From: Vladimir Oltean <vladimir.oltean@nxp.com>
To: netdev@vger.kernel.org
Cc: "David S. Miller" <davem@davemloft.net>,
"Eric Dumazet" <edumazet@google.com>,
"Jakub Kicinski" <kuba@kernel.org>,
"Paolo Abeni" <pabeni@redhat.com>, "Andrew Lunn" <andrew@lunn.ch>,
"Florian Fainelli" <f.fainelli@gmail.com>,
"Luiz Angelo Daros de Luca" <luizluca@gmail.com>,
"Alvin Šipraga" <alsi@bang-olufsen.dk>,
"Madhuri Sripada" <madhuri.sripada@microchip.com>,
"Marcin Wojtas" <mw@semihalf.com>,
"Linus Walleij" <linus.walleij@linaro.org>,
"Tobias Waldekranz" <tobias@waldekranz.com>,
"Arun Ramadoss" <arun.ramadoss@microchip.com>,
"Russell King (Oracle)" <rmk+kernel@armlinux.org.uk>,
"Jonathan Corbet" <corbet@lwn.net>
Subject: [PATCH net 1/4] docs: net: dsa: document the tagger-owned storage mechanism
Date: Fri, 8 Dec 2023 21:35:15 +0200 [thread overview]
Message-ID: <20231208193518.2018114-2-vladimir.oltean@nxp.com> (raw)
In-Reply-To: <20231208193518.2018114-1-vladimir.oltean@nxp.com>
Introduced 2 years ago in commit dc452a471dba ("net: dsa: introduce
tagger-owned storage for private and shared data"), the tagger-owned
storage mechanism has recently sparked some discussions which denote a
general lack of developer understanding / awareness of it. There was
also a bug in the ksz switch driver which indicates the same thing.
Admittedly, it is also not obvious to see the design constraints that
led to the creation of such a complicated mechanism.
Here are some paragraphs that explain what it's about.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
---
Documentation/networking/dsa/dsa.rst | 59 ++++++++++++++++++++++++++++
1 file changed, 59 insertions(+)
diff --git a/Documentation/networking/dsa/dsa.rst b/Documentation/networking/dsa/dsa.rst
index 7b2e69cd7ef0..0c326a42eb81 100644
--- a/Documentation/networking/dsa/dsa.rst
+++ b/Documentation/networking/dsa/dsa.rst
@@ -221,6 +221,44 @@ receive all frames regardless of the value of the MAC DA. This can be done by
setting the ``promisc_on_conduit`` property of the ``struct dsa_device_ops``.
Note that this assumes a DSA-unaware conduit driver, which is the norm.
+Separation between tagging protocol and switch drivers
+------------------------------------------------------
+
+Sometimes it is desirable to test the behavior of a given conduit interface
+with a given switch protocol, to see how it responds to checksum offloading,
+padding with tail tags, increased MTU, how the hardware parser sees DSA-tagged
+frames, etc.
+
+To achieve that, any tagging protocol driver may be used with ``dsa_loop``
+(this requires modifying the ``dsa_loop_get_protocol()`` function
+implementation). Therefore, tagging protocol drivers must not assume that they
+are used only in conjunction with a particular switch driver. Concretely, the
+tagging protocol driver should make no assumptions about the type of
+``ds->priv``, and its core functionality should only rely on the data
+structures offered by the DSA core for all switches (``struct dsa_switch``,
+``struct dsa_port`` etc).
+
+Additionally, tagging protocol drivers must not depend on symbols exported by
+any particular switch control path driver. Doing so would create a circular
+dependency, because DSA, on behalf of the switch driver, already requests the
+appropriate tagging protocol driver module to be loaded.
+
+Nonetheless, there are exceptional situations when switch-specific processing
+is required in a tagging protocol driver. In some cases the tagger needs a
+place to hold state; in other cases, the packet transmission procedure may
+involve accessing switch registers. The tagger may also be processing packets
+which are not destined for the network stack but for the switch driver's
+management logic, and thus, the switch driver should have a handler for these
+management frames.
+
+A mechanism, called tagger-owned storage (in reference to ``ds->tagger_data``),
+exists, which permits tagging protocol drivers to allocate memory for each
+switch that they connect to. Each tagging protocol driver may define its own
+contract with switch drivers as to what this data structure contains.
+Through the ``struct dsa_device_ops`` methods ``connect()`` and ``disconnect()``,
+tagging protocol drivers are given the possibility to manage the
+``ds->tagger_data`` pointer of any switch that they connect to.
+
Conduit network devices
-----------------------
@@ -624,6 +662,27 @@ Switch configuration
case, further calls to ``get_tag_protocol`` should report the protocol in
current use.
+- ``connect_tag_protocol``: optional method to notify the switch driver that a
+ tagging protocol driver has connected to this switch. Depending on the
+ contract established by the protocol given in the ``proto`` argument, the
+ tagger-owned storage (``ds->tagger_data``) may be expected to contain a
+ pointer to a data structure specific to the tagging protocol. This data
+ structure may contain function pointers to packet handlers that the switch
+ driver registers with the tagging protocol. If interested in these packets,
+ the switch driver must cast the ``ds->tagger_data`` pointer to the data type
+ established by the tagging protocol, and assign the packet handler function
+ pointers to methods that it owns. Since the memory pointed to by
+ ``ds->tagger_data`` is owned by the tagging protocol, the switch driver must
+ assume by convention that it has been allocated, and this method is only
+ provided for making initial adjustments to the contents of ``ds->tagger_data``.
+ It is also the reason why no ``disconnect_tag_protocol()`` counterpart is
+ provided. Additionally, a tagging protocol driver which makes use of
+ tagger-owned storage must not assume that the connected switch has
+ implemented the ``connect_tag_protocol()`` method (it may connect to a
+ ``dsa_loop`` switch, which does not). Therefore, a tagging protocol may
+ always rely on ``ds->tagger_data``, but it must treat the packet handlers
+ provided by the switch in this method as optional.
+
- ``setup``: setup function for the switch, this function is responsible for setting
up the ``dsa_switch_ops`` private structure with all it needs: register maps,
interrupts, mutexes, locks, etc. This function is also expected to properly
--
2.34.1
next prev parent reply other threads:[~2023-12-08 19:36 UTC|newest]
Thread overview: 24+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-12-08 19:35 [PATCH net 0/4] Add some history to the DSA documentation Vladimir Oltean
2023-12-08 19:35 ` Vladimir Oltean [this message]
2023-12-08 22:15 ` [PATCH net 1/4] docs: net: dsa: document the tagger-owned storage mechanism Linus Walleij
2023-12-08 22:32 ` Florian Fainelli
2023-12-10 13:37 ` Alvin Šipraga
2023-12-08 19:35 ` [PATCH net 2/4] docs: net: dsa: update platform_data documentation Vladimir Oltean
2023-12-08 22:19 ` Linus Walleij
2023-12-09 0:52 ` Vladimir Oltean
2023-12-08 22:33 ` Florian Fainelli
2023-12-10 13:37 ` Alvin Šipraga
2023-12-08 19:35 ` [PATCH net 3/4] docs: net: dsa: update user MDIO bus documentation Vladimir Oltean
2023-12-08 22:22 ` Linus Walleij
2023-12-08 22:36 ` Florian Fainelli
2023-12-09 1:22 ` Vladimir Oltean
2023-12-09 21:49 ` Linus Walleij
2023-12-10 13:48 ` Alvin Šipraga
2023-12-11 14:35 ` Vladimir Oltean
2023-12-13 5:30 ` Luiz Angelo Daros de Luca
2023-12-13 12:06 ` Vladimir Oltean
2023-12-08 19:35 ` [PATCH net 4/4] docs: net: dsa: replace TODO section with info about history and devel ideas Vladimir Oltean
2023-12-08 22:40 ` Linus Walleij
2023-12-08 23:03 ` Florian Fainelli
2023-12-09 1:58 ` Vladimir Oltean
2023-12-11 17:29 ` Florian Fainelli
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=20231208193518.2018114-2-vladimir.oltean@nxp.com \
--to=vladimir.oltean@nxp.com \
--cc=alsi@bang-olufsen.dk \
--cc=andrew@lunn.ch \
--cc=arun.ramadoss@microchip.com \
--cc=corbet@lwn.net \
--cc=davem@davemloft.net \
--cc=edumazet@google.com \
--cc=f.fainelli@gmail.com \
--cc=kuba@kernel.org \
--cc=linus.walleij@linaro.org \
--cc=luizluca@gmail.com \
--cc=madhuri.sripada@microchip.com \
--cc=mw@semihalf.com \
--cc=netdev@vger.kernel.org \
--cc=pabeni@redhat.com \
--cc=rmk+kernel@armlinux.org.uk \
--cc=tobias@waldekranz.com \
/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