If device_register() fails, put_device() should be called to drop the device reference. Thus add put_device() after subsys_unregister label and change device_unregister() to device_del() in fail label.
Found by code review.
Fixes: 6c9d29095264 ("power: cmm - convert sysdev_class to a regular subsystem") Cc: stable@vger.kernel.org Signed-off-by: Haoxiang Li lihaoxiang@isrc.iscas.ac.cn --- arch/powerpc/platforms/pseries/cmm.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/arch/powerpc/platforms/pseries/cmm.c b/arch/powerpc/platforms/pseries/cmm.c index 4cbbe2ee58ab..0666d3300bdb 100644 --- a/arch/powerpc/platforms/pseries/cmm.c +++ b/arch/powerpc/platforms/pseries/cmm.c @@ -419,8 +419,9 @@ static int cmm_sysfs_register(struct device *dev) fail: while (--i >= 0) device_remove_file(dev, cmm_attrs[i]); - device_unregister(dev); + device_del(dev); subsys_unregister: + put_device(dev); bus_unregister(&cmm_subsys); return rc; }
On Mon, Dec 22, 2025 at 11:12:25AM +0800, Haoxiang Li wrote:
If device_register() fails, put_device() should be called to drop the device reference. Thus add put_device() after subsys_unregister label and change device_unregister() to device_del() in fail label.
Found by code review.
Fixes: 6c9d29095264 ("power: cmm - convert sysdev_class to a regular subsystem") Cc: stable@vger.kernel.org Signed-off-by: Haoxiang Li lihaoxiang@isrc.iscas.ac.cn
arch/powerpc/platforms/pseries/cmm.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/arch/powerpc/platforms/pseries/cmm.c b/arch/powerpc/platforms/pseries/cmm.c index 4cbbe2ee58ab..0666d3300bdb 100644 --- a/arch/powerpc/platforms/pseries/cmm.c +++ b/arch/powerpc/platforms/pseries/cmm.c @@ -419,8 +419,9 @@ static int cmm_sysfs_register(struct device *dev) fail: while (--i >= 0) device_remove_file(dev, cmm_attrs[i]);
- device_unregister(dev);
- device_del(dev);
subsys_unregister:
- put_device(dev); bus_unregister(&cmm_subsys); return rc;
}
this does not look to be correct, how was it tested?
Also, why not fix all of this up properly by calling the correct functions so that you don't have to manually add/remove the sysfs files? That would resolve the "problem" you seem to think is here in a much simpler way, and also fix the real bug in this function (i.e. it races with userspace and looses.)
thanks,
greg k-h
linux-stable-mirror@lists.linaro.org