On ppc64, the PowerVM hypervisor runs with limited memory and a VCPU
creation during hotplug may fail during kvm_ioctl for KVM_CREATE_VCPU,
leading to termination of guest since errp is set to &error_fatal while
calling kvm_init_vcpu. This unexpected behaviour can be avoided by
pre-creating and parking vcpu on success or return error otherwise.
This enables graceful error delivery for any vcpu hotplug failures while
the guest can keep running.

Also introducing KVM AccelCPUClass to init cpu_target_realize for kvm.

Tested OK by repeatedly doing a hotplug/unplug of vcpus as below:

 #virsh setvcpus hotplug 40
 #virsh setvcpus hotplug 70
error: internal error: unable to execute QEMU command 'device_add':
kvmppc_cpu_realize: vcpu hotplug failed with -12

Reported-by: Anushree Mathur <anushree.mathur@linux.vnet.ibm.com>
Suggested-by: Shivaprasad G Bhat <sbhat@linux.ibm.com>
Suggested-by: Vaibhav Jain <vaibhav@linux.ibm.com>
Signed-off by: Harsh Prateek Bora <harshpb@linux.ibm.com>
Tested-by: Anushree Mathur <anushree.mathur@linux.vnet.ibm.com>
---
 target/ppc/kvm.c | 41 +++++++++++++++++++++++++++++++++++++++++
 1 file changed, 41 insertions(+)

diff --git a/target/ppc/kvm.c b/target/ppc/kvm.c
index 63930d4a77..8e5a7c3d2d 100644
--- a/target/ppc/kvm.c
+++ b/target/ppc/kvm.c
@@ -48,6 +48,8 @@
 #include "qemu/mmap-alloc.h"
 #include "elf.h"
 #include "sysemu/kvm_int.h"
+#include "accel/kvm/kvm-cpus.h"
+#include "hw/core/accel-cpu.h"
 
 #define PROC_DEVTREE_CPU      "/proc/device-tree/cpus/"
 
@@ -2339,6 +2341,25 @@ static void alter_insns(uint64_t *word, uint64_t flags, bool on)
     }
 }
 
+static bool kvmppc_cpu_realize(CPUState *cs, Error **errp)
+{
+    int ret;
+    const char *vcpu_str = (cs->parent_obj.hotplugged == true) ?
+                           "hotplug" : "create";
+    cs->cpu_index = cpu_get_free_index();
+
+    POWERPC_CPU(cs)->vcpu_id = cs->cpu_index;
+
+    /* create and park to fail gracefully in case vcpu hotplug fails */
+    ret = kvm_create_and_park_vcpu(cs);
+    if (ret) {
+        error_setg(errp, "%s: vcpu %s failed with %d",
+                         __func__, vcpu_str, ret);
+        return false;
+    }
+    return true;
+}
+
 static void kvmppc_host_cpu_class_init(ObjectClass *oc, void *data)
 {
     PowerPCCPUClass *pcc = POWERPC_CPU_CLASS(oc);
@@ -2959,3 +2980,23 @@ void kvmppc_set_reg_tb_offset(PowerPCCPU *cpu, int64_t tb_offset)
 void kvm_arch_accel_class_init(ObjectClass *oc)
 {
 }
+
+static void kvm_cpu_accel_class_init(ObjectClass *oc, void *data)
+{
+    AccelCPUClass *acc = ACCEL_CPU_CLASS(oc);
+
+    acc->cpu_target_realize = kvmppc_cpu_realize;
+}
+
+static const TypeInfo kvm_cpu_accel_type_info = {
+    .name = ACCEL_CPU_NAME("kvm"),
+
+    .parent = TYPE_ACCEL_CPU,
+    .class_init = kvm_cpu_accel_class_init,
+    .abstract = true,
+};
+static void kvm_cpu_accel_register_types(void)
+{
+    type_register_static(&kvm_cpu_accel_type_info);
+}
+type_init(kvm_cpu_accel_register_types);
-- 
2.39.3

On Thu May 23, 2024 at 5:26 PM AEST, Harsh Prateek Bora wrote:
> On ppc64, the PowerVM hypervisor runs with limited memory and a VCPU
> creation during hotplug may fail during kvm_ioctl for KVM_CREATE_VCPU,
> leading to termination of guest since errp is set to &error_fatal while
> calling kvm_init_vcpu. This unexpected behaviour can be avoided by
> pre-creating and parking vcpu on success or return error otherwise.
> This enables graceful error delivery for any vcpu hotplug failures while
> the guest can keep running.
>
> Also introducing KVM AccelCPUClass to init cpu_target_realize for kvm.
>
> Tested OK by repeatedly doing a hotplug/unplug of vcpus as below:
>
>  #virsh setvcpus hotplug 40
>  #virsh setvcpus hotplug 70
> error: internal error: unable to execute QEMU command 'device_add':
> kvmppc_cpu_realize: vcpu hotplug failed with -12
>
> Reported-by: Anushree Mathur <anushree.mathur@linux.vnet.ibm.com>
> Suggested-by: Shivaprasad G Bhat <sbhat@linux.ibm.com>
> Suggested-by: Vaibhav Jain <vaibhav@linux.ibm.com>
> Signed-off by: Harsh Prateek Bora <harshpb@linux.ibm.com>
> Tested-by: Anushree Mathur <anushree.mathur@linux.vnet.ibm.com>
> ---
>  target/ppc/kvm.c | 41 +++++++++++++++++++++++++++++++++++++++++
>  1 file changed, 41 insertions(+)
>
> diff --git a/target/ppc/kvm.c b/target/ppc/kvm.c
> index 63930d4a77..8e5a7c3d2d 100644
> --- a/target/ppc/kvm.c
> +++ b/target/ppc/kvm.c
> @@ -48,6 +48,8 @@
>  #include "qemu/mmap-alloc.h"
>  #include "elf.h"
>  #include "sysemu/kvm_int.h"
> +#include "accel/kvm/kvm-cpus.h"
> +#include "hw/core/accel-cpu.h"
>  
>  #define PROC_DEVTREE_CPU      "/proc/device-tree/cpus/"
>  
> @@ -2339,6 +2341,25 @@ static void alter_insns(uint64_t *word, uint64_t flags, bool on)
>      }
>  }
>  
> +static bool kvmppc_cpu_realize(CPUState *cs, Error **errp)
> +{
> +    int ret;
> +    const char *vcpu_str = (cs->parent_obj.hotplugged == true) ?
> +                           "hotplug" : "create";
> +    cs->cpu_index = cpu_get_free_index();
> +
> +    POWERPC_CPU(cs)->vcpu_id = cs->cpu_index;
> +
> +    /* create and park to fail gracefully in case vcpu hotplug fails */

The only thing that wasn't immediately clear to me is that in the
machine init path, this results in qemu termination, and in the
hotplug path it results in graceful hotplug failure. That is the
behaviour we want, maybe just expand the comment slightly to be
more explicit about it. E.g.,

> +    ret = kvm_create_and_park_vcpu(cs);
> +    if (ret) {
           /*
	    * This causes QEMU to terminate if initial CPU creation
	    * fails, and CPU hotplug failure if the error happens
	    * there.
	    */
> +        error_setg(errp, "%s: vcpu %s failed with %d",
> +                         __func__, vcpu_str, ret);
> +        return false;
> +    }
> +    return true;
> +}
> +

Otherwise looks good.

Reviewed-by: Nicholas Piggin <npiggin@gmail.com>

>  static void kvmppc_host_cpu_class_init(ObjectClass *oc, void *data)
>  {
>      PowerPCCPUClass *pcc = POWERPC_CPU_CLASS(oc);
> @@ -2959,3 +2980,23 @@ void kvmppc_set_reg_tb_offset(PowerPCCPU *cpu, int64_t tb_offset)
>  void kvm_arch_accel_class_init(ObjectClass *oc)
>  {
>  }
> +
> +static void kvm_cpu_accel_class_init(ObjectClass *oc, void *data)
> +{
> +    AccelCPUClass *acc = ACCEL_CPU_CLASS(oc);
> +
> +    acc->cpu_target_realize = kvmppc_cpu_realize;
> +}
> +
> +static const TypeInfo kvm_cpu_accel_type_info = {
> +    .name = ACCEL_CPU_NAME("kvm"),
> +
> +    .parent = TYPE_ACCEL_CPU,
> +    .class_init = kvm_cpu_accel_class_init,
> +    .abstract = true,
> +};
> +static void kvm_cpu_accel_register_types(void)
> +{
> +    type_register_static(&kvm_cpu_accel_type_info);
> +}
> +type_init(kvm_cpu_accel_register_types);