$20 sent to whoever gives me the fix to this problem.

[kgroombr]

Seriously, I will send $20 to the person who gives me the fix to this.

As I posted in a previous post, I was getting severe lockups with my system. I have now fixed the lockups but at the expense of a system that runs slower than molasses in January.

The system:

Abit NF7-S 2.0 with the BIOS 18 update (latest one as of 11 Aug). Barton 2500+, Maxtor 160GB and Seagate 80GB on a promise ultra 133 tx2 controller. LG DVD on EIDE #1 of the nforce2 and a Zip 250 on EIDE #2 of the nforce2.

Video card: Xtasy Geforce3 Ti500
Kworld BT878 capture card
Advansys SCSI controller

Kernel is 2.4.22-pre2 (something between 2.4.21 and this pre2 kernel fixed my lockups, but now it runs super slow).

For some reason I have to have ACPI turned on in order for my lockups to go away. This causes my system to run super slow and here is what I think is going on:

Everytime I added a device that required an interrupt (using modprobe), my system would get progressively slower. At one point it was so slow that it took (no joke) about 45 minutes to load X.

Here is /proc/interrupts immediately after booting:

CPU0
0: 21576 IO-APIC-edge timer
1: 132 IO-APIC-edge keyboard
2: 0 XT-PIC cascade
6: 6 IO-APIC-edge floppy
8: 1 IO-APIC-edge rtc
9: 0 IO-APIC-level acpi
12: 4745 IO-APIC-edge PS/2 Mouse
14: 7 IO-APIC-edge ide0
15: 22 IO-APIC-edge ide1
16: 23970871 IO-APIC-level bttv
18: 25148787 IO-APIC-level ide2, ide3
19: 19967973 IO-APIC-level nvidia
22: 10012602 IO-APIC-level NVidia nForce2
NMI: 0
LOC: 21524
ERR: 51
MIS: 12207


Anything loaded in the APIC irq range is totally wacked. IRQ 22 is my sound, and I noticed that it was the only one that currently not changing.

Here is /proc/interrupts only moments later:

CPU0
0: 35578 IO-APIC-edge timer
1: 1111 IO-APIC-edge keyboard
2: 0 XT-PIC cascade
6: 13 IO-APIC-edge floppy
8: 1 IO-APIC-edge rtc
9: 0 IO-APIC-level acpi
12: 5185 IO-APIC-edge PS/2 Mouse
14: 7 IO-APIC-edge ide0
15: 22 IO-APIC-edge ide1
16: 40344618 IO-APIC-level bttv
18: 41495938 IO-APIC-level ide2, ide3
19: 36337043 IO-APIC-level nvidia
22: 10012602 IO-APIC-level NVidia nForce2
NMI: 0
LOC: 35490
ERR: 53
MIS: 12345

Checking out my processes, I found a aplay process that appeared to be hung up (plays the intro music when logging into KDE). I killed the process and checked out /proc/interrupts again:

CPU0
0: 41843 IO-APIC-edge timer
1: 1568 IO-APIC-edge keyboard
2: 0 XT-PIC cascade
6: 20 IO-APIC-edge floppy
8: 1 IO-APIC-edge rtc
9: 0 IO-APIC-level acpi
12: 5185 IO-APIC-edge PS/2 Mouse
14: 7 IO-APIC-edge ide0
15: 22 IO-APIC-edge ide1
16: 46664621 IO-APIC-level bttv
18: 47809364 IO-APIC-level ide2, ide3
19: 42653097 IO-APIC-level nvidia
22: 21695941 IO-APIC-level NVidia nForce2
NMI: 0
LOC: 41735
ERR: 53
MIS: 12401

In order to run these, I had to rmmod by SCSI card (for my CDRW), and my USB ports). If I didn't, the system would be even so slow it would have taken me forever to do this test.

Again, if I turn APIC off, the system runs super fast, but unfortunately will lock in in about two to three minutes.

Sound will sometimes play a fraction of a second of the sound, but then appears to hang.

It appears that the IRQs go berzerk with APIC.

Again, $20 goes to the person that gives me the correct fix to this problem. No hack, and nothing like "buy another motherboard".

Need more info, just let me know. More than happy to provide.

Thanks,
Ken

[Cao]

Ok, I'm not using the 2.4.x series, so the best advice I can give you is this: grab the ACPI update patch from acpi.sf.net. Andrew DeQuincey worked very hard to fix the ACPI IRQ routing and APIC support for nForce2 chipsets, his work was accepted by ACPI developers and is already in 2.6.x series. Marcelo tends to be slow in catching up with ACPI, so the code is not so new in 2.4.x and maybe his work isn't in 2.4.22-pre yet.

Even if his code is already merged and didn't fix your problem, he's the one you should talk to.

Or maybe the easy way out could work for you: "noacpi pci=noacpi" in your kernel parameters.

BTW: if this solves your problem, Andrew is the one you should send the money to.

[kgroombr]

Thanks for taking a look. Turning off acpi with noapci or pci=noacpi does make the system run fast as it should, but it causes it to lock up hard in two to three minutes.

I tried the acpi patches with the 2.6-test3 kernel with no success. Didn't know there were updated ones for the 2.4.X series.

Thanks, if the patches work, I will send him the money.

Ken

[Cao]

If 2.6.0-test3 didn't solve your problems, the ACPI patches won't work either, as it has always the latest ACPI code. I have a setup very similar to yours and don't have any problem. So, I think the fault is in the Promise controller.

Did you test messing around with hdparm options?
Accessing data on the CD-ROM or Zip (as playing a .wav sound with aplay) suffer the same problem?
Disabling DMA does any good? (hdX=nodma)
APIC enabled without ACPI IRQ routing hangs too?

[kgroombr]

Way ahead of you on that. I have removed everything from this system and put one hard drive right on the nforce EIDE controller. The only card on it was my video card. Reloaded Linux just on that one drive. Used hdparm to turn on dma, turn off dma, try everything that I knew caused stablity problems with hard drives and dma.

I reloaded windows 2000 on this, and run two stress test programs (not one, but two at the same time that consistently banged the CDROM, hard drives, video, network, etc). It ran for three straight days non-stop without a hiccup.

Just another little addtion, when acpi is on (currently the only way my system runs stable, but awefully slow), I get a message when I boot:

..MP-BIOS bug: 8254 timer not connected to the IO-APIC

and sometimes, but not too often, when I shutdown, I get this message:

APIC error on CPU0: 02(02)

Google searches on those didn't turn up anything really definitive.

[pzgren]

[Cao]

[alexschr]

try booting with noapic and EDIT: acpi=off
oh eehh btw : if anyone know how to fix this problem post the solution here to to double your money

[kgroombr]

It is strange that 2.4.22-rc2 does't lockup, but if I run it with pci=noapic or acpi=off, my lockup problem comes back. Most of the people I know with this problem say that interrupts are somehow causing the problem. I have to agree with that from what I see now. Everybody that I have found with this problem have the nforce2 chipset revision 162.

Thanks,
Ken

[hchu]