[PATCH net-next] hv_netvsc: Move subchannel waiting to rndis_filter_device_remove()

[PATCH net-next] hv_netvsc: Move subchannel waiting to rndis_filter_device_remove()

[Haiyang Zhang]

During hot add, vmbus_device_register() is called from vmbus_onoffer(), on
the same workqueue as the subchannel offer message work-queue, so 
subchannel offer won't be processed until the vmbus_device_register()/...
/netvsc_probe() is done.
Also, vmbus_device_register() is called with channel_mutex locked, which
prevents subchannel processing too. So the "waiting for sub-channel 
processing" will not success in hot add case. But, in usual module loading,
the netvsc_probe() is called from different code path, and doesn't fail.

This patch resolves the deadlock during NIC hot-add, and speeds up NIC 
loading time.

Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Reviewed-by: K. Y. Srinivasan <kys@microsoft.com>
---
 drivers/net/hyperv/rndis_filter.c |   19 +++++++++++++------
 1 files changed, 13 insertions(+), 6 deletions(-)

diff --git a/drivers/net/hyperv/rndis_filter.c b/drivers/net/hyperv/rndis_filter.c
index a37bbda..47d07c5 100644
--- a/drivers/net/hyperv/rndis_filter.c
+++ b/drivers/net/hyperv/rndis_filter.c
@@ -1175,22 +1175,18 @@ int rndis_filter_device_add(struct hv_device *dev,
 	ret = rndis_filter_set_rss_param(rndis_device, net_device->num_chn);
 
 	/*
-	 * Wait for the host to send us the sub-channel offers.
+	 * Set the number of sub-channels to be received.
 	 */
 	spin_lock_irqsave(&net_device->sc_lock, flags);
 	sc_delta = num_rss_qs - (net_device->num_chn - 1);
 	net_device->num_sc_offered -= sc_delta;
 	spin_unlock_irqrestore(&net_device->sc_lock, flags);
 
-	while (net_device->num_sc_offered != 0) {
-		t = wait_for_completion_timeout(&net_device->channel_init_wait, 10*HZ);
-		if (t == 0)
-			WARN(1, "Netvsc: Waiting for sub-channel processing");
-	}
 out:
 	if (ret) {
 		net_device->max_chn = 1;
 		net_device->num_chn = 1;
+		net_device->num_sc_offered = 0;
 	}
 
 	return 0; /* return 0 because primary channel can be used alone */
@@ -1204,6 +1200,17 @@ void rndis_filter_device_remove(struct hv_device *dev)
 {
 	struct netvsc_device *net_dev = hv_get_drvdata(dev);
 	struct rndis_device *rndis_dev = net_dev->extension;
+	unsigned long t;
+
+	/* If not all subchannel offers are complete, wait for them until
+	 * completion to avoid race.
+	 */
+	while (net_dev->num_sc_offered > 0) {
+		t = wait_for_completion_timeout(&net_dev->channel_init_wait,
+						10 * HZ);
+		if (t == 0)
+			WARN(1, "Netvsc: Waiting for sub-channel processing");
+	}
 
 	/* Halt and release the rndis device */
 	rndis_filter_halt_device(rndis_dev);
-- 
1.7.4.1

Re: [PATCH net-next] hv_netvsc: Move subchannel waiting to rndis_filter_device_remove()

[David Miller]

From: Haiyang Zhang <haiyangz@microsoft.com>
Date: Fri,  4 Mar 2016 15:07:43 -0800

> During hot add, vmbus_device_register() is called from vmbus_onoffer(), on
> the same workqueue as the subchannel offer message work-queue, so 
> subchannel offer won't be processed until the vmbus_device_register()/...
> /netvsc_probe() is done.
> Also, vmbus_device_register() is called with channel_mutex locked, which
> prevents subchannel processing too. So the "waiting for sub-channel 
> processing" will not success in hot add case. But, in usual module loading,
> the netvsc_probe() is called from different code path, and doesn't fail.
> 
> This patch resolves the deadlock during NIC hot-add, and speeds up NIC 
> loading time.
> 
> Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
> Reviewed-by: K. Y. Srinivasan <kys@microsoft.com>

Applied, thanks.