[RFC MM] speculative page fault

[KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>]

This is just a toy patch inspied by on Christoph's mmap_sem works.
Only for my hobby, now.

Not well tested. So please look into only if you have time.

My multi-thread page fault test program shows some improvement.
But I doubt my test ;) Do you have recommended benchmarks for parallel page-faults ?

Counting # of page faults per 60sec. See page-faults. bigger is better.
Test on x86-64 8cpus.

[Before]
  474441.541914  task-clock-msecs         #      7.906 CPUs
          10318  context-switches         #      0.000 M/sec
             10  CPU-migrations           #      0.000 M/sec
       15816787  page-faults              #      0.033 M/sec
  1485219138381  cycles                   #   3130.458 M/sec  (scaled from 69.99%)
   295669524399  instructions             #      0.199 IPC    (scaled from 79.98%)
    57658291915  branches                 #    121.529 M/sec  (scaled from 79.98%)
      798567455  branch-misses            #      1.385 %      (scaled from 79.98%)
     2458780947  cache-references         #      5.182 M/sec  (scaled from 20.02%)
      844605496  cache-misses             #      1.780 M/sec  (scaled from 20.02%)

[After]
471166.582784  task-clock-msecs         #      7.852 CPUs
          10378  context-switches         #      0.000 M/sec
             10  CPU-migrations           #      0.000 M/sec
       37950235  page-faults              #      0.081 M/sec
  1463000664470  cycles                   #   3105.060 M/sec  (scaled from 70.32%)
   346531590054  instructions             #      0.237 IPC    (scaled from 80.20%)
    63309364882  branches                 #    134.367 M/sec  (scaled from 80.19%)
      448256258  branch-misses            #      0.708 %      (scaled from 80.20%)
     2601112130  cache-references         #      5.521 M/sec  (scaled from 19.81%)
      872978619  cache-misses             #      1.853 M/sec  (scaled from 19.80%)


Main concept of this patch is
 - Do page fault without taking mm->mmap_sem until some modification in vma happens.
 - All page fault via get_user_pages() should have to take mmap_sem.
 - find_vma()/rb_tree must be walked under proper locks. For avoiding that, use
   per-thread cache.

It seems I don't have enough time to update this, more.
So, I dump patches here just for share.

Thanks,
-Kame

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>