[PATCH v6 2/5] x86: Add SFrame support for x86 architecture

[Adhemerval Zanella Netto]

On 02/07/25 04:23, Florian Weimer wrote:
> * Adhemerval Zanella Netto:
> 
>>> +  if (*(unsigned char *)(pc+0) == 0x48
>>> +      && *(unsigned long long *)(pc+1) == RT_SIGRETURN_SYSCALL)
>>
>> This is strictly UB, although x86_64 supports unaligned access.  You can 
>> use something like:
>>
>> #define __read_unaligned_t(__type, __ptr) ({                            \
>>         const struct { type x; } __attribute__((__packed__)) * __pptr = \
>>           (typeof(_pptr))(__ptr);                                       \
>>         __pptr->x;                                                      \
>> })
>> #define read_unaligned(ptr)   __read_unaligned_t(typeof(*(ptr)), (ptr))
>>
>> Or the memcpy to a temporary (as we do on some code internally).  Afaik 
>> the packet attribute tend to generate better code.
> 
> I would just use memcmp.  GCC will expand it inline if the ordering is
> not used.

On most architectures that supports unaligned memory operations it will be
the same, however on architectures that do not the packed generates slight 
better code [1] (on sparc the packed code avoids the store/load on the 
temporary variable).

[1] https://godbolt.org/z/eE3qeP1Kc