mmci: U300 "sync with blockend" broken for multi-block?

[Linus Walleij]

2011/1/14 Linus Walleij <linus.ml.walleij at gmail.com>:

> What it problematic with the reads *and* writes is that
> sometimes there us a MCI_DATABLOCKEND missing, so the
> blocks simply don't add up! This is the real bug that
> the sync code was trying to solve, and not in a very
> good way.

Here is a follow-up explanation as to why the code as it
is today actually works: the sync code has a bug. It is
intended to check whether the *last* blockend interrupt
*and* the dataend interrupt has occurred. However it
actually checks whether *any* blockend interrupt and
the datend interrupt has occured.

Thus it works, with the side effect of ACK:ing multiblock
transfers on U300 while the host->data_xfered value is lower
that what was requested - the blocks have been read
but not all are accounted for, because there are blockend
IRQs missing.

I'll follow up with a patch fixing the *real* issue, in an
atleast a little more elegant way :-/

Yours,
Linus Walleij