Re: [PATCH v5 net-next] net: mana: Add a driver for Microsoft Azure Network Adapter (MANA)

Re: [PATCH v5 net-next] net: mana: Add a driver for Microsoft Azure Network Adapter (MANA)

[Jakub Kicinski]

On Mon, 12 Apr 2021 19:35:09 -0700 Dexuan Cui wrote:
> +	apc->port_st_save = apc->port_is_up;
> +	apc->port_is_up = false;
> +	apc->start_remove = true;
> +
> +	/* Ensure port state updated before txq state */
> +	smp_wmb();
> +
> +	netif_tx_disable(ndev);

In your napi poll method there is no barrier between port_is_up check
and netif_tx_queue_stopped().

> +	netif_carrier_off(ndev);
> +
> +	/* No packet can be transmitted now since apc->port_is_up is false.
> +	 * There is still a tiny chance that mana_poll_tx_cq() can re-enable
> +	 * a txq because it may not timely see apc->port_is_up being cleared
> +	 * to false, but it doesn't matter since mana_start_xmit() drops any
> +	 * new packets due to apc->port_is_up being false.
> +	 *
> +	 * Drain all the in-flight TX packets
> +	 */
> +	for (i = 0; i < apc->num_queues; i++) {
> +		txq = &apc->tx_qp[i].txq;
> +
> +		while (atomic_read(&txq->pending_sends) > 0)
> +			usleep_range(1000, 2000);
> +	}

> +		/* All cleanup actions should stay after rtnl_lock(), otherwise
> +		 * other functions may access partially cleaned up data.
> +		 */
> +		rtnl_lock();
> +
> +		mana_detach(ndev);
> +
> +		unregister_netdevice(ndev);
> +
> +		rtnl_unlock();

I find the resource management somewhat strange. Why is mana_attach()
and mana_detach() called at probe/remove time, and not when the
interface is brought up? Presumably when the user ifdowns the interface
there is no point holding the resources? Your open/close methods are
rather empty.

> +	if ((eq_addr & PAGE_MASK) != eq_addr)
> +		return -EINVAL;
> +
> +	if ((cq_addr & PAGE_MASK) != cq_addr)
> +		return -EINVAL;
> +
> +	if ((rq_addr & PAGE_MASK) != rq_addr)
> +		return -EINVAL;
> +
> +	if ((sq_addr & PAGE_MASK) != sq_addr)
> +		return -EINVAL;

PAGE_ALIGNED()

RE: [PATCH v5 net-next] net: mana: Add a driver for Microsoft Azure Network Adapter (MANA)

[Dexuan Cui]

> From: Jakub Kicinski <kuba@kernel.org>
> Sent: Tuesday, April 13, 2021 12:03 PM
> 
> On Mon, 12 Apr 2021 19:35:09 -0700 Dexuan Cui wrote:
> > +	apc->port_st_save = apc->port_is_up;
> > +	apc->port_is_up = false;
> > +	apc->start_remove = true;
> > +
> > +	/* Ensure port state updated before txq state */
> > +	smp_wmb();
> > +
> > +	netif_tx_disable(ndev);
> 
> In your napi poll method there is no barrier between port_is_up check
> and netif_tx_queue_stopped().

Thanks for spotting this! We'll make the below change:

--- a/drivers/net/ethernet/microsoft/mana/mana_en.c
+++ b/drivers/net/ethernet/microsoft/mana/mana_en.c
@@ -938,16 +938,19 @@ static void mana_poll_tx_cq(struct mana_cq *cq)
        avail_space = mana_gd_wq_avail_space(gdma_wq);

        /* Ensure tail updated before checking q stop */
        smp_mb();

        net_txq = txq->net_txq;
        txq_stopped = netif_tx_queue_stopped(net_txq);

+       /* Ensure checking txq_stopped before apc->port_is_up. */
+       smp_rmb();
+
        if (txq_stopped && apc->port_is_up && avail_space >= MAX_TX_WQE_SIZE) {
                netif_tx_wake_queue(net_txq);
                apc->eth_stats.wake_queue++;
        }

        if (atomic_sub_return(pkt_transmitted, &txq->pending_sends) < 0)
                WARN_ON_ONCE(1);
 }

> > +	netif_carrier_off(ndev);
> > +
> > +	/* No packet can be transmitted now since apc->port_is_up is false.
> > +	 * There is still a tiny chance that mana_poll_tx_cq() can re-enable
> > +	 * a txq because it may not timely see apc->port_is_up being cleared
> > +	 * to false, but it doesn't matter since mana_start_xmit() drops any
> > +	 * new packets due to apc->port_is_up being false.
> > +	 *
> > +	 * Drain all the in-flight TX packets
> > +	 */
> > +	for (i = 0; i < apc->num_queues; i++) {
> > +		txq = &apc->tx_qp[i].txq;
> > +
> > +		while (atomic_read(&txq->pending_sends) > 0)
> > +			usleep_range(1000, 2000);
> > +	}
> 
> > +		/* All cleanup actions should stay after rtnl_lock(), otherwise
> > +		 * other functions may access partially cleaned up data.
> > +		 */
> > +		rtnl_lock();
> > +
> > +		mana_detach(ndev);
> > +
> > +		unregister_netdevice(ndev);
> > +
> > +		rtnl_unlock();
> 
> I find the resource management somewhat strange. Why is mana_attach()
> and mana_detach() called at probe/remove time, and not when the
> interface is brought up? Presumably when the user ifdowns the interface
> there is no point holding the resources? Your open/close methods are
> rather empty.

Thanks for the suggestion! Will move the functions to open/close().

> > +	if ((eq_addr & PAGE_MASK) != eq_addr)
> > +		return -EINVAL;
> > +
> > +	if ((cq_addr & PAGE_MASK) != cq_addr)
> > +		return -EINVAL;
> > +
> > +	if ((rq_addr & PAGE_MASK) != rq_addr)
> > +		return -EINVAL;
> > +
> > +	if ((sq_addr & PAGE_MASK) != sq_addr)
> > +		return -EINVAL;

Will change to PAGE_ALIGNED(). 

Thanks,
Dexuan