Re: [Linux-fbdev-devel] lxfb driver regression

[Jens R. <JRottmann@LiPPERT-AT.de>]

Andres Salomon wrote:
> The pixclock that's specified is 17460 (from the olpc dcon table)
> [...] Prior to your patch, this was:  { 0x00004286, 56250 },
> [...] Your patch adds:  { 0x00014170, 57375 },

Yes, of course that was my first idea, too. And this was my motivation for
sending you the cut down patch without the > 25 MHz entries. But I'm almost
prepared to bet this is not the root of the evil.

As I said, this does not explain why the resolution seems to switch to 800x640,
because lx_set_clock() does not return any feedback. I somehow doubt you'll get
a white screen just by raising the dotclock by 2 % (and getting it _closer_ to
the intended value)?

And before I wrote my mail yesterday, I checked the dotclock for the 0x00004286
and 0x00014170 PLL values with a scope, and both settings produced a fine signal
with the expected frequencies.

Anyway, I hope I'm wrong, then at least we'd have a solution for this issue.

> BTW, where did the intermediate values in your table come from?

Took the original table from lxfb_ops.c, everywhere set DIV4 and divided
frequency by 4, merged it back into the original table, dropping all lines
within 300 kHz of an preexisting setting.
I wrote a small script to do this, to avoid typos when mangling the table
(assuming the original table was ok).

> The LX data book shows the DOTPLL equation (6.14.2.14), but very
> little documentation of DIV4 or the usable values below 15MHz.

LX Data Book, p. 554:
"DIV4: Divide by 4. When set, the PLL output is divided by 4 before clocking the
logic."

Sounds like a simple divider circuit _behind_ the PLL, so as long as the
original frequency is ok, the DIV4-ed should be, too.

As I said, I assumed the original table in lxfb_ops.c to contain usable values.
And all of the few settings I checked with the scope looked ok.

Jordan Crouse wrote:
> ... the original value (56.250), which in retrospect is probably
> a better choice for ths display.

So in case the increased dotclock in fact causes the white screen, it might be
cleaner to lower the _requested_ dotclock in the OLPC DCON mode timing to 56.250
MHz instead of cutting out PLL settings.

> I don't think we really need divided signals > 24Mhz.

Well, in my opinion it would be nice to have some intermediate steps in the
available dotclocks. However, I don't actually need those for the 320x240 panel,
so I don't care much.

Regards,
Jens