devlink: expand the devlink-info documentation

We are having multiple review cycles with all vendors trying
to implement devlink-info. Let's expand the documentation with
more information about what's implemented and motivation behind
this interface in an attempt to make the implementations easier.

Describe what each info section is supposed to contain, and make
some references to other HW interfaces (PCI caps).

Document how firmware management is expected to look, to make
it clear how devlink-info and devlink-flash work in concert.

Name some future work.

v2: - improve wording
v3: - improve wording

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

1 files changed, 117 insertions, 16 deletions

diff --git a/Documentation/networking/devlink/devlink-info.rst b/Documentation/networking/devlink/devlink-info.rst
index b701899b7f42..e5e5e89f8f7a 100644
--- a/Documentation/networking/devlink/devlink-info.rst
+++ b/Documentation/networking/devlink/devlink-info.rst
@@ -5,34 +5,119 @@ Devlink Info
 ============
 
 The ``devlink-info`` mechanism enables device drivers to report device
-information in a generic fashion. It is extensible, and enables exporting
-even device or driver specific information.
+(hardware and firmware) information in a standard, extensible fashion.
 
-devlink supports representing the following types of versions
+The original motivation for the ``devlink-info`` API was twofold:
 
-.. list-table:: List of version types
+ - making it possible to automate device and firmware management in a fleet
+   of machines in a vendor-independent fashion (see also
+   :ref:`Documentation/networking/devlink/devlink-flash.rst <devlink_flash>`);
+ - name the per component FW versions (as opposed to the crowded ethtool
+   version string).
+
+``devlink-info`` supports reporting multiple types of objects. Reporting driver
+versions is generally discouraged - here, and via any other Linux API.
+
+.. list-table:: List of top level info objects
    :widths: 5 95
 
-   * - Type
+   * - Name
      - Description
+   * - ``driver``
+     - Name of the currently used device driver, also available through sysfs.
+
+   * - ``serial_number``
+     - Serial number of the device.
+
+       This is usually the serial number of the ASIC, also often available
+       in PCI config space of the device in the *Device Serial Number*
+       capability.
+
+       The serial number should be unique per physical device.
+       Sometimes the serial number of the device is only 48 bits long (the
+       length of the Ethernet MAC address), and since PCI DSN is 64 bits long
+       devices pad or encode additional information into the serial number.
+       One example is adding port ID or PCI interface ID in the extra two bytes.
+       Drivers should make sure to strip or normalize any such padding
+       or interface ID, and report only the part of the serial number
+       which uniquely identifies the hardware. In other words serial number
+       reported for two ports of the same device or on two hosts of
+       a multi-host device should be identical.
+
+       .. note:: ``devlink-info`` API should be extended with a new field
+	  if devices want to report board/product serial number (often
+	  reported in PCI *Vital Product Data* capability).
+
    * - ``fixed``
-     - Represents fixed versions, which cannot change. For example,
+     - Group for hardware identifiers, and versions of components
+       which are not field-updatable.
+
+       Versions in this section identify the device design. For example,
        component identifiers or the board version reported in the PCI VPD.
+       Data in ``devlink-info`` should be broken into the smallest logical
+       components, e.g. PCI VPD may concatenate various information
+       to form the Part Number string, while in ``devlink-info`` all parts
+       should be reported as separate items.
+
+       This group must not contain any frequently changing identifiers,
+       such as serial numbers. See
+       :ref:`Documentation/networking/devlink/devlink-flash.rst <devlink_flash>`
+       to understand why.
+
    * - ``running``
-     - Represents the version of the currently running component. For
-       example the running version of firmware. These versions generally
-       only update after a reboot.
+     - Group for information about currently running software/firmware.
+       These versions often only update after a reboot, sometimes device reset.
+
    * - ``stored``
-     - Represents the version of a component as stored, such as after a
-       flash update. Stored values should update to reflect changes in the
-       flash even if a reboot has not yet occurred.
+     - Group for software/firmware versions in device flash.
+
+       Stored values must update to reflect changes in the flash even
+       if reboot has not yet occurred. If device is not capable of updating
+       ``stored`` versions when new software is flashed, it must not report
+       them.
+
+Each version can be reported at most once in each version group. Firmware
+components stored on the flash should feature in both the ``running`` and
+``stored`` sections, if device is capable of reporting ``stored`` versions
+(see :ref:`Documentation/networking/devlink/devlink-flash.rst <devlink_flash>`).
+In case software/firmware components are loaded from the disk (e.g.
+``/lib/firmware``) only the running version should be reported via
+the kernel API.
 
 Generic Versions
 ================
 
 It is expected that drivers use the following generic names for exporting
-version information. Other information may be exposed using driver-specific
-names, but these should be documented in the driver-specific file.
+version information. If a generic name for a given component doesn't exist yet,
+driver authors should consult existing driver-specific versions and attempt
+reuse. As last resort, if a component is truly unique, using driver-specific
+names is allowed, but these should be documented in the driver-specific file.
+
+All versions should try to use the following terminology:
+
+.. list-table:: List of common version suffixes
+   :widths: 10 90
+
+   * - Name
+     - Description
+   * - ``id``, ``revision``
+     - Identifiers of designs and revision, mostly used for hardware versions.
+
+   * - ``api``
+     - Version of API between components. API items are usually of limited
+       value to the user, and can be inferred from other versions by the vendor,
+       so adding API versions is generally discouraged as noise.
+
+   * - ``bundle_id``
+     - Identifier of a distribution package which was flashed onto the device.
+       This is an attribute of a firmware package which covers multiple versions
+       for ease of managing firmware images (see
+       :ref:`Documentation/networking/devlink/devlink-flash.rst <devlink_flash>`).
+
+       ``bundle_id`` can appear in both ``running`` and ``stored`` versions,
+       but it must not be reported if any of the components covered by the
+       ``bundle_id`` was changed and no longer matches the version from
+       the bundle.
 
 board.id
 --------
@@ -52,7 +137,7 @@ ASIC design identifier.
 asic.rev
 --------
 
-ASIC design revision.
+ASIC design revision/stepping.
 
 board.manufacture
 -----------------
@@ -91,7 +176,8 @@ Network Controller Sideband Interface.
 fw.psid
 -------
 
-Unique identifier of the firmware parameter set.
+Unique identifier of the firmware parameter set. These are usually
+parameters of a particular board, defined at manufacturing time.
 
 fw.roce
 -------
@@ -103,3 +189,18 @@ fw.bundle_id
 ------------
 
 Unique identifier of the entire firmware bundle.
+
+Future work
+===========
+
+The following extensions could be useful:
+
+ - product serial number - NIC boards often get labeled with a board serial
+   number rather than ASIC serial number; it'd be useful to add board serial
+   numbers to the API if they can be retrieved from the device;
+
+ - on-disk firmware file names - drivers list the file names of firmware they
+   may need to load onto devices via the ``MODULE_FIRMWARE()`` macro. These,
+   however, are per module, rather than per device. It'd be useful to list
+   the names of firmware files the driver will try to load for a given device,
+   in order of priority.