Re: [PATCH 0/6 net-next] rhashtable fixes

[Thomas Graf <tgraf@suug.ch>]

On 01/30/15 at 05:10pm, Ying Xue wrote:
> Hi Thomas,
> 
> I make sure that my local net-next tree is synchronized to the latest
> version in which the commit fe6a043c535acfec8f8e554536c87923dcb45097
> ("rhashtable: rhashtable_remove() must unlink in both tbl and
> future_tbl") is already contained, and then I manually applied the whole
> series patches. But when I repeatedly run the test case I originally
> posted, soft lockup happens. Please see its relevant log:

Right, I see the same soft lockup. Interestingly I cannot trigger it
with the rht test code. I can only trigger it with your Netlink socket
creation stress test. It is definitely related to the deferred worker,
when I disable growing, then the bug disappears. I think that the
expansion leaves a race open in which remove cannot find certain entries
(I verified this by adding a BUG_ON() when rhashtable_remove() could not
find a match). This then keeps an entry on the list which has already
been freed.

However, I think this was present before these fixes but hidden as the
lockup requires a lot more iterations of your stress test on my
machine.