[openib-general] Re: [PATCH fixed] ipoib: device removal races

[Michael S. Tsirkin]

Quoting r. Roland Dreier <rolandd at cisco.com>:
> Subject: Re: [PATCH fixed] ipoib: device removal races
> 
>     Michael> Here's fix for this theoretical race (I didnt see it
>     Michael> triggered in real life).  This needs to be applied in
>     Michael> addition to my previous patch, which fixes a crash I
>     Michael> actually see in the lab.
> 
> Thanks.  For this patch, why do we need to use a different workqueue
> rather than sharing the existing IPoIB workqueue?

Because we down/up the device upon an event.
We need to flush ipoib_workqueue when we down the
device, and this cant be done from inside the same work queue.

>     Michael> Roland, I think at least the previous one-line patch
>     Michael> should go in to 2.6.13. Do you have it?
> 
> Yes, I have it in my queue.  I think it's too late to add this to
> 2.6.13, since it's not really critical, but we can propose it for
> 2.6.13.1.

You mean people dont unload the module all that much?


-- 
MST