netdev - Re: [RFC v2 net-next 06/10] net/sched: Introduce the TBS Qdisc

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <5015404f-3883-61d3-f686-bd54c650e9ea@intel.com>
Date:   Tue, 23 Jan 2018 14:01:39 -0800
From:   Jesus Sanchez-Palencia <jesus.sanchez-palencia@...el.com>
To:     Jamal Hadi Salim <jhs@...atatu.com>, netdev@...r.kernel.org
Cc:     xiyou.wangcong@...il.com, jiri@...nulli.us,
        vinicius.gomes@...el.com, richardcochran@...il.com,
        intel-wired-lan@...ts.osuosl.org, anna-maria@...utronix.de,
        henrik@...tad.us, tglx@...utronix.de, john.stultz@...aro.org,
        andre.guedes@...el.com, ivan.briano@...el.com,
        levi.pearson@...man.com
Subject: Re: [RFC v2 net-next 06/10] net/sched: Introduce the TBS Qdisc

Hi,


On 01/18/2018 05:35 AM, Jamal Hadi Salim wrote:
> On 18-01-17 06:06 PM, Jesus Sanchez-Palencia wrote:
>> From: Vinicius Costa Gomes <vinicius.gomes@...el.com>
>>
>> TBS (Time Based Scheduler) uses the information added earlier in this
>> series (the socket option SO_TXTIME and the new role of
>> sk_buff->tstamp) to schedule traffic transmission based on absolute
>> time.
>>
>> For some workloads, just bandwidth enforcement is not enough, and
>> precise control of the transmission of packets is necessary.
>>
>> Example:
>>
>> $ tc qdisc replace dev enp2s0 parent root handle 100 mqprio num_tc 3 \
> 
> handle 100:0 ?
> 
>>             map 2 2 1 0 2 2 2 2 2 2 2 2 2 2 2 2 queues 1@0 1@1 2@2 hw 0
>>
>> $ tc qdisc add dev enp2s0 parent 100:1 tbs delta 60000 clockid 11 offload 1
>>
>>
>> In this example, the Qdisc will try to enable offloading (offload 1)
>> the control of the transmission time to the network adapter, the
>> time stamp in socket are in reference to the clockid '11' (CLOCK_TAI)
>> and packets leave the Qdisc "delta" (60000) nanoseconds before its
>> transmission time.
>>
> 
> 
>> When offloading is disabled, the network adapter will ignore the
>> sk_buff time stamp, and so, the transmission time will be only "best
>> effort" from the Qdisc.
>>
> 
> General comments:
> 1) iproute2: Avoid magic numbers like 1 or 11 please; "offload"
> (without 1) and "TAI" will be more human friendly.


Sure, we'll change both parameters.


> 
> 2) Experience shows that adding padding fields in the control structs
> implies they will _never ever_ be used. That was not design intent
> for netlink but over years shit like that has happened.
> Maybe look at using a 32 bitmap? It is more "future proof".
> You seem to only have 2-3 flags but it gives you opportunity
> to add more changes later. If you are 100% sure youll never need
> it - then maybe just move the tc_tbs_qopt::offload to the end of
> of the struct.


Ok, we are looking into it.


> 
> 3)It would be helpful for debugging to increment some stats other
> than drop counters on enqueu/dequeue obsolete packet drop. Maybe use
> overlimits for the dequeu drops (in addition)?


Yes, sure, that's a good idea.


> 
> 4) I may be misreading things - but did you need to reset the
> watchdog on dequeue? It is already being kicked for every incoming packet.


The watchdog timer must always be set to the next deadline. Since both enqueue()
and dequeue() modify the tree structure and may have it rebalanced, we need to
make sure the watchdog has been 're-set' to the next deadline. i.e. After a
dequeue, we need to re-arm the timer for the next head of the queue, so that is
why. Does that make sense?

I'm now thinking that naming that helper as reset_watchdog() can be misleading,
so we are considering naming it as rearm_watchdog() instead.


Thanks,
Jesus


> 
> cheers,
> jamal