I applied Marco's patch to the 4.9 release source. It did not patch cleanly (perhaps Marco's patch was against CVS and not the 4.9 release tag), but I was able to apply the patches by hand without any trouble.
With the patch in place and ASPM enabled, I was unable to replicate the bug. I transferred about 10GB worth of data with no problems. I will continue to run this kernel for the next few weeks. I also ran the same tests against the patched 4.9 kernel with ASPM disabled in the BIOS. This also worked fine, as it had before. It is interesting to note that with ASPM enabled, transfer rates were roughly 200 KB/sec slower, though I was still able to sustain transfer rates in excess of 7.2 MB/sec in either configuration. Perhaps this is an artifact of ASPM, I'm not sure. I also noticed that this bug exists on 4.8 (contrary to what I initially reported), though it is not as readily reproducible. Like 4.9, disabling ASPM in the bios seems to cure the problem. In summary, it appears that Marco's patch fixes the bug, or disabling ASPM does too. Thanks for the great work! Colin
