venus: ensure shmem is attached to renderer before use for guest vram

For guest vram, there's already roundtrip to protect device memory alloc
ordering. This change adds the same protection for shmem used in below
scenarios and optimize to wait for new shmem only.
- reply shmem
- indirect upload shmem
- cmd stream shmem

Signed-off-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28147>

src/virtio/vulkan/vn_command_buffer.c

@@ -1037,6 +1037,9 @@ vn_cmd_submit(struct vn_command_buffer *cmd)
       return;
    }
 
+   if (vn_cs_encoder_needs_roundtrip(&cmd->cs))
+      vn_ring_roundtrip(ring);
+
    if (vn_ring_submit_command_simple(ring, &cmd->cs) != VK_SUCCESS) {
       cmd->state = VN_COMMAND_BUFFER_STATE_INVALID;
       return;

src/virtio/vulkan/vn_cs.c

@@ -296,3 +296,19 @@ vn_cs_encoder_commit(struct vn_cs_encoder *enc)
 
    vn_cs_encoder_sanity_check(enc);
 }
+
+bool
+vn_cs_encoder_needs_roundtrip(struct vn_cs_encoder *enc)
+{
+   if (!enc->instance->renderer->info.has_guest_vram)
+      return false;
+
+   /* check pinned shmems for new shmem allocs */
+   for (uint32_t i = 0; i < enc->buffer_count; i++) {
+      const struct vn_renderer_shmem *shmem = enc->buffers[i].shmem;
+      if (shmem && !shmem->cache_timestamp)
+         return true;
+   }
+
+   return false;
+}

src/virtio/vulkan/vn_cs.h

@@ -192,6 +192,9 @@ vn_cs_encoder_write(struct vn_cs_encoder *enc,
 void
 vn_cs_encoder_commit(struct vn_cs_encoder *enc);
 
+bool
+vn_cs_encoder_needs_roundtrip(struct vn_cs_encoder *enc);
+
 static inline void
 vn_cs_decoder_init(struct vn_cs_decoder *dec, const void *data, size_t size)
 {

src/virtio/vulkan/vn_ring.c

@@ -569,6 +569,9 @@ vn_ring_cs_upload_locked(struct vn_ring *ring, const struct vn_cs_encoder *cs)
    vn_cs_encoder_write(upload, cs_size, cs_data, cs_size);
    vn_cs_encoder_commit(upload);
 
+   if (vn_cs_encoder_needs_roundtrip(upload))
+      vn_ring_roundtrip(ring);
+
    return upload;
 }
 
@@ -657,6 +660,10 @@ vn_ring_submit_command(struct vn_ring *ring,
          ring->instance, submit->reply_size, &reply_offset);
       if (!submit->reply_shmem)
          return;
+
+      if (ring->instance->renderer->info.has_guest_vram &&
+          !submit->reply_shmem->cache_timestamp)
+         vn_ring_roundtrip(ring);
    }
 
    mtx_lock(&ring->mutex);