| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <aXj28ZAsPhp7s5sm@kernel.org>
Date: Tue, 27 Jan 2026 19:33:37 +0200
From: Mike Rapoport <rppt@...nel.org>
To: Jordan Richards <jordanrichards@...gle.com>
Cc: Pasha Tatashin <pasha.tatashin@...een.com>,
Pratyush Yadav <pratyush@...nel.org>, Shuah Khan <shuah@...nel.org>,
Jason Miu <jasonmiu@...gle.com>,
David Matlack <dmatlack@...gle.com>, linux-mm@...ck.org,
linux-kernel@...r.kernel.org, linux-kselftest@...r.kernel.org
Subject: Re: [PATCH] selftests/liveupdate: add end to end test infrastructure
and scripts
On Thu, Jan 22, 2026 at 09:44:27PM +0000, Jordan Richards wrote:
> From: Pasha Tatashin <pasha.tatashin@...een.com>
>
> Add the end to end testing infrastructure required to verify the
> liveupdate feature. This includes a custom init process, a test
> orchestration script, and a batch runner.
>
> The framework consists of:
>
> init.c:
> A lightweight init process that manages the kexec lifecycle.
> It mounts necessary filesystems, determines the current execution
> stage (1 or 2) via the kernel command line, and handles the
> kexec_file_load() sequence to transition between kernels.
>
> luo_test.sh:
> The primary KTAP-compliant test driver. It handles:
> - Kernel configuration merging and building.
> - Cross-compilation detection for x86_64 and arm64.
> - Generation of the initrd containing the test binary and init.
> - QEMU execution with automatic accelerator detection (KVM, HVF,
> or TCG).
>
> run.sh:
> A wrapper script to discover and execute all `luo_*.c`
> tests across supported architectures, providing a summary of
> pass/fail/skip results.
>
> Signed-off-by: Pasha Tatashin <pasha.tatashin@...een.com>
> Co-developed-by: Jordan Richards <jordanrichards@...gle.com>
> Signed-off-by: Jordan Richards <jordanrichards@...gle.com>
> ---
> Changelog since luo v7 [1]:
> - Build test binaries with `-nostdlib -nostdinc`
> - Use minimal per-arch config instead of defconfig
> - Unhandled test errors now cause the test to fail instead of skip
>
> [1] https://lore.kernel.org/all/20251122222351.1059049-20-pasha.tatashin@soleen.com/
...
> diff --git a/tools/testing/selftests/liveupdate/.gitignore b/tools/testing/selftests/liveupdate/.gitignore
> index 661827083ab6..7dc1e8aec44c 100644
> --- a/tools/testing/selftests/liveupdate/.gitignore
> +++ b/tools/testing/selftests/liveupdate/.gitignore
> @@ -6,4 +6,6 @@
> !*.sh
> !.gitignore
> !config
> +!config.aarch64
> +!config.x86_64
> !Makefile
Hmm, I missed it when tools/testing/selftests/liveupdate/ was posted.
I'm not a huge fun of negative logic in .gitignore.
Why can't we just exclude the patterns we don't want to track?
> +static int kexec_load(void)
> +{
> + char cmdline[COMMAND_LINE_SIZE];
> + int kernel_fd, initrd_fd, err;
> + ssize_t len;
> + int fd;
> +
> + fd = open("/proc/cmdline", O_RDONLY);
> + if (fd < 0) {
> + fprintf(stderr, "INIT: Failed to read /proc/cmdline\n");
> +
> + return -1;
> + }
> +
> + len = read(fd, cmdline, sizeof(cmdline) - 1);
> + close(fd);
> + if (len < 0)
> + return -1;
> +
> + cmdline[len] = 0;
> + if (len > 0 && cmdline[len - 1] == '\n')
> + cmdline[len - 1] = 0;
> +
> + strncat(cmdline, " luo_stage=2", sizeof(cmdline) - strlen(cmdline) - 1);
> +
> + kernel_fd = open(KERNEL_IMAGE, O_RDONLY);
> + if (kernel_fd < 0) {
> + fprintf(stderr, "INIT: Failed to open kernel image\n");
> + return -1;
> + }
> +
> + initrd_fd = open(INITRD_IMAGE, O_RDONLY);
> + if (initrd_fd < 0) {
> + fprintf(stderr, "INIT: Failed to open initrd image\n");
> + close(kernel_fd);
> + return -1;
> + }
> +
> + err = kexec_file_load(kernel_fd, initrd_fd, strlen(cmdline) + 1,
> + cmdline, 0);
> +
> + close(initrd_fd);
> + close(kernel_fd);
> +
> + return err ? : 0;
Just return err?
> diff --git a/tools/testing/selftests/liveupdate/luo_test.sh b/tools/testing/selftests/liveupdate/luo_test.sh
> new file mode 100755
> index 000000000000..90ecb16e87bb
> --- /dev/null
> +++ b/tools/testing/selftests/liveupdate/luo_test.sh
...
> +function detect_cross_compile() {
> + local target=$1
> + local host=$(uname -m)
> +
This function works fine if you run luo_test.sh directly or have cross
compilers named the way it expects in $PATH.
But if I run
CROSS_COMPILE=~/cross/gcc-13.2.0-nolibc/aarch64-linux/bin/aarch64-linux- ./run.sh
on x86, x86 tests fail
> + if [ -n "$CROSS_COMPILE" ]; then
> + return
> + fi
> +
> + [[ "$host" == "arm64" ]] && host="aarch64"
> + [[ "$target" == "arm64" ]] && target="aarch64"
> +
> + if [[ "$host" == "$target" ]]; then
> + CROSS_COMPILE=""
> + return
> + fi
> +
> + local candidate=""
> + case "$target" in
> + aarch64) candidate="aarch64-linux-gnu-" ;;
> + x86_64) candidate="x86_64-linux-gnu-" ;;
> + *) skip "Auto-detection for target '$target' not supported. Please set CROSS_COMPILE manually." ;;
> + esac
> +
> + if command -v "${candidate}gcc" &> /dev/null; then
> + CROSS_COMPILE="$candidate"
> + else
> + skip "Compiler '${candidate}gcc' not found. Please install it (e.g., 'apt install gcc-aarch64-linux-gnu') or set CROSS_COMPILE."
> + fi
> +}
> +
> +function build_kernel() {
> + local build_dir=$1
> + local make_cmd=$2
> + local kimage=$3
> + local target_arch=$4
> +
> + local kconfig="$build_dir/.config"
> + local common_conf="$test_dir/config"
> + local arch_conf="$test_dir/config.$target_arch"
> +
> + echo "# Building kernel in: $build_dir"
> +
> + local fragments=()
> +
> + if [[ -f "$common_conf" ]]; then
> + fragments+=("$common_conf")
> + fi
> +
> + if [[ -f "$arch_conf" ]]; then
> + fragments+=("$arch_conf")
> + fi
I think the common and arch config fragments are required and we can just
assign fragments directly and run merge_config.sh.
> +
> + if [[ ${#fragments[@]} > 1 ]]; then
> + echo $fragments
> + "$kernel_dir/scripts/kconfig/merge_config.sh" \
> + -Q -m -O "$build_dir" "${fragments[@]}" >> /dev/null
> + else
> + cp ${fragments[0]} $kconfig
> + fi
> + cat $kconfig
> +
> + $make_cmd olddefconfig
> + $make_cmd "$kimage"
> + $make_cmd headers_install INSTALL_HDR_PATH="$headers_dir"
> +}
...
> +function run_qemu() {
> + local qemu_cmd=$1
> + local cmdline=$2
> + local kernel_path=$3
> + local serial="$workspace_dir/qemu.serial"
> +
> + local accel="-accel tcg"
> + local host_machine=$(uname -m)
> +
> + [[ "$host_machine" == "arm64" ]] && host_machine="aarch64"
> + [[ "$host_machine" == "x86_64" ]] && host_machine="x86_64"
> +
> + if [[ "$qemu_cmd" == *"$host_machine"* ]]; then
> + if [ -w /dev/kvm ]; then
> + accel="-accel kvm"
> + fi
> + fi
Do we care that much about qemu warnings about invalid accelerator to have
this logic here?
-accel kvm -accel hvf -accel tcg
seems to cover all bases.
> +
> + cmdline="$cmdline liveupdate=on panic=-1"
> +
...
> diff --git a/tools/testing/selftests/liveupdate/luo_test_utils.c b/tools/testing/selftests/liveupdate/luo_test_utils.c
> index 3c8721c505df..7ee80b6ed4cb 100644
> --- a/tools/testing/selftests/liveupdate/luo_test_utils.c
> +++ b/tools/testing/selftests/liveupdate/luo_test_utils.c
> @@ -13,6 +13,7 @@
> #include <getopt.h>
> #include <fcntl.h>
> #include <unistd.h>
> +#include <sys.h>
This breaks running normal make:
luo_test_utils.c:16:10: fatal error: sys.h: No such file or directory
16 | #include <sys.h>
| ^~~~~~~
NOLIBC specific includes and calls should be guarded with #ifdef NOLIBC
> #include <sys/ioctl.h>
> #include <sys/syscall.h>
> #include <sys/mman.h>
> @@ -21,8 +22,20 @@
> #include <errno.h>
> #include <stdarg.h>
>
> +#include <linux/unistd.h>
> +
> #include "luo_test_utils.h"
>
> +int sys_ftruncate(int fd, off_t length)
> +{
> + return my_syscall2(__NR_ftruncate, fd, length);
> +}
> +
> +int ftruncate(int fd, off_t length)
> +{
> + return __sysret(sys_ftruncate(fd, length));
> +}
These should be added to nolibc I suppose.
> +
,,,
> diff --git a/tools/testing/selftests/liveupdate/run.sh b/tools/testing/selftests/liveupdate/run.sh
> new file mode 100755
> index 000000000000..3f6b29a26648
> --- /dev/null
> +++ b/tools/testing/selftests/liveupdate/run.sh
> @@ -0,0 +1,68 @@
> +#!/bin/bash
> +# SPDX-License-Identifier: GPL-2.0
> +
> +OUTPUT_DIR="results_$(date +%Y%m%d_%H%M%S)"
I don't think that putting the results in the current directory rather than
in SCRIPT_DIR or in an explicitly named directory is a good idea.
> +SCRIPT_DIR=$(dirname "$(realpath "$0")")
> +TEST_RUNNER="$SCRIPT_DIR/luo_test.sh"
> +
> +TARGETS=("x86_64" "aarch64")
> +
> +GREEN='\033[0;32m'
> +RED='\033[0;31m'
> +YELLOW='\033[1;33m'
> +NC='\033[0m'
> +
> +PASSED=()
> +FAILED=()
> +SKIPPED=()
> +
> +mkdir -p "$OUTPUT_DIR"
> +
> +TEST_NAMES=()
> +while IFS= read -r file; do
> + TEST_NAMES+=("$(basename "$file" .c)")
> +done < <(find "$SCRIPT_DIR" -maxdepth 1 -name "luo_*.c" ! -name "luo_test_utils.c")
I don't like name based detection of tests. Listing them explicitly seems a
viable option.
> +
> +if [ ${#TEST_NAMES[@]} -eq 0 ]; then
> + echo "No tests found in $SCRIPT_DIR"
> + exit 1
> +fi
> +
--
Sincerely yours,
Mike.
Powered by blists - more mailing lists