Re: [PATCH v3] usb: dwc3: gadget: trb_dequeue is not updated properly

[Felipe Balbi]

Hi,

fei.yang@xxxxxxxxx writes:
> From: Fei Yang <fei.yang@xxxxxxxxx>
>
> If scatter-gather operation is allowed, a large USB request is split into
> multiple TRBs. These TRBs are chained up by setting DWC3_TRB_CTRL_CHN bit
> except the last one which has DWC3_TRB_CTRL_IOC bit set instead.
> Since only the last TRB has IOC set for the whole USB request, the
> dwc3_gadget_ep_reclaim_trb_sg() gets called only once after the last TRB
> completes and all the TRBs allocated for this request are supposed to be
> reclaimed. However that is not what the current code does.
>
> dwc3_gadget_ep_reclaim_trb_sg() is trying to reclaim all the TRBs in the
> following for-loop,
> 	for_each_sg(sg, s, pending, i) {
> 		trb = &dep->trb_pool[dep->trb_dequeue];
>
>                 if (trb->ctrl & DWC3_TRB_CTRL_HWO)
>                         break;
>
>                 req->sg = sg_next(s);
>                 req->num_pending_sgs--;
>
>                 ret = dwc3_gadget_ep_reclaim_completed_trb(dep, req,
>                                 trb, event, status, chain);
>                 if (ret)
>                         break;
>         }
> but since the interrupt comes only after the last TRB completes, the
> event->status has DEPEVT_STATUS_IOC bit set, so that the for-loop ends for
> the first TRB due to dwc3_gadget_ep_reclaim_completed_trb() returns 1.
> 	if (event->status & DEPEVT_STATUS_IOC)
> 		return 1;
>
> This patch addresses the issue by checking each TRB in function
> dwc3_gadget_ep_reclaim_trb_sg() and maing sure the chained ones are properly
> reclaimed. dwc3_gadget_ep_reclaim_completed_trb() will return 1 Only for the
> last TRB.
>
> Signed-off-by: Fei Yang <fei.yang@xxxxxxxxx>
> Cc: stable <stable@xxxxxxxxxxxxxxx>
> ---
> v2: Better solution is to reclaim chained TRBs in dwc3_gadget_ep_reclaim_trb_sg()
>     and leave the last TRB to the dwc3_gadget_ep_reclaim_completed_trb().
> v3: Checking DWC3_TRB_CTRL_CHN bit for each TRB instead, and making sure that
>     dwc3_gadget_ep_reclaim_completed_trb() returns 1 only for the last TRB.
> ---
>  drivers/usb/dwc3/gadget.c | 11 ++++++++---
>  1 file changed, 8 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/usb/dwc3/gadget.c b/drivers/usb/dwc3/gadget.c
> index 173f532..88eed49 100644
> --- a/drivers/usb/dwc3/gadget.c
> +++ b/drivers/usb/dwc3/gadget.c
> @@ -2394,7 +2394,7 @@ static int dwc3_gadget_ep_reclaim_completed_trb(struct dwc3_ep *dep,
>  	if (event->status & DEPEVT_STATUS_SHORT && !chain)
>  		return 1;
>  
> -	if (event->status & DEPEVT_STATUS_IOC)
> +	if (event->status & DEPEVT_STATUS_IOC && !chain)
>  		return 1;

This will break the situation when we have more SGs than available
TRBs. In that case we set IOC before the last so we have time to update
transfer to append more TRBs.

Please, send me tracepoints

> @@ -2404,11 +2404,12 @@ static int dwc3_gadget_ep_reclaim_trb_sg(struct dwc3_ep *dep,
>  		struct dwc3_request *req, const struct dwc3_event_depevt *event,
>  		int status)
>  {
> -	struct dwc3_trb *trb = &dep->trb_pool[dep->trb_dequeue];
> +	struct dwc3_trb *trb;

should be part of another patch. This is a cleanup that has nothing to
do with this fix.

>  	struct scatterlist *sg = req->sg;
>  	struct scatterlist *s;
>  	unsigned int pending = req->num_pending_sgs;
>  	unsigned int i;
> +	int chain = false;

this could be defined inside for_each_sg() loop like this:

	int chain = trb->ctrl & DWC3_TRB_CTRL_CHN;

> @@ -2419,9 +2420,13 @@ static int dwc3_gadget_ep_reclaim_trb_sg(struct dwc3_ep *dep,
>  
>  		req->sg = sg_next(s);
>  		req->num_pending_sgs--;
> +		if (trb->ctrl & DWC3_TRB_CTRL_CHN)
> +			chain = true;
> +		else
> +			chain = false;
>  
>  		ret = dwc3_gadget_ep_reclaim_completed_trb(dep, req,
> -				trb, event, status, true);
> +				trb, event, status, chain);

this is definitely a valid fix :-) I'm not convinced about that IOC &&
!chain above, however. Also, if "chain" is always trb->ctrl &
DWC3_TRB_CTRL_CHN, we can get rid of that argument altogether and have
the callee handle it internally, but that's something else, subject to
another patch.

-- 
balbi

Attachment: signature.asc
Description: PGP signature