Make run_qemu.sh usable on 64-bit Arm archtecutre systems #140

ikitayama · 2025-04-01T03:01:20Z

This PR is tested on my Ubuntu arm64 port VM with:

qemu=~/projects/qemu/build/qemu-system-aarch64 ../run_qemu/run_qemu.sh -r none --no-cxl --no-kvm --no-hmat -p 1S --no-kvm

the system about to boot.

marc-hb

Thanks for sharing this early prototype! I'm making this a draft because we can't just replace x86 with ARM and drop the former :-)

I also see some changes that really do not look ARM-specific.

Please break down this PR and re-submit smaller, more focused pieces; start with the simplest and easiest changes.

You can either use git branches or something like git push HEAD~x, see more advice here:

https://github.com/zephyrproject-rtos/zephyr/pull/83839/files

marc-hb · 2025-04-01T22:04:10Z

mkosi.ubuntu.default.tmpl

@@ -33,5 +33,6 @@ Packages=
  libtracefs-dev
  libtraceevent-dev
  libsystemd-dev
+  systemd


What specific failure does this fix? I really don't see how this could not already be included by default, the system would by definition not boot otherwise. I've tested this file without this line with both Ubuntu 22 and 24 and never had any issue.

See: https://github.com/pmem/run_qemu/issues/141 I am at this moment Im unable to launch a system with the specific --cxl-test option. Surely adding --cxl doesn't prevent system booting though.

UPDATE: problem is fixed now and this was not the way to fix it.

marc-hb · 2025-04-01T22:12:24Z

.github/workflows/main.yml

@@ -34,7 +34,7 @@ jobs:
          - os: ubuntu-24.04
            img_distro: ubuntu
            img_rel: noble
-        arch: [x86_64]
+        arch: [arm64]


This is a list, you must append not replace.

Ack. I'll fix.

marc-hb · 2025-04-01T22:13:11Z

debian_vars.sh

@@ -2,4 +2,5 @@
 # SPDX-License-Identifier: CC0-1.0
 # Copyright (C) 2023 Intel Corporation. All rights reserved.

-ovmf_path="/usr/share/OVMF/"
+#ovmf_path="/usr/share/OVMF/"
+ovmf_path="/usr/share/qemu-efi/"


Mmmmm... I thought you were only testing Ubuntu, did you test Debian too?

If the previous value is incorrect then don't leave it commented out.

My oversite - I didn't test run_qemu.sh with Debian, so I'll drop in the next spin.

marc-hb · 2025-04-01T22:14:33Z

run_qemu.sh

@@ -19,7 +19,8 @@ pmem_final_size="$((pmem_size + pmem_label_size))"
 selftests_home=root/built-selftests
 : "${mkosi_bin:=mkosi}"
 mkosi_opts=("-i" "-f")
-console="ttyS0"
+#console="ttyS0"
+console="ttyAMA0"


I really wonder why ttyS0 is x86-specific...?

Maybe there is a need to make this more flexible/configurable but it is really related to ARM?

In any case we can't just replace one hardcoding with another.

marc-hb · 2025-04-01T22:16:40Z

run_qemu.sh

-		"memory_hotplug.memmap_on_memory=force"
+		#"initcall_debug"
+		#"log_buf_len=20M"
+		#"memory_hotplug.memmap_on_memory=force"


I don't see how selinux=0 could hurt any architecture.

yes, disabling SELinux is a good thing other might complain but so be it. Will fix.

marc-hb · 2025-04-01T22:17:47Z

run_qemu.sh

@@ -729,7 +731,7 @@ build_kernel_cmdline()
 		# This requires systemd v251 or above (commit 4b9a4b017, April 2022)
 		kcmd+=(
 			systemd.set_credential=agetty.autologin:root
-			systemd.set_credential=login.noauth:yes
+			systemd.set_credential=login.noauth:yes\"


Whichever issue this helps with, unbalanced quotes inside a single constant can't be the solution :-)

Turned out, -append does not require a pair of """s, if I set -cpu explicitly in my case "max". In your q35 machine case it was not needed I suppose.

marc-hb · 2025-04-01T22:18:45Z

run_qemu.sh

@@ -1355,10 +1357,11 @@ get_ovmf_binaries()
 	if ! [ -e "OVMF_CODE.fd" ] && ! [ -e "OVMF_VARS.fd" ]; then
 		if [ ! -f "$ovmf_path/OVMF_CODE.fd" ]; then
 			echo "OVMF binaries not found, please install '[edk2-]ovmf' or similar, 'edk2-shell', ..."
-			exit 1
+			#exit 1


Should this code support AAVMF instead?

WIll remove exit 1 should be there to baill out.

marc-hb · 2025-04-01T22:19:57Z

run_qemu.sh

 accel="kvm"

+arch=$(uname -m)


Are you running ARM on x86 or ARM on ARM? I can see both being useful.

I'm on my MacBook Pro powered by M1, then on it I create an Ubuntu 24.10 as you suggested, VM. So 64-bit Arm on 64-bit Arm. That's my answer. Do you guys want to support say like riscv base system launch an X86 VM and do execute run_qemu.sh by the way?

marc-hb · 2025-04-01T22:20:33Z

run_qemu.sh

@@ -1455,7 +1478,7 @@ prepare_qcmd()
 {
 	# this step may expect files to be present at the toplevel, so run
 	# it before dropping into the builddir
-	build_kernel_cmdline "/dev/sda2"
+	build_kernel_cmdline "/dev/vda2"


I don't see how this can be ARM-specific...

marc-hb · 2025-04-01T22:24:46Z

ubuntu_vars.sh

@@ -3,3 +3,4 @@
 # Copyright (C) 2023 Intel Corporation. All rights reserved.

 ovmf_path="/usr/share/OVMF/"
+aavmf_path="/usr/share/AAVMF/"


I struggled to find which package contains this file, maybe add a few comments?

Surely, I'll add.

realm@machine-1:~/projects/cxl$ apt-get show qemu-efi-aarch64 E: Invalid operation show realm@machine-1:~/projects/cxl$ realm@machine-1:~/projects/cxl$ apt-cache show qemu-efi-aarch64 Package: qemu-efi-aarch64 Architecture: all Version: 2024.05-2ubuntu0.1 Multi-Arch: foreign Priority: extra Section: misc Source: edk2 Origin: Ubuntu Maintainer: Ubuntu Developers <[email protected]> Original-Maintainer: Debian QEMU Team <[email protected]> Bugs: https://bugs.launchpad.net/ubuntu/+filebug Installed-Size: 329778 Suggests: qemu-system-arm Breaks: qemu-efi (<< 0~20161202.7bbe0b3e-2) Filename: pool/main/e/edk2/qemu-efi-aarch64_2024.05-2ubuntu0.1_all.deb Size: 4101360 MD5sum: 321614f7e5cf568ea467e0466d0333c9 SHA1: edf3bf59385a2dad9d561e5da63a3426d68d1ee9 SHA256: d7059e3bf40b802e700cfd6be62ae627a035ae5b9432e4b354f1ded390f9ba90 SHA512: 57bb702923f3fb816620a8a5b1e987e8110b72e9cd78d4eb6b6adf8532d9b2f1b1a9db56e48f9457ede704be5146809273e9eb91a9afaea7547aba5642630952 Homepage: http://www.tianocore.org Description-en: UEFI firmware for 64-bit ARM virtual machines qemu-efi-aarch64 is a build of EDK II for 64-bit ARM virtual machines. It includes full support for UEFI, including Secure Boot. Description-md5: 0191997e8e7dbde0e9009d3719a10aea

This reverts commit add62e2.

marc-hb · 2025-04-04T02:30:01Z

@ikitayama I see you just pushed some reverts of commits already in this PR.

Just FYI, all pmem git repos are used the "traditional" git way, not the usual GitHub way.

https://docs.zephyrproject.org/latest/contribute/contributor_expectations.html#pr-requirements

...
During development, commits may include intermediary changes (e.g., partial implementations, temporary files, or debugging code). These should be squashed or rewritten before submitting the pull request. Remove non-final artifacts, such as:

Temporary renaming of files that are later renamed again.

Code that was rewritten or significantly changed in later commits.

Ensure Clean History Before Submission
...

ikitayama · 2025-04-04T02:34:43Z

@marc-hb sorry about this. Should I close this PR? As this is not merged as is, and I've made smaller chunks of PRs (too small as you pointed out though) already and after an multi architectures support is added I'll take it from there.

marc-hb · 2025-04-04T02:36:41Z

You don't need to close this PR, sorry for the misunderstanding. You can keep it to share ideas, prototypes, anything as it's in "draft" status now. This was just general advice FYI for other, non-draft PRs

Initial support for q35 and "virt". Default to the host architecture. Progress towards supporting ARM64 as discussed in pmem#140, pmem#143 and others. Signed-off-by: Marc Herbert <[email protected]>

Progress towards supporting ARM64 as discussed in pmem#140, pmem#143 and others. Signed-off-by: Marc Herbert <[email protected]>

Progress towards supporting ARM64 as discussed in pmem#140, pmem#143 and others. Signed-off-by: Marc Herbert <[email protected]> Signed-off-by: Itaru Kitayama [email protected] Tested-by: Itaru Kitayama [email protected]

Progress towards supporting ARM64 as discussed in #140, #143 and others. Signed-off-by: Marc Herbert <[email protected]> Signed-off-by: Itaru Kitayama [email protected] Tested-by: Itaru Kitayama [email protected]

Itaru Kitayama added 8 commits March 31, 2025 08:10

Changes needed for the 64-bit Arm emulation.

add62e2

Dev commit 2.

fc3967a

Add "aavmf_path" variable in fedora_vars.sh file for aarch64.

07bd369

Always set -cpu option on aarch64 otherwise Linux won't boot.

5f451d0

Dev commit 3.

498f180

Remove dax=none kernel command line option.

d85bd7b

Don't terminate the kernel command line with a "\"".

d89f887

Pass EFI binaries to the QEMU options depending on the architecture of

801890e

the system about to boot.

ikitayama requested a review from stellarhopper as a code owner April 1, 2025 03:01

Fixes typo and add EFI path check.

db8b68f

marc-hb marked this pull request as draft April 1, 2025 22:21

marc-hb reviewed Apr 1, 2025

View reviewed changes

Itaru Kitayama added 2 commits April 2, 2025 18:38

Don't touch the GitHub Workflow files for now.

acba7d3

Revert "Changes needed for the 64-bit Arm emulation."

f006134

This reverts commit add62e2.

This was referenced Apr 2, 2025

Set machine other than "q35" if not on X86. #143

Draft

Add new aamvf_path variable in ubuntu_vars.sh file #139

Closed

The -r img option fails on arm64 #136

Open

Remove aamvf_path for now.

9139245

marc-hb mentioned this pull request Apr 9, 2025

Add initial support for arm64 "virt". Default to host architecture #160

Merged

marc-hb added a commit to marc-hb/run_qemu that referenced this pull request Apr 9, 2025

Add initial support for arm64 "virt". Default to host architecture.

cca3ad2

Progress towards supporting ARM64 as discussed in pmem#140, pmem#143 and others. Signed-off-by: Marc Herbert <[email protected]>

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Make run_qemu.sh usable on 64-bit Arm archtecutre systems #140

Make run_qemu.sh usable on 64-bit Arm archtecutre systems #140

ikitayama commented Apr 1, 2025

marc-hb left a comment

marc-hb Apr 1, 2025

ikitayama Apr 1, 2025

marc-hb Apr 8, 2025

marc-hb Apr 1, 2025

ikitayama Apr 1, 2025 •

edited

Loading

marc-hb Apr 1, 2025

ikitayama Apr 1, 2025

marc-hb Apr 1, 2025

marc-hb Apr 1, 2025

ikitayama Apr 1, 2025

marc-hb Apr 1, 2025

ikitayama Apr 1, 2025

marc-hb Apr 1, 2025

ikitayama Apr 1, 2025

marc-hb Apr 1, 2025

ikitayama Apr 1, 2025

marc-hb Apr 1, 2025

marc-hb Apr 1, 2025

ikitayama Apr 1, 2025

marc-hb commented Apr 4, 2025

ikitayama commented Apr 4, 2025

marc-hb commented Apr 4, 2025

Make run_qemu.sh usable on 64-bit Arm archtecutre systems #140

Are you sure you want to change the base?

Make run_qemu.sh usable on 64-bit Arm archtecutre systems #140

Conversation

ikitayama commented Apr 1, 2025

marc-hb left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ikitayama Apr 1, 2025 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

marc-hb commented Apr 4, 2025

ikitayama commented Apr 4, 2025

marc-hb commented Apr 4, 2025

ikitayama Apr 1, 2025 •

edited

Loading