Re: [PATCH v8 04/20] KVM: selftests: Initialize the default/global pRNG during kvm_selftest_init()

[Sean Christopherson <seanjc@google.com>]

On Fri, Jun 26, 2026, sashiko-bot@kernel.org wrote:
> > diff --git a/tools/testing/selftests/kvm/lib/kvm_util.c b/tools/testing/selftests/kvm/lib/kvm_util.c
> > index 875030c22d07b..1016865d3f7a2 100644
> > --- a/tools/testing/selftests/kvm/lib/kvm_util.c
> > +++ b/tools/testing/selftests/kvm/lib/kvm_util.c
> > @@ -24,6 +24,13 @@ u32 kvm_random_seed;
> >  struct kvm_random_state kvm_rng;
> >  static u32 last_kvm_seed;
> >  
> > +static void kvm_seed_rng(u32 seed)
> > +{
> > +	kvm_random_seed = last_kvm_seed = seed;
> > +	pr_info("Random seed: 0x%x\n", kvm_random_seed);
> > +	kvm_rng = new_kvm_random_state(kvm_random_seed);
> > +}
> > +
> 
> [Severity: Medium]
> Does making this function static prevent tests from reseeding the PRNG
> early?
> 
> If a test draws from kvm_rng before __vm_create() is called (for example,
> generating gsi and vector at the start of main() in irq_test), the test
> will use the initial seed from the constructor.
> 
> Since kvm_seed_rng() is static and only automatically reseeded during
> __vm_create(), how can developers override the seed to reproduce CI failures
> for these pre-VM random variables?

By hacking selftests.  Yes, it's ugly, but AFAIK we haven't yet had a failure
that's actually been dependent on the RNG seed, and plumbing in a way to set the
seed for every test would be painful.