Not sure if it would disrupt the code flow to explain why we're checking for the ai_dsm TRIM bit first. The bug report explains this nicely, wasn't sure if it should end up in the (seemingly sparsely-commented) sd code as well. Not a deal-breaker, though.