On Tue, May 04, 2021 at 02:30:50PM -0700, Eric Biggers wrote:

quoted

So maybe the memcpy to a local copy is the better way to go, and
hopefully the C compiler will optimize away the local copy on
architectures where it is safe to do so.  And in the unlikely case
that it is a performance bottleneck, we could add a -DUBSAN when
configure --enable-ubsan is in force, which switches in the memcpy
when only when ubsan is enabled.

These days the memcpy() approach does get optimized properly.  armv6 and armv7
with gcc used to be a notable exception, but it got fixed in gcc 6
(https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67366).

Just to be clear (looking at the latest patches on the list which are copying
whole structs), by "the memcpy() approach does get optimized properly", I meant
that it gets optimized properly in implementations of get_unaligned_le16(),
get_unaligned_le32(), put_unaligned_le32(), etc., where a single word (or less
than a word) is loaded or stored.  I don't know how reliably the compilers will
optimize out the copy if you memcpy() a whole struct instead of a single word.

Even if they don't optimize it out, I don't expect that it would be a
performance problem in this context, so it's probably still fine to solve the
problem.  But I just wanted to clarify what I meant here.

- Eric