Re: per_cpu_ptr_to_phys() failure on UV x86_64

[Cliff Wickman <cpw@sgi.com>]

On Thu, Jun 17, 2010 at 07:35:16PM +0200, Tejun Heo wrote:
> On 06/17/2010 07:08 PM, Tejun Heo wrote:
> > (scratching head...) So, that means it's given an address for which
> > !pcpu_addr_in_first_chunk() but outside of vmalloc area.  Strange.
> > I'll find out what's going on.
> 
> Does the following patch work?  The original patch assumed that @addr
> would be the address of the base cpu which isn't true.  I only compile
> tested the patch so it might be broken (sorry, I gotta go somewhere
> now) but this should be the right direction.

Yes, your patch works.  I tested it on a 32p UV system.

-Cliff
 
> diff --git a/mm/percpu.c b/mm/percpu.c
> index 46485e1..8956155 100644
> --- a/mm/percpu.c
> +++ b/mm/percpu.c
> @@ -978,14 +978,23 @@ bool is_kernel_percpu_address(unsigned long addr)
>   */
>  phys_addr_t per_cpu_ptr_to_phys(void *addr)
>  {
> -	if (pcpu_addr_in_first_chunk(addr)) {
> -		if ((unsigned long)addr < VMALLOC_START ||
> -		    (unsigned long)addr >= VMALLOC_END)
> -			return __pa(addr);
> -		else
> -			return page_to_phys(vmalloc_to_page(addr));
> -	} else
> -		return page_to_phys(pcpu_addr_to_page(addr));
> +	void __percpu *base = __addr_to_pcpu_ptr(pcpu_base_addr);
> +	unsigned int cpu;
> +
> +	for_each_possible_cpu(cpu) {
> +		void *start = per_cpu_ptr(base, cpu);
> +
> +		if (addr >= start && addr < start + pcpu_unit_size) {
> +			/* in the first chunk */
> +			if ((unsigned long)addr < VMALLOC_START ||
> +			    (unsigned long)addr >= VMALLOC_END)
> +				return __pa(addr);
> +			else
> +				return page_to_phys(vmalloc_to_page(addr));
> +		}
> +	}
> +	/* in one of the other chunks */
> +	return page_to_phys(pcpu_addr_to_page(addr));
>  }
> 
>  static inline size_t pcpu_calc_fc_sizes(size_t static_size,
> 
> 
> -- 
> tejun

-- 
Cliff Wickman
SGI
cpw@sgi.com
(651) 683-3824