[ofa-general] Re: [RFC] [PATCH v3] IB/ipoib: Add bonding support to IPoIB

[Moni Shoua]

Michael S. Tsirkin wrote:
>> Quoting Moni Shoua <monisonlists at gmail.com>:
>>
>>> Where does the claim that module unload is unsafe by definition
>>> come from? Weren't the races solved in 2.6 with the new in-kernel
>>> loader?
>> Well, what happens with bonding and IPoIB speaks for itself, doesn't it?
> 
> Not to my eyes.
> 
>> If I can unload a module that is being referenced from the outside
> 
> This just means there's a bug in the specific module(s).
> In this case it is either ipoib or bonding module (or both).
> 
>> then I am not protected by the kernel from doing something wrong.
> 
> Yes, this is really fundamental in kernel programming.
> That's what makes it interesting.
> 
I'm not trying to avoid the challenge/fun of kernel programming  but only to fix one bug
without introducing the other fixes at the same time.
My opinion is that IPoIB shouldn't assume that n->dev is an IPoIB device
because event if we fix the dependency bug between ib_ipoib and bonding this assumption
is still wrong.

One more thing...
We can tell the customer that unloading modules is allowed but that they have to 
do it in the right order and anyway,