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 Intended use 155 ------------ 156 157 Drivers that opt to use this API first need to identify which of the above 3 158 quirk combinations (for a total of 8) match what the hardware documentation 159 describes. Then they should wrap the packing() function, creating a new 160 xxx_packing() that calls it using the proper QUIRK_* one-hot bits set. 161 162 The packing() function returns an int-encoded error code, which protects the 163 programmer against incorrect API use. The errors are not expected to occur 164 during runtime, therefore it is reasonable for xxx_packing() to return void 165 and simply swallow those errors. Optionally it can dump stack or print the 166 error description.
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.