xref: /dpdk/doc/guides/nics/fail_safe.rst (revision d80e42cce4c7017ed8c99dabb8ae444a492acc1c)
1..  SPDX-License-Identifier: BSD-3-Clause
2    Copyright 2017 6WIND S.A.
3
4Fail-safe poll mode driver library
5==================================
6
7The Fail-safe poll mode driver library (**librte_pmd_failsafe**) is a virtual
8device that allows using any device supporting hotplug (sudden device removal
9and plugging on its bus), without modifying other components relying on such
10device (application, other PMDs).
11
12Additionally to the Seamless Hotplug feature, the Fail-safe PMD offers the
13ability to redirect operations to secondary devices when the primary has been
14removed from the system.
15
16.. note::
17
18   The library is enabled by default. You can enable it or disable it manually
19   by setting the ``CONFIG_RTE_LIBRTE_PMD_FAILSAFE`` configuration option.
20
21Features
22--------
23
24The Fail-safe PMD only supports a limited set of features. If you plan to use a
25device underneath the Fail-safe PMD with a specific feature, this feature must
26be supported by the Fail-safe PMD to avoid throwing any error.
27
28A notable exception is the device removal feature. The fail-safe PMD being a
29virtual device, it cannot currently be removed in the sense of a specific bus
30hotplug, like for PCI for example. It will however enable this feature for its
31sub-device automatically, detecting those that are capable and register the
32relevant callback for such event.
33
34Check the feature matrix for the complete set of supported features.
35
36Compilation option
37------------------
38
39This option can be modified in the ``$RTE_TARGET/build/.config`` file.
40
41- ``CONFIG_RTE_LIBRTE_PMD_FAILSAFE`` (default **y**)
42
43  Toggle compiling librte_pmd_failsafe.
44
45Using the Fail-safe PMD from the EAL command line
46-------------------------------------------------
47
48The Fail-safe PMD can be used like most other DPDK virtual devices, by passing a
49``--vdev`` parameter to the EAL when starting the application. The device name
50must start with the *net_failsafe* prefix, followed by numbers or letters. This
51name must be unique for each device. Each fail-safe instance must have at least one
52sub-device, up to ``RTE_MAX_ETHPORTS-1``.
53
54A sub-device can be any legal DPDK device, including possibly another fail-safe
55instance.
56
57Fail-safe command line parameters
58~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
59
60- **dev(<iface>)** parameter
61
62  This parameter allows the user to define a sub-device. The ``<iface>`` part of
63  this parameter must be a valid device definition. It could be the argument
64  provided to any ``-w`` device specification or the argument that would be
65  given to a ``--vdev`` parameter (including a fail-safe).
66  Enclosing the device definition within parenthesis here allows using
67  additional sub-device parameters if need be. They will be passed on to the
68  sub-device.
69
70.. note::
71
72   In case of whitelist sub-device probed by EAL, fail-safe PMD will take the device
73   as is, which means that EAL device options are taken in this case.
74   When trying to use a PCI device automatically probed in blacklist mode,
75   the syntax for the fail-safe must be with the full PCI id:
76   Domain:Bus:Device.Function. See the usage example section.
77
78- **exec(<shell command>)** parameter
79
80  This parameter allows the user to provide a command to the fail-safe PMD to
81  execute and define a sub-device.
82  It is done within a regular shell context.
83  The first line of its output is read by the fail-safe PMD and otherwise
84  interpreted as if passed by the regular **dev** parameter.
85  Any other line is discarded.
86  If the command fail or output an incorrect string, the sub-device is not
87  initialized.
88  All commas within the ``shell command`` are replaced by spaces before
89  executing the command. This helps using scripts to specify devices.
90
91- **fd(<file descriptor number>)** parameter
92
93  This parameter reads a device definition from an arbitrary file descriptor
94  number in ``<iface>`` format as described above.
95
96  The file descriptor is read in non-blocking mode and is never closed in
97  order to take only the last line into account (unlike ``exec()``) at every
98  probe attempt.
99
100- **mac** parameter [MAC address]
101
102  This parameter allows the user to set a default MAC address to the fail-safe
103  and all of its sub-devices.
104  If no default mac address is provided, the fail-safe PMD will read the MAC
105  address of the first of its sub-device to be successfully probed and use it as
106  its default MAC address, trying to set it to all of its other sub-devices.
107  If no sub-device was successfully probed at initialization, then a random MAC
108  address is generated, that will be subsequently applied to all sub-device once
109  they are probed.
110
111- **hotplug_poll** parameter [UINT64] (default **2000**)
112
113  This parameter allows the user to configure the amount of time in milliseconds
114  between two slave upkeep round.
115
116Usage example
117~~~~~~~~~~~~~
118
119This section shows some example of using **testpmd** with a fail-safe PMD.
120
121#. To build a PMD and configure DPDK, refer to the document
122   :ref:`compiling and testing a PMD for a NIC <pmd_build_and_test>`.
123
124#. Start testpmd. The slave device should be blacklisted from normal EAL
125   operations to avoid probing it twice when in PCI blacklist mode.
126
127   .. code-block:: console
128
129      $RTE_TARGET/build/app/testpmd -c 0xff -n 4 \
130         --vdev 'net_failsafe0,mac=de:ad:be:ef:01:02,dev(84:00.0),dev(net_ring0)' \
131         -b 84:00.0 -b 00:04.0 -- -i
132
133   If the slave device being used is not blacklisted, it will be probed by the
134   EAL first. When the fail-safe then tries to initialize it the probe operation
135   fails.
136
137   Note that PCI blacklist mode is the default PCI operating mode.
138
139#. Alternatively, it can be used alongside any other device in whitelist mode.
140
141   .. code-block:: console
142
143      $RTE_TARGET/build/app/testpmd -c 0xff -n 4 \
144         --vdev 'net_failsafe0,mac=de:ad:be:ef:01:02,dev(84:00.0),dev(net_ring0)' \
145         -w 81:00.0 -- -i
146
147#. Start testpmd using a flexible device definition
148
149   .. code-block:: console
150
151      $RTE_TARGET/build/app/testpmd -c 0xff -n 4 --no-pci \
152         --vdev='net_failsafe0,exec(echo 84:00.0)' -- -i
153
154#. Start testpmd, automatically probing the device 84:00.0 and using it with
155   the fail-safe.
156
157   .. code-block:: console
158
159      $RTE_TARGET/build/app/testpmd -c 0xff -n 4 \
160         --vdev 'net_failsafe0,dev(0000:84:00.0),dev(net_ring0)' -- -i
161
162
163Using the Fail-safe PMD from an application
164-------------------------------------------
165
166This driver strives to be as seamless as possible to existing applications, in
167order to propose the hotplug functionality in the easiest way possible.
168
169Care must be taken, however, to respect the **ether** API concerning device
170access, and in particular, using the ``RTE_ETH_FOREACH_DEV`` macro to iterate
171over ethernet devices, instead of directly accessing them or by writing one's
172own device iterator.
173
174Plug-in feature
175---------------
176
177A sub-device can be defined without existing on the system when the fail-safe
178PMD is initialized. Upon probing this device, the fail-safe PMD will detect its
179absence and postpone its use. It will then register for a periodic check on any
180missing sub-device.
181
182During this time, the fail-safe PMD can be used normally, configured and told to
183emit and receive packets. It will store any applied configuration, and try to
184apply it upon the probing of its missing sub-device. After this configuration
185pass, the new sub-device will be synchronized with other sub-devices, i.e. be
186started if the fail-safe PMD has been started by the user before.
187
188Plug-out feature
189----------------
190
191A sub-device supporting the device removal event can be removed from its bus at
192any time. The fail-safe PMD will register a callback for such event and react
193accordingly. It will try to safely stop, close and uninit the sub-device having
194emitted this event, allowing it to free its eventual resources.
195
196Fail-safe glossary
197------------------
198
199Fallback device : Secondary device
200    The fail-safe will fail-over onto this device when the preferred device is
201    absent.
202
203Preferred device : Primary device
204    The first declared sub-device in the fail-safe parameters.
205    When this device is plugged, it is always used as emitting device.
206    It is the main sub-device and is used as target for configuration
207    operations if there is any ambiguity.
208
209Upkeep round
210    Periodical process when slaves are serviced. Each devices having a state
211    different to that of the fail-safe device itself, is synchronized with it.
212    Additionally, each slave having the remove flag set are cleaned-up.
213
214Slave
215    In the context of the fail-safe PMD, synonymous to sub-device.
216
217Sub-device
218    A device being utilized by the fail-safe PMD.
219    This is another PMD running underneath the fail-safe PMD.
220    Any sub-device can disappear at any time. The fail-safe will ensure
221    that the device removal happens gracefully.
222