The only vendor that seems to take open firmware seriously seems to be Raptor, they are even blocking switching from open firmware POWER9 systems to POWER10 CPUs because of non-open RAM training code.

https://raptorcs.com/

The Power10 issue isn't just memory training. There's a programmable processor in the memory interface that runs a signed blob.

That's a much more serious problem -- potentially even worse than the Intel ME since the ME can't possibly intercept every single memory access without the user noticing the slowdown.

Is POWER10 usable *at all* without that blob?

Does the programmable processor itself enforce that signature?

Is there a way to modify the keyring for that verification?

> Is POWER10 usable at all without that blob?

No.

You might be able to use the on-die cache as RAM (most bootloaders do this), but that's less than 128MB. For the amount that these things cost nobody is going to want to use them with that little memory.

Yeah, if you're looking for an open-source workstation solution, my advice is simple: Give up. Buy a Mac.

To the extent that open source is tolerated at all by OEMs, it will be behind someone's sandbox, whether that be some sort of virtual machine or proprietary baseband firmware that runs as super, super root. Eventually, proprietary firmware will verify and attest that the software you run, from power on to end user application code, is authorized and approved, or the machine will refuse to run it. The CPU will shut off after 30s if it cannot find a vendor-signed, approved boot firmware. All for "security reasons", of course; whose security is not specified.

Today if you want to run entirely open source including firmware, your best shot is a prohibitively expensive workstation with a POWER CPU. That's what all of "general purpose computing" will look like soon; only possible with extremely expensive, boutique equipment. Eventually the profit margins on boutique hardware will crumble and no one without a large corporate or government account will be able to get their hands on hardware uncontrolled by someone's proprietary firmware or OS. That's just the economics of the situation.

So yeah, just buy a Mac. Apple may be a multitrillion dollar company, but a) they own very nearly EVERY piece of IP inside that Mac, and when they say "jump", the supply chain asks "how high", meaning that unlike other OEMs, Apple is beholden to NO ONE; b) they are the only OEM that comes remotely close to giving a shit about you, the end user. It's hard to overstate what a miraculous company Apple truly is.

Good points. But it also means when apple tells me to "bend over", I would have to say "how low? Please lubricate?".

My hope is for frame.work, pine64 and others to get to a point where risc-v or ppc models would be offered.

I gave up 20 years ago, for me Linux was only a means to an end, running UNIX on the cheap at home.

I have more hours delivering stuff into production using Solaris, HP-UX and Aix, than GNU/Linux.

In fact, had Microsoft been serious with their POSIX subsystem for Windows NT, I most likely wouldn't have bothered with Linux during the university, which for a big part of our lectures was still using DG/UX.

The success of macOS and Microsoft's WSL endeavors prove that most people only care to have some kind of POSIX support, regardless of the actual kernel.

Really hoping for RISC-V to take off and become a viable and mainstream alternative for consumer desktop machines so we can have a modern, fully open, and customizable computer.

RISC-V doesn't mandate an implementation's hardware and firmware stack be fully open, does it?

Not at all

I would love to believe in this, but in practice RISC-V has only gained traction in 32-bit microcontrollers, from companies that simply don't want to pay for an ARM license.

If there is a riscv64 server/workstation out there that I can buy, today, I would love to order one.

SiFive are the closest to that.

Which is what is so depressing.

They're the "closest", but not remotely close.

They have more RAM than my current desktop, so probably they are good enough for many people :)

Can you be more specific? Like which model you mean?

AFAICT all the SiFive products that are available to purchase and have workstation-ish amounts of RAM require boot blobs. I would be happy to be wrong about this.

https://forums.sifive.com/t/remaining-agency-issues-to-be-so...

Apparently there was a very limited run of dev boards for which no blobs are needed, and then they promptly discontinued that product and replaced it with a (confusingly-similarly-named) blobbified product that you can pre-order via mouser (listed as "backordered, no ETA"). This product is already marked EOL which is not encouraging.

I was just referring to their latest dev boards, which people have put in standard mini-ITX compatible cases, plugged in GPUs and used as workstations. Zero idea what blobs this requires. For eg:

https://drewdevault.com/2022/01/15/2022-01-15-The-RISC-V-exp...

> Today I’m writing this on the HiFive Unmatched

That board's firmware is not open-source at all. See the link in my previous comment.

> Zero idea what blobs this requires.

You've stumbled into the wrong thread, my friend. This thread is about open-source firmware. The stuff you're hyping does not deliver that.

I hardly stumbled into this thread, I was the one who posted it to HN.

I am definitely not hyping anything, least of all SiFive.

This subthread started with the need for RISC-V workstations, I simply pointed out the closest thing to one of those. Of course I didn't mention the firmware situation, which was a mistake, but I also didn't know about the situation.

Isin't risc mainly for embedded systems and low power usage?

ARM is a RISC architecture and enough to power Apple's Macbooks. Low power can also indicate high efficiency.

Reading this after the last Intel acquisition of Linutronix makes one wonder if Intel's plan is to force Linux systems to run their own set of tight closed kernel/patches, which of course would be insane in any security sensitive environment.

To be honest, it's not like Intel historically has had osf in mind more or less ever when designing their firmware.

And in a way I suppose it makes some sense, very few (laymen) care about open firmware.

But this is another bleak piece of reality being nailed down the into the open source coffin.

Don't worry peeps, RISC-V softcore porn killed the Wintel marriage.

Hahaha, my sides are splitting! Hahaha!

And the future is open-source firmware.

The catch (of the day)? Strong trademark protection and certification to be included in the global supply chain.

Print that out and trace it on your ass Mr. Stallman.

Hahaha. I'm sad.

Good thing there are alternatives.