Hi,

Thanks for your PR.

I didn't see in kernel's iio_enable_buffers() watermark setting error is ignored, so even if the driver returns error it will still works.

	if (indio_dev->info->hwfifo_set_watermark)
		indio_dev->info->hwfifo_set_watermark(indio_dev,
			config->watermark);

Because the actual value of watermark and length will decide if otherone's value can be successfully written:

• length must >= watermark
• watermark must <= legnth
  Otherwise a Invalid argument will happen when you write the value (you can test simplyly with echo)

It's not sufficient to only write watermark once.

For example:

1. length=100, watermark=100 from previous run
2. New run want length=50
3. local_write_dev_attr(dev, "buffer/length", 50) will fail will Invalid argument since it's smaller than watermark.

I think you need to reset `watermark to 1 before setting length.

This is incorrect. If the watermark is higher than the new buffer length, the kernel will reduce it to the buffer length. See: https://github.com/torvalds/linux/blob/master/drivers/iio/industrialio-buffer.c#L667-L668

This is incorrect. If the watermark is higher than the new buffer length, the kernel will reduce it to the buffer length.

Hum... Maybe I've some memory corruption... will try it later.

I've tested creating buffer with size up & down, it's working :)

@HiFiPhile And it also works if the buffer size isn't a multiple of the watermark?

With a size of 110 and watermark of 20 it seems working. I don't have a high speed device to test if last conversion will be cut off, in low speed device since kernel buffer has a 4*length it should be ok.

But 110*4 is divisible by 20 :)
Could you try e.g. a buffer size of 22 samples? I just want to make sure it doesn't lock up.

With a length of 123 and watermark of 20 it still works :)

Great, thanks. I sent patches upstream to fix the two drivers that errored with too high watermarks.