-
Notifications
You must be signed in to change notification settings - Fork 0
Fedora Recipe
Historical reference. This is the retained September 25, 2026 bring-up record, not a new hardware result. Canonical source at fb50ccee853009703d3a68dad147e74ed245b19b. The original snapshot table below supplies this reference's hardware and stack metadata. Only the manifest-enumerated Markdown links and the single
HOWTOcheckout path (the correction already documented in the repository README) are adapted. All other source bytes, including commands and limitations, are retained. Installation orientation | Current recovery authority | Home
For experienced Linux bring-up users: stock non-UEFI Bianbu/U-Boot on a SpacemiT K3 CoM260 IFX/Firefly carrier, Fedora on NVMe, a locally built dldo4-patched host, and the persistent Fedora guest used for our interactive QEMU/KVM sessions.
Every U-Boot choice below is temporary. Never run env save or saveenv.
Reset returns to the board's existing saved boot path, which may be Bianbu
rather than either Fedora kernel.
This is bring-up material, not a production or hostile-guest isolation claim. Keep the bootstrap host on a trusted, isolated LAN and change its shipped credentials before exposing it to another network.
| Item | Recorded value |
|---|---|
| Hardware | CoM260/Firefly, 8 GiB RAM; WD Red SN700 500 GB in the 2280 slot, imaged through a USB NVMe enclosure |
| Firmware | Bianbu Minimal K3 v4.0.1, non-UEFI U-Boot |
| Bootstrap | Fedora 44 Omni, July 31 image, 7.1.5-201.0.riscv64.omni.fc44.riscv64
|
| KVM host | 7.3.0-rc4-k3-kvm-host-a1 |
| QEMU | Fedora 10.2.2-1.fc44
|
| Guest | Fedora 44 Cloud, June 4 image, 6.19.8-200.0.riscv64.fc43.riscv64, 4 vCPUs / 4 GiB |
This is the September 25, 2026 setup, not a "latest" recommendation. Image
checksums are enforced below and by the guest preparer. The pinned host
source, 14-patch order, config, and refresh rules are in
files/kernel/README.md.
Use Bash and simple paths without spaces. Commands identify their machine;
variables do not cross SSH sessions or tmux windows. These are interactive
steps, not one script: stop on any failed check rather than pasting onward.
Do not enable set -u in the interactive Fedora shell; its prompt hook uses
an unset variable.
Have Git, SSH, tmux, picocom, curl, xz, zstd, dtc, util-linux, udevadm, GNU
coreutils, and partprobe (parted) available before preparing the SSD.
Build and guest-preparation dependencies are listed in their sections.
# workstation
export WORK="$HOME/k3-fedora-howto"
mkdir -p "$WORK"
git clone https://github.com/xjamesmorris/k3-com260-info.git "$WORK/k3-com260-info"
export HOWTO="$WORK/k3-com260-info/k3-com260-fedora-howto"
export SERIAL=/dev/serial/by-id/usb-REPLACE_WITH_YOUR_ADAPTER
export NVME_DISK=/dev/disk/by-id/usb-REPLACE_WITH_YOUR_NVME
export K3_SSH=fedora@k3-host.example
export GUEST_KEY="$HOME/.ssh/k3-fedora-guest"
cd "$HOWTO"Replace the placeholders. NVME_DISK must be the enclosure's stable whole
disk ID. K3_SSH must be an explicit fedora@operator-chosen-host; the
transfer helper intentionally uses that image account and /home/fedora.
Connect a 3.3 V USB serial adapter to RX, TX, and ground only. Use 115200 8N1, no flow control:
# workstation
tmux new-session -s k3-serial
export SERIAL=/dev/serial/by-id/usb-REPLACE_WITH_YOUR_ADAPTER
picocom --baud 115200 --flow n "$SERIAL"Confirm that Bianbu 4.0.1 U-Boot appears and autoboot can be interrupted. Keep physical reset or power-cycle access. If this works, do not reflash firmware or erase UFS. Only for actual recovery, use the public K3-CoM260 recovery guide; factory flashing erases NOR and UFS, including the saved environment, but does not erase NVMe.
Download the pinned image; its checksum is checked again inside the write step:
# workstation
set -o pipefail
export OMNI_IMAGE="$WORK/Fedora-Server-Host-Omni-44-20260731.0.riscv64.raw.xz"
curl --fail --location --output "$OMNI_IMAGE" https://dl.fedoraproject.org/pub/alt/risc-v/release/44/Server/riscv64/images/Fedora-Server-Host-Omni-44-20260731.0.riscv64.raw.xz
printf '%s %s\n' bcbf6fcadc506b4b09ea46c88bf461ce3fb31bd54f125da3604969c3c06d0923 "$OMNI_IMAGE" |
sha256sum --check --strict -
xz --test -- "$OMNI_IMAGE"The write below destroys the selected disk. Positively identify the stable ID, model, size, serial, and USB transport:
# workstation
export NVME_NODE
NVME_NODE=$(readlink -e -- "$NVME_DISK")
test -b "$NVME_NODE"
test "$(lsblk -dnro TYPE "$NVME_NODE")" = disk
test "$(lsblk -dnro TRAN "$NVME_NODE")" = usb
ls -l "$NVME_DISK"
lsblk -o NAME,PATH,SIZE,MODEL,SERIAL,TRAN,FSTYPE,MOUNTPOINTS "$NVME_NODE"
udevadm info --query=property --name="$NVME_NODE" | grep -E '^(ID_MODEL|ID_SERIAL|ID_WWN|ID_BUS)='
swapon --show=NAME,TYPE,SIZEUnmount each listed filesystem with sudo umount /dev/ACTUAL_PARTITION;
disable any target-disk swap with sudo swapoff /dev/ACTUAL_SWAP_PARTITION.
Use the actual devices shown above, not those literal example names.
Do not disconnect or move the enclosure between inspection and writing.
Run the following block as a unit. Its subshell stops on a failed identity,
mount/swap, or checksum check before dd; strict mode does not leak into the
interactive shell. lsblk includes active swap as [SWAP].
# workstation
(
set -euo pipefail
set -x
: "${NVME_DISK:?}" "${NVME_NODE:?}" "${OMNI_IMAGE:?}"
test "$(readlink -e -- "$NVME_DISK")" = "$NVME_NODE"
test -b "$NVME_NODE"
test "$(lsblk -dnro TYPE "$NVME_NODE")" = disk
test "$(lsblk -dnro TRAN "$NVME_NODE")" = usb
mountpoints=$(lsblk -nro MOUNTPOINTS "$NVME_NODE")
if [[ "$mountpoints" =~ [^[:space:]] ]]; then
printf 'Refusing an in-use disk (mounts or swap):\n%s\n' "$mountpoints" >&2
exit 1
fi
printf '%s %s\n' bcbf6fcadc506b4b09ea46c88bf461ce3fb31bd54f125da3604969c3c06d0923 "$OMNI_IMAGE" |
sha256sum --check --strict -
xz --test -- "$OMNI_IMAGE"
lsblk -dn -o PATH,SIZE,MODEL,SERIAL,TRAN "$NVME_NODE"
xz --decompress --stdout -- "$OMNI_IMAGE" |
sudo dd of="$NVME_NODE" bs=16M oflag=direct conv=fsync status=progress
)Only after that block succeeds, discover the new partitions:
# workstation
sync
sudo partprobe "$NVME_NODE"
udevadm settle
export NVME_BOOT="${NVME_DISK}-part2" NVME_ROOT="${NVME_DISK}-part3"
test -b "$NVME_BOOT" && test -b "$NVME_ROOT"
lsblk -o NAME,PATH,SIZE,FSTYPE,UUID,MOUNTPOINTS "$NVME_NODE"
export ROOT_UUID
ROOT_UUID=$(sudo blkid -s UUID -o value "$NVME_ROOT")
test -n "$ROOT_UUID"
printf 'NVMe root UUID: %s\n' "$ROOT_UUID"The recorded layout is partition 1 ESP, partition 2 ext4 /boot, and
partition 3 Btrfs root with subvolume root. Use the UUID just discovered,
never a historical value.
Mount /boot, inspect its BLS entry, and select matching boot files:
# workstation
export BOOT_MNT="$WORK/omni-boot"
mkdir -p "$BOOT_MNT"
sudo mount -o rw "$NVME_BOOT" "$BOOT_MNT"
sudo find "$BOOT_MNT/loader/entries" -maxdepth 1 -type f -name '*.conf' \
-exec sed -n '/^\(title\|version\|linux\|initrd\|options\) /p' {} +
export BOOT_KREL=7.1.5-201.0.riscv64.omni.fc44.riscv64
export VMLINUX="$BOOT_MNT/vmlinuz-$BOOT_KREL"
export INITRAMFS="$BOOT_MNT/initramfs-$BOOT_KREL.img"
export DTB="$BOOT_MNT/dtb-$BOOT_KREL/spacemit/k3-com260-ifx.dtb"
export RAW_IMAGE="$BOOT_MNT/Image-$BOOT_KREL"
export FIXED_DTB="$BOOT_MNT/dtb-$BOOT_KREL/spacemit/k3-com260-bootstrap.dtb"
test -f "$VMLINUX" && test -f "$INITRAMFS" && test -f "$DTB"
sudo sha256sum "$VMLINUX" "$INITRAMFS" "$DTB" > "$WORK/omni-original-boot.sha256"Stop if the BLS entry or paths do not match this pinned kernel. Fedora protects BLS entries and initramfs files; their inspection needs privilege.
The packaged vmlinuz is a zstd EFI zboot container; booti needs its raw
RISC-V Image. Parse and bound the little-endian payload:
# workstation
test "$(od -An -tx1 -N8 "$VMLINUX" | tr -d '[:space:]')" = 4d5a00007a696d67
test "$(od -An -tx1 -j24 -N5 "$VMLINUX" | tr -d '[:space:]')" = 7a73746400
OFF=$(od --endian=little -An -tu4 -j8 -N4 "$VMLINUX" | tr -d '[:space:]')
SIZE=$(od --endian=little -An -tu4 -j12 -N4 "$VMLINUX" | tr -d '[:space:]')
RAW_SIZE=$(od --endian=little -An -tu4 -j "$((OFF + SIZE))" -N4 "$VMLINUX" | tr -d '[:space:]')
test "$OFF" -ge 4096
test "$SIZE" -gt 0
test "$((OFF + SIZE + 4))" -le "$(stat -c %s "$VMLINUX")"
PAYLOAD="$WORK/$BOOT_KREL.zstd"
RAW_TMP="$WORK/Image-$BOOT_KREL"
test ! -e "$PAYLOAD" && test ! -e "$RAW_TMP" && test ! -e "$RAW_IMAGE"
dd if="$VMLINUX" of="$PAYLOAD" bs=1M iflag=skip_bytes,count_bytes skip="$OFF" count="$SIZE" status=progress
zstd --test --quiet -- "$PAYLOAD"
zstd --decompress --stdout --quiet -- "$PAYLOAD" > "$RAW_TMP"
test "$(od -An -tx1 -j48 -N8 "$RAW_TMP" | tr -d '[:space:]')" = 5249534356000000
HEADER_SIZE=$(od --endian=little -An -tu8 -j16 -N8 "$RAW_TMP" | tr -d '[:space:]')
test "$(stat -c %s "$RAW_TMP")" -eq "$RAW_SIZE"
test "$HEADER_SIZE" -eq "$RAW_SIZE"
sudo install -m 0644 -- "$RAW_TMP" "$RAW_IMAGE"Leave the original DTB untouched. Decompile it, add
regulator-always-on; inside both dldo4 and the root-level hub-reset
node, then build a separately named bootstrap DTB:
# workstation
DTS="$WORK/k3-com260-bootstrap.dts"
DTB_TMP="$WORK/k3-com260-bootstrap.dtb"
test ! -e "$DTS" && test ! -e "$DTB_TMP" && test ! -e "$FIXED_DTB"
dtc -I dtb -O dts -o "$DTS" "$DTB"
grep -n -E 'dldo4|hub-reset|regulator-always-on' "$DTS"
${EDITOR:-vi} "$DTS"
dtc -I dts -O dtb -o "$DTB_TMP" "$DTS"
dtc -I dtb -O dts "$DTB_TMP" | grep -n -E 'dldo4|hub-reset|regulator-always-on'
sudo install -m 0644 -- "$DTB_TMP" "$FIXED_DTB"Inspect the regenerated DTS and confirm the property in both nodes. A DTB round trip is not byte-identical, and this successful two-property workaround does not prove both changes independently necessary. The later source-built host carries the isolated dldo4 fix plus PCIe support, not the unrelated USB/hub series. Without the workaround, bring-up hit a hard lockup around unused-regulator cleanup, roughly 30-45 seconds into boot; an early login prompt is not enough.
Record actual decimal and U-Boot hexadecimal sizes, then unmount:
# workstation
for file in "$FIXED_DTB" "$RAW_IMAGE" "$INITRAMFS"; do
bytes=$(stat -c %s "$file")
printf '%s: %u bytes, %x hex\n' "$(basename "$file")" "$bytes" "$bytes"
done | tee "$WORK/bootstrap-load-sizes.txt"
sync
sudo umount "$BOOT_MNT"
test -z "$(findmnt -rn -S "$NVME_BOOT" -o TARGET)"Install the unmounted SSD in the carrier's 2280 slot; this recipe assumes it is the only NVMe device. Remove microSD cards for this route: non-vendor SD layouts stopped the stock BootROM before serial output during bring-up.
Interrupt U-Boot, replace the UUID, enter one line at a time, and compare all
three reported load sizes with bootstrap-load-sizes.txt. Before loading,
use the printed addresses and file sizes to check that the three memory
ranges [address, address + size) do not overlap. An empty 2230 slot may
report PCIE-2: Link down; that is not failure of the occupied 2280 slot.
# U-Boot
printenv fdt_addr_r kernel_addr_r ramdisk_addr_r bootcmd
setenv bootrel 7.1.5-201.0.riscv64.omni.fc44.riscv64
setenv rootuuid REPLACE_WITH_DISCOVERED_UUID
nvme scan
load nvme 0:2 ${fdt_addr_r} dtb-${bootrel}/spacemit/k3-com260-bootstrap.dtb
printenv fileaddr filesize
load nvme 0:2 ${kernel_addr_r} Image-${bootrel}
printenv fileaddr filesize
load nvme 0:2 ${ramdisk_addr_r} initramfs-${bootrel}.img
setenv initrd_size ${filesize}
printenv fileaddr filesize initrd_size
setenv bootargs "root=UUID=${rootuuid} rootflags=subvol=root clk_ignore_unused console=ttyS0,115200n8"
printenv bootargs
booti ${kernel_addr_r} ${ramdisk_addr_r}:${initrd_size} ${fdt_addr_r}
Stop before booti on an undefined address, failed load, size mismatch,
continuation prompt, malformed command line, or wrong UUID. Save nothing.
The pinned image shipped with fedora / linux. On the isolated LAN, log in
over serial, run passwd, record
sudo ssh-keygen -lf /etc/ssh/ssh_host_ed25519_key.pub, and inspect
ip -brief address, uname -r, and findmnt /. Install a normal workstation
public key for fedora, verify the SSH host key against the serial
fingerprint, and test $K3_SSH; never disable host-key checking. Do not
launch guests on this old Omni kernel.
Use normal kernel build dependencies, Clang/LLD and LLVM tools, GNU
riscv64-linux-gnu- binutils, dtc, pahole, Git, and rsync. No earlier
build is required. git am needs your committer identity. Allow disk space
for a full Linux history plus the separate build:
# workstation
export KERNEL_REPO="$WORK/linux-kvm-riscv"
export HOST_OUTPUT="$WORK/k3-kvm-host-a1"
test ! -e "$KERNEL_REPO" && test ! -e "$HOST_OUTPUT"
git clone https://github.com/kvm-riscv/linux.git "$KERNEL_REPO"
test -z "$(git -C "$KERNEL_REPO" status --porcelain)"
git -C "$KERNEL_REPO" var GIT_COMMITTER_IDENT
cd "$HOWTO"
OUTPUT_ROOT="$HOST_OUTPUT" LOCAL_TAG=-k3-kvm-host-a1 JOBS="$(nproc)" \
./scripts/build-k3-kvm-host.sh "$KERNEL_REPO" "$HOWTO/files/kernel/config.seed"
test "$(cat "$HOST_OUTPUT/stage/kernelrelease")" = 7.3.0-rc4-k3-kvm-host-a1
(cd "$HOST_OUTPUT/stage" && sha256sum --check --strict SHA256SUMS)The helper fetches the exact public commit into a private ref, creates a
detached worktree under the fresh output root, applies all 14 patches, and
verifies the prepared tree. It builds with
ARCH=riscv LLVM=1 LLVM_IAS=0 CROSS_COMPILE=riscv64-linux-gnu-; GNU as
avoids the tested module relocation problem, while split module BTF remains
disabled because the tested Clang/pahole output was rejected by this kernel.
If the object or tree cannot be reproduced, stop rather than advancing a ref.
Transfer from the workstation and install locally on the K3:
# workstation
export HOST_SUFFIX=k3-kvm-host-a1
cd "$HOWTO"
./scripts/transfer-omni-baseline-to-k3.sh "$HOST_OUTPUT/stage" "$K3_SSH" "$HOST_SUFFIX"# K3 bootstrap host
export HOST_SUFFIX=k3-kvm-host-a1
export HOST_STAGE="/home/fedora/kernel-stage-$HOST_SUFFIX"
sudo "$HOST_STAGE/k3-install-manual-kernel.sh" "$HOST_STAGE" "$HOST_SUFFIX"The installer adds the new kernel beside the running one and preserves
/boot/dtb. Derive the live root UUID and actual candidate sizes:
# K3 bootstrap host
export TESTREL=7.3.0-rc4-k3-kvm-host-a1
ROOT_SOURCE=$(findmnt -nro SOURCE /)
ROOT_DEVICE=${ROOT_SOURCE%%\[*}
ROOT_UUID=$(sudo blkid -s UUID -o value "$ROOT_DEVICE")
test -n "$ROOT_UUID"
printf 'root UUID: %s\n' "$ROOT_UUID"
for file in "/boot/dtb-$TESTREL/spacemit/k3-com260-ifx.dtb" "/boot/Image-$TESTREL" "/boot/initramfs-$TESTREL.img"; do
bytes=$(stat -c %s "$file")
printf '%s: %u bytes, %x hex\n' "$file" "$bytes" "$bytes"
doneAt an attended reset, interrupt U-Boot and select the matching installed
triple. Check non-overlapping load ranges again and compare each load with
the sizes above. boot_mode=nor retains the recorded vendor token; Linux
may report it as an unknown parameter.
# U-Boot
printenv fdt_addr_r kernel_addr_r ramdisk_addr_r bootcmd
setenv testrel 7.3.0-rc4-k3-kvm-host-a1
setenv rootuuid REPLACE_WITH_DISCOVERED_UUID
nvme scan
load nvme 0:2 ${fdt_addr_r} dtb-${testrel}/spacemit/k3-com260-ifx.dtb
printenv fileaddr filesize
load nvme 0:2 ${kernel_addr_r} Image-${testrel}
printenv fileaddr filesize
load nvme 0:2 ${ramdisk_addr_r} initramfs-${testrel}.img
setenv initrd_size ${filesize}
printenv fileaddr filesize initrd_size
setenv bootargs "root=UUID=${rootuuid} rootflags=subvol=root clk_ignore_unused console=ttyS0,115200n8 boot_mode=nor"
printenv bootargs
booti ${kernel_addr_r} ${ramdisk_addr_r}:${initrd_size} ${fdt_addr_r}
Again, save nothing and select this kernel manually after every reset. Once Fedora returns:
# K3 KVM host
test "$(uname -r)" = 7.3.0-rc4-k3-kvm-host-a1
sudo dnf install qemu-system-riscv-core qemu-img python3 rsync tmux
findmnt -no SOURCE,FSTYPE,OPTIONS /
sudo dmesg --color=never | grep -iE 'hypervisor extension|G-stage|VMID|AIA'
test -c /dev/kvm
rpm -q qemu-system-riscv-core qemu-img
/usr/bin/qemu-system-riscv64 --version
test -r /dev/kvm && test -w /dev/kvmThe recorded endpoint used distro QEMU 10.2.2-1.fc44. Run it as fedora;
if /dev/kvm is owned by group kvm but inaccessible, use
sudo usermod -aG kvm fedora and log in again. Otherwise inspect the normal
udev/group policy; never make the device world-writable.
The workstation needs qemu-img, libguestfs virt-copy-out,
cloud-localds, Python 3, curl, zstd, and ssh-keygen. Select or create
a key, retaining its private half only on the workstation:
# workstation
export GUEST_KEY="$HOME/.ssh/k3-fedora-guest"
test -f "$GUEST_KEY" && test -f "$GUEST_KEY.pub" || ssh-keygen -t ed25519 -f "$GUEST_KEY"
export GUEST_NAME=k3-fedora-44-20260604-a1
export GUEST_STAGE="$WORK/$GUEST_NAME"
test ! -e "$GUEST_STAGE"
cd "$HOWTO"
./scripts/prepare-riscv-fedora-guest.sh "$GUEST_STAGE" "$GUEST_KEY.pub"
(cd "$GUEST_STAGE" && sha256sum --check --strict SHA256SUMS)The fresh stage contains the pinned base, NoCloud seed, validated raw guest
Image, initramfs, and a persistent overlay.qcow2 whose relative backing
file is base.img. Transfer the complete tree with sparse-file handling:
# workstation
ssh "$K3_SSH" "test ! -e '/home/fedora/$GUEST_NAME' && test ! -e '/home/fedora/$GUEST_NAME.partial' && mkdir -m 0700 '/home/fedora/$GUEST_NAME.partial'"
rsync -aS --info=progress2 -- "$GUEST_STAGE/" "$K3_SSH:/home/fedora/$GUEST_NAME.partial/"
rsync -a -- "$HOWTO/scripts/k3-run-fedora-guest.sh" "$K3_SSH:/home/fedora/"
ssh "$K3_SSH" "set -eu; cd '/home/fedora/$GUEST_NAME.partial'; sha256sum --check --strict SHA256SUMS; cd /home/fedora; mv -- '$GUEST_NAME.partial' '$GUEST_NAME'"Do not rename or separate base.img and overlay.qcow2.
Start tmux new-session -s k3-fedora-guest on the K3. These CPU overrides and
kernel arguments are compatibility workarounds, not tuning preferences.
Use the entire environment for this host/QEMU/guest combination: the helper's
generic bare-host defaults are not the validated Fedora guest recipe.
Leave Sstc enabled as recorded. Re-enter the block in every launch shell:
# K3 KVM host, inside tmux
export RUNNER=/home/fedora/k3-run-fedora-guest.sh
export GUEST_DIR=/home/fedora/k3-fedora-44-20260604-a1
export K3_FEDORA_AIA=plic
export K3_FEDORA_CPU='host,svpbmt=false,zicbom=false,zicbop=false,zicboz=false'
export K3_FEDORA_VCPUS=4 K3_FEDORA_MEMORY=4G K3_FEDORA_SSH_PORT=2222
export K3_FEDORA_EXTRA_APPEND='earlycon=sbi unaligned_scalar_speed=slow unaligned_vector_speed=unsupported'
test "$(uname -r)" = 7.3.0-rc4-k3-kvm-host-a1 &&
"$RUNNER" run "$GUEST_DIR" /usr/bin/qemu-system-riscv64Keep these settings, including enabled Sstc, on subsequent launches. The
recorded interactive session used resume:
# K3 KVM host, after a clean shutdown
test "$(uname -r)" = 7.3.0-rc4-k3-kvm-host-a1 &&
"$RUNNER" resume "$GUEST_DIR" /usr/bin/qemu-system-riscv64resume boots the same persistent disk; it does not restore a VM snapshot.
For status from another K3 shell:
# K3 KVM host, second shell
~/k3-run-fedora-guest.sh status ~/k3-fedora-44-20260604-a1The guest's port 22 is exposed only on K3 loopback port 2222. Tunnel from one workstation shell, then connect from another with the same seed identity:
# workstation; keep running
export K3_SSH=fedora@k3-host.example
ssh -N -o ExitOnForwardFailure=yes -o ForwardAgent=no \
-L 127.0.0.1:2222:127.0.0.1:2222 "$K3_SSH"# workstation, second shell
export GUEST_KEY="$HOME/.ssh/k3-fedora-guest"
ssh -o IdentitiesOnly=yes -o HostKeyAlias=k3-fedora-44-20260604-a1 \
-i "$GUEST_KEY" -p 2222 fedora@127.0.0.1Never use StrictHostKeyChecking=no. Verify the K3 key over serial. For the
new guest key, use a trusted inspection path or record an explicit
trust-on-first-use decision over this verified tunnel. Do not copy the guest
private key to the K3 or assume agent forwarding.
# guest
sudo dnf install fastfetch
fastfetch
systemd-detect-virt --vm
nproc
findmnt -no SOURCE,FSTYPE,OPTIONS /
sudo journalctl -k -b | grep -iE 'SBI specification|SBI implementation|earlycon|PLIC|IMSIC|APLIC'fastfetch is a summary, not proof of KVM. The helper requires -accel kvm
and /dev/kvm; it never falls back to software emulation.
The seeded fedora account is password-locked. From authenticated guest SSH,
run sudo passwd fedora only if serial login is wanted. Prefer
sudo systemctl poweroff from that SSH session, let foreground QEMU exit,
then use resume. Do not use Ctrl-a x, QMP quit, SIGKILL, or force-kill
for normal shutdown.
The optional runner stop action needs a usable guest identity or agent on
the K3; it does not acquire the workstation's private key. After PLIC has
booted and shut down cleanly, AIA is the tested alternative:
# K3 KVM host; other exported settings unchanged
test "$(uname -r)" = 7.3.0-rc4-k3-kvm-host-a1 &&
K3_FEDORA_AIA=aplic-imsic "$RUNNER" resume "$GUEST_DIR" /usr/bin/qemu-system-riscv64Keep the Image, initramfs, DTB, modules, and release name matched. Packaged kernel updates need another raw-Image extraction and deliberate U-Boot selection; they do not update this direct-boot path automatically.
Display, external USB, fan/thermal policy, suspend, VFIO, and long-duration stability are outside this recipe. Other carriers, UEFI/libvirt, and newer image/kernel/toolchain combinations are alternatives, not validated substitutions. The historical stages ran on hardware; this portable bundle has not been rerun end to end. License: GPL-2.0-only.