The ror32 implementation (word >> shift) | (word << (32 - shift) has
undefined behaviour if shift is outside the [1, 31] range. Similarly
for the 64 bit variants. Most callers pass a compile-time
constant (naturally in that range), but there's an UBSAN report that
these may actually be called with a shift count of 0.
Instead of special-casing that, we can make them DTRT for all values
of shift while also avoiding UB. For some reason, this was already
partly done for rol32 (which was well-defined for [0, 31]). gcc 8
recognizes these patterns as rotates, so for example
__u32 rol32(__u32 word, unsigned int shift)
{
return (word << (shift & 31)) | (word >> ((-shift) & 31));
}
compiles to
0000000000000020 <rol32>:
20: 89 f8 mov %edi,%eax
22: 89 f1 mov %esi,%ecx
24: d3 c0 rol %cl,%eax
26: c3 retq
Older compilers unfortunately do not do as well, but this only affects
the small minority of users that don't pass constants.
Due to integer promotions, ro[lr]8 were already well-defined for
shifts in [0, 8], and ro[lr]16 were mostly well-defined for shifts in
[0, 16] (only mostly - u16 gets promoted to _signed_ int, so if bit 15
is set, word << 16 is undefined). For consistency, update those as
well.
Reported-by: Ido Schimmel <[email protected]>
Cc: Vadim Pasternak <[email protected]>
Signed-off-by: Rasmus Villemoes <[email protected]>
---
include/linux/bitops.h | 16 ++++++++--------
1 file changed, 8 insertions(+), 8 deletions(-)
diff --git a/include/linux/bitops.h b/include/linux/bitops.h
index 602af23b98c7..cf074bce3eb3 100644
--- a/include/linux/bitops.h
+++ b/include/linux/bitops.h
@@ -60,7 +60,7 @@ static __always_inline unsigned long hweight_long(unsigned long w)
*/
static inline __u64 rol64(__u64 word, unsigned int shift)
{
- return (word << shift) | (word >> (64 - shift));
+ return (word << (shift & 63)) | (word >> ((-shift) & 63));
}
/**
@@ -70,7 +70,7 @@ static inline __u64 rol64(__u64 word, unsigned int shift)
*/
static inline __u64 ror64(__u64 word, unsigned int shift)
{
- return (word >> shift) | (word << (64 - shift));
+ return (word >> (shift & 63)) | (word << ((-shift) & 63));
}
/**
@@ -80,7 +80,7 @@ static inline __u64 ror64(__u64 word, unsigned int shift)
*/
static inline __u32 rol32(__u32 word, unsigned int shift)
{
- return (word << shift) | (word >> ((-shift) & 31));
+ return (word << (shift & 31)) | (word >> ((-shift) & 31));
}
/**
@@ -90,7 +90,7 @@ static inline __u32 rol32(__u32 word, unsigned int shift)
*/
static inline __u32 ror32(__u32 word, unsigned int shift)
{
- return (word >> shift) | (word << (32 - shift));
+ return (word >> (shift & 31)) | (word << ((-shift) & 31));
}
/**
@@ -100,7 +100,7 @@ static inline __u32 ror32(__u32 word, unsigned int shift)
*/
static inline __u16 rol16(__u16 word, unsigned int shift)
{
- return (word << shift) | (word >> (16 - shift));
+ return (word << (shift & 15)) | (word >> ((-shift) & 15));
}
/**
@@ -110,7 +110,7 @@ static inline __u16 rol16(__u16 word, unsigned int shift)
*/
static inline __u16 ror16(__u16 word, unsigned int shift)
{
- return (word >> shift) | (word << (16 - shift));
+ return (word >> (shift & 15)) | (word << ((-shift) & 15));
}
/**
@@ -120,7 +120,7 @@ static inline __u16 ror16(__u16 word, unsigned int shift)
*/
static inline __u8 rol8(__u8 word, unsigned int shift)
{
- return (word << shift) | (word >> (8 - shift));
+ return (word << (shift & 7)) | (word >> ((-shift) & 7));
}
/**
@@ -130,7 +130,7 @@ static inline __u8 rol8(__u8 word, unsigned int shift)
*/
static inline __u8 ror8(__u8 word, unsigned int shift)
{
- return (word >> shift) | (word << (8 - shift));
+ return (word >> (shift & 7)) | (word << ((-shift) & 7));
}
/**
--
2.20.1
On Wed, Apr 10, 2019 at 11:19:06PM +0200, Rasmus Villemoes wrote:
> The ror32 implementation (word >> shift) | (word << (32 - shift) has
> undefined behaviour if shift is outside the [1, 31] range. Similarly
> for the 64 bit variants. Most callers pass a compile-time
> constant (naturally in that range), but there's an UBSAN report that
> these may actually be called with a shift count of 0.
>
> Instead of special-casing that, we can make them DTRT for all values
> of shift while also avoiding UB. For some reason, this was already
> partly done for rol32 (which was well-defined for [0, 31]). gcc 8
> recognizes these patterns as rotates, so for example
>
> __u32 rol32(__u32 word, unsigned int shift)
> {
> return (word << (shift & 31)) | (word >> ((-shift) & 31));
> }
>
> compiles to
>
> 0000000000000020 <rol32>:
> 20: 89 f8 mov %edi,%eax
> 22: 89 f1 mov %esi,%ecx
> 24: d3 c0 rol %cl,%eax
> 26: c3 retq
>
> Older compilers unfortunately do not do as well, but this only affects
> the small minority of users that don't pass constants.
>
> Due to integer promotions, ro[lr]8 were already well-defined for
> shifts in [0, 8], and ro[lr]16 were mostly well-defined for shifts in
> [0, 16] (only mostly - u16 gets promoted to _signed_ int, so if bit 15
> is set, word << 16 is undefined). For consistency, update those as
> well.
>
> Reported-by: Ido Schimmel <[email protected]>
> Cc: Vadim Pasternak <[email protected]>
> Signed-off-by: Rasmus Villemoes <[email protected]>
Tested-by: Ido Schimmel <[email protected]>
Thanks!
On Wed, Apr 10, 2019 at 11:19:06PM +0200, Rasmus Villemoes wrote:
> The ror32 implementation (word >> shift) | (word << (32 - shift) has
> undefined behaviour if shift is outside the [1, 31] range. Similarly
> for the 64 bit variants. Most callers pass a compile-time
> constant (naturally in that range), but there's an UBSAN report that
> these may actually be called with a shift count of 0.
>
> Instead of special-casing that, we can make them DTRT for all values
> of shift while also avoiding UB. For some reason, this was already
> partly done for rol32 (which was well-defined for [0, 31]). gcc 8
> recognizes these patterns as rotates, so for example
>
> __u32 rol32(__u32 word, unsigned int shift)
> {
> return (word << (shift & 31)) | (word >> ((-shift) & 31));
> }
>
> compiles to
>
> 0000000000000020 <rol32>:
> 20: 89 f8 mov %edi,%eax
> 22: 89 f1 mov %esi,%ecx
> 24: d3 c0 rol %cl,%eax
> 26: c3 retq
>
> Older compilers unfortunately do not do as well, but this only affects
> the small minority of users that don't pass constants.
>
> Due to integer promotions, ro[lr]8 were already well-defined for
> shifts in [0, 8], and ro[lr]16 were mostly well-defined for shifts in
> [0, 16] (only mostly - u16 gets promoted to _signed_ int, so if bit 15
> is set, word << 16 is undefined). For consistency, update those as
> well.
>
> Reported-by: Ido Schimmel <[email protected]>
> Cc: Vadim Pasternak <[email protected]>
> Signed-off-by: Rasmus Villemoes <[email protected]>
> ---
> include/linux/bitops.h | 16 ++++++++--------
> 1 file changed, 8 insertions(+), 8 deletions(-)
Reviewed-by: Will Deacon <[email protected]>
I guess it would be possible to roll some of this up into macros using
sizeof, but perhaps that would make things even more difficult for the
compiler.
Will