Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Feedback-ID: iccd040f4:Fastmail
From:   Zi Yan <zi.yan@sent.com>
To:     linux-mm@kvack.org
Cc:     David Hildenbrand <david@redhat.com>,
        Matthew Wilcox <willy@infradead.org>,
        Vlastimil Babka <vbabka@suse.cz>,
        "Kirill A . Shutemov" <kirill.shutemov@linux.intel.com>,
        Mike Kravetz <mike.kravetz@oracle.com>,
        John Hubbard <jhubbard@nvidia.com>,
        Yang Shi <shy828301@gmail.com>,
        David Rientjes <rientjes@google.com>,
        James Houghton <jthoughton@google.com>,
        Mike Rapoport <rppt@kernel.org>, linux-kernel@vger.kernel.org
Subject: [RFC PATCH v2 04/12] mm: adapt deferred struct page init to new MAX_ORDER.
Date:   Thu, 11 Aug 2022 19:16:35 -0400
Message-Id: <20220811231643.1012912-5-zi.yan@sent.com>
In-Reply-To: <20220811231643.1012912-1-zi.yan@sent.com>
References: <20220811231643.1012912-1-zi.yan@sent.com>
Reply-To: Zi Yan <ziy@nvidia.com>
MIME-Version: 1.0
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

From: Zi Yan <ziy@nvidia.com>

deferred_init only initializes first section of a zone and defers the
rest and the rest of the zone will be initialized in size of a section.
When MAX_ORDER grows beyond a section size, early_page_uninitialised()
did not prevent pages beyond first section from initialization, since it
only checked the starting pfn and assumes MAX_ORDER is smaller than
a section size. In addition, deferred_init_maxorder() uses
MAX_ORDER_NR_PAGES as the initialization unit, which can cause the
initialized chunk of memory overlapping with other initialization jobs.

For the first issue, make early_page_uninitialised() decrease the order
for non-deferred memory initialization when it is bigger than first
section. For the second issue, when adjust pfn alignment in
deferred_init_maxorder(), make sure the alignment is not bigger than
a section size.

Signed-off-by: Zi Yan <ziy@nvidia.com>
---
 mm/internal.h   |  2 +-
 mm/memblock.c   |  6 ++++--
 mm/page_alloc.c | 26 +++++++++++++++++++-------
 3 files changed, 24 insertions(+), 10 deletions(-)

diff --git a/mm/internal.h b/mm/internal.h
index 1433e3a6fdd0..cbe745670c6e 100644
--- a/mm/internal.h
+++ b/mm/internal.h
@@ -355,7 +355,7 @@ extern int __isolate_free_page(struct page *page, unsig=
ned int order);
 extern void __putback_isolated_page(struct page *page, unsigned int order,
 				    int mt);
 extern void memblock_free_pages(struct page *page, unsigned long pfn,
-					unsigned int order);
+					unsigned int *order);
 extern void __free_pages_core(struct page *page, unsigned int order);
 extern void prep_compound_page(struct page *page, unsigned int order);
 extern void post_alloc_hook(struct page *page, unsigned int order,
diff --git a/mm/memblock.c b/mm/memblock.c
index d1525463c05e..dc2ce6df8fe3 100644
--- a/mm/memblock.c
+++ b/mm/memblock.c
@@ -1640,7 +1640,9 @@ void __init memblock_free_late(phys_addr_t base, phys=
_addr_t size)
 	end =3D PFN_DOWN(base + size);
=20
 	for (; cursor < end; cursor++) {
-		memblock_free_pages(pfn_to_page(cursor), cursor, 0);
+		unsigned int order =3D 0;
+
+		memblock_free_pages(pfn_to_page(cursor), cursor, &order);
 		totalram_pages_inc();
 	}
 }
@@ -2035,7 +2037,7 @@ static void __init __free_pages_memory(unsigned long =
start, unsigned long end)
 		while (start + (1UL << order) > end)
 			order--;
=20
-		memblock_free_pages(pfn_to_page(start), start, order);
+		memblock_free_pages(pfn_to_page(start), start, &order);
=20
 		start +=3D (1UL << order);
 	}
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 07ad8074950f..3f3af7cd5164 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -463,13 +463,19 @@ static inline bool deferred_pages_enabled(void)
 }
=20
 /* Returns true if the struct page for the pfn is uninitialised */
-static inline bool __meminit early_page_uninitialised(unsigned long pfn)
+static inline bool __meminit early_page_uninitialised(unsigned long pfn, u=
nsigned int *order)
 {
 	int nid =3D early_pfn_to_nid(pfn);
=20
 	if (node_online(nid) && pfn >=3D NODE_DATA(nid)->first_deferred_pfn)
 		return true;
=20
+	/* clamp down order to not exceed first_deferred_pfn */
+	if (order)
+		*order =3D min_t(unsigned int,
+			       *order,
+			       ilog2(NODE_DATA(nid)->first_deferred_pfn - pfn));
+
 	return false;
 }
=20
@@ -515,7 +521,7 @@ static inline bool deferred_pages_enabled(void)
 	return false;
 }
=20
-static inline bool early_page_uninitialised(unsigned long pfn)
+static inline bool early_page_uninitialised(unsigned long pfn, unsigned in=
t *order)
 {
 	return false;
 }
@@ -1644,7 +1650,7 @@ static void __meminit init_reserved_page(unsigned lon=
g pfn)
 	pg_data_t *pgdat;
 	int nid, zid;
=20
-	if (!early_page_uninitialised(pfn))
+	if (!early_page_uninitialised(pfn, NULL))
 		return;
=20
 	nid =3D early_pfn_to_nid(pfn);
@@ -1800,11 +1806,11 @@ int __meminit early_pfn_to_nid(unsigned long pfn)
 #endif /* CONFIG_NUMA */
=20
 void __init memblock_free_pages(struct page *page, unsigned long pfn,
-							unsigned int order)
+							unsigned int *order)
 {
-	if (early_page_uninitialised(pfn))
+	if (early_page_uninitialised(pfn, order))
 		return;
-	__free_pages_core(page, order);
+	__free_pages_core(page, *order);
 }
=20
 /*
@@ -2030,7 +2036,13 @@ static unsigned long __init
 deferred_init_maxorder(u64 *i, struct zone *zone, unsigned long *start_pfn,
 		       unsigned long *end_pfn)
 {
-	unsigned long mo_pfn =3D ALIGN(*start_pfn + 1, MAX_ORDER_NR_PAGES);
+	/*
+	 * deferred_init_memmap_chunk gives out jobs with max size to
+	 * PAGES_PER_SECTION. Do not align mo_pfn beyond that.
+	 */
+	unsigned long align =3D min_t(unsigned long,
+				MAX_ORDER_NR_PAGES, PAGES_PER_SECTION);
+	unsigned long mo_pfn =3D ALIGN(*start_pfn + 1, align);
 	unsigned long spfn =3D *start_pfn, epfn =3D *end_pfn;
 	unsigned long nr_pages =3D 0;
 	u64 j =3D *i;
--=20
2.35.1