DPDK patches and discussions
 help / color / mirror / Atom feed
From: "Chilikin, Andrey" <andrey.chilikin@intel.com>
To: Vladimir Medvedkin <medvedkinv@gmail.com>
Cc: "dev@dpdk.org" <dev@dpdk.org>
Subject: Re: [dpdk-dev] [PATCH v2] Add toeplitz hash algorithm used by RSS
Date: Thu, 7 May 2015 11:38:16 +0000	[thread overview]
Message-ID: <AAC06825A3B29643AF5372F5E0DDF053350C7DF4@IRSMSX106.ger.corp.intel.com> (raw)
In-Reply-To: <CANDrEH=uUjgE8aVsFJ6qSWnh3bZfXjKyyx3xfQFfjiHV0DWw6A@mail.gmail.com>

Hi Vladimir,

Yes, at the moment NICs support limited input sets for hash calculation, but why limit SW for the same sets if it can be done in more general way and be easily scalable for HW updates? Using limited input set for RSS is not a feature of Toeplitz hash, but limitation of HW. I believe that general Toeplitz function will be more appropriate – it will cover input sets currently supported by HW and also will be easily scalable for future HW. Also, talking about different NICs – Niantic and Fortville, for example, have hash keys of different length, so rte_softrss() function should take into account hash key’s length as well.
Regards,
Andrey


From: Vladimir Medvedkin [mailto:medvedkinv@gmail.com]
Sent: Thursday, May 7, 2015 11:28 AM
To: Chilikin, Andrey
Cc: dev@dpdk.org
Subject: Re: [dpdk-dev] [PATCH v2] Add toeplitz hash algorithm used by RSS

Hi Andrey,

The main goal of this new functions is to calculate the hash which is equal to the hash of the NIC.
According to XL710 datasheet table 7-5 for sctp input set consists of IP4-S, IP4-D, SCTP-Verification-Tag. I don't see any NIC that uses QinQ or single vlan tag, ip proto number, tunnel id, vxlan, etc for calculating RSS hash. If it appear we can always update union rte_thash_tuple.
I think it should be like:
struct rte_ports {
        uint16_t dport;
        uint16_t sport;
};

union rte_thash_l4 {
        struct          rte_ports ports;
        uint32_t        sctp_tag;
};
struct rte_ipv4_tuple {
        uint32_t        src_addr;
        uint32_t        dst_addr;
        union rte_thash_l4 l4;
};
If it is necessary to distribute packets according to non standart tuples I think it's more appropriate to use crc32 or jhash because of speed. rte_softrss_be consumes 400-500 clocks for each 4-byte input at E3 1230v1@3.2GHz<mailto:1230v1@3.2GHz>. This means for ipv4+tcp it consumes ~1500 clocks.
If you or someone still think you need general toeplitz hash I'll add it.
Regards,
Vladimir


2015-05-05 19:03 GMT+03:00 Chilikin, Andrey <andrey.chilikin@intel.com<mailto:andrey.chilikin@intel.com>>:
Hi Vladimir,

Why limit Toeplitz hash calculation to predefined tuples and length? Should it be more general, something like
rte_softrss_be(void *input, uint32_t input_len, const uint8_t *rss_key) to enable hash calculation for an input of any size? It would be useful for distributing packets using some non-standard tuples, like hashing on QinQ or adding IP protocol to hash calculation to separate UDP and TCP flows or even some other fields from a packet, for example, tunnel ID from VXLAN headers. By the way, i40e already supports RSS for SCTP in addition to TCP and UDP and includes Verification Tag as well as SCTP source and destination ports for RSS hash.

Regards,
Andrey

> -----Original Message-----
> From: dev [mailto:dev-bounces@dpdk.org<mailto:dev-bounces@dpdk.org>] On Behalf Of Vladimir
> Medvedkin
> Sent: Tuesday, May 5, 2015 2:20 PM
> To: dev@dpdk.org<mailto:dev@dpdk.org>
> Subject: [dpdk-dev] [PATCH v2] Add toeplitz hash algorithm used by RSS
>
> Software implementation of the Toeplitz hash function used by RSS.
> Can be used either for packet distribution on single queue NIC or for
> simulating of RSS computation on specific NIC (for example after GRE header
> decapsulating).
>
> v2 changes
> - Add ipv6 support
> - Various style fixes
>
> Signed-off-by: Vladimir Medvedkin <medvedkinv@gmail.com<mailto:medvedkinv@gmail.com>>
> ---
>  lib/librte_hash/Makefile    |   1 +
>  lib/librte_hash/rte_thash.h | 209
> ++++++++++++++++++++++++++++++++++++++++++++
>  2 files changed, 210 insertions(+)
>  create mode 100644 lib/librte_hash/rte_thash.h
>
> diff --git a/lib/librte_hash/Makefile b/lib/librte_hash/Makefile index
> 3696cb1..981230b 100644
> --- a/lib/librte_hash/Makefile
> +++ b/lib/librte_hash/Makefile
> @@ -49,6 +49,7 @@ SRCS-$(CONFIG_RTE_LIBRTE_HASH) += rte_fbk_hash.c
> SYMLINK-$(CONFIG_RTE_LIBRTE_HASH)-include := rte_hash.h  SYMLINK-
> $(CONFIG_RTE_LIBRTE_HASH)-include += rte_hash_crc.h  SYMLINK-
> $(CONFIG_RTE_LIBRTE_HASH)-include += rte_jhash.h
> +SYMLINK-$(CONFIG_RTE_LIBRTE_HASH)-include += rte_thash.h
>  SYMLINK-$(CONFIG_RTE_LIBRTE_HASH)-include += rte_fbk_hash.h
>
>  # this lib needs eal
> diff --git a/lib/librte_hash/rte_thash.h b/lib/librte_hash/rte_thash.h new file
> mode 100644 index 0000000..42c7bf6
> --- /dev/null
> +++ b/lib/librte_hash/rte_thash.h
> @@ -0,0 +1,209 @@
> +/*-
> + *   BSD LICENSE
> + *
> + *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
> + *   All rights reserved.
> + *
> + *   Redistribution and use in source and binary forms, with or without
> + *   modification, are permitted provided that the following conditions
> + *   are met:
> + *
> + *     * Redistributions of source code must retain the above copyright
> + *       notice, this list of conditions and the following disclaimer.
> + *     * Redistributions in binary form must reproduce the above copyright
> + *       notice, this list of conditions and the following disclaimer in
> + *       the documentation and/or other materials provided with the
> + *       distribution.
> + *     * Neither the name of Intel Corporation nor the names of its
> + *       contributors may be used to endorse or promote products derived
> + *       from this software without specific prior written permission.
> + *
> + *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND
> CONTRIBUTORS
> + *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT
> NOT
> + *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
> FITNESS FOR
> + *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
> COPYRIGHT
> + *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
> INCIDENTAL,
> + *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
> NOT
> + *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
> OF USE,
> + *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
> AND ON ANY
> + *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR
> TORT
> + *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF
> THE USE
> + *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
> DAMAGE.
> + */
> +
> +#ifndef _RTE_THASH_H
> +#define _RTE_THASH_H
> +
> +/**
> + * @file
> + *
> + * toeplitz hash functions.
> + */
> +
> +#ifdef __cplusplus
> +extern "C" {
> +#endif
> +
> +/**
> + * Software implementation of the Toeplitz hash function used by RSS.
> + * Can be used either for packet distribution on single queue NIC
> + * or for simulating of RSS computation on specific NIC (for example
> + * after GRE header decapsulating)
> + */
> +
> +#include <stdint.h>
> +#include <rte_byteorder.h>
> +#include <rte_vect.h>
> +
> +#ifdef __SSE3__
> +static const __m128i bswap_mask = {0x0405060700010203,
> +0x0C0D0E0F08090A0B}; #endif
> +
> +enum rte_thash_len {
> +     RTE_THASH_V4_L3 = 2,    /*calculate hash of ipv4 header
> only*/
> +     RTE_THASH_V4_L4 = 3,    /*calculate hash of ipv4 + transport
> headers*/
> +     RTE_THASH_V6_L3 = 8,    /*calculate hash of ipv6 header only
> */
> +     RTE_THASH_V6_L4 = 9,    /*calculate hash of ipv6 + transport
> headers */
> +};
> +
> +/**
> + * IPv4 tuple
> + * addreses and ports have to be CPU byte order  */ struct
> +rte_ipv4_tuple {
> +     uint32_t        src_addr;
> +     uint32_t        dst_addr;
> +     uint16_t        dport;
> +     uint16_t        sport;
> +};
> +
> +/**
> + * IPv6 tuple
> + * Addresses have to be filled by rte_thash_load_v6_addr()
> + * ports have to be CPU byte order
> + */
> +struct rte_ipv6_tuple {
> +     uint8_t         src_addr[16];
> +     uint8_t         dst_addr[16];
> +     uint16_t        dport;
> +     uint16_t        sport;
> +};
> +
> +union rte_thash_tuple {
> +     struct rte_ipv4_tuple   v4;
> +     struct rte_ipv6_tuple   v6;
> +} __attribute__((aligned(16)));
> +
> +/**
> + * Prepare special converted key to use with rte_softrss_be()
> + * @param orig
> + *   pointer to original RSS key
> + * @param targ
> + *   pointer to target RSS key
> + * @param len
> + *   RSS key length
> + */
> +static inline void
> +rte_convert_rss_key(const uint32_t *orig, uint32_t *targ, int len) {
> +     int i;
> +
> +     for (i = 0; i < (len >> 2); i++) {
> +             targ[i] = rte_be_to_cpu_32(orig[i]);
> +     }
> +}
> +
> +/**
> + * Prepare and load IPv6 address
> + * @param orig
> + *   Pointer to ipv6 address inside ipv6_hdr
> + * @param targ
> + *   Pointer to ipv6 address inside rte_ipv6_tuple
> + */
> +static inline void
> +rte_thash_load_v6_addr(const uint8_t *orig, uint8_t *targ) { #ifdef
> +__SSE3__
> +     __m128i ipv6 = _mm_loadu_si128((const __m128i *)orig);
> +     *(__m128i *)targ = _mm_shuffle_epi8(ipv6, bswap_mask); #else
> +     int i;
> +
> +     for (i = 0; i < 4; i++) {
> +             *((uint32_t *)targ + i) =
> +                     rte_be_to_cpu_32(*((const uint32_t *)orig + i));
> +     }
> +#endif
> +}
> +
> +/**
> + * Generic implementation. Can be used with original rss_key
> + * @param input_tuple
> + *   Pointer to rte_thash_tuple union
> + * @param input_len
> + *   Length of input_tuple in 4-bytes chunks
> + *   RTE_THASH_V4_L3:        calculate hash of IPv4 src address and IPv4 dst
> address
> + *   RTE_THASH_V4_L4 calculate hash of IPv4 adresses and TCP|UDP
> ports
> + *   RTE_THASH_V6_L3:        calculate hash of IPv6 src address and IPv4 dst
> address
> + *   RTE_THASH_V6_L4 calculate hash of IPv6 adresses and TCP|UDP
> ports
> + * @param rss_key
> + *   Pointer to RSS hash key.
> + * @return
> + *   Calculated hash value.
> + */
> +static inline uint32_t
> +rte_softrss(union rte_thash_tuple *input_tuple, enum rte_thash_len
> input_len,
> +             const uint8_t *rss_key)
> +{
> +     uint32_t i, j, ret = 0;
> +
> +     for (j = 0; j < input_len; j++) {
> +             for (i = 0; i < 32; i++) {
> +                     if (((uint32_t *)input_tuple)[j] & (1 << (31 - i))) {
> +                             ret ^= rte_cpu_to_be_32(((const uint32_t
> *)rss_key)[j]) << i |
> +
>       (uint32_t)((uint64_t)(rte_cpu_to_be_32(((const uint32_t
> *)rss_key)[j + 1])) >> (32 - i));
> +                     }
> +             }
> +     }
> +     return ret;
> +}
> +
> +/**
> + * Optimized implementation.
> + * If you want the calculated hash value matches NIC RSS value
> + * you have to use special converted key.
> + * @param input_tuple
> + *   Pointer to rte_thash_tuple union
> + * @param input_len
> + *   Length of input_tuple in 4-bytes chunks
> + *   RTE_THASH_V4_L3:        calculate hash of IPv4 src address and IPv4 dst
> address
> + *   RTE_THASH_V4_L4 calculate hash of IPv4 adresses and TCP|UDP
> ports
> + *   RTE_THASH_V6_L3:        calculate hash of IPv6 src address and IPv4 dst
> address
> + *   RTE_THASH_V6_L4 calculate hash of IPv6 adresses and TCP|UDP
> ports
> + * @param *rss_key
> + *   Pointer to RSS hash key.
> + * @return
> + *   Calculated hash value.
> + */
> +static inline uint32_t
> +rte_softrss_be(union rte_thash_tuple *input_tuple, enum rte_thash_len
> input_len,
> +             const uint8_t *rss_key)
> +{
> +     uint32_t i, j, ret = 0;
> +
> +     for (j = 0; j < input_len; j++) {
> +             for (i = 0; i < 32; i++) {
> +                     if (((uint32_t *)input_tuple)[j] & (1 << (31 - i))) {
> +                             ret ^= ((const uint32_t *)rss_key)[j] << i |
> +                                     (uint32_t)((uint64_t)(((const uint32_t
> *)rss_key)[j + 1]) >> (32 - i));
> +                     }
> +             }
> +     }
> +     return ret;
> +}
> +
> +#ifdef __cplusplus
> +}
> +#endif
> +
> +#endif /* _RTE_THASH_H */
> --
> 1.8.3.2


  reply	other threads:[~2015-05-07 11:38 UTC|newest]

Thread overview: 37+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2015-04-08 19:06 [dpdk-dev] [PATCH] Add toeplitz hash algorithm Vladimir Medvedkin
2015-04-08 22:24 ` Stephen Hemminger
2015-04-09 12:50   ` Vladimir Medvedkin
2015-04-09  6:37 ` Gleb Natapov
2015-04-09 12:57   ` Vladimir Medvedkin
2015-05-05 13:20 ` [dpdk-dev] [PATCH v2] Add toeplitz hash algorithm used by RSS Vladimir Medvedkin
2015-05-05 16:03   ` Chilikin, Andrey
2015-05-07 10:28     ` Vladimir Medvedkin
2015-05-07 11:38       ` Chilikin, Andrey [this message]
2015-05-08 14:58         ` Vladimir Medvedkin
2015-05-08 14:58   ` Vladimir Medvedkin
2015-06-03 14:07     ` Thomas Monjalon
2015-06-16  9:07       ` Thomas Monjalon
2015-06-16 10:36         ` Bruce Richardson
2015-06-16 12:29     ` Bruce Richardson
2015-06-16 19:26       ` Vladimir Medvedkin
2015-06-19 14:55     ` [dpdk-dev] [PATCH v4] " Vladimir Medvedkin
2015-06-19 15:59       ` Richardson, Bruce
2015-06-19 16:14         ` Vladimir Medvedkin
2015-06-29 12:18           ` Bruce Richardson
2015-06-19 16:33         ` Mcnamara, John
2015-06-19 17:31       ` [dpdk-dev] [PATCH v5] " Vladimir Medvedkin
2015-06-29 12:40         ` Bruce Richardson
2015-06-30 12:14           ` Vladimir Medvedkin
2015-06-30 12:48             ` Bruce Richardson
2015-06-30 23:40         ` [dpdk-dev] [PATCH v6] " Vladimir Medvedkin
2015-07-01 15:29           ` Bruce Richardson
2015-07-01 21:23             ` Thomas Monjalon
2015-07-22  7:55           ` Tony Lu
2015-07-27 11:57             ` Vladimir Medvedkin
2015-07-29  5:01               ` Qiu, Michael
2015-07-29 14:00                 ` Vladimir Medvedkin
2015-06-19 17:32       ` [dpdk-dev] [PATCH v3] Add unit test for thash library Vladimir Medvedkin
2015-06-29 12:42         ` Bruce Richardson
2015-06-30 23:41         ` [dpdk-dev] [PATCH v4] " Vladimir Medvedkin
2015-07-01 15:30           ` Bruce Richardson
2015-07-01 21:28             ` Thomas Monjalon

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=AAC06825A3B29643AF5372F5E0DDF053350C7DF4@IRSMSX106.ger.corp.intel.com \
    --to=andrey.chilikin@intel.com \
    --cc=dev@dpdk.org \
    --cc=medvedkinv@gmail.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).