Checking 'hlt' instruction - failed (MORE)

[GianPiero Puccioni]

Hi,

thanks for the reply

> I believe the kernel uses the "hlt" instruction to reduce power
> consumption and system bus competition in its idle loop.  For 
> some reason, in interrupt doesn't seem to wake up some CPUs properly, 
> so this option was added.  Instead of simply hlt'ing with interrupts 
> enabled, the kernel just spins in a tight loop waiting for an interrupt
> to happen.
> 
> Try this:
> 
> 1) Change the line "kernel.sysrq=0" to "kernel.sysrq=1" in the file
>    "/etc/sysctl.conf". 
> 
> 2) As root:
> 
> 	# echo 1 > /proc/sys/kernel/sysrq
> 
OK, did that. As I said it takes some time for the system to hang so
when it happen I'll report.

[...]

> A) A system deadlock.  Try Alt-SysRq-s and then Alt-Sysrq-b to safely
>    reboot your system.  Try disabling some of your modules.
> 
There isn't much running, in fact all the machines are just a beowulf
cluster with minimal stuff on them. The modules running are

Module                  Size  Used by    Not tainted
autofs                 13780   0  (autoclean) (unused)
nfs                    89912   0  (autoclean)
lockd                  60656   0  (autoclean) [nfs]
sunrpc                 90876   0  (autoclean) [nfs lockd]
e100                   58468   1
floppy                 58908   0  (autoclean)
sg                     37580   0  (autoclean) (unused)
scsi_mod              111528   1  (autoclean) [sg]
microcode               5024   0  (autoclean)
ext3                   74148   2
jbd                    56624   2  [ext3]

I suppose I could rmmod floppy sg and scsi_mod, I have no idea why sg
and scsi are running as the machine has only an IDE HD (no CD or burner
or anything SCSIish on it).

> B) Broken hardware.
> 
> C) A buggy BIOS that is incorrectly handling interrupt assignments or
>    power management (try adding "apm=off" and "acpi=off" to the
>    kernel boot arguments).
> 
hmm, If I try this and then boot without no-hlt, should it start if this
is the problem? I have another machine to play with waiting for the
first to hang.

> D) Some other problem ;-)
> 
Hope not!

> Hope this gives you a direction to proceed.
> 

Oh yea, thanks!

GianPiero