Re: [PATCH 01/10] i2c: add and export of_get_i2c_adapter_by_node() interface

On Wed, Jul 08, 2015 at 04:31:37PM +0300, Vladimir Zapolskiy wrote:

Hi Thierry,

On 08.07.2015 16:11, Thierry Reding wrote:

...

On Wed, Jul 08, 2015 at 03:59:12PM +0300, Vladimir Zapolskiy wrote:

...

of_find_i2c_adapter_by_node() call requires quite often missing put_device(), and i2c_put_adapter() releases a device locked by i2c_get_adapter() only. In general module_put(adapter->owner) and put_device(dev) are not interchangeable.

This is a common error reproduction scenario as a result of the misusage described above (for clearness this is run on iMX6 platform with HDMI and I2C bus drivers compiled as kernel modules):

root@mx6q:~# lsmod | grep i2c
i2c_imx                10213  0
root@mx6q:~# lsmod | grep dw_hdmi_imx
dw_hdmi_imx             3631  0
dw_hdmi                11846  1 dw_hdmi_imx
imxdrm                  8674  3 dw_hdmi_imx,imx_ipuv3_crtc,imx_ldb
drm_kms_helper        113765  5 dw_hdmi,imxdrm,imx_ipuv3_crtc,imx_ldb
root@mx6q:~# rmmod dw_hdmi_imx
root@mx6q:~# lsmod | grep i2c
i2c_imx                10213  -1

                             ^^^^^

root@mx6q:~# rmmod i2c_imx
rmmod: ERROR: Module i2c_imx is in use

To fix existing users of these interfaces and to avoid any further confusion and misusage in future, add one more interface of_get_i2c_adapter_by_node(), it is similar to i2c_get_adapter() in sense that an I2C bus device driver found and locked by user can be correctly unlocked by i2c_put_adapter().

Signed-off-by: Vladimir Zapolskiy vladimir_zapolskiy@mentor.com

The change is based on RFC http://www.spinics.net/lists/linux-i2c/msg20257.html

• added new exported function declaration in include/linux/i2c.h
• added put_device(dev) call right inside of_get_i2c_adapter_by_node()
• corrected authorship of the change

drivers/i2c/i2c-core.c | 20 ++++++++++++++++++++ include/linux/i2c.h | 6 ++++++ 2 files changed, 26 insertions(+)

diff --git a/drivers/i2c/i2c-core.c b/drivers/i2c/i2c-core.c index 069a41f..0d902ab 100644 --- a/drivers/i2c/i2c-core.c +++ b/drivers/i2c/i2c-core.c @@ -1356,6 +1356,26 @@ struct i2c_adapter *of_find_i2c_adapter_by_node(struct device_node *node) return i2c_verify_adapter(dev); } EXPORT_SYMBOL(of_find_i2c_adapter_by_node);

+struct i2c_adapter *of_get_i2c_adapter_by_node(struct device_node *node) +{

• struct device *dev;
• struct i2c_adapter *adapter;
• dev = bus_find_device(&i2c_bus_type, NULL, node,

• 	      of_dev_node_match);
  

• if (!dev)

• return NULL;
  

• adapter = i2c_verify_adapter(dev);
• if (adapter && !try_module_get(adapter->owner))

• adapter = NULL;
  

• put_device(dev);

I don't think this is correct. Users still need to keep a reference to the device, otherwise it can simply disappear even if the module stays around (think sysfs bind/unbind attributes).

Looking at i2c_put_adapter() it seems like it would need to do more than just drop the module reference. Then again, that probably means that we need to add a get_device() somewhere in i2c_get_adapter() to balance the put_device() in i2c_put_adapter().

it makes sense for me, thanks for momentary review.

I'm hesitating to add put_device(dev) to i2c_put_adapter() etc. in this series though. After development and testing I would like to send another preceding independent change updating i2c_get_adapter(), i2c_put_adapter() and clients (or if you wish you can do it), then I'll rebase 01/10 on top of it, the rest most probably is unchanged.

I think that would make sense, yes.

Thierry