diff options
Diffstat (limited to 'include/drm/gpu_scheduler.h')
-rw-r--r-- | include/drm/gpu_scheduler.h | 319 |
1 files changed, 262 insertions, 57 deletions
diff --git a/include/drm/gpu_scheduler.h b/include/drm/gpu_scheduler.h index 589be851f8a1..2ae4fd62e01c 100644 --- a/include/drm/gpu_scheduler.h +++ b/include/drm/gpu_scheduler.h @@ -27,21 +27,35 @@ #include <drm/spsc_queue.h> #include <linux/dma-fence.h> #include <linux/completion.h> +#include <linux/xarray.h> +#include <linux/workqueue.h> #define MAX_WAIT_SCHED_ENTITY_Q_EMPTY msecs_to_jiffies(1000) +/** + * DRM_SCHED_FENCE_DONT_PIPELINE - Prefent dependency pipelining + * + * Setting this flag on a scheduler fence prevents pipelining of jobs depending + * on this fence. In other words we always insert a full CPU round trip before + * dependen jobs are pushed to the hw queue. + */ +#define DRM_SCHED_FENCE_DONT_PIPELINE DMA_FENCE_FLAG_USER_BITS + +struct drm_gem_object; + struct drm_gpu_scheduler; struct drm_sched_rq; +/* These are often used as an (initial) index + * to an array, and as such should start at 0. + */ enum drm_sched_priority { DRM_SCHED_PRIORITY_MIN, - DRM_SCHED_PRIORITY_LOW = DRM_SCHED_PRIORITY_MIN, DRM_SCHED_PRIORITY_NORMAL, - DRM_SCHED_PRIORITY_HIGH_SW, - DRM_SCHED_PRIORITY_HIGH_HW, + DRM_SCHED_PRIORITY_HIGH, DRM_SCHED_PRIORITY_KERNEL, - DRM_SCHED_PRIORITY_MAX, - DRM_SCHED_PRIORITY_INVALID = -1, + + DRM_SCHED_PRIORITY_COUNT, DRM_SCHED_PRIORITY_UNSET = -2 }; @@ -49,55 +63,147 @@ enum drm_sched_priority { * struct drm_sched_entity - A wrapper around a job queue (typically * attached to the DRM file_priv). * - * @list: used to append this struct to the list of entities in the - * runqueue. - * @rq: runqueue on which this entity is currently scheduled. - * @sched_list: A list of schedulers (drm_gpu_schedulers). - * Jobs from this entity can be scheduled on any scheduler - * on this list. - * @num_sched_list: number of drm_gpu_schedulers in the sched_list. - * @rq_lock: lock to modify the runqueue to which this entity belongs. - * @job_queue: the list of jobs of this entity. - * @fence_seq: a linearly increasing seqno incremented with each - * new &drm_sched_fence which is part of the entity. - * @fence_context: a unique context for all the fences which belong - * to this entity. - * The &drm_sched_fence.scheduled uses the - * fence_context but &drm_sched_fence.finished uses - * fence_context + 1. - * @dependency: the dependency fence of the job which is on the top - * of the job queue. - * @cb: callback for the dependency fence above. - * @guilty: points to ctx's guilty. - * @fini_status: contains the exit status in case the process was signalled. - * @last_scheduled: points to the finished fence of the last scheduled job. - * @last_user: last group leader pushing a job into the entity. - * @stopped: Marks the enity as removed from rq and destined for termination. - * @entity_idle: Signals when enityt is not in use - * * Entities will emit jobs in order to their corresponding hardware * ring, and the scheduler will alternate between entities based on * scheduling policy. */ struct drm_sched_entity { + /** + * @list: + * + * Used to append this struct to the list of entities in the runqueue + * @rq under &drm_sched_rq.entities. + * + * Protected by &drm_sched_rq.lock of @rq. + */ struct list_head list; + + /** + * @rq: + * + * Runqueue on which this entity is currently scheduled. + * + * FIXME: Locking is very unclear for this. Writers are protected by + * @rq_lock, but readers are generally lockless and seem to just race + * with not even a READ_ONCE. + */ struct drm_sched_rq *rq; + + /** + * @sched_list: + * + * A list of schedulers (struct drm_gpu_scheduler). Jobs from this entity can + * be scheduled on any scheduler on this list. + * + * This can be modified by calling drm_sched_entity_modify_sched(). + * Locking is entirely up to the driver, see the above function for more + * details. + * + * This will be set to NULL if &num_sched_list equals 1 and @rq has been + * set already. + * + * FIXME: This means priority changes through + * drm_sched_entity_set_priority() will be lost henceforth in this case. + */ struct drm_gpu_scheduler **sched_list; + + /** + * @num_sched_list: + * + * Number of drm_gpu_schedulers in the @sched_list. + */ unsigned int num_sched_list; + + /** + * @priority: + * + * Priority of the entity. This can be modified by calling + * drm_sched_entity_set_priority(). Protected by &rq_lock. + */ enum drm_sched_priority priority; + + /** + * @rq_lock: + * + * Lock to modify the runqueue to which this entity belongs. + */ spinlock_t rq_lock; + /** + * @job_queue: the list of jobs of this entity. + */ struct spsc_queue job_queue; + /** + * @fence_seq: + * + * A linearly increasing seqno incremented with each new + * &drm_sched_fence which is part of the entity. + * + * FIXME: Callers of drm_sched_job_arm() need to ensure correct locking, + * this doesn't need to be atomic. + */ atomic_t fence_seq; + + /** + * @fence_context: + * + * A unique context for all the fences which belong to this entity. The + * &drm_sched_fence.scheduled uses the fence_context but + * &drm_sched_fence.finished uses fence_context + 1. + */ uint64_t fence_context; + /** + * @dependency: + * + * The dependency fence of the job which is on the top of the job queue. + */ struct dma_fence *dependency; + + /** + * @cb: + * + * Callback for the dependency fence above. + */ struct dma_fence_cb cb; + + /** + * @guilty: + * + * Points to entities' guilty. + */ atomic_t *guilty; + + /** + * @last_scheduled: + * + * Points to the finished fence of the last scheduled job. Only written + * by the scheduler thread, can be accessed locklessly from + * drm_sched_job_arm() iff the queue is empty. + */ struct dma_fence *last_scheduled; + + /** + * @last_user: last group leader pushing a job into the entity. + */ struct task_struct *last_user; + + /** + * @stopped: + * + * Marks the enity as removed from rq and destined for + * termination. This is set by calling drm_sched_entity_flush() and by + * drm_sched_fini(). + */ bool stopped; + + /** + * @entity_idle: + * + * Signals when entity is not in use, used to sequence entity cleanup in + * drm_sched_entity_fini(). + */ struct completion entity_idle; }; @@ -169,10 +275,11 @@ struct drm_sched_fence *to_drm_sched_fence(struct dma_fence *f); * struct drm_sched_job - A job to be run by an entity. * * @queue_node: used to append this struct to the queue of jobs in an entity. + * @list: a job participates in a "pending" and "done" lists. * @sched: the scheduler instance on which this job is scheduled. * @s_fence: contains the fences for the scheduling of job. * @finish_cb: the callback for the finished fence. - * @node: used to append this struct to the @drm_gpu_scheduler.ring_mirror_list. + * @work: Helper to reschdeule job kill to different context. * @id: a unique id assigned to each job scheduled on the scheduler. * @karma: increment on every hang caused by this job. If this exceeds the hang * limit of the scheduler then the job is marked guilty and will not @@ -187,34 +294,66 @@ struct drm_sched_fence *to_drm_sched_fence(struct dma_fence *f); */ struct drm_sched_job { struct spsc_node queue_node; + struct list_head list; struct drm_gpu_scheduler *sched; struct drm_sched_fence *s_fence; - struct dma_fence_cb finish_cb; - struct list_head node; + + /* + * work is used only after finish_cb has been used and will not be + * accessed anymore. + */ + union { + struct dma_fence_cb finish_cb; + struct work_struct work; + }; + uint64_t id; atomic_t karma; enum drm_sched_priority s_priority; - struct drm_sched_entity *entity; + struct drm_sched_entity *entity; struct dma_fence_cb cb; + /** + * @dependencies: + * + * Contains the dependencies as struct dma_fence for this job, see + * drm_sched_job_add_dependency() and + * drm_sched_job_add_implicit_dependencies(). + */ + struct xarray dependencies; + + /** @last_dependency: tracks @dependencies as they signal */ + unsigned long last_dependency; }; static inline bool drm_sched_invalidate_job(struct drm_sched_job *s_job, int threshold) { - return (s_job && atomic_inc_return(&s_job->karma) > threshold); + return s_job && atomic_inc_return(&s_job->karma) > threshold; } +enum drm_gpu_sched_stat { + DRM_GPU_SCHED_STAT_NONE, /* Reserve 0 */ + DRM_GPU_SCHED_STAT_NOMINAL, + DRM_GPU_SCHED_STAT_ENODEV, +}; + /** - * struct drm_sched_backend_ops + * struct drm_sched_backend_ops - Define the backend operations + * called by the scheduler * - * Define the backend operations called by the scheduler, - * these functions should be implemented in driver side. + * These functions should be implemented in the driver side. */ struct drm_sched_backend_ops { /** - * @dependency: Called when the scheduler is considering scheduling - * this job next, to get another struct dma_fence for this job to - * block on. Once it returns NULL, run_job() may be called. + * @dependency: + * + * Called when the scheduler is considering scheduling this job next, to + * get another struct dma_fence for this job to block on. Once it + * returns NULL, run_job() may be called. + * + * If a driver exclusively uses drm_sched_job_add_dependency() and + * drm_sched_job_add_implicit_dependencies() this can be ommitted and + * left as NULL. */ struct dma_fence *(*dependency)(struct drm_sched_job *sched_job, struct drm_sched_entity *s_entity); @@ -228,10 +367,48 @@ struct drm_sched_backend_ops { struct dma_fence *(*run_job)(struct drm_sched_job *sched_job); /** - * @timedout_job: Called when a job has taken too long to execute, - * to trigger GPU recovery. + * @timedout_job: Called when a job has taken too long to execute, + * to trigger GPU recovery. + * + * This method is called in a workqueue context. + * + * Drivers typically issue a reset to recover from GPU hangs, and this + * procedure usually follows the following workflow: + * + * 1. Stop the scheduler using drm_sched_stop(). This will park the + * scheduler thread and cancel the timeout work, guaranteeing that + * nothing is queued while we reset the hardware queue + * 2. Try to gracefully stop non-faulty jobs (optional) + * 3. Issue a GPU reset (driver-specific) + * 4. Re-submit jobs using drm_sched_resubmit_jobs() + * 5. Restart the scheduler using drm_sched_start(). At that point, new + * jobs can be queued, and the scheduler thread is unblocked + * + * Note that some GPUs have distinct hardware queues but need to reset + * the GPU globally, which requires extra synchronization between the + * timeout handler of the different &drm_gpu_scheduler. One way to + * achieve this synchronization is to create an ordered workqueue + * (using alloc_ordered_workqueue()) at the driver level, and pass this + * queue to drm_sched_init(), to guarantee that timeout handlers are + * executed sequentially. The above workflow needs to be slightly + * adjusted in that case: + * + * 1. Stop all schedulers impacted by the reset using drm_sched_stop() + * 2. Try to gracefully stop non-faulty jobs on all queues impacted by + * the reset (optional) + * 3. Issue a GPU reset on all faulty queues (driver-specific) + * 4. Re-submit jobs on all schedulers impacted by the reset using + * drm_sched_resubmit_jobs() + * 5. Restart all schedulers that were stopped in step #1 using + * drm_sched_start() + * + * Return DRM_GPU_SCHED_STAT_NOMINAL, when all is normal, + * and the underlying driver has started or completed recovery. + * + * Return DRM_GPU_SCHED_STAT_ENODEV, if the device is no longer + * available, i.e. has been unplugged. */ - void (*timedout_job)(struct drm_sched_job *sched_job); + enum drm_gpu_sched_stat (*timedout_job)(struct drm_sched_job *sched_job); /** * @free_job: Called once the job's finished fence has been signaled @@ -241,7 +418,7 @@ struct drm_sched_backend_ops { }; /** - * struct drm_gpu_scheduler + * struct drm_gpu_scheduler - scheduler instance-specific data * * @ops: backend operations provided by the driver. * @hw_submission_limit: the max size of the hardware queue. @@ -255,16 +432,19 @@ struct drm_sched_backend_ops { * finished. * @hw_rq_count: the number of jobs currently in the hardware queue. * @job_id_count: used to assign unique id to the each job. + * @timeout_wq: workqueue used to queue @work_tdr * @work_tdr: schedules a delayed call to @drm_sched_job_timedout after the * timeout interval is over. * @thread: the kthread on which the scheduler which run. - * @ring_mirror_list: the list of jobs which are currently in the job queue. - * @job_list_lock: lock to protect the ring_mirror_list. + * @pending_list: the list of jobs which are currently in the job queue. + * @job_list_lock: lock to protect the pending_list. * @hang_limit: once the hangs by a job crosses this limit then it is marked - * guilty and it will be considered for scheduling further. + * guilty and it will no longer be considered for scheduling. * @score: score to help loadbalancer pick a idle sched + * @_score: score used when the driver doesn't provide one * @ready: marks if the underlying HW is ready to work * @free_guilty: A hit to time out handler to free the guilty job. + * @dev: system &struct device * * One scheduler is implemented for each hardware ring. */ @@ -273,36 +453,55 @@ struct drm_gpu_scheduler { uint32_t hw_submission_limit; long timeout; const char *name; - struct drm_sched_rq sched_rq[DRM_SCHED_PRIORITY_MAX]; + struct drm_sched_rq sched_rq[DRM_SCHED_PRIORITY_COUNT]; wait_queue_head_t wake_up_worker; wait_queue_head_t job_scheduled; atomic_t hw_rq_count; atomic64_t job_id_count; + struct workqueue_struct *timeout_wq; struct delayed_work work_tdr; struct task_struct *thread; - struct list_head ring_mirror_list; + struct list_head pending_list; spinlock_t job_list_lock; int hang_limit; - atomic_t score; + atomic_t *score; + atomic_t _score; bool ready; bool free_guilty; + struct device *dev; }; int drm_sched_init(struct drm_gpu_scheduler *sched, const struct drm_sched_backend_ops *ops, - uint32_t hw_submission, unsigned hang_limit, long timeout, - const char *name); + uint32_t hw_submission, unsigned hang_limit, + long timeout, struct workqueue_struct *timeout_wq, + atomic_t *score, const char *name, struct device *dev); void drm_sched_fini(struct drm_gpu_scheduler *sched); int drm_sched_job_init(struct drm_sched_job *job, struct drm_sched_entity *entity, void *owner); +void drm_sched_job_arm(struct drm_sched_job *job); +int drm_sched_job_add_dependency(struct drm_sched_job *job, + struct dma_fence *fence); +int drm_sched_job_add_implicit_dependencies(struct drm_sched_job *job, + struct drm_gem_object *obj, + bool write); + + +void drm_sched_entity_modify_sched(struct drm_sched_entity *entity, + struct drm_gpu_scheduler **sched_list, + unsigned int num_sched_list); + void drm_sched_job_cleanup(struct drm_sched_job *job); void drm_sched_wakeup(struct drm_gpu_scheduler *sched); void drm_sched_stop(struct drm_gpu_scheduler *sched, struct drm_sched_job *bad); void drm_sched_start(struct drm_gpu_scheduler *sched, bool full_recovery); void drm_sched_resubmit_jobs(struct drm_gpu_scheduler *sched); +void drm_sched_resubmit_jobs_ext(struct drm_gpu_scheduler *sched, int max); void drm_sched_increase_karma(struct drm_sched_job *bad); +void drm_sched_reset_karma(struct drm_sched_job *bad); +void drm_sched_increase_karma_ext(struct drm_sched_job *bad, int type); bool drm_sched_dependency_optimized(struct dma_fence* fence, struct drm_sched_entity *entity); void drm_sched_fault(struct drm_gpu_scheduler *sched); @@ -323,19 +522,25 @@ void drm_sched_entity_fini(struct drm_sched_entity *entity); void drm_sched_entity_destroy(struct drm_sched_entity *entity); void drm_sched_entity_select_rq(struct drm_sched_entity *entity); struct drm_sched_job *drm_sched_entity_pop_job(struct drm_sched_entity *entity); -void drm_sched_entity_push_job(struct drm_sched_job *sched_job, - struct drm_sched_entity *entity); +void drm_sched_entity_push_job(struct drm_sched_job *sched_job); void drm_sched_entity_set_priority(struct drm_sched_entity *entity, enum drm_sched_priority priority); bool drm_sched_entity_is_ready(struct drm_sched_entity *entity); -struct drm_sched_fence *drm_sched_fence_create( +struct drm_sched_fence *drm_sched_fence_alloc( struct drm_sched_entity *s_entity, void *owner); +void drm_sched_fence_init(struct drm_sched_fence *fence, + struct drm_sched_entity *entity); +void drm_sched_fence_free(struct drm_sched_fence *fence); + void drm_sched_fence_scheduled(struct drm_sched_fence *fence); void drm_sched_fence_finished(struct drm_sched_fence *fence); unsigned long drm_sched_suspend_timeout(struct drm_gpu_scheduler *sched); void drm_sched_resume_timeout(struct drm_gpu_scheduler *sched, unsigned long remaining); +struct drm_gpu_scheduler * +drm_sched_pick_best(struct drm_gpu_scheduler **sched_list, + unsigned int num_sched_list); #endif |