From patchwork Tue Apr 19 21:28:04 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 8bit X-Patchwork-Submitter: Adhemerval Zanella Netto X-Patchwork-Id: 53048 Return-Path: X-Original-To: patchwork@sourceware.org Delivered-To: patchwork@sourceware.org Received: from server2.sourceware.org (localhost [IPv6:::1]) by sourceware.org (Postfix) with ESMTP id 135103857341 for ; Tue, 19 Apr 2022 21:29:29 +0000 (GMT) DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org 135103857341 DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=sourceware.org; s=default; t=1650403769; bh=Ly2hXH5LcjbtKnibIPOc2eFpYhIBanms9Qijc/lU+6U=; h=To:Subject:Date:In-Reply-To:References:List-Id:List-Unsubscribe: List-Archive:List-Post:List-Help:List-Subscribe:From:Reply-To:Cc: From; b=SZZglaAs2wJsGcED52pxwvRpZgTutt1AjDRv9FO77rpugYjaClCWW7Sp4i8tCSTQr JBrwQmUCPBkEyABbGUMmVmKAjuWdRfnr2LIiD45Dbo5BEUUHuJrkXP/7+0E/72k1+f qNwHAyzGwZZg9487VaMpHiA4k3gUBk669XtN7D+4= X-Original-To: libc-alpha@sourceware.org Delivered-To: libc-alpha@sourceware.org Received: from mail-oa1-x33.google.com (mail-oa1-x33.google.com [IPv6:2001:4860:4864:20::33]) by sourceware.org (Postfix) with ESMTPS id 571803858D3C for ; Tue, 19 Apr 2022 21:28:22 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org 571803858D3C Received: by mail-oa1-x33.google.com with SMTP id 586e51a60fabf-e5e8523fcbso7383155fac.10 for ; Tue, 19 Apr 2022 14:28:22 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; h=x-gm-message-state:from:to:cc:subject:date:message-id:in-reply-to :references:mime-version:content-transfer-encoding; bh=Ly2hXH5LcjbtKnibIPOc2eFpYhIBanms9Qijc/lU+6U=; b=MbhcLwoduITbpHHWBnQE6pp8I+5T7cZNVxlhBrXpkJ0dg2dTt6Gy3h6X9NdEEJjygl cWWX1naoTikbFuKQbIZzuIYMOyWCVsMf+nt1l1s9eXBgUaMjlTo6CPXh2FKdnxiIzwmU 8p5hXH4oKLHPc+6B9lU+ccBSCABRDYdBaO+FCj+yshHKUzTmUiPtRr8PWFF3EMGpuV3f K2EKA9/Cne5U7Ghnsw6fPazvn7wX5nHH+B5I5PiHM/cQ7FeUbNxSOdCYp6EbscPdUUWF Ah2kEpi0ppb9aN7tOIsbamni6qYxGRXjllvpstVnYtseHxwR9iJisaDDZ23yoEE32Hfk l/+A== X-Gm-Message-State: AOAM531VtkilAoEFpT3ORFZqvHyYRSVkMdbgZCgbDyWnuHjHRjJj1oIM xYwpFu+9dURxDNGZ2iRntSLMQWqvaIbfyA== X-Google-Smtp-Source: ABdhPJyosbC0ZTdjZmBfNgGH+gjgsnD2JZw3At2N2Ak9/mos1nyBSD8O3qf9hi5Kw/PrUmKsXGwAmg== X-Received: by 2002:a05:6870:2189:b0:e6:bd5:635a with SMTP id l9-20020a056870218900b000e60bd5635amr268256oae.32.1650403700266; Tue, 19 Apr 2022 14:28:20 -0700 (PDT) Received: from birita.. ([2804:431:c7ca:c9d0:98f6:7aed:2f61:2745]) by smtp.gmail.com with ESMTPSA id nf9-20020a056871460900b000e2c44ca8edsm5473321oab.6.2022.04.19.14.28.18 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Tue, 19 Apr 2022 14:28:19 -0700 (PDT) To: libc-alpha@sourceware.org Subject: [PATCH v3 1/9] stdlib: Add arc4random, arc4random_buf, and arc4random_uniform (BZ #4417) Date: Tue, 19 Apr 2022 18:28:04 -0300 Message-Id: <20220419212812.2688764-2-adhemerval.zanella@linaro.org> X-Mailer: git-send-email 2.32.0 In-Reply-To: <20220419212812.2688764-1-adhemerval.zanella@linaro.org> References: <20220419212812.2688764-1-adhemerval.zanella@linaro.org> MIME-Version: 1.0 X-Spam-Status: No, score=-12.4 required=5.0 tests=BAYES_00, DKIM_SIGNED, DKIM_VALID, DKIM_VALID_AU, DKIM_VALID_EF, GIT_PATCH_0, KAM_SHORT, RCVD_IN_DNSWL_NONE, SPF_HELO_NONE, SPF_PASS, TXREP, T_SCC_BODY_TEXT_LINE autolearn=ham autolearn_force=no version=3.4.4 X-Spam-Checker-Version: SpamAssassin 3.4.4 (2020-01-24) on server2.sourceware.org X-BeenThere: libc-alpha@sourceware.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Libc-alpha mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-Patchwork-Original-From: Adhemerval Zanella via Libc-alpha From: Adhemerval Zanella Netto Reply-To: Adhemerval Zanella Cc: Florian Weimer Errors-To: libc-alpha-bounces+patchwork=sourceware.org@sourceware.org Sender: "Libc-alpha" The implementation is based on scalar Chacha20, with global cache and locking. It uses getrandom or /dev/urandom as fallback to get the initial entropy, and reseeds the internal state on every 16MB of consumed buffer. It maintains an internal buffer which consumes at maximum one page on most systems (assuming minimum of 4k pages). The internal buf optimizes the cipher encrypt calls, by amortize arc4random calls (where both function call and locks cost are the dominating factor). The ChaCha20 implementation is based on the RFC8439 [1], with last step that XOR with the input omited. Since the input stream will either zero bytes (initial state) or the PRNG output itself this step does not add any extra entropy. The arc4random_uniform is based on previous work by Florian Weimer. Checked on x86_64-linux-gnu, aarch64-linux, and powerpc64le-linux-gnu. Co-authored-by: Florian Weimer [1] https://datatracker.ietf.org/doc/html/rfc8439 --- NEWS | 4 +- include/stdlib.h | 13 + posix/fork.c | 2 + stdlib/Makefile | 2 + stdlib/Versions | 5 + stdlib/arc4random.c | 245 ++++++++++++++++++ stdlib/arc4random_uniform.c | 152 +++++++++++ stdlib/chacha20.c | 163 ++++++++++++ stdlib/stdlib.h | 14 + sysdeps/generic/not-cancel.h | 2 + sysdeps/mach/hurd/i386/libc.abilist | 3 + sysdeps/mach/hurd/not-cancel.h | 3 + sysdeps/unix/sysv/linux/aarch64/libc.abilist | 3 + sysdeps/unix/sysv/linux/alpha/libc.abilist | 3 + sysdeps/unix/sysv/linux/arc/libc.abilist | 3 + sysdeps/unix/sysv/linux/arm/be/libc.abilist | 3 + sysdeps/unix/sysv/linux/arm/le/libc.abilist | 3 + sysdeps/unix/sysv/linux/csky/libc.abilist | 3 + sysdeps/unix/sysv/linux/hppa/libc.abilist | 3 + sysdeps/unix/sysv/linux/i386/libc.abilist | 3 + sysdeps/unix/sysv/linux/ia64/libc.abilist | 3 + .../sysv/linux/m68k/coldfire/libc.abilist | 3 + .../unix/sysv/linux/m68k/m680x0/libc.abilist | 3 + .../sysv/linux/microblaze/be/libc.abilist | 3 + .../sysv/linux/microblaze/le/libc.abilist | 3 + .../sysv/linux/mips/mips32/fpu/libc.abilist | 3 + .../sysv/linux/mips/mips32/nofpu/libc.abilist | 3 + .../sysv/linux/mips/mips64/n32/libc.abilist | 3 + .../sysv/linux/mips/mips64/n64/libc.abilist | 3 + sysdeps/unix/sysv/linux/nios2/libc.abilist | 3 + sysdeps/unix/sysv/linux/not-cancel.h | 7 + sysdeps/unix/sysv/linux/or1k/libc.abilist | 3 + .../linux/powerpc/powerpc32/fpu/libc.abilist | 3 + .../powerpc/powerpc32/nofpu/libc.abilist | 3 + .../linux/powerpc/powerpc64/be/libc.abilist | 3 + .../linux/powerpc/powerpc64/le/libc.abilist | 3 + .../unix/sysv/linux/riscv/rv32/libc.abilist | 3 + .../unix/sysv/linux/riscv/rv64/libc.abilist | 3 + .../unix/sysv/linux/s390/s390-32/libc.abilist | 3 + .../unix/sysv/linux/s390/s390-64/libc.abilist | 3 + sysdeps/unix/sysv/linux/sh/be/libc.abilist | 3 + sysdeps/unix/sysv/linux/sh/le/libc.abilist | 3 + .../sysv/linux/sparc/sparc32/libc.abilist | 3 + .../sysv/linux/sparc/sparc64/libc.abilist | 3 + .../unix/sysv/linux/x86_64/64/libc.abilist | 3 + .../unix/sysv/linux/x86_64/x32/libc.abilist | 3 + 46 files changed, 713 insertions(+), 1 deletion(-) create mode 100644 stdlib/arc4random.c create mode 100644 stdlib/arc4random_uniform.c create mode 100644 stdlib/chacha20.c diff --git a/NEWS b/NEWS index 4b6d9de2b5..4d9d95b35b 100644 --- a/NEWS +++ b/NEWS @@ -9,7 +9,9 @@ Version 2.36 Major new features: - [Add new features here] +* The functions arc4random, arc4random_buf, arc4random_uniform have been + added. The functions use a cryptographic pseudo-random number generator + based on ChaCha20 initilized with entropy from kernel. Deprecated and removed features, and other changes affecting compatibility: diff --git a/include/stdlib.h b/include/stdlib.h index 1c6f70b082..055f9d2965 100644 --- a/include/stdlib.h +++ b/include/stdlib.h @@ -144,6 +144,19 @@ libc_hidden_proto (__ptsname_r) libc_hidden_proto (grantpt) libc_hidden_proto (unlockpt) +__typeof (arc4random) __arc4random; +libc_hidden_proto (__arc4random); +__typeof (arc4random_buf) __arc4random_buf; +libc_hidden_proto (__arc4random_buf); +__typeof (arc4random_uniform) __arc4random_uniform; +libc_hidden_proto (__arc4random_uniform); +extern void __arc4random_buf_internal (void *buffer, size_t len) + attribute_hidden; +/* Called from the fork function to reinitialize the internal lock in thte + child process. This avoids deadlocks if fork is called in multi-threaded + processes. */ +extern void __arc4random_fork_subprocess (void) attribute_hidden; + extern double __strtod_internal (const char *__restrict __nptr, char **__restrict __endptr, int __group) __THROW __nonnull ((1)) __wur; diff --git a/posix/fork.c b/posix/fork.c index 6b50c091f9..87d8329b46 100644 --- a/posix/fork.c +++ b/posix/fork.c @@ -96,6 +96,8 @@ __libc_fork (void) &nss_database_data); } + call_function_static_weak (__arc4random_fork_subprocess); + /* Reset the lock the dynamic loader uses to protect its data. */ __rtld_lock_initialize (GL(dl_load_lock)); diff --git a/stdlib/Makefile b/stdlib/Makefile index 60fc59c12c..9f9cc1bd7f 100644 --- a/stdlib/Makefile +++ b/stdlib/Makefile @@ -53,6 +53,8 @@ routines := \ a64l \ abort \ abs \ + arc4random \ + arc4random_uniform \ at_quick_exit \ atof \ atoi \ diff --git a/stdlib/Versions b/stdlib/Versions index 5e9099a153..d09a308fb5 100644 --- a/stdlib/Versions +++ b/stdlib/Versions @@ -136,6 +136,11 @@ libc { strtof32; strtof64; strtof32x; strtof32_l; strtof64_l; strtof32x_l; } + GLIBC_2.36 { + arc4random; + arc4random_buf; + arc4random_uniform; + } GLIBC_PRIVATE { # functions which have an additional interface since they are # are cancelable. diff --git a/stdlib/arc4random.c b/stdlib/arc4random.c new file mode 100644 index 0000000000..cddb0e405a --- /dev/null +++ b/stdlib/arc4random.c @@ -0,0 +1,245 @@ +/* Pseudo Random Number Generator based on ChaCha20. + Copyright (C) 2020 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + . */ + +#include +#include +#include +#include +#include +#include +#include +#include + +/* Besides the cipher state 'ctx', it keeps two counters: 'have' is the + current valid bytes not yet consumed in 'buf', while 'count' is the maximum + number of bytes until a reseed. + + Both the initial seed an reseed tries to obtain entropy from the kernel + and abort the process if none could be obtained. + + The state 'buf' improves the usage of the cipher call, allowing to call + optimized implementations (if the archictecture provides it) and optimize + arc4random calls (since only multiple call it will encrypt the next block). + */ + +/* Maximum number bytes until reseed (16 MB). */ +#define CHACHE_RESEED_SIZE (16 * 1024 * 1024) +/* Internal buffer size in bytes (1KB). */ +#define CHACHA20_BUFSIZE (8 * CHACHA20_BLOCK_SIZE) + +#include + +static struct arc4random_state +{ + uint32_t ctx[CHACHA20_STATE_LEN]; + size_t have; + size_t count; + uint8_t buf[CHACHA20_BUFSIZE]; +} *state; + +/* Indicate that MADV_WIPEONFORK is supported by the kernel and thus + it does not require to clear the internal state. */ +static bool __arc4random_wipeonfork = false; + +__libc_lock_define_initialized (, __arc4random_lock); + +/* Called from the fork function to reset the state if MADV_WIPEONFORK is + not supported and to reinit the internal lock. */ +void +__arc4random_fork_subprocess (void) +{ + if (__arc4random_wipeonfork && state != NULL) + memset (state, 0, sizeof (struct arc4random_state)); + + __libc_lock_init (__arc4random_lock); +} + +static void +arc4random_allocate_failure (void) +{ + __libc_fatal ("Fatal glibc error: Cannot allocate memory for arc4random\n"); +} + +static void +arc4random_getrandom_failure (void) +{ + __libc_fatal ("Fatal glibc error: Cannot get entropy for arc4random\n"); +} + +/* Fork detection is done by checking if MADV_WIPEONFORK supported. If not + the fork callback will reset the state on the fork call. It does not + handle direct clone calls, nor vfork or _Fork (arc4random is not + async-signal-safe due the internal lock usage). */ +static void +arc4random_init (uint8_t *buf, size_t len) +{ + state = __mmap (NULL, sizeof (struct arc4random_state), + PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_PRIVATE, -1, 0); + if (state == MAP_FAILED) + arc4random_allocate_failure (); + +#ifdef MADV_WIPEONFORK + int r = __madvise (state, sizeof (struct arc4random_state), MADV_WIPEONFORK); + if (r == 0) + __arc4random_wipeonfork = true; + else if (errno != EINVAL) + arc4random_allocate_failure (); +#endif + + chacha20_init (state->ctx, buf, buf + CHACHA20_KEY_SIZE); +} + +#define min(x,y) (((x) > (y)) ? (y) : (x)) + +static void +arc4random_rekey (uint8_t *rnd, size_t rndlen) +{ + memset (state->buf, 0, sizeof state->buf); + chacha20_crypt (state->ctx, state->buf, state->buf, sizeof state->buf); + + /* Mix some extra entropy if provided. */ + if (rnd != NULL) + { + size_t m = min (rndlen, CHACHA20_KEY_SIZE + CHACHA20_IV_SIZE); + for (size_t i = 0; i < m; i++) + state->buf[i] ^= rnd[i]; + } + + /* Immediately reinit for backtracking resistance. */ + chacha20_init (state->ctx, state->buf, state->buf + CHACHA20_KEY_SIZE); + memset (state->buf, 0, CHACHA20_KEY_SIZE + CHACHA20_IV_SIZE); + state->have = sizeof (state->buf) - (CHACHA20_KEY_SIZE + CHACHA20_IV_SIZE); +} + +static void +arc4random_getentropy (uint8_t *rnd, size_t len) +{ + if (__getrandomn_nocancel (rnd, len, GRND_NONBLOCK) == len) + return; + + int fd = __open64_nocancel ("/dev/urandom", O_RDONLY); + if (fd != -1) + { + unsigned char *p = rnd; + unsigned char *end = p + len; + do + { + ssize_t ret = TEMP_FAILURE_RETRY (__read_nocancel (fd, p, end - p)); + if (ret <= 0) + arc4random_getrandom_failure (); + p += ret; + } + while (p < end); + + if (__close_nocancel (fd) != 0) + return; + } + arc4random_getrandom_failure (); +} + +/* Either allocates the state buffer or reinit it by reseeding the cipher + state with kernel entropy. */ +static void +arc4random_stir (void) +{ + uint8_t rnd[CHACHA20_KEY_SIZE + CHACHA20_IV_SIZE]; + arc4random_getentropy (rnd, sizeof rnd); + + if (state == NULL) + arc4random_init (rnd, sizeof rnd); + else + arc4random_rekey (rnd, sizeof rnd); + + explicit_bzero (rnd, sizeof rnd); + + state->have = 0; + memset (state->buf, 0, sizeof state->buf); + state->count = CHACHE_RESEED_SIZE; +} + +static void +arc4random_check_stir (size_t len) +{ + if (state == NULL || state->count < len) + arc4random_stir (); + if (state->count <= len) + state->count = 0; + else + state->count -= len; +} + +void +__arc4random_buf_internal (void *buffer, size_t len) +{ + arc4random_check_stir (len); + + while (len > 0) + { + if (state->have > 0) + { + size_t m = min (len, state->have); + uint8_t *ks = state->buf + sizeof (state->buf) - state->have; + memcpy (buffer, ks, m); + memset (ks, 0, m); + buffer += m; + len -= m; + state->have -= m; + } + if (state->have == 0) + arc4random_rekey (NULL, 0); + } +} + +void +__arc4random_buf (void *buffer, size_t len) +{ + __libc_lock_lock (__arc4random_lock); + __arc4random_buf_internal (buffer, len); + __libc_lock_unlock (__arc4random_lock); +} +libc_hidden_def (__arc4random_buf) +weak_alias (__arc4random_buf, arc4random_buf) + + +static uint32_t +__arc4random_internal (void) +{ + uint32_t r; + + arc4random_check_stir (sizeof (uint32_t)); + if (state->have < sizeof (uint32_t)) + arc4random_rekey (NULL, 0); + uint8_t *ks = state->buf + sizeof (state->buf) - state->have; + memcpy (&r, ks, sizeof (uint32_t)); + memset (ks, 0, sizeof (uint32_t)); + state->have -= sizeof (uint32_t); + + return r; +} + +uint32_t +__arc4random (void) +{ + uint32_t r; + __libc_lock_lock (__arc4random_lock); + r = __arc4random_internal (); + __libc_lock_unlock (__arc4random_lock); + return r; +} +libc_hidden_def (__arc4random) +weak_alias (__arc4random, arc4random) diff --git a/stdlib/arc4random_uniform.c b/stdlib/arc4random_uniform.c new file mode 100644 index 0000000000..96ffe62df1 --- /dev/null +++ b/stdlib/arc4random_uniform.c @@ -0,0 +1,152 @@ +/* Random pseudo generator numbers between 0 and 2**-31 (inclusive) + uniformly distributed but with an upper_bound. + Copyright (C) 2022 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + . */ + +#include +#include +#include +#include + +/* Return the number of bytes which cover values up to the limit. */ +__attribute__ ((const)) +static uint32_t +byte_count (uint32_t n) +{ + if (n <= (1U << 8)) + return 1; + else if (n <= (1U << 16)) + return 2; + else if (n <= (1U << 24)) + return 3; + else + return 4; +} + +/* Fill the lower bits of the result with randomness, according to the + number of bytes requested. */ +static void +random_bytes (uint32_t *result, uint32_t byte_count) +{ + *result = 0; + unsigned char *ptr = (unsigned char *) result; + if (__BYTE_ORDER == __BIG_ENDIAN) + ptr += 4 - byte_count; + __arc4random_buf_internal (ptr, byte_count); +} + +static uint32_t +compute_uniform (uint32_t n) +{ + if (n <= 1) + /* There is no valid return value for a zero limit, and 0 is the + only possible result for limit 1. */ + return 0; + + /* The bits variable serves as a source for bits. Prefetch the + minimum number of bytes needed. */ + unsigned count = byte_count (n); + uint32_t bits_length = count * CHAR_BIT; + uint32_t bits; + random_bytes (&bits, count); + + /* Powers of two are easy. */ + if (powerof2 (n)) + return bits & (n - 1); + + /* The general case. This algorithm follows Jérémie Lumbroso, + Optimal Discrete Uniform Generation from Coin Flips, and + Applications (2013), who credits Donald E. Knuth and Andrew + C. Yao, The complexity of nonuniform random number generation + (1976), for solving the general case. + + The implementation below unrolls the initialization stage of the + loop, where v is less than n. */ + + /* Use 64-bit variables even though the intermediate results are + never larger that 33 bits. This ensures the code easier to + compile on 64-bit architectures. */ + uint64_t v; + uint64_t c; + + /* Initialize v and c. v is the smallest power of 2 which is larger + than n.*/ + { + uint32_t log2p1 = 32 - __builtin_clz (n); + v = 1ULL << log2p1; + c = bits & (v - 1); + bits >>= log2p1; + bits_length -= log2p1; + } + + /* At the start of the loop, c is uniformly distributed within the + half-open interval [0, v), and v < 2n < 2**33. */ + while (true) + { + if (v >= n) + { + /* If the candidate is less than n, accept it. */ + if (c < n) + /* c is uniformly distributed on [0, n). */ + return c; + else + { + /* c is uniformly distributed on [n, v). */ + v -= n; + c -= n; + /* The distribution was shifted, so c is uniformly + distributed on [0, v) again. */ + } + } + /* v < n here. */ + + /* Replenish the bit source if necessary. */ + if (bits_length == 0) + { + /* Overwrite the least significant byte. */ + random_bytes (&bits, 1); + bits_length = CHAR_BIT; + } + + /* Double the range. No overflow because v < n < 2**32. */ + v *= 2; + /* v < 2n here. */ + + /* Extract a bit and append it to c. c remains less than v and + thus 2**33. */ + c = (c << 1) | (bits & 1); + bits >>= 1; + --bits_length; + + /* At this point, c is uniformly distributed on [0, v) again, + and v < 2n < 2**33. */ + } +} + +__libc_lock_define (extern , __arc4random_lock attribute_hidden) + +uint32_t +__arc4random_uniform (uint32_t upper_bound) +{ + uint32_t r; + __libc_lock_lock (__arc4random_lock); + r = compute_uniform (upper_bound); + __libc_lock_unlock (__arc4random_lock); + return r; +} +libc_hidden_def (__arc4random_uniform) +weak_alias (__arc4random_uniform, arc4random_uniform) diff --git a/stdlib/chacha20.c b/stdlib/chacha20.c new file mode 100644 index 0000000000..af4ffa9860 --- /dev/null +++ b/stdlib/chacha20.c @@ -0,0 +1,163 @@ +/* Generic ChaCha20 implementation (used on arc4random). + Copyright (C) 2022 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + . */ + +#include +#include +#include +#include +#include + +/* 32-bit stream position, then 96-bit nonce. */ +#define CHACHA20_IV_SIZE 16 +#define CHACHA20_KEY_SIZE 32 + +#define CHACHA20_BLOCK_SIZE 64 +#define CHACHA20_BLOCK_WORDS (CHACHA20_BLOCK_SIZE / sizeof (uint32_t)) + +#define CHACHA20_STATE_LEN 16 + +/* Defining CHACHA20_XOR_FINAL issues the final XOR using the input as defined + Sby RFC8439. Since the input stream will either zero bytes (initial state) + or the PRNG output itself this step does not add any extra entropy. */ + +enum chacha20_constants +{ + CHACHA20_CONSTANT_EXPA = 0x61707865U, + CHACHA20_CONSTANT_ND_3 = 0x3320646eU, + CHACHA20_CONSTANT_2_BY = 0x79622d32U, + CHACHA20_CONSTANT_TE_K = 0x6b206574U +}; + +static inline uint32_t +read_unaligned_32 (const uint8_t *p) +{ + uint32_t r; + memcpy (&r, p, sizeof (r)); + return r; +} + +static inline void +write_unaligned_32 (uint8_t *p, uint32_t v) +{ + memcpy (p, &v, sizeof (v)); +} + +#if __BYTE_ORDER == __BIG_ENDIAN +# define read_unaligned_le32(p) __builtin_bswap32 (read_unaligned_32 (p)) +# define set_state(v) __builtin_bswap32 ((v)) +#else +# define read_unaligned_le32(p) read_unaligned_32 ((p)) +# define set_state(v) (v) +#endif + +static inline void +chacha20_init (uint32_t *state, const uint8_t *key, const uint8_t *iv) +{ + state[0] = CHACHA20_CONSTANT_EXPA; + state[1] = CHACHA20_CONSTANT_ND_3; + state[2] = CHACHA20_CONSTANT_2_BY; + state[3] = CHACHA20_CONSTANT_TE_K; + + state[4] = read_unaligned_le32 (key + 0 * sizeof (uint32_t)); + state[5] = read_unaligned_le32 (key + 1 * sizeof (uint32_t)); + state[6] = read_unaligned_le32 (key + 2 * sizeof (uint32_t)); + state[7] = read_unaligned_le32 (key + 3 * sizeof (uint32_t)); + state[8] = read_unaligned_le32 (key + 4 * sizeof (uint32_t)); + state[9] = read_unaligned_le32 (key + 5 * sizeof (uint32_t)); + state[10] = read_unaligned_le32 (key + 6 * sizeof (uint32_t)); + state[11] = read_unaligned_le32 (key + 7 * sizeof (uint32_t)); + + state[12] = read_unaligned_le32 (iv + 0 * sizeof (uint32_t)); + state[13] = read_unaligned_le32 (iv + 1 * sizeof (uint32_t)); + state[14] = read_unaligned_le32 (iv + 2 * sizeof (uint32_t)); + state[15] = read_unaligned_le32 (iv + 3 * sizeof (uint32_t)); +} + +static inline uint32_t +rotl32 (unsigned int shift, uint32_t word) +{ + return (word << (shift & 31)) | (word >> ((-shift) & 31)); +} + +#define QROUND(x0, x1, x2, x3) \ + do { \ + x0 = x0 + x1; x3 = rotl32 (16, (x0 ^ x3)); \ + x2 = x2 + x3; x1 = rotl32 (12, (x1 ^ x2)); \ + x0 = x0 + x1; x3 = rotl32 (8, (x0 ^ x3)); \ + x2 = x2 + x3; x1 = rotl32 (7, (x1 ^ x2)); \ + } while(0) + +static inline void +chacha20_block (uint32_t *state, uint32_t *stream) +{ + uint32_t x[CHACHA20_STATE_LEN]; + memcpy (x, state, sizeof x); + + for (int i = 0; i < 20; i += 2) + { + QROUND (x[0], x[4], x[8], x[12]); + QROUND (x[1], x[5], x[9], x[13]); + QROUND (x[2], x[6], x[10], x[14]); + QROUND (x[3], x[7], x[11], x[15]); + + QROUND (x[0], x[5], x[10], x[15]); + QROUND (x[1], x[6], x[11], x[12]); + QROUND (x[2], x[7], x[8], x[13]); + QROUND (x[3], x[4], x[9], x[14]); + } + + /* Unroll the loop a bit. */ + for (int i = 0; i < CHACHA20_BLOCK_WORDS / 4; i++) + { + stream[i*4+0] = set_state (x[i*4+0] + state[i*4+0]); + stream[i*4+1] = set_state (x[i*4+1] + state[i*4+1]); + stream[i*4+2] = set_state (x[i*4+2] + state[i*4+2]); + stream[i*4+3] = set_state (x[i*4+3] + state[i*4+3]); + } + + state[12]++; +} + +static void +chacha20_crypt (uint32_t *state, uint8_t *dst, const uint8_t *src, + size_t bytes) +{ + uint32_t stream[CHACHA20_BLOCK_WORDS]; + + while (bytes >= CHACHA20_BLOCK_SIZE) + { + chacha20_block (state, stream); +#ifdef CHACHA20_XOR_FINAL + for (int i = 0; i < CHACHA20_BLOCK_WORDS; i++) + stream[i] ^= read_unaligned_32 (&src[i * sizeof (uint32_t)]); +#endif + memcpy (dst, stream, CHACHA20_BLOCK_SIZE); + bytes -= CHACHA20_BLOCK_SIZE; + dst += CHACHA20_BLOCK_SIZE; + src += CHACHA20_BLOCK_SIZE; + } + if (bytes != 0) + { + chacha20_block (state, stream); +#ifdef CHACHA20_XOR_FINAL + for (int i = 0; i < CHACHA20_BLOCK_WORDS; i++) + stream[i] ^= read_unaligned_32 (&src[i * sizeof (uint32_t)]); +#endif + memcpy (dst, stream, bytes); + } +} diff --git a/stdlib/stdlib.h b/stdlib/stdlib.h index bf7cd438e1..f2b0c83c12 100644 --- a/stdlib/stdlib.h +++ b/stdlib/stdlib.h @@ -485,6 +485,7 @@ extern unsigned short int *seed48 (unsigned short int __seed16v[3]) extern void lcong48 (unsigned short int __param[7]) __THROW __nonnull ((1)); # ifdef __USE_MISC +# include /* Data structure for communication with thread safe versions. This type is to be regarded as opaque. It's only exported because users have to allocate objects of this type. */ @@ -533,6 +534,19 @@ extern int seed48_r (unsigned short int __seed16v[3], extern int lcong48_r (unsigned short int __param[7], struct drand48_data *__buffer) __THROW __nonnull ((1, 2)); + +/* Return a random integer between zero and 2**31-1 (inclusive). */ +extern uint32_t arc4random (void) + __THROW __wur; + +/* Fill the buffer with random data. */ +extern void arc4random_buf (void *__buf, size_t __size) + __THROW __nonnull ((1)); + +/* Return a random number between zero (inclusive) and the specified + limit (exclusive). */ +extern uint32_t arc4random_uniform (uint32_t __upper_bound) + __THROW __wur; # endif /* Use misc. */ #endif /* Use misc or X/Open. */ diff --git a/sysdeps/generic/not-cancel.h b/sysdeps/generic/not-cancel.h index 2104efeb54..f4882a9ffd 100644 --- a/sysdeps/generic/not-cancel.h +++ b/sysdeps/generic/not-cancel.h @@ -48,5 +48,7 @@ (void) __writev (fd, iov, n) #define __fcntl64_nocancel(fd, cmd, ...) \ __fcntl64 (fd, cmd, __VA_ARGS__) +#define __getrandomn_nocancel(buf, size, flags) \ + __getrandom (buf, size, flags) #endif /* NOT_CANCEL_H */ diff --git a/sysdeps/mach/hurd/i386/libc.abilist b/sysdeps/mach/hurd/i386/libc.abilist index 4dc87e9061..7bd565103b 100644 --- a/sysdeps/mach/hurd/i386/libc.abilist +++ b/sysdeps/mach/hurd/i386/libc.abilist @@ -2289,6 +2289,9 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 close_range F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F GLIBC_2.4 __fgets_unlocked_chk F diff --git a/sysdeps/mach/hurd/not-cancel.h b/sysdeps/mach/hurd/not-cancel.h index 6ec92ced84..39edfe76b6 100644 --- a/sysdeps/mach/hurd/not-cancel.h +++ b/sysdeps/mach/hurd/not-cancel.h @@ -74,6 +74,9 @@ __typeof (__fcntl) __fcntl_nocancel; #define __fcntl64_nocancel(...) \ __fcntl_nocancel (__VA_ARGS__) +#define __getrandomn_nocancel(buf, size, flags) \ + __getrandom (buf, size, flags) + #if IS_IN (libc) hidden_proto (__close_nocancel) hidden_proto (__close_nocancel_nostatus) diff --git a/sysdeps/unix/sysv/linux/aarch64/libc.abilist b/sysdeps/unix/sysv/linux/aarch64/libc.abilist index 1b63d9e447..f8f38bb205 100644 --- a/sysdeps/unix/sysv/linux/aarch64/libc.abilist +++ b/sysdeps/unix/sysv/linux/aarch64/libc.abilist @@ -2616,3 +2616,6 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 epoll_pwait2 F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F diff --git a/sysdeps/unix/sysv/linux/alpha/libc.abilist b/sysdeps/unix/sysv/linux/alpha/libc.abilist index e7e4cf7d2a..9de1726de0 100644 --- a/sysdeps/unix/sysv/linux/alpha/libc.abilist +++ b/sysdeps/unix/sysv/linux/alpha/libc.abilist @@ -2713,6 +2713,9 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 epoll_pwait2 F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F GLIBC_2.4 _IO_fprintf F GLIBC_2.4 _IO_printf F GLIBC_2.4 _IO_sprintf F diff --git a/sysdeps/unix/sysv/linux/arc/libc.abilist b/sysdeps/unix/sysv/linux/arc/libc.abilist index bc3d228e31..16e2532838 100644 --- a/sysdeps/unix/sysv/linux/arc/libc.abilist +++ b/sysdeps/unix/sysv/linux/arc/libc.abilist @@ -2377,3 +2377,6 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 epoll_pwait2 F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F diff --git a/sysdeps/unix/sysv/linux/arm/be/libc.abilist b/sysdeps/unix/sysv/linux/arm/be/libc.abilist index db7039c4ab..ae9e465088 100644 --- a/sysdeps/unix/sysv/linux/arm/be/libc.abilist +++ b/sysdeps/unix/sysv/linux/arm/be/libc.abilist @@ -496,6 +496,9 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 epoll_pwait2 F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F GLIBC_2.4 _Exit F GLIBC_2.4 _IO_2_1_stderr_ D 0xa0 GLIBC_2.4 _IO_2_1_stdin_ D 0xa0 diff --git a/sysdeps/unix/sysv/linux/arm/le/libc.abilist b/sysdeps/unix/sysv/linux/arm/le/libc.abilist index d2add4fb49..b669f43194 100644 --- a/sysdeps/unix/sysv/linux/arm/le/libc.abilist +++ b/sysdeps/unix/sysv/linux/arm/le/libc.abilist @@ -493,6 +493,9 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 epoll_pwait2 F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F GLIBC_2.4 _Exit F GLIBC_2.4 _IO_2_1_stderr_ D 0xa0 GLIBC_2.4 _IO_2_1_stdin_ D 0xa0 diff --git a/sysdeps/unix/sysv/linux/csky/libc.abilist b/sysdeps/unix/sysv/linux/csky/libc.abilist index 355d72a30c..42daa90248 100644 --- a/sysdeps/unix/sysv/linux/csky/libc.abilist +++ b/sysdeps/unix/sysv/linux/csky/libc.abilist @@ -2652,3 +2652,6 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 epoll_pwait2 F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F diff --git a/sysdeps/unix/sysv/linux/hppa/libc.abilist b/sysdeps/unix/sysv/linux/hppa/libc.abilist index 3df39bb28c..090be20f53 100644 --- a/sysdeps/unix/sysv/linux/hppa/libc.abilist +++ b/sysdeps/unix/sysv/linux/hppa/libc.abilist @@ -2601,6 +2601,9 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 epoll_pwait2 F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F GLIBC_2.4 __fgets_unlocked_chk F diff --git a/sysdeps/unix/sysv/linux/i386/libc.abilist b/sysdeps/unix/sysv/linux/i386/libc.abilist index c4da358f80..6b7cf064bb 100644 --- a/sysdeps/unix/sysv/linux/i386/libc.abilist +++ b/sysdeps/unix/sysv/linux/i386/libc.abilist @@ -2785,6 +2785,9 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 epoll_pwait2 F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F GLIBC_2.4 __fgets_unlocked_chk F diff --git a/sysdeps/unix/sysv/linux/ia64/libc.abilist b/sysdeps/unix/sysv/linux/ia64/libc.abilist index 241bac70ea..3e766f64dd 100644 --- a/sysdeps/unix/sysv/linux/ia64/libc.abilist +++ b/sysdeps/unix/sysv/linux/ia64/libc.abilist @@ -2551,6 +2551,9 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 epoll_pwait2 F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F GLIBC_2.4 __fgets_unlocked_chk F diff --git a/sysdeps/unix/sysv/linux/m68k/coldfire/libc.abilist b/sysdeps/unix/sysv/linux/m68k/coldfire/libc.abilist index 78bf372b72..c0b99199a8 100644 --- a/sysdeps/unix/sysv/linux/m68k/coldfire/libc.abilist +++ b/sysdeps/unix/sysv/linux/m68k/coldfire/libc.abilist @@ -497,6 +497,9 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 epoll_pwait2 F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F GLIBC_2.4 _Exit F GLIBC_2.4 _IO_2_1_stderr_ D 0x98 GLIBC_2.4 _IO_2_1_stdin_ D 0x98 diff --git a/sysdeps/unix/sysv/linux/m68k/m680x0/libc.abilist b/sysdeps/unix/sysv/linux/m68k/m680x0/libc.abilist index 00df5c901f..4d0be7c86d 100644 --- a/sysdeps/unix/sysv/linux/m68k/m680x0/libc.abilist +++ b/sysdeps/unix/sysv/linux/m68k/m680x0/libc.abilist @@ -2728,6 +2728,9 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 epoll_pwait2 F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F GLIBC_2.4 __fgets_unlocked_chk F diff --git a/sysdeps/unix/sysv/linux/microblaze/be/libc.abilist b/sysdeps/unix/sysv/linux/microblaze/be/libc.abilist index e8118569c3..b944680ede 100644 --- a/sysdeps/unix/sysv/linux/microblaze/be/libc.abilist +++ b/sysdeps/unix/sysv/linux/microblaze/be/libc.abilist @@ -2701,3 +2701,6 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 epoll_pwait2 F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F diff --git a/sysdeps/unix/sysv/linux/microblaze/le/libc.abilist b/sysdeps/unix/sysv/linux/microblaze/le/libc.abilist index c0d2373e64..28f7d19983 100644 --- a/sysdeps/unix/sysv/linux/microblaze/le/libc.abilist +++ b/sysdeps/unix/sysv/linux/microblaze/le/libc.abilist @@ -2698,3 +2698,6 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 epoll_pwait2 F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F diff --git a/sysdeps/unix/sysv/linux/mips/mips32/fpu/libc.abilist b/sysdeps/unix/sysv/linux/mips/mips32/fpu/libc.abilist index 2d0fd04f54..3da7cdaca5 100644 --- a/sysdeps/unix/sysv/linux/mips/mips32/fpu/libc.abilist +++ b/sysdeps/unix/sysv/linux/mips/mips32/fpu/libc.abilist @@ -2693,6 +2693,9 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 epoll_pwait2 F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F GLIBC_2.4 __fgets_unlocked_chk F diff --git a/sysdeps/unix/sysv/linux/mips/mips32/nofpu/libc.abilist b/sysdeps/unix/sysv/linux/mips/mips32/nofpu/libc.abilist index e39ccfb312..9fe87f15be 100644 --- a/sysdeps/unix/sysv/linux/mips/mips32/nofpu/libc.abilist +++ b/sysdeps/unix/sysv/linux/mips/mips32/nofpu/libc.abilist @@ -2691,6 +2691,9 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 epoll_pwait2 F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F GLIBC_2.4 __fgets_unlocked_chk F diff --git a/sysdeps/unix/sysv/linux/mips/mips64/n32/libc.abilist b/sysdeps/unix/sysv/linux/mips/mips64/n32/libc.abilist index 1e900f86e4..c14fca2111 100644 --- a/sysdeps/unix/sysv/linux/mips/mips64/n32/libc.abilist +++ b/sysdeps/unix/sysv/linux/mips/mips64/n32/libc.abilist @@ -2699,6 +2699,9 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 epoll_pwait2 F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F GLIBC_2.4 __fgets_unlocked_chk F diff --git a/sysdeps/unix/sysv/linux/mips/mips64/n64/libc.abilist b/sysdeps/unix/sysv/linux/mips/mips64/n64/libc.abilist index 9145ba7931..a363830226 100644 --- a/sysdeps/unix/sysv/linux/mips/mips64/n64/libc.abilist +++ b/sysdeps/unix/sysv/linux/mips/mips64/n64/libc.abilist @@ -2602,6 +2602,9 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 epoll_pwait2 F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F GLIBC_2.4 __fgets_unlocked_chk F diff --git a/sysdeps/unix/sysv/linux/nios2/libc.abilist b/sysdeps/unix/sysv/linux/nios2/libc.abilist index e95d60d926..89b6f98667 100644 --- a/sysdeps/unix/sysv/linux/nios2/libc.abilist +++ b/sysdeps/unix/sysv/linux/nios2/libc.abilist @@ -2740,3 +2740,6 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 epoll_pwait2 F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F diff --git a/sysdeps/unix/sysv/linux/not-cancel.h b/sysdeps/unix/sysv/linux/not-cancel.h index 75b9e0ee1e..be5df35927 100644 --- a/sysdeps/unix/sysv/linux/not-cancel.h +++ b/sysdeps/unix/sysv/linux/not-cancel.h @@ -67,6 +67,13 @@ __writev_nocancel_nostatus (int fd, const struct iovec *iov, int iovcnt) INTERNAL_SYSCALL_CALL (writev, fd, iov, iovcnt); } +static inline int +__getrandomn_nocancel (void *buf, size_t buflen, unsigned int flags) +{ + return INTERNAL_SYSCALL_CALL (getrandom, buf, buflen, flags); +} + + /* Uncancelable fcntl. */ __typeof (__fcntl) __fcntl64_nocancel; diff --git a/sysdeps/unix/sysv/linux/or1k/libc.abilist b/sysdeps/unix/sysv/linux/or1k/libc.abilist index ca934e374b..94c0ff9526 100644 --- a/sysdeps/unix/sysv/linux/or1k/libc.abilist +++ b/sysdeps/unix/sysv/linux/or1k/libc.abilist @@ -2123,3 +2123,6 @@ GLIBC_2.35 wprintf F GLIBC_2.35 write F GLIBC_2.35 writev F GLIBC_2.35 wscanf F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F diff --git a/sysdeps/unix/sysv/linux/powerpc/powerpc32/fpu/libc.abilist b/sysdeps/unix/sysv/linux/powerpc/powerpc32/fpu/libc.abilist index 3820b9f235..d6188de00b 100644 --- a/sysdeps/unix/sysv/linux/powerpc/powerpc32/fpu/libc.abilist +++ b/sysdeps/unix/sysv/linux/powerpc/powerpc32/fpu/libc.abilist @@ -2755,6 +2755,9 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 epoll_pwait2 F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F GLIBC_2.4 _IO_fprintf F GLIBC_2.4 _IO_printf F GLIBC_2.4 _IO_sprintf F diff --git a/sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/libc.abilist b/sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/libc.abilist index 464dc27fcd..8201230059 100644 --- a/sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/libc.abilist +++ b/sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/libc.abilist @@ -2788,6 +2788,9 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 epoll_pwait2 F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F GLIBC_2.4 _IO_fprintf F GLIBC_2.4 _IO_printf F GLIBC_2.4 _IO_sprintf F diff --git a/sysdeps/unix/sysv/linux/powerpc/powerpc64/be/libc.abilist b/sysdeps/unix/sysv/linux/powerpc/powerpc64/be/libc.abilist index 2f7e58747f..623505d783 100644 --- a/sysdeps/unix/sysv/linux/powerpc/powerpc64/be/libc.abilist +++ b/sysdeps/unix/sysv/linux/powerpc/powerpc64/be/libc.abilist @@ -2510,6 +2510,9 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 epoll_pwait2 F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F GLIBC_2.4 _IO_fprintf F GLIBC_2.4 _IO_printf F GLIBC_2.4 _IO_sprintf F diff --git a/sysdeps/unix/sysv/linux/powerpc/powerpc64/le/libc.abilist b/sysdeps/unix/sysv/linux/powerpc/powerpc64/le/libc.abilist index 4f3043d913..23b0d83408 100644 --- a/sysdeps/unix/sysv/linux/powerpc/powerpc64/le/libc.abilist +++ b/sysdeps/unix/sysv/linux/powerpc/powerpc64/le/libc.abilist @@ -2812,3 +2812,6 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 epoll_pwait2 F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F diff --git a/sysdeps/unix/sysv/linux/riscv/rv32/libc.abilist b/sysdeps/unix/sysv/linux/riscv/rv32/libc.abilist index 84b6ac815a..a72e8ed9cc 100644 --- a/sysdeps/unix/sysv/linux/riscv/rv32/libc.abilist +++ b/sysdeps/unix/sysv/linux/riscv/rv32/libc.abilist @@ -2379,3 +2379,6 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 epoll_pwait2 F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F diff --git a/sysdeps/unix/sysv/linux/riscv/rv64/libc.abilist b/sysdeps/unix/sysv/linux/riscv/rv64/libc.abilist index 4d5c19c56a..f3faecc2ae 100644 --- a/sysdeps/unix/sysv/linux/riscv/rv64/libc.abilist +++ b/sysdeps/unix/sysv/linux/riscv/rv64/libc.abilist @@ -2579,3 +2579,6 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 epoll_pwait2 F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F diff --git a/sysdeps/unix/sysv/linux/s390/s390-32/libc.abilist b/sysdeps/unix/sysv/linux/s390/s390-32/libc.abilist index 7c5ee8d569..105e5a9231 100644 --- a/sysdeps/unix/sysv/linux/s390/s390-32/libc.abilist +++ b/sysdeps/unix/sysv/linux/s390/s390-32/libc.abilist @@ -2753,6 +2753,9 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 epoll_pwait2 F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F GLIBC_2.4 _IO_fprintf F GLIBC_2.4 _IO_printf F GLIBC_2.4 _IO_sprintf F diff --git a/sysdeps/unix/sysv/linux/s390/s390-64/libc.abilist b/sysdeps/unix/sysv/linux/s390/s390-64/libc.abilist index 50de0b46cf..c08c6c8301 100644 --- a/sysdeps/unix/sysv/linux/s390/s390-64/libc.abilist +++ b/sysdeps/unix/sysv/linux/s390/s390-64/libc.abilist @@ -2547,6 +2547,9 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 epoll_pwait2 F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F GLIBC_2.4 _IO_fprintf F GLIBC_2.4 _IO_printf F GLIBC_2.4 _IO_sprintf F diff --git a/sysdeps/unix/sysv/linux/sh/be/libc.abilist b/sysdeps/unix/sysv/linux/sh/be/libc.abilist index 66fba013ca..8ec1005644 100644 --- a/sysdeps/unix/sysv/linux/sh/be/libc.abilist +++ b/sysdeps/unix/sysv/linux/sh/be/libc.abilist @@ -2608,6 +2608,9 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 epoll_pwait2 F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F GLIBC_2.4 __fgets_unlocked_chk F diff --git a/sysdeps/unix/sysv/linux/sh/le/libc.abilist b/sysdeps/unix/sysv/linux/sh/le/libc.abilist index 38703f8aa0..5d776576f9 100644 --- a/sysdeps/unix/sysv/linux/sh/le/libc.abilist +++ b/sysdeps/unix/sysv/linux/sh/le/libc.abilist @@ -2605,6 +2605,9 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 epoll_pwait2 F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F GLIBC_2.4 __fgets_unlocked_chk F diff --git a/sysdeps/unix/sysv/linux/sparc/sparc32/libc.abilist b/sysdeps/unix/sysv/linux/sparc/sparc32/libc.abilist index 6df55eb765..f5f07f612e 100644 --- a/sysdeps/unix/sysv/linux/sparc/sparc32/libc.abilist +++ b/sysdeps/unix/sysv/linux/sparc/sparc32/libc.abilist @@ -2748,6 +2748,9 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 epoll_pwait2 F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F GLIBC_2.4 _IO_fprintf F GLIBC_2.4 _IO_printf F GLIBC_2.4 _IO_sprintf F diff --git a/sysdeps/unix/sysv/linux/sparc/sparc64/libc.abilist b/sysdeps/unix/sysv/linux/sparc/sparc64/libc.abilist index b90569d881..be687ebe02 100644 --- a/sysdeps/unix/sysv/linux/sparc/sparc64/libc.abilist +++ b/sysdeps/unix/sysv/linux/sparc/sparc64/libc.abilist @@ -2574,6 +2574,9 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 epoll_pwait2 F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F GLIBC_2.4 __fgets_unlocked_chk F diff --git a/sysdeps/unix/sysv/linux/x86_64/64/libc.abilist b/sysdeps/unix/sysv/linux/x86_64/64/libc.abilist index e88b0f101f..7f456fbb55 100644 --- a/sysdeps/unix/sysv/linux/x86_64/64/libc.abilist +++ b/sysdeps/unix/sysv/linux/x86_64/64/libc.abilist @@ -2525,6 +2525,9 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 epoll_pwait2 F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F GLIBC_2.4 __confstr_chk F GLIBC_2.4 __fgets_chk F GLIBC_2.4 __fgets_unlocked_chk F diff --git a/sysdeps/unix/sysv/linux/x86_64/x32/libc.abilist b/sysdeps/unix/sysv/linux/x86_64/x32/libc.abilist index e0755272eb..c737201248 100644 --- a/sysdeps/unix/sysv/linux/x86_64/x32/libc.abilist +++ b/sysdeps/unix/sysv/linux/x86_64/x32/libc.abilist @@ -2631,3 +2631,6 @@ GLIBC_2.35 __memcmpeq F GLIBC_2.35 _dl_find_object F GLIBC_2.35 epoll_pwait2 F GLIBC_2.35 posix_spawn_file_actions_addtcsetpgrp_np F +GLIBC_2.36 arc4random F +GLIBC_2.36 arc4random_buf F +GLIBC_2.36 arc4random_uniform F From patchwork Tue Apr 19 21:28:05 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Adhemerval Zanella Netto X-Patchwork-Id: 53049 Return-Path: X-Original-To: patchwork@sourceware.org Delivered-To: patchwork@sourceware.org Received: from server2.sourceware.org (localhost [IPv6:::1]) by sourceware.org (Postfix) with ESMTP id 0E42E3857342 for ; Tue, 19 Apr 2022 21:30:17 +0000 (GMT) DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org 0E42E3857342 DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=sourceware.org; s=default; t=1650403817; bh=EHGgke4tI+PI9Bdk8aQ/mZjSA7ID0wh/nPrEmdQS3SU=; h=To:Subject:Date:In-Reply-To:References:List-Id:List-Unsubscribe: List-Archive:List-Post:List-Help:List-Subscribe:From:Reply-To:Cc: From; b=wGDXM9zdwjbYJvWOGhmfk/51CCHtP0KI7krltVlGPurZkuiLM3jRObm8ne17TKFQ5 iMUiNCMUanXxn/iW/LZZBFs+oi5Su2CNv+Nm5aZpffvgiFHtbqPYM168ueh2GcuZOL vhbJK8nBP5JKsfkCpT7HawQNGWUjEXnfcnNDe63Y= X-Original-To: libc-alpha@sourceware.org Delivered-To: libc-alpha@sourceware.org Received: from mail-ot1-x335.google.com (mail-ot1-x335.google.com [IPv6:2607:f8b0:4864:20::335]) by sourceware.org (Postfix) with ESMTPS id 05D1F385741D for ; Tue, 19 Apr 2022 21:28:24 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org 05D1F385741D Received: by mail-ot1-x335.google.com with SMTP id r14-20020a9d750e000000b00605446d683eso6613099otk.10 for ; Tue, 19 Apr 2022 14:28:23 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; h=x-gm-message-state:from:to:cc:subject:date:message-id:in-reply-to :references:mime-version:content-transfer-encoding; bh=EHGgke4tI+PI9Bdk8aQ/mZjSA7ID0wh/nPrEmdQS3SU=; b=jgwZLw1AWD3BnMSTXiF7nOZNAz/dN/HI73y2n5ddpADyKV2ByGNufJzXYI1n2gwp/t BGPthpLGNyDU9byNDOHdldLjluU5PQJxyEgpayO/LABA1BoiyYGOBfZuRkpVj1C/QwM9 vZtMqJUp8ZVYXHW/eBFwsbKzxYYp7IHvg3iKyLm6nxdPQWmlZmf5znpQv3v4bTQZx0yl UKuE42cmtcgaWVI2zvDaVtvqKodvDCnDf4XB4Gskb6Zp7zRobyonxq+lMjfeq8yX85NO VWunRReJ/+fh8qMlUjCR3whH8jBdxa/IidVlvF62oKq2cjC7N1y8pkI/aqJYa1JyfgNC Hxeg== X-Gm-Message-State: AOAM531lNdzC+NGM5kwVLj6XnD5JpcBxnLKZr8CMOWaxNNPyHM6QslTE IOrbIU4X+mXEY7itekfRKhGdeFtgoPy1WA== X-Google-Smtp-Source: ABdhPJy1i8SUzztDnBX+jnyY+1f8wjzjGXgZTtypxGzD60cr1//SNUu3eU7axlS6lE+9HYhosPfTDg== X-Received: by 2002:a9d:17cb:0:b0:604:350:7fd3 with SMTP id j69-20020a9d17cb000000b0060403507fd3mr6720708otj.291.1650403702584; Tue, 19 Apr 2022 14:28:22 -0700 (PDT) Received: from birita.. ([2804:431:c7ca:c9d0:98f6:7aed:2f61:2745]) by smtp.gmail.com with ESMTPSA id nf9-20020a056871460900b000e2c44ca8edsm5473321oab.6.2022.04.19.14.28.21 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Tue, 19 Apr 2022 14:28:22 -0700 (PDT) To: libc-alpha@sourceware.org Subject: [PATCH v3 2/9] stdlib: Add arc4random tests Date: Tue, 19 Apr 2022 18:28:05 -0300 Message-Id: <20220419212812.2688764-3-adhemerval.zanella@linaro.org> X-Mailer: git-send-email 2.32.0 In-Reply-To: <20220419212812.2688764-1-adhemerval.zanella@linaro.org> References: <20220419212812.2688764-1-adhemerval.zanella@linaro.org> MIME-Version: 1.0 X-Spam-Status: No, score=-12.5 required=5.0 tests=BAYES_00, DKIM_SIGNED, DKIM_VALID, DKIM_VALID_AU, DKIM_VALID_EF, GIT_PATCH_0, KAM_SHORT, RCVD_IN_DNSWL_NONE, SPF_HELO_NONE, SPF_PASS, TXREP, T_SCC_BODY_TEXT_LINE autolearn=ham autolearn_force=no version=3.4.4 X-Spam-Checker-Version: SpamAssassin 3.4.4 (2020-01-24) on server2.sourceware.org X-BeenThere: libc-alpha@sourceware.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Libc-alpha mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-Patchwork-Original-From: Adhemerval Zanella via Libc-alpha From: Adhemerval Zanella Netto Reply-To: Adhemerval Zanella Cc: Florian Weimer Errors-To: libc-alpha-bounces+patchwork=sourceware.org@sourceware.org Sender: "Libc-alpha" The basic tst-arc4random-chacha20.c checks if the output of ChaCha20 implementation matches the reference test vectors from RFC8439. The tst-arc4random-fork.c check if subprocesses generate distinct streams of randomness (if fork handling is done correctly). The tst-arc4random-stats.c is a statistical test to the randomness of arc4random, arc4random_buf, and arc4random_uniform. The tst-arc4random-thread.c check if threads generate distinct streams of randomness (if function are thread-safe). Checked on x86_64-linux-gnu, aarch64-linux, and powerpc64le-linux-gnu. Co-authored-by: Florian Weimer --- stdlib/Makefile | 7 + stdlib/tst-arc4random-chacha20.c | 166 ++++++++++++++++++ stdlib/tst-arc4random-fork.c | 174 +++++++++++++++++++ stdlib/tst-arc4random-stats.c | 146 ++++++++++++++++ stdlib/tst-arc4random-thread.c | 278 +++++++++++++++++++++++++++++++ 5 files changed, 771 insertions(+) create mode 100644 stdlib/tst-arc4random-chacha20.c create mode 100644 stdlib/tst-arc4random-fork.c create mode 100644 stdlib/tst-arc4random-stats.c create mode 100644 stdlib/tst-arc4random-thread.c diff --git a/stdlib/Makefile b/stdlib/Makefile index 9f9cc1bd7f..c29faf2e43 100644 --- a/stdlib/Makefile +++ b/stdlib/Makefile @@ -183,6 +183,9 @@ tests := \ testmb2 \ testrand \ testsort \ + tst-arc4random-fork \ + tst-arc4random-stats \ + tst-arc4random-thread \ tst-at_quick_exit \ tst-atexit \ tst-atof1 \ @@ -243,6 +246,7 @@ tests := \ # tests tests-internal := \ + tst-arc4random-chacha20 \ tst-strtod1i \ tst-strtod3 \ tst-strtod4 \ @@ -252,6 +256,7 @@ tests-internal := \ # tests-internal tests-static := \ + tst-arc4random-chacha20 \ tst-secure-getenv \ # tests-static @@ -271,6 +276,8 @@ LDLIBS-test-cxa_atexit-race = $(shared-thread-library) LDLIBS-test-cxa_atexit-race2 = $(shared-thread-library) LDLIBS-test-on_exit-race = $(shared-thread-library) LDLIBS-tst-canon-bz26341 = $(shared-thread-library) +LDLIBS-tst-arc4random-fork = $(shared-thread-library) +LDLIBS-tst-arc4random-thread = $(shared-thread-library) LDLIBS-test-dlclose-exit-race = $(shared-thread-library) LDFLAGS-test-dlclose-exit-race = $(LDFLAGS-rdynamic) diff --git a/stdlib/tst-arc4random-chacha20.c b/stdlib/tst-arc4random-chacha20.c new file mode 100644 index 0000000000..dd0ef6d8ba --- /dev/null +++ b/stdlib/tst-arc4random-chacha20.c @@ -0,0 +1,166 @@ +/* Basic tests for chacha20 cypher used in arc4random. + Copyright (C) 2022 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + . */ + +#include +#include + +/* It does not define CHACHA20_XOR_FINAL to check what glibc actual uses. */ +#define CHACHA20_BUFSIZE (8 * CHACHA20_BLOCK_SIZE) +#include + +static int +do_test (void) +{ + const uint8_t key[CHACHA20_KEY_SIZE] = + { + 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, + 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, + 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, + 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, + }; + const uint8_t iv[CHACHA20_IV_SIZE] = + { + 0x0, 0x0, 0x0, 0x0, + 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, + }; + const uint8_t expected1[CHACHA20_BUFSIZE] = + { + 0x76, 0xb8, 0xe0, 0xad, 0xa0, 0xf1, 0x3d, 0x90, 0x40, 0x5d, 0x6a, + 0xe5, 0x53, 0x86, 0xbd, 0x28, 0xbd, 0xd2, 0x19, 0xb8, 0xa0, 0x8d, + 0xed, 0x1a, 0xa8, 0x36, 0xef, 0xcc, 0x8b, 0x77, 0x0d, 0xc7, 0xda, + 0x41, 0x59, 0x7c, 0x51, 0x57, 0x48, 0x8d, 0x77, 0x24, 0xe0, 0x3f, + 0xb8, 0xd8, 0x4a, 0x37, 0x6a, 0x43, 0xb8, 0xf4, 0x15, 0x18, 0xa1, + 0x1c, 0xc3, 0x87, 0xb6, 0x69, 0xb2, 0xee, 0x65, 0x86, 0x9f, 0x07, + 0xe7, 0xbe, 0x55, 0x51, 0x38, 0x7a, 0x98, 0xba, 0x97, 0x7c, 0x73, + 0x2d, 0x08, 0x0d, 0xcb, 0x0f, 0x29, 0xa0, 0x48, 0xe3, 0x65, 0x69, + 0x12, 0xc6, 0x53, 0x3e, 0x32, 0xee, 0x7a, 0xed, 0x29, 0xb7, 0x21, + 0x76, 0x9c, 0xe6, 0x4e, 0x43, 0xd5, 0x71, 0x33, 0xb0, 0x74, 0xd8, + 0x39, 0xd5, 0x31, 0xed, 0x1f, 0x28, 0x51, 0x0a, 0xfb, 0x45, 0xac, + 0xe1, 0x0a, 0x1f, 0x4b, 0x79, 0x4d, 0x6f, 0x2d, 0x09, 0xa0, 0xe6, + 0x63, 0x26, 0x6c, 0xe1, 0xae, 0x7e, 0xd1, 0x08, 0x19, 0x68, 0xa0, + 0x75, 0x8e, 0x71, 0x8e, 0x99, 0x7b, 0xd3, 0x62, 0xc6, 0xb0, 0xc3, + 0x46, 0x34, 0xa9, 0xa0, 0xb3, 0x5d, 0x01, 0x27, 0x37, 0x68, 0x1f, + 0x7b, 0x5d, 0x0f, 0x28, 0x1e, 0x3a, 0xfd, 0xe4, 0x58, 0xbc, 0x1e, + 0x73, 0xd2, 0xd3, 0x13, 0xc9, 0xcf, 0x94, 0xc0, 0x5f, 0xf3, 0x71, + 0x62, 0x40, 0xa2, 0x48, 0xf2, 0x13, 0x20, 0xa0, 0x58, 0xd7, 0xb3, + 0x56, 0x6b, 0xd5, 0x20, 0xda, 0xaa, 0x3e, 0xd2, 0xbf, 0x0a, 0xc5, + 0xb8, 0xb1, 0x20, 0xfb, 0x85, 0x27, 0x73, 0xc3, 0x63, 0x97, 0x34, + 0xb4, 0x5c, 0x91, 0xa4, 0x2d, 0xd4, 0xcb, 0x83, 0xf8, 0x84, 0x0d, + 0x2e, 0xed, 0xb1, 0x58, 0x13, 0x10, 0x62, 0xac, 0x3f, 0x1f, 0x2c, + 0xf8, 0xff, 0x6d, 0xcd, 0x18, 0x56, 0xe8, 0x6a, 0x1e, 0x6c, 0x31, + 0x67, 0x16, 0x7e, 0xe5, 0xa6, 0x88, 0x74, 0x2b, 0x47, 0xc5, 0xad, + 0xfb, 0x59, 0xd4, 0xdf, 0x76, 0xfd, 0x1d, 0xb1, 0xe5, 0x1e, 0xe0, + 0x3b, 0x1c, 0xa9, 0xf8, 0x2a, 0xca, 0x17, 0x3e, 0xdb, 0x8b, 0x72, + 0x93, 0x47, 0x4e, 0xbe, 0x98, 0x0f, 0x90, 0x4d, 0x10, 0xc9, 0x16, + 0x44, 0x2b, 0x47, 0x83, 0xa0, 0xe9, 0x84, 0x86, 0x0c, 0xb6, 0xc9, + 0x57, 0xb3, 0x9c, 0x38, 0xed, 0x8f, 0x51, 0xcf, 0xfa, 0xa6, 0x8a, + 0x4d, 0xe0, 0x10, 0x25, 0xa3, 0x9c, 0x50, 0x45, 0x46, 0xb9, 0xdc, + 0x14, 0x06, 0xa7, 0xeb, 0x28, 0x15, 0x1e, 0x51, 0x50, 0xd7, 0xb2, + 0x04, 0xba, 0xa7, 0x19, 0xd4, 0xf0, 0x91, 0x02, 0x12, 0x17, 0xdb, + 0x5c, 0xf1, 0xb5, 0xc8, 0x4c, 0x4f, 0xa7, 0x1a, 0x87, 0x96, 0x10, + 0xa1, 0xa6, 0x95, 0xac, 0x52, 0x7c, 0x5b, 0x56, 0x77, 0x4a, 0x6b, + 0x8a, 0x21, 0xaa, 0xe8, 0x86, 0x85, 0x86, 0x8e, 0x09, 0x4c, 0xf2, + 0x9e, 0xf4, 0x09, 0x0a, 0xf7, 0xa9, 0x0c, 0xc0, 0x7e, 0x88, 0x17, + 0xaa, 0x52, 0x87, 0x63, 0x79, 0x7d, 0x3c, 0x33, 0x2b, 0x67, 0xca, + 0x4b, 0xc1, 0x10, 0x64, 0x2c, 0x21, 0x51, 0xec, 0x47, 0xee, 0x84, + 0xcb, 0x8c, 0x42, 0xd8, 0x5f, 0x10, 0xe2, 0xa8, 0xcb, 0x18, 0xc3, + 0xb7, 0x33, 0x5f, 0x26, 0xe8, 0xc3, 0x9a, 0x12, 0xb1, 0xbc, 0xc1, + 0x70, 0x71, 0x77, 0xb7, 0x61, 0x38, 0x73, 0x2e, 0xed, 0xaa, 0xb7, + 0x4d, 0xa1, 0x41, 0x0f, 0xc0, 0x55, 0xea, 0x06, 0x8c, 0x99, 0xe9, + 0x26, 0x0a, 0xcb, 0xe3, 0x37, 0xcf, 0x5d, 0x3e, 0x00, 0xe5, 0xb3, + 0x23, 0x0f, 0xfe, 0xdb, 0x0b, 0x99, 0x07, 0x87, 0xd0, 0xc7, 0x0e, + 0x0b, 0xfe, 0x41, 0x98, 0xea, 0x67, 0x58, 0xdd, 0x5a, 0x61, 0xfb, + 0x5f, 0xec, 0x2d, 0xf9, 0x81, 0xf3, 0x1b, 0xef, 0xe1, 0x53, 0xf8, + 0x1d, 0x17, 0x16, 0x17, 0x84, 0xdb + }; + + const uint8_t expected2[CHACHA20_BUFSIZE] = + { + 0x1c, 0x88, 0x22, 0xd5, 0x3c, 0xd1, 0xee, 0x7d, 0xb5, 0x32, 0x36, + 0x48, 0x28, 0xbd, 0xf4, 0x04, 0xb0, 0x40, 0xa8, 0xdc, 0xc5, 0x22, + 0xf3, 0xd3, 0xd9, 0x9a, 0xec, 0x4b, 0x80, 0x57, 0xed, 0xb8, 0x50, + 0x09, 0x31, 0xa2, 0xc4, 0x2d, 0x2f, 0x0c, 0x57, 0x08, 0x47, 0x10, + 0x0b, 0x57, 0x54, 0xda, 0xfc, 0x5f, 0xbd, 0xb8, 0x94, 0xbb, 0xef, + 0x1a, 0x2d, 0xe1, 0xa0, 0x7f, 0x8b, 0xa0, 0xc4, 0xb9, 0x19, 0x30, + 0x10, 0x66, 0xed, 0xbc, 0x05, 0x6b, 0x7b, 0x48, 0x1e, 0x7a, 0x0c, + 0x46, 0x29, 0x7b, 0xbb, 0x58, 0x9d, 0x9d, 0xa5, 0xb6, 0x75, 0xa6, + 0x72, 0x3e, 0x15, 0x2e, 0x5e, 0x63, 0xa4, 0xce, 0x03, 0x4e, 0x9e, + 0x83, 0xe5, 0x8a, 0x01, 0x3a, 0xf0, 0xe7, 0x35, 0x2f, 0xb7, 0x90, + 0x85, 0x14, 0xe3, 0xb3, 0xd1, 0x04, 0x0d, 0x0b, 0xb9, 0x63, 0xb3, + 0x95, 0x4b, 0x63, 0x6b, 0x5f, 0xd4, 0xbf, 0x6d, 0x0a, 0xad, 0xba, + 0xf8, 0x15, 0x7d, 0x06, 0x2a, 0xcb, 0x24, 0x18, 0xc1, 0x76, 0xa4, + 0x75, 0x51, 0x1b, 0x35, 0xc3, 0xf6, 0x21, 0x8a, 0x56, 0x68, 0xea, + 0x5b, 0xc6, 0xf5, 0x4b, 0x87, 0x82, 0xf8, 0xb3, 0x40, 0xf0, 0x0a, + 0xc1, 0xbe, 0xba, 0x5e, 0x62, 0xcd, 0x63, 0x2a, 0x7c, 0xe7, 0x80, + 0x9c, 0x72, 0x56, 0x08, 0xac, 0xa5, 0xef, 0xbf, 0x7c, 0x41, 0xf2, + 0x37, 0x64, 0x3f, 0x06, 0xc0, 0x99, 0x72, 0x07, 0x17, 0x1d, 0xe8, + 0x67, 0xf9, 0xd6, 0x97, 0xbf, 0x5e, 0xa6, 0x01, 0x1a, 0xbc, 0xce, + 0x6c, 0x8c, 0xdb, 0x21, 0x13, 0x94, 0xd2, 0xc0, 0x2d, 0xd0, 0xfb, + 0x60, 0xdb, 0x5a, 0x2c, 0x17, 0xac, 0x3d, 0xc8, 0x58, 0x78, 0xa9, + 0x0b, 0xed, 0x38, 0x09, 0xdb, 0xb9, 0x6e, 0xaa, 0x54, 0x26, 0xfc, + 0x8e, 0xae, 0x0d, 0x2d, 0x65, 0xc4, 0x2a, 0x47, 0x9f, 0x08, 0x86, + 0x48, 0xbe, 0x2d, 0xc8, 0x01, 0xd8, 0x2a, 0x36, 0x6f, 0xdd, 0xc0, + 0xef, 0x23, 0x42, 0x63, 0xc0, 0xb6, 0x41, 0x7d, 0x5f, 0x9d, 0xa4, + 0x18, 0x17, 0xb8, 0x8d, 0x68, 0xe5, 0xe6, 0x71, 0x95, 0xc5, 0xc1, + 0xee, 0x30, 0x95, 0xe8, 0x21, 0xf2, 0x25, 0x24, 0xb2, 0x0b, 0xe4, + 0x1c, 0xeb, 0x59, 0x04, 0x12, 0xe4, 0x1d, 0xc6, 0x48, 0x84, 0x3f, + 0xa9, 0xbf, 0xec, 0x7a, 0x3d, 0xcf, 0x61, 0xab, 0x05, 0x41, 0x57, + 0x33, 0x16, 0xd3, 0xfa, 0x81, 0x51, 0x62, 0x93, 0x03, 0xfe, 0x97, + 0x41, 0x56, 0x2e, 0xd0, 0x65, 0xdb, 0x4e, 0xbc, 0x00, 0x50, 0xef, + 0x55, 0x83, 0x64, 0xae, 0x81, 0x12, 0x4a, 0x28, 0xf5, 0xc0, 0x13, + 0x13, 0x23, 0x2f, 0xbc, 0x49, 0x6d, 0xfd, 0x8a, 0x25, 0x68, 0x65, + 0x7b, 0x68, 0x6d, 0x72, 0x14, 0x38, 0x2a, 0x1a, 0x00, 0x90, 0x30, + 0x17, 0xdd, 0xa9, 0x69, 0x87, 0x84, 0x42, 0xba, 0x5a, 0xff, 0xf6, + 0x61, 0x3f, 0x55, 0x3c, 0xbb, 0x23, 0x3c, 0xe4, 0x6d, 0x9a, 0xee, + 0x93, 0xa7, 0x87, 0x6c, 0xf5, 0xe9, 0xe8, 0x29, 0x12, 0xb1, 0x8c, + 0xad, 0xf0, 0xb3, 0x43, 0x27, 0xb2, 0xe0, 0x42, 0x7e, 0xcf, 0x66, + 0xb7, 0xce, 0xb7, 0xc0, 0x91, 0x8d, 0xc4, 0x7b, 0xdf, 0xf1, 0x2a, + 0x06, 0x2a, 0xdf, 0x07, 0x13, 0x30, 0x09, 0xce, 0x7a, 0x5e, 0x5c, + 0x91, 0x7e, 0x01, 0x68, 0x30, 0x61, 0x09, 0xb7, 0xcb, 0x49, 0x65, + 0x3a, 0x6d, 0x2c, 0xae, 0xf0, 0x05, 0xde, 0x78, 0x3a, 0x9a, 0x9b, + 0xfe, 0x05, 0x38, 0x1e, 0xd1, 0x34, 0x8d, 0x94, 0xec, 0x65, 0x88, + 0x6f, 0x9c, 0x0b, 0x61, 0x9c, 0x52, 0xc5, 0x53, 0x38, 0x00, 0xb1, + 0x6c, 0x83, 0x61, 0x72, 0xb9, 0x51, 0x82, 0xdb, 0xc5, 0xee, 0xc0, + 0x42, 0xb8, 0x9e, 0x22, 0xf1, 0x1a, 0x08, 0x5b, 0x73, 0x9a, 0x36, + 0x11, 0xcd, 0x8d, 0x83, 0x60, 0x18 + }; + + /* Check with the expected internal arc4random keystream buffer. Some + architecture optimization expectes a buffer with a minimum size of + multiple of ChaCha20 blocksize, so they might not be prepared to + handle smaller buffers. */ + + uint8_t output[CHACHA20_BUFSIZE]; + + uint32_t state[CHACHA20_STATE_LEN]; + chacha20_init (state, key, iv); + + /* Check with the initial state. */ + uint8_t input[CHACHA20_BUFSIZE] = { 0 }; + + chacha20_crypt (state, output, input, CHACHA20_BUFSIZE); + TEST_COMPARE_BLOB (output, sizeof output, expected1, CHACHA20_BUFSIZE); + + /* And on the next round. */ + chacha20_crypt (state, output, input, CHACHA20_BUFSIZE); + TEST_COMPARE_BLOB (output, sizeof output, expected2, CHACHA20_BUFSIZE); + + return 0; +} + +#include diff --git a/stdlib/tst-arc4random-fork.c b/stdlib/tst-arc4random-fork.c new file mode 100644 index 0000000000..cd8852c8d3 --- /dev/null +++ b/stdlib/tst-arc4random-fork.c @@ -0,0 +1,174 @@ +/* Test that subprocesses generate distinct streams of randomness. + Copyright (C) 2022 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + . */ + +/* Collect random data from subprocesses and check that all the + results are unique. */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include + +/* Perform multiple runs. The subsequent runs start with an + already-initialized random number generator. (The number 1500 was + seen to reproduce failures reliable in case of a race condition in + the fork detection code.) */ +enum { runs = 1500 }; + +/* One hundred processes in total. This should be high enough to + expose any issues, but low enough not to tax the overall system too + much. */ +enum { subprocesses = 49 }; + +/* The total number of processes. */ +enum { processes = subprocesses + 1 }; + +/* Number of bytes of randomness to generate per process. Large + enough to make false positive duplicates extremely unlikely. */ +enum { random_size = 16 }; + +/* Generated bytes of randomness. */ +struct result +{ + unsigned char bytes[random_size]; +}; + +/* Shared across all processes. */ +static struct shared_data +{ + pthread_barrier_t barrier; + struct result results[runs][processes]; +} *shared_data; + +/* Invoked to collect data from a subprocess. */ +static void +subprocess (int run, int process_index) +{ + xpthread_barrier_wait (&shared_data->barrier); + arc4random_buf (shared_data->results[run][process_index].bytes, random_size); +} + +/* Used to sort the results. */ +struct index +{ + int run; + int process_index; +}; + +/* Used to sort an array of struct index values. */ +static int +index_compare (const void *left1, const void *right1) +{ + const struct index *left = left1; + const struct index *right = right1; + + return memcmp (shared_data->results[left->run][left->process_index].bytes, + shared_data->results[right->run][right->process_index].bytes, + random_size); +} + +static int +do_test (void) +{ + shared_data = support_shared_allocate (sizeof (*shared_data)); + { + pthread_barrierattr_t attr; + xpthread_barrierattr_init (&attr); + xpthread_barrierattr_setpshared (&attr, PTHREAD_PROCESS_SHARED); + xpthread_barrier_init (&shared_data->barrier, &attr, processes); + xpthread_barrierattr_destroy (&attr); + } + + /* Collect random data. */ + for (int run = 0; run < runs; ++run) + { +#if 0 + if (run == runs / 2) + { + /* In the middle, desynchronize the block cache by consuming + an odd number of bytes. */ + char buf; + arc4random_buf (&buf, 1); + } +#endif + + pid_t pids[subprocesses]; + for (int process_index = 0; process_index < subprocesses; + ++process_index) + { + pids[process_index] = xfork (); + if (pids[process_index] == 0) + { + subprocess (run, process_index); + _exit (0); + } + } + + /* Trigger all subprocesses. Also add data from the parent + process. */ + subprocess (run, subprocesses); + + for (int process_index = 0; process_index < subprocesses; + ++process_index) + { + int status; + xwaitpid (pids[process_index], &status, 0); + if (status != 0) + FAIL_EXIT1 ("subprocess index %d (PID %d) exit status %d\n", + process_index, (int) pids[process_index], status); + } + } + + /* Check for duplicates. */ + struct index indexes[runs * processes]; + for (int run = 0; run < runs; ++run) + for (int process_index = 0; process_index < processes; ++process_index) + indexes[run * processes + process_index] + = (struct index) { .run = run, .process_index = process_index }; + qsort (indexes, array_length (indexes), sizeof (indexes[0]), index_compare); + for (size_t i = 1; i < array_length (indexes); ++i) + { + if (index_compare (indexes + i - 1, indexes + i) == 0) + { + support_record_failure (); + unsigned char *bytes + = shared_data->results[indexes[i].run] + [indexes[i].process_index].bytes; + char *quoted = support_quote_blob (bytes, random_size); + printf ("error: duplicate randomness data: \"%s\"\n" + " run %d, subprocess %d\n" + " run %d, subprocess %d\n", + quoted, indexes[i - 1].run, indexes[i - 1].process_index, + indexes[i].run, indexes[i].process_index); + free (quoted); + } + } + + xpthread_barrier_destroy (&shared_data->barrier); + support_shared_free (shared_data); + shared_data = NULL; + + return 0; +} + +#include diff --git a/stdlib/tst-arc4random-stats.c b/stdlib/tst-arc4random-stats.c new file mode 100644 index 0000000000..9747180c99 --- /dev/null +++ b/stdlib/tst-arc4random-stats.c @@ -0,0 +1,146 @@ +/* Statistical tests for arc4random-related functions. + Copyright (C) 2022 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + . */ + +#include +#include +#include +#include +#include +#include + +enum +{ + arc4random_key_size = 32 +}; + +struct key +{ + unsigned char data[arc4random_key_size]; +}; + +/* With 12,000 keys, the probability that a byte in a predetermined + position does not have a predetermined value in all generated keys + is about 4e-21. The probability that this happens with any of the + 16 * 256 possible byte position/values is 1.6e-17. This results in + an acceptably low false-positive rate. */ +enum { key_count = 12000 }; + +static struct key keys[key_count]; + +/* Used to perform the distribution check. */ +static int byte_counts[arc4random_key_size][256]; + +/* Bail out after this many failures. */ +enum { failure_limit = 100 }; + +static void +find_stuck_bytes (bool (*func) (unsigned char *key)) +{ + memset (&keys, 0xcc, sizeof (keys)); + + int failures = 0; + for (int key = 0; key < key_count; ++key) + { + while (true) + { + if (func (keys[key].data)) + break; + ++failures; + if (failures >= failure_limit) + { + printf ("warning: bailing out after %d failures\n", failures); + return; + } + } + } + printf ("info: key generation finished with %d failures\n", failures); + + memset (&byte_counts, 0, sizeof (byte_counts)); + for (int key = 0; key < key_count; ++key) + for (int pos = 0; pos < arc4random_key_size; ++pos) + ++byte_counts[pos][keys[key].data[pos]]; + + for (int pos = 0; pos < arc4random_key_size; ++pos) + for (int byte = 0; byte < 256; ++byte) + if (byte_counts[pos][byte] == 0) + { + support_record_failure (); + printf ("error: byte %d never appeared at position %d\n", byte, pos); + } +} + +/* Test adapter for arc4random. */ +static bool +generate_arc4random (unsigned char *key) +{ + uint32_t words[arc4random_key_size / 4]; + _Static_assert (sizeof (words) == arc4random_key_size, "sizeof (words)"); + + for (int i = 0; i < array_length (words); ++i) + words[i] = arc4random (); + memcpy (key, &words, arc4random_key_size); + return true; +} + +/* Test adapter for arc4random_buf. */ +static bool +generate_arc4random_buf (unsigned char *key) +{ + arc4random_buf (key, arc4random_key_size); + return true; +} + +/* Test adapter for arc4random_uniform. */ +static bool +generate_arc4random_uniform (unsigned char *key) +{ + for (int i = 0; i < arc4random_key_size; ++i) + key[i] = arc4random_uniform (256); + return true; +} + +/* Test adapter for arc4random_uniform with argument 257. This means + that byte 0 happens more often, but we do not perform such a + statistcal check, so the test will still pass */ +static bool +generate_arc4random_uniform_257 (unsigned char *key) +{ + for (int i = 0; i < arc4random_key_size; ++i) + key[i] = arc4random_uniform (257); + return true; +} + +static int +do_test (void) +{ + puts ("info: arc4random implementation test"); + find_stuck_bytes (generate_arc4random); + + puts ("info: arc4random_buf implementation test"); + find_stuck_bytes (generate_arc4random_buf); + + puts ("info: arc4random_uniform implementation test"); + find_stuck_bytes (generate_arc4random_uniform); + + puts ("info: arc4random_uniform implementation test (257 variant)"); + find_stuck_bytes (generate_arc4random_uniform_257); + + return 0; +} + +#include diff --git a/stdlib/tst-arc4random-thread.c b/stdlib/tst-arc4random-thread.c new file mode 100644 index 0000000000..b122eaa826 --- /dev/null +++ b/stdlib/tst-arc4random-thread.c @@ -0,0 +1,278 @@ +/* Test that threads generate distinct streams of randomness. + Copyright (C) 2018 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + . */ + +#include +#include +#include +#include +#include +#include +#include + +/* Number of arc4random_buf calls per thread. */ +enum { count_per_thread = 5000 }; + +/* Number of threads computing randomness. */ +enum { inner_threads = 5 }; + +/* Number of threads launching other threads. Chosen as to not to + overload the system. */ +enum { outer_threads = 7 }; + +/* Number of launching rounds performed by the outer threads. */ +enum { outer_rounds = 10 }; + +/* Maximum number of bytes generated in an arc4random call. */ +enum { max_size = 32 }; + +/* Sizes generated by threads. Must be long enough to be unique with + high probability. */ +static const int sizes[] = { 12, 15, 16, 17, 24, 31, max_size }; + +/* Data structure to capture randomness results. */ +struct blob +{ + unsigned int size; + int thread_id; + unsigned int index; + unsigned char bytes[max_size]; +}; + +#define DYNARRAY_STRUCT dynarray_blob +#define DYNARRAY_ELEMENT struct blob +#define DYNARRAY_PREFIX dynarray_blob_ +#include + +/* Sort blob elements by length first, then by comparing the data + member. */ +static int +compare_blob (const void *left1, const void *right1) +{ + const struct blob *left = left1; + const struct blob *right = right1; + + if (left->size != right->size) + /* No overflow due to limited range. */ + return left->size - right->size; + return memcmp (left->bytes, right->bytes, left->size); +} + +/* Used to store the global result. */ +static pthread_mutex_t global_result_lock = PTHREAD_MUTEX_INITIALIZER; +static struct dynarray_blob global_result; + +/* Copy data to the global result, with locking. */ +static void +copy_result_to_global (struct dynarray_blob *result) +{ + xpthread_mutex_lock (&global_result_lock); + size_t old_size = dynarray_blob_size (&global_result); + TEST_VERIFY_EXIT + (dynarray_blob_resize (&global_result, + old_size + dynarray_blob_size (result))); + memcpy (dynarray_blob_begin (&global_result) + old_size, + dynarray_blob_begin (result), + dynarray_blob_size (result) * sizeof (struct blob)); + xpthread_mutex_unlock (&global_result_lock); +} + +/* Used to assign unique thread IDs. Accessed atomically. */ +static int next_thread_id; + +static void * +inner_thread (void *unused) +{ + /* Use local result to avoid global lock contention while generating + randomness. */ + struct dynarray_blob result; + dynarray_blob_init (&result); + + int thread_id = __atomic_fetch_add (&next_thread_id, 1, __ATOMIC_RELAXED); + + /* Determine the sizes to be used by this thread. */ + int size_slot = thread_id % (array_length (sizes) + 1); + bool switch_sizes = size_slot == array_length (sizes); + if (switch_sizes) + size_slot = 0; + + /* Compute the random blobs. */ + for (int i = 0; i < count_per_thread; ++i) + { + struct blob *place = dynarray_blob_emplace (&result); + TEST_VERIFY_EXIT (place != NULL); + place->size = sizes[size_slot]; + place->thread_id = thread_id; + place->index = i; + arc4random_buf (place->bytes, place->size); + + if (switch_sizes) + size_slot = (size_slot + 1) % array_length (sizes); + } + + /* Store the blobs in the global result structure. */ + copy_result_to_global (&result); + + dynarray_blob_free (&result); + + return NULL; +} + +/* Launch the inner threads and wait for their termination. */ +static void * +outer_thread (void *unused) +{ + for (int round = 0; round < outer_rounds; ++round) + { + pthread_t threads[inner_threads]; + + for (int i = 0; i < inner_threads; ++i) + threads[i] = xpthread_create (NULL, inner_thread, NULL); + + for (int i = 0; i < inner_threads; ++i) + xpthread_join (threads[i]); + } + + return NULL; +} + +static bool termination_requested; + +/* Call arc4random_buf to fill one blob with 16 bytes. */ +static void * +get_one_blob_thread (void *closure) +{ + struct blob *result = closure; + result->size = 16; + arc4random_buf (result->bytes, result->size); + return NULL; +} + +/* Invoked from fork_thread to actually obtain randomness data. */ +static void +fork_thread_subprocess (void *closure) +{ + struct blob *shared_result = closure; + + pthread_t thr1 = xpthread_create + (NULL, get_one_blob_thread, shared_result + 1); + pthread_t thr2 = xpthread_create + (NULL, get_one_blob_thread, shared_result + 2); + get_one_blob_thread (shared_result); + xpthread_join (thr1); + xpthread_join (thr2); +} + +/* Continuously fork subprocesses to obtain a little bit of + randomness. */ +static void * +fork_thread (void *unused) +{ + struct dynarray_blob result; + dynarray_blob_init (&result); + + /* Three blobs from each subprocess. */ + struct blob *shared_result + = support_shared_allocate (3 * sizeof (*shared_result)); + + while (!__atomic_load_n (&termination_requested, __ATOMIC_RELAXED)) + { + /* Obtain the results from a subprocess. */ + support_isolate_in_subprocess (fork_thread_subprocess, shared_result); + + for (int i = 0; i < 3; ++i) + { + struct blob *place = dynarray_blob_emplace (&result); + TEST_VERIFY_EXIT (place != NULL); + place->size = shared_result[i].size; + place->thread_id = -1; + place->index = i; + memcpy (place->bytes, shared_result[i].bytes, place->size); + } + } + + support_shared_free (shared_result); + + copy_result_to_global (&result); + dynarray_blob_free (&result); + + return NULL; +} + +/* Launch the outer threads and wait for their termination. */ +static void +run_outer_threads (void) +{ + /* Special thread that continuously calls fork. */ + pthread_t fork_thread_id = xpthread_create (NULL, fork_thread, NULL); + + pthread_t threads[outer_threads]; + for (int i = 0; i < outer_threads; ++i) + threads[i] = xpthread_create (NULL, outer_thread, NULL); + + for (int i = 0; i < outer_threads; ++i) + xpthread_join (threads[i]); + + __atomic_store_n (&termination_requested, true, __ATOMIC_RELAXED); + xpthread_join (fork_thread_id); +} + +static int +do_test (void) +{ + dynarray_blob_init (&global_result); + int expected_blobs + = count_per_thread * inner_threads * outer_threads * outer_rounds; + printf ("info: minimum of %d blob results expected\n", expected_blobs); + + run_outer_threads (); + + /* The forking thread delivers a non-deterministic number of + results, which is why expected_blobs is only a minimun number of + results. */ + printf ("info: %zu blob results observed\n", + dynarray_blob_size (&global_result)); + TEST_VERIFY (dynarray_blob_size (&global_result) >= expected_blobs); + + /* Verify that there are no duplicates. */ + qsort (dynarray_blob_begin (&global_result), + dynarray_blob_size (&global_result), + sizeof (struct blob), compare_blob); + struct blob *end = dynarray_blob_end (&global_result); + for (struct blob *p = dynarray_blob_begin (&global_result) + 1; + p < end; ++p) + { + if (compare_blob (p - 1, p) == 0) + { + support_record_failure (); + char *quoted = support_quote_blob (p->bytes, p->size); + printf ("error: duplicate blob: \"%s\" (%d bytes)\n", + quoted, (int) p->size); + printf (" first source: thread %d, index %u\n", + p[-1].thread_id, p[-1].index); + printf (" second source: thread %d, index %u\n", + p[0].thread_id, p[0].index); + free (quoted); + } + } + + dynarray_blob_free (&global_result); + + return 0; +} + +#include From patchwork Tue Apr 19 21:28:06 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Adhemerval Zanella Netto X-Patchwork-Id: 53050 Return-Path: X-Original-To: patchwork@sourceware.org Delivered-To: patchwork@sourceware.org Received: from server2.sourceware.org (localhost [IPv6:::1]) by sourceware.org (Postfix) with ESMTP id DD73C385735D for ; Tue, 19 Apr 2022 21:30:59 +0000 (GMT) DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org DD73C385735D DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=sourceware.org; s=default; t=1650403859; bh=DbOz/Ryb45xxu6D/qLKLRm8/5hqWAy7vsTdJ/jJbnds=; h=To:Subject:Date:In-Reply-To:References:List-Id:List-Unsubscribe: List-Archive:List-Post:List-Help:List-Subscribe:From:Reply-To: From; b=C2ndL6NVbdDUgZZlonSaFp3SnekyvU5GqSmcco6y2j/s9GMPmb1M1pUAMbVM3b2Ef DKKJFkgV31MfnhD4ptrWlhKDuWEZ/jgCnWkeW98Jg90uyqZNpGY83sD5pIryqgHugx Tgxxl48ZZouFdiLA8++OnHxhLqVSGEmGVqBJaS/A= X-Original-To: libc-alpha@sourceware.org Delivered-To: libc-alpha@sourceware.org Received: from mail-oi1-x22e.google.com (mail-oi1-x22e.google.com [IPv6:2607:f8b0:4864:20::22e]) by sourceware.org (Postfix) with ESMTPS id 2FFA03857827 for ; Tue, 19 Apr 2022 21:28:27 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org 2FFA03857827 Received: by mail-oi1-x22e.google.com with SMTP id e4so14627oif.2 for ; Tue, 19 Apr 2022 14:28:27 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; h=x-gm-message-state:from:to:subject:date:message-id:in-reply-to :references:mime-version:content-transfer-encoding; bh=DbOz/Ryb45xxu6D/qLKLRm8/5hqWAy7vsTdJ/jJbnds=; b=pntIJEdPH6vvn9UQ8JYuBZ6NMiAL+plysAgazG2nJj8cm/5HQddEpvl7+JTPviTDiS eKaIxCIXZuTqvfsNPt0/OUTLL3WkV7IVzYpoqHXmdZl6yF5BgmAAQlTjxX7f4Rq5mkPb 2c0c46RFHzydQfXf98PoFV9wpUiPGxocNjBUntFnBbd0IufU9ZbMSbTqPP/BsYqo/ozD FZP4tqaaq6wmZAiu3LxRc22iBGuh90K3erhwRE7V2CB3Fsx8SsTDBPhmPr/t2vbA6Nlg +VRVFBlkAVlSAzDVkqpu1iLHtsbkN6Gdos1jpq9k/AQYlJVtBy2+TBGWb+V6omhri0TS raqg== X-Gm-Message-State: AOAM532dYQ8kWdlozBSibZuHXS7cIfvu5Lph5AogRLNO8a5Ia4JTmmQD r7Ui2LDYQKdLqByHewspSQrl63MXVhe9AA== X-Google-Smtp-Source: ABdhPJzCfG/24zMgJmp1h0Yny5ZZbd+G8yb614OIZkhAT8bI3qmtiiC/rdgOpaSm3fTzgRwRtrnuTw== X-Received: by 2002:a05:6808:1a94:b0:322:a7aa:e89d with SMTP id bm20-20020a0568081a9400b00322a7aae89dmr262205oib.262.1650403704164; Tue, 19 Apr 2022 14:28:24 -0700 (PDT) Received: from birita.. ([2804:431:c7ca:c9d0:98f6:7aed:2f61:2745]) by smtp.gmail.com with ESMTPSA id nf9-20020a056871460900b000e2c44ca8edsm5473321oab.6.2022.04.19.14.28.22 for (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Tue, 19 Apr 2022 14:28:23 -0700 (PDT) To: libc-alpha@sourceware.org Subject: [PATCH v3 3/9] benchtests: Add arc4random benchtest Date: Tue, 19 Apr 2022 18:28:06 -0300 Message-Id: <20220419212812.2688764-4-adhemerval.zanella@linaro.org> X-Mailer: git-send-email 2.32.0 In-Reply-To: <20220419212812.2688764-1-adhemerval.zanella@linaro.org> References: <20220419212812.2688764-1-adhemerval.zanella@linaro.org> MIME-Version: 1.0 X-Spam-Status: No, score=-12.5 required=5.0 tests=BAYES_00, DKIM_SIGNED, DKIM_VALID, DKIM_VALID_AU, DKIM_VALID_EF, GIT_PATCH_0, KAM_SHORT, RCVD_IN_DNSWL_NONE, SPF_HELO_NONE, SPF_PASS, TXREP, T_SCC_BODY_TEXT_LINE autolearn=ham autolearn_force=no version=3.4.4 X-Spam-Checker-Version: SpamAssassin 3.4.4 (2020-01-24) on server2.sourceware.org X-BeenThere: libc-alpha@sourceware.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Libc-alpha mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-Patchwork-Original-From: Adhemerval Zanella via Libc-alpha From: Adhemerval Zanella Netto Reply-To: Adhemerval Zanella Errors-To: libc-alpha-bounces+patchwork=sourceware.org@sourceware.org Sender: "Libc-alpha" It shows both throughput (total bytes obtained in the test duration) and latecy for both arc4random and arc4random_buf with different sizes. Checked on x86_64-linux-gnu, aarch64-linux, and powerpc64le-linux-gnu. --- benchtests/Makefile | 6 +- benchtests/bench-arc4random.c | 224 ++++++++++++++++++++++++++++++++++ 2 files changed, 229 insertions(+), 1 deletion(-) create mode 100644 benchtests/bench-arc4random.c diff --git a/benchtests/Makefile b/benchtests/Makefile index 8dfca592fd..50b96dd71f 100644 --- a/benchtests/Makefile +++ b/benchtests/Makefile @@ -111,8 +111,12 @@ bench-string := \ ffsll \ # bench-string +bench-stdlib := \ + arc4random \ +# bench-stdlib + ifeq (${BENCHSET},) -bench := $(bench-math) $(bench-pthread) $(bench-string) +bench := $(bench-math) $(bench-pthread) $(bench-string) $(bench-stdlib) else bench := $(foreach B,$(filter bench-%,${BENCHSET}), ${${B}}) endif diff --git a/benchtests/bench-arc4random.c b/benchtests/bench-arc4random.c new file mode 100644 index 0000000000..626f2ba48c --- /dev/null +++ b/benchtests/bench-arc4random.c @@ -0,0 +1,224 @@ +/* arc4random benchmarks. + Copyright (C) 2022 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + . */ + +#include "bench-timing.h" +#include "json-lib.h" +#include +#include +#include +#include +#include +#include +#include +#include +#include + +/* Prevent compiler to optimize away call. */ +#define DO_NOT_OPTIMIZE(__value) \ + ({ \ + __typeof (__value) __v = (__value); \ + asm volatile("" : : "r,m" (__v) : "memory");\ + }) + +static volatile sig_atomic_t timer_finished; + +static void timer_callback (int unused) +{ + timer_finished = 1; +} + +static timer_t timer; + +/* Run for approximately DURATION seconds, and it does not matter who + receive the signal (so not need to mask it on main thread). */ +static void +timer_start (void) +{ + timer_finished = 0; + timer = support_create_timer (DURATION, 0, false, timer_callback); +} +static void +timer_stop (void) +{ + support_delete_timer (timer); +} + +static const uint32_t sizes[] = { 0, 16, 32, 48, 64, 80, 96, 112, 128 }; + +static double +bench_throughput (void) +{ + uint64_t n = 0; + + struct timespec start, end; + clock_gettime (CLOCK_MONOTONIC, &start); + while (1) + { + DO_NOT_OPTIMIZE (arc4random ()); + n++; + + if (timer_finished == 1) + break; + } + clock_gettime (CLOCK_MONOTONIC, &end); + struct timespec diff = timespec_sub (end, start); + + double total = (double) n * sizeof (uint32_t); + double duration = (double) diff.tv_sec + + (double) diff.tv_nsec / TIMESPEC_HZ; + + return total / duration; +} + +static double +bench_latency (void) +{ + timing_t start, stop, cur; + const size_t iters = 1024; + + TIMING_NOW (start); + for (size_t i = 0; i < iters; i++) + DO_NOT_OPTIMIZE (arc4random ()); + TIMING_NOW (stop); + + TIMING_DIFF (cur, start, stop); + + return (double) (cur) / (double) iters; +} + +static double +bench_buf_throughput (size_t len) +{ + uint8_t buf[len]; + uint64_t n = 0; + + struct timespec start, end; + clock_gettime (CLOCK_MONOTONIC, &start); + while (1) + { + arc4random_buf (buf, len); + n++; + + if (timer_finished == 1) + break; + } + clock_gettime (CLOCK_MONOTONIC, &end); + struct timespec diff = timespec_sub (end, start); + + double total = (double) n * len; + double duration = (double) diff.tv_sec + + (double) diff.tv_nsec / TIMESPEC_HZ; + + return total / duration; +} + +static double +bench_buf_latency (size_t len) +{ + timing_t start, stop, cur; + const size_t iters = 1024; + + uint8_t buf[len]; + + TIMING_NOW (start); + for (size_t i = 0; i < iters; i++) + arc4random_buf (buf, len); + TIMING_NOW (stop); + + TIMING_DIFF (cur, start, stop); + + return (double) (cur) / (double) iters; +} + +static void +bench_singlethread (json_ctx_t *json_ctx) +{ + json_element_object_begin (json_ctx); + + json_array_begin (json_ctx, "throughput"); + for (int i = 0; i < array_length (sizes); i++) + { + timer_start (); + double r = sizes[i] == 0 + ? bench_throughput () : bench_buf_throughput (sizes[i]); + timer_stop (); + + json_element_double (json_ctx, r); + } + json_array_end (json_ctx); + + json_array_begin (json_ctx, "latency"); + for (int i = 0; i < array_length (sizes); i++) + { + timer_start (); + double r = sizes[i] == 0 + ? bench_latency () : bench_buf_latency (sizes[i]); + timer_stop (); + + json_element_double (json_ctx, r); + } + json_array_end (json_ctx); + + json_element_object_end (json_ctx); +} + +static void +run_bench (json_ctx_t *json_ctx, const char *name, + char *const*fnames, size_t fnameslen, + void (*bench)(json_ctx_t *ctx)) +{ + json_attr_object_begin (json_ctx, name); + json_array_begin (json_ctx, "functions"); + for (int i = 0; i < fnameslen; i++) + json_element_string (json_ctx, fnames[i]); + json_array_end (json_ctx); + + json_array_begin (json_ctx, "results"); + bench (json_ctx); + json_array_end (json_ctx); + json_attr_object_end (json_ctx); +} + +static int +do_test (void) +{ + char *fnames[array_length (sizes)]; + for (int i = 0; i < array_length (sizes); i++) + if (sizes[i] == 0) + fnames[i] = xasprintf ("arc4random"); + else + fnames[i] = xasprintf ("arc4random_buf(%u)", sizes[i]); + + json_ctx_t json_ctx; + json_init (&json_ctx, 0, stdout); + + json_document_begin (&json_ctx); + json_attr_string (&json_ctx, "timing_type", TIMING_TYPE); + + run_bench (&json_ctx, "single-thread", fnames, array_length (fnames), + bench_singlethread); + + json_document_end (&json_ctx); + + for (int i = 0; i < array_length (sizes); i++) + free (fnames[i]); + + return 0; +} + +#include From patchwork Tue Apr 19 21:28:07 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Adhemerval Zanella Netto X-Patchwork-Id: 53051 Return-Path: X-Original-To: patchwork@sourceware.org Delivered-To: patchwork@sourceware.org Received: from server2.sourceware.org (localhost [IPv6:::1]) by sourceware.org (Postfix) with ESMTP id 5B2193857340 for ; Tue, 19 Apr 2022 21:31:47 +0000 (GMT) DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org 5B2193857340 DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=sourceware.org; s=default; t=1650403907; bh=uqekK+PYyq7VI4FftvQ6xI2B1vxudbHkn3IHCnWYqPs=; h=To:Subject:Date:In-Reply-To:References:List-Id:List-Unsubscribe: List-Archive:List-Post:List-Help:List-Subscribe:From:Reply-To: From; b=IZNMH/qrKyumWuKz+ANuTV0bQyIWk4Uzj+M9exBPTfhGF3zs+JSE1BqUESQgU8ieS snDcdQMlb2RnvDaSslncyHPCgrbzsv117A2pfA1e17J+18w8xyH+MUAsbb9E17DJ2J ObityLPXfFxWQcdZegowluOmWDAUIC+wcYEj1V/o= X-Original-To: libc-alpha@sourceware.org Delivered-To: libc-alpha@sourceware.org Received: from mail-oi1-x22a.google.com (mail-oi1-x22a.google.com [IPv6:2607:f8b0:4864:20::22a]) by sourceware.org (Postfix) with ESMTPS id B989A3857344 for ; Tue, 19 Apr 2022 21:28:29 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org B989A3857344 Received: by mail-oi1-x22a.google.com with SMTP id r85so14370695oie.7 for ; Tue, 19 Apr 2022 14:28:29 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; h=x-gm-message-state:from:to:subject:date:message-id:in-reply-to :references:mime-version:content-transfer-encoding; bh=uqekK+PYyq7VI4FftvQ6xI2B1vxudbHkn3IHCnWYqPs=; b=dESmwb9k+MxHSZ2uuDCe3pmOm6XLvd2uTKHHtZkJitdmRHxJwV+Px+TDY8ar50XT5B 9zFuqFyBC1SvDapzQ7Z9AJruWbmRmE1ncy12Yp8To/nHP7I/PbmsgWLb30kxjmqHYgdO hdLRkutObyHcS8Iqx8Im7vOyXx/gv1DHZBs9onin13msEwRtUvfl+tyBS2KiTI9dZWmR 5bvubZ16210HOfqnafhUO7JXvDEiz1tFsOBMTBp+Bk+x39IfBf+rBGpeZj4GkTLKNbVi Ub9jNh52dS/LiZwvGI19qcyDt1DaIS18nfTt0u+HdsMUto0IEwEArtn1ShKlMOrRsVwT lWVw== X-Gm-Message-State: AOAM532n5Getw/f9CAoG++CLNFn+KDpalReBa+6/nKydRQKUPL27FsL2 bJWb9HLCSO0vDkrwxOrImJiR3cGhRjJS+Q== X-Google-Smtp-Source: ABdhPJxVxCK6y3zTbK6HIMSioTeBSbwOjBp4jHskAJ1fQxxxScV3gr9vVt4n7Cm0slhrMGFsn8bryQ== X-Received: by 2002:a05:6808:2008:b0:2da:5b12:83ff with SMTP id q8-20020a056808200800b002da5b1283ffmr266508oiw.216.1650403706873; Tue, 19 Apr 2022 14:28:26 -0700 (PDT) Received: from birita.. ([2804:431:c7ca:c9d0:98f6:7aed:2f61:2745]) by smtp.gmail.com with ESMTPSA id nf9-20020a056871460900b000e2c44ca8edsm5473321oab.6.2022.04.19.14.28.24 for (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Tue, 19 Apr 2022 14:28:25 -0700 (PDT) To: libc-alpha@sourceware.org Subject: [PATCH v3 4/9] aarch64: Add optimized chacha20 Date: Tue, 19 Apr 2022 18:28:07 -0300 Message-Id: <20220419212812.2688764-5-adhemerval.zanella@linaro.org> X-Mailer: git-send-email 2.32.0 In-Reply-To: <20220419212812.2688764-1-adhemerval.zanella@linaro.org> References: <20220419212812.2688764-1-adhemerval.zanella@linaro.org> MIME-Version: 1.0 X-Spam-Status: No, score=-11.9 required=5.0 tests=BAYES_00, DKIM_SIGNED, DKIM_VALID, DKIM_VALID_AU, DKIM_VALID_EF, GIT_PATCH_0, KAM_ASCII_DIVIDERS, KAM_NUMSUBJECT, KAM_SHORT, RCVD_IN_DNSWL_NONE, SPF_HELO_NONE, SPF_PASS, TXREP, T_SCC_BODY_TEXT_LINE autolearn=ham autolearn_force=no version=3.4.4 X-Spam-Checker-Version: SpamAssassin 3.4.4 (2020-01-24) on server2.sourceware.org X-BeenThere: libc-alpha@sourceware.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Libc-alpha mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-Patchwork-Original-From: Adhemerval Zanella via Libc-alpha From: Adhemerval Zanella Netto Reply-To: Adhemerval Zanella Errors-To: libc-alpha-bounces+patchwork=sourceware.org@sourceware.org Sender: "Libc-alpha" It adds vectorized ChaCha20 implementation based on libgcrypt cipher/chacha20-aarch64.S. It is used as default and only little-endian is supported (BE uses generic fallback code). As for generic implementation, the last step that XOR with the input is omited. On a Neoverse-N1 it shows the following improvements (using formatted bench-arc4random data): GENERIC Function MB/s -------------------------------------------------- arc4random [single-thread] 149.74 arc4random_buf(16) [single-thread] 259.47 arc4random_buf(32) [single-thread] 325.34 arc4random_buf(48) [single-thread] 347.49 arc4random_buf(64) [single-thread] 361.60 arc4random_buf(80) [single-thread] 371.36 arc4random_buf(96) [single-thread] 383.19 arc4random_buf(112) [single-thread] 386.03 arc4random_buf(128) [single-thread] 388.69 -------------------------------------------------- OPTIMIZED Function MB/s -------------------------------------------------- arc4random [single-thread] 154.98 arc4random_buf(16) [single-thread] 342.63 arc4random_buf(32) [single-thread] 485.91 arc4random_buf(48) [single-thread] 539.95 arc4random_buf(64) [single-thread] 593.38 arc4random_buf(80) [single-thread] 629.45 arc4random_buf(96) [single-thread] 655.78 arc4random_buf(112) [single-thread] 670.54 arc4random_buf(128) [single-thread] 681.65 -------------------------------------------------- Checked on aarch64-linux-gnu. --- LICENSES | 20 ++ stdlib/chacha20.c | 8 +- sysdeps/aarch64/Makefile | 4 + sysdeps/aarch64/chacha20-neon.S | 323 ++++++++++++++++++++++++++++++++ sysdeps/aarch64/chacha20_arch.h | 40 ++++ sysdeps/generic/chacha20_arch.h | 24 +++ 6 files changed, 417 insertions(+), 2 deletions(-) create mode 100644 sysdeps/aarch64/chacha20-neon.S create mode 100644 sysdeps/aarch64/chacha20_arch.h create mode 100644 sysdeps/generic/chacha20_arch.h diff --git a/LICENSES b/LICENSES index 530893b1dc..7288d281dc 100644 --- a/LICENSES +++ b/LICENSES @@ -389,3 +389,23 @@ Copyright 2001 by Stephen L. Moshier You should have received a copy of the GNU Lesser General Public License along with this library; if not, see . */ + +sysdeps/aarch64/chacha20.S imports code from libgcrypt, with the +following notices: + +Copyright (C) 2017-2019 Jussi Kivilinna + +This file is part of Libgcrypt. + +Libgcrypt is free software; you can redistribute it and/or modify +it under the terms of the GNU Lesser General Public License as +published by the Free Software Foundation; either version 2.1 of +the License, or (at your option) any later version. + +Libgcrypt is distributed in the hope that it will be useful, +but WITHOUT ANY WARRANTY; without even the implied warranty of +MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +GNU Lesser General Public License for more details. + +You should have received a copy of the GNU Lesser General Public +License along with this program; if not, see . diff --git a/stdlib/chacha20.c b/stdlib/chacha20.c index af4ffa9860..fea4994169 100644 --- a/stdlib/chacha20.c +++ b/stdlib/chacha20.c @@ -134,8 +134,9 @@ chacha20_block (uint32_t *state, uint32_t *stream) } static void -chacha20_crypt (uint32_t *state, uint8_t *dst, const uint8_t *src, - size_t bytes) +__attribute_maybe_unused__ +chacha20_crypt_generic (uint32_t *state, uint8_t *dst, const uint8_t *src, + size_t bytes) { uint32_t stream[CHACHA20_BLOCK_WORDS]; @@ -161,3 +162,6 @@ chacha20_crypt (uint32_t *state, uint8_t *dst, const uint8_t *src, memcpy (dst, stream, bytes); } } + +/* Get the architecture optimized version. */ +#include diff --git a/sysdeps/aarch64/Makefile b/sysdeps/aarch64/Makefile index 7183895d04..812cdc2c00 100644 --- a/sysdeps/aarch64/Makefile +++ b/sysdeps/aarch64/Makefile @@ -50,6 +50,10 @@ ifeq ($(subdir),csu) gen-as-const-headers += tlsdesc.sym endif +ifeq ($(subdir),stdlib) +sysdep_routines += chacha20-neon +endif + ifeq ($(subdir),gmon) CFLAGS-mcount.c += -mgeneral-regs-only endif diff --git a/sysdeps/aarch64/chacha20-neon.S b/sysdeps/aarch64/chacha20-neon.S new file mode 100644 index 0000000000..f5652d5062 --- /dev/null +++ b/sysdeps/aarch64/chacha20-neon.S @@ -0,0 +1,323 @@ +/* Optimized AArch64 implementation of ChaCha20 cipher. + Copyright (C) 2022 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + . */ + +#include + +/* Only LE is supported. */ +#ifdef __AARCH64EL__ + +/* Based on D. J. Bernstein reference implementation at + http://cr.yp.to/chacha.html: + + chacha-regs.c version 20080118 + D. J. Bernstein + Public domain. */ + +#define GET_DATA_POINTER(reg, name) \ + adrp reg, name ; \ + add reg, reg, :lo12:name + +/* 'ret' instruction replacement for straight-line speculation mitigation */ +#define ret_spec_stop \ + ret; dsb sy; isb; + +.cpu generic+simd + +.text + +/* register macros */ +#define INPUT x0 +#define DST x1 +#define SRC x2 +#define NBLKS x3 +#define ROUND x4 +#define INPUT_CTR x5 +#define INPUT_POS x6 +#define CTR x7 + +/* vector registers */ +#define X0 v16 +#define X4 v17 +#define X8 v18 +#define X12 v19 + +#define X1 v20 +#define X5 v21 + +#define X9 v22 +#define X13 v23 +#define X2 v24 +#define X6 v25 + +#define X3 v26 +#define X7 v27 +#define X11 v28 +#define X15 v29 + +#define X10 v30 +#define X14 v31 + +#define VCTR v0 +#define VTMP0 v1 +#define VTMP1 v2 +#define VTMP2 v3 +#define VTMP3 v4 +#define X12_TMP v5 +#define X13_TMP v6 +#define ROT8 v7 + +/********************************************************************** + helper macros + **********************************************************************/ + +#define _(...) __VA_ARGS__ + +#define vpunpckldq(s1, s2, dst) \ + zip1 dst.4s, s2.4s, s1.4s; + +#define vpunpckhdq(s1, s2, dst) \ + zip2 dst.4s, s2.4s, s1.4s; + +#define vpunpcklqdq(s1, s2, dst) \ + zip1 dst.2d, s2.2d, s1.2d; + +#define vpunpckhqdq(s1, s2, dst) \ + zip2 dst.2d, s2.2d, s1.2d; + +/* 4x4 32-bit integer matrix transpose */ +#define transpose_4x4(x0, x1, x2, x3, t1, t2, t3) \ + vpunpckhdq(x1, x0, t2); \ + vpunpckldq(x1, x0, x0); \ + \ + vpunpckldq(x3, x2, t1); \ + vpunpckhdq(x3, x2, x2); \ + \ + vpunpckhqdq(t1, x0, x1); \ + vpunpcklqdq(t1, x0, x0); \ + \ + vpunpckhqdq(x2, t2, x3); \ + vpunpcklqdq(x2, t2, x2); + +#define clear(x) \ + movi x.16b, #0; + +/********************************************************************** + 4-way chacha20 + **********************************************************************/ + +#define XOR(d,s1,s2) \ + eor d.16b, s2.16b, s1.16b; + +#define PLUS(ds,s) \ + add ds.4s, ds.4s, s.4s; + +#define ROTATE4(dst1,dst2,dst3,dst4,c,src1,src2,src3,src4) \ + shl dst1.4s, src1.4s, #(c); \ + shl dst2.4s, src2.4s, #(c); \ + shl dst3.4s, src3.4s, #(c); \ + shl dst4.4s, src4.4s, #(c); \ + sri dst1.4s, src1.4s, #(32 - (c)); \ + sri dst2.4s, src2.4s, #(32 - (c)); \ + sri dst3.4s, src3.4s, #(32 - (c)); \ + sri dst4.4s, src4.4s, #(32 - (c)); + +#define ROTATE4_8(dst1,dst2,dst3,dst4,src1,src2,src3,src4) \ + tbl dst1.16b, {src1.16b}, ROT8.16b; \ + tbl dst2.16b, {src2.16b}, ROT8.16b; \ + tbl dst3.16b, {src3.16b}, ROT8.16b; \ + tbl dst4.16b, {src4.16b}, ROT8.16b; + +#define ROTATE4_16(dst1,dst2,dst3,dst4,src1,src2,src3,src4) \ + rev32 dst1.8h, src1.8h; \ + rev32 dst2.8h, src2.8h; \ + rev32 dst3.8h, src3.8h; \ + rev32 dst4.8h, src4.8h; + +#define QUARTERROUND4(a1,b1,c1,d1,a2,b2,c2,d2,a3,b3,c3,d3,a4,b4,c4,d4,ign,tmp1,tmp2,tmp3,tmp4) \ + PLUS(a1,b1); PLUS(a2,b2); \ + PLUS(a3,b3); PLUS(a4,b4); \ + XOR(tmp1,d1,a1); XOR(tmp2,d2,a2); \ + XOR(tmp3,d3,a3); XOR(tmp4,d4,a4); \ + ROTATE4_16(d1, d2, d3, d4, tmp1, tmp2, tmp3, tmp4); \ + PLUS(c1,d1); PLUS(c2,d2); \ + PLUS(c3,d3); PLUS(c4,d4); \ + XOR(tmp1,b1,c1); XOR(tmp2,b2,c2); \ + XOR(tmp3,b3,c3); XOR(tmp4,b4,c4); \ + ROTATE4(b1, b2, b3, b4, 12, tmp1, tmp2, tmp3, tmp4) \ + PLUS(a1,b1); PLUS(a2,b2); \ + PLUS(a3,b3); PLUS(a4,b4); \ + XOR(tmp1,d1,a1); XOR(tmp2,d2,a2); \ + XOR(tmp3,d3,a3); XOR(tmp4,d4,a4); \ + ROTATE4_8(d1, d2, d3, d4, tmp1, tmp2, tmp3, tmp4) \ + PLUS(c1,d1); PLUS(c2,d2); \ + PLUS(c3,d3); PLUS(c4,d4); \ + XOR(tmp1,b1,c1); XOR(tmp2,b2,c2); \ + XOR(tmp3,b3,c3); XOR(tmp4,b4,c4); \ + ROTATE4(b1, b2, b3, b4, 7, tmp1, tmp2, tmp3, tmp4) \ + +.align 4 +L(__chacha20_blocks4_data_inc_counter): + .long 0,1,2,3 + +.align 4 +L(__chacha20_blocks4_data_rot8): + .byte 3,0,1,2 + .byte 7,4,5,6 + .byte 11,8,9,10 + .byte 15,12,13,14 + +.hidden __chacha20_neon_blocks4 +ENTRY (__chacha20_neon_blocks4) + /* input: + * x0: input + * x1: dst + * x2: src + * x3: nblks (multiple of 4) + */ + + GET_DATA_POINTER(CTR, L(__chacha20_blocks4_data_rot8)) + add INPUT_CTR, INPUT, #(12*4); + ld1 {ROT8.16b}, [CTR]; + GET_DATA_POINTER(CTR, L(__chacha20_blocks4_data_inc_counter)) + mov INPUT_POS, INPUT; + ld1 {VCTR.16b}, [CTR]; + +L(loop4): + /* Construct counter vectors X12 and X13 */ + + ld1 {X15.16b}, [INPUT_CTR]; + mov ROUND, #20; + ld1 {VTMP1.16b-VTMP3.16b}, [INPUT_POS]; + + dup X12.4s, X15.s[0]; + dup X13.4s, X15.s[1]; + ldr CTR, [INPUT_CTR]; + add X12.4s, X12.4s, VCTR.4s; + dup X0.4s, VTMP1.s[0]; + dup X1.4s, VTMP1.s[1]; + dup X2.4s, VTMP1.s[2]; + dup X3.4s, VTMP1.s[3]; + dup X14.4s, X15.s[2]; + cmhi VTMP0.4s, VCTR.4s, X12.4s; + dup X15.4s, X15.s[3]; + add CTR, CTR, #4; /* Update counter */ + dup X4.4s, VTMP2.s[0]; + dup X5.4s, VTMP2.s[1]; + dup X6.4s, VTMP2.s[2]; + dup X7.4s, VTMP2.s[3]; + sub X13.4s, X13.4s, VTMP0.4s; + dup X8.4s, VTMP3.s[0]; + dup X9.4s, VTMP3.s[1]; + dup X10.4s, VTMP3.s[2]; + dup X11.4s, VTMP3.s[3]; + mov X12_TMP.16b, X12.16b; + mov X13_TMP.16b, X13.16b; + str CTR, [INPUT_CTR]; + +L(round2): + subs ROUND, ROUND, #2 + QUARTERROUND4(X0, X4, X8, X12, X1, X5, X9, X13, + X2, X6, X10, X14, X3, X7, X11, X15, + tmp:=,VTMP0,VTMP1,VTMP2,VTMP3) + QUARTERROUND4(X0, X5, X10, X15, X1, X6, X11, X12, + X2, X7, X8, X13, X3, X4, X9, X14, + tmp:=,VTMP0,VTMP1,VTMP2,VTMP3) + b.ne L(round2); + + ld1 {VTMP0.16b, VTMP1.16b}, [INPUT_POS], #32; + + PLUS(X12, X12_TMP); /* INPUT + 12 * 4 + counter */ + PLUS(X13, X13_TMP); /* INPUT + 13 * 4 + counter */ + + dup VTMP2.4s, VTMP0.s[0]; /* INPUT + 0 * 4 */ + dup VTMP3.4s, VTMP0.s[1]; /* INPUT + 1 * 4 */ + dup X12_TMP.4s, VTMP0.s[2]; /* INPUT + 2 * 4 */ + dup X13_TMP.4s, VTMP0.s[3]; /* INPUT + 3 * 4 */ + PLUS(X0, VTMP2); + PLUS(X1, VTMP3); + PLUS(X2, X12_TMP); + PLUS(X3, X13_TMP); + + dup VTMP2.4s, VTMP1.s[0]; /* INPUT + 4 * 4 */ + dup VTMP3.4s, VTMP1.s[1]; /* INPUT + 5 * 4 */ + dup X12_TMP.4s, VTMP1.s[2]; /* INPUT + 6 * 4 */ + dup X13_TMP.4s, VTMP1.s[3]; /* INPUT + 7 * 4 */ + ld1 {VTMP0.16b, VTMP1.16b}, [INPUT_POS]; + mov INPUT_POS, INPUT; + PLUS(X4, VTMP2); + PLUS(X5, VTMP3); + PLUS(X6, X12_TMP); + PLUS(X7, X13_TMP); + + dup VTMP2.4s, VTMP0.s[0]; /* INPUT + 8 * 4 */ + dup VTMP3.4s, VTMP0.s[1]; /* INPUT + 9 * 4 */ + dup X12_TMP.4s, VTMP0.s[2]; /* INPUT + 10 * 4 */ + dup X13_TMP.4s, VTMP0.s[3]; /* INPUT + 11 * 4 */ + dup VTMP0.4s, VTMP1.s[2]; /* INPUT + 14 * 4 */ + dup VTMP1.4s, VTMP1.s[3]; /* INPUT + 15 * 4 */ + PLUS(X8, VTMP2); + PLUS(X9, VTMP3); + PLUS(X10, X12_TMP); + PLUS(X11, X13_TMP); + PLUS(X14, VTMP0); + PLUS(X15, VTMP1); + + transpose_4x4(X0, X1, X2, X3, VTMP0, VTMP1, VTMP2); + transpose_4x4(X4, X5, X6, X7, VTMP0, VTMP1, VTMP2); + transpose_4x4(X8, X9, X10, X11, VTMP0, VTMP1, VTMP2); + transpose_4x4(X12, X13, X14, X15, VTMP0, VTMP1, VTMP2); + + subs NBLKS, NBLKS, #4; + + st1 {X0.16b,X4.16B,X8.16b, X12.16b}, [DST], #64 + st1 {X1.16b,X5.16b}, [DST], #32; + st1 {X9.16b, X13.16b, X2.16b, X6.16b}, [DST], #64 + st1 {X10.16b,X14.16b}, [DST], #32; + st1 {X3.16b, X7.16b, X11.16b, X15.16b}, [DST], #64; + + b.ne L(loop4); + + /* clear the used vector registers and stack */ + clear(VTMP0); + clear(VTMP1); + clear(VTMP2); + clear(VTMP3); + clear(X12_TMP); + clear(X13_TMP); + clear(X0); + clear(X1); + clear(X2); + clear(X3); + clear(X4); + clear(X5); + clear(X6); + clear(X7); + clear(X8); + clear(X9); + clear(X10); + clear(X11); + clear(X12); + clear(X13); + clear(X14); + clear(X15); + + eor x0, x0, x0 + ret_spec_stop +END (__chacha20_neon_blocks4) + +#endif diff --git a/sysdeps/aarch64/chacha20_arch.h b/sysdeps/aarch64/chacha20_arch.h new file mode 100644 index 0000000000..9febee7bb6 --- /dev/null +++ b/sysdeps/aarch64/chacha20_arch.h @@ -0,0 +1,40 @@ +/* Chacha20 implementation, used on arc4random. + Copyright (C) 2022 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + . */ + +#include +#include + +unsigned int __chacha20_neon_blocks4 (uint32_t *state, uint8_t *dst, + const uint8_t *src, size_t nblks) + attribute_hidden; + +static void +chacha20_crypt (uint32_t *state, uint8_t *dst, const uint8_t *src, + size_t bytes) +{ + _Static_assert (CHACHA20_BUFSIZE % 4 == 0, + "CHACHA20_BUFSIZE not multiple of 4"); + _Static_assert (CHACHA20_BUFSIZE > CHACHA20_BLOCK_SIZE * 4, + "CHACHA20_BUFSIZE <= CHACHA20_BLOCK_SIZE * 4"); +#ifdef __AARCH64EL__ + __chacha20_neon_blocks4 (state, dst, src, + CHACHA20_BUFSIZE / CHACHA20_BLOCK_SIZE); +#else + chacha20_crypt_generic (state, dst, src, bytes); +#endif +} diff --git a/sysdeps/generic/chacha20_arch.h b/sysdeps/generic/chacha20_arch.h new file mode 100644 index 0000000000..efad41d034 --- /dev/null +++ b/sysdeps/generic/chacha20_arch.h @@ -0,0 +1,24 @@ +/* Chacha20 implementation, generic interface for encrypt. + Copyright (C) 2022 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + . */ + +static inline void +chacha20_crypt (uint32_t *state, uint8_t *dst, const uint8_t *src, + size_t bytes) +{ + chacha20_crypt_generic (state, dst, src, bytes); +} From patchwork Tue Apr 19 21:28:08 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Adhemerval Zanella Netto X-Patchwork-Id: 53052 Return-Path: X-Original-To: patchwork@sourceware.org Delivered-To: patchwork@sourceware.org Received: from server2.sourceware.org (localhost [IPv6:::1]) by sourceware.org (Postfix) with ESMTP id 27CE23857836 for ; Tue, 19 Apr 2022 21:32:30 +0000 (GMT) DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org 27CE23857836 DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=sourceware.org; s=default; t=1650403950; bh=6CmFQs4vjqNh5lg4hELNDjOjnhS5rVjjihu0hEVUDOA=; h=To:Subject:Date:In-Reply-To:References:List-Id:List-Unsubscribe: List-Archive:List-Post:List-Help:List-Subscribe:From:Reply-To: From; b=AZN2BcQmU3s1vHZgycyXaKua2EaGLkLccjEvG2GWe0uIBiArw4W/aV01K4mCD/Ziu MQVFsWDejtZlxSpBhS5YEkLD/umh+jbji+deGdyIVfiUT4hEekHy10YZcsIV5GAjfK kd3DvXOfni3HT7ay1vEse5s3VvzzlxHbxM9/05VA= X-Original-To: libc-alpha@sourceware.org Delivered-To: libc-alpha@sourceware.org Received: from mail-oi1-x22c.google.com (mail-oi1-x22c.google.com [IPv6:2607:f8b0:4864:20::22c]) by sourceware.org (Postfix) with ESMTPS id 2DAC13857404 for ; Tue, 19 Apr 2022 21:28:31 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org 2DAC13857404 Received: by mail-oi1-x22c.google.com with SMTP id w127so19547999oig.10 for ; Tue, 19 Apr 2022 14:28:31 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; h=x-gm-message-state:from:to:subject:date:message-id:in-reply-to :references:mime-version:content-transfer-encoding; bh=6CmFQs4vjqNh5lg4hELNDjOjnhS5rVjjihu0hEVUDOA=; b=SdLXo3BgrgbOedHX0nMVB6zqgt3IMgwpueUWY9T+92KSzRRoFCM6ohjxGjmnDLv8Mb LNDbDXj/HKqN5vFGg0V+V1980UI7vysesUMnjpCl6pmQKIPcTSLaMEDUBn8Xmw8ijH6p ORjbDqH/QJ6XLvpC/xolemXuJLOQBEdR1r3hssciNLJV9mF753aGCruOFk+yuEA41UGW uenBTO45W39qDmtYrtU3iTqoPwOCorBsacmVbhnkOpFZTskjfREQ5KcRceldixUfK+Co Fc0BbdGZRWVFZaF/TNIA/Jts57+up0oOXUy0eO+OGE8PfGrMvjUSFBwGwJQNsrPqGM1M IdkQ== X-Gm-Message-State: AOAM530bDP32VIOWYHdiwXVQgDKBJM8qM9hgQYOyyYbJf58joR3Ssk4J QKGHMPYddKhyXpGPrdrH0kbcKnRLLYaa1A== X-Google-Smtp-Source: ABdhPJz7IWZhPvqHcEWCsa461m8YQ/+qeeWqceykUgLh4y6IWmnoYZTtpxxQd498axqwOfOStYNpHQ== X-Received: by 2002:a05:6808:d46:b0:2ec:f0ff:f8c8 with SMTP id w6-20020a0568080d4600b002ecf0fff8c8mr286791oik.56.1650403708440; Tue, 19 Apr 2022 14:28:28 -0700 (PDT) Received: from birita.. ([2804:431:c7ca:c9d0:98f6:7aed:2f61:2745]) by smtp.gmail.com with ESMTPSA id nf9-20020a056871460900b000e2c44ca8edsm5473321oab.6.2022.04.19.14.28.27 for (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Tue, 19 Apr 2022 14:28:27 -0700 (PDT) To: libc-alpha@sourceware.org Subject: [PATCH v3 5/9] x86: Add SSE2 optimized chacha20 Date: Tue, 19 Apr 2022 18:28:08 -0300 Message-Id: <20220419212812.2688764-6-adhemerval.zanella@linaro.org> X-Mailer: git-send-email 2.32.0 In-Reply-To: <20220419212812.2688764-1-adhemerval.zanella@linaro.org> References: <20220419212812.2688764-1-adhemerval.zanella@linaro.org> MIME-Version: 1.0 X-Spam-Status: No, score=-10.9 required=5.0 tests=BAYES_00, DKIM_SIGNED, DKIM_VALID, DKIM_VALID_AU, DKIM_VALID_EF, GIT_PATCH_0, KAM_ASCII_DIVIDERS, KAM_NUMSUBJECT, KAM_SHORT, RCVD_IN_DNSWL_NONE, SCC_10_SHORT_WORD_LINES, SCC_5_SHORT_WORD_LINES, SPF_HELO_NONE, SPF_PASS, TXREP, T_SCC_BODY_TEXT_LINE autolearn=ham autolearn_force=no version=3.4.4 X-Spam-Checker-Version: SpamAssassin 3.4.4 (2020-01-24) on server2.sourceware.org X-BeenThere: libc-alpha@sourceware.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Libc-alpha mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-Patchwork-Original-From: Adhemerval Zanella via Libc-alpha From: Adhemerval Zanella Netto Reply-To: Adhemerval Zanella Errors-To: libc-alpha-bounces+patchwork=sourceware.org@sourceware.org Sender: "Libc-alpha" It adds vectorized ChaCha20 implementation based on libgcrypt cipher/chacha20-amd64-ssse3.S. It replaces the ROTATE_SHUF_2 (which uses pshufb) by ROTATE2 and thus making the original implementation SSE2. As for generic implementation, the last step that XOR with the input is omited. On a Ryzen 9 5900X it shows the following improvements (using formatted bench-arc4random data): GENERIC Function MB/s -------------------------------------------------- arc4random [single-thread] 384.43 arc4random_buf(16) [single-thread] 465.00 arc4random_buf(32) [single-thread] 528.32 arc4random_buf(48) [single-thread] 546.26 arc4random_buf(64) [single-thread] 560.57 arc4random_buf(80) [single-thread] 562.37 arc4random_buf(96) [single-thread] 572.05 arc4random_buf(112) [single-thread] 573.12 arc4random_buf(128) [single-thread] 578.12 -------------------------------------------------- SSE2: -------------------------------------------------- arc4random [single-thread] 637.06 arc4random_buf(16) [single-thread] 856.62 arc4random_buf(32) [single-thread] 1129.41 arc4random_buf(48) [single-thread] 1260.61 arc4random_buf(64) [single-thread] 1330.56 arc4random_buf(80) [single-thread] 1353.84 arc4random_buf(96) [single-thread] 1376.53 arc4random_buf(112) [single-thread] 1405.74 arc4random_buf(128) [single-thread] 1422.59 -------------------------------------------------- Checked on x86_64-linux-gnu. --- LICENSES | 4 +- sysdeps/x86_64/Makefile | 6 + sysdeps/x86_64/chacha20-sse2.S | 311 +++++++++++++++++++++++++++++++++ sysdeps/x86_64/chacha20_arch.h | 38 ++++ 4 files changed, 357 insertions(+), 2 deletions(-) create mode 100644 sysdeps/x86_64/chacha20-sse2.S create mode 100644 sysdeps/x86_64/chacha20_arch.h diff --git a/LICENSES b/LICENSES index 7288d281dc..415991e208 100644 --- a/LICENSES +++ b/LICENSES @@ -390,8 +390,8 @@ Copyright 2001 by Stephen L. Moshier License along with this library; if not, see . */ -sysdeps/aarch64/chacha20.S imports code from libgcrypt, with the -following notices: +sysdeps/aarch64/chacha20.S and sysdeps/x86_64/chacha20-sse2.S +import code from libgcrypt, with the following notices: Copyright (C) 2017-2019 Jussi Kivilinna diff --git a/sysdeps/x86_64/Makefile b/sysdeps/x86_64/Makefile index 79365aff2a..c8fbc30857 100644 --- a/sysdeps/x86_64/Makefile +++ b/sysdeps/x86_64/Makefile @@ -5,6 +5,12 @@ ifeq ($(subdir),csu) gen-as-const-headers += link-defines.sym endif +ifeq ($(subdir),stdlib) +sysdep_routines += \ + chacha20-sse2 \ + # sysdep_routines +endif + ifeq ($(subdir),gmon) sysdep_routines += _mcount # We cannot compile _mcount.S with -pg because that would create diff --git a/sysdeps/x86_64/chacha20-sse2.S b/sysdeps/x86_64/chacha20-sse2.S new file mode 100644 index 0000000000..3a4cb7d2ea --- /dev/null +++ b/sysdeps/x86_64/chacha20-sse2.S @@ -0,0 +1,311 @@ +/* Optimized SSE2 implementation of ChaCha20 cipher. + Copyright (C) 2022 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + . */ + +/* Based on D. J. Bernstein reference implementation at + http://cr.yp.to/chacha.html: + + chacha-regs.c version 20080118 + D. J. Bernstein + Public domain. */ + +#include + +#ifdef PIC +# define rRIP (%rip) +#else +# define rRIP +#endif + +/* 'ret' instruction replacement for straight-line speculation mitigation */ +#define ret_spec_stop \ + ret; int3; + +/* register macros */ +#define INPUT %rdi +#define DST %rsi +#define SRC %rdx +#define NBLKS %rcx +#define ROUND %eax + +/* stack structure */ +#define STACK_VEC_X12 (16) +#define STACK_VEC_X13 (16 + STACK_VEC_X12) +#define STACK_TMP (16 + STACK_VEC_X13) +#define STACK_TMP1 (16 + STACK_TMP) +#define STACK_TMP2 (16 + STACK_TMP1) + +#define STACK_MAX (16 + STACK_TMP2) + +/* vector registers */ +#define X0 %xmm0 +#define X1 %xmm1 +#define X2 %xmm2 +#define X3 %xmm3 +#define X4 %xmm4 +#define X5 %xmm5 +#define X6 %xmm6 +#define X7 %xmm7 +#define X8 %xmm8 +#define X9 %xmm9 +#define X10 %xmm10 +#define X11 %xmm11 +#define X12 %xmm12 +#define X13 %xmm13 +#define X14 %xmm14 +#define X15 %xmm15 + +/********************************************************************** + helper macros + **********************************************************************/ + +/* 4x4 32-bit integer matrix transpose */ +#define TRANSPOSE_4x4(x0, x1, x2, x3, t1, t2, t3) \ + movdqa x0, t2; \ + punpckhdq x1, t2; \ + punpckldq x1, x0; \ + \ + movdqa x2, t1; \ + punpckldq x3, t1; \ + punpckhdq x3, x2; \ + \ + movdqa x0, x1; \ + punpckhqdq t1, x1; \ + punpcklqdq t1, x0; \ + \ + movdqa t2, x3; \ + punpckhqdq x2, x3; \ + punpcklqdq x2, t2; \ + movdqa t2, x2; + +/* fill xmm register with 32-bit value from memory */ +#define PBROADCASTD(mem32, xreg) \ + movd mem32, xreg; \ + pshufd $0, xreg, xreg; + +#define CLEAR(x) pxor x,x; + +/********************************************************************** + 4-way chacha20 + **********************************************************************/ + +#define ROTATE2(v1,v2,c,tmp1,tmp2) \ + movdqa v1, tmp1; \ + movdqa v2, tmp2; \ + psrld $(32 - (c)), v1; \ + pslld $(c), tmp1; \ + paddb tmp1, v1; \ + psrld $(32 - (c)), v2; \ + pslld $(c), tmp2; \ + paddb tmp2, v2; + +#define XOR(ds,s) \ + pxor s, ds; + +#define PLUS(ds,s) \ + paddd s, ds; + +#define QUARTERROUND2(a1,b1,c1,d1,a2,b2,c2,d2,ign,tmp1,tmp2) \ + PLUS(a1,b1); PLUS(a2,b2); XOR(d1,a1); XOR(d2,a2); \ + ROTATE2(d1, d2, 16, tmp1, tmp2); \ + PLUS(c1,d1); PLUS(c2,d2); XOR(b1,c1); XOR(b2,c2); \ + ROTATE2(b1, b2, 12, tmp1, tmp2); \ + PLUS(a1,b1); PLUS(a2,b2); XOR(d1,a1); XOR(d2,a2); \ + ROTATE2(d1, d2, 8, tmp1, tmp2); \ + PLUS(c1,d1); PLUS(c2,d2); XOR(b1,c1); XOR(b2,c2); \ + ROTATE2(b1, b2, 7, tmp1, tmp2); + + .section .text.sse2,"ax",@progbits + +chacha20_data: + .align 16 +L(counter1): + .long 1,0,0,0 +L(inc_counter): + .long 0,1,2,3 +L(unsigned_cmp): + .long 0x80000000,0x80000000,0x80000000,0x80000000 + + .hidden __chacha20_sse2_blocks4 +ENTRY (__chacha20_sse2_blocks4) + /* input: + * %rdi: input + * %rsi: dst + * %rdx: src + * %rcx: nblks (multiple of 4) + */ + + pushq %rbp; + cfi_adjust_cfa_offset(8); + cfi_rel_offset(rbp, 0) + movq %rsp, %rbp; + cfi_def_cfa_register(%rbp); + + subq $STACK_MAX, %rsp; + andq $~15, %rsp; + +L(loop4): + mov $20, ROUND; + + /* Construct counter vectors X12 and X13 */ + movdqa L(inc_counter) rRIP, X0; + movdqa L(unsigned_cmp) rRIP, X2; + PBROADCASTD((12 * 4)(INPUT), X12); + PBROADCASTD((13 * 4)(INPUT), X13); + paddd X0, X12; + movdqa X12, X1; + pxor X2, X0; + pxor X2, X1; + pcmpgtd X1, X0; + psubd X0, X13; + movdqa X12, (STACK_VEC_X12)(%rsp); + movdqa X13, (STACK_VEC_X13)(%rsp); + + /* Load vectors */ + PBROADCASTD((0 * 4)(INPUT), X0); + PBROADCASTD((1 * 4)(INPUT), X1); + PBROADCASTD((2 * 4)(INPUT), X2); + PBROADCASTD((3 * 4)(INPUT), X3); + PBROADCASTD((4 * 4)(INPUT), X4); + PBROADCASTD((5 * 4)(INPUT), X5); + PBROADCASTD((6 * 4)(INPUT), X6); + PBROADCASTD((7 * 4)(INPUT), X7); + PBROADCASTD((8 * 4)(INPUT), X8); + PBROADCASTD((9 * 4)(INPUT), X9); + PBROADCASTD((10 * 4)(INPUT), X10); + PBROADCASTD((11 * 4)(INPUT), X11); + PBROADCASTD((14 * 4)(INPUT), X14); + PBROADCASTD((15 * 4)(INPUT), X15); + movdqa X11, (STACK_TMP)(%rsp); + movdqa X15, (STACK_TMP1)(%rsp); + +L(round2_4): + QUARTERROUND2(X0, X4, X8, X12, X1, X5, X9, X13, tmp:=,X11,X15) + movdqa (STACK_TMP)(%rsp), X11; + movdqa (STACK_TMP1)(%rsp), X15; + movdqa X8, (STACK_TMP)(%rsp); + movdqa X9, (STACK_TMP1)(%rsp); + QUARTERROUND2(X2, X6, X10, X14, X3, X7, X11, X15, tmp:=,X8,X9) + QUARTERROUND2(X0, X5, X10, X15, X1, X6, X11, X12, tmp:=,X8,X9) + movdqa (STACK_TMP)(%rsp), X8; + movdqa (STACK_TMP1)(%rsp), X9; + movdqa X11, (STACK_TMP)(%rsp); + movdqa X15, (STACK_TMP1)(%rsp); + QUARTERROUND2(X2, X7, X8, X13, X3, X4, X9, X14, tmp:=,X11,X15) + sub $2, ROUND; + jnz L(round2_4); + + /* tmp := X15 */ + movdqa (STACK_TMP)(%rsp), X11; + PBROADCASTD((0 * 4)(INPUT), X15); + PLUS(X0, X15); + PBROADCASTD((1 * 4)(INPUT), X15); + PLUS(X1, X15); + PBROADCASTD((2 * 4)(INPUT), X15); + PLUS(X2, X15); + PBROADCASTD((3 * 4)(INPUT), X15); + PLUS(X3, X15); + PBROADCASTD((4 * 4)(INPUT), X15); + PLUS(X4, X15); + PBROADCASTD((5 * 4)(INPUT), X15); + PLUS(X5, X15); + PBROADCASTD((6 * 4)(INPUT), X15); + PLUS(X6, X15); + PBROADCASTD((7 * 4)(INPUT), X15); + PLUS(X7, X15); + PBROADCASTD((8 * 4)(INPUT), X15); + PLUS(X8, X15); + PBROADCASTD((9 * 4)(INPUT), X15); + PLUS(X9, X15); + PBROADCASTD((10 * 4)(INPUT), X15); + PLUS(X10, X15); + PBROADCASTD((11 * 4)(INPUT), X15); + PLUS(X11, X15); + movdqa (STACK_VEC_X12)(%rsp), X15; + PLUS(X12, X15); + movdqa (STACK_VEC_X13)(%rsp), X15; + PLUS(X13, X15); + movdqa X13, (STACK_TMP)(%rsp); + PBROADCASTD((14 * 4)(INPUT), X15); + PLUS(X14, X15); + movdqa (STACK_TMP1)(%rsp), X15; + movdqa X14, (STACK_TMP1)(%rsp); + PBROADCASTD((15 * 4)(INPUT), X13); + PLUS(X15, X13); + movdqa X15, (STACK_TMP2)(%rsp); + + /* Update counter */ + addq $4, (12 * 4)(INPUT); + + TRANSPOSE_4x4(X0, X1, X2, X3, X13, X14, X15); + movdqu X0, (64 * 0 + 16 * 0)(DST) + movdqu X1, (64 * 1 + 16 * 0)(DST) + movdqu X2, (64 * 2 + 16 * 0)(DST) + movdqu X3, (64 * 3 + 16 * 0)(DST) + TRANSPOSE_4x4(X4, X5, X6, X7, X0, X1, X2); + movdqa (STACK_TMP)(%rsp), X13; + movdqa (STACK_TMP1)(%rsp), X14; + movdqa (STACK_TMP2)(%rsp), X15; + movdqu X4, (64 * 0 + 16 * 1)(DST) + movdqu X5, (64 * 1 + 16 * 1)(DST) + movdqu X6, (64 * 2 + 16 * 1)(DST) + movdqu X7, (64 * 3 + 16 * 1)(DST) + TRANSPOSE_4x4(X8, X9, X10, X11, X0, X1, X2); + movdqu X8, (64 * 0 + 16 * 2)(DST) + movdqu X9, (64 * 1 + 16 * 2)(DST) + movdqu X10, (64 * 2 + 16 * 2)(DST) + movdqu X11, (64 * 3 + 16 * 2)(DST) + TRANSPOSE_4x4(X12, X13, X14, X15, X0, X1, X2); + movdqu X12, (64 * 0 + 16 * 3)(DST) + movdqu X13, (64 * 1 + 16 * 3)(DST) + movdqu X14, (64 * 2 + 16 * 3)(DST) + movdqu X15, (64 * 3 + 16 * 3)(DST) + + sub $4, NBLKS; + lea (4 * 64)(DST), DST; + lea (4 * 64)(SRC), SRC; + jnz L(loop4); + + /* CLEAR the used vector registers and stack */ + CLEAR(X0); + movdqa X0, (STACK_VEC_X12)(%rsp); + movdqa X0, (STACK_VEC_X13)(%rsp); + movdqa X0, (STACK_TMP)(%rsp); + movdqa X0, (STACK_TMP1)(%rsp); + movdqa X0, (STACK_TMP2)(%rsp); + CLEAR(X1); + CLEAR(X2); + CLEAR(X3); + CLEAR(X4); + CLEAR(X5); + CLEAR(X6); + CLEAR(X7); + CLEAR(X8); + CLEAR(X9); + CLEAR(X10); + CLEAR(X11); + CLEAR(X12); + CLEAR(X13); + CLEAR(X14); + CLEAR(X15); + + /* eax zeroed by round loop. */ + leave; + cfi_adjust_cfa_offset(-8) + cfi_def_cfa_register(%rsp); + ret_spec_stop; +END (__chacha20_sse2_blocks4) diff --git a/sysdeps/x86_64/chacha20_arch.h b/sysdeps/x86_64/chacha20_arch.h new file mode 100644 index 0000000000..5738c840a9 --- /dev/null +++ b/sysdeps/x86_64/chacha20_arch.h @@ -0,0 +1,38 @@ +/* Chacha20 implementation, used on arc4random. + Copyright (C) 2022 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + . */ + +#include +#include +#include + +unsigned int __chacha20_sse2_blocks4 (uint32_t *state, uint8_t *dst, + const uint8_t *src, size_t nblks) + attribute_hidden; + +static inline void +chacha20_crypt (uint32_t *state, uint8_t *dst, const uint8_t *src, + size_t bytes) +{ + _Static_assert (CHACHA20_BUFSIZE % 4 == 0, + "CHACHA20_BUFSIZE not multiple of 4"); + _Static_assert (CHACHA20_BUFSIZE >= CHACHA20_BLOCK_SIZE * 4, + "CHACHA20_BUFSIZE <= CHACHA20_BLOCK_SIZE * 4"); + + __chacha20_sse2_blocks4 (state, dst, src, + CHACHA20_BUFSIZE / CHACHA20_BLOCK_SIZE); +} From patchwork Tue Apr 19 21:28:09 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Adhemerval Zanella Netto X-Patchwork-Id: 53053 Return-Path: X-Original-To: patchwork@sourceware.org Delivered-To: patchwork@sourceware.org Received: from server2.sourceware.org (localhost [IPv6:::1]) by sourceware.org (Postfix) with ESMTP id EACB93857836 for ; Tue, 19 Apr 2022 21:33:12 +0000 (GMT) DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org EACB93857836 DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=sourceware.org; s=default; t=1650403993; bh=gERMFarbnIEHboXioVfR8Iy+48hhIqrKJoIIjHCSLlk=; h=To:Subject:Date:In-Reply-To:References:List-Id:List-Unsubscribe: List-Archive:List-Post:List-Help:List-Subscribe:From:Reply-To: From; b=Cg/mmXVvUiTyMNh2lYdCn+DQfV9BKL2tJSyMusdlUxdiKNdfJnh9TMycv558WLqIF iwr0j41I5pUrrJTP9PwfJMi7nxY03PL5VY869AO38DNzqsSxGjfTAJFvTAFlh0aQtk tMyZlliungXyagEQitpUCeJknyUvzbqxLjP8aciQ= X-Original-To: libc-alpha@sourceware.org Delivered-To: libc-alpha@sourceware.org Received: from mail-ot1-x32a.google.com (mail-ot1-x32a.google.com [IPv6:2607:f8b0:4864:20::32a]) by sourceware.org (Postfix) with ESMTPS id 6C7FA3857348 for ; Tue, 19 Apr 2022 21:28:31 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org 6C7FA3857348 Received: by mail-ot1-x32a.google.com with SMTP id i3-20020a056830010300b00605468119c3so6341843otp.11 for ; Tue, 19 Apr 2022 14:28:31 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; h=x-gm-message-state:from:to:subject:date:message-id:in-reply-to :references:mime-version:content-transfer-encoding; bh=gERMFarbnIEHboXioVfR8Iy+48hhIqrKJoIIjHCSLlk=; b=vIqqtplAPencd3JE9a7izNG4fp8uq539irqG0N/LPt/hIt9mAqwTrJZIZ2PpI60elG hKU0u0GwryxWNqOPVUwbk9GdYx+bHZlSXTZeVVYZwupdNR8bHbw4I2eyXw9k/Ddpzmy3 qz0BB8KQ1G059cIEfppJ7MEQ4fCf2FjMz2MrLMkFGFyFitRUd1bN1oA5mCYp1gzXwFRl lO43is2TU3UhFniZOrHHTFew8nNWzNLvk7h5PM2Oil1/6iCJkQYNwaJXHwlOxBmZUmDy oc0um028WClqd3Zgh9IZmSO3R7wPbZplIzqdtJskhZ3N+PfCmNlKNIDoSVeI0pZeEGYQ sLIA== X-Gm-Message-State: AOAM530rlHJ30bBPUcu+phjUU4dceN8beRlQlnGPBjL6RpOukGq7exvz M4SwFfF/aK2sOCdOZ2h4cB5Va2WWMITgvg== X-Google-Smtp-Source: ABdhPJyBV3ND0GDCpKfgd16a/hxi3Jfk5Ev5TolqVr1COCZg+43XXoNR+CFlrA3jdCa903l0krr6Pw== X-Received: by 2002:a05:6830:402c:b0:605:4eed:d8df with SMTP id i12-20020a056830402c00b006054eedd8dfmr3441342ots.319.1650403710201; Tue, 19 Apr 2022 14:28:30 -0700 (PDT) Received: from birita.. ([2804:431:c7ca:c9d0:98f6:7aed:2f61:2745]) by smtp.gmail.com with ESMTPSA id nf9-20020a056871460900b000e2c44ca8edsm5473321oab.6.2022.04.19.14.28.28 for (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Tue, 19 Apr 2022 14:28:29 -0700 (PDT) To: libc-alpha@sourceware.org Subject: [PATCH v3 6/9] x86: Add AVX2 optimized chacha20 Date: Tue, 19 Apr 2022 18:28:09 -0300 Message-Id: <20220419212812.2688764-7-adhemerval.zanella@linaro.org> X-Mailer: git-send-email 2.32.0 In-Reply-To: <20220419212812.2688764-1-adhemerval.zanella@linaro.org> References: <20220419212812.2688764-1-adhemerval.zanella@linaro.org> MIME-Version: 1.0 X-Spam-Status: No, score=-11.4 required=5.0 tests=BAYES_00, DKIM_SIGNED, DKIM_VALID, DKIM_VALID_AU, DKIM_VALID_EF, GIT_PATCH_0, KAM_ASCII_DIVIDERS, KAM_NUMSUBJECT, KAM_SHORT, RCVD_IN_DNSWL_NONE, SCC_5_SHORT_WORD_LINES, SPF_HELO_NONE, SPF_PASS, TXREP, T_SCC_BODY_TEXT_LINE autolearn=ham autolearn_force=no version=3.4.4 X-Spam-Checker-Version: SpamAssassin 3.4.4 (2020-01-24) on server2.sourceware.org X-BeenThere: libc-alpha@sourceware.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Libc-alpha mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-Patchwork-Original-From: Adhemerval Zanella via Libc-alpha From: Adhemerval Zanella Netto Reply-To: Adhemerval Zanella Errors-To: libc-alpha-bounces+patchwork=sourceware.org@sourceware.org Sender: "Libc-alpha" It adds vectorized ChaCha20 implementation based on libgcrypt cipher/chacha20-amd64-avx2.S. It is used only if AVX2 is supported and enabled by the architecture. As for generic implementation, the last step that XOR with the input is omited. On a Ryzen 9 5900X it shows the following improvements (using formatted bench-arc4random data): SSE2: -------------------------------------------------- arc4random [single-thread] 637.06 arc4random_buf(16) [single-thread] 856.62 arc4random_buf(32) [single-thread] 1129.41 arc4random_buf(48) [single-thread] 1260.61 arc4random_buf(64) [single-thread] 1330.56 arc4random_buf(80) [single-thread] 1353.84 arc4random_buf(96) [single-thread] 1376.53 arc4random_buf(112) [single-thread] 1405.74 arc4random_buf(128) [single-thread] 1422.59 -------------------------------------------------- AVX2: Function MB/s -------------------------------------------------- arc4random [single-thread] 809.53 arc4random_buf(16) [single-thread] 1242.56 arc4random_buf(32) [single-thread] 1915.90 arc4random_buf(48) [single-thread] 2230.03 arc4random_buf(64) [single-thread] 2429.68 arc4random_buf(80) [single-thread] 2489.70 arc4random_buf(96) [single-thread] 2598.88 arc4random_buf(112) [single-thread] 2699.93 arc4random_buf(128) [single-thread] 2747.31 Checked on x86_64-linux-gnu. --- LICENSES | 5 +- sysdeps/x86_64/Makefile | 1 + sysdeps/x86_64/chacha20-avx2.S | 313 +++++++++++++++++++++++++++++++++ sysdeps/x86_64/chacha20_arch.h | 22 ++- 4 files changed, 333 insertions(+), 8 deletions(-) create mode 100644 sysdeps/x86_64/chacha20-avx2.S diff --git a/LICENSES b/LICENSES index 415991e208..05a5c07fcf 100644 --- a/LICENSES +++ b/LICENSES @@ -390,8 +390,9 @@ Copyright 2001 by Stephen L. Moshier License along with this library; if not, see . */ -sysdeps/aarch64/chacha20.S and sysdeps/x86_64/chacha20-sse2.S -import code from libgcrypt, with the following notices: +sysdeps/aarch64/chacha20.S, sysdeps/x86_64/chacha20-sse2.S, and +sysdeps/x86_64/chacha20-avx2.S import code from libgcrypt, with the +following notices: Copyright (C) 2017-2019 Jussi Kivilinna diff --git a/sysdeps/x86_64/Makefile b/sysdeps/x86_64/Makefile index c8fbc30857..0fa8897404 100644 --- a/sysdeps/x86_64/Makefile +++ b/sysdeps/x86_64/Makefile @@ -8,6 +8,7 @@ endif ifeq ($(subdir),stdlib) sysdep_routines += \ chacha20-sse2 \ + chacha20-avx2 \ # sysdep_routines endif diff --git a/sysdeps/x86_64/chacha20-avx2.S b/sysdeps/x86_64/chacha20-avx2.S new file mode 100644 index 0000000000..fb76865890 --- /dev/null +++ b/sysdeps/x86_64/chacha20-avx2.S @@ -0,0 +1,313 @@ +/* Optimized AVX2 implementation of ChaCha20 cipher. + Copyright (C) 2022 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + . */ + +#include + +/* Based on D. J. Bernstein reference implementation at + http://cr.yp.to/chacha.html: + + chacha-regs.c version 20080118 + D. J. Bernstein + Public domain. */ + +#ifdef PIC +# define rRIP (%rip) +#else +# define rRIP +#endif + +/* register macros */ +#define INPUT %rdi +#define DST %rsi +#define SRC %rdx +#define NBLKS %rcx +#define ROUND %eax + +/* stack structure */ +#define STACK_VEC_X12 (32) +#define STACK_VEC_X13 (32 + STACK_VEC_X12) +#define STACK_TMP (32 + STACK_VEC_X13) +#define STACK_TMP1 (32 + STACK_TMP) + +#define STACK_MAX (32 + STACK_TMP1) + +/* vector registers */ +#define X0 %ymm0 +#define X1 %ymm1 +#define X2 %ymm2 +#define X3 %ymm3 +#define X4 %ymm4 +#define X5 %ymm5 +#define X6 %ymm6 +#define X7 %ymm7 +#define X8 %ymm8 +#define X9 %ymm9 +#define X10 %ymm10 +#define X11 %ymm11 +#define X12 %ymm12 +#define X13 %ymm13 +#define X14 %ymm14 +#define X15 %ymm15 + +#define X0h %xmm0 +#define X1h %xmm1 +#define X2h %xmm2 +#define X3h %xmm3 +#define X4h %xmm4 +#define X5h %xmm5 +#define X6h %xmm6 +#define X7h %xmm7 +#define X8h %xmm8 +#define X9h %xmm9 +#define X10h %xmm10 +#define X11h %xmm11 +#define X12h %xmm12 +#define X13h %xmm13 +#define X14h %xmm14 +#define X15h %xmm15 + +/********************************************************************** + helper macros + **********************************************************************/ + +/* 4x4 32-bit integer matrix transpose */ +#define transpose_4x4(x0,x1,x2,x3,t1,t2) \ + vpunpckhdq x1, x0, t2; \ + vpunpckldq x1, x0, x0; \ + \ + vpunpckldq x3, x2, t1; \ + vpunpckhdq x3, x2, x2; \ + \ + vpunpckhqdq t1, x0, x1; \ + vpunpcklqdq t1, x0, x0; \ + \ + vpunpckhqdq x2, t2, x3; \ + vpunpcklqdq x2, t2, x2; + +/* 2x2 128-bit matrix transpose */ +#define transpose_16byte_2x2(x0,x1,t1) \ + vmovdqa x0, t1; \ + vperm2i128 $0x20, x1, x0, x0; \ + vperm2i128 $0x31, x1, t1, x1; + +/********************************************************************** + 8-way chacha20 + **********************************************************************/ + +#define ROTATE2(v1,v2,c,tmp) \ + vpsrld $(32 - (c)), v1, tmp; \ + vpslld $(c), v1, v1; \ + vpaddb tmp, v1, v1; \ + vpsrld $(32 - (c)), v2, tmp; \ + vpslld $(c), v2, v2; \ + vpaddb tmp, v2, v2; + +#define ROTATE_SHUF_2(v1,v2,shuf) \ + vpshufb shuf, v1, v1; \ + vpshufb shuf, v2, v2; + +#define XOR(ds,s) \ + vpxor s, ds, ds; + +#define PLUS(ds,s) \ + vpaddd s, ds, ds; + +#define QUARTERROUND2(a1,b1,c1,d1,a2,b2,c2,d2,ign,tmp1,\ + interleave_op1,interleave_op2,\ + interleave_op3,interleave_op4) \ + vbroadcasti128 .Lshuf_rol16 rRIP, tmp1; \ + interleave_op1; \ + PLUS(a1,b1); PLUS(a2,b2); XOR(d1,a1); XOR(d2,a2); \ + ROTATE_SHUF_2(d1, d2, tmp1); \ + interleave_op2; \ + PLUS(c1,d1); PLUS(c2,d2); XOR(b1,c1); XOR(b2,c2); \ + ROTATE2(b1, b2, 12, tmp1); \ + vbroadcasti128 .Lshuf_rol8 rRIP, tmp1; \ + interleave_op3; \ + PLUS(a1,b1); PLUS(a2,b2); XOR(d1,a1); XOR(d2,a2); \ + ROTATE_SHUF_2(d1, d2, tmp1); \ + interleave_op4; \ + PLUS(c1,d1); PLUS(c2,d2); XOR(b1,c1); XOR(b2,c2); \ + ROTATE2(b1, b2, 7, tmp1); + + .section .text.avx2, "ax", @progbits + .align 32 +chacha20_data: +L(shuf_rol16): + .byte 2,3,0,1,6,7,4,5,10,11,8,9,14,15,12,13 +L(shuf_rol8): + .byte 3,0,1,2,7,4,5,6,11,8,9,10,15,12,13,14 +L(inc_counter): + .byte 0,1,2,3,4,5,6,7 +L(unsigned_cmp): + .long 0x80000000 + + .hidden __chacha20_avx2_blocks8 +ENTRY (__chacha20_avx2_blocks8) + /* input: + * %rdi: input + * %rsi: dst + * %rdx: src + * %rcx: nblks (multiple of 8) + */ + vzeroupper; + + pushq %rbp; + cfi_adjust_cfa_offset(8); + cfi_rel_offset(rbp, 0) + movq %rsp, %rbp; + cfi_def_cfa_register(rbp); + + subq $STACK_MAX, %rsp; + andq $~31, %rsp; + +L(loop8): + mov $20, ROUND; + + /* Construct counter vectors X12 and X13 */ + vpmovzxbd L(inc_counter) rRIP, X0; + vpbroadcastd L(unsigned_cmp) rRIP, X2; + vpbroadcastd (12 * 4)(INPUT), X12; + vpbroadcastd (13 * 4)(INPUT), X13; + vpaddd X0, X12, X12; + vpxor X2, X0, X0; + vpxor X2, X12, X1; + vpcmpgtd X1, X0, X0; + vpsubd X0, X13, X13; + vmovdqa X12, (STACK_VEC_X12)(%rsp); + vmovdqa X13, (STACK_VEC_X13)(%rsp); + + /* Load vectors */ + vpbroadcastd (0 * 4)(INPUT), X0; + vpbroadcastd (1 * 4)(INPUT), X1; + vpbroadcastd (2 * 4)(INPUT), X2; + vpbroadcastd (3 * 4)(INPUT), X3; + vpbroadcastd (4 * 4)(INPUT), X4; + vpbroadcastd (5 * 4)(INPUT), X5; + vpbroadcastd (6 * 4)(INPUT), X6; + vpbroadcastd (7 * 4)(INPUT), X7; + vpbroadcastd (8 * 4)(INPUT), X8; + vpbroadcastd (9 * 4)(INPUT), X9; + vpbroadcastd (10 * 4)(INPUT), X10; + vpbroadcastd (11 * 4)(INPUT), X11; + vpbroadcastd (14 * 4)(INPUT), X14; + vpbroadcastd (15 * 4)(INPUT), X15; + vmovdqa X15, (STACK_TMP)(%rsp); + +L(round2): + QUARTERROUND2(X0, X4, X8, X12, X1, X5, X9, X13, tmp:=,X15,,,,) + vmovdqa (STACK_TMP)(%rsp), X15; + vmovdqa X8, (STACK_TMP)(%rsp); + QUARTERROUND2(X2, X6, X10, X14, X3, X7, X11, X15, tmp:=,X8,,,,) + QUARTERROUND2(X0, X5, X10, X15, X1, X6, X11, X12, tmp:=,X8,,,,) + vmovdqa (STACK_TMP)(%rsp), X8; + vmovdqa X15, (STACK_TMP)(%rsp); + QUARTERROUND2(X2, X7, X8, X13, X3, X4, X9, X14, tmp:=,X15,,,,) + sub $2, ROUND; + jnz L(round2); + + vmovdqa X8, (STACK_TMP1)(%rsp); + + /* tmp := X15 */ + vpbroadcastd (0 * 4)(INPUT), X15; + PLUS(X0, X15); + vpbroadcastd (1 * 4)(INPUT), X15; + PLUS(X1, X15); + vpbroadcastd (2 * 4)(INPUT), X15; + PLUS(X2, X15); + vpbroadcastd (3 * 4)(INPUT), X15; + PLUS(X3, X15); + vpbroadcastd (4 * 4)(INPUT), X15; + PLUS(X4, X15); + vpbroadcastd (5 * 4)(INPUT), X15; + PLUS(X5, X15); + vpbroadcastd (6 * 4)(INPUT), X15; + PLUS(X6, X15); + vpbroadcastd (7 * 4)(INPUT), X15; + PLUS(X7, X15); + transpose_4x4(X0, X1, X2, X3, X8, X15); + transpose_4x4(X4, X5, X6, X7, X8, X15); + vmovdqa (STACK_TMP1)(%rsp), X8; + transpose_16byte_2x2(X0, X4, X15); + transpose_16byte_2x2(X1, X5, X15); + transpose_16byte_2x2(X2, X6, X15); + transpose_16byte_2x2(X3, X7, X15); + vmovdqa (STACK_TMP)(%rsp), X15; + vmovdqu X0, (64 * 0 + 16 * 0)(DST) + vmovdqu X1, (64 * 1 + 16 * 0)(DST) + vpbroadcastd (8 * 4)(INPUT), X0; + PLUS(X8, X0); + vpbroadcastd (9 * 4)(INPUT), X0; + PLUS(X9, X0); + vpbroadcastd (10 * 4)(INPUT), X0; + PLUS(X10, X0); + vpbroadcastd (11 * 4)(INPUT), X0; + PLUS(X11, X0); + vmovdqa (STACK_VEC_X12)(%rsp), X0; + PLUS(X12, X0); + vmovdqa (STACK_VEC_X13)(%rsp), X0; + PLUS(X13, X0); + vpbroadcastd (14 * 4)(INPUT), X0; + PLUS(X14, X0); + vpbroadcastd (15 * 4)(INPUT), X0; + PLUS(X15, X0); + vmovdqu X2, (64 * 2 + 16 * 0)(DST) + vmovdqu X3, (64 * 3 + 16 * 0)(DST) + + /* Update counter */ + addq $8, (12 * 4)(INPUT); + + transpose_4x4(X8, X9, X10, X11, X0, X1); + transpose_4x4(X12, X13, X14, X15, X0, X1); + vmovdqu X4, (64 * 4 + 16 * 0)(DST) + vmovdqu X5, (64 * 5 + 16 * 0)(DST) + transpose_16byte_2x2(X8, X12, X0); + transpose_16byte_2x2(X9, X13, X0); + transpose_16byte_2x2(X10, X14, X0); + transpose_16byte_2x2(X11, X15, X0); + vmovdqu X6, (64 * 6 + 16 * 0)(DST) + vmovdqu X7, (64 * 7 + 16 * 0)(DST) + vmovdqu X8, (64 * 0 + 16 * 2)(DST) + vmovdqu X9, (64 * 1 + 16 * 2)(DST) + vmovdqu X10, (64 * 2 + 16 * 2)(DST) + vmovdqu X11, (64 * 3 + 16 * 2)(DST) + vmovdqu X12, (64 * 4 + 16 * 2)(DST) + vmovdqu X13, (64 * 5 + 16 * 2)(DST) + vmovdqu X14, (64 * 6 + 16 * 2)(DST) + vmovdqu X15, (64 * 7 + 16 * 2)(DST) + + sub $8, NBLKS; + lea (8 * 64)(DST), DST; + lea (8 * 64)(SRC), SRC; + jnz L(loop8); + + /* clear the used vector registers and stack */ + vpxor X0, X0, X0; + vmovdqa X0, (STACK_VEC_X12)(%rsp); + vmovdqa X0, (STACK_VEC_X13)(%rsp); + vmovdqa X0, (STACK_TMP)(%rsp); + vmovdqa X0, (STACK_TMP1)(%rsp); + vzeroall; + + /* eax zeroed by round loop. */ + leave; + cfi_adjust_cfa_offset(-8) + cfi_def_cfa_register(%rsp); + ret; + int3; +END(__chacha20_avx2_blocks8) diff --git a/sysdeps/x86_64/chacha20_arch.h b/sysdeps/x86_64/chacha20_arch.h index 5738c840a9..bfdc6c0a36 100644 --- a/sysdeps/x86_64/chacha20_arch.h +++ b/sysdeps/x86_64/chacha20_arch.h @@ -23,16 +23,26 @@ unsigned int __chacha20_sse2_blocks4 (uint32_t *state, uint8_t *dst, const uint8_t *src, size_t nblks) attribute_hidden; +unsigned int __chacha20_avx2_blocks8 (uint32_t *state, uint8_t *dst, + const uint8_t *src, size_t nblks) + attribute_hidden; static inline void chacha20_crypt (uint32_t *state, uint8_t *dst, const uint8_t *src, size_t bytes) { - _Static_assert (CHACHA20_BUFSIZE % 4 == 0, - "CHACHA20_BUFSIZE not multiple of 4"); - _Static_assert (CHACHA20_BUFSIZE >= CHACHA20_BLOCK_SIZE * 4, - "CHACHA20_BUFSIZE <= CHACHA20_BLOCK_SIZE * 4"); + _Static_assert (CHACHA20_BUFSIZE % 4 == 0 && CHACHA20_BUFSIZE % 8 == 0, + "CHACHA20_BUFSIZE not multiple of 4 or 8"); + _Static_assert (CHACHA20_BUFSIZE >= CHACHA20_BLOCK_SIZE * 8, + "CHACHA20_BUFSIZE < CHACHA20_BLOCK_SIZE * 8"); + const struct cpu_features* cpu_features = __get_cpu_features (); - __chacha20_sse2_blocks4 (state, dst, src, - CHACHA20_BUFSIZE / CHACHA20_BLOCK_SIZE); + /* AVX2 version uses vzeroupper, so disable it if RTM is enabled. */ + if (CPU_FEATURE_USABLE_P (cpu_features, AVX2) + && !CPU_FEATURES_ARCH_P (cpu_features, Prefer_No_VZEROUPPER)) + __chacha20_avx2_blocks8 (state, dst, src, + CHACHA20_BUFSIZE / CHACHA20_BLOCK_SIZE); + else + __chacha20_sse2_blocks4 (state, dst, src, + CHACHA20_BUFSIZE / CHACHA20_BLOCK_SIZE); } From patchwork Tue Apr 19 21:28:10 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Adhemerval Zanella Netto X-Patchwork-Id: 53054 Return-Path: X-Original-To: patchwork@sourceware.org Delivered-To: patchwork@sourceware.org Received: from server2.sourceware.org (localhost [IPv6:::1]) by sourceware.org (Postfix) with ESMTP id 44E8F385734B for ; Tue, 19 Apr 2022 21:34:00 +0000 (GMT) DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org 44E8F385734B DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=sourceware.org; s=default; t=1650404040; bh=dEkNzUzhF4enCUuY6OAe2zUAQ4qAL9Ni4/Up6Yv6kPc=; h=To:Subject:Date:In-Reply-To:References:List-Id:List-Unsubscribe: List-Archive:List-Post:List-Help:List-Subscribe:From:Reply-To: From; b=Fa705w4cefeS8yXZx3SPaRBkWzAUOGxOuHANdlNpkGKX8J62tseOxCLp4HC2QtfRz ZdREa3l70qpozu/30ryjTZ5dXWsnSJE9pe4Xw0d5c2sFhcVQ3yr60NKS2h3qLv38nW Y6lBZVnOdNS6c1RJVzx4f5kGCfxFcvCcsfoKXklI= X-Original-To: libc-alpha@sourceware.org Delivered-To: libc-alpha@sourceware.org Received: from mail-ot1-x32c.google.com (mail-ot1-x32c.google.com [IPv6:2607:f8b0:4864:20::32c]) by sourceware.org (Postfix) with ESMTPS id D8DC23857341 for ; Tue, 19 Apr 2022 21:28:32 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org D8DC23857341 Received: by mail-ot1-x32c.google.com with SMTP id i3-20020a056830010300b00605468119c3so6341878otp.11 for ; Tue, 19 Apr 2022 14:28:32 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; h=x-gm-message-state:from:to:subject:date:message-id:in-reply-to :references:mime-version:content-transfer-encoding; bh=dEkNzUzhF4enCUuY6OAe2zUAQ4qAL9Ni4/Up6Yv6kPc=; b=VcJcvTMjIfgTQlm3z9YuYqcxxlOFNCdbB/xXs6qrKb9gicxFvaZC5pG8WbejUINoi8 w1/nfy6/XgXcS4Uz2g/Cw4ASMiKB/FPtBxNSyTxaci50greW/umnvrnlPM6VHP6n0i6x ml1alj/UNwtqXJBux3DQZsRltnCMppQa11azJBTO8m/T8MlabzgmCrqb+nirVFPzb3ik dpUEkOGgBS9gRa80e4bYV5S42uy/P5NLnwKPBccKOJLnlhW6xD6dJ2QfqFCq73c6FvLi 3fTz44p7mpjU2Lwf5Munp4a5xwtfiRs+tJB1z6+RgeGJeARYHHsL6GWcq145xa/yJF4X 3HRA== X-Gm-Message-State: AOAM531LuA4wlZ3hd1qPtbfcRNwnTb+8aO/H78zqqLolDaaRDcM247MV mljpVh6oLDJ/+O8WiozXA7b8cDRQEPr6Ag== X-Google-Smtp-Source: ABdhPJzJSXWERX3oSzfNJRnYlwpzfMumIQDnV66jPHJm/61ASILen+pB4SN8CVdFMUAbeg5Mna2XLQ== X-Received: by 2002:a05:6830:1012:b0:605:4fd8:c849 with SMTP id a18-20020a056830101200b006054fd8c849mr3182431otp.73.1650403711715; Tue, 19 Apr 2022 14:28:31 -0700 (PDT) Received: from birita.. ([2804:431:c7ca:c9d0:98f6:7aed:2f61:2745]) by smtp.gmail.com with ESMTPSA id nf9-20020a056871460900b000e2c44ca8edsm5473321oab.6.2022.04.19.14.28.30 for (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Tue, 19 Apr 2022 14:28:31 -0700 (PDT) To: libc-alpha@sourceware.org Subject: [PATCH v3 7/9] powerpc64: Add optimized chacha20 Date: Tue, 19 Apr 2022 18:28:10 -0300 Message-Id: <20220419212812.2688764-8-adhemerval.zanella@linaro.org> X-Mailer: git-send-email 2.32.0 In-Reply-To: <20220419212812.2688764-1-adhemerval.zanella@linaro.org> References: <20220419212812.2688764-1-adhemerval.zanella@linaro.org> MIME-Version: 1.0 X-Spam-Status: No, score=-11.4 required=5.0 tests=BAYES_00, DKIM_SIGNED, DKIM_VALID, DKIM_VALID_AU, DKIM_VALID_EF, GIT_PATCH_0, KAM_ASCII_DIVIDERS, KAM_NUMSUBJECT, KAM_SHORT, RCVD_IN_DNSWL_NONE, SCC_5_SHORT_WORD_LINES, SPF_HELO_NONE, SPF_PASS, TXREP, T_SCC_BODY_TEXT_LINE autolearn=ham autolearn_force=no version=3.4.4 X-Spam-Checker-Version: SpamAssassin 3.4.4 (2020-01-24) on server2.sourceware.org X-BeenThere: libc-alpha@sourceware.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Libc-alpha mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-Patchwork-Original-From: Adhemerval Zanella via Libc-alpha From: Adhemerval Zanella Netto Reply-To: Adhemerval Zanella Errors-To: libc-alpha-bounces+patchwork=sourceware.org@sourceware.org Sender: "Libc-alpha" It adds vectorized ChaCha20 implementation based on libgcrypt cipher/chacha20-ppc.c. It targets POWER8 and it is used on default for LE. On a POWER8 it shows the following improvements (using formatted bench-arc4random data): GENERIC (powerpc64-linux-gnu) Function MB/s -------------------------------------------------- arc4random [single-thread] 71.08 arc4random_buf(16) [single-thread] 141.26 arc4random_buf(32) [single-thread] 198.31 arc4random_buf(48) [single-thread] 226.78 arc4random_buf(64) [single-thread] 246.69 arc4random_buf(80) [single-thread] 257.23 arc4random_buf(96) [single-thread] 268.06 arc4random_buf(112) [single-thread] 274.50 arc4random_buf(128) [single-thread] 279.56 -------------------------------------------------- POWER8 Function MB/s -------------------------------------------------- arc4random [single-thread] 84.68 arc4random_buf(16) [single-thread] 210.75 arc4random_buf(32) [single-thread] 366.11 arc4random_buf(48) [single-thread] 471.99 arc4random_buf(64) [single-thread] 567.06 arc4random_buf(80) [single-thread] 633.79 arc4random_buf(96) [single-thread] 693.16 arc4random_buf(112) [single-thread] 737.77 arc4random_buf(128) [single-thread] 774.38 -------------------------------------------------- Checked on powerpc64-linux-gnu and powerpc64le-linux-gnu. --- LICENSES | 6 +- sysdeps/powerpc/powerpc64/Makefile | 3 + sysdeps/powerpc/powerpc64/chacha20-ppc.c | 236 ++++++++++++++++++++++ sysdeps/powerpc/powerpc64/chacha20_arch.h | 47 +++++ 4 files changed, 289 insertions(+), 3 deletions(-) create mode 100644 sysdeps/powerpc/powerpc64/chacha20-ppc.c create mode 100644 sysdeps/powerpc/powerpc64/chacha20_arch.h diff --git a/LICENSES b/LICENSES index 05a5c07fcf..1c6c5d73e6 100644 --- a/LICENSES +++ b/LICENSES @@ -390,9 +390,9 @@ Copyright 2001 by Stephen L. Moshier License along with this library; if not, see . */ -sysdeps/aarch64/chacha20.S, sysdeps/x86_64/chacha20-sse2.S, and -sysdeps/x86_64/chacha20-avx2.S import code from libgcrypt, with the -following notices: +sysdeps/aarch64/chacha20.S, sysdeps/x86_64/chacha20-sse2.S, +sysdeps/x86_64/chacha20-avx2.S, and sysdeps/powerpc/powerpc64/chacha20-ppc.c +import code from libgcrypt, with the following notices: Copyright (C) 2017-2019 Jussi Kivilinna diff --git a/sysdeps/powerpc/powerpc64/Makefile b/sysdeps/powerpc/powerpc64/Makefile index 679d5e49ba..18943ef09e 100644 --- a/sysdeps/powerpc/powerpc64/Makefile +++ b/sysdeps/powerpc/powerpc64/Makefile @@ -66,6 +66,9 @@ tst-setjmp-bug21895-static-ENV = \ endif ifeq ($(subdir),stdlib) +sysdep_routines += chacha20-ppc +CFLAGS-chacha20-ppc.c += -mcpu=power8 + CFLAGS-tst-ucontext-ppc64-vscr.c += -maltivec tests += tst-ucontext-ppc64-vscr endif diff --git a/sysdeps/powerpc/powerpc64/chacha20-ppc.c b/sysdeps/powerpc/powerpc64/chacha20-ppc.c new file mode 100644 index 0000000000..e2567c379a --- /dev/null +++ b/sysdeps/powerpc/powerpc64/chacha20-ppc.c @@ -0,0 +1,236 @@ +/* Optimized PowerPC implementation of ChaCha20 cipher. + Copyright (C) 2022 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + . */ + +#include +#include +#include +#include +#include + +typedef vector unsigned char vector16x_u8; +typedef vector unsigned int vector4x_u32; +typedef vector unsigned long long vector2x_u64; + +#if __BYTE_ORDER == __BIG_ENDIAN +static const vector16x_u8 le_bswap_const = + { 3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12 }; +#endif + +static inline vector4x_u32 +vec_rol_elems (vector4x_u32 v, unsigned int idx) +{ +#if __BYTE_ORDER != __BIG_ENDIAN + return vec_sld (v, v, (16 - (4 * idx)) & 15); +#else + return vec_sld (v, v, (4 * idx) & 15); +#endif +} + +static inline vector4x_u32 +vec_load_le (unsigned long offset, const unsigned char *ptr) +{ + vector4x_u32 vec; + vec = vec_vsx_ld (offset, (const uint32_t *)ptr); +#if __BYTE_ORDER == __BIG_ENDIAN + vec = (vector4x_u32) vec_perm ((vector16x_u8)vec, (vector16x_u8)vec, + le_bswap_const); +#endif + return vec; +} + +static inline void +vec_store_le (vector4x_u32 vec, unsigned long offset, unsigned char *ptr) +{ +#if __BYTE_ORDER == __BIG_ENDIAN + vec = (vector4x_u32)vec_perm((vector16x_u8)vec, (vector16x_u8)vec, + le_bswap_const); +#endif + vec_vsx_st (vec, offset, (uint32_t *)ptr); +} + + +static inline vector4x_u32 +vec_add_ctr_u64 (vector4x_u32 v, vector4x_u32 a) +{ +#if __BYTE_ORDER == __BIG_ENDIAN + static const vector16x_u8 swap32 = + { 4, 5, 6, 7, 0, 1, 2, 3, 12, 13, 14, 15, 8, 9, 10, 11 }; + vector2x_u64 vec, add, sum; + + vec = (vector2x_u64)vec_perm ((vector16x_u8)v, (vector16x_u8)v, swap32); + add = (vector2x_u64)vec_perm ((vector16x_u8)a, (vector16x_u8)a, swap32); + sum = vec + add; + return (vector4x_u32)vec_perm ((vector16x_u8)sum, (vector16x_u8)sum, swap32); +#else + return (vector4x_u32)((vector2x_u64)(v) + (vector2x_u64)(a)); +#endif +} + +/********************************************************************** + 4-way chacha20 + **********************************************************************/ + +#define ROTATE(v1,rolv) \ + __asm__ ("vrlw %0,%1,%2\n\t" : "=v" (v1) : "v" (v1), "v" (rolv)) + +#define PLUS(ds,s) \ + ((ds) += (s)) + +#define XOR(ds,s) \ + ((ds) ^= (s)) + +#define ADD_U64(v,a) \ + (v = vec_add_ctr_u64(v, a)) + +/* 4x4 32-bit integer matrix transpose */ +#define transpose_4x4(x0, x1, x2, x3) ({ \ + vector4x_u32 t1 = vec_mergeh(x0, x2); \ + vector4x_u32 t2 = vec_mergel(x0, x2); \ + vector4x_u32 t3 = vec_mergeh(x1, x3); \ + x3 = vec_mergel(x1, x3); \ + x0 = vec_mergeh(t1, t3); \ + x1 = vec_mergel(t1, t3); \ + x2 = vec_mergeh(t2, x3); \ + x3 = vec_mergel(t2, x3); \ + }) + +#define QUARTERROUND2(a1,b1,c1,d1,a2,b2,c2,d2) \ + PLUS(a1,b1); PLUS(a2,b2); XOR(d1,a1); XOR(d2,a2); \ + ROTATE(d1, rotate_16); ROTATE(d2, rotate_16); \ + PLUS(c1,d1); PLUS(c2,d2); XOR(b1,c1); XOR(b2,c2); \ + ROTATE(b1, rotate_12); ROTATE(b2, rotate_12); \ + PLUS(a1,b1); PLUS(a2,b2); XOR(d1,a1); XOR(d2,a2); \ + ROTATE(d1, rotate_8); ROTATE(d2, rotate_8); \ + PLUS(c1,d1); PLUS(c2,d2); XOR(b1,c1); XOR(b2,c2); \ + ROTATE(b1, rotate_7); ROTATE(b2, rotate_7); + +unsigned int attribute_hidden +__chacha20_power8_blocks4 (uint32_t *state, uint8_t *dst, const uint8_t *src, + size_t nblks) +{ + vector4x_u32 counters_0123 = { 0, 1, 2, 3 }; + vector4x_u32 counter_4 = { 4, 0, 0, 0 }; + vector4x_u32 rotate_16 = { 16, 16, 16, 16 }; + vector4x_u32 rotate_12 = { 12, 12, 12, 12 }; + vector4x_u32 rotate_8 = { 8, 8, 8, 8 }; + vector4x_u32 rotate_7 = { 7, 7, 7, 7 }; + vector4x_u32 state0, state1, state2, state3; + vector4x_u32 v0, v1, v2, v3, v4, v5, v6, v7; + vector4x_u32 v8, v9, v10, v11, v12, v13, v14, v15; + vector4x_u32 tmp; + int i; + + /* Force preload of constants to vector registers. */ + __asm__ ("": "+v" (counters_0123) :: "memory"); + __asm__ ("": "+v" (counter_4) :: "memory"); + __asm__ ("": "+v" (rotate_16) :: "memory"); + __asm__ ("": "+v" (rotate_12) :: "memory"); + __asm__ ("": "+v" (rotate_8) :: "memory"); + __asm__ ("": "+v" (rotate_7) :: "memory"); + + state0 = vec_vsx_ld (0 * 16, state); + state1 = vec_vsx_ld (1 * 16, state); + state2 = vec_vsx_ld (2 * 16, state); + state3 = vec_vsx_ld (3 * 16, state); + + do + { + v0 = vec_splat (state0, 0); + v1 = vec_splat (state0, 1); + v2 = vec_splat (state0, 2); + v3 = vec_splat (state0, 3); + v4 = vec_splat (state1, 0); + v5 = vec_splat (state1, 1); + v6 = vec_splat (state1, 2); + v7 = vec_splat (state1, 3); + v8 = vec_splat (state2, 0); + v9 = vec_splat (state2, 1); + v10 = vec_splat (state2, 2); + v11 = vec_splat (state2, 3); + v12 = vec_splat (state3, 0); + v13 = vec_splat (state3, 1); + v14 = vec_splat (state3, 2); + v15 = vec_splat (state3, 3); + + v12 += counters_0123; + v13 -= vec_cmplt (v12, counters_0123); + + for (i = 20; i > 0; i -= 2) + { + QUARTERROUND2 (v0, v4, v8, v12, v1, v5, v9, v13) + QUARTERROUND2 (v2, v6, v10, v14, v3, v7, v11, v15) + QUARTERROUND2 (v0, v5, v10, v15, v1, v6, v11, v12) + QUARTERROUND2 (v2, v7, v8, v13, v3, v4, v9, v14) + } + + v0 += vec_splat (state0, 0); + v1 += vec_splat (state0, 1); + v2 += vec_splat (state0, 2); + v3 += vec_splat (state0, 3); + v4 += vec_splat (state1, 0); + v5 += vec_splat (state1, 1); + v6 += vec_splat (state1, 2); + v7 += vec_splat (state1, 3); + v8 += vec_splat (state2, 0); + v9 += vec_splat (state2, 1); + v10 += vec_splat (state2, 2); + v11 += vec_splat (state2, 3); + tmp = vec_splat( state3, 0); + tmp += counters_0123; + v12 += tmp; + v13 += vec_splat (state3, 1) - vec_cmplt (tmp, counters_0123); + v14 += vec_splat (state3, 2); + v15 += vec_splat (state3, 3); + ADD_U64 (state3, counter_4); + + transpose_4x4 (v0, v1, v2, v3); + transpose_4x4 (v4, v5, v6, v7); + transpose_4x4 (v8, v9, v10, v11); + transpose_4x4 (v12, v13, v14, v15); + + vec_store_le (v0, (64 * 0 + 16 * 0), dst); + vec_store_le (v1, (64 * 1 + 16 * 0), dst); + vec_store_le (v2, (64 * 2 + 16 * 0), dst); + vec_store_le (v3, (64 * 3 + 16 * 0), dst); + + vec_store_le (v4, (64 * 0 + 16 * 1), dst); + vec_store_le (v5, (64 * 1 + 16 * 1), dst); + vec_store_le (v6, (64 * 2 + 16 * 1), dst); + vec_store_le (v7, (64 * 3 + 16 * 1), dst); + + vec_store_le (v8, (64 * 0 + 16 * 2), dst); + vec_store_le (v9, (64 * 1 + 16 * 2), dst); + vec_store_le (v10, (64 * 2 + 16 * 2), dst); + vec_store_le (v11, (64 * 3 + 16 * 2), dst); + + vec_store_le (v12, (64 * 0 + 16 * 3), dst); + vec_store_le (v13, (64 * 1 + 16 * 3), dst); + vec_store_le (v14, (64 * 2 + 16 * 3), dst); + vec_store_le (v15, (64 * 3 + 16 * 3), dst); + + src += 4*64; + dst += 4*64; + + nblks -= 4; + } + while (nblks); + + vec_vsx_st (state3, 3 * 16, state); + + return 0; +} diff --git a/sysdeps/powerpc/powerpc64/chacha20_arch.h b/sysdeps/powerpc/powerpc64/chacha20_arch.h new file mode 100644 index 0000000000..a18115392f --- /dev/null +++ b/sysdeps/powerpc/powerpc64/chacha20_arch.h @@ -0,0 +1,47 @@ +/* PowerPC optimization for ChaCha20. + Copyright (C) 2022 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + . */ + +#include +#include + +unsigned int __chacha20_power8_blocks4 (uint32_t *state, uint8_t *dst, + const uint8_t *src, size_t nblks) + attribute_hidden; + +static void +chacha20_crypt (uint32_t *state, uint8_t *dst, + const uint8_t *src, size_t bytes) +{ + _Static_assert (CHACHA20_BUFSIZE % 4 == 0, + "CHACHA20_BUFSIZE not multiple of 4"); + _Static_assert (CHACHA20_BUFSIZE >= CHACHA20_BLOCK_SIZE * 4, + "CHACHA20_BUFSIZE < CHACHA20_BLOCK_SIZE * 4"); + +#ifdef __LITTLE_ENDIAN__ + __chacha20_power8_blocks4 (state, dst, src, + CHACHA20_BUFSIZE / CHACHA20_BLOCK_SIZE); +#else + unsigned long int hwcap = GLRO(dl_hwcap); + unsigned long int hwcap2 = GLRO(dl_hwcap2); + if (hwcap2 & PPC_FEATURE2_ARCH_2_07 && hwcap & PPC_FEATURE_HAS_ALTIVEC) + __chacha20_power8_blocks4 (state, dst, src, + CHACHA20_BUFSIZE / CHACHA20_BLOCK_SIZE); + else + chacha20_crypt_generic (state, dst, src, bytes); +#endif +} From patchwork Tue Apr 19 21:28:11 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Adhemerval Zanella Netto X-Patchwork-Id: 53056 Return-Path: X-Original-To: patchwork@sourceware.org Delivered-To: patchwork@sourceware.org Received: from server2.sourceware.org (localhost [IPv6:::1]) by sourceware.org (Postfix) with ESMTP id C222E3857C50 for ; Tue, 19 Apr 2022 21:35:30 +0000 (GMT) DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org C222E3857C50 DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=sourceware.org; s=default; t=1650404130; bh=waO4GmtpIo8oQXWSP+2yhbSXWcGXtrF3UKV6pMhhu84=; h=To:Subject:Date:In-Reply-To:References:List-Id:List-Unsubscribe: List-Archive:List-Post:List-Help:List-Subscribe:From:Reply-To: From; b=SpkQodVO+2bICN62ZMxqDEd2L7MfV2IgLZqychLVnoRMT5JVRERqK3xaieCe1lIHB aFujGzemO4h8ZpPv55O0LT4liB3Ztrpbfhhuv69mKQFIUJaeBG2mA2SEhIh5iWTJ9V HCF7zAVc0U9K7dfPBtghu0keWaNmanEMV4mbAoWc= X-Original-To: libc-alpha@sourceware.org Delivered-To: libc-alpha@sourceware.org Received: from mail-oi1-x231.google.com (mail-oi1-x231.google.com [IPv6:2607:f8b0:4864:20::231]) by sourceware.org (Postfix) with ESMTPS id EE17F3857836 for ; Tue, 19 Apr 2022 21:28:35 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org EE17F3857836 Received: by mail-oi1-x231.google.com with SMTP id s16so50344oie.0 for ; Tue, 19 Apr 2022 14:28:35 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; h=x-gm-message-state:from:to:subject:date:message-id:in-reply-to :references:mime-version:content-transfer-encoding; bh=waO4GmtpIo8oQXWSP+2yhbSXWcGXtrF3UKV6pMhhu84=; b=wuVdmgfdd5bzNVAfP00Wgu8pJPuzG9YuXjKC9c1URodUAgQgcVMT2AB452ElzxW6z7 ll89OxMElkvmDiXTZXVWIwfTNC061t3JA44eodNwBG7du7BOQQVkg9vZX/04EddYSYcU MG6j/rBrUvCtbrK5f+ZrlBaSf3ltOops4wLpIUusqn1keUDhOjcEqJdW/UtxRdR+s30B +OfxibonrToyyNTl/Mn2EBivpefS+xAE4dbSks8Ad9s+uiMMQEZ/ZNPRgjmN4iEf4HNw EV0K2maCOBy7q9jbbMQXS3fNhKDUaDxwZeQG/A4N5nD/ybHGysC8H/L8jf1ed1McufF+ /TYw== X-Gm-Message-State: AOAM5306KNFFnaNk7/I+Spe5sVXpB6eppPR7UPzZf55AyJRCyy9oBYrP 8787GwCUVKnqOxDZ0gO2nqrn49buR1QEOQ== X-Google-Smtp-Source: ABdhPJzhDkQUvQv7RZ4Yjr/CWUFwh9vZqGqZqhilgF65+72RkyLgsT8OX1hLsWBWhxX/7P1Eck0S5A== X-Received: by 2002:a05:6808:138e:b0:322:2dcd:dc7d with SMTP id c14-20020a056808138e00b003222dcddc7dmr264153oiw.159.1650403713171; Tue, 19 Apr 2022 14:28:33 -0700 (PDT) Received: from birita.. ([2804:431:c7ca:c9d0:98f6:7aed:2f61:2745]) by smtp.gmail.com with ESMTPSA id nf9-20020a056871460900b000e2c44ca8edsm5473321oab.6.2022.04.19.14.28.31 for (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Tue, 19 Apr 2022 14:28:32 -0700 (PDT) To: libc-alpha@sourceware.org Subject: [PATCH v3 8/9] s390x: Add optimized chacha20 Date: Tue, 19 Apr 2022 18:28:11 -0300 Message-Id: <20220419212812.2688764-9-adhemerval.zanella@linaro.org> X-Mailer: git-send-email 2.32.0 In-Reply-To: <20220419212812.2688764-1-adhemerval.zanella@linaro.org> References: <20220419212812.2688764-1-adhemerval.zanella@linaro.org> MIME-Version: 1.0 X-Spam-Status: No, score=-10.9 required=5.0 tests=BAYES_00, DKIM_SIGNED, DKIM_VALID, DKIM_VALID_AU, DKIM_VALID_EF, GIT_PATCH_0, KAM_ASCII_DIVIDERS, KAM_NUMSUBJECT, KAM_SHORT, RCVD_IN_DNSWL_NONE, SCC_10_SHORT_WORD_LINES, SCC_5_SHORT_WORD_LINES, SPF_HELO_NONE, SPF_PASS, TXREP, T_SCC_BODY_TEXT_LINE autolearn=ham autolearn_force=no version=3.4.4 X-Spam-Checker-Version: SpamAssassin 3.4.4 (2020-01-24) on server2.sourceware.org X-BeenThere: libc-alpha@sourceware.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Libc-alpha mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-Patchwork-Original-From: Adhemerval Zanella via Libc-alpha From: Adhemerval Zanella Netto Reply-To: Adhemerval Zanella Errors-To: libc-alpha-bounces+patchwork=sourceware.org@sourceware.org Sender: "Libc-alpha" It adds vectorized ChaCha20 implementation based on libgcrypt cipher/chacha20-s390x.S. On a z15 it shows the following improvements (using formatted bench-arc4random data): GENERIC Function MB/s -------------------------------------------------- arc4random [single-thread] 150.27 arc4random_buf(16) [single-thread] 239.11 arc4random_buf(32) [single-thread] 268.36 arc4random_buf(48) [single-thread] 311.75 arc4random_buf(64) [single-thread] 300.16 arc4random_buf(80) [single-thread] 329.82 arc4random_buf(96) [single-thread] 305.98 arc4random_buf(112) [single-thread] 334.56 arc4random_buf(128) [single-thread] 313.29 -------------------------------------------------- VX Function MB/s -------------------------------------------------- arc4random [single-thread] 239.59 arc4random_buf(16) [single-thread] 552.42 arc4random_buf(32) [single-thread] 854.64 arc4random_buf(48) [single-thread] 1048.83 arc4random_buf(64) [single-thread] 1184.54 arc4random_buf(80) [single-thread] 1282.09 arc4random_buf(96) [single-thread] 1363.40 arc4random_buf(112) [single-thread] 1420.99 arc4random_buf(128) [single-thread] 1464.38 -------------------------------------------------- Checked on s390x-linux-gnu. --- LICENSES | 5 +- sysdeps/s390/s390-64/Makefile | 4 + sysdeps/s390/s390-64/chacha20-vx.S | 564 +++++++++++++++++++++++++++ sysdeps/s390/s390-64/chacha20_arch.h | 45 +++ 4 files changed, 616 insertions(+), 2 deletions(-) create mode 100644 sysdeps/s390/s390-64/chacha20-vx.S create mode 100644 sysdeps/s390/s390-64/chacha20_arch.h diff --git a/LICENSES b/LICENSES index 1c6c5d73e6..12485408c7 100644 --- a/LICENSES +++ b/LICENSES @@ -391,8 +391,9 @@ Copyright 2001 by Stephen L. Moshier . */ sysdeps/aarch64/chacha20.S, sysdeps/x86_64/chacha20-sse2.S, -sysdeps/x86_64/chacha20-avx2.S, and sysdeps/powerpc/powerpc64/chacha20-ppc.c -import code from libgcrypt, with the following notices: +sysdeps/x86_64/chacha20-avx2.S, sysdeps/powerpc/powerpc64/chacha20-ppc.c, +and sysdeps/s390/s390-64/chacha20-vx.S import code from libgcrypt, with +the following notices: Copyright (C) 2017-2019 Jussi Kivilinna diff --git a/sysdeps/s390/s390-64/Makefile b/sysdeps/s390/s390-64/Makefile index 66ed844e68..5c50fb117e 100644 --- a/sysdeps/s390/s390-64/Makefile +++ b/sysdeps/s390/s390-64/Makefile @@ -67,3 +67,7 @@ tests-container += tst-glibc-hwcaps-cache endif endif # $(subdir) == elf + +ifeq ($(subdir),stdlib) +sysdep_routines += chacha20-vx +endif diff --git a/sysdeps/s390/s390-64/chacha20-vx.S b/sysdeps/s390/s390-64/chacha20-vx.S new file mode 100644 index 0000000000..5123e3c064 --- /dev/null +++ b/sysdeps/s390/s390-64/chacha20-vx.S @@ -0,0 +1,564 @@ +/* Optimized s390x implementation of ChaCha20 cipher. + Copyright (C) 2022 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + . */ + +#include + +#ifdef HAVE_S390_VX_ASM_SUPPORT + +/* CFA expressions are used for pointing CFA and registers to + * SP relative offsets. */ +# define DW_REGNO_SP 15 + +/* Fixed length encoding used for integers for now. */ +# define DW_SLEB128_7BIT(value) \ + 0x00|((value) & 0x7f) +# define DW_SLEB128_28BIT(value) \ + 0x80|((value)&0x7f), \ + 0x80|(((value)>>7)&0x7f), \ + 0x80|(((value)>>14)&0x7f), \ + 0x00|(((value)>>21)&0x7f) + +# define cfi_cfa_on_stack(rsp_offs,cfa_depth) \ + .cfi_escape \ + 0x0f, /* DW_CFA_def_cfa_expression */ \ + DW_SLEB128_7BIT(11), /* length */ \ + 0x7f, /* DW_OP_breg15, rsp + constant */ \ + DW_SLEB128_28BIT(rsp_offs), \ + 0x06, /* DW_OP_deref */ \ + 0x23, /* DW_OP_plus_constu */ \ + DW_SLEB128_28BIT((cfa_depth)+160) + +.machine "z13+vx" +.text + +.balign 16 +.Lconsts: +.Lwordswap: + .byte 12, 13, 14, 15, 8, 9, 10, 11, 4, 5, 6, 7, 0, 1, 2, 3 +.Lbswap128: + .byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0 +.Lbswap32: + .byte 3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12 +.Lone: + .long 0, 0, 0, 1 +.Ladd_counter_0123: + .long 0, 1, 2, 3 +.Ladd_counter_4567: + .long 4, 5, 6, 7 + +/* register macros */ +#define INPUT %r2 +#define DST %r3 +#define SRC %r4 +#define NBLKS %r0 +#define ROUND %r1 + +/* stack structure */ + +#define STACK_FRAME_STD (8 * 16 + 8 * 4) +#define STACK_FRAME_F8_F15 (8 * 8) +#define STACK_FRAME_Y0_Y15 (16 * 16) +#define STACK_FRAME_CTR (4 * 16) +#define STACK_FRAME_PARAMS (6 * 8) + +#define STACK_MAX (STACK_FRAME_STD + STACK_FRAME_F8_F15 + \ + STACK_FRAME_Y0_Y15 + STACK_FRAME_CTR + \ + STACK_FRAME_PARAMS) + +#define STACK_F8 (STACK_MAX - STACK_FRAME_F8_F15) +#define STACK_F9 (STACK_F8 + 8) +#define STACK_F10 (STACK_F9 + 8) +#define STACK_F11 (STACK_F10 + 8) +#define STACK_F12 (STACK_F11 + 8) +#define STACK_F13 (STACK_F12 + 8) +#define STACK_F14 (STACK_F13 + 8) +#define STACK_F15 (STACK_F14 + 8) +#define STACK_Y0_Y15 (STACK_F8 - STACK_FRAME_Y0_Y15) +#define STACK_CTR (STACK_Y0_Y15 - STACK_FRAME_CTR) +#define STACK_INPUT (STACK_CTR - STACK_FRAME_PARAMS) +#define STACK_DST (STACK_INPUT + 8) +#define STACK_SRC (STACK_DST + 8) +#define STACK_NBLKS (STACK_SRC + 8) +#define STACK_POCTX (STACK_NBLKS + 8) +#define STACK_POSRC (STACK_POCTX + 8) + +#define STACK_G0_H3 STACK_Y0_Y15 + +/* vector registers */ +#define A0 %v0 +#define A1 %v1 +#define A2 %v2 +#define A3 %v3 + +#define B0 %v4 +#define B1 %v5 +#define B2 %v6 +#define B3 %v7 + +#define C0 %v8 +#define C1 %v9 +#define C2 %v10 +#define C3 %v11 + +#define D0 %v12 +#define D1 %v13 +#define D2 %v14 +#define D3 %v15 + +#define E0 %v16 +#define E1 %v17 +#define E2 %v18 +#define E3 %v19 + +#define F0 %v20 +#define F1 %v21 +#define F2 %v22 +#define F3 %v23 + +#define G0 %v24 +#define G1 %v25 +#define G2 %v26 +#define G3 %v27 + +#define H0 %v28 +#define H1 %v29 +#define H2 %v30 +#define H3 %v31 + +#define IO0 E0 +#define IO1 E1 +#define IO2 E2 +#define IO3 E3 +#define IO4 F0 +#define IO5 F1 +#define IO6 F2 +#define IO7 F3 + +#define S0 G0 +#define S1 G1 +#define S2 G2 +#define S3 G3 + +#define TMP0 H0 +#define TMP1 H1 +#define TMP2 H2 +#define TMP3 H3 + +#define X0 A0 +#define X1 A1 +#define X2 A2 +#define X3 A3 +#define X4 B0 +#define X5 B1 +#define X6 B2 +#define X7 B3 +#define X8 C0 +#define X9 C1 +#define X10 C2 +#define X11 C3 +#define X12 D0 +#define X13 D1 +#define X14 D2 +#define X15 D3 + +#define Y0 E0 +#define Y1 E1 +#define Y2 E2 +#define Y3 E3 +#define Y4 F0 +#define Y5 F1 +#define Y6 F2 +#define Y7 F3 +#define Y8 G0 +#define Y9 G1 +#define Y10 G2 +#define Y11 G3 +#define Y12 H0 +#define Y13 H1 +#define Y14 H2 +#define Y15 H3 + +/********************************************************************** + helper macros + **********************************************************************/ + +#define _ /*_*/ + +#define CLEAR(x,...) vzero x; + +#define START_STACK(last_r) \ + lgr %r0, %r15; \ + lghi %r1, ~15; \ + stmg %r6, last_r, 6 * 8(%r15); \ + aghi %r0, -STACK_MAX; \ + ngr %r0, %r1; \ + lgr %r1, %r15; \ + cfi_def_cfa_register(1); \ + lgr %r15, %r0; \ + stg %r1, 0(%r15); \ + cfi_cfa_on_stack(0, 0); \ + std %f8, STACK_F8(%r15); \ + std %f9, STACK_F9(%r15); \ + std %f10, STACK_F10(%r15); \ + std %f11, STACK_F11(%r15); \ + std %f12, STACK_F12(%r15); \ + std %f13, STACK_F13(%r15); \ + std %f14, STACK_F14(%r15); \ + std %f15, STACK_F15(%r15); + +#define END_STACK(last_r) \ + lg %r1, 0(%r15); \ + ld %f8, STACK_F8(%r15); \ + ld %f9, STACK_F9(%r15); \ + ld %f10, STACK_F10(%r15); \ + ld %f11, STACK_F11(%r15); \ + ld %f12, STACK_F12(%r15); \ + ld %f13, STACK_F13(%r15); \ + ld %f14, STACK_F14(%r15); \ + ld %f15, STACK_F15(%r15); \ + lmg %r6, last_r, 6 * 8(%r1); \ + lgr %r15, %r1; \ + cfi_def_cfa_register(DW_REGNO_SP); + +#define PLUS(dst,src) \ + vaf dst, dst, src; + +#define XOR(dst,src) \ + vx dst, dst, src; + +#define ROTATE(v1,c) \ + verllf v1, v1, (c)(0); + +#define WORD_ROTATE(v1,s) \ + vsldb v1, v1, v1, ((s) * 4); + +#define DST_8(OPER, I, J) \ + OPER(A##I, J); OPER(B##I, J); OPER(C##I, J); OPER(D##I, J); \ + OPER(E##I, J); OPER(F##I, J); OPER(G##I, J); OPER(H##I, J); + +/********************************************************************** + round macros + **********************************************************************/ + +/********************************************************************** + 8-way chacha20 ("vertical") + **********************************************************************/ + +#define QUARTERROUND4_V8_POLY(x0,x1,x2,x3,x4,x5,x6,x7,\ + x8,x9,x10,x11,x12,x13,x14,x15,\ + y0,y1,y2,y3,y4,y5,y6,y7,\ + y8,y9,y10,y11,y12,y13,y14,y15,\ + op1,op2,op3,op4,op5,op6,op7,op8,\ + op9,op10,op11,op12) \ + op1; \ + PLUS(x0, x1); PLUS(x4, x5); \ + PLUS(x8, x9); PLUS(x12, x13); \ + PLUS(y0, y1); PLUS(y4, y5); \ + PLUS(y8, y9); PLUS(y12, y13); \ + op2; \ + XOR(x3, x0); XOR(x7, x4); \ + XOR(x11, x8); XOR(x15, x12); \ + XOR(y3, y0); XOR(y7, y4); \ + XOR(y11, y8); XOR(y15, y12); \ + op3; \ + ROTATE(x3, 16); ROTATE(x7, 16); \ + ROTATE(x11, 16); ROTATE(x15, 16); \ + ROTATE(y3, 16); ROTATE(y7, 16); \ + ROTATE(y11, 16); ROTATE(y15, 16); \ + op4; \ + PLUS(x2, x3); PLUS(x6, x7); \ + PLUS(x10, x11); PLUS(x14, x15); \ + PLUS(y2, y3); PLUS(y6, y7); \ + PLUS(y10, y11); PLUS(y14, y15); \ + op5; \ + XOR(x1, x2); XOR(x5, x6); \ + XOR(x9, x10); XOR(x13, x14); \ + XOR(y1, y2); XOR(y5, y6); \ + XOR(y9, y10); XOR(y13, y14); \ + op6; \ + ROTATE(x1,12); ROTATE(x5,12); \ + ROTATE(x9,12); ROTATE(x13,12); \ + ROTATE(y1,12); ROTATE(y5,12); \ + ROTATE(y9,12); ROTATE(y13,12); \ + op7; \ + PLUS(x0, x1); PLUS(x4, x5); \ + PLUS(x8, x9); PLUS(x12, x13); \ + PLUS(y0, y1); PLUS(y4, y5); \ + PLUS(y8, y9); PLUS(y12, y13); \ + op8; \ + XOR(x3, x0); XOR(x7, x4); \ + XOR(x11, x8); XOR(x15, x12); \ + XOR(y3, y0); XOR(y7, y4); \ + XOR(y11, y8); XOR(y15, y12); \ + op9; \ + ROTATE(x3,8); ROTATE(x7,8); \ + ROTATE(x11,8); ROTATE(x15,8); \ + ROTATE(y3,8); ROTATE(y7,8); \ + ROTATE(y11,8); ROTATE(y15,8); \ + op10; \ + PLUS(x2, x3); PLUS(x6, x7); \ + PLUS(x10, x11); PLUS(x14, x15); \ + PLUS(y2, y3); PLUS(y6, y7); \ + PLUS(y10, y11); PLUS(y14, y15); \ + op11; \ + XOR(x1, x2); XOR(x5, x6); \ + XOR(x9, x10); XOR(x13, x14); \ + XOR(y1, y2); XOR(y5, y6); \ + XOR(y9, y10); XOR(y13, y14); \ + op12; \ + ROTATE(x1,7); ROTATE(x5,7); \ + ROTATE(x9,7); ROTATE(x13,7); \ + ROTATE(y1,7); ROTATE(y5,7); \ + ROTATE(y9,7); ROTATE(y13,7); + +#define QUARTERROUND4_V8(x0,x1,x2,x3,x4,x5,x6,x7,x8,x9,x10,x11,x12,x13,x14,x15,\ + y0,y1,y2,y3,y4,y5,y6,y7,y8,y9,y10,y11,y12,y13,y14,y15) \ + QUARTERROUND4_V8_POLY(x0,x1,x2,x3,x4,x5,x6,x7,\ + x8,x9,x10,x11,x12,x13,x14,x15,\ + y0,y1,y2,y3,y4,y5,y6,y7,\ + y8,y9,y10,y11,y12,y13,y14,y15,\ + ,,,,,,,,,,,) + +#define TRANSPOSE_4X4_2(v0,v1,v2,v3,va,vb,vc,vd,tmp0,tmp1,tmp2,tmpa,tmpb,tmpc) \ + vmrhf tmp0, v0, v1; \ + vmrhf tmp1, v2, v3; \ + vmrlf tmp2, v0, v1; \ + vmrlf v3, v2, v3; \ + vmrhf tmpa, va, vb; \ + vmrhf tmpb, vc, vd; \ + vmrlf tmpc, va, vb; \ + vmrlf vd, vc, vd; \ + vpdi v0, tmp0, tmp1, 0; \ + vpdi v1, tmp0, tmp1, 5; \ + vpdi v2, tmp2, v3, 0; \ + vpdi v3, tmp2, v3, 5; \ + vpdi va, tmpa, tmpb, 0; \ + vpdi vb, tmpa, tmpb, 5; \ + vpdi vc, tmpc, vd, 0; \ + vpdi vd, tmpc, vd, 5; + +.balign 8 +.globl __chacha20_s390x_vx_blocks8 +ENTRY (__chacha20_s390x_vx_blocks8) + /* input: + * %r2: input + * %r3: dst + * %r4: src + * %r5: nblks (multiple of 8) + */ + + START_STACK(%r8); + lgr NBLKS, %r5; + + larl %r7, .Lconsts; + + /* Load counter. */ + lg %r8, (12 * 4)(INPUT); + rllg %r8, %r8, 32; + +.balign 4 + /* Process eight chacha20 blocks per loop. */ +.Lloop8: + vlm Y0, Y3, 0(INPUT); + + slgfi NBLKS, 8; + lghi ROUND, (20 / 2); + + /* Construct counter vectors X12/X13 & Y12/Y13. */ + vl X4, (.Ladd_counter_0123 - .Lconsts)(%r7); + vl Y4, (.Ladd_counter_4567 - .Lconsts)(%r7); + vrepf Y12, Y3, 0; + vrepf Y13, Y3, 1; + vaccf X5, Y12, X4; + vaccf Y5, Y12, Y4; + vaf X12, Y12, X4; + vaf Y12, Y12, Y4; + vaf X13, Y13, X5; + vaf Y13, Y13, Y5; + + vrepf X0, Y0, 0; + vrepf X1, Y0, 1; + vrepf X2, Y0, 2; + vrepf X3, Y0, 3; + vrepf X4, Y1, 0; + vrepf X5, Y1, 1; + vrepf X6, Y1, 2; + vrepf X7, Y1, 3; + vrepf X8, Y2, 0; + vrepf X9, Y2, 1; + vrepf X10, Y2, 2; + vrepf X11, Y2, 3; + vrepf X14, Y3, 2; + vrepf X15, Y3, 3; + + /* Store counters for blocks 0-7. */ + vstm X12, X13, (STACK_CTR + 0 * 16)(%r15); + vstm Y12, Y13, (STACK_CTR + 2 * 16)(%r15); + + vlr Y0, X0; + vlr Y1, X1; + vlr Y2, X2; + vlr Y3, X3; + vlr Y4, X4; + vlr Y5, X5; + vlr Y6, X6; + vlr Y7, X7; + vlr Y8, X8; + vlr Y9, X9; + vlr Y10, X10; + vlr Y11, X11; + vlr Y14, X14; + vlr Y15, X15; + + /* Update and store counter. */ + agfi %r8, 8; + rllg %r5, %r8, 32; + stg %r5, (12 * 4)(INPUT); + +.balign 4 +.Lround2_8: + QUARTERROUND4_V8(X0, X4, X8, X12, X1, X5, X9, X13, + X2, X6, X10, X14, X3, X7, X11, X15, + Y0, Y4, Y8, Y12, Y1, Y5, Y9, Y13, + Y2, Y6, Y10, Y14, Y3, Y7, Y11, Y15); + QUARTERROUND4_V8(X0, X5, X10, X15, X1, X6, X11, X12, + X2, X7, X8, X13, X3, X4, X9, X14, + Y0, Y5, Y10, Y15, Y1, Y6, Y11, Y12, + Y2, Y7, Y8, Y13, Y3, Y4, Y9, Y14); + brctg ROUND, .Lround2_8; + + /* Store blocks 4-7. */ + vstm Y0, Y15, STACK_Y0_Y15(%r15); + + /* Load counters for blocks 0-3. */ + vlm Y0, Y1, (STACK_CTR + 0 * 16)(%r15); + + lghi ROUND, 1; + j .Lfirst_output_4blks_8; + +.balign 4 +.Lsecond_output_4blks_8: + /* Load blocks 4-7. */ + vlm X0, X15, STACK_Y0_Y15(%r15); + + /* Load counters for blocks 4-7. */ + vlm Y0, Y1, (STACK_CTR + 2 * 16)(%r15); + + lghi ROUND, 0; + +.balign 4 + /* Output four chacha20 blocks per loop. */ +.Lfirst_output_4blks_8: + vlm Y12, Y15, 0(INPUT); + PLUS(X12, Y0); + PLUS(X13, Y1); + vrepf Y0, Y12, 0; + vrepf Y1, Y12, 1; + vrepf Y2, Y12, 2; + vrepf Y3, Y12, 3; + vrepf Y4, Y13, 0; + vrepf Y5, Y13, 1; + vrepf Y6, Y13, 2; + vrepf Y7, Y13, 3; + vrepf Y8, Y14, 0; + vrepf Y9, Y14, 1; + vrepf Y10, Y14, 2; + vrepf Y11, Y14, 3; + vrepf Y14, Y15, 2; + vrepf Y15, Y15, 3; + PLUS(X0, Y0); + PLUS(X1, Y1); + PLUS(X2, Y2); + PLUS(X3, Y3); + PLUS(X4, Y4); + PLUS(X5, Y5); + PLUS(X6, Y6); + PLUS(X7, Y7); + PLUS(X8, Y8); + PLUS(X9, Y9); + PLUS(X10, Y10); + PLUS(X11, Y11); + PLUS(X14, Y14); + PLUS(X15, Y15); + + vl Y15, (.Lbswap32 - .Lconsts)(%r7); + TRANSPOSE_4X4_2(X0, X1, X2, X3, X4, X5, X6, X7, + Y9, Y10, Y11, Y12, Y13, Y14); + TRANSPOSE_4X4_2(X8, X9, X10, X11, X12, X13, X14, X15, + Y9, Y10, Y11, Y12, Y13, Y14); + + vlm Y0, Y14, 0(SRC); + vperm X0, X0, X0, Y15; + vperm X1, X1, X1, Y15; + vperm X2, X2, X2, Y15; + vperm X3, X3, X3, Y15; + vperm X4, X4, X4, Y15; + vperm X5, X5, X5, Y15; + vperm X6, X6, X6, Y15; + vperm X7, X7, X7, Y15; + vperm X8, X8, X8, Y15; + vperm X9, X9, X9, Y15; + vperm X10, X10, X10, Y15; + vperm X11, X11, X11, Y15; + vperm X12, X12, X12, Y15; + vperm X13, X13, X13, Y15; + vperm X14, X14, X14, Y15; + vperm X15, X15, X15, Y15; + vl Y15, (15 * 16)(SRC); + + XOR(Y0, X0); + XOR(Y1, X4); + XOR(Y2, X8); + XOR(Y3, X12); + XOR(Y4, X1); + XOR(Y5, X5); + XOR(Y6, X9); + XOR(Y7, X13); + XOR(Y8, X2); + XOR(Y9, X6); + XOR(Y10, X10); + XOR(Y11, X14); + XOR(Y12, X3); + XOR(Y13, X7); + XOR(Y14, X11); + XOR(Y15, X15); + vstm Y0, Y15, 0(DST); + + aghi SRC, 256; + aghi DST, 256; + + clgije ROUND, 1, .Lsecond_output_4blks_8; + + clgijhe NBLKS, 8, .Lloop8; + + /* Clear the used vector registers. */ + DST_8(CLEAR, 0, _); + DST_8(CLEAR, 1, _); + DST_8(CLEAR, 2, _); + DST_8(CLEAR, 3, _); + + /* Clear sensitive data in stack. */ + vlm Y0, Y15, STACK_Y0_Y15(%r15); + vlm Y0, Y3, STACK_CTR(%r15); + + END_STACK(%r8); + xgr %r2, %r2; + br %r14; +END (__chacha20_s390x_vx_blocks8) + +#endif /* HAVE_S390_VX_ASM_SUPPORT */ diff --git a/sysdeps/s390/s390-64/chacha20_arch.h b/sysdeps/s390/s390-64/chacha20_arch.h new file mode 100644 index 0000000000..78252c5488 --- /dev/null +++ b/sysdeps/s390/s390-64/chacha20_arch.h @@ -0,0 +1,45 @@ +/* s390x optimization for ChaCha20.VE_S390_VX_ASM_SUPPORT + Copyright (C) 2022 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + . */ + +#include +#include +#include + +unsigned int __chacha20_s390x_vx_blocks8 (uint32_t *state, uint8_t *dst, + const uint8_t *src, size_t nblks) + attribute_hidden; + +static inline void +chacha20_crypt (uint32_t *state, uint8_t *dst, const uint8_t *src, + size_t bytes) +{ +#ifdef HAVE_S390_VX_ASM_SUPPORT + _Static_assert (CHACHA20_BUFSIZE % 8 == 0, + "CHACHA20_BUFSIZE not multiple of 8"); + _Static_assert (CHACHA20_BUFSIZE >= CHACHA20_BLOCK_SIZE * 8, + "CHACHA20_BUFSIZE < CHACHA20_BLOCK_SIZE * 8"); + + if (GLRO(dl_hwcap) & HWCAP_S390_VX) + { + __chacha20_s390x_vx_blocks8 (state, dst, src, + CHACHA20_BUFSIZE / CHACHA20_BLOCK_SIZE); + return; + } +#endif + chacha20_crypt_generic (state, dst, src, bytes); +} From patchwork Tue Apr 19 21:28:12 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Adhemerval Zanella Netto X-Patchwork-Id: 53055 Return-Path: X-Original-To: patchwork@sourceware.org Delivered-To: patchwork@sourceware.org Received: from server2.sourceware.org (localhost [IPv6:::1]) by sourceware.org (Postfix) with ESMTP id DE04D3857827 for ; Tue, 19 Apr 2022 21:34:47 +0000 (GMT) DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org DE04D3857827 DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=sourceware.org; s=default; t=1650404087; bh=2CP20qY5p9O5mMCEL3+EMYoyw6qZK6pwYmU6euwiUI8=; h=To:Subject:Date:In-Reply-To:References:List-Id:List-Unsubscribe: List-Archive:List-Post:List-Help:List-Subscribe:From:Reply-To: From; b=XYcuBOEhD02Arcy31kUJk+2zkPwEta+6HNeSXop2jqaimmrqTI0IuwW0TJuroBWmP aUAVOb6fjC/YSUBveApHMJKgtKM5xJeZ0Cj6NGJ1ap199cxoy+eGQBVkpg0pQJyMCi AoG/oK2NQQL3wLDeJDYOW1+xQn59QUU193jcI7kc= X-Original-To: libc-alpha@sourceware.org Delivered-To: libc-alpha@sourceware.org Received: from mail-ot1-x32d.google.com (mail-ot1-x32d.google.com [IPv6:2607:f8b0:4864:20::32d]) by sourceware.org (Postfix) with ESMTPS id 056363857348 for ; Tue, 19 Apr 2022 21:28:36 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org 056363857348 Received: by mail-ot1-x32d.google.com with SMTP id l9-20020a056830268900b006054381dd35so6685377otu.4 for ; Tue, 19 Apr 2022 14:28:35 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; h=x-gm-message-state:from:to:subject:date:message-id:in-reply-to :references:mime-version:content-transfer-encoding; bh=2CP20qY5p9O5mMCEL3+EMYoyw6qZK6pwYmU6euwiUI8=; b=vyFp/RzXvQtPYT14QrIEpNBiGpxnFI9KD+NN8RgTpAJ5bKjs1kAgGP8ZSAIW/5SlBg 16Ug35ehdONxo88uCsoq1qzzHsRu7JQlGNG0zs+VHwJ3gYFuD6P/MvX9RjlpWr5QSSq/ Qf8+hYgnNTQxbRo6A9XTSpJ8hE91C5jlQH/E39kTjoq2D5VnBGkGJD98TNE9Jdr+e0jZ VIKwWfUItCtI+gUHJ1d+sv6LM72z6AwC3wRZ745YYckp4Myb43B7dc4wTRTZ4Qyes+jg kXyfwgDZRBPPE7mwxUBIcJ523BySufdXq0bIUSF8kDMLeQD21r9UZgAM0i7GnmC3+Z1E SvAg== X-Gm-Message-State: AOAM530W8V3pWU6QKBUCjBbk0rynKWxsLiiZrhPrlrsfpPN/75FKbH7z R3N2sYVfLfiIyCKEl3TUju7arJBm2faosg== X-Google-Smtp-Source: ABdhPJzmXlVVwz8BPDg8UHKP/C+LTpvSI1hOj4uC86D25RplmPQMtK5y7f73lnqON3+KdKTPApgejg== X-Received: by 2002:a9d:58c3:0:b0:5e6:d8f7:c18c with SMTP id s3-20020a9d58c3000000b005e6d8f7c18cmr6571268oth.364.1650403714635; Tue, 19 Apr 2022 14:28:34 -0700 (PDT) Received: from birita.. ([2804:431:c7ca:c9d0:98f6:7aed:2f61:2745]) by smtp.gmail.com with ESMTPSA id nf9-20020a056871460900b000e2c44ca8edsm5473321oab.6.2022.04.19.14.28.33 for (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Tue, 19 Apr 2022 14:28:34 -0700 (PDT) To: libc-alpha@sourceware.org Subject: [PATCH v3 9/9] stdlib: Add TLS optimization to arc4random Date: Tue, 19 Apr 2022 18:28:12 -0300 Message-Id: <20220419212812.2688764-10-adhemerval.zanella@linaro.org> X-Mailer: git-send-email 2.32.0 In-Reply-To: <20220419212812.2688764-1-adhemerval.zanella@linaro.org> References: <20220419212812.2688764-1-adhemerval.zanella@linaro.org> MIME-Version: 1.0 X-Spam-Status: No, score=-12.2 required=5.0 tests=BAYES_00, DKIM_SIGNED, DKIM_VALID, DKIM_VALID_AU, DKIM_VALID_EF, GIT_PATCH_0, KAM_ASCII_DIVIDERS, KAM_SHORT, RCVD_IN_DNSWL_NONE, SPF_HELO_NONE, SPF_PASS, TXREP, T_SCC_BODY_TEXT_LINE autolearn=ham autolearn_force=no version=3.4.4 X-Spam-Checker-Version: SpamAssassin 3.4.4 (2020-01-24) on server2.sourceware.org X-BeenThere: libc-alpha@sourceware.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Libc-alpha mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-Patchwork-Original-From: Adhemerval Zanella via Libc-alpha From: Adhemerval Zanella Netto Reply-To: Adhemerval Zanella Errors-To: libc-alpha-bounces+patchwork=sourceware.org@sourceware.org Sender: "Libc-alpha" The arc4random state is moved to TCB, so there is no allocation failure. It adds about 592 bytes struct pthread. Now that the state is thread private within a shared struct, the MADV_WIPEONFORK usage is removed. The cipher state reset is done solely by the atfork internal handler. The state is also cleared on thread exit iff it was initialized (so if arc4random is not called it is not touched). Although it is lock-free, arc4random is still not async-signal-safe (the per thread state is not updated atomically). On x86_64 using AVX2 it shows a slight better performance: From -------------------------------------------------- arc4random [single-thread] 809.53 arc4random_buf(16) [single-thread] 1242.56 arc4random_buf(32) [single-thread] 1915.90 arc4random_buf(48) [single-thread] 2230.03 arc4random_buf(64) [single-thread] 2429.68 arc4random_buf(80) [single-thread] 2489.70 arc4random_buf(96) [single-thread] 2598.88 arc4random_buf(112) [single-thread] 2699.93 arc4random_buf(128) [single-thread] 2747.31 To MB/s -------------------------------------------------- arc4random [single-thread] 941.54 arc4random_buf(16) [single-thread] 1409.39 arc4random_buf(32) [single-thread] 2056.17 arc4random_buf(48) [single-thread] 2367.13 arc4random_buf(64) [single-thread] 2551.44 arc4random_buf(80) [single-thread] 2601.38 arc4random_buf(96) [single-thread] 2710.21 arc4random_buf(112) [single-thread] 2797.86 arc4random_buf(128) [single-thread] 2846.12 -------------------------------------------------- However it shows a large speed up specially on architecture with most costly atomics. For instance, on a aarch64 Neoverse N1: From MB/s -------------------------------------------------- arc4random [single-thread] 154.98 arc4random_buf(16) [single-thread] 342.63 arc4random_buf(32) [single-thread] 485.91 arc4random_buf(48) [single-thread] 539.95 arc4random_buf(64) [single-thread] 593.38 arc4random_buf(80) [single-thread] 629.45 arc4random_buf(96) [single-thread] 655.78 arc4random_buf(112) [single-thread] 670.54 arc4random_buf(128) [single-thread] 681.65 -------------------------------------------------- To MB/s -------------------------------------------------- arc4random [single-thread] 335.94 arc4random_buf(16) [single-thread] 498.69 arc4random_buf(32) [single-thread] 612.24 arc4random_buf(48) [single-thread] 655.77 arc4random_buf(64) [single-thread] 691.97 arc4random_buf(80) [single-thread] 701.68 arc4random_buf(96) [single-thread] 710.35 arc4random_buf(112) [single-thread] 714.23 arc4random_buf(128) [single-thread] 722.13 -------------------------------------------------- Checked on x86_64-linux-gnu. --- nptl/allocatestack.c | 5 +- stdlib/arc4random.c | 137 +++++++------------------ stdlib/arc4random.h | 45 ++++++++ stdlib/arc4random_uniform.c | 8 +- stdlib/chacha20.c | 3 - stdlib/tst-arc4random-chacha20.c | 2 +- sysdeps/generic/tls-internal-struct.h | 3 + sysdeps/unix/sysv/linux/tls-internal.h | 27 ++++- 8 files changed, 115 insertions(+), 115 deletions(-) create mode 100644 stdlib/arc4random.h diff --git a/nptl/allocatestack.c b/nptl/allocatestack.c index 01a282f3f6..ada65d40c2 100644 --- a/nptl/allocatestack.c +++ b/nptl/allocatestack.c @@ -32,6 +32,7 @@ #include #include #include +#include /* Default alignment of stack. */ #ifndef STACK_ALIGN @@ -127,7 +128,7 @@ get_cached_stack (size_t *sizep, void **memp) result->exiting = false; __libc_lock_init (result->exit_lock); - result->tls_state = (struct tls_internal_t) { 0 }; + __glibc_tls_internal_init (&result->tls_state); /* Clear the DTV. */ dtv_t *dtv = GET_DTV (TLS_TPADJ (result)); @@ -559,6 +560,8 @@ allocate_stack (const struct pthread_attr *attr, struct pthread **pdp, #endif pd->robust_head.list = &pd->robust_head; + __glibc_tls_internal_init (&pd->tls_state); + /* We place the thread descriptor at the end of the stack. */ *pdp = pd; diff --git a/stdlib/arc4random.c b/stdlib/arc4random.c index cddb0e405a..6144275c08 100644 --- a/stdlib/arc4random.c +++ b/stdlib/arc4random.c @@ -16,14 +16,15 @@ License along with the GNU C Library; if not, see . */ +#include #include -#include #include #include #include #include #include #include +#include /* Besides the cipher state 'ctx', it keeps two counters: 'have' is the current valid bytes not yet consumed in 'buf', while 'count' is the maximum @@ -37,42 +38,16 @@ arc4random calls (since only multiple call it will encrypt the next block). */ -/* Maximum number bytes until reseed (16 MB). */ -#define CHACHE_RESEED_SIZE (16 * 1024 * 1024) -/* Internal buffer size in bytes (1KB). */ -#define CHACHA20_BUFSIZE (8 * CHACHA20_BLOCK_SIZE) - #include -static struct arc4random_state -{ - uint32_t ctx[CHACHA20_STATE_LEN]; - size_t have; - size_t count; - uint8_t buf[CHACHA20_BUFSIZE]; -} *state; - -/* Indicate that MADV_WIPEONFORK is supported by the kernel and thus - it does not require to clear the internal state. */ -static bool __arc4random_wipeonfork = false; - -__libc_lock_define_initialized (, __arc4random_lock); - -/* Called from the fork function to reset the state if MADV_WIPEONFORK is - not supported and to reinit the internal lock. */ +/* Called from the fork function to reset the state. */ void __arc4random_fork_subprocess (void) { - if (__arc4random_wipeonfork && state != NULL) - memset (state, 0, sizeof (struct arc4random_state)); - - __libc_lock_init (__arc4random_lock); -} - -static void -arc4random_allocate_failure (void) -{ - __libc_fatal ("Fatal glibc error: Cannot allocate memory for arc4random\n"); + struct arc4random_state *state = &__glibc_tls_internal()->rnd_state; + memset (state, 0, sizeof (struct arc4random_state)); + /* Force key init. */ + state->count = -1; } static void @@ -81,33 +56,10 @@ arc4random_getrandom_failure (void) __libc_fatal ("Fatal glibc error: Cannot get entropy for arc4random\n"); } -/* Fork detection is done by checking if MADV_WIPEONFORK supported. If not - the fork callback will reset the state on the fork call. It does not - handle direct clone calls, nor vfork or _Fork (arc4random is not - async-signal-safe due the internal lock usage). */ -static void -arc4random_init (uint8_t *buf, size_t len) -{ - state = __mmap (NULL, sizeof (struct arc4random_state), - PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_PRIVATE, -1, 0); - if (state == MAP_FAILED) - arc4random_allocate_failure (); - -#ifdef MADV_WIPEONFORK - int r = __madvise (state, sizeof (struct arc4random_state), MADV_WIPEONFORK); - if (r == 0) - __arc4random_wipeonfork = true; - else if (errno != EINVAL) - arc4random_allocate_failure (); -#endif - - chacha20_init (state->ctx, buf, buf + CHACHA20_KEY_SIZE); -} - #define min(x,y) (((x) > (y)) ? (y) : (x)) static void -arc4random_rekey (uint8_t *rnd, size_t rndlen) +arc4random_rekey (struct arc4random_state *state, uint8_t *rnd, size_t rndlen) { memset (state->buf, 0, sizeof state->buf); chacha20_crypt (state->ctx, state->buf, state->buf, sizeof state->buf); @@ -152,41 +104,41 @@ arc4random_getentropy (uint8_t *rnd, size_t len) arc4random_getrandom_failure (); } -/* Either allocates the state buffer or reinit it by reseeding the cipher - state with kernel entropy. */ -static void -arc4random_stir (void) +/* Reinit the thread context by reseeding the cipher state with kernel + entropy. */ +static struct arc4random_state * +arc4random_check_stir (size_t len) { - uint8_t rnd[CHACHA20_KEY_SIZE + CHACHA20_IV_SIZE]; - arc4random_getentropy (rnd, sizeof rnd); + struct arc4random_state *state = &__glibc_tls_internal()->rnd_state; - if (state == NULL) - arc4random_init (rnd, sizeof rnd); - else - arc4random_rekey (rnd, sizeof rnd); + if (state->count < len || state->count == -1) + { + uint8_t rnd[CHACHA20_KEY_SIZE + CHACHA20_IV_SIZE]; + arc4random_getentropy (rnd, sizeof rnd); - explicit_bzero (rnd, sizeof rnd); + if (state->count > CHACHE_RESEED_SIZE) + chacha20_init (state->ctx, rnd, rnd + CHACHA20_KEY_SIZE); + else + arc4random_rekey (state, rnd, sizeof rnd); - state->have = 0; - memset (state->buf, 0, sizeof state->buf); - state->count = CHACHE_RESEED_SIZE; -} + explicit_bzero (rnd, sizeof rnd); -static void -arc4random_check_stir (size_t len) -{ - if (state == NULL || state->count < len) - arc4random_stir (); + state->have = 0; + memset (state->buf, 0, sizeof state->buf); + state->count = CHACHE_RESEED_SIZE; + } if (state->count <= len) state->count = 0; else state->count -= len; + + return state; } void -__arc4random_buf_internal (void *buffer, size_t len) +__arc4random_buf (void *buffer, size_t len) { - arc4random_check_stir (len); + struct arc4random_state *state = arc4random_check_stir (len); while (len > 0) { @@ -201,29 +153,20 @@ __arc4random_buf_internal (void *buffer, size_t len) state->have -= m; } if (state->have == 0) - arc4random_rekey (NULL, 0); + arc4random_rekey (state, NULL, 0); } } - -void -__arc4random_buf (void *buffer, size_t len) -{ - __libc_lock_lock (__arc4random_lock); - __arc4random_buf_internal (buffer, len); - __libc_lock_unlock (__arc4random_lock); -} libc_hidden_def (__arc4random_buf) weak_alias (__arc4random_buf, arc4random_buf) - -static uint32_t -__arc4random_internal (void) +uint32_t +__arc4random (void) { uint32_t r; - arc4random_check_stir (sizeof (uint32_t)); + struct arc4random_state *state = arc4random_check_stir (sizeof (uint32_t)); if (state->have < sizeof (uint32_t)) - arc4random_rekey (NULL, 0); + arc4random_rekey (state, NULL, 0); uint8_t *ks = state->buf + sizeof (state->buf) - state->have; memcpy (&r, ks, sizeof (uint32_t)); memset (ks, 0, sizeof (uint32_t)); @@ -231,15 +174,5 @@ __arc4random_internal (void) return r; } - -uint32_t -__arc4random (void) -{ - uint32_t r; - __libc_lock_lock (__arc4random_lock); - r = __arc4random_internal (); - __libc_lock_unlock (__arc4random_lock); - return r; -} libc_hidden_def (__arc4random) weak_alias (__arc4random, arc4random) diff --git a/stdlib/arc4random.h b/stdlib/arc4random.h new file mode 100644 index 0000000000..40672299d0 --- /dev/null +++ b/stdlib/arc4random.h @@ -0,0 +1,45 @@ +/* Arc4random definition used on TLS. + Copyright (C) 2022 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + . */ + +#ifndef _CHACHA20_H +#define _CHACHA20_H + +#include +#include + +/* Internal ChaCha20 state. */ +#define CHACHA20_STATE_LEN 16 +#define CHACHA20_BLOCK_SIZE 64 + +/* Maximum number bytes until reseed (16 MB). */ +#define CHACHE_RESEED_SIZE (16 * 1024 * 1024) + +/* Internal arc4random buffer, used on each feedback step so offer some + backtracking protection and to allow better used of vectorized + chacha20 implementations. */ +#define CHACHA20_BUFSIZE (8 * CHACHA20_BLOCK_SIZE) + +struct arc4random_state +{ + uint32_t ctx[CHACHA20_STATE_LEN]; + size_t have; + size_t count; + uint8_t buf[CHACHA20_BUFSIZE]; +}; + +#endif diff --git a/stdlib/arc4random_uniform.c b/stdlib/arc4random_uniform.c index 96ffe62df1..7d0140c375 100644 --- a/stdlib/arc4random_uniform.c +++ b/stdlib/arc4random_uniform.c @@ -46,7 +46,7 @@ random_bytes (uint32_t *result, uint32_t byte_count) unsigned char *ptr = (unsigned char *) result; if (__BYTE_ORDER == __BIG_ENDIAN) ptr += 4 - byte_count; - __arc4random_buf_internal (ptr, byte_count); + __arc4random_buf (ptr, byte_count); } static uint32_t @@ -142,11 +142,7 @@ __libc_lock_define (extern , __arc4random_lock attribute_hidden) uint32_t __arc4random_uniform (uint32_t upper_bound) { - uint32_t r; - __libc_lock_lock (__arc4random_lock); - r = compute_uniform (upper_bound); - __libc_lock_unlock (__arc4random_lock); - return r; + return compute_uniform (upper_bound); } libc_hidden_def (__arc4random_uniform) weak_alias (__arc4random_uniform, arc4random_uniform) diff --git a/stdlib/chacha20.c b/stdlib/chacha20.c index fea4994169..0fb55c0fa3 100644 --- a/stdlib/chacha20.c +++ b/stdlib/chacha20.c @@ -26,11 +26,8 @@ #define CHACHA20_IV_SIZE 16 #define CHACHA20_KEY_SIZE 32 -#define CHACHA20_BLOCK_SIZE 64 #define CHACHA20_BLOCK_WORDS (CHACHA20_BLOCK_SIZE / sizeof (uint32_t)) -#define CHACHA20_STATE_LEN 16 - /* Defining CHACHA20_XOR_FINAL issues the final XOR using the input as defined Sby RFC8439. Since the input stream will either zero bytes (initial state) or the PRNG output itself this step does not add any extra entropy. */ diff --git a/stdlib/tst-arc4random-chacha20.c b/stdlib/tst-arc4random-chacha20.c index dd0ef6d8ba..614e6e0736 100644 --- a/stdlib/tst-arc4random-chacha20.c +++ b/stdlib/tst-arc4random-chacha20.c @@ -16,11 +16,11 @@ License along with the GNU C Library; if not, see . */ +#include #include #include /* It does not define CHACHA20_XOR_FINAL to check what glibc actual uses. */ -#define CHACHA20_BUFSIZE (8 * CHACHA20_BLOCK_SIZE) #include static int diff --git a/sysdeps/generic/tls-internal-struct.h b/sysdeps/generic/tls-internal-struct.h index d76c715a96..5d0e2fba53 100644 --- a/sysdeps/generic/tls-internal-struct.h +++ b/sysdeps/generic/tls-internal-struct.h @@ -19,10 +19,13 @@ #ifndef _TLS_INTERNAL_STRUCT_H #define _TLS_INTERNAL_STRUCT_H 1 +#include + struct tls_internal_t { char *strsignal_buf; char *strerror_l_buf; + struct arc4random_state rnd_state; }; #endif diff --git a/sysdeps/unix/sysv/linux/tls-internal.h b/sysdeps/unix/sysv/linux/tls-internal.h index f7a1a62135..16ff836d05 100644 --- a/sysdeps/unix/sysv/linux/tls-internal.h +++ b/sysdeps/unix/sysv/linux/tls-internal.h @@ -22,6 +22,19 @@ #include #include +static inline void +__glibc_tls_internal_init (struct tls_internal_t *tls_state) +{ + tls_state->strsignal_buf = NULL; + tls_state->strerror_l_buf = NULL; + + /* Force key init on created threads. There is no need to clear the + initial state since it will be done either by allocation a new + stack (through mmap with MAP_ANONYMOUS) or by the free function + below). */ + tls_state->rnd_state.count = -1; +} + static inline struct tls_internal_t * __glibc_tls_internal (void) { @@ -31,8 +44,18 @@ __glibc_tls_internal (void) static inline void __glibc_tls_internal_free (void) { - free (THREAD_SELF->tls_state.strsignal_buf); - free (THREAD_SELF->tls_state.strerror_l_buf); + struct pthread *self = THREAD_SELF; + free (self->tls_state.strsignal_buf); + free (self->tls_state.strerror_l_buf); + if (self->tls_state.rnd_state.count != -1) + { + /* Clear any lingering random state prior so if the thread stack + is cached it won't leak any data. */ + memset (&self->tls_state.rnd_state, 0, + sizeof self->tls_state.rnd_state); + /* Force key init on created threads. */ + self->tls_state.rnd_state.count = -1; + } } #endif