> Il giorno 13 feb 2018, alle ore 06:02, Mike Galbraith <efault@xxxxxx> ha scritto: > > On Mon, 2018-02-12 at 13:48 +0100, Oleksandr Natalenko wrote: >> Hi. >> >> 12.02.2018 12:22, Mike Galbraith wrote: >>> How does that work without someone actually submitting patches? CC >>> stable and pass along a conveniently sorted cherry-pick list? >> >> From my experience, Greg is fine with the list of hashes, but he needs >> an approval from the relevant subtree maintainer (Jens in this case). > > Oh dear. I'm a bit reluctant to even mention this, buuut, when > creating that hash list, I extracted/applied them to 4.15.3 via quilt > to make sure that they really do slip right in.. and later, while > building two fat enterprise config kernels concurrently, trees living > in separate spinning rust buckets, box IO stalled on me. Not good. > Hard reboot, clean up and try it again, it reproduced. Switch to > master, that reproduced. Extra not good. > > I found that rather odd, because I had earlier done this to 4.15.. > > patches/block-bfq:-fix-occurrences-of-request-finish-method-s-old-name.patch > patches/block-bfq:-add-requeue-request-hook.patch > > ..ie take only the rename, wedge the fix in on top of it, beat on box, > it worked fine. This morning, I go back to only those two, and box is > again (at least pretending to be) perfectly happy to build the same > trees it hung on twice in a row yesterday, it _seems_ to be completely > uninterested in stalling no matter what I do. > > So, something not so funny is still going on, here at least. Test > results imply that something in that hash list _may_ not be all that > desirable a thing to send to stable, but who knows, some playful > gremlin may well be pulling my chain, will let box stall on me again a > few seconds after I write that, and poke send :) > > Ok IO gremlin, go for it... > Any chance your gremlin grants you some OOPS? BTW, the only other commit in that series that affects bfq interaction with the rest of the system is: a52a69ea89dc block, bfq: limit tags for writes and async I/O And the other commits that do something beyond changing some calculation are: 0d52af590552 block, bfq: release oom-queue ref to root group on exit 52257ffbfcaf block, bfq: put async queues for root bfq groups too Thanks, Paolo > -Mike