Re: [PATCH part1 v5 5/7] PCI: Add pci_dummy_ops to isolate pci devicetemporarily

[Yijing Wang]

>>>> +	spin_lock_irqsave(&pci_freeze_lock, flags);
>>>> +	ops = pci_bus_set_ops(bus, &pci_dummy_ops);
>>>> +	bus->save_ops = ops;
>>>> +	spin_unlock_irqrestore(&pci_freeze_lock, flags);
>>>
>>> Against what exactly are you locking here?
>>
>> I want to use this spin lock to serialize freeze device and unfreeze device.
> 
> Yes, but against what? I am sorry I should have been more explicit.
> You are using these functions only in pci_scan_single_device()


Hi Oliver, thanks very much for your detailed analysis. My original intention to use
pci_freeze_lock to serialize pci_bus_freeze_device() and pci_bus_unfreeze_device(),
because I think these two functions maybe used in other places, although currently
only used in pci_scan_single_device().
Like:

CPU A         					         CPU B
pci_bus_freeze_device()                                pci_bus_unfreeze_device()
  pci_bus_set_ops(bus, &pci_dummy_ops);
  							 pci_bus_set_ops(bus, bus->save_ops);   ---> here, save_ops is NULL, it's bad.
  bus->save_ops = ops;


> 
> 
> CPU A					CPU B
> pci_bus_freeze_device()			wait
> bus->save_ops = ops {valid}		wait
> ...					pci_bus_freeze_device()
> wait					bus->save_ops = ops
> 					{pci_dummy_ops !}
> pci_bus_unfreeze_device()		wait
> pci_bus_set_ops(bus, bus->save_ops)
> 
> You see the problem?
> 

Yes, this is a issue, good catch. I should add a refcount to avoid this situation.

> If this function ever races with itself, the locking is useless.
> If it doesn't race with itself, the locking is not needed.
> If this function can really race with itself, you need a refcount
> for freezing.

Thanks again!




> 
> 
> 
> .
> 


-- 
Thanks!
Yijing

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/