Re: [PATCH RFC] futex: avoid false sharing between hb->chain and the bucket lock

[Breno Leitao]

On Wed, Jun 10, 2026 at 01:25:46PM +0200, Peter Zijlstra wrote:
> On Wed, Jun 10, 2026 at 01:22:34PM +0200, Thomas Gleixner wrote:
> > On Tue, Jun 09 2026 at 22:18, Peter Zijlstra wrote:
> > > On Tue, Jun 09, 2026 at 10:11:17PM +0200, Peter Zijlstra wrote:
> > >> Anyway, how does something like the below work for you? It's a total
> > >> hack job, but it (sorta) builds and runs.
> > >> 
> > >
> > > Please use this one, I spotted a silly bug.
> > 
> > So I ran this on two machines.
> > 
> > SKL dual socket 112 threads:
> > 
> > 		Baseline	Patched
> > 
> > shared (16k)	1571857 	1641435         + 4.4%
> > autosize (512)	 646390 	 903371         +39.7%
> > -b 256		 464395 	 587014         +26.4%
> > -b 512		 715687 	 995943         +39.2%
> > -b 1024		 995085 	1396328         +40.3%
> > -b 2048		1293114         1668395         +29.0%
> > -b 4096		2124438 	2240228         + 5.5%
> > 
> > Zen3 dual socket 256 threads:
> > 
> > 		Baseline	Patched
> > 
> > shared	(16k)	1275840		1381279	 	+ 8.2%
> > autosize (512)	1252745		1482179		+18.3%
> > -b 256		 856274		 955455		+11.5%
> > -b 512		1267490		1544010		+21.8%
> > -b 1024		1424013		1625424		+14.1%
> > -b 2048		1505181		1669342		+10.9%
> > -b 4096		1465993		1688932		+15.2%
> 
> I suppose that means I'd better go make it prettier and survive
> randconfig :-)

I've Peter it here on the same machine I used earlier 176-thread AMD EPYC host,
10s perf bench futex hash per run, baseline = parent commit (acb7500801e98):

                       Baseline       Patched      Delta
  shared (16 buckets)  1,230,599      1,368,655    +11.2%
  autosize (1024)      1,285,440      1,556,946    +21.1%
  -b 256               1,341,471      1,520,303    +13.3%
  -b 512               1,438,330      1,599,319    +11.2%
  -b 1024              1,443,772      1,622,493    +12.4%
  -b 2048              1,472,108      1,643,975    +11.7%
  -b 4096              1,333,098      1,570,897    +17.8%

Stderr was 0.06%-0.22% across the board, so the deltas are well
outside noise.

The trade Peter sketched holds up here: no extra futex memory
cost, and we still recover most of what padding the bucket would
have bought.

Really good, thanks for your this patch,
--breno