xref: /dpdk/doc/guides/prog_guide/toeplitz_hash_lib.rst (revision 9c30a6f3c9a456e8111a2b1e5f6c2c02a62025b6)
1534fe5f3SVladimir Medvedkin..  SPDX-License-Identifier: BSD-3-Clause
2534fe5f3SVladimir Medvedkin    Copyright(c) 2021 Intel Corporation.
3534fe5f3SVladimir Medvedkin
4534fe5f3SVladimir MedvedkinToeplitz Hash Library
5534fe5f3SVladimir Medvedkin=====================
6534fe5f3SVladimir Medvedkin
7534fe5f3SVladimir MedvedkinDPDK provides a Toeplitz Hash Library
8534fe5f3SVladimir Medvedkinto calculate the Toeplitz hash function and to use its properties.
9534fe5f3SVladimir MedvedkinThe Toeplitz hash function is commonly used in a wide range of NICs
10534fe5f3SVladimir Medvedkinto calculate the RSS hash sum to spread the traffic among the queues.
11534fe5f3SVladimir Medvedkin
12534fe5f3SVladimir Medvedkin.. _figure_rss_queue_assign:
13534fe5f3SVladimir Medvedkin
14534fe5f3SVladimir Medvedkin.. figure:: img/rss_queue_assign.*
15534fe5f3SVladimir Medvedkin
16534fe5f3SVladimir Medvedkin   RSS queue assignment example
17534fe5f3SVladimir Medvedkin
18534fe5f3SVladimir Medvedkin
19534fe5f3SVladimir MedvedkinToeplitz hash function API
20534fe5f3SVladimir Medvedkin--------------------------
21534fe5f3SVladimir Medvedkin
22534fe5f3SVladimir MedvedkinThere are two functions that provide calculation of the Toeplitz hash sum:
23534fe5f3SVladimir Medvedkin
24534fe5f3SVladimir Medvedkin* ``rte_softrss()``
25534fe5f3SVladimir Medvedkin* ``rte_softrss_be()``
26534fe5f3SVladimir Medvedkin
27534fe5f3SVladimir MedvedkinBoth of these functions take the parameters:
28534fe5f3SVladimir Medvedkin
29534fe5f3SVladimir Medvedkin* A pointer to the tuple, containing fields extracted from the packet.
30534fe5f3SVladimir Medvedkin* A length of this tuple counted in double words.
31534fe5f3SVladimir Medvedkin* A pointer to the RSS hash key corresponding to the one installed on the NIC.
32534fe5f3SVladimir Medvedkin
33534fe5f3SVladimir MedvedkinBoth functions expect the tuple to be in "host" byte order
34534fe5f3SVladimir Medvedkinand a multiple of 4 bytes in length.
35534fe5f3SVladimir MedvedkinThe ``rte_softrss()`` function expects the ``rss_key``
36534fe5f3SVladimir Medvedkinto be exactly the same as the one installed on the NIC.
37534fe5f3SVladimir MedvedkinThe ``rte_softrss_be`` function is a faster implementation,
38534fe5f3SVladimir Medvedkinbut it expects ``rss_key`` to be converted to the host byte order.
3928ebff11SVladimir Medvedkin
4028ebff11SVladimir Medvedkin
4128ebff11SVladimir MedvedkinPredictable RSS
4228ebff11SVladimir Medvedkin---------------
4328ebff11SVladimir Medvedkin
4428ebff11SVladimir MedvedkinIn some use cases it is useful to have a way to find partial collisions of the
4528ebff11SVladimir MedvedkinToeplitz hash function. In figure :numref:`figure_rss_queue_assign` only a few
4628ebff11SVladimir Medvedkinof the least significant bits (LSB) of the hash value are used to indicate an
4728ebff11SVladimir Medvedkinentry in the RSS Redirection Table (ReTa) and thus the index of the queue. So,
4828ebff11SVladimir Medvedkinin this case it would be useful to find another tuple whose hash has the same
4928ebff11SVladimir MedvedkinLSB's as the hash from the original tuple.
5028ebff11SVladimir Medvedkin
5128ebff11SVladimir MedvedkinFor example:
5228ebff11SVladimir Medvedkin
5328ebff11SVladimir Medvedkin- In the case of SNAT (Source Network Address Translation) it is possible to
5428ebff11SVladimir Medvedkin  find a special source port number on translation so that the hash of
5528ebff11SVladimir Medvedkin  returning packets, of the given connection, will have desired LSB's.
5628ebff11SVladimir Medvedkin- In the case of MPLS (Multiprotocol Label Switching), if the MPLS tag is used
5728ebff11SVladimir Medvedkin  in the hash calculation, the Label Switching router can allocate a special
5828ebff11SVladimir Medvedkin  MPLS tag to bind an LSP (Label Switching Path) to a given queue. This method
5928ebff11SVladimir Medvedkin  can be used with the allocation of IPSec SPI, VXLan VNI, etc., to bind the
6028ebff11SVladimir Medvedkin  tunnel to the desired queue.
6128ebff11SVladimir Medvedkin- In the case of a TCP stack, a special source port could be chosen for
6228ebff11SVladimir Medvedkin  outgoing connections, such that the response packets will be assigned to the
6328ebff11SVladimir Medvedkin  desired queue.
6428ebff11SVladimir Medvedkin
6528ebff11SVladimir MedvedkinThis functionality is provided by the API shown below.
6628ebff11SVladimir MedvedkinThe API consists of 3 parts:
6728ebff11SVladimir Medvedkin
6828ebff11SVladimir Medvedkin* Create the thash context.
6928ebff11SVladimir Medvedkin
7028ebff11SVladimir Medvedkin* Create the thash helper, associated with a context.
7128ebff11SVladimir Medvedkin
7228ebff11SVladimir Medvedkin* Use the helper run time to calculate the adjustable bits of the tuple to
7328ebff11SVladimir Medvedkin  ensure a collision.
7428ebff11SVladimir Medvedkin
7528ebff11SVladimir Medvedkin
7628ebff11SVladimir MedvedkinThash context
7728ebff11SVladimir Medvedkin~~~~~~~~~~~~~
7828ebff11SVladimir Medvedkin
7928ebff11SVladimir MedvedkinThe function ``rte_thash_init_ctx()`` initializes the context struct
8028ebff11SVladimir Medvedkinassociated with a particular NIC or a set of NICs. It expects:
8128ebff11SVladimir Medvedkin
8228ebff11SVladimir Medvedkin* The log2 value of the size of the RSS redirection table for the
8328ebff11SVladimir Medvedkin  corresponding NIC. It reflects the number of least significant bits of the
8428ebff11SVladimir Medvedkin  hash value to produce a collision for.
8528ebff11SVladimir Medvedkin
8628ebff11SVladimir Medvedkin* A predefined RSS hash key. This is optional, if ``NULL`` then a random key
8728ebff11SVladimir Medvedkin  will be initialized.
8828ebff11SVladimir Medvedkin
8928ebff11SVladimir Medvedkin* The length of the RSS hash key. This value is usually hardware/driver
9028ebff11SVladimir Medvedkin  specific and can be found in the NIC datasheet.
9128ebff11SVladimir Medvedkin
9228ebff11SVladimir Medvedkin* Optional flags, as shown below.
9328ebff11SVladimir Medvedkin
9428ebff11SVladimir MedvedkinSupported flags:
9528ebff11SVladimir Medvedkin
9628ebff11SVladimir Medvedkin* ``RTE_THASH_IGNORE_PERIOD_OVERFLOW`` - By default, and for security reasons,
9728ebff11SVladimir Medvedkin  the library prohibits generating a repeatable sequence in the hash key. This
9828ebff11SVladimir Medvedkin  flag disables such checking. The flag is mainly used for testing in the lab
9928ebff11SVladimir Medvedkin  to generate an RSS hash key with a uniform hash distribution, if the input
10028ebff11SVladimir Medvedkin  traffic also has a uniform distribution.
10128ebff11SVladimir Medvedkin
10228ebff11SVladimir Medvedkin* ``RTE_THASH_MINIMAL_SEQ`` - By default, the library generates a special bit
10328ebff11SVladimir Medvedkin  sequence in the hash key for all the bits of the subtuple. However, the
10428ebff11SVladimir Medvedkin  collision generation task requires only the ``log2(RETA_SZ)`` bits in the
10528ebff11SVladimir Medvedkin  subtuple. This flag forces the minimum bit sequence in the hash key to be
10628ebff11SVladimir Medvedkin  generated for the required ``log2(RETA_SZ)`` least significant bits of the
10728ebff11SVladimir Medvedkin  subtuple. The flag can be used in the case of a relatively large number of
10828ebff11SVladimir Medvedkin  helpers that may overlap with their corresponding bit sequences of RSS hash
10928ebff11SVladimir Medvedkin  keys.
11028ebff11SVladimir Medvedkin
11128ebff11SVladimir Medvedkin
11228ebff11SVladimir MedvedkinThash helper
11328ebff11SVladimir Medvedkin~~~~~~~~~~~~
11428ebff11SVladimir Medvedkin
11528ebff11SVladimir MedvedkinThe function ``rte_thash_add_helper()`` initializes the helper struct
11628ebff11SVladimir Medvedkinassociated with a given context and a part of a target tuple of interest which
11728ebff11SVladimir Medvedkincould be altered to produce a hash collision. On success it writes a specially
11828ebff11SVladimir Medvedkincalculated bit sequence into the RSS hash key which is stored in the context
11928ebff11SVladimir Medvedkinand calculates a table with values to be XORed with a subtuple.
12028ebff11SVladimir Medvedkin
12128ebff11SVladimir MedvedkinIt expects:
12228ebff11SVladimir Medvedkin
12328ebff11SVladimir Medvedkin* A pointer to the Thash context to be associated with.
12428ebff11SVladimir Medvedkin
12528ebff11SVladimir Medvedkin* A length of the subtuple to be modified. The length is counted in bits.
12628ebff11SVladimir Medvedkin
12728ebff11SVladimir Medvedkin* An offset of the subtuple to be modified from the beginning of the tuple. It
12828ebff11SVladimir Medvedkin  is also counted in bits.
12928ebff11SVladimir Medvedkin
13028ebff11SVladimir Medvedkin.. note::
13128ebff11SVladimir Medvedkin
13228ebff11SVladimir Medvedkin   Adding a helper changes the key stored in the corresponding context. So the
13328ebff11SVladimir Medvedkin   updated RSS hash key must be uploaded into the NIC after creating all the
13428ebff11SVladimir Medvedkin   required helpers.
13528ebff11SVladimir Medvedkin
13628ebff11SVladimir Medvedkin
13728ebff11SVladimir MedvedkinCalculation of the complementary bits to adjust the subtuple
13828ebff11SVladimir Medvedkin~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
13928ebff11SVladimir Medvedkin
14028ebff11SVladimir MedvedkinThe ``rte_thash_get_complement()`` function returns a special bit sequence
14128ebff11SVladimir Medvedkinwith length ``N = log2(rss_reta_sz)`` (for the ``rss_reta_sz`` provided at
14228ebff11SVladimir Medvedkincontext initialization) to be xored with N least significant bits of the
14328ebff11SVladimir Medvedkinsubtuple.
14428ebff11SVladimir Medvedkin
14528ebff11SVladimir MedvedkinIt expects:
14628ebff11SVladimir Medvedkin
14728ebff11SVladimir Medvedkin* A corresponding helper created for a given subtuple of the tuple.
14828ebff11SVladimir Medvedkin
14928ebff11SVladimir Medvedkin* A hash value of the tuple we want to alter.
15028ebff11SVladimir Medvedkin
15128ebff11SVladimir Medvedkin* The desired LSB's of the hash value the user expects to have.
15228ebff11SVladimir Medvedkin
15328ebff11SVladimir MedvedkinAfter the returned bit sequence has been XORed with the subtuple, the resulted
15428ebff11SVladimir MedvedkinLSB's of the new hash value, calculated from the altered tuple, will be the
15528ebff11SVladimir Medvedkinsame as in ``desired_hash``.
15628ebff11SVladimir Medvedkin
15728ebff11SVladimir Medvedkin
15828ebff11SVladimir MedvedkinAdjust tuple API
15928ebff11SVladimir Medvedkin~~~~~~~~~~~~~~~~~
16028ebff11SVladimir Medvedkin
16128ebff11SVladimir MedvedkinThe ``rte_thash_get_complement()`` function is a user-friendly wrapper around
16228ebff11SVladimir Medvedkina number of other functions. It alters a passed tuple to meet the above
16328ebff11SVladimir Medvedkinmentioned requirements around the desired hash LSB's.
16428ebff11SVladimir Medvedkin
16528ebff11SVladimir MedvedkinIt expects:
16628ebff11SVladimir Medvedkin
16728ebff11SVladimir Medvedkin* A Thash context and helper.
16828ebff11SVladimir Medvedkin
16928ebff11SVladimir Medvedkin* A pointer to the tuple to be changed.
17028ebff11SVladimir Medvedkin
17128ebff11SVladimir Medvedkin* The length of the tuple.
17228ebff11SVladimir Medvedkin
17328ebff11SVladimir Medvedkin* A callback function and its userdata to check the tuple after it has been
17428ebff11SVladimir Medvedkin  changed.
17528ebff11SVladimir Medvedkin
17628ebff11SVladimir Medvedkin* The number of attempts to change the tuple. Basically, it makes sense if
17728ebff11SVladimir Medvedkin  there is a callback and a limit on the number of attempts to change the
17828ebff11SVladimir Medvedkin  tuple, if the callback function returns an error.
17928ebff11SVladimir Medvedkin
18028ebff11SVladimir Medvedkin
18128ebff11SVladimir MedvedkinUse case example
182*9c30a6f3SHenry Nadeau----------------
18328ebff11SVladimir Medvedkin
18428ebff11SVladimir MedvedkinThere could be a number of different use cases, such as NAT, TCP stack, MPLS
18528ebff11SVladimir Medvedkintag allocation, etc. In the following we will consider a SNAT application.
18628ebff11SVladimir Medvedkin
18728ebff11SVladimir MedvedkinPackets of a single bidirectional flow belonging to different directions can
18828ebff11SVladimir Medvedkinend up being assigned to different queues and thus processed by different
18928ebff11SVladimir Medvedkinlcores, as shown in :numref:`figure_predictable_snat_1`:
19028ebff11SVladimir Medvedkin
19128ebff11SVladimir Medvedkin.. _figure_predictable_snat_1:
19228ebff11SVladimir Medvedkin
19328ebff11SVladimir Medvedkin.. figure:: img/predictable_snat_1.*
19428ebff11SVladimir Medvedkin
19528ebff11SVladimir Medvedkin   Bidirectional flow packets distribution in general
19628ebff11SVladimir Medvedkin
19728ebff11SVladimir MedvedkinThat leads to a situation where the same packet flow can be shared between two
19828ebff11SVladimir Medvedkincores. Such a situation is not ideal from a performance perspective and
19928ebff11SVladimir Medvedkinrequires extra synchronization efforts that might lead to various performance
20028ebff11SVladimir Medvedkinpenalties, for example:
20128ebff11SVladimir Medvedkin
20228ebff11SVladimir Medvedkin* The connections table is global so locking/RCU on the flow insertion/removal
20328ebff11SVladimir Medvedkin  is required.
20428ebff11SVladimir Medvedkin
20528ebff11SVladimir Medvedkin* Connection metadata must be protected to avoid race conditions.
20628ebff11SVladimir Medvedkin
20728ebff11SVladimir Medvedkin* More cache pressure if a single connection metadata is kept in different
20828ebff11SVladimir Medvedkin  L1/L2 caches of a different CPU core.
20928ebff11SVladimir Medvedkin
21028ebff11SVladimir Medvedkin* Cache pressure/less cache locality on packet handover to the different cores.
21128ebff11SVladimir Medvedkin
21228ebff11SVladimir MedvedkinWe can avoid all these penalties if it can be guaranteed that packets
21328ebff11SVladimir Medvedkinbelonging to one bidirectional flow will be assigned to the same queue, as
21428ebff11SVladimir Medvedkinshown in :numref:`figure_predictable_snat_2`:
21528ebff11SVladimir Medvedkin
21628ebff11SVladimir Medvedkin.. _figure_predictable_snat_2:
21728ebff11SVladimir Medvedkin
21828ebff11SVladimir Medvedkin.. figure:: img/predictable_snat_2.*
21928ebff11SVladimir Medvedkin
22028ebff11SVladimir Medvedkin   Bidirectional flow packets distribution with predictable RSS
22128ebff11SVladimir Medvedkin
22228ebff11SVladimir Medvedkin
22328ebff11SVladimir MedvedkinTo achieve this in a SNAT scenario it is possible to choose a source port not
22428ebff11SVladimir Medvedkinrandomly, but using the predictable RSS library to produce a partial hash
22528ebff11SVladimir Medvedkincollision. This is shown in the code below.
22628ebff11SVladimir Medvedkin
22728ebff11SVladimir Medvedkin.. code-block:: c
22828ebff11SVladimir Medvedkin
22928ebff11SVladimir Medvedkin   int key_len = 40; /* The default Niantic RSS key length. */
23028ebff11SVladimir Medvedkin
23128ebff11SVladimir Medvedkin   /** The default Niantic RSS reta size = 2^7 entries, LSBs of hash value are
23228ebff11SVladimir Medvedkin    *  used as an indexes in RSS ReTa. */
23328ebff11SVladimir Medvedkin   int reta_sz = 7;
23428ebff11SVladimir Medvedkin   int ret;
23528ebff11SVladimir Medvedkin   struct rte_thash_ctx *ctx;
23628ebff11SVladimir Medvedkin
23728ebff11SVladimir Medvedkin   uint8_t initial_key[key_len] = {0}; /* Default empty key. */
23828ebff11SVladimir Medvedkin
23928ebff11SVladimir Medvedkin   /* Create and initialize a new thash context. */
24028ebff11SVladimir Medvedkin   ctx = rte_thash_init_ctx("SNAT", key_len, reta_sz, initial_key, 0);
24128ebff11SVladimir Medvedkin
24228ebff11SVladimir Medvedkin   /** Add a helper and specify the variable tuple part and its length. In the
24328ebff11SVladimir Medvedkin    *  SNAT case we want to choose a new source port on SNAT translation in a
24428ebff11SVladimir Medvedkin    *  way that the reverse tuple will have the same LSBs as the original
24528ebff11SVladimir Medvedkin    *  direction tuple so that the selected source port will be the
24628ebff11SVladimir Medvedkin    *  destination port on reply.
24728ebff11SVladimir Medvedkin    */
24828ebff11SVladimir Medvedkin   ret = rte_thash_add_helper(ctx, "snat", sizeof(uint16_t) * 8,
24928ebff11SVladimir Medvedkin                              offsetof(union rte_thash_tuple, v4.dport) * 8);
25028ebff11SVladimir Medvedkin
25128ebff11SVladimir Medvedkin   if (ret != 0)
25228ebff11SVladimir Medvedkin       return ret;
25328ebff11SVladimir Medvedkin
25428ebff11SVladimir Medvedkin   /* Get handler of the required helper. */
25528ebff11SVladimir Medvedkin   struct rte_thash_subtuple_helper *h = rte_thash_get_helper(ctx, "snat");
25628ebff11SVladimir Medvedkin
25728ebff11SVladimir Medvedkin   /** After calling rte_thash_add_helper() the initial_key passed on ctx
25828ebff11SVladimir Medvedkin    *  creation has been changed so we get the new one.
25928ebff11SVladimir Medvedkin    */
26028ebff11SVladimir Medvedkin   uint8_t *new_key = rte_thash_get_key(ctx);
26128ebff11SVladimir Medvedkin
26228ebff11SVladimir Medvedkin   union rte_thash_tuple tuple, rev_tuple;
26328ebff11SVladimir Medvedkin
26428ebff11SVladimir Medvedkin   /* A complete tuple from the packet. */
26528ebff11SVladimir Medvedkin   complete_tuple(mbuf, &tuple);
26628ebff11SVladimir Medvedkin
26728ebff11SVladimir Medvedkin   /* Calculate the RSS hash or get it from mbuf->hash.rss. */
26828ebff11SVladimir Medvedkin   uint32_t orig_hash = rte_softrss((uint32_t *)&tuple, RTE_THASH_V4_L4_LEN, new_key);
26928ebff11SVladimir Medvedkin
27028ebff11SVladimir Medvedkin   /** Complete the reverse tuple by translating the SRC address and swapping
27128ebff11SVladimir Medvedkin    *  src and dst addresses and ports.
27228ebff11SVladimir Medvedkin    */
27328ebff11SVladimir Medvedkin   get_rev_tuple(&rev_tuple, &tuple, new_ip);
27428ebff11SVladimir Medvedkin
27528ebff11SVladimir Medvedkin   /* Calculate the expected rss hash for the reverse tuple. */
27628ebff11SVladimir Medvedkin   uint32_t rev_hash = rte_softrss((uint32_t *)&rev_tuple, RTE_THASH_V4_L4_LEN, new_key);
27728ebff11SVladimir Medvedkin
27828ebff11SVladimir Medvedkin   /* Get the adjustment bits for the src port to get a new port. */
27928ebff11SVladimir Medvedkin   uint32_t adj = rte_thash_get_compliment(h, rev_hash, orig_hash);
28028ebff11SVladimir Medvedkin
28128ebff11SVladimir Medvedkin   /* Adjust the source port bits. */
28228ebff11SVladimir Medvedkin   uint16_t new_sport = tuple.v4.sport ^ adj;
28328ebff11SVladimir Medvedkin
28428ebff11SVladimir Medvedkin   /* Make an actual packet translation. */
28528ebff11SVladimir Medvedkin   do_snat(mbuf, new_ip, new_sport);
286