On Mon, Mar 01, 2010 at 07:07:29PM +0100, Peter Zijlstra wrote: > On Mon, 2010-03-01 at 18:58 +0100, Andrea Arcangeli wrote: > > > > But nothing risks to break at build time, simply any arch with > > transparent hugepage support also has to implement > > __get_user_pages_fast. Disabling irq and using __get_user_pages_fast > > looked the best way to serialize against split_huge_page here (rather > > than taking locks). > > PowerPC uses rcu-free'd pagetables for gup_fast() iirc, so no need to > disable IRQs there. btw, powerpc right now doesn't support transparent hugepages. IRQs have to be disabled to stop pmdp_splitting_flush_notify on x86, for the same reason why x86 has to disable irq in gup-fast. I surely can try to abstract it before there is a second arch involved, but it'd be more natural in the process of porting to an arch that wants rcu_read_lock instead of local_irq_disable. -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxxx For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>