On Wed, Oct 22, 2008 at 02:30:50PM +0800, Xu, Anthony wrote:
> Isaku Yamahata wrote:
> > On Wed, Oct 22, 2008 at 02:11:52PM +0800, Xu, Anthony wrote:
> >> Isaku Yamahata wrote:
> >>> On Wed, Oct 22, 2008 at 01:56:05PM +0800, Xu, Anthony wrote:
> >>>> Yes, it is not SMP-safe there is lock for p2m.
> >>>> Modifying p2m is not a frequent operation, why not add a lock for
> >>>> it?
> >>> It is frequent to read p2m table. So lockless approach was adopted
> >>> for scalability. It doesn't make sense to lock around only writer
> >>> side.
> >> If only add write lock for p2m, is there any bad impact/senario?
> >> Can you explain more details?
> > Generally lock should protect both readers and writers.
> > So locking around only writers doesn't make sense.
> So you can use read/write lock, multiple reader one writer.
> Because write is very rare, it will not impact performance, but it makes code
> in mm.c clear and easy to modify.
> I think that's why read/write lock exist.
Yes, that's quite right.
It's another discussion to go for reader/writer lock or
to keep the current lockless approach.
Xen-ia64-devel mailing list