[COMMITTED] gas: sframe: do not test whether offsetT exceeds INT64_MIN..INT64_MAX
Jan Beulich
jbeulich@suse.com
Wed Jan 14 11:02:06 GMT 2026
On 14.01.2026 11:18, Jens Remus wrote:
> On 1/14/2026 8:34 AM, Jan Beulich wrote:
>> On 13.01.2026 20:12, Indu Bhagat wrote:
>>> On 1/13/26 8:18 AM, Jens Remus wrote:
>>>> On 1/13/2026 5:10 PM, Jan Beulich wrote:
>>>>> On 13.01.2026 17:07, Jens Remus wrote:
>>>>>> On 1/13/2026 4:52 PM, Jan Beulich wrote:
>>>>>>> On 13.01.2026 16:46, Jens Remus wrote:
>>>>>>>> On 1/13/2026 4:37 PM, Jan Beulich wrote:
>>>>>>>>> On 13.01.2026 16:27, Jens Remus wrote:
>>>>>>>>>> On 1/13/2026 2:59 PM, Jan Beulich wrote:
>>>>>>>>>>> On 13.01.2026 14:24, Jens Remus wrote:
>>>>>>>>>>>> A value of type offsetT, which is either a signed 32-bit or 64-bit
>>>>>>>>>>>> integer, cannot exceed the range of INT64_MIN..INT64_MAX. This
>>>>>>>>>>>> resolves the following compile error:
>>>>>>>>>>>>
>>>>>>>>>>>> ../../binutils-gdb/gas/gen-sframe.c: In function ‘get_offset_size_in_bytes’:
>>>>>>>>>>>> ../../binutils-gdb/gas/gen-sframe.c:213:45: error: comparison is always true due to limited range of data type [-Werror=type-limits]
>>>>>>>>>>>> 213 | else if ((sizeof (offsetT) > 4) && (value <= INT64_MAX && value >= INT64_MIN))
>>>>>>>>>>>> | ^~
>>>>>>>>>>>> ../../binutils-gdb/gas/gen-sframe.c:213:67: error: comparison is always true due to limited range of data type [-Werror=type-limits]
>>>>>>>>>>>> 213 | else if ((sizeof (offsetT) > 4) && (value <= INT64_MAX && value >= INT64_MIN))
>>>>>>>>>>>
>>>
>>> Sorry for not detecting this in review. I remembered adding those
>>> typecasts for addressing some warnings in the past, but for the life of
>>> me, I couldnt explain to myself why would I have seen any warnings...
>>>
>>>>>>>>>>> This same error ...
>>>>>>>>>>>
>>>>>>>>>>>> --- a/gas/gen-sframe.c
>>>>>>>>>>>> +++ b/gas/gen-sframe.c
>>>>>>>>>>>> @@ -210,7 +210,7 @@ get_offset_size_in_bytes (offsetT value)
>>>>>>>>>>>> size = 2;
>>>>>>>>>>>> else if (value <= INT32_MAX && value >= INT32_MIN)
>>>>>>>>>>>
>>>>>>>>>>> ... will trigger here on a !BFD64 build.
>>>>>>>>>>
>>>>>>>>>> Argh! Good catch! Thanks for letting me know! How do I resolve this
>>>>>>>>>> properly? Shall I commit another fix "gas: sframe: do not test whether
>>>>>>>>>> 32-bit offsetT exceeds INT32_MIN..INT32_MAX" or shall I better revert
>>>>>>>>>> both commits and start fresh with a v3 review of the initial patch?
>>>>>>>>>
>>>>>>>>> Incrementally is fine with me.
>>>>>>>>>
>>>>>>>>>> I propose the following as solution:
>>>>>>>>>>
>>>>>>>>>> if (value <= INT8_MAX && value >= INT8_MIN)
>>>>>>>>>> size = 1;
>>>>>>>>>> else if (value <= INT16_MAX && value >= INT16_MIN)
>>>>>>>>>> size = 2;
>>>>>>>>>> else if ((sizeof (offsetT) == 4) || (value <= INT32_MAX && value >= INT32_MIN))
>>>>>>>>>
>>>>>>>>> Depending on the internal workings of compilers, this may still trigger
>>>>>>>>> the same diagnostic. I.e. depending on whether the rhs of the || is
>>>>>>>>> eliminated ahead of or after when these kinds of warnings are issued.
>>>>>>>>> A pre-processor conditional involving BFD64 may be better.
>>>>>>>>
>>>>>>>> Is the following patch ok?
>>>>>>>
>>>>>>> In principle yes; the part I need to defer to Indu or you is whether ...
>>>>>>>
>>>>>>>> --- a/gas/gen-sframe.c
>>>>>>>> +++ b/gas/gen-sframe.c
>>>>>>>> @@ -208,10 +208,15 @@ get_offset_size_in_bytes (offsetT value)
>>>>>>>> size = 1;
>>>>>>>> else if (value <= INT16_MAX && value >= INT16_MIN)
>>>>>>>> size = 2;
>>>>>>>> +#ifdef BFD64
>>>>>>>> else if (value <= INT32_MAX && value >= INT32_MIN)
>>>>>>>> size = 4;
>>>>>>>> - else if (sizeof (offsetT) > 4)
>>>>>>>> + else
>>>>>>>> size = 8;
>>>>>>>> +#else
>>>>>>>> + else
>>>>>>>> + size = 4;
>>>>>>>> +#endif
>>>>>>>
>>>>>>> ... 8-byte sizes really are entirely impossible to use when !BFD64. Since
>>>>>>> it's an offsetT that is passed in, this would seem logical, but I'd rather
>>>>>>> have Indu / you double-check.
>>>>>>
>>>>>> Condensed excerpt from gas/as.h:
>>>>>>
>>>>>> #ifdef BFD64
>>>>>> typedef int64_t bfd_signed_vma;
>>>>>> #else /* not BFD64 */
>>>>>> typedef int32_t bfd_signed_vma;
>>>>>> #endif /* not BFD64 */
>>>>>>
>>>>>> typedef bfd_signed_vma offsetT;
>>>>>>
>>>>>> If BFD64, then offsetT is a signed 64-bit integer. Otherwise it is a
>>>>>> signed 32-bit integer. As a result it is impossible for !BFD64 to have
>>>>>> a 64-bit offsetT.
>>>>>
>>>>> Sure, hence my reference to offsetT in my earlier reply.
>>>>>
>>>>>> Or were you concerned whether SFrame would be broken for !BFD64 targets
>>>>>> in general? AFAIK SFrame does only allow 8-bit, 16-bit, and 32-bit
>>>>>> SFrame offsets. 64-bit ones are not supported:
>>>>>>
>>>>>> Excerpt from gas/gen-sframe.c:
>>>>>>
>>>>>> #define SFRAME_FRE_OFFSET_FUNC_MAP_INDEX_1B 0 /* SFRAME_FRE_OFFSET_1B. */
>>>>>> #define SFRAME_FRE_OFFSET_FUNC_MAP_INDEX_2B 1 /* SFRAME_FRE_OFFSET_2B. */
>>>>>> #define SFRAME_FRE_OFFSET_FUNC_MAP_INDEX_4B 2 /* SFRAME_FRE_OFFSET_4B. */
>>>>>> #define SFRAME_FRE_OFFSET_FUNC_MAP_INDEX_8B 3 /* Not supported in SFrame. */
>>>>>
>>>>> Then why is size 8 even a possible return value of the function?
>>>> Honestly I don't know. Indu, could you please shed some light on this?
>>>>
>>>
>>> It was just to keep keep the function get_offset_size_in_bytes () usable
>>> for any offsetT value (and doing only what the function name hints). We
>>> do error checking/failure in callers of get_offset_size_in_bytes ().
>>>
>>>> Given SFrame does not use offsetT for the respective offsets, would it
>>>> instead make sense to change the function to accept a "int value" or
>>>> "int32_t value" (instead of "offsetT value") and remove the 64-bit case?
>>>>
>>>> static unsigned int
>>>> get_offset_size_in_bytes (int value)
>>>> {
>>>> unsigned int size = 0;
>>>>
>>>> if (value <= INT8_MAX && value >= INT8_MIN)
>>>> size = 1;
>>>> else if (value <= INT16_MAX && value >= INT16_MIN)
>>>> size = 2;
>>>> else
>>>> size = 4;
>>>>
>>>> return size;
>>>> }
>>>>
>>>
>>> Wouldnt the compiler then warn about the narrowing of arg value which is
>>> offsetT in the rest of gen-sframe.c ?
>>
>> Not unless that kind of warning is explicitly enabled, afaik. Such "silent"
>> truncation occurs all over the place if I'm not mistaken. Question is whether
>> callers may actually pass values that fit in offsetT but not in int32_t
>> (implying BFD64). In that case maybe take Jens's suggestion above but keep
>> the parameter at offsetT and add
>>
>> int32_t val = value;
>>
>> gas_assert(val == value);
>>
>> (provided passing of such arguments really is a bug in the caller).
>>
>> Yet another possible alternative:
>
> I like that! Simple and easy to understand. I'll send a new patch for review.
>
>>
>> static unsigned int
>> get_offset_size_in_bytes (offsetT value)
>> {
>> unsigned int size = 0;
>>
>> if (value == (int8_t)value)
>> size = 1;
>> else if (value == (int16_t)value)
>> size = 2;
>> else if (value == (int32_t)value)
>> size = 4;
>> else if (sizeof (value) > 4)
>
> Can't this be simply:
>
> else
>
> If offsetT is a signed 32-bit integer, then (value == (int32_t)value)
> must be true.
It could, yes. I'm not quite sure why I added it back in.
Jan
More information about the Binutils
mailing list