| 1 | ================================================
|
| 2 | Generic bitfield packing and unpacking functions
|
| 3 | ================================================
|
| 4 |
|
| 5 | Problem statement
|
| 6 | -----------------
|
| 7 |
|
| 8 | When working with hardware, one has to choose between several approaches of
|
| 9 | interfacing with it.
|
| 10 | One can memory-map a pointer to a carefully crafted struct over the hardware
|
| 11 | device's memory region, and access its fields as struct members (potentially
|
| 12 | declared as bitfields). But writing code this way would make it less portable,
|
| 13 | due to potential endianness mismatches between the CPU and the hardware device.
|
| 14 | Additionally, one has to pay close attention when translating register
|
| 15 | definitions from the hardware documentation into bit field indices for the
|
| 16 | structs. Also, some hardware (typically networking equipment) tends to group
|
| 17 | its register fields in ways that violate any reasonable word boundaries
|
| 18 | (sometimes even 64 bit ones). This creates the inconvenience of having to
|
| 19 | define "high" and "low" portions of register fields within the struct.
|
| 20 | A more robust alternative to struct field definitions would be to extract the
|
| 21 | required fields by shifting the appropriate number of bits. But this would
|
| 22 | still not protect from endianness mismatches, except if all memory accesses
|
| 23 | were performed byte-by-byte. Also the code can easily get cluttered, and the
|
| 24 | high-level idea might get lost among the many bit shifts required.
|
| 25 | Many drivers take the bit-shifting approach and then attempt to reduce the
|
| 26 | clutter with tailored macros, but more often than not these macros take
|
| 27 | shortcuts that still prevent the code from being truly portable.
|
| 28 |
|
| 29 | The solution
|
| 30 | ------------
|
| 31 |
|
| 32 | This API deals with 2 basic operations:
|
| 33 |
|
| 34 | - Packing a CPU-usable number into a memory buffer (with hardware
|
| 35 | constraints/quirks)
|
| 36 | - Unpacking a memory buffer (which has hardware constraints/quirks)
|
| 37 | into a CPU-usable number.
|
| 38 |
|
| 39 | The API offers an abstraction over said hardware constraints and quirks,
|
| 40 | over CPU endianness and therefore between possible mismatches between
|
| 41 | the two.
|
| 42 |
|
| 43 | The basic unit of these API functions is the u64. From the CPU's
|
| 44 | perspective, bit 63 always means bit offset 7 of byte 7, albeit only
|
| 45 | logically. The question is: where do we lay this bit out in memory?
|
| 46 |
|
| 47 | The following examples cover the memory layout of a packed u64 field.
|
| 48 | The byte offsets in the packed buffer are always implicitly 0, 1, ... 7.
|
| 49 | What the examples show is where the logical bytes and bits sit.
|
| 50 |
|
| 51 | 1. Normally (no quirks), we would do it like this:
|
| 52 |
|
| 53 | ::
|
| 54 |
|
| 55 | 63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32
|
| 56 | 7 6 5 4
|
| 57 | 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
|
| 58 | 3 2 1 0
|
| 59 |
|
| 60 | That is, the MSByte (7) of the CPU-usable u64 sits at memory offset 0, and the
|
| 61 | LSByte (0) of the u64 sits at memory offset 7.
|
| 62 | This corresponds to what most folks would regard to as "big endian", where
|
| 63 | bit i corresponds to the number 2^i. This is also referred to in the code
|
| 64 | comments as "logical" notation.
|
| 65 |
|
| 66 |
|
| 67 | 2. If QUIRK_MSB_ON_THE_RIGHT is set, we do it like this:
|
| 68 |
|
| 69 | ::
|
| 70 |
|
| 71 | 56 57 58 59 60 61 62 63 48 49 50 51 52 53 54 55 40 41 42 43 44 45 46 47 32 33 34 35 36 37 38 39
|
| 72 | 7 6 5 4
|
| 73 | 24 25 26 27 28 29 30 31 16 17 18 19 20 21 22 23 8 9 10 11 12 13 14 15 0 1 2 3 4 5 6 7
|
| 74 | 3 2 1 0
|
| 75 |
|
| 76 | That is, QUIRK_MSB_ON_THE_RIGHT does not affect byte positioning, but
|
| 77 | inverts bit offsets inside a byte.
|
| 78 |
|
| 79 |
|
| 80 | 3. If QUIRK_LITTLE_ENDIAN is set, we do it like this:
|
| 81 |
|
| 82 | ::
|
| 83 |
|
| 84 | 39 38 37 36 35 34 33 32 47 46 45 44 43 42 41 40 55 54 53 52 51 50 49 48 63 62 61 60 59 58 57 56
|
| 85 | 4 5 6 7
|
| 86 | 7 6 5 4 3 2 1 0 15 14 13 12 11 10 9 8 23 22 21 20 19 18 17 16 31 30 29 28 27 26 25 24
|
| 87 | 0 1 2 3
|
| 88 |
|
| 89 | Therefore, QUIRK_LITTLE_ENDIAN means that inside the memory region, every
|
| 90 | byte from each 4-byte word is placed at its mirrored position compared to
|
| 91 | the boundary of that word.
|
| 92 |
|
| 93 | 4. If QUIRK_MSB_ON_THE_RIGHT and QUIRK_LITTLE_ENDIAN are both set, we do it
|
| 94 | like this:
|
| 95 |
|
| 96 | ::
|
| 97 |
|
| 98 | 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63
|
| 99 | 4 5 6 7
|
| 100 | 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
|
| 101 | 0 1 2 3
|
| 102 |
|
| 103 |
|
| 104 | 5. If just QUIRK_LSW32_IS_FIRST is set, we do it like this:
|
| 105 |
|
| 106 | ::
|
| 107 |
|
| 108 | 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
|
| 109 | 3 2 1 0
|
| 110 | 63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32
|
| 111 | 7 6 5 4
|
| 112 |
|
| 113 | In this case the 8 byte memory region is interpreted as follows: first
|
| 114 | 4 bytes correspond to the least significant 4-byte word, next 4 bytes to
|
| 115 | the more significant 4-byte word.
|
| 116 |
|
| 117 |
|
| 118 | 6. If QUIRK_LSW32_IS_FIRST and QUIRK_MSB_ON_THE_RIGHT are set, we do it like
|
| 119 | this:
|
| 120 |
|
| 121 | ::
|
| 122 |
|
| 123 | 24 25 26 27 28 29 30 31 16 17 18 19 20 21 22 23 8 9 10 11 12 13 14 15 0 1 2 3 4 5 6 7
|
| 124 | 3 2 1 0
|
| 125 | 56 57 58 59 60 61 62 63 48 49 50 51 52 53 54 55 40 41 42 43 44 45 46 47 32 33 34 35 36 37 38 39
|
| 126 | 7 6 5 4
|
| 127 |
|
| 128 |
|
| 129 | 7. If QUIRK_LSW32_IS_FIRST and QUIRK_LITTLE_ENDIAN are set, it looks like
|
| 130 | this:
|
| 131 |
|
| 132 | ::
|
| 133 |
|
| 134 | 7 6 5 4 3 2 1 0 15 14 13 12 11 10 9 8 23 22 21 20 19 18 17 16 31 30 29 28 27 26 25 24
|
| 135 | 0 1 2 3
|
| 136 | 39 38 37 36 35 34 33 32 47 46 45 44 43 42 41 40 55 54 53 52 51 50 49 48 63 62 61 60 59 58 57 56
|
| 137 | 4 5 6 7
|
| 138 |
|
| 139 |
|
| 140 | 8. If QUIRK_LSW32_IS_FIRST, QUIRK_LITTLE_ENDIAN and QUIRK_MSB_ON_THE_RIGHT
|
| 141 | are set, it looks like this:
|
| 142 |
|
| 143 | ::
|
| 144 |
|
| 145 | 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
|
| 146 | 0 1 2 3
|
| 147 | 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63
|
| 148 | 4 5 6 7
|
| 149 |
|
| 150 |
|
| 151 | We always think of our offsets as if there were no quirk, and we translate
|
| 152 | them afterwards, before accessing the memory region.
|
| 153 |
|
| 154 | Note on buffer lengths not multiple of 4
|
| 155 | ----------------------------------------
|
| 156 |
|
| 157 | To deal with memory layout quirks where groups of 4 bytes are laid out "little
|
| 158 | endian" relative to each other, but "big endian" within the group itself, the
|
| 159 | concept of groups of 4 bytes is intrinsic to the packing API (not to be
|
| 160 | confused with the memory access, which is performed byte by byte, though).
|
| 161 |
|
| 162 | With buffer lengths not multiple of 4, this means one group will be incomplete.
|
| 163 | Depending on the quirks, this may lead to discontinuities in the bit fields
|
| 164 | accessible through the buffer. The packing API assumes discontinuities were not
|
| 165 | the intention of the memory layout, so it avoids them by effectively logically
|
| 166 | shortening the most significant group of 4 octets to the number of octets
|
| 167 | actually available.
|
| 168 |
|
| 169 | Example with a 31 byte sized buffer given below. Physical buffer offsets are
|
| 170 | implicit, and increase from left to right within a group, and from top to
|
| 171 | bottom within a column.
|
| 172 |
|
| 173 | No quirks:
|
| 174 |
|
| 175 | ::
|
| 176 |
|
| 177 | 31 29 28 | Group 7 (most significant)
|
| 178 | 27 26 25 24 | Group 6
|
| 179 | 23 22 21 20 | Group 5
|
| 180 | 19 18 17 16 | Group 4
|
| 181 | 15 14 13 12 | Group 3
|
| 182 | 11 10 9 8 | Group 2
|
| 183 | 7 6 5 4 | Group 1
|
| 184 | 3 2 1 0 | Group 0 (least significant)
|
| 185 |
|
| 186 | QUIRK_LSW32_IS_FIRST:
|
| 187 |
|
| 188 | ::
|
| 189 |
|
| 190 | 3 2 1 0 | Group 0 (least significant)
|
| 191 | 7 6 5 4 | Group 1
|
| 192 | 11 10 9 8 | Group 2
|
| 193 | 15 14 13 12 | Group 3
|
| 194 | 19 18 17 16 | Group 4
|
| 195 | 23 22 21 20 | Group 5
|
| 196 | 27 26 25 24 | Group 6
|
| 197 | 30 29 28 | Group 7 (most significant)
|
| 198 |
|
| 199 | QUIRK_LITTLE_ENDIAN:
|
| 200 |
|
| 201 | ::
|
| 202 |
|
| 203 | 30 28 29 | Group 7 (most significant)
|
| 204 | 24 25 26 27 | Group 6
|
| 205 | 20 21 22 23 | Group 5
|
| 206 | 16 17 18 19 | Group 4
|
| 207 | 12 13 14 15 | Group 3
|
| 208 | 8 9 10 11 | Group 2
|
| 209 | 4 5 6 7 | Group 1
|
| 210 | 0 1 2 3 | Group 0 (least significant)
|
| 211 |
|
| 212 | QUIRK_LITTLE_ENDIAN | QUIRK_LSW32_IS_FIRST:
|
| 213 |
|
| 214 | ::
|
| 215 |
|
| 216 | 0 1 2 3 | Group 0 (least significant)
|
| 217 | 4 5 6 7 | Group 1
|
| 218 | 8 9 10 11 | Group 2
|
| 219 | 12 13 14 15 | Group 3
|
| 220 | 16 17 18 19 | Group 4
|
| 221 | 20 21 22 23 | Group 5
|
| 222 | 24 25 26 27 | Group 6
|
| 223 | 28 29 30 | Group 7 (most significant)
|
| 224 |
|
| 225 | Intended use
|
| 226 | ------------
|
| 227 |
|
| 228 | Drivers that opt to use this API first need to identify which of the above 3
|
| 229 | quirk combinations (for a total of 8) match what the hardware documentation
|
| 230 | describes.
|
| 231 |
|
| 232 | There are 3 supported usage patterns, detailed below.
|
| 233 |
|
| 234 | packing()
|
| 235 | ^^^^^^^^^
|
| 236 |
|
| 237 | This API function is deprecated.
|
| 238 |
|
| 239 | The packing() function returns an int-encoded error code, which protects the
|
| 240 | programmer against incorrect API use. The errors are not expected to occur
|
| 241 | during runtime, therefore it is reasonable to wrap packing() into a custom
|
| 242 | function which returns void and swallows those errors. Optionally it can
|
| 243 | dump stack or print the error description.
|
| 244 |
|
| 245 | .. code-block:: c
|
| 246 |
|
| 247 | void my_packing(void *buf, u64 *val, int startbit, int endbit,
|
| 248 | size_t len, enum packing_op op)
|
| 249 | {
|
| 250 | int err;
|
| 251 |
|
| 252 | /* Adjust quirks accordingly */
|
| 253 | err = packing(buf, val, startbit, endbit, len, op, QUIRK_LSW32_IS_FIRST);
|
| 254 | if (likely(!err))
|
| 255 | return;
|
| 256 |
|
| 257 | if (err == -EINVAL) {
|
| 258 | pr_err("Start bit (%d) expected to be larger than end (%d)\n",
|
| 259 | startbit, endbit);
|
| 260 | } else if (err == -ERANGE) {
|
| 261 | if ((startbit - endbit + 1) > 64)
|
| 262 | pr_err("Field %d-%d too large for 64 bits!\n",
|
| 263 | startbit, endbit);
|
| 264 | else
|
| 265 | pr_err("Cannot store %llx inside bits %d-%d (would truncate)\n",
|
| 266 | *val, startbit, endbit);
|
| 267 | }
|
| 268 | dump_stack();
|
| 269 | }
|
| 270 |
|
| 271 | pack() and unpack()
|
| 272 | ^^^^^^^^^^^^^^^^^^^
|
| 273 |
|
| 274 | These are const-correct variants of packing(), and eliminate the last "enum
|
| 275 | packing_op op" argument.
|
| 276 |
|
| 277 | Calling pack(...) is equivalent, and preferred, to calling packing(..., PACK).
|
| 278 |
|
| 279 | Calling unpack(...) is equivalent, and preferred, to calling packing(..., UNPACK).
|
| 280 |
|
| 281 | pack_fields() and unpack_fields()
|
| 282 | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
| 283 |
|
| 284 | The library exposes optimized functions for the scenario where there are many
|
| 285 | fields represented in a buffer, and it encourages consumer drivers to avoid
|
| 286 | repetitive calls to pack() and unpack() for each field, but instead use
|
| 287 | pack_fields() and unpack_fields(), which reduces the code footprint.
|
| 288 |
|
| 289 | These APIs use field definitions in arrays of ``struct packed_field_u8`` or
|
| 290 | ``struct packed_field_u16``, allowing consumer drivers to minimize the size
|
| 291 | of these arrays according to their custom requirements.
|
| 292 |
|
| 293 | The pack_fields() and unpack_fields() API functions are actually macros which
|
| 294 | automatically select the appropriate function at compile time, based on the
|
| 295 | type of the fields array passed in.
|
| 296 |
|
| 297 | An additional benefit over pack() and unpack() is that sanity checks on the
|
| 298 | field definitions are handled at compile time with ``BUILD_BUG_ON`` rather
|
| 299 | than only when the offending code is executed. These functions return void and
|
| 300 | wrapping them to handle unexpected errors is not necessary.
|
| 301 |
|
| 302 | It is recommended, but not required, that you wrap your packed buffer into a
|
| 303 | structured type with a fixed size. This generally makes it easier for the
|
| 304 | compiler to enforce that the correct size buffer is used.
|
| 305 |
|
| 306 | Here is an example of how to use the fields APIs:
|
| 307 |
|
| 308 | .. code-block:: c
|
| 309 |
|
| 310 | /* Ordering inside the unpacked structure is flexible and can be different
|
| 311 | * from the packed buffer. Here, it is optimized to reduce padding.
|
| 312 | */
|
| 313 | struct data {
|
| 314 | u64 field3;
|
| 315 | u32 field4;
|
| 316 | u16 field1;
|
| 317 | u8 field2;
|
| 318 | };
|
| 319 |
|
| 320 | #define SIZE 13
|
| 321 |
|
| 322 | typedef struct __packed { u8 buf[SIZE]; } packed_buf_t;
|
| 323 |
|
| 324 | static const struct packed_field_u8 fields[] = {
|
| 325 | PACKED_FIELD(100, 90, struct data, field1),
|
| 326 | PACKED_FIELD(90, 87, struct data, field2),
|
| 327 | PACKED_FIELD(86, 30, struct data, field3),
|
| 328 | PACKED_FIELD(29, 0, struct data, field4),
|
| 329 | };
|
| 330 |
|
| 331 | void unpack_your_data(const packed_buf_t *buf, struct data *unpacked)
|
| 332 | {
|
| 333 | BUILD_BUG_ON(sizeof(*buf) != SIZE);
|
| 334 |
|
| 335 | unpack_fields(buf, sizeof(*buf), unpacked, fields,
|
| 336 | QUIRK_LITTLE_ENDIAN);
|
| 337 | }
|
| 338 |
|
| 339 | void pack_your_data(const struct data *unpacked, packed_buf_t *buf)
|
| 340 | {
|
| 341 | BUILD_BUG_ON(sizeof(*buf) != SIZE);
|
| 342 |
|
| 343 | pack_fields(buf, sizeof(*buf), unpacked, fields,
|
| 344 | QUIRK_LITTLE_ENDIAN);
|
| 345 | }
|