drm/amdgpu: implement lru amdgpu_queue_mgr policy for compute v4

Use an LRU policy to map usermode rings to HW compute queues.

Most compute clients use one queue, and usually the first queue
available. This results in poor pipe/queue work distribution when
multiple compute apps are running. In most cases pipe 0 queue 0 is
the only queue that gets used.

In order to better distribute work across multiple HW queues, we adopt
a policy to map the usermode ring ids to the LRU HW queue.

This fixes a large majority of multi-app compute workloads sharing the
same HW queue, even though 7 other queues are available.

v2: use ring->funcs->type instead of ring->hw_ip
v3: remove amdgpu_queue_mapper_funcs
v4: change ring_lru_list_lock to spinlock, grab only once in lru_get()

Signed-off-by: Andres Rodriguez <andresx7@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

1 files changed, 37 insertions, 1 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_queue_mgr.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_queue_mgr.c
index c13a55352db6..4073f072f6c4 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_queue_mgr.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_queue_mgr.c
@@ -100,6 +100,40 @@ static int amdgpu_identity_map(struct amdgpu_device *adev,
         return amdgpu_update_cached_map(mapper, ring, *out_ring);
 }
 
+static enum amdgpu_ring_type amdgpu_hw_ip_to_ring_type(int hw_ip)
+{
+        switch (hw_ip) {
+        case AMDGPU_HW_IP_GFX:
+                return AMDGPU_RING_TYPE_GFX;
+        case AMDGPU_HW_IP_COMPUTE:
+                return AMDGPU_RING_TYPE_COMPUTE;
+        case AMDGPU_HW_IP_DMA:
+                return AMDGPU_RING_TYPE_SDMA;
+        case AMDGPU_HW_IP_UVD:
+                return AMDGPU_RING_TYPE_UVD;
+        case AMDGPU_HW_IP_VCE:
+                return AMDGPU_RING_TYPE_VCE;
+        default:
+                DRM_ERROR("Invalid HW IP specified %d\n", hw_ip);
+                return -1;
+        }
+}
+
+static int amdgpu_lru_map(struct amdgpu_device *adev,
+                          struct amdgpu_queue_mapper *mapper,
+                          int user_ring,
+                          struct amdgpu_ring **out_ring)
+{
+        int r;
+        int ring_type = amdgpu_hw_ip_to_ring_type(mapper->hw_ip);
+
+        r = amdgpu_ring_lru_get(adev, ring_type, out_ring);
+        if (r)
+                return r;
+
+        return amdgpu_update_cached_map(mapper, user_ring, *out_ring);
+}
+
 /**
  * amdgpu_queue_mgr_init - init an amdgpu_queue_mgr struct
  *
@@ -230,7 +264,6 @@ int amdgpu_queue_mgr_map(struct amdgpu_device *adev,
 
         switch (mapper->hw_ip) {
         case AMDGPU_HW_IP_GFX:
-        case AMDGPU_HW_IP_COMPUTE:
         case AMDGPU_HW_IP_DMA:
         case AMDGPU_HW_IP_UVD:
         case AMDGPU_HW_IP_VCE:
@@ -239,6 +272,9 @@ int amdgpu_queue_mgr_map(struct amdgpu_device *adev,
         case AMDGPU_HW_IP_VCN_ENC:
                 r = amdgpu_identity_map(adev, mapper, ring, out_ring);
                 break;
+        case AMDGPU_HW_IP_COMPUTE:
+                r = amdgpu_lru_map(adev, mapper, ring, out_ring);
+                break;
         default:
                 *out_ring = NULL;
                 r = -EINVAL;