KSUN + SUSFS + DroidSpaces — verified booting, containers working
Pre-release
peridot 6.1.175 + KernelSU-Next v3.3.0 + SUSFS v2.2.0 + DroidSpaces (LXC-style
containers, ravindu644/Droidspaces-OSS).
Built from 88c1263bc9deacb5d6933e2e7839a44336ab0f94 on peridot-6.1.175 —
verified directly: CI compiled this exact commit, the zip is a real AnyKernel3
package (Image + anykernel.sh), kernel.release =
6.1.175-android14-11-ga3b9c44908dd-ab13320413.
Two earlier builds on this feature bootlooped
Both compiled cleanly and still failed on-device. Neither failure was captured
with diagnostics at the time, so what follows is the strongest evidence
available, not a confirmed root cause:
- 1st attempt had
CONFIG_CGROUP_DEVICE/CONFIG_CGROUP_PIDSenabled, which
growenum cgroup_subsys_idand resizestruct css_set.subsys[]— measured via
hand-builtgenksyms, this shifts the exported-symbol ABI checksum for
__put_task_structand everything reachable fromtask_struct. - 2nd attempt had
CONFIG_BRIDGE_NETFILTER/CONFIG_NF_TABLESenabled, which
independently shifted the checksum for 113 of 115 exports in
kernel/sched/core.c(wake_up_process,sched_setscheduler,
set_cpus_allowed_ptr,runqueues, ...).
Removing each pair, in turn, was followed by a stable boot. That correlation is
solid — a controlled two-step bisection, not a guess. The exact mechanism is
less certain than it might sound: on-device evidence from this successful boot
(below) shows this vendor kernel tolerates ~1791 symbol-CRC disagreements as
non-fatal "...but ignore" warnings by default, so a plain "CRC mismatch means
hard module rejection" story doesn't fully hold up. Something about those two
config pairs broke this specific device's boot; a checksum shift is the
measurable fact, not a confirmed causal chain to the failure.
The CONFIG_SYSVIPC KABI relocation (scripts/droidspaces/integrate.sh) was not
the cause of either bootloop — confirmed clean via direct CRC comparison against
the boot-tested base, both times, and again in the raw log below.
Raw on-device verification
Reported via a separate Claude session with adb access to the physical device.
Commands and their literal output, unedited:
Full raw log (uname, dmesg, namespace stress test, live Alpine container)
################################################################################
# Device: Xiaomi peridot (POCO F6, SM8635) — A/B slot _b
# Kernel: Theettam / KernelSU-Next 3.3.0 + SUSFS v2.2.0 + DroidSpaces (SYSVIPC)
# Built: Fri Jul 17 15:59:37 UTC 2026 (commit 88c1263bc9de, peridot-6.1.175)
################################################################################
===================== CHECK 1 =====================
$ adb shell su -c 'uname -r'
6.1.175-android14-11-ga3b9c44908dd-ab13320413
===================== CHECK 2 =====================
$ adb shell su -c 'dmesg | grep -iE "module verification failed|invalid module format|unknown symbol"'
<-- empty: zero hard module-load rejections
$ adb shell su -c 'dmesg | grep -ic "disagrees about version"'
1791 <-- all "...but ignore" (module loads anyway);
this kernel tolerates CRC drift as non-fatal
on a large number of symbols by default
===================== CHECK 3 =====================
$ adb shell su -c 'dmesg | grep -iE "kernel panic|Oops|BUG:|Call trace|WARNING:"' | head -30
[ 1.435480] pmic_glink_debug: disagrees about version of symbol _dev_err, but ignore...
[ 1.435524] pmic_glink_debug: disagrees about version of symbol put_device, but ignore...
[ 1.438328] qcom_iommu_debug: disagrees about version of symbol _dev_err, but ignore...
[ 1.438403] qcom_iommu_debug: disagrees about version of symbol dev_driver_string, but ignore...
[ 1.503417] spmi_pmic_arb_debug: disagrees about version of symbol _dev_err, but ignore...
[ 1.503433] spmi_pmic_arb_debug: disagrees about version of symbol _dev_info, but ignore...
[ 1.503461] spmi_pmic_arb_debug: disagrees about version of symbol put_device, but ignore...
[ 1.503612] spmi_pmic_arb_debug 10b14000.qcom,spmi-debug: SPMI PMIC arbiter debug bus disabled by fuse
[ 1.507479] qti_battery_debug: disagrees about version of symbol _dev_err, but ignore...
[ 1.688899] qcom_cpufreq_hw_debug: disagrees about version of symbol _dev_err, but ignore...
[ 1.935618] mtdoops: pares mtd_dt, mem_address =0xa7000000, mem_size =0x400000
[ 1.935620] mtdoops: pares mtd_dt, pmsg_size =0x200000,console-size =0x200000
[ 1.997878] mtdoops: ready 4, 1389 (no erase)
[ 1.997881] mtdoops: Attached to MTD device 0
[ 1.997884] block2mtd: mtd0: [/dev/block/by-name/oops] erase_size = 2048KiB [2097152]
[ 4.061737] WARNING: Unprivileged eBPF is enabled, data leaks possible via Spectre v2 BHB attacks!
[ 901.295275] capability: warning: `wpa_supplicant' uses 32-bit capabilities (legacy support in use)
[ 1050.987097] [GTP-INF][brl_irq_enbale:463] warning: irq depth imbalance!
# no kernel panic, no real Oops, no real BUG: — remaining WARNINGs are stock/benign
===================== CHECK 4 (fork + namespace stress via unshare) =====================
$ adb shell su -c 'unshare -m -i -p -U -r -f /system/bin/sh /data/local/tmp/ctest.sh'
=== inside new namespaces ===
pid_in_ns=1
forked_reaped=300
proc_pids_visible_in_ns=3
sysvipc_in_ns=msg,sem,shm,
uid_in_ns=0
=== done ===
$ adb shell su -c 'ls /proc | grep -c "^[0-9][0-9]*$"' # host, for contrast
764
===================== CHECK 5 (live Alpine container "hi", PID 22215) =====================
$ adb shell su -c '/data/local/Droidspaces/bin/droidspaces --name=hi run cat /etc/os-release'
NAME="Alpine Linux"
ID=alpine
VERSION_ID=3.23.5
PRETTY_NAME="Alpine Linux v3.23"
$ adb shell su -c '/data/local/Droidspaces/bin/droidspaces --name=hi run uname -a'
Linux hi 6.1.175-android14-11-ga3b9c44908dd-ab13320413 #1 SMP PREEMPT Fri Jul 17 15:59:37 UTC 2026 aarch64 GNU/Linux
$ adb shell su -c '/data/local/Droidspaces/bin/droidspaces --name=hi run ps -ef'
UID PID PPID C STIME TTY TIME CMD
root 1 0 0 16:49 ? 00:00:00 /sbin/init
dhcpcd 132 1 0 16:49 ? 00:00:00 dhcpcd: [manager] [ip4] [ip6]
root 135 132 0 16:49 ? 00:00:00 dhcpcd: [privileged proxy]
dhcpcd 136 132 0 16:49 ? 00:00:00 dhcpcd: [network proxy]
dhcpcd 137 132 0 16:49 ? 00:00:00 dhcpcd: [control proxy]
root 168 1 0 16:49 ? 00:00:00 sshd: /usr/sbin/sshd [listener] 0 of 10-100 startups
root 175 1 0 16:49 console 00:00:00 /sbin/getty 38400 console
dhcpcd 176 135 0 16:49 ? 00:00:00 dhcpcd: [BPF ARP] eth0 172.28.36.75
root 201 0 0 16:49 ? 00:00:00 ps -ef
$ adb shell su -c 'cat /proc/22215/root/etc/alpine-release'
3.23.5
$ adb shell su -c 'for p in $(ls /proc | grep -E "^[0-9]+$"); do readlink /proc/$p/ns/pid; done | sort | uniq -c | sort -rn'
770 pid:[4026531836] <-- host PID namespace
8 pid:[4026535174] <-- container "hi" PID namespace (8 procs)
$ adb shell su -c 'for n in pid mnt net ipc uts user cgroup; do echo "$n host=$(readlink /proc/1/ns/$n) ctnr=$(readlink /proc/22215/ns/$n)"; done'
pid host=pid:[4026531836] ctnr=pid:[4026535174] # ISOLATED
mnt host=mnt:[4026532614] ctnr=mnt:[4026535264] # ISOLATED
net host=net:[4026531840] ctnr=net:[4026535175] # ISOLATED
ipc host=ipc:[4026531839] ctnr=ipc:[4026535172] # ISOLATED (SYSVIPC)
uts host=uts:[4026531838] ctnr=uts:[4026535171] # ISOLATED
user host=user:[4026531837] ctnr=user:[4026531837] # shared (allow_userns=0 by config)
cgroup host=cgroup:[4026531835] ctnr=cgroup:[4026535173] # ISOLATED
Reading it:
- No hard module-load rejection (
unknown symbol/invalid module format/
module verification failed) — empty - No panic, real Oops, or real BUG since boot
- 300 fork/exit cycles inside a PID+USER namespace under
unshare— the
task_structalloc/__put_task_structpath, zero faults - Real namespace isolation: pid/mnt/net/ipc/uts/cgroup separated from host (
ls /procshows 3 processes in-namespace vs 764 on host);usernamespace is
shared by DroidSpaces config (allow_userns=0), not a kernel limitation - A real Alpine Linux 3.23.5 container running via the DroidSpaces binary, with
init+dhcpcd+sshd+getty— confirmed by reading the container's actual
rootfs at/proc/<pid>/root/etc/alpine-release, independent of the tool's own
reporting
What this does not cover
One device, one boot slot, one test session. Confirms every vendor module that
loads on this hardware loads without a hard rejection — does not cover code
paths only exercised by workloads that weren't run, or other SoC variants.
Known limitations vs the full DroidSpaces feature list
No per-device cgroup allowlisting, no --pids-limit enforcement, no in-container
nftables/bridge-netfilter. Basic container networking (the already-enabled
veth/bridge/NAT path) is unaffected — Alpine's dhcpcd got a lease in the
test above.
Still EXPERIMENTAL. Flash with a backup and fastboot recovery ready.