[ofa-general] Re: [PATCH] IB/ipoib: drain cq in dev_stop

[Michael S. Tsirkin]

> Quoting Roland Dreier <rdreier at cisco.com>:
> Subject: Re: [PATCH] IB/ipoib: drain cq in dev_stop
> 
>  > +	if (unlikely(wc->status != IB_WC_SUCCESS || flush)) {
> 
> Now we have two things to test here, which means we hurt our fast path
> for the rare case.
> 
> What if we overwrite any status of IB_WC_SUCCESS with IB_WC_FLUSH_ERR
> if we're draining a CQ?  I don't see anything obviously wrong with
> this (on top of the patches I just applied):
> 
> diff --git a/drivers/infiniband/ulp/ipoib/ipoib_ib.c b/drivers/infiniband/ulp/ipoib/ipoib_ib.c
> index 8404f05..e24ccb4 100644
> --- a/drivers/infiniband/ulp/ipoib/ipoib_ib.c
> +++ b/drivers/infiniband/ulp/ipoib/ipoib_ib.c
> @@ -557,6 +557,14 @@ void ipoib_drain_cq(struct net_device *dev)
>  	do {
>  		n = ib_poll_cq(priv->cq, IPOIB_NUM_WC, priv->ibwc);
>  		for (i = 0; i < n; ++i) {
> +			/*
> +			 * Convert any successful completions to flush
> +			 * errors to avoid passing packets up the
> +			 * stack after bringing the device down.
> +			 */
> +			if (priv->ibwc[i].status == IB_WC_SUCCESS)
> +				priv->ibwc[i].status = IB_WC_WR_FLUSH_ERR;
> +
>  			if (priv->ibwc[i].wr_id & IPOIB_CM_OP_SRQ)
>  				ipoib_cm_handle_rx_wc(dev, priv->ibwc + i);
>  			else if (priv->ibwc[i].wr_id & IPOIB_OP_RECV)

I love this! Go for it.

-- 
MST