[PATCH 57/73] drm/amdgpu/mes: use ring for kernel queue submission
Alex Deucher
alexander.deucher at amd.com
Fri Apr 29 17:46:08 UTC 2022
From: Jack Xiao <Jack.Xiao at amd.com>
Use ring as the front end for kernel queue submission.
Signed-off-by: Jack Xiao <Jack.Xiao at amd.com>
Acked-by: Christian König <christian.koenig at amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang at amd.com>
Signed-off-by: Alex Deucher <alexander.deucher at amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c | 93 +++++++++++++++++++++++++
drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h | 5 ++
2 files changed, 98 insertions(+)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c
index 4e99adcfbb0e..827391fcb2a3 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c
@@ -753,3 +753,96 @@ int amdgpu_mes_ctx_get_offs(struct amdgpu_ring *ring, unsigned int id_offs)
WARN_ON(1);
return -EINVAL;
}
+
+int amdgpu_mes_add_ring(struct amdgpu_device *adev, int gang_id,
+ int queue_type, int idx,
+ struct amdgpu_mes_ctx_data *ctx_data,
+ struct amdgpu_ring **out)
+{
+ struct amdgpu_ring *ring;
+ struct amdgpu_mes_gang *gang;
+ struct amdgpu_mes_queue_properties qprops = {0};
+ int r, queue_id, pasid;
+
+ mutex_lock(&adev->mes.mutex);
+ gang = idr_find(&adev->mes.gang_id_idr, gang_id);
+ if (!gang) {
+ DRM_ERROR("gang id %d doesn't exist\n", gang_id);
+ mutex_unlock(&adev->mes.mutex);
+ return -EINVAL;
+ }
+ pasid = gang->process->pasid;
+
+ ring = kzalloc(sizeof(struct amdgpu_ring), GFP_KERNEL);
+ if (!ring) {
+ mutex_unlock(&adev->mes.mutex);
+ return -ENOMEM;
+ }
+
+ ring->ring_obj = NULL;
+ ring->use_doorbell = true;
+ ring->is_mes_queue = true;
+ ring->mes_ctx = ctx_data;
+ ring->idx = idx;
+ ring->no_scheduler = true;
+
+ if (queue_type == AMDGPU_RING_TYPE_COMPUTE) {
+ int offset = offsetof(struct amdgpu_mes_ctx_meta_data,
+ compute[ring->idx].mec_hpd);
+ ring->eop_gpu_addr =
+ amdgpu_mes_ctx_get_offs_gpu_addr(ring, offset);
+ }
+
+ switch (queue_type) {
+ case AMDGPU_RING_TYPE_GFX:
+ ring->funcs = adev->gfx.gfx_ring[0].funcs;
+ break;
+ case AMDGPU_RING_TYPE_COMPUTE:
+ ring->funcs = adev->gfx.compute_ring[0].funcs;
+ break;
+ case AMDGPU_RING_TYPE_SDMA:
+ ring->funcs = adev->sdma.instance[0].ring.funcs;
+ break;
+ default:
+ BUG();
+ }
+
+ r = amdgpu_ring_init(adev, ring, 1024, NULL, 0,
+ AMDGPU_RING_PRIO_DEFAULT, NULL);
+ if (r)
+ goto clean_up_memory;
+
+ amdgpu_mes_ring_to_queue_props(adev, ring, &qprops);
+
+ dma_fence_wait(gang->process->vm->last_update, false);
+ dma_fence_wait(ctx_data->meta_data_va->last_pt_update, false);
+ mutex_unlock(&adev->mes.mutex);
+
+ r = amdgpu_mes_add_hw_queue(adev, gang_id, &qprops, &queue_id);
+ if (r)
+ goto clean_up_ring;
+
+ ring->hw_queue_id = queue_id;
+ ring->doorbell_index = qprops.doorbell_off;
+
+ if (queue_type == AMDGPU_RING_TYPE_GFX)
+ sprintf(ring->name, "gfx_%d.%d.%d", pasid, gang_id, queue_id);
+ else if (queue_type == AMDGPU_RING_TYPE_COMPUTE)
+ sprintf(ring->name, "compute_%d.%d.%d", pasid, gang_id,
+ queue_id);
+ else if (queue_type == AMDGPU_RING_TYPE_SDMA)
+ sprintf(ring->name, "sdma_%d.%d.%d", pasid, gang_id,
+ queue_id);
+ else
+ BUG();
+
+ *out = ring;
+ return 0;
+
+clean_up_ring:
+ amdgpu_ring_fini(ring);
+clean_up_memory:
+ kfree(ring);
+ mutex_unlock(&adev->mes.mutex);
+ return r;
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h
index 36684416f277..1fe5c869f37e 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h
@@ -254,4 +254,9 @@ int amdgpu_mes_add_hw_queue(struct amdgpu_device *adev, int gang_id,
int *queue_id);
int amdgpu_mes_remove_hw_queue(struct amdgpu_device *adev, int queue_id);
+int amdgpu_mes_add_ring(struct amdgpu_device *adev, int gang_id,
+ int queue_type, int idx,
+ struct amdgpu_mes_ctx_data *ctx_data,
+ struct amdgpu_ring **out);
+
#endif /* __AMDGPU_MES_H__ */
--
2.35.1
More information about the amd-gfx
mailing list