Re: [PATCH] x86/video: Preserve the default VGA fallback for legacy modes

[Thomas Zimmermann <tzimmermann@suse.de>]

Hi

Am 29.10.26 um 17:23 schrieb Hengyu Liang:
> Commit 596b3678326d ("x86/video: Only fall back to vga_default_device()
> without screen info") made video_is_primary_device() rely exclusively
> on screen_info resources whenever any are available.
>
> However, screen_info_resources() also returns legacy I/O and memory
> ranges for VGA text modes. These ranges are not PCI BARs, so the resource
> lookup fails and the default VGA fallback is skipped.

If the BIOS booted in text mode or VGA graphics mode, the rsp device is 
still the primary display device. So the current matching code is not 
necessarily incorrect. (But also not necessarily correct either. :)


>
> On a SeaBIOS QEMU guest with bochs and VKMS, the boot VGA loses its DRM
> boot_display attribute and fbcon leaves the console on VKMS. This can
> be checked with:
>
>      cat /sys/class/drm/card1/boot_display
>      stty -F /dev/tty1 size
>
> Without the fix, boot_display is missing and tty1 is 48 rows by 128
> columns. With it, boot_display contains 1 and tty1 is 50 rows by 160
> columns.
>
> Use resource-based selection only for linear framebuffers. Fall back to
> vga_default_device() for legacy display modes, while keeping EFI and
> VESA framebuffer ownership authoritative.
>
> Fixes: 596b3678326d ("x86/video: Only fall back to vga_default_device() without screen info")
> Cc: stable@vger.kernel.org
> Signed-off-by: Hengyu Liang <hengyul@cs.unc.edu>
> ---
> Tested in QEMU/KVM with identical KASAN configs: SeaBIOS VGA text mode
> (with and without VKMS), VESA, and OVMF with one and two virtual GPUs.
> All 26 guest-only helper checks pass, including EFI/VESA framebuffer
> ownership, missing framebuffer resources, and addresses above 4 GiB.
> Also build-tested x86-32 with/without CONFIG_SCREEN_INFO, and x86-64
> without CONFIG_SCREEN_INFO.
>
>   arch/x86/video/video-common.c | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/arch/x86/video/video-common.c b/arch/x86/video/video-common.c
> index 8ed82fff7638..b8f83346dfe8 100644
> --- a/arch/x86/video/video-common.c
> +++ b/arch/x86/video/video-common.c
> @@ -45,7 +45,7 @@ bool video_is_primary_device(struct device *dev)
>   
>   #ifdef CONFIG_SCREEN_INFO
>   	numres = screen_info_resources(si, res, ARRAY_SIZE(res));
> -	if (numres > 0) {
> +	if (numres > 0 && __screen_info_has_lfb(screen_info_video_type(si))) {
>   		for (i = 0; i < numres; ++i) {
>   			if (!(res[i].flags & IORESOURCE_MEM))
>   				continue;

Further below in that code is a call to pci_find_resource(). [1]  Why 
does it match with vkms?  Vkms is not even on a PCI device.

[1] 
https://elixir.bootlin.com/linux/v7.3/source/arch/x86/video/video-common.c#L53

Best regards
Thomas


>
> base-commit: c339f4010dea8b0d754f84ace70cd1b089878ef8

-- 
--
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Frankenstr. 146, 90461 Nürnberg, Germany, www.suse.com
GF: Stefan Gaiser, Jochen Jaser, Abhinav Puri, (HRB 36809, AG Nürnberg)