Hi David, Arnd, LoongArch,
In a linux-next defconfig LLVM=1 build today I got:
./include/asm-generic/tlb.h:629:10: error: parameter 'ptep' set
but not used [-Werror,-Wunused-but-set-parameter]
629 | pte_t *ptep, unsigned int nr, unsigned long address)
| ^
Indeed, in loongarch, `__tlb_remove_tlb_entry` does not do anything.
This seems the same that Arnd reported for arm64:
https://lore.kernel.org/all/[email protected]/
So perhaps the loongarch's one should also be changed into an static inline?
I hope that helps!
Cheers,
Miguel
On 16.04.24 12:26, Miguel Ojeda wrote:
> Hi David, Arnd, LoongArch,
>
> In a linux-next defconfig LLVM=1 build today I got:
>
> ./include/asm-generic/tlb.h:629:10: error: parameter 'ptep' set
> but not used [-Werror,-Wunused-but-set-parameter]
> 629 | pte_t *ptep, unsigned int nr, unsigned long address)
> | ^
>
> Indeed, in loongarch, `__tlb_remove_tlb_entry` does not do anything.
> This seems the same that Arnd reported for arm64:
>
> https://lore.kernel.org/all/[email protected]/
>
> So perhaps the loongarch's one should also be changed into an static inline?
4d5bf0b6183f79ea361dd506365d2a471270735c is already part of v6.9-rc1. How come
we see that only now on linux-next?
I assume we should see the same on upstream Linux with LLVM=1, correct?
If so, we should likely just drop that completely and rely on the asm-generic one:
diff --git a/arch/loongarch/include/asm/tlb.h b/arch/loongarch/include/asm/tlb.h
index da7a3b5b9374a..e071f5e9e8580 100644
--- a/arch/loongarch/include/asm/tlb.h
+++ b/arch/loongarch/include/asm/tlb.h
@@ -132,8 +132,6 @@ static __always_inline void invtlb_all(u32 op, u32 info, u64 addr)
);
}
-#define __tlb_remove_tlb_entry(tlb, ptep, address) do { } while (0)
-
static void tlb_flush(struct mmu_gather *tlb);
#define tlb_flush tlb_flush
Thanks!
--
Cheers,
David / dhildenb
On Tue, Apr 16, 2024, at 15:51, David Hildenbrand wrote:
> On 16.04.24 12:26, Miguel Ojeda wrote:
>> Hi David, Arnd, LoongArch,
>>
>> In a linux-next defconfig LLVM=1 build today I got:
>>
>> ./include/asm-generic/tlb.h:629:10: error: parameter 'ptep' set
>> but not used [-Werror,-Wunused-but-set-parameter]
>> 629 | pte_t *ptep, unsigned int nr, unsigned long address)
>> | ^
>>
>> Indeed, in loongarch, `__tlb_remove_tlb_entry` does not do anything.
>> This seems the same that Arnd reported for arm64:
>>
>> https://lore.kernel.org/all/[email protected]/
>>
>> So perhaps the loongarch's one should also be changed into an static inline?
>
> 4d5bf0b6183f79ea361dd506365d2a471270735c is already part of v6.9-rc1. How come
> we see that only now on linux-next?
Andrew merged my patch to enable -Wextra yesterday, and it appears
that this one fell through the cracks with my testing, either I
missed the combination of loongarch with clang, or I last tested
it before your patches got merged.
> I assume we should see the same on upstream Linux with LLVM=1, correct?
On upstream, it only shows up with 'make W=1'.
> If so, we should likely just drop that completely and rely on the
> asm-generic one:
>
> diff --git a/arch/loongarch/include/asm/tlb.h
> b/arch/loongarch/include/asm/tlb.h
> index da7a3b5b9374a..e071f5e9e8580 100644
> --- a/arch/loongarch/include/asm/tlb.h
> +++ b/arch/loongarch/include/asm/tlb.h
> @@ -132,8 +132,6 @@ static __always_inline void invtlb_all(u32 op, u32
> info, u64 addr)
> );
> }
>
> -#define __tlb_remove_tlb_entry(tlb, ptep, address) do { } while (0)
> -
> static void tlb_flush(struct mmu_gather *tlb);
Yes, this looks like the best solution, and I can confirm that this
addresses the warning on linux-next.
Tested-by: Arnd Bergmann <[email protected]>
On Tue, Apr 16, 2024 at 10:14 PM Arnd Bergmann <[email protected]> wrote:
>
> On Tue, Apr 16, 2024, at 15:51, David Hildenbrand wrote:
> > On 16.04.24 12:26, Miguel Ojeda wrote:
> >> Hi David, Arnd, LoongArch,
> >>
> >> In a linux-next defconfig LLVM=1 build today I got:
> >>
> >> ./include/asm-generic/tlb.h:629:10: error: parameter 'ptep' set
> >> but not used [-Werror,-Wunused-but-set-parameter]
> >> 629 | pte_t *ptep, unsigned int nr, unsigned long address)
> >> | ^
> >>
> >> Indeed, in loongarch, `__tlb_remove_tlb_entry` does not do anything.
> >> This seems the same that Arnd reported for arm64:
> >>
> >> https://lore.kernel.org/all/[email protected]/
> >>
> >> So perhaps the loongarch's one should also be changed into an static inline?
> >
> > 4d5bf0b6183f79ea361dd506365d2a471270735c is already part of v6.9-rc1. How come
> > we see that only now on linux-next?
>
> Andrew merged my patch to enable -Wextra yesterday, and it appears
> that this one fell through the cracks with my testing, either I
> missed the combination of loongarch with clang, or I last tested
> it before your patches got merged.
>
> > I assume we should see the same on upstream Linux with LLVM=1, correct?
>
> On upstream, it only shows up with 'make W=1'.
>
> > If so, we should likely just drop that completely and rely on the
> > asm-generic one:
> >
> > diff --git a/arch/loongarch/include/asm/tlb.h
> > b/arch/loongarch/include/asm/tlb.h
> > index da7a3b5b9374a..e071f5e9e8580 100644
> > --- a/arch/loongarch/include/asm/tlb.h
> > +++ b/arch/loongarch/include/asm/tlb.h
> > @@ -132,8 +132,6 @@ static __always_inline void invtlb_all(u32 op, u32
> > info, u64 addr)
> > );
> > }
> >
> > -#define __tlb_remove_tlb_entry(tlb, ptep, address) do { } while (0)
> > -
> > static void tlb_flush(struct mmu_gather *tlb);
>
> Yes, this looks like the best solution, and I can confirm that this
> addresses the warning on linux-next.
Emmm, this should be removed in the first place because x86 removed it
at 5.12...
Huacai
>
> Tested-by: Arnd Bergmann <[email protected]>