diff mbox

[net-next,1/6] enic: Bug fix: use safe queue shutdown in dev->stop

Message ID 20091219020941.2745.47459.stgit@savbu-pc100.cisco.com
State Superseded, archived
Delegated to: David Miller
Headers show

Commit Message

Scott Feldman Dec. 19, 2009, 2:09 a.m. UTC
enic: Bug fix: use safe queue shutdown in dev->stop

Fix dev->stop shutdown bug where driver was stopping xmit queue and then
disabling intrs.  Fix is to disable intrs first and then stop the xmit
queue, otherwise an interrupt could cause the queue to be rewoken.  Also,
no need to explicitly do queue servicing because queues are cleaned and
reset back to initial state at end of dev->stop.  Servicing queues also
had the side-effect of also rewakening the xmit queue, which is not what
we want.

Signed-off-by: Scott Feldman <scofeldm@cisco.com>
Signed-off-by: Vasanthy Kolluri <vkolluri@cisco.com>
---
 drivers/net/enic/enic.h      |    2 +-
 drivers/net/enic/enic_main.c |   44 +++++++-----------------------------------
 2 files changed, 8 insertions(+), 38 deletions(-)


--
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

Comments

Ben Hutchings Dec. 19, 2009, 3:34 p.m. UTC | #1
On Fri, 2009-12-18 at 18:09 -0800, Scott Feldman wrote:
> enic: Bug fix: use safe queue shutdown in dev->stop
> 
> Fix dev->stop shutdown bug where driver was stopping xmit queue and then
> disabling intrs.  Fix is to disable intrs first and then stop the xmit
> queue, otherwise an interrupt could cause the queue to be rewoken.  Also,
> no need to explicitly do queue servicing because queues are cleaned and
> reset back to initial state at end of dev->stop.  Servicing queues also
> had the side-effect of also rewakening the xmit queue, which is not what
> we want.
[...]
> @@ -1409,16 +1382,18 @@ static int enic_stop(struct net_device *netdev)
>  	unsigned int i;
>  	int err;
>  
> +	for (i = 0; i < enic->intr_count; i++)
> +		vnic_intr_mask(&enic->intr[i]);
> +

I think you need to use synchronize_irq() after this.

>  	del_timer_sync(&enic->notify_timer);
>  
>  	spin_lock(&enic->devcmd_lock);
>  	vnic_dev_disable(enic->vdev);
>  	spin_unlock(&enic->devcmd_lock);
>  	napi_disable(&enic->napi);
> -	netif_stop_queue(netdev);
> -
> -	for (i = 0; i < enic->intr_count; i++)
> -		vnic_intr_mask(&enic->intr[i]);
> +	netif_tx_disable(netdev);
> +	msleep(10);

This sleep is suspicious.

> +	netif_carrier_off(netdev);
[...]

And this should be unnecessary.

Ben.
Scott Feldman Dec. 19, 2009, 8:44 p.m. UTC | #2
On 12/19/09 7:34 AM, "Ben Hutchings" <bhutchings@solarflare.com> wrote:

> On Fri, 2009-12-18 at 18:09 -0800, Scott Feldman wrote:
>> enic: Bug fix: use safe queue shutdown in dev->stop
>> 
>> Fix dev->stop shutdown bug where driver was stopping xmit queue and then
>> disabling intrs.  Fix is to disable intrs first and then stop the xmit
>> queue, otherwise an interrupt could cause the queue to be rewoken.  Also,
>> no need to explicitly do queue servicing because queues are cleaned and
>> reset back to initial state at end of dev->stop.  Servicing queues also
>> had the side-effect of also rewakening the xmit queue, which is not what
>> we want.
> [...]
>> @@ -1409,16 +1382,18 @@ static int enic_stop(struct net_device *netdev)
>> unsigned int i;
>> int err;
>>  
>> + for (i = 0; i < enic->intr_count; i++)
>> +  vnic_intr_mask(&enic->intr[i]);
>> +
> 
> I think you need to use synchronize_irq() after this.

I think you're right.  That's what I want: all intrs masked and no IRQs
running.  I also need to add a PIO read to flush the PIO write to mask.

I want intrs shut down from here on.
 
>> del_timer_sync(&enic->notify_timer);
>>  
>> spin_lock(&enic->devcmd_lock);
>> vnic_dev_disable(enic->vdev);
>> spin_unlock(&enic->devcmd_lock);
>> napi_disable(&enic->napi);
>> - netif_stop_queue(netdev);
>> -
>> - for (i = 0; i < enic->intr_count; i++)
>> -  vnic_intr_mask(&enic->intr[i]);
>> + netif_tx_disable(netdev);
>> + msleep(10);
> 
> This sleep is suspicious.

Ya, that can go.  Netif_tx_disable grabs netif_tx_lock, so we should be
synced with xmit_hard_start.  Forgot to delete that when we had
netif_stop_queue.
 
>> + netif_carrier_off(netdev);
> [...]
> 
> And this should be unnecessary.

We don't have any way to get true link status when the interface is down, so
this forces us to always report link status down when the interface is down.

-scott

--
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/enic/enic.h b/drivers/net/enic/enic.h
index e1c2076..dfd6024 100644
--- a/drivers/net/enic/enic.h
+++ b/drivers/net/enic/enic.h
@@ -34,7 +34,7 @@ 
 
 #define DRV_NAME		"enic"
 #define DRV_DESCRIPTION		"Cisco 10G Ethernet Driver"
-#define DRV_VERSION		"1.1.0.100"
+#define DRV_VERSION		"1.1.0.241"
 #define DRV_COPYRIGHT		"Copyright 2008-2009 Cisco Systems, Inc"
 #define PFX			DRV_NAME ": "
 
diff --git a/drivers/net/enic/enic_main.c b/drivers/net/enic/enic_main.c
index f875751..496e8b6 100644
--- a/drivers/net/enic/enic_main.c
+++ b/drivers/net/enic/enic_main.c
@@ -24,6 +24,7 @@ 
 #include <linux/types.h>
 #include <linux/init.h>
 #include <linux/workqueue.h>
+#include <linux/delay.h>
 #include <linux/pci.h>
 #include <linux/netdevice.h>
 #include <linux/etherdevice.h>
@@ -1084,34 +1085,6 @@  static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
 	return 0;
 }
 
-static void enic_rq_drop_buf(struct vnic_rq *rq,
-	struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
-	int skipped, void *opaque)
-{
-	struct enic *enic = vnic_dev_priv(rq->vdev);
-	struct sk_buff *skb = buf->os_buf;
-
-	if (skipped)
-		return;
-
-	pci_unmap_single(enic->pdev, buf->dma_addr,
-		buf->len, PCI_DMA_FROMDEVICE);
-
-	dev_kfree_skb_any(skb);
-}
-
-static int enic_rq_service_drop(struct vnic_dev *vdev, struct cq_desc *cq_desc,
-	u8 type, u16 q_number, u16 completed_index, void *opaque)
-{
-	struct enic *enic = vnic_dev_priv(vdev);
-
-	vnic_rq_service(&enic->rq[q_number], cq_desc,
-		completed_index, VNIC_RQ_RETURN_DESC,
-		enic_rq_drop_buf, opaque);
-
-	return 0;
-}
-
 static int enic_poll(struct napi_struct *napi, int budget)
 {
 	struct enic *enic = container_of(napi, struct enic, napi);
@@ -1409,16 +1382,18 @@  static int enic_stop(struct net_device *netdev)
 	unsigned int i;
 	int err;
 
+	for (i = 0; i < enic->intr_count; i++)
+		vnic_intr_mask(&enic->intr[i]);
+
 	del_timer_sync(&enic->notify_timer);
 
 	spin_lock(&enic->devcmd_lock);
 	vnic_dev_disable(enic->vdev);
 	spin_unlock(&enic->devcmd_lock);
 	napi_disable(&enic->napi);
-	netif_stop_queue(netdev);
-
-	for (i = 0; i < enic->intr_count; i++)
-		vnic_intr_mask(&enic->intr[i]);
+	netif_tx_disable(netdev);
+	msleep(10);
+	netif_carrier_off(netdev);
 
 	for (i = 0; i < enic->wq_count; i++) {
 		err = vnic_wq_disable(&enic->wq[i]);
@@ -1436,11 +1411,6 @@  static int enic_stop(struct net_device *netdev)
 	spin_unlock(&enic->devcmd_lock);
 	enic_free_intr(enic);
 
-	(void)vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
-		-1, enic_rq_service_drop, NULL);
-	(void)vnic_cq_service(&enic->cq[ENIC_CQ_WQ],
-		-1, enic_wq_service, NULL);
-
 	for (i = 0; i < enic->wq_count; i++)
 		vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
 	for (i = 0; i < enic->rq_count; i++)