[PATCH v2 1/9] usb: dwc3: fix PHY disable sequence

[PATCH v2 1/9] usb: dwc3: fix PHY disable sequence

[Johan Hovold]

Generic PHYs must be powered-off before they can be tore down.

Similarly, suspending legacy PHYs after having powered them off makes no
sense.

Fix the dwc3_core_exit() (e.g. called during suspend) and open-coded
dwc3_probe() error-path sequences that got this wrong.

Note that this makes dwc3_core_exit() match the dwc3_core_init() error
path with respect to powering off the PHYs.

Fixes: 03c1fd622f72 ("usb: dwc3: core: add phy cleanup for probe error handling")
Fixes: c499ff71ff2a ("usb: dwc3: core: re-factor init and exit paths")
Cc: stable@vger.kernel.org      # 4.8
Reviewed-by: Andrew Halaney <ahalaney@redhat.com>
Reviewed-by: Matthias Kaehlcke <mka@chromium.org>
Signed-off-by: Johan Hovold <johan+linaro@kernel.org>
---
 drivers/usb/dwc3/core.c | 19 ++++++++++---------
 1 file changed, 10 insertions(+), 9 deletions(-)

diff --git a/drivers/usb/dwc3/core.c b/drivers/usb/dwc3/core.c
index c5c238ab3083..16d1f328775f 100644
--- a/drivers/usb/dwc3/core.c
+++ b/drivers/usb/dwc3/core.c
@@ -833,15 +833,16 @@ static void dwc3_core_exit(struct dwc3 *dwc)
 {
 	dwc3_event_buffers_cleanup(dwc);
 
+	usb_phy_set_suspend(dwc->usb2_phy, 1);
+	usb_phy_set_suspend(dwc->usb3_phy, 1);
+	phy_power_off(dwc->usb2_generic_phy);
+	phy_power_off(dwc->usb3_generic_phy);
+
 	usb_phy_shutdown(dwc->usb2_phy);
 	usb_phy_shutdown(dwc->usb3_phy);
 	phy_exit(dwc->usb2_generic_phy);
 	phy_exit(dwc->usb3_generic_phy);
 
-	usb_phy_set_suspend(dwc->usb2_phy, 1);
-	usb_phy_set_suspend(dwc->usb3_phy, 1);
-	phy_power_off(dwc->usb2_generic_phy);
-	phy_power_off(dwc->usb3_generic_phy);
 	dwc3_clk_disable(dwc);
 	reset_control_assert(dwc->reset);
 }
@@ -1879,16 +1880,16 @@ static int dwc3_probe(struct platform_device *pdev)
 	dwc3_debugfs_exit(dwc);
 	dwc3_event_buffers_cleanup(dwc);
 
-	usb_phy_shutdown(dwc->usb2_phy);
-	usb_phy_shutdown(dwc->usb3_phy);
-	phy_exit(dwc->usb2_generic_phy);
-	phy_exit(dwc->usb3_generic_phy);
-
 	usb_phy_set_suspend(dwc->usb2_phy, 1);
 	usb_phy_set_suspend(dwc->usb3_phy, 1);
 	phy_power_off(dwc->usb2_generic_phy);
 	phy_power_off(dwc->usb3_generic_phy);
 
+	usb_phy_shutdown(dwc->usb2_phy);
+	usb_phy_shutdown(dwc->usb3_phy);
+	phy_exit(dwc->usb2_generic_phy);
+	phy_exit(dwc->usb3_generic_phy);
+
 	dwc3_ulpi_exit(dwc);
 
 err4:
-- 
2.35.1

Re: [PATCH v2 1/9] usb: dwc3: fix PHY disable sequence

[Manivannan Sadhasivam]

On Thu, Aug 04, 2022 at 05:09:53PM +0200, Johan Hovold wrote:
> Generic PHYs must be powered-off before they can be tore down.
> 
> Similarly, suspending legacy PHYs after having powered them off makes no
> sense.
> 
> Fix the dwc3_core_exit() (e.g. called during suspend) and open-coded
> dwc3_probe() error-path sequences that got this wrong.
> 
> Note that this makes dwc3_core_exit() match the dwc3_core_init() error
> path with respect to powering off the PHYs.
> 
> Fixes: 03c1fd622f72 ("usb: dwc3: core: add phy cleanup for probe error handling")
> Fixes: c499ff71ff2a ("usb: dwc3: core: re-factor init and exit paths")
> Cc: stable@vger.kernel.org      # 4.8
> Reviewed-by: Andrew Halaney <ahalaney@redhat.com>
> Reviewed-by: Matthias Kaehlcke <mka@chromium.org>
> Signed-off-by: Johan Hovold <johan+linaro@kernel.org>

Reviewed-by: Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org>

Thanks,
Mani

> ---
>  drivers/usb/dwc3/core.c | 19 ++++++++++---------
>  1 file changed, 10 insertions(+), 9 deletions(-)
> 
> diff --git a/drivers/usb/dwc3/core.c b/drivers/usb/dwc3/core.c
> index c5c238ab3083..16d1f328775f 100644
> --- a/drivers/usb/dwc3/core.c
> +++ b/drivers/usb/dwc3/core.c
> @@ -833,15 +833,16 @@ static void dwc3_core_exit(struct dwc3 *dwc)
>  {
>  	dwc3_event_buffers_cleanup(dwc);
>  
> +	usb_phy_set_suspend(dwc->usb2_phy, 1);
> +	usb_phy_set_suspend(dwc->usb3_phy, 1);
> +	phy_power_off(dwc->usb2_generic_phy);
> +	phy_power_off(dwc->usb3_generic_phy);
> +
>  	usb_phy_shutdown(dwc->usb2_phy);
>  	usb_phy_shutdown(dwc->usb3_phy);
>  	phy_exit(dwc->usb2_generic_phy);
>  	phy_exit(dwc->usb3_generic_phy);
>  
> -	usb_phy_set_suspend(dwc->usb2_phy, 1);
> -	usb_phy_set_suspend(dwc->usb3_phy, 1);
> -	phy_power_off(dwc->usb2_generic_phy);
> -	phy_power_off(dwc->usb3_generic_phy);
>  	dwc3_clk_disable(dwc);
>  	reset_control_assert(dwc->reset);
>  }
> @@ -1879,16 +1880,16 @@ static int dwc3_probe(struct platform_device *pdev)
>  	dwc3_debugfs_exit(dwc);
>  	dwc3_event_buffers_cleanup(dwc);
>  
> -	usb_phy_shutdown(dwc->usb2_phy);
> -	usb_phy_shutdown(dwc->usb3_phy);
> -	phy_exit(dwc->usb2_generic_phy);
> -	phy_exit(dwc->usb3_generic_phy);
> -
>  	usb_phy_set_suspend(dwc->usb2_phy, 1);
>  	usb_phy_set_suspend(dwc->usb3_phy, 1);
>  	phy_power_off(dwc->usb2_generic_phy);
>  	phy_power_off(dwc->usb3_generic_phy);
>  
> +	usb_phy_shutdown(dwc->usb2_phy);
> +	usb_phy_shutdown(dwc->usb3_phy);
> +	phy_exit(dwc->usb2_generic_phy);
> +	phy_exit(dwc->usb3_generic_phy);
> +
>  	dwc3_ulpi_exit(dwc);
>  
>  err4:
> -- 
> 2.35.1
> 

-- 
மணிவண்ணன் சதாசிவம்

[PATCH v2 4/9] usb: dwc3: qcom: fix use-after-free on runtime-PM wakeup

[Johan Hovold]

The Qualcomm dwc3 runtime-PM implementation checks the xhci
platform-device pointer in the wakeup-interrupt handler to determine
whether the controller is in host mode and if so triggers a resume.

After a role switch in OTG mode the xhci platform-device would have been
freed and the next wakeup from runtime suspend would access the freed
memory.

Note that role switching is executed from a freezable workqueue, which
guarantees that the pointer is stable during suspend.

Also note that runtime PM has been broken since commit 2664deb09306
("usb: dwc3: qcom: Honor wakeup enabled/disabled state"), which
incidentally also prevents this issue from being triggered.

Fixes: a4333c3a6ba9 ("usb: dwc3: Add Qualcomm DWC3 glue driver")
Cc: stable@vger.kernel.org      # 4.18
Signed-off-by: Johan Hovold <johan+linaro@kernel.org>
---

Changes in v2
 - new patch

 drivers/usb/dwc3/dwc3-qcom.c | 14 +++++++++++++-
 drivers/usb/dwc3/host.c      |  1 +
 2 files changed, 14 insertions(+), 1 deletion(-)

diff --git a/drivers/usb/dwc3/dwc3-qcom.c b/drivers/usb/dwc3/dwc3-qcom.c
index e9364141661b..6884026b9fad 100644
--- a/drivers/usb/dwc3/dwc3-qcom.c
+++ b/drivers/usb/dwc3/dwc3-qcom.c
@@ -298,6 +298,14 @@ static void dwc3_qcom_interconnect_exit(struct dwc3_qcom *qcom)
 	icc_put(qcom->icc_path_apps);
 }
 
+/* Only usable in contexts where the role can not change. */
+static bool dwc3_qcom_is_host(struct dwc3_qcom *qcom)
+{
+	struct dwc3 *dwc = platform_get_drvdata(qcom->dwc3);
+
+	return dwc->xhci;
+}
+
 static enum usb_device_speed dwc3_qcom_read_usb2_speed(struct dwc3_qcom *qcom)
 {
 	struct dwc3 *dwc = platform_get_drvdata(qcom->dwc3);
@@ -460,7 +468,11 @@ static irqreturn_t qcom_dwc3_resume_irq(int irq, void *data)
 	if (qcom->pm_suspended)
 		return IRQ_HANDLED;
 
-	if (dwc->xhci)
+	/*
+	 * This is safe as role switching is done from a freezable workqueue
+	 * and the wakeup interrupts are disabled as part of resume.
+	 */
+	if (dwc3_qcom_is_host(qcom))
 		pm_runtime_resume(&dwc->xhci->dev);
 
 	return IRQ_HANDLED;
diff --git a/drivers/usb/dwc3/host.c b/drivers/usb/dwc3/host.c
index f56c30cf151e..f6f13e7f1ba1 100644
--- a/drivers/usb/dwc3/host.c
+++ b/drivers/usb/dwc3/host.c
@@ -135,4 +135,5 @@ int dwc3_host_init(struct dwc3 *dwc)
 void dwc3_host_exit(struct dwc3 *dwc)
 {
 	platform_device_unregister(dwc->xhci);
+	dwc->xhci = NULL;
 }
-- 
2.35.1

Re: [PATCH v2 4/9] usb: dwc3: qcom: fix use-after-free on runtime-PM wakeup

[Matthias Kaehlcke]

On Thu, Aug 04, 2022 at 05:09:56PM +0200, Johan Hovold wrote:
> The Qualcomm dwc3 runtime-PM implementation checks the xhci
> platform-device pointer in the wakeup-interrupt handler to determine
> whether the controller is in host mode and if so triggers a resume.
> 
> After a role switch in OTG mode the xhci platform-device would have been
> freed and the next wakeup from runtime suspend would access the freed
> memory.
> 
> Note that role switching is executed from a freezable workqueue, which
> guarantees that the pointer is stable during suspend.
> 
> Also note that runtime PM has been broken since commit 2664deb09306
> ("usb: dwc3: qcom: Honor wakeup enabled/disabled state"), which
> incidentally also prevents this issue from being triggered.
> 
> Fixes: a4333c3a6ba9 ("usb: dwc3: Add Qualcomm DWC3 glue driver")
> Cc: stable@vger.kernel.org      # 4.18
> Signed-off-by: Johan Hovold <johan+linaro@kernel.org>

Reviewed-by: Matthias Kaehlcke <mka@chromium.org>

Re: [PATCH v2 4/9] usb: dwc3: qcom: fix use-after-free on runtime-PM wakeup

[Manivannan Sadhasivam]

On Thu, Aug 04, 2022 at 05:09:56PM +0200, Johan Hovold wrote:
> The Qualcomm dwc3 runtime-PM implementation checks the xhci
> platform-device pointer in the wakeup-interrupt handler to determine
> whether the controller is in host mode and if so triggers a resume.
> 
> After a role switch in OTG mode the xhci platform-device would have been
> freed and the next wakeup from runtime suspend would access the freed
> memory.
> 
> Note that role switching is executed from a freezable workqueue, which
> guarantees that the pointer is stable during suspend.
> 
> Also note that runtime PM has been broken since commit 2664deb09306
> ("usb: dwc3: qcom: Honor wakeup enabled/disabled state"), which
> incidentally also prevents this issue from being triggered.
> 
> Fixes: a4333c3a6ba9 ("usb: dwc3: Add Qualcomm DWC3 glue driver")
> Cc: stable@vger.kernel.org      # 4.18
> Signed-off-by: Johan Hovold <johan+linaro@kernel.org>

It'd be good to mention the introduction of dwc3_qcom_is_host() function.
Initially I thought it is used in a single place, but going through the rest of
the patches reveals that it is used later on.

Reviewed-by: Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org>

Thanks,
Mani

> ---
> 
> Changes in v2
>  - new patch
> 
>  drivers/usb/dwc3/dwc3-qcom.c | 14 +++++++++++++-
>  drivers/usb/dwc3/host.c      |  1 +
>  2 files changed, 14 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/usb/dwc3/dwc3-qcom.c b/drivers/usb/dwc3/dwc3-qcom.c
> index e9364141661b..6884026b9fad 100644
> --- a/drivers/usb/dwc3/dwc3-qcom.c
> +++ b/drivers/usb/dwc3/dwc3-qcom.c
> @@ -298,6 +298,14 @@ static void dwc3_qcom_interconnect_exit(struct dwc3_qcom *qcom)
>  	icc_put(qcom->icc_path_apps);
>  }
>  
> +/* Only usable in contexts where the role can not change. */
> +static bool dwc3_qcom_is_host(struct dwc3_qcom *qcom)
> +{
> +	struct dwc3 *dwc = platform_get_drvdata(qcom->dwc3);
> +
> +	return dwc->xhci;
> +}
> +
>  static enum usb_device_speed dwc3_qcom_read_usb2_speed(struct dwc3_qcom *qcom)
>  {
>  	struct dwc3 *dwc = platform_get_drvdata(qcom->dwc3);
> @@ -460,7 +468,11 @@ static irqreturn_t qcom_dwc3_resume_irq(int irq, void *data)
>  	if (qcom->pm_suspended)
>  		return IRQ_HANDLED;
>  
> -	if (dwc->xhci)
> +	/*
> +	 * This is safe as role switching is done from a freezable workqueue
> +	 * and the wakeup interrupts are disabled as part of resume.
> +	 */
> +	if (dwc3_qcom_is_host(qcom))
>  		pm_runtime_resume(&dwc->xhci->dev);
>  
>  	return IRQ_HANDLED;
> diff --git a/drivers/usb/dwc3/host.c b/drivers/usb/dwc3/host.c
> index f56c30cf151e..f6f13e7f1ba1 100644
> --- a/drivers/usb/dwc3/host.c
> +++ b/drivers/usb/dwc3/host.c
> @@ -135,4 +135,5 @@ int dwc3_host_init(struct dwc3 *dwc)
>  void dwc3_host_exit(struct dwc3 *dwc)
>  {
>  	platform_device_unregister(dwc->xhci);
> +	dwc->xhci = NULL;
>  }
> -- 
> 2.35.1
> 

-- 
மணிவண்ணன் சதாசிவம்

Re: [PATCH v2 4/9] usb: dwc3: qcom: fix use-after-free on runtime-PM wakeup

[Johan Hovold]

On Sat, Aug 06, 2022 at 08:03:11PM +0530, Manivannan Sadhasivam wrote:
> On Thu, Aug 04, 2022 at 05:09:56PM +0200, Johan Hovold wrote:
> > The Qualcomm dwc3 runtime-PM implementation checks the xhci
> > platform-device pointer in the wakeup-interrupt handler to determine
> > whether the controller is in host mode and if so triggers a resume.
> > 
> > After a role switch in OTG mode the xhci platform-device would have been
> > freed and the next wakeup from runtime suspend would access the freed
> > memory.
> > 
> > Note that role switching is executed from a freezable workqueue, which
> > guarantees that the pointer is stable during suspend.
> > 
> > Also note that runtime PM has been broken since commit 2664deb09306
> > ("usb: dwc3: qcom: Honor wakeup enabled/disabled state"), which
> > incidentally also prevents this issue from being triggered.
> > 
> > Fixes: a4333c3a6ba9 ("usb: dwc3: Add Qualcomm DWC3 glue driver")
> > Cc: stable@vger.kernel.org      # 4.18
> > Signed-off-by: Johan Hovold <johan+linaro@kernel.org>
> 
> It'd be good to mention the introduction of dwc3_qcom_is_host() function.
> Initially I thought it is used in a single place, but going through the rest of
> the patches reveals that it is used later on.

I think the helper is warranted on its own as it serves as documentation
of the underlying assumptions that this code relies on.

> > +/* Only usable in contexts where the role can not change. */
> > +static bool dwc3_qcom_is_host(struct dwc3_qcom *qcom)
> > +{
> > +	struct dwc3 *dwc = platform_get_drvdata(qcom->dwc3);
> > +
> > +	return dwc->xhci;
> > +}
> > +
> >  static enum usb_device_speed dwc3_qcom_read_usb2_speed(struct dwc3_qcom *qcom)
> >  {
> >  	struct dwc3 *dwc = platform_get_drvdata(qcom->dwc3);
> > @@ -460,7 +468,11 @@ static irqreturn_t qcom_dwc3_resume_irq(int irq, void *data)
> >  	if (qcom->pm_suspended)
> >  		return IRQ_HANDLED;
> >  
> > -	if (dwc->xhci)
> > +	/*
> > +	 * This is safe as role switching is done from a freezable workqueue
> > +	 * and the wakeup interrupts are disabled as part of resume.
> > +	 */
> > +	if (dwc3_qcom_is_host(qcom))
> >  		pm_runtime_resume(&dwc->xhci->dev);
> >  
> >  	return IRQ_HANDLED;
> > diff --git a/drivers/usb/dwc3/host.c b/drivers/usb/dwc3/host.c
> > index f56c30cf151e..f6f13e7f1ba1 100644
> > --- a/drivers/usb/dwc3/host.c
> > +++ b/drivers/usb/dwc3/host.c
> > @@ -135,4 +135,5 @@ int dwc3_host_init(struct dwc3 *dwc)
> >  void dwc3_host_exit(struct dwc3 *dwc)
> >  {
> >  	platform_device_unregister(dwc->xhci);
> > +	dwc->xhci = NULL;
> >  }
> > -- 
> > 2.35.1

Johan

Re: [PATCH v2 4/9] usb: dwc3: qcom: fix use-after-free on runtime-PM wakeup

[Manivannan Sadhasivam]

On Sat, Aug 06, 2022 at 06:08:51PM +0200, Johan Hovold wrote:
> On Sat, Aug 06, 2022 at 08:03:11PM +0530, Manivannan Sadhasivam wrote:
> > On Thu, Aug 04, 2022 at 05:09:56PM +0200, Johan Hovold wrote:
> > > The Qualcomm dwc3 runtime-PM implementation checks the xhci
> > > platform-device pointer in the wakeup-interrupt handler to determine
> > > whether the controller is in host mode and if so triggers a resume.
> > > 
> > > After a role switch in OTG mode the xhci platform-device would have been
> > > freed and the next wakeup from runtime suspend would access the freed
> > > memory.
> > > 
> > > Note that role switching is executed from a freezable workqueue, which
> > > guarantees that the pointer is stable during suspend.
> > > 
> > > Also note that runtime PM has been broken since commit 2664deb09306
> > > ("usb: dwc3: qcom: Honor wakeup enabled/disabled state"), which
> > > incidentally also prevents this issue from being triggered.
> > > 
> > > Fixes: a4333c3a6ba9 ("usb: dwc3: Add Qualcomm DWC3 glue driver")
> > > Cc: stable@vger.kernel.org      # 4.18
> > > Signed-off-by: Johan Hovold <johan+linaro@kernel.org>
> > 
> > It'd be good to mention the introduction of dwc3_qcom_is_host() function.
> > Initially I thought it is used in a single place, but going through the rest of
> > the patches reveals that it is used later on.
> 
> I think the helper is warranted on its own as it serves as documentation
> of the underlying assumptions that this code relies on.
> 

That's even better.

Thanks,
Mani

> > > +/* Only usable in contexts where the role can not change. */
> > > +static bool dwc3_qcom_is_host(struct dwc3_qcom *qcom)
> > > +{
> > > +	struct dwc3 *dwc = platform_get_drvdata(qcom->dwc3);
> > > +
> > > +	return dwc->xhci;
> > > +}
> > > +
> > >  static enum usb_device_speed dwc3_qcom_read_usb2_speed(struct dwc3_qcom *qcom)
> > >  {
> > >  	struct dwc3 *dwc = platform_get_drvdata(qcom->dwc3);
> > > @@ -460,7 +468,11 @@ static irqreturn_t qcom_dwc3_resume_irq(int irq, void *data)
> > >  	if (qcom->pm_suspended)
> > >  		return IRQ_HANDLED;
> > >  
> > > -	if (dwc->xhci)
> > > +	/*
> > > +	 * This is safe as role switching is done from a freezable workqueue
> > > +	 * and the wakeup interrupts are disabled as part of resume.
> > > +	 */
> > > +	if (dwc3_qcom_is_host(qcom))
> > >  		pm_runtime_resume(&dwc->xhci->dev);
> > >  
> > >  	return IRQ_HANDLED;
> > > diff --git a/drivers/usb/dwc3/host.c b/drivers/usb/dwc3/host.c
> > > index f56c30cf151e..f6f13e7f1ba1 100644
> > > --- a/drivers/usb/dwc3/host.c
> > > +++ b/drivers/usb/dwc3/host.c
> > > @@ -135,4 +135,5 @@ int dwc3_host_init(struct dwc3 *dwc)
> > >  void dwc3_host_exit(struct dwc3 *dwc)
> > >  {
> > >  	platform_device_unregister(dwc->xhci);
> > > +	dwc->xhci = NULL;
> > >  }
> > > -- 
> > > 2.35.1
> 
> Johan

-- 
மணிவண்ணன் சதாசிவம்