[PATCH 2/8] drm/amdgpu: add framework for HW specific priority settings v8

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Am 09.06.2017 um 00:06 schrieb Andres Rodriguez:
> Add an initial framework for changing the HW priorities of rings. The
> framework allows requesting priority changes for the lifetime of an
> amdgpu_job. After the job completes the priority will decay to the next
> lowest priority for which a request is still valid.
>
> A new ring function set_priority() can now be populated to take care of
> the HW specific programming sequence for priority changes.
>
> v2: set priority before emitting IB, and take a ref on amdgpu_job
> v3: use AMD_SCHED_PRIORITY_* instead of AMDGPU_CTX_PRIORITY_*
> v4: plug amdgpu_ring_restore_priority_cb into amdgpu_job_free_cb
> v5: use atomic for tracking job priorities instead of last_job
> v6: rename amdgpu_ring_priority_[get/put]() and align parameters
> v7: replace spinlocks with mutexes for KIQ compatibility
> v8: raise ring priority during cs_ioctl, instead of job_run
>
> Signed-off-by: Andres Rodriguez <andresx7 at gmail.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c        |  2 +
>   drivers/gpu/drm/amd/amdgpu/amdgpu_job.c       |  4 ++
>   drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c      | 76 ++++++++++++++++++++++++++-
>   drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h      | 15 ++++++
>   drivers/gpu/drm/amd/scheduler/gpu_scheduler.h |  7 +++
>   5 files changed, 103 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
> index b17635c..d84d026 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
> @@ -996,30 +996,32 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
>   	if (r) {
>   		amdgpu_job_free(job);
>   		return r;
>   	}
>   
>   	job->owner = p->filp;
>   	job->fence_ctx = entity->fence_context;
>   	p->fence = dma_fence_get(&job->base.s_fence->finished);
>   	cs->out.handle = amdgpu_ctx_add_fence(p->ctx, ring, p->fence);
>   	job->uf_sequence = cs->out.handle;
>   	amdgpu_job_free_resources(job);
>   	amdgpu_cs_parser_fini(p, 0, true);
>   
>   	trace_amdgpu_cs_ioctl(job);
>   	amd_sched_entity_push_job(&job->base);
> +	amdgpu_ring_priority_get(job->ring,
> +				 amd_sched_get_job_priority(&job->base));

After the call to amd_sched_entity_push_job() we no longer own the 
pointer, so the job could be already be freed when you dereference it.

I suggest to just move the call to amdgpu_ring_priority_get() before the 
one to amd_sched_entity_push_job(), cause the whole thing is racy in the 
first place.

Christian.

>   
>   	return 0;
>   }
>   
>   int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
>   {
>   	struct amdgpu_device *adev = dev->dev_private;
>   	struct amdgpu_fpriv *fpriv = filp->driver_priv;
>   	union drm_amdgpu_cs *cs = data;
>   	struct amdgpu_cs_parser parser = {};
>   	bool reserved_buffers = false;
>   	int i, r;
>   
>   	if (!adev->accel_working)
>   		return -EBUSY;
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
> index 3d641e1..f10ce0d 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
> @@ -89,30 +89,31 @@ void amdgpu_job_free_resources(struct amdgpu_job *job)
>   {
>   	struct dma_fence *f;
>   	unsigned i;
>   
>   	/* use sched fence if available */
>   	f = job->base.s_fence ? &job->base.s_fence->finished : job->fence;
>   
>   	for (i = 0; i < job->num_ibs; ++i)
>   		amdgpu_ib_free(job->adev, &job->ibs[i], f);
>   }
>   
>   static void amdgpu_job_free_cb(struct amd_sched_job *s_job)
>   {
>   	struct amdgpu_job *job = container_of(s_job, struct amdgpu_job, base);
>   
> +	amdgpu_ring_priority_put(job->ring, amd_sched_get_job_priority(s_job));
>   	dma_fence_put(job->fence);
>   	amdgpu_sync_free(&job->sync);
>   	amdgpu_sync_free(&job->dep_sync);
>   	amdgpu_sync_free(&job->sched_sync);
>   	kfree(job);
>   }
>   
>   void amdgpu_job_free(struct amdgpu_job *job)
>   {
>   	amdgpu_job_free_resources(job);
>   
>   	dma_fence_put(job->fence);
>   	amdgpu_sync_free(&job->sync);
>   	amdgpu_sync_free(&job->dep_sync);
>   	amdgpu_sync_free(&job->sched_sync);
> @@ -126,30 +127,32 @@ int amdgpu_job_submit(struct amdgpu_job *job, struct amdgpu_ring *ring,
>   	int r;
>   	job->ring = ring;
>   
>   	if (!f)
>   		return -EINVAL;
>   
>   	r = amd_sched_job_init(&job->base, &ring->sched, entity, owner);
>   	if (r)
>   		return r;
>   
>   	job->owner = owner;
>   	job->fence_ctx = entity->fence_context;
>   	*f = dma_fence_get(&job->base.s_fence->finished);
>   	amdgpu_job_free_resources(job);
>   	amd_sched_entity_push_job(&job->base);
> +	amdgpu_ring_priority_get(job->ring,
> +				 amd_sched_get_job_priority(&job->base));
>   
>   	return 0;
>   }
>   
>   static struct dma_fence *amdgpu_job_dependency(struct amd_sched_job *sched_job)
>   {
>   	struct amdgpu_job *job = to_amdgpu_job(sched_job);
>   	struct amdgpu_vm *vm = job->vm;
>   
>   	struct dma_fence *fence = amdgpu_sync_get_fence(&job->dep_sync);
>   	int r;
>   
>   	if (amd_sched_dependency_optimized(fence, sched_job->s_entity)) {
>   		r = amdgpu_sync_fence(job->adev, &job->sched_sync, fence);
>   		if (r)
> @@ -189,25 +192,26 @@ static struct dma_fence *amdgpu_job_run(struct amd_sched_job *sched_job)
>   
>   	trace_amdgpu_sched_run_job(job);
>   	if (job->vm)
>   		fpriv = container_of(job->vm, struct amdgpu_fpriv, vm);
>   	/* skip ib schedule when vram is lost */
>   	if (fpriv && amdgpu_kms_vram_lost(job->adev, fpriv))
>   		DRM_ERROR("Skip scheduling IBs!\n");
>   	else {
>   		r = amdgpu_ib_schedule(job->ring, job->num_ibs, job->ibs, job, &fence);
>   		if (r)
>   			DRM_ERROR("Error scheduling IBs (%d)\n", r);
>   	}
>   	/* if gpu reset, hw fence will be replaced here */
>   	dma_fence_put(job->fence);
>   	job->fence = dma_fence_get(fence);
> +
>   	amdgpu_job_free_resources(job);
>   	return fence;
>   }
>   
>   const struct amd_sched_backend_ops amdgpu_sched_ops = {
>   	.dependency = amdgpu_job_dependency,
>   	.run_job = amdgpu_job_run,
>   	.timedout_job = amdgpu_job_timedout,
>   	.free_job = amdgpu_job_free_cb
>   };
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c
> index 75165e0..2d8b20a7 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c
> @@ -143,45 +143,114 @@ void amdgpu_ring_commit(struct amdgpu_ring *ring)
>    * amdgpu_ring_undo - reset the wptr
>    *
>    * @ring: amdgpu_ring structure holding ring information
>    *
>    * Reset the driver's copy of the wptr (all asics).
>    */
>   void amdgpu_ring_undo(struct amdgpu_ring *ring)
>   {
>   	ring->wptr = ring->wptr_old;
>   
>   	if (ring->funcs->end_use)
>   		ring->funcs->end_use(ring);
>   }
>   
>   /**
> + * amdgpu_ring_priority_put - restore a ring's priority
> + *
> + * @ring: amdgpu_ring structure holding the information
> + * @priority: target priority
> + *
> + * Release a request for executing at @priority
> + */
> +void amdgpu_ring_priority_put(struct amdgpu_ring *ring,
> +			      enum amd_sched_priority priority)
> +{
> +	int i;
> +
> +	if (!ring->funcs->set_priority)
> +		return;
> +
> +	if (atomic_dec_return(&ring->num_jobs[priority]) > 0)
> +		return;
> +
> +	/* no need to restore if the job is already at the lowest priority */
> +	if (priority == AMD_SCHED_PRIORITY_NORMAL)
> +		return;
> +
> +	mutex_lock(&ring->priority_mutex);
> +	/* something higher prio is executing, no need to decay */
> +	if (ring->priority > priority)
> +		goto out_unlock;
> +
> +	/* decay priority to the next level with a job available */
> +	for (i = priority; i >= AMD_SCHED_PRIORITY_MIN; i--) {
> +		if (i == AMD_SCHED_PRIORITY_NORMAL
> +				|| atomic_read(&ring->num_jobs[i])) {
> +			ring->priority = i;
> +			ring->funcs->set_priority(ring, i);
> +			break;
> +		}
> +	}
> +
> +out_unlock:
> +	mutex_unlock(&ring->priority_mutex);
> +}
> +
> +/**
> + * amdgpu_ring_priority_get - change the ring's priority
> + *
> + * @ring: amdgpu_ring structure holding the information
> + * @priority: target priority
> + *
> + * Request a ring's priority to be raised to @priority (refcounted).
> + */
> +void amdgpu_ring_priority_get(struct amdgpu_ring *ring,
> +			      enum amd_sched_priority priority)
> +{
> +	if (!ring->funcs->set_priority)
> +		return;
> +
> +	atomic_inc(&ring->num_jobs[priority]);
> +
> +	mutex_lock(&ring->priority_mutex);
> +	if (priority <= ring->priority)
> +		goto out_unlock;
> +
> +	ring->priority = priority;
> +	ring->funcs->set_priority(ring, priority);
> +
> +out_unlock:
> +	mutex_unlock(&ring->priority_mutex);
> +}
> +
> +/**
>    * amdgpu_ring_init - init driver ring struct.
>    *
>    * @adev: amdgpu_device pointer
>    * @ring: amdgpu_ring structure holding ring information
>    * @max_ndw: maximum number of dw for ring alloc
>    * @nop: nop packet for this ring
>    *
>    * Initialize the driver information for the selected ring (all asics).
>    * Returns 0 on success, error on failure.
>    */
>   int amdgpu_ring_init(struct amdgpu_device *adev, struct amdgpu_ring *ring,
>   		     unsigned max_dw, struct amdgpu_irq_src *irq_src,
>   		     unsigned irq_type)
>   {
> -	int r;
> +	int r, i;
>   
>   	if (ring->adev == NULL) {
>   		if (adev->num_rings >= AMDGPU_MAX_RINGS)
>   			return -EINVAL;
>   
>   		ring->adev = adev;
>   		ring->idx = adev->num_rings++;
>   		adev->rings[ring->idx] = ring;
>   		r = amdgpu_fence_driver_init_ring(ring,
>   			amdgpu_sched_hw_submission);
>   		if (r)
>   			return r;
>   	}
>   
>   	if (ring->funcs->support_64bit_ptrs) {
> @@ -243,33 +312,38 @@ int amdgpu_ring_init(struct amdgpu_device *adev, struct amdgpu_ring *ring,
>   	/* Allocate ring buffer */
>   	if (ring->ring_obj == NULL) {
>   		r = amdgpu_bo_create_kernel(adev, ring->ring_size, PAGE_SIZE,
>   					    AMDGPU_GEM_DOMAIN_GTT,
>   					    &ring->ring_obj,
>   					    &ring->gpu_addr,
>   					    (void **)&ring->ring);
>   		if (r) {
>   			dev_err(adev->dev, "(%d) ring create failed\n", r);
>   			return r;
>   		}
>   		amdgpu_ring_clear_ring(ring);
>   	}
>   
>   	ring->max_dw = max_dw;
> +	ring->priority = AMD_SCHED_PRIORITY_NORMAL;
> +	mutex_init(&ring->priority_mutex);
>   	INIT_LIST_HEAD(&ring->lru_list);
>   	amdgpu_ring_lru_touch(adev, ring);
>   
> +	for (i = 0; i < AMD_SCHED_PRIORITY_MAX; ++i)
> +		atomic_set(&ring->num_jobs[i], 0);
> +
>   	if (amdgpu_debugfs_ring_init(adev, ring)) {
>   		DRM_ERROR("Failed to register debugfs file for rings !\n");
>   	}
>   
>   	return 0;
>   }
>   
>   /**
>    * amdgpu_ring_fini - tear down the driver ring struct.
>    *
>    * @adev: amdgpu_device pointer
>    * @ring: amdgpu_ring structure holding ring information
>    *
>    * Tear down the driver information for the selected ring (all asics).
>    */
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
> index bc8dec9..7348769 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
> @@ -12,30 +12,31 @@
>    * all copies or substantial portions of the Software.
>    *
>    * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
>    * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
>    * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
>    * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
>    * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
>    * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
>    * OTHER DEALINGS IN THE SOFTWARE.
>    *
>    * Authors: Christian König
>    */
>   #ifndef __AMDGPU_RING_H__
>   #define __AMDGPU_RING_H__
>   
> +#include <drm/amdgpu_drm.h>
>   #include "gpu_scheduler.h"
>   
>   /* max number of rings */
>   #define AMDGPU_MAX_RINGS		18
>   #define AMDGPU_MAX_GFX_RINGS		1
>   #define AMDGPU_MAX_COMPUTE_RINGS	8
>   #define AMDGPU_MAX_VCE_RINGS		3
>   #define AMDGPU_MAX_UVD_ENC_RINGS	2
>   
>   /* some special values for the owner field */
>   #define AMDGPU_FENCE_OWNER_UNDEFINED	((void*)0ul)
>   #define AMDGPU_FENCE_OWNER_VM		((void*)1ul)
>   
>   #define AMDGPU_FENCE_FLAG_64BIT         (1 << 0)
>   #define AMDGPU_FENCE_FLAG_INT           (1 << 1)
> @@ -44,30 +45,31 @@ enum amdgpu_ring_type {
>   	AMDGPU_RING_TYPE_GFX,
>   	AMDGPU_RING_TYPE_COMPUTE,
>   	AMDGPU_RING_TYPE_SDMA,
>   	AMDGPU_RING_TYPE_UVD,
>   	AMDGPU_RING_TYPE_VCE,
>   	AMDGPU_RING_TYPE_KIQ,
>   	AMDGPU_RING_TYPE_UVD_ENC,
>   	AMDGPU_RING_TYPE_VCN_DEC,
>   	AMDGPU_RING_TYPE_VCN_ENC
>   };
>   
>   struct amdgpu_device;
>   struct amdgpu_ring;
>   struct amdgpu_ib;
>   struct amdgpu_cs_parser;
> +struct amdgpu_job;
>   
>   /*
>    * Fences.
>    */
>   struct amdgpu_fence_driver {
>   	uint64_t			gpu_addr;
>   	volatile uint32_t		*cpu_addr;
>   	/* sync_seq is protected by ring emission lock */
>   	uint32_t			sync_seq;
>   	atomic_t			last_seq;
>   	bool				initialized;
>   	struct amdgpu_irq_src		*irq_src;
>   	unsigned			irq_type;
>   	struct timer_list		fallback_timer;
>   	unsigned			num_fences_mask;
> @@ -135,30 +137,33 @@ struct amdgpu_ring_funcs {
>   	void (*insert_nop)(struct amdgpu_ring *ring, uint32_t count);
>   	void (*insert_start)(struct amdgpu_ring *ring);
>   	void (*insert_end)(struct amdgpu_ring *ring);
>   	/* pad the indirect buffer to the necessary number of dw */
>   	void (*pad_ib)(struct amdgpu_ring *ring, struct amdgpu_ib *ib);
>   	unsigned (*init_cond_exec)(struct amdgpu_ring *ring);
>   	void (*patch_cond_exec)(struct amdgpu_ring *ring, unsigned offset);
>   	/* note usage for clock and power gating */
>   	void (*begin_use)(struct amdgpu_ring *ring);
>   	void (*end_use)(struct amdgpu_ring *ring);
>   	void (*emit_switch_buffer) (struct amdgpu_ring *ring);
>   	void (*emit_cntxcntl) (struct amdgpu_ring *ring, uint32_t flags);
>   	void (*emit_rreg)(struct amdgpu_ring *ring, uint32_t reg);
>   	void (*emit_wreg)(struct amdgpu_ring *ring, uint32_t reg, uint32_t val);
>   	void (*emit_tmz)(struct amdgpu_ring *ring, bool start);
> +	/* priority functions */
> +	void (*set_priority) (struct amdgpu_ring *ring,
> +			      enum amd_sched_priority priority);
>   };
>   
>   struct amdgpu_ring {
>   	struct amdgpu_device		*adev;
>   	const struct amdgpu_ring_funcs	*funcs;
>   	struct amdgpu_fence_driver	fence_drv;
>   	struct amd_gpu_scheduler	sched;
>   	struct list_head		lru_list;
>   
>   	struct amdgpu_bo	*ring_obj;
>   	volatile uint32_t	*ring;
>   	unsigned		rptr_offs;
>   	u64			wptr;
>   	u64			wptr_old;
>   	unsigned		ring_size;
> @@ -175,40 +180,50 @@ struct amdgpu_ring {
>   	struct amdgpu_bo	*mqd_obj;
>   	uint64_t                mqd_gpu_addr;
>   	void                    *mqd_ptr;
>   	uint64_t                eop_gpu_addr;
>   	u32			doorbell_index;
>   	bool			use_doorbell;
>   	unsigned		wptr_offs;
>   	unsigned		fence_offs;
>   	uint64_t		current_ctx;
>   	char			name[16];
>   	unsigned		cond_exe_offs;
>   	u64			cond_exe_gpu_addr;
>   	volatile u32		*cond_exe_cpu_addr;
>   	unsigned		vm_inv_eng;
>   	bool			has_compute_vm_bug;
> +
> +	atomic_t		num_jobs[AMD_SCHED_PRIORITY_MAX];
> +	struct mutex		priority_mutex;
> +	/* protected by priority_mutex */
> +	int			priority;
> +
>   #if defined(CONFIG_DEBUG_FS)
>   	struct dentry *ent;
>   #endif
>   };
>   
>   int amdgpu_ring_alloc(struct amdgpu_ring *ring, unsigned ndw);
>   void amdgpu_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count);
>   void amdgpu_ring_generic_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib);
>   void amdgpu_ring_commit(struct amdgpu_ring *ring);
>   void amdgpu_ring_undo(struct amdgpu_ring *ring);
> +void amdgpu_ring_priority_get(struct amdgpu_ring *ring,
> +			      enum amd_sched_priority priority);
> +void amdgpu_ring_priority_put(struct amdgpu_ring *ring,
> +			      enum amd_sched_priority priority);
>   int amdgpu_ring_init(struct amdgpu_device *adev, struct amdgpu_ring *ring,
>   		     unsigned ring_size, struct amdgpu_irq_src *irq_src,
>   		     unsigned irq_type);
>   void amdgpu_ring_fini(struct amdgpu_ring *ring);
>   int amdgpu_ring_lru_get(struct amdgpu_device *adev, int type, int *blacklist,
>   			int num_blacklist, struct amdgpu_ring **ring);
>   void amdgpu_ring_lru_touch(struct amdgpu_device *adev, struct amdgpu_ring *ring);
>   static inline void amdgpu_ring_clear_ring(struct amdgpu_ring *ring)
>   {
>   	int i = 0;
>   	while (i <= ring->buf_mask)
>   		ring->ring[i++] = ring->funcs->nop;
>   
>   }
>   
> diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
> index 38e622c..dbcaa2e 100644
> --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
> +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
> @@ -158,16 +158,23 @@ int amd_sched_fence_slab_init(void);
>   void amd_sched_fence_slab_fini(void);
>   
>   struct amd_sched_fence *amd_sched_fence_create(
>   	struct amd_sched_entity *s_entity, void *owner);
>   void amd_sched_fence_scheduled(struct amd_sched_fence *fence);
>   void amd_sched_fence_finished(struct amd_sched_fence *fence);
>   int amd_sched_job_init(struct amd_sched_job *job,
>   		       struct amd_gpu_scheduler *sched,
>   		       struct amd_sched_entity *entity,
>   		       void *owner);
>   void amd_sched_hw_job_reset(struct amd_gpu_scheduler *sched);
>   void amd_sched_job_recovery(struct amd_gpu_scheduler *sched);
>   bool amd_sched_dependency_optimized(struct dma_fence* fence,
>   				    struct amd_sched_entity *entity);
>   void amd_sched_job_kickout(struct amd_sched_job *s_job);
> +
> +static inline enum amd_sched_priority
> +amd_sched_get_job_priority(struct amd_sched_job *job)
> +{
> +	return (job->s_entity->rq - job->sched->sched_rq);
> +}
> +
>   #endif




[Index of Archives]     [Linux USB Devel]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux