From: Shreyansh Jain <shreyansh.jain@nxp.com>
To: Thomas Monjalon <thomas@monjalon.net>, <dev@dpdk.org>
Cc: Yuanhan Liu <yliu@fridaylinux.org>,
Adrien Mazarguil <adrien.mazarguil@6wind.com>,
Ciara Loftus <ciara.loftus@intel.com>,
Kevin Traynor <ktraynor@redhat.com>, <stephen@networkplumber.org>,
<ferruh.yigit@intel.com>
Subject: Re: [dpdk-dev] standardize device identification
Date: Fri, 22 Dec 2017 12:31:03 +0530 [thread overview]
Message-ID: <023a90af-bfcd-328d-4ed9-783ee76203fd@nxp.com> (raw)
In-Reply-To: <7044959.u7szEIarlR@xps>
On Thursday 21 December 2017 03:32 AM, Thomas Monjalon wrote:
> Changing the title and adding more comments inline:
>
> 19/12/2017 00:05, Thomas Monjalon:
>> Let's summarize and resume this thread.
>>
>> We need a generic syntax to describe a device.
>> This syntax can be used
>> - before initializing the device (i.e. whitelist/blacklist)
>> - or after the initialization (e.g. user config)
>>
>> We need to answer 4 questions:
>> 1/ what are the separators (comma, colon, etc)?
>> 2/ how to distinguish a device identification from a configuration?
>> 3/ what are the mandatory parts?
>> 4/ what can be the optional properties?
>>
>> 30/11/2017 08:35, Yuanhan Liu:
>>> What this patch proposes is to use "name[,mac]" syntax. "name" is the
>>> PCI id for pci device. For vdev, it's the vdev name given by user. The
>>> reason "mac" is needed is for some devices (say ConnectX-3), 2 ports
>>> (in a single NIC) have the same PCI id.
>>
>> Based on the feedbacks we had, I suggest a syntax where everything is
>> optional key/value pairs, and split in 3 categories:
>> - bus (pci, vdev, vmbus, fslmc, etc)
>> - class (eth, crypto)
>> - driver (i40e, mlx5, virtio, etc)
>
> The key/value pair describing the category scope is mandatory
> and must be the first pair in the category properties.
> Example: bus=pci, must be placed before id=0000:01:00.0
>
>> Between categories, the separator is a slash.
Why is a '/' required as a separator? Are you expecting the key in
key,value pair to duplicate across categories?
>> Inside a category, the separator is a comma.
>> Inside a key/value pair, the separator is an equal sign.
sounds reasonable to me.
>>
>> It may look like this:
>> bus=BUS_NAME,id=BUS_ID/class=CLASS_NAME,dev_port=PORT_NUM,mac=MAC_ADDRESS/driver=DRIVER_NAME,driverspecificproperty=VALUE
If I take cue from fslmc and dpaa bus:
for fslmc: bus=fslmc,id=dpni.1
bus=fslmc,id=dpsec.1
for dpaa: bus=dpaa,id=fm1-mac1
bus=dpaa,id=dpaa-sec1
So, at least from fslmc/dpaa perspective, above fits.
Just want to highlight: in some cases the device names can contain ',' -
but then, that can be handled at bus scan level.
For dpaa bus, the device identified from platform identifiers contains a
longer ',' separated string - which is then stripped to form a name like
'fm1-mac1' before being added to device->name.
>>
>> A device is identified when every properties are matched.
So, a ovs-dpdk user would have to know a dpdk bus to identify a device
to plug into a OVS bridge?
>> Before device is probed, only the bus category is relevant.
>> For the simple PCI whitelist, it means moving from
>> -w 0000:01:00.0
>> to
>> -w bus=pci,id=0000:01:00.0
OK
>>
>> It is possible to mix some settings in these devargs syntax if the keys
>> are differents. Example: mac= is for identification by MAC, whereas
>> newmac= would be for specifying a MAC address to set.
>>
>> Agreement?
in principle, yes.
just need clarity on '/' as a separator.
>
> Yuanhan is proposing to use this syntax in OVS option dpdk-devargs:
> https://mail.openvswitch.org/pipermail/ovs-dev/2017-December/342273.html
>
> Please, any feedback or approval that this syntax is good?
>
next prev parent reply other threads:[~2017-12-22 6:47 UTC|newest]
Thread overview: 28+ messages / expand[flat|nested] mbox.gz Atom feed top
2017-11-30 7:35 [dpdk-dev] [PATCH] [RFC] ether: standardize getting the port by name Yuanhan Liu
2017-11-30 17:15 ` Stephen Hemminger
2017-11-30 17:35 ` Thomas Monjalon
2017-11-30 21:21 ` Stephen Hemminger
2017-11-30 21:44 ` Thomas Monjalon
2017-12-01 9:47 ` Gaëtan Rivet
2017-12-04 13:55 ` Yuanhan Liu
2017-12-05 11:04 ` Adrien Mazarguil
2017-12-05 13:20 ` Thomas Monjalon
2017-12-05 13:58 ` Yuanhan Liu
2017-12-05 15:28 ` Thomas Monjalon
2017-12-05 17:22 ` Adrien Mazarguil
2017-12-06 15:49 ` Yuanhan Liu
2017-12-18 22:25 ` Thomas Monjalon
2017-12-18 22:30 ` Stephen Hemminger
2017-12-18 22:41 ` Thomas Monjalon
2017-12-18 23:05 ` Thomas Monjalon
2017-12-20 22:02 ` [dpdk-dev] standardize device identification Thomas Monjalon
2017-12-22 7:01 ` Shreyansh Jain [this message]
2017-12-22 9:00 ` Thomas Monjalon
2018-01-05 7:52 ` Finn Christensen
2018-01-05 8:39 ` Thomas Monjalon
2018-01-05 11:09 ` Finn Christensen
2018-01-05 12:01 ` Thomas Monjalon
2018-01-05 14:14 ` Finn Christensen
2018-01-05 15:34 ` Thomas Monjalon
2018-01-05 20:32 ` Finn Christensen
2018-01-16 20:20 ` [dpdk-dev] [PATCH] [RFC] ether: standardize getting the port by name Ferruh Yigit
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=023a90af-bfcd-328d-4ed9-783ee76203fd@nxp.com \
--to=shreyansh.jain@nxp.com \
--cc=adrien.mazarguil@6wind.com \
--cc=ciara.loftus@intel.com \
--cc=dev@dpdk.org \
--cc=ferruh.yigit@intel.com \
--cc=ktraynor@redhat.com \
--cc=stephen@networkplumber.org \
--cc=thomas@monjalon.net \
--cc=yliu@fridaylinux.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).