Re: [PATCH 9/15] arm: Set again stack pointer as CFA reg when popping if necessary

[Richard Earnshaw <Richard.Earnshaw@foss.arm.com>]

On 09/01/2023 14:58, Andrea Corallo via Gcc-patches wrote:
> Andrea Corallo via Gcc-patches <gcc-patches@gcc.gnu.org> writes:
> 
>> Richard Earnshaw <Richard.Earnshaw@foss.arm.com> writes:
>>
>>> On 27/09/2022 16:24, Kyrylo Tkachov via Gcc-patches wrote:
>>>>
>>>>> -----Original Message-----
>>>>> From: Andrea Corallo <andrea.corallo@arm.com>
>>>>> Sent: Tuesday, September 27, 2022 11:06 AM
>>>>> To: Kyrylo Tkachov <Kyrylo.Tkachov@arm.com>
>>>>> Cc: Andrea Corallo via Gcc-patches <gcc-patches@gcc.gnu.org>; Richard
>>>>> Earnshaw <Richard.Earnshaw@arm.com>; nd <nd@arm.com>
>>>>> Subject: Re: [PATCH 9/15] arm: Set again stack pointer as CFA reg when
>>>>> popping if necessary
>>>>>
>>>>> Kyrylo Tkachov <Kyrylo.Tkachov@arm.com> writes:
>>>>>
>>>>>> Hi Andrea,
>>>>>>
>>>>>>> -----Original Message-----
>>>>>>> From: Gcc-patches <gcc-patches-
>>>>>>> bounces+kyrylo.tkachov=arm.com@gcc.gnu.org> On Behalf Of Andrea
>>>>>>> Corallo via Gcc-patches
>>>>>>> Sent: Friday, August 12, 2022 4:34 PM
>>>>>>> To: Andrea Corallo via Gcc-patches <gcc-patches@gcc.gnu.org>
>>>>>>> Cc: Richard Earnshaw <Richard.Earnshaw@arm.com>; nd <nd@arm.com>
>>>>>>> Subject: [PATCH 9/15] arm: Set again stack pointer as CFA reg when
>>>>> popping
>>>>>>> if necessary
>>>>>>>
>>>>>>> Hi all,
>>>>>>>
>>>>>>> this patch enables 'arm_emit_multi_reg_pop' to set again the stack
>>>>>>> pointer as CFA reg when popping if this is necessary.
>>>>>>>
>>>>>>
>>>>>>   From what I can tell from similar functions this is correct, but could you
>>>>> elaborate on why this change is needed for my understanding please?
>>>>>> Thanks,
>>>>>> Kyrill
>>>>>
>>>>> Hi Kyrill,
>>>>>
>>>>> sure, if the frame pointer was set, than it is the current CFA register.
>>>>> If we request to adjust the current CFA register offset indicating it
>>>>> being SP (while it's actually FP) that is indeed not correct and the
>>>>> incoherence we will be detected by an assertion in the dwarf emission
>>>>> machinery.
>>>> Thanks,  the patch is ok
>>>> Kyrill
>>>>
>>>>>
>>>>> Best Regards
>>>>>
>>>>>     Andrea
>>>
>>> Hmm, wait.  Why would a multi-reg pop be updating the stack pointer?
>>
>> Hi Richard,
>>
>> not sure I understand, isn't any pop updating SP by definition?
> 
> 
> Back on this,
> 
> compiling:
> 
> =======
> int i;
> 
> void foo (int);
> 
> int bar()
> {
>    foo (i);
>    return 0;
> }
> =======
> 
> With -march=armv8.1-m.main+fp -mbranch-protection=pac-ret+leaf -mthumb -O0 -g
> 
> Produces the following asm for bar.
> 
> bar:
> 	@ args = 0, pretend = 0, frame = 0
> 	@ frame_needed = 1, uses_anonymous_args = 0
> 	pac	ip, lr, sp
> 	push	{r3, r7, ip, lr}
> 	add	r7, sp, #0
> 	ldr	r3, .L3
> 	ldr	r3, [r3]
> 	mov	r0, r3
> 	bl	foo
> 	movs	r3, #0
> 	mov	r0, r3
> 	pop	{r3, r7, ip, lr}
> 	aut	ip, lr, sp
> 	bx	lr
> 
> The offending instruction causing the ICE (without this patch) when
> emitting dwarf is "pop {r3, r7, ip, lr}".
> 
> The current CFA reg when emitting the multipop is R7 (the frame
> pointer).  If is not the multipop that has the duty to restore SP as
> current CFA here which other instruction should do it?

Ah, OK.  I think this is a special case, though, because in this 
specific case the frame pointer (r7) and the stack pointer point to the 
same place.  This means that in the epilogue we don't start by restoring 
SP from FP (at which point we tell the dwarf code that the frame is back 
in SP again).

For example, if I have:


int i;

void foo (int, int*);

int bar()
{
   int j[10];
   foo (i,j);
   return 0;
}


then the epilogue sequence starts with:

         adds    r7, r7, #40
         .cfi_def_cfa_offset 8
         mov     sp, r7
         .cfi_def_cfa_register 13

And then the pop works correctly as-is.

But I'm not convinced that this is enough anyway, you cause the compiler 
to output a directive that changes the CFA pointer back to r13, but you 
don't output anything that changes the CFA offset.  So I think this 
means that the CFA state machine ends up pointing to the wrong location, 
but it's hard to tell as you haven't shown the CFA directives in your 
example above.

> 
> Best Regards
> 
>    Andrea

R.