News for hackers. Stuff that matters.
Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by phh on Tuesday August 31, 2021 @04:27AM (#28365221)
Step -1: Forbid Linux Kernel Modules for security reasons

Step 1: Require OEM's userspace drivers to use stable ABI (Android 8.0-9.0) exposed over a bus

Step 2: Require OEMs to use your way of doing secure boot chain (Android 10)

Step 3: Require OEM's kernel-space drivers to use stable ABI, and force the usage of Linux Kernel Modules (Android 11 -- we already lost the GPL here)

Step 3b: Start implementing VM support. Find a nifty marketing reason to have VMs ("Privacy Core")

Step 4: Require OEMs to use your own signed kernel (Android 12)

Step 5: Have said micro-kernel load/exec original kernel's executables

Step 5: Say that for security's-sake, it will be better of the nifty VM if its OS is a micro-kernel

Step 6: Require OEMs that your own signed kernel also runs as hypervisor

Step 7: Expose bus defined at step 1 across the VM. Use it to move various components from original kernel to the new micro-kernel, "for security reasons"

Step 8: Have said micro-kernel load/exec original kernel's kernel-space drivers (stable thanks to step 3)

Step 9: OTA kernel defined at step 4 to new micro-kernel

Step 10: Extinguished.

Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by gman83 on Tuesday August 31, 2021 @05:41AM (#28365765)
How would Google switching their platform's kernel from Linux to Fuchsia "extinguish" Linux? I'm pretty sure we'll be fine without Google.
Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by coldtea on Tuesday August 31, 2021 @06:04AM (#28365932)
>How would Google switching their platform's kernel from Linux to Fuchsia "extinguish" Linux?

In that nobody would use it on mobile without continued development from Google (Android development can easily pause after say a 5-10 year transition period) and without Google apps and services on it.

At best, you'd have Samsung or some Chinese company keeping the flame alive for a short while.

After that everything now Android will be Fucscia based.

Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by user-the-name on Tuesday August 31, 2021 @07:29AM (#28366506)
They weren't interested in supporting Linux before Android.

They aren't going to be interested in supporting Linux after Android.

Google hasn't extinguished anything. Linux just didn't offer a compelling reason for companies to support it.

Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by coldtea on Tuesday August 31, 2021 @01:38PM (#28371059)
Your argument amounts to: "Before X happened, and after X had stopped happening, X wasn't happening".

Logically it's very strong. It's also neither here, nor there.

The "extinguish" part refers to the transition from Linux on mobile being a thing thanks to Android (on billions of devices) to Linux on mobile not being a thing anymore (or being a small niche) in a post-Android/Fucscia world.

It doesn't matter if Linux on mobile wasn't a thing before Android, the same way when defending a murder "Well, the victim wasn't alive before it was born either, judge" is not an argument.

Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by user-the-name on Tuesday August 31, 2021 @01:45PM (#28371136)
That is not at all what "embrace, extend, extinguish" refers to.

What you are describing is what happens when an inferior technology is replaced by a better one.

Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by coldtea on Tuesday August 31, 2021 @06:28PM (#28374448)
>That is not at all what "embrace, extend, extinguish" refers to.

I'll agree with that.

It still is "use X when convenient, profit off X, build your own replacement piecemeal, throw X away and let it die" though.

Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by user-the-name on Tuesday August 31, 2021 @07:25PM (#28374832)
> It still is "use X when convenient, profit off X, build your own replacement piecemeal, throw X away and let it die" though.

Pretty sure doing that falls under the essential software freedoms, though, doesn't it.

Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by coldtea on Wednesday September 01, 2021 @04:50PM (#28385362)
Doesn't EEE also fall under the essential software freedoms?
Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by fsflover on Tuesday August 31, 2021 @05:51AM (#28365841)
Google can use its monopoly position to force everyone from Android/Linux.
Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by scoopertrooper on Tuesday August 31, 2021 @06:26AM (#28366065)
Can they though? How would they even do that? The whole concept behind Android is to extend the reach of Google's advertising and service offerings into the mobile ecosystem.

This largely depends on them having the goodwill of the phone manufacturers in not de-googling their distributions. How would Google feel about Samsung switching the default search engine to Bing?

Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by fsflover on Tuesday August 31, 2021 @06:29AM (#28366076)
They can do it in the same way they did with XMPP.
Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by izacus on Tuesday August 31, 2021 @06:43AM (#28366167)
How exactly? They can't even persuade Samsung to not apply their hideous skin on Android and add their own bloatware.
Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by fsflover on Tuesday August 31, 2021 @06:59AM (#28366266)
Google couldn't care less what Samsung is doing.
Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by pjmlp on Tuesday August 31, 2021 @07:49AM (#28366689)
Why should they? They want OEM diversification of Android ecosystem.

And for what matters, non-techies love Samsung´s UI skin.

Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by user-the-name on Tuesday August 31, 2021 @05:21AM (#28365609)
Replacing an existing solution with a solution that better fits your requirements is not "Embrace, Extend, Extinguish", it's just good engineering.

Google doesn't have an obligation to use Linux. Linux doesn't offer what they need, so they stopped using it.

Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by huibf on Tuesday August 31, 2021 @05:24AM (#28365634)
Most of the people who complain about stuff like this are those who run custom, degoogled Android ROMs and think they are entitled to Google's work forever.
Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by phh on Tuesday August 31, 2021 @05:43AM (#28365774)
It's technically true that I run custom degoogled Android ROM.

However, I'm MAKING my own Android ROM. I actually do the job Google implied they would do when they announced Project Treble with Android 8.0 (which is actually providing working new Android versions on older Android vendors).

I'm guessing you don't know my project, so I'm making GSIs, which make Android's Project Treble actually usable to thousands of various devices. If you take Google's Project Treble as-is, you get /at best/ something that is not even a smartphone (you get no in-call audio, assuming you get calls at all, because it doesn't provide VoLTE support). I'm making it actually daily-drivable.

Many users are saying that I'm doing Google's job there, and even some Googlers said so. But of course, it's not in Google's nor the OEM's best interests to do that, which is way this job needs to be done by other people.

Also, I sent dozens of patches to AOSP, with very very small success rate. (While my contributions to Linux were pretty smooth)

But yes, I'm entitled to Google's work.

Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by izacus on Tuesday August 31, 2021 @06:47AM (#28366196)
> But yes, I'm entitled to Google's work.

I'm pretty sure Google will disagree on that :P

Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by phh on Tuesday August 31, 2021 @08:26AM (#28367050)
My line could have been misinterpreted, and I'm not sure how I'm supposed to interpret yours, so I'll explicit it just in case, because my previous post can be understood both ways:

No, I don't feel entitled to Google's work, the quote was sarcasm.

My point of showing my contributions is that I can do stuff on my own, that I'm happy that Android is opensource, and I do my best to contribute to it, in the opensource spirit. But if it stops being opensource, then so be it (it'd be a bigger shot in the foot for Google than for me anyway), I'm not simply leeching on Google's back, like previous comments assumed.

Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by approxim8ion on Tuesday August 31, 2021 @08:57AM (#28367443)
I'm entitled to Google's work because they made their code open source.

They did it because they wanted to be entitled to someone else's work (the free components they built it on). No one told them they couldn't be proprietary from the ground up.

Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by ducktective on Tuesday August 31, 2021 @04:56AM (#28365453)
Would Google even care about EEEing Linux? I might be naive here but I think they have technical reasons to switch to Fuchsia which is an RTOS...

Android phones don't have any semblance of FOSS/Linux/DIY sprint in them anyway...

Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by phh on Tuesday August 31, 2021 @05:19AM (#28365598)
> I might be naive here but I think they have technical reasons to switch to Fuchsia which is an RTOS...

Nowadays, Linux's realtime is pretty good. Steven Rostedt has extensive documentations on how to "debug" realtime on Linux. You know what's the number one source of non-realtime-ness in Linux? It's parts of hardware that Linux simply can't control. He found that on many PCs, the BIOS triggers SMI (to refresh the RAM IIRC), and said SMI blocks the CPU.

In case you're thinking "Surely this is an x86-only mess", well, sorry, but that's wrong. It's even worse for embedded systems with shared memory bus, where modem has higher priority over the bus, and can decides to stall any memory access made by the CPU.

Switching to Fuchsia won't change much wrt real-time. (Possibly with Fuchsia, it is harder for OEMs to footgun their drivers, which would still be an improvement)

I think the reason Android would switch to Fuchsia is simply that Google is too big to rely on the community, and most engineers there don't have an opensource mind, so overall it makes it more convenient for Google to get rid of 3rd parties. Yes, it is a technical reason.

I agree that extinguishing is not Google's goal (which is why I didn't add embrace/extend). It is just a side-effect.

Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by xorcist on Tuesday August 31, 2021 @05:53AM (#28365854)
> it is harder for OEMs to footgun their drivers,

Too little imagination applied here. They will find a way. They always do.

Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by zeusk on Tuesday August 31, 2021 @05:53AM (#28365853)
Fuchsia is definitely not an RTOS, it's an embedded system of sort. I had some experience with little kernel which served the basis for zircon, so I followed the project for a bit - they have no real time guarantees from what I could tell.
Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by saagarjha on Tuesday August 31, 2021 @06:50AM (#28366219)
Little Kernel? I mean yes, that was designed for embedded, but Google has rewritten a lot of it. What makes you think that it can’t be used for other things now?
Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by pankajdoharey on Tuesday August 31, 2021 @12:34PM (#28370146)
You gotta have RT Scheduler to begin with in order to call it RT.
Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by goodthenandnow on Thursday September 09, 2021 @10:25AM (#28470025)
They have a deadline profile available for threads, and plans to implement an actual deadline scheduler. So, this is confusing.
Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by huibf on Tuesday August 31, 2021 @05:18AM (#28365590)
You forget the part where all of this was necessary precisely because OEMs couldn't be bothered to update their drivers to work with more recent versions of the kernel.
Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by phh on Tuesday August 31, 2021 @05:31AM (#28365690)
Which kernel is latest high-end Pixel running?

Three years ago, gregkh did a presentation about LTS in Android. Guess who was using latest LTS kernels. Google? Nope. Sony and Essential are who.

Guess who upstreamed support for their phones in Linux kernel? Google? Still nope, again it's Sony.

How long was Pixel 1 supported? Ah yes, a whopping 3 years. (Same for Pixel 2 I believe). Who does better? Fairphone (circa 100 overall employees, and not many are software engineers), and nVidia.

Guess who provides major kernel upgrades? Sony (but not Google-certified sadly), and nVidia.

So far, Google hasn't shown any sign they are more capable than average OEMs to upgrade my devices.

Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by izacus on Tuesday August 31, 2021 @06:06AM (#28365943)
> Which kernel is latest high-end Pixel running?

The one Qualcomm supports in their BSP package for Pixels SoC.

You seem to not understand who exactly provides kernels for Android and other embedded devices.

Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by phh on Tuesday August 31, 2021 @08:17AM (#28366952)
> You seem to not understand who exactly provides kernels for Android and other embedded devices.

Oh, I perfectly understand how those things work, I've Google-certified more smartphones than I can count (which means more than 50), and that's merely one line of my resume.

In my comment, I mention Sony who upgraded their major kernel version. How does it fit your mental model of embedded system's BSPs?

I currently work in a company who provides software upgrades for a 10 years-old STB. This 10 years-old STB (that's older than the Nexus 4) is currently running Linux 5.4, and is planned to run Linux 5.14 in a few months, at millions of customers' home.

Are we bigger than Google? Gosh no, by orders of magnitude. Are we spending a lot of time on this? Well nope, we don't even need one person full-time to upgrade kernel version. Are we paying that person a lot? Nope, they could probably move to Google 500 meters away and have at the very least double salary.

Of course SoC-vendor stopped providing us with upgrades aeons ago.

But, contrary to Google, we actually have monetary incentives to keep supporting this device.

Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by monocasa on Tuesday August 31, 2021 @06:23AM (#28366046)
That's their point. Google, for their own phones, put next to no effort upstreaming the code, but accepted the vendor status quo.

If they can't even wrangle this for their own phones, how is going to a more permissive license and embracing proprietary modules going to make the situation better?

Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by izacus on Tuesday August 31, 2021 @06:45AM (#28366185)
I'm not sure where you got that "point". Microsoft couldn't wrangle this either and they've still ended up with a wildly successful operating system which simply provided a stable ABI.

Why are you so darn sure that they're wrong and the model that's been failing for the last 15+ years is really the right one?

Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by monocasa on Tuesday August 31, 2021 @07:01AM (#28366280)
Microsoft's kernel is also famously behind the times because they can't update key parts because of the stable API. Like how the scheduler only knows about banks of 64 cores at a time, and you have to manually load balance your threads from user space in some cases. Or how they couldn't change their VFS to compete with Linux's expectations because they reversed the caching model. The NT filesystem cache sits between user space and the filesystem so internal metadata isn't cached by default but instead needs manual work from each day driver that's always not quite implemented well in practice. On Linux the cache conceptually sits between the FS and the block device so metadata is actually cached 'for free'. Because of these problems it was faster to literally run a whole other kernel for WSL2 rather than the apples to apples of WSL1.
Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by huibf on Tuesday August 31, 2021 @06:35AM (#28366118)
What matters is having a stable API, not having the code. Nobody needs the code if there's nothing to fix.
Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by monocasa on Tuesday August 31, 2021 @06:43AM (#28366165)
And when we have an example of perfect phone APIs that don't need updates and perfectly written drivers to back those APIs then we can start to entertain that idea.
Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by rfoo on Tuesday August 31, 2021 @08:13AM (#28366905)
Stable does not mean perfect. It means putting a version number on it, don't fucking change it pointlessly, and decouple it from the other parts of the kernel.
Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by monocasa on Tuesday August 31, 2021 @12:20PM (#28369923)
Linux kernel API does have a version number on it...

And this stuff isn't like an http API. I gave examples above about how NT has fallen behind Linux because of their stable API and binary drivers.

Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by trulyrandom on Tuesday August 31, 2021 @05:46AM (#28365802)
My Pixel 5 with the latest update installed is running Linux 4.19. Pretty old, but still supported until 2024 according to kernel.org. Still, a more recent kernel release would be preferable of course.
Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by fsflover on Tuesday August 31, 2021 @05:26AM (#28365647)
One could require FLOSS drivers instead, so the community would have a possibility to update them later. This is exactly how it works on GNU/Linux phones, Librem 5 and Pinephone.
Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by izacus on Tuesday August 31, 2021 @06:46AM (#28366192)
Ah yes, just "require it". So simple. That's why we're just flooded with fully FLOSS supported hardware right?
Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by fsflover on Tuesday August 31, 2021 @06:58AM (#28366258)
Did I miss a big player requiring it? What are you talking about?
Re:RFC-0082: Runnning unmodified Linux programs on Fuchsia by monocasa on Tuesday August 31, 2021 @05:28AM (#28365667)
I'm not sure how switching to a more permissive driver model and licence where now you won't even get the source to these drivers will make the situation better.