Xen devel - May 2011 - [PATCH V2] xen/balloon: Memory hotplug support for Xen balloon driver

Memory hotplug support for Xen balloon driver. It should be
mentioned that hotplugged memory is not onlined automatically.
It should be onlined by user through standard sysfs interface.

This patch applies to Linus'' git tree, v2.6.39-rc5 tag with a few
prerequisite patches available at https://lkml.org/lkml/2011/5/2/339
and at https://lkml.org/lkml/2011/3/28/98.

I have received notice that previous series of patches broke machine
migration under Xen. I am going to confirm that and solve that problem
ASAP (#define ASAP LATER_THEN_SOONER_HOWEVER_IT_WILL_BE_DONE). I do not
have received any notices about other problems till now.

Signed-off-by: Daniel Kiper <dkiper@net-space.pl>
---
 drivers/xen/Kconfig   |   10 ++++
 drivers/xen/balloon.c |  139 ++++++++++++++++++++++++++++++++++++++++++++++++-
 include/xen/balloon.h |    4 ++
 3 files changed, 151 insertions(+), 2 deletions(-)

diff --git a/drivers/xen/Kconfig b/drivers/xen/Kconfig
index a59638b..bd8cfae 100644
--- a/drivers/xen/Kconfig
+++ b/drivers/xen/Kconfig
@@ -9,6 +9,16 @@ config XEN_BALLOON
 	  the system to expand the domain''s memory allocation, or
alternatively
 	  return unneeded memory to the system.
 
+config XEN_BALLOON_MEMORY_HOTPLUG
+	bool "Memory hotplug support for Xen balloon driver"
+	default n
+	depends on XEN_BALLOON && MEMORY_HOTPLUG
+	help
+	  Memory hotplug support for Xen balloon driver allows expanding memory
+	  available for the system above limit declared at system startup.
+	  It is very useful on critical systems which require long
+	  run without rebooting.
+
 config XEN_SCRUB_PAGES
 	bool "Scrub pages before returning them to system"
 	depends on XEN_BALLOON
diff --git a/drivers/xen/balloon.c b/drivers/xen/balloon.c
index f54290b..e16cf26 100644
--- a/drivers/xen/balloon.c
+++ b/drivers/xen/balloon.c
@@ -4,6 +4,12 @@
  * Copyright (c) 2003, B Dragovic
  * Copyright (c) 2003-2004, M Williamson, K Fraser
  * Copyright (c) 2005 Dan M. Smith, IBM Corporation
+ * Copyright (c) 2010 Daniel Kiper
+ *
+ * Memory hotplug support was written by Daniel Kiper. Work on
+ * it was sponsored by Google under Google Summer of Code 2010
+ * program. Jeremy Fitzhardinge from Xen.org was the mentor for
+ * this project.
  *
  * This program is free software; you can redistribute it and/or
  * modify it under the terms of the GNU General Public License version 2
@@ -40,6 +46,9 @@
 #include <linux/mutex.h>
 #include <linux/list.h>
 #include <linux/gfp.h>
+#include <linux/notifier.h>
+#include <linux/memory.h>
+#include <linux/memory_hotplug.h>
 
 #include <asm/page.h>
 #include <asm/pgalloc.h>
@@ -194,6 +203,87 @@ static enum bp_state update_schedule(enum bp_state state)
 	return BP_EAGAIN;
 }
 
+#ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
+static long current_credit(void)
+{
+	return balloon_stats.target_pages - balloon_stats.current_pages -
+		balloon_stats.hotplug_pages;
+}
+
+static bool balloon_is_inflated(void)
+{
+	if (balloon_stats.balloon_low || balloon_stats.balloon_high ||
+			balloon_stats.balloon_hotplug)
+		return true;
+	else
+		return false;
+}
+
+/*
+ * reserve_additional_memory() adds memory region of size >= credit above
+ * max_pfn. New region is section aligned and size is modified to be multiple
+ * of section size. Those features allow optimal use of address space and
+ * establish proper alignment when this function is called first time after
+ * boot (last section not fully populated at boot time contains unused memory
+ * pages with PG_reserved bit not set; online_pages_range() does not allow page
+ * onlining in whole range if first onlined page does not have PG_reserved
+ * bit set). Real size of added memory is established at page onlining stage.
+ */
+
+static enum bp_state reserve_additional_memory(long credit)
+{
+	int nid, rc;
+	u64 hotplug_start_paddr;
+	unsigned long balloon_hotplug = credit;
+
+	hotplug_start_paddr = PFN_PHYS(SECTION_ALIGN_UP(max_pfn));
+	balloon_hotplug = round_up(balloon_hotplug, PAGES_PER_SECTION);
+	nid = memory_add_physaddr_to_nid(hotplug_start_paddr);
+
+	rc = add_memory(nid, hotplug_start_paddr, balloon_hotplug <<
PAGE_SHIFT);
+
+	if (rc) {
+		pr_info("xen_balloon: %s: add_memory() failed: %i\n", __func__,
rc);
+		return BP_EAGAIN;
+	}
+
+	balloon_hotplug -= credit;
+
+	balloon_stats.hotplug_pages += credit;
+	balloon_stats.balloon_hotplug = balloon_hotplug;
+
+	return BP_DONE;
+}
+
+static void xen_online_page(struct page *page)
+{
+	__online_page_set_limits(page);
+
+	mutex_lock(&balloon_mutex);
+
+	__balloon_append(page);
+
+	if (balloon_stats.hotplug_pages)
+		--balloon_stats.hotplug_pages;
+	else
+		--balloon_stats.balloon_hotplug;
+
+	mutex_unlock(&balloon_mutex);
+}
+
+static int xen_memory_notifier(struct notifier_block *nb, unsigned long val,
void *v)
+{
+	if (val == MEM_ONLINE)
+		schedule_delayed_work(&balloon_worker, 0);
+
+	return NOTIFY_OK;
+}
+
+static struct notifier_block xen_memory_nb = {
+	.notifier_call = xen_memory_notifier,
+	.priority = 0
+};
+#else
 static long current_credit(void)
 {
 	unsigned long target = balloon_stats.target_pages;
@@ -206,6 +296,21 @@ static long current_credit(void)
 	return target - balloon_stats.current_pages;
 }
 
+static bool balloon_is_inflated(void)
+{
+	if (balloon_stats.balloon_low || balloon_stats.balloon_high)
+		return true;
+	else
+		return false;
+}
+
+static enum bp_state reserve_additional_memory(long credit)
+{
+	balloon_stats.target_pages = balloon_stats.current_pages;
+	return BP_DONE;
+}
+#endif /* CONFIG_XEN_BALLOON_MEMORY_HOTPLUG */
+
 static enum bp_state increase_reservation(unsigned long nr_pages)
 {
 	int rc;
@@ -217,6 +322,15 @@ static enum bp_state increase_reservation(unsigned long
nr_pages)
 		.domid        = DOMID_SELF
 	};
 
+#ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
+	if (!balloon_stats.balloon_low && !balloon_stats.balloon_high) {
+		nr_pages = min(nr_pages, balloon_stats.balloon_hotplug);
+		balloon_stats.hotplug_pages += nr_pages;
+		balloon_stats.balloon_hotplug -= nr_pages;
+		return BP_DONE;
+	}
+#endif
+
 	if (nr_pages > ARRAY_SIZE(frame_list))
 		nr_pages = ARRAY_SIZE(frame_list);
 
@@ -279,6 +393,15 @@ static enum bp_state decrease_reservation(unsigned long
nr_pages, gfp_t gfp)
 		.domid        = DOMID_SELF
 	};
 
+#ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
+	if (balloon_stats.hotplug_pages) {
+		nr_pages = min(nr_pages, balloon_stats.hotplug_pages);
+		balloon_stats.hotplug_pages -= nr_pages;
+		balloon_stats.balloon_hotplug += nr_pages;
+		return BP_DONE;
+	}
+#endif
+
 	if (nr_pages > ARRAY_SIZE(frame_list))
 		nr_pages = ARRAY_SIZE(frame_list);
 
@@ -340,8 +463,12 @@ static void balloon_process(struct work_struct *work)
 	do {
 		credit = current_credit();
 
-		if (credit > 0)
-			state = increase_reservation(credit);
+		if (credit > 0) {
+			if (balloon_is_inflated())
+				state = increase_reservation(credit);
+			else
+				state = reserve_additional_memory(credit);
+		}
 
 		if (credit < 0)
 			state = decrease_reservation(-credit, GFP_BALLOON);
@@ -448,6 +575,14 @@ static int __init balloon_init(void)
 	balloon_stats.retry_count = 1;
 	balloon_stats.max_retry_count = RETRY_UNLIMITED;
 
+#ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
+	balloon_stats.hotplug_pages = 0;
+	balloon_stats.balloon_hotplug = 0;
+
+	register_online_page_callback(&xen_online_page);
+	register_memory_notifier(&xen_memory_nb);
+#endif
+
 	/*
 	 * Initialise the balloon with excess memory space.  We need
 	 * to make sure we don''t add memory which doesn''t exist or
diff --git a/include/xen/balloon.h b/include/xen/balloon.h
index a2b22f0..aeca6ae 100644
--- a/include/xen/balloon.h
+++ b/include/xen/balloon.h
@@ -15,6 +15,10 @@ struct balloon_stats {
 	unsigned long max_schedule_delay;
 	unsigned long retry_count;
 	unsigned long max_retry_count;
+#ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
+	unsigned long hotplug_pages;
+	unsigned long balloon_hotplug;
+#endif
 };
 
 extern struct balloon_stats balloon_stats;
-- 
1.5.6.5

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

On Tue, 2011-05-03 at 00:01 +0200, Daniel Kiper wrote:
> @@ -448,6 +575,14 @@ static int __init balloon_init(void)
>         balloon_stats.retry_count = 1;
>         balloon_stats.max_retry_count = RETRY_UNLIMITED;
> 
> +#ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
> +       balloon_stats.hotplug_pages = 0;
> +       balloon_stats.balloon_hotplug = 0;
> +
> +       register_online_page_callback(&xen_online_page);
> +       register_memory_notifier(&xen_memory_nb);
> +#endif 
This is 100% static, apparently.  XEN_BALLOON can''t be a module, so I
still don''t see the point of having the un/register stuff.  

-- Dave



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

On Tue, May 03, 2011 at 09:32:33AM -0700, Dave Hansen
wrote:
> On Tue, 2011-05-03 at 00:01 +0200, Daniel Kiper wrote:
> > @@ -448,6 +575,14 @@ static int __init balloon_init(void)
> >         balloon_stats.retry_count = 1;
> >         balloon_stats.max_retry_count = RETRY_UNLIMITED;
> >
> > +#ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
> > +       balloon_stats.hotplug_pages = 0;
> > +       balloon_stats.balloon_hotplug = 0;
> > +
> > +       register_online_page_callback(&xen_online_page);
> > +       register_memory_notifier(&xen_memory_nb);
> > +#endif 
>
> This is 100% static, apparently.  XEN_BALLOON can''t be a module,
so I
> still don''t see the point of having the un/register stuff.  
You are right to some extent. However, xen_online_page() is registered
as page onlining function only on Xen hypervisor. On bare metal
generic_online_page() is only valid page onlining function.

Additionally, I think this callback mechanism enable other balloon
implementations (KVM, VMware, ...) to easily integrate with memory
hotplug. If it comes true (I am going to propose relevant solution
maybe with more generic balloon driver for Linux Kernel somewhen;
I have some ideas, however, I must focus on more important issues
for me now) proper page onlining function (for Xen, KVM, ...) should
be registered at boot time or module load/unload (after hypervisor
detection). That is why I am insisting on run time solution.
It is an investment into the future.

Daniel

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

On Tue, May 03, 2011 at 12:01:48AM +0200, Daniel Kiper
wrote:
> Memory hotplug support for Xen balloon driver. It should be
> mentioned that hotplugged memory is not onlined automatically.
> It should be onlined by user through standard sysfs interface.
> 
> This patch applies to Linus'' git tree, v2.6.39-rc5 tag with a few
> prerequisite patches available at https://lkml.org/lkml/2011/5/2/339
> and at https://lkml.org/lkml/2011/3/28/98.
The patch looks good. How do I use it? Should the writeup or the
Kconfig include a little section on how to online the memory?

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