Re: [kvm-unit-tests PATCH v2 3/8] RFC: s390x: Define STACK_FRAME_INT_SIZE macro

[Janosch Frank <frankja@linux.ibm.com>]

On 2/17/21 4:38 PM, Thomas Huth wrote:
> On 17/02/2021 15.41, Janosch Frank wrote:
>> Using sizeof is safer than using magic constants. However, it doesn't
>> really fit into asm-offsets.h as it's not an offset so I'm happy to
>> receive suggestions on where to put it.
>>
>> Signed-off-by: Janosch Frank <frankja@linux.ibm.com>
>> ---
>>   lib/s390x/asm-offsets.c | 1 +
>>   s390x/macros.S          | 4 ++--
>>   2 files changed, 3 insertions(+), 2 deletions(-)
>>
>> diff --git a/lib/s390x/asm-offsets.c b/lib/s390x/asm-offsets.c
>> index 96cb21cf..2658b59a 100644
>> --- a/lib/s390x/asm-offsets.c
>> +++ b/lib/s390x/asm-offsets.c
>> @@ -86,6 +86,7 @@ int main(void)
>>   	OFFSET(STACK_FRAME_INT_CRS, stack_frame_int, crs);
>>   	OFFSET(STACK_FRAME_INT_GRS0, stack_frame_int, grs0);
>>   	OFFSET(STACK_FRAME_INT_GRS1, stack_frame_int, grs1);
>> +	DEFINE(STACK_FRAME_INT_SIZE, sizeof(struct stack_frame_int));
>>   
>>   	return 0;
>>   }
>> diff --git a/s390x/macros.S b/s390x/macros.S
>> index d7eeeb55..a7d62c6f 100644
>> --- a/s390x/macros.S
>> +++ b/s390x/macros.S
>> @@ -43,14 +43,14 @@
>>   /* Save registers on the stack (r15), so we can have stacked interrupts. */
>>   	.macro SAVE_REGS_STACK
>>   	/* Allocate a full stack frame */
>> -	slgfi   %r15, 32 * 8 + 4 * 8
>> +	slgfi   %r15, STACK_FRAME_INT_SIZE
>>   	/* Store registers r0 to r14 on the stack */
>>   	stmg    %r2, %r15, STACK_FRAME_INT_GRS0(%r15)
>>   	stg     %r0, STACK_FRAME_INT_GRS1(%r15)
>>   	stg     %r1, STACK_FRAME_INT_GRS1 + 8(%r15)
>>   	/* Store the gr15 value before we allocated the new stack */
>>   	lgr     %r0, %r15
>> -	algfi   %r0, 32 * 8 + 4 * 8
>> +	algfi   %r0, STACK_FRAME_INT_SIZE
> 
> Ah, well, that of course fixes the problem that I had with the previous 
> patch. I'd suggest to merge it into patch 2.

That was the plan anyway, I had so much pain until I yanked out the int
offsets in favor of the macros. :)

Did you have time to read the commit message?
I'm not completely convinced that asm-offset.c is the right place for
the DEFINE() so I kept this patch as a discussion starter.

> 
>   Thomas
>