UPSTREAM: lib: Add xxhash module
authorNick Terrell <terrelln@fb.com>
Fri, 4 Aug 2017 20:19:17 +0000 (13:19 -0700)
committerivanmeler <i_ivan@windowslive.com>
Wed, 13 Apr 2022 21:13:23 +0000 (21:13 +0000)
Adds xxhash kernel module with xxh32 and xxh64 hashes. xxhash is an
extremely fast non-cryptographic hash algorithm for checksumming.
The zstd compression and decompression modules added in the next patch
require xxhash. I extracted it out from zstd since it is useful on its
own. I copied the code from the upstream XXHash source repository and
translated it into kernel style. I ran benchmarks and tests in the kernel
and tests in userland.

I benchmarked xxhash as a special character device. I ran in four modes,
no-op, xxh32, xxh64, and crc32. The no-op mode simply copies the data to
kernel space and ignores it. The xxh32, xxh64, and crc32 modes compute
hashes on the copied data. I also ran it with four different buffer sizes.
The benchmark file is located in the upstream zstd source repository under
`contrib/linux-kernel/xxhash_test.c` [1].

I ran the benchmarks on a Ubuntu 14.04 VM with 2 cores and 4 GiB of RAM.
The VM is running on a MacBook Pro with a 3.1 GHz Intel Core i7 processor,
16 GB of RAM, and a SSD. I benchmarked using the file `filesystem.squashfs`
from `ubuntu-16.10-desktop-amd64.iso`, which is 1,536,217,088 B large.
Run the following commands for the benchmark:

    modprobe xxhash_test
    mknod xxhash_test c 245 0
    time cp filesystem.squashfs xxhash_test

The time is reported by the time of the userland `cp`.
The GB/s is computed with

    1,536,217,008 B / time(buffer size, hash)

which includes the time to copy from userland.
The Normalized GB/s is computed with

    1,536,217,088 B / (time(buffer size, hash) - time(buffer size, none)).

| Buffer Size (B) | Hash  | Time (s) | GB/s | Adjusted GB/s |
|-----------------|-------|----------|------|---------------|
|            1024 | none  |    0.408 | 3.77 |             - |
|            1024 | xxh32 |    0.649 | 2.37 |          6.37 |
|            1024 | xxh64 |    0.542 | 2.83 |         11.46 |
|            1024 | crc32 |    1.290 | 1.19 |          1.74 |
|            4096 | none  |    0.380 | 4.04 |             - |
|            4096 | xxh32 |    0.645 | 2.38 |          5.79 |
|            4096 | xxh64 |    0.500 | 3.07 |         12.80 |
|            4096 | crc32 |    1.168 | 1.32 |          1.95 |
|            8192 | none  |    0.351 | 4.38 |             - |
|            8192 | xxh32 |    0.614 | 2.50 |          5.84 |
|            8192 | xxh64 |    0.464 | 3.31 |         13.60 |
|            8192 | crc32 |    1.163 | 1.32 |          1.89 |
|           16384 | none  |    0.346 | 4.43 |             - |
|           16384 | xxh32 |    0.590 | 2.60 |          6.30 |
|           16384 | xxh64 |    0.466 | 3.30 |         12.80 |
|           16384 | crc32 |    1.183 | 1.30 |          1.84 |

Tested in userland using the test-suite in the zstd repo under
`contrib/linux-kernel/test/XXHashUserlandTest.cpp` [2] by mocking the
kernel functions. A line in each branch of every function in `xxhash.c`
was commented out to ensure that the test-suite fails. Additionally
tested while testing zstd and with SMHasher [3].

[1] https://phabricator.intern.facebook.com/P57526246
[2] https://github.com/facebook/zstd/blob/dev/contrib/linux-kernel/test/XXHashUserlandTest.cpp
[3] https://github.com/aappleby/smhasher

zstd source repository: https://github.com/facebook/zstd
XXHash source repository: https://github.com/cyan4973/xxhash

Signed-off-by: Nick Terrell <terrelln@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
(cherry picked from commit 5d2405227a9eaea48e8cc95756a06d407b11f141)
Signed-off-by: Peter Kalauskas <peskal@google.com>
Bug: 112488418
Change-Id: I4b63e96457f17cf455591e8f35058dacd7aa9004

include/linux/xxhash.h [new file with mode: 0644]
lib/Kconfig
lib/Makefile
lib/xxhash.c [new file with mode: 0644]

diff --git a/include/linux/xxhash.h b/include/linux/xxhash.h
new file mode 100644 (file)
index 0000000..9e1f42c
--- /dev/null
@@ -0,0 +1,236 @@
+/*
+ * xxHash - Extremely Fast Hash algorithm
+ * Copyright (C) 2012-2016, Yann Collet.
+ *
+ * BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+ *
+ * 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.
+ *
+ * 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.
+ *
+ * This program is free software; you can redistribute it and/or modify it under
+ * the terms of the GNU General Public License version 2 as published by the
+ * Free Software Foundation. This program is dual-licensed; you may select
+ * either version 2 of the GNU General Public License ("GPL") or BSD license
+ * ("BSD").
+ *
+ * You can contact the author at:
+ * - xxHash homepage: http://cyan4973.github.io/xxHash/
+ * - xxHash source repository: https://github.com/Cyan4973/xxHash
+ */
+
+/*
+ * Notice extracted from xxHash homepage:
+ *
+ * xxHash is an extremely fast Hash algorithm, running at RAM speed limits.
+ * It also successfully passes all tests from the SMHasher suite.
+ *
+ * Comparison (single thread, Windows Seven 32 bits, using SMHasher on a Core 2
+ * Duo @3GHz)
+ *
+ * Name            Speed       Q.Score   Author
+ * xxHash          5.4 GB/s     10
+ * CrapWow         3.2 GB/s      2       Andrew
+ * MumurHash 3a    2.7 GB/s     10       Austin Appleby
+ * SpookyHash      2.0 GB/s     10       Bob Jenkins
+ * SBox            1.4 GB/s      9       Bret Mulvey
+ * Lookup3         1.2 GB/s      9       Bob Jenkins
+ * SuperFastHash   1.2 GB/s      1       Paul Hsieh
+ * CityHash64      1.05 GB/s    10       Pike & Alakuijala
+ * FNV             0.55 GB/s     5       Fowler, Noll, Vo
+ * CRC32           0.43 GB/s     9
+ * MD5-32          0.33 GB/s    10       Ronald L. Rivest
+ * SHA1-32         0.28 GB/s    10
+ *
+ * Q.Score is a measure of quality of the hash function.
+ * It depends on successfully passing SMHasher test set.
+ * 10 is a perfect score.
+ *
+ * A 64-bits version, named xxh64 offers much better speed,
+ * but for 64-bits applications only.
+ * Name     Speed on 64 bits    Speed on 32 bits
+ * xxh64       13.8 GB/s            1.9 GB/s
+ * xxh32        6.8 GB/s            6.0 GB/s
+ */
+
+#ifndef XXHASH_H
+#define XXHASH_H
+
+#include <linux/types.h>
+
+/*-****************************
+ * Simple Hash Functions
+ *****************************/
+
+/**
+ * xxh32() - calculate the 32-bit hash of the input with a given seed.
+ *
+ * @input:  The data to hash.
+ * @length: The length of the data to hash.
+ * @seed:   The seed can be used to alter the result predictably.
+ *
+ * Speed on Core 2 Duo @ 3 GHz (single thread, SMHasher benchmark) : 5.4 GB/s
+ *
+ * Return:  The 32-bit hash of the data.
+ */
+uint32_t xxh32(const void *input, size_t length, uint32_t seed);
+
+/**
+ * xxh64() - calculate the 64-bit hash of the input with a given seed.
+ *
+ * @input:  The data to hash.
+ * @length: The length of the data to hash.
+ * @seed:   The seed can be used to alter the result predictably.
+ *
+ * This function runs 2x faster on 64-bit systems, but slower on 32-bit systems.
+ *
+ * Return:  The 64-bit hash of the data.
+ */
+uint64_t xxh64(const void *input, size_t length, uint64_t seed);
+
+/*-****************************
+ * Streaming Hash Functions
+ *****************************/
+
+/*
+ * These definitions are only meant to allow allocation of XXH state
+ * statically, on stack, or in a struct for example.
+ * Do not use members directly.
+ */
+
+/**
+ * struct xxh32_state - private xxh32 state, do not use members directly
+ */
+struct xxh32_state {
+       uint32_t total_len_32;
+       uint32_t large_len;
+       uint32_t v1;
+       uint32_t v2;
+       uint32_t v3;
+       uint32_t v4;
+       uint32_t mem32[4];
+       uint32_t memsize;
+};
+
+/**
+ * struct xxh32_state - private xxh64 state, do not use members directly
+ */
+struct xxh64_state {
+       uint64_t total_len;
+       uint64_t v1;
+       uint64_t v2;
+       uint64_t v3;
+       uint64_t v4;
+       uint64_t mem64[4];
+       uint32_t memsize;
+};
+
+/**
+ * xxh32_reset() - reset the xxh32 state to start a new hashing operation
+ *
+ * @state: The xxh32 state to reset.
+ * @seed:  Initialize the hash state with this seed.
+ *
+ * Call this function on any xxh32_state to prepare for a new hashing operation.
+ */
+void xxh32_reset(struct xxh32_state *state, uint32_t seed);
+
+/**
+ * xxh32_update() - hash the data given and update the xxh32 state
+ *
+ * @state:  The xxh32 state to update.
+ * @input:  The data to hash.
+ * @length: The length of the data to hash.
+ *
+ * After calling xxh32_reset() call xxh32_update() as many times as necessary.
+ *
+ * Return:  Zero on success, otherwise an error code.
+ */
+int xxh32_update(struct xxh32_state *state, const void *input, size_t length);
+
+/**
+ * xxh32_digest() - produce the current xxh32 hash
+ *
+ * @state: Produce the current xxh32 hash of this state.
+ *
+ * A hash value can be produced at any time. It is still possible to continue
+ * inserting input into the hash state after a call to xxh32_digest(), and
+ * generate new hashes later on, by calling xxh32_digest() again.
+ *
+ * Return: The xxh32 hash stored in the state.
+ */
+uint32_t xxh32_digest(const struct xxh32_state *state);
+
+/**
+ * xxh64_reset() - reset the xxh64 state to start a new hashing operation
+ *
+ * @state: The xxh64 state to reset.
+ * @seed:  Initialize the hash state with this seed.
+ */
+void xxh64_reset(struct xxh64_state *state, uint64_t seed);
+
+/**
+ * xxh64_update() - hash the data given and update the xxh64 state
+ * @state:  The xxh64 state to update.
+ * @input:  The data to hash.
+ * @length: The length of the data to hash.
+ *
+ * After calling xxh64_reset() call xxh64_update() as many times as necessary.
+ *
+ * Return:  Zero on success, otherwise an error code.
+ */
+int xxh64_update(struct xxh64_state *state, const void *input, size_t length);
+
+/**
+ * xxh64_digest() - produce the current xxh64 hash
+ *
+ * @state: Produce the current xxh64 hash of this state.
+ *
+ * A hash value can be produced at any time. It is still possible to continue
+ * inserting input into the hash state after a call to xxh64_digest(), and
+ * generate new hashes later on, by calling xxh64_digest() again.
+ *
+ * Return: The xxh64 hash stored in the state.
+ */
+uint64_t xxh64_digest(const struct xxh64_state *state);
+
+/*-**************************
+ * Utils
+ ***************************/
+
+/**
+ * xxh32_copy_state() - copy the source state into the destination state
+ *
+ * @src: The source xxh32 state.
+ * @dst: The destination xxh32 state.
+ */
+void xxh32_copy_state(struct xxh32_state *dst, const struct xxh32_state *src);
+
+/**
+ * xxh64_copy_state() - copy the source state into the destination state
+ *
+ * @src: The source xxh64 state.
+ * @dst: The destination xxh64 state.
+ */
+void xxh64_copy_state(struct xxh64_state *dst, const struct xxh64_state *src);
+
+#endif /* XXHASH_H */
index 48635688046f3fa86a5b46fdfb7eca3fa7b9c4c5..45f6b279d4a06a28a3f2524e943f4aa34948bbd8 100644 (file)
@@ -185,6 +185,9 @@ config CRC8
          when they need to do cyclic redundancy check according CRC8
          algorithm. Module will be called crc8.
 
+config XXHASH
+       tristate
+
 config AUDIT_GENERIC
        bool
        depends on AUDIT && !AUDIT_ARCH
index b9ad86add71dc7ca8fd07a1dd9bac00ee8fcc88a..eb5af4a37c7f40fabc88eabf920d66d6998e7de5 100644 (file)
@@ -93,6 +93,7 @@ obj-$(CONFIG_CRC32)   += crc32.o
 obj-$(CONFIG_CRC7)     += crc7.o
 obj-$(CONFIG_LIBCRC32C)        += libcrc32c.o
 obj-$(CONFIG_CRC8)     += crc8.o
+obj-$(CONFIG_XXHASH)   += xxhash.o
 obj-$(CONFIG_GENERIC_ALLOCATOR) += genalloc.o
 
 obj-$(CONFIG_842_COMPRESS) += 842/
diff --git a/lib/xxhash.c b/lib/xxhash.c
new file mode 100644 (file)
index 0000000..aa61e2a
--- /dev/null
@@ -0,0 +1,500 @@
+/*
+ * xxHash - Extremely Fast Hash algorithm
+ * Copyright (C) 2012-2016, Yann Collet.
+ *
+ * BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+ *
+ * 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.
+ *
+ * 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.
+ *
+ * This program is free software; you can redistribute it and/or modify it under
+ * the terms of the GNU General Public License version 2 as published by the
+ * Free Software Foundation. This program is dual-licensed; you may select
+ * either version 2 of the GNU General Public License ("GPL") or BSD license
+ * ("BSD").
+ *
+ * You can contact the author at:
+ * - xxHash homepage: http://cyan4973.github.io/xxHash/
+ * - xxHash source repository: https://github.com/Cyan4973/xxHash
+ */
+
+#include <asm/unaligned.h>
+#include <linux/errno.h>
+#include <linux/compiler.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/string.h>
+#include <linux/xxhash.h>
+
+/*-*************************************
+ * Macros
+ **************************************/
+#define xxh_rotl32(x, r) ((x << r) | (x >> (32 - r)))
+#define xxh_rotl64(x, r) ((x << r) | (x >> (64 - r)))
+
+#ifdef __LITTLE_ENDIAN
+# define XXH_CPU_LITTLE_ENDIAN 1
+#else
+# define XXH_CPU_LITTLE_ENDIAN 0
+#endif
+
+/*-*************************************
+ * Constants
+ **************************************/
+static const uint32_t PRIME32_1 = 2654435761U;
+static const uint32_t PRIME32_2 = 2246822519U;
+static const uint32_t PRIME32_3 = 3266489917U;
+static const uint32_t PRIME32_4 =  668265263U;
+static const uint32_t PRIME32_5 =  374761393U;
+
+static const uint64_t PRIME64_1 = 11400714785074694791ULL;
+static const uint64_t PRIME64_2 = 14029467366897019727ULL;
+static const uint64_t PRIME64_3 =  1609587929392839161ULL;
+static const uint64_t PRIME64_4 =  9650029242287828579ULL;
+static const uint64_t PRIME64_5 =  2870177450012600261ULL;
+
+/*-**************************
+ *  Utils
+ ***************************/
+void xxh32_copy_state(struct xxh32_state *dst, const struct xxh32_state *src)
+{
+       memcpy(dst, src, sizeof(*dst));
+}
+EXPORT_SYMBOL(xxh32_copy_state);
+
+void xxh64_copy_state(struct xxh64_state *dst, const struct xxh64_state *src)
+{
+       memcpy(dst, src, sizeof(*dst));
+}
+EXPORT_SYMBOL(xxh64_copy_state);
+
+/*-***************************
+ * Simple Hash Functions
+ ****************************/
+static uint32_t xxh32_round(uint32_t seed, const uint32_t input)
+{
+       seed += input * PRIME32_2;
+       seed = xxh_rotl32(seed, 13);
+       seed *= PRIME32_1;
+       return seed;
+}
+
+uint32_t xxh32(const void *input, const size_t len, const uint32_t seed)
+{
+       const uint8_t *p = (const uint8_t *)input;
+       const uint8_t *b_end = p + len;
+       uint32_t h32;
+
+       if (len >= 16) {
+               const uint8_t *const limit = b_end - 16;
+               uint32_t v1 = seed + PRIME32_1 + PRIME32_2;
+               uint32_t v2 = seed + PRIME32_2;
+               uint32_t v3 = seed + 0;
+               uint32_t v4 = seed - PRIME32_1;
+
+               do {
+                       v1 = xxh32_round(v1, get_unaligned_le32(p));
+                       p += 4;
+                       v2 = xxh32_round(v2, get_unaligned_le32(p));
+                       p += 4;
+                       v3 = xxh32_round(v3, get_unaligned_le32(p));
+                       p += 4;
+                       v4 = xxh32_round(v4, get_unaligned_le32(p));
+                       p += 4;
+               } while (p <= limit);
+
+               h32 = xxh_rotl32(v1, 1) + xxh_rotl32(v2, 7) +
+                       xxh_rotl32(v3, 12) + xxh_rotl32(v4, 18);
+       } else {
+               h32 = seed + PRIME32_5;
+       }
+
+       h32 += (uint32_t)len;
+
+       while (p + 4 <= b_end) {
+               h32 += get_unaligned_le32(p) * PRIME32_3;
+               h32 = xxh_rotl32(h32, 17) * PRIME32_4;
+               p += 4;
+       }
+
+       while (p < b_end) {
+               h32 += (*p) * PRIME32_5;
+               h32 = xxh_rotl32(h32, 11) * PRIME32_1;
+               p++;
+       }
+
+       h32 ^= h32 >> 15;
+       h32 *= PRIME32_2;
+       h32 ^= h32 >> 13;
+       h32 *= PRIME32_3;
+       h32 ^= h32 >> 16;
+
+       return h32;
+}
+EXPORT_SYMBOL(xxh32);
+
+static uint64_t xxh64_round(uint64_t acc, const uint64_t input)
+{
+       acc += input * PRIME64_2;
+       acc = xxh_rotl64(acc, 31);
+       acc *= PRIME64_1;
+       return acc;
+}
+
+static uint64_t xxh64_merge_round(uint64_t acc, uint64_t val)
+{
+       val = xxh64_round(0, val);
+       acc ^= val;
+       acc = acc * PRIME64_1 + PRIME64_4;
+       return acc;
+}
+
+uint64_t xxh64(const void *input, const size_t len, const uint64_t seed)
+{
+       const uint8_t *p = (const uint8_t *)input;
+       const uint8_t *const b_end = p + len;
+       uint64_t h64;
+
+       if (len >= 32) {
+               const uint8_t *const limit = b_end - 32;
+               uint64_t v1 = seed + PRIME64_1 + PRIME64_2;
+               uint64_t v2 = seed + PRIME64_2;
+               uint64_t v3 = seed + 0;
+               uint64_t v4 = seed - PRIME64_1;
+
+               do {
+                       v1 = xxh64_round(v1, get_unaligned_le64(p));
+                       p += 8;
+                       v2 = xxh64_round(v2, get_unaligned_le64(p));
+                       p += 8;
+                       v3 = xxh64_round(v3, get_unaligned_le64(p));
+                       p += 8;
+                       v4 = xxh64_round(v4, get_unaligned_le64(p));
+                       p += 8;
+               } while (p <= limit);
+
+               h64 = xxh_rotl64(v1, 1) + xxh_rotl64(v2, 7) +
+                       xxh_rotl64(v3, 12) + xxh_rotl64(v4, 18);
+               h64 = xxh64_merge_round(h64, v1);
+               h64 = xxh64_merge_round(h64, v2);
+               h64 = xxh64_merge_round(h64, v3);
+               h64 = xxh64_merge_round(h64, v4);
+
+       } else {
+               h64  = seed + PRIME64_5;
+       }
+
+       h64 += (uint64_t)len;
+
+       while (p + 8 <= b_end) {
+               const uint64_t k1 = xxh64_round(0, get_unaligned_le64(p));
+
+               h64 ^= k1;
+               h64 = xxh_rotl64(h64, 27) * PRIME64_1 + PRIME64_4;
+               p += 8;
+       }
+
+       if (p + 4 <= b_end) {
+               h64 ^= (uint64_t)(get_unaligned_le32(p)) * PRIME64_1;
+               h64 = xxh_rotl64(h64, 23) * PRIME64_2 + PRIME64_3;
+               p += 4;
+       }
+
+       while (p < b_end) {
+               h64 ^= (*p) * PRIME64_5;
+               h64 = xxh_rotl64(h64, 11) * PRIME64_1;
+               p++;
+       }
+
+       h64 ^= h64 >> 33;
+       h64 *= PRIME64_2;
+       h64 ^= h64 >> 29;
+       h64 *= PRIME64_3;
+       h64 ^= h64 >> 32;
+
+       return h64;
+}
+EXPORT_SYMBOL(xxh64);
+
+/*-**************************************************
+ * Advanced Hash Functions
+ ***************************************************/
+void xxh32_reset(struct xxh32_state *statePtr, const uint32_t seed)
+{
+       /* use a local state for memcpy() to avoid strict-aliasing warnings */
+       struct xxh32_state state;
+
+       memset(&state, 0, sizeof(state));
+       state.v1 = seed + PRIME32_1 + PRIME32_2;
+       state.v2 = seed + PRIME32_2;
+       state.v3 = seed + 0;
+       state.v4 = seed - PRIME32_1;
+       memcpy(statePtr, &state, sizeof(state));
+}
+EXPORT_SYMBOL(xxh32_reset);
+
+void xxh64_reset(struct xxh64_state *statePtr, const uint64_t seed)
+{
+       /* use a local state for memcpy() to avoid strict-aliasing warnings */
+       struct xxh64_state state;
+
+       memset(&state, 0, sizeof(state));
+       state.v1 = seed + PRIME64_1 + PRIME64_2;
+       state.v2 = seed + PRIME64_2;
+       state.v3 = seed + 0;
+       state.v4 = seed - PRIME64_1;
+       memcpy(statePtr, &state, sizeof(state));
+}
+EXPORT_SYMBOL(xxh64_reset);
+
+int xxh32_update(struct xxh32_state *state, const void *input, const size_t len)
+{
+       const uint8_t *p = (const uint8_t *)input;
+       const uint8_t *const b_end = p + len;
+
+       if (input == NULL)
+               return -EINVAL;
+
+       state->total_len_32 += (uint32_t)len;
+       state->large_len |= (len >= 16) | (state->total_len_32 >= 16);
+
+       if (state->memsize + len < 16) { /* fill in tmp buffer */
+               memcpy((uint8_t *)(state->mem32) + state->memsize, input, len);
+               state->memsize += (uint32_t)len;
+               return 0;
+       }
+
+       if (state->memsize) { /* some data left from previous update */
+               const uint32_t *p32 = state->mem32;
+
+               memcpy((uint8_t *)(state->mem32) + state->memsize, input,
+                       16 - state->memsize);
+
+               state->v1 = xxh32_round(state->v1, get_unaligned_le32(p32));
+               p32++;
+               state->v2 = xxh32_round(state->v2, get_unaligned_le32(p32));
+               p32++;
+               state->v3 = xxh32_round(state->v3, get_unaligned_le32(p32));
+               p32++;
+               state->v4 = xxh32_round(state->v4, get_unaligned_le32(p32));
+               p32++;
+
+               p += 16-state->memsize;
+               state->memsize = 0;
+       }
+
+       if (p <= b_end - 16) {
+               const uint8_t *const limit = b_end - 16;
+               uint32_t v1 = state->v1;
+               uint32_t v2 = state->v2;
+               uint32_t v3 = state->v3;
+               uint32_t v4 = state->v4;
+
+               do {
+                       v1 = xxh32_round(v1, get_unaligned_le32(p));
+                       p += 4;
+                       v2 = xxh32_round(v2, get_unaligned_le32(p));
+                       p += 4;
+                       v3 = xxh32_round(v3, get_unaligned_le32(p));
+                       p += 4;
+                       v4 = xxh32_round(v4, get_unaligned_le32(p));
+                       p += 4;
+               } while (p <= limit);
+
+               state->v1 = v1;
+               state->v2 = v2;
+               state->v3 = v3;
+               state->v4 = v4;
+       }
+
+       if (p < b_end) {
+               memcpy(state->mem32, p, (size_t)(b_end-p));
+               state->memsize = (uint32_t)(b_end-p);
+       }
+
+       return 0;
+}
+EXPORT_SYMBOL(xxh32_update);
+
+uint32_t xxh32_digest(const struct xxh32_state *state)
+{
+       const uint8_t *p = (const uint8_t *)state->mem32;
+       const uint8_t *const b_end = (const uint8_t *)(state->mem32) +
+               state->memsize;
+       uint32_t h32;
+
+       if (state->large_len) {
+               h32 = xxh_rotl32(state->v1, 1) + xxh_rotl32(state->v2, 7) +
+                       xxh_rotl32(state->v3, 12) + xxh_rotl32(state->v4, 18);
+       } else {
+               h32 = state->v3 /* == seed */ + PRIME32_5;
+       }
+
+       h32 += state->total_len_32;
+
+       while (p + 4 <= b_end) {
+               h32 += get_unaligned_le32(p) * PRIME32_3;
+               h32 = xxh_rotl32(h32, 17) * PRIME32_4;
+               p += 4;
+       }
+
+       while (p < b_end) {
+               h32 += (*p) * PRIME32_5;
+               h32 = xxh_rotl32(h32, 11) * PRIME32_1;
+               p++;
+       }
+
+       h32 ^= h32 >> 15;
+       h32 *= PRIME32_2;
+       h32 ^= h32 >> 13;
+       h32 *= PRIME32_3;
+       h32 ^= h32 >> 16;
+
+       return h32;
+}
+EXPORT_SYMBOL(xxh32_digest);
+
+int xxh64_update(struct xxh64_state *state, const void *input, const size_t len)
+{
+       const uint8_t *p = (const uint8_t *)input;
+       const uint8_t *const b_end = p + len;
+
+       if (input == NULL)
+               return -EINVAL;
+
+       state->total_len += len;
+
+       if (state->memsize + len < 32) { /* fill in tmp buffer */
+               memcpy(((uint8_t *)state->mem64) + state->memsize, input, len);
+               state->memsize += (uint32_t)len;
+               return 0;
+       }
+
+       if (state->memsize) { /* tmp buffer is full */
+               uint64_t *p64 = state->mem64;
+
+               memcpy(((uint8_t *)p64) + state->memsize, input,
+                       32 - state->memsize);
+
+               state->v1 = xxh64_round(state->v1, get_unaligned_le64(p64));
+               p64++;
+               state->v2 = xxh64_round(state->v2, get_unaligned_le64(p64));
+               p64++;
+               state->v3 = xxh64_round(state->v3, get_unaligned_le64(p64));
+               p64++;
+               state->v4 = xxh64_round(state->v4, get_unaligned_le64(p64));
+
+               p += 32 - state->memsize;
+               state->memsize = 0;
+       }
+
+       if (p + 32 <= b_end) {
+               const uint8_t *const limit = b_end - 32;
+               uint64_t v1 = state->v1;
+               uint64_t v2 = state->v2;
+               uint64_t v3 = state->v3;
+               uint64_t v4 = state->v4;
+
+               do {
+                       v1 = xxh64_round(v1, get_unaligned_le64(p));
+                       p += 8;
+                       v2 = xxh64_round(v2, get_unaligned_le64(p));
+                       p += 8;
+                       v3 = xxh64_round(v3, get_unaligned_le64(p));
+                       p += 8;
+                       v4 = xxh64_round(v4, get_unaligned_le64(p));
+                       p += 8;
+               } while (p <= limit);
+
+               state->v1 = v1;
+               state->v2 = v2;
+               state->v3 = v3;
+               state->v4 = v4;
+       }
+
+       if (p < b_end) {
+               memcpy(state->mem64, p, (size_t)(b_end-p));
+               state->memsize = (uint32_t)(b_end - p);
+       }
+
+       return 0;
+}
+EXPORT_SYMBOL(xxh64_update);
+
+uint64_t xxh64_digest(const struct xxh64_state *state)
+{
+       const uint8_t *p = (const uint8_t *)state->mem64;
+       const uint8_t *const b_end = (const uint8_t *)state->mem64 +
+               state->memsize;
+       uint64_t h64;
+
+       if (state->total_len >= 32) {
+               const uint64_t v1 = state->v1;
+               const uint64_t v2 = state->v2;
+               const uint64_t v3 = state->v3;
+               const uint64_t v4 = state->v4;
+
+               h64 = xxh_rotl64(v1, 1) + xxh_rotl64(v2, 7) +
+                       xxh_rotl64(v3, 12) + xxh_rotl64(v4, 18);
+               h64 = xxh64_merge_round(h64, v1);
+               h64 = xxh64_merge_round(h64, v2);
+               h64 = xxh64_merge_round(h64, v3);
+               h64 = xxh64_merge_round(h64, v4);
+       } else {
+               h64  = state->v3 + PRIME64_5;
+       }
+
+       h64 += (uint64_t)state->total_len;
+
+       while (p + 8 <= b_end) {
+               const uint64_t k1 = xxh64_round(0, get_unaligned_le64(p));
+
+               h64 ^= k1;
+               h64 = xxh_rotl64(h64, 27) * PRIME64_1 + PRIME64_4;
+               p += 8;
+       }
+
+       if (p + 4 <= b_end) {
+               h64 ^= (uint64_t)(get_unaligned_le32(p)) * PRIME64_1;
+               h64 = xxh_rotl64(h64, 23) * PRIME64_2 + PRIME64_3;
+               p += 4;
+       }
+
+       while (p < b_end) {
+               h64 ^= (*p) * PRIME64_5;
+               h64 = xxh_rotl64(h64, 11) * PRIME64_1;
+               p++;
+       }
+
+       h64 ^= h64 >> 33;
+       h64 *= PRIME64_2;
+       h64 ^= h64 >> 29;
+       h64 *= PRIME64_3;
+       h64 ^= h64 >> 32;
+
+       return h64;
+}
+EXPORT_SYMBOL(xxh64_digest);
+
+MODULE_LICENSE("Dual BSD/GPL");
+MODULE_DESCRIPTION("xxHash");