* Sebastian Andrzej Siewior [off-list ref] [140904 01:46]:

On 09/03/2014 08:39 PM, Tony Lindgren wrote:

quoted

quoted

good.txt and bad.txt are from the late_initcall.

 $ diff -u good.txt bad.txt

--- good.txt    2014-09-03 10:29:58.920317368 +0200
+++ bad.txt     2014-09-03 10:28:57.064313222 +0200

Hmm can you check that you have good.txt and bad.txt the
right way? I'd assume you need VAUX2 or VAUX3 enabled
not disabled for the MMC to work?

No, it was correct. If you look at the complete file you will notice
the - which removes the mmc detect/mount in the bad case and + which
adds the -110 error

...
 

With that patch it seems it is a little harder to trigger. It is
usually every other boot that fails. Here a diff between two that
worked (say good-v1 vs good-v2):

...
 

It took mmc a little longer to detect but it worked. And the content of
the three registers seem not to matter _or_ it was dumped before MMC
got active.

...
 

I didn't change a thing, I just pressed reset. As you see the content
of the TWL registers don't seem to be that important since it was the
same at the time of the dump. And the mmc didn't came back.

OK. I guess that means that it's most likely some timing related issue,
or an issue with some regulator not being ready by the time the MMC
probes. Or it's a bug somewhere that I'm not figuring out.

Looking at the twl-regulator.c, twl4030reg_enable just sets the group
bit and we're not checking any status and don't have startup-delay-us
for them.

At least phy-twl4030-usb.c has twl4030_i2c_write_u8_verify(), I wonder
if adding a read back of the register to twl-regulator.c would help?

Regards,

Tony