Re: [RFC PATCH 4/5] mm: truncate: split huge page cache page to a non-zero order if possible.

[Zi Yan <ziy@nvidia.com>]

[-- Attachment #1: Type: text/plain, Size: 1333 bytes --]

On 21 Mar 2022, at 18:32, Roman Gushchin wrote:

> On Mon, Mar 21, 2022 at 10:21:27AM -0400, Zi Yan wrote:
>> From: Zi Yan <ziy@nvidia.com>
>>
>> To minimize the number of pages after a huge page truncation, we do not
>> need to split it all the way down to order-0. The huge page has at most
>> three parts, the part before offset, the part to be truncated, the part
>> remaining at the end. Find the greatest common power of two multiplier of
>> the non-zero values of them as the new order, so we can split the huge
>> page to this order and keep the remaining pages as large and as few as
>> possible.
>
> Would you mind please to describe the algorithm in more details?

Sure.

During truncation, there can be three parts in a huge page:
1. the _offset_ from the beginning of the huge page,
2. the _length_ of the to-be-truncated part,
3. the _remaining_ part after the to-be-truncated part.

the size of the split huge page need to be the greatest common divisor
of the non-zero ones of three after being rounded down to power of two.

OK, I actually find there is a gcd function. I think the algorithm can
simplified to

new_order = ilog2(gcd(gcd(offset, length), remaining)) - PAGE_SHIFT;

I will update the code, the commit message, and the comment in the next
version.

Thank you for the comment.

--
Best Regards,
Yan, Zi

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 854 bytes --]