Comment 17 for bug 1681439

OK, I can probably fix the alignment and use the new undo function to at least improve the characteristics of this failure considerably. Thanks for the pointer on the new iov function.

I see two potential problems still:

(1) The IDE device will allow partial transfers to succeed, by doing partial sector writes. AFAIUI the IDE state machine is supposed to do full sector or nothing at all, but maybe I am wrong about that being a requirement. It is at least not a regression, exactly. I can file a separate bug for this.

(2) The invalid device inputs are just completely unknown to me right now, and I don't have any hardware in my own house to test this, so I have to deprioritize that until I can get back into the office and regain access to more testing equipment. HELP WANTED, if anyone has a PCI BMDMA that they can orchestrate in a virtual environment to prod it for how it handles certain errant inputs.