Slashdot Mirror
Linux 4.17 Kernel Offers Better Intel Power-Savings While Dropping Old CPUs (phoronix.com)
An anonymous reader writes: Linus Torvalds has released Linux 4.17-rc1. This kernel comes with a significant amount of new capabilities as outlined by the Linux 4.17 feature overview. Among the new features are AMDGPU WattMan support, Intel HDCP support, Vega 12 GPU enablement, NVIDIA Xavier SoC support, removal of obsolete CPU architectures, and even better support for the original Macintosh PowerBook 100 series. Phoronix testing has also revealed measurable power savings improvements and better power efficiency on Intel hardware. The kernel is expected to be stabilized by June.
These aren't exactly mainline architectures they are dropping and they are being dropped due to the current lack of active use.
They were dropped because there are no developers to maintain the code, If you want to pick up the slack, I'm sure the community will be more then willing to rummage through their junk-piles to send you testing hardware
I'm curious what's considered an "obsolete CPU architecture" if a Powerbook 100 is still supported.
things like Power4, Blackfin, M32R etc.
I'm curious what's considered an "obsolete CPU architecture" if a Powerbook 100 is still supported.
Obsolete = Nobody is stepping up to maintain the code.
Obviously someone still cares about the Powerbook 100 enough to do the maintenance work.
Proprietary software drops support when they no longer care. Open source drops support when you no longer care.
I tried that, but it broke most of the features in ettercap, which is fundamental to my work flow.
Kernel 2.6 works just fine thank you very much sincerely my bosses sticking with Redhat 5 and 6.
http://saveie6.com/
Anything before Skylake. Oh wait nevermind I thought we were talking about Windows 10.
http://saveie6.com/
Things like Unicore, Hexagon, S+core, OpenRISC, M32R, Cris i.e. stuff most people didn't even heard about.
The long version at (as always) excellent LWN:
https://lwn.net/Articles/74807... and
https://lwn.net/Articles/74929...
:wq
Of course, in the case of the processors being dropped, you just about have to be a developer yourself to have any interest in continued Linux support. It's not like you could ever just put a DVD in a drive and install Linux on anything that has one of those processors.
So if there is anyone out there who wishes support would continue, they weren't interested enough to actually do anything about it.
What kind of mental gymnastics do you have to go through to keep 68000 but drop the much newer Blackfin DSP
No mental gymnastics required. You just need a developer willing to support the 68k, and not have a developer for the Blackfin.
The article summary didn't do it justice and implied they might be dropping out some form of support for older chips regardless of architecture such as older X86 generations ..... That's not the case.
The 68k is kind of baseline for that tree, and relatively simple even for people without specialized knowledge, so it may stick around for a while. It's harder to maintain more complex architectures where you need instruction order and cache management to think about as well.
If builders built buildings the way programmers wrote programs, then the first woodpecker would destroy civilization.
If that's the case, just say you're dropping unmaintained architectures due to lack of interest or active maintainers. Talking about dropping "obsolete" architectures is somewhat misleading when the architectures being dropped are still in being manufactured or developed, but undeniably obsolete architectures are still being actively supported.
Win 10 and any post 7 is an inferior experience.
If builders built buildings the way programmers wrote programs, then the first woodpecker would destroy civilization.
Typically they'll leave old hardware in of anyone says they are USING it. It won't necessarily see new development. I've seen several lkml posts asking "is anyone using _____? If not, we'll remove it."
Typically old hardware isn't dropped if we know someone is using it. Of course there are exceptions when a stack gets a major rewrite. If there is discussion about dropping support for something you use, just let us know on lkml.
NTFS doesn't seem that stable on Windows 10. On Linux, when a filesystem error occurs, the filesystem is set as read-only until reboot, which prevents further errors, and really encourages people to fix the problem. On Windows 10, when a filesystem error occurs, a directory may give a "Permission Denied" status message when you try to read a file, rename it, etc., but otherwise the system runs as usual. This means errors go unreported, and the system just feels vaguely untrustworthy. Often the directory in question doesn't even exist after a reboot, but the system just says "Permission Denied" after every operation. (This is with Administrator permissions.)
I've had to reboot at least three times in the past week because various operations required file deletion and Windows wasn't cooperating. That's not to mention the common obvious problems, like Windows and other Microsoft software filling up the C: drive instead of the location it was installed.
So I wouldn't say the Windows distro is smooth sailing.
A cat can't teach a dog to bark.
I'm curious what's considered an "obsolete CPU architecture" if a Powerbook 100 is still supported.
The full list of dropped CPUs is Blackfin, CRIS, FRV, M32R, Metag, MN10300, Score, and Tile. Also under consideration are Unicore32 and Hexagon, but they are not officially gone yet. Apparently this change removes about half a million lines of code, a substantial reduction in complexity. I had never heard of any of these before and I suspect most other people haven't either, so I don't think they will be missed.
Actually, how does Linux support the 68K CPU? It doesn't have an MMU that's required. I know there was a version of Linux called uCLinux that worked on MMU-less (and MMU-lite) processors like the 68K, but I've found it quite unstable. No hardware support for protection means a bad pointer literally will take down the system.
So it was neat, but completely pointless in the end. All it took was one bus error and you'd be rebooting.
Now, there were later 68K CPUs with MMUs (I believe the 68020 had an external MMU chip, but the 68030 was fully capable of Linux). Heck, there were a couple of bad instructions in the 68k that made a "modern" OS impossible (i.e., you can call them from user mode) - that's why it was revised into the 68010 to allow proper user-supervisor mode separation.
I had never heard of any of these before and I suspect most other people haven't either, so I don't think they will be missed.
Blackfin I've heard of, but don't know what it is. Tile is the instruction set used by Tilera's parallel CPU, which had 64 cores on a chip about 10 years ago. Looks like they switched the thing over to use ARM cores after the first couple of generations.
If you're an audio engineer, the Blackfin is used by the legendary Bricasti Model 7 (actually, the M7 uses six Blackfin processors). It's also used in devices like the DR-70D. I'm actually surprised they are dropping support for it.
Just FYI, Mikrotik's latest Cloud Core Routers use the Tile CPUs. I sure hope this mean's Tile isn't the end of the line for Mikrotik!
I suppose if they'd said "unmaintained" instead of "obsolete" it would have been a tautology. Perhaps "unpopular"?
Confucius say, "Find worm in apple - bad. Find half a worm - worse."
Were likely feature-complete anyways.
None of them supported USB or later versions of USB, PCIe, etc. Meaning none of them have been gaining anything other than bloat from newer iterations of the kernel. Furthermore some of them may have been broken thanks to the minimum kernel size exceeding the maximum kernel image the bootloader could load (sparc ipx was limited to 1 meg maximum kernel size, which was impossible with later 2.4/2.6 kernels and the minimum features for bringup, notably ipv4 support.)
While I am disappointed in linux for doing this, the real solution is breaking out the arch-specific kernel code and possibly some of the bus code, then making the driver packages rely on defined interfaces to the kernel packages. Doing so would both make the arch specific kernels easier to maintain, as well as making the driver easier to maintain and verify against specific breakage of the kernel code, which has been happening consistently for 20+ years now. Almost no arches other than x86 have consistently been bug-free out of the box in a real kernel release in that long. Sparc, Alpha, Power, etc haven't been that reliable and anything embedded has usually been considered horribly broken. Google for old mailing list entries if you doubt me.
code may not change, but new bugs and vulnerabilities are always discovered, if you don't have people actively supporting that code base then it is both a stability and security risk as well as pointless bloat which there is already enough of.
Soo it is yet not bug fixed. it become stable till june so is it under developer option.
@galaxyunlocks
... these days. Even before flakey audio and bluetooth management.
Seriously. Every other OS get's an easy 8 hours out of todays regular portable hardware, only Linux barely scrapes 4 hours. I'm a big Linux and FOSS fan but this is a problem that is really annoying and needs fixing ASAP. Windows, macOS and even Chrome have been on top of this for the better part of a decade and the Linux kernel still wastes gobs of energy. Unacceptable. This update is a step in the right direction. I hope it continues that way. Thumbs up for the kernel crew.
We suffer more in our imagination than in reality. - Seneca
things like Power4, Blackfin, M32R etc.
Which are all newer than the motorola 68000 in the powerbook 100.
"Newer" has nothing to do with "less obsolete". There is a link in TFA, describing eight dropped architectures as "without active users". POWER4 was dropped because POWER4 support is broken since two years already and no fix or rework available (and apparently no urge to fix it).
So yes, a newer architecture can be obsolete when an older one is still alive and kicking.
Seriously. Every other OS get's an easy 8 hours out of todays regular portable hardware, only Linux barely scrapes 4 hours... Windows, macOS and even Chrome have been on top of this for the better part of a decade and the Linux kernel still wastes gobs of energy.
you do know that chromeOS uses the linux kernel?
On a long enough timeline, the survival rate for everyone drops to zero.
Albeit no doubt modified. However those modifications should have been pushed upstream to the main line.
The power wastage in desktop systems probably comes from the bloated GUIs that tend to come as the default on linux these days (hello Gnome and KDE) which seem to have their processes permantly at the top of the "top" cpu usage list if my system is anything to go by.
fwiw a lot of that can be fixed by turning down the brightness on your laptop screen. The backlight draws a lot of power, and Apple aggressively dims the screen.
"First they came for the slanderers and i said nothing."
Heck, there were a couple of bad instructions in the 68k that made a "modern" OS impossible (i.e., you can call them from user mode) - that's why it was revised into the 68010 to allow proper user-supervisor mode separation.
I think this is what you are referring to: https://en.wikipedia.org/wiki/...
Next time you see a long word, go upstairs and ask your mom what it means.
Confucius say, "Find worm in apple - bad. Find half a worm - worse."
It says "100 series" in the summary but I was curious and followed the trail on Phoronix.
The mention of a specific model is the Powerbook 190, produced several years later - whose specs on wikipedia have a 68LC040 (no FPU) and up to a whopping 40GB of RAM.
Whether that's backward compatible with other models is anyone's guess.
Instead of rebooting, open Process Explorer (https://docs.microsoft.com/en-us/sysinternals/downloads/process-explorer)
hit ctrl+f and type in the file name or folder name
click a result, it'll highlight a file handle in the main window
right-click on it and select close handle
To much abuse, ridicule and disbelief, I’ve posted here many times about my problems with Linux on the desktop...
This hasn’t been one of them. I have an Acer Cloudbook which gets 12-17h per charge. It can’t hibernate properly, crashes on resume from suspend, bit the fact that it gets such wicked battery life while ON, means I just leave it running in my bag all day.
Although my audio stopped working, no fricking clue why. Who has time to deal with this stuff?
Really? Wow, so it IS an advanced system that can do things it wasn't even designed for.
We used to have a Bill of Rights. Now, with the rights gone, all we have left is the bill.
How is it a problem?
That is not what obsolete means. That is what un-maintained means. If I am the only person using and maintaining the source for hardware I developed my self, then it would perhaps still be maintained, even if it is obsolete.
If nobody maintains the source even when there are millions of users, then it would be dropped. Hardly obsolete.
Yes, very often obsolete hardware will become un-maintained, as there are not any developers that have the hardware or the will to maintain it.
Obsolete =/= un-maintained.
Don't fight for your country, if your country does not fight for you.
Windows 10 is not designed to provide an inferior experience
So, this was just a happy accident?
This sig left unintentionally blank.
You're right about the stability; that's just the nature of the beast. When they refer to supporting the 68000, they're probably using a NOMMU system which is a quirky thing but within its several limits it DOES work. As you noted, some features of modern CPUs and MMUs such as memory protection are not available. uClinux's kernel work was folded into the Kernel during the 2.5.x series. As of 2016 they were still alive and kicking.
Getting modded down for posting facts is shit.
If they want to speak some other language, fine. But if they want to claim to be speaking English, then they ought to use the word "obsolete" to mean what it actually means, and not something else when there is a perfectly useful word that means what they are trying to say.
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
junk pile of hardware? that would be so nice, I currently mostly need an AAUI adapter (Apple AUI, Ethernet) https://www.youtube.com/watch?... If you find other goodies, let me know. Other most wanted is some Sgi Octane HD tray, or this Mac 8100/80 HD and CD-ROM tray: https://rene.rebe.de/wishlist/
Blackfin was discussed. It's not being removed at this time.
A manufacturer using Blackfin, or someone who wants to keep support in newer kernels, will need to step up, though, or it will be removed soon.
Go to Reddit, shill; your efforts are guaranteed to be more successful there.
Nice!
Still, there are processors which were introduced late and are out of production now, while older designs are still made and in use. Thus my point still stands, despite your nitpicking (and remember, not everything that gets obsolete was produced).
I *might* have one. I'll have to look to see if I still have it or if I recycled it last purge.
Dragonball is still in use and is 68K based. It is also more likely to run a Linux OS than the Blackfin or M32R which usually run an RTOS.
Also, a lot of enthusiast systems run 68K CPU's, but I know of no enthusiast systems at all based around Blackfin or M32R. And since enthusiasts are likely to be contributing to Linux, support works itself out. Therefore support for some obsolete CPU's is dropped, while support for other obsolete CPU's remain.
No mental gymnastics required.
It's also used in devices like the DR-70D [tascam.com].
Is that running Linux, though? I was under the impression that Tascam used a "true" embedded system without kernel/userspace separation.
The 68000 have no MMU internally and no support for virtual memory even with an external MMU, the "bad" instructions just exposed system state making pure virtualization impossible. But virtualization isn't a requirement for a modern OS and x86 fixed their abstraction leakages much later (20+ years) and in a much more complex way (but with advantages).
The 68010 was hardware compatible with the 68000 but fixed both the virtualization problem and the MMU problem (the MMU itself had to be added externally).
And the CPU's are no longer used in the context of Linux, thus obsolete in the context of Linux.
English takes context into account. You should too.
Which is quite interesting, given my experience with a Thinkpad Carbon X1 4th gen. I get around 10 hours on that running Windows, and can squeeze that to almost an extra hour if I really push hard on running nothing but Scrivener - but that requires Windows Update to behave, which it usually doesn't, pushing battery life below 9 hours.
Running Linux, with Scrivener in Wine, I get 12-14 hours on it without even trying.
In both cases WiFi is on, screen brightness at near minimum, and I'm using it pretty much constantly.
So now I have removed Windows.
Ok, maybe from the mainstream kernel, but space missions use old hardware and you don't encourage recycling eWaste this way.
Maybe have a tree of obsolete architectures, where those who want old stuff can go.
Better yet, since architectures are split off anyway, have a subtree for each architecture (inline or otherwise). Let arch-specific distros merge the architectures they want and no others, let developers cleanly see what does what. It should make adding new architectures to (and removing old ones from) the mainline easier if the mainline had the arch section assembled.
Regardless of whether anyone thinks either is sensible or even sane, I might have a go at building a mega-arch (triumphal arch?) project just for kicks. Well, the DEC VAX arch needs to be less obscure, assuming there's still a copy out there.
It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
For most purposes, it's not necessary to test it on hardware. We know the underlying functions work and that side-effects are minimal. The requirements are that the API presented matches the API the kernel uses, the semantics are as expected and that the code compiles. That requires a continuous integration tool and a way to describe the semantics.
You can get the software for free. The effort is a bit more, depending on how much it is worth.
It won't prove things work, but it will prove things will probably work and that they should work. If you're using obsolete hardware, you can expect to have to make mods, this would place bounds on that.
It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
Same as with any other kernel function - no verifiable use.
The problem is that this buggers up my plan to travel back in time and upload the latest Linux source into Linus Torvalds' head when he was a student. The architectures removed were the only ones around then. How's that going to convince him, it he can't run it?
It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
The Powerbook 190 shares nothing with the rest of the Powerbook 100 series (160,165,170,180) and is more like the 5300. It was a stopgap "value" model until PPC machines came down in price. All the rest of the 100 series (except for the original 100) have a MMU equipped 68030 CPU in them.
I have an Apple IIe with an applecat modem.
Now Im showing my age. lol
Checklist:
1. Are you running KDE, Gnome or Unity? If so, bury them with dignity at the bottom of the garden and install Enlightenment.
2. Are you running SystemD? Ditto and install OpenBSD's init or Gentoo's OpenRC.
3. Is the kernel tuned? It should not contain things you don't use. If power is a problem, that is the governor you should select. Run with the worst latency you can afford because you need more power to get equal performance with low latency.
4. You might want to use the hoard memory allocator over Glibc's malloc for some apps - you can set this using the dynamic library path.
5. You might want to use a lightweight libc to reduce system strain.
6. You might want to experiment with apps to see what causes netsnmp to scream in pain at the load levels. It's a good monitor. Once you've identified the problem apps, place them in a cgroup that is more restricted on resources.
Should you have to do this? No. Fact is, you do, because distro authors are going for mass appeal rather than usability. They always have, which is why MCC was the last distro I actually liked using.
It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
The biggest issues with power use is the hardware drivers. Many times open source drivers are written without any involvement of the hardware manufacture's company. The manufacturers often don't publish the open hardware/firmware specifications. If they do, they can be incomplete. To get the full specs, you often have to be apart of some big company and sign an NDA agreement. This is very incompatible with open source code.
This leaves devices driver writers often black boxing the driver writing process so it is easy to miss things or not knowing if they've written the driver in the best or most efficient way possible. Most of the time it feels like we're lucky if it works and is stable.
This is why it is important to support companies the actively support open source driver releases. Binary blobs don't count. They're products often cost more, but if you're running Linux, you're more likely going to have a better experience.
No good deed goes unpunished.
I have an Apple IIe with an applecat modem.
Now Im showing my age. lol
I have an Apple 1.
Anyone interested?
"Wrong priorities man. Take over the world *first*, and then you can have others fix all your computer problems."
Not sure if this is a shot at humor, but I think you're bang-on with this.
Technology for me and many people on this site was a hobby once.
Sure, I can install a custom kernel to troubleshoot my hibernate issues, follow discussion forums for my laptop for audio issues, etc.. or I can do work.
What OS are you running? I used to have issues with older Ubuntu and Debian and basically anything that uses old PulseAudio. Most have switched to ALSA which isn't too much better but works. I used to manually have to strip pulse and force the OS to use ALSA. But those days are fortunately over. And for ease of use Ubuntu 16.04 and Debian 9 from experience.
And the CPU's are no longer used in the context of Linux, thus obsolete in the context of Linux.
Or at least, nobody spoke up for them on LKML.
The thing is, there are architectures supported by Linux which can only reasonably be called obsolete. What's relevant here is that these are becoming unsupported. Most architectures being dropped by Linux were already obsoleted in the eyes of all but hobbyists long ago.
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
I don't get the complaint. Chances are the stuff in the new kernels won't even work on legacy hardware. Put your big boy pants on, download an older flavor and get some back ported security patches and you're in the game.. Problem solved.. Linux on old hardware is easy, sometimes. The hard part is finding a distro that works with little modification. If you're trying to run Linux on old hardware you're already doing it for fun anyways because if it was for production you would already.
The powerbook 100 isn't and couldn't be supported--it's a 68000, with no potential for an emu (other than that bizarre hack that used two, and stopped one mid-instruction on a fault, using the other to fix things before resuming).
Oddas it sounds, the Powerbook 100 was not part of the Powerbook 100 series.
Apple basically handed the Macintosh Portable to Sony and said "shrink it." The result got called the 100.
There was one insignificant difference, other than size, iirc, but I forget what it was.
hawk, who still has his MacPortable, albeit spread across a couple of shelves
I'm curious what's considered an "obsolete CPU architecture" if a Powerbook 100 is still supported.
The question is "obsolete" for who? Obsolete depends on who defines it.
Obsolete due to lack of users and lack of working hardware?
Obsolete due to lack of vendor support?
Obsolete due to lack of maintainers?
Ultimately in every traditional version of obsolescence one group decides quite arbitrarily when something becomes obsolete and it has nothing to do with the orderly discontinuation of products in a linear timeline.
If I am the only person using and maintaining the source for hardware I developed my self, then it would perhaps still be maintained, even if it is obsolete.
You're describing a vendor actively maintaining a product in active use. That by its nature means it's not obsolete.
I read the release notes, most of it was already broken on those dropped architectures.
I'd say "nonworking and unmaintained"
Always the way, isn't it? I just threw out a bunch of Quadra 400 & 800 trays a few weeks back.
It isn't a full 040. The 'LC', low cost, in the model number refers specifically to a variation where they left out the FPU.
Power4 is an expensive niche processor, as are the others... The number of people who have access to such hardware, let alone the desire and knowledge to run linux on them are extremely small.
The m68k series on the other hand was extremely widespread used in all kinds of devices, including devices with niche followings like certain apple models and the amiga etc.
On the other hand, power4 servers were expensive when new, rarely crop up on ebay and the few people who would buy them typically do so to run AIX which doesn't run on any cheaper or more widely available hardware.
Ergo there are far more people interested in running old m68k machines than old power4 servers.
http://spamdecoy.net - free throwaway anonymous email - avoid spam!
Thanks for the thought, but the open file handles are a separate problem.
I should clarify, the error messages may say "permission denied", but the longer more expressive errors say the directory doesn't exist. Yet it is still returned as a member of the parent directory, so various operations will try to operate on the "ghost directory", and fail. It's not only obscure code which fails, but Windows Explorer is unable to delete or rename a directory which contains any such ghost directory.
If the problem were open file handles, I would expect that killing processes would solve it, and also that the only error would be "permission denied", and further, that the error would only occur when trying to rename or delete a file. Hence, there is a filesystem bug.
A cat can't teach a dog to bark.
The spirit is willing but the flash is week.
Ezekiel 23:20
Wrong priorities man. Take over the world *first*, and then you can have others fix all your computer problems.
If you run the world and someone else runs your computer, who really runs the world?
The architectures that were dropped are Blackfin, CRIS, FRV, M32R, Metag, MN10300, Score, Tile, and POWER4.
Blackfin is a DSP-oriented processor made by Analog Devices. It is not normally used for general purpose computing, so its applications rarely call for a full OS running on it. It's used in embedded applications, not as a user-visible CPU.
POWER4 was used in IBM mainframes circa 2004, and was never found in any Apple computers. The PowerPC 970 that was used in some Macs, which was derived from POWER4, is still supported.
The rest are architectures that were only used in embedded systems and have been obsolete for many years.
In summary, almost nobody is affected. A few of the old mainframes with POWER4 processors may still be in use and being used to run Linux. The embedded systems rarely get kernel upgrades anyway, so they can just stay with the last version that supported them.
Blackfin is still being made. But hardly anybody ever ran Linux on it. A common usage scenario for Blackfin is as a specialized DSP processor for a system that also contains another CPU (these days most likely ARM) that runs a full OS.