venus: emulate a second graphics queue on Android

Starting from Android 14 (Android U), framework HWUI has required a
second graphics queue to avoid racing between webview and skiavk. For
non-Android, we leave the second queue emulation behind a debug option.

Signed-off-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30985>

src/virtio/vulkan/vn_common.c

@@ -32,6 +32,7 @@ static const struct debug_control vn_debug_options[] = {
    { "cache", VN_DEBUG_CACHE },
    { "no_sparse", VN_DEBUG_NO_SPARSE },
    { "no_gpl", VN_DEBUG_NO_GPL },
+   { "second_queue", VN_DEBUG_SECOND_QUEUE },
    { NULL, 0 },
    /* clang-format on */
 };

src/virtio/vulkan/vn_common.h

@@ -112,6 +112,7 @@ enum vn_debug {
    VN_DEBUG_CACHE = 1ull << 6,
    VN_DEBUG_NO_SPARSE = 1ull << 7,
    VN_DEBUG_NO_GPL = 1ull << 8,
+   VN_DEBUG_SECOND_QUEUE = 1ull << 9,
 };
 
 enum vn_perf {

src/virtio/vulkan/vn_device.c

@@ -40,7 +40,8 @@ static VkResult
 vn_queue_init(struct vn_device *dev,
               struct vn_queue *queue,
               const VkDeviceQueueCreateInfo *queue_info,
-              uint32_t queue_index)
+              uint32_t queue_index,
+              struct vn_queue *shared_queue)
 {
    VkResult result =
       vn_queue_base_init(&queue->base, &dev->base, queue_info, queue_index);
@@ -49,6 +50,16 @@ vn_queue_init(struct vn_device *dev,
 
    vn_cached_storage_init(&queue->storage, &dev->base.base.alloc);
 
+   if (dev->physical_device->emulate_second_queue ==
+          queue_info->queueFamilyIndex &&
+       shared_queue != NULL) {
+      assert(queue_index > 0);
+      queue->emulated = true;
+      queue->base.id = shared_queue->base.id;
+      queue->ring_idx = shared_queue->ring_idx;
+      return VK_SUCCESS;
+   }
+
    const int ring_idx = vn_instance_acquire_ring_idx(dev->instance);
    if (ring_idx < 0) {
       vn_log(dev->instance, "failed binding VkQueue to renderer timeline");
@@ -92,13 +103,15 @@ vn_device_init_queues(struct vn_device *dev,
       return VK_ERROR_OUT_OF_HOST_MEMORY;
 
    count = 0;
+   struct vn_queue *shared_queue = NULL;
    for (uint32_t i = 0; i < create_info->queueCreateInfoCount; i++) {
       VkResult result;
 
       const VkDeviceQueueCreateInfo *queue_info =
          &create_info->pQueueCreateInfos[i];
       for (uint32_t j = 0; j < queue_info->queueCount; j++) {
-         result = vn_queue_init(dev, &queues[count], queue_info, j);
+         result =
+            vn_queue_init(dev, &queues[count], queue_info, j, shared_queue);
          if (result != VK_SUCCESS) {
             for (uint32_t k = 0; k < count; k++)
                vn_queue_fini(&queues[k]);
@@ -107,7 +120,7 @@ vn_device_init_queues(struct vn_device *dev,
             return result;
          }
 
-         count++;
+         shared_queue = &queues[count++];
       }
    }
 
@@ -470,8 +483,25 @@ vn_device_init(struct vn_device *dev,
    if (group && group->physicalDeviceCount)
       dev->device_mask = (1 << group->physicalDeviceCount) - 1;
 
+   VkDeviceCreateInfo final_create_info = *create_info;
+   STACK_ARRAY(VkDeviceQueueCreateInfo, queue_infos,
+               create_info->queueCreateInfoCount);
+   for (uint32_t i = 0; i < create_info->queueCreateInfoCount; i++) {
+      const VkDeviceQueueCreateInfo *queue_info =
+         &create_info->pQueueCreateInfos[i];
+      if (queue_info->queueFamilyIndex ==
+             physical_dev->emulate_second_queue &&
+          queue_info->queueCount == 2) {
+         typed_memcpy(queue_infos, create_info->pQueueCreateInfos,
+                      create_info->queueCreateInfoCount);
+         final_create_info.pQueueCreateInfos = queue_infos;
+         queue_infos[i].queueCount = 1;
+         break;
+      }
+   }
    result = vn_call_vkCreateDevice(dev->primary_ring, physical_dev_handle,
-                                   create_info, NULL, &dev_handle);
+                                   &final_create_info, NULL, &dev_handle);
+   STACK_ARRAY_FINISH(queue_infos);
 
    /* free the fixed extensions here since no longer needed below */
    if (create_info == &local_create_info)
@@ -612,7 +642,8 @@ vn_DestroyDevice(VkDevice device, const VkAllocationCallbacks *pAllocator)
     * are still bound to the queues in the renderer.
     */
    for (uint32_t i = 0; i < dev->queue_count; i++) {
-      vn_instance_release_ring_idx(dev->instance, dev->queues[i].ring_idx);
+      if (!dev->queues[i].emulated)
+         vn_instance_release_ring_idx(dev->instance, dev->queues[i].ring_idx);
    }
 
    vk_free(alloc, dev->queues);

src/virtio/vulkan/vn_physical_device.c

@@ -678,6 +678,27 @@ vn_physical_device_init_queue_family_properties(
    vn_call_vkGetPhysicalDeviceQueueFamilyProperties2(
       ring, vn_physical_device_to_handle(physical_dev), &count, props);
 
+#if DETECT_OS_ANDROID && ANDROID_API_LEVEL >= 34
+   /* Starting from Android 14 (Android U), framework HWUI has required a
+    * second graphics queue to avoid racing between webview and skiavk.
+    */
+   static const bool require_second_queue = true;
+#else
+   const bool require_second_queue = VN_DEBUG(SECOND_QUEUE);
+#endif
+   physical_dev->emulate_second_queue = -1;
+   for (uint32_t i = 0; i < count; i++) {
+      if (props[i].queueFamilyProperties.queueFlags & VK_QUEUE_GRAPHICS_BIT) {
+         if (require_second_queue &&
+             props[i].queueFamilyProperties.queueCount < 2) {
+            props[i].queueFamilyProperties.queueCount = 2;
+            physical_dev->emulate_second_queue = i;
+         }
+
+         break;
+      }
+   }
+
    /* Filter out queue families that exclusively support sparse binding as
     * we need additional support for submitting feedback commands
     */

src/virtio/vulkan/vn_physical_device.h

@@ -83,6 +83,10 @@ struct vn_physical_device {
    VkQueueFamilyProperties2 *queue_family_properties;
    uint32_t queue_family_count;
    bool sparse_binding_disabled;
+   /* Track the queue family index to emulate a second queue. -1 means no
+    * emulation is needed.
+    */
+   int emulate_second_queue;
 
    VkPhysicalDeviceMemoryProperties memory_properties;
 

src/virtio/vulkan/vn_queue.h

@@ -16,6 +16,9 @@
 struct vn_queue {
    struct vn_queue_base base;
 
+   /* emulated queue shares base queue id and ring_idx with another queue */
+   bool emulated;
+
    /* only used if renderer supports multiple timelines */
    uint32_t ring_idx;