Re: [viro-vfs:work.dcache2] [__dentry_kill()] 1b738f196e: stress-ng.sysinfo.ops_per_sec -27.2% regression

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On 12/6/23, Al Viro <viro@xxxxxxxxxxxxxxxxxx> wrote:
> On Wed, Dec 06, 2023 at 05:30:10PM +0100, Mateusz Guzik wrote:
>
>> > What the hell is going on?  Was ->d_lock on parent serving as a throttle
>> > and reducing
>> > the access rate to something badly contended down the road?  I don't see
>> > anything
>> > of that sort in the profile changes, though...
>
>> Not an outlandish claim would be that after you stopped taking one of
>> them, spinning went down and more traffic was put on locks which *can*
>> put their consumers off cpu (and which *do* do it in this test).
>
> That's about the only guess I've got (see above), but if that's the case,
> which lock would
> that be?
>
>> All that said I think it would help if these reports started including
>> off cpu time (along with bt) at least for spawned threads.
>

Given backtracs I posted it's at least root->kernfs_iattr_rwsem (see
kernfs_iop_permission), but there may be more.

Ultimately I expect would be easy to answer if one was to rebuild with lockstat.

I may end up doing it tomorrow if there is a problem testing this on your end.

-- 
Mateusz Guzik <mjguzik gmail.com>




[Index of Archives]     [Kernel Newbies]     [Security]     [Netfilter]     [Bugtraq]     [Linux FS]     [Yosemite Forum]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux RAID]     [Samba]     [Video 4 Linux]     [Device Mapper]     [Linux Resources]

  Powered by Linux