linux/Documentation/core-api/packing.rst
Vladimir Oltean a636ba5e86 lib: packing: adjust definitions and implementation for arbitrary buffer lengths
Jacob Keller has a use case for packing() in the intel/ice networking
driver, but it cannot be used as-is.

Simply put, the API quirks for LSW32_IS_FIRST and LITTLE_ENDIAN are
naively implemented with the undocumented assumption that the buffer
length must be a multiple of 4. All calculations of group offsets and
offsets of bytes within groups assume that this is the case. But in the
ice case, this does not hold true. For example, packing into a buffer
of 22 bytes would yield wrong results, but pretending it was a 24 byte
buffer would work.

Rather than requiring such hacks, and leaving a big question mark when
it comes to discontinuities in the accessible bit fields of such buffer,
we should extend the packing API to support this use case.

It turns out that we can keep the design in terms of groups of 4 bytes,
but also make it work if the total length is not a multiple of 4.
Just like before, imagine the buffer as a big number, and its most
significant bytes (the ones that would make up to a multiple of 4) are
missing. Thus, with a big endian (no quirks) interpretation of the
buffer, those most significant bytes would be absent from the beginning
of the buffer, and with a LSW32_IS_FIRST interpretation, they would be
absent from the end of the buffer. The LITTLE_ENDIAN quirk, in the
packing() API world, only affects byte ordering within groups of 4.
Thus, it does not change which bytes are missing. Only the significance
of the remaining bytes within the (smaller) group.

No change intended for buffer sizes which are multiples of 4. Tested
with the sja1105 driver and with downstream unit tests.

Link: https://lore.kernel.org/netdev/a0338310-e66c-497c-bc1f-a597e50aa3ff@intel.com/
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Tested-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://patch.msgid.link/20241002-packing-kunit-tests-and-split-pack-unpack-v2-2-8373e551eae3@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2024-10-03 15:32:03 -07:00

238 lines
10 KiB
ReStructuredText

================================================
Generic bitfield packing and unpacking functions
================================================
Problem statement
-----------------
When working with hardware, one has to choose between several approaches of
interfacing with it.
One can memory-map a pointer to a carefully crafted struct over the hardware
device's memory region, and access its fields as struct members (potentially
declared as bitfields). But writing code this way would make it less portable,
due to potential endianness mismatches between the CPU and the hardware device.
Additionally, one has to pay close attention when translating register
definitions from the hardware documentation into bit field indices for the
structs. Also, some hardware (typically networking equipment) tends to group
its register fields in ways that violate any reasonable word boundaries
(sometimes even 64 bit ones). This creates the inconvenience of having to
define "high" and "low" portions of register fields within the struct.
A more robust alternative to struct field definitions would be to extract the
required fields by shifting the appropriate number of bits. But this would
still not protect from endianness mismatches, except if all memory accesses
were performed byte-by-byte. Also the code can easily get cluttered, and the
high-level idea might get lost among the many bit shifts required.
Many drivers take the bit-shifting approach and then attempt to reduce the
clutter with tailored macros, but more often than not these macros take
shortcuts that still prevent the code from being truly portable.
The solution
------------
This API deals with 2 basic operations:
- Packing a CPU-usable number into a memory buffer (with hardware
constraints/quirks)
- Unpacking a memory buffer (which has hardware constraints/quirks)
into a CPU-usable number.
The API offers an abstraction over said hardware constraints and quirks,
over CPU endianness and therefore between possible mismatches between
the two.
The basic unit of these API functions is the u64. From the CPU's
perspective, bit 63 always means bit offset 7 of byte 7, albeit only
logically. The question is: where do we lay this bit out in memory?
The following examples cover the memory layout of a packed u64 field.
The byte offsets in the packed buffer are always implicitly 0, 1, ... 7.
What the examples show is where the logical bytes and bits sit.
1. Normally (no quirks), we would do it like this:
::
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
7 6 5 4
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
3 2 1 0
That is, the MSByte (7) of the CPU-usable u64 sits at memory offset 0, and the
LSByte (0) of the u64 sits at memory offset 7.
This corresponds to what most folks would regard to as "big endian", where
bit i corresponds to the number 2^i. This is also referred to in the code
comments as "logical" notation.
2. If QUIRK_MSB_ON_THE_RIGHT is set, we do it like this:
::
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
7 6 5 4
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
3 2 1 0
That is, QUIRK_MSB_ON_THE_RIGHT does not affect byte positioning, but
inverts bit offsets inside a byte.
3. If QUIRK_LITTLE_ENDIAN is set, we do it like this:
::
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
4 5 6 7
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
0 1 2 3
Therefore, QUIRK_LITTLE_ENDIAN means that inside the memory region, every
byte from each 4-byte word is placed at its mirrored position compared to
the boundary of that word.
4. If QUIRK_MSB_ON_THE_RIGHT and QUIRK_LITTLE_ENDIAN are both set, we do it
like this:
::
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
4 5 6 7
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
0 1 2 3
5. If just QUIRK_LSW32_IS_FIRST is set, we do it like this:
::
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
3 2 1 0
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
7 6 5 4
In this case the 8 byte memory region is interpreted as follows: first
4 bytes correspond to the least significant 4-byte word, next 4 bytes to
the more significant 4-byte word.
6. If QUIRK_LSW32_IS_FIRST and QUIRK_MSB_ON_THE_RIGHT are set, we do it like
this:
::
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
3 2 1 0
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
7 6 5 4
7. If QUIRK_LSW32_IS_FIRST and QUIRK_LITTLE_ENDIAN are set, it looks like
this:
::
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
0 1 2 3
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
4 5 6 7
8. If QUIRK_LSW32_IS_FIRST, QUIRK_LITTLE_ENDIAN and QUIRK_MSB_ON_THE_RIGHT
are set, it looks like this:
::
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
0 1 2 3
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
4 5 6 7
We always think of our offsets as if there were no quirk, and we translate
them afterwards, before accessing the memory region.
Note on buffer lengths not multiple of 4
----------------------------------------
To deal with memory layout quirks where groups of 4 bytes are laid out "little
endian" relative to each other, but "big endian" within the group itself, the
concept of groups of 4 bytes is intrinsic to the packing API (not to be
confused with the memory access, which is performed byte by byte, though).
With buffer lengths not multiple of 4, this means one group will be incomplete.
Depending on the quirks, this may lead to discontinuities in the bit fields
accessible through the buffer. The packing API assumes discontinuities were not
the intention of the memory layout, so it avoids them by effectively logically
shortening the most significant group of 4 octets to the number of octets
actually available.
Example with a 31 byte sized buffer given below. Physical buffer offsets are
implicit, and increase from left to right within a group, and from top to
bottom within a column.
No quirks:
::
31 29 28 | Group 7 (most significant)
27 26 25 24 | Group 6
23 22 21 20 | Group 5
19 18 17 16 | Group 4
15 14 13 12 | Group 3
11 10 9 8 | Group 2
7 6 5 4 | Group 1
3 2 1 0 | Group 0 (least significant)
QUIRK_LSW32_IS_FIRST:
::
3 2 1 0 | Group 0 (least significant)
7 6 5 4 | Group 1
11 10 9 8 | Group 2
15 14 13 12 | Group 3
19 18 17 16 | Group 4
23 22 21 20 | Group 5
27 26 25 24 | Group 6
30 29 28 | Group 7 (most significant)
QUIRK_LITTLE_ENDIAN:
::
30 28 29 | Group 7 (most significant)
24 25 26 27 | Group 6
20 21 22 23 | Group 5
16 17 18 19 | Group 4
12 13 14 15 | Group 3
8 9 10 11 | Group 2
4 5 6 7 | Group 1
0 1 2 3 | Group 0 (least significant)
QUIRK_LITTLE_ENDIAN | QUIRK_LSW32_IS_FIRST:
::
0 1 2 3 | Group 0 (least significant)
4 5 6 7 | Group 1
8 9 10 11 | Group 2
12 13 14 15 | Group 3
16 17 18 19 | Group 4
20 21 22 23 | Group 5
24 25 26 27 | Group 6
28 29 30 | Group 7 (most significant)
Intended use
------------
Drivers that opt to use this API first need to identify which of the above 3
quirk combinations (for a total of 8) match what the hardware documentation
describes. Then they should wrap the packing() function, creating a new
xxx_packing() that calls it using the proper QUIRK_* one-hot bits set.
The packing() function returns an int-encoded error code, which protects the
programmer against incorrect API use. The errors are not expected to occur
during runtime, therefore it is reasonable for xxx_packing() to return void
and simply swallow those errors. Optionally it can dump stack or print the
error description.