Re: question about usb_rebind_intf

[Alan Stern]

On Fri, 17 Nov 2017, Mathias Nyman wrote:

> On 17.11.2017 19:09, Greg Kroah-Hartman wrote:
> > On Fri, Nov 17, 2017 at 09:35:51AM -0700, Jerry Snitselaar wrote:
> >> Should this skip warning that the rebind failed if device_attach
> >> is returning -EPROBE_DEFER? If I do something like 'rtcwake -m mem -s 30'
> >> on a laptop I have here I will see a couple "rebind failed: -517" messages
> >> as it comes back out of suspend. Since the device probe eventually happens
> >> once probes are not deferred wondering if this warning this be given in that
> >> case.
> >>
> >>
> >> diff --git a/drivers/usb/core/driver.c b/drivers/usb/core/driver.c
> >> index 64262a9a8829..5d3408010112 100644
> >> --- a/drivers/usb/core/driver.c
> >> +++ b/drivers/usb/core/driver.c
> >> @@ -1070,7 +1070,7 @@ static void usb_rebind_intf(struct usb_interface *intf)
> >>         if (!intf->dev.power.is_prepared) {
> >>                 intf->needs_binding = 0;
> >>                 rc = device_attach(&intf->dev);
> >> -               if (rc < 0)
> >> +               if (rc < 0 && rc != -EPROBE_DEFER)
> >>                         dev_warn(&intf->dev, "rebind failed: %d\n", rc);
> >>         }
> >> }
> > 
> > What USB driver is returning -EPROBE_DEFER to cause this to be an issue?
> > Shouldn't that really only be for "platform" drivers and the like?  USB
> > interface drivers should all be "self-contained" within reason.
> > 
> 
> I see this as well with btusb driver in resume if port was reset.
> 
> drivers/base/dd.c really_probe() returns -EPROBE_DEFER if it's called before
> device_unblock_probing() is called.
> 
> drivers/base/power/main/dpm_complete() will unlock probing after it has
> finished resuming all devices and called all the pm_ops .complete callbacks.
> 
> The usb_device_pm_ops .complete callback will try to rebind the interface driver
> if device was reset at resume, leading to them -EPROBE_DEFER.
>   
> I guess we either need to rework a few things, or remove that warning.

I agree, and the patch seems like a good idea for now.

The real fix would be to change the interface drivers by adding proper 
support for reset-resume.  Otherwise there will always be a time window 
following resume during which the interface is non-functional.

Alan Stern