Re: kernel crash/disc errors when unbinding USB devices

[Greg Kroah-Hartman]

On Sun, Jan 23, 2022 at 12:59:21PM +0100, Helge Deller wrote:
> On 1/23/22 12:07, Greg Kroah-Hartman wrote:
> > On Sun, Jan 23, 2022 at 11:40:01AM +0100, Helge Deller wrote:
> >> On all kernels 5.15.x and 5.16.x I noticed that resetting
> >> the USB devices with this shell script:
> >>
> >>        for i in $(ls /sys/bus/pci/drivers/ahci/|grep :)
> >>          do
> >>          echo $i
> >>          echo $i >/sys/bus/pci/drivers/ahci/unbind
> >>          sleep 1
> >>          echo $i >/sys/bus/pci/drivers/ahci/bind
> >>         done
> >
> > That is dangerous to do, why do this?  All of your block devices might
> > have disappeard.
> >
> >>         # reseting USB3 ports (if there none you'll get errors)
> >>         for i in $(ls /sys/bus/pci/drivers/xhci_hcd/|grep :)
> >>          do
> >>          echo $i
> >>          echo $i >/sys/bus/pci/drivers/xhci_hcd/unbind
> >>          sleep 1
> >>          echo $i >/sys/bus/pci/drivers/xhci_hcd/bind
> >>         done
> >
> > Again, why do this at all?
> 
> I need to reset some of the USB devices after a suspend/resume cycle.

The devices, or the host controllers?  They are different things.

> The problem is, that some of the USB devices are
> handed over into a running VirtualBox VM and after
> the suspend/resume they need to be virtually plugged out/in
> so that the running Windows VM will reconnect them.

unbind/bind is a very harsh way of doing this.  but do it on the USB
devics, not the host controllers.

> If you search in the internet, you will find many places
> where this unbind/bind process is mentioned, e.g.:
> https://askubuntu.com/questions/645/how-do-you-reset-a-usb-device-from-the-command-line
> This procedure did worked in the past.

Never trust the internet :)

But note, there is a usbreset program mentioned there, which is part of
'usbutils' and should already be installed on your machine.  Why not
just use that?

> I think the main problem is, that it's somehow unexpected that
> the SATA controllers are logically attached to the USB
> controllers.

Huh?  No they are not.

'ahci' is a SATA controller.
'xhci' is a USB controller.

Two totally different things.

> So, if you believe to unbind "just" the USB devices you
> suddenly unbind SATA controllers too.

Again, no, you are unbinding two different things here.  And a USB host
controller is different than individual USB devices as well.

> >> immediately leads to a non-functional system, because the discs
> >> face I/O errors and will switch to read-only mode.
> >
> > Sure, the disks are gone, this is expected.
> 
> Ok.
> 
> >> Effectively I have to reboot the machine.
> >
> > I'm amazed it keeps working at all.
> 
> :-)
> 
> >> It's reproduceable for me on two completely different x86 machines with
> >> different USB devices.
> >
> > What do you expect this to do instead?  You just disabled all block
> > controllers in your system, followed by all USB controllers.  And then
> > attempted to add them back, but given that the old devices still had
> > userspace references on them, the devices will come back with different
> > names and so you need to mount them again.
> >
> > So this looks like it is working as intended.  Just don't do this :)
> 
> Ok, sure, but I think it's unexpected that the SATA ports are listed
> beneath the USB controllers, beside USB devices.

Where are they listed like that?

> If I run "lsusb", the SATA controllers don't show up either.

That is because they are independant and have nothing to do with each
otner.

> They are listed when I run "lspci".
> Shouldn't they maybe show up somewhere else, e.g.
> /sys/bus/pci/drivers/sata/*

The controllers are PCI drivers.  The block devices they create are
scsi devices (through the ata system).  This has nothing to do with USB.

confused,

greg k-h