doc: update RSS action with best effort
diff mbox series

Message ID 1596464905-43867-1-git-send-email-orika@mellanox.com
State Superseded, archived
Headers show
Series
  • doc: update RSS action with best effort
Related show

Checks

Context Check Description
ci/travis-robot success Travis build: passed
ci/Intel-compilation success Compilation OK
ci/checkpatch success coding style OK

Commit Message

Ori Kam Aug. 3, 2020, 2:28 p.m. UTC
Using the rte_flow action RSS types field,
may result in an undefined outcome.

For example selecting both UDP and TCP,
selecting TCP RSS type but the pattern is targeting UDP traffic.
another option is that the PMD doesn't support all requested types.

Until now, it wasn't clear what will happen in such cases.
This commit clarify this issue by stating that the PMD
will work in the best-effort mode.

Signed-off-by: Ori Kam <orika@mellanox.com>
---
 doc/guides/prog_guide/rte_flow.rst | 10 ++++++++++
 1 file changed, 10 insertions(+)

Comments

Andrew Rybchenko Aug. 3, 2020, 2:33 p.m. UTC | #1
On 8/3/20 5:28 PM, Ori Kam wrote:
> Using the rte_flow action RSS types field,
> may result in an undefined outcome.
> 
> For example selecting both UDP and TCP,
> selecting TCP RSS type but the pattern is targeting UDP traffic.
> another option is that the PMD doesn't support all requested types.
> 
> Until now, it wasn't clear what will happen in such cases.
> This commit clarify this issue by stating that the PMD
> will work in the best-effort mode.
> 
> Signed-off-by: Ori Kam <orika@mellanox.com>
> ---
>  doc/guides/prog_guide/rte_flow.rst | 10 ++++++++++
>  1 file changed, 10 insertions(+)
> 
> diff --git a/doc/guides/prog_guide/rte_flow.rst b/doc/guides/prog_guide/rte_flow.rst
> index 3e5cd1e..d000560 100644
> --- a/doc/guides/prog_guide/rte_flow.rst
> +++ b/doc/guides/prog_guide/rte_flow.rst
> @@ -1735,6 +1735,16 @@ unspecified "best-effort" settings from the underlying PMD, which depending
>  on the flow rule, may result in anything ranging from empty (single queue)
>  to all-inclusive RSS.
>  
> +Best effort will be used, in case there is a conflict inside the rule.
> +The conflict can be the result of:
> +
> +- Conflicting RSS types, for example setting both UDP and TCP.
> +
> +- Conflicting between RSS types and the requested pattern to match,
> +  for example matching on UDP and hashing RSS on TCP.

IMHO it is not a problem at all, but good to clarify anyway.

> +
> +- Hashing on types that are not supported by the PMD.

Shouldn't it return error to the caller?

> +
>  Note: RSS hash result is stored in the ``hash.rss`` mbuf field which
>  overlaps ``hash.fdir.lo``. Since `Action: MARK`_ sets the ``hash.fdir.hi``
>  field only, both can be requested simultaneously.
>
Ori Kam Aug. 3, 2020, 3:22 p.m. UTC | #2
Hi Andrew,

> -----Original Message-----
> From: Andrew Rybchenko <arybchenko@solarflare.com>
> 
> On 8/3/20 5:28 PM, Ori Kam wrote:
> > Using the rte_flow action RSS types field,
> > may result in an undefined outcome.
> >
> > For example selecting both UDP and TCP,
> > selecting TCP RSS type but the pattern is targeting UDP traffic.
> > another option is that the PMD doesn't support all requested types.
> >
> > Until now, it wasn't clear what will happen in such cases.
> > This commit clarify this issue by stating that the PMD
> > will work in the best-effort mode.
> >
> > Signed-off-by: Ori Kam <orika@mellanox.com>
> > ---
> >  doc/guides/prog_guide/rte_flow.rst | 10 ++++++++++
> >  1 file changed, 10 insertions(+)
> >
> > diff --git a/doc/guides/prog_guide/rte_flow.rst
> b/doc/guides/prog_guide/rte_flow.rst
> > index 3e5cd1e..d000560 100644
> > --- a/doc/guides/prog_guide/rte_flow.rst
> > +++ b/doc/guides/prog_guide/rte_flow.rst
> > @@ -1735,6 +1735,16 @@ unspecified "best-effort" settings from the
> underlying PMD, which depending
> >  on the flow rule, may result in anything ranging from empty (single queue)
> >  to all-inclusive RSS.
> >
> > +Best effort will be used, in case there is a conflict inside the rule.
> > +The conflict can be the result of:
> > +
> > +- Conflicting RSS types, for example setting both UDP and TCP.
> > +
> > +- Conflicting between RSS types and the requested pattern to match,
> > +  for example matching on UDP and hashing RSS on TCP.
> 
> IMHO it is not a problem at all, but good to clarify anyway.
> 
Agree this patch is just for clarification.

> > +
> > +- Hashing on types that are not supported by the PMD.
> 
> Shouldn't it return error to the caller?
> 
That’s depends, if for example the application requested eth and IPv4,
and the PMD can do only IPv4 so according to this, the PMD will just do IPv4.

> > +
> >  Note: RSS hash result is stored in the ``hash.rss`` mbuf field which
> >  overlaps ``hash.fdir.lo``. Since `Action: MARK`_ sets the ``hash.fdir.hi``
> >  field only, both can be requested simultaneously.
> >
Andrew Rybchenko Aug. 3, 2020, 3:33 p.m. UTC | #3
On 8/3/20 6:22 PM, Ori Kam wrote:
> Hi Andrew,
>
>> -----Original Message-----
>> From: Andrew Rybchenko <arybchenko@solarflare.com>
>>
>> On 8/3/20 5:28 PM, Ori Kam wrote:
>>> Using the rte_flow action RSS types field,
>>> may result in an undefined outcome.
>>>
>>> For example selecting both UDP and TCP,
>>> selecting TCP RSS type but the pattern is targeting UDP traffic.
>>> another option is that the PMD doesn't support all requested types.
>>>
>>> Until now, it wasn't clear what will happen in such cases.
>>> This commit clarify this issue by stating that the PMD
>>> will work in the best-effort mode.
>>>
>>> Signed-off-by: Ori Kam <orika@mellanox.com>
>>> ---
>>>  doc/guides/prog_guide/rte_flow.rst | 10 ++++++++++
>>>  1 file changed, 10 insertions(+)
>>>
>>> diff --git a/doc/guides/prog_guide/rte_flow.rst
>> b/doc/guides/prog_guide/rte_flow.rst
>>> index 3e5cd1e..d000560 100644
>>> --- a/doc/guides/prog_guide/rte_flow.rst
>>> +++ b/doc/guides/prog_guide/rte_flow.rst
>>> @@ -1735,6 +1735,16 @@ unspecified "best-effort" settings from the
>> underlying PMD, which depending
>>>  on the flow rule, may result in anything ranging from empty (single queue)
>>>  to all-inclusive RSS.
>>>
>>> +Best effort will be used, in case there is a conflict inside the rule.
>>> +The conflict can be the result of:
>>> +
>>> +- Conflicting RSS types, for example setting both UDP and TCP.
>>> +
>>> +- Conflicting between RSS types and the requested pattern to match,
>>> +  for example matching on UDP and hashing RSS on TCP.
>> IMHO it is not a problem at all, but good to clarify anyway.
>>
> Agree this patch is just for clarification.
>
>>> +
>>> +- Hashing on types that are not supported by the PMD.
>> Shouldn't it return error to the caller?
>>
> That’s depends, if for example the application requested eth and IPv4,
> and the PMD can do only IPv4 so according to this, the PMD will just do IPv4.

I think it is a bad behaviour. Application has required information to
provide correct parameters and therefore incorrect should be rejected.
Am I missing something?

>>> +
>>>  Note: RSS hash result is stored in the ``hash.rss`` mbuf field which
>>>  overlaps ``hash.fdir.lo``. Since `Action: MARK`_ sets the ``hash.fdir.hi``
>>>  field only, both can be requested simultaneously.
>>>
David Marchand Aug. 3, 2020, 3:40 p.m. UTC | #4
On Mon, Aug 3, 2020 at 5:23 PM Ori Kam <orika@mellanox.com> wrote:
> > > +
> > > +- Hashing on types that are not supported by the PMD.
> >
> > Shouldn't it return error to the caller?
> >
> That’s depends, if for example the application requested eth and IPv4,
> and the PMD can do only IPv4 so according to this, the PMD will just do IPv4.

We should have some validation in ethdev to catch this.
Is it not the case?
Ori Kam Aug. 3, 2020, 3:47 p.m. UTC | #5
> -----Original Message-----
> From: Andrew Rybchenko <arybchenko@solarflare.com>
> 
> On 8/3/20 6:22 PM, Ori Kam wrote:
> > Hi Andrew,
> >
> >> -----Original Message-----
> >> From: Andrew Rybchenko <arybchenko@solarflare.com>
> >>
> >> On 8/3/20 5:28 PM, Ori Kam wrote:
> >>> Using the rte_flow action RSS types field,
> >>> may result in an undefined outcome.
> >>>
> >>> For example selecting both UDP and TCP,
> >>> selecting TCP RSS type but the pattern is targeting UDP traffic.
> >>> another option is that the PMD doesn't support all requested types.
> >>>
> >>> Until now, it wasn't clear what will happen in such cases.
> >>> This commit clarify this issue by stating that the PMD
> >>> will work in the best-effort mode.
> >>>
> >>> Signed-off-by: Ori Kam <orika@mellanox.com>
> >>> ---
> >>>  doc/guides/prog_guide/rte_flow.rst | 10 ++++++++++
> >>>  1 file changed, 10 insertions(+)
> >>>
> >>> diff --git a/doc/guides/prog_guide/rte_flow.rst
> >> b/doc/guides/prog_guide/rte_flow.rst
> >>> index 3e5cd1e..d000560 100644
> >>> --- a/doc/guides/prog_guide/rte_flow.rst
> >>> +++ b/doc/guides/prog_guide/rte_flow.rst
> >>> @@ -1735,6 +1735,16 @@ unspecified "best-effort" settings from the
> >> underlying PMD, which depending
> >>>  on the flow rule, may result in anything ranging from empty (single queue)
> >>>  to all-inclusive RSS.
> >>>
> >>> +Best effort will be used, in case there is a conflict inside the rule.
> >>> +The conflict can be the result of:
> >>> +
> >>> +- Conflicting RSS types, for example setting both UDP and TCP.
> >>> +
> >>> +- Conflicting between RSS types and the requested pattern to match,
> >>> +  for example matching on UDP and hashing RSS on TCP.
> >> IMHO it is not a problem at all, but good to clarify anyway.
> >>
> > Agree this patch is just for clarification.
> >
> >>> +
> >>> +- Hashing on types that are not supported by the PMD.
> >> Shouldn't it return error to the caller?
> >>
> > That’s depends, if for example the application requested eth and IPv4,
> > and the PMD can do only IPv4 so according to this, the PMD will just do IPv4.
> 
> I think it is a bad behaviour. Application has required information to
> provide correct parameters and therefore incorrect should be rejected.
> Am I missing something?
> 
In RTE flow you can't know what are the parameters that are supported.
One option is to fail the rule, but this means failing the rule even if some of them are supported.
Or we can choose to fail if all the types are unsupported, but this will result in miss match between
adding two types and adding just one type. For example the PMD doesn't support RSS on eth
and the application request RSS on ETH and IPv4 so the PMD will do IPv4 RSS,
while if the user selects only ETH the rule will fail.
Same this is exactly the same case as miss match the application requests something
that the PMD can't supply. 
In any case I think this is the current behavior of the PMD so this just clarify it.

> >>> +
> >>>  Note: RSS hash result is stored in the ``hash.rss`` mbuf field which
> >>>  overlaps ``hash.fdir.lo``. Since `Action: MARK`_ sets the ``hash.fdir.hi``
> >>>  field only, both can be requested simultaneously.
> >>>
Ori Kam Aug. 3, 2020, 3:49 p.m. UTC | #6
Hi David,

> -----Original Message-----
> From: David Marchand <david.marchand@redhat.com>
> 
> On Mon, Aug 3, 2020 at 5:23 PM Ori Kam <orika@mellanox.com> wrote:
> > > > +
> > > > +- Hashing on types that are not supported by the PMD.
> > >
> > > Shouldn't it return error to the caller?
> > >
> > That’s depends, if for example the application requested eth and IPv4,
> > and the PMD can do only IPv4 so according to this, the PMD will just do IPv4.
> 
> We should have some validation in ethdev to catch this.
> Is it not the case?
> 
Like I said in my reply to Andrew, in rte_flow we don't have such caps.
Maybe we should think about adding them for the RSS case, but again it is tricky
What if one RSS type is supported depending on the flow or other types?

> 
> --
> David Marchand
Andrew Rybchenko Aug. 3, 2020, 3:55 p.m. UTC | #7
On 8/3/20 6:49 PM, Ori Kam wrote:
> Hi David,
>
>> -----Original Message-----
>> From: David Marchand <david.marchand@redhat.com>
>>
>> On Mon, Aug 3, 2020 at 5:23 PM Ori Kam <orika@mellanox.com> wrote:
>>>>> +
>>>>> +- Hashing on types that are not supported by the PMD.
>>>> Shouldn't it return error to the caller?
>>>>
>>> That’s depends, if for example the application requested eth and IPv4,
>>> and the PMD can do only IPv4 so according to this, the PMD will just do IPv4.
>> We should have some validation in ethdev to catch this.
>> Is it not the case?
>>
> Like I said in my reply to Andrew, in rte_flow we don't have such caps.
> Maybe we should think about adding them for the RSS case, but again it is tricky
> What if one RSS type is supported depending on the flow or other types?

Also I'd like to add that ethdev layer is dummy for rte_flow API.
It does not parse pattern/actions etc. May be should change it one day.
Andrew Rybchenko Aug. 3, 2020, 4:03 p.m. UTC | #8
On 8/3/20 6:47 PM, Ori Kam wrote:
> 
> 
>> -----Original Message-----
>> From: Andrew Rybchenko <arybchenko@solarflare.com>
>>
>> On 8/3/20 6:22 PM, Ori Kam wrote:
>>> Hi Andrew,
>>>
>>>> -----Original Message-----
>>>> From: Andrew Rybchenko <arybchenko@solarflare.com>
>>>>
>>>> On 8/3/20 5:28 PM, Ori Kam wrote:
>>>>> Using the rte_flow action RSS types field,
>>>>> may result in an undefined outcome.
>>>>>
>>>>> For example selecting both UDP and TCP,
>>>>> selecting TCP RSS type but the pattern is targeting UDP traffic.
>>>>> another option is that the PMD doesn't support all requested types.
>>>>>
>>>>> Until now, it wasn't clear what will happen in such cases.
>>>>> This commit clarify this issue by stating that the PMD
>>>>> will work in the best-effort mode.
>>>>>
>>>>> Signed-off-by: Ori Kam <orika@mellanox.com>
>>>>> ---
>>>>>  doc/guides/prog_guide/rte_flow.rst | 10 ++++++++++
>>>>>  1 file changed, 10 insertions(+)
>>>>>
>>>>> diff --git a/doc/guides/prog_guide/rte_flow.rst
>>>> b/doc/guides/prog_guide/rte_flow.rst
>>>>> index 3e5cd1e..d000560 100644
>>>>> --- a/doc/guides/prog_guide/rte_flow.rst
>>>>> +++ b/doc/guides/prog_guide/rte_flow.rst
>>>>> @@ -1735,6 +1735,16 @@ unspecified "best-effort" settings from the
>>>> underlying PMD, which depending
>>>>>  on the flow rule, may result in anything ranging from empty (single queue)
>>>>>  to all-inclusive RSS.
>>>>>
>>>>> +Best effort will be used, in case there is a conflict inside the rule.
>>>>> +The conflict can be the result of:
>>>>> +
>>>>> +- Conflicting RSS types, for example setting both UDP and TCP.
>>>>> +
>>>>> +- Conflicting between RSS types and the requested pattern to match,
>>>>> +  for example matching on UDP and hashing RSS on TCP.
>>>> IMHO it is not a problem at all, but good to clarify anyway.
>>>>
>>> Agree this patch is just for clarification.
>>>
>>>>> +
>>>>> +- Hashing on types that are not supported by the PMD.
>>>> Shouldn't it return error to the caller?
>>>>
>>> That’s depends, if for example the application requested eth and IPv4,
>>> and the PMD can do only IPv4 so according to this, the PMD will just do IPv4.
>>
>> I think it is a bad behaviour. Application has required information to
>> provide correct parameters and therefore incorrect should be rejected.
>> Am I missing something?
>>
> In RTE flow you can't know what are the parameters that are supported.
> One option is to fail the rule, but this means failing the rule even if some of them are supported.
> Or we can choose to fail if all the types are unsupported, but this will result in miss match between
> adding two types and adding just one type. For example the PMD doesn't support RSS on eth
> and the application request RSS on ETH and IPv4 so the PMD will do IPv4 RSS,
> while if the user selects only ETH the rule will fail.
> Same this is exactly the same case as miss match the application requests something
> that the PMD can't supply. 
> In any case I think this is the current behavior of the PMD so this just clarify it.

I'm sorry, but the PMD is hardly applicable here.
Do you mean the "mlx5" PMD? Or have you reviewed all PMDs?

No I understand that the topic is more complicated, but
covering complex cases should not spoil simple.
I think that information provided in dev_info should
specify all supported protocols/fields.
Yes, it is acceptable to behave best effort if requested
field is supported in general, but it not for the flow.
Please, make it a bit clearer in the suggested description.

> 
>>>>> +
>>>>>  Note: RSS hash result is stored in the ``hash.rss`` mbuf field which
>>>>>  overlaps ``hash.fdir.lo``. Since `Action: MARK`_ sets the ``hash.fdir.hi``
>>>>>  field only, both can be requested simultaneously.
>>>>>
>
Thomas Monjalon Aug. 3, 2020, 4:12 p.m. UTC | #9
03/08/2020 17:55, Andrew Rybchenko:
> On 8/3/20 6:49 PM, Ori Kam wrote:
> > From: David Marchand <david.marchand@redhat.com>
> >> On Mon, Aug 3, 2020 at 5:23 PM Ori Kam <orika@mellanox.com> wrote:
> >>>>> +
> >>>>> +- Hashing on types that are not supported by the PMD.
> >>>> Shouldn't it return error to the caller?
> >>>>
> >>> That’s depends, if for example the application requested eth and IPv4,
> >>> and the PMD can do only IPv4 so according to this, the PMD will just do IPv4.
> >> We should have some validation in ethdev to catch this.
> >> Is it not the case?
> >>
> > Like I said in my reply to Andrew, in rte_flow we don't have such caps.
> > Maybe we should think about adding them for the RSS case, but again it is tricky
> > What if one RSS type is supported depending on the flow or other types?
> 
> Also I'd like to add that ethdev layer is dummy for rte_flow API.
> It does not parse pattern/actions etc. May be should change it one day.

For now, what we can do is to have "best effort" (sic) checks.
In the case we are 100% sure a rule cannot be fully supported,
I think we should reject the rule.
If there are more complex cases to handle, we can accept the rule
and support it with "best effort" handling.

The other appropriate action is to implement tests in DTS
to check the behaviour of the PMDs, validating the compliance
of the accepted rules with real flows.
Ori Kam Aug. 3, 2020, 4:53 p.m. UTC | #10
Hi Andrew,

> -----Original Message-----
> From: Andrew Rybchenko <arybchenko@solarflare.com>
> 
> On 8/3/20 6:47 PM, Ori Kam wrote:
> >
> >
> >> -----Original Message-----
> >> From: Andrew Rybchenko <arybchenko@solarflare.com>
> >>
> >> On 8/3/20 6:22 PM, Ori Kam wrote:
> >>> Hi Andrew,
> >>>
> >>>> -----Original Message-----
> >>>> From: Andrew Rybchenko <arybchenko@solarflare.com>
> >>>>
> >>>> On 8/3/20 5:28 PM, Ori Kam wrote:
> >>>>> Using the rte_flow action RSS types field,
> >>>>> may result in an undefined outcome.
> >>>>>
> >>>>> For example selecting both UDP and TCP,
> >>>>> selecting TCP RSS type but the pattern is targeting UDP traffic.
> >>>>> another option is that the PMD doesn't support all requested types.
> >>>>>
> >>>>> Until now, it wasn't clear what will happen in such cases.
> >>>>> This commit clarify this issue by stating that the PMD
> >>>>> will work in the best-effort mode.
> >>>>>
> >>>>> Signed-off-by: Ori Kam <orika@mellanox.com>
> >>>>> ---
> >>>>>  doc/guides/prog_guide/rte_flow.rst | 10 ++++++++++
> >>>>>  1 file changed, 10 insertions(+)
> >>>>>
> >>>>> diff --git a/doc/guides/prog_guide/rte_flow.rst
> >>>> b/doc/guides/prog_guide/rte_flow.rst
> >>>>> index 3e5cd1e..d000560 100644
> >>>>> --- a/doc/guides/prog_guide/rte_flow.rst
> >>>>> +++ b/doc/guides/prog_guide/rte_flow.rst
> >>>>> @@ -1735,6 +1735,16 @@ unspecified "best-effort" settings from the
> >>>> underlying PMD, which depending
> >>>>>  on the flow rule, may result in anything ranging from empty (single
> queue)
> >>>>>  to all-inclusive RSS.
> >>>>>
> >>>>> +Best effort will be used, in case there is a conflict inside the rule.
> >>>>> +The conflict can be the result of:
> >>>>> +
> >>>>> +- Conflicting RSS types, for example setting both UDP and TCP.
> >>>>> +
> >>>>> +- Conflicting between RSS types and the requested pattern to match,
> >>>>> +  for example matching on UDP and hashing RSS on TCP.
> >>>> IMHO it is not a problem at all, but good to clarify anyway.
> >>>>
> >>> Agree this patch is just for clarification.
> >>>
> >>>>> +
> >>>>> +- Hashing on types that are not supported by the PMD.
> >>>> Shouldn't it return error to the caller?
> >>>>
> >>> That’s depends, if for example the application requested eth and IPv4,
> >>> and the PMD can do only IPv4 so according to this, the PMD will just do
> IPv4.
> >>
> >> I think it is a bad behaviour. Application has required information to
> >> provide correct parameters and therefore incorrect should be rejected.
> >> Am I missing something?
> >>
> > In RTE flow you can't know what are the parameters that are supported.
> > One option is to fail the rule, but this means failing the rule even if some of
> them are supported.
> > Or we can choose to fail if all the types are unsupported, but this will result in
> miss match between
> > adding two types and adding just one type. For example the PMD doesn't
> support RSS on eth
> > and the application request RSS on ETH and IPv4 so the PMD will do IPv4 RSS,
> > while if the user selects only ETH the rule will fail.
> > Same this is exactly the same case as miss match the application requests
> something
> > that the PMD can't supply.
> > In any case I think this is the current behavior of the PMD so this just clarify it.
> 
> I'm sorry, but the PMD is hardly applicable here.
> Do you mean the "mlx5" PMD? Or have you reviewed all PMDs?
> 
I assume that each PMD as its own limitations, I'm sure that no
PMD can support all possible RSS. But maybe there is such perfect PMD.

> No I understand that the topic is more complicated, but
> covering complex cases should not spoil simple.
> I think that information provided in dev_info should
> specify all supported protocols/fields.
> Yes, it is acceptable to behave best effort if requested
> field is supported in general, but it not for the flow.
> Please, make it a bit clearer in the suggested description.
> 
So what about the following wording: (after the current part)
If requested RSS hash type is not supported, and no supported hash function
is requested then the PMD may reject the flow.

> >
> >>>>> +
> >>>>>  Note: RSS hash result is stored in the ``hash.rss`` mbuf field which
> >>>>>  overlaps ``hash.fdir.lo``. Since `Action: MARK`_ sets the ``hash.fdir.hi``
> >>>>>  field only, both can be requested simultaneously.
> >>>>>
> >
Ferruh Yigit Sept. 14, 2020, 2:29 p.m. UTC | #11
On 8/3/2020 4:55 PM, Andrew Rybchenko wrote:
> On 8/3/20 6:49 PM, Ori Kam wrote:
>> Hi David,
>>
>>> -----Original Message-----
>>> From: David Marchand <david.marchand@redhat.com>
>>>
>>> On Mon, Aug 3, 2020 at 5:23 PM Ori Kam <orika@mellanox.com> wrote:
>>>>>> +
>>>>>> +- Hashing on types that are not supported by the PMD.
>>>>> Shouldn't it return error to the caller?
>>>>>
>>>> That’s depends, if for example the application requested eth and IPv4,
>>>> and the PMD can do only IPv4 so according to this, the PMD will just do IPv4.
>>> We should have some validation in ethdev to catch this.
>>> Is it not the case?
>>>
>> Like I said in my reply to Andrew, in rte_flow we don't have such caps.
>> Maybe we should think about adding them for the RSS case, but again it is tricky
>> What if one RSS type is supported depending on the flow or other types?
> 
> Also I'd like to add that ethdev layer is dummy for rte_flow API.
> It does not parse pattern/actions etc. May be should change it one day.
> 

As far as I remember not having capabilities was the design decision,
PMD does the validation via 'rte_flow_validate()'.

Indeed it can be quite complex to have that kind of capability
reporting, and without something like that there is not much to do in
ethdev layer.

Patch
diff mbox series

diff --git a/doc/guides/prog_guide/rte_flow.rst b/doc/guides/prog_guide/rte_flow.rst
index 3e5cd1e..d000560 100644
--- a/doc/guides/prog_guide/rte_flow.rst
+++ b/doc/guides/prog_guide/rte_flow.rst
@@ -1735,6 +1735,16 @@  unspecified "best-effort" settings from the underlying PMD, which depending
 on the flow rule, may result in anything ranging from empty (single queue)
 to all-inclusive RSS.
 
+Best effort will be used, in case there is a conflict inside the rule.
+The conflict can be the result of:
+
+- Conflicting RSS types, for example setting both UDP and TCP.
+
+- Conflicting between RSS types and the requested pattern to match,
+  for example matching on UDP and hashing RSS on TCP.
+
+- Hashing on types that are not supported by the PMD.
+
 Note: RSS hash result is stored in the ``hash.rss`` mbuf field which
 overlaps ``hash.fdir.lo``. Since `Action: MARK`_ sets the ``hash.fdir.hi``
 field only, both can be requested simultaneously.