[v1,12/12] mm: make MAX_ORDER a kernel boot time parameter.

Commit Message

Zi Yan Sept. 22, 2022, 1:12 a.m. UTC

From: Zi Yan <[e-mail]>

With the new buddy_alloc_max_order, users can specify larger MAX_ORDER
than set in CONFIG_ARCH_MAX_ORDER or CONFIG_SET_MAX_ORDER.
It can be set any value >= CONFIG_ARCH_MAX_ORDER or CONFIG_SET_MAX_ORDER,
but < 256 (limited by vmscan scan_control and per-cpu free page list).

Signed-off-by: Zi Yan <[e-mail]>
Cc: Jonathan Corbet <[e-mail]>
Cc: "Paul E. McKenney" <[e-mail]>
Cc: Randy Dunlap <[e-mail]>
Cc: Thomas Gleixner <[e-mail]>
Cc: Vlastimil Babka <[e-mail]>
Cc: [e-mail]
Cc: [e-mail]
Cc: [e-mail]
---
 .../admin-guide/kernel-parameters.txt         |  5 +++
 include/linux/mmzone.h                        | 10 +++++-
 mm/Kconfig                                    | 13 ++++++++
 mm/page_alloc.c                               | 31 +++++++++++++++++++
 mm/vmscan.c                                   |  1 -
 5 files changed, 58 insertions(+), 2 deletions(-)

Patch

diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
index 5f633844daac..eb0dd8a78205 100644
--- a/Documentation/admin-guide/kernel-parameters.txt
+++ b/Documentation/admin-guide/kernel-parameters.txt
@@ -494,6 +494,11 @@ 
 	bttv.pll=	See Documentation/admin-guide/media/bttv.rst
 	bttv.tuner=
 
+	buddy_alloc_max_order=	[KNL] This parameter adjusts the size of largest
+			pages that can be allocated from kernel buddy allocator. The largest
+			page size is 2^buddy_alloc_max_order * PAGE_SIZE.
+			Format: integer
+
 	bulk_remove=off	[PPC]  This parameter disables the use of the pSeries
 			firmware feature for flushing multiple hpte entries
 			at a time.
diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
index 19fca391f635..5669191d15dc 100644
--- a/include/linux/mmzone.h
+++ b/include/linux/mmzone.h
@@ -35,6 +35,14 @@ 
 #define MIN_MAX_ORDER MAX_ORDER
 #endif
 
+/* remap MAX_ORDER to buddy_alloc_max_order for boot time adjustment */
+#ifdef CONFIG_BOOT_TIME_MAX_ORDER
+/* Defined in mm/page_alloc.c */
+extern int buddy_alloc_max_order;
+#undef MAX_ORDER
+#define MAX_ORDER buddy_alloc_max_order
+#endif /* CONFIG_BOOT_TIME_MAX_ORDER */
+
 #define MAX_ORDER_NR_PAGES (1 << MAX_ORDER)
 
 /*
@@ -1600,7 +1608,7 @@  static inline bool movable_only_nodes(nodemask_t *nodes)
  * contiguous, thus > section size pages can be allocated and manipulated
  * without worrying about non-contiguous struct page.
  */
-#ifndef CONFIG_SET_MAX_ORDER
+#if !defined(CONFIG_SET_MAX_ORDER) && !defined(CONFIG_BOOT_TIME_MAX_ORDER)
 #if (MAX_ORDER + PAGE_SHIFT) > SECTION_SIZE_BITS
 #error Allocator MAX_ORDER exceeds SECTION_SIZE
 #endif
diff --git a/mm/Kconfig b/mm/Kconfig
index 9c7280acd528..3e6b61ba9fec 100644
--- a/mm/Kconfig
+++ b/mm/Kconfig
@@ -449,6 +449,19 @@  config SET_MAX_ORDER
 	  increase this value. A value of 10 means that the largest free memory
 	  block is 2^10 pages.
 
+config BOOT_TIME_MAX_ORDER
+	bool "Set maximum order of buddy allocator at boot time"
+	depends on SPARSEMEM_VMEMMAP && (ARCH_FORCE_MAX_ORDER != 0 || SET_MAX_ORDER != 0)
+	help
+	  It enables users to set the maximum order of buddy allocator at system
+	  boot time instead of a static MACRO set at compilation time. Systems with
+	  a lot of memory might want to allocate large pages whereas it is much
+	  less feasible and desirable for systems with less memory. This option
+	  allows different systems to control the largest page they want to
+	  allocate. By default, MAX_ORDER will be set to ARCH_FORCE_MAX_ORDER or
+	  SET_MAX_ORDER, whichever is non-zero, when the boot time parameter is not
+	  set. The maximum of MAX_ORDER is currently limited at 256.
+
 config HAVE_MEMBLOCK_PHYS_MAP
 	bool
 
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index ba7c284ba3d3..9eacdf3a37c4 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -9720,3 +9720,34 @@  bool has_managed_dma(void)
 	return false;
 }
 #endif /* CONFIG_ZONE_DMA */
+
+#ifdef CONFIG_BOOT_TIME_MAX_ORDER
+int buddy_alloc_max_order = MIN_MAX_ORDER;
+EXPORT_SYMBOL(buddy_alloc_max_order);
+
+static int __init buddy_alloc_set(char *val)
+{
+	int ret;
+	unsigned long max_order;
+
+	ret = kstrtoul(val, 10, &max_order);
+
+	if (ret < 0)
+		return -EINVAL;
+
+	/*
+	 * max_order is also limited at below locations:
+	 * 1. scan_control in mm/vmscan.c uses s8 field for order, max_order cannot
+	 * be bigger than S8_MAX before the field is changed.
+	 * 2. free_pcppages_bulk has max_order upper limit.
+	 */
+	if (max_order > MIN_MAX_ORDER && max_order <= S8_MAX)
+		buddy_alloc_max_order = max_order;
+	else
+		buddy_alloc_max_order = MIN_MAX_ORDER;
+
+	return 0;
+}
+
+early_param("buddy_alloc_max_order", buddy_alloc_set);
+#endif /* CONFIG_BOOT_TIME_MAX_ORDER */
diff --git a/mm/vmscan.c b/mm/vmscan.c
index a8fd6300fa7e..009632243398 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -6623,7 +6623,6 @@  unsigned long try_to_free_pages(struct zonelist *zonelist, int order,
 	 * scan_control uses s8 fields for order, priority, and reclaim_idx.
 	 * Confirm they are large enough for max values.
 	 */
-	BUILD_BUG_ON(MAX_ORDER > S8_MAX);
 	BUILD_BUG_ON(DEF_PRIORITY > S8_MAX);
 	BUILD_BUG_ON(MAX_NR_ZONES > S8_MAX);