Re: [PATCH] psi: Treat ksm swapping in copy as memstall

[Johannes Weiner <hannes@cmpxchg.org>]

Hello Yang,

On Sun, Jan 16, 2022 at 03:21:51PM +0000, cgel.zte@gmail.com wrote:
> From: Yang Yang <yang.yang29@zte.com.cn>
> 
> When faults in from swap what used to be a ksm page and that page
> had been swapped in before, system has to make a copy. Obviously
> this kind of copy is related to high memory pressure, so we treat
> it as memstall. Although ksm page merging is not because of high
> memory pressure.
> 
> Information of this new kind of stall will help psi to account
> memory pressure more precise.

Thanks for your patch. I'm curious if you have a concrete use case
where this makes a difference, or if this is something you found while
reading the code?

> Signed-off-by: Yang Yang <yang.yang29@zte.com.cn>
> ---
>  mm/ksm.c | 4 ++++
>  1 file changed, 4 insertions(+)
> 
> diff --git a/mm/ksm.c b/mm/ksm.c
> index 4a7f8614e57d..d4ec6773f9b8 100644
> --- a/mm/ksm.c
> +++ b/mm/ksm.c
> @@ -39,6 +39,7 @@
>  #include <linux/freezer.h>
>  #include <linux/oom.h>
>  #include <linux/numa.h>
> +#include <linux/psi.h>
>  
>  #include <asm/tlbflush.h>
>  #include "internal.h"
> @@ -2569,6 +2570,7 @@ struct page *ksm_might_need_to_copy(struct page *page,
>  {
>  	struct anon_vma *anon_vma = page_anon_vma(page);
>  	struct page *new_page;
> +	unsigned long pflags;
>  
>  	if (PageKsm(page)) {
>  		if (page_stable_node(page) &&
> @@ -2583,6 +2585,7 @@ struct page *ksm_might_need_to_copy(struct page *page,
>  	if (!PageUptodate(page))
>  		return page;		/* let do_swap_page report the error */
>  
> +	psi_memstall_enter(&pflags);
>  	new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
>  	if (new_page &&
>  	    mem_cgroup_charge(page_folio(new_page), vma->vm_mm, GFP_KERNEL)) {
> @@ -2600,6 +2603,7 @@ struct page *ksm_might_need_to_copy(struct page *page,
>  #endif
>  	}
>  
> +	psi_memstall_leave(&pflags);

This does unconditional stall accounting for a swapin operation. But
if you take a look at workingset_refault() -> folio_wait_bit_common(),
we only count memory stalls when the page is thrashing, not when it's
a transitionary refault (which happen even when there is enough memory
to hold the workingset). You need to check PageWorkingset() at least.

But again I'd be curious first if this is a practical concern. Swapins
should be IO dominated - or in the case of zswap dominated by the
decompression. Does a page copy really matter?