mt76x02: avoid status_list.lock and sta->rate_ctrl_lock dependency
authorStanislaw Gruszka <sgruszka@redhat.com>
Fri, 5 Apr 2019 11:42:56 +0000 (13:42 +0200)
committerKalle Valo <kvalo@codeaurora.org>
Fri, 12 Apr 2019 18:32:40 +0000 (21:32 +0300)
Move ieee80211_tx_status_ext() outside of status_list lock section
in order to avoid locking dependency and possible deadlock reposed by
LOCKDEP in below warning.

Also do mt76_tx_status_lock() just before it's needed.

[  440.224832] WARNING: possible circular locking dependency detected
[  440.224833] 5.1.0-rc2+ #22 Not tainted
[  440.224834] ------------------------------------------------------
[  440.224835] kworker/u16:28/2362 is trying to acquire lock:
[  440.224836] 0000000089b8cacf (&(&q->lock)->rlock#2){+.-.}, at: mt76_wake_tx_queue+0x4c/0xb0 [mt76]
[  440.224842]
               but task is already holding lock:
[  440.224842] 000000002cfedc59 (&(&sta->lock)->rlock){+.-.}, at: ieee80211_stop_tx_ba_cb+0x32/0x1f0 [mac80211]
[  440.224863]
               which lock already depends on the new lock.

[  440.224863]
               the existing dependency chain (in reverse order) is:
[  440.224864]
               -> #3 (&(&sta->lock)->rlock){+.-.}:
[  440.224869]        _raw_spin_lock_bh+0x34/0x40
[  440.224880]        ieee80211_start_tx_ba_session+0xe4/0x3d0 [mac80211]
[  440.224894]        minstrel_ht_get_rate+0x45c/0x510 [mac80211]
[  440.224906]        rate_control_get_rate+0xc1/0x140 [mac80211]
[  440.224918]        ieee80211_tx_h_rate_ctrl+0x195/0x3c0 [mac80211]
[  440.224930]        ieee80211_xmit_fast+0x26d/0xa50 [mac80211]
[  440.224942]        __ieee80211_subif_start_xmit+0xfc/0x310 [mac80211]
[  440.224954]        ieee80211_subif_start_xmit+0x38/0x390 [mac80211]
[  440.224956]        dev_hard_start_xmit+0xb8/0x300
[  440.224957]        __dev_queue_xmit+0x7d4/0xbb0
[  440.224968]        ip6_finish_output2+0x246/0x860 [ipv6]
[  440.224978]        mld_sendpack+0x1bd/0x360 [ipv6]
[  440.224987]        mld_ifc_timer_expire+0x1a4/0x2f0 [ipv6]
[  440.224989]        call_timer_fn+0x89/0x2a0
[  440.224990]        run_timer_softirq+0x1bd/0x4d0
[  440.224992]        __do_softirq+0xdb/0x47c
[  440.224994]        irq_exit+0xfa/0x100
[  440.224996]        smp_apic_timer_interrupt+0x9a/0x220
[  440.224997]        apic_timer_interrupt+0xf/0x20
[  440.224999]        cpuidle_enter_state+0xc1/0x470
[  440.225000]        do_idle+0x21a/0x260
[  440.225001]        cpu_startup_entry+0x19/0x20
[  440.225004]        start_secondary+0x135/0x170
[  440.225006]        secondary_startup_64+0xa4/0xb0
[  440.225007]
               -> #2 (&(&sta->rate_ctrl_lock)->rlock){+.-.}:
[  440.225009]        _raw_spin_lock_bh+0x34/0x40
[  440.225022]        rate_control_tx_status+0x4f/0xb0 [mac80211]
[  440.225031]        ieee80211_tx_status_ext+0x142/0x1a0 [mac80211]
[  440.225035]        mt76x02_send_tx_status+0x2e4/0x340 [mt76x02_lib]
[  440.225037]        mt76x02_tx_status_data+0x31/0x40 [mt76x02_lib]
[  440.225040]        mt76u_tx_status_data+0x51/0xa0 [mt76_usb]
[  440.225042]        process_one_work+0x237/0x5d0
[  440.225043]        worker_thread+0x3c/0x390
[  440.225045]        kthread+0x11d/0x140
[  440.225046]        ret_from_fork+0x3a/0x50
[  440.225047]
               -> #1 (&(&list->lock)->rlock#8){+.-.}:
[  440.225049]        _raw_spin_lock_bh+0x34/0x40
[  440.225052]        mt76_tx_status_skb_add+0x51/0x100 [mt76]
[  440.225054]        mt76x02u_tx_prepare_skb+0xbd/0x116 [mt76x02_usb]
[  440.225056]        mt76u_tx_queue_skb+0x5f/0x180 [mt76_usb]
[  440.225058]        mt76_tx+0x93/0x190 [mt76]
[  440.225070]        ieee80211_tx_frags+0x148/0x210 [mac80211]
[  440.225081]        __ieee80211_tx+0x75/0x1b0 [mac80211]
[  440.225092]        ieee80211_tx+0xde/0x110 [mac80211]
[  440.225105]        __ieee80211_tx_skb_tid_band+0x72/0x90 [mac80211]
[  440.225122]        ieee80211_send_auth+0x1f3/0x360 [mac80211]
[  440.225141]        ieee80211_auth.cold.40+0x6c/0x100 [mac80211]
[  440.225156]        ieee80211_mgd_auth.cold.50+0x132/0x15f [mac80211]
[  440.225171]        cfg80211_mlme_auth+0x149/0x360 [cfg80211]
[  440.225181]        nl80211_authenticate+0x273/0x2e0 [cfg80211]
[  440.225183]        genl_family_rcv_msg+0x196/0x3a0
[  440.225184]        genl_rcv_msg+0x47/0x8e
[  440.225185]        netlink_rcv_skb+0x3a/0xf0
[  440.225187]        genl_rcv+0x24/0x40
[  440.225188]        netlink_unicast+0x16d/0x210
[  440.225189]        netlink_sendmsg+0x204/0x3b0
[  440.225191]        sock_sendmsg+0x36/0x40
[  440.225193]        ___sys_sendmsg+0x259/0x2b0
[  440.225194]        __sys_sendmsg+0x47/0x80
[  440.225196]        do_syscall_64+0x60/0x1f0
[  440.225197]        entry_SYSCALL_64_after_hwframe+0x49/0xbe
[  440.225198]
               -> #0 (&(&q->lock)->rlock#2){+.-.}:
[  440.225200]        lock_acquire+0xb9/0x1a0
[  440.225202]        _raw_spin_lock_bh+0x34/0x40
[  440.225204]        mt76_wake_tx_queue+0x4c/0xb0 [mt76]
[  440.225215]        ieee80211_agg_start_txq+0xe8/0x2b0 [mac80211]
[  440.225225]        ieee80211_stop_tx_ba_cb+0xb8/0x1f0 [mac80211]
[  440.225235]        ieee80211_ba_session_work+0x1c1/0x2f0 [mac80211]
[  440.225236]        process_one_work+0x237/0x5d0
[  440.225237]        worker_thread+0x3c/0x390
[  440.225239]        kthread+0x11d/0x140
[  440.225240]        ret_from_fork+0x3a/0x50
[  440.225240]
               other info that might help us debug this:

[  440.225241] Chain exists of:
                 &(&q->lock)->rlock#2 --> &(&sta->rate_ctrl_lock)->rlock --> &(&sta->lock)->rlock

[  440.225243]  Possible unsafe locking scenario:

[  440.225244]        CPU0                    CPU1
[  440.225244]        ----                    ----
[  440.225245]   lock(&(&sta->lock)->rlock);
[  440.225245]                                lock(&(&sta->rate_ctrl_lock)->rlock);
[  440.225246]                                lock(&(&sta->lock)->rlock);
[  440.225247]   lock(&(&q->lock)->rlock#2);
[  440.225248]
                *** DEADLOCK ***

[  440.225249] 5 locks held by kworker/u16:28/2362:
[  440.225250]  #0: 0000000048fcd291 ((wq_completion)phy0){+.+.}, at: process_one_work+0x1b5/0x5d0
[  440.225252]  #1: 00000000f1c6828f ((work_completion)(&sta->ampdu_mlme.work)){+.+.}, at: process_one_work+0x1b5/0x5d0
[  440.225254]  #2: 00000000433d2b2c (&sta->ampdu_mlme.mtx){+.+.}, at: ieee80211_ba_session_work+0x5c/0x2f0 [mac80211]
[  440.225265]  #3: 000000002cfedc59 (&(&sta->lock)->rlock){+.-.}, at: ieee80211_stop_tx_ba_cb+0x32/0x1f0 [mac80211]
[  440.225276]  #4: 000000009d7b9a44 (rcu_read_lock){....}, at: ieee80211_agg_start_txq+0x33/0x2b0 [mac80211]
[  440.225286]
               stack backtrace:
[  440.225288] CPU: 2 PID: 2362 Comm: kworker/u16:28 Not tainted 5.1.0-rc2+ #22
[  440.225289] Hardware name: LENOVO 20KGS23S0P/20KGS23S0P, BIOS N23ET55W (1.30 ) 08/31/2018
[  440.225300] Workqueue: phy0 ieee80211_ba_session_work [mac80211]
[  440.225301] Call Trace:
[  440.225304]  dump_stack+0x85/0xc0
[  440.225306]  print_circular_bug.isra.38.cold.58+0x15c/0x195
[  440.225307]  check_prev_add.constprop.48+0x5f0/0xc00
[  440.225309]  ? check_prev_add.constprop.48+0x39d/0xc00
[  440.225311]  ? __lock_acquire+0x41d/0x1100
[  440.225312]  __lock_acquire+0xd98/0x1100
[  440.225313]  ? __lock_acquire+0x41d/0x1100
[  440.225315]  lock_acquire+0xb9/0x1a0
[  440.225317]  ? mt76_wake_tx_queue+0x4c/0xb0 [mt76]
[  440.225319]  _raw_spin_lock_bh+0x34/0x40
[  440.225321]  ? mt76_wake_tx_queue+0x4c/0xb0 [mt76]
[  440.225323]  mt76_wake_tx_queue+0x4c/0xb0 [mt76]
[  440.225334]  ieee80211_agg_start_txq+0xe8/0x2b0 [mac80211]
[  440.225344]  ieee80211_stop_tx_ba_cb+0xb8/0x1f0 [mac80211]
[  440.225354]  ieee80211_ba_session_work+0x1c1/0x2f0 [mac80211]
[  440.225356]  process_one_work+0x237/0x5d0
[  440.225358]  worker_thread+0x3c/0x390
[  440.225359]  ? wq_calc_node_cpumask+0x70/0x70
[  440.225360]  kthread+0x11d/0x140
[  440.225362]  ? kthread_create_on_node+0x40/0x40
[  440.225363]  ret_from_fork+0x3a/0x50

Cc: stable@vger.kernel.org
Fixes: 88046b2c9f6d ("mt76: add support for reporting tx status with skb")
Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
Acked-by: Felix Fietkau <nbd@nbd.name>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
drivers/net/wireless/mediatek/mt76/mt76x02_mac.c

index 9ed231abe91676119d751b06cfa995a7f5dd716c..4fe5a83ca5a41713d894a4210fe5ef0d68e47e17 100644 (file)
@@ -466,7 +466,6 @@ void mt76x02_send_tx_status(struct mt76x02_dev *dev,
                return;
 
        rcu_read_lock();
-       mt76_tx_status_lock(mdev, &list);
 
        if (stat->wcid < ARRAY_SIZE(dev->mt76.wcid))
                wcid = rcu_dereference(dev->mt76.wcid[stat->wcid]);
@@ -479,6 +478,8 @@ void mt76x02_send_tx_status(struct mt76x02_dev *dev,
                                          drv_priv);
        }
 
+       mt76_tx_status_lock(mdev, &list);
+
        if (wcid) {
                if (stat->pktid >= MT_PACKET_ID_FIRST)
                        status.skb = mt76_tx_status_skb_get(mdev, wcid,
@@ -498,7 +499,9 @@ void mt76x02_send_tx_status(struct mt76x02_dev *dev,
                if (*update == 0 && stat_val == stat_cache &&
                    stat->wcid == msta->status.wcid && msta->n_frames < 32) {
                        msta->n_frames++;
-                       goto out;
+                       mt76_tx_status_unlock(mdev, &list);
+                       rcu_read_unlock();
+                       return;
                }
 
                mt76x02_mac_fill_tx_status(dev, status.info, &msta->status,
@@ -514,11 +517,10 @@ void mt76x02_send_tx_status(struct mt76x02_dev *dev,
 
        if (status.skb)
                mt76_tx_status_skb_done(mdev, status.skb, &list);
-       else
-               ieee80211_tx_status_ext(mt76_hw(dev), &status);
-
-out:
        mt76_tx_status_unlock(mdev, &list);
+
+       if (!status.skb)
+               ieee80211_tx_status_ext(mt76_hw(dev), &status);
        rcu_read_unlock();
 }