Hacker News new | past | comments | ask | show | jobs | submit login

> software proprietary bits

Only the DDR4 memory training blob (which someone could eventually rewrite).




Where are they getting the LTE baseband software from?


They're not of course. They just treat the modem as walled off peripheral black box like all the other rest of the hardware. They may load firmware on boot for LTE/Wifi/Bluetooth like most Linux laptops do. Depending on your opinion on the topic that may be free enough or not but it should be safe as none of those can access the CPU/RAM directly.


I see. The question here is not if there will be proprietary software: there will be. The question is which of that proprietary software runs on the CPU.

Is the idea that all of the software in the kernel to interface with these devices is open source?

I also don’t get the hangup over PCIe (vs USB). DMA with an IOMMU can be made fairly secure (and has obvious perf benefits).


> I see. The question here is not if there will be proprietary software: there will be. The question is which of that proprietary software runs on the CPU.

I'm pretty sure no closed source software runs on the CPU. As for the hardware that depends on your definition. Hardware often has firmware in ROM. That you now feed that firmware on bootup instead of it being in ROM just allows you to update the firmware from the manufacturer. If that's running proprietary software depends on your definition.


I think the main reason is that powering off a device (for their hardware kill switches) connected via USB is more reliable than powering off a device connected via PCIe.

There's also just more layers of security when using USB as opposed to a single layer with PCIe (IOMMU -- which is secure as far as I know but I'd prefer to be safe rather than sorry in this case).


Besides IOMMU bugs, which silicon vendors often subtly add to their errata too much for my comfort; an IOMMU adds security against DMA attacks in theory. I specifically say in theory because oftentimes vendors either don't configure it correctly or leave it totally unconfigured. Additionally, things like multiplexing on the same bus further complicates things


It depends on the board design. It costs more, but you can have controlled power line for a single device, controllable with a gpio.

Edit: it has a "radio hardware killswitch" ; so it's well designed.


That makes sense.

I thought of the kill switch thing after commenting… eGPUs (necessarily) have the ability to be unplugged while running, but I think it was a lot of work on the part of the OSes to make that work well.


What would be a benefit to putting the modems on PCIe bus? I don’t see any positives, only drawbacks.


Reminds me of a joke I heard:

What’s the difference between a person and a cellphone?

A person only has two arms.


They don't have the baseband on the same chip, the modem is connected over something like USB (and there's probably gonna be a physical kill switch).

IIRC this is fine even under FSF "Respects your Freedom" rules




Consider applying for YC's Spring batch! Applications are open till Feb 11.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: