On 26 January 2015 at 18:45, Doug Anderson [off-list ref] wrote:

Ulf,

On Mon, Jan 26, 2015 at 7:15 AM, Ulf Hansson [off-list ref] wrote:

quoted

On 26 January 2015 at 12:19, Addy Ke [off-list ref] wrote:

quoted

We need to take the card pointer in execute_tuning() for mmc_send_status(),

mmc_send_status() is an mmc core function, not intended for host's to call.

quoted

but mmc->card is NULL in tuning state. So we need change the first parameter
of execute_tuning() to card pointer(struct mmc_card * card).

So, why do we need this?

I asked Addy to post upstream against mmc_send_tuning(), but I guess
he didn't (he posted against Alex's NAKed patch instead).

...when I talked to him about it, Addy was asserting that when tuning
fails it is important (at least on dw_mmc on rk3288) that we wait for
the card to stop being busy and that the way to detect was using
mmc_send_status().

So, could that be due to the internal logic of the error handling in
dw_mmc driver? Or you think this is a generic issue?

According to the specifications (eMMC and SD) both states that the
tuning command has an R1 response. So, there shouldn't be any busy
signalling involved - at least according to spec.

That would mean that against upstream you'd need to change
mmc_send_tuning() to take in the card as well (or move the "host->card
= card" assignment to before UHS init, which seems less desirable?)

What do you think about that?  Is there a better solution?

Why do we need to change mmc_send_tuning()? I thought the issue was
that mmc_send_status(), which currently takes "card" as a parameter.

Kind regards
Uffe