On 5/25/20 2:57 PM, Mark Brown wrote: > On Mon, May 25, 2020 at 02:41:31PM +0200, Marc Kleine-Budde wrote: >> On 5/25/20 1:31 PM, Mark Brown wrote: > >>> This isn't something that every individual driver should be doing, such >>> rewriting should happen in the core so that everything sees the benefit. > >> The core could merge several half duplex transfers (until there's as cs_change) >> into a single full duplex transfer. > > Yes, that is what I am suggesting. Where in the SPI stack do you see such a "merge" function? One point to clarify is when and where to allocate and free the memory for the contiguous full duplex buffers. >> I think it's not easy to detect and reliable to split a full duplex transfer >> into half duplex ones. How can you tell, if the controller is supposed to tx 0x0 >> or actually receive. > > I don't understand how that could possibly work or why it would make > sense? ACK, I was just thinking loud about options. >> I think spi_write_then_read() can be extended to generate one full duplex >> transfer instead on two half duplex ones it does a memcpy() anyways. > > This has the same problem as doing it in any other driver code - it > causes a needless incompatibility with three wire and single duplex > devices. What about the note "portable code should never use this for more than 32 bytes" in spi_write_then_read()? The CAN driver in question may read more than 32 bytes of data. >> To get a feeling for the use cases, this is what I do in the regmap read >> function of a (not yet mainlined) CAN SPI driver. > > Like I say it's probably better if code like this gets pushed into the > SPI core where we've got more information about what the controller can > do and there's more win from doing the tuning since more devices and > systems can take advantage of it. ACK Marc -- Pengutronix e.K. | Marc Kleine-Budde | Embedded Linux | https://www.pengutronix.de | Vertretung West/Dortmund | Phone: +49-231-2826-924 | Amtsgericht Hildesheim, HRA 2686 | Fax: +49-5121-206917-5555 |