Linux Kernel 2.6.3 Has Been Released [updated]

justinarthur writes "At 04:36 UTC, Linux kernel version 2.6.3 has been made available. As is typical, downloaders are advised to utilize a mirror upon file availability. There are many changes from version 2.6.2, including recent ALSA patches, XFS fixes, and updates in many other areas." Update: 02/18 14:15 GMT by T : Peter Willis points out that kernel 2.4.25 (changelog) was also released, and writes "Incidentally, a security advisory dated today states there is an exploit in kernels up to 2.4.24 and 2.6.2, but the two releases today don't seem to reflect any changes, so get ready to patch up as soon as a patch pops up. More details on the vulnerability here."

Comments

SCO's code is available for perusal

[ObviousGuy]

Wait until 2.6.4 when that code has been removed.

Re:SCO's code is available for perusal

[BrynM]

Wait until 2.6.4 when that code has been removed.

Yep, the ink's still wet and SCO will have to amend their filing. The wheels of justice churn onward... *cough* *cough* er... The hamster wheels in Darl's head keep churning. My mistake. Sorry. Sometimes the sqeeky justice wheels sound just like hamster wheels.

Re:SCO's code is available for perusal

[warmcat]

Many a true word is spoken in jest... copied from this Yahoo SCOX messageboard post (thanks to jgabriel66 and kevinmcb2004 for finding the site)

The full URL for the parent package is here:

http://linuxupdate.sco.com/scolinux/update/RPMS.up dates/glibc-devel-2.2.5-213.i5 86.rpm

Here is the kernel mentioned in the grandparent post

http://linuxupdate.sco.com/scolinux/update/RPMS.up dates/kernel-source-2.4.21-138 .i586.rpm

(as mentioned in the parents they have set up some kind of HTTP auth, but blank credentials work fine)

What is bizarre is that the files are still being updated while all this crap is going on. For example http://linuxupdate.sco.com/scolinux/update/RPMS.up dates/mc-4.5.55-719.i586.rpm is dated 17th Feb 2004.

Further the RPMs are cryptographically signed with GPG keys for both

SuSE Package Signing Key [9C800ACA]
Caldera Security [C4970D31]

This is proof that the files are deliberately approved and issued by Caldera/SCOX, signing packages is the most protected step that can happen in a company for its sources and binaries.

Further when I look at /usr/src/linux-2.4.21-138/COPYING I see... the GPL v2 with a note at the top by Linus as usual.

Its like the arms and legs of SCOX don't know what the 455h013 is saying. HEY IDIOTS, you're still GIVING AWAY on your websites, under the GPL, the stuff you're trying to shake down companies for!! AND ITS GPG-SIGNED BY YOU THAT IT IS OKAY!!!!

Re:SCO's code is available for perusal

[warmcat]

They have kernel source in a kernel-source package in the updates dir, the distro sources and apparatently by the date the rest of the update sources are here:

http://linuxupdate.sco.com/scolinux/SRPMS/

The thing is they are compliant to the GPL :-)

Today they put up a new web-based "buy a year's license for the SCO IP" thing on their website. But why bother when you can license any SCO IP under the GPL irrevocably and for free, from the same website? Its just madness from SCO.

Re:SCO's code is available for perusal

[Anonymous Coward]

I bet there is a Linux person on the inside, working for SCO, thinking "hehehehehehehe that will show em" to himself right now...

Cheers to the nameless hero.

Great work!

[Michael Iatrou]

$ cat ChangeLog-2.6.2 | grep -xP "<.*@.*>" | wc -l
1105

$ cat ChangeLog-2.6.3 | grep -xP "<.*@.*>" | wc -l
1260

155 more bug fixes from the last release... (yeah, I know, not a really sophisticated method...)

Re:Great work!

[Feztaa]

gazbo mentioned the unneccessary use of cat, you actually don't need any pipes at all:

$ grep -xP "<.*@.*>" ChangeLog-2.6.2 -c
1105

$ grep -xP "<.*@.*>" ChangeLog-2.6.3 -c
1260

Re:Great work!

[cpeterso]

so Linux is getting buggier as more bugs get fixed??

Gentoo disappointment...

[dmayle]

I must admit to being a little disappointed right now with my Gentoo installation. I saw the article here, ran an emerge sync, and 2.6.3 is still marked as unstable. I guess I'll have to wait 15 minutes before I can install it... ;)

On a slightly related sidenote, whichever developer it was who "broke" support for the 105th key (the "Europe" key) in 2.6.1 should be drawn and quartered. It took me forever to figure out why my tilde key wasn't working. I created a text file called tilde with the character in it so that everytime I needed a tilde, I could copy and paste it... Aaarrggh

Re:Gentoo disappointment...

[Dahan]

At least I know that my fscking keyboard will work properly.

You may use Windows XP, but your keyboard must be running Linux, if it's always having to run fsck.

Re:Gentoo disappointment...

[vojtech]

That was me, thanks for the torture offer. ;)

Btw, the problem was that 2.6.1 actually supported the 103rd European key, and users couldn't cope with that. Now (2.6.2+) it's treated like a regular backslash key, even though they're two different keys in reality.

Re:Gentoo disappointment...

[dmayle]

I don't think it was that 2.6.1 supported the 103rd key, but that X doesn't. When I had 2.6.1 installed, X couldn't distinguish the key from the PrtScrn key, so there was no way to change the map file to support the key...

Re:Gentoo disappointment...

[Anonymous Coward]

That was me, thanks for the torture offer. ;)

You ought to get some kind of an award for that - smallest bug causing the most amount of people to tear their hair out :)

Re:Gentoo disappointment...

[GuyWithLag]

Humm... I have a multimedia keyboard on the PS/2 port and some keys make the kernel print a message and not forwading the keypress to X. Should I send info to you?

OnTopic: I'm hoping they've fixed BTTV support, module works just fine on 2.6.0, causes a modprobe segfault and OOPS on 2.6.2 (and will kill init when compiled in ....). Yeah, I have an old Miro PCTV card.....

Changes include

[halfnerd]

KernelTrap reported "large merges" to 2.6.3-rc2, including:

network driver updates, compiler warning fixes, PPC updates, a major ALSA update and SCSI updates, NFSv4 update, XFS fixes, ARM and sparc updates

Re:Changes include

[halfnerd]

And as if that wasn't enough, -rc4 introduced these:

out-of-the-box ppc/ppc64 support, complete with G5 support (64-bit), large radeonfb driver updates, IDE oops fixes (and cleanups), and a SELinux update

Slashdot really should consider linking to kerneltrap for each kernel update that has to be reported...

Summary Redefined by Changelog

[Red Pointy Tail]

Summary of changes from v2.6.2 to v2.6.3

At a tidy 9472 lines, I think the word 'summary' needs a new definition...

Re:Summary Redefined by Changelog

[cassidyc]

that would be summary, as in not the full unabridged version.

CJC

Re:Summary Redefined by Changelog

[lithiumcloud]

Which is called a 'patch'

distributions

[HoneyBunchesOfGoats]

Any idea when we'll start seeing 2.6 as the kernel included in the big distributions?

Re:distributions

[JET 666]

after 2.6.10

Re:distributions

[FooAtWFU]

I hear that it's the kernel in Fedora Core 2 test release 1.

Re:distributions

[RodeoBoy]

Fedora is scheduled for april, Debian and Suse by the summer. Start preparing now for the new install.

Re:distributions

[damiam]

Debian and Suse by the summer.

If by "Debian" you mean testing/unstable, then it's already there (and has been for a while). If you mean stable, it won't be default in the next release ("sarge"), so it'll be a few more years. Either way, "scheduled for this summer" is pretty far off the mark.

Re:distributions

[Malek the Damned]

I'm pretty sure it's planned for Mandrake 10.0

Re:distributions

[Anonymous Coward]

Industry sources are reporting Debian will include the 2.6 kernel as soon as Duke Nukem Forever is released :-)

Changelogs..

[noselasd]

Those that wants detailed changelogs, or just wants to follow the very very latest changes/additions to the kernel source tree can do so here
2.4 kernel tracking can be done here

Damn it!

[Rysc]

I still haven't upgraded to 2.6.2!

I grabbed the patch, applied it, reconfigured, recompiled, and set up grub. I've been waiting and waiting but still have not suffered a single crash, so I have been unable to justify rebooting.

Is any distro already with 2.6 kernel series?

[ron_lima]

Is there already any distro with the kernel 2.6 series? I'm still waiting for the response of the market before going for kernel 2.6 yet. My 2.4.24 is really stable and running fine, but I'm anxious to see the threads improvements in 2.6...

Re:Is any distro already with 2.6 kernel series?

[mshiltonj]

Fedora Core 2 is running the 2.6 kernel. I hear from first impressions that the performance
improvements are nothing short of stunning, but I haven't tried it yet.

Re:Is any distro already with 2.6 kernel series?

[Ender Ryan]

To really see the threads performance improvements, you will need to recompile glibc with support for it. From what I understand, this will break a number of programs and cause some programs to not compile, so it's not really considered stable yet by most people.

OTOH, even without rebuilding glibc to take advantage of the new kernel-based pthread support, I have found really nice interactivity improvements on the desktop. It's really quite nice, I wouldn't want to go back to 2.4.

Re:Is any distro already with 2.6 kernel series?

[noselasd]

Well, FC2 isn't really released yet, its only test versionsout now.
FC2 will be released April 19.

Re:Is any distro already with 2.6 kernel series?

[activesynapsis]

I'm running 2.6.0 and the multitasking performance is easily noticed on my SMP machine. My samba shared folders transfer to windows boxes faster, and the screen updates in X with both processors pegged are fluid, instead of jerky like they were with 2.4.x.

Re:Is any distro already with 2.6 kernel series?

[nbdy]

One thing not clear is that what kernel version 2.6.1? 2.6.2? 2.6.3? will Fedora Core 2 use?

Re:Is any distro already with 2.6 kernel series?

[sbryant]

Is there already any distro with the kernel 2.6 series?

Check out the current SuSE distribution (9.0) - that came with a RC for 2.6.0. I'm not sure whether newer 2.6.x kernels have been made available via online update, as I'm not running SuSE 9.0 myself. Alternatively, the 9.1 release will probably be out in April, and I'd be very surprised if it's not based on 2.6.

-- Steve

Re:This is great

[nbdy]

Just curious, how you get the 10%? System boot time, Executing application,...? btw, I mod your first post up...

TODO:

[gazbo]

Locate and insert link to the "unnecessary use of cat" awards page.

And I don't mean having feline carnal knowledge.

Re:TODO:

[tcopeland]

> "unnecessary use of cat" awards page

Done.

Many's the time I've been guily of a

[tom@hal tom]$ cat pmdexport.xml | wc -l
42031
[tom@hal tom]$

when a

[tom@hal tom]$ wc -l pmdexport.xml
42031 pmdexport.xml
[tom@hal tom]$

would have done just as well...

Re:TODO:

[Anonymous Coward]

Here you go..
This is also a useless use of wc:
$ grep -cxP "" ChangeLog-2.6.3
1260

Re:TODO:

[Anonymous Coward]

> Locate and insert link to the "unnecessary use of cat" awards page.

using cat means you can switch sources to something other than cat. probably the worst thing about the unix toolchain is not having every command fed with a pipe (there could always be some syntax sugar for "cat xyz |" in the shell). Not only does it make the commands more orthogonal, it makes the I/O async for free.

If you want to search in reverse, all you have to do is change cat to tac.

Re:TODO:

[Anonymous Coward]

One prog to read the lines, one prog to find them,
One prog to count the lines and to the console bind them.

vojtech!

[Ender Ryan]

Hey, good to see a kernel developer here on our beloved /.! Anyway, I just wanted to say thanks for all the joystick driver support over the years, it's been really excellent for my home set-top-box, among other things. So thanks for filling that niche.

And thanks for personally helping me customize one of the joystick drivers once-upon-a-time many moons ago.

Nforce fix went in

[motown]

[ACPI] nforce2 timer lockup from Maciej W. Rozycki

Oooh, does that mean I can finally enable both APIC and ACPI support in the kernel without experiencing lockups on my Nforce2-based system? I've been waiting for quite a while for this patch to go in. :)

But weren't there supposed to be two (complementary) patches for this problem out there?

len.brown@intel.com

Thanks Intel guy, for allowing this AMD fix to go in. :D

Radeon users will be happy

[CRCulver]

Previous kernels have never worked entirely well with my Mobile Radeon M6-P graphics card. Switching from radeon-powered X to radeonfb-powered console, or changing the resolution within radeonfb, would occasionally cause the screen to get all fuzzy, making me switch back and forth until it looked normal again.

But with the 2.6.3 kernel, there's a kickass new radeonfb driver that doesn't have any of these problems, and has improved collaboration with the BIOS to decide certain settings. No longer will peers think Linux must suck because my screen occasionally gets fuzzy.

However, I'm still only getting 435fps in glxgears with a 16MB graphics card at 1024x768 with DRI definitely on. Is this normal for a sucky notebook display card, or is there a problem with Linux's radeon gl support or my settings?

Re:Radeon users will be happy

[HoldmyCauls]

435 is a low number to you? For how long have you been working at Pixar? My Radeon Mobility M7 *occasionally* gets 70 or higher at that resolution, with that amount of video RAM. BTW, the reason it never *feels* like 435 fps is probably because you're running ATi's glxgears which somehow does *wonderfully* for testing. Tuxracer, RtCW, Unreal2003, whatever else you're using, don't have the same mysterious code that ATi's glxgears does, IIUC.

Re:Radeon users will be happy

[green pizza]

However, I'm still only getting 435fps in glxgears with a 16MB graphics card at 1024x768 with DRI definitely on.

I've heard that glxgears is a terrible benchmark, but it piqued my curiousity enough to fire it up on my PowerBook (Radeon Mobility 9600, Mac OS X 10.3.2 w/ optional X11 installed). With the window geometry set to 1024x768 I'm getting ~1800 to ~2700 fps, it really bounces around. Given that glxgears doesn't use any texturing nor any other wizbang new features, I'm guessing that something's not right on your end. Maybe it's even rendering with software?

Re:Radeon users will be happy

[green pizza]

435 is a low number to you? For how long have you been working at Pixar? My Radeon Mobility M7 *occasionally* gets 70 or higher at that resolution, with that amount of video RAM.

70?? And video ram should have nothing to do with it, glxgears is just blasting out pixels like a machine gun. Unless Apple did some sort of magical OpenGL kung-fu to their X11 implementation, I'd say that something isn't right on your end. I rarely get less than 1000fps reported from glx gears in a big window (though that drops way down when I go full sreen).

Re:Radeon users will be happy

[Carnildo]

Maybe it's even rendering with software?

I got 20fps with software (MesaGL) rendering four years ago on a 400MHz K6-II. On any relatively recent computer, not even software rendering should be that slow.

Re:Radeon users will be happy

[Jacek Poplawski]

Please stop using glxgears as benchmark. This application draws just few triangles, without textures. Try to use glxgears in software and hardware mode. Then try any serious app, and compare.

Re:Radeon users will be happy

[CRCulver]

No, I'm using the bog-standard XFree86 glxgears. I've never used any of ATi's stuff. At least now I know from other posts that glxgears is pretty worthless for benchmarking.

I20 drivers

[DJayC]

I'm just waiting for the I20 drivers to finally compile cleanly in 2.6. I haven't been able to compile 2.6 on a server here because it requires I20 support that is apparently broken. Bummer...

Debian and Slackware already patched

[mst76]

According to the changelogs, Debian and Slackware already have the patches for the bounds checking error in place. I didn't check the other distributions. (Or rather, I don't know where to check most of them quickly.)

It still crashes :(

[motown]

The bug's still there. :(

The lockup can be easily triggered by running "hdparm -t /dev/hdX" (hdX being your harddrive) from the root prompt a couple of times.

I know, I know, it's easy (as well as unfair) for me to bitch and moan, since I'm not a kernel developer. But knowing that there are patches out there that could solve this problem, but still haven't made it into the kernel, it's frustrating.

For what it's worth: I haven't noticed any speed advantages between an APIC-enabled and non-APIC-enabled kernel, so I guess it's not too much of a loss.

Does anybody here have any better experiences with this?

Re:It still crashes :(

[Viduliya]

Yes, I've tried this too. System locked up just as you said with the 'hdparm -t /dev/hdX'.

What was fixed by this fix from the 2.6.3 kernel changlog?

[ACPI] nforce2 timer lockup from Maciej W. Rozycki

Can anyone please let me know where I can find a patch that can fix this problem properly?

Thank you

Vulnerabitily fixed in 2.6.3 and 2.4.25

[Shanes]

At least acording to Linus .

Re:Vulnerabitily fixed in 2.6.3 and 2.4.25

[graf0z]

... and 2.4.25

From the ChangeLog of kernel-2.4.25:

Summary of changes from v2.4.24-pre3 to v2.4.25-pre4 [...] Andrea Arcangeli: malicious users of mremap() syscall can gain priviledges

Date of patch-2.4.25-pre4-pre5 (did not find pre3-pre4) ist Jan-15! Why did it take so long to get the crowd informed? Same thing happened with the other do_mremap() bug.

Hacker's guide:

• watch changelog of testkernels
• wait for words like "malicious" or "exploit"
• have fun ...

/graf0z.

HFS+?

[Anonymous Coward]

The 2.6 series doesn't have support for HFS+ disks yet. (2.4 did.) Anybody feel like porting the latest HFS+ driver to 2.6? :-)

I think I speak for everyone

[eddy]

Sometimes the sqeeky justice wheels sound just like hamster wheels.

..when I say "Squeeky wheel gets the kick! Go for the eyes Boo!"

ACPI cure for 2.4.25 HOW-TO

[Quietti]

Many people complained that kernels since 2.4.22 broke ACPI support on their hardware.

The cause is a brand new ACPI implementation which has a cutoff date of January 1st 2001. If your computer's BIOS is older than that, any ACPI support that might be present will be completely ignored by the kernel. ACPI hacker Len Brown explains that while the cutoff date is indeed arbitrary, it was already being used by certain distributions who noticed a pattern in when BIOSes with broken ACPI support where manufactured, so the ACPI hackers stuck by that concensus.

If you know for fact that ACPI worked fine on your computer until 2.4.21, you can enable it again: the cure is to put acpi=force in your bootloader configuration options.

Len also noted that there might eventually be a whitelist of older BIOS versions whose ACPI support is spotless. If you feel that your motherboard is one of those that should be whitelisted, file a bug at Kernel.org. Len makes absolutely no promise whatsoever that such a whitelist will ever be implemented, but still leaves the door open for people to manifest their interest via the above bug report form.

Obligatory Gentoo

[MarcQuadra]

I just totally rebuilt my systems with gcc-3.3.2 / glibc-2.3.2 / kernel-headers-2.6.1, I had to drop a few packages, but overall everything works splendidly. I'm in that teensy little group of people who actually have an entire system built against the NPTL thread libraries.

Honestly, I don't notice much of a difference, just some 'different' behavior when I'm loading the system. It's not like things compile ten times faster or mozilla opens in much less time. I noticed that my system 'loads' differently now, full-speed network transfers tend to take all the unused CPU power, but only when no other processes want it. Before, under 2.4, full-speed net transfers would only tax my CPU to about 10%, but I suppose having the slightly greater throughput at the cost of 'junk' cycles makes sense, it's probably a high-res timer kicking in to make sure the NIC is 'properly fed'.

This weekend I'll be rebuilding my PPC and x86 machines AGAIN against gcc-3.3.3 / glibc-2.3.2 NPTL / linux-headers-2.6.3, I think I'll give the advanced IO-APIC a try too. I'm hoping 2.6.3 will fix some of those broken packages on PPC. Wish me luck!