public inbox for linux-kernel@vger.kernel.org
 help / color / mirror / Atom feed
* Panics need better handling
@ 2004-06-14  2:23 ndiamond
  2004-06-14  4:21 ` Willy Tarreau
  2004-06-14  7:28 ` Panics need better handling Helge Hafting
  0 siblings, 2 replies; 10+ messages in thread
From: ndiamond @ 2004-06-14  2:23 UTC (permalink / raw)
  To: linux-kernel

The following information is visible on
a screen after a panic.  Usually even
this amount of information is not
visible.  It was visible because of an
intuitive guess that a panic was going
to happen within minutes of that guess.
A very lucky occurence resulted in one
of the lines being displayed which is
rarely displayed even when other panic
lines are visible.  I am not asking for
help in solving this particular panic,
I am asking for help in general, in
getting information displayed when it
needs to be displayed.

EIP is at __insmod_binfmt_misc_S.data_L548 [binfmt_misc] 0x20e7 (2.4.20-custom)
eax: e47f1600   ebx: e4017e80   ecx: 00000000   edx: e7883000
esi: 04000001   edi: c034df54   ebp: 0000000b   esp: c034df10
ds: 0068   es: 0068   ss: 0068
Process swapper (pid: 0, stackpage=c034d000)
Stack: c010aab5 0000000b e47f1600 c034df54 c036fe80 0000000b c29c6c00 00000580
       c010ac54 0000000b c034df54 c29c6c00 00000000 00000000 00000000 00000000
       c010d768 00000000 00000000 000005c1 00000000 00000000 00000000 c034dfbc
Call Trace:   [<c010aab5>] handle_IRQ_event [kernel] 0x45 (0xc034df10))
[<c010ac54>] do_IRQ [kernel] 0x84 (0xc034df30))
[<c010d768>] call_do_IRQ [kernel] 0x5 (0xc034df50))
[<c0110068>] dump_fpu [kernel] 0x48 (0xc034df70))
[<c01157ee>] apm_bios_call_simple [kernel] 0x8e (0xc034df7c))
[<c0115937>] apm_do_idle [kernel] 0x27 (0xc034dfa8))
[<c0115a8b>] apm_cpu_idle [kernel] 0xbb (0xc034dfc0))
[<c01159d0>] apm_cpu_idle [kernel] 0x0 (0xc034dfc4))
[<c0107040>] default_idle [kernel] 0x0 (0xc034dfd0))
[<c01070f2>] cpu_idle [kernel] 0x42 (0xc034dfd4))
[<c0105000>] stext [kernel] 0x0 (0xc034dfe0))


Code:  Bad EIP value.
 <0>Kernel panic: Aiee, killing interrupt handler!
In interrupt handler - not syncing
_       <-- (cursor is there)


I'm not sure which of these complaints
to prioritize.

When a panic is visible, there are two
blank lines, a redundant line saying
Kernel panic Aieee, a redundant line
saying In interrupt handler, and a
redundant line with the cursor only.
If these 5 lines were removed, there
could be 5 more lines of trace
information.  Actually there usually
are 5 more lines of trace information,
actually usually more than that, but
they get scrolled off the top of the
screen.  The line saying "EIP is at" is
almost always among the lines that get
scrolled off the screen, but in this
case we were lucky that this line just
barely got a chance to survive.  And
usually the EIP isn't so badly invalid
as this one is, usually it's really
useful information that gets scrolled
off the top of the screen.  Is there
any chance in getting the 24 most
important lines of panic information
displayed last, and putting the cursor
at the end of the 24th line thereof, so
that 24 valuable lines of panic
information can be visible?

Also, usually the panic information
isn't visible at all.  Certain famous
frequent blue screens of death are
famous and frequent partly because of
famous frequent crashes, but also
partly because the kernel puts the
screen back into VGA text mode before
displaying the text.  In Linux, usually
XFree86 is running, and usually the
only visible effect is a freeze.  How
many of these freezes really have panic
information written to an invisible VGA
buffer and never displayed?  When a
panic occurs, we need to shove off
XFree86 and put the display into VGA
text mode.  Even if this makes Linux
black screens of death become as famous
and frequent as certain blue ones, we
need to get the information.

(Apologies if there are editing errors
in the above, which are not visible in
the web interface being used for this
submission.  I used to send submissions
from my real e-mail account, and as a
result was rewarded with 60 spams per
day.  I'm not going to send submissions
from my new real e-mail account.)

^ permalink raw reply	[flat|nested] 10+ messages in thread
* Re: Panics need better handling
@ 2004-06-14  5:19 ndiamond
  2004-06-14  5:45 ` Willy Tarreau
  0 siblings, 1 reply; 10+ messages in thread
From: ndiamond @ 2004-06-14  5:19 UTC (permalink / raw)
  To: linux-kernel

Willy Tarreau replied to me:

>> Is there
>> any chance in getting the 24 most
>> important lines of panic information
>> displayed last, and putting the cursor
>> at the end of the 24th line thereof, so
>> that 24 valuable lines of panic
>> information can be visible?
>
> You could try kmsgdump, which Randy Dunlap ported to 2.6
[...]
> Clearly what you need it seems,

Partly, yes it looks clearly what I
need (though I need it for 2.4).

But surely every developer or maintainer
of every driver or other part of the
kernel also has a clear need for every
Linux user to install this.  I am not
the only one who needs to get these
reports, right?  Shouldn't this be in
the main kernel tree by now, and enabled
by default?


^ permalink raw reply	[flat|nested] 10+ messages in thread
* Re: Panics need better handling
@ 2004-06-14  5:52 ndiamond
  0 siblings, 0 replies; 10+ messages in thread
From: ndiamond @ 2004-06-14  5:52 UTC (permalink / raw)
  To: linux-kernel

Willy Tarreau replied to me:

>> But surely every developer or maintainer
>> of every driver or other part of the
>> kernel also has a clear need for every
>> Linux user to install this. I am not
>> the only one who needs to get these
>> reports, right? Shouldn't this be in
>> the main kernel tree by now, and enabled
>> by default?
>
> Well, yes and no. Yes because it's useful, no because there are so many
> other useful tools which would largely replace it and be more complete
> (kdb, lkcd, ...) that one could wonder why it's in the kernel at all.

Every developer or maintainer of every
driver or other part of the kernel wants
every Linux user to use kdb etc.?
I sure hope not.

> Since it's mainly useful to developers, and not too much intrusive, people
> who need it can easily apply it to their tree.

The information contained in the panic
reports is mainly useful to developers.
By not making the information visible
by default to end users, developers
avoid getting reports from end users.
Is this really what we want?  Do we
want to get reports only from failures
that we personally experience?

Although a particular big company is
famous for that attitude (including
not allowing end users to submit bug
reports unless we pay 4,200 yen to
make the submission), it's interesting
that their X Windows version sometimes
offers to call home with a crash report
after rebooting.  They're no longer
100% dedicated to closing their ears
to failures in the field.  Maybe Linux
isn't ready for this yet, but surely
we should not discourage reports from
end users?

^ permalink raw reply	[flat|nested] 10+ messages in thread
* RE: Panics need better handling
@ 2004-06-14 14:20 Cress, Andrew R
  0 siblings, 0 replies; 10+ messages in thread
From: Cress, Andrew R @ 2004-06-14 14:20 UTC (permalink / raw)
  To: John Bradford, Helge Hafting, ndiamond; +Cc: linux-kernel


For Intel servers, there is some help in 2.4.  It is now included within
the OpenIPMI driver.
It saves the panic info into a firmware log.
See http://sourceforge.net/projects/openipmi/ and http://panicsel.sf.net
(more info, plus a 'showsel' utility to view the firmware log).

The parameter to save this isn't turned on by default in OpenIPMI, but
it is there in 2.4 kernels (CONFIG_IPMI_PANIC_EVENT).  

Andy Cress

-----Original Message-----
From: linux-kernel-owner@vger.kernel.org
[mailto:linux-kernel-owner@vger.kernel.org] On Behalf Of John Bradford
Sent: Monday, June 14, 2004 3:44 AM
To: Helge Hafting; ndiamond@despammed.com
Cc: linux-kernel@vger.kernel.org
Subject: Re: Panics need better handling


Quote from Helge Hafting <helgehaf@aitel.hist.no>:
> ndiamond@despammed.com wrote:
> 
> > I am not asking for
> >help in solving this particular panic,
> >I am asking for help in general, in
> >getting information displayed when it
> >needs to be displayed.
> >  
> >
> I have struggled with this from time to time.  Wanting to
> report a trace, but it is too long for the screen. 
> 
> Using a framebuffer console helps a lot.  I use 1280x1024 resolution,
> and 8x8 characters.  The resulting 160x128 console isn't
> that fun to _work_ with, but most panics/oopses fit.  I rarely
> work at the console anyway.  If you do, consider making two almost
> identical kernels where console font size is the only difference.
(The
> extra compile takes very little time.)  Then use the small-font kernel
> when debugging.

On the other hand, if like me you use a text-based console almost
exclusively,
then the best course of action is probably to buy a real serial
terminal, (or
several :-) ), and configure one of them as the console.  Then you can
basically ignore the VGA display completely.

John.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel"
in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

^ permalink raw reply	[flat|nested] 10+ messages in thread

end of thread, other threads:[~2004-06-14 14:23 UTC | newest]

Thread overview: 10+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2004-06-14  2:23 Panics need better handling ndiamond
2004-06-14  4:21 ` Willy Tarreau
2004-06-14  4:34   ` Norberto Bensa
2004-06-14 13:46     ` [PATCH] kmsgdump-2.6.6.patch (was: Panics need better handling) Kalin KOZHUHAROV
2004-06-14  7:28 ` Panics need better handling Helge Hafting
2004-06-14  7:44   ` John Bradford
  -- strict thread matches above, loose matches on Subject: below --
2004-06-14  5:19 ndiamond
2004-06-14  5:45 ` Willy Tarreau
2004-06-14  5:52 ndiamond
2004-06-14 14:20 Cress, Andrew R

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox