Talk:Linux kernel/Archive 4

Linux Kernel Architecture
I think there should be an article on Linux Kernel Architecture, describing various parts, algorithms used and so on. There are many pages in Linux Kernel category that this page could link to. 130.119.248.52 (talk) 10:21, 12 November 2010 (UTC)


 * The great thing about Wikipedia is that as long as you have reliable references that you can cite, then you can start that article. - Ahunt (talk) 13:50, 12 November 2010 (UTC)

Version number problems
In kernel 3.0.4 a linux personality was introduced that would return version number of 2.6.40+x for all 3.x kernels, since it was found that many userspace tools have problems with the 3.x versioning scheme (they were all expecting 2.x or even 2.6.x versions). since kernel.org is down, I cant link to the changelog entry — Preceding unsigned comment added by 213.61.9.75 (talk) 12:04, 12 September 2011 (UTC)

Completely free distros
User "Ahunt" has now twice reverted this edit. Makes no sense to me. Palosirkka (talk) 10:10, 28 September 2011 (UTC)


 * Thanks for bringing this to the talk page. I delinked the words "completely free distributions" from List of Linux distributions endorsed by the Free Software Foundation because the the two subjects are not the same thing. The piping of the link in that way implies that List of Linux distributions endorsed by the Free Software Foundation = "completely free distributions". It is more than possible for someone to create a free version without FSF endorsement. By linking these two concepts it looks as if Wikipedia is endorsing the FSF as the only organization that can pronounce a distro or kernel version as "free". If you really want to include the link then this would be better worded as something like: "In response, the FSFLA started a project, Linux-libre, to create a completely free kernel without proprietary objects, which is used by some completely free distributions, such as those endorsed by the Free Software Foundation." - Ahunt (talk) 12:06, 28 September 2011 (UTC)


 * Should somebody create a distro that is completely free, it will be FSF endorsed very fast. And to many people software freedom is defined by FSF, that is the origin of the term after all. I can live with your suggestion however. Palosirkka (talk) 08:33, 30 September 2011 (UTC)


 * I agree with what you said, it is really an issue of links going where the reader expects them to and your change to the article looks fine now. - Ahunt (talk)


 * If anyone ever gets to know the proportion of Linux (in code lines or size?) that is libre, it would be nice for it to be mentioned in the article.
 * I have opened a thread for that question to be discussed (http://trisquel.info/en/forum/percentage-linux-kernel-libre), although no answers yet.
 * Twipley (talk) 16:34, 26 September 2012 (UTC)

Features list
I am really missing a complete features list do describe advanced features like hotswapping cpu(s), allowing a newer kernel to take over without restaring, the taskscheduler etc etc.... — Preceding unsigned comment added by 80.213.183.242 (talk)
 * You and me both. I revived the article Linux kernel API, there is enough space there to write some clean text on the features of the "User-kernel API" that offers that much more then other kernels. Currently (2013) the Wikipedia is still full of references to Unix-like and less to POSIX and a lot of language relating to the history, but very little regarding the current state of affairs, especially from a technological POV. This means, that by only reading the Wikipedia, a reader won't comprehend the immense differences between the Linux kernel, its features and the other kernel's features and short-comings. See e.g. the article on Kernel-based Virtual Machine, this doesn't nearly represent the features of KVM nor the notable investments flowing into its further development! Alone the state of the article is a joke. ScotXW (talk) 10:30, 6 November 2013 (UTC)

No reference to Lions'? Why?
The the Linux vs Minix debate is mentioned. Although Linux borrowed the Minix file system, there was no similarity in the code or code structure. Anyone who worked with the .9x versions like I did would appreciate that early Linux contained the same bugs as commercial kernels derived from Unix v6. As Lions' Commentary was widely available and used to teach Unix academicall it would be natural to assume Linus was heavily influenced by the code examples in Lions'. Shjacks45 (talk) 12:16, 11 March 2014 (UTC)

Article needs to explain whether support for architectures has been (or may be) dropped
The article states that version 1.2 added support for Alpha, SPARC, and MIPS architectures; version 2.2 added support for the m68k and PowerPC architectures, etc. What the article doesn't explain is, GPS Pilot (talk) 20:23, 19 November 2012 (UTC)
 * Does the latest version retain support for all of these architectures, or has support for some of them been dropped?
 * Is there a policy of forever retaining support, or might some rarely-used architectures be dropped in the future?
 * You are right. And there is the article: List of Linux supported computer architectures. Obviously it was written by some morons, since the dumb list does not distinguish between an instruction set, a microarchitecture and other stuff, that requires code to "be supported". It also didn't mention the software needed to translate source-code into machine code, e.g. GNU Compiler Collection, LLVM/Clang, etc. We should maybe


 * 1) stop calling software, that has been compiled for a specific computer architecture, a port, like Debian does it,
 * 2) transform the list into a table,
 * 3) bring it up-to-date and
 * 4) add references.
 * From the top of my head I can tell you, that the specific support for i386 was dropped in Linux kernel 3.8. Irrelevant to me, but, yes, it should be documented. User:ScotXW t@lk 20:45, 4 June 2014 (UTC)

The role of a kernel
If you read Template_talk:OS, do agree with me? In case you do, do you think that this article documents the kernel internals well enough, so that a reader would know where to look for which components, in case, he would like to inform himself about e.g. evdev or kernel components of the free and open-source graphics device drivers?
 * 1) Does some refreshable braille display require specific code in the Linux kernel, because Sony's DualShock 4 seem to do, see 3.15 mainlines support for DualShock 4 User:ScotXW t@lk 20:55, 4 June 2014 (UTC)
 * 2) Do you have other questions, that these article should answer? User:ScotXW t@lk 20:55, 4 June 2014 (UTC)

Te reduce the number of references in the article, I suggest to use   it links to https://www.kernel.org/category/releases.html as sole reference for the official long-term stables? Unofficial LTS are not listed there. User:ScotXW t@lk 22:08, 3 July 2014 (UTC)

Kernel 3.17 RC1 added
I added the new kernel 3.17 RC1 to Maintenance, so please if something is bad or wrong only fix.

Thanks. --Patrios (talk) 22:26, 16 August 2014 (UTC)Patrios

Linux kernel most popular for sure
I of course "disagree" with this revert as I put the info in. I decided against putting it in the Linux article as, for me at least, "Linux" mostly means "desktop Linux" and not Android. It would give the wrong idea that (desktop) Linux operating system(s) are the most popular in the world. Only the Android variant is it seems.

In case Android is not the most popular OS as of now by installed base; and only the most popular OS sold in the last two years, mobile OS or not, then at least the Linux kernel is because of huge numbers of routers in addition to Android. The kernel is also the most popular non-Android embedded operating system segment as a whole (possibly not in the smaller RTOS embedded space where the kernel is also used). [I had a source for that, but Android only and routers makes this rock solid.]

I agree, I could have put it elsewhere, not in the lead, but in my lazyness I didn't and then I would have put the same info in twice. I'll try again and only summarize in the lead.

The lead seems to be an appropriate place, compared to say the lead in Linux: "As of June 2013, more than 95% of the world's 500 fastest supercomputers run some variant of Linux". comp.arch (talk) 16:44, 23 October 2014 (UTC)


 * FWIW, I'd say that's the way to go – providing market share / popularity info as part of the article, and then summing it up in the lead section.  That way, lead section would get just one sentence, what would be pretty much what lead sections should do.  I'll also see to expand the lead section further, if you agree, so it sums up the whole article. &mdash; Dsimic (talk | contribs) 17:04, 23 October 2014 (UTC)


 * , I wasn't sure if from your edit summary if you disagreed on the content, so I didn't want to make the change I discussed doing above and have now done. I've been burned by going fast before and I agree with your response and if that is only issue I think we are ok with each other :) comp.arch (talk) 21:24, 23 October 2014 (UTC)


 * Yeah, we're pretty much on the same page. :) Sorry for my not-so-understandable edit summary, it's usually not that easy to squeeze a more complex thought into a single line. :)  Went ahead and  a bit your additions, please check it out –  hope you're fine with that. &mdash; Dsimic (talk | contribs) 19:54, 26 October 2014 (UTC)

Popularity is unclear
I'm starting a convo here, we'll see where it leads. The section on popularity is unclear. First in the lead:
 * The Linux kernel is the most widely used operating system kernel in the world really needs a citation. Why?

Because the popularity section doesn't support it. There are some references to sales of Android devices, but that's a long way from establishing popularity. First off, sales of new Android devices doesn't automatically mean all other existing devices wandered off into the wilderness. Secondly, according to the Android article, there is some debate if Android can even be considered Linux anymore, or some derivative. Thirdly, smart phones and routers aren't the only gadgets with operating systems. TVs, my cable modem, my car, etc.

So to be the "most widely used operating system", Linux (and it's derivative Android), would have to be in use more than combined: Windows, Mac OS, *BSD, QNX, FOS, iOS, the REAL IOS (Cisco), IOS's cousin CatOS, whatever lives inside HP printers (and for that matter Lexmark, Dell, IBM, etc), whatever is powering my Kia, whatever Nokia and Blackberry that are still in use, Foxboro IA, Modicon, Allen Bradley etc industrial automation systems, VxWorks, etc. Can that claim be accurately made? No, of course it can't. So if you want to say something like "Linux is the most widely used mobile operating system", I guess you can do that, because the references seem to support this (barely, as they cite Android sales), but please stop asserting it's the most widely used operating system, because right now, that statement is unfounded.

--Robert.Labrie (talk) 03:00, 2 January 2015 (UTC)

And phone switches and hard wired telephony systems like Nortel, Lucent, Avaya, Cisco, etc. So many operating systems... --Robert.Labrie (talk) 16:50, 2 January 2015 (UTC)


 * Hello! Went ahead and  the lead section a bit, what should be much better anyway instead of trying to hunt down some references.  By the way, old version of the lede didn't state that "Linux is the most widely used operating system", instead it said that the "Linux kernel is the most widely used operating system kernel"; thus, the "is Android Linux" debate didn't apply at all. &mdash; Dsimic (talk | contribs) 17:42, 2 January 2015 (UTC)


 * I wouldn't have put this in at the time unless I was pretty sure myself (but see new sources I just put in for Windows and Android). Information is very hard to find. What would credibly be more popular? Nobody is saying FreeBSD is more popular say. Feel free to give me other candidate OSes for laser printers (there are probably many different, what would possibly be most popular?) say or other embedded systems. There just are not that many laser printers.. compared to smartphones. I've looked into RTOSes, VxWorks at 1.5 billions is far from enough (not even beating smartphones only) to make it most popular and they say they are most popular (commercial) RTOS. comp.arch (talk) 10:54, 20 February 2015 (UTC)


 * : "would have to be in use more than combined: Windows, Mac OS, *BSD, QNX, FOS, iOS, the REAL IOS (Cisco), [..]" no - would not have to be used/installed in more than all other combined! First Windows has their own kernel. More popular means than any one other kernel. Android however can't be thought of as a separate kernel from Linux kernel (this article is about the kernel, not a Linux distro - full "operating systems"). Android has a larger installed base than Windows, maybe not higher than Windows, iOS and Mac OS X combined - yet - but soon. It already sells more than those combined per quarter for many years now. iOS and Mac OS X have kind of the same kernel and you could add the rest of FreeBSD to it. Could you claim that would be bigger than Linux kernel? FreeBSD outside of iOS would have to be much more used that iOS plus Mac OS X. Use of FreeBSD is much less than Linux in servers (and I *guess* routers), would you think Playstation would make up the difference? Ok to revert Dsimic's edit based on a misunderstanding? comp.arch (talk) 11:13, 20 February 2015 (UTC)


 * Regarding TVs, as a major consumer product that use an OS (still some old ones around not needing one..). Most probably the Linux kernel is dominating in that sector; used in all the popular brands of TV or TV-related stuff, I know of: Android TV (and former Google TV, webOS, FireTV, upcoming Tizen-based TVs from Samsung. There just isn't much money in using something other than the no-cost Linux kernel. See also: Comparison of digital media players. If you want software to run in your TV look there, Linux based. Some of that software might work in say FreeBSD (not sure if many TVs use it though), just do not see it used outside of PlayStation (know size - Xbox Windows-based about as popular) and Wii [U](?) and Apple TV. None of these matter unless FreeBSD would be hugely popular somewhere else. "the PC gaming segment is already twice the size of the console gaming market – and growing" Highest selling console: "PlayStation 2	Sony	2000	>155 million" See: list of million-selling game consoles. Xbox (all variants combined) are not that many to make Windows, that is including PCs, more popular than the Linux kernel. comp.arch (talk) 12:13, 20 February 2015 (UTC)


 * Regarding printers, it's really hard to find what OS they run, but I found a link saying Canon is the biggest manufacturer and tried to find out. At least their PIXMA printer has been hacked to run DOOM (prehacked printer has a web interface -likely Apache running on Linux?). That is a port of the Linux DOOM source code, implying the printer runs on Linux (at least after the hacking). Not sure if it's helpful to dig more into individual printers or manufacturers. comp.arch (talk) 12:44, 20 February 2015 (UTC)

Is Linux (hard) real-time? Just not by default?
"Through the use of the real-time Linux patch PREEMPT_RT, [..] Partial mainline integration of real-time Linux already brings some functionality to Linux mainline.[68] Preemption improves latency, increases responsiveness, and makes Linux more suitable for desktop and real-time applications. [..]

Additional scheduling policy known as SCHED_DEADLINE, implementing the earliest deadline first algorithm (EDF), was added to the Linux scheduler in version 3.14 of the Linux kernel mainline, released on 30 March 2014."

Doesn't SCHED_DEADLINE make Linux fully real-time? Does it need a patch (such as PREEMT_RT) that isn't mainlined? If not it seems that info is a little inconsistent. Possibly SCHED_DEADLINE needs to be enabled globally and/or at least for the programs that need real-time behavior. Not saying the article needs to explain to much (WP is not a HOWTO), still a little curious and the article shouldn't at least be misleading giving the impression that you need a patch and/or recompile. Neither should it give the impression Linux isn't real-time if not commonly enabled? Can you say in Android? comp.arch (talk) 10:47, 20 February 2015 (UTC)


 * In a few words, SCHED_DEADLINE by itself doesn't make Linux kernel even nearly real-time. &mdash; Dsimic (talk | contribs) 23:30, 20 February 2015 (UTC)


 * Let me put it this way, is it at least as good as Windows (or Symbian, that claims real-time, (or the Amiga)) out-of the box? :) It has a real-time scheduling class.. I guess you mean not hard real-time, yes (I know it requires right hardware also..). "SCHED_DEADLINE is a scheduling class for resource-reservation real-time CPU scheduler in the Linux kernel." I haven't looked to much into it, but what's the point if not to be (at least soft) real-time? What I meant, is Linux out-of-the box real-time capable (without recompiling)? In some similar sense to the old absolute priority preemptive-multitasking Amiga? Note the Amiga didn't have virtual memory. That can be a problem (OOM, not as much in Android, but can be disabled?) XFS had real-time I/O capabilities (at least on IRIX) and is now default in Red Hat.. E.g. does it have at least soft real-time? For the hardware, hard real-time is getting more difficult with caches etc. — Preceding unsigned comment added by Comp.arch (talk) 19:38, 22 February 2015‎ (UTC)


 * Well, as described here, it all depends of how hard is really hard. :) A better question is how realtime is the vanilla Linux kernel (which SCHED_DEADLINE is part of), without additional patches –  and the answer is given in the previously linked source and this one.  In a few words, to a certain extent, yes, but it can be made much more realtime by applying the PREEMPT_RT patch, which unfortunately slowly goes out of steam.  Regarding specialized hardware, there's pretty much no such thing nowadays, and that assumption dates back from the SGI and its specialized hardware that made it possible to reserve I/O bandwidth, which could be dedicated to XFS realtime subvolumes; though, without that hardware (which went belly up long time ago), those XFS realtime subvolumes do nothing special even if they work at all. :) &mdash; Dsimic (talk | contribs) 22:43, 22 February 2015 (UTC)


 * Thanks, didn't know about special SGI (disk or disk-related) hardware. XFS was made for disks ("spinning rust"), in the context of flash memory the limitations of disk and design desisions of XFS (and the general elevator algorithm) seem now invalid. With carefully chosen hardware (and a chosen made for real-time application?) and config, Linux seems not to bad, at least not worse than any other OS (at least desktop/mobile class). Maybe even better? comp.arch (talk) 10:06, 24 February 2015 (UTC)


 * Well, Linux isn't too bad in that regard, of course, but at the same time isn't good enough to be used in a control unit for car anti-lock brakes, for example. :) That's what "how hard is really hard" actually refers to. &mdash; Dsimic (talk | contribs) 23:51, 24 February 2015 (UTC)


 * Yes, I know, Linus says so :) That Linux kernel isn't most appropriate, but really he blames the hardware mostly if I recall (this was pre-PREEMPT_RT-patch if I recall, said it would get out-dated as not much point for desktop, it IS available, but picked up on SCHED_DEADLINE as standard, and not yet the differences for (not hard) real-time). It's just not his highest priority. Desktop is (and servers). I know personally how absymal the [kernel]/swapping/Firefox/Chrome can be. I blame it mostly on swapping and the browsers, but the OS should cope with bad browsers.. Already tried disabling swap. It helps, then the browsers crash sooner.. Android is probably a good alternative to traditional Linux distros.. Doesn't rely on paging/swapping (same as the Amiga). Am I not right? Where the kernel is mostly used (non-microcontrollers), I'm not sure there are any better alternatives. comp.arch (talk) 11:55, 26 February 2015 (UTC)


 * Hm, I'm not sure how much blame can be put onto the hardware itself, except when there simply isn't enough RAM for a particular use, for example. :) Other operating systems aren't much different in that regard; for example, running Firefox under Windows 7 on a slow computer with 2 GB of RAM yields quite poor results, while running it on fast hardware with lots of RAM and an SSD, under the same version of Windows, turns the whole thing into a very smooth experience.  By the way, Android actually optionally uses swap space, but not in a traditional way –  please see our zram article and this post. &mdash; Dsimic (talk | contribs) 00:50, 27 February 2015 (UTC)

Reflist column width
Now that I've completed fleshing out the 2.6 kernel series release history, I've turned my attention to the references. On my personal computer, my monitor has a horizontal resolution of 1920 pixels and with the current reflist having columns set to 30em it makes the reflist very "busy". What I'm suggesting is changing the reflist & the "further reading" section to using a column width of 40em. I know it makes these two sections a bit longer, but it makes these sections much easier to read IMO. #FF9600 talk 00:32, 4 March 2015 (UTC)
 * Hello! Well, 30em is a widely used value, so we should pretty much stick to it. &mdash; Dsimic (talk | contribs) 03:40, 4 March 2015 (UTC)
 * Thanks for the response, but I would like to point out two things. Per the columns section on the template documentation of, it says:


 * In addition, per Other stuff exists, just because other articles have their reflist set to that column width, doesn't mean that it "works" for this article. #FF9600  talk 05:06, 4 March 2015 (UTC)
 * Could you, please, clarify a bit what do you mean by the 30em width making the "References" section very "busy"? I've tried how 40em looks, and to me there isn't that much difference.  Maybe you'd like to provide two screenshots for a comparison? &mdash; Dsimic (talk | contribs) 05:39, 4 March 2015 (UTC)
 * Comparison Gallery. #FF9600  talk 14:45, 4 March 2015 (UTC)
 * This is quite strange, as your 40em screenshot looks the same as I see the "References" section in its current state, with 30em columns, though on a 1366-pixels-wide screen. Thus, I see what you're speaking about, but it seems that the columns width isn't applied everywhere in the same way.  Which browser do you use?  Could you try resizing the browser window to see the effects? &mdash; Dsimic (talk | contribs) 15:00, 4 March 2015 (UTC)
 * How column width works is that the width you specify is the minimum width that columns should use (when specifying a width opposed to a set amount of columns). So the wider your display area is, the potential for more columns. In addition, it will not add additional columns, unless all columns meet the specified requested minimum width. The default for most browsers is 1em equals 16px (12pt). This would make the minimum width at 30em is 480px wide; while 40em is 640px wide. The Mediawiki skin Vector also has 1.5em (24px) padding to the left & right of each article and the sidebar has a width of 111em (176px). So in my case, with a width of 1920px and a loss of 12.5em (200px) for the sidebar & horizontal padding of the article, I end up with 4 columns with a width of 28.875em (430px). In your case, at a width of 1366px wide, you end up with 3 columns with a width of 24.292em (388.67px). While my column width is actually wider, a column takes up 22.4% of my horizontal width; while a column takes up 28.5% of your horizontal width. #FF9600  talk 15:24, 4 March 2015 (UTC)
 * But, I see two columns in the "References" section, not three, so it seems that (according to your calculation) the available space for three 24.292em columns turns into two 24.292 × 3 / 2 = 36.438em columns. However, I agree that making the columns wider, by changing their width to 40em, would totally make the references more readable. &mdash; Dsimic (talk | contribs) 15:36, 4 March 2015 (UTC)