Comment 6 for bug 1305406

I have tested the latest upstream kernel even more, and I have noticed something that may or may not help. I have found a "degradation effect" in which the total length of time the connection is hung within a given time frame increases the longer the system is running. To illustrate this effect, here are time frames from a sample system run in which I saw the aforementioned messages in the system log:

130.027554-277.358228 seconds since startup (147.330674 seconds long)
279.384357-299.437885 seconds since startup (20.053528 seconds long)
51257.090718-51261.102655 seconds since startup (4.011937 seconds long)
51575.931991-51579.948176 seconds since startup (4.016185 seconds long)
53066.713218-53099.649674 seconds since startup (32.936456 seconds long)
68674.880550-68738.963402 seconds since startup (64.082852 seconds long)
83473.503976-83619.708034 seconds since startup (146.204058 seconds long)
84739.202166-84739.204135 seconds since startup (0.001969 of a second long)
87142.062085-87198.136332 seconds since startup (56.074247 seconds long)
. . .

Suppose we group these time frames into 40,000 second time intervals. Between 0 and 40,000 seconds since startup, there are 56.074247 seconds worth of these messages. Between 40,000 and 80,000 seconds since startup, there are 105.04743 seconds worth of these messages. Between 80,000 and 120,000 seconds since startup, there are at least 202.280274 seconds worth of these messages.