On 10.05.2021 08:36, Christoph Hellwig wrote:
> On Sat, May 08, 2021 at 12:29:15AM +0200, Heiner Kallweit wrote:
>> +
>> + if (len == 4) {
>> + put_unaligned_le32(val, buf);
>> + } else {
>> + cpu_to_le32s(&val);
>> + memcpy(buf, (u8 *)&val + skip, len);
>
> cpu_to_le32s is a horrible API that breaks endianess annotations.
>
The endian-adjusted value is used just in the next line, so I think
there's no risk of wrong usage.
But yes, this function converts to le32 w/o using __bitwise.
Therefore I understand the concern.
> Is the intent of this code to only put 16 bits in? Why not something
> like:
>
> switch (len) {
> case 4:
> put_unaligned_le32(val, buf);
> break;
> case 2:
> put_unaligned_le16(val, buf + 2);
> break;
> case 1:
> buf[3] = val;
> break;
> }
>
len can have any value 1 .. 4. Also the proposal doesn't consider
the skip value.
