Re: [PATCH v3 1/6] iio: introduce IIO_DECLARE_BUFFER_WITH_TS macros

[David Lechner <dlechner@baylibre.com>]

On 4/26/25 6:35 AM, Jonathan Cameron wrote:
> On Fri, 25 Apr 2025 16:08:43 -0500
> David Lechner <dlechner@baylibre.com> wrote:
> 

...

>> @@ -777,6 +779,42 @@ static inline void *iio_device_get_drvdata(const struct iio_dev *indio_dev)
>>   * them safe for use with non-coherent DMA.
>>   */
>>  #define IIO_DMA_MINALIGN ARCH_DMA_MINALIGN
>> +
>> +#define __IIO_DECLARE_BUFFER_WITH_TS(type, name, count) \
>> +	static_assert(count); \
> 
> Why do we care if count is 0?  Or is intent to check if is constant?
> If the thought is we don't care either way about 0 (as rather nonsensical)
> and this will fail to compile if not constant, then perhaps a comment would
> avoid future confusion?

I would be inclined to just leave out the check. But yes, it is just checking
that count is constant and we don't expect 0.

> 
>> +	type name[ALIGN((count), sizeof(s64) / sizeof(type)) + sizeof(s64) / sizeof(type)]
>> +
>> +/**
>> + * IIO_DECLARE_BUFFER_WITH_TS() - Declare a buffer with timestamp
>> + * @type: element type of the buffer
>> + * @name: identifier name of the buffer
>> + * @count: number of elements in the buffer
>> + *
>> + * Declares a buffer that is safe to use with iio_push_to_buffer_with_ts(). In
>> + * addition to allocating enough space for @count elements of @type, it also
>> + * allocates space for a s64 timestamp at the end of the buffer and ensures
>> + * proper alignment of the timestamp.
>> + */
>> +#define IIO_DECLARE_BUFFER_WITH_TS(type, name, count) \
>> +	__IIO_DECLARE_BUFFER_WITH_TS(type, name, count) __aligned(sizeof(s64))
>> +
>> +/**
>> + * IIO_DECLARE_DMA_BUFFER_WITH_TS() - Declare a DMA-aligned buffer with timestamp
>> + * @type: element type of the buffer
>> + * @name: identifier name of the buffer
>> + * @count: number of elements in the buffer
>> + *
>> + * Same as IIO_DECLARE_BUFFER_WITH_TS(), but is uses __aligned(IIO_DMA_MINALIGN)
>> + * to ensure that the buffer doesn't share cachelines with anything that comes
>> + * before it in a struct. This should not be used for stack-allocated buffers
>> + * as stack memory cannot generally be used for DMA.
>> + */
>> +#define IIO_DECLARE_DMA_BUFFER_WITH_TS(type, name, count) \
>> +	__IIO_DECLARE_BUFFER_WITH_TS(type, name, count) __aligned(IIO_DMA_MINALIGN)
>> +
>> +static_assert(IIO_DMA_MINALIGN % sizeof(s64) == 0,
> That message isn't super helpful if seen in a compile log as we aren't reading the code here
> "IIO_DECLARE_DMA_BUFFER_WITH_TS() assumes that ...
> 
>> +	"macros above assume that IIO_DMA_MINALIGN also ensures s64 timestamp alignment");
>> +

Seems we actually have an arch (openrisc) that triggers this [1]. This arch
doesn't define ARCH_DMA_MINALIGN so it falls back to:

#define ARCH_DMA_MINALIGN __alignof__(unsigned long long)

Apparently this is only of those 32-bit arches that only does 4 byte alignment.
From the official docs [2]:

	Current OR32 implementations (OR1200) do not implement 8 byte alignment,
	but do require 4 byte alignment. Therefore the Application Binary
	Interface (chapter 16) uses 4 byte alignment for 8 byte types. Future
	extensions such as ORVDX64 may require natural alignment.

[1]: https://lore.kernel.org/linux-iio/20250425-iio-introduce-iio_declare_buffer_with_ts-v3-0-f12df1bff248@baylibre.com/T/#m91e0332673438793ff76949037ff40a34765ca30
[2]: https://openrisc.io/or1k.html


It looks like this could work (it compiles for me):

	__aligned(MAX(IIO_DMA_MINALIGN, sizeof(s64)))

If that is OK we could leave out the static_assert(), unless we think there
could be an arch with IIO_DMA_MINALIGN not a power of 2?!