Date: Thu, 30 May 2024 08:27:15 -0400
From: Johannes Weiner <hannes@...xchg.org>
To: Usama Arif <usamaarif642@...il.com>
Cc: akpm@...ux-foundation.org, yosryahmed@...gle.com, nphamcs@...il.com,
	chengming.zhou@...ux.dev, linux-mm@...ck.org,
	linux-kernel@...r.kernel.org, kernel-team@...a.com,
	Hugh Dickins <hughd@...gle.com>, Huang Ying <ying.huang@...el.com>
Subject: Re: [PATCH 1/2] mm: store zero pages to be swapped out in a bitmap

On Thu, May 30, 2024 at 11:19:07AM +0100, Usama Arif wrote:
> Approximately 10-20% of pages to be swapped out are zero pages [1].
> Rather than reading/writing these pages to flash resulting
> in increased I/O and flash wear, a bitmap can be used to mark these
> pages as zero at write time, and the pages can be filled at
> read time if the bit corresponding to the page is set.
> With this patch, NVMe writes in Meta server fleet decreased
> by almost 10% with conventional swap setup (zswap disabled).
> 
> [1]https://lore.kernel.org/all/20171018104832epcms5p1b2232e2236258de3d03d1344dde9fce0@epcms5p1/
> 
> Signed-off-by: Usama Arif <usamaarif642@...il.com>

This is awesome.

> ---
>  include/linux/swap.h |  1 +
>  mm/page_io.c         | 86 ++++++++++++++++++++++++++++++++++++++++++--
>  mm/swapfile.c        | 10 ++++++
>  3 files changed, 95 insertions(+), 2 deletions(-)
> 
> diff --git a/include/linux/swap.h b/include/linux/swap.h
> index a11c75e897ec..e88563978441 100644
> --- a/include/linux/swap.h
> +++ b/include/linux/swap.h
> @@ -299,6 +299,7 @@ struct swap_info_struct {
>  	signed char	type;		/* strange name for an index */
>  	unsigned int	max;		/* extent of the swap_map */
>  	unsigned char *swap_map;	/* vmalloc'ed array of usage counts */
> +	unsigned long *zeromap;		/* vmalloc'ed bitmap to track zero pages */

One bit per swap slot, so 1 / (4096 * 8) = 0.003% static memory
overhead for configured swap space. That seems reasonable for what
appears to be a fairly universal 10% reduction in swap IO.

An alternative implementation would be to reserve a bit in
swap_map. This would be no overhead at idle, but would force
continuation counts earlier on heavily shared page tables, and AFAICS
would get complicated in terms of locking, whereas this one is pretty
simple (atomic ops protect the map, swapcache lock protects the bit).

So I prefer this version. But a few comments below:

>  	struct swap_cluster_info *cluster_info; /* cluster info. Only for SSD */
>  	struct swap_cluster_list free_clusters; /* free clusters list */
>  	unsigned int lowest_bit;	/* index of first free in swap_map */
> diff --git a/mm/page_io.c b/mm/page_io.c
> index a360857cf75d..ab043b4ad577 100644
> --- a/mm/page_io.c
> +++ b/mm/page_io.c
> @@ -172,6 +172,77 @@ int generic_swapfile_activate(struct swap_info_struct *sis,
>  	goto out;
>  }
>  
> +static bool is_folio_page_zero_filled(struct folio *folio, int i)
> +{
> +	unsigned long *page;
> +	unsigned int pos;
> +	bool ret = false;
> +
> +	page = kmap_local_folio(folio, i * PAGE_SIZE);
> +	for (pos = 0; pos < PAGE_SIZE / sizeof(*page); pos++) {
> +		if (page[pos] != 0)
> +			goto out;
> +	}
> +	ret = true;
> +out:
> +	kunmap_local(page);
> +	return ret;
> +}
> +
> +static bool is_folio_zero_filled(struct folio *folio)
> +{
> +	unsigned int i;
> +
> +	for (i = 0; i < folio_nr_pages(folio); i++) {
> +		if (!is_folio_page_zero_filled(folio, i))
> +			return false;
> +	}
> +	return true;
> +}
> +
> +static void folio_page_zero_fill(struct folio *folio, int i)
> +{
> +	unsigned long *page;
> +
> +	page = kmap_local_folio(folio, i * PAGE_SIZE);
> +	memset_l(page, 0, PAGE_SIZE / sizeof(unsigned long));
> +	kunmap_local(page);
> +}
> +
> +static void folio_zero_fill(struct folio *folio)
> +{
> +	unsigned int i;
> +
> +	for (i = 0; i < folio_nr_pages(folio); i++)
> +		folio_page_zero_fill(folio, i);
> +}
> +
> +static void swap_zeromap_folio_set(struct folio *folio)
> +{
> +	struct swap_info_struct *sis = swp_swap_info(folio->swap);
> +	swp_entry_t entry;
> +	unsigned int i;
> +
> +	for (i = 0; i < folio_nr_pages(folio); i++) {
> +		entry = page_swap_entry(folio_page(folio, i));
> +		bitmap_set(sis->zeromap, swp_offset(entry), 1);

This should be set_bit(). bitmap_set() isn't atomic, so it would
corrupt the map on concurrent swapping of other zero pages. And you
don't need a range op here anyway.

> +	}
> +}
> +
> +static bool swap_zeromap_folio_test(struct folio *folio)
> +{
> +	struct swap_info_struct *sis = swp_swap_info(folio->swap);
> +	swp_entry_t entry;
> +	unsigned int i;
> +
> +	for (i = 0; i < folio_nr_pages(folio); i++) {
> +		entry = page_swap_entry(folio_page(folio, i));
> +		if (!test_bit(swp_offset(entry), sis->zeromap))
> +			return false;
> +	}
> +	return true;
> +}
> +
>  /*
>   * We may have stale swap cache pages in memory: notice
>   * them here and get rid of the unnecessary final write.
> @@ -195,6 +266,14 @@ int swap_writepage(struct page *page, struct writeback_control *wbc)
>  		folio_unlock(folio);
>  		return ret;
>  	}
> +
> +	if (is_folio_zero_filled(folio)) {
> +		swap_zeromap_folio_set(folio);
> +		folio_start_writeback(folio);
> +		folio_unlock(folio);
> +		folio_end_writeback(folio);
> +		return 0;
> +	}

You need to clear the zeromap bit in the else branch.

Userspace can change the contents of a swapcached page, which
redirties the page and forces an overwrite of the slot when the page
is reclaimed again.

So if the page goes from zeroes to something else and then gets
reclaimed again, a subsequent swapin would read the stale zeroes.

>  	if (zswap_store(folio)) {
>  		folio_start_writeback(folio);
>  		folio_unlock(folio);
> @@ -515,8 +594,11 @@ void swap_read_folio(struct folio *folio, bool synchronous,
>  		psi_memstall_enter(&pflags);
>  	}
>  	delayacct_swapin_start();
> -
> -	if (zswap_load(folio)) {
> +	if (swap_zeromap_folio_test(folio)) {
> +		folio_zero_fill(folio);
> +		folio_mark_uptodate(folio);
> +		folio_unlock(folio);
> +	} else if (zswap_load(folio)) {
>  		folio_mark_uptodate(folio);
>  		folio_unlock(folio);
>  	} else if (data_race(sis->flags & SWP_FS_OPS)) {
> diff --git a/mm/swapfile.c b/mm/swapfile.c
> index f1e559e216bd..3f00a1cce5ba 100644
> --- a/mm/swapfile.c
> +++ b/mm/swapfile.c
> @@ -461,6 +461,7 @@ static void swap_cluster_schedule_discard(struct swap_info_struct *si,
>  	 */
>  	memset(si->swap_map + idx * SWAPFILE_CLUSTER,
>  			SWAP_MAP_BAD, SWAPFILE_CLUSTER);
> +	bitmap_clear(si->zeromap, idx * SWAPFILE_CLUSTER, SWAPFILE_CLUSTER);

AFAICS this needs to be atomic as well. The swap_info and cluster are
locked, which stabilizes si->swap_map, but zeromap can see updates
from concurrent swap_writepage() and swap_read_folio() on other slots.

I think you need to use a loop over clear_bit(). Please also add a
comment with the above.

>  
>  	cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx);
>  
> @@ -498,6 +499,7 @@ static void swap_do_scheduled_discard(struct swap_info_struct *si)
>  		__free_cluster(si, idx);
>  		memset(si->swap_map + idx * SWAPFILE_CLUSTER,
>  				0, SWAPFILE_CLUSTER);
> +		bitmap_clear(si->zeromap, idx * SWAPFILE_CLUSTER, SWAPFILE_CLUSTER);

Same.

>  		unlock_cluster(ci);
>  	}
>  }
> @@ -1336,6 +1338,7 @@ static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry)
>  	count = p->swap_map[offset];
>  	VM_BUG_ON(count != SWAP_HAS_CACHE);
>  	p->swap_map[offset] = 0;
> +	bitmap_clear(p->zeromap, offset, 1);

This too needs to be atomic, IOW clear_bit().

Otherwise this looks good to me.

>  	dec_cluster_info_page(p, p->cluster_info, offset);
>  	unlock_cluster(ci);
>  
> @@ -2597,6 +2600,7 @@ SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
>  	free_percpu(p->cluster_next_cpu);
>  	p->cluster_next_cpu = NULL;
>  	vfree(swap_map);
> +	bitmap_free(p->zeromap);
>  	kvfree(cluster_info);
>  	/* Destroy swap account information */
>  	swap_cgroup_swapoff(p->type);
> @@ -3123,6 +3127,12 @@ SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
>  		goto bad_swap_unlock_inode;
>  	}
>  
> +	p->zeromap = bitmap_zalloc(maxpages, GFP_KERNEL);
> +	if (!p->zeromap) {
> +		error = -ENOMEM;
> +		goto bad_swap_unlock_inode;
> +	}
> +
>  	if (p->bdev && bdev_stable_writes(p->bdev))
>  		p->flags |= SWP_STABLE_WRITES;
>  
> -- 
> 2.43.0

Powered by blists - more mailing lists