Hi,
On Mon, Oct 10, 2011 at 08:15:29PM -0400, Alan Stern wrote:
> > > > > In fact we do already have sibling lists. They are maintained as part
> > > > > of the device_private structure. What we are missing is a
> > > > > device_for_each_sibling() routine. It could be added pretty easily; it
> > > > > would be similar to device_for_each_child().
> > > >
> > > > care to point out where is ?
> > > >
> > > > 68 struct device_private {
> > > > 69 struct klist klist_children;
> > > > 70 struct klist_node knode_parent;
> > > -------------^ Here. The "parent" in the name refers to where the
> > > head of the list is stored.
> > >
> > > > 71 struct klist_node knode_driver;
> > > > 72 struct klist_node knode_bus;
> > > > 73 void *driver_data;
> > > > 74 struct device *device;
> > > > 75 };
> > >
> > > From device_add():
> > >
> > > if (parent)
> > > klist_add_tail(&dev->p->knode_parent,
> > > &parent->p->klist_children);
> >
> > that's a parent -> child relationship. What we have on this case is:
> >
> > -------------- ---------------
> > | | | | |\
> > | UHH | clocks, etc | USBTLL | | |
> > | | <==========> | | <======> | | <====> ports
> > | ------- | | (Transceiver- | | |
> > | | EHCI | | | less Link) | |/
> > | ------- | | | Port MUX
> > | | | |
> > | ------- | | |
> > | | OHCI | | | |
> > | ------- | | |
> > | | | |
> > -------------- ---------------
> >
> > It doesn't shown here, but the TLL link is completely optional. It's
> > mainly used for modem integration, IIRC. Still, if we're using TLL, EHCI
> > and OHCI will depend on a clock provided by the USBTLL block.
> >
> > Clearly, USBTLL isn't either a parent of UHH, nor a parent of EHCI/OHCI
> > blocks. We can, from a code perspective, make USBTLL into a parent of
> > UHH to make things simpler, but this will mean that calling
> > pm_runtime_get() will also unconditionaly turn on TLL clock, unless we
> > add some nasty hacks to allow TLL know if *HCI port is in TLL mode.
> >
> > That's why I decided for making TLL and UHH siblings, because that's a
> > closer relationship than parent-child.
> >
> > Can you see the problem now ?
>
> Okay, now I understand better. The word "sibling" implies that the two
> objects have the same parent, so a different word would describe this
> relationship better. Something like "friend" or "associate".
>
> Or maybe, following Paul's suggestion, the driver core doesn't have to
> be changed at all.
I see... I just thought that if there are other similar cases, it might
make sense to have a more generic way to make those two devices talk to
each other. But if you all agree that an EXPORT_SYMBOL_GPL() is enough,
then it's ok ;-)
thanks
--
balbi
Attachment:
signature.asc
Description: Digital signature
