Dave Chinner wrote: > > OTOH (contradicting myself here), writepages, in essence writebacks, are > > performed by per-BDI flusher threads which are kicked by the mm code in > > low memory situations, as opposed to the thread performing the allocation. > > > > As Tetsuo pointed out, direct reclaims are the real problematic scenarios. > > Sure, but I've been saying for more than 10 years we need to get rid > of direct reclaim because it's horribly inefficient when there's > lots of concurrent allocation pressure, not to mention it's full of > deadlock scenarios like this. > > Really, though I'm tired of having the same arguments over and over > again about architectural problems that people just don't seem to > understand or want to fix. > Yeah, it is sad that developers are not interested in lowmem situation. Suspect the MM subsystem when your Linux system hung up!? https://elinux.org/images/4/49/CELFJP-Jamboree63-handa-en.pdf
