From: Yang Shi <firstname.lastname@example.org> To: Hugh Dickins <email@example.com> Cc: Andrew Morton <firstname.lastname@example.org>, Shakeel Butt <email@example.com>, "Kirill A. Shutemov" <firstname.lastname@example.org>, Miaohe Lin <email@example.com>, Mike Kravetz <firstname.lastname@example.org>, Michal Hocko <email@example.com>, Rik van Riel <firstname.lastname@example.org>, Christoph Hellwig <email@example.com>, Matthew Wilcox <firstname.lastname@example.org>, "Eric W. Biederman" <email@example.com>, Alexey Gladkov <firstname.lastname@example.org>, Chris Wilson <email@example.com>, Matthew Auld <firstname.lastname@example.org>, Linux FS-devel Mailing List <email@example.com>, Linux Kernel Mailing List <firstname.lastname@example.org>, email@example.com, Linux MM <firstname.lastname@example.org> Subject: Re: [PATCH 06/16] huge tmpfs: shmem_is_huge(vma, inode, index) Date: Thu, 12 Aug 2021 11:19:34 -0700 [thread overview] Message-ID: <CAHbLzkpd1r1kLhNP7=Una_Fxpdgx7vE9aeyBgqHRE8M5e9j-qQ@mail.gmail.com> (raw) In-Reply-To: <CAHbLzkou+6m+htMNzSQrHfd6U0yURWiewK=Pvg30XSdiWemail@example.com> On Fri, Aug 6, 2021 at 10:57 AM Yang Shi <firstname.lastname@example.org> wrote: > > On Thu, Aug 5, 2021 at 10:43 PM Hugh Dickins <email@example.com> wrote: > > > > On Thu, 5 Aug 2021, Yang Shi wrote: > > > > > > By rereading the code, I think you are correct. Both cases do work > > > correctly without leaking. And the !CONFIG_NUMA case may carry the > > > huge page indefinitely. > > > > > > I think it is because khugepaged may collapse memory for another NUMA > > > node in the next loop, so it doesn't make too much sense to carry the > > > huge page, but it may be an optimization for !CONFIG_NUMA case. > > > > Yes, that is its intention. > > > > > > > > However, as I mentioned in earlier email the new pcp implementation > > > could cache THP now, so we might not need keep this convoluted logic > > > anymore. Just free the page if collapse is failed then re-allocate > > > THP. The carried THP might improve the success rate a little bit but I > > > doubt how noticeable it would be, may be not worth for the extra > > > complexity at all. > > > > It would be great if the new pcp implementation is good enough to > > get rid of khugepaged's confusing NUMA=y/NUMA=n differences; and all > > the *hpage stuff too, I hope. That would be a welcome cleanup. > > The other question is if that optimization is worth it nowadays or > not. I bet not too many users build NUMA=n kernel nowadays even though > the kernel is actually running on a non-NUMA machine. Some small > devices may run NUMA=n kernel, but I don't think they actually use > THP. So such code complexity could be removed from this point of view > too. > > > > > > > > Collapse failure is not uncommon and leaking huge pages gets noticed. > > > > After writing that, I realized how I'm almost always testing a NUMA=y > > kernel (though on non-NUMA machines), and seldom try the NUMA=n build. > > So did so to check no leak, indeed; but was surprised, when comparing > > vmstats, that the NUMA=n run had done 5 times as much thp_collapse_alloc > > as the NUMA=y run. I've merely made a note to look into that one day: > > maybe it was just a one-off oddity, or maybe the incrementing of stats > > is wrong down one path or the other. I came up with a patch to remove !CONFIG_NUMA case, and my test found the same problem. NUMA=n run had done 5 times as much thp_collapse_alloc as NUMA=y run with vanilla kernel just exactly as what you saw. A quick look shows the huge page allocation timing is different for the two cases. For NUMA=n, the huge page is allocated by khugepaged_prealloc_page() before scanning the address space, so it means huge page may be allocated even though there is no suitable range for collapsing. Then the page would be just freed if khugepaged already made enough progress then try to reallocate again. The problem should be more noticeable if you have a shorter scan interval (scan_sleep_millisecs). I set it to 100ms for my test. We could carry the huge page across scan passes for NUMA=n, but this would make the code more complicated. I don't think it is really worth, so just removing the special case for NUMA=n sounds more reasonable to me. > > Yeah, probably. > > > > > Hugh
next prev parent reply other threads:[~2021-08-12 18:19 UTC|newest] Thread overview: 45+ messages / expand[flat|nested] mbox.gz Atom feed top 2021-07-30 7:22 [PATCH 00/16] tmpfs: HUGEPAGE and MEM_LOCK fcntls and memfds Hugh Dickins 2021-07-30 7:25 ` [PATCH 01/16] huge tmpfs: fix fallocate(vanilla) advance over huge pages Hugh Dickins 2021-07-30 21:36 ` Yang Shi 2021-08-01 3:38 ` Hugh Dickins 2021-08-02 20:36 ` Yang Shi 2021-07-30 7:28 ` [PATCH 02/16] huge tmpfs: fix split_huge_page() after FALLOC_FL_KEEP_SIZE Hugh Dickins 2021-07-30 23:48 ` Yang Shi 2021-07-30 7:30 ` [PATCH 03/16] huge tmpfs: remove shrinklist addition from shmem_setattr() Hugh Dickins 2021-07-30 21:50 ` Yang Shi 2021-07-30 7:36 ` [PATCH 04/16] huge tmpfs: revert shmem's use of transhuge_vma_enabled() Hugh Dickins 2021-07-30 21:56 ` Yang Shi 2021-08-01 4:01 ` Hugh Dickins 2021-08-02 20:39 ` Yang Shi 2021-07-30 7:39 ` [PATCH 05/16] huge tmpfs: move shmem_huge_enabled() upwards Hugh Dickins 2021-07-30 21:57 ` Yang Shi 2021-07-30 7:42 ` [PATCH 06/16] huge tmpfs: shmem_is_huge(vma, inode, index) Hugh Dickins 2021-07-30 23:34 ` Yang Shi 2021-08-01 5:22 ` Hugh Dickins 2021-08-01 5:37 ` Hugh Dickins 2021-08-02 21:14 ` Yang Shi 2021-08-04 8:28 ` Hugh Dickins 2021-08-04 19:01 ` Yang Shi 2021-08-06 5:21 ` Hugh Dickins 2021-08-06 17:41 ` Yang Shi 2021-08-05 23:04 ` Yang Shi 2021-08-06 5:43 ` Hugh Dickins 2021-08-06 17:57 ` Yang Shi 2021-08-12 18:19 ` Yang Shi [this message] 2021-07-30 7:45 ` [PATCH 07/16] memfd: memfd_create(name, MFD_HUGEPAGE) for shmem huge pages Hugh Dickins 2021-08-04 14:03 ` Kirill A. Shutemov 2021-08-06 3:33 ` Hugh Dickins 2021-07-30 7:48 ` [PATCH 08/16] huge tmpfs: fcntl(fd, F_HUGEPAGE) and fcntl(fd, F_NOHUGEPAGE) Hugh Dickins 2021-08-04 14:08 ` Kirill A. Shutemov 2021-08-06 4:34 ` Hugh Dickins 2021-07-30 7:51 ` [PATCH 09/16] huge tmpfs: decide stat.st_blksize by shmem_is_huge() Hugh Dickins 2021-07-30 23:40 ` Yang Shi 2021-07-30 7:55 ` [PATCH 10/16] tmpfs: fcntl(fd, F_MEM_LOCK) to memlock a tmpfs file Hugh Dickins 2021-08-03 1:38 ` Matthew Wilcox 2021-08-04 9:15 ` Hugh Dickins 2021-07-30 7:57 ` [PATCH 11/16] tmpfs: fcntl(fd, F_MEM_LOCKED) to test if memlocked Hugh Dickins 2021-07-30 8:00 ` [PATCH 12/16] tmpfs: refuse memlock when fallocated beyond i_size Hugh Dickins 2021-07-30 8:03 ` [PATCH 13/16] mm: bool user_shm_lock(loff_t size, struct ucounts *) Hugh Dickins 2021-07-30 8:06 ` [PATCH 14/16] mm: user_shm_lock(,,getuc) and user_shm_unlock(,,putuc) Hugh Dickins 2021-07-30 8:09 ` [PATCH 15/16] tmpfs: permit changing size of memlocked file Hugh Dickins 2021-07-30 8:13 ` [PATCH 16/16] memfd: memfd_create(name, MFD_MEM_LOCK) for memlocked shmem Hugh Dickins
Reply instructions: You may reply publicly to this message via plain-text email using any one of the following methods: * Save the following mbox file, import it into your mail client, and reply-to-all from there: mbox Avoid top-posting and favor interleaved quoting: https://en.wikipedia.org/wiki/Posting_style#Interleaved_style * Reply using the --to, --cc, and --in-reply-to switches of git-send-email(1): git send-email \ --in-reply-to='CAHbLzkpd1r1kLhNP7=Una_Fxpdgx7vE9aeyBgqHRE8M5e9j-qQ@mail.gmail.com' \ --firstname.lastname@example.org \ --email@example.com \ --firstname.lastname@example.org \ --email@example.com \ --firstname.lastname@example.org \ --email@example.com \ --firstname.lastname@example.org \ --email@example.com \ --firstname.lastname@example.org \ --email@example.com \ --firstname.lastname@example.org \ --email@example.com \ --firstname.lastname@example.org \ --email@example.com \ --firstname.lastname@example.org \ --email@example.com \ --firstname.lastname@example.org \ --email@example.com \ --firstname.lastname@example.org \ --subject='Re: [PATCH 06/16] huge tmpfs: shmem_is_huge(vma, inode, index)' \ /path/to/YOUR_REPLY https://kernel.org/pub/software/scm/git/docs/git-send-email.html * If your mail client supports setting the In-Reply-To header via mailto: links, try the mailto: link
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for NNTP newsgroup(s).