On Sat, 22 Apr 2023 at 12:03, Thomas Weißschuh <tho...@t-8ch.de> wrote:
>
> On 2023-04-22 02:58:07+0200, Ilya Leoshkevich wrote:
> > Make sure values are stored in memory as little-endian regardless of
> > the host endianness.
> >
> > Signed-off-by: Ilya Leoshkevich <i...@linux.ibm.com>
> > ---
> > tests/tcg/multiarch/system/memory.c | 24 ++++++++++++++++++++++++
> > 1 file changed, 24 insertions(+)
> >
> > diff --git a/tests/tcg/multiarch/system/memory.c
> > b/tests/tcg/multiarch/system/memory.c
> > index 214f7d4f54b..8ef6666b440 100644
> > --- a/tests/tcg/multiarch/system/memory.c
> > +++ b/tests/tcg/multiarch/system/memory.c
> > @@ -121,6 +121,9 @@ static void init_test_data_u16(int offset)
> > for (i = 0; i < max; i++) {
> > uint8_t low = count++, high = count++;
> > word = BYTE_SHIFT(high, 1) | BYTE_SHIFT(low, 0);
> > +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
> > + word = __builtin_bswap16(word);
> > +#endif
>
> These looks like a usecase for cpu_to_le16() and friends.
I'm not sure this test code has direct access to those, but
the general principle is right.
It is particularly odd that we carefully manually put
together a word by shifting-and-ORing together the
various bytes in it, and then separately do a byteswap.
thanks
-- PMM
