diff mbox

[dpdk-dev,1/2] rte_ethdev: update link status (speed, duplex, link_up) after rte_eth_dev_start

Message ID 1415381511-43364-2-git-send-email-jyu@vmware.com (mailing list archive)
State Accepted, archived
Headers show

Commit Message

Jia Yu Nov. 7, 2014, 5:31 p.m. UTC
Since LSR interrupt is disabled by pmd drivers, link status in rte_eth_device is always down.
Bond slave_configure() enables LSR interrupt on devices to get notification if link status
changes. However, the LSC interrupt at device start time is still lost.

In this fix, call link_update to read link status from hardware
register at device start time.

Issue:
Change-Id: Ib57a1c9114f922485c7b0f4338bfe7b3d3f87d65
Signed-off-by: Jia Yu <jyu@vmware.com>
---
 lib/librte_ether/rte_ethdev.c | 4 ++++
 1 file changed, 4 insertions(+)

Comments

Helin Zhang Nov. 12, 2014, 3:57 a.m. UTC | #1
Hi Jia

> -----Original Message-----
> From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Jia Yu
> Sent: Saturday, November 8, 2014 1:32 AM
> To: dev@dpdk.org
> Subject: [dpdk-dev] [PATCH 1/2] rte_ethdev: update link status (speed, duplex,
> link_up) after rte_eth_dev_start
> 
> Since LSR interrupt is disabled by pmd drivers, link status in rte_eth_device is
> always down.
If LSC interrupt is disabled by default, it will poll the link status during the initialization
or in dev_start, and then the link status should he correct. If I am not wrong.

> Bond slave_configure() enables LSR interrupt on devices to get notification if link
> status changes. However, the LSC interrupt at device start time is still lost.
Before enabling interrupt for LSC, the link status should be polled. So after the port
startup, the link status should be there.

> 
> In this fix, call link_update to read link status from hardware register at device
> start time.
Could you help to explain this code changes a bit more? Why we need it?

> 
> Issue:
> Change-Id: Ib57a1c9114f922485c7b0f4338bfe7b3d3f87d65
> Signed-off-by: Jia Yu <jyu@vmware.com>
> ---
>  lib/librte_ether/rte_ethdev.c | 4 ++++
>  1 file changed, 4 insertions(+)
> 
> diff --git a/lib/librte_ether/rte_ethdev.c b/lib/librte_ether/rte_ethdev.c index
> ff1c769..6c01b02 100644
> --- a/lib/librte_ether/rte_ethdev.c
> +++ b/lib/librte_ether/rte_ethdev.c
> @@ -869,6 +869,10 @@ rte_eth_dev_start(uint8_t port_id)
> 
>  	rte_eth_dev_config_restore(port_id);
> 
> +	if (dev->data->dev_conf.intr_conf.lsc != 0) {
> +		FUNC_PTR_OR_ERR_RET(*dev->dev_ops->link_update, -ENOTSUP);
> +		(*dev->dev_ops->link_update)(dev, 0);
> +	}
>  	return 0;
>  }
> 
> --
> 1.9.1

Regards,
Helin
Thomas Monjalon Jan. 30, 2015, 10:28 a.m. UTC | #2
Jia, any news on this patchset?

2014-11-12 03:57, Zhang, Helin:
> Hi Jia
> 
> > -----Original Message-----
> > From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Jia Yu
> > Sent: Saturday, November 8, 2014 1:32 AM
> > To: dev@dpdk.org
> > Subject: [dpdk-dev] [PATCH 1/2] rte_ethdev: update link status (speed, duplex,
> > link_up) after rte_eth_dev_start
> > 
> > Since LSR interrupt is disabled by pmd drivers, link status in rte_eth_device is
> > always down.
> If LSC interrupt is disabled by default, it will poll the link status during the initialization
> or in dev_start, and then the link status should he correct. If I am not wrong.
> 
> > Bond slave_configure() enables LSR interrupt on devices to get notification if link
> > status changes. However, the LSC interrupt at device start time is still lost.
> Before enabling interrupt for LSC, the link status should be polled. So after the port
> startup, the link status should be there.
> 
> > 
> > In this fix, call link_update to read link status from hardware register at device
> > start time.
> Could you help to explain this code changes a bit more? Why we need it?
> 
> > 
> > Issue:
> > Change-Id: Ib57a1c9114f922485c7b0f4338bfe7b3d3f87d65
> > Signed-off-by: Jia Yu <jyu@vmware.com>
> > ---
> >  lib/librte_ether/rte_ethdev.c | 4 ++++
> >  1 file changed, 4 insertions(+)
> > 
> > diff --git a/lib/librte_ether/rte_ethdev.c b/lib/librte_ether/rte_ethdev.c index
> > ff1c769..6c01b02 100644
> > --- a/lib/librte_ether/rte_ethdev.c
> > +++ b/lib/librte_ether/rte_ethdev.c
> > @@ -869,6 +869,10 @@ rte_eth_dev_start(uint8_t port_id)
> > 
> >  	rte_eth_dev_config_restore(port_id);
> > 
> > +	if (dev->data->dev_conf.intr_conf.lsc != 0) {
> > +		FUNC_PTR_OR_ERR_RET(*dev->dev_ops->link_update, -ENOTSUP);
> > +		(*dev->dev_ops->link_update)(dev, 0);
> > +	}
> >  	return 0;
> >  }
> > 
> > --
> > 1.9.1
> 
> Regards,
> Helin
Jia Yu Feb. 3, 2015, 8 a.m. UTC | #3
My answer to Helin¹s comments:

This patch is needed for bond slave devices or other devices, when LSC
interrupt is enabled.
 1. slave_configure()  -> slave_eth_dev->Š.lsc = 1

 2. rte_eth_link_get() reads dev_link from eth_dev, when lsc interrupt is
enabled. However, the dev_link on eth_dev has not be initialized and
showed link down state. This patch initializes the device¹s dev_link at
rte_eth_dev_start time.

Please let me know if you have further questions/comments.

Thanks,
Jia

On 1/30/15, 2:28 AM, "Thomas Monjalon" <thomas.monjalon@6wind.com> wrote:

>Jia, any news on this patchset?
>
>2014-11-12 03:57, Zhang, Helin:
>> Hi Jia
>> 
>> > -----Original Message-----
>> > From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Jia Yu
>> > Sent: Saturday, November 8, 2014 1:32 AM
>> > To: dev@dpdk.org
>> > Subject: [dpdk-dev] [PATCH 1/2] rte_ethdev: update link status
>>(speed, duplex,
>> > link_up) after rte_eth_dev_start
>> > 
>> > Since LSR interrupt is disabled by pmd drivers, link status in
>>rte_eth_device is
>> > always down.
>> If LSC interrupt is disabled by default, it will poll the link status
>>during the initialization
>> or in dev_start, and then the link status should he correct. If I am
>>not wrong.
>> 
>> > Bond slave_configure() enables LSR interrupt on devices to get
>>notification if link
>> > status changes. However, the LSC interrupt at device start time is
>>still lost.
>> Before enabling interrupt for LSC, the link status should be polled. So
>>after the port
>> startup, the link status should be there.
>> 
>> > 
>> > In this fix, call link_update to read link status from hardware
>>register at device
>> > start time.
>> Could you help to explain this code changes a bit more? Why we need it?
>> 
>> > 
>> > Issue:
>> > Change-Id: Ib57a1c9114f922485c7b0f4338bfe7b3d3f87d65
>> > Signed-off-by: Jia Yu <jyu@vmware.com>
>> > ---
>> >  lib/librte_ether/rte_ethdev.c | 4 ++++
>> >  1 file changed, 4 insertions(+)
>> > 
>> > diff --git a/lib/librte_ether/rte_ethdev.c
>>b/lib/librte_ether/rte_ethdev.c index
>> > ff1c769..6c01b02 100644
>> > --- a/lib/librte_ether/rte_ethdev.c
>> > +++ b/lib/librte_ether/rte_ethdev.c
>> > @@ -869,6 +869,10 @@ rte_eth_dev_start(uint8_t port_id)
>> > 
>> >  	rte_eth_dev_config_restore(port_id);
>> > 
>> > +	if (dev->data->dev_conf.intr_conf.lsc != 0) {
>> > +		FUNC_PTR_OR_ERR_RET(*dev->dev_ops->link_update, -ENOTSUP);
>> > +		(*dev->dev_ops->link_update)(dev, 0);
>> > +	}
>> >  	return 0;
>> >  }
>> > 
>> > --
>> > 1.9.1
>> 
>> Regards,
>> Helin
>
Helin Zhang Feb. 3, 2015, 8:35 a.m. UTC | #4
> -----Original Message-----
> From: Jia Yu [mailto:jyu@vmware.com]
> Sent: Tuesday, February 3, 2015 4:00 PM
> To: Zhang, Helin
> Cc: dev@dpdk.org; Thomas Monjalon
> Subject: Re: [dpdk-dev] [PATCH 1/2] rte_ethdev: update link status (speed,
> duplex, link_up) after rte_eth_dev_start
> 
> My answer to Helin¹s comments:
> 
> This patch is needed for bond slave devices or other devices, when LSC
> interrupt is enabled.
>  1. slave_configure()  -> slave_eth_dev->Š.lsc = 1
> 
>  2. rte_eth_link_get() reads dev_link from eth_dev, when lsc interrupt is
> enabled. However, the dev_link on eth_dev has not be initialized and showed
> link down state. This patch initializes the device¹s dev_link at
> rte_eth_dev_start time.
So the link update is for bond only? But your code changes is in rte_ethdev, it will
be used for all PMDs. For hardware NIC (e.g. 82599), this link update is not needed
at all. It just need to wait the link status change event. So can those code changes
be put in librte_pmd_bond, but not in librte_ether?

Regards,
Helin

> 
> Please let me know if you have further questions/comments.
> 
> Thanks,
> Jia
> 
> On 1/30/15, 2:28 AM, "Thomas Monjalon" <thomas.monjalon@6wind.com>
> wrote:
> 
> >Jia, any news on this patchset?
> >
> >2014-11-12 03:57, Zhang, Helin:
> >> Hi Jia
> >>
> >> > -----Original Message-----
> >> > From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Jia Yu
> >> > Sent: Saturday, November 8, 2014 1:32 AM
> >> > To: dev@dpdk.org
> >> > Subject: [dpdk-dev] [PATCH 1/2] rte_ethdev: update link status
> >>(speed, duplex,
> >> > link_up) after rte_eth_dev_start
> >> >
> >> > Since LSR interrupt is disabled by pmd drivers, link status in
> >>rte_eth_device is
> >> > always down.
> >> If LSC interrupt is disabled by default, it will poll the link status
> >>during the initialization  or in dev_start, and then the link status
> >>should he correct. If I am not wrong.
> >>
> >> > Bond slave_configure() enables LSR interrupt on devices to get
> >>notification if link
> >> > status changes. However, the LSC interrupt at device start time is
> >>still lost.
> >> Before enabling interrupt for LSC, the link status should be polled.
> >>So after the port  startup, the link status should be there.
> >>
> >> >
> >> > In this fix, call link_update to read link status from hardware
> >>register at device
> >> > start time.
> >> Could you help to explain this code changes a bit more? Why we need it?
> >>
> >> >
> >> > Issue:
> >> > Change-Id: Ib57a1c9114f922485c7b0f4338bfe7b3d3f87d65
> >> > Signed-off-by: Jia Yu <jyu@vmware.com>
> >> > ---
> >> >  lib/librte_ether/rte_ethdev.c | 4 ++++
> >> >  1 file changed, 4 insertions(+)
> >> >
> >> > diff --git a/lib/librte_ether/rte_ethdev.c
> >>b/lib/librte_ether/rte_ethdev.c index
> >> > ff1c769..6c01b02 100644
> >> > --- a/lib/librte_ether/rte_ethdev.c
> >> > +++ b/lib/librte_ether/rte_ethdev.c
> >> > @@ -869,6 +869,10 @@ rte_eth_dev_start(uint8_t port_id)
> >> >
> >> >  	rte_eth_dev_config_restore(port_id);
> >> >
> >> > +	if (dev->data->dev_conf.intr_conf.lsc != 0) {
> >> > +		FUNC_PTR_OR_ERR_RET(*dev->dev_ops->link_update,
> -ENOTSUP);
> >> > +		(*dev->dev_ops->link_update)(dev, 0);
> >> > +	}
> >> >  	return 0;
> >> >  }
> >> >
> >> > --
> >> > 1.9.1
> >>
> >> Regards,
> >> Helin
> >
Jia Yu Feb. 3, 2015, 6:53 p.m. UTC | #5
Helin,

Thanks for comment. Any device that enabled LSC needs this fix, otherwise
they all need to call link_update separately. We cannot assume that only
Bond enables LSC interrupt.

The fix will not be used for all PMDs, as it explicitly checks if
(dev->data->dev_conf.intr_conf.lsc != 0). Therefore, for hardware NIC
(e.g. 82599) that disabled lsc by default, the link_update callback will
not be executed. Please let me know if you have other concerns.

Thanks,
Jia

On 2/3/15, 12:35 AM, "Zhang, Helin" <helin.zhang@intel.com> wrote:

>

>

>> -----Original Message-----

>> From: Jia Yu [mailto:jyu@vmware.com]

>> Sent: Tuesday, February 3, 2015 4:00 PM

>> To: Zhang, Helin

>> Cc: dev@dpdk.org; Thomas Monjalon

>> Subject: Re: [dpdk-dev] [PATCH 1/2] rte_ethdev: update link status

>>(speed,

>> duplex, link_up) after rte_eth_dev_start

>> 

>> My answer to Helin¹s comments:

>> 

>> This patch is needed for bond slave devices or other devices, when LSC

>> interrupt is enabled.

>>  1. slave_configure()  -> slave_eth_dev->Š.lsc = 1

>> 

>>  2. rte_eth_link_get() reads dev_link from eth_dev, when lsc interrupt

>>is

>> enabled. However, the dev_link on eth_dev has not be initialized and

>>showed

>> link down state. This patch initializes the device¹s dev_link at

>> rte_eth_dev_start time.

>So the link update is for bond only? But your code changes is in

>rte_ethdev, it will

>be used for all PMDs. For hardware NIC (e.g. 82599), this link update is

>not needed

>at all. It just need to wait the link status change event. So can those

>code changes

>be put in librte_pmd_bond, but not in librte_ether?

>

>Regards,

>Helin

>

>> 

>> Please let me know if you have further questions/comments.

>> 

>> Thanks,

>> Jia

>> 

>> On 1/30/15, 2:28 AM, "Thomas Monjalon" <thomas.monjalon@6wind.com>

>> wrote:

>> 

>> >Jia, any news on this patchset?

>> >

>> >2014-11-12 03:57, Zhang, Helin:

>> >> Hi Jia

>> >>

>> >> > -----Original Message-----

>> >> > From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Jia Yu

>> >> > Sent: Saturday, November 8, 2014 1:32 AM

>> >> > To: dev@dpdk.org

>> >> > Subject: [dpdk-dev] [PATCH 1/2] rte_ethdev: update link status

>> >>(speed, duplex,

>> >> > link_up) after rte_eth_dev_start

>> >> >

>> >> > Since LSR interrupt is disabled by pmd drivers, link status in

>> >>rte_eth_device is

>> >> > always down.

>> >> If LSC interrupt is disabled by default, it will poll the link status

>> >>during the initialization  or in dev_start, and then the link status

>> >>should he correct. If I am not wrong.

>> >>

>> >> > Bond slave_configure() enables LSR interrupt on devices to get

>> >>notification if link

>> >> > status changes. However, the LSC interrupt at device start time is

>> >>still lost.

>> >> Before enabling interrupt for LSC, the link status should be polled.

>> >>So after the port  startup, the link status should be there.

>> >>

>> >> >

>> >> > In this fix, call link_update to read link status from hardware

>> >>register at device

>> >> > start time.

>> >> Could you help to explain this code changes a bit more? Why we need

>>it?

>> >>

>> >> >

>> >> > Issue:

>> >> > Change-Id: Ib57a1c9114f922485c7b0f4338bfe7b3d3f87d65

>> >> > Signed-off-by: Jia Yu <jyu@vmware.com>

>> >> > ---

>> >> >  lib/librte_ether/rte_ethdev.c | 4 ++++

>> >> >  1 file changed, 4 insertions(+)

>> >> >

>> >> > diff --git a/lib/librte_ether/rte_ethdev.c

>> >>b/lib/librte_ether/rte_ethdev.c index

>> >> > ff1c769..6c01b02 100644

>> >> > --- a/lib/librte_ether/rte_ethdev.c

>> >> > +++ b/lib/librte_ether/rte_ethdev.c

>> >> > @@ -869,6 +869,10 @@ rte_eth_dev_start(uint8_t port_id)

>> >> >

>> >> >  	rte_eth_dev_config_restore(port_id);

>> >> >

>> >> > +	if (dev->data->dev_conf.intr_conf.lsc != 0) {

>> >> > +		FUNC_PTR_OR_ERR_RET(*dev->dev_ops->link_update,

>> -ENOTSUP);

>> >> > +		(*dev->dev_ops->link_update)(dev, 0);

>> >> > +	}

>> >> >  	return 0;

>> >> >  }

>> >> >

>> >> > --

>> >> > 1.9.1

>> >>

>> >> Regards,

>> >> Helin

>> >

>
Helin Zhang Feb. 4, 2015, 12:37 a.m. UTC | #6
Hi Jia

> -----Original Message-----

> From: Jia Yu [mailto:jyu@vmware.com]

> Sent: Wednesday, February 4, 2015 2:53 AM

> To: Zhang, Helin

> Cc: dev@dpdk.org; Thomas Monjalon

> Subject: Re: [dpdk-dev] [PATCH 1/2] rte_ethdev: update link status (speed,

> duplex, link_up) after rte_eth_dev_start

> 

> Helin,

> 

> Thanks for comment. Any device that enabled LSC needs this fix, otherwise

> they all need to call link_update separately. We cannot assume that only Bond

> enables LSC interrupt.

> 

> The fix will not be used for all PMDs, as it explicitly checks if

> (dev->data->dev_conf.intr_conf.lsc != 0). Therefore, for hardware NIC (e.g.

> 82599) that disabled lsc by default, the link_update callback will not be

> executed. Please let me know if you have other concerns.

I understood. I mean if (lsc == 1), for 82599, it can report lsc event correctly after startup.
But if a manually link update is executed, no link change can be found when a lsc
event is received in application.
Your fix is to solve the problem in bond only, am I correct? My comment is that can we
avoid doing that for 82599, Fortville, etc? As some of DPDK customers may not want it if
lsc is enabled.

Regards,
Helin

> 

> Thanks,

> Jia

> 

> On 2/3/15, 12:35 AM, "Zhang, Helin" <helin.zhang@intel.com> wrote:

> 

> >

> >

> >> -----Original Message-----

> >> From: Jia Yu [mailto:jyu@vmware.com]

> >> Sent: Tuesday, February 3, 2015 4:00 PM

> >> To: Zhang, Helin

> >> Cc: dev@dpdk.org; Thomas Monjalon

> >> Subject: Re: [dpdk-dev] [PATCH 1/2] rte_ethdev: update link status

> >>(speed,  duplex, link_up) after rte_eth_dev_start

> >>

> >> My answer to Helin¹s comments:

> >>

> >> This patch is needed for bond slave devices or other devices, when

> >> LSC interrupt is enabled.

> >>  1. slave_configure()  -> slave_eth_dev->Š.lsc = 1

> >>

> >>  2. rte_eth_link_get() reads dev_link from eth_dev, when lsc

> >>interrupt is  enabled. However, the dev_link on eth_dev has not be

> >>initialized and showed  link down state. This patch initializes the

> >>device¹s dev_link at  rte_eth_dev_start time.

> >So the link update is for bond only? But your code changes is in

> >rte_ethdev, it will be used for all PMDs. For hardware NIC (e.g.

> >82599), this link update is not needed at all. It just need to wait the

> >link status change event. So can those code changes be put in

> >librte_pmd_bond, but not in librte_ether?

> >

> >Regards,

> >Helin

> >

> >>

> >> Please let me know if you have further questions/comments.

> >>

> >> Thanks,

> >> Jia

> >>

> >> On 1/30/15, 2:28 AM, "Thomas Monjalon" <thomas.monjalon@6wind.com>

> >> wrote:

> >>

> >> >Jia, any news on this patchset?

> >> >

> >> >2014-11-12 03:57, Zhang, Helin:

> >> >> Hi Jia

> >> >>

> >> >> > -----Original Message-----

> >> >> > From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Jia Yu

> >> >> > Sent: Saturday, November 8, 2014 1:32 AM

> >> >> > To: dev@dpdk.org

> >> >> > Subject: [dpdk-dev] [PATCH 1/2] rte_ethdev: update link status

> >> >>(speed, duplex,

> >> >> > link_up) after rte_eth_dev_start

> >> >> >

> >> >> > Since LSR interrupt is disabled by pmd drivers, link status in

> >> >>rte_eth_device is

> >> >> > always down.

> >> >> If LSC interrupt is disabled by default, it will poll the link

> >> >>status during the initialization  or in dev_start, and then the

> >> >>link status should he correct. If I am not wrong.

> >> >>

> >> >> > Bond slave_configure() enables LSR interrupt on devices to get

> >> >>notification if link

> >> >> > status changes. However, the LSC interrupt at device start time

> >> >> > is

> >> >>still lost.

> >> >> Before enabling interrupt for LSC, the link status should be polled.

> >> >>So after the port  startup, the link status should be there.

> >> >>

> >> >> >

> >> >> > In this fix, call link_update to read link status from hardware

> >> >>register at device

> >> >> > start time.

> >> >> Could you help to explain this code changes a bit more? Why we

> >> >> need

> >>it?

> >> >>

> >> >> >

> >> >> > Issue:

> >> >> > Change-Id: Ib57a1c9114f922485c7b0f4338bfe7b3d3f87d65

> >> >> > Signed-off-by: Jia Yu <jyu@vmware.com>

> >> >> > ---

> >> >> >  lib/librte_ether/rte_ethdev.c | 4 ++++

> >> >> >  1 file changed, 4 insertions(+)

> >> >> >

> >> >> > diff --git a/lib/librte_ether/rte_ethdev.c

> >> >>b/lib/librte_ether/rte_ethdev.c index

> >> >> > ff1c769..6c01b02 100644

> >> >> > --- a/lib/librte_ether/rte_ethdev.c

> >> >> > +++ b/lib/librte_ether/rte_ethdev.c

> >> >> > @@ -869,6 +869,10 @@ rte_eth_dev_start(uint8_t port_id)

> >> >> >

> >> >> >  	rte_eth_dev_config_restore(port_id);

> >> >> >

> >> >> > +	if (dev->data->dev_conf.intr_conf.lsc != 0) {

> >> >> > +		FUNC_PTR_OR_ERR_RET(*dev->dev_ops->link_update,

> >> -ENOTSUP);

> >> >> > +		(*dev->dev_ops->link_update)(dev, 0);

> >> >> > +	}

> >> >> >  	return 0;

> >> >> >  }

> >> >> >

> >> >> > --

> >> >> > 1.9.1

> >> >>

> >> >> Regards,

> >> >> Helin

> >> >

> >
diff mbox

Patch

diff --git a/lib/librte_ether/rte_ethdev.c b/lib/librte_ether/rte_ethdev.c
index ff1c769..6c01b02 100644
--- a/lib/librte_ether/rte_ethdev.c
+++ b/lib/librte_ether/rte_ethdev.c
@@ -869,6 +869,10 @@  rte_eth_dev_start(uint8_t port_id)
 
 	rte_eth_dev_config_restore(port_id);
 
+	if (dev->data->dev_conf.intr_conf.lsc != 0) {
+		FUNC_PTR_OR_ERR_RET(*dev->dev_ops->link_update, -ENOTSUP);
+		(*dev->dev_ops->link_update)(dev, 0);
+	}
 	return 0;
 }