Re: [PATCH] kdb: Reject zero BYTESPERWORD before calculating the repeat count

From: Doug Anderson

Date: Fri Nov 13 2026 - 12:23:51 EST


Hi,

On Fri, Nov 13, 2026 at 3:12 AM Alex Kazinsky <a.kazinsky@xxxxxxxx> wrote:
>
> Setting BYTESPERWORD to zero in the kdb environment causes kdb_md()
> to divide by zero when calculating the initial repeat count. The
> existing width validation happens after this division.
>
> Reject a zero width with KDB_BADWIDTH before calculating the repeat
> count.
>
> Found by Linux Verification Center (linuxtesting.org) with SVACE static
> analysis tool.
>
> Fixes: 5d5314d6795f ("kdb: core for kgdb back end (1 of 2)")
> Signed-off-by: Alex Kazinsky <a.kazinsky@xxxxxxxx>
> ---
> kernel/debug/kdb/kdb_main.c | 3 +++
> 1 file changed, 3 insertions(+)
>
> diff --git a/kernel/debug/kdb/kdb_main.c b/kernel/debug/kdb/kdb_main.c
> index 40c332d2179d..c3cbeb25f6fb 100644
> --- a/kernel/debug/kdb/kdb_main.c
> +++ b/kernel/debug/kdb/kdb_main.c
> @@ -1527,6 +1527,9 @@ static int kdb_md(int argc, const char **argv)
> kdbgetintenv("RADIX", &radix);
> kdbgetintenv("BYTESPERWORD", &bytesperword);
>
> + if (!bytesperword)
> + return KDB_BADWIDTH;
> +

Could you check `bytesperword <= 0` rather than `!bytesperword` (and
update the commit message accordingly)?

AI-generated details, in case they are interesting to you:

kdbgetintenv() parses an `unsigned long` with kstrtoul() and casts it to
`int`, so on 64-bit systems `set BYTESPERWORD=0xffffffff` gives
`bytesperword = -1`. Combined with `set MDCOUNT=0x8000000` (so `mdcount
* 16` wraps to `INT_MIN`, `-2147483648`), `mdcount * 16 / bytesperword`
computes `INT_MIN / -1`. Because `+2147483648` overflows a 32-bit signed
`int` (`INT_MAX` is `2147483647`), x86's signed `idiv` instruction
raises the same `#DE` divide error exception as dividing by zero.

-Doug