> From: Jack Bond-Preston [mailto:[email protected]]
> Sent: Monday, 16 November 2026 06.09
> 
> On 13/11/2026 18:07, Morten Brørup wrote:
> >> If the function rte_combine64ms1b() returns the maximum possible
> >> uint64_t
> >> value, it will result in an overflow.
> >>
> >> To avoid this, add a check.
> >>
> >> Found by Linux Verification Center (linuxtesting.org) with SVACE.
> >>
> >> Fixes: 7253e3d2a5 ("bitops: move bit count functions")
> >> Cc: [email protected]
> >>
> >> Signed-off-by: Alexey Simakov <[email protected]>
> >> Signed-off-by: Mark Bulatov <[email protected]>
> >> ---
> >>   lib/eal/include/rte_bitops.h | 3 +++
> >>   1 file changed, 3 insertions(+)
> >>
> >> diff --git a/lib/eal/include/rte_bitops.h
> >> b/lib/eal/include/rte_bitops.h
> >> index 45255b68aa..d2ed0c3c61 100644
> >> --- a/lib/eal/include/rte_bitops.h
> >> +++ b/lib/eal/include/rte_bitops.h
> >> @@ -1307,6 +1307,9 @@ rte_align64pow2(uint64_t v)
> >>    v--;
> >>    v = rte_combine64ms1b(v);
> >>
> >> +  if (v == UINT64_MAX)
> >> +          v--;
> >> +
> >>    return v + 1;
> >>   }
> >
> > NAK.
> >
> > The overflow bug is present, also in rte_align32pow2().
> >
> > But the fix is wrong; it does not return a power-of-two value.
> >
> > There is no solution to fix this overflow.
> > The only possible fix is to specify in the function documentation
> that the parameter must not exceed 1 << 63 (0x8000000000000000).
> >
> > -Morten
> 
> You are right Morten, sorry about that. No point avoiding the overflow
> if the return value doesn't even fulfil the purpose of the function
> itself. Also, I realised the provided fix would change the behaviour
> for
> the input value of 0 - currently this returns 0, but with this patch it
> would return UINT64_MAX.
> 
> I guess the change should look something like this:
> -----
> 
> [PATCH] bitops: document input value range for rte_align[32,64]pow2
> 
> rte_align32pow2 and rte_align62pow2 will overflow and return incorrect
> values for input values exceeding 1<<31 and 1<<63, respectively. There
> is no way to return a valid result here: the next power of two above
> these values is unrepresentable with 32/64 bit unsigned integers.
> 
> Adjust the documentation for these functions to clarify that input
> values above these limits will return invalid results, and thus should
> not be provided to these functions. Additionally, document an existing
> edge case of the implementation: when the input is 0, it returns 0 (not
> a power of 2).
> ---
>   lib/eal/include/rte_bitops.h | 22 ++++++++++++++++++----
>   1 file changed, 18 insertions(+), 4 deletions(-)
> 
> diff --git a/lib/eal/include/rte_bitops.h
> b/lib/eal/include/rte_bitops.h
> index 45255b68aa..59dcf1e028 100644
> --- a/lib/eal/include/rte_bitops.h
> +++ b/lib/eal/include/rte_bitops.h
> @@ -1261,10 +1261,17 @@ rte_is_power_of_2(uint32_t n)
>    * Aligns input parameter to the next power of 2
>    *
>    * @param x
> - *   The integer value to align
> + *   The integer value to align. Must be less than or equal to 1 << 31
> + *   (0x80000000).
>    *
>    * @return
> - *   The smallest power of 2 which is greater than or equal to @c x.
> + *   The smallest power of 2 which is greater than or equal to @c x,
> or
> 0 if
> + *   @c v is 0.
> + *
> + * @note If the input parameter exceeds 1 << 31 (0x80000000), then the
> expected
> + *       result is unrepresentable and the return value is invalid.
> + * @note For the input value 0, this function returns 0, which is not
> a
> power of
> + *       2.
>    */
>   static inline uint32_t
>   rte_align32pow2(uint32_t x)
> @@ -1296,10 +1303,17 @@ rte_align32prevpow2(uint32_t x)
>    * Aligns 64b input parameter to the next power of 2
>    *
>    * @param v
> - *   The 64b value to align
> + *   The 64b value to align. Must be less than or equal to 1 << 63
> + *   (0x8000000000000000).
>    *
>    * @return
> - *   The smallest power of 2 which is greater than or equal to @c v.
> + *   The smallest power of 2 which is greater than or equal to @c v,
> or
> 0 if
> + *   @c v is 0.
> + *
> + * @note If the input parameter exceeds 1 << 63 (0x8000000000000000),
> then the
> + *       expected result is unrepresentable and the return value is
> invalid.
> + * @note For the input value 0, this function returns 0, which is not
> a
> power of
> + *       2.
>    */
>   static inline uint64_t
>   rte_align64pow2(uint64_t v)
> --
> 2.43.0

Yes, please submit a v2; you may  add:
Acked-by: Morten Brørup <[email protected]>

Reply via email to