DWC OTG TX path optimisation for 11-current

[Andreas Schwarz]

On 24.09.15, Svatopluk Kraus wrote:

> I finally found a way how to get something what could be analyzed
> considering the problem with smcs warnings, slow console response, and
> other oddities. And it was Michal Meloun who figured out how the
> problem comes up. He was so enthusiastic and fast that he got me no
> chance, no chance at all to figure it out myself. ;)) Thanks goes to
> Bob Prohaska too who pointed me to stress2 which triggers the problem
> even with all KTR classes compiled in kernel.

This could explain why the legacy BSD scheduler was so stable compared to 
ULE. I'll patch and start my buildworld loop test (which will take a week
(10 loops)).

-asc