diff options
Diffstat (limited to 'include/drm/ttm')
| -rw-r--r-- | include/drm/ttm/ttm_bo_api.h | 77 | ||||
| -rw-r--r-- | include/drm/ttm/ttm_bo_driver.h | 270 | ||||
| -rw-r--r-- | include/drm/ttm/ttm_execbuf_util.h | 18 | ||||
| -rw-r--r-- | include/drm/ttm/ttm_lock.h | 2 | ||||
| -rw-r--r-- | include/drm/ttm/ttm_memory.h | 3 | ||||
| -rw-r--r-- | include/drm/ttm/ttm_object.h | 89 | ||||
| -rw-r--r-- | include/drm/ttm/ttm_page_alloc.h | 17 | ||||
| -rw-r--r-- | include/drm/ttm/ttm_placement.h | 3 |
8 files changed, 303 insertions, 176 deletions
diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h index 974c8f801c3..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; @@ -124,11 +125,15 @@ struct ttm_mem_reg { * * @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_kernel + ttm_bo_type_kernel, + ttm_bo_type_sg }; struct ttm_tt; @@ -137,12 +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. @@ -151,7 +153,6 @@ 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. * @mem: structure describing current placement. * @persistent_swap_storage: Usually the swap storage is deleted for buffers * pinned in physical memory. If this behaviour is not desired, this member @@ -162,20 +163,13 @@ struct ttm_tt; * @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 @@ -196,11 +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; /** @@ -209,10 +201,9 @@ struct ttm_buffer_object { struct kref kref; struct kref list_kref; - wait_queue_head_t event_queue; /** - * Members protected by the bo::reserved lock. + * Members protected by the bo::resv::reserved lock. */ struct ttm_mem_reg mem; @@ -234,15 +225,6 @@ struct ttm_buffer_object { struct list_head ddestroy; struct list_head swap; struct list_head io_reserve_lru; - uint32_t val_seq; - bool seq_valid; - - /** - * Members protected by the bdev::lru_lock - * only when written to. - */ - - atomic_t reserved; /** * Members protected by struct buffer_object_device::fence_lock @@ -251,17 +233,10 @@ struct ttm_buffer_object { * 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 @@ -271,6 +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; }; /** @@ -336,7 +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 @@ -349,7 +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); /** @@ -423,8 +403,9 @@ 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. @@ -466,8 +447,6 @@ size_t ttm_bo_dma_acc_size(struct ttm_bo_device *bdev, * @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. * @persistent_swap_storage: Usually the swap storage is deleted for buffers @@ -499,23 +478,20 @@ 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 *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. * @persistent_swap_storage: Usually the swap storage is deleted for buffers @@ -538,7 +514,6 @@ 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 *persistent_swap_storage, struct ttm_buffer_object **p_bo); @@ -728,5 +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 d43e892307f..a5183da3ef9 100644 --- a/include/drm/ttm/ttm_bo_driver.h +++ b/include/drm/ttm/ttm_bo_driver.h @@ -30,16 +30,17 @@ #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 { /** @@ -81,6 +82,7 @@ struct ttm_backend_func { #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, @@ -116,8 +118,8 @@ struct ttm_tt { struct page **pages; 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 file *swap_storage; enum ttm_caching_state caching_state; enum { @@ -395,7 +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); /** @@ -423,10 +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); @@ -518,12 +520,10 @@ struct ttm_bo_global { * @man: An array of mem_type_managers. * @fence_lock: Protects the synchronizing members on *all* bos belonging * to this device. - * @addr_space_mm: Range manager for the device address space. + * @vma_manager: Address space manager * lru_lock: Spinlock that protects the buffer+device lru lists and * ddestroy lists. * @val_seq: Current validation sequence. - * @nice_mode: Try nicely to wait for buffer idle when cleaning a manager. - * If a GPU lockup has been detected, this is forced to 0. * @dev_mapping: A pointer to the struct address_space representing the * device address space. * @wq: Work queue structure for the delayed delete workqueue. @@ -538,14 +538,13 @@ 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. @@ -557,7 +556,6 @@ struct ttm_bo_device { * Protected by load / firstopen / lastclose /unload sync. */ - bool nice_mode; struct address_space *dev_mapping; /* @@ -683,6 +681,15 @@ 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 *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 */ @@ -707,7 +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 @@ -723,27 +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); - extern void ttm_bo_global_release(struct drm_global_reference *ref); extern int ttm_bo_global_init(struct drm_global_reference *ref); @@ -755,6 +747,7 @@ extern int ttm_bo_device_release(struct ttm_bo_device *bdev); * @bdev: A pointer to a struct ttm_bo_device to initialize. * @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. @@ -766,6 +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); /** @@ -790,6 +784,55 @@ 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: @@ -797,8 +840,8 @@ extern void ttm_mem_io_unlock(struct ttm_mem_type_manager *man); * @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 @@ -809,19 +852,10 @@ extern void ttm_mem_io_unlock(struct ttm_mem_type_manager *man); * to make room for a buffer already reserved. (Buffers are reserved before * they are evicted). The following algorithm prevents such deadlocks from * occurring: - * 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, + * 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, @@ -836,90 +870,118 @@ extern void ttm_mem_io_unlock(struct ttm_mem_type_manager *man); * 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 - * -EDEADLK: Bo already reserved using @sequence. This error code will only - * be returned if @use_sequence is set to 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_locked: - * + * ttm_bo_reserve_slowpath: * @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. - * - * Must be called with struct ttm_bo_global::lru_lock held, - * and will not remove reserved buffers from the lru lists. - * The function may release the LRU spinlock if it needs to sleep. - * Otherwise identical to ttm_bo_reserve. + * @sequence: Set (@bo)->sequence to this value after lock * - * Returns: - * -EAGAIN: The reservation may cause a deadlock. - * Release all buffer reservations, wait for @bo to become unreserved and - * try again. (only if use_sequence == 1). - * -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 - * -EDEADLK: Bo already reserved using @sequence. This error code will only - * be returned if @use_sequence is set to true. + * 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. */ -extern int ttm_bo_reserve_locked(struct ttm_buffer_object *bo, - bool interruptible, - bool no_wait, bool use_sequence, - uint32_t sequence); +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 - * + * __ttm_bo_unreserve * @bo: A pointer to a struct ttm_buffer_object. * - * Unreserve a previous reservation of @bo. + * Unreserve a previous reservation of @bo where the buffer object is + * already on lru lists. */ -extern void ttm_bo_unreserve(struct ttm_buffer_object *bo); +static inline void __ttm_bo_unreserve(struct ttm_buffer_object *bo) +{ + ww_mutex_unlock(&bo->resv->lock); +} /** - * ttm_bo_unreserve_locked + * ttm_bo_unreserve * * @bo: A pointer to a struct ttm_buffer_object. * * Unreserve a previous reservation of @bo. - * Needs to be called with struct ttm_bo_global::lru_lock held. */ -extern void ttm_bo_unreserve_locked(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. * @@ -934,15 +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. * @@ -957,8 +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 @@ -974,10 +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. * @@ -991,9 +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 diff --git a/include/drm/ttm/ttm_execbuf_util.h b/include/drm/ttm/ttm_execbuf_util.h index 26cc7f9ffa4..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> /** @@ -39,8 +39,6 @@ * * @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. * @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. @@ -50,7 +48,6 @@ 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; @@ -60,17 +57,21 @@ struct ttm_validate_buffer { /** * 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. * * Tries to reserve bos pointed to by the list entries for validation. @@ -93,11 +94,13 @@ extern void ttm_eu_backoff_reservation(struct list_head *list); * has failed. */ -extern int ttm_eu_reserve_buffers(struct list_head *list); +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. * @@ -107,6 +110,7 @@ extern int ttm_eu_reserve_buffers(struct list_head *list); * */ -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 2e7f0c941b5..2902beb5f68 100644 --- a/include/drm/ttm/ttm_lock.h +++ b/include/drm/ttm/ttm_lock.h @@ -49,7 +49,7 @@ #ifndef _TTM_LOCK_H_ #define _TTM_LOCK_H_ -#include "ttm/ttm_object.h" +#include <ttm/ttm_object.h> #include <linux/wait.h> #include <linux/atomic.h> diff --git a/include/drm/ttm/ttm_memory.h b/include/drm/ttm/ttm_memory.h index 26c1f78d136..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> @@ -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 e46054e5255..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, @@ -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 * @@ -162,14 +190,26 @@ 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 referencing a struct ttm_base_object. @@ -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 5fe27400d17..49a828425fa 100644 --- a/include/drm/ttm/ttm_page_alloc.h +++ b/include/drm/ttm/ttm_page_alloc.h @@ -26,8 +26,10 @@ #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. @@ -62,7 +64,7 @@ extern void ttm_pool_unpopulate(struct ttm_tt *ttm); extern int ttm_page_alloc_debugfs(struct seq_file *m, void *data); -#ifdef CONFIG_SWIOTLB +#if defined(CONFIG_SWIOTLB) || defined(CONFIG_INTEL_IOMMU) /** * Initialize pool allocator. */ @@ -94,6 +96,15 @@ 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 | \ |