Linux 2.6.27 Out

diegocgteleline.es writes "Linux 2.6.27 has been released. It adds a new filesystem (UBIFS) for 'pure' flash-based storage, the page-cache is now lockless, much improved Direct I/O scalability and performance, delayed allocation support for ext4, multiqueue networking, data integrity support in the block layer, a function tracer, a mmio tracer, sysprof support, improved webcam support, support for the Intel wifi 5000 series and RTL8187B network cards, a new ath9k driver for the Atheros AR5008 and AR9001 chipsets, more new drivers, and many other improvements and fixes. Full list of changes can be found here."

Comments

Not in upcoming Debian

[gringer]

It's a shame this won't be in the upcoming Lenny release of Debian. The in-kernel support for heaps of webcams via gspca is a very nice user-visible element of this release.

http://release.debian.org/emails/release-update-200808

Although, I guess they made the decision for 2.6.26 before they realised that a September release would be an impossible target.

Re:Not in upcoming Debian

[Bent Mind]

To each their own. However, I always preferred having the driver just be there when I need it. I always found it annoying, under Windows, to have to hunt down drivers. Especially when you have a hundred similar devices that have the same binary driver blob (same chipset) but require a hundred different INF files because every company that assembles a board insists on having a unique driver download. Then you can throw in driver signing that makes life even more difficult.

Linux drivers are much easier to deal with.

This is a huge amount of work

[QuantumG]

In only 3 months, all of this code has been completed and reviewed by multiple developers. This happens *every* three months. The pace at which the Linux kernel is moving and yet still maintaining quality is incredible. It is clearly the case that the Linux kernel has hit a new kind of critical mass and is now a form of software development that has never been seen before. The sheer number of people involved changes what is possible. If you suggested that every single change to the codebase be reviewed by multiple developers in a traditional proprietary software development house you would be, rightly, laughed at. There simply isn't the resources.

Re:This is a huge amount of work

[FooAtWFU]

If you suggested that every single change to the codebase be reviewed by multiple developers in a traditional proprietary software development house you would be, rightly, laughed at. There simply isn't the resources.

Where I work, it's called "pair programming".

(If two programmers is enough to count as "multiple". Also, bug fixes are supposed to get an additional diff check.)

If you do it right, you not only save time by not-writing bugs you don't have to fix later, but you can also avoid wasting all sorts of time (writing the feature wrong, going down paths that could lead to disaster, or spinning your wheels and banging your head when you can't figure out something stupid like feeding rrdtool deltas when it expects raw counters...), and you can bring new developers up to speed on a code base very very quickly.

Re:This is a huge amount of work

[QuantumG]

Do you have this hardware? Any chance you could narrow down the versions it works on and the versions it doesn't?

This is a general problem with kernel development.. if you don't have the hardware, it's a bitch to test. Please do contribute your findings.

Re:This is a huge amount of work

[cryptoluddite]

Do you have this hardware? Any chance you could narrow down the versions it works on and the versions it doesn't?

Same hardware as this guy:
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/266927

System is at work... I would test except there are not any easy options for doing so there. Also, I realize that you can't be expected to fix hardware problems where you don't have the hardware... in fact I've personally seen code fail on one system and run perfectly on the exact same spec hardware sitting right next to it, with exactly the same software (ghosted).

Mostly I'm just pointing out that there are longstanding problems in linux... the original fanboy post was way over the top.

Re:This is a huge amount of work

[RzUpAnmsCwrds]

If you suggested that every single change to the codebase be reviewed by multiple developers in a traditional proprietary software development house you would be, rightly, laughed at.

When I was at Microsoft, that's exactly how it worked. All code had to be reviewed and approved by the feature owner and the PM. There was also a team that reviewed any changes to the common libraries, in addition to the PM.

In addition, to actually get code checked in, it had to pass FxCop (code standards verification tool), not break the build, and not break any of the build verification tests (BVTs).

Mind you, I worked in the test team. Developers have to go through all of the same steps, and then their code also gets tested by the test team.

Re:This is a huge amount of work

[QuantumG]

When I worked at VMware we had to get code reviews for every checkin. Code reviews are literally the only thing that has been shown to consistently improve quality. Of course, it's not just code reviews.. it's also attitude. If you're accepting of stuff being broken because it is "in development" then that's what you'll get. On the other hand, if you have a tight knit small team working on the same stuff then you can get similar quality by just maintaining pace and having lots of communication through the code.. but that doesn't scale.. this does.

Re:This is a huge amount of work

[RiotingPacifist]

Did you bother reading the bug report:

...it seems linked to the HDA Intel chipset, although I do not have this problem in Fedora or PCLinuxOS."

Its a ubuntu problem not a kernel problem, i would have guessed it was pulseaudio/alsa problem and not a kernel based problem too.

Re:This is a huge amount of work

[paulbd]

There are also longstanding issues with Intel HDA hardware ... this supposed "standard" isn't really a standard at all. It has huge amounts of slop for mobo makers to futz around with the pinouts, and indeed, there are at least as many variants on HDA h/w as seen by the kernel as there are major laptop models. The windows drivers work because of collaboration with mobo makers who provide the info about how they specifically wired up the pinouts. The linux ones are a case of trial-and-error. There are thousands (or even millions) of Intel HDA users on linux who do not have your problem, another whole set who do, and and even bigger set who have different problems with this godforsaken "standard" h/w.

AR5008

[log0n]

Excellent! Macbook & Pro users can finally have wifi support.

'pure' flash devices

[Chris Pimlott]

Before you get all excited about running UBIFS on your USB drive, take note: UBI is not for consumer flash media. These devices already incorporate hardware to hide their flash nature so they look like a plain old block device to your OS. UBI is for pure flash devices that directly expose the quirks and distinct characteristics of the underlying media.

So what kind of flash hardware is this for? Embedded devices, apparently. But maybe as flash storage becomes more common, more devices will support raw access?

Re:'pure' flash devices

[schwaang]

It seems quite likely that OLPC will largely replace jffs2 with UBI for the internal nand on the XO. Good news. Maybe this will apply to the Asus eee as and other solid-state drives as well.

Re:'pure' flash devices

[david.given]

How about SD cards? They appear to be rather low on circuitry.

No, SD cards still have an on-board microcontroller. If you take the lid off, there are usually two chips in there: one's the flash itself, the other's the microcontroller.

(SD cards are awesome if you're a homebrewer. They speak a high-level protocol over a very simple four-wire serial interface. It clocks down far enough that it's possible to hook one up to, say, a C64 or Spectrum by just connecting it to some spare I/O pins and wiggling them manually. You can then read and write 512 byte sectors by sending the appropriate command, and you don't have to worry about any of that horrible flash stuff.)

Embedded devices for sure

[Weaselmancer]

Yeah, embedded devices definitely. It'll be awfully nice to see simple flash chips soldered onto a board rather than someone bolting an SD or compact flash socket onto them just so you can have a boot device.

Fragile, more expensive, and adds another physical item that can break. And not only that, but you can drop about 20-30 dollars worth of non-essential hardware from your design and still be on target. If you do any embedded work you know how big 20 dollars worth of hardware savings is. This new driver is *huge*.

Re:ACPI

[Anonymous Coward]

> Any chance that this will fix some of the ACPI problems with Linux?

Just to be clear, ACPI problems are motherboard problems, not Linux problems.

If the ACPI function of your motherboard is correct and compliant with the ACPI specification, Linux will work just fine.

Part of the motherboard ACPI problem is that Windows expects, and uses, some functions within ACPI that are not compliant with the ACPI specification ... you know the drill: embrace, extend, obscure, try to screw the opposition ...

Fortunately with ACPI we have not quite yet got to the "extinguish" phase.

Re:Linux 2.6.27 Out

[frankm_slashdot]

sad part is i just pre-ordered the openbsd 4.4 cd set... hah. im not sure if i should be proud or ashamed.

then again, i sometimes think im the last of the right-os-for-the-job heretics... openbsd on my firewall. solaris (with zfs) on my fileserver... mac os x on my main desktop... (i dabble in photoshop and video.. mostly failed fark contests. ha) and windows vista on my macbook pro (along side of os x of course)... because i do a lot of autocad/solidworks stuff on the side. my webserver runs gentoo..

i guess you could call me a glutton for punishment.

Re:Thank you Linus.

[JohnFluxx]

I know you joke, but on average he merges four code bases (patches) per day. That is not trivial by any measure.

Re:ACPI

[TheNetAvenger]

Part of the motherboard ACPI problem is that Windows expects, and uses, some functions within ACPI that are not compliant with the ACPI specification ... you know the drill: embrace, extend, obscure, try to screw the opposition

Yet Windows works around more 'crap' ACPI implementations than it 'takes advantage of' non-compliant specifications.

This is really a goofy argument, as there is very little mainboard ACPI implementations that are Windows specific, let alone off spec to be Windows specific.

Instead you find crap Motherboards that still have exceptions for OS/2 RAM usage, non-Windows features like VGA palette crawling, cobbled Sx states, and horrid USB support for 'legacy' OS methods that Windows hasn't used in 10 years. (Yes we know these are not all ACPI specific)

I'm sure it is fun to blame windows for ACPI sucking and Linux's support of ACPI sucking.

The bottom line is, ACPI tends to suck, and Linux doesn't have the development resources to make it work in all circumstances, even though it does a pretty good job. Apple has trouble with their hardware, yet have few model, moved to EFI and still have some of the same inconsistent behavior Linux and Windows users encounter or messed up combinations of hardware.

As for ACPI, MS tried to push the industry on ACPI and move past it back in the 90s, and it was hobbists that were using non-Windows OSes like Linux that screamed and stopped EFI type suggestions from taking hold. MS shoved for legacy free BIOS concepts, and there is some hardware even out there that used a generic proprietary EFI type of legacy free BIOS system, go look at Toshiba laptops from 2002 that required OS level drivers, as there was no traditional BIOS. They also didn't have legacy ISA or older device support and could boot WindowsXP in less than 10secs on some machines, and return from a full hibernate in under 2 secs because of no BIOS time delay.

Just to blow your argument to the side, crap like this link would not exist if Windows did have more control over ACPI compliance as you suggest. http://support.microsoft.com/kb/831691

Specifications and variations in the specification is an area that 'logic' would dictate that the OSS model would be supreme; however, in reality, the complexity and diversity of the implementations favors larger production OSes like Windows where exceptions have to be implemented, and a large vendor like Microsoft can force Motherboard companies to clean up their crappy implementations or work around them, as Windows often does.

One of the biggest bitches users had with Vista and hibernation and Standby were because of Vista adhereing to the specifications and trying to force vendors to do the same, so that S1,S2,S3 etc were consistent. Instead MS had to write a bunch of 'exception' code for motherboards and even up until SP1 was still adding code to deal with crappy motherboard implementations to get the hibernation and standby back in line so that hybrid sleep could work consistently.

Microsoft doesn't have control over the hardware markets like people assume they do, and never really have. If they did, they would not have had to resort to proprietary hardware for the XBox 360, as some of the hardware specifications in the console are things MS shoved for in the PC market years before. Just an example would be unified shaders, and this didn't finally get shoved to PC users until Vista's DX10 required them, even though the benefits of a more agnostic GPU shader system was known years and years ago.

Kernel changes I wanted in as of v1.0

[dayton967]

Way back when, I made a comment that the kernel should, be more modular, that was done. But I also stated that the development of the Kernel, and the Drivers should be seperated.

This is to save on these massive downloads required these days, also to allow for faster development of both kernels, and drivers.

One requirement of this, would be to build out driver stubs, so that there would be standardize the communication between the kernel and the drivers.

- Some of the benefits would be to have faster development schedules.
- Reduce the downloads.
- Provide a method for Hardware modules to communicate with the kernel. Allows for commercial modules to be used, and to provide a method for the kernels to be developed without kernel specific code.
- Removes the requirement for kernel specific modules. Some hardware doesn't have even upto date drivers because of changes with the kernel. (VMWare has this problem with the VMWare-Tools, considering the code hasn't changed that much there is at least 2 #ifdef's for the 2.6.* kernels).
- Allows for urgent updates of individual drivers. eg. e1000
- Distributions would upgrade more frequently, instead of back porting some fixes.
- Reduced bandwidth requirements, don't have to download a 50-60M tar.gz for the kernel, or 17+M for the kernel.
- Ultimately, it would eliminate a person from making a change in an area of the kernel, that affects many other modules, which results in changes in those modules or bugs in those modules.

All of this would allow for greater development speed, improved security, reduction of bugs.

Stability version?

[Spazmania]

When is the next stability-focused version (like 2.6.16) due out?

Intel e1000e bug fixed?

[vidarlo]

This bug could've been a showstopper. It essentially ruined your intel e1000e ethernet card, by overwriting the firmware. They've not patched it, according to LWN:

It is worth noting that, as of this writing, 2.6.27 does not contain a fix for the e1000e hardware corruption bug. What it does contain, though, is a series of patches which will prevent that bug from actually damaging the hardware. That makes the kernel safer to run, which is an important step in the right direction.

What does that mean? Obviously, it should not ruin your ethernet card anymore, but will e1000e work very well with this kernel? Or what?

Since this is a pretty high-profile bug it's strange it ain't mentioned in the summary. E1000e is a very popular gigabit ethernet chip from Intel, and actual hardware corruption is serious and (luckily) rare.