[PATCH] usb: core: stop USB enumeration if too many retries

[PATCH] usb: core: stop USB enumeration if too many retries

[Ray Chi]

If a broken accessory connected to a USB host, usbcore might
keep doing enumeration retries and it will take a long time to
cause system unstable.

This patch provides a quirk to specific USB ports of the hub to
stop USB enumeration if needed.

Signed-off-by: Ray Chi <raychi@google.com>
---
 drivers/usb/core/hub.c | 33 +++++++++++++++++++++++++++++++++
 include/linux/usb.h    |  3 +++
 2 files changed, 36 insertions(+)

diff --git a/drivers/usb/core/hub.c b/drivers/usb/core/hub.c
index 2633acde7ac1..0f4097440ffb 100644
--- a/drivers/usb/core/hub.c
+++ b/drivers/usb/core/hub.c
@@ -3081,6 +3081,28 @@ static int hub_port_reset(struct usb_hub *hub, int port1,
 	return status;
 }
 
+/* Stop enumerate if the port met errors and quirk is set */
+static bool hub_port_stop_enumerate(struct usb_hub *hub, int port1, int retries)
+{
+	struct usb_port *port_dev = hub->ports[port1 - 1];
+	struct usb_device *hdev = hub->hdev;
+
+	if (retries < (PORT_INIT_TRIES - 1) / 2)
+		return false;
+
+	/*
+	 * Some USB hosts can't take a long time to keep doing enumeration
+	 * retry. After doing half of the retries, we would turn off the port
+	 * power to stop enumeration if the quirk is set.
+	 */
+	if (port_dev->quirks & USB_PORT_QUIRK_STOP_ENUM) {
+		usb_hub_set_port_power(hdev, hub, port1, false);
+		return true;
+	}
+
+	return false;
+}
+
 /* Check if a port is power on */
 int usb_port_is_power_on(struct usb_hub *hub, unsigned int portstatus)
 {
@@ -4855,6 +4877,9 @@ hub_port_init(struct usb_hub *hub, struct usb_device *udev, int port1,
 					buf->bMaxPacketSize0;
 			kfree(buf);
 
+			if (r < 0 && (port_dev->quirks & USB_PORT_QUIRK_STOP_ENUM))
+				goto fail;
+
 			retval = hub_port_reset(hub, port1, udev, delay, false);
 			if (retval < 0)		/* error or disconnect */
 				goto fail;
@@ -5387,6 +5412,9 @@ static void hub_port_connect(struct usb_hub *hub, int port1, u16 portstatus,
 		if ((status == -ENOTCONN) || (status == -ENOTSUPP))
 			break;
 
+		if (hub_port_stop_enumerate(hub, port1, i))
+			break;
+
 		/* When halfway through our retry count, power-cycle the port */
 		if (i == (PORT_INIT_TRIES - 1) / 2) {
 			dev_info(&port_dev->dev, "attempt power cycle\n");
@@ -5934,6 +5962,9 @@ static int usb_reset_and_verify_device(struct usb_device *udev)
 		ret = hub_port_init(parent_hub, udev, port1, i);
 		if (ret >= 0 || ret == -ENOTCONN || ret == -ENODEV)
 			break;
+
+		if (hub_port_stop_enumerate(parent_hub, port1, i))
+			goto stop_enumerate;
 	}
 	mutex_unlock(hcd->address0_mutex);
 
@@ -6022,6 +6053,8 @@ static int usb_reset_and_verify_device(struct usb_device *udev)
 	udev->bos = bos;
 	return 0;
 
+stop_enumerate:
+	mutex_unlock(hcd->address0_mutex);
 re_enumerate:
 	usb_release_bos_descriptor(udev);
 	udev->bos = bos;
diff --git a/include/linux/usb.h b/include/linux/usb.h
index f7a9914fc97f..fc0fef58c706 100644
--- a/include/linux/usb.h
+++ b/include/linux/usb.h
@@ -490,6 +490,9 @@ enum usb_port_connect_type {
 /* Decrease TRSTRCY to 10ms during device enumeration. */
 #define USB_PORT_QUIRK_FAST_ENUM	BIT(1)
 
+/* Stop the enumeration for the given port if there are too many failures*/
+#define USB_PORT_QUIRK_STOP_ENUM	BIT(2)
+
 /*
  * USB 2.0 Link Power Management (LPM) parameters.
  */
-- 
2.37.2.672.g94769d06f0-goog

Re: [PATCH] usb: core: stop USB enumeration if too many retries

[Alan Stern]

On Fri, Aug 26, 2022 at 03:58:39PM +0800, Ray Chi wrote:
> If a broken accessory connected to a USB host, usbcore might
> keep doing enumeration retries and it will take a long time to
> cause system unstable.
> 
> This patch provides a quirk to specific USB ports of the hub to
> stop USB enumeration if needed.

Why only to specific ports?

> Signed-off-by: Ray Chi <raychi@google.com>
> ---
>  drivers/usb/core/hub.c | 33 +++++++++++++++++++++++++++++++++
>  include/linux/usb.h    |  3 +++
>  2 files changed, 36 insertions(+)
> 
> diff --git a/drivers/usb/core/hub.c b/drivers/usb/core/hub.c
> index 2633acde7ac1..0f4097440ffb 100644
> --- a/drivers/usb/core/hub.c
> +++ b/drivers/usb/core/hub.c
> @@ -3081,6 +3081,28 @@ static int hub_port_reset(struct usb_hub *hub, int port1,
>  	return status;
>  }
>  
> +/* Stop enumerate if the port met errors and quirk is set */
> +static bool hub_port_stop_enumerate(struct usb_hub *hub, int port1, int retries)
> +{
> +	struct usb_port *port_dev = hub->ports[port1 - 1];
> +	struct usb_device *hdev = hub->hdev;
> +
> +	if (retries < (PORT_INIT_TRIES - 1) / 2)
> +		return false;
> +
> +	/*
> +	 * Some USB hosts can't take a long time to keep doing enumeration
> +	 * retry. After doing half of the retries, we would turn off the port
> +	 * power to stop enumeration if the quirk is set.
> +	 */
> +	if (port_dev->quirks & USB_PORT_QUIRK_STOP_ENUM) {
> +		usb_hub_set_port_power(hdev, hub, port1, false);

Why turn the port power off?  Aren't there better ways to stop the 
enumeration attempts?  When will the power ever get turned back on?

Why not use the initial_descriptor_timeout module parameter for this 
purpose?  That's the sort of thing it was meant for.

Alan Stern

Re: [PATCH] usb: core: stop USB enumeration if too many retries

[Ray Chi]

On Fri, Aug 26, 2022 at 10:55 PM Alan Stern <stern@rowland.harvard.edu> wrote:
>
> On Fri, Aug 26, 2022 at 03:58:39PM +0800, Ray Chi wrote:
> > If a broken accessory connected to a USB host, usbcore might
> > keep doing enumeration retries and it will take a long time to
> > cause system unstable.
> >
> > This patch provides a quirk to specific USB ports of the hub to
> > stop USB enumeration if needed.
>
> Why only to specific ports?

The specific port means it is connected to a broken accessory.

>
> > Signed-off-by: Ray Chi <raychi@google.com>
> > ---
> >  drivers/usb/core/hub.c | 33 +++++++++++++++++++++++++++++++++
> >  include/linux/usb.h    |  3 +++
> >  2 files changed, 36 insertions(+)
> >
> > diff --git a/drivers/usb/core/hub.c b/drivers/usb/core/hub.c
> > index 2633acde7ac1..0f4097440ffb 100644
> > --- a/drivers/usb/core/hub.c
> > +++ b/drivers/usb/core/hub.c
> > @@ -3081,6 +3081,28 @@ static int hub_port_reset(struct usb_hub *hub, int port1,
> >       return status;
> >  }
> >
> > +/* Stop enumerate if the port met errors and quirk is set */
> > +static bool hub_port_stop_enumerate(struct usb_hub *hub, int port1, int retries)
> > +{
> > +     struct usb_port *port_dev = hub->ports[port1 - 1];
> > +     struct usb_device *hdev = hub->hdev;
> > +
> > +     if (retries < (PORT_INIT_TRIES - 1) / 2)
> > +             return false;
> > +
> > +     /*
> > +      * Some USB hosts can't take a long time to keep doing enumeration
> > +      * retry. After doing half of the retries, we would turn off the port
> > +      * power to stop enumeration if the quirk is set.
> > +      */
> > +     if (port_dev->quirks & USB_PORT_QUIRK_STOP_ENUM) {
> > +             usb_hub_set_port_power(hdev, hub, port1, false);
>
> Why turn the port power off?  Aren't there better ways to stop the
> enumeration attempts?  When will the power ever get turned back on?
>

Because the broken accessory is still connected to the port. Even if we stop
the enumeration attempts, the port change event will issue a new port
initialization.

The implementation is used for dual-role devices, the port power could
be turned on
when the host mode restarts again.

> Why not use the initial_descriptor_timeout module parameter for this
> purpose?  That's the sort of thing it was meant for.
>

As I mentioned above, the usbcore driver will keep doing enumeration attempts if
the broken accessory is still connected. It never stops. This is why I
want to turn
off the port to stop enumeration.

> Alan Stern

Thanks,
Ray

Re: [PATCH] usb: core: stop USB enumeration if too many retries

[Alan Stern]

On Sat, Aug 27, 2022 at 01:53:45AM +0800, Ray Chi wrote:
> On Fri, Aug 26, 2022 at 10:55 PM Alan Stern <stern@rowland.harvard.edu> wrote:
> >
> > On Fri, Aug 26, 2022 at 03:58:39PM +0800, Ray Chi wrote:
> > > If a broken accessory connected to a USB host, usbcore might
> > > keep doing enumeration retries and it will take a long time to
> > > cause system unstable.
> > >
> > > This patch provides a quirk to specific USB ports of the hub to
> > > stop USB enumeration if needed.
> >
> > Why only to specific ports?
> 
> The specific port means it is connected to a broken accessory.

Then this patch should be about connections to broken accessories, not 
too many retries.  The quirk should make the hub driver permanently 
ignore the port; whether the power is on or off doesn't matter.

Also, you might want to check whether the port connection is fixed 
rather than hot-unpluggable.  If the broken accessory can be unplugged 
from the port then you don't want to disable the port.

> > > Signed-off-by: Ray Chi <raychi@google.com>
> > > ---
> > >  drivers/usb/core/hub.c | 33 +++++++++++++++++++++++++++++++++
> > >  include/linux/usb.h    |  3 +++
> > >  2 files changed, 36 insertions(+)
> > >
> > > diff --git a/drivers/usb/core/hub.c b/drivers/usb/core/hub.c
> > > index 2633acde7ac1..0f4097440ffb 100644
> > > --- a/drivers/usb/core/hub.c
> > > +++ b/drivers/usb/core/hub.c
> > > @@ -3081,6 +3081,28 @@ static int hub_port_reset(struct usb_hub *hub, int port1,
> > >       return status;
> > >  }
> > >
> > > +/* Stop enumerate if the port met errors and quirk is set */
> > > +static bool hub_port_stop_enumerate(struct usb_hub *hub, int port1, int retries)
> > > +{
> > > +     struct usb_port *port_dev = hub->ports[port1 - 1];
> > > +     struct usb_device *hdev = hub->hdev;
> > > +
> > > +     if (retries < (PORT_INIT_TRIES - 1) / 2)
> > > +             return false;
> > > +
> > > +     /*
> > > +      * Some USB hosts can't take a long time to keep doing enumeration
> > > +      * retry. After doing half of the retries, we would turn off the port
> > > +      * power to stop enumeration if the quirk is set.
> > > +      */
> > > +     if (port_dev->quirks & USB_PORT_QUIRK_STOP_ENUM) {
> > > +             usb_hub_set_port_power(hdev, hub, port1, false);
> >
> > Why turn the port power off?  Aren't there better ways to stop the
> > enumeration attempts?  When will the power ever get turned back on?
> >
> 
> Because the broken accessory is still connected to the port. Even if we stop
> the enumeration attempts, the port change event will issue a new port
> initialization.

If you know the accessory is broken then there's no reason to pay 
attention to the port change events in the first place.  You can simply 
make port_event() avoid calling hub_port_connect_change() if the quirk 
is set.

Alan Stern

> The implementation is used for dual-role devices, the port power could
> be turned on
> when the host mode restarts again.
> 
> > Why not use the initial_descriptor_timeout module parameter for this
> > purpose?  That's the sort of thing it was meant for.
> >
> 
> As I mentioned above, the usbcore driver will keep doing enumeration attempts if
> the broken accessory is still connected. It never stops. This is why I
> want to turn
> off the port to stop enumeration.
> 
> > Alan Stern
> 
> Thanks,
> Ray

Re: [PATCH] usb: core: stop USB enumeration if too many retries

[Greg KH]

On Fri, Aug 26, 2022 at 03:58:39PM +0800, Ray Chi wrote:
> If a broken accessory connected to a USB host, usbcore might
> keep doing enumeration retries and it will take a long time to
> cause system unstable.
> 
> This patch provides a quirk to specific USB ports of the hub to
> stop USB enumeration if needed.

Where does it ever allow the port to handle new devices in the future if
the device is removed and then a new one is added back?  Or is the port
just now dead for forever?

thanks,

greg k-h

Re: [PATCH] usb: core: stop USB enumeration if too many retries

[Ray Chi]

On Tue, Aug 30, 2022 at 8:44 PM Greg KH <gregkh@linuxfoundation.org> wrote:
>
> On Fri, Aug 26, 2022 at 03:58:39PM +0800, Ray Chi wrote:
> > If a broken accessory connected to a USB host, usbcore might
> > keep doing enumeration retries and it will take a long time to
> > cause system unstable.
> >
> > This patch provides a quirk to specific USB ports of the hub to
> > stop USB enumeration if needed.
>
> Where does it ever allow the port to handle new devices in the future if
> the device is removed and then a new one is added back?  Or is the port
> just now dead for forever?
>

I modified the patch according to Alan's suggestion, so the port will
be working again
after clearing the quirk with the v2 patch.

> thanks,
>
> greg k-h

Thanks,
Ray