On Wed, Sep 18, 2019 at 08:17:36AM +0800, kernel test robot wrote:
> Greeting,
>
> FYI, we noticed a -24.5% regression of aim9.dir_rtns_1.ops_per_sec due to
> commit:
>
>
> commit: c25aa432ff56e179bf5414edff3aa430d2b260c0 ("Fix the locking in
> dcache_readdir() and friends")
> https://kernel.googlesource.com/pub/scm/linux/kernel/git/next/linux-next.git
> master
... and that, folks, is ->d_lock on the parent. I.e. the original reason for
trying to do the damn thing lockless.
I'd really like to do a lockless variant, properly. First we'd need to sort
the misuses of d_subdirs/d_child, though - there are several outright bugs
in the mainline. And right now it looks like it wants to grow a series of
cleanups in autofs. Oh, well...
