Re: [PATCH] fbdev: pm3fb: Fix potential divide by zero
From: Helge Deller <deller@gmx.de>
Date: 2025-07-27 16:28:35
Also in:
dri-devel, lkml
Hi Alex, On 6/12/25 11:29, Geert Uytterhoeven wrote:
On Wed, 11 Jun 2025 at 18:12, Alex Guo [off-list ref] wrote:quoted
quoted
On Sat, 7 Jun 2025 at 22:14, Alex Guo [off-list ref] wrote:quoted
variable var->pixclock can be set by user. In case it equals to zero, divide by zero would occur in pm3fb_check_var. Similar crashes have happened in other fbdev drivers. There is no check and modification on var->pixclock along the call chain to pm3fb_check_var. So we fix this by checking whether 'pixclock' is zero. Similar commit: commit 16844e58704 ("video: fbdev: tridentfb: Error out if 'pixclock' equals zero") Signed-off-by: Alex Guo <redacted>Thanks for your patch, which is now commit 59d1fc7b3e1ae9d4 ("fbdev: pm3fb: fix potential divide by zero") in fbdev/for-next.quoted
--- a/drivers/video/fbdev/pm3fb.c +++ b/drivers/video/fbdev/pm3fb.c@@ -998,6 +998,9 @@ static int pm3fb_check_var(struct fb_var_screeninfo *var, struct fb_info *info) return -EINVAL; } + if (!var->pixclock) + return -EINVAL;While this fixes the crash, this is correct behavior for an fbdev driver. When a value is invalid, it should be rounded up to a valid value instead, if possible.Thanks for your confirmation and suggestions. I added this patch based on existing checks on var->pixclock in other drivers, such as savagefb_check_var, nvidiafb_check_var, etc. Are you suggesting that it is better to replace an invalid value (var->pixclock == 0) with a default valid value, instead of returning -EINVAL?Indeed.quoted
If so, could you advise what a suitable default value would be for this case?The answer is hidden in the existing check below:quoted
quoted
quoted
+ if (PICOS2KHZ(var->pixclock) > PM3_MAX_PIXCLOCK) { DPRINTK("pixclock too high (%ldKHz)\n", PICOS2KHZ(var->pixclock)); return -EINVAL; }It can be replaced by: if (var->pixclock <= KHZ2PICOS(PM3_MAX_PIXCLOCK)) var->pixclock = KHZ2PICOS(PM3_MAX_PIXCLOCK) + 1; The "+ 1" is needed because of rounding.
You sent a whole bunch of patches [1] which check pixclock against zero, but you don't set the default value as Geert pointed out above. Can you maybe revise your patches accordingly? Helge [1] https://patchwork.kernel.org/project/linux-fbdev/list/