Re: [PATCH] 2.5: push BKL out of llseek

[Linus Torvalds]

On Tue, 29 Jan 2002, Andrew Morton wrote:
>
> And llseek is *fast*.  If we're seeing significant
> lock contention in there then adding a schedule() is
> likely to turn Anton into one unhappy dbencher.

I'd agree, except I doubt there is every much contention on the _same_
file.

The reason llseek() ends up being so clear on the profiles is that it's a
very common system call under certain loads _and_ it uses a shared lock
for everything.

Also note the correctness issue (ie serialization on i_size), although
that is only an issue for SEEK_END (and maybe the lock should only be
gotten for that case. I'd love to hear what Al thinks..

		Linus