Jack Steiner wrote:

>On Sat, May 01, 2004 at 02:17:04PM -0700, William Lee Irwin III wrote:
>> On Sat, May 01, 2004 at 07:08:05AM -0500, Jack Steiner wrote:
>> > On a 512p idle 2.6.5 system, each cpu spends ~6% of the time in the kernel
>> > RCU code. The time is spent contending for shared cache lines.
>> 
>> Would something like this help cacheline contention? This uses the
>> per_cpu data areas to hold per-cpu booleans for needing switches.
>> Untested/uncompiled.
>  
>
> [snip]
>
>We use 64MB granules. The percpu data structures on the individual nodes 
>are separated by addresses that differ by many GB. A scan of all percpu 
>data structures requires a TLB entry for each node.  This is costly & 
>trashes the TLB.  (Our max system size is currently 256 nodes).
>  
>
rcu_cpu_mask serves two purposes:
- if a bit for a cpu is set, then the cpu must report quiescent states. Thus all cpus continuously poll their own bits.
- If a cpu notices that it completed a quiescent state, it clears it's own bit, checks all bits and completes the grace period. This happens under a global spinlock.

What about splitting that into two variables?
A global counter could be used to inform the cpus that they should look for quiescent states. The cacheline would be mostly read-only, it shouldn't cause trashing.

Attached is a patch - it boots uniprocessor i386. What do you think? It should remove the hotspot from scheduler_tick entirely.

--
	Manfred