mm: Fix for movable_node boot option

From: Masayoshi Mizuma <[email protected]>

commit 124049decbb1 ("x86/e820: put !E820_TYPE_RAM regions into memblock.reserved") breaks movable_node kernel option because it changed the memory gap range to reserved memblock. So, the node is marked as Normal zone even if the SRAT has Hot plaggable affinity.

    =====================================================================
    kernel: BIOS-e820: [mem 0x0000180000000000-0x0000180fffffffff] usable
    kernel: BIOS-e820: [mem 0x00001c0000000000-0x00001c0fffffffff] usable
    ...
    kernel: reserved[0x12]#011[0x0000181000000000-0x00001bffffffffff], 0x000003f000000000 bytes flags: 0x0
    ...
    kernel: ACPI: SRAT: Node 2 PXM 6 [mem 0x180000000000-0x1bffffffffff] hotplug
    kernel: ACPI: SRAT: Node 3 PXM 7 [mem 0x1c0000000000-0x1fffffffffff] hotplug
    ...
    kernel: Movable zone start for each node
    kernel:  Node 3: 0x00001c0000000000
    kernel: Early memory node ranges
    ...
    =====================================================================

Naoya's v1 patch [*] fixes the original issue and this movable_node issue doesn't occur.
Let's revert commit 124049decbb1 ("x86/e820: put !E820_TYPE_RAM regions into memblock.reserved") and apply the v1 patch.

[*] https://lkml.org/lkml/2018/6/13/27

Signed-off-by Masayoshi Mizuma <[email protected]>

Reviewed-by Pavel Tatashin <[email protected]>

---
 arch/x86/kernel/e820.c | 15 +++------------
 1 file changed, 3 insertions(+), 12 deletions(-)

Quoted 29 lines · Masayoshi Mizuma

From: Masayoshi Mizuma <[email protected]>

commit 124049decbb1 ("x86/e820: put !E820_TYPE_RAM regions into memblock.reserved") breaks movable_node kernel option because it changed the memory gap range to reserved memblock. So, the node is marked as Normal zone even if the SRAT has Hot plaggable affinity.

    =====================================================================
    kernel: BIOS-e820: [mem 0x0000180000000000-0x0000180fffffffff] usable
    kernel: BIOS-e820: [mem 0x00001c0000000000-0x00001c0fffffffff] usable
    ...
    kernel: reserved[0x12]#011[0x0000181000000000-0x00001bffffffffff], 0x000003f000000000 bytes flags: 0x0
    ...
    kernel: ACPI: SRAT: Node 2 PXM 6 [mem 0x180000000000-0x1bffffffffff] hotplug
    kernel: ACPI: SRAT: Node 3 PXM 7 [mem 0x1c0000000000-0x1fffffffffff] hotplug
    ...
    kernel: Movable zone start for each node
    kernel:  Node 3: 0x00001c0000000000
    kernel: Early memory node ranges
    ...
    =====================================================================

Naoya's v1 patch [*] fixes the original issue and this movable_node issue doesn't occur.
Let's revert commit 124049decbb1 ("x86/e820: put !E820_TYPE_RAM regions into memblock.reserved") and apply the v1 patch.

[*] https://lkml.org/lkml/2018/6/13/27

Quoted 5 lines

Signed-off-by Masayoshi Mizuma <[email protected]>

Reviewed-by Pavel Tatashin <[email protected]>

---
 arch/x86/kernel/e820.c | 15 +++------------
 1 file changed, 3 insertions(+), 12 deletions(-)

Bad ordering which introduces the bug and thus breaks bisection of related issues: the fixes should come first, then the revert of the unnecessary or bad fix.

From: Pavel Tatashin <[email protected]>

When checking for valid pfns in zero_resv_unavail(), it is not necessary to verify that pfns within pageblock_nr_pages ranges are valid, only the first one needs to be checked. This is because memory for pages are allocated in contiguous chunks that contain pageblock_nr_pages struct pages.

Signed-off-by Pavel Tatashin <[email protected]>

Reviewed-off-by Masayoshi Mizuma <[email protected]>

---
 mm/page_alloc.c | 46 ++++++++++++++++++++++++++--------------------
 1 file changed, 26 insertions(+), 20 deletions(-)

mm/page_alloc.c +26 -20

index 3b9d89e..bd5b7e4 100644
@@ -6440,6 +6440,29 @@ void __init free_area_init_node(int nid, unsigned long *zones_size,
 }
 
 #if defined(CONFIG_HAVE_MEMBLOCK) && !defined(CONFIG_FLAT_NODE_MEM_MAP)
+
+/*
+ * Zero all valid struct pages in range [spfn, epfn), return number of struct
+ * pages zeroed
+ */
+static u64 zero_pfn_range(unsigned long spfn, unsigned long epfn)
+{
+	unsigned long pfn;
+	u64 pgcnt = 0;
+
+	for (pfn = spfn; pfn < epfn; pfn++) {
+		if (!pfn_valid(ALIGN_DOWN(pfn, pageblock_nr_pages))) {
+			pfn = ALIGN_DOWN(pfn, pageblock_nr_pages)
+				+ pageblock_nr_pages - 1;
+			continue;
+		}
+		mm_zero_struct_page(pfn_to_page(pfn));
+		pgcnt++;
+	}
+
+	return pgcnt;
+}
+
 /*
  * Only struct pages that are backed by physical memory are zeroed and
  * initialized by going through __init_single_page(). But, there are some

@@ -6455,7 +6478,6 @@ void __init free_area_init_node(int nid, unsigned long *zones_size,
 void __init zero_resv_unavail(void)
 {
 	phys_addr_t start, end;
-	unsigned long pfn;
 	u64 i, pgcnt;
 	phys_addr_t next = 0;
 

@@ -6465,34 +6487,18 @@ void __init zero_resv_unavail(void)
 	pgcnt = 0;
 	for_each_mem_range(i, &memblock.memory, NULL,
 			NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end, NULL) {
-		if (next < start) {
-			for (pfn = PFN_DOWN(next); pfn < PFN_UP(start); pfn++) {
-				if (!pfn_valid(ALIGN_DOWN(pfn, pageblock_nr_pages)))
-					continue;
-				mm_zero_struct_page(pfn_to_page(pfn));
-				pgcnt++;
-			}
-		}
+		if (next < start)
+			pgcnt += zero_pfn_range(PFN_DOWN(next), PFN_UP(start));
 		next = end;
 	}
-	for (pfn = PFN_DOWN(next); pfn < max_pfn; pfn++) {
-		if (!pfn_valid(ALIGN_DOWN(pfn, pageblock_nr_pages)))
-			continue;
-		mm_zero_struct_page(pfn_to_page(pfn));
-		pgcnt++;
-	}
-
+	pgcnt += zero_pfn_range(PFN_DOWN(next), max_pfn);
 
 	/*
 	 * Struct pages that do not have backing memory. This could be because
 	 * firmware is using some of this memory, or for some other reasons.
-	 * Once memblock is changed so such behaviour is not allowed: i.e.
-	 * list of "reserved" memory must be a subset of list of "memory", then
-	 * this code can be removed.
 	 */
 	if (pgcnt)
 		pr_info("Zeroed struct page in unavailable ranges: %lld pages", pgcnt);
-
 }
 #endif /* CONFIG_HAVE_MEMBLOCK && !CONFIG_FLAT_NODE_MEM_MAP */
 

Signature · 1 line

2.18.0

Quoted 7 lines · Masayoshi Mizuma

From: Pavel Tatashin <[email protected]>

When checking for valid pfns in zero_resv_unavail(), it is not necessary to verify that pfns within pageblock_nr_pages ranges are valid, only the first one needs to be checked. This is because memory for pages are allocated in contiguous chunks that contain pageblock_nr_pages struct pages.

Quoted 2 lines

Signed-off-by Pavel Tatashin <[email protected]>

Reviewed-off-by Masayoshi Mizuma <[email protected]>

According to convention, review tag is formatted like "Reviewed-by: ...",

Otherwise, looks good to me.

Acked-by Naoya Horiguchi <[email protected]>

Quoted 89 lines

---
 mm/page_alloc.c | 46 ++++++++++++++++++++++++++--------------------
 1 file changed, 26 insertions(+), 20 deletions(-)

mm/page_alloc.c +26 -20

index 3b9d89e..bd5b7e4 100644
@@ -6440,6 +6440,29 @@ void __init free_area_init_node(int nid, unsigned long *zones_size,
 }
 
 #if defined(CONFIG_HAVE_MEMBLOCK) && !defined(CONFIG_FLAT_NODE_MEM_MAP)
+
+/*
+ * Zero all valid struct pages in range [spfn, epfn), return number of struct
+ * pages zeroed
+ */
+static u64 zero_pfn_range(unsigned long spfn, unsigned long epfn)
+{
+	unsigned long pfn;
+	u64 pgcnt = 0;
+
+	for (pfn = spfn; pfn < epfn; pfn++) {
+		if (!pfn_valid(ALIGN_DOWN(pfn, pageblock_nr_pages))) {
+			pfn = ALIGN_DOWN(pfn, pageblock_nr_pages)
+				+ pageblock_nr_pages - 1;
+			continue;
+		}
+		mm_zero_struct_page(pfn_to_page(pfn));
+		pgcnt++;
+	}
+
+	return pgcnt;
+}
+
 /*
  * Only struct pages that are backed by physical memory are zeroed and
  * initialized by going through __init_single_page(). But, there are some

@@ -6455,7 +6478,6 @@ void __init free_area_init_node(int nid, unsigned long *zones_size,
 void __init zero_resv_unavail(void)
 {
 	phys_addr_t start, end;
-	unsigned long pfn;
 	u64 i, pgcnt;
 	phys_addr_t next = 0;
 

@@ -6465,34 +6487,18 @@ void __init zero_resv_unavail(void)
 	pgcnt = 0;
 	for_each_mem_range(i, &memblock.memory, NULL,
 			NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end, NULL) {
-		if (next < start) {
-			for (pfn = PFN_DOWN(next); pfn < PFN_UP(start); pfn++) {
-				if (!pfn_valid(ALIGN_DOWN(pfn, pageblock_nr_pages)))
-					continue;
-				mm_zero_struct_page(pfn_to_page(pfn));
-				pgcnt++;
-			}
-		}
+		if (next < start)
+			pgcnt += zero_pfn_range(PFN_DOWN(next), PFN_UP(start));
 		next = end;
 	}
-	for (pfn = PFN_DOWN(next); pfn < max_pfn; pfn++) {
-		if (!pfn_valid(ALIGN_DOWN(pfn, pageblock_nr_pages)))
-			continue;
-		mm_zero_struct_page(pfn_to_page(pfn));
-		pgcnt++;
-	}
-
+	pgcnt += zero_pfn_range(PFN_DOWN(next), max_pfn);
 
 	/*
 	 * Struct pages that do not have backing memory. This could be because
 	 * firmware is using some of this memory, or for some other reasons.
-	 * Once memblock is changed so such behaviour is not allowed: i.e.
-	 * list of "reserved" memory must be a subset of list of "memory", then
-	 * this code can be removed.
 	 */
 	if (pgcnt)
 		pr_info("Zeroed struct page in unavailable ranges: %lld pages", pgcnt);
-
 }
 #endif /* CONFIG_HAVE_MEMBLOCK && !CONFIG_FLAT_NODE_MEM_MAP */
 

Signature · 1 line

2.18.0

This patch series are the fix for movable_node boot option issue which was introduced by commit 124049decbb1 ("x86/e820:
put !E820_TYPE_RAM regions into memblock.reserved").

First patch, revert the commit. Second and third patch fix the original issue.

Quoted 6 lines · Masayoshi Mizuma

This patch series are the fix for movable_node boot option issue which was introduced by commit 124049decbb1 ("x86/e820:
put !E820_TYPE_RAM regions into memblock.reserved").

First patch, revert the commit. Second and third patch fix the original issue.

Can the mm folks please comment on this?

Quoted 6 lines · Thomas Gleixner

Quoted 4 lines · Masayoshi Mizuma

This patch series are the fix for movable_node boot option issue which was introduced by commit 124049decbb1 ("x86/e820:
put !E820_TYPE_RAM regions into memblock.reserved").

Quoted 4 lines

Quoted 2 lines

First patch, revert the commit. Second and third patch fix the original issue.

Can the mm folks please comment on this?

I was under impression that Pavel who authored the original change which got reverted here has reviewed these patches.

Quoted 10 lines · Thomas Gleixner

Quoted 6 lines · Masayoshi Mizuma

This patch series are the fix for movable_node boot option issue which was introduced by commit 124049decbb1 ("x86/e820:
put !E820_TYPE_RAM regions into memblock.reserved").

First patch, revert the commit. Second and third patch fix the original issue.

Can the mm folks please comment on this?

I was under impression that Pavel who authored the original change which got reverted here has reviewed these patches.