linux-kernel - Re: [PATCH 1/1] mm/madvise: enhance lazyfreeing with mTHP in madvise

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-Id: <20240226083714.26187-1-ioworker0@gmail.com>
Date: Mon, 26 Feb 2024 16:37:14 +0800
From: Lance Yang <ioworker0@...il.com>
To: 21cnbao@...il.com
Cc: akpm@...ux-foundation.org,
	david@...hat.com,
	ioworker0@...il.com,
	linux-kernel@...r.kernel.org,
	linux-mm@...ck.org,
	mhocko@...e.com,
	minchan@...nel.org,
	peterx@...hat.com,
	ryan.roberts@....com,
	shy828301@...il.com,
	songmuchun@...edance.com,
	wangkefeng.wang@...wei.com,
	zokeefe@...gle.com,
	fengwei.yin@...el.com
Subject: Re: [PATCH 1/1] mm/madvise: enhance lazyfreeing with mTHP in madvise_free

Hey Barry,

Thanks for taking time to review!

On Mon, Feb 26, 2024 at 12:00 PM Barry Song <21cnbao@...il.com> wrote:
[...]
> On Mon, Feb 26, 2024 at 1:33 AM Lance Yang <ioworker0@...il.com> wrote:
[...]
> We did something similar on MADV_PAGEOUT[1]
> [1] https://lore.kernel.org/linux-mm/20240118111036.72641-7-21cnbao@gmail.com/

Thanks for providing the link above.

[...]
> > +                        * Avoid unnecessary folio splitting if the large
> > +                        * folio is entirely within the given range.
> > +                        */
> > +                       folio_test_clear_dirty(folio);
> > +                       folio_unlock(folio);
> > +                       for (; addr != next_addr; pte++, addr += PAGE_SIZE) {
> > +                               ptent = ptep_get(pte);
> > +                               if (pte_young(ptent) || pte_dirty(ptent)) {
> > +                                       ptent = ptep_get_and_clear_full(
> > +                                               mm, addr, pte, tlb->fullmm);
> > +                                       ptent = pte_mkold(ptent);
> > +                                       ptent = pte_mkclean(ptent);
> > +                                       set_pte_at(mm, addr, pte, ptent);
> > +                                       tlb_remove_tlb_entry(tlb, pte, addr);
> > +                               }
>
> The code works under the assumption the large folio is entirely mapped
> in all PTEs in the range. This is not always true.
>
> This won't work in some cases as some PTEs might be mapping to the
> large folios. some others might have been unmapped or mapped
> to different folios.
>
> so in MADV_PAGEOUT, we have a function to check the folio is
> really entirely mapped:
>
> +static inline bool pte_range_cont_mapped(unsigned long start_pfn,
> + pte_t *start_pte, unsigned long start_addr, int nr)
> +{
> +              int i;
> +              pte_t pte_val;
> +
> +              for (i = 0; i < nr; i++) {
> +                           pte_val = ptep_get(start_pte + i);
> +
> +                           if (pte_none(pte_val))
> +                                        return false;
> +
> +                           if (pte_pfn(pte_val) != (start_pfn + i))
> +                                        return false;
> +              }
> +
> +              return true;
> +}

Thanks for providing the information; it's very helpful to me!
I made some changes. Would you mind taking another look, please?

As a diff against this patch.

diff --git a/mm/madvise.c b/mm/madvise.c
index bcbf56595a2e..255d2f329be4 100644
--- a/mm/madvise.c
+++ b/mm/madvise.c
@@ -616,6 +616,18 @@ static long madvise_pageout(struct vm_area_struct *vma,
 	return 0;
 }
 
+static inline bool pte_range_cont_mapped(pte_t *pte, unsigned long nr)
+{
+	pte_t pte_val;
+	unsigned long pfn = pte_pfn(pte);
+	for (int i = 0; i < nr; i++) {
+		pte_val = ptep_get(pte + i);
+		if (pte_none(pte_val) || pte_pfn(pte_val) != (pfn + i))
+			return false;
+	}
+	return true;
+}
+
 static int madvise_free_pte_range(pmd_t *pmd, unsigned long addr,
 				unsigned long end, struct mm_walk *walk)
 
@@ -676,20 +688,25 @@ static int madvise_free_pte_range(pmd_t *pmd, unsigned long addr,
 		 */
 		if (folio_test_large(folio)) {
 			int err;
-			unsigned long next_addr, align;
+			unsigned long nr, next_addr, align;
 
 			if (folio_estimated_sharers(folio) != 1 ||
 			    !folio_trylock(folio))
 				goto skip_large_folio;
 
-			align = folio_nr_pages(folio) * PAGE_SIZE;
+			nr = folio_nr_pages(folio);
+			align = nr * PAGE_SIZE;
 			next_addr = ALIGN_DOWN(addr + align, align);
 
 			/*
-			 * If we mark only the subpages as lazyfree,
-			 * split the large folio.
+			 * If we mark only the subpages as lazyfree, or
+			 * if there is a cow folio associated with this folio,
+			 * or if this folio is not really entirely mapped,
+			 * then split the large folio.
 			 */
-			if (next_addr > end || next_addr - addr != align)
+			if (next_addr > end || next_addr - addr != align ||
+			    folio_total_mapcount(folio) != nr ||
+			    pte_range_cont_mapped(pte, nr))
 				goto split_large_folio;
 
 			/*
---

Thanks again for your time!

Best,
Lance