[PATCH v4 1/2] memblock: drop for_each_memblock_type() and open code its users

[PATCH v4 1/2] memblock: drop for_each_memblock_type() and open code its users

[Tarun Sahu]

for_each_memblock_type() always starts the iteration at index 0 and its
body is a trivial 'for' loop, so it hides very little. It also gets in
the way of iterating from an arbitrary index, which the next patch
needs.

Remove the macro and open code its three users: memblock_add_range(),
memblock_isolate_range() and memblock_dump(). While at it, move the
region pointer into the loop body scope.

No functional change.

Reviewed-by: Pratyush Yadav <pratyush@kernel.org>
Signed-off-by: Tarun Sahu <tarunsahu@google.com>
---
 mm/memblock.c | 17 ++++++-----------
 1 file changed, 6 insertions(+), 11 deletions(-)

diff --git a/mm/memblock.c b/mm/memblock.c
index f5aaa38cfa31..59dda7d085f3 100644
--- a/mm/memblock.c
+++ b/mm/memblock.c
@@ -155,11 +155,6 @@ struct memblock_type physmem = {
  */
 static __refdata struct memblock_type *memblock_memory = &memblock.memory;
 
-#define for_each_memblock_type(i, memblock_type, rgn)			\
-	for (i = 0, rgn = &memblock_type->regions[0];			\
-	     i < memblock_type->cnt;					\
-	     i++, rgn = &memblock_type->regions[i])
-
 #define memblock_dbg(fmt, ...)						\
 	do {								\
 		if (memblock_debug)					\
@@ -615,7 +610,6 @@ static int __init_memblock memblock_add_range(struct memblock_type *type,
 	phys_addr_t obase = base;
 	phys_addr_t end = base + memblock_cap_size(base, &size);
 	int idx, nr_new, start_rgn = -1, end_rgn;
-	struct memblock_region *rgn;
 
 	if (!size)
 		return 0;
@@ -651,7 +645,8 @@ static int __init_memblock memblock_add_range(struct memblock_type *type,
 	base = obase;
 	nr_new = 0;
 
-	for_each_memblock_type(idx, type, rgn) {
+	for (idx = 0; idx < type->cnt; idx++) {
+		struct memblock_region *rgn = &type->regions[idx];
 		phys_addr_t rbase = rgn->base;
 		phys_addr_t rend = rbase + rgn->size;
 
@@ -815,7 +810,6 @@ static int __init_memblock memblock_isolate_range(struct memblock_type *type,
 {
 	phys_addr_t end = base + memblock_cap_size(base, &size);
 	int idx;
-	struct memblock_region *rgn;
 
 	*start_rgn = *end_rgn = 0;
 
@@ -827,7 +821,8 @@ static int __init_memblock memblock_isolate_range(struct memblock_type *type,
 		if (memblock_double_array(type, base, size) < 0)
 			return -ENOMEM;
 
-	for_each_memblock_type(idx, type, rgn) {
+	for (idx = 0; idx < type->cnt; idx++) {
+		struct memblock_region *rgn = &type->regions[idx];
 		phys_addr_t rbase = rgn->base;
 		phys_addr_t rend = rbase + rgn->size;
 
@@ -2194,11 +2189,11 @@ static void __init_memblock memblock_dump(struct memblock_type *type)
 	phys_addr_t base, end, size;
 	enum memblock_flags flags;
 	int idx;
-	struct memblock_region *rgn;
 
 	pr_info(" %s.cnt  = 0x%lx\n", type->name, type->cnt);
 
-	for_each_memblock_type(idx, type, rgn) {
+	for (idx = 0; idx < type->cnt; idx++) {
+		struct memblock_region *rgn = &type->regions[idx];
 		char nid_buf[32] = "";
 
 		base = rgn->base;
-- 
2.56.0.385.gd3acb90ef8-goog

[PATCH v4 2/2] memblock: use binary search to locate candidate regions

[Tarun Sahu]

Use binary search (memblock_bsearch_start) in memblock_add_range() and
memblock_isolate_range() to locate candidate regions instead of linearly
scanning from index 0.

Under heavy memory fragmentation (such as KHO page preservation registering
hundreds of thousands of disjoint folios), scanning from index 0 on every
insertion and isolation results in O(N^2) complexity, causing boot-time
memory retrieval to take several minutes (~268s for 393k pages).

Using binary search reduces the candidate search complexity to O(log N)
(from O(N)), cutting KHO memory retrieval time from ~268s to ~50ms.

memblock_search() open codes the same binary search, so reimplement it on
top of the new helper.

Signed-off-by: Tarun Sahu <tarunsahu@google.com>
Reviewed-by: Pratyush Yadav <pratyush@kernel.org>
---
 mm/memblock.c | 45 +++++++++++++++++++++++++++++++--------------
 1 file changed, 31 insertions(+), 14 deletions(-)

diff --git a/mm/memblock.c b/mm/memblock.c
index 59dda7d085f3..614d0a55dd70 100644
--- a/mm/memblock.c
+++ b/mm/memblock.c
@@ -586,6 +586,29 @@ static void __init_memblock memblock_insert_region(struct memblock_type *type,
 	type->total_size += size;
 }
 
+/**
+ * __memblock_search - Find the first region index where rend > base
+ * @type: memblock type to search
+ * @base: base physical address of the candidate range
+ *
+ * Returns the first region index that could potentially overlap @base.
+ */
+static int __init_memblock __memblock_search(struct memblock_type *type,
+						  phys_addr_t base)
+{
+	int mid, low = 0;
+	int high = type->cnt;
+
+	while (low < high) {
+		mid = (low + high) / 2;
+		if (type->regions[mid].base + type->regions[mid].size <= base)
+			low = mid + 1;
+		else
+			high = mid;
+	}
+	return low;
+}
+
 /**
  * memblock_add_range - add new memblock region
  * @type: memblock type to add new region into
@@ -644,8 +667,9 @@ static int __init_memblock memblock_add_range(struct memblock_type *type,
 	 */
 	base = obase;
 	nr_new = 0;
+	idx = __memblock_search(type, base);
 
-	for (idx = 0; idx < type->cnt; idx++) {
+	for (; idx < type->cnt; idx++) {
 		struct memblock_region *rgn = &type->regions[idx];
 		phys_addr_t rbase = rgn->base;
 		phys_addr_t rend = rbase + rgn->size;
@@ -821,7 +845,9 @@ static int __init_memblock memblock_isolate_range(struct memblock_type *type,
 		if (memblock_double_array(type, base, size) < 0)
 			return -ENOMEM;
 
-	for (idx = 0; idx < type->cnt; idx++) {
+	idx = __memblock_search(type, base);
+
+	for (; idx < type->cnt; idx++) {
 		struct memblock_region *rgn = &type->regions[idx];
 		phys_addr_t rbase = rgn->base;
 		phys_addr_t rend = rbase + rgn->size;
@@ -2062,19 +2088,10 @@ void __init memblock_mem_limit_remove_map(phys_addr_t limit)
 
 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
 {
-	unsigned int left = 0, right = type->cnt;
-
-	do {
-		unsigned int mid = (right + left) / 2;
+	int idx = __memblock_search(type, addr);
 
-		if (addr < type->regions[mid].base)
-			right = mid;
-		else if (addr >= (type->regions[mid].base +
-				  type->regions[mid].size))
-			left = mid + 1;
-		else
-			return mid;
-	} while (left < right);
+	if (idx < type->cnt && addr >= type->regions[idx].base)
+		return idx;
 	return -1;
 }
 
-- 
2.56.0.385.gd3acb90ef8-goog

Re: [PATCH v4 2/2] memblock: use binary search to locate candidate regions

[sashiko-bot]

> Use binary search (memblock_bsearch_start) in memblock_add_range() and
> memblock_isolate_range() to locate candidate regions instead of linearly
> scanning from index 0.
> 
> Under heavy memory fragmentation (such as KHO page preservation registering
> hundreds of thousands of disjoint folios), scanning from index 0 on every
> insertion and isolation results in O(N^2) complexity, causing boot-time
> memory retrieval to take several minutes (~268s for 393k pages).
> 
> Using binary search reduces the candidate search complexity to O(log N)
> (from O(N)), cutting KHO memory retrieval time from ~268s to ~50ms.
> 
> memblock_search() open codes the same binary search, so reimplement it on
> top of the new helper.
> 
> Signed-off-by: Tarun Sahu <tarunsahu@google.com>

Sashiko has reviewed this patch and found no issues. It looks great!

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261008190828.3221718-1-tarunsahu@google.com?part=2

Re: [PATCH v4 1/2] memblock: drop for_each_memblock_type() and open code its users

[sashiko-bot]

> for_each_memblock_type() always starts the iteration at index 0 and its
> body is a trivial 'for' loop, so it hides very little. It also gets in
> the way of iterating from an arbitrary index, which the next patch
> needs.
> 
> Remove the macro and open code its three users: memblock_add_range(),
> memblock_isolate_range() and memblock_dump(). While at it, move the
> region pointer into the loop body scope.
> 
> No functional change.
> 
> Reviewed-by: Pratyush Yadav <pratyush@kernel.org>
> Signed-off-by: Tarun Sahu <tarunsahu@google.com>

Sashiko has reviewed this patch and found no issues. It looks great!

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261008190828.3221718-1-tarunsahu@google.com?part=1