> Quoting Moni Shoua <[EMAIL PROTECTED]>:
> Subject: Re: [openib-general] [RFC] [PATCH v2] IB/ipoib: Add bonding support 
> to IPoIB
> 
> Michael S. Tsirkin wrote:
> >> Quoting Moni Shoua <[EMAIL PROTECTED]>:
> >> Subject: Re: [openib-general] [RFC] [PATCH v2] IB/ipoib: Add bonding 
> >> support to IPoIB
> >>
> >> Michael S. Tsirkin wrote:
> >>>> Quoting Moni Shoua <[EMAIL PROTECTED]>:
> >>>> Subject: Re: [openib-general] [RFC] [PATCH v2] IB/ipoib: Add bonding 
> >>>> support to IPoIB
> >>>>
> >>>> This version of the patch tracks the allocs and releases of ipoib_neigh 
> >>>> and
> >>>>  keeps a list of them.  Before IPoIB net device unregisters the list is 
> >>>> passed
> >>>> to destroy ipoib_neighs that ride on on a bond neighbour.
> >>>>
> >>>> This is a replacement to the method of scanning the arp and ndisc
> >>>> tables.
> >>> Why does the list need to be global?
> >>> We already have a per-device list of paths, and each of these in turn
> >>> has a list of neighbours. Can't this be used?
> >>>
> >> OK, It's a good point but coming to think of it now I have a question
> >>
> >> When a device unregisters ipoib_stop() is called and all ipoib_neighs are 
> >> destroyed.
> >>    Isn't it enough to ensure that ipoib_neigh_destructor will not try to
> >>    "touch" one of the ib devs?  or in other words: Isn't it that the work 
> >> to
> >>    clean ipoib_neighs is unnecessary?
> >>
> Michael, Do you agree that destroying the ipoib_neighs 
> through (a call trace that starts with) ipoib_stop() is enough for safety
> and that there is no need to do that just before calling to 
> unregister_netdev() ?

Well, you have to make sure no one is caching either the device pointer
or the destructor/setup pointers.

-- 
MST

_______________________________________________
general mailing list
[email protected]
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general

To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general

Reply via email to