Re: [PATCH v3 1/1] usb: xhci: Add Clear_TT_Buffer

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

On 30.4.2019 6.06, Jim Lin wrote:
> USB 2.0 specification chapter 11.17.5 says "as part of endpoint halt
> processing for full-/low-speed endpoints connected via a TT, the host
> software must use the Clear_TT_Buffer request to the TT to ensure
> that the buffer is not in the busy state".

Good point, xhci isn't making sure TT buffers get cleared when they should.

> 
> In our case, a full-speed speaker (ConferenceCam) is behind a high-
> speed hub (ConferenceCam Connect), sometimes once we get STALL on a
> request we may continue to get STALL with the folllowing requests,
> like Set_Interface.
> 
> Here we add Clear_TT_Buffer for the following Set_Interface requests
> to get ACK successfully.
> 
> Signed-off-by: Jim Lin <jilin@nvidia.com>
> ---
> v2: xhci_clear_tt_buffer_complete: add static, shorter indentation
>      , remove its claiming in xhci.h
> v3: Add description for clearing_tt (xhci.h)
> 
>   drivers/usb/host/xhci-ring.c | 28 ++++++++++++++++++++++++++++
>   drivers/usb/host/xhci.c      |  7 +++++++
>   drivers/usb/host/xhci.h      |  2 ++
>   3 files changed, 37 insertions(+)
> 
> diff --git a/drivers/usb/host/xhci-ring.c b/drivers/usb/host/xhci-ring.c
> index 9215a28dad40..02b1ebad81e7 100644
> --- a/drivers/usb/host/xhci-ring.c
> +++ b/drivers/usb/host/xhci-ring.c
> @@ -1786,6 +1786,33 @@ struct xhci_segment *trb_in_td(struct xhci_hcd *xhci,
>   	return NULL;
>   }
>   
> +static void xhci_clear_hub_tt_buffer(struct xhci_hcd *xhci,
> +	unsigned int slot_id, struct xhci_td *td)
> +{
> +	struct xhci_virt_device *dev;
> +	struct xhci_slot_ctx *slot_ctx;
> +	int saved_devnum;
> +
> +	/*
> +	 * As part of low/full-speed endpoint-halt processing
> +	 * we must clear the TT buffer (USB 2.0 specification 11.17.5).
> +	 */
> +	if (td->urb->dev->tt && !usb_pipeint(td->urb->pipe) &&
> +	    (td->urb->dev->tt->hub != xhci_to_hcd(xhci)->self.root_hub) &&
> +	    !xhci->clearing_tt) {
> +		xhci->clearing_tt = 1;

one xhci->clearing_tt under is not enough, there might be several HS hubs, or
multi TT hubs with halted endpoints at the same time that need TT clearing.

How about a flag per endpoint?

For example Aadding a EP_CLEARING_TT flag for ep_state in struct xhci_virt_ep?
just like EP_STOP_CMD_PENDING, or EP_HALTED

> +		dev = xhci->devs[slot_id];
> +		slot_ctx = xhci_get_slot_ctx(xhci, dev->out_ctx);
> +		/* Update devnum temporarily for usb_hub_clear_tt_buffer */
> +		saved_devnum = td->urb->dev->devnum;
> +		td->urb->dev->devnum = (int) le32_to_cpu(slot_ctx->dev_state) &
> +			DEV_ADDR_MASK;

Changing the struct usb_device devnum on the fly seems like a bit of a hack, and probably
causes issues elsewhere. Devnum is tied to uevents, usbfs, sysfs etc.

We need another solution, some options:

- Let usb_hub_clear_tt_buffer() figure out address and not just use devnum if host == xhci.
- Add address to struct usb_device, (would have both address and devnum), use it when needed.
- Provide address as parameter to usb_clear_tt_buffer() (api change, changes other host drivers)
- Force devnum to be same as address, usb core can't choose address for xhci devices.

-Mathias