Skip to contentSkip to conversation

USB: dwc2: Allow wakeup from suspend; enable for rk3288-veyron

PATCH In review Douglas Anderson 2 participants 5 messages May 20, 2019 – Jun 13, 2019

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

v3 cover USB: dwc2: Allow wakeup from suspend; enable for rk3288-veyron

This is a re-post of the last 3 patches of a series I posted earlier at:
https://lkml.kernel.org/r/[email protected]

The first two patches were applied but the last three weren't because they didn't apply at the time. They apply fine now so are ready to land.

Patch #2 has a slight difference in v3 here to fix build robot reported error.

Changes in v3:
- Fixed kbuild test robot error.

Changes in v2:
- Rebased to mainline atop rk3288 remote wake quirk series.
- rk3288-veyron dts patch new for v2.

Douglas Anderson (3):
Documentation: dt-bindings: Add snps,need-phy-for-wake for dwc2 USB
USB: dwc2: Don't turn off the usbphy in suspend if wakeup is enabled
ARM: dts: rockchip: Allow wakeup from rk3288-veyron's dwc2 USB ports

 .../devicetree/bindings/usb/dwc2.txt          |  3 +++
 arch/arm/boot/dts/rk3288-veyron.dtsi          |  2 ++
 drivers/usb/dwc2/core.h                       |  8 +++++++
 drivers/usb/dwc2/hcd.c                        | 19 +++++++++++++++
 drivers/usb/dwc2/platform.c                   | 23 ++++++++++++++++---
 5 files changed, 52 insertions(+), 3 deletions(-)
Signature · 1 line
2.21.0.1020.gf2820cf01a-goog
#1 lore raw

PATCH v3 1/3 Documentation: dt-bindings: Add snps,need-phy-for-wake for dwc2 USB

Some SoCs with a dwc2 USB controller may need to keep the PHY on to support remote wakeup. Allow specifying this as a device tree property.

---
For relevant prior discussion on this patch, see:

https://lkml.kernel.org/r/[email protected]

I didn't make any changes from the prior version since I never found out what Rob thought of my previous arguments. If folks want a change, perhaps they could choose from these options:

1. Assume that all dwc2 hosts would like to keep their PHY on for suspend if there's a USB wakeup enabled, thus we totally drop this binding. This doesn't seem super great to me since I'd bet that many devices that use dwc2 weren't designed for USB wakeup (they may not keep enough clocks or rails on) so we might be wasting power for nothing.
2. Rename this property to "snps,wakeup-from-suspend-with-phy" to make it more obvious that this property is intended both to document that wakeup from suspend is possible and that we need the PHY for said wakeup.
3. Rename this property to "snps,can-wakeup-from-suspend" and assume it's implicit that if we can wakeup from suspend that we need to keep the PHY on. If/when someone shows that a device exists using dwc2 where we can wakeup from suspend without the PHY they can add a new property.

NOTE FOR REPOST:
- In v2 Rob said [1] he'd prefer something based on the SoC compatibility string, but that doesn't work because not all boards will have the regulator setup / board design / suspend logic necessary to make this work.

[1] https://lkml.kernel.org/r/20190430012328.GA25660@bogus

Changes in v3: None
Changes in v2: None

 Documentation/devicetree/bindings/usb/dwc2.txt | 3 +++
 1 file changed, 3 insertions(+)
Documentation/devicetree/bindings/usb/dwc2.txt +3 -0
index 49eac0dc86b0..aafff3a6904d 100644
@@ -42,6 +42,8 @@ Refer to phy/phy-bindings.txt for generic phy consumer properties
 - g-rx-fifo-size: size of rx fifo size in gadget mode.
 - g-np-tx-fifo-size: size of non-periodic tx fifo size in gadget mode.
 - g-tx-fifo-size: size of periodic tx fifo per endpoint (except ep0) in gadget mode.
+- snps,need-phy-for-wake: If present indicates that the phy needs to be left
+                          on for remote wakeup during suspend.
 - snps,reset-phy-on-wake: If present indicates that we need to reset the PHY when
                           we detect a wakeup.  This is due to a hardware errata.
 
@@ -58,4 +60,5 @@ Example:
 		clock-names = "otg";
 		phys = <&usbphy>;
 		phy-names = "usb2-phy";
+		snps,need-phy-for-wake;
         };
Signature · 1 line
2.21.0.1020.gf2820cf01a-goog
maintainer Reviewed-by on v3 1/3
#2 lore raw

Re: [PATCH v3 1/3] Documentation: dt-bindings: Add snps,need-phy-for-wake for dwc2 USB

Quoted 36 lines · Douglas Anderson

Some SoCs with a dwc2 USB controller may need to keep the PHY on to support remote wakeup. Allow specifying this as a device tree property.

---
For relevant prior discussion on this patch, see:

https://lkml.kernel.org/r/[email protected]

I didn't make any changes from the prior version since I never found out what Rob thought of my previous arguments. If folks want a change, perhaps they could choose from these options:

1. Assume that all dwc2 hosts would like to keep their PHY on for suspend if there's a USB wakeup enabled, thus we totally drop this binding. This doesn't seem super great to me since I'd bet that many devices that use dwc2 weren't designed for USB wakeup (they may not keep enough clocks or rails on) so we might be wasting power for nothing.
2. Rename this property to "snps,wakeup-from-suspend-with-phy" to make it more obvious that this property is intended both to document that wakeup from suspend is possible and that we need the PHY for said wakeup.
3. Rename this property to "snps,can-wakeup-from-suspend" and assume it's implicit that if we can wakeup from suspend that we need to keep the PHY on. If/when someone shows that a device exists using dwc2 where we can wakeup from suspend without the PHY they can add a new property.

NOTE FOR REPOST:
- In v2 Rob said [1] he'd prefer something based on the SoC compatibility string, but that doesn't work because not all boards will have the regulator setup / board design / suspend logic necessary to make this work.

Quoted 9 lines

[1] https://lkml.kernel.org/r/20190430012328.GA25660@bogus

Changes in v3: None
Changes in v2: None

 Documentation/devicetree/bindings/usb/dwc2.txt | 3 +++
 1 file changed, 3 insertions(+)
#3 lore raw

PATCH v3 2/3 USB: dwc2: Don't turn off the usbphy in suspend if wakeup is enabled

In reply to Douglas Anderson

If the 'snps,need-phy-for-wake' is set in the device tree then:

- We know that we can wakeup, so call device_set_wakeup_capable(). The USB core will use this knowledge to enable wakeup by default.
- We know that we should keep the PHY on during suspend if something on our root hub needs remote wakeup. This requires the patch (USB:
Export usb_wakeup_enabled_descendants()). Note that we don't keep the PHY on at suspend time if it's not needed because it would be a power draw.

If we later find some users of dwc2 that can support wakeup without keeping the PHY on we may want to add a way to call device_set_wakeup_capable() without keeping the PHY on at suspend time.

---
For relevant prior discussion of this idea, see:

https://lkml.kernel.org/r/[email protected]

If I'm reading all the responses correctly folks were of the opinion that this patch is still the right way to go.

Changes in v3:
- Fixed kbuild test robot error.

Changes in v2:
- Rebased to mainline atop rk3288 remote wake quirk series.

 drivers/usb/dwc2/core.h     |  8 ++++++++
 drivers/usb/dwc2/hcd.c      | 19 +++++++++++++++++++
 drivers/usb/dwc2/platform.c | 23 ++++++++++++++++++++---
 3 files changed, 47 insertions(+), 3 deletions(-)
drivers/usb/dwc2/core.h +8 -0
index 152ac41dfb2d..d08d070a0fb6 100644
@@ -861,6 +861,9 @@ struct dwc2_hregs_backup {
  * @hibernated:		True if core is hibernated
  * @reset_phy_on_wake:	Quirk saying that we should assert PHY reset on a
  *			remote wakeup.
+ * @phy_off_for_suspend: Status of whether we turned the PHY off at suspend.
+ * @need_phy_for_wake:	Quirk saying that we should keep the PHY on at
+ *			suspend if we need USB to wake us up.
  * @frame_number:       Frame number read from the core. For both device
  *			and host modes. The value ranges are from 0
  *			to HFNUM_MAX_FRNUM.
@@ -1049,6 +1052,8 @@ struct dwc2_hsotg {
 	unsigned int ll_hw_enabled:1;
 	unsigned int hibernated:1;
 	unsigned int reset_phy_on_wake:1;
+	unsigned int need_phy_for_wake:1;
+	unsigned int phy_off_for_suspend:1;
 	u16 frame_number;
 
 	struct phy *phy;
@@ -1438,6 +1443,7 @@ int dwc2_restore_host_registers(struct dwc2_hsotg *hsotg);
 int dwc2_host_enter_hibernation(struct dwc2_hsotg *hsotg);
 int dwc2_host_exit_hibernation(struct dwc2_hsotg *hsotg,
 			       int rem_wakeup, int reset);
+bool dwc2_host_can_poweroff_phy(struct dwc2_hsotg *dwc2);
 static inline void dwc2_host_schedule_phy_reset(struct dwc2_hsotg *hsotg)
 { schedule_work(&hsotg->phy_reset_work); }
 #else
@@ -1463,6 +1469,8 @@ static inline int dwc2_host_enter_hibernation(struct dwc2_hsotg *hsotg)
 static inline int dwc2_host_exit_hibernation(struct dwc2_hsotg *hsotg,
 					     int rem_wakeup, int reset)
 { return 0; }
+static inline bool dwc2_host_can_poweroff_phy(struct dwc2_hsotg *dwc2)
+{ return false; }
 static inline void dwc2_host_schedule_phy_reset(struct dwc2_hsotg *hsotg) {}
 
 #endif
drivers/usb/dwc2/hcd.c +19 -0
index b50ec3714fd8..cc918fe7ddb8 100644
@@ -5578,3 +5578,22 @@ int dwc2_host_exit_hibernation(struct dwc2_hsotg *hsotg, int rem_wakeup,
 	dev_dbg(hsotg->dev, "Host hibernation restore complete\n");
 	return ret;
 }
+
+bool dwc2_host_can_poweroff_phy(struct dwc2_hsotg *dwc2)
+{
+	struct usb_device *root_hub = dwc2_hsotg_to_hcd(dwc2)->self.root_hub;
+
+	/* If the controller isn't allowed to wakeup then we can power off. */
+	if (!device_may_wakeup(dwc2->dev))
+		return true;
+
+	/*
+	 * We don't want to power off the PHY if something under the
+	 * root hub has wakeup enabled.
+	 */
+	if (usb_wakeup_enabled_descendants(root_hub))
+		return false;
+
+	/* No reason to keep the PHY powered, so allow poweroff */
+	return true;
+}
drivers/usb/dwc2/platform.c +20 -3
index d10a7f8daec3..3e6c3c8a32ff 100644
@@ -447,6 +447,10 @@ static int dwc2_driver_probe(struct platform_device *dev)
 	if (retval)
 		goto error;
 
+	hsotg->need_phy_for_wake =
+		of_property_read_bool(dev->dev.of_node,
+				      "snps,need-phy-for-wake");
+
 	/*
 	 * Reset before dwc2_get_hwparams() then it could get power-on real
 	 * reset value form registers.
@@ -478,6 +482,14 @@ static int dwc2_driver_probe(struct platform_device *dev)
 		hsotg->gadget_enabled = 1;
 	}
 
+	/*
+	 * If we need PHY for wakeup we must be wakeup capable.
+	 * When we have a device that can wake without the PHY we
+	 * can adjust this condition.
+	 */
+	if (hsotg->need_phy_for_wake)
+		device_set_wakeup_capable(&dev->dev, true);
+
 	hsotg->reset_phy_on_wake =
 		of_property_read_bool(dev->dev.of_node,
 				      "snps,reset-phy-on-wake");
@@ -516,13 +528,17 @@ static int dwc2_driver_probe(struct platform_device *dev)
 static int __maybe_unused dwc2_suspend(struct device *dev)
 {
 	struct dwc2_hsotg *dwc2 = dev_get_drvdata(dev);
+	bool is_device_mode = dwc2_is_device_mode(dwc2);
 	int ret = 0;
 
-	if (dwc2_is_device_mode(dwc2))
+	if (is_device_mode)
 		dwc2_hsotg_suspend(dwc2);
 
-	if (dwc2->ll_hw_enabled)
+	if (dwc2->ll_hw_enabled &&
+	    (is_device_mode || dwc2_host_can_poweroff_phy(dwc2))) {
 		ret = __dwc2_lowlevel_hw_disable(dwc2);
+		dwc2->phy_off_for_suspend = true;
+	}
 
 	return ret;
 }
@@ -532,11 +548,12 @@ static int __maybe_unused dwc2_resume(struct device *dev)
 	struct dwc2_hsotg *dwc2 = dev_get_drvdata(dev);
 	int ret = 0;
 
-	if (dwc2->ll_hw_enabled) {
+	if (dwc2->phy_off_for_suspend && dwc2->ll_hw_enabled) {
 		ret = __dwc2_lowlevel_hw_enable(dwc2);
 		if (ret)
 			return ret;
 	}
+	dwc2->phy_off_for_suspend = false;
 
 	if (dwc2_is_device_mode(dwc2))
 		ret = dwc2_hsotg_resume(dwc2);
Signature · 1 line
2.21.0.1020.gf2820cf01a-goog
#4 lore raw

PATCH v3 3/3 ARM: dts: rockchip: Allow wakeup from rk3288-veyron's dwc2 USB ports

In reply to Douglas Anderson

We want to be able to wake from USB if a device is plugged in that wants remote wakeup. Enable it on both dwc2 controllers.

NOTE: this is added specifically to veyron and not to rk3288 in general since it's not known whether all rk3288 boards are designed to support USB wakeup. It is plausible that some boards could shut down important rails in S3.

Also note that currently wakeup doesn't seem to happen unless you use the "deep" suspend mode (where SDRAM is turned off). Presumably the shallow suspend mode is gating some sort of clock that's important but I couldn't easily figure out how to get it working.

---

Changes in v3: None
Changes in v2:
- rk3288-veyron dts patch new for v2.

 arch/arm/boot/dts/rk3288-veyron.dtsi | 2 ++
 1 file changed, 2 insertions(+)
arch/arm/boot/dts/rk3288-veyron.dtsi +2 -0
index 1252522392c7..1d8bfed7830c 100644
@@ -424,6 +424,7 @@
 
 &usb_host1 {
 	status = "okay";
+	snps,need-phy-for-wake;
 };
 
 &usb_otg {
@@ -432,6 +433,7 @@
 	assigned-clocks = <&cru SCLK_USBPHY480M_SRC>;
 	assigned-clock-parents = <&usbphy0>;
 	dr_mode = "host";
+	snps,need-phy-for-wake;
 };
 
 &vopb {
Signature · 1 line
2.21.0.1020.gf2820cf01a-goog