Re: [PATCH] readahead: enforce full sync mmap readahead size

[Nick Piggin <npiggin@suse.de>]

On Sun, Apr 12, 2009 at 08:15:12AM -0700, Linus Torvalds wrote:
> 
> 
> On Sun, 12 Apr 2009, Wu Fengguang wrote:
> >
> > Now that we do readahead for sequential mmap reads, here is
> > a simple evaluation of the impacts, and one further optimization.
> 
> Hmm.
> 
> Wu, I just went through your latest (?) series of 1-9 and they all looked 
> (a) quite small and (b) all of them looked like good cleanups. 
> 
> And not only do they look good, you seem to have numbers to back it all up 
> too.

They look pretty good to me too.

 
> In other words, I'd really prefer to merge this sooner rather than later. 
> There just doesn't seem to be any reason _not_ to. Is there any reason to 
> not just take this? I realize that it's past -rc1, but this is way smaller 
> and saner-looking than the average patch that makes it in past -rc1.
> 
> Besides, it was originally posted before -rc1, and the last series didn't 
> have the much more intrusive page-fault-retry patches. I'd leave those for 
> the next merge window, but the read-ahead series (1-9 plus this final 
> one-liner) seem to be pure improvement - both in code readability _and_ in 
> numbers - with no real contentious issues.
> 
> No?

I guess untested code, especially with heuristics, can always cause non
intuitive problems for some people. So the other side of the argument
is what's the harm in putting them in -mm until next merge window?

That said, I like these kinds of things, so I don't object :)