On Tue, 11 Jun 2019, Felipe Balbi wrote:

quoted

quoted

quoted

quoted

quoted

So for "severe" issues, yes, we should do this, but perhaps not for all
of the "normal" things we see when a device is yanked out of the system
and the like.

Then what counts as a "severe" issue?  Anything besides enumeration 
failure?

Not that I can think of at the moment, other than the other recently
added KOBJ_CHANGE issue.  I'm sure we have other "hard failure" issues
in the USB stack that people will want exposed over time.

From an XHCI standpoint, Transaction Errors might be one thing. They
happen rarely and are a strong indication that the bus itself is
bad. Either bad cable, misbehaving PHYs, improper power management, etc.

Don't you also get transaction errors if the user unplugs a device in 
the middle of a transfer?  That's not the sort of thing we want to sent 
notifications about.

Mathias, do we get Transaction Error if user removes cable during a
transfer? I thought we would just get Port Status Change with CC bit
cleared, no?

Even if xHCI doesn't give Transaction Errors when a cable is unplugged 
during a transfer, other host controllers do.  Sometimes quite a lot -- 
they continue to occur until the kernel polls the parent hub's 
interrupt ep and learns that the port is disconnected, which can take 
up to 250 ms.

Alan Stern