[wplug] My computer is going slightly insane

[Bryan J. Smith]

On Sat, 2007-08-18 at 02:29 -0400, Mackenzie Morgan wrote:
> I don't know when it started as I've been away at school until May,
> but probably 5 times this summer I've walked over to my computer and
> the monitor was in suspend (ok sure...normal if it's been a while),
> but moving the mouse doesn't bring anything back.  The "any" key
> doesn't work either.  Dropping to a virtual terminal wakes the monitor
> back up, but all I see is...well, a mess.

Do you get a mess normally when you drop to a virtual terminal?

I.e., I've had issues with Intel GPU framebuffers separating the virtual
terminal framebuffer from X11.  Sometimes booting with a vga= line (or
without if you are already) into framebuffer for the virtual consoles
solves the problem.

Especially on my older notebooks with the i855GM, although I still have
some issues with those in general, even on Fedora 7 or the latest
Xubuntu.

> It looks like Thursday night when Ted demo'd what happens when you
> cat /dev/urandom and let it go so long the terminal corrupts.  When he
> did that though, he fixed it with the reset command.  That doesn't
> help in this case.  It redraws the mess.

Again, it's a GPU framebuffer issue, not a terminfo (which is Ted's
example).

It's possibly an Intel suspend/timing support issue.  Understand that
Intel, despite the praises its receives, does _not_ share a great amount
of its memory management (part of the reason why the Linux drivers
cannot complete with the Windows versions on 3D) and feature support in
its drivers (especially on notebooks and external output).  I've run
into this on every i8xx/9xx series product I've ever used.

> The terminal can still, technically, be used, provided you don't make
> a typo.  Usually I'd hit the power button and force shut down then
> turn it back on, but this time I tried actually logging in and sending
> "sudo reboot" which worked so yay for not potentially hurting things
> with forced shutdowns.  
> Distro:  Ubuntu 7.04 (Feisty)
> Kernel: 2.6.20
> I've been bugging some IRC channels, and now I'm bugging you, but
> here's things I've been asked or had suggested as the cause:
> Q1: ATI graphics with fglrx driver? 
> A1: No, it's Intel graphics, and old ones at that (so no, the new
> X3000/i965 thing that's a bit tricky isn't in here)
> Q2: Funky ACPI interrupts from a hardware suspend button?
> A2: I don't have a hardware suspend button, so I doubt it

Just because you don't have a hardware suspend button doesn't mean the
system isn't suspending.

ACPI is almost a non-implemented started, largely because Microsoft
tries to get PC OEMs to change the "non-standard" every Windows release.
This is not so much against Linux, but to prevent old Windows releases
from working on newer hardware and vice-versa.  E.g., I have original
Windows XP releases and latter version 2002 releases that work on some
but not others, typically because of age and due to differing ACPI
implementation.

The Linux kernel gives you the option to disable ACPI, and that can
often fix many things (although it can break other support).

> Q3: Power management settings for suspend or hibernate with poorly
> supported hardware?
> A3: It's all very generic hardware, and suspend/hibernate are disabled

In the BIOS?  Good.

But how about your KDE or GNOME settings?  ;)

> Q4: Anything weird in kern.log or /var/log/messages? 
> A4: Not that I see.  kern.log has absolutely no events registered
> between around 10 AM and when I rebooted it at 6:13 PM, and the last
> user logged out at 5:05 PM, and /var/log/messages just has 3 lines of
> - - MARK - - between those times. 
> The most reasonable suggestion I've had so far was failing hardware.
> I think my onboard graphics could be going (but very slowly) since for
> the last 2 or 3 years weird things would sometimes happen.  On Windows
> that meant I'd suddenly have 640x480 for a resolution and it'd be in
> 8-bit color.

That's just Windows self-screwing up.  ;)

> On Ubuntu, that means it acts like someone's playing with the width
> button on the bottom of the monitor because it sorta shrinks in about
> 20px on each side and then back to normal, and squish, normal, etc
> repeatedly for a couple seconds.

That could be Windows and X11 "fighting" over programmed modes in the
monitor.  I.e., Windows and X11 are using different scan ranges, and
each time you boot between then, one "resets" the monitor mode for the
other.

[ Did I mention I love DVI to avoid that non-sense? ;-]

> Could failing hardware cause this, and would the graphics card be
> critical enough to do it, or should I start worrying about my HDD or
> some other important part?  If it's not likely to be hardware, what
> else could cause it? 

Power.  Power.  Power.  Just on Friday, I had a situation that I didn't
think was power-related, but it was.

I had a PCIe x8 prototype storage controller in a server chassis, which
has its own firmware and console interfaces.  The chassis would boot,
and the PCIe x8 prototype board would boot its IOP384 and get to a
RedBoot prompt, but just hang after a few seconds (retrospectively, it
was probably as it powered up its second core, select peripherals,
etc...).

A fellow engineer finally just made the suggestion that it could be a
power issue.  I figured it could be, because a typical PCIe slot isn't
rated for more than 25-50W (PCIe x16 graphics slots are rated 125-150W),
but he suggested I just start pulling loads on the system itself.  As
much as I didn't think it would be that (I believed it may more likely
be adequate power in the slot itself), I did -- fans, disks and, at one
point, even the memory in the main system.

Tada!  The RedBoot prompt on the storage controller stopped freezing.

Root cause:  the efficiency of the host server SSI EEB EPS12V power
supply was just inadequate for delivering [probably] the +3.3V liner.
Unplugging any +3.3V loads (including memory) took care of it.


-- 
Bryan J. Smith         Professional, Technical Annoyance
mailto:b.j.smith at ieee.org   http://thebs413.blogspot.com
--------------------------------------------------------
        Fission Power:  An Inconvenient Solution