doc: refine ethernet and VLAN flow rule items

  Specified pattern may be translated in different manner.
For example the pattern "eth / ipv4" can be translated to match
untagged packets only, since the pattern doesn't specify a vlan item.
It can also be translated to match both tagged and untagged packets,
for the same reason.
This patch updates the rte_flow documentation to clearly specify the
required pattern to use.
For example:
To match tagged ipv4 packets, the pattern "eth type is 0x8100 /
vlan / ipv4 / end" should be used.
To match untagged ipv4 packets, the pattern "eth type is 0x0800 /
ipv4 / end" should be used.
To match both tagged and untagged packets, the pattern "eth / end"
should be used.

Signed-off-by: Dekel Peled <dekelp@mellanox.com>
---
 doc/guides/prog_guide/rte_flow.rst | 8 ++++++++
 lib/librte_ethdev/rte_flow.h       | 9 +++++++++
 2 files changed, 17 insertions(+)
  

On 4/23/20 9:30 PM, Dekel Peled wrote:
> Specified pattern may be translated in different manner.
> For example the pattern "eth / ipv4" can be translated to match
> untagged packets only, since the pattern doesn't specify a vlan item.

vlan -> VLAN

> It can also be translated to match both tagged and untagged packets,
> for the same reason.
> This patch updates the rte_flow documentation to clearly specify the
> required pattern to use.
> For example:
> To match tagged ipv4 packets, the pattern "eth type is 0x8100 /
> vlan / ipv4 / end" should be used.

Isn't eth / vlan / ipv4 /end sufficient? What's the difference?
I guess later should allow any VLAN TPID, but it is greyish
since it is HW dependent.

> To match untagged ipv4 packets, the pattern "eth type is 0x0800 /
> ipv4 / end" should be used.

What about eth / ipv4 / end?
Does usage of ipv4 assume that EtherType is 0x0800?

> To match both tagged and untagged packets, the pattern "eth / end"
> should be used.

The interesting question is what should be used if I want
either tagged or untagged IPv4 packets. I think it worse
to mention to make the picture complete.

> Signed-off-by: Dekel Peled <dekelp@mellanox.com>
> ---
>  doc/guides/prog_guide/rte_flow.rst | 8 ++++++++
>  lib/librte_ethdev/rte_flow.h       | 9 +++++++++
>  2 files changed, 17 insertions(+)
> 
> diff --git a/doc/guides/prog_guide/rte_flow.rst b/doc/guides/prog_guide/rte_flow.rst
> index cf4368e..0d1c305 100644
> --- a/doc/guides/prog_guide/rte_flow.rst
> +++ b/doc/guides/prog_guide/rte_flow.rst
> @@ -905,6 +905,12 @@ so-called layer 2.5 pattern items such as ``RTE_FLOW_ITEM_TYPE_VLAN``. In
>  the latter case, ``type`` refers to that of the outer header, with the inner
>  EtherType/TPID provided by the subsequent pattern item. This is the same
>  order as on the wire.
> +If the ``type`` field contains a TPID value, then only tagged packets will match
> +the pattern.

Shouldn't we emphasis that "tagged packets with specified TPID
will match the pattern." since tagged packets could have
various TPIDs.

> +If the ``type`` field contains another EtherType value, then only untagged
> +packets will match the pattern.

I'm afraid "another EtherType" is too ambiguous.
"non-TPID EtherType" is ambiguous as well and HW
dependent. May be it is better to remove the sentence
completely.

> +If the ``ETH`` item is the only item in the pattern, and the ``type`` field is
> +not specified, then both tagged and untagged packets will match the pattern.
>  
>  - ``dst``: destination MAC.
>  - ``src``: source MAC.
> @@ -919,6 +925,8 @@ Matches an 802.1Q/ad VLAN tag.
>  The corresponding standard outer EtherType (TPID) values are
>  ``RTE_ETHER_TYPE_VLAN`` or ``RTE_ETHER_TYPE_QINQ``. It can be overridden by the
>  preceding pattern item.
> +If a ``VLAN`` item is present in the pattern, then only tagged packets will
> +match the pattern.
>  
>  - ``tci``: tag control information.
>  - ``inner_type``: inner EtherType or TPID.
> diff --git a/lib/librte_ethdev/rte_flow.h b/lib/librte_ethdev/rte_flow.h
> index 132b44e..178e87e 100644
> --- a/lib/librte_ethdev/rte_flow.h
> +++ b/lib/librte_ethdev/rte_flow.h
> @@ -710,6 +710,13 @@ struct rte_flow_item_raw {
>   * the latter case, @p type refers to that of the outer header, with the
>   * inner EtherType/TPID provided by the subsequent pattern item. This is the
>   * same order as on the wire.
> + * If the @p type field contains a TPID value, then only tagged packets will
> + * match the pattern.
> + * If the @p type field contains another EtherType value, then only untagged
> + * packets will match the pattern.
> + * If the @p ETH item is the only item in the pattern, and the @p type field
> + * is not specified, then both tagged and untagged packets will match the
> + * pattern.
>   */
>  struct rte_flow_item_eth {
>  	struct rte_ether_addr dst; /**< Destination MAC. */
> @@ -734,6 +741,8 @@ struct rte_flow_item_eth {
>   * The corresponding standard outer EtherType (TPID) values are
>   * RTE_ETHER_TYPE_VLAN or RTE_ETHER_TYPE_QINQ. It can be overridden by
>   * the preceding pattern item.
> + * If a @p VLAN item is present in the pattern, then only tagged packets will
> + * match the pattern.
>   */
>  struct rte_flow_item_vlan {
>  	rte_be16_t tci; /**< Tag control information. */
>
  

Thanks, PSB.

> -----Original Message-----
> From: Andrew Rybchenko <arybchenko@solarflare.com>
> Sent: Saturday, April 25, 2020 5:00 PM
> To: Dekel Peled <dekelp@mellanox.com>; Ori Kam <orika@mellanox.com>;
> john.mcnamara@intel.com; marko.kovacevic@intel.com; Thomas Monjalon
> <thomas@monjalon.net>; ferruh.yigit@intel.com
> Cc: dev@dpdk.org; Asaf Penso <asafp@mellanox.com>
> Subject: Re: [PATCH] doc: refine ethernet and VLAN flow rule items
> 
> On 4/23/20 9:30 PM, Dekel Peled wrote:
> > Specified pattern may be translated in different manner.
> > For example the pattern "eth / ipv4" can be translated to match
> > untagged packets only, since the pattern doesn't specify a vlan item.
> 
> vlan -> VLAN

I will change to uppercase.

> 
> > It can also be translated to match both tagged and untagged packets,
> > for the same reason.
> > This patch updates the rte_flow documentation to clearly specify the
> > required pattern to use.
> > For example:
> > To match tagged ipv4 packets, the pattern "eth type is 0x8100 / vlan /
> > ipv4 / end" should be used.
> 
> Isn't eth / vlan / ipv4 /end sufficient? What's the difference?
> I guess later should allow any VLAN TPID, but it is greyish since it is HW
> dependent.

In the example I wanted to show explicit rule, to emphasize the importance of detailed pattern structure.

> 
> > To match untagged ipv4 packets, the pattern "eth type is 0x0800 /
> > ipv4 / end" should be used.
> 
> What about eth / ipv4 / end?
> Does usage of ipv4 assume that EtherType is 0x0800?

Same as above.

> 
> > To match both tagged and untagged packets, the pattern "eth / end"
> > should be used.
> 
> The interesting question is what should be used if I want either tagged or
> untagged IPv4 packets. I think it worse to mention to make the picture
> complete.

To match any IPV4 packet, either tagged or untagged, need to apply two rules.
One for tagged packets and the other for untagged packets.
I will add this example as well.

> 
> > Signed-off-by: Dekel Peled <dekelp@mellanox.com>
> > ---
> >  doc/guides/prog_guide/rte_flow.rst | 8 ++++++++
> >  lib/librte_ethdev/rte_flow.h       | 9 +++++++++
> >  2 files changed, 17 insertions(+)
> >
> > diff --git a/doc/guides/prog_guide/rte_flow.rst
> > b/doc/guides/prog_guide/rte_flow.rst
> > index cf4368e..0d1c305 100644
> > --- a/doc/guides/prog_guide/rte_flow.rst
> > +++ b/doc/guides/prog_guide/rte_flow.rst
> > @@ -905,6 +905,12 @@ so-called layer 2.5 pattern items such as
> > ``RTE_FLOW_ITEM_TYPE_VLAN``. In  the latter case, ``type`` refers to
> > that of the outer header, with the inner  EtherType/TPID provided by
> > the subsequent pattern item. This is the same  order as on the wire.
> > +If the ``type`` field contains a TPID value, then only tagged packets
> > +will match the pattern.
> 
> Shouldn't we emphasis that "tagged packets with specified TPID will match
> the pattern." since tagged packets could have various TPIDs.

Agree, I will update.

> 
> > +If the ``type`` field contains another EtherType value, then only
> > +untagged packets will match the pattern.
> 
> I'm afraid "another EtherType" is too ambiguous.
> "non-TPID EtherType" is ambiguous as well and HW dependent. May be it is
> better to remove the sentence completely.

I think this sentence is important in order to emphasize the untagged packets case.
How about "Otherwise, only untagged packets will match the pattern."?

> 
> > +If the ``ETH`` item is the only item in the pattern, and the ``type``
> > +field is not specified, then both tagged and untagged packets will match
> the pattern.
> >
> >  - ``dst``: destination MAC.
> >  - ``src``: source MAC.
> > @@ -919,6 +925,8 @@ Matches an 802.1Q/ad VLAN tag.
> >  The corresponding standard outer EtherType (TPID) values are
> > ``RTE_ETHER_TYPE_VLAN`` or ``RTE_ETHER_TYPE_QINQ``. It can be
> > overridden by the  preceding pattern item.
> > +If a ``VLAN`` item is present in the pattern, then only tagged
> > +packets will match the pattern.
> >
> >  - ``tci``: tag control information.
> >  - ``inner_type``: inner EtherType or TPID.
> > diff --git a/lib/librte_ethdev/rte_flow.h
> > b/lib/librte_ethdev/rte_flow.h index 132b44e..178e87e 100644
> > --- a/lib/librte_ethdev/rte_flow.h
> > +++ b/lib/librte_ethdev/rte_flow.h
> > @@ -710,6 +710,13 @@ struct rte_flow_item_raw {
> >   * the latter case, @p type refers to that of the outer header, with the
> >   * inner EtherType/TPID provided by the subsequent pattern item. This is
> the
> >   * same order as on the wire.
> > + * If the @p type field contains a TPID value, then only tagged
> > + packets will
> > + * match the pattern.
> > + * If the @p type field contains another EtherType value, then only
> > + untagged
> > + * packets will match the pattern.
> > + * If the @p ETH item is the only item in the pattern, and the @p
> > + type field
> > + * is not specified, then both tagged and untagged packets will match
> > + the
> > + * pattern.
> >   */
> >  struct rte_flow_item_eth {
> >  	struct rte_ether_addr dst; /**< Destination MAC. */ @@ -734,6
> +741,8
> > @@ struct rte_flow_item_eth {
> >   * The corresponding standard outer EtherType (TPID) values are
> >   * RTE_ETHER_TYPE_VLAN or RTE_ETHER_TYPE_QINQ. It can be
> overridden by
> >   * the preceding pattern item.
> > + * If a @p VLAN item is present in the pattern, then only tagged
> > + packets will
> > + * match the pattern.
> >   */
> >  struct rte_flow_item_vlan {
> >  	rte_be16_t tci; /**< Tag control information. */
> >
  

On Sun, 26 Apr 2020 09:18:54 +0000
Dekel Peled <dekelp@mellanox.com> wrote:

> Thanks, PSB.
> 
> > -----Original Message-----
> > From: Andrew Rybchenko <arybchenko@solarflare.com>
> > Sent: Saturday, April 25, 2020 5:00 PM
> > To: Dekel Peled <dekelp@mellanox.com>; Ori Kam <orika@mellanox.com>;
> > john.mcnamara@intel.com; marko.kovacevic@intel.com; Thomas Monjalon
> > <thomas@monjalon.net>; ferruh.yigit@intel.com
> > Cc: dev@dpdk.org; Asaf Penso <asafp@mellanox.com>
> > Subject: Re: [PATCH] doc: refine ethernet and VLAN flow rule items
> > 
> > On 4/23/20 9:30 PM, Dekel Peled wrote:  
> > > Specified pattern may be translated in different manner.
> > > For example the pattern "eth / ipv4" can be translated to match
> > > untagged packets only, since the pattern doesn't specify a vlan item.  
> > 
> > vlan -> VLAN  
> 
> I will change to uppercase.
> 
> >   
> > > It can also be translated to match both tagged and untagged packets,
> > > for the same reason.
> > > This patch updates the rte_flow documentation to clearly specify the
> > > required pattern to use.
> > > For example:
> > > To match tagged ipv4 packets, the pattern "eth type is 0x8100 / vlan /
> > > ipv4 / end" should be used.  
> > 
> > Isn't eth / vlan / ipv4 /end sufficient? What's the difference?
> > I guess later should allow any VLAN TPID, but it is greyish since it is HW
> > dependent.  
> 
> In the example I wanted to show explicit rule, to emphasize the importance of detailed pattern structure.
> 
> >   
> > > To match untagged ipv4 packets, the pattern "eth type is 0x0800 /
> > > ipv4 / end" should be used.  
> > 
> > What about eth / ipv4 / end?
> > Does usage of ipv4 assume that EtherType is 0x0800?  
> 
> Same as above.
> 
> >   
> > > To match both tagged and untagged packets, the pattern "eth / end"
> > > should be used.  
> > 
> > The interesting question is what should be used if I want either tagged or
> > untagged IPv4 packets. I think it worse to mention to make the picture
> > complete.  
> 
> To match any IPV4 packet, either tagged or untagged, need to apply two rules.
> One for tagged packets and the other for untagged packets.
> I will add this example as well.
> 
> >   
> > > Signed-off-by: Dekel Peled <dekelp@mellanox.com>

All this reminds me that "code is the best documentation" and there
is no working code that does a complete software implementation of the
rte_flow engine. This means the actual interpretation of what the rte_flow
means is left to documentation and hardware vendors choices in implementation.
  

Hi Stephen,

> -----Original Message-----
> From: Stephen Hemminger <stephen@networkplumber.org>
> Sent: Sunday, April 26, 2020 8:02 PM
> To: Dekel Peled <dekelp@mellanox.com>
> Cc: Andrew Rybchenko <arybchenko@solarflare.com>; Ori Kam
> <orika@mellanox.com>; john.mcnamara@intel.com;
> marko.kovacevic@intel.com; Thomas Monjalon <thomas@monjalon.net>;
> ferruh.yigit@intel.com; dev@dpdk.org; Asaf Penso <asafp@mellanox.com>
> Subject: Re: [dpdk-dev] [PATCH] doc: refine ethernet and VLAN flow rule items
> 
> On Sun, 26 Apr 2020 09:18:54 +0000
> Dekel Peled <dekelp@mellanox.com> wrote:
> 
> > Thanks, PSB.
> >
> > > -----Original Message-----
> > > From: Andrew Rybchenko <arybchenko@solarflare.com>
> > > Sent: Saturday, April 25, 2020 5:00 PM
> > > To: Dekel Peled <dekelp@mellanox.com>; Ori Kam <orika@mellanox.com>;
> > > john.mcnamara@intel.com; marko.kovacevic@intel.com; Thomas Monjalon
> > > <thomas@monjalon.net>; ferruh.yigit@intel.com
> > > Cc: dev@dpdk.org; Asaf Penso <asafp@mellanox.com>
> > > Subject: Re: [PATCH] doc: refine ethernet and VLAN flow rule items
> > >
> > > On 4/23/20 9:30 PM, Dekel Peled wrote:
> > > > Specified pattern may be translated in different manner.
> > > > For example the pattern "eth / ipv4" can be translated to match
> > > > untagged packets only, since the pattern doesn't specify a vlan item.
> > >
> > > vlan -> VLAN
> >
> > I will change to uppercase.
> >
> > >
> > > > It can also be translated to match both tagged and untagged packets,
> > > > for the same reason.
> > > > This patch updates the rte_flow documentation to clearly specify the
> > > > required pattern to use.
> > > > For example:
> > > > To match tagged ipv4 packets, the pattern "eth type is 0x8100 / vlan /
> > > > ipv4 / end" should be used.
> > >
> > > Isn't eth / vlan / ipv4 /end sufficient? What's the difference?
> > > I guess later should allow any VLAN TPID, but it is greyish since it is HW
> > > dependent.
> >
> > In the example I wanted to show explicit rule, to emphasize the importance of
> detailed pattern structure.
> >
> > >
> > > > To match untagged ipv4 packets, the pattern "eth type is 0x0800 /
> > > > ipv4 / end" should be used.
> > >
> > > What about eth / ipv4 / end?
> > > Does usage of ipv4 assume that EtherType is 0x0800?
> >
> > Same as above.
> >
> > >
> > > > To match both tagged and untagged packets, the pattern "eth / end"
> > > > should be used.
> > >
> > > The interesting question is what should be used if I want either tagged or
> > > untagged IPv4 packets. I think it worse to mention to make the picture
> > > complete.
> >
> > To match any IPV4 packet, either tagged or untagged, need to apply two
> rules.
> > One for tagged packets and the other for untagged packets.
> > I will add this example as well.
> >
> > >
> > > > Signed-off-by: Dekel Peled <dekelp@mellanox.com>
> 
> All this reminds me that "code is the best documentation" and there
> is no working code that does a complete software implementation of the
> rte_flow engine. This means the actual interpretation of what the rte_flow
> means is left to documentation and hardware vendors choices in
> implementation.

I agree with you, that is why I think this patch is important. 

Best,
Ori