Le 21/12/2021 à 20:36, Segher Boessenkool a écrit :
> On Tue, Dec 21, 2021 at 05:49:33PM +0000, Christophe Leroy wrote:
>> diff --git a/arch/powerpc/kernel/vdso32/datapage.S
>> b/arch/powerpc/kernel/vdso32/datapage.S
>> index 65244416ab94..db8e167f0166 100644
>> --- a/arch/powerpc/kernel/vdso32/datapage.S
>> +++ b/arch/powerpc/kernel/vdso32/datapage.S
>> @@ -30,11 +30,15 @@ V_FUNCTION_BEGIN(__kernel_get_syscall_map)
>> mr. r4,r3
>> get_datapage r3
>> mtlr r12
>> +#ifdef __powerpc64__
>> + addi r3,r3,CFG_SYSCALL_MAP64
>> +#else
>> addi r3,r3,CFG_SYSCALL_MAP32
>> +#endif
>> + crclr cr0*4+so
>> beqlr
>> li r0,NR_syscalls
>> stw r0,0(r4)
>> - crclr cr0*4+so
>> blr
>
> This now clears cr0.3 if r3 is zero on entry. Is that on purpose?
All crclr where added by commit 5d66da3d71e6 ("[PATCH] powerpc: Make the
vDSO functions set error code (#2)")
It looks like it was properly added to ppc64's version of
__kernel_get_syscall_map() but for ppc32 CR.SO was left undefined in
case of early exit of __kernel_get_syscall_map()
So yes, I think the change is on purpose.
>
> Rest looks good.
>
> Reviewed-by: Segher Boessenkool <seg...@kernel.crashing.org>
>
Thanks
Christophe
