[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <42d0cfed-2f47-2d81-9654-aa9f017860ac@intel.com>
Date: Fri, 14 Jul 2017 03:31:01 -0700
From: "Nambiar, Amritha" <amritha.nambiar@...el.com>
To: Jamal Hadi Salim <jhs@...atatu.com>,
intel-wired-lan@...ts.osuosl.org, jeffrey.t.kirsher@...el.com
Cc: alexander.h.duyck@...el.com, kiran.patil@...el.com,
netdev@...r.kernel.org, mitch.a.williams@...el.com,
alexander.duyck@...il.com, neerav.parikh@...el.com,
sridhar.samudrala@...el.com, carolyn.wyborny@...el.com,
Roman Mashak <mrv@...atatu.com>
Subject: Re: [PATCH 0/6] Configuring traffic classes via new hardware offload
mechanism in tc/mqprio
On 7/14/2017 12:56 AM, Jamal Hadi Salim wrote:
> Hi Amritha,
>
> On 17-07-11 06:18 AM, Amritha Nambiar wrote:
>> The following series introduces a new hardware offload mode in
>> tc/mqprio where the TCs, the queue configurations and
>> bandwidth rate limits are offloaded to the hardware. The existing
>> mqprio framework is extended to configure the queue counts and
>> layout and also added support for rate limiting. This is achieved
>> by setting the value 2 for the "hw" mode. Legacy devices can fall
>> back to the existing setup supporting hw mode 1 where only the
>> TCs are offloaded, however if hw mode 2 is specified, then this
>> type of offload fails to initialize if the requested mode is not
>> supported. The i40e driver enables the new mqprio hardware offload
>> mechanism factoring the TCs, queue configuration and bandwidth
>> rates by creating HW channel VSIs.
>>
>> In this new mode, the priority to traffic class mapping and the
>> user specified queue ranges are used to configure the traffic
>> class when the 'hw' option is set to 2. This is achieved by
>> creating HW channels(VSI). A new channel is created for each
>> of the traffic class configuration offloaded via mqprio
>> framework except for the first TC (TC0) which is for the main
>> VSI. TC0 for the main VSI is also reconfigured as per user
>> provided queue parameters. Finally, bandwidth rate limits are
>> set on these traffic classes through the mqprio offload
>> framework by sending these rates in addition to the number of
>> TCs and the queue configurations.
>>
>> Example:
>> # tc qdisc add dev eth0 root mqprio num_tc 2 map 0 0 0 0 1 1 1 1\
>> queues 4@0 4@4 min_rate 0Mbit 0Mbit max_rate 55Mbit 60Mbit hw 2
>>
>> To dump the bandwidth rates:
>>
>> # tc qdisc show dev eth0
>> qdisc mqprio 804a: root tc 2 map 0 0 0 0 1 1 1 1 0 0 0 0 0 0 0 0
>> queues:(0:3) (4:7)
>> min rates:0bit 0bit
>> max rates:55Mbit 60Mbit
>
> My only concern is usability. Was already cryptic with hw "1" and
> now we introduce another magic number "2". Could we not have two
> have some human friendly nouns? I know 1 means offload qos and two
> mean offload rate (+qos?).
I agree this could be less friendly, I was trying to retain the existing
conventions. Yes, hw "2" offloads qos and rates.
>
> I have some questions:
> BTW, what would tc class show dev eth0 display in this case?
> Above seems to select queue groupings for a specific rates which
> is a short-cut to say assigning each queue via something like TBF
> via "tc class" semantics (but way more cryptic than tc class;->)
Dumping the classes under mqprio qdisc will not display any of the
queue configurations or rates. MQPRIO is classless and does not support
change_class operation like HTB does to add a class to the qdisc and set
a rate for the class.
>
> cheers,
> jamal
>
>
Powered by blists - more mailing lists