Re: [mm-unstable PATCH v4 4/9] mm, hwpoison, hugetlb: support saving mechanism of raw error pages

["HORIGUCHI NAOYA(堀口 直也)" <naoya.horiguchi@nec.com>]

On Wed, Jul 06, 2022 at 10:37:50AM +0800, Miaohe Lin wrote:
> On 2022/7/4 9:33, Naoya Horiguchi wrote:
> > From: Naoya Horiguchi <naoya.horiguchi@nec.com>
> > 
> > When handling memory error on a hugetlb page, the error handler tries to
> > dissolve and turn it into 4kB pages.  If it's successfully dissolved,
> > PageHWPoison flag is moved to the raw error page, so that's all right.
> > However, dissolve sometimes fails, then the error page is left as
> > hwpoisoned hugepage. It's useful if we can retry to dissolve it to save
> > healthy pages, but that's not possible now because the information about
> > where the raw error pages is lost.
> > 
> > Use the private field of a few tail pages to keep that information.  The
> > code path of shrinking hugepage pool uses this info to try delayed dissolve.
> > In order to remember multiple errors in a hugepage, a singly-linked list
> > originated from SUBPAGE_INDEX_HWPOISON-th tail page is constructed.  Only
> > simple operations (adding an entry or clearing all) are required and the
> > list is assumed not to be very long, so this simple data structure should
> > be enough.
> > 
> > If we failed to save raw error info, the hwpoison hugepage has errors on
> > unknown subpage, then this new saving mechanism does not work any more,
> > so disable saving new raw error info and freeing hwpoison hugepages.
> > 
> > Signed-off-by: Naoya Horiguchi <naoya.horiguchi@nec.com>
> > ---
...
> > diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> > index 66bb39e0fce8..ccd470f0194c 100644
> > --- a/mm/hugetlb.c
> > +++ b/mm/hugetlb.c
> > @@ -1535,17 +1535,15 @@ static void __update_and_free_page(struct hstate *h, struct page *page)
> >  	if (hstate_is_gigantic(h) && !gigantic_page_runtime_supported())
> >  		return;
> >  
> > -	if (hugetlb_vmemmap_restore(h, page)) {
> > -		spin_lock_irq(&hugetlb_lock);
> > -		/*
> > -		 * If we cannot allocate vmemmap pages, just refuse to free the
> > -		 * page and put the page back on the hugetlb free list and treat
> > -		 * as a surplus page.
> > -		 */
> > -		add_hugetlb_page(h, page, true);
> > -		spin_unlock_irq(&hugetlb_lock);
> > -		return;
> > -	}
> > +	if (hugetlb_vmemmap_restore(h, page))
> > +		goto fail;
> > +
> > +	/*
> > +	 * Move PageHWPoison flag from head page to the raw error pages,
> > +	 * which makes any healthy subpages reusable.
> > +	 */
> > +	if (unlikely(PageHWPoison(page) && hugetlb_clear_page_hwpoison(page)))
> > +		goto fail;
> 
> IIUC, HPageVmemmapOptimized must have been cleared via hugetlb_vmemmap_restore above. So
> VM_BUG_ON_PAGE(!HPageVmemmapOptimized(page), page) in add_hugetlb_page will be triggered
> if we go to fail here. add_hugetlb_page is expected to be called when we cannot allocate
> vmemmap pages.

Thanks a lot, you're right. I shouldn't simply factor the failure path with
the goto label.  I think that it's hard to undo hugetlb_vmemmap_restore(), so
checking HPageRawHwpUnreliable() before hugetlb_vmemmap_restore(), then
try to hugetlb_clear_page_hwpoison() after it (where tail pages are available).

> 
> >  
> >  	for (i = 0; i < pages_per_huge_page(h);
> >  	     i++, subpage = mem_map_next(subpage, page, i)) {
> > @@ -1566,6 +1564,16 @@ static void __update_and_free_page(struct hstate *h, struct page *page)
> >  	} else {
> >  		__free_pages(page, huge_page_order(h));
> >  	}
> > +	return;
> > +fail:
> > +	spin_lock_irq(&hugetlb_lock);
> > +	/*
> > +	 * If we cannot allocate vmemmap pages or cannot identify raw hwpoison
> > +	 * subpages reliably, just refuse to free the page and put the page
> > +	 * back on the hugetlb free list and treat as a surplus page.
> > +	 */
> > +	add_hugetlb_page(h, page, true);
> > +	spin_unlock_irq(&hugetlb_lock);
> >  }
> >  
> >  /*
> > @@ -2109,15 +2117,6 @@ int dissolve_free_huge_page(struct page *page)
> >  		 */
> >  		rc = hugetlb_vmemmap_restore(h, head);
> >  		if (!rc) {
> > -			/*
> > -			 * Move PageHWPoison flag from head page to the raw
> > -			 * error page, which makes any subpages rather than
> > -			 * the error page reusable.
> > -			 */
> > -			if (PageHWPoison(head) && page != head) {
> > -				SetPageHWPoison(page);
> > -				ClearPageHWPoison(head);
> > -			}
> >  			update_and_free_page(h, head, false);
> >  		} else {
> >  			spin_lock_irq(&hugetlb_lock);
> > diff --git a/mm/memory-failure.c b/mm/memory-failure.c
> > index c9931c676335..53bf7486a245 100644
> > --- a/mm/memory-failure.c
> > +++ b/mm/memory-failure.c
> > @@ -1664,6 +1664,82 @@ int mf_dax_kill_procs(struct address_space *mapping, pgoff_t index,
> >  EXPORT_SYMBOL_GPL(mf_dax_kill_procs);
> >  #endif /* CONFIG_FS_DAX */
> >  
> > +/*
> > + * Struct raw_hwp_page represents information about "raw error page",
> > + * constructing singly linked list originated from ->private field of
> > + * SUBPAGE_INDEX_HWPOISON-th tail page.
> > + */
> > +struct raw_hwp_page {
> > +	struct llist_node node;
> > +	struct page *page;
> > +};
> > +
> > +static inline struct llist_head *raw_hwp_list_head(struct page *hpage)
> > +{
> > +	return (struct llist_head *)&page_private(hpage + SUBPAGE_INDEX_HWPOISON);
> > +}
> > +
> > +static inline int hugetlb_set_page_hwpoison(struct page *hpage,
> > +					struct page *page)
> > +{
> > +	struct llist_head *head;
> > +	struct raw_hwp_page *raw_hwp;
> > +	struct llist_node *t, *tnode;
> > +	int ret;
> > +
> > +	/*
> > +	 * Once the hwpoison hugepage has lost reliable raw error info,
> > +	 * there is little meaning to keep additional error info precisely,
> > +	 * so skip to add additional raw error info.
> > +	 */
> > +	if (HPageRawHwpUnreliable(hpage))
> > +		return -EHWPOISON;
> 
> If we return here, num_poisoned_pages can't reflect the all hwpoisoned hugepages?

No, it can't now.  Currently we try to (and fail) count only the number
hwpoison subpages with raw_hwp_info.  If we want to track all corrupted
pages (including hwpoison subpages without raw_hwp_info), maybe running the
following part first in this function could make it better.

	ret = TestSetPageHWPoison(hpage) ? -EHWPOISON : 0;
	/* the first error event will be counted in action_result(). */
	if (ret)
		num_poisoned_pages_inc();

But I like the option you suggest below.

> 
> > +	head = raw_hwp_list_head(hpage);
> > +	llist_for_each_safe(tnode, t, head->first) {
> > +		struct raw_hwp_page *p = container_of(tnode, struct raw_hwp_page, node);
> > +
> > +		if (p->page == page)
> > +			return -EHWPOISON;
> > +	}
> > +
> > +	ret = TestSetPageHWPoison(hpage) ? -EHWPOISON : 0;
> > +	/* the first error event will be counted in action_result(). */
> > +	if (ret)
> > +		num_poisoned_pages_inc();
> > +
> > +	raw_hwp = kmalloc(sizeof(struct raw_hwp_page), GFP_ATOMIC);
> > +	if (raw_hwp) {
> > +		raw_hwp->page = page;
> > +		llist_add(&raw_hwp->node, head);
> 
> IMHO, we might need to do num_poisoned_pages_inc here because we decrement the
> num_poisoned_pages according to the llist length.

Yes, if we'd like to count only hwpoisoned subpages with raw_hwp_info,
doing num_poisoned_pages_inc here is fine.

> > +	} else {
> > +		/*
> > +		 * Failed to save raw error info.  We no longer trace all
> > +		 * hwpoisoned subpages, and we need refuse to free/dissolve
> > +		 * this hwpoisoned hugepage.
> > +		 */
> > +		SetHPageRawHwpUnreliable(hpage);
> > +	}
> > +	return ret;
> > +}
> > +
> > +inline int hugetlb_clear_page_hwpoison(struct page *hpage)
> 
> off-the-topic: Is "inline" needed here? I see hugetlb_clear_page_hwpoison is "extern" above.

Maybe not, this code is not performance-sensitive and actually this inline
is a leftover of updates in the previous versions. I'll remove this.

> 
> > +{
> > +	struct llist_head *head;
> > +	struct llist_node *t, *tnode;
> > +
> > +	if (!HPageRawHwpUnreliable(hpage))
> > +		ClearPageHWPoison(hpage);
> > +	head = raw_hwp_list_head(hpage);
> > +	llist_for_each_safe(tnode, t, head->first) {
> > +		struct raw_hwp_page *p = container_of(tnode, struct raw_hwp_page, node);
> > +
> > +		SetPageHWPoison(p->page);
> 
> IMHO, in HPageRawHwpUnreliable(hpage) case, it's better not to do SetPageHWPoison here.
> Because hugepage won't be dissolved and thus we cannot write any data to some tail struct
> pages if HugeTLB Vmemmap Optimization is enabled. Freeing the memory here should be enough.

This is a good point, too. Current version surely does not work with HVO, so
I think of simply giving up clearing hwpoison when HPageVmemmapOptimized is
true. And I should leave some inline comment about this.

> > +		kfree(p);
> > +	}
> > +	llist_del_all(head);
> > +	return 0;
> > +}
> 
> I thought num_poisoned_pages_dec is missing and return value is unneeded. But this is changed
> in next patch. So it should be fine here.

OK.

> 
> > +
> >  /*
> >   * Called from hugetlb code with hugetlb_lock held.
> >   *
> > @@ -1698,7 +1774,7 @@ int __get_huge_page_for_hwpoison(unsigned long pfn, int flags)
> >  		goto out;
> >  	}
> >  
> > -	if (TestSetPageHWPoison(head)) {
> > +	if (hugetlb_set_page_hwpoison(head, page)) {
> >  		ret = -EHWPOISON;
> >  		goto out;
> >  	}
> > @@ -1751,7 +1827,7 @@ static int try_memory_failure_hugetlb(unsigned long pfn, int flags, int *hugetlb
> >  	lock_page(head);
> >  
> >  	if (hwpoison_filter(p)) {
> > -		ClearPageHWPoison(head);
> > +		hugetlb_clear_page_hwpoison(head);
> >  		res = -EOPNOTSUPP;
> >  		goto out;
> >  	}
> > 
> 
> Many thanks for your hard work. :)

Thanks for the detailed review and feedback.

- Naoya Horiguchi