lkml.org 
[lkml]   [2019]   [Jul]   [11]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH V3 3/5] drm/vkms: Decouple crc operations from composer
On Tue, Jun 25, 2019 at 10:37:58PM -0300, Rodrigo Siqueira wrote:
> In the vkms_composer.c, some of the functions related to CRC and compose
> have interdependence between each other. This patch reworks some
> functions inside vkms_composer to make crc and composer computation
> decoupled.
>
> This patch is preparation work for making vkms able to support new
> features.
>
> Signed-off-by: Rodrigo Siqueira <rodrigosiqueiramelo@gmail.com>
> ---
> drivers/gpu/drm/vkms/vkms_composer.c | 49 ++++++++++++++++------------
> 1 file changed, 29 insertions(+), 20 deletions(-)
>
> diff --git a/drivers/gpu/drm/vkms/vkms_composer.c b/drivers/gpu/drm/vkms/vkms_composer.c
> index eb7ea8be1f98..51a270514219 100644
> --- a/drivers/gpu/drm/vkms/vkms_composer.c
> +++ b/drivers/gpu/drm/vkms/vkms_composer.c
> @@ -105,35 +105,31 @@ static void compose_cursor(struct vkms_composer *cursor_composer,
> primary_composer, cursor_composer);
> }
>
> -static uint32_t _vkms_get_crc(struct vkms_composer *primary_composer,
> - struct vkms_composer *cursor_composer)
> +static int compose_planes(void **vaddr_out,
> + struct vkms_composer *primary_composer,
> + struct vkms_composer *cursor_composer)
> {
> struct drm_framebuffer *fb = &primary_composer->fb;
> struct drm_gem_object *gem_obj = drm_gem_fb_get_obj(fb, 0);
> struct vkms_gem_object *vkms_obj = drm_gem_to_vkms_gem(gem_obj);
> - void *vaddr_out = kzalloc(vkms_obj->gem.size, GFP_KERNEL);
> - u32 crc = 0;
>
> - if (!vaddr_out) {
> - DRM_ERROR("Failed to allocate memory for output frame.");
> - return 0;
> + if (!*vaddr_out) {
> + *vaddr_out = kzalloc(vkms_obj->gem.size, GFP_KERNEL);

Uh allocating memory here isn't great, since you effectily can't handle
the error at all. Also for big resolutions kzalloc will fall back to
kvmalloc I think, which is rather expensive to set up.

But I guess this is a preexisting issue, so welp.

What I would do is pull out the allocation at least, so that
compose_planes really only dos composing, can never fail because it
doesn't need to allocate memory.

> + if (!*vaddr_out) {
> + DRM_ERROR("Cannot allocate memory for output frame.");
> + return -ENOMEM;
> + }
> }
>
> - if (WARN_ON(!vkms_obj->vaddr)) {
> - kfree(vaddr_out);
> - return crc;
> - }
> + if (WARN_ON(!vkms_obj->vaddr))
> + return -EINVAL;
>
> - memcpy(vaddr_out, vkms_obj->vaddr, vkms_obj->gem.size);
> + memcpy(*vaddr_out, vkms_obj->vaddr, vkms_obj->gem.size);
>
> if (cursor_composer)
> - compose_cursor(cursor_composer, primary_composer, vaddr_out);
> + compose_cursor(cursor_composer, primary_composer, *vaddr_out);
>
> - crc = compute_crc(vaddr_out, primary_composer);
> -
> - kfree(vaddr_out);
> -
> - return crc;
> + return 0;
> }
>
> /**
> @@ -154,9 +150,11 @@ void vkms_composer_worker(struct work_struct *work)
> struct vkms_output *out = drm_crtc_to_vkms_output(crtc);
> struct vkms_composer *primary_composer = NULL;
> struct vkms_composer *cursor_composer = NULL;
> + void *vaddr_out = NULL;
> u32 crc32 = 0;
> u64 frame_start, frame_end;
> bool crc_pending;
> + int ret;
>
> spin_lock_irq(&out->composer_lock);
> frame_start = crtc_state->frame_start;
> @@ -180,14 +178,25 @@ void vkms_composer_worker(struct work_struct *work)
> if (crtc_state->num_active_planes == 2)
> cursor_composer = crtc_state->active_planes[1]->composer;
>
> - if (primary_composer)
> - crc32 = _vkms_get_crc(primary_composer, cursor_composer);
> + if (!primary_composer)
> + return;
> +
> + ret = compose_planes(&vaddr_out, primary_composer, cursor_composer);
> + if (ret) {
> + if (ret == -EINVAL)
> + kfree(vaddr_out);
> + return;
> + }
> +
> + crc32 = compute_crc(vaddr_out, primary_composer);
>
> /*
> * The worker can fall behind the vblank hrtimer, make sure we catch up.
> */
> while (frame_start <= frame_end)
> drm_crtc_add_crc_entry(crtc, true, frame_start++, &crc32);
> +
> + kfree(vaddr_out);

Especially since you're freeing the memory _outside_ of compose_planes.

Aside: This all kinda doesn't go in the right direction for
high-performance composing, so I guess I need to get started with typing
up what that should look like.
-Daniel

> }
>
> static const char * const pipe_crc_sources[] = {"auto"};
> --
> 2.21.0

--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch

\
 
 \ /
  Last update: 2019-07-11 10:20    [W:0.136 / U:55.320 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site