[PATCH v15 0/5] riscv: Add fine-tuned checksum functions

[Charlie Jenkins]

Each architecture generally implements fine-tuned checksum functions to
leverage the instruction set. This patch adds the main checksum
functions that are used in networking. Tested on QEMU, this series
allows the CHECKSUM_KUNIT tests to complete an average of 50.9% faster.

This patch takes heavy use of the Zbb extension using alternatives
patching.

To test this patch, enable the configs for KUNIT, then CHECKSUM_KUNIT.

I have attempted to make these functions as optimal as possible, but I
have not ran anything on actual riscv hardware. My performance testing
has been limited to inspecting the assembly, running the algorithms on
x86 hardware, and running in QEMU.

ip_fast_csum is a relatively small function so even though it is
possible to read 64 bits at a time on compatible hardware, the
bottleneck becomes the clean up and setup code so loading 32 bits at a
time is actually faster.

Relies on https://lore.kernel.org/lkml/20230920193801.3035093-1-evan@xxxxxxxxxxxx/

---
    
The algorithm proposed to replace the default csum_fold can be seen to
compute the same result by running all 2^32 possible inputs.
    
static inline unsigned int ror32(unsigned int word, unsigned int shift)
{
	return (word >> (shift & 31)) | (word << ((-shift) & 31));
}

unsigned short csum_fold(unsigned int csum)
{
	unsigned int sum = csum;
	sum = (sum & 0xffff) + (sum >> 16);
	sum = (sum & 0xffff) + (sum >> 16);
	return ~sum;
}

unsigned short csum_fold_arc(unsigned int csum)
{
	return ((~csum - ror32(csum, 16)) >> 16);
}

int main()
{
	unsigned int start = 0x0;
	do {
		if (csum_fold(start) != csum_fold_arc(start)) {
			printf("Not the same %u\n", start);
			return -1;
		}
		start += 1;
	} while(start != 0x0);
	printf("The same\n");
	return 0;
}

Cc: Paul Walmsley <paul.walmsley@xxxxxxxxxx>
Cc: Albert Ou <aou@xxxxxxxxxxxxxxxxx>
Cc: Arnd Bergmann <arnd@xxxxxxxx>
To: Charlie Jenkins <charlie@xxxxxxxxxxxx>
To: Palmer Dabbelt <palmer@xxxxxxxxxxx>
To: Conor Dooley <conor@xxxxxxxxxx>
To: Samuel Holland <samuel.holland@xxxxxxxxxx>
To: David Laight <David.Laight@xxxxxxxxxx>
To: Xiao Wang <xiao.w.wang@xxxxxxxxx>
To: Evan Green <evan@xxxxxxxxxxxx>
To: Guo Ren <guoren@xxxxxxxxxx> 
To: linux-riscv@xxxxxxxxxxxxxxxxxxx
To: linux-kernel@xxxxxxxxxxxxxxx
To: linux-arch@xxxxxxxxxxxxxxx
Signed-off-by: Charlie Jenkins <charlie@xxxxxxxxxxxx>

---
Changes in v15:
- Create modify_unaligned_access_branches to consolidate duplicate code
  (Evan)
- Link to v14: https://lore.kernel.org/r/20231227-optimize_checksum-v14-0-ddfd48016566@xxxxxxxxxxxx

Changes in v14:
- Update misaligned static branch when CPUs are hotplugged (Guo)
- Leave off Evan's reviewed-by on patch 2 since it was completely
  re-written
- Link to v13: https://lore.kernel.org/r/20231220-optimize_checksum-v13-0-a73547e1cad8@xxxxxxxxxxxx

Changes in v13:
- Move cast from patch 4 to patch 3
- Link to v12: https://lore.kernel.org/r/20231212-optimize_checksum-v12-0-419a4ba6d666@xxxxxxxxxxxx

Changes in v12:
- Rebase onto 6.7-rc5
- Add performance stats in the cover letter
- Link to v11: https://lore.kernel.org/r/20231117-optimize_checksum-v11-0-7d9d954fe361@xxxxxxxxxxxx

Changes in v11:
- Extensive modifications to comply to sparse
- Organize include statements (Xiao)
- Add csum_ipv6_magic to commit message (Xiao)
- Remove extraneous len statement (Xiao)
- Add kasan_check_read call (Xiao)
- Improve comment field checksum.h (Xiao)
- Consolidate "buff" and "len" into one parameter "end" (Xiao)
- Link to v10: https://lore.kernel.org/r/20231101-optimize_checksum-v10-0-a498577bb969@xxxxxxxxxxxx

Changes in v10:
- Move tests that were riscv-specific to be arch agnostic (Arnd)
- Link to v9: https://lore.kernel.org/r/20231031-optimize_checksum-v9-0-ea018e69b229@xxxxxxxxxxxx

Changes in v9:
- Use ror64 (Xiao)
- Move do_csum and csum_ipv6_magic headers to patch 4 (Xiao)
- Remove word "IP" from checksum headers (Xiao)
- Swap to using ifndef CONFIG_32BIT instead of ifdef CONFIG_64BIT (Xiao)
- Run no alignment code when buff is aligned (Xiao)
- Consolidate two do_csum implementations overlap into do_csum_common
- Link to v8: https://lore.kernel.org/r/20231027-optimize_checksum-v8-0-feb7101d128d@xxxxxxxxxxxx

Changes in v8:
- Speedups of 12% without Zbb and 21% with Zbb when cpu supports fast
  misaligned accesses for do_csum
- Various formatting updates
- Patch now relies on https://lore.kernel.org/lkml/20230920193801.3035093-1-evan@xxxxxxxxxxxx/
- Link to v7: https://lore.kernel.org/r/20230919-optimize_checksum-v7-0-06c7d0ddd5d6@xxxxxxxxxxxx

Changes in v7:
- Included linux/bitops.h in asm-generic/checksum.h to use ror (Conor)
- Optimized loop in do_csum (David)
- Used ror instead of shifting (David)
- Unfortunately had to reintroduce ifdefs because gcc is not smart
  enough to not throw warnings on code that will never execute
- Use ifdef instead of IS_ENABLED on __LITTLE_ENDIAN because IS_ENABLED
  does not work on that
- Only optimize for zbb when alternatives is enabled in do_csum
- Link to v6: https://lore.kernel.org/r/20230915-optimize_checksum-v6-0-14a6cf61c618@xxxxxxxxxxxx

Changes in v6:
- Fix accuracy of commit message for csum_fold
- Fix indentation
- Link to v5: https://lore.kernel.org/r/20230914-optimize_checksum-v5-0-c95b82a2757e@xxxxxxxxxxxx

Changes in v5:
- Drop vector patches
- Check ZBB enabled before doing any ZBB code (Conor)
- Check endianness in IS_ENABLED
- Revert to the simpler non-tree based version of ipv6_csum_magic since
  David pointed out that the tree based version is not better.
- Link to v4: https://lore.kernel.org/r/20230911-optimize_checksum-v4-0-77cc2ad9e9d7@xxxxxxxxxxxx

Changes in v4:
- Suggestion by David Laight to use an improved checksum used in
  arch/arc.
- Eliminates zero-extension on rv32, but not on rv64.
- Reduces data dependency which should improve execution speed on
  rv32 and rv64
- Still passes CHECKSUM_KUNIT and RISCV_CHECKSUM_KUNIT on rv32 and
  rv64 with and without zbb.
- Link to v3: https://lore.kernel.org/r/20230907-optimize_checksum-v3-0-c502d34d9d73@xxxxxxxxxxxx

Changes in v3:
- Use riscv_has_extension_likely and has_vector where possible (Conor)
- Reduce ifdefs by using IS_ENABLED where possible (Conor)
- Use kernel_vector_begin in the vector code (Samuel)
- Link to v2: https://lore.kernel.org/r/20230905-optimize_checksum-v2-0-ccd658db743b@xxxxxxxxxxxx

Changes in v2:
- After more benchmarking, rework functions to improve performance.
- Remove tests that overlapped with the already existing checksum
  tests and make tests more extensive.
- Use alternatives to activate code with Zbb and vector extensions
- Link to v1: https://lore.kernel.org/r/20230826-optimize_checksum-v1-0-937501b4522a@xxxxxxxxxxxx

---
Charlie Jenkins (5):
      asm-generic: Improve csum_fold
      riscv: Add static key for misaligned accesses
      riscv: Add checksum header
      riscv: Add checksum library
      kunit: Add tests for csum_ipv6_magic and ip_fast_csum

 arch/riscv/include/asm/checksum.h   |  93 ++++++++++
 arch/riscv/include/asm/cpufeature.h |   2 +
 arch/riscv/kernel/cpufeature.c      |  90 +++++++++-
 arch/riscv/lib/Makefile             |   1 +
 arch/riscv/lib/csum.c               | 326 ++++++++++++++++++++++++++++++++++++
 include/asm-generic/checksum.h      |   6 +-
 lib/checksum_kunit.c                | 284 ++++++++++++++++++++++++++++++-
 7 files changed, 795 insertions(+), 7 deletions(-)
---
base-commit: a39b6ac3781d46ba18193c9dbb2110f31e9bffe9
change-id: 20230804-optimize_checksum-db145288ac21
-- 
- Charlie