On 09.09.2015 13:20, Gerd Hoffmann wrote:
> Add virglrenderer library detection.  Add 3d mode to virtio-gpu,
> wire up virglrenderer library.  When in 3d mode render using the
> new context management and texture scanout callbacks.
> 
> Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
> ---
>  configure                      |  32 +++
>  hw/display/Makefile.objs       |   6 +-
>  hw/display/virtio-gpu-3d.c     | 598 +++++++++++++++++++++++++++++++++++++++++
>  hw/display/virtio-gpu.c        | 130 ++++++++-
>  include/hw/virtio/virtio-gpu.h |  22 +-
>  trace-events                   |   8 +
>  6 files changed, 785 insertions(+), 11 deletions(-)
>  create mode 100644 hw/display/virtio-gpu-3d.c

As always when it comes to virtio-gpu, I'm just a very, very naive
reviewer, thus I only have "local" things to complain about.

[snip]

> diff --git a/hw/display/virtio-gpu-3d.c b/hw/display/virtio-gpu-3d.c
> new file mode 100644
> index 0000000..ac62fb4
> --- /dev/null
> +++ b/hw/display/virtio-gpu-3d.c
> @@ -0,0 +1,598 @@

[snip]

> +static void virgl_resource_detach_backing(VirtIOGPU *g,
> +                                          struct virtio_gpu_ctrl_command *cmd)
> +{
> +    struct virtio_gpu_resource_detach_backing detach_rb;
> +    struct iovec *res_iovs = NULL;
> +    int num_iovs = 0;
> +
> +    VIRTIO_GPU_FILL_CMD(detach_rb);
> +    trace_virtio_gpu_cmd_res_back_detach(detach_rb.resource_id);
> +
> +    virgl_renderer_resource_detach_iov(detach_rb.resource_id,
> +                                       &res_iovs,
> +                                       &num_iovs);
> +    if (res_iovs == NULL || num_iovs == 0) {
> +        return;
> +    }
> +    virtio_gpu_cleanup_mapping_iov(res_iovs, num_iovs);

Is res_iovs leaked here?

> +}

[snip]

> diff --git a/hw/display/virtio-gpu.c b/hw/display/virtio-gpu.c
> index a67d927..4dbdd12 100644
> --- a/hw/display/virtio-gpu.c
> +++ b/hw/display/virtio-gpu.c

[snip]

> @@ -45,6 +62,27 @@ static void update_cursor_data_simple(VirtIOGPU *g,
>             pixels * sizeof(uint32_t));
>  }
>  
> +#ifdef CONFIG_VIRGL
> +
> +static void update_cursor_data_virgl(VirtIOGPU *g,
> +                                     struct virtio_gpu_scanout *s,
> +                                     uint32_t resource_id)
> +{
> +    uint32_t width, height;
> +    uint32_t pixels, *data;
> +
> +    data = virgl_renderer_get_cursor_data(resource_id, &width, &height);
> +    if (!data) {
> +        return;
> +    }
> +
> +    pixels = s->current_cursor->width * s->current_cursor->height;
> +    memcpy(s->current_cursor->data, data, pixels * sizeof(uint32_t));
> +    free(data);

width and height are unused; should they be compared against
s->current_cursor->{width,height} to spot discrepancies?

> +}
> +
> +#endif
> +
>  static void update_cursor(VirtIOGPU *g, struct virtio_gpu_update_cursor *cursor)
>  {
>      struct virtio_gpu_scanout *s;

[snip]

> @@ -92,9 +131,23 @@ static void virtio_gpu_set_config(VirtIODevice *vdev, const uint8_t *config)
>  static uint64_t virtio_gpu_get_features(VirtIODevice *vdev, uint64_t features,
>                                          Error **errp)
>  {
> +    VirtIOGPU *g = VIRTIO_GPU(vdev);
> +
> +    if (virtio_gpu_virgl_enabled(g->conf)) {
> +        features |= (1 << VIRTIO_GPU_FEATURE_VIRGL);
> +    }
>      return features;
>  }
>  
> +static void virtio_gpu_set_features(VirtIODevice *vdev, uint64_t features)
> +{
> +    static const uint32_t virgl = (1 << VIRTIO_GPU_FEATURE_VIRGL);
> +    VirtIOGPU *g = VIRTIO_GPU(vdev);
> +
> +    g->use_virgl_renderer = ((features & virgl) == virgl);

Could a non-well-behaving guest just set this feature bit even if it was
not reported by virtio_gpu_get_features() because it has been disabled?
This would then result in g->use_virgl_renderer being set to true, which
would have two implications:

(1) In case CONFIG_VIRGL is not defined, no implication right now. But
    maybe in the future there may be cases where the flag is queried
    even with !defined(CONFIG_VIRGL).
(2) virgl may be disabled even with CONFIG_VIRGL being defined; namely
    with HOST_WORDS_BIGENDIAN. That would probably be a problem, because
    we'd then call the virgl functions even though we meant not to.

Max