Hi!
[ Upstream commit c34f674c8875235725c3ef86147a627f165d23b4 ]
ksz_read64() currently does some dubious byte-swapping on the two halves of a 64-bit register, and then only returns the high bits. Replace this with a straightforward expression.
The code indeed is very strange, but there are just 2 users, and they will now receive byteswapped values, right? If it worked before, it will be broken.
The old code swaps the bytes within each 32-bit word, attempts to concatenate them into a 64-bit word, then swaps the bytes within the 64-bit word. There is no need for byte-swapping, only (on little- endian platforms) a word-swap, which is what the new code does.
Did this get enough testing for -stable?
Yes, I actually developed and tested all the ksz8795 changes in 5.10 before forward-porting to mainline.
Is hw little endian or high endian or...?
The hardware is big-endian and regmap handles any necessary byte-swapping for values up to 32 bits.
Note that ksz_write64() still contains the strange code, at least in 5.10.
It's unnecessarily complex, but it does work.
Thanks for the explanations and sorry for the noise. Indeed ksz_write64() is quite obfuscated, but I can't see a problem.
Best regards, Pavel