On 02/05/2016 04:32 PM, Andy Shevchenko wrote: > On Fri, Feb 5, 2016 at 4:30 PM, Lars-Peter Clausen <lars@xxxxxxxxxx> wrote: >>>> + clk_div = DIV_ROUND_UP(clk_get_rate(spi_engine->ref_clk), >>>> + xfer->speed_hz * 2); >>> >>>> + if (clk_div > 255) >>>> + clk_div = 255; >>>> + else if (clk_div > 0) >>>> + clk_div -= 1; >>> >>> 255 is okay, 254 is not, 253- is okay. Why 254 is so special? >> >> I don't see that. The condition is > 255, so everything greater or equal >> than 256 gets mapped to 255. Everything else to x - 1, so 255 to 254, 254 to >> 253. > > Ah, you are right. > >>>> + for (i = 0; i < m; i++) >>>> + writel_relaxed(buf[i], addr); >>> >>> writesl() ? >> >> Hm, maybe. Does it really have the same semantics? > > It's a loop inside, so it has a count parameter. Otherwise it should be similar. It seems to have different semantics in terms of endianness and ordering. > >>>> +static int spi_engine_transfer_one_message(struct spi_master *master, >>>> + struct spi_message *msg) >>> >>> And you are not using transfer_one() because of..? >> >> transfer_one() does flow control in software. Execution is passed back to >> software after each transfer and it also takes care of handling the chip >> select assertion/deassertion as well as the delays. It's useful for hardware >> which does not support hardware flow control. In this case the hardware >> supports flow control including chip-select logic as well as delays. Making >> use of this generates far less context switches per message and also has >> predictable timings between transfers within a message. > > Maybe the hardware flow control implementation is needed? But wouldn't that be transfer_one_message? -- To unsubscribe from this list: send the line "unsubscribe devicetree" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
