Re: [PATCH 1/3] Per-cpu UP unification

[Andrew Morton]

Rusty Russell <[e-mail]> wrote:
>
> [ Untested on ia64, but fairly trivial if I've broken something ].
> 
> Name: Unification of per-cpu headers for non-SMP
> Author: Rusty Russell
> Status: Trivial

I applied all these to the ppc64 kernel (you missed ppc64 btw) and it dies.

Quite late in boot, during floppy_init->init_disk_stats->alloc_percpu.

I'm reduced to debugging with printk on ppc64.  __alloc_percpu() calls
new_block(), loops around and then dies in here:

#define D() printk("at %s:%d\n", __FILE__, __LINE__)

			D();
			/* Transfer extra to previous block. */
			if (b->size[i-1] < 0)
				b->size[i-1] -= extra;
			else
				b->size[i-1] += extra;
			b->size[i] -= extra;
			ptr += extra;

			D();

Not sure what happened - no oops, no xmon, no sysrq, no nuthin.  It even
manages to lock up minicom on the other end of the cable.  Impressed.