Re: [PATCH v4] target/s390x/kvm: Enable adapter interruption suppression again

[Thomas Huth]

On 21/01/2020 17.48, David Hildenbrand wrote:
> On 21.01.20 17:33, Thomas Huth wrote:
>> The AIS feature has been disabled late in the v2.10 development cycle since
>> there were some issues with migration (see commit 3f2d07b3b01ea61126b -
>> "s390x/ais: for 2.10 stable: disable ais facility"). We originally wanted
>> to enable it again for newer machine types, but apparently we forgot to do
>> this so far. Let's do it for the new s390-ccw-virtio-5.0 machine now.
>>
>> While at it, also add a more verbose comment why we need the *_allowed()
>> wrappers in s390-virtio-ccw.c.
>>
>> Buglink: https://bugzilla.redhat.com/show_bug.cgi?id=1756946
>> Reviewed-by: David Hildenbrand <address@hidden>
>> Tested-by: Matthew Rosato <address@hidden>
>> Signed-off-by: Thomas Huth <address@hidden>
>> ---
>>  v4: Use kvm_kernel_irqchip_allowed() for avoiding problems when running
>>      with -machine s390-ccw-virtio,kernel_irqchip=off
>>
>>  hw/s390x/s390-virtio-ccw.c         | 20 +++++++++++++++++---
>>  include/hw/s390x/s390-virtio-ccw.h |  3 +++
>>  target/s390x/kvm.c                 |  9 ++++++---
>>  3 files changed, 26 insertions(+), 6 deletions(-)
>>
>> diff --git a/hw/s390x/s390-virtio-ccw.c b/hw/s390x/s390-virtio-ccw.c
>> index e0e28139a2..76254e8447 100644
>> --- a/hw/s390x/s390-virtio-ccw.c
>> +++ b/hw/s390x/s390-virtio-ccw.c
>> @@ -452,6 +452,7 @@ static void ccw_machine_class_init(ObjectClass *oc, void 
>> *data)
>>      s390mc->cpu_model_allowed = true;
>>      s390mc->css_migration_enabled = true;
>>      s390mc->hpage_1m_allowed = true;
>> +    s390mc->kvm_ais_allowed = true;
>>      mc->init = ccw_init;
>>      mc->reset = s390_machine_reset;
>>      mc->hot_add_cpu = s390_hot_add_cpu;
>> @@ -505,6 +506,14 @@ static inline void machine_set_dea_key_wrap(Object 
>> *obj, bool value,
>>  
>>  static S390CcwMachineClass *current_mc;
>>  
>> +/*
>> + * Get the class of the s390-ccw-virtio machine that is currently in use.
>> + * Note: libvirt is using the "none" machine to probe for the features of 
>> the
>> + * host CPU, so in case this is called with the "none" machine, the function
>> + * returns the TYPE_S390_CCW_MACHINE base class. In this base class, all the
>> + * various "*_allowed" variables are enabled, so that the *_allowed() 
>> wrappers
>> + * below return the correct default value for the "none" machine.
>> + */
>>  static S390CcwMachineClass *get_machine_class(void)
>>  {
>>      if (unlikely(!current_mc)) {
>> @@ -521,22 +530,24 @@ static S390CcwMachineClass *get_machine_class(void)
>>  
>>  bool ri_allowed(void)
>>  {
>> -    /* for "none" machine this results in true */
>>      return get_machine_class()->ri_allowed;
>>  }
>>  
>>  bool cpu_model_allowed(void)
>>  {
>> -    /* for "none" machine this results in true */
>>      return get_machine_class()->cpu_model_allowed;
>>  }
>>  
>>  bool hpage_1m_allowed(void)
>>  {
>> -    /* for "none" machine this results in true */
>>      return get_machine_class()->hpage_1m_allowed;
>>  }
>>  
>> +bool kvm_ais_allowed(void)
>> +{
>> +    return get_machine_class()->kvm_ais_allowed;
>> +}
>> +
>>  static char *machine_get_loadparm(Object *obj, Error **errp)
>>  {
>>      S390CcwMachineState *ms = S390_CCW_MACHINE(obj);
>> @@ -658,8 +669,11 @@ static void 
>> ccw_machine_4_2_instance_options(MachineState *machine)
>>  
>>  static void ccw_machine_4_2_class_options(MachineClass *mc)
>>  {
>> +    S390CcwMachineClass *s390mc = S390_MACHINE_CLASS(mc);
>> +
>>      ccw_machine_5_0_class_options(mc);
>>      compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
>> +    s390mc->kvm_ais_allowed = false;
>>  }
>>  DEFINE_CCW_MACHINE(4_2, "4.2", false);
>>  
>> diff --git a/include/hw/s390x/s390-virtio-ccw.h 
>> b/include/hw/s390x/s390-virtio-ccw.h
>> index 8aa27199c9..e3ba3b88b1 100644
>> --- a/include/hw/s390x/s390-virtio-ccw.h
>> +++ b/include/hw/s390x/s390-virtio-ccw.h
>> @@ -40,6 +40,7 @@ typedef struct S390CcwMachineClass {
>>      bool cpu_model_allowed;
>>      bool css_migration_enabled;
>>      bool hpage_1m_allowed;
>> +    bool kvm_ais_allowed;
>>  } S390CcwMachineClass;
>>  
>>  /* runtime-instrumentation allowed by the machine */
>> @@ -48,6 +49,8 @@ bool ri_allowed(void);
>>  bool cpu_model_allowed(void);
>>  /* 1M huge page mappings allowed by the machine */
>>  bool hpage_1m_allowed(void);
>> +/* adapter-interrupt suppression allowed by the machine? */
>> +bool kvm_ais_allowed(void);
>>  
>>  /**
>>   * Returns true if (vmstate based) migration of the channel subsystem
>> diff --git a/target/s390x/kvm.c b/target/s390x/kvm.c
>> index 15260aeb9a..1602a2c33d 100644
>> --- a/target/s390x/kvm.c
>> +++ b/target/s390x/kvm.c
>> @@ -365,10 +365,13 @@ int kvm_arch_init(MachineState *ms, KVMState *s)
>>      /*
>>       * The migration interface for ais was introduced with kernel 4.13
>>       * but the capability itself had been active since 4.12. As migration
>> -     * support is considered necessary let's disable ais in the 2.10
>> -     * machine.
>> +     * support is considered necessary, we only try to enable this for
>> +     * newer machine types if KVM_CAP_S390_AIS_MIGRATION is available.
>>       */
>> -    /* kvm_vm_enable_cap(s, KVM_CAP_S390_AIS, 0); */
>> +    if (kvm_ais_allowed() && kvm_kernel_irqchip_allowed() &&
>> +        kvm_check_extension(s, KVM_CAP_S390_AIS_MIGRATION)) {
>> +        kvm_vm_enable_cap(s, KVM_CAP_S390_AIS, 0);
>> +    }
> 
> I just remembered that the availability of CPU features should in
> general not depend on the selected machine. We only have compat handling
> for pre-cpu-model-support machines (e.g., vx).
> 
> The issue is, that the probed host cpu model is otherwise not guaranteed
> to run om a selected machine and you get misleading errors.
> 
> Can someone remind me why we need kvm_ais_allowed() at all and cannot
> simply rely on cpu model checks to properly handle this (like all other
> features)?

Sorry, I don't quite get what you mean here. Which other features are
you talking about? KVM_CAP_S390_RI and KVM_CAP_S390_GS are enabled in a
very similar way...

 Thomas