lkml.org 
[lkml]   [2014]   [Jun]   [2]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
SubjectRe: [PATCH v2] introduce atomic_pointer to fix a race condition in cancelable mcs spinlocks
From
On Mon, Jun 2, 2014 at 9:25 AM, Peter Zijlstra <peterz@infradead.org> wrote:
>
> And I can't say I'm a particular fan of these ops either, as alternative
> I'm almost inclined to just exclude parisc from using opt spinning.

Please do.

There is no way in hell that we should introduce a magic new
atomic_pointer thing for parisc. And the idea somebody had to change
ACCESS_ONCE() for PA-RISC (I'm not going to go back to find who to
blame) is just horribly wrong too, since it's not even necessary for
any normal use: the special "load-and-store-zero" instruction isn't
actually used for "real" data, it's used only for the special
spinlocks afaik, so doing it for all ACCESS_ONCE() users would be
wrong even on PA-RISC. For any normal data, the usual "just load the
value, making sure the compiler doesn't reload it" is perfectly fine -
even on PA-RISC.

Now, if PA-RISC was a major architecture, we'd have to figure this
out. But as it is, PA-RISC is just about the shittiest RISC ever
invented (with original sparc being a strong contender), and let's
face it, nobody really uses it. It's a "fun project", but it is not
something that we should use to mess up either ACCESS_ONCE() or the
MCS locks.

Just make PA-RISC use its own locks, not any of the new fancy ones.

Linus


\
 
 \ /
  Last update: 2014-06-02 19:41    [W:1.390 / U:0.004 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site