Re: [PATCH v1 2/3] arm/mpu: Provide and populate MPU C data structures

["Orzel, Michal" <michal.orzel@xxxxxxx>]

On 05/06/2025 10:58, Julien Grall wrote:
> Hi Michal,
> 
> On 05/06/2025 08:44, Orzel, Michal wrote:
>>
>>
>> On 04/06/2025 19:43, Ayan Kumar Halder wrote:
>>> Do the arm32 equivalent initialization for commit id ca5df936c4.
>> This is not a good commit msg.
>> Also, we somewhat require passing 12 char long IDs.
> 
> We are following the same convention as Linux. IIRC this was updated 
> because there was some collision with 10 characters in Linux (not sure 
> if we have seen it in Xen yet).
> 
> [...]
> 
>>> diff --git a/xen/arch/arm/include/asm/mpu/regions.inc 
>>> b/xen/arch/arm/include/asm/mpu/regions.inc
>>> index 6b8c233e6c..943bcda346 100644
>>> --- a/xen/arch/arm/include/asm/mpu/regions.inc
>>> +++ b/xen/arch/arm/include/asm/mpu/regions.inc
>>> @@ -24,7 +24,13 @@
>>>   #define XEN_MPUMAP_ENTRY_SHIFT  0x3     /* 8 byte structure */
>>>   
>>>   .macro store_pair reg1, reg2, dst
>>> -    .word 0xe7f000f0                    /* unimplemented */
>>> +    str \reg1, [\dst]
>>> +    add \dst, \dst, #4
>>> +    str \reg2, [\dst]
>> AFAIR there is STM instruction to do the same
> 
> AFAICT, one issue with stm is the ordering is forced by the instruction 
> rather than the user. So \reg1 could be stored first.
Yes, I think it stores based on register number. So if you do {r2, r1} it will
still store r1 first.

> 
> I think it would be better to use "strd". It still has restriction
> (the two registers need to have contiguous index). But I think that 
> would be better if we want to reduce the number of instructions.
Ok

~Michal