On Mon, Jul 09, 2018 at 06:46:49PM +0300, Haneen Mohammed wrote:
> This patch map/unmap GEM backing memory to kernel address space
> in prepare/cleanup_fb respectively and cache the virtual address
> for later use.
>
> Signed-off-by: Haneen Mohammed <[email protected]>
> ---
> drivers/gpu/drm/vkms/vkms_plane.c | 41 +++++++++++++++++++++++++++++++
> 1 file changed, 41 insertions(+)
>
> diff --git a/drivers/gpu/drm/vkms/vkms_plane.c
> b/drivers/gpu/drm/vkms/vkms_plane.c
> index f7f63143f6d0..f464eb0e05dd 100644
> --- a/drivers/gpu/drm/vkms/vkms_plane.c
> +++ b/drivers/gpu/drm/vkms/vkms_plane.c
> @@ -9,6 +9,7 @@
> #include "vkms_drv.h"
> #include <drm/drm_plane_helper.h>
> #include <drm/drm_atomic_helper.h>
> +#include <drm/drm_gem_framebuffer_helper.h>
>
> static const struct drm_plane_funcs vkms_plane_funcs = {
> .update_plane = drm_atomic_helper_update_plane,
> @@ -30,9 +31,49 @@ static void vkms_primary_plane_update(struct drm_plane
> *plane,
> {
> }
>
> +static int vkms_prepare_fb(struct drm_plane *plane,
> + struct drm_plane_state *state)
> +{
> + struct drm_gem_object *gem_obj;
> + struct vkms_gem_object *vkms_obj;
> +
> + if (!state->fb)
> + return 0;
> +
> + gem_obj = drm_gem_fb_get_obj(state->fb, 0);
> + vkms_obj = container_of(gem_obj, struct vkms_gem_object, gem);
> + vkms_obj->vaddr = vkms_gem_vmap(gem_obj);
> +
> + if (!vkms_obj->vaddr)
> + DRM_INFO("vmap failed\n");
> +
> + return drm_gem_fb_prepare_fb(plane, state);
> +}
> +
> +static void vkms_cleanup_fb(struct drm_plane *plane,
> + struct drm_plane_state *old_state)
> +{
> + struct drm_gem_object *gem_obj;
> + struct vkms_gem_object *vkms_obj;
> +
> + if (!old_state->fb)
> + return;
> +
> + gem_obj = drm_gem_fb_get_obj(old_state->fb, 0);
> + vkms_obj = container_of(gem_obj, struct vkms_gem_object, gem);
> +
> + if (vkms_obj && vkms_obj->pages) {
> + vunmap(vkms_obj->vaddr);
Please also clear obj->vaddr here.
> + drm_gem_put_pages(gem_obj, vkms_obj->pages, false, true);
> + vkms_obj->pages = NULL;
I think it'd be good to also have a vkms_gem_vunmap helper for this, for
symmetry and prettier code.
There's another issue: cleanup/prepare_fb can be called multiple times on
the same fb. This doesn't really happen for the current vkms case where we
only have 1 plane, but if you have more than one plane you can attach the
same fb to all of them. This means we need a vmap counter (probably best
put into the vkms_gem_v(un)map functions), and only create the actual vmap
when the counter goes from 0 -> 1, and release it only when it goes down
to 0 finally. Usually it's called vmap_use_count or so.
Ofc that needs to be protected by the buffer lock, so you need to rework
your locking a bit and pull it out into the vmap/vunmap functions.
Cheers, Daniel
> + }
> +}
> +
> static const struct drm_plane_helper_funcs vkms_primary_helper_funcs = {
> .atomic_check = vkms_plane_atomic_check,
> .atomic_update = vkms_primary_plane_update,
> + .prepare_fb = vkms_prepare_fb,
> + .cleanup_fb = vkms_cleanup_fb,
> };
>
> struct drm_plane *vkms_plane_init(struct vkms_device *vkmsdev)
> --
> 2.17.1
>
--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
