[PATCH v1 2/2] qemu: driver: perform CPU model expansion on single CPU during baseline

[Collin Walling]

On 9/24/20 3:30 AM, Jiri Denemark wrote:
> On Wed, Sep 23, 2020 at 22:26:29 -0400, Collin Walling wrote:
>> When executing the hypervisor-cpu-baseline command and if there is only
>> a single CPU definition present in the XML file, then libvirt will
>> print an unhelpful message:
>>
>> "error: An error occurred, but the cause is unknown"
>>
>> This is due to no CPU definition ever being "baselined", since the
>> API expects at least two CPU models.
>>
>> Let's fix this by performing a CPU model expansion on the single CPU
>> definition and returning the result to the caller.
> 
> Ah, so when host-passthrough CPU is passed to the baseline API, we
> should report an error. So this patch is actually trying to handle
> single CPU definition with a non-empty <model> element specified. Good,
> as the API is of course supposed to work in this case. It should
> basically return the CPU definition passed to it with unsupported
> features disabled.
> 
> Normally, expansion should only be done when requested by the
> corresponding API flag. The simplest fix would be just returning the
> original CPU definition if only one was passed to baseline. But the
> result might not be the correct one as it could include unsupported

Right. In fact, if the model cannot be baselined (due to only a single
CPU provided to the API), and we DON'T call expansion here, the result
will just be the same verbatim XML that the user provided... which may
or may not be correct...

Calling expansion on the model will guarantee a sane response.

> features. So is static expansion the thing that will return the expected
> result? If so, this patch is mostly correct...

Precisely. A static expansion will return the CPU model and a
non-exhaustive list of features associated with the model.

> 
>> Signed-off-by: Collin Walling <walling at linux.ibm.com>
>> ---
>>  src/qemu/qemu_driver.c | 17 +++++++++--------
>>  1 file changed, 9 insertions(+), 8 deletions(-)
>>
>> diff --git a/src/qemu/qemu_driver.c b/src/qemu/qemu_driver.c
>> index 427d2419f3..97a960a769 100644
>> --- a/src/qemu/qemu_driver.c
>> +++ b/src/qemu/qemu_driver.c
>> @@ -12472,6 +12472,7 @@ qemuConnectCPUModelBaseline(virQEMUCapsPtr qemuCaps,
>>      g_autoptr(qemuProcessQMP) proc = NULL;
>>      g_autoptr(virCPUDef) baseline = NULL;
>>      qemuMonitorCPUModelInfoPtr result = NULL;
>> +    qemuMonitorCPUModelExpansionType expansion_type;
>>      size_t i;
>>  
>>      if (!(proc = qemuProcessQMPNew(virQEMUCapsGetBinary(qemuCaps),
>> @@ -12503,15 +12504,15 @@ qemuConnectCPUModelBaseline(virQEMUCapsPtr qemuCaps,
>>              return NULL;
>>      }
>>  
>> -    if (expand_features) {
> 
> While full expansion has to always be performed on the baseline result,
> static expansion should only be called when ncpus == 1. In other words,
> rather than removing this condition, you should change it to
> 
>        if (expand_features || ncpus == 1) {
> 

Sounds good. I'll replace patch #1 with something that simply checks for
a missing model name within baseline and prints a helpful error, and
make the appropriate changes to this one.

>> -        if (qemuMonitorGetCPUModelExpansion(proc->mon,
>> -                                            QEMU_MONITOR_CPU_MODEL_EXPANSION_FULL,
>> -                                            baseline, true, false, &result) < 0)
>> -            return NULL;
>> +    expansion_type = expand_features ? QEMU_MONITOR_CPU_MODEL_EXPANSION_FULL
>> +                                     : QEMU_MONITOR_CPU_MODEL_EXPANSION_STATIC;
>>  
>> -        if (qemuConnectStealCPUModelFromInfo(baseline, &result) < 0)
>> -            return NULL;
>> -    }
>> +    if (qemuMonitorGetCPUModelExpansion(proc->mon, expansion_type,
>> +                                        baseline, true, false, &result) < 0)
>> +        return NULL;
>> +
>> +    if (qemuConnectStealCPUModelFromInfo(baseline, &result) < 0)
>> +        return NULL;
>>  
>>      return g_steal_pointer(&baseline);
>>  }
> 
> Jirka
> 

Thanks!

-- 
Regards,
Collin

Stay safe and stay healthy