[PATCH 3/6] spi: bcm2835: fill FIFO before enabling interrupts to

[Martin Sperl]

On 2015-04-07 18:28, Mark Brown wrote:
> On Tue, Apr 07, 2015 at 09:39:03AM -0600, Stephen Warren wrote:
>> On 04/07/2015 12:25 AM, Martin Sperl wrote:
>> ... then not touch any CS-related register for the entire transfer. There
>> shouldn't be a need to enable/disable IRQs during the transfer; just leave
>> them enabled the entire time, until all bytes have been transferred.
>
> We will need to make sure /CS is kept asserted between transfers in a
> message too.
The problem is here that you need to touch the CS-register when an
individual spi-transfer is finished, as only that way you can disable
the (level)interrupts, which would fire immediately again when you exit
chip-select.

So you are at the risk of it happening at that phase as well...
and during the next spi_transfer_one you need to reenable the
interrupts again.

Anyway: my latest incarnation of the driver does now implement full
cs_gpio operation with native-cs getting translated during the
new spi_register_cs phase when registering the bus using the (new)
function-pointer register_cs in spi_master.
So this issue goes away and the code becomes smaller at the same time.

This is also used to set up chip_selects to their defaults levels prior
to any activity on the SPI bus.