[PATCH v2] usb: host: xhci: wait CNR when doing xhci resume

[PATCH v2] usb: host: xhci: wait CNR when doing xhci resume

[Rick Tseng]

NVIDIA 3.1 xHCI card would lose power when moving power state into D3Cold.
Thus we need to wait CNR bit to clear when xhci resmue as xhci init.

Signed-off-by: Rick Tseng <rtseng@nvidia.com>
---
 drivers/usb/host/xhci.c | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/drivers/usb/host/xhci.c b/drivers/usb/host/xhci.c
index 03d1e55..6c7102c 100644
--- a/drivers/usb/host/xhci.c
+++ b/drivers/usb/host/xhci.c
@@ -1108,6 +1108,15 @@ int xhci_resume(struct xhci_hcd *xhci, bool hibernated)
 		hibernated = true;
 
 	if (!hibernated) {
+		/* Some xHC would lose power during suspend, so wait for
+		 * controller ready from resume as xHC init.
+		 */
+		if (xhci_handshake(&xhci->op_regs->status,
+				   STS_CNR, 0, 10 * 1000 * 1000)) {
+			xhci_warn(xhci, "WARN: xHC timeout for CNR clear\n");
+			spin_unlock_irq(&xhci->lock);
+			return -ETIMEDOUT;
+		}
 		/* step 1: restore register */
 		xhci_restore_registers(xhci);
 		/* step 2: initialize command ring buffer */
-- 
2.1.4

Re: [PATCH v2] usb: host: xhci: wait CNR when doing xhci resume

[Mathias Nyman]

On 6.9.2019 8.36, Rick Tseng wrote:
> NVIDIA 3.1 xHCI card would lose power when moving power state into D3Cold.
> Thus we need to wait CNR bit to clear when xhci resmue as xhci init.
> 
> Signed-off-by: Rick Tseng <rtseng@nvidia.com>
> ---
>   drivers/usb/host/xhci.c | 9 +++++++++
>   1 file changed, 9 insertions(+)
> 
> diff --git a/drivers/usb/host/xhci.c b/drivers/usb/host/xhci.c
> index 03d1e55..6c7102c 100644
> --- a/drivers/usb/host/xhci.c
> +++ b/drivers/usb/host/xhci.c
> @@ -1108,6 +1108,15 @@ int xhci_resume(struct xhci_hcd *xhci, bool hibernated)
>   		hibernated = true;
>   
>   	if (!hibernated) {
> +		/* Some xHC would lose power during suspend, so wait for
> +		 * controller ready from resume as xHC init.
> +		 */
> +		if (xhci_handshake(&xhci->op_regs->status,
> +				   STS_CNR, 0, 10 * 1000 * 1000)) {
> +			xhci_warn(xhci, "WARN: xHC timeout for CNR clear\n");
> +			spin_unlock_irq(&xhci->lock);
> +			return -ETIMEDOUT;
> +		}

xhci_handshake() can return -ENODEV in case controller is not reachable (still in PCI D3).
Would be better to just show and return what xhci_handshake() returns.
I know there are places where the existing code doesn't do this, but it would be
better to add it for new code

ret = xhci_handshake(CNR bit clear)
if (ret) {
	xhci_warn(xhci, "Controller not ready at resume %d\n", ret);
	unlock()
	return ret
}

-Mathias