Build date: 1757800803 - Sat Sep 13 22:00:03 UTC 2025
Build cvs date: 1757781775 - Sat Sep 13 16:42:55 UTC 2025
Build id: 2025-09-14.1
Build tags: amd64-regress ports sysupgrade

Applied the following diff(s):
/home/anton/tmp/robsd/src-sys-em.diff
/home/anton/tmp/robsd/src-sys-uhidev-sispm.diff
/home/anton/tmp/robsd/src-sysupgrade.diff

P sys/arch/amd64/amd64/vmm_machdep.c
P sys/arch/amd64/include/hibernate_var.h
P sys/netinet6/nd6.c
P sys/netinet6/nd6.h
M usr.sbin/bgpd/session.c

commit VmXIlI0jVguCdIoO
Author: dv
Date: 2025/09/13 16:42:55

  vmm(4): emulate PKRU XSAVE area and features.

  vmm(4) has supported guests using memory protection keys based on
  OpenBSD's implementation on Intel amd64 devices. Recent releases
  of Linux distros use kernel builds expecting to save and restore
  the PKRU state using XSAVE/XRSTOR. This leads to Linux guests
  panicking on boot as they expect to see certain CPUID features.

  This change adds support to the CPUID emulation to describe the
  proper size and offset required for PKRU state so a guest can choose
  to use XSAVE/XRSTOR to handle PKRU on context switches.

  This does not change how the OpenBSD host uses XSAVE/XRSTOR. vmm(4)
  continues to directly save and restore PKRU using WRPKRU/RDPKRU
  instructions on guest entry and exit.

  ok mlarkin@

  sys/arch/amd64/amd64/vmm_machdep.c

commit L01rNxC6XYSfD2Dq
Author: mlarkin
Date: 2025/09/13 13:43:47

  Increase the size of the unhibernate chunk table

  The unhibernate chunk table was previously capable of managing machines up
  to approximately 102GB RAM. On machines with more memory than this,
  unhibernate failed. This doubles the size of the chunktable to account
  for larger RAM sizes.

  ok krw, deraadt

  sys/arch/amd64/include/hibernate_var.h

commit d8lCIA8ExwyhwD8b
Author: bluhm
Date: 2025/09/12 23:02:36

  Iterate over ND6 llinfo list without net lock.

  Make looping over llinfo list in nd6_timer() and nd6_purge() MP
  safe.  Use an iterator like in ARP timer.  With that traversal will
  not interfere with any inserts into or removes from the list.  In
  nd6_timer() take net lock only after a timeout has occured.  Between
  releasing the mutex and grabbing the net lock, another thread may
  free the ND6 entry.  To avoid this race, check RTF_LLINFO in
  nd6_llinfo_timer() while holding the exclusive net lock.

  OK mvs@

  sys/netinet6/nd6.c
  sys/netinet6/nd6.h