Re: [PATCH 1/3] xen/arm64: head: Move earlyprintk 'hex' string to .rodata.str

[Michal Orzel <michal.orzel@xxxxxxx>]

Hi,

On 21/11/2023 19:31, Julien Grall wrote:
> 
> 
> Hi Michal,
> 
> On 21/11/2023 17:18, Michal Orzel wrote:
>>
>>
>> On 21/11/2023 18:04, Julien Grall wrote:
>>>
>>>
>>> On 21/11/2023 17:00, Michal Orzel wrote:
>>>> Hi Julien,
>>>
>>> Hi,
>>>
>>>> On 21/11/2023 17:09, Julien Grall wrote:
>>>>>
>>>>>
>>>>> Hi Michal,
>>>>>
>>>>> On 21/11/2023 09:45, Michal Orzel wrote:
>>>>>> At the moment, the 'hex' string is placed right after the 'putn'
>>>>>> function in the .text section. This is because of the limited range
>>>>>> (+/- 1MB) of PC relative 'adr' instruction 'putn' uses. Modify it to use
>>>>>> 'adr_l' instead (range extended to +/- 4GB) and move the string to
>>>>>> .rodata.str. This way all the earlyprintk messages will be part of 
>>>>>> .rodata
>>>>>> and the behavior will be consistent with what we already do on arm32.
>>>>>
>>>>> This will be correct today, but I am actually thinking to move 'hex' to
>>>>> .rodata.idmap so it can be used while we are on the 1:1 mapping and/or
>>>>> temporary mapping.
>>>> So you are planning on extending your series to also cover arm64?
>>>
>>> It is not in my soonish plan. But you are arguing that this patch is for
>>> consistency with arm32. This will not be after my series.
>> Hmm, consistency was not the only reason for sending this patch. It's a 
>> beneficial side effect
>> given that putn implementations are very similar on arm32/arm64.
>> Are you saying that
>> moving a string that is const from .text to .rodata is not a good change?
> 
> .rodata is better but I would rather prefer if this is moved to
> .rodata.idmap directly.
> 
>>
>>>
>>> And I would not change arm64 just for consistency because I don't view
>>> it as necessary. The boot code is already not the same.
>> I don't recall trying to make the entire boot code consistent. But if there 
>> are parts that
>> are almost identical, I think symmetry is welcomed. It helps reading the 
>> code, at least for me.
> 
> If you want symmetry, then I am happy to switch to .rodata.idmap. But I
> would rather not first move to .rodata and then in a month time switch
> to .rodata.idmap.
Ok, fair enough.

~Michal