On 11/11/2010 04:16 PM, Greg KH wrote:
> On Thu, Nov 11, 2010 at 04:05:43PM +0200, Boaz Harrosh wrote:
>> On 11/11/2010 02:04 PM, Greg KH wrote:
>>> On Thu, Nov 11, 2010 at 11:59:28AM +0200, Boaz Harrosh wrote:
>>>> 4. Exactly Like 3 but without the extra kref member
>>>> Only x_put() changes and x_kref_release() now receives
>>>> an x_object
>>>>
>>>> int x_put(struct object_x *x)
>>>> {
>>>> if (kobject_put(&x->kobj) == 1)
>>>> // Like above [3] x_kref_release()
>>>> x_kref_release(x);
>>>> }
>>>
>>> This is racy, please never do this.
>>>
>>
>> The last ref belongs to the core code. 1 means there are no
>> more external clients on the object. So it can not race with
>> decrements. But I guess there is a possibility that it can
>> race with new increments.
>
> Exactly.
>
>> If it is the case that new increments
>> can only come from, say, sysfs access, then if we call the
>> x_put() == 1 after we are unregistered from sysfs and no new
>> users are allowed then the counter can only go down and we
>> have the last reference. No?
>
> Just don't do this, it's not worth it and will break over time when
> others mess with the code.
>
> Also note that kobject_put() does not even return a value, so the code
> above will not even compile, let alone work.
>
OK Point taken, it is fragile. So there is option [3] then, with the extra
kref. I think I've seen other places with this approach.
> thanks,
>
> greg k-h
Thanks
Boaz
--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
