Re: [PATCH 00/10] iio: prefer aligned_s64 timestamp (round 1)

[David Lechner <dlechner@baylibre.com>]

On 4/18/25 2:58 PM, David Lechner wrote:
> While reviewing the recent conversion to iio_push_to_buffers_with_ts(),
> I found it very time-consuming to check the correctness of the buffers
> passed to that function when they used an array with extra room at the
> end for a timestamp. And we still managed find a few that were wrongly
> sized or not properly aligned despite several efforts in the past to
> audit these for correctness already.
> 
> Even though these ones look to be correct, it will still be easier for
> future readers of the code if we follow the pattern of using a struct
> with the array and timestamp instead.
> 
> For example, it is much easier to see that:
> 
> struct {
> 	__be32 data[3];
> 	aligned_s64 timestamp;
> } buffer;
> 
After sending [1], I realized that some (perhaps many) of these would actually
be a better candidate for the proposed IIO_DECLARE_BUFFER_WITH_TS macro rather
that converting to the struct style as above.

Case in point: if the driver using that struct allows reading only one channel,
then the offset of the timestamp when doing iio_push_to_buffers_with_ts() would
be 8 bytes, not 16, so the struct would not always be the correct layout.

As long as the driver doesn't access the timestamp member of the struct, it
doesn't really matter, but this could be a bit misleading to anyone who might
unknowing try to use it in the future.

[1]: https://lore.kernel.org/linux-iio/20250418-iio-introduce-iio_declare_buffer_with_ts-v1-0-ee0c62a33a0f@baylibre.com/