LKML Archive on lore.kernel.org
help / color / mirror / Atom feed
* [PATCH R2 7/7] xen/balloon: Xen memory balloon driver with memory hotplug support
@ 2010-12-29 17:07 Daniel Kiper
  2011-01-03 22:33 ` Dave Hansen
  0 siblings, 1 reply; 4+ messages in thread
From: Daniel Kiper @ 2010-12-29 17:07 UTC (permalink / raw)
  To: ian.campbell, akpm, andi.kleen, haicheng.li, fengguang.wu,
	jeremy, konrad.wilk, dan.magenheimer, v.tolstov, xen-devel,
	linux-kernel, linux-mm

Features and fixes:
  - new version of memory hotplug patch which supports
    among others memory allocation policies during errors
    (try until success or stop at first error),
  - this version of patch was tested with tmem
    (selfballooning and frontswap) and works
    very well with it,
  - some other minor fixes.

Signed-off-by: Daniel Kiper <dkiper@net-space.pl>
---
 drivers/xen/Kconfig   |   10 ++
 drivers/xen/balloon.c |  222 ++++++++++++++++++++++++++++++++++++++++++++++---
 2 files changed, 221 insertions(+), 11 deletions(-)

diff --git a/drivers/xen/Kconfig b/drivers/xen/Kconfig
index 60d71e9..ada8ef5 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 "Xen memory balloon driver with memory hotplug support"
+	default n
+	depends on XEN_BALLOON && MEMORY_HOTPLUG
+	help
+	  Xen memory balloon driver with memory hotplug support 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 f105e67..73abacd 100644
--- a/drivers/xen/balloon.c
+++ b/drivers/xen/balloon.c
@@ -6,6 +6,7 @@
  * 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
  *
  * This program is free software; you can redistribute it and/or
  * modify it under the terms of the GNU General Public License version 2
@@ -44,6 +45,7 @@
 #include <linux/list.h>
 #include <linux/sysdev.h>
 #include <linux/gfp.h>
+#include <linux/memory.h>
 
 #include <asm/page.h>
 #include <asm/pgalloc.h>
@@ -79,6 +81,9 @@ enum bp_state {
 	BP_HUNGRY
 };
 
+#define MH_POLICY_TRY_UNTIL_SUCCESS	0
+#define MH_POLICY_STOP_AT_FIRST_ERROR	1
+
 struct balloon_stats {
 	/* We aim for 'current allocation' == 'target allocation'. */
 	unsigned long current_pages;
@@ -88,6 +93,10 @@ struct balloon_stats {
 	unsigned long balloon_high;
 	unsigned long schedule_delay;
 	unsigned long max_schedule_delay;
+#ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
+	unsigned long boot_max_pfn;
+	unsigned long mh_policy;
+#endif
 };
 
 static DEFINE_MUTEX(balloon_mutex);
@@ -201,18 +210,195 @@ static void update_schedule_delay(enum bp_state state)
 	balloon_stats.schedule_delay = new_schedule_delay;
 }
 
-static unsigned long current_target(void)
+#ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
+static int allocate_memory_resource(struct resource **r, unsigned long nr_pages)
 {
-	unsigned long target = balloon_stats.target_pages;
+	int rc;
+	resource_size_t r_min, r_size;
+
+	/*
+	 * Look for first unused memory region starting at page
+	 * boundary. Skip last memory section created at boot time
+	 * becuase it may contains unused memory pages with PG_reserved
+	 * bit not set (online_pages require PG_reserved bit set).
+	 */
+
+	*r = kzalloc(sizeof(struct resource), GFP_KERNEL);
 
-	target = min(target,
-		     balloon_stats.current_pages +
-		     balloon_stats.balloon_low +
-		     balloon_stats.balloon_high);
+	if (!*r)
+		return -ENOMEM;
 
-	return target;
+	(*r)->name = "System RAM";
+	(*r)->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
+	r_min = PFN_PHYS(section_nr_to_pfn(pfn_to_section_nr(balloon_stats.boot_max_pfn) + 1));
+	r_size = nr_pages << PAGE_SHIFT;
+
+	rc = allocate_resource(&iomem_resource, *r, r_size, r_min,
+					ULONG_MAX, PAGE_SIZE, NULL, NULL);
+
+	if (rc < 0) {
+		kfree(*r);
+		*r = NULL;
+	}
+
+	return rc;
 }
 
+static void adjust_memory_resource(struct resource **r, unsigned long nr_pages)
+{
+	if ((*r)->end + 1 - (nr_pages << PAGE_SHIFT) == (*r)->start) {
+		BUG_ON(release_resource(*r) < 0);
+		kfree(*r);
+		*r = NULL;
+		return;
+	}
+
+	BUG_ON(adjust_resource(*r, (*r)->start, (*r)->end + 1 - (*r)->start -
+				(nr_pages << PAGE_SHIFT)) < 0);
+}
+
+static int allocate_additional_memory(struct resource *r, unsigned long nr_pages)
+{
+	int rc;
+	struct xen_memory_reservation reservation = {
+		.address_bits = 0,
+		.extent_order = 0,
+		.domid        = DOMID_SELF
+	};
+	unsigned long flags, i, pfn, pfn_start;
+
+	if (!nr_pages)
+		return 0;
+
+	pfn_start = PFN_UP(r->end) - nr_pages;
+
+	if (nr_pages > ARRAY_SIZE(frame_list))
+		nr_pages = ARRAY_SIZE(frame_list);
+
+	for (i = 0, pfn = pfn_start; i < nr_pages; ++i, ++pfn)
+		frame_list[i] = pfn;
+
+	set_xen_guest_handle(reservation.extent_start, frame_list);
+	reservation.nr_extents = nr_pages;
+
+	spin_lock_irqsave(&xen_reservation_lock, flags);
+
+	rc = HYPERVISOR_memory_op(XENMEM_populate_physmap, &reservation);
+
+	if (rc <= 0)
+		return (rc < 0) ? rc : -ENOMEM;
+
+	for (i = 0, pfn = pfn_start; i < rc; ++i, ++pfn) {
+		BUG_ON(!xen_feature(XENFEAT_auto_translated_physmap) &&
+		       phys_to_machine_mapping_valid(pfn));
+		set_phys_to_machine(pfn, frame_list[i]);
+	}
+
+	spin_unlock_irqrestore(&xen_reservation_lock, flags);
+
+	return rc;
+}
+
+static void hotplug_allocated_memory(struct resource **r)
+{
+	int nid, rc;
+	resource_size_t r_size;
+	struct memory_block *mem;
+	unsigned long pfn;
+
+	r_size = (*r)->end + 1 - (*r)->start;
+	nid = memory_add_physaddr_to_nid((*r)->start);
+
+	rc = add_registered_memory(nid, (*r)->start, r_size);
+
+	if (rc) {
+		pr_err("%s: add_registered_memory: Memory hotplug failed: %i\n",
+			__func__, rc);
+		balloon_stats.target_pages = balloon_stats.current_pages;
+		*r = NULL;
+		return;
+	}
+
+	if (xen_pv_domain())
+		for (pfn = PFN_DOWN((*r)->start); pfn < PFN_UP((*r)->end); ++pfn)
+			if (!PageHighMem(pfn_to_page(pfn)))
+				BUG_ON(HYPERVISOR_update_va_mapping(
+					(unsigned long)__va(pfn << PAGE_SHIFT),
+					mfn_pte(pfn_to_mfn(pfn), PAGE_KERNEL), 0));
+
+	rc = online_pages(PFN_DOWN((*r)->start), r_size >> PAGE_SHIFT);
+
+	if (rc) {
+		pr_err("%s: online_pages: Failed: %i\n", __func__, rc);
+		balloon_stats.target_pages = balloon_stats.current_pages;
+		*r = NULL;
+		return;
+	}
+
+	for (pfn = PFN_DOWN((*r)->start); pfn < PFN_UP((*r)->end); pfn += PAGES_PER_SECTION) {
+		mem = find_memory_block(__pfn_to_section(pfn));
+		BUG_ON(!mem);
+		BUG_ON(!present_section_nr(mem->phys_index));
+		mutex_lock(&mem->state_mutex);
+		mem->state = MEM_ONLINE;
+		mutex_unlock(&mem->state_mutex);
+	}
+
+	balloon_stats.current_pages += r_size >> PAGE_SHIFT;
+
+	*r = NULL;
+}
+
+static enum bp_state request_additional_memory(long credit)
+{
+	int rc;
+	static struct resource *r;
+	static unsigned long pages_left;
+
+	if ((credit <= 0 || balloon_stats.balloon_low ||
+				balloon_stats.balloon_high) && !r)
+		return BP_DONE;
+
+	if (!r) {
+		rc = allocate_memory_resource(&r, credit);
+
+		if (rc)
+			return BP_ERROR;
+
+		pages_left = credit;
+	}
+
+	rc = allocate_additional_memory(r, pages_left);
+
+	if (rc < 0) {
+		if (balloon_stats.mh_policy == MH_POLICY_TRY_UNTIL_SUCCESS)
+			return BP_ERROR;
+
+		adjust_memory_resource(&r, pages_left);
+
+		if (!r)
+			return BP_ERROR;
+	} else {
+		pages_left -= rc;
+
+		if (pages_left)
+			return BP_HUNGRY;
+	}
+
+	hotplug_allocated_memory(&r);
+
+	return BP_DONE;
+}
+#else
+static enum bp_state request_additional_memory(long credit)
+{
+	if (balloon_stats.balloon_low && balloon_stats.balloon_high &&
+			balloon_stats.target_pages > balloon_stats.current_pages)
+		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)
 {
 	enum bp_state state = BP_DONE;
@@ -358,15 +544,17 @@ static enum bp_state decrease_reservation(unsigned long nr_pages)
  */
 static void balloon_process(struct work_struct *work)
 {
-	enum bp_state rc, state = BP_DONE;
+	enum bp_state rc, state;
 	long credit;
 
 	mutex_lock(&balloon_mutex);
 
 	do {
-		credit = current_target() - balloon_stats.current_pages;
+		credit = balloon_stats.target_pages - balloon_stats.current_pages;
+
+		state = request_additional_memory(credit);
 
-		if (credit > 0) {
+		if (credit > 0 && state == BP_DONE) {
 			rc = increase_reservation(credit);
 			state = (rc == BP_ERROR) ? BP_ERROR : state;
 		}
@@ -458,6 +646,11 @@ static int __init balloon_init(void)
 	balloon_stats.schedule_delay = 1;
 	balloon_stats.max_schedule_delay = 32;
 
+#ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
+	balloon_stats.boot_max_pfn = max_pfn;
+	balloon_stats.mh_policy = MH_POLICY_STOP_AT_FIRST_ERROR;
+#endif
+
 	register_balloon(&balloon_sysdev);
 
 	/* Initialise the balloon with excess memory space. */
@@ -501,6 +694,10 @@ BALLOON_SHOW(high_kb, "%lu\n", PAGES2KB(balloon_stats.balloon_high));
 static SYSDEV_ULONG_ATTR(schedule_delay, 0644, balloon_stats.schedule_delay);
 static SYSDEV_ULONG_ATTR(max_schedule_delay, 0644, balloon_stats.max_schedule_delay);
 
+#ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
+static SYSDEV_ULONG_ATTR(memory_hotplug_policy, 0644, balloon_stats.mh_policy);
+#endif
+
 static ssize_t show_target_kb(struct sys_device *dev, struct sysdev_attribute *attr,
 			      char *buf)
 {
@@ -563,7 +760,10 @@ static struct sysdev_attribute *balloon_attrs[] = {
 	&attr_target_kb,
 	&attr_target,
 	&attr_schedule_delay.attr,
-	&attr_max_schedule_delay.attr
+	&attr_max_schedule_delay.attr,
+#ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
+	&attr_memory_hotplug_policy.attr
+#endif
 };
 
 static struct attribute *balloon_info_attrs[] = {
-- 
1.4.4.4


^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: [PATCH R2 7/7] xen/balloon: Xen memory balloon driver with memory hotplug support
  2010-12-29 17:07 [PATCH R2 7/7] xen/balloon: Xen memory balloon driver with memory hotplug support Daniel Kiper
@ 2011-01-03 22:33 ` Dave Hansen
  2011-01-11 17:51   ` Daniel Kiper
  2011-02-03 15:31   ` Daniel Kiper
  0 siblings, 2 replies; 4+ messages in thread
From: Dave Hansen @ 2011-01-03 22:33 UTC (permalink / raw)
  To: Daniel Kiper
  Cc: ian.campbell, akpm, andi.kleen, haicheng.li, fengguang.wu,
	jeremy, konrad.wilk, dan.magenheimer, v.tolstov, xen-devel,
	linux-kernel, linux-mm

On Wed, 2010-12-29 at 18:07 +0100, Daniel Kiper wrote:
> +config XEN_BALLOON_MEMORY_HOTPLUG
> +	bool "Xen memory balloon driver with memory hotplug support"
> +	default n
> +	depends on XEN_BALLOON && MEMORY_HOTPLUG
> +	help
> +	  Xen memory balloon driver with memory hotplug support 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.

This might be better phrased as "Memory hotplug support for Xen balloon
driver".  It might otherwise confuse people about whether they're seeing
some kind of choice or an _enhancement_ to the existing driver.

Also, why bother even making this a config option?  What are the
downsides if it was always compiled in?  You could even make it a
non-prompting Kconfig option and just automatically turn it on with
XEN_BALLOON && MEMORY_HOTPLUG.


> +static int allocate_memory_resource(struct resource **r, unsigned long nr_pages)
>  {
> -	unsigned long target = balloon_stats.target_pages;
> +	int rc;
> +	resource_size_t r_min, r_size;
> +
> +	/*
> +	 * Look for first unused memory region starting at page
> +	 * boundary. Skip last memory section created at boot time
> +	 * becuase it may contains unused memory pages with PG_reserved
> +	 * bit not set (online_pages require PG_reserved bit set).
> +	 */
> +
> +	*r = kzalloc(sizeof(struct resource), GFP_KERNEL);
> 
> -	target = min(target,
> -		     balloon_stats.current_pages +
> -		     balloon_stats.balloon_low +
> -		     balloon_stats.balloon_high);
> +	if (!*r)
> +		return -ENOMEM;
> 
> -	return target;
> +	(*r)->name = "System RAM";
> +	(*r)->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
> +	r_min = PFN_PHYS(section_nr_to_pfn(pfn_to_section_nr(balloon_stats.boot_max_pfn) + 1));
> +	r_size = nr_pages << PAGE_SHIFT;
> +
> +	rc = allocate_resource(&iomem_resource, *r, r_size, r_min,
> +					ULONG_MAX, PAGE_SIZE, NULL, NULL);
> +
> +	if (rc < 0) {
> +		kfree(*r);
> +		*r = NULL;
> +	}
> +
> +	return rc;
>  }

The double-pointer stuff here ends up looking a little funky.  Is there
any reason you don't just use ERR_PTRs?  That might look a bit more
sane.

> +static void adjust_memory_resource(struct resource **r, unsigned long nr_pages)
> +{
> +	if ((*r)->end + 1 - (nr_pages << PAGE_SHIFT) == (*r)->start) {
> +		BUG_ON(release_resource(*r) < 0);

In some kernels, people do:

	#define BUG_ON(...) do{}while(0)

to save space.  If anyone ever does that with this code, it'll break
horribly.  It's also hard to read these.  So, please break logic actions
_out_ of the BUG_ON() arguments.

That's repeated in quite a few places in here.  Make sure to go get them
all.

It also isn't evident what this patch set is trying to do until you get
down to this 7/7 patch.  You might want to put a more complete
description in 0/7.

-- Dave


^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: [PATCH R2 7/7] xen/balloon: Xen memory balloon driver with memory hotplug support
  2011-01-03 22:33 ` Dave Hansen
@ 2011-01-11 17:51   ` Daniel Kiper
  2011-02-03 15:31   ` Daniel Kiper
  1 sibling, 0 replies; 4+ messages in thread
From: Daniel Kiper @ 2011-01-11 17:51 UTC (permalink / raw)
  To: Dave Hansen
  Cc: Daniel Kiper, ian.campbell, akpm, andi.kleen, haicheng.li,
	fengguang.wu, jeremy, konrad.wilk, dan.magenheimer, v.tolstov,
	xen-devel, linux-kernel, linux-mm

Hi everybody,

Sorry for delays. I could not devote fully my time to this
project :-((( (especially at the beginning of this year).
I am going to post all replies and new version of patch next week.
Please be patient.

Daniel

^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: [PATCH R2 7/7] xen/balloon: Xen memory balloon driver with memory hotplug support
  2011-01-03 22:33 ` Dave Hansen
  2011-01-11 17:51   ` Daniel Kiper
@ 2011-02-03 15:31   ` Daniel Kiper
  1 sibling, 0 replies; 4+ messages in thread
From: Daniel Kiper @ 2011-02-03 15:31 UTC (permalink / raw)
  To: Dave Hansen
  Cc: Daniel Kiper, ian.campbell, akpm, andi.kleen, haicheng.li,
	fengguang.wu, jeremy, konrad.wilk, dan.magenheimer, v.tolstov,
	xen-devel, linux-kernel, linux-mm

On Mon, Jan 03, 2011 at 02:33:22PM -0800, Dave Hansen wrote:
> On Wed, 2010-12-29 at 18:07 +0100, Daniel Kiper wrote:
> > +config XEN_BALLOON_MEMORY_HOTPLUG
> > +	bool "Xen memory balloon driver with memory hotplug support"
> > +	default n
> > +	depends on XEN_BALLOON && MEMORY_HOTPLUG
> > +	help
> > +	  Xen memory balloon driver with memory hotplug support 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.
>
> This might be better phrased as "Memory hotplug support for Xen balloon
> driver".  It might otherwise confuse people about whether they're seeing
> some kind of choice or an _enhancement_ to the existing driver.

Thx. Done. I will send new patch release today.

> Also, why bother even making this a config option?  What are the
> downsides if it was always compiled in?  You could even make it a
> non-prompting Kconfig option and just automatically turn it on with
> XEN_BALLOON && MEMORY_HOTPLUG.

At this stage of development I think it is better to leave this
as an option for user. Later when this future will be mature it could
be removed. However, I realized that this solution does not give
possibility for user to disable this future at runtime. Maybe
it shoudl be as boot/sysfs option. I am not sure it is required or not.
Currently it is not implemented.

> > +static int allocate_memory_resource(struct resource **r, unsigned long nr_pages)
> >  {
> > -	unsigned long target = balloon_stats.target_pages;
> > +	int rc;
> > +	resource_size_t r_min, r_size;
> > +
> > +	/*
> > +	 * Look for first unused memory region starting at page
> > +	 * boundary. Skip last memory section created at boot time
> > +	 * becuase it may contains unused memory pages with PG_reserved
> > +	 * bit not set (online_pages require PG_reserved bit set).
> > +	 */
> > +
> > +	*r = kzalloc(sizeof(struct resource), GFP_KERNEL);
> >
> > -	target = min(target,
> > -		     balloon_stats.current_pages +
> > -		     balloon_stats.balloon_low +
> > -		     balloon_stats.balloon_high);
> > +	if (!*r)
> > +		return -ENOMEM;
> >
> > -	return target;
> > +	(*r)->name = "System RAM";
> > +	(*r)->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
> > +	r_min = PFN_PHYS(section_nr_to_pfn(pfn_to_section_nr(balloon_stats.boot_max_pfn) + 1));
> > +	r_size = nr_pages << PAGE_SHIFT;
> > +
> > +	rc = allocate_resource(&iomem_resource, *r, r_size, r_min,
> > +					ULONG_MAX, PAGE_SIZE, NULL, NULL);
> > +
> > +	if (rc < 0) {
> > +		kfree(*r);
> > +		*r = NULL;
> > +	}
> > +
> > +	return rc;
> >  }
>
> The double-pointer stuff here ends up looking a little funky.  Is there
> any reason you don't just use ERR_PTRs?  That might look a bit more
> sane.

After another review I removed all double-pointer stuff (it was
very simple, however, if I read many times code written by myself
it was very difficult to discover that). Now I think it is much better.
Thanks.

> > +static void adjust_memory_resource(struct resource **r, unsigned long nr_pages)
> > +{
> > +	if ((*r)->end + 1 - (nr_pages << PAGE_SHIFT) == (*r)->start) {
> > +		BUG_ON(release_resource(*r) < 0);
>
> In some kernels, people do:
>
> 	#define BUG_ON(...) do{}while(0)
>
> to save space.  If anyone ever does that with this code, it'll break
> horribly.  It's also hard to read these.  So, please break logic actions
> _out_ of the BUG_ON() arguments.
>
> That's repeated in quite a few places in here.  Make sure to go get them
> all.

Done.

> It also isn't evident what this patch set is trying to do until you get
> down to this 7/7 patch.  You might want to put a more complete
> description in 0/7.

Will be done. :-))) Thank you for your help.

Daniel

^ permalink raw reply	[flat|nested] 4+ messages in thread

end of thread, other threads:[~2011-02-03 15:32 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2010-12-29 17:07 [PATCH R2 7/7] xen/balloon: Xen memory balloon driver with memory hotplug support Daniel Kiper
2011-01-03 22:33 ` Dave Hansen
2011-01-11 17:51   ` Daniel Kiper
2011-02-03 15:31   ` Daniel Kiper

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).