usb: dwc2: avoid leaking DMA channels on disconnection

When the HCD is disconnected, the DMA transfers still in-flight were cleaned-up
but the count of available DMA channels (e.g. available_host_channels) was not
reset.
The pool of DMA channels can be depleted when doing unclean
disconnection of USB peripherals, and reaches the point where no
transfer was possible until the next reboot/reload of the driver.

Tested by putting a programmable USB mux on the port and randomly
plugging/unpluging a USB HUB with USB mass-storage key, USB-audio and
USB-ethernet dongle connected to its downstream ports, and also doing the
disconnection early while the devices are still enumerating to get more URBs
in-flight.
After the patch, the devices are still enumerating after thousands of cycles,
while the port was totally dead before.

Signed-off-by: Vincent Palatin <vpalatin@chromium.org>
Acked-by: John Youn <johnyoun@synopsys.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>

diff --git a/drivers/usb/dwc2/hcd.c b/drivers/usb/dwc2/hcd.c
index c78c8740db1d7855967d441a21ae1ffafe3d26fc..559b55e5debb02e84c76434ba03af649b99ca410 100644 (file)
--- a/drivers/usb/dwc2/hcd.c
+++ b/drivers/usb/dwc2/hcd.c
@@ -257,6 +257,14 @@ static void dwc2_hcd_cleanup_channels(struct dwc2_hsotg *hsotg)
                 */
                channel->qh = NULL;
        }
+       /* All channels have been freed, mark them available */
+       if (hsotg->core_params->uframe_sched > 0) {
+               hsotg->available_host_channels =
+                       hsotg->core_params->host_channels;
+       } else {
+               hsotg->non_periodic_channels = 0;
+               hsotg->periodic_channels = 0;
+       }
 }
 
 /**