nvme: Add support for Autonomous Power State Transition

[eseipi]

APST is an optional NVMe power-saving feature that allows devices to autonomously enter higher non-operational power states after a certain amount of idle time, reducing the controller's overall power consumption.

On my laptop with a Toshiba XG5, the ability to adjust APST settings allowed me to nearly double the battery life (from about 6 to 11 hours) and reduce the temperature of both the CPU and the controller by about 20°C.

At the moment, the code is very simple: the driver checks two new kenv variables (hw.nvme.apst_enable and hw.nvme.apst_data) at initialization and submit the Set Features command with those values, if any (for more info see 5.1.25.1.6 of NVM Express Base Specification rev. 2.1). If only apst_enable is set, the transition data is retrieved from the controller itself, but AFAIK the spec doesn't say anything about its presence being mandatory. Right now I only have two devices, and both have sane defaults.

Linux is doing the whole thing differently: each time it generates the data based on available power states, some set of user-provided parameters such as latency and tolerance, and a bit of heuristic. I used similar approach at first, but after afedorov@ feedback and a little more thought, moved all that kind of stuff to userspace (D47381). In either case, any further comments/ideas on this matter would be greatly appreciated.

[bsdimp]

Generally, I like this. I'm unsure of what we should do by default...

[wigneddoom]

This is a great feature. But for some reason, I think that we can implement it in user space - nvmecontrol(8)

[eseipi]

Though my approach to configuring APST closely resembles the Linux implementation, I agree that its design could be improved. In particular, I totally agree with @wigneddoom on moving as much functionality as possible out of kernel, though I'm still considering the best way to achieve this. I've marked this PR as a draft, so I can take some time to give it more thought. Please remove the needs-review label for now.

[eseipi]

I've created a revision (D47381) for nvmecontrol(8), which leaves just two things here: toggling (hw.nvme.apst_enable) and data overriding (hw.nvme.apst_data). This saves us from having to hook userspace when (re)initializing the controller, so I think it's reasonable to have this in the driver.

[eseipi]

Fixes:

• merge conflict with f08746a;
• poor memory management on nvme_ctrlr_cmd_get_feature failure;
• couple of style issues.