Btrfs To Ship Multiple Performance Improvements In The Next Linux Kernel

[treba]

What world I am living in you ask?

How sweet. What about

https://w3techs.com/technologies/comparison/os-linux,os-windows

https://www.zdnet.com/article/linux-now-dominates-azure/

Don't know what happens in the Netherlands, but it certainly doesn't reflects global website shares.

[ssokolow]

It can, and does, run Linux binaries. You can even install apt on it and use Ubuntu or Debian repos.

I'm familiar with what Android is capable of. Your argument has the same flaw as claiming that iOS and MacOS are the same thing because they both use the XNU kernel... or claiming that, because WSL exists, the kernel is the only thing keeping Windows from being unarguably considered a Linux distro.

[aht0]

How sweet. What about

https://w3techs.com/technologies/comparison/os-linux,os-windows

https://www.zdnet.com/article/linux-now-dominates-azure/

Don't know what happens in the Netherlands, but it certainly doesn't reflects global website shares.

Yeah, EXACTLY what I wrote about

...
Nothing stops Linux enthusiasts here from generalizing based upon W3Techs "Comparison of the usage of Linux vs Windows for websites" and present it as a solid proof of overwhelming general Linux rule of ALL servers. When in fact it only reflects operating systems of web servers and nothing more, nothing less. And it has been done here in Phoronix countless times. Ignoring various kinds of servers and other devices sitting behind firewalls in Intranet, disconnected from Internet. The link I posted does count these normally "invisible" machines in.

One party or group commonly making one mistake has no bearing on methodological flaws or misrepresentation by another party or group. They're wrong too.

Look above? Are you claiming that guy presenting statistics covering pure web server market share as a general success-story ain't "methodological flaw" in itself?
Which is more wrong: presenting statistics of "ONE particular kind of servers (and there are ton of different kinds) as a total market share of all" or taking one pretty generic free country and extrapolating from there? How big could the differences be, going from country to country? 5%?10%? Even if it was 20% better for Linux on the link I posted earlier, results would still favor Windows Server's in general. There's just no flippin' good solid free alternative to certain Microsoft software lots of companies need - so it remains logical result. Yeah, Linux rules web servers. Does it rule every kind of server? Hell no.

You are whining about statistics from Netherlands, arguing it cannot cover the trends of whole world. Do you reasonably expect people going throughout the world from door-to-door..? It's mission impossible. Extrapolating from one free country gives you better results though than believing that web server OS market share statistics is Holy Truth for all kinds of servers, workstations and devices. If so, it's a case of believing what you want to believe - lying to yourself.

[jacob]

I'm familiar with what Android is capable of. Your argument has the same flaw as claiming that iOS and MacOS are the same thing because they both use the XNU kernel... or claiming that, because WSL exists, the kernel is the only thing keeping Windows from being unarguably considered a Linux distro.

Linux is the equivalent of XNU, not MacOS so actually you are right even if you believe that what you say is wrong. Both iOS and MacOS run XNU and by the same token, both Android and, say, Ubuntu run Linux. No-one says the two are the same platform, but the point is that Linux is not a platform. Android is, and Ubuntu is, and both use Linux in the same way and to the same extent.

[ssokolow]

Linux is the equivalent of XNU, not MacOS so actually you are right even if you believe that what you say is wrong. Both iOS and MacOS run XNU and by the same token, both Android and, say, Ubuntu run Linux. No-one says the two are the same platform, but the point is that Linux is not a platform. Android is, and Ubuntu is, and both use Linux in the same way and to the same extent.

OK, I'll be more explicit. People use "Linux" as as shorthand for "The glibc/Linux platform, and the musl/Linux platform, and the uClibc/Linux platform, and any platform which uses X11 on top of Linux as its primary interface." Android doesn't meet their definition of "Linux" because it's using Bionic and a non-X11 UI layer and, on a design philosophy level, Google treats POSIX as an implementation detail that could be factored out if inconvenient. (ie. POSIX and details of the Linux kernel like /proc aren't part of "the Android platform ABI", unlike "Linux" systems.)

Heck, that's what a lot of people believe Fuchsia to be an effort toward in the long term. A means to factor Linux out of Android to improve upgradability in the face of the hardware vendor ecosystem's behaviour... which won't change the fact that it's Android from the perspective of end-users running it as intended (locked bootloader, un-rooted, and using apps from Google Play).

Also, regarding your most recent comment:

1. I'm making a statement about established habits and, since I'm just "the messenger", trying to argue me out of it is pointless.
2. The kind of argument you're using never works anyway because it comes across as sophistry. People just get pissed off at you when you "should know what I mean" but, instead of addressing their actual point, you're just stripping away the words they use to discuss their point without actually proving it invalid.

[ssokolow]

Look above? Are you claiming that guy presenting statistics covering pure web server market share as a general success-story ain't "methodological flaw" in itself?

In the excerpt you quoted to write that reply, I said that it's also a methodological flaw.

How big could the differences be, going from country to country?

Being specific to a single country is just the most obvious concern.

You are whining about statistics from Netherlands

Please don't assume my behaviours or motivations.

Do you reasonably expect people going throughout the world from door-to-door..? It's mission impossible. Extrapolating from one free country gives you better results though than believing that web server OS market share statistics is Holy Truth for all kinds of servers, workstations and devices. If so, it's a case of believing what you want to believe - lying to yourself.

No, it's a case of recognizing that you want to take a representative sample before you generalize about the population under study. They were taking a representative sample of the Netherlands, which may be a biased sample when generalized to the world.

A list of countries is just another statistical population which you need to identify a representative sample for.

Even just splitting it between the U.S. and the Netherlands while keeping the number of respondents the same may produce more accurate results.

Anyway, I never said I believed Linux has won out in that sphere. I just said that, with such vague methodology in the article and such sloppy generalization of it, I don't trust any conclusions drawn from it because who knows what assumptions may be lurking.

For a concrete example, they don't say what they've done to control for the "Linux has no license fees, so nobody bothered to talk to management before setting up a server for their project" effect and, given that the thrust of the article is one Windows version vs. another, I'm worried they may not have done anything.

Another example: They don't say what instructions they gave to respondents regarding how to count on-premises vs. co-located vs. cloud. For all I know, that survey could be only counting on-premises servers, which would naturally skew heavily in favour of Windows, since things like Active Directory servers and CIFS file servers are the classic example of something you can't co-locate or outsource to Amazon or Google's cloud.

I can easily imagine a bunch of these little things adding up to skew Linux to be 51%+ of the total, so it would be irresponsible to generalize in the way you're asking.

Heck, regarding that "does management know?" aspect, I have a recent example:

Large companies have no clue what their employees are doing. I discussed this flaw a CISO of a giant payment processing company. He assured me this wouldn’t be a problem, as their employees weren’t supposed to communicate through Slack. They had their own intranet set up to handle these things. I proved him wrong by joining 8 rogue Slack channels actively used by 332 employees all around the globe. I ended up getting a $5,000 bounty for it.

[jacob]

OK, I'll be more explicit. People use "Linux" as as shorthand for "The glibc/Linux platform, and the musl/Linux platform, and the uClibc/Linux platform, and any platform which uses X11 on top of Linux as its primary interface." Android doesn't meet their definition of "Linux" because it's using Bionic and a non-X11 UI layer and, on a design philosophy level, Google treats POSIX as an implementation detail that could be factored out if inconvenient. (ie. POSIX and details of the Linux kernel like /proc aren't part of "the Android platform ABI", unlike "Linux" systems.)

Also, regarding your most recent comment:

1. I'm making a statement about established habits and, since I'm just "the messenger", trying to argue me out of it is pointless.
2. The kind of argument you're using never works anyway because it comes across as sophistry. People just get pissed off at you when you "should know what I mean" but, instead of addressing their actual point, you're just stripping away the words they use to discuss their point without actually proving it invalid.

I don't know which "people" use Linux as a shorthand for a purported "glibc/Linux platform", in fact as far as I can tell no-one even uses such a term. Richard Stallman always repeats that "Linux" is a kernel, and he insists on calling one particular use of that kernel "GNU/Linux". Linus Torvalds insists that Linux is a kernel, and is on record saying that his best hope for a "Linux desktop" is.... ChromeOS. Debian defines itself as a "free software operating system" and doesn't even use the word Linux at all. Ubuntu states that is "an open source software operating system". Etc. Fedora and Mageia, by the way, don't even use X11 (not by default, at any rate) and my own laptop, although it's presently running Ubuntu, doesn't use X11 either except through Xwayland when needed - which is the same as running X11 on Android. Musl or uClibc-based distros, to the extent they have any statistical relevance at all, are as far removed from a glibc-based one as Bionic.

I short you seem to have your own definition of what "Linux" means, but that is only your own, it's not the definition of the developers of Linux, or the developers of the GNU environment, nor the distributors and, simply by observation, it's also not something implicitly agreed on within the community. You claim that Android is not part of some alleged "Linux platform" but yes, I will state that it's an invalid point, because no such platform exists; there is a myriad of loosely related but variously different platforms that share come components. Many share the Linux kernel (this includes Android, ChromeOS and various special distros), many share the kernel plus other parts of their stacks, and some share most components but NOT the kernel (kFreeBSD, WSL etc.). But there is fundamentally no more difference between Android and <your favorite distro> than between one distro that uses Wayland, KDE, RPM and a kernel with voluntary preemption, and one that uses GNOME, X11, deb/snap and a kernel with mandatory preemption. So to get back to the original point about what it means to "run Linux", it really means the same thing as "to run GNOME" or even "to run the minesweeper": simply to run any platform where Linux happens to be part of the OS.

Incidentally, this also brings us back to the actual discussion, which is about Linux (as in, kernel), Btrfs (still the kernel!) and their use on the market. Nothing there was related to any libc implementation and since the OP's point I was responding to was about servers, X11, Wayland or any other graphical stack doesn't make any difference there either.

[ssokolow]

I short you seem to have your own definition of what "Linux" means, but that is only your own, it's not the definition of the developers of Linux, or the developers of the GNU environment, nor the distributors and, simply by observation, it's also not something implicitly agreed on within the community. You claim that Android is not part of some alleged "Linux platform" but yes, I will state that it's an invalid point, because no such platform exists; there is a myriad of loosely related but variously different platforms that share come components. Many share the Linux kernel (this includes Android, ChromeOS and various special distros), many share the kernel plus other parts of their stacks, and some share most components but NOT the kernel (kFreeBSD, WSL etc.). But there is fundamentally no more difference between Android and <your favorite distro> than between one distro that uses Wayland, KDE, RPM and a kernel with voluntary preemption, and one that uses GNOME, X11, deb/snap and a kernel with mandatory preemption. So to get back to the original point about what it means to "run Linux", it really means the same thing as "to run GNOME" or even "to run the minesweeper": simply to run any platform where Linux happens to be part of the OS.

Again, you're arguing against the term, rather than trying to understand the concept that the other party has applied it to, which only pisses people off because it comes across as "wasting their time arguing semantics" when they want to discuss whatever underlying concept they borrowed it as a concise label for. (And, as I said, I'm just talking about what I've observed others doing.)

You're very unlikely to reach a common understanding by doing that.

Different groups use "Linux" to mean different things.

Enthusiasts who game tend to define "Linux" as "X11/glibc/Linux/x86" because they define it based on what characteristics a distro must have for their "Linux" Steam or GOG.com games to natively run.

People who work in the embedded space tend to define "Linux" something like "Platforms based on a Linux kernel, with a primary API that's a superset of POSIX (this distinguishes it from "Android") and an ecosystem of source packages they associate with Linux, like busybox/toybox and lighttpd."

People who want a desktop/workstation but aren't wedded to closed-source software tend to define it something like "Any distro/arch combination that combines the Linux kernel with support for X11/Wayland-based DEs like KDE, GNOME, Xfce, etc." (This is how they use it for my OpenPandora handheld, which can run either an Ångstrom-based distro with Xfce for the default desktop or Android.)

etc. etc. etc.

Heck, when a lot of people say "Android" what they're actually intuitively saying is "Android on ARM" because they're conceptualizing it based on software compatibility and Android/x86 wouldn't run their NDK-based software.

Incidentally, this also brings us back to the actual discussion, which is about Linux (as in, kernel), Btrfs (still the kernel!) and their use on the market. Nothing there was related to any libc implementation and since the OP's point I was responding to was about servers, X11, Wayland or any other graphical stack doesn't make any difference there either.

This thread of discussion began with your response to aht0's "bah.. 75% mobiles run Android, not Linux." I've been explaining aht0's perspective and why you're not likely to get a satisfactory outcome unless you understand it and at least acknowledge the underlying intent.

(In short, aht0 was saying "bah... 75% of mobiles run Android. The availability of the ecosystem I associate with Linux is comparable to what Wine brings on x86 distros and is not an officially supported option. Linux is a mere implementation detail.")

[jacob]

Again, you're arguing against the term, rather than trying to understand the concept that the other party has applied it to, which only pisses people off because it comes across as "wasting their time arguing semantics" when they want to discuss whatever underlying concept they borrowed it as a concise label for. (And, as I said, I'm just talking about what I've observed others doing.)

You're very unlikely to reach a common understanding by doing that.

My point is that the concept is bogus. Let's recap, shall we. The original argument was whether Linux (talking about kernel development) is "successful" and whether Android is a proof of that "success". An objection was that no, because it supposedly can't run Linux binaries.

I pointed out that it can.

Then the argument suddenly became all right, but "Linux" actually means glibc (hence abandoning all relevance to the subject of the success of the Linux kernel) or any other obscure POSIX C library except Bionic, plus X11 graphics. That is despite the fact that Bionic is also an open-source, (mostly) POSIX C library and is no different in that regard from other "alternative" implementations such as musl & others.

I pointed that that was bogus, among other reasons because several systems that no-one would deny are "Linux" don't in fact use X11 any more.

So you extended your definition of "Linux" to anything that has any libc other than Bionic, and X11 OR Wayland.

You are quite right that Linux is used, among other things, for embedded systems. The funny thing is that, leaving aside the fact that some embedded applications are actually Android, there is a number of embedded "Linux" implementations that don't use X11 or Wayland, but Mir. On the other hand, in Amibian, I don't know for sure which underlying display server it uses but that's irrelevant, because the UI the user actually interacts with is Amiga's Workbench... So are we going to change the definition again to say that Linux means Linux kernel plus any combination of libraries provided that it's not Bionic, and any graphics stack or even no graphics at all, but not Surface Flinger? Why not specifically Surface Flinger? Because we say so?

In conclusion: yes, I argue against the term AND the concept, because the concept a) appears purely arbitrary and not rooted in any technical reality; b) is contrary to the documented intent of Linux developers, and GNU developers, and even Android developers; and c.) it doesn't seem to be possible to define it in any other way than "anything and everything except Android, just because."

Different groups use "Linux" to mean different things.

Precisely, and the only unifying factor among all those different things is the Linux kernel itself. So can we agree now that all these applications reflect somehow the success of Linux in the broad software world?

Enthusiasts who game tend to define "Linux" as "X11/glibc/Linux/x86" because they define it based on what characteristics a distro must have for their "Linux" Steam or GOG.com games to natively run.

Some may use that definition but that doesn't mean it's not bogus. According to it, using Raspbian, or Debian, or Ubuntu, or Arch etc... on a Raspberry Pi is thus somehow not Linux.

People who work in the embedded space tend to define "Linux" something like "Platforms based on a Linux kernel, with a primary API that's a superset of POSIX (this distinguishes it from "Android") and an ecosystem of source packages they associate with Linux, like busybox/toybox and lighttpd."

Android does have a POSIX API. Ok, Bionic is not 100% POSIX compliant yet, but it's its stated goal.
However once again, we are talking about the kernel. POSIX means nothing with regards to the kernel. So POSIX is absolutely irrelevant to the discussion at hand but if you insist that talking about Linux must involve POSIX for whatever reason, does a Linux-based platform whose developer states that POSIX compliance is a chief goal count, or not?

Same thing for busybox. The piece of software you obtain when you download Linux from kernel.org has no relationship at all with busybox. You can use busybox on Windows, that doesn't make it Linux, does it? On the other hand, it is very common to use busybox on Android. Much more than, say, on Ubuntu, Fedora or Debian.

People who want a desktop/workstation but aren't wedded to closed-source software tend to define it something like "Any distro/arch combination that combines the Linux kernel with support for X11/Wayland-based DEs like KDE, GNOME, Xfce, etc." (This is how they use it for my OpenPandora handheld, which can run either an Ångstrom-based distro with Xfce for the default desktop or Android.)

Well people who want a desktop workstation are not going to use a system designed for mobile devices, that much is clear. And vice versa. But they are not going to use Ubuntu Server or RHEL either. They are also not going to use a container-oriented system like Clear Linux. Meanwhile, people building a supercomputer with 8K cores are not going to run Xubuntu or KDE Neon on it. So at what point exactly does it cease to be "Linux"?

Heck, when a lot of people say "Android" what they're actually intuitively saying is "Android on ARM" because they're conceptualizing it based on software compatibility and Android/x86 wouldn't run their NDK-based software.

Yes. So that is relevant as far as the popularity of ARM vs x86 goes. On the other hand, in both cases they run the Linux kernel, so whether Android is more popular on ARM or x86, it means nothing for the popularity of Linux.

Now iPhones also use ARM processors, so there we can talk about the relative popularity of Linux vs Darwin. FIne?

[ssokolow]

My point is that the concept is bogus. Let's recap, shall we. The original argument was whether Linux (talking about kernel development) is "successful" and whether Android is a proof of that "success". An objection was that no, because it supposedly can't run Linux binaries.

I pointed out that it can.

Then the argument suddenly became all right, but "Linux" actually means glibc (hence abandoning all relevance to the subject of the success of the Linux kernel) or any other obscure POSIX C library except Bionic, plus X11 graphics. That is despite the fact that Bionic is also an open-source, (mostly) POSIX C library and is no different in that regard from other "alternative" implementations such as musl & others.

And, all along, my point has been that, if you dismiss the concept as bogus without trying to reach a common understanding, you'll just be wasting your own time by interacting with people who see your perspective as an attempt to dismiss their concerns without addressing them using cheap terminological trickery.

If you really want to argue the "Linux refers to the kernel and Android uses Linux" point you're making, rather than just accepting a statement like "Android isn't Linux" for the sake of moving the discussion forward, you have to start with something like "What about Android makes it not Linux?" so you're acknowledging their point and trying to find a common understanding to argue from.

(That's also more likely to help you convince them of your point because you can follow up with thought-provoking questions like "If Android isn't Linux, what about something like Alpine Linux which also doesn't use glibc?". The best way to change people's minds is to let them reach the conclusion themselves.)

I pointed that that was bogus, among other reasons because several systems that no-one would deny are "Linux" don't in fact use X11 any more.

So you extended your definition of "Linux" to anything that has any libc other than Bionic, and X11 OR Wayland.

No, I clarified that people who use "Linux" to mean more than the kernel (again, it's not my definition) don't all use "Linux" consistently, but that the ways they do use it tend to correlate to what group they are a member of.

(ie. What is and isn't "Linux" is defined in large part by what the individual needs it to do and what they need to contrast it with.)

That said, I will admit that I'm sometimes guilty of saying "Linux distros" with an implicit "except for Android" (when my meaning is clear in context) for the same reasons that terms like "atheist" and "NoSQL" come around. (Android, or belief in God/gods, or SQL-based storage systems have such an overwhelming presence in their respective spheres of discourse that it's useful to omit them and then talk about the entirety of the Venn diagram that's left behind.)

You are quite right that Linux is used, among other things, for embedded systems. The funny thing is that, leaving aside the fact that some embedded applications are actually Android, there is a number of embedded "Linux" implementations that don't use X11 or Wayland, but Mir.

How's that at all odd? I specifically said in my last post that people in the embedded space tend to use a definition of "Linux" that is more based around Linux's superset of POSIX being the primary supported API for userland development rather than the Android API.

Heck, we both neglected to mention that, in the embedded space, it's common for applications to be written using Qt's support for speaking directly to APIs like the Linux framebuffer or EGL without a window server or compositor in between.

On the other hand, in Amibian, I don't know for sure which underlying display server it uses but that's irrelevant, because the UI the user actually interacts with is Amiga's Workbench... So are we going to change the definition again to say that Linux means Linux kernel plus any combination of libraries provided that it's not Bionic, and any graphics stack or even no graphics at all, but not Surface Flinger? Why not specifically Surface Flinger? Because we say so?

I haven't looked at Amibian, but from your description of it, I can't even guess how others would refer to it. I do know that everybody I've met who knows of Debian GNU/kFreeBSD considers it to be too nuanced a thing to be referred to as either "Linux" (because it lacks the Linux kernel) or "FreeBSD" (because it lacks the FreeBSD userland).

I'm guessing that something similar would be at play with Amibian, with it not falling under the heading of "Linux" or "Amiga".

In conclusion: yes, I argue against the term AND the concept, because the concept a) appears purely arbitrary and not rooted in any technical reality; b) is contrary to the documented intent of Linux developers, and GNU developers, and even Android developers; and c.) it doesn't seem to be possible to define it in any other way than "anything and everything except Android, just because."

I think we've been talking at cross-purposes... ironically, for reasons similar to why I came into the conversation in the first place.

My point has always been that, for lack of specific terminology, various people use "Linux" to refer to various groupings of systems, the bounds of which they believe will be obvious from context.

It seems to work for many people every day, so, if you try to argue them out of it directly, they'll push back because, from their perspective, the problem is minor enough that you must be a pedant who possibly has too much time on his hands.

Hence my earlier comment that you have to acknowledge their use of the term before you can argue against it. If they believe it's a case of miscommunication or deflection, you'll never get anywhere. Conversely, if they believe that you understand their intent and want an honest discussion over a point of disagreement, then you can actually start to discuss things productively.

Precisely, and the only unifying factor among all those different things is the Linux kernel itself. So can we agree now that all these applications reflect somehow the success of Linux in the broad software world?

The success of the Linux kernel was never in question... however, I will agree with what aht0 seemed to be pushing at. To use my own terminology, "Who cares about Linux? Depending on the discussion in question, what I care about is the success of either Linux-based POSIX-centric OSes (eg. my OpenPandora handheld) or OSes that meet the Linux+glibc+x86 combination needed to run GOG.com games, and Android's official APIs are neither POSIX-centric nor x86+glibc." (People have tried for years to come up with a term that encapsulates what I care about in the broader case, but UNIX describes a certification an Linux iterates rapidly enough to make that cost-prohibitive, Unixy is an adjective, and things like Lunix sound too adolescent to catch on.)

Some may use that definition but that doesn't mean it's not bogus. According to it, using Raspbian, or Debian, or Ubuntu, or Arch etc... on a Raspberry Pi is thus somehow not Linux.

Again, different groups use "Linux" to mean different things. To the people who use the Raspberry Pi, of course those distros are "Linux" because "it'd be silly" to stop calling it Linux just because the CPU architecture changed and, besides, they need something to collectively refer to Linux distros in contrast to other things you can boot on your Pi, like RISC OS. To people who are talking about game compatibility, they'll say something like "Linux on x86" rather than "Linux" once it becomes clear that other CPU architectures have entered the discussion.

Android does have a POSIX API. Ok, Bionic is not 100% POSIX compliant yet, but it's its stated goal.
However once again, we are talking about the kernel. POSIX means nothing with regards to the kernel. So POSIX is absolutely irrelevant to the discussion at hand but if you insist that talking about Linux must involve POSIX for whatever reason, does a Linux-based platform whose developer states that POSIX compliance is a chief goal count, or not?

That's why I say that, often, people are disqualifying Android for not having its primary API be POSIX-centric. They perceive Android as an ecosystem that exists on top of the Linux kernel but cares about as much about it as Windows applications running on Wine while whatever they're calling "Linux" is much more closely wedded to the POSIX API and other trappings of Unixy OSes.

Same thing for busybox. The piece of software you obtain when you download Linux from kernel.org has no relationship at all with busybox. You can use busybox on Windows, that doesn't make it Linux, does it? On the other hand, it is very common to use busybox on Android. Much more than, say, on Ubuntu, Fedora or Debian.

I'm not even sure why you're talking about busybox here. Because I gave it as an example of one of characteristics embedded developers associate with a "Linux" platform?

Well people who want a desktop workstation are not going to use a system designed for mobile devices, that much is clear. And vice versa. But they are not going to use Ubuntu Server or RHEL either. They are also not going to use a container-oriented system like Clear Linux. Meanwhile, people building a supercomputer with 8K cores are not going to run Xubuntu or KDE Neon on it. So at what point exactly does it cease to be "Linux"?

Again, different groups define "Linux" differently because it's just a convenient shorthand, often defined more by what it's not than what it is. (eg. embedded developers talking about Linux rather than Android and meaning "let's explore alternative options like Wayland, Mir, and tying Qt directly to EGL". Hence my comparison to words like "atheist" and "NoSQL")

Yes. So that is relevant as far as the popularity of ARM vs x86 goes. On the other hand, in both cases they run the Linux kernel, so whether Android is more popular on ARM or x86, it means nothing for the popularity of Linux.

Now iPhones also use ARM processors, so there we can talk about the relative popularity of Linux vs Darwin. FIne?

No, because it was never about the kernel. It was about the platform and people who contrast Linux with Android see "Linux" as referring to something (perhaps a nebulous ecosystem in potentia) that would be built around the Linux kernel's POSIX APIs, expose something like Wayland or Mir as an officially supported API, and not treat non-Dalvik graphical applications as second-class citizens.