Skip to contentSkip to conversation

gpio-vbus: support disabling D+ pullup on suspend

PATCH Superseded Dmitry Eremin-Solenikov 5 participants 19 messages Jun 22, 2011 – Jun 28, 2011

Subsystems (4)

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/2 gpio-vbus: support disabling D+ pullup on suspend

Some platforms would like to disable D+ pullup on suspend, to drain as low power, as possible. E.g. this was requested by mioa701 board maintainers.

---
 drivers/usb/otg/gpio_vbus.c   |   32 ++++++++++++++++++++++++++++++++
 include/linux/usb/gpio_vbus.h |    1 +
 2 files changed, 33 insertions(+), 0 deletions(-)
drivers/usb/otg/gpio_vbus.c +32 -0
index 52733d9..44527bd 100644
@@ -327,6 +327,34 @@ static int __exit gpio_vbus_remove(struct platform_device *pdev)
 	return 0;
 }
 
+#ifdef CONFIG_PM
+static int gpio_vbus_suspend(struct platform_device *pdev, pm_message_t state)
+{
+	struct gpio_vbus_data *gpio_vbus = platform_get_drvdata(pdev);
+	struct gpio_vbus_mach_info *pdata = gpio_vbus->dev->platform_data;
+
+	if (gpio_vbus->otg.gadget && pdata->disconnect_on_suspend) {
+		/* optionally disable D+ pullup */
+		if (gpio_is_valid(pdata->gpio_pullup))
+			gpio_set_value(pdata->gpio_pullup,
+					pdata->gpio_pullup_inverted);
+
+		set_vbus_draw(gpio_vbus, 0);
+	}
+	return 0;
+}
+
+static int gpio_vbus_resume(struct platform_device *pdev)
+{
+	struct gpio_vbus_data *gpio_vbus = platform_get_drvdata(pdev);
+
+	if (gpio_vbus->otg.gadget)
+		schedule_work(&gpio_vbus->work);
+
+	return 0;
+}
+#endif
+
 /* NOTE:  the gpio-vbus device may *NOT* be hotplugged */
 
 MODULE_ALIAS("platform:gpio-vbus");
@@ -337,6 +365,10 @@ static struct platform_driver gpio_vbus_driver = {
 		.owner = THIS_MODULE,
 	},
 	.remove  = __exit_p(gpio_vbus_remove),
+#ifdef CONFIG_PM
+	.suspend	= gpio_vbus_suspend,
+	.resume		= gpio_vbus_resume
+#endif
 };
 
 static int __init gpio_vbus_init(void)
include/linux/usb/gpio_vbus.h +1 -0
index d9f03cc..2faa38d 100644
@@ -27,4 +27,5 @@ struct gpio_vbus_mach_info {
 	int gpio_pullup;
 	bool gpio_vbus_inverted;
 	bool gpio_pullup_inverted;
+	bool disconnect_on_suspend;
 };
Signature · 1 line
1.7.5.3
#1 lore raw

PATCH 2/2 mioa701: move gpio-pullup functionality to gpio-vbus

gpio-vbus can (and should?) handle D+ pullup gpio for us. Move gpio-pullup handling to gpio-vbus and stop providing udc_info as it's empty now.

---
 arch/arm/mach-pxa/mioa701.c |    7 +------
 1 files changed, 1 insertions(+), 6 deletions(-)
arch/arm/mach-pxa/mioa701.c +1 -6
index bcbd359..e712c9d 100644
@@ -402,14 +402,10 @@ static void gsm_exit(void)
 /*
  * USB UDC
  */
-static struct pxa2xx_udc_mach_info mioa701_udc_info = {
-	.gpio_pullup	  = GPIO22_USB_ENABLE,
-};
-
 struct gpio_vbus_mach_info gpio_vbus_data = {
 	.gpio_vbus = GPIO13_nUSB_DETECT,
 	.gpio_vbus_inverted = 1,
-	.gpio_pullup = -1,
+	.gpio_pullup = GPIO22_USB_ENABLE,
 };
 
 /*
@@ -770,7 +766,6 @@ static void __init mioa701_machine_init(void)
 	pxa_set_fb_info(NULL, &mioa701_pxafb_info);
 	pxa_set_mci_info(&mioa701_mci_info);
 	pxa_set_keypad_info(&mioa701_keypad_info);
-	pxa_set_udc_info(&mioa701_udc_info);
 	pxa_set_ac97_info(&mioa701_ac97_info);
 	pm_power_off = mioa701_poweroff;
 	arm_pm_restart = mioa701_restart;
Signature · 1 line
1.7.5.3
#2 lore raw

PATCH 2/2 mioa701: move gpio-pullup functionality to gpio-vbus

Oops... See bellow,

Quoted 16 lines · Dmitry Eremin-Solenikov

gpio-vbus can (and should?) handle D+ pullup gpio for us. Move gpio-pullup handling
to gpio-vbus and stop providing udc_info as it's empty now.

---
 arch/arm/mach-pxa/mioa701.c |    7 +------
 1 files changed, 1 insertions(+), 6 deletions(-)
arch/arm/mach-pxa/mioa701.c +0 -0
index bcbd359..e712c9d 100644
@@ -402,14 +402,10 @@ static void gsm_exit(void)
 /*
  * USB UDC

arch/arm/mach-pxa/mioa701.c static void gsm_exit(void) · line 408

Quoted 10 lines
arch/arm/mach-pxa/mioa701.c +1 -5
  */
-static struct pxa2xx_udc_mach_info mioa701_udc_info = {
-	.gpio_pullup	  = GPIO22_USB_ENABLE,
-};
-
 struct gpio_vbus_mach_info gpio_vbus_data = {
 	.gpio_vbus = GPIO13_nUSB_DETECT,
 	.gpio_vbus_inverted = 1,
-	.gpio_pullup = -1,
+	.gpio_pullup = GPIO22_USB_ENABLE,

This should be also
+ .disconnect_on_suspend = 1,

Quoted 15 lines
arch/arm/mach-pxa/mioa701.c +0 -2
 };
 
 /*
@@ -770,7 +766,6 @@ static void __init mioa701_machine_init(void)
 	pxa_set_fb_info(NULL, &mioa701_pxafb_info);
 	pxa_set_mci_info(&mioa701_mci_info);
 	pxa_set_keypad_info(&mioa701_keypad_info);
-	pxa_set_udc_info(&mioa701_udc_info);
 	pxa_set_ac97_info(&mioa701_ac97_info);
 	pm_power_off = mioa701_poweroff;
 	arm_pm_restart = mioa701_restart;
--

1.7.5.3

Signature · 2 lines
With best wishes
Dmitry
#3 lore raw

PATCH 1/2 gpio-vbus: support disabling D+ pullup on suspend

In reply to Dmitry Eremin-Solenikov

Hi,

Quoted 3 lines · Dmitry Eremin-Solenikov

Some platforms would like to disable D+ pullup on suspend, to drain as low power, as possible. E.g. this was requested by mioa701 board maintainers.

I think this makes sense to many platforms, but by doing so, you would loose connection to the Host PC, so you need to make sure your device isn't been used before you go down this road.

Quoted 30 lines
---
 drivers/usb/otg/gpio_vbus.c   |   32 ++++++++++++++++++++++++++++++++
 include/linux/usb/gpio_vbus.h |    1 +
 2 files changed, 33 insertions(+), 0 deletions(-)
drivers/usb/otg/gpio_vbus.c +16 -0
index 52733d9..44527bd 100644
@@ -327,6 +327,34 @@ static int __exit gpio_vbus_remove(struct platform_device *pdev)
 	return 0;
 }
 
+#ifdef CONFIG_PM
+static int gpio_vbus_suspend(struct platform_device *pdev, pm_message_t state)
+{
+	struct gpio_vbus_data *gpio_vbus = platform_get_drvdata(pdev);
+	struct gpio_vbus_mach_info *pdata = gpio_vbus->dev->platform_data;
+
+	if (gpio_vbus->otg.gadget && pdata->disconnect_on_suspend) {
+		/* optionally disable D+ pullup */
+		if (gpio_is_valid(pdata->gpio_pullup))
+			gpio_set_value(pdata->gpio_pullup,
+					pdata->gpio_pullup_inverted);
+
+		set_vbus_draw(gpio_vbus, 0);
+	}
+	return 0;
+}

drivers/usb/otg/gpio_vbus.c static int __exit gpio_vbus_remove(struct platform_device *pdev) · line 355

Quoted 10 lines
drivers/usb/otg/gpio_vbus.c +10 -0
+
+static int gpio_vbus_resume(struct platform_device *pdev)
+{
+	struct gpio_vbus_data *gpio_vbus = platform_get_drvdata(pdev);
+
+	if (gpio_vbus->otg.gadget)
+		schedule_work(&gpio_vbus->work);
+
+	return 0;
+}

actually, the correct way would be to use dev_pm_ops.

drivers/usb/otg/gpio_vbus.c static struct platform_driver gpio_vbus_driver = { · line 371

Quoted 13 lines
drivers/usb/otg/gpio_vbus.c +6 -0
+#endif
+
 /* NOTE:  the gpio-vbus device may *NOT* be hotplugged */
 
 MODULE_ALIAS("platform:gpio-vbus");
@@ -337,6 +365,10 @@ static struct platform_driver gpio_vbus_driver = {
 		.owner = THIS_MODULE,
 	},
 	.remove  = __exit_p(gpio_vbus_remove),
+#ifdef CONFIG_PM
+	.suspend	= gpio_vbus_suspend,
+	.resume		= gpio_vbus_resume
+#endif

also, avoid the ifdef on the driver structure.

Signature · 8 lines
balbi
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 490 bytes
Desc: Digital signature
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20110622/caa12764/attachment.sig>
#4 lore raw

PATCH 1/2 gpio-vbus: support disabling D+ pullup on suspend

Quoted 10 lines · Felipe Balbi

Hi,

Quoted 3 lines · Dmitry Eremin-Solenikov

Some platforms would like to disable D+ pullup on suspend, to drain as low power, as possible. E.g. this was requested by mioa701 board maintainers.

I think this makes sense to many platforms, but by doing so, you would loose connection to the Host PC, so you need to make sure your device isn't been used before you go down this road.

I've thought about this. Should UDC driver should somehow call into OTG layer on suspend? My understanding is that otg_set_suspend isn't the call that should be done here, is it true?

My idea was that board can ask for D+ disabling, knowing itself the behaviour of the platform driver on suspend (e.g. PXA27x does disable UDC on suspend, but I dunno what effect this will cause on Host PC).

Quoted 36 lines
Quoted 36 lines
---
 drivers/usb/otg/gpio_vbus.c   |   32 ++++++++++++++++++++++++++++++++
 include/linux/usb/gpio_vbus.h |    1 +
 2 files changed, 33 insertions(+), 0 deletions(-)
drivers/usb/otg/gpio_vbus.c +20 -0
index 52733d9..44527bd 100644
@@ -327,6 +327,34 @@ static int __exit gpio_vbus_remove(struct
 platform_device *pdev)
 	return 0;
 }
 
+#ifdef CONFIG_PM
+static int gpio_vbus_suspend(struct platform_device *pdev, pm_message_t
 state)
+{
+	struct gpio_vbus_data *gpio_vbus = platform_get_drvdata(pdev);
+	struct gpio_vbus_mach_info *pdata = gpio_vbus->dev->platform_data;
+
+	if (gpio_vbus->otg.gadget && pdata->disconnect_on_suspend) {
+		/* optionally disable D+ pullup */
+		if (gpio_is_valid(pdata->gpio_pullup))
+			gpio_set_value(pdata->gpio_pullup,
+					pdata->gpio_pullup_inverted);
+
+		set_vbus_draw(gpio_vbus, 0);
+	}
+	return 0;
+}
+
+static int gpio_vbus_resume(struct platform_device *pdev)
+{
+	struct gpio_vbus_data *gpio_vbus = platform_get_drvdata(pdev);
Quoted 8 lines

drivers/usb/otg/gpio_vbus.c static int __exit gpio_vbus_remove(struct platform_device *pdev) · line 355

Quoted 6 lines
drivers/usb/otg/gpio_vbus.c +6 -0
+
+	if (gpio_vbus->otg.gadget)
+		schedule_work(&gpio_vbus->work);
+
+	return 0;
+}

actually, the correct way would be to use dev_pm_ops.

Could I use SIMPLE_DEV_PM_OPS here?

Quoted 5 lines
Quoted 5 lines
drivers/usb/otg/gpio_vbus.c +2 -0
+#endif
+
 /* NOTE:  the gpio-vbus device may *NOT* be hotplugged */
Quoted 11 lines

drivers/usb/otg/gpio_vbus.c static struct platform_driver gpio_vbus_driver = { · line 371

Quoted 9 lines

MODULE_ALIAS("platform:gpio-vbus");

drivers/usb/otg/gpio_vbus.c +4 -0
@@ -337,6 +365,10 @@ static struct platform_driver gpio_vbus_driver = {
 		.owner = THIS_MODULE,
 	},
 	.remove  = __exit_p(gpio_vbus_remove),
+#ifdef CONFIG_PM
+	.suspend	= gpio_vbus_suspend,
+	.resume		= gpio_vbus_resume
+#endif

also, avoid the ifdef on the driver structure.

Ack. Was just C&P from gpio-vbus, but it's not an excuse to follow bad style.

Signature · 2 lines
With best wishes
Dmitry
#5 lore raw

PATCH 1/2 gpio-vbus: support disabling D+ pullup on suspend

Hi,

Quoted 12 lines · Dmitry Eremin-Solenikov
Quoted 10 lines · Felipe Balbi

Hi,

Quoted 3 lines · Dmitry Eremin-Solenikov

Some platforms would like to disable D+ pullup on suspend, to drain as low power, as possible. E.g. this was requested by mioa701 board maintainers.

I think this makes sense to many platforms, but by doing so, you would loose connection to the Host PC, so you need to make sure your device isn't been used before you go down this road.

Quoted 7 lines

I've thought about this. Should UDC driver should somehow call into OTG layer on suspend? My understanding is that otg_set_suspend isn't the call that should be done here, is it true?

My idea was that board can ask for D+ disabling, knowing itself the behaviour of the platform driver on suspend (e.g. PXA27x does disable UDC on suspend, but I dunno what effect this will cause on Host PC).

Host PC will only see the device disconnecting. So, what happens if the user has mounted file systems when you decide to go into suspend ?

Quoted 46 lines
Quoted 37 lines
Quoted 37 lines
---
 drivers/usb/otg/gpio_vbus.c   |   32 ++++++++++++++++++++++++++++++++
 include/linux/usb/gpio_vbus.h |    1 +
 2 files changed, 33 insertions(+), 0 deletions(-)
drivers/usb/otg/gpio_vbus.c +20 -0
index 52733d9..44527bd 100644
@@ -327,6 +327,34 @@ static int __exit gpio_vbus_remove(struct
 platform_device *pdev)
 	return 0;
 }
 
+#ifdef CONFIG_PM
+static int gpio_vbus_suspend(struct platform_device *pdev, pm_message_t
 state)
+{
+	struct gpio_vbus_data *gpio_vbus = platform_get_drvdata(pdev);
+	struct gpio_vbus_mach_info *pdata = gpio_vbus->dev->platform_data;
+
+	if (gpio_vbus->otg.gadget && pdata->disconnect_on_suspend) {
+		/* optionally disable D+ pullup */
+		if (gpio_is_valid(pdata->gpio_pullup))
+			gpio_set_value(pdata->gpio_pullup,
+					pdata->gpio_pullup_inverted);
+
+		set_vbus_draw(gpio_vbus, 0);
+	}
+	return 0;
+}
+
+static int gpio_vbus_resume(struct platform_device *pdev)
+{
+	struct gpio_vbus_data *gpio_vbus = platform_get_drvdata(pdev);
Quoted 8 lines

drivers/usb/otg/gpio_vbus.c static int __exit gpio_vbus_remove(struct platform_device *pdev) · line 355

Quoted 6 lines
drivers/usb/otg/gpio_vbus.c +6 -0
+
+	if (gpio_vbus->otg.gadget)
+		schedule_work(&gpio_vbus->work);
+
+	return 0;
+}

actually, the correct way would be to use dev_pm_ops.

Quoted 1 line

Could I use SIMPLE_DEV_PM_OPS here?

for sure ;-)

Quoted 16 lines
Quoted 4 lines
Quoted 4 lines
drivers/usb/otg/gpio_vbus.c +2 -0
+#endif
+
 /* NOTE:  the gpio-vbus device may *NOT* be hotplugged */
Quoted 11 lines

drivers/usb/otg/gpio_vbus.c static struct platform_driver gpio_vbus_driver = { · line 371

Quoted 9 lines

MODULE_ALIAS("platform:gpio-vbus");

drivers/usb/otg/gpio_vbus.c +4 -0
@@ -337,6 +365,10 @@ static struct platform_driver gpio_vbus_driver = {
 		.owner = THIS_MODULE,
 	},
 	.remove  = __exit_p(gpio_vbus_remove),
+#ifdef CONFIG_PM
+	.suspend	= gpio_vbus_suspend,
+	.resume		= gpio_vbus_resume
+#endif

also, avoid the ifdef on the driver structure.

Quoted 2 lines

Ack. Was just C&P from gpio-vbus, but it's not an excuse to follow bad style.

see what e.g the musb driver does to avoid the ifdef (drivers/usb/musb/musb_core.c)

Signature · 8 lines
balbi
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 490 bytes
Desc: Digital signature
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20110622/53a18614/attachment.sig>
#6 lore raw

PATCH 1/2 gpio-vbus: support disabling D+ pullup on suspend

Quoted 19 lines · Felipe Balbi

Hi,

Quoted 16 lines · Dmitry Eremin-Solenikov
Quoted 10 lines · Felipe Balbi

Hi,

Quoted 3 lines · Dmitry Eremin-Solenikov

Some platforms would like to disable D+ pullup on suspend, to drain as low power, as possible. E.g. this was requested by mioa701 board maintainers.

I think this makes sense to many platforms, but by doing so, you would loose connection to the Host PC, so you need to make sure your device isn't been used before you go down this road.

I've thought about this. Should UDC driver should somehow call into OTG layer on suspend? My understanding is that otg_set_suspend isn't the call that should be done here, is it true?

Quoted 8 lines
Quoted 5 lines

My idea was that board can ask for D+ disabling, knowing itself the behaviour
of the platform driver on suspend (e.g. PXA27x does disable UDC on suspend,
but I dunno what effect this will cause on Host PC).

Host PC will only see the device disconnecting. So, what happens if the user has mounted file systems when you decide to go into suspend ?

What happens if user has mounted filesystems when I decide to pull out the cable?
I agree with you generally, but I'd like to hear any suggestions.

Signature · 2 lines
With best wishes
Dmitry
#7 lore raw

PATCH 1/2 gpio-vbus: support disabling D+ pullup on suspend

Hi,

Quoted 26 lines · Dmitry Eremin-Solenikov
Quoted 25 lines · Felipe Balbi

Hi,

Quoted 21 lines · Dmitry Eremin-Solenikov
Quoted 10 lines · Felipe Balbi

Hi,

Quoted 3 lines · Dmitry Eremin-Solenikov

Some platforms would like to disable D+ pullup on suspend, to drain as low power, as possible. E.g. this was requested by mioa701 board maintainers.

I think this makes sense to many platforms, but by doing so, you would loose connection to the Host PC, so you need to make sure your device isn't been used before you go down this road.

I've thought about this. Should UDC driver should somehow call into OTG layer on suspend? My understanding is that otg_set_suspend isn't the call that should be done here, is it true?

My idea was that board can ask for D+ disabling, knowing itself the behaviour
of the platform driver on suspend (e.g. PXA27x does disable UDC on suspend,
but I dunno what effect this will cause on Host PC).

Quoted 5 lines
Quoted 2 lines

Host PC will only see the device disconnecting. So, what happens if the user has mounted file systems when you decide to go into suspend ?

What happens if user has mounted filesystems when I decide to pull out the cable?

for starters you could have filesystem corruption. In short, the best way would be to return -EBUSY in your suspend if the cable is still attached.

You could use osme VBUS IRQ to toggle a driver flag which, if true, would return -EBUSY on suspend().

Quoted 1 line

I agree with you generally, but I'd like to hear any suggestions.

I'm not sure how to solve this, but OTOH the original code already did this, just on a different way, right ?

Signature · 8 lines
balbi
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 490 bytes
Desc: Digital signature
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20110622/7a844ca7/attachment.sig>
#8 lore raw

PATCH 1/2 gpio-vbus: support disabling D+ pullup on suspend

Quoted 32 lines · Felipe Balbi

Hi,

Quoted 29 lines · Dmitry Eremin-Solenikov
Quoted 27 lines · Felipe Balbi

Hi,

Quoted 16 lines · Dmitry Eremin-Solenikov
Quoted 10 lines · Felipe Balbi

Hi,

Quoted 3 lines · Dmitry Eremin-Solenikov

Some platforms would like to disable D+ pullup on suspend, to drain as low power, as possible. E.g. this was requested by mioa701 board maintainers.

I think this makes sense to many platforms, but by doing so, you would loose connection to the Host PC, so you need to make sure your device isn't been used before you go down this road.

I've thought about this. Should UDC driver should somehow call into OTG layer on suspend? My understanding is that otg_set_suspend isn't the call that should be done here, is it true?

Quoted 5 lines

My idea was that board can ask for D+ disabling, knowing itself the behaviour
of the platform driver on suspend (e.g. PXA27x does disable UDC on suspend,
but I dunno what effect this will cause on Host PC).

Host PC will only see the device disconnecting. So, what happens if the user has mounted file systems when you decide to go into suspend ?

Quoted 6 lines
Quoted 2 lines

What happens if user has mounted filesystems when I decide to pull out the cable?

for starters you could have filesystem corruption. In short, the best way would be to return -EBUSY in your suspend if the cable is still attached.

That was a rhetorical question. Basically there are plenty of situations and cases (cable is not attached; cable attached, but no gadget driver; cable attached, block gadget driver, but filesystems aren't mounted; cable attached, block gadget driver , filesystem mounted, but host is also suspended, etc.).

Quoted 2 lines

You could use osme VBUS IRQ to toggle a driver flag which, if true, would return -EBUSY on suspend().

I'm more and more thinking that this handling this -EBUSY isn't a task of gpio-vbus, but rather of some higher level driver. I'd assume that if I hit this point, all previous drivers (which depend on this transceiver, so registered later) permit suspending at this moment, so everything is OK :)

Quoted 5 lines
Quoted 2 lines

I agree with you generally, but I'd like to hear any suggestions.

I'm not sure how to solve this, but OTOH the original code already did this, just on a different way, right ?

Yes. pxa27x udc driver disables D+ pullup on suspend and that's the behaviour asked from me by Robert Jarzmik in comments to first cleanup patch serie for pxa27x UDC driver.

Signature · 2 lines
With best wishes
Dmitry
#9 lore raw

PATCH 1/2 gpio-vbus: support disabling D+ pullup on suspend

Hi,

Quoted 9 lines · Dmitry Eremin-Solenikov
Quoted 2 lines

You could use osme VBUS IRQ to toggle a driver flag which, if true, would return -EBUSY on suspend().

I'm more and more thinking that this handling this -EBUSY isn't a task of gpio-vbus, but rather of some higher level driver. I'd assume that
if I hit this point, all previous drivers (which depend on this transceiver, so registered later) permit suspending at this moment, so everything is OK :)

the thing is that today we don't have the "higher level driver" see that the OTG/transceiver framework (if you can call it a framework) is just a static global pointer which people set. So, you need to have per-transceiver solutions, unfortunately :-(

Quoted 8 lines
Quoted 4 lines
Quoted 1 line

I agree with you generally, but I'd like to hear any suggestions.

I'm not sure how to solve this, but OTOH the original code already did this, just on a different way, right ?

Yes. pxa27x udc driver disables D+ pullup on suspend and that's the behaviour asked from me by Robert Jarzmik in comments to first cleanup patch serie for pxa27x UDC driver.

ok... so, go ahead but keep in mind you could end up in a bad situation ;-)

Signature · 8 lines
balbi
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 490 bytes
Desc: Digital signature
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20110622/50aed419/attachment.sig>
#10 lore raw

PATCH 1/2 gpio-vbus: support disabling D+ pullup on suspend

In reply to Felipe Balbi
Quoted 10 lines · Felipe Balbi

Hi,

Quoted 3 lines · Dmitry Eremin-Solenikov

Some platforms would like to disable D+ pullup on suspend, to drain as low power, as possible. E.g. this was requested by mioa701 board maintainers.

I think this makes sense to many platforms, but by doing so, you would loose connection to the Host PC, so you need to make sure your device isn't been used before you go down this road.

In fact, something like this is _necessary_ for all UDC/PHY drivers unless the device can guarantee that it will automatically wake up from suspend in time to service a USB packet (note that the window for responding to a packet is only a few microseconds). Otherwise the device would appear to the host to be unresponsive and broken -- better to do a clean disconnect.

If suspending the device while it is in use would cause problems ... then don't suspend it when it is in use!

Alan Stern

#11 lore raw

PATCH 1/2 gpio-vbus: support disabling D+ pullup on suspend

Hi,

Quoted 20 lines · Alan Stern
Quoted 10 lines · Felipe Balbi

Hi,

Quoted 3 lines · Dmitry Eremin-Solenikov

Some platforms would like to disable D+ pullup on suspend, to drain as low power, as possible. E.g. this was requested by mioa701 board maintainers.

I think this makes sense to many platforms, but by doing so, you would loose connection to the Host PC, so you need to make sure your device isn't been used before you go down this road.

In fact, something like this is _necessary_ for all UDC/PHY drivers unless the device can guarantee that it will automatically wake up from suspend in time to service a USB packet (note that the window for responding to a packet is only a few microseconds). Otherwise the device would appear to the host to be unresponsive and broken -- better to do a clean disconnect.

Quoted 2 lines

If suspending the device while it is in use would cause problems ... then don't suspend it when it is in use!

I second your thoughts, but today we don't have enough infrastructure to communicate between PHY and Link, so a clean solution isn't possible, right ?

Should we block this patch due to that ?

Signature · 8 lines
balbi
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 490 bytes
Desc: Digital signature
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20110622/555e71af/attachment.sig>
#12 lore raw

PATCH 1/2 gpio-vbus: support disabling D+ pullup on suspend

Quoted 7 lines · Felipe Balbi
Quoted 7 lines

In fact, something like this is _necessary_ for all UDC/PHY drivers unless the device can guarantee that it will automatically wake up from suspend in time to service a USB packet (note that the window for responding to a packet is only a few microseconds). Otherwise the device would appear to the host to be unresponsive and broken -- better to do a clean disconnect.

Quoted 8 lines
Quoted 2 lines

If suspending the device while it is in use would cause problems ... then don't suspend it when it is in use!

I second your thoughts, but today we don't have enough infrastructure to communicate between PHY and Link, so a clean solution isn't possible, right ?

Should we block this patch due to that ?

No, the patch is appropriate.

We don't need better communication. If g_mass_storage (for example) knows that the device is in use, it can block suspends by returning -EBUSY from its own suspend callback. The UDC driver doesn't need to worry about these matters; it should assume that such things have already been handled elsewhere. That's what Dmitry meant when he was talking about a "higher level driver" -- maybe "lower level" would have been a better choice of words. :-)

Until recently, ordering of the suspend callbacks didn't present any problem. The gadget device was a child of the UDC device and therefore its suspend callback would always be invoked first.

With the new UDC framework, I don't know if this is true any more. It's something to consider.

Alan Stern

#13 lore raw

PATCH 1/2 gpio-vbus: support disabling D+ pullup on suspend

Hi,

Quoted 28 lines · Alan Stern
Quoted 7 lines · Felipe Balbi
Quoted 7 lines

In fact, something like this is _necessary_ for all UDC/PHY drivers unless the device can guarantee that it will automatically wake up from suspend in time to service a USB packet (note that the window for responding to a packet is only a few microseconds). Otherwise the device would appear to the host to be unresponsive and broken -- better to do a clean disconnect.

Quoted 8 lines
Quoted 2 lines

If suspending the device while it is in use would cause problems ... then don't suspend it when it is in use!

I second your thoughts, but today we don't have enough infrastructure to communicate between PHY and Link, so a clean solution isn't possible, right ?

Should we block this patch due to that ?

No, the patch is appropriate.

We don't need better communication. If g_mass_storage (for example) knows that the device is in use, it can block suspends by returning -EBUSY from its own suspend callback. The UDC driver doesn't need to worry about these matters; it should assume that such things have already been handled elsewhere. That's what Dmitry meant when he was talking about a "higher level driver" -- maybe "lower level" would have been a better choice of words. :-)

Quoted 6 lines

Until recently, ordering of the suspend callbacks didn't present any problem. The gadget device was a child of the UDC device and therefore its suspend callback would always be invoked first.

With the new UDC framework, I don't know if this is true any more. It's something to consider.

it's still true, but with one extra device in the middle. e.g.:
musb is parent to udc-core which is parent to g_zero.

Signature · 8 lines
balbi
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 490 bytes
Desc: Digital signature
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20110622/4af441b7/attachment.sig>
maintainer
#14 lore raw

PATCH 1/2 gpio-vbus: support disabling D+ pullup on suspend

Quoted 4 lines · Felipe Balbi

Hi,

Quoted 1 line · Alan Stern

No, the patch is appropriate.

Indeed.

Quoted 8 lines
Quoted 8 lines

We don't need better communication. If g_mass_storage (for example) knows that the device is in use, it can block suspends by returning -EBUSY from its own suspend callback. The UDC driver doesn't need to worry about these matters; it should assume that such things have already been handled elsewhere. That's what Dmitry meant when he was talking about a "higher level driver" -- maybe "lower level" would have been a better choice of words. :-)

The suspend at driver level should not care about filesystem in use, etc ... A disconnected cable cannot be prevented by the kernel, and the effect in fine is the same as the suspend.

Quoted 6 lines
Quoted 6 lines

Until recently, ordering of the suspend callbacks didn't present any problem. The gadget device was a child of the UDC device and therefore its suspend callback would always be invoked first.

With the new UDC framework, I don't know if this is true any more. It's something to consider.

That true also for the UDC driver, which should be suspended before the transciever, so that it can complete its shutdown properly. That's what bothers me with the patch.

And we should not confuse the 2 suspends :
(1) driver suspend, which happens at suspendToRam or suspendToDisk
(the one we're discussing here)
(2) USB suspend (which if I remember correctly is a special USB
command while the USB bus remains powered).

The order of suspend in (1) requires :
- gadget (gzero, gether, ...) suspended first
- UDC suspended second
- transceiver suspended last

The order of suspend in (2) doesn't require anything AFAIR.

For order in (1) :
- gadget before UDC should be enforced by the framework, as the UDC
usb_gadget_probe_driver() method calls device_add() on the gadget<
- UDC before transceiver is enforced in some drivers by directly
manipulating the gpio line.

So Dimitry, do you have an idea as how to guarantee order of suspend in (1), between UDC driver and gpio_vbus ? Maybe an otg_*() call would do the trick ?

Cheers.

--
Robert

#15 lore raw

PATCH 1/2 gpio-vbus: support disabling D+ pullup on suspend

Quoted 31 lines · Robert Jarzmik
Quoted 7 lines · Felipe Balbi
Quoted 6 lines · Alan Stern

Until recently, ordering of the suspend callbacks didn't present any problem. The gadget device was a child of the UDC device and therefore its suspend callback would always be invoked first.

With the new UDC framework, I don't know if this is true any more. It's something to consider.

That true also for the UDC driver, which should be suspended before the transciever, so that it can complete its shutdown properly. That's what bothers me with the patch.

And we should not confuse the 2 suspends :
(1) driver suspend, which happens at suspendToRam or suspendToDisk
(the one we're discussing here)
(2) USB suspend (which if I remember correctly is a special USB
command while the USB bus remains powered).

The order of suspend in (1) requires :
- gadget (gzero, gether, ...) suspended first
- UDC suspended second
- transceiver suspended last

The order of suspend in (2) doesn't require anything AFAIR.

For order in (1) :
- gadget before UDC should be enforced by the framework, as the UDC
usb_gadget_probe_driver() method calls device_add() on the gadget<
- UDC before transceiver is enforced in some drivers by directly
manipulating the gpio line.

Quoted 3 lines

So Dimitry, do you have an idea as how to guarantee order of suspend in (1), between UDC driver and gpio_vbus ? Maybe an otg_*() call would do the trick ?

That is simple: UDC driver acquires transceiver via otg_*() calls, so transceiver
is already registered at the point of probing UDC. I'll check the order of calls during suspend though.

Signature · 2 lines
With best wishes
Dmitry
#16 lore raw

PATCH 1/2 gpio-vbus: support disabling D+ pullup on suspend

In reply to Robert Jarzmik
Quoted 14 lines · Robert Jarzmik
Quoted 4 lines · Felipe Balbi

Hi,

Quoted 1 line · Alan Stern

No, the patch is appropriate.

Indeed.

Quoted 8 lines
Quoted 8 lines

We don't need better communication. If g_mass_storage (for example) knows that the device is in use, it can block suspends by returning -EBUSY from its own suspend callback. The UDC driver doesn't need to worry about these matters; it should assume that such things have already been handled elsewhere. That's what Dmitry meant when he was talking about a "higher level driver" -- maybe "lower level" would have been a better choice of words. :-)

Quoted 3 lines

The suspend at driver level should not care about filesystem in use, etc ... A disconnected cable cannot be prevented by the kernel, and the effect in fine is the same as the suspend.

This is a little ironic, because the mass-storage gadget driver has no idea whether a filesystem is in use or not. All it knows is whether it is currently connected to a host. Besides, it's questionable whether the gadget driver should prevent the gadget from suspending. This is a policy matter which must be decided by the user, not by the kernel.

Nevertheless... The fact that the kernel can do nothing about a disconnected cable shouldn't stop us from handling system sleep correctly. After all, the kernel _can_ do something about that.

Alan Stern

maintainer
#17 lore raw

PATCH 1/2 gpio-vbus: support disabling D+ pullup on suspend

Quoted 3 lines · Alan Stern

Nevertheless... The fact that the kernel can do nothing about a disconnected cable shouldn't stop us from handling system sleep correctly. After all, the kernel _can_ do something about that.

Well, it depends of which level of driver we speek about.
I was speeking about UDC driver, ie. a hardware driver. My point was that the hardware driver should only care about the hardware register manipulations, and check their success. If they succeed, then it should suspend itself without care of filesystem not in sync.

I didn't meant that the filesystem driver shouldn't care about the suspend, or even the mass-storage gadget. I don't have a strong opinion on these higher levels. The question if the kernel can do something about it is much more complicated :
- should userspace sync/umount and check filesystem before suspending
(as umount usb storage before suspending)
- should the kernel block a suspend because a usb key was "forgotten"
in an USB slot ?

For these questions, I'll let others battle. For the hardware related drivers, I'm pretty convinced all they should care about is the success or failure of hardware suspend manipulation, and the correct suspend order.

Cheers.

--
Robert

#18 lore raw

PATCH 1/2 gpio-vbus: support disabling D+ pullup on suspend

In reply to Robert Jarzmik
Quoted 24 lines · Robert Jarzmik
Quoted 4 lines · Felipe Balbi

Hi,

Quoted 1 line · Alan Stern

No, the patch is appropriate.

Indeed.

Quoted 8 lines
Quoted 8 lines

We don't need better communication. If g_mass_storage (for example) knows that the device is in use, it can block suspends by returning -EBUSY from its own suspend callback. The UDC driver doesn't need to worry about these matters; it should assume that such things have already been handled elsewhere. That's what Dmitry meant when he was talking about a "higher level driver" -- maybe "lower level" would have been a better choice of words. :-)

The suspend at driver level should not care about filesystem in use, etc ... A disconnected cable cannot be prevented by the kernel, and the effect in fine is the same as the suspend.

Quoted 6 lines
Quoted 6 lines

Until recently, ordering of the suspend callbacks didn't present any problem. The gadget device was a child of the UDC device and therefore its suspend callback would always be invoked first.

With the new UDC framework, I don't know if this is true any more. It's something to consider.

Quoted 7 lines

That true also for the UDC driver, which should be suspended before the transciever, so that it can complete its shutdown properly. That's what bothers me with the patch.

And we should not confuse the 2 suspends :
(1) driver suspend, which happens at suspendToRam or suspendToDisk
(the one we're discussing here)

I think Dmitry's patch is for this system suspend, and less power consumption is very important for system design, esp for embedded device.

Quoted 2 lines

(2) USB suspend (which if I remember correctly is a special USB
command while the USB bus remains powered).

This suspend is USB suspend in USB 2.0 spec, it is usually triggered by Usb controller after bus idle 3ms. It should not disable D+ pull-up, or the bus will not be J, and the host will think the device has disconnected.

Quoted 27 lines

The order of suspend in (1) requires :
- gadget (gzero, gether, ...) suspended first
- UDC suspended second
- transceiver suspended last

The order of suspend in (2) doesn't require anything AFAIR.

For order in (1) :
- gadget before UDC should be enforced by the framework, as the UDC
usb_gadget_probe_driver() method calls device_add() on the gadget<
- UDC before transceiver is enforced in some drivers by directly
manipulating the gpio line.

So Dimitry, do you have an idea as how to guarantee order of suspend in (1), between UDC driver and gpio_vbus ? Maybe an otg_*() call would do the trick ?

Cheers.

--
Robert
--
To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo at vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html

Signature · 2 lines
Best Regards,
Peter Chen