Messages in this thread | | | Date | Thu, 5 Sep 2013 12:21:10 -0700 | Subject | Re: [PATCH] lockref: remove cpu_relax() again | From | Linus Torvalds <> |
| |
On Thu, Sep 5, 2013 at 11:57 AM, Luck, Tony <tony.luck@intel.com> wrote: > > Not "tons", just two. You can ask for "acquire" or "release" semantics, > there is no relaxed option.
Seriously? You can't just do a cache-coherent cmpxchg without extra serialization? Oh well.
> Worse still - early processor implementations actually just ignored > the acquire/release and did a full fence all the time. Unfortunately > this meant a lot of badly written code that used .acq when they really > wanted .rel became legacy out in the wild - so when we made a cpu > that strictly did the .acq or .rel ... all that code started breaking - so > we had to back-pedal and keep the "legacy" behavior of a full fence :-(
Ugh. Can you try what happens with the weaker release-semantics performance-wise for that code? Do it *just* for the lockref code..
Linus
| |