Architectures/ARM/AArch64

In other languages: English (en) русский (ru) 中文（中国大陆） (zh-cn) [edit]

Have a question? Join the Fedora ARM team on IRC in #fedora-arm on Freenode

Current status

AArch64 is now being built in koji (Fedora Rawhide/F21). Most packages are now built, but a few still require attention. If you would like to help, please check for failed builds in Fedora ARM koji.

If you just want to try Fedora for AArch64 on the Foundation Model (simulator), please see the QuickStart guide.

Introduction

ARM chips have been wildly successful in embedded applications, cell phones and in tablet devices, but until recently the ARM Architecture was able only to address 32-bits of physical memory. As part of a push into the nascent ARM server (Hyperscale) market (as well as for use in more complex traditional embedded applications), ARM have extended the 32-bit Architecture to support 40-bit physical addressing in the Cortex-A15 processor. This was a big step, and several vendors have shipped successful 32-bit ARM server platforms based upon the LPAE 40-bit technology within the Cortex-A15, the older Cortex-A9 processor, and also using custom designs conforming to the 32-bit architecture.

A broader push into the Enterprise space requires the introduction of a 64-bit ARM Architecture. This happens with the introduction of the new AArch64 execution state in version 8 of the ARM Architecture. AArch64 is not a pure extension of the 32-bit ARM Architecture, but is rather a new architecture state within ARMv8, complete with a new A64 instruction set. It is born out of many years of research into modern architecture design. Importantly, AArch64 as a separate execution state means that some future processors may not support the older AArch32 execution state. Although initial 64-bit ARM processors will be fully backwards compatible, we are taking the bold and forward-looking step of treating AArch64 as the only execution state on ARMv8 processors. We will not be supporting 32-bit execution on these systems. This will allow for many beneficial implementation trade-offs to be made, such as the use of a larger 64K page size by default, and will result in 64-bit clean ARM server systems free from legacy code. It is important to make this distinction immediately such that there is an opportunity for 64-bit only server systems to emerge within the next few years. Not having to implement a full 32-bit pipeline in addition to the newer 64-bit architecture will increase energy efficiency in future ARM server systems. Think of AArch64 as part of a natural progression of supported ARM Architectures in the Fedora ARM project: armv5tel, armv7hl, aarch64. The new architecture is named aarch64 consistent with ARM's own choice of upstream naming, and in respect of ARM's desires to separate the ARM trademark from the architecture name.

You can find out more about the | ARMv8 Architecture on ARM's website.

This page documents the initiative to bring Fedora support to the AArch64 execution state within the ARMv8 architecture, starting with the bootstrapping of Fedora 17. Future releases of Fedora will support both ARM-implemented 64-bit processor cores - such as the new Cortex-A57 - as well as custom 64-bit cores from third parties like Applied Micro. A demonstration of the initial bootstrap has already been given on the latter hardware. The goal is to support an initial Fedora 19 remix on real ARMv8 silicon, with a more complete Fedora 21 release. A formal ARMv8 build system is established. We will also collaborate with both Linaro and the Linaro Enterprise Group, as part of a broader Linux community effort to support the new 64-bit architecture.

Target flags

No special flags are required at this time.

Hardware floating point

Hardware floating point in ARMv8 processors is considered standard and not optional. Hence, no special options are required.

Bootstrap history

If you are interested in how AArch64 was done in Fedora then bootstrap page has information about it.