Messages in this thread | | | Date | Tue, 11 Nov 2014 22:16:15 +0100 | From | Jan Kara <> | Subject | Re: [BUG] mm/page-writeback.c: divide by zero in pos_ratio_polynom not fixed |
| |
On Tue 11-11-14 16:04:48, Steven Rostedt wrote: > On Tue, 11 Nov 2014 21:15:39 +0100 > Jan Kara <jack@suse.cz> wrote: > > > > So I was looking into this but I have to say I don't understand where is > > the problem. The registers clearly show that x_intercept - bdi_setpoint + 1 > > == 0 (in 32-bit arithmetics). Given: > > x_intercept = bdi_setpoint + span > > > > We have that span + 1 == 0 and that means that: > > ((thresh - bdi_thresh + 8 * write_bw) * (u64)x >> 16) == -1 (again in > > 32-bit arithmetics). But I don't see how that can realistically happen... > > > > Is this reproducible at all? > > > > Unfortunately not. It only happened once, and I haven't been able to > reproduce it again. BTW, how much memory does the machine have and what is /proc/sys/vm/dirty_ratio and /proc/sys/vm/dirty_background_ratio (or corresponding dirty_bytes, dirty_background_bytes if you are using them)?
Honza -- Jan Kara <jack@suse.cz> SUSE Labs, CR
| |