[Intel-gfx] limiting modes on 4k monitors on Haswell ULT

[Ville Syrjälä]

On Thu, Oct 02, 2014 at 08:56:54PM +1000, Dave Airlie wrote:
> On 2 October 2014 18:40, Ville Syrjälä <ville.syrjala at linux.intel.com> wrote:
> > On Thu, Oct 02, 2014 at 10:00:50AM +0200, Daniel Vetter wrote:
> >> On Thu, Oct 02, 2014 at 01:30:48PM +1000, Dave Airlie wrote:
> >> > Hey guys,
> >> >
> >> > so I have a haswell ULT laptop (lenovo t440s), and got access to a
> >> > Samsung single panel 4k monitor (no MST).
> >> >
> >> > Now we detect the monitor fine, but unfortunately the ULT hsw can't
> >> > run it over DP, as it has clock limits in place. However we still
> >> > offer the 60hz mode and we pick it by default, ensuing black screens.
> >> >
> >> > I've tested the same monitor with a haswell in a t540p and it works
> >> > fine due to having the higher limits in place.
> >>
> >> We have the IS_HSW_ULX check in intel_dp_max_link_bw, and that's used by
> >> both compute_config and mode_valid. So it should work, and from a quick
> >> look I don't see any bugs. But obviously something is busted.
> >>
> >> Can you please printk-augment intel_dp_mode_valid and check what's going
> >> wrong there?
> >
> > Or maybe this is about cdclk instead of link clock? I believe the max
> > cdclk ULT can support is 450MHz. I'm not sure if the default would be
> > 450MHz or 337.5MHz. Also we still don't have support for changing that
> > during runtime even though I regularly try to trick people into
> > implementing it. What's the pixel clock of the mode that fails?
> 
> yup it appears to max cdclk, which is 450Mhz on the ULT and 540Mhz on
> normal hsw.
> 
> Modeline "3840x2160R"  533.00  3840 3888 3920 4000  2160 2163 2168
> 2222 +hsync -vsync
> 
> cvt gives me that for reduced blank, so around that I guess, I can get
> the exact mode line from logs tomorrow if required.

Just FYI I went ahead and implemented something for dynamically
changing the cdclk frequency on HSW/BDW. But I only have a HSW-ULT
on me currently so I couldn't actually test it.

While doing that I added cdclk extraction support to all the
platforms for which I could find sane docs. For the rest I went with
a best guess, which means after my patches we can rely on
dev_priv->max_cdclk_freq on all platforms. That should make it
trivial to add some .mode_valid() checks based on the max cdclk,
but I didn't actually implemnt such checks yet.

Here's the code (includes a whole pile of cdclk related stuff)
for the brave:
git://gitorious.org/vsyrjala/linux.git cdclk_3

-- 
Ville Syrjälä
Intel OTC