[prev in list] [next in list] [prev in thread] [next in thread] 

List:       linux-smp
Subject:    APIC problems on 2.2.2 w/ Ga686DLX
From:       hausutzu () t-bird ! in-berlin ! de (Utz-Uwe Haus)
Date:       1999-02-23 22:43:14
[Download RAW message or body]

Hi everyone,

I'm the one of the poor people for whom the IO-APIC on the GA686DLX
still does not work: Under high load the interrupts seem to be delivered
incorrectly to the busy device (eth, scsi) and the timer gets way out of
sync with real time so that even xntp flakes out.
I had settled for 2.1.131-ac13 (IIRC) with the noapic option, since that
was a clean inplementation of my 'fix' of forcing the interrupts to be
treated as XT.
I'd still be interested in resolving the problem, but start to believe
it's just a problem with this board rev or so (since there are people
with problems on this board, and others without).

The noapic option however does not work on 2.2.2: The usual APIC-table
is not printed, but the PCI->APIC transform as in

Feb 23 22:47:49 cookie kernel: PCI->APIC IRQ transform: (B0,I10,P0) -> 18
Feb 23 22:47:49 cookie kernel: PCI->APIC IRQ transform: (B0,I12,P0) -> 16

happens anyway !? This does not work (not that I expected it to),
resulting in scsi timeouts when it tries to detect the disks (therefor
no complete bootlog, sorry).

And just for the record:
GA-686DLX, current BIOS update done
2xPII-266, 128MB
on-board AIC7xxx
Cnet Tulip-21140A 10/100Mb ethernet, using tulip driver
Cnet Isa-NE2000 ethernet

and the symptoms are:
when creating heavy ethernet traffic on the 100Mb card (tcpspray in both
directions for a few minutes continuously, maybe a few pings with large
packets) after maybe 10-15 minutes the ethernet card locks up: it does
not receive any packets anymore, tcpdump shows arp request for the other
host going out. Interrupts on the device are no longer generated,
rmmod/insmod does not cure problem, reboot does :|
When I replace the tulip by a 3c905B I get 'transmit timed out' errors
but otherwise the same behaviour. A realtek 8139 showes the same
behaviour, so it's most probably not ethernet-card related.
When not stress testing, the machine runs maybe 12h to 1 day, and flakes
out with the same problem, or scsi command timed out & abort, leading to
fs damage.
All the time the system time ist sometimes doubling it's speed, drifting
up to 32s per 64s-xntpd interval, which xntp refuses to correct. 

Can anybody shed light on this? at least fix the noapic option?

Greetings
Utz

-- 
Utz-Uwe Haus                                       hausutzu@t-bird.in-berlin.de
PGP key available, Fingerprint:                  or                 haus@zib.de
	 1024/6AD23BE1 --  3E 0D 3B 81 30 BC 5F 1A  DF 60 8E D7 C5 11 F3 83
 "... dont' worry, base 8 is just like base 10 -- if you're missing 2 fingers"

-
Linux SMP list: FIRST see FAQ at http://www.irisa.fr/prive/mentre/smp-faq/
To Unsubscribe: send "unsubscribe linux-smp" to majordomo@vger.rutgers.edu

[prev in list] [next in list] [prev in thread] [next in thread]