Hi Ori,

On 6/2/21 4:32 PM, Ori Kam wrote:

Hi Ivan,

quoted

-----Original Message-----
From: Ivan Malov <redacted>

Hi Ori,

Your review efforts are much appreciated. I understand your concern
about the partial item/action coverage, but there are some points to be
considered when addressing it:
- It's anyway hardly possible to use the printed flow directly in
testpmd if it contains "opaque", or "PMD-specific", items/actions in
terms of the tunnel offload model. These items/actions have to be
omitted when printing the flow, and their absence in the resulting
string means that copy/pasting the flow to testpmd isn't helpful in this
particular case.

I fully agree with you that some of the rules can't be printed. That is why.
I'm not sure having partial solution is the way to go.

Sorry, I disagree that possibility to cover 99% and
impossibility to cover just 1% is the reason to discard.

If OVS for example cares about
some of the item/action, maybe this log should be on their part.

OvS does and as far as I can see already has bugs there.
Of course, nobody says that it is testpmd-complient format,
but it definitely looks so.

Anyway, it sounds strange do duplicate the functionality in
many-many DPDK apps. Of course, it removes the headache
from DPDK maintainers, but it is hardly friendly to DPDK
community in general.

quoted

- There's action ENCAP which also can't be fully represented by the tool
in question, simply because it has no parameters. In tespmd, one first
has to issue "set vxlan" command to configure the encap. header, whilst
"vxlan" token in the flow rule string just refers to the previously set
encap. parameters. The suggested flow print helper can't reliably print
these two components ("set vxlan" and the flow rule itself) as they
belong to different testpmd command strings.

Again, I agree with you but like my above answer, do we want a partial solution 
in DPDK?

My answer is YES.

quoted

As you might see, completeness of the solution wouldn't necessarily be
reachable, even if full item/action coverage was provided.

As for the item/action coverage itself, it's rather controversial. On
the one hand, yes, we should probably try to cover more items and
actions in the suggested patch, to the extent allowed by our current
priorities. But on the other hand, the existing coverage might not be
that poor: it's fairly elaborate and at least allows to print the most
common flow rules.

That is my main issue you are going to push something that is good for you
and maybe some other cases, but it can't be used by all application, even with
the most basic commands like encap.

Isn't it fair: if one wants to use something, be ready to help
and extend it. No pain, no gain :) Jokes aside - we're ready to
support "the most basic commands", just list it, but do not say
everything is basic. "everything" will delay the feature and
simply unfair to demand (IMHO).

IMHO, the feature would make flow API more friendly and easier
to debug, report bugs etc.

quoted

Yes, macros and some other cunning ways to cover more flow specifics
might come in handy, but, at the same time, can be rather error prone.
Sometimes it's more robust to just write the code out in full.

I'm always in favor of easy of extra complex but too hard is also not good.

Thanks,
Ori

quoted

Thank you.

On 30/05/2021 10:27, Ori Kam wrote:

quoted

Hi Ivan,

First nice idea and thanks for the picking up the ball.

Before a detail review,
The main thing I'm concerned about is that this print will be partially

supported,

quoted

I know that you covered this issue by printing unknown for unsupported

item/actions,

quoted

but this will mean that it is enough that one item/action is not supported

and already the

quoted

flow can't be used in testpmd.
To get full support it means that the developer needs to add such print

with each new

quoted

item/action. I agree it is possible, but it has high overhead for each feature.

Maybe we should somehow create a macros for the prints or other easier

to support ways.

quoted

For example, just printing the ipv4 has 7 function calls inside of it each one

with error checking,

quoted

and I'm not counting the dedicated functions.



Best,
Ori

quoted

-----Original Message-----
From: Ivan Malov <redacted>
Sent: Thursday, May 27, 2021 11:25 AM
To: dev@dpdk.org
Cc: NBU-Contact-Thomas Monjalon <redacted>; Ferruh

Yigit

quoted

quoted

[off-list ref]; Andrew Rybchenko
[off-list ref]; Ori Kam [off-list ref]; Ray
Kinsella [off-list ref]; Neil Horman [off-list ref]
Subject: [RFC PATCH] ethdev: add support for testpmd-compliant flow

rule

quoted

quoted

dumping

DPDK applications (for example, OvS) or tests which use RTE flow API

need to

quoted

quoted

log created or rejected flow rules to help to recognise what goes right or
wrong. From this standpoint, testpmd-compliant format is nice for the
purpose because it allows to copy-paste the flow rules and debug using
testpmd.

Recognisable pattern items:
VOID, VF, PF, PHY_PORT, PORT_ID, ETH, VLAN, IPV4, IPV6, UDP, TCP,

VXLAN,

quoted

quoted

NVGRE, GENEVE, MARK, PPPOES, PPPOED.

Recognisable actions:
VOID, JUMP, MARK, FLAG, QUEUE, DROP, COUNT, RSS, PF, VF,

PHY_PORT,

quoted

quoted

PORT_ID, OF_POP_VLAN, OF_PUSH_VLAN, OF_SET_VLAN_VID,
OF_SET_VLAN_PCP, VXLAN_ENCAP, VXLAN_DECAP.

Recognisable RSS types (action RSS):
IPV4, FRAG_IPV4, NONFRAG_IPV4_TCP, NONFRAG_IPV4_UDP,
NONFRAG_IPV4_OTHER, IPV6, FRAG_IPV6, NONFRAG_IPV6_TCP,
NONFRAG_IPV6_UDP, NONFRAG_IPV6_OTHER, IPV6_EX, IPV6_TCP_EX,
IPV6_UDP_EX, L3_SRC_ONLY, L3_DST_ONLY, L4_SRC_ONLY,

L4_DST_ONLY.

quoted

quoted

Unrecognised parts of the flow specification are represented by tokens
"{unknown}" and "{unknown bits}". Interested parties are welcome to
extend this tool to recognise more items and actions.

Signed-off-by: Ivan Malov <redacted>