From: Shimmer Huang <[email protected]>
Linux xHCI driver does not check the CEC bit in register PORTSC when
handling port status events. If Port Configure Error for root hub port
occurs, CEC bit in PORTSC would be set by xHC and remains 1. This
happends when the root port fails to configure its link partner,
e.g. the port fails to exchange port capabilities information using
Port Capability LMPs.
Then the Port Status Change Events will be blocked until all status
change bits(CEC is one of the change bits) are cleared('0') (refer to
xHCI spec 4.19.2). Otherwise, the port status change event for this
root port will not be generated anymore, then root port would look
like “dead” for user and can’t be recovered until a Host Controller
Reset(HCRST)
This patch is to check CEC bit and clear the CEC bit if it's set to 1
in function handle_port_status().
[[email protected]: ported from 3.10 and added more
explanations(from Shimmer) in the commit log]
Signed-off-by: Guang Shen <[email protected]>
Signed-off-by: Shimmer Huang <[email protected]>
Signed-off-by: Nadav Haklai <[email protected]>
Reviewed-by: Yehuda Yitschak <[email protected]>
Signed-off-by: Gregory CLEMENT <[email protected]>
Cc: <[email protected]>
---
Hi,
usually I tried to add a kernel version for the stable team, but for
this patch I don't know since when it makes sens to apply it.
Gregory
drivers/usb/host/xhci-ring.c | 7 +++++++
1 file changed, 7 insertions(+)
diff --git a/drivers/usb/host/xhci-ring.c b/drivers/usb/host/xhci-ring.c
index e692e769c50c..45d8dd7e07f3 100644
--- a/drivers/usb/host/xhci-ring.c
+++ b/drivers/usb/host/xhci-ring.c
@@ -1541,6 +1541,13 @@ static void handle_port_status(struct xhci_hcd *xhci,
port_id);
temp = readl(port_array[faked_port_index]);
+
+ if (temp & PORT_CEC) {
+ xhci_dbg(xhci, "port failed to configure its link partner.\n");
+ xhci_test_and_clear_bit(xhci, port_array,
+ faked_port_index, PORT_CEC);
+ }
+
if (hcd->state == HC_STATE_SUSPENDED) {
xhci_dbg(xhci, "resume root hub\n");
usb_hcd_resume_root_hub(hcd);
--
2.1.0
On 06.02.2015 17:20, Gregory CLEMENT wrote:
> From: Shimmer Huang <[email protected]>
>
> Linux xHCI driver does not check the CEC bit in register PORTSC when
> handling port status events. If Port Configure Error for root hub port
> occurs, CEC bit in PORTSC would be set by xHC and remains 1. This
> happends when the root port fails to configure its link partner,
> e.g. the port fails to exchange port capabilities information using
> Port Capability LMPs.
>
> Then the Port Status Change Events will be blocked until all status
> change bits(CEC is one of the change bits) are cleared('0') (refer to
> xHCI spec 4.19.2). Otherwise, the port status change event for this
> root port will not be generated anymore, then root port would look
> like “dead” for user and can’t be recovered until a Host Controller
> Reset(HCRST)
>
> This patch is to check CEC bit and clear the CEC bit if it's set to 1
> in function handle_port_status().
>
> [[email protected]: ported from 3.10 and added more
> explanations(from Shimmer) in the commit log]
>
> Signed-off-by: Guang Shen <[email protected]>
> Signed-off-by: Shimmer Huang <[email protected]>
> Signed-off-by: Nadav Haklai <[email protected]>
> Reviewed-by: Yehuda Yitschak <[email protected]>
> Signed-off-by: Gregory CLEMENT <[email protected]>
> Cc: <[email protected]>
> ---
> Hi,
>
> usually I tried to add a kernel version for the stable team, but for
> this patch I don't know since when it makes sens to apply it.
>
> Gregory
>
>
> drivers/usb/host/xhci-ring.c | 7 +++++++
> 1 file changed, 7 insertions(+)
>
> diff --git a/drivers/usb/host/xhci-ring.c b/drivers/usb/host/xhci-ring.c
> index e692e769c50c..45d8dd7e07f3 100644
> --- a/drivers/usb/host/xhci-ring.c
> +++ b/drivers/usb/host/xhci-ring.c
> @@ -1541,6 +1541,13 @@ static void handle_port_status(struct xhci_hcd *xhci,
> port_id);
>
> temp = readl(port_array[faked_port_index]);
> +
> + if (temp & PORT_CEC) {
> + xhci_dbg(xhci, "port failed to configure its link partner.\n");
> + xhci_test_and_clear_bit(xhci, port_array,
> + faked_port_index, PORT_CEC);
> + }
> +
> if (hcd->state == HC_STATE_SUSPENDED) {
> xhci_dbg(xhci, "resume root hub\n");
> usb_hcd_resume_root_hub(hcd);
>
Nice catch,
Looks like we're not handling the Config Error Change (CEC) at all in xhci.
I think we instead need to set a Config Error in the return status of the GetPortStatus
requests by setting USB_PORT_FEAT_C_PORT_CONFIG_ERROR in xhci_get_port_status().
This should cause a ClearPortFeature request, where we then can clear the bit in
xhci_clear_port_change_bit()
So going the long way round informing the usb core about the error instead of
immediately clearing the bit.
-Mathias