Re: linux-next: build warnings after merge of the kvm-arm-fixes tree

[Marc Zyngier]

On Tue, 07 Jul 2026 13:11:55 +0100,
Mark Brown <broonie@xxxxxxxxxx> wrote:
> 
> Hi all,
> 
> After merging the kvm-arm-fixes tree, today's linux-next build
> (arm64 defconfig) started generating warnings:
> 
> In file included from /tmp/next/build/include/linux/limits.h:7,
>                  from /tmp/next/build/include/linux/overflow.h:6,
>                  from /tmp/next/build/include/linux/bits.h:32,
>                  from /tmp/next/build/arch/arm64/include/asm/kvm_pgtable.h:10,
>                  from /tmp/next/build/arch/arm64/kvm/hyp/pgtable.c:11:
> /tmp/next/build/arch/arm64/kvm/hyp/pgtable.c: In function 'kvm_pgtable_stage2_relax_perms':
> /tmp/next/build/include/vdso/limits.h:8:25: warning: overflow in conversion from 'int' to 's8' {aka 'signed char'} changes value from '2147483647' to '-1' [-Woverflow]
>     8 | #define INT_MAX         ((int)(~0U >> 1))
>       |                         ^
> /tmp/next/build/arch/arm64/include/asm/tlbflush.h:131:33: note: in expansion of macro 'INT_MAX'
>   131 | #define TLBI_TTL_UNKNOWN        INT_MAX
>       |                                 ^~~~~~~
> /tmp/next/build/arch/arm64/kvm/hyp/pgtable.c:1361:20: note: in expansion of macro 'TLBI_TTL_UNKNOWN'
>  1361 |         s8 level = TLBI_TTL_UNKNOWN;
>       |                    ^~~~~~~~~~~~~~~~
> In file included from /tmp/next/build/include/linux/limits.h:7,
>                  from /tmp/next/build/include/linux/overflow.h:6,
>                  from /tmp/next/build/include/linux/bits.h:32,
>                  from /tmp/next/build/arch/arm64/include/asm/kvm_pgtable.h:10,
>                  from /tmp/next/build/arch/arm64/kvm/hyp/nvhe/../pgtable.c:11:
> /tmp/next/build/arch/arm64/kvm/hyp/nvhe/../pgtable.c: In function 'kvm_pgtable_stage2_relax_perms':
> /tmp/next/build/include/vdso/limits.h:8:25: warning: overflow in conversion from 'int' to 's8' {aka 'signed char'} changes value from '2147483647' to '-1' [-Woverflow]
>     8 | #define INT_MAX         ((int)(~0U >> 1))
>       |                         ^
> /tmp/next/build/arch/arm64/include/asm/tlbflush.h:131:33: note: in expansion of macro 'INT_MAX'
>   131 | #define TLBI_TTL_UNKNOWN        INT_MAX
>       |                                 ^~~~~~~
> /tmp/next/build/arch/arm64/kvm/hyp/nvhe/../pgtable.c:1361:20: note: in expansion of macro 'TLBI_TTL_UNKNOWN'
>  1361 |         s8 level = TLBI_TTL_UNKNOWN;
>       |                    ^~~~~~~~~~~~~~~~
> 
> 
> Caused by commit
> 
>    100baf0184896 (KVM: arm64: Ensure level is always initialized when relaxing perms)
> 
> I imagine this will cause an allyesconfig build failure when I get to
> that due to -Werror.

Crap, I don't know how I missed that one! And the whole thing is
pretty dodgy anyway:

- level starts as s8
- passed to __kvm_tlb_flush_vmid_ipa_nsh() as int
- passed to __tlbi_level() as u32
- compared to the value 3

The problem is that we are conflating the level of a descriptor and
the TTL, which may or may not be related.

I'm minded to queue something like this on top. Oliver?

Thanks for the heads up,

	M.

diff --git a/arch/arm64/kvm/hyp/pgtable.c b/arch/arm64/kvm/hyp/pgtable.c
index 8754c99c22f2f..70dceb20dfada 100644
--- a/arch/arm64/kvm/hyp/pgtable.c
+++ b/arch/arm64/kvm/hyp/pgtable.c
@@ -1356,7 +1356,7 @@ int kvm_pgtable_stage2_relax_perms(struct kvm_pgtable *pgt, u64 addr,
 				   enum kvm_pgtable_prot prot, enum kvm_pgtable_walk_flags flags)
 {
 	kvm_pte_t xn = 0, set = 0, clr = 0;
-	s8 level = TLBI_TTL_UNKNOWN;
+	s8 level;
 	int ret;
 
 	if (prot & KVM_PTE_LEAF_ATTR_HI_SW)
@@ -1379,7 +1379,8 @@ int kvm_pgtable_stage2_relax_perms(struct kvm_pgtable *pgt, u64 addr,
 
 	ret = stage2_update_leaf_attrs(pgt, addr, 1, set, clr, NULL, &level, flags);
 	if (!ret || ret == -EAGAIN)
-		kvm_call_hyp(__kvm_tlb_flush_vmid_ipa_nsh, pgt->mmu, addr, level);
+		kvm_call_hyp(__kvm_tlb_flush_vmid_ipa_nsh, pgt->mmu, addr,
+			     (ret == -EAGAIN) ? TLBI_TTL_UNKNOWN : level);
 	return ret;
 }
 

-- 
Without deviation from the norm, progress is not possible.