[Intel-gfx] [RFC] [PATCH 00/38] Haswell

[Daniel Vetter]

On Wed, Mar 21, 2012 at 10:09:35PM -0300, Eugeni Dodonov wrote:
> Hi folks,
> 
> To contribute to everyones curiosity and further enrich Phoronix articles
> :), this is the initial Haswell enablement patchset for basic modesetting
> support over VGA.
> 
> This is actually the first time I am doing such enablement, and the
> combination of HSW/LPT changes proved to be quite tricky, yet also very cool
> to do. But certainly, while I tried to split the patches in the best way to
> 'tell a story', I suspect lots and lots of things could be improved. So
> bikeshedding is much welcomed! And of course, there are still a bunch of
> things missing, but those can be done in parallel now.
> 
> As for those patches, I organized them in the following way:
>  - The first part adds new hardware information and new registers which we'll
>    need to make it work (a.k.a, 'The Pledge'). This goes from patch 01 until
>    patch 21.
>  - The second part does small changes to existent functions to make them work
>    on Haswell ('The Turn'). This goes from patch 22 to patch 28.  Haswell is
>    not quite Gen7, but is shares lots of stuff with what we already have, so
>    there was no point in reinventing the wheel just now.
>  - The third part makes things work by enabling power wells, DDI and FDI
>    connections, PCH-specific stuff and so on ('The Prestige'). This is from
>    patch 29 to patch 35. Those are the scariest ones, but I couldn't split them
>    better and still maintain logical consistency among them.  Initially, those
>    were 24 small patches, but it was very hard to keep track of them, so I
>    combined them into just 6 for this patchset.
>  - And finally, there are 2 last patches ('The Prestige Debug' :)) which won't
>    be present in the final series, but are helpful at least for me to trace
>    where things go wrong. So if you run into issues and want to send me a
>    dmesg, those traces will make my life so much easier.
> 
> I didn't wanted to go much deeper with HSW for now because I would end up
> doing a massive code reordering to maintain my mental sanity, and as we
> already discussed by email earlier today Jesse is running into same problems
> as well. So perhaps we should just settle on doing some cleanup in our
> intel_display module before adding more platform-specific stuff?

Awesome work and nice patch series. To satisfy you're expectation, I've
dropped a few bikesheds on it. Now my gut says that this needs some more
work before we can merge hsw support, but there are quite a few things
that I'd like to pick up earlier (just so you don't have to carry so many
patches):

- the separate register definition patches are rather nice, so as soon as
  someone with enough coffee has cross-checked them with Bspec I could
  merge them, maybe already next week.

- some of the infrastucture work (like adding the has_pch_split feature
  work) could go in early, too. Luckily hsw seems to be less a
  frankenstein monster than vlv, which nicely reflects in much fewer pain
  points when integrating it in our code.

- for merging real hsw support (and especially the pci ids) I think we
  should wait until digital outputs also work somewhat.

Yours, Daniel
-- 
Daniel Vetter
Mail: daniel at ffwll.ch
Mobile: +41 (0)79 365 57 48