Re: [ltt-dev] [PATCH] Fix dirty page accounting in redirty_page_for_writepage()

[Mathieu Desnoyers <compudj@krystal.dyndns.org>]

* Christoph Lameter (cl@linux.com) wrote:
> On Fri, 1 May 2009, Mathieu Desnoyers wrote:
> 
> > So, back to my original point : do you agree on the usefulness of
> > separating fallback irq-disabling from the per-cpu atomic construct ?
> 
> No. Percpu operations are used for statistics and are like atomic
> operations. Aggregation of these leads to a can of worms that we better
> leave unopened.
> 
> > x86 would map :
> >
> > percpu_irqsave/restore to "nothing".
> > percpu_add_return_irq to xadd instruction. It is irq-safe by design.
> >
> > Other architectures (fallback) would map
> >
> > percpu_irqsave/restore to local_irq_save/restore.
> > percpu_add_return_irq to var += value; return var;
> 
> Shudder.... We have explored those types of macros before (while doing
> fastpath optimization for SLUB) and it significant increases the
> complexity. People may add additional instructions in between and now
> interrupts off could be on or off depending on the architecture. Sometimes
> percpu_irqsave does nothing. Very difficult to ensure that the usage is
> correct.
> 
> And we have barely any usage case for such macros.
> 

Then do you have a better idea on how to deal with
__inc_zone_state/inc_zone_state without duplicating the code and without
adding useless local_irq_save/restore on x86 ?

Mathieu

-- 
Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68