[RFC PATCH] lib: Introduce generic __cmpxchg_u64() and use it where needed

[Dmitry Vyukov]

On Thu, Nov 1, 2018 at 6:18 PM, Peter Zijlstra <peterz at infradead.org> wrote:
>> > > > > > My one question (and the reason why I went with cmpxchg() in the
>> > > > > > first place) would be about the overflow behaviour for
>> > > > > > atomic_fetch_inc() and friends. I believe those functions should
>> > > > > > be OK on x86, so that when we overflow the counter, it behaves
>> > > > > > like an unsigned value and wraps back around.  Is that the case
>> > > > > > for all architectures?
>> > > > > >
>> > > > > > i.e. are atomic_t/atomic64_t always guaranteed to behave like
>> > > > > > u32/u64 on increment?
>> > > > > >
>> > > > > > I could not find any documentation that explicitly stated that
>> > > > > > they should.
>> > > > >
>> > > > > Peter, Will, I understand that the atomic_t/atomic64_t ops are
>> > > > > required to wrap per 2's-complement. IIUC the refcount code relies
>> > > > > on this.
>> > > > >
>> > > > > Can you confirm?
>> > > >
>> > > > There is quite a bit of core code that hard assumes 2s-complement.
>> > > > Not only for atomics but for any signed integer type. Also see the
>> > > > kernel using -fno-strict-overflow which implies -fwrapv, which
>> > > > defines signed overflow to behave like 2s-complement (and rids us of
>> > > > that particular UB).
>> > >
>> > > Fair enough, but there have also been bugfixes to explicitly fix unsafe
>> > > C standards assumptions for signed integers. See, for instance commit
>> > > 5a581b367b5d "jiffies: Avoid undefined behavior from signed overflow"
>> > > from Paul McKenney.
>> >
>> > Yes, I feel Paul has been to too many C/C++ committee meetings and got
>> > properly paranoid. Which isn't always a bad thing :-)
>>
>> Even the C standard defines 2s complement for atomics.
>
> Ooh good to know.
>
>> Just not for
>> normal arithmetic, where yes, signed overflow is UB.  And yes, I do
>> know about -fwrapv, but I would like to avoid at least some copy-pasta
>> UB from my kernel code to who knows what user-mode environment.  :-/
>>
>> At least where it is reasonably easy to do so.
>
> Fair enough I suppose; I just always make sure to include the same
> -fknobs for the userspace thing when I lift code.
>
>> And there is a push to define C++ signed arithmetic as 2s complement,
>> but there are still 1s complement systems with C compilers.  Just not
>> C++ compilers.  Legacy...
>
> *groan*; how about those ancient hardwares keep using ancient compilers
> and we all move on to the 70s :-)
>
>> > But for us using -fno-strict-overflow which actually defines signed
>> > overflow, I myself am really not worried. I'm also not sure if KASAN has
>> > been taught about this, or if it will still (incorrectly) warn about UB
>> > for signed types.
>>
>> UBSAN gave me a signed-overflow warning a few days ago.  Which I have
>> fixed, even though 2s complement did the right thing.  I am also taking
>> advantage of the change to use better naming.
>
> Oh too many *SANs I suppose; and yes, if you can make the code better,
> why not.

If there is a warning that we don't want to see at all, then we can
disable it. It supposed to be a useful tool, rather than a thing in
itself that lives own life. We already I think removed 1 particularly
noisy warning and made another optional via a config.
But the thing with overflows is that, even if it's defined, it's not
necessary the intended behavior. For example, take allocation size
calculation done via unsigned size_t. If it overflows it does not help
if C defines result or not, it still gives a user controlled write
primitive. We've seen similar cases with timeout/deadline calculation
in kernel, we really don't want it to just wrap modulo-2, right. Some
user-space projects even test with unsigned overflow warnings or
implicit truncation warnings, which are formally legal, but frequently
bugs.