Hi Sascha, On 26.09.24 15:15, Sascha Hauer wrote: > This adds support for the Rockchip VOP2 video core found on the Rockchip > SoCs RK3566, RK3568 and RK3588. The code is based on the Linux driver > and has been heavily stripped down for barebox. Support for the cluster > windows has been removed, also support for YUV modes has been dropped. > > Signed-off-by: Sascha Hauer <s.hauer@xxxxxxxxxxxxxx> With below described adjustments: Tested-by: Ahmad Fatoum <a.fatoum@xxxxxxxxxxxxxx> # rk3566 HDMI > +config DRIVER_VIDEO_ROCKCHIP_HDMI > + bool "Rockchip HDMI driver" > + select VIDEO_VPL > + select OFTREE > + depends on DRIVER_VIDEO_EDID > + depends on DRIVER_VIDEO_DW_HDMI Turn these two into select? Makes it easier to enable. > +static int dw_hdmi_rockchip_probe(struct device *dev) > +{ > + struct dw_hdmi_plat_data *plat_data; > + struct rockchip_hdmi *hdmi; > + int ret; > + > + hdmi = xzalloc(sizeof(*hdmi)); > + > + ret = dev_get_drvdata(dev, (const void **)&plat_data); > + if (ret) > + return ret; We should really phase out dev_get_drvdata. For one, it doesn't do the same thing as in Linux and those casts are ugly. Can you use device_get_match_data instead? > + ret = rockchip_hdmi_parse_dt(hdmi); > + if (ret) { > + if (ret != -EPROBE_DEFER) > + dev_err(hdmi->dev, "Unable to parse OF data\n"); dev_err_probe here and below would be nice, but I assume it's that way in Linux too? > +static struct vop2_win *vop2_find_unused_win(struct vop2 *vop2) > +{ > + int i; > + > + for (i = 0; i < vop2->registered_num_wins; i++) { > + struct vop2_win *win = &vop2->win[i]; I think the logic here is wrong. In a below hunk, we already have the same loop and now we set i = 0 and restart. That means that the same win can be used twice. win has struct fb_info as element though, so in that case, the same framebuffer is registered twice, which leads to an error message on my rk3566. You may want to turn this into a: static struct vop2_win *vop2_next_unused_win(struct vop2 *vop2, int *i) and then start from i. The i++ in the main loop will take care to advance i, so no win is reused again. > + nvp = 0; > + for (i = 0; i < vop2->registered_num_wins; i++) { > + struct vop2_win *win = &vop2->win[i]; > + > + if (vop2->data->soc_id == 3566) { > + /* > + * On RK3566 these windows don't have an independent > + * framebuffer. They share the framebuffer with smart0, > + * esmart0 and cluster0 respectively. > + */ > + switch (win->data->phys_id) { > + case ROCKCHIP_VOP2_SMART1: > + case ROCKCHIP_VOP2_ESMART1: > + case ROCKCHIP_VOP2_CLUSTER1: > + continue; > + } I don't get any overlay framebuffers registered on rk3566. You know what might be the reason? > + for (i = 0; i < vop2->registered_num_wins; i++) { > + struct vop2_win *win = &vop2->win[i]; > + struct vop2_video_port *vp = win->vp; > + int j; > + > + if (win->type != DRM_PLANE_TYPE_PRIMARY) > + continue; On rk3566, you may end up with vp == NULL here, so you probably want to add a `|| !vp` into the above if condition. > + > + ret = vop2_register_plane(vp, win); > + if (ret) > + continue; > + > + for (j = 0; j < overlay_per_vp; j++) { > + win = vop2_find_unused_win(vop2); > + if (!win) > + break; > + > + win->vp = vp; > + > + ret = vop2_register_plane(vp, win); > + if (ret) > + return ret; > + } > + vp->nlayers = j + 1; > + } > + > + return 0; > +} > + > +struct driver vop2_driver = { > + .probe = vop2_bind, > + .name = "rockchip-vop2", > + .of_compatible = vop2_dt_match, > +}; > +late_platform_driver(vop2_driver); Does this have to be a late driver? Can't we let deep probe take care of the dependency handling? > diff --git a/drivers/video/rockchip/rockchip_vop_reg.c b/drivers/video/rockchip/rockchip_vop_reg.c This file is unused. Cheers, Ahmad -- Pengutronix e.K. | | Steuerwalder Str. 21 | http://www.pengutronix.de/ | 31137 Hildesheim, Germany | Phone: +49-5121-206917-0 | Amtsgericht Hildesheim, HRA 2686 | Fax: +49-5121-206917-5555 |