[PATCH] i2c: i2c-core: do not use bus internal data

[PATCH] i2c: i2c-core: do not use bus internal data

[Sudip Mukherjee]

The variable p is a data structure which is used by the driver core
internally and it is not expected that busses will be directly accessing
these driver core internal only data.

Signed-off-by: Sudip Mukherjee <sudip.mukherjee@codethink.co.uk>
---

Reference of Greg's comment about it at:
https://lkml.org/lkml/2016/3/5/171

 drivers/i2c/i2c-core.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/drivers/i2c/i2c-core.c b/drivers/i2c/i2c-core.c
index 2949ab3..2f31fb5 100644
--- a/drivers/i2c/i2c-core.c
+++ b/drivers/i2c/i2c-core.c
@@ -73,6 +73,7 @@ static struct device_type i2c_client_type;
 static int i2c_detect(struct i2c_adapter *adapter, struct i2c_driver *driver);
 
 static struct static_key i2c_trace_msg = STATIC_KEY_INIT_FALSE;
+static bool is_registered;
 
 void i2c_transfer_trace_reg(void)
 {
@@ -1529,7 +1530,7 @@ static int i2c_register_adapter(struct i2c_adapter *adap)
 	int res = 0;
 
 	/* Can't register until after driver model init */
-	if (unlikely(WARN_ON(!i2c_bus_type.p))) {
+	if (unlikely(WARN_ON(!is_registered))) {
 		res = -EAGAIN;
 		goto out_list;
 	}
@@ -1926,7 +1927,7 @@ int i2c_register_driver(struct module *owner, struct i2c_driver *driver)
 	int res;
 
 	/* Can't register until after driver model init */
-	if (unlikely(WARN_ON(!i2c_bus_type.p)))
+	if (unlikely(WARN_ON(!is_registered)))
 		return -EAGAIN;
 
 	/* add the driver to the list of i2c drivers in the driver core */
@@ -2118,6 +2119,7 @@ static int __init i2c_init(void)
 	if (IS_ENABLED(CONFIG_OF_DYNAMIC))
 		WARN_ON(of_reconfig_notifier_register(&i2c_of_notifier));
 
+	is_registered = true;
 	return 0;
 
 class_err:
-- 
1.9.1

Re: [PATCH] i2c: i2c-core: do not use bus internal data

[Greg KH]

On Mon, Mar 07, 2016 at 05:19:17PM +0530, Sudip Mukherjee wrote:
> The variable p is a data structure which is used by the driver core
> internally and it is not expected that busses will be directly accessing
> these driver core internal only data.
> 
> Signed-off-by: Sudip Mukherjee <sudip.mukherjee@codethink.co.uk>
> ---
> 
> Reference of Greg's comment about it at:
> https://lkml.org/lkml/2016/3/5/171
> 
>  drivers/i2c/i2c-core.c | 6 ++++--
>  1 file changed, 4 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/i2c/i2c-core.c b/drivers/i2c/i2c-core.c
> index 2949ab3..2f31fb5 100644
> --- a/drivers/i2c/i2c-core.c
> +++ b/drivers/i2c/i2c-core.c
> @@ -73,6 +73,7 @@ static struct device_type i2c_client_type;
>  static int i2c_detect(struct i2c_adapter *adapter, struct i2c_driver *driver);
>  
>  static struct static_key i2c_trace_msg = STATIC_KEY_INIT_FALSE;
> +static bool is_registered;
>  
>  void i2c_transfer_trace_reg(void)
>  {
> @@ -1529,7 +1530,7 @@ static int i2c_register_adapter(struct i2c_adapter *adap)
>  	int res = 0;
>  
>  	/* Can't register until after driver model init */
> -	if (unlikely(WARN_ON(!i2c_bus_type.p))) {
> +	if (unlikely(WARN_ON(!is_registered))) {
>  		res = -EAGAIN;
>  		goto out_list;
>  	}

Minor nit, likely/unlikely should only be used on very "hot paths" where
the difference if it is not included can be measured.  the "register a
device" path is not "hot" at all.

That being said, the patch is fine with me:

Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

Re: [PATCH] i2c: i2c-core: do not use bus internal data

[Sudip Mukherjee]

On Monday 07 March 2016 10:27 PM, Greg KH wrote:
> On Mon, Mar 07, 2016 at 05:19:17PM +0530, Sudip Mukherjee wrote:
>> The variable p is a data structure which is used by the driver core
>> internally and it is not expected that busses will be directly accessing
>> these driver core internal only data.
>>
>> Signed-off-by: Sudip Mukherjee <sudip.mukherjee@codethink.co.uk>
>> ---
>>
>> Reference of Greg's comment about it at:
>> https://lkml.org/lkml/2016/3/5/171
>>
>>   drivers/i2c/i2c-core.c | 6 ++++--
>>   1 file changed, 4 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/i2c/i2c-core.c b/drivers/i2c/i2c-core.c
>> index 2949ab3..2f31fb5 100644
>> --- a/drivers/i2c/i2c-core.c
>> +++ b/drivers/i2c/i2c-core.c
>> @@ -73,6 +73,7 @@ static struct device_type i2c_client_type;
>>   static int i2c_detect(struct i2c_adapter *adapter, struct i2c_driver *driver);
>>
>>   static struct static_key i2c_trace_msg = STATIC_KEY_INIT_FALSE;
>> +static bool is_registered;
>>
>>   void i2c_transfer_trace_reg(void)
>>   {
>> @@ -1529,7 +1530,7 @@ static int i2c_register_adapter(struct i2c_adapter *adap)
>>   	int res = 0;
>>
>>   	/* Can't register until after driver model init */
>> -	if (unlikely(WARN_ON(!i2c_bus_type.p))) {
>> +	if (unlikely(WARN_ON(!is_registered))) {
>>   		res = -EAGAIN;
>>   		goto out_list;
>>   	}
>
> Minor nit, likely/unlikely should only be used on very "hot paths" where
> the difference if it is not included can be measured.  the "register a
> device" path is not "hot" at all.

I can submit a patch afterwards to remove the "unlikely" if that is ok 
with Wolfram.

regards
sudip

Re: [PATCH] i2c: i2c-core: do not use bus internal data

[Wolfram Sang]

[-- Attachment #1: Type: text/plain, Size: 387 bytes --]

On Mon, Mar 07, 2016 at 05:19:17PM +0530, Sudip Mukherjee wrote:
> The variable p is a data structure which is used by the driver core
> internally and it is not expected that busses will be directly accessing
> these driver core internal only data.
> 
> Signed-off-by: Sudip Mukherjee <sudip.mukherjee@codethink.co.uk>

Removed the unlikely() and applied to for-next, thanks!


[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH] i2c: i2c-core: do not use bus internal data

[Thierry Reding]

[-- Attachment #1: Type: text/plain, Size: 2228 bytes --]

On Mon, Mar 07, 2016 at 05:19:17PM +0530, Sudip Mukherjee wrote:
> The variable p is a data structure which is used by the driver core
> internally and it is not expected that busses will be directly accessing
> these driver core internal only data.
> 
> Signed-off-by: Sudip Mukherjee <sudip.mukherjee@codethink.co.uk>
> ---
> 
> Reference of Greg's comment about it at:
> https://lkml.org/lkml/2016/3/5/171
> 
>  drivers/i2c/i2c-core.c | 6 ++++--
>  1 file changed, 4 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/i2c/i2c-core.c b/drivers/i2c/i2c-core.c
> index 2949ab3..2f31fb5 100644
> --- a/drivers/i2c/i2c-core.c
> +++ b/drivers/i2c/i2c-core.c
> @@ -73,6 +73,7 @@ static struct device_type i2c_client_type;
>  static int i2c_detect(struct i2c_adapter *adapter, struct i2c_driver *driver);
>  
>  static struct static_key i2c_trace_msg = STATIC_KEY_INIT_FALSE;
> +static bool is_registered;
>  
>  void i2c_transfer_trace_reg(void)
>  {
> @@ -1529,7 +1530,7 @@ static int i2c_register_adapter(struct i2c_adapter *adap)
>  	int res = 0;
>  
>  	/* Can't register until after driver model init */
> -	if (unlikely(WARN_ON(!i2c_bus_type.p))) {
> +	if (unlikely(WARN_ON(!is_registered))) {
>  		res = -EAGAIN;
>  		goto out_list;
>  	}
> @@ -1926,7 +1927,7 @@ int i2c_register_driver(struct module *owner, struct i2c_driver *driver)
>  	int res;
>  
>  	/* Can't register until after driver model init */
> -	if (unlikely(WARN_ON(!i2c_bus_type.p)))
> +	if (unlikely(WARN_ON(!is_registered)))
>  		return -EAGAIN;
>  
>  	/* add the driver to the list of i2c drivers in the driver core */
> @@ -2118,6 +2119,7 @@ static int __init i2c_init(void)
>  	if (IS_ENABLED(CONFIG_OF_DYNAMIC))
>  		WARN_ON(of_reconfig_notifier_register(&i2c_of_notifier));
>  
> +	is_registered = true;
>  	return 0;
>  
>  class_err:

This doesn't work. I see a number of these WARN_ON()s trigger and I
think the reason is that i2c_init() always fails now. The cause seems to
be that i2c_init() calls i2c_add_driver(&dummy_driver), which will now
always fail, because is_register is set to true *after* that call. There
is no way I see I2C working at all after this patch.

Thierry

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH] i2c: i2c-core: do not use bus internal data

[Thierry Reding]

[-- Attachment #1: Type: text/plain, Size: 3061 bytes --]

On Mon, Mar 14, 2016 at 10:18:19AM +0100, Thierry Reding wrote:
> On Mon, Mar 07, 2016 at 05:19:17PM +0530, Sudip Mukherjee wrote:
> > The variable p is a data structure which is used by the driver core
> > internally and it is not expected that busses will be directly accessing
> > these driver core internal only data.
> > 
> > Signed-off-by: Sudip Mukherjee <sudip.mukherjee@codethink.co.uk>
> > ---
> > 
> > Reference of Greg's comment about it at:
> > https://lkml.org/lkml/2016/3/5/171
> > 
> >  drivers/i2c/i2c-core.c | 6 ++++--
> >  1 file changed, 4 insertions(+), 2 deletions(-)
> > 
> > diff --git a/drivers/i2c/i2c-core.c b/drivers/i2c/i2c-core.c
> > index 2949ab3..2f31fb5 100644
> > --- a/drivers/i2c/i2c-core.c
> > +++ b/drivers/i2c/i2c-core.c
> > @@ -73,6 +73,7 @@ static struct device_type i2c_client_type;
> >  static int i2c_detect(struct i2c_adapter *adapter, struct i2c_driver *driver);
> >  
> >  static struct static_key i2c_trace_msg = STATIC_KEY_INIT_FALSE;
> > +static bool is_registered;
> >  
> >  void i2c_transfer_trace_reg(void)
> >  {
> > @@ -1529,7 +1530,7 @@ static int i2c_register_adapter(struct i2c_adapter *adap)
> >  	int res = 0;
> >  
> >  	/* Can't register until after driver model init */
> > -	if (unlikely(WARN_ON(!i2c_bus_type.p))) {
> > +	if (unlikely(WARN_ON(!is_registered))) {
> >  		res = -EAGAIN;
> >  		goto out_list;
> >  	}
> > @@ -1926,7 +1927,7 @@ int i2c_register_driver(struct module *owner, struct i2c_driver *driver)
> >  	int res;
> >  
> >  	/* Can't register until after driver model init */
> > -	if (unlikely(WARN_ON(!i2c_bus_type.p)))
> > +	if (unlikely(WARN_ON(!is_registered)))
> >  		return -EAGAIN;
> >  
> >  	/* add the driver to the list of i2c drivers in the driver core */
> > @@ -2118,6 +2119,7 @@ static int __init i2c_init(void)
> >  	if (IS_ENABLED(CONFIG_OF_DYNAMIC))
> >  		WARN_ON(of_reconfig_notifier_register(&i2c_of_notifier));
> >  
> > +	is_registered = true;
> >  	return 0;
> >  
> >  class_err:
> 
> This doesn't work. I see a number of these WARN_ON()s trigger and I
> think the reason is that i2c_init() always fails now. The cause seems to
> be that i2c_init() calls i2c_add_driver(&dummy_driver), which will now
> always fail, because is_register is set to true *after* that call. There
> is no way I see I2C working at all after this patch.

FWIW, the below on top of your patch seems to fix things for me.

Thierry

--- >8 ---
diff --git a/drivers/i2c/i2c-core.c b/drivers/i2c/i2c-core.c
index f4726cdbb06a..f8723a474e28 100644
--- a/drivers/i2c/i2c-core.c
+++ b/drivers/i2c/i2c-core.c
@@ -2112,6 +2112,8 @@ static int __init i2c_init(void)
 		goto bus_err;
 	}
 #endif
+	is_registered = true;
+
 	retval = i2c_add_driver(&dummy_driver);
 	if (retval)
 		goto class_err;
@@ -2119,7 +2121,6 @@ static int __init i2c_init(void)
 	if (IS_ENABLED(CONFIG_OF_DYNAMIC))
 		WARN_ON(of_reconfig_notifier_register(&i2c_of_notifier));
 
-	is_registered = true;
 	return 0;
 
 class_err:

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH] i2c: i2c-core: do not use bus internal data

[Wolfram Sang]

[-- Attachment #1: Type: text/plain, Size: 483 bytes --]


> This doesn't work. I see a number of these WARN_ON()s trigger and I
> think the reason is that i2c_init() always fails now. The cause seems to
> be that i2c_init() calls i2c_add_driver(&dummy_driver), which will now
> always fail, because is_register is set to true *after* that call. There
> is no way I see I2C working at all after this patch.

Same conclusion here. Too much trust applied to the original patch, mea
culpa and sorry! Will send the fixup (tested!) in a minute.


[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH] i2c: i2c-core: do not use bus internal data

[Thierry Reding]

[-- Attachment #1: Type: text/plain, Size: 687 bytes --]

On Mon, Mar 14, 2016 at 10:28:10AM +0100, Wolfram Sang wrote:
> 
> > This doesn't work. I see a number of these WARN_ON()s trigger and I
> > think the reason is that i2c_init() always fails now. The cause seems to
> > be that i2c_init() calls i2c_add_driver(&dummy_driver), which will now
> > always fail, because is_register is set to true *after* that call. There
> > is no way I see I2C working at all after this patch.
> 
> Same conclusion here. Too much trust applied to the original patch, mea
> culpa and sorry! Will send the fixup (tested!) in a minute.

No worries. There's nothing like good old runtime testing for making
sure things really do work. =)

Thierry

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH] i2c: i2c-core: do not use bus internal data

[Sudeep Holla]

On Mon, Mar 14, 2016 at 9:27 AM, Thierry Reding
<thierry.reding@gmail.com> wrote:
> On Mon, Mar 14, 2016 at 10:18:19AM +0100, Thierry Reding wrote:
>> On Mon, Mar 07, 2016 at 05:19:17PM +0530, Sudip Mukherjee wrote:
>> > The variable p is a data structure which is used by the driver core
>> > internally and it is not expected that busses will be directly accessing
>> > these driver core internal only data.
>> >
>> > Signed-off-by: Sudip Mukherjee <sudip.mukherjee@codethink.co.uk>
>> > ---
>> >
>> > Reference of Greg's comment about it at:
>> > https://lkml.org/lkml/2016/3/5/171
>> >
>> >  drivers/i2c/i2c-core.c | 6 ++++--
>> >  1 file changed, 4 insertions(+), 2 deletions(-)
>> >
>> > diff --git a/drivers/i2c/i2c-core.c b/drivers/i2c/i2c-core.c
>> > index 2949ab3..2f31fb5 100644
>> > --- a/drivers/i2c/i2c-core.c
>> > +++ b/drivers/i2c/i2c-core.c
>> > @@ -73,6 +73,7 @@ static struct device_type i2c_client_type;
>> >  static int i2c_detect(struct i2c_adapter *adapter, struct i2c_driver *driver);
>> >
>> >  static struct static_key i2c_trace_msg = STATIC_KEY_INIT_FALSE;
>> > +static bool is_registered;
>> >
>> >  void i2c_transfer_trace_reg(void)
>> >  {
>> > @@ -1529,7 +1530,7 @@ static int i2c_register_adapter(struct i2c_adapter *adap)
>> >     int res = 0;
>> >
>> >     /* Can't register until after driver model init */
>> > -   if (unlikely(WARN_ON(!i2c_bus_type.p))) {
>> > +   if (unlikely(WARN_ON(!is_registered))) {
>> >             res = -EAGAIN;
>> >             goto out_list;
>> >     }
>> > @@ -1926,7 +1927,7 @@ int i2c_register_driver(struct module *owner, struct i2c_driver *driver)
>> >     int res;
>> >
>> >     /* Can't register until after driver model init */
>> > -   if (unlikely(WARN_ON(!i2c_bus_type.p)))
>> > +   if (unlikely(WARN_ON(!is_registered)))
>> >             return -EAGAIN;
>> >
>> >     /* add the driver to the list of i2c drivers in the driver core */
>> > @@ -2118,6 +2119,7 @@ static int __init i2c_init(void)
>> >     if (IS_ENABLED(CONFIG_OF_DYNAMIC))
>> >             WARN_ON(of_reconfig_notifier_register(&i2c_of_notifier));
>> >
>> > +   is_registered = true;
>> >     return 0;
>> >
>> >  class_err:
>>
>> This doesn't work. I see a number of these WARN_ON()s trigger and I
>> think the reason is that i2c_init() always fails now. The cause seems to
>> be that i2c_init() calls i2c_add_driver(&dummy_driver), which will now
>> always fail, because is_register is set to true *after* that call. There
>> is no way I see I2C working at all after this patch.

Exactly, and this is resulting in recursive failures on dev platform ending up
in boot failure

>
> FWIW, the below on top of your patch seems to fix things for me.
>

I too came up with same patch, good that I searched before posting it out.
FWIW, it fixes the recursive fault at boot on my arm64 platform.

Regards,
Sudeep

Re: [PATCH] i2c: i2c-core: do not use bus internal data

[Sudip Mukherjee]

On Mon, Mar 14, 2016 at 10:28:10AM +0100, Wolfram Sang wrote:
> 
> > This doesn't work. I see a number of these WARN_ON()s trigger and I
> > think the reason is that i2c_init() always fails now. The cause seems to
> > be that i2c_init() calls i2c_add_driver(&dummy_driver), which will now
> > always fail, because is_register is set to true *after* that call. There
> > is no way I see I2C working at all after this patch.
> 
> Same conclusion here. Too much trust applied to the original patch, mea
> culpa and sorry! Will send the fixup (tested!) in a minute.

Sorry for the mess. Fengguang Wu did send me a warning, but since I was
travelling i could not do much with it.
Sorry again.

regards
sudip