Re: [PATCH 4/5] KVM: arm64: vgic-v3: Simplify initial GICv3 configuration sampling

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:
- [Medium] The `noinstr` function `kvm_patch_ich_vtr_el2()` calls a non-`noinstr` function (`vgic_v3_broken_seis()`), causing a `noinstr` section violation.
- [Medium] `WARN_ON` inside the `noinstr` alternative patching callback `kvm_patch_ich_vtr_el2()` violates instrumentation rules and risks deadlocks.
--

--- Patch [4]: [PATCH 4/5] KVM: arm64: vgic-v3: Simplify initial GICv3 configuration sampling ---
commit bf26cd3d96184399493f95667d1786d9b2823890
Author: Marc Zyngier <maz@kernel.org>

KVM: arm64: vgic-v3: Simplify initial GICv3 configuration sampling

This commit simplifies the initial GICv3 configuration sampling by returning
early if GICv5 CPU interface support is present. It also inlines a workaround
for a broken locally generated SEI directly into the alternative patching
callback.

[ ... ]

> diff --git a/arch/arm64/kvm/vgic/vgic-v3.c b/arch/arm64/kvm/vgic/vgic-v3.c
> index ef62dd498f5cf..68d4a486df0a1 100644
> --- a/arch/arm64/kvm/vgic/vgic-v3.c
> +++ b/arch/arm64/kvm/vgic/vgic-v3.c
> @@ -906,6 +906,9 @@ void noinstr kvm_patch_ich_vtr_el2(struct alt_instr *alt,
>  
>  	vtr = res.a1;
>  
> +	if (vgic_v3_broken_seis())
> +		vtr &= ~ICH_VTR_EL2_SEIS;

[Severity: Medium]
Does this code call an instrumentable function from within a noinstr section?

Looking at vgic_v3_broken_seis(), it calls is_midr_in_range_list(). Calling an
instrumentable function from a noinstr function like kvm_patch_ich_vtr_el2()
breaks safety invariants, which can result in objtool build warnings or
ftrace/kprobe recursion deadlocks if the patching code is traced.

[Severity: Medium]
Also, just before this addition in kvm_patch_ich_vtr_el2(), there is a WARN_ON
call:

	if (WARN_ON(res.a0 == HVC_STUB_ERR))
		return;

	vtr = res.a1;

Is it safe to use WARN_ON inside a noinstr alternative patching callback?

Alternative patching callbacks like kvm_patch_ich_vtr_el2() typically run under
stop_machine() with CPUs parked and interrupts disabled. If WARN_ON triggers,
it invokes the warning exception machinery which can attempt to acquire console
locks. If a parked CPU holds these locks, it could cause a hard deadlock.
Additionally, WARN_ON expands to instrumentable paths, directly violating
noinstr requirements.

> +
>  	/* Compute target register */
>  	oinsn = le32_to_cpu(*origptr);
>  	rd = aarch64_insn_decode_register(AARCH64_INSN_REGTYPE_RD, oinsn);

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260720132220.2143486-1-maz@kernel.org?part=4