Re: [PATCH v2 4/7] usb: xhci: Workaround undercalculated BW for fullspeed BI

[Mathias Nyman <mathias.nyman@linux.intel.com>]

On 10.4.2021 3.47, Thinh Nguyen wrote:
> DWC_usb31 host version 1.90a and prior undercalculates the bandwidth
> available for interrupt endpoints. The controller will return bandwidth
> error on config endpoint commands if there are already 6 or more
> fullspeed interrupt endpoints with bInterval of 4 (or 4ms) associated
> with a single fullspeed bus instance (BI).
> 
> To workaround this, configure and use the endpoint at a shorter
> interrupt interval. Lower the ep_ctx interval from 5 to 4 (or 2ms)
> for interrupt endpoints of the fullspeed BI. Note: we have not observed
> functional impact to the fullspeed devices by lowering the interrupt
> service interval (at least for a few devices that we tested).
> 
> To simplify the workaround, let's just check and apply the workaround if
> the endpoint is a fullspeed interrupt endpoint with interval of 4ms and
> if the top parent device is also operating in fullspeed (i.e. associated
> with fullspeed BI).
> 
> Signed-off-by: Thinh Nguyen <Thinh.Nguyen@synopsys.com>
> ---
> Changes in v2:
> - None
> 
> Note:
> Checkpatch will give a warning below for getting top_dev:
> 	WARNING: suspect code indent for conditional statements
> 
> Since this logic is done everywhere else in the driver, I'm keeping it
> consistent here.
> 
>  drivers/usb/host/xhci-mem.c     | 23 +++++++++++++++++++++++
>  include/linux/usb/xhci-quirks.h |  1 +
>  2 files changed, 24 insertions(+)
> 
> diff --git a/drivers/usb/host/xhci-mem.c b/drivers/usb/host/xhci-mem.c
> index 1053b43008e4..e01d0ddb012a 100644
> --- a/drivers/usb/host/xhci-mem.c
> +++ b/drivers/usb/host/xhci-mem.c
> @@ -1463,6 +1463,29 @@ int xhci_endpoint_init(struct xhci_hcd *xhci,
>  		}
>  	}
>  
> +	/*
> +	 * Check for undercalculated bandwidth quirk for interrupt endpoints
> +	 * associated with fullspeed BI.
> +	 */
> +	if ((xhci->quirks & XHCI_LIMIT_FS_BI_INTR_EP) &&
> +	    usb_endpoint_xfer_int(&ep->desc) &&
> +	    udev->speed == USB_SPEED_FULL &&
> +	    interval == 5) {
> +		struct usb_device *top_dev;
> +
> +		for (top_dev = udev;
> +		     top_dev->parent && top_dev->parent->parent;
> +		     top_dev = top_dev->parent)
> +			/* Found device below root hub */;
> +
> +		/*
> +		 * If the top device is operating at fullspeed, then the
> +		 * controller is using fullspeed BI for this device.
> +		 */
> +		if (top_dev->speed == USB_SPEED_FULL)
> +			interval = 4;
> +	}
> +


If we write a new xhci-snps driver then adjusting the interval could be done in the
.add_endpoint override, after calling xhci_add_endpoint()

If not then something like this should work.
Maybe turn it into a separate function for readability.

xhci_tune_interval_quirk(xhci, udev, ep, &interval)

-Mathias