On Thu, Jun 14, 2012 at 10:57 AM, Minchan Kim <minchan@xxxxxxxxxx> wrote: > Hi KOSAKI, > > Sorry for late response. > Let me ask a question about description. > > On Thu, Jun 14, 2012 at 04:13:12AM -0400, kosaki.motohiro@xxxxxxxxx wrote: >> From: KOSAKI Motohiro <kosaki.motohiro@xxxxxxxxxxxxxx> >> >> Currently, do_try_to_free_pages() can enter livelock. Because of, >> now vmscan has two conflicted policies. >> >> 1) kswapd sleep when it couldn't reclaim any page when reaching >> priority 0. This is because to avoid kswapd() infinite >> loop. That said, kswapd assume direct reclaim makes enough >> free pages to use either regular page reclaim or oom-killer. >> This logic makes kswapd -> direct-reclaim dependency. >> 2) direct reclaim continue to reclaim without oom-killer until >> kswapd turn on zone->all_unreclaimble. This is because >> to avoid too early oom-kill. >> This logic makes direct-reclaim -> kswapd dependency. >> >> In worst case, direct-reclaim may continue to page reclaim forever >> when kswapd sleeps forever. > > I have tried imagined scenario you mentioned above with code level but > unfortunately I got failed. > If kswapd can't meet high watermark on order-0, it doesn't sleep if I don't miss something. pgdat_balanced() doesn't recognized zone. Therefore kswapd may sleep if node has multiple zones. Hm ok, I realized my descriptions was slightly misleading. priority 0 is not needed. bakance_pddat() calls pgdat_balanced() every priority. Most easy case is, movable zone has a lot of free pages and normal zone has no reclaimable page. btw, current pgdat_balanced() logic seems not correct. kswapd should sleep only if every zones have much free pages than high water mark _and_ 25% of present pages in node are free. > So if kswapd sleeps, it means we already have enough order-0 free pages. > Hmm, could you describe scenario you found in detail with code level? > > Anyway, as I look at your patch, I can't find any problem. > I just want to understand scenario you mentioned completely in my head. > Maybe It can help making description clear. > -- 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/ . Don't email: <a href