Re: [CRASH] in tulip driver?

[Jussi Laako <jussi.laako@kolumbus.fi>]

[-- Attachment #1: Type: text/plain, Size: 1402 bytes --]

On Thu, 2002-07-11 at 06:34, Andrew Morton wrote:
>
> whoops.  I think the "-ll" means low-latency.  But the only
> finger I have in that pie is:
> 
> --- 2.4.19-pre6/drivers/char/random.c~low-latency       Fri Apr  5 12:11:17 2002
> +++ 2.4.19-pre6-akpm/drivers/char/random.c      Fri Apr  5 12:11:17 2002
> @@ -1369,6 +1369,11 @@ static ssize_t extract_entropy(struct en
>                 buf += i;
>                 ret += i;
>                 add_timer_randomness(&extract_timer_state, nbytes);
> +#if LOWLATENCY_NEEDED
> +               /* This can happen in softirq's, but that's what we want */
> +               if (conditional_schedule_needed())
> +                       break;
> +#endif
>         }
>  
>         /* Wipe data just returned from memory */
> 
> So it's a bit of a mystery.  It seems to think that it has
> EXTRACT_ENTROPY_USER.

Whoops, thanks, I found the bug. My fault...

That "break;" breaks some (apparently broken) programs that don't expect
read of /dev/urandom to return early. For security resons (to get
identical behaviour compared to the original kernel) I made a fix that
someone proposed. That fix is apparently broken on some rare situations
which seem to be difficult to trigger (requires high overall irq rates
with network load). Now I'm going to remove that part completely and see
what happens next...


	- Jussi Laako


[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 232 bytes --]