venus: allow renderer submissions on per-context virtgpu rings

For submissions to renderers that support multiple timelines, put
them on the virtgpu fencing timeline (dma fence context) specified by
the VkQueue's bound ring_idx. CPU-sync'd renderer submissions
can be sent in the same manner by using ring_idx = 0.

Signed-off-by: Ryan Neph <ryanneph@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19691>

diff --git a/src/virtio/vulkan/vn_renderer.h b/src/virtio/vulkan/vn_renderer.h
index 6143c90..c08bc1e 100644 (file)
--- a/src/virtio/vulkan/vn_renderer.h
+++ b/src/virtio/vulkan/vn_renderer.h
@@ -86,20 +86,18 @@ struct vn_renderer_submit_batch {
    size_t cs_size;
 
    /*
-    * Submit cs to the virtual sync queue identified by sync_queue_index.  The
-    * virtual queue is assumed to be associated with the physical VkQueue
-    * identified by vk_queue_id.  After the execution completes on the
-    * VkQueue, the virtual sync queue is signaled.
+    * Submit cs to the timeline identified by ring_idx. A timeline is
+    * typically associated with a physical VkQueue and bound to the ring_idx
+    * during VkQueue creation. After execution completes on the VkQueue, the
+    * timeline sync point is signaled.
     *
-    * sync_queue_index must be less than max_sync_queue_count.
-    *
-    * vk_queue_id specifies the object id of a VkQueue.
-    *
-    * When sync_queue_cpu is true, it specifies the special CPU sync queue,
-    * and sync_queue_index/vk_queue_id are ignored.  TODO revisit this later
+    * ring_idx 0 is reserved for the context-specific CPU timeline. sync
+    * points on the CPU timeline are signaled immediately after command
+    * processing by the renderer.
     */
-   uint32_t sync_queue_index;
-   bool sync_queue_cpu;
+   uint32_t ring_idx;
+
+   // TODO remove once vtest supports multiple timelines
    vn_object_id vk_queue_id;
 
    /* syncs to update when the virtual sync queue is signaled */

diff --git a/src/virtio/vulkan/vn_renderer_util.c b/src/virtio/vulkan/vn_renderer_util.c
index d2654b9..4c8809a 100644 (file)
--- a/src/virtio/vulkan/vn_renderer_util.c
+++ b/src/virtio/vulkan/vn_renderer_util.c
@@ -21,7 +21,7 @@ vn_renderer_submit_simple_sync(struct vn_renderer *renderer,
          &(const struct vn_renderer_submit_batch){
             .cs_data = cs_data,
             .cs_size = cs_size,
-            .sync_queue_cpu = true,
+            .ring_idx = 0, /* CPU ring */
             .syncs = &sync,
             .sync_values = &(const uint64_t){ 1 },
             .sync_count = 1,

diff --git a/src/virtio/vulkan/vn_renderer_util.h b/src/virtio/vulkan/vn_renderer_util.h
index a579790..8fd8747 100644 (file)
--- a/src/virtio/vulkan/vn_renderer_util.h
+++ b/src/virtio/vulkan/vn_renderer_util.h
@@ -27,6 +27,7 @@ vn_renderer_submit_simple(struct vn_renderer *renderer,
          &(const struct vn_renderer_submit_batch){
             .cs_data = cs_data,
             .cs_size = cs_size,
+            .ring_idx = 0, /* CPU ring */
          },
       .batch_count = 1,
    };

diff --git a/src/virtio/vulkan/vn_renderer_virtgpu.c b/src/virtio/vulkan/vn_renderer_virtgpu.c
index 306e943..fdf0850 100644 (file)
--- a/src/virtio/vulkan/vn_renderer_virtgpu.c
+++ b/src/virtio/vulkan/vn_renderer_virtgpu.c
@@ -509,6 +509,8 @@ sim_submit_alloc_gem_handles(struct vn_renderer_bo *const *bos,
 static int
 sim_submit(struct virtgpu *gpu, const struct vn_renderer_submit *submit)
 {
+   const bool use_ring_idx = gpu->base.info.supports_multiple_timelines;
+
    /* TODO replace submit->bos by submit->gem_handles to avoid malloc/loop */
    uint32_t *gem_handles = NULL;
    if (submit->bo_count) {
@@ -523,11 +525,13 @@ sim_submit(struct virtgpu *gpu, const struct vn_renderer_submit *submit)
       const struct vn_renderer_submit_batch *batch = &submit->batches[i];
 
       struct drm_virtgpu_execbuffer args = {
-         .flags = batch->sync_count ? VIRTGPU_EXECBUF_FENCE_FD_OUT : 0,
+         .flags = (batch->sync_count ? VIRTGPU_EXECBUF_FENCE_FD_OUT : 0) |
+                  (use_ring_idx ? VIRTGPU_EXECBUF_RING_IDX : 0),
          .size = batch->cs_size,
          .command = (uintptr_t)batch->cs_data,
          .bo_handles = (uintptr_t)gem_handles,
          .num_bo_handles = submit->bo_count,
+         .ring_idx = (use_ring_idx ? batch->ring_idx : 0),
       };
 
       ret = drmIoctl(gpu->fd, DRM_IOCTL_VIRTGPU_EXECBUFFER, &args);
@@ -539,7 +543,7 @@ sim_submit(struct virtgpu *gpu, const struct vn_renderer_submit *submit)
       if (batch->sync_count) {
          ret = sim_submit_signal_syncs(gpu, args.fence_fd, batch->syncs,
                                        batch->sync_values, batch->sync_count,
-                                       batch->sync_queue_cpu);
+                                       batch->ring_idx == 0);
          close(args.fence_fd);
          if (ret)
             break;

diff --git a/src/virtio/vulkan/vn_renderer_vtest.c b/src/virtio/vulkan/vn_renderer_vtest.c
index 8d37a65..0cb475d 100644 (file)
--- a/src/virtio/vulkan/vn_renderer_vtest.c
+++ b/src/virtio/vulkan/vn_renderer_vtest.c
@@ -540,9 +540,9 @@ vtest_vcmd_submit_cmd2(struct vtest *vtest,
          .sync_offset = sync_offset / sizeof(uint32_t),
          .sync_count = batch->sync_count,
       };
-      if (!batch->sync_queue_cpu) {
+      if (vtest->base.info.supports_multiple_timelines) {
          dst.flags = VCMD_SUBMIT_CMD2_FLAG_SYNC_QUEUE;
-         dst.sync_queue_index = batch->sync_queue_index;
+         dst.sync_queue_index = batch->ring_idx;
          dst.sync_queue_id = batch->vk_queue_id;
       }
       vtest_write(vtest, &dst, sizeof(dst));