[PATCH] usb: dwc3: qcom: Check for L2 only when suspending a host with wakeup

[PATCH] usb: dwc3: qcom: Check for L2 only when suspending a host with wakeup

[Fabricio Akio via B4 Relay]

From: Fabricio Akio <fabricioakio@gmail.com>

In peripheral mode the core disconnects and powers the PHY off before
the glue suspends, so the link never enters L2 and every system
suspend logs "HS-PHY not in L2". Check only for a host with wakeup,
the one case that parks the PHY in L2.

Signed-off-by: Fabricio Akio <fabricioakio@gmail.com>
---
Tested on a Motorola Moto X4 (XT1900-6, SDM630) with the sdm660-mainline
7.2.3 tree, in peripheral mode with the cable attached: the "HS-PHY not
in L2" message appears on every system suspend without this patch, and
not over three consecutive cycles with it. Host mode is untested, since
this board runs the controller in peripheral mode. The board binds
dwc3-qcom-legacy, so the same change in dwc3-qcom.c is build-tested
only. Applies unchanged to next-20260928.
---
 drivers/usb/dwc3/dwc3-qcom-legacy.c | 16 ++++++++++++----
 drivers/usb/dwc3/dwc3-qcom.c        | 16 ++++++++++++----
 2 files changed, 24 insertions(+), 8 deletions(-)

diff --git a/drivers/usb/dwc3/dwc3-qcom-legacy.c b/drivers/usb/dwc3/dwc3-qcom-legacy.c
index d3fad0fcfdac..ba3f70dea4ab 100644
--- a/drivers/usb/dwc3/dwc3-qcom-legacy.c
+++ b/drivers/usb/dwc3/dwc3-qcom-legacy.c
@@ -431,10 +431,18 @@ static int dwc3_qcom_suspend(struct dwc3_qcom *qcom, bool wakeup)
 	if (qcom->is_suspended)
 		return 0;
 
-	for (i = 0; i < qcom->num_ports; i++) {
-		val = readl(qcom->qscratch_base + pwr_evnt_irq_stat_reg[i]);
-		if (!(val & PWR_EVNT_LPM_IN_L2_MASK))
-			dev_err(qcom->dev, "port-%d HS-PHY not in L2\n", i + 1);
+	/*
+	 * Only a host with wakeup enabled keeps the HS PHY powered and parks
+	 * it in L2 across suspend. A peripheral disconnects from the bus and
+	 * powers the PHY off first, so its link never enters L2.
+	 */
+	if (dwc3_qcom_is_host(qcom) && wakeup) {
+		for (i = 0; i < qcom->num_ports; i++) {
+			val = readl(qcom->qscratch_base + pwr_evnt_irq_stat_reg[i]);
+			if (!(val & PWR_EVNT_LPM_IN_L2_MASK))
+				dev_err(qcom->dev, "port-%d HS-PHY not in L2\n",
+					i + 1);
+		}
 	}
 
 	for (i = qcom->num_clocks - 1; i >= 0; i--)
diff --git a/drivers/usb/dwc3/dwc3-qcom.c b/drivers/usb/dwc3/dwc3-qcom.c
index ac68b4218b56..e0920da280ce 100644
--- a/drivers/usb/dwc3/dwc3-qcom.c
+++ b/drivers/usb/dwc3/dwc3-qcom.c
@@ -343,10 +343,18 @@ static int dwc3_qcom_suspend(struct dwc3_qcom *qcom, bool wakeup)
 	if (qcom->is_suspended)
 		return 0;
 
-	for (i = 0; i < qcom->num_ports; i++) {
-		val = readl(qcom->qscratch_base + pwr_evnt_irq_stat_reg[i]);
-		if (!(val & PWR_EVNT_LPM_IN_L2_MASK))
-			dev_err(qcom->dev, "port-%d HS-PHY not in L2\n", i + 1);
+	/*
+	 * Only a host with wakeup enabled keeps the HS PHY powered and parks
+	 * it in L2 across suspend. A peripheral disconnects from the bus and
+	 * powers the PHY off first, so its link never enters L2.
+	 */
+	if (dwc3_qcom_is_host(qcom) && wakeup) {
+		for (i = 0; i < qcom->num_ports; i++) {
+			val = readl(qcom->qscratch_base + pwr_evnt_irq_stat_reg[i]);
+			if (!(val & PWR_EVNT_LPM_IN_L2_MASK))
+				dev_err(qcom->dev, "port-%d HS-PHY not in L2\n",
+					i + 1);
+		}
 	}
 	clk_bulk_disable_unprepare(qcom->num_clocks, qcom->clks);
 

---
base-commit: 6375e61c01e93e35ee7acd336a689ac1fae4b509
change-id: 20260928-dwc3-qcom-l2-check-e3b55acefb62

Best regards,
--  
Fabricio Akio <fabricioakio@gmail.com>

Re: [PATCH] usb: dwc3: qcom: Check for L2 only when suspending a host with wakeup

[Huang Wei]

Hi Fabricio,

This matches what I'd expect from the PHY behaviour - in peripheral
mode the device disconnects before the glue suspends, so there is no
link left to park in L2, and the unconditional check is just log noise
on every suspend. Restricting it to the host-with-wakeup case that
actually relies on L2 retention looks right.

The untested host path is low risk in a specific way: for host with
wakeup the check runs exactly as before, so the patch only silences
the cases where L2 is not expected. And since the runtime PM paths
pass wakeup unconditionally, host runtime suspend still gets the
check, which seems to be the intended behaviour.

Tested on real hardware in the mode that was noisy - nice.

Reviewed-by: Huang Wei <huangwei@kylinos.cn>

Thanks,
Huang Wei

Re: [PATCH] usb: dwc3: qcom: Check for L2 only when suspending a host with wakeup

[Thinh Nguyen]

On Mon, Sep 28, 2026, Fabricio Akio via B4 Relay wrote:
> From: Fabricio Akio <fabricioakio@gmail.com>
> 
> In peripheral mode the core disconnects and powers the PHY off before
> the glue suspends, so the link never enters L2 and every system
> suspend logs "HS-PHY not in L2". Check only for a host with wakeup,
> the one case that parks the PHY in L2.
> 
> Signed-off-by: Fabricio Akio <fabricioakio@gmail.com>
> ---
> Tested on a Motorola Moto X4 (XT1900-6, SDM630) with the sdm660-mainline
> 7.2.3 tree, in peripheral mode with the cable attached: the "HS-PHY not
> in L2" message appears on every system suspend without this patch, and
> not over three consecutive cycles with it. Host mode is untested, since
> this board runs the controller in peripheral mode. The board binds
> dwc3-qcom-legacy, so the same change in dwc3-qcom.c is build-tested
> only. Applies unchanged to next-20260928.
> ---
>  drivers/usb/dwc3/dwc3-qcom-legacy.c | 16 ++++++++++++----
>  drivers/usb/dwc3/dwc3-qcom.c        | 16 ++++++++++++----
>  2 files changed, 24 insertions(+), 8 deletions(-)
> 
> diff --git a/drivers/usb/dwc3/dwc3-qcom-legacy.c b/drivers/usb/dwc3/dwc3-qcom-legacy.c
> index d3fad0fcfdac..ba3f70dea4ab 100644
> --- a/drivers/usb/dwc3/dwc3-qcom-legacy.c
> +++ b/drivers/usb/dwc3/dwc3-qcom-legacy.c
> @@ -431,10 +431,18 @@ static int dwc3_qcom_suspend(struct dwc3_qcom *qcom, bool wakeup)
>  	if (qcom->is_suspended)
>  		return 0;
>  
> -	for (i = 0; i < qcom->num_ports; i++) {
> -		val = readl(qcom->qscratch_base + pwr_evnt_irq_stat_reg[i]);
> -		if (!(val & PWR_EVNT_LPM_IN_L2_MASK))
> -			dev_err(qcom->dev, "port-%d HS-PHY not in L2\n", i + 1);
> +	/*
> +	 * Only a host with wakeup enabled keeps the HS PHY powered and parks
> +	 * it in L2 across suspend. A peripheral disconnects from the bus and
> +	 * powers the PHY off first, so its link never enters L2.
> +	 */
> +	if (dwc3_qcom_is_host(qcom) && wakeup) {
> +		for (i = 0; i < qcom->num_ports; i++) {
> +			val = readl(qcom->qscratch_base + pwr_evnt_irq_stat_reg[i]);
> +			if (!(val & PWR_EVNT_LPM_IN_L2_MASK))
> +				dev_err(qcom->dev, "port-%d HS-PHY not in L2\n",
> +					i + 1);
> +		}
>  	}
>  
>  	for (i = qcom->num_clocks - 1; i >= 0; i--)
> diff --git a/drivers/usb/dwc3/dwc3-qcom.c b/drivers/usb/dwc3/dwc3-qcom.c
> index ac68b4218b56..e0920da280ce 100644
> --- a/drivers/usb/dwc3/dwc3-qcom.c
> +++ b/drivers/usb/dwc3/dwc3-qcom.c
> @@ -343,10 +343,18 @@ static int dwc3_qcom_suspend(struct dwc3_qcom *qcom, bool wakeup)
>  	if (qcom->is_suspended)
>  		return 0;
>  
> -	for (i = 0; i < qcom->num_ports; i++) {
> -		val = readl(qcom->qscratch_base + pwr_evnt_irq_stat_reg[i]);
> -		if (!(val & PWR_EVNT_LPM_IN_L2_MASK))
> -			dev_err(qcom->dev, "port-%d HS-PHY not in L2\n", i + 1);
> +	/*
> +	 * Only a host with wakeup enabled keeps the HS PHY powered and parks
> +	 * it in L2 across suspend. A peripheral disconnects from the bus and
> +	 * powers the PHY off first, so its link never enters L2.
> +	 */
> +	if (dwc3_qcom_is_host(qcom) && wakeup) {
> +		for (i = 0; i < qcom->num_ports; i++) {
> +			val = readl(qcom->qscratch_base + pwr_evnt_irq_stat_reg[i]);
> +			if (!(val & PWR_EVNT_LPM_IN_L2_MASK))
> +				dev_err(qcom->dev, "port-%d HS-PHY not in L2\n",
> +					i + 1);
> +		}
>  	}
>  	clk_bulk_disable_unprepare(qcom->num_clocks, qcom->clks);
>  
> 
> ---
> base-commit: 6375e61c01e93e35ee7acd336a689ac1fae4b509
> change-id: 20260928-dwc3-qcom-l2-check-e3b55acefb62
> 
> Best regards,
> --  
> Fabricio Akio <fabricioakio@gmail.com>
> 
> 

Acked-by: Thinh Nguyen <Thinh.Nguyen@synopsys.com>

Thanks,
Thinh

Re: [PATCH] usb: dwc3: qcom: Check for L2 only when suspending a host with wakeup

[Joonhoe Kim]

On Mon, Sep 28, 2026 at 03:16:28PM -0300, Fabricio Akio wrote:
> Host mode is untested, since this board runs the controller in
> peripheral mode.

I tried the host side on a Lenovo Legion Tab Y700 Gen 5 (SM8850,
dwc3-qcom), v7.3-rc4 plus board patches, s2idle with an RTC wakeup:

 - host with wakeup disabled (USB 2.0 hub, USB 3 flash drive behind
   it): "HS-PHY not in L2" on every suspend before, none with the
   patch; the devices are still there after resume
 - host with wakeup enabled: no message, the port does reach L2
 - peripheral (connected to a PC): no message

Tested-by: Joonhoe Kim <26rote@gmail.com>

Re: [PATCH] usb: dwc3: qcom: Check for L2 only when suspending a host with wakeup

[Joonhoe Kim]

On Mon, Sep 28, 2026 at 03:16:28PM -0300, Fabricio Akio wrote:
> Host mode is untested, since this board runs the controller in
> peripheral mode.

I tried the host side on a Lenovo Legion Tab Y700 Gen 5 (SM8850,
dwc3-qcom), v7.3-rc4 plus board patches, s2idle with an RTC wakeup:

 - host with wakeup disabled (USB 2.0 hub, USB 3 flash drive behind
   it): "HS-PHY not in L2" on every suspend before, none with the
   patch; the devices are still there after resume
 - host with wakeup enabled: no message, the port does reach L2
 - peripheral (connected to a PC): no message

Tested-by: Joonhoe Kim <26rote@gmail.com>

Re: [PATCH] usb: dwc3: qcom: Check for L2 only when suspending a host with wakeup

[Pankaj Patil]

On 9/28/2026 11:46 PM, Fabricio Akio via B4 Relay wrote:
> From: Fabricio Akio <fabricioakio@gmail.com>
> 
> In peripheral mode the core disconnects and powers the PHY off before
> the glue suspends, so the link never enters L2 and every system
> suspend logs "HS-PHY not in L2". Check only for a host with wakeup,
> the one case that parks the PHY in L2.
> 
> Signed-off-by: Fabricio Akio <fabricioakio@gmail.com>
> ---
> Tested on a Motorola Moto X4 (XT1900-6, SDM630) with the sdm660-mainline
> 7.2.3 tree, in peripheral mode with the cable attached: the "HS-PHY not
> in L2" message appears on every system suspend without this patch, and
> not over three consecutive cycles with it. Host mode is untested, since
> this board runs the controller in peripheral mode. The board binds
> dwc3-qcom-legacy, so the same change in dwc3-qcom.c is build-tested
> only. Applies unchanged to next-20260928.
> ---
Tested on Glymur-CRD
Tested-by: Pankaj Patil <pankaj.patil@oss.qualcomm.com>

Reviewed-by: Pankaj Patil <pankaj.patil@oss.qualcomm.com>

[...]

Re: [PATCH] usb: dwc3: qcom: Check for L2 only when suspending a host with wakeup

[Konrad Dybcio]

On 9/28/26 8:16 PM, Fabricio Akio via B4 Relay wrote:
> From: Fabricio Akio <fabricioakio@gmail.com>
> 
> In peripheral mode the core disconnects and powers the PHY off before
> the glue suspends, so the link never enters L2 and every system
> suspend logs "HS-PHY not in L2". Check only for a host with wakeup,
> the one case that parks the PHY in L2.
> 
> Signed-off-by: Fabricio Akio <fabricioakio@gmail.com>
> ---

I think we should be able to retain the link in L2 in device
mode too upon suspend, but that needs more plumbing on the
clock/gdsc side? (+Cc)

Konrad

Re: [PATCH] usb: dwc3: qcom: Check for L2 only when suspending a host with wakeup

[Peter Chen]

On 26-10-21 15:44:52, Konrad Dybcio wrote:
> On 9/28/26 8:16 PM, Fabricio Akio via B4 Relay wrote:
> > From: Fabricio Akio <fabricioakio@gmail.com>
> > 
> > In peripheral mode the core disconnects and powers the PHY off before
> > the glue suspends, so the link never enters L2 and every system
> > suspend logs "HS-PHY not in L2". Check only for a host with wakeup,
> > the one case that parks the PHY in L2.
> > 
> > Signed-off-by: Fabricio Akio <fabricioakio@gmail.com>
> > ---
> 
> I think we should be able to retain the link in L2 in device
> mode too upon suspend, but that needs more plumbing on the
> clock/gdsc side? (+Cc)
> 

From what I see, dwc3 core just keeps gadget (device mode) disconnecting
when dwc3 core enters suspend. The reason is probably dwc3 core wakeup signal
(through .dwc3_gadget_wakeup or dwc3_gadget_func_wakeup) is send by software.
When the system has suspended, no software has run, so can't trigger wakeup.

For runtime suspend, current dwc3 core doesn't consider link suspend event
(eg, L2/U3) as low power event, and doesn't put dwc3 core entering runtime
suspend, so the link is always L2 or U3.

Thinh/Wesley, if I understand wrong, please correct me.

-- 

Thanks,
Peter Chen

Re: [PATCH] usb: dwc3: qcom: Check for L2 only when suspending a host with wakeup

[Thinh Nguyen]

On Wed, Oct 21, 2026, Peter Chen wrote:
> On 26-10-21 15:44:52, Konrad Dybcio wrote:
> > On 9/28/26 8:16 PM, Fabricio Akio via B4 Relay wrote:
> > > From: Fabricio Akio <fabricioakio@gmail.com>
> > > 
> > > In peripheral mode the core disconnects and powers the PHY off before
> > > the glue suspends, so the link never enters L2 and every system
> > > suspend logs "HS-PHY not in L2". Check only for a host with wakeup,
> > > the one case that parks the PHY in L2.
> > > 
> > > Signed-off-by: Fabricio Akio <fabricioakio@gmail.com>
> > > ---
> > 
> > I think we should be able to retain the link in L2 in device
> > mode too upon suspend, but that needs more plumbing on the
> > clock/gdsc side? (+Cc)
> > 
> 
> From what I see, dwc3 core just keeps gadget (device mode) disconnecting
> when dwc3 core enters suspend. The reason is probably dwc3 core wakeup signal
> (through .dwc3_gadget_wakeup or dwc3_gadget_func_wakeup) is send by software.
> When the system has suspended, no software has run, so can't trigger wakeup.

Remote wakeup isn't the limitation. The host decides when to put the
link to L2/U3, and the device must still respond to a host resume or
reset while suspended. If the core is powered down and its state is
lost, it can't do that. To support this, we'd need to implement device
hibernation.

> 
> For runtime suspend, current dwc3 core doesn't consider link suspend event
> (eg, L2/U3) as low power event, and doesn't put dwc3 core entering runtime
> suspend, so the link is always L2 or U3.
> 

Right, dwc3 won't runtime suspend while connected, regardless of the
link state.

Fabricio's patch is fine as is.

Thanks,
Thinh