On 08/05/14 19:34, Nicolas Pitre wrote:

quoted

It allows us to synchronize with another CPU that may be inside
gic_raise_softirq(). If the other CPU was in that function then this CPU
would wait until it was done sending the IPI to continue along and
reroute them. If the other CPU was just about to grab the sgi lock then
we would guarantee that the CPU would see the new gic_cpu_map value and
thus any redirection is not necessary.

OK I get it now.

quoted

I hoped that the commit text explained this.

I'm possibly not bright enough to get it the first time.

quoted

Honestly it probably isn't a noticeable performance boost either way 
but I think this is the best we can do.

Sure, agreed.

Ok, so which patch is preferred?

-- 
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
hosted by The Linux Foundation