Re: [PATCH] mm: prevent divide error for small values of vm_dirty_bytes

[Andrea Righi <righi.andrea@gmail.com>]

On Wed, Apr 29, 2009 at 02:26:11AM -0700, David Rientjes wrote:
> On Wed, 29 Apr 2009, Andrea Righi wrote:
> 
> > Avoid to set less than two pages for vm_dirty_byte: this is necessary to avoid
> > potential division by 0 (like the following) in get_dirty_limits().
> > 
> 
> Where exactly is the divide by 0 in get_dirty_limits()?

in the evaluation of the BDI's share ratio:
...
	if (bdi) {
		u64 bdi_dirty;
		long numerator, denominator;

		/*
		 * Calculate this BDI's share of the dirty ratio.
		 */
		bdi_writeout_fraction(bdi, &numerator, &denominator);

		bdi_dirty = (dirty * (100 - bdi_min_ratio)) / 100;
		bdi_dirty *= numerator;
		do_div(bdi_dirty, denominator);
		^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
...

More exactly:
/home/righiandr/Software/linux/git/linux-2.6/mm/page-writeback.c:474
    1186:       48 89 f0                mov    %rsi,%rax
    1189:       48 f7 f7                div    %rdi

And looking at the trace:

[   49.952195] RAX: 00000000000000c0 RBX: ffff88001de03b80 RCX: 28f5c28f5c28f5c3
[   49.952195] RDX: 0000000000000000 RSI: 00000000000000c0 RDI: 0000000000000000
                                                           ^^^^^^^^^^^^^^^^^^^^^
> 
> Is this because we forced the background threshold to be half of the dirty 
> threshold when its setting is greater?

No (see above).

-Andrea