Re: [PATCH 1/1] driver,usb: Fix a warning in uhci-hcd driver

[ZhenHua]

On 04/25/2013 10:54 PM, Alan Stern wrote:
> On Thu, 25 Apr 2013, ZhenHua wrote:
>
> > > > > +#define UHCI_SUSPENDRH_RETRY_MAX      10
> > > > > +#define UHCI_SUSPENDRH_RETRY_DELAY    100
> > > Why is the delay set to 100 us?  Isn't that excessively large?  How
> > > long does it take for this controller to go into suspend?
> > This controller will take about 200~400 us, but I am not sure how long
> > other devices will take.
> > I set interval to 100 us,  so it will save more time.
> A 400-us delay is fairly long.  It would be better to avoid it
The device needs about 200~400 us to get stopped, not OS.
For other devices, it will not wait.
> entirely.
>
> > > Why are these variables u16?  Why not int?
> > uhci_readw will return u16.
> That's not a good reason, since u16 fits perfectly well inside an
> int.  But never mind...
>
> > > Anyway, a better approach would be not to add a delay loop at all.
> > > Instead, change this test:
> > >
> > > 	if (!auto_stop && !(uhci_readw(uhci, USBSTS) & USBSTS_HCH)) {
> > > 		uhci->rh_state = UHCI_RH_SUSPENDING;
> > > 		spin_unlock_irq(&uhci->lock);
> > > 		msleep(1);
> > > 		spin_lock_irq(&uhci->lock);
> > > 		if (uhci->dead)
> > > 			return;
> > > 	}
> > >
> > > When the iLo controller is present, make the "if" statement always
> > > succeed.  Then you'll get a whole 1-ms delay.
> > This will cause more operation and more time for other devices.
> Actually what I wrote was wrong anyway.  I forgot that when auto_stop
> is set, the routine is not allowed to sleep.
>
> A better way to solve your problem is to change uhci_hub_status_data().
> In the UHCI_RH_RUNNING_NODEVS case, change the line that says
>
> 		else if (time_after_eq(jiffies, uhci->auto_stop_time))
>
> to
>
> 		else if (time_after_eq(jiffies, uhci->auto_stop_time) &&
> 				!uhci->no_auto_stops)
>
> where uhci->no_auto_stops is a new bitflag that you set inside
> uhci_pci_init() if you detect that the controller is an iLo virtual
> UHCI controller.
>
> This way there will always be a 1-ms delay, so the slow controller will
> suspend successfully.  And other types of host controllers won't be
> affected, because the no_auto_stops flag won't get set for them.
>
> Alan Stern

I  think it is a good idea, and the logic of the code may be more 
clear.  I will do some test on my system.

Thanks
Zhen-Hua

--
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/