Forums before death by AOL, social media and spammers... "We can't have nice things"
|    linux.debian.kernel    |    Debian kernel discussions    |    2,884 messages    |
[   << oldest   |   < older   |   list   |   newer >   |   newest >>   ]
|    Message 1,930 of 2,884    |
|    Salvatore Bonaccorso to All    |
|    Bug#1120058: [regression] 0c58a97f919c (    |
|    14 Dec 25 15:30:01    |
   
   XPost: linux.debian.bugs.dist, linux.kernel   
   From: carnil@debian.org   
      
   Hi Joanne,   
      
   In Debian J. Neuschäfer reported an issue where after 0c58a97f919c   
   ("fuse: remove tmp folio for writebacks and internal rb tree") a   
   specific, but admittely not very minimal workload, involving podman   
   contains, x2goserver and a openjdk application restults in   
   suspend-to-ram hang.   
      
   The report is at https://bugs.debian.org/1120058 and information on   
   bisection and the test setup follows:   
      
   On Sun, Nov 30, 2025 at 02:11:13PM +0100, J. Neuschäfer wrote:   
   > On Fri, Nov 28, 2025 at 09:10:25PM +0100, Salvatore Bonaccorso wrote:   
   > > Control: found -1 6.17.8-1   
   > >   
   > > Hi,   
   > >   
   > > On Fri, Nov 28, 2025 at 11:50:48AM +0100, J. Neuschäfer wrote:   
   > > > On Wed, Nov 05, 2025 at 06:09:43AM +0100, Salvatore Bonaccorso wrote:   
   > [...]   
   > > > I can reproduce the bug fairly reliably on 6.16/17 by running a specific   
   > > > podman container plus x2go (not entirely sure which parts of this is   
   > > > necessary).   
   > >   
   > > Okay if you have a very reliable way to reproduce it, would you be   
   > > open to make "your hands bit dirty" and do some bisecting on the   
   > > issue?   
   >   
   > Thank you for your detailed instructions! I've already started and completed   
   > the git bisect run in the meantime. I had to restart a few times due to   
   > mistakes, but I was able to identify the following upstream commit as the   
   > commit that introduced the issue:   
   >   
   > https://git.kernel.org/linus/0c58a97f919c24fe4245015f4375a39ff05665b6   
   >   
   > fuse: remove tmp folio for writebacks and internal rb tree   
   >   
   > The relevant commit history is as follows:   
   >   
   > * 2619a6d413f4c3 Merge tag 'fuse-update-6.16' of git://git   
   kernel.org/pub/scm/linux/kernel/git/mszeredi/fuse <-- bad   
   > |\   
   > | * dabb9039102879 fuse: increase readdir buffer size   
   > | * 467e245d47e666 readdir: supply dir_context.count as readdir buffer   
   size hint   
   > | * c31f91c6af96a5 fuse: don't allow signals to interrupt getdents copying   
   > | * f3cb8bd908c72e fuse: support large folios for writeback   
   > | * 906354c87f4917 fuse: support large folios for readahead   
   > | * ff7c3ee4842d87 fuse: support large folios for queued writes   
   > | * c91440c89fbd9d fuse: support large folios for stores   
   > | * cacc0645bcad3e fuse: support large folios for symlinks   
   > | * 351a24eb48209b fuse: support large folios for folio reads   
   > | * d60a6015e1a284 fuse: support large folios for writethrough writes   
   > | * 63c69ad3d18a80 fuse: refactor fuse_fill_write_pages()   
   > | * 3568a956932621 fuse: support large folios for retrieves   
   > | * 394244b24fdd09 fuse: support copying large folios   
   > | * f09222980d7751 fs: fuse: add dev id to /dev/fuse fdinfo   
   > | * 18ee43c398af0b docs: filesystems: add fuse-passthrough.rst   
   > | * 767c4b82715ad3 MAINTAINERS: update filter of FUSE documentation   
   > | * 69efbff69f89c9 fuse: fix race between concurrent setattrs from   
   multiple nodes   
   > | * 0c58a97f919c24 fuse: remove tmp folio for writebacks and internal rb   
   tree <-- first bad commit   
   > | * 0c4f8ed498cea1 mm: skip folio reclaim in legacy memcg contexts for   
   deadlockable mappings   
   > | * 4fea593e625cd5 fuse: optimize over-io-uring request expiration check   
   > | * 03a3617f92c2a7 fuse: use boolean bit-fields in struct fuse_copy_state   
   > | * a5c4983bb90759 fuse: Convert 'write' to a bit-field in struct   
   fuse_copy_state   
   > | * 2396356a945bb0 fuse: add more control over cache invalidation behaviour   
   > | * faa794dd2e17e7 fuse: Move prefaulting out of hot write path   
   > | * 0486b1832dc386 fuse: change 'unsigned' to 'unsigned int'   
   > * 0fb34422b5c223 Merge tag 'vfs-6.16-rc1.netfs' of git://g   
   t.kernel.org/pub/scm/linux/kernel/git/vfs/vfs <-- good   
   >   
   > The first and last commits shown are merge commits done by Linus Torvalds.   
   The   
   > fuse-update branch was based on v6.15-rc1, under which I can't run my test   
   due   
   > to an unrelated bug, so I ended up merging in 0fb34422b5c223 to test the   
   > commits within the fuse-update branch. e.g.:   
   >   
   > git reset --hard 394244b24fdd09 && git merge 0fb34422b5c223 && make clean   
   && make   
   >   
   >   
   > I have also verified that the issue still happens on v6.18-rc7 but I wasn't   
   > able to revert 0c58a97f919 on top of this release, because a trivial revert   
   > is not possible.   
   >   
   > My test case consists of a few parts:   
   >   
   > - A podman container based on the "debian:13" image (which points to   
   > docker.io/library/debian via /etc/containers/registries.con   
   .d/shortnames.conf),   
   > where I installed x2goserver and a openjdk-21-based application; It runs   
   the   
   > OpenSSH server and port 22 is exposed as localhost:2001   
   > - x2goclient to start a desktop session in the container   
   >   
   > Source code: https://codeberg.org/neuschaefer/re-workspace   
   >   
   > I suspect, but haven't verified, that the X server in the container somehow   
   > uses the FUSE-emulated filesystem in the container to create a file that is   
   > used with mmap (perhaps to create shared pages as frame buffers).   
   >   
   >   
   > Raw bisect notes:   
   >   
   > good:   
   > - v6.12.48+deb13-amd64   
   > - v6.12.59   
   > - v6.12   
   > - v6.14   
   > - v6.15-1304-g14418ddcc2c205   
   > - v6.15-10380-gec71f661a572   
   > - v6.15-10888-gb509c16e1d7cba   
   > - v6.15-rc7-357-g8e86e73626527e   
   > - v6.15-10933-g4c3b7df7844340   
   > - v6.15-10954-gd00a83477e7a8f   
   > - v6.15-rc7-366-g438e22801b1958 (CONFIG_X86_5LEVEL=y)   
   > - v6.15-rc4-126-g07212d16adc7a0   
      
   [continued in next message]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)   
|
[   << oldest   |   < older   |   list   |   newer >   |   newest >>   ]
(c) 1994, bbs@darkrealms.ca