Upstreaming Linux kernel support for the Snapdragon X Elite | Hacker News Reader

Upstreaming Linux kernel support for the Snapdragon X Elite

qualcomm.com

174 pointsdblitt67 comments

I am excited about the prospect of Linux on another powerful aarch64 platform like Snapdragon Elite.

But, as far as I can tell from the diagram on the link shared, you will boot into EL1 and not EL2. This means that you cannot run a hardware accelerated VM on KVM (via something like qemu).

This makes a Snapdragon Linux laptop not as useful. BTW Asahi Linux on Apple Silicon enters on EL2 which allows qemu+KVM.

Entering on EL1 instead of EL2 seems to be be an outstanding issue with current Snapdragon based Linux laptops too. Can anybody correct me here if I'm wrong ?

> In short, our roadmap for the next six months includes work in these areas:

> End-to-end hardware video decoding, on Firefox and Chrome

> Implementation of the libcamera-SoftISP camera solution

> GPU and CPU performance optimizations

> Power optimizations (Suspend/DCVS)

> Making our firmware openly available (in Linux-firmware)

> Access to easy installers (Ubuntu and Debian)

"Making our firmware openly available (in Linux-firmware)" is a big one. How are firmware updates currently distributed ? People using Linux Laptops probably don't want to be signing into some Qualcomm website to get latest firmware updates. Also downloading firmware updates from some random link either would not instill a lot of confidence either.

I feel these 2 big items need to be addressed before Linux on Snapdragon can be a truly attractive option.

> "Making our firmware openly available (in Linux-firmware)" is a big one. How are firmware updates currently distributed ?

The "linux-firmware" repo is the standard place to put it, my reading is that Qualcomm are doing this in the right way.

The last time I checked, audio and video firmware wasn't there.

> The "linux-firmware" repo is the standard place to put it, my reading is that Qualcomm are doing this in the right way.

Indeed, Linux-firmware is the correct way to distribute. Qualcomm is currently _not_ doing this for all their Snapdragon Elite firmware (please read the linked article).

I think that for many use-cases containers already replaced VMs even for developers. And you obviously can't run accelerated x86 VMs either way. So probably won't be a big issue.

> And you obviously can't run accelerated x86 VMs either way.

But Qualcomm should allow users to run accelerated aarch64 VMs on a Snapdragon Elite (just like users can on Asahi Linux).

> I think that for many use-cases containers already replaced VMs even for developers.

If you're happy with running containers then you'll be fine. However, many developers like VMs for the extra isolation they bring and other features like running a full Linux distribution. Example: You may want to run Ubuntu on your Snapdragon Elite system but also want to run a full aarch64 NixOS within a VM.

Worst case there's always UserModeLinux.

«linux-firmware» is the goto place to dump firmware that goes to all linux distributions

Yes, "linux-firmware" is the way to do it -- however Qualcomm is currently NOT doing it for all their Snapdragon Elite firmware (as per the linked article). This is on their TODO list.

Tangentially related but I wish we had something similar to LVFS but for IoT devices.

That way, you would only need to have one hub (Homey, Home Assistant, IKEA, Apple TV, etc.) and it would be able to keep all your stuff up to date and control everything via Matter. Right now, we are slowly moving towards a future where any hub can control any device, but you still need to run the vendor hub if you want firmware updates.

The ARM vendors are annoyingly stingy about providing access to this. You’ll probably have to use Qualcomm’s proprietary hypervisor and even then it’s not clear whether they’ll give you enough access to run a VM.

> You’ll probably have to use Qualcomm’s proprietary hypervisor

This one?

https://github.com/quic/gunyah-hypervisor

Yes

Pointer for those not familiar with Arm exception levels: https://developer.arm.com/documentation/102412/0103/Privileg...

You'll have EL2 on Snapdragon X Elite.

If it's using the design that I prototyped while at Qualcomm, it's jumping to EL2 at ExitBootServices when the boot to Linux at EL2 option is enabled.

Didn't look at a release firmware yet to see what was the final impl, but was told that it did ship :)

The boot flow diagram on the article indicates the user gets dropped off into EL1 and not EL2, sadly though. The article is from Qualcomm so I would assume this the current official situation.

Out of the box that's true.

The system firmware will ship with 3 boot modes selectable via the setup interface:

- Windows (this one has the Windows tcblauncher escalated to EL2 through Secure Launch)

- Linux (this one stays at EL1)

- Linux w/ KVM (which jumps to EL2 before kernel handover)

This would be wonderful. However one would need to verify when these laptops actually start shipping. Let’s hope for the best !

Is this a "arm system ready " uefi firmware? Curious if you tested ESXi for aarch64 edition.

Arm SystemReady SR/ES assumes sane ACPI tables. That's something that Snapdragon X (1st-gen) very much doesn't have. The ACPI tables present there are pretty much only usable for Windows if you want full functionality.

Yeah ESXi-Arm is bootable, but a number of patches were required.

This is very cool, sure the performance isn’t as good as Apple’s stuff but a reasonably performant processor with official Linux support is very cool

Apple Silicon is officially supported in Linux.

I think by "official" they meant support from the manufacturer.

Laptops with this chip are not shipping yet, so it’s a bit early to judge performance.

There are some early benchmarks which put its performance behind iPad:

- Snapdragon X Elite (2,574 / 12,562)

- iPad M4 (3,630 / 13,060)

What benchmark is this? Geekbench?

Geekbench is notoriously inconsistent across platforms. Not saying the iPad chip isn't faster, but I wouldn't judge by Geekbench scores alone...

Do you have a source for Geekbench 6 being inconsistent across platforms?

What device was this benchmark run on?

Since we’re talking about Linux support, the iPad benchmarks seem irrelevant. It’s a completely different kernel.

Asahi Linux benchmarks on an M3 MacBook Air would be the closest thing here.

This is a CPU benchmark, not a test of what OS is running on it.

It’s common to have 5-10% differences on benchmarks between Linux and Windows on the same hardware. iPadOS is even more different (for example, no swap, and many other similar fundamental optimizations for single-app usage).

It seems very unlikely that Geekbench would swap

According to the asahi website, support is currently available for m1 and m2 macs only[1].

Time will tell about m3 macs.

[1]: https://asahilinux.org/fedora/#device-support

You can't benchmark a chip. It all depends on how much power and heat dissipation the actual device can provide. When the MacBook Pro comes out it should run circles round the iPad. And there are no Snapdragon X devices yet

Sure you can. Measure performance when not thermally throttled and at the lowest speed / no boost. All real devices will fall in between those numbers.

The metric I'm interested in is mostly system performance, not CPU performance.

How fast can a system compile my projects that is.

Looks like by next LTS (for kernel 6.10 and 6.11) we'll have some good support for these chips. I wonder if any OEMs will make any Linux laptops for us to buy.

I really hope system76 takes a look at them

There is the Huawei Qingyun. It's about 500-600 USD on Taobao

That's Kirin, not Snapdragon

I wouldn't recommend Huawei for anything. Also Lenovo is a chinese company.

I think it's best to wait for System76 or someone like Asus to get there.

I used a Huawei laptop and it's great. Huawei's build quality is significantly better than Asus. Not Apple level, but maybe half way there. Lenovo is also pretty good in quality (depending on the model though). Asus laptops are either cheap plastic or extremely thin metal that flexes - even on top end models. They also have a bad track record with support for their Tinkerboards. Not sure how the country of origin matters

Technically they might be fine, but there are geopolitical reasons to avoid chinese products.

I guess you need to avoid using any phone, tablet or laptop then unless all its chips are built in Taiwan. I know Apple Silicon is built there but what about other chips in their hardware?

For some things, we have more options than for others. It makes sense to make more choices where more choices exist.

Some chips are not a concern, but a CPU with its firmware most certainly is. NICs would be too, for obvious reasons.

For example, simple gate chips, or I/O chips are likely not a concern.

Also, many other places in the world than Taiwan make chips, they just make a lot.

It's a moot point if they can't get access to the chips.

don't forget that ASUS is so bad with warranties that GN not only once but twice now called them out and there are YouTubers who refuse to review ASUS devices.

System76 uses Clevo and Sager notebooks with little modifications. Both of them are purely Chinese. With your logic anything produced after ca. 1995 is unusable.

Clevo seems to be Taiwanese, not (mainland) Chinese?

As someone who has worked with Qualcomm extensively and developed products based on their QSDK for years, this thread is way too optimistic.

Qualcomm is heavy on sales and marketing and all of these promises have been made before on previous silicon releases. Their management hates open-source and only permits what is absolutely necessary. I am not hopeful for the future on these devices. They will be locked down and closed source in various ways that people just don't know about yet.

Qualcomm hyped these CPUs since years showing results in benchmarks. We still don't have laptops with Snapdragon X Elite but meanwhile competition is releasing improved chips, see M4.

I wonder if it won't be a little late when the laptops will be finally available to consumers.

They won't be late. There's a significant amount of people who cannot use macs for work. If Qualcomm succeeds with the x86 translation, they'll still be the only real option for that segment, even if they're behind M4.

(Also likely for people who don't have $$$ to throw around for hardware)

I think if you expect these to land in budget devices; you're going to be sorely disappointed.

I don't expect the first one to do that. But the next ones definitely. And it doesn't have to be "budget" straight away; there's a whole range of "normal" prices between that and Apple.

They're going to end up in extremely shit high end devices which will be canned after 6 months of poor sales and bad reviews.

It's always the same song and dance with Windows on ARM.

Microsoft _might_ make interesting Surface models out of it and price them in a somewhat reasonable fashion. Bad reviews will be done by people expecting x86/M4 performance with Windows and MacOS. But what is needed is non-Mac stable and open ARM Linux platform that is more powerful than a RPi5.

The chips themselves are reportedly quite cheap relative to comparable Intel CPUs.

I promise that for every one of these posts from me, I suppress the urge multiple other times. But it would be amazing to see things like this published as a Nix flake instead of some debian image chucked over the wall. Seeing the flake would be a definitive, at-a-glance, reference for what is upstream and what is being pulled from some BSP.

Hopefully there's some decently performing AMD64 emulation available on linux, and it doesn't end up exclusive to windows. It's not as necessary as it is on most other OSs, bit it'd help pull along any random bits that don't have native binaries.

It supports MTE, right? Can't wait for that to become mainstream.

This is great news for those invested in ARM CPUs. I, for one, purchased a Lenovo X13s when it first came out for around 1,900 EUR. Unfortunately, my experience with it on Windows was subpar, and it was even worse on Linux. At the same time, I bought my wife a MacBook Air (not sure if it was an M1 or M2), and it was snappy and very pleasant to use. I bought the Lenovo based on the hype from benchmarks, and it seemed like a capable system initially. However, I couldn't tolerate all its limitations and ended up returning it, as it performed no better than a Chromebook I had bought earlier.

I'm hopeful that Snapdragon will offer an alternative ARM platform for laptops that can handle more than just browsing. As consumers, we need options, and the more, the merrier. I'm still undecided about the short-term success of Snapdragon. For now, I'm betting/waiting on the MacBook Air with an M4 as my daily driver, although I do prefer the Lenovo ThinkPad format.

For me that list looks not that good compared to Asahi Linux for M1. No GPU, no camera, no suspend/resume.... I won't be spending my money on that platform until that sorted out. And judging from the other ARM platforms that means most likely never. Over-hyped and under-delivered.

Given that all those things are from the list of features they plan to ship in the next two kernel releases, "never" seems a little pessimistic. They aren't even selling laptops with this chip yet!

This is the schedule for upstreaming/"main-lining". Realistically, customers using these SoCs will be building on a forked linux BSP (Board Support Package) with all of these included.

To be fair, these might be server platforms and Chromebooks, nobody is buying ARM notebooks to run their favourite FEM package.

> Chromebooks

Qualcomm has been shipping closed-source modules for Android mobiles for years. Support of the another similar platform changes nothing. Adreno GPUs exist for many years and upstream support is deplorable. Nothing was stopping Chromebooks making ARM/Adreno support in upstream ever. Page is a marketing and not a binding. Continue to downvote all you want. Qualcomm and open source doesn't belong to the same sentence - history of the alternative ROM for Androids is a living poof.

Reply on news.ycombinator.com