Re: [GIT PULL] x86/build changes for v4.17

[James Y Knight]

On Wed, Apr 4, 2018 at 12:59 PM Greg KH <gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
> Here is another horrible work around that was needed to get clang to
> stop generating invalid code, beaec533fc27 ("llist: clang: introduce
> member_address_is_nonnull()")  That one caused a lot of odd failures by
> users, I wonder what else is lurking in that same code pattern.  It's a
> hard one to debug...

I would note that this issue is an entirely different issue from the
null-pointer-deref-is-undefined-behavior optimizations, even though it may
seem superficially similar. For _this_ issue, the behavior at question is
that the compiler assumes that objects are contiguous in memory, and thus
that "&struct_pointer->member_at_nonzero_offset != NULL" is always true. I
don't really see clang ever getting a flag to stop assuming that objects
are contiguous.

Note that clang does actually emit an on-by-default warning for a situation
analogous to this:
   warning: comparison of address of 'p->sub' not equal to a null pointer is
always true [-Wtautological-pointer-compare]
   if (&p->sub != NULL) {
        ~~~^~~    ~~~~
...but unfortunately that warning is suppressed when it occurs in a
macro-expansion, so the llist_for_each_entry error was silent.

(OTOH, I _do_ expect clang to eventually gain support for a flag to treat
NULL-pointer deref as a legal operation instead of UB. I'm not really sure
it makes sense for the linux kernel, but it's a useful feature for the
compiler to provide in any case, so why not...)