Thank you Barry (and Gustavo) for providing closure on this one. :)
On Thu, Sep 12, 2019 at 04:18:00PM -0500, Barry Arndt wrote:
One of the problems was introduced in 4.12, and the other in 4.15.
For posterity I should note that I first encountered the issue back in 4.9 and bisected it to dc16b553c949e81f37555777dc7bab66d78285a7. It looks like the first commit you mentioned largely undoes that one.
In any case I confirmed that cherry-picking those two patches on to an affected kernel does fix the issue for me. Thank you!