[PATCH v2] powerpc/uaccess: Use flexible addressing with __put_user()/__get_user()

[Christophe Leroy]

Le 08/07/2020 à 06:49, Christophe Leroy a écrit :
> 
> 
> Le 07/07/2020 à 21:02, Christophe Leroy a écrit :
>>
>>
>> Le 07/07/2020 à 14:44, Christophe Leroy a écrit :
>>>
>>>
>>> Le 30/06/2020 à 03:19, Michael Ellerman a écrit :
>>>> Michael Ellerman <mpe at ellerman.id.au> writes:
>>>>> Christophe Leroy <christophe.leroy at csgroup.eu> writes:
>>>>>> Hi Michael,
>>>>>>
>>>>>> I see this patch is marked as "defered" in patchwork, but I can't see
>>>>>> any related discussion. Is it normal ?
>>>>>
>>>>> Because it uses the "m<>" constraint which didn't work on GCC 4.6.
>>>>>
>>>>> https://github.com/linuxppc/issues/issues/297
>>>>>
>>>>> So we should be able to pick it up for v5.9 hopefully.
>>>>
>>>> It seems to break the build with the kernel.org 4.9.4 compiler and
>>>> corenet64_smp_defconfig:
>>>
>>> Most likely a GCC bug ?
>>>
>>> It seems the problem vanishes with patch 
>>> https://patchwork.ozlabs.org/project/linuxppc-dev/patch/173de3b659fa3a5f126a0eb170522cccd909950f.1594125164.git.christophe.leroy@csgroup.eu/ 
>>>
>>
>> Same kind of issue in signal_64.c now.
>>
>> The following patch fixes it: 
>> https://patchwork.ozlabs.org/project/linuxppc-dev/patch/810bd8840ef990a200f58c9dea9abe767ca02a3a.1594146723.git.christophe.leroy@csgroup.eu/ 
>>
>>
> 
> This time I confirm, with the two above mentioned patches, it builds OK 
> with 4.9, see 
> http://kisskb.ellerman.id.au/kisskb/head/810bd8840ef990a200f58c9dea9abe767ca02a3a/ 
> 
> 

I see you've merged those patches that make the issue disappear, yet not 
this patch yet. I guess you are still a bit chilly about it, so I split 
it in two parts for you to safely take patch 1 as soon as possible while 
handling the "m<>" constraint subject more carefully via 
https://github.com/linuxppc/issues/issues/297 in a later stage.

Anyway, it seems that GCC doesn't make much use of the "m<>" and the 
pre-update form. Most of the benefit of flexible addressing seems to be 
achieved with patch 1 ie without the "m<>" constraint and update form.

Christophe