Re: [RFC 0/6]mm: add new LRU list for MADV_FREE pages
From: Johannes Weiner
Date: Tue Jan 31 2017 - 14:00:11 EST
On Sun, Jan 29, 2017 at 09:51:17PM -0800, Shaohua Li wrote:
> We are trying to use MADV_FREE in jemalloc. Several issues are found. Without
> solving the issues, jemalloc can't use the MADV_FREE feature.
> - Doesn't support system without swap enabled. Because if swap is off, we can't
> or can't efficiently age anonymous pages. And since MADV_FREE pages are mixed
> with other anonymous pages, we can't reclaim MADV_FREE pages. In current
> implementation, MADV_FREE will fallback to MADV_DONTNEED without swap enabled.
> But in our environment, a lot of machines don't enable swap. This will prevent
> our setup using MADV_FREE.
> - Increases memory pressure. page reclaim bias file pages reclaim against
> anonymous pages. This doesn't make sense for MADV_FREE pages, because those
> pages could be freed easily and refilled with very slight penality. Even page
> reclaim doesn't bias file pages, there is still an issue, because MADV_FREE
> pages and other anonymous pages are mixed together. To reclaim a MADV_FREE
> page, we probably must scan a lot of other anonymous pages, which is
> inefficient. In our test, we usually see oom with MADV_FREE enabled and nothing
> without it.
Fully agreed, the anon LRU is a bad place for these pages.
> For the first two issues, introducing a new LRU list for MADV_FREE pages could
> solve the issues. We can directly reclaim MADV_FREE pages without writting them
> out to swap, so the first issue could be fixed. If only MADV_FREE pages are in
> the new list, page reclaim can easily reclaim such pages without interference
> of file or anonymous pages. The memory pressure issue will disappear.
Do we actually need a new page flag and a special LRU for them? These
pages are basically like clean cache pages at that point. What do you
think about clearing their PG_swapbacked flag on MADV_FREE and moving
them to the inactive file list? The way isolate+putback works should
not even need much modification, something like clear_page_mlock().
When the reclaim scanner finds anon && dirty && !swapbacked, it can
again set PG_swapbacked and goto keep_locked to move the page back
into the anon LRU to get reclaimed according to swapping rules.
> For the third issue, we can add a separate RSS count for MADV_FREE pages. The
> count will be increased in madvise syscall and decreased in page reclaim (eg,
> unmap). One issue is activate_page(). A MADV_FREE page can be promoted to
> active page there. But there isn't mm_struct context at that place. Iterating
> vma there sounds too silly. The patchset don't fix this issue yet. Hopefully
> somebody can share a hint how to fix this issue.
This problem also goes away if we use the file LRUs.