Re: [PATCH 6/6] ARCv2: spinlock/rwlock: Reset retry delay when starting a new spin-wait cycle

[Peter Zijlstra]

On Mon, Aug 03, 2015 at 02:40:13PM +0000, Vineet Gupta wrote:
> On Monday 03 August 2015 05:14 PM, Peter Zijlstra wrote:
> > On Mon, Aug 03, 2015 at 03:33:08PM +0530, Vineet Gupta wrote:
> >> > A spin lock could be available momentarily, but the SCOND to actually
> >> > acquire it might still fail due to concurrent update from other core(s).
> >> > To elide hardware lock, the sequence is retried after a "delay" which is
> >> > increased expoenntially to get a nice backoff behaviour.
> >> > 
> >> > However, this could cause the delay counter to get to a high value. Thus
> >> > when the next spin cycle restarts, reset the counter back to starting
> >> > value of 1.
> > Cute.. fwiw, did you look at what Sparc64 does?
> >
> 
> Can't really comprehend what's special there - are you referring to the special
> branching or to the out of line slow path code for 32 bit version.

Its the only arch that I can remember seeing decay logic in.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/