Skip to contentSkip to conversation

Add support for high priority scheduling in amdgpu v9

PATCH New Andres Rodriguez 1 participant 4 messages May 9, 2017 1 message not in the archive

Subsystems (3)
lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): 20170509213951.2904-1-andresx7-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org

#2 lore raw

PATCH 1/3 drm/amdgpu: add parameter to allocate high priority contexts v9

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
* AMDGPU_CTX_PRIORITY_NORMAL
* AMDGPU_CTX_PRIORITY_LOW

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.

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

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c       | 38 ++++++++++++++++++++++++---
 drivers/gpu/drm/amd/scheduler/gpu_scheduler.h |  4 ++-
 include/uapi/drm/amdgpu_drm.h                 |  8 +++++-
 3 files changed, 44 insertions(+), 6 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +34 -4
index 1969f27..e0bb76a 100644
@@ -25,11 +25,19 @@
 #include <drm/drmP.h>
 #include "amdgpu.h"
 
-static int amdgpu_ctx_init(struct amdgpu_device *adev, struct amdgpu_ctx *ctx)
+static int amdgpu_ctx_init(struct amdgpu_device *adev,
+			   enum amd_sched_priority priority,
+			   struct amdgpu_ctx *ctx)
 {
 	unsigned i, j;
 	int r;
 
+	if (priority < 0 || priority >= AMD_SCHED_PRIORITY_MAX)
+		return -EINVAL;
+
+	if (priority >= AMD_SCHED_PRIORITY_HIGH && !capable(CAP_SYS_NICE))
+		return -EACCES;
+
 	memset(ctx, 0, sizeof(*ctx));
 	ctx->adev = adev;
 	kref_init(&ctx->refcount);
@@ -51,7 +59,7 @@ 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];
 		r = amd_sched_entity_init(&ring->sched, &ctx->rings[i].entity,
 					  rq, amdgpu_sched_jobs);
 		if (r)
@@ -96,6 +104,7 @@ static void amdgpu_ctx_fini(struct amdgpu_ctx *ctx)
 
 static int amdgpu_ctx_alloc(struct amdgpu_device *adev,
 			    struct amdgpu_fpriv *fpriv,
+			    enum amd_sched_priority priority,
 			    uint32_t *id)
 {
 	struct amdgpu_ctx_mgr *mgr = &fpriv->ctx_mgr;
@@ -113,8 +122,9 @@ 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, ctx);
 	if (r) {
 		idr_remove(&mgr->ctx_handles, *id);
 		*id = 0;
@@ -184,11 +194,27 @@ 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:
+		return AMD_SCHED_PRIORITY_HIGH;
+	case AMDGPU_CTX_PRIORITY_NORMAL:
+		return AMD_SCHED_PRIORITY_NORMAL;
+	case AMDGPU_CTX_PRIORITY_LOW:
+		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;
@@ -196,10 +222,14 @@ 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, priority, &id);
 		args->out.alloc.ctx_id = id;
 		break;
 	case AMDGPU_CTX_OP_FREE_CTX:
drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +3 -1
index 0255c7f..fc2d4bd 100644
@@ -109,7 +109,9 @@ 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,
 	AMD_SCHED_PRIORITY_KERNEL,
 	AMD_SCHED_PRIORITY_MAX
 };
include/uapi/drm/amdgpu_drm.h +7 -1
index 56ceb3d..df1f4cb 100644
@@ -162,13 +162,19 @@ union drm_amdgpu_bo_list {
 /* unknown cause */
 #define AMDGPU_CTX_UNKNOWN_RESET	3
 
+/* Context priority level */
+#define AMDGPU_CTX_PRIORITY_LOW         -1023
+#define AMDGPU_CTX_PRIORITY_NORMAL      0
+/* Selecting a priority above NORMAL requires CAP_SYS_ADMIN */
+#define AMDGPU_CTX_PRIORITY_HIGH        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 {
Signature · 6 lines
2.9.3

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
#3 lore raw

PATCH 2/3 drm/amdgpu: add framework for HW specific priority settings v7

In reply to a message not in the archive

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

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_job.c       |  7 +++
 drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c      | 78 ++++++++++++++++++++++++++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h      | 15 ++++++
 drivers/gpu/drm/amd/scheduler/gpu_scheduler.h |  7 +++
 4 files changed, 106 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_job.c +7 -0
index c3cfeb3..42f60d5 100644
@@ -95,6 +95,7 @@ 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);
 	kfree(job);
@@ -169,6 +170,11 @@ static struct dma_fence *amdgpu_job_run(struct amd_sched_job *sched_job)
 
 	BUG_ON(amdgpu_sync_peek_fence(&job->sync, NULL));
 
+	r = amdgpu_ring_priority_get(job->ring,
+				     amd_sched_get_job_priority(&job->base));
+	if (r)
+		DRM_ERROR("Failed to set job priority (%d)\n", r);
+
 	trace_amdgpu_sched_run_job(job);
 	r = amdgpu_ib_schedule(job->ring, job->num_ibs, job->ibs, job, &fence);
 	if (r)
@@ -177,6 +183,7 @@ 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;
 }
drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c +77 -1

99 lines not on this page · raw message

drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h +15 -0
index 10d6692..168ad93 100644
@@ -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 */
@@ -54,6 +55,7 @@ struct amdgpu_device;
 struct amdgpu_ring;
 struct amdgpu_ib;
 struct amdgpu_cs_parser;
+struct amdgpu_job;
 
 /*
  * Fences.
@@ -142,6 +144,9 @@ struct amdgpu_ring_funcs {
 	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);
+	/* priority functions */
+	void (*set_priority) (struct amdgpu_ring *ring,
+			      enum amd_sched_priority priority);
 };
 
 struct amdgpu_ring {
@@ -181,6 +186,12 @@ struct amdgpu_ring {
 	u64			cond_exe_gpu_addr;
 	volatile u32		*cond_exe_cpu_addr;
 	unsigned		vm_inv_eng;
+
+	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
@@ -193,6 +204,10 @@ 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);
+int 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);
drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +7 -0
index fc2d4bd..9dc19ed 100644
@@ -160,4 +160,11 @@ int amd_sched_job_init(struct amd_sched_job *job,
 		       void *owner);
 void amd_sched_hw_job_reset(struct amd_gpu_scheduler *sched);
 void amd_sched_job_recovery(struct amd_gpu_scheduler *sched);
+
+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
Signature · 6 lines
2.9.3

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
#4 lore raw

PATCH 3/3 drm/amdgpu: implement ring set_priority for gfx_v8 compute v8

In reply to a message not in the archive

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 CUs bouncing between GFX and high priority compute and introducing further latency, we reserve CUs 2+ for high priority compute on-demand.

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.

---
 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      | 113 +++++++++++++++++++++++++++++
 3 files changed, 119 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +4 -0
index 8ae8671..25c51ac 100644
@@ -1097,6 +1097,10 @@ 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,
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +2 -0
index c93516a..7fb70d5 100644
@@ -1826,6 +1826,7 @@ 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;
@@ -1854,6 +1855,7 @@ 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);
drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c +113 -0

125 lines not on this page · raw message

Signature · 6 lines
2.9.3

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx