On Fri, Aug 12, 2011 at 06:57:41PM -0700, Paul E. McKenney wrote: > But if we are getting much below 100 milliseconds, we need to rethink > this. The delay may be low in some corner case but this still benefits by running it only once. You can mmap() bzero, mremap(+4096) (if mremap moves the pages to a not aligned 2m address, it forces a split_huge_page, an hardware issue) and all pages will be splitted in potentially less than 100msec if they're only a few. At least we'll be running synchronize_rcu only once instead of for every hugepage, as long as it runs only once I guess we're ok. Normally it shouldn't happen so fast. My current /proc/vmstat says there are 271 splits for 97251 THP allocated and they're not so likely to have happened within 100msec. -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/ Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>