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

[Shrikanth Hegde]

On 2/17/26 3:56 PM, Christian Zigotzky wrote:
> On 17/02/26 11:17 Madhavan Srinivasan wrote:
>>
>> On 2/17/26 3:14 PM, Christian Zigotzky wrote:
>>> On 17/02/26 10:03, Shrikanth Hegde wrote:
>>>>
>>>>
>>>> On 2/17/26 2:22 PM, Christian Zigotzky wrote:
>>>>> Hi,
>>>>>
>>>>> Thank you for your patch. I tested the latest git kernel with your 
>>>>> patch today. Unfortunately the system freezes while booting the VM 
>>>>> with KVM HV.
>>>>>
>>>>> Cheers,
>>>>> Christian
>>>>>
>>>>>
>>>> Would it be possible to share the console log or boot log of KVM to the
>>>> point it freezes?
>>>
>>> Unfortunately there aren't any error messages.
>>>
>>> I also tried it with:
>>>
>>> gdb --args qemu-system-ppc64 -M ppce500 -cpu e5500 -m 1024 -kernel 
>>> uImage-7.0-preempt -drive format=raw,file=void-live-powerpc-20230317- 
>>> updated2.img,index=0,if=virtio -netdev user,id=mynet0 -device virtio- 
>>> net,netdev=mynet0 -append "rw root=/dev/vda" -device virtio-gpu-pci - 
>>> device virtio-mouse-pci -device virtio-keyboard-pci -enable-kvm -smp 4
>>>
>>> and with gdb in the guest:
>>>
>>> qemu-system-ppc64 -s -S -M ppce500 -cpu e5500 -m 1024 -kernel 
>>> uImage-7.0-preempt -drive format=raw,file=void-live-powerpc-20230317- 
>>> updated2.img,index=0,if=virtio -netdev user,id=mynet0 -device virtio- 
>>> net,netdev=mynet0 -append "rw root=/dev/vda" -device virtio-gpu-pci - 
>>> device virtio-mouse-pci -device virtio-keyboard-pci -enable-kvm -smp 4
>>>
>>> The system freezed without any error messages.

Anything on host console?

>> can you please share the config.
>> I tried kvm-hv in my power8/power9 and seems to boot fine without any 
>> issues with today's upstream
>> Maddy
> Here are the configs.
> 
> With PREEMPT: https://github.com/chzigotzky/kernels/blob/ 
> d50d259ca63dd2b3b9a9eaabc82e5cc69c8cc9f0/configs/e5500_defconfig
> 
> Without PREEMPT: https://github.com/chzigotzky/kernels/blob/main/ 
> configs/e5500_defconfig
> 
> - Christian

I looked at the code. there was one more place where cond_resched was 
there.

If possible give the below try.
(Note: Didn;t replicate your issue)

---

diff --git a/arch/powerpc/kvm/book3s_hv.c b/arch/powerpc/kvm/book3s_hv.c
index 7667563fb9ff..bf0ca49c9915 100644
--- a/arch/powerpc/kvm/book3s_hv.c
+++ b/arch/powerpc/kvm/book3s_hv.c
@@ -4825,10 +4825,13 @@ static int kvmppc_run_vcpu(struct kvm_vcpu *vcpu)
  		vc->runner = vcpu;
  		if (n_ceded == vc->n_runnable) {
  			kvmppc_vcore_blocked(vc);
-		} else if (need_resched()) {
+		} else if (need_resched() || spin_needbreak(&vc->lock)) {
  			kvmppc_vcore_preempt(vc);
  			/* Let something else run */
-			cond_resched_lock(&vc->lock);
+			spin_unlock(&vc->lock);
+			schedule();
+			//cond_resched_lock(&vc->lock);
+			spin_lock(&vc->lock);
  			if (vc->vcore_state == VCORE_PREEMPT)
  				kvmppc_vcore_end_preempt(vc);
  		} else {
@@ -4901,7 +4904,7 @@ 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;
  		}