From 09f2921b5dacaf79b391652cecd606be4dd69f50 Mon Sep 17 00:00:00 2001 From: Julius Werner Date: Tue, 29 Sep 2015 13:51:35 -0700 Subject: cbfs: Add LZ4 in-place decompression support for pre-RAM stages This patch ports the LZ4 decompression code that debuted in libpayload last year to coreboot for use in CBFS stages (upgrading the base algorithm to LZ4's dev branch to access the new in-place decompression checks). This is especially useful for pre-RAM stages in constrained SRAM-based systems, which previously could not be compressed due to the size requirements of the LZMA scratchpad and bounce buffer. The LZ4 algorithm offers a very lean decompressor function and in-place decompression support to achieve roughly the same boot speed gains (trading compression ratio for decompression time) with nearly no memory overhead. For now we only activate it for the stages that had previously not been compressed at all on non-XIP (read: non-x86) boards. In the future we may also consider replacing LZMA completely for certain boards, since which algorithm wins out on boot speed depends on board-specific parameters (architecture, processor speed, SPI transfer rate, etc.). BRANCH=None BUG=None TEST=Built and booted Oak, Jerry, Nyan and Falco. Measured boot time on Oak to be about ~20ms faster (cutting load times for affected stages almost in half). Change-Id: Iec256c0e6d585d1b69985461939884a54e3ab900 Signed-off-by: Julius Werner Reviewed-on: https://review.coreboot.org/13638 Tested-by: build bot (Jenkins) Reviewed-by: Aaron Durbin --- src/commonlib/Makefile.inc | 5 + src/commonlib/include/commonlib/cbfs_serialized.h | 1 + src/commonlib/include/commonlib/compression.h | 34 +++ .../include/commonlib/timestamp_serialized.h | 2 + src/commonlib/lz4.c.inc | 280 +++++++++++++++++++++ src/commonlib/lz4_wrapper.c | 193 ++++++++++++++ 6 files changed, 515 insertions(+) create mode 100644 src/commonlib/include/commonlib/compression.h create mode 100644 src/commonlib/lz4.c.inc create mode 100644 src/commonlib/lz4_wrapper.c (limited to 'src/commonlib') diff --git a/src/commonlib/Makefile.inc b/src/commonlib/Makefile.inc index 19f9ba3b8d..2752922067 100644 --- a/src/commonlib/Makefile.inc +++ b/src/commonlib/Makefile.inc @@ -16,3 +16,8 @@ verstage-y += cbfs.c romstage-y += cbfs.c ramstage-y += cbfs.c smm-y += cbfs.c + +bootblock-y += lz4_wrapper.c +verstage-y += lz4_wrapper.c +romstage-y += lz4_wrapper.c +ramstage-y += lz4_wrapper.c diff --git a/src/commonlib/include/commonlib/cbfs_serialized.h b/src/commonlib/include/commonlib/cbfs_serialized.h index bea5d6bb9d..c1a1c3b542 100644 --- a/src/commonlib/include/commonlib/cbfs_serialized.h +++ b/src/commonlib/include/commonlib/cbfs_serialized.h @@ -56,6 +56,7 @@ #define CBFS_COMPRESS_NONE 0 #define CBFS_COMPRESS_LZMA 1 +#define CBFS_COMPRESS_LZ4 2 /** These are standard component types for well known components (i.e - those that coreboot needs to consume. diff --git a/src/commonlib/include/commonlib/compression.h b/src/commonlib/include/commonlib/compression.h new file mode 100644 index 0000000000..428ee42e65 --- /dev/null +++ b/src/commonlib/include/commonlib/compression.h @@ -0,0 +1,34 @@ +/* + * This file is part of the coreboot project. + * + * Copyright 2016 Google Inc. + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; version 2 of the License. + * + * This program 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 General Public License for more details. + */ + +#ifndef _COMMONLIB_COMPRESSION_H_ +#define _COMMONLIB_COMPRESSION_H_ + +#include + +/* Decompresses an LZ4F image (multiple LZ4 blocks with frame header) from src + * to dst, ensuring that it doesn't read more than srcn bytes and doesn't write + * more than dstn. Buffer sizes must stay below 2GB. Can decompress files loaded + * to the end of a buffer in-place, as long as buffer is larger than the final + * output size. (Usually just a few bytes, but may be up to (8 + dstn/255) in + * worst case. Will reliably return an error if buffer was too small.) + * Returns amount of decompressed bytes, or 0 on error. + */ +size_t ulz4fn(const void *src, size_t srcn, void *dst, size_t dstn); + +/* Same as ulz4fn() but does not perform any bounds checks. */ +size_t ulz4f(const void *src, void *dst); + +#endif /* _COMMONLIB_COMPRESSION_H_ */ diff --git a/src/commonlib/include/commonlib/timestamp_serialized.h b/src/commonlib/include/commonlib/timestamp_serialized.h index 6ce087924d..c665964a0d 100644 --- a/src/commonlib/include/commonlib/timestamp_serialized.h +++ b/src/commonlib/include/commonlib/timestamp_serialized.h @@ -47,6 +47,8 @@ enum timestamp_id { TS_END_COPYROM = 14, TS_START_ULZMA = 15, TS_END_ULZMA = 16, + TS_START_ULZ4F = 17, + TS_END_ULZ4F = 18, TS_DEVICE_ENUMERATE = 30, TS_DEVICE_CONFIGURE = 40, TS_DEVICE_ENABLE = 50, diff --git a/src/commonlib/lz4.c.inc b/src/commonlib/lz4.c.inc new file mode 100644 index 0000000000..b3be4e5b44 --- /dev/null +++ b/src/commonlib/lz4.c.inc @@ -0,0 +1,280 @@ +/* + LZ4 - Fast LZ compression algorithm + Copyright (C) 2011-2015, Yann Collet. + + BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php) + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are + met: + + * Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above + copyright notice, this list of conditions and the following disclaimer + in the documentation and/or other materials provided with the + distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + + You can contact the author at : + - LZ4 source repository : https://github.com/Cyan4973/lz4 + - LZ4 public forum : https://groups.google.com/forum/#!forum/lz4c +*/ + + +/************************************** +* Reading and writing into memory +**************************************/ + +/* customized variant of memcpy, which can overwrite up to 7 bytes beyond dstEnd */ +static void LZ4_wildCopy(void* dstPtr, const void* srcPtr, void* dstEnd) +{ + BYTE* d = (BYTE*)dstPtr; + const BYTE* s = (const BYTE*)srcPtr; + BYTE* const e = (BYTE*)dstEnd; + +#if 0 + const size_t l2 = 8 - (((size_t)d) & (sizeof(void*)-1)); + LZ4_copy8(d,s); if (d>e-9) return; + d+=l2; s+=l2; +#endif /* join to align */ + + do { LZ4_copy8(d,s); d+=8; s+=8; } while (d= op) && (ip < oend)); + + + /* Special cases */ + if ((partialDecoding) && (oexit> oend-MFLIMIT)) oexit = oend-MFLIMIT; /* targetOutputSize too high => decode everything */ + if ((endOnInput) && (unlikely(outputSize==0))) return ((inputSize==1) && (*ip==0)) ? 0 : -1; /* Empty output buffer */ + if ((!endOnInput) && (unlikely(outputSize==0))) return (*ip==0?1:-1); + + + /* Main Loop */ + while (1) + { + unsigned token; + size_t length; + const BYTE* match; + size_t offset; + + if (unlikely((inPlaceDecode) && (op + WILDCOPYLENGTH > ip))) goto _output_error; /* output stream ran over input stream */ + + /* get literal length */ + token = *ip++; + if ((length=(token>>ML_BITS)) == RUN_MASK) + { + unsigned s; + do + { + s = *ip++; + length += s; + } + while ( likely(endOnInput ? ip(partialDecoding?oexit:oend-MFLIMIT)) || (ip+length>iend-(2+1+LASTLITERALS))) ) + || ((!endOnInput) && (cpy>oend-WILDCOPYLENGTH))) + { + if (partialDecoding) + { + if (cpy > oend) goto _output_error; /* Error : write attempt beyond end of output buffer */ + if ((endOnInput) && (ip+length > iend)) goto _output_error; /* Error : read attempt beyond end of input buffer */ + } + else + { + if ((!endOnInput) && (cpy != oend)) goto _output_error; /* Error : block decoding must stop exactly there */ + if ((endOnInput) && ((ip+length != iend) || (cpy > oend))) goto _output_error; /* Error : input must be consumed */ + } + memmove(op, ip, length); + ip += length; + op += length; + break; /* Necessarily EOF, due to parsing restrictions */ + } + LZ4_wildCopy(op, ip, cpy); + ip += length; op = cpy; + + /* get offset */ + offset = LZ4_readLE16(ip); ip+=2; + match = op - offset; + if ((checkOffset) && (unlikely(match < lowLimit))) goto _output_error; /* Error : offset outside buffers */ + + /* get matchlength */ + length = token & ML_MASK; + if (length == ML_MASK) + { + unsigned s; + do + { + if ((endOnInput) && (ip > iend-LASTLITERALS)) goto _output_error; + s = *ip++; + length += s; + } while (s==255); + if ((safeDecode) && unlikely((size_t)(op+length)<(size_t)op)) goto _output_error; /* overflow detection */ + } + length += MINMATCH; + + /* check external dictionary */ + if ((dict==usingExtDict) && (match < lowPrefix)) + { + if (unlikely(op+length > oend-LASTLITERALS)) goto _output_error; /* doesn't respect parsing restriction */ + + if (length <= (size_t)(lowPrefix-match)) + { + /* match can be copied as a single segment from external dictionary */ + match = dictEnd - (lowPrefix-match); + memmove(op, match, length); op += length; + } + else + { + /* match encompass external dictionary and current block */ + size_t copySize = (size_t)(lowPrefix-match); + memcpy(op, dictEnd - copySize, copySize); + op += copySize; + copySize = length - copySize; + if (copySize > (size_t)(op-lowPrefix)) /* overlap copy */ + { + BYTE* const endOfMatch = op + copySize; + const BYTE* copyFrom = lowPrefix; + while (op < endOfMatch) *op++ = *copyFrom++; + } + else + { + memcpy(op, lowPrefix, copySize); + op += copySize; + } + } + continue; + } + + /* copy match within block */ + cpy = op + length; + if (unlikely(offset<8)) + { + const int dec64 = dec64table[offset]; + op[0] = match[0]; + op[1] = match[1]; + op[2] = match[2]; + op[3] = match[3]; + match += dec32table[offset]; + memcpy(op+4, match, 4); + match -= dec64; + } else { LZ4_copy8(op, match); match+=8; } + op += 8; + + if (unlikely(cpy>oend-12)) + { + BYTE* const oCopyLimit = oend-(WILDCOPYLENGTH-1); + if (cpy > oend-LASTLITERALS) goto _output_error; /* Error : last LASTLITERALS bytes must be literals (uncompressed) */ + if (op < oCopyLimit) + { + LZ4_wildCopy(op, match, oCopyLimit); + match += oCopyLimit - op; + op = oCopyLimit; + } + while (op +#include +#include +#include +#include + +/* LZ4 comes with its own supposedly portable memory access functions, but they + * seem to be very inefficient in practice (at least on ARM64). Since coreboot + * knows about endinaness and allows some basic assumptions (such as unaligned + * access support), we can easily write the ones we need ourselves. */ +static uint16_t LZ4_readLE16(const void *src) +{ + return read_le16(src); +} +static void LZ4_copy8(void *dst, const void *src) +{ +/* ARM32 needs to be a special snowflake to prevent GCC from coalescing the + * access into LDRD/STRD (which don't support unaligned accesses). */ +#ifdef __arm__ /* ARMv < 6 doesn't support unaligned accesses at all. */ + #if defined(__COREBOOT_ARM_ARCH__) && __COREBOOT_ARM_ARCH__ < 6 + int i; + for (i = 0; i < 8; i++) + ((uint8_t *)dst)[i] = ((uint8_t *)src)[i]; + #else + uint32_t x0, x1; + asm volatile ( + "ldr %[x0], [%[src]]\n\t" + "ldr %[x1], [%[src], #4]\n\t" + "str %[x0], [%[dst]]\n\t" + "str %[x1], [%[dst], #4]\n\t" + : [x0]"=r"(x0), [x1]"=r"(x1) + : [src]"r"(src), [dst]"r"(dst) + : "memory" ); + #endif +#else + *(uint64_t *)dst = *(const uint64_t *)src; +#endif +} + +typedef uint8_t BYTE; +typedef uint16_t U16; +typedef uint32_t U32; +typedef int32_t S32; +typedef uint64_t U64; + +#define FORCE_INLINE static inline __attribute__((always_inline)) +#define likely(expr) __builtin_expect((expr) != 0, 1) +#define unlikely(expr) __builtin_expect((expr) != 0, 0) + +/* Unaltered (just removed unrelated code) from github.com/Cyan4973/lz4/dev. */ +#include "lz4.c.inc" /* #include for inlining, do not link! */ + +#define LZ4F_MAGICNUMBER 0x184D2204 + +struct lz4_frame_header { + uint32_t magic; + union { + uint8_t flags; + struct { + uint8_t reserved0 : 2; + uint8_t has_content_checksum : 1; + uint8_t has_content_size : 1; + uint8_t has_block_checksum : 1; + uint8_t independent_blocks : 1; + uint8_t version : 2; + }; + }; + union { + uint8_t block_descriptor; + struct { + uint8_t reserved1 : 4; + uint8_t max_block_size : 3; + uint8_t reserved2 : 1; + }; + }; + /* + uint64_t content_size iff has_content_size is set */ + /* + uint8_t header_checksum */ +} __attribute__((packed)); + +struct lz4_block_header { + union { + uint32_t raw; + struct { + uint32_t size : 31; + uint32_t not_compressed : 1; + }; + }; + /* + size bytes of data */ + /* + uint32_t block_checksum iff has_block_checksum is set */ +} __attribute__((packed)); + +size_t ulz4fn(const void *src, size_t srcn, void *dst, size_t dstn) +{ + const void *in = src; + void *out = dst; + size_t out_size = 0; + int has_block_checksum; + + { /* With in-place decompression the header may become invalid later. */ + const struct lz4_frame_header *h = in; + + if (srcn < sizeof(*h) + sizeof(uint64_t) + sizeof(uint8_t)) + return 0; /* input overrun */ + + /* We assume there's always only a single, standard frame. */ + if (read_le32(&h->magic) != LZ4F_MAGICNUMBER || h->version != 1) + return 0; /* unknown format */ + if (h->reserved0 || h->reserved1 || h->reserved2) + return 0; /* reserved must be zero */ + if (!h->independent_blocks) + return 0; /* we don't support block dependency */ + has_block_checksum = h->has_block_checksum; + + in += sizeof(*h); + if (h->has_content_size) + in += sizeof(uint64_t); + in += sizeof(uint8_t); + } + + while (1) { + struct lz4_block_header b = { .raw = read_le32(in) }; + in += sizeof(struct lz4_block_header); + + if ((size_t)(in - src) + b.size > srcn) + break; /* input overrun */ + + if (!b.size) { + out_size = out - dst; + break; /* decompression successful */ + } + + if (b.not_compressed) { + size_t size = MIN((uint32_t)b.size, dst + dstn - out); + memcpy(out, in, size); + if (size < b.size) + break; /* output overrun */ + else + out += size; + } else { + /* constant folding essential, do not touch params! */ + int ret = LZ4_decompress_generic(in, out, b.size, + dst + dstn - out, endOnInputSize, + full, 0, noDict, out, NULL, 0); + if (ret < 0) + break; /* decompression error */ + else + out += ret; + } + + in += b.size; + if (has_block_checksum) + in += sizeof(uint32_t); + } + + return out_size; +} + +size_t ulz4f(const void *src, void *dst) +{ + /* LZ4 uses signed size parameters, so can't just use ((u32)-1) here. */ + return ulz4fn(src, 1*GiB, dst, 1*GiB); +} -- cgit v1.2.3