Patchwork e1000e: stop unnecessary polling when using msi-x

login
register
mail settings
Submitter Andy Gospodarek
Date June 18, 2009, 9:57 p.m.
Message ID <20090618215737.GF8515@gospo.rdu.redhat.com>
Download mbox | patch
Permalink /patch/28882/
State Accepted
Delegated to: David Miller
Headers show

Comments

Andy Gospodarek - June 18, 2009, 9:57 p.m.
The last hunk of this commit:

    commit 12d04a3c12b420f23398b4d650127642469a60a6
    Author: Alexander Duyck <alexander.h.duyck@intel.com>
    Date:   Wed Mar 25 22:05:03 2009 +0000

        e1000e: commonize tx cleanup routine to match e1000 & igb

changed the logic for determining if we should call napi_complete or
not at then end of a napi poll.

If the NIC is using MSI-X with no work to do in ->poll, net_rx_action
can just spin indefinitely on older kernels and for 2 jiffies on newer
kernels since napi_complete is never called and budget isn't
decremented.

Discovered and verified while testing driver backport to an older
kernel.

Signed-off-by: Andy Gospodarek <andy@greyhouse.net>

---

 netdev.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)


--
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
Alexander Duyck - June 18, 2009, 10:06 p.m.
Andy Gospodarek wrote:
> The last hunk of this commit:
> 
>     commit 12d04a3c12b420f23398b4d650127642469a60a6
>     Author: Alexander Duyck <alexander.h.duyck@intel.com>
>     Date:   Wed Mar 25 22:05:03 2009 +0000
> 
>         e1000e: commonize tx cleanup routine to match e1000 & igb
> 
> changed the logic for determining if we should call napi_complete or
> not at then end of a napi poll.
> 
> If the NIC is using MSI-X with no work to do in ->poll, net_rx_action
> can just spin indefinitely on older kernels and for 2 jiffies on newer
> kernels since napi_complete is never called and budget isn't
> decremented.
> 
> Discovered and verified while testing driver backport to an older
> kernel.
> 
> Signed-off-by: Andy Gospodarek <andy@greyhouse.net>
> 
> ---
> 
>  netdev.c |    2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/net/e1000e/netdev.c b/drivers/net/e1000e/netdev.c
> index 677f604..679885a 100644
> --- a/drivers/net/e1000e/netdev.c
> +++ b/drivers/net/e1000e/netdev.c
> @@ -1997,7 +1997,7 @@ static int e1000_clean(struct napi_struct *napi, int budget)
>  	struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
>  	struct e1000_hw *hw = &adapter->hw;
>  	struct net_device *poll_dev = adapter->netdev;
> -	int tx_cleaned = 0, work_done = 0;
> +	int tx_cleaned = 1, work_done = 0;
>  
>  	adapter = netdev_priv(poll_dev);
> 
> 
This is most certainly a bug, and the fix is correct.  The logic for 
tx_cleaned is supposed to default to true in the event that there are no 
   tx queues cleaned as part of the polling routine.

Acked-by: Alexander Duyck <alexander.h.duyck@intel.com>
--
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
David Miller - June 19, 2009, 7:24 a.m.
From: Alexander Duyck <alexander.h.duyck@intel.com>
Date: Thu, 18 Jun 2009 15:06:36 -0700

> Andy Gospodarek wrote:
>> The last hunk of this commit:
>>     commit 12d04a3c12b420f23398b4d650127642469a60a6
>>     Author: Alexander Duyck <alexander.h.duyck@intel.com>
>>     Date:   Wed Mar 25 22:05:03 2009 +0000
>>         e1000e: commonize tx cleanup routine to match e1000 & igb
>> changed the logic for determining if we should call napi_complete or
>> not at then end of a napi poll.
>> If the NIC is using MSI-X with no work to do in ->poll, net_rx_action
>> can just spin indefinitely on older kernels and for 2 jiffies on newer
>> kernels since napi_complete is never called and budget isn't
>> decremented.
>> Discovered and verified while testing driver backport to an older
>> kernel.
>> Signed-off-by: Andy Gospodarek <andy@greyhouse.net>
>> ---
>>  netdev.c |    2 +-
>>  1 file changed, 1 insertion(+), 1 deletion(-)
>> diff --git a/drivers/net/e1000e/netdev.c b/drivers/net/e1000e/netdev.c
>> index 677f604..679885a 100644
>> --- a/drivers/net/e1000e/netdev.c
>> +++ b/drivers/net/e1000e/netdev.c
>> @@ -1997,7 +1997,7 @@ static int e1000_clean(struct napi_struct *napi,
>> int budget)
>>  	struct e1000_adapter *adapter = container_of(napi, struct
>>  	e1000_adapter, napi);
>>  	struct e1000_hw *hw = &adapter->hw;
>>  	struct net_device *poll_dev = adapter->netdev;
>> -	int tx_cleaned = 0, work_done = 0;
>> +	int tx_cleaned = 1, work_done = 0;
>>   	adapter = netdev_priv(poll_dev);
>> 
> This is most certainly a bug, and the fix is correct.  The logic for
> tx_cleaned is supposed to default to true in the event that there are
> no tx queues cleaned as part of the polling routine.
> 
> Acked-by: Alexander Duyck <alexander.h.duyck@intel.com>

Applied, thanks.
--
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
Andy Gospodarek - June 19, 2009, 3:06 p.m.
On Fri, Jun 19, 2009 at 12:24:54AM -0700, David Miller wrote:
> From: Alexander Duyck <alexander.h.duyck@intel.com>
> Date: Thu, 18 Jun 2009 15:06:36 -0700
> 
> > Andy Gospodarek wrote:
> >> The last hunk of this commit:
> >>     commit 12d04a3c12b420f23398b4d650127642469a60a6
> >>     Author: Alexander Duyck <alexander.h.duyck@intel.com>
> >>     Date:   Wed Mar 25 22:05:03 2009 +0000
> >>         e1000e: commonize tx cleanup routine to match e1000 & igb
> >> changed the logic for determining if we should call napi_complete or
> >> not at then end of a napi poll.
> >> If the NIC is using MSI-X with no work to do in ->poll, net_rx_action
> >> can just spin indefinitely on older kernels and for 2 jiffies on newer
> >> kernels since napi_complete is never called and budget isn't
> >> decremented.
> >> Discovered and verified while testing driver backport to an older
> >> kernel.
> >> Signed-off-by: Andy Gospodarek <andy@greyhouse.net>
> >> ---
> >>  netdev.c |    2 +-
> >>  1 file changed, 1 insertion(+), 1 deletion(-)
> >> diff --git a/drivers/net/e1000e/netdev.c b/drivers/net/e1000e/netdev.c
> >> index 677f604..679885a 100644
> >> --- a/drivers/net/e1000e/netdev.c
> >> +++ b/drivers/net/e1000e/netdev.c
> >> @@ -1997,7 +1997,7 @@ static int e1000_clean(struct napi_struct *napi,
> >> int budget)
> >>  	struct e1000_adapter *adapter = container_of(napi, struct
> >>  	e1000_adapter, napi);
> >>  	struct e1000_hw *hw = &adapter->hw;
> >>  	struct net_device *poll_dev = adapter->netdev;
> >> -	int tx_cleaned = 0, work_done = 0;
> >> +	int tx_cleaned = 1, work_done = 0;
> >>   	adapter = netdev_priv(poll_dev);
> >> 
> > This is most certainly a bug, and the fix is correct.  The logic for
> > tx_cleaned is supposed to default to true in the event that there are
> > no tx queues cleaned as part of the polling routine.
> > 
> > Acked-by: Alexander Duyck <alexander.h.duyck@intel.com>
> 
> Applied, thanks.

Thanks, Dave.

I did some more testing with 2.6.30 and ksoftirqd uses 100% of the cpu
once the device is up.  With this patch it does not, so it should
probably go into 2.6.29-stable as well.

--
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
David Miller - June 22, 2009, 8:52 a.m.
From: Andy Gospodarek <andy@greyhouse.net>
Date: Fri, 19 Jun 2009 11:06:34 -0400

> I did some more testing with 2.6.30 and ksoftirqd uses 100% of the cpu
> once the device is up.  With this patch it does not, so it should
> probably go into 2.6.29-stable as well.

Ok, will queue it up.
--
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

Patch

diff --git a/drivers/net/e1000e/netdev.c b/drivers/net/e1000e/netdev.c
index 677f604..679885a 100644
--- a/drivers/net/e1000e/netdev.c
+++ b/drivers/net/e1000e/netdev.c
@@ -1997,7 +1997,7 @@  static int e1000_clean(struct napi_struct *napi, int budget)
 	struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
 	struct e1000_hw *hw = &adapter->hw;
 	struct net_device *poll_dev = adapter->netdev;
-	int tx_cleaned = 0, work_done = 0;
+	int tx_cleaned = 1, work_done = 0;
 
 	adapter = netdev_priv(poll_dev);