Re: [PATCH v2] usb: Retry port status check on resume to work around RH bugs

From: Alan Stern
Date: Tue Jan 27 2015 - 11:17:08 EST


On Mon, 26 Jan 2015, Julius Werner wrote:

> The EHCI controller on the RK3288 SoC is violating basic parts of the
> USB spec and thereby unable to properly resume a suspended port. It does
> not start SOF generation within 3ms of finishing resume signaling, so
> the attached device will drop off the bus again. This is a particular
> problem with runtime PM, where accessing the device will trigger a
> resume that immediately makes it unavailabe (and reenumerate with a new

Typo: "unavailabe"

> handle).
>
> Thankfully, the persist feature is generally able to work around stuff
> like that. Unfortunately, it doesn't quite work in this particular case
> because the controller will turn off the CurrentConnectStatus bit for an
> instant while the device is reconnecting, which causes the kernel to
> conclude that it permanently disappeared. This patch adds a tiny retry
> mechanism to the core port resume code which will catch this case and
> shouldn't have any notable impact on other controllers.
>
> Signed-off-by: Julius Werner <jwerner@xxxxxxxxxxxx>
> ---
> drivers/usb/core/hub.c | 13 +++++++++++--
> 1 file changed, 11 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/usb/core/hub.c b/drivers/usb/core/hub.c
> index aeb50bb..d1d0a66 100644
> --- a/drivers/usb/core/hub.c
> +++ b/drivers/usb/core/hub.c
> @@ -2896,10 +2896,12 @@ static int port_is_suspended(struct usb_hub *hub, unsigned portstatus)
> */
> static int check_port_resume_type(struct usb_device *udev,
> struct usb_hub *hub, int port1,
> - int status, unsigned portchange, unsigned portstatus)
> + int status, u16 portchange, u16 portstatus)
> {
> struct usb_port *port_dev = hub->ports[port1 - 1];
> + int retries = 3;
>
> +retry:

I prefer to indent statement labels by one space, so that diff doesn't
see them as function declarations.

> /* Is a warm reset needed to recover the connection? */
> if (status == 0 && udev->reset_resume
> && hub_port_warm_reset_required(hub, port1, portstatus)) {
> @@ -2909,8 +2911,15 @@ static int check_port_resume_type(struct usb_device *udev,
> else if (status || port_is_suspended(hub, portstatus) ||
> !port_is_power_on(hub, portstatus) ||
> !(portstatus & USB_PORT_STAT_CONNECTION)) {
> - if (status >= 0)
> + if (status >= 0) {

Thinking about this more carefully, it looks like the only thing you
care about here is the !(portstatus & USB_PORT_STAT_CONNECTION) case.
For everything else, you don't need to retry.

So why not separate out that one case?

else if (status || port_is_suspended(hub, portstatus) ||
!port_is_power_on(hub, portstatus)) {
if (status >= 0)
status = -ENODEV;
} else if (!(portstatus & USB_PORT_STAT_CONNECTION)) {
retry stuff...
}

> + if (retries--) {
> + usleep_range(200, 300);
> + status = hub_port_status(hub, port1,
> + &portstatus, &portchange);
> + goto retry;
> + }
> status = -ENODEV;
> + }
> }
>
> /* Can't do a normal resume if the port isn't enabled,

Alan Stern

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/