Re: [Linux PPC] KVM PR and KVM HV do not work if the kernel was compiled with PREEMPT

[Shrikanth Hegde <sshegde@linux.ibm.com>]

On 2/14/26 7:32 PM, Christian Zigotzky wrote:
> Hello,
> 
> KVM PR and KVM HV do not work if the kernel was compiled with PREEMPT.
> 
> The entire FSL Cyrus+ board freezes when using KVM HV with PREEMPT.
> 
> The guest kernel doesn't boot if we use KVM PR with a PREEMPT kernel on 
> the PA Semi Nemo board.
> 
> We were previously able to disable PREEMPT in the kernel configuration, 
> but the latest git kernels now enable it by default and it is no longer 
> possible to disable it.
> 
> I created a patch for disabling PREEMPT today. [1]
> 
> Is it possible to let us decide whether to activate PREEMPT or not?
> 
> Thanks in advance,
> 
> Christian
> 
> [1] https://raw.githubusercontent.com/chzigotzky/kernels/ 
> a74fa6179eaeafcea7ad89f0e61c30ace038daf2/patches/X1000/ 
> Kconfig.preempt.patch
> [2] Bug report: https://github.com/chzigotzky/kernels/issues/19


Hi.

Do you have any trace why it is stuck? That would be useful.



My initial take is cond_resched is a nop. So we might be stuck there.
Eventually it should have come out though.

Could you please give the below patch a try and let me know?
Note: This likely still needs lazy bit handling. So keep in preempt=full.
(Not tested)


diff --git a/arch/powerpc/kvm/book3s_hv.c b/arch/powerpc/kvm/book3s_hv.c
index 7667563fb9ff..fe215d1177fe 100644
--- a/arch/powerpc/kvm/book3s_hv.c
+++ b/arch/powerpc/kvm/book3s_hv.c
@@ -4901,7 +4901,7 @@ int kvmhv_run_single_vcpu(struct kvm_vcpu *vcpu, u64 time_limit,
         }
  
         if (need_resched())
-               cond_resched();
+               schedule();
  
         kvmppc_update_vpas(vcpu);
  
diff --git a/arch/powerpc/kvm/powerpc.c b/arch/powerpc/kvm/powerpc.c
index 9a89a6d98f97..54963c1d8b58 100644
--- a/arch/powerpc/kvm/powerpc.c
+++ b/arch/powerpc/kvm/powerpc.c
@@ -86,7 +86,7 @@ int kvmppc_prepare_to_enter(struct kvm_vcpu *vcpu)
         while (true) {
                 if (need_resched()) {
                         local_irq_enable();
-                       cond_resched();
+                       schedule();
                         hard_irq_disable();
                         continue;
                 }