Re: [patch] voluntary-preempt-2.6.8-rc2-L2 PS2 keyboard gone south

Re: [patch] voluntary-preempt-2.6.8-rc2-L2 PS2 keyboard gone south

[Shane Shrybman]

Twice while using -L2 my IBM PS2 keyboard has become completely
non-responsive. USB mouse and everything else seems to be fine, but no
LEDs or anything from the keyboard.

On both occasions the last key I hit on the keyboard was numlock and the
numlock did not come on and I had to reboot after that.

UP, x86, gcc 2.95, scsi + ide, bttv

Resolved in M5?

Regards,

Shane

Re: [patch] voluntary-preempt-2.6.8-rc2-L2 PS2 keyboard gone south

[Ingo Molnar]

* Shane Shrybman <shrybman@aei.ca> wrote:

> Twice while using -L2 my IBM PS2 keyboard has become completely
> non-responsive. USB mouse and everything else seems to be fine, but no
> LEDs or anything from the keyboard.
> 
> On both occasions the last key I hit on the keyboard was numlock and the
> numlock did not come on and I had to reboot after that.
> 
> UP, x86, gcc 2.95, scsi + ide, bttv
> 
> Resolved in M5?

M5 does that differently, yes - so could you try it? If you still get
problems, does this fix it:

	echo 0 > /proc/irq/1/*/threaded

(this turns the IRQ 1 thread off.)

	Ingo

Re: [patch] voluntary-preempt-2.6.8-rc2-L2 PS2 keyboard gone south

[Shane Shrybman]

Hi,

On Fri, 2004-07-30 at 11:20, Ingo Molnar wrote:
> * Shane Shrybman <shrybman@aei.ca> wrote:
> 
> > Twice while using -L2 my IBM PS2 keyboard has become completely
> > non-responsive. USB mouse and everything else seems to be fine, but no
> > LEDs or anything from the keyboard.
> > 
> > On both occasions the last key I hit on the keyboard was numlock and the
> > numlock did not come on and I had to reboot after that.
> > 
> > UP, x86, gcc 2.95, scsi + ide, bttv
> > 
> > Resolved in M5?
> 
> M5 does that differently, yes - so could you try it? If you still get
> problems, does this fix it:
> 

Ok, M5 locked up the whole machine within a few seconds of starting X.

> 	echo 0 > /proc/irq/1/*/threaded
> 

I rebooted and tried the above command, but the keyboard became
unresponsive again soon after starting X.

> (this turns the IRQ 1 thread off.)
> 
> 	Ingo
> 

Shane

Re: [patch] voluntary-preempt-2.6.8-rc2-L2 PS2 keyboard gone south

[Felipe Alfaro Solana]

On Fri, 2004-07-30 at 13:38 -0400, Shane Shrybman wrote:

> > M5 does that differently, yes - so could you try it? If you still get
> > problems, does this fix it:
> 
> Ok, M5 locked up the whole machine within a few seconds of starting X.

Me too, with voluntary-preempt=3... It seems I can trigger this randomly
by heavily moving the mouse around while logging in into my KDE session.

However, with voluntary-preempt=2 I've been unable to lock the machine
yet.

Re: [patch] voluntary-preempt-2.6.8-rc2-L2 PS2 keyboard gone south

[Lee Revell]

On Fri, 2004-07-30 at 19:21, Felipe Alfaro Solana wrote:
> On Fri, 2004-07-30 at 13:38 -0400, Shane Shrybman wrote:
> 
> > > M5 does that differently, yes - so could you try it? If you still get
> > > problems, does this fix it:
> > 
> > Ok, M5 locked up the whole machine within a few seconds of starting X.
> 
> Me too, with voluntary-preempt=3... It seems I can trigger this randomly
> by heavily moving the mouse around while logging in into my KDE session.
> 
> However, with voluntary-preempt=2 I've been unable to lock the machine
> yet.

It looks like this is a mouse problem, I have a PS/2 keyboard and USB
mouse and have not had any problems yet with M5.  I also found that with
L2, I could toggle Caps Lock fast enough to get significantly 'ahead' of
it, this no longer happens with M5.

Lee

Re: [patch] voluntary-preempt-2.6.8-rc2-L2 PS2 keyboard gone south

[Felipe Alfaro Solana]

On Fri, 2004-07-30 at 20:05 -0400, Lee Revell wrote:
> On Fri, 2004-07-30 at 19:21, Felipe Alfaro Solana wrote:
> > On Fri, 2004-07-30 at 13:38 -0400, Shane Shrybman wrote:
> > 
> > > > M5 does that differently, yes - so could you try it? If you still get
> > > > problems, does this fix it:
> > > 
> > > Ok, M5 locked up the whole machine within a few seconds of starting X.
> > 
> > Me too, with voluntary-preempt=3... It seems I can trigger this randomly
> > by heavily moving the mouse around while logging in into my KDE session.
> > 
> > However, with voluntary-preempt=2 I've been unable to lock the machine
> > yet.
> 
> It looks like this is a mouse problem, I have a PS/2 keyboard and USB
> mouse and have not had any problems yet with M5.  I also found that with
> L2, I could toggle Caps Lock fast enough to get significantly 'ahead' of
> it, this no longer happens with M5.

I have a PS/2 keyboard and a USB mouse.

Re: [patch] voluntary-preempt-2.6.8-rc2-L2 PS2 keyboard gone south

[Lee Revell]

On Fri, 2004-07-30 at 21:13, Felipe Alfaro Solana wrote:
> On Fri, 2004-07-30 at 20:05 -0400, Lee Revell wrote:
> > On Fri, 2004-07-30 at 19:21, Felipe Alfaro Solana wrote:
> > > On Fri, 2004-07-30 at 13:38 -0400, Shane Shrybman wrote:
> > > 
> > > > > M5 does that differently, yes - so could you try it? If you still get
> > > > > problems, does this fix it:
> > > > 
> > > > Ok, M5 locked up the whole machine within a few seconds of starting X.
> > > 
> > > Me too, with voluntary-preempt=3... It seems I can trigger this randomly
> > > by heavily moving the mouse around while logging in into my KDE session.
> > > 
> > > However, with voluntary-preempt=2 I've been unable to lock the machine
> > > yet.
> > 
> > It looks like this is a mouse problem, I have a PS/2 keyboard and USB
> > mouse and have not had any problems yet with M5.  I also found that with
> > L2, I could toggle Caps Lock fast enough to get significantly 'ahead' of
> > it, this no longer happens with M5.
> 
> I have a PS/2 keyboard and a USB mouse.

Weird.  This is my setup also, and I have had no problems since
installing M5.

Try:

        Option          "NoAccel"

in section "Device" of your XF86Config.  Also try commenting out:

	Load		"dri"

in the "Module" section.

This will ensure that the X server is not accessing hardware directly. 
Normally this should not be a problem but a buggy video driver can cause
problems.  On my machine, having 2D acceleration enabled caused
interrupts from other devices to be lost when dragging a window.

Also, is the machine pingable after it locks up?

Lee

Re: [patch] voluntary-preempt-2.6.8-rc2-L2 PS2 keyboard gone south

[Felipe Alfaro Solana]

On Fri, 2004-07-30 at 23:57 -0400, Lee Revell wrote:
> On Fri, 2004-07-30 at 21:13, Felipe Alfaro Solana wrote:
> > On Fri, 2004-07-30 at 20:05 -0400, Lee Revell wrote:
> > > On Fri, 2004-07-30 at 19:21, Felipe Alfaro Solana wrote:
> > > > On Fri, 2004-07-30 at 13:38 -0400, Shane Shrybman wrote:
> > > > 
> > > > > > M5 does that differently, yes - so could you try it? If you still get
> > > > > > problems, does this fix it:
> > > > > 
> > > > > Ok, M5 locked up the whole machine within a few seconds of starting X.
> > > > 
> > > > Me too, with voluntary-preempt=3... It seems I can trigger this randomly
> > > > by heavily moving the mouse around while logging in into my KDE session.
> > > > 
> > > > However, with voluntary-preempt=2 I've been unable to lock the machine
> > > > yet.
> > > 
> > > It looks like this is a mouse problem, I have a PS/2 keyboard and USB
> > > mouse and have not had any problems yet with M5.  I also found that with
> > > L2, I could toggle Caps Lock fast enough to get significantly 'ahead' of
> > > it, this no longer happens with M5.
> > 
> > I have a PS/2 keyboard and a USB mouse.
> 
> Weird.  This is my setup also, and I have had no problems since
> installing M5.
> 
> Try:
> 
>         Option          "NoAccel"
> 
> in section "Device" of your XF86Config.  Also try commenting out:
> 
> 	Load		"dri"
> 
> in the "Module" section.
> 
> This will ensure that the X server is not accessing hardware directly. 
> Normally this should not be a problem but a buggy video driver can cause
> problems.  On my machine, having 2D acceleration enabled caused
> interrupts from other devices to be lost when dragging a window.
> 
> Also, is the machine pingable after it locks up?

Curiously, the machine is pingable, but I can't ssh/telnet into it: no
data is received from the locked machine.

Re: [patch] voluntary-preempt-2.6.8-rc2-L2 PS2 keyboard gone south

[Lee Revell]

On Sat, 2004-07-31 at 20:24, Felipe Alfaro Solana wrote:
> On Fri, 2004-07-30 at 23:57 -0400, Lee Revell wrote:
> > On Fri, 2004-07-30 at 21:13, Felipe Alfaro Solana wrote:
> > > On Fri, 2004-07-30 at 20:05 -0400, Lee Revell wrote:
> > > > On Fri, 2004-07-30 at 19:21, Felipe Alfaro Solana wrote:
> > > > > On Fri, 2004-07-30 at 13:38 -0400, Shane Shrybman wrote:
> > > > > 
> > > > > > > M5 does that differently, yes - so could you try it? If you still get
> > > > > > > problems, does this fix it:
> > > > > > 
> > > > > > Ok, M5 locked up the whole machine within a few seconds of starting X.
> > > > > 
> > > > > Me too, with voluntary-preempt=3... It seems I can trigger this randomly
> > > > > by heavily moving the mouse around while logging in into my KDE session.
> > > > > 
> > > > > However, with voluntary-preempt=2 I've been unable to lock the machine
> > > > > yet.
> > > > 
> > > > It looks like this is a mouse problem, I have a PS/2 keyboard and USB
> > > > mouse and have not had any problems yet with M5.  I also found that with
> > > > L2, I could toggle Caps Lock fast enough to get significantly 'ahead' of
> > > > it, this no longer happens with M5.
> > > 
> > > I have a PS/2 keyboard and a USB mouse.
> > 
> > Weird.  This is my setup also, and I have had no problems since
> > installing M5.
> > 
> > Try:
> > 
> >         Option          "NoAccel"
> > 
> > in section "Device" of your XF86Config.  Also try commenting out:
> > 
> > 	Load		"dri"
> > 
> > in the "Module" section.
> > 
> > This will ensure that the X server is not accessing hardware directly. 
> > Normally this should not be a problem but a buggy video driver can cause
> > problems.  On my machine, having 2D acceleration enabled caused
> > interrupts from other devices to be lost when dragging a window.
> > 
> > Also, is the machine pingable after it locks up?
> 
> Curiously, the machine is pingable, but I can't ssh/telnet into it: no
> data is received from the locked machine.
> 

Hmm.  Maybe the problem is filesystem or disk related.  Is your root
partition on IDE or SCSI?  It could be a SCSI problem.

Are you running software or hardware RAID?

This is the same behavior you get when, for example, a disk dies.

Lee

Re: [patch] voluntary-preempt-2.6.8-rc2-L2 PS2 keyboard gone south

[Felipe Alfaro Solana]

On Sat, 2004-07-31 at 20:32 -0400, Lee Revell wrote:
> On Sat, 2004-07-31 at 20:24, Felipe Alfaro Solana wrote:
> > On Fri, 2004-07-30 at 23:57 -0400, Lee Revell wrote:
> > > On Fri, 2004-07-30 at 21:13, Felipe Alfaro Solana wrote:
> > > > On Fri, 2004-07-30 at 20:05 -0400, Lee Revell wrote:
> > > > > On Fri, 2004-07-30 at 19:21, Felipe Alfaro Solana wrote:
> > > > > > On Fri, 2004-07-30 at 13:38 -0400, Shane Shrybman wrote:
> > > > > > 
> > > > > > > > M5 does that differently, yes - so could you try it? If you still get
> > > > > > > > problems, does this fix it:
> > > > > > > 
> > > > > > > Ok, M5 locked up the whole machine within a few seconds of starting X.
> > > > > > 
> > > > > > Me too, with voluntary-preempt=3... It seems I can trigger this randomly
> > > > > > by heavily moving the mouse around while logging in into my KDE session.
> > > > > > 
> > > > > > However, with voluntary-preempt=2 I've been unable to lock the machine
> > > > > > yet.
> > > > > 
> > > > > It looks like this is a mouse problem, I have a PS/2 keyboard and USB
> > > > > mouse and have not had any problems yet with M5.  I also found that with
> > > > > L2, I could toggle Caps Lock fast enough to get significantly 'ahead' of
> > > > > it, this no longer happens with M5.
> > > > 
> > > > I have a PS/2 keyboard and a USB mouse.
> > > 
> > > Weird.  This is my setup also, and I have had no problems since
> > > installing M5.
> > > 
> > > Try:
> > > 
> > >         Option          "NoAccel"
> > > 
> > > in section "Device" of your XF86Config.  Also try commenting out:
> > > 
> > > 	Load		"dri"
> > > 
> > > in the "Module" section.
> > > 
> > > This will ensure that the X server is not accessing hardware directly. 
> > > Normally this should not be a problem but a buggy video driver can cause
> > > problems.  On my machine, having 2D acceleration enabled caused
> > > interrupts from other devices to be lost when dragging a window.
> > > 
> > > Also, is the machine pingable after it locks up?
> > 
> > Curiously, the machine is pingable, but I can't ssh/telnet into it: no
> > data is received from the locked machine.
> 
> Hmm.  Maybe the problem is filesystem or disk related.  Is your root
> partition on IDE or SCSI?  It could be a SCSI problem.

No SCSI. Only 2 IDE drives, 1 ATAPI DVD and 1 ATAPI CD-RW.

> Are you running software or hardware RAID?

No.

> This is the same behavior you get when, for example, a disk dies.

I have 2 identical UDMA 120GB Seagate disks which are 1 month old, so I
guess it's not a problem with bad hardware.

It turns out that my lock ups where related to APIC. Disabling APIC
seems to get rid of those lock ups.

Re: [patch] voluntary-preempt-2.6.8-rc2-L2 PS2 keyboard gone south

[Ingo Molnar]

On Sun, 1 Aug 2004, Felipe Alfaro Solana wrote:

> It turns out that my lock ups where related to APIC. Disabling APIC
> seems to get rid of those lock ups.

ah ... that makes sense. It is most likely level-triggered IO-APIC
interrupts that break. Could you try to turn off redirection for every
interrupt that says 'IO-APIC-level' in /proc/interrupts - does that
stabilize things? (besides the 'noapic' workaround you already
discovered.) How does your /proc/interrupts look like btw?

the 'XT-PIC' interrupt controller (i8259A) is fine for redirection. (and
that's what i used for testing almost exclusively.)

	Ingo

Re: [patch] voluntary-preempt-2.6.8-rc2-L2 PS2 keyboard gone south

[Shane Shrybman]

On Fri, 2004-07-30 at 20:05, Lee Revell wrote: 
> On Fri, 2004-07-30 at 19:21, Felipe Alfaro Solana wrote:
> > On Fri, 2004-07-30 at 13:38 -0400, Shane Shrybman wrote:
> > 
> > > > M5 does that differently, yes - so could you try it? If you still get
> > > > problems, does this fix it:
> > > 
> > > Ok, M5 locked up the whole machine within a few seconds of starting X.
> > 
> > Me too, with voluntary-preempt=3... It seems I can trigger this randomly
> > by heavily moving the mouse around while logging in into my KDE session.
> > 
> > However, with voluntary-preempt=2 I've been unable to lock the machine
> > yet.
> 
> It looks like this is a mouse problem, I have a PS/2 keyboard and USB
> mouse and have not had any problems yet with M5. 

I have a usb mouse and PS/2 keyboard as well. The mouse kept functioning
and the keyboard didn't, except when the whole machine hung. I am
recompiling M5 from fresh sources to confirm the behaviour there.

Hmm, M5 still no go here. Keyboard and display updates stop working
quickly once in X, although the mouse is able to keep moving. This was
with voluntary-preempt=2 this time.

Shane

Re: [patch] voluntary-preempt-2.6.8-rc2-L2 PS2 keyboard gone south

[Lee Revell]

On Sat, 2004-07-31 at 20:36, Felipe Alfaro Solana wrote:
> On Sat, 2004-07-31 at 13:56 -0400, Lee Revell wrote:
> > On Sat, 2004-07-31 at 05:48, Felipe Alfaro Solana wrote:
> > > On Sat, 2004-07-31 at 00:03 -0400, Lee Revell wrote:
> > > > Shane & Felipe,
> > > > 
> > > > I am removing LKML and Ingo from the cc: list so we can compare
> > > > hardware, as I definitely am not seeing this problem.  Maybe we can
> > > > track it down to a difference there.  What hardware are you guys using?
> > 
> > > I'm running 2.6.8-rc2-bk9 plus a bunch of other patches, like Con's
> > > Staircase Scheduler and Ingo's voluntary-preempt (of course). Kernel is
> > > compiled with ACPI + APIC.
> > 
> > Hmm.  Maybe it's an ACPI/APIC problem?  I do not have these enabled.  It
> > could also be an interaction between some other patch you have.
> > 
> > Have you tried with 2.6.8-rc2-M5, without the other patches?  Booting
> > with acpi=off?
> 
> OK, OK :-) I've been running 2.6.8-rc2-mm1-M5 with ACPI but without APIC
> for more than ten minutes), compiling kernels, sending mails with
> Evolution and it hasn't locked up yet. Crossfingers. Should we report
> this to Ingo and Andrew?
> 
> Anyways, I'll keep on running this puppy to see if this behavior is
> consistent.
> 
> # grep . /proc/irq/*/*/threaded
> /proc/irq/11/eth0/threaded:1
> /proc/irq/12/Intel 82801BA-ICH2/threaded:1
> /proc/irq/14/ide0/threaded:1
> /proc/irq/15/ide1/threaded:1
> /proc/irq/1/i8042/threaded:1
> /proc/irq/5/uhci_hcd/threaded:1
> /proc/irq/8/rtc/threaded:1
> /proc/irq/9/acpi/threaded:1
> /proc/irq/9/uhci_hcd/threaded:!
> 
> # grep . /proc/sys/kernel/*_preemption
> /proc/sys/kernel/voluntary_preemption:3
> /proc/sys/kernel/preemption:1

The next thing I was going to suggest was the software RAID.  You appear
to have a RAID 0 or 1 with one disk on irq14 and one on irq15.  I am not
sure how interrupts are handled by Linux IDE RAID, but it seems like
this would be tricky.  I bet the threading is screwing up the
synchronization between the devices, and you end up with one waiting
forever for the other - the possibilities for lockup are endless.

However, I have no idea what you would do about this.  You could try
making those irqs non-threaded, but you'd have to do it one at a time,
and I would *definitely* expect weird stuff with one threaded and one
not.

I will have to defer to someone with more kernel expertise.

Lee

Re: [patch] voluntary-preempt-2.6.8-rc2-L2 PS2 keyboard gone south

[Felipe Alfaro Solana]

On Sat, 2004-07-31 at 20:46 -0400, Lee Revell wrote:

> > OK, OK :-) I've been running 2.6.8-rc2-mm1-M5 with ACPI but without APIC
> > for more than ten minutes), compiling kernels, sending mails with
> > Evolution and it hasn't locked up yet. Crossfingers. Should we report
> > this to Ingo and Andrew?
> > 
> > Anyways, I'll keep on running this puppy to see if this behavior is
> > consistent.
> > 
> > # grep . /proc/irq/*/*/threaded
> > /proc/irq/11/eth0/threaded:1
> > /proc/irq/12/Intel 82801BA-ICH2/threaded:1
> > /proc/irq/14/ide0/threaded:1
> > /proc/irq/15/ide1/threaded:1
> > /proc/irq/1/i8042/threaded:1
> > /proc/irq/5/uhci_hcd/threaded:1
> > /proc/irq/8/rtc/threaded:1
> > /proc/irq/9/acpi/threaded:1
> > /proc/irq/9/uhci_hcd/threaded:!
> > 
> > # grep . /proc/sys/kernel/*_preemption
> > /proc/sys/kernel/voluntary_preemption:3
> > /proc/sys/kernel/preemption:1
> 
> The next thing I was going to suggest was the software RAID.  You appear
> to have a RAID 0 or 1 with one disk on irq14 and one on irq15.  I am not
> sure how interrupts are handled by Linux IDE RAID, but it seems like
> this would be tricky.  I bet the threading is screwing up the
> synchronization between the devices, and you end up with one waiting
> forever for the other - the possibilities for lockup are endless.

No, I'm not using any kind of RAID (eiher HW or SW) at all. Simply, two
identical, independent 120GB drives that only have one thing in common:
they are attached to the same IDE channel (the primary IDE channel). The
CD/RW and DVD are attached to the secondary IDE channel.

Re: [patch] voluntary-preempt-2.6.8-rc2-L2 PS2 keyboard gone south

[Lee Revell]

On Fri, 2004-07-30 at 10:06, Shane Shrybman wrote:
> Twice while using -L2 my IBM PS2 keyboard has become completely
> non-responsive. USB mouse and everything else seems to be fine, but no
> LEDs or anything from the keyboard.
> 
> On both occasions the last key I hit on the keyboard was numlock and the
> numlock did not come on and I had to reboot after that.
> 
> UP, x86, gcc 2.95, scsi + ide, bttv
> 

This happened to me, also twice necessitating a reboot.  I am pretty
sure I did *not* hit Num Lock last, though the system was under load -
multiple builds going on, jackd running, and video playback.  I tried to
toggle Num Lock to see if the machine was really locked hard, and it
worked for a while (though it did not go on/off exactly once for each
time I hit it), then stopped responding.

This happened with the timer + soundcard IRQs direct, and with timer +
soundcard + RTC direct.

Lee

Re: [patch] voluntary-preempt-2.6.8-rc2-L2 PS2 keyboard gone south

[Shane Shrybman]

On Fri, 2004-07-30 at 14:03, Lee Revell wrote:
> On Fri, 2004-07-30 at 10:06, Shane Shrybman wrote:
> > Twice while using -L2 my IBM PS2 keyboard has become completely
> > non-responsive. USB mouse and everything else seems to be fine, but no
> > LEDs or anything from the keyboard.
> > 
> > On both occasions the last key I hit on the keyboard was numlock and the
> > numlock did not come on and I had to reboot after that.
> > 
> > UP, x86, gcc 2.95, scsi + ide, bttv
> > 
> 
> This happened to me, also twice necessitating a reboot.  I am pretty
> sure I did *not* hit Num Lock last, though the system was under load -
> multiple builds going on, jackd running, and video playback.  I tried to
> toggle Num Lock to see if the machine was really locked hard, and it
> worked for a while (though it did not go on/off exactly once for each
> time I hit it), then stopped responding.
> 

Did this happen with both the rc2-L2 and rc2-M5 kernels for you too?

Ingo, I have been running a previous version ( -I1 I think ) of
voluntary-preempt patch successfully for a few days, so I think the
problem was introduced since the -I1 version.

Shane

Re: [patch] voluntary-preempt-2.6.8-rc2-L2 PS2 keyboard gone south

[Lee Revell]

On Fri, 2004-07-30 at 17:31, Shane Shrybman wrote:
> On Fri, 2004-07-30 at 14:03, Lee Revell wrote:
> > On Fri, 2004-07-30 at 10:06, Shane Shrybman wrote:
> > > Twice while using -L2 my IBM PS2 keyboard has become completely
> > > non-responsive. USB mouse and everything else seems to be fine, but no
> > > LEDs or anything from the keyboard.
> > > 
> > > On both occasions the last key I hit on the keyboard was numlock and the
> > > numlock did not come on and I had to reboot after that.
> > > 
> > > UP, x86, gcc 2.95, scsi + ide, bttv
> > > 
> > 
> > This happened to me, also twice necessitating a reboot.  I am pretty
> > sure I did *not* hit Num Lock last, though the system was under load -
> > multiple builds going on, jackd running, and video playback.  I tried to
> > toggle Num Lock to see if the machine was really locked hard, and it
> > worked for a while (though it did not go on/off exactly once for each
> > time I hit it), then stopped responding.
> > 
> 
> Did this happen with both the rc2-L2 and rc2-M5 kernels for you too?
> 

Have not tried M5 yet (still building).  An improved fix was posted for
the PS/2 problem recently which uses a semaphore and work queues rather
than Ingo's original quick fix of just adding a lock-break.  If the
problem is indeed keyboard-related, this may help.

I also have not determined whether the system freezes completely or
whether it is just the input layer that hangs.  Next time this happens I
will try to ssh in.

Lee