diff options
Diffstat (limited to 'include/drm/ttm')
| -rw-r--r-- | include/drm/ttm/ttm_bo_api.h | 184 | ||||
| -rw-r--r-- | include/drm/ttm/ttm_bo_driver.h | 636 | ||||
| -rw-r--r-- | include/drm/ttm/ttm_execbuf_util.h | 27 | ||||
| -rw-r--r-- | include/drm/ttm/ttm_lock.h | 4 | ||||
| -rw-r--r-- | include/drm/ttm/ttm_memory.h | 5 | ||||
| -rw-r--r-- | include/drm/ttm/ttm_object.h | 93 | ||||
| -rw-r--r-- | include/drm/ttm/ttm_page_alloc.h | 88 | ||||
| -rw-r--r-- | include/drm/ttm/ttm_placement.h | 3 |
8 files changed, 678 insertions, 362 deletions
diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h index 5afa5b52063..7526c5bf561 100644 --- a/include/drm/ttm/ttm_bo_api.h +++ b/include/drm/ttm/ttm_bo_api.h @@ -31,14 +31,15 @@ #ifndef _TTM_BO_API_H_ #define _TTM_BO_API_H_ -#include "drm_hashtab.h" +#include <drm/drm_hashtab.h> +#include <drm/drm_vma_manager.h> #include <linux/kref.h> #include <linux/list.h> #include <linux/wait.h> #include <linux/mutex.h> #include <linux/mm.h> -#include <linux/rbtree.h> #include <linux/bitmap.h> +#include <linux/reservation.h> struct ttm_bo_device; @@ -50,10 +51,10 @@ struct drm_mm_node; * * @fpfn: first valid page frame number to put the object * @lpfn: last valid page frame number to put the object - * @num_placement: number of prefered placements - * @placement: prefered placements - * @num_busy_placement: number of prefered placements when need to evict buffer - * @busy_placement: prefered placements when need to evict buffer + * @num_placement: number of preferred placements + * @placement: preferred placements + * @num_busy_placement: number of preferred placements when need to evict buffer + * @busy_placement: preferred placements when need to evict buffer * * Structure indicating the placement you request for an object. */ @@ -74,6 +75,8 @@ struct ttm_placement { * @is_iomem: is this io memory ? * @size: size in byte * @offset: offset from the base address + * @io_reserved_vm: The VM system has a refcount in @io_reserved_count + * @io_reserved_count: Refcounting the numbers of callers to ttm_mem_io_reserve * * Structure indicating the bus placement of an object. */ @@ -83,7 +86,8 @@ struct ttm_bus_placement { unsigned long size; unsigned long offset; bool is_iomem; - bool io_reserved; + bool io_reserved_vm; + uint64_t io_reserved_count; }; @@ -119,18 +123,17 @@ struct ttm_mem_reg { * be mmapped by user space. Each of these bos occupy a slot in the * device address space, that can be used for normal vm operations. * - * @ttm_bo_type_user: These are user-space memory areas that are made - * available to the GPU by mapping the buffer pages into the GPU aperture - * space. These buffers cannot be mmaped from the device address space. - * * @ttm_bo_type_kernel: These buffers are like ttm_bo_type_device buffers, * but they cannot be accessed from user-space. For kernel-only use. + * + * @ttm_bo_type_sg: Buffer made from dmabuf sg table shared with another + * driver. */ enum ttm_bo_type { ttm_bo_type_device, - ttm_bo_type_user, - ttm_bo_type_kernel + ttm_bo_type_kernel, + ttm_bo_type_sg }; struct ttm_tt; @@ -139,12 +142,9 @@ struct ttm_tt; * struct ttm_buffer_object * * @bdev: Pointer to the buffer object device structure. - * @buffer_start: The virtual user-space start address of ttm_bo_type_user - * buffers. * @type: The bo type. * @destroy: Destruction function. If NULL, kfree is used. * @num_pages: Actual number of pages. - * @addr_space_offset: Address space offset. * @acc_size: Accounted size for this object. * @kref: Reference count of this buffer object. When this refcount reaches * zero, the object is put on the delayed delete list. @@ -153,32 +153,23 @@ struct ttm_tt; * Lru lists may keep one refcount, the delayed delete list, and kref != 0 * keeps one refcount. When this refcount reaches zero, * the object is destroyed. - * @event_queue: Queue for processes waiting on buffer object status change. - * @lock: spinlock protecting mostly synchronization members. * @mem: structure describing current placement. - * @persistant_swap_storage: Usually the swap storage is deleted for buffers + * @persistent_swap_storage: Usually the swap storage is deleted for buffers * pinned in physical memory. If this behaviour is not desired, this member - * holds a pointer to a persistant shmem object. + * holds a pointer to a persistent shmem object. * @ttm: TTM structure holding system pages. * @evicted: Whether the object was evicted without user-space knowing. * @cpu_writes: For synchronization. Number of cpu writers. * @lru: List head for the lru list. * @ddestroy: List head for the delayed destroy list. * @swap: List head for swap LRU list. - * @val_seq: Sequence of the validation holding the @reserved lock. - * Used to avoid starvation when many processes compete to validate the - * buffer. This member is protected by the bo_device::lru_lock. - * @seq_valid: The value of @val_seq is valid. This value is protected by - * the bo_device::lru_lock. - * @reserved: Deadlock-free lock used for synchronization state transitions. - * @sync_obj_arg: Opaque argument to synchronization object function. * @sync_obj: Pointer to a synchronization object. * @priv_flags: Flags describing buffer object internal state. - * @vm_rb: Rb node for the vm rb tree. - * @vm_node: Address space manager node. + * @vma_node: Address space manager node. * @offset: The current GPU offset, which can have different meanings * depending on the memory type. For SYSTEM type memory, it should be 0. * @cur_placement: Hint of current placement. + * @wu_mutex: Wait unreserved mutex. * * Base class for TTM buffer object, that deals with data placement and CPU * mappings. GPU mappings are really up to the driver, but for simpler GPUs @@ -199,11 +190,9 @@ struct ttm_buffer_object { struct ttm_bo_global *glob; struct ttm_bo_device *bdev; - unsigned long buffer_start; enum ttm_bo_type type; void (*destroy) (struct ttm_buffer_object *); unsigned long num_pages; - uint64_t addr_space_offset; size_t acc_size; /** @@ -212,15 +201,13 @@ struct ttm_buffer_object { struct kref kref; struct kref list_kref; - wait_queue_head_t event_queue; - spinlock_t lock; /** - * Members protected by the bo::reserved lock. + * Members protected by the bo::resv::reserved lock. */ struct ttm_mem_reg mem; - struct file *persistant_swap_storage; + struct file *persistent_swap_storage; struct ttm_tt *ttm; bool evicted; @@ -237,34 +224,19 @@ struct ttm_buffer_object { struct list_head lru; struct list_head ddestroy; struct list_head swap; - uint32_t val_seq; - bool seq_valid; + struct list_head io_reserve_lru; /** - * Members protected by the bdev::lru_lock - * only when written to. - */ - - atomic_t reserved; - - /** - * Members protected by the bo::lock + * Members protected by struct buffer_object_device::fence_lock * In addition, setting sync_obj to anything else * than NULL requires bo::reserved to be held. This allows for - * checking NULL while reserved but not holding bo::lock. + * checking NULL while reserved but not holding the mentioned lock. */ - void *sync_obj_arg; void *sync_obj; unsigned long priv_flags; - /** - * Members protected by the bdev::vm_lock - */ - - struct rb_node vm_rb; - struct drm_mm_node *vm_node; - + struct drm_vma_offset_node vma_node; /** * Special members that are protected by the reserve lock @@ -274,6 +246,12 @@ struct ttm_buffer_object { unsigned long offset; uint32_t cur_placement; + + struct sg_table *sg; + + struct reservation_object *resv; + struct reservation_object ttm_resv; + struct mutex wu_mutex; }; /** @@ -339,7 +317,6 @@ extern int ttm_bo_wait(struct ttm_buffer_object *bo, bool lazy, * @bo: The buffer object. * @placement: Proposed placement for the buffer object. * @interruptible: Sleep interruptible if sleeping. - * @no_wait_reserve: Return immediately if other buffers are busy. * @no_wait_gpu: Return immediately if the GPU is busy. * * Changes placement and caching policy of the buffer object @@ -352,7 +329,7 @@ extern int ttm_bo_wait(struct ttm_buffer_object *bo, bool lazy, */ extern int ttm_bo_validate(struct ttm_buffer_object *bo, struct ttm_placement *placement, - bool interruptible, bool no_wait_reserve, + bool interruptible, bool no_wait_gpu); /** @@ -364,6 +341,44 @@ extern int ttm_bo_validate(struct ttm_buffer_object *bo, */ extern void ttm_bo_unref(struct ttm_buffer_object **bo); + +/** + * ttm_bo_list_ref_sub + * + * @bo: The buffer object. + * @count: The number of references with which to decrease @bo::list_kref; + * @never_free: The refcount should not reach zero with this operation. + * + * Release @count lru list references to this buffer object. + */ +extern void ttm_bo_list_ref_sub(struct ttm_buffer_object *bo, int count, + bool never_free); + +/** + * ttm_bo_add_to_lru + * + * @bo: The buffer object. + * + * Add this bo to the relevant mem type lru and, if it's backed by + * system pages (ttms) to the swap list. + * This function must be called with struct ttm_bo_global::lru_lock held, and + * is typically called immediately prior to unreserving a bo. + */ +extern void ttm_bo_add_to_lru(struct ttm_buffer_object *bo); + +/** + * ttm_bo_del_from_lru + * + * @bo: The buffer object. + * + * Remove this bo from all lru lists used to lookup and reserve an object. + * This function must be called with struct ttm_bo_global::lru_lock held, + * and is usually called just immediately after the bo has been reserved to + * avoid recursive reservation from lru lists. + */ +extern int ttm_bo_del_from_lru(struct ttm_buffer_object *bo); + + /** * ttm_bo_lock_delayed_workqueue * @@ -388,15 +403,16 @@ extern void ttm_bo_unlock_delayed_workqueue(struct ttm_bo_device *bdev, * @no_wait: Return immediately if buffer is busy. * * Synchronizes a buffer object for CPU RW access. This means - * blocking command submission that affects the buffer and - * waiting for buffer idle. This lock is recursive. + * command submission that affects the buffer will return -EBUSY + * until ttm_bo_synccpu_write_release is called. + * * Returns * -EBUSY if the buffer is busy and no_wait is true. * -ERESTARTSYS if interrupted by a signal. */ - extern int ttm_bo_synccpu_write_grab(struct ttm_buffer_object *bo, bool no_wait); + /** * ttm_bo_synccpu_write_release: * @@ -407,6 +423,22 @@ ttm_bo_synccpu_write_grab(struct ttm_buffer_object *bo, bool no_wait); extern void ttm_bo_synccpu_write_release(struct ttm_buffer_object *bo); /** + * ttm_bo_acc_size + * + * @bdev: Pointer to a ttm_bo_device struct. + * @bo_size: size of the buffer object in byte. + * @struct_size: size of the structure holding buffer object datas + * + * Returns size to account for a buffer object + */ +size_t ttm_bo_acc_size(struct ttm_bo_device *bdev, + unsigned long bo_size, + unsigned struct_size); +size_t ttm_bo_dma_acc_size(struct ttm_bo_device *bdev, + unsigned long bo_size, + unsigned struct_size); + +/** * ttm_bo_init * * @bdev: Pointer to a ttm_bo_device struct. @@ -415,13 +447,11 @@ extern void ttm_bo_synccpu_write_release(struct ttm_buffer_object *bo); * @type: Requested type of buffer object. * @flags: Initial placement flags. * @page_alignment: Data alignment in pages. - * @buffer_start: Virtual address of user space data backing a - * user buffer object. * @interruptible: If needing to sleep to wait for GPU resources, * sleep interruptible. - * @persistant_swap_storage: Usually the swap storage is deleted for buffers + * @persistent_swap_storage: Usually the swap storage is deleted for buffers * pinned in physical memory. If this behaviour is not desired, this member - * holds a pointer to a persistant shmem object. Typically, this would + * holds a pointer to a persistent shmem object. Typically, this would * point to the shmem object backing a GEM object if TTM is used to back a * GEM user interface. * @acc_size: Accounted size for this object. @@ -432,6 +462,10 @@ extern void ttm_bo_synccpu_write_release(struct ttm_buffer_object *bo); * together with the @destroy function, * enables driver-specific objects derived from a ttm_buffer_object. * On successful return, the object kref and list_kref are set to 1. + * If a failure occurs, the function will call the @destroy function, or + * kfree() if @destroy is NULL. Thus, after a failure, dereferencing @bo is + * illegal and will likely cause memory corruption. + * * Returns * -ENOMEM: Out of memory. * -EINVAL: Invalid placement flags. @@ -444,27 +478,25 @@ extern int ttm_bo_init(struct ttm_bo_device *bdev, enum ttm_bo_type type, struct ttm_placement *placement, uint32_t page_alignment, - unsigned long buffer_start, bool interrubtible, - struct file *persistant_swap_storage, + struct file *persistent_swap_storage, size_t acc_size, + struct sg_table *sg, void (*destroy) (struct ttm_buffer_object *)); + /** - * ttm_bo_synccpu_object_init + * ttm_bo_create * * @bdev: Pointer to a ttm_bo_device struct. - * @bo: Pointer to a ttm_buffer_object to be initialized. * @size: Requested size of buffer object. * @type: Requested type of buffer object. - * @flags: Initial placement flags. + * @placement: Initial placement. * @page_alignment: Data alignment in pages. - * @buffer_start: Virtual address of user space data backing a - * user buffer object. * @interruptible: If needing to sleep while waiting for GPU resources, * sleep interruptible. - * @persistant_swap_storage: Usually the swap storage is deleted for buffers + * @persistent_swap_storage: Usually the swap storage is deleted for buffers * pinned in physical memory. If this behaviour is not desired, this member - * holds a pointer to a persistant shmem object. Typically, this would + * holds a pointer to a persistent shmem object. Typically, this would * point to the shmem object backing a GEM object if TTM is used to back a * GEM user interface. * @p_bo: On successful completion *p_bo points to the created object. @@ -482,9 +514,8 @@ extern int ttm_bo_create(struct ttm_bo_device *bdev, enum ttm_bo_type type, struct ttm_placement *placement, uint32_t page_alignment, - unsigned long buffer_start, bool interruptible, - struct file *persistant_swap_storage, + struct file *persistent_swap_storage, struct ttm_buffer_object **p_bo); /** @@ -618,9 +649,6 @@ extern int ttm_bo_kmap(struct ttm_buffer_object *bo, unsigned long start_page, extern void ttm_bo_kunmap(struct ttm_bo_kmap_obj *map); -#if 0 -#endif - /** * ttm_fbdev_mmap - mmap fbdev memory backed by a ttm buffer object. * @@ -675,5 +703,5 @@ extern ssize_t ttm_bo_io(struct ttm_bo_device *bdev, struct file *filp, size_t count, loff_t *f_pos, bool write); extern void ttm_bo_swapout_all(struct ttm_bo_device *bdev); - +extern int ttm_bo_wait_unreserved(struct ttm_buffer_object *bo); #endif diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h index d01b4ddbdc5..a5183da3ef9 100644 --- a/include/drm/ttm/ttm_bo_driver.h +++ b/include/drm/ttm/ttm_bo_driver.h @@ -30,99 +30,59 @@ #ifndef _TTM_BO_DRIVER_H_ #define _TTM_BO_DRIVER_H_ -#include "ttm/ttm_bo_api.h" -#include "ttm/ttm_memory.h" -#include "ttm/ttm_module.h" -#include "drm_mm.h" -#include "drm_global.h" -#include "linux/workqueue.h" -#include "linux/fs.h" -#include "linux/spinlock.h" - -struct ttm_backend; +#include <ttm/ttm_bo_api.h> +#include <ttm/ttm_memory.h> +#include <ttm/ttm_module.h> +#include <ttm/ttm_placement.h> +#include <drm/drm_mm.h> +#include <drm/drm_global.h> +#include <drm/drm_vma_manager.h> +#include <linux/workqueue.h> +#include <linux/fs.h> +#include <linux/spinlock.h> +#include <linux/reservation.h> struct ttm_backend_func { /** - * struct ttm_backend_func member populate - * - * @backend: Pointer to a struct ttm_backend. - * @num_pages: Number of pages to populate. - * @pages: Array of pointers to ttm pages. - * @dummy_read_page: Page to be used instead of NULL pages in the - * array @pages. - * - * Populate the backend with ttm pages. Depending on the backend, - * it may or may not copy the @pages array. - */ - int (*populate) (struct ttm_backend *backend, - unsigned long num_pages, struct page **pages, - struct page *dummy_read_page); - /** - * struct ttm_backend_func member clear - * - * @backend: Pointer to a struct ttm_backend. - * - * This is an "unpopulate" function. Release all resources - * allocated with populate. - */ - void (*clear) (struct ttm_backend *backend); - - /** * struct ttm_backend_func member bind * - * @backend: Pointer to a struct ttm_backend. + * @ttm: Pointer to a struct ttm_tt. * @bo_mem: Pointer to a struct ttm_mem_reg describing the * memory type and location for binding. * * Bind the backend pages into the aperture in the location * indicated by @bo_mem. This function should be able to handle - * differences between aperture- and system page sizes. + * differences between aperture and system page sizes. */ - int (*bind) (struct ttm_backend *backend, struct ttm_mem_reg *bo_mem); + int (*bind) (struct ttm_tt *ttm, struct ttm_mem_reg *bo_mem); /** * struct ttm_backend_func member unbind * - * @backend: Pointer to a struct ttm_backend. + * @ttm: Pointer to a struct ttm_tt. * * Unbind previously bound backend pages. This function should be - * able to handle differences between aperture- and system page sizes. + * able to handle differences between aperture and system page sizes. */ - int (*unbind) (struct ttm_backend *backend); + int (*unbind) (struct ttm_tt *ttm); /** * struct ttm_backend_func member destroy * - * @backend: Pointer to a struct ttm_backend. + * @ttm: Pointer to a struct ttm_tt. * - * Destroy the backend. + * Destroy the backend. This will be call back from ttm_tt_destroy so + * don't call ttm_tt_destroy from the callback or infinite loop. */ - void (*destroy) (struct ttm_backend *backend); + void (*destroy) (struct ttm_tt *ttm); }; -/** - * struct ttm_backend - * - * @bdev: Pointer to a struct ttm_bo_device. - * @flags: For driver use. - * @func: Pointer to a struct ttm_backend_func that describes - * the backend methods. - * - */ - -struct ttm_backend { - struct ttm_bo_device *bdev; - uint32_t flags; - struct ttm_backend_func *func; -}; - -#define TTM_PAGE_FLAG_USER (1 << 1) -#define TTM_PAGE_FLAG_USER_DIRTY (1 << 2) #define TTM_PAGE_FLAG_WRITE (1 << 3) #define TTM_PAGE_FLAG_SWAPPED (1 << 4) -#define TTM_PAGE_FLAG_PERSISTANT_SWAP (1 << 5) +#define TTM_PAGE_FLAG_PERSISTENT_SWAP (1 << 5) #define TTM_PAGE_FLAG_ZERO_ALLOC (1 << 6) #define TTM_PAGE_FLAG_DMA32 (1 << 7) +#define TTM_PAGE_FLAG_SG (1 << 8) enum ttm_caching_state { tt_uncached, @@ -133,19 +93,15 @@ enum ttm_caching_state { /** * struct ttm_tt * + * @bdev: Pointer to a struct ttm_bo_device. + * @func: Pointer to a struct ttm_backend_func that describes + * the backend methods. * @dummy_read_page: Page to map where the ttm_tt page array contains a NULL * pointer. * @pages: Array of pages backing the data. - * @first_himem_page: Himem pages are put last in the page array, which - * enables us to run caching attribute changes on only the first part - * of the page array containing lomem pages. This is the index of the - * first himem page. - * @last_lomem_page: Index of the last lomem page in the page array. * @num_pages: Number of pages in the page array. * @bdev: Pointer to the current struct ttm_bo_device. * @be: Pointer to the ttm backend. - * @tsk: The task for user ttm. - * @start: virtual address for user ttm. * @swap_storage: Pointer to shmem struct file for swap storage. * @caching_state: The current caching state of the pages. * @state: The current binding state of the pages. @@ -156,16 +112,14 @@ enum ttm_caching_state { */ struct ttm_tt { + struct ttm_bo_device *bdev; + struct ttm_backend_func *func; struct page *dummy_read_page; struct page **pages; - long first_himem_page; - long last_lomem_page; uint32_t page_flags; unsigned long num_pages; + struct sg_table *sg; /* for SG objects via dma-buf */ struct ttm_bo_global *glob; - struct ttm_backend *be; - struct task_struct *tsk; - unsigned long start; struct file *swap_storage; enum ttm_caching_state caching_state; enum { @@ -175,10 +129,111 @@ struct ttm_tt { } state; }; +/** + * struct ttm_dma_tt + * + * @ttm: Base ttm_tt struct. + * @dma_address: The DMA (bus) addresses of the pages + * @pages_list: used by some page allocation backend + * + * This is a structure holding the pages, caching- and aperture binding + * status for a buffer object that isn't backed by fixed (VRAM / AGP) + * memory. + */ +struct ttm_dma_tt { + struct ttm_tt ttm; + dma_addr_t *dma_address; + struct list_head pages_list; +}; + #define TTM_MEMTYPE_FLAG_FIXED (1 << 0) /* Fixed (on-card) PCI memory */ #define TTM_MEMTYPE_FLAG_MAPPABLE (1 << 1) /* Memory mappable */ #define TTM_MEMTYPE_FLAG_CMA (1 << 3) /* Can't map aperture */ +struct ttm_mem_type_manager; + +struct ttm_mem_type_manager_func { + /** + * struct ttm_mem_type_manager member init + * + * @man: Pointer to a memory type manager. + * @p_size: Implementation dependent, but typically the size of the + * range to be managed in pages. + * + * Called to initialize a private range manager. The function is + * expected to initialize the man::priv member. + * Returns 0 on success, negative error code on failure. + */ + int (*init)(struct ttm_mem_type_manager *man, unsigned long p_size); + + /** + * struct ttm_mem_type_manager member takedown + * + * @man: Pointer to a memory type manager. + * + * Called to undo the setup done in init. All allocated resources + * should be freed. + */ + int (*takedown)(struct ttm_mem_type_manager *man); + + /** + * struct ttm_mem_type_manager member get_node + * + * @man: Pointer to a memory type manager. + * @bo: Pointer to the buffer object we're allocating space for. + * @placement: Placement details. + * @mem: Pointer to a struct ttm_mem_reg to be filled in. + * + * This function should allocate space in the memory type managed + * by @man. Placement details if + * applicable are given by @placement. If successful, + * @mem::mm_node should be set to a non-null value, and + * @mem::start should be set to a value identifying the beginning + * of the range allocated, and the function should return zero. + * If the memory region accommodate the buffer object, @mem::mm_node + * should be set to NULL, and the function should return 0. + * If a system error occurred, preventing the request to be fulfilled, + * the function should return a negative error code. + * + * Note that @mem::mm_node will only be dereferenced by + * struct ttm_mem_type_manager functions and optionally by the driver, + * which has knowledge of the underlying type. + * + * This function may not be called from within atomic context, so + * an implementation can and must use either a mutex or a spinlock to + * protect any data structures managing the space. + */ + int (*get_node)(struct ttm_mem_type_manager *man, + struct ttm_buffer_object *bo, + struct ttm_placement *placement, + struct ttm_mem_reg *mem); + + /** + * struct ttm_mem_type_manager member put_node + * + * @man: Pointer to a memory type manager. + * @mem: Pointer to a struct ttm_mem_reg to be filled in. + * + * This function frees memory type resources previously allocated + * and that are identified by @mem::mm_node and @mem::start. May not + * be called from within atomic context. + */ + void (*put_node)(struct ttm_mem_type_manager *man, + struct ttm_mem_reg *mem); + + /** + * struct ttm_mem_type_manager member debug + * + * @man: Pointer to a memory type manager. + * @prefix: Prefix to be used in printout to identify the caller. + * + * This function is called to print out the state of the memory + * type manager to aid debugging of out-of-memory conditions. + * It may not be called from within atomic context. + */ + void (*debug)(struct ttm_mem_type_manager *man, const char *prefix); +}; + /** * struct ttm_mem_type_manager * @@ -193,29 +248,21 @@ struct ttm_tt { * as defined in ttm_placement_common.h * @default_caching: The default caching policy used for a buffer object * placed in this memory type if the user doesn't provide one. - * @manager: The range manager used for this memory type. FIXME: If the aperture - * has a page size different from the underlying system, the granularity - * of this manager should take care of this. But the range allocating code - * in ttm_bo.c needs to be modified for this. + * @func: structure pointer implementing the range manager. See above + * @priv: Driver private closure for @func. + * @io_reserve_mutex: Mutex optionally protecting shared io_reserve structures + * @use_io_reserve_lru: Use an lru list to try to unreserve io_mem_regions + * reserved by the TTM vm system. + * @io_reserve_lru: Optional lru list for unreserving io mem regions. + * @io_reserve_fastpath: Only use bdev::driver::io_mem_reserve to obtain + * static information. bdev::driver::io_mem_free is never used. * @lru: The lru list for this memory type. * * This structure is used to identify and manage memory types for a device. * It's set up by the ttm_bo_driver::init_mem_type method. */ -struct ttm_mem_type_manager; -struct ttm_mem_type_manager_func { - int (*init)(struct ttm_mem_type_manager *man, unsigned long p_size); - int (*takedown)(struct ttm_mem_type_manager *man); - int (*get_node)(struct ttm_mem_type_manager *man, - struct ttm_buffer_object *bo, - struct ttm_placement *placement, - struct ttm_mem_reg *mem); - void (*put_node)(struct ttm_mem_type_manager *man, - struct ttm_mem_reg *mem); - void (*debug)(struct ttm_mem_type_manager *man, const char *prefix); -}; struct ttm_mem_type_manager { struct ttm_bo_device *bdev; @@ -231,14 +278,22 @@ struct ttm_mem_type_manager { uint64_t size; uint32_t available_caching; uint32_t default_caching; + const struct ttm_mem_type_manager_func *func; + void *priv; + struct mutex io_reserve_mutex; + bool use_io_reserve_lru; + bool io_reserve_fastpath; /* - * Protected by the bdev->lru_lock. - * TODO: Consider one lru_lock per ttm_mem_type_manager. - * Plays ill with list removal, though. + * Protected by @io_reserve_mutex: */ - const struct ttm_mem_type_manager_func *func; - void *priv; + + struct list_head io_reserve_lru; + + /* + * Protected by the global->lru_lock. + */ + struct list_head lru; }; @@ -263,15 +318,42 @@ struct ttm_mem_type_manager { struct ttm_bo_driver { /** - * struct ttm_bo_driver member create_ttm_backend_entry + * ttm_tt_create + * + * @bdev: pointer to a struct ttm_bo_device: + * @size: Size of the data needed backing. + * @page_flags: Page flags as identified by TTM_PAGE_FLAG_XX flags. + * @dummy_read_page: See struct ttm_bo_device. + * + * Create a struct ttm_tt to back data with system memory pages. + * No pages are actually allocated. + * Returns: + * NULL: Out of memory. + */ + struct ttm_tt *(*ttm_tt_create)(struct ttm_bo_device *bdev, + unsigned long size, + uint32_t page_flags, + struct page *dummy_read_page); + + /** + * ttm_tt_populate * - * @bdev: The buffer object device. + * @ttm: The struct ttm_tt to contain the backing pages. * - * Create a driver specific struct ttm_backend. + * Allocate all backing pages + * Returns: + * -ENOMEM: Out of memory. */ + int (*ttm_tt_populate)(struct ttm_tt *ttm); - struct ttm_backend *(*create_ttm_backend_entry) - (struct ttm_bo_device *bdev); + /** + * ttm_tt_unpopulate + * + * @ttm: The struct ttm_tt to contain the backing pages. + * + * Free all backing page + */ + void (*ttm_tt_unpopulate)(struct ttm_tt *ttm); /** * struct ttm_bo_driver member invalidate_caches @@ -315,7 +397,7 @@ struct ttm_bo_driver { */ int (*move) (struct ttm_buffer_object *bo, bool evict, bool interruptible, - bool no_wait_reserve, bool no_wait_gpu, + bool no_wait_gpu, struct ttm_mem_reg *new_mem); /** @@ -343,10 +425,10 @@ struct ttm_bo_driver { * documentation. */ - bool (*sync_obj_signaled) (void *sync_obj, void *sync_arg); - int (*sync_obj_wait) (void *sync_obj, void *sync_arg, + bool (*sync_obj_signaled) (void *sync_obj); + int (*sync_obj_wait) (void *sync_obj, bool lazy, bool interruptible); - int (*sync_obj_flush) (void *sync_obj, void *sync_arg); + int (*sync_obj_flush) (void *sync_obj); void (*sync_obj_unref) (void **sync_obj); void *(*sync_obj_ref) (void *sync_obj); @@ -389,9 +471,6 @@ struct ttm_bo_global_ref { * @dummy_read_page: Pointer to a dummy page used for mapping requests * of unpopulated pages. * @shrink: A shrink callback object used for buffer object swap. - * @ttm_bo_extra_size: Extra size (sizeof(struct ttm_buffer_object) excluded) - * used by a buffer object. This is excluding page arrays and backing pages. - * @ttm_bo_size: This is @ttm_bo_extra_size + sizeof(struct ttm_buffer_object). * @device_list_mutex: Mutex protecting the device list. * This mutex is held while traversing the device list for pm options. * @lru_lock: Spinlock protecting the bo subsystem lru lists. @@ -409,8 +488,6 @@ struct ttm_bo_global { struct ttm_mem_global *mem_glob; struct page *dummy_read_page; struct ttm_mem_shrink shrink; - size_t ttm_bo_extra_size; - size_t ttm_bo_size; struct mutex device_list_mutex; spinlock_t lru_lock; @@ -441,11 +518,12 @@ struct ttm_bo_global { * * @driver: Pointer to a struct ttm_bo_driver struct setup by the driver. * @man: An array of mem_type_managers. - * @addr_space_mm: Range manager for the device address space. + * @fence_lock: Protects the synchronizing members on *all* bos belonging + * to this device. + * @vma_manager: Address space manager * lru_lock: Spinlock that protects the buffer+device lru lists and * ddestroy lists. - * @nice_mode: Try nicely to wait for buffer idle when cleaning a manager. - * If a GPU lockup has been detected, this is forced to 0. + * @val_seq: Current validation sequence. * @dev_mapping: A pointer to the struct address_space representing the * device address space. * @wq: Work queue structure for the delayed delete workqueue. @@ -460,24 +538,24 @@ struct ttm_bo_device { struct list_head device_list; struct ttm_bo_global *glob; struct ttm_bo_driver *driver; - rwlock_t vm_lock; struct ttm_mem_type_manager man[TTM_NUM_MEM_TYPES]; + spinlock_t fence_lock; + /* - * Protected by the vm lock. + * Protected by internal locks. */ - struct rb_root addr_space_rb; - struct drm_mm addr_space_mm; + struct drm_vma_offset_manager vma_manager; /* * Protected by the global:lru lock. */ struct list_head ddestroy; + uint32_t val_seq; /* * Protected by load / firstopen / lastclose /unload sync. */ - bool nice_mode; struct address_space *dev_mapping; /* @@ -507,8 +585,9 @@ ttm_flag_masked(uint32_t *old, uint32_t new, uint32_t mask) } /** - * ttm_tt_create + * ttm_tt_init * + * @ttm: The struct ttm_tt. * @bdev: pointer to a struct ttm_bo_device: * @size: Size of the data needed backing. * @page_flags: Page flags as identified by TTM_PAGE_FLAG_XX flags. @@ -519,28 +598,22 @@ ttm_flag_masked(uint32_t *old, uint32_t new, uint32_t mask) * Returns: * NULL: Out of memory. */ -extern struct ttm_tt *ttm_tt_create(struct ttm_bo_device *bdev, - unsigned long size, - uint32_t page_flags, - struct page *dummy_read_page); +extern int ttm_tt_init(struct ttm_tt *ttm, struct ttm_bo_device *bdev, + unsigned long size, uint32_t page_flags, + struct page *dummy_read_page); +extern int ttm_dma_tt_init(struct ttm_dma_tt *ttm_dma, struct ttm_bo_device *bdev, + unsigned long size, uint32_t page_flags, + struct page *dummy_read_page); /** - * ttm_tt_set_user: + * ttm_tt_fini * - * @ttm: The struct ttm_tt to populate. - * @tsk: A struct task_struct for which @start is a valid user-space address. - * @start: A valid user-space address. - * @num_pages: Size in pages of the user memory area. + * @ttm: the ttm_tt structure. * - * Populate a struct ttm_tt with a user-space memory area after first pinning - * the pages backing it. - * Returns: - * !0: Error. + * Free memory of ttm_tt structure */ - -extern int ttm_tt_set_user(struct ttm_tt *ttm, - struct task_struct *tsk, - unsigned long start, unsigned long num_pages); +extern void ttm_tt_fini(struct ttm_tt *ttm); +extern void ttm_dma_tt_fini(struct ttm_dma_tt *ttm_dma); /** * ttm_ttm_bind: @@ -553,20 +626,11 @@ extern int ttm_tt_set_user(struct ttm_tt *ttm, extern int ttm_tt_bind(struct ttm_tt *ttm, struct ttm_mem_reg *bo_mem); /** - * ttm_tt_populate: - * - * @ttm: The struct ttm_tt to contain the backing pages. - * - * Add backing pages to all of @ttm - */ -extern int ttm_tt_populate(struct ttm_tt *ttm); - -/** * ttm_ttm_destroy: * * @ttm: The struct ttm_tt. * - * Unbind, unpopulate and destroy a struct ttm_tt. + * Unbind, unpopulate and destroy common struct ttm_tt. */ extern void ttm_tt_destroy(struct ttm_tt *ttm); @@ -580,19 +644,13 @@ extern void ttm_tt_destroy(struct ttm_tt *ttm); extern void ttm_tt_unbind(struct ttm_tt *ttm); /** - * ttm_ttm_destroy: + * ttm_tt_swapin: * * @ttm: The struct ttm_tt. - * @index: Index of the desired page. - * - * Return a pointer to the struct page backing @ttm at page - * index @index. If the page is unpopulated, one will be allocated to - * populate that index. * - * Returns: - * NULL on OOM. + * Swap in a previously swap out ttm_tt. */ -extern struct page *ttm_tt_get_page(struct ttm_tt *ttm, int index); +extern int ttm_tt_swapin(struct ttm_tt *ttm); /** * ttm_tt_cache_flush: @@ -621,7 +679,16 @@ extern void ttm_tt_cache_flush(struct page *pages[], unsigned long num_pages); */ extern int ttm_tt_set_placement_caching(struct ttm_tt *ttm, uint32_t placement); extern int ttm_tt_swapout(struct ttm_tt *ttm, - struct file *persistant_swap_storage); + struct file *persistent_swap_storage); + +/** + * ttm_tt_unpopulate - free pages from a ttm + * + * @ttm: Pointer to the ttm_tt structure + * + * Calls the driver method to free all pages from a ttm + */ +extern void ttm_tt_unpopulate(struct ttm_tt *ttm); /* * ttm_bo.c @@ -647,7 +714,6 @@ extern bool ttm_mem_reg_is_pci(struct ttm_bo_device *bdev, * @proposed_placement: Proposed new placement for the buffer object. * @mem: A struct ttm_mem_reg. * @interruptible: Sleep interruptible when sliping. - * @no_wait_reserve: Return immediately if other buffers are busy. * @no_wait_gpu: Return immediately if the GPU is busy. * * Allocate memory space for the buffer object pointed to by @bo, using @@ -663,52 +729,13 @@ extern int ttm_bo_mem_space(struct ttm_buffer_object *bo, struct ttm_placement *placement, struct ttm_mem_reg *mem, bool interruptible, - bool no_wait_reserve, bool no_wait_gpu); + bool no_wait_gpu); extern void ttm_bo_mem_put(struct ttm_buffer_object *bo, struct ttm_mem_reg *mem); extern void ttm_bo_mem_put_locked(struct ttm_buffer_object *bo, struct ttm_mem_reg *mem); -/** - * ttm_bo_wait_for_cpu - * - * @bo: Pointer to a struct ttm_buffer_object. - * @no_wait: Don't sleep while waiting. - * - * Wait until a buffer object is no longer sync'ed for CPU access. - * Returns: - * -EBUSY: Buffer object was sync'ed for CPU access. (only if no_wait == 1). - * -ERESTARTSYS: An interruptible sleep was interrupted by a signal. - */ - -extern int ttm_bo_wait_cpu(struct ttm_buffer_object *bo, bool no_wait); - -/** - * ttm_bo_pci_offset - Get the PCI offset for the buffer object memory. - * - * @bo Pointer to a struct ttm_buffer_object. - * @bus_base On return the base of the PCI region - * @bus_offset On return the byte offset into the PCI region - * @bus_size On return the byte size of the buffer object or zero if - * the buffer object memory is not accessible through a PCI region. - * - * Returns: - * -EINVAL if the buffer object is currently not mappable. - * 0 otherwise. - */ - -extern int ttm_bo_pci_offset(struct ttm_bo_device *bdev, - struct ttm_mem_reg *mem, - unsigned long *bus_base, - unsigned long *bus_offset, - unsigned long *bus_size); - -extern int ttm_mem_io_reserve(struct ttm_bo_device *bdev, - struct ttm_mem_reg *mem); -extern void ttm_mem_io_free(struct ttm_bo_device *bdev, - struct ttm_mem_reg *mem); - extern void ttm_bo_global_release(struct drm_global_reference *ref); extern int ttm_bo_global_init(struct drm_global_reference *ref); @@ -718,8 +745,9 @@ extern int ttm_bo_device_release(struct ttm_bo_device *bdev); * ttm_bo_device_init * * @bdev: A pointer to a struct ttm_bo_device to initialize. - * @mem_global: A pointer to an initialized struct ttm_mem_global. + * @glob: A pointer to an initialized struct ttm_bo_global. * @driver: A pointer to a struct ttm_bo_driver set up by the caller. + * @mapping: The address space to use for this bo. * @file_page_offset: Offset into the device address space that is available * for buffer data. This ensures compatibility with other users of the * address space. @@ -731,6 +759,7 @@ extern int ttm_bo_device_release(struct ttm_bo_device *bdev); extern int ttm_bo_device_init(struct ttm_bo_device *bdev, struct ttm_bo_global *glob, struct ttm_bo_driver *driver, + struct address_space *mapping, uint64_t file_page_offset, bool need_dma32); /** @@ -741,13 +770,78 @@ extern int ttm_bo_device_init(struct ttm_bo_device *bdev, extern void ttm_bo_unmap_virtual(struct ttm_buffer_object *bo); /** + * ttm_bo_unmap_virtual + * + * @bo: tear down the virtual mappings for this BO + * + * The caller must take ttm_mem_io_lock before calling this function. + */ +extern void ttm_bo_unmap_virtual_locked(struct ttm_buffer_object *bo); + +extern int ttm_mem_io_reserve_vm(struct ttm_buffer_object *bo); +extern void ttm_mem_io_free_vm(struct ttm_buffer_object *bo); +extern int ttm_mem_io_lock(struct ttm_mem_type_manager *man, + bool interruptible); +extern void ttm_mem_io_unlock(struct ttm_mem_type_manager *man); + +extern void ttm_bo_del_sub_from_lru(struct ttm_buffer_object *bo); +extern void ttm_bo_add_to_lru(struct ttm_buffer_object *bo); + +/** + * __ttm_bo_reserve: + * + * @bo: A pointer to a struct ttm_buffer_object. + * @interruptible: Sleep interruptible if waiting. + * @no_wait: Don't sleep while trying to reserve, rather return -EBUSY. + * @use_ticket: If @bo is already reserved, Only sleep waiting for + * it to become unreserved if @ticket->stamp is older. + * + * Will not remove reserved buffers from the lru lists. + * Otherwise identical to ttm_bo_reserve. + * + * Returns: + * -EDEADLK: The reservation may cause a deadlock. + * Release all buffer reservations, wait for @bo to become unreserved and + * try again. (only if use_sequence == 1). + * -ERESTARTSYS: A wait for the buffer to become unreserved was interrupted by + * a signal. Release all buffer reservations and return to user-space. + * -EBUSY: The function needed to sleep, but @no_wait was true + * -EALREADY: Bo already reserved using @ticket. This error code will only + * be returned if @use_ticket is set to true. + */ +static inline int __ttm_bo_reserve(struct ttm_buffer_object *bo, + bool interruptible, + bool no_wait, bool use_ticket, + struct ww_acquire_ctx *ticket) +{ + int ret = 0; + + if (no_wait) { + bool success; + if (WARN_ON(ticket)) + return -EBUSY; + + success = ww_mutex_trylock(&bo->resv->lock); + return success ? 0 : -EBUSY; + } + + if (interruptible) + ret = ww_mutex_lock_interruptible(&bo->resv->lock, ticket); + else + ret = ww_mutex_lock(&bo->resv->lock, ticket); + if (ret == -EINTR) + return -ERESTARTSYS; + return ret; +} + +/** * ttm_bo_reserve: * * @bo: A pointer to a struct ttm_buffer_object. * @interruptible: Sleep interruptible if waiting. * @no_wait: Don't sleep while trying to reserve, rather return -EBUSY. - * @use_sequence: If @bo is already reserved, Only sleep waiting for - * it to become unreserved if @sequence < (@bo)->sequence. + * @use_ticket: If @bo is already reserved, Only sleep waiting for + * it to become unreserved if @ticket->stamp is older. * * Locks a buffer object for validation. (Or prevents other processes from * locking it for validation) and removes it from lru lists, while taking @@ -757,20 +851,11 @@ extern void ttm_bo_unmap_virtual(struct ttm_buffer_object *bo); * different order, either by will or as a result of a buffer being evicted * to make room for a buffer already reserved. (Buffers are reserved before * they are evicted). The following algorithm prevents such deadlocks from - * occuring: - * 1) Buffers are reserved with the lru spinlock held. Upon successful - * reservation they are removed from the lru list. This stops a reserved buffer - * from being evicted. However the lru spinlock is released between the time - * a buffer is selected for eviction and the time it is reserved. - * Therefore a check is made when a buffer is reserved for eviction, that it - * is still the first buffer in the lru list, before it is removed from the - * list. @check_lru == 1 forces this check. If it fails, the function returns - * -EINVAL, and the caller should then choose a new buffer to evict and repeat - * the procedure. - * 2) Processes attempting to reserve multiple buffers other than for eviction, + * occurring: + * Processes attempting to reserve multiple buffers other than for eviction, * (typically execbuf), should first obtain a unique 32-bit * validation sequence number, - * and call this function with @use_sequence == 1 and @sequence == the unique + * and call this function with @use_ticket == 1 and @ticket->stamp == the unique * sequence number. If upon call of this function, the buffer object is already * reserved, the validation sequence is checked against the validation * sequence of the process currently reserving the buffer, @@ -785,15 +870,74 @@ extern void ttm_bo_unmap_virtual(struct ttm_buffer_object *bo); * will eventually succeed, preventing both deadlocks and starvation. * * Returns: - * -EAGAIN: The reservation may cause a deadlock. + * -EDEADLK: The reservation may cause a deadlock. * Release all buffer reservations, wait for @bo to become unreserved and * try again. (only if use_sequence == 1). * -ERESTARTSYS: A wait for the buffer to become unreserved was interrupted by * a signal. Release all buffer reservations and return to user-space. + * -EBUSY: The function needed to sleep, but @no_wait was true + * -EALREADY: Bo already reserved using @ticket. This error code will only + * be returned if @use_ticket is set to true. */ -extern int ttm_bo_reserve(struct ttm_buffer_object *bo, - bool interruptible, - bool no_wait, bool use_sequence, uint32_t sequence); +static inline int ttm_bo_reserve(struct ttm_buffer_object *bo, + bool interruptible, + bool no_wait, bool use_ticket, + struct ww_acquire_ctx *ticket) +{ + int ret; + + WARN_ON(!atomic_read(&bo->kref.refcount)); + + ret = __ttm_bo_reserve(bo, interruptible, no_wait, use_ticket, ticket); + if (likely(ret == 0)) + ttm_bo_del_sub_from_lru(bo); + + return ret; +} + +/** + * ttm_bo_reserve_slowpath: + * @bo: A pointer to a struct ttm_buffer_object. + * @interruptible: Sleep interruptible if waiting. + * @sequence: Set (@bo)->sequence to this value after lock + * + * This is called after ttm_bo_reserve returns -EAGAIN and we backed off + * from all our other reservations. Because there are no other reservations + * held by us, this function cannot deadlock any more. + */ +static inline int ttm_bo_reserve_slowpath(struct ttm_buffer_object *bo, + bool interruptible, + struct ww_acquire_ctx *ticket) +{ + int ret = 0; + + WARN_ON(!atomic_read(&bo->kref.refcount)); + + if (interruptible) + ret = ww_mutex_lock_slow_interruptible(&bo->resv->lock, + ticket); + else + ww_mutex_lock_slow(&bo->resv->lock, ticket); + + if (likely(ret == 0)) + ttm_bo_del_sub_from_lru(bo); + else if (ret == -EINTR) + ret = -ERESTARTSYS; + + return ret; +} + +/** + * __ttm_bo_unreserve + * @bo: A pointer to a struct ttm_buffer_object. + * + * Unreserve a previous reservation of @bo where the buffer object is + * already on lru lists. + */ +static inline void __ttm_bo_unreserve(struct ttm_buffer_object *bo) +{ + ww_mutex_unlock(&bo->resv->lock); +} /** * ttm_bo_unreserve @@ -802,30 +946,42 @@ extern int ttm_bo_reserve(struct ttm_buffer_object *bo, * * Unreserve a previous reservation of @bo. */ -extern void ttm_bo_unreserve(struct ttm_buffer_object *bo); +static inline void ttm_bo_unreserve(struct ttm_buffer_object *bo) +{ + if (!(bo->mem.placement & TTM_PL_FLAG_NO_EVICT)) { + spin_lock(&bo->glob->lru_lock); + ttm_bo_add_to_lru(bo); + spin_unlock(&bo->glob->lru_lock); + } + __ttm_bo_unreserve(bo); +} /** - * ttm_bo_wait_unreserved - * + * ttm_bo_unreserve_ticket * @bo: A pointer to a struct ttm_buffer_object. + * @ticket: ww_acquire_ctx used for reserving * - * Wait for a struct ttm_buffer_object to become unreserved. - * This is typically used in the execbuf code to relax cpu-usage when - * a potential deadlock condition backoff. + * Unreserve a previous reservation of @bo made with @ticket. */ -extern int ttm_bo_wait_unreserved(struct ttm_buffer_object *bo, - bool interruptible); +static inline void ttm_bo_unreserve_ticket(struct ttm_buffer_object *bo, + struct ww_acquire_ctx *t) +{ + ttm_bo_unreserve(bo); +} /* * ttm_bo_util.c */ +int ttm_mem_io_reserve(struct ttm_bo_device *bdev, + struct ttm_mem_reg *mem); +void ttm_mem_io_free(struct ttm_bo_device *bdev, + struct ttm_mem_reg *mem); /** * ttm_bo_move_ttm * * @bo: A pointer to a struct ttm_buffer_object. * @evict: 1: This is an eviction. Don't try to pipeline. - * @no_wait_reserve: Return immediately if other buffers are busy. * @no_wait_gpu: Return immediately if the GPU is busy. * @new_mem: struct ttm_mem_reg indicating where to move. * @@ -840,15 +996,14 @@ extern int ttm_bo_wait_unreserved(struct ttm_buffer_object *bo, */ extern int ttm_bo_move_ttm(struct ttm_buffer_object *bo, - bool evict, bool no_wait_reserve, - bool no_wait_gpu, struct ttm_mem_reg *new_mem); + bool evict, bool no_wait_gpu, + struct ttm_mem_reg *new_mem); /** * ttm_bo_move_memcpy * * @bo: A pointer to a struct ttm_buffer_object. * @evict: 1: This is an eviction. Don't try to pipeline. - * @no_wait_reserve: Return immediately if other buffers are busy. * @no_wait_gpu: Return immediately if the GPU is busy. * @new_mem: struct ttm_mem_reg indicating where to move. * @@ -863,8 +1018,8 @@ extern int ttm_bo_move_ttm(struct ttm_buffer_object *bo, */ extern int ttm_bo_move_memcpy(struct ttm_buffer_object *bo, - bool evict, bool no_wait_reserve, - bool no_wait_gpu, struct ttm_mem_reg *new_mem); + bool evict, bool no_wait_gpu, + struct ttm_mem_reg *new_mem); /** * ttm_bo_free_old_node @@ -880,10 +1035,7 @@ extern void ttm_bo_free_old_node(struct ttm_buffer_object *bo); * * @bo: A pointer to a struct ttm_buffer_object. * @sync_obj: A sync object that signals when moving is complete. - * @sync_obj_arg: An argument to pass to the sync object idle / wait - * functions. * @evict: This is an evict move. Don't return until the buffer is idle. - * @no_wait_reserve: Return immediately if other buffers are busy. * @no_wait_gpu: Return immediately if the GPU is busy. * @new_mem: struct ttm_mem_reg indicating where to move. * @@ -897,9 +1049,7 @@ extern void ttm_bo_free_old_node(struct ttm_buffer_object *bo); extern int ttm_bo_move_accel_cleanup(struct ttm_buffer_object *bo, void *sync_obj, - void *sync_obj_arg, - bool evict, bool no_wait_reserve, - bool no_wait_gpu, + bool evict, bool no_wait_gpu, struct ttm_mem_reg *new_mem); /** * ttm_io_prot @@ -919,17 +1069,25 @@ extern const struct ttm_mem_type_manager_func ttm_bo_manager_func; #include <linux/agp_backend.h> /** - * ttm_agp_backend_init + * ttm_agp_tt_create * * @bdev: Pointer to a struct ttm_bo_device. * @bridge: The agp bridge this device is sitting on. + * @size: Size of the data needed backing. + * @page_flags: Page flags as identified by TTM_PAGE_FLAG_XX flags. + * @dummy_read_page: See struct ttm_bo_device. + * * * Create a TTM backend that uses the indicated AGP bridge as an aperture * for TT memory. This function uses the linux agpgart interface to * bind and unbind memory backing a ttm_tt. */ -extern struct ttm_backend *ttm_agp_backend_init(struct ttm_bo_device *bdev, - struct agp_bridge_data *bridge); +extern struct ttm_tt *ttm_agp_tt_create(struct ttm_bo_device *bdev, + struct agp_bridge_data *bridge, + unsigned long size, uint32_t page_flags, + struct page *dummy_read_page); +int ttm_agp_tt_populate(struct ttm_tt *ttm); +void ttm_agp_tt_unpopulate(struct ttm_tt *ttm); #endif #endif diff --git a/include/drm/ttm/ttm_execbuf_util.h b/include/drm/ttm/ttm_execbuf_util.h index cd2c475da9e..16db7d01a33 100644 --- a/include/drm/ttm/ttm_execbuf_util.h +++ b/include/drm/ttm/ttm_execbuf_util.h @@ -31,7 +31,7 @@ #ifndef _TTM_EXECBUF_UTIL_H_ #define _TTM_EXECBUF_UTIL_H_ -#include "ttm/ttm_bo_api.h" +#include <ttm/ttm_bo_api.h> #include <linux/list.h> /** @@ -39,34 +39,40 @@ * * @head: list head for thread-private list. * @bo: refcounted buffer object pointer. - * @new_sync_obj_arg: New sync_obj_arg for @bo, to be used once - * adding a new sync object. - * @reservied: Indicates whether @bo has been reserved for validation. + * @reserved: Indicates whether @bo has been reserved for validation. + * @removed: Indicates whether @bo has been removed from lru lists. + * @put_count: Number of outstanding references on bo::list_kref. + * @old_sync_obj: Pointer to a sync object about to be unreferenced */ struct ttm_validate_buffer { struct list_head head; struct ttm_buffer_object *bo; - void *new_sync_obj_arg; bool reserved; + bool removed; + int put_count; + void *old_sync_obj; }; /** * function ttm_eu_backoff_reservation * + * @ticket: ww_acquire_ctx from reserve call * @list: thread private list of ttm_validate_buffer structs. * * Undoes all buffer validation reservations for bos pointed to by * the list entries. */ -extern void ttm_eu_backoff_reservation(struct list_head *list); +extern void ttm_eu_backoff_reservation(struct ww_acquire_ctx *ticket, + struct list_head *list); /** * function ttm_eu_reserve_buffers * + * @ticket: [out] ww_acquire_ctx filled in by call, or NULL if only + * non-blocking reserves should be tried. * @list: thread private list of ttm_validate_buffer structs. - * @val_seq: A unique sequence number. * * Tries to reserve bos pointed to by the list entries for validation. * If the function returns 0, all buffers are marked as "unfenced", @@ -88,11 +94,13 @@ extern void ttm_eu_backoff_reservation(struct list_head *list); * has failed. */ -extern int ttm_eu_reserve_buffers(struct list_head *list, uint32_t val_seq); +extern int ttm_eu_reserve_buffers(struct ww_acquire_ctx *ticket, + struct list_head *list); /** * function ttm_eu_fence_buffer_objects. * + * @ticket: ww_acquire_ctx from reserve call * @list: thread private list of ttm_validate_buffer structs. * @sync_obj: The new sync object for the buffers. * @@ -102,6 +110,7 @@ extern int ttm_eu_reserve_buffers(struct list_head *list, uint32_t val_seq); * */ -extern void ttm_eu_fence_buffer_objects(struct list_head *list, void *sync_obj); +extern void ttm_eu_fence_buffer_objects(struct ww_acquire_ctx *ticket, + struct list_head *list, void *sync_obj); #endif diff --git a/include/drm/ttm/ttm_lock.h b/include/drm/ttm/ttm_lock.h index 81ba0b0b891..2902beb5f68 100644 --- a/include/drm/ttm/ttm_lock.h +++ b/include/drm/ttm/ttm_lock.h @@ -49,9 +49,9 @@ #ifndef _TTM_LOCK_H_ #define _TTM_LOCK_H_ -#include "ttm/ttm_object.h" +#include <ttm/ttm_object.h> #include <linux/wait.h> -#include <asm/atomic.h> +#include <linux/atomic.h> /** * struct ttm_lock diff --git a/include/drm/ttm/ttm_memory.h b/include/drm/ttm/ttm_memory.h index b199170b3c2..72dcbe81dd0 100644 --- a/include/drm/ttm/ttm_memory.h +++ b/include/drm/ttm/ttm_memory.h @@ -30,6 +30,7 @@ #include <linux/workqueue.h> #include <linux/spinlock.h> +#include <linux/bug.h> #include <linux/wait.h> #include <linux/errno.h> #include <linux/kobject.h> @@ -41,7 +42,7 @@ * @do_shrink: The callback function. * * Arguments to the do_shrink functions are intended to be passed using - * inheritance. That is, the argument class derives from struct ttm_mem_srink, + * inheritance. That is, the argument class derives from struct ttm_mem_shrink, * and can be accessed using container_of(). */ @@ -59,7 +60,6 @@ struct ttm_mem_shrink { * for the GPU, and this will otherwise block other workqueue tasks(?) * At this point we use only a single-threaded workqueue. * @work: The workqueue callback for the shrink queue. - * @queue: Wait queue for processes suspended waiting for memory. * @lock: Lock to protect the @shrink - and the memory accounting members, * that is, essentially the whole structure with some exceptions. * @zones: Array of pointers to accounting zones. @@ -79,7 +79,6 @@ struct ttm_mem_global { struct ttm_mem_shrink *shrink; struct workqueue_struct *swap_queue; struct work_struct work; - wait_queue_head_t queue; spinlock_t lock; struct ttm_mem_zone *zones[TTM_MEM_MAX_ZONES]; unsigned int num_zones; diff --git a/include/drm/ttm/ttm_object.h b/include/drm/ttm/ttm_object.h index 0d9db099978..ed953f98f0e 100644 --- a/include/drm/ttm/ttm_object.h +++ b/include/drm/ttm/ttm_object.h @@ -38,8 +38,10 @@ #define _TTM_OBJECT_H_ #include <linux/list.h> -#include "drm_hashtab.h" +#include <drm/drm_hashtab.h> #include <linux/kref.h> +#include <linux/rcupdate.h> +#include <linux/dma-buf.h> #include <ttm/ttm_memory.h> /** @@ -76,6 +78,7 @@ enum ttm_object_type { ttm_fence_type, ttm_buffer_type, ttm_lock_type, + ttm_prime_type, ttm_driver_type0 = 256, ttm_driver_type1, ttm_driver_type2, @@ -111,7 +114,7 @@ struct ttm_object_device; * * @ref_obj_release: A function to be called when a reference object * with another ttm_ref_type than TTM_REF_USAGE is deleted. - * this function may, for example, release a lock held by a user-space + * This function may, for example, release a lock held by a user-space * process. * * This struct is intended to be used as a base struct for objects that @@ -120,6 +123,7 @@ struct ttm_object_device; */ struct ttm_base_object { + struct rcu_head rhead; struct drm_hash_item hash; enum ttm_object_type object_type; bool shareable; @@ -130,6 +134,30 @@ struct ttm_base_object { enum ttm_ref_type ref_type); }; + +/** + * struct ttm_prime_object - Modified base object that is prime-aware + * + * @base: struct ttm_base_object that we derive from + * @mutex: Mutex protecting the @dma_buf member. + * @size: Size of the dma_buf associated with this object + * @real_type: Type of the underlying object. Needed since we're setting + * the value of @base::object_type to ttm_prime_type + * @dma_buf: Non ref-coutned pointer to a struct dma_buf created from this + * object. + * @refcount_release: The underlying object's release method. Needed since + * we set @base::refcount_release to our own release method. + */ + +struct ttm_prime_object { + struct ttm_base_object base; + struct mutex mutex; + size_t size; + enum ttm_object_type real_type; + struct dma_buf *dma_buf; + void (*refcount_release) (struct ttm_base_object **); +}; + /** * ttm_base_object_init * @@ -162,17 +190,29 @@ extern int ttm_base_object_init(struct ttm_object_file *tfile, * @key: Hash key * * Looks up a struct ttm_base_object with the key @key. - * Also verifies that the object is visible to the application, by - * comparing the @tfile argument and checking the object shareable flag. */ extern struct ttm_base_object *ttm_base_object_lookup(struct ttm_object_file *tfile, uint32_t key); /** + * ttm_base_object_lookup_for_ref + * + * @tdev: Pointer to a struct ttm_object_device. + * @key: Hash key + * + * Looks up a struct ttm_base_object with the key @key. + * This function should only be used when the struct tfile associated with the + * caller doesn't yet have a reference to the base object. + */ + +extern struct ttm_base_object * +ttm_base_object_lookup_for_ref(struct ttm_object_device *tdev, uint32_t key); + +/** * ttm_base_object_unref * - * @p_base: Pointer to a pointer referncing a struct ttm_base_object. + * @p_base: Pointer to a pointer referencing a struct ttm_base_object. * * Decrements the base object refcount and clears the pointer pointed to by * p_base. @@ -190,6 +230,8 @@ extern void ttm_base_object_unref(struct ttm_base_object **p_base); * @existed: Upon completion, indicates that an identical reference object * already existed, and the refcount was upped on that object instead. * + * Checks that the base object is shareable and adds a ref object to it. + * * Adding a ref object to a base object is basically like referencing the * base object, but a user-space application holds the reference. When the * file corresponding to @tfile is closed, all its reference objects are @@ -202,6 +244,10 @@ extern void ttm_base_object_unref(struct ttm_base_object **p_base); extern int ttm_ref_object_add(struct ttm_object_file *tfile, struct ttm_base_object *base, enum ttm_ref_type ref_type, bool *existed); + +extern bool ttm_ref_object_exists(struct ttm_object_file *tfile, + struct ttm_base_object *base); + /** * ttm_ref_object_base_unref * @@ -246,14 +292,18 @@ extern void ttm_object_file_release(struct ttm_object_file **p_tfile); /** * ttm_object device init - initialize a struct ttm_object_device * + * @mem_glob: struct ttm_mem_global for memory accounting. * @hash_order: Order of hash table used to hash the base objects. + * @ops: DMA buf ops for prime objects of this device. * * This function is typically called on device initialization to prepare * data structures needed for ttm base and ref objects. */ -extern struct ttm_object_device *ttm_object_device_init - (struct ttm_mem_global *mem_glob, unsigned int hash_order); +extern struct ttm_object_device * +ttm_object_device_init(struct ttm_mem_global *mem_glob, + unsigned int hash_order, + const struct dma_buf_ops *ops); /** * ttm_object_device_release - release data held by a ttm_object_device @@ -268,4 +318,33 @@ extern struct ttm_object_device *ttm_object_device_init extern void ttm_object_device_release(struct ttm_object_device **p_tdev); +#define ttm_base_object_kfree(__object, __base)\ + kfree_rcu(__object, __base.rhead) + +extern int ttm_prime_object_init(struct ttm_object_file *tfile, + size_t size, + struct ttm_prime_object *prime, + bool shareable, + enum ttm_object_type type, + void (*refcount_release) + (struct ttm_base_object **), + void (*ref_obj_release) + (struct ttm_base_object *, + enum ttm_ref_type ref_type)); + +static inline enum ttm_object_type +ttm_base_object_type(struct ttm_base_object *base) +{ + return (base->object_type == ttm_prime_type) ? + container_of(base, struct ttm_prime_object, base)->real_type : + base->object_type; +} +extern int ttm_prime_fd_to_handle(struct ttm_object_file *tfile, + int fd, u32 *handle); +extern int ttm_prime_handle_to_fd(struct ttm_object_file *tfile, + uint32_t handle, uint32_t flags, + int *prime_fd); + +#define ttm_prime_object_kfree(__obj, __prime) \ + kfree_rcu(__obj, __prime.base.rhead) #endif diff --git a/include/drm/ttm/ttm_page_alloc.h b/include/drm/ttm/ttm_page_alloc.h index 116821448c3..49a828425fa 100644 --- a/include/drm/ttm/ttm_page_alloc.h +++ b/include/drm/ttm/ttm_page_alloc.h @@ -26,45 +26,85 @@ #ifndef TTM_PAGE_ALLOC #define TTM_PAGE_ALLOC -#include "ttm_bo_driver.h" -#include "ttm_memory.h" +#include <drm/ttm/ttm_bo_driver.h> +#include <drm/ttm/ttm_memory.h> +struct device; + +/** + * Initialize pool allocator. + */ +int ttm_page_alloc_init(struct ttm_mem_global *glob, unsigned max_pages); /** - * Get count number of pages from pool to pages list. + * Free pool allocator. + */ +void ttm_page_alloc_fini(void); + +/** + * ttm_pool_populate: * - * @pages: heado of empty linked list where pages are filled. - * @flags: ttm flags for page allocation. - * @cstate: ttm caching state for the page. - * @count: number of pages to allocate. + * @ttm: The struct ttm_tt to contain the backing pages. + * + * Add backing pages to all of @ttm */ -int ttm_get_pages(struct list_head *pages, - int flags, - enum ttm_caching_state cstate, - unsigned count); +extern int ttm_pool_populate(struct ttm_tt *ttm); + /** - * Put linked list of pages to pool. + * ttm_pool_unpopulate: * - * @pages: list of pages to free. - * @page_count: number of pages in the list. Zero can be passed for unknown - * count. - * @flags: ttm flags for page allocation. - * @cstate: ttm caching state. + * @ttm: The struct ttm_tt which to free backing pages. + * + * Free all pages of @ttm + */ +extern void ttm_pool_unpopulate(struct ttm_tt *ttm); + +/** + * Output the state of pools to debugfs file */ -void ttm_put_pages(struct list_head *pages, - unsigned page_count, - int flags, - enum ttm_caching_state cstate); +extern int ttm_page_alloc_debugfs(struct seq_file *m, void *data); + + +#if defined(CONFIG_SWIOTLB) || defined(CONFIG_INTEL_IOMMU) /** * Initialize pool allocator. */ -int ttm_page_alloc_init(struct ttm_mem_global *glob, unsigned max_pages); +int ttm_dma_page_alloc_init(struct ttm_mem_global *glob, unsigned max_pages); + /** * Free pool allocator. */ -void ttm_page_alloc_fini(void); +void ttm_dma_page_alloc_fini(void); /** * Output the state of pools to debugfs file */ -extern int ttm_page_alloc_debugfs(struct seq_file *m, void *data); +extern int ttm_dma_page_alloc_debugfs(struct seq_file *m, void *data); + +extern int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev); +extern void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, struct device *dev); + +#else +static inline int ttm_dma_page_alloc_init(struct ttm_mem_global *glob, + unsigned max_pages) +{ + return -ENODEV; +} + +static inline void ttm_dma_page_alloc_fini(void) { return; } + +static inline int ttm_dma_page_alloc_debugfs(struct seq_file *m, void *data) +{ + return 0; +} +static inline int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, + struct device *dev) +{ + return -ENOMEM; +} +static inline void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma, + struct device *dev) +{ +} +#endif + #endif diff --git a/include/drm/ttm/ttm_placement.h b/include/drm/ttm/ttm_placement.h index c84ff153a56..8ed44f9bbdf 100644 --- a/include/drm/ttm/ttm_placement.h +++ b/include/drm/ttm/ttm_placement.h @@ -65,6 +65,8 @@ * reference the buffer. * TTM_PL_FLAG_NO_EVICT means that the buffer may never * be evicted to make room for other buffers. + * TTM_PL_FLAG_TOPDOWN requests to be placed from the + * top of the memory area, instead of the bottom. */ #define TTM_PL_FLAG_CACHED (1 << 16) @@ -72,6 +74,7 @@ #define TTM_PL_FLAG_WC (1 << 18) #define TTM_PL_FLAG_SHARED (1 << 20) #define TTM_PL_FLAG_NO_EVICT (1 << 21) +#define TTM_PL_FLAG_TOPDOWN (1 << 22) #define TTM_PL_MASK_CACHING (TTM_PL_FLAG_CACHED | \ TTM_PL_FLAG_UNCACHED | \ |