On 08.07.2025 19:32, Stefano Stabellini wrote:
> On Tue, 8 Jul 2025, Jan Beulich wrote:
>> On 08.07.2025 00:00, Stefano Stabellini wrote:
>>> On Tue, 24 Jun 2025, Jan Beulich wrote:
>>>> On 24.06.2025 02:20, victorm.l...@amd.com wrote:
>>>>> From: Nicola Vetrini <nicola.vetr...@bugseng.com>
>>>>>
>>>>> Use {get,put}_unaligned_t to ensure that reads and writes are
>>>>> safe to perform even on potentially misaligned pointers.
>>>>
>>>> Also applicable to the Arm patch: Please can such patches mention the
>>>> main subject of the rule, not just the number?
>>>
>>> +1
>>>
>>>
>>>> Overall I'm unconvinced we really want or need this on x86; I'm curious
>>>> what Andrew and Roger think.
>>>
>>> To be honest, I had a similar reaction to you, which is why I suggested
>>> on Matrix to:
>>>
>>> - deviate the rule in its entirety on x86
>>> - deviate the rule for all mappings except for devmem mappings on ARM
>>>
>>> Leaving aside ARM for a second (this is exactly the kind of very
>>> arch-specific behavior that is OK to device differently per
>>> architecture), would you be OK with deviating the rule in its entirety on
>>> x86?
>>>
>>> Or do you prefer to continue with this patch?
>>
>> Neither. Imo globally deviating rules needs to be done with care. There
>> are, in principle, misaligned accesses in x86 which can be made fault
>> (and I think this was mentioned before). We want to know of such risks.
>> Hence for a rule like this one more fine grained deviation is on order,
>> imo.
>
> What fine grained deviation do you have in mind?
Ones for almost(?) everything that is having actual code changes right now
in this series.
Jan
