Re: [PATCH v6 11/15] mm: define clear_pages(), clear_user_pages()

[Ankur Arora <ankur.a.arora@oracle.com>]

Ankur Arora <ankur.a.arora@oracle.com> writes:

> David Hildenbrand <david@redhat.com> writes:
>
>> On 02.09.25 10:08, Ankur Arora wrote:

[ ... ]

>>> + * clear_pages() - clear kernel page range.
>>> + * @addr: start address of page range
>>> + * @npages: number of pages
>>> + *
>>> + * Assumes that (@addr, +@npages) references a kernel region.
>>> + * Like clear_page(), this does absolutely no exception handling.
>>> + */
>>> +static inline void clear_pages(void *addr, unsigned int npages)
>>> +{
>>> +	for (int i = 0; i < npages; i++)
>>> +		clear_page(addr + i * PAGE_SIZE);
>>
>> If we know that we will clear at least one page (which we can document)
>>
>> do {
>> 	clear_page(addr);
>> 	addr += PAGE_SIZE;
>> } while (--npages);
>>
>> Similarly for the case below.
>
> Ack. Though how about the following instead? Slightly less clear but
> probably better suited for caching the likely access pattern.
>
>    addr += (npages - 1) * PAGE_SIZE;
>    do {
>            clear_page(addr);
>            addr -= PAGE_SIZE;
>    } while (--npages);

Decided against using this reverse access pattern. I would need
to replicate it at three places  (clear_pages(), clear_user_pages() and
clear_highpages()) and this kind of policy should probably be dictated
from folio_zero_user() or clear_contig_pages() etc.

--
ankur