On Fri, 19 Jun 2009 16:31:41 -0400
Bob Copeland <me@bobcopeland.com> wrote:

> On Fri, Jun 19, 2009 at 3:24 PM, Pierre Ossman<pierre@ossman.eu> wrote:
> > I'm afraid I don't understand the problem. If you power up the card
> > before the MMC system has started, it should still properly find the
> > card and activate the wl12xx driver once things are in sync.
> 
> Okay a little more background:
> 
> This is with the out-of-tree Android msm_sdcc host driver.  There are
> various implicit assumptions in module ordering.  For example, one of
> the platform callbacks sets a "virtual card select" flag, and then
> tries to notify the MMC host via another callback to do
> mmc_detect_change().  But that code runs before the MMC driver is
> loaded so the callback isn't yet registered and no one gets the
> notification.
> 

This does not sound very Linux:y. Module loading order should not foul
up things. If it does, then there is some missing dependency somewhere.

> > Do you want me to merge it?
> 
> It can go via the wireless tree if that's ok with you, since the
> SPI driver already lives there and this all needs to be rebased
> on Kalle Valo's latest changes anyway.
> 
> It'd be nice to get msm_sdcc upstream though.
> 

Yeah, sorry. Of course it should go through wireless. I was just in
merge-mode from all the MMC patches thrown my way. :)


I noticed one more thing though. You call wl12xx_irq() on interrupts,
and that's all you do. So I have to assume that calls back to
wl12xx_sdio_read(). That in turn calls sdio_claim_host(), which will
cause the system to deadlock as that lock is not recursive (yeah I
know, patches welcome).

Has this code actually been tested? It doesn't seem like it can work to
me.

Rgds
-- 
     -- Pierre Ossman

  WARNING: This correspondence is being monitored by the
  Swedish government. Make sure your server uses encryption
  for SMTP traffic and consider using PGP for end-to-end
  encryption.