Re: [PATCH] percpu updates

[Brian Gerst <bgerst@didntduck.org>]

Andrew Morton wrote:
> Brian Gerst wrote:
> 
>>These patches convert some of the existing arrays based on NR_CPUS to
>>use the new per cpu code.
>>
>>...
>>-extern struct page_state {
>>+struct page_state {
>>        unsigned long nr_dirty;
>>        unsigned long nr_locked;
>>        unsigned long nr_pagecache;
>>-} ____cacheline_aligned_in_smp page_states[NR_CPUS];
>>+};
>>+
>>+extern struct page_state __per_cpu_data page_states;
> 
> 
> When I did this a couple of weeks back it failed in
> mysterious ways and I ended up parking it.  Failure
> symptoms included negative numbers being reported in
> /proc/meminfo for "Locked" and "Dirty".
> 
> How well has this been tested?  (If the answer
> is "not very" then please wait until I've tested
> it out...)
> 
> -
> 

Well, the answer is not very.  I don't have an SMP machine to do 
thorough testing on.  The best I can do is boot an SMP kernel on a UP 
machine.  I did check the disassembly of vmlinux, and it looked like it 
would work as advertised.

-- 

						Brian Gerst