https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=287872

            Bug ID: 287872
           Summary: boot time regression related to netlink and devd
           Product: Base System
           Version: 14.3-STABLE
          Hardware: Any
                OS: Any
            Status: New
          Severity: Affects Some People
          Priority: ---
         Component: kern
          Assignee: n...@freebsd.org
          Reporter: eu...@freebsd.org
                CC: melif...@freebsd.org

I have an old HP ProLiant DL120 G6 server that ran FreeBSD 13.5-STABLE/amd64
just fine until today I updated it to 14.3-STABLE/amd64 using sources of
stable/14 as of June, 11. The upgrade process went just fine but after reboot
start of devd took over 17.6 minutes, then system proceeded to boot normally
and now runs just fine. This is UFS2-only system.

CPU: Intel(R) Xeon(R) CPU X3470 @ 2.93GHz (2926.10-MHz K8-class CPU)

Mem: 1716M Active, 2244M Inact, 690M Wired, 326M Buf, 11G Free
Swap: 14G Total, 14G Free

I have a custom rc.d script that starts before devd and creates over 800 vlans
with integrated bge0 interface as a parent. 

For a test, in multiuser mode I destroy all vlans, stop devd, rerun my script
to recreate all vlans and use "service devd start", and the problem reproduces
and mentioned command takes same long time again.

During that time, "top -SHPz" shows:

  PID USERNAME    PRI NICE   SIZE    RES STATE    C   TIME    WCPU COMMAND
   12 root        -60    -     0B   336K CPU1     1  22:53  97,01% intr{irq16:
ehci0 bge0}
    0 root          8    -     0B   720K CPU3     3   0:01   5,27%
kernel{netlink_socket (PID}
    2 root        -60    -     0B    64K WAIT     2   0:12   0,77% clock{clock
(0)}
15765 root         68    0    21M  7808K wait     1   0:00   0,11% devd

Note, there are less than 200 kilobytes/s of traffic over bge0 meantime. After
"service devd start" finishes, the load of "intr{irq16: ehci0 bge0}" drops to
about zero.

More details follow.

-- 
You are receiving this mail because:
You are the assignee for the bug.

Reply via email to