On Fri, 2020-04-03 at 10:28 -0400, Mark Asselstine wrote: > I was looking into this as part of > https://bugzilla.kernel.org/show_bug.cgi?id=206971 and had a similar > fix in flight. My concern was that queue_owner being used outside of > the RCU might be an issue as now you have no guaranty that the > eventual use of sta->txq[tid] in iwl_mvm_free_inactive_queue() is > going to be valid. The only way to work around this is instead of > storing queue_owner, store mvmtxq = iwl_mvm_txq_from_tid(sta, i), then > adjust iwl_mvm_free_inactive_queue(), iwl_mvm_disable_txq() and > whatnot to take struct iwl_mvm_txq * instead of struct ieee80211_sta > *. If you open the bug you will see the latest version of my work as > the attached patch. I am not an RCU expert so I am curious to hear > your thoughts. I asked Johannes to check your comment. For now, I'm going to drop it from v2 of this patchset, so we can go ahead with the other patches. -- Cheers, Luca.
