Re: Add a reset interface

[Chris Wilson <chris@chris-wilson.co.uk>]

On Mon, 31 Jan 2011 06:03:25 +1000, Dave Airlie <airlied@gmail.com> wrote:
> It seems like an interface we could use, I just wasn't sure how a regression fix
> for something in. 38 isn't a revert but requires a whole new interface
> in the core
> drm to fix. Perhaps if you can enlighten me on that I'd be happier.

We've made several mistakes along the way, which can be surmised as
forgetting to reset state upon resume. (The same bug affects the GPU reset
paths, but I don't expect to receive regression reports following a
hang...)

The bug fixes are themselves just a couple of missing lines (excluding
the many more lines actually required to access the right structures),
but the failure is systematic, and I think there may be more bugs (or at
least susceptible to the same bugs). And so it will be useful to have the
infrastructure in place to fix those bugs.

The other aspect is that reverting the commits responsible reintroduces
undesirable behaviour during normal operation; flickering during output
polling and fdi training failure (due to multiple calls). The failures upon
resume are by comparison the lesser of the evils.

But trivial to fix.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre