devlink: Add Documentation/networking/devlink-health.txt
authorAya Levin <ayal@mellanox.com>
Thu, 17 Jan 2019 21:59:20 +0000 (23:59 +0200)
committerDavid S. Miller <davem@davemloft.net>
Fri, 18 Jan 2019 22:51:23 +0000 (14:51 -0800)
This patch adds a new file to add information about devlink health
mechanism.

Signed-off-by: Aya Levin <ayal@mellanox.com>
Signed-off-by: Eran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Documentation/networking/devlink-health.txt [new file with mode: 0644]

diff --git a/Documentation/networking/devlink-health.txt b/Documentation/networking/devlink-health.txt
new file mode 100644 (file)
index 0000000..1db3fbe
--- /dev/null
@@ -0,0 +1,86 @@
+The health mechanism is targeted for Real Time Alerting, in order to know when
+something bad had happened to a PCI device
+- Provide alert debug information
+- Self healing
+- If problem needs vendor support, provide a way to gather all needed debugging
+  information.
+
+The main idea is to unify and centralize driver health reports in the
+generic devlink instance and allow the user to set different
+attributes of the health reporting and recovery procedures.
+
+The devlink health reporter:
+Device driver creates a "health reporter" per each error/health type.
+Error/Health type can be a known/generic (eg pci error, fw error, rx/tx error)
+or unknown (driver specific).
+For each registered health reporter a driver can issue error/health reports
+asynchronously. All health reports handling is done by devlink.
+Device driver can provide specific callbacks for each "health reporter", e.g.
+ - Recovery procedures
+ - Diagnostics and object dump procedures
+ - OOB initial parameters
+Different parts of the driver can register different types of health reporters
+with different handlers.
+
+Once an error is reported, devlink health will do the following actions:
+  * A log is being send to the kernel trace events buffer
+  * Health status and statistics are being updated for the reporter instance
+  * Object dump is being taken and saved at the reporter instance (as long as
+    there is no other dump which is already stored)
+  * Auto recovery attempt is being done. Depends on:
+    - Auto-recovery configuration
+    - Grace period vs. time passed since last recover
+
+The user interface:
+User can access/change each reporter's parameters and driver specific callbacks
+via devlink, e.g per error type (per health reporter)
+ - Configure reporter's generic parameters (like: disable/enable auto recovery)
+ - Invoke recovery procedure
+ - Run diagnostics
+ - Object dump
+
+The devlink health interface (via netlink):
+DEVLINK_CMD_HEALTH_REPORTER_GET
+  Retrieves status and configuration info per DEV and reporter.
+DEVLINK_CMD_HEALTH_REPORTER_SET
+  Allows reporter-related configuration setting.
+DEVLINK_CMD_HEALTH_REPORTER_RECOVER
+  Triggers a reporter's recovery procedure.
+DEVLINK_CMD_HEALTH_REPORTER_DIAGNOSE
+  Retrieves diagnostics data from a reporter on a device.
+DEVLINK_CMD_HEALTH_REPORTER_DUMP_GET
+  Retrieves the last stored dump. Devlink health
+  saves a single dump. If an dump is not already stored by the devlink
+  for this reporter, devlink generates a new dump.
+  dump output is defined by the reporter.
+DEVLINK_CMD_HEALTH_REPORTER_DUMP_CLEAR
+  Clears the last saved dump file for the specified reporter.
+
+
+                                               netlink
+                                      +--------------------------+
+                                      |                          |
+                                      |            +             |
+                                      |            |             |
+                                      +--------------------------+
+                                                   |request for ops
+                                                   |(diagnose,
+ mlx5_core                             devlink     |recover,
+                                                   |dump)
++--------+                            +--------------------------+
+|        |                            |    reporter|             |
+|        |                            |  +---------v----------+  |
+|        |   ops execution            |  |                    |  |
+|     <----------------------------------+                    |  |
+|        |                            |  |                    |  |
+|        |                            |  + ^------------------+  |
+|        |                            |    | request for ops     |
+|        |                            |    | (recover, dump)     |
+|        |                            |    |                     |
+|        |                            |  +-+------------------+  |
+|        |     health report          |  | health handler     |  |
+|        +------------------------------->                    |  |
+|        |                            |  +--------------------+  |
+|        |     health reporter create |                          |
+|        +---------------------------->                          |
++--------+                            +--------------------------+