Re: [PATCH] compile-time configurable NR_CPUS

[William Lee Irwin III]

On Thu, 2002-08-29 at 17:42, William Lee Irwin III wrote:
>> Could you make CONFIG_NR_CPUS only for non-NUMA-Q systems and hardwire
>> it to 32 for NUMA-Q, as the bugs in io_apic.c don't have fixes yet and
>> NUMA-Q's have enough IO-APIC's to trigger the bugs.

On Thu, Aug 29, 2002 at 05:44:20PM -0400, Robert Love wrote:
> Linus has not shown any interest in merging, so it is a non-issue at the
> moment...

devfs doesn't hold a candle to io_apic.c

I did 3 runs on a 32x last night, and got 3 panics not present in 2.4:
(1) "Recompile kernel with bigger MAX_IO_APICS!.\n",
	so I bumped up MAX_IO_APICS to "impossibly huge"
(2) "Max APIC ID exceeded!\n", so I removed the physid reprogramming
(3) "ran out of interrupt sources!",
	and I chugged a stiff drink, gave up, & went to bed (this is evil)

Reducing NR_CPUS tends to trigger some kind of physical APIC ID
reprogramming panic() (msg #2 above) that doesn't normally happen.

Cheers,
Bill