Re: [PATCH v2 3/7] fbcon: Implement struct consw.con_unblank

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] fbcon: missing mode restore on VT switch due to par->graphics desync
--

commit c18deeb283bd933a2a74f5f89b8adc5baff05746
Author: Thomas Zimmermann <tzimmermann@suse.de>

fbcon: Implement struct consw.con_unblank

Implement con_unblank from struct consw with the !blank branches from
fbcon_blank(). The new function is called fbcon_unblank(). Remove the
duplicated code from fbcon_blank().

> diff --git a/drivers/video/fbdev/core/fbcon.c b/drivers/video/fbdev/core/fbcon.c
> index caad4e06944ea..6b6f162c90dd3 100644
> --- a/drivers/video/fbdev/core/fbcon.c
> +++ b/drivers/video/fbdev/core/fbcon.c
> @@ -2293,32 +2293,53 @@ static bool fbcon_blank(struct vc_data *vc, enum vesa_blank_mode blank,
[ ... ]
> +static bool fbcon_unblank(struct vc_data *vc)
> +{
> +	struct fb_info *info = fbcon_info_from_console(vc->vc_num);
> +	struct fbcon_par *par = info->fbcon_par;
> +	bool mode_switch = par->graphics && vc->vc_mode == KD_TEXT;

[Severity: High]
Can par->graphics become out of sync and prevent a required mode restore?

If vt_kdsetmode(KD_GRAPHICS) is called on a background console, it returns
early without calling do_blank_screen(), leaving par->graphics as false.

When switching to this VT, if the previous foreground VT was also in
KD_GRAPHICS mode, complete_change_console() in drivers/tty/vt/vt_ioctl.c skips
do_blank_screen() because old_vc_mode matches vc->vc_mode:

	if (old_vc_mode != vc->vc_mode) {
		if (vc->vc_mode == KD_TEXT)
			do_unblank_screen(1);
		else
			do_blank_screen(1);
	}

Because do_blank_screen() is skipped, par->graphics remains false for the new
framebuffer. When the application later exits and calls ioctl(KDSETMODE,
KD_TEXT), fbcon_unblank() is triggered.

Since par->graphics is false, mode_switch evaluates to false, skipping
fb_set_var() with FB_ACTIVATE_KD_TEXT. Without FB_ACTIVATE_KD_TEXT, DRM
helpers (like drm_fb_helper_set_par()) won't force a mode restore before the
DRM master drops its locks, which could leave the display frozen or corrupted.

> +
> +	if (mode_switch) {
> +		struct fb_var_screeninfo var = info->var;
> +
> +		var.activate = FB_ACTIVATE_NOW | FB_ACTIVATE_FORCE | FB_ACTIVATE_KD_TEXT;
> +		fb_set_var(info, &var);
> +		par->graphics = false;
> +		par->var = info->var;
> +	}

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261016093357.33294-1-tzimmermann@suse.de?part=3