This fixes the recent commit:
Drivers: hv: vmbus: Further improve CPU affiliation logic
Without the fix, reloading hv_netvsc hangs the guest.
Signed-off-by: Dexuan Cui <decui@xxxxxxxxxxxxx>
---
Change since V1:
clear the related bit(s) in hv_process_channel_removal() rather
than in init_vp_index().
drivers/hv/channel_mgmt.c | 17 +++++++++++++++++
1 file changed, 17 insertions(+)
diff --git a/drivers/hv/channel_mgmt.c b/drivers/hv/channel_mgmt.c
index 2f9aead..652afd1 100644
--- a/drivers/hv/channel_mgmt.c
+++ b/drivers/hv/channel_mgmt.c
@@ -204,6 +204,8 @@ void hv_process_channel_removal(struct vmbus_channel *channel, u32 relid)
spin_lock_irqsave(&vmbus_connection.channel_lock, flags);
list_del(&channel->listentry);
spin_unlock_irqrestore(&vmbus_connection.channel_lock, flags);
+
+ primary_channel = channel;
} else {
primary_channel = channel->primary_channel;
spin_lock_irqsave(&primary_channel->lock, flags);
@@ -211,6 +213,14 @@ void hv_process_channel_removal(struct vmbus_channel *channel, u32 relid)
primary_channel->num_sc--;
spin_unlock_irqrestore(&primary_channel->lock, flags);
}
+
+ /*
+ * We need to free the bit for init_vp_index() to work in the case
+ * of sub-channel, when we reload drivers like hv_netvsc.
+ */
+ cpumask_clear_cpu(channel->target_cpu,
+ &primary_channel->alloced_cpus_in_node);
+
free_channel(channel);
}
@@ -458,6 +468,13 @@ static void init_vp_index(struct vmbus_channel *channel, const uuid_le *type_gui
continue;
}
+ /*
+ * NOTE: in the case of sub-channel, we clear the sub-channel
+ * related bit(s) in primary->alloced_cpus_in_node in
+ * hv_process_channel_removal(), so when we reload drivers
+ * like hv_netvsc in SMP guest, here we're able to re-allocate
+ * bit from primary->alloced_cpus_in_node.
+ */
if (!cpumask_test_cpu(cur_cpu,
&primary->alloced_cpus_in_node)) {
cpumask_set_cpu(cur_cpu,
--
2.1.0
_______________________________________________
devel mailing list
devel@xxxxxxxxxxxxxxxxxxxxxx
http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel
