--- wikisrc/ports/evbarm.mdwn 2017/10/21 22:43:03 1.43 +++ wikisrc/ports/evbarm.mdwn 2018/11/22 01:18:49 1.60 @@ -1,18 +1,16 @@ [[!template id=port port="evbarm" port_alt="arm" -port_var1="arm" -port_var2="armeb" -port_var3="earm" -port_var4="earmeb" -port_var5="earmv6hf" -port_var6="earmv7hf" -port_var7="earmv7hfeb" -port_var_install_notes="evbarm-arm" -cur_rel="7.1" -future_rel="8.0" -changes_cur="7.0" -changes_future="8.0" +port_var1="earm" +port_var2="earmeb" +port_var3="earmv6hf" +port_var4="earmv7hf" +port_var5="earmv7hfeb" +port_var_install_notes="evbarm-earm" +cur_rel="8.0" +future_rel="9.0" +changes_cur="8.0" +changes_future="9.0" thumbnail="http://www.netbsd.org/images/ports/evbarm/adi_brh.gif" about=""" NetBSD/evbarm is the port of NetBSD to various evaluation and prototyping @@ -24,25 +22,94 @@ Matt Thomas is the maintainer of NetBSD/ ### CPU types -The evbarm port can be built with a variety of CPU options. There are -three main variables: the instruction set, the endianness, and whether -there is hardware floating point. By default the CPU type is "earm", -and this implies little endian (el when explicitly stated), and soft -(emulated) floating point. Another example, suitable for Raspberry PI -2, is earmv7hf, which is the v7 instruction support, little endian, -and hardware floating point. +The evbarm port can be built with a variety of CPU options, corresponding to the +[large array of ARM CPU architectures](https://en.wikipedia.org/wiki/ARM_architecture#Cores). +There are +four main variables: the word size, the instruction set, the +endianness, and whether there is hardware floating point. By default +the CPU type is "earm", and this implies aarch32 (32-bit), \todo cpu +architecture, little endian (el when explicitly stated), and soft +(Emulated) floating point. Another example, suitable for Raspberry PI +2, is earmv7hf, which is aarch32, the v7 instruction set, little +endian, and hardware floating point. Typically, various boards are best compiled with a CPU type that matches the board's CPU and floating point support, but generally a lower CPU instruction set version is workable on a newer board. See build.sh and look for aliases for the evbarm port. +Through NetBSD 8, the evbarm port has supported exclusively the +aarch32 (32-bit CPU) sub-family of the ARM architecture. Some +processors, such as many supporting the armv8 CPU architecture, also +support a 64-bit instruction set, referred to as aarch64. This is +sometimes referred to as a distinct port, [[NetBSD/aarch64|aarch64]], +with code in src/sys/arch/aarch64, but it is built as the evbarm port +with aarch64 cpu type, and available as the alias evbarm64. + +Note that MACHINE_ARCH=aarch64 currently refers to the A64 instruction +set and the aarch64 architecture, built for the armv8 architecture. +(Note also that armv8 is the first architecture to support aarch64, so +this will not be an issue until at least armv9.) + +#### ABI types + +There are two basic ABIs on ARM. One, called oabi, assumed a +particular kind of hardware floating point (FPA). This results in +faulting any floating-point instructions for kernel emulation on a +vast number of CPus, which is very slow. A newer one, called eabi, +has two variants. Both have stricter alignment rules, tending to 8 +byte rather than 4 bytes for 8-byte types (but actually read the specs +if you care). The one without "hf" emulates floating point without +causing traps/emulation, and "hf" uses VFP instructions, which are +present on modern CPUs. See the +[TS-7200](https://wiki.embeddedarm.com/wiki/EABI_vs_OABI) and +[Debian](https://wiki.debian.org/ArmEabiPort) documentation. + +Now, EABI is normal, and OABI is crufty. The only real reason NetBSD +retains OABI support is binary compatibility with older releases. The +"arm" and "armeb" MACHINE_ARCH targets are OABI; the rest of the +targets, all having "earm" are EABI. + +\todo CHECK THIS: The "aarch64" MACHINE_ARCH target is an EABI variant. + +### Relationship of MACHINE_ARCH to official ARM terminology + +Note that these are all little endian, and have big endian variants +with a "eb" sufix. + +[[!table data=\"\"\\" +MACHINE_ARCH |bits | ARM architecture version |ABI +arm |32 |? |oabi +earm |32 |armv4 (effectively an alias) |eabi +earmv4 |32 |armv4 (no thumb, so ok on strongarm) | eabi +earmv5 |32 |armv5t |eabi +earmv6 |32 |armv6 |eabi +earmv7 |32 |armv7 |eabi +aarch64 |64 |armv8 |\todo ? eabi +\"\"\"]] + +\todo Explain why, if we have armv4, and this is confusing, we still have earm as a MACHINE_ARCH. + +\todo Explain why aarch64 is a MACHINE_ARCH, when it seems like it +should be something like armv8hf_64. + +\todo Explain if MACHINE_ARCH values correspond to a particular +argument to some CPU selection command in gcc (and/or clang). + ### Kernels and userland The evbarm userland can be used on any system that can run code of the CPU type used for the build. Typically, a particular board requires a kernel for that board. +### anita and qemu + +anita can be used to test builds. (In addition to anita, install qemu and dtb-arm-vexpress from pkgsrc.) The release subdirectory should follow the naming convention on the autobuild cluster, used below. + + - evbarm-earmv7hf uses "qemu-system-arm -M vexpress-a15" + - evbarm-aarch64 uses "qemu-system-aarch64 -M virt" + - Information on how to test emulated versions of other specific hardware is welcome. + ### Board specific information - [[Allwinner sunxi family SoCs|Allwinner]] - [[BeagleBone and BeagleBone Black|BeagleBone]] @@ -254,9 +321,6 @@ Support for NVIDIA [[Tegra]] K1 SoCs is ### Raspberry Pi Foundation **Raspberry Pi**/**Raspberry Pi 2**/**Raspberry Pi 3** The [[Raspberry Pi]] is a low-cost credit-card-sized computer from the Raspberry Pi Foundation. The Raspberry Pi, Pi 2, and Pi 3 are supported. -### Rockchip PX2/RK3066/RK3188/RK3188+ -Various [[Rockchip]] family SoCs are supported by the ROCKCHIP kernel. - ### Samsung **SMDK2410** The SMDK2410 is the reference platform for the Samsung **S3C2410** processor,