Problems after recent changes to meson-gx-mmc driver

[Jerome Brunet]

On Sun, 2017-09-10 at 18:20 +0200, Heiner Kallweit wrote:
> Am 10.09.2017 um 17:08 schrieb Jerome Brunet:
> > On Sat, 2017-09-09 at 22:20 +0200, Heiner Kallweit wrote:
> > > I checked further and setting the tx clock phase back to 0 fixes the issue
> > > for
> > > me.
> > > ("mmc: meson-gx: change default tx phase" changes it from 0 to 270.)
> > > But as you write 0 seems to break certain other systems.
> > 
> > That was my second guess ...
> > 
> > As I mentioned in the commit message, 270 is working fine for the setups I
> > have
> > tested but I always wondered if that would be the case for every possible
> > setups/boards/modes.
> > 
> > Would you mind testing 90 and 180 as well with your setup ? I'll make
> > another
> > pass on the different setups I have access to. Please stick to hs200 and
> > drop
> > hs400 for this test. I'm still unsure if doubling the clock after doing the
> > tuning may affect the phase tuning ... lets keep that out of the way for
> > now.
> > 
> 
> I tested the other tx clock settings with HS200/200MHz.
> 
> 0: No errors
> 90: 6 CRC errors, otherwise system works normal.
> 180: Lots of CRC errors, but system still works.
> 270: So many errors that root file system gets corrupted and is mounted r/o.
> 
> Seems like we won't find a tx clock phase working on all systems.
> So maybe the tuning needs to be extended to check all tx / rx clock
> phase combinations.

Well, I kind of had this in mind when writing the new tuning function. Tuning
the Tx phase should be fairly simple, a call like:

meson_mmc_clk_phase_tuning(mmc, opcode, host->tx_clk);

added in meson_mmc_execute_tuning() should do the trick.

That being said If you are getting file system errors then tuning must have
succeed first, even if the tx phase is not good for you.
Not sure if cycling on the Tx phase in the tuning function will help then.
It will select the center of the working window, so maybe ... It's worth trying

Another approach, if this difference is due to the PCB layout, line routing,
etc, maybe the Tx phase could be provided as a DT param ... maybe

> 
> IIRC I went with a fixed tx clock phase because other combinations of
> tx / rx clock phase selected by an experimental tuning algorithm
> worked fine when tuning but produced errors later.