Followup to: <Pine.LNX.4.33.0203081252450.1412-100000@penguin.transmeta.com>
By author: Linus Torvalds <torvalds@transmeta.com>
In newsgroup: linux.dev.kernel
>
> On Fri, 8 Mar 2002, Alan Cox wrote:
> >
> > Can we go to cache line alignment - for an array of locks thats clearly
> > advantageous
>
> I disagree about the "clearly". Firstly, the cacheline alignment is CPU
> dependent, so on some CPU's it's 32 bytes (or even 16), on others it is
> 128 bytes.
>
> Secondly, a lot of locking is actually done inside a single thread, and
> false sharing doesn't happen much - so keeping the locks dense can be
> quite advantageous.
>
> The cases where false sharing _does_ happen and are a problem should be
> for the application writer to worry about, not for the kernel to force.
>
> So I think 8 bytes is plenty fine enough - with 16 bytes a remote
> possibility (I don't think it is needed, but it gives you som epadding for
> future expansion). And people who have arrays and find false sharing to be
> a problem can fix it themselves.
>
> I personally don't find arrays of locks very common. It's much more common
> to have arrays of data structures that _contain_ locks (eg things like
> having hash tables etc with a per-hashchain lock) and then those container
> structures may want to be cacheline aligned, but the locks themselves
> should not need to be.
>
Also, on UP this is all a waste.
-hpa
-- <hpa@transmeta.com> at work, <hpa@zytor.com> in private! "Unix gives you enough rope to shoot yourself in the foot." http://www.zytor.com/~hpa/puzzle.txt <amsp@zytor.com> - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Fri Mar 15 2002 - 22:00:10 EST