Re: [PATCH] Check for changed device descriptors when a connection-change occurs before validating the connection.

From: David Heinzelmann <heinzelmann.david@gmail.com>
To: Greg KH <gregkh@linuxfoundation.org>
Cc: linux-usb@vger.kernel.org
Subject: Re: [PATCH] Check for changed device descriptors when a connection-change occurs before validating the connection.
Date: Fri, 20 Sep 2019 17:33:57 +0200	[thread overview]
Message-ID: <20190920153357.GB5913@dhe-pc> (raw)
In-Reply-To: <20190920121538.GA549982@kroah.com>

On Fri, Sep 20, 2019 at 02:15:38PM +0200, Greg KH wrote:
> On Fri, Sep 20, 2019 at 03:17:26PM +0200, David Heinzelmann wrote:
> > Hi,
> > 
> > sorry for the wrong patch format.
> 
> No problem, that's normal.  But please do not top-post on linux mailing
> lists.
> 
> > I am trying to detect a change. At the moment I think the change could be ignored if a
> > port connection-change occurs and the port status has again the 'PORT_CONNECTION' bit set. 
> > 
> > I have a fx3 device which does a re-enumeration after a firmware download. This is working
> > as expected and I am seeing a 'remove event' and a 'add event' monitoring via udevadm. But
> > if I connect multiple devices at the same time via an usb hub I am sometimes not receiving
> > a 'remove event' and 'add event' for a single device.
> 
> Sounds like a broken hub :)
> 

I tried different hubs but I forgot to mention that it is also possible to trigger the issue
without a hub if I reboot the devices via software at the same time. 

> > I think the problem could be that when a device disconnects and the port connection-change
> > occurs and before the 'PORT_CONNECTION' bit is checked the device could already be
> > reconnected and the 'PORT_CONNECTION' bit is set. Therefore I think it is not correct to
> > resuscitate the exisiting device.
> 
> Does your patch actually fix the issue?  When a fx3 device downloads
> firmware and re-enumerates, it should come back as a totally different
> device, which will fail this check, right?  So I don't see how this
> fixes the issues with your devices.
> 

With the patch I do not have the issue anymore. After re-enumerate the device comes back with the same
VID/PID but with a different device descriptor. Therefore the check will fail and hub_port_connect is
called which initiates a device disconnect and connect. Without this 'reconnect' lsusb still shows me 
the old device descriptor and I am not able to communicate with the device. 

> Unless all of the devices reset themselves at the same time and the hub
> doesn't like that and can't notice that it happened?
> 
> If you use a different hub, does that work properly?
> 

There is no difference if an other hub is used. It also happens without a hub when the devices are
rebooted via software. My thoughts on this is that when the device re-enumerates and the device
descriptor has changed a device disconnect and connect should be initiated instead of doing nothing?

If I understand it correctly the resuscitation is used for handling port enable-changes occured by EMI.
But when the device is doing a re-enumeration there should be no resuscitation.

David