[v5] app/testpmd: fix the default RSS key configuration
diff mbox series

Message ID 1602765662-43299-1-git-send-email-oulijun@huawei.com
State Changes Requested
Delegated to: Ferruh Yigit
Headers show
Series
  • [v5] app/testpmd: fix the default RSS key configuration
Related show

Checks

Context Check Description
ci/iol-mellanox-Performance success Performance Testing PASS
ci/iol-intel-Performance success Performance Testing PASS
ci/iol-testing fail Testing issues
ci/iol-intel-Functional success Functional Testing PASS
ci/iol-broadcom-Functional success Functional Testing PASS
ci/travis-robot warning Travis build: failed
ci/iol-broadcom-Performance success Performance Testing PASS
ci/Intel-compilation success Compilation OK
ci/checkpatch success coding style OK

Commit Message

Lijun Ou Oct. 15, 2020, 12:41 p.m. UTC
When start the testpmd, the pmd driver initializes the RSS
configuration. Generally, the recommended RSS hash key is
used as the default key in the driver. In addition, the
default key is different from the default RSS flow in testpmd
without specifying RSS hash key. So. if you do not specify
the RSS key when creating an RSS rule, the testpmd uses the
default key as the default RSS key of the RSS rule. As a result,
you may mistakenly consider that the RSS key in use is the valid
default key of the NIC, actually, the key and the valid default
key of the NIC are two values.

Consider the follow usage with testpmd:
1. first, startup testpmd:
testpmd> show port 0 rss-hash key
RSS functions:
  all ipv4-frag ipv4-other ipv6-frag ipv6-other ip
RSS key:
-6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA38030F
-20C6A42B73BBEAC01FA
2. create a rss rule
testpmd> flow create 0 ingress pattern eth / ipv4 / udp / end actions rss \
types ipv4-udp end queues end / end

3. show rss-hash key
testpmd> show port 0 rss-hash key
RSS functions:
  all ipv4-udp udp
RSS key:
-74657374706D6427732064656661756C74205253532068617368206B65792C206F
-76657272696465

In order to solve the above problems, it use the NIC valid default
RSS key instead of the testpmd dummy RSS key in the flow configuration
when the RSS key is not specified in the flow rule. If the NIC RSS key
is invalid, it will use testpmd dummy RSS key as the default key.

Fixes: ac8d22de2394 ("ethdev: flatten RSS configuration in flow API")
Cc: stable@dpdk.org

Signed-off-by: Lijun Ou <oulijun@huawei.com>
Reviewed-by: Phil Yang <phil.yang@arm.com>
---
V4->V5:
-rewrite the commit log
-add reviewed-by

V3->V4:
-fix checkpatch warning and shorter commit content.

V2->V3:
-fix checkpatch warning.

V1->V2:
-fix the commit.
---
 app/test-pmd/cmdline_flow.c | 8 ++++++++
 1 file changed, 8 insertions(+)

Comments

Ferruh Yigit Oct. 15, 2020, 1:52 p.m. UTC | #1
On 10/15/2020 1:41 PM, Lijun Ou wrote:
> When start the testpmd, the pmd driver initializes the RSS
> configuration. Generally, the recommended RSS hash key is
> used as the default key in the driver. In addition, the
> default key is different from the default RSS flow in testpmd
> without specifying RSS hash key. So. if you do not specify
> the RSS key when creating an RSS rule, the testpmd uses the
> default key as the default RSS key of the RSS rule. As a result,
> you may mistakenly consider that the RSS key in use is the valid
> default key of the NIC, actually, the key and the valid default
> key of the NIC are two values.
> 
> Consider the follow usage with testpmd:
> 1. first, startup testpmd:
> testpmd> show port 0 rss-hash key
> RSS functions:
>    all ipv4-frag ipv4-other ipv6-frag ipv6-other ip
> RSS key:
> -6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA38030F
> -20C6A42B73BBEAC01FA
> 2. create a rss rule
> testpmd> flow create 0 ingress pattern eth / ipv4 / udp / end actions rss \
> types ipv4-udp end queues end / end
> 
> 3. show rss-hash key
> testpmd> show port 0 rss-hash key
> RSS functions:
>    all ipv4-udp udp
> RSS key:
> -74657374706D6427732064656661756C74205253532068617368206B65792C206F
> -76657272696465
> 
> In order to solve the above problems, it use the NIC valid default
> RSS key instead of the testpmd dummy RSS key in the flow configuration
> when the RSS key is not specified in the flow rule. If the NIC RSS key
> is invalid, it will use testpmd dummy RSS key as the default key.
> 
> Fixes: ac8d22de2394 ("ethdev: flatten RSS configuration in flow API")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Lijun Ou <oulijun@huawei.com>
> Reviewed-by: Phil Yang <phil.yang@arm.com>
> ---
> V4->V5:
> -rewrite the commit log
> -add reviewed-by

Hi Lijun,

There were multiple other comments, it seems they are not addressed but only 
updated the commit log, can you please check comments to prev versions.

Before going into the details, my question was what happens if default key not 
provided at all?
It seems this has been already tried by Ophir [1], later reverted back [2] 
bringing the initial issue back.

According commit, the reason of revert is to support following command:
"flow create 0 <pattern> actions rss queues 0 1 end key_len 40 / end"

@Ophir, @Lijun,
Can we ignore the 'key_len' if the 'key' is not supported and solve current 
issue as initially intended ([1])?


[1] https://git.dpdk.org/dpdk/commit/?id=a4391f8bae8

[2] https://git.dpdk.org/dpdk/commit/?id=f3698c3d09a
Lijun Ou Oct. 15, 2020, 2:04 p.m. UTC | #2
在 2020/10/15 21:52, Ferruh Yigit 写道:
> On 10/15/2020 1:41 PM, Lijun Ou wrote:
>> When start the testpmd, the pmd driver initializes the RSS
>> configuration. Generally, the recommended RSS hash key is
>> used as the default key in the driver. In addition, the
>> default key is different from the default RSS flow in testpmd
>> without specifying RSS hash key. So. if you do not specify
>> the RSS key when creating an RSS rule, the testpmd uses the
>> default key as the default RSS key of the RSS rule. As a result,
>> you may mistakenly consider that the RSS key in use is the valid
>> default key of the NIC, actually, the key and the valid default
>> key of the NIC are two values.
>>
>> Consider the follow usage with testpmd:
>> 1. first, startup testpmd:
>> testpmd> show port 0 rss-hash key
>> RSS functions:
>>    all ipv4-frag ipv4-other ipv6-frag ipv6-other ip
>> RSS key:
>> -6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA38030F
>> -20C6A42B73BBEAC01FA
>> 2. create a rss rule
>> testpmd> flow create 0 ingress pattern eth / ipv4 / udp / end actions 
>> rss \
>> types ipv4-udp end queues end / end
>>
>> 3. show rss-hash key
>> testpmd> show port 0 rss-hash key
>> RSS functions:
>>    all ipv4-udp udp
>> RSS key:
>> -74657374706D6427732064656661756C74205253532068617368206B65792C206F
>> -76657272696465
>>
>> In order to solve the above problems, it use the NIC valid default
>> RSS key instead of the testpmd dummy RSS key in the flow configuration
>> when the RSS key is not specified in the flow rule. If the NIC RSS key
>> is invalid, it will use testpmd dummy RSS key as the default key.
>>
>> Fixes: ac8d22de2394 ("ethdev: flatten RSS configuration in flow API")
>> Cc: stable@dpdk.org
>>
>> Signed-off-by: Lijun Ou <oulijun@huawei.com>
>> Reviewed-by: Phil Yang <phil.yang@arm.com>
>> ---
>> V4->V5:
>> -rewrite the commit log
>> -add reviewed-by
> 
> Hi Lijun,
> 
> There were multiple other comments, it seems they are not addressed but 
> only updated the commit log, can you please check comments to prev 
> versions.
> 
> Before going into the details, my question was what happens if default 
> key not provided at all?
> It seems this has been already tried by Ophir [1], later reverted back 
> [2] bringing the initial issue back.
> 
> According commit, the reason of revert is to support following command:
> "flow create 0 <pattern> actions rss queues 0 1 end key_len 40 / end"
> 
> @Ophir, @Lijun,
> Can we ignore the 'key_len' if the 'key' is not supported and solve 
> current issue as initially intended ([1])?
> 
Hi, Ferruh
   I have discussed with Phil Yang about the problem in [1]. I think 
there may be other problems with the idea and there is no better 
solution. and we need to remove key_len definition from rte_rss_conf 
structure. They don't have a plan. And [1] was eventually reverted.

Thanks
Lijun Ou
> 
> [1] https://git.dpdk.org/dpdk/commit/?id=a4391f8bae8
> 
> [2] https://git.dpdk.org/dpdk/commit/?id=f3698c3d09a
> .
>
Ferruh Yigit Oct. 15, 2020, 2:43 p.m. UTC | #3
On 10/15/2020 3:04 PM, oulijun wrote:
> 
> 
> 在 2020/10/15 21:52, Ferruh Yigit 写道:
>> On 10/15/2020 1:41 PM, Lijun Ou wrote:
>>> When start the testpmd, the pmd driver initializes the RSS
>>> configuration. Generally, the recommended RSS hash key is
>>> used as the default key in the driver. In addition, the
>>> default key is different from the default RSS flow in testpmd
>>> without specifying RSS hash key. So. if you do not specify
>>> the RSS key when creating an RSS rule, the testpmd uses the
>>> default key as the default RSS key of the RSS rule. As a result,
>>> you may mistakenly consider that the RSS key in use is the valid
>>> default key of the NIC, actually, the key and the valid default
>>> key of the NIC are two values.
>>>
>>> Consider the follow usage with testpmd:
>>> 1. first, startup testpmd:
>>> testpmd> show port 0 rss-hash key
>>> RSS functions:
>>>    all ipv4-frag ipv4-other ipv6-frag ipv6-other ip
>>> RSS key:
>>> -6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA38030F
>>> -20C6A42B73BBEAC01FA
>>> 2. create a rss rule
>>> testpmd> flow create 0 ingress pattern eth / ipv4 / udp / end actions rss \
>>> types ipv4-udp end queues end / end
>>>
>>> 3. show rss-hash key
>>> testpmd> show port 0 rss-hash key
>>> RSS functions:
>>>    all ipv4-udp udp
>>> RSS key:
>>> -74657374706D6427732064656661756C74205253532068617368206B65792C206F
>>> -76657272696465
>>>
>>> In order to solve the above problems, it use the NIC valid default
>>> RSS key instead of the testpmd dummy RSS key in the flow configuration
>>> when the RSS key is not specified in the flow rule. If the NIC RSS key
>>> is invalid, it will use testpmd dummy RSS key as the default key.
>>>
>>> Fixes: ac8d22de2394 ("ethdev: flatten RSS configuration in flow API")
>>> Cc: stable@dpdk.org
>>>
>>> Signed-off-by: Lijun Ou <oulijun@huawei.com>
>>> Reviewed-by: Phil Yang <phil.yang@arm.com>
>>> ---
>>> V4->V5:
>>> -rewrite the commit log
>>> -add reviewed-by
>>
>> Hi Lijun,
>>
>> There were multiple other comments, it seems they are not addressed but only 
>> updated the commit log, can you please check comments to prev versions.
>>
>> Before going into the details, my question was what happens if default key not 
>> provided at all?
>> It seems this has been already tried by Ophir [1], later reverted back [2] 
>> bringing the initial issue back.
>>
>> According commit, the reason of revert is to support following command:
>> "flow create 0 <pattern> actions rss queues 0 1 end key_len 40 / end"
>>
>> @Ophir, @Lijun,
>> Can we ignore the 'key_len' if the 'key' is not supported and solve current 
>> issue as initially intended ([1])?
>>
> Hi, Ferruh
>    I have discussed with Phil Yang about the problem in [1]. I think there may 
> be other problems with the idea and there is no better solution. and we need to 
> remove key_len definition from rte_rss_conf structure. They don't have a plan. 
> And [1] was eventually reverted.
> 

Why ignoring 'key_len' (set it to zero) when there is no 'key' provided doesn't 
work?

> Thanks
> Lijun Ou
>>
>> [1] https://git.dpdk.org/dpdk/commit/?id=a4391f8bae8
>>
>> [2] https://git.dpdk.org/dpdk/commit/?id=f3698c3d09a
>> .
>>
Ferruh Yigit Oct. 15, 2020, 4:05 p.m. UTC | #4
On 10/15/2020 3:43 PM, Ferruh Yigit wrote:
> On 10/15/2020 3:04 PM, oulijun wrote:
>>
>>
>> 在 2020/10/15 21:52, Ferruh Yigit 写道:
>>> On 10/15/2020 1:41 PM, Lijun Ou wrote:
>>>> When start the testpmd, the pmd driver initializes the RSS
>>>> configuration. Generally, the recommended RSS hash key is
>>>> used as the default key in the driver. In addition, the
>>>> default key is different from the default RSS flow in testpmd
>>>> without specifying RSS hash key. So. if you do not specify
>>>> the RSS key when creating an RSS rule, the testpmd uses the
>>>> default key as the default RSS key of the RSS rule. As a result,
>>>> you may mistakenly consider that the RSS key in use is the valid
>>>> default key of the NIC, actually, the key and the valid default
>>>> key of the NIC are two values.
>>>>
>>>> Consider the follow usage with testpmd:
>>>> 1. first, startup testpmd:
>>>> testpmd> show port 0 rss-hash key
>>>> RSS functions:
>>>>    all ipv4-frag ipv4-other ipv6-frag ipv6-other ip
>>>> RSS key:
>>>> -6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA38030F
>>>> -20C6A42B73BBEAC01FA
>>>> 2. create a rss rule
>>>> testpmd> flow create 0 ingress pattern eth / ipv4 / udp / end actions rss \
>>>> types ipv4-udp end queues end / end
>>>>
>>>> 3. show rss-hash key
>>>> testpmd> show port 0 rss-hash key
>>>> RSS functions:
>>>>    all ipv4-udp udp
>>>> RSS key:
>>>> -74657374706D6427732064656661756C74205253532068617368206B65792C206F
>>>> -76657272696465
>>>>
>>>> In order to solve the above problems, it use the NIC valid default
>>>> RSS key instead of the testpmd dummy RSS key in the flow configuration
>>>> when the RSS key is not specified in the flow rule. If the NIC RSS key
>>>> is invalid, it will use testpmd dummy RSS key as the default key.
>>>>
>>>> Fixes: ac8d22de2394 ("ethdev: flatten RSS configuration in flow API")
>>>> Cc: stable@dpdk.org
>>>>
>>>> Signed-off-by: Lijun Ou <oulijun@huawei.com>
>>>> Reviewed-by: Phil Yang <phil.yang@arm.com>
>>>> ---
>>>> V4->V5:
>>>> -rewrite the commit log
>>>> -add reviewed-by
>>>
>>> Hi Lijun,
>>>
>>> There were multiple other comments, it seems they are not addressed but only 
>>> updated the commit log, can you please check comments to prev versions.
>>>
>>> Before going into the details, my question was what happens if default key 
>>> not provided at all?
>>> It seems this has been already tried by Ophir [1], later reverted back [2] 
>>> bringing the initial issue back.
>>>
>>> According commit, the reason of revert is to support following command:
>>> "flow create 0 <pattern> actions rss queues 0 1 end key_len 40 / end"
>>>
>>> @Ophir, @Lijun,
>>> Can we ignore the 'key_len' if the 'key' is not supported and solve current 
>>> issue as initially intended ([1])?
>>>
>> Hi, Ferruh
>>    I have discussed with Phil Yang about the problem in [1]. I think there may 
>> be other problems with the idea and there is no better solution. and we need 
>> to remove key_len definition from rte_rss_conf structure. They don't have a 
>> plan. And [1] was eventually reverted.
>>
> 
> Why ignoring 'key_len' (set it to zero) when there is no 'key' provided doesn't 
> work?
> 

What do you think [1] + following update, will it work?

  diff --git a/lib/librte_ethdev/rte_flow.c b/lib/librte_ethdev/rte_flow.c
  index ee4f3464fe..e7789c87b3 100644
  --- a/lib/librte_ethdev/rte_flow.c
  +++ b/lib/librte_ethdev/rte_flow.c
  @@ -535,7 +535,7 @@ rte_flow_conv_action_conf(void *buf, const size_t size,
                             }),
                             size > sizeof(*dst.rss) ? sizeof(*dst.rss) : size);
                  off = sizeof(*dst.rss);
  -               if (src.rss->key_len) {
  +               if (src.rss->key_len && src.rss->key) {
                          off = RTE_ALIGN_CEIL(off, sizeof(*dst.rss->key));
                          tmp = sizeof(*src.rss->key) * src.rss->key_len;
                          if (size >= off + tmp)

>> Thanks
>> Lijun Ou
>>>
>>> [1] https://git.dpdk.org/dpdk/commit/?id=a4391f8bae8
>>>
>>> [2] https://git.dpdk.org/dpdk/commit/?id=f3698c3d09a
>>> .
>>>
>
Ophir Munk Oct. 15, 2020, 11:21 p.m. UTC | #5
> From: Ferruh Yigit <ferruh.yigit@intel.com>
> >>> On 10/15/2020 1:41 PM, Lijun Ou wrote:
> >>>> When start the testpmd, the pmd driver initializes the RSS
> >>>> configuration. Generally, the recommended RSS hash key is used as
> >>>> the default key in the driver. In addition, the default key is
> >>>> different from the default RSS flow in testpmd without specifying
> >>>> RSS hash key. So. if you do not specify the RSS key when creating
> >>>> an RSS rule, the testpmd uses the default key as the default RSS
> >>>> key of the RSS rule. As a result, you may mistakenly consider that
> >>>> the RSS key in use is the valid default key of the NIC, actually,
> >>>> the key and the valid default key of the NIC are two values.
> >>>>
> >>>> Consider the follow usage with testpmd:
> >>>> 1. first, startup testpmd:
> >>>> testpmd> show port 0 rss-hash key
> >>>> RSS functions:
> >>>>    all ipv4-frag ipv4-other ipv6-frag ipv6-other ip RSS key:
> >>>> -
> 6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA3803
> 0F
> >>>> -20C6A42B73BBEAC01FA
> >>>> 2. create a rss rule
> >>>> testpmd> flow create 0 ingress pattern eth / ipv4 / udp / end
> >>>> testpmd> actions rss \
> >>>> types ipv4-udp end queues end / end
> >>>>
> >>>> 3. show rss-hash key
> >>>> testpmd> show port 0 rss-hash key
> >>>> RSS functions:
> >>>>    all ipv4-udp udp
> >>>> RSS key:
> >>>> -
> 74657374706D6427732064656661756C74205253532068617368206B65792C
> 206F
> >>>> -76657272696465
> >>>>
> >>>> In order to solve the above problems, it use the NIC valid default
> >>>> RSS key instead of the testpmd dummy RSS key in the flow
> >>>> configuration when the RSS key is not specified in the flow rule.
> >>>> If the NIC RSS key is invalid, it will use testpmd dummy RSS key as the
> default key.
> >>>>
> >>>> Fixes: ac8d22de2394 ("ethdev: flatten RSS configuration in flow
> >>>> API")
> >>>> Cc: stable@dpdk.org
> >>>>
> >>>> Signed-off-by: Lijun Ou <oulijun@huawei.com>
> >>>> Reviewed-by: Phil Yang <phil.yang@arm.com>
> >>>> ---
> >>>> V4->V5:
> >>>> -rewrite the commit log
> >>>> -add reviewed-by
> >>>
> >>> Hi Lijun,
> >>>
> >>> There were multiple other comments, it seems they are not addressed
> >>> but only updated the commit log, can you please check comments to
> prev versions.
> >>>
> >>> Before going into the details, my question was what happens if
> >>> default key not provided at all?
> >>> It seems this has been already tried by Ophir [1], later reverted
> >>> back [2] bringing the initial issue back.
> >>>
> >>> According commit, the reason of revert is to support following
> command:
> >>> "flow create 0 <pattern> actions rss queues 0 1 end key_len 40 / end"
> >>>
> >>> @Ophir, @Lijun,
> >>> Can we ignore the 'key_len' if the 'key' is not supported and solve
> >>> current issue as initially intended ([1])?
> >>>
> >> Hi, Ferruh
> >>    I have discussed with Phil Yang about the problem in [1]. I think
> >> there may be other problems with the idea and there is no better
> >> solution. and we need to remove key_len definition from rte_rss_conf
> >> structure. They don't have a plan. And [1] was eventually reverted.
> >>
> >
> > Why ignoring 'key_len' (set it to zero) when there is no 'key'
> > provided doesn't work?
> >
> 
> What do you think [1] + following update, will it work?
> 
>   diff --git a/lib/librte_ethdev/rte_flow.c b/lib/librte_ethdev/rte_flow.c
>   index ee4f3464fe..e7789c87b3 100644
>   --- a/lib/librte_ethdev/rte_flow.c
>   +++ b/lib/librte_ethdev/rte_flow.c
>   @@ -535,7 +535,7 @@ rte_flow_conv_action_conf(void *buf, const size_t
> size,
>                              }),
>                              size > sizeof(*dst.rss) ? sizeof(*dst.rss) : size);
>                   off = sizeof(*dst.rss);
>   -               if (src.rss->key_len) {
>   +               if (src.rss->key_len && src.rss->key) {
>                           off = RTE_ALIGN_CEIL(off, sizeof(*dst.rss->key));
>                           tmp = sizeof(*src.rss->key) * src.rss->key_len;
>                           if (size >= off + tmp)
> 

Ferruh, your suggestion ([1] + update) looks correct. I also verified it on mlx5 PMD.
Advantage: it's a generic fix for all dpdk applications using rte_flows (not just testpmd). 
It reduces code.
With this fix the responsibility of handling key==NULL and/or len==0 is moved to the PMDs (which is good).
 
With regard to Lijun patch - I liked the approach of overriding the default testpmd key with the default PMD key.
But it only addresses testpmd. More code was added. 
It seems OK to call rte_eth_dev_rss_hash_conf_get() as part of parsing RSS, but it would feel more confident if we could confirm it for all the PMDs (by testing) or at least review the PMDs rss_hash_conf_get() implementations.

> >> Thanks
> >> Lijun Ou
> >>>
Ferruh Yigit Oct. 15, 2020, 11:53 p.m. UTC | #6
On 10/16/2020 12:21 AM, Ophir Munk wrote:
>> From: Ferruh Yigit <ferruh.yigit@intel.com>
>>>>> On 10/15/2020 1:41 PM, Lijun Ou wrote:
>>>>>> When start the testpmd, the pmd driver initializes the RSS
>>>>>> configuration. Generally, the recommended RSS hash key is used as
>>>>>> the default key in the driver. In addition, the default key is
>>>>>> different from the default RSS flow in testpmd without specifying
>>>>>> RSS hash key. So. if you do not specify the RSS key when creating
>>>>>> an RSS rule, the testpmd uses the default key as the default RSS
>>>>>> key of the RSS rule. As a result, you may mistakenly consider that
>>>>>> the RSS key in use is the valid default key of the NIC, actually,
>>>>>> the key and the valid default key of the NIC are two values.
>>>>>>
>>>>>> Consider the follow usage with testpmd:
>>>>>> 1. first, startup testpmd:
>>>>>> testpmd> show port 0 rss-hash key
>>>>>> RSS functions:
>>>>>>     all ipv4-frag ipv4-other ipv6-frag ipv6-other ip RSS key:
>>>>>> -
>> 6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA3803
>> 0F
>>>>>> -20C6A42B73BBEAC01FA
>>>>>> 2. create a rss rule
>>>>>> testpmd> flow create 0 ingress pattern eth / ipv4 / udp / end
>>>>>> testpmd> actions rss \
>>>>>> types ipv4-udp end queues end / end
>>>>>>
>>>>>> 3. show rss-hash key
>>>>>> testpmd> show port 0 rss-hash key
>>>>>> RSS functions:
>>>>>>     all ipv4-udp udp
>>>>>> RSS key:
>>>>>> -
>> 74657374706D6427732064656661756C74205253532068617368206B65792C
>> 206F
>>>>>> -76657272696465
>>>>>>
>>>>>> In order to solve the above problems, it use the NIC valid default
>>>>>> RSS key instead of the testpmd dummy RSS key in the flow
>>>>>> configuration when the RSS key is not specified in the flow rule.
>>>>>> If the NIC RSS key is invalid, it will use testpmd dummy RSS key as the
>> default key.
>>>>>>
>>>>>> Fixes: ac8d22de2394 ("ethdev: flatten RSS configuration in flow
>>>>>> API")
>>>>>> Cc: stable@dpdk.org
>>>>>>
>>>>>> Signed-off-by: Lijun Ou <oulijun@huawei.com>
>>>>>> Reviewed-by: Phil Yang <phil.yang@arm.com>
>>>>>> ---
>>>>>> V4->V5:
>>>>>> -rewrite the commit log
>>>>>> -add reviewed-by
>>>>>
>>>>> Hi Lijun,
>>>>>
>>>>> There were multiple other comments, it seems they are not addressed
>>>>> but only updated the commit log, can you please check comments to
>> prev versions.
>>>>>
>>>>> Before going into the details, my question was what happens if
>>>>> default key not provided at all?
>>>>> It seems this has been already tried by Ophir [1], later reverted
>>>>> back [2] bringing the initial issue back.
>>>>>
>>>>> According commit, the reason of revert is to support following
>> command:
>>>>> "flow create 0 <pattern> actions rss queues 0 1 end key_len 40 / end"
>>>>>
>>>>> @Ophir, @Lijun,
>>>>> Can we ignore the 'key_len' if the 'key' is not supported and solve
>>>>> current issue as initially intended ([1])?
>>>>>
>>>> Hi, Ferruh
>>>>     I have discussed with Phil Yang about the problem in [1]. I think
>>>> there may be other problems with the idea and there is no better
>>>> solution. and we need to remove key_len definition from rte_rss_conf
>>>> structure. They don't have a plan. And [1] was eventually reverted.
>>>>
>>>
>>> Why ignoring 'key_len' (set it to zero) when there is no 'key'
>>> provided doesn't work?
>>>
>>
>> What do you think [1] + following update, will it work?
>>
>>    diff --git a/lib/librte_ethdev/rte_flow.c b/lib/librte_ethdev/rte_flow.c
>>    index ee4f3464fe..e7789c87b3 100644
>>    --- a/lib/librte_ethdev/rte_flow.c
>>    +++ b/lib/librte_ethdev/rte_flow.c
>>    @@ -535,7 +535,7 @@ rte_flow_conv_action_conf(void *buf, const size_t
>> size,
>>                               }),
>>                               size > sizeof(*dst.rss) ? sizeof(*dst.rss) : size);
>>                    off = sizeof(*dst.rss);
>>    -               if (src.rss->key_len) {
>>    +               if (src.rss->key_len && src.rss->key) {
>>                            off = RTE_ALIGN_CEIL(off, sizeof(*dst.rss->key));
>>                            tmp = sizeof(*src.rss->key) * src.rss->key_len;
>>                            if (size >= off + tmp)
>>
> 
> Ferruh, your suggestion ([1] + update) looks correct. I also verified it on mlx5 PMD.
> Advantage: it's a generic fix for all dpdk applications using rte_flows (not just testpmd).
> It reduces code.
> With this fix the responsibility of handling key==NULL and/or len==0 is moved to the PMDs (which is good).
>   
> With regard to Lijun patch - I liked the approach of overriding the default testpmd key with the default PMD key.
> But it only addresses testpmd. More code was added.
> It seems OK to call rte_eth_dev_rss_hash_conf_get() as part of parsing RSS, but it would feel more confident if we could confirm it for all the PMDs (by testing) or at least review the PMDs rss_hash_conf_get() implementations.
> 

Lijun's idea can work. There was a problem in implementation related to the key 
size assumption, which can be fixed.

Even it is fixed, when user gives a rss rule without a key, we are getting key 
from device and feeding same key back to device, this is unnecessary I think.
When user didn't provide a key, rss rule shouldn't touch the key at all.

Complication was when user provides key_len without a key, I think both ignoring 
or returning error in this case is OK.
Ajit Khaparde Oct. 16, 2020, 12:14 a.m. UTC | #7
On Thu, Oct 15, 2020 at 4:54 PM Ferruh Yigit <ferruh.yigit@intel.com> wrote:
>
> On 10/16/2020 12:21 AM, Ophir Munk wrote:
> >> From: Ferruh Yigit <ferruh.yigit@intel.com>
> >>>>> On 10/15/2020 1:41 PM, Lijun Ou wrote:
> >>>>>> When start the testpmd, the pmd driver initializes the RSS
> >>>>>> configuration. Generally, the recommended RSS hash key is used as
> >>>>>> the default key in the driver. In addition, the default key is
> >>>>>> different from the default RSS flow in testpmd without specifying
> >>>>>> RSS hash key. So. if you do not specify the RSS key when creating
> >>>>>> an RSS rule, the testpmd uses the default key as the default RSS
> >>>>>> key of the RSS rule. As a result, you may mistakenly consider that
> >>>>>> the RSS key in use is the valid default key of the NIC, actually,
> >>>>>> the key and the valid default key of the NIC are two values.
> >>>>>>
> >>>>>> Consider the follow usage with testpmd:
> >>>>>> 1. first, startup testpmd:
> >>>>>> testpmd> show port 0 rss-hash key
> >>>>>> RSS functions:
> >>>>>>     all ipv4-frag ipv4-other ipv6-frag ipv6-other ip RSS key:
> >>>>>> -
> >> 6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA3803
> >> 0F
> >>>>>> -20C6A42B73BBEAC01FA
> >>>>>> 2. create a rss rule
> >>>>>> testpmd> flow create 0 ingress pattern eth / ipv4 / udp / end
> >>>>>> testpmd> actions rss \
> >>>>>> types ipv4-udp end queues end / end
> >>>>>>
> >>>>>> 3. show rss-hash key
> >>>>>> testpmd> show port 0 rss-hash key
> >>>>>> RSS functions:
> >>>>>>     all ipv4-udp udp
> >>>>>> RSS key:
> >>>>>> -
> >> 74657374706D6427732064656661756C74205253532068617368206B65792C
> >> 206F
> >>>>>> -76657272696465
> >>>>>>
> >>>>>> In order to solve the above problems, it use the NIC valid default
> >>>>>> RSS key instead of the testpmd dummy RSS key in the flow
> >>>>>> configuration when the RSS key is not specified in the flow rule.
> >>>>>> If the NIC RSS key is invalid, it will use testpmd dummy RSS key as the
> >> default key.
> >>>>>>
> >>>>>> Fixes: ac8d22de2394 ("ethdev: flatten RSS configuration in flow
> >>>>>> API")
> >>>>>> Cc: stable@dpdk.org
> >>>>>>
> >>>>>> Signed-off-by: Lijun Ou <oulijun@huawei.com>
> >>>>>> Reviewed-by: Phil Yang <phil.yang@arm.com>
> >>>>>> ---
> >>>>>> V4->V5:
> >>>>>> -rewrite the commit log
> >>>>>> -add reviewed-by
> >>>>>
> >>>>> Hi Lijun,
> >>>>>
> >>>>> There were multiple other comments, it seems they are not addressed
> >>>>> but only updated the commit log, can you please check comments to
> >> prev versions.
> >>>>>
> >>>>> Before going into the details, my question was what happens if
> >>>>> default key not provided at all?
> >>>>> It seems this has been already tried by Ophir [1], later reverted
> >>>>> back [2] bringing the initial issue back.
> >>>>>
> >>>>> According commit, the reason of revert is to support following
> >> command:
> >>>>> "flow create 0 <pattern> actions rss queues 0 1 end key_len 40 / end"
> >>>>>
> >>>>> @Ophir, @Lijun,
> >>>>> Can we ignore the 'key_len' if the 'key' is not supported and solve
> >>>>> current issue as initially intended ([1])?
> >>>>>
> >>>> Hi, Ferruh
> >>>>     I have discussed with Phil Yang about the problem in [1]. I think
> >>>> there may be other problems with the idea and there is no better
> >>>> solution. and we need to remove key_len definition from rte_rss_conf
> >>>> structure. They don't have a plan. And [1] was eventually reverted.
> >>>>
> >>>
> >>> Why ignoring 'key_len' (set it to zero) when there is no 'key'
> >>> provided doesn't work?
> >>>
> >>
> >> What do you think [1] + following update, will it work?
> >>
> >>    diff --git a/lib/librte_ethdev/rte_flow.c b/lib/librte_ethdev/rte_flow.c
> >>    index ee4f3464fe..e7789c87b3 100644
> >>    --- a/lib/librte_ethdev/rte_flow.c
> >>    +++ b/lib/librte_ethdev/rte_flow.c
> >>    @@ -535,7 +535,7 @@ rte_flow_conv_action_conf(void *buf, const size_t
> >> size,
> >>                               }),
> >>                               size > sizeof(*dst.rss) ? sizeof(*dst.rss) : size);
> >>                    off = sizeof(*dst.rss);
> >>    -               if (src.rss->key_len) {
> >>    +               if (src.rss->key_len && src.rss->key) {
> >>                            off = RTE_ALIGN_CEIL(off, sizeof(*dst.rss->key));
> >>                            tmp = sizeof(*src.rss->key) * src.rss->key_len;
> >>                            if (size >= off + tmp)
> >>
> >
> > Ferruh, your suggestion ([1] + update) looks correct. I also verified it on mlx5 PMD.
> > Advantage: it's a generic fix for all dpdk applications using rte_flows (not just testpmd).
> > It reduces code.
> > With this fix the responsibility of handling key==NULL and/or len==0 is moved to the PMDs (which is good).
> >
> > With regard to Lijun patch - I liked the approach of overriding the default testpmd key with the default PMD key.
> > But it only addresses testpmd. More code was added.
> > It seems OK to call rte_eth_dev_rss_hash_conf_get() as part of parsing RSS, but it would feel more confident if we could confirm it for all the PMDs (by testing) or at least review the PMDs rss_hash_conf_get() implementations.
> >
>
> Lijun's idea can work. There was a problem in implementation related to the key
> size assumption, which can be fixed.
>
> Even it is fixed, when user gives a rss rule without a key, we are getting key
> from device and feeding same key back to device, this is unnecessary I think.
> When user didn't provide a key, rss rule shouldn't touch the key at all.
+1
But can we add this as expected behavior in the rte_flow document.

>
> Complication was when user provides key_len without a key, I think both ignoring
> or returning error in this case is OK.
Let's return an error if key_len is provided without a key.
Ophir Munk Oct. 16, 2020, 6:46 a.m. UTC | #8
> -----Original Message-----
> From: Ferruh Yigit <ferruh.yigit@intel.com>
<..>
> >>
> >
> > Ferruh, your suggestion ([1] + update) looks correct. I also verified it on
> mlx5 PMD.
> > Advantage: it's a generic fix for all dpdk applications using rte_flows (not
> just testpmd).
> > It reduces code.
> > With this fix the responsibility of handling key==NULL and/or len==0 is
> moved to the PMDs (which is good).
> >
> > With regard to Lijun patch - I liked the approach of overriding the default
> testpmd key with the default PMD key.
> > But it only addresses testpmd. More code was added.
> > It seems OK to call rte_eth_dev_rss_hash_conf_get() as part of parsing
> RSS, but it would feel more confident if we could confirm it for all the PMDs
> (by testing) or at least review the PMDs rss_hash_conf_get()
> implementations.
> >
> 
> Lijun's idea can work. There was a problem in implementation related to the
> key size assumption, which can be fixed.
> 
> Even it is fixed, when user gives a rss rule without a key, we are getting key
> from device and feeding same key back to device, this is unnecessary I think.

I agree.

> When user didn't provide a key, rss rule shouldn't touch the key at all.

Agreed as well.

> Complication was when user provides key_len without a key, I think both
> ignoring or returning error in this case is OK.

I think that in general flow rules should arrive "as is" to the PMD which has its
validation and translation APIs to handle all cases.
Lijun Ou Oct. 16, 2020, 9:39 a.m. UTC | #9
Hi, all
   Can the DPDK travis-CI check be configured locally? If yes, please 
provide guidance on the configuration check method.

Thanks
Lijun Ou

在 2020/10/16 3:56, 0-day Robot 写道:
> From: robot@bytheb.org
> 
> Test-Label: travis-robot
> Test-Status: WARNING
> http://dpdk.org/patch/80899
> 
> _Travis build: failed_
> Build URL: https://travis-ci.com/ovsrobot/dpdk/builds/190253613
> .
>
David Marchand Oct. 16, 2020, 9:41 a.m. UTC | #10
On Fri, Oct 16, 2020 at 11:40 AM oulijun <oulijun@huawei.com> wrote:
>
> Hi, all
>    Can the DPDK travis-CI check be configured locally? If yes, please
> provide guidance on the configuration check method.
>
> Thanks
> Lijun Ou
>
> 在 2020/10/16 3:56, 0-day Robot 写道:
> > From: robot@bytheb.org
> >
> > Test-Label: travis-robot
> > Test-Status: WARNING
> > http://dpdk.org/patch/80899
> >
> > _Travis build: failed_
> > Build URL: https://travis-ci.com/ovsrobot/dpdk/builds/190253613

The issue here should be fixed with
https://patchwork.dpdk.org/patch/81068/ which I am looking at.
Lijun Ou Oct. 16, 2020, 10:04 a.m. UTC | #11
在 2020/10/16 7:53, Ferruh Yigit 写道:
> On 10/16/2020 12:21 AM, Ophir Munk wrote:
>>> From: Ferruh Yigit <ferruh.yigit@intel.com>
>>>>>> On 10/15/2020 1:41 PM, Lijun Ou wrote:
>>>>>>> When start the testpmd, the pmd driver initializes the RSS
>>>>>>> configuration. Generally, the recommended RSS hash key is used as
>>>>>>> the default key in the driver. In addition, the default key is
>>>>>>> different from the default RSS flow in testpmd without specifying
>>>>>>> RSS hash key. So. if you do not specify the RSS key when creating
>>>>>>> an RSS rule, the testpmd uses the default key as the default RSS
>>>>>>> key of the RSS rule. As a result, you may mistakenly consider that
>>>>>>> the RSS key in use is the valid default key of the NIC, actually,
>>>>>>> the key and the valid default key of the NIC are two values.
>>>>>>>
>>>>>>> Consider the follow usage with testpmd:
>>>>>>> 1. first, startup testpmd:
>>>>>>> testpmd> show port 0 rss-hash key
>>>>>>> RSS functions:
>>>>>>>     all ipv4-frag ipv4-other ipv6-frag ipv6-other ip RSS key:
>>>>>>> -
>>> 6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA3803
>>> 0F
>>>>>>> -20C6A42B73BBEAC01FA
>>>>>>> 2. create a rss rule
>>>>>>> testpmd> flow create 0 ingress pattern eth / ipv4 / udp / end
>>>>>>> testpmd> actions rss \
>>>>>>> types ipv4-udp end queues end / end
>>>>>>>
>>>>>>> 3. show rss-hash key
>>>>>>> testpmd> show port 0 rss-hash key
>>>>>>> RSS functions:
>>>>>>>     all ipv4-udp udp
>>>>>>> RSS key:
>>>>>>> -
>>> 74657374706D6427732064656661756C74205253532068617368206B65792C
>>> 206F
>>>>>>> -76657272696465
>>>>>>>
>>>>>>> In order to solve the above problems, it use the NIC valid default
>>>>>>> RSS key instead of the testpmd dummy RSS key in the flow
>>>>>>> configuration when the RSS key is not specified in the flow rule.
>>>>>>> If the NIC RSS key is invalid, it will use testpmd dummy RSS key 
>>>>>>> as the
>>> default key.
>>>>>>>
>>>>>>> Fixes: ac8d22de2394 ("ethdev: flatten RSS configuration in flow
>>>>>>> API")
>>>>>>> Cc: stable@dpdk.org
>>>>>>>
>>>>>>> Signed-off-by: Lijun Ou <oulijun@huawei.com>
>>>>>>> Reviewed-by: Phil Yang <phil.yang@arm.com>
>>>>>>> ---
>>>>>>> V4->V5:
>>>>>>> -rewrite the commit log
>>>>>>> -add reviewed-by
>>>>>>
>>>>>> Hi Lijun,
>>>>>>
>>>>>> There were multiple other comments, it seems they are not addressed
>>>>>> but only updated the commit log, can you please check comments to
>>> prev versions.
>>>>>>
>>>>>> Before going into the details, my question was what happens if
>>>>>> default key not provided at all?
>>>>>> It seems this has been already tried by Ophir [1], later reverted
>>>>>> back [2] bringing the initial issue back.
>>>>>>
>>>>>> According commit, the reason of revert is to support following
>>> command:
>>>>>> "flow create 0 <pattern> actions rss queues 0 1 end key_len 40 / end"
>>>>>>
>>>>>> @Ophir, @Lijun,
>>>>>> Can we ignore the 'key_len' if the 'key' is not supported and solve
>>>>>> current issue as initially intended ([1])?
>>>>>>
>>>>> Hi, Ferruh
>>>>>     I have discussed with Phil Yang about the problem in [1]. I think
>>>>> there may be other problems with the idea and there is no better
>>>>> solution. and we need to remove key_len definition from rte_rss_conf
>>>>> structure. They don't have a plan. And [1] was eventually reverted.
>>>>>
>>>>
>>>> Why ignoring 'key_len' (set it to zero) when there is no 'key'
>>>> provided doesn't work?
>>>>
>>>
>>> What do you think [1] + following update, will it work?
>>>
>>>    diff --git a/lib/librte_ethdev/rte_flow.c 
>>> b/lib/librte_ethdev/rte_flow.c
>>>    index ee4f3464fe..e7789c87b3 100644
>>>    --- a/lib/librte_ethdev/rte_flow.c
>>>    +++ b/lib/librte_ethdev/rte_flow.c
>>>    @@ -535,7 +535,7 @@ rte_flow_conv_action_conf(void *buf, const size_t
>>> size,
>>>                               }),
>>>                               size > sizeof(*dst.rss) ? 
>>> sizeof(*dst.rss) : size);
>>>                    off = sizeof(*dst.rss);
>>>    -               if (src.rss->key_len) {
>>>    +               if (src.rss->key_len && src.rss->key) {
>>>                            off = RTE_ALIGN_CEIL(off, 
>>> sizeof(*dst.rss->key));
>>>                            tmp = sizeof(*src.rss->key) * 
>>> src.rss->key_len;
>>>                            if (size >= off + tmp)
>>>
>>
>> Ferruh, your suggestion ([1] + update) looks correct. I also verified 
>> it on mlx5 PMD.
>> Advantage: it's a generic fix for all dpdk applications using 
>> rte_flows (not just testpmd).
>> It reduces code.
>> With this fix the responsibility of handling key==NULL and/or len==0 
>> is moved to the PMDs (which is good).
>> With regard to Lijun patch - I liked the approach of overriding the 
>> default testpmd key with the default PMD key.
>> But it only addresses testpmd. More code was added.
>> It seems OK to call rte_eth_dev_rss_hash_conf_get() as part of parsing 
>> RSS, but it would feel more confident if we could confirm it for all 
>> the PMDs (by testing) or at least review the PMDs rss_hash_conf_get() 
>> implementations.
>>
> 
> Lijun's idea can work. There was a problem in implementation related to 
> the key size assumption, which can be fixed.
> 
> Even it is fixed, when user gives a rss rule without a key, we are 
> getting key from device and feeding same key back to device, this is 
> unnecessary I think.
> When user didn't provide a key, rss rule shouldn't touch the key at all.
> 
Do you mean that the driver should not configure the key to the hardware 
when the RSS key is not specified?
If so, we cannot identify when the user does not specify the RSS key to 
determine whether to configure the key.
In hardware design, most vendors do not configure keys for hardware 
independently, which may be associated with other RSS config parameters.
I think it would be reasonable for us to reconfigure the RSS key with 
the default value configured and valid in the hardware as the default 
value if the user does not specify the RSS key.
> Complication was when user provides key_len without a key, I think both 
> ignoring or returning error in this case is OK.
I agree. However, I think it is meaningless to expose the RSS key length 
to users. Do you consider deleting the RSS key length directly?
> .
>
Lijun Ou Oct. 16, 2020, 10:05 a.m. UTC | #12
在 2020/10/16 14:46, Ophir Munk 写道:
> 
> 
>> -----Original Message-----
>> From: Ferruh Yigit <ferruh.yigit@intel.com>
> <..>
>>>>
>>>
>>> Ferruh, your suggestion ([1] + update) looks correct. I also verified it on
>> mlx5 PMD.
>>> Advantage: it's a generic fix for all dpdk applications using rte_flows (not
>> just testpmd).
>>> It reduces code.
>>> With this fix the responsibility of handling key==NULL and/or len==0 is
>> moved to the PMDs (which is good).
>>>
>>> With regard to Lijun patch - I liked the approach of overriding the default
>> testpmd key with the default PMD key.
>>> But it only addresses testpmd. More code was added.
>>> It seems OK to call rte_eth_dev_rss_hash_conf_get() as part of parsing
>> RSS, but it would feel more confident if we could confirm it for all the PMDs
>> (by testing) or at least review the PMDs rss_hash_conf_get()
>> implementations.
>>>
>>
>> Lijun's idea can work. There was a problem in implementation related to the
>> key size assumption, which can be fixed.
>>
>> Even it is fixed, when user gives a rss rule without a key, we are getting key
>> from device and feeding same key back to device, this is unnecessary I think.
> 
> I agree.
> 
>> When user didn't provide a key, rss rule shouldn't touch the key at all.
> 
> Agreed as well.
> 
>> Complication was when user provides key_len without a key, I think both
>> ignoring or returning error in this case is OK.
> 
> I think that in general flow rules should arrive "as is" to the PMD which has its
> validation and translation APIs to handle all cases.
> 
Do you agree with [1] + Ferruh update solution?
Ferruh Yigit Oct. 16, 2020, 10:57 a.m. UTC | #13
On 10/16/2020 11:04 AM, oulijun wrote:
> 
> 
> 在 2020/10/16 7:53, Ferruh Yigit 写道:
>> On 10/16/2020 12:21 AM, Ophir Munk wrote:
>>>> From: Ferruh Yigit <ferruh.yigit@intel.com>
>>>>>>> On 10/15/2020 1:41 PM, Lijun Ou wrote:
>>>>>>>> When start the testpmd, the pmd driver initializes the RSS
>>>>>>>> configuration. Generally, the recommended RSS hash key is used as
>>>>>>>> the default key in the driver. In addition, the default key is
>>>>>>>> different from the default RSS flow in testpmd without specifying
>>>>>>>> RSS hash key. So. if you do not specify the RSS key when creating
>>>>>>>> an RSS rule, the testpmd uses the default key as the default RSS
>>>>>>>> key of the RSS rule. As a result, you may mistakenly consider that
>>>>>>>> the RSS key in use is the valid default key of the NIC, actually,
>>>>>>>> the key and the valid default key of the NIC are two values.
>>>>>>>>
>>>>>>>> Consider the follow usage with testpmd:
>>>>>>>> 1. first, startup testpmd:
>>>>>>>> testpmd> show port 0 rss-hash key
>>>>>>>> RSS functions:
>>>>>>>>     all ipv4-frag ipv4-other ipv6-frag ipv6-other ip RSS key:
>>>>>>>> -
>>>> 6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA3803
>>>> 0F
>>>>>>>> -20C6A42B73BBEAC01FA
>>>>>>>> 2. create a rss rule
>>>>>>>> testpmd> flow create 0 ingress pattern eth / ipv4 / udp / end
>>>>>>>> testpmd> actions rss \
>>>>>>>> types ipv4-udp end queues end / end
>>>>>>>>
>>>>>>>> 3. show rss-hash key
>>>>>>>> testpmd> show port 0 rss-hash key
>>>>>>>> RSS functions:
>>>>>>>>     all ipv4-udp udp
>>>>>>>> RSS key:
>>>>>>>> -
>>>> 74657374706D6427732064656661756C74205253532068617368206B65792C
>>>> 206F
>>>>>>>> -76657272696465
>>>>>>>>
>>>>>>>> In order to solve the above problems, it use the NIC valid default
>>>>>>>> RSS key instead of the testpmd dummy RSS key in the flow
>>>>>>>> configuration when the RSS key is not specified in the flow rule.
>>>>>>>> If the NIC RSS key is invalid, it will use testpmd dummy RSS key as the
>>>> default key.
>>>>>>>>
>>>>>>>> Fixes: ac8d22de2394 ("ethdev: flatten RSS configuration in flow
>>>>>>>> API")
>>>>>>>> Cc: stable@dpdk.org
>>>>>>>>
>>>>>>>> Signed-off-by: Lijun Ou <oulijun@huawei.com>
>>>>>>>> Reviewed-by: Phil Yang <phil.yang@arm.com>
>>>>>>>> ---
>>>>>>>> V4->V5:
>>>>>>>> -rewrite the commit log
>>>>>>>> -add reviewed-by
>>>>>>>
>>>>>>> Hi Lijun,
>>>>>>>
>>>>>>> There were multiple other comments, it seems they are not addressed
>>>>>>> but only updated the commit log, can you please check comments to
>>>> prev versions.
>>>>>>>
>>>>>>> Before going into the details, my question was what happens if
>>>>>>> default key not provided at all?
>>>>>>> It seems this has been already tried by Ophir [1], later reverted
>>>>>>> back [2] bringing the initial issue back.
>>>>>>>
>>>>>>> According commit, the reason of revert is to support following
>>>> command:
>>>>>>> "flow create 0 <pattern> actions rss queues 0 1 end key_len 40 / end"
>>>>>>>
>>>>>>> @Ophir, @Lijun,
>>>>>>> Can we ignore the 'key_len' if the 'key' is not supported and solve
>>>>>>> current issue as initially intended ([1])?
>>>>>>>
>>>>>> Hi, Ferruh
>>>>>>     I have discussed with Phil Yang about the problem in [1]. I think
>>>>>> there may be other problems with the idea and there is no better
>>>>>> solution. and we need to remove key_len definition from rte_rss_conf
>>>>>> structure. They don't have a plan. And [1] was eventually reverted.
>>>>>>
>>>>>
>>>>> Why ignoring 'key_len' (set it to zero) when there is no 'key'
>>>>> provided doesn't work?
>>>>>
>>>>
>>>> What do you think [1] + following update, will it work?
>>>>
>>>>    diff --git a/lib/librte_ethdev/rte_flow.c b/lib/librte_ethdev/rte_flow.c
>>>>    index ee4f3464fe..e7789c87b3 100644
>>>>    --- a/lib/librte_ethdev/rte_flow.c
>>>>    +++ b/lib/librte_ethdev/rte_flow.c
>>>>    @@ -535,7 +535,7 @@ rte_flow_conv_action_conf(void *buf, const size_t
>>>> size,
>>>>                               }),
>>>>                               size > sizeof(*dst.rss) ? sizeof(*dst.rss) : 
>>>> size);
>>>>                    off = sizeof(*dst.rss);
>>>>    -               if (src.rss->key_len) {
>>>>    +               if (src.rss->key_len && src.rss->key) {
>>>>                            off = RTE_ALIGN_CEIL(off, sizeof(*dst.rss->key));
>>>>                            tmp = sizeof(*src.rss->key) * src.rss->key_len;
>>>>                            if (size >= off + tmp)
>>>>
>>>
>>> Ferruh, your suggestion ([1] + update) looks correct. I also verified it on 
>>> mlx5 PMD.
>>> Advantage: it's a generic fix for all dpdk applications using rte_flows (not 
>>> just testpmd).
>>> It reduces code.
>>> With this fix the responsibility of handling key==NULL and/or len==0 is moved 
>>> to the PMDs (which is good).
>>> With regard to Lijun patch - I liked the approach of overriding the default 
>>> testpmd key with the default PMD key.
>>> But it only addresses testpmd. More code was added.
>>> It seems OK to call rte_eth_dev_rss_hash_conf_get() as part of parsing RSS, 
>>> but it would feel more confident if we could confirm it for all the PMDs (by 
>>> testing) or at least review the PMDs rss_hash_conf_get() implementations.
>>>
>>
>> Lijun's idea can work. There was a problem in implementation related to the 
>> key size assumption, which can be fixed.
>>
>> Even it is fixed, when user gives a rss rule without a key, we are getting key 
>> from device and feeding same key back to device, this is unnecessary I think.
>> When user didn't provide a key, rss rule shouldn't touch the key at all.
>>
> Do you mean that the driver should not configure the key to the hardware when 
> the RSS key is not specified?

We are talking about 'key' in RSS rte flow rule, not generally configuring the 
device for RSS.

If a specific rss rte flow rule, lets say to filter some defined packets to some 
specific queues, don't have 'key' as a part of rule, I am suggesting not touch 
'key' configuration of the device.

> If so, we cannot identify when the user does not specify the RSS key to 
> determine whether to configure the key.

I think we can. If the rule has 'key' attribute, driver can use that key to 
program the device, if 'key' attribute is missing, don't do anything related to 
the rss key.

> In hardware design, most vendors do not configure keys for hardware 
> independently, which may be associated with other RSS config parameters.
> I think it would be reasonable for us to reconfigure the RSS key with the 
> default value configured and valid in the hardware as the default value if the 
> user does not specify the RSS key.

There are already APIs to update the RSS configuration, like RSS key, hash 
functions etc.. They are independent from the rte flow RSS rule.
Application should configure RSS according needs using those APIs.

The question here is about each RSS rte flow rule that can update the key. Am I 
missing something?

>> Complication was when user provides key_len without a key, I think both 
>> ignoring or returning error in this case is OK.
> I agree. However, I think it is meaningless to expose the RSS key length to 
> users. Do you consider deleting the RSS key length directly?

Isn't both 'key' and 'key_len' needed to program the RSS key? Can the driver 
know the size of the 'key' without 'key_len'?
And driver should verify these provided values.
Ophir Munk Oct. 16, 2020, 2:59 p.m. UTC | #14
Ferruh Yigit <ferruh.yigit@intel.com> wrote:
> Sent: Friday, October 16, 2020 1:57 PM
> >> Lijun's idea can work. There was a problem in implementation related
> >> to the key size assumption, which can be fixed.
> >>
> >> Even it is fixed, when user gives a rss rule without a key, we are
> >> getting key from device and feeding same key back to device, this is
> unnecessary I think.
> >> When user didn't provide a key, rss rule shouldn't touch the key at all.
> >>
> > Do you mean that the driver should not configure the key to the
> > hardware when the RSS key is not specified?
> 

I would consider 2 user types:
1. A user who knows the HW capabilities and wish to get a distribution by specifying an explicit key.
2. A clueless user that is not familiar with the NIC HW. For example, a switching application which runs on any NIC. In that case the switching application will not configure a key. The HW driver can decide to configure the default key if not configured.
On the rte flow level we should communicate to the PMD that no key was specified.
Inside the PMD any decision can be taken (e.g. configuring the default key).
All of this is supported by PMD call backs to rte calls such as flow_validate(), flow_translate().

> We are talking about 'key' in RSS rte flow rule, not generally configuring the
> device for RSS.
> 
> If a specific rss rte flow rule, lets say to filter some defined packets to some
> specific queues, don't have 'key' as a part of rule, I am suggesting not touch
> 'key' configuration of the device.
> 
> > If so, we cannot identify when the user does not specify the RSS key
> > to determine whether to configure the key.
> 
> I think we can. If the rule has 'key' attribute, driver can use that key to
> program the device, if 'key' attribute is missing, don't do anything related to
> the rss key.

The driver is free to take any decision when a key is missing.
 
> > In hardware design, most vendors do not configure keys for hardware
> > independently, which may be associated with other RSS config parameters.
> > I think it would be reasonable for us to reconfigure the RSS key with
> > the default value configured and valid in the hardware as the default
> > value if the user does not specify the RSS key.
> 
> There are already APIs to update the RSS configuration, like RSS key, hash
> functions etc.. They are independent from the rte flow RSS rule.
> Application should configure RSS according needs using those APIs.
> 
> The question here is about each RSS rte flow rule that can update the key.
> Am I missing something?
> 
> >> Complication was when user provides key_len without a key, I think
> >> both ignoring or returning error in this case is OK.
> > I agree. However, I think it is meaningless to expose the RSS key
> > length to users. Do you consider deleting the RSS key length directly?
> 
> Isn't both 'key' and 'key_len' needed to program the RSS key? Can the driver
> know the size of the 'key' without 'key_len'?
> And driver should verify these provided values.

The key_len must be explicit. Key itself is just a pointer to an array of bytes.
Ophir Munk Oct. 16, 2020, 3:13 p.m. UTC | #15
> -----Original Message-----
> From: oulijun <oulijun@huawei.com>
> Sent: Friday, October 16, 2020 1:06 PM
ndle all cases.
> >
> Do you agree with [1] + Ferruh update solution?

Yes.
In that case your PMD may get a NULL RSS key. Then you can override it with your default key (if HW is not already configured with the default).
Lijun Ou Oct. 20, 2020, 9 a.m. UTC | #16
在 2020/10/16 18:57, Ferruh Yigit 写道:
> On 10/16/2020 11:04 AM, oulijun wrote:
>>
>>
>> 在 2020/10/16 7:53, Ferruh Yigit 写道:
>>> On 10/16/2020 12:21 AM, Ophir Munk wrote:
>>>>> From: Ferruh Yigit <ferruh.yigit@intel.com>
>>>>>>>> On 10/15/2020 1:41 PM, Lijun Ou wrote:
>>>>>>>>> When start the testpmd, the pmd driver initializes the RSS
>>>>>>>>> configuration. Generally, the recommended RSS hash key is used as
>>>>>>>>> the default key in the driver. In addition, the default key is
>>>>>>>>> different from the default RSS flow in testpmd without specifying
>>>>>>>>> RSS hash key. So. if you do not specify the RSS key when creating
>>>>>>>>> an RSS rule, the testpmd uses the default key as the default RSS
>>>>>>>>> key of the RSS rule. As a result, you may mistakenly consider that
>>>>>>>>> the RSS key in use is the valid default key of the NIC, actually,
>>>>>>>>> the key and the valid default key of the NIC are two values.
>>>>>>>>>
>>>>>>>>> Consider the follow usage with testpmd:
>>>>>>>>> 1. first, startup testpmd:
>>>>>>>>> testpmd> show port 0 rss-hash key
>>>>>>>>> RSS functions:
>>>>>>>>>     all ipv4-frag ipv4-other ipv6-frag ipv6-other ip RSS key:
>>>>>>>>> -
>>>>> 6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA3803
>>>>> 0F
>>>>>>>>> -20C6A42B73BBEAC01FA
>>>>>>>>> 2. create a rss rule
>>>>>>>>> testpmd> flow create 0 ingress pattern eth / ipv4 / udp / end
>>>>>>>>> testpmd> actions rss \
>>>>>>>>> types ipv4-udp end queues end / end
>>>>>>>>>
>>>>>>>>> 3. show rss-hash key
>>>>>>>>> testpmd> show port 0 rss-hash key
>>>>>>>>> RSS functions:
>>>>>>>>>     all ipv4-udp udp
>>>>>>>>> RSS key:
>>>>>>>>> -
>>>>> 74657374706D6427732064656661756C74205253532068617368206B65792C
>>>>> 206F
>>>>>>>>> -76657272696465
>>>>>>>>>
>>>>>>>>> In order to solve the above problems, it use the NIC valid default
>>>>>>>>> RSS key instead of the testpmd dummy RSS key in the flow
>>>>>>>>> configuration when the RSS key is not specified in the flow rule.
>>>>>>>>> If the NIC RSS key is invalid, it will use testpmd dummy RSS 
>>>>>>>>> key as the
>>>>> default key.
>>>>>>>>>
>>>>>>>>> Fixes: ac8d22de2394 ("ethdev: flatten RSS configuration in flow
>>>>>>>>> API")
>>>>>>>>> Cc: stable@dpdk.org
>>>>>>>>>
>>>>>>>>> Signed-off-by: Lijun Ou <oulijun@huawei.com>
>>>>>>>>> Reviewed-by: Phil Yang <phil.yang@arm.com>
>>>>>>>>> ---
>>>>>>>>> V4->V5:
>>>>>>>>> -rewrite the commit log
>>>>>>>>> -add reviewed-by
>>>>>>>>
>>>>>>>> Hi Lijun,
>>>>>>>>
>>>>>>>> There were multiple other comments, it seems they are not addressed
>>>>>>>> but only updated the commit log, can you please check comments to
>>>>> prev versions.
>>>>>>>>
>>>>>>>> Before going into the details, my question was what happens if
>>>>>>>> default key not provided at all?
>>>>>>>> It seems this has been already tried by Ophir [1], later reverted
>>>>>>>> back [2] bringing the initial issue back.
>>>>>>>>
>>>>>>>> According commit, the reason of revert is to support following
>>>>> command:
>>>>>>>> "flow create 0 <pattern> actions rss queues 0 1 end key_len 40 / 
>>>>>>>> end"
>>>>>>>>
>>>>>>>> @Ophir, @Lijun,
>>>>>>>> Can we ignore the 'key_len' if the 'key' is not supported and solve
>>>>>>>> current issue as initially intended ([1])?
>>>>>>>>
>>>>>>> Hi, Ferruh
>>>>>>>     I have discussed with Phil Yang about the problem in [1]. I 
>>>>>>> think
>>>>>>> there may be other problems with the idea and there is no better
>>>>>>> solution. and we need to remove key_len definition from rte_rss_conf
>>>>>>> structure. They don't have a plan. And [1] was eventually reverted.
>>>>>>>
>>>>>>
>>>>>> Why ignoring 'key_len' (set it to zero) when there is no 'key'
>>>>>> provided doesn't work?
>>>>>>
>>>>>
>>>>> What do you think [1] + following update, will it work?
>>>>>
>>>>>    diff --git a/lib/librte_ethdev/rte_flow.c 
>>>>> b/lib/librte_ethdev/rte_flow.c
>>>>>    index ee4f3464fe..e7789c87b3 100644
>>>>>    --- a/lib/librte_ethdev/rte_flow.c
>>>>>    +++ b/lib/librte_ethdev/rte_flow.c
>>>>>    @@ -535,7 +535,7 @@ rte_flow_conv_action_conf(void *buf, const 
>>>>> size_t
>>>>> size,
>>>>>                               }),
>>>>>                               size > sizeof(*dst.rss) ? 
>>>>> sizeof(*dst.rss) : size);
>>>>>                    off = sizeof(*dst.rss);
>>>>>    -               if (src.rss->key_len) {
>>>>>    +               if (src.rss->key_len && src.rss->key) {
>>>>>                            off = RTE_ALIGN_CEIL(off, 
>>>>> sizeof(*dst.rss->key));
>>>>>                            tmp = sizeof(*src.rss->key) * 
>>>>> src.rss->key_len;
>>>>>                            if (size >= off + tmp)
>>>>>
>>>>
>>>> Ferruh, your suggestion ([1] + update) looks correct. I also 
>>>> verified it on mlx5 PMD.
>>>> Advantage: it's a generic fix for all dpdk applications using 
>>>> rte_flows (not just testpmd).
>>>> It reduces code.
>>>> With this fix the responsibility of handling key==NULL and/or len==0 
>>>> is moved to the PMDs (which is good).
>>>> With regard to Lijun patch - I liked the approach of overriding the 
>>>> default testpmd key with the default PMD key.
>>>> But it only addresses testpmd. More code was added.
>>>> It seems OK to call rte_eth_dev_rss_hash_conf_get() as part of 
>>>> parsing RSS, but it would feel more confident if we could confirm it 
>>>> for all the PMDs (by testing) or at least review the PMDs 
>>>> rss_hash_conf_get() implementations.
>>>>
>>>
>>> Lijun's idea can work. There was a problem in implementation related 
>>> to the key size assumption, which can be fixed.
>>>
>>> Even it is fixed, when user gives a rss rule without a key, we are 
>>> getting key from device and feeding same key back to device, this is 
>>> unnecessary I think.
>>> When user didn't provide a key, rss rule shouldn't touch the key at all.
>>>
>> Do you mean that the driver should not configure the key to the 
>> hardware when the RSS key is not specified?
> 
> We are talking about 'key' in RSS rte flow rule, not generally 
> configuring the device for RSS.
> 
Yes, I know. However, I am talking about some implementations. The 
configuration interface may be the same as that for configuring RSS
parameters of the device in general mode.
> If a specific rss rte flow rule, lets say to filter some defined packets 
> to some specific queues, don't have 'key' as a part of rule, I am 
> suggesting not touch 'key' configuration of the device.
> 
Yes, I agree. I think your point of view is very reasonable and a more
appropriate implementation. However, for the sake of simplicity and
other considerations in the architecture design of some devices, the
hardware may support reasonable hybrid configuration for paramter 
configuration in the RSS config, rather than independent configuration
for hw, that is, maybe touch A of rss configuration of the device and
must touch B of rss configuration. As a result, If the preceding 
scenario occurs, it is reasonable to configure an RSS config that does 
not change and specified.What do you think?
>> If so, we cannot identify when the user does not specify the RSS key 
>> to determine whether to configure the key.
> 
> I think we can. If the rule has 'key' attribute, driver can use that key 
> to program the device, if 'key' attribute is missing, don't do anything 
> related to the rss key.
> 
Yes, if 'key' attribute is missing, don't do anything related to the rss 
key. It's more reasonable.
But, If he can't do that, I think it is reasonable to configure the 
default key that already exists on the device. The only disadvantage is 
wasteful.
>> In hardware design, most vendors do not configure keys for hardware 
>> independently, which may be associated with other RSS config parameters.
>> I think it would be reasonable for us to reconfigure the RSS key with 
>> the default value configured and valid in the hardware as the default 
>> value if the user does not specify the RSS key.
> 
> There are already APIs to update the RSS configuration, like RSS key, 
> hash functions etc.. They are independent from the rte flow RSS rule.
> Application should configure RSS according needs using those APIs.
> 
Yes, should do this. Are there any unreasonable users who simply do not 
know these APIs or do not want to use them and want to configure some 
parameters through the rte flow RSS rule without changing other parameters?
> The question here is about each RSS rte flow rule that can update the 
> key. Am I missing something?
I don't think you've misunderstood the general situation.
Even if the RSS key is not specified in a rule, the driver uses the 
default key value to re-touch the device.
> 
>>> Complication was when user provides key_len without a key, I think 
>>> both ignoring or returning error in this case is OK.
>> I agree. However, I think it is meaningless to expose the RSS key 
>> length to users. Do you consider deleting the RSS key length directly?
> 
> Isn't both 'key' and 'key_len' needed to program the RSS key? Can the 
> driver know the size of the 'key' without 'key_len'?
> And driver should verify these provided values.
> 
I know and agree.It is only during the analysis of the [1] scheme, from 
the time it was proposed to the time it was withdrawn, that I found some 
guys in the community questioned the significance of key_len.

Back to the subject, what is your plan for the solution in the patch?
> .
>
Ferruh Yigit Oct. 20, 2020, 10:02 a.m. UTC | #17
On 10/20/2020 10:00 AM, oulijun wrote:
> 
> 
> 在 2020/10/16 18:57, Ferruh Yigit 写道:
>> On 10/16/2020 11:04 AM, oulijun wrote:
>>>
>>>
>>> 在 2020/10/16 7:53, Ferruh Yigit 写道:
>>>> On 10/16/2020 12:21 AM, Ophir Munk wrote:
>>>>>> From: Ferruh Yigit <ferruh.yigit@intel.com>
>>>>>>>>> On 10/15/2020 1:41 PM, Lijun Ou wrote:
>>>>>>>>>> When start the testpmd, the pmd driver initializes the RSS
>>>>>>>>>> configuration. Generally, the recommended RSS hash key is used as
>>>>>>>>>> the default key in the driver. In addition, the default key is
>>>>>>>>>> different from the default RSS flow in testpmd without specifying
>>>>>>>>>> RSS hash key. So. if you do not specify the RSS key when creating
>>>>>>>>>> an RSS rule, the testpmd uses the default key as the default RSS
>>>>>>>>>> key of the RSS rule. As a result, you may mistakenly consider that
>>>>>>>>>> the RSS key in use is the valid default key of the NIC, actually,
>>>>>>>>>> the key and the valid default key of the NIC are two values.
>>>>>>>>>>
>>>>>>>>>> Consider the follow usage with testpmd:
>>>>>>>>>> 1. first, startup testpmd:
>>>>>>>>>> testpmd> show port 0 rss-hash key
>>>>>>>>>> RSS functions:
>>>>>>>>>>     all ipv4-frag ipv4-other ipv6-frag ipv6-other ip RSS key:
>>>>>>>>>> -
>>>>>> 6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA3803
>>>>>> 0F
>>>>>>>>>> -20C6A42B73BBEAC01FA
>>>>>>>>>> 2. create a rss rule
>>>>>>>>>> testpmd> flow create 0 ingress pattern eth / ipv4 / udp / end
>>>>>>>>>> testpmd> actions rss \
>>>>>>>>>> types ipv4-udp end queues end / end
>>>>>>>>>>
>>>>>>>>>> 3. show rss-hash key
>>>>>>>>>> testpmd> show port 0 rss-hash key
>>>>>>>>>> RSS functions:
>>>>>>>>>>     all ipv4-udp udp
>>>>>>>>>> RSS key:
>>>>>>>>>> -
>>>>>> 74657374706D6427732064656661756C74205253532068617368206B65792C
>>>>>> 206F
>>>>>>>>>> -76657272696465
>>>>>>>>>>
>>>>>>>>>> In order to solve the above problems, it use the NIC valid default
>>>>>>>>>> RSS key instead of the testpmd dummy RSS key in the flow
>>>>>>>>>> configuration when the RSS key is not specified in the flow rule.
>>>>>>>>>> If the NIC RSS key is invalid, it will use testpmd dummy RSS key as the
>>>>>> default key.
>>>>>>>>>>
>>>>>>>>>> Fixes: ac8d22de2394 ("ethdev: flatten RSS configuration in flow
>>>>>>>>>> API")
>>>>>>>>>> Cc: stable@dpdk.org
>>>>>>>>>>
>>>>>>>>>> Signed-off-by: Lijun Ou <oulijun@huawei.com>
>>>>>>>>>> Reviewed-by: Phil Yang <phil.yang@arm.com>
>>>>>>>>>> ---
>>>>>>>>>> V4->V5:
>>>>>>>>>> -rewrite the commit log
>>>>>>>>>> -add reviewed-by
>>>>>>>>>
>>>>>>>>> Hi Lijun,
>>>>>>>>>
>>>>>>>>> There were multiple other comments, it seems they are not addressed
>>>>>>>>> but only updated the commit log, can you please check comments to
>>>>>> prev versions.
>>>>>>>>>
>>>>>>>>> Before going into the details, my question was what happens if
>>>>>>>>> default key not provided at all?
>>>>>>>>> It seems this has been already tried by Ophir [1], later reverted
>>>>>>>>> back [2] bringing the initial issue back.
>>>>>>>>>
>>>>>>>>> According commit, the reason of revert is to support following
>>>>>> command:
>>>>>>>>> "flow create 0 <pattern> actions rss queues 0 1 end key_len 40 / end"
>>>>>>>>>
>>>>>>>>> @Ophir, @Lijun,
>>>>>>>>> Can we ignore the 'key_len' if the 'key' is not supported and solve
>>>>>>>>> current issue as initially intended ([1])?
>>>>>>>>>
>>>>>>>> Hi, Ferruh
>>>>>>>>     I have discussed with Phil Yang about the problem in [1]. I think
>>>>>>>> there may be other problems with the idea and there is no better
>>>>>>>> solution. and we need to remove key_len definition from rte_rss_conf
>>>>>>>> structure. They don't have a plan. And [1] was eventually reverted.
>>>>>>>>
>>>>>>>
>>>>>>> Why ignoring 'key_len' (set it to zero) when there is no 'key'
>>>>>>> provided doesn't work?
>>>>>>>
>>>>>>
>>>>>> What do you think [1] + following update, will it work?
>>>>>>
>>>>>>    diff --git a/lib/librte_ethdev/rte_flow.c b/lib/librte_ethdev/rte_flow.c
>>>>>>    index ee4f3464fe..e7789c87b3 100644
>>>>>>    --- a/lib/librte_ethdev/rte_flow.c
>>>>>>    +++ b/lib/librte_ethdev/rte_flow.c
>>>>>>    @@ -535,7 +535,7 @@ rte_flow_conv_action_conf(void *buf, const size_t
>>>>>> size,
>>>>>>                               }),
>>>>>>                               size > sizeof(*dst.rss) ? sizeof(*dst.rss) : 
>>>>>> size);
>>>>>>                    off = sizeof(*dst.rss);
>>>>>>    -               if (src.rss->key_len) {
>>>>>>    +               if (src.rss->key_len && src.rss->key) {
>>>>>>                            off = RTE_ALIGN_CEIL(off, sizeof(*dst.rss->key));
>>>>>>                            tmp = sizeof(*src.rss->key) * src.rss->key_len;
>>>>>>                            if (size >= off + tmp)
>>>>>>
>>>>>
>>>>> Ferruh, your suggestion ([1] + update) looks correct. I also verified it on 
>>>>> mlx5 PMD.
>>>>> Advantage: it's a generic fix for all dpdk applications using rte_flows 
>>>>> (not just testpmd).
>>>>> It reduces code.
>>>>> With this fix the responsibility of handling key==NULL and/or len==0 is 
>>>>> moved to the PMDs (which is good).
>>>>> With regard to Lijun patch - I liked the approach of overriding the default 
>>>>> testpmd key with the default PMD key.
>>>>> But it only addresses testpmd. More code was added.
>>>>> It seems OK to call rte_eth_dev_rss_hash_conf_get() as part of parsing RSS, 
>>>>> but it would feel more confident if we could confirm it for all the PMDs 
>>>>> (by testing) or at least review the PMDs rss_hash_conf_get() implementations.
>>>>>
>>>>
>>>> Lijun's idea can work. There was a problem in implementation related to the 
>>>> key size assumption, which can be fixed.
>>>>
>>>> Even it is fixed, when user gives a rss rule without a key, we are getting 
>>>> key from device and feeding same key back to device, this is unnecessary I 
>>>> think.
>>>> When user didn't provide a key, rss rule shouldn't touch the key at all.
>>>>
>>> Do you mean that the driver should not configure the key to the hardware when 
>>> the RSS key is not specified?
>>
>> We are talking about 'key' in RSS rte flow rule, not generally configuring the 
>> device for RSS.
>>
> Yes, I know. However, I am talking about some implementations. The configuration 
> interface may be the same as that for configuring RSS
> parameters of the device in general mode.
>> If a specific rss rte flow rule, lets say to filter some defined packets to 
>> some specific queues, don't have 'key' as a part of rule, I am suggesting not 
>> touch 'key' configuration of the device.
>>
> Yes, I agree. I think your point of view is very reasonable and a more
> appropriate implementation. However, for the sake of simplicity and
> other considerations in the architecture design of some devices, the
> hardware may support reasonable hybrid configuration for paramter configuration 
> in the RSS config, rather than independent configuration
> for hw, that is, maybe touch A of rss configuration of the device and
> must touch B of rss configuration. As a result, If the preceding scenario 
> occurs, it is reasonable to configure an RSS config that does not change and 
> specified.What do you think?
>>> If so, we cannot identify when the user does not specify the RSS key to 
>>> determine whether to configure the key.
>>
>> I think we can. If the rule has 'key' attribute, driver can use that key to 
>> program the device, if 'key' attribute is missing, don't do anything related 
>> to the rss key.
>>
> Yes, if 'key' attribute is missing, don't do anything related to the rss key. 
> It's more reasonable.
> But, If he can't do that, I think it is reasonable to configure the default key 
> that already exists on the device. The only disadvantage is wasteful.
>>> In hardware design, most vendors do not configure keys for hardware 
>>> independently, which may be associated with other RSS config parameters.
>>> I think it would be reasonable for us to reconfigure the RSS key with the 
>>> default value configured and valid in the hardware as the default value if 
>>> the user does not specify the RSS key.
>>
>> There are already APIs to update the RSS configuration, like RSS key, hash 
>> functions etc.. They are independent from the rte flow RSS rule.
>> Application should configure RSS according needs using those APIs.
>>
> Yes, should do this. Are there any unreasonable users who simply do not know 
> these APIs or do not want to use them and want to configure some parameters 
> through the rte flow RSS rule without changing other parameters?
>> The question here is about each RSS rte flow rule that can update the key. Am 
>> I missing something?
> I don't think you've misunderstood the general situation.
> Even if the RSS key is not specified in a rule, the driver uses the default key 
> value to re-touch the device.
>>
>>>> Complication was when user provides key_len without a key, I think both 
>>>> ignoring or returning error in this case is OK.
>>> I agree. However, I think it is meaningless to expose the RSS key length to 
>>> users. Do you consider deleting the RSS key length directly?
>>
>> Isn't both 'key' and 'key_len' needed to program the RSS key? Can the driver 
>> know the size of the 'key' without 'key_len'?
>> And driver should verify these provided values.
>>
> I know and agree.It is only during the analysis of the [1] scheme, from the time 
> it was proposed to the time it was withdrawn, that I found some guys in the 
> community questioned the significance of key_len.
> 
> Back to the subject, what is your plan for the solution in the patch?

I think, Ophir's old reverted patch [1] + 'rte_flow_conv_action_conf()' update I 
proposed above can work. Is this working for you? If so can you please send a patch?


[1] https://git.dpdk.org/dpdk/commit/?id=a4391f8bae8
Lijun Ou Oct. 20, 2020, 1:35 p.m. UTC | #18
在 2020/10/20 18:02, Ferruh Yigit 写道:
> On 10/20/2020 10:00 AM, oulijun wrote:
>>
>>
>> 在 2020/10/16 18:57, Ferruh Yigit 写道:
>>> On 10/16/2020 11:04 AM, oulijun wrote:
>>>>
>>>>
>>>> 在 2020/10/16 7:53, Ferruh Yigit 写道:
>>>>> On 10/16/2020 12:21 AM, Ophir Munk wrote:
>>>>>>> From: Ferruh Yigit <ferruh.yigit@intel.com>
>>>>>>>>>> On 10/15/2020 1:41 PM, Lijun Ou wrote:
>>>>>>>>>>> When start the testpmd, the pmd driver initializes the RSS
>>>>>>>>>>> configuration. Generally, the recommended RSS hash key is 
>>>>>>>>>>> used as
>>>>>>>>>>> the default key in the driver. In addition, the default key is
>>>>>>>>>>> different from the default RSS flow in testpmd without 
>>>>>>>>>>> specifying
>>>>>>>>>>> RSS hash key. So. if you do not specify the RSS key when 
>>>>>>>>>>> creating
>>>>>>>>>>> an RSS rule, the testpmd uses the default key as the default RSS
>>>>>>>>>>> key of the RSS rule. As a result, you may mistakenly consider 
>>>>>>>>>>> that
>>>>>>>>>>> the RSS key in use is the valid default key of the NIC, 
>>>>>>>>>>> actually,
>>>>>>>>>>> the key and the valid default key of the NIC are two values.
>>>>>>>>>>>
>>>>>>>>>>> Consider the follow usage with testpmd:
>>>>>>>>>>> 1. first, startup testpmd:
>>>>>>>>>>> testpmd> show port 0 rss-hash key
>>>>>>>>>>> RSS functions:
>>>>>>>>>>>     all ipv4-frag ipv4-other ipv6-frag ipv6-other ip RSS key:
>>>>>>>>>>> -
>>>>>>> 6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA3803
>>>>>>> 0F
>>>>>>>>>>> -20C6A42B73BBEAC01FA
>>>>>>>>>>> 2. create a rss rule
>>>>>>>>>>> testpmd> flow create 0 ingress pattern eth / ipv4 / udp / end
>>>>>>>>>>> testpmd> actions rss \
>>>>>>>>>>> types ipv4-udp end queues end / end
>>>>>>>>>>>
>>>>>>>>>>> 3. show rss-hash key
>>>>>>>>>>> testpmd> show port 0 rss-hash key
>>>>>>>>>>> RSS functions:
>>>>>>>>>>>     all ipv4-udp udp
>>>>>>>>>>> RSS key:
>>>>>>>>>>> -
>>>>>>> 74657374706D6427732064656661756C74205253532068617368206B65792C
>>>>>>> 206F
>>>>>>>>>>> -76657272696465
>>>>>>>>>>>
>>>>>>>>>>> In order to solve the above problems, it use the NIC valid 
>>>>>>>>>>> default
>>>>>>>>>>> RSS key instead of the testpmd dummy RSS key in the flow
>>>>>>>>>>> configuration when the RSS key is not specified in the flow 
>>>>>>>>>>> rule.
>>>>>>>>>>> If the NIC RSS key is invalid, it will use testpmd dummy RSS 
>>>>>>>>>>> key as the
>>>>>>> default key.
>>>>>>>>>>>
>>>>>>>>>>> Fixes: ac8d22de2394 ("ethdev: flatten RSS configuration in flow
>>>>>>>>>>> API")
>>>>>>>>>>> Cc: stable@dpdk.org
>>>>>>>>>>>
>>>>>>>>>>> Signed-off-by: Lijun Ou <oulijun@huawei.com>
>>>>>>>>>>> Reviewed-by: Phil Yang <phil.yang@arm.com>
>>>>>>>>>>> ---
>>>>>>>>>>> V4->V5:
>>>>>>>>>>> -rewrite the commit log
>>>>>>>>>>> -add reviewed-by
>>>>>>>>>>
>>>>>>>>>> Hi Lijun,
>>>>>>>>>>
>>>>>>>>>> There were multiple other comments, it seems they are not 
>>>>>>>>>> addressed
>>>>>>>>>> but only updated the commit log, can you please check comments to
>>>>>>> prev versions.
>>>>>>>>>>
>>>>>>>>>> Before going into the details, my question was what happens if
>>>>>>>>>> default key not provided at all?
>>>>>>>>>> It seems this has been already tried by Ophir [1], later reverted
>>>>>>>>>> back [2] bringing the initial issue back.
>>>>>>>>>>
>>>>>>>>>> According commit, the reason of revert is to support following
>>>>>>> command:
>>>>>>>>>> "flow create 0 <pattern> actions rss queues 0 1 end key_len 40 
>>>>>>>>>> / end"
>>>>>>>>>>
>>>>>>>>>> @Ophir, @Lijun,
>>>>>>>>>> Can we ignore the 'key_len' if the 'key' is not supported and 
>>>>>>>>>> solve
>>>>>>>>>> current issue as initially intended ([1])?
>>>>>>>>>>
>>>>>>>>> Hi, Ferruh
>>>>>>>>>     I have discussed with Phil Yang about the problem in [1]. I 
>>>>>>>>> think
>>>>>>>>> there may be other problems with the idea and there is no better
>>>>>>>>> solution. and we need to remove key_len definition from 
>>>>>>>>> rte_rss_conf
>>>>>>>>> structure. They don't have a plan. And [1] was eventually 
>>>>>>>>> reverted.
>>>>>>>>>
>>>>>>>>
>>>>>>>> Why ignoring 'key_len' (set it to zero) when there is no 'key'
>>>>>>>> provided doesn't work?
>>>>>>>>
>>>>>>>
>>>>>>> What do you think [1] + following update, will it work?
>>>>>>>
>>>>>>>    diff --git a/lib/librte_ethdev/rte_flow.c 
>>>>>>> b/lib/librte_ethdev/rte_flow.c
>>>>>>>    index ee4f3464fe..e7789c87b3 100644
>>>>>>>    --- a/lib/librte_ethdev/rte_flow.c
>>>>>>>    +++ b/lib/librte_ethdev/rte_flow.c
>>>>>>>    @@ -535,7 +535,7 @@ rte_flow_conv_action_conf(void *buf, const 
>>>>>>> size_t
>>>>>>> size,
>>>>>>>                               }),
>>>>>>>                               size > sizeof(*dst.rss) ? 
>>>>>>> sizeof(*dst.rss) : size);
>>>>>>>                    off = sizeof(*dst.rss);
>>>>>>>    -               if (src.rss->key_len) {
>>>>>>>    +               if (src.rss->key_len && src.rss->key) {
>>>>>>>                            off = RTE_ALIGN_CEIL(off, 
>>>>>>> sizeof(*dst.rss->key));
>>>>>>>                            tmp = sizeof(*src.rss->key) * 
>>>>>>> src.rss->key_len;
>>>>>>>                            if (size >= off + tmp)
>>>>>>>
>>>>>>
>>>>>> Ferruh, your suggestion ([1] + update) looks correct. I also 
>>>>>> verified it on mlx5 PMD.
>>>>>> Advantage: it's a generic fix for all dpdk applications using 
>>>>>> rte_flows (not just testpmd).
>>>>>> It reduces code.
>>>>>> With this fix the responsibility of handling key==NULL and/or 
>>>>>> len==0 is moved to the PMDs (which is good).
>>>>>> With regard to Lijun patch - I liked the approach of overriding 
>>>>>> the default testpmd key with the default PMD key.
>>>>>> But it only addresses testpmd. More code was added.
>>>>>> It seems OK to call rte_eth_dev_rss_hash_conf_get() as part of 
>>>>>> parsing RSS, but it would feel more confident if we could confirm 
>>>>>> it for all the PMDs (by testing) or at least review the PMDs 
>>>>>> rss_hash_conf_get() implementations.
>>>>>>
>>>>>
>>>>> Lijun's idea can work. There was a problem in implementation 
>>>>> related to the key size assumption, which can be fixed.
>>>>>
>>>>> Even it is fixed, when user gives a rss rule without a key, we are 
>>>>> getting key from device and feeding same key back to device, this 
>>>>> is unnecessary I think.
>>>>> When user didn't provide a key, rss rule shouldn't touch the key at 
>>>>> all.
>>>>>
>>>> Do you mean that the driver should not configure the key to the 
>>>> hardware when the RSS key is not specified?
>>>
>>> We are talking about 'key' in RSS rte flow rule, not generally 
>>> configuring the device for RSS.
>>>
>> Yes, I know. However, I am talking about some implementations. The 
>> configuration interface may be the same as that for configuring RSS
>> parameters of the device in general mode.
>>> If a specific rss rte flow rule, lets say to filter some defined 
>>> packets to some specific queues, don't have 'key' as a part of rule, 
>>> I am suggesting not touch 'key' configuration of the device.
>>>
>> Yes, I agree. I think your point of view is very reasonable and a more
>> appropriate implementation. However, for the sake of simplicity and
>> other considerations in the architecture design of some devices, the
>> hardware may support reasonable hybrid configuration for paramter 
>> configuration in the RSS config, rather than independent configuration
>> for hw, that is, maybe touch A of rss configuration of the device and
>> must touch B of rss configuration. As a result, If the preceding 
>> scenario occurs, it is reasonable to configure an RSS config that does 
>> not change and specified.What do you think?
>>>> If so, we cannot identify when the user does not specify the RSS key 
>>>> to determine whether to configure the key.
>>>
>>> I think we can. If the rule has 'key' attribute, driver can use that 
>>> key to program the device, if 'key' attribute is missing, don't do 
>>> anything related to the rss key.
>>>
>> Yes, if 'key' attribute is missing, don't do anything related to the 
>> rss key. It's more reasonable.
>> But, If he can't do that, I think it is reasonable to configure the 
>> default key that already exists on the device. The only disadvantage 
>> is wasteful.
>>>> In hardware design, most vendors do not configure keys for hardware 
>>>> independently, which may be associated with other RSS config 
>>>> parameters.
>>>> I think it would be reasonable for us to reconfigure the RSS key 
>>>> with the default value configured and valid in the hardware as the 
>>>> default value if the user does not specify the RSS key.
>>>
>>> There are already APIs to update the RSS configuration, like RSS key, 
>>> hash functions etc.. They are independent from the rte flow RSS rule.
>>> Application should configure RSS according needs using those APIs.
>>>
>> Yes, should do this. Are there any unreasonable users who simply do 
>> not know these APIs or do not want to use them and want to configure 
>> some parameters through the rte flow RSS rule without changing other 
>> parameters?
>>> The question here is about each RSS rte flow rule that can update the 
>>> key. Am I missing something?
>> I don't think you've misunderstood the general situation.
>> Even if the RSS key is not specified in a rule, the driver uses the 
>> default key value to re-touch the device.
>>>
>>>>> Complication was when user provides key_len without a key, I think 
>>>>> both ignoring or returning error in this case is OK.
>>>> I agree. However, I think it is meaningless to expose the RSS key 
>>>> length to users. Do you consider deleting the RSS key length directly?
>>>
>>> Isn't both 'key' and 'key_len' needed to program the RSS key? Can the 
>>> driver know the size of the 'key' without 'key_len'?
>>> And driver should verify these provided values.
>>>
>> I know and agree.It is only during the analysis of the [1] scheme, 
>> from the time it was proposed to the time it was withdrawn, that I 
>> found some guys in the community questioned the significance of key_len.
>>
>> Back to the subject, what is your plan for the solution in the patch?
> 
> I think, Ophir's old reverted patch [1] + 'rte_flow_conv_action_conf()' 
> update I proposed above can work. Is this working for you? If so can you 
> please send a patch?
> 
> 
> [1] https://git.dpdk.org/dpdk/commit/?id=a4391f8bae8
> .
Hi, Ferruh
    I've just used [1] + 'rte_flow_conv_action_conf()' for testing based 
on the Kunpeng920 NIC plation and use hns3 pmd driver. it can be resolve 
the RSS key question.
    testpmd> show port 0 rss-hash key
RSS functions:
  all ipv4-frag ipv4-other ipv6-frag ipv6-other ip
RSS key:
6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA38030F20C6A42B73BBEAC01FA
testpmd> flow create 0 ingress pattern end actions rss func 
symmetric_toeplitz types all end / end
0000:7d:00.0 hns3_config_rss_filter(): modified RSS types based on 
hardware support, requested:7f83fffc configured:3ef8
0000:7d:00.0 hns3_parse_rss_key(): no valid RSS key config, using default
Flow rule #0 created
testpmd> show port 0 rss-hash key
RSS functions:
  all ipv4-frag ipv4-tcp ipv4-udp ipv4-sctp ipv4-other ipv6-frag 
ipv6-tcp ipv6-udp ipv6-sctp ipv6-other ip udp tcp sctp
RSS key:
6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA38030F20C6A42B73BBEAC01FA

However, I also tested the case. I am not specified the RSS key and 
specified the key_len when create a RSS flow rule, it can create success.
testpmd> flow create 0 ingress pattern end actions rss queues 0 1 end 
key_len 40 / end
0000:7d:00.0 hns3_config_rss_filter(): modified RSS types based on 
hardware support, requested:a38c configured:2288
0000:7d:00.0 hns3_parse_rss_key(): no valid RSS key config, using default
Flow rule #0 created
testpmd> show port 0 rss-hash key
RSS functions:
  all ipv4-frag ipv4-other ipv6-frag ipv6-other ip
RSS key:
6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA38030F20C6A42B73BBEAC01FA

>
Ferruh Yigit Oct. 20, 2020, 2:34 p.m. UTC | #19
On 10/20/2020 2:35 PM, oulijun wrote:
> 
> 
> 在 2020/10/20 18:02, Ferruh Yigit 写道:
>> On 10/20/2020 10:00 AM, oulijun wrote:
>>>
>>>
>>> 在 2020/10/16 18:57, Ferruh Yigit 写道:
>>>> On 10/16/2020 11:04 AM, oulijun wrote:
>>>>>
>>>>>
>>>>> 在 2020/10/16 7:53, Ferruh Yigit 写道:
>>>>>> On 10/16/2020 12:21 AM, Ophir Munk wrote:
>>>>>>>> From: Ferruh Yigit <ferruh.yigit@intel.com>
>>>>>>>>>>> On 10/15/2020 1:41 PM, Lijun Ou wrote:
>>>>>>>>>>>> When start the testpmd, the pmd driver initializes the RSS
>>>>>>>>>>>> configuration. Generally, the recommended RSS hash key is used as
>>>>>>>>>>>> the default key in the driver. In addition, the default key is
>>>>>>>>>>>> different from the default RSS flow in testpmd without specifying
>>>>>>>>>>>> RSS hash key. So. if you do not specify the RSS key when creating
>>>>>>>>>>>> an RSS rule, the testpmd uses the default key as the default RSS
>>>>>>>>>>>> key of the RSS rule. As a result, you may mistakenly consider that
>>>>>>>>>>>> the RSS key in use is the valid default key of the NIC, actually,
>>>>>>>>>>>> the key and the valid default key of the NIC are two values.
>>>>>>>>>>>>
>>>>>>>>>>>> Consider the follow usage with testpmd:
>>>>>>>>>>>> 1. first, startup testpmd:
>>>>>>>>>>>> testpmd> show port 0 rss-hash key
>>>>>>>>>>>> RSS functions:
>>>>>>>>>>>>     all ipv4-frag ipv4-other ipv6-frag ipv6-other ip RSS key:
>>>>>>>>>>>> -
>>>>>>>> 6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA3803
>>>>>>>> 0F
>>>>>>>>>>>> -20C6A42B73BBEAC01FA
>>>>>>>>>>>> 2. create a rss rule
>>>>>>>>>>>> testpmd> flow create 0 ingress pattern eth / ipv4 / udp / end
>>>>>>>>>>>> testpmd> actions rss \
>>>>>>>>>>>> types ipv4-udp end queues end / end
>>>>>>>>>>>>
>>>>>>>>>>>> 3. show rss-hash key
>>>>>>>>>>>> testpmd> show port 0 rss-hash key
>>>>>>>>>>>> RSS functions:
>>>>>>>>>>>>     all ipv4-udp udp
>>>>>>>>>>>> RSS key:
>>>>>>>>>>>> -
>>>>>>>> 74657374706D6427732064656661756C74205253532068617368206B65792C
>>>>>>>> 206F
>>>>>>>>>>>> -76657272696465
>>>>>>>>>>>>
>>>>>>>>>>>> In order to solve the above problems, it use the NIC valid default
>>>>>>>>>>>> RSS key instead of the testpmd dummy RSS key in the flow
>>>>>>>>>>>> configuration when the RSS key is not specified in the flow rule.
>>>>>>>>>>>> If the NIC RSS key is invalid, it will use testpmd dummy RSS key as the
>>>>>>>> default key.
>>>>>>>>>>>>
>>>>>>>>>>>> Fixes: ac8d22de2394 ("ethdev: flatten RSS configuration in flow
>>>>>>>>>>>> API")
>>>>>>>>>>>> Cc: stable@dpdk.org
>>>>>>>>>>>>
>>>>>>>>>>>> Signed-off-by: Lijun Ou <oulijun@huawei.com>
>>>>>>>>>>>> Reviewed-by: Phil Yang <phil.yang@arm.com>
>>>>>>>>>>>> ---
>>>>>>>>>>>> V4->V5:
>>>>>>>>>>>> -rewrite the commit log
>>>>>>>>>>>> -add reviewed-by
>>>>>>>>>>>
>>>>>>>>>>> Hi Lijun,
>>>>>>>>>>>
>>>>>>>>>>> There were multiple other comments, it seems they are not addressed
>>>>>>>>>>> but only updated the commit log, can you please check comments to
>>>>>>>> prev versions.
>>>>>>>>>>>
>>>>>>>>>>> Before going into the details, my question was what happens if
>>>>>>>>>>> default key not provided at all?
>>>>>>>>>>> It seems this has been already tried by Ophir [1], later reverted
>>>>>>>>>>> back [2] bringing the initial issue back.
>>>>>>>>>>>
>>>>>>>>>>> According commit, the reason of revert is to support following
>>>>>>>> command:
>>>>>>>>>>> "flow create 0 <pattern> actions rss queues 0 1 end key_len 40 / end"
>>>>>>>>>>>
>>>>>>>>>>> @Ophir, @Lijun,
>>>>>>>>>>> Can we ignore the 'key_len' if the 'key' is not supported and solve
>>>>>>>>>>> current issue as initially intended ([1])?
>>>>>>>>>>>
>>>>>>>>>> Hi, Ferruh
>>>>>>>>>>     I have discussed with Phil Yang about the problem in [1]. I think
>>>>>>>>>> there may be other problems with the idea and there is no better
>>>>>>>>>> solution. and we need to remove key_len definition from rte_rss_conf
>>>>>>>>>> structure. They don't have a plan. And [1] was eventually reverted.
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>> Why ignoring 'key_len' (set it to zero) when there is no 'key'
>>>>>>>>> provided doesn't work?
>>>>>>>>>
>>>>>>>>
>>>>>>>> What do you think [1] + following update, will it work?
>>>>>>>>
>>>>>>>>    diff --git a/lib/librte_ethdev/rte_flow.c b/lib/librte_ethdev/rte_flow.c
>>>>>>>>    index ee4f3464fe..e7789c87b3 100644
>>>>>>>>    --- a/lib/librte_ethdev/rte_flow.c
>>>>>>>>    +++ b/lib/librte_ethdev/rte_flow.c
>>>>>>>>    @@ -535,7 +535,7 @@ rte_flow_conv_action_conf(void *buf, const size_t
>>>>>>>> size,
>>>>>>>>                               }),
>>>>>>>>                               size > sizeof(*dst.rss) ? sizeof(*dst.rss) 
>>>>>>>> : size);
>>>>>>>>                    off = sizeof(*dst.rss);
>>>>>>>>    -               if (src.rss->key_len) {
>>>>>>>>    +               if (src.rss->key_len && src.rss->key) {
>>>>>>>>                            off = RTE_ALIGN_CEIL(off, 
>>>>>>>> sizeof(*dst.rss->key));
>>>>>>>>                            tmp = sizeof(*src.rss->key) * src.rss->key_len;
>>>>>>>>                            if (size >= off + tmp)
>>>>>>>>
>>>>>>>
>>>>>>> Ferruh, your suggestion ([1] + update) looks correct. I also verified it 
>>>>>>> on mlx5 PMD.
>>>>>>> Advantage: it's a generic fix for all dpdk applications using rte_flows 
>>>>>>> (not just testpmd).
>>>>>>> It reduces code.
>>>>>>> With this fix the responsibility of handling key==NULL and/or len==0 is 
>>>>>>> moved to the PMDs (which is good).
>>>>>>> With regard to Lijun patch - I liked the approach of overriding the 
>>>>>>> default testpmd key with the default PMD key.
>>>>>>> But it only addresses testpmd. More code was added.
>>>>>>> It seems OK to call rte_eth_dev_rss_hash_conf_get() as part of parsing 
>>>>>>> RSS, but it would feel more confident if we could confirm it for all the 
>>>>>>> PMDs (by testing) or at least review the PMDs rss_hash_conf_get() 
>>>>>>> implementations.
>>>>>>>
>>>>>>
>>>>>> Lijun's idea can work. There was a problem in implementation related to 
>>>>>> the key size assumption, which can be fixed.
>>>>>>
>>>>>> Even it is fixed, when user gives a rss rule without a key, we are getting 
>>>>>> key from device and feeding same key back to device, this is unnecessary I 
>>>>>> think.
>>>>>> When user didn't provide a key, rss rule shouldn't touch the key at all.
>>>>>>
>>>>> Do you mean that the driver should not configure the key to the hardware 
>>>>> when the RSS key is not specified?
>>>>
>>>> We are talking about 'key' in RSS rte flow rule, not generally configuring 
>>>> the device for RSS.
>>>>
>>> Yes, I know. However, I am talking about some implementations. The 
>>> configuration interface may be the same as that for configuring RSS
>>> parameters of the device in general mode.
>>>> If a specific rss rte flow rule, lets say to filter some defined packets to 
>>>> some specific queues, don't have 'key' as a part of rule, I am suggesting 
>>>> not touch 'key' configuration of the device.
>>>>
>>> Yes, I agree. I think your point of view is very reasonable and a more
>>> appropriate implementation. However, for the sake of simplicity and
>>> other considerations in the architecture design of some devices, the
>>> hardware may support reasonable hybrid configuration for paramter 
>>> configuration in the RSS config, rather than independent configuration
>>> for hw, that is, maybe touch A of rss configuration of the device and
>>> must touch B of rss configuration. As a result, If the preceding scenario 
>>> occurs, it is reasonable to configure an RSS config that does not change and 
>>> specified.What do you think?
>>>>> If so, we cannot identify when the user does not specify the RSS key to 
>>>>> determine whether to configure the key.
>>>>
>>>> I think we can. If the rule has 'key' attribute, driver can use that key to 
>>>> program the device, if 'key' attribute is missing, don't do anything related 
>>>> to the rss key.
>>>>
>>> Yes, if 'key' attribute is missing, don't do anything related to the rss key. 
>>> It's more reasonable.
>>> But, If he can't do that, I think it is reasonable to configure the default 
>>> key that already exists on the device. The only disadvantage is wasteful.
>>>>> In hardware design, most vendors do not configure keys for hardware 
>>>>> independently, which may be associated with other RSS config parameters.
>>>>> I think it would be reasonable for us to reconfigure the RSS key with the 
>>>>> default value configured and valid in the hardware as the default value if 
>>>>> the user does not specify the RSS key.
>>>>
>>>> There are already APIs to update the RSS configuration, like RSS key, hash 
>>>> functions etc.. They are independent from the rte flow RSS rule.
>>>> Application should configure RSS according needs using those APIs.
>>>>
>>> Yes, should do this. Are there any unreasonable users who simply do not know 
>>> these APIs or do not want to use them and want to configure some parameters 
>>> through the rte flow RSS rule without changing other parameters?
>>>> The question here is about each RSS rte flow rule that can update the key. 
>>>> Am I missing something?
>>> I don't think you've misunderstood the general situation.
>>> Even if the RSS key is not specified in a rule, the driver uses the default 
>>> key value to re-touch the device.
>>>>
>>>>>> Complication was when user provides key_len without a key, I think both 
>>>>>> ignoring or returning error in this case is OK.
>>>>> I agree. However, I think it is meaningless to expose the RSS key length to 
>>>>> users. Do you consider deleting the RSS key length directly?
>>>>
>>>> Isn't both 'key' and 'key_len' needed to program the RSS key? Can the driver 
>>>> know the size of the 'key' without 'key_len'?
>>>> And driver should verify these provided values.
>>>>
>>> I know and agree.It is only during the analysis of the [1] scheme, from the 
>>> time it was proposed to the time it was withdrawn, that I found some guys in 
>>> the community questioned the significance of key_len.
>>>
>>> Back to the subject, what is your plan for the solution in the patch?
>>
>> I think, Ophir's old reverted patch [1] + 'rte_flow_conv_action_conf()' update 
>> I proposed above can work. Is this working for you? If so can you please send 
>> a patch?
>>
>>
>> [1] https://git.dpdk.org/dpdk/commit/?id=a4391f8bae8
>> .
> Hi, Ferruh
>     I've just used [1] + 'rte_flow_conv_action_conf()' for testing based on the 
> Kunpeng920 NIC plation and use hns3 pmd driver. it can be resolve the RSS key 
> question.
>     testpmd> show port 0 rss-hash key
> RSS functions:
>   all ipv4-frag ipv4-other ipv6-frag ipv6-other ip
> RSS key:
> 6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA38030F20C6A42B73BBEAC01FA
> testpmd> flow create 0 ingress pattern end actions rss func symmetric_toeplitz 
> types all end / end
> 0000:7d:00.0 hns3_config_rss_filter(): modified RSS types based on hardware 
> support, requested:7f83fffc configured:3ef8
> 0000:7d:00.0 hns3_parse_rss_key(): no valid RSS key config, using default
> Flow rule #0 created
> testpmd> show port 0 rss-hash key
> RSS functions:
>   all ipv4-frag ipv4-tcp ipv4-udp ipv4-sctp ipv4-other ipv6-frag ipv6-tcp 
> ipv6-udp ipv6-sctp ipv6-other ip udp tcp sctp
> RSS key:
> 6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA38030F20C6A42B73BBEAC01FA
> 
> However, I also tested the case. I am not specified the RSS key and specified 
> the key_len when create a RSS flow rule, it can create success.
> testpmd> flow create 0 ingress pattern end actions rss queues 0 1 end key_len 40 
> / end
> 0000:7d:00.0 hns3_config_rss_filter(): modified RSS types based on hardware 
> support, requested:a38c configured:2288
> 0000:7d:00.0 hns3_parse_rss_key(): no valid RSS key config, using default
> Flow rule #0 created
> testpmd> show port 0 rss-hash key
> RSS functions:
>   all ipv4-frag ipv4-other ipv6-frag ipv6-other ip
> RSS key:
> 6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA38030F20C6A42B73BBEAC01FA
> 

So are you happy with this solution?
Lijun Ou Oct. 21, 2020, 8:19 a.m. UTC | #20
在 2020/10/20 22:34, Ferruh Yigit 写道:
> On 10/20/2020 2:35 PM, oulijun wrote:
>>
>>
>> 在 2020/10/20 18:02, Ferruh Yigit 写道:
>>> On 10/20/2020 10:00 AM, oulijun wrote:
>>>>
>>>>
>>>> 在 2020/10/16 18:57, Ferruh Yigit 写道:
>>>>> On 10/16/2020 11:04 AM, oulijun wrote:
>>>>>>
>>>>>>
>>>>>> 在 2020/10/16 7:53, Ferruh Yigit 写道:
>>>>>>> On 10/16/2020 12:21 AM, Ophir Munk wrote:
>>>>>>>>> From: Ferruh Yigit <ferruh.yigit@intel.com>
>>>>>>>>>>>> On 10/15/2020 1:41 PM, Lijun Ou wrote:
>>>>>>>>>>>>> When start the testpmd, the pmd driver initializes the RSS
>>>>>>>>>>>>> configuration. Generally, the recommended RSS hash key is 
>>>>>>>>>>>>> used as
>>>>>>>>>>>>> the default key in the driver. In addition, the default key is
>>>>>>>>>>>>> different from the default RSS flow in testpmd without 
>>>>>>>>>>>>> specifying
>>>>>>>>>>>>> RSS hash key. So. if you do not specify the RSS key when 
>>>>>>>>>>>>> creating
>>>>>>>>>>>>> an RSS rule, the testpmd uses the default key as the 
>>>>>>>>>>>>> default RSS
>>>>>>>>>>>>> key of the RSS rule. As a result, you may mistakenly 
>>>>>>>>>>>>> consider that
>>>>>>>>>>>>> the RSS key in use is the valid default key of the NIC, 
>>>>>>>>>>>>> actually,
>>>>>>>>>>>>> the key and the valid default key of the NIC are two values.
>>>>>>>>>>>>>
>>>>>>>>>>>>> Consider the follow usage with testpmd:
>>>>>>>>>>>>> 1. first, startup testpmd:
>>>>>>>>>>>>> testpmd> show port 0 rss-hash key
>>>>>>>>>>>>> RSS functions:
>>>>>>>>>>>>>     all ipv4-frag ipv4-other ipv6-frag ipv6-other ip RSS key:
>>>>>>>>>>>>> -
>>>>>>>>> 6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA3803
>>>>>>>>> 0F
>>>>>>>>>>>>> -20C6A42B73BBEAC01FA
>>>>>>>>>>>>> 2. create a rss rule
>>>>>>>>>>>>> testpmd> flow create 0 ingress pattern eth / ipv4 / udp / end
>>>>>>>>>>>>> testpmd> actions rss \
>>>>>>>>>>>>> types ipv4-udp end queues end / end
>>>>>>>>>>>>>
>>>>>>>>>>>>> 3. show rss-hash key
>>>>>>>>>>>>> testpmd> show port 0 rss-hash key
>>>>>>>>>>>>> RSS functions:
>>>>>>>>>>>>>     all ipv4-udp udp
>>>>>>>>>>>>> RSS key:
>>>>>>>>>>>>> -
>>>>>>>>> 74657374706D6427732064656661756C74205253532068617368206B65792C
>>>>>>>>> 206F
>>>>>>>>>>>>> -76657272696465
>>>>>>>>>>>>>
>>>>>>>>>>>>> In order to solve the above problems, it use the NIC valid 
>>>>>>>>>>>>> default
>>>>>>>>>>>>> RSS key instead of the testpmd dummy RSS key in the flow
>>>>>>>>>>>>> configuration when the RSS key is not specified in the flow 
>>>>>>>>>>>>> rule.
>>>>>>>>>>>>> If the NIC RSS key is invalid, it will use testpmd dummy 
>>>>>>>>>>>>> RSS key as the
>>>>>>>>> default key.
>>>>>>>>>>>>>
>>>>>>>>>>>>> Fixes: ac8d22de2394 ("ethdev: flatten RSS configuration in 
>>>>>>>>>>>>> flow
>>>>>>>>>>>>> API")
>>>>>>>>>>>>> Cc: stable@dpdk.org
>>>>>>>>>>>>>
>>>>>>>>>>>>> Signed-off-by: Lijun Ou <oulijun@huawei.com>
>>>>>>>>>>>>> Reviewed-by: Phil Yang <phil.yang@arm.com>
>>>>>>>>>>>>> ---
>>>>>>>>>>>>> V4->V5:
>>>>>>>>>>>>> -rewrite the commit log
>>>>>>>>>>>>> -add reviewed-by
>>>>>>>>>>>>
>>>>>>>>>>>> Hi Lijun,
>>>>>>>>>>>>
>>>>>>>>>>>> There were multiple other comments, it seems they are not 
>>>>>>>>>>>> addressed
>>>>>>>>>>>> but only updated the commit log, can you please check 
>>>>>>>>>>>> comments to
>>>>>>>>> prev versions.
>>>>>>>>>>>>
>>>>>>>>>>>> Before going into the details, my question was what happens if
>>>>>>>>>>>> default key not provided at all?
>>>>>>>>>>>> It seems this has been already tried by Ophir [1], later 
>>>>>>>>>>>> reverted
>>>>>>>>>>>> back [2] bringing the initial issue back.
>>>>>>>>>>>>
>>>>>>>>>>>> According commit, the reason of revert is to support following
>>>>>>>>> command:
>>>>>>>>>>>> "flow create 0 <pattern> actions rss queues 0 1 end key_len 
>>>>>>>>>>>> 40 / end"
>>>>>>>>>>>>
>>>>>>>>>>>> @Ophir, @Lijun,
>>>>>>>>>>>> Can we ignore the 'key_len' if the 'key' is not supported 
>>>>>>>>>>>> and solve
>>>>>>>>>>>> current issue as initially intended ([1])?
>>>>>>>>>>>>
>>>>>>>>>>> Hi, Ferruh
>>>>>>>>>>>     I have discussed with Phil Yang about the problem in [1]. 
>>>>>>>>>>> I think
>>>>>>>>>>> there may be other problems with the idea and there is no better
>>>>>>>>>>> solution. and we need to remove key_len definition from 
>>>>>>>>>>> rte_rss_conf
>>>>>>>>>>> structure. They don't have a plan. And [1] was eventually 
>>>>>>>>>>> reverted.
>>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> Why ignoring 'key_len' (set it to zero) when there is no 'key'
>>>>>>>>>> provided doesn't work?
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>> What do you think [1] + following update, will it work?
>>>>>>>>>
>>>>>>>>>    diff --git a/lib/librte_ethdev/rte_flow.c 
>>>>>>>>> b/lib/librte_ethdev/rte_flow.c
>>>>>>>>>    index ee4f3464fe..e7789c87b3 100644
>>>>>>>>>    --- a/lib/librte_ethdev/rte_flow.c
>>>>>>>>>    +++ b/lib/librte_ethdev/rte_flow.c
>>>>>>>>>    @@ -535,7 +535,7 @@ rte_flow_conv_action_conf(void *buf, 
>>>>>>>>> const size_t
>>>>>>>>> size,
>>>>>>>>>                               }),
>>>>>>>>>                               size > sizeof(*dst.rss) ? 
>>>>>>>>> sizeof(*dst.rss) : size);
>>>>>>>>>                    off = sizeof(*dst.rss);
>>>>>>>>>    -               if (src.rss->key_len) {
>>>>>>>>>    +               if (src.rss->key_len && src.rss->key) {
>>>>>>>>>                            off = RTE_ALIGN_CEIL(off, 
>>>>>>>>> sizeof(*dst.rss->key));
>>>>>>>>>                            tmp = sizeof(*src.rss->key) * 
>>>>>>>>> src.rss->key_len;
>>>>>>>>>                            if (size >= off + tmp)
>>>>>>>>>
>>>>>>>>
>>>>>>>> Ferruh, your suggestion ([1] + update) looks correct. I also 
>>>>>>>> verified it on mlx5 PMD.
>>>>>>>> Advantage: it's a generic fix for all dpdk applications using 
>>>>>>>> rte_flows (not just testpmd).
>>>>>>>> It reduces code.
>>>>>>>> With this fix the responsibility of handling key==NULL and/or 
>>>>>>>> len==0 is moved to the PMDs (which is good).
>>>>>>>> With regard to Lijun patch - I liked the approach of overriding 
>>>>>>>> the default testpmd key with the default PMD key.
>>>>>>>> But it only addresses testpmd. More code was added.
>>>>>>>> It seems OK to call rte_eth_dev_rss_hash_conf_get() as part of 
>>>>>>>> parsing RSS, but it would feel more confident if we could 
>>>>>>>> confirm it for all the PMDs (by testing) or at least review the 
>>>>>>>> PMDs rss_hash_conf_get() implementations.
>>>>>>>>
>>>>>>>
>>>>>>> Lijun's idea can work. There was a problem in implementation 
>>>>>>> related to the key size assumption, which can be fixed.
>>>>>>>
>>>>>>> Even it is fixed, when user gives a rss rule without a key, we 
>>>>>>> are getting key from device and feeding same key back to device, 
>>>>>>> this is unnecessary I think.
>>>>>>> When user didn't provide a key, rss rule shouldn't touch the key 
>>>>>>> at all.
>>>>>>>
>>>>>> Do you mean that the driver should not configure the key to the 
>>>>>> hardware when the RSS key is not specified?
>>>>>
>>>>> We are talking about 'key' in RSS rte flow rule, not generally 
>>>>> configuring the device for RSS.
>>>>>
>>>> Yes, I know. However, I am talking about some implementations. The 
>>>> configuration interface may be the same as that for configuring RSS
>>>> parameters of the device in general mode.
>>>>> If a specific rss rte flow rule, lets say to filter some defined 
>>>>> packets to some specific queues, don't have 'key' as a part of 
>>>>> rule, I am suggesting not touch 'key' configuration of the device.
>>>>>
>>>> Yes, I agree. I think your point of view is very reasonable and a more
>>>> appropriate implementation. However, for the sake of simplicity and
>>>> other considerations in the architecture design of some devices, the
>>>> hardware may support reasonable hybrid configuration for paramter 
>>>> configuration in the RSS config, rather than independent configuration
>>>> for hw, that is, maybe touch A of rss configuration of the device and
>>>> must touch B of rss configuration. As a result, If the preceding 
>>>> scenario occurs, it is reasonable to configure an RSS config that 
>>>> does not change and specified.What do you think?
>>>>>> If so, we cannot identify when the user does not specify the RSS 
>>>>>> key to determine whether to configure the key.
>>>>>
>>>>> I think we can. If the rule has 'key' attribute, driver can use 
>>>>> that key to program the device, if 'key' attribute is missing, 
>>>>> don't do anything related to the rss key.
>>>>>
>>>> Yes, if 'key' attribute is missing, don't do anything related to the 
>>>> rss key. It's more reasonable.
>>>> But, If he can't do that, I think it is reasonable to configure the 
>>>> default key that already exists on the device. The only disadvantage 
>>>> is wasteful.
>>>>>> In hardware design, most vendors do not configure keys for 
>>>>>> hardware independently, which may be associated with other RSS 
>>>>>> config parameters.
>>>>>> I think it would be reasonable for us to reconfigure the RSS key 
>>>>>> with the default value configured and valid in the hardware as the 
>>>>>> default value if the user does not specify the RSS key.
>>>>>
>>>>> There are already APIs to update the RSS configuration, like RSS 
>>>>> key, hash functions etc.. They are independent from the rte flow 
>>>>> RSS rule.
>>>>> Application should configure RSS according needs using those APIs.
>>>>>
>>>> Yes, should do this. Are there any unreasonable users who simply do 
>>>> not know these APIs or do not want to use them and want to configure 
>>>> some parameters through the rte flow RSS rule without changing other 
>>>> parameters?
>>>>> The question here is about each RSS rte flow rule that can update 
>>>>> the key. Am I missing something?
>>>> I don't think you've misunderstood the general situation.
>>>> Even if the RSS key is not specified in a rule, the driver uses the 
>>>> default key value to re-touch the device.
>>>>>
>>>>>>> Complication was when user provides key_len without a key, I 
>>>>>>> think both ignoring or returning error in this case is OK.
>>>>>> I agree. However, I think it is meaningless to expose the RSS key 
>>>>>> length to users. Do you consider deleting the RSS key length 
>>>>>> directly?
>>>>>
>>>>> Isn't both 'key' and 'key_len' needed to program the RSS key? Can 
>>>>> the driver know the size of the 'key' without 'key_len'?
>>>>> And driver should verify these provided values.
>>>>>
>>>> I know and agree.It is only during the analysis of the [1] scheme, 
>>>> from the time it was proposed to the time it was withdrawn, that I 
>>>> found some guys in the community questioned the significance of 
>>>> key_len.
>>>>
>>>> Back to the subject, what is your plan for the solution in the patch?
>>>
>>> I think, Ophir's old reverted patch [1] + 
>>> 'rte_flow_conv_action_conf()' update I proposed above can work. Is 
>>> this working for you? If so can you please send a patch?
>>>
>>>
>>> [1] https://git.dpdk.org/dpdk/commit/?id=a4391f8bae8
>>> .
>> Hi, Ferruh
>>     I've just used [1] + 'rte_flow_conv_action_conf()' for testing 
>> based on the Kunpeng920 NIC plation and use hns3 pmd driver. it can be 
>> resolve the RSS key question.
>>     testpmd> show port 0 rss-hash key
>> RSS functions:
>>   all ipv4-frag ipv4-other ipv6-frag ipv6-other ip
>> RSS key:
>> 6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA38030F20C6A42B73BBEAC01FA 
>>
>> testpmd> flow create 0 ingress pattern end actions rss func 
>> symmetric_toeplitz types all end / end
>> 0000:7d:00.0 hns3_config_rss_filter(): modified RSS types based on 
>> hardware support, requested:7f83fffc configured:3ef8
>> 0000:7d:00.0 hns3_parse_rss_key(): no valid RSS key config, using default
>> Flow rule #0 created
>> testpmd> show port 0 rss-hash key
>> RSS functions:
>>   all ipv4-frag ipv4-tcp ipv4-udp ipv4-sctp ipv4-other ipv6-frag 
>> ipv6-tcp ipv6-udp ipv6-sctp ipv6-other ip udp tcp sctp
>> RSS key:
>> 6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA38030F20C6A42B73BBEAC01FA 
>>
>>
>> However, I also tested the case. I am not specified the RSS key and 
>> specified the key_len when create a RSS flow rule, it can create success.
>> testpmd> flow create 0 ingress pattern end actions rss queues 0 1 end 
>> key_len 40 / end
>> 0000:7d:00.0 hns3_config_rss_filter(): modified RSS types based on 
>> hardware support, requested:a38c configured:2288
>> 0000:7d:00.0 hns3_parse_rss_key(): no valid RSS key config, using default
>> Flow rule #0 created
>> testpmd> show port 0 rss-hash key
>> RSS functions:
>>   all ipv4-frag ipv4-other ipv6-frag ipv6-other ip
>> RSS key:
>> 6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA38030F20C6A42B73BBEAC01FA 
>>
>>
> 
> So are you happy with this solution?
> .
Yes, thanks.
>
Ferruh Yigit Oct. 21, 2020, 9:38 a.m. UTC | #21
On 10/21/2020 9:19 AM, oulijun wrote:
> 
> 
> 在 2020/10/20 22:34, Ferruh Yigit 写道:
>> On 10/20/2020 2:35 PM, oulijun wrote:
>>>
>>>
>>> 在 2020/10/20 18:02, Ferruh Yigit 写道:
>>>> On 10/20/2020 10:00 AM, oulijun wrote:
>>>>>
>>>>>
>>>>> 在 2020/10/16 18:57, Ferruh Yigit 写道:
>>>>>> On 10/16/2020 11:04 AM, oulijun wrote:
>>>>>>>
>>>>>>>
>>>>>>> 在 2020/10/16 7:53, Ferruh Yigit 写道:
>>>>>>>> On 10/16/2020 12:21 AM, Ophir Munk wrote:
>>>>>>>>>> From: Ferruh Yigit <ferruh.yigit@intel.com>
>>>>>>>>>>>>> On 10/15/2020 1:41 PM, Lijun Ou wrote:
>>>>>>>>>>>>>> When start the testpmd, the pmd driver initializes the RSS
>>>>>>>>>>>>>> configuration. Generally, the recommended RSS hash key is used as
>>>>>>>>>>>>>> the default key in the driver. In addition, the default key is
>>>>>>>>>>>>>> different from the default RSS flow in testpmd without specifying
>>>>>>>>>>>>>> RSS hash key. So. if you do not specify the RSS key when creating
>>>>>>>>>>>>>> an RSS rule, the testpmd uses the default key as the default RSS
>>>>>>>>>>>>>> key of the RSS rule. As a result, you may mistakenly consider that
>>>>>>>>>>>>>> the RSS key in use is the valid default key of the NIC, actually,
>>>>>>>>>>>>>> the key and the valid default key of the NIC are two values.
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> Consider the follow usage with testpmd:
>>>>>>>>>>>>>> 1. first, startup testpmd:
>>>>>>>>>>>>>> testpmd> show port 0 rss-hash key
>>>>>>>>>>>>>> RSS functions:
>>>>>>>>>>>>>>     all ipv4-frag ipv4-other ipv6-frag ipv6-other ip RSS key:
>>>>>>>>>>>>>> -
>>>>>>>>>> 6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA3803
>>>>>>>>>> 0F
>>>>>>>>>>>>>> -20C6A42B73BBEAC01FA
>>>>>>>>>>>>>> 2. create a rss rule
>>>>>>>>>>>>>> testpmd> flow create 0 ingress pattern eth / ipv4 / udp / end
>>>>>>>>>>>>>> testpmd> actions rss \
>>>>>>>>>>>>>> types ipv4-udp end queues end / end
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> 3. show rss-hash key
>>>>>>>>>>>>>> testpmd> show port 0 rss-hash key
>>>>>>>>>>>>>> RSS functions:
>>>>>>>>>>>>>>     all ipv4-udp udp
>>>>>>>>>>>>>> RSS key:
>>>>>>>>>>>>>> -
>>>>>>>>>> 74657374706D6427732064656661756C74205253532068617368206B65792C
>>>>>>>>>> 206F
>>>>>>>>>>>>>> -76657272696465
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> In order to solve the above problems, it use the NIC valid default
>>>>>>>>>>>>>> RSS key instead of the testpmd dummy RSS key in the flow
>>>>>>>>>>>>>> configuration when the RSS key is not specified in the flow rule.
>>>>>>>>>>>>>> If the NIC RSS key is invalid, it will use testpmd dummy RSS key 
>>>>>>>>>>>>>> as the
>>>>>>>>>> default key.
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> Fixes: ac8d22de2394 ("ethdev: flatten RSS configuration in flow
>>>>>>>>>>>>>> API")
>>>>>>>>>>>>>> Cc: stable@dpdk.org
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> Signed-off-by: Lijun Ou <oulijun@huawei.com>
>>>>>>>>>>>>>> Reviewed-by: Phil Yang <phil.yang@arm.com>
>>>>>>>>>>>>>> ---
>>>>>>>>>>>>>> V4->V5:
>>>>>>>>>>>>>> -rewrite the commit log
>>>>>>>>>>>>>> -add reviewed-by
>>>>>>>>>>>>>
>>>>>>>>>>>>> Hi Lijun,
>>>>>>>>>>>>>
>>>>>>>>>>>>> There were multiple other comments, it seems they are not addressed
>>>>>>>>>>>>> but only updated the commit log, can you please check comments to
>>>>>>>>>> prev versions.
>>>>>>>>>>>>>
>>>>>>>>>>>>> Before going into the details, my question was what happens if
>>>>>>>>>>>>> default key not provided at all?
>>>>>>>>>>>>> It seems this has been already tried by Ophir [1], later reverted
>>>>>>>>>>>>> back [2] bringing the initial issue back.
>>>>>>>>>>>>>
>>>>>>>>>>>>> According commit, the reason of revert is to support following
>>>>>>>>>> command:
>>>>>>>>>>>>> "flow create 0 <pattern> actions rss queues 0 1 end key_len 40 / end"
>>>>>>>>>>>>>
>>>>>>>>>>>>> @Ophir, @Lijun,
>>>>>>>>>>>>> Can we ignore the 'key_len' if the 'key' is not supported and solve
>>>>>>>>>>>>> current issue as initially intended ([1])?
>>>>>>>>>>>>>
>>>>>>>>>>>> Hi, Ferruh
>>>>>>>>>>>>     I have discussed with Phil Yang about the problem in [1]. I think
>>>>>>>>>>>> there may be other problems with the idea and there is no better
>>>>>>>>>>>> solution. and we need to remove key_len definition from rte_rss_conf
>>>>>>>>>>>> structure. They don't have a plan. And [1] was eventually reverted.
>>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> Why ignoring 'key_len' (set it to zero) when there is no 'key'
>>>>>>>>>>> provided doesn't work?
>>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> What do you think [1] + following update, will it work?
>>>>>>>>>>
>>>>>>>>>>    diff --git a/lib/librte_ethdev/rte_flow.c 
>>>>>>>>>> b/lib/librte_ethdev/rte_flow.c
>>>>>>>>>>    index ee4f3464fe..e7789c87b3 100644
>>>>>>>>>>    --- a/lib/librte_ethdev/rte_flow.c
>>>>>>>>>>    +++ b/lib/librte_ethdev/rte_flow.c
>>>>>>>>>>    @@ -535,7 +535,7 @@ rte_flow_conv_action_conf(void *buf, const size_t
>>>>>>>>>> size,
>>>>>>>>>>                               }),
>>>>>>>>>>                               size > sizeof(*dst.rss) ? 
>>>>>>>>>> sizeof(*dst.rss) : size);
>>>>>>>>>>                    off = sizeof(*dst.rss);
>>>>>>>>>>    -               if (src.rss->key_len) {
>>>>>>>>>>    +               if (src.rss->key_len && src.rss->key) {
>>>>>>>>>>                            off = RTE_ALIGN_CEIL(off, 
>>>>>>>>>> sizeof(*dst.rss->key));
>>>>>>>>>>                            tmp = sizeof(*src.rss->key) * 
>>>>>>>>>> src.rss->key_len;
>>>>>>>>>>                            if (size >= off + tmp)
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>> Ferruh, your suggestion ([1] + update) looks correct. I also verified 
>>>>>>>>> it on mlx5 PMD.
>>>>>>>>> Advantage: it's a generic fix for all dpdk applications using rte_flows 
>>>>>>>>> (not just testpmd).
>>>>>>>>> It reduces code.
>>>>>>>>> With this fix the responsibility of handling key==NULL and/or len==0 is 
>>>>>>>>> moved to the PMDs (which is good).
>>>>>>>>> With regard to Lijun patch - I liked the approach of overriding the 
>>>>>>>>> default testpmd key with the default PMD key.
>>>>>>>>> But it only addresses testpmd. More code was added.
>>>>>>>>> It seems OK to call rte_eth_dev_rss_hash_conf_get() as part of parsing 
>>>>>>>>> RSS, but it would feel more confident if we could confirm it for all 
>>>>>>>>> the PMDs (by testing) or at least review the PMDs rss_hash_conf_get() 
>>>>>>>>> implementations.
>>>>>>>>>
>>>>>>>>
>>>>>>>> Lijun's idea can work. There was a problem in implementation related to 
>>>>>>>> the key size assumption, which can be fixed.
>>>>>>>>
>>>>>>>> Even it is fixed, when user gives a rss rule without a key, we are 
>>>>>>>> getting key from device and feeding same key back to device, this is 
>>>>>>>> unnecessary I think.
>>>>>>>> When user didn't provide a key, rss rule shouldn't touch the key at all.
>>>>>>>>
>>>>>>> Do you mean that the driver should not configure the key to the hardware 
>>>>>>> when the RSS key is not specified?
>>>>>>
>>>>>> We are talking about 'key' in RSS rte flow rule, not generally configuring 
>>>>>> the device for RSS.
>>>>>>
>>>>> Yes, I know. However, I am talking about some implementations. The 
>>>>> configuration interface may be the same as that for configuring RSS
>>>>> parameters of the device in general mode.
>>>>>> If a specific rss rte flow rule, lets say to filter some defined packets 
>>>>>> to some specific queues, don't have 'key' as a part of rule, I am 
>>>>>> suggesting not touch 'key' configuration of the device.
>>>>>>
>>>>> Yes, I agree. I think your point of view is very reasonable and a more
>>>>> appropriate implementation. However, for the sake of simplicity and
>>>>> other considerations in the architecture design of some devices, the
>>>>> hardware may support reasonable hybrid configuration for paramter 
>>>>> configuration in the RSS config, rather than independent configuration
>>>>> for hw, that is, maybe touch A of rss configuration of the device and
>>>>> must touch B of rss configuration. As a result, If the preceding scenario 
>>>>> occurs, it is reasonable to configure an RSS config that does not change 
>>>>> and specified.What do you think?
>>>>>>> If so, we cannot identify when the user does not specify the RSS key to 
>>>>>>> determine whether to configure the key.
>>>>>>
>>>>>> I think we can. If the rule has 'key' attribute, driver can use that key 
>>>>>> to program the device, if 'key' attribute is missing, don't do anything 
>>>>>> related to the rss key.
>>>>>>
>>>>> Yes, if 'key' attribute is missing, don't do anything related to the rss 
>>>>> key. It's more reasonable.
>>>>> But, If he can't do that, I think it is reasonable to configure the default 
>>>>> key that already exists on the device. The only disadvantage is wasteful.
>>>>>>> In hardware design, most vendors do not configure keys for hardware 
>>>>>>> independently, which may be associated with other RSS config parameters.
>>>>>>> I think it would be reasonable for us to reconfigure the RSS key with the 
>>>>>>> default value configured and valid in the hardware as the default value 
>>>>>>> if the user does not specify the RSS key.
>>>>>>
>>>>>> There are already APIs to update the RSS configuration, like RSS key, hash 
>>>>>> functions etc.. They are independent from the rte flow RSS rule.
>>>>>> Application should configure RSS according needs using those APIs.
>>>>>>
>>>>> Yes, should do this. Are there any unreasonable users who simply do not 
>>>>> know these APIs or do not want to use them and want to configure some 
>>>>> parameters through the rte flow RSS rule without changing other parameters?
>>>>>> The question here is about each RSS rte flow rule that can update the key. 
>>>>>> Am I missing something?
>>>>> I don't think you've misunderstood the general situation.
>>>>> Even if the RSS key is not specified in a rule, the driver uses the default 
>>>>> key value to re-touch the device.
>>>>>>
>>>>>>>> Complication was when user provides key_len without a key, I think both 
>>>>>>>> ignoring or returning error in this case is OK.
>>>>>>> I agree. However, I think it is meaningless to expose the RSS key length 
>>>>>>> to users. Do you consider deleting the RSS key length directly?
>>>>>>
>>>>>> Isn't both 'key' and 'key_len' needed to program the RSS key? Can the 
>>>>>> driver know the size of the 'key' without 'key_len'?
>>>>>> And driver should verify these provided values.
>>>>>>
>>>>> I know and agree.It is only during the analysis of the [1] scheme, from the 
>>>>> time it was proposed to the time it was withdrawn, that I found some guys 
>>>>> in the community questioned the significance of key_len.
>>>>>
>>>>> Back to the subject, what is your plan for the solution in the patch?
>>>>
>>>> I think, Ophir's old reverted patch [1] + 'rte_flow_conv_action_conf()' 
>>>> update I proposed above can work. Is this working for you? If so can you 
>>>> please send a patch?
>>>>
>>>>
>>>> [1] https://git.dpdk.org/dpdk/commit/?id=a4391f8bae8
>>>> .
>>> Hi, Ferruh
>>>     I've just used [1] + 'rte_flow_conv_action_conf()' for testing based on 
>>> the Kunpeng920 NIC plation and use hns3 pmd driver. it can be resolve the RSS 
>>> key question.
>>>     testpmd> show port 0 rss-hash key
>>> RSS functions:
>>>   all ipv4-frag ipv4-other ipv6-frag ipv6-other ip
>>> RSS key:
>>> 6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA38030F20C6A42B73BBEAC01FA
>>> testpmd> flow create 0 ingress pattern end actions rss func 
>>> symmetric_toeplitz types all end / end
>>> 0000:7d:00.0 hns3_config_rss_filter(): modified RSS types based on hardware 
>>> support, requested:7f83fffc configured:3ef8
>>> 0000:7d:00.0 hns3_parse_rss_key(): no valid RSS key config, using default
>>> Flow rule #0 created
>>> testpmd> show port 0 rss-hash key
>>> RSS functions:
>>>   all ipv4-frag ipv4-tcp ipv4-udp ipv4-sctp ipv4-other ipv6-frag ipv6-tcp 
>>> ipv6-udp ipv6-sctp ipv6-other ip udp tcp sctp
>>> RSS key:
>>> 6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA38030F20C6A42B73BBEAC01FA
>>>
>>> However, I also tested the case. I am not specified the RSS key and specified 
>>> the key_len when create a RSS flow rule, it can create success.
>>> testpmd> flow create 0 ingress pattern end actions rss queues 0 1 end key_len 
>>> 40 / end
>>> 0000:7d:00.0 hns3_config_rss_filter(): modified RSS types based on hardware 
>>> support, requested:a38c configured:2288
>>> 0000:7d:00.0 hns3_parse_rss_key(): no valid RSS key config, using default
>>> Flow rule #0 created
>>> testpmd> show port 0 rss-hash key
>>> RSS functions:
>>>   all ipv4-frag ipv4-other ipv6-frag ipv6-other ip
>>> RSS key:
>>> 6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA38030F20C6A42B73BBEAC01FA
>>>
>>
>> So are you happy with this solution?
>> .
> Yes, thanks.

cool, will you able to send a new version? Thanks.
Lijun Ou Oct. 21, 2020, 10:07 a.m. UTC | #22
在 2020/10/21 17:38, Ferruh Yigit 写道:
> On 10/21/2020 9:19 AM, oulijun wrote:
>>
>>
>> 在 2020/10/20 22:34, Ferruh Yigit 写道:
>>> On 10/20/2020 2:35 PM, oulijun wrote:
>>>>
>>>>
>>>> 在 2020/10/20 18:02, Ferruh Yigit 写道:
>>>>> On 10/20/2020 10:00 AM, oulijun wrote:
>>>>>>
>>>>>>
>>>>>> 在 2020/10/16 18:57, Ferruh Yigit 写道:
>>>>>>> On 10/16/2020 11:04 AM, oulijun wrote:
>>>>>>>>
>>>>>>>>
>>>>>>>> 在 2020/10/16 7:53, Ferruh Yigit 写道:
>>>>>>>>> On 10/16/2020 12:21 AM, Ophir Munk wrote:
>>>>>>>>>>> From: Ferruh Yigit <ferruh.yigit@intel.com>
>>>>>>>>>>>>>> On 10/15/2020 1:41 PM, Lijun Ou wrote:
>>>>>>>>>>>>>>> When start the testpmd, the pmd driver initializes the RSS
>>>>>>>>>>>>>>> configuration. Generally, the recommended RSS hash key is 
>>>>>>>>>>>>>>> used as
>>>>>>>>>>>>>>> the default key in the driver. In addition, the default 
>>>>>>>>>>>>>>> key is
>>>>>>>>>>>>>>> different from the default RSS flow in testpmd without 
>>>>>>>>>>>>>>> specifying
>>>>>>>>>>>>>>> RSS hash key. So. if you do not specify the RSS key when 
>>>>>>>>>>>>>>> creating
>>>>>>>>>>>>>>> an RSS rule, the testpmd uses the default key as the 
>>>>>>>>>>>>>>> default RSS
>>>>>>>>>>>>>>> key of the RSS rule. As a result, you may mistakenly 
>>>>>>>>>>>>>>> consider that
>>>>>>>>>>>>>>> the RSS key in use is the valid default key of the NIC, 
>>>>>>>>>>>>>>> actually,
>>>>>>>>>>>>>>> the key and the valid default key of the NIC are two values.
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> Consider the follow usage with testpmd:
>>>>>>>>>>>>>>> 1. first, startup testpmd:
>>>>>>>>>>>>>>> testpmd> show port 0 rss-hash key
>>>>>>>>>>>>>>> RSS functions:
>>>>>>>>>>>>>>>     all ipv4-frag ipv4-other ipv6-frag ipv6-other ip RSS 
>>>>>>>>>>>>>>> key:
>>>>>>>>>>>>>>> -
>>>>>>>>>>> 6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA3803
>>>>>>>>>>> 0F
>>>>>>>>>>>>>>> -20C6A42B73BBEAC01FA
>>>>>>>>>>>>>>> 2. create a rss rule
>>>>>>>>>>>>>>> testpmd> flow create 0 ingress pattern eth / ipv4 / udp / 
>>>>>>>>>>>>>>> end
>>>>>>>>>>>>>>> testpmd> actions rss \
>>>>>>>>>>>>>>> types ipv4-udp end queues end / end
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> 3. show rss-hash key
>>>>>>>>>>>>>>> testpmd> show port 0 rss-hash key
>>>>>>>>>>>>>>> RSS functions:
>>>>>>>>>>>>>>>     all ipv4-udp udp
>>>>>>>>>>>>>>> RSS key:
>>>>>>>>>>>>>>> -
>>>>>>>>>>> 74657374706D6427732064656661756C74205253532068617368206B65792C
>>>>>>>>>>> 206F
>>>>>>>>>>>>>>> -76657272696465
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> In order to solve the above problems, it use the NIC 
>>>>>>>>>>>>>>> valid default
>>>>>>>>>>>>>>> RSS key instead of the testpmd dummy RSS key in the flow
>>>>>>>>>>>>>>> configuration when the RSS key is not specified in the 
>>>>>>>>>>>>>>> flow rule.
>>>>>>>>>>>>>>> If the NIC RSS key is invalid, it will use testpmd dummy 
>>>>>>>>>>>>>>> RSS key as the
>>>>>>>>>>> default key.
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> Fixes: ac8d22de2394 ("ethdev: flatten RSS configuration 
>>>>>>>>>>>>>>> in flow
>>>>>>>>>>>>>>> API")
>>>>>>>>>>>>>>> Cc: stable@dpdk.org
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> Signed-off-by: Lijun Ou <oulijun@huawei.com>
>>>>>>>>>>>>>>> Reviewed-by: Phil Yang <phil.yang@arm.com>
>>>>>>>>>>>>>>> ---
>>>>>>>>>>>>>>> V4->V5:
>>>>>>>>>>>>>>> -rewrite the commit log
>>>>>>>>>>>>>>> -add reviewed-by
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> Hi Lijun,
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> There were multiple other comments, it seems they are not 
>>>>>>>>>>>>>> addressed
>>>>>>>>>>>>>> but only updated the commit log, can you please check 
>>>>>>>>>>>>>> comments to
>>>>>>>>>>> prev versions.
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> Before going into the details, my question was what 
>>>>>>>>>>>>>> happens if
>>>>>>>>>>>>>> default key not provided at all?
>>>>>>>>>>>>>> It seems this has been already tried by Ophir [1], later 
>>>>>>>>>>>>>> reverted
>>>>>>>>>>>>>> back [2] bringing the initial issue back.
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> According commit, the reason of revert is to support 
>>>>>>>>>>>>>> following
>>>>>>>>>>> command:
>>>>>>>>>>>>>> "flow create 0 <pattern> actions rss queues 0 1 end 
>>>>>>>>>>>>>> key_len 40 / end"
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> @Ophir, @Lijun,
>>>>>>>>>>>>>> Can we ignore the 'key_len' if the 'key' is not supported 
>>>>>>>>>>>>>> and solve
>>>>>>>>>>>>>> current issue as initially intended ([1])?
>>>>>>>>>>>>>>
>>>>>>>>>>>>> Hi, Ferruh
>>>>>>>>>>>>>     I have discussed with Phil Yang about the problem in 
>>>>>>>>>>>>> [1]. I think
>>>>>>>>>>>>> there may be other problems with the idea and there is no 
>>>>>>>>>>>>> better
>>>>>>>>>>>>> solution. and we need to remove key_len definition from 
>>>>>>>>>>>>> rte_rss_conf
>>>>>>>>>>>>> structure. They don't have a plan. And [1] was eventually 
>>>>>>>>>>>>> reverted.
>>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> Why ignoring 'key_len' (set it to zero) when there is no 'key'
>>>>>>>>>>>> provided doesn't work?
>>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> What do you think [1] + following update, will it work?
>>>>>>>>>>>
>>>>>>>>>>>    diff --git a/lib/librte_ethdev/rte_flow.c 
>>>>>>>>>>> b/lib/librte_ethdev/rte_flow.c
>>>>>>>>>>>    index ee4f3464fe..e7789c87b3 100644
>>>>>>>>>>>    --- a/lib/librte_ethdev/rte_flow.c
>>>>>>>>>>>    +++ b/lib/librte_ethdev/rte_flow.c
>>>>>>>>>>>    @@ -535,7 +535,7 @@ rte_flow_conv_action_conf(void *buf, 
>>>>>>>>>>> const size_t
>>>>>>>>>>> size,
>>>>>>>>>>>                               }),
>>>>>>>>>>>                               size > sizeof(*dst.rss) ? 
>>>>>>>>>>> sizeof(*dst.rss) : size);
>>>>>>>>>>>                    off = sizeof(*dst.rss);
>>>>>>>>>>>    -               if (src.rss->key_len) {
>>>>>>>>>>>    +               if (src.rss->key_len && src.rss->key) {
>>>>>>>>>>>                            off = RTE_ALIGN_CEIL(off, 
>>>>>>>>>>> sizeof(*dst.rss->key));
>>>>>>>>>>>                            tmp = sizeof(*src.rss->key) * 
>>>>>>>>>>> src.rss->key_len;
>>>>>>>>>>>                            if (size >= off + tmp)
>>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> Ferruh, your suggestion ([1] + update) looks correct. I also 
>>>>>>>>>> verified it on mlx5 PMD.
>>>>>>>>>> Advantage: it's a generic fix for all dpdk applications using 
>>>>>>>>>> rte_flows (not just testpmd).
>>>>>>>>>> It reduces code.
>>>>>>>>>> With this fix the responsibility of handling key==NULL and/or 
>>>>>>>>>> len==0 is moved to the PMDs (which is good).
>>>>>>>>>> With regard to Lijun patch - I liked the approach of 
>>>>>>>>>> overriding the default testpmd key with the default PMD key.
>>>>>>>>>> But it only addresses testpmd. More code was added.
>>>>>>>>>> It seems OK to call rte_eth_dev_rss_hash_conf_get() as part of 
>>>>>>>>>> parsing RSS, but it would feel more confident if we could 
>>>>>>>>>> confirm it for all the PMDs (by testing) or at least review 
>>>>>>>>>> the PMDs rss_hash_conf_get() implementations.
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>> Lijun's idea can work. There was a problem in implementation 
>>>>>>>>> related to the key size assumption, which can be fixed.
>>>>>>>>>
>>>>>>>>> Even it is fixed, when user gives a rss rule without a key, we 
>>>>>>>>> are getting key from device and feeding same key back to 
>>>>>>>>> device, this is unnecessary I think.
>>>>>>>>> When user didn't provide a key, rss rule shouldn't touch the 
>>>>>>>>> key at all.
>>>>>>>>>
>>>>>>>> Do you mean that the driver should not configure the key to the 
>>>>>>>> hardware when the RSS key is not specified?
>>>>>>>
>>>>>>> We are talking about 'key' in RSS rte flow rule, not generally 
>>>>>>> configuring the device for RSS.
>>>>>>>
>>>>>> Yes, I know. However, I am talking about some implementations. The 
>>>>>> configuration interface may be the same as that for configuring RSS
>>>>>> parameters of the device in general mode.
>>>>>>> If a specific rss rte flow rule, lets say to filter some defined 
>>>>>>> packets to some specific queues, don't have 'key' as a part of 
>>>>>>> rule, I am suggesting not touch 'key' configuration of the device.
>>>>>>>
>>>>>> Yes, I agree. I think your point of view is very reasonable and a 
>>>>>> more
>>>>>> appropriate implementation. However, for the sake of simplicity and
>>>>>> other considerations in the architecture design of some devices, the
>>>>>> hardware may support reasonable hybrid configuration for paramter 
>>>>>> configuration in the RSS config, rather than independent 
>>>>>> configuration
>>>>>> for hw, that is, maybe touch A of rss configuration of the device and
>>>>>> must touch B of rss configuration. As a result, If the preceding 
>>>>>> scenario occurs, it is reasonable to configure an RSS config that 
>>>>>> does not change and specified.What do you think?
>>>>>>>> If so, we cannot identify when the user does not specify the RSS 
>>>>>>>> key to determine whether to configure the key.
>>>>>>>
>>>>>>> I think we can. If the rule has 'key' attribute, driver can use 
>>>>>>> that key to program the device, if 'key' attribute is missing, 
>>>>>>> don't do anything related to the rss key.
>>>>>>>
>>>>>> Yes, if 'key' attribute is missing, don't do anything related to 
>>>>>> the rss key. It's more reasonable.
>>>>>> But, If he can't do that, I think it is reasonable to configure 
>>>>>> the default key that already exists on the device. The only 
>>>>>> disadvantage is wasteful.
>>>>>>>> In hardware design, most vendors do not configure keys for 
>>>>>>>> hardware independently, which may be associated with other RSS 
>>>>>>>> config parameters.
>>>>>>>> I think it would be reasonable for us to reconfigure the RSS key 
>>>>>>>> with the default value configured and valid in the hardware as 
>>>>>>>> the default value if the user does not specify the RSS key.
>>>>>>>
>>>>>>> There are already APIs to update the RSS configuration, like RSS 
>>>>>>> key, hash functions etc.. They are independent from the rte flow 
>>>>>>> RSS rule.
>>>>>>> Application should configure RSS according needs using those APIs.
>>>>>>>
>>>>>> Yes, should do this. Are there any unreasonable users who simply 
>>>>>> do not know these APIs or do not want to use them and want to 
>>>>>> configure some parameters through the rte flow RSS rule without 
>>>>>> changing other parameters?
>>>>>>> The question here is about each RSS rte flow rule that can update 
>>>>>>> the key. Am I missing something?
>>>>>> I don't think you've misunderstood the general situation.
>>>>>> Even if the RSS key is not specified in a rule, the driver uses 
>>>>>> the default key value to re-touch the device.
>>>>>>>
>>>>>>>>> Complication was when user provides key_len without a key, I 
>>>>>>>>> think both ignoring or returning error in this case is OK.
>>>>>>>> I agree. However, I think it is meaningless to expose the RSS 
>>>>>>>> key length to users. Do you consider deleting the RSS key length 
>>>>>>>> directly?
>>>>>>>
>>>>>>> Isn't both 'key' and 'key_len' needed to program the RSS key? Can 
>>>>>>> the driver know the size of the 'key' without 'key_len'?
>>>>>>> And driver should verify these provided values.
>>>>>>>
>>>>>> I know and agree.It is only during the analysis of the [1] scheme, 
>>>>>> from the time it was proposed to the time it was withdrawn, that I 
>>>>>> found some guys in the community questioned the significance of 
>>>>>> key_len.
>>>>>>
>>>>>> Back to the subject, what is your plan for the solution in the patch?
>>>>>
>>>>> I think, Ophir's old reverted patch [1] + 
>>>>> 'rte_flow_conv_action_conf()' update I proposed above can work. Is 
>>>>> this working for you? If so can you please send a patch?
>>>>>
>>>>>
>>>>> [1] https://git.dpdk.org/dpdk/commit/?id=a4391f8bae8
>>>>> .
>>>> Hi, Ferruh
>>>>     I've just used [1] + 'rte_flow_conv_action_conf()' for testing 
>>>> based on the Kunpeng920 NIC plation and use hns3 pmd driver. it can 
>>>> be resolve the RSS key question.
>>>>     testpmd> show port 0 rss-hash key
>>>> RSS functions:
>>>>   all ipv4-frag ipv4-other ipv6-frag ipv6-other ip
>>>> RSS key:
>>>> 6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA38030F20C6A42B73BBEAC01FA 
>>>>
>>>> testpmd> flow create 0 ingress pattern end actions rss func 
>>>> symmetric_toeplitz types all end / end
>>>> 0000:7d:00.0 hns3_config_rss_filter(): modified RSS types based on 
>>>> hardware support, requested:7f83fffc configured:3ef8
>>>> 0000:7d:00.0 hns3_parse_rss_key(): no valid RSS key config, using 
>>>> default
>>>> Flow rule #0 created
>>>> testpmd> show port 0 rss-hash key
>>>> RSS functions:
>>>>   all ipv4-frag ipv4-tcp ipv4-udp ipv4-sctp ipv4-other ipv6-frag 
>>>> ipv6-tcp ipv6-udp ipv6-sctp ipv6-other ip udp tcp sctp
>>>> RSS key:
>>>> 6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA38030F20C6A42B73BBEAC01FA 
>>>>
>>>>
>>>> However, I also tested the case. I am not specified the RSS key and 
>>>> specified the key_len when create a RSS flow rule, it can create 
>>>> success.
>>>> testpmd> flow create 0 ingress pattern end actions rss queues 0 1 
>>>> end key_len 40 / end
>>>> 0000:7d:00.0 hns3_config_rss_filter(): modified RSS types based on 
>>>> hardware support, requested:a38c configured:2288
>>>> 0000:7d:00.0 hns3_parse_rss_key(): no valid RSS key config, using 
>>>> default
>>>> Flow rule #0 created
>>>> testpmd> show port 0 rss-hash key
>>>> RSS functions:
>>>>   all ipv4-frag ipv4-other ipv6-frag ipv6-other ip
>>>> RSS key:
>>>> 6D5A56DA255B0EC24167253D43A38FB0D0CA2BCBAE7B30B477CB2DA38030F20C6A42B73BBEAC01FA 
>>>>
>>>>
>>>
>>> So are you happy with this solution?
>>> .
>> Yes, thanks.
> 
> cool, will you able to send a new version? Thanks.
> .
Yes
>

Patch
diff mbox series

diff --git a/app/test-pmd/cmdline_flow.c b/app/test-pmd/cmdline_flow.c
index 84bba0f..578555e 100644
--- a/app/test-pmd/cmdline_flow.c
+++ b/app/test-pmd/cmdline_flow.c
@@ -4735,6 +4735,7 @@  parse_vc_action_rss(struct context *ctx, const struct token *token,
 		action_rss_data->queue[i] = i;
 	if (!port_id_is_invalid(ctx->port, DISABLED_WARN) &&
 	    ctx->port != (portid_t)RTE_PORT_ALL) {
+		struct rte_eth_rss_conf rss_conf = {0};
 		struct rte_eth_dev_info info;
 		int ret2;
 
@@ -4745,6 +4746,13 @@  parse_vc_action_rss(struct context *ctx, const struct token *token,
 		action_rss_data->conf.key_len =
 			RTE_MIN(sizeof(action_rss_data->key),
 				info.hash_key_size);
+
+		rss_conf.rss_key_len = sizeof(action_rss_data->key);
+		rss_conf.rss_key = action_rss_data->key;
+		ret2 = rte_eth_dev_rss_hash_conf_get(ctx->port, &rss_conf);
+		if (ret2 != 0)
+			return ret2;
+		action_rss_data->conf.key = rss_conf.rss_key;
 	}
 	action->conf = &action_rss_data->conf;
 	return ret;