[PATCH v3 03/12] usb: ehci-hcd: notify phy when connect change

Alan Stern stern at rowland.harvard.edu
Wed Jun 6 03:30:08 EST 2012


On Tue, 5 Jun 2012, Marek Vasut wrote:

> > > > Do the tests in the opposite order and add "unlikely":
> > > > 			if (unlikely(ehci->transceiver &&
> > > > 			
> > > > 					(pstatus & PORT_CSC)) {
> > > > 
> > > > That way it will fail more quickly on systems where it doesn't apply or
> > > > for unaffected ports.
> > > 
> > > Does this unlikely() have any effect on ARM/MIPS/PPC, where this chipidea
> > > IP is used ? Or is there some x86 device sporting this IP too?
> > 
> > I don't know what unlikely() does on architectures other than x86.
> 
> It should be optimized out to nothing on anything else but x86. On x86 it is a 
> hint for the branch predictor. Or that's what I believe it is.

That's right.  Then it should have no effect on ARM/MIPS/PPC.

> > And
> > I haven't heard of any x86 systems that would need to use this code.
> >
> > On the other hand, port-status changes don't occur very frequently.  A
> > little time penalty one way or the other won't make much difference.
> 
> I'm not opposed, just curious :)

No big deal either way.  But the order of the tests should be switched, 
because on most systems, ehci->transceiver will be NULL.

Alan Stern



More information about the devicetree-discuss mailing list