Re: [spi-devel-general] [PATCH][2.16.17-rc1-mm2] spi: Added spi master driver for Freescale MPC83xx SPI controller

[Kumar Gala <galak@kernel.crashing.org>]

On Apr 10, 2006, at 3:03 PM, Vitaly Wool wrote:

> Hi,
>
> Kumar Gala wrote:
>
>> +irqreturn_t mpc83xx_spi_irq(s32 irq, void *context_data,
>> +			    struct pt_regs * ptregs)
>> +{
>> +	struct mpc83xx_spi *mpc83xx_spi = context_data;
>> +	u32 event;
>> +	irqreturn_t ret = IRQ_NONE;
>> +
>> +	/* Get interrupt events(tx/rx) */
>> +	event = mpc83xx_spi_read_reg(&mpc83xx_spi->base->event);
>> +
>> +	/* We need handle RX first */
>> +	if (event & SPIE_NE) {
>> +		u32 rx_data = mpc83xx_spi_read_reg(&mpc83xx_spi->base->receive);
>> +
>> +		if (mpc83xx_spi->rx)
>> +			mpc83xx_spi->get_rx(rx_data, mpc83xx_spi);
>> +
>> +		ret = IRQ_HANDLED;
>> +	}
>> +
>> +	if ((event & SPIE_NF) == 0)
>> +		/* spin until TX is done */
>> +		while (((event =
>> +			 mpc83xx_spi_read_reg(&mpc83xx_spi->base->event)) &
>> +						SPIE_NF) == 0)
>> +			 ;
>>
> This is a potentially endless loop so two questions here.
> First, did you do any measurements on how long it can loop here?
> The, what if a bus error occurs and this bit is never set?

In my measurements the if is never true, so we never spin.

I wasn't particular happy about the spinning for ever, wasn't sure  
what would be better.  I need to ensure we've gotten both a TX & RX  
event before transmitting the next character.  I'm open to  
suggestions on how to do this better.

>> +
>> +	mpc83xx_spi->count -= 1;
>> +	if (mpc83xx_spi->count) {
>> +		if (mpc83xx_spi->tx) {
>> +			u32 word = mpc83xx_spi->get_tx(mpc83xx_spi);
>> +			mpc83xx_spi_write_reg(&mpc83xx_spi->base->transmit,
>> +					      word);
>> +		}
>> +	} else {
>> +		complete(&mpc83xx_spi->done);
>> +	}
>>
> So, if it's not SPIE_NF, it's not marked as HANDLED?

I wasn't sure what the best thing here was.  In truth we will only  
get an interrupt if SPIE_NF is set so it should always be handled.