Messages in this thread | | | Subject | Re: [RESEND PATCH 1/2 -mm] mm: account lazy free pages separately | From | Yang Shi <> | Date | Wed, 14 Aug 2019 21:54:23 -0700 |
| |
On 8/14/19 5:55 AM, Vlastimil Babka wrote: > On 8/12/19 7:00 PM, Yang Shi wrote: >>> I can see that memcg rss size was the primary problem David was looking >>> at. But MemAvailable will not help with that, right? Moreover is >> Yes, but David actually would like to have memcg MemAvailable (the >> accounter like the global one), which should be counted like the global >> one and should account per memcg deferred split THP properly. >> >>> accounting the full THP correct? What if subpages are still mapped? >> "Deferred split" definitely doesn't mean they are free. When memory >> pressure is hit, they would be split, then the unmapped normal pages >> would be freed. So, when calculating MemAvailable, they are not >> accounted 100%, but like "available += lazyfree - min(lazyfree / 2, >> wmark_low)", just like how page cache is accounted. >> >> We could get more accurate account, i.e. checking each sub page's >> mapcount when accounting, but it may change before shrinker start >> scanning. So, just use the ballpark estimation to trade off the >> complexity for accurate accounting. > If we know the mapcounts in the moment the deferred split is initiated (I > suppose there has to be a iteration over all subpages already?), we could get > the exact number to adjust the counter with, and also store the number somewhere > (e.g. a unused field in first/second tail page, I think we already do that for > something). Then in the shrinker we just read that number to adjust the counter > back. Then we can ignore the subpage mapping changes before shrinking happens, > they shouldn't change the situation significantly, and importantly we we will be > safe from counter imbalance thanks to the stored number.
Thanks, I'm going to look into this approach. Thanks for the suggestion again.
| |