Flock 2025 Status Of ARM Devices On Fedora

I think we're almost ready to go. I'm excited the video actually just worked. I've never used this um laptop on a projector before. Um yeah, so I mean following on Dan from Dan and stuck between Dan and lunch, I will try and get this done as quick as possible. Um I haven't practiced this at all. I was asked to literally do this last night to fill in. Um so um let's go. Um I'll give myself an intro. Peter Robinson, I've worked in the Fedora ecosystem forever. Um I was directly involved in um various um bootstraps. So, I was involved in armv5 and then armv7 hard float and then arm 64. Um I maintain the Raspberry Pi support in Fedora and a whole lot of other arm stuff. Um as always, it's a fast-moving ecosystem. So, I thought um we'll just give a quick update because I'm always getting questions about various other bits and pieces. Um I worked at Red Hat for 12 years. I've recently moved to um themselves. Um not working on anything to do with Fedora at arm. Um but sort of still um actively involved in driving a whole lot of the open-source ecosystem behind the scenes. Um so support for various devices in Fedora. Um Raspberry Pi, it's the one that everyone knows and loves. It's widely available. Um Pi 3, mostly done. Pi 4, also mostly done. Um Some interesting sort of peripherals are starting to land upstream. So, things like hardware video encode decode, uh camera support, various other bits and pieces. Um, the Raspberry Pi 5, um, it works in Fedora 42. It's not useful in Fedora 42. Um, so we boot, we've got serial console, we've got micro SD storage, um, and general compute, um, it's a device that if you want to hack on the hardware and the kernel and stuff like that, you can start to do that on Fedora. Um, we're getting PCI support soon. The big blocker to make it useful is the Pi Foundation has decided, and if you read tea leaves and various other bits and pieces, you can probably work that, I mean, they've been from the Pi 1 all the way through to the Pi 5, um, they've been using Broadcom as the hardware, like the SOC vendor. Um, with the Pi 5, they split a whole lot of stuff out into a auxiliary chip, like they refer to it as a Southbridge chip, if you look at a traditional sort of X86 motherboard, where a lot of their um, you know, USB, network, audio, etc., etc., hang off this RP1 chip. So, a lot of the things that sort of just work from the Pi 2 to the Pi 3 to the Pi 4 are no longer there on the Pi 5, because they're using their own stuff, and they have this weird chip called the RP1 that contains all of this IP, and we're waiting for that support to land upstream. And most of the IP on that is like standard IP that's used across the ecosystem. So, once that sort of root device is there, a whole lot of things should just light up and work. Um the problem is waiting for it to get upstream. Um so, there are patches out there. It's moving forward nowhere near as fast as I or the community would like, but we're getting there. Um we have PCI working now um for 43. If the RP1 stuff lands in 617, um I will probably beg the Fedora kernel maintainer to pull that back. So, we're almost there. Um one of the things that GNOME and a bunch of the desktops started doing is auto suspend, and that's a big problem. Um someone in the community um started working on auto suspend, got it mostly there for Pi 3, and decided it was way too hard and walked away. Um so, yeah. But, we're getting there. The Pi Foundation is getting better. Like, they're the ones that are driving the video encode decode stuff upstream. They're leading a lot of the stuff in like the complex camera stacks and things like that. They are improving. They just have a lot of catch-up to do. And like every other company, they're a small company with limited resources, so they have to focus stuff. Um we're getting there. It's not perfect, but well, we are where we are. Nvidia. Um so, there are multiple Jetson generations. There's multiple devices coming out onto the market that vaguely look like Jetson devices, but are massive big like $3,000 workstation things. Like, they had Project Digits, which is now I think called DGX Spark, um things like that. Most of the IP in those devices actually look very similar to a Jetson. Um the Jetson Orin, which is the current generation of Jetson, is well supported. It's the first device where they support multi-purpose distributions, uh like multiple distributions rather than just their own. Um that was six or seven years of my work with Nvidia while I was at Red Hat to get that. Um the GPU driver is still out of tree. Um but the vast majority of the rest of the stuff just works on Fedora. You know, you can run a standard EDK2 uh UEFI, install it in the standard way, um encryption, various other bits and pieces, and it generally just works. You then need to pull down the GPU drivers and like the AI drivers, but, you know, that's no different to the rest of the Nvidia ecosystem there. And the nice thing about um Orin is they have a like a nano version, which is quite cheap and quite powerful, right up to the AGX version, which is a lot more expensive and powerful, but they all basically look the same. They're ARM 8.2 class of devices, so much more modern than a lot of the other ARM cores that are widely available. Um and you know, they're a good powerful device. Um Qualcomm um I think the Qualcomm stuff is very similar to the Raspberry Pi. There's a huge amount of interest in it because of the Windows on ARM laptops. Um the laptops are in a various state of support. Um the IoT SOCs are actually very well supported. Um Qualcomm is actively engaging in the system ready stuff. Their teams are actively testing Fedora on these devices and we're working at with them to sort of get firmwares in place so they just work. Um where with the laptops there's a bunch of work happening behind the scenes to improve the firmware to be more Linux friendly. Um and like with all big companies that just takes time. So hopefully soon they will improve. Um the device I have here that I'm presenting on is um X13S. That was the first generation where they act actively supported. Um we worked with Lenovo and Qualcomm to get the firmware enabled so that it just works. Um we're getting there on the newer generations and we're hopefully getting to a point where before long Linux support should be just there. Um but you know, it's I know there's a bunch of people in Fedora that are using the newer generation Qualcomm laptops and they work pretty well, but there's a lot of quirks involved. Um but [snorts] there when you get them up and running they're nice powerful devices. They typically just work. They need some more work around things like suspend resume. This will do suspend to idle, but you just sort of like not the deepest suspend mode, but I'll quite happily get 15 plus hours of battery life running Linux on this device. Um Rockchip's the older Rockchip's are really well supported. Like literally I don't think there's any bit of IP in these devices we don't support now. Cameras, all sorts of stuff just work. Um the newer 3500 series mostly works. Um improving rapidly. Uh GPU support is all there. Um there's bits of the firmware we can't distribute in Fedora. Um so, in the I think it was the 42 cycle, I started doing copper builds where we can deal with it there. Um so, it's a bit more complex to deal with. I'm working to improve that in the 43 life cycle. And we're even starting to get things like NPU support and various other bits and pieces there. There's a whole bunch of really interesting devices here, like handheld gaming devices, um e-paper tablets, various other bits and pieces where we're starting to get interested in the community, how do we support this? And it's like, "Ooh." You know, um and but like the base support is actually very good there. Um the GPU is a Mali GPU. It works really well out of the box. Um there's lots of optimization being done on that driver. Um and yeah, I think these devices are nice and cheap. Uh the uh the 3588 is pretty powerful. Um another 8.2 um 8.2 platform. Um and there's like newer platforms that are announced that aren't available yet, but you know, um and then there's a whole raft of other um stuff. So, AMD's Zynq uh have these relatively powerful devices with honking great FPGAs on board. Um TI has some interesting stuff around their PRUs and their real-time stuff. Um used extensively in robotics and those sort of use cases. Um working with a whole lot of those vendors as part of the system ready program and a platform called One Lab. Um One Lab has Fedora as one of the key distros that they test all of the hardware against. Um so through that program at Linaro, we're getting TIs, Xilinx, Renesas, and a whole lot of other hardware vendors that are just testing their devices on Fedora. Um filing tickets to get drivers enabled and things fixed, and everything just works. Um so there's a bunch of stuff there um around like real-time, co-processors, and FPGAs, um where we're seeing users actually do a whole lot of fun and interesting stuff with robotics, and a whole lot of things that I just wish I had the time and the money to play with. Um next-gen platforms. Um the Red Cedar Orion 6 is the first of the Arm v9 platforms out in the market. Um it's relatively uh reasonably priced for I think a 12-core um Arm v9 platform. Up to 64 gig of RAM, I think 64 gigs is about a $400 device. Um with the ACPI firmware, it just works with Fedora 42. Um that doesn't give you GPU and a whole lot of the fun stuff. Um so that's working with upstream. There's a whole lot of people in Arm internally that are interested in this cuz it's the first Arm v9 non-data center, non-phone sort of general-purpose platform. So I mean, I worked years ago with some of the Arm people to um enable things like PAC, BTI in Fedora, which is some really good security features that come with some of the arm CPUs these days and I think it's been there for like 5 years and we've never really been able to like widely test it because there's not that many sort of decent community devices out there and um pleasantly surprised that when we actually tested it on Fedora, it all just worked. Yay. Um there's also some like um earlier this year announced the Cortex-A320 which is a low-end um V9 device. I'm hoping like hell that it's going to quickly supplant um like the Cortex-A53s that are um 8.0 and actually really long in the tooth these days. Um there's a whole lot of rumored and interesting devices coming down the um pipe. I look forward to those sort of being out in the ecosystem. Um and yeah, so there's a bunch of interesting stuff there. Um and a lot of this stuff is um with the One Lab and things like that actually being tested on Fedora. Um so I'm hoping like as these things come to light um with a lot of the you know standardized arm IP out there now that a lot of them should actually just work. Um so you know it's ultimately um Yes, watch this space. Um yeah, so a few other bits and pieces. Um cloud and servers mostly just work. Um lots of people in the ecosystem, you know, Amazon, Microsoft, Google in the cloud space just working on optimizing their platforms for various different pieces. Um a lot of the boot firmware for the cheaper lower-end devices that we're supporting in Fedora are now getting a lot of useful and interesting um user the improvements like boot menus, um secure boot, HTTP boot, various other bits and pieces. Um arm is thankfully now engaging more and improving the Mali GPUs. Um the camera ecosystem is slowly improving. Um like the Pi Foundation is very much leading that charge, um which is nice. Qualcomm and and Nvidia less so, um but you know, things are getting better there. Um we're starting to see quite a community around some of the NPU stuff. Um the first devices are now supported in Mesa, and I'm working and looking at how we can integrate that with some of the AI workflows. So, things like llama.cpp and a whole bunch of stuff I'm sure Dan Dan went through on his talk. Um things like Vulcan compute are starting to get very interesting. Um I've been running and playing with Vulcan compute on this laptop a bit. Um and RustiCAL as well. Um so, [snorts] there's like I mean, they're not superpowered devices like a honking great Nvidia GPU, but you know, it's just enough sort of AI where you can actually do with voice models and things like that some pretty cool um stuff like um I'm looking at how I can possibly use some of these arm devices with NPUs and voice detection to do sort of Amazon Alexa stuff stuff on Fedora so that I don't have to um open my internal home to Amazon or Google or other clouds listening to me all the time. Um So, so yeah, um I'm leaving Apple to the AI people. Um you know, I've said to a bunch of them at different times, you know, we can collaborate there, let me know. Um but you know, generally they're doing very well. Um, a bunch of work behind the scenes with the Qualcomm team um, and their open source team to improve things for the Windows on ARM devices running Linux. Um, and so, hopefully some of their internal firmware work and things like that, um, as I mentioned, um, gets out to devices soon. Um, and yeah, so that's most of the ARM device ecosystem support in Fedora. Um, I was given very short notice for this presentation. Um, I was asked sort of late yesterday afternoon if I could fill in a slot. Um, so, do we have any questions? >> And so, can you explain a little bit about your plan for the proprietary firmwares? Are you going to bundle those with something like U-Boot and move U-Boot to Copper? Or are you going to have a separate firmwares just there for in this Copper and U-Boot still the same upstream for Fedora? >> Um, so the vast majority, like at the moment, the Rockchip 3500 series devices are in the Copper. The proprietary bit in that is the memory training driver, but it includes, like, that is part built as part of a firmware stack which includes U-Boot and other bits. So, there's um, ARM trusted firmware, U-Boot, um, a firmware for the power management co-processor, so the SCP, plus the memory training. So, for the devices we can support directly in Fedora, that will remain in Fedora proper. For the devices like the Rockchip 3500 series, um, we need to do it for the TI devices, we need to do it for a couple of the others, those ones alone will be built in copper um and separate because we can't ship them in Fedora proper, but they are like the same U-Boot builds and things like that. Um Initially, I'm going to be building the TI firmwares. It looks like TI may take a bunch of them over and just ship that as part of their device. Um at which point when that happens, I'm just going to drop them because we don't need to ship them. Um And um yeah, so it'll be SOC by SOC and it may be that like someone can rewrite the memory training as the like there's a bunch of memory training drivers in U-Boot proper that are open. Like so any of the devices pre um 3500 series in Rockchip, so the 3399s and the 3328s, things like that, were actually all open drivers. Um so if we get in a And so it seems like DDR5 for some reason is complicated and the drivers aren't open yet. Um and it may be they can be opened at some point. Maybe there's reasons they can't be. Um and so at the moment, it's a If it's fully open, it'll be in Fedora. If it's not, I'm going to do it in copper because at least that way we can support those while it's less than ideal, we can at least support those devices so people get a relatively good user experience. >> So, uh we had a brief discussion on on Wednesday about handhelds and I told you that on the low end I don't think there's a lot of interest to run Fedora on those, but on the high end, especially last year, there are two or three manufacturers that are coming out with these devices that run uh the Snapdragon 8 Gen 2 chips that high-end phones run. And for those, I think they also have a bit of a bigger screen. There might be an interest to run Fedora. So, can you comment a bit on Snapdragon support itself and >> I'm sorry, the Snapdragon support was very much in the Qualcomm So, Windows on ARM stuff. Google is moving Android to a new firmware bootloader called I think it's GUP or something like that. And that's actually a UEFI application that runs on a UEFI firmware. If they support that, we will be able to support Fedora actually quite well on those devices. There's a lot of work in like the ARM ecosystem. So, like Linaro has a team called the Qualcomm landing team that like is a collaboration between Linaro and Qualcomm to support devices in Linux and Android and all sorts of things like that. So, I'm not going to say for sure, but I am pretty certain we could support Fedora on those devices. And so, you know, if there's you know, if we can have a conversation with those manufacturers, yeah, I would say it's likely possible because a lot of the like depend like the 8cx Gen 3, which I think is in this laptop, in Linux is really well supported. And so, you know, if there's a handheld with that, it it wouldn't be a lot of effort. >> So, like many ARM devices are still maybe for legacy reasons basically booting from SD cards and not being actually like installed in the normal way. Do you think this ever going away or is this there always being this split like a laptop is just installed but on this arm devices I need to fiddle with SD cards and MMC and stuff like that? >> It depends. Um so if you look at like the Jetson platforms, um they have standard UEFI firmwares on standard on flash. You shove whatever storage you would like in them NVMe or whatever and you can install Fedora in the normal way. Um the lower end devices, I don't think it will go away just simply because it's um around cost. But there's a like the Red Cedar devices, a large chunk of them already come with SPI firmware and work in the more traditional X86 way. Um so and I've worked with a couple of the hardware manufacturers and said, "This is why we'd like SPI flash where we can put the firmware on there because it gives people a more easy to get started experience." And I think generally we're moving more in that direction as the devices become more capable. Um the hardware vendors are starting to see the value of spending, you know, a few extra cents on that on the device to get that experience. Um and I'm talking with a couple of the vendors as well about how we can actually with that SPI flash actually get them to do a high quality firmware that gets updates through Linux vendor firmware service. Um and you know, and so like that's something I'm working on. Some of those conversations are slow burner conversations that happen behind the scenes. Um At Linaro Connect recently, there was a discussion around CRA. Um and one of those device manufacturers went, "Oh, this doesn't apply to us." To which one of the CRA specialists said, "Do you sell your devices in Europe?" And they said, "Yes." "Do they run software?" And they said, "Yes." "You are affected by this." And the response was almost tears. Um and so, I think things like that, if a device vendor can say, "We support this tiny little firmware, and this is how we update it through Linux vendor firmware service, and by doing that, we can support Fedora and any distro that someone wants, that large chunk of software that is the Linux distro is the Linux distro's problem to deal with, and all we have to deal with is this tiny shimmer firmware that makes it more interesting. It's cheaper for them to put that chip on there and do a Linux vendor firmware and make um and partner with Linux distros for the rest of it. Um and so, there's things like that, which I think actually will quickly drive vendors to doing that, cuz it gives them this nice abstraction. So, it's an interesting law that's scaring the out of everyone. Um and um but I think from our point of view, it will end up being quite positive for that device support sort of thing. So, yeah, there's like there's a bunch of like slow-burning discussions there, where I wouldn't be surprised where I've, you know, had discussions with vendors and said like, "This is how you do it. This is, you know, various you know, reasons for it." And then suddenly devices start to appear working like that. >> Still have time for one question? No question then? >> Everyone wants lunch. >> Probably thank you. Thank you, Peter. >> Thank you.