This patch fixes an issue introduced by patch:
72c973d usb: gadget: add usb_endpoint_descriptor to struct usb_ep
Without this patch we see a kworker taking 100% CPU, after this sequence:
- Connect gadget to a windows host
- load g_ether
- ifconfig up <ip>; ifconfig down; ifconfig up
- ping <windows host>
The "ifconfig down" results in calling eth_stop(), which will call
usb_ep_disable() and, if the carrier is still ok, usb_ep_enable():
usb_ep_disable(link->in_ep);
usb_ep_disable(link->out_ep);
if (netif_carrier_ok(net)) {
usb_ep_enable(link->in_ep);
usb_ep_enable(link->out_ep);
}
The ep should stay enabled, but will not, as ep_disable set the desc
pointer to NULL, therefore the subsequent ep_enable will fail. This leads
to permanent rescheduling of the eth_work() worker as usb_ep_queue()
(called by the worker) will fail due to the unconfigured endpoint.
We fix this issue by saving the ep descriptors and re-assign them before
usb_ep_enable().
Cc: Tatyana Brokhman <tlinder@codeaurora.org>
Signed-off-by: Michael Grzeschik <m.grzeschik@pengutronix.de>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
Cc: stable <stable@vger.kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
spin_lock_irqsave(&dev->lock, flags);
if (dev->port_usb) {
struct gether *link = dev->port_usb;
+ const struct usb_endpoint_descriptor *in;
+ const struct usb_endpoint_descriptor *out;
if (link->close)
link->close(link);
* their own pace; the network stack can handle old packets.
* For the moment we leave this here, since it works.
*/
+ in = link->in_ep->desc;
+ out = link->out_ep->desc;
usb_ep_disable(link->in_ep);
usb_ep_disable(link->out_ep);
if (netif_carrier_ok(net)) {
DBG(dev, "host still using in/out endpoints\n");
+ link->in_ep->desc = in;
+ link->out_ep->desc = out;
usb_ep_enable(link->in_ep);
usb_ep_enable(link->out_ep);
}