Re: kern/59663 (ffs_snapshot_read -> uvm_fault (or pool page empty) doing dump/restore with snapshot)

["Henryk Paluch via gnats" <gnats-admin%NetBSD.org@localhost>]

The following reply was made to PR kern/59663; it has been noted by GNATS.

From: Henryk Paluch <hpaluch%seznam.cz@localhost>
To: gnats-bugs%netbsd.org@localhost
Cc: 
Subject: Re: kern/59663 (ffs_snapshot_read -> uvm_fault (or pool page empty)
 doing dump/restore with snapshot)
Date: Wed, 22 Jul 2026 21:03:15 +0200

 I performed manual kernel bisection (building kernel from CVS for each 
 tag and booting on target machine where fssconfig & dd tests were run
 
 Example build for RC3 (on machine called nb11rc6-fss-src - kernel builder)
 
 cd /usr/src/
 cvs -q -z2 -d anoncvs%anoncvs.NetBSD.org@localhost:/cvsroot update -r 
 netbsd-11-0-RC3 -dP
 rm -rf /usr/src/sys/arch/amd64/compile/HP
 cd /usr/src/sys/arch/amd64/conf
 config HP # see notes below
 cd ../compile/HP
 make depend && make -j 1C
 gzip -9v netbsd.gdb netbsd netbsd-HP.debug netbsd.map
 scp *.gz tgt:kernel-11.0_RC3 # copy to TARGET machine where crashes are 
 tested, machine called nb11rc6-fss-tgt
 
 My config (called HP) has these changes from GENERIC (enabled all debug 
 options):
 
 options       DIAGNOSTIC      # inexpensive kernel consistency checks
 options       DEBUG           # expensive debugging checks/support
 options       LOCKDEBUG       # expensive locking checks/support
 
 Both build machine and target machine is official 
 NetBSD-11.0_RC6-amd64-dvd.iso, testing Target just boots different RC 
 kernels. Both machines are VMs under Linux KVM. Target is using Host 
 snapshot to ensure that after crash it is rolled back to consistent state.
 
 Results are following (and confusing):
 
 * 11.0_RC1 quick crash (matches this report)
 * 11.0_RC2 skipped (not part of bisection)
 * 11.0_RC3 quick crash (matches this report)
 * 11.0_RC4 quick crash (matches this report) - LAST ONE
 
 here happened something in sources that significantly reduced pool 
 corruption panics
 
 * 11.0_RC5 nearly stable - pool panics are very rare (but still exist)
 * 11.0_RC6 no panic occurred so far, but kernel freezes after few 
 minutes of dd (only local console shows keystrokes, but everything else 
 is frozen) - is seems that logout from other console triggers it(?)
 
 Below is "systat vmstat" session when it last updated on RC6:
 
      3 users    Load  0.97  0.39  0.15                  Wed Jul 22 18:34:59
 
 Proc:r  d  s        Csw  Traps SysCal  Intr   Soft  Fault     PAGING 
 SWAPPING
       1     1      55568          3113    47    291            in  out 
 in  out
                                                          ops
    67.8% Sy   0.0% Us   0.0% Ni   0.0% In  32.2% Id    pages
 |    |    |    |    |    |    |    |    |    |    |
 ================================== 
   forks
   
    fkppw
 Anon        22024   1%   zero                147 Interrupts 
   fksvm
 Exec        10772    %   wired                   TLB shootdown 
   pwait
 File        31468   1%   inact               100 cpu0 timer 
   relck
 Meta       297308  -1%   bufs   148439           ioapic0 pin 4 
   rlkok
   (kB)        real   swaponly      free           ioapic0 pin 1 
    noram
 Active      64264              1535108        46 msix0 vec 1 
   ndcpy
 Namei         Sys-cache     Proc-cache           msix1 vec 0 
   fltcp
      Calls     hits    %     hits     %           ioapic0 pin 22 
    zfod
          6        6  100                        1 msix2 vec 1 
    cow
   
 512 fmin
    Disks:     ld0     cd0    fss0 
 682 ftarg
   seeks 
    itarg
   xfers         2           36002 
    flnan
   bytes       34K          72004K 
    pdfre
   %busy                      50.0 
    pdscn
 
 
 Note: that Meta percent (-1%) is quite suspicious...
 
 I'm now a bit confused at what direction should I go to get closer to 
 cause of these problems. Any clues?