diff --git a/drivers/gpu/drm/Kconfig b/drivers/gpu/drm/Kconfig index f5d46e7199d48d03c12bfbc7b866540e607cddf1..64a57adc0a2c13ca485f822c4da92f70052f1812 100644 --- a/drivers/gpu/drm/Kconfig +++ b/drivers/gpu/drm/Kconfig @@ -18,6 +18,14 @@ menuconfig DRM details. You should also select and configure AGP (/dev/agpgart) support. +config DRM_TTM + tristate "TTM memory manager" + depends on DRM + help + GPU memory management subsystem for devices with multiple + GPU memory types. Will be enabled automatically if a device driver + uses it. + config DRM_TDFX tristate "3dfx Banshee/Voodoo3+" depends on DRM && PCI diff --git a/drivers/gpu/drm/Makefile b/drivers/gpu/drm/Makefile index 4ec5061fa584aab4d070deabcf776de0994d3720..4e89ab08b7b8d9eecf6fccc6e16901f5e41f4ed9 100644 --- a/drivers/gpu/drm/Makefile +++ b/drivers/gpu/drm/Makefile @@ -26,4 +26,4 @@ obj-$(CONFIG_DRM_I915) += i915/ obj-$(CONFIG_DRM_SIS) += sis/ obj-$(CONFIG_DRM_SAVAGE)+= savage/ obj-$(CONFIG_DRM_VIA) +=via/ - +obj-$(CONFIG_DRM_TTM) += ttm/ diff --git a/drivers/gpu/drm/ttm/Makefile b/drivers/gpu/drm/ttm/Makefile new file mode 100644 index 0000000000000000000000000000000000000000..b0a9de7a57c22b65d043d895f049a4afb775fe76 --- /dev/null +++ b/drivers/gpu/drm/ttm/Makefile @@ -0,0 +1,8 @@ +# +# Makefile for the drm device driver. This driver provides support for the + +ccflags-y := -Iinclude/drm +ttm-y := ttm_agp_backend.o ttm_memory.o ttm_tt.o ttm_bo.o \ + ttm_bo_util.o ttm_bo_vm.o ttm_module.o ttm_global.o + +obj-$(CONFIG_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 0000000000000000000000000000000000000000..e8f6d2229d8c36bb6c297d6fc7a6b1ca576f39e5 --- /dev/null +++ b/drivers/gpu/drm/ttm/ttm_agp_backend.c @@ -0,0 +1,150 @@ +/************************************************************************** + * + * 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 + * Keith Packard. + */ + +#include "ttm/ttm_module.h" +#include "ttm/ttm_bo_driver.h" +#ifdef TTM_HAS_AGP +#include "ttm/ttm_placement.h" +#include +#include +#include +#include + +struct ttm_agp_backend { + struct ttm_backend backend; + struct agp_memory *mem; + struct agp_bridge_data *bridge; +}; + +static int ttm_agp_populate(struct ttm_backend *backend, + unsigned long num_pages, struct page **pages, + struct page *dummy_read_page) +{ + struct ttm_agp_backend *agp_be = + container_of(backend, struct ttm_agp_backend, backend); + struct page **cur_page, **last_page = pages + num_pages; + struct agp_memory *mem; + + mem = agp_allocate_memory(agp_be->bridge, num_pages, AGP_USER_MEMORY); + if (unlikely(mem == NULL)) + return -ENOMEM; + + mem->page_count = 0; + for (cur_page = pages; cur_page < last_page; ++cur_page) { + struct page *page = *cur_page; + if (!page) + page = dummy_read_page; + + mem->memory[mem->page_count++] = + phys_to_gart(page_to_phys(page)); + } + agp_be->mem = mem; + return 0; +} + +static int ttm_agp_bind(struct ttm_backend *backend, struct ttm_mem_reg *bo_mem) +{ + struct ttm_agp_backend *agp_be = + container_of(backend, struct ttm_agp_backend, backend); + struct agp_memory *mem = agp_be->mem; + int cached = (bo_mem->placement & TTM_PL_FLAG_CACHED); + int ret; + + mem->is_flushed = 1; + mem->type = (cached) ? AGP_USER_CACHED_MEMORY : AGP_USER_MEMORY; + + ret = agp_bind_memory(mem, bo_mem->mm_node->start); + if (ret) + printk(KERN_ERR TTM_PFX "AGP Bind memory failed.\n"); + + return ret; +} + +static int ttm_agp_unbind(struct ttm_backend *backend) +{ + struct ttm_agp_backend *agp_be = + container_of(backend, struct ttm_agp_backend, backend); + + if (agp_be->mem->is_bound) + return agp_unbind_memory(agp_be->mem); + else + return 0; +} + +static void ttm_agp_clear(struct ttm_backend *backend) +{ + struct ttm_agp_backend *agp_be = + container_of(backend, struct ttm_agp_backend, backend); + struct agp_memory *mem = agp_be->mem; + + if (mem) { + ttm_agp_unbind(backend); + agp_free_memory(mem); + } + agp_be->mem = NULL; +} + +static void ttm_agp_destroy(struct ttm_backend *backend) +{ + struct ttm_agp_backend *agp_be = + container_of(backend, struct ttm_agp_backend, backend); + + if (agp_be->mem) + ttm_agp_clear(backend); + kfree(agp_be); +} + +static struct ttm_backend_func ttm_agp_func = { + .populate = ttm_agp_populate, + .clear = ttm_agp_clear, + .bind = ttm_agp_bind, + .unbind = ttm_agp_unbind, + .destroy = ttm_agp_destroy, +}; + +struct ttm_backend *ttm_agp_backend_init(struct ttm_bo_device *bdev, + struct agp_bridge_data *bridge) +{ + 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->backend.func = &ttm_agp_func; + agp_be->backend.bdev = bdev; + return &agp_be->backend; +} +EXPORT_SYMBOL(ttm_agp_backend_init); + +#endif diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c new file mode 100644 index 0000000000000000000000000000000000000000..1587aeca7bea73d20ec27ced01de75bb28acdacf --- /dev/null +++ b/drivers/gpu/drm/ttm/ttm_bo.c @@ -0,0 +1,1698 @@ +/************************************************************************** + * + * 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 + */ + +#include "ttm/ttm_module.h" +#include "ttm/ttm_bo_driver.h" +#include "ttm/ttm_placement.h" +#include +#include +#include +#include +#include +#include + +#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 void ttm_bo_unmap_virtual(struct ttm_buffer_object *bo); +static int ttm_bo_swapout(struct ttm_mem_shrink *shrink); + +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; + + 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); + if (bo->destroy) + bo->destroy(bo); + else { + ttm_mem_global_free(bdev->mem_glob, bo->acc_size, false); + kfree(bo); + } +} + +int ttm_bo_wait_unreserved(struct ttm_buffer_object *bo, bool interruptible) +{ + + if (interruptible) { + int ret = 0; + + ret = wait_event_interruptible(bo->event_queue, + atomic_read(&bo->reserved) == 0); + if (unlikely(ret != 0)) + return -ERESTART; + } else { + wait_event(bo->event_queue, atomic_read(&bo->reserved) == 0); + } + return 0; +} + +static 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(!atomic_read(&bo->reserved)); + + 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, &bdev->swap_lru); + kref_get(&bo->list_kref); + } + } +} + +/** + * Call with the lru_lock held. + */ + +static 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_locked(struct ttm_buffer_object *bo, + bool interruptible, + bool no_wait, bool use_sequence, uint32_t sequence) +{ + struct ttm_bo_device *bdev = bo->bdev; + int ret; + + while (unlikely(atomic_cmpxchg(&bo->reserved, 0, 1) != 0)) { + if (use_sequence && bo->seq_valid && + (sequence - bo->val_seq < (1 << 31))) { + return -EAGAIN; + } + + if (no_wait) + return -EBUSY; + + spin_unlock(&bdev->lru_lock); + ret = ttm_bo_wait_unreserved(bo, interruptible); + spin_lock(&bdev->lru_lock); + + if (unlikely(ret)) + return ret; + } + + if (use_sequence) { + bo->val_seq = sequence; + bo->seq_valid = true; + } else { + bo->seq_valid = false; + } + + return 0; +} +EXPORT_SYMBOL(ttm_bo_reserve); + +static void ttm_bo_ref_bug(struct kref *list_kref) +{ + BUG(); +} + +int ttm_bo_reserve(struct ttm_buffer_object *bo, + bool interruptible, + bool no_wait, bool use_sequence, uint32_t sequence) +{ + struct ttm_bo_device *bdev = bo->bdev; + int put_count = 0; + int ret; + + spin_lock(&bdev->lru_lock); + ret = ttm_bo_reserve_locked(bo, interruptible, no_wait, use_sequence, + sequence); + if (likely(ret == 0)) + put_count = ttm_bo_del_from_lru(bo); + spin_unlock(&bdev->lru_lock); + + while (put_count--) + kref_put(&bo->list_kref, ttm_bo_ref_bug); + + return ret; +} + +void ttm_bo_unreserve(struct ttm_buffer_object *bo) +{ + struct ttm_bo_device *bdev = bo->bdev; + + spin_lock(&bdev->lru_lock); + ttm_bo_add_to_lru(bo); + atomic_set(&bo->reserved, 0); + wake_up_all(&bo->event_queue); + spin_unlock(&bdev->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; + int ret = 0; + uint32_t page_flags = 0; + + TTM_ASSERT_LOCKED(&bo->mutex); + bo->ttm = NULL; + + 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 = ttm_tt_create(bdev, bo->num_pages << PAGE_SHIFT, + page_flags, bdev->dummy_read_page); + if (unlikely(bo->ttm == NULL)) + ret = -ENOMEM; + break; + case ttm_bo_type_user: + bo->ttm = ttm_tt_create(bdev, bo->num_pages << PAGE_SHIFT, + page_flags | TTM_PAGE_FLAG_USER, + bdev->dummy_read_page); + if (unlikely(bo->ttm == NULL)) + ret = -ENOMEM; + break; + + ret = ttm_tt_set_user(bo->ttm, current, + bo->buffer_start, bo->num_pages); + if (unlikely(ret != 0)) + ttm_tt_destroy(bo->ttm); + break; + default: + printk(KERN_ERR TTM_PFX "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) +{ + 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)) + ttm_bo_unmap_virtual(bo); + + /* + * Create and bind a ttm if required. + */ + + if (!(new_man->flags & TTM_MEMTYPE_FLAG_FIXED) && (bo->ttm == NULL)) { + ret = ttm_bo_add_ttm(bo, false); + if (ret) + goto out_err; + + ret = ttm_tt_set_placement_caching(bo->ttm, mem->placement); + if (ret) + return ret; + + 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) { + + struct ttm_mem_reg *old_mem = &bo->mem; + uint32_t save_flags = old_mem->placement; + + *old_mem = *mem; + mem->mm_node = NULL; + ttm_flag_masked(&save_flags, mem->placement, + TTM_PL_MASK_MEMTYPE); + goto moved; + } + + } + + if (!(old_man->flags & TTM_MEMTYPE_FLAG_FIXED) && + !(new_man->flags & TTM_MEMTYPE_FLAG_FIXED)) + ret = ttm_bo_move_ttm(bo, evict, no_wait, mem); + else if (bdev->driver->move) + ret = bdev->driver->move(bo, evict, interruptible, + no_wait, mem); + else + ret = ttm_bo_move_memcpy(bo, evict, no_wait, mem); + + if (ret) + goto out_err; + +moved: + if (bo->evicted) { + ret = bdev->driver->invalidate_caches(bdev, bo->mem.placement); + if (ret) + printk(KERN_ERR TTM_PFX "Can not flush read caches\n"); + bo->evicted = false; + } + + if (bo->mem.mm_node) { + spin_lock(&bo->lock); + bo->offset = (bo->mem.mm_node->start << PAGE_SHIFT) + + bdev->man[bo->mem.mem_type].gpu_offset; + bo->cur_placement = bo->mem.placement; + spin_unlock(&bo->lock); + } + + 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; +} + +/** + * If bo idle, remove from delayed- and lru lists, and unref. + * If not idle, and already on delayed list, do nothing. + * If not idle, and not on delayed list, put on delayed list, + * up the list_kref and schedule a delayed list check. + */ + +static int ttm_bo_cleanup_refs(struct ttm_buffer_object *bo, bool remove_all) +{ + struct ttm_bo_device *bdev = bo->bdev; + struct ttm_bo_driver *driver = bdev->driver; + int ret; + + spin_lock(&bo->lock); + (void) ttm_bo_wait(bo, false, false, !remove_all); + + if (!bo->sync_obj) { + int put_count; + + spin_unlock(&bo->lock); + + spin_lock(&bdev->lru_lock); + ret = ttm_bo_reserve_locked(bo, false, false, false, 0); + BUG_ON(ret); + if (bo->ttm) + ttm_tt_unbind(bo->ttm); + + if (!list_empty(&bo->ddestroy)) { + list_del_init(&bo->ddestroy); + kref_put(&bo->list_kref, ttm_bo_ref_bug); + } + if (bo->mem.mm_node) { + drm_mm_put_block(bo->mem.mm_node); + bo->mem.mm_node = NULL; + } + put_count = ttm_bo_del_from_lru(bo); + spin_unlock(&bdev->lru_lock); + + atomic_set(&bo->reserved, 0); + + while (put_count--) + kref_put(&bo->list_kref, ttm_bo_release_list); + + return 0; + } + + spin_lock(&bdev->lru_lock); + if (list_empty(&bo->ddestroy)) { + void *sync_obj = bo->sync_obj; + void *sync_obj_arg = bo->sync_obj_arg; + + kref_get(&bo->list_kref); + list_add_tail(&bo->ddestroy, &bdev->ddestroy); + spin_unlock(&bdev->lru_lock); + spin_unlock(&bo->lock); + + if (sync_obj) + driver->sync_obj_flush(sync_obj, sync_obj_arg); + schedule_delayed_work(&bdev->wq, + ((HZ / 100) < 1) ? 1 : HZ / 100); + ret = 0; + + } else { + spin_unlock(&bdev->lru_lock); + spin_unlock(&bo->lock); + ret = -EBUSY; + } + + return ret; +} + +/** + * 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_buffer_object *entry, *nentry; + struct list_head *list, *next; + int ret; + + spin_lock(&bdev->lru_lock); + list_for_each_safe(list, next, &bdev->ddestroy) { + entry = list_entry(list, struct ttm_buffer_object, ddestroy); + nentry = NULL; + + /* + * Protect the next list entry from destruction while we + * unlock the lru_lock. + */ + + if (next != &bdev->ddestroy) { + nentry = list_entry(next, struct ttm_buffer_object, + ddestroy); + kref_get(&nentry->list_kref); + } + kref_get(&entry->list_kref); + + spin_unlock(&bdev->lru_lock); + ret = ttm_bo_cleanup_refs(entry, remove_all); + kref_put(&entry->list_kref, ttm_bo_release_list); + + spin_lock(&bdev->lru_lock); + if (nentry) { + bool next_onlist = !list_empty(next); + spin_unlock(&bdev->lru_lock); + kref_put(&nentry->list_kref, ttm_bo_release_list); + spin_lock(&bdev->lru_lock); + /* + * Someone might have raced us and removed the + * next entry from the list. We don't bother restarting + * list traversal. + */ + + if (!next_onlist) + break; + } + if (ret) + break; + } + ret = !list_empty(&bdev->ddestroy); + spin_unlock(&bdev->lru_lock); + + 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; + + 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_bo_cleanup_refs(bo, false); + kref_put(&bo->list_kref, ttm_bo_release_list); + write_lock(&bdev->vm_lock); +} + +void ttm_bo_unref(struct ttm_buffer_object **p_bo) +{ + struct ttm_buffer_object *bo = *p_bo; + struct ttm_bo_device *bdev = bo->bdev; + + *p_bo = NULL; + write_lock(&bdev->vm_lock); + kref_put(&bo->kref, ttm_bo_release); + write_unlock(&bdev->vm_lock); +} +EXPORT_SYMBOL(ttm_bo_unref); + +static int ttm_bo_evict(struct ttm_buffer_object *bo, unsigned mem_type, + bool interruptible, bool no_wait) +{ + int ret = 0; + struct ttm_bo_device *bdev = bo->bdev; + struct ttm_mem_reg evict_mem; + uint32_t proposed_placement; + + if (bo->mem.mem_type != mem_type) + goto out; + + spin_lock(&bo->lock); + ret = ttm_bo_wait(bo, false, interruptible, no_wait); + spin_unlock(&bo->lock); + + if (ret && ret != -ERESTART) { + printk(KERN_ERR TTM_PFX "Failed to expire sync object before " + "buffer eviction.\n"); + goto out; + } + + BUG_ON(!atomic_read(&bo->reserved)); + + evict_mem = bo->mem; + evict_mem.mm_node = NULL; + + proposed_placement = bdev->driver->evict_flags(bo); + + ret = ttm_bo_mem_space(bo, proposed_placement, + &evict_mem, interruptible, no_wait); + if (unlikely(ret != 0 && ret != -ERESTART)) + ret = ttm_bo_mem_space(bo, TTM_PL_FLAG_SYSTEM, + &evict_mem, interruptible, no_wait); + + if (ret) { + if (ret != -ERESTART) + printk(KERN_ERR TTM_PFX + "Failed to find memory space for " + "buffer 0x%p eviction.\n", bo); + goto out; + } + + ret = ttm_bo_handle_move_mem(bo, &evict_mem, true, interruptible, + no_wait); + if (ret) { + if (ret != -ERESTART) + printk(KERN_ERR TTM_PFX "Buffer eviction failed\n"); + goto out; + } + + spin_lock(&bdev->lru_lock); + if (evict_mem.mm_node) { + drm_mm_put_block(evict_mem.mm_node); + evict_mem.mm_node = NULL; + } + spin_unlock(&bdev->lru_lock); + bo->evicted = true; +out: + return ret; +} + +/** + * 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_bo_device *bdev, + struct ttm_mem_reg *mem, + uint32_t mem_type, + bool interruptible, bool no_wait) +{ + struct drm_mm_node *node; + struct ttm_buffer_object *entry; + struct ttm_mem_type_manager *man = &bdev->man[mem_type]; + struct list_head *lru; + unsigned long num_pages = mem->num_pages; + int put_count = 0; + int ret; + +retry_pre_get: + ret = drm_mm_pre_get(&man->manager); + if (unlikely(ret != 0)) + return ret; + + spin_lock(&bdev->lru_lock); + do { + node = drm_mm_search_free(&man->manager, num_pages, + mem->page_alignment, 1); + if (node) + break; + + lru = &man->lru; + if (list_empty(lru)) + break; + + entry = list_first_entry(lru, struct ttm_buffer_object, lru); + kref_get(&entry->list_kref); + + ret = + ttm_bo_reserve_locked(entry, interruptible, no_wait, + false, 0); + + if (likely(ret == 0)) + put_count = ttm_bo_del_from_lru(entry); + + spin_unlock(&bdev->lru_lock); + + if (unlikely(ret != 0)) + return ret; + + while (put_count--) + kref_put(&entry->list_kref, ttm_bo_ref_bug); + + ret = ttm_bo_evict(entry, mem_type, interruptible, no_wait); + + ttm_bo_unreserve(entry); + + kref_put(&entry->list_kref, ttm_bo_release_list); + if (ret) + return ret; + + spin_lock(&bdev->lru_lock); + } while (1); + + if (!node) { + spin_unlock(&bdev->lru_lock); + return -ENOMEM; + } + + node = drm_mm_get_block_atomic(node, num_pages, mem->page_alignment); + if (unlikely(!node)) { + spin_unlock(&bdev->lru_lock); + goto retry_pre_get; + } + + spin_unlock(&bdev->lru_lock); + mem->mm_node = node; + mem->mem_type = mem_type; + return 0; +} + +static bool ttm_bo_mt_compatible(struct ttm_mem_type_manager *man, + bool disallow_fixed, + uint32_t mem_type, + uint32_t mask, uint32_t *res_mask) +{ + uint32_t cur_flags = ttm_bo_type_flags(mem_type); + + if ((man->flags & TTM_MEMTYPE_FLAG_FIXED) && disallow_fixed) + return false; + + if ((cur_flags & mask & TTM_PL_MASK_MEM) == 0) + return false; + + if ((mask & man->available_caching) == 0) + return false; + if (mask & man->default_caching) + cur_flags |= man->default_caching; + else if (mask & TTM_PL_FLAG_CACHED) + cur_flags |= TTM_PL_FLAG_CACHED; + else if (mask & TTM_PL_FLAG_WC) + cur_flags |= TTM_PL_FLAG_WC; + else + cur_flags |= TTM_PL_FLAG_UNCACHED; + + *res_mask = 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, + uint32_t proposed_placement, + struct ttm_mem_reg *mem, + bool interruptible, bool no_wait) +{ + struct ttm_bo_device *bdev = bo->bdev; + struct ttm_mem_type_manager *man; + + uint32_t num_prios = bdev->driver->num_mem_type_prio; + const uint32_t *prios = bdev->driver->mem_type_prio; + uint32_t i; + uint32_t mem_type = TTM_PL_SYSTEM; + uint32_t cur_flags = 0; + bool type_found = false; + bool type_ok = false; + bool has_eagain = false; + struct drm_mm_node *node = NULL; + int ret; + + mem->mm_node = NULL; + for (i = 0; i < num_prios; ++i) { + mem_type = prios[i]; + man = &bdev->man[mem_type]; + + type_ok = ttm_bo_mt_compatible(man, + bo->type == ttm_bo_type_user, + mem_type, proposed_placement, + &cur_flags); + + if (!type_ok) + continue; + + if (mem_type == TTM_PL_SYSTEM) + break; + + if (man->has_type && man->use_type) { + type_found = true; + do { + ret = drm_mm_pre_get(&man->manager); + if (unlikely(ret)) + return ret; + + spin_lock(&bdev->lru_lock); + node = drm_mm_search_free(&man->manager, + mem->num_pages, + mem->page_alignment, + 1); + if (unlikely(!node)) { + spin_unlock(&bdev->lru_lock); + break; + } + node = drm_mm_get_block_atomic(node, + mem->num_pages, + mem-> + page_alignment); + spin_unlock(&bdev->lru_lock); + } while (!node); + } + if (node) + break; + } + + if ((type_ok && (mem_type == TTM_PL_SYSTEM)) || node) { + mem->mm_node = node; + mem->mem_type = mem_type; + mem->placement = cur_flags; + return 0; + } + + if (!type_found) + return -EINVAL; + + num_prios = bdev->driver->num_mem_busy_prio; + prios = bdev->driver->mem_busy_prio; + + for (i = 0; i < num_prios; ++i) { + mem_type = prios[i]; + man = &bdev->man[mem_type]; + + if (!man->has_type) + continue; + + if (!ttm_bo_mt_compatible(man, + bo->type == ttm_bo_type_user, + mem_type, + proposed_placement, &cur_flags)) + continue; + + ret = ttm_bo_mem_force_space(bdev, mem, mem_type, + interruptible, no_wait); + + if (ret == 0 && mem->mm_node) { + mem->placement = cur_flags; + return 0; + } + + if (ret == -ERESTART) + has_eagain = true; + } + + ret = (has_eagain) ? -ERESTART : -ENOMEM; + return ret; +} +EXPORT_SYMBOL(ttm_bo_mem_space); + +int ttm_bo_wait_cpu(struct ttm_buffer_object *bo, bool no_wait) +{ + int ret = 0; + + if ((atomic_read(&bo->cpu_writers) > 0) && no_wait) + return -EBUSY; + + ret = wait_event_interruptible(bo->event_queue, + atomic_read(&bo->cpu_writers) == 0); + + if (ret == -ERESTARTSYS) + ret = -ERESTART; + + return ret; +} + +int ttm_bo_move_buffer(struct ttm_buffer_object *bo, + uint32_t proposed_placement, + bool interruptible, bool no_wait) +{ + struct ttm_bo_device *bdev = bo->bdev; + int ret = 0; + struct ttm_mem_reg mem; + + BUG_ON(!atomic_read(&bo->reserved)); + + /* + * 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(&bo->lock); + ret = ttm_bo_wait(bo, false, interruptible, no_wait); + spin_unlock(&bo->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; + + /* + * Determine where to move the buffer. + */ + + ret = ttm_bo_mem_space(bo, proposed_placement, &mem, + interruptible, no_wait); + if (ret) + goto out_unlock; + + ret = ttm_bo_handle_move_mem(bo, &mem, false, interruptible, no_wait); + +out_unlock: + if (ret && mem.mm_node) { + spin_lock(&bdev->lru_lock); + drm_mm_put_block(mem.mm_node); + spin_unlock(&bdev->lru_lock); + } + return ret; +} + +static int ttm_bo_mem_compat(uint32_t proposed_placement, + struct ttm_mem_reg *mem) +{ + if ((proposed_placement & mem->placement & TTM_PL_MASK_MEM) == 0) + return 0; + if ((proposed_placement & mem->placement & TTM_PL_MASK_CACHING) == 0) + return 0; + + return 1; +} + +int ttm_buffer_object_validate(struct ttm_buffer_object *bo, + uint32_t proposed_placement, + bool interruptible, bool no_wait) +{ + int ret; + + BUG_ON(!atomic_read(&bo->reserved)); + bo->proposed_placement = proposed_placement; + + TTM_DEBUG("Proposed placement 0x%08lx, Old flags 0x%08lx\n", + (unsigned long)proposed_placement, + (unsigned long)bo->mem.placement); + + /* + * Check whether we need to move buffer. + */ + + if (!ttm_bo_mem_compat(bo->proposed_placement, &bo->mem)) { + ret = ttm_bo_move_buffer(bo, bo->proposed_placement, + interruptible, no_wait); + if (ret) { + if (ret != -ERESTART) + printk(KERN_ERR TTM_PFX + "Failed moving buffer. " + "Proposed placement 0x%08x\n", + bo->proposed_placement); + if (ret == -ENOMEM) + printk(KERN_ERR TTM_PFX + "Out of aperture space or " + "DRM memory quota.\n"); + return ret; + } + } + + /* + * 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; + } + /* + * Validation has succeeded, move the access and other + * non-mapping-related flag bits from the proposed flags to + * the active flags + */ + + ttm_flag_masked(&bo->mem.placement, bo->proposed_placement, + ~TTM_PL_MASK_MEMTYPE); + + return 0; +} +EXPORT_SYMBOL(ttm_buffer_object_validate); + +int +ttm_bo_check_placement(struct ttm_buffer_object *bo, + uint32_t set_flags, uint32_t clr_flags) +{ + uint32_t new_mask = set_flags | clr_flags; + + if ((bo->type == ttm_bo_type_user) && + (clr_flags & TTM_PL_FLAG_CACHED)) { + printk(KERN_ERR TTM_PFX + "User buffers require cache-coherent memory.\n"); + return -EINVAL; + } + + if (!capable(CAP_SYS_ADMIN)) { + if (new_mask & TTM_PL_FLAG_NO_EVICT) { + printk(KERN_ERR TTM_PFX "Need to be root to modify" + " NO_EVICT status.\n"); + return -EINVAL; + } + + if ((clr_flags & bo->mem.placement & TTM_PL_MASK_MEMTYPE) && + (bo->mem.placement & TTM_PL_FLAG_NO_EVICT)) { + printk(KERN_ERR TTM_PFX + "Incompatible memory specification" + " for NO_EVICT buffer.\n"); + return -EINVAL; + } + } + return 0; +} + +int ttm_buffer_object_init(struct ttm_bo_device *bdev, + struct ttm_buffer_object *bo, + unsigned long size, + enum ttm_bo_type type, + uint32_t flags, + uint32_t page_alignment, + unsigned long buffer_start, + bool interruptible, + struct file *persistant_swap_storage, + size_t acc_size, + void (*destroy) (struct ttm_buffer_object *)) +{ + int ret = 0; + unsigned long num_pages; + + size += buffer_start & ~PAGE_MASK; + num_pages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; + if (num_pages == 0) { + printk(KERN_ERR TTM_PFX "Illegal buffer object size.\n"); + return -EINVAL; + } + bo->destroy = destroy; + + spin_lock_init(&bo->lock); + 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); + bo->bdev = bdev; + bo->type = type; + bo->num_pages = num_pages; + 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->buffer_start = buffer_start & PAGE_MASK; + bo->priv_flags = 0; + bo->mem.placement = (TTM_PL_FLAG_SYSTEM | TTM_PL_FLAG_CACHED); + bo->seq_valid = false; + bo->persistant_swap_storage = persistant_swap_storage; + bo->acc_size = acc_size; + + ret = ttm_bo_check_placement(bo, flags, 0ULL); + if (unlikely(ret != 0)) + goto out_err; + + /* + * If no caching attributes are set, accept any form of caching. + */ + + if ((flags & TTM_PL_MASK_CACHING) == 0) + flags |= TTM_PL_MASK_CACHING; + + /* + * For ttm_bo_type_device buffers, allocate + * address space from the device. + */ + + if (bo->type == ttm_bo_type_device) { + ret = ttm_bo_setup_vm(bo); + if (ret) + goto out_err; + } + + ret = ttm_buffer_object_validate(bo, flags, 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_buffer_object_init); + +static inline size_t ttm_bo_size(struct ttm_bo_device *bdev, + unsigned long num_pages) +{ + size_t page_array_size = (num_pages * sizeof(void *) + PAGE_SIZE - 1) & + PAGE_MASK; + + return bdev->ttm_bo_size + 2 * page_array_size; +} + +int ttm_buffer_object_create(struct ttm_bo_device *bdev, + unsigned long size, + enum ttm_bo_type type, + uint32_t flags, + uint32_t page_alignment, + unsigned long buffer_start, + bool interruptible, + struct file *persistant_swap_storage, + struct ttm_buffer_object **p_bo) +{ + struct ttm_buffer_object *bo; + int ret; + struct ttm_mem_global *mem_glob = bdev->mem_glob; + + size_t acc_size = + ttm_bo_size(bdev, (size + PAGE_SIZE - 1) >> PAGE_SHIFT); + ret = ttm_mem_global_alloc(mem_glob, acc_size, false, false, false); + if (unlikely(ret != 0)) + return ret; + + bo = kzalloc(sizeof(*bo), GFP_KERNEL); + + if (unlikely(bo == NULL)) { + ttm_mem_global_free(mem_glob, acc_size, false); + return -ENOMEM; + } + + ret = ttm_buffer_object_init(bdev, bo, size, type, flags, + page_alignment, buffer_start, + interruptible, + persistant_swap_storage, acc_size, NULL); + if (likely(ret == 0)) + *p_bo = bo; + + return ret; +} + +static int ttm_bo_leave_list(struct ttm_buffer_object *bo, + uint32_t mem_type, bool allow_errors) +{ + int ret; + + spin_lock(&bo->lock); + ret = ttm_bo_wait(bo, false, false, false); + spin_unlock(&bo->lock); + + if (ret && allow_errors) + goto out; + + if (bo->mem.mem_type == mem_type) + ret = ttm_bo_evict(bo, mem_type, false, false); + + if (ret) { + if (allow_errors) { + goto out; + } else { + ret = 0; + printk(KERN_ERR TTM_PFX "Cleanup eviction failed\n"); + } + } + +out: + return ret; +} + +static int ttm_bo_force_list_clean(struct ttm_bo_device *bdev, + struct list_head *head, + unsigned mem_type, bool allow_errors) +{ + struct ttm_buffer_object *entry; + int ret; + int put_count; + + /* + * Can't use standard list traversal since we're unlocking. + */ + + spin_lock(&bdev->lru_lock); + + while (!list_empty(head)) { + entry = list_first_entry(head, struct ttm_buffer_object, lru); + kref_get(&entry->list_kref); + ret = ttm_bo_reserve_locked(entry, false, false, false, 0); + put_count = ttm_bo_del_from_lru(entry); + spin_unlock(&bdev->lru_lock); + while (put_count--) + kref_put(&entry->list_kref, ttm_bo_ref_bug); + BUG_ON(ret); + ret = ttm_bo_leave_list(entry, mem_type, allow_errors); + ttm_bo_unreserve(entry); + kref_put(&entry->list_kref, ttm_bo_release_list); + spin_lock(&bdev->lru_lock); + } + + spin_unlock(&bdev->lru_lock); + + return 0; +} + +int ttm_bo_clean_mm(struct ttm_bo_device *bdev, unsigned mem_type) +{ + struct ttm_mem_type_manager *man = &bdev->man[mem_type]; + int ret = -EINVAL; + + if (mem_type >= TTM_NUM_MEM_TYPES) { + printk(KERN_ERR TTM_PFX "Illegal memory type %d\n", mem_type); + return ret; + } + + if (!man->has_type) { + printk(KERN_ERR TTM_PFX "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, &man->lru, mem_type, false); + + spin_lock(&bdev->lru_lock); + if (drm_mm_clean(&man->manager)) + drm_mm_takedown(&man->manager); + else + ret = -EBUSY; + + spin_unlock(&bdev->lru_lock); + } + + 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) { + printk(KERN_ERR TTM_PFX + "Illegal memory manager memory type %u.\n", + mem_type); + return -EINVAL; + } + + if (!man->has_type) { + printk(KERN_ERR TTM_PFX + "Memory type %u has not been initialized.\n", + mem_type); + return 0; + } + + return ttm_bo_force_list_clean(bdev, &man->lru, mem_type, true); +} +EXPORT_SYMBOL(ttm_bo_evict_mm); + +int ttm_bo_init_mm(struct ttm_bo_device *bdev, unsigned type, + unsigned long p_offset, unsigned long p_size) +{ + int ret = -EINVAL; + struct ttm_mem_type_manager *man; + + if (type >= TTM_NUM_MEM_TYPES) { + printk(KERN_ERR TTM_PFX "Illegal memory type %d\n", type); + return ret; + } + + man = &bdev->man[type]; + if (man->has_type) { + printk(KERN_ERR TTM_PFX + "Memory manager already initialized for type %d\n", + type); + return ret; + } + + ret = bdev->driver->init_mem_type(bdev, type, man); + if (ret) + return ret; + + ret = 0; + if (type != TTM_PL_SYSTEM) { + if (!p_size) { + printk(KERN_ERR TTM_PFX + "Zero size memory manager type %d\n", + type); + return ret; + } + ret = drm_mm_init(&man->manager, p_offset, 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); + +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; + + 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; + printk(KERN_ERR TTM_PFX + "DRM memory manager type %d " + "is not clean.\n", i); + } + man->has_type = false; + } + } + + if (!cancel_delayed_work(&bdev->wq)) + flush_scheduled_work(); + + while (ttm_bo_delayed_delete(bdev, true)) + ; + + spin_lock(&bdev->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(&bdev->lru_lock); + + ttm_mem_unregister_shrink(bdev->mem_glob, &bdev->shrink); + 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); + + __free_page(bdev->dummy_read_page); + return ret; +} +EXPORT_SYMBOL(ttm_bo_device_release); + +/* + * This function is intended to be called on drm driver load. + * If you decide to call it from firstopen, you must protect the call + * from a potentially racing ttm_bo_driver_finish in lastclose. + * (This may happen on X server restart). + */ + +int ttm_bo_device_init(struct ttm_bo_device *bdev, + struct ttm_mem_global *mem_glob, + struct ttm_bo_driver *driver, uint64_t file_page_offset) +{ + int ret = -EINVAL; + + bdev->dummy_read_page = NULL; + rwlock_init(&bdev->vm_lock); + spin_lock_init(&bdev->lru_lock); + + bdev->driver = driver; + bdev->mem_glob = mem_glob; + + memset(bdev->man, 0, sizeof(bdev->man)); + + bdev->dummy_read_page = alloc_page(__GFP_ZERO | GFP_DMA32); + if (unlikely(bdev->dummy_read_page == NULL)) { + ret = -ENOMEM; + goto out_err0; + } + + /* + * Initialize the system memory buffer type. + * Other types need to be driver / IOCTL initialized. + */ + ret = ttm_bo_init_mm(bdev, TTM_PL_SYSTEM, 0, 0); + if (unlikely(ret != 0)) + goto out_err1; + + bdev->addr_space_rb = RB_ROOT; + ret = drm_mm_init(&bdev->addr_space_mm, file_page_offset, 0x10000000); + if (unlikely(ret != 0)) + goto out_err2; + + INIT_DELAYED_WORK(&bdev->wq, ttm_bo_delayed_workqueue); + bdev->nice_mode = true; + INIT_LIST_HEAD(&bdev->ddestroy); + INIT_LIST_HEAD(&bdev->swap_lru); + bdev->dev_mapping = NULL; + ttm_mem_init_shrink(&bdev->shrink, ttm_bo_swapout); + ret = ttm_mem_register_shrink(mem_glob, &bdev->shrink); + if (unlikely(ret != 0)) { + printk(KERN_ERR TTM_PFX + "Could not register buffer object swapout.\n"); + goto out_err2; + } + + bdev->ttm_bo_extra_size = + ttm_round_pot(sizeof(struct ttm_tt)) + + ttm_round_pot(sizeof(struct ttm_backend)); + + bdev->ttm_bo_size = bdev->ttm_bo_extra_size + + ttm_round_pot(sizeof(struct ttm_buffer_object)); + + return 0; +out_err2: + ttm_bo_clean_mm(bdev, 0); +out_err1: + __free_page(bdev->dummy_read_page); +out_err0: + 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; +} + +int ttm_bo_pci_offset(struct ttm_bo_device *bdev, + struct ttm_mem_reg *mem, + unsigned long *bus_base, + unsigned long *bus_offset, unsigned long *bus_size) +{ + struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type]; + + *bus_size = 0; + if (!(man->flags & TTM_MEMTYPE_FLAG_MAPPABLE)) + return -EINVAL; + + if (ttm_mem_reg_is_pci(bdev, mem)) { + *bus_offset = mem->mm_node->start << PAGE_SHIFT; + *bus_size = mem->num_pages << PAGE_SHIFT; + *bus_base = man->io_offset; + } + + return 0; +} + +void ttm_bo_unmap_virtual(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); +} + +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; + void *sync_obj; + void *sync_obj_arg; + int ret = 0; + + if (likely(bo->sync_obj == NULL)) + return 0; + + while (bo->sync_obj) { + + if (driver->sync_obj_signaled(bo->sync_obj, bo->sync_obj_arg)) { + void *tmp_obj = bo->sync_obj; + bo->sync_obj = NULL; + clear_bit(TTM_BO_PRIV_FLAG_MOVING, &bo->priv_flags); + spin_unlock(&bo->lock); + driver->sync_obj_unref(&tmp_obj); + spin_lock(&bo->lock); + continue; + } + + if (no_wait) + return -EBUSY; + + sync_obj = driver->sync_obj_ref(bo->sync_obj); + sync_obj_arg = bo->sync_obj_arg; + spin_unlock(&bo->lock); + ret = driver->sync_obj_wait(sync_obj, sync_obj_arg, + lazy, interruptible); + if (unlikely(ret != 0)) { + driver->sync_obj_unref(&sync_obj); + spin_lock(&bo->lock); + return ret; + } + spin_lock(&bo->lock); + if (likely(bo->sync_obj == sync_obj && + bo->sync_obj_arg == sync_obj_arg)) { + void *tmp_obj = bo->sync_obj; + bo->sync_obj = NULL; + clear_bit(TTM_BO_PRIV_FLAG_MOVING, + &bo->priv_flags); + spin_unlock(&bo->lock); + driver->sync_obj_unref(&sync_obj); + driver->sync_obj_unref(&tmp_obj); + spin_lock(&bo->lock); + } + } + return 0; +} +EXPORT_SYMBOL(ttm_bo_wait); + +void ttm_bo_unblock_reservation(struct ttm_buffer_object *bo) +{ + atomic_set(&bo->reserved, 0); + wake_up_all(&bo->event_queue); +} + +int ttm_bo_block_reservation(struct ttm_buffer_object *bo, bool interruptible, + bool no_wait) +{ + int ret; + + while (unlikely(atomic_cmpxchg(&bo->reserved, 0, 1) != 0)) { + if (no_wait) + return -EBUSY; + else if (interruptible) { + ret = wait_event_interruptible + (bo->event_queue, atomic_read(&bo->reserved) == 0); + if (unlikely(ret != 0)) + return -ERESTART; + } else { + wait_event(bo->event_queue, + atomic_read(&bo->reserved) == 0); + } + } + return 0; +} + +int ttm_bo_synccpu_write_grab(struct ttm_buffer_object *bo, bool no_wait) +{ + int ret = 0; + + /* + * Using ttm_bo_reserve instead of ttm_bo_block_reservation + * 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(&bo->lock); + ret = ttm_bo_wait(bo, false, true, no_wait); + spin_unlock(&bo->lock); + if (likely(ret == 0)) + atomic_inc(&bo->cpu_writers); + ttm_bo_unreserve(bo); + return ret; +} + +void ttm_bo_synccpu_write_release(struct ttm_buffer_object *bo) +{ + if (atomic_dec_and_test(&bo->cpu_writers)) + wake_up_all(&bo->event_queue); +} + +/** + * 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_device *bdev = + container_of(shrink, struct ttm_bo_device, 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(&bdev->lru_lock); + while (ret == -EBUSY) { + if (unlikely(list_empty(&bdev->swap_lru))) { + spin_unlock(&bdev->lru_lock); + return -EBUSY; + } + + bo = list_first_entry(&bdev->swap_lru, + struct ttm_buffer_object, swap); + kref_get(&bo->list_kref); + + /** + * Reserve buffer. Since we unlock while sleeping, we need + * to re-check that nobody removed us from the swap-list while + * we slept. + */ + + ret = ttm_bo_reserve_locked(bo, false, true, false, 0); + if (unlikely(ret == -EBUSY)) { + spin_unlock(&bdev->lru_lock); + ttm_bo_wait_unreserved(bo, false); + kref_put(&bo->list_kref, ttm_bo_release_list); + spin_lock(&bdev->lru_lock); + } + } + + BUG_ON(ret != 0); + put_count = ttm_bo_del_from_lru(bo); + spin_unlock(&bdev->lru_lock); + + while (put_count--) + kref_put(&bo->list_kref, ttm_bo_ref_bug); + + /** + * Wait for GPU, then move to system cached. + */ + + spin_lock(&bo->lock); + ret = ttm_bo_wait(bo, false, false, false); + spin_unlock(&bo->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. + */ + + ret = ttm_tt_swapout(bo->ttm, bo->persistant_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->shrink) == 0) + ; +} 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 0000000000000000000000000000000000000000..517c8455963312e16c6a515dfc8fed21de16fec3 --- /dev/null +++ b/drivers/gpu/drm/ttm/ttm_bo_util.c @@ -0,0 +1,561 @@ +/************************************************************************** + * + * 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 + */ + +#include "ttm/ttm_bo_driver.h" +#include "ttm/ttm_placement.h" +#include +#include +#include +#include +#include +#include + +void ttm_bo_free_old_node(struct ttm_buffer_object *bo) +{ + struct ttm_mem_reg *old_mem = &bo->mem; + + if (old_mem->mm_node) { + spin_lock(&bo->bdev->lru_lock); + drm_mm_put_block(old_mem->mm_node); + spin_unlock(&bo->bdev->lru_lock); + } + old_mem->mm_node = NULL; +} + +int ttm_bo_move_ttm(struct ttm_buffer_object *bo, + bool evict, bool no_wait, struct ttm_mem_reg *new_mem) +{ + struct ttm_tt *ttm = bo->ttm; + struct ttm_mem_reg *old_mem = &bo->mem; + uint32_t save_flags = old_mem->placement; + 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; + save_flags = old_mem->placement; + } + + 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; + ttm_flag_masked(&save_flags, new_mem->placement, TTM_PL_MASK_MEMTYPE); + return 0; +} +EXPORT_SYMBOL(ttm_bo_move_ttm); + +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]; + unsigned long bus_offset; + unsigned long bus_size; + unsigned long bus_base; + int ret; + void *addr; + + *virtual = NULL; + ret = ttm_bo_pci_offset(bdev, mem, &bus_base, &bus_offset, &bus_size); + if (ret || bus_size == 0) + return ret; + + if (!(man->flags & TTM_MEMTYPE_FLAG_NEEDS_IOREMAP)) + addr = (void *)(((u8 *) man->io_addr) + bus_offset); + else { + if (mem->placement & TTM_PL_FLAG_WC) + addr = ioremap_wc(bus_base + bus_offset, bus_size); + else + addr = ioremap_nocache(bus_base + bus_offset, bus_size); + if (!addr) + 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 && (man->flags & TTM_MEMTYPE_FLAG_NEEDS_IOREMAP)) + iounmap(virtual); +} + +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) +{ + struct page *d = ttm_tt_get_page(ttm, page); + void *dst; + + if (!d) + return -ENOMEM; + + src = (void *)((unsigned long)src + (page << PAGE_SHIFT)); + dst = kmap(d); + if (!dst) + return -ENOMEM; + + memcpy_fromio(dst, src, PAGE_SIZE); + kunmap(d); + return 0; +} + +static int ttm_copy_ttm_io_page(struct ttm_tt *ttm, void *dst, + unsigned long page) +{ + struct page *s = ttm_tt_get_page(ttm, page); + void *src; + + if (!s) + return -ENOMEM; + + dst = (void *)((unsigned long)dst + (page << PAGE_SHIFT)); + src = kmap(s); + if (!src) + return -ENOMEM; + + memcpy_toio(dst, src, PAGE_SIZE); + kunmap(s); + return 0; +} + +int ttm_bo_move_memcpy(struct ttm_buffer_object *bo, + bool evict, bool no_wait, 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; + uint32_t save_flags = old_mem->placement; + 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; + + add = 0; + dir = 1; + + if ((old_mem->mem_type == new_mem->mem_type) && + (new_mem->mm_node->start < + old_mem->mm_node->start + old_mem->mm_node->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) + ret = ttm_copy_ttm_io_page(ttm, new_iomap, page); + else if (new_iomap == NULL) + ret = ttm_copy_io_ttm_page(ttm, old_iomap, page); + else + ret = ttm_copy_io_page(new_iomap, old_iomap, page); + if (ret) + goto out1; + } + mb(); +out2: + ttm_bo_free_old_node(bo); + + *old_mem = *new_mem; + new_mem->mm_node = NULL; + ttm_flag_masked(&save_flags, new_mem->placement, TTM_PL_MASK_MEMTYPE); + + 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, new_mem, new_iomap); +out: + ttm_mem_reg_iounmap(bdev, &old_copy, old_iomap); + 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 = kzalloc(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. + */ + + spin_lock_init(&fbo->lock); + init_waitqueue_head(&fbo->event_queue); + INIT_LIST_HEAD(&fbo->ddestroy); + INIT_LIST_HEAD(&fbo->lru); + INIT_LIST_HEAD(&fbo->swap); + fbo->vm_node = NULL; + + fbo->sync_obj = driver->sync_obj_ref(bo->sync_obj); + if (fbo->mem.mm_node) + fbo->mem.mm_node->private = (void *)fbo; + kref_init(&fbo->list_kref); + kref_init(&fbo->kref); + fbo->destroy = &ttm_transfered_destroy; + + *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__) + if (!(caching_flags & TTM_PL_FLAG_CACHED)) + tmp = pgprot_noncached(tmp); +#endif + return tmp; +} + +static int ttm_bo_ioremap(struct ttm_buffer_object *bo, + unsigned long bus_base, + unsigned long bus_offset, + unsigned long bus_size, + struct ttm_bo_kmap_obj *map) +{ + struct ttm_bo_device *bdev = bo->bdev; + struct ttm_mem_reg *mem = &bo->mem; + struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type]; + + if (!(man->flags & TTM_MEMTYPE_FLAG_NEEDS_IOREMAP)) { + map->bo_kmap_type = ttm_bo_map_premapped; + map->virtual = (void *)(((u8 *) man->io_addr) + bus_offset); + } else { + map->bo_kmap_type = ttm_bo_map_iomap; + if (mem->placement & TTM_PL_FLAG_WC) + map->virtual = ioremap_wc(bus_base + bus_offset, + bus_size); + else + map->virtual = ioremap_nocache(bus_base + bus_offset, + bus_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; + struct page *d; + int i; + + BUG_ON(!ttm); + 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_tt_get_page(ttm, start_page); + map->virtual = kmap(map->page); + } else { + /* + * Populate the part we're mapping; + */ + for (i = start_page; i < start_page + num_pages; ++i) { + d = ttm_tt_get_page(ttm, i); + if (!d) + return -ENOMEM; + } + + /* + * We need to use vmap to get the desired page protection + * or to make the buffer object look contigous. + */ + 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) +{ + int ret; + unsigned long bus_base; + unsigned long bus_offset; + unsigned long bus_size; + + BUG_ON(!list_empty(&bo->swap)); + map->virtual = NULL; + 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 + ret = ttm_bo_pci_offset(bo->bdev, &bo->mem, &bus_base, + &bus_offset, &bus_size); + if (ret) + return ret; + if (bus_size == 0) { + return ttm_bo_kmap_ttm(bo, start_page, num_pages, map); + } else { + bus_offset += start_page << PAGE_SHIFT; + bus_size = num_pages << PAGE_SHIFT; + return ttm_bo_ioremap(bo, bus_base, bus_offset, bus_size, map); + } +} +EXPORT_SYMBOL(ttm_bo_kmap); + +void ttm_bo_kunmap(struct ttm_bo_kmap_obj *map) +{ + 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(); + } + map->virtual = NULL; + map->page = NULL; +} +EXPORT_SYMBOL(ttm_bo_kunmap); + +int ttm_bo_pfn_prot(struct ttm_buffer_object *bo, + unsigned long dst_offset, + unsigned long *pfn, pgprot_t *prot) +{ + struct ttm_mem_reg *mem = &bo->mem; + struct ttm_bo_device *bdev = bo->bdev; + unsigned long bus_offset; + unsigned long bus_size; + unsigned long bus_base; + int ret; + ret = ttm_bo_pci_offset(bdev, mem, &bus_base, &bus_offset, + &bus_size); + if (ret) + return -EINVAL; + if (bus_size != 0) + *pfn = (bus_base + bus_offset + dst_offset) >> PAGE_SHIFT; + else + if (!bo->ttm) + return -EINVAL; + else + *pfn = page_to_pfn(ttm_tt_get_page(bo->ttm, + dst_offset >> + PAGE_SHIFT)); + *prot = (mem->placement & TTM_PL_FLAG_CACHED) ? + PAGE_KERNEL : ttm_io_prot(mem->placement, PAGE_KERNEL); + + return 0; +} + +int ttm_bo_move_accel_cleanup(struct ttm_buffer_object *bo, + void *sync_obj, + void *sync_obj_arg, + bool evict, bool no_wait, + 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; + uint32_t save_flags = old_mem->placement; + struct ttm_buffer_object *ghost_obj; + void *tmp_obj = NULL; + + spin_lock(&bo->lock); + if (bo->sync_obj) { + tmp_obj = bo->sync_obj; + bo->sync_obj = NULL; + } + bo->sync_obj = driver->sync_obj_ref(sync_obj); + bo->sync_obj_arg = sync_obj_arg; + if (evict) { + ret = ttm_bo_wait(bo, false, false, false); + spin_unlock(&bo->lock); + driver->sync_obj_unref(&bo->sync_obj); + + if (ret) + return ret; + + ttm_bo_free_old_node(bo); + if ((man->flags & TTM_MEMTYPE_FLAG_FIXED) && + (bo->ttm != NULL)) { + ttm_tt_unbind(bo->ttm); + ttm_tt_destroy(bo->ttm); + bo->ttm = NULL; + } + } 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(&bo->lock); + + 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; + ttm_flag_masked(&save_flags, new_mem->placement, TTM_PL_MASK_MEMTYPE); + 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 0000000000000000000000000000000000000000..27b146c54fbcb4d41c137ea9fb15a4a873ee068f --- /dev/null +++ b/drivers/gpu/drm/ttm/ttm_bo_vm.c @@ -0,0 +1,454 @@ +/************************************************************************** + * + * 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 + */ + +#include +#include +#include +#include +#include +#include +#include +#include + +#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 bus_base; + unsigned long bus_offset; + unsigned long bus_size; + unsigned long page_offset; + unsigned long page_last; + unsigned long pfn; + struct ttm_tt *ttm = NULL; + struct page *page; + int ret; + int i; + bool is_iomem; + unsigned long address = (unsigned long)vmf->virtual_address; + int retval = VM_FAULT_NOPAGE; + + /* + * 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; + } + + /* + * Wait for buffer data in transit, due to a pipelined + * move. + */ + + spin_lock(&bo->lock); + if (test_bit(TTM_BO_PRIV_FLAG_MOVING, &bo->priv_flags)) { + ret = ttm_bo_wait(bo, false, true, false); + spin_unlock(&bo->lock); + if (unlikely(ret != 0)) { + retval = (ret != -ERESTART) ? + VM_FAULT_SIGBUS : VM_FAULT_NOPAGE; + goto out_unlock; + } + } else + spin_unlock(&bo->lock); + + + ret = ttm_bo_pci_offset(bdev, &bo->mem, &bus_base, &bus_offset, + &bus_size); + if (unlikely(ret != 0)) { + retval = VM_FAULT_SIGBUS; + goto out_unlock; + } + + is_iomem = (bus_size != 0); + + page_offset = ((address - vma->vm_start) >> PAGE_SHIFT) + + bo->vm_node->start - vma->vm_pgoff; + page_last = ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) + + bo->vm_node->start - vma->vm_pgoff; + + if (unlikely(page_offset >= bo->num_pages)) { + retval = VM_FAULT_SIGBUS; + goto out_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 (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); + } + + /* + * Speculatively prefault a number of pages. Only error on + * first page. + */ + + for (i = 0; i < TTM_BO_VM_NUM_PREFAULT; ++i) { + + if (is_iomem) + pfn = ((bus_base + bus_offset) >> PAGE_SHIFT) + + page_offset; + else { + page = ttm_tt_get_page(ttm, page_offset); + if (unlikely(!page && i == 0)) { + retval = VM_FAULT_OOM; + goto out_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_unlock; + + } + + address += PAGE_SIZE; + if (unlikely(++page_offset >= page_last)) + break; + } + +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 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->vm_end - vma->vm_start) >> PAGE_SHIFT); + if (likely(bo != NULL)) + ttm_bo_reference(bo); + read_unlock(&bdev->vm_lock); + + if (unlikely(bo == NULL)) { + printk(KERN_ERR TTM_PFX + "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_RESERVED | VM_IO | VM_MIXEDMAP | VM_DONTEXPAND; + 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_RESERVED | 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 -ERESTART: + ret = -EINTR; + goto out_unref; + 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 -ERESTART: + return -EINTR; + 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_global.c b/drivers/gpu/drm/ttm/ttm_global.c new file mode 100644 index 0000000000000000000000000000000000000000..0b14eb1972b81bc5c95d0d53bc9a54d1634bcdec --- /dev/null +++ b/drivers/gpu/drm/ttm/ttm_global.c @@ -0,0 +1,114 @@ +/************************************************************************** + * + * Copyright 2008-2009 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 + */ + +#include "ttm/ttm_module.h" +#include +#include +#include + +struct ttm_global_item { + struct mutex mutex; + void *object; + int refcount; +}; + +static struct ttm_global_item glob[TTM_GLOBAL_NUM]; + +void ttm_global_init(void) +{ + int i; + + for (i = 0; i < TTM_GLOBAL_NUM; ++i) { + struct ttm_global_item *item = &glob[i]; + mutex_init(&item->mutex); + item->object = NULL; + item->refcount = 0; + } +} + +void ttm_global_release(void) +{ + int i; + for (i = 0; i < TTM_GLOBAL_NUM; ++i) { + struct ttm_global_item *item = &glob[i]; + BUG_ON(item->object != NULL); + BUG_ON(item->refcount != 0); + } +} + +int ttm_global_item_ref(struct ttm_global_reference *ref) +{ + int ret; + struct ttm_global_item *item = &glob[ref->global_type]; + void *object; + + mutex_lock(&item->mutex); + if (item->refcount == 0) { + item->object = kmalloc(ref->size, GFP_KERNEL); + if (unlikely(item->object == NULL)) { + ret = -ENOMEM; + goto out_err; + } + + ref->object = item->object; + ret = ref->init(ref); + if (unlikely(ret != 0)) + goto out_err; + + ++item->refcount; + } + ref->object = item->object; + object = item->object; + mutex_unlock(&item->mutex); + return 0; +out_err: + kfree(item->object); + mutex_unlock(&item->mutex); + item->object = NULL; + return ret; +} +EXPORT_SYMBOL(ttm_global_item_ref); + +void ttm_global_item_unref(struct ttm_global_reference *ref) +{ + struct ttm_global_item *item = &glob[ref->global_type]; + + mutex_lock(&item->mutex); + BUG_ON(item->refcount == 0); + BUG_ON(ref->object != item->object); + if (--item->refcount == 0) { + ref->release(ref); + kfree(item->object); + item->object = NULL; + } + mutex_unlock(&item->mutex); +} +EXPORT_SYMBOL(ttm_global_item_unref); + diff --git a/drivers/gpu/drm/ttm/ttm_memory.c b/drivers/gpu/drm/ttm/ttm_memory.c new file mode 100644 index 0000000000000000000000000000000000000000..87323d4ff68d862a6d883321857d49a7eb430cb7 --- /dev/null +++ b/drivers/gpu/drm/ttm/ttm_memory.c @@ -0,0 +1,234 @@ +/************************************************************************** + * + * 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 "ttm/ttm_memory.h" +#include +#include +#include +#include +#include + +#define TTM_PFX "[TTM] " +#define TTM_MEMORY_ALLOC_RETRIES 4 + +/** + * 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_workqueue, + uint64_t extra) +{ + int ret; + struct ttm_mem_shrink *shrink; + uint64_t target; + uint64_t total_target; + + spin_lock(&glob->lock); + if (glob->shrink == NULL) + goto out; + + if (from_workqueue) { + target = glob->swap_limit; + total_target = glob->total_memory_swap_limit; + } else if (capable(CAP_SYS_ADMIN)) { + total_target = glob->emer_total_memory; + target = glob->emer_memory; + } else { + total_target = glob->max_total_memory; + target = glob->max_memory; + } + + total_target = (extra >= total_target) ? 0 : total_target - extra; + target = (extra >= target) ? 0 : target - extra; + + while (glob->used_memory > target || + glob->used_total_memory > total_target) { + 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); +} + +int ttm_mem_global_init(struct ttm_mem_global *glob) +{ + struct sysinfo si; + uint64_t mem; + + spin_lock_init(&glob->lock); + glob->swap_queue = create_singlethread_workqueue("ttm_swap"); + INIT_WORK(&glob->work, ttm_shrink_work); + init_waitqueue_head(&glob->queue); + + si_meminfo(&si); + + mem = si.totalram - si.totalhigh; + mem *= si.mem_unit; + + glob->max_memory = mem >> 1; + glob->emer_memory = (mem >> 1) + (mem >> 2); + glob->swap_limit = glob->max_memory - (mem >> 3); + glob->used_memory = 0; + glob->used_total_memory = 0; + glob->shrink = NULL; + + mem = si.totalram; + mem *= si.mem_unit; + + glob->max_total_memory = mem >> 1; + glob->emer_total_memory = (mem >> 1) + (mem >> 2); + + glob->total_memory_swap_limit = glob->max_total_memory - (mem >> 3); + + printk(KERN_INFO TTM_PFX "TTM available graphics memory: %llu MiB\n", + glob->max_total_memory >> 20); + printk(KERN_INFO TTM_PFX "TTM available object memory: %llu MiB\n", + glob->max_memory >> 20); + + return 0; +} +EXPORT_SYMBOL(ttm_mem_global_init); + +void ttm_mem_global_release(struct ttm_mem_global *glob) +{ + printk(KERN_INFO TTM_PFX "Used total memory is %llu bytes.\n", + (unsigned long long)glob->used_total_memory); + flush_workqueue(glob->swap_queue); + destroy_workqueue(glob->swap_queue); + glob->swap_queue = NULL; +} +EXPORT_SYMBOL(ttm_mem_global_release); + +static inline void ttm_check_swapping(struct ttm_mem_global *glob) +{ + bool needs_swapping; + + spin_lock(&glob->lock); + needs_swapping = (glob->used_memory > glob->swap_limit || + glob->used_total_memory > + glob->total_memory_swap_limit); + spin_unlock(&glob->lock); + + if (unlikely(needs_swapping)) + (void)queue_work(glob->swap_queue, &glob->work); + +} + +void ttm_mem_global_free(struct ttm_mem_global *glob, + uint64_t amount, bool himem) +{ + spin_lock(&glob->lock); + glob->used_total_memory -= amount; + if (!himem) + glob->used_memory -= amount; + wake_up_all(&glob->queue); + spin_unlock(&glob->lock); +} + +static int ttm_mem_global_reserve(struct ttm_mem_global *glob, + uint64_t amount, bool himem, bool reserve) +{ + uint64_t limit; + uint64_t lomem_limit; + int ret = -ENOMEM; + + spin_lock(&glob->lock); + + if (capable(CAP_SYS_ADMIN)) { + limit = glob->emer_total_memory; + lomem_limit = glob->emer_memory; + } else { + limit = glob->max_total_memory; + lomem_limit = glob->max_memory; + } + + if (unlikely(glob->used_total_memory + amount > limit)) + goto out_unlock; + if (unlikely(!himem && glob->used_memory + amount > lomem_limit)) + goto out_unlock; + + if (reserve) { + glob->used_total_memory += amount; + if (!himem) + glob->used_memory += amount; + } + ret = 0; +out_unlock: + spin_unlock(&glob->lock); + ttm_check_swapping(glob); + + return ret; +} + +int ttm_mem_global_alloc(struct ttm_mem_global *glob, uint64_t memory, + bool no_wait, bool interruptible, bool himem) +{ + int count = TTM_MEMORY_ALLOC_RETRIES; + + while (unlikely(ttm_mem_global_reserve(glob, memory, himem, true) + != 0)) { + if (no_wait) + return -ENOMEM; + if (unlikely(count-- == 0)) + return -ENOMEM; + ttm_shrink(glob, false, memory + (memory >> 2) + 16); + } + + return 0; +} + +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; +} diff --git a/drivers/gpu/drm/ttm/ttm_module.c b/drivers/gpu/drm/ttm/ttm_module.c new file mode 100644 index 0000000000000000000000000000000000000000..59ce8191d5847e564783e8edcd6187851653b255 --- /dev/null +++ b/drivers/gpu/drm/ttm/ttm_module.c @@ -0,0 +1,50 @@ +/************************************************************************** + * + * 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 + * Jerome Glisse + */ +#include +#include + +static int __init ttm_init(void) +{ + ttm_global_init(); + return 0; +} + +static void __exit ttm_exit(void) +{ + ttm_global_release(); +} + +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_tt.c b/drivers/gpu/drm/ttm/ttm_tt.c new file mode 100644 index 0000000000000000000000000000000000000000..c27ab3a877adbac9c80f37ce9a1fc4698f5f8a8a --- /dev/null +++ b/drivers/gpu/drm/ttm/ttm_tt.c @@ -0,0 +1,635 @@ +/************************************************************************** + * + * 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 + */ + +#include +#include +#include +#include +#include +#include +#include +#include "ttm/ttm_module.h" +#include "ttm/ttm_bo_driver.h" +#include "ttm/ttm_placement.h" + +static int ttm_tt_swapin(struct ttm_tt *ttm); + +#if defined(CONFIG_X86) +static void ttm_tt_clflush_page(struct page *page) +{ + uint8_t *page_virtual; + unsigned int i; + + if (unlikely(page == NULL)) + return; + + page_virtual = kmap_atomic(page, KM_USER0); + + for (i = 0; i < PAGE_SIZE; i += boot_cpu_data.x86_clflush_size) + clflush(page_virtual + i); + + kunmap_atomic(page_virtual, KM_USER0); +} + +static void ttm_tt_cache_flush_clflush(struct page *pages[], + unsigned long num_pages) +{ + unsigned long i; + + mb(); + for (i = 0; i < num_pages; ++i) + ttm_tt_clflush_page(*pages++); + mb(); +} +#else +static void ttm_tt_ipi_handler(void *null) +{ + ; +} +#endif + +void ttm_tt_cache_flush(struct page *pages[], unsigned long num_pages) +{ + +#if defined(CONFIG_X86) + if (cpu_has_clflush) { + ttm_tt_cache_flush_clflush(pages, num_pages); + return; + } +#else + if (on_each_cpu(ttm_tt_ipi_handler, NULL, 1) != 0) + printk(KERN_ERR TTM_PFX + "Timed out waiting for drm cache flush.\n"); +#endif +} + +/** + * Allocates storage for pointers to the pages that back the ttm. + * + * Uses kmalloc if possible. Otherwise falls back to vmalloc. + */ +static void ttm_tt_alloc_page_directory(struct ttm_tt *ttm) +{ + unsigned long size = ttm->num_pages * sizeof(*ttm->pages); + ttm->pages = NULL; + + if (size <= PAGE_SIZE) + ttm->pages = kzalloc(size, GFP_KERNEL); + + if (!ttm->pages) { + ttm->pages = vmalloc_user(size); + if (ttm->pages) + ttm->page_flags |= TTM_PAGE_FLAG_VMALLOC; + } +} + +static void ttm_tt_free_page_directory(struct ttm_tt *ttm) +{ + if (ttm->page_flags & TTM_PAGE_FLAG_VMALLOC) { + vfree(ttm->pages); + ttm->page_flags &= ~TTM_PAGE_FLAG_VMALLOC; + } else { + kfree(ttm->pages); + } + ttm->pages = NULL; +} + +static struct page *ttm_tt_alloc_page(unsigned page_flags) +{ + if (page_flags & TTM_PAGE_FLAG_ZERO_ALLOC) + return alloc_page(GFP_HIGHUSER | __GFP_ZERO); + + return alloc_page(GFP_HIGHUSER); +} + +static void ttm_tt_free_user_pages(struct ttm_tt *ttm) +{ + int write; + int dirty; + struct page *page; + int i; + struct ttm_backend *be = ttm->be; + + BUG_ON(!(ttm->page_flags & TTM_PAGE_FLAG_USER)); + write = ((ttm->page_flags & TTM_PAGE_FLAG_WRITE) != 0); + dirty = ((ttm->page_flags & TTM_PAGE_FLAG_USER_DIRTY) != 0); + + if (be) + be->func->clear(be); + + for (i = 0; i < ttm->num_pages; ++i) { + page = ttm->pages[i]; + if (page == NULL) + continue; + + if (page == ttm->dummy_read_page) { + BUG_ON(write); + continue; + } + + if (write && dirty && !PageReserved(page)) + set_page_dirty_lock(page); + + ttm->pages[i] = NULL; + ttm_mem_global_free(ttm->bdev->mem_glob, PAGE_SIZE, false); + put_page(page); + } + ttm->state = tt_unpopulated; + ttm->first_himem_page = ttm->num_pages; + ttm->last_lomem_page = -1; +} + +static struct page *__ttm_tt_get_page(struct ttm_tt *ttm, int index) +{ + struct page *p; + struct ttm_bo_device *bdev = ttm->bdev; + struct ttm_mem_global *mem_glob = bdev->mem_glob; + int ret; + + while (NULL == (p = ttm->pages[index])) { + p = ttm_tt_alloc_page(ttm->page_flags); + + if (!p) + return NULL; + + if (PageHighMem(p)) { + ret = + ttm_mem_global_alloc(mem_glob, PAGE_SIZE, + false, false, true); + if (unlikely(ret != 0)) + goto out_err; + ttm->pages[--ttm->first_himem_page] = p; + } else { + ret = + ttm_mem_global_alloc(mem_glob, PAGE_SIZE, + false, false, false); + if (unlikely(ret != 0)) + goto out_err; + ttm->pages[++ttm->last_lomem_page] = p; + } + } + return p; +out_err: + put_page(p); + return NULL; +} + +struct page *ttm_tt_get_page(struct ttm_tt *ttm, int index) +{ + int ret; + + if (unlikely(ttm->page_flags & TTM_PAGE_FLAG_SWAPPED)) { + ret = ttm_tt_swapin(ttm); + if (unlikely(ret != 0)) + return NULL; + } + return __ttm_tt_get_page(ttm, index); +} + +int ttm_tt_populate(struct ttm_tt *ttm) +{ + struct page *page; + unsigned long i; + struct ttm_backend *be; + int ret; + + if (ttm->state != tt_unpopulated) + return 0; + + if (unlikely(ttm->page_flags & TTM_PAGE_FLAG_SWAPPED)) { + ret = ttm_tt_swapin(ttm); + if (unlikely(ret != 0)) + return ret; + } + + be = ttm->be; + + for (i = 0; i < ttm->num_pages; ++i) { + page = __ttm_tt_get_page(ttm, i); + if (!page) + return -ENOMEM; + } + + be->func->populate(be, ttm->num_pages, ttm->pages, + ttm->dummy_read_page); + ttm->state = tt_unbound; + return 0; +} + +#ifdef CONFIG_X86 +static inline int ttm_tt_set_page_caching(struct page *p, + enum ttm_caching_state c_state) +{ + if (PageHighMem(p)) + return 0; + + switch (c_state) { + case tt_cached: + return set_pages_wb(p, 1); + case tt_wc: + return set_memory_wc((unsigned long) page_address(p), 1); + default: + return set_pages_uc(p, 1); + } +} +#else /* CONFIG_X86 */ +static inline int ttm_tt_set_page_caching(struct page *p, + enum ttm_caching_state c_state) +{ + 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 (c_state != tt_cached) { + ret = ttm_tt_populate(ttm); + if (unlikely(ret != 0)) + return ret; + } + + if (ttm->caching_state == tt_cached) + ttm_tt_cache_flush(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, 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, + 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); +} + +static void ttm_tt_free_alloced_pages(struct ttm_tt *ttm) +{ + int i; + struct page *cur_page; + struct ttm_backend *be = ttm->be; + + if (be) + be->func->clear(be); + (void)ttm_tt_set_caching(ttm, tt_cached); + for (i = 0; i < ttm->num_pages; ++i) { + cur_page = ttm->pages[i]; + ttm->pages[i] = NULL; + if (cur_page) { + if (page_count(cur_page) != 1) + printk(KERN_ERR TTM_PFX + "Erroneous page count. " + "Leaking pages.\n"); + ttm_mem_global_free(ttm->bdev->mem_glob, PAGE_SIZE, + PageHighMem(cur_page)); + __free_page(cur_page); + } + } + ttm->state = tt_unpopulated; + ttm->first_himem_page = ttm->num_pages; + ttm->last_lomem_page = -1; +} + +void ttm_tt_destroy(struct ttm_tt *ttm) +{ + struct ttm_backend *be; + + if (unlikely(ttm == NULL)) + return; + + be = ttm->be; + if (likely(be != NULL)) { + be->func->destroy(be); + ttm->be = NULL; + } + + if (likely(ttm->pages != NULL)) { + if (ttm->page_flags & TTM_PAGE_FLAG_USER) + ttm_tt_free_user_pages(ttm); + else + ttm_tt_free_alloced_pages(ttm); + + ttm_tt_free_page_directory(ttm); + } + + if (!(ttm->page_flags & TTM_PAGE_FLAG_PERSISTANT_SWAP) && + ttm->swap_storage) + fput(ttm->swap_storage); + + kfree(ttm); +} + +int ttm_tt_set_user(struct ttm_tt *ttm, + struct task_struct *tsk, + unsigned long start, unsigned long num_pages) +{ + struct mm_struct *mm = tsk->mm; + int ret; + int write = (ttm->page_flags & TTM_PAGE_FLAG_WRITE) != 0; + struct ttm_mem_global *mem_glob = ttm->bdev->mem_glob; + + BUG_ON(num_pages != ttm->num_pages); + BUG_ON((ttm->page_flags & TTM_PAGE_FLAG_USER) == 0); + + /** + * Account user pages as lowmem pages for now. + */ + + ret = ttm_mem_global_alloc(mem_glob, num_pages * PAGE_SIZE, + false, false, false); + if (unlikely(ret != 0)) + return ret; + + down_read(&mm->mmap_sem); + ret = get_user_pages(tsk, mm, start, num_pages, + write, 0, ttm->pages, NULL); + up_read(&mm->mmap_sem); + + if (ret != num_pages && write) { + ttm_tt_free_user_pages(ttm); + ttm_mem_global_free(mem_glob, num_pages * PAGE_SIZE, false); + return -ENOMEM; + } + + ttm->tsk = tsk; + ttm->start = start; + ttm->state = tt_unbound; + + return 0; +} + +struct ttm_tt *ttm_tt_create(struct ttm_bo_device *bdev, unsigned long size, + uint32_t page_flags, struct page *dummy_read_page) +{ + struct ttm_bo_driver *bo_driver = bdev->driver; + struct ttm_tt *ttm; + + if (!bo_driver) + return NULL; + + ttm = kzalloc(sizeof(*ttm), GFP_KERNEL); + if (!ttm) + return NULL; + + ttm->bdev = bdev; + + ttm->num_pages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; + ttm->first_himem_page = ttm->num_pages; + ttm->last_lomem_page = -1; + ttm->caching_state = tt_cached; + ttm->page_flags = page_flags; + + ttm->dummy_read_page = dummy_read_page; + + ttm_tt_alloc_page_directory(ttm); + if (!ttm->pages) { + ttm_tt_destroy(ttm); + printk(KERN_ERR TTM_PFX "Failed allocating page table\n"); + return NULL; + } + ttm->be = bo_driver->create_ttm_backend_entry(bdev); + if (!ttm->be) { + ttm_tt_destroy(ttm); + printk(KERN_ERR TTM_PFX "Failed creating ttm backend entry\n"); + return NULL; + } + ttm->state = tt_unpopulated; + return ttm; +} + +void ttm_tt_unbind(struct ttm_tt *ttm) +{ + int ret; + struct ttm_backend *be = ttm->be; + + if (ttm->state == tt_bound) { + ret = be->func->unbind(be); + BUG_ON(ret); + ttm->state = tt_unbound; + } +} + +int ttm_tt_bind(struct ttm_tt *ttm, struct ttm_mem_reg *bo_mem) +{ + int ret = 0; + struct ttm_backend *be; + + if (!ttm) + return -EINVAL; + + if (ttm->state == tt_bound) + return 0; + + be = ttm->be; + + ret = ttm_tt_populate(ttm); + if (ret) + return ret; + + ret = be->func->bind(be, bo_mem); + if (ret) { + printk(KERN_ERR TTM_PFX "Couldn't bind backend.\n"); + return ret; + } + + ttm->state = tt_bound; + + if (ttm->page_flags & TTM_PAGE_FLAG_USER) + ttm->page_flags |= TTM_PAGE_FLAG_USER_DIRTY; + return 0; +} +EXPORT_SYMBOL(ttm_tt_bind); + +static 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; + void *from_virtual; + void *to_virtual; + int i; + int ret; + + if (ttm->page_flags & TTM_PAGE_FLAG_USER) { + ret = ttm_tt_set_user(ttm, ttm->tsk, ttm->start, + ttm->num_pages); + if (unlikely(ret != 0)) + return ret; + + ttm->page_flags &= ~TTM_PAGE_FLAG_SWAPPED; + return 0; + } + + swap_storage = ttm->swap_storage; + BUG_ON(swap_storage == NULL); + + swap_space = swap_storage->f_path.dentry->d_inode->i_mapping; + + for (i = 0; i < ttm->num_pages; ++i) { + from_page = read_mapping_page(swap_space, i, NULL); + if (IS_ERR(from_page)) + goto out_err; + to_page = __ttm_tt_get_page(ttm, i); + if (unlikely(to_page == NULL)) + goto out_err; + + preempt_disable(); + from_virtual = kmap_atomic(from_page, KM_USER0); + to_virtual = kmap_atomic(to_page, KM_USER1); + memcpy(to_virtual, from_virtual, PAGE_SIZE); + kunmap_atomic(to_virtual, KM_USER1); + kunmap_atomic(from_virtual, KM_USER0); + preempt_enable(); + page_cache_release(from_page); + } + + if (!(ttm->page_flags & TTM_PAGE_FLAG_PERSISTANT_SWAP)) + fput(swap_storage); + ttm->swap_storage = NULL; + ttm->page_flags &= ~TTM_PAGE_FLAG_SWAPPED; + + return 0; +out_err: + ttm_tt_free_alloced_pages(ttm); + return -ENOMEM; +} + +int ttm_tt_swapout(struct ttm_tt *ttm, struct file *persistant_swap_storage) +{ + struct address_space *swap_space; + struct file *swap_storage; + struct page *from_page; + struct page *to_page; + void *from_virtual; + void *to_virtual; + int i; + + BUG_ON(ttm->state != tt_unbound && ttm->state != tt_unpopulated); + BUG_ON(ttm->caching_state != tt_cached); + + /* + * For user buffers, just unpin the pages, as there should be + * vma references. + */ + + if (ttm->page_flags & TTM_PAGE_FLAG_USER) { + ttm_tt_free_user_pages(ttm); + ttm->page_flags |= TTM_PAGE_FLAG_SWAPPED; + ttm->swap_storage = NULL; + return 0; + } + + if (!persistant_swap_storage) { + swap_storage = shmem_file_setup("ttm swap", + ttm->num_pages << PAGE_SHIFT, + 0); + if (unlikely(IS_ERR(swap_storage))) { + printk(KERN_ERR "Failed allocating swap storage.\n"); + return -ENOMEM; + } + } else + swap_storage = persistant_swap_storage; + + swap_space = swap_storage->f_path.dentry->d_inode->i_mapping; + + for (i = 0; i < ttm->num_pages; ++i) { + from_page = ttm->pages[i]; + if (unlikely(from_page == NULL)) + continue; + to_page = read_mapping_page(swap_space, i, NULL); + if (unlikely(to_page == NULL)) + goto out_err; + + preempt_disable(); + from_virtual = kmap_atomic(from_page, KM_USER0); + to_virtual = kmap_atomic(to_page, KM_USER1); + memcpy(to_virtual, from_virtual, PAGE_SIZE); + kunmap_atomic(to_virtual, KM_USER1); + kunmap_atomic(from_virtual, KM_USER0); + preempt_enable(); + set_page_dirty(to_page); + mark_page_accessed(to_page); + page_cache_release(to_page); + } + + ttm_tt_free_alloced_pages(ttm); + ttm->swap_storage = swap_storage; + ttm->page_flags |= TTM_PAGE_FLAG_SWAPPED; + if (persistant_swap_storage) + ttm->page_flags |= TTM_PAGE_FLAG_PERSISTANT_SWAP; + + return 0; +out_err: + if (!persistant_swap_storage) + fput(swap_storage); + + return -ENOMEM; +} diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h new file mode 100644 index 0000000000000000000000000000000000000000..cd22ab4b495c2ea51cf5ca98d5b0e93c3edbe33d --- /dev/null +++ b/include/drm/ttm/ttm_bo_api.h @@ -0,0 +1,618 @@ +/************************************************************************** + * + * 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 + */ + +#ifndef _TTM_BO_API_H_ +#define _TTM_BO_API_H_ + +#include "drm_hashtab.h" +#include +#include +#include +#include +#include +#include +#include + +struct ttm_bo_device; + +struct drm_mm_node; + +/** + * struct ttm_mem_reg + * + * @mm_node: Memory manager node. + * @size: Requested size of memory region. + * @num_pages: Actual size of memory region in pages. + * @page_alignment: Page alignment. + * @placement: Placement flags. + * + * Structure indicating the placement and space resources used by a + * buffer object. + */ + +struct ttm_mem_reg { + struct drm_mm_node *mm_node; + unsigned long size; + unsigned long num_pages; + uint32_t page_alignment; + uint32_t mem_type; + uint32_t placement; +}; + +/** + * enum ttm_bo_type + * + * @ttm_bo_type_device: These are 'normal' buffers that can + * be mmapped by user space. Each of these bos occupy a slot in the + * device address space, that can be used for normal vm operations. + * + * @ttm_bo_type_user: These are user-space memory areas that are made + * available to the GPU by mapping the buffer pages into the GPU aperture + * space. These buffers cannot be mmaped from the device address space. + * + * @ttm_bo_type_kernel: These buffers are like ttm_bo_type_device buffers, + * but they cannot be accessed from user-space. For kernel-only use. + */ + +enum ttm_bo_type { + ttm_bo_type_device, + ttm_bo_type_user, + ttm_bo_type_kernel +}; + +struct ttm_tt; + +/** + * struct ttm_buffer_object + * + * @bdev: Pointer to the buffer object device structure. + * @buffer_start: The virtual user-space start address of ttm_bo_type_user + * buffers. + * @type: The bo type. + * @destroy: Destruction function. If NULL, kfree is used. + * @num_pages: Actual number of pages. + * @addr_space_offset: Address space offset. + * @acc_size: Accounted size for this object. + * @kref: Reference count of this buffer object. When this refcount reaches + * zero, the object is put on the delayed delete list. + * @list_kref: List reference count of this buffer object. This member is + * used to avoid destruction while the buffer object is still on a list. + * Lru lists may keep one refcount, the delayed delete list, and kref != 0 + * keeps one refcount. When this refcount reaches zero, + * the object is destroyed. + * @event_queue: Queue for processes waiting on buffer object status change. + * @lock: spinlock protecting mostly synchronization members. + * @proposed_placement: Proposed placement for the buffer. Changed only by the + * creator prior to validation as opposed to bo->mem.proposed_flags which is + * changed by the implementation prior to a buffer move if it wants to outsmart + * the buffer creator / user. This latter happens, for example, at eviction. + * @mem: structure describing current placement. + * @persistant_swap_storage: Usually the swap storage is deleted for buffers + * pinned in physical memory. If this behaviour is not desired, this member + * holds a pointer to a persistant shmem object. + * @ttm: TTM structure holding system pages. + * @evicted: Whether the object was evicted without user-space knowing. + * @cpu_writes: For synchronization. Number of cpu writers. + * @lru: List head for the lru list. + * @ddestroy: List head for the delayed destroy list. + * @swap: List head for swap LRU list. + * @val_seq: Sequence of the validation holding the @reserved lock. + * Used to avoid starvation when many processes compete to validate the + * buffer. This member is protected by the bo_device::lru_lock. + * @seq_valid: The value of @val_seq is valid. This value is protected by + * the bo_device::lru_lock. + * @reserved: Deadlock-free lock used for synchronization state transitions. + * @sync_obj_arg: Opaque argument to synchronization object function. + * @sync_obj: Pointer to a synchronization object. + * @priv_flags: Flags describing buffer object internal state. + * @vm_rb: Rb node for the vm rb tree. + * @vm_node: Address space manager node. + * @offset: The current GPU offset, which can have different meanings + * depending on the memory type. For SYSTEM type memory, it should be 0. + * @cur_placement: Hint of current placement. + * + * Base class for TTM buffer object, that deals with data placement and CPU + * mappings. GPU mappings are really up to the driver, but for simpler GPUs + * the driver can usually use the placement offset @offset directly as the + * GPU virtual address. For drivers implementing multiple + * GPU memory manager contexts, the driver should manage the address space + * in these contexts separately and use these objects to get the correct + * placement and caching for these GPU maps. This makes it possible to use + * these objects for even quite elaborate memory management schemes. + * The destroy member, the API visibility of this object makes it possible + * to derive driver specific types. + */ + +struct ttm_buffer_object { + /** + * Members constant at init. + */ + + struct ttm_bo_device *bdev; + unsigned long buffer_start; + enum ttm_bo_type type; + void (*destroy) (struct ttm_buffer_object *); + unsigned long num_pages; + uint64_t addr_space_offset; + size_t acc_size; + + /** + * Members not needing protection. + */ + + struct kref kref; + struct kref list_kref; + wait_queue_head_t event_queue; + spinlock_t lock; + + /** + * Members protected by the bo::reserved lock. + */ + + uint32_t proposed_placement; + struct ttm_mem_reg mem; + struct file *persistant_swap_storage; + struct ttm_tt *ttm; + bool evicted; + + /** + * Members protected by the bo::reserved lock only when written to. + */ + + atomic_t cpu_writers; + + /** + * Members protected by the bdev::lru_lock. + */ + + struct list_head lru; + struct list_head ddestroy; + struct list_head swap; + uint32_t val_seq; + bool seq_valid; + + /** + * Members protected by the bdev::lru_lock + * only when written to. + */ + + atomic_t reserved; + + + /** + * Members protected by the bo::lock + */ + + void *sync_obj_arg; + void *sync_obj; + unsigned long priv_flags; + + /** + * Members protected by the bdev::vm_lock + */ + + struct rb_node vm_rb; + struct drm_mm_node *vm_node; + + + /** + * Special members that are protected by the reserve lock + * and the bo::lock when written to. Can be read with + * either of these locks held. + */ + + unsigned long offset; + uint32_t cur_placement; +}; + +/** + * struct ttm_bo_kmap_obj + * + * @virtual: The current kernel virtual address. + * @page: The page when kmap'ing a single page. + * @bo_kmap_type: Type of bo_kmap. + * + * Object describing a kernel mapping. Since a TTM bo may be located + * in various memory types with various caching policies, the + * mapping can either be an ioremap, a vmap, a kmap or part of a + * premapped region. + */ + +struct ttm_bo_kmap_obj { + void *virtual; + struct page *page; + enum { + ttm_bo_map_iomap, + ttm_bo_map_vmap, + ttm_bo_map_kmap, + ttm_bo_map_premapped, + } bo_kmap_type; +}; + +/** + * ttm_bo_reference - reference a struct ttm_buffer_object + * + * @bo: The buffer object. + * + * Returns a refcounted pointer to a buffer object. + */ + +static inline struct ttm_buffer_object * +ttm_bo_reference(struct ttm_buffer_object *bo) +{ + kref_get(&bo->kref); + return bo; +} + +/** + * ttm_bo_wait - wait for buffer idle. + * + * @bo: The buffer object. + * @interruptible: Use interruptible wait. + * @no_wait: Return immediately if buffer is busy. + * + * This function must be called with the bo::mutex held, and makes + * sure any previous rendering to the buffer is completed. + * Note: It might be necessary to block validations before the + * wait by reserving the buffer. + * Returns -EBUSY if no_wait is true and the buffer is busy. + * Returns -ERESTART if interrupted by a signal. + */ +extern int ttm_bo_wait(struct ttm_buffer_object *bo, bool lazy, + bool interruptible, bool no_wait); +/** + * ttm_buffer_object_validate + * + * @bo: The buffer object. + * @proposed_placement: Proposed_placement for the buffer object. + * @interruptible: Sleep interruptible if sleeping. + * @no_wait: Return immediately if the buffer is busy. + * + * Changes placement and caching policy of the buffer object + * according to bo::proposed_flags. + * Returns + * -EINVAL on invalid proposed_flags. + * -ENOMEM on out-of-memory condition. + * -EBUSY if no_wait is true and buffer busy. + * -ERESTART if interrupted by a signal. + */ +extern int ttm_buffer_object_validate(struct ttm_buffer_object *bo, + uint32_t proposed_placement, + bool interruptible, bool no_wait); +/** + * ttm_bo_unref + * + * @bo: The buffer object. + * + * Unreference and clear a pointer to a buffer object. + */ +extern void ttm_bo_unref(struct ttm_buffer_object **bo); + +/** + * ttm_bo_synccpu_write_grab + * + * @bo: The buffer object: + * @no_wait: Return immediately if buffer is busy. + * + * Synchronizes a buffer object for CPU RW access. This means + * blocking command submission that affects the buffer and + * waiting for buffer idle. This lock is recursive. + * Returns + * -EBUSY if the buffer is busy and no_wait is true. + * -ERESTART if interrupted by a signal. + */ + +extern int +ttm_bo_synccpu_write_grab(struct ttm_buffer_object *bo, bool no_wait); +/** + * ttm_bo_synccpu_write_release: + * + * @bo : The buffer object. + * + * Releases a synccpu lock. + */ +extern void ttm_bo_synccpu_write_release(struct ttm_buffer_object *bo); + +/** + * ttm_buffer_object_init + * + * @bdev: Pointer to a ttm_bo_device struct. + * @bo: Pointer to a ttm_buffer_object to be initialized. + * @size: Requested size of buffer object. + * @type: Requested type of buffer object. + * @flags: Initial placement flags. + * @page_alignment: Data alignment in pages. + * @buffer_start: Virtual address of user space data backing a + * user buffer object. + * @interruptible: If needing to sleep to wait for GPU resources, + * sleep interruptible. + * @persistant_swap_storage: Usually the swap storage is deleted for buffers + * pinned in physical memory. If this behaviour is not desired, this member + * holds a pointer to a persistant shmem object. Typically, this would + * point to the shmem object backing a GEM object if TTM is used to back a + * GEM user interface. + * @acc_size: Accounted size for this object. + * @destroy: Destroy function. Use NULL for kfree(). + * + * This function initializes a pre-allocated struct ttm_buffer_object. + * As this object may be part of a larger structure, this function, + * together with the @destroy function, + * enables driver-specific objects derived from a ttm_buffer_object. + * On successful return, the object kref and list_kref are set to 1. + * Returns + * -ENOMEM: Out of memory. + * -EINVAL: Invalid placement flags. + * -ERESTART: Interrupted by signal while sleeping waiting for resources. + */ + +extern int ttm_buffer_object_init(struct ttm_bo_device *bdev, + struct ttm_buffer_object *bo, + unsigned long size, + enum ttm_bo_type type, + uint32_t flags, + uint32_t page_alignment, + unsigned long buffer_start, + bool interrubtible, + struct file *persistant_swap_storage, + size_t acc_size, + void (*destroy) (struct ttm_buffer_object *)); +/** + * ttm_bo_synccpu_object_init + * + * @bdev: Pointer to a ttm_bo_device struct. + * @bo: Pointer to a ttm_buffer_object to be initialized. + * @size: Requested size of buffer object. + * @type: Requested type of buffer object. + * @flags: Initial placement flags. + * @page_alignment: Data alignment in pages. + * @buffer_start: Virtual address of user space data backing a + * user buffer object. + * @interruptible: If needing to sleep while waiting for GPU resources, + * sleep interruptible. + * @persistant_swap_storage: Usually the swap storage is deleted for buffers + * pinned in physical memory. If this behaviour is not desired, this member + * holds a pointer to a persistant shmem object. Typically, this would + * point to the shmem object backing a GEM object if TTM is used to back a + * GEM user interface. + * @p_bo: On successful completion *p_bo points to the created object. + * + * This function allocates a ttm_buffer_object, and then calls + * ttm_buffer_object_init on that object. + * The destroy function is set to kfree(). + * Returns + * -ENOMEM: Out of memory. + * -EINVAL: Invalid placement flags. + * -ERESTART: Interrupted by signal while waiting for resources. + */ + +extern int ttm_buffer_object_create(struct ttm_bo_device *bdev, + unsigned long size, + enum ttm_bo_type type, + uint32_t flags, + uint32_t page_alignment, + unsigned long buffer_start, + bool interruptible, + struct file *persistant_swap_storage, + struct ttm_buffer_object **p_bo); + +/** + * ttm_bo_check_placement + * + * @bo: the buffer object. + * @set_flags: placement flags to set. + * @clr_flags: placement flags to clear. + * + * Performs minimal validity checking on an intended change of + * placement flags. + * Returns + * -EINVAL: Intended change is invalid or not allowed. + */ + +extern int ttm_bo_check_placement(struct ttm_buffer_object *bo, + uint32_t set_flags, uint32_t clr_flags); + +/** + * ttm_bo_init_mm + * + * @bdev: Pointer to a ttm_bo_device struct. + * @mem_type: The memory type. + * @p_offset: offset for managed area in pages. + * @p_size: size managed area in pages. + * + * Initialize a manager for a given memory type. + * Note: if part of driver firstopen, it must be protected from a + * potentially racing lastclose. + * Returns: + * -EINVAL: invalid size or memory type. + * -ENOMEM: Not enough memory. + * May also return driver-specified errors. + */ + +extern int ttm_bo_init_mm(struct ttm_bo_device *bdev, unsigned type, + unsigned long p_offset, unsigned long p_size); +/** + * ttm_bo_clean_mm + * + * @bdev: Pointer to a ttm_bo_device struct. + * @mem_type: The memory type. + * + * Take down a manager for a given memory type after first walking + * the LRU list to evict any buffers left alive. + * + * Normally, this function is part of lastclose() or unload(), and at that + * point there shouldn't be any buffers left created by user-space, since + * there should've been removed by the file descriptor release() method. + * However, before this function is run, make sure to signal all sync objects, + * and verify that the delayed delete queue is empty. The driver must also + * make sure that there are no NO_EVICT buffers present in this memory type + * when the call is made. + * + * If this function is part of a VT switch, the caller must make sure that + * there are no appications currently validating buffers before this + * function is called. The caller can do that by first taking the + * struct ttm_bo_device::ttm_lock in write mode. + * + * Returns: + * -EINVAL: invalid or uninitialized memory type. + * -EBUSY: There are still buffers left in this memory type. + */ + +extern int ttm_bo_clean_mm(struct ttm_bo_device *bdev, unsigned mem_type); + +/** + * ttm_bo_evict_mm + * + * @bdev: Pointer to a ttm_bo_device struct. + * @mem_type: The memory type. + * + * Evicts all buffers on the lru list of the memory type. + * This is normally part of a VT switch or an + * out-of-memory-space-due-to-fragmentation handler. + * The caller must make sure that there are no other processes + * currently validating buffers, and can do that by taking the + * struct ttm_bo_device::ttm_lock in write mode. + * + * Returns: + * -EINVAL: Invalid or uninitialized memory type. + * -ERESTART: The call was interrupted by a signal while waiting to + * evict a buffer. + */ + +extern int ttm_bo_evict_mm(struct ttm_bo_device *bdev, unsigned mem_type); + +/** + * ttm_kmap_obj_virtual + * + * @map: A struct ttm_bo_kmap_obj returned from ttm_bo_kmap. + * @is_iomem: Pointer to an integer that on return indicates 1 if the + * virtual map is io memory, 0 if normal memory. + * + * Returns the virtual address of a buffer object area mapped by ttm_bo_kmap. + * If *is_iomem is 1 on return, the virtual address points to an io memory area, + * that should strictly be accessed by the iowriteXX() and similar functions. + */ + +static inline void *ttm_kmap_obj_virtual(struct ttm_bo_kmap_obj *map, + bool *is_iomem) +{ + *is_iomem = (map->bo_kmap_type == ttm_bo_map_iomap || + map->bo_kmap_type == ttm_bo_map_premapped); + return map->virtual; +} + +/** + * ttm_bo_kmap + * + * @bo: The buffer object. + * @start_page: The first page to map. + * @num_pages: Number of pages to map. + * @map: pointer to a struct ttm_bo_kmap_obj representing the map. + * + * Sets up a kernel virtual mapping, using ioremap, vmap or kmap to the + * data in the buffer object. The ttm_kmap_obj_virtual function can then be + * used to obtain a virtual address to the data. + * + * Returns + * -ENOMEM: Out of memory. + * -EINVAL: Invalid range. + */ + +extern int ttm_bo_kmap(struct ttm_buffer_object *bo, unsigned long start_page, + unsigned long num_pages, struct ttm_bo_kmap_obj *map); + +/** + * ttm_bo_kunmap + * + * @map: Object describing the map to unmap. + * + * Unmaps a kernel map set up by ttm_bo_kmap. + */ + +extern void ttm_bo_kunmap(struct ttm_bo_kmap_obj *map); + +#if 0 +#endif + +/** + * ttm_fbdev_mmap - mmap fbdev memory backed by a ttm buffer object. + * + * @vma: vma as input from the fbdev mmap method. + * @bo: The bo backing the address space. The address space will + * have the same size as the bo, and start at offset 0. + * + * This function is intended to be called by the fbdev mmap method + * if the fbdev address space is to be backed by a bo. + */ + +extern int ttm_fbdev_mmap(struct vm_area_struct *vma, + struct ttm_buffer_object *bo); + +/** + * ttm_bo_mmap - mmap out of the ttm device address space. + * + * @filp: filp as input from the mmap method. + * @vma: vma as input from the mmap method. + * @bdev: Pointer to the ttm_bo_device with the address space manager. + * + * This function is intended to be called by the device mmap method. + * if the device address space is to be backed by the bo manager. + */ + +extern int ttm_bo_mmap(struct file *filp, struct vm_area_struct *vma, + struct ttm_bo_device *bdev); + +/** + * ttm_bo_io + * + * @bdev: Pointer to the struct ttm_bo_device. + * @filp: Pointer to the struct file attempting to read / write. + * @wbuf: User-space pointer to address of buffer to write. NULL on read. + * @rbuf: User-space pointer to address of buffer to read into. + * Null on write. + * @count: Number of bytes to read / write. + * @f_pos: Pointer to current file position. + * @write: 1 for read, 0 for write. + * + * This function implements read / write into ttm buffer objects, and is + * intended to + * be called from the fops::read and fops::write method. + * Returns: + * See man (2) write, man(2) read. In particular, + * the function may return -EINTR if + * interrupted by a signal. + */ + +extern 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); + +extern void ttm_bo_swapout_all(struct ttm_bo_device *bdev); + +#endif diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h new file mode 100644 index 0000000000000000000000000000000000000000..62ed733c52a23898a893c51bf05eb80e17995bf4 --- /dev/null +++ b/include/drm/ttm/ttm_bo_driver.h @@ -0,0 +1,867 @@ +/************************************************************************** + * + * 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 + */ +#ifndef _TTM_BO_DRIVER_H_ +#define _TTM_BO_DRIVER_H_ + +#include "ttm/ttm_bo_api.h" +#include "ttm/ttm_memory.h" +#include "drm_mm.h" +#include "linux/workqueue.h" +#include "linux/fs.h" +#include "linux/spinlock.h" + +struct ttm_backend; + +struct ttm_backend_func { + /** + * struct ttm_backend_func member populate + * + * @backend: Pointer to a struct ttm_backend. + * @num_pages: Number of pages to populate. + * @pages: Array of pointers to ttm pages. + * @dummy_read_page: Page to be used instead of NULL pages in the + * array @pages. + * + * Populate the backend with ttm pages. Depending on the backend, + * it may or may not copy the @pages array. + */ + int (*populate) (struct ttm_backend *backend, + unsigned long num_pages, struct page **pages, + struct page *dummy_read_page); + /** + * struct ttm_backend_func member clear + * + * @backend: Pointer to a struct ttm_backend. + * + * This is an "unpopulate" function. Release all resources + * allocated with populate. + */ + void (*clear) (struct ttm_backend *backend); + + /** + * struct ttm_backend_func member bind + * + * @backend: Pointer to a struct ttm_backend. + * @bo_mem: Pointer to a struct ttm_mem_reg describing the + * memory type and location for binding. + * + * Bind the backend pages into the aperture in the location + * indicated by @bo_mem. This function should be able to handle + * differences between aperture- and system page sizes. + */ + int (*bind) (struct ttm_backend *backend, struct ttm_mem_reg *bo_mem); + + /** + * struct ttm_backend_func member unbind + * + * @backend: Pointer to a struct ttm_backend. + * + * Unbind previously bound backend pages. This function should be + * able to handle differences between aperture- and system page sizes. + */ + int (*unbind) (struct ttm_backend *backend); + + /** + * struct ttm_backend_func member destroy + * + * @backend: Pointer to a struct ttm_backend. + * + * Destroy the backend. + */ + void (*destroy) (struct ttm_backend *backend); +}; + +/** + * struct ttm_backend + * + * @bdev: Pointer to a struct ttm_bo_device. + * @flags: For driver use. + * @func: Pointer to a struct ttm_backend_func that describes + * the backend methods. + * + */ + +struct ttm_backend { + struct ttm_bo_device *bdev; + uint32_t flags; + struct ttm_backend_func *func; +}; + +#define TTM_PAGE_FLAG_VMALLOC (1 << 0) +#define TTM_PAGE_FLAG_USER (1 << 1) +#define TTM_PAGE_FLAG_USER_DIRTY (1 << 2) +#define TTM_PAGE_FLAG_WRITE (1 << 3) +#define TTM_PAGE_FLAG_SWAPPED (1 << 4) +#define TTM_PAGE_FLAG_PERSISTANT_SWAP (1 << 5) +#define TTM_PAGE_FLAG_ZERO_ALLOC (1 << 6) + +enum ttm_caching_state { + tt_uncached, + tt_wc, + tt_cached +}; + +/** + * struct ttm_tt + * + * @dummy_read_page: Page to map where the ttm_tt page array contains a NULL + * pointer. + * @pages: Array of pages backing the data. + * @first_himem_page: Himem pages are put last in the page array, which + * enables us to run caching attribute changes on only the first part + * of the page array containing lomem pages. This is the index of the + * first himem page. + * @last_lomem_page: Index of the last lomem page in the page array. + * @num_pages: Number of pages in the page array. + * @bdev: Pointer to the current struct ttm_bo_device. + * @be: Pointer to the ttm backend. + * @tsk: The task for user ttm. + * @start: virtual address for user ttm. + * @swap_storage: Pointer to shmem struct file for swap storage. + * @caching_state: The current caching state of the pages. + * @state: The current binding state of the pages. + * + * This is a structure holding the pages, caching- and aperture binding + * status for a buffer object that isn't backed by fixed (VRAM / AGP) + * memory. + */ + +struct ttm_tt { + struct page *dummy_read_page; + struct page **pages; + long first_himem_page; + long last_lomem_page; + uint32_t page_flags; + unsigned long num_pages; + struct ttm_bo_device *bdev; + struct ttm_backend *be; + struct task_struct *tsk; + unsigned long start; + struct file *swap_storage; + enum ttm_caching_state caching_state; + enum { + tt_bound, + tt_unbound, + tt_unpopulated, + } state; +}; + +#define TTM_MEMTYPE_FLAG_FIXED (1 << 0) /* Fixed (on-card) PCI memory */ +#define TTM_MEMTYPE_FLAG_MAPPABLE (1 << 1) /* Memory mappable */ +#define TTM_MEMTYPE_FLAG_NEEDS_IOREMAP (1 << 2) /* Fixed memory needs ioremap + before kernel access. */ +#define TTM_MEMTYPE_FLAG_CMA (1 << 3) /* Can't map aperture */ + +/** + * struct ttm_mem_type_manager + * + * @has_type: The memory type has been initialized. + * @use_type: The memory type is enabled. + * @flags: TTM_MEMTYPE_XX flags identifying the traits of the memory + * managed by this memory type. + * @gpu_offset: If used, the GPU offset of the first managed page of + * fixed memory or the first managed location in an aperture. + * @io_offset: The io_offset of the first managed page of IO memory or + * the first managed location in an aperture. For TTM_MEMTYPE_FLAG_CMA + * memory, this should be set to NULL. + * @io_size: The size of a managed IO region (fixed memory or aperture). + * @io_addr: Virtual kernel address if the io region is pre-mapped. For + * TTM_MEMTYPE_FLAG_NEEDS_IOREMAP there is no pre-mapped io map and + * @io_addr should be set to NULL. + * @size: Size of the managed region. + * @available_caching: A mask of available caching types, TTM_PL_FLAG_XX, + * as defined in ttm_placement_common.h + * @default_caching: The default caching policy used for a buffer object + * placed in this memory type if the user doesn't provide one. + * @manager: The range manager used for this memory type. FIXME: If the aperture + * has a page size different from the underlying system, the granularity + * of this manager should take care of this. But the range allocating code + * in ttm_bo.c needs to be modified for this. + * @lru: The lru list for this memory type. + * + * This structure is used to identify and manage memory types for a device. + * It's set up by the ttm_bo_driver::init_mem_type method. + */ + +struct ttm_mem_type_manager { + + /* + * No protection. Constant from start. + */ + + bool has_type; + bool use_type; + uint32_t flags; + unsigned long gpu_offset; + unsigned long io_offset; + unsigned long io_size; + void *io_addr; + uint64_t size; + uint32_t available_caching; + uint32_t default_caching; + + /* + * Protected by the bdev->lru_lock. + * TODO: Consider one lru_lock per ttm_mem_type_manager. + * Plays ill with list removal, though. + */ + + struct drm_mm manager; + struct list_head lru; +}; + +/** + * struct ttm_bo_driver + * + * @mem_type_prio: Priority array of memory types to place a buffer object in + * if it fits without evicting buffers from any of these memory types. + * @mem_busy_prio: Priority array of memory types to place a buffer object in + * if it needs to evict buffers to make room. + * @num_mem_type_prio: Number of elements in the @mem_type_prio array. + * @num_mem_busy_prio: Number of elements in the @num_mem_busy_prio array. + * @create_ttm_backend_entry: Callback to create a struct ttm_backend. + * @invalidate_caches: Callback to invalidate read caches when a buffer object + * has been evicted. + * @init_mem_type: Callback to initialize a struct ttm_mem_type_manager + * structure. + * @evict_flags: Callback to obtain placement flags when a buffer is evicted. + * @move: Callback for a driver to hook in accelerated functions to + * move a buffer. + * If set to NULL, a potentially slow memcpy() move is used. + * @sync_obj_signaled: See ttm_fence_api.h + * @sync_obj_wait: See ttm_fence_api.h + * @sync_obj_flush: See ttm_fence_api.h + * @sync_obj_unref: See ttm_fence_api.h + * @sync_obj_ref: See ttm_fence_api.h + */ + +struct ttm_bo_driver { + const uint32_t *mem_type_prio; + const uint32_t *mem_busy_prio; + uint32_t num_mem_type_prio; + uint32_t num_mem_busy_prio; + + /** + * struct ttm_bo_driver member create_ttm_backend_entry + * + * @bdev: The buffer object device. + * + * Create a driver specific struct ttm_backend. + */ + + struct ttm_backend *(*create_ttm_backend_entry) + (struct ttm_bo_device *bdev); + + /** + * struct ttm_bo_driver member invalidate_caches + * + * @bdev: the buffer object device. + * @flags: new placement of the rebound buffer object. + * + * A previosly evicted buffer has been rebound in a + * potentially new location. Tell the driver that it might + * consider invalidating read (texture) caches on the next command + * submission as a consequence. + */ + + int (*invalidate_caches) (struct ttm_bo_device *bdev, uint32_t flags); + int (*init_mem_type) (struct ttm_bo_device *bdev, uint32_t type, + struct ttm_mem_type_manager *man); + /** + * struct ttm_bo_driver member evict_flags: + * + * @bo: the buffer object to be evicted + * + * Return the bo flags for a buffer which is not mapped to the hardware. + * These will be placed in proposed_flags so that when the move is + * finished, they'll end up in bo->mem.flags + */ + + uint32_t(*evict_flags) (struct ttm_buffer_object *bo); + /** + * struct ttm_bo_driver member move: + * + * @bo: the buffer to move + * @evict: whether this motion is evicting the buffer from + * the graphics address space + * @interruptible: Use interruptible sleeps if possible when sleeping. + * @no_wait: whether this should give up and return -EBUSY + * if this move would require sleeping + * @new_mem: the new memory region receiving the buffer + * + * Move a buffer between two memory regions. + */ + int (*move) (struct ttm_buffer_object *bo, + bool evict, bool interruptible, + bool no_wait, struct ttm_mem_reg *new_mem); + + /** + * struct ttm_bo_driver_member verify_access + * + * @bo: Pointer to a buffer object. + * @filp: Pointer to a struct file trying to access the object. + * + * Called from the map / write / read methods to verify that the + * caller is permitted to access the buffer object. + * This member may be set to NULL, which will refuse this kind of + * access for all buffer objects. + * This function should return 0 if access is granted, -EPERM otherwise. + */ + int (*verify_access) (struct ttm_buffer_object *bo, + struct file *filp); + + /** + * In case a driver writer dislikes the TTM fence objects, + * the driver writer can replace those with sync objects of + * his / her own. If it turns out that no driver writer is + * using these. I suggest we remove these hooks and plug in + * fences directly. The bo driver needs the following functionality: + * See the corresponding functions in the fence object API + * documentation. + */ + + bool (*sync_obj_signaled) (void *sync_obj, void *sync_arg); + int (*sync_obj_wait) (void *sync_obj, void *sync_arg, + bool lazy, bool interruptible); + int (*sync_obj_flush) (void *sync_obj, void *sync_arg); + void (*sync_obj_unref) (void **sync_obj); + void *(*sync_obj_ref) (void *sync_obj); +}; + +#define TTM_NUM_MEM_TYPES 8 + +#define TTM_BO_PRIV_FLAG_MOVING 0 /* Buffer object is moving and needs + idling before CPU mapping */ +#define TTM_BO_PRIV_FLAG_MAX 1 +/** + * struct ttm_bo_device - Buffer object driver device-specific data. + * + * @mem_glob: Pointer to a struct ttm_mem_global object for accounting. + * @driver: Pointer to a struct ttm_bo_driver struct setup by the driver. + * @count: Current number of buffer object. + * @pages: Current number of pinned pages. + * @dummy_read_page: Pointer to a dummy page used for mapping requests + * of unpopulated pages. + * @shrink: A shrink callback object used for buffre object swap. + * @ttm_bo_extra_size: Extra size (sizeof(struct ttm_buffer_object) excluded) + * used by a buffer object. This is excluding page arrays and backing pages. + * @ttm_bo_size: This is @ttm_bo_extra_size + sizeof(struct ttm_buffer_object). + * @man: An array of mem_type_managers. + * @addr_space_mm: Range manager for the device address space. + * lru_lock: Spinlock that protects the buffer+device lru lists and + * ddestroy lists. + * @nice_mode: Try nicely to wait for buffer idle when cleaning a manager. + * If a GPU lockup has been detected, this is forced to 0. + * @dev_mapping: A pointer to the struct address_space representing the + * device address space. + * @wq: Work queue structure for the delayed delete workqueue. + * + */ + +struct ttm_bo_device { + + /* + * Constant after bo device init / atomic. + */ + + struct ttm_mem_global *mem_glob; + struct ttm_bo_driver *driver; + struct page *dummy_read_page; + struct ttm_mem_shrink shrink; + + size_t ttm_bo_extra_size; + size_t ttm_bo_size; + + rwlock_t vm_lock; + /* + * Protected by the vm lock. + */ + struct ttm_mem_type_manager man[TTM_NUM_MEM_TYPES]; + struct rb_root addr_space_rb; + struct drm_mm addr_space_mm; + + /* + * Might want to change this to one lock per manager. + */ + spinlock_t lru_lock; + /* + * Protected by the lru lock. + */ + struct list_head ddestroy; + struct list_head swap_lru; + + /* + * Protected by load / firstopen / lastclose /unload sync. + */ + + bool nice_mode; + struct address_space *dev_mapping; + + /* + * Internal protection. + */ + + struct delayed_work wq; +}; + +/** + * ttm_flag_masked + * + * @old: Pointer to the result and original value. + * @new: New value of bits. + * @mask: Mask of bits to change. + * + * Convenience function to change a number of bits identified by a mask. + */ + +static inline uint32_t +ttm_flag_masked(uint32_t *old, uint32_t new, uint32_t mask) +{ + *old ^= (*old ^ new) & mask; + return *old; +} + +/** + * ttm_tt_create + * + * @bdev: pointer to a struct ttm_bo_device: + * @size: Size of the data needed backing. + * @page_flags: Page flags as identified by TTM_PAGE_FLAG_XX flags. + * @dummy_read_page: See struct ttm_bo_device. + * + * Create a struct ttm_tt to back data with system memory pages. + * No pages are actually allocated. + * Returns: + * NULL: Out of memory. + */ +extern struct ttm_tt *ttm_tt_create(struct ttm_bo_device *bdev, + unsigned long size, + uint32_t page_flags, + struct page *dummy_read_page); + +/** + * ttm_tt_set_user: + * + * @ttm: The struct ttm_tt to populate. + * @tsk: A struct task_struct for which @start is a valid user-space address. + * @start: A valid user-space address. + * @num_pages: Size in pages of the user memory area. + * + * Populate a struct ttm_tt with a user-space memory area after first pinning + * the pages backing it. + * Returns: + * !0: Error. + */ + +extern int ttm_tt_set_user(struct ttm_tt *ttm, + struct task_struct *tsk, + unsigned long start, unsigned long num_pages); + +/** + * ttm_ttm_bind: + * + * @ttm: The struct ttm_tt containing backing pages. + * @bo_mem: The struct ttm_mem_reg identifying the binding location. + * + * Bind the pages of @ttm to an aperture location identified by @bo_mem + */ +extern int ttm_tt_bind(struct ttm_tt *ttm, struct ttm_mem_reg *bo_mem); + +/** + * ttm_ttm_destroy: + * + * @ttm: The struct ttm_tt. + * + * Unbind, unpopulate and destroy a struct ttm_tt. + */ +extern void ttm_tt_destroy(struct ttm_tt *ttm); + +/** + * ttm_ttm_unbind: + * + * @ttm: The struct ttm_tt. + * + * Unbind a struct ttm_tt. + */ +extern void ttm_tt_unbind(struct ttm_tt *ttm); + +/** + * ttm_ttm_destroy: + * + * @ttm: The struct ttm_tt. + * @index: Index of the desired page. + * + * Return a pointer to the struct page backing @ttm at page + * index @index. If the page is unpopulated, one will be allocated to + * populate that index. + * + * Returns: + * NULL on OOM. + */ +extern struct page *ttm_tt_get_page(struct ttm_tt *ttm, int index); + +/** + * ttm_tt_cache_flush: + * + * @pages: An array of pointers to struct page:s to flush. + * @num_pages: Number of pages to flush. + * + * Flush the data of the indicated pages from the cpu caches. + * This is used when changing caching attributes of the pages from + * cache-coherent. + */ +extern void ttm_tt_cache_flush(struct page *pages[], unsigned long num_pages); + +/** + * ttm_tt_set_placement_caching: + * + * @ttm A struct ttm_tt the backing pages of which will change caching policy. + * @placement: Flag indicating the desired caching policy. + * + * This function will change caching policy of any default kernel mappings of + * the pages backing @ttm. If changing from cached to uncached or + * write-combined, + * all CPU caches will first be flushed to make sure the data of the pages + * hit RAM. This function may be very costly as it involves global TLB + * and cache flushes and potential page splitting / combining. + */ +extern int ttm_tt_set_placement_caching(struct ttm_tt *ttm, uint32_t placement); +extern int ttm_tt_swapout(struct ttm_tt *ttm, + struct file *persistant_swap_storage); + +/* + * ttm_bo.c + */ + +/** + * ttm_mem_reg_is_pci + * + * @bdev: Pointer to a struct ttm_bo_device. + * @mem: A valid struct ttm_mem_reg. + * + * Returns true if the memory described by @mem is PCI memory, + * false otherwise. + */ +extern bool ttm_mem_reg_is_pci(struct ttm_bo_device *bdev, + struct ttm_mem_reg *mem); + +/** + * ttm_bo_mem_space + * + * @bo: Pointer to a struct ttm_buffer_object. the data of which + * we want to allocate space for. + * @proposed_placement: Proposed new placement for the buffer object. + * @mem: A struct ttm_mem_reg. + * @interruptible: Sleep interruptible when sliping. + * @no_wait: Don't sleep waiting for space to become available. + * + * Allocate memory space for the buffer object pointed to by @bo, using + * the placement flags in @mem, potentially evicting other idle buffer objects. + * This function may sleep while waiting for space to become available. + * Returns: + * -EBUSY: No space available (only if no_wait == 1). + * -ENOMEM: Could not allocate memory for the buffer object, either due to + * fragmentation or concurrent allocators. + * -ERESTART: An interruptible sleep was interrupted by a signal. + */ +extern int ttm_bo_mem_space(struct ttm_buffer_object *bo, + uint32_t proposed_placement, + struct ttm_mem_reg *mem, + bool interruptible, bool no_wait); +/** + * ttm_bo_wait_for_cpu + * + * @bo: Pointer to a struct ttm_buffer_object. + * @no_wait: Don't sleep while waiting. + * + * Wait until a buffer object is no longer sync'ed for CPU access. + * Returns: + * -EBUSY: Buffer object was sync'ed for CPU access. (only if no_wait == 1). + * -ERESTART: An interruptible sleep was interrupted by a signal. + */ + +extern int ttm_bo_wait_cpu(struct ttm_buffer_object *bo, bool no_wait); + +/** + * ttm_bo_pci_offset - Get the PCI offset for the buffer object memory. + * + * @bo Pointer to a struct ttm_buffer_object. + * @bus_base On return the base of the PCI region + * @bus_offset On return the byte offset into the PCI region + * @bus_size On return the byte size of the buffer object or zero if + * the buffer object memory is not accessible through a PCI region. + * + * Returns: + * -EINVAL if the buffer object is currently not mappable. + * 0 otherwise. + */ + +extern int ttm_bo_pci_offset(struct ttm_bo_device *bdev, + struct ttm_mem_reg *mem, + unsigned long *bus_base, + unsigned long *bus_offset, + unsigned long *bus_size); + +extern int ttm_bo_device_release(struct ttm_bo_device *bdev); + +/** + * ttm_bo_device_init + * + * @bdev: A pointer to a struct ttm_bo_device to initialize. + * @mem_global: A pointer to an initialized struct ttm_mem_global. + * @driver: A pointer to a struct ttm_bo_driver set up by the caller. + * @file_page_offset: Offset into the device address space that is available + * for buffer data. This ensures compatibility with other users of the + * address space. + * + * Initializes a struct ttm_bo_device: + * Returns: + * !0: Failure. + */ +extern int ttm_bo_device_init(struct ttm_bo_device *bdev, + struct ttm_mem_global *mem_glob, + struct ttm_bo_driver *driver, + uint64_t file_page_offset); + +/** + * ttm_bo_reserve: + * + * @bo: A pointer to a struct ttm_buffer_object. + * @interruptible: Sleep interruptible if waiting. + * @no_wait: Don't sleep while trying to reserve, rather return -EBUSY. + * @use_sequence: If @bo is already reserved, Only sleep waiting for + * it to become unreserved if @sequence < (@bo)->sequence. + * + * Locks a buffer object for validation. (Or prevents other processes from + * locking it for validation) and removes it from lru lists, while taking + * a number of measures to prevent deadlocks. + * + * Deadlocks may occur when two processes try to reserve multiple buffers in + * different order, either by will or as a result of a buffer being evicted + * to make room for a buffer already reserved. (Buffers are reserved before + * they are evicted). The following algorithm prevents such deadlocks from + * occuring: + * 1) Buffers are reserved with the lru spinlock held. Upon successful + * reservation they are removed from the lru list. This stops a reserved buffer + * from being evicted. However the lru spinlock is released between the time + * a buffer is selected for eviction and the time it is reserved. + * Therefore a check is made when a buffer is reserved for eviction, that it + * is still the first buffer in the lru list, before it is removed from the + * list. @check_lru == 1 forces this check. If it fails, the function returns + * -EINVAL, and the caller should then choose a new buffer to evict and repeat + * the procedure. + * 2) Processes attempting to reserve multiple buffers other than for eviction, + * (typically execbuf), should first obtain a unique 32-bit + * validation sequence number, + * and call this function with @use_sequence == 1 and @sequence == the unique + * sequence number. If upon call of this function, the buffer object is already + * reserved, the validation sequence is checked against the validation + * sequence of the process currently reserving the buffer, + * and if the current validation sequence is greater than that of the process + * holding the reservation, the function returns -EAGAIN. Otherwise it sleeps + * waiting for the buffer to become unreserved, after which it retries + * reserving. + * The caller should, when receiving an -EAGAIN error + * release all its buffer reservations, wait for @bo to become unreserved, and + * then rerun the validation with the same validation sequence. This procedure + * will always guarantee that the process with the lowest validation sequence + * will eventually succeed, preventing both deadlocks and starvation. + * + * Returns: + * -EAGAIN: The reservation may cause a deadlock. + * Release all buffer reservations, wait for @bo to become unreserved and + * try again. (only if use_sequence == 1). + * -ERESTART: A wait for the buffer to become unreserved was interrupted by + * a signal. Release all buffer reservations and return to user-space. + */ +extern int ttm_bo_reserve(struct ttm_buffer_object *bo, + bool interruptible, + bool no_wait, bool use_sequence, uint32_t sequence); + +/** + * ttm_bo_unreserve + * + * @bo: A pointer to a struct ttm_buffer_object. + * + * Unreserve a previous reservation of @bo. + */ +extern void ttm_bo_unreserve(struct ttm_buffer_object *bo); + +/** + * ttm_bo_wait_unreserved + * + * @bo: A pointer to a struct ttm_buffer_object. + * + * Wait for a struct ttm_buffer_object to become unreserved. + * This is typically used in the execbuf code to relax cpu-usage when + * a potential deadlock condition backoff. + */ +extern int ttm_bo_wait_unreserved(struct ttm_buffer_object *bo, + bool interruptible); + +/** + * ttm_bo_block_reservation + * + * @bo: A pointer to a struct ttm_buffer_object. + * @interruptible: Use interruptible sleep when waiting. + * @no_wait: Don't sleep, but rather return -EBUSY. + * + * Block reservation for validation by simply reserving the buffer. + * This is intended for single buffer use only without eviction, + * and thus needs no deadlock protection. + * + * Returns: + * -EBUSY: If no_wait == 1 and the buffer is already reserved. + * -ERESTART: If interruptible == 1 and the process received a signal + * while sleeping. + */ +extern int ttm_bo_block_reservation(struct ttm_buffer_object *bo, + bool interruptible, bool no_wait); + +/** + * ttm_bo_unblock_reservation + * + * @bo: A pointer to a struct ttm_buffer_object. + * + * Unblocks reservation leaving lru lists untouched. + */ +extern void ttm_bo_unblock_reservation(struct ttm_buffer_object *bo); + +/* + * ttm_bo_util.c + */ + +/** + * ttm_bo_move_ttm + * + * @bo: A pointer to a struct ttm_buffer_object. + * @evict: 1: This is an eviction. Don't try to pipeline. + * @no_wait: Never sleep, but rather return with -EBUSY. + * @new_mem: struct ttm_mem_reg indicating where to move. + * + * Optimized move function for a buffer object with both old and + * new placement backed by a TTM. The function will, if successful, + * free any old aperture space, and set (@new_mem)->mm_node to NULL, + * and update the (@bo)->mem placement flags. If unsuccessful, the old + * data remains untouched, and it's up to the caller to free the + * memory space indicated by @new_mem. + * Returns: + * !0: Failure. + */ + +extern int ttm_bo_move_ttm(struct ttm_buffer_object *bo, + bool evict, bool no_wait, + struct ttm_mem_reg *new_mem); + +/** + * ttm_bo_move_memcpy + * + * @bo: A pointer to a struct ttm_buffer_object. + * @evict: 1: This is an eviction. Don't try to pipeline. + * @no_wait: Never sleep, but rather return with -EBUSY. + * @new_mem: struct ttm_mem_reg indicating where to move. + * + * Fallback move function for a mappable buffer object in mappable memory. + * The function will, if successful, + * free any old aperture space, and set (@new_mem)->mm_node to NULL, + * and update the (@bo)->mem placement flags. If unsuccessful, the old + * data remains untouched, and it's up to the caller to free the + * memory space indicated by @new_mem. + * Returns: + * !0: Failure. + */ + +extern int ttm_bo_move_memcpy(struct ttm_buffer_object *bo, + bool evict, + bool no_wait, struct ttm_mem_reg *new_mem); + +/** + * ttm_bo_free_old_node + * + * @bo: A pointer to a struct ttm_buffer_object. + * + * Utility function to free an old placement after a successful move. + */ +extern void ttm_bo_free_old_node(struct ttm_buffer_object *bo); + +/** + * ttm_bo_move_accel_cleanup. + * + * @bo: A pointer to a struct ttm_buffer_object. + * @sync_obj: A sync object that signals when moving is complete. + * @sync_obj_arg: An argument to pass to the sync object idle / wait + * functions. + * @evict: This is an evict move. Don't return until the buffer is idle. + * @no_wait: Never sleep, but rather return with -EBUSY. + * @new_mem: struct ttm_mem_reg indicating where to move. + * + * Accelerated move function to be called when an accelerated move + * has been scheduled. The function will create a new temporary buffer object + * representing the old placement, and put the sync object on both buffer + * objects. After that the newly created buffer object is unref'd to be + * destroyed when the move is complete. This will help pipeline + * buffer moves. + */ + +extern int ttm_bo_move_accel_cleanup(struct ttm_buffer_object *bo, + void *sync_obj, + void *sync_obj_arg, + bool evict, bool no_wait, + struct ttm_mem_reg *new_mem); +/** + * ttm_io_prot + * + * @c_state: Caching state. + * @tmp: Page protection flag for a normal, cached mapping. + * + * Utility function that returns the pgprot_t that should be used for + * setting up a PTE with the caching model indicated by @c_state. + */ +extern pgprot_t ttm_io_prot(enum ttm_caching_state c_state, pgprot_t tmp); + +#if (defined(CONFIG_AGP) || (defined(CONFIG_AGP_MODULE) && defined(MODULE))) +#define TTM_HAS_AGP +#include + +/** + * ttm_agp_backend_init + * + * @bdev: Pointer to a struct ttm_bo_device. + * @bridge: The agp bridge this device is sitting on. + * + * Create a TTM backend that uses the indicated AGP bridge as an aperture + * for TT memory. This function uses the linux agpgart interface to + * bind and unbind memory backing a ttm_tt. + */ +extern struct ttm_backend *ttm_agp_backend_init(struct ttm_bo_device *bdev, + struct agp_bridge_data *bridge); +#endif + +#endif diff --git a/include/drm/ttm/ttm_memory.h b/include/drm/ttm/ttm_memory.h new file mode 100644 index 0000000000000000000000000000000000000000..d8b8f042c4f19d79fa4bcab79592c55e69d294e7 --- /dev/null +++ b/include/drm/ttm/ttm_memory.h @@ -0,0 +1,153 @@ +/************************************************************************** + * + * 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. + * + **************************************************************************/ + +#ifndef TTM_MEMORY_H +#define TTM_MEMORY_H + +#include +#include +#include +#include + +/** + * struct ttm_mem_shrink - callback to shrink TTM memory usage. + * + * @do_shrink: The callback function. + * + * Arguments to the do_shrink functions are intended to be passed using + * inheritance. That is, the argument class derives from struct ttm_mem_srink, + * and can be accessed using container_of(). + */ + +struct ttm_mem_shrink { + int (*do_shrink) (struct ttm_mem_shrink *); +}; + +/** + * struct ttm_mem_global - Global memory accounting structure. + * + * @shrink: A single callback to shrink TTM memory usage. Extend this + * to a linked list to be able to handle multiple callbacks when needed. + * @swap_queue: A workqueue to handle shrinking in low memory situations. We + * need a separate workqueue since it will spend a lot of time waiting + * for the GPU, and this will otherwise block other workqueue tasks(?) + * At this point we use only a single-threaded workqueue. + * @work: The workqueue callback for the shrink queue. + * @queue: Wait queue for processes suspended waiting for memory. + * @lock: Lock to protect the @shrink - and the memory accounting members, + * that is, essentially the whole structure with some exceptions. + * @emer_memory: Lowmem memory limit available for root. + * @max_memory: Lowmem memory limit available for non-root. + * @swap_limit: Lowmem memory limit where the shrink workqueue kicks in. + * @used_memory: Currently used lowmem memory. + * @used_total_memory: Currently used total (lowmem + highmem) memory. + * @total_memory_swap_limit: Total memory limit where the shrink workqueue + * kicks in. + * @max_total_memory: Total memory available to non-root processes. + * @emer_total_memory: Total memory available to root processes. + * + * Note that this structure is not per device. It should be global for all + * graphics devices. + */ + +struct ttm_mem_global { + struct ttm_mem_shrink *shrink; + struct workqueue_struct *swap_queue; + struct work_struct work; + wait_queue_head_t queue; + spinlock_t lock; + uint64_t emer_memory; + uint64_t max_memory; + uint64_t swap_limit; + uint64_t used_memory; + uint64_t used_total_memory; + uint64_t total_memory_swap_limit; + uint64_t max_total_memory; + uint64_t emer_total_memory; +}; + +/** + * ttm_mem_init_shrink - initialize a struct ttm_mem_shrink object + * + * @shrink: The object to initialize. + * @func: The callback function. + */ + +static inline void ttm_mem_init_shrink(struct ttm_mem_shrink *shrink, + int (*func) (struct ttm_mem_shrink *)) +{ + shrink->do_shrink = func; +} + +/** + * ttm_mem_register_shrink - register a struct ttm_mem_shrink object. + * + * @glob: The struct ttm_mem_global object to register with. + * @shrink: An initialized struct ttm_mem_shrink object to register. + * + * Returns: + * -EBUSY: There's already a callback registered. (May change). + */ + +static inline int ttm_mem_register_shrink(struct ttm_mem_global *glob, + struct ttm_mem_shrink *shrink) +{ + spin_lock(&glob->lock); + if (glob->shrink != NULL) { + spin_unlock(&glob->lock); + return -EBUSY; + } + glob->shrink = shrink; + spin_unlock(&glob->lock); + return 0; +} + +/** + * ttm_mem_unregister_shrink - unregister a struct ttm_mem_shrink object. + * + * @glob: The struct ttm_mem_global object to unregister from. + * @shrink: A previously registert struct ttm_mem_shrink object. + * + */ + +static inline void ttm_mem_unregister_shrink(struct ttm_mem_global *glob, + struct ttm_mem_shrink *shrink) +{ + spin_lock(&glob->lock); + BUG_ON(glob->shrink != shrink); + glob->shrink = NULL; + spin_unlock(&glob->lock); +} + +extern int ttm_mem_global_init(struct ttm_mem_global *glob); +extern void ttm_mem_global_release(struct ttm_mem_global *glob); +extern int ttm_mem_global_alloc(struct ttm_mem_global *glob, uint64_t memory, + bool no_wait, bool interruptible, bool himem); +extern void ttm_mem_global_free(struct ttm_mem_global *glob, + uint64_t amount, bool himem); +extern size_t ttm_round_pot(size_t size); +#endif diff --git a/include/drm/ttm/ttm_module.h b/include/drm/ttm/ttm_module.h new file mode 100644 index 0000000000000000000000000000000000000000..889a4c7958ae7cd02769db1899d1a0b25689587d --- /dev/null +++ b/include/drm/ttm/ttm_module.h @@ -0,0 +1,58 @@ +/************************************************************************** + * + * Copyright 2008-2009 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 + */ + +#ifndef _TTM_MODULE_H_ +#define _TTM_MODULE_H_ + +#include + +#define TTM_PFX "[TTM]" + +enum ttm_global_types { + TTM_GLOBAL_TTM_MEM = 0, + TTM_GLOBAL_TTM_BO, + TTM_GLOBAL_TTM_OBJECT, + TTM_GLOBAL_NUM +}; + +struct ttm_global_reference { + enum ttm_global_types global_type; + size_t size; + void *object; + int (*init) (struct ttm_global_reference *); + void (*release) (struct ttm_global_reference *); +}; + +extern void ttm_global_init(void); +extern void ttm_global_release(void); +extern int ttm_global_item_ref(struct ttm_global_reference *ref); +extern void ttm_global_item_unref(struct ttm_global_reference *ref); + +#endif /* _TTM_MODULE_H_ */ diff --git a/include/drm/ttm/ttm_placement.h b/include/drm/ttm/ttm_placement.h new file mode 100644 index 0000000000000000000000000000000000000000..c84ff153a564258f5e71404bfb50589b313c56f0 --- /dev/null +++ b/include/drm/ttm/ttm_placement.h @@ -0,0 +1,92 @@ +/************************************************************************** + * + * 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 + */ + +#ifndef _TTM_PLACEMENT_H_ +#define _TTM_PLACEMENT_H_ +/* + * Memory regions for data placement. + */ + +#define TTM_PL_SYSTEM 0 +#define TTM_PL_TT 1 +#define TTM_PL_VRAM 2 +#define TTM_PL_PRIV0 3 +#define TTM_PL_PRIV1 4 +#define TTM_PL_PRIV2 5 +#define TTM_PL_PRIV3 6 +#define TTM_PL_PRIV4 7 +#define TTM_PL_PRIV5 8 +#define TTM_PL_SWAPPED 15 + +#define TTM_PL_FLAG_SYSTEM (1 << TTM_PL_SYSTEM) +#define TTM_PL_FLAG_TT (1 << TTM_PL_TT) +#define TTM_PL_FLAG_VRAM (1 << TTM_PL_VRAM) +#define TTM_PL_FLAG_PRIV0 (1 << TTM_PL_PRIV0) +#define TTM_PL_FLAG_PRIV1 (1 << TTM_PL_PRIV1) +#define TTM_PL_FLAG_PRIV2 (1 << TTM_PL_PRIV2) +#define TTM_PL_FLAG_PRIV3 (1 << TTM_PL_PRIV3) +#define TTM_PL_FLAG_PRIV4 (1 << TTM_PL_PRIV4) +#define TTM_PL_FLAG_PRIV5 (1 << TTM_PL_PRIV5) +#define TTM_PL_FLAG_SWAPPED (1 << TTM_PL_SWAPPED) +#define TTM_PL_MASK_MEM 0x0000FFFF + +/* + * Other flags that affects data placement. + * TTM_PL_FLAG_CACHED indicates cache-coherent mappings + * if available. + * TTM_PL_FLAG_SHARED means that another application may + * reference the buffer. + * TTM_PL_FLAG_NO_EVICT means that the buffer may never + * be evicted to make room for other buffers. + */ + +#define TTM_PL_FLAG_CACHED (1 << 16) +#define TTM_PL_FLAG_UNCACHED (1 << 17) +#define TTM_PL_FLAG_WC (1 << 18) +#define TTM_PL_FLAG_SHARED (1 << 20) +#define TTM_PL_FLAG_NO_EVICT (1 << 21) + +#define TTM_PL_MASK_CACHING (TTM_PL_FLAG_CACHED | \ + TTM_PL_FLAG_UNCACHED | \ + TTM_PL_FLAG_WC) + +#define TTM_PL_MASK_MEMTYPE (TTM_PL_MASK_MEM | TTM_PL_MASK_CACHING) + +/* + * Access flags to be used for CPU- and GPU- mappings. + * The idea is that the TTM synchronization mechanism will + * allow concurrent READ access and exclusive write access. + * Currently GPU- and CPU accesses are exclusive. + */ + +#define TTM_ACCESS_READ (1 << 0) +#define TTM_ACCESS_WRITE (1 << 1) + +#endif