1 .. SPDX-License-Identifier: BSD-3-Clause
2 Copyright 2018 The DPDK contributors
4 ABI and API Deprecation
5 =======================
7 See the guidelines document for details of the :doc:`ABI policy
8 <../contributing/abi_policy>`. API and ABI deprecation notices are to be posted
14 * make: Support for building DPDK with "make" has been deprecated and
15 support will be removed in the 20.11 release. From 20.11 onwards, DPDK
16 should be built using meson and ninja. For basic instructions see the
17 `Quick-start Guide <https://core.dpdk.org/doc/quick-start/>`_ on the
18 website or the `Getting Started Guide
19 <https://doc.dpdk.org/guides/linux_gsg/build_dpdk.html>`_ document.
21 * meson: The minimum supported version of meson for configuring and building
22 DPDK will be increased to v0.47.1 (from 0.41) from DPDK 19.05 onwards. For
23 those users with a version earlier than 0.47.1, an updated copy of meson
24 can be got using the ``pip``, or ``pip3``, tool for downloading python
27 * kvargs: The function ``rte_kvargs_process`` will get a new parameter
28 for returning key match count. It will ease handling of no-match case.
30 * eal: The function ``rte_eal_remote_launch`` will return new error codes
31 after read or write error on the pipe, instead of calling ``rte_panic``.
33 * eal: The ``rte_logs`` struct and global symbol will be made private to
34 remove it from the externally visible ABI and allow it to be updated in the
37 * eal: The ``rte_dev_event`` structure will be made private to the EAL as no
38 public API makes use of it.
40 * rte_atomicNN_xxx: These APIs do not take memory order parameter. This does
41 not allow for writing optimized code for all the CPU architectures supported
42 in DPDK. DPDK will adopt C11 atomic operations semantics and provide wrappers
43 using C11 atomic built-ins. These wrappers must be used for patches that
44 need to be merged in 20.08 onwards. This change will not introduce any
45 performance degradation.
47 * rte_smp_*mb: These APIs provide full barrier functionality. However, many
48 use cases do not require full barriers. To support such use cases, DPDK will
49 adopt C11 barrier semantics and provide wrappers using C11 atomic built-ins.
50 These wrappers must be used for patches that need to be merged in 20.08
51 onwards. This change will not introduce any performance degradation.
53 * rte_cio_*mb: Since the IO barriers for ARMv8 platforms are relaxed from DSB
54 to DMB, rte_cio_*mb APIs provide the same functionality as rte_io_*mb
55 APIs (taking all platforms into consideration). rte_io_*mb APIs should be
56 used in the place of rte_cio_*mb APIs. The rte_cio_*mb APIs will be
57 deprecated in 20.11 release.
59 * igb_uio: In the view of reducing the kernel dependency from the main tree,
60 as a first step, the Technical Board decided to move ``igb_uio``
61 kernel module to the dpdk-kmods repository in the /linux/igb_uio/ directory
63 Minutes of Technical Board Meeting of `2019-11-06
64 <https://mails.dpdk.org/archives/dev/2019-November/151763.html>`_.
66 * lib: will fix extending some enum/define breaking the ABI. There are multiple
67 samples in DPDK that enum/define terminated with a ``.*MAX.*`` value which is
68 used by iterators, and arrays holding these values are sized with this
69 ``.*MAX.*`` value. So extending this enum/define increases the ``.*MAX.*``
70 value which increases the size of the array and depending on how/where the
71 array is used this may break the ABI.
72 ``RTE_ETH_FLOW_MAX`` is one sample of the mentioned case, adding a new flow
73 type will break the ABI because of ``flex_mask[RTE_ETH_FLOW_MAX]`` array
74 usage in following public struct hierarchy:
75 ``rte_eth_fdir_flex_conf -> rte_fdir_conf -> rte_eth_conf (in the middle)``.
76 Need to identify this kind of usages and fix in 20.11, otherwise this blocks
77 us extending existing enum/define.
78 One solution can be using a fixed size array instead of ``.*MAX.*`` value.
80 * pci: The PCI resources map API (``pci_map_resource`` and
81 ``pci_unmap_resource``) was not abstracting the Unix mmap flags (see the
82 workaround for Windows support implemented in the commit
83 9d2b24593724 ("pci: keep API compatibility with mmap values")).
84 This API will be removed from the public API in 20.11 and moved to the PCI
85 bus driver along with the PCI resources lists and associated structures
86 (``pci_map``, ``pci_msix_table``, ``mapped_pci_resource`` and
87 ``mapped_pci_res_list``).
88 With this removal, there won't be a need for the mentioned workaround which
91 * pci: The ``rte_kernel_driver`` enum defined in rte_dev.h will be made private
92 to the PCI subsystem as it is used only by the PCI bus driver and PCI
94 The associated field ``kdrv`` in the ethdev ``rte_eth_dev_data`` structure
95 will be removed as it gave no useful abstracted information to the
96 applications and had no user (neither internal nor external).
98 * mbuf: Some fields will be converted to dynamic API in DPDK 20.11
99 in order to reserve more space for the dynamic fields, as explained in
100 `this presentation <https://www.youtube.com/watch?v=Ttl6MlhmzWY>`_.
101 The following static fields will be moved as dynamic:
104 - ``userdata`` / ``udata64``
107 As a consequence, the layout of the ``struct rte_mbuf`` will be re-arranged,
108 avoiding impact on vectorized implementation of the driver datapaths,
109 while evaluating performance gains of a better use of the first cache line.
111 The deprecated unioned fields ``buf_physaddr`` and ``refcnt_atomic``
112 (as explained below) will be removed in DPDK 20.11.
114 * mbuf: ``refcnt_atomic`` member in structures ``rte_mbuf`` and
115 ``rte_mbuf_ext_shared_info`` is of type ``rte_atomic16_t``.
116 Due to adoption of C11 atomic builtins, the field ``refcnt_atomic``
117 will be replaced with ``refcnt`` of type ``uint16_t`` in DPDK 20.11.
119 * ethdev: Split the ``struct eth_dev_ops`` struct to hide it as much as possible
120 will be done in 20.11.
121 Currently the ``struct eth_dev_ops`` struct is accessible by the application
122 because some inline functions, like ``rte_eth_tx_descriptor_status()``,
123 access the struct directly.
124 The struct will be separate in two, the ops used by inline functions will be
125 moved next to Rx/Tx burst functions, rest of the ``struct eth_dev_ops`` struct
126 will be moved to header file for drivers to hide it from applications.
128 * ethdev: the legacy filter API, including
129 ``rte_eth_dev_filter_supported()``, ``rte_eth_dev_filter_ctrl()`` as well
130 as filter types MACVLAN, ETHERTYPE, FLEXIBLE, SYN, NTUPLE, TUNNEL, FDIR,
131 HASH and L2_TUNNEL, is superseded by the generic flow API (rte_flow) in
132 PMDs that implement the latter.
133 The legacy API will be removed in DPDK 20.11.
135 * ethdev: The flow director API, including ``rte_eth_conf.fdir_conf`` field,
136 and the related structures (``rte_fdir_*`` and ``rte_eth_fdir_*``),
137 will be removed in DPDK 20.11.
139 * ethdev: The legacy L2 tunnel filtering API is deprecated as the rest of
140 the legacy filtering API.
141 The functions ``rte_eth_dev_l2_tunnel_eth_type_conf`` and
142 ``rte_eth_dev_l2_tunnel_offload_set`` which were not marked as deprecated,
143 will be removed in DPDK 20.11.
145 * ethdev: Update API functions returning ``void`` to return ``int`` with
146 negative errno values to indicate various error conditions (e.g.
147 invalid port ID, unsupported operation, failed operation):
149 - ``rte_eth_dev_stop``
150 - ``rte_eth_dev_close``
152 * ethdev: The temporary flag RTE_ETH_DEV_CLOSE_REMOVE will be removed in 20.11.
153 As a consequence, the new behaviour introduced in 18.11 will be effective
154 for all drivers: generic port resources are freed on close operation.
155 Private resources are expected to be released in the ``dev_close`` callback.
156 More details in http://inbox.dpdk.org/dev/5248162.j6AOsuQRmx@thomas/
158 * ethdev: New offload flags ``DEV_RX_OFFLOAD_FLOW_MARK`` will be added in 19.11.
159 This will allow application to enable or disable PMDs from updating
160 ``rte_mbuf::hash::fdir``.
161 This scheme will allow PMDs to avoid writes to ``rte_mbuf`` fields on Rx and
162 thereby improve Rx performance if application wishes do so.
163 In 19.11 PMDs will still update the field even when the offload is not
166 * ethdev: ``rx_descriptor_done`` dev_ops and ``rte_eth_rx_descriptor_done``
167 will be deprecated in 20.11 and will be removed in 21.11.
168 Existing ``rte_eth_rx_descriptor_status`` and ``rte_eth_tx_descriptor_status``
169 APIs can be used as replacement.
171 * ethdev: The port mirroring API can be replaced with a more fine grain flow API.
172 The structs ``rte_eth_mirror_conf``, ``rte_eth_vlan_mirror`` and the functions
173 ``rte_eth_mirror_rule_set``, ``rte_eth_mirror_rule_reset`` will be marked
174 as deprecated in DPDK 20.11, along with the associated macros ``ETH_MIRROR_*``.
175 This API will be fully removed in DPDK 21.11.
177 * ethdev: Some internal APIs for driver usage are exported in the .map file.
178 Now DPDK has ``__rte_internal`` marker so we can mark internal APIs and move
179 them to the INTERNAL block in .map. Although these APIs are internal it will
180 break the ABI checks, that is why change is planned for 20.11.
181 The list of internal APIs are mainly ones listed in ``rte_ethdev_driver.h``.
183 * traffic manager: All traffic manager API's in ``rte_tm.h`` were mistakenly made
184 ABI stable in the v19.11 release. The TM maintainer and other contributors have
185 agreed to keep the TM APIs as experimental in expectation of additional spec
186 improvements. Therefore, all APIs in ``rte_tm.h`` will be marked back as
187 experimental in v20.11 DPDK release. For more details, please see `the thread
188 <https://mails.dpdk.org/archives/dev/2020-April/164970.html>`_.
190 * pmd_dpaa: The API ``rte_pmd_dpaa_set_tx_loopback`` will have extended
191 ``port_id`` definition from ``uint8_t`` to ``uint16_t``.
193 * security: The API ``rte_security_session_create`` takes only single mempool
194 for session and session private data. So the application need to create
195 mempool for twice the number of sessions needed and will also lead to
196 wastage of memory as session private data need more memory compared to session.
197 Hence the API will be modified to take two mempool pointers - one for session
198 and one for private data.
200 * cryptodev: ``RTE_CRYPTO_AEAD_LIST_END`` from ``enum rte_crypto_aead_algorithm``,
201 ``RTE_CRYPTO_CIPHER_LIST_END`` from ``enum rte_crypto_cipher_algorithm`` and
202 ``RTE_CRYPTO_AUTH_LIST_END`` from ``enum rte_crypto_auth_algorithm``
205 * cryptodev: support for using IV with all sizes is added, J0 still can
206 be used but only when IV length in following structs ``rte_crypto_auth_xform``,
207 ``rte_crypto_aead_xform`` is set to zero. When IV length is greater or equal
208 to one it means it represents IV, when is set to zero it means J0 is used
209 directly, in this case 16 bytes of J0 need to be passed.
211 * eventdev: Following structures will be modified to support DLB PMD
212 and future extensions:
214 - ``rte_event_dev_info``
215 - ``rte_event_dev_config``
216 - ``rte_event_port_conf``
218 Patches containing justification, documentation, and proposed modifications
221 - https://patches.dpdk.org/patch/71457/
222 - https://patches.dpdk.org/patch/71456/
224 * rawdev: The rawdev APIs which take a device-specific structure as
225 parameter directly, or indirectly via a "private" pointer inside another
226 structure, will be modified to take an additional parameter of the
227 structure size. The affected APIs will include ``rte_rawdev_info_get``,
228 ``rte_rawdev_configure``, ``rte_rawdev_queue_conf_get`` and
229 ``rte_rawdev_queue_setup``.
231 * sched: To allow more traffic classes, flexible mapping of pipe queues to
232 traffic classes, and subport level configuration of pipes and queues
233 changes will be made to macros, data structures and API functions defined
234 in "rte_sched.h". These changes are aligned to improvements suggested in the
235 RFC https://mails.dpdk.org/archives/dev/2018-November/120035.html.
237 * metrics: The function ``rte_metrics_init`` will have a non-void return
238 in order to notify errors instead of calling ``rte_exit``.
240 * power: ``rte_power_set_env`` function will no longer return 0 on attempt
241 to set new power environment if power environment was already initialized.
242 In this case the function will return -1 unless the environment is unset first
243 (using ``rte_power_unset_env``). Other function usage scenarios will not change.
245 * python: Since the beginning of 2020, Python 2 has officially reached
246 end-of-support: https://www.python.org/doc/sunset-python-2/.
247 Python 2 support will be completely removed in 20.11.
248 In 20.08, explicit deprecation warnings will be displayed when running
249 scripts with Python 2.
251 * dpdk-setup.sh: This old script relies on deprecated stuff, and especially
252 ``make``. Given environments are too much variables for such a simple script,
253 it will be removed in DPDK 20.11.
254 Some useful parts may be converted into specific scripts.