Re: [External] Re: [PATCH v3 05/21] mm/hugetlb: Introduce pgtable allocation/freeing helpers

[Mike Kravetz]

On 11/12/20 4:35 PM, Mike Kravetz wrote:
> On 11/10/20 7:41 PM, Muchun Song wrote:
>> On Wed, Nov 11, 2020 at 8:47 AM Mike Kravetz <mike.kravetz@xxxxxxxxxx> wrote:
>>>
>>> On 11/8/20 6:10 AM, Muchun Song wrote:
>>> I am reading the code incorrectly it does not appear page->lru (of the huge
>>> page) is being used for this purpose.  Is that correct?
>>>
>>> If it is correct, would using page->lru of the huge page make this code
>>> simpler?  I am just missing the reason why you are using
>>> page_huge_pte(page)->lru
>>
>> For 1GB HugeTLB pages, we should pre-allocate more than one page
>> table. So I use a linked list. The page_huge_pte(page) is the list head.
>> Because the page->lru shares storage with page->pmd_huge_pte.
> 
> Sorry, but I do not understand the statement page->lru shares storage with
> page->pmd_huge_pte.  Are you saying they are both in head struct page of
> the huge page?
> 
> Here is what I was suggesting.  If we just use page->lru for the list
> then vmemmap_pgtable_prealloc() could be coded like the following:
> 
> static int vmemmap_pgtable_prealloc(struct hstate *h, struct page *page)
> {
> 	struct page *pte_page, *t_page;
> 	unsigned int nr = pgtable_pages_to_prealloc_per_hpage(h);
> 
> 	if (!nr)
> 		return 0;
> 
> 	/* Store preallocated pages on huge page lru list */
> 	INIT_LIST_HEAD(&page->lru);
> 
> 	while (nr--) {
> 		pte_t *pte_p;
> 
> 		pte_p = pte_alloc_one_kernel(&init_mm);
> 		if (!pte_p)
> 			goto out;
> 		list_add(&virt_to_page(pte_p)->lru, &page->lru);
> 	}
> 
> 	return 0;
> out:
> 	list_for_each_entry_safe(pte_page, t_page, &page->lru, lru)

Forgot the list_del(&pte_page->lru)
Perhaps it is not simpler after all. :)
-- 
Mike Kravetz

> 		pte_free_kernel(&init_mm, page_to_virt(pte_page));
> 	return -ENOMEM;
> }
> 
> By doing this we could eliminate the routines,
> vmemmap_pgtable_init()
> vmemmap_pgtable_deposit()
> vmemmap_pgtable_withdraw()
> and simply use the list manipulation routines.
> 
> To me, that looks simpler than the proposed code in this patch.
>