[PATCH v2 3/3] Linux: Add tests that check that TLS and rseq area are separate

[Michael Jeanson]

On 2025-01-14 10:22, Michael Jeanson wrote:
> On 2025-01-14 01:40, Florian Weimer wrote:
>> * Michael Jeanson:
>>
>>> On 2025-01-13 07:33, Florian Weimer wrote:
>>>> +static void
>>>> +add_rseq (void)
>>>> +{
>>>> +  if (__rseq_size > 0)
>>>> +    add_range ("", "rseq area",
>>>> +               (char *) __thread_pointer () + __rseq_offset, __rseq_size);
>>>> +}
>>>
>>> Since there is always an rseq area of at least RSEQ_AREA_SIZE_INITIAL
>>> (32 bytes) allocated even when rseq is disabled, we should always add
>>> the range to the test, something like that :
>>>
>>>  +static void
>>>  +add_rseq (void)
>>>  +{
>>>  +  add_range ("", "rseq area", (char *) __thread_pointer () + __rseq_offset,
>>>  +  	MAX (__rseq_size, RSEQ_AREA_SIZE_INITIAL));
>>>  +}
>>
>> Or perhaps round up the size to a multiple of 32?
> 
> Right, '__rseq_size' rounded up to '_rseq_align' will give the actual
> allocation including the padding, even when rseq is disabled. However,
> I'm not sure if we can access '_rseq_aling' in the tests.

Well, answering my own question, we can just use:

  size_t rseq_align = MAX (getauxval (AT_RSEQ_ALIGN), RSEQ_MIN_ALIGN);

> 
>>
>> Thanks,
>> Florian
>>
>