[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>:
>> Subject: Re: Re: [RFC] [PATCH v3] IB/ipoib: Add bonding support to?IPoIB
>>
>> Michael S. Tsirkin wrote:
>>>> With the current solution it is still unsafe to unload ib_ipoib before bonding but
>>>> I tend to agree with Michael and my opinion now is that this should be fixed in the
>>>> upper layer (i.e. bonding).
>>> This looks simple. Will we get to see the patch to core bonding code as well
>>> soon? I guess bonding support in OFED will need to be patched somehow?
>>>
>> Sure. I am working on it now (making ib_ipoib safer to remove before bonding
>> does) and I would be happy to share when I'm
>> done.  However, I think that this is a separate issue and should not prevent
>> from this patch to get in.  Don't you agree?
> 
> There's no rush I guess - Roland's on vacation so we have time to
> review how everything works together.
> 
>> BTW: It is often claimed to module unload is unsafe by definition and is not 
>> a production issue. Doesn't this claim make the unload issue "less critical"?
> 
> I know our users depend on module unload being stable.
> Module unload remains the simplest way to test hotplug,
> so bugs there might hide real issues.
> 
ib_ipoib can still be unloaded even when bonding is used (it is just that bonding
should be removed first).
About hotplug:  it is still working well and correct me if I'm wrong but my way to test 
hotplug is to unload ib_mthca which is not affected by the presence of bonding.

As I said, I would be happy to share the work on bonding and get reviews about it but 
I still think that this patch alone can be a first step.
> 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?
If I can unload a module that is being referenced from the outside then
I am not protected by the kernel from doing something wrong.