Re: [PATCH 12/12] hugetlb: batch TLB flushes when restoring vmemmap

[Muchun Song <muchun.song@linux.dev>]

On 2023/8/26 03:04, Mike Kravetz wrote:
> Update the hugetlb_vmemmap_restore path to take a 'batch' parameter that
> indicates restoration is happening on a batch of pages.  When set, use
> the existing mechanism (VMEMMAP_REMAP_BULK_PAGES) to delay TLB flushing.
> The routine hugetlb_vmemmap_restore_folios is the only user of this new
> batch parameter and it will perform a global flush after all vmemmap is
> restored.
>
> Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
> ---
>   mm/hugetlb_vmemmap.c | 37 +++++++++++++++++++++++--------------
>   1 file changed, 23 insertions(+), 14 deletions(-)
>
> diff --git a/mm/hugetlb_vmemmap.c b/mm/hugetlb_vmemmap.c
> index a2fc7b03ac6b..d6e7440b9507 100644
> --- a/mm/hugetlb_vmemmap.c
> +++ b/mm/hugetlb_vmemmap.c
> @@ -479,17 +479,19 @@ static int alloc_vmemmap_page_list(unsigned long start, unsigned long end,
>    * @end:	end address of the vmemmap virtual address range that we want to
>    *		remap.
>    * @reuse:	reuse address.
> + * @bulk:	bulk operation, batch TLB flushes
>    *
>    * Return: %0 on success, negative error code otherwise.
>    */
>   static int vmemmap_remap_alloc(unsigned long start, unsigned long end,
> -			       unsigned long reuse)
> +			       unsigned long reuse, bool bulk)

I'd like to let vmemmap_remap_alloc pass VMEMMAP_REMAP_BULK_PAGES directly,
in which case, we do not need to change this function if we want to 
introduce
another flag in the future. I mean that change "bool bulk" to "unsigned 
long flags".

>   {
>   	LIST_HEAD(vmemmap_pages);
>   	struct vmemmap_remap_walk walk = {
>   		.remap_pte	= vmemmap_restore_pte,
>   		.reuse_addr	= reuse,
>   		.vmemmap_pages	= &vmemmap_pages,
> +		.flags		= !bulk ? 0 : VMEMMAP_REMAP_BULK_PAGES,
>   	};
>   
>   	/* See the comment in the vmemmap_remap_free(). */
> @@ -511,17 +513,7 @@ EXPORT_SYMBOL(hugetlb_optimize_vmemmap_key);
>   static bool vmemmap_optimize_enabled = IS_ENABLED(CONFIG_HUGETLB_PAGE_OPTIMIZE_VMEMMAP_DEFAULT_ON);
>   core_param(hugetlb_free_vmemmap, vmemmap_optimize_enabled, bool, 0);
>   
> -/**
> - * hugetlb_vmemmap_restore - restore previously optimized (by
> - *			     hugetlb_vmemmap_optimize()) vmemmap pages which
> - *			     will be reallocated and remapped.
> - * @h:		struct hstate.
> - * @head:	the head page whose vmemmap pages will be restored.
> - *
> - * Return: %0 if @head's vmemmap pages have been reallocated and remapped,
> - * negative error code otherwise.
> - */
> -int hugetlb_vmemmap_restore(const struct hstate *h, struct page *head)
> +int __hugetlb_vmemmap_restore(const struct hstate *h, struct page *head, bool bulk)

The same as here.

>   {
>   	int ret;
>   	unsigned long vmemmap_start = (unsigned long)head, vmemmap_end;
> @@ -541,7 +533,7 @@ int hugetlb_vmemmap_restore(const struct hstate *h, struct page *head)
>   	 * When a HugeTLB page is freed to the buddy allocator, previously
>   	 * discarded vmemmap pages must be allocated and remapping.
>   	 */
> -	ret = vmemmap_remap_alloc(vmemmap_start, vmemmap_end, vmemmap_reuse);
> +	ret = vmemmap_remap_alloc(vmemmap_start, vmemmap_end, vmemmap_reuse, bulk);
>   	if (!ret) {
>   		ClearHPageVmemmapOptimized(head);
>   		static_branch_dec(&hugetlb_optimize_vmemmap_key);
> @@ -550,12 +542,29 @@ int hugetlb_vmemmap_restore(const struct hstate *h, struct page *head)
>   	return ret;
>   }
>   
> +/**
> + * hugetlb_vmemmap_restore - restore previously optimized (by
> + *			     hugetlb_vmemmap_optimize()) vmemmap pages which
> + *			     will be reallocated and remapped.
> + * @h:		struct hstate.
> + * @head:	the head page whose vmemmap pages will be restored.
> + *
> + * Return: %0 if @head's vmemmap pages have been reallocated and remapped,
> + * negative error code otherwise.
> + */
> +int hugetlb_vmemmap_restore(const struct hstate *h, struct page *head)
> +{
> +	return __hugetlb_vmemmap_restore(h, head, false);
> +}
> +
>   void hugetlb_vmemmap_restore_folios(const struct hstate *h, struct list_head *folio_list)
>   {
>   	struct folio *folio;
>   
>   	list_for_each_entry(folio, folio_list, lru)
> -		hugetlb_vmemmap_restore(h, &folio->page);
> +		(void)__hugetlb_vmemmap_restore(h, &folio->page, true);

Pass VMEMMAP_REMAP_BULK_PAGES directly here.

Thanks.

> +
> +	flush_tlb_kernel_range(0, TLB_FLUSH_ALL);
>   }
>   
>   /* Return true iff a HugeTLB whose vmemmap should and can be optimized. */