Re: [PATCH] KVM: Simplify backwards dirty ring batch check

[Paolo Bonzini <pbonzini@redhat.com>]

On 9/17/26 18:09, Hao Peng wrote:
>> ... I love it, but it's unreadable. The "mask << -delta >> -delta"
>> basically means no overflow on shift left. So just write
>>
>> if (!check_shl_overflow(mask, -delta, shifted_mask)) {
>>      cur_offset = next_offset;
>>      mask = shifted_mask | 1;
>> }
>>
> Yes, check_shl_overflow() expresses the intent much more clearly than
>   the leading-zero calculation.
> 
>   I will use it together with an unsigned shift calculated directly from
>    cur_offset and next_offset.  That also avoids evaluating -delta when
>    delta is S64_MIN.

That would require a 2^(64+12) bit address space, so it cannot happen I 
think?

>    Something along these lines:
> 
>            unsigned long shifted_mask;
>            u64 shift = abs_diff(next_offset, cur_offset);
> 
>            if (next_offset < cur_offset &&
>                !check_shl_overflow(mask, shift, &shifted_mask)) {
>                    cur_offset = next_offset;
>                    mask = shifted_mask | BIT(0);
>                    continue;
>            }

shift can be just cur_offset - next_offset here, and it can be placed in 
the call to check_shl_overflow().

Paolo