2013-06-27 23:10:56

by William Gulland

Subject: [PATCH] usb: Clear both buffers when clearing a control transfer TT buffer.

Control transfers have both IN and OUT (or SETUP) packets, so when
clearing TT buffers for a control transfer it's necessary to send
two HUB_CLEAR_TT_BUFFER requests to the hub.

Signed-off-by: William Gulland <[email protected]>
---
drivers/usb/core/hub.c | 9 +++++++++
1 file changed, 9 insertions(+)

diff --git a/drivers/usb/core/hub.c b/drivers/usb/core/hub.c
index feef935..868ad74 100644
--- a/drivers/usb/core/hub.c
+++ b/drivers/usb/core/hub.c
@@ -668,6 +668,15 @@ resubmit:
static inline int
hub_clear_tt_buffer (struct usb_device *hdev, u16 devinfo, u16 tt)
{
+ /* Need to clear both directions for control ep */
+ if (((devinfo >> 11) & USB_ENDPOINT_XFERTYPE_MASK) ==
+ USB_ENDPOINT_XFER_CONTROL) {
+ int status = usb_control_msg(hdev, usb_sndctrlpipe(hdev, 0),
+ HUB_CLEAR_TT_BUFFER, USB_RT_PORT,
+ devinfo ^ 0x8000, tt, NULL, 0, 1000);
+ if (status)
+ return status;
+ }
return usb_control_msg(hdev, usb_sndctrlpipe(hdev, 0),
HUB_CLEAR_TT_BUFFER, USB_RT_PORT, devinfo,
tt, NULL, 0, 1000);
--
1.8.3

2013-06-28 13:54:21

by Sergei Shtylyov

Subject: Re: [PATCH] usb: Clear both buffers when clearing a control transfer TT buffer.

Hello.

On 28-06-2013 3:10, William Gulland wrote:

> Control transfers have both IN and OUT (or SETUP) packets, so when
> clearing TT buffers for a control transfer it's necessary to send
> two HUB_CLEAR_TT_BUFFER requests to the hub.

> Signed-off-by: William Gulland <[email protected]>
> ---
> drivers/usb/core/hub.c | 9 +++++++++
> 1 file changed, 9 insertions(+)

> diff --git a/drivers/usb/core/hub.c b/drivers/usb/core/hub.c
> index feef935..868ad74 100644
> --- a/drivers/usb/core/hub.c
> +++ b/drivers/usb/core/hub.c
> @@ -668,6 +668,15 @@ resubmit:
> static inline int
> hub_clear_tt_buffer (struct usb_device *hdev, u16 devinfo, u16 tt)
> {
> + /* Need to clear both directions for control ep */
> + if (((devinfo >> 11) & USB_ENDPOINT_XFERTYPE_MASK) ==
> + USB_ENDPOINT_XFER_CONTROL) {
> + int status = usb_control_msg(hdev, usb_sndctrlpipe(hdev, 0),
> + HUB_CLEAR_TT_BUFFER, USB_RT_PORT,
> + devinfo ^ 0x8000, tt, NULL, 0, 1000);

Could you use consistent indentation style with the below call?
Also, empty line wouldn't hurt after declaration...

> + if (status)
> + return status;
> + }
> return usb_control_msg(hdev, usb_sndctrlpipe(hdev, 0),
> HUB_CLEAR_TT_BUFFER, USB_RT_PORT, devinfo,
> tt, NULL, 0, 1000);

WBR, Sergei

2013-06-28 15:56:28

by Alan Stern

Subject: Re: [PATCH] usb: Clear both buffers when clearing a control transfer TT buffer.

On Thu, 27 Jun 2013, William Gulland wrote:

> Control transfers have both IN and OUT (or SETUP) packets, so when
> clearing TT buffers for a control transfer it's necessary to send
> two HUB_CLEAR_TT_BUFFER requests to the hub.

What makes you think this is necessary? Have you found any hardware
that requires this?

Although the USB spec doesn't say much about Clear-TT-Buffer requests,
the text in section 11.17.1 and Figure 11-47 seems to indicate that a
second Clear-TT-Buffer shouldn't be needed. The spec says that TT
buffers are matched based on the device address, endpoint number, and
endpoint direction -- but the direction is used only for bulk
endpoints, not for control endpoints.

In particular, a control endpoint uses only one TT buffer for both the
IN and OUT directions. So I don't think two Clear-TT-Buffers are
needed.

Alan Stern

2013-07-01 20:41:19

by Alan Stern

Subject: Re: [PATCH] usb: Clear both buffers when clearing a control transfer TT buffer.

On Thu, 27 Jun 2013, William Gulland wrote:

> Control transfers have both IN and OUT (or SETUP) packets, so when
> clearing TT buffers for a control transfer it's necessary to send
> two HUB_CLEAR_TT_BUFFER requests to the hub.
>
> Signed-off-by: William Gulland <[email protected]>
> ---
> drivers/usb/core/hub.c | 9 +++++++++
> 1 file changed, 9 insertions(+)
>
> diff --git a/drivers/usb/core/hub.c b/drivers/usb/core/hub.c
> index feef935..868ad74 100644
> --- a/drivers/usb/core/hub.c
> +++ b/drivers/usb/core/hub.c
> @@ -668,6 +668,15 @@ resubmit:
> static inline int
> hub_clear_tt_buffer (struct usb_device *hdev, u16 devinfo, u16 tt)
> {
> + /* Need to clear both directions for control ep */
> + if (((devinfo >> 11) & USB_ENDPOINT_XFERTYPE_MASK) ==
> + USB_ENDPOINT_XFER_CONTROL) {
> + int status = usb_control_msg(hdev, usb_sndctrlpipe(hdev, 0),
> + HUB_CLEAR_TT_BUFFER, USB_RT_PORT,
> + devinfo ^ 0x8000, tt, NULL, 0, 1000);
> + if (status)
> + return status;
> + }
> return usb_control_msg(hdev, usb_sndctrlpipe(hdev, 0),
> HUB_CLEAR_TT_BUFFER, USB_RT_PORT, devinfo,
> tt, NULL, 0, 1000);

Well, this won't hurt much. And since it does seem to fix a hardware
problem,

Acked-by: Alan Stern <[email protected]>