Re: [RFC][PATCH 0/5] Optimize percpu-rwsem

[Linus Torvalds]

On Tue, May 26, 2015 at 11:42 AM, Davidlohr Bueso <dave@xxxxxxxxxxxx> wrote:
>
> Instead of dropping percpu-rwsem, I was thinking we could instead look
> for opportunities to convert new users, for instance shinkers, where the
> write lock is also taken just for register and unregister purposes,
> similar to uprobes.

So if there really are useful use cases for this, I don't object to
the patch. It seems to just improve on a currently very low-usage
locking primitive.

And it's not like I conceptually mind the notion of a percpu rwsem, I
just hate seeing specialty locking that isn't really worth it.

Because as it is, with the current single use, I don't think it's even
worth improving on.

I _would_ ask that people who are looking at this also look at our
"lglock" thing. It's pretty much *exactly* the same thing, except for
spinlocks, and that one too has exactly two users (the documentation
states that the only user is stop_machine, but in fact file locking
does too).

Because that is another example of a complete failure of a locking
primitive that was just too specialized to be worth it.

                       Linus
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/