Future Intel Systems To Reportedly Be Even Less Friendly For Open-Source Firmware

[icon]

To provide some thoughts about the original post: From a firmware developer's perspective, Intel moving further away from free firmware is pretty much expected. I believe it's actually part of their overall open-source strategy. One has to consider the whole software eco-system to see this: Intel is supporting Linux nowadays which is quite nice of course. However AFAICS, they only support it as open-source not as free software. Every time they want to support new silicon in Linux, they can decide if they will open source it or if they will put it into a firmware blob and hide the details from Linux developers. Of course, for this they need a proprietary firmware.

The issue is not very visible on consumer platforms where Intel doesn't want to hide things that happen at OS runtime. However, on Intel-based servers for instance, there are advanced features that they don't contribute as open-source. RAS comes to mind, that's Reliability, Availability and Serviceability. A set of features that is close to hardware parts that they traditionally keep under wraps (e.g. memory controller). So they put such OS features into SMM blobs instead of the kernel. I once joked that without blobs, Linux on Intel-based servers must be unreliable, unavailable and unmaintainable Actually, I don't know how well it would work without blobs, I don't work with servers, so please take this information with a grain of salt.

IMHO, this is an eco-system problem and we can't hope silicon vendors who profit from the status quo to fix this. And it doesn't matter if they produce RISC-V, ARM, OpenPower or x86, to some degree these issues exist everywhere. I don't want to blame OS developers, but there are some unsettling oddities in that direction. For instance, Linux-libre is practically rewarding proprietary firmware: They drop code for silicon where the vendor is honest about the blob situation and keep code for silicon where the vendor hides blobs in the firmware. I don't know how to fix this, free operating systems always have to compromise to support hardware that isn't designated for free software. I hope Linux will at some point have enough leverage and use it to abandon OS blobs hidden in firmware.

[Old Grouch]

It appears I am unable to post. Hi ho. And so it goes...

[Old Grouch]

The below text was a continuation of the conversation https://www.phoronix.com/forums/foru...11#post1311211, but for some reason shows up as 'unapproved'. Is this something I managed inadvertently to do, or does it mean that a moderator didn't like the text?

The direction taken with POWER10 is indeed unfortunate, and I hope that it is a glitch, and not just a cynical 'poisoning of the well'. Personally, I am not after a competitive general purpose cpu / computing device: rather one capable of doing basic cryptographic tasks with no known backdoorable glitches/design choices. As long as opaque BLOBs are part of the boot process or general operation of hardware, you have no guarantee that you are the owner - someone else can be doing 'stuff'.

The problem is, such adequately secure hardware can be used by people with less than pure motives, which gives a good arguing point for governments to ban, or at least impede its use: providing instead devices that are subject to the whims of the copyright rent-seekers and security services. The argument being: if you want to use adequately trustworthy and secure computing devices, you must have something to hide, and therefore be either a terrorist or child abuser.

Of course, if you trust your government, the question is moot. On the other hand, as Ukrainian people are finding out, you can't always predict what a future government might look like, and a future government may not be as benign as you hope. Good encryption on trustworthy hardware able to withstand nation-state attacks is needed, otherwise all of your important information, such as proof of identity, financial records, health records etc. can be wiped out or worse, subverted.

[Veto]

But the direction towards RAM-with-firmware is unlikely to change. Outside IBM, companies are developing CXL. Eventually all RAM sticks will move to serial interface. Just like how all SSD and HDD contain binary blobs, so will all the RAM sticks on board.

Indeed! If it was not for Rambus, we would likely already be there.

[billyswong]

There is indication that the problems with POWER10 are a direct result of the coronavirus pandemic and a resulting shortage of manpower at IBM. They didn't have time to develop a DDR4 to OMI bridge chip and so outsourced it to micron who used a synopsys PHY. IBM didn't have time to develop their own PCIe 5.0 host controller and so used a synopsys SERDES. In both cases, that means using the synopsys firmware that comes along with those IP blocks.

This is likely to be a temporary blip as the pandemic should be well and truly over by the time POWER11 comes around (or POWER12 for that matter). IBM took an enormous step moving the POWER ISA out, along with all of the associated patents, into a subsidiary of the Linux Foundation. I expect they will continue to honour that commitment.

This is a sharp contrast to sifive (for example) whose chips require blobs to boot, right from the birth of the company onwards.

But the direction towards RAM-with-firmware is unlikely to change. Outside IBM, companies are developing CXL. Eventually all RAM sticks will move to serial interface. Just like how all SSD and HDD contain binary blobs, so will all the RAM sticks on board.

[Old Grouch]

There is indication that the problems with POWER10 are a direct result of the coronavirus pandemic and a resulting shortage of manpower at IBM. They didn't have time to develop a DDR4 to OMI bridge chip and so outsourced it to micron who used a synopsys PHY. IBM didn't have time to develop their own PCIe 5.0 host controller and so used a synopsys SERDES. In both cases, that means using the synopsys firmware that comes along with those IP blocks.

This is likely to be a temporary blip as the pandemic should be well and truly over by the time POWER11 comes around (or POWER12 for that matter). IBM took an enormous step moving the POWER ISA out, along with all of the associated patents, into a subsidiary of the Linux Foundation. I expect they will continue to honour that commitment.

This is a sharp contrast to sifive (for example) whose chips require blobs to boot, right from the birth of the company onwards.

The direction taken with POWER10 is indeed unfortunate, and I hope that it is a glitch, and not just a cynical 'poisoning of the well'. Personally, I am not after a competitive general purpose cpu / computing device: rather one capable of doing basic cryptographic tasks with no known backdoorable glitches/design choices. As long as opaque BLOBs are part of the boot process or general operation of hardware, you have no guarantee that you are the owner - someone else can be doing 'stuff'.

The problem is, such adequately secure hardware can be used by people with less than pure motives, which gives a good arguing point for governments to ban, or at least impede its use: providing instead devices that are subject to the whims of the copyright rent-seekers and security services. The argument being: if you want to use adequately trustworthy and secure computing devices, you must have something to hide, and therefore be either a terrorist or child abuser.

Of course, if you trust your government, the question is moot. On the other hand, as Ukrainian people are finding out, you can't always predict what a future government might look like, and a future government may not be as benign as you hope. Good encryption on trustworthy hardware able to withstand nation-state attacks is needed, otherwise all of your important information, such as proof of identity, financial records, health records etc. can be wiped out or worse, subverted.

[Developer12]

There is indication that the problems with POWER10 are a direct result of the coronavirus pandemic and a resulting shortage of manpower at IBM. They didn't have time to develop a DDR4 to OMI bridge chip and so outsourced it to micron who used a synopsys PHY. IBM didn't have time to develop their own PCIe 5.0 host controller and so used a synopsys SERDES. In both cases, that means using the synopsys firmware that comes along with those IP blocks.

This is likely to be a temporary blip as the pandemic should be well and truly over by the time POWER11 comes around (or POWER12 for that matter). IBM took an enormous step moving the POWER ISA out, along with all of the associated patents, into a subsidiary of the Linux Foundation. I expect they will continue to honour that commitment.

This is a sharp contrast to sifive (for example) whose chips require blobs to boot, right from the birth of the company onwards.

[jacob]

Well, you are not wrong, but a viable royalty-free ISA is one step along the long road to FLOSS computing software and hardware. There are well known problems with the RISC-V foundation and its approach, but unfortunately, the POWER architecture has its own problems: POWER 10 requiring non-free BLOBS in its memory modules: https://www.phoronix.com/scan.php?pa...0-Blobs-Raptor

I suppose I'm slightly guilty of the 'if you wish for it hard enough, it will appear' school of thought, but I am also an eventualist: eventually the benefits of the FLOSS approach will be sufficiently obvious for it to become the majority approach. It might not happen in my lifetime though. I still use the money in my wallet to buy 'open' solutions, where possible. I'm looking forward to the Android/iOS effective duopoly being dismantled and for open, standards based information processing and communications protocols being used, including in telephony - again, might not happen in my lifetime: but so long as I can contribute to the eventual success, I will. There will be setbacks and diversions on the way, but the goal is clear.

I think if there ever is any improvement on this front, it will come from legislation. When IBM released the original PC, each unit sold came with a folder that contained the full source code of the BIOS. This should be mandated. I really don't expect the US to enforce any law like that, but the EU potentially might.

[Old Grouch]

There is nothing even remotely competitive there and probably won't be for a long time if ever. But that's not even the main issue. People keep bringing up RISC-V as some kind of saviour but I just don't see how's that supposed to work. Just because the ISA is royalty free doesn't mean the chips themselves including the microcode, the necessary BIOS etc can't be strictly proprietary. In fact it's the opposite since it's not a copyleft licence. Just like FreeBSD itself might be FOSS, the products derived from it are entirely proprietary and even overtly FOSS-hostile. It's the same here, the RISC-V products that are available now are every bit as closed as ARM.

Well, you are not wrong, but a viable royalty-free ISA is one step along the long road to FLOSS computing software and hardware. There are well known problems with the RISC-V foundation and its approach, but unfortunately, the POWER architecture has its own problems: POWER 10 requiring non-free BLOBS in its memory modules: https://www.phoronix.com/scan.php?pa...0-Blobs-Raptor

I suppose I'm slightly guilty of the 'if you wish for it hard enough, it will appear' school of thought, but I am also an eventualist: eventually the benefits of the FLOSS approach will be sufficiently obvious for it to become the majority approach. It might not happen in my lifetime though. I still use the money in my wallet to buy 'open' solutions, where possible. I'm looking forward to the Android/iOS effective duopoly being dismantled and for open, standards based information processing and communications protocols being used, including in telephony - again, might not happen in my lifetime: but so long as I can contribute to the eventual success, I will. There will be setbacks and diversions on the way, but the goal is clear.

[jacob]

RISC-V maybe. It'll be underpowered and costly.

There is nothing even remotely competitive there and probably won't be for a long time if ever. But that's not even the main issue. People keep bringing up RISC-V as some kind of saviour but I just don't see how's that supposed to work. Just because the ISA is royalty free doesn't mean the chips themselves including the microcode, the necessary BIOS etc can't be strictly proprietary. In fact it's the opposite since it's not a copyleft licence. Just like FreeBSD itself might be FOSS, the products derived from it are entirely proprietary and even overtly FOSS-hostile. It's the same here, the RISC-V products that are available now are every bit as closed as ARM.