On Mon, Apr 24, 2017 at 12:55:21PM +0200, Geert Uytterhoeven wrote:
> On Fri, Apr 14, 2017 at 7:39 AM, Jiada Wang <jiada_wang@mentor.com> wrote:

> > Our use case is to use spidev as an interface to communicate with external
> > SPI master devices.
> > meanwhile the SPI bus controller can also act as master device to send data
> > to other
> > SPI slave devices on the board.

> That sounds a bit hackish to me. SPI was never meant to be a multi-master bus.
> While it can be done, you will need external synchronization (signals) to
> avoid conflicts between the SPI masters.

> > I found in your implementation, SPI bus controller is limited to either work
> > in master mode or
> > slave mode, is there any reasoning to not configure SPI mode based on SPI
> > devices use case?

> If you really need both master and slave support, you can use 2 subnodes
> in DT, the first representing the master, the second the slave.

> Mark, what's your opinion about this?

That sounds like a mess...   we *could* put the slave flag on the device
rather than the controller I guess but there's also going to need to be
something representing whatever avoids collisions on the bus somewhere.