]> git.ipfire.org Git - thirdparty/linux.git/commitdiff
iavf: get rid of the crit lock
authorPrzemek Kitszel <przemyslaw.kitszel@intel.com>
Fri, 4 Apr 2025 10:23:21 +0000 (12:23 +0200)
committerTony Nguyen <anthony.l.nguyen@intel.com>
Tue, 3 Jun 2025 16:48:03 +0000 (09:48 -0700)
Get rid of the crit lock.
That frees us from the error prone logic of try_locks.

Thanks to netdev_lock() by Jakub it is now easy, and in most cases we were
protected by it already - replace crit lock by netdev lock when it was not
the case.

Lockdep reports that we should cancel the work under crit_lock [splat1],
and that was the scheme we have mostly followed since [1] by Slawomir.
But when that is done we still got into deadlocks [splat2]. So instead
we should look at the bigger problem, namely "weird locking/scheduling"
of the iavf. The first step to fix that is to remove the crit lock.
I will followup with a -next series that simplifies scheduling/tasks.

Cancel the work without netdev lock (weird unlock+lock scheme),
to fix the [splat2] (which would be totally ugly if we would kept
the crit lock).

Extend protected part of iavf_watchdog_task() to include scheduling
more work.

Note that the removed comment in iavf_reset_task() was misplaced,
it belonged to inside of the removed if condition, so it's gone now.

[splat1] - w/o this patch - The deadlock during VF removal:
     WARNING: possible circular locking dependency detected
     sh/3825 is trying to acquire lock:
      ((work_completion)(&(&adapter->watchdog_task)->work)){+.+.}-{0:0}, at: start_flush_work+0x1a1/0x470
          but task is already holding lock:
      (&adapter->crit_lock){+.+.}-{4:4}, at: iavf_remove+0xd1/0x690 [iavf]
          which lock already depends on the new lock.

[splat2] - when cancelling work under crit lock, w/o this series,
   see [2] for the band aid attempt
    WARNING: possible circular locking dependency detected
    sh/3550 is trying to acquire lock:
    ((wq_completion)iavf){+.+.}-{0:0}, at: touch_wq_lockdep_map+0x26/0x90
        but task is already holding lock:
    (&dev->lock){+.+.}-{4:4}, at: iavf_remove+0xa6/0x6e0 [iavf]
        which lock already depends on the new lock.

[1] fc2e6b3b132a ("iavf: Rework mutexes for better synchronisation")
[2] https://github.com/pkitszel/linux/commit/52dddbfc2bb60294083f5711a158a

Fixes: d1639a17319b ("iavf: fix a deadlock caused by rtnl and driver's lock circular dependencies")
Signed-off-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>
Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com>
Tested-by: Rafal Romanowski <rafal.romanowski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
drivers/net/ethernet/intel/iavf/iavf.h
drivers/net/ethernet/intel/iavf/iavf_ethtool.c
drivers/net/ethernet/intel/iavf/iavf_main.c

index 9de3e0ba37316c801bc8300e9eaf1a81502abcbd..f7a98ff43a57fb66f2c3d38929b90e94fe7e021f 100644 (file)
@@ -268,7 +268,6 @@ struct iavf_adapter {
        struct list_head vlan_filter_list;
        int num_vlan_filters;
        struct list_head mac_filter_list;
-       struct mutex crit_lock;
        /* Lock to protect accesses to MAC and VLAN lists */
        spinlock_t mac_vlan_list_lock;
        char misc_vector_name[IFNAMSIZ + 9];
index 03d86fe80ad9104af8bcd539b30dfa58717c1fcd..2b2b315205b5e06d828a91bc13ea809b7edc0e51 100644 (file)
@@ -1258,7 +1258,6 @@ static int iavf_add_fdir_ethtool(struct iavf_adapter *adapter, struct ethtool_rx
 {
        struct ethtool_rx_flow_spec *fsp = &cmd->fs;
        struct iavf_fdir_fltr *fltr;
-       int count = 50;
        int err;
 
        netdev_assert_locked(adapter->netdev);
@@ -1281,14 +1280,6 @@ static int iavf_add_fdir_ethtool(struct iavf_adapter *adapter, struct ethtool_rx
        if (!fltr)
                return -ENOMEM;
 
-       while (!mutex_trylock(&adapter->crit_lock)) {
-               if (--count == 0) {
-                       kfree(fltr);
-                       return -EINVAL;
-               }
-               udelay(1);
-       }
-
        err = iavf_add_fdir_fltr_info(adapter, fsp, fltr);
        if (!err)
                err = iavf_fdir_add_fltr(adapter, fltr);
@@ -1296,7 +1287,6 @@ static int iavf_add_fdir_ethtool(struct iavf_adapter *adapter, struct ethtool_rx
        if (err)
                kfree(fltr);
 
-       mutex_unlock(&adapter->crit_lock);
        return err;
 }
 
@@ -1439,9 +1429,9 @@ iavf_set_adv_rss_hash_opt(struct iavf_adapter *adapter,
 {
        struct iavf_adv_rss *rss_old, *rss_new;
        bool rss_new_add = false;
-       int count = 50, err = 0;
        bool symm = false;
        u64 hash_flds;
+       int err = 0;
        u32 hdrs;
 
        netdev_assert_locked(adapter->netdev);
@@ -1469,15 +1459,6 @@ iavf_set_adv_rss_hash_opt(struct iavf_adapter *adapter,
                return -EINVAL;
        }
 
-       while (!mutex_trylock(&adapter->crit_lock)) {
-               if (--count == 0) {
-                       kfree(rss_new);
-                       return -EINVAL;
-               }
-
-               udelay(1);
-       }
-
        spin_lock_bh(&adapter->adv_rss_lock);
        rss_old = iavf_find_adv_rss_cfg_by_hdrs(adapter, hdrs);
        if (rss_old) {
@@ -1506,8 +1487,6 @@ iavf_set_adv_rss_hash_opt(struct iavf_adapter *adapter,
        if (!err)
                iavf_schedule_aq_request(adapter, IAVF_FLAG_AQ_ADD_ADV_RSS_CFG);
 
-       mutex_unlock(&adapter->crit_lock);
-
        if (!rss_new_add)
                kfree(rss_new);
 
index bf8c7baf2ab8ae31a2cb07f58f53c7beec40f173..2c0bb41809a4149be58c735c02495e7c35b39536 100644 (file)
@@ -1287,9 +1287,7 @@ static void iavf_configure(struct iavf_adapter *adapter)
 /**
  * iavf_up_complete - Finish the last steps of bringing up a connection
  * @adapter: board private structure
- *
- * Expects to be called while holding crit_lock.
- **/
+ */
 static void iavf_up_complete(struct iavf_adapter *adapter)
 {
        netdev_assert_locked(adapter->netdev);
@@ -1412,9 +1410,7 @@ static void iavf_clear_adv_rss_conf(struct iavf_adapter *adapter)
 /**
  * iavf_down - Shutdown the connection processing
  * @adapter: board private structure
- *
- * Expects to be called while holding crit_lock.
- **/
+ */
 void iavf_down(struct iavf_adapter *adapter)
 {
        struct net_device *netdev = adapter->netdev;
@@ -2029,22 +2025,21 @@ err:
  * iavf_finish_config - do all netdev work that needs RTNL
  * @work: our work_struct
  *
- * Do work that needs both RTNL and crit_lock.
- **/
+ * Do work that needs RTNL.
+ */
 static void iavf_finish_config(struct work_struct *work)
 {
        struct iavf_adapter *adapter;
-       bool locks_released = false;
+       bool netdev_released = false;
        int pairs, err;
 
        adapter = container_of(work, struct iavf_adapter, finish_config);
 
        /* Always take RTNL first to prevent circular lock dependency;
-        * The dev->lock is needed to update the queue number
+        * the dev->lock (== netdev lock) is needed to update the queue number.
         */
        rtnl_lock();
        netdev_lock(adapter->netdev);
-       mutex_lock(&adapter->crit_lock);
 
        if ((adapter->flags & IAVF_FLAG_SETUP_NETDEV_FEATURES) &&
            adapter->netdev->reg_state == NETREG_REGISTERED &&
@@ -2063,22 +2058,21 @@ static void iavf_finish_config(struct work_struct *work)
                netif_set_real_num_tx_queues(adapter->netdev, pairs);
 
                if (adapter->netdev->reg_state != NETREG_REGISTERED) {
-                       mutex_unlock(&adapter->crit_lock);
                        netdev_unlock(adapter->netdev);
-                       locks_released = true;
+                       netdev_released = true;
                        err = register_netdevice(adapter->netdev);
                        if (err) {
                                dev_err(&adapter->pdev->dev, "Unable to register netdev (%d)\n",
                                        err);
 
                                /* go back and try again.*/
-                               mutex_lock(&adapter->crit_lock);
+                               netdev_lock(adapter->netdev);
                                iavf_free_rss(adapter);
                                iavf_free_misc_irq(adapter);
                                iavf_reset_interrupt_capability(adapter);
                                iavf_change_state(adapter,
                                                  __IAVF_INIT_CONFIG_ADAPTER);
-                               mutex_unlock(&adapter->crit_lock);
+                               netdev_unlock(adapter->netdev);
                                goto out;
                        }
                }
@@ -2094,10 +2088,8 @@ static void iavf_finish_config(struct work_struct *work)
        }
 
 out:
-       if (!locks_released) {
-               mutex_unlock(&adapter->crit_lock);
+       if (!netdev_released)
                netdev_unlock(adapter->netdev);
-       }
        rtnl_unlock();
 }
 
@@ -2924,7 +2916,6 @@ static int iavf_watchdog_step(struct iavf_adapter *adapter)
        u32 reg_val;
 
        netdev_assert_locked(adapter->netdev);
-       lockdep_assert_held(&adapter->crit_lock);
 
        if (adapter->flags & IAVF_FLAG_PF_COMMS_FAILED)
                iavf_change_state(adapter, __IAVF_COMM_FAILED);
@@ -3044,22 +3035,7 @@ static void iavf_watchdog_task(struct work_struct *work)
        int msec_delay;
 
        netdev_lock(netdev);
-       if (!mutex_trylock(&adapter->crit_lock)) {
-               if (adapter->state == __IAVF_REMOVE) {
-                       netdev_unlock(netdev);
-                       return;
-               }
-
-               msec_delay = 20;
-               goto restart_watchdog;
-       }
-
        msec_delay = iavf_watchdog_step(adapter);
-
-       mutex_unlock(&adapter->crit_lock);
-restart_watchdog:
-       netdev_unlock(netdev);
-
        /* note that we schedule a different task */
        if (adapter->state >= __IAVF_DOWN)
                queue_work(adapter->wq, &adapter->adminq_task);
@@ -3067,6 +3043,7 @@ restart_watchdog:
        if (msec_delay != IAVF_NO_RESCHED)
                queue_delayed_work(adapter->wq, &adapter->watchdog_task,
                                   msecs_to_jiffies(msec_delay));
+       netdev_unlock(netdev);
 }
 
 /**
@@ -3074,8 +3051,7 @@ restart_watchdog:
  * @adapter: board private structure
  *
  * Set communication failed flag and free all resources.
- * NOTE: This function is expected to be called with crit_lock being held.
- **/
+ */
 static void iavf_disable_vf(struct iavf_adapter *adapter)
 {
        struct iavf_mac_filter *f, *ftmp;
@@ -3183,17 +3159,7 @@ static void iavf_reset_task(struct work_struct *work)
        int i = 0, err;
        bool running;
 
-       /* When device is being removed it doesn't make sense to run the reset
-        * task, just return in such a case.
-        */
        netdev_lock(netdev);
-       if (!mutex_trylock(&adapter->crit_lock)) {
-               if (adapter->state != __IAVF_REMOVE)
-                       queue_work(adapter->wq, &adapter->reset_task);
-
-               netdev_unlock(netdev);
-               return;
-       }
 
        iavf_misc_irq_disable(adapter);
        if (adapter->flags & IAVF_FLAG_RESET_NEEDED) {
@@ -3238,7 +3204,6 @@ static void iavf_reset_task(struct work_struct *work)
                dev_err(&adapter->pdev->dev, "Reset never finished (%x)\n",
                        reg_val);
                iavf_disable_vf(adapter);
-               mutex_unlock(&adapter->crit_lock);
                netdev_unlock(netdev);
                return; /* Do not attempt to reinit. It's dead, Jim. */
        }
@@ -3382,7 +3347,6 @@ continue_reset:
        adapter->flags &= ~IAVF_FLAG_REINIT_ITR_NEEDED;
 
        wake_up(&adapter->reset_waitqueue);
-       mutex_unlock(&adapter->crit_lock);
        netdev_unlock(netdev);
 
        return;
@@ -3393,7 +3357,6 @@ reset_err:
        }
        iavf_disable_vf(adapter);
 
-       mutex_unlock(&adapter->crit_lock);
        netdev_unlock(netdev);
        dev_err(&adapter->pdev->dev, "failed to allocate resources during reinit\n");
 }
@@ -3406,6 +3369,7 @@ static void iavf_adminq_task(struct work_struct *work)
 {
        struct iavf_adapter *adapter =
                container_of(work, struct iavf_adapter, adminq_task);
+       struct net_device *netdev = adapter->netdev;
        struct iavf_hw *hw = &adapter->hw;
        struct iavf_arq_event_info event;
        enum virtchnl_ops v_op;
@@ -3413,13 +3377,7 @@ static void iavf_adminq_task(struct work_struct *work)
        u32 val, oldval;
        u16 pending;
 
-       if (!mutex_trylock(&adapter->crit_lock)) {
-               if (adapter->state == __IAVF_REMOVE)
-                       return;
-
-               queue_work(adapter->wq, &adapter->adminq_task);
-               goto out;
-       }
+       netdev_lock(netdev);
 
        if (adapter->flags & IAVF_FLAG_PF_COMMS_FAILED)
                goto unlock;
@@ -3486,8 +3444,7 @@ static void iavf_adminq_task(struct work_struct *work)
 freedom:
        kfree(event.msg_buf);
 unlock:
-       mutex_unlock(&adapter->crit_lock);
-out:
+       netdev_unlock(netdev);
        /* re-enable Admin queue interrupt cause */
        iavf_misc_irq_enable(adapter);
 }
@@ -4180,8 +4137,8 @@ static int iavf_configure_clsflower(struct iavf_adapter *adapter,
                                    struct flow_cls_offload *cls_flower)
 {
        int tc = tc_classid_to_hwtc(adapter->netdev, cls_flower->classid);
-       struct iavf_cloud_filter *filter = NULL;
-       int err = -EINVAL, count = 50;
+       struct iavf_cloud_filter *filter;
+       int err;
 
        if (tc < 0) {
                dev_err(&adapter->pdev->dev, "Invalid traffic class\n");
@@ -4191,17 +4148,10 @@ static int iavf_configure_clsflower(struct iavf_adapter *adapter,
        filter = kzalloc(sizeof(*filter), GFP_KERNEL);
        if (!filter)
                return -ENOMEM;
-
-       while (!mutex_trylock(&adapter->crit_lock)) {
-               if (--count == 0) {
-                       kfree(filter);
-                       return err;
-               }
-               udelay(1);
-       }
-
        filter->cookie = cls_flower->cookie;
 
+       netdev_lock(adapter->netdev);
+
        /* bail out here if filter already exists */
        spin_lock_bh(&adapter->cloud_filter_list_lock);
        if (iavf_find_cf(adapter, &cls_flower->cookie)) {
@@ -4235,7 +4185,7 @@ err:
        if (err)
                kfree(filter);
 
-       mutex_unlock(&adapter->crit_lock);
+       netdev_unlock(adapter->netdev);
        return err;
 }
 
@@ -4539,28 +4489,13 @@ static int iavf_open(struct net_device *netdev)
                return -EIO;
        }
 
-       while (!mutex_trylock(&adapter->crit_lock)) {
-               /* If we are in __IAVF_INIT_CONFIG_ADAPTER state the crit_lock
-                * is already taken and iavf_open is called from an upper
-                * device's notifier reacting on NETDEV_REGISTER event.
-                * We have to leave here to avoid dead lock.
-                */
-               if (adapter->state == __IAVF_INIT_CONFIG_ADAPTER)
-                       return -EBUSY;
-
-               usleep_range(500, 1000);
-       }
-
-       if (adapter->state != __IAVF_DOWN) {
-               err = -EBUSY;
-               goto err_unlock;
-       }
+       if (adapter->state != __IAVF_DOWN)
+               return -EBUSY;
 
        if (adapter->state == __IAVF_RUNNING &&
            !test_bit(__IAVF_VSI_DOWN, adapter->vsi.state)) {
                dev_dbg(&adapter->pdev->dev, "VF is already open.\n");
-               err = 0;
-               goto err_unlock;
+               return 0;
        }
 
        /* allocate transmit descriptors */
@@ -4579,9 +4514,7 @@ static int iavf_open(struct net_device *netdev)
                goto err_req_irq;
 
        spin_lock_bh(&adapter->mac_vlan_list_lock);
-
        iavf_add_filter(adapter, adapter->hw.mac.addr);
-
        spin_unlock_bh(&adapter->mac_vlan_list_lock);
 
        /* Restore filters that were removed with IFF_DOWN */
@@ -4594,8 +4527,6 @@ static int iavf_open(struct net_device *netdev)
 
        iavf_irq_enable(adapter, true);
 
-       mutex_unlock(&adapter->crit_lock);
-
        return 0;
 
 err_req_irq:
@@ -4605,8 +4536,6 @@ err_setup_rx:
        iavf_free_all_rx_resources(adapter);
 err_setup_tx:
        iavf_free_all_tx_resources(adapter);
-err_unlock:
-       mutex_unlock(&adapter->crit_lock);
 
        return err;
 }
@@ -4630,12 +4559,8 @@ static int iavf_close(struct net_device *netdev)
 
        netdev_assert_locked(netdev);
 
-       mutex_lock(&adapter->crit_lock);
-
-       if (adapter->state <= __IAVF_DOWN_PENDING) {
-               mutex_unlock(&adapter->crit_lock);
+       if (adapter->state <= __IAVF_DOWN_PENDING)
                return 0;
-       }
 
        set_bit(__IAVF_VSI_DOWN, adapter->vsi.state);
        /* We cannot send IAVF_FLAG_AQ_GET_OFFLOAD_VLAN_V2_CAPS before
@@ -4666,7 +4591,6 @@ static int iavf_close(struct net_device *netdev)
        iavf_change_state(adapter, __IAVF_DOWN_PENDING);
        iavf_free_traffic_irqs(adapter);
 
-       mutex_unlock(&adapter->crit_lock);
        netdev_unlock(netdev);
 
        /* We explicitly don't free resources here because the hardware is
@@ -4685,11 +4609,10 @@ static int iavf_close(struct net_device *netdev)
                                    msecs_to_jiffies(500));
        if (!status)
                netdev_warn(netdev, "Device resources not yet released\n");
-
        netdev_lock(netdev);
-       mutex_lock(&adapter->crit_lock);
+
        adapter->aq_required |= aq_to_restore;
-       mutex_unlock(&adapter->crit_lock);
+
        return 0;
 }
 
@@ -5198,17 +5121,16 @@ iavf_shaper_set(struct net_shaper_binding *binding,
        struct iavf_adapter *adapter = netdev_priv(binding->netdev);
        const struct net_shaper_handle *handle = &shaper->handle;
        struct iavf_ring *tx_ring;
-       int ret = 0;
+       int ret;
 
        netdev_assert_locked(adapter->netdev);
 
-       mutex_lock(&adapter->crit_lock);
        if (handle->id >= adapter->num_active_queues)
-               goto unlock;
+               return 0;
 
        ret = iavf_verify_shaper(binding, shaper, extack);
        if (ret)
-               goto unlock;
+               return ret;
 
        tx_ring = &adapter->tx_rings[handle->id];
 
@@ -5218,9 +5140,7 @@ iavf_shaper_set(struct net_shaper_binding *binding,
 
        adapter->aq_required |= IAVF_FLAG_AQ_CONFIGURE_QUEUES_BW;
 
-unlock:
-       mutex_unlock(&adapter->crit_lock);
-       return ret;
+       return 0;
 }
 
 static int iavf_shaper_del(struct net_shaper_binding *binding,
@@ -5232,9 +5152,8 @@ static int iavf_shaper_del(struct net_shaper_binding *binding,
 
        netdev_assert_locked(adapter->netdev);
 
-       mutex_lock(&adapter->crit_lock);
        if (handle->id >= adapter->num_active_queues)
-               goto unlock;
+               return 0;
 
        tx_ring = &adapter->tx_rings[handle->id];
        tx_ring->q_shaper.bw_min = 0;
@@ -5243,8 +5162,6 @@ static int iavf_shaper_del(struct net_shaper_binding *binding,
 
        adapter->aq_required |= IAVF_FLAG_AQ_CONFIGURE_QUEUES_BW;
 
-unlock:
-       mutex_unlock(&adapter->crit_lock);
        return 0;
 }
 
@@ -5505,10 +5422,6 @@ static int iavf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
                goto err_alloc_qos_cap;
        }
 
-       /* set up the locks for the AQ, do this only once in probe
-        * and destroy them only once in remove
-        */
-       mutex_init(&adapter->crit_lock);
        mutex_init(&hw->aq.asq_mutex);
        mutex_init(&hw->aq.arq_mutex);
 
@@ -5578,9 +5491,7 @@ static int iavf_suspend(struct device *dev_d)
        running = netif_running(netdev);
        if (running)
                rtnl_lock();
-
        netdev_lock(netdev);
-       mutex_lock(&adapter->crit_lock);
 
        if (running)
                iavf_down(adapter);
@@ -5588,7 +5499,6 @@ static int iavf_suspend(struct device *dev_d)
        iavf_free_misc_irq(adapter);
        iavf_reset_interrupt_capability(adapter);
 
-       mutex_unlock(&adapter->crit_lock);
        netdev_unlock(netdev);
        if (running)
                rtnl_unlock();
@@ -5668,20 +5578,20 @@ static void iavf_remove(struct pci_dev *pdev)
         * There are flows where register/unregister netdev may race.
         */
        while (1) {
-               mutex_lock(&adapter->crit_lock);
+               netdev_lock(netdev);
                if (adapter->state == __IAVF_RUNNING ||
                    adapter->state == __IAVF_DOWN ||
                    adapter->state == __IAVF_INIT_FAILED) {
-                       mutex_unlock(&adapter->crit_lock);
+                       netdev_unlock(netdev);
                        break;
                }
                /* Simply return if we already went through iavf_shutdown */
                if (adapter->state == __IAVF_REMOVE) {
-                       mutex_unlock(&adapter->crit_lock);
+                       netdev_unlock(netdev);
                        return;
                }
 
-               mutex_unlock(&adapter->crit_lock);
+               netdev_unlock(netdev);
                usleep_range(500, 1000);
        }
        cancel_delayed_work_sync(&adapter->watchdog_task);
@@ -5691,7 +5601,6 @@ static void iavf_remove(struct pci_dev *pdev)
                unregister_netdev(netdev);
 
        netdev_lock(netdev);
-       mutex_lock(&adapter->crit_lock);
        dev_info(&adapter->pdev->dev, "Removing device\n");
        iavf_change_state(adapter, __IAVF_REMOVE);
 
@@ -5707,9 +5616,11 @@ static void iavf_remove(struct pci_dev *pdev)
 
        iavf_misc_irq_disable(adapter);
        /* Shut down all the garbage mashers on the detention level */
+       netdev_unlock(netdev);
        cancel_work_sync(&adapter->reset_task);
        cancel_delayed_work_sync(&adapter->watchdog_task);
        cancel_work_sync(&adapter->adminq_task);
+       netdev_lock(netdev);
 
        adapter->aq_required = 0;
        adapter->flags &= ~IAVF_FLAG_REINIT_ITR_NEEDED;
@@ -5727,8 +5638,6 @@ static void iavf_remove(struct pci_dev *pdev)
        /* destroy the locks only once, here */
        mutex_destroy(&hw->aq.arq_mutex);
        mutex_destroy(&hw->aq.asq_mutex);
-       mutex_unlock(&adapter->crit_lock);
-       mutex_destroy(&adapter->crit_lock);
        netdev_unlock(netdev);
 
        iounmap(hw->hw_addr);