Re: [MM] Make mm counters per cpu instead of atomic V2

[Christoph Lameter]

On Sat, 7 Nov 2009, KAMEZAWA Hiroyuki wrote:

> And allocate mm->usage only when the first CLONE_THREAD is specified.

Ok.

> if (mm->usage)
>     access per cpu
> else
>     atomic_long_xxx

If we just have one thread: Do we need atomic access at all?

> and read operation will be
>
>     val = atomic_read(mm->rss);
>     if (mm->usage)
>         for_each_possible_cpu()....

or
   val = m->rss
   for_each_cpu(cpu) val+= percpu ...


> ==
> Does "if" seems too costly ?

The above method would avoid the if.

> If this idea is bad, I think moving mm_counter to task_struct from
> mm_struct and doing slow-sync is an idea instead of percpu.

Yeah then the access is effectively percpu as long as preempt is disabled.

But then for the mmap_writer_lock we would need to traverse a doubly
linked list to add up the counters. Bad caching on that one and we would
have to lock the list too. Sigh.

> kswapd and reclaim routine can update mm_struct's counter, directly.
> Readers just read mm_struct's counter.

Would work for rss counters but not for avoiding the rw semaphore I guess.

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/