linux-kernel - [PATCH 0/3] riscv: word-at-a-time: improve find

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Message-ID: <20260113122457.27507-1-jszhang@kernel.org>
Date: Tue, 13 Jan 2026 20:24:54 +0800
From: Jisheng Zhang <jszhang@...nel.org>
To: Paul Walmsley <pjw@...nel.org>,
	Palmer Dabbelt <palmer@...belt.com>,
	Albert Ou <aou@...s.berkeley.edu>,
	Alexandre Ghiti <alex@...ti.fr>
Cc: linux-riscv@...ts.infradead.org,
	linux-kernel@...r.kernel.org
Subject: [PATCH 0/3] riscv: word-at-a-time: improve find_zero()

Currently, there are two problems with riscv find_zero():

1. When !RISCV_ISA_ZBB, the generic fls64() bring non-optimal code.

But in word-at-a-time case, we don't have to go with fls64() code path,
instead, we can fallback to the generic word-at-a-time implementaion.

What's more, the fls64() brings non-necessary zero bits couting for
RV32. In fact, fls() is enough.

2. Similar as 1, the generic fls64() also brings non-optimal code when
RISCV_ISA_ZBB=y but HW doesn't support Zbb.

So this series tries to improve find_zero() by falling back to generic
word-at-a-time implementaion where necessary. We dramatically reduce
the instructions of find_zero() from 33 to 8! Also testing with the
micro-benchamrk in patch1 shows that the performance is improved by
about 1150%!

After that, we improve find_zero() for Zbb further by applying similar
optimization as Linus did in commit f915a3e5b018 ("arm64:
word-at-a-time: improve byte count calculations for LE"), so that
we share the similar improvements:

"The difference between the old and the new implementation is that
"count_zero()" ends up scheduling better because it is being done on a
value that is available earlier (before the final mask).

But more importantly, it can be implemented without the insane semantics
of the standard bit finding helpers that have the off-by-one issue and
have to special-case the zero mask situation."

On RV64 w/ Zbb, the new "find_zero()" ends up just "ctz" plus the shift
right that then ends up being subsumed by the "add to final length".
Reduce the total instructions from 7 to 3!

But I have no HW platform which supports Zbb, so I can't get the
performance improvement numbers by the last patch, only built and
tested the patch on QEMU.

Jisheng Zhang (3):
  riscv: word-at-a-time: improve find_zero() for !RISCV_ISA_ZBB
  riscv: word-at-a-time: improve find_zero() without Zbb
  riscv: word-at-a-time: improve find_zero() for Zbb

 arch/riscv/include/asm/word-at-a-time.h | 47 +++++++++++++++++++++++--
 1 file changed, 44 insertions(+), 3 deletions(-)

-- 
2.51.0