RE: [Xen-devel] [PATCH 1/2] Xen acpi pad implement

[Liu, Jinsong]

Konrad Rzeszutek Wilk wrote:
> On Tue, Oct 30, 2012 at 03:18:59PM +0000, Liu, Jinsong wrote:
>> Konrad Rzeszutek Wilk wrote:
>>>>> +config XEN_ACPI_PAD_STUB
>>>>> +	bool
>>>>> +	depends on XEN_DOM0 && X86_64 && ACPI
>>>>> +	default n
>>>>> +
>>>> 
>>>> This Kconfig is pointless, if CONFIG_XEN_ACPI_PAD_STUB = n, native
>>>> pad would successfully registerred, and then mwait #UD (we would
>>>> revert df88b2d96e36d9a9e325bfcd12eb45671cbbc937, right?). So xen
>>>> stub logic should unconditionally built-in kernel.
>>> 
>>> 
>>> Potentially. Keep in mind that there is no need to built this if the
>>> kernel is not built with ACPI.
>> 
>> Sure, 'obj-$(CONFIG_XEN_DOM0) +=' is enough.
>> (XEN_DOM0 depends on ACPI).
>> 
>>>>> +subsys_initcall(xen_acpi_pad_stub_init);
>>>> 
>>>> I'm still confused. In this way there are xen-acpi-pad-stub.c and
>>>> xen-acpi-pad.c, and you want to let xen-acpi-pad loaded as module,
>>>> right? how can xen-acpi-pad logic work when it was insmoded?
>>> 
>>> Via the register/unregister calls that this provides? Or does ACPI
>>> bus drivers get immediately called once the call
>>> acpi_bus_register_driver? 
>> 
>> But when xen stub driver registerred, real xen pad ops has not been
>> hooked to stub ops. 
>> 
>>> 
>>> Or can one 'poke' the 'add' and 'remove' calls so that once the
>>> "true" PAD driver is loaded it will restart the ops->add call?
>> 
>> I think we'd better not to use xen pad stub approach. Technically
>> it's complicated, say, how to match xen_acpi_pad driver w/ pad
>> device? when and how to invoke .add method? how to avoid native pad
>> loading risk? etc. I didn't find its obivous advantages, so how
>> about keep simpler approach?    
> 
> OK. Lets go with that one for right now. The one thing I don't like
> about it is that it is built-in. It would be so much better if it was
> a module, but I am just not sure how to make that work with the
> acpi_pad. Unless the mwait CPUID capability is not exported (so drop
> your patch 1 that reverts a commit). 

Yes, but considering mwait case and its benefit to hypervisor, xen pad built-in kernel seems to be the price we have to pay. At least currently I didn't find better way.

> 
> Perhaps there is a way to make it a module/built-in with some Kconfig
> magic options? Say if ACPI_PROCESSOR_AGGREGATOR is not set, then we
> can 
> make it a module. But if ACPI_PROCESSOR_AGGREGATOR=m|y then we do it
> as built-in?

Hmm, seems it's hard to express this logic in Kconfig, and that would make xen pad logic very complicated :)

Thanks,
Jinsong

--
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/