[1/2] app/testpmd: fix scatter offload configuration

Message ID 1564403817-13438-1-git-send-email-matan@mellanox.com
State New
Delegated to: Ferruh Yigit
Headers show
Series
  • [1/2] app/testpmd: fix scatter offload configuration
Related show

Checks

Context Check Description
ci/Intel-compilation success Compilation OK
ci/iol-Compile-Testing success Compile Testing PASS
ci/checkpatch success coding style OK

Commit Message

Matan Azrad July 29, 2019, 12:36 p.m.
When the mbuf data size cannot contain the maximum Rx packet length with
the mbuf headroom, a packet should be scattered in more than one mbuf.

The application did not configure scatter offload in the above case.

Enable the Rx scatter offload in the above case.

Fixes: 33f9630fc23d ("app/testpmd: create mbuf based on max supported segments")
Cc: stable@dpdk.org

Signed-off-by: Matan Azrad <matan@mellanox.com>
---
 app/test-pmd/testpmd.c | 11 +++++++++++
 1 file changed, 11 insertions(+)

Comments

Matan Azrad July 30, 2019, 9 a.m. | #1
Hi all

Any review?

> -----Original Message-----
> From: dev <dev-bounces@dpdk.org> On Behalf Of Matan Azrad
> Sent: Monday, July 29, 2019 3:37 PM
> To: Wenzhuo Lu <wenzhuo.lu@intel.com>; Jingjing Wu
> <jingjing.wu@intel.com>
> Cc: dev@dpdk.org; stable@dpdk.org
> Subject: [dpdk-dev] [PATCH 1/2] app/testpmd: fix scatter offload
> configuration
> 
> When the mbuf data size cannot contain the maximum Rx packet length with
> the mbuf headroom, a packet should be scattered in more than one mbuf.
> 
> The application did not configure scatter offload in the above case.
> 
> Enable the Rx scatter offload in the above case.
> 
> Fixes: 33f9630fc23d ("app/testpmd: create mbuf based on max supported
> segments")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Matan Azrad <matan@mellanox.com>
> ---
>  app/test-pmd/testpmd.c | 11 +++++++++++
>  1 file changed, 11 insertions(+)
> 
> diff --git a/app/test-pmd/testpmd.c b/app/test-pmd/testpmd.c index
> 518865a..4ae70ef 100644
> --- a/app/test-pmd/testpmd.c
> +++ b/app/test-pmd/testpmd.c
> @@ -1191,6 +1191,17 @@ struct extmem_param {
>  				warning = 1;
>  			}
>  		}
> +		if (rx_mode.max_rx_pkt_len + RTE_PKTMBUF_HEADROOM
> >
> +		    mbuf_data_size) {
> +			if (port->dev_info.rx_queue_offload_capa &
> +			    DEV_RX_OFFLOAD_SCATTER)
> +				port->dev_conf.rxmode.offloads |=
> +						DEV_RX_OFFLOAD_SCATTER;
> +			else
> +				TESTPMD_LOG(WARNING, "Configure
> scatter is"
> +					    " needed and cannot be
> configured"
> +					    " in the port %u\n", pid);
> +		}
>  	}
> 
>  	if (warning)
> --
> 1.8.3.1
Moti Haimovsky July 30, 2019, 11:36 a.m. | #2
> > -----Original Message-----
> > From: dev <dev-bounces@dpdk.org> On Behalf Of Matan Azrad
> > Sent: Monday, July 29, 2019 3:37 PM
> > To: Wenzhuo Lu <wenzhuo.lu@intel.com>; Jingjing Wu
> > <jingjing.wu@intel.com>
> > Cc: dev@dpdk.org; stable@dpdk.org
> > Subject: [dpdk-dev] [PATCH 1/2] app/testpmd: fix scatter offload
> > configuration
> >
> > When the mbuf data size cannot contain the maximum Rx packet length
> > with the mbuf headroom, a packet should be scattered in more than one
> mbuf.
> >
> > The application did not configure scatter offload in the above case.
> >
> > Enable the Rx scatter offload in the above case.
> >
> > Fixes: 33f9630fc23d ("app/testpmd: create mbuf based on max supported
> > segments")
> > Cc: stable@dpdk.org
> >
> > Signed-off-by: Matan Azrad <matan@mellanox.com>
Acked-by: Moti Haimovsky <motih@mellanox.com>
> > ---
> >  app/test-pmd/testpmd.c | 11 +++++++++++
> >  1 file changed, 11 insertions(+)
> >
> > diff --git a/app/test-pmd/testpmd.c b/app/test-pmd/testpmd.c index
> > 518865a..4ae70ef 100644
> > --- a/app/test-pmd/testpmd.c
> > +++ b/app/test-pmd/testpmd.c
> > @@ -1191,6 +1191,17 @@ struct extmem_param {
> >  				warning = 1;
> >  			}
> >  		}
> > +		if (rx_mode.max_rx_pkt_len + RTE_PKTMBUF_HEADROOM
> > >
> > +		    mbuf_data_size) {
> > +			if (port->dev_info.rx_queue_offload_capa &
> > +			    DEV_RX_OFFLOAD_SCATTER)
> > +				port->dev_conf.rxmode.offloads |=
> > +						DEV_RX_OFFLOAD_SCATTER;
> > +			else
> > +				TESTPMD_LOG(WARNING, "Configure
> > scatter is"
> > +					    " needed and cannot be
> > configured"
> > +					    " in the port %u\n", pid);
> > +		}
> >  	}
> >
> >  	if (warning)
> > --
> > 1.8.3.1
Ferruh Yigit July 30, 2019, 1:09 p.m. | #3
On 7/29/2019 1:36 PM, Matan Azrad wrote:
> When the mbuf data size cannot contain the maximum Rx packet length with
> the mbuf headroom, a packet should be scattered in more than one mbuf.
> 
> The application did not configure scatter offload in the above case.
> 
> Enable the Rx scatter offload in the above case.
> 
> Fixes: 33f9630fc23d ("app/testpmd: create mbuf based on max supported segments")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Matan Azrad <matan@mellanox.com>

Deferring the patchset to next release, they were late anyway and not actually
fixing a defect, safer to defer than getting them in rc3.
Matan Azrad July 30, 2019, 1:17 p.m. | #4
Hi Ferruh

From: Ferruh Yigit
> Sent: Tuesday, July 30, 2019 4:09 PM
> To: Matan Azrad <matan@mellanox.com>; Wenzhuo Lu
> <wenzhuo.lu@intel.com>; Jingjing Wu <jingjing.wu@intel.com>
> Cc: dev@dpdk.org; stable@dpdk.org
> Subject: Re: [dpdk-dev] [PATCH 1/2] app/testpmd: fix scatter offload
> configuration
> 
> On 7/29/2019 1:36 PM, Matan Azrad wrote:
> > When the mbuf data size cannot contain the maximum Rx packet length
> > with the mbuf headroom, a packet should be scattered in more than one
> mbuf.
> >
> > The application did not configure scatter offload in the above case.
> >
> > Enable the Rx scatter offload in the above case.
> >
> > Fixes: 33f9630fc23d ("app/testpmd: create mbuf based on max supported
> > segments")
> > Cc: stable@dpdk.org
> >
> > Signed-off-by: Matan Azrad <matan@mellanox.com>
> 
> Deferring the patchset to next release, they were late anyway and not
> actually fixing a defect, safer to defer than getting them in rc3.

Yes this patch came late for RC3 but it is a fix.

What are you concerns here?
Why don't you think defect found?

What's about RC4?

Matan
Ferruh Yigit July 30, 2019, 3:21 p.m. | #5
On 7/30/2019 2:17 PM, Matan Azrad wrote:
> Hi Ferruh
> 
> From: Ferruh Yigit
>> Sent: Tuesday, July 30, 2019 4:09 PM
>> To: Matan Azrad <matan@mellanox.com>; Wenzhuo Lu
>> <wenzhuo.lu@intel.com>; Jingjing Wu <jingjing.wu@intel.com>
>> Cc: dev@dpdk.org; stable@dpdk.org
>> Subject: Re: [dpdk-dev] [PATCH 1/2] app/testpmd: fix scatter offload
>> configuration
>>
>> On 7/29/2019 1:36 PM, Matan Azrad wrote:
>>> When the mbuf data size cannot contain the maximum Rx packet length
>>> with the mbuf headroom, a packet should be scattered in more than one
>> mbuf.
>>>
>>> The application did not configure scatter offload in the above case.
>>>
>>> Enable the Rx scatter offload in the above case.
>>>
>>> Fixes: 33f9630fc23d ("app/testpmd: create mbuf based on max supported
>>> segments")
>>> Cc: stable@dpdk.org
>>>
>>> Signed-off-by: Matan Azrad <matan@mellanox.com>
>>
>> Deferring the patchset to next release, they were late anyway and not
>> actually fixing a defect, safer to defer than getting them in rc3.
> 
> Yes this patch came late for RC3 but it is a fix.
> 
> What are you concerns here?
> Why don't you think defect found?

First patch changes the behavior, when mbuf size is larger than configured size
and user didn't provided the scatter offload, should test application
automatically enable it? It may or not, but this is the change of the behavior,
I think not a fix.

And second patch adds more detail into the statistics, so I believe it is clear
that it is not a fix.

The concern is getting changes very close to release, to balance between risk
and benefit of the feature. I don't see any reason why these changes can't wait
next release, so I don't see any reason to get the risk.

> 
> What's about RC4?

No, it is even worse, there will be only a little testing after rc4 and a little
time before release.
Matan Azrad July 30, 2019, 3:56 p.m. | #6
Hi Ferruh

 From: Ferruh Yigit
> Sent: Tuesday, July 30, 2019 6:22 PM
> To: Matan Azrad <matan@mellanox.com>; Wenzhuo Lu
> <wenzhuo.lu@intel.com>; Jingjing Wu <jingjing.wu@intel.com>
> Cc: dev@dpdk.org; stable@dpdk.org
> Subject: Re: [dpdk-dev] [PATCH 1/2] app/testpmd: fix scatter offload
> configuration
> 
> On 7/30/2019 2:17 PM, Matan Azrad wrote:
> > Hi Ferruh
> >
> > From: Ferruh Yigit
> >> Sent: Tuesday, July 30, 2019 4:09 PM
> >> To: Matan Azrad <matan@mellanox.com>; Wenzhuo Lu
> >> <wenzhuo.lu@intel.com>; Jingjing Wu <jingjing.wu@intel.com>
> >> Cc: dev@dpdk.org; stable@dpdk.org
> >> Subject: Re: [dpdk-dev] [PATCH 1/2] app/testpmd: fix scatter offload
> >> configuration
> >>
> >> On 7/29/2019 1:36 PM, Matan Azrad wrote:
> >>> When the mbuf data size cannot contain the maximum Rx packet length
> >>> with the mbuf headroom, a packet should be scattered in more than
> >>> one
> >> mbuf.
> >>>
> >>> The application did not configure scatter offload in the above case.
> >>>
> >>> Enable the Rx scatter offload in the above case.
> >>>
> >>> Fixes: 33f9630fc23d ("app/testpmd: create mbuf based on max
> >>> supported
> >>> segments")
> >>> Cc: stable@dpdk.org
> >>>
> >>> Signed-off-by: Matan Azrad <matan@mellanox.com>
> >>
> >> Deferring the patchset to next release, they were late anyway and not
> >> actually fixing a defect, safer to defer than getting them in rc3.
> >
> > Yes this patch came late for RC3 but it is a fix.
> >
> > What are you concerns here?
> > Why don't you think defect found?
> 
> First patch changes the behavior, when mbuf size is larger than configured
> size and user didn't provided the scatter offload, should test application
> automatically enable it?

No, only when the mbuf size is smaller than the max_rx_pkt_len with headroom.
If scatter is not enabled in the above case, how can the PMD provide a packet with max_rx_pkt_len size? 

I think not enabling scatter in this case it is a user conflict in configuration and should raise an error in the PMD.  Maybe even in ethdev layer.

> It may or not, but this is the change of the behavior, I
> think not a fix.
> 
> And second patch adds more detail into the statistics, so I believe it is clear
> that it is not a fix.
 
 Agree, this can wait.

> The concern is getting changes very close to release, to balance between risk
> and benefit of the feature. I don't see any reason why these changes can't
> wait next release, so I don't see any reason to get the risk.

When  I changed the default max_rx_pkt_len and mbuf size in LRO testing I met this issue.

By default scatter will not be enabled.


> > What's about RC4?
> 
> No, it is even worse, there will be only a little testing after rc4 and a little time
> before release.

So, I hope it will be integrated in RC3.
Ferruh Yigit July 30, 2019, 5:28 p.m. | #7
On 7/30/2019 4:56 PM, Matan Azrad wrote:
> Hi Ferruh
> 
>  From: Ferruh Yigit
>> Sent: Tuesday, July 30, 2019 6:22 PM
>> To: Matan Azrad <matan@mellanox.com>; Wenzhuo Lu
>> <wenzhuo.lu@intel.com>; Jingjing Wu <jingjing.wu@intel.com>
>> Cc: dev@dpdk.org; stable@dpdk.org
>> Subject: Re: [dpdk-dev] [PATCH 1/2] app/testpmd: fix scatter offload
>> configuration
>>
>> On 7/30/2019 2:17 PM, Matan Azrad wrote:
>>> Hi Ferruh
>>>
>>> From: Ferruh Yigit
>>>> Sent: Tuesday, July 30, 2019 4:09 PM
>>>> To: Matan Azrad <matan@mellanox.com>; Wenzhuo Lu
>>>> <wenzhuo.lu@intel.com>; Jingjing Wu <jingjing.wu@intel.com>
>>>> Cc: dev@dpdk.org; stable@dpdk.org
>>>> Subject: Re: [dpdk-dev] [PATCH 1/2] app/testpmd: fix scatter offload
>>>> configuration
>>>>
>>>> On 7/29/2019 1:36 PM, Matan Azrad wrote:
>>>>> When the mbuf data size cannot contain the maximum Rx packet length
>>>>> with the mbuf headroom, a packet should be scattered in more than
>>>>> one
>>>> mbuf.
>>>>>
>>>>> The application did not configure scatter offload in the above case.
>>>>>
>>>>> Enable the Rx scatter offload in the above case.
>>>>>
>>>>> Fixes: 33f9630fc23d ("app/testpmd: create mbuf based on max
>>>>> supported
>>>>> segments")
>>>>> Cc: stable@dpdk.org
>>>>>
>>>>> Signed-off-by: Matan Azrad <matan@mellanox.com>
>>>>
>>>> Deferring the patchset to next release, they were late anyway and not
>>>> actually fixing a defect, safer to defer than getting them in rc3.
>>>
>>> Yes this patch came late for RC3 but it is a fix.
>>>
>>> What are you concerns here?
>>> Why don't you think defect found?
>>
>> First patch changes the behavior, when mbuf size is larger than configured
>> size and user didn't provided the scatter offload, should test application
>> automatically enable it?
> 
> No, only when the mbuf size is smaller than the max_rx_pkt_len with headroom.
> If scatter is not enabled in the above case, how can the PMD provide a packet with max_rx_pkt_len size? 
> 
> I think not enabling scatter in this case it is a user conflict in configuration and should raise an error in the PMD.  Maybe even in ethdev layer.
> 
>> It may or not, but this is the change of the behavior, I
>> think not a fix.
>>
>> And second patch adds more detail into the statistics, so I believe it is clear
>> that it is not a fix.
>  
>  Agree, this can wait.
> 
>> The concern is getting changes very close to release, to balance between risk
>> and benefit of the feature. I don't see any reason why these changes can't
>> wait next release, so I don't see any reason to get the risk.
> 
> When  I changed the default max_rx_pkt_len and mbuf size in LRO testing I met this issue.
> 
> By default scatter will not be enabled.

I think it is still arguable if scatter should be enabled by default, but isn't
there a way in testpmd to enable scatter explicitly? If so you have a way to
test LRO.

> 
> 
>>> What's about RC4?
>>
>> No, it is even worse, there will be only a little testing after rc4 and a little time
>> before release.
> 
> So, I hope it will be integrated in RC3.
>
Matan Azrad July 30, 2019, 6:34 p.m. | #8
From: Ferruh Yigit 
> On 7/30/2019 4:56 PM, Matan Azrad wrote:
> > Hi Ferruh
> >
> >  From: Ferruh Yigit
> >> Sent: Tuesday, July 30, 2019 6:22 PM
> >> To: Matan Azrad <matan@mellanox.com>; Wenzhuo Lu
> >> <wenzhuo.lu@intel.com>; Jingjing Wu <jingjing.wu@intel.com>
> >> Cc: dev@dpdk.org; stable@dpdk.org
> >> Subject: Re: [dpdk-dev] [PATCH 1/2] app/testpmd: fix scatter offload
> >> configuration
> >>
> >> On 7/30/2019 2:17 PM, Matan Azrad wrote:
> >>> Hi Ferruh
> >>>
> >>> From: Ferruh Yigit
> >>>> Sent: Tuesday, July 30, 2019 4:09 PM
> >>>> To: Matan Azrad <matan@mellanox.com>; Wenzhuo Lu
> >>>> <wenzhuo.lu@intel.com>; Jingjing Wu <jingjing.wu@intel.com>
> >>>> Cc: dev@dpdk.org; stable@dpdk.org
> >>>> Subject: Re: [dpdk-dev] [PATCH 1/2] app/testpmd: fix scatter
> >>>> offload configuration
> >>>>
> >>>> On 7/29/2019 1:36 PM, Matan Azrad wrote:
> >>>>> When the mbuf data size cannot contain the maximum Rx packet
> >>>>> length with the mbuf headroom, a packet should be scattered in
> >>>>> more than one
> >>>> mbuf.
> >>>>>
> >>>>> The application did not configure scatter offload in the above case.
> >>>>>
> >>>>> Enable the Rx scatter offload in the above case.
> >>>>>
> >>>>> Fixes: 33f9630fc23d ("app/testpmd: create mbuf based on max
> >>>>> supported
> >>>>> segments")
> >>>>> Cc: stable@dpdk.org
> >>>>>
> >>>>> Signed-off-by: Matan Azrad <matan@mellanox.com>
> >>>>
> >>>> Deferring the patchset to next release, they were late anyway and
> >>>> not actually fixing a defect, safer to defer than getting them in rc3.
> >>>
> >>> Yes this patch came late for RC3 but it is a fix.
> >>>
> >>> What are you concerns here?
> >>> Why don't you think defect found?
> >>
> >> First patch changes the behavior, when mbuf size is larger than
> >> configured size and user didn't provided the scatter offload, should
> >> test application automatically enable it?
> >
> > No, only when the mbuf size is smaller than the max_rx_pkt_len with
> headroom.
> > If scatter is not enabled in the above case, how can the PMD provide a
> packet with max_rx_pkt_len size?
> >

Answer here?

> > I think not enabling scatter in this case it is a user conflict in configuration
> and should raise an error in the PMD.  Maybe even in ethdev layer.
> >
> >> It may or not, but this is the change of the behavior, I think not a
> >> fix.
> >>
> >> And second patch adds more detail into the statistics, so I believe
> >> it is clear that it is not a fix.
> >
> >  Agree, this can wait.
> >
> >> The concern is getting changes very close to release, to balance
> >> between risk and benefit of the feature. I don't see any reason why
> >> these changes can't wait next release, so I don't see any reason to get the
> risk.
> >
> > When  I changed the default max_rx_pkt_len and mbuf size in LRO testing I
> met this issue.
> >
> > By default scatter will not be enabled.
> 
> I think it is still arguable if scatter should be enabled by default,

I meant that with this patch it will not be enabled by default due to the default values of mbuf size and max_rx_pkt_len.

> but isn't there a way in testpmd to enable scatter explicitly? If so you have a way to test LRO.

Yes there is a way.

This patch is just the right way to do it.
Ferruh Yigit July 30, 2019, 6:55 p.m. | #9
On 7/30/2019 7:34 PM, Matan Azrad wrote:
> 
> 
> From: Ferruh Yigit 
>> On 7/30/2019 4:56 PM, Matan Azrad wrote:
>>> Hi Ferruh
>>>
>>>  From: Ferruh Yigit
>>>> Sent: Tuesday, July 30, 2019 6:22 PM
>>>> To: Matan Azrad <matan@mellanox.com>; Wenzhuo Lu
>>>> <wenzhuo.lu@intel.com>; Jingjing Wu <jingjing.wu@intel.com>
>>>> Cc: dev@dpdk.org; stable@dpdk.org
>>>> Subject: Re: [dpdk-dev] [PATCH 1/2] app/testpmd: fix scatter offload
>>>> configuration
>>>>
>>>> On 7/30/2019 2:17 PM, Matan Azrad wrote:
>>>>> Hi Ferruh
>>>>>
>>>>> From: Ferruh Yigit
>>>>>> Sent: Tuesday, July 30, 2019 4:09 PM
>>>>>> To: Matan Azrad <matan@mellanox.com>; Wenzhuo Lu
>>>>>> <wenzhuo.lu@intel.com>; Jingjing Wu <jingjing.wu@intel.com>
>>>>>> Cc: dev@dpdk.org; stable@dpdk.org
>>>>>> Subject: Re: [dpdk-dev] [PATCH 1/2] app/testpmd: fix scatter
>>>>>> offload configuration
>>>>>>
>>>>>> On 7/29/2019 1:36 PM, Matan Azrad wrote:
>>>>>>> When the mbuf data size cannot contain the maximum Rx packet
>>>>>>> length with the mbuf headroom, a packet should be scattered in
>>>>>>> more than one
>>>>>> mbuf.
>>>>>>>
>>>>>>> The application did not configure scatter offload in the above case.
>>>>>>>
>>>>>>> Enable the Rx scatter offload in the above case.
>>>>>>>
>>>>>>> Fixes: 33f9630fc23d ("app/testpmd: create mbuf based on max
>>>>>>> supported
>>>>>>> segments")
>>>>>>> Cc: stable@dpdk.org
>>>>>>>
>>>>>>> Signed-off-by: Matan Azrad <matan@mellanox.com>
>>>>>>
>>>>>> Deferring the patchset to next release, they were late anyway and
>>>>>> not actually fixing a defect, safer to defer than getting them in rc3.
>>>>>
>>>>> Yes this patch came late for RC3 but it is a fix.
>>>>>
>>>>> What are you concerns here?
>>>>> Why don't you think defect found?
>>>>
>>>> First patch changes the behavior, when mbuf size is larger than
>>>> configured size and user didn't provided the scatter offload, should
>>>> test application automatically enable it?
>>>
>>> No, only when the mbuf size is smaller than the max_rx_pkt_len with
>> headroom.
>>> If scatter is not enabled in the above case, how can the PMD provide a
>> packet with max_rx_pkt_len size?
>>>
> 
> Answer here?

Is it because drivers also "automatically" enable scattered Rx based on other
values? - which is also open to discussion I think.

> 
>>> I think not enabling scatter in this case it is a user conflict in configuration
>> and should raise an error in the PMD.  Maybe even in ethdev layer.
>>>
>>>> It may or not, but this is the change of the behavior, I think not a
>>>> fix.
>>>>
>>>> And second patch adds more detail into the statistics, so I believe
>>>> it is clear that it is not a fix.
>>>
>>>  Agree, this can wait.
>>>
>>>> The concern is getting changes very close to release, to balance
>>>> between risk and benefit of the feature. I don't see any reason why
>>>> these changes can't wait next release, so I don't see any reason to get the
>> risk.
>>>
>>> When  I changed the default max_rx_pkt_len and mbuf size in LRO testing I
>> met this issue.
>>>
>>> By default scatter will not be enabled.
>>
>> I think it is still arguable if scatter should be enabled by default,
> 
> I meant that with this patch it will not be enabled by default due to the default values of mbuf size and max_rx_pkt_len.

I mean the same thing indeed, still I believe arguable.

> 
>> but isn't there a way in testpmd to enable scatter explicitly? If so you have a way to test LRO.
> 
> Yes there is a way.
> 
> This patch is just the right way to do it.
> 

Good to know it is not blocking anyone, patch can be reviewed by its maintainers
and discussed more.
Matan Azrad July 31, 2019, 6:11 a.m. | #10
Hi Ferruh

From: Ferruh Yigit
> On 7/30/2019 7:34 PM, Matan Azrad wrote:
> >
> >
> > From: Ferruh Yigit
> >> On 7/30/2019 4:56 PM, Matan Azrad wrote:
> >>> Hi Ferruh
> >>>
> >>>  From: Ferruh Yigit
> >>>> Sent: Tuesday, July 30, 2019 6:22 PM
> >>>> To: Matan Azrad <matan@mellanox.com>; Wenzhuo Lu
> >>>> <wenzhuo.lu@intel.com>; Jingjing Wu <jingjing.wu@intel.com>
> >>>> Cc: dev@dpdk.org; stable@dpdk.org
> >>>> Subject: Re: [dpdk-dev] [PATCH 1/2] app/testpmd: fix scatter
> >>>> offload configuration
> >>>>
> >>>> On 7/30/2019 2:17 PM, Matan Azrad wrote:
> >>>>> Hi Ferruh
> >>>>>
> >>>>> From: Ferruh Yigit
> >>>>>> Sent: Tuesday, July 30, 2019 4:09 PM
> >>>>>> To: Matan Azrad <matan@mellanox.com>; Wenzhuo Lu
> >>>>>> <wenzhuo.lu@intel.com>; Jingjing Wu <jingjing.wu@intel.com>
> >>>>>> Cc: dev@dpdk.org; stable@dpdk.org
> >>>>>> Subject: Re: [dpdk-dev] [PATCH 1/2] app/testpmd: fix scatter
> >>>>>> offload configuration
> >>>>>>
> >>>>>> On 7/29/2019 1:36 PM, Matan Azrad wrote:
> >>>>>>> When the mbuf data size cannot contain the maximum Rx packet
> >>>>>>> length with the mbuf headroom, a packet should be scattered in
> >>>>>>> more than one
> >>>>>> mbuf.
> >>>>>>>
> >>>>>>> The application did not configure scatter offload in the above case.
> >>>>>>>
> >>>>>>> Enable the Rx scatter offload in the above case.
> >>>>>>>
> >>>>>>> Fixes: 33f9630fc23d ("app/testpmd: create mbuf based on max
> >>>>>>> supported
> >>>>>>> segments")
> >>>>>>> Cc: stable@dpdk.org
> >>>>>>>
> >>>>>>> Signed-off-by: Matan Azrad <matan@mellanox.com>
> >>>>>>
> >>>>>> Deferring the patchset to next release, they were late anyway and
> >>>>>> not actually fixing a defect, safer to defer than getting them in rc3.
> >>>>>
> >>>>> Yes this patch came late for RC3 but it is a fix.
> >>>>>
> >>>>> What are you concerns here?
> >>>>> Why don't you think defect found?
> >>>>
> >>>> First patch changes the behavior, when mbuf size is larger than
> >>>> configured size and user didn't provided the scatter offload,
> >>>> should test application automatically enable it?
> >>>
> >>> No, only when the mbuf size is smaller than the max_rx_pkt_len with
> >> headroom.
> >>> If scatter is not enabled in the above case, how can the PMD provide
> >>> a
> >> packet with max_rx_pkt_len size?
> >>>
> >
> > Answer here?
> 
> Is it because drivers also "automatically" enable scattered Rx based on other
> values?

Scatter is a defined RX offload.
Like other offloads I think it always should be explicitly set by the user if he wants it, and vice versa.
If the user doesn't configure it, the PMD should not scatter packets because the user doesn't expect multi-mbuf packets in datapath and maybe even doesn't handle it.

So, I think the above case is a user conflict in Rx configuration and like other conflicts it should cause an error.
In MLX5 PMD, an error will be returned from Rx setup.

Patch

diff --git a/app/test-pmd/testpmd.c b/app/test-pmd/testpmd.c
index 518865a..4ae70ef 100644
--- a/app/test-pmd/testpmd.c
+++ b/app/test-pmd/testpmd.c
@@ -1191,6 +1191,17 @@  struct extmem_param {
 				warning = 1;
 			}
 		}
+		if (rx_mode.max_rx_pkt_len + RTE_PKTMBUF_HEADROOM >
+		    mbuf_data_size) {
+			if (port->dev_info.rx_queue_offload_capa &
+			    DEV_RX_OFFLOAD_SCATTER)
+				port->dev_conf.rxmode.offloads |=
+						DEV_RX_OFFLOAD_SCATTER;
+			else
+				TESTPMD_LOG(WARNING, "Configure scatter is"
+					    " needed and cannot be configured"
+					    " in the port %u\n", pid);
+		}
 	}
 
 	if (warning)