0); +} + +static bool __ttm_vt_lock(struct ttm_lock *lock) +{ + bool locked = false; + + spin_lock(&lock->lock); + if (lock->rw == 0) { + lock->flags &= ~TTM_VT_LOCK_PENDING; + lock->flags |= TTM_VT_LOCK; + locked = true; + } else { + lock->flags |= TTM_VT_LOCK_PENDING; + } + spin_unlock(&lock->lock); + return locked; +} + +int ttm_vt_lock(struct ttm_lock *lock, + bool interruptible, + struct ttm_object_file *tfile) +{ + int ret = 0; + + if (interruptible) { + ret = wait_event_interruptible(lock->queue, + __ttm_vt_lock(lock)); + if (unlikely(ret != 0)) { + spin_lock(&lock->lock); + lock->flags &= ~TTM_VT_LOCK_PENDING; + wake_up_all(&lock->queue); + spin_unlock(&lock->lock); + return ret; + } + } else + wait_event(lock->queue, __ttm_vt_lock(lock)); + + /* + * Add a base-object, the destructor of which will + * make sure the lock is released if the client dies + * while holding it. + */ + + ret = ttm_base_object_init(tfile, &lock->base, false, + ttm_lock_type, &ttm_vt_lock_remove, NULL); + if (ret) + (void)__ttm_vt_unlock(lock); + else + lock->vt_holder = tfile; + + return ret; +} +EXPORT_SYMBOL(ttm_vt_lock); + +int ttm_vt_unlock(struct ttm_lock *lock) +{ + return ttm_ref_object_base_unref(lock->vt_holder, + lock->base.hash.key, TTM_REF_USAGE); +} +EXPORT_SYMBOL(ttm_vt_unlock); + +void ttm_suspend_unlock(struct ttm_lock *lock) +{ + spin_lock(&lock->lock); + lock->flags &= ~TTM_SUSPEND_LOCK; + wake_up_all(&lock->queue); + spin_unlock(&lock->lock); +} +EXPORT_SYMBOL(ttm_suspend_unlock); + +static bool __ttm_suspend_lock(struct ttm_lock *lock) +{ + bool locked = false; + + spin_lock(&lock->lock); + if (lock->rw == 0) { + lock->flags &= ~TTM_SUSPEND_LOCK_PENDING; + lock->flags |= TTM_SUSPEND_LOCK; + locked = true; + } else { + lock->flags |= TTM_SUSPEND_LOCK_PENDING; + } + spin_unlock(&lock->lock); + return locked; +} + +void ttm_suspend_lock(struct ttm_lock *lock) +{ + wait_event(lock->queue, __ttm_suspend_lock(lock)); +} +EXPORT_SYMBOL(ttm_suspend_lock); diff --git a/drivers/gpu/drm/ttm/ttm_memory.c b/drivers/gpu/drm/ttm/ttm_memory.c new file mode 100644 index 0000000..dbc2def --- /dev/null +++ b/drivers/gpu/drm/ttm/ttm_memory.c @@ -0,0 +1,601 @@ +/************************************************************************** + * + * Copyright (c) 2006-2009 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. + * + **************************************************************************/ + +#define pr_fmt(fmt) "[TTM] " fmt + +#include <drm/ttm/ttm_memory.h> +#include <drm/ttm/ttm_module.h> +#include <drm/ttm/ttm_page_alloc.h> +#include <linux/spinlock.h> +#include <linux/sched.h> +#include <linux/wait.h> +#include <linux/mm.h> +#include <linux/module.h> +#include <linux/slab.h> + +#define TTM_MEMORY_ALLOC_RETRIES 4 + +struct ttm_mem_zone { + struct kobject kobj; + struct ttm_mem_global *glob; + const char *name; + uint64_t zone_mem; + uint64_t emer_mem; + uint64_t max_mem; + uint64_t swap_limit; + uint64_t used_mem; +}; + +static struct attribute ttm_mem_sys = { + .name = "zone_memory", + .mode = S_IRUGO +}; +static struct attribute ttm_mem_emer = { + .name = "emergency_memory", + .mode = S_IRUGO | S_IWUSR +}; +static struct attribute ttm_mem_max = { + .name = "available_memory", + .mode = S_IRUGO | S_IWUSR +}; +static struct attribute ttm_mem_swap = { + .name = "swap_limit", + .mode = S_IRUGO | S_IWUSR +}; +static struct attribute ttm_mem_used = { + .name = "used_memory", + .mode = S_IRUGO +}; + +static void ttm_mem_zone_kobj_release(struct kobject *kobj) +{ + struct ttm_mem_zone *zone = + container_of(kobj, struct ttm_mem_zone, kobj); + + pr_info("Zone %7s: Used memory at exit: %llu kiB\n", + zone->name, (unsigned long long)zone->used_mem >> 10); + kfree(zone); +} + +static ssize_t ttm_mem_zone_show(struct kobject *kobj, + struct attribute *attr, + char *buffer) +{ + struct ttm_mem_zone *zone = + container_of(kobj, struct ttm_mem_zone, kobj); + uint64_t val = 0; + + spin_lock(&zone->glob->lock); + if (attr == &ttm_mem_sys) + val = zone->zone_mem; + else if (attr == &ttm_mem_emer) + val = zone->emer_mem; + else if (attr == &ttm_mem_max) + val = zone->max_mem; + else if (attr == &ttm_mem_swap) + val = zone->swap_limit; + else if (attr == &ttm_mem_used) + val = zone->used_mem; + spin_unlock(&zone->glob->lock); + + return snprintf(buffer, PAGE_SIZE, "%llu\n", + (unsigned long long) val >> 10); +} + +static void ttm_check_swapping(struct ttm_mem_global *glob); + +static ssize_t ttm_mem_zone_store(struct kobject *kobj, + struct attribute *attr, + const char *buffer, + size_t size) +{ + struct ttm_mem_zone *zone = + container_of(kobj, struct ttm_mem_zone, kobj); + int chars; + unsigned long val; + uint64_t val64; + + chars = sscanf(buffer, "%lu", &val); + if (chars == 0) + return size; + + val64 = val; + val64 <<= 10; + + spin_lock(&zone->glob->lock); + if (val64 > zone->zone_mem) + val64 = zone->zone_mem; + if (attr == &ttm_mem_emer) { + zone->emer_mem = val64; + if (zone->max_mem > val64) + zone->max_mem = val64; + } else if (attr == &ttm_mem_max) { + zone->max_mem = val64; + if (zone->emer_mem < val64) + zone->emer_mem = val64; + } else if (attr == &ttm_mem_swap) + zone->swap_limit = val64; + spin_unlock(&zone->glob->lock); + + ttm_check_swapping(zone->glob); + + return size; +} + +static struct attribute *ttm_mem_zone_attrs[] = { + &ttm_mem_sys, + &ttm_mem_emer, + &ttm_mem_max, + &ttm_mem_swap, + &ttm_mem_used, + NULL +}; + +static const struct sysfs_ops ttm_mem_zone_ops = { + .show = &ttm_mem_zone_show, + .store = &ttm_mem_zone_store +}; + +static struct kobj_type ttm_mem_zone_kobj_type = { + .release = &ttm_mem_zone_kobj_release, + .sysfs_ops = &ttm_mem_zone_ops, + .default_attrs = ttm_mem_zone_attrs, +}; + +static void ttm_mem_global_kobj_release(struct kobject *kobj) +{ + struct ttm_mem_global *glob = + container_of(kobj, struct ttm_mem_global, kobj); + + kfree(glob); +} + +static struct kobj_type ttm_mem_glob_kobj_type = { + .release = &ttm_mem_global_kobj_release, +}; + +static bool ttm_zones_above_swap_target(struct ttm_mem_global *glob, + bool from_wq, uint64_t extra) +{ + unsigned int i; + struct ttm_mem_zone *zone; + uint64_t target; + + for (i = 0; i < glob->num_zones; ++i) { + zone = glob->zones[i]; + + if (from_wq) + target = zone->swap_limit; + else if (capable(CAP_SYS_ADMIN)) + target = zone->emer_mem; + else + target = zone->max_mem; + + target = (extra > target) ? 0ULL : target; + + if (zone->used_mem > target) + return true; + } + return false; +} + +/** + * At this point we only support a single shrink callback. + * Extend this if needed, perhaps using a linked list of callbacks. + * Note that this function is reentrant: + * many threads may try to swap out at any given time. + */ + +static void ttm_shrink(struct ttm_mem_global *glob, bool from_wq, + uint64_t extra) +{ + int ret; + struct ttm_mem_shrink *shrink; + + spin_lock(&glob->lock); + if (glob->shrink == NULL) + goto out; + + while (ttm_zones_above_swap_target(glob, from_wq, extra)) { + shrink = glob->shrink; + spin_unlock(&glob->lock); + ret = shrink->do_shrink(shrink); + spin_lock(&glob->lock); + if (unlikely(ret != 0)) + goto out; + } +out: + spin_unlock(&glob->lock); +} + + + +static void ttm_shrink_work(struct work_struct *work) +{ + struct ttm_mem_global *glob = + container_of(work, struct ttm_mem_global, work); + + ttm_shrink(glob, true, 0ULL); +} + +static int ttm_mem_init_kernel_zone(struct ttm_mem_global *glob, + const struct sysinfo *si) +{ + struct ttm_mem_zone *zone = kzalloc(sizeof(*zone), GFP_KERNEL); + uint64_t mem; + int ret; + + if (unlikely(!zone)) + return -ENOMEM; + + mem = si->totalram - si->totalhigh; + mem *= si->mem_unit; + + zone->name = "kernel"; + zone->zone_mem = mem; + zone->max_mem = mem >> 1; + zone->emer_mem = (mem >> 1) + (mem >> 2); + zone->swap_limit = zone->max_mem - (mem >> 3); + zone->used_mem = 0; + zone->glob = glob; + glob->zone_kernel = zone; + ret = kobject_init_and_add( + &zone->kobj, &ttm_mem_zone_kobj_type, &glob->kobj, zone->name); + if (unlikely(ret != 0)) { + kobject_put(&zone->kobj); + return ret; + } + glob->zones[glob->num_zones++] = zone; + return 0; +} + +#ifdef CONFIG_HIGHMEM +static int ttm_mem_init_highmem_zone(struct ttm_mem_global *glob, + const struct sysinfo *si) +{ + struct ttm_mem_zone *zone; + uint64_t mem; + int ret; + + if (si->totalhigh == 0) + return 0; + + zone = kzalloc(sizeof(*zone), GFP_KERNEL); + if (unlikely(!zone)) + return -ENOMEM; + + mem = si->totalram; + mem *= si->mem_unit; + + zone->name = "highmem"; + zone->zone_mem = mem; + zone->max_mem = mem >> 1; + zone->emer_mem = (mem >> 1) + (mem >> 2); + zone->swap_limit = zone->max_mem - (mem >> 3); + zone->used_mem = 0; + zone->glob = glob; + glob->zone_highmem = zone; + ret = kobject_init_and_add( + &zone->kobj, &ttm_mem_zone_kobj_type, &glob->kobj, zone->name); + if (unlikely(ret != 0)) { + kobject_put(&zone->kobj); + return ret; + } + glob->zones[glob->num_zones++] = zone; + return 0; +} +#else +static int ttm_mem_init_dma32_zone(struct ttm_mem_global *glob, + const struct sysinfo *si) +{ + struct ttm_mem_zone *zone = kzalloc(sizeof(*zone), GFP_KERNEL); + uint64_t mem; + int ret; + + if (unlikely(!zone)) + return -ENOMEM; + + mem = si->totalram; + mem *= si->mem_unit; + + /** + * No special dma32 zone needed. + */ + + if (mem <= ((uint64_t) 1ULL << 32)) { + kfree(zone); + return 0; + } + + /* + * Limit max dma32 memory to 4GB for now + * until we can figure out how big this + * zone really is. + */ + + mem = ((uint64_t) 1ULL << 32); + zone->name = "dma32"; + zone->zone_mem = mem; + zone->max_mem = mem >> 1; + zone->emer_mem = (mem >> 1) + (mem >> 2); + zone->swap_limit = zone->max_mem - (mem >> 3); + zone->used_mem = 0; + zone->glob = glob; + glob->zone_dma32 = zone; + ret = kobject_init_and_add( + &zone->kobj, &ttm_mem_zone_kobj_type, &glob->kobj, zone->name); + if (unlikely(ret != 0)) { + kobject_put(&zone->kobj); + return ret; + } + glob->zones[glob->num_zones++] = zone; + return 0; +} +#endif + +int ttm_mem_global_init(struct ttm_mem_global *glob) +{ + struct sysinfo si; + int ret; + int i; + struct ttm_mem_zone *zone; + + spin_lock_init(&glob->lock); + glob->swap_queue = create_singlethread_workqueue("ttm_swap"); + INIT_WORK(&glob->work, ttm_shrink_work); + ret = kobject_init_and_add( + &glob->kobj, &ttm_mem_glob_kobj_type, ttm_get_kobj(), "memory_accounting"); + if (unlikely(ret != 0)) { + kobject_put(&glob->kobj); + return ret; + } + + si_meminfo(&si); + + ret = ttm_mem_init_kernel_zone(glob, &si); + if (unlikely(ret != 0)) + goto out_no_zone; +#ifdef CONFIG_HIGHMEM + ret = ttm_mem_init_highmem_zone(glob, &si); + if (unlikely(ret != 0)) + goto out_no_zone; +#else + ret = ttm_mem_init_dma32_zone(glob, &si); + if (unlikely(ret != 0)) + goto out_no_zone; +#endif + for (i = 0; i < glob->num_zones; ++i) { + zone = glob->zones[i]; + pr_info("Zone %7s: Available graphics memory: %llu kiB\n", + zone->name, (unsigned long long)zone->max_mem >> 10); + } + ttm_page_alloc_init(glob, glob->zone_kernel->max_mem/(2*PAGE_SIZE)); + ttm_dma_page_alloc_init(glob, glob->zone_kernel->max_mem/(2*PAGE_SIZE)); + return 0; +out_no_zone: + ttm_mem_global_release(glob); + return ret; +} +EXPORT_SYMBOL(ttm_mem_global_init); + +void ttm_mem_global_release(struct ttm_mem_global *glob) +{ + unsigned int i; + struct ttm_mem_zone *zone; + + /* let the page allocator first stop the shrink work. */ + ttm_page_alloc_fini(); + ttm_dma_page_alloc_fini(); + + flush_workqueue(glob->swap_queue); + destroy_workqueue(glob->swap_queue); + glob->swap_queue = NULL; + for (i = 0; i < glob->num_zones; ++i) { + zone = glob->zones[i]; + kobject_del(&zone->kobj); + kobject_put(&zone->kobj); + } + kobject_del(&glob->kobj); + kobject_put(&glob->kobj); +} +EXPORT_SYMBOL(ttm_mem_global_release); + +static void ttm_check_swapping(struct ttm_mem_global *glob) +{ + bool needs_swapping = false; + unsigned int i; + struct ttm_mem_zone *zone; + + spin_lock(&glob->lock); + for (i = 0; i < glob->num_zones; ++i) { + zone = glob->zones[i]; + if (zone->used_mem > zone->swap_limit) { + needs_swapping = true; + break; + } + } + + spin_unlock(&glob->lock); + + if (unlikely(needs_swapping)) + (void)queue_work(glob->swap_queue, &glob->work); + +} + +static void ttm_mem_global_free_zone(struct ttm_mem_global *glob, + struct ttm_mem_zone *single_zone, + uint64_t amount) +{ + unsigned int i; + struct ttm_mem_zone *zone; + + spin_lock(&glob->lock); + for (i = 0; i < glob->num_zones; ++i) { + zone = glob->zones[i]; + if (single_zone && zone != single_zone) + continue; + zone->used_mem -= amount; + } + spin_unlock(&glob->lock); +} + +void ttm_mem_global_free(struct ttm_mem_global *glob, + uint64_t amount) +{ + return ttm_mem_global_free_zone(glob, NULL, amount); +} +EXPORT_SYMBOL(ttm_mem_global_free); + +static int ttm_mem_global_reserve(struct ttm_mem_global *glob, + struct ttm_mem_zone *single_zone, + uint64_t amount, bool reserve) +{ + uint64_t limit; + int ret = -ENOMEM; + unsigned int i; + struct ttm_mem_zone *zone; + + spin_lock(&glob->lock); + for (i = 0; i < glob->num_zones; ++i) { + zone = glob->zones[i]; + if (single_zone && zone != single_zone) + continue; + + limit = (capable(CAP_SYS_ADMIN)) ? + zone->emer_mem : zone->max_mem; + + if (zone->used_mem > limit) + goto out_unlock; + } + + if (reserve) { + for (i = 0; i < glob->num_zones; ++i) { + zone = glob->zones[i]; + if (single_zone && zone != single_zone) + continue; + zone->used_mem += amount; + } + } + + ret = 0; +out_unlock: + spin_unlock(&glob->lock); + ttm_check_swapping(glob); + + return ret; +} + + +static int ttm_mem_global_alloc_zone(struct ttm_mem_global *glob, + struct ttm_mem_zone *single_zone, + uint64_t memory, + bool no_wait, bool interruptible) +{ + int count = TTM_MEMORY_ALLOC_RETRIES; + + while (unlikely(ttm_mem_global_reserve(glob, + single_zone, + memory, true) + != 0)) { + if (no_wait) + return -ENOMEM; + if (unlikely(count-- == 0)) + return -ENOMEM; + ttm_shrink(glob, false, memory + (memory >> 2) + 16); + } + + return 0; +} + +int ttm_mem_global_alloc(struct ttm_mem_global *glob, uint64_t memory, + bool no_wait, bool interruptible) +{ + /** + * Normal allocations of kernel memory are registered in + * all zones. + */ + + return ttm_mem_global_alloc_zone(glob, NULL, memory, no_wait, + interruptible); +} +EXPORT_SYMBOL(ttm_mem_global_alloc); + +int ttm_mem_global_alloc_page(struct ttm_mem_global *glob, + struct page *page, + bool no_wait, bool interruptible) +{ + + struct ttm_mem_zone *zone = NULL; + + /** + * Page allocations may be registed in a single zone + * only if highmem or !dma32. + */ + +#ifdef CONFIG_HIGHMEM + if (PageHighMem(page) && glob->zone_highmem != NULL) + zone = glob->zone_highmem; +#else + if (glob->zone_dma32 && page_to_pfn(page) > 0x00100000UL) + zone = glob->zone_kernel; +#endif + return ttm_mem_global_alloc_zone(glob, zone, PAGE_SIZE, no_wait, + interruptible); +} + +void ttm_mem_global_free_page(struct ttm_mem_global *glob, struct page *page) +{ + struct ttm_mem_zone *zone = NULL; + +#ifdef CONFIG_HIGHMEM + if (PageHighMem(page) && glob->zone_highmem != NULL) + zone = glob->zone_highmem; +#else + if (glob->zone_dma32 && page_to_pfn(page) > 0x00100000UL) + zone = glob->zone_kernel; +#endif + ttm_mem_global_free_zone(glob, zone, PAGE_SIZE); +} + + +size_t ttm_round_pot(size_t size) +{ + if ((size & (size - 1)) == 0) + return size; + else if (size > PAGE_SIZE) + return PAGE_ALIGN(size); + else { + size_t tmp_size = 4; + + while (tmp_size < size) + tmp_size <<= 1; + + return tmp_size; + } + return 0; +} +EXPORT_SYMBOL(ttm_round_pot); diff --git a/drivers/gpu/drm/ttm/ttm_module.c b/drivers/gpu/drm/ttm/ttm_module.c new file mode 100644 index 0000000..d7f92fe --- /dev/null +++ b/drivers/gpu/drm/ttm/ttm_module.c @@ -0,0 +1,102 @@ +/************************************************************************** + * + * Copyright (c) 2006-2009 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 <thellstrom-at-vmware-dot-com> + * Jerome Glisse + */ +#include <linux/module.h> +#include <linux/device.h> +#include <linux/sched.h> +#include <drm/ttm/ttm_module.h> +#include <drm/drm_sysfs.h> + +static DECLARE_WAIT_QUEUE_HEAD(exit_q); +atomic_t device_released; + +static struct device_type ttm_drm_class_type = { + .name = "ttm", + /** + * Add pm ops here. + */ +}; + +static void ttm_drm_class_device_release(struct device *dev) +{ + atomic_set(&device_released, 1); + wake_up_all(&exit_q); +} + +static struct device ttm_drm_class_device = { + .type = &ttm_drm_class_type, + .release = &ttm_drm_class_device_release +}; + +struct kobject *ttm_get_kobj(void) +{ + struct kobject *kobj = &ttm_drm_class_device.kobj; + BUG_ON(kobj == NULL); + return kobj; +} + +static int __init ttm_init(void) +{ + int ret; + + ret = dev_set_name(&ttm_drm_class_device, "ttm"); + if (unlikely(ret != 0)) + return ret; + + atomic_set(&device_released, 0); + ret = drm_class_device_register(&ttm_drm_class_device); + if (unlikely(ret != 0)) + goto out_no_dev_reg; + + return 0; +out_no_dev_reg: + atomic_set(&device_released, 1); + wake_up_all(&exit_q); + return ret; +} + +static void __exit ttm_exit(void) +{ + drm_class_device_unregister(&ttm_drm_class_device); + + /** + * Refuse to unload until the TTM device is released. + * Not sure this is 100% needed. + */ + + wait_event(exit_q, atomic_read(&device_released) == 1); +} + +module_init(ttm_init); +module_exit(ttm_exit); + +MODULE_AUTHOR("Thomas Hellstrom, Jerome Glisse"); +MODULE_DESCRIPTION("TTM memory manager subsystem (for DRM device)"); +MODULE_LICENSE("GPL and additional rights"); diff --git a/drivers/gpu/drm/ttm/ttm_object.c b/drivers/gpu/drm/ttm/ttm_object.c new file mode 100644 index 0000000..58a5f32 --- /dev/null +++ b/drivers/gpu/drm/ttm/ttm_object.c @@ -0,0 +1,454 @@ +/************************************************************************** + * + * Copyright (c) 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 <thellstrom-at-vmware-dot-com> + */ +/** @file ttm_ref_object.c + * + * Base- and reference object implementation for the various + * ttm objects. Implements reference counting, minimal security checks + * and release on file close. + */ + +/** + * struct ttm_object_file + * + * @tdev: Pointer to the ttm_object_device. + * + * @lock: Lock that protects the ref_list list and the + * ref_hash hash tables. + * + * @ref_list: List of ttm_ref_objects to be destroyed at + * file release. + * + * @ref_hash: Hash tables of ref objects, one per ttm_ref_type, + * for fast lookup of ref objects given a base object. + */ + +#define pr_fmt(fmt) "[TTM] " fmt + +#include <drm/ttm/ttm_object.h> +#include <drm/ttm/ttm_module.h> +#include <linux/list.h> +#include <linux/spinlock.h> +#include <linux/slab.h> +#include <linux/module.h> +#include <linux/atomic.h> + +struct ttm_object_file { + struct ttm_object_device *tdev; + rwlock_t lock; + struct list_head ref_list; + struct drm_open_hash ref_hash[TTM_REF_NUM]; + struct kref refcount; +}; + +/** + * struct ttm_object_device + * + * @object_lock: lock that protects the object_hash hash table. + * + * @object_hash: hash table for fast lookup of object global names. + * + * @object_count: Per device object count. + * + * This is the per-device data structure needed for ttm object management. + */ + +struct ttm_object_device { + spinlock_t object_lock; + struct drm_open_hash object_hash; + atomic_t object_count; + struct ttm_mem_global *mem_glob; +}; + +/** + * struct ttm_ref_object + * + * @hash: Hash entry for the per-file object reference hash. + * + * @head: List entry for the per-file list of ref-objects. + * + * @kref: Ref count. + * + * @obj: Base object this ref object is referencing. + * + * @ref_type: Type of ref object. + * + * This is similar to an idr object, but it also has a hash table entry + * that allows lookup with a pointer to the referenced object as a key. In + * that way, one can easily detect whether a base object is referenced by + * a particular ttm_object_file. It also carries a ref count to avoid creating + * multiple ref objects if a ttm_object_file references the same base + * object more than once. + */ + +struct ttm_ref_object { + struct drm_hash_item hash; + struct list_head head; + struct kref kref; + enum ttm_ref_type ref_type; + struct ttm_base_object *obj; + struct ttm_object_file *tfile; +}; + +static inline struct ttm_object_file * +ttm_object_file_ref(struct ttm_object_file *tfile) +{ + kref_get(&tfile->refcount); + return tfile; +} + +static void ttm_object_file_destroy(struct kref *kref) +{ + struct ttm_object_file *tfile = + container_of(kref, struct ttm_object_file, refcount); + + kfree(tfile); +} + + +static inline void ttm_object_file_unref(struct ttm_object_file **p_tfile) +{ + struct ttm_object_file *tfile = *p_tfile; + + *p_tfile = NULL; + kref_put(&tfile->refcount, ttm_object_file_destroy); +} + + +int ttm_base_object_init(struct ttm_object_file *tfile, + struct ttm_base_object *base, + bool shareable, + enum ttm_object_type object_type, + void (*refcount_release) (struct ttm_base_object **), + void (*ref_obj_release) (struct ttm_base_object *, + enum ttm_ref_type ref_type)) +{ + struct ttm_object_device *tdev = tfile->tdev; + int ret; + + base->shareable = shareable; + base->tfile = ttm_object_file_ref(tfile); + base->refcount_release = refcount_release; + base->ref_obj_release = ref_obj_release; + base->object_type = object_type; + kref_init(&base->refcount); + spin_lock(&tdev->object_lock); + ret = drm_ht_just_insert_please_rcu(&tdev->object_hash, + &base->hash, + (unsigned long)base, 31, 0, 0); + spin_unlock(&tdev->object_lock); + if (unlikely(ret != 0)) + goto out_err0; + + ret = ttm_ref_object_add(tfile, base, TTM_REF_USAGE, NULL); + if (unlikely(ret != 0)) + goto out_err1; + + ttm_base_object_unref(&base); + + return 0; +out_err1: + spin_lock(&tdev->object_lock); + (void)drm_ht_remove_item_rcu(&tdev->object_hash, &base->hash); + spin_unlock(&tdev->object_lock); +out_err0: + return ret; +} +EXPORT_SYMBOL(ttm_base_object_init); + +static void ttm_release_base(struct kref *kref) +{ + struct ttm_base_object *base = + container_of(kref, struct ttm_base_object, refcount); + struct ttm_object_device *tdev = base->tfile->tdev; + + spin_lock(&tdev->object_lock); + (void)drm_ht_remove_item_rcu(&tdev->object_hash, &base->hash); + spin_unlock(&tdev->object_lock); + + /* + * Note: We don't use synchronize_rcu() here because it's far + * too slow. It's up to the user to free the object using + * call_rcu() or ttm_base_object_kfree(). + */ + + if (base->refcount_release) { + ttm_object_file_unref(&base->tfile); + base->refcount_release(&base); + } +} + +void ttm_base_object_unref(struct ttm_base_object **p_base) +{ + struct ttm_base_object *base = *p_base; + + *p_base = NULL; + + kref_put(&base->refcount, ttm_release_base); +} +EXPORT_SYMBOL(ttm_base_object_unref); + +struct ttm_base_object *ttm_base_object_lookup(struct ttm_object_file *tfile, + uint32_t key) +{ + struct ttm_object_device *tdev = tfile->tdev; + struct ttm_base_object *base; + struct drm_hash_item *hash; + int ret; + + rcu_read_lock(); + ret = drm_ht_find_item_rcu(&tdev->object_hash, key, &hash); + + if (likely(ret == 0)) { + base = drm_hash_entry(hash, struct ttm_base_object, hash); + ret = kref_get_unless_zero(&base->refcount) ? 0 : -EINVAL; + } + rcu_read_unlock(); + + if (unlikely(ret != 0)) + return NULL; + + if (tfile != base->tfile && !base->shareable) { + pr_err("Attempted access of non-shareable object\n"); + ttm_base_object_unref(&base); + return NULL; + } + + return base; +} +EXPORT_SYMBOL(ttm_base_object_lookup); + +int ttm_ref_object_add(struct ttm_object_file *tfile, + struct ttm_base_object *base, + enum ttm_ref_type ref_type, bool *existed) +{ + struct drm_open_hash *ht = &tfile->ref_hash[ref_type]; + struct ttm_ref_object *ref; + struct drm_hash_item *hash; + struct ttm_mem_global *mem_glob = tfile->tdev->mem_glob; + int ret = -EINVAL; + + if (existed != NULL) + *existed = true; + + while (ret == -EINVAL) { + read_lock(&tfile->lock); + ret = drm_ht_find_item(ht, base->hash.key, &hash); + + if (ret == 0) { + ref = drm_hash_entry(hash, struct ttm_ref_object, hash); + kref_get(&ref->kref); + read_unlock(&tfile->lock); + break; + } + + read_unlock(&tfile->lock); + ret = ttm_mem_global_alloc(mem_glob, sizeof(*ref), + false, false); + if (unlikely(ret != 0)) + return ret; + ref = kmalloc(sizeof(*ref), GFP_KERNEL); + if (unlikely(ref == NULL)) { + ttm_mem_global_free(mem_glob, sizeof(*ref)); + return -ENOMEM; + } + + ref->hash.key = base->hash.key; + ref->obj = base; + ref->tfile = tfile; + ref->ref_type = ref_type; + kref_init(&ref->kref); + + write_lock(&tfile->lock); + ret = drm_ht_insert_item(ht, &ref->hash); + + if (likely(ret == 0)) { + list_add_tail(&ref->head, &tfile->ref_list); + kref_get(&base->refcount); + write_unlock(&tfile->lock); + if (existed != NULL) + *existed = false; + break; + } + + write_unlock(&tfile->lock); + BUG_ON(ret != -EINVAL); + + ttm_mem_global_free(mem_glob, sizeof(*ref)); + kfree(ref); + } + + return ret; +} +EXPORT_SYMBOL(ttm_ref_object_add); + +static void ttm_ref_object_release(struct kref *kref) +{ + struct ttm_ref_object *ref = + container_of(kref, struct ttm_ref_object, kref); + struct ttm_base_object *base = ref->obj; + struct ttm_object_file *tfile = ref->tfile; + struct drm_open_hash *ht; + struct ttm_mem_global *mem_glob = tfile->tdev->mem_glob; + + ht = &tfile->ref_hash[ref->ref_type]; + (void)drm_ht_remove_item(ht, &ref->hash); + list_del(&ref->head); + write_unlock(&tfile->lock); + + if (ref->ref_type != TTM_REF_USAGE && base->ref_obj_release) + base->ref_obj_release(base, ref->ref_type); + + ttm_base_object_unref(&ref->obj); + ttm_mem_global_free(mem_glob, sizeof(*ref)); + kfree(ref); + write_lock(&tfile->lock); +} + +int ttm_ref_object_base_unref(struct ttm_object_file *tfile, + unsigned long key, enum ttm_ref_type ref_type) +{ + struct drm_open_hash *ht = &tfile->ref_hash[ref_type]; + struct ttm_ref_object *ref; + struct drm_hash_item *hash; + int ret; + + write_lock(&tfile->lock); + ret = drm_ht_find_item(ht, key, &hash); + if (unlikely(ret != 0)) { + write_unlock(&tfile->lock); + return -EINVAL; + } + ref = drm_hash_entry(hash, struct ttm_ref_object, hash); + kref_put(&ref->kref, ttm_ref_object_release); + write_unlock(&tfile->lock); + return 0; +} +EXPORT_SYMBOL(ttm_ref_object_base_unref); + +void ttm_object_file_release(struct ttm_object_file **p_tfile) +{ + struct ttm_ref_object *ref; + struct list_head *list; + unsigned int i; + struct ttm_object_file *tfile = *p_tfile; + + *p_tfile = NULL; + write_lock(&tfile->lock); + + /* + * Since we release the lock within the loop, we have to + * restart it from the beginning each time. + */ + + while (!list_empty(&tfile->ref_list)) { + list = tfile->ref_list.next; + ref = list_entry(list, struct ttm_ref_object, head); + ttm_ref_object_release(&ref->kref); + } + + for (i = 0; i < TTM_REF_NUM; ++i) + drm_ht_remove(&tfile->ref_hash[i]); + + write_unlock(&tfile->lock); + ttm_object_file_unref(&tfile); +} +EXPORT_SYMBOL(ttm_object_file_release); + +struct ttm_object_file *ttm_object_file_init(struct ttm_object_device *tdev, + unsigned int hash_order) +{ + struct ttm_object_file *tfile = kmalloc(sizeof(*tfile), GFP_KERNEL); + unsigned int i; + unsigned int j = 0; + int ret; + + if (unlikely(tfile == NULL)) + return NULL; + + rwlock_init(&tfile->lock); + tfile->tdev = tdev; + kref_init(&tfile->refcount); + INIT_LIST_HEAD(&tfile->ref_list); + + for (i = 0; i < TTM_REF_NUM; ++i) { + ret = drm_ht_create(&tfile->ref_hash[i], hash_order); + if (ret) { + j = i; + goto out_err; + } + } + + return tfile; +out_err: + for (i = 0; i < j; ++i) + drm_ht_remove(&tfile->ref_hash[i]); + + kfree(tfile); + + return NULL; +} +EXPORT_SYMBOL(ttm_object_file_init); + +struct ttm_object_device *ttm_object_device_init(struct ttm_mem_global + *mem_glob, + unsigned int hash_order) +{ + struct ttm_object_device *tdev = kmalloc(sizeof(*tdev), GFP_KERNEL); + int ret; + + if (unlikely(tdev == NULL)) + return NULL; + + tdev->mem_glob = mem_glob; + spin_lock_init(&tdev->object_lock); + atomic_set(&tdev->object_count, 0); + ret = drm_ht_create(&tdev->object_hash, hash_order); + + if (likely(ret == 0)) + return tdev; + + kfree(tdev); + return NULL; +} +EXPORT_SYMBOL(ttm_object_device_init); + +void ttm_object_device_release(struct ttm_object_device **p_tdev) +{ + struct ttm_object_device *tdev = *p_tdev; + + *p_tdev = NULL; + + spin_lock(&tdev->object_lock); + drm_ht_remove(&tdev->object_hash); + spin_unlock(&tdev->object_lock); + + kfree(tdev); +} +EXPORT_SYMBOL(ttm_object_device_release); diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc.c b/drivers/gpu/drm/ttm/ttm_page_alloc.c new file mode 100644 index 0000000..bd2a3b4 --- /dev/null +++ b/drivers/gpu/drm/ttm/ttm_page_alloc.c @@ -0,0 +1,919 @@ +/* + * Copyright (c) Red Hat Inc. + + * 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 AUTHORS OR COPYRIGHT HOLDERS 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: Dave Airlie <airlied@redhat.com> + * Jerome Glisse <jglisse@redhat.com> + * Pauli Nieminen <suokkos@gmail.com> + */ + +/* simple list based uncached page pool + * - Pool collects resently freed pages for reuse + * - Use page->lru to keep a free list + * - doesn't track currently in use pages + */ + +#define pr_fmt(fmt) "[TTM] " fmt + +#include <linux/list.h> +#include <linux/spinlock.h> +#include <linux/highmem.h> +#include <linux/mm_types.h> +#include <linux/module.h> +#include <linux/mm.h> +#include <linux/seq_file.h> /* for seq_printf */ +#include <linux/slab.h> +#include <linux/dma-mapping.h> + +#include <linux/atomic.h> + +#include <drm/ttm/ttm_bo_driver.h> +#include <drm/ttm/ttm_page_alloc.h> + +#ifdef TTM_HAS_AGP +#include <asm/agp.h> +#endif + +#define NUM_PAGES_TO_ALLOC (PAGE_SIZE/sizeof(struct page *)) +#define SMALL_ALLOCATION 16 +#define FREE_ALL_PAGES (~0U) +/* times are in msecs */ +#define PAGE_FREE_INTERVAL 1000 + +/** + * struct ttm_page_pool - Pool to reuse recently allocated uc/wc pages. + * + * @lock: Protects the shared pool from concurrnet access. Must be used with + * irqsave/irqrestore variants because pool allocator maybe called from + * delayed work. + * @fill_lock: Prevent concurrent calls to fill. + * @list: Pool of free uc/wc pages for fast reuse. + * @gfp_flags: Flags to pass for alloc_page. + * @npages: Number of pages in pool. + */ +struct ttm_page_pool { + spinlock_t lock; + bool fill_lock; + struct list_head list; + gfp_t gfp_flags; + unsigned npages; + char *name; + unsigned long nfrees; + unsigned long nrefills; +}; + +/** + * Limits for the pool. They are handled without locks because only place where + * they may change is in sysfs store. They won't have immediate effect anyway + * so forcing serialization to access them is pointless. + */ + +struct ttm_pool_opts { + unsigned alloc_size; + unsigned max_size; + unsigned small; +}; + +#define NUM_POOLS 4 + +/** + * struct ttm_pool_manager - Holds memory pools for fst allocation + * + * Manager is read only object for pool code so it doesn't need locking. + * + * @free_interval: minimum number of jiffies between freeing pages from pool. + * @page_alloc_inited: reference counting for pool allocation. + * @work: Work that is used to shrink the pool. Work is only run when there is + * some pages to free. + * @small_allocation: Limit in number of pages what is small allocation. + * + * @pools: All pool objects in use. + **/ +struct ttm_pool_manager { + struct kobject kobj; + struct shrinker mm_shrink; + struct ttm_pool_opts options; + + union { + struct ttm_page_pool pools[NUM_POOLS]; + struct { + struct ttm_page_pool wc_pool; + struct ttm_page_pool uc_pool; + struct ttm_page_pool wc_pool_dma32; + struct ttm_page_pool uc_pool_dma32; + } ; + }; +}; + +static struct attribute ttm_page_pool_max = { + .name = "pool_max_size", + .mode = S_IRUGO | S_IWUSR +}; +static struct attribute ttm_page_pool_small = { + .name = "pool_small_allocation", + .mode = S_IRUGO | S_IWUSR +}; +static struct attribute ttm_page_pool_alloc_size = { + .name = "pool_allocation_size", + .mode = S_IRUGO | S_IWUSR +}; + +static struct attribute *ttm_pool_attrs[] = { + &ttm_page_pool_max, + &ttm_page_pool_small, + &ttm_page_pool_alloc_size, + NULL +}; + +static void ttm_pool_kobj_release(struct kobject *kobj) +{ + struct ttm_pool_manager *m = + container_of(kobj, struct ttm_pool_manager, kobj); + kfree(m); +} + +static ssize_t ttm_pool_store(struct kobject *kobj, + struct attribute *attr, const char *buffer, size_t size) +{ + struct ttm_pool_manager *m = + container_of(kobj, struct ttm_pool_manager, kobj); + int chars; + unsigned val; + chars = sscanf(buffer, "%u", &val); + if (chars == 0) + return size; + + /* Convert kb to number of pages */ + val = val / (PAGE_SIZE >> 10); + + if (attr == &ttm_page_pool_max) + m->options.max_size = val; + else if (attr == &ttm_page_pool_small) + m->options.small = val; + else if (attr == &ttm_page_pool_alloc_size) { + if (val > NUM_PAGES_TO_ALLOC*8) { + pr_err("Setting allocation size to %lu is not allowed. Recommended size is %lu\n", + NUM_PAGES_TO_ALLOC*(PAGE_SIZE >> 7), + NUM_PAGES_TO_ALLOC*(PAGE_SIZE >> 10)); + return size; + } else if (val > NUM_PAGES_TO_ALLOC) { + pr_warn("Setting allocation size to larger than %lu is not recommended\n", + NUM_PAGES_TO_ALLOC*(PAGE_SIZE >> 10)); + } + m->options.alloc_size = val; + } + + return size; +} + +static ssize_t ttm_pool_show(struct kobject *kobj, + struct attribute *attr, char *buffer) +{ + struct ttm_pool_manager *m = + container_of(kobj, struct ttm_pool_manager, kobj); + unsigned val = 0; + + if (attr == &ttm_page_pool_max) + val = m->options.max_size; + else if (attr == &ttm_page_pool_small) + val = m->options.small; + else if (attr == &ttm_page_pool_alloc_size) + val = m->options.alloc_size; + + val = val * (PAGE_SIZE >> 10); + + return snprintf(buffer, PAGE_SIZE, "%u\n", val); +} + +static const struct sysfs_ops ttm_pool_sysfs_ops = { + .show = &ttm_pool_show, + .store = &ttm_pool_store, +}; + +static struct kobj_type ttm_pool_kobj_type = { + .release = &ttm_pool_kobj_release, + .sysfs_ops = &ttm_pool_sysfs_ops, + .default_attrs = ttm_pool_attrs, +}; + +static struct ttm_pool_manager *_manager; + +#ifndef CONFIG_X86 +static int set_pages_array_wb(struct page **pages, int addrinarray) +{ +#ifdef TTM_HAS_AGP + int i; + + for (i = 0; i < addrinarray; i++) + unmap_page_from_agp(pages[i]); +#endif + return 0; +} + +static int set_pages_array_wc(struct page **pages, int addrinarray) +{ +#ifdef TTM_HAS_AGP + int i; + + for (i = 0; i < addrinarray; i++) + map_page_into_agp(pages[i]); +#endif + return 0; +} + +static int set_pages_array_uc(struct page **pages, int addrinarray) +{ +#ifdef TTM_HAS_AGP + int i; + + for (i = 0; i < addrinarray; i++) + map_page_into_agp(pages[i]); +#endif + return 0; +} +#endif + +/** + * Select the right pool or requested caching state and ttm flags. */ +static struct ttm_page_pool *ttm_get_pool(int flags, + enum ttm_caching_state cstate) +{ + int pool_index; + + if (cstate == tt_cached) + return NULL; + + if (cstate == tt_wc) + pool_index = 0x0; + else + pool_index = 0x1; + + if (flags & TTM_PAGE_FLAG_DMA32) + pool_index |= 0x2; + + return &_manager->pools[pool_index]; +} + +/* set memory back to wb and free the pages. */ +static void ttm_pages_put(struct page *pages[], unsigned npages) +{ + unsigned i; + if (set_pages_array_wb(pages, npages)) + pr_err("Failed to set %d pages to wb!\n", npages); + for (i = 0; i < npages; ++i) + __free_page(pages[i]); +} + +static void ttm_pool_update_free_locked(struct ttm_page_pool *pool, + unsigned freed_pages) +{ + pool->npages -= freed_pages; + pool->nfrees += freed_pages; +} + +/** + * Free pages from pool. + * + * To prevent hogging the ttm_swap process we only free NUM_PAGES_TO_ALLOC + * number of pages in one go. + * + * @pool: to free the pages from + * @free_all: If set to true will free all pages in pool + **/ +static int ttm_page_pool_free(struct ttm_page_pool *pool, unsigned nr_free) +{ + unsigned long irq_flags; + struct page *p; + struct page **pages_to_free; + unsigned freed_pages = 0, + npages_to_free = nr_free; + + if (NUM_PAGES_TO_ALLOC < nr_free) + npages_to_free = NUM_PAGES_TO_ALLOC; + + pages_to_free = kmalloc(npages_to_free * sizeof(struct page *), + GFP_KERNEL); + if (!pages_to_free) { + pr_err("Failed to allocate memory for pool free operation\n"); + return 0; + } + +restart: + spin_lock_irqsave(&pool->lock, irq_flags); + + list_for_each_entry_reverse(p, &pool->list, lru) { + if (freed_pages >= npages_to_free) + break; + + pages_to_free[freed_pages++] = p; + /* We can only remove NUM_PAGES_TO_ALLOC at a time. */ + if (freed_pages >= NUM_PAGES_TO_ALLOC) { + /* remove range of pages from the pool */ + __list_del(p->lru.prev, &pool->list); + + ttm_pool_update_free_locked(pool, freed_pages); + /** + * Because changing page caching is costly + * we unlock the pool to prevent stalling. + */ + spin_unlock_irqrestore(&pool->lock, irq_flags); + + ttm_pages_put(pages_to_free, freed_pages); + if (likely(nr_free != FREE_ALL_PAGES)) + nr_free -= freed_pages; + + if (NUM_PAGES_TO_ALLOC >= nr_free) + npages_to_free = nr_free; + else + npages_to_free = NUM_PAGES_TO_ALLOC; + + freed_pages = 0; + + /* free all so restart the processing */ + if (nr_free) + goto restart; + + /* Not allowed to fall through or break because + * following context is inside spinlock while we are + * outside here. + */ + goto out; + + } + } + + /* remove range of pages from the pool */ + if (freed_pages) { + __list_del(&p->lru, &pool->list); + + ttm_pool_update_free_locked(pool, freed_pages); + nr_free -= freed_pages; + } + + spin_unlock_irqrestore(&pool->lock, irq_flags); + + if (freed_pages) + ttm_pages_put(pages_to_free, freed_pages); +out: + kfree(pages_to_free); + return nr_free; +} + +/* Get good estimation how many pages are free in pools */ +static int ttm_pool_get_num_unused_pages(void) +{ + unsigned i; + int total = 0; + for (i = 0; i < NUM_POOLS; ++i) + total += _manager->pools[i].npages; + + return total; +} + +/** + * Callback for mm to request pool to reduce number of page held. + */ +static int ttm_pool_mm_shrink(struct shrinker *shrink, + struct shrink_control *sc) +{ + static atomic_t start_pool = ATOMIC_INIT(0); + unsigned i; + unsigned pool_offset = atomic_add_return(1, &start_pool); + struct ttm_page_pool *pool; + int shrink_pages = sc->nr_to_scan; + + pool_offset = pool_offset % NUM_POOLS; + /* select start pool in round robin fashion */ + for (i = 0; i < NUM_POOLS; ++i) { + unsigned nr_free = shrink_pages; + if (shrink_pages == 0) + break; + pool = &_manager->pools[(i + pool_offset)%NUM_POOLS]; + shrink_pages = ttm_page_pool_free(pool, nr_free); + } + /* return estimated number of unused pages in pool */ + return ttm_pool_get_num_unused_pages(); +} + +static void ttm_pool_mm_shrink_init(struct ttm_pool_manager *manager) +{ + manager->mm_shrink.shrink = &ttm_pool_mm_shrink; + manager->mm_shrink.seeks = 1; + register_shrinker(&manager->mm_shrink); +} + +static void ttm_pool_mm_shrink_fini(struct ttm_pool_manager *manager) +{ + unregister_shrinker(&manager->mm_shrink); +} + +static int ttm_set_pages_caching(struct page **pages, + enum ttm_caching_state cstate, unsigned cpages) +{ + int r = 0; + /* Set page caching */ + switch (cstate) { + case tt_uncached: + r = set_pages_array_uc(pages, cpages); + if (r) + pr_err("Failed to set %d pages to uc!\n", cpages); + break; + case tt_wc: + r = set_pages_array_wc(pages, cpages); + if (r) + pr_err("Failed to set %d pages to wc!\n", cpages); + break; + default: + break; + } + return r; +} + +/** + * Free pages the pages that failed to change the caching state. If there is + * any pages that have changed their caching state already put them to the + * pool. + */ +static void ttm_handle_caching_state_failure(struct list_head *pages, + int ttm_flags, enum ttm_caching_state cstate, + struct page **failed_pages, unsigned cpages) +{ + unsigned i; + /* Failed pages have to be freed */ + for (i = 0; i < cpages; ++i) { + list_del(&failed_pages[i]->lru); + __free_page(failed_pages[i]); + } +} + +/** + * Allocate new pages with correct caching. + * + * This function is reentrant if caller updates count depending on number of + * pages returned in pages array. + */ +static int ttm_alloc_new_pages(struct list_head *pages, gfp_t gfp_flags, + int ttm_flags, enum ttm_caching_state cstate, unsigned count) +{ + struct page **caching_array; + struct page *p; + int r = 0; + unsigned i, cpages; + unsigned max_cpages = min(count, + (unsigned)(PAGE_SIZE/sizeof(struct page *))); + + /* allocate array for page caching change */ + caching_array = kmalloc(max_cpages*sizeof(struct page *), GFP_KERNEL); + + if (!caching_array) { + pr_err("Unable to allocate table for new pages\n"); + return -ENOMEM; + } + + for (i = 0, cpages = 0; i < count; ++i) { + p = alloc_page(gfp_flags); + + if (!p) { + pr_err("Unable to get page %u\n", i); + + /* store already allocated pages in the pool after + * setting the caching state */ + if (cpages) { + r = ttm_set_pages_caching(caching_array, + cstate, cpages); + if (r) + ttm_handle_caching_state_failure(pages, + ttm_flags, cstate, + caching_array, cpages); + } + r = -ENOMEM; + goto out; + } + +#ifdef CONFIG_HIGHMEM + /* gfp flags of highmem page should never be dma32 so we + * we should be fine in such case + */ + if (!PageHighMem(p)) +#endif + { + caching_array[cpages++] = p; + if (cpages == max_cpages) { + + r = ttm_set_pages_caching(caching_array, + cstate, cpages); + if (r) { + ttm_handle_caching_state_failure(pages, + ttm_flags, cstate, + caching_array, cpages); + goto out; + } + cpages = 0; + } + } + + list_add(&p->lru, pages); + } + + if (cpages) { + r = ttm_set_pages_caching(caching_array, cstate, cpages); + if (r) + ttm_handle_caching_state_failure(pages, + ttm_flags, cstate, + caching_array, cpages); + } +out: + kfree(caching_array); + + return r; +} + +/** + * Fill the given pool if there aren't enough pages and the requested number of + * pages is small. + */ +static void ttm_page_pool_fill_locked(struct ttm_page_pool *pool, + int ttm_flags, enum ttm_caching_state cstate, unsigned count, + unsigned long *irq_flags) +{ + struct page *p; + int r; + unsigned cpages = 0; + /** + * Only allow one pool fill operation at a time. + * If pool doesn't have enough pages for the allocation new pages are + * allocated from outside of pool. + */ + if (pool->fill_lock) + return; + + pool->fill_lock = true; + + /* If allocation request is small and there are not enough + * pages in a pool we fill the pool up first. */ + if (count < _manager->options.small + && count > pool->npages) { + struct list_head new_pages; + unsigned alloc_size = _manager->options.alloc_size; + + /** + * Can't change page caching if in irqsave context. We have to + * drop the pool->lock. + */ + spin_unlock_irqrestore(&pool->lock, *irq_flags); + + INIT_LIST_HEAD(&new_pages); + r = ttm_alloc_new_pages(&new_pages, pool->gfp_flags, ttm_flags, + cstate, alloc_size); + spin_lock_irqsave(&pool->lock, *irq_flags); + + if (!r) { + list_splice(&new_pages, &pool->list); + ++pool->nrefills; + pool->npages += alloc_size; + } else { + pr_err("Failed to fill pool (%p)\n", pool); + /* If we have any pages left put them to the pool. */ + list_for_each_entry(p, &pool->list, lru) { + ++cpages; + } + list_splice(&new_pages, &pool->list); + pool->npages += cpages; + } + + } + pool->fill_lock = false; +} + +/** + * Cut 'count' number of pages from the pool and put them on the return list. + * + * @return count of pages still required to fulfill the request. + */ +static unsigned ttm_page_pool_get_pages(struct ttm_page_pool *pool, + struct list_head *pages, + int ttm_flags, + enum ttm_caching_state cstate, + unsigned count) +{ + unsigned long irq_flags; + struct list_head *p; + unsigned i; + + spin_lock_irqsave(&pool->lock, irq_flags); + ttm_page_pool_fill_locked(pool, ttm_flags, cstate, count, &irq_flags); + + if (count >= pool->npages) { + /* take all pages from the pool */ + list_splice_init(&pool->list, pages); + count -= pool->npages; + pool->npages = 0; + goto out; + } + /* find the last pages to include for requested number of pages. Split + * pool to begin and halve it to reduce search space. */ + if (count <= pool->npages/2) { + i = 0; + list_for_each(p, &pool->list) { + if (++i == count) + break; + } + } else { + i = pool->npages + 1; + list_for_each_prev(p, &pool->list) { + if (--i == count) + break; + } + } + /* Cut 'count' number of pages from the pool */ + list_cut_position(pages, &pool->list, p); + pool->npages -= count; + count = 0; +out: + spin_unlock_irqrestore(&pool->lock, irq_flags); + return count; +} + +/* Put all pages in pages list to correct pool to wait for reuse */ +static void ttm_put_pages(struct page **pages, unsigned npages, int flags, + enum ttm_caching_state cstate) +{ + unsigned long irq_flags; + struct ttm_page_pool *pool = ttm_get_pool(flags, cstate); + unsigned i; + + if (pool == NULL) { + /* No pool for this memory type so free the pages */ + for (i = 0; i < npages; i++) { + if (pages[i]) { + if (page_count(pages[i]) != 1) + pr_err("Erroneous page count. Leaking pages.\n"); + __free_page(pages[i]); + pages[i] = NULL; + } + } + return; + } + + spin_lock_irqsave(&pool->lock, irq_flags); + for (i = 0; i < npages; i++) { + if (pages[i]) { + if (page_count(pages[i]) != 1) + pr_err("Erroneous page count. Leaking pages.\n"); + list_add_tail(&pages[i]->lru, &pool->list); + pages[i] = NULL; + pool->npages++; + } + } + /* Check that we don't go over the pool limit */ + npages = 0; + if (pool->npages > _manager->options.max_size) { + npages = pool->npages - _manager->options.max_size; + /* free at least NUM_PAGES_TO_ALLOC number of pages + * to reduce calls to set_memory_wb */ + if (npages < NUM_PAGES_TO_ALLOC) + npages = NUM_PAGES_TO_ALLOC; + } + spin_unlock_irqrestore(&pool->lock, irq_flags); + if (npages) + ttm_page_pool_free(pool, npages); +} + +/* + * On success pages list will hold count number of correctly + * cached pages. + */ +static int ttm_get_pages(struct page **pages, unsigned npages, int flags, + enum ttm_caching_state cstate) +{ + struct ttm_page_pool *pool = ttm_get_pool(flags, cstate); + struct list_head plist; + struct page *p = NULL; + gfp_t gfp_flags = GFP_USER; + unsigned count; + int r; + + /* set zero flag for page allocation if required */ + if (flags & TTM_PAGE_FLAG_ZERO_ALLOC) + gfp_flags |= __GFP_ZERO; + + /* No pool for cached pages */ + if (pool == NULL) { + if (flags & TTM_PAGE_FLAG_DMA32) + gfp_flags |= GFP_DMA32; + else + gfp_flags |= GFP_HIGHUSER; + + for (r = 0; r < npages; ++r) { + p = alloc_page(gfp_flags); + if (!p) { + + pr_err("Unable to allocate page\n"); + return -ENOMEM; + } + + pages[r] = p; + } + return 0; + } + + /* combine zero flag to pool flags */ + gfp_flags |= pool->gfp_flags; + + /* First we take pages from the pool */ + INIT_LIST_HEAD(&plist); + npages = ttm_page_pool_get_pages(pool, &plist, flags, cstate, npages); + count = 0; + list_for_each_entry(p, &plist, lru) { + pages[count++] = p; + } + + /* clear the pages coming from the pool if requested */ + if (flags & TTM_PAGE_FLAG_ZERO_ALLOC) { + list_for_each_entry(p, &plist, lru) { + if (PageHighMem(p)) + clear_highpage(p); + else + clear_page(page_address(p)); + } + } + + /* If pool didn't have enough pages allocate new one. */ + if (npages > 0) { + /* ttm_alloc_new_pages doesn't reference pool so we can run + * multiple requests in parallel. + **/ + INIT_LIST_HEAD(&plist); + r = ttm_alloc_new_pages(&plist, gfp_flags, flags, cstate, npages); + list_for_each_entry(p, &plist, lru) { + pages[count++] = p; + } + if (r) { + /* If there is any pages in the list put them back to + * the pool. */ + pr_err("Failed to allocate extra pages for large request\n"); + ttm_put_pages(pages, count, flags, cstate); + return r; + } + } + + return 0; +} + +static void ttm_page_pool_init_locked(struct ttm_page_pool *pool, int flags, + char *name) +{ + spin_lock_init(&pool->lock); + pool->fill_lock = false; + INIT_LIST_HEAD(&pool->list); + pool->npages = pool->nfrees = 0; + pool->gfp_flags = flags; + pool->name = name; +} + +int ttm_page_alloc_init(struct ttm_mem_global *glob, unsigned max_pages) +{ + int ret; + + WARN_ON(_manager); + + pr_info("Initializing pool allocator\n"); + + _manager = kzalloc(sizeof(*_manager), GFP_KERNEL); + + ttm_page_pool_init_locked(&_manager->wc_pool, GFP_HIGHUSER, "wc"); + + ttm_page_pool_init_locked(&_manager->uc_pool, GFP_HIGHUSER, "uc"); + + ttm_page_pool_init_locked(&_manager->wc_pool_dma32, + GFP_USER | GFP_DMA32, "wc dma"); + + ttm_page_pool_init_locked(&_manager->uc_pool_dma32, + GFP_USER | GFP_DMA32, "uc dma"); + + _manager->options.max_size = max_pages; + _manager->options.small = SMALL_ALLOCATION; + _manager->options.alloc_size = NUM_PAGES_TO_ALLOC; + + ret = kobject_init_and_add(&_manager->kobj, &ttm_pool_kobj_type, + &glob->kobj, "pool"); + if (unlikely(ret != 0)) { + kobject_put(&_manager->kobj); + _manager = NULL; + return ret; + } + + ttm_pool_mm_shrink_init(_manager); + + return 0; +} + +void ttm_page_alloc_fini(void) +{ + int i; + + pr_info("Finalizing pool allocator\n"); + ttm_pool_mm_shrink_fini(_manager); + + for (i = 0; i < NUM_POOLS; ++i) + ttm_page_pool_free(&_manager->pools[i], FREE_ALL_PAGES); + + kobject_put(&_manager->kobj); + _manager = NULL; +} + +int ttm_pool_populate(struct ttm_tt *ttm) +{ + struct ttm_mem_global *mem_glob = ttm->glob->mem_glob; + unsigned i; + int ret; + + if (ttm->state != tt_unpopulated) + return 0; + + for (i = 0; i < ttm->num_pages; ++i) { + ret = ttm_get_pages(&ttm->pages[i], 1, + ttm->page_flags, + ttm->caching_state); + if (ret != 0) { + ttm_pool_unpopulate(ttm); + return -ENOMEM; + } + + ret = ttm_mem_global_alloc_page(mem_glob, ttm->pages[i], + false, false); + if (unlikely(ret != 0)) { + ttm_pool_unpopulate(ttm); + return -ENOMEM; + } + } + + if (unlikely(ttm->page_flags & TTM_PAGE_FLAG_SWAPPED)) { + ret = ttm_tt_swapin(ttm); + if (unlikely(ret != 0)) { + ttm_pool_unpopulate(ttm); + return ret; + } + } + + ttm->state = tt_unbound; + return 0; +} +EXPORT_SYMBOL(ttm_pool_populate); + +void ttm_pool_unpopulate(struct ttm_tt *ttm) +{ + unsigned i; + + for (i = 0; i < ttm->num_pages; ++i) { + if (ttm->pages[i]) { + ttm_mem_global_free_page(ttm->glob->mem_glob, + ttm->pages[i]); + ttm_put_pages(&ttm->pages[i], 1, + ttm->page_flags, + ttm->caching_state); + } + } + ttm->state = tt_unpopulated; +} +EXPORT_SYMBOL(ttm_pool_unpopulate); + +int ttm_page_alloc_debugfs(struct seq_file *m, void *data) +{ + struct ttm_page_pool *p; + unsigned i; + char *h[] = {"pool", "refills", "pages freed", "size"}; + if (!_manager) { + seq_printf(m, "No pool allocator running.\n"); + return 0; + } + seq_printf(m, "%6s %12s %13s %8s\n", + h[0], h[1], h[2], h[3]); + for (i = 0; i < NUM_POOLS; ++i) { + p = &_manager->pools[i]; + + seq_printf(m, "%6s %12ld %13ld %8d\n", + p->name, p->nrefills, + p->nfrees, p->npages); + } + return 0; +} +EXPORT_SYMBOL(ttm_page_alloc_debugfs); diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c new file mode 100644 index 0000000..b8b3943 --- /dev/null +++ b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c @@ -0,0 +1,1131 @@ +/* + * Copyright 2011 (c) Oracle Corp. + + * 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 AUTHORS OR COPYRIGHT HOLDERS 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. + * + * Author: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> + */ + +/* + * A simple DMA pool losely based on dmapool.c. It has certain advantages + * over the DMA pools: + * - Pool collects resently freed pages for reuse (and hooks up to + * the shrinker). + * - Tracks currently in use pages + * - Tracks whether the page is UC, WB or cached (and reverts to WB + * when freed). + */ + +#define pr_fmt(fmt) "[TTM] " fmt + +#include <linux/dma-mapping.h> +#include <linux/list.h> +#include <linux/seq_file.h> /* for seq_printf */ +#include <linux/slab.h> +#include <linux/spinlock.h> +#include <linux/highmem.h> +#include <linux/mm_types.h> +#include <linux/module.h> +#include <linux/mm.h> +#include <linux/atomic.h> +#include <linux/device.h> +#include <linux/kthread.h> +#include <drm/ttm/ttm_bo_driver.h> +#include <drm/ttm/ttm_page_alloc.h> +#ifdef TTM_HAS_AGP +#include <asm/agp.h> +#endif + +#define NUM_PAGES_TO_ALLOC (PAGE_SIZE/sizeof(struct page *)) +#define SMALL_ALLOCATION 4 +#define FREE_ALL_PAGES (~0U) +/* times are in msecs */ +#define IS_UNDEFINED (0) +#define IS_WC (1<<1) +#define IS_UC (1<<2) +#define IS_CACHED (1<<3) +#define IS_DMA32 (1<<4) + +enum pool_type { + POOL_IS_UNDEFINED, + POOL_IS_WC = IS_WC, + POOL_IS_UC = IS_UC, + POOL_IS_CACHED = IS_CACHED, + POOL_IS_WC_DMA32 = IS_WC | IS_DMA32, + POOL_IS_UC_DMA32 = IS_UC | IS_DMA32, + POOL_IS_CACHED_DMA32 = IS_CACHED | IS_DMA32, +}; +/* + * The pool structure. There are usually six pools: + * - generic (not restricted to DMA32): + * - write combined, uncached, cached. + * - dma32 (up to 2^32 - so up 4GB): + * - write combined, uncached, cached. + * for each 'struct device'. The 'cached' is for pages that are actively used. + * The other ones can be shrunk by the shrinker API if neccessary. + * @pools: The 'struct device->dma_pools' link. + * @type: Type of the pool + * @lock: Protects the inuse_list and free_list from concurrnet access. Must be + * used with irqsave/irqrestore variants because pool allocator maybe called + * from delayed work. + * @inuse_list: Pool of pages that are in use. The order is very important and + * it is in the order that the TTM pages that are put back are in. + * @free_list: Pool of pages that are free to be used. No order requirements. + * @dev: The device that is associated with these pools. + * @size: Size used during DMA allocation. + * @npages_free: Count of available pages for re-use. + * @npages_in_use: Count of pages that are in use. + * @nfrees: Stats when pool is shrinking. + * @nrefills: Stats when the pool is grown. + * @gfp_flags: Flags to pass for alloc_page. + * @name: Name of the pool. + * @dev_name: Name derieved from dev - similar to how dev_info works. + * Used during shutdown as the dev_info during release is unavailable. + */ +struct dma_pool { + struct list_head pools; /* The 'struct device->dma_pools link */ + enum pool_type type; + spinlock_t lock; + struct list_head inuse_list; + struct list_head free_list; + struct device *dev; + unsigned size; + unsigned npages_free; + unsigned npages_in_use; + unsigned long nfrees; /* Stats when shrunk. */ + unsigned long nrefills; /* Stats when grown. */ + gfp_t gfp_flags; + char name[13]; /* "cached dma32" */ + char dev_name[64]; /* Constructed from dev */ +}; + +/* + * The accounting page keeping track of the allocated page along with + * the DMA address. + * @page_list: The link to the 'page_list' in 'struct dma_pool'. + * @vaddr: The virtual address of the page + * @dma: The bus address of the page. If the page is not allocated + * via the DMA API, it will be -1. + */ +struct dma_page { + struct list_head page_list; + void *vaddr; + struct page *p; + dma_addr_t dma; +}; + +/* + * Limits for the pool. They are handled without locks because only place where + * they may change is in sysfs store. They won't have immediate effect anyway + * so forcing serialization to access them is pointless. + */ + +struct ttm_pool_opts { + unsigned alloc_size; + unsigned max_size; + unsigned small; +}; + +/* + * Contains the list of all of the 'struct device' and their corresponding + * DMA pools. Guarded by _mutex->lock. + * @pools: The link to 'struct ttm_pool_manager->pools' + * @dev: The 'struct device' associated with the 'pool' + * @pool: The 'struct dma_pool' associated with the 'dev' + */ +struct device_pools { + struct list_head pools; + struct device *dev; + struct dma_pool *pool; +}; + +/* + * struct ttm_pool_manager - Holds memory pools for fast allocation + * + * @lock: Lock used when adding/removing from pools + * @pools: List of 'struct device' and 'struct dma_pool' tuples. + * @options: Limits for the pool. + * @npools: Total amount of pools in existence. + * @shrinker: The structure used by [un|]register_shrinker + */ +struct ttm_pool_manager { + struct mutex lock; + struct list_head pools; + struct ttm_pool_opts options; + unsigned npools; + struct shrinker mm_shrink; + struct kobject kobj; +}; + +static struct ttm_pool_manager *_manager; + +static struct attribute ttm_page_pool_max = { + .name = "pool_max_size", + .mode = S_IRUGO | S_IWUSR +}; +static struct attribute ttm_page_pool_small = { + .name = "pool_small_allocation", + .mode = S_IRUGO | S_IWUSR +}; +static struct attribute ttm_page_pool_alloc_size = { + .name = "pool_allocation_size", + .mode = S_IRUGO | S_IWUSR +}; + +static struct attribute *ttm_pool_attrs[] = { + &ttm_page_pool_max, + &ttm_page_pool_small, + &ttm_page_pool_alloc_size, + NULL +}; + +static void ttm_pool_kobj_release(struct kobject *kobj) +{ + struct ttm_pool_manager *m = + container_of(kobj, struct ttm_pool_manager, kobj); + kfree(m); +} + +static ssize_t ttm_pool_store(struct kobject *kobj, struct attribute *attr, + const char *buffer, size_t size) +{ + struct ttm_pool_manager *m = + container_of(kobj, struct ttm_pool_manager, kobj); + int chars; + unsigned val; + chars = sscanf(buffer, "%u", &val); + if (chars == 0) + return size; + + /* Convert kb to number of pages */ + val = val / (PAGE_SIZE >> 10); + + if (attr == &ttm_page_pool_max) + m->options.max_size = val; + else if (attr == &ttm_page_pool_small) + m->options.small = val; + else if (attr == &ttm_page_pool_alloc_size) { + if (val > NUM_PAGES_TO_ALLOC*8) { + pr_err("Setting allocation size to %lu is not allowed. Recommended size is %lu\n", + NUM_PAGES_TO_ALLOC*(PAGE_SIZE >> 7), + NUM_PAGES_TO_ALLOC*(PAGE_SIZE >> 10)); + return size; + } else if (val > NUM_PAGES_TO_ALLOC) { + pr_warn("Setting allocation size to larger than %lu is not recommended\n", + NUM_PAGES_TO_ALLOC*(PAGE_SIZE >> 10)); + } + m->options.alloc_size = val; + } + + return size; +} + +static ssize_t ttm_pool_show(struct kobject *kobj, struct attribute *attr, + char *buffer) +{ + struct ttm_pool_manager *m = + container_of(kobj, struct ttm_pool_manager, kobj); + unsigned val = 0; + + if (attr == &ttm_page_pool_max) + val = m->options.max_size; + else if (attr == &ttm_page_pool_small) + val = m->options.small; + else if (attr == &ttm_page_pool_alloc_size) + val = m->options.alloc_size; + + val = val * (PAGE_SIZE >> 10); + + return snprintf(buffer, PAGE_SIZE, "%u\n", val); +} + +static const struct sysfs_ops ttm_pool_sysfs_ops = { + .show = &ttm_pool_show, + .store = &ttm_pool_store, +}; + +static struct kobj_type ttm_pool_kobj_type = { + .release = &ttm_pool_kobj_release, + .sysfs_ops = &ttm_pool_sysfs_ops, + .default_attrs = ttm_pool_attrs, +}; + +#ifndef CONFIG_X86 +static int set_pages_array_wb(struct page **pages, int addrinarray) +{ +#ifdef TTM_HAS_AGP + int i; + + for (i = 0; i < addrinarray; i++) + unmap_page_from_agp(pages[i]); +#endif + return 0; +} + +static int set_pages_array_wc(struct page **pages, int addrinarray) +{ +#ifdef TTM_HAS_AGP + int i; + + for (i = 0; i < addrinarray; i++) + map_page_into_agp(pages[i]); +#endif + return 0; +} + +static int set_pages_array_uc(struct page **pages, int addrinarray) +{ +#ifdef TTM_HAS_AGP + int i; + + for (i = 0; i < addrinarray; i++) + map_page_into_agp(pages[i]); +#endif + return 0; +} +#endif /* for !CONFIG_X86 */ + +static int ttm_set_pages_caching(struct dma_pool *pool, + struct page **pages, unsigned cpages) +{ + int r = 0; + /* Set page caching */ + if (pool->type & IS_UC) { + r = set_pages_array_uc(pages, cpages); + if (r) + pr_err("%s: Failed to set %d pages to uc!\n", + pool->dev_name, cpages); + } + if (pool->type & IS_WC) { + r = set_pages_array_wc(pages, cpages); + if (r) + pr_err("%s: Failed to set %d pages to wc!\n", + pool->dev_name, cpages); + } + return r; +} + +static void __ttm_dma_free_page(struct dma_pool *pool, struct dma_page *d_page) +{ + dma_addr_t dma = d_page->dma; + dma_free_coherent(pool->dev, pool->size, d_page->vaddr, dma); + + kfree(d_page); + d_page = NULL; +} +static struct dma_page *__ttm_dma_alloc_page(struct dma_pool *pool) +{ + struct dma_page *d_page; + + d_page = kmalloc(sizeof(struct dma_page), GFP_KERNEL); + if (!d_page) + return NULL; + + d_page->vaddr = dma_alloc_coherent(pool->dev, pool->size, + &d_page->dma, + pool->gfp_flags); + if (d_page->vaddr) + d_page->p = virt_to_page(d_page->vaddr); + else { + kfree(d_page); + d_page = NULL; + } + return d_page; +} +static enum pool_type ttm_to_type(int flags, enum ttm_caching_state cstate) +{ + enum pool_type type = IS_UNDEFINED; + + if (flags & TTM_PAGE_FLAG_DMA32) + type |= IS_DMA32; + if (cstate == tt_cached) + type |= IS_CACHED; + else if (cstate == tt_uncached) + type |= IS_UC; + else + type |= IS_WC; + + return type; +} + +static void ttm_pool_update_free_locked(struct dma_pool *pool, + unsigned freed_pages) +{ + pool->npages_free -= freed_pages; + pool->nfrees += freed_pages; + +} + +/* set memory back to wb and free the pages. */ +static void ttm_dma_pages_put(struct dma_pool *pool, struct list_head *d_pages, + struct page *pages[], unsigned npages) +{ + struct dma_page *d_page, *tmp; + + /* Don't set WB on WB page pool. */ + if (npages && !(pool->type & IS_CACHED) && + set_pages_array_wb(pages, npages)) + pr_err("%s: Failed to set %d pages to wb!\n", + pool->dev_name, npages); + + list_for_each_entry_safe(d_page, tmp, d_pages, page_list) { + list_del(&d_page->page_list); + __ttm_dma_free_page(pool, d_page); + } +} + +static void ttm_dma_page_put(struct dma_pool *pool, struct dma_page *d_page) +{ + /* Don't set WB on WB page pool. */ + if (!(pool->type & IS_CACHED) && set_pages_array_wb(&d_page->p, 1)) + pr_err("%s: Failed to set %d pages to wb!\n", + pool->dev_name, 1); + + list_del(&d_page->page_list); + __ttm_dma_free_page(pool, d_page); +} + +/* + * Free pages from pool. + * + * To prevent hogging the ttm_swap process we only free NUM_PAGES_TO_ALLOC + * number of pages in one go. + * + * @pool: to free the pages from + * @nr_free: If set to true will free all pages in pool + **/ +static unsigned ttm_dma_page_pool_free(struct dma_pool *pool, unsigned nr_free) +{ + unsigned long irq_flags; + struct dma_page *dma_p, *tmp; + struct page **pages_to_free; + struct list_head d_pages; + unsigned freed_pages = 0, + npages_to_free = nr_free; + + if (NUM_PAGES_TO_ALLOC < nr_free) + npages_to_free = NUM_PAGES_TO_ALLOC; +#if 0 + if (nr_free > 1) { + pr_debug("%s: (%s:%d) Attempting to free %d (%d) pages\n", + pool->dev_name, pool->name, current->pid, + npages_to_free, nr_free); + } +#endif + pages_to_free = kmalloc(npages_to_free * sizeof(struct page *), + GFP_KERNEL); + + if (!pages_to_free) { + pr_err("%s: Failed to allocate memory for pool free operation\n", + pool->dev_name); + return 0; + } + INIT_LIST_HEAD(&d_pages); +restart: + spin_lock_irqsave(&pool->lock, irq_flags); + + /* We picking the oldest ones off the list */ + list_for_each_entry_safe_reverse(dma_p, tmp, &pool->free_list, + page_list) { + if (freed_pages >= npages_to_free) + break; + + /* Move the dma_page from one list to another. */ + list_move(&dma_p->page_list, &d_pages); + + pages_to_free[freed_pages++] = dma_p->p; + /* We can only remove NUM_PAGES_TO_ALLOC at a time. */ + if (freed_pages >= NUM_PAGES_TO_ALLOC) { + + ttm_pool_update_free_locked(pool, freed_pages); + /** + * Because changing page caching is costly + * we unlock the pool to prevent stalling. + */ + spin_unlock_irqrestore(&pool->lock, irq_flags); + + ttm_dma_pages_put(pool, &d_pages, pages_to_free, + freed_pages); + + INIT_LIST_HEAD(&d_pages); + + if (likely(nr_free != FREE_ALL_PAGES)) + nr_free -= freed_pages; + + if (NUM_PAGES_TO_ALLOC >= nr_free) + npages_to_free = nr_free; + else + npages_to_free = NUM_PAGES_TO_ALLOC; + + freed_pages = 0; + + /* free all so restart the processing */ + if (nr_free) + goto restart; + + /* Not allowed to fall through or break because + * following context is inside spinlock while we are + * outside here. + */ + goto out; + + } + } + + /* remove range of pages from the pool */ + if (freed_pages) { + ttm_pool_update_free_locked(pool, freed_pages); + nr_free -= freed_pages; + } + + spin_unlock_irqrestore(&pool->lock, irq_flags); + + if (freed_pages) + ttm_dma_pages_put(pool, &d_pages, pages_to_free, freed_pages); +out: + kfree(pages_to_free); + return nr_free; +} + +static void ttm_dma_free_pool(struct device *dev, enum pool_type type) +{ + struct device_pools *p; + struct dma_pool *pool; + + if (!dev) + return; + + mutex_lock(&_manager->lock); + list_for_each_entry_reverse(p, &_manager->pools, pools) { + if (p->dev != dev) + continue; + pool = p->pool; + if (pool->type != type) + continue; + + list_del(&p->pools); + kfree(p); + _manager->npools--; + break; + } + list_for_each_entry_reverse(pool, &dev->dma_pools, pools) { + if (pool->type != type) + continue; + /* Takes a spinlock.. */ + ttm_dma_page_pool_free(pool, FREE_ALL_PAGES); + WARN_ON(((pool->npages_in_use + pool->npages_free) != 0)); + /* This code path is called after _all_ references to the + * struct device has been dropped - so nobody should be + * touching it. In case somebody is trying to _add_ we are + * guarded by the mutex. */ + list_del(&pool->pools); + kfree(pool); + break; + } + mutex_unlock(&_manager->lock); +} + +/* + * On free-ing of the 'struct device' this deconstructor is run. + * Albeit the pool might have already been freed earlier. + */ +static void ttm_dma_pool_release(struct device *dev, void *res) +{ + struct dma_pool *pool = *(struct dma_pool **)res; + + if (pool) + ttm_dma_free_pool(dev, pool->type); +} + +static int ttm_dma_pool_match(struct device *dev, void *res, void *match_data) +{ + return *(struct dma_pool **)res == match_data; +} + +static struct dma_pool *ttm_dma_pool_init(struct device *dev, gfp_t flags, + enum pool_type type) +{ + char *n[] = {"wc", "uc", "cached", " dma32", "unknown",}; + enum pool_type t[] = {IS_WC, IS_UC, IS_CACHED, IS_DMA32, IS_UNDEFINED}; + struct device_pools *sec_pool = NULL; + struct dma_pool *pool = NULL, **ptr; + unsigned i; + int ret = -ENODEV; + char *p; + + if (!dev) + return NULL; + + ptr = devres_alloc(ttm_dma_pool_release, sizeof(*ptr), GFP_KERNEL); + if (!ptr) + return NULL; + + ret = -ENOMEM; + + pool = kmalloc_node(sizeof(struct dma_pool), GFP_KERNEL, + dev_to_node(dev)); + if (!pool) + goto err_mem; + + sec_pool = kmalloc_node(sizeof(struct device_pools), GFP_KERNEL, + dev_to_node(dev)); + if (!sec_pool) + goto err_mem; + + INIT_LIST_HEAD(&sec_pool->pools); + sec_pool->dev = dev; + sec_pool->pool = pool; + + INIT_LIST_HEAD(&pool->free_list); + INIT_LIST_HEAD(&pool->inuse_list); + INIT_LIST_HEAD(&pool->pools); + spin_lock_init(&pool->lock); + pool->dev = dev; + pool->npages_free = pool->npages_in_use = 0; + pool->nfrees = 0; + pool->gfp_flags = flags; + pool->size = PAGE_SIZE; + pool->type = type; + pool->nrefills = 0; + p = pool->name; + for (i = 0; i < 5; i++) { + if (type & t[i]) { + p += snprintf(p, sizeof(pool->name) - (p - pool->name), + "%s", n[i]); + } + } + *p = 0; + /* We copy the name for pr_ calls b/c when dma_pool_destroy is called + * - the kobj->name has already been deallocated.*/ + snprintf(pool->dev_name, sizeof(pool->dev_name), "%s %s", + dev_driver_string(dev), dev_name(dev)); + mutex_lock(&_manager->lock); + /* You can get the dma_pool from either the global: */ + list_add(&sec_pool->pools, &_manager->pools); + _manager->npools++; + /* or from 'struct device': */ + list_add(&pool->pools, &dev->dma_pools); + mutex_unlock(&_manager->lock); + + *ptr = pool; + devres_add(dev, ptr); + + return pool; +err_mem: + devres_free(ptr); + kfree(sec_pool); + kfree(pool); + return ERR_PTR(ret); +} + +static struct dma_pool *ttm_dma_find_pool(struct device *dev, + enum pool_type type) +{ + struct dma_pool *pool, *tmp, *found = NULL; + + if (type == IS_UNDEFINED) + return found; + + /* NB: We iterate on the 'struct dev' which has no spinlock, but + * it does have a kref which we have taken. The kref is taken during + * graphic driver loading - in the drm_pci_init it calls either + * pci_dev_get or pci_register_driver which both end up taking a kref + * on 'struct device'. + * + * On teardown, the graphic drivers end up quiescing the TTM (put_pages) + * and calls the dev_res deconstructors: ttm_dma_pool_release. The nice + * thing is at that point of time there are no pages associated with the + * driver so this function will not be called. + */ + list_for_each_entry_safe(pool, tmp, &dev->dma_pools, pools) { + if (pool->type != type) + continue; + found = pool; + break; + } + return found; +} + +/* + * Free pages the pages that failed to change the caching state. If there + * are pages that have changed their caching state already put them to the + * pool. + */ +static void ttm_dma_handle_caching_state_failure(struct dma_pool *pool, + struct list_head *d_pages, + struct page **failed_pages, + unsigned cpages) +{ + struct dma_page *d_page, *tmp; + struct page *p; + unsigned i = 0; + + p = failed_pages[0]; + if (!p) + return; + /* Find the failed page. */ + list_for_each_entry_safe(d_page, tmp, d_pages, page_list) { + if (d_page->p != p) + continue; + /* .. and then progress over the full list. */ + list_del(&d_page->page_list); + __ttm_dma_free_page(pool, d_page); + if (++i < cpages) + p = failed_pages[i]; + else + break; + } + +} + +/* + * Allocate 'count' pages, and put 'need' number of them on the + * 'pages' and as well on the 'dma_address' starting at 'dma_offset' offset. + * The full list of pages should also be on 'd_pages'. + * We return zero for success, and negative numbers as errors. + */ +static int ttm_dma_pool_alloc_new_pages(struct dma_pool *pool, + struct list_head *d_pages, + unsigned count) +{ + struct page **caching_array; + struct dma_page *dma_p; + struct page *p; + int r = 0; + unsigned i, cpages; + unsigned max_cpages = min(count, + (unsigned)(PAGE_SIZE/sizeof(struct page *))); + + /* allocate array for page caching change */ + caching_array = kmalloc(max_cpages*sizeof(struct page *), GFP_KERNEL); + + if (!caching_array) { + pr_err("%s: Unable to allocate table for new pages\n", + pool->dev_name); + return -ENOMEM; + } + + if (count > 1) { + pr_debug("%s: (%s:%d) Getting %d pages\n", + pool->dev_name, pool->name, current->pid, count); + } + + for (i = 0, cpages = 0; i < count; ++i) { + dma_p = __ttm_dma_alloc_page(pool); + if (!dma_p) { + pr_err("%s: Unable to get page %u\n", + pool->dev_name, i); + + /* store already allocated pages in the pool after + * setting the caching state */ + if (cpages) { + r = ttm_set_pages_caching(pool, caching_array, + cpages); + if (r) + ttm_dma_handle_caching_state_failure( + pool, d_pages, caching_array, + cpages); + } + r = -ENOMEM; + goto out; + } + p = dma_p->p; +#ifdef CONFIG_HIGHMEM + /* gfp flags of highmem page should never be dma32 so we + * we should be fine in such case + */ + if (!PageHighMem(p)) +#endif + { + caching_array[cpages++] = p; + if (cpages == max_cpages) { + /* Note: Cannot hold the spinlock */ + r = ttm_set_pages_caching(pool, caching_array, + cpages); + if (r) { + ttm_dma_handle_caching_state_failure( + pool, d_pages, caching_array, + cpages); + goto out; + } + cpages = 0; + } + } + list_add(&dma_p->page_list, d_pages); + } + + if (cpages) { + r = ttm_set_pages_caching(pool, caching_array, cpages); + if (r) + ttm_dma_handle_caching_state_failure(pool, d_pages, + caching_array, cpages); + } +out: + kfree(caching_array); + return r; +} + +/* + * @return count of pages still required to fulfill the request. + */ +static int ttm_dma_page_pool_fill_locked(struct dma_pool *pool, + unsigned long *irq_flags) +{ + unsigned count = _manager->options.small; + int r = pool->npages_free; + + if (count > pool->npages_free) { + struct list_head d_pages; + + INIT_LIST_HEAD(&d_pages); + + spin_unlock_irqrestore(&pool->lock, *irq_flags); + + /* Returns how many more are neccessary to fulfill the + * request. */ + r = ttm_dma_pool_alloc_new_pages(pool, &d_pages, count); + + spin_lock_irqsave(&pool->lock, *irq_flags); + if (!r) { + /* Add the fresh to the end.. */ + list_splice(&d_pages, &pool->free_list); + ++pool->nrefills; + pool->npages_free += count; + r = count; + } else { + struct dma_page *d_page; + unsigned cpages = 0; + + pr_err("%s: Failed to fill %s pool (r:%d)!\n", + pool->dev_name, pool->name, r); + + list_for_each_entry(d_page, &d_pages, page_list) { + cpages++; + } + list_splice_tail(&d_pages, &pool->free_list); + pool->npages_free += cpages; + r = cpages; + } + } + return r; +} + +/* + * @return count of pages still required to fulfill the request. + * The populate list is actually a stack (not that is matters as TTM + * allocates one page at a time. + */ +static int ttm_dma_pool_get_pages(struct dma_pool *pool, + struct ttm_dma_tt *ttm_dma, + unsigned index) +{ + struct dma_page *d_page; + struct ttm_tt *ttm = &ttm_dma->ttm; + unsigned long irq_flags; + int count, r = -ENOMEM; + + spin_lock_irqsave(&pool->lock, irq_flags); + count = ttm_dma_page_pool_fill_locked(pool, &irq_flags); + if (count) { + d_page = list_first_entry(&pool->free_list, struct dma_page, page_list); + ttm->pages[index] = d_page->p; + ttm_dma->dma_address[index] = d_page->dma; + list_move_tail(&d_page->page_list, &ttm_dma->pages_list); + r = 0; + pool->npages_in_use += 1; + pool->npages_free -= 1; + } + spin_unlock_irqrestore(&pool->lock, irq_flags); + return r; +} + +/* + * On success pages list will hold count number of correctly + * cached pages. On failure will hold the negative return value (-ENOMEM, etc). + */ +int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev) +{ + struct ttm_tt *ttm = &ttm_dma->ttm; + struct ttm_mem_global *mem_glob = ttm->glob->mem_glob; + struct dma_pool *pool; + enum pool_type type; + unsigned i; + gfp_t gfp_flags; + int ret; + + if (ttm->state != tt_unpopulated) + return 0; + + type = ttm_to_type(ttm->page_flags, ttm->caching_state); + if (ttm->page_flags & TTM_PAGE_FLAG_DMA32) + gfp_flags = GFP_USER | GFP_DMA32; + else + gfp_flags = GFP_HIGHUSER; + if (ttm->page_flags & TTM_PAGE_FLAG_ZERO_ALLOC) + gfp_flags |= __GFP_ZERO; + + pool = ttm_dma_find_pool(dev, type); + if (!pool) { + pool = ttm_dma_pool_init(dev, gfp_flags, type); + if (IS_ERR_OR_NULL(pool)) { + return -ENOMEM; + } + } + + INIT_LIST_HEAD(&ttm_dma->pages_list); + for (i = 0; i < ttm->num_pages; ++i) { + ret = ttm_dma_pool_get_pages(pool, ttm_dma, i); + if (ret != 0) { + ttm_dma_unpopulate(ttm_dma, dev); + return -ENOMEM; + } + + ret = ttm_mem_global_alloc_page(mem_glob, ttm->pages[i], + false, false); + if (unlikely(ret != 0)) { + ttm_dma_unpopulate(ttm_dma, dev); + return -ENOMEM; + } + } + + if (unlikely(ttm->page_flags & TTM_PAGE_FLAG_SWAPPED)) { + ret = ttm_tt_swapin(ttm); + if (unlikely(ret != 0)) { + ttm_dma_unpopulate(ttm_dma, dev); + return ret; + } + } + + ttm->state = tt_unbound; + return 0; +} +EXPORT_SYMBOL_GPL(ttm_dma_populate); + +/* Get good estimation how many pages are free in pools */ +static int ttm_dma_pool_get_num_unused_pages(void) +{ + struct device_pools *p; + unsigned total = 0; + + mutex_lock(&_manager->lock); + list_for_each_entry(p, &_manager->pools, pools) + total += p->pool->npages_free; + mutex_unlock(&_manager->lock); + return total; +} + +/* Put all pages in pages list to correct pool to wait for reuse */ +void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev) +{ + struct ttm_tt *ttm = &ttm_dma->ttm; + struct dma_pool *pool; + struct dma_page *d_page, *next; + enum pool_type type; + bool is_cached = false; + unsigned count = 0, i, npages = 0; + unsigned long irq_flags; + + type = ttm_to_type(ttm->page_flags, ttm->caching_state); + pool = ttm_dma_find_pool(dev, type); + if (!pool) + return; + + is_cached = (ttm_dma_find_pool(pool->dev, + ttm_to_type(ttm->page_flags, tt_cached)) == pool); + + /* make sure pages array match list and count number of pages */ + list_for_each_entry(d_page, &ttm_dma->pages_list, page_list) { + ttm->pages[count] = d_page->p; + count++; + } + + spin_lock_irqsave(&pool->lock, irq_flags); + pool->npages_in_use -= count; + if (is_cached) { + pool->nfrees += count; + } else { + pool->npages_free += count; + list_splice(&ttm_dma->pages_list, &pool->free_list); + npages = count; + if (pool->npages_free > _manager->options.max_size) { + npages = pool->npages_free - _manager->options.max_size; + /* free at least NUM_PAGES_TO_ALLOC number of pages + * to reduce calls to set_memory_wb */ + if (npages < NUM_PAGES_TO_ALLOC) + npages = NUM_PAGES_TO_ALLOC; + } + } + spin_unlock_irqrestore(&pool->lock, irq_flags); + + if (is_cached) { + list_for_each_entry_safe(d_page, next, &ttm_dma->pages_list, page_list) { + ttm_mem_global_free_page(ttm->glob->mem_glob, + d_page->p); + ttm_dma_page_put(pool, d_page); + } + } else { + for (i = 0; i < count; i++) { + ttm_mem_global_free_page(ttm->glob->mem_glob, + ttm->pages[i]); + } + } + + INIT_LIST_HEAD(&ttm_dma->pages_list); + for (i = 0; i < ttm->num_pages; i++) { + ttm->pages[i] = NULL; + ttm_dma->dma_address[i] = 0; + } + + /* shrink pool if necessary (only on !is_cached pools)*/ + if (npages) + ttm_dma_page_pool_free(pool, npages); + ttm->state = tt_unpopulated; +} +EXPORT_SYMBOL_GPL(ttm_dma_unpopulate); + +/** + * Callback for mm to request pool to reduce number of page held. + */ +static int ttm_dma_pool_mm_shrink(struct shrinker *shrink, + struct shrink_control *sc) +{ + static atomic_t start_pool = ATOMIC_INIT(0); + unsigned idx = 0; + unsigned pool_offset = atomic_add_return(1, &start_pool); + unsigned shrink_pages = sc->nr_to_scan; + struct device_pools *p; + + if (list_empty(&_manager->pools)) + return 0; + + mutex_lock(&_manager->lock); + pool_offset = pool_offset % _manager->npools; + list_for_each_entry(p, &_manager->pools, pools) { + unsigned nr_free; + + if (!p->dev) + continue; + if (shrink_pages == 0) + break; + /* Do it in round-robin fashion. */ + if (++idx < pool_offset) + continue; + nr_free = shrink_pages; + shrink_pages = ttm_dma_page_pool_free(p->pool, nr_free); + pr_debug("%s: (%s:%d) Asked to shrink %d, have %d more to go\n", + p->pool->dev_name, p->pool->name, current->pid, + nr_free, shrink_pages); + } + mutex_unlock(&_manager->lock); + /* return estimated number of unused pages in pool */ + return ttm_dma_pool_get_num_unused_pages(); +} + +static void ttm_dma_pool_mm_shrink_init(struct ttm_pool_manager *manager) +{ + manager->mm_shrink.shrink = &ttm_dma_pool_mm_shrink; + manager->mm_shrink.seeks = 1; + register_shrinker(&manager->mm_shrink); +} + +static void ttm_dma_pool_mm_shrink_fini(struct ttm_pool_manager *manager) +{ + unregister_shrinker(&manager->mm_shrink); +} + +int ttm_dma_page_alloc_init(struct ttm_mem_global *glob, unsigned max_pages) +{ + int ret = -ENOMEM; + + WARN_ON(_manager); + + pr_info("Initializing DMA pool allocator\n"); + + _manager = kzalloc(sizeof(*_manager), GFP_KERNEL); + if (!_manager) + goto err; + + mutex_init(&_manager->lock); + INIT_LIST_HEAD(&_manager->pools); + + _manager->options.max_size = max_pages; + _manager->options.small = SMALL_ALLOCATION; + _manager->options.alloc_size = NUM_PAGES_TO_ALLOC; + + /* This takes care of auto-freeing the _manager */ + ret = kobject_init_and_add(&_manager->kobj, &ttm_pool_kobj_type, + &glob->kobj, "dma_pool"); + if (unlikely(ret != 0)) { + kobject_put(&_manager->kobj); + goto err; + } + ttm_dma_pool_mm_shrink_init(_manager); + return 0; +err: + return ret; +} + +void ttm_dma_page_alloc_fini(void) +{ + struct device_pools *p, *t; + + pr_info("Finalizing DMA pool allocator\n"); + ttm_dma_pool_mm_shrink_fini(_manager); + + list_for_each_entry_safe_reverse(p, t, &_manager->pools, pools) { + dev_dbg(p->dev, "(%s:%d) Freeing.\n", p->pool->name, + current->pid); + WARN_ON(devres_destroy(p->dev, ttm_dma_pool_release, + ttm_dma_pool_match, p->pool)); + ttm_dma_free_pool(p->dev, p->pool->type); + } + kobject_put(&_manager->kobj); + _manager = NULL; +} + +int ttm_dma_page_alloc_debugfs(struct seq_file *m, void *data) +{ + struct device_pools *p; + struct dma_pool *pool = NULL; + char *h[] = {"pool", "refills", "pages freed", "inuse", "available", + "name", "virt", "busaddr"}; + + if (!_manager) { + seq_printf(m, "No pool allocator running.\n"); + return 0; + } + seq_printf(m, "%13s %12s %13s %8s %8s %8s\n", + h[0], h[1], h[2], h[3], h[4], h[5]); + mutex_lock(&_manager->lock); + list_for_each_entry(p, &_manager->pools, pools) { + struct device *dev = p->dev; + if (!dev) + continue; + pool = p->pool; + seq_printf(m, "%13s %12ld %13ld %8d %8d %8s\n", + pool->name, pool->nrefills, + pool->nfrees, pool->npages_in_use, + pool->npages_free, + pool->dev_name); + } + mutex_unlock(&_manager->lock); + return 0; +} +EXPORT_SYMBOL_GPL(ttm_dma_page_alloc_debugfs); diff --git a/drivers/gpu/drm/ttm/ttm_tt.c b/drivers/gpu/drm/ttm/ttm_tt.c new file mode 100644 index 0000000..210d503 --- /dev/null +++ b/drivers/gpu/drm/ttm/ttm_tt.c @@ -0,0 +1,377 @@ +/************************************************************************** + * + * Copyright (c) 2006-2009 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 <thellstrom-at-vmware-dot-com> + */ + +#define pr_fmt(fmt) "[TTM] " fmt + +#include <linux/sched.h> +#include <linux/highmem.h> +#include <linux/pagemap.h> +#include <linux/shmem_fs.h> +#include <linux/file.h> +#include <linux/swap.h> +#include <linux/slab.h> +#include <linux/export.h> +#include <drm/drm_cache.h> +#include <drm/drm_mem_util.h> +#include <drm/ttm/ttm_module.h> +#include <drm/ttm/ttm_bo_driver.h> +#include <drm/ttm/ttm_placement.h> +#include <drm/ttm/ttm_page_alloc.h> + +/** + * Allocates storage for pointers to the pages that back the ttm. + */ +static void ttm_tt_alloc_page_directory(struct ttm_tt *ttm) +{ + ttm->pages = drm_calloc_large(ttm->num_pages, sizeof(void*)); +} + +static void ttm_dma_tt_alloc_page_directory(struct ttm_dma_tt *ttm) +{ + ttm->ttm.pages = drm_calloc_large(ttm->ttm.num_pages, sizeof(void*)); + ttm->dma_address = drm_calloc_large(ttm->ttm.num_pages, + sizeof(*ttm->dma_address)); +} + +#ifdef CONFIG_X86 +static inline int ttm_tt_set_page_caching(struct page *p, + enum ttm_caching_state c_old, + enum ttm_caching_state c_new) +{ + int ret = 0; + + if (PageHighMem(p)) + return 0; + + if (c_old != tt_cached) { + /* p isn't in the default caching state, set it to + * writeback first to free its current memtype. */ + + ret = set_pages_wb(p, 1); + if (ret) + return ret; + } + + if (c_new == tt_wc) + ret = set_memory_wc((unsigned long) page_address(p), 1); + else if (c_new == tt_uncached) + ret = set_pages_uc(p, 1); + + return ret; +} +#else /* CONFIG_X86 */ +static inline int ttm_tt_set_page_caching(struct page *p, + enum ttm_caching_state c_old, + enum ttm_caching_state c_new) +{ + return 0; +} +#endif /* CONFIG_X86 */ + +/* + * Change caching policy for the linear kernel map + * for range of pages in a ttm. + */ + +static int ttm_tt_set_caching(struct ttm_tt *ttm, + enum ttm_caching_state c_state) +{ + int i, j; + struct page *cur_page; + int ret; + + if (ttm->caching_state == c_state) + return 0; + + if (ttm->state == tt_unpopulated) { + /* Change caching but don't populate */ + ttm->caching_state = c_state; + return 0; + } + + if (ttm->caching_state == tt_cached) + drm_clflush_pages(ttm->pages, ttm->num_pages); + + for (i = 0; i < ttm->num_pages; ++i) { + cur_page = ttm->pages[i]; + if (likely(cur_page != NULL)) { + ret = ttm_tt_set_page_caching(cur_page, + ttm->caching_state, + c_state); + if (unlikely(ret != 0)) + goto out_err; + } + } + + ttm->caching_state = c_state; + + return 0; + +out_err: + for (j = 0; j < i; ++j) { + cur_page = ttm->pages[j]; + if (likely(cur_page != NULL)) { + (void)ttm_tt_set_page_caching(cur_page, c_state, + ttm->caching_state); + } + } + + return ret; +} + +int ttm_tt_set_placement_caching(struct ttm_tt *ttm, uint32_t placement) +{ + enum ttm_caching_state state; + + if (placement & TTM_PL_FLAG_WC) + state = tt_wc; + else if (placement & TTM_PL_FLAG_UNCACHED) + state = tt_uncached; + else + state = tt_cached; + + return ttm_tt_set_caching(ttm, state); +} +EXPORT_SYMBOL(ttm_tt_set_placement_caching); + +void ttm_tt_destroy(struct ttm_tt *ttm) +{ + if (unlikely(ttm == NULL)) + return; + + if (ttm->state == tt_bound) { + ttm_tt_unbind(ttm); + } + + if (ttm->state == tt_unbound) { + ttm->bdev->driver->ttm_tt_unpopulate(ttm); + } + + if (!(ttm->page_flags & TTM_PAGE_FLAG_PERSISTENT_SWAP) && + ttm->swap_storage) + fput(ttm->swap_storage); + + ttm->swap_storage = NULL; + ttm->func->destroy(ttm); +} + +int ttm_tt_init(struct ttm_tt *ttm, struct ttm_bo_device *bdev, + unsigned long size, uint32_t page_flags, + struct page *dummy_read_page) +{ + ttm->bdev = bdev; + ttm->glob = bdev->glob; + ttm->num_pages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; + ttm->caching_state = tt_cached; + ttm->page_flags = page_flags; + ttm->dummy_read_page = dummy_read_page; + ttm->state = tt_unpopulated; + ttm->swap_storage = NULL; + + ttm_tt_alloc_page_directory(ttm); + if (!ttm->pages) { + ttm_tt_destroy(ttm); + pr_err("Failed allocating page table\n"); + return -ENOMEM; + } + return 0; +} +EXPORT_SYMBOL(ttm_tt_init); + +void ttm_tt_fini(struct ttm_tt *ttm) +{ + drm_free_large(ttm->pages); + ttm->pages = NULL; +} +EXPORT_SYMBOL(ttm_tt_fini); + +int ttm_dma_tt_init(struct ttm_dma_tt *ttm_dma, struct ttm_bo_device *bdev, + unsigned long size, uint32_t page_flags, + struct page *dummy_read_page) +{ + struct ttm_tt *ttm = &ttm_dma->ttm; + + ttm->bdev = bdev; + ttm->glob = bdev->glob; + ttm->num_pages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; + ttm->caching_state = tt_cached; + ttm->page_flags = page_flags; + ttm->dummy_read_page = dummy_read_page; + ttm->state = tt_unpopulated; + ttm->swap_storage = NULL; + + INIT_LIST_HEAD(&ttm_dma->pages_list); + ttm_dma_tt_alloc_page_directory(ttm_dma); + if (!ttm->pages || !ttm_dma->dma_address) { + ttm_tt_destroy(ttm); + pr_err("Failed allocating page table\n"); + return -ENOMEM; + } + return 0; +} +EXPORT_SYMBOL(ttm_dma_tt_init); + +void ttm_dma_tt_fini(struct ttm_dma_tt *ttm_dma) +{ + struct ttm_tt *ttm = &ttm_dma->ttm; + + drm_free_large(ttm->pages); + ttm->pages = NULL; + drm_free_large(ttm_dma->dma_address); + ttm_dma->dma_address = NULL; +} +EXPORT_SYMBOL(ttm_dma_tt_fini); + +void ttm_tt_unbind(struct ttm_tt *ttm) +{ + int ret; + + if (ttm->state == tt_bound) { + ret = ttm->func->unbind(ttm); + BUG_ON(ret); + ttm->state = tt_unbound; + } +} + +int ttm_tt_bind(struct ttm_tt *ttm, struct ttm_mem_reg *bo_mem) +{ + int ret = 0; + + if (!ttm) + return -EINVAL; + + if (ttm->state == tt_bound) + return 0; + + ret = ttm->bdev->driver->ttm_tt_populate(ttm); + if (ret) + return ret; + + ret = ttm->func->bind(ttm, bo_mem); + if (unlikely(ret != 0)) + return ret; + + ttm->state = tt_bound; + + return 0; +} +EXPORT_SYMBOL(ttm_tt_bind); + +int ttm_tt_swapin(struct ttm_tt *ttm) +{ + struct address_space *swap_space; + struct file *swap_storage; + struct page *from_page; + struct page *to_page; + int i; + int ret = -ENOMEM; + + swap_storage = ttm->swap_storage; + BUG_ON(swap_storage == NULL); + + swap_space = file_inode(swap_storage)->i_mapping; + + for (i = 0; i < ttm->num_pages; ++i) { + from_page = shmem_read_mapping_page(swap_space, i); + if (IS_ERR(from_page)) { + ret = PTR_ERR(from_page); + goto out_err; + } + to_page = ttm->pages[i]; + if (unlikely(to_page == NULL)) + goto out_err; + + copy_highpage(to_page, from_page); + page_cache_release(from_page); + } + + if (!(ttm->page_flags & TTM_PAGE_FLAG_PERSISTENT_SWAP)) + fput(swap_storage); + ttm->swap_storage = NULL; + ttm->page_flags &= ~TTM_PAGE_FLAG_SWAPPED; + + return 0; +out_err: + return ret; +} + +int ttm_tt_swapout(struct ttm_tt *ttm, struct file *persistent_swap_storage) +{ + struct address_space *swap_space; + struct file *swap_storage; + struct page *from_page; + struct page *to_page; + int i; + int ret = -ENOMEM; + + BUG_ON(ttm->state != tt_unbound && ttm->state != tt_unpopulated); + BUG_ON(ttm->caching_state != tt_cached); + + if (!persistent_swap_storage) { + swap_storage = shmem_file_setup("ttm swap", + ttm->num_pages << PAGE_SHIFT, + 0); + if (unlikely(IS_ERR(swap_storage))) { + pr_err("Failed allocating swap storage\n"); + return PTR_ERR(swap_storage); + } + } else + swap_storage = persistent_swap_storage; + + swap_space = file_inode(swap_storage)->i_mapping; + + for (i = 0; i < ttm->num_pages; ++i) { + from_page = ttm->pages[i]; + if (unlikely(from_page == NULL)) + continue; + to_page = shmem_read_mapping_page(swap_space, i); + if (unlikely(IS_ERR(to_page))) { + ret = PTR_ERR(to_page); + goto out_err; + } + copy_highpage(to_page, from_page); + set_page_dirty(to_page); + mark_page_accessed(to_page); + page_cache_release(to_page); + } + + ttm->bdev->driver->ttm_tt_unpopulate(ttm); + ttm->swap_storage = swap_storage; + ttm->page_flags |= TTM_PAGE_FLAG_SWAPPED; + if (persistent_swap_storage) + ttm->page_flags |= TTM_PAGE_FLAG_PERSISTENT_SWAP; + + return 0; +out_err: + if (!persistent_swap_storage) + fput(swap_storage); + + return ret; +} |