Re: nforce2 lockups

[Mikael Pettersson <mikpe@csd.uu.se>]

On Sat, 23 Aug 2003 14:48:06 +0200, Patrick Dreker <patrick@dreker.de> wrote:
>Am Samstag, 23. August 2003 14:20 schrieb Mikael Pettersson <mikpe@csd.uu.se> 
>zum Thema Re: nforce2 lockups:
>> On Sat, 23 Aug 2003 10:41:46 +0900, kenton.groombridge@us.army.mil wrote:
>> Passing nolapic to a kernel which doesn't recognise it causes it to
>> simply be passed through to init, with no error message.
>> So either you used non-standard versions of 2.4.22-rc2/2.6.0-test3,
>> or nolapic wasn't the thing that fixed your nforce2 board.
>It probably was a combination of the other measures mentioned in my mail then. 
>I have (had) the same problems, and one has to completely avoid the local 
>APIC it seems. Passing noapic and disabling APIC Mode in the BIOS did not do 
>that (2.6.0-test3 + acpi20030730). 
...
>> "noapic" (note: no "l") might very well fix your board, but that's
>see above. noapic had no effect on the freezes. On boot it still said "found 
>and enabling local APIC"

Of course it did. "noapic" and BIOS APIC mode relate to the I/O-APIC,
not the local APIC.

My guess is that your BIOS or graphics card can't handle the local
APIC, presumably due to a crap SMM# handler or you using APM's
CPU_IDLE or DISPLAY_BLANK options.

So the solution for your board is to forcibly avoid the local APIC.
A DMI blacklist entry would do that, as would configuring the
kernel without local APIC support (!SMP && !UP_APIC).

Kernel 2.6.0-test4 supports "nolapic", so that's an option too.
I'll send that patch to Marcelo for 2.4.23-pre so it should be
in 2.4 eventually.

As for getting the DMI blacklist rule in, you or someone else
with that board has to run dmidecode and prepare & test a patch.

/Mikael