On Fri, Nov 02, 2018 at 12:01:22PM -0400, Sasha Levin wrote:
On Fri, Nov 02, 2018 at 02:45:42AM +0000, Dexuan Cui wrote:
From: Roman Gushchin guro@fb.com Sent: Thursday, November 1, 2018 17:58
On Fri, Nov 02, 2018 at 12:16:02AM +0000, Dexuan Cui wrote: Hello, Dexuan!
A couple of issues has been revealed recently, here are fixes (hashes are from the next tree):
5f4b04528b5f mm: don't reclaim inodes with many attached pages 5a03b371ad6a mm: handle no memcg case in memcg_kmem_charge() properly
These two patches should be added to the serie.
Thanks for the new info!
Re stable backporting, I'd really wait for some time. Memory reclaim is a quite complex and fragile area, so even if patches are correct by themselves, they can easily cause a regression by revealing some other issues (as it was with the inode reclaim case).
I totally agree. I'm now just wondering if there is any temporary workaround, even if that means we have to run the kernel with some features disabled or with a suboptimal performance?
I'm not sure what workload you're seeing it on, but if you could merge these 7 patches and see that it solves the problem you're seeing and doesn't cause any regressions it'll be a useful test for the rest of us.
AFAIK, with Roman's patches backported to Ubuntu version of 4.15, the problem reported at [1] is solved.
[1] https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1792349
-- Thanks, Sasha