Messages in this thread | | | Date | Fri, 7 Feb 2014 13:28:37 +0100 | From | Peter Zijlstra <> | Subject | Re: [PATCH] Convert powerpc simple spinlocks into ticket locks |
| |
On Fri, Feb 07, 2014 at 12:49:49PM +0100, Torsten Duwe wrote: > On Fri, Feb 07, 2014 at 11:45:30AM +0100, Peter Zijlstra wrote: > > > > That might need to be lhz too, I'm confused on all the load variants. > > ;-) > > > > unlock: > > > lhz %0, 0, &tail > > > addic %0, %0, 1 > > No carry with this one, I'd say.
Right you are, add immediate it is.
> Besides, unlock increments the head.
No, unlock increments the tail, lock increments the head and waits until the tail matches the pre-inc value.
That said, why do the atomic_inc() primitives do an carry add? (that's where I borrowed it from).
> > > lwsync > > > sth %0, 0, &tail > > > > > Given the beauty and simplicity of this, may I ask Ingo: > you signed off 314cdbefd1fd0a7acf3780e9628465b77ea6a836; > can you explain why head and tail must live on the same cache > line? Or is it just a space saver? I just ported it to ppc, > I didn't think about alternatives.
spinlock_t should, ideally, be 32bits.
> What about > > atomic_t tail; > volatile int head; ? > > Admittedly, that's usually 8 bytes instead of 4...
That still won't straddle a cacheline unless you do weird alignement things which will bloat all the various data structures more still.
Anyway, you can do a version with lwarx/stwcx if you're looking get rid of lharx.
| |