Re: [qemu-s390x] [Qemu-devel] [PATCH v1 2/4] s390x/zpci: use hotplug_dev instead of looking up the host bridge

[David Hildenbrand]

On 08.11.18 12:07, Cornelia Huck wrote:
> On Wed, 7 Nov 2018 15:28:31 -0500
> Collin Walling <address@hidden> wrote:
> 
>> On 11/5/18 6:50 AM, David Hildenbrand wrote:
>>> On 05.11.18 12:40, Christian Borntraeger wrote:  
>>>>
>>>>
>>>> On 11/05/2018 12:37 PM, David Hildenbrand wrote:  
>>>>> On 05.11.18 12:21, Cornelia Huck wrote:  
>>>>>> On Mon,  5 Nov 2018 12:03:11 +0100
>>>>>> David Hildenbrand <address@hidden> wrote:
>>>>>>  
>>>>>>> We directly have it in our hands.
>>>>>>>
>>>>>>> Signed-off-by: David Hildenbrand <address@hidden>
>>>>>>> ---
>>>>>>>  hw/s390x/s390-pci-bus.c | 4 ++--
>>>>>>>  1 file changed, 2 insertions(+), 2 deletions(-)
>>>>>>>
>>>>>>> diff --git a/hw/s390x/s390-pci-bus.c b/hw/s390x/s390-pci-bus.c
>>>>>>> index 1eaae3aca6..68660eac74 100644
>>>>>>> --- a/hw/s390x/s390-pci-bus.c
>>>>>>> +++ b/hw/s390x/s390-pci-bus.c
>>>>>>> @@ -814,9 +814,9 @@ static bool s390_pci_alloc_idx(S390pciState *s, 
>>>>>>> S390PCIBusDevice *pbdev)
>>>>>>>  static void s390_pcihost_plug(HotplugHandler *hotplug_dev, DeviceState 
>>>>>>> *dev,
>>>>>>>                                Error **errp)
>>>>>>>  {
>>>>>>> +    S390pciState *s = S390_PCI_HOST_BRIDGE(hotplug_dev);
>>>>>>>      PCIDevice *pdev = NULL;
>>>>>>>      S390PCIBusDevice *pbdev = NULL;
>>>>>>> -    S390pciState *s = s390_get_phb();
>>>>>>>  
>>>>>>>      if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_BRIDGE)) {
>>>>>>>          BusState *bus;
>>>>>>> @@ -924,11 +924,11 @@ static void s390_pcihost_timer_cb(void *opaque)
>>>>>>>  static void s390_pcihost_unplug(HotplugHandler *hotplug_dev, 
>>>>>>> DeviceState *dev,
>>>>>>>                                  Error **errp)
>>>>>>>  {
>>>>>>> +    S390pciState *s = S390_PCI_HOST_BRIDGE(hotplug_dev);
>>>>>>>      PCIDevice *pci_dev = NULL;
>>>>>>>      PCIBus *bus;
>>>>>>>      int32_t devfn;
>>>>>>>      S390PCIBusDevice *pbdev = NULL;
>>>>>>> -    S390pciState *s = s390_get_phb();
>>>>>>>  
>>>>>>>      if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_BRIDGE)) {
>>>>>>>          error_setg(errp, "PCI bridge hot unplug currently not 
>>>>>>> supported");  
>>>>>>
>>>>>> Not sure whether that is an improvement (s390_get_phb() caches the
>>>>>> value, and is called from multiple other places as well.)
>>>>>>  
>>>>>
>>>>> Looking up a variable that is directly passed as an argument doesn't
>>>>> look clean to me.  
>>>>
>>>> I think there was a reason for this caching, namely that qom resolution can
>>>> be quite expensive. For the hotplug case this obviously does not matter but
>>>> for all the other cases it might. So do we really want to have different 
>>>> places use different methods?
>>>>  
>>>
>>> Caching resolution is fine (as that is expensive), caching a downcast is
>>> as far as I remember not necessary. Especially, as you said, for hotplug
>>> handlers.
> 
> Yes, the complete QOM cast was the expensive thing AFAIR.
> 
>>>
>>> Anyhow, if there are strong feelings to this change, I can drop this
>>> patch. There are certainly more important things to do in zPCI hotplug code.
>>>
>>>   
>>
>> Truthfully, I'm not in favor of one over the other. As long as the device 
>> handlers
>> are consistent, I think either is fine.
> 
> I don't feel *that* strong about this change here, either :) Your call.
> 
>>
>> However, it would be nice if at some point during plug we cache the PHB 
>> somewhere.
>> That would be some sort of best-of-both-worlds approach.
> 
> Not sure if caching-from-a-downcast would be conceptionally clean. I'd
> vote for either taking this patch or dropping it completely.
> 

Yes, caching at that point, I would consider premature optimization.

I'll drag this patch along and let the maintainers decide what to do :)

(later we should forward the actual value to e.g. event injection code,
so we don't have to look it up multiple times during one call)

-- 

Thanks,

David / dhildenb