Re: [PATCH v4] riscv: lib: Fix ZBB strnlen wrap-around regression on huge counts

[shao.mingyin]

Hi Paul,

Gentle ping on this one - it has picked up more tags since posting, and

should be ready to be picked up:

  Acked-by:     Michael Neuling <mikey@xxxxxxxxxxx>

  Reviewed-by:  Aurelien Jarno <aurelien@xxxxxxxxxxx>

  Tested-by:    Troy Mitchell <troy.mitchell@xxxxxxxxx>

  Suggested-by: David Laight <david.laight.linux@xxxxxxxxx>

  Suggested-by: Qingfang Deng <qingfang.deng@xxxxxxxxx>

It stays a minimal fix: only the two boundary-computation instructions

in the ZBB path are replaced (18 insertions, 2 deletions, no new

registers), so the scanning loop itself is unchanged.

This one matters for stable - the regression is in v7.1.10+ and v7.2:

strnlen(s, SIZE_MAX) returns 8 for any 8+-byte aligned string and

truncates names built through FORTIFY strcat/strlcat (device-mapper's

sysfs name "live-base" becomes "live-bas"), which broke blivet/anaconda

installs and LVM on RISC-V systems.  It has since been reproduced

independently on a Fedora 45 riscv64 image with ZBB enabled, where the

fix restores correct behaviour.

Cc: stable is set, and the minimal form should backport cleanly to

7.1.y.  Happy to rebase onto riscv/fixes or adjust anything if needed.

Thanks,

Shao Mingyin