2014-10-14 20:14:30

by David Cohen

[permalink] [raw]
Subject: [PATCH] usb: dwc3: be more verbose on ERRATIC_ERROR interrupt

ERRATIC_ERROR interrupt is an event that needs more attention from
developers than currently implemented, since this indicates a serious
stability issue. The only way to get warned about it is by selecting the
maximum driver's verbosity.

This patch increases a bit the error's verbosity.

Signed-off-by: David Cohen <[email protected]>
---
drivers/usb/dwc3/gadget.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/usb/dwc3/gadget.c b/drivers/usb/dwc3/gadget.c
index 3818b26bfc05..132e761d62e4 100644
--- a/drivers/usb/dwc3/gadget.c
+++ b/drivers/usb/dwc3/gadget.c
@@ -2484,7 +2484,8 @@ static void dwc3_gadget_interrupt(struct dwc3 *dwc,
dev_vdbg(dwc->dev, "Start of Periodic Frame\n");
break;
case DWC3_DEVICE_EVENT_ERRATIC_ERROR:
- dev_vdbg(dwc->dev, "Erratic Error\n");
+ WARN_ON_ONCE(1);
+ dev_dbg(dwc->dev, "Erratic Error\n");
break;
case DWC3_DEVICE_EVENT_CMD_CMPL:
dev_vdbg(dwc->dev, "Command Complete\n");
--
2.1.0


2014-10-15 05:00:32

by Felipe Balbi

[permalink] [raw]
Subject: Re: [PATCH] usb: dwc3: be more verbose on ERRATIC_ERROR interrupt

Hi,

On Tue, Oct 14, 2014 at 01:15:05PM -0700, David Cohen wrote:
> ERRATIC_ERROR interrupt is an event that needs more attention from
> developers than currently implemented, since this indicates a serious
> stability issue. The only way to get warned about it is by selecting the
> maximum driver's verbosity.
>
> This patch increases a bit the error's verbosity.
>
> Signed-off-by: David Cohen <[email protected]>
> ---
> drivers/usb/dwc3/gadget.c | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/usb/dwc3/gadget.c b/drivers/usb/dwc3/gadget.c
> index 3818b26bfc05..132e761d62e4 100644
> --- a/drivers/usb/dwc3/gadget.c
> +++ b/drivers/usb/dwc3/gadget.c
> @@ -2484,7 +2484,8 @@ static void dwc3_gadget_interrupt(struct dwc3 *dwc,
> dev_vdbg(dwc->dev, "Start of Periodic Frame\n");
> break;
> case DWC3_DEVICE_EVENT_ERRATIC_ERROR:
> - dev_vdbg(dwc->dev, "Erratic Error\n");
> + WARN_ON_ONCE(1);
> + dev_dbg(dwc->dev, "Erratic Error\n");

how about:
WARN_ONCE(true, "Erratic Error\n");

instead ?

> break;
> case DWC3_DEVICE_EVENT_CMD_CMPL:
> dev_vdbg(dwc->dev, "Command Complete\n");
> --
> 2.1.0
>

--
balbi


Attachments:
(No filename) (1.16 kB)
signature.asc (819.00 B)
Digital signature
Download all attachments

2014-10-15 16:57:07

by David Cohen

[permalink] [raw]
Subject: Re: [PATCH] usb: dwc3: be more verbose on ERRATIC_ERROR interrupt

On Wed, Oct 15, 2014 at 12:00:22AM -0500, Felipe Balbi wrote:
> Hi,
>
> On Tue, Oct 14, 2014 at 01:15:05PM -0700, David Cohen wrote:
> > ERRATIC_ERROR interrupt is an event that needs more attention from
> > developers than currently implemented, since this indicates a serious
> > stability issue. The only way to get warned about it is by selecting the
> > maximum driver's verbosity.
> >
> > This patch increases a bit the error's verbosity.
> >
> > Signed-off-by: David Cohen <[email protected]>
> > ---
> > drivers/usb/dwc3/gadget.c | 3 ++-
> > 1 file changed, 2 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/usb/dwc3/gadget.c b/drivers/usb/dwc3/gadget.c
> > index 3818b26bfc05..132e761d62e4 100644
> > --- a/drivers/usb/dwc3/gadget.c
> > +++ b/drivers/usb/dwc3/gadget.c
> > @@ -2484,7 +2484,8 @@ static void dwc3_gadget_interrupt(struct dwc3 *dwc,
> > dev_vdbg(dwc->dev, "Start of Periodic Frame\n");
> > break;
> > case DWC3_DEVICE_EVENT_ERRATIC_ERROR:
> > - dev_vdbg(dwc->dev, "Erratic Error\n");
> > + WARN_ON_ONCE(1);
> > + dev_dbg(dwc->dev, "Erratic Error\n");
>
> how about:
> WARN_ONCE(true, "Erratic Error\n");
>
> instead ?

When erratic error event happens, in my experience it is usually
followed but many of them in a row. It may end up too verbose.

Br, David

>
> > break;
> > case DWC3_DEVICE_EVENT_CMD_CMPL:
> > dev_vdbg(dwc->dev, "Command Complete\n");
> > --
> > 2.1.0
> >
>
> --
> balbi

2014-10-15 17:01:57

by David Cohen

[permalink] [raw]
Subject: Re: [PATCH] usb: dwc3: be more verbose on ERRATIC_ERROR interrupt

On Wed, Oct 15, 2014 at 09:57:18AM -0700, David Cohen wrote:
> On Wed, Oct 15, 2014 at 12:00:22AM -0500, Felipe Balbi wrote:
> > Hi,
> >
> > On Tue, Oct 14, 2014 at 01:15:05PM -0700, David Cohen wrote:
> > > ERRATIC_ERROR interrupt is an event that needs more attention from
> > > developers than currently implemented, since this indicates a serious
> > > stability issue. The only way to get warned about it is by selecting the
> > > maximum driver's verbosity.
> > >
> > > This patch increases a bit the error's verbosity.
> > >
> > > Signed-off-by: David Cohen <[email protected]>
> > > ---
> > > drivers/usb/dwc3/gadget.c | 3 ++-
> > > 1 file changed, 2 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/drivers/usb/dwc3/gadget.c b/drivers/usb/dwc3/gadget.c
> > > index 3818b26bfc05..132e761d62e4 100644
> > > --- a/drivers/usb/dwc3/gadget.c
> > > +++ b/drivers/usb/dwc3/gadget.c
> > > @@ -2484,7 +2484,8 @@ static void dwc3_gadget_interrupt(struct dwc3 *dwc,
> > > dev_vdbg(dwc->dev, "Start of Periodic Frame\n");
> > > break;
> > > case DWC3_DEVICE_EVENT_ERRATIC_ERROR:
> > > - dev_vdbg(dwc->dev, "Erratic Error\n");
> > > + WARN_ON_ONCE(1);
> > > + dev_dbg(dwc->dev, "Erratic Error\n");
> >
> > how about:
> > WARN_ONCE(true, "Erratic Error\n");
> >
> > instead ?
>
> When erratic error event happens, in my experience it is usually
> followed but many of them in a row. It may end up too verbose.

Oops, sorry for my confusion. I read WARN(), not WARN_ONCE() when I read
for the first time :)

I let the dev_dbg() because it may be useful to get all the occurrences
in case debug is enabled. But WARN_ONCE() is fine too if you prefer.

Br, David

>
> Br, David
>
> >
> > > break;
> > > case DWC3_DEVICE_EVENT_CMD_CMPL:
> > > dev_vdbg(dwc->dev, "Command Complete\n");
> > > --
> > > 2.1.0
> > >
> >
> > --
> > balbi
>
>

2014-10-15 17:06:23

by Felipe Balbi

[permalink] [raw]
Subject: Re: [PATCH] usb: dwc3: be more verbose on ERRATIC_ERROR interrupt

On Wed, Oct 15, 2014 at 10:01:45AM -0700, David Cohen wrote:
> On Wed, Oct 15, 2014 at 09:57:18AM -0700, David Cohen wrote:
> > On Wed, Oct 15, 2014 at 12:00:22AM -0500, Felipe Balbi wrote:
> > > Hi,
> > >
> > > On Tue, Oct 14, 2014 at 01:15:05PM -0700, David Cohen wrote:
> > > > ERRATIC_ERROR interrupt is an event that needs more attention from
> > > > developers than currently implemented, since this indicates a serious
> > > > stability issue. The only way to get warned about it is by selecting the
> > > > maximum driver's verbosity.
> > > >
> > > > This patch increases a bit the error's verbosity.
> > > >
> > > > Signed-off-by: David Cohen <[email protected]>
> > > > ---
> > > > drivers/usb/dwc3/gadget.c | 3 ++-
> > > > 1 file changed, 2 insertions(+), 1 deletion(-)
> > > >
> > > > diff --git a/drivers/usb/dwc3/gadget.c b/drivers/usb/dwc3/gadget.c
> > > > index 3818b26bfc05..132e761d62e4 100644
> > > > --- a/drivers/usb/dwc3/gadget.c
> > > > +++ b/drivers/usb/dwc3/gadget.c
> > > > @@ -2484,7 +2484,8 @@ static void dwc3_gadget_interrupt(struct dwc3 *dwc,
> > > > dev_vdbg(dwc->dev, "Start of Periodic Frame\n");
> > > > break;
> > > > case DWC3_DEVICE_EVENT_ERRATIC_ERROR:
> > > > - dev_vdbg(dwc->dev, "Erratic Error\n");
> > > > + WARN_ON_ONCE(1);
> > > > + dev_dbg(dwc->dev, "Erratic Error\n");
> > >
> > > how about:
> > > WARN_ONCE(true, "Erratic Error\n");
> > >
> > > instead ?
> >
> > When erratic error event happens, in my experience it is usually
> > followed but many of them in a row. It may end up too verbose.
>
> Oops, sorry for my confusion. I read WARN(), not WARN_ONCE() when I read
> for the first time :)
>
> I let the dev_dbg() because it may be useful to get all the occurrences
> in case debug is enabled. But WARN_ONCE() is fine too if you prefer.

We might WARN_ONCE() and forcibly disconnect from host as the platform
worn't work anymore after Erratic Error happens. Then just make that
clear with "Erratic Error: disconnecting from host\n" or something like
that.

The only way to fix erratic error (by fixing code - probably PHY driver)
will require rebooting the platform anyway. At a minimum, you'd have to
unload and reload dwc3.ko, just make sure (if you can reproduce erratic
error easily) modprobe -r dwc3 && modprobe dwc3 still causes dwc3 to
reconnect to host.

cheers

--
balbi


Attachments:
(No filename) (2.32 kB)
signature.asc (819.00 B)
Digital signature
Download all attachments

2014-10-15 18:36:22

by David Cohen

[permalink] [raw]
Subject: Re: [PATCH] usb: dwc3: be more verbose on ERRATIC_ERROR interrupt

On Wed, Oct 15, 2014 at 12:06:10PM -0500, Felipe Balbi wrote:
> On Wed, Oct 15, 2014 at 10:01:45AM -0700, David Cohen wrote:
> > On Wed, Oct 15, 2014 at 09:57:18AM -0700, David Cohen wrote:
> > > On Wed, Oct 15, 2014 at 12:00:22AM -0500, Felipe Balbi wrote:
> > > > Hi,
> > > >
> > > > On Tue, Oct 14, 2014 at 01:15:05PM -0700, David Cohen wrote:
> > > > > ERRATIC_ERROR interrupt is an event that needs more attention from
> > > > > developers than currently implemented, since this indicates a serious
> > > > > stability issue. The only way to get warned about it is by selecting the
> > > > > maximum driver's verbosity.
> > > > >
> > > > > This patch increases a bit the error's verbosity.
> > > > >
> > > > > Signed-off-by: David Cohen <[email protected]>
> > > > > ---
> > > > > drivers/usb/dwc3/gadget.c | 3 ++-
> > > > > 1 file changed, 2 insertions(+), 1 deletion(-)
> > > > >
> > > > > diff --git a/drivers/usb/dwc3/gadget.c b/drivers/usb/dwc3/gadget.c
> > > > > index 3818b26bfc05..132e761d62e4 100644
> > > > > --- a/drivers/usb/dwc3/gadget.c
> > > > > +++ b/drivers/usb/dwc3/gadget.c
> > > > > @@ -2484,7 +2484,8 @@ static void dwc3_gadget_interrupt(struct dwc3 *dwc,
> > > > > dev_vdbg(dwc->dev, "Start of Periodic Frame\n");
> > > > > break;
> > > > > case DWC3_DEVICE_EVENT_ERRATIC_ERROR:
> > > > > - dev_vdbg(dwc->dev, "Erratic Error\n");
> > > > > + WARN_ON_ONCE(1);
> > > > > + dev_dbg(dwc->dev, "Erratic Error\n");
> > > >
> > > > how about:
> > > > WARN_ONCE(true, "Erratic Error\n");
> > > >
> > > > instead ?
> > >
> > > When erratic error event happens, in my experience it is usually
> > > followed but many of them in a row. It may end up too verbose.
> >
> > Oops, sorry for my confusion. I read WARN(), not WARN_ONCE() when I read
> > for the first time :)
> >
> > I let the dev_dbg() because it may be useful to get all the occurrences
> > in case debug is enabled. But WARN_ONCE() is fine too if you prefer.
>
> We might WARN_ONCE() and forcibly disconnect from host as the platform
> worn't work anymore after Erratic Error happens. Then just make that
> clear with "Erratic Error: disconnecting from host\n" or something like
> that.

That works.
I assume calling dwc3_gadget_disconnect_interrupt() from
DWC3_DEVICE_EVENT_ERRATIC_ERROR should be enough to disconnect the
gadget and propagate the status to userspace.

>
> The only way to fix erratic error (by fixing code - probably PHY driver)
> will require rebooting the platform anyway. At a minimum, you'd have to
> unload and reload dwc3.ko, just make sure (if you can reproduce erratic
> error easily) modprobe -r dwc3 && modprobe dwc3 still causes dwc3 to
> reconnect to host.

It does. Forcing soft reset on whole IP is how databook describes to
solve it.

Br, David

>
> cheers
>
> --
> balbi

2014-10-15 18:55:52

by Felipe Balbi

[permalink] [raw]
Subject: Re: [PATCH] usb: dwc3: be more verbose on ERRATIC_ERROR interrupt

On Wed, Oct 15, 2014 at 11:36:33AM -0700, David Cohen wrote:
> On Wed, Oct 15, 2014 at 12:06:10PM -0500, Felipe Balbi wrote:
> > On Wed, Oct 15, 2014 at 10:01:45AM -0700, David Cohen wrote:
> > > On Wed, Oct 15, 2014 at 09:57:18AM -0700, David Cohen wrote:
> > > > On Wed, Oct 15, 2014 at 12:00:22AM -0500, Felipe Balbi wrote:
> > > > > Hi,
> > > > >
> > > > > On Tue, Oct 14, 2014 at 01:15:05PM -0700, David Cohen wrote:
> > > > > > ERRATIC_ERROR interrupt is an event that needs more attention from
> > > > > > developers than currently implemented, since this indicates a serious
> > > > > > stability issue. The only way to get warned about it is by selecting the
> > > > > > maximum driver's verbosity.
> > > > > >
> > > > > > This patch increases a bit the error's verbosity.
> > > > > >
> > > > > > Signed-off-by: David Cohen <[email protected]>
> > > > > > ---
> > > > > > drivers/usb/dwc3/gadget.c | 3 ++-
> > > > > > 1 file changed, 2 insertions(+), 1 deletion(-)
> > > > > >
> > > > > > diff --git a/drivers/usb/dwc3/gadget.c b/drivers/usb/dwc3/gadget.c
> > > > > > index 3818b26bfc05..132e761d62e4 100644
> > > > > > --- a/drivers/usb/dwc3/gadget.c
> > > > > > +++ b/drivers/usb/dwc3/gadget.c
> > > > > > @@ -2484,7 +2484,8 @@ static void dwc3_gadget_interrupt(struct dwc3 *dwc,
> > > > > > dev_vdbg(dwc->dev, "Start of Periodic Frame\n");
> > > > > > break;
> > > > > > case DWC3_DEVICE_EVENT_ERRATIC_ERROR:
> > > > > > - dev_vdbg(dwc->dev, "Erratic Error\n");
> > > > > > + WARN_ON_ONCE(1);
> > > > > > + dev_dbg(dwc->dev, "Erratic Error\n");
> > > > >
> > > > > how about:
> > > > > WARN_ONCE(true, "Erratic Error\n");
> > > > >
> > > > > instead ?
> > > >
> > > > When erratic error event happens, in my experience it is usually
> > > > followed but many of them in a row. It may end up too verbose.
> > >
> > > Oops, sorry for my confusion. I read WARN(), not WARN_ONCE() when I read
> > > for the first time :)
> > >
> > > I let the dev_dbg() because it may be useful to get all the occurrences
> > > in case debug is enabled. But WARN_ONCE() is fine too if you prefer.
> >
> > We might WARN_ONCE() and forcibly disconnect from host as the platform
> > worn't work anymore after Erratic Error happens. Then just make that
> > clear with "Erratic Error: disconnecting from host\n" or something like
> > that.
>
> That works.
> I assume calling dwc3_gadget_disconnect_interrupt() from
> DWC3_DEVICE_EVENT_ERRATIC_ERROR should be enough to disconnect the
> gadget and propagate the status to userspace.
>
> >
> > The only way to fix erratic error (by fixing code - probably PHY driver)
> > will require rebooting the platform anyway. At a minimum, you'd have to
> > unload and reload dwc3.ko, just make sure (if you can reproduce erratic
> > error easily) modprobe -r dwc3 && modprobe dwc3 still causes dwc3 to
> > reconnect to host.
>
> It does. Forcing soft reset on whole IP is how databook describes to
> solve it.

alright, then please send a patch and I'll take for v3.19

--
balbi


Attachments:
(No filename) (2.97 kB)
signature.asc (819.00 B)
Digital signature
Download all attachments