From Bugzilla Helper: User-Agent: Mozilla/4.61 [en] (OS/2; U) After sucessful installation of RH7.1 SMP, the machine would not reboot into SMP. It either crashes after: ENABLING IO-APIC IRQs ... changing IO-APIC physical APIC ID to 2...ok Synchronizing Arb IDs ..TIMER: Vector=49 pin1=2 pin2=0 or after this: Warning: unexpected IO-APIC calibrating APIC timer..... ...CPU clock speed is 1004.6878 MHz ...host bus clock speed is 133.9582 MHz cpu:0, clocks:1339582, slice:446527 CPU0<T0:1339568, T1:893040, D:1, S:446527, C:1339582> Reboot into single CPU mode works OK. Reproducible: Always Steps to Reproduce: 1. Install RH7.1 on a ASUS CUV4X-DLS 2xCPU machine 2. Reboot 3. Reboot crashes. The only way out is Reset. Hardware: ASUS CUV4X-DLS MoBo, 2xPIII/1GHz, 1GB RAM
I can confirm this is a known 2.4 problem. I don't currently know of a cure.
Ingo: any ideas ?
Could you please let me know, when do you expect this bug to be resolved? I have two machines that must go into production shortly. If the bug is not going to be resolved in the near future, I will have to contact my supplier and ask him to change ASUS MoBo's for TYAN. Thanks
One more question: Can I downgrade to 2.2.19 kernel? Will it work with R7.1 distribution, or should I install RH7.0?
All the configscripts and such are tuned to a 2.4 kernel. I would recommend using 7.0 if you need a 2.2.19 kernel.
I dont know of any cure either, yet. It's tough to debug these sort of problems remotely. One suspect could be some sort of 32-bit overflow due to the high frequency of the CPU - is it possible to somehow slow the CPU's speed down in the BIOS? another thing to try would be to add printks to the boot code, to see exactly where the lockup happens. question: does the system boot the SMP kernel properly if you add numcpus=1 as a boot option? It could be a startup problem of the second CPU.
The boot of the SMP kernel with "numcpus=1" crashes exactly the same way as it does without it.
I tried to slow down the CPUs to 500MHz. The crash was exactly the same "Unexpected IO-APIC".
I just tried to run at 800 MHz CPU's and 100 MHz bus. The crash was the same. I do not believe it has anything to do with speed. I am replacing the ASUS MoBo's with TYAN S2507 Tiger 230. I will report the result.
TYAN S2507 Tiger 230 MoBo, though having the same chipset as ASUS, does _not_ have the problem. 2.4 SMP kernel booting and running just fine.
This makes it _REALLY_ sound like a BIOS problem then. Or the quality of wiring, you never know.
I have same problem with dual 600M P3s on an ASUS CUV4X-D with 1G. Why is this same hardware able to boot RH7.0 SMP 2.2 kernels but is unable to boot RH7.1 SMP 2.4.2-2? I have tried nodma, noapm, nousb and numcpus=1 with no help. UP is OK.
Hey, I just compiled kernel 2.4.4 with SMP option and it works! dmesg looks a little different but I get "PCI_>APIC IRQ transform: (B1,I10,PO) -> 18" etc. 2.4.4 has the serial driver I am looking for, too. dmesg says "usb.c: registered a new driver hub" "found SMP MP-table at 000f5460" "Processors: 2" and "WARNING: unexpected IO-APIC, please mail to linux-smp.org". I don't see anything about DMA, what should I look for or did I loose this for IDE?
How bizarre. Last night I booted SMP kernel 2.4.4 twice and left it on overnight. This morning there's a blank screen but the monitor is still on full power and it won't wake up. I have to hold the power button in for a few seconds before it will shutoff the computer. Then it wouldn't boot SMP kernel 2.4.4 the first few tries. Then it booted. I rebooted with noapm option, guess I'll see how well that takes.
noapic might be a better option to try
please check whether 2.4.4 fixes the boot problem for everyone. The 'hangs after a night' problem is likely an unrelated issue.
noapic cleans up alot while the MB BIOS is set to "Plug and Play OS = Y". Some of the SMP FAQ recommends "Plug and Play OS = N" and this is the fix I needed to get kernel 2.4.4 to boot cleanly and my soundcard is assigned to IRQ17 (without noapic boot option). With noapic boot option the soundcard is assigned IRQ5. Kernel 2.4.4 has a .config option for the VIA VT82C* pci chipsets used on ASUS CUV4X-D and this will enable UDMA(100).
For me, adding 'append="noapic"' to the smp kernel section of lilo.conf was sufficient for everything to work with a stock RH 7.1 installation. Tony
SUCCESS! It appears that disabling MPS 1.4 in the motherboard's BIOS solves the problem. As I went hunting for more info on this problem, I ran across this message from LKML in March that may inform ... http://www.cs.helsinki.fi/linux/linux-kernel/2001-10/0217.html
Alternative workaround: If the secondary on-board IDE controlled is dissabled (Advanced->Chip Config) then it is possible to boot both the 2.4.6-3.1smp (Roswell-1) and 2.4.7-2smp (Roswell-2) kernels without appending 'noapic'. I haven't tried the stock 7.1 kernel.