Re: [PATCH] - Optional method to purge the TLB on SN systems

[Jack Steiner <steiner@sgi.com>]

On Tue, Mar 27, 2007 at 01:24:27PM -0700, Luck, Tony wrote:
> +smp_flush_tlb_cpumask (cpumask_t xcpumask)
> +{
> +	unsigned short counts[NR_CPUS];
> 
> Even as a "short" ... that is still a lot of stack.  Possibly
> a new place to fail when NR_CPUS gets bumped up again?

I was also somwhat concerned about that, too. At NR_CPUS\x1024, it
uses 2k bytes whick is big but I think ok. Even at NR_CPUS 48,
I think is is still ok but I'm getting a little nervous. 

FWIW, slab.c has a "struct array_cache *new[NR_CPUS]" on the
stack.

Also, net/core/utils.c has "unsigned long seed[NR_CPUS]".

I'm only adding 1/4 as much space. Can I get away with it or
should I allocate a node-local chunk of memory & keep a pointer to
it in per-cpu data?



> 
> +#define IA64_IPI_LOCAL_TLB_FLUSH	0xfc	/* SMP reschedule */
>  #define IA64_IPI_RESCHEDULE		0xfd	/* SMP reschedule */
> 
> Cut & pasted the comment when duping this line?

Whoops :-) 
Fixed....