Re: Interrupt storm

On Wed, 30 Mar 2005 12:54:18 -0500 (EST), Andre Guibert de Bruet
<andy_at_siliconlandmark.com> wrote:
> 
> On Wed, 30 Mar 2005, Dan Cojocar wrote:
> > On Wed, 30 Mar 2005 10:53:48 -0500 (EST), Andre Guibert de Bruet wrote:
> >> On Wed, 30 Mar 2005, Dan Cojocar wrote:
> >>>
> >>> I tried running GENERIC and i get the following lor and watchdog
> >>> timeout, no interrupt storms:
> >> --- 8< --- <snip> --- 8< ---
> >>>   Here is the full dmesg: http://cs.ubbcluj.ro/~dan/dmesg.txt.generic
> >>>   I tried also my kernel with apic, i don't get interrupt storms but
> >>> i get watchdog timeout message just like with GENERIC.
> >>> Here is the full dmesg: http://cs.ubbcluj.ro/~dan/dmesg.txt.apic
> >>>   And finally i tried my kernel without apic and without acpi, it's
> >>> running just fine, no interrupt storms, no watchdog timeout, but i
> >>> don't have acpi :(, here is the final
> >>> dmesg: http://cs.ubbcluj.ro/~dan/dmesg.txt.noacpi.
> >>
> >> It looks like your motherboard's DSDT is doing something hokey. I would
> >> complain to the board manufacturer about this. For the timebeing, you can
> >> dump the DSDT into ASL format, modify it, compile it back to DSDT and load
> >> a proper version on system startup. If this is your first time, you
> >> probably want to run "acpidump -d > my.asl" and make the resulting file
> >> available somewhere.
> >>
> >> Would you also mind sharing the kernel config file that was used to build
> >> the kernel that gaves you the mother of all interrupt parties?
> >>
> >    Here is my asl: http://cs.ubbcluj.ro/~dan/my.asl
> >    And here is my kernel config: http://cs.ubbcluj.ro/~dan/FREE
> 
> I am looking at your kernel config file and it looks pretty much like a
> stripped down version of GENERIC. I do have some observations:

Yes it's a stripped version of GENERIC.

> - "device agp" is commented out. Is there any particular reason for this?
> Does the machine not boot with it on? nForce chipsets provide generic agp
> support.

  I use agp as a module.

> - atapicam is currently broken on CURRENT. See sos_at_'s recent ata-mkIII
> headsup announcement for additional details.

My current is from yesterday, sos did his commit today, that is why i
have atapicam.
 
> You might also want to take note of the following:
> 
> - You are specifying a "maxusers" parameter. This is no longer required
> for normal operation. The kernel does a pretty good job of auto-tuning
> itself. I presume there is a reason for having selected such a low number.
> - The default scheduler in CURRENT is 4BSD not ULE (This might cause you
> to encounter edge cases down the road).

    I'm running with ULE for some time and it's very stable, i use ULE
on all my machines and i don't have problems. I will do a test with
4BSD but i don't think
that this is the problem.

> - You commented out "device random". This causes your system to use
> alternate (and arguably less secure) entropy sources for its PRNG.

   random is a module too.

> 
> I haven't really taken a look at the ASL yet. There are people on this
> list that are better qualified than myself that could lend you a helping
> hand with it in far less time than I ever could. I added one such person
> on the CC list... :-)

   Thank you,
            Dan