Showstopper bugs and PRs

• Any issues found by syzbot may potentially crash the kernel and should be investigated and fixed.
• PR #55272: userland watchdog processes may be outstalled (regression since -9)
• PR #55466: rump_server processes busy looping can stall a machine (regression since -9)
• topdown uvm mmap assertion failure (PR upcoming) - PMAP_PREFER can cause KASSERT failure (might be a gcc bug, only reproducable on sh3)
• PR #56329: nvme(4) takes long time to umount analyzed: the number of dirty vnodes has been bumped, no real performance regression
• handling of randomness and seed file need to be decided (more below, waiting for randot)
• autoconfig API changes need to be backed out or made type-safe see this backout request
• PR #56653: kernel crash in ipv6 fragment reassembly
• PR #56713: kqueue/kevent does not work with null mounts

Regressions since netbsd-9

• PR #54913: fexecve(2) panics in chroot(8)
• PR #55370: Invalid modification/connection time with mount_null(8)
• PR #55489: Infinite loop in "rump ifconfig bridge0 create"
• PR #55509: npm install -g electron crashes the kernel
• PR #55567: tcp-send slows down to slow single byte transfers
• PR #55997: lang/perl5 fails to build on NetBSD-current under COMPAT_NETBSD32 (Perl bug)
• PR #56073: reading from /dev/pad panics the kernel with an assertion failure
• PR #56113: short reads on AF_UNIX/SOCK_STREAM sockets return early, breaks jackd (maybe a jack bug, fixed in newer jack...)
• PR #55667: XEN3_DOM0 fails to boot on Supermicro AS-2113S-WN24RT
• PR #56291: XEN3_DOM0 nvme abysmal performance
• PR #56404: ACPI poweroff fails
• PR #56117: New test failures on amd64 since switch to gcc10
• PR #56414: cmake hang on kqueue
• PR #56573: kernel memory leak in i915drmkms

PRs worth fixing

• PR #51734: hdaudio(4) "RIRB timeout" interrupt storm on boot
• PR #53173: "go test net/http" locks up the machine
• PR #53155: Wedge after <12h uptime when >2 bnx network interfaces in use
• PR #53291: GPT prevents installation
• PR #53128: Boot says "default boot twice, skipping..."
• PR #54345: kernel diagnostic assertion "xfer != NULL" in wdc.c
• PR #54727: writing a large file causes unreasonable system behaviour (should be fixed, needs to be checked) (fixed in -current, no backport to netbsd-9 planned)
• PR #54761: nvme corruption on GENERIC without DIAGNOSTIC
• PR #54771: nvme(4) fails to attach (not a showstopper)
• PR #54914: fcntl(F_GETPATH) is broken in chroot
• PR #55042: kernel panics when ZFS filesystem is exported with NFS
• PR #55458: entropy: WrAnRdNcING no nstropy in eed
• PR #55510: pg_jobc going negative and crashing the kernel
• PR #55533: mmap(2) fails when called with a non-zero hint
• PR #55892: npf cannot handle large tables
• PR #55941: NetBSD 9+ /boot does not work
• PR #56330: amd64 gdb issues

Test Regressions since netbsd-9

Some of this issues are older (so "regression" is slightly wrong), but it was necessary to disable tests to not break whole test setups. This are likely effects caused by bugs in the tests or RUMP, so not considered show stoppers.

• PR #49141: lib/librumpclient/t_exec/threxec test randomly fails
• PR #50350: rump/rumpkern/t_sp/stress_{long,short} fail
• PR #55304: lib/librumpclient/t_exec/threxec, rump/rumpkern/t_sp/reconnect now fail
• PR #55356: stress_killer test case now fails on real hardware
• PR #55338: rump threxec test broken
• PR #55632: tests/net/ipsec/t_ipsec_misc: timeout tests fail randomly

Ongoing projects and unmerged branches

• Wifi renewal on hg - will not make it into mainline before the branch, but is planned to be merged into -current shortly after branching for netbsd-10
• Updating drmkms to Linux 5.6 - has been merged to HEAD
• Removing PF - will not happen before the branch
  • pf was deprecated in 9 so it could be removed in 10.
  • what is left to do to improve npf?
    • ftp-proxy(8) does not support npf but does support ipf and pf
    • altq(4)
    • shared rulesets ("ipf groups")?
    • address subset selection?
    • update NetBSD Guide from ipf and document migrations
    • note that most of the above are ipf things, not pf things.
    • ...
• Waiting for Randot
  • we need to decide on how randomness would work before branch.
  • Riastradh proposed getrandom(p,n,0) should not block. this needs to be done still.
  • getentropy is missing from NetBSD but will be in POSIX.
  • suggested installer changes
  • rndctl(8) includes misleading features that no longer work with the current entropy behavior (disable/enable estimation, etc). it should maybe also print the number of collected samples in -l.

Current status and timeline

• final pre-branch wrapup, no big changes w/o prior releng/core involvement please!
• top 6 architectures below 10 failures, two at zero - we are getting there!
• branch planed May 11 or 12

Last Test Results overview

For all tests, see releng's tests page.

Test	Last checked	Failures	Comments
i386, qemu, official b5 runs	2022-05-10	0
evbarm64el-aarch64, real hardware	2022-05-10	0
XEN, i386 and amd64	2022-05-07	7
macppc, real hardware	2022-05-09	7	PR 55272, PR 55338
sparc64, real hardware	2022-05-09	7
amd64, qemu, official b5 runs	2022-05-10	9
sparc64, qemu, low memory	2022-05-09	17
evbearmv7hfeb, real hardware	2022-05-09	18
evbarmv5, real hardware	2022-05-09	56	userland watchdog starvation PR 55272
landisk, SH4, real hardware, low memory	2022-05-09	57