2.5.59-mm5: cpu1 not working
A message that is not in the archive (sent off-list or not yet received): [email protected]
Quoted 16 lines · On Mon, 27 Jan 2003 [email protected] wrote:
Quoted 11 lines
Quoted 7 lines · On Mon, 27 Jan 2003 16:34:22 -0500 (EST) Bill Davidsen <[email protected]> wrote:
That's the thing I would expect to see if you used 'noapic' and watchdog. I posted over the weekend that I have been seeing some inobvious results to IPC benchmarking with scombinations of noapic and watchdog, but I didn't snap the interrupts. I'll be happy to add that to my list of stuff to try next weekend, but the box is not a toy during the week, and I would have a five hour round trip drive if a reboot failed, so I'll pass on trying it until I'll in the same room.
Quoted 4 lines
no "noapic" here:
kernel
/boot/linux-2.5.59-mm5root=/dev/hda5 ro vga=0x30a profile=2 nmi_watchdog=1
Quoted 9 lines
There was a seperate thread about this a couple of days ago. It boils down to this: if there isn't sufficient load on your CPUs then it is both easier and faster to have one CPU handle all of the interrupts, if there is sufficient load on your CPUs then the kirq patch distributes the interrupts accordingly. I can verify that this behavior is consisent with what I see on my dual Xeon box (shows up as 4 ht capable CPUs). As long as your second CPU is showing up in
/proc/cpuinfoand in top then the CPU is working it just isn't having to handle any interrupts yet.
The reason I mentioned noapic is that when running with the apic I see a very good balance on a system with minimal load other than CPU. Since there was a count on NMI the watchdog was clearly on. But on my test system, I just check the count and the balance is very good:
Script started on Mon Jan 27 22:38:37 2003
bash-2.05a$ uname -a
Linux bilbo.tmr.com 2.5.59 #6 SMP Sat Jan 25 19:46:18 EST 2003 i686 unknown bash-2.05a$ cat /proc/interrupts
CPU0 CPU1
0: 88705542 91864773 IO-APIC-edge timer 1: 853 850 IO-APIC-edge i8042 2: 0 0 XT-PIC cascade 3: 7752 7269 IO-APIC-edge NE2000 8: 1 1 IO-APIC-edge rtc 9: 0 0 IO-APIC-level uhci-hcd 11: 24218 28443 IO-APIC-level ide2 12: 26 26 IO-APIC-edge i8042 14: 664226 665346 IO-APIC-edge ide0 15: 1 1 IO-APIC-edge ide1 NMI: 0 0 LOC: 180596403 180596410 ERR: 25 MIS: 0
bash-2.05a$ exit
Script done on Mon Jan 27 22:39:00 2003
The system is up >24hr, so the irq counts reflect very little io. Running noapic obviously puts all the ints on CPU0.
Signature · 3 lines
bill davidsen <[email protected]> CTO, TMR Associates, Inc Doing interesting things with little computers since 1979.