From: Pavel Vazharov <freakpv@gmail.com>
To: Stephen Hemminger <stephen@networkplumber.org>
Cc: users <users@dpdk.org>
Subject: Re: IPv4 flows per queue on top of RSS for ixgbe and i40e drivers
Date: Wed, 5 Feb 2025 18:53:32 +0200 [thread overview]
Message-ID: <CAK9EM1_1LkAvTSYVwCKBWeK2rft-rM4wk6PmuTx+Q4OHC0uGfA@mail.gmail.com> (raw)
In-Reply-To: <20250205084957.02e14a3e@hermes.local>
On Wed, Feb 5, 2025 at 6:50 PM Stephen Hemminger
<stephen@networkplumber.org> wrote:
>
> On Wed, 5 Feb 2025 18:21:33 +0200
> Pavel Vazharov <freakpv@gmail.com> wrote:
>
> > Hi there,
> >
> > We have a proxy application on top of DPDK where we use a symmetric
> > RSS key to receive the packets for a given connection in both
> > directions on a single queue. In addition to that, we've few IPv4
> > addresses where we need to receive their traffic on particular queues.
> > We use the rte flows functionality with RTE_FLOW_ACTION_TYPE_QUEUE to
> > achieve that and it seemed to work with the DPDK ixgbe driver.
> > However, today we tried the same application on top of DPDK i40e
> > driver and this setup doesn't seem to work there. It prints the
> > following errors:
> > `i40e_flow_add_del_fdir_filter(): Conflict with existing flow director rules!`
> > It seems that the i40e driver doesn't allow adding flow rules on top
> > of the already set RSS.
> >
> > Can somebody suggest a way to achieve what we need with i40e: to use a
> > symmetric RSS key for most of the traffic but to redirect the traffic
> > for a few specific IPv4 addresses to particular queues?
> >
> > Thanks,
> > Pavel.
>
> If you are mixing RSS and rte_flow the results are not well defined.
> Many drivers treat all active queues (including those used by rte_flow)
> as candidates for RSS.
>
> If you want to mix, the the safe way is:
> - don't enable RSS in the device config (rx_mode)
> - define an rte_flow rule with RTE_FLOW_ACTION_TYPE_RSS
> with a match all
> - define a rte_flow rule with RTE_FLOW_ACTION_TYPE_QUEUE
> that matches the IP
> and set rule priorities so that specific IP rule matches before
> the match all.
Thank you for this idea. I'll give it a try.
> You still maybe at risk of hardware quirks and driver incompatibilities
> which is part of the problem with rte_flow.
next prev parent reply other threads:[~2025-02-05 16:53 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-02-05 16:21 Pavel Vazharov
2025-02-05 16:49 ` Stephen Hemminger
2025-02-05 16:53 ` Pavel Vazharov [this message]
2025-02-11 7:45 ` Pavel Vazharov
2025-02-11 10:58 ` Pavel Vazharov
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=CAK9EM1_1LkAvTSYVwCKBWeK2rft-rM4wk6PmuTx+Q4OHC0uGfA@mail.gmail.com \
--to=freakpv@gmail.com \
--cc=stephen@networkplumber.org \
--cc=users@dpdk.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).