Skip to contentSkip to conversation

pvops: fix "xm save -c" issue

PATCH In review Kenji Wakamiya 5 participants 12 messages Dec 14, 2010 – Jan 25, 2011

Subsystems (6)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 pvops: fix "xm save -c" issue

Hi,

I'm investigating the issue of "xm save -c" in case of PV guests. Then, I tried to prevent calling dpm_resume_end() when suspend was canceled. It seems to work fine about blk and net.
How about this?

Thanks,

Signature · 1 line
Kenji Wakamiya
drivers/xen/manage.c +1 -2
index 0b50906..3dcc270 100644
@@ -148,11 +148,10 @@ out_resume:
 	if (!cancelled) {
 		xen_arch_resume();
 		xs_resume();
+		dpm_resume_end(PMSG_RESUME);
 	} else
 		xs_suspend_cancel();
 
-	dpm_resume_end(PMSG_RESUME);
-
 	/* Make sure timer events get retriggered on all CPUs */
 	clock_was_set();
 

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xensource.com/xen-devel

#1 lore raw
Quoted 6 lines · Kenji Wakamiya

Hi,

I'm investigating the issue of "xm save -c" in case of PV guests. Then, I tried to prevent calling dpm_resume_end() when suspend was canceled. It seems to work fine about blk and net.
How about this?

Could you give more details of what the failure is?

Quoted 31 lines

Thanks,

Signature · 1 line
Kenji Wakamiya
drivers/xen/manage.c +1 -2
index 0b50906..3dcc270 100644
@@ -148,11 +148,10 @@ out_resume:
 	if (!cancelled) {
 		xen_arch_resume();
 		xs_resume();
+		dpm_resume_end(PMSG_RESUME);
 	} else
 		xs_suspend_cancel();
 
-	dpm_resume_end(PMSG_RESUME);
-
 	/* Make sure timer events get retriggered on all CPUs */
 	clock_was_set();
 

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xensource.com/xen-devel

#2 lore raw

Hi Konrad, thank you for your response,

Quoted 10 lines · Konrad Rzeszutek Wilk
Quoted 6 lines · Kenji Wakamiya

Hi,

I'm investigating the issue of "xm save -c" in case of PV guests. Then, I tried to prevent calling dpm_resume_end() when suspend was canceled. It seems to work fine about blk and net.
How about this?

Could you give more details of what the failure is?

"xm save -c" saves domain state to storage and leaves it running after creating snapshot. When this command is executed for a pvops guest (xen/stable-2.6.32.x), the guest will hang up after accessing to devices.

When the guest is unapused, stop_machine() in manage.c returns as the suspend is cancelled. In that case, I think dpm_resume_end() should not be called after stop_machine().

I tested vbd and net, the guest did not hang.
But I'm not sure if this is a right way...

Thanks,

Quoted 31 lines
Quoted 30 lines

Thanks,

Signature · 1 line
Kenji Wakamiya
drivers/xen/manage.c +1 -2
index 0b50906..3dcc270 100644
@@ -148,11 +148,10 @@ out_resume:
 	if (!cancelled) {
 		xen_arch_resume();
 		xs_resume();
+		dpm_resume_end(PMSG_RESUME);
 	} else
 		xs_suspend_cancel();
 
-	dpm_resume_end(PMSG_RESUME);
-
 	/* Make sure timer events get retriggered on all CPUs */
 	clock_was_set();
 

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xensource.com/xen-devel

#3 lore raw
Quoted 19 lines · Kenji Wakamiya

Hi Konrad, thank you for your response,

Quoted 10 lines · Konrad Rzeszutek Wilk
Quoted 6 lines · Kenji Wakamiya

Hi,

I'm investigating the issue of "xm save -c" in case of PV guests. Then, I tried to prevent calling dpm_resume_end() when suspend was canceled. It seems to work fine about blk and net.
How about this?

Could you give more details of what the failure is?

"xm save -c" saves domain state to storage and leaves it running after creating snapshot. When this command is executed for a pvops guest (xen/stable-2.6.32.x), the guest will hang up after accessing to devices.

Quoted 5 lines

When the guest is unapused, stop_machine() in manage.c returns as the suspend is cancelled. In that case, I think dpm_resume_end() should not be called after stop_machine().

I tested vbd and net, the guest did not hang.

OK. Does 'xm save' (so no -c) and then resume work with this patch (I would think so, but I am curious whether you tested this).

Quoted 1 line

But I'm not sure if this is a right way...

Looks OK to me, just want to make sure that the normal 'xm save' still works after this.

#4 lore raw

Hi Konrad,

Quoted 6 lines · Konrad Rzeszutek Wilk
Quoted 5 lines

When the guest is unapused, stop_machine() in manage.c returns as the suspend is cancelled. In that case, I think dpm_resume_end() should not be called after stop_machine().

I tested vbd and net, the guest did not hang.

Quoted 7 lines

OK. Does 'xm save' (so no -c) and then resume work with this patch (I would think so, but I am curious whether you tested this).

Quoted 1 line

But I'm not sure if this is a right way...

Looks OK to me, just want to make sure that the normal 'xm save' still works after this.

I tested normal 'xm save' with this patch, and made sure it works well.

When normal save/restore is performed, the domain is suspended without cancel, unlike 'save -c'. So, dpm_resume_end() is called. This scenario haven't changed even after applying this patch.

Thanks,

Signature · 1 line
Kenji Wakamiya
#5 lore raw
Quoted 14 lines · Kenji Wakamiya

Hi Konrad,

Quoted 11 lines · Konrad Rzeszutek Wilk
Quoted 5 lines

When the guest is unapused, stop_machine() in manage.c returns as the suspend is cancelled. In that case, I think dpm_resume_end() should not be called after stop_machine().

I tested vbd and net, the guest did not hang.

OK. Does 'xm save' (so no -c) and then resume work with this patch (I would think so, but I am curious whether you tested this).

Quoted 1 line

But I'm not sure if this is a right way...

Quoted 8 lines
Quoted 2 lines

Looks OK to me, just want to make sure that the normal 'xm save' still works after this.

I tested normal 'xm save' with this patch, and made sure it works well.

When normal save/restore is performed, the domain is suspended without cancel, unlike 'save -c'. So, dpm_resume_end() is called. This scenario haven't changed even after applying this patch.

With this change how is the effect of dpm_suspend_start undone in the suspend cancelled case?

Currently we have
dpm_suspend_start(PMSG_SUSPEND)
xs_suspend
dpm_suspend_noirq(PMSG_SUSPEND)
SUSPEND
dpm_resume_noirq(PMSG_RESUME)
xs_resume or xs_supend_cancel
dpm_resume_end(PMSG_RESUME)

Which seems nicely nested and logical but by only calling dpm_resume_end in the non-cancelled case we seem to be unbalancing things.

Do we need some sort of dpm_resume_cancel, or some way of pushing the cancelled flag down into the individual xenbus_device.resume handlers?

Should we maybe simply be using a difference PMSG_XXX in the cancelled case? Is this what one of PMSG_RESTORE or PMSG_RECOVER means?

Looks like to propagate the PMSG_* to the actual device resume functions we would need to provide a pm_ops for the struct bus xenbus_frontend instead of relying on the legacy handlers. This is probably a independently good idea anyway.

Ian.

#6 lore raw
Quoted 42 lines · Ian Campbell
Quoted 14 lines · Kenji Wakamiya

Hi Konrad,

Quoted 11 lines · Konrad Rzeszutek Wilk
Quoted 5 lines

When the guest is unapused, stop_machine() in manage.c returns as the suspend is cancelled. In that case, I think dpm_resume_end() should not be called after stop_machine().

I tested vbd and net, the guest did not hang.

OK. Does 'xm save' (so no -c) and then resume work with this patch (I would think so, but I am curious whether you tested this).

Quoted 1 line

But I'm not sure if this is a right way...

Quoted 8 lines
Quoted 2 lines

Looks OK to me, just want to make sure that the normal 'xm save' still works after this.

I tested normal 'xm save' with this patch, and made sure it works well.

When normal save/restore is performed, the domain is suspended without cancel, unlike 'save -c'. So, dpm_resume_end() is called. This scenario haven't changed even after applying this patch.

With this change how is the effect of dpm_suspend_start undone in the suspend cancelled case?

Currently we have
dpm_suspend_start(PMSG_SUSPEND)
xs_suspend
dpm_suspend_noirq(PMSG_SUSPEND)
SUSPEND
dpm_resume_noirq(PMSG_RESUME)
xs_resume or xs_supend_cancel
dpm_resume_end(PMSG_RESUME)

Which seems nicely nested and logical but by only calling dpm_resume_end in the non-cancelled case we seem to be unbalancing things.

Do we need some sort of dpm_resume_cancel, or some way of pushing the cancelled flag down into the individual xenbus_device.resume handlers?

Quoted 7 lines

Should we maybe simply be using a difference PMSG_XXX in the cancelled case? Is this what one of PMSG_RESTORE or PMSG_RECOVER means?

Looks like to propagate the PMSG_* to the actual device resume functions we would need to provide a pm_ops for the struct bus xenbus_frontend instead of relying on the legacy handlers. This is probably a independently good idea anyway.

ping?

Kenji any ideas or patches to address Ian's comments?

#7 lore raw

Hi Konrand, and sorry for very late response.

Quoted 21 lines · Konrad Rzeszutek Wilk
Quoted 21 lines

With this change how is the effect of dpm_suspend_start undone in the suspend cancelled case?

Currently we have
dpm_suspend_start(PMSG_SUSPEND)
xs_suspend
dpm_suspend_noirq(PMSG_SUSPEND)
SUSPEND
dpm_resume_noirq(PMSG_RESUME)
xs_resume or xs_supend_cancel
dpm_resume_end(PMSG_RESUME)

Which seems nicely nested and logical but by only calling dpm_resume_end in the non-cancelled case we seem to be unbalancing things.

Do we need some sort of dpm_resume_cancel, or some way of pushing the cancelled flag down into the individual xenbus_device.resume handlers?

Should we maybe simply be using a difference PMSG_XXX in the cancelled case? Is this what one of PMSG_RESTORE or PMSG_RECOVER means?

Quoted 8 lines
Quoted 4 lines

Looks like to propagate the PMSG_* to the actual device resume functions we would need to provide a pm_ops for the struct bus xenbus_frontend instead of relying on the legacy handlers. This is probably a independently good idea anyway.

ping?

Kenji any ideas or patches to address Ian's comments?

My colleague made a patch which reflected Ian's comments, so I will ask him to post it. Please wait a little.

Thanks,
Kenji

#8 lore raw

Hi,

The following patch fixes 'xm save -c' issue.
We defined 'PMSG_CANCEL' message for suspend cancel situation and suspend_cancel handler in pm_ops struct.
If the suspend_cancel is defined, suspend_cancel() is called instead of resume().

Thanks,
KAZ

From: Kenji Wakamiya <[email protected]>
Subject: Re: [Xen-devel] [PATCH] pvops: fix "xm save -c" issue
Date: Fri, 21 Jan 2011 14:35:09 +0900

Quoted 36 lines

Hi Konrand, and sorry for very late response.

Quoted 21 lines · Konrad Rzeszutek Wilk
Quoted 21 lines

With this change how is the effect of dpm_suspend_start undone in the suspend cancelled case?

Currently we have
dpm_suspend_start(PMSG_SUSPEND)
xs_suspend
dpm_suspend_noirq(PMSG_SUSPEND)
SUSPEND
dpm_resume_noirq(PMSG_RESUME)
xs_resume or xs_supend_cancel
dpm_resume_end(PMSG_RESUME)

Which seems nicely nested and logical but by only calling dpm_resume_end in the non-cancelled case we seem to be unbalancing things.

Do we need some sort of dpm_resume_cancel, or some way of pushing the cancelled flag down into the individual xenbus_device.resume handlers?

Should we maybe simply be using a difference PMSG_XXX in the cancelled case? Is this what one of PMSG_RESTORE or PMSG_RECOVER means?

Quoted 8 lines
Quoted 4 lines

Looks like to propagate the PMSG_* to the actual device resume functions we would need to provide a pm_ops for the struct bus xenbus_frontend instead of relying on the legacy handlers. This is probably a independently good idea anyway.

ping?

Kenji any ideas or patches to address Ian's comments?

My colleague made a patch which reflected Ian's comments, so I will ask him to post it. Please wait a little.

Quoted 8 lines

Thanks,
Kenji

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xensource.com/xen-devel

drivers/base/power/main.c +9 -0
index 8aa2443..b743e8f 100644
@@ -181,6 +181,13 @@ static int pm_op(struct device *dev,
 			suspend_report_result(ops->suspend, error);
 		}
 		break;
+	case PM_EVENT_CANCEL:
+		if (ops->suspend_cancel) {
+			error = ops->suspend_cancel(dev);
+			suspend_report_result(ops->suspend_cancel, error);
+			break;
+		}
+		/* Fall through */
 	case PM_EVENT_RESUME:
 		if (ops->resume) {
 			error = ops->resume(dev);
@@ -291,6 +298,8 @@ static char *pm_verb(int event)
 	switch (event) {
 	case PM_EVENT_SUSPEND:
 		return "suspend";
+	case PM_EVENT_CANCEL:
+		return "cancel";
 	case PM_EVENT_RESUME:
 		return "resume";
 	case PM_EVENT_FREEZE:
drivers/net/xen-netfront.c +1 -1
index 3f71199..22c6288 100644
@@ -1293,7 +1293,7 @@ static void xennet_disconnect_backend(struct netfront_info *info)
 	info->rx.sring = NULL;
 }
 
-static int netfront_suspend(struct xenbus_device *dev, pm_message_t state)
+static int netfront_suspend(struct xenbus_device *dev)
 {
 	struct netfront_info *info = dev_get_drvdata(&dev->dev);
 	struct hrtimer *timer = &info->smart_poll.timer;
drivers/xen/manage.c +4 -3
index 0b50906..845afb8 100644
@@ -148,10 +148,11 @@ out_resume:
 	if (!cancelled) {
 		xen_arch_resume();
 		xs_resume();
-	} else
+		dpm_resume_end(PMSG_RESUME);
+	} else {
 		xs_suspend_cancel();
-
-	dpm_resume_end(PMSG_RESUME);
+		dpm_resume_end(PMSG_CANCEL);
+	}
 
 	/* Make sure timer events get retriggered on all CPUs */
 	clock_was_set();
drivers/xen/xenbus/xenbus_probe.c +10 -2
index 3a83ba2..019337a 100644
@@ -575,7 +575,7 @@ void xenbus_dev_changed(const char *node, struct xen_bus_type *bus)
 }
 EXPORT_SYMBOL_GPL(xenbus_dev_changed);
 
-int xenbus_dev_suspend(struct device *dev, pm_message_t state)
+int xenbus_dev_suspend(struct device *dev)
 {
 	int err = 0;
 	struct xenbus_driver *drv;
@@ -587,7 +587,7 @@ int xenbus_dev_suspend(struct device *dev, pm_message_t state)
 		return 0;
 	drv = to_xenbus_driver(dev->driver);
 	if (drv->suspend)
-		err = drv->suspend(xdev, state);
+		err = drv->suspend(xdev);
 	if (err)
 		printk(KERN_WARNING
 		       "xenbus: suspend %s failed: %i\n", dev_name(dev), err);
@@ -595,6 +595,14 @@ int xenbus_dev_suspend(struct device *dev, pm_message_t state)
 }
 EXPORT_SYMBOL_GPL(xenbus_dev_suspend);
 
+int xenbus_dev_suspend_cancel(struct device *dev)
+{
+	/* Do nothing */
+	DPRINTK("cancel");
+	return 0;
+}
+EXPORT_SYMBOL_GPL(xenbus_dev_suspend_cancel);
+
 int xenbus_dev_resume(struct device *dev)
 {
 	int err;
drivers/xen/xenbus/xenbus_probe.h +2 -1
index 0e5fc4c..bf4a793 100644
@@ -62,7 +62,8 @@ extern void xenbus_dev_changed(const char *node, struct xen_bus_type *bus);
 
 extern void xenbus_dev_shutdown(struct device *_dev);
 
-extern int xenbus_dev_suspend(struct device *dev, pm_message_t state);
+extern int xenbus_dev_suspend(struct device *dev);
+extern int xenbus_dev_suspend_cancel(struct device *dev);
 extern int xenbus_dev_resume(struct device *dev);
 
 extern void xenbus_otherend_changed(struct xenbus_watch *watch,
drivers/xen/xenbus/xenbus_probe_frontend.c +6 -2
index 5413248..928fc10 100644
@@ -82,6 +82,11 @@ static struct device_attribute xenbus_frontend_dev_attrs[] = {
 	__ATTR_NULL
 };
 
+static struct dev_pm_ops xenbus_pm_ops = {
+	.suspend        = xenbus_dev_suspend,
+	.suspend_cancel = xenbus_dev_suspend_cancel,
+	.resume         = xenbus_dev_resume,
+};
 
 static struct xen_bus_type xenbus_frontend = {
 	.root = "device",
@@ -98,8 +103,7 @@ static struct xen_bus_type xenbus_frontend = {
 		.shutdown = xenbus_dev_shutdown,
 		.dev_attrs= xenbus_frontend_dev_attrs,
 
-		.suspend  = xenbus_dev_suspend,
-		.resume   = xenbus_dev_resume,
+		.pm       = &xenbus_pm_ops,
 	},
 };
 
include/linux/pm.h +3 -0
index 3b7e04b..aa89839 100644
@@ -197,6 +197,7 @@ struct dev_pm_ops {
 	int (*prepare)(struct device *dev);
 	void (*complete)(struct device *dev);
 	int (*suspend)(struct device *dev);
+	int (*suspend_cancel)(struct device *dev);
 	int (*resume)(struct device *dev);
 	int (*freeze)(struct device *dev);
 	int (*thaw)(struct device *dev);
@@ -291,6 +292,7 @@ struct dev_pm_ops name = { \
 #define PM_EVENT_USER		0x0100
 #define PM_EVENT_REMOTE		0x0200
 #define PM_EVENT_AUTO		0x0400
+#define PM_EVENT_CANCEL		0x0800
 
 #define PM_EVENT_SLEEP		(PM_EVENT_SUSPEND | PM_EVENT_HIBERNATE)
 #define PM_EVENT_USER_SUSPEND	(PM_EVENT_USER | PM_EVENT_SUSPEND)
@@ -308,6 +310,7 @@ struct dev_pm_ops name = { \
 #define PMSG_THAW	((struct pm_message){ .event = PM_EVENT_THAW, })
 #define PMSG_RESTORE	((struct pm_message){ .event = PM_EVENT_RESTORE, })
 #define PMSG_RECOVER	((struct pm_message){ .event = PM_EVENT_RECOVER, })
+#define PMSG_CANCEL	((struct pm_message){ .event = PM_EVENT_CANCEL, })
 #define PMSG_USER_SUSPEND	((struct pm_message) \
 					{ .event = PM_EVENT_USER_SUSPEND, })
 #define PMSG_USER_RESUME	((struct pm_message) \
include/xen/xenbus.h +1 -1
index 542ca7c..23e7f25 100644
@@ -91,7 +91,7 @@ struct xenbus_driver {
 	void (*otherend_changed)(struct xenbus_device *dev,
 				 enum xenbus_state backend_state);
 	int (*remove)(struct xenbus_device *dev);
-	int (*suspend)(struct xenbus_device *dev, pm_message_t state);
+	int (*suspend)(struct xenbus_device *dev);
 	int (*resume)(struct xenbus_device *dev);
 	int (*uevent)(struct xenbus_device *, struct kobj_uevent_env *);
 	struct device_driver driver;

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xensource.com/xen-devel

#9 lore raw
Quoted 7 lines · Kazuhiro

Hi,

The following patch fixes 'xm save -c' issue.
We defined 'PMSG_CANCEL' message for suspend cancel situation and suspend_cancel handler in pm_ops struct.
If the suspend_cancel is defined, suspend_cancel() is called instead of resume().

Thanks. I like the general shape of this patch but the core pm infrastructure change needs to be split out and sent upstream via the power management maintainer. I think this is Rafael J. Wysocki <[email protected]> and the [email protected] list.

If you also split out the Xen conversion to dev_pm_ops, without the suspend_cancel bit, then I think we can take that bit now and then you can resend the bit which adds the suspend_cancel bits once the core stuff is upstream.

Ian.

Quoted 56 lines

Thanks,
KAZ

From: Kenji Wakamiya <[email protected]>
Subject: Re: [Xen-devel] [PATCH] pvops: fix "xm save -c" issue
Date: Fri, 21 Jan 2011 14:35:09 +0900

Quoted 44 lines

Hi Konrand, and sorry for very late response.

Quoted 21 lines · Konrad Rzeszutek Wilk
Quoted 21 lines

With this change how is the effect of dpm_suspend_start undone in the suspend cancelled case?

Currently we have
dpm_suspend_start(PMSG_SUSPEND)
xs_suspend
dpm_suspend_noirq(PMSG_SUSPEND)
SUSPEND
dpm_resume_noirq(PMSG_RESUME)
xs_resume or xs_supend_cancel
dpm_resume_end(PMSG_RESUME)

Which seems nicely nested and logical but by only calling dpm_resume_end in the non-cancelled case we seem to be unbalancing things.

Do we need some sort of dpm_resume_cancel, or some way of pushing the cancelled flag down into the individual xenbus_device.resume handlers?

Should we maybe simply be using a difference PMSG_XXX in the cancelled case? Is this what one of PMSG_RESTORE or PMSG_RECOVER means?

Quoted 8 lines
Quoted 4 lines

Looks like to propagate the PMSG_* to the actual device resume functions we would need to provide a pm_ops for the struct bus xenbus_frontend instead of relying on the legacy handlers. This is probably a independently good idea anyway.

ping?

Kenji any ideas or patches to address Ian's comments?

My colleague made a patch which reflected Ian's comments, so I will ask him to post it. Please wait a little.

Thanks,
Kenji

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xensource.com/xen-devel

#10 lore raw

Hi Ian,

I split the patch into pm infrastructure part and xen part. Actually, I've never post to linux community, so I'm not sure what to do.

Thanks,
KAZ

From: Ian Campbell <[email protected]>
Subject: Re: [Xen-devel] [PATCH] pvops: fix "xm save -c" issue
Date: Fri, 21 Jan 2011 09:14:30 +0000

Quoted 14 lines · Kazuhiro
Quoted 7 lines

Hi,

The following patch fixes 'xm save -c' issue.
We defined 'PMSG_CANCEL' message for suspend cancel situation and suspend_cancel handler in pm_ops struct.
If the suspend_cancel is defined, suspend_cancel() is called instead of resume().

Thanks. I like the general shape of this patch but the core pm infrastructure change needs to be split out and sent upstream via the power management maintainer. I think this is Rafael J. Wysocki <[email protected]> and the [email protected] list.

Quoted 65 lines

If you also split out the Xen conversion to dev_pm_ops, without the suspend_cancel bit, then I think we can take that bit now and then you can resend the bit which adds the suspend_cancel bits once the core stuff is upstream.

Ian.

Quoted 56 lines

Thanks,
KAZ

From: Kenji Wakamiya <[email protected]>
Subject: Re: [Xen-devel] [PATCH] pvops: fix "xm save -c" issue
Date: Fri, 21 Jan 2011 14:35:09 +0900

Quoted 44 lines

Hi Konrand, and sorry for very late response.

Quoted 21 lines · Konrad Rzeszutek Wilk
Quoted 21 lines

With this change how is the effect of dpm_suspend_start undone in the suspend cancelled case?

Currently we have
dpm_suspend_start(PMSG_SUSPEND)
xs_suspend
dpm_suspend_noirq(PMSG_SUSPEND)
SUSPEND
dpm_resume_noirq(PMSG_RESUME)
xs_resume or xs_supend_cancel
dpm_resume_end(PMSG_RESUME)

Which seems nicely nested and logical but by only calling dpm_resume_end in the non-cancelled case we seem to be unbalancing things.

Do we need some sort of dpm_resume_cancel, or some way of pushing the cancelled flag down into the individual xenbus_device.resume handlers?

Should we maybe simply be using a difference PMSG_XXX in the cancelled case? Is this what one of PMSG_RESTORE or PMSG_RECOVER means?

Quoted 8 lines
Quoted 4 lines

Looks like to propagate the PMSG_* to the actual device resume functions we would need to provide a pm_ops for the struct bus xenbus_frontend instead of relying on the legacy handlers. This is probably a independently good idea anyway.

ping?

Kenji any ideas or patches to address Ian's comments?

My colleague made a patch which reflected Ian's comments, so I will ask him to post it. Please wait a little.

Thanks,
Kenji

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xensource.com/xen-devel

drivers/base/power/main.c +9 -0
index 8aa2443..b743e8f 100644
@@ -181,6 +181,13 @@ static int pm_op(struct device *dev,
 			suspend_report_result(ops->suspend, error);
 		}
 		break;
+	case PM_EVENT_CANCEL:
+		if (ops->suspend_cancel) {
+			error = ops->suspend_cancel(dev);
+			suspend_report_result(ops->suspend_cancel, error);
+			break;
+		}
+		/* Fall through */
 	case PM_EVENT_RESUME:
 		if (ops->resume) {
 			error = ops->resume(dev);
@@ -291,6 +298,8 @@ static char *pm_verb(int event)
 	switch (event) {
 	case PM_EVENT_SUSPEND:
 		return "suspend";
+	case PM_EVENT_CANCEL:
+		return "cancel";
 	case PM_EVENT_RESUME:
 		return "resume";
 	case PM_EVENT_FREEZE:
include/linux/pm.h +3 -0
index 3b7e04b..aa89839 100644
@@ -197,6 +197,7 @@ struct dev_pm_ops {
 	int (*prepare)(struct device *dev);
 	void (*complete)(struct device *dev);
 	int (*suspend)(struct device *dev);
+	int (*suspend_cancel)(struct device *dev);
 	int (*resume)(struct device *dev);
 	int (*freeze)(struct device *dev);
 	int (*thaw)(struct device *dev);
@@ -291,6 +292,7 @@ struct dev_pm_ops name = { \
 #define PM_EVENT_USER		0x0100
 #define PM_EVENT_REMOTE		0x0200
 #define PM_EVENT_AUTO		0x0400
+#define PM_EVENT_CANCEL		0x0800
 
 #define PM_EVENT_SLEEP		(PM_EVENT_SUSPEND | PM_EVENT_HIBERNATE)
 #define PM_EVENT_USER_SUSPEND	(PM_EVENT_USER | PM_EVENT_SUSPEND)
@@ -308,6 +310,7 @@ struct dev_pm_ops name = { \
 #define PMSG_THAW	((struct pm_message){ .event = PM_EVENT_THAW, })
 #define PMSG_RESTORE	((struct pm_message){ .event = PM_EVENT_RESTORE, })
 #define PMSG_RECOVER	((struct pm_message){ .event = PM_EVENT_RECOVER, })
+#define PMSG_CANCEL	((struct pm_message){ .event = PM_EVENT_CANCEL, })
 #define PMSG_USER_SUSPEND	((struct pm_message) \
 					{ .event = PM_EVENT_USER_SUSPEND, })
 #define PMSG_USER_RESUME	((struct pm_message) \
drivers/net/xen-netfront.c +1 -1
index 3f71199..22c6288 100644
@@ -1293,7 +1293,7 @@ static void xennet_disconnect_backend(struct netfront_info *info)
 	info->rx.sring = NULL;
 }
 
-static int netfront_suspend(struct xenbus_device *dev, pm_message_t state)
+static int netfront_suspend(struct xenbus_device *dev)
 {
 	struct netfront_info *info = dev_get_drvdata(&dev->dev);
 	struct hrtimer *timer = &info->smart_poll.timer;
drivers/xen/manage.c +4 -3
index 0b50906..845afb8 100644
@@ -148,10 +148,11 @@ out_resume:
 	if (!cancelled) {
 		xen_arch_resume();
 		xs_resume();
-	} else
+		dpm_resume_end(PMSG_RESUME);
+	} else {
 		xs_suspend_cancel();
-
-	dpm_resume_end(PMSG_RESUME);
+		dpm_resume_end(PMSG_CANCEL);
+	}
 
 	/* Make sure timer events get retriggered on all CPUs */
 	clock_was_set();
drivers/xen/xenbus/xenbus_probe.c +10 -2
index 3a83ba2..019337a 100644
@@ -575,7 +575,7 @@ void xenbus_dev_changed(const char *node, struct xen_bus_type *bus)
 }
 EXPORT_SYMBOL_GPL(xenbus_dev_changed);
 
-int xenbus_dev_suspend(struct device *dev, pm_message_t state)
+int xenbus_dev_suspend(struct device *dev)
 {
 	int err = 0;
 	struct xenbus_driver *drv;
@@ -587,7 +587,7 @@ int xenbus_dev_suspend(struct device *dev, pm_message_t state)
 		return 0;
 	drv = to_xenbus_driver(dev->driver);
 	if (drv->suspend)
-		err = drv->suspend(xdev, state);
+		err = drv->suspend(xdev);
 	if (err)
 		printk(KERN_WARNING
 		       "xenbus: suspend %s failed: %i\n", dev_name(dev), err);
@@ -595,6 +595,14 @@ int xenbus_dev_suspend(struct device *dev, pm_message_t state)
 }
 EXPORT_SYMBOL_GPL(xenbus_dev_suspend);
 
+int xenbus_dev_suspend_cancel(struct device *dev)
+{
+	/* Do nothing */
+	DPRINTK("cancel");
+	return 0;
+}
+EXPORT_SYMBOL_GPL(xenbus_dev_suspend_cancel);
+
 int xenbus_dev_resume(struct device *dev)
 {
 	int err;
drivers/xen/xenbus/xenbus_probe.h +2 -1
index 0e5fc4c..bf4a793 100644
@@ -62,7 +62,8 @@ extern void xenbus_dev_changed(const char *node, struct xen_bus_type *bus);
 
 extern void xenbus_dev_shutdown(struct device *_dev);
 
-extern int xenbus_dev_suspend(struct device *dev, pm_message_t state);
+extern int xenbus_dev_suspend(struct device *dev);
+extern int xenbus_dev_suspend_cancel(struct device *dev);
 extern int xenbus_dev_resume(struct device *dev);
 
 extern void xenbus_otherend_changed(struct xenbus_watch *watch,
drivers/xen/xenbus/xenbus_probe_frontend.c +6 -2
index 5413248..928fc10 100644
@@ -82,6 +82,11 @@ static struct device_attribute xenbus_frontend_dev_attrs[] = {
 	__ATTR_NULL
 };
 
+static struct dev_pm_ops xenbus_pm_ops = {
+	.suspend        = xenbus_dev_suspend,
+	.suspend_cancel = xenbus_dev_suspend_cancel,
+	.resume         = xenbus_dev_resume,
+};
 
 static struct xen_bus_type xenbus_frontend = {
 	.root = "device",
@@ -98,8 +103,7 @@ static struct xen_bus_type xenbus_frontend = {
 		.shutdown = xenbus_dev_shutdown,
 		.dev_attrs= xenbus_frontend_dev_attrs,
 
-		.suspend  = xenbus_dev_suspend,
-		.resume   = xenbus_dev_resume,
+		.pm       = &xenbus_pm_ops,
 	},
 };
 
include/xen/xenbus.h +1 -1
index 542ca7c..23e7f25 100644
@@ -91,7 +91,7 @@ struct xenbus_driver {
 	void (*otherend_changed)(struct xenbus_device *dev,
 				 enum xenbus_state backend_state);
 	int (*remove)(struct xenbus_device *dev);
-	int (*suspend)(struct xenbus_device *dev, pm_message_t state);
+	int (*suspend)(struct xenbus_device *dev);
 	int (*resume)(struct xenbus_device *dev);
 	int (*uevent)(struct xenbus_device *, struct kobj_uevent_env *);
 	struct device_driver driver;

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xensource.com/xen-devel

#11 lore raw
Quoted 5 lines · Kazuhiro

Hi Ian,

I split the patch into pm infrastructure part and xen part. Actually, I've never post to linux community, so I'm not sure what to do.

Documentation/SubmittingPatches in the Linux source tree has some good guidance.

In general you need to post your patches inline (not as attachments which is a difference from the Xen maintainer's policy) with a changelog message, in this case explaining the need for the new event type and explaining the use cases etc, and a signed-off-by.

You should send to the subsystem maintainer and the relevant mailing list, the MAINTAINERS file in the Linux source would help figure out who they are but in this case I think it is Rafael J. Wysocki <[email protected]> and the [email protected] list. You should cc xen-devel as well.

It will be interesting to see what upstream make of the use case. I suspect there might be some common ground with the ability to abort a hibernation attempt on native for example.

Ian.

Quoted 91 lines

Thanks,
KAZ

From: Ian Campbell <[email protected]>
Subject: Re: [Xen-devel] [PATCH] pvops: fix "xm save -c" issue
Date: Fri, 21 Jan 2011 09:14:30 +0000

Quoted 79 lines · Kazuhiro
Quoted 7 lines

Hi,

The following patch fixes 'xm save -c' issue.
We defined 'PMSG_CANCEL' message for suspend cancel situation and suspend_cancel handler in pm_ops struct.
If the suspend_cancel is defined, suspend_cancel() is called instead of resume().

Thanks. I like the general shape of this patch but the core pm infrastructure change needs to be split out and sent upstream via the power management maintainer. I think this is Rafael J. Wysocki <[email protected]> and the [email protected] list.

If you also split out the Xen conversion to dev_pm_ops, without the suspend_cancel bit, then I think we can take that bit now and then you can resend the bit which adds the suspend_cancel bits once the core stuff is upstream.

Ian.

Quoted 56 lines

Thanks,
KAZ

From: Kenji Wakamiya <[email protected]>
Subject: Re: [Xen-devel] [PATCH] pvops: fix "xm save -c" issue
Date: Fri, 21 Jan 2011 14:35:09 +0900

Quoted 44 lines

Hi Konrand, and sorry for very late response.

Quoted 21 lines · Konrad Rzeszutek Wilk
Quoted 21 lines

With this change how is the effect of dpm_suspend_start undone in the suspend cancelled case?

Currently we have
dpm_suspend_start(PMSG_SUSPEND)
xs_suspend
dpm_suspend_noirq(PMSG_SUSPEND)
SUSPEND
dpm_resume_noirq(PMSG_RESUME)
xs_resume or xs_supend_cancel
dpm_resume_end(PMSG_RESUME)

Which seems nicely nested and logical but by only calling dpm_resume_end in the non-cancelled case we seem to be unbalancing things.

Do we need some sort of dpm_resume_cancel, or some way of pushing the cancelled flag down into the individual xenbus_device.resume handlers?

Should we maybe simply be using a difference PMSG_XXX in the cancelled case? Is this what one of PMSG_RESTORE or PMSG_RECOVER means?

Quoted 8 lines
Quoted 4 lines

Looks like to propagate the PMSG_* to the actual device resume functions we would need to provide a pm_ops for the struct bus xenbus_frontend instead of relying on the legacy handlers. This is probably a independently good idea anyway.

ping?

Kenji any ideas or patches to address Ian's comments?

My colleague made a patch which reflected Ian's comments, so I will ask him to post it. Please wait a little.

Thanks,
Kenji

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xensource.com/xen-devel