diff options
author | Thomas Hellstrom <thellstrom@vmware.com> | 2009-06-10 15:20:19 +0200 |
---|---|---|
committer | Dave Airlie <airlied@redhat.com> | 2009-06-15 09:37:57 +1000 |
commit | ba4e7d973dd09b66912ac4c0856add8b0703a997 (patch) | |
tree | 32a87edb83a427ffd22645c5f77e6cec8be4e719 /drivers/gpu | |
parent | e6c03c5b40314d787f7053f631594d6b1bd609e8 (diff) | |
download | op-kernel-dev-ba4e7d973dd09b66912ac4c0856add8b0703a997.zip op-kernel-dev-ba4e7d973dd09b66912ac4c0856add8b0703a997.tar.gz |
drm: Add the TTM GPU memory manager subsystem.
TTM is a GPU memory manager subsystem designed for use with GPU
devices with various memory types (On-card VRAM, AGP,
PCI apertures etc.). It's essentially a helper library that assists
the DRM driver in creating and managing persistent buffer objects.
TTM manages placement of data and CPU map setup and teardown on
data movement. It can also optionally manage synchronization of
data on a per-buffer-object level.
TTM takes care to provide an always valid virtual user-space address
to a buffer object which makes user-space sub-allocation of
big buffer objects feasible.
TTM uses a fine-grained per buffer-object locking scheme, taking
care to release all relevant locks when waiting for the GPU.
Although this implies some locking overhead, it's probably a big
win for devices with multiple command submission mechanisms, since
the lock contention will be minimal.
TTM can be used with whatever user-space interface the driver
chooses, including GEM. It's used by the upcoming Radeon KMS DRM driver
and is also the GPU memory management core of various new experimental
DRM drivers.
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Diffstat (limited to 'drivers/gpu')
-rw-r--r-- | drivers/gpu/drm/Kconfig | 8 | ||||
-rw-r--r-- | drivers/gpu/drm/Makefile | 2 | ||||
-rw-r--r-- | drivers/gpu/drm/ttm/Makefile | 8 | ||||
-rw-r--r-- | drivers/gpu/drm/ttm/ttm_agp_backend.c | 150 | ||||
-rw-r--r-- | drivers/gpu/drm/ttm/ttm_bo.c | 1698 | ||||
-rw-r--r-- | drivers/gpu/drm/ttm/ttm_bo_util.c | 561 | ||||
-rw-r--r-- | drivers/gpu/drm/ttm/ttm_bo_vm.c | 454 | ||||
-rw-r--r-- | drivers/gpu/drm/ttm/ttm_global.c | 114 | ||||
-rw-r--r-- | drivers/gpu/drm/ttm/ttm_memory.c | 234 | ||||
-rw-r--r-- | drivers/gpu/drm/ttm/ttm_module.c | 50 | ||||
-rw-r--r-- | drivers/gpu/drm/ttm/ttm_tt.c | 635 |
11 files changed, 3913 insertions, 1 deletions
diff --git a/drivers/gpu/drm/Kconfig b/drivers/gpu/drm/Kconfig index f5d46e7..64a57ad 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 4ec5061..4e89ab0 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-$(CONFIG_DRM_TTM) += ttm/ diff --git a/drivers/gpu/drm/ttm/Makefile b/drivers/gpu/drm/ttm/Makefile new file mode 100644 index 0000000..b0a9de7 --- /dev/null +++ b/drivers/gpu/drm/ttm/Makefile @@ -0,0 +1,8 @@ +# +# Makefile for the drm device driver. This driver provides support for the + +ccflags-y := -Iinclude/drm +ttm-y := ttm_agp_backend.o ttm_memory.o ttm_tt.o ttm_bo.o \ + ttm_bo_util.o ttm_bo_vm.o ttm_module.o ttm_global.o + +obj-$(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 0000000..e8f6d22 --- /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 <thellstrom-at-vmware-dot-com> + * Keith Packard. + */ + +#include "ttm/ttm_module.h" +#include "ttm/ttm_bo_driver.h" +#ifdef TTM_HAS_AGP +#include "ttm/ttm_placement.h" +#include <linux/agp_backend.h> +#include <linux/module.h> +#include <linux/io.h> +#include <asm/agp.h> + +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 0000000..1587aeca --- /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 <thellstrom-at-vmware-dot-com> + */ + +#include "ttm/ttm_module.h" +#include "ttm/ttm_bo_driver.h" +#include "ttm/ttm_placement.h" +#include <linux/jiffies.h> +#include <linux/slab.h> +#include <linux/sched.h> +#include <linux/mm.h> +#include <linux/file.h> +#include <linux/module.h> + +#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 0000000..517c845 --- /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 <thellstrom-at-vmware-dot-com> + */ + +#include "ttm/ttm_bo_driver.h" +#include "ttm/ttm_placement.h" +#include <linux/io.h> +#include <linux/highmem.h> +#include <linux/wait.h> +#include <linux/vmalloc.h> +#include <linux/version.h> +#include <linux/module.h> + +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 0000000..27b146c --- /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 <thellstrom-at-vmware-dot-com> + */ + +#include <ttm/ttm_module.h> +#include <ttm/ttm_bo_driver.h> +#include <ttm/ttm_placement.h> +#include <linux/mm.h> +#include <linux/version.h> +#include <linux/rbtree.h> +#include <linux/module.h> +#include <linux/uaccess.h> + +#define TTM_BO_VM_NUM_PREFAULT 16 + +static struct ttm_buffer_object *ttm_bo_vm_lookup_rb(struct ttm_bo_device *bdev, + unsigned long page_start, + unsigned long num_pages) +{ + struct rb_node *cur = bdev->addr_space_rb.rb_node; + unsigned long cur_offset; + struct ttm_buffer_object *bo; + struct ttm_buffer_object *best_bo = NULL; + + while (likely(cur != NULL)) { + bo = rb_entry(cur, struct ttm_buffer_object, vm_rb); + cur_offset = bo->vm_node->start; + if (page_start >= cur_offset) { + cur = cur->rb_right; + best_bo = bo; + if (page_start == cur_offset) + break; + } else + cur = cur->rb_left; + } + + if (unlikely(best_bo == NULL)) + return NULL; + + if (unlikely((best_bo->vm_node->start + best_bo->num_pages) < + (page_start + num_pages))) + return NULL; + + return best_bo; +} + +static int ttm_bo_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf) +{ + struct ttm_buffer_object *bo = (struct ttm_buffer_object *) + vma->vm_private_data; + struct ttm_bo_device *bdev = bo->bdev; + unsigned long 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 0000000..0b14eb1 --- /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 <thellstrom-at-vmware-dot-com> + */ + +#include "ttm/ttm_module.h" +#include <linux/mutex.h> +#include <linux/slab.h> +#include <linux/module.h> + +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 0000000..87323d4 --- /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 <linux/spinlock.h> +#include <linux/sched.h> +#include <linux/wait.h> +#include <linux/mm.h> +#include <linux/module.h> + +#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 0000000..59ce819 --- /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 <thellstrom-at-vmware-dot-com> + * Jerome Glisse + */ +#include <linux/module.h> +#include <ttm/ttm_module.h> + +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 0000000..c27ab3a --- /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 <thellstrom-at-vmware-dot-com> + */ + +#include <linux/version.h> +#include <linux/vmalloc.h> +#include <linux/sched.h> +#include <linux/highmem.h> +#include <linux/pagemap.h> +#include <linux/file.h> +#include <linux/swap.h> +#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; +} |