BBB MMC / SD detection instability with U-Boot 2014.04 (CPU 1GHz)

[Ian Lepore]

On Thu, 2014-05-22 at 20:46 +0900, SAITOU Toshihide wrote:
> In message: <CADH-AwGb36EUknNofdch1Q4Pn8GAN+Ep9SdiJ_f7Q2v9e4kW1g at mail.gmail.com>
>             Winston Smith <smith.winston.101 at gmail.com> writes:
> > On Wed, May 21, 2014 at 11:20 AM, SAITOU Toshihide <toshi at ruby.ocn.ne.jp> wrote:
> >> If abort like
> >>
> >>   musbotg0: TI AM335X USBSS v0.0.13
> >>   Fatal kernel mode data abort: 'External Non-Linefetch Abort (S)'
> >>   trapframe: 0xc0a2eb60
> > 
> > I see this with the 1Ghz uboot, it occurs about 50% of the time, see:
> > 
> > http://comments.gmane.org/gmane.os.freebsd.devel.arm/8200
> 
> Although it is an ad hoc workaround but ``usb start'' at u-boot command
> prompt (someone mentioned before) or add device_printf("!\n") before
> ``rev = USBSS_READ4(sc, USBSS_REVREG);'' in the musbotg_attach of
> am335x_usbss.c prevent this panic for me.

An 'external non-linefetch abort' on a TI chip usually means that the
clocks for a device never got turned on and you attempted to read or
write a register in that device.  If 'usb start' makes the problem go
away, that tends to confirm that thought.

The thing is, I don't understand why adding a printf to the code with no
other changes would help in any way.  I though maybe it was adding some
delay to allow the clock-start call to take effect, but the clock enable
call is after the USBSS_REVREG read, and that seems wrong.

Does it fix the problem to move the ti_prcm_clk_enable() call to be
before the USBSS_REVREG read in attach?

-- Ian