Message-ID: <20180628132241.GA2177@nanopsycho.orion>
Date:   Thu, 28 Jun 2018 15:22:41 +0200
From:   Jiri Pirko <jiri@...nulli.us>
To:     Jamal Hadi Salim <jhs@...atatu.com>
Cc:     netdev@...r.kernel.org, davem@...emloft.net,
        xiyou.wangcong@...il.com, jakub.kicinski@...ronome.com,
        simon.horman@...ronome.com, john.hurley@...ronome.com,
        dsahern@...il.com, mlxsw@...lanox.com
Subject: Re: [patch net-next v2 0/9] net: sched: introduce chain templates
 support with offloading to mlxsw

Thu, Jun 28, 2018 at 03:13:30PM CEST, jhs@...atatu.com wrote:
>
>On 26/06/18 03:59 AM, Jiri Pirko wrote:
>> From: Jiri Pirko <jiri@...lanox.com>
>> 
>> For the TC clsact offload these days, some of HW drivers need
>> to hold a magic ball. The reason is, with the first inserted rule inside
>> HW they need to guess what fields will be used for the matching. If
>> later on this guess proves to be wrong and user adds a filter with a
>> different field to match, there's a problem. Mlxsw resolves it now with
>> couple of patterns. Those try to cover as many match fields as possible.
>> This aproach is far from optimal, both performance-wise and scale-wise.
>> Also, there is a combination of filters that in certain order won't
>> succeed.
>> 
>>
>> Most of the time, when user inserts filters in chain, he knows right away
>> how the filters are going to look like - what type and option will they
>> have. For example, he knows that he will only insert filters of type
>> flower matching destination IP address. He can specify a template that
>> would cover all the filters in the chain.
>> 
>
>Is this just restricted to hardware offload? Example it will make sense
>for u32 in s/ware as well (i.e flexible TCAM like TCAM based
>classification). i.e it is possible that rules the user enters
>end up being worst case a linked list lookup, yes? And allocating
>space for a tuple that is not in use is a waste of space.

I'm afraid I don't understand clearly what you say. This is not
restricted to hw offload. The templates apply to all filters, no matter
if they are offloaded or not.


>
>If yes, then I would reword the above as something like:
>
>For very flexible classifiers such as TCAM based ones,
>one could add arbitrary tuple rules which tend to be inefficient both
>from a space and lookup performance. One approach, taken by Mlxsw,
>is to assume a multi filter tuple arrangement which is inefficient
>from a space perspective when the user-specified rules dont make
>use of pre-provisioned tuple space.
>Typically users already know what tuples are of interest to them:
>for example for ipv4 route lookup purposes they may just want to
>lookup destination IP with a specified mask etc.
>This feature allows user to provide good hints to the classifier to
>optimize.
>
>
>> This patchset is providing the possibility to user to provide such
>> template  to kernel and propagate it all the way down to device
>> drivers.
>> 
>> See the examples below.
>> 
>> Create dummy device with clsact first:
>> # ip link add type dummy
>> # tc qdisc add dev dummy0 clsact
>> 
>> There is no template assigned by default:
>> # tc filter template show dev dummy0 ingress
>> 
>> Add a template of type flower allowing to insert rules matching on last
>> 2 bytes of destination mac address:
>> # tc filter template add dev dummy0 ingress proto ip flower dst_mac 00:00:00:00:00:00/00:00:00:00:FF:FF
>> 
>> The template is now showed in the list:
>> # tc filter template show dev dummy0 ingress
>> filter flower chain 0
>>    dst_mac 00:00:00:00:00:00/00:00:00:00:ff:ff
>>    eth_type ipv4
>> 
>> Add another template, this time for chain number 22:
>> # tc filter template add dev dummy0 ingress proto ip chain 22 flower dst_ip 0.0.0.0/16
>> # tc filter template show dev dummy0 ingress
>> filter flower chain 0
>>    dst_mac 00:00:00:00:00:00/00:00:00:00:ff:ff
>>    eth_type ipv4
>> filter flower chain 22
>>    eth_type ipv4
>>    dst_ip 0.0.0.0/16
>> 
>> Add a filter that fits the template:
>> # tc filter add dev dummy0 ingress proto ip flower dst_mac aa:bb:cc:dd:ee:ff/00:00:00:00:00:0F action drop
>> 
>> Addition of filters that does not fit the template would fail:
>> # tc filter add dev dummy0 ingress proto ip flower dst_mac aa:11:22:33:44:55/00:00:00:FF:00:00 action drop
>> Error: Mask does not fit the template.
>> We have an error talking to the kernel, -1
>> # tc filter add dev dummy0 ingress proto ip flower dst_ip 10.0.0.1 action drop
>> Error: Mask does not fit the template.
>> We have an error talking to the kernel, -1
>> 
>> Additions of filters to chain 22:
>> # tc filter add dev dummy0 ingress proto ip chain 22 flower dst_ip 10.0.0.1/8 action drop
>> # tc filter add dev dummy0 ingress proto ip chain 22 flower dst_ip 10.0.0.1 action drop
>> Error: Mask does not fit the template.
>> We have an error talking to the kernel, -1
>> # tc filter add dev dummy0 ingress proto ip chain 22 flower dst_ip 10.0.0.1/24 action drop
>> Error: Mask does not fit the template.
>> We have an error talking to the kernel, -1
>> 
>> Removal of a template from non-empty chain would fail:
>> # tc filter template del dev dummy0 ingress
>> Error: The chain is not empty, unable to delete template.
>> We have an error talking to the kernel, -1
>> 
>> Once the chain is flushed, the template could be removed:
>> # tc filter del dev dummy0 ingress
>> # tc filter template del dev dummy0 ingress
>> 
>
>BTW: unlike the other comments on this - I think the syntax above
>is fine ;-> Chain are already either explicitly or are implicitly
>(case of chain 0) specified.
>
>Assuming that one cant add a new template to a chain if it already
>has at least one filter (even if no template has been added).
>
>I like it - it may help making u32 more friendly to humans in some
>cases.
>
>cheers,
>jamal

Powered by blists - more mailing lists