> On Tue, 14 Apr 1998 16:18:46 -0700 (PDT),
> Linus Torvalds <torvalds@transmeta.com> wrote:
> >No. The lock is not protecting any filesystem, it is _only_ protecting the
> >actual IO request. It's quite ok to drop the lock, although I've told
> >people who are worried about latency issues to no worry about those yet,
> >as we have the bigger picture of correctness that is the first priority.
> >
> >You can drop the lock whenever you don't access any io-request lists, and
> >when you aren't mucking around with any data structures (and aren't
> >caching any data structure info).
>
> Colour me confused :). If the lock can be dropped when we don't access
> any io-request lists, why is it held on entry to dev->request_fn? If
> it is safe to drop the lock during loopback, it should be safe to drop
> it before any request_fn and reacquire only when needed.
We could drop it before calling request_fn(), but 99% of all uses woul
dhave to immediately get it again, because the first thing the
request_fn() is doing is looking at the request list.
So the prudent thing is to _not_ drop it, but allow special cases like
loopback to drop it when they need to.
Linus
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu