Skip to contentSkip to conversation

usb: musb: fix bug in musb_start_urb

PATCH NAK Ajay Kumar Gupta 4 participants 4 messages Feb 6, 2009 – Feb 24, 2009 1 message not in the archive

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 2/3 usb: musb: fix bug in musb_start_urb

urb->transfer_buffer_length and urb->transfer_buffer should be updated based on urb->actual_length.For a fresh and first time urb, actual_length will be zero but for urbs which has been stopped and restarted (as bulk nak scheme does) actual_length may not be zero.

---
 drivers/usb/musb/musb_host.c |    4 ++--
 1 files changed, 2 insertions(+), 2 deletions(-)
drivers/usb/musb/musb_host.c +2 -2
index a776951..1a78f00 100644
@@ -201,8 +201,8 @@ musb_start_urb(struct musb *musb, int is_in, struct musb_qh *qh)
 		len = urb->iso_frame_desc[0].length;
 		break;
 	default:		/* bulk, interrupt */
-		buf = urb->transfer_buffer;
-		len = urb->transfer_buffer_length;
+		buf = urb->transfer_buffer + urb->actual_length;
+		len = urb->transfer_buffer_length - urb->actual_length;
 	}
 
 	DBG(4, "qh %p urb %p dev%d ep%d%s%s, hw_ep %d, %p/%d\n",
Signature · 6 lines
1.5.6

--
To unsubscribe from this list: send the line "unsubscribe linux-usb" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

A message that is not in the archive (sent off-list or not yet received): 1233921146-4046-1-git-send-email-ajay.gupta-l0cyMroinI0@public.gmane.org

#2 lore raw
Quoted 6 lines · Ajay Kumar Gupta

urb->transfer_buffer_length and urb->transfer_buffer should be updated based on urb->actual_length.For a fresh and first time urb, actual_length will be zero but for urbs which has been stopped and restarted (as bulk nak scheme does) actual_length may not be zero.

NAK, this is not a problem for the current driver since URBs do not ever
get restarted. Also, musb_host_tx() doesn't update urb->actual_length -- please fix it too. Also, you must not clear qh->iso_idx when restarting -- you must not start ISO transfer all over again too. Also, you should not set musb->ep0_state to MUSB_EP0_START again in this case (I agree that control transfers will remain not restartable from an arbitatry place even then). If you're trying to make musb_start_urb() able to re-start, please be consistent.

WBR, Sergei
--
To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

#3 lore raw
Quoted 11 lines · Sergei Shtylyov
Quoted 6 lines · Ajay Kumar Gupta

urb->transfer_buffer_length and urb->transfer_buffer should be updated based on urb->actual_length.For a fresh and first time urb, actual_length will be zero but for urbs which has been stopped and restarted (as bulk nak scheme does) actual_length may not be zero.

NAK, this is not a problem for the current driver since URBs do not ever
get restarted.

Resolvable by changing the patch description to not say this is a "fix".

However, since this is a two line change, I think I'll just merge this with the patch adding the bulk RX retry logic.

Quoted 2 lines

Also, musb_host_tx() doesn't update urb->actual_length -- please fix it too.

That would be a good point if the retry patch touched any TX paths. But it doesn't.

Quoted 4 lines

Also, you must not clear qh->iso_idx when restarting -- you must not start ISO transfer all over again too. Also, you should not set musb->ep0_state to MUSB_EP0_START again in this case (I agree that control transfers will remain not restartable from an arbitatry place even then).

But the [3/3] patch only adds NAK timeout support for bulk RX. And ISO transfers can't NAK in the first place.

Plus, as noted in a comment you could see in this patch, this only touches (re)start for bulk/interrupt transfers. Not ISO; not control.

- Dave

#4 lore raw

Hello.

Quoted 12 lines · David Brownell
Quoted 11 lines
Quoted 6 lines · Ajay Kumar Gupta

urb->transfer_buffer_length and urb->transfer_buffer should be updated based on urb->actual_length.For a fresh and first time urb, actual_length will be zero but for urbs which has been stopped and restarted (as bulk nak scheme does) actual_length may not be zero.

NAK, this is not a problem for the current driver since URBs do not ever
get restarted.

Quoted 6 lines

Resolvable by changing the patch description to not say this is a "fix".

However, since this is a two line change, I think I'll just merge this with the patch adding the bulk RX retry logic.

Yes, that's what should've been done from the start.

Quoted 4 lines
Quoted 1 line

Also, musb_host_tx() doesn't update urb->actual_length -- please fix it too.

That would be a good point if the retry patch touched any TX paths. But it doesn't.

If one teaches musb_start_urb() to restart, that one should at least be
consistent I think.

Quoted 8 lines
Quoted 1 line

Also, you must not clear qh->iso_idx when restarting -- you must not start ISO transfer all over again too. Also, you should not set musb->ep0_state to MUSB_EP0_START again in this case (I agree that control transfers will remain not restartable from an arbitatry place even then).

But the [3/3] patch only adds NAK timeout support for bulk RX. And ISO transfers can't NAK in the first place.

Plus, as noted in a comment you could see in this patch, this only touches (re)start for bulk/interrupt transfers. Not ISO; not control.

All I was asking for was a bit of consistency. Note that I have already
done all the changes that I requested for and can post them.
URB restart is also going to be used for the interrupt transfers BTW.

Quoted 1 line

- Dave

WBR, Sergei