src/commonlib/bsd/lz4_wrapper.c

   1 /* SPDX-License-Identifier: BSD-3-Clause OR GPL-2.0-only */
   2
   3 #include <commonlib/bsd/compression.h>
   4 #include <commonlib/bsd/helpers.h>
   5 #include <commonlib/bsd/sysincludes.h>
   6 #include <stdint.h>
   7 #include <string.h>
   8 #include <endian.h>
   9
  10 /*
  11  * RISC-V and older ARM architectures do not mandate support for misaligned access.
  12  * Our le16toh and friends functions assume misaligned access support. Writing the access
  13  * like this causes the compiler to generate instructions using misaligned access (or not)
  14  * depending on the architecture. So there is no performance penalty for platforms supporting
  15  * misaligned access.
  16  */
  17 static uint16_t LZ4_readLE16(const void *src)
  18 {
  19         return *((const uint8_t *)src + 1) << 8
  20               | *(const uint8_t *)src;
  21 }
  22
  23 static uint32_t LZ4_readLE32(const void *src)
  24 {
  25         return *((const uint8_t *)src + 3) << 24
  26              | *((const uint8_t *)src + 2) << 16
  27              | *((const uint8_t *)src + 1) << 8
  28              |  *(const uint8_t *)src;
  29 }
  30
  31 static void LZ4_copy8(void *dst, const void *src)
  32 {
  33 /* ARM32 needs to be a special snowflake to prevent GCC from coalescing the
  34  * access into LDRD/STRD (which don't support unaligned accesses). */
  35 #ifdef __arm__  /* ARMv < 6 doesn't support unaligned accesses at all. */
  36         #if defined(__COREBOOT_ARM_ARCH__) && __COREBOOT_ARM_ARCH__ < 6
  37                 int i;
  38                 for (i = 0; i < 8; i++)
  39                         ((uint8_t *)dst)[i] = ((uint8_t *)src)[i];
  40         #else
  41                 uint32_t x0, x1;
  42                 __asm__ ("ldr %[x0], [%[src]]"
  43                         : [x0]"=r"(x0)
  44                         : [src]"r"(src), "m"(*(const uint32_t *)src));
  45                 __asm__ ("ldr %[x1], [%[src], #4]"
  46                         : [x1]"=r"(x1)
  47                         : [src]"r"(src), "m"(*(const uint32_t *)(src + 4)));
  48                 __asm__ ("str %[x0], [%[dst]]"
  49                         : "=m"(*(uint32_t *)dst)
  50                         : [x0]"r"(x0), [dst]"r"(dst));
  51                 __asm__ ("str %[x1], [%[dst], #4]"
  52                         : "=m"(*(uint32_t *)(dst + 4))
  53                         : [x1]"r"(x1), [dst]"r"(dst));
  54         #endif
  55 #elif defined(__riscv)
  56         /* RISC-V implementations may trap on any unaligned access. */
  57         int i;
  58         for (i = 0; i < 8; i++)
  59                 ((uint8_t *)dst)[i] = ((uint8_t *)src)[i];
  60 #else
  61         *(uint64_t *)dst = *(const uint64_t *)src;
  62 #endif
  63 }
  64
  65 typedef  uint8_t BYTE;
  66 typedef uint16_t U16;
  67 typedef uint32_t U32;
  68 typedef  int32_t S32;
  69 typedef uint64_t U64;
  70
  71 #define FORCE_INLINE static __always_inline
  72 #define likely(expr) __builtin_expect((expr) != 0, 1)
  73 #define unlikely(expr) __builtin_expect((expr) != 0, 0)
  74
  75 /* Unaltered (just removed unrelated code) from github.com/Cyan4973/lz4/dev. */
  76 #include "lz4.c.inc"    /* #include for inlining, do not link! */
  77
  78 #define LZ4F_MAGICNUMBER 0x184D2204
  79
  80 /* Bit field endianness is implementation-defined. Use masks instead.
  81  * https://stackoverflow.com/a/6044223 */
  82 #define RESERVED0               0x03
  83 #define HAS_CONTENT_CHECKSUM    0x04
  84 #define HAS_CONTENT_SIZE        0x08
  85 #define HAS_BLOCK_CHECKSUM      0x10
  86 #define INDEPENDENT_BLOCKS      0x20
  87 #define VERSION                 0xC0
  88 #define VERSION_SHIFT           6
  89
  90 #define RESERVED1_2             0x8F
  91 #define MAX_BLOCK_SIZE          0x70
  92
  93 struct lz4_frame_header {
  94         uint32_t magic;
  95         uint8_t flags;
  96         uint8_t block_descriptor;
  97         /* + uint64_t content_size iff has_content_size is set */
  98         /* + uint8_t header_checksum */
  99 } __packed;
 100
 101 #define BH_SIZE                 0x7FFFFFFF
 102 #define NOT_COMPRESSED          0x80000000
 103
 104 struct lz4_block_header {
 105         uint32_t raw;
 106         /* + size bytes of data */
 107         /* + uint32_t block_checksum iff has_block_checksum is set */
 108 } __packed;
 109
 110 size_t ulz4fn(const void *src, size_t srcn, void *dst, size_t dstn)
 111 {
 112         const void *in = src;
 113         void *out = dst;
 114         size_t out_size = 0;
 115         int has_block_checksum;
 116
 117         { /* With in-place decompression the header may become invalid later. */
 118                 const struct lz4_frame_header *h = in;
 119
 120                 if (srcn < sizeof(*h) + sizeof(uint64_t) + sizeof(uint8_t))
 121                         return 0;       /* input overrun */
 122
 123                 /* We assume there's always only a single, standard frame. */
 124                 if (LZ4_readLE32(&h->magic) != LZ4F_MAGICNUMBER
 125                     || (h->flags & VERSION) != (1 << VERSION_SHIFT))
 126                         return 0;       /* unknown format */
 127                 if ((h->flags & RESERVED0) || (h->block_descriptor & RESERVED1_2))
 128                         return 0;       /* reserved must be zero */
 129                 if (!(h->flags & INDEPENDENT_BLOCKS))
 130                         return 0;       /* we don't support block dependency */
 131                 has_block_checksum = h->flags & HAS_BLOCK_CHECKSUM;
 132
 133                 in += sizeof(*h);
 134                 if (h->flags & HAS_CONTENT_SIZE)
 135                         in += sizeof(uint64_t);
 136                 in += sizeof(uint8_t);
 137         }
 138
 139         while (1) {
 140                 if ((size_t)(in - src) + sizeof(struct lz4_block_header) > srcn)
 141                         break;          /* input overrun */
 142
 143                 struct lz4_block_header b = {
 144                         .raw = LZ4_readLE32((const uint32_t *)in)
 145                 };
 146                 in += sizeof(struct lz4_block_header);
 147
 148                 if ((size_t)(in - src) + (b.raw & BH_SIZE) > srcn)
 149                         break;                  /* input overrun */
 150
 151                 if (!(b.raw & BH_SIZE)) {
 152                         out_size = out - dst;
 153                         break;                  /* decompression successful */
 154                 }
 155
 156                 if (b.raw & NOT_COMPRESSED) {
 157                         size_t size = MIN((uintptr_t)(b.raw & BH_SIZE), (uintptr_t)dst
 158                                 + dstn - (uintptr_t)out);
 159                         memcpy(out, in, size);
 160                         if (size < (b.raw & BH_SIZE))
 161                                 break;          /* output overrun */
 162                         out += size;
 163                 } else {
 164                         /* constant folding essential, do not touch params! */
 165                         int ret = LZ4_decompress_generic(in, out, (b.raw & BH_SIZE),
 166                                         dst + dstn - out, endOnInputSize,
 167                                         full, 0, noDict, out, NULL, 0);
 168                         if (ret < 0)
 169                                 break;          /* decompression error */
 170                         out += ret;
 171                 }
 172
 173                 in += (b.raw & BH_SIZE);
 174                 if (has_block_checksum)
 175                         in += sizeof(uint32_t);
 176         }
 177
 178         return out_size;
 179 }
 180
 181 size_t ulz4f(const void *src, void *dst)
 182 {
 183         /* LZ4 uses signed size parameters, so can't just use ((u32)-1) here. */
 184         return ulz4fn(src, 1*GiB, dst, 1*GiB);
 185 }