venus: make device memory alloc async

Add a new perf option NO_ASYNC_MEM_ALLOC. Track the ring seqno of the
memory alloc command, and do async ring wait to ensure:
- memory allocation is before resource creation
- memory import is before resource destroy

Signed-off-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25611>

diff --git a/src/virtio/vulkan/vn_common.c b/src/virtio/vulkan/vn_common.c
index 90620f1..2dddd76 100644 (file)
--- a/src/virtio/vulkan/vn_common.c
+++ b/src/virtio/vulkan/vn_common.c
@@ -48,6 +48,7 @@ static const struct debug_control vn_perf_options[] = {
    { "no_cmd_batching", VN_PERF_NO_CMD_BATCHING },
    { "no_timeline_sem_feedback", VN_PERF_NO_TIMELINE_SEM_FEEDBACK },
    { "no_query_feedback", VN_PERF_NO_QUERY_FEEDBACK },
+   { "no_async_mem_alloc", VN_PERF_NO_ASYNC_MEM_ALLOC },
    { NULL, 0 },
    /* clang-format on */
 };

diff --git a/src/virtio/vulkan/vn_common.h b/src/virtio/vulkan/vn_common.h
index 774d344..7c51c61 100644 (file)
--- a/src/virtio/vulkan/vn_common.h
+++ b/src/virtio/vulkan/vn_common.h
@@ -118,6 +118,7 @@ enum vn_perf {
    VN_PERF_NO_CMD_BATCHING = 1ull << 6,
    VN_PERF_NO_TIMELINE_SEM_FEEDBACK = 1ull << 7,
    VN_PERF_NO_QUERY_FEEDBACK = 1ull << 8,
+   VN_PERF_NO_ASYNC_MEM_ALLOC = 1ull << 9,
 };
 
 typedef uint64_t vn_object_id;

diff --git a/src/virtio/vulkan/vn_device_memory.c b/src/virtio/vulkan/vn_device_memory.c
index 4149990..cfb32c7 100644 (file)
--- a/src/virtio/vulkan/vn_device_memory.c
+++ b/src/virtio/vulkan/vn_device_memory.c
@@ -28,8 +28,20 @@ vn_device_memory_alloc_simple(struct vn_device *dev,
 {
    VkDevice dev_handle = vn_device_to_handle(dev);
    VkDeviceMemory mem_handle = vn_device_memory_to_handle(mem);
-   return vn_call_vkAllocateMemory(dev->instance, dev_handle, alloc_info,
-                                   NULL, &mem_handle);
+   if (VN_PERF(NO_ASYNC_MEM_ALLOC)) {
+      return vn_call_vkAllocateMemory(dev->instance, dev_handle, alloc_info,
+                                      NULL, &mem_handle);
+   }
+
+   struct vn_instance_submit_command instance_submit;
+   vn_submit_vkAllocateMemory(dev->instance, 0, dev_handle, alloc_info, NULL,
+                              &mem_handle, &instance_submit);
+   if (!instance_submit.ring_seqno_valid)
+      return VK_ERROR_OUT_OF_HOST_MEMORY;
+
+   mem->bo_ring_seqno_valid = true;
+   mem->bo_ring_seqno = instance_submit.ring_seqno;
+   return VK_SUCCESS;
 }
 
 static inline void
@@ -42,6 +54,48 @@ vn_device_memory_free_simple(struct vn_device *dev,
 }
 
 static VkResult
+vn_device_memory_wait_alloc(struct vn_device *dev,
+                            struct vn_device_memory *mem)
+{
+   if (!mem->bo_ring_seqno_valid)
+      return VK_SUCCESS;
+
+   /* fine to false it here since renderer submission failure is fatal */
+   mem->bo_ring_seqno_valid = false;
+
+   uint32_t local_data[8];
+   struct vn_cs_encoder local_enc =
+      VN_CS_ENCODER_INITIALIZER_LOCAL(local_data, sizeof(local_data));
+   vn_encode_vkWaitRingSeqnoMESA(&local_enc, 0, dev->instance->ring.id,
+                                 mem->bo_ring_seqno);
+   return vn_renderer_submit_simple(dev->renderer, local_data,
+                                    vn_cs_encoder_get_len(&local_enc));
+}
+
+static inline VkResult
+vn_device_memory_bo_init(struct vn_device *dev,
+                         struct vn_device_memory *mem,
+                         VkExternalMemoryHandleTypeFlags external_handles)
+{
+   VkResult result = vn_device_memory_wait_alloc(dev, mem);
+   if (result != VK_SUCCESS)
+      return result;
+
+   return vn_renderer_bo_create_from_device_memory(
+      dev->renderer, mem->size, mem->base.id, mem->type.propertyFlags,
+      external_handles, &mem->base_bo);
+}
+
+static inline void
+vn_device_memory_bo_fini(struct vn_device *dev, struct vn_device_memory *mem)
+{
+   if (mem->base_bo) {
+      vn_device_memory_wait_alloc(dev, mem);
+      vn_renderer_bo_unref(dev->renderer, mem->base_bo);
+   }
+}
+
+static VkResult
 vn_device_memory_pool_grow_alloc(struct vn_device *dev,
                                  uint32_t mem_type_index,
                                  VkDeviceSize size,
@@ -68,9 +122,7 @@ vn_device_memory_pool_grow_alloc(struct vn_device *dev,
    if (result != VK_SUCCESS)
       goto obj_fini;
 
-   result = vn_renderer_bo_create_from_device_memory(
-      dev->renderer, mem->size, mem->base.id, mem->type.propertyFlags, 0,
-      &mem->base_bo);
+   result = vn_device_memory_bo_init(dev, mem, 0);
    if (result != VK_SUCCESS)
       goto mem_free;
 
@@ -342,9 +394,7 @@ vn_device_memory_alloc_export(struct vn_device *dev,
    if (result != VK_SUCCESS)
       return result;
 
-   result = vn_renderer_bo_create_from_device_memory(
-      dev->renderer, mem->size, mem->base.id, mem->type.propertyFlags,
-      external_handles, &mem->base_bo);
+   result = vn_device_memory_bo_init(dev, mem, external_handles);
    if (result != VK_SUCCESS) {
       vn_device_memory_free_simple(dev, mem);
       return result;
@@ -587,8 +637,8 @@ vn_FreeMemory(VkDevice device,
    if (mem->base_memory) {
       vn_device_memory_pool_unref(dev, mem->base_memory);
    } else {
-      if (mem->base_bo)
-         vn_renderer_bo_unref(dev->renderer, mem->base_bo);
+      /* ensure renderer side import still sees the resource */
+      vn_device_memory_bo_fini(dev, mem);
 
       if (mem->bo_roundtrip_seqno_valid)
          vn_instance_wait_roundtrip(dev->instance, mem->bo_roundtrip_seqno);
@@ -646,9 +696,7 @@ vn_MapMemory(VkDevice device,
     * the extension.
     */
    if (need_bo) {
-      result = vn_renderer_bo_create_from_device_memory(
-         dev->renderer, mem->size, mem->base.id, mem->type.propertyFlags, 0,
-         &mem->base_bo);
+      result = vn_device_memory_bo_init(dev, mem, 0);
       if (result != VK_SUCCESS)
          return vn_error(dev->instance, result);
    }

diff --git a/src/virtio/vulkan/vn_device_memory.h b/src/virtio/vulkan/vn_device_memory.h
index cc4d470..3e0e0c0 100644 (file)
--- a/src/virtio/vulkan/vn_device_memory.h
+++ b/src/virtio/vulkan/vn_device_memory.h
@@ -32,6 +32,21 @@ struct vn_device_memory {
    /* non-NULL when mappable or external */
    struct vn_renderer_bo *base_bo;
 
+   /* ensure renderer side resource create is called after vkAllocateMemory
+    *
+    * 1. driver submits vkAllocateMemory (alloc) via ring for a ring seqno
+    * 2. driver submits via vq to wait for above ring to reach the seqno
+    * 3. driver creates virtgpu bo from renderer VkDeviceMemory
+    *
+    * ensure renderer side resource destroy is called after vkAllocateMemory
+    *
+    * 1. driver submits vkAllocateMemory (import) via ring for a ring seqno
+    * 2. driver submits via vq to wait for above ring to reach the seqno
+    * 3. driver destroys virtgpu bo
+    */
+   bool bo_ring_seqno_valid;
+   uint32_t bo_ring_seqno;
+
    /* ensure renderer side vkFreeMemory is called after vkGetMemoryFdKHR
     *
     * 1. driver creates virtgpu bo from renderer VkDeviceMemory