[lkml]   [2019]   [Apr]   [25]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: powerpc hugepage leak caused by 576ed913 "block: use bio_add_page in bio_iov_iter_get_pages"
On Fri, Apr 26, 2019 at 11:05:17AM +1000, David Gibson wrote:
> On Thu, Apr 25, 2019 at 08:19:58AM +0200, Christoph Hellwig wrote:
> > Just curious: What exact trees do you see this with? This area
> > changed a lot with the multipage bvec support, and subsequent fixes.
> So, I tried it with 576ed913 itself and with 576ed913^ to verify that
> it didn't happen there. The problem also occurred with Linus' tree as
> of when I started bisecting, which appears to have been 444fe991.
> Actually, come to that, here's the whole bisect log in case it's
> helpful:
> # git bisect log
> git bisect start
> # good: [bebc6082da0a9f5d47a1ea2edc099bf671058bd4] Linux 4.14
> git bisect good bebc6082da0a9f5d47a1ea2edc099bf671058bd4
> # bad: [444fe991353987c1c9bc5ab1f903d01f1b4ad415] Merge tag 'riscv-for-linus-5.1-rc6' of git://
> git bisect bad 444fe991353987c1c9bc5ab1f903d01f1b4ad415
> # good: [399c4129eba6145924ab90363352b7bdcd554751] Merge tag 'pxa-for-4.19-dma_slave_map' of
> git bisect good 399c4129eba6145924ab90363352b7bdcd554751
> # bad: [73b6f96cbc0162787bcbdac5f538167084c8d605] Merge branch 'drm-fixes-4.20' of git:// into drm-fixes
> git bisect bad 73b6f96cbc0162787bcbdac5f538167084c8d605
> # good: [85a585918fb4122ad26b6febaec5c3c90bf2535c] Merge tag 'loadpin-security-next' of into next-loadpin
> git bisect good 85a585918fb4122ad26b6febaec5c3c90bf2535c
> # bad: [3acbd2de6bc3af215c6ed7732dfc097d1e238503] Merge tag 'sound-4.20-rc1' of git://
> git bisect bad 3acbd2de6bc3af215c6ed7732dfc097d1e238503
> # good: [8f18da47211554f1ef674fef627c05f23b75a8e0] Merge branch 'master' of git://
> git bisect good 8f18da47211554f1ef674fef627c05f23b75a8e0
> # bad: [0d1b82cd8ac2e8856ae9045c97782ac1c359929c] Merge branch 'ras-core-for-linus' of git://
> git bisect bad 0d1b82cd8ac2e8856ae9045c97782ac1c359929c
> # bad: [1650ac53066577a5e83fe3e9d992c9311597ff8c] Merge tag 'mmc-v4.20' of git://
> git bisect bad 1650ac53066577a5e83fe3e9d992c9311597ff8c
> # bad: [6ab9e09238fdfd742fe23b81e2d385a1cab49d9b] Merge tag 'for-4.20/block-20181021' of git://
> git bisect bad 6ab9e09238fdfd742fe23b81e2d385a1cab49d9b
> # good: [528985117126f11beea339cf39120ee99da04cd2] Merge tag 'arm64-upstream' of git://
> git bisect good 528985117126f11beea339cf39120ee99da04cd2
> # bad: [2cf99bbd106f89fc72f778e8ad9d5538f1ef939b] lightnvm: pblk: add helpers for chunk addresses
> git bisect bad 2cf99bbd106f89fc72f778e8ad9d5538f1ef939b
> # bad: [33b14f67a4e1eabd219fd6543da8f15ed86b641c] nvme: register ns_id attributes as default sysfs groups
> git bisect bad 33b14f67a4e1eabd219fd6543da8f15ed86b641c
> # bad: [27ca1d4ed04ea29dc77b47190a3cc82697023e76] block: move req_gap_back_merge to blk.h
> git bisect bad 27ca1d4ed04ea29dc77b47190a3cc82697023e76
> # bad: [07b05bcc3213ac9f8c28c9d835b4bf3d5798cc60] blkcg: convert blkg_lookup_create to find closest blkg
> git bisect bad 07b05bcc3213ac9f8c28c9d835b4bf3d5798cc60
> # good: [cbeb869a3d1110450186b738199963c5e68c2a71] block, bfq: correctly charge and reset entity service in all cases
> git bisect good cbeb869a3d1110450186b738199963c5e68c2a71
> # bad: [576ed9135489c723fb39b97c4e2c73428d06dd78] block: use bio_add_page in bio_iov_iter_get_pages
> git bisect bad 576ed9135489c723fb39b97c4e2c73428d06dd78
> # good: [c8765de0adfcaaf4ffb2d951e07444f00ffa9453] blok, bfq: do not plug I/O if all queues are weight-raised
> git bisect good c8765de0adfcaaf4ffb2d951e07444f00ffa9453
> # first bad commit: [576ed9135489c723fb39b97c4e2c73428d06dd78] block: use bio_add_page in bio_iov_iter_get_pages
> The problem also occurred with the RHEL8 downstream kernel tree.
> That's based on 4.18, but has 576ed913 backported.
> > So I'd be really curious if it can be reproduced with Jens' latest
> > block for-5.2 tree (which should be in latest linux-next).
> I'll see if I can try that when I next get access to the machine.

Ok, I've now had a chance to test the next-20190423 tree.

I can still reproduce the problem: in fact it is substantially worse,
and somewhat more consistent.

Previously I usually lost 2-3 hugepages per run, though I'd
occasionally seen other values between 0 and 8. With the next tree, I
lost 46 hugepages on most runs, though I also saw 45 and 48

Hope this is illuminating.

David Gibson | I'll have my music baroque, and my code
david AT | minimalist, thank you. NOT _the_ _other_
| _way_ _around_!
[unhandled content-type:application/pgp-signature]
 \ /
  Last update: 2019-04-26 05:57    [W:0.136 / U:0.356 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site