MMIO and gcc re-ordering issue

[Paul Mackerras]

Nick Piggin writes:

> > I just wish we had even one actual example of things going wrong with
> > the current rules we have on powerpc to motivate changing to this
> > model.
> 
> ~/usr/src/linux-2.6> git grep test_and_set_bit drivers/ | wc -l
> 506
> How sure are you that none of those forms part of a cobbled-together
> locking scheme that hopes to constrain some IO access?

My comment was precisely about the fact that this sort of argument is
actually FUD.  I want one example that is demonstrably wrong, not just
a "how sure are you".

> But surely you have to audit the drivers anyway to ensure they are OK
> with the current powerpc scheme. In which case, once you have audited
> them and know they are safe, you can easily convert them to the even
> _faster_ __readl/__writel, and just add the appropriate barriers.

The trouble is that as code gets maintained, using __writel + explicit
barriers is more fragile because some people will change the code, or
add new code, without understanding the barriers.  So whenever a
driver gets converted to using __writel + barriers, we will end up
having to watch every change that goes into it forever.  Whereas with
the current scheme there's a much smaller set of gotchas to watch out
for, and the gotchas are things that already raise red flags, such as
open-coded locking and any sort of "clever" lockless scheme.

Paul.