On Tue, Jul 30, 2024 at 12:26 AM Shung-Hsi Yu [off-list ref] wrote:
[...]

That is great to hear and really boost the level of confidence. Though I
did made an update[1] to the patch such that implementation of
negative_bit_floor() is change from

        v &= v >> 1;
        v &= v >> 2;
        v &= v >> 4;
        v &= v >> 8;
        v &= v >> 16;
        v &= v >> 32;
        return v;

to one that closer resembles tnum_range()

        u8 bits = fls64(~v); /* find most-significant unset bit */
        u64 delta;

        /* special case, needed because 1ULL << 64 is undefined */
        if (bits > 63)
                return 0;

        delta = (1ULL << bits) - 1;
        return ~delta;

This [1] is indeed the version of the patch that we checked: the one that
uses fls and fls64 in negative_bit_floor and negative32_bit_floor .
I replied here because you had CCed us in this thread.

Note that for checking in Agni, the implementation of fls and fls64 were
borrowed from asm-generic [2,3,4].

Having said that, the patch [1] looks good to me.
Tested-by: Harishankar Vishwanathan <redacted>

[1]: https://lore.kernel.org/bpf/20240719081702.137173-1-shung-hsi.yu@suse.com/ (local)
[2]: https://elixir.bootlin.com/linux/v6.10/source/include/asm-generic/bitops/fls.h#L43
[3]: https://elixir.bootlin.com/linux/v6.10/source/include/asm-generic/bitops/fls64.h#L19
[4]: https://elixir.bootlin.com/linux/v6.10/source/include/asm-generic/bitops/__fls.h#L45