It's not limited to Linux - the problem is that, while you can go around resetting increasingly large parts of the IO stack, lots of devices in the path may or

It's not limited to Linux - the problem is that, while you can go around resetting increasingly large parts of the IO stack, lots of devices in the path may or may not actually respond to "off" or "reset" when they're in a bad state, so a drive being in a bad state can cause the entire IO path it's in to go south for the winter if nobody is smart enough in the path. This is one of the actual reasons that enterprise drives can be better - almost all of them support an equivalent of TLER (time-limited error recovery), which is basically a programmable timeout on read/write errors. Most parts of the IO stack, hardware and software, on every platform I've used deal a lot better with explicit errors than a device that hasn't appeared to vanish but is acting like a black hole.