Hi, all,
It's been a really long time since I've had much to say hereabouts,
but as I'm in the middle of an upgrade cycle (12.4 to 13.2) I wanted
to post about an issue I ran into. On both of my workstations, my
custom kernel would hang at boot. I didn't see this on either of the
servers that I had already upgraded. As I was bored at home today, I
tried booting a GENERIC kernel, built from the same source tree
(13.2-RELEASE-p1) as my custom kernel, and it booted just fine.
I don't have the ability to do serial console on either of my
workstations, nor any sort of network debugging, but when I did a
verbose boot on the office workstation, it didn't show anything
interesting. However, at home, I noticed that the hang occurred
immediately after attach of:
hwpstate_intel0: <Intel Speed Shift> on cpu0
hwpstate_intel1: <Intel Speed Shift> on cpu1
The first time I pressed a key on this machine's PS/2 keyboard, it got
one step further:
hwpstate_intel2: <Intel Speed Shift> on cpu2
This is a 6-core, 12-thread system, and the working kernel gets all
the way to
hwpstate_intel11: <Intel Speed Shift> on cpu11
nearly instantly.
I took the working GENERIC configuration and pared it down to make a
new custom kernel, and it worked (I'm using it right now). So I
compared the working and broken configurations, and noticed the
following options were present in the working configuration and not in
the broken one:
options EARLY_AP_STARTUP
options GZIO
options IICHID_SAMPLING
options KDB
options KDB_TRACE
options NUMA
options SCSI_DELAY=5000
options SC_PIXEL_MODE
options VESA
options ZSTDIO
The first one, EARLY_AP_STARTUP, stood out to me as likely related to
the problem -- most of the other options involve hardware or features
that this machine doesn't use, but I could easily imagine that
configuring power state controls on CPUs that haven't been started yet
might fail. This option isn't mentioned anywhere in UPDATING, and the
comment in GENERIC isn't espcially helpful, but I have a suspicion
that this option is now effectively mandatory, at least if `cpufreq`
is compiled into the kernel (as it is on all of my kernels and in
GENERIC as well). To be 100% certain I should build the old config
with just that option enabled, and maybe I'll try that on my work
desktop since I still need to finish the upgrade there.
This option was apparently added in 2016 by jhb@, and in his
PHabricator description, he wrote:
As a transition aid, the new behavior is moved under a new
kernel option (EARLY_AP_STARTUP). This will allow the option
to be turned off if need be during initial testing. I hope to
enable this on x86 by default in a followup commit and to have
all platforms moved over before 11.0. Once the transition is
complete, the option will be removed along with the
!EARLY_AP_STARTUP code.
Apparently we got all the way to 13.2 and this never happened. It
should probably get at least a mention in UPDATING for anyone else who
hasn't tripped over this.
-GAWollman
