[PATCH Linux 2.6.31] Add hugepage support to the balloon driver

[Dave McCracken <dcm@mccr.org>]

[-- Attachment #1: Type: Text/Plain, Size: 339 bytes --]


This patch adds a boot command line switch "balloon_hugepages" that makes the 
balloon driver work entirely in hugepages.

This version of the code checks to make sure each page is contiguous at the 
machine level.  If it is not, it frees each page individually.  When the huge 
page is reallocated it will be contiguous.

Dave McCracken

[-- Attachment #2: linux-balloon-2.6.31-1.patch --]
[-- Type: text/x-patch, Size: 9404 bytes --]

--- 2.6.31/./drivers/xen/balloon.c	2009-09-09 17:13:59.000000000 -0500
+++ 2.6.31-balloon/./drivers/xen/balloon.c	2009-10-28 07:21:01.000000000 -0500
@@ -58,7 +58,7 @@
 #include <xen/features.h>
 #include <xen/page.h>
 
-#define PAGES2KB(_p) ((_p)<<(PAGE_SHIFT-10))
+#define PAGES2KB(_p) ((_p)<<(PAGE_SHIFT+balloon_order-10))
 
 #define BALLOON_CLASS_NAME "xen_memory"
 
@@ -93,6 +93,14 @@ static DEFINE_SPINLOCK(balloon_lock);
 
 static struct balloon_stats balloon_stats;
 
+/*
+ * Work in pages of this order.  Can be either 0 for normal pages
+ * or 9 for hugepages.
+ */
+static int balloon_order;
+static unsigned long balloon_npages;
+static unsigned long discontig_frame_list[PAGE_SIZE / sizeof(unsigned long)];
+
 /* We increase/decrease in batches which fit in a page */
 static unsigned long frame_list[PAGE_SIZE / sizeof(unsigned long)];
 
@@ -124,10 +132,41 @@ static struct timer_list balloon_timer;
 static void scrub_page(struct page *page)
 {
 #ifdef CONFIG_XEN_SCRUB_PAGES
-	clear_highpage(page);
+	int i;
+
+	for (i = 0; i < balloon_npages; i++)
+		clear_highpage(page++);
 #endif
 }
 
+static void free_discontig_frame(void)
+{
+	int rc;
+	struct xen_memory_reservation reservation = {
+		.address_bits = 0,
+		.domid        = DOMID_SELF,
+		.nr_extents   = balloon_npages,
+		.extent_order = 0
+	};
+
+	set_xen_guest_handle(reservation.extent_start, discontig_frame_list);
+	rc = HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation);
+	BUG_ON(rc != balloon_npages);
+}
+
+static unsigned long shrink_frame(unsigned long nr_pages)
+{
+	unsigned long i, j;
+
+	for (i = 0, j = 0; i < nr_pages; i++, j++) {
+		if (frame_list[i] == 0)
+			j++;
+		if (i != j)
+			frame_list[i] = frame_list[j];
+	}
+	return i;
+}
+
 /* balloon_append: add the given page to the balloon. */
 static void balloon_append(struct page *page)
 {
@@ -197,12 +236,11 @@ static unsigned long current_target(void
 
 static int increase_reservation(unsigned long nr_pages)
 {
-	unsigned long  pfn, i, flags;
+	unsigned long  pfn, mfn, i, j, flags;
 	struct page   *page;
 	long           rc;
 	struct xen_memory_reservation reservation = {
 		.address_bits = 0,
-		.extent_order = 0,
 		.domid        = DOMID_SELF
 	};
 
@@ -214,12 +252,14 @@ static int increase_reservation(unsigned
 	page = balloon_first_page();
 	for (i = 0; i < nr_pages; i++) {
 		BUG_ON(page == NULL);
-		frame_list[i] = page_to_pfn(page);;
+		frame_list[i] = page_to_pfn(page);
 		page = balloon_next_page(page);
 	}
 
 	set_xen_guest_handle(reservation.extent_start, frame_list);
 	reservation.nr_extents = nr_pages;
+	reservation.extent_order = balloon_order;
+
 	rc = HYPERVISOR_memory_op(XENMEM_populate_physmap, &reservation);
 	if (rc < nr_pages) {
 		if (rc > 0) {
@@ -242,19 +282,22 @@ static int increase_reservation(unsigned
 		BUG_ON(page == NULL);
 
 		pfn = page_to_pfn(page);
+		mfn = frame_list[i];
 		BUG_ON(!xen_feature(XENFEAT_auto_translated_physmap) &&
 		       phys_to_machine_mapping_valid(pfn));
 
-		set_phys_to_machine(pfn, frame_list[i]);
+		for (j = 0; j < balloon_npages; j++, pfn++, mfn++) {
+			set_phys_to_machine(pfn, mfn);
 
-		/* Link back into the page tables if not highmem. */
-		if (pfn < max_low_pfn) {
-			int ret;
-			ret = HYPERVISOR_update_va_mapping(
-				(unsigned long)__va(pfn << PAGE_SHIFT),
-				mfn_pte(frame_list[i], PAGE_KERNEL),
-				0);
-			BUG_ON(ret);
+			/* Link back into the page tables if not highmem. */
+			if (pfn < max_low_pfn) {
+				int ret;
+				ret = HYPERVISOR_update_va_mapping(
+					(unsigned long)__va(pfn << PAGE_SHIFT),
+					mfn_pte(mfn, PAGE_KERNEL),
+					0);
+				BUG_ON(ret);
+			}
 		}
 
 		/* Relinquish the page back to the allocator. */
@@ -264,7 +307,7 @@ static int increase_reservation(unsigned
 	}
 
 	balloon_stats.current_pages += nr_pages;
-	totalram_pages = balloon_stats.current_pages;
+	totalram_pages = balloon_stats.current_pages << balloon_order;
 
  out:
 	spin_unlock_irqrestore(&balloon_lock, flags);
@@ -274,13 +317,13 @@ static int increase_reservation(unsigned
 
 static int decrease_reservation(unsigned long nr_pages)
 {
-	unsigned long  pfn, i, flags;
+	unsigned long  pfn, lpfn, mfn, i, j, flags;
 	struct page   *page;
 	int            need_sleep = 0;
-	int ret;
+	int		discontig, discontig_free;
+	int		ret;
 	struct xen_memory_reservation reservation = {
 		.address_bits = 0,
-		.extent_order = 0,
 		.domid        = DOMID_SELF
 	};
 
@@ -288,7 +331,7 @@ static int decrease_reservation(unsigned
 		nr_pages = ARRAY_SIZE(frame_list);
 
 	for (i = 0; i < nr_pages; i++) {
-		if ((page = alloc_page(GFP_BALLOON)) == NULL) {
+		if ((page = alloc_pages(GFP_BALLOON, balloon_order)) == NULL) {
 			nr_pages = i;
 			need_sleep = 1;
 			break;
@@ -298,14 +341,6 @@ static int decrease_reservation(unsigned
 		frame_list[i] = pfn_to_mfn(pfn);
 
 		scrub_page(page);
-
-		if (!PageHighMem(page)) {
-			ret = HYPERVISOR_update_va_mapping(
-				(unsigned long)__va(pfn << PAGE_SHIFT),
-				__pte_ma(0), 0);
-			BUG_ON(ret);
-                }
-
 	}
 
 	/* Ensure that ballooned highmem pages don't have kmaps. */
@@ -315,20 +350,42 @@ static int decrease_reservation(unsigned
 	spin_lock_irqsave(&balloon_lock, flags);
 
 	/* No more mappings: invalidate P2M and add to balloon. */
+	discontig = 0;
 	for (i = 0; i < nr_pages; i++) {
-		pfn = mfn_to_pfn(frame_list[i]);
-		set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
+		mfn = frame_list[i];
+		lpfn = pfn = mfn_to_pfn(mfn);
 		balloon_append(pfn_to_page(pfn));
+		discontig_free = 0;
+		for (j = 0; j < balloon_npages; j++, lpfn++, mfn++) {
+			if ((discontig_frame_list[j] = pfn_to_mfn(lpfn)) != mfn)
+				discontig_free = 1;
+
+			set_phys_to_machine(lpfn, INVALID_P2M_ENTRY);
+			if (!PageHighMem(page)) {
+				ret = HYPERVISOR_update_va_mapping(
+					(unsigned long)__va(lpfn << PAGE_SHIFT),
+					__pte_ma(0), 0);
+				BUG_ON(ret);
+			}
+		}
+		if (discontig_free) {
+			free_discontig_frame();
+			frame_list[i] = 0;
+			discontig = 1;
+		}
 	}
+	balloon_stats.current_pages -= nr_pages;
+	totalram_pages = balloon_stats.current_pages << balloon_order;
+
+	if (discontig)
+		nr_pages = shrink_frame(nr_pages);
 
 	set_xen_guest_handle(reservation.extent_start, frame_list);
 	reservation.nr_extents   = nr_pages;
+	reservation.extent_order = balloon_order;
 	ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation);
 	BUG_ON(ret != nr_pages);
 
-	balloon_stats.current_pages -= nr_pages;
-	totalram_pages = balloon_stats.current_pages;
-
 	spin_unlock_irqrestore(&balloon_lock, flags);
 
 	return need_sleep;
@@ -397,7 +454,7 @@ static void watch_target(struct xenbus_w
 	/* The given memory/target value is in KiB, so it needs converting to
 	 * pages. PAGE_SHIFT converts bytes to pages, hence PAGE_SHIFT - 10.
 	 */
-	balloon_set_new_target(new_target >> (PAGE_SHIFT - 10));
+	balloon_set_new_target(new_target >> ((PAGE_SHIFT - 10) + balloon_order));
 }
 
 static int balloon_init_watcher(struct notifier_block *notifier,
@@ -423,10 +480,13 @@ static int __init balloon_init(void)
 	if (!xen_pv_domain())
 		return -ENODEV;
 
-	pr_info("xen_balloon: Initialising balloon driver.\n");
+	pr_info("xen_balloon: Initialising balloon driver with page order %d.\n",
+		balloon_order);
 
-	balloon_stats.current_pages = min(xen_start_info->nr_pages, max_pfn);
-	totalram_pages   = balloon_stats.current_pages;
+	balloon_npages = 1 << balloon_order;
+
+	balloon_stats.current_pages = (min(xen_start_info->nr_pages, max_pfn)) >> balloon_order;
+	totalram_pages   = balloon_stats.current_pages << balloon_order;
 	balloon_stats.target_pages  = balloon_stats.current_pages;
 	balloon_stats.balloon_low   = 0;
 	balloon_stats.balloon_high  = 0;
@@ -440,10 +500,12 @@ static int __init balloon_init(void)
 	register_balloon(&balloon_sysdev);
 
 	/* Initialise the balloon with excess memory space. */
-	for (pfn = xen_start_info->nr_pages; pfn < max_pfn; pfn++) {
-		page = pfn_to_page(pfn);
-		if (!PageReserved(page))
-			balloon_append(page);
+	for (pfn = xen_start_info->nr_pages; pfn < max_pfn; pfn += balloon_npages) {
+		if (pfn_valid(pfn)) {
+			page = pfn_to_page(pfn);
+			if (!PageReserved(page))
+				balloon_append(page);
+		}
 	}
 
 	target_watch.callback = watch_target;
@@ -464,6 +526,14 @@ static void balloon_exit(void)
 
 module_exit(balloon_exit);
 
+static int __init balloon_parse_huge(char *s)
+{
+	balloon_order = 9;
+	return 1;
+}
+
+__setup("balloon_hugepages", balloon_parse_huge);
+
 #define BALLOON_SHOW(name, format, args...)				\
 	static ssize_t show_##name(struct sys_device *dev,		\
 				   struct sysdev_attribute *attr,	\
@@ -500,7 +570,7 @@ static ssize_t store_target_kb(struct sy
 
 	target_bytes = simple_strtoull(buf, &endchar, 0) * 1024;
 
-	balloon_set_new_target(target_bytes >> PAGE_SHIFT);
+	balloon_set_new_target(target_bytes >> (PAGE_SHIFT + balloon_order));
 
 	return count;
 }
@@ -514,7 +584,7 @@ static ssize_t show_target(struct sys_de
 {
 	return sprintf(buf, "%llu\n",
 		       (unsigned long long)balloon_stats.target_pages
-		       << PAGE_SHIFT);
+		       << (PAGE_SHIFT + balloon_order));
 }
 
 static ssize_t store_target(struct sys_device *dev,
@@ -530,7 +600,7 @@ static ssize_t store_target(struct sys_d
 
 	target_bytes = memparse(buf, &endchar);
 
-	balloon_set_new_target(target_bytes >> PAGE_SHIFT);
+	balloon_set_new_target(target_bytes >> (PAGE_SHIFT + balloon_order));
 
 	return count;
 }

[-- Attachment #3: Type: text/plain, Size: 138 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel