Not All Of The IBM POWER10 Firmware Is Currently Open-Source

[ksec]

Both blobs are unsurprisingly connected to IP core designer Synopsys. I suspect they put some non-disclosure clause in contract.

Exactly. But let's blame IBM for it. There are probably enough POWER 10 material when you could design this without using those two IPs.

[pal666]

They couldn't or didn't want to keep their main advantage.

they never had an advantage. firmware is implementation detail of hardware, hardware is closed, no amount of free firmware will open it, it still can spy on you

[hiryu]

From #talos-workstation on FreeNode (well, liberal now), I was told a while back is that P10 is a casualty of covid. It sounds like IBM ended up cutting some corners. Sounds like P11 shouldn't have this problem.

Why covid would result in closed source firmware? I don't know... They had to be rather right lipped as this was non-public information. But maybe some of the P10 components were licensed from a 3rd due to a decrease in available manpower.

Hopefully this can still be rectified. I would LOVE to upgrade to P10! I have a Talos II in a colo and a blackbird based system at home. There were definitely some rough edges in the beginning but P9 support in Linux and especially in FreeBSD have really matured

[Developer12]

Both blobs are unsurprisingly connected to IP core designer Synopsys. I suspect they put some non-disclosure clause in contract.

If this is true, then I feel quite confident in saying that POWER10 will never, ever be free. Not unless IBM redesigns the chip to cut out the IP used in their I/O processor.

Even then, we'd need some outfit to make an open DDR4 bridge chip too.

[billyswong]

If this is true, then I feel quite confident in saying that POWER10 will never, ever be free. Not unless IBM redesigns the chip to cut out the IP used in their I/O processor.

Even then, we'd need some outfit to make an open DDR4 bridge chip too.

I heard that their new OMI RAM requires firmware in the RAM stick, like how SATA/NVMe SSD need firmware in itself. Good luck manufacturing one's own open firmware RAM stick for such a non-mainstream platform. We still don't have open firmware GPU even though such GPU, connecting via PCIe, will be compatible to all computer platforms and have a much larger market.

[phoron]

As opposed to the other options... oh wait, there are none.

How did you read me ? My other option is P9.

[phoron]

they never had an advantage.

But they did sell P9. They must have had some advantage.

firmware is implementation detail of hardware,

You can look at it however you want. In some advertisements which presumably work, they will just tell you that phone has an quadcore CPU, so much RAM and so many megapixels. Which lenses, which cores, how fast RAM, are all implementation details the buyer shouldn't worry about. The OS is android, so you can install java apps and the browser displays webs, so no need to care whether is 32 or 64 bit, these are all implementation details.

The difference is that loadable firmware can be changed. So it is software and not hardware.

hardware is closed, no amount of free firmware will open it, it still can spy on you

That's true. I find it great not to use hardware. But if you use it, it's a problem you can't avoid (unless instead of buying a computer you buy a whole industrial line with chip fabs, pick and place machines, test equipment, etc., and build your own hardware, which nobody can afford and if anyone could they would still need to trust employees).

Not adding suspicious firmware to that suspicious hardware is something you can avoid and something that I want to avoid. Just like crossing the street at a pedestrian crossing with a green light does reduce (not eliminate) the risk of a car running over me, even if I still have the risk of a heart attack. I reduce the risks I can, no those that I can't.

[phoron]

I heard that their new OMI RAM requires firmware in the RAM stick, like how SATA/NVMe SSD need firmware in itself. Good luck manufacturing one's own open firmware RAM stick for such a non-mainstream platform. We still don't have open firmware GPU even though such GPU, connecting via PCIe, will be compatible to all computer platforms and have a much larger market.

I guess there's always a way. A talos motherboard includes a FPGA because that was better than a custom ASIC for the volume. Raptor would presumably have choosen an off the self component if possible.maybe you can build some circuit to interface a P10 with some off-the-self sticks. Or maybe one can reverse engineer the firmware and publish free software to replace it. I don't know the licenses involved. For me the problem is not if there's a way. It's that it's probably too much effort if even possible, and that when I spend that money I want it to go to vendors that care to offer what I want, not to vendors I have to work around (or pay extra because someone has had to). If I have to use reverse engineered free software, then at least I look for cheap hardware (preferably second hand) and direct less money to those providers.

[hiryu]

You can find a bit more detail about the issues here:

https://www.talospace.com/2021/09/its-not-just-omi-thats-trouble-with.html

[Developer12]

I heard that their new OMI RAM requires firmware in the RAM stick, like how SATA/NVMe SSD need firmware in itself. Good luck manufacturing one's own open firmware RAM stick for such a non-mainstream platform. We still don't have open firmware GPU even though such GPU, connecting via PCIe, will be compatible to all computer platforms and have a much larger market.

The OMI memory lanes need a bridge chip to connect to regular DDR4 RAM. There is only one manufacturer of bridge chips, and they used a Synopsys IP block in the design that requires firmware. Synopsys doesn't like to give up it's secrets.

But it gets worse. They also have a firmware blob on the CPU, running some kind of I/O processor. If the comment I was referencing is correct, it's also based on Synopsys IP. If I had to make a total guess, I would wonder if it's a Synopsys SERDES or something being used for the PCIe 5.0. The blob firmware on github has some small hints that it might be some kind of calibration routine. (oh, and it's 16 bit, and zero-padded which makes me think it's not encrypted though it might be signed/checksummed; hard to tell at this stage)

Strangely, it seems IBM's automated tooling licenced the file for the latter as apache, which I doubt is correct.