Re: [PATCH] usb: avoid PM error messages during resume if a device was disconnected

["Rafael J. Wysocki" <rjw@sisk.pl>]

First, sorry for the delayed response.  Frans has just reminded me about this
thread.

On Tuesday 24 March 2009, Alan Stern wrote:
> On Mon, 23 Mar 2009, Frans Pop wrote:
> 
> > > Or do you think maybe it would be better to move this test up into the
> > > PM core?  After all, other subsystems will face the same issue.  I
> > > think that would be the best approach.  Yes?
> > 
> > I did look at that option, but implementing it in the USB subsystem seemed
> > more logical to me, for example as other subsystems possibly would want to
> > display an info message.
> 
> They still can...
> 
> > And is -ENODEV safe to ignore in all cases? Would there be other errors that
> > should be ignored too?
> 
> In general, the PM core ignores _all_ errors during resume -- in the
> sense that it doesn't try to recover from them or do anything to handle
> them.  All it does is put a message in the log.
> 
> So your question becomes: For which errors should a message be added to 
> the system log?  The most logical answer seems to be that we want an 
> error message whenever something bad or unexpected occurs.
> 
> Removal of a hot-unpluggable device isn't really bad or unexpected.  
> Removal of a non-hot-unpluggable device might be bad, but on the other
> hand the system isn't really "hot" while it is suspended.  Besides, the
> appropriate subsystem can print an error message.  Furthermore the
> kernel can't easily tell which devices are hot-unpluggable and which
> aren't.
> 
> Anything else amounts to failure resuming a device that still exists.  
> As such, it probably deserves an error message.

Returning 0 from usb_external_resume_device() if the device is not present
any more doesn't seem wrong.  It's not really an error condition, IMO, because
it's rather normal that the devices may be removed while suspended.

OTOH, I don't think we can ignore -ENODEV universally in the core, because
its meaning may depend on the bus type.  For example, for PCI it sometimes
means a hardware problem has occured (other than the device being not present
any more).

> > if Rafael would be happy with a generic test for -ENODEV, it could be done.
> > If not, maybe some other special error code would need to be used but then
> > you'd still need to test in the subsystem to set that error.
> > Disadvantage is also that it would make resume_device() and related PM
> > driver core functions quite a bit less clean than they currently are.
> > 
> > Implementing the test in USB was quite a bit simpler (for me at least ;-)
> 
> We should get Rafael's opinion.

I'd vote in favor of the Frans' patch, at least for now.

So, Frans, please resubmit with the changelog modified as requested by Alan.

Best,
Rafael