Linux kernel 2.6.25 ARM平台正式挑战X86
时间：2008-04-20 来源：linux论坛
Linus Torvalds日前宣布了最新的Linux kernel 2.6.25，在该版本的内核发布过程中，居然附带了一份多大7.5M的内核变更说明，从2.6.25开始，ARM平台正式受到支持，开始和X86在未来的个人计算设备的竞争走向了前台。同时该版本的内核也包含有大量的实时Linux、桌面Linux以及服务器相关的变化。
Linus Torvalds has annointed a blockbuster new kernel that required some 7.5 megabytes just for its changelog. The 2.6.25 release sets the stage for a war between ARM and x86 for the personal computing environment of the future, while also merging tons of improvements for real-time devices, desktops, and servers.
In his lkml post (Linux kernel mailing list) announcing the release, Torvalds referred interested readers to learn about the major additions at Kernel Newbies, which listed salient points as follows:
* MEI/Panasonic MN10300/AM33 architecture support
* Marvell Orion architecture support
* A new interface for more accurate measurement of process memory usage
* A 'memory resource controller' for controlling the memory usage of groups of processes
* Realtime group scheduling
* A tool for measuring high latencies called latencytop
* ACPI thermal regulation
* Timer event notifications through file descriptors
* An alternative MAC security framework called SMACK
* An ext4 update
* BRK and PIE-executable address space randomization
* RCU preemption support
* FIFO spinlocks in x86
* EFI support in x86-64
* A new network protocol called CAN
* Initial ATI r500 DRI/DRM support
* Improved device support and many other small improvements
At the risk of summarizing a summary, real-time improvements are prominent, but Linux's big "real-time patch" still remains outside the mainline tree, as do several things that have to be merged first, like changes to soft IRQs and ISRs. Folks who want real-time Linux early can still get it from MontaVista, though -- or, from Red Hat and SuSE, which have recently switched from using the "processor shielding" technology they previously licensed from Washington, D.C.-area hardware vendor Concurrent.
Some features like the latencytop diagnostics tool and group scheduling will bring benefits to the Linux desktop, by helping engineers ensure that process groups required for, say, multimedia streaming have the CPU access they need for adequate performance. Servers and network infrastructure equipment are not left out, with new mainline implementation of ACPI thermal regulation and SMACK.
In the embedded space, there is lots to be excited about. There's a mainline implementation of CAN (controller area network), the bus that links the sensors to the engine control unit (ECU) in most modern automobiles. A light, real-time networking protocol, CAN is also popular in industrial computing. Support for CAN in mainline will help automakers and others adopt Linux, since it will make forward kernel migrations much less painful and bring more vendor neutrality across the market.
The real biggie, though, could well be support for Marvel's Orion architecture. Orion chips are currently the most powerful ARM-based processors that are widely available at commodity pricing, thanks to their widespread use in consumer NAS devices. Orion chips can support up to 2GB of RAM, making them very useful for natively compiling ARM software (gcc is a real memory hog). The lack of native development hardware has probably been the single biggest factor keeping the ARM architecture a "second class software citizen" to x86.
The stage is set
Kernel.org for Orion combined with cheap development hardware will radically facilitate porting software from x86 to ARM. This comes at a time when x86 is wading into the embedded space, with Intel's Atom and Via's Isaiah chips marching jointly under the standard of better software support than less power-thirsty chips based on embedded architectures like ARM.
Linux came to servers first. In 1999, Linus said he wanted to win the desktop, but that embedded was the nearest land of opportunity for Linux. It took a few years, but today (and for the last two or three), Linux has been the top embedded OS in new project starts, according to many market research firms. So the desktop is next, right?
Not exactly. A funny thing happened, on the way to the desktop. The desktop became embedded. Today, mainstream computing devices like the Asus EEE already have solid state disks, smaller form-factors, and other hallmarks of traditional "embedded" devices. There are even a few laptop-like devices showing up with ARM-based processors. Tomorrow, the personal computing device of choice may well be a pocketable iPhone-like device capable of plugging into any handy KVM (keyboard, mouse, and video display terminal).
Currently overwhelmed by traffic, the Kernel Newbies article detailing the new release's salient improvements may be available here. It may also be available here (Google cache, which may also be overwhelmed at times)
Linus Torvalds has annointed a blockbuster new kernel that required some 7.5 megabytes just for its changelog. The 2.6.25 release sets the stage for a war between ARM and x86 for the personal computing environment of the future, while also merging tons of improvements for real-time devices, desktops, and servers.
In his lkml post (Linux kernel mailing list) announcing the release, Torvalds referred interested readers to learn about the major additions at Kernel Newbies, which listed salient points as follows:
* MEI/Panasonic MN10300/AM33 architecture support
* Marvell Orion architecture support
* A new interface for more accurate measurement of process memory usage
* A 'memory resource controller' for controlling the memory usage of groups of processes
* Realtime group scheduling
* A tool for measuring high latencies called latencytop
* ACPI thermal regulation
* Timer event notifications through file descriptors
* An alternative MAC security framework called SMACK
* An ext4 update
* BRK and PIE-executable address space randomization
* RCU preemption support
* FIFO spinlocks in x86
* EFI support in x86-64
* A new network protocol called CAN
* Initial ATI r500 DRI/DRM support
* Improved device support and many other small improvements
At the risk of summarizing a summary, real-time improvements are prominent, but Linux's big "real-time patch" still remains outside the mainline tree, as do several things that have to be merged first, like changes to soft IRQs and ISRs. Folks who want real-time Linux early can still get it from MontaVista, though -- or, from Red Hat and SuSE, which have recently switched from using the "processor shielding" technology they previously licensed from Washington, D.C.-area hardware vendor Concurrent.
Some features like the latencytop diagnostics tool and group scheduling will bring benefits to the Linux desktop, by helping engineers ensure that process groups required for, say, multimedia streaming have the CPU access they need for adequate performance. Servers and network infrastructure equipment are not left out, with new mainline implementation of ACPI thermal regulation and SMACK.
In the embedded space, there is lots to be excited about. There's a mainline implementation of CAN (controller area network), the bus that links the sensors to the engine control unit (ECU) in most modern automobiles. A light, real-time networking protocol, CAN is also popular in industrial computing. Support for CAN in mainline will help automakers and others adopt Linux, since it will make forward kernel migrations much less painful and bring more vendor neutrality across the market.
The real biggie, though, could well be support for Marvel's Orion architecture. Orion chips are currently the most powerful ARM-based processors that are widely available at commodity pricing, thanks to their widespread use in consumer NAS devices. Orion chips can support up to 2GB of RAM, making them very useful for natively compiling ARM software (gcc is a real memory hog). The lack of native development hardware has probably been the single biggest factor keeping the ARM architecture a "second class software citizen" to x86.
The stage is set
Kernel.org for Orion combined with cheap development hardware will radically facilitate porting software from x86 to ARM. This comes at a time when x86 is wading into the embedded space, with Intel's Atom and Via's Isaiah chips marching jointly under the standard of better software support than less power-thirsty chips based on embedded architectures like ARM.
Linux came to servers first. In 1999, Linus said he wanted to win the desktop, but that embedded was the nearest land of opportunity for Linux. It took a few years, but today (and for the last two or three), Linux has been the top embedded OS in new project starts, according to many market research firms. So the desktop is next, right?
Not exactly. A funny thing happened, on the way to the desktop. The desktop became embedded. Today, mainstream computing devices like the Asus EEE already have solid state disks, smaller form-factors, and other hallmarks of traditional "embedded" devices. There are even a few laptop-like devices showing up with ARM-based processors. Tomorrow, the personal computing device of choice may well be a pocketable iPhone-like device capable of plugging into any handy KVM (keyboard, mouse, and video display terminal).
Currently overwhelmed by traffic, the Kernel Newbies article detailing the new release's salient improvements may be available here. It may also be available here (Google cache, which may also be overwhelmed at times)
相关阅读 更多 +
