Re: [PATCH] proc: Avoid a thundering herd of threads freeing proc dentries

[Junxiao Bi <junxiao.bi@oracle.com>]

On 6/19/20 10:24 AM, ebiederm@xmission.com wrote:

> Junxiao Bi <junxiao.bi@oracle.com> writes:
>
>> Hi Eric,
>>
>> The patch didn't improve lock contention.
> Which raises the question where is the lock contention coming from.
>
> Especially with my first variant.  Only the last thread to be reaped
> would free up anything in the cache.
>
> Can you comment out the call to proc_flush_pid entirely?

Still high lock contention. Collect the following hot path.

     74.90%     0.01%  proc_race 
[kernel.kallsyms]                               [k] 
entry_SYSCALL_64_after_hwframe
             |
              --74.89%--entry_SYSCALL_64_after_hwframe
                        |
                         --74.88%--do_syscall_64
                                   |
                                   |--69.70%--exit_to_usermode_loop
                                   |          |
                                   |           --69.70%--do_signal
                                   |                     |
                                   | --69.69%--get_signal
                                   | |
                                   | |--56.30%--do_group_exit
                                   | |          |
                                   | |           --56.30%--do_exit
                                   | |                     |
                                   | |                     
|--27.50%--_raw_write_lock_irq
                                   | |                     |          |
                                   | |                     | 
--27.47%--queued_write_lock_slowpath
                                   | |                     
|                     |
                                   | |                     | 
--27.18%--native_queued_spin_lock_slowpath
                                   | |                     |
                                   | |                     
|--26.10%--release_task.part.20
                                   | |                     |          |
                                   | |                     |           
--25.60%--_raw_write_lock_irq
                                   | |                     
|                     |
                                   | |                     | 
--25.56%--queued_write_lock_slowpath
                                   | |                     
|                                |
                                   | |                     | 
--25.23%--native_queued_spin_lock_slowpath
                                   | |                     |
                                   | |                      --0.56%--mmput
                                   | |                                |
                                   | |                                 
--0.55%--exit_mmap
                                   | |
|                                 --13.31%--_raw_spin_lock_irq
|                                           |
| --13.28%--native_queued_spin_lock_slowpath
                                   |

Thanks,

Junxiao.

>
> That will rule out the proc_flush_pid in d_invalidate entirely.
>
> The only candidate I can think of d_invalidate aka (proc_flush_pid) vs ps.
>
> Eric