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
   9 /* LZ4 comes with its own supposedly portable memory access functions, but they
  10  * seem to be very inefficient in practice (at least on ARM64). Since coreboot
  11  * knows about endinaness and allows some basic assumptions (such as unaligned
  12  * access support), we can easily write the ones we need ourselves. */
  13 static uint16_t LZ4_readLE16(const void *src)
  14 {
  15         return le16toh(*(const uint16_t *)src);
  16 }
  17 static void LZ4_copy8(void *dst, const void *src)
  18 {
  19 /* ARM32 needs to be a special snowflake to prevent GCC from coalescing the
  20  * access into LDRD/STRD (which don't support unaligned accesses). */
  21 #ifdef __arm__  /* ARMv < 6 doesn't support unaligned accesses at all. */
  22         #if defined(__COREBOOT_ARM_ARCH__) && __COREBOOT_ARM_ARCH__ < 6
  23                 int i;
  24                 for (i = 0; i < 8; i++)
  25                         ((uint8_t *)dst)[i] = ((uint8_t *)src)[i];
  26         #else
  27                 uint32_t x0, x1;
  28                 __asm__ ("ldr %[x0], [%[src]]"
  29                         : [x0]"=r"(x0)
  30                         : [src]"r"(src), "m"(*(const uint32_t *)src));
  31                 __asm__ ("ldr %[x1], [%[src], #4]"
  32                         : [x1]"=r"(x1)
  33                         : [src]"r"(src), "m"(*(const uint32_t *)(src + 4)));
  34                 __asm__ ("str %[x0], [%[dst]]"
  35                         : "=m"(*(uint32_t *)dst)
  36                         : [x0]"r"(x0), [dst]"r"(dst));
  37                 __asm__ ("str %[x1], [%[dst], #4]"
  38                         : "=m"(*(uint32_t *)(dst + 4))
  39                         : [x1]"r"(x1), [dst]"r"(dst));
  40         #endif
  41 #elif defined(__riscv)
  42         /* RISC-V implementations may trap on any unaligned access. */
  43         int i;
  44         for (i = 0; i < 8; i++)
  45                 ((uint8_t *)dst)[i] = ((uint8_t *)src)[i];
  46 #else
  47         *(uint64_t *)dst = *(const uint64_t *)src;
  48 #endif
  49 }
  50
  51 typedef  uint8_t BYTE;
  52 typedef uint16_t U16;
  53 typedef uint32_t U32;
  54 typedef  int32_t S32;
  55 typedef uint64_t U64;
  56
  57 #define FORCE_INLINE static __always_inline
  58 #define likely(expr) __builtin_expect((expr) != 0, 1)
  59 #define unlikely(expr) __builtin_expect((expr) != 0, 0)
  60
  61 /* Unaltered (just removed unrelated code) from github.com/Cyan4973/lz4/dev. */
  62 #include "lz4.c.inc"    /* #include for inlining, do not link! */
  63
  64 #define LZ4F_MAGICNUMBER 0x184D2204
  65
  66 /* Bit field endianness is implementation-defined. Use masks instead.
  67  * https://stackoverflow.com/a/6044223 */
  68 #define RESERVED0               0x03
  69 #define HAS_CONTENT_CHECKSUM    0x04
  70 #define HAS_CONTENT_SIZE        0x08
  71 #define HAS_BLOCK_CHECKSUM      0x10
  72 #define INDEPENDENT_BLOCKS      0x20
  73 #define VERSION                 0xC0
  74 #define VERSION_SHIFT           6
  75
  76 #define RESERVED1_2             0x8F
  77 #define MAX_BLOCK_SIZE          0x70
  78
  79 struct lz4_frame_header {
  80         uint32_t magic;
  81         uint8_t flags;
  82         uint8_t block_descriptor;
  83         /* + uint64_t content_size iff has_content_size is set */
  84         /* + uint8_t header_checksum */
  85 } __packed;
  86
  87 #define BH_SIZE                 0x7FFFFFFF
  88 #define NOT_COMPRESSED          0x80000000
  89
  90 struct lz4_block_header {
  91         uint32_t raw;
  92         /* + size bytes of data */
  93         /* + uint32_t block_checksum iff has_block_checksum is set */
  94 } __packed;
  95
  96 size_t ulz4fn(const void *src, size_t srcn, void *dst, size_t dstn)
  97 {
  98         const void *in = src;
  99         void *out = dst;
 100         size_t out_size = 0;
 101         int has_block_checksum;
 102
 103         { /* With in-place decompression the header may become invalid later. */
 104                 const struct lz4_frame_header *h = in;
 105
 106                 if (srcn < sizeof(*h) + sizeof(uint64_t) + sizeof(uint8_t))
 107                         return 0;       /* input overrun */
 108
 109                 /* We assume there's always only a single, standard frame. */
 110                 if (le32toh(h->magic) != LZ4F_MAGICNUMBER
 111                     || (h->flags & VERSION) != (1 << VERSION_SHIFT))
 112                         return 0;       /* unknown format */
 113                 if ((h->flags & RESERVED0) || (h->block_descriptor & RESERVED1_2))
 114                         return 0;       /* reserved must be zero */
 115                 if (!(h->flags & INDEPENDENT_BLOCKS))
 116                         return 0;       /* we don't support block dependency */
 117                 has_block_checksum = h->flags & HAS_BLOCK_CHECKSUM;
 118
 119                 in += sizeof(*h);
 120                 if (h->flags & HAS_CONTENT_SIZE)
 121                         in += sizeof(uint64_t);
 122                 in += sizeof(uint8_t);
 123         }
 124
 125         while (1) {
 126                 if ((size_t)(in - src) + sizeof(struct lz4_block_header) > srcn)
 127                         break;          /* input overrun */
 128
 129                 struct lz4_block_header b = {
 130                         .raw = le32toh(*(const uint32_t *)in)
 131                 };
 132                 in += sizeof(struct lz4_block_header);
 133
 134                 if ((size_t)(in - src) + (b.raw & BH_SIZE) > srcn)
 135                         break;                  /* input overrun */
 136
 137                 if (!(b.raw & BH_SIZE)) {
 138                         out_size = out - dst;
 139                         break;                  /* decompression successful */
 140                 }
 141
 142                 if (b.raw & NOT_COMPRESSED) {
 143                         size_t size = MIN((uintptr_t)(b.raw & BH_SIZE), (uintptr_t)dst
 144                                 + dstn - (uintptr_t)out);
 145                         memcpy(out, in, size);
 146                         if (size < (b.raw & BH_SIZE))
 147                                 break;          /* output overrun */
 148                         out += size;
 149                 } else {
 150                         /* constant folding essential, do not touch params! */
 151                         int ret = LZ4_decompress_generic(in, out, (b.raw & BH_SIZE),
 152                                         dst + dstn - out, endOnInputSize,
 153                                         full, 0, noDict, out, NULL, 0);
 154                         if (ret < 0)
 155                                 break;          /* decompression error */
 156                         out += ret;
 157                 }
 158
 159                 in += (b.raw & BH_SIZE);
 160                 if (has_block_checksum)
 161                         in += sizeof(uint32_t);
 162         }
 163
 164         return out_size;
 165 }
 166
 167 size_t ulz4f(const void *src, void *dst)
 168 {
 169         /* LZ4 uses signed size parameters, so can't just use ((u32)-1) here. */
 170         return ulz4fn(src, 1*GiB, dst, 1*GiB);
 171 }