diff mbox

[5/5] net: sh_eth: use NAPI

Message ID 4F62FDDA.9070708@renesas.com
State Changes Requested, archived
Delegated to: David Miller
Headers show

Commit Message

Yoshihiro Shimoda March 16, 2012, 8:46 a.m. UTC 
This patch modifies the driver to use NAPI.

Signed-off-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>
---
 drivers/net/ethernet/renesas/sh_eth.c |   88 +++++++++++++++++++++++----------
 drivers/net/ethernet/renesas/sh_eth.h |    3 +
 2 files changed, 64 insertions(+), 27 deletions(-)

Comments

David Miller March 16, 2012, 9:15 a.m. UTC | #1 
From: "Shimoda, Yoshihiro" <yoshihiro.shimoda.uh@renesas.com>
Date: Fri, 16 Mar 2012 17:46:18 +0900

sh_eth_interrupt takes mdp->lock, and:

> +static int sh_eth_poll(struct napi_struct *napi, int budget)
> +{
> +	struct sh_eth_private *mdp = container_of(napi, struct sh_eth_private,
> +						  napi);
> +	struct net_device *ndev = mdp->ndev;
> +	struct sh_eth_cpu_data *cd = mdp->cd;
> +	int work_done = 0, txfree_num;
> +	u32 intr_status = sh_eth_read(ndev, EESR);
> +
> +	spin_lock(&mdp->lock);

sh_eth_poll() runs from software interrupt context, therefore
this can deadlock.

Even though you turned off interrupts in sh_eth_interrupt to
enable NAPI mode, this can still happen, interrupts can be
stuck in the interrupt controller, another device can be on
the same interrupt line, etc.  Therefore you must handle
this properly.

I would suggest _NOT_ fixing this by taking the lock with interrupts
disabled in sh_eth_poll(), that defeats the whole prupose of
converting to NAPI and this would also require you to change back
the dev_kfree_skb() to dev_kfree_skb_irq().

Instead, do what other drivers do, make your interrupt handler run
completely lockless and have a sophisticated quiescence sequence
when you want to sync interrupts off and make sure no async contexts
are still running in the interrupt handler.  See tg3.c for one
example of this.

I'm toss this entire patch series, it needs a lot more work.
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Yoshihiro Shimoda March 16, 2012, 10:46 a.m. UTC | #2 
2012/03/16 18:15, David Miller wrote:
> From: "Shimoda, Yoshihiro" <yoshihiro.shimoda.uh@renesas.com>
> Date: Fri, 16 Mar 2012 17:46:18 +0900
> 
> sh_eth_interrupt takes mdp->lock, and:
> 
>> +static int sh_eth_poll(struct napi_struct *napi, int budget)
>> +{
>> +	struct sh_eth_private *mdp = container_of(napi, struct sh_eth_private,
>> +						  napi);
>> +	struct net_device *ndev = mdp->ndev;
>> +	struct sh_eth_cpu_data *cd = mdp->cd;
>> +	int work_done = 0, txfree_num;
>> +	u32 intr_status = sh_eth_read(ndev, EESR);
>> +
>> +	spin_lock(&mdp->lock);
> 
> sh_eth_poll() runs from software interrupt context, therefore
> this can deadlock.

Thank you for the point. I will fix it.

> Even though you turned off interrupts in sh_eth_interrupt to
> enable NAPI mode, this can still happen, interrupts can be
> stuck in the interrupt controller, another device can be on
> the same interrupt line, etc.  Therefore you must handle
> this properly.

I see. I wrote a comment below in sh_eth_interrupt, but this
is a wrong comment.

+			/* Disable all interrupts */
+			sh_eth_write(ndev, 0, EESIPR);

It only disable the interrupts of the controller's channel.
It doesn't disable other devices. So, I will fix the comment.

> I would suggest _NOT_ fixing this by taking the lock with interrupts
> disabled in sh_eth_poll(), that defeats the whole prupose of
> converting to NAPI and this would also require you to change back
> the dev_kfree_skb() to dev_kfree_skb_irq().

I understood it. I will fix it.

> Instead, do what other drivers do, make your interrupt handler run
> completely lockless and have a sophisticated quiescence sequence
> when you want to sync interrupts off and make sure no async contexts
> are still running in the interrupt handler.  See tg3.c for one
> example of this.

Thank you for your suggestion. I will see the tg3.c.

> I'm toss this entire patch series, it needs a lot more work.

I understood it. I wlll resubmit entire patch series after I fix the code.

Best regards,
Yoshihiro Shimoda
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
diff mbox

Patch

diff --git a/drivers/net/ethernet/renesas/sh_eth.c b/drivers/net/ethernet/renesas/sh_eth.c
index 62458d9..02878af 100644
--- a/drivers/net/ethernet/renesas/sh_eth.c
+++ b/drivers/net/ethernet/renesas/sh_eth.c
@@ -871,7 +871,7 @@  static int sh_eth_txfree(struct net_device *ndev)
 		if (mdp->tx_skbuff[entry]) {
 			dma_unmap_single(&ndev->dev, txdesc->addr,
 					 txdesc->buffer_length, DMA_TO_DEVICE);
-			dev_kfree_skb_irq(mdp->tx_skbuff[entry]);
+			dev_kfree_skb(mdp->tx_skbuff[entry]);
 			mdp->tx_skbuff[entry] = NULL;
 			freeNum++;
 		}
@@ -886,7 +886,7 @@  static int sh_eth_txfree(struct net_device *ndev)
 }

 /* Packet receive function */
-static int sh_eth_rx(struct net_device *ndev)
+static int sh_eth_rx(struct net_device *ndev, int *work, int budget)
 {
 	struct sh_eth_private *mdp = netdev_priv(ndev);
 	struct sh_eth_rxdesc *rxdesc;
@@ -898,7 +898,8 @@  static int sh_eth_rx(struct net_device *ndev)
 	u32 desc_status;

 	rxdesc = &mdp->rx_ring[entry];
-	while (!(rxdesc->status & cpu_to_edmac(mdp, RD_RACT))) {
+	while (!(rxdesc->status & cpu_to_edmac(mdp, RD_RACT)) &&
+	       *work < budget) {
 		desc_status = edmac_to_cpu(mdp, rxdesc->status);
 		pkt_len = rxdesc->frame_length;

@@ -934,13 +935,16 @@  static int sh_eth_rx(struct net_device *ndev)
 				skb_reserve(skb, NET_IP_ALIGN);
 			skb_put(skb, pkt_len);
 			skb->protocol = eth_type_trans(skb, ndev);
-			netif_rx(skb);
-			ndev->stats.rx_packets++;
-			ndev->stats.rx_bytes += pkt_len;
+			if (netif_receive_skb(skb) == NET_RX_DROP) {
+				ndev->stats.rx_dropped++;
+			} else {
+				ndev->stats.rx_packets++;
+				ndev->stats.rx_bytes += pkt_len;
+			}
 		}
-		rxdesc->status |= cpu_to_edmac(mdp, RD_RACT);
 		entry = (++mdp->cur_rx) % RX_RING_SIZE;
 		rxdesc = &mdp->rx_ring[entry];
+		(*work)++;
 	}

 	/* Refill the Rx ring buffers. */
@@ -972,7 +976,7 @@  static int sh_eth_rx(struct net_device *ndev)

 	/* Restart Rx engine if stopped. */
 	/* If we don't need to check status, don't. -KDU */
-	if (!(sh_eth_read(ndev, EDRRR) & EDRRR_R)) {
+	if (*work < budget && !(sh_eth_read(ndev, EDRRR) & EDRRR_R)) {
 		/* fix the values for the next receiving */
 		mdp->cur_rx = mdp->dirty_rx = (sh_eth_read(ndev, RDFAR) -
 					       sh_eth_read(ndev, RDLAR)) >> 4;
@@ -1128,38 +1132,61 @@  static irqreturn_t sh_eth_interrupt(int irq, void *netdev)

 	/* Get interrpt stat */
 	intr_status = sh_eth_read(ndev, EESR);
-	/* Clear interrupt */
 	if (intr_status & (EESR_FRC | EESR_RMAF | EESR_RRF |
 			EESR_RTLF | EESR_RTSF | EESR_PRE | EESR_CERF |
 			cd->tx_check | cd->eesr_err_check)) {
-		sh_eth_write(ndev, intr_status, EESR);
+		if (napi_schedule_prep(&mdp->napi)) {
+			/* Disable all interrupts */
+			sh_eth_write(ndev, 0, EESIPR);
+			__napi_schedule(&mdp->napi);
+		}
 		ret = IRQ_HANDLED;
-	} else
-		goto other_irq;
-
-	if (intr_status & (EESR_FRC | /* Frame recv*/
-			EESR_RMAF | /* Multi cast address recv*/
-			EESR_RRF  | /* Bit frame recv */
-			EESR_RTLF | /* Long frame recv*/
-			EESR_RTSF | /* short frame recv */
-			EESR_PRE  | /* PHY-LSI recv error */
-			EESR_CERF)){ /* recv frame CRC error */
-		sh_eth_rx(ndev);
 	}

-	/* Tx Check */
-	if (intr_status & cd->tx_check) {
-		sh_eth_txfree(ndev);
+	spin_unlock(&mdp->lock);
+
+	return ret;
+}
+
+static int sh_eth_poll(struct napi_struct *napi, int budget)
+{
+	struct sh_eth_private *mdp = container_of(napi, struct sh_eth_private,
+						  napi);
+	struct net_device *ndev = mdp->ndev;
+	struct sh_eth_cpu_data *cd = mdp->cd;
+	int work_done = 0, txfree_num;
+	u32 intr_status = sh_eth_read(ndev, EESR);
+
+	spin_lock(&mdp->lock);
+
+	/* Clear interrupt flags */
+	sh_eth_write(ndev, intr_status, EESR);
+
+	/* check txdesc */
+	txfree_num = sh_eth_txfree(ndev);
+	if (txfree_num)
 		netif_wake_queue(ndev);
-	}

+	/* check rxdesc */
+	sh_eth_rx(ndev, &work_done, budget);
+
+	/* check error flags */
 	if (intr_status & cd->eesr_err_check)
 		sh_eth_error(ndev, intr_status);

-other_irq:
+	/* get current interrupt flags */
+	intr_status = sh_eth_read(ndev, EESR);
 	spin_unlock(&mdp->lock);

-	return ret;
+	/* check whether the controller doesn't have any events */
+	if (!txfree_num && !(intr_status & cd->eesr_err_check) &&
+	    work_done < budget) {
+		napi_complete(napi);
+		/* Enable all interrupts */
+		sh_eth_write(ndev, cd->eesipr_value, EESIPR);
+	}
+
+	return work_done;
 }

 /* PHY state control function */
@@ -1380,6 +1407,8 @@  static int sh_eth_open(struct net_device *ndev)

 	pm_runtime_get_sync(&mdp->pdev->dev);

+	napi_enable(&mdp->napi);
+
 	ret = request_irq(ndev->irq, sh_eth_interrupt,
 #if defined(CONFIG_CPU_SUBTYPE_SH7763) || \
 	defined(CONFIG_CPU_SUBTYPE_SH7764) || \
@@ -1520,6 +1549,8 @@  static int sh_eth_close(struct net_device *ndev)
 		phy_disconnect(mdp->phydev);
 	}

+	napi_disable(&mdp->napi);
+
 	free_irq(ndev->irq, ndev);

 	/* Free all the skbuffs in the Rx queue. */
@@ -2152,6 +2183,9 @@  static int sh_eth_drv_probe(struct platform_device *pdev)
 #endif
 	sh_eth_set_default_cpu_data(mdp->cd);

+	mdp->ndev = ndev;
+	netif_napi_add(ndev, &mdp->napi, sh_eth_poll, SH_ETH_NAPI_WEIGHT);
+
 	/* set function */
 	ndev->netdev_ops = &sh_eth_netdev_ops;
 	SET_ETHTOOL_OPS(ndev, &sh_eth_ethtool_ops);
diff --git a/drivers/net/ethernet/renesas/sh_eth.h b/drivers/net/ethernet/renesas/sh_eth.h
index 667bdd2..9f4c91e 100644
--- a/drivers/net/ethernet/renesas/sh_eth.h
+++ b/drivers/net/ethernet/renesas/sh_eth.h
@@ -31,6 +31,7 @@ 
 #define PKT_BUF_SZ		1538
 #define SH_ETH_TSU_TIMEOUT_MS	500
 #define SH_ETH_TSU_CAM_ENTRIES	32
+#define SH_ETH_NAPI_WEIGHT	32

 enum {
 	/* E-DMAC registers */
@@ -716,6 +717,8 @@  struct sh_eth_private {
 	int duplex;
 	int port;		/* for TSU */
 	int vlan_num_ids;	/* for VLAN tag filter */
+	struct napi_struct napi;
+	struct net_device *ndev;

 	unsigned no_ether_link:1;
 	unsigned ether_link_active_low:1;