[PATCHv2] mmci: work with only one irq

[Linus Walleij]

2010/10/11 Russell King - ARM Linux <linux at arm.linux.org.uk>:
> On Mon, Oct 11, 2010 at 05:36:41AM +0530, Rabin Vincent wrote:
>> @@ -276,6 +292,10 @@ mmci_data_irq(struct mmci_host *host, struct mmc_data *data,
>>       if (status & MCI_DATAEND) {
>>               mmci_stop_data(host);
>>
>> +             /* MCI_DATABLOCKEND not used with single irq */
>> +             if (host->singleirq && !data->error)
>> +                     host->data_xfered = data->blksz * data->blocks;
>> +
>
> I'm uneasy about this.  If we don't get datablockend interrupts, then
> when there's an error we don't know how much data was transferred, and
> with big requests, we're going to have a very long time spent retrying
> block by block.

I have sent a respin of Rabins patch (he's on vacation, so I got his
permission to fix this up) which removes this offending handling of
MCI_DATABLOCKEND, it's now also in the patch tracker as 6311/2.

On top of this I have updated my patch for out-of-order IRQ arrival
to cover also this case and renamed it
"handle broken MCI_DATABLOCKEND hardware"
it will do away with the confusing U300 #ifdef and adds a comment
block that explains what the problem really is on the Ux hardwares,
basically that under some circumstances in the U300 and under
all circumstances in the Ux500, the MCI_DATABLOCKEND flag
is totally unreliable and "hangs" at 1.

Yours,
Linus Walleij