Hi Pierre,
I've been trying to debug some card detection problems in the atmel-mci
driver. Sometimes, when I remove a card, the event doesn't seem to get
detected properly, and the MMC core thinks the card is still there.
When I re-insert the card, the MMC core thinks the card is gone.
I've tried to add a debouncing timer to avoid glitches on the card
detect pin. This helps a bit, but it does not eliminate the problem
altogether.
It seems like the problem only occurs if the card is removed very
slowly. If I increase the debouncing or detection delay, I have to
remove the card more slowly to trigger the problem.
I think the real problem is that the card detection interrupt triggers
while there is still electrical contact with the card. So when the MMC
core tries to send a SEND_STATUS command to check if it's still there,
the card will respond even though it's about to be removed. Since there
will be no more interrupts as the card is completely removed, the MMC
core will never notice that the card is gone.
When the card is reinserted, the MMC core will try to send a
SEND_STATUS command again. This time, the card won't respond because it
is in the "idle" state, and the MMC core will think the card is gone.
In order to fix this problem, I think I need a way to tell the MMC core
that the card really is gone and that there's no point trying to
communicate with it. Is there any way I can do that?
Haavard
On Thu, 21 Feb 2008 19:46:20 +0100
Haavard Skinnemoen <[email protected]> wrote:
> In order to fix this problem, I think I need a way to tell the MMC core
> that the card really is gone and that there's no point trying to
> communicate with it. Is there any way I can do that?
Never mind, I figured it out. I can simply fail the command right away
with -ENOMEDIUM if I know the card isn't present. Seems to work like a
charm.
Haavard
On Thu, 21 Feb 2008 19:46:20 +0100
Haavard Skinnemoen <[email protected]> wrote:
>
> When the card is reinserted, the MMC core will try to send a
> SEND_STATUS command again. This time, the card won't respond because it
> is in the "idle" state, and the MMC core will think the card is gone.
>
> In order to fix this problem, I think I need a way to tell the MMC core
> that the card really is gone and that there's no point trying to
> communicate with it. Is there any way I can do that?
>
As you found out, you can. But the MMC core doesn't treat it
differently.
What is really needed is that the core should make an extra pass and
see if the card has been replaced by something else (and not just check
if it is gone). It's on my todo, but I haven't had time to implement it.
Rgds
Pierre