Re: kobjects: mark cleaned up kobjects as unitialized

[eran liberty]

Hi Greg,

On Wed, Nov 25, 2009 at 2:35 PM, Greg KH <gregkh@xxxxxxx> wrote:
> On Wed, Nov 25, 2009 at 11:27:58AM +0200, eran liberty wrote:
>> Hi Greg & Balaji,
>>
>> After diving into the LDKM and failed to spot the point where you
>> actually un-initialize the 'state_initialized' of a kobject... and
>> since I have statically allocated object which trip over this very
>> same trap...
>
> Ah, there's your problem, don't statically allocate a kobject.  Fix that
> and your issue goes away, right?

right... but... I want to :).

Is there a Linux directive saying 'thus shall not statically allocate
(or reuse in any other way) kobjects"?

>
>> Google-ing for others who fell into this trap, I found your thread/patch at:
>> http://lkml.org/lkml/2008/3/8/155
>> and
>> http://lkml.indiana.edu/hypermail/linux/kernel/0902.0/01969.html
>>
>> I noticed this patch did not make it into the mainline.
>>
>> Is this patch still valid?

Why was your proposed patch not merged?
Is there some logic behind not having it?

>> Is there some other, better way to do it by the book?
>
> Do not statically allocate a kobject.
>
>> Right now I by-pass the problem by memset-ing the whole object after I
>> release it... but I feel this is a bit brutal.

Assuming I will keep it static and clear the status_initialize bit (by
memset-ing the whole object) after the kobject was released... am I
doing something wrong? should i expect other bad phenomena?

>
> You should be freeing your memory in your release function.
>
Should I free the object itself in release() function?
In OO-like thinking I would expect release() to be a cleanup function
for the device to be used, while the kfree() be done by the same
object which did the kmalloc()

> Do you have a pointer to your code somewhere?
>

I will pack something and send you (i dont think Mr. vger will
tolerate attachments :) )
but basically you can demonstrate the problem with this simple code:

struct device dev;
while(1) {
  dev->release = dummy_release_function_that_does_nothing;
  device_register(&dev); /* will fail on second iteration without the
memset!! ?? */
  device_unregister(&dev);
  memset(&dev,0,sizeof(struct device)); /* should be unnecessary  */
}

> thanks,
>
> greg k-h
>

thanks,
-- Liberty
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/