Hacker News Reader

Comment by antsoul | Hacker News Reader

Thanks to Alyssa Rosenzweig and the Panfrost team for reverse-engineering the Mali GPU.

We can start moving out of the Intel/AMD backdoors and start to equip ourselves and our communities with trustworthy computers : ARM-based for now, RISC-V-based for later.

View on HN

I am sorry, but how does an open ISA guarantee lack of backdoors?

Technically, the x64 ISA is also open. It is patented to death, but its open. There is a free document describing every detail of it on Intel's website (AMD has one too).

Can't speak for OP, but maybe they mean things like Intel Management Engine, or cryptographicly locked bootloaders (not a backdoor, obviously, but a nuisance) with keys we can't control. So the backdoor is "feature", not a conspiracy.

And we are starting to see riscv chips with similar functions. That ISA is open doesn't mean the firmware must be open. And we have already riscv devices with additional homegrown hardware security "features".

Of course, you could avoid manufacturers that put this type of security crap in their chips. But with that logic you can already avoid Intel ME by using CPUs from Via...

(The point I'm trying to make is that an open ISA doesn't really solve all our problems. I feel Berkeley folks are overpromising here)

If the problem is the lack of trustful manufacturers, then RISC-V simplify the work of any fab that want to become this trustful manufacturer.

When we talk about security on Intel CPU, it's tragic. When we talk about security about RISC-V, it's a complex problem. That's a massive improvement. But yes, the work is far from over.

I'm not a fan at all of cryptocurrencies, but there are some interesting cryptographic ideas behind Polkadot (with Jeff Burdges) and Golem Project (with Joanna Rutkowska) to use those networks security to bring trust back to the laptop/phone level.

> We can start moving out of the Intel/AMD backdoors and start to equip ourselves and our communities with trustworthy computers : ARM-based for now, RISC-V-based for later.

What is your objection (if any) to the POWER9 architecture? Granted, I'm not sure if there is a decent laptop with the chip, but it is certainly a viable competitor to AMD and Intel in the desktop space.

Price.

I know Thalos has some FSF-certified boards based on POWER9 CPU and want to offer a laptop soon, but I would not expect anything under 3000$.

Won't that be only short term ? Or is the Mali GPU a generic thing that future iterations will not break the reverse engineering ?

There are three major revisions of Mali GPUs: Utgard (-4), Midgard (-T) and Bifrost (-G*).

There were 2 reverse engineering efforts. Lima (Utgard) and Panfrost project (Midgard and Bitfrost). Both were included in Mesa 19.1.

And the "long term" is not about "every pinebook will have a reverse-engineered GPU in the future", but : "if you buy these GPU, you know every compilers you write that makes special uses of the specific GPU capabilities will never need you to install the new proprietary CUDA driver to continue your optimization, no blob in this part..."

Also : You can run Linux-libre, thus riding the FSF train (GuixSD) and stop caring about not-free hardware.

What assures the trustworthiness of this RK3399 platform?

The same thing that "assures the trustworthiness" of every other computer ever made: nothing. What is the purpose of asking a question so vague as to be unanswerable?

> What is the purpose

Challenging a claim made by the previous poster?

It's not "so vague as to be unanswerable". It's an explicit response to the implication that non-Intel/AMD systems somehow lack backdoors. What's the absolute proof that this laptop (or some future RISC-V based one) lacks a backdoor?

That question isn't better. Of course there is no absolute proof. That doesn't mean the situation isn't significantly better then on AMD/Intel.

1. Large portions of the firmware and bootloader are open source. RAM initialization currently requires a blob from rockchip, but that too may not be needed in the future.

2. The GPU is usable with open source graphics drivers.

3. There is no equivalent to the IME or PSP that is present on x86 platforms.

3. ..not exactly, but it's open, so you could use it if you are able to use it :-)

https://github.com/rockchip-linux/rk_tee_user/blob/develop-n...

https://github.com/OP-TEE

https://www.op-tee.org/

https://optee.readthedocs.io/en/latest/

My idea is that a pool of free-hardware computer can be used to build up some community networks and clusters of hardware sharing their resources.

If you run a GNUnet node on a RISC-V router running seL4 [0] to log into your bank account, then it's pretty useless if your laptop runs an Intel CPU that access your keystrokes. Or if you use a cellphone.

But with those RK3399 platform, it's another level of security. I'd say that the pinephone / pinebook will not ruin your potential for network security. And that the trust is defined by the fab, and thus if Pinestore (and their fabs... are they located in Shenzhen ?) is doubtful, at least their design is getting easier to copy and make a trustful version yourself.

Also, I want GuixSD with GNU Hurd and seL4 sharing GPU resources for Vulkan rendering in the community. Not all this is ready yet, but I could never dream about it without the ARM phones/laptops/SoC.

[0]https://fosdem.org/2020/schedule/event/uk_sel4/

Reply on news.ycombinator.com