[PATCH 0/8] allow DRM_MASTER to change client's priorities v3

[PATCH 0/8] allow DRM_MASTER to change client's priorities v3

[Andres Rodriguez]

Just a rebase of v2 since it went a little stale.

Series available in the wip-process-priorities-v3 branch at:
git://people.freedesktop.org/~lostgoat/linux

Gpuvis trace with normal priority running steamvr, hellovr_vulkan and ssao
demo:
https://drive.google.com/open?id=0B2ygSoZuj3IMdFJ4bTZPcmh5aHc

Gpuvis trace with compute at high priority:
https://drive.google.com/open?id=0B2ygSoZuj3IMYzBFUUN0dXV4bGs

Full trace data vailable here:
https://drive.google.com/open?id=0B2ygSoZuj3IMWGUzV25yNGtyRFU

For validation, use the wip-process-priorities-v3 branch of:
mesa: git@github.com:lostgoat/mesa.git
libdrm: git@github.com:lostgoat/libdrm.git
vk-demos: git@github.com:lostgoat/Vulkan.git

Run the following two vk-demos simultaneously:
./ssao
./computeparticles -high-priority

Andres Rodriguez (8):
  drm/amdgpu: add parameter to allocate high priority contexts v11
  drm/amdgpu: add framework for HW specific priority settings v9
  drm/amdgpu: implement ring set_priority for gfx_v8 compute v9
  drm/amdgpu: make amdgpu_to_sched_priority detect invalid parameters
  drm/amd/sched: allow clients to edit an entity's rq v2
  drm/amdgpu: introduce AMDGPU_CTX_PRIORITY_UNSET
  drm/amdgpu: add plumbing for ctx priority changes v2
  drm/amdgpu: add interface for editing a foreign process's priority v3

 drivers/gpu/drm/amd/amdgpu/Makefile           |   2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu.h           |  10 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c        |   2 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c       |  75 +++++++++++++++++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c    |   2 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_job.c       |   4 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c       |   2 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c      |  76 +++++++++++++++++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h      |  15 ++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c     | 109 ++++++++++++++++++++++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_sched.h     |  34 ++++++++
 drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c         |  99 +++++++++++++++++++++++
 drivers/gpu/drm/amd/scheduler/gpu_scheduler.c |  24 +++++-
 drivers/gpu/drm/amd/scheduler/gpu_scheduler.h |  19 ++++-
 include/uapi/drm/amdgpu_drm.h                 |  28 ++++++-
 15 files changed, 489 insertions(+), 12 deletions(-)
 create mode 100644 drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
 create mode 100644 drivers/gpu/drm/amd/amdgpu/amdgpu_sched.h

-- 
2.9.3

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 1/8] drm/amdgpu: add parameter to allocate high priority contexts v11

[Andres Rodriguez]

Add a new context creation parameter to express a global context priority.

The priority ranking in descending order is as follows:
 * AMDGPU_CTX_PRIORITY_HIGH_HW
 * AMDGPU_CTX_PRIORITY_HIGH_SW
 * AMDGPU_CTX_PRIORITY_NORMAL
 * AMDGPU_CTX_PRIORITY_LOW_SW
 * AMDGPU_CTX_PRIORITY_LOW_HW

The driver will attempt to schedule work to the hardware according to
the priorities. No latency or throughput guarantees are provided by
this patch.

This interface intends to service the EGL_IMG_context_priority
extension, and vulkan equivalents.

Setting a priority above NORMAL requires CAP_SYS_NICE or DRM_MASTER.

v2: Instead of using flags, repurpose __pad
v3: Swap enum values of _NORMAL _HIGH for backwards compatibility
v4: Validate usermode priority and store it
v5: Move priority validation into amdgpu_ctx_ioctl(), headline reword
v6: add UAPI note regarding priorities requiring CAP_SYS_ADMIN
v7: remove ctx->priority
v8: added AMDGPU_CTX_PRIORITY_LOW, s/CAP_SYS_ADMIN/CAP_SYS_NICE
v9: change the priority parameter to __s32
v10: split priorities into _SW and _HW
v11: Allow DRM_MASTER without CAP_SYS_NICE

Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Andres Rodriguez <andresx7@gmail.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c       | 61 +++++++++++++++++++++++++--
 drivers/gpu/drm/amd/scheduler/gpu_scheduler.h |  5 ++-
 include/uapi/drm/amdgpu_drm.h                 | 10 ++++-
 3 files changed, 70 insertions(+), 6 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
index a11e443..9ec85d5 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
@@ -23,13 +23,40 @@
  */
 
 #include <drm/drmP.h>
+#include <drm/drm_auth.h>
 #include "amdgpu.h"
 
-static int amdgpu_ctx_init(struct amdgpu_device *adev, struct amdgpu_ctx *ctx)
+static int amdgpu_ctx_priority_permit(struct drm_file *filp,
+				      enum amd_sched_priority priority)
+{
+	/* NORMAL and below are accessible by everyone */
+	if (priority <= AMD_SCHED_PRIORITY_NORMAL)
+		return 0;
+
+	if (capable(CAP_SYS_NICE))
+		return 0;
+
+	if (drm_is_current_master(filp))
+		return 0;
+
+	return -EACCES;
+}
+
+static int amdgpu_ctx_init(struct amdgpu_device *adev,
+			   enum amd_sched_priority priority,
+			   struct drm_file *filp,
+			   struct amdgpu_ctx *ctx)
 {
 	unsigned i, j;
 	int r;
 
+	if (priority < 0 || priority >= AMD_SCHED_PRIORITY_MAX)
+		return -EINVAL;
+
+	r = amdgpu_ctx_priority_permit(filp, priority);
+	if (r)
+		return r;
+
 	memset(ctx, 0, sizeof(*ctx));
 	ctx->adev = adev;
 	kref_init(&ctx->refcount);
@@ -51,7 +78,7 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev, struct amdgpu_ctx *ctx)
 		struct amdgpu_ring *ring = adev->rings[i];
 		struct amd_sched_rq *rq;
 
-		rq = &ring->sched.sched_rq[AMD_SCHED_PRIORITY_NORMAL];
+		rq = &ring->sched.sched_rq[priority];
 
 		if (ring == &adev->gfx.kiq.ring)
 			continue;
@@ -100,6 +127,8 @@ static void amdgpu_ctx_fini(struct amdgpu_ctx *ctx)
 
 static int amdgpu_ctx_alloc(struct amdgpu_device *adev,
 			    struct amdgpu_fpriv *fpriv,
+			    struct drm_file *filp,
+			    enum amd_sched_priority priority,
 			    uint32_t *id)
 {
 	struct amdgpu_ctx_mgr *mgr = &fpriv->ctx_mgr;
@@ -117,8 +146,9 @@ static int amdgpu_ctx_alloc(struct amdgpu_device *adev,
 		kfree(ctx);
 		return r;
 	}
+
 	*id = (uint32_t)r;
-	r = amdgpu_ctx_init(adev, ctx);
+	r = amdgpu_ctx_init(adev, priority, filp, ctx);
 	if (r) {
 		idr_remove(&mgr->ctx_handles, *id);
 		*id = 0;
@@ -188,11 +218,30 @@ static int amdgpu_ctx_query(struct amdgpu_device *adev,
 	return 0;
 }
 
+static enum amd_sched_priority amdgpu_to_sched_priority(int amdgpu_priority)
+{
+	switch (amdgpu_priority) {
+	case AMDGPU_CTX_PRIORITY_HIGH_HW:
+		return AMD_SCHED_PRIORITY_HIGH_HW;
+	case AMDGPU_CTX_PRIORITY_HIGH_SW:
+		return AMD_SCHED_PRIORITY_HIGH_SW;
+	case AMDGPU_CTX_PRIORITY_NORMAL:
+		return AMD_SCHED_PRIORITY_NORMAL;
+	case AMDGPU_CTX_PRIORITY_LOW_SW:
+	case AMDGPU_CTX_PRIORITY_LOW_HW:
+		return AMD_SCHED_PRIORITY_LOW;
+	default:
+		WARN(1, "Invalid context priority %d\n", amdgpu_priority);
+		return AMD_SCHED_PRIORITY_NORMAL;
+	}
+}
+
 int amdgpu_ctx_ioctl(struct drm_device *dev, void *data,
 		     struct drm_file *filp)
 {
 	int r;
 	uint32_t id;
+	enum amd_sched_priority priority;
 
 	union drm_amdgpu_ctx *args = data;
 	struct amdgpu_device *adev = dev->dev_private;
@@ -200,10 +249,14 @@ int amdgpu_ctx_ioctl(struct drm_device *dev, void *data,
 
 	r = 0;
 	id = args->in.ctx_id;
+	priority = amdgpu_to_sched_priority(args->in.priority);
+
+	if (priority >= AMD_SCHED_PRIORITY_MAX)
+		return -EINVAL;
 
 	switch (args->in.op) {
 	case AMDGPU_CTX_OP_ALLOC_CTX:
-		r = amdgpu_ctx_alloc(adev, fpriv, &id);
+		r = amdgpu_ctx_alloc(adev, fpriv, filp, priority, &id);
 		args->out.alloc.ctx_id = id;
 		break;
 	case AMDGPU_CTX_OP_FREE_CTX:
diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
index f9d8f28..38e622c 100644
--- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
+++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
@@ -115,7 +115,10 @@ struct amd_sched_backend_ops {
 
 enum amd_sched_priority {
 	AMD_SCHED_PRIORITY_MIN,
-	AMD_SCHED_PRIORITY_NORMAL = AMD_SCHED_PRIORITY_MIN,
+	AMD_SCHED_PRIORITY_LOW = AMD_SCHED_PRIORITY_MIN,
+	AMD_SCHED_PRIORITY_NORMAL,
+	AMD_SCHED_PRIORITY_HIGH_SW,
+	AMD_SCHED_PRIORITY_HIGH_HW,
 	AMD_SCHED_PRIORITY_KERNEL,
 	AMD_SCHED_PRIORITY_MAX
 };
diff --git a/include/uapi/drm/amdgpu_drm.h b/include/uapi/drm/amdgpu_drm.h
index d59e6a4..d93e988 100644
--- a/include/uapi/drm/amdgpu_drm.h
+++ b/include/uapi/drm/amdgpu_drm.h
@@ -166,13 +166,21 @@ union drm_amdgpu_bo_list {
 /* unknown cause */
 #define AMDGPU_CTX_UNKNOWN_RESET	3
 
+/* Context priority level */
+#define AMDGPU_CTX_PRIORITY_LOW_HW      -1023
+#define AMDGPU_CTX_PRIORITY_LOW_SW      -512
+#define AMDGPU_CTX_PRIORITY_NORMAL      0
+/* Selecting a priority above NORMAL requires CAP_SYS_NICE or DRM_MASTER */
+#define AMDGPU_CTX_PRIORITY_HIGH_SW     512
+#define AMDGPU_CTX_PRIORITY_HIGH_HW     1023
+
 struct drm_amdgpu_ctx_in {
 	/** AMDGPU_CTX_OP_* */
 	__u32	op;
 	/** For future use, no flags defined so far */
 	__u32	flags;
 	__u32	ctx_id;
-	__u32	_pad;
+	__s32	priority;
 };
 
 union drm_amdgpu_ctx_out {
-- 
2.9.3

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 2/8] drm/amdgpu: add framework for HW specific priority settings v9

[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
v9: priority_get() before push_job()

Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Andres Rodriguez <andresx7@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 5840d07..bc8a403 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -1164,6 +1164,8 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
 	cs->out.handle = amdgpu_ctx_add_fence(p->ctx, ring, p->fence);
 	job->uf_sequence = cs->out.handle;
 	amdgpu_job_free_resources(job);
+	amdgpu_ring_priority_get(job->ring,
+				 amd_sched_get_job_priority(&job->base));
 
 	trace_amdgpu_cs_ioctl(job);
 	amd_sched_entity_push_job(&job->base);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
index 4510627..83d1343 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c
@@ -103,6 +103,7 @@ 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);
@@ -139,6 +140,8 @@ int amdgpu_job_submit(struct amdgpu_job *job, struct amdgpu_ring *ring,
 	job->fence_ctx = entity->fence_context;
 	*f = dma_fence_get(&job->base.s_fence->finished);
 	amdgpu_job_free_resources(job);
+	amdgpu_ring_priority_get(job->ring,
+				 amd_sched_get_job_priority(&job->base));
 	amd_sched_entity_push_job(&job->base);
 
 	return 0;
@@ -203,6 +206,7 @@ static struct dma_fence *amdgpu_job_run(struct amd_sched_job *sched_job)
 	/* 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;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c
index 5ce6528..331ed8a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c
@@ -155,6 +155,75 @@ void amdgpu_ring_undo(struct amdgpu_ring *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
@@ -169,7 +238,7 @@ 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;
 	int sched_hw_submission = amdgpu_sched_hw_submission;
 
 	/* Set the hw submission limit higher for KIQ because
@@ -247,9 +316,14 @@ int amdgpu_ring_init(struct amdgpu_device *adev, struct amdgpu_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");
 	}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
index 322d2529..a76c3629 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
@@ -24,6 +24,7 @@
 #ifndef __AMDGPU_RING_H__
 #define __AMDGPU_RING_H__
 
+#include <drm/amdgpu_drm.h>
 #include "gpu_scheduler.h"
 
 /* max number of rings */
@@ -56,6 +57,7 @@ struct amdgpu_device;
 struct amdgpu_ring;
 struct amdgpu_ib;
 struct amdgpu_cs_parser;
+struct amdgpu_job;
 
 /*
  * Fences.
@@ -147,6 +149,9 @@ struct amdgpu_ring_funcs {
 	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 {
@@ -187,6 +192,12 @@ struct amdgpu_ring {
 	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
@@ -197,6 +208,10 @@ 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);
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
@@ -170,4 +170,11 @@ 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
-- 
2.9.3

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 3/8] drm/amdgpu: implement ring set_priority for gfx_v8 compute v9

[Andres Rodriguez]

Programming CP_HQD_QUEUE_PRIORITY enables a queue to take priority over
other queues on the same pipe. Multiple queues on a pipe are timesliced
so this gives us full precedence over other queues.

Programming CP_HQD_PIPE_PRIORITY changes the SPI_ARB_PRIORITY of the
wave as follows:
        0x2: CS_H
        0x1: CS_M
        0x0: CS_L

The SPI block will then dispatch work according to the policy set by
SPI_ARB_PRIORITY. In the current policy CS_H is higher priority than
gfx.

In order to prevent getting stuck in loops of resources bouncing between
GFX and high priority compute and introducing further latency, we
statically reserve a portion of the pipe.

v2: fix srbm_select to ring->queue and use ring->funcs->type
v3: use AMD_SCHED_PRIORITY_* instead of AMDGPU_CTX_PRIORITY_*
v4: switch int to enum amd_sched_priority
v5: corresponding changes for srbm_lock
v6: change CU reservation to PIPE_PERCENT allocation
v7: use kiq instead of MMIO
v8: back to MMIO, and make the implementation sleep safe.
v9: corresponding changes for splitting HIGH into _HW/_SW

Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Andres Rodriguez <andresx7@gmail.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h        |  4 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |  2 +
 drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c      | 99 ++++++++++++++++++++++++++++++
 3 files changed, 105 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index d62a35e..cedfc96 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -1023,6 +1023,10 @@ struct amdgpu_gfx {
 	bool                            in_suspend;
 	/* NGG */
 	struct amdgpu_ngg		ngg;
+
+	/* pipe reservation */
+	struct mutex			pipe_reserve_mutex;
+	DECLARE_BITMAP			(pipe_reserve_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
 };
 
 int amdgpu_ib_get(struct amdgpu_device *adev, struct amdgpu_vm *vm,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index b0109eb..e07750c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -2012,6 +2012,7 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 	adev->vm_manager.vm_pte_num_rings = 0;
 	adev->gart.gart_funcs = NULL;
 	adev->fence_context = dma_fence_context_alloc(AMDGPU_MAX_RINGS);
+	bitmap_zero(adev->gfx.pipe_reserve_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
 
 	adev->smc_rreg = &amdgpu_invalid_rreg;
 	adev->smc_wreg = &amdgpu_invalid_wreg;
@@ -2040,6 +2041,7 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 	mutex_init(&adev->pm.mutex);
 	mutex_init(&adev->gfx.gpu_clock_mutex);
 	mutex_init(&adev->srbm_mutex);
+	mutex_init(&adev->gfx.pipe_reserve_mutex);
 	mutex_init(&adev->grbm_idx_mutex);
 	mutex_init(&adev->mn_lock);
 	hash_init(adev->mn_hash);
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
index 666a154..3b565df 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
@@ -6365,6 +6365,104 @@ static void gfx_v8_0_ring_set_wptr_compute(struct amdgpu_ring *ring)
 	WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
 }
 
+static void gfx_v8_0_ring_set_pipe_percent(struct amdgpu_ring *ring,
+					   bool acquire)
+{
+	struct amdgpu_device *adev = ring->adev;
+	int pipe_num, tmp, reg;
+	int pipe_percent = acquire ? SPI_WCL_PIPE_PERCENT_GFX__VALUE_MASK : 0x1;
+
+	pipe_num = ring->me * adev->gfx.mec.num_pipe_per_mec + ring->pipe;
+
+	/* first me only has 2 entries, GFX and HP3D */
+	if (ring->me > 0)
+		pipe_num -= 2;
+
+	reg = mmSPI_WCL_PIPE_PERCENT_GFX + pipe_num;
+	tmp = RREG32(reg);
+	tmp = REG_SET_FIELD(tmp, SPI_WCL_PIPE_PERCENT_GFX, VALUE, pipe_percent);
+	WREG32(reg, tmp);
+}
+
+static void gfx_v8_0_pipe_reserve_resources(struct amdgpu_device *adev,
+					    struct amdgpu_ring *ring,
+					    bool acquire)
+{
+	int i, pipe;
+	bool reserve;
+	struct amdgpu_ring *iring;
+
+	mutex_lock(&adev->gfx.pipe_reserve_mutex);
+	pipe = amdgpu_gfx_queue_to_bit(adev, ring->me, ring->pipe, 0);
+	if (acquire)
+		set_bit(pipe, adev->gfx.pipe_reserve_bitmap);
+	else
+		clear_bit(pipe, adev->gfx.pipe_reserve_bitmap);
+
+	if (!bitmap_weight(adev->gfx.pipe_reserve_bitmap, AMDGPU_MAX_COMPUTE_QUEUES)) {
+		/* Clear all reservations - everyone reacquires all resources */
+		for (i = 0; i < adev->gfx.num_gfx_rings; ++i)
+			gfx_v8_0_ring_set_pipe_percent(&adev->gfx.gfx_ring[i],
+						       true);
+
+		for (i = 0; i < adev->gfx.num_compute_rings; ++i)
+			gfx_v8_0_ring_set_pipe_percent(&adev->gfx.compute_ring[i],
+						       true);
+	} else {
+		/* Lower all pipes without a current reservation */
+		for (i = 0; i < adev->gfx.num_gfx_rings; ++i) {
+			iring = &adev->gfx.gfx_ring[i];
+			pipe = amdgpu_gfx_queue_to_bit(adev,
+						       iring->me,
+						       iring->pipe,
+						       0);
+			reserve = test_bit(pipe, adev->gfx.pipe_reserve_bitmap);
+			gfx_v8_0_ring_set_pipe_percent(iring, reserve);
+		}
+
+		for (i = 0; i < adev->gfx.num_compute_rings; ++i) {
+			iring = &adev->gfx.compute_ring[i];
+			pipe = amdgpu_gfx_queue_to_bit(adev,
+						       iring->me,
+						       iring->pipe,
+						       0);
+			reserve = test_bit(pipe, adev->gfx.pipe_reserve_bitmap);
+			gfx_v8_0_ring_set_pipe_percent(iring, reserve);
+		}
+	}
+
+	mutex_unlock(&adev->gfx.pipe_reserve_mutex);
+}
+
+static void gfx_v8_0_hqd_set_priority(struct amdgpu_device *adev,
+				      struct amdgpu_ring *ring,
+				      bool acquire)
+{
+	uint32_t pipe_priority = acquire ? 0x2 : 0x0;
+	uint32_t queue_priority = acquire ? 0xf : 0x0;
+
+	mutex_lock(&adev->srbm_mutex);
+	vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
+
+	WREG32(mmCP_HQD_PIPE_PRIORITY, pipe_priority);
+	WREG32(mmCP_HQD_QUEUE_PRIORITY, queue_priority);
+
+	vi_srbm_select(adev, 0, 0, 0, 0);
+	mutex_unlock(&adev->srbm_mutex);
+}
+static void gfx_v8_0_ring_set_priority_compute(struct amdgpu_ring *ring,
+					       enum amd_sched_priority priority)
+{
+	struct amdgpu_device *adev = ring->adev;
+	bool acquire = priority == AMD_SCHED_PRIORITY_HIGH_HW;
+
+	if (ring->funcs->type != AMDGPU_RING_TYPE_COMPUTE)
+		return;
+
+	gfx_v8_0_hqd_set_priority(adev, ring, acquire);
+	gfx_v8_0_pipe_reserve_resources(adev, ring, acquire);
+}
+
 static void gfx_v8_0_ring_emit_fence_compute(struct amdgpu_ring *ring,
 					     u64 addr, u64 seq,
 					     unsigned flags)
@@ -6810,6 +6908,7 @@ static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_compute = {
 	.test_ib = gfx_v8_0_ring_test_ib,
 	.insert_nop = amdgpu_ring_insert_nop,
 	.pad_ib = amdgpu_ring_generic_pad_ib,
+	.set_priority = gfx_v8_0_ring_set_priority_compute,
 };
 
 static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_kiq = {
-- 
2.9.3

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 4/8] drm/amdgpu: make amdgpu_to_sched_priority detect invalid parameters

[Andres Rodriguez]

Returning invalid priorities as _NORMAL is a backwards compatibility
quirk of amdgpu_ctx_ioctl(). Move this detail one layer up where it
belongs.

Signed-off-by: Andres Rodriguez <andresx7@gmail.com>
Acked-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c       | 8 +++++---
 drivers/gpu/drm/amd/scheduler/gpu_scheduler.h | 3 ++-
 2 files changed, 7 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
index 9ec85d5..fbf15dd 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
@@ -232,7 +232,7 @@ static enum amd_sched_priority amdgpu_to_sched_priority(int amdgpu_priority)
 		return AMD_SCHED_PRIORITY_LOW;
 	default:
 		WARN(1, "Invalid context priority %d\n", amdgpu_priority);
-		return AMD_SCHED_PRIORITY_NORMAL;
+		return AMD_SCHED_PRIORITY_INVALID;
 	}
 }
 
@@ -251,8 +251,10 @@ int amdgpu_ctx_ioctl(struct drm_device *dev, void *data,
 	id = args->in.ctx_id;
 	priority = amdgpu_to_sched_priority(args->in.priority);
 
-	if (priority >= AMD_SCHED_PRIORITY_MAX)
-		return -EINVAL;
+	/* For backwards compatibility reasons, we need to accept
+	 * ioctls with garbage in the priority field */
+	if (priority == AMD_SCHED_PRIORITY_INVALID)
+		priority = AMD_SCHED_PRIORITY_NORMAL;
 
 	switch (args->in.op) {
 	case AMDGPU_CTX_OP_ALLOC_CTX:
diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
index dbcaa2e..da040bc 100644
--- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
+++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
@@ -120,7 +120,8 @@ enum amd_sched_priority {
 	AMD_SCHED_PRIORITY_HIGH_SW,
 	AMD_SCHED_PRIORITY_HIGH_HW,
 	AMD_SCHED_PRIORITY_KERNEL,
-	AMD_SCHED_PRIORITY_MAX
+	AMD_SCHED_PRIORITY_MAX,
+	AMD_SCHED_PRIORITY_INVALID = -1
 };
 
 /**
-- 
2.9.3

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 5/8] drm/amd/sched: allow clients to edit an entity's rq v2

[Andres Rodriguez]

This is useful for changing an entity's priority at runtime.

v2: don't modify the order of amd_sched_entity members

Signed-off-by: Andres Rodriguez <andresx7@gmail.com>
Acked-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 24 ++++++++++++++++++++++--
 drivers/gpu/drm/amd/scheduler/gpu_scheduler.h |  3 +++
 2 files changed, 25 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c
index 97c94f9..a5adde1 100644
--- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c
+++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c
@@ -133,6 +133,7 @@ int amd_sched_entity_init(struct amd_gpu_scheduler *sched,
 	entity->rq = rq;
 	entity->sched = sched;
 
+	spin_lock_init(&entity->rq_lock);
 	spin_lock_init(&entity->queue_lock);
 	r = kfifo_alloc(&entity->job_queue, jobs * sizeof(void *), GFP_KERNEL);
 	if (r)
@@ -204,7 +205,6 @@ static bool amd_sched_entity_is_ready(struct amd_sched_entity *entity)
 void amd_sched_entity_fini(struct amd_gpu_scheduler *sched,
 			   struct amd_sched_entity *entity)
 {
-	struct amd_sched_rq *rq = entity->rq;
 	int r;
 
 	if (!amd_sched_entity_is_initialized(sched, entity))
@@ -218,7 +218,7 @@ void amd_sched_entity_fini(struct amd_gpu_scheduler *sched,
 	else
 		r = wait_event_killable(sched->job_scheduled,
 					amd_sched_entity_is_idle(entity));
-	amd_sched_rq_remove_entity(rq, entity);
+	amd_sched_entity_set_rq(entity, NULL);
 	if (r) {
 		struct amd_sched_job *job;
 
@@ -251,6 +251,24 @@ static void amd_sched_entity_clear_dep(struct dma_fence *f, struct dma_fence_cb
 	dma_fence_put(f);
 }
 
+void amd_sched_entity_set_rq(struct amd_sched_entity *entity,
+			     struct amd_sched_rq *rq)
+{
+	if (entity->rq == rq)
+		return;
+
+	spin_lock(&entity->rq_lock);
+
+	if (entity->rq)
+		amd_sched_rq_remove_entity(entity->rq, entity);
+
+	entity->rq = rq;
+	if (rq)
+		amd_sched_rq_add_entity(rq, entity);
+
+	spin_unlock(&entity->rq_lock);
+}
+
 bool amd_sched_dependency_optimized(struct dma_fence* fence,
 				    struct amd_sched_entity *entity)
 {
@@ -348,7 +366,9 @@ static bool amd_sched_entity_in(struct amd_sched_job *sched_job)
 	/* first job wakes up scheduler */
 	if (first) {
 		/* Add the entity to the run queue */
+		spin_lock(&entity->rq_lock);
 		amd_sched_rq_add_entity(entity->rq, entity);
+		spin_unlock(&entity->rq_lock);
 		amd_sched_wakeup(sched);
 	}
 	return added;
diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
index da040bc..4b528f7 100644
--- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
+++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
@@ -39,6 +39,7 @@ struct amd_sched_rq;
 struct amd_sched_entity {
 	struct list_head		list;
 	struct amd_sched_rq		*rq;
+	spinlock_t			rq_lock;
 	struct amd_gpu_scheduler	*sched;
 
 	spinlock_t			queue_lock;
@@ -154,6 +155,8 @@ int amd_sched_entity_init(struct amd_gpu_scheduler *sched,
 void amd_sched_entity_fini(struct amd_gpu_scheduler *sched,
 			   struct amd_sched_entity *entity);
 void amd_sched_entity_push_job(struct amd_sched_job *sched_job);
+void amd_sched_entity_set_rq(struct amd_sched_entity *entity,
+			     struct amd_sched_rq *rq);
 
 int amd_sched_fence_slab_init(void);
 void amd_sched_fence_slab_fini(void);
-- 
2.9.3

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 6/8] drm/amdgpu: introduce AMDGPU_CTX_PRIORITY_UNSET

[Andres Rodriguez]

Use _INVALID to identify bad parameters and _UNSET to represent the
lack of interest in a specific value.

Signed-off-by: Andres Rodriguez <andresx7@gmail.com>
Acked-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c       | 2 ++
 drivers/gpu/drm/amd/scheduler/gpu_scheduler.h | 3 ++-
 include/uapi/drm/amdgpu_drm.h                 | 1 +
 3 files changed, 5 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
index fbf15dd..e4de0fb 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
@@ -230,6 +230,8 @@ static enum amd_sched_priority amdgpu_to_sched_priority(int amdgpu_priority)
 	case AMDGPU_CTX_PRIORITY_LOW_SW:
 	case AMDGPU_CTX_PRIORITY_LOW_HW:
 		return AMD_SCHED_PRIORITY_LOW;
+	case AMDGPU_CTX_PRIORITY_UNSET:
+		return AMD_SCHED_PRIORITY_UNSET;
 	default:
 		WARN(1, "Invalid context priority %d\n", amdgpu_priority);
 		return AMD_SCHED_PRIORITY_INVALID;
diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
index 4b528f7..52c8e54 100644
--- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
+++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
@@ -122,7 +122,8 @@ enum amd_sched_priority {
 	AMD_SCHED_PRIORITY_HIGH_HW,
 	AMD_SCHED_PRIORITY_KERNEL,
 	AMD_SCHED_PRIORITY_MAX,
-	AMD_SCHED_PRIORITY_INVALID = -1
+	AMD_SCHED_PRIORITY_INVALID = -1,
+	AMD_SCHED_PRIORITY_UNSET = -2
 };
 
 /**
diff --git a/include/uapi/drm/amdgpu_drm.h b/include/uapi/drm/amdgpu_drm.h
index d93e988..127797a 100644
--- a/include/uapi/drm/amdgpu_drm.h
+++ b/include/uapi/drm/amdgpu_drm.h
@@ -167,6 +167,7 @@ union drm_amdgpu_bo_list {
 #define AMDGPU_CTX_UNKNOWN_RESET	3
 
 /* Context priority level */
+#define AMDGPU_CTX_PRIORITY_UNSET       -2048
 #define AMDGPU_CTX_PRIORITY_LOW_HW      -1023
 #define AMDGPU_CTX_PRIORITY_LOW_SW      -512
 #define AMDGPU_CTX_PRIORITY_NORMAL      0
-- 
2.9.3

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 7/8] drm/amdgpu: add plumbing for ctx priority changes v2

[Andres Rodriguez]

Introduce amdgpu_ctx_priority_override(). A mechanism to override a
context's priority.

An override can be terminated by setting the override to
AMD_SCHED_PRIORITY_UNSET.

v2: change refcounted interface for a direct set

Signed-off-by: Andres Rodriguez <andresx7@gmail.com>
Acked-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h     |  6 +++++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c | 29 +++++++++++++++++++++++++++++
 2 files changed, 34 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index cedfc96..ca2df51 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -722,7 +722,9 @@ struct amdgpu_ctx {
 	spinlock_t		ring_lock;
 	struct dma_fence	**fences;
 	struct amdgpu_ctx_ring	rings[AMDGPU_MAX_RINGS];
-	bool preamble_presented;
+	bool 			preamble_presented;
+	enum amd_sched_priority init_priority;
+	enum amd_sched_priority override_priority;
 };
 
 struct amdgpu_ctx_mgr {
@@ -739,6 +741,8 @@ uint64_t amdgpu_ctx_add_fence(struct amdgpu_ctx *ctx, struct amdgpu_ring *ring,
 			      struct dma_fence *fence);
 struct dma_fence *amdgpu_ctx_get_fence(struct amdgpu_ctx *ctx,
 				   struct amdgpu_ring *ring, uint64_t seq);
+void amdgpu_ctx_priority_override(struct amdgpu_ctx *ctx,
+				  enum amd_sched_priority priority);
 
 int amdgpu_ctx_ioctl(struct drm_device *dev, void *data,
 		     struct drm_file *filp);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
index e4de0fb..bf05180 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
@@ -72,6 +72,8 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
 	}
 
 	ctx->reset_counter = atomic_read(&adev->gpu_reset_counter);
+	ctx->init_priority = priority;
+	ctx->override_priority = AMD_SCHED_PRIORITY_UNSET;
 
 	/* create context entity for each ring */
 	for (i = 0; i < adev->num_rings; i++) {
@@ -360,6 +362,33 @@ struct dma_fence *amdgpu_ctx_get_fence(struct amdgpu_ctx *ctx,
 	return fence;
 }
 
+void amdgpu_ctx_priority_override(struct amdgpu_ctx *ctx,
+				  enum amd_sched_priority priority)
+{
+	int i;
+	struct amdgpu_device *adev = ctx->adev;
+	struct amd_sched_rq *rq;
+	struct amd_sched_entity *entity;
+	struct amdgpu_ring *ring;
+	enum amd_sched_priority ctx_prio;
+
+	ctx->override_priority = priority;
+
+	ctx_prio = (ctx->override_priority == AMD_SCHED_PRIORITY_UNSET) ?
+			ctx->init_priority : ctx->override_priority;
+
+	for (i = 0; i < adev->num_rings; i++) {
+		ring = adev->rings[i];
+		entity = &ctx->rings[i].entity;
+		rq = &ring->sched.sched_rq[ctx_prio];
+
+		if (ring->funcs->type == AMDGPU_RING_TYPE_KIQ)
+			continue;
+
+		amd_sched_entity_set_rq(entity, rq);
+	}
+}
+
 void amdgpu_ctx_mgr_init(struct amdgpu_ctx_mgr *mgr)
 {
 	mutex_init(&mgr->lock);
-- 
2.9.3

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 8/8] drm/amdgpu: add interface for editing a foreign process's priority v3

[Andres Rodriguez]

The AMDGPU_SCHED_OP_PROCESS_PRIORITY_OVERRIDE ioctls are used to set
the priority of a different process in the current system.

When a request is dropped, the process's contexts will be
restored to the priority specified at context creation time.

A request can be dropped by setting the override priority to
AMDGPU_CTX_PRIORITY_UNSET.

An fd is used to identify the remote process. This is simpler than
passing a pid number, which is vulnerable to re-use, etc.

This functionality is limited to DRM_MASTER since abuse of this
interface can have a negative impact on the system's performance.

v2: removed unused output structure
v3: change refcounted interface for a regular set operation

Signed-off-by: Andres Rodriguez <andresx7@gmail.com>
Acked-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/Makefile       |   2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c   |  21 +-----
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c   |   2 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c | 109 ++++++++++++++++++++++++++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_sched.h |  34 ++++++++++
 include/uapi/drm/amdgpu_drm.h             |  17 +++++
 6 files changed, 164 insertions(+), 21 deletions(-)
 create mode 100644 drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
 create mode 100644 drivers/gpu/drm/amd/amdgpu/amdgpu_sched.h

diff --git a/drivers/gpu/drm/amd/amdgpu/Makefile b/drivers/gpu/drm/amd/amdgpu/Makefile
index 25a95c9..ef9a3b6 100644
--- a/drivers/gpu/drm/amd/amdgpu/Makefile
+++ b/drivers/gpu/drm/amd/amdgpu/Makefile
@@ -25,7 +25,7 @@ amdgpu-y += amdgpu_device.o amdgpu_kms.o \
 	amdgpu_prime.o amdgpu_vm.o amdgpu_ib.o amdgpu_pll.o \
 	amdgpu_ucode.o amdgpu_bo_list.o amdgpu_ctx.o amdgpu_sync.o \
 	amdgpu_gtt_mgr.o amdgpu_vram_mgr.o amdgpu_virt.o amdgpu_atomfirmware.o \
-	amdgpu_queue_mgr.o amdgpu_vf_error.o
+	amdgpu_queue_mgr.o amdgpu_vf_error.o amdgpu_sched.o
 
 # add asic specific block
 amdgpu-$(CONFIG_DRM_AMDGPU_CIK)+= cik.o cik_ih.o kv_smc.o kv_dpm.o \
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
index bf05180..97aafc2 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
@@ -25,6 +25,7 @@
 #include <drm/drmP.h>
 #include <drm/drm_auth.h>
 #include "amdgpu.h"
+#include "amdgpu_sched.h"
 
 static int amdgpu_ctx_priority_permit(struct drm_file *filp,
 				      enum amd_sched_priority priority)
@@ -220,26 +221,6 @@ static int amdgpu_ctx_query(struct amdgpu_device *adev,
 	return 0;
 }
 
-static enum amd_sched_priority amdgpu_to_sched_priority(int amdgpu_priority)
-{
-	switch (amdgpu_priority) {
-	case AMDGPU_CTX_PRIORITY_HIGH_HW:
-		return AMD_SCHED_PRIORITY_HIGH_HW;
-	case AMDGPU_CTX_PRIORITY_HIGH_SW:
-		return AMD_SCHED_PRIORITY_HIGH_SW;
-	case AMDGPU_CTX_PRIORITY_NORMAL:
-		return AMD_SCHED_PRIORITY_NORMAL;
-	case AMDGPU_CTX_PRIORITY_LOW_SW:
-	case AMDGPU_CTX_PRIORITY_LOW_HW:
-		return AMD_SCHED_PRIORITY_LOW;
-	case AMDGPU_CTX_PRIORITY_UNSET:
-		return AMD_SCHED_PRIORITY_UNSET;
-	default:
-		WARN(1, "Invalid context priority %d\n", amdgpu_priority);
-		return AMD_SCHED_PRIORITY_INVALID;
-	}
-}
-
 int amdgpu_ctx_ioctl(struct drm_device *dev, void *data,
 		     struct drm_file *filp)
 {
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
index 79d9ab4..dff4e54 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
@@ -28,6 +28,7 @@
 #include <drm/drmP.h>
 #include "amdgpu.h"
 #include <drm/amdgpu_drm.h>
+#include "amdgpu_sched.h"
 #include "amdgpu_uvd.h"
 #include "amdgpu_vce.h"
 
@@ -1020,6 +1021,7 @@ const struct drm_ioctl_desc amdgpu_ioctls_kms[] = {
 	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_CREATE, amdgpu_gem_create_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
 	DRM_IOCTL_DEF_DRV(AMDGPU_CTX, amdgpu_ctx_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
 	DRM_IOCTL_DEF_DRV(AMDGPU_VM, amdgpu_vm_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(AMDGPU_SCHED, amdgpu_sched_ioctl, DRM_MASTER),
 	DRM_IOCTL_DEF_DRV(AMDGPU_BO_LIST, amdgpu_bo_list_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
 	/* KMS */
 	DRM_IOCTL_DEF_DRV(AMDGPU_GEM_MMAP, amdgpu_gem_mmap_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
new file mode 100644
index 0000000..cd12330
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
@@ -0,0 +1,109 @@
+/*
+ * Copyright 2017 Valve Corporation
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND 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: Andres Rodriguez <andresx7@gmail.com>
+ */
+
+#include <linux/fdtable.h>
+#include <linux/pid.h>
+#include <drm/amdgpu_drm.h>
+#include "amdgpu.h"
+
+#include "amdgpu_vm.h"
+
+enum amd_sched_priority amdgpu_to_sched_priority(int amdgpu_priority)
+{
+	switch (amdgpu_priority) {
+	case AMDGPU_CTX_PRIORITY_HIGH_HW:
+		return AMD_SCHED_PRIORITY_HIGH_HW;
+	case AMDGPU_CTX_PRIORITY_HIGH_SW:
+		return AMD_SCHED_PRIORITY_HIGH_SW;
+	case AMDGPU_CTX_PRIORITY_NORMAL:
+		return AMD_SCHED_PRIORITY_NORMAL;
+	case AMDGPU_CTX_PRIORITY_LOW_SW:
+	case AMDGPU_CTX_PRIORITY_LOW_HW:
+		return AMD_SCHED_PRIORITY_LOW;
+	case AMDGPU_CTX_PRIORITY_UNSET:
+		return AMD_SCHED_PRIORITY_UNSET;
+	default:
+		WARN(1, "Invalid context priority %d\n", amdgpu_priority);
+		return AMD_SCHED_PRIORITY_INVALID;
+	}
+}
+
+static int amdgpu_sched_process_priority_override(struct amdgpu_device *adev,
+						  int fd,
+						  enum amd_sched_priority priority)
+{
+	struct file *filp = fcheck(fd);
+	struct drm_file *file;
+	struct pid *pid;
+	struct amdgpu_fpriv *fpriv;
+	struct amdgpu_ctx *ctx;
+	uint32_t id;
+
+	if (!filp)
+		return -EINVAL;
+
+	pid = get_pid(((struct drm_file *)filp->private_data)->pid);
+
+	mutex_lock(&adev->ddev->filelist_mutex);
+	list_for_each_entry(file, &adev->ddev->filelist, lhead) {
+		if (file->pid != pid)
+			continue;
+
+		fpriv = file->driver_priv;
+		idr_for_each_entry(&fpriv->ctx_mgr.ctx_handles, ctx, id)
+				amdgpu_ctx_priority_override(ctx, priority);
+	}
+	mutex_unlock(&adev->ddev->filelist_mutex);
+
+	put_pid(pid);
+
+	return 0;
+}
+
+int amdgpu_sched_ioctl(struct drm_device *dev, void *data,
+		       struct drm_file *filp)
+{
+	union drm_amdgpu_sched *args = data;
+	struct amdgpu_device *adev = dev->dev_private;
+	enum amd_sched_priority priority;
+	int r;
+
+	priority = amdgpu_to_sched_priority(args->in.priority);
+	if (args->in.flags || priority == AMD_SCHED_PRIORITY_INVALID)
+		return -EINVAL;
+
+	switch (args->in.op) {
+	case AMDGPU_SCHED_OP_PROCESS_PRIORITY_OVERRIDE:
+		r = amdgpu_sched_process_priority_override(adev,
+							   args->in.fd,
+							   priority);
+		break;
+	default:
+		DRM_ERROR("Invalid sched op specified: %d\n", args->in.op);
+		r = -EINVAL;
+		break;
+	}
+
+	return r;
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.h
new file mode 100644
index 0000000..b28c067
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.h
@@ -0,0 +1,34 @@
+/*
+ * Copyright 2017 Valve Corporation
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND 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: Andres Rodriguez <andresx7@gmail.com>
+ */
+
+#ifndef __AMDGPU_SCHED_H__
+#define __AMDGPU_SCHED_H__
+
+#include <drm/drmP.h>
+
+enum amd_sched_priority amdgpu_to_sched_priority(int amdgpu_priority);
+int amdgpu_sched_ioctl(struct drm_device *dev, void *data,
+		       struct drm_file *filp);
+
+#endif // __AMDGPU_SCHED_H__
diff --git a/include/uapi/drm/amdgpu_drm.h b/include/uapi/drm/amdgpu_drm.h
index 127797a..96fcde8 100644
--- a/include/uapi/drm/amdgpu_drm.h
+++ b/include/uapi/drm/amdgpu_drm.h
@@ -52,6 +52,7 @@ extern "C" {
 #define DRM_AMDGPU_GEM_USERPTR		0x11
 #define DRM_AMDGPU_WAIT_FENCES		0x12
 #define DRM_AMDGPU_VM			0x13
+#define DRM_AMDGPU_SCHED		0x14
 
 #define DRM_IOCTL_AMDGPU_GEM_CREATE	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_CREATE, union drm_amdgpu_gem_create)
 #define DRM_IOCTL_AMDGPU_GEM_MMAP	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_MMAP, union drm_amdgpu_gem_mmap)
@@ -67,6 +68,7 @@ extern "C" {
 #define DRM_IOCTL_AMDGPU_GEM_USERPTR	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_GEM_USERPTR, struct drm_amdgpu_gem_userptr)
 #define DRM_IOCTL_AMDGPU_WAIT_FENCES	DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_WAIT_FENCES, union drm_amdgpu_wait_fences)
 #define DRM_IOCTL_AMDGPU_VM		DRM_IOWR(DRM_COMMAND_BASE + DRM_AMDGPU_VM, union drm_amdgpu_vm)
+#define DRM_IOCTL_AMDGPU_SCHED		DRM_IOW(DRM_COMMAND_BASE + DRM_AMDGPU_SCHED, union drm_amdgpu_sched)
 
 #define AMDGPU_GEM_DOMAIN_CPU		0x1
 #define AMDGPU_GEM_DOMAIN_GTT		0x2
@@ -225,6 +227,21 @@ union drm_amdgpu_vm {
 	struct drm_amdgpu_vm_out out;
 };
 
+/* sched ioctl */
+#define AMDGPU_SCHED_OP_PROCESS_PRIORITY_OVERRIDE	1
+
+struct drm_amdgpu_sched_in {
+	/* AMDGPU_SCHED_OP_* */
+	__u32	op;
+	__u32	fd;
+	__s32	priority;
+	__u32	flags;
+};
+
+union drm_amdgpu_sched {
+	struct drm_amdgpu_sched_in in;
+};
+
 /*
  * This is not a reliable API and you should expect it to fail for any
  * number of reasons and have fallback path that do not use userptr to
-- 
2.9.3

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 7/8] drm/amdgpu: add plumbing for ctx priority changes v2

[Andres Rodriguez]

Introduce amdgpu_ctx_priority_override(). A mechanism to override a
context's priority.

An override can be terminated by setting the override to
AMD_SCHED_PRIORITY_UNSET.

v2: change refcounted interface for a direct set

Signed-off-by: Andres Rodriguez <andresx7@gmail.com>
Acked-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h     |  6 +++++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c | 29 +++++++++++++++++++++++++++++
 2 files changed, 34 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 479a1e9..7251823 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -825,7 +825,9 @@ struct amdgpu_ctx {
 	spinlock_t		ring_lock;
 	struct dma_fence	**fences;
 	struct amdgpu_ctx_ring	rings[AMDGPU_MAX_RINGS];
-	bool preamble_presented;
+	bool 			preamble_presented;
+	enum amd_sched_priority init_priority;
+	enum amd_sched_priority override_priority;
 };
 
 struct amdgpu_ctx_mgr {
@@ -842,6 +844,8 @@ uint64_t amdgpu_ctx_add_fence(struct amdgpu_ctx *ctx, struct amdgpu_ring *ring,
 			      struct dma_fence *fence);
 struct dma_fence *amdgpu_ctx_get_fence(struct amdgpu_ctx *ctx,
 				   struct amdgpu_ring *ring, uint64_t seq);
+void amdgpu_ctx_priority_override(struct amdgpu_ctx *ctx,
+				  enum amd_sched_priority priority);
 
 int amdgpu_ctx_ioctl(struct drm_device *dev, void *data,
 		     struct drm_file *filp);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
index e4de0fb..bf05180 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
@@ -72,6 +72,8 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
 	}
 
 	ctx->reset_counter = atomic_read(&adev->gpu_reset_counter);
+	ctx->init_priority = priority;
+	ctx->override_priority = AMD_SCHED_PRIORITY_UNSET;
 
 	/* create context entity for each ring */
 	for (i = 0; i < adev->num_rings; i++) {
@@ -360,6 +362,33 @@ struct dma_fence *amdgpu_ctx_get_fence(struct amdgpu_ctx *ctx,
 	return fence;
 }
 
+void amdgpu_ctx_priority_override(struct amdgpu_ctx *ctx,
+				  enum amd_sched_priority priority)
+{
+	int i;
+	struct amdgpu_device *adev = ctx->adev;
+	struct amd_sched_rq *rq;
+	struct amd_sched_entity *entity;
+	struct amdgpu_ring *ring;
+	enum amd_sched_priority ctx_prio;
+
+	ctx->override_priority = priority;
+
+	ctx_prio = (ctx->override_priority == AMD_SCHED_PRIORITY_UNSET) ?
+			ctx->init_priority : ctx->override_priority;
+
+	for (i = 0; i < adev->num_rings; i++) {
+		ring = adev->rings[i];
+		entity = &ctx->rings[i].entity;
+		rq = &ring->sched.sched_rq[ctx_prio];
+
+		if (ring->funcs->type == AMDGPU_RING_TYPE_KIQ)
+			continue;
+
+		amd_sched_entity_set_rq(entity, rq);
+	}
+}
+
 void amdgpu_ctx_mgr_init(struct amdgpu_ctx_mgr *mgr)
 {
 	mutex_init(&mgr->lock);
-- 
2.9.3

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 7/8] drm/amdgpu: add plumbing for ctx priority changes v2

[Andres Rodriguez]

Introduce amdgpu_ctx_priority_override(). A mechanism to override a
context's priority.

An override can be terminated by setting the override to
AMD_SCHED_PRIORITY_UNSET.

v2: change refcounted interface for a direct set

Signed-off-by: Andres Rodriguez <andresx7@gmail.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h     |  6 +++++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c | 29 +++++++++++++++++++++++++++++
 2 files changed, 34 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 7c6fca6..0a24c1f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -825,7 +825,9 @@ struct amdgpu_ctx {
 	spinlock_t		ring_lock;
 	struct dma_fence	**fences;
 	struct amdgpu_ctx_ring	rings[AMDGPU_MAX_RINGS];
-	bool preamble_presented;
+	bool 			preamble_presented;
+	enum amd_sched_priority init_priority;
+	enum amd_sched_priority override_priority;
 };
 
 struct amdgpu_ctx_mgr {
@@ -842,6 +844,8 @@ uint64_t amdgpu_ctx_add_fence(struct amdgpu_ctx *ctx, struct amdgpu_ring *ring,
 			      struct dma_fence *fence);
 struct dma_fence *amdgpu_ctx_get_fence(struct amdgpu_ctx *ctx,
 				   struct amdgpu_ring *ring, uint64_t seq);
+void amdgpu_ctx_priority_override(struct amdgpu_ctx *ctx,
+				  enum amd_sched_priority priority);
 
 int amdgpu_ctx_ioctl(struct drm_device *dev, void *data,
 		     struct drm_file *filp);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
index e4de0fb..bf05180 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
@@ -72,6 +72,8 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
 	}
 
 	ctx->reset_counter = atomic_read(&adev->gpu_reset_counter);
+	ctx->init_priority = priority;
+	ctx->override_priority = AMD_SCHED_PRIORITY_UNSET;
 
 	/* create context entity for each ring */
 	for (i = 0; i < adev->num_rings; i++) {
@@ -360,6 +362,33 @@ struct dma_fence *amdgpu_ctx_get_fence(struct amdgpu_ctx *ctx,
 	return fence;
 }
 
+void amdgpu_ctx_priority_override(struct amdgpu_ctx *ctx,
+				  enum amd_sched_priority priority)
+{
+	int i;
+	struct amdgpu_device *adev = ctx->adev;
+	struct amd_sched_rq *rq;
+	struct amd_sched_entity *entity;
+	struct amdgpu_ring *ring;
+	enum amd_sched_priority ctx_prio;
+
+	ctx->override_priority = priority;
+
+	ctx_prio = (ctx->override_priority == AMD_SCHED_PRIORITY_UNSET) ?
+			ctx->init_priority : ctx->override_priority;
+
+	for (i = 0; i < adev->num_rings; i++) {
+		ring = adev->rings[i];
+		entity = &ctx->rings[i].entity;
+		rq = &ring->sched.sched_rq[ctx_prio];
+
+		if (ring->funcs->type == AMDGPU_RING_TYPE_KIQ)
+			continue;
+
+		amd_sched_entity_set_rq(entity, rq);
+	}
+}
+
 void amdgpu_ctx_mgr_init(struct amdgpu_ctx_mgr *mgr)
 {
 	mutex_init(&mgr->lock);
-- 
2.9.3

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx