Date
1 - 11 of 11
Building Yocto on M1 Mac
|
Abhijeet Tripathi
Hello yocto-devs, I'm new to the yocto project and trying to set up a build environment on my M1 Mac which is arm64 architecture. I'm following the steps as mentioned in below link: But it is always pulling up containers based on amd64 architecture. Are there any updated steps/link to set up a build environment specific to M1 Mac(arm64)? Thanks, Abhijeet
|
|
|
|
Anders Montonen
Hi,
Looking at Docker Hub, none of the published Crops containers are built for ARM64. You can try building the containers yourself, making sure to pick base distros that are available for ARM64. Note that there are plenty of reports of Docker performing poorly on Macs, probably because of the file system implementation. You may be better off using a full VM via e.g. UTM. -a
|
|
|
|
Abhijeet Tripathi
On Sun, Feb 13, 2022 at 7:09 PM Anders Montonen <Anders.Montonen@...> wrote:
I initially started with UTM but the problem with virtual machines is that we can only use half the resources of the machine. This makes the builds slower, so I was looking at the container way. Can you point me to any documentation which I can use to build customized containers for yocto builds? Thanks, Abhijeet
|
|
|
|
Alexander Kanavin
Even if you do manage to start a build, performance won't be great.
Yocto, by design, builds the entire stack from source, so a lot of CPU cores is essential. Don't spend your time waiting, get an appropriately specced Linux workstation, or an account on a Linux server that your organization might have for the purpose. Alex On Sun, 13 Feb 2022 at 15:56, Abhijeet Tripathi <abhijeettripathi3003@...> wrote:
|
|
|
|
Josef Holzmayr
Howdy!
While that is perfectly true from the pure technical facts, I have to add that building in Parallels on a M1 is ok-ish. You have to crank up the VM way beyond the defaults, to something like 6-8core, 16+X GB RAM, but then it’s bearable for the occasional user. When you can’t get your hands on a proper environment, its certainly an option. Greetz
|
|
|
|
Anders Montonen
-a
|
|
|
|
Robert Joslyn
On Feb 13, 2022, at 6:56 AM, Abhijeet Tripathi <abhijeettripathi3003@...> wrote:A Dockerfile like this works for me: FROM ubuntu:20.04 ARG DEBIAN_FRONTEND=noninteractive ENV LANG=en_US.UTF-8 RUN apt-get update \ && apt-get -y install \ locales \ sudo \ vim-tiny \ && sed -i '/en_US.UTF-8/s/^# //g' /etc/locale.gen \ && locale-gen \ && apt-get -y install \ binutils \ build-essential \ chrpath \ cpio \ diffstat \ gawk \ git \ lz4 \ python3 \ python3-distutils \ wget \ zstd \ && rm -rf /var/lib/apt/lists/* RUN useradd -m -G sudo --uid=1000 -s /bin/bash yocto RUN install -d -o yocto -g yocto /yocto USER yocto Save that to a file called “Dockerfile”. You can build a container called “yocto” with: docker build -t yocto . You can run this with something like: docker run —rm —mount type=volume,src=yocto,dst=/yocto -it yocto Once in the container, go to /yocto to do work within a Docker volume. I have an M1 MacBook Air and can use this container to run builds. As others have said, it’s not fast, but it does work if it’s what you have. Normally I don’t do builds on the MacBook, I usually ssh into my Linux desktop and do my work there. You got me curious though, so I did a quick comparison of building on my MacBook Air and my desktop. Using the container built with that Dockerfile, I ran this sequence (the download is separate to avoid download time, which is highly variable): git clone https://git.yoctoproject.org/poky.git -b honister cd poky . oe-init-build-env bitbake core-image-minimal —runonly=fetch rm -rf tmp/ sstate-cache/ time bitbake core-image-minimal On my 2020 M1 MacBook Air (8 cores, 16 GB RAM, docker using 8 cores and 8 GB), the build took 84 minutes. My desktop with an AMD Ryzen 9 3950X (16 cores, 64 GB RAM) it takes 21 minutes. Docker performance has always been bad for me on MacOS, so it wouldn’t surprise me if it’s faster to use a normal VM than docker (which uses a VM internally too). Robert
|
|
|
|
Ross Burton <ross@...>
On Mon, 14 Feb 2022 at 00:17, Robert Joslyn
<robert.joslyn@...> wrote: On my 2020 M1 MacBook Air (8 cores, 16 GB RAM, docker using 8 cores and 8 GB), the build took 84 minutes. My desktop with an AMD Ryzen 9 3950X (16 cores, 64 GB RAM) it takes 21 minutes.Try doing the MBA build with the laptop suspended in the air on a couple of pencils. As the MBA has no fan you might find that increasing airflow around the laptop makes a big difference. And giving the container more RAM/cores will help, obviously, there's no reason to limit it to half the power. Ross
|
|
|
|
Abhijeet Tripathi
On Mon, Feb 14, 2022 at 5:32 PM Ross Burton <ross@...> wrote: On Mon, 14 Feb 2022 at 00:17, Robert Joslyn Thanks for the pointers, let me try it out on my Mac. Will try to use max cpu cores and RAM to boost the performance a bit. Regards, Abhijeet
|
|
|
|
Abhijeet Tripathi
On Mon, Feb 14, 2022 at 1:29 AM Josef Holzmayr <jester@...> wrote:
I agree that a powerful Linux workstation will really help but right nowI do not have access to any of it. I Will try to check if I can get one, but in the meantime I would like to start on my Mac so that I can continue my learning. Thanks, Abhijeet
|
|
|
|
Robert Joslyn
On Feb 14, 2022, at 4:02 AM, Ross Burton <ross@...> wrote:Giving docker all 16 GB of memory made the build take 124 minutes. I assume this is due to the VM docker uses fighting for memory with the build and the host, slowing things down. I’m sure someone could find an optimal amount of memory to give the docker VM for best times. I did these builds with the MBA on its side, which should be a little better than it just being flat on a desk (or sitting on pencils). A fan would also help, but lack of a fan is one of my favorite features! This was also done at room temperature, 68 F/20 C, I’m sure it’d be a bit faster if I went outside where it’s below freezing, but I’ll let someone else run that test :-D The MBA is obviously not intended for this kind of workload, but it does at least work. Robert
|
|
|