fcoe: convert to kworker
authorSebastian Andrzej Siewior <bigeasy@linutronix.de>
Tue, 12 Apr 2016 15:16:54 +0000 (17:16 +0200)
committerMartin K. Petersen <martin.petersen@oracle.com>
Wed, 13 Jul 2016 03:16:31 +0000 (23:16 -0400)
The driver creates its own per-CPU threads which are updated based on
CPU hotplug events. It is also possible to use kworkers and remove some
of the kthread infrastrucure.

The code checked ->thread to decide if there is an active per-CPU
thread. By using the kworker infrastructure this is no longer
possible (or required). The thread pointer is saved in `kthread' instead
of `thread' so anything trying to use thread is caught by the
compiler. Currently only the bnx2fc driver is using struct fcoe_percpu_s
and the kthread member.

After a CPU went offline, we may still enqueue items on the "offline"
CPU. This isn't much of a problem. The work will be done on a random
CPU. The allocated crc_eof_page page won't be cleaned up. It is probably
expected that the CPU comes up at some point so it should not be a
problem. The crc_eof_page memory is released of course once the module
is removed.

This patch was only compile-tested due to -ENODEV.

Cc: Vasu Dev <vasu.dev@intel.com>
Cc: "James E.J. Bottomley" <jejb@linux.vnet.ibm.com>
Cc: "Martin K. Petersen" <martin.petersen@oracle.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: fcoe-devel@open-fcoe.org
Cc: linux-scsi@vger.kernel.org
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Tested-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
drivers/scsi/bnx2fc/bnx2fc_fcoe.c
drivers/scsi/fcoe/fcoe.c
include/scsi/libfcoe.h

index a188199..d6800af 100644 (file)
@@ -486,7 +486,7 @@ static int bnx2fc_rcv(struct sk_buff *skb, struct net_device *dev,
 
        __skb_queue_tail(&bg->fcoe_rx_list, skb);
        if (bg->fcoe_rx_list.qlen == 1)
-               wake_up_process(bg->thread);
+               wake_up_process(bg->kthread);
 
        spin_unlock(&bg->fcoe_rx_list.lock);
 
@@ -2715,7 +2715,7 @@ static int __init bnx2fc_mod_init(void)
        }
        wake_up_process(l2_thread);
        spin_lock_bh(&bg->fcoe_rx_list.lock);
-       bg->thread = l2_thread;
+       bg->kthread = l2_thread;
        spin_unlock_bh(&bg->fcoe_rx_list.lock);
 
        for_each_possible_cpu(cpu) {
@@ -2788,8 +2788,8 @@ static void __exit bnx2fc_mod_exit(void)
        /* Destroy global thread */
        bg = &bnx2fc_global;
        spin_lock_bh(&bg->fcoe_rx_list.lock);
-       l2_thread = bg->thread;
-       bg->thread = NULL;
+       l2_thread = bg->kthread;
+       bg->kthread = NULL;
        while ((skb = __skb_dequeue(&bg->fcoe_rx_list)) != NULL)
                kfree_skb(skb);
 
index 0efe711..f7c7ccc 100644 (file)
@@ -67,9 +67,6 @@ static DEFINE_MUTEX(fcoe_config_mutex);
 
 static struct workqueue_struct *fcoe_wq;
 
-/* fcoe_percpu_clean completion.  Waiter protected by fcoe_create_mutex */
-static DECLARE_COMPLETION(fcoe_flush_completion);
-
 /* fcoe host list */
 /* must only by accessed under the RTNL mutex */
 static LIST_HEAD(fcoe_hostlist);
@@ -80,7 +77,6 @@ static int fcoe_reset(struct Scsi_Host *);
 static int fcoe_xmit(struct fc_lport *, struct fc_frame *);
 static int fcoe_rcv(struct sk_buff *, struct net_device *,
                    struct packet_type *, struct net_device *);
-static int fcoe_percpu_receive_thread(void *);
 static void fcoe_percpu_clean(struct fc_lport *);
 static int fcoe_link_ok(struct fc_lport *);
 
@@ -107,7 +103,6 @@ static int fcoe_ddp_setup(struct fc_lport *, u16, struct scatterlist *,
 static int fcoe_ddp_done(struct fc_lport *, u16);
 static int fcoe_ddp_target(struct fc_lport *, u16, struct scatterlist *,
                           unsigned int);
-static int fcoe_cpu_callback(struct notifier_block *, unsigned long, void *);
 static int fcoe_dcb_app_notification(struct notifier_block *notifier,
                                     ulong event, void *ptr);
 
@@ -136,11 +131,6 @@ static struct notifier_block fcoe_notifier = {
        .notifier_call = fcoe_device_notification,
 };
 
-/* notification function for CPU hotplug events */
-static struct notifier_block fcoe_cpu_notifier = {
-       .notifier_call = fcoe_cpu_callback,
-};
-
 /* notification function for DCB events */
 static struct notifier_block dcb_notifier = {
        .notifier_call = fcoe_dcb_app_notification,
@@ -1245,152 +1235,21 @@ static int __exit fcoe_if_exit(void)
        return 0;
 }
 
-/**
- * fcoe_percpu_thread_create() - Create a receive thread for an online CPU
- * @cpu: The CPU index of the CPU to create a receive thread for
- */
-static void fcoe_percpu_thread_create(unsigned int cpu)
+static void fcoe_thread_cleanup_local(unsigned int cpu)
 {
-       struct fcoe_percpu_s *p;
-       struct task_struct *thread;
-
-       p = &per_cpu(fcoe_percpu, cpu);
-
-       thread = kthread_create_on_node(fcoe_percpu_receive_thread,
-                                       (void *)p, cpu_to_node(cpu),
-                                       "fcoethread/%d", cpu);
-
-       if (likely(!IS_ERR(thread))) {
-               kthread_bind(thread, cpu);
-               wake_up_process(thread);
-
-               spin_lock_bh(&p->fcoe_rx_list.lock);
-               p->thread = thread;
-               spin_unlock_bh(&p->fcoe_rx_list.lock);
-       }
-}
-
-/**
- * fcoe_percpu_thread_destroy() - Remove the receive thread of a CPU
- * @cpu: The CPU index of the CPU whose receive thread is to be destroyed
- *
- * Destroys a per-CPU Rx thread. Any pending skbs are moved to the
- * current CPU's Rx thread. If the thread being destroyed is bound to
- * the CPU processing this context the skbs will be freed.
- */
-static void fcoe_percpu_thread_destroy(unsigned int cpu)
-{
-       struct fcoe_percpu_s *p;
-       struct task_struct *thread;
        struct page *crc_eof;
-       struct sk_buff *skb;
-#ifdef CONFIG_SMP
-       struct fcoe_percpu_s *p0;
-       unsigned targ_cpu = get_cpu();
-#endif /* CONFIG_SMP */
-
-       FCOE_DBG("Destroying receive thread for CPU %d\n", cpu);
+       struct fcoe_percpu_s *p;
 
-       /* Prevent any new skbs from being queued for this CPU. */
-       p = &per_cpu(fcoe_percpu, cpu);
+       p = per_cpu_ptr(&fcoe_percpu, cpu);
        spin_lock_bh(&p->fcoe_rx_list.lock);
-       thread = p->thread;
-       p->thread = NULL;
        crc_eof = p->crc_eof_page;
        p->crc_eof_page = NULL;
        p->crc_eof_offset = 0;
        spin_unlock_bh(&p->fcoe_rx_list.lock);
 
-#ifdef CONFIG_SMP
-       /*
-        * Don't bother moving the skb's if this context is running
-        * on the same CPU that is having its thread destroyed. This
-        * can easily happen when the module is removed.
-        */
-       if (cpu != targ_cpu) {
-               p0 = &per_cpu(fcoe_percpu, targ_cpu);
-               spin_lock_bh(&p0->fcoe_rx_list.lock);
-               if (p0->thread) {
-                       FCOE_DBG("Moving frames from CPU %d to CPU %d\n",
-                                cpu, targ_cpu);
-
-                       while ((skb = __skb_dequeue(&p->fcoe_rx_list)) != NULL)
-                               __skb_queue_tail(&p0->fcoe_rx_list, skb);
-                       spin_unlock_bh(&p0->fcoe_rx_list.lock);
-               } else {
-                       /*
-                        * The targeted CPU is not initialized and cannot accept
-                        * new  skbs. Unlock the targeted CPU and drop the skbs
-                        * on the CPU that is going offline.
-                        */
-                       while ((skb = __skb_dequeue(&p->fcoe_rx_list)) != NULL)
-                               kfree_skb(skb);
-                       spin_unlock_bh(&p0->fcoe_rx_list.lock);
-               }
-       } else {
-               /*
-                * This scenario occurs when the module is being removed
-                * and all threads are being destroyed. skbs will continue
-                * to be shifted from the CPU thread that is being removed
-                * to the CPU thread associated with the CPU that is processing
-                * the module removal. Once there is only one CPU Rx thread it
-                * will reach this case and we will drop all skbs and later
-                * stop the thread.
-                */
-               spin_lock_bh(&p->fcoe_rx_list.lock);
-               while ((skb = __skb_dequeue(&p->fcoe_rx_list)) != NULL)
-                       kfree_skb(skb);
-               spin_unlock_bh(&p->fcoe_rx_list.lock);
-       }
-       put_cpu();
-#else
-       /*
-        * This a non-SMP scenario where the singular Rx thread is
-        * being removed. Free all skbs and stop the thread.
-        */
-       spin_lock_bh(&p->fcoe_rx_list.lock);
-       while ((skb = __skb_dequeue(&p->fcoe_rx_list)) != NULL)
-               kfree_skb(skb);
-       spin_unlock_bh(&p->fcoe_rx_list.lock);
-#endif
-
-       if (thread)
-               kthread_stop(thread);
-
        if (crc_eof)
                put_page(crc_eof);
-}
-
-/**
- * fcoe_cpu_callback() - Handler for CPU hotplug events
- * @nfb:    The callback data block
- * @action: The event triggering the callback
- * @hcpu:   The index of the CPU that the event is for
- *
- * This creates or destroys per-CPU data for fcoe
- *
- * Returns NOTIFY_OK always.
- */
-static int fcoe_cpu_callback(struct notifier_block *nfb,
-                            unsigned long action, void *hcpu)
-{
-       unsigned cpu = (unsigned long)hcpu;
-
-       switch (action) {
-       case CPU_ONLINE:
-       case CPU_ONLINE_FROZEN:
-               FCOE_DBG("CPU %x online: Create Rx thread\n", cpu);
-               fcoe_percpu_thread_create(cpu);
-               break;
-       case CPU_DEAD:
-       case CPU_DEAD_FROZEN:
-               FCOE_DBG("CPU %x offline: Remove Rx thread\n", cpu);
-               fcoe_percpu_thread_destroy(cpu);
-               break;
-       default:
-               break;
-       }
-       return NOTIFY_OK;
+       flush_work(&p->work);
 }
 
 /**
@@ -1509,26 +1368,6 @@ static int fcoe_rcv(struct sk_buff *skb, struct net_device *netdev,
 
        fps = &per_cpu(fcoe_percpu, cpu);
        spin_lock(&fps->fcoe_rx_list.lock);
-       if (unlikely(!fps->thread)) {
-               /*
-                * The targeted CPU is not ready, let's target
-                * the first CPU now. For non-SMP systems this
-                * will check the same CPU twice.
-                */
-               FCOE_NETDEV_DBG(netdev, "CPU is online, but no receive thread "
-                               "ready for incoming skb- using first online "
-                               "CPU.\n");
-
-               spin_unlock(&fps->fcoe_rx_list.lock);
-               cpu = cpumask_first(cpu_online_mask);
-               fps = &per_cpu(fcoe_percpu, cpu);
-               spin_lock(&fps->fcoe_rx_list.lock);
-               if (!fps->thread) {
-                       spin_unlock(&fps->fcoe_rx_list.lock);
-                       goto err;
-               }
-       }
-
        /*
         * We now have a valid CPU that we're targeting for
         * this skb. We also have this receive thread locked,
@@ -1543,8 +1382,7 @@ static int fcoe_rcv(struct sk_buff *skb, struct net_device *netdev,
         * in softirq context.
         */
        __skb_queue_tail(&fps->fcoe_rx_list, skb);
-       if (fps->thread->state == TASK_INTERRUPTIBLE)
-               wake_up_process(fps->thread);
+       schedule_work_on(cpu, &fps->work);
        spin_unlock(&fps->fcoe_rx_list.lock);
 
        return NET_RX_SUCCESS;
@@ -1713,15 +1551,6 @@ static int fcoe_xmit(struct fc_lport *lport, struct fc_frame *fp)
 }
 
 /**
- * fcoe_percpu_flush_done() - Indicate per-CPU queue flush completion
- * @skb: The completed skb (argument required by destructor)
- */
-static void fcoe_percpu_flush_done(struct sk_buff *skb)
-{
-       complete(&fcoe_flush_completion);
-}
-
-/**
  * fcoe_filter_frames() - filter out bad fcoe frames, i.e. bad CRC
  * @lport: The local port the frame was received on
  * @fp:           The received frame
@@ -1792,8 +1621,7 @@ static void fcoe_recv_frame(struct sk_buff *skb)
        fr = fcoe_dev_from_skb(skb);
        lport = fr->fr_dev;
        if (unlikely(!lport)) {
-               if (skb->destructor != fcoe_percpu_flush_done)
-                       FCOE_NETDEV_DBG(skb->dev, "NULL lport in skb\n");
+               FCOE_NETDEV_DBG(skb->dev, "NULL lport in skb\n");
                kfree_skb(skb);
                return;
        }
@@ -1857,40 +1685,28 @@ drop:
 }
 
 /**
- * fcoe_percpu_receive_thread() - The per-CPU packet receive thread
- * @arg: The per-CPU context
+ * fcoe_receive_work() - The per-CPU worker
+ * @work: The work struct
  *
- * Return: 0 for success
  */
-static int fcoe_percpu_receive_thread(void *arg)
+static void fcoe_receive_work(struct work_struct *work)
 {
-       struct fcoe_percpu_s *p = arg;
+       struct fcoe_percpu_s *p;
        struct sk_buff *skb;
        struct sk_buff_head tmp;
 
+       p = container_of(work, struct fcoe_percpu_s, work);
        skb_queue_head_init(&tmp);
 
-       set_user_nice(current, MIN_NICE);
-
-       while (!kthread_should_stop()) {
-
-               spin_lock_bh(&p->fcoe_rx_list.lock);
-               skb_queue_splice_init(&p->fcoe_rx_list, &tmp);
-
-               if (!skb_queue_len(&tmp)) {
-                       set_current_state(TASK_INTERRUPTIBLE);
-                       spin_unlock_bh(&p->fcoe_rx_list.lock);
-                       schedule();
-                       continue;
-               }
-
-               spin_unlock_bh(&p->fcoe_rx_list.lock);
+       spin_lock_bh(&p->fcoe_rx_list.lock);
+       skb_queue_splice_init(&p->fcoe_rx_list, &tmp);
+       spin_unlock_bh(&p->fcoe_rx_list.lock);
 
-               while ((skb = __skb_dequeue(&tmp)) != NULL)
-                       fcoe_recv_frame(skb);
+       if (!skb_queue_len(&tmp))
+               return;
 
-       }
-       return 0;
+       while ((skb = __skb_dequeue(&tmp)))
+               fcoe_recv_frame(skb);
 }
 
 /**
@@ -2450,36 +2266,19 @@ static int fcoe_link_ok(struct fc_lport *lport)
  *
  * Must be called with fcoe_create_mutex held to single-thread completion.
  *
- * This flushes the pending skbs by adding a new skb to each queue and
- * waiting until they are all freed.  This assures us that not only are
- * there no packets that will be handled by the lport, but also that any
- * threads already handling packet have returned.
+ * This flushes the pending skbs by flush the work item for each CPU. The work
+ * item on each possible CPU is flushed because we may have used the per-CPU
+ * struct of an offline CPU.
  */
 static void fcoe_percpu_clean(struct fc_lport *lport)
 {
        struct fcoe_percpu_s *pp;
-       struct sk_buff *skb;
        unsigned int cpu;
 
        for_each_possible_cpu(cpu) {
                pp = &per_cpu(fcoe_percpu, cpu);
 
-               if (!pp->thread || !cpu_online(cpu))
-                       continue;
-
-               skb = dev_alloc_skb(0);
-               if (!skb)
-                       continue;
-
-               skb->destructor = fcoe_percpu_flush_done;
-
-               spin_lock_bh(&pp->fcoe_rx_list.lock);
-               __skb_queue_tail(&pp->fcoe_rx_list, skb);
-               if (pp->fcoe_rx_list.qlen == 1)
-                       wake_up_process(pp->thread);
-               spin_unlock_bh(&pp->fcoe_rx_list.lock);
-
-               wait_for_completion(&fcoe_flush_completion);
+               flush_work(&pp->work);
        }
 }
 
@@ -2625,22 +2424,11 @@ static int __init fcoe_init(void)
        mutex_lock(&fcoe_config_mutex);
 
        for_each_possible_cpu(cpu) {
-               p = &per_cpu(fcoe_percpu, cpu);
+               p = per_cpu_ptr(&fcoe_percpu, cpu);
+               INIT_WORK(&p->work, fcoe_receive_work);
                skb_queue_head_init(&p->fcoe_rx_list);
        }
 
-       cpu_notifier_register_begin();
-
-       for_each_online_cpu(cpu)
-               fcoe_percpu_thread_create(cpu);
-
-       /* Initialize per CPU interrupt thread */
-       rc = __register_hotcpu_notifier(&fcoe_cpu_notifier);
-       if (rc)
-               goto out_free;
-
-       cpu_notifier_register_done();
-
        /* Setup link change notification */
        fcoe_dev_setup();
 
@@ -2652,12 +2440,6 @@ static int __init fcoe_init(void)
        return 0;
 
 out_free:
-       for_each_online_cpu(cpu) {
-               fcoe_percpu_thread_destroy(cpu);
-       }
-
-       cpu_notifier_register_done();
-
        mutex_unlock(&fcoe_config_mutex);
        destroy_workqueue(fcoe_wq);
        return rc;
@@ -2690,14 +2472,8 @@ static void __exit fcoe_exit(void)
        }
        rtnl_unlock();
 
-       cpu_notifier_register_begin();
-
-       for_each_online_cpu(cpu)
-               fcoe_percpu_thread_destroy(cpu);
-
-       __unregister_hotcpu_notifier(&fcoe_cpu_notifier);
-
-       cpu_notifier_register_done();
+       for_each_possible_cpu(cpu)
+               fcoe_thread_cleanup_local(cpu);
 
        mutex_unlock(&fcoe_config_mutex);
 
index de7e3ee..c6fbbb6 100644 (file)
@@ -319,14 +319,16 @@ struct fcoe_transport {
 
 /**
  * struct fcoe_percpu_s - The context for FCoE receive thread(s)
- * @thread:        The thread context
+ * @kthread:       The thread context (used by bnx2fc)
+ * @work:          The work item (used by fcoe)
  * @fcoe_rx_list:   The queue of pending packets to process
  * @page:          The memory page for calculating frame trailer CRCs
  * @crc_eof_offset: The offset into the CRC page pointing to available
  *                 memory for a new trailer
  */
 struct fcoe_percpu_s {
-       struct task_struct *thread;
+       struct task_struct *kthread;
+       struct work_struct work;
        struct sk_buff_head fcoe_rx_list;
        struct page *crc_eof_page;
        int crc_eof_offset;