Drivers: hv: vmbus: Perform device register in the per-channel work element

This patch is a continuation of the rescind handling cleanup work. We cannot
block in the global message handling work context especially if we are blocking
waiting for the host to wake us up. I would like to thank
Dexuan Cui <decui@microsoft.com> for observing this problem.

The current char-next branch is broken and this patch fixes
the bug.

Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

drivers/hv/channel_mgmt.c

@@ -23,6 +23,7 @@
 #include <linux/kernel.h>
 #include <linux/sched.h>
 #include <linux/wait.h>
+#include <linux/delay.h>
 #include <linux/mm.h>
 #include <linux/slab.h>
 #include <linux/list.h>
@@ -37,6 +38,10 @@ struct vmbus_channel_message_table_entry {
 	void (*message_handler)(struct vmbus_channel_message_header *msg);
 };
 
+struct vmbus_rescind_work {
+	struct work_struct work;
+	struct vmbus_channel *channel;
+};
 
 /**
  * vmbus_prep_negotiate_resp() - Create default response for Hyper-V Negotiate message
@@ -134,20 +139,6 @@ fw_error:
 
 EXPORT_SYMBOL_GPL(vmbus_prep_negotiate_resp);
 
-static void vmbus_process_device_unregister(struct work_struct *work)
-{
-	struct device *dev;
-	struct vmbus_channel *channel = container_of(work,
-							struct vmbus_channel,
-							work);
-
-	dev = get_device(&channel->device_obj->device);
-	if (dev) {
-		vmbus_device_unregister(channel->device_obj);
-		put_device(dev);
-	}
-}
-
 static void vmbus_sc_creation_cb(struct work_struct *work)
 {
 	struct vmbus_channel *newchannel = container_of(work,
@@ -220,6 +211,40 @@ static void free_channel(struct vmbus_channel *channel)
 	queue_work(vmbus_connection.work_queue, &channel->work);
 }
 
+static void process_rescind_fn(struct work_struct *work)
+{
+	struct vmbus_rescind_work *rc_work;
+	struct vmbus_channel *channel;
+	struct device *dev;
+
+	rc_work = container_of(work, struct vmbus_rescind_work, work);
+	channel = rc_work->channel;
+
+	/*
+	 * We have already acquired a reference on the channel
+	 * and so it cannot vanish underneath us.
+	 * It is possible (while very unlikely) that we may
+	 * get here while the processing of the initial offer
+	 * is still not complete. Deal with this situation by
+	 * just waiting until the channel is in the correct state.
+	 */
+
+	while (channel->work.func != release_channel)
+		msleep(1000);
+
+	if (channel->device_obj) {
+		dev = get_device(&channel->device_obj->device);
+		if (dev) {
+			vmbus_device_unregister(channel->device_obj);
+			put_device(dev);
+		}
+	} else {
+		hv_process_channel_removal(channel,
+					   channel->offermsg.child_relid);
+	}
+	kfree(work);
+}
+
 static void percpu_channel_enq(void *arg)
 {
 	struct vmbus_channel *channel = arg;
@@ -282,6 +307,46 @@ void vmbus_free_channels(void)
 	}
 }
 
+static void vmbus_do_device_register(struct work_struct *work)
+{
+	struct hv_device *device_obj;
+	int ret;
+	unsigned long flags;
+	struct vmbus_channel *newchannel = container_of(work,
+						     struct vmbus_channel,
+						     work);
+
+	ret = vmbus_device_register(newchannel->device_obj);
+	if (ret != 0) {
+		pr_err("unable to add child device object (relid %d)\n",
+			newchannel->offermsg.child_relid);
+		spin_lock_irqsave(&vmbus_connection.channel_lock, flags);
+		list_del(&newchannel->listentry);
+		device_obj = newchannel->device_obj;
+		newchannel->device_obj = NULL;
+		spin_unlock_irqrestore(&vmbus_connection.channel_lock, flags);
+
+		if (newchannel->target_cpu != get_cpu()) {
+			put_cpu();
+			smp_call_function_single(newchannel->target_cpu,
+					 percpu_channel_deq, newchannel, true);
+		} else {
+			percpu_channel_deq(newchannel);
+			put_cpu();
+		}
+
+		kfree(device_obj);
+		if (!newchannel->rescind) {
+			free_channel(newchannel);
+			return;
+		}
+	}
+	/*
+	 * The next state for this channel is to be freed.
+	 */
+	INIT_WORK(&newchannel->work, release_channel);
+}
+
 /*
  * vmbus_process_offer - Process the offer by creating a channel/device
  * associated with this offer
@@ -291,7 +356,6 @@ static void vmbus_process_offer(struct vmbus_channel *newchannel)
 	struct vmbus_channel *channel;
 	bool fnew = true;
 	bool enq = false;
-	int ret;
 	unsigned long flags;
 
 	/* Make sure this is a new offer */
@@ -393,16 +457,13 @@ static void vmbus_process_offer(struct vmbus_channel *newchannel)
 	 * Add the new device to the bus. This will kick off device-driver
 	 * binding which eventually invokes the device driver's AddDevice()
 	 * method.
+	 * Invoke this call on the per-channel work context.
+	 * Until we return from this function, rescind offer message
+	 * cannot be processed as we are running on the global message
+	 * handling work.
 	 */
-	ret = vmbus_device_register(newchannel->device_obj);
+	INIT_WORK(&newchannel->work, vmbus_do_device_register);
-	if (ret != 0) {
+	queue_work(newchannel->controlwq, &newchannel->work);
-		pr_err("unable to add child device object (relid %d)\n",
-			   newchannel->offermsg.child_relid);
-
-		kfree(newchannel->device_obj);
-		goto err_deq_chan;
-	}
-
 	return;
 
 err_deq_chan:
@@ -556,33 +617,31 @@ static void vmbus_onoffer_rescind(struct vmbus_channel_message_header *hdr)
 {
 	struct vmbus_channel_rescind_offer *rescind;
 	struct vmbus_channel *channel;
-	unsigned long flags;
+	struct vmbus_rescind_work *rc_work;
 
 	rescind = (struct vmbus_channel_rescind_offer *)hdr;
-	channel = relid2channel(rescind->child_relid);
+	channel = relid2channel(rescind->child_relid, true);
 
 	if (channel == NULL) {
 		hv_process_channel_removal(NULL, rescind->child_relid);
 		return;
 	}
 
-	spin_lock_irqsave(&channel->lock, flags);
+	/*
-	channel->rescind = true;
+	 * We have acquired a reference on the channel and have posted
-	spin_unlock_irqrestore(&channel->lock, flags);
+	 * the rescind state. Perform further cleanup in a work context
-
+	 * that is different from the global work context in which
-	if (channel->device_obj) {
+	 * we process messages from the host (we are currently executing
-		/*
+	 * on that global context.
-		 * We will have to unregister this device from the
+	 */
-		 * driver core. Do this in the per-channel work context.
+	rc_work = kzalloc(sizeof(struct vmbus_rescind_work), GFP_KERNEL);
-		 * Note that we are currently executing on the global
+	if (!rc_work) {
-		 * workq for handling messages from the host.
+		pr_err("Unable to allocate memory for rescind processing ");
-		 */
+		return;
-		INIT_WORK(&channel->work, vmbus_process_device_unregister);
-		queue_work(channel->controlwq, &channel->work);
-	} else {
-		hv_process_channel_removal(channel,
-					   channel->offermsg.child_relid);
 	}
+	rc_work->channel = channel;
+	INIT_WORK(&rc_work->work, process_rescind_fn);
+	schedule_work(&rc_work->work);
 }
 
 /*

drivers/hv/connection.c

@@ -270,7 +270,7 @@ static struct vmbus_channel *pcpu_relid2channel(u32 relid)
  * relid2channel - Get the channel object given its
  * child relative id (ie channel id)
  */
-struct vmbus_channel *relid2channel(u32 relid)
+struct vmbus_channel *relid2channel(u32 relid, bool rescind)
 {
 	struct vmbus_channel *channel;
 	struct vmbus_channel *found_channel  = NULL;
@@ -282,6 +282,8 @@ struct vmbus_channel *relid2channel(u32 relid)
 	list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
 		if (channel->offermsg.child_relid == relid) {
 			found_channel = channel;
+			if (rescind)
+				found_channel->rescind = true;
 			break;
 		} else if (!list_empty(&channel->sc_list)) {
 			/*
@@ -292,6 +294,8 @@ struct vmbus_channel *relid2channel(u32 relid)
 							sc_list);
 				if (cur_sc->offermsg.child_relid == relid) {
 					found_channel = cur_sc;
+					if (rescind)
+						found_channel->rescind = true;
 					break;
 				}
 			}

drivers/hv/hyperv_vmbus.h

@@ -698,7 +698,7 @@ void vmbus_device_unregister(struct hv_device *device_obj);
 /* VmbusChildDeviceDestroy( */
 /* struct hv_device *); */
 
-struct vmbus_channel *relid2channel(u32 relid);
+struct vmbus_channel *relid2channel(u32 relid, bool rescind);
 
 void vmbus_free_channels(void);