[RFC PATCH 0/2] allow bootmem to be freed to allocator late

[RFC PATCH 0/2] allow bootmem to be freed to allocator late

[Chris Wright]

Currently there is no way to release bootmem once the bootmem allocator
frees all unreserved memory.  This adds the ability to free reserved
pages directly to the page allocator after the bootmem allocator metadata
is already freed.  It's limited in scope since it's still all marked
__init, and creates a new entry point free_bootmem_late rather than
trying to do this automatically in free_bootmem.  Hence the RFC...

With this we are able to do something like allocate swiotlb, and then
free it later if we discover we had a hw iommu that doesn't need swiotlb.

 include/linux/bootmem.h |    1 +
 mm/bootmem.c            |   98 ++++++++++++++++++++++++++++++++++++----------
 2 files changed, 77 insertions(+), 22 deletions(-)

thanks,
-chris

[RFC PATCH 1/2] bootmem: refactor free_all_bootmem_core

[Chris Wright]

[-- Attachment #1: bootmem-break-out-free_pages_bootmem-loop.patch --]
[-- Type: text/plain, Size: 3359 bytes --]

Move the loop that frees all bootmem pages back to page allocator into
its own function.  This should have not functional effect and allows the
function to be reused later.

Cc: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Signed-off-by: Chris Wright <chrisw@sous-sol.org>
---
 mm/bootmem.c |   61 +++++++++++++++++++++++++++++++++++++++-------------------
 1 files changed, 41 insertions(+), 20 deletions(-)

diff --git a/mm/bootmem.c b/mm/bootmem.c
index 555d5d2..94ef2e7 100644
--- a/mm/bootmem.c
+++ b/mm/bootmem.c
@@ -143,17 +143,22 @@ unsigned long __init init_bootmem(unsigned long start, unsigned long pages)
 	return init_bootmem_core(NODE_DATA(0)->bdata, start, 0, pages);
 }
 
-static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
+/**
+ * free_bootmem_pages - frees bootmem pages to page allocator
+ * @start: start pfn
+ * @end: end pfn
+ * @map: bootmem bitmap of reserved pages
+ *
+ * This will free the pages in the range @start to @end, making them
+ * available to the page allocator.  The @map will be used to skip
+ * reserved pages.  Returns the count of pages freed.
+ */
+static unsigned long __init free_bootmem_pages(unsigned long start,
+					       unsigned long end,
+					       unsigned long *map)
 {
+	unsigned long cursor, count = 0;
 	int aligned;
-	struct page *page;
-	unsigned long start, end, pages, count = 0;
-
-	if (!bdata->node_bootmem_map)
-		return 0;
-
-	start = bdata->node_min_pfn;
-	end = bdata->node_low_pfn;
 
 	/*
 	 * If the start is aligned to the machines wordsize, we might
@@ -161,27 +166,25 @@ static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
 	 */
 	aligned = !(start & (BITS_PER_LONG - 1));
 
-	bdebug("nid=%td start=%lx end=%lx aligned=%d\n",
-		bdata - bootmem_node_data, start, end, aligned);
+	for (cursor = start; cursor < end; cursor += BITS_PER_LONG) {
+		unsigned long idx, vec;
 
-	while (start < end) {
-		unsigned long *map, idx, vec;
-
-		map = bdata->node_bootmem_map;
-		idx = start - bdata->node_min_pfn;
+		idx = cursor - start;
 		vec = ~map[idx / BITS_PER_LONG];
 
-		if (aligned && vec == ~0UL && start + BITS_PER_LONG < end) {
+		if (aligned && vec == ~0UL && cursor + BITS_PER_LONG < end) {
 			int order = ilog2(BITS_PER_LONG);
 
-			__free_pages_bootmem(pfn_to_page(start), order);
+			__free_pages_bootmem(pfn_to_page(cursor), order);
 			count += BITS_PER_LONG;
 		} else {
 			unsigned long off = 0;
 
 			while (vec && off < BITS_PER_LONG) {
 				if (vec & 1) {
-					page = pfn_to_page(start + off);
+					struct page *page;
+
+					page = pfn_to_page(cursor + off);
 					__free_pages_bootmem(page, 0);
 					count++;
 				}
@@ -189,8 +192,26 @@ static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
 				off++;
 			}
 		}
-		start += BITS_PER_LONG;
 	}
+	return count;
+}
+
+static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
+{
+	struct page *page;
+	unsigned long start, end, *map, pages, count = 0;
+
+	if (!bdata->node_bootmem_map)
+		return 0;
+
+	start = bdata->node_min_pfn;
+	end = bdata->node_low_pfn;
+	map = bdata->node_bootmem_map;
+
+	bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data,
+		start, end);
+
+	count = free_bootmem_pages(start, end, map);
 
 	page = virt_to_page(bdata->node_bootmem_map);
 	pages = bdata->node_low_pfn - bdata->node_min_pfn;

[RFC PATCH 2/2] bootmem: add free_bootmem_late

[Chris Wright]

[-- Attachment #1: bootmem-make-free_pages_bootmem-generic.patch --]
[-- Type: text/plain, Size: 3135 bytes --]

Add a new function for freeing bootmem after the bootmem allocator has
been released and the unreserved pages given to the page allocator.
This allows us to reserve bootmem and then release it if we later
discover it was not needed.

Cc: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Signed-off-by: Chris Wright <chrisw@sous-sol.org>
---
 include/linux/bootmem.h |    1 +
 mm/bootmem.c            |   43 ++++++++++++++++++++++++++++++++++++++-----
 2 files changed, 39 insertions(+), 5 deletions(-)

--- a/include/linux/bootmem.h
+++ b/include/linux/bootmem.h
@@ -53,6 +53,7 @@ extern void free_bootmem_node(pg_data_t 
 			      unsigned long addr,
 			      unsigned long size);
 extern void free_bootmem(unsigned long addr, unsigned long size);
+extern void free_bootmem_late(unsigned long addr, unsigned long size);
 
 /*
  * Flags for reserve_bootmem (also if CONFIG_HAVE_ARCH_BOOTMEM_NODE,
--- a/mm/bootmem.c
+++ b/mm/bootmem.c
@@ -151,7 +151,9 @@ unsigned long __init init_bootmem(unsign
  *
  * This will free the pages in the range @start to @end, making them
  * available to the page allocator.  The @map will be used to skip
- * reserved pages.  Returns the count of pages freed.
+ * reserved pages.  In the case that @map is NULL, the bootmem allocator
+ * is already free and the range is contiguous.  Returns the count of
+ * pages freed.
  */
 static unsigned long __init free_bootmem_pages(unsigned long start,
 					       unsigned long end,
@@ -164,13 +166,23 @@ static unsigned long __init free_bootmem
 	 * If the start is aligned to the machines wordsize, we might
 	 * be able to free pages in bulks of that order.
 	 */
-	aligned = !(start & (BITS_PER_LONG - 1));
+	if (map)
+		aligned = !(start & (BITS_PER_LONG - 1));
+	else
+		aligned = 1;
 
 	for (cursor = start; cursor < end; cursor += BITS_PER_LONG) {
-		unsigned long idx, vec;
+		unsigned long vec;
 
-		idx = cursor - start;
-		vec = ~map[idx / BITS_PER_LONG];
+		if (map) {
+			unsigned long idx = cursor - start;
+			vec = ~map[idx / BITS_PER_LONG];
+		} else {
+			if (end - cursor >= BITS_PER_LONG)
+				vec = ~0UL;
+			else
+				vec = (1UL << (end - cursor)) - 1;
+		}
 
 		if (aligned && vec == ~0UL && cursor + BITS_PER_LONG < end) {
 			int order = ilog2(BITS_PER_LONG);
@@ -387,6 +399,27 @@ void __init free_bootmem(unsigned long a
 }
 
 /**
+ * free_bootmem_late - free bootmem pages directly to page allocator
+ * @addr: starting address of the range
+ * @size: size of the range in bytes
+ *
+ * This is only useful when the bootmem allocator has already been torn
+ * down, but we are still initializing the system.  Pages are given directly
+ * to the page allocator, no bootmem metadata is updated because it is gone.
+ */
+void __init free_bootmem_late(unsigned long addr, unsigned long size)
+{
+	unsigned long start, end;
+
+	kmemleak_free_part(__va(addr), size);
+
+	start = PFN_UP(addr);
+	end = PFN_DOWN(addr + size);
+
+	totalram_pages += free_bootmem_pages(start, end, NULL);
+}
+
+/**
  * reserve_bootmem_node - mark a page range as reserved
  * @pgdat: node the range resides on
  * @physaddr: starting address of the range