Re: [PATCH v5 4/6] mm/page_isolation: remove migratetype from move_freepages_block_isolate()

[Vlastimil Babka <vbabka@suse.cz>]

On 5/23/25 21:12, Zi Yan wrote:
> Since migratetype is no longer overwritten during pageblock isolation,
> moving a pageblock out of MIGRATE_ISOLATE no longer needs a new
> migratetype.
> 
> Add pageblock_isolate_and_move_free_pages() and
> pageblock_unisolate_and_move_free_pages() to be explicit about the page
> isolation operations. Both share the common code in
> __move_freepages_block_isolate(), which is renamed from
> move_freepages_block_isolate().
> 
> Add toggle_pageblock_isolate() to flip pageblock isolation bit in
> __move_freepages_block_isolate().
> 
> Make set_pageblock_migratetype() only accept non MIGRATE_ISOLATE types,
> so that one should use set_pageblock_isolate() to isolate pageblocks.
> As a result, move pageblock migratetype code out of
> __move_freepages_block().
> 
> Signed-off-by: Zi Yan <ziy@nvidia.com>
> ---
>  include/linux/page-isolation.h |  5 +-
>  mm/page_alloc.c                | 97 ++++++++++++++++++++++++++++------
>  mm/page_isolation.c            | 21 ++++----
>  3 files changed, 92 insertions(+), 31 deletions(-)
> 
> diff --git a/include/linux/page-isolation.h b/include/linux/page-isolation.h
> index 14c6a5f691c2..7241a6719618 100644
> --- a/include/linux/page-isolation.h
> +++ b/include/linux/page-isolation.h
> @@ -44,10 +44,9 @@ static inline void set_pageblock_isolate(struct page *page)
>  void __meminit init_pageblock_migratetype(struct page *page,
>  					  enum migratetype migratetype,
>  					  bool isolate);
> -void set_pageblock_migratetype(struct page *page, enum migratetype migratetype);
>  
> -bool move_freepages_block_isolate(struct zone *zone, struct page *page,
> -				  int migratetype);
> +bool pageblock_isolate_and_move_free_pages(struct zone *zone, struct page *page);
> +bool pageblock_unisolate_and_move_free_pages(struct zone *zone, struct page *page);
>  
>  int start_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
>  			     int migratetype, int flags);
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index 8fcbd7fa13c2..44a08b1a9de4 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -524,13 +524,36 @@ void clear_pfnblock_bit(const struct page *page, unsigned long pfn,
>  	__clear_bit(bitidx + pb_bit, bitmap_word);
>  }
>  
> +#ifdef CONFIG_MEMORY_ISOLATION
> +/**
> + * toggle_pfnblock_bit - Toggle a standalone bit of a pageblock
> + * @page: The page within the block of interest
> + * @pfn: The target page frame number
> + * @pb_bit: pageblock bit to toggle
> + */
> +static void toggle_pfnblock_bit(const struct page *page, unsigned long pfn,
> +		      enum pageblock_bits pb_bit)
> +{
> +	unsigned long *bitmap_word;
> +	unsigned long bitidx;
> +
> +	if (WARN_ON_ONCE(pb_bit <= PB_migrate_end ||
> +			 pb_bit >= __NR_PAGEBLOCK_BITS))
> +		return;
> +
> +	get_pfnblock_bitmap_bitidx(page, pfn, &bitmap_word, &bitidx);
> +
> +	__change_bit(bitidx + pb_bit, bitmap_word);

Again the non-atomic variant, but actually below I suggest we drop this.

> +}
> +#endif
> +
>  /**
>   * set_pageblock_migratetype - Set the migratetype of a pageblock
>   * @page: The page within the block of interest
>   * @migratetype: migratetype to set
>   */
> -__always_inline void set_pageblock_migratetype(struct page *page,
> -					       enum migratetype migratetype)
> +static void set_pageblock_migratetype(struct page *page,
> +				      enum migratetype migratetype)
>  {
>  	unsigned long mask = MIGRATETYPE_MASK;
>  
> @@ -540,11 +563,15 @@ __always_inline void set_pageblock_migratetype(struct page *page,
>  
>  #ifdef CONFIG_MEMORY_ISOLATION
>  	if (migratetype == MIGRATE_ISOLATE) {
> -		set_pfnblock_bit(page, page_to_pfn(page), PB_migrate_isolate);
> +		VM_WARN_ONCE(1,
> +			"Use set_pageblock_isolate() for pageblock isolation");
>  		return;
>  	}
>  	/* change mask to clear PB_migrate_isolate if it is set */
>  	mask = MIGRATETYPE_AND_ISO_MASK;
> +	VM_WARN_ONCE(get_pfnblock_bit(page, page_to_pfn(page),
> +				      PB_migrate_isolate),
> +		     "Use clear_pageblock_isolate() to unisolate pageblock");
>  #endif

We might be too paranoid with the warnings given these are all local
functions to this file so risk of misuse should be low. Maybe we could
remove later...

>  	__set_pfnblock_flags_mask(page, page_to_pfn(page),
>  				  (unsigned long)migratetype, mask);
> @@ -1931,8 +1958,8 @@ static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
>  #endif
>  
>  /*
> - * Change the type of a block and move all its free pages to that
> - * type's freelist.
> + * Move all free pages of a block to new type's freelist. Caller needs to
> + * change the block type.
>   */
>  static int __move_freepages_block(struct zone *zone, unsigned long start_pfn,
>  				  int old_mt, int new_mt)
> @@ -1964,8 +1991,6 @@ static int __move_freepages_block(struct zone *zone, unsigned long start_pfn,
>  		pages_moved += 1 << order;
>  	}
>  
> -	set_pageblock_migratetype(pfn_to_page(start_pfn), new_mt);
> -
>  	return pages_moved;
>  }
>  
> @@ -2023,11 +2048,16 @@ static int move_freepages_block(struct zone *zone, struct page *page,
>  				int old_mt, int new_mt)
>  {
>  	unsigned long start_pfn;
> +	int res;
>  
>  	if (!prep_move_freepages_block(zone, page, &start_pfn, NULL, NULL))
>  		return -1;
>  
> -	return __move_freepages_block(zone, start_pfn, old_mt, new_mt);
> +	res = __move_freepages_block(zone, start_pfn, old_mt, new_mt);
> +	set_pageblock_migratetype(pfn_to_page(start_pfn), new_mt);
> +
> +	return res;
> +
>  }
>  
>  #ifdef CONFIG_MEMORY_ISOLATION
> @@ -2055,11 +2085,16 @@ static unsigned long find_large_buddy(unsigned long start_pfn)
>  	return start_pfn;
>  }
>  
> +static inline void toggle_pageblock_isolate(struct page *page)
> +{
> +	toggle_pfnblock_bit(page, page_to_pfn(page), PB_migrate_isolate);
> +}

I'm wary about the togle action, as we should always know what action we
want to do anyway. So we could just add a "bool isolate" parameter and call
set or clear explicitly? Allows for some hypothetical DEBUG_VM checks too
(pageblock is not already in the state we want it to be).