usb: chipidea: tegra: fix hardlock with invalid dr mode
diff mbox series

Message ID 20200127023548.27107-1-pgwipeout@gmail.com
State New
Headers show
Series
  • usb: chipidea: tegra: fix hardlock with invalid dr mode
Related show

Commit Message

Peter Geis Jan. 27, 2020, 2:35 a.m. UTC
The ci_hdrc_tegra driver does not currently support dual role mode, but
it does not explicitly prevent its use.
Certain devices support dual role mode, but not automatic role switch.
This can cause the device to lock up during initialization of the
driver.

Detect this state by checking for the extcon phandle when dual role mode
is set to otg.
If it doesn't exist request the driver to only enable manual role
switching.

Fixes: dfebb5f ("usb: chipidea: Add support for Tegra20/30/114/124")
Signed-off-by: Peter Geis <pgwipeout@gmail.com>
---
 drivers/usb/chipidea/ci_hdrc_tegra.c | 7 +++++++
 1 file changed, 7 insertions(+)

Comments

Greg KH Jan. 27, 2020, 7:17 a.m. UTC | #1
On Sun, Jan 26, 2020 at 09:35:48PM -0500, Peter Geis wrote:
> The ci_hdrc_tegra driver does not currently support dual role mode, but
> it does not explicitly prevent its use.
> Certain devices support dual role mode, but not automatic role switch.
> This can cause the device to lock up during initialization of the
> driver.
> 
> Detect this state by checking for the extcon phandle when dual role mode
> is set to otg.
> If it doesn't exist request the driver to only enable manual role
> switching.
> 
> Fixes: dfebb5f ("usb: chipidea: Add support for Tegra20/30/114/124")

Please use 12 digits for kernel sha1 values, this should be:
Fixes: dfebb5f43a78 ("usb: chipidea: Add support for Tegra20/30/114/124")

As proof of the need of this, just using the values you gave here causes
git to complain:

$ git show dfebb5f
error: short SHA1 dfebb5f is ambiguous
hint: The candidates are:
hint:   dfebb5f43a78 commit 2017-08-16 - usb: chipidea: Add support for Tegra20/30/114/124
hint:   dfebb5fec744 tree
fatal: ambiguous argument 'dfebb5f': unknown revision or path not in the working tree.

thanks,

greg k-h
Peter Geis Jan. 27, 2020, 3:25 p.m. UTC | #2
On Mon, Jan 27, 2020 at 2:18 AM Greg KH <gregkh@linuxfoundation.org> wrote:
>
> On Sun, Jan 26, 2020 at 09:35:48PM -0500, Peter Geis wrote:
> > The ci_hdrc_tegra driver does not currently support dual role mode, but
> > it does not explicitly prevent its use.
> > Certain devices support dual role mode, but not automatic role switch.
> > This can cause the device to lock up during initialization of the
> > driver.
> >
> > Detect this state by checking for the extcon phandle when dual role mode
> > is set to otg.
> > If it doesn't exist request the driver to only enable manual role
> > switching.
> >
> > Fixes: dfebb5f ("usb: chipidea: Add support for Tegra20/30/114/124")
>
> Please use 12 digits for kernel sha1 values, this should be:
> Fixes: dfebb5f43a78 ("usb: chipidea: Add support for Tegra20/30/114/124")

Understood, thank you.
I will ensure that is the case in the future.

>
> As proof of the need of this, just using the values you gave here causes
> git to complain:
>
> $ git show dfebb5f
> error: short SHA1 dfebb5f is ambiguous
> hint: The candidates are:
> hint:   dfebb5f43a78 commit 2017-08-16 - usb: chipidea: Add support for Tegra20/30/114/124
> hint:   dfebb5fec744 tree
> fatal: ambiguous argument 'dfebb5f': unknown revision or path not in the working tree.
>
> thanks,
>
> greg k-h
Dmitry Osipenko Jan. 27, 2020, 3:38 p.m. UTC | #3
27.01.2020 18:25, Peter Geis пишет:
> On Mon, Jan 27, 2020 at 2:18 AM Greg KH <gregkh@linuxfoundation.org> wrote:
>>
>> On Sun, Jan 26, 2020 at 09:35:48PM -0500, Peter Geis wrote:
>>> The ci_hdrc_tegra driver does not currently support dual role mode, but
>>> it does not explicitly prevent its use.
>>> Certain devices support dual role mode, but not automatic role switch.
>>> This can cause the device to lock up during initialization of the
>>> driver.
>>>
>>> Detect this state by checking for the extcon phandle when dual role mode
>>> is set to otg.
>>> If it doesn't exist request the driver to only enable manual role
>>> switching.
>>>
>>> Fixes: dfebb5f ("usb: chipidea: Add support for Tegra20/30/114/124")
>>
>> Please use 12 digits for kernel sha1 values, this should be:
>> Fixes: dfebb5f43a78 ("usb: chipidea: Add support for Tegra20/30/114/124")
> 
> Understood, thank you.
> I will ensure that is the case in the future.

Please also take a look at the kernel's scripts/checkpatch.pl, running
'scripts/checkpatch.pl --strict path_to.patch' will help to catch such
things.
Peter Chen Jan. 31, 2020, 5:27 a.m. UTC | #4
On 20-01-26 21:35:48, Peter Geis wrote:
> The ci_hdrc_tegra driver does not currently support dual role mode, but
> it does not explicitly prevent its use.
> Certain devices support dual role mode, but not automatic role switch.
> This can cause the device to lock up during initialization of the
> driver.

If the driver only supports peripheral mode, you could set dr_mode as
peripheral-only at glue layer, it would not be override by core.c.
See ci_get_platdata.

But one thing I could not understand, if the driver doesn't support
dual-role, how could you do manual role switch?

> 
> Detect this state by checking for the extcon phandle when dual role mode
> is set to otg.
> If it doesn't exist request the driver to only enable manual role
> switching.
> 
> Fixes: dfebb5f ("usb: chipidea: Add support for Tegra20/30/114/124")
> Signed-off-by: Peter Geis <pgwipeout@gmail.com>
> ---
>  drivers/usb/chipidea/ci_hdrc_tegra.c | 7 +++++++
>  1 file changed, 7 insertions(+)
> 
> diff --git a/drivers/usb/chipidea/ci_hdrc_tegra.c b/drivers/usb/chipidea/ci_hdrc_tegra.c
> index 7455df0ede49..2f6f6ce3e8f5 100644
> --- a/drivers/usb/chipidea/ci_hdrc_tegra.c
> +++ b/drivers/usb/chipidea/ci_hdrc_tegra.c
> @@ -89,6 +89,13 @@ static int tegra_udc_probe(struct platform_device *pdev)
>  	udc->data.usb_phy = udc->phy;
>  	udc->data.capoffset = DEF_CAPOFFSET;
>  
> +	/* check the dual mode and warn about bad configurations */
> +	if (usb_get_dr_mode(&pdev->dev) == USB_DR_MODE_OTG &&
> +	   !of_property_read_bool(pdev->dev.of_node, "extcon")) {
> +		dev_warn(&pdev->dev, "no extcon registered, otg unavailable");
> +		udc->data.flags |= CI_HDRC_DUAL_ROLE_NOT_OTG;
> +	}
> +

The CI_HDRC_DUAL_ROLE_NOT_OTG flag may not be suitable here, please see
comments for it.
 
>  	udc->dev = ci_hdrc_add_device(&pdev->dev, pdev->resource,
>  				      pdev->num_resources, &udc->data);
>  	if (IS_ERR(udc->dev)) {
> -- 
> 2.17.1
>
Peter Geis Jan. 31, 2020, 8:43 p.m. UTC | #5
On Fri, Jan 31, 2020 at 12:27 AM Peter Chen <peter.chen@nxp.com> wrote:
>
> On 20-01-26 21:35:48, Peter Geis wrote:
> > The ci_hdrc_tegra driver does not currently support dual role mode, but
> > it does not explicitly prevent its use.
> > Certain devices support dual role mode, but not automatic role switch.
> > This can cause the device to lock up during initialization of the
> > driver.
>
> If the driver only supports peripheral mode, you could set dr_mode as
> peripheral-only at glue layer, it would not be override by core.c.
> See ci_get_platdata.

The mode is set via the device tree dr_mode property.
Even though the current tegra_udc driver does not currently support
mode switching, board device tree files such as the apalis-eval and
colibri-eval-v3 have dr_mode set to otg.

They also seem to be missing the extcon phandle, which means automatic
mode switching is not possible?

>
> But one thing I could not understand, if the driver doesn't support
> dual-role, how could you do manual role switch?

Manual role switching is conducted via debugfs,
/sys/kernel/debug/usb/ci_hdrc.0/role

>
> >
> > Detect this state by checking for the extcon phandle when dual role mode
> > is set to otg.
> > If it doesn't exist request the driver to only enable manual role
> > switching.
> >
> > Fixes: dfebb5f ("usb: chipidea: Add support for Tegra20/30/114/124")
> > Signed-off-by: Peter Geis <pgwipeout@gmail.com>
> > ---
> >  drivers/usb/chipidea/ci_hdrc_tegra.c | 7 +++++++
> >  1 file changed, 7 insertions(+)
> >
> > diff --git a/drivers/usb/chipidea/ci_hdrc_tegra.c b/drivers/usb/chipidea/ci_hdrc_tegra.c
> > index 7455df0ede49..2f6f6ce3e8f5 100644
> > --- a/drivers/usb/chipidea/ci_hdrc_tegra.c
> > +++ b/drivers/usb/chipidea/ci_hdrc_tegra.c
> > @@ -89,6 +89,13 @@ static int tegra_udc_probe(struct platform_device *pdev)
> >       udc->data.usb_phy = udc->phy;
> >       udc->data.capoffset = DEF_CAPOFFSET;
> >
> > +     /* check the dual mode and warn about bad configurations */
> > +     if (usb_get_dr_mode(&pdev->dev) == USB_DR_MODE_OTG &&
> > +        !of_property_read_bool(pdev->dev.of_node, "extcon")) {
> > +             dev_warn(&pdev->dev, "no extcon registered, otg unavailable");
> > +             udc->data.flags |= CI_HDRC_DUAL_ROLE_NOT_OTG;
> > +     }
> > +
>
> The CI_HDRC_DUAL_ROLE_NOT_OTG flag may not be suitable here, please see
> comments for it.

I've dug around the various mailing lists and I fail to find any
reference to why this is not a valid use case.
The commit message specifically references dual role capable devices
without proper otg implementations, such as missing the otgsc
register.

My current use case is the Ouya, which deadlocks the kernel durning
probe if the otg capable usb controller is set to dr_mode=otg.
It works with this flag set.
Unfortunately there isn't a property for dr_mode set to non_otg_dr_mode.

I found a post stating that the driver blindly touches registers in
otg mode, which leads to the deadlock if those registers are read only
or non-existent,
Eventually someone should look into why this deadlock occurs and make
a proper fix that applies to all users.
Unfortunately I do not have the knowledge yet to accomplish this task.

With some simple modifications to the tegra_udc driver it is possible
to get host mode working, vice using the tegra-ehci driver.
At this point role switch works

I also managed to move all of the functionality of the tegra-ehci
driver into the tegra-udc driver.
Unfortunately it doesn't behave correctly under some cases, so I never
submitted it.

Do you have a recommendation for handling this?

>
> >       udc->dev = ci_hdrc_add_device(&pdev->dev, pdev->resource,
> >                                     pdev->num_resources, &udc->data);
> >       if (IS_ERR(udc->dev)) {
> > --
> > 2.17.1
> >
>
> --
>
> Thanks,
> Peter Chen
Peter Chen Feb. 1, 2020, 6:13 a.m. UTC | #6
On 20-01-31 15:43:24, Peter Geis wrote:
> On Fri, Jan 31, 2020 at 12:27 AM Peter Chen <peter.chen@nxp.com> wrote:
> >
> > On 20-01-26 21:35:48, Peter Geis wrote:
> > > The ci_hdrc_tegra driver does not currently support dual role mode, but
> > > it does not explicitly prevent its use.
> > > Certain devices support dual role mode, but not automatic role switch.
> > > This can cause the device to lock up during initialization of the
> > > driver.
> >
> > If the driver only supports peripheral mode, you could set dr_mode as
> > peripheral-only at glue layer, it would not be override by core.c.
> > See ci_get_platdata.
> 
> The mode is set via the device tree dr_mode property.
> Even though the current tegra_udc driver does not currently support
> mode switching, board device tree files such as the apalis-eval and
> colibri-eval-v3 have dr_mode set to otg.
> 

If you would like fixing wrong dts setting issue, you could do below:

diff --git a/drivers/usb/chipidea/ci_hdrc_tegra.c b/drivers/usb/chipidea/ci_hdrc_tegra.c
index 0c9911d44ee5..5d85363ad0f4 100644
--- a/drivers/usb/chipidea/ci_hdrc_tegra.c
+++ b/drivers/usb/chipidea/ci_hdrc_tegra.c
@@ -95,6 +95,7 @@ static int tegra_udc_probe(struct platform_device *pdev)
 	udc->data.flags = soc->flags;
 	udc->data.usb_phy = udc->phy;
 	udc->data.capoffset = DEF_CAPOFFSET;
+	udc->data.dr_mode = USB_DR_MODE_PERIPHERAL;
 
 	udc->dev = ci_hdrc_add_device(&pdev->dev, pdev->resource,
 				      pdev->num_resources, &udc->data);

> They also seem to be missing the extcon phandle, which means automatic
> mode switching is not possible?
> 
> >
> > But one thing I could not understand, if the driver doesn't support
> > dual-role, how could you do manual role switch?
> 
> Manual role switching is conducted via debugfs,
> /sys/kernel/debug/usb/ci_hdrc.0/role
> 
> >
> > >
> > > Detect this state by checking for the extcon phandle when dual role mode
> > > is set to otg.
> > > If it doesn't exist request the driver to only enable manual role
> > > switching.
> > >
> > > Fixes: dfebb5f ("usb: chipidea: Add support for Tegra20/30/114/124")
> > > Signed-off-by: Peter Geis <pgwipeout@gmail.com>
> > > ---
> > >  drivers/usb/chipidea/ci_hdrc_tegra.c | 7 +++++++
> > >  1 file changed, 7 insertions(+)
> > >
> > > diff --git a/drivers/usb/chipidea/ci_hdrc_tegra.c b/drivers/usb/chipidea/ci_hdrc_tegra.c
> > > index 7455df0ede49..2f6f6ce3e8f5 100644
> > > --- a/drivers/usb/chipidea/ci_hdrc_tegra.c
> > > +++ b/drivers/usb/chipidea/ci_hdrc_tegra.c
> > > @@ -89,6 +89,13 @@ static int tegra_udc_probe(struct platform_device *pdev)
> > >       udc->data.usb_phy = udc->phy;
> > >       udc->data.capoffset = DEF_CAPOFFSET;
> > >
> > > +     /* check the dual mode and warn about bad configurations */
> > > +     if (usb_get_dr_mode(&pdev->dev) == USB_DR_MODE_OTG &&
> > > +        !of_property_read_bool(pdev->dev.of_node, "extcon")) {
> > > +             dev_warn(&pdev->dev, "no extcon registered, otg unavailable");
> > > +             udc->data.flags |= CI_HDRC_DUAL_ROLE_NOT_OTG;
> > > +     }
> > > +
> >
> > The CI_HDRC_DUAL_ROLE_NOT_OTG flag may not be suitable here, please see
> > comments for it.
> 
> I've dug around the various mailing lists and I fail to find any
> reference to why this is not a valid use case.
> The commit message specifically references dual role capable devices
> without proper otg implementations, such as missing the otgsc
> register.
> 
> My current use case is the Ouya, which deadlocks the kernel durning
> probe if the otg capable usb controller is set to dr_mode=otg.
> It works with this flag set.
> Unfortunately there isn't a property for dr_mode set to non_otg_dr_mode.
> 
> I found a post stating that the driver blindly touches registers in
> otg mode, which leads to the deadlock if those registers are read only
> or non-existent,
> Eventually someone should look into why this deadlock occurs and make
> a proper fix that applies to all users.
> Unfortunately I do not have the knowledge yet to accomplish this task.
> 
> With some simple modifications to the tegra_udc driver it is possible
> to get host mode working, vice using the tegra-ehci driver.
> At this point role switch works
> 
> I also managed to move all of the functionality of the tegra-ehci
> driver into the tegra-udc driver.
> Unfortunately it doesn't behave correctly under some cases, so I never
> submitted it.
> 
> Do you have a recommendation for handling this?

If you would like adding host function in chipidea driver for tegra, and
want dual-role switch to work, first, you may need to know which
dual-role switch you need:
- Through controller's id and vbus pins
- Through extcon, eg,GPIO.
- Through usb-role-switch class, eg, Type-C controller
- Through sysfs, /sys/bus/platform/devices/ci_hdrc.0/role

The first three are automatically switch, the last one is manually.
Current chipidea core should support all above switches, maybe there are
some limitations for them, the fixes are welcome.

Second, you may check if touch otgsc will hang or deadlock the system.
If you can't touch otgsc when portsc.phcd = 0, you may need the flag
CI_HDRC_DUAL_ROLE_NOT_OTG, afaik, few SoCs can't touch otgsc if it
supports dual-role.

When you add host function, you may set dr_mode as host-only first, it
could make thing easier.
Peter Geis Feb. 3, 2020, 5:22 p.m. UTC | #7
Resending due to html screwup, apologies.

On Sat, Feb 1, 2020 at 1:13 AM Peter Chen <peter.chen@nxp.com> wrote:
>
> On 20-01-31 15:43:24, Peter Geis wrote:
> > On Fri, Jan 31, 2020 at 12:27 AM Peter Chen <peter.chen@nxp.com> wrote:
> > >
> > > On 20-01-26 21:35:48, Peter Geis wrote:
> > > > The ci_hdrc_tegra driver does not currently support dual role mode, but
> > > > it does not explicitly prevent its use.
> > > > Certain devices support dual role mode, but not automatic role switch.
> > > > This can cause the device to lock up during initialization of the
> > > > driver.
> > >
> > > If the driver only supports peripheral mode, you could set dr_mode as
> > > peripheral-only at glue layer, it would not be override by core.c.
> > > See ci_get_platdata.
> >
> > The mode is set via the device tree dr_mode property.
> > Even though the current tegra_udc driver does not currently support
> > mode switching, board device tree files such as the apalis-eval and
> > colibri-eval-v3 have dr_mode set to otg.
> >
>
> If you would like fixing wrong dts setting issue, you could do below:
>
> diff --git a/drivers/usb/chipidea/ci_hdrc_tegra.c b/drivers/usb/chipidea/ci_hdrc_tegra.c
> index 0c9911d44ee5..5d85363ad0f4 100644
> --- a/drivers/usb/chipidea/ci_hdrc_tegra.c
> +++ b/drivers/usb/chipidea/ci_hdrc_tegra.c
> @@ -95,6 +95,7 @@ static int tegra_udc_probe(struct platform_device *pdev)
>         udc->data.flags = soc->flags;
>         udc->data.usb_phy = udc->phy;
>         udc->data.capoffset = DEF_CAPOFFSET;
> +       udc->data.dr_mode = USB_DR_MODE_PERIPHERAL;
>
>         udc->dev = ci_hdrc_add_device(&pdev->dev, pdev->resource,
>                                       pdev->num_resources, &udc->data);

I do not wish to set the dr_mode manually, since this could break
things on devices I do not own.
Future work is needed to address the tegra_udc capabilities, as well
as fully correct the hang issue.
This patch merely aims to fix a corner case.

>
> > They also seem to be missing the extcon phandle, which means automatic
> > mode switching is not possible?
> >
> > >
> > > But one thing I could not understand, if the driver doesn't support
> > > dual-role, how could you do manual role switch?
> >
> > Manual role switching is conducted via debugfs,
> > /sys/kernel/debug/usb/ci_hdrc.0/role
> >
> > >
> > > >
> > > > Detect this state by checking for the extcon phandle when dual role mode
> > > > is set to otg.
> > > > If it doesn't exist request the driver to only enable manual role
> > > > switching.
> > > >
> > > > Fixes: dfebb5f ("usb: chipidea: Add support for Tegra20/30/114/124")
> > > > Signed-off-by: Peter Geis <pgwipeout@gmail.com>
> > > > ---
> > > >  drivers/usb/chipidea/ci_hdrc_tegra.c | 7 +++++++
> > > >  1 file changed, 7 insertions(+)
> > > >
> > > > diff --git a/drivers/usb/chipidea/ci_hdrc_tegra.c b/drivers/usb/chipidea/ci_hdrc_tegra.c
> > > > index 7455df0ede49..2f6f6ce3e8f5 100644
> > > > --- a/drivers/usb/chipidea/ci_hdrc_tegra.c
> > > > +++ b/drivers/usb/chipidea/ci_hdrc_tegra.c
> > > > @@ -89,6 +89,13 @@ static int tegra_udc_probe(struct platform_device *pdev)
> > > >       udc->data.usb_phy = udc->phy;
> > > >       udc->data.capoffset = DEF_CAPOFFSET;
> > > >
> > > > +     /* check the dual mode and warn about bad configurations */
> > > > +     if (usb_get_dr_mode(&pdev->dev) == USB_DR_MODE_OTG &&
> > > > +        !of_property_read_bool(pdev->dev.of_node, "extcon")) {
> > > > +             dev_warn(&pdev->dev, "no extcon registered, otg unavailable");
> > > > +             udc->data.flags |= CI_HDRC_DUAL_ROLE_NOT_OTG;
> > > > +     }
> > > > +
> > >
> > > The CI_HDRC_DUAL_ROLE_NOT_OTG flag may not be suitable here, please see
> > > comments for it.
> >
> > I've dug around the various mailing lists and I fail to find any
> > reference to why this is not a valid use case.
> > The commit message specifically references dual role capable devices
> > without proper otg implementations, such as missing the otgsc
> > register.
> >
> > My current use case is the Ouya, which deadlocks the kernel durning
> > probe if the otg capable usb controller is set to dr_mode=otg.
> > It works with this flag set.
> > Unfortunately there isn't a property for dr_mode set to non_otg_dr_mode.
> >
> > I found a post stating that the driver blindly touches registers in
> > otg mode, which leads to the deadlock if those registers are read only
> > or non-existent,
> > Eventually someone should look into why this deadlock occurs and make
> > a proper fix that applies to all users.
> > Unfortunately I do not have the knowledge yet to accomplish this task.
> >
> > With some simple modifications to the tegra_udc driver it is possible
> > to get host mode working, vice using the tegra-ehci driver.
> > At this point role switch works
> >
> > I also managed to move all of the functionality of the tegra-ehci
> > driver into the tegra-udc driver.
> > Unfortunately it doesn't behave correctly under some cases, so I never
> > submitted it.
> >
> > Do you have a recommendation for handling this?
>
> If you would like adding host function in chipidea driver for tegra, and
> want dual-role switch to work, first, you may need to know which
> dual-role switch you need:
> - Through controller's id and vbus pins
> - Through extcon, eg,GPIO.
> - Through usb-role-switch class, eg, Type-C controller
> - Through sysfs, /sys/bus/platform/devices/ci_hdrc.0/role
>
> The first three are automatically switch, the last one is manually.
> Current chipidea core should support all above switches, maybe there are
> some limitations for them, the fixes are welcome.

AFAIK reading through the comments, the chipidea driver intended to
exclusively use extcon for automatic role switching, please correct me
if I'm wrong here.
USB-C support is not a likely scenario, as currently the chipidea
driver only supports usb 2.0.

>
> Second, you may check if touch otgsc will hang or deadlock the system.
> If you can't touch otgsc when portsc.phcd = 0, you may need the flag
> CI_HDRC_DUAL_ROLE_NOT_OTG, afaik, few SoCs can't touch otgsc if it
> supports dual-role.

I added some traces to the driver, and it doesn't actually appear to
be a register read/write that is breaking things. (Not directly
anyways).
The hang occurs after it enumerates the usb gadgets and hub.
Still trying to track down exactly where the hang occurs.

>
> When you add host function, you may set dr_mode as host-only first, it
> could make thing easier.
>
> --
>
> Thanks,
> Peter Chen
Peter Chen Feb. 4, 2020, 7:44 a.m. UTC | #8
On 20-02-03 12:22:37, Peter Geis wrote:
> > diff --git a/drivers/usb/chipidea/ci_hdrc_tegra.c b/drivers/usb/chipidea/ci_hdrc_tegra.c
> > index 0c9911d44ee5..5d85363ad0f4 100644
> > --- a/drivers/usb/chipidea/ci_hdrc_tegra.c
> > +++ b/drivers/usb/chipidea/ci_hdrc_tegra.c
> > @@ -95,6 +95,7 @@ static int tegra_udc_probe(struct platform_device *pdev)
> >         udc->data.flags = soc->flags;
> >         udc->data.usb_phy = udc->phy;
> >         udc->data.capoffset = DEF_CAPOFFSET;
> > +       udc->data.dr_mode = USB_DR_MODE_PERIPHERAL;
> >
> >         udc->dev = ci_hdrc_add_device(&pdev->dev, pdev->resource,
> >                                       pdev->num_resources, &udc->data);
> 
> I do not wish to set the dr_mode manually, since this could break
> things on devices I do not own.
> Future work is needed to address the tegra_udc capabilities, as well
> as fully correct the hang issue.
> This patch merely aims to fix a corner case.
> 
> >
> > > They also seem to be missing the extcon phandle, which means automatic
> > > mode switching is not possible?
> > >
> > > >
> > > > But one thing I could not understand, if the driver doesn't support
> > > > dual-role, how could you do manual role switch?
> > >
> > > Manual role switching is conducted via debugfs,
> > > /sys/kernel/debug/usb/ci_hdrc.0/role
> > >
> > > >
> > > > >
> > > > > Detect this state by checking for the extcon phandle when dual role mode
> > > > > is set to otg.
> > > > > If it doesn't exist request the driver to only enable manual role
> > > > > switching.
> > > > >
> > > > > Fixes: dfebb5f ("usb: chipidea: Add support for Tegra20/30/114/124")
> > > > > Signed-off-by: Peter Geis <pgwipeout@gmail.com>
> > > > > ---
> > > > >  drivers/usb/chipidea/ci_hdrc_tegra.c | 7 +++++++
> > > > >  1 file changed, 7 insertions(+)
> > > > >
> > > > > diff --git a/drivers/usb/chipidea/ci_hdrc_tegra.c b/drivers/usb/chipidea/ci_hdrc_tegra.c
> > > > > index 7455df0ede49..2f6f6ce3e8f5 100644
> > > > > --- a/drivers/usb/chipidea/ci_hdrc_tegra.c
> > > > > +++ b/drivers/usb/chipidea/ci_hdrc_tegra.c
> > > > > @@ -89,6 +89,13 @@ static int tegra_udc_probe(struct platform_device *pdev)
> > > > >       udc->data.usb_phy = udc->phy;
> > > > >       udc->data.capoffset = DEF_CAPOFFSET;
> > > > >
> > > > > +     /* check the dual mode and warn about bad configurations */
> > > > > +     if (usb_get_dr_mode(&pdev->dev) == USB_DR_MODE_OTG &&
> > > > > +        !of_property_read_bool(pdev->dev.of_node, "extcon")) {
> > > > > +             dev_warn(&pdev->dev, "no extcon registered, otg unavailable");
> > > > > +             udc->data.flags |= CI_HDRC_DUAL_ROLE_NOT_OTG;
> > > > > +     }
> > > > > +
> > > >
> > > > The CI_HDRC_DUAL_ROLE_NOT_OTG flag may not be suitable here, please see
> > > > comments for it.
> > >
> > > I've dug around the various mailing lists and I fail to find any
> > > reference to why this is not a valid use case.
> > > The commit message specifically references dual role capable devices
> > > without proper otg implementations, such as missing the otgsc
> > > register.
> > >
> > > My current use case is the Ouya, which deadlocks the kernel durning
> > > probe if the otg capable usb controller is set to dr_mode=otg.
> > > It works with this flag set.
> > > Unfortunately there isn't a property for dr_mode set to non_otg_dr_mode.
> > >
> > > I found a post stating that the driver blindly touches registers in
> > > otg mode, which leads to the deadlock if those registers are read only
> > > or non-existent,
> > > Eventually someone should look into why this deadlock occurs and make
> > > a proper fix that applies to all users.
> > > Unfortunately I do not have the knowledge yet to accomplish this task.
> > >
> > > With some simple modifications to the tegra_udc driver it is possible
> > > to get host mode working, vice using the tegra-ehci driver.
> > > At this point role switch works
> > >
> > > I also managed to move all of the functionality of the tegra-ehci
> > > driver into the tegra-udc driver.
> > > Unfortunately it doesn't behave correctly under some cases, so I never
> > > submitted it.
> > >
> > > Do you have a recommendation for handling this?
> >
> > If you would like adding host function in chipidea driver for tegra, and
> > want dual-role switch to work, first, you may need to know which
> > dual-role switch you need:
> > - Through controller's id and vbus pins
> > - Through extcon, eg,GPIO.
> > - Through usb-role-switch class, eg, Type-C controller
> > - Through sysfs, /sys/bus/platform/devices/ci_hdrc.0/role
> >
> > The first three are automatically switch, the last one is manually.
> > Current chipidea core should support all above switches, maybe there are
> > some limitations for them, the fixes are welcome.
> 
> AFAIK reading through the comments, the chipidea driver intended to
> exclusively use extcon for automatic role switching, please correct me
> if I'm wrong here.

No, extcon is supported.

> USB-C support is not a likely scenario, as currently the chipidea
> driver only supports usb 2.0.

USB-C is a connector, not related to USB HW revision. For USB 2.0
USB-C solution, there is no SS TX/RX connected on connector.

> 
> >
> > Second, you may check if touch otgsc will hang or deadlock the system.
> > If you can't touch otgsc when portsc.phcd = 0, you may need the flag
> > CI_HDRC_DUAL_ROLE_NOT_OTG, afaik, few SoCs can't touch otgsc if it
> > supports dual-role.
> 
> I added some traces to the driver, and it doesn't actually appear to
> be a register read/write that is breaking things. (Not directly
> anyways).
> The hang occurs after it enumerates the usb gadgets and hub.
> Still trying to track down exactly where the hang occurs.
> 

Try to see if it is related to runtime power management.
Peter Geis Feb. 4, 2020, 2:35 p.m. UTC | #9
On Tue, Feb 4, 2020 at 2:44 AM Peter Chen <peter.chen@nxp.com> wrote:
>
> On 20-02-03 12:22:37, Peter Geis wrote:
> > > diff --git a/drivers/usb/chipidea/ci_hdrc_tegra.c b/drivers/usb/chipidea/ci_hdrc_tegra.c
> > > index 0c9911d44ee5..5d85363ad0f4 100644
> > > --- a/drivers/usb/chipidea/ci_hdrc_tegra.c
> > > +++ b/drivers/usb/chipidea/ci_hdrc_tegra.c
> > > @@ -95,6 +95,7 @@ static int tegra_udc_probe(struct platform_device *pdev)
> > >         udc->data.flags = soc->flags;
> > >         udc->data.usb_phy = udc->phy;
> > >         udc->data.capoffset = DEF_CAPOFFSET;
> > > +       udc->data.dr_mode = USB_DR_MODE_PERIPHERAL;
> > >
> > >         udc->dev = ci_hdrc_add_device(&pdev->dev, pdev->resource,
> > >                                       pdev->num_resources, &udc->data);
> >
> > I do not wish to set the dr_mode manually, since this could break
> > things on devices I do not own.
> > Future work is needed to address the tegra_udc capabilities, as well
> > as fully correct the hang issue.
> > This patch merely aims to fix a corner case.
> >
> > >
> > > > They also seem to be missing the extcon phandle, which means automatic
> > > > mode switching is not possible?
> > > >
> > > > >
> > > > > But one thing I could not understand, if the driver doesn't support
> > > > > dual-role, how could you do manual role switch?
> > > >
> > > > Manual role switching is conducted via debugfs,
> > > > /sys/kernel/debug/usb/ci_hdrc.0/role
> > > >
> > > > >
> > > > > >
> > > > > > Detect this state by checking for the extcon phandle when dual role mode
> > > > > > is set to otg.
> > > > > > If it doesn't exist request the driver to only enable manual role
> > > > > > switching.
> > > > > >
> > > > > > Fixes: dfebb5f ("usb: chipidea: Add support for Tegra20/30/114/124")
> > > > > > Signed-off-by: Peter Geis <pgwipeout@gmail.com>
> > > > > > ---
> > > > > >  drivers/usb/chipidea/ci_hdrc_tegra.c | 7 +++++++
> > > > > >  1 file changed, 7 insertions(+)
> > > > > >
> > > > > > diff --git a/drivers/usb/chipidea/ci_hdrc_tegra.c b/drivers/usb/chipidea/ci_hdrc_tegra.c
> > > > > > index 7455df0ede49..2f6f6ce3e8f5 100644
> > > > > > --- a/drivers/usb/chipidea/ci_hdrc_tegra.c
> > > > > > +++ b/drivers/usb/chipidea/ci_hdrc_tegra.c
> > > > > > @@ -89,6 +89,13 @@ static int tegra_udc_probe(struct platform_device *pdev)
> > > > > >       udc->data.usb_phy = udc->phy;
> > > > > >       udc->data.capoffset = DEF_CAPOFFSET;
> > > > > >
> > > > > > +     /* check the dual mode and warn about bad configurations */
> > > > > > +     if (usb_get_dr_mode(&pdev->dev) == USB_DR_MODE_OTG &&
> > > > > > +        !of_property_read_bool(pdev->dev.of_node, "extcon")) {
> > > > > > +             dev_warn(&pdev->dev, "no extcon registered, otg unavailable");
> > > > > > +             udc->data.flags |= CI_HDRC_DUAL_ROLE_NOT_OTG;
> > > > > > +     }
> > > > > > +
> > > > >
> > > > > The CI_HDRC_DUAL_ROLE_NOT_OTG flag may not be suitable here, please see
> > > > > comments for it.
> > > >
> > > > I've dug around the various mailing lists and I fail to find any
> > > > reference to why this is not a valid use case.
> > > > The commit message specifically references dual role capable devices
> > > > without proper otg implementations, such as missing the otgsc
> > > > register.
> > > >
> > > > My current use case is the Ouya, which deadlocks the kernel durning
> > > > probe if the otg capable usb controller is set to dr_mode=otg.
> > > > It works with this flag set.
> > > > Unfortunately there isn't a property for dr_mode set to non_otg_dr_mode.
> > > >
> > > > I found a post stating that the driver blindly touches registers in
> > > > otg mode, which leads to the deadlock if those registers are read only
> > > > or non-existent,
> > > > Eventually someone should look into why this deadlock occurs and make
> > > > a proper fix that applies to all users.
> > > > Unfortunately I do not have the knowledge yet to accomplish this task.
> > > >
> > > > With some simple modifications to the tegra_udc driver it is possible
> > > > to get host mode working, vice using the tegra-ehci driver.
> > > > At this point role switch works
> > > >
> > > > I also managed to move all of the functionality of the tegra-ehci
> > > > driver into the tegra-udc driver.
> > > > Unfortunately it doesn't behave correctly under some cases, so I never
> > > > submitted it.
> > > >
> > > > Do you have a recommendation for handling this?
> > >
> > > If you would like adding host function in chipidea driver for tegra, and
> > > want dual-role switch to work, first, you may need to know which
> > > dual-role switch you need:
> > > - Through controller's id and vbus pins
> > > - Through extcon, eg,GPIO.
> > > - Through usb-role-switch class, eg, Type-C controller
> > > - Through sysfs, /sys/bus/platform/devices/ci_hdrc.0/role
> > >
> > > The first three are automatically switch, the last one is manually.
> > > Current chipidea core should support all above switches, maybe there are
> > > some limitations for them, the fixes are welcome.
> >
> > AFAIK reading through the comments, the chipidea driver intended to
> > exclusively use extcon for automatic role switching, please correct me
> > if I'm wrong here.
>
> No, extcon is supported.
>
> > USB-C support is not a likely scenario, as currently the chipidea
> > driver only supports usb 2.0.
>
> USB-C is a connector, not related to USB HW revision. For USB 2.0
> USB-C solution, there is no SS TX/RX connected on connector.
>
> >
> > >
> > > Second, you may check if touch otgsc will hang or deadlock the system.
> > > If you can't touch otgsc when portsc.phcd = 0, you may need the flag
> > > CI_HDRC_DUAL_ROLE_NOT_OTG, afaik, few SoCs can't touch otgsc if it
> > > supports dual-role.
> >
> > I added some traces to the driver, and it doesn't actually appear to
> > be a register read/write that is breaking things. (Not directly
> > anyways).
> > The hang occurs after it enumerates the usb gadgets and hub.
> > Still trying to track down exactly where the hang occurs.
> >
>
> Try to see if it is related to runtime power management.

That's it!
There doesn't appear to be a method to disable PM inside the chipidea driver.
Do you have a suggestion on how to do it, aside from the global method?

>
> --
>
> Thanks,
> Peter Chen
Peter Chen Feb. 5, 2020, 11:17 a.m. UTC | #10
> > > > Second, you may check if touch otgsc will hang or deadlock the system.
> > > > If you can't touch otgsc when portsc.phcd = 0, you may need the
> > > > flag CI_HDRC_DUAL_ROLE_NOT_OTG, afaik, few SoCs can't touch otgsc
> > > > if it supports dual-role.
> > >
> > > I added some traces to the driver, and it doesn't actually appear to
> > > be a register read/write that is breaking things. (Not directly
> > > anyways).
> > > The hang occurs after it enumerates the usb gadgets and hub.
> > > Still trying to track down exactly where the hang occurs.
> > >
> >
> > Try to see if it is related to runtime power management.
> 
> That's it!
> There doesn't appear to be a method to disable PM inside the chipidea driver.
> Do you have a suggestion on how to do it, aside from the global method?
> 
 

See drivers/usb/chipidea/ci_hdrc_imx.c for reference please.

Peter

Patch
diff mbox series

diff --git a/drivers/usb/chipidea/ci_hdrc_tegra.c b/drivers/usb/chipidea/ci_hdrc_tegra.c
index 7455df0ede49..2f6f6ce3e8f5 100644
--- a/drivers/usb/chipidea/ci_hdrc_tegra.c
+++ b/drivers/usb/chipidea/ci_hdrc_tegra.c
@@ -89,6 +89,13 @@  static int tegra_udc_probe(struct platform_device *pdev)
 	udc->data.usb_phy = udc->phy;
 	udc->data.capoffset = DEF_CAPOFFSET;
 
+	/* check the dual mode and warn about bad configurations */
+	if (usb_get_dr_mode(&pdev->dev) == USB_DR_MODE_OTG &&
+	   !of_property_read_bool(pdev->dev.of_node, "extcon")) {
+		dev_warn(&pdev->dev, "no extcon registered, otg unavailable");
+		udc->data.flags |= CI_HDRC_DUAL_ROLE_NOT_OTG;
+	}
+
 	udc->dev = ci_hdrc_add_device(&pdev->dev, pdev->resource,
 				      pdev->num_resources, &udc->data);
 	if (IS_ERR(udc->dev)) {