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

[Andrey Ryabinin]

On 11/01/2018 07:32 PM, Peter Zijlstra wrote:
> On Thu, Nov 01, 2018 at 03:22:15PM +0000, Trond Myklebust wrote:
>> On Thu, 2018-11-01 at 15:59 +0100, Peter Zijlstra wrote:
>>> On Thu, Nov 01, 2018 at 01:18:46PM +0000, Mark Rutland 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 :-)
> 
> 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 warns about signed overflows despite -fno-strict-overflow if gcc version is < 8.
I have learned recently that UBSAN in GCC 8 ignores signed overflows if -fno-strict-overflow of fwrapv is used.

$ cat signed_overflow.c 
#include <stdio.h>

__attribute__((noinline))
int foo(int a, int b)
{
        return a+b;
}

int main(void)
{
        int a = 0x7fffffff;
        int b = 2;
        printf("%d\n", foo(a,b));
        return 0;
}

$ gcc-8.2.0 -fsanitize=signed-integer-overflow signed_overflow.c && ./a.out 
signed_overflow.c:6:10: runtime error: signed integer overflow: 2147483647 + 2 cannot be represented in type 'int'
-2147483647
$ gcc-8.2.0 -fno-strict-overflow -fsanitize=signed-integer-overflow signed_overflow.c && ./a.out 
-2147483647
$ gcc-7.3.0 -fsanitize=signed-integer-overflow signed_overflow.c && ./a.out
signed_overflow.c:6:10: runtime error: signed integer overflow: 2147483647 + 2 cannot be represented in type 'int'
-2147483647
$ gcc-7.3.0 -fno-strict-overflow -fsanitize=signed-integer-overflow signed_overflow.c && ./a.out
signed_overflow.c:6:10: runtime error: signed integer overflow: 2147483647 + 2 cannot be represented in type 'int'
-2147483647


clang behaves the same way as GCC 8:
$ clang -fsanitize=signed-integer-overflow signed_overflow.c && ./a.out 
signed_overflow.c:6:10: runtime error: signed integer overflow: 2147483647 + 2 cannot be represented in type 'int'
-2147483647
$ clang -fno-strict-overflow -fsanitize=signed-integer-overflow signed_overflow.c && ./a.out 
-2147483647


We can always just drop -fsanitize=signed-integer-overflow if it considered too noisy.
Although it did catch some real bugs.