Re: [PATCH v3 1/4] mm,hwpoison,hugetlb,memory_hotplug: hotremove memory section with hwpoisoned hugepage

[Miaohe Lin <linmiaohe@huawei.com>]

On 2022/9/21 17:13, Naoya Horiguchi wrote:
> From: Naoya Horiguchi <naoya.horiguchi@nec.com>
> 
> HWPoisoned page is not supposed to be accessed once marked, but currently
> such accesses can happen during memory hotremove because do_migrate_range()
> can be called before dissolve_free_huge_pages() is called.
> 
> Clear HPageMigratable for hwpoisoned hugepages to prevent them from being
> migrated.  This should be done in hugetlb_lock to avoid race against
> isolate_hugetlb().
> 
> get_hwpoison_huge_page() needs to have a flag to show it's called from
> unpoison to take refcount of hwpoisoned hugepages, so add it.
> 
> Reported-by: Miaohe Lin <linmiaohe@huawei.com>
> Signed-off-by: Naoya Horiguchi <naoya.horiguchi@nec.com>

Thanks for your work, Naoya. Maybe something to improve below.

> ---
> ChangeLog v2 -> v3
> - move to the approach of clearing HPageMigratable instead of shifting
>   dissolve_free_huge_pages.
> ---
>  include/linux/hugetlb.h |  4 ++--
>  mm/hugetlb.c            |  4 ++--
>  mm/memory-failure.c     | 12 ++++++++++--
>  3 files changed, 14 insertions(+), 6 deletions(-)
> 

<snip>

> @@ -7267,7 +7267,7 @@ int get_hwpoison_huge_page(struct page *page, bool *hugetlb)
>  		*hugetlb = true;
>  		if (HPageFreed(page))
>  			ret = 0;
> -		else if (HPageMigratable(page))
> +		else if (HPageMigratable(page) || unpoison)

Is unpoison_memory() expected to restore the HPageMigratable flag as well ?

>  			ret = get_page_unless_zero(page);
>  		else
>  			ret = -EBUSY;
> diff --git a/mm/memory-failure.c b/mm/memory-failure.c
> index 145bb561ddb3..5942e1c0407e 100644
> --- a/mm/memory-failure.c
> +++ b/mm/memory-failure.c
> @@ -1244,7 +1244,7 @@ static int __get_hwpoison_page(struct page *page, unsigned long flags)
>  	int ret = 0;
>  	bool hugetlb = false;
>  
> -	ret = get_hwpoison_huge_page(head, &hugetlb);
> +	ret = get_hwpoison_huge_page(head, &hugetlb, false);
>  	if (hugetlb)
>  		return ret;
>  
> @@ -1334,7 +1334,7 @@ static int __get_unpoison_page(struct page *page)
>  	int ret = 0;
>  	bool hugetlb = false;
>  
> -	ret = get_hwpoison_huge_page(head, &hugetlb);
> +	ret = get_hwpoison_huge_page(head, &hugetlb, true);
>  	if (hugetlb)
>  		return ret;
>  
> @@ -1815,6 +1815,13 @@ int __get_huge_page_for_hwpoison(unsigned long pfn, int flags)
>  		goto out;
>  	}
>  
> +	/*
> +	 * Clearing HPageMigratable for hwpoisoned hugepages to prevent them
> +	 * from being migrated by memory hotremove.
> +	 */
> +	if (count_increased)
> +		ClearHPageMigratable(head);

I believe this can prevent hwpoisoned hugepages from being migrated though there still be some windows.

> +
>  	return ret;
>  out:
>  	if (count_increased)
> @@ -1862,6 +1869,7 @@ static int try_memory_failure_hugetlb(unsigned long pfn, int flags, int *hugetlb
>  
>  	if (hwpoison_filter(p)) {
>  		hugetlb_clear_page_hwpoison(head);
> +		SetHPageMigratable(head);

Would we set HPageMigratable flag for free hugetlb pages here? IIUC, they're not expected to have this flag set.

Thanks,
Miaohe Lin

>  		unlock_page(head);
>  		if (res == 1)
>  			put_page(head);
>