[PATCH v2 00/28] hugetlb/CMA improvements for large systems

[PATCH v2 00/28] hugetlb/CMA improvements for large systems

[Frank van der Linden]

On large systems, we observed some issues with hugetlb and CMA:

1) When specifying a large number of hugetlb boot pages (hugepages=
   on the commandline), the kernel may run out of memory before it
   even gets to HVO. For example, if you have a 3072G system, and
   want to use 3024 1G hugetlb pages for VMs, that should leave
   you plenty of space for the hypervisor, provided you have the
   hugetlb vmemmap optimization (HVO) enabled. However, since
   the vmemmap pages are always allocated first, and then later
   in boot freed, you will actually run yourself out of memory
   before you can do HVO. This means not getting all the hugetlb
   pages you want, and worse, failure to boot if there is an
   allocation failure in the system from which it can't recover.

2) There is a system setup where you might want to use hugetlb_cma
   with a large value (say, again, 3024 out of 3072G like above),
   and then lower that if system usage allows it, to make room
   for non-hugetlb processes. For this, a variation of the problem
   above applies: the kernel runs out of unmovable space to allocate
   from before you finish boot, since your CMA area takes up all
   the space.

3) CMA wants to use one big contiguous area for allocations. Which
   fails if you have the aforementioned 3T system with a gap in the
   middle of physical memory (like the < 40bits BIOS DMA area seen on
   some AMD systems). You then won't be able to set up a CMA area for
   one of the NUMA nodes, leading to loss of half of your hugetlb
   CMA area.

4) Under the scenario mentioned in 2), when trying to grow the
   number of hugetlb pages after dropping it for a while, new
   CMA allocations may fail occasionally. This is not unexpected,
   some transient references on pages may prevent cma_alloc
   from succeeding under memory pressure. However, the hugetlb
   code then falls back to a normal contiguous alloc, which may
   end up succeeding. This is not always desired behavior. If
   you have a large CMA area, then the kernel has a restricted
   amount of memory it can do unmovable allocations from (a well
   known issue). A normal contiguous alloc may eat further in to
   this space.


To resolve these issues, do the following:

* Add hooks to the section init code to do custom initialization
  of memmap pages. Hugetlb bootmem (memblock) allocated pages can
  then be pre-HVOed. This avoids allocating a large number of
  vmemmap pages early in boot, only to have them be freed again
  later, and also avoids running out of memory as described under 1).
  Using these hooks for hugetlb is optional. It requires moving
  hugetlb bootmem allocation to an earlier spot by the architecture.
  This has been enabled on x86.
* hugetlb_cma doesn't care about the CMA area it uses being one
  large contiguous range. Multiple smaller ranges are fine. The
  only requirements are that the areas should be on one NUMA node,
  and individual gigantic pages should be allocatable from them. So,
  implement multi-range support for CMA, avoiding issue 3).
* Introduce a hugetlb_cma_only option on the commandline. This only
  allows allocations from CMA for gigantic pages, if hugetlb_cma=
  is also specified.
* With hugetlb_cma_only active, it also makes sense to be able to
  pre-allocate gigantic hugetlb pages at boot time from the CMA
  area(s). Add a rudimentary early CMA allocation interface, that
  just grabs a piece of memblock-allocated space from the CMA
  area, which gets marked as allocated in the CMA bitmap
  when the CMA area is initialized. With this, hugepages= can
  be supported with hugetlb_cma=, making scenario 2) work.

Additionally, fix some minor bugs, with one worth mentioning:
since hugetlb gigantic bootmem pages are allocated by memblock,
they may span multiple zones, as memblock doesn't (and mostly
can't) know about zones. This can cause problems. A hugetlb page
spanning multiple zones is bad, and it's worse with HVO, when
the de-HVO step effectively sneakily re-assigns pages to a
different zone than originally configured, since the tail pages
all inherit the zone from the first 60 tail pages. This condition
is not common, but can be easily reproduced using ZONE_MOVABLE.
To fix this, add checks to see if gigantic bootmem pages intersect
with multiple zones, and do not use them if they do, giving them
back to the page allocator instead.

The first patch is kind of along for the ride, except that
maintaining an available_count for a CMA area is convenient
for the multiple range support.

v2:
* Add missing CMA debugfs code.
* Minor cleanups in hugetlb_cma changes.
* Move hugetlb_cma code to its own file to further clean
  things up.

Frank van der Linden (28):
  mm/cma: export total and free number of pages for CMA areas
  mm, cma: support multiple contiguous ranges, if requested
  mm/cma: introduce cma_intersects function
  mm, hugetlb: use cma_declare_contiguous_multi
  mm/hugetlb: fix round-robin bootmem allocation
  mm/hugetlb: remove redundant __ClearPageReserved
  mm/hugetlb: use online nodes for bootmem allocation
  mm/hugetlb: convert cmdline parameters from setup to early
  x86/mm: make register_page_bootmem_memmap handle PTE mappings
  mm/bootmem_info: export register_page_bootmem_memmap
  mm/sparse: allow for alternate vmemmap section init at boot
  mm/hugetlb: set migratetype for bootmem folios
  mm: define __init_reserved_page_zone function
  mm/hugetlb: check bootmem pages for zone intersections
  mm/sparse: add vmemmap_*_hvo functions
  mm/hugetlb: deal with multiple calls to hugetlb_bootmem_alloc
  mm/hugetlb: move huge_boot_pages list init to hugetlb_bootmem_alloc
  mm/hugetlb: add pre-HVO framework
  mm/hugetlb_vmemmap: fix hugetlb_vmemmap_restore_folios definition
  mm/hugetlb: do pre-HVO for bootmem allocated pages
  x86/setup: call hugetlb_bootmem_alloc early
  x86/mm: set ARCH_WANT_SPARSEMEM_VMEMMAP_PREINIT
  mm/cma: simplify zone intersection check
  mm/cma: introduce a cma validate function
  mm/cma: introduce interface for early reservations
  mm/hugetlb: add hugetlb_cma_only cmdline option
  mm/hugetlb: enable bootmem allocation from CMA areas
  mm/hugetlb: move hugetlb CMA code in to its own file

 Documentation/ABI/testing/sysfs-kernel-mm-cma |  13 +
 .../admin-guide/kernel-parameters.txt         |   7 +
 arch/powerpc/include/asm/book3s/64/hugetlb.h  |   6 +
 arch/powerpc/mm/hugetlbpage.c                 |   1 +
 arch/powerpc/mm/init_64.c                     |   1 +
 arch/s390/mm/init.c                           |  13 +-
 arch/x86/Kconfig                              |   1 +
 arch/x86/kernel/setup.c                       |   4 +-
 arch/x86/mm/init_64.c                         |  16 +-
 include/linux/bootmem_info.h                  |   7 +
 include/linux/cma.h                           |   9 +
 include/linux/hugetlb.h                       |  35 +
 include/linux/mm.h                            |  13 +-
 include/linux/mmzone.h                        |  35 +
 mm/Kconfig                                    |   8 +
 mm/Makefile                                   |   3 +
 mm/bootmem_info.c                             |   4 +-
 mm/cma.c                                      | 749 +++++++++++++++---
 mm/cma.h                                      |  46 +-
 mm/cma_debug.c                                |  61 +-
 mm/cma_sysfs.c                                |  20 +
 mm/hugetlb.c                                  | 566 +++++++------
 mm/hugetlb_cma.c                              | 258 ++++++
 mm/hugetlb_cma.h                              |  55 ++
 mm/hugetlb_vmemmap.c                          | 199 ++++-
 mm/hugetlb_vmemmap.h                          |  23 +-
 mm/internal.h                                 |  19 +
 mm/mm_init.c                                  |  78 +-
 mm/sparse-vmemmap.c                           | 168 +++-
 mm/sparse.c                                   |  87 +-
 30 files changed, 2016 insertions(+), 489 deletions(-)
 create mode 100644 mm/hugetlb_cma.c
 create mode 100644 mm/hugetlb_cma.h

-- 
2.48.1.262.g85cc9f2d1e-goog

[PATCH v2 01/28] mm/cma: export total and free number of pages for CMA areas

[Frank van der Linden]

In addition to the number of allocations and releases, system
management software may like to be aware of the size of CMA
areas, and how many pages are available in it. This information
is currently not available, so export it in total_page and
available_pages, respectively.

The name 'available_pages' was picked over 'free_pages' because
'free' implies that the pages are unused. But they might not
be, they just haven't been used by cma_alloc

The number of available pages is tracked regardless of
CONFIG_CMA_SYSFS, allowing for a few minor shortcuts in
the code, avoiding bitmap operations.

Signed-off-by: Frank van der Linden <fvdl@google.com>
---
 Documentation/ABI/testing/sysfs-kernel-mm-cma | 13 +++++++++++
 mm/cma.c                                      | 22 ++++++++++++++-----
 mm/cma.h                                      |  1 +
 mm/cma_debug.c                                |  5 +----
 mm/cma_sysfs.c                                | 20 +++++++++++++++++
 5 files changed, 51 insertions(+), 10 deletions(-)

diff --git a/Documentation/ABI/testing/sysfs-kernel-mm-cma b/Documentation/ABI/testing/sysfs-kernel-mm-cma
index dfd755201142..aaf2a5d8b13b 100644
--- a/Documentation/ABI/testing/sysfs-kernel-mm-cma
+++ b/Documentation/ABI/testing/sysfs-kernel-mm-cma
@@ -29,3 +29,16 @@ Date:		Feb 2024
 Contact:	Anshuman Khandual <anshuman.khandual@arm.com>
 Description:
 		the number of pages CMA API succeeded to release
+
+What:		/sys/kernel/mm/cma/<cma-heap-name>/total_pages
+Date:		Jun 2024
+Contact:	Frank van der Linden <fvdl@google.com>
+Description:
+		The size of the CMA area in pages.
+
+What:		/sys/kernel/mm/cma/<cma-heap-name>/available_pages
+Date:		Jun 2024
+Contact:	Frank van der Linden <fvdl@google.com>
+Description:
+		The number of pages in the CMA area that are still
+		available for CMA allocation.
diff --git a/mm/cma.c b/mm/cma.c
index de5bc0c81fc2..95a8788e54d3 100644
--- a/mm/cma.c
+++ b/mm/cma.c
@@ -86,6 +86,7 @@ static void cma_clear_bitmap(struct cma *cma, unsigned long pfn,
 
 	spin_lock_irqsave(&cma->lock, flags);
 	bitmap_clear(cma->bitmap, bitmap_no, bitmap_count);
+	cma->available_count += count;
 	spin_unlock_irqrestore(&cma->lock, flags);
 }
 
@@ -133,7 +134,7 @@ static void __init cma_activate_area(struct cma *cma)
 			free_reserved_page(pfn_to_page(pfn));
 	}
 	totalcma_pages -= cma->count;
-	cma->count = 0;
+	cma->available_count = cma->count = 0;
 	pr_err("CMA area %s could not be activated\n", cma->name);
 }
 
@@ -206,7 +207,7 @@ int __init cma_init_reserved_mem(phys_addr_t base, phys_addr_t size,
 		snprintf(cma->name, CMA_MAX_NAME,  "cma%d\n", cma_area_count);
 
 	cma->base_pfn = PFN_DOWN(base);
-	cma->count = size >> PAGE_SHIFT;
+	cma->available_count = cma->count = size >> PAGE_SHIFT;
 	cma->order_per_bit = order_per_bit;
 	*res_cma = cma;
 	cma_area_count++;
@@ -390,7 +391,7 @@ static void cma_debug_show_areas(struct cma *cma)
 {
 	unsigned long next_zero_bit, next_set_bit, nr_zero;
 	unsigned long start = 0;
-	unsigned long nr_part, nr_total = 0;
+	unsigned long nr_part;
 	unsigned long nbits = cma_bitmap_maxno(cma);
 
 	spin_lock_irq(&cma->lock);
@@ -402,12 +403,12 @@ static void cma_debug_show_areas(struct cma *cma)
 		next_set_bit = find_next_bit(cma->bitmap, nbits, next_zero_bit);
 		nr_zero = next_set_bit - next_zero_bit;
 		nr_part = nr_zero << cma->order_per_bit;
-		pr_cont("%s%lu@%lu", nr_total ? "+" : "", nr_part,
+		pr_cont("%s%lu@%lu", start ? "+" : "", nr_part,
 			next_zero_bit);
-		nr_total += nr_part;
 		start = next_zero_bit + nr_zero;
 	}
-	pr_cont("=> %lu free of %lu total pages\n", nr_total, cma->count);
+	pr_cont("=> %lu free of %lu total pages\n", cma->available_count,
+			cma->count);
 	spin_unlock_irq(&cma->lock);
 }
 
@@ -444,6 +445,14 @@ static struct page *__cma_alloc(struct cma *cma, unsigned long count,
 
 	for (;;) {
 		spin_lock_irq(&cma->lock);
+		/*
+		 * If the request is larger than the available number
+		 * of pages, stop right away.
+		 */
+		if (count > cma->available_count) {
+			spin_unlock_irq(&cma->lock);
+			break;
+		}
 		bitmap_no = bitmap_find_next_zero_area_off(cma->bitmap,
 				bitmap_maxno, start, bitmap_count, mask,
 				offset);
@@ -452,6 +461,7 @@ static struct page *__cma_alloc(struct cma *cma, unsigned long count,
 			break;
 		}
 		bitmap_set(cma->bitmap, bitmap_no, bitmap_count);
+		cma->available_count -= count;
 		/*
 		 * It's safe to drop the lock here. We've marked this region for
 		 * our exclusive use. If the migration fails we will take the
diff --git a/mm/cma.h b/mm/cma.h
index 8485ef893e99..3dd3376ae980 100644
--- a/mm/cma.h
+++ b/mm/cma.h
@@ -13,6 +13,7 @@ struct cma_kobject {
 struct cma {
 	unsigned long   base_pfn;
 	unsigned long   count;
+	unsigned long	available_count;
 	unsigned long   *bitmap;
 	unsigned int order_per_bit; /* Order of pages represented by one bit */
 	spinlock_t	lock;
diff --git a/mm/cma_debug.c b/mm/cma_debug.c
index 602fff89b15f..89236f22230a 100644
--- a/mm/cma_debug.c
+++ b/mm/cma_debug.c
@@ -34,13 +34,10 @@ DEFINE_DEBUGFS_ATTRIBUTE(cma_debugfs_fops, cma_debugfs_get, NULL, "%llu\n");
 static int cma_used_get(void *data, u64 *val)
 {
 	struct cma *cma = data;
-	unsigned long used;
 
 	spin_lock_irq(&cma->lock);
-	/* pages counter is smaller than sizeof(int) */
-	used = bitmap_weight(cma->bitmap, (int)cma_bitmap_maxno(cma));
+	*val = cma->count - cma->available_count;
 	spin_unlock_irq(&cma->lock);
-	*val = (u64)used << cma->order_per_bit;
 
 	return 0;
 }
diff --git a/mm/cma_sysfs.c b/mm/cma_sysfs.c
index f50db3973171..97acd3e5a6a5 100644
--- a/mm/cma_sysfs.c
+++ b/mm/cma_sysfs.c
@@ -62,6 +62,24 @@ static ssize_t release_pages_success_show(struct kobject *kobj,
 }
 CMA_ATTR_RO(release_pages_success);
 
+static ssize_t total_pages_show(struct kobject *kobj,
+					  struct kobj_attribute *attr, char *buf)
+{
+	struct cma *cma = cma_from_kobj(kobj);
+
+	return sysfs_emit(buf, "%lu\n", cma->count);
+}
+CMA_ATTR_RO(total_pages);
+
+static ssize_t available_pages_show(struct kobject *kobj,
+					  struct kobj_attribute *attr, char *buf)
+{
+	struct cma *cma = cma_from_kobj(kobj);
+
+	return sysfs_emit(buf, "%lu\n", cma->available_count);
+}
+CMA_ATTR_RO(available_pages);
+
 static void cma_kobj_release(struct kobject *kobj)
 {
 	struct cma *cma = cma_from_kobj(kobj);
@@ -75,6 +93,8 @@ static struct attribute *cma_attrs[] = {
 	&alloc_pages_success_attr.attr,
 	&alloc_pages_fail_attr.attr,
 	&release_pages_success_attr.attr,
+	&total_pages_attr.attr,
+	&available_pages_attr.attr,
 	NULL,
 };
 ATTRIBUTE_GROUPS(cma);
-- 
2.48.1.262.g85cc9f2d1e-goog

[PATCH v2 02/28] mm, cma: support multiple contiguous ranges, if requested

[Frank van der Linden]

Currently, CMA manages one range of physically contiguous memory.
Creation of larger CMA areas with hugetlb_cma may run in to gaps
in physical memory, so that they are not able to allocate that
contiguous physical range from memblock when creating the CMA
area.

This can happen, for example, on an AMD system with > 1TB of memory,
where there will be a gap just below the 1TB (40bit DMA) line. If
you have set aside most of memory for potential hugetlb CMA allocation,
cma_declare_contiguous_nid will fail.

hugetlb_cma doesn't need the entire area to be one physically
contiguous range. It just cares about being able to get physically
contiguous chunks of a certain size (e.g. 1G), and it is fine
to have the CMA area backed by multiple physical ranges, as
long as it gets 1G contiguous allocations.

Multi-range support is implemented by introducing an array of
ranges, instead of just one big one. Each range has its own bitmap.
Effectively, the allocate and release operations work as before,
just per-range. So, instead of going through one large bitmap, they
now go through a number of smaller ones.

The maximum number of supported ranges is 8, as defined in
CMA_MAX_RANGES.

Since some current users of CMA expect a CMA area to just use one
physically contiguous range, only allow for multiple ranges if a
new interface, cma_declare_contiguous_nid_multi, is used. The other
interfaces will work like before, creating only CMA areas with
1 range.

cma_declare_contiguous_nid_multi works as follows, mimicking the
default "bottom-up, above 4G" reservation approach:

0) Try cma_declare_contiguous_nid, which will use only one
   region. If this succeeds, return. This makes sure that for
   all the cases that currently work, the behavior remains
   unchanged even if the caller switches from
   cma_declare_contiguous_nid to cma_declare_contiguous_nid_multi.
1) Select the largest free memblock ranges above 4G, with
   a maximum number of CMA_MAX_RANGES.
2) If we did not find at most CMA_MAX_RANGES that add
   up to the total size requested, return -ENOMEM.
3) Sort the selected ranges by base address.
4) Reserve them bottom-up until we get what we wanted.

Signed-off-by: Frank van der Linden <fvdl@google.com>
---
 include/linux/cma.h |   3 +
 mm/cma.c            | 604 +++++++++++++++++++++++++++++++++++---------
 mm/cma.h            |  27 +-
 mm/cma_debug.c      |  56 ++--
 4 files changed, 552 insertions(+), 138 deletions(-)

diff --git a/include/linux/cma.h b/include/linux/cma.h
index d15b64f51336..863427c27dc2 100644
--- a/include/linux/cma.h
+++ b/include/linux/cma.h
@@ -40,6 +40,9 @@ static inline int __init cma_declare_contiguous(phys_addr_t base,
 	return cma_declare_contiguous_nid(base, size, limit, alignment,
 			order_per_bit, fixed, name, res_cma, NUMA_NO_NODE);
 }
+extern int __init cma_declare_contiguous_multi(phys_addr_t size,
+			phys_addr_t align, unsigned int order_per_bit,
+			const char *name, struct cma **res_cma, int nid);
 extern int cma_init_reserved_mem(phys_addr_t base, phys_addr_t size,
 					unsigned int order_per_bit,
 					const char *name,
diff --git a/mm/cma.c b/mm/cma.c
index 95a8788e54d3..c20255161642 100644
--- a/mm/cma.c
+++ b/mm/cma.c
@@ -18,6 +18,7 @@
 
 #include <linux/memblock.h>
 #include <linux/err.h>
+#include <linux/list.h>
 #include <linux/mm.h>
 #include <linux/sizes.h>
 #include <linux/slab.h>
@@ -35,9 +36,16 @@ struct cma cma_areas[MAX_CMA_AREAS];
 unsigned int cma_area_count;
 static DEFINE_MUTEX(cma_mutex);
 
+static int __init __cma_declare_contiguous_nid(phys_addr_t base,
+			phys_addr_t size, phys_addr_t limit,
+			phys_addr_t alignment, unsigned int order_per_bit,
+			bool fixed, const char *name, struct cma **res_cma,
+			int nid);
+
 phys_addr_t cma_get_base(const struct cma *cma)
 {
-	return PFN_PHYS(cma->base_pfn);
+	WARN_ON_ONCE(cma->nranges != 1);
+	return PFN_PHYS(cma->ranges[0].base_pfn);
 }
 
 unsigned long cma_get_size(const struct cma *cma)
@@ -63,9 +71,10 @@ static unsigned long cma_bitmap_aligned_mask(const struct cma *cma,
  * The value returned is represented in order_per_bits.
  */
 static unsigned long cma_bitmap_aligned_offset(const struct cma *cma,
+					       const struct cma_memrange *cmr,
 					       unsigned int align_order)
 {
-	return (cma->base_pfn & ((1UL << align_order) - 1))
+	return (cmr->base_pfn & ((1UL << align_order) - 1))
 		>> cma->order_per_bit;
 }
 
@@ -75,46 +84,57 @@ static unsigned long cma_bitmap_pages_to_bits(const struct cma *cma,
 	return ALIGN(pages, 1UL << cma->order_per_bit) >> cma->order_per_bit;
 }
 
-static void cma_clear_bitmap(struct cma *cma, unsigned long pfn,
-			     unsigned long count)
+static void cma_clear_bitmap(struct cma *cma, const struct cma_memrange *cmr,
+			     unsigned long pfn, unsigned long count)
 {
 	unsigned long bitmap_no, bitmap_count;
 	unsigned long flags;
 
-	bitmap_no = (pfn - cma->base_pfn) >> cma->order_per_bit;
+	bitmap_no = (pfn - cmr->base_pfn) >> cma->order_per_bit;
 	bitmap_count = cma_bitmap_pages_to_bits(cma, count);
 
 	spin_lock_irqsave(&cma->lock, flags);
-	bitmap_clear(cma->bitmap, bitmap_no, bitmap_count);
+	bitmap_clear(cmr->bitmap, bitmap_no, bitmap_count);
 	cma->available_count += count;
 	spin_unlock_irqrestore(&cma->lock, flags);
 }
 
 static void __init cma_activate_area(struct cma *cma)
 {
-	unsigned long base_pfn = cma->base_pfn, pfn;
+	unsigned long pfn, base_pfn;
+	int allocrange, r;
 	struct zone *zone;
+	struct cma_memrange *cmr;
+
+	for (allocrange = 0; allocrange < cma->nranges; allocrange++) {
+		cmr = &cma->ranges[allocrange];
+		cmr->bitmap = bitmap_zalloc(cma_bitmap_maxno(cma, cmr),
+					    GFP_KERNEL);
+		if (!cmr->bitmap)
+			goto cleanup;
+	}
 
-	cma->bitmap = bitmap_zalloc(cma_bitmap_maxno(cma), GFP_KERNEL);
-	if (!cma->bitmap)
-		goto out_error;
+	for (r = 0; r < cma->nranges; r++) {
+		cmr = &cma->ranges[r];
+		base_pfn = cmr->base_pfn;
 
-	/*
-	 * alloc_contig_range() requires the pfn range specified to be in the
-	 * same zone. Simplify by forcing the entire CMA resv range to be in the
-	 * same zone.
-	 */
-	WARN_ON_ONCE(!pfn_valid(base_pfn));
-	zone = page_zone(pfn_to_page(base_pfn));
-	for (pfn = base_pfn + 1; pfn < base_pfn + cma->count; pfn++) {
-		WARN_ON_ONCE(!pfn_valid(pfn));
-		if (page_zone(pfn_to_page(pfn)) != zone)
-			goto not_in_zone;
-	}
+		/*
+		 * alloc_contig_range() requires the pfn range specified
+		 * to be in the same zone. Simplify by forcing the entire
+		 * CMA resv range to be in the same zone.
+		 */
+		WARN_ON_ONCE(!pfn_valid(base_pfn));
+		zone = page_zone(pfn_to_page(base_pfn));
+		for (pfn = base_pfn + 1; pfn < base_pfn + cmr->count; pfn++) {
+			WARN_ON_ONCE(!pfn_valid(pfn));
+			if (page_zone(pfn_to_page(pfn)) != zone)
+				goto cleanup;
+		}
 
-	for (pfn = base_pfn; pfn < base_pfn + cma->count;
-	     pfn += pageblock_nr_pages)
-		init_cma_reserved_pageblock(pfn_to_page(pfn));
+		for (pfn = base_pfn; pfn < base_pfn + cmr->count;
+		     pfn += pageblock_nr_pages)
+			init_cma_reserved_pageblock(pfn_to_page(pfn));
+	}
 
 	spin_lock_init(&cma->lock);
 
@@ -125,13 +145,19 @@ static void __init cma_activate_area(struct cma *cma)
 
 	return;
 
-not_in_zone:
-	bitmap_free(cma->bitmap);
-out_error:
+cleanup:
+	for (r = 0; r < allocrange; r++)
+		bitmap_free(cma->ranges[r].bitmap);
+
 	/* Expose all pages to the buddy, they are useless for CMA. */
 	if (!cma->reserve_pages_on_error) {
-		for (pfn = base_pfn; pfn < base_pfn + cma->count; pfn++)
-			free_reserved_page(pfn_to_page(pfn));
+		for (r = 0; r < allocrange; r++) {
+			cmr = &cma->ranges[r];
+			for (pfn = cmr->base_pfn;
+			     pfn < cmr->base_pfn + cmr->count;
+			     pfn++)
+				free_reserved_page(pfn_to_page(pfn));
+		}
 	}
 	totalcma_pages -= cma->count;
 	cma->available_count = cma->count = 0;
@@ -154,6 +180,43 @@ void __init cma_reserve_pages_on_error(struct cma *cma)
 	cma->reserve_pages_on_error = true;
 }
 
+static int __init cma_new_area(const char *name, phys_addr_t size,
+			       unsigned int order_per_bit,
+			       struct cma **res_cma)
+{
+	struct cma *cma;
+
+	if (cma_area_count == ARRAY_SIZE(cma_areas)) {
+		pr_err("Not enough slots for CMA reserved regions!\n");
+		return -ENOSPC;
+	}
+
+	/*
+	 * Each reserved area must be initialised later, when more kernel
+	 * subsystems (like slab allocator) are available.
+	 */
+	cma = &cma_areas[cma_area_count];
+	cma_area_count++;
+
+	if (name)
+		snprintf(cma->name, CMA_MAX_NAME, name);
+	else
+		snprintf(cma->name, CMA_MAX_NAME,  "cma%d\n", cma_area_count);
+
+	cma->available_count = cma->count = size >> PAGE_SHIFT;
+	cma->order_per_bit = order_per_bit;
+	*res_cma = cma;
+	totalcma_pages += cma->count;
+
+	return 0;
+}
+
+static void __init cma_drop_area(struct cma *cma)
+{
+	totalcma_pages -= cma->count;
+	cma_area_count--;
+}
+
 /**
  * cma_init_reserved_mem() - create custom contiguous area from reserved memory
  * @base: Base address of the reserved area
@@ -172,13 +235,9 @@ int __init cma_init_reserved_mem(phys_addr_t base, phys_addr_t size,
 				 struct cma **res_cma)
 {
 	struct cma *cma;
+	int ret;
 
 	/* Sanity checks */
-	if (cma_area_count == ARRAY_SIZE(cma_areas)) {
-		pr_err("Not enough slots for CMA reserved regions!\n");
-		return -ENOSPC;
-	}
-
 	if (!size || !memblock_is_region_reserved(base, size))
 		return -EINVAL;
 
@@ -195,25 +254,261 @@ int __init cma_init_reserved_mem(phys_addr_t base, phys_addr_t size,
 	if (!IS_ALIGNED(base | size, CMA_MIN_ALIGNMENT_BYTES))
 		return -EINVAL;
 
+	ret = cma_new_area(name, size, order_per_bit, &cma);
+	if (ret != 0)
+		return ret;
+
+	cma->ranges[0].base_pfn = PFN_DOWN(base);
+	cma->ranges[0].count = cma->count;
+	cma->nranges = 1;
+
+	*res_cma = cma;
+
+	return 0;
+}
+
+/*
+ * Structure used while walking physical memory ranges and finding out
+ * which one(s) to use for a CMA area.
+ */
+struct cma_init_memrange {
+	phys_addr_t base;
+	phys_addr_t size;
+	struct list_head list;
+};
+
+/*
+ * Work array used during CMA initialization.
+ */
+static struct cma_init_memrange memranges[CMA_MAX_RANGES] __initdata;
+
+static bool __init revsizecmp(struct cma_init_memrange *mlp,
+			      struct cma_init_memrange *mrp)
+{
+	return mlp->size > mrp->size;
+}
+
+static bool __init basecmp(struct cma_init_memrange *mlp,
+			   struct cma_init_memrange *mrp)
+{
+	return mlp->base < mrp->base;
+}
+
+/*
+ * Helper function to create sorted lists.
+ */
+static void __init list_insert_sorted(
+	struct list_head *ranges,
+	struct cma_init_memrange *mrp,
+	bool (*cmp)(struct cma_init_memrange *lh, struct cma_init_memrange *rh))
+{
+	struct list_head *mp;
+	struct cma_init_memrange *mlp;
+
+	if (list_empty(ranges))
+		list_add(&mrp->list, ranges);
+	{
+		list_for_each(mp, ranges) {
+			mlp = list_entry(mp, struct cma_init_memrange, list);
+			if (cmp(mlp, mrp))
+				break;
+		}
+		__list_add(&mrp->list, mlp->list.prev, &mlp->list);
+	}
+}
+
+/*
+ * Create CMA areas with a total size of @total_size. A normal allocation
+ * for one area is tried first. If that fails, the biggest memblock
+ * ranges above 4G are selected, and allocated bottom up.
+ *
+ * The complexity here is not great, but this function will only be
+ * called during boot, and the lists operated on have fewer than
+ * CMA_MAX_RANGES elements (default value: 8).
+ */
+int __init cma_declare_contiguous_multi(phys_addr_t total_size,
+			phys_addr_t align, unsigned int order_per_bit,
+			const char *name, struct cma **res_cma, int nid)
+{
+	phys_addr_t start, end;
+	phys_addr_t size, sizesum, sizeleft;
+	struct cma_init_memrange *mrp, *mlp, *failed;
+	struct cma_memrange *cmrp;
+	LIST_HEAD(ranges);
+	LIST_HEAD(final_ranges);
+	struct list_head *mp, *next;
+	int ret, nr = 1;
+	u64 i;
+	struct cma *cma;
+
 	/*
-	 * Each reserved area must be initialised later, when more kernel
-	 * subsystems (like slab allocator) are available.
+	 * First, try it the normal way, producing just one range.
 	 */
-	cma = &cma_areas[cma_area_count];
+	ret = __cma_declare_contiguous_nid(0, total_size, 0, align,
+			order_per_bit, false, name, res_cma, nid);
+	if (ret != -ENOMEM)
+		goto out;
 
-	if (name)
-		snprintf(cma->name, CMA_MAX_NAME, name);
-	else
-		snprintf(cma->name, CMA_MAX_NAME,  "cma%d\n", cma_area_count);
+	/*
+	 * Couldn't find one range that fits our needs, so try multiple
+	 * ranges.
+	 *
+	 * No need to do the alignment checks here, the call to
+	 * cma_declare_contiguous_nid above would have caught
+	 * any issues. With the checks, we know that:
+	 *
+	 * - @align is a power of 2
+	 * - @align is >= pageblock alignment
+	 * - @size is aligned to @align and to @order_per_bit
+	 *
+	 * So, as long as we create ranges that have a base
+	 * aligned to @align, and a size that is aligned to
+	 * both @align and @order_to_bit, things will work out.
+	 */
+	nr = 0;
+	sizesum = 0;
+	failed = NULL;
 
-	cma->base_pfn = PFN_DOWN(base);
-	cma->available_count = cma->count = size >> PAGE_SHIFT;
-	cma->order_per_bit = order_per_bit;
+	ret = cma_new_area(name, total_size, order_per_bit, &cma);
+	if (ret != 0)
+		goto out;
+
+	align = max_t(phys_addr_t, align, CMA_MIN_ALIGNMENT_BYTES);
+	/*
+	 * Create a list of ranges above 4G, largest range first.
+	 */
+	for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &start, &end, NULL) {
+		if (start < SZ_4G)
+			continue;
+
+		start = ALIGN(start, align);
+		if (start >= end)
+			continue;
+
+		end = ALIGN_DOWN(end, align);
+		if (end <= start)
+			continue;
+
+		size = end - start;
+		size = ALIGN_DOWN(size, (PAGE_SIZE << order_per_bit));
+		if (!size)
+			continue;
+		sizesum += size;
+
+		pr_debug("consider %016llx - %016llx\n", (u64)start, (u64)end);
+
+		/*
+		 * If we don't yet have used the maximum number of
+		 * areas, grab a new one.
+		 *
+		 * If we can't use anymore, see if this range is not
+		 * smaller than the smallest one already recorded. If
+		 * not, re-use the smallest element.
+		 */
+		if (nr < CMA_MAX_RANGES)
+			mrp = &memranges[nr++];
+		else {
+			mrp = list_last_entry(&ranges,
+					      struct cma_init_memrange, list);
+			if (size < mrp->size)
+				continue;
+			list_del(&mrp->list);
+			sizesum -= mrp->size;
+			pr_debug("deleted %016llx - %016llx from the list\n",
+				(u64)mrp->base, (u64)mrp->base + size);
+		}
+		mrp->base = start;
+		mrp->size = size;
+
+		/*
+		 * Now do a sorted insert.
+		 */
+		list_insert_sorted(&ranges, mrp, revsizecmp);
+		pr_debug("added %016llx - %016llx to the list\n",
+		    (u64)mrp->base, (u64)mrp->base + size);
+		pr_debug("total size now %llu\n", (u64)sizesum);
+	}
+
+	/*
+	 * There is not enough room in the CMA_MAX_RANGES largest
+	 * ranges, so bail out.
+	 */
+	if (sizesum < total_size) {
+		cma_drop_area(cma);
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	/*
+	 * Found ranges that provide enough combined space.
+	 * Now, sorted them by address, smallest first, because we
+	 * want to mimic a bottom-up memblock allocation.
+	 */
+	sizesum = 0;
+	list_for_each_safe(mp, next, &ranges) {
+		mlp = list_entry(mp, struct cma_init_memrange, list);
+		list_del(mp);
+		list_insert_sorted(&final_ranges, mlp, basecmp);
+		sizesum += mlp->size;
+		if (sizesum >= total_size)
+			break;
+	}
+
+	/*
+	 * Walk the final list, and add a CMA range for
+	 * each range, possibly not using the last one fully.
+	 */
+	nr = 0;
+	sizeleft = total_size;
+	list_for_each(mp, &final_ranges) {
+		mlp = list_entry(mp, struct cma_init_memrange, list);
+		size = min(sizeleft, mlp->size);
+		if (memblock_reserve(mlp->base, size)) {
+			/*
+			 * Unexpected error. Could go on to
+			 * the next one, but just abort to
+			 * be safe.
+			 */
+			failed = mlp;
+			break;
+		}
+
+		pr_debug("created region %d: %016llx - %016llx\n",
+		    nr, (u64)mlp->base, (u64)mlp->base + size);
+		cmrp = &cma->ranges[nr++];
+		cmrp->base_pfn = PHYS_PFN(mlp->base);
+		cmrp->count = size >> PAGE_SHIFT;
+
+		sizeleft -= size;
+		if (sizeleft == 0)
+			break;
+	}
+
+	if (failed) {
+		list_for_each(mp, &final_ranges) {
+			mlp = list_entry(mp, struct cma_init_memrange, list);
+			if (mlp == failed)
+				break;
+			memblock_phys_free(mlp->base, mlp->size);
+		}
+		cma_drop_area(cma);
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	cma->nranges = nr;
 	*res_cma = cma;
-	cma_area_count++;
-	totalcma_pages += cma->count;
 
-	return 0;
+out:
+	if (ret != 0)
+		pr_err("Failed to reserve %lu MiB\n",
+			(unsigned long)total_size / SZ_1M);
+	else
+		pr_info("Reserved %lu MiB in %d range%s\n",
+			(unsigned long)total_size / SZ_1M, nr,
+			nr > 1 ? "s" : "");
+
+	return ret;
 }
 
 /**
@@ -241,6 +536,26 @@ int __init cma_declare_contiguous_nid(phys_addr_t base,
 			phys_addr_t alignment, unsigned int order_per_bit,
 			bool fixed, const char *name, struct cma **res_cma,
 			int nid)
+{
+	int ret;
+
+	ret = __cma_declare_contiguous_nid(base, size, limit, alignment,
+			order_per_bit, fixed, name, res_cma, nid);
+	if (ret != 0)
+		pr_err("Failed to reserve %ld MiB\n",
+				(unsigned long)size / SZ_1M);
+	else
+		pr_info("Reserved %ld MiB at %pa\n",
+				(unsigned long)size / SZ_1M, &base);
+
+	return ret;
+}
+
+static int __init __cma_declare_contiguous_nid(phys_addr_t base,
+			phys_addr_t size, phys_addr_t limit,
+			phys_addr_t alignment, unsigned int order_per_bit,
+			bool fixed, const char *name, struct cma **res_cma,
+			int nid)
 {
 	phys_addr_t memblock_end = memblock_end_of_DRAM();
 	phys_addr_t highmem_start;
@@ -273,10 +588,9 @@ int __init cma_declare_contiguous_nid(phys_addr_t base,
 	/* Sanitise input arguments. */
 	alignment = max_t(phys_addr_t, alignment, CMA_MIN_ALIGNMENT_BYTES);
 	if (fixed && base & (alignment - 1)) {
-		ret = -EINVAL;
 		pr_err("Region at %pa must be aligned to %pa bytes\n",
 			&base, &alignment);
-		goto err;
+		return -EINVAL;
 	}
 	base = ALIGN(base, alignment);
 	size = ALIGN(size, alignment);
@@ -294,10 +608,9 @@ int __init cma_declare_contiguous_nid(phys_addr_t base,
 	 * low/high memory boundary.
 	 */
 	if (fixed && base < highmem_start && base + size > highmem_start) {
-		ret = -EINVAL;
 		pr_err("Region at %pa defined on low/high memory boundary (%pa)\n",
 			&base, &highmem_start);
-		goto err;
+		return -EINVAL;
 	}
 
 	/*
@@ -309,18 +622,16 @@ int __init cma_declare_contiguous_nid(phys_addr_t base,
 		limit = memblock_end;
 
 	if (base + size > limit) {
-		ret = -EINVAL;
 		pr_err("Size (%pa) of region at %pa exceeds limit (%pa)\n",
 			&size, &base, &limit);
-		goto err;
+		return -EINVAL;
 	}
 
 	/* Reserve memory */
 	if (fixed) {
 		if (memblock_is_region_reserved(base, size) ||
 		    memblock_reserve(base, size) < 0) {
-			ret = -EBUSY;
-			goto err;
+			return -EBUSY;
 		}
 	} else {
 		phys_addr_t addr = 0;
@@ -357,10 +668,8 @@ int __init cma_declare_contiguous_nid(phys_addr_t base,
 		if (!addr) {
 			addr = memblock_alloc_range_nid(size, alignment, base,
 					limit, nid, true);
-			if (!addr) {
-				ret = -ENOMEM;
-				goto err;
-			}
+			if (!addr)
+				return -ENOMEM;
 		}
 
 		/*
@@ -373,75 +682,67 @@ int __init cma_declare_contiguous_nid(phys_addr_t base,
 
 	ret = cma_init_reserved_mem(base, size, order_per_bit, name, res_cma);
 	if (ret)
-		goto free_mem;
-
-	pr_info("Reserved %ld MiB at %pa on node %d\n", (unsigned long)size / SZ_1M,
-		&base, nid);
-	return 0;
+		memblock_phys_free(base, size);
 
-free_mem:
-	memblock_phys_free(base, size);
-err:
-	pr_err("Failed to reserve %ld MiB on node %d\n", (unsigned long)size / SZ_1M,
-	       nid);
 	return ret;
 }
 
 static void cma_debug_show_areas(struct cma *cma)
 {
 	unsigned long next_zero_bit, next_set_bit, nr_zero;
-	unsigned long start = 0;
+	unsigned long start;
 	unsigned long nr_part;
-	unsigned long nbits = cma_bitmap_maxno(cma);
+	unsigned long nbits;
+	int r;
+	struct cma_memrange *cmr;
 
 	spin_lock_irq(&cma->lock);
 	pr_info("number of available pages: ");
-	for (;;) {
-		next_zero_bit = find_next_zero_bit(cma->bitmap, nbits, start);
-		if (next_zero_bit >= nbits)
-			break;
-		next_set_bit = find_next_bit(cma->bitmap, nbits, next_zero_bit);
-		nr_zero = next_set_bit - next_zero_bit;
-		nr_part = nr_zero << cma->order_per_bit;
-		pr_cont("%s%lu@%lu", start ? "+" : "", nr_part,
-			next_zero_bit);
-		start = next_zero_bit + nr_zero;
+	for (r = 0; r < cma->nranges; r++) {
+		cmr = &cma->ranges[r];
+
+		start = 0;
+		nbits = cma_bitmap_maxno(cma, cmr);
+
+		pr_info("range %d: ", r);
+		for (;;) {
+			next_zero_bit = find_next_zero_bit(cmr->bitmap,
+							   nbits, start);
+			if (next_zero_bit >= nbits)
+				break;
+			next_set_bit = find_next_bit(cmr->bitmap, nbits,
+						     next_zero_bit);
+			nr_zero = next_set_bit - next_zero_bit;
+			nr_part = nr_zero << cma->order_per_bit;
+			pr_cont("%s%lu@%lu", start ? "+" : "", nr_part,
+				next_zero_bit);
+			start = next_zero_bit + nr_zero;
+		}
+		pr_info("\n");
 	}
 	pr_cont("=> %lu free of %lu total pages\n", cma->available_count,
 			cma->count);
 	spin_unlock_irq(&cma->lock);
 }
 
-static struct page *__cma_alloc(struct cma *cma, unsigned long count,
-				unsigned int align, gfp_t gfp)
+static int cma_range_alloc(struct cma *cma, struct cma_memrange *cmr,
+				unsigned long count, unsigned int align,
+				struct page **pagep, gfp_t gfp)
 {
 	unsigned long mask, offset;
 	unsigned long pfn = -1;
 	unsigned long start = 0;
 	unsigned long bitmap_maxno, bitmap_no, bitmap_count;
-	unsigned long i;
+	int ret = -EBUSY;
 	struct page *page = NULL;
-	int ret = -ENOMEM;
-	const char *name = cma ? cma->name : NULL;
-
-	trace_cma_alloc_start(name, count, align);
-
-	if (!cma || !cma->count || !cma->bitmap)
-		return page;
-
-	pr_debug("%s(cma %p, name: %s, count %lu, align %d)\n", __func__,
-		(void *)cma, cma->name, count, align);
-
-	if (!count)
-		return page;
 
 	mask = cma_bitmap_aligned_mask(cma, align);
-	offset = cma_bitmap_aligned_offset(cma, align);
-	bitmap_maxno = cma_bitmap_maxno(cma);
+	offset = cma_bitmap_aligned_offset(cma, cmr, align);
+	bitmap_maxno = cma_bitmap_maxno(cma, cmr);
 	bitmap_count = cma_bitmap_pages_to_bits(cma, count);
 
 	if (bitmap_count > bitmap_maxno)
-		return page;
+		goto out;
 
 	for (;;) {
 		spin_lock_irq(&cma->lock);
@@ -453,14 +754,14 @@ static struct page *__cma_alloc(struct cma *cma, unsigned long count,
 			spin_unlock_irq(&cma->lock);
 			break;
 		}
-		bitmap_no = bitmap_find_next_zero_area_off(cma->bitmap,
+		bitmap_no = bitmap_find_next_zero_area_off(cmr->bitmap,
 				bitmap_maxno, start, bitmap_count, mask,
 				offset);
 		if (bitmap_no >= bitmap_maxno) {
 			spin_unlock_irq(&cma->lock);
 			break;
 		}
-		bitmap_set(cma->bitmap, bitmap_no, bitmap_count);
+		bitmap_set(cmr->bitmap, bitmap_no, bitmap_count);
 		cma->available_count -= count;
 		/*
 		 * It's safe to drop the lock here. We've marked this region for
@@ -469,7 +770,7 @@ static struct page *__cma_alloc(struct cma *cma, unsigned long count,
 		 */
 		spin_unlock_irq(&cma->lock);
 
-		pfn = cma->base_pfn + (bitmap_no << cma->order_per_bit);
+		pfn = cmr->base_pfn + (bitmap_no << cma->order_per_bit);
 		mutex_lock(&cma_mutex);
 		ret = alloc_contig_range(pfn, pfn + count, MIGRATE_CMA, gfp);
 		mutex_unlock(&cma_mutex);
@@ -478,7 +779,7 @@ static struct page *__cma_alloc(struct cma *cma, unsigned long count,
 			break;
 		}
 
-		cma_clear_bitmap(cma, pfn, count);
+		cma_clear_bitmap(cma, cmr, pfn, count);
 		if (ret != -EBUSY)
 			break;
 
@@ -490,6 +791,48 @@ static struct page *__cma_alloc(struct cma *cma, unsigned long count,
 		/* try again with a bit different memory target */
 		start = bitmap_no + mask + 1;
 	}
+out:
+	*pagep = page;
+	return ret;
+}
+
+/**
+ * cma_alloc() - allocate pages from contiguous area
+ * @cma:   Contiguous memory region for which the allocation is performed.
+ * @count: Requested number of pages.
+ * @align: Requested alignment of pages (in PAGE_SIZE order).
+ * @no_warn: Avoid printing message about failed allocation
+ *
+ * This function allocates part of contiguous memory on specific
+ * contiguous memory area.
+ */
+static struct page *__cma_alloc(struct cma *cma, unsigned long count,
+		       unsigned int align, gfp_t gfp)
+{
+	struct page *page = NULL;
+	int ret = -ENOMEM, r;
+	unsigned long i;
+	const char *name = cma ? cma->name : NULL;
+
+	trace_cma_alloc_start(name, count, align);
+
+	if (!cma || !cma->count)
+		return page;
+
+	pr_debug("%s(cma %p, name: %s, count %lu, align %d)\n", __func__,
+		(void *)cma, cma->name, count, align);
+
+	if (!count)
+		return page;
+
+	for (r = 0; r < cma->nranges; r++) {
+		page = NULL;
+
+		ret = cma_range_alloc(cma, &cma->ranges[r], count, align,
+				       &page, gfp);
+		if (ret != -EBUSY || page)
+			break;
+	}
 
 	/*
 	 * CMA can allocate multiple page blocks, which results in different
@@ -508,7 +851,8 @@ static struct page *__cma_alloc(struct cma *cma, unsigned long count,
 	}
 
 	pr_debug("%s(): returned %p\n", __func__, page);
-	trace_cma_alloc_finish(name, pfn, page, count, align, ret);
+	trace_cma_alloc_finish(name, page ? page_to_pfn(page) : 0,
+			       page, count, align, ret);
 	if (page) {
 		count_vm_event(CMA_ALLOC_SUCCESS);
 		cma_sysfs_account_success_pages(cma, count);
@@ -551,20 +895,31 @@ struct folio *cma_alloc_folio(struct cma *cma, int order, gfp_t gfp)
 bool cma_pages_valid(struct cma *cma, const struct page *pages,
 		     unsigned long count)
 {
-	unsigned long pfn;
+	unsigned long pfn, end;
+	int r;
+	struct cma_memrange *cmr;
+	bool ret;
 
-	if (!cma || !pages)
+	if (!cma || !pages || count > cma->count)
 		return false;
 
 	pfn = page_to_pfn(pages);
+	ret = false;
 
-	if (pfn < cma->base_pfn || pfn >= cma->base_pfn + cma->count) {
-		pr_debug("%s(page %p, count %lu)\n", __func__,
-						(void *)pages, count);
-		return false;
+	for (r = 0; r < cma->nranges; r++) {
+		cmr = &cma->ranges[r];
+		end = cmr->base_pfn + cmr->count;
+		if (pfn >= cmr->base_pfn && pfn < end) {
+			ret = pfn + count <= end;
+			break;
+		}
 	}
 
-	return true;
+	if (!ret)
+		pr_debug("%s(page %p, count %lu)\n",
+				__func__, (void *)pages, count);
+
+	return ret;
 }
 
 /**
@@ -580,19 +935,32 @@ bool cma_pages_valid(struct cma *cma, const struct page *pages,
 bool cma_release(struct cma *cma, const struct page *pages,
 		 unsigned long count)
 {
-	unsigned long pfn;
+	struct cma_memrange *cmr;
+	unsigned long pfn, end_pfn;
+	int r;
+
+	pr_debug("%s(page %p, count %lu)\n", __func__, (void *)pages, count);
 
 	if (!cma_pages_valid(cma, pages, count))
 		return false;
 
-	pr_debug("%s(page %p, count %lu)\n", __func__, (void *)pages, count);
-
 	pfn = page_to_pfn(pages);
+	end_pfn = pfn + count;
+
+	for (r = 0; r < cma->nranges; r++) {
+		cmr = &cma->ranges[r];
+		if (pfn >= cmr->base_pfn &&
+		    pfn < (cmr->base_pfn + cmr->count)) {
+			VM_BUG_ON(end_pfn > cmr->base_pfn + cmr->count);
+			break;
+		}
+	}
 
-	VM_BUG_ON(pfn + count > cma->base_pfn + cma->count);
+	if (r == cma->nranges)
+		return false;
 
 	free_contig_range(pfn, count);
-	cma_clear_bitmap(cma, pfn, count);
+	cma_clear_bitmap(cma, cmr, pfn, count);
 	cma_sysfs_account_release_pages(cma, count);
 	trace_cma_release(cma->name, pfn, pages, count);
 
diff --git a/mm/cma.h b/mm/cma.h
index 3dd3376ae980..5f39dd1aac91 100644
--- a/mm/cma.h
+++ b/mm/cma.h
@@ -10,19 +10,35 @@ struct cma_kobject {
 	struct cma *cma;
 };
 
+/*
+ * Multi-range support. This can be useful if the size of the allocation
+ * is not expected to be larger than the alignment (like with hugetlb_cma),
+ * and the total amount of memory requested, while smaller than the total
+ * amount of memory available, is large enough that it doesn't fit in a
+ * single physical memory range because of memory holes.
+ */
+struct cma_memrange {
+	unsigned long base_pfn;
+	unsigned long count;
+	unsigned long *bitmap;
+#ifdef CONFIG_CMA_DEBUGFS
+	struct debugfs_u32_array dfs_bitmap;
+#endif
+};
+#define CMA_MAX_RANGES 8
+
 struct cma {
-	unsigned long   base_pfn;
 	unsigned long   count;
 	unsigned long	available_count;
-	unsigned long   *bitmap;
 	unsigned int order_per_bit; /* Order of pages represented by one bit */
 	spinlock_t	lock;
 #ifdef CONFIG_CMA_DEBUGFS
 	struct hlist_head mem_head;
 	spinlock_t mem_head_lock;
-	struct debugfs_u32_array dfs_bitmap;
 #endif
 	char name[CMA_MAX_NAME];
+	int nranges;
+	struct cma_memrange ranges[CMA_MAX_RANGES];
 #ifdef CONFIG_CMA_SYSFS
 	/* the number of CMA page successful allocations */
 	atomic64_t nr_pages_succeeded;
@@ -39,9 +55,10 @@ struct cma {
 extern struct cma cma_areas[MAX_CMA_AREAS];
 extern unsigned int cma_area_count;
 
-static inline unsigned long cma_bitmap_maxno(struct cma *cma)
+static inline unsigned long cma_bitmap_maxno(struct cma *cma,
+		struct cma_memrange *cmr)
 {
-	return cma->count >> cma->order_per_bit;
+	return cmr->count >> cma->order_per_bit;
 }
 
 #ifdef CONFIG_CMA_SYSFS
diff --git a/mm/cma_debug.c b/mm/cma_debug.c
index 89236f22230a..400f589756ba 100644
--- a/mm/cma_debug.c
+++ b/mm/cma_debug.c
@@ -46,17 +46,26 @@ DEFINE_DEBUGFS_ATTRIBUTE(cma_used_fops, cma_used_get, NULL, "%llu\n");
 static int cma_maxchunk_get(void *data, u64 *val)
 {
 	struct cma *cma = data;
+	struct cma_memrange *cmr;
 	unsigned long maxchunk = 0;
-	unsigned long start, end = 0;
-	unsigned long bitmap_maxno = cma_bitmap_maxno(cma);
+	unsigned long start, end;
+	unsigned long bitmap_maxno;
+	int r;
 
 	spin_lock_irq(&cma->lock);
-	for (;;) {
-		start = find_next_zero_bit(cma->bitmap, bitmap_maxno, end);
-		if (start >= bitmap_maxno)
-			break;
-		end = find_next_bit(cma->bitmap, bitmap_maxno, start);
-		maxchunk = max(end - start, maxchunk);
+	for (r = 0; r < cma->nranges; r++) {
+		cmr = &cma->ranges[r];
+		bitmap_maxno = cma_bitmap_maxno(cma, cmr);
+		end = 0;
+		for (;;) {
+			start = find_next_zero_bit(cmr->bitmap,
+						   bitmap_maxno, end);
+			if (start >= bitmap_maxno)
+				break;
+			end = find_next_bit(cmr->bitmap, bitmap_maxno,
+					    start);
+			maxchunk = max(end - start, maxchunk);
+		}
 	}
 	spin_unlock_irq(&cma->lock);
 	*val = (u64)maxchunk << cma->order_per_bit;
@@ -159,24 +168,41 @@ DEFINE_DEBUGFS_ATTRIBUTE(cma_alloc_fops, NULL, cma_alloc_write, "%llu\n");
 
 static void cma_debugfs_add_one(struct cma *cma, struct dentry *root_dentry)
 {
-	struct dentry *tmp;
+	struct dentry *tmp, *dir, *rangedir;
+	int r;
+	char rdirname[3];
+	struct cma_memrange *cmr;
 
 	tmp = debugfs_create_dir(cma->name, root_dentry);
 
 	debugfs_create_file("alloc", 0200, tmp, cma, &cma_alloc_fops);
 	debugfs_create_file("free", 0200, tmp, cma, &cma_free_fops);
-	debugfs_create_file("base_pfn", 0444, tmp,
-			    &cma->base_pfn, &cma_debugfs_fops);
 	debugfs_create_file("count", 0444, tmp, &cma->count, &cma_debugfs_fops);
 	debugfs_create_file("order_per_bit", 0444, tmp,
 			    &cma->order_per_bit, &cma_debugfs_fops);
 	debugfs_create_file("used", 0444, tmp, cma, &cma_used_fops);
 	debugfs_create_file("maxchunk", 0444, tmp, cma, &cma_maxchunk_fops);
 
-	cma->dfs_bitmap.array = (u32 *)cma->bitmap;
-	cma->dfs_bitmap.n_elements = DIV_ROUND_UP(cma_bitmap_maxno(cma),
-						  BITS_PER_BYTE * sizeof(u32));
-	debugfs_create_u32_array("bitmap", 0444, tmp, &cma->dfs_bitmap);
+	rangedir = debugfs_create_dir("ranges", tmp);
+	for (r = 0; r < cma->nranges; r++) {
+		cmr = &cma->ranges[r];
+		snprintf(rdirname, sizeof(rdirname), "%d", r);
+		dir = debugfs_create_dir(rdirname, rangedir);
+		debugfs_create_file("base_pfn", 0444, dir,
+			    &cmr->base_pfn, &cma_debugfs_fops);
+		cmr->dfs_bitmap.array = (u32 *)cmr->bitmap;
+		cmr->dfs_bitmap.n_elements =
+			DIV_ROUND_UP(cma_bitmap_maxno(cma, cmr),
+					BITS_PER_BYTE * sizeof(u32));
+		debugfs_create_u32_array("bitmap", 0444, dir,
+				&cmr->dfs_bitmap);
+	}
+
+	/*
+	 * Backward compatible symlinks to range 0 for base_pfn and bitmap.
+	 */
+	debugfs_create_symlink("base_pfn", tmp, "ranges/0/base_pfn");
+	debugfs_create_symlink("bitmap", tmp, "ranges/0/bitmap");
 }
 
 static int __init cma_debugfs_init(void)
-- 
2.48.1.262.g85cc9f2d1e-goog

[PATCH v2 03/28] mm/cma: introduce cma_intersects function

[Frank van der Linden]

Now that CMA areas can have multiple physical ranges,
code can't assume a CMA struct represents a base_pfn
plus a size, as returned from cma_get_base.

Most cases are ok though, since they all explicitly
refer to CMA areas that were created using existing
interfaces (cma_declare_contiguous_nid or
cma_init_reserved_mem), which guarantees they have just
one physical range.

An exception is the s390 code, which walks all CMA
ranges to see if they intersect with a range of memory
that is about to be hotremoved. So, in the future,
it might run in to multi-range areas. To keep this check
working, define a cma_intersects function. This just checks
if a physaddr range intersects any of the ranges.
Use it in the s390 check.

Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: linux-s390@vger.kernel.org
Signed-off-by: Frank van der Linden <fvdl@google.com>
---
 arch/s390/mm/init.c | 13 +++++--------
 include/linux/cma.h |  1 +
 mm/cma.c            | 21 +++++++++++++++++++++
 3 files changed, 27 insertions(+), 8 deletions(-)

diff --git a/arch/s390/mm/init.c b/arch/s390/mm/init.c
index f2298f7a3f21..d88cb1c13f7d 100644
--- a/arch/s390/mm/init.c
+++ b/arch/s390/mm/init.c
@@ -239,16 +239,13 @@ struct s390_cma_mem_data {
 static int s390_cma_check_range(struct cma *cma, void *data)
 {
 	struct s390_cma_mem_data *mem_data;
-	unsigned long start, end;
 
 	mem_data = data;
-	start = cma_get_base(cma);
-	end = start + cma_get_size(cma);
-	if (end < mem_data->start)
-		return 0;
-	if (start >= mem_data->end)
-		return 0;
-	return -EBUSY;
+
+	if (cma_intersects(cma, mem_data->start, mem_data->end))
+		return -EBUSY;
+
+	return 0;
 }
 
 static int s390_cma_mem_notifier(struct notifier_block *nb,
diff --git a/include/linux/cma.h b/include/linux/cma.h
index 863427c27dc2..03d85c100dcc 100644
--- a/include/linux/cma.h
+++ b/include/linux/cma.h
@@ -53,6 +53,7 @@ extern bool cma_pages_valid(struct cma *cma, const struct page *pages, unsigned
 extern bool cma_release(struct cma *cma, const struct page *pages, unsigned long count);
 
 extern int cma_for_each_area(int (*it)(struct cma *cma, void *data), void *data);
+extern bool cma_intersects(struct cma *cma, unsigned long start, unsigned long end);
 
 extern void cma_reserve_pages_on_error(struct cma *cma);
 
diff --git a/mm/cma.c b/mm/cma.c
index c20255161642..1704d5be6a07 100644
--- a/mm/cma.c
+++ b/mm/cma.c
@@ -988,3 +988,24 @@ int cma_for_each_area(int (*it)(struct cma *cma, void *data), void *data)
 
 	return 0;
 }
+
+bool cma_intersects(struct cma *cma, unsigned long start, unsigned long end)
+{
+	int r;
+	struct cma_memrange *cmr;
+	unsigned long rstart, rend;
+
+	for (r = 0; r < cma->nranges; r++) {
+		cmr = &cma->ranges[r];
+
+		rstart = PFN_PHYS(cmr->base_pfn);
+		rend = PFN_PHYS(cmr->base_pfn + cmr->count);
+		if (end < rstart)
+			continue;
+		if (start >= rend)
+			continue;
+		return true;
+	}
+
+	return false;
+}
-- 
2.48.1.262.g85cc9f2d1e-goog

Re: [PATCH v2 03/28] mm/cma: introduce cma_intersects function

[Alexander Gordeev]

On Wed, Jan 29, 2025 at 10:41:32PM +0000, Frank van der Linden wrote:
> Now that CMA areas can have multiple physical ranges,
> code can't assume a CMA struct represents a base_pfn
> plus a size, as returned from cma_get_base.
> 
> Most cases are ok though, since they all explicitly
> refer to CMA areas that were created using existing
> interfaces (cma_declare_contiguous_nid or
> cma_init_reserved_mem), which guarantees they have just
> one physical range.
> 
> An exception is the s390 code, which walks all CMA
> ranges to see if they intersect with a range of memory
> that is about to be hotremoved. So, in the future,
> it might run in to multi-range areas. To keep this check
> working, define a cma_intersects function. This just checks
> if a physaddr range intersects any of the ranges.
> Use it in the s390 check.
> 
> Cc: Heiko Carstens <hca@linux.ibm.com>
> Cc: Vasily Gorbik <gor@linux.ibm.com>
> Cc: Alexander Gordeev <agordeev@linux.ibm.com>
> Cc: linux-s390@vger.kernel.org
> Signed-off-by: Frank van der Linden <fvdl@google.com>
> ---
>  arch/s390/mm/init.c | 13 +++++--------
>  include/linux/cma.h |  1 +
>  mm/cma.c            | 21 +++++++++++++++++++++
>  3 files changed, 27 insertions(+), 8 deletions(-)

Acked-by: Alexander Gordeev <agordeev@linux.ibm.com>

[PATCH v2 04/28] mm, hugetlb: use cma_declare_contiguous_multi

[Frank van der Linden]

hugetlb_cma is fine with using multiple CMA ranges, as long as it
can get its gigantic pages allocated from them. So, use
cma_declare_contiguous_multi to allow for multiple ranges,
increasing the chances of getting what we want on systems
with gaps in physical memory.

Signed-off-by: Frank van der Linden <fvdl@google.com>
---
 mm/hugetlb.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index 3b25b69aa94f..bc8af09a3105 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -7738,9 +7738,8 @@ void __init hugetlb_cma_reserve(int order)
 		 * may be returned to CMA allocator in the case of
 		 * huge page demotion.
 		 */
-		res = cma_declare_contiguous_nid(0, size, 0,
-					PAGE_SIZE << order,
-					HUGETLB_PAGE_ORDER, false, name,
+		res = cma_declare_contiguous_multi(size, PAGE_SIZE << order,
+					HUGETLB_PAGE_ORDER, name,
 					&hugetlb_cma[nid], nid);
 		if (res) {
 			pr_warn("hugetlb_cma: reservation failed: err %d, node %d",
-- 
2.48.1.262.g85cc9f2d1e-goog

[PATCH v2 05/28] mm/hugetlb: fix round-robin bootmem allocation

[Frank van der Linden]

Commit b5389086ad7b ("hugetlbfs: extend the definition of hugepages parameter to support node allocation")
changed the NUMA_NO_NODE round-robin allocation behavior in case of a
failure to allocate from one NUMA node. The code originally moved on to
the next node to try again, but now it immediately breaks out of the loop.

Restore the original behavior.

Fixes: b5389086ad7b ("hugetlbfs: extend the definition of hugepages parameter to support node allocation")
Cc: Zhenguo Yao <yaozhenguo1@gmail.com>
Signed-off-by: Frank van der Linden <fvdl@google.com>
---
 mm/hugetlb.c | 22 +++++++++++++---------
 1 file changed, 13 insertions(+), 9 deletions(-)

diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index bc8af09a3105..18d308d5df6d 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -3156,16 +3156,13 @@ int __alloc_bootmem_huge_page(struct hstate *h, int nid)
 		m = memblock_alloc_try_nid_raw(
 				huge_page_size(h), huge_page_size(h),
 				0, MEMBLOCK_ALLOC_ACCESSIBLE, node);
-		/*
-		 * Use the beginning of the huge page to store the
-		 * huge_bootmem_page struct (until gather_bootmem
-		 * puts them into the mem_map).
-		 */
-		if (!m)
-			return 0;
-		goto found;
+		if (m)
+			break;
 	}
 
+	if (!m)
+		return 0;
+
 found:
 
 	/*
@@ -3177,7 +3174,14 @@ int __alloc_bootmem_huge_page(struct hstate *h, int nid)
 	 */
 	memblock_reserved_mark_noinit(virt_to_phys((void *)m + PAGE_SIZE),
 		huge_page_size(h) - PAGE_SIZE);
-	/* Put them into a private list first because mem_map is not up yet */
+	/*
+	 * Use the beginning of the huge page to store the
+	 * huge_bootmem_page struct (until gather_bootmem
+	 * puts them into the mem_map).
+	 *
+	 * Put them into a private list first because mem_map
+	 * is not up yet.
+	 */
 	INIT_LIST_HEAD(&m->list);
 	list_add(&m->list, &huge_boot_pages[node]);
 	m->hstate = h;
-- 
2.48.1.262.g85cc9f2d1e-goog

[PATCH v2 06/28] mm/hugetlb: remove redundant __ClearPageReserved

[Frank van der Linden]

In hugetlb_folio_init_tail_vmemmap, the reserved flag is cleared for
the tail page just before it is zeroed out, which is redundant.
Remove the __ClearPageReserved call.

Signed-off-by: Frank van der Linden <fvdl@google.com>
---
 mm/hugetlb.c | 1 -
 1 file changed, 1 deletion(-)

diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index 18d308d5df6d..196359254cfb 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -3202,7 +3202,6 @@ static void __init hugetlb_folio_init_tail_vmemmap(struct folio *folio,
 	for (pfn = head_pfn + start_page_number; pfn < end_pfn; pfn++) {
 		struct page *page = pfn_to_page(pfn);
 
-		__ClearPageReserved(folio_page(folio, pfn - head_pfn));
 		__init_single_page(page, pfn, zone, nid);
 		prep_compound_tail((struct page *)folio, pfn - head_pfn);
 		ret = page_ref_freeze(page, 1);
-- 
2.48.1.262.g85cc9f2d1e-goog

[PATCH v2 07/28] mm/hugetlb: use online nodes for bootmem allocation

[Frank van der Linden]

Later commits will move hugetlb bootmem allocation to earlier
in init, when N_MEMORY has not yet been set on nodes. Use
online nodes instead. At most, this wastes just a few cycles
once during boot (and most likely none).

Signed-off-by: Frank van der Linden <fvdl@google.com>
---
 mm/hugetlb.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index 196359254cfb..20d54eaf2bad 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -3152,7 +3152,7 @@ int __alloc_bootmem_huge_page(struct hstate *h, int nid)
 		goto found;
 	}
 	/* allocate from next node when distributing huge pages */
-	for_each_node_mask_to_alloc(&h->next_nid_to_alloc, nr_nodes, node, &node_states[N_MEMORY]) {
+	for_each_node_mask_to_alloc(&h->next_nid_to_alloc, nr_nodes, node, &node_states[N_ONLINE]) {
 		m = memblock_alloc_try_nid_raw(
 				huge_page_size(h), huge_page_size(h),
 				0, MEMBLOCK_ALLOC_ACCESSIBLE, node);
@@ -4550,8 +4550,8 @@ void __init hugetlb_add_hstate(unsigned int order)
 	for (i = 0; i < MAX_NUMNODES; ++i)
 		INIT_LIST_HEAD(&h->hugepage_freelists[i]);
 	INIT_LIST_HEAD(&h->hugepage_activelist);
-	h->next_nid_to_alloc = first_memory_node;
-	h->next_nid_to_free = first_memory_node;
+	h->next_nid_to_alloc = first_online_node;
+	h->next_nid_to_free = first_online_node;
 	snprintf(h->name, HSTATE_NAME_LEN, "hugepages-%lukB",
 					huge_page_size(h)/SZ_1K);
 
-- 
2.48.1.262.g85cc9f2d1e-goog

[PATCH v2 08/28] mm/hugetlb: convert cmdline parameters from setup to early

[Frank van der Linden]

Convert the cmdline parameters (hugepagesz, hugepages,
default_hugepagesz and hugetlb_free_vmemmap) to early
parameters.

Since parse_early_param might run before MMU setups
on some platforms (powerpc), validation of huge page
sizes as specified in command line parameters would
fail. So instead, for the hstate-related values,
just record the them and parse them on demand, from
hugetlb_bootmem_alloc.

The allocation of hugetlb bootmem pages is now done in
hugetlb_bootmem_alloc, which is called explicitly at the
start of mm_core_init().  core_initcall would be too late,
as that happens with memblock already torn down.

This change will allow earlier allocation and initialization
of bootmem hugetlb pages later on.

No functional change intended.

Signed-off-by: Frank van der Linden <fvdl@google.com>
---
 include/linux/hugetlb.h |   6 ++
 mm/hugetlb.c            | 133 +++++++++++++++++++++++++++++++---------
 mm/hugetlb_vmemmap.c    |   6 +-
 mm/mm_init.c            |   3 +
 4 files changed, 119 insertions(+), 29 deletions(-)

diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h
index ec8c0ccc8f95..9cd7c9dacb88 100644
--- a/include/linux/hugetlb.h
+++ b/include/linux/hugetlb.h
@@ -174,6 +174,8 @@ struct address_space *hugetlb_folio_mapping_lock_write(struct folio *folio);
 extern int sysctl_hugetlb_shm_group;
 extern struct list_head huge_boot_pages[MAX_NUMNODES];
 
+void hugetlb_bootmem_alloc(void);
+
 /* arch callbacks */
 
 #ifndef CONFIG_HIGHPTE
@@ -1250,6 +1252,10 @@ static inline bool hugetlbfs_pagecache_present(
 {
 	return false;
 }
+
+static inline void hugetlb_bootmem_alloc(void)
+{
+}
 #endif	/* CONFIG_HUGETLB_PAGE */
 
 static inline spinlock_t *huge_pte_lock(struct hstate *h,
diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index 20d54eaf2bad..c16ed9790022 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -40,6 +40,7 @@
 #include <asm/page.h>
 #include <asm/pgalloc.h>
 #include <asm/tlb.h>
+#include <asm/setup.h>
 
 #include <linux/io.h>
 #include <linux/hugetlb.h>
@@ -62,6 +63,24 @@ static unsigned long hugetlb_cma_size __initdata;
 
 __initdata struct list_head huge_boot_pages[MAX_NUMNODES];
 
+/*
+ * Due to ordering constraints across the init code for various
+ * architectures, hugetlb hstate cmdline parameters can't simply
+ * be early_param. early_param might call the setup function
+ * before valid hugetlb page sizes are determined, leading to
+ * incorrect rejection of valid hugepagesz= options.
+ *
+ * So, record the parameters early and consume them whenever the
+ * init code is ready for them, by calling hugetlb_parse_params().
+ */
+
+/* one (hugepagesz=,hugepages=) pair per hstate, one default_hugepagesz */
+#define HUGE_MAX_CMDLINE_ARGS	(2 * HUGE_MAX_HSTATE + 1)
+struct hugetlb_cmdline {
+	char *val;
+	int (*setup)(char *val);
+};
+
 /* for command line parsing */
 static struct hstate * __initdata parsed_hstate;
 static unsigned long __initdata default_hstate_max_huge_pages;
@@ -69,6 +88,20 @@ static bool __initdata parsed_valid_hugepagesz = true;
 static bool __initdata parsed_default_hugepagesz;
 static unsigned int default_hugepages_in_node[MAX_NUMNODES] __initdata;
 
+static char hstate_cmdline_buf[COMMAND_LINE_SIZE] __initdata;
+static int hstate_cmdline_index __initdata;
+static struct hugetlb_cmdline hugetlb_params[HUGE_MAX_CMDLINE_ARGS] __initdata;
+static int hugetlb_param_index __initdata;
+static __init int hugetlb_add_param(char *s, int (*setup)(char *val));
+static __init void hugetlb_parse_params(void);
+
+#define hugetlb_early_param(str, func) \
+static __init int func##args(char *s) \
+{ \
+	return hugetlb_add_param(s, func); \
+} \
+early_param(str, func##args)
+
 /*
  * Protects updates to hugepage_freelists, hugepage_activelist, nr_huge_pages,
  * free_huge_pages, and surplus_huge_pages.
@@ -3488,6 +3521,8 @@ static void __init hugetlb_hstate_alloc_pages(struct hstate *h)
 
 		for (i = 0; i < MAX_NUMNODES; i++)
 			INIT_LIST_HEAD(&huge_boot_pages[i]);
+		h->next_nid_to_alloc = first_online_node;
+		h->next_nid_to_free = first_online_node;
 		initialized = true;
 	}
 
@@ -4550,8 +4585,6 @@ void __init hugetlb_add_hstate(unsigned int order)
 	for (i = 0; i < MAX_NUMNODES; ++i)
 		INIT_LIST_HEAD(&h->hugepage_freelists[i]);
 	INIT_LIST_HEAD(&h->hugepage_activelist);
-	h->next_nid_to_alloc = first_online_node;
-	h->next_nid_to_free = first_online_node;
 	snprintf(h->name, HSTATE_NAME_LEN, "hugepages-%lukB",
 					huge_page_size(h)/SZ_1K);
 
@@ -4576,6 +4609,42 @@ static void __init hugepages_clear_pages_in_node(void)
 	}
 }
 
+static __init int hugetlb_add_param(char *s, int (*setup)(char *))
+{
+	size_t len;
+	char *p;
+
+	if (hugetlb_param_index >= HUGE_MAX_CMDLINE_ARGS)
+		return -EINVAL;
+
+	len = strlen(s) + 1;
+	if (len + hstate_cmdline_index > sizeof(hstate_cmdline_buf))
+		return -EINVAL;
+
+	p = &hstate_cmdline_buf[hstate_cmdline_index];
+	memcpy(p, s, len);
+	hstate_cmdline_index += len;
+
+	hugetlb_params[hugetlb_param_index].val = p;
+	hugetlb_params[hugetlb_param_index].setup = setup;
+
+	hugetlb_param_index++;
+
+	return 0;
+}
+
+static __init void hugetlb_parse_params(void)
+{
+	int i;
+	struct hugetlb_cmdline *hcp;
+
+	for (i = 0; i < hugetlb_param_index; i++) {
+		hcp = &hugetlb_params[i];
+
+		hcp->setup(hcp->val);
+	}
+}
+
 /*
  * hugepages command line processing
  * hugepages normally follows a valid hugepagsz or default_hugepagsz
@@ -4595,7 +4664,7 @@ static int __init hugepages_setup(char *s)
 	if (!parsed_valid_hugepagesz) {
 		pr_warn("HugeTLB: hugepages=%s does not follow a valid hugepagesz, ignoring\n", s);
 		parsed_valid_hugepagesz = true;
-		return 1;
+		return -EINVAL;
 	}
 
 	/*
@@ -4649,24 +4718,16 @@ static int __init hugepages_setup(char *s)
 		}
 	}
 
-	/*
-	 * Global state is always initialized later in hugetlb_init.
-	 * But we need to allocate gigantic hstates here early to still
-	 * use the bootmem allocator.
-	 */
-	if (hugetlb_max_hstate && hstate_is_gigantic(parsed_hstate))
-		hugetlb_hstate_alloc_pages(parsed_hstate);
-
 	last_mhp = mhp;
 
-	return 1;
+	return 0;
 
 invalid:
 	pr_warn("HugeTLB: Invalid hugepages parameter %s\n", p);
 	hugepages_clear_pages_in_node();
-	return 1;
+	return -EINVAL;
 }
-__setup("hugepages=", hugepages_setup);
+hugetlb_early_param("hugepages", hugepages_setup);
 
 /*
  * hugepagesz command line processing
@@ -4685,7 +4746,7 @@ static int __init hugepagesz_setup(char *s)
 
 	if (!arch_hugetlb_valid_size(size)) {
 		pr_err("HugeTLB: unsupported hugepagesz=%s\n", s);
-		return 1;
+		return -EINVAL;
 	}
 
 	h = size_to_hstate(size);
@@ -4700,7 +4761,7 @@ static int __init hugepagesz_setup(char *s)
 		if (!parsed_default_hugepagesz ||  h != &default_hstate ||
 		    default_hstate.max_huge_pages) {
 			pr_warn("HugeTLB: hugepagesz=%s specified twice, ignoring\n", s);
-			return 1;
+			return -EINVAL;
 		}
 
 		/*
@@ -4710,14 +4771,14 @@ static int __init hugepagesz_setup(char *s)
 		 */
 		parsed_hstate = h;
 		parsed_valid_hugepagesz = true;
-		return 1;
+		return 0;
 	}
 
 	hugetlb_add_hstate(ilog2(size) - PAGE_SHIFT);
 	parsed_valid_hugepagesz = true;
-	return 1;
+	return 0;
 }
-__setup("hugepagesz=", hugepagesz_setup);
+hugetlb_early_param("hugepagesz", hugepagesz_setup);
 
 /*
  * default_hugepagesz command line input
@@ -4731,14 +4792,14 @@ static int __init default_hugepagesz_setup(char *s)
 	parsed_valid_hugepagesz = false;
 	if (parsed_default_hugepagesz) {
 		pr_err("HugeTLB: default_hugepagesz previously specified, ignoring %s\n", s);
-		return 1;
+		return -EINVAL;
 	}
 
 	size = (unsigned long)memparse(s, NULL);
 
 	if (!arch_hugetlb_valid_size(size)) {
 		pr_err("HugeTLB: unsupported default_hugepagesz=%s\n", s);
-		return 1;
+		return -EINVAL;
 	}
 
 	hugetlb_add_hstate(ilog2(size) - PAGE_SHIFT);
@@ -4755,17 +4816,33 @@ static int __init default_hugepagesz_setup(char *s)
 	 */
 	if (default_hstate_max_huge_pages) {
 		default_hstate.max_huge_pages = default_hstate_max_huge_pages;
-		for_each_online_node(i)
-			default_hstate.max_huge_pages_node[i] =
-				default_hugepages_in_node[i];
-		if (hstate_is_gigantic(&default_hstate))
-			hugetlb_hstate_alloc_pages(&default_hstate);
+		/*
+		 * Since this is an early parameter, we can't check
+		 * NUMA node state yet, so loop through MAX_NUMNODES.
+		 */
+		for (i = 0; i < MAX_NUMNODES; i++) {
+			if (default_hugepages_in_node[i] != 0)
+				default_hstate.max_huge_pages_node[i] =
+					default_hugepages_in_node[i];
+		}
 		default_hstate_max_huge_pages = 0;
 	}
 
-	return 1;
+	return 0;
+}
+hugetlb_early_param("default_hugepagesz", default_hugepagesz_setup);
+
+void __init hugetlb_bootmem_alloc(void)
+{
+	struct hstate *h;
+
+	hugetlb_parse_params();
+
+	for_each_hstate(h) {
+		if (hstate_is_gigantic(h))
+			hugetlb_hstate_alloc_pages(h);
+	}
 }
-__setup("default_hugepagesz=", default_hugepagesz_setup);
 
 static unsigned int allowed_mems_nr(struct hstate *h)
 {
diff --git a/mm/hugetlb_vmemmap.c b/mm/hugetlb_vmemmap.c
index 7735972add01..5b484758f813 100644
--- a/mm/hugetlb_vmemmap.c
+++ b/mm/hugetlb_vmemmap.c
@@ -444,7 +444,11 @@ DEFINE_STATIC_KEY_FALSE(hugetlb_optimize_vmemmap_key);
 EXPORT_SYMBOL(hugetlb_optimize_vmemmap_key);
 
 static bool vmemmap_optimize_enabled = IS_ENABLED(CONFIG_HUGETLB_PAGE_OPTIMIZE_VMEMMAP_DEFAULT_ON);
-core_param(hugetlb_free_vmemmap, vmemmap_optimize_enabled, bool, 0);
+static int __init hugetlb_vmemmap_optimize_param(char *buf)
+{
+	return kstrtobool(buf, &vmemmap_optimize_enabled);
+}
+early_param("hugetlb_free_vmemmap", hugetlb_vmemmap_optimize_param);
 
 static int __hugetlb_vmemmap_restore_folio(const struct hstate *h,
 					   struct folio *folio, unsigned long flags)
diff --git a/mm/mm_init.c b/mm/mm_init.c
index 2630cc30147e..d2dee53e95dd 100644
--- a/mm/mm_init.c
+++ b/mm/mm_init.c
@@ -30,6 +30,7 @@
 #include <linux/crash_dump.h>
 #include <linux/execmem.h>
 #include <linux/vmstat.h>
+#include <linux/hugetlb.h>
 #include "internal.h"
 #include "slab.h"
 #include "shuffle.h"
@@ -2641,6 +2642,8 @@ static void __init mem_init_print_info(void)
  */
 void __init mm_core_init(void)
 {
+	hugetlb_bootmem_alloc();
+
 	/* Initializations relying on SMP setup */
 	BUILD_BUG_ON(MAX_ZONELISTS > 2);
 	build_all_zonelists(NULL);
-- 
2.48.1.262.g85cc9f2d1e-goog

[PATCH v2 09/28] x86/mm: make register_page_bootmem_memmap handle PTE mappings

[Frank van der Linden]

register_page_bootmem_memmap expects that vmemmap pages handed
to it are PMD-mapped, and that the number of pages to call
get_page_bootmem on is PMD-aligned.

This is currently a correct assumption, but will no longer be
true once pre-HVO of hugetlb pages is implemented.

Make it handle PTE-mapped vmemmap pages and a nr_pages argument
that is not necessarily PAGES_PER_SECTION.

Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Frank van der Linden <fvdl@google.com>
---
 arch/x86/mm/init_64.c | 16 ++++++----------
 1 file changed, 6 insertions(+), 10 deletions(-)

diff --git a/arch/x86/mm/init_64.c b/arch/x86/mm/init_64.c
index 01ea7c6df303..e7572af639a4 100644
--- a/arch/x86/mm/init_64.c
+++ b/arch/x86/mm/init_64.c
@@ -1599,11 +1599,12 @@ void register_page_bootmem_memmap(unsigned long section_nr,
 		}
 		get_page_bootmem(section_nr, pud_page(*pud), MIX_SECTION_INFO);
 
-		if (!boot_cpu_has(X86_FEATURE_PSE)) {
+		pmd = pmd_offset(pud, addr);
+		if (pmd_none(*pmd))
+			continue;
+
+		if (!boot_cpu_has(X86_FEATURE_PSE) || !pmd_leaf(*pmd)) {
 			next = (addr + PAGE_SIZE) & PAGE_MASK;
-			pmd = pmd_offset(pud, addr);
-			if (pmd_none(*pmd))
-				continue;
 			get_page_bootmem(section_nr, pmd_page(*pmd),
 					 MIX_SECTION_INFO);
 
@@ -1614,12 +1615,7 @@ void register_page_bootmem_memmap(unsigned long section_nr,
 					 SECTION_INFO);
 		} else {
 			next = pmd_addr_end(addr, end);
-
-			pmd = pmd_offset(pud, addr);
-			if (pmd_none(*pmd))
-				continue;
-
-			nr_pmd_pages = 1 << get_order(PMD_SIZE);
+			nr_pmd_pages = (next - addr) >> PAGE_SHIFT;
 			page = pmd_page(*pmd);
 			while (nr_pmd_pages--)
 				get_page_bootmem(section_nr, page++,
-- 
2.48.1.262.g85cc9f2d1e-goog

[PATCH v2 10/28] mm/bootmem_info: export register_page_bootmem_memmap

[Frank van der Linden]

If other mm code wants to use this function for early
memmap inialization (on the platforms that have it),
it should be made available properly, not just
unconditionally in mm.h

Make this function available for such cases.

Signed-off-by: Frank van der Linden <fvdl@google.com>
---
 arch/powerpc/mm/init_64.c    | 1 +
 include/linux/bootmem_info.h | 7 +++++++
 include/linux/mm.h           | 3 ---
 3 files changed, 8 insertions(+), 3 deletions(-)

diff --git a/arch/powerpc/mm/init_64.c b/arch/powerpc/mm/init_64.c
index d96bbc001e73..c2d99d68d40e 100644
--- a/arch/powerpc/mm/init_64.c
+++ b/arch/powerpc/mm/init_64.c
@@ -41,6 +41,7 @@
 #include <linux/libfdt.h>
 #include <linux/memremap.h>
 #include <linux/memory.h>
+#include <linux/bootmem_info.h>
 
 #include <asm/pgalloc.h>
 #include <asm/page.h>
diff --git a/include/linux/bootmem_info.h b/include/linux/bootmem_info.h
index d8a8d245824a..4c506e76a808 100644
--- a/include/linux/bootmem_info.h
+++ b/include/linux/bootmem_info.h
@@ -18,6 +18,8 @@ enum bootmem_type {
 
 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
 void __init register_page_bootmem_info_node(struct pglist_data *pgdat);
+void register_page_bootmem_memmap(unsigned long section_nr, struct page *map,
+				  unsigned long nr_pages);
 
 void get_page_bootmem(unsigned long info, struct page *page,
 		enum bootmem_type type);
@@ -58,6 +60,11 @@ static inline void register_page_bootmem_info_node(struct pglist_data *pgdat)
 {
 }
 
+static inline void register_page_bootmem_memmap(unsigned long section_nr,
+		struct page *map, unsigned long nr_pages)
+{
+}
+
 static inline void put_page_bootmem(struct page *page)
 {
 }
diff --git a/include/linux/mm.h b/include/linux/mm.h
index 7b1068ddcbb7..6dfc41b461af 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -3918,9 +3918,6 @@ static inline bool vmemmap_can_optimize(struct vmem_altmap *altmap,
 }
 #endif
 
-void register_page_bootmem_memmap(unsigned long section_nr, struct page *map,
-				  unsigned long nr_pages);
-
 enum mf_flags {
 	MF_COUNT_INCREASED = 1 << 0,
 	MF_ACTION_REQUIRED = 1 << 1,
-- 
2.48.1.262.g85cc9f2d1e-goog

[PATCH v2 11/28] mm/sparse: allow for alternate vmemmap section init at boot

[Frank van der Linden]

Add functions that are called just before the per-section memmap
is initialized and just before the memmap page structures are
initialized. They are called sparse_vmemmap_init_nid_early and
sparse_vmemmap_init_nid_late, respectively.

This allows for mm subsystems to add calls to initialize memmap
and page structures in a specific way, if using SPARSEMEM_VMEMMAP.
Specifically, hugetlb can pre-HVO bootmem allocated pages that
way, so that no time and resources are wasted on allocating vmemmap
pages, only to free them later (and possibly unnecessarily running
the system out of memory in the process).

Refactor some code and export a few convenience functions for
external use.

In sparse_init_nid, skip any sections that are already initialized,
e.g. they have been initialized by sparse_vmemmap_init_nid_early
already.

The hugetlb code to use these functions will be added in a later
commit.

Export section_map_size, as any alternate memmap init code
will want to use it.

THe config option to enable this is SPARSEMEM_VMEMMAP_PREINIT,
which is dependent on and architecture-specific option,
ARCH_WANT_SPARSEMEM_VMEMMAP_PREINIT. This is done because
a section flag is used, and the number of flags available
is architecture-dependent (see mmzone.h). Architecures can
decide if there is room for the flag and enable the option.
Fortunately, as of right now, all sparse vmemmap using
architectures do have room.

Signed-off-by: Frank van der Linden <fvdl@google.com>
---
 include/linux/mm.h     |  1 +
 include/linux/mmzone.h | 35 +++++++++++++++++
 mm/Kconfig             |  8 ++++
 mm/bootmem_info.c      |  4 +-
 mm/mm_init.c           |  3 ++
 mm/sparse-vmemmap.c    | 23 +++++++++++
 mm/sparse.c            | 87 ++++++++++++++++++++++++++++++++----------
 7 files changed, 139 insertions(+), 22 deletions(-)

diff --git a/include/linux/mm.h b/include/linux/mm.h
index 6dfc41b461af..df83653ed6e3 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -3828,6 +3828,7 @@ static inline void print_vma_addr(char *prefix, unsigned long rip)
 #endif
 
 void *sparse_buffer_alloc(unsigned long size);
+unsigned long section_map_size(void);
 struct page * __populate_section_memmap(unsigned long pfn,
 		unsigned long nr_pages, int nid, struct vmem_altmap *altmap,
 		struct dev_pagemap *pgmap);
diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
index 9540b41894da..44ecb2f90db4 100644
--- a/include/linux/mmzone.h
+++ b/include/linux/mmzone.h
@@ -1933,6 +1933,9 @@ enum {
 	SECTION_IS_EARLY_BIT,
 #ifdef CONFIG_ZONE_DEVICE
 	SECTION_TAINT_ZONE_DEVICE_BIT,
+#endif
+#ifdef CONFIG_SPARSEMEM_VMEMMAP_PREINIT
+	SECTION_IS_VMEMMAP_PREINIT_BIT,
 #endif
 	SECTION_MAP_LAST_BIT,
 };
@@ -1944,6 +1947,9 @@ enum {
 #ifdef CONFIG_ZONE_DEVICE
 #define SECTION_TAINT_ZONE_DEVICE	BIT(SECTION_TAINT_ZONE_DEVICE_BIT)
 #endif
+#ifdef CONFIG_SPARSEMEM_VMEMMAP_PREINIT
+#define SECTION_IS_VMEMMAP_PREINIT	BIT(SECTION_IS_VMEMMAP_PREINIT_BIT)
+#endif
 #define SECTION_MAP_MASK		(~(BIT(SECTION_MAP_LAST_BIT) - 1))
 #define SECTION_NID_SHIFT		SECTION_MAP_LAST_BIT
 
@@ -1998,6 +2004,30 @@ static inline int online_device_section(struct mem_section *section)
 }
 #endif
 
+#ifdef CONFIG_SPARSEMEM_VMEMMAP_PREINIT
+static inline int preinited_vmemmap_section(struct mem_section *section)
+{
+	return (section &&
+		(section->section_mem_map & SECTION_IS_VMEMMAP_PREINIT));
+}
+
+void sparse_vmemmap_init_nid_early(int nid);
+void sparse_vmemmap_init_nid_late(int nid);
+
+#else
+static inline int preinited_vmemmap_section(struct mem_section *section)
+{
+	return 0;
+}
+static inline void sparse_vmemmap_init_nid_early(int nid)
+{
+}
+
+static inline void sparse_vmemmap_init_nid_late(int nid)
+{
+}
+#endif
+
 static inline int online_section_nr(unsigned long nr)
 {
 	return online_section(__nr_to_section(nr));
@@ -2035,6 +2065,9 @@ static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
 }
 #endif
 
+void sparse_init_early_section(int nid, struct page *map, unsigned long pnum,
+			       unsigned long flags);
+
 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
 /**
  * pfn_valid - check if there is a valid memory map entry for a PFN
@@ -2116,6 +2149,8 @@ void sparse_init(void);
 #else
 #define sparse_init()	do {} while (0)
 #define sparse_index_init(_sec, _nid)  do {} while (0)
+#define sparse_vmemmap_init_nid_early(_nid, _use) do {} while (0)
+#define sparse_vmemmap_init_nid_late(_nid) do {} while (0)
 #define pfn_in_present_section pfn_valid
 #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
 #endif /* CONFIG_SPARSEMEM */
diff --git a/mm/Kconfig b/mm/Kconfig
index 1b501db06417..f984dd928ce7 100644
--- a/mm/Kconfig
+++ b/mm/Kconfig
@@ -489,6 +489,14 @@ config SPARSEMEM_VMEMMAP
 	  SPARSEMEM_VMEMMAP uses a virtually mapped memmap to optimise
 	  pfn_to_page and page_to_pfn operations.  This is the most
 	  efficient option when sufficient kernel resources are available.
+
+config ARCH_WANT_SPARSEMEM_VMEMMAP_PREINIT
+	bool
+
+config SPARSEMEM_VMEMMAP_PREINIT
+	bool "Early init of sparse memory virtual memmap"
+	depends on SPARSEMEM_VMEMMAP && ARCH_WANT_SPARSEMEM_VMEMMAP_PREINIT
+	default y
 #
 # Select this config option from the architecture Kconfig, if it is preferred
 # to enable the feature of HugeTLB/dev_dax vmemmap optimization.
diff --git a/mm/bootmem_info.c b/mm/bootmem_info.c
index 95f288169a38..b0e2a9fa641f 100644
--- a/mm/bootmem_info.c
+++ b/mm/bootmem_info.c
@@ -88,7 +88,9 @@ static void __init register_page_bootmem_info_section(unsigned long start_pfn)
 
 	memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
 
-	register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
+	if (!preinited_vmemmap_section(ms))
+		register_page_bootmem_memmap(section_nr, memmap,
+				PAGES_PER_SECTION);
 
 	usage = ms->usage;
 	page = virt_to_page(usage);
diff --git a/mm/mm_init.c b/mm/mm_init.c
index d2dee53e95dd..9f1e41c3dde6 100644
--- a/mm/mm_init.c
+++ b/mm/mm_init.c
@@ -1862,6 +1862,9 @@ void __init free_area_init(unsigned long *max_zone_pfn)
 		}
 	}
 
+	for_each_node_state(nid, N_MEMORY)
+		sparse_vmemmap_init_nid_late(nid);
+
 	calc_nr_kernel_pages();
 	memmap_init();
 
diff --git a/mm/sparse-vmemmap.c b/mm/sparse-vmemmap.c
index 3287ebadd167..8751c46c35e4 100644
--- a/mm/sparse-vmemmap.c
+++ b/mm/sparse-vmemmap.c
@@ -470,3 +470,26 @@ struct page * __meminit __populate_section_memmap(unsigned long pfn,
 
 	return pfn_to_page(pfn);
 }
+
+#ifdef CONFIG_SPARSEMEM_VMEMMAP_PREINIT
+/*
+ * This is called just before initializing sections for a NUMA node.
+ * Any special initialization that needs to be done before the
+ * generic initialization can be done from here. Sections that
+ * are initialized in hooks called from here will be skipped by
+ * the generic initialization.
+ */
+void __init sparse_vmemmap_init_nid_early(int nid)
+{
+}
+
+/*
+ * This is called just before the initialization of page structures
+ * through memmap_init. Zones are now initialized, so any work that
+ * needs to be done that needs zone information can be done from
+ * here.
+ */
+void __init sparse_vmemmap_init_nid_late(int nid)
+{
+}
+#endif
diff --git a/mm/sparse.c b/mm/sparse.c
index 133b033d0cba..ee0234a77c7f 100644
--- a/mm/sparse.c
+++ b/mm/sparse.c
@@ -408,13 +408,13 @@ static void __init check_usemap_section_nr(int nid,
 #endif /* CONFIG_MEMORY_HOTREMOVE */
 
 #ifdef CONFIG_SPARSEMEM_VMEMMAP
-static unsigned long __init section_map_size(void)
+unsigned long __init section_map_size(void)
 {
 	return ALIGN(sizeof(struct page) * PAGES_PER_SECTION, PMD_SIZE);
 }
 
 #else
-static unsigned long __init section_map_size(void)
+unsigned long __init section_map_size(void)
 {
 	return PAGE_ALIGN(sizeof(struct page) * PAGES_PER_SECTION);
 }
@@ -495,6 +495,44 @@ void __weak __meminit vmemmap_populate_print_last(void)
 {
 }
 
+static void *sparse_usagebuf __meminitdata;
+static void *sparse_usagebuf_end __meminitdata;
+
+/*
+ * Helper function that is used for generic section initialization, and
+ * can also be used by any hooks added above.
+ */
+void __init sparse_init_early_section(int nid, struct page *map,
+				      unsigned long pnum, unsigned long flags)
+{
+	BUG_ON(!sparse_usagebuf || sparse_usagebuf >= sparse_usagebuf_end);
+	check_usemap_section_nr(nid, sparse_usagebuf);
+	sparse_init_one_section(__nr_to_section(pnum), pnum, map,
+			sparse_usagebuf, SECTION_IS_EARLY | flags);
+	sparse_usagebuf = (void *)sparse_usagebuf + mem_section_usage_size();
+}
+
+static int __init sparse_usage_init(int nid, unsigned long map_count)
+{
+	unsigned long size;
+
+	size = mem_section_usage_size() * map_count;
+	sparse_usagebuf = sparse_early_usemaps_alloc_pgdat_section(
+				NODE_DATA(nid), size);
+	if (!sparse_usagebuf) {
+		sparse_usagebuf_end = NULL;
+		return -ENOMEM;
+	}
+
+	sparse_usagebuf_end = sparse_usagebuf + size;
+	return 0;
+}
+
+static void __init sparse_usage_fini(void)
+{
+	sparse_usagebuf = sparse_usagebuf_end = NULL;
+}
+
 /*
  * Initialize sparse on a specific node. The node spans [pnum_begin, pnum_end)
  * And number of present sections in this node is map_count.
@@ -503,47 +541,54 @@ static void __init sparse_init_nid(int nid, unsigned long pnum_begin,
 				   unsigned long pnum_end,
 				   unsigned long map_count)
 {
-	struct mem_section_usage *usage;
 	unsigned long pnum;
 	struct page *map;
+	struct mem_section *ms;
 
-	usage = sparse_early_usemaps_alloc_pgdat_section(NODE_DATA(nid),
-			mem_section_usage_size() * map_count);
-	if (!usage) {
+	if (sparse_usage_init(nid, map_count)) {
 		pr_err("%s: node[%d] usemap allocation failed", __func__, nid);
 		goto failed;
 	}
+
 	sparse_buffer_init(map_count * section_map_size(), nid);
+
+	sparse_vmemmap_init_nid_early(nid);
+
 	for_each_present_section_nr(pnum_begin, pnum) {
 		unsigned long pfn = section_nr_to_pfn(pnum);
 
 		if (pnum >= pnum_end)
 			break;
 
-		map = __populate_section_memmap(pfn, PAGES_PER_SECTION,
-				nid, NULL, NULL);
-		if (!map) {
-			pr_err("%s: node[%d] memory map backing failed. Some memory will not be available.",
-			       __func__, nid);
-			pnum_begin = pnum;
-			sparse_buffer_fini();
-			goto failed;
+		ms = __nr_to_section(pnum);
+		if (!preinited_vmemmap_section(ms)) {
+			map = __populate_section_memmap(pfn, PAGES_PER_SECTION,
+					nid, NULL, NULL);
+			if (!map) {
+				pr_err("%s: node[%d] memory map backing failed. Some memory will not be available.",
+				       __func__, nid);
+				pnum_begin = pnum;
+				sparse_usage_fini();
+				sparse_buffer_fini();
+				goto failed;
+			}
+			sparse_init_early_section(nid, map, pnum, 0);
 		}
-		check_usemap_section_nr(nid, usage);
-		sparse_init_one_section(__nr_to_section(pnum), pnum, map, usage,
-				SECTION_IS_EARLY);
-		usage = (void *) usage + mem_section_usage_size();
 	}
+	sparse_usage_fini();
 	sparse_buffer_fini();
 	return;
 failed:
-	/* We failed to allocate, mark all the following pnums as not present */
+	/*
+	 * We failed to allocate, mark all the following pnums as not present,
+	 * except the ones already initialized earlier.
+	 */
 	for_each_present_section_nr(pnum_begin, pnum) {
-		struct mem_section *ms;
-
 		if (pnum >= pnum_end)
 			break;
 		ms = __nr_to_section(pnum);
+		if (!preinited_vmemmap_section(ms))
+			ms->section_mem_map = 0;
 		ms->section_mem_map = 0;
 	}
 }
-- 
2.48.1.262.g85cc9f2d1e-goog

[PATCH v2 12/28] mm/hugetlb: set migratetype for bootmem folios

[Frank van der Linden]

The pageblocks that back memblock allocated hugetlb folios might
not have the migrate type set, in the CONFIG_DEFERRED_STRUCT_PAGE_INIT
case.

memblock allocated hugetlb folios might be given to the buddy allocator
eventually (if nr_hugepages is lowered), so make sure that the migrate
type for the pageblocks contained in them is set when initializing them.
Set it to the default that memmap init also uses (MIGRATE_MOVABLE).

Signed-off-by: Frank van der Linden <fvdl@google.com>
---
 mm/hugetlb.c | 22 ++++++++++++++++++++++
 1 file changed, 22 insertions(+)

diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index c16ed9790022..e5ca5cf2c6fd 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -36,6 +36,7 @@
 #include <linux/memory.h>
 #include <linux/mm_inline.h>
 #include <linux/padata.h>
+#include <linux/page-isolation.h>
 
 #include <asm/page.h>
 #include <asm/pgalloc.h>
@@ -3258,6 +3259,26 @@ static void __init hugetlb_folio_init_vmemmap(struct folio *folio,
 	prep_compound_head((struct page *)folio, huge_page_order(h));
 }
 
+/*
+ * memblock-allocated pageblocks might not have the migrate type set
+ * if marked with the 'noinit' flag. Set it to the default (MIGRATE_MOVABLE)
+ * here.
+ *
+ * Note that this will not write the page struct, it is ok (and necessary)
+ * to do this on vmemmap optimized folios.
+ */
+static void __init hugetlb_bootmem_init_migratetype(struct folio *folio,
+							  struct hstate *h)
+{
+	unsigned long nr_pages = pages_per_huge_page(h), i;
+
+	WARN_ON_ONCE(!pageblock_aligned(folio_pfn(folio)));
+
+	for (i = 0; i < nr_pages; i += pageblock_nr_pages)
+		set_pageblock_migratetype(folio_page(folio, i),
+					  MIGRATE_MOVABLE);
+}
+
 static void __init prep_and_add_bootmem_folios(struct hstate *h,
 					struct list_head *folio_list)
 {
@@ -3279,6 +3300,7 @@ static void __init prep_and_add_bootmem_folios(struct hstate *h,
 					HUGETLB_VMEMMAP_RESERVE_PAGES,
 					pages_per_huge_page(h));
 		}
+		hugetlb_bootmem_init_migratetype(folio, h);
 		/* Subdivide locks to achieve better parallel performance */
 		spin_lock_irqsave(&hugetlb_lock, flags);
 		__prep_account_new_huge_page(h, folio_nid(folio));
-- 
2.48.1.262.g85cc9f2d1e-goog

[PATCH v2 13/28] mm: define __init_reserved_page_zone function

[Frank van der Linden]

Sometimes page structs must be unconditionally initialized
as reserved, regardless of DEFERRED_STRUCT_PAGE_INIT.

Define a function, __init_reserved_page_zone, containing
code that already did all of the work in init_reserved_page,
and make it available for use.

Signed-off-by: Frank van der Linden <fvdl@google.com>
---
 mm/internal.h |  1 +
 mm/mm_init.c  | 38 +++++++++++++++++++++++---------------
 2 files changed, 24 insertions(+), 15 deletions(-)

diff --git a/mm/internal.h b/mm/internal.h
index 109ef30fee11..57662141930e 100644
--- a/mm/internal.h
+++ b/mm/internal.h
@@ -1448,6 +1448,7 @@ static inline bool pte_needs_soft_dirty_wp(struct vm_area_struct *vma, pte_t pte
 
 void __meminit __init_single_page(struct page *page, unsigned long pfn,
 				unsigned long zone, int nid);
+void __meminit __init_reserved_page_zone(unsigned long pfn, int nid);
 
 /* shrinker related functions */
 unsigned long shrink_slab(gfp_t gfp_mask, int nid, struct mem_cgroup *memcg,
diff --git a/mm/mm_init.c b/mm/mm_init.c
index 9f1e41c3dde6..925ed6564572 100644
--- a/mm/mm_init.c
+++ b/mm/mm_init.c
@@ -650,6 +650,28 @@ static inline void fixup_hashdist(void)
 static inline void fixup_hashdist(void) {}
 #endif /* CONFIG_NUMA */
 
+/*
+ * Initialize a reserved page unconditionally, finding its zone first.
+ */
+void __meminit __init_reserved_page_zone(unsigned long pfn, int nid)
+{
+	pg_data_t *pgdat;
+	int zid;
+
+	pgdat = NODE_DATA(nid);
+
+	for (zid = 0; zid < MAX_NR_ZONES; zid++) {
+		struct zone *zone = &pgdat->node_zones[zid];
+
+		if (zone_spans_pfn(zone, pfn))
+			break;
+	}
+	__init_single_page(pfn_to_page(pfn), pfn, zid, nid);
+
+	if (pageblock_aligned(pfn))
+		set_pageblock_migratetype(pfn_to_page(pfn), MIGRATE_MOVABLE);
+}
+
 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
 static inline void pgdat_set_deferred_range(pg_data_t *pgdat)
 {
@@ -708,24 +730,10 @@ defer_init(int nid, unsigned long pfn, unsigned long end_pfn)
 
 static void __meminit init_reserved_page(unsigned long pfn, int nid)
 {
-	pg_data_t *pgdat;
-	int zid;
-
 	if (early_page_initialised(pfn, nid))
 		return;
 
-	pgdat = NODE_DATA(nid);
-
-	for (zid = 0; zid < MAX_NR_ZONES; zid++) {
-		struct zone *zone = &pgdat->node_zones[zid];
-
-		if (zone_spans_pfn(zone, pfn))
-			break;
-	}
-	__init_single_page(pfn_to_page(pfn), pfn, zid, nid);
-
-	if (pageblock_aligned(pfn))
-		set_pageblock_migratetype(pfn_to_page(pfn), MIGRATE_MOVABLE);
+	__init_reserved_page_zone(pfn, nid);
 }
 #else
 static inline void pgdat_set_deferred_range(pg_data_t *pgdat) {}
-- 
2.48.1.262.g85cc9f2d1e-goog

[PATCH v2 14/28] mm/hugetlb: check bootmem pages for zone intersections

[Frank van der Linden]

Bootmem hugetlb pages are allocated using memblock, which isn't
(and mostly can't be) aware of zones.

So, they may end up crossing zone boundaries. This would create
confusion, a hugetlb page that is part of multiple zones is bad.
Worse, HVO might then end up stealthily re-assigning pages to a
different zone when a hugetlb page is freed, since the tail page
structures beyond the first vmemmap page would inherit the zone
of the first page structures.

While the chance of this happening is low, you can definitely
create a configuration where this happens (especially using
ZONE_MOVABLE).

To avoid this issue, check if bootmem hugetlb pages intersect
with multiple zones during the gather phase, and discard
them, handing them to the page allocator, if they do. Record
the number of invalid bootmem pages per node and subtract them
from the number of available pages at the end, making it easier
to do these checks in multiple places later on.

Signed-off-by: Frank van der Linden <fvdl@google.com>
---
 mm/hugetlb.c  | 61 +++++++++++++++++++++++++++++++++++++++++++++++++--
 mm/internal.h |  2 ++
 mm/mm_init.c  | 25 +++++++++++++++++++++
 3 files changed, 86 insertions(+), 2 deletions(-)

diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index e5ca5cf2c6fd..a0a87d1a8569 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -63,6 +63,7 @@ static unsigned long hugetlb_cma_size_in_node[MAX_NUMNODES] __initdata;
 static unsigned long hugetlb_cma_size __initdata;
 
 __initdata struct list_head huge_boot_pages[MAX_NUMNODES];
+__initdata unsigned long hstate_boot_nrinvalid[HUGE_MAX_HSTATE];
 
 /*
  * Due to ordering constraints across the init code for various
@@ -3309,6 +3310,44 @@ static void __init prep_and_add_bootmem_folios(struct hstate *h,
 	}
 }
 
+static bool __init hugetlb_bootmem_page_zones_valid(int nid,
+						    struct huge_bootmem_page *m)
+{
+	unsigned long start_pfn;
+	bool valid;
+
+	start_pfn = virt_to_phys(m) >> PAGE_SHIFT;
+
+	valid = !pfn_range_intersects_zones(nid, start_pfn,
+			pages_per_huge_page(m->hstate));
+	if (!valid)
+		hstate_boot_nrinvalid[hstate_index(m->hstate)]++;
+
+	return valid;
+}
+
+/*
+ * Free a bootmem page that was found to be invalid (intersecting with
+ * multiple zones).
+ *
+ * Since it intersects with multiple zones, we can't just do a free
+ * operation on all pages at once, but instead have to walk all
+ * pages, freeing them one by one.
+ */
+static void __init hugetlb_bootmem_free_invalid_page(int nid, struct page *page,
+					     struct hstate *h)
+{
+	unsigned long npages = pages_per_huge_page(h);
+	unsigned long pfn;
+
+	while (npages--) {
+		pfn = page_to_pfn(page);
+		__init_reserved_page_zone(pfn, nid);
+		free_reserved_page(page);
+		page++;
+	}
+}
+
 /*
  * Put bootmem huge pages into the standard lists after mem_map is up.
  * Note: This only applies to gigantic (order > MAX_PAGE_ORDER) pages.
@@ -3316,14 +3355,25 @@ static void __init prep_and_add_bootmem_folios(struct hstate *h,
 static void __init gather_bootmem_prealloc_node(unsigned long nid)
 {
 	LIST_HEAD(folio_list);
-	struct huge_bootmem_page *m;
+	struct huge_bootmem_page *m, *tm;
 	struct hstate *h = NULL, *prev_h = NULL;
 
-	list_for_each_entry(m, &huge_boot_pages[nid], list) {
+	list_for_each_entry_safe(m, tm, &huge_boot_pages[nid], list) {
 		struct page *page = virt_to_page(m);
 		struct folio *folio = (void *)page;
 
 		h = m->hstate;
+		if (!hugetlb_bootmem_page_zones_valid(nid, m)) {
+			/*
+			 * Can't use this page. Initialize the
+			 * page structures if that hasn't already
+			 * been done, and give them to the page
+			 * allocator.
+			 */
+			hugetlb_bootmem_free_invalid_page(nid, page, h);
+			continue;
+		}
+
 		/*
 		 * It is possible to have multiple huge page sizes (hstates)
 		 * in this list.  If so, process each size separately.
@@ -3595,13 +3645,20 @@ static void __init hugetlb_init_hstates(void)
 static void __init report_hugepages(void)
 {
 	struct hstate *h;
+	unsigned long nrinvalid;
 
 	for_each_hstate(h) {
 		char buf[32];
 
+		nrinvalid = hstate_boot_nrinvalid[hstate_index(h)];
+		h->max_huge_pages -= nrinvalid;
+
 		string_get_size(huge_page_size(h), 1, STRING_UNITS_2, buf, 32);
 		pr_info("HugeTLB: registered %s page size, pre-allocated %ld pages\n",
 			buf, h->free_huge_pages);
+		if (nrinvalid)
+			pr_info("HugeTLB: %s page size: %lu invalid page%s discarded\n",
+					buf, nrinvalid, nrinvalid > 1 ? "s" : "");
 		pr_info("HugeTLB: %d KiB vmemmap can be freed for a %s page\n",
 			hugetlb_vmemmap_optimizable_size(h) / SZ_1K, buf);
 	}
diff --git a/mm/internal.h b/mm/internal.h
index 57662141930e..63fda9bb9426 100644
--- a/mm/internal.h
+++ b/mm/internal.h
@@ -658,6 +658,8 @@ static inline struct page *pageblock_pfn_to_page(unsigned long start_pfn,
 }
 
 void set_zone_contiguous(struct zone *zone);
+bool pfn_range_intersects_zones(int nid, unsigned long start_pfn,
+			   unsigned long nr_pages);
 
 static inline void clear_zone_contiguous(struct zone *zone)
 {
diff --git a/mm/mm_init.c b/mm/mm_init.c
index 925ed6564572..f7d5b4fe1ae9 100644
--- a/mm/mm_init.c
+++ b/mm/mm_init.c
@@ -2287,6 +2287,31 @@ void set_zone_contiguous(struct zone *zone)
 	zone->contiguous = true;
 }
 
+/*
+ * Check if a PFN range intersects multiple zones on one or more
+ * NUMA nodes. Specify the @nid argument if it is known that this
+ * PFN range is on one node, NUMA_NO_NODE otherwise.
+ */
+bool pfn_range_intersects_zones(int nid, unsigned long start_pfn,
+			   unsigned long nr_pages)
+{
+	struct zone *zone, *izone = NULL;
+
+	for_each_zone(zone) {
+		if (nid != NUMA_NO_NODE && zone_to_nid(zone) != nid)
+			continue;
+
+		if (zone_intersects(zone, start_pfn, nr_pages)) {
+			if (izone != NULL)
+				return true;
+			izone = zone;
+		}
+
+	}
+
+	return false;
+}
+
 static void __init mem_init_print_info(void);
 void __init page_alloc_init_late(void)
 {
-- 
2.48.1.262.g85cc9f2d1e-goog

[PATCH v2 15/28] mm/sparse: add vmemmap_*_hvo functions

[Frank van der Linden]

Add a few functions to enable early HVO:

vmemmap_populate_hvo
vmemmap_undo_hvo
vmemmap_wrprotect_hvo

The populate and undo functions are expected to be used in early
init, from the sparse_init_nid_early() function. The wrprotect
function is to be used, potentially, later.

To implement these functions, mostly re-use the existing
compound pages vmemmap logic used by DAX. vmemmap_populate_address
has its argument changed a bit in this commit: the page structure
passed in to be reused in the mapping is replaced by a PFN and a
flag. The flag indicates whether an extra ref should be taken on
the vmemmap page containing the head page structure. Taking the
ref is appropriate to for DAX / ZONE_DEVICE, but not for HugeTLB
HVO.

The HugeTLB vmemmap optimization maps tail page structure pages
read-only. The vmemmap_wrprotect_hvo function that does this is
implemented separately, because it cannot be guaranteed that reserved
page structures will not be write accessed during memory initialization.
Even with CONFIG_DEFERRED_STRUCT_PAGE_INIT, they might still be
written to (if they are at the bottom of a zone). So,
vmemmap_populate_hvo leaves the tail page structure pages RW initially,
and then later during initialization, after memmap init is fully done,
vmemmap_wrprotect_hvo must be called to finish the job.

Subsequent commits will use these functions for early HugeTLB HVO.

Signed-off-by: Frank van der Linden <fvdl@google.com>
---
 include/linux/mm.h  |   9 ++-
 mm/sparse-vmemmap.c | 141 +++++++++++++++++++++++++++++++++++++++-----
 2 files changed, 135 insertions(+), 15 deletions(-)

diff --git a/include/linux/mm.h b/include/linux/mm.h
index df83653ed6e3..0463c062fd7a 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -3837,7 +3837,8 @@ p4d_t *vmemmap_p4d_populate(pgd_t *pgd, unsigned long addr, int node);
 pud_t *vmemmap_pud_populate(p4d_t *p4d, unsigned long addr, int node);
 pmd_t *vmemmap_pmd_populate(pud_t *pud, unsigned long addr, int node);
 pte_t *vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node,
-			    struct vmem_altmap *altmap, struct page *reuse);
+			    struct vmem_altmap *altmap, unsigned long ptpfn,
+			    unsigned long flags);
 void *vmemmap_alloc_block(unsigned long size, int node);
 struct vmem_altmap;
 void *vmemmap_alloc_block_buf(unsigned long size, int node,
@@ -3853,6 +3854,12 @@ int vmemmap_populate_hugepages(unsigned long start, unsigned long end,
 			       int node, struct vmem_altmap *altmap);
 int vmemmap_populate(unsigned long start, unsigned long end, int node,
 		struct vmem_altmap *altmap);
+int vmemmap_populate_hvo(unsigned long start, unsigned long end, int node,
+			 unsigned long headsize);
+int vmemmap_undo_hvo(unsigned long start, unsigned long end, int node,
+		     unsigned long headsize);
+void vmemmap_wrprotect_hvo(unsigned long start, unsigned long end, int node,
+			  unsigned long headsize);
 void vmemmap_populate_print_last(void);
 #ifdef CONFIG_MEMORY_HOTPLUG
 void vmemmap_free(unsigned long start, unsigned long end,
diff --git a/mm/sparse-vmemmap.c b/mm/sparse-vmemmap.c
index 8751c46c35e4..bee22ca93654 100644
--- a/mm/sparse-vmemmap.c
+++ b/mm/sparse-vmemmap.c
@@ -30,6 +30,13 @@
 
 #include <asm/dma.h>
 #include <asm/pgalloc.h>
+#include <asm/tlbflush.h>
+
+/*
+ * Flags for vmemmap_populate_range and friends.
+ */
+/* Get a ref on the head page struct page, for ZONE_DEVICE compound pages */
+#define VMEMMAP_POPULATE_PAGEREF	0x0001
 
 #include "internal.h"
 
@@ -144,17 +151,18 @@ void __meminit vmemmap_verify(pte_t *pte, int node,
 
 pte_t * __meminit vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node,
 				       struct vmem_altmap *altmap,
-				       struct page *reuse)
+				       unsigned long ptpfn, unsigned long flags)
 {
 	pte_t *pte = pte_offset_kernel(pmd, addr);
 	if (pte_none(ptep_get(pte))) {
 		pte_t entry;
 		void *p;
 
-		if (!reuse) {
+		if (!ptpfn) {
 			p = vmemmap_alloc_block_buf(PAGE_SIZE, node, altmap);
 			if (!p)
 				return NULL;
+			ptpfn = PHYS_PFN(__pa(p));
 		} else {
 			/*
 			 * When a PTE/PMD entry is freed from the init_mm
@@ -165,10 +173,10 @@ pte_t * __meminit vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node,
 			 * and through vmemmap_populate_compound_pages() when
 			 * slab is available.
 			 */
-			get_page(reuse);
-			p = page_to_virt(reuse);
+			if (flags & VMEMMAP_POPULATE_PAGEREF)
+				get_page(pfn_to_page(ptpfn));
 		}
-		entry = pfn_pte(__pa(p) >> PAGE_SHIFT, PAGE_KERNEL);
+		entry = pfn_pte(ptpfn, PAGE_KERNEL);
 		set_pte_at(&init_mm, addr, pte, entry);
 	}
 	return pte;
@@ -238,7 +246,8 @@ pgd_t * __meminit vmemmap_pgd_populate(unsigned long addr, int node)
 
 static pte_t * __meminit vmemmap_populate_address(unsigned long addr, int node,
 					      struct vmem_altmap *altmap,
-					      struct page *reuse)
+					      unsigned long ptpfn,
+					      unsigned long flags)
 {
 	pgd_t *pgd;
 	p4d_t *p4d;
@@ -258,7 +267,7 @@ static pte_t * __meminit vmemmap_populate_address(unsigned long addr, int node,
 	pmd = vmemmap_pmd_populate(pud, addr, node);
 	if (!pmd)
 		return NULL;
-	pte = vmemmap_pte_populate(pmd, addr, node, altmap, reuse);
+	pte = vmemmap_pte_populate(pmd, addr, node, altmap, ptpfn, flags);
 	if (!pte)
 		return NULL;
 	vmemmap_verify(pte, node, addr, addr + PAGE_SIZE);
@@ -269,13 +278,15 @@ static pte_t * __meminit vmemmap_populate_address(unsigned long addr, int node,
 static int __meminit vmemmap_populate_range(unsigned long start,
 					    unsigned long end, int node,
 					    struct vmem_altmap *altmap,
-					    struct page *reuse)
+					    unsigned long ptpfn,
+					    unsigned long flags)
 {
 	unsigned long addr = start;
 	pte_t *pte;
 
 	for (; addr < end; addr += PAGE_SIZE) {
-		pte = vmemmap_populate_address(addr, node, altmap, reuse);
+		pte = vmemmap_populate_address(addr, node, altmap,
+					       ptpfn, flags);
 		if (!pte)
 			return -ENOMEM;
 	}
@@ -286,7 +297,107 @@ static int __meminit vmemmap_populate_range(unsigned long start,
 int __meminit vmemmap_populate_basepages(unsigned long start, unsigned long end,
 					 int node, struct vmem_altmap *altmap)
 {
-	return vmemmap_populate_range(start, end, node, altmap, NULL);
+	return vmemmap_populate_range(start, end, node, altmap, 0, 0);
+}
+
+/*
+ * Undo populate_hvo, and replace it with a normal base page mapping.
+ * Used in memory init in case a HVO mapping needs to be undone.
+ *
+ * This can happen when it is discovered that a memblock allocated
+ * hugetlb page spans multiple zones, which can only be verified
+ * after zones have been initialized.
+ *
+ * We know that:
+ * 1) The first @headsize / PAGE_SIZE vmemmap pages were individually
+ *    allocated through memblock, and mapped.
+ *
+ * 2) The rest of the vmemmap pages are mirrors of the last head page.
+ */
+int __meminit vmemmap_undo_hvo(unsigned long addr, unsigned long end,
+				      int node, unsigned long headsize)
+{
+	unsigned long maddr, pfn;
+	pte_t *pte;
+	int headpages;
+
+	/*
+	 * Should only be called early in boot, so nothing will
+	 * be accessing these page structures.
+	 */
+	WARN_ON(!early_boot_irqs_disabled);
+
+	headpages = headsize >> PAGE_SHIFT;
+
+	/*
+	 * Clear mirrored mappings for tail page structs.
+	 */
+	for (maddr = addr + headsize; maddr < end; maddr += PAGE_SIZE) {
+		pte = virt_to_kpte(maddr);
+		pte_clear(&init_mm, maddr, pte);
+	}
+
+	/*
+	 * Clear and free mappings for head page and first tail page
+	 * structs.
+	 */
+	for (maddr = addr; headpages-- > 0; maddr += PAGE_SIZE) {
+		pte = virt_to_kpte(maddr);
+		pfn = pte_pfn(ptep_get(pte));
+		pte_clear(&init_mm, maddr, pte);
+		memblock_phys_free(PFN_PHYS(pfn), PAGE_SIZE);
+	}
+
+	flush_tlb_kernel_range(addr, end);
+
+	return vmemmap_populate(addr, end, node, NULL);
+}
+
+/*
+ * Write protect the mirrored tail page structs for HVO. This will be
+ * called from the hugetlb code when gathering and initializing the
+ * memblock allocated gigantic pages. The write protect can't be
+ * done earlier, since it can't be guaranteed that the reserved
+ * page structures will not be written to during initialization,
+ * even if CONFIG_DEFERRED_STRUCT_PAGE_INIT is enabled.
+ *
+ * The PTEs are known to exist, and nothing else should be touching
+ * these pages. The caller is responsible for any TLB flushing.
+ */
+void vmemmap_wrprotect_hvo(unsigned long addr, unsigned long end,
+				    int node, unsigned long headsize)
+{
+	unsigned long maddr;
+	pte_t *pte;
+
+	for (maddr = addr + headsize; maddr < end; maddr += PAGE_SIZE) {
+		pte = virt_to_kpte(maddr);
+		ptep_set_wrprotect(&init_mm, maddr, pte);
+	}
+}
+
+/*
+ * Populate vmemmap pages HVO-style. The first page contains the head
+ * page and needed tail pages, the other ones are mirrors of the first
+ * page.
+ */
+int __meminit vmemmap_populate_hvo(unsigned long addr, unsigned long end,
+				       int node, unsigned long headsize)
+{
+	pte_t *pte;
+	unsigned long maddr;
+
+	for (maddr = addr; maddr < addr + headsize; maddr += PAGE_SIZE) {
+		pte = vmemmap_populate_address(maddr, node, NULL, 0, 0);
+		if (!pte)
+			return -ENOMEM;
+	}
+
+	/*
+	 * Reuse the last page struct page mapped above for the rest.
+	 */
+	return vmemmap_populate_range(maddr, end, node, NULL,
+					pte_pfn(ptep_get(pte)), 0);
 }
 
 void __weak __meminit vmemmap_set_pmd(pmd_t *pmd, void *p, int node,
@@ -409,7 +520,8 @@ static int __meminit vmemmap_populate_compound_pages(unsigned long start_pfn,
 		 * with just tail struct pages.
 		 */
 		return vmemmap_populate_range(start, end, node, NULL,
-					      pte_page(ptep_get(pte)));
+					      pte_pfn(ptep_get(pte)),
+					      VMEMMAP_POPULATE_PAGEREF);
 	}
 
 	size = min(end - start, pgmap_vmemmap_nr(pgmap) * sizeof(struct page));
@@ -417,13 +529,13 @@ static int __meminit vmemmap_populate_compound_pages(unsigned long start_pfn,
 		unsigned long next, last = addr + size;
 
 		/* Populate the head page vmemmap page */
-		pte = vmemmap_populate_address(addr, node, NULL, NULL);
+		pte = vmemmap_populate_address(addr, node, NULL, 0, 0);
 		if (!pte)
 			return -ENOMEM;
 
 		/* Populate the tail pages vmemmap page */
 		next = addr + PAGE_SIZE;
-		pte = vmemmap_populate_address(next, node, NULL, NULL);
+		pte = vmemmap_populate_address(next, node, NULL, 0, 0);
 		if (!pte)
 			return -ENOMEM;
 
@@ -433,7 +545,8 @@ static int __meminit vmemmap_populate_compound_pages(unsigned long start_pfn,
 		 */
 		next += PAGE_SIZE;
 		rc = vmemmap_populate_range(next, last, node, NULL,
-					    pte_page(ptep_get(pte)));
+					    pte_pfn(ptep_get(pte)),
+					    VMEMMAP_POPULATE_PAGEREF);
 		if (rc)
 			return -ENOMEM;
 	}
-- 
2.48.1.262.g85cc9f2d1e-goog

[PATCH v2 16/28] mm/hugetlb: deal with multiple calls to hugetlb_bootmem_alloc

[Frank van der Linden]

Architectures that want pre-HVO of hugetlb vmemmap pages
will need to call hugetlb_bootmem_alloc from an earlier
spot in boot (before sparse_init). To facilitate some
architectures doing this, protect hugetlb_bootmem_alloc
against multiple calls.

Also provide a helper function to check if it's been called,
so that the early HVO code, to be added later, can see if there
is anything to do.

Signed-off-by: Frank van der Linden <fvdl@google.com>
---
 include/linux/hugetlb.h |  6 ++++++
 mm/hugetlb.c            | 12 ++++++++++++
 2 files changed, 18 insertions(+)

diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h
index 9cd7c9dacb88..5061279e5f73 100644
--- a/include/linux/hugetlb.h
+++ b/include/linux/hugetlb.h
@@ -175,6 +175,7 @@ extern int sysctl_hugetlb_shm_group;
 extern struct list_head huge_boot_pages[MAX_NUMNODES];
 
 void hugetlb_bootmem_alloc(void);
+bool hugetlb_bootmem_allocated(void);
 
 /* arch callbacks */
 
@@ -1256,6 +1257,11 @@ static inline bool hugetlbfs_pagecache_present(
 static inline void hugetlb_bootmem_alloc(void)
 {
 }
+
+static inline bool hugetlb_bootmem_allocated(void)
+{
+	return false;
+}
 #endif	/* CONFIG_HUGETLB_PAGE */
 
 static inline spinlock_t *huge_pte_lock(struct hstate *h,
diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index a0a87d1a8569..0a27659d9290 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -4911,16 +4911,28 @@ static int __init default_hugepagesz_setup(char *s)
 }
 hugetlb_early_param("default_hugepagesz", default_hugepagesz_setup);
 
+static bool __hugetlb_bootmem_allocated __initdata;
+
+bool __init hugetlb_bootmem_allocated(void)
+{
+	return __hugetlb_bootmem_allocated;
+}
+
 void __init hugetlb_bootmem_alloc(void)
 {
 	struct hstate *h;
 
+	if (__hugetlb_bootmem_allocated)
+		return;
+
 	hugetlb_parse_params();
 
 	for_each_hstate(h) {
 		if (hstate_is_gigantic(h))
 			hugetlb_hstate_alloc_pages(h);
 	}
+
+	__hugetlb_bootmem_allocated = true;
 }
 
 static unsigned int allowed_mems_nr(struct hstate *h)
-- 
2.48.1.262.g85cc9f2d1e-goog

[PATCH v2 17/28] mm/hugetlb: move huge_boot_pages list init to hugetlb_bootmem_alloc

[Frank van der Linden]

Instead of initializing the per-node hugetlb bootmem pages list
from the alloc function, we can now do it in a somewhat cleaner
way, since there is an explicit hugetlb_bootmem_alloc function.
Initialize the lists there.

Signed-off-by: Frank van der Linden <fvdl@google.com>
---
 mm/hugetlb.c | 19 +++++++------------
 1 file changed, 7 insertions(+), 12 deletions(-)

diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index 0a27659d9290..7879e772c0d9 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -3579,7 +3579,6 @@ static unsigned long __init hugetlb_pages_alloc_boot(struct hstate *h)
 static void __init hugetlb_hstate_alloc_pages(struct hstate *h)
 {
 	unsigned long allocated;
-	static bool initialized __initdata;
 
 	/* skip gigantic hugepages allocation if hugetlb_cma enabled */
 	if (hstate_is_gigantic(h) && hugetlb_cma_size) {
@@ -3587,17 +3586,6 @@ static void __init hugetlb_hstate_alloc_pages(struct hstate *h)
 		return;
 	}
 
-	/* hugetlb_hstate_alloc_pages will be called many times, initialize huge_boot_pages once */
-	if (!initialized) {
-		int i = 0;
-
-		for (i = 0; i < MAX_NUMNODES; i++)
-			INIT_LIST_HEAD(&huge_boot_pages[i]);
-		h->next_nid_to_alloc = first_online_node;
-		h->next_nid_to_free = first_online_node;
-		initialized = true;
-	}
-
 	/* do node specific alloc */
 	if (hugetlb_hstate_alloc_pages_specific_nodes(h))
 		return;
@@ -4921,13 +4909,20 @@ bool __init hugetlb_bootmem_allocated(void)
 void __init hugetlb_bootmem_alloc(void)
 {
 	struct hstate *h;
+	int i;
 
 	if (__hugetlb_bootmem_allocated)
 		return;
 
+	for (i = 0; i < MAX_NUMNODES; i++)
+		INIT_LIST_HEAD(&huge_boot_pages[i]);
+
 	hugetlb_parse_params();
 
 	for_each_hstate(h) {
+		h->next_nid_to_alloc = first_online_node;
+		h->next_nid_to_free = first_online_node;
+
 		if (hstate_is_gigantic(h))
 			hugetlb_hstate_alloc_pages(h);
 	}
-- 
2.48.1.262.g85cc9f2d1e-goog

[PATCH v2 18/28] mm/hugetlb: add pre-HVO framework

[Frank van der Linden]

Define flags for pre-HVOed bootmem hugetlb pages, and act on them.

The most important flag is the HVO flag, signalling that a bootmem
allocated gigantic page has already been HVO-ed. If this flag is
seen by the hugetlb bootmem gather code, the page is marked
as HVO optimized. The HVO code will then not try to optimize
it again. Instead, it will just map the tail page mirror pages
read-only, completing the HVO steps.

No functional change, as nothing sets the flags yet.

Signed-off-by: Frank van der Linden <fvdl@google.com>
---
 arch/powerpc/mm/hugetlbpage.c |  1 +
 include/linux/hugetlb.h       |  4 +++
 mm/hugetlb.c                  | 24 ++++++++++++++++-
 mm/hugetlb_vmemmap.c          | 50 +++++++++++++++++++++++++++++++++--
 mm/hugetlb_vmemmap.h          | 15 +++++++++++
 5 files changed, 91 insertions(+), 3 deletions(-)

diff --git a/arch/powerpc/mm/hugetlbpage.c b/arch/powerpc/mm/hugetlbpage.c
index 6b043180220a..d3c1b749dcfc 100644
--- a/arch/powerpc/mm/hugetlbpage.c
+++ b/arch/powerpc/mm/hugetlbpage.c
@@ -113,6 +113,7 @@ static int __init pseries_alloc_bootmem_huge_page(struct hstate *hstate)
 	gpage_freearray[nr_gpages] = 0;
 	list_add(&m->list, &huge_boot_pages[0]);
 	m->hstate = hstate;
+	m->flags = 0;
 	return 1;
 }
 
diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h
index 5061279e5f73..10a7ce2b95e1 100644
--- a/include/linux/hugetlb.h
+++ b/include/linux/hugetlb.h
@@ -681,8 +681,12 @@ struct hstate {
 struct huge_bootmem_page {
 	struct list_head list;
 	struct hstate *hstate;
+	unsigned long flags;
 };
 
+#define HUGE_BOOTMEM_HVO		0x0001
+#define HUGE_BOOTMEM_ZONES_VALID	0x0002
+
 int isolate_or_dissolve_huge_page(struct page *page, struct list_head *list);
 int replace_free_hugepage_folios(unsigned long start_pfn, unsigned long end_pfn);
 struct folio *alloc_hugetlb_folio(struct vm_area_struct *vma,
diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index 7879e772c0d9..b48f8638c9af 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -3220,6 +3220,7 @@ int __alloc_bootmem_huge_page(struct hstate *h, int nid)
 	INIT_LIST_HEAD(&m->list);
 	list_add(&m->list, &huge_boot_pages[node]);
 	m->hstate = h;
+	m->flags = 0;
 	return 1;
 }
 
@@ -3287,7 +3288,7 @@ static void __init prep_and_add_bootmem_folios(struct hstate *h,
 	struct folio *folio, *tmp_f;
 
 	/* Send list for bulk vmemmap optimization processing */
-	hugetlb_vmemmap_optimize_folios(h, folio_list);
+	hugetlb_vmemmap_optimize_bootmem_folios(h, folio_list);
 
 	list_for_each_entry_safe(folio, tmp_f, folio_list, lru) {
 		if (!folio_test_hugetlb_vmemmap_optimized(folio)) {
@@ -3316,6 +3317,13 @@ static bool __init hugetlb_bootmem_page_zones_valid(int nid,
 	unsigned long start_pfn;
 	bool valid;
 
+	if (m->flags & HUGE_BOOTMEM_ZONES_VALID) {
+		/*
+		 * Already validated, skip check.
+		 */
+		return true;
+	}
+
 	start_pfn = virt_to_phys(m) >> PAGE_SHIFT;
 
 	valid = !pfn_range_intersects_zones(nid, start_pfn,
@@ -3348,6 +3356,11 @@ static void __init hugetlb_bootmem_free_invalid_page(int nid, struct page *page,
 	}
 }
 
+static bool __init hugetlb_bootmem_page_prehvo(struct huge_bootmem_page *m)
+{
+	return (m->flags & HUGE_BOOTMEM_HVO);
+}
+
 /*
  * Put bootmem huge pages into the standard lists after mem_map is up.
  * Note: This only applies to gigantic (order > MAX_PAGE_ORDER) pages.
@@ -3388,6 +3401,15 @@ static void __init gather_bootmem_prealloc_node(unsigned long nid)
 		hugetlb_folio_init_vmemmap(folio, h,
 					   HUGETLB_VMEMMAP_RESERVE_PAGES);
 		init_new_hugetlb_folio(h, folio);
+
+		if (hugetlb_bootmem_page_prehvo(m))
+			/*
+			 * If pre-HVO was done, just set the
+			 * flag, the HVO code will then skip
+			 * this folio.
+			 */
+			folio_set_hugetlb_vmemmap_optimized(folio);
+
 		list_add(&folio->lru, &folio_list);
 
 		/*
diff --git a/mm/hugetlb_vmemmap.c b/mm/hugetlb_vmemmap.c
index 5b484758f813..be6b33ecbc8e 100644
--- a/mm/hugetlb_vmemmap.c
+++ b/mm/hugetlb_vmemmap.c
@@ -649,14 +649,39 @@ static int hugetlb_vmemmap_split_folio(const struct hstate *h, struct folio *fol
 	return vmemmap_remap_split(vmemmap_start, vmemmap_end, vmemmap_reuse);
 }
 
-void hugetlb_vmemmap_optimize_folios(struct hstate *h, struct list_head *folio_list)
+static void __hugetlb_vmemmap_optimize_folios(struct hstate *h,
+					      struct list_head *folio_list,
+					      bool boot)
 {
 	struct folio *folio;
+	int nr_to_optimize;
 	LIST_HEAD(vmemmap_pages);
 	unsigned long flags = VMEMMAP_REMAP_NO_TLB_FLUSH | VMEMMAP_SYNCHRONIZE_RCU;
 
+	nr_to_optimize = 0;
 	list_for_each_entry(folio, folio_list, lru) {
-		int ret = hugetlb_vmemmap_split_folio(h, folio);
+		int ret;
+		unsigned long spfn, epfn;
+
+		if (boot && folio_test_hugetlb_vmemmap_optimized(folio)) {
+			/*
+			 * Already optimized by pre-HVO, just map the
+			 * mirrored tail page structs RO.
+			 */
+			spfn = (unsigned long)&folio->page;
+			epfn = spfn + pages_per_huge_page(h);
+			vmemmap_wrprotect_hvo(spfn, epfn, folio_nid(folio),
+					HUGETLB_VMEMMAP_RESERVE_SIZE);
+			register_page_bootmem_memmap(pfn_to_section_nr(spfn),
+					&folio->page,
+					HUGETLB_VMEMMAP_RESERVE_SIZE);
+			static_branch_inc(&hugetlb_optimize_vmemmap_key);
+			continue;
+		}
+
+		nr_to_optimize++;
+
+		ret = hugetlb_vmemmap_split_folio(h, folio);
 
 		/*
 		 * Spliting the PMD requires allocating a page, thus lets fail
@@ -668,6 +693,16 @@ void hugetlb_vmemmap_optimize_folios(struct hstate *h, struct list_head *folio_l
 			break;
 	}
 
+	if (!nr_to_optimize)
+		/*
+		 * All pre-HVO folios, nothing left to do. It's ok if
+		 * there is a mix of pre-HVO and not yet HVO-ed folios
+		 * here, as __hugetlb_vmemmap_optimize_folio() will
+		 * skip any folios that already have the optimized flag
+		 * set, see vmemmap_should_optimize_folio().
+		 */
+		goto out;
+
 	flush_tlb_all();
 
 	list_for_each_entry(folio, folio_list, lru) {
@@ -693,10 +728,21 @@ void hugetlb_vmemmap_optimize_folios(struct hstate *h, struct list_head *folio_l
 		}
 	}
 
+out:
 	flush_tlb_all();
 	free_vmemmap_page_list(&vmemmap_pages);
 }
 
+void hugetlb_vmemmap_optimize_folios(struct hstate *h, struct list_head *folio_list)
+{
+	__hugetlb_vmemmap_optimize_folios(h, folio_list, false);
+}
+
+void hugetlb_vmemmap_optimize_bootmem_folios(struct hstate *h, struct list_head *folio_list)
+{
+	__hugetlb_vmemmap_optimize_folios(h, folio_list, true);
+}
+
 static const struct ctl_table hugetlb_vmemmap_sysctls[] = {
 	{
 		.procname	= "hugetlb_optimize_vmemmap",
diff --git a/mm/hugetlb_vmemmap.h b/mm/hugetlb_vmemmap.h
index 2fcae92d3359..a6354a27e63f 100644
--- a/mm/hugetlb_vmemmap.h
+++ b/mm/hugetlb_vmemmap.h
@@ -24,6 +24,8 @@ long hugetlb_vmemmap_restore_folios(const struct hstate *h,
 					struct list_head *non_hvo_folios);
 void hugetlb_vmemmap_optimize_folio(const struct hstate *h, struct folio *folio);
 void hugetlb_vmemmap_optimize_folios(struct hstate *h, struct list_head *folio_list);
+void hugetlb_vmemmap_optimize_bootmem_folios(struct hstate *h, struct list_head *folio_list);
+
 
 static inline unsigned int hugetlb_vmemmap_size(const struct hstate *h)
 {
@@ -64,6 +66,19 @@ static inline void hugetlb_vmemmap_optimize_folios(struct hstate *h, struct list
 {
 }
 
+static inline void hugetlb_vmemmap_init_early(int nid)
+{
+}
+
+static inline void hugetlb_vmemmap_init_late(int nid)
+{
+}
+
+static inline void hugetlb_vmemmap_optimize_bootmem_folios(struct hstate *h,
+						struct list_head *folio_list)
+{
+}
+
 static inline unsigned int hugetlb_vmemmap_optimizable_size(const struct hstate *h)
 {
 	return 0;
-- 
2.48.1.262.g85cc9f2d1e-goog

[PATCH v2 19/28] mm/hugetlb_vmemmap: fix hugetlb_vmemmap_restore_folios definition

[Frank van der Linden]

Make the hugetlb_vmemmap_restore_folios definition inline for
the !CONFIG_HUGETLB_PAGE_OPTIMIZE_VMEMMAP case, so that including
this file in files other than hugetlb_vmemmap.c will work.

Fixes: cfb8c75099db ("hugetlb: perform vmemmap restoration on a list of pages")
Signed-off-by: Frank van der Linden <fvdl@google.com>
---
 mm/hugetlb_vmemmap.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/mm/hugetlb_vmemmap.h b/mm/hugetlb_vmemmap.h
index a6354a27e63f..926b8b27b5cb 100644
--- a/mm/hugetlb_vmemmap.h
+++ b/mm/hugetlb_vmemmap.h
@@ -50,7 +50,7 @@ static inline int hugetlb_vmemmap_restore_folio(const struct hstate *h, struct f
 	return 0;
 }
 
-static long hugetlb_vmemmap_restore_folios(const struct hstate *h,
+static inline long hugetlb_vmemmap_restore_folios(const struct hstate *h,
 					struct list_head *folio_list,
 					struct list_head *non_hvo_folios)
 {
-- 
2.48.1.262.g85cc9f2d1e-goog

[PATCH v2 20/28] mm/hugetlb: do pre-HVO for bootmem allocated pages

[Frank van der Linden]

For large systems, the overhead of vmemmap pages for hugetlb
is substantial. It's about 1.5% of memory, which is about
45G for a 3T system. If you want to configure most of that
system for hugetlb (e.g. to use as backing memory for VMs),
there is a chance of running out of memory on boot, even
though you know that the 45G will become available later.

To avoid this scenario, and since it's a waste to first
allocate and then free that 45G during boot, do pre-HVO
for hugetlb bootmem allocated pages ('gigantic' pages).

pre-HVO is done by adding functions that are called from
sparse_init_nid_early and sparse_init_nid_late. The first
is called before memmap allocation, so it takes care of
allocating memmap HVO-style. The second verifies that all
bootmem pages look good, specifically it checks that they
do not intersect with multiple zones. This can only be done
from sparse_init_nid_late path, when zones have been
initialized.

The hugetlb page size must be aligned to the section size,
and aligned to the size of memory described by the number
of page structures contained in one PMD (since pre-HVO
is not prepared to split PMDs). This should be true for
most 'gigantic' pages, it is for 1G pages on x86, where
both of these alignment requirements are 128M.

This will only have an effect if hugetlb_bootmem_alloc was
called early in boot. If not, it won't do anything, and
HVO for bootmem hugetlb pages works as before.

Signed-off-by: Frank van der Linden <fvdl@google.com>
---
 include/linux/hugetlb.h |   2 +
 mm/hugetlb.c            |   4 +-
 mm/hugetlb_vmemmap.c    | 143 ++++++++++++++++++++++++++++++++++++++++
 mm/hugetlb_vmemmap.h    |   6 ++
 mm/sparse-vmemmap.c     |   4 ++
 5 files changed, 157 insertions(+), 2 deletions(-)

diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h
index 10a7ce2b95e1..2512463bca49 100644
--- a/include/linux/hugetlb.h
+++ b/include/linux/hugetlb.h
@@ -687,6 +687,8 @@ struct huge_bootmem_page {
 #define HUGE_BOOTMEM_HVO		0x0001
 #define HUGE_BOOTMEM_ZONES_VALID	0x0002
 
+bool hugetlb_bootmem_page_zones_valid(int nid, struct huge_bootmem_page *m);
+
 int isolate_or_dissolve_huge_page(struct page *page, struct list_head *list);
 int replace_free_hugepage_folios(unsigned long start_pfn, unsigned long end_pfn);
 struct folio *alloc_hugetlb_folio(struct vm_area_struct *vma,
diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index b48f8638c9af..5af544960052 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -3311,8 +3311,8 @@ static void __init prep_and_add_bootmem_folios(struct hstate *h,
 	}
 }
 
-static bool __init hugetlb_bootmem_page_zones_valid(int nid,
-						    struct huge_bootmem_page *m)
+bool __init hugetlb_bootmem_page_zones_valid(int nid,
+					     struct huge_bootmem_page *m)
 {
 	unsigned long start_pfn;
 	bool valid;
diff --git a/mm/hugetlb_vmemmap.c b/mm/hugetlb_vmemmap.c
index be6b33ecbc8e..9a99dfa3c495 100644
--- a/mm/hugetlb_vmemmap.c
+++ b/mm/hugetlb_vmemmap.c
@@ -743,6 +743,149 @@ void hugetlb_vmemmap_optimize_bootmem_folios(struct hstate *h, struct list_head
 	__hugetlb_vmemmap_optimize_folios(h, folio_list, true);
 }
 
+#ifdef CONFIG_SPARSEMEM_VMEMMAP_PREINIT
+
+/* Return true of a bootmem allocated HugeTLB page should be pre-HVO-ed */
+static bool vmemmap_should_optimize_bootmem_page(struct huge_bootmem_page *m)
+{
+	unsigned long section_size, psize, pmd_vmemmap_size;
+	phys_addr_t paddr;
+
+	if (!READ_ONCE(vmemmap_optimize_enabled))
+		return false;
+
+	if (!hugetlb_vmemmap_optimizable(m->hstate))
+		return false;
+
+	psize = huge_page_size(m->hstate);
+	paddr = virt_to_phys(m);
+
+	/*
+	 * Pre-HVO only works if the bootmem huge page
+	 * is aligned to the section size.
+	 */
+	section_size = (1UL << PA_SECTION_SHIFT);
+	if (!IS_ALIGNED(paddr, section_size) ||
+	    !IS_ALIGNED(psize, section_size))
+		return false;
+
+	/*
+	 * The pre-HVO code does not deal with splitting PMDS,
+	 * so the bootmem page must be aligned to the number
+	 * of base pages that can be mapped with one vmemmap PMD.
+	 */
+	pmd_vmemmap_size = (PMD_SIZE / (sizeof(struct page))) << PAGE_SHIFT;
+	if (!IS_ALIGNED(paddr, pmd_vmemmap_size) ||
+	    !IS_ALIGNED(psize, pmd_vmemmap_size))
+		return false;
+
+	return true;
+}
+
+/*
+ * Initialize memmap section for a gigantic page, HVO-style.
+ */
+void __init hugetlb_vmemmap_init_early(int nid)
+{
+	unsigned long psize, paddr, section_size;
+	unsigned long ns, i, pnum, pfn, nr_pages;
+	unsigned long start, end;
+	struct huge_bootmem_page *m = NULL;
+	void *map;
+
+	/*
+	 * Noting to do if bootmem pages were not allocated
+	 * early in boot, or if HVO wasn't enabled in the
+	 * first place.
+	 */
+	if (!hugetlb_bootmem_allocated())
+		return;
+
+	if (!READ_ONCE(vmemmap_optimize_enabled))
+		return;
+
+	section_size = (1UL << PA_SECTION_SHIFT);
+
+	list_for_each_entry(m, &huge_boot_pages[nid], list) {
+		if (!vmemmap_should_optimize_bootmem_page(m))
+			continue;
+
+		nr_pages = pages_per_huge_page(m->hstate);
+		psize = nr_pages << PAGE_SHIFT;
+		paddr = virt_to_phys(m);
+		pfn = PHYS_PFN(paddr);
+		map = pfn_to_page(pfn);
+		start = (unsigned long)map;
+		end = start + nr_pages * sizeof(struct page);
+
+		if (vmemmap_populate_hvo(start, end, nid,
+					HUGETLB_VMEMMAP_RESERVE_SIZE) < 0)
+			continue;
+
+		memmap_boot_pages_add(HUGETLB_VMEMMAP_RESERVE_SIZE / PAGE_SIZE);
+
+		pnum = pfn_to_section_nr(pfn);
+		ns = psize / section_size;
+
+		for (i = 0; i < ns; i++) {
+			sparse_init_early_section(nid, map, pnum,
+					SECTION_IS_VMEMMAP_PREINIT);
+			map += section_map_size();
+			pnum++;
+		}
+
+		m->flags |= HUGE_BOOTMEM_HVO;
+	}
+}
+
+void __init hugetlb_vmemmap_init_late(int nid)
+{
+	struct huge_bootmem_page *m, *tm;
+	unsigned long phys, nr_pages, start, end;
+	unsigned long pfn, nr_mmap;
+	struct hstate *h;
+	void *map;
+
+	if (!hugetlb_bootmem_allocated())
+		return;
+
+	if (!READ_ONCE(vmemmap_optimize_enabled))
+		return;
+
+	list_for_each_entry_safe(m, tm, &huge_boot_pages[nid], list) {
+		if (!(m->flags & HUGE_BOOTMEM_HVO))
+			continue;
+
+		phys = virt_to_phys(m);
+		h = m->hstate;
+		pfn = PHYS_PFN(phys);
+		nr_pages = pages_per_huge_page(h);
+
+		if (!hugetlb_bootmem_page_zones_valid(nid, m)) {
+			/*
+			 * Oops, the hugetlb page spans multiple zones.
+			 * Remove it from the list, and undo HVO.
+			 */
+			list_del(&m->list);
+
+			map = pfn_to_page(pfn);
+
+			start = (unsigned long)map;
+			end = start + nr_pages * sizeof(struct page);
+
+			vmemmap_undo_hvo(start, end, nid,
+					 HUGETLB_VMEMMAP_RESERVE_SIZE);
+			nr_mmap = end - start - HUGETLB_VMEMMAP_RESERVE_SIZE;
+			memmap_boot_pages_add(DIV_ROUND_UP(nr_mmap, PAGE_SIZE));
+
+			memblock_phys_free(phys, huge_page_size(h));
+			continue;
+		} else
+			m->flags |= HUGE_BOOTMEM_ZONES_VALID;
+	}
+}
+#endif
+
 static const struct ctl_table hugetlb_vmemmap_sysctls[] = {
 	{
 		.procname	= "hugetlb_optimize_vmemmap",
diff --git a/mm/hugetlb_vmemmap.h b/mm/hugetlb_vmemmap.h
index 926b8b27b5cb..0031e49b12f7 100644
--- a/mm/hugetlb_vmemmap.h
+++ b/mm/hugetlb_vmemmap.h
@@ -9,6 +9,8 @@
 #ifndef _LINUX_HUGETLB_VMEMMAP_H
 #define _LINUX_HUGETLB_VMEMMAP_H
 #include <linux/hugetlb.h>
+#include <linux/io.h>
+#include <linux/memblock.h>
 
 /*
  * Reserve one vmemmap page, all vmemmap addresses are mapped to it. See
@@ -25,6 +27,10 @@ long hugetlb_vmemmap_restore_folios(const struct hstate *h,
 void hugetlb_vmemmap_optimize_folio(const struct hstate *h, struct folio *folio);
 void hugetlb_vmemmap_optimize_folios(struct hstate *h, struct list_head *folio_list);
 void hugetlb_vmemmap_optimize_bootmem_folios(struct hstate *h, struct list_head *folio_list);
+#ifdef CONFIG_SPARSEMEM_VMEMMAP_PREINIT
+void hugetlb_vmemmap_init_early(int nid);
+void hugetlb_vmemmap_init_late(int nid);
+#endif
 
 
 static inline unsigned int hugetlb_vmemmap_size(const struct hstate *h)
diff --git a/mm/sparse-vmemmap.c b/mm/sparse-vmemmap.c
index bee22ca93654..29647fd3d606 100644
--- a/mm/sparse-vmemmap.c
+++ b/mm/sparse-vmemmap.c
@@ -32,6 +32,8 @@
 #include <asm/pgalloc.h>
 #include <asm/tlbflush.h>
 
+#include "hugetlb_vmemmap.h"
+
 /*
  * Flags for vmemmap_populate_range and friends.
  */
@@ -594,6 +596,7 @@ struct page * __meminit __populate_section_memmap(unsigned long pfn,
  */
 void __init sparse_vmemmap_init_nid_early(int nid)
 {
+	hugetlb_vmemmap_init_early(nid);
 }
 
 /*
@@ -604,5 +607,6 @@ void __init sparse_vmemmap_init_nid_early(int nid)
  */
 void __init sparse_vmemmap_init_nid_late(int nid)
 {
+	hugetlb_vmemmap_init_late(nid);
 }
 #endif
-- 
2.48.1.262.g85cc9f2d1e-goog

[PATCH v2 21/28] x86/setup: call hugetlb_bootmem_alloc early

[Frank van der Linden]

Call hugetlb_bootmem_allloc in an earlier spot in setup, after
hugelb_cma_reserve. This will make vmemmap preinit of
the sections covered by the allocated hugetlb pages possible.

Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Frank van der Linden <fvdl@google.com>
---
 arch/x86/kernel/setup.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/arch/x86/kernel/setup.c b/arch/x86/kernel/setup.c
index cebee310e200..ff8604007b08 100644
--- a/arch/x86/kernel/setup.c
+++ b/arch/x86/kernel/setup.c
@@ -1108,8 +1108,10 @@ void __init setup_arch(char **cmdline_p)
 	initmem_init();
 	dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT);
 
-	if (boot_cpu_has(X86_FEATURE_GBPAGES))
+	if (boot_cpu_has(X86_FEATURE_GBPAGES)) {
 		hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
+		hugetlb_bootmem_alloc();
+	}
 
 	/*
 	 * Reserve memory for crash kernel after SRAT is parsed so that it
-- 
2.48.1.262.g85cc9f2d1e-goog

[PATCH v2 22/28] x86/mm: set ARCH_WANT_SPARSEMEM_VMEMMAP_PREINIT

[Frank van der Linden]

Now that hugetlb bootmem pages are allocated earlier,
and available for section preinit (HVO-style), set
ARCH_WANT_SPARSEMEM_VMEMMAP_PREINIT for x86_64, so
that is can be done.

This enables pre-HVO on x86_64.

Signed-off-by: Frank van der Linden <fvdl@google.com>
---
 arch/x86/Kconfig | 1 +
 1 file changed, 1 insertion(+)

diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index 87198d957e2f..ccef99c0a2ba 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -146,6 +146,7 @@ config X86
 	select ARCH_WANT_LD_ORPHAN_WARN
 	select ARCH_WANT_OPTIMIZE_DAX_VMEMMAP	if X86_64
 	select ARCH_WANT_OPTIMIZE_HUGETLB_VMEMMAP	if X86_64
+	select ARCH_WANT_SPARSEMEM_VMEMMAP_PREINIT if X86_64
 	select ARCH_WANTS_THP_SWAP		if X86_64
 	select ARCH_HAS_PARANOID_L1D_FLUSH
 	select BUILDTIME_TABLE_SORT
-- 
2.48.1.262.g85cc9f2d1e-goog

[PATCH v2 23/28] mm/cma: simplify zone intersection check

[Frank van der Linden]

cma_activate_area walks all pages in the area, checking
their zone individually to see if the area resides in
more than one zone.

Make this a little more efficient by using the recently
introduced pfn_range_intersects_zones() function. Store
the NUMA node id (if any) in the cma structure to facilitate
this.

Signed-off-by: Frank van der Linden <fvdl@google.com>
---
 mm/cma.c | 13 ++++++-------
 mm/cma.h |  2 ++
 2 files changed, 8 insertions(+), 7 deletions(-)

diff --git a/mm/cma.c b/mm/cma.c
index 1704d5be6a07..6ad631c9fdca 100644
--- a/mm/cma.c
+++ b/mm/cma.c
@@ -103,7 +103,6 @@ static void __init cma_activate_area(struct cma *cma)
 {
 	unsigned long pfn, base_pfn;
 	int allocrange, r;
-	struct zone *zone;
 	struct cma_memrange *cmr;
 
 	for (allocrange = 0; allocrange < cma->nranges; allocrange++) {
@@ -124,12 +123,8 @@ static void __init cma_activate_area(struct cma *cma)
 		 * CMA resv range to be in the same zone.
 		 */
 		WARN_ON_ONCE(!pfn_valid(base_pfn));
-		zone = page_zone(pfn_to_page(base_pfn));
-		for (pfn = base_pfn + 1; pfn < base_pfn + cmr->count; pfn++) {
-			WARN_ON_ONCE(!pfn_valid(pfn));
-			if (page_zone(pfn_to_page(pfn)) != zone)
-				goto cleanup;
-		}
+		if (pfn_range_intersects_zones(cma->nid, base_pfn, cmr->count))
+			goto cleanup;
 
 		for (pfn = base_pfn; pfn < base_pfn + cmr->count;
 		     pfn += pageblock_nr_pages)
@@ -261,6 +256,7 @@ int __init cma_init_reserved_mem(phys_addr_t base, phys_addr_t size,
 	cma->ranges[0].base_pfn = PFN_DOWN(base);
 	cma->ranges[0].count = cma->count;
 	cma->nranges = 1;
+	cma->nid = NUMA_NO_NODE;
 
 	*res_cma = cma;
 
@@ -497,6 +493,7 @@ int __init cma_declare_contiguous_multi(phys_addr_t total_size,
 	}
 
 	cma->nranges = nr;
+	cma->nid = nid;
 	*res_cma = cma;
 
 out:
@@ -684,6 +681,8 @@ static int __init __cma_declare_contiguous_nid(phys_addr_t base,
 	if (ret)
 		memblock_phys_free(base, size);
 
+	(*res_cma)->nid = nid;
+
 	return ret;
 }
 
diff --git a/mm/cma.h b/mm/cma.h
index 5f39dd1aac91..ff79dba5508c 100644
--- a/mm/cma.h
+++ b/mm/cma.h
@@ -50,6 +50,8 @@ struct cma {
 	struct cma_kobject *cma_kobj;
 #endif
 	bool reserve_pages_on_error;
+	/* NUMA node (NUMA_NO_NODE if unspecified) */
+	int nid;
 };
 
 extern struct cma cma_areas[MAX_CMA_AREAS];
-- 
2.48.1.262.g85cc9f2d1e-goog

[PATCH v2 24/28] mm/cma: introduce a cma validate function

[Frank van der Linden]

Define a function to check if a CMA area is valid, which means:
do its ranges not cross any zone boundaries. Store the result
in the newly created flags for each CMA area, so that multiple
calls are dealt with.

This allows for checking the validity of a CMA area early, which
is needed later in order to be able to allocate hugetlb bootmem
pages from it with pre-HVO.

Signed-off-by: Frank van der Linden <fvdl@google.com>
---
 include/linux/cma.h |  5 ++++
 mm/cma.c            | 60 ++++++++++++++++++++++++++++++++++++---------
 mm/cma.h            |  8 +++++-
 3 files changed, 60 insertions(+), 13 deletions(-)

diff --git a/include/linux/cma.h b/include/linux/cma.h
index 03d85c100dcc..62d9c1cf6326 100644
--- a/include/linux/cma.h
+++ b/include/linux/cma.h
@@ -60,6 +60,7 @@ extern void cma_reserve_pages_on_error(struct cma *cma);
 #ifdef CONFIG_CMA
 struct folio *cma_alloc_folio(struct cma *cma, int order, gfp_t gfp);
 bool cma_free_folio(struct cma *cma, const struct folio *folio);
+bool cma_validate_zones(struct cma *cma);
 #else
 static inline struct folio *cma_alloc_folio(struct cma *cma, int order, gfp_t gfp)
 {
@@ -70,6 +71,10 @@ static inline bool cma_free_folio(struct cma *cma, const struct folio *folio)
 {
 	return false;
 }
+static inline bool cma_validate_zones(struct cma *cma)
+{
+	return false;
+}
 #endif
 
 #endif
diff --git a/mm/cma.c b/mm/cma.c
index 6ad631c9fdca..41248dee7197 100644
--- a/mm/cma.c
+++ b/mm/cma.c
@@ -99,6 +99,49 @@ static void cma_clear_bitmap(struct cma *cma, const struct cma_memrange *cmr,
 	spin_unlock_irqrestore(&cma->lock, flags);
 }
 
+/*
+ * Check if a CMA area contains no ranges that intersect with
+ * multiple zones. Store the result in the flags in case
+ * this gets called more than once.
+ */
+bool cma_validate_zones(struct cma *cma)
+{
+	int r;
+	unsigned long base_pfn;
+	struct cma_memrange *cmr;
+	bool valid_bit_set;
+
+	/*
+	 * If already validated, return result of previous check.
+	 * Either the valid or invalid bit will be set if this
+	 * check has already been done. If neither is set, the
+	 * check has not been performed yet.
+	 */
+	valid_bit_set = test_bit(CMA_ZONES_VALID, &cma->flags);
+	if (valid_bit_set || test_bit(CMA_ZONES_INVALID, &cma->flags))
+		return valid_bit_set;
+
+	for (r = 0; r < cma->nranges; r++) {
+		cmr = &cma->ranges[r];
+		base_pfn = cmr->base_pfn;
+
+		/*
+		 * alloc_contig_range() requires the pfn range specified
+		 * to be in the same zone. Simplify by forcing the entire
+		 * CMA resv range to be in the same zone.
+		 */
+		WARN_ON_ONCE(!pfn_valid(base_pfn));
+		if (pfn_range_intersects_zones(cma->nid, base_pfn, cmr->count)) {
+			set_bit(CMA_ZONES_INVALID, &cma->flags);
+			return false;
+		}
+	}
+
+	set_bit(CMA_ZONES_VALID, &cma->flags);
+
+	return true;
+}
+
 static void __init cma_activate_area(struct cma *cma)
 {
 	unsigned long pfn, base_pfn;
@@ -113,19 +156,12 @@ static void __init cma_activate_area(struct cma *cma)
 			goto cleanup;
 	}
 
+	if (!cma_validate_zones(cma))
+		goto cleanup;
+
 	for (r = 0; r < cma->nranges; r++) {
 		cmr = &cma->ranges[r];
 		base_pfn = cmr->base_pfn;
-
-		/*
-		 * alloc_contig_range() requires the pfn range specified
-		 * to be in the same zone. Simplify by forcing the entire
-		 * CMA resv range to be in the same zone.
-		 */
-		WARN_ON_ONCE(!pfn_valid(base_pfn));
-		if (pfn_range_intersects_zones(cma->nid, base_pfn, cmr->count))
-			goto cleanup;
-
 		for (pfn = base_pfn; pfn < base_pfn + cmr->count;
 		     pfn += pageblock_nr_pages)
 			init_cma_reserved_pageblock(pfn_to_page(pfn));
@@ -145,7 +181,7 @@ static void __init cma_activate_area(struct cma *cma)
 		bitmap_free(cma->ranges[r].bitmap);
 
 	/* Expose all pages to the buddy, they are useless for CMA. */
-	if (!cma->reserve_pages_on_error) {
+	if (!test_bit(CMA_RESERVE_PAGES_ON_ERROR, &cma->flags)) {
 		for (r = 0; r < allocrange; r++) {
 			cmr = &cma->ranges[r];
 			for (pfn = cmr->base_pfn;
@@ -172,7 +208,7 @@ core_initcall(cma_init_reserved_areas);
 
 void __init cma_reserve_pages_on_error(struct cma *cma)
 {
-	cma->reserve_pages_on_error = true;
+	set_bit(CMA_RESERVE_PAGES_ON_ERROR, &cma->flags);
 }
 
 static int __init cma_new_area(const char *name, phys_addr_t size,
diff --git a/mm/cma.h b/mm/cma.h
index ff79dba5508c..bddc84b3cd96 100644
--- a/mm/cma.h
+++ b/mm/cma.h
@@ -49,11 +49,17 @@ struct cma {
 	/* kobject requires dynamic object */
 	struct cma_kobject *cma_kobj;
 #endif
-	bool reserve_pages_on_error;
+	unsigned long flags;
 	/* NUMA node (NUMA_NO_NODE if unspecified) */
 	int nid;
 };
 
+enum cma_flags {
+	CMA_RESERVE_PAGES_ON_ERROR,
+	CMA_ZONES_VALID,
+	CMA_ZONES_INVALID,
+};
+
 extern struct cma cma_areas[MAX_CMA_AREAS];
 extern unsigned int cma_area_count;
 
-- 
2.48.1.262.g85cc9f2d1e-goog

[PATCH v2 25/28] mm/cma: introduce interface for early reservations

[Frank van der Linden]

It can be desirable to reserve memory in a CMA area before
it is activated, early in boot. Such reservations would
effectively be memblock allocations, but they can be
returned to the CMA area later. This functionality can
be used to allow hugetlb bootmem allocations from a
hugetlb CMA area.

A new interface, cma_reserve_early is introduced. This allows
for pageblock-aligned reservations. These reservations are
skipped during the initial handoff of pages in a CMA area
to the buddy allocator. The caller is responsible for making
sure that the page structures are set up, and that the migrate
type is set correctly, as with other memblock allocations that
stick around. If the CMA area fails to activate (because it
intersects with multiple zones), the reserved memory is not
given to the buddy allocator, the caller needs to take care
of that.

Signed-off-by: Frank van der Linden <fvdl@google.com>
---
 mm/cma.c      | 83 ++++++++++++++++++++++++++++++++++++++++++++++-----
 mm/cma.h      |  8 +++++
 mm/internal.h | 16 ++++++++++
 mm/mm_init.c  |  9 ++++++
 4 files changed, 109 insertions(+), 7 deletions(-)

diff --git a/mm/cma.c b/mm/cma.c
index 41248dee7197..2b1e264e4e99 100644
--- a/mm/cma.c
+++ b/mm/cma.c
@@ -144,9 +144,10 @@ bool cma_validate_zones(struct cma *cma)
 
 static void __init cma_activate_area(struct cma *cma)
 {
-	unsigned long pfn, base_pfn;
+	unsigned long pfn, end_pfn;
 	int allocrange, r;
 	struct cma_memrange *cmr;
+	unsigned long bitmap_count, count;
 
 	for (allocrange = 0; allocrange < cma->nranges; allocrange++) {
 		cmr = &cma->ranges[allocrange];
@@ -161,8 +162,13 @@ static void __init cma_activate_area(struct cma *cma)
 
 	for (r = 0; r < cma->nranges; r++) {
 		cmr = &cma->ranges[r];
-		base_pfn = cmr->base_pfn;
-		for (pfn = base_pfn; pfn < base_pfn + cmr->count;
+		if (cmr->early_pfn != cmr->base_pfn) {
+			count = cmr->early_pfn - cmr->base_pfn;
+			bitmap_count = cma_bitmap_pages_to_bits(cma, count);
+			bitmap_set(cmr->bitmap, 0, bitmap_count);
+		}
+
+		for (pfn = cmr->early_pfn; pfn < cmr->base_pfn + cmr->count;
 		     pfn += pageblock_nr_pages)
 			init_cma_reserved_pageblock(pfn_to_page(pfn));
 	}
@@ -173,6 +179,7 @@ static void __init cma_activate_area(struct cma *cma)
 	INIT_HLIST_HEAD(&cma->mem_head);
 	spin_lock_init(&cma->mem_head_lock);
 #endif
+	set_bit(CMA_ACTIVATED, &cma->flags);
 
 	return;
 
@@ -184,9 +191,8 @@ static void __init cma_activate_area(struct cma *cma)
 	if (!test_bit(CMA_RESERVE_PAGES_ON_ERROR, &cma->flags)) {
 		for (r = 0; r < allocrange; r++) {
 			cmr = &cma->ranges[r];
-			for (pfn = cmr->base_pfn;
-			     pfn < cmr->base_pfn + cmr->count;
-			     pfn++)
+			end_pfn = cmr->base_pfn + cmr->count;
+			for (pfn = cmr->early_pfn; pfn < end_pfn; pfn++)
 				free_reserved_page(pfn_to_page(pfn));
 		}
 	}
@@ -290,6 +296,7 @@ int __init cma_init_reserved_mem(phys_addr_t base, phys_addr_t size,
 		return ret;
 
 	cma->ranges[0].base_pfn = PFN_DOWN(base);
+	cma->ranges[0].early_pfn = PFN_DOWN(base);
 	cma->ranges[0].count = cma->count;
 	cma->nranges = 1;
 	cma->nid = NUMA_NO_NODE;
@@ -509,6 +516,7 @@ int __init cma_declare_contiguous_multi(phys_addr_t total_size,
 		    nr, (u64)mlp->base, (u64)mlp->base + size);
 		cmrp = &cma->ranges[nr++];
 		cmrp->base_pfn = PHYS_PFN(mlp->base);
+		cmrp->early_pfn = cmrp->base_pfn;
 		cmrp->count = size >> PAGE_SHIFT;
 
 		sizeleft -= size;
@@ -540,7 +548,6 @@ int __init cma_declare_contiguous_multi(phys_addr_t total_size,
 		pr_info("Reserved %lu MiB in %d range%s\n",
 			(unsigned long)total_size / SZ_1M, nr,
 			nr > 1 ? "s" : "");
-
 	return ret;
 }
 
@@ -1044,3 +1051,65 @@ bool cma_intersects(struct cma *cma, unsigned long start, unsigned long end)
 
 	return false;
 }
+
+/*
+ * Very basic function to reserve memory from a CMA area that has not
+ * yet been activated. This is expected to be called early, when the
+ * system is single-threaded, so there is no locking. The alignment
+ * checking is restrictive - only pageblock-aligned areas
+ * (CMA_MIN_ALIGNMENT_BYTES) may be reserved through this function.
+ * This keeps things simple, and is enough for the current use case.
+ *
+ * The CMA bitmaps have not yet been allocated, so just start
+ * reserving from the bottom up, using a PFN to keep track
+ * of what has been reserved. Unreserving is not possible.
+ *
+ * The caller is responsible for initializing the page structures
+ * in the area properly, since this just points to memblock-allocated
+ * memory. The caller should subsequently use init_cma_pageblock to
+ * set the migrate type and CMA stats  the pageblocks that were reserved.
+ *
+ * If the CMA area fails to activate later, memory obtained through
+ * this interface is not handed to the page allocator, this is
+ * the responsibility of the caller (e.g. like normal memblock-allocated
+ * memory).
+ */
+void __init *cma_reserve_early(struct cma *cma, unsigned long size)
+{
+	int r;
+	struct cma_memrange *cmr;
+	unsigned long available;
+	void *ret = NULL;
+
+	if (!cma || !cma->count)
+		return NULL;
+	/*
+	 * Can only be called early in init.
+	 */
+	if (test_bit(CMA_ACTIVATED, &cma->flags))
+		return NULL;
+
+	if (!IS_ALIGNED(size, CMA_MIN_ALIGNMENT_BYTES))
+		return NULL;
+
+	if (!IS_ALIGNED(size, (PAGE_SIZE << cma->order_per_bit)))
+		return NULL;
+
+	size >>= PAGE_SHIFT;
+
+	if (size > cma->available_count)
+		return NULL;
+
+	for (r = 0; r < cma->nranges; r++) {
+		cmr = &cma->ranges[r];
+		available = cmr->count - (cmr->early_pfn - cmr->base_pfn);
+		if (size <= available) {
+			ret = phys_to_virt(PFN_PHYS(cmr->early_pfn));
+			cmr->early_pfn += size;
+			cma->available_count -= size;
+			return ret;
+		}
+	}
+
+	return ret;
+}
diff --git a/mm/cma.h b/mm/cma.h
index bddc84b3cd96..df7fc623b7a6 100644
--- a/mm/cma.h
+++ b/mm/cma.h
@@ -16,9 +16,16 @@ struct cma_kobject {
  * and the total amount of memory requested, while smaller than the total
  * amount of memory available, is large enough that it doesn't fit in a
  * single physical memory range because of memory holes.
+ *
+ * Fields:
+ *   @base_pfn: physical address of range
+ *   @early_pfn: first PFN not reserved through cma_reserve_early
+ *   @count: size of range
+ *   @bitmap: bitmap of allocated (1 << order_per_bit)-sized chunks.
  */
 struct cma_memrange {
 	unsigned long base_pfn;
+	unsigned long early_pfn;
 	unsigned long count;
 	unsigned long *bitmap;
 #ifdef CONFIG_CMA_DEBUGFS
@@ -58,6 +65,7 @@ enum cma_flags {
 	CMA_RESERVE_PAGES_ON_ERROR,
 	CMA_ZONES_VALID,
 	CMA_ZONES_INVALID,
+	CMA_ACTIVATED,
 };
 
 extern struct cma cma_areas[MAX_CMA_AREAS];
diff --git a/mm/internal.h b/mm/internal.h
index 63fda9bb9426..8318c8e6e589 100644
--- a/mm/internal.h
+++ b/mm/internal.h
@@ -848,6 +848,22 @@ void init_cma_reserved_pageblock(struct page *page);
 
 #endif /* CONFIG_COMPACTION || CONFIG_CMA */
 
+struct cma;
+
+#ifdef CONFIG_CMA
+void *cma_reserve_early(struct cma *cma, unsigned long size);
+void init_cma_pageblock(struct page *page);
+#else
+static inline void *cma_reserve_early(struct cma *cma, unsigned long size)
+{
+	return NULL;
+}
+static inline void init_cma_pageblock(struct page *page)
+{
+}
+#endif
+
+
 int find_suitable_fallback(struct free_area *area, unsigned int order,
 			int migratetype, bool only_stealable, bool *can_steal);
 
diff --git a/mm/mm_init.c b/mm/mm_init.c
index f7d5b4fe1ae9..f31260fd393e 100644
--- a/mm/mm_init.c
+++ b/mm/mm_init.c
@@ -2263,6 +2263,15 @@ void __init init_cma_reserved_pageblock(struct page *page)
 	adjust_managed_page_count(page, pageblock_nr_pages);
 	page_zone(page)->cma_pages += pageblock_nr_pages;
 }
+/*
+ * Similar to above, but only set the migrate type and stats.
+ */
+void __init init_cma_pageblock(struct page *page)
+{
+	set_pageblock_migratetype(page, MIGRATE_CMA);
+	adjust_managed_page_count(page, pageblock_nr_pages);
+	page_zone(page)->cma_pages += pageblock_nr_pages;
+}
 #endif
 
 void set_zone_contiguous(struct zone *zone)
-- 
2.48.1.262.g85cc9f2d1e-goog

[PATCH v2 26/28] mm/hugetlb: add hugetlb_cma_only cmdline option

[Frank van der Linden]

Add an option to force hugetlb gigantic pages to be allocated using
CMA only (if hugetlb_cma is enabled). This avoids a fallback to
allocation from the rest of system memory if the CMA allocation
fails. This makes the size of hugetlb_cma a hard upper boundary
for gigantic hugetlb page allocations.

This is useful because, with a large CMA area, the kernel's
unmovable allocations will have less room to work with and it
is undesirable for new hugetlb gigantic page allocations
to be done from that remaining area. It will eat in to the space
available for unmovable allocations, leading to unwanted system
behavior (OOMs because the kernel fails to do unmovable allocations).

So, with this enabled, an administrator can force a hard upper
bound for runtime gigantic page allocations, and have more
predictable system behavior.

Signed-off-by: Frank van der Linden <fvdl@google.com>
---
 Documentation/admin-guide/kernel-parameters.txt |  7 +++++++
 mm/hugetlb.c                                    | 14 ++++++++++++++
 2 files changed, 21 insertions(+)

diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
index fb8752b42ec8..eb56b251ce10 100644
--- a/Documentation/admin-guide/kernel-parameters.txt
+++ b/Documentation/admin-guide/kernel-parameters.txt
@@ -1892,6 +1892,13 @@
 			hugepages using the CMA allocator. If enabled, the
 			boot-time allocation of gigantic hugepages is skipped.
 
+	hugetlb_cma_only=
+			[HW,CMA,EARLY] When allocating new HugeTLB pages, only
+			try to allocate from the CMA areas.
+
+			This option does nothing if hugetlb_cma= is not also
+			specified.
+
 	hugetlb_free_vmemmap=
 			[KNL] Requires CONFIG_HUGETLB_PAGE_OPTIMIZE_VMEMMAP
 			enabled.
diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index 5af544960052..c227d0b9cf1e 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -60,6 +60,7 @@ struct hstate hstates[HUGE_MAX_HSTATE];
 static struct cma *hugetlb_cma[MAX_NUMNODES];
 static unsigned long hugetlb_cma_size_in_node[MAX_NUMNODES] __initdata;
 #endif
+static bool hugetlb_cma_only;
 static unsigned long hugetlb_cma_size __initdata;
 
 __initdata struct list_head huge_boot_pages[MAX_NUMNODES];
@@ -1511,6 +1512,9 @@ static struct folio *alloc_gigantic_folio(struct hstate *h, gfp_t gfp_mask,
 	}
 #endif
 	if (!folio) {
+		if (hugetlb_cma_only)
+			return NULL;
+
 		folio = folio_alloc_gigantic(order, gfp_mask, nid, nodemask);
 		if (!folio)
 			return NULL;
@@ -4732,6 +4736,9 @@ static __init void hugetlb_parse_params(void)
 
 		hcp->setup(hcp->val);
 	}
+
+	if (!hugetlb_cma_size)
+		hugetlb_cma_only = false;
 }
 
 /*
@@ -7844,6 +7851,13 @@ static int __init cmdline_parse_hugetlb_cma(char *p)
 
 early_param("hugetlb_cma", cmdline_parse_hugetlb_cma);
 
+static int __init cmdline_parse_hugetlb_cma_only(char *p)
+{
+	return kstrtobool(p, &hugetlb_cma_only);
+}
+
+early_param("hugetlb_cma_only", cmdline_parse_hugetlb_cma_only);
+
 void __init hugetlb_cma_reserve(int order)
 {
 	unsigned long size, reserved, per_node;
-- 
2.48.1.262.g85cc9f2d1e-goog

[PATCH v2 27/28] mm/hugetlb: enable bootmem allocation from CMA areas

[Frank van der Linden]

If hugetlb_cma_only is enabled, we know that hugetlb pages
can only be allocated from CMA. Now that there is an interface
to do early reservations from a CMA area (returning memblock
memory), it can be used to allocate hugetlb pages from CMA.

This also allows for doing pre-HVO on these pages (if enabled).

Make sure to initialize the page structures and associated data
correctly. Create a flag to signal that a hugetlb page has been
allocated from CMA to make things a little easier.

Some configurations of powerpc have a special hugetlb bootmem
allocator, so introduce a boolean arch_specific_huge_bootmem_alloc
that returns true if such an allocator is present. In that case,
CMA bootmem allocations can't be used, so check that function
before trying.

Cc: Madhavan Srinivasan <maddy@linux.ibm.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: linuxppc-dev@lists.ozlabs.org
Signed-off-by: Frank van der Linden <fvdl@google.com>
---
 arch/powerpc/include/asm/book3s/64/hugetlb.h |   6 +
 include/linux/hugetlb.h                      |  17 +++
 mm/hugetlb.c                                 | 121 ++++++++++++++-----
 3 files changed, 113 insertions(+), 31 deletions(-)

diff --git a/arch/powerpc/include/asm/book3s/64/hugetlb.h b/arch/powerpc/include/asm/book3s/64/hugetlb.h
index f0bba9c5f9c3..bb786694dd26 100644
--- a/arch/powerpc/include/asm/book3s/64/hugetlb.h
+++ b/arch/powerpc/include/asm/book3s/64/hugetlb.h
@@ -94,4 +94,10 @@ static inline int check_and_get_huge_psize(int shift)
 	return mmu_psize;
 }
 
+#define arch_has_huge_bootmem_alloc arch_has_huge_bootmem_alloc
+
+static inline bool arch_has_huge_bootmem_alloc(void)
+{
+	return (firmware_has_feature(FW_FEATURE_LPAR) && !radix_enabled());
+}
 #endif
diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h
index 2512463bca49..6c6546b54934 100644
--- a/include/linux/hugetlb.h
+++ b/include/linux/hugetlb.h
@@ -591,6 +591,7 @@ enum hugetlb_page_flags {
 	HPG_freed,
 	HPG_vmemmap_optimized,
 	HPG_raw_hwp_unreliable,
+	HPG_cma,
 	__NR_HPAGEFLAGS,
 };
 
@@ -650,6 +651,7 @@ HPAGEFLAG(Temporary, temporary)
 HPAGEFLAG(Freed, freed)
 HPAGEFLAG(VmemmapOptimized, vmemmap_optimized)
 HPAGEFLAG(RawHwpUnreliable, raw_hwp_unreliable)
+HPAGEFLAG(Cma, cma)
 
 #ifdef CONFIG_HUGETLB_PAGE
 
@@ -678,14 +680,18 @@ struct hstate {
 	char name[HSTATE_NAME_LEN];
 };
 
+struct cma;
+
 struct huge_bootmem_page {
 	struct list_head list;
 	struct hstate *hstate;
 	unsigned long flags;
+	struct cma *cma;
 };
 
 #define HUGE_BOOTMEM_HVO		0x0001
 #define HUGE_BOOTMEM_ZONES_VALID	0x0002
+#define HUGE_BOOTMEM_CMA		0x0004
 
 bool hugetlb_bootmem_page_zones_valid(int nid, struct huge_bootmem_page *m);
 
@@ -823,6 +829,17 @@ static inline pte_t arch_make_huge_pte(pte_t entry, unsigned int shift,
 }
 #endif
 
+#ifndef arch_has_huge_bootmem_alloc
+/*
+ * Some architectures do their own bootmem allocation, so they can't use
+ * early CMA allocation.
+ */
+static inline bool arch_has_huge_bootmem_alloc(void)
+{
+	return false;
+}
+#endif
+
 static inline struct hstate *folio_hstate(struct folio *folio)
 {
 	VM_BUG_ON_FOLIO(!folio_test_hugetlb(folio), folio);
diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index c227d0b9cf1e..5a3e9f7deaba 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -132,8 +132,10 @@ static void hugetlb_free_folio(struct folio *folio)
 #ifdef CONFIG_CMA
 	int nid = folio_nid(folio);
 
-	if (cma_free_folio(hugetlb_cma[nid], folio))
+	if (folio_test_hugetlb_cma(folio)) {
+		WARN_ON_ONCE(!cma_free_folio(hugetlb_cma[nid], folio));
 		return;
+	}
 #endif
 	folio_put(folio);
 }
@@ -1509,6 +1511,9 @@ static struct folio *alloc_gigantic_folio(struct hstate *h, gfp_t gfp_mask,
 					break;
 			}
 		}
+
+		if (folio)
+			folio_set_hugetlb_cma(folio);
 	}
 #endif
 	if (!folio) {
@@ -3175,6 +3180,53 @@ struct folio *alloc_hugetlb_folio(struct vm_area_struct *vma,
 	return ERR_PTR(-ENOSPC);
 }
 
+static bool __init hugetlb_early_cma(struct hstate *h)
+{
+	if (arch_has_huge_bootmem_alloc())
+		return false;
+
+	return (hstate_is_gigantic(h) && hugetlb_cma_only);
+}
+
+static __init void *alloc_bootmem(struct hstate *h, int nid)
+{
+	struct huge_bootmem_page *m;
+	unsigned long flags;
+	struct cma *cma;
+
+#ifdef CONFIG_CMA
+	if (hugetlb_early_cma(h)) {
+		flags = HUGE_BOOTMEM_CMA;
+		cma = hugetlb_cma[nid];
+		m = cma_reserve_early(cma, huge_page_size(h));
+	} else
+#endif
+	{
+		flags = 0;
+		cma = NULL;
+		m = memblock_alloc_try_nid_raw(huge_page_size(h),
+			huge_page_size(h), 0, MEMBLOCK_ALLOC_ACCESSIBLE, nid);
+	}
+
+	if (m) {
+		/*
+		 * Use the beginning of the huge page to store the
+		 * huge_bootmem_page struct (until gather_bootmem
+		 * puts them into the mem_map).
+		 *
+		 * Put them into a private list first because mem_map
+		 * is not up yet.
+		 */
+		INIT_LIST_HEAD(&m->list);
+		list_add(&m->list, &huge_boot_pages[nid]);
+		m->hstate = h;
+		m->flags = flags;
+		m->cma = cma;
+	}
+
+	return m;
+}
+
 int alloc_bootmem_huge_page(struct hstate *h, int nid)
 	__attribute__ ((weak, alias("__alloc_bootmem_huge_page")));
 int __alloc_bootmem_huge_page(struct hstate *h, int nid)
@@ -3184,17 +3236,14 @@ int __alloc_bootmem_huge_page(struct hstate *h, int nid)
 
 	/* do node specific alloc */
 	if (nid != NUMA_NO_NODE) {
-		m = memblock_alloc_try_nid_raw(huge_page_size(h), huge_page_size(h),
-				0, MEMBLOCK_ALLOC_ACCESSIBLE, nid);
+		m = alloc_bootmem(h, node);
 		if (!m)
 			return 0;
 		goto found;
 	}
 	/* allocate from next node when distributing huge pages */
 	for_each_node_mask_to_alloc(&h->next_nid_to_alloc, nr_nodes, node, &node_states[N_ONLINE]) {
-		m = memblock_alloc_try_nid_raw(
-				huge_page_size(h), huge_page_size(h),
-				0, MEMBLOCK_ALLOC_ACCESSIBLE, node);
+		m = alloc_bootmem(h, node);
 		if (m)
 			break;
 	}
@@ -3203,7 +3252,6 @@ int __alloc_bootmem_huge_page(struct hstate *h, int nid)
 		return 0;
 
 found:
-
 	/*
 	 * Only initialize the head struct page in memmap_init_reserved_pages,
 	 * rest of the struct pages will be initialized by the HugeTLB
@@ -3213,18 +3261,6 @@ int __alloc_bootmem_huge_page(struct hstate *h, int nid)
 	 */
 	memblock_reserved_mark_noinit(virt_to_phys((void *)m + PAGE_SIZE),
 		huge_page_size(h) - PAGE_SIZE);
-	/*
-	 * Use the beginning of the huge page to store the
-	 * huge_bootmem_page struct (until gather_bootmem
-	 * puts them into the mem_map).
-	 *
-	 * Put them into a private list first because mem_map
-	 * is not up yet.
-	 */
-	INIT_LIST_HEAD(&m->list);
-	list_add(&m->list, &huge_boot_pages[node]);
-	m->hstate = h;
-	m->flags = 0;
 	return 1;
 }
 
@@ -3265,13 +3301,25 @@ static void __init hugetlb_folio_init_vmemmap(struct folio *folio,
 	prep_compound_head((struct page *)folio, huge_page_order(h));
 }
 
+static bool __init hugetlb_bootmem_page_prehvo(struct huge_bootmem_page *m)
+{
+	return m->flags & HUGE_BOOTMEM_HVO;
+}
+
+static bool __init hugetlb_bootmem_page_earlycma(struct huge_bootmem_page *m)
+{
+	return m->flags & HUGE_BOOTMEM_CMA;
+}
+
 /*
  * memblock-allocated pageblocks might not have the migrate type set
  * if marked with the 'noinit' flag. Set it to the default (MIGRATE_MOVABLE)
- * here.
+ * here, or MIGRATE_CMA if this was a page allocated through an early CMA
+ * reservation.
  *
- * Note that this will not write the page struct, it is ok (and necessary)
- * to do this on vmemmap optimized folios.
+ * In case of vmemmap optimized folios, the tail vmemmap pages are mapped
+ * read-only, but that's ok - for sparse vmemmap this does not write to
+ * the page structure.
  */
 static void __init hugetlb_bootmem_init_migratetype(struct folio *folio,
 							  struct hstate *h)
@@ -3280,9 +3328,13 @@ static void __init hugetlb_bootmem_init_migratetype(struct folio *folio,
 
 	WARN_ON_ONCE(!pageblock_aligned(folio_pfn(folio)));
 
-	for (i = 0; i < nr_pages; i += pageblock_nr_pages)
-		set_pageblock_migratetype(folio_page(folio, i),
+	for (i = 0; i < nr_pages; i += pageblock_nr_pages) {
+		if (folio_test_hugetlb_cma(folio))
+			init_cma_pageblock(folio_page(folio, i));
+		else
+			set_pageblock_migratetype(folio_page(folio, i),
 					  MIGRATE_MOVABLE);
+	}
 }
 
 static void __init prep_and_add_bootmem_folios(struct hstate *h,
@@ -3328,10 +3380,16 @@ bool __init hugetlb_bootmem_page_zones_valid(int nid,
 		return true;
 	}
 
+	if (hugetlb_bootmem_page_earlycma(m)) {
+		valid = cma_validate_zones(m->cma);
+		goto out;
+	}
+
 	start_pfn = virt_to_phys(m) >> PAGE_SHIFT;
 
 	valid = !pfn_range_intersects_zones(nid, start_pfn,
 			pages_per_huge_page(m->hstate));
+out:
 	if (!valid)
 		hstate_boot_nrinvalid[hstate_index(m->hstate)]++;
 
@@ -3360,11 +3418,6 @@ static void __init hugetlb_bootmem_free_invalid_page(int nid, struct page *page,
 	}
 }
 
-static bool __init hugetlb_bootmem_page_prehvo(struct huge_bootmem_page *m)
-{
-	return (m->flags & HUGE_BOOTMEM_HVO);
-}
-
 /*
  * Put bootmem huge pages into the standard lists after mem_map is up.
  * Note: This only applies to gigantic (order > MAX_PAGE_ORDER) pages.
@@ -3414,6 +3467,9 @@ static void __init gather_bootmem_prealloc_node(unsigned long nid)
 			 */
 			folio_set_hugetlb_vmemmap_optimized(folio);
 
+		if (hugetlb_bootmem_page_earlycma(m))
+			folio_set_hugetlb_cma(folio);
+
 		list_add(&folio->lru, &folio_list);
 
 		/*
@@ -3606,8 +3662,11 @@ static void __init hugetlb_hstate_alloc_pages(struct hstate *h)
 {
 	unsigned long allocated;
 
-	/* skip gigantic hugepages allocation if hugetlb_cma enabled */
-	if (hstate_is_gigantic(h) && hugetlb_cma_size) {
+	/*
+	 * Skip gigantic hugepages allocation if early CMA
+	 * reservations are not available.
+	 */
+	if (hstate_is_gigantic(h) && hugetlb_cma_size && !hugetlb_early_cma(h)) {
 		pr_warn_once("HugeTLB: hugetlb_cma is enabled, skip boot time allocation\n");
 		return;
 	}
-- 
2.48.1.262.g85cc9f2d1e-goog

[PATCH v2 28/28] mm/hugetlb: move hugetlb CMA code in to its own file

[Frank van der Linden]

hugetlb.c contained a number of CONFIG_CMA ifdefs, and
the code inside them was large enough to merit being in
its own file, so move it, cleaning up things a bit.

Hide some direct variable access behind functions to
accomodate the move.

No functional change intended.

Signed-off-by: Frank van der Linden <fvdl@google.com>
---
 mm/Makefile      |   3 +
 mm/hugetlb.c     | 252 +++------------------------------------------
 mm/hugetlb_cma.c | 258 +++++++++++++++++++++++++++++++++++++++++++++++
 mm/hugetlb_cma.h |  55 ++++++++++
 4 files changed, 332 insertions(+), 236 deletions(-)
 create mode 100644 mm/hugetlb_cma.c
 create mode 100644 mm/hugetlb_cma.h

diff --git a/mm/Makefile b/mm/Makefile
index 850386a67b3e..810ccd45d270 100644
--- a/mm/Makefile
+++ b/mm/Makefile
@@ -79,6 +79,9 @@ obj-$(CONFIG_SWAP)	+= page_io.o swap_state.o swapfile.o swap_slots.o
 obj-$(CONFIG_ZSWAP)	+= zswap.o
 obj-$(CONFIG_HAS_DMA)	+= dmapool.o
 obj-$(CONFIG_HUGETLBFS)	+= hugetlb.o
+ifdef CONFIG_CMA
+obj-$(CONFIG_HUGETLBFS)	+= hugetlb_cma.o
+endif
 obj-$(CONFIG_HUGETLB_PAGE_OPTIMIZE_VMEMMAP)	+= hugetlb_vmemmap.o
 obj-$(CONFIG_NUMA) 	+= mempolicy.o
 obj-$(CONFIG_SPARSEMEM)	+= sparse.o
diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index 5a3e9f7deaba..6e296f16116d 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -50,19 +50,13 @@
 #include <linux/page_owner.h>
 #include "internal.h"
 #include "hugetlb_vmemmap.h"
+#include "hugetlb_cma.h"
 #include <linux/page-isolation.h>
 
 int hugetlb_max_hstate __read_mostly;
 unsigned int default_hstate_idx;
 struct hstate hstates[HUGE_MAX_HSTATE];
 
-#ifdef CONFIG_CMA
-static struct cma *hugetlb_cma[MAX_NUMNODES];
-static unsigned long hugetlb_cma_size_in_node[MAX_NUMNODES] __initdata;
-#endif
-static bool hugetlb_cma_only;
-static unsigned long hugetlb_cma_size __initdata;
-
 __initdata struct list_head huge_boot_pages[MAX_NUMNODES];
 __initdata unsigned long hstate_boot_nrinvalid[HUGE_MAX_HSTATE];
 
@@ -129,14 +123,11 @@ static struct resv_map *vma_resv_map(struct vm_area_struct *vma);
 
 static void hugetlb_free_folio(struct folio *folio)
 {
-#ifdef CONFIG_CMA
-	int nid = folio_nid(folio);
-
 	if (folio_test_hugetlb_cma(folio)) {
-		WARN_ON_ONCE(!cma_free_folio(hugetlb_cma[nid], folio));
+		hugetlb_cma_free_folio(folio);
 		return;
 	}
-#endif
+
 	folio_put(folio);
 }
 
@@ -1493,31 +1484,9 @@ static struct folio *alloc_gigantic_folio(struct hstate *h, gfp_t gfp_mask,
 	if (nid == NUMA_NO_NODE)
 		nid = numa_mem_id();
 retry:
-	folio = NULL;
-#ifdef CONFIG_CMA
-	{
-		int node;
-
-		if (hugetlb_cma[nid])
-			folio = cma_alloc_folio(hugetlb_cma[nid], order, gfp_mask);
-
-		if (!folio && !(gfp_mask & __GFP_THISNODE)) {
-			for_each_node_mask(node, *nodemask) {
-				if (node == nid || !hugetlb_cma[node])
-					continue;
-
-				folio = cma_alloc_folio(hugetlb_cma[node], order, gfp_mask);
-				if (folio)
-					break;
-			}
-		}
-
-		if (folio)
-			folio_set_hugetlb_cma(folio);
-	}
-#endif
+	folio = hugetlb_cma_alloc_folio(h, gfp_mask, nid, nodemask);
 	if (!folio) {
-		if (hugetlb_cma_only)
+		if (hugetlb_cma_exclusive_alloc())
 			return NULL;
 
 		folio = folio_alloc_gigantic(order, gfp_mask, nid, nodemask);
@@ -3180,32 +3149,19 @@ struct folio *alloc_hugetlb_folio(struct vm_area_struct *vma,
 	return ERR_PTR(-ENOSPC);
 }
 
-static bool __init hugetlb_early_cma(struct hstate *h)
-{
-	if (arch_has_huge_bootmem_alloc())
-		return false;
-
-	return (hstate_is_gigantic(h) && hugetlb_cma_only);
-}
-
 static __init void *alloc_bootmem(struct hstate *h, int nid)
 {
 	struct huge_bootmem_page *m;
-	unsigned long flags;
-	struct cma *cma;
 
-#ifdef CONFIG_CMA
-	if (hugetlb_early_cma(h)) {
-		flags = HUGE_BOOTMEM_CMA;
-		cma = hugetlb_cma[nid];
-		m = cma_reserve_early(cma, huge_page_size(h));
-	} else
-#endif
-	{
-		flags = 0;
-		cma = NULL;
+	if (hugetlb_early_cma(h))
+		m = hugetlb_cma_alloc_bootmem(h, nid);
+	else {
 		m = memblock_alloc_try_nid_raw(huge_page_size(h),
 			huge_page_size(h), 0, MEMBLOCK_ALLOC_ACCESSIBLE, nid);
+		if (m) {
+			m->flags = 0;
+			m->cma = NULL;
+		}
 	}
 
 	if (m) {
@@ -3220,8 +3176,6 @@ static __init void *alloc_bootmem(struct hstate *h, int nid)
 		INIT_LIST_HEAD(&m->list);
 		list_add(&m->list, &huge_boot_pages[nid]);
 		m->hstate = h;
-		m->flags = flags;
-		m->cma = cma;
 	}
 
 	return m;
@@ -3666,7 +3620,8 @@ static void __init hugetlb_hstate_alloc_pages(struct hstate *h)
 	 * Skip gigantic hugepages allocation if early CMA
 	 * reservations are not available.
 	 */
-	if (hstate_is_gigantic(h) && hugetlb_cma_size && !hugetlb_early_cma(h)) {
+	if (hstate_is_gigantic(h) && hugetlb_cma_total_size() &&
+	    !hugetlb_early_cma(h)) {
 		pr_warn_once("HugeTLB: hugetlb_cma is enabled, skip boot time allocation\n");
 		return;
 	}
@@ -3703,7 +3658,7 @@ static void __init hugetlb_init_hstates(void)
 		 */
 		if (hstate_is_gigantic(h) && !gigantic_page_runtime_supported())
 			continue;
-		if (hugetlb_cma_size && h->order <= HUGETLB_PAGE_ORDER)
+		if (hugetlb_cma_total_size() && h->order <= HUGETLB_PAGE_ORDER)
 			continue;
 		for_each_hstate(h2) {
 			if (h2 == h)
@@ -4605,14 +4560,6 @@ static void hugetlb_register_all_nodes(void) { }
 
 #endif
 
-#ifdef CONFIG_CMA
-static void __init hugetlb_cma_check(void);
-#else
-static inline __init void hugetlb_cma_check(void)
-{
-}
-#endif
-
 static void __init hugetlb_sysfs_init(void)
 {
 	struct hstate *h;
@@ -4796,8 +4743,7 @@ static __init void hugetlb_parse_params(void)
 		hcp->setup(hcp->val);
 	}
 
-	if (!hugetlb_cma_size)
-		hugetlb_cma_only = false;
+	hugetlb_cma_validate_params();
 }
 
 /*
@@ -7867,169 +7813,3 @@ void hugetlb_unshare_all_pmds(struct vm_area_struct *vma)
 	hugetlb_unshare_pmds(vma, ALIGN(vma->vm_start, PUD_SIZE),
 			ALIGN_DOWN(vma->vm_end, PUD_SIZE));
 }
-
-#ifdef CONFIG_CMA
-static bool cma_reserve_called __initdata;
-
-static int __init cmdline_parse_hugetlb_cma(char *p)
-{
-	int nid, count = 0;
-	unsigned long tmp;
-	char *s = p;
-
-	while (*s) {
-		if (sscanf(s, "%lu%n", &tmp, &count) != 1)
-			break;
-
-		if (s[count] == ':') {
-			if (tmp >= MAX_NUMNODES)
-				break;
-			nid = array_index_nospec(tmp, MAX_NUMNODES);
-
-			s += count + 1;
-			tmp = memparse(s, &s);
-			hugetlb_cma_size_in_node[nid] = tmp;
-			hugetlb_cma_size += tmp;
-
-			/*
-			 * Skip the separator if have one, otherwise
-			 * break the parsing.
-			 */
-			if (*s == ',')
-				s++;
-			else
-				break;
-		} else {
-			hugetlb_cma_size = memparse(p, &p);
-			break;
-		}
-	}
-
-	return 0;
-}
-
-early_param("hugetlb_cma", cmdline_parse_hugetlb_cma);
-
-static int __init cmdline_parse_hugetlb_cma_only(char *p)
-{
-	return kstrtobool(p, &hugetlb_cma_only);
-}
-
-early_param("hugetlb_cma_only", cmdline_parse_hugetlb_cma_only);
-
-void __init hugetlb_cma_reserve(int order)
-{
-	unsigned long size, reserved, per_node;
-	bool node_specific_cma_alloc = false;
-	int nid;
-
-	/*
-	 * HugeTLB CMA reservation is required for gigantic
-	 * huge pages which could not be allocated via the
-	 * page allocator. Just warn if there is any change
-	 * breaking this assumption.
-	 */
-	VM_WARN_ON(order <= MAX_PAGE_ORDER);
-	cma_reserve_called = true;
-
-	if (!hugetlb_cma_size)
-		return;
-
-	for (nid = 0; nid < MAX_NUMNODES; nid++) {
-		if (hugetlb_cma_size_in_node[nid] == 0)
-			continue;
-
-		if (!node_online(nid)) {
-			pr_warn("hugetlb_cma: invalid node %d specified\n", nid);
-			hugetlb_cma_size -= hugetlb_cma_size_in_node[nid];
-			hugetlb_cma_size_in_node[nid] = 0;
-			continue;
-		}
-
-		if (hugetlb_cma_size_in_node[nid] < (PAGE_SIZE << order)) {
-			pr_warn("hugetlb_cma: cma area of node %d should be at least %lu MiB\n",
-				nid, (PAGE_SIZE << order) / SZ_1M);
-			hugetlb_cma_size -= hugetlb_cma_size_in_node[nid];
-			hugetlb_cma_size_in_node[nid] = 0;
-		} else {
-			node_specific_cma_alloc = true;
-		}
-	}
-
-	/* Validate the CMA size again in case some invalid nodes specified. */
-	if (!hugetlb_cma_size)
-		return;
-
-	if (hugetlb_cma_size < (PAGE_SIZE << order)) {
-		pr_warn("hugetlb_cma: cma area should be at least %lu MiB\n",
-			(PAGE_SIZE << order) / SZ_1M);
-		hugetlb_cma_size = 0;
-		return;
-	}
-
-	if (!node_specific_cma_alloc) {
-		/*
-		 * If 3 GB area is requested on a machine with 4 numa nodes,
-		 * let's allocate 1 GB on first three nodes and ignore the last one.
-		 */
-		per_node = DIV_ROUND_UP(hugetlb_cma_size, nr_online_nodes);
-		pr_info("hugetlb_cma: reserve %lu MiB, up to %lu MiB per node\n",
-			hugetlb_cma_size / SZ_1M, per_node / SZ_1M);
-	}
-
-	reserved = 0;
-	for_each_online_node(nid) {
-		int res;
-		char name[CMA_MAX_NAME];
-
-		if (node_specific_cma_alloc) {
-			if (hugetlb_cma_size_in_node[nid] == 0)
-				continue;
-
-			size = hugetlb_cma_size_in_node[nid];
-		} else {
-			size = min(per_node, hugetlb_cma_size - reserved);
-		}
-
-		size = round_up(size, PAGE_SIZE << order);
-
-		snprintf(name, sizeof(name), "hugetlb%d", nid);
-		/*
-		 * Note that 'order per bit' is based on smallest size that
-		 * may be returned to CMA allocator in the case of
-		 * huge page demotion.
-		 */
-		res = cma_declare_contiguous_multi(size, PAGE_SIZE << order,
-					HUGETLB_PAGE_ORDER, name,
-					&hugetlb_cma[nid], nid);
-		if (res) {
-			pr_warn("hugetlb_cma: reservation failed: err %d, node %d",
-				res, nid);
-			continue;
-		}
-
-		reserved += size;
-		pr_info("hugetlb_cma: reserved %lu MiB on node %d\n",
-			size / SZ_1M, nid);
-
-		if (reserved >= hugetlb_cma_size)
-			break;
-	}
-
-	if (!reserved)
-		/*
-		 * hugetlb_cma_size is used to determine if allocations from
-		 * cma are possible.  Set to zero if no cma regions are set up.
-		 */
-		hugetlb_cma_size = 0;
-}
-
-static void __init hugetlb_cma_check(void)
-{
-	if (!hugetlb_cma_size || cma_reserve_called)
-		return;
-
-	pr_warn("hugetlb_cma: the option isn't supported by current arch\n");
-}
-
-#endif /* CONFIG_CMA */
diff --git a/mm/hugetlb_cma.c b/mm/hugetlb_cma.c
new file mode 100644
index 000000000000..3ea9cd0f6b9f
--- /dev/null
+++ b/mm/hugetlb_cma.c
@@ -0,0 +1,258 @@
+// SPDX-License-Identifier: GPL-2.0-only
+
+#include <linux/mm.h>
+#include <linux/cma.h>
+#include <linux/compiler.h>
+#include <linux/mm_inline.h>
+
+#include <asm/page.h>
+#include <asm/setup.h>
+
+#include <linux/hugetlb.h>
+#include "internal.h"
+#include "hugetlb_cma.h"
+
+
+static struct cma *hugetlb_cma[MAX_NUMNODES];
+static unsigned long hugetlb_cma_size_in_node[MAX_NUMNODES] __initdata;
+static bool hugetlb_cma_only;
+static unsigned long hugetlb_cma_size __initdata;
+
+void hugetlb_cma_free_folio(struct folio *folio)
+{
+	int nid = folio_nid(folio);
+
+	WARN_ON_ONCE(!cma_free_folio(hugetlb_cma[nid], folio));
+}
+
+
+struct folio *hugetlb_cma_alloc_folio(struct hstate *h, gfp_t gfp_mask,
+				      int nid, nodemask_t *nodemask)
+{
+	int node;
+	int order = huge_page_order(h);
+	struct folio *folio = NULL;
+
+	if (hugetlb_cma[nid])
+		folio = cma_alloc_folio(hugetlb_cma[nid], order, gfp_mask);
+
+	if (!folio && !(gfp_mask & __GFP_THISNODE)) {
+		for_each_node_mask(node, *nodemask) {
+			if (node == nid || !hugetlb_cma[node])
+				continue;
+
+			folio = cma_alloc_folio(hugetlb_cma[node], order, gfp_mask);
+			if (folio)
+				break;
+		}
+	}
+
+	if (folio)
+		folio_set_hugetlb_cma(folio);
+
+	return folio;
+}
+
+struct huge_bootmem_page * __init
+hugetlb_cma_alloc_bootmem(struct hstate *h, int nid)
+{
+	struct cma *cma;
+	struct huge_bootmem_page *m;
+
+	cma = hugetlb_cma[nid];
+	m = cma_reserve_early(cma, huge_page_size(h));
+	if (m) {
+		m->flags = HUGE_BOOTMEM_CMA;
+		m->cma = cma;
+	}
+
+	return m;
+}
+
+
+static bool cma_reserve_called __initdata;
+
+static int __init cmdline_parse_hugetlb_cma(char *p)
+{
+	int nid, count = 0;
+	unsigned long tmp;
+	char *s = p;
+
+	while (*s) {
+		if (sscanf(s, "%lu%n", &tmp, &count) != 1)
+			break;
+
+		if (s[count] == ':') {
+			if (tmp >= MAX_NUMNODES)
+				break;
+			nid = array_index_nospec(tmp, MAX_NUMNODES);
+
+			s += count + 1;
+			tmp = memparse(s, &s);
+			hugetlb_cma_size_in_node[nid] = tmp;
+			hugetlb_cma_size += tmp;
+
+			/*
+			 * Skip the separator if have one, otherwise
+			 * break the parsing.
+			 */
+			if (*s == ',')
+				s++;
+			else
+				break;
+		} else {
+			hugetlb_cma_size = memparse(p, &p);
+			break;
+		}
+	}
+
+	return 0;
+}
+
+early_param("hugetlb_cma", cmdline_parse_hugetlb_cma);
+
+static int __init cmdline_parse_hugetlb_cma_only(char *p)
+{
+	return kstrtobool(p, &hugetlb_cma_only);
+}
+
+early_param("hugetlb_cma_only", cmdline_parse_hugetlb_cma_only);
+
+void __init hugetlb_cma_reserve(int order)
+{
+	unsigned long size, reserved, per_node;
+	bool node_specific_cma_alloc = false;
+	int nid;
+
+	/*
+	 * HugeTLB CMA reservation is required for gigantic
+	 * huge pages which could not be allocated via the
+	 * page allocator. Just warn if there is any change
+	 * breaking this assumption.
+	 */
+	VM_WARN_ON(order <= MAX_PAGE_ORDER);
+	cma_reserve_called = true;
+
+	if (!hugetlb_cma_size)
+		return;
+
+	for (nid = 0; nid < MAX_NUMNODES; nid++) {
+		if (hugetlb_cma_size_in_node[nid] == 0)
+			continue;
+
+		if (!node_online(nid)) {
+			pr_warn("hugetlb_cma: invalid node %d specified\n", nid);
+			hugetlb_cma_size -= hugetlb_cma_size_in_node[nid];
+			hugetlb_cma_size_in_node[nid] = 0;
+			continue;
+		}
+
+		if (hugetlb_cma_size_in_node[nid] < (PAGE_SIZE << order)) {
+			pr_warn("hugetlb_cma: cma area of node %d should be at least %lu MiB\n",
+				nid, (PAGE_SIZE << order) / SZ_1M);
+			hugetlb_cma_size -= hugetlb_cma_size_in_node[nid];
+			hugetlb_cma_size_in_node[nid] = 0;
+		} else {
+			node_specific_cma_alloc = true;
+		}
+	}
+
+	/* Validate the CMA size again in case some invalid nodes specified. */
+	if (!hugetlb_cma_size)
+		return;
+
+	if (hugetlb_cma_size < (PAGE_SIZE << order)) {
+		pr_warn("hugetlb_cma: cma area should be at least %lu MiB\n",
+			(PAGE_SIZE << order) / SZ_1M);
+		hugetlb_cma_size = 0;
+		return;
+	}
+
+	if (!node_specific_cma_alloc) {
+		/*
+		 * If 3 GB area is requested on a machine with 4 numa nodes,
+		 * let's allocate 1 GB on first three nodes and ignore the last one.
+		 */
+		per_node = DIV_ROUND_UP(hugetlb_cma_size, nr_online_nodes);
+		pr_info("hugetlb_cma: reserve %lu MiB, up to %lu MiB per node\n",
+			hugetlb_cma_size / SZ_1M, per_node / SZ_1M);
+	}
+
+	reserved = 0;
+	for_each_online_node(nid) {
+		int res;
+		char name[CMA_MAX_NAME];
+
+		if (node_specific_cma_alloc) {
+			if (hugetlb_cma_size_in_node[nid] == 0)
+				continue;
+
+			size = hugetlb_cma_size_in_node[nid];
+		} else {
+			size = min(per_node, hugetlb_cma_size - reserved);
+		}
+
+		size = round_up(size, PAGE_SIZE << order);
+
+		snprintf(name, sizeof(name), "hugetlb%d", nid);
+		/*
+		 * Note that 'order per bit' is based on smallest size that
+		 * may be returned to CMA allocator in the case of
+		 * huge page demotion.
+		 */
+		res = cma_declare_contiguous_multi(size, PAGE_SIZE << order,
+					HUGETLB_PAGE_ORDER, name,
+					&hugetlb_cma[nid], nid);
+		if (res) {
+			pr_warn("hugetlb_cma: reservation failed: err %d, node %d",
+				res, nid);
+			continue;
+		}
+
+		reserved += size;
+		pr_info("hugetlb_cma: reserved %lu MiB on node %d\n",
+			size / SZ_1M, nid);
+
+		if (reserved >= hugetlb_cma_size)
+			break;
+	}
+
+	if (!reserved)
+		/*
+		 * hugetlb_cma_size is used to determine if allocations from
+		 * cma are possible.  Set to zero if no cma regions are set up.
+		 */
+		hugetlb_cma_size = 0;
+}
+
+void __init hugetlb_cma_check(void)
+{
+	if (!hugetlb_cma_size || cma_reserve_called)
+		return;
+
+	pr_warn("hugetlb_cma: the option isn't supported by current arch\n");
+}
+
+bool hugetlb_cma_exclusive_alloc(void)
+{
+	return hugetlb_cma_only;
+}
+
+unsigned long __init hugetlb_cma_total_size(void)
+{
+	return hugetlb_cma_size;
+}
+
+void __init hugetlb_cma_validate_params(void)
+{
+	if (!hugetlb_cma_size)
+		hugetlb_cma_only = false;
+}
+
+bool __init hugetlb_early_cma(struct hstate *h)
+{
+	if (arch_has_huge_bootmem_alloc())
+		return false;
+
+	return hstate_is_gigantic(h) && hugetlb_cma_only;
+}
diff --git a/mm/hugetlb_cma.h b/mm/hugetlb_cma.h
new file mode 100644
index 000000000000..92eb7530fe9e
--- /dev/null
+++ b/mm/hugetlb_cma.h
@@ -0,0 +1,55 @@
+// SPDX-License-Identifier: GPL-2.0
+#ifndef _LINUX_HUGETLB_CMA_H
+#define _LINUX_HUGETLB_CMA_H
+
+#ifdef CONFIG_CMA
+void hugetlb_cma_free_folio(struct folio *folio);
+struct folio *hugetlb_cma_alloc_folio(struct hstate *h, gfp_t gfp_mask,
+				      int nid, nodemask_t *nodemask);
+struct huge_bootmem_page *hugetlb_cma_alloc_bootmem(struct hstate *h, int nid);
+void hugetlb_cma_check(void);
+bool hugetlb_cma_exclusive_alloc(void);
+unsigned long hugetlb_cma_total_size(void);
+void hugetlb_cma_validate_params(void);
+bool hugetlb_early_cma(struct hstate *h);
+#else
+static inline void hugetlb_cma_free_folio(struct folio *folio)
+{
+}
+
+static inline struct folio *hugetlb_cma_alloc_folio(struct hstate *h,
+	    gfp_t gfp_mask, int nid, nodemask_t *nodemask)
+{
+	return NULL;
+}
+
+static inline
+struct huge_bootmem_page *hugetlb_cma_alloc_bootmem(struct hstate *h, int nid)
+{
+	return NULL;
+}
+
+static inline void hugetlb_cma_check(void)
+{
+}
+
+static inline bool hugetlb_cma_exclusive_alloc(void)
+{
+	return false;
+}
+
+static inline unsigned long hugetlb_cma_total_size(void)
+{
+	return 0;
+}
+
+static inline void hugetlb_cma_validate_params(void)
+{
+}
+
+static inline bool hugetlb_early_cma(struct hstate *h)
+{
+	return false;
+}
+#endif
+#endif
-- 
2.48.1.262.g85cc9f2d1e-goog