Re: [lkp] [sched/preempt] 103637a5b94: 12.4% fsmark.app_overhead, -1.4% fsmark.files_per_sec

2015-07-26 Thread Peter Zijlstra
On Sat, Jul 25, 2015 at 07:32:03PM +0300, Konstantin Khlebnikov wrote: > I guess that's because of cond_resched_lock() usage in nfs_scan_commit_list(). > My patch fixes cond_resched_lock() which never worked since v3.13. Yeah, I wasn't surprised to see preemptions go up with this patch, not total

Re: [lkp] [sched/preempt] 103637a5b94: 12.4% fsmark.app_overhead, -1.4% fsmark.files_per_sec

2015-07-25 Thread Konstantin Khlebnikov
I guess that's because of cond_resched_lock() usage in nfs_scan_commit_list(). My patch fixes cond_resched_lock() which never worked since v3.13. On Sat, Jul 25, 2015 at 9:09 AM, kernel test robot wrote: > FYI, we noticed the below changes on > > git://internal_merge_and_test_tree > revert-10363

[lkp] [sched/preempt] 103637a5b94: 12.4% fsmark.app_overhead, -1.4% fsmark.files_per_sec

2015-07-24 Thread kernel test robot
FYI, we noticed the below changes on git://internal_merge_and_test_tree revert-103637a5b947af7e6abb8d19b341acff6fbf1ec3-103637a5b947af7e6abb8d19b341acff6fbf1ec3 commit 103637a5b947af7e6abb8d19b341acff6fbf1ec3 ("sched/preempt: Fix cond_resched_lock() and cond_resched_softirq()") ===