Re: [RFC] New system device API

On Mon, 9 Jun 2003 09:18:03 -0700 (PDT), Patrick Mochel wrote:
>> >-EXPORT_SYMBOL(device_lapic);
>> 
>> Why did you ignore the 'not static' comment, and why remove
>> the EXPORT? They're there for a reason...
>
>Because it only appeared to be there to be able to set the parent device
>for hierarchically dependent devices. Assuming that is not needed, then 
>the EXPORT_SYMBOL() shouldn't be needed, and that device can be declared 
>statically. Right? 

Yes, they were only there so a child device could set its
parent pointer.

>> Unless I'm missing something, you've just broken the hierarchical
>> relationship that exists between the local APIC device and its
>> client devices (NMI watchdog, oprofile [which you didn't convert],
>> and perfctr [not merged into Linus' tree]).
>
>I'm aware of the necessary ancstral relationship, and I should have 
>mentioned this in the first email. Proper ancestral order is maintained by 
>virtue of the fact that child devices are registered after parent devices. 
>Because of this, they are inserted into the list of system devices after 
>their parents, so by walking the list in reverse order, you are guaranteed 
>to suspend/shutdown the devices in the right order. 

Ok. I'm used to having this spelled out explicitly, but walking the
list in insertion order / reverse insertion order should work too.

/Mikael