Linux Kernel 4.5 Officially Released

prisoninmate writes: Yes, you're reading it right, after being in development for the past two months, Linux kernel 4.5 is finally here in its final production version. It is internally dubbed "Blurry Fish Butt" and received a total of seven RC builds since January 25, 2016. Prominent features of Linux kernel 4.5 include the implementation of initial support for the AMD PowerPlay power management technology, bringing high performance to the AMDGPU open-source driver for Radeon GPUs, scalability improvements in the free space handling of the Btrfs file system, and better epoll multithreaded scalability. The sources are now available for download from kernel.org. Update: 03/14 13:24 GMT by T : Reader diegocg lists some other notable features (a new copy_file_range() system call that allows to make copies of files without transferring data through userspace; support GCC's Undefined Behavior Sanitizer (-fsanitize=undefined); Forwarded Error Correction support in the device-mapper's verity target; support for the MADV_FREE flag in madvise(); the new cgroup unified hierarchy is considered stable; scalability improvements for SO_REUSEPORT UDP sockets; scalability improvements for epoll, and better memory accounting of sockets in the memory controller), and links to an explanation of the changes at Kernel Newbies.

Comments

Still pretty crusty on laptops

[jones_supa]

I recently did a quick survey in Reddit on people's experience on suspend/hibernate, and I may summarize it simply by saying that Linux is not the best performer in this area. :D It's a shame that such an important laptop feature works so poorly. Some might say that it's because OEMs do not "support ACPI spec properly", but in practice most PCs don't... It could be more practical to just find the patterns that Windows uses, and imitate them.

One really weird thing is also that backlight adjustment requests are sent to both ACPI and GPU, which causes double backlight adjustment events on many laptops.

People fight about SystemD, various open source licenses, differences between DEs, filesystems, but at the same time there's these fundamental problems which should get way more attention. Sometimes it feels like we are in a house arguing what kind of wallpapers bring the best experience, while that same wall is infested with mold inside.

Some people still talk like this is supposed to be the hi-tech kernel that breathes new life to my PC. Are they blind to all this stuff happening?

Re:Still pretty crusty on laptops

[Bert64]

Suspend and hibernate works just fine on laptops designed to run linux (e.g. chromebooks), the same can be said of macosx - suspend and hibernate is perfectly reliable on apple laptops, but is usually flakey on a hackintosh.

Linux already has various kludges to emulate the nonstandard way in which windows handles power management, but laptops also often come with customised model-specific drivers so even if you run windows you often still have problems if you run the default drivers or drivers for the chipsets rather than the specific laptop model.
The lower end laptop makers also make things difficult for users by varying the hardware in the same model, when looking at laptops recently i was told that a given model could have any one of 3 different wifi and ethernet chipsets, and that i wouldn't know which until i physically took delivery of the laptop... They will guarantee that you get "an 802.11ac wireless card" and "a gigabit ethernet", but the performance, range, stability or cpu usage can vary wildly between chipsets as can compatibility with linux or other systems and even (albeit quite niche) features like wireless monitor or master modes are not available with some chipsets.
The chipsets in use for various components were always an important factor for me when deciding what to purchase.

Yes it's a huge nasty mess!

Re:Still pretty crusty on laptops

[fuzzyfuzzyfungus]

It's worth noting, in terms of 'difficulty of dealing with undocumented binary blobs and ACPI in general', that Microsoft's own designed, developed, and blessed Surface Pro and Surface Book products have been dogged by power management issues; and that's with hardware hand picked by Microsoft, an OS built by Microsoft, and drivers and firmware either written by Microsoft or written for Microsoft by vendors who do most of their driver development work to support Microsoft OSes.

Obviously "but look at the other guy!" isn't an argument against the fact that Linux on laptops indeed has issues; it just provides some perspective on what a ghastly mess PC power management is. If Microsoft's own "Our OEMs are making us look bad, so here's a kick in the ass and a reminder of what kind of products we want in the PC space" product can't power-manage properly, that doesn't imply positive things about the difficulty for the Linux kernel team of getting power management to work correctly on some random vendor's apathetic attempt to shove a laptop out the door for as little money as possible.

Re:Why branch when you don't intend to support it?

[Kjella]

As I recall, the older versions of the kernel added some features and offered others as third party patches. The 2.0.x, 2.2.x, 2.4.x, and 2.6.x branches were supported for a very long time, (...) This was a very successful development model for over a decade and I don't understand why that's changed. Arguably, it would require fewer work since there wouldn't be as many branches to maintain.

Actually it wasn't. Distros were massively cherry-picking from the odd-numbered branches creating huge variations from the stock kernel, creating strange bugs and making the big jumps was a huge pain because the more relaxed requirements to put it in a development branch led to poor quality. Linus tightened ship and basically said do development in your own branch, when it's ready merge it to the released kernel and ~2 months after the merge window closes it'll be released instead of years like the old kernel. No more hodge-podge development kernels full of half-assed changes. It got distros to work more on the upstream kernel than their own variations, leading to more manpower and higher quality in the core project. It was a great success.