[RFC,V2 2/5] gas: aarch64: suppport CFI directive .cfi_mte_tagged_frame

[Indu Bhagat]

On 4/11/25 2:49 AM, Richard Sandiford wrote:
> Indu Bhagat <indu.bhagat@oracle.com> writes:
>> Process a new aarch64-specific CFI directive: .cfi_mte_tagged_frame
>> (LLVM uses this CFI directive already).  The CFI directive, when present
>> for a function, indicates that the stack frame for the function may
>> modify the MTE tags of the stack space it uses.  The assembler emits
>> char 'G' in the CIE augmentation string to indicate the same.
>>
>> TBD:
>>    - Whats that comment style in c-aarch64.texi ?
> 
> I think it's just alphabetical.  The new entry would then go under:
> 
> @c CCCCCCCCCCCCCCCCCCCCCCCCCC
> 
> and before:
> 
> @cindex @code{.cpu} directive, AArch64
> 
> 

OK.

>>   /* Extra equivalence checks required by AArch64 when selecting the correct cie
>>      for some fde.  Currently only used to check for quivalence between keys used
>>      to sign ther return address.  */
>> -#define tc_cie_fde_equivalent_extra(cie, fde) (cie->pauth_key == fde->pauth_key)
>> +#define tc_cie_fde_equivalent_extra(cie, fde) ((cie->pauth_key == fde->pauth_key) \
>> +					      && (cie->memtag_frame_p = fde->memtag_frame_p))
> 
> Should be == rather than =.
> 
> It would be good to extend the testcase so that it would have caught this.
> 

Swapping the order of the two functions in the testcase (with the first 
without .cfi_mte_tagged_frame and the second with .cfi_mte_tagged_frame) 
was enough to help catch this.

I have amended the current testcase now.

Thanks!

> Otherwise it looks good to me.
> 
> Thanks,
> Richard