On 2024/05/02 4:52, Dmitry Osipenko wrote:
On 5/1/24 22:38, Dmitry Osipenko wrote:
On 5/1/24 22:31, Dmitry Osipenko wrote:
On 4/27/24 10:12, Akihiko Odaki wrote:
int virtio_gpu_virgl_get_num_capsets(VirtIOGPU *g)
{
uint32_t capset2_max_ver, capset2_max_size;
+
+ if (g->capset_ids) {
Move capset_ids initialization to virtio_gpu_virgl_init() to save this
conditional.
Capsets are used before virgl is inited. At first guest queries virtio
device features and then enables virgl only if capset is available.
While virgl itself is initialized when first virtio command is
processed. I.e. it's not possible to move to virtio_gpu_virgl_init.
Though no, capsets aren't part of device features. I'll move it to
virtio_gpu_virgl_init, thanks.
Number of capsets actually is a part of generic virtio device cfg
descriptor. Capsets initialization can't be moved without probing
capsets twice, i.e. not worthwhile.
I see. Then I suggest replacing virtio_gpu_virgl_get_num_capsets() with
a function that returns GArray of capset IDs.
virtio_gpu_gl_device_realize() will assign the returned GArray to
g->capset_ids. virtio_gpu_gl_device_unrealize(), which doesn't exist
yet, will free g->capset_ids later.
This way, you won't need the conditional, and it will be clear that a
GArray allocation happens in virtio_gpu_gl_device_realize() and is
matched with the deallocation in virtio_gpu_gl_device_unrealize().
