Re: MemoryFence()

Andrew Fish <afish@...>

On Feb 5, 2021, at 10:47 AM, Laszlo Ersek <lersek@...> wrote:

On 02/05/21 19:32, Paolo Bonzini wrote:

Unaligned pointers are already undefined behavior so you can ignore how
they are dealt with at the processor level.
My question was unclearly asked, sorry. Let's say we have a UINT32 at an
address that's not a multiple of 4, but a multiple of 2. A pointer to
that UINT32 is "acceptably aligned" on x86, but not "naturally aligned".
Dereferencing the pointer is not undefined (my reading of C99 suggests
that alignment requirements are implementation-defined), but I don't
know if the atomicity guarantee holds.

Clang treats unaligned pointers as undefined behavior (UB). I’m not sure if this is due to C11? What I know is clang choses to NOT optimize away UB from alignment errors, but if you run the address sanitizer you get errors.

For fun one day I turned on the clang address sanitizer with our edk2 firmware and had it emit UD2 on faults so it did not require a runtime. It turned out our debugger stub had lots of alignment issues :( so I kind of gave up at that point :).


Andrew Fish

Another example; we may have a pointer to a packed structure, and we
might want to poke at a UINT32 field in that structure. Not through a
naked pointer-to-UINT32 of course, which would throw away the
packed-ness, but really through the pointer-to-the-whole-packed-struct.


Join to automatically receive all group messages.