Re: [PATCH v2 2/7] x86: use long long for 64-bit atomic ops

From: Dmitry Vyukov
Date: Mon May 29 2017 - 10:45:21 EST


On Sun, May 28, 2017 at 11:34 AM, <hpa@xxxxxxxxx> wrote:
> On May 28, 2017 2:29:32 AM PDT, Dmitry Vyukov <dvyukov@xxxxxxxxxx> wrote:
>>On Sun, May 28, 2017 at 1:02 AM, <hpa@xxxxxxxxx> wrote:
>>> On May 26, 2017 12:09:04 PM PDT, Dmitry Vyukov <dvyukov@xxxxxxxxxx>
>>wrote:
>>>>Some 64-bit atomic operations use 'long long' as operand/return type
>>>>(e.g. asm-generic/atomic64.h, arch/x86/include/asm/atomic64_32.h);
>>>>while others use 'long' (e.g. arch/x86/include/asm/atomic64_64.h).
>>>>This makes it impossible to write portable code.
>>>>For example, there is no format specifier that prints result of
>>>>atomic64_read() without warnings. atomic64_try_cmpxchg() is almost
>>>>impossible to use in portable fashion because it requires either
>>>>'long *' or 'long long *' as argument depending on arch.
>>>>
>>>>Switch arch/x86/include/asm/atomic64_64.h to 'long long'.
>>>>
>>>>Signed-off-by: Dmitry Vyukov <dvyukov@xxxxxxxxxx>
>>>>Cc: Mark Rutland <mark.rutland@xxxxxxx>
>>>>Cc: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
>>>>Cc: Will Deacon <will.deacon@xxxxxxx>
>>>>Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
>>>>Cc: Andrey Ryabinin <aryabinin@xxxxxxxxxxxxx>
>>>>Cc: Ingo Molnar <mingo@xxxxxxxxxx>
>>>>Cc: kasan-dev@xxxxxxxxxxxxxxxx
>>>>Cc: linux-mm@xxxxxxxxx
>>>>Cc: linux-kernel@xxxxxxxxxxxxxxx
>>>>Cc: x86@xxxxxxxxxx
>>>>
>>>>---
>>>>Changes since v1:
>>>> - reverted stray s/long/long long/ replace in comment
>>>> - added arch/s390 changes to fix build errors/warnings
>>>>---

[snip]

>>> NAK - this is what u64/s64 is for.
>>
>>
>>Hi,
>>
>>Patch 3 adds atomic-instrumented.h which now contains:
>>
>>+static __always_inline long long atomic64_read(const atomic64_t *v)
>>+{
>>+ return arch_atomic64_read(v);
>>+}
>>
>>without this patch that will become
>>
>>+static __always_inline s64 atomic64_read(const atomic64_t *v)
>>
>>Right?
>
> Yes.


I see that s64 is not the same as long on x86_64 (long long vs long),
so it's still not possible to e.g. portably print a result of
atomic_read(). But it's a separate issue that we don't need to solve
now.

Also all wrappers like:

void atomic64_set(atomic64_t *v, s64 i)
{
arch_atomic64_set(v, i);
}

lead to type conversions, but at least my compiler does not bark on it.

The only remaining problem is with atomic64_try_cmpxchg, which is
simply not possible to use now (not possible to declare the *old
type).
I will need something along the following lines to fix it (then
callers can use s64* for old).
Sounds good?


--- a/arch/x86/include/asm/atomic64_64.h
+++ b/arch/x86/include/asm/atomic64_64.h
@@ -177,7 +177,7 @@ static inline long
arch_atomic64_cmpxchg(atomic64_t *v, long old, long new)
}

#define arch_atomic64_try_cmpxchg arch_atomic64_try_cmpxchg
-static __always_inline bool arch_atomic64_try_cmpxchg(atomic64_t *v,
long *old, long new)
+static __always_inline bool arch_atomic64_try_cmpxchg(atomic64_t *v,
s64 *old, long new)
{
return try_cmpxchg(&v->counter, old, new);
}
@@ -198,7 +198,7 @@ static inline long arch_atomic64_xchg(atomic64_t
*v, long new)
*/
static inline bool arch_atomic64_add_unless(atomic64_t *v, long a, long u)
{
- long c = arch_atomic64_read(v);
+ s64 c = arch_atomic64_read(v);
do {
if (unlikely(c == u))
return false;
@@ -217,7 +217,7 @@ static inline bool
arch_atomic64_add_unless(atomic64_t *v, long a, long u)
*/
static inline long arch_atomic64_dec_if_positive(atomic64_t *v)
{
- long dec, c = arch_atomic64_read(v);
+ s64 dec, c = arch_atomic64_read(v);
do {
dec = c - 1;
if (unlikely(dec < 0))
@@ -236,7 +236,7 @@ static inline void arch_atomic64_and(long i, atomic64_t *v)

static inline long arch_atomic64_fetch_and(long i, atomic64_t *v)
{
- long val = arch_atomic64_read(v);
+ s64 val = arch_atomic64_read(v);

do {
} while (!arch_atomic64_try_cmpxchg(v, &val, val & i));
@@ -253,7 +253,7 @@ static inline void arch_atomic64_or(long i, atomic64_t *v)

static inline long arch_atomic64_fetch_or(long i, atomic64_t *v)
{
- long val = arch_atomic64_read(v);
+ s64 val = arch_atomic64_read(v);

do {
} while (!arch_atomic64_try_cmpxchg(v, &val, val | i));
@@ -270,7 +270,7 @@ static inline void arch_atomic64_xor(long i, atomic64_t *v)

static inline long arch_atomic64_fetch_xor(long i, atomic64_t *v)
{
- long val = arch_atomic64_read(v);
+ s64 val = arch_atomic64_read(v);

do {
} while (!arch_atomic64_try_cmpxchg(v, &val, val ^ i));
diff --git a/arch/x86/include/asm/cmpxchg.h b/arch/x86/include/asm/cmpxchg.h
index e8cf95908fe5..9e2faa85eb02 100644
--- a/arch/x86/include/asm/cmpxchg.h
+++ b/arch/x86/include/asm/cmpxchg.h
@@ -157,7 +157,7 @@ extern void __add_wrong_size(void)
#define __raw_try_cmpxchg(_ptr, _pold, _new, size, lock) \
({ \
bool success; \
- __typeof__(_ptr) _old = (_pold); \
+ __typeof__(_ptr) _old = (__typeof__(_ptr))(_pold); \
__typeof__(*(_ptr)) __old = *_old; \
__typeof__(*(_ptr)) __new = (_new); \
switch (size) { \