vmbus_teardown_gpadl() can result in infinite wait when it is called
on 5 second timeout in vmbus_open(). The issue is caused by the fact
that gpadl teardown operation won't ever succeed for an opened channel
and the timeout isn't always enough. Increase the timeout to 15 seconds
and add sending CHANNELMSG_CLOSECHANNEL message on timeout to make sure
the channel is closed.
The other possible solution for the issue would be to remove the timeout
in vmbus_open() making our wait infinite.
Signed-off-by: Vitaly Kuznetsov <vkuznets@xxxxxxxxxx>
---
drivers/hv/channel.c | 14 +++++++++++++-
1 file changed, 13 insertions(+), 1 deletion(-)
diff --git a/drivers/hv/channel.c b/drivers/hv/channel.c
index 6a8660d..becb9e8 100644
--- a/drivers/hv/channel.c
+++ b/drivers/hv/channel.c
@@ -69,6 +69,7 @@ int vmbus_open(struct vmbus_channel *newchannel, u32 send_ringbuffer_size,
void (*onchannelcallback)(void *context), void *context)
{
struct vmbus_channel_open_channel *open_msg;
+ struct vmbus_channel_close_channel *close_msg;
struct vmbus_channel_msginfo *open_info = NULL;
void *in, *out;
unsigned long flags;
@@ -183,8 +184,19 @@ int vmbus_open(struct vmbus_channel *newchannel, u32 send_ringbuffer_size,
goto error1;
}
- t = wait_for_completion_timeout(&open_info->waitevent, 5*HZ);
+ t = wait_for_completion_timeout(&open_info->waitevent, 15*HZ);
if (t == 0) {
+ /*
+ * We won't be able to tear down the gpadl handle if the
+ * channel finally gets openned, send channel close message
+ * to be on the safe side.
+ */
+ close_msg = &newchannel->close_msg.msg;
+ close_msg->header.msgtype = CHANNELMSG_CLOSECHANNEL;
+ close_msg->child_relid = newchannel->offermsg.child_relid;
+ vmbus_post_msg(close_msg,
+ sizeof(struct vmbus_channel_close_channel));
+
err = -ETIMEDOUT;
goto error1;
}
--
2.5.5
_______________________________________________
devel mailing list
devel@xxxxxxxxxxxxxxxxxxxxxx
http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel
