Re: [PATCH v3 2/5] ksm: support unsharing zero pages placed by KSM

[David Hildenbrand]

On 11.10.22 04:22, xu.xin.sc@xxxxxxxxx wrote:
> From: xu xin <xu.xin16@xxxxxxxxxx>
>
> use_zero_pages may be very useful, not just because of cache colouring
> as described in doc, but also because use_zero_pages can accelerate
> merging empty pages when there are plenty of empty pages (full of zeros)
> as the time of page-by-page comparisons (unstable_tree_search_insert) is
> saved.
>
> But when enabling use_zero_pages, madvise(addr, len, MADV_UNMERGEABLE) and
> other ways (like write 2 to /sys/kernel/mm/ksm/run) to trigger unsharing
> will *not* unshare the shared zeropage as placed by KSM (which may be
> against the MADV_UNMERGEABLE documentation at least).
>
> To not blindly unshare all shared zero_pages in applicable VMAs, the patch
> introduces a dedicated flag ZERO_PAGE_FLAG to mark the rmap_items of those
> shared zero_pages. and guarantee that these rmap_items will be not freed
> during the time of zero_pages not being writing, so we can only unshare
> the *KSM-placed* zero_pages.
>
> The patch will not degrade the performance of use_zero_pages as it doesn't
> change the way of merging empty pages in use_zero_pages's feature.
>
> Fixes: e86c59b1b12d ("mm/ksm: improve deduplication of zero pages with colouring")
> Reported-by: David Hildenbrand <david@xxxxxxxxxx>
> Cc: Claudio Imbrenda <imbrenda@xxxxxxxxxxxxx>
> Cc: Xuexin Jiang <jiang.xuexin@xxxxxxxxxx>
> Signed-off-by: xu xin <xu.xin16@xxxxxxxxxx>
> Co-developed-by: Xiaokai Ran <ran.xiaokai@xxxxxxxxxx>
> Signed-off-by: Xiaokai Ran <ran.xiaokai@xxxxxxxxxx>
> Co-developed-by: Yang Yang <yang.yang29@xxxxxxxxxx>
> Signed-off-by: Yang Yang <yang.yang29@xxxxxxxxxx>
> Signed-off-by: xu xin <xu.xin16@xxxxxxxxxx>
> ---
>   mm/ksm.c | 136 ++++++++++++++++++++++++++++++++++++++++++-------------
>   1 file changed, 105 insertions(+), 31 deletions(-)
>
> diff --git a/mm/ksm.c b/mm/ksm.c
> index 13c60f1071d8..e351d7b6d15e 100644
> --- a/mm/ksm.c
> +++ b/mm/ksm.c
> @@ -213,6 +213,7 @@ struct ksm_rmap_item {
>   #define SEQNR_MASK	0x0ff	/* low bits of unstable tree seqnr */
>   #define UNSTABLE_FLAG	0x100	/* is a node of the unstable tree */
>   #define STABLE_FLAG	0x200	/* is listed from the stable tree */
> +#define ZERO_PAGE_FLAG 0x400 /* is zero page placed by KSM */
>   
>   /* The stable and unstable tree heads */
>   static struct rb_root one_stable_tree[1] = { RB_ROOT };
> @@ -381,14 +382,6 @@ static inline struct ksm_rmap_item *alloc_rmap_item(void)
>   	return rmap_item;
>   }
>   
> -static inline void free_rmap_item(struct ksm_rmap_item *rmap_item)
> -{
> -	ksm_rmap_items--;
> -	rmap_item->mm->ksm_rmap_items--;
> -	rmap_item->mm = NULL;	/* debug safety */
> -	kmem_cache_free(rmap_item_cache, rmap_item);
> -}
> -
>   static inline struct ksm_stable_node *alloc_stable_node(void)
>   {
>   	/*
> @@ -420,7 +413,8 @@ static inline bool ksm_test_exit(struct mm_struct *mm)
>   }
>   
>   /*
> - * We use break_ksm to break COW on a ksm page: it's a stripped down
> + * We use break_ksm to break COW on a ksm page or KSM-placed zero page (only
> + * happen when enabling use_zero_pages): it's a stripped down
>    *
>    *	if (get_user_pages(addr, 1, FOLL_WRITE, &page, NULL) == 1)
>    *		put_page(page);
> @@ -434,7 +428,8 @@ static inline bool ksm_test_exit(struct mm_struct *mm)
>    * of the process that owns 'vma'.  We also do not want to enforce
>    * protection keys here anyway.
>    */
> -static int break_ksm(struct vm_area_struct *vma, unsigned long addr)
> +static int break_ksm(struct vm_area_struct *vma, unsigned long addr,
> +				     bool ksm_check_bypass)
>   {
>   	struct page *page;
>   	vm_fault_t ret = 0;
> @@ -449,6 +444,16 @@ static int break_ksm(struct vm_area_struct *vma, unsigned long addr)
>   			ret = handle_mm_fault(vma, addr,
>   					      FAULT_FLAG_WRITE | FAULT_FLAG_REMOTE,
>   					      NULL);
> +		else if (ksm_check_bypass && is_zero_pfn(page_to_pfn(page))) {
> +			/*
> +			 * Although it's not ksm page, it's zero page as placed by
> +			 * KSM use_zero_page, so we should unshare it when
> +			 * ksm_check_bypass is true.
> +			 */
> +			ret = handle_mm_fault(vma, addr,
> +						  FAULT_FLAG_WRITE | FAULT_FLAG_REMOTE,
> +						  NULL);
> +		}

Please don't duplicate that page fault triggering code.

Also, please be aware that this collides with

https://lkml.kernel.org/r/20221021101141.84170-1-david@xxxxxxxxxx

Adjustments should be comparatively easy.

--
Thanks,

David / dhildenb