Re: [RFC 3/3] mm: make swapin readahead to improve thp collapse rate

[Andrew Morton]

On Sun, 14 Jun 2015 18:04:43 +0300 Ebru Akagunduz <ebru.akagunduz@xxxxxxxxx> wrote:

> This patch makes swapin readahead to improve thp collapse rate.
> When khugepaged scanned pages, there can be a few of the pages
> in swap area.
> 
> With the patch THP can collapse 4kB pages into a THP when
> there are up to max_ptes_swap swap ptes in a 2MB range.
> 
> The patch was tested with a test program that allocates
> 800MB of memory, writes to it, and then sleeps. I force
> the system to swap out all. Afterwards, the test program
> touches the area by writing, it skips a page in each
> 20 pages of the area.
> 
> Without the patch, system did not swap in readahead.
> THP rate was %47 of the program of the memory, it
> did not change over time.
> 
> With this patch, after 10 minutes of waiting khugepaged had
> collapsed %99 of the program's memory.
> 
> ...
>
> +/*
> + * Bring missing pages in from swap, to complete THP collapse.
> + * Only done if khugepaged_scan_pmd believes it is worthwhile.
> + *
> + * Called and returns without pte mapped or spinlocks held,
> + * but with mmap_sem held to protect against vma changes.
> + */
> +
> +static void __collapse_huge_page_swapin(struct mm_struct *mm,
> +					struct vm_area_struct *vma,
> +					unsigned long address, pmd_t *pmd,
> +					pte_t *pte)
> +{
> +	unsigned long _address;
> +	pte_t pteval = *pte;
> +	int swap_pte = 0;
> +
> +	pte = pte_offset_map(pmd, address);
> +	for (_address = address; _address < address + HPAGE_PMD_NR*PAGE_SIZE;
> +	     pte++, _address += PAGE_SIZE) {
> +		pteval = *pte;
> +		if (is_swap_pte(pteval)) {
> +			swap_pte++;
> +			do_swap_page(mm, vma, _address, pte, pmd, 0x0, pteval);
> +			/* pte is unmapped now, we need to map it */
> +			pte = pte_offset_map(pmd, _address);
> +		}
> +	}
> +	pte--;
> +	pte_unmap(pte);
> +	trace_mm_collapse_huge_page_swapin(mm, vma->vm_start, swap_pte);
> +}

This is doing a series of synchronous reads.  That will be sloooow on
spinning disks.

This function should be significantly faster if it first gets all the
necessary I/O underway.  I don't think we have a function which exactly
does this.  Perhaps generalise swapin_readahead() or open-code
something like

	blk_start_plug(...);
	for (_address = address; _address < address + HPAGE_PMD_NR*PAGE_SIZE;
	     pte++, _address += PAGE_SIZE) {
		if (is_swap_pte(*pte)) {
			read_swap_cache_async(...);
		}
	}
	blk_finish_plug(...);


If you do make a change such as this, please benchmark its effects. 
Not on SSD ;)

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/