ec4678f6 (build 6bbb994) fails to boot with zygisk enabled

[alphapuggle]

ec478f6 should fix it.

artifact will be generated here. would appreciate if anyone running magisk on grapheneos could test this.

Originally posted by @pixincreate in #30

Enabling zygisk from this artifact causes the device (Pixel 8 Pro Graphene 2026071101) to fail to boot. It never gives up even after like half an hour, but won't turn back on unless powered back on in safe mode.

Unfortunately I have no logs to give you, unless there's a method for getting logcat in boot on Graphene

(Other issue is locked)

[pixincreate]

do you have any modules enabled? safe booting worked? you've the bootloader unlocked, right? you flashed the init_boot.img, right?

one hunch that i have is the memd which is fixed in this branch. that commit is not in https://github.com/pixincreate/Magisk/commit/ec478f6acb8806bb86f7eb12c36a4f8e53379b98 since i'm waiting for upstream to fix it.

can you follow below attached bug report template and try to fetch the logs?

## READ BEFORE OPENING ISSUES

All bug reports require you to **USE DEBUG BUILDS**. Please include the version name and version code in the bug report.

If you experience a bootloop, attach a `dmesg` (kernel logs) when the device refuse to boot. This may very likely require a custom kernel on some devices as `last_kmsg` or `pstore ramoops` are usually not enabled by default. In addition, please also upload the result of `cat /proc/mounts` when your device is working correctly **WITHOUT MAGISK**.

If you experience issues during installation, in recovery, upload the recovery logs, or in Magisk, upload the install logs. Please also upload the `boot.img` or `recovery.img` that you are using for patching.

If you experience a crash of Magisk app, dump the full `logcat` **when the crash happens**.

If you experience other issues related to Magisk, upload `magisk.log`, and preferably also include a boot `logcat` (start dumping `logcat` when the device boots up)

**DO NOT** open issues regarding root detection.

**DO NOT** ask for instructions.

**DO NOT** report issues if you have any modules installed.

Without following the rules above, your issue will be closed without explanation.

also, consider checking these:

https://xdaforums.com/t/guide-january-3-2024-root-pixel-8-pro-unlock-bootloader-pass-safetynet-both-slots-bootable-more.4638510/
https://xdaforums.com/t/july-7-2026-cp2a-260705-006-global-unlock-bootloader-root-pixel-8-pro-husky-stable-firmware.4633839/

[alphapuggle]

Tried with both modules enabled & disabled with no changes.
Safe booting disabled zygisk and was able to finish booting after rebooting back into normal mode
Updated via Custota, which should flash init_boot.img on update.

Pulled the artifact from that branch (https://github.com/pixincreate/Magisk/actions/runs/28891775617) and was able to boot with zygisk enabled, but says "no" in the preview and throws a bunch of zygisk64: Cannot hook method: nativeForkAndSpecialize ILIILILILLLLZLLZZLLZZZ

magisk_log_2026-07-14T13.15.15.log

[pixincreate]

yes. that is older one. try this build and let me know?

https://github.com/pixincreate/Magisk/actions/runs/29260066494

[alphapuggle]

Ah, that is the one I originally tried; I didn't realize the changes from the other branch were already in main

[pixincreate]

i checked the attached log. it is from 31aebdfe, the older memfd-only build, not the newer 6bbb994/ec478f6 signature build. it reaches boot-complete and confirms that the older build cannot match grapheneos’s current nativeForkAndSpecialize signature.

the newer build contains the matching jni signatures but does not contain the memfd selinux patch, and we currently have no logs from its failed boot. recent grapheneos changes also introduced a separate zygote_compat process without changing the jni descriptors again.

the next useful test is a build containing both changes (artifact link below), tested with all magisk modules removed. before rebooting, please run adb wait-for-device logcat -b all -v threadtime > graphene-boot.log so a bootloop log can be captured.

https://github.com/pixincreate/Magisk/actions/runs/29400247149

[alphapuggle]

Yes sorry I didn't make it quite clear which one I was running to get that log. I could only grab that because it successfully booted.

I tried the new artifact and it still refuses to boot, but afterwards I found & loaded a module to dump the boot logs to disk

Attached are logs for 5e658f0

debug-boot-logcat.log
debug-boot-dmesg.log

[pixincreate]

thanks for the logs!

the updated jni signature is matching, and the memfd selinux rule is present.

i have a hunch that the exec-spawned processes specialize the current process without forking, while zygisk currently performs its own fork beforehand which might be causing graphene to replay the command to enter an invalid parent-process path and terminate with IllegalStateException. not exactly sure..
i don't have much experience here either. have asked my ai bot to look after the grapheneos recent code changes in the framework base and compare that with zygisk.

if my hunch turns out to be true then, i’ll have to add a guarded in-place specialization path for grapheneos exec spawning while retaining the normal fork path for backward compatibility.

in the mean time, can you try disabling exec spawning for the app and see (this app level option is available in 2026071101, not sure about older versions) if that helps?

[artgunz]

It didn't help me with the same issue (latest grapheneos, magisk app from pull #34 ) for Pixel9a

BTW: exec spawning disable for magisk app cause to 'spawn' that message many times, instead of one time before:
'zygisk64: Cannot hook method: nativeForkAndSpecialize ILIILILILLLLZLLZZLLZZZ'

Thanks in advance for helping, @pixincreate

FYI: also I tried the #35 build - no luck as well

[pixincreate]

thanks for testing. a couple of clarifications:

pr #35 only adds ci monitoring and it does not contain the zygisk fix from PR #34. if you direct-installed #35 after #34, please reinstall #34.

please install the #34 apk, use magisk → install → direct Install (or repatch the boot image), reboot, and confirm the native build:

adb shell su -c 'magisk -c; magisk -v; magisk -V'
adb shell getprop ro.build.version.incremental

the magisk log should identify commit e251a34a. please also attach the complete /cache/magisk.log.

the ILI... message confirms jni hook registration failed, but that displayed signature is art’s non-standard format, so the full log is needed to identify the mismatch.

graphene’s newly announced changes do not modify the monitored jni signatures. once their new release is available, please retest #34 on it as well. disabling exec spawning (as i suggested before) only applies to the magisk app does not disable it system-wide.

[alphapuggle]

Doesn't bootloop with e7ec5dd (#34 artifact) but zygisk doesn't enable either. Didn't see anything in the logs about it even attempting to load here. Tested 2026071101 and 2026071500 (01 security update wasn't available when testing)

Download.zip