[Mellanox] [Smartswitch] Kill stale initramfs udevd process before updating udevd rules#26343
Merged
liat-grozovik merged 2 commits intosonic-net:masterfrom Mar 25, 2026
Merged
Conversation
Signed-off-by: Hemanth Kumar Tirupati <tirupatihemanthkumar@gmail.com>
Signed-off-by: Hemanth Kumar Tirupati <tirupatihemanthkumar@gmail.com>
Collaborator
|
/azp run Azure.sonic-buildimage |
|
Azure Pipelines successfully started running 1 pipeline(s). |
Contributor
There was a problem hiding this comment.
Pull request overview
This PR addresses SmartSwitch reboot instability on DPU-based platforms by preventing a stale initramfs systemd-udevd process from surviving switch_root and crashing when udev rules are updated in the real root filesystem.
Changes:
- Added
kill_stale_udevd()to detect and terminate leftover initramfssystemd-udevd --daemonprocesses before writing DPU udev rules. - Updated the script header (shebang and copyright/SPDX metadata).
dgsudharsan
approved these changes
Mar 23, 2026
arlakshm
pushed a commit
to arlakshm/sonic-buildimage
that referenced
this pull request
Apr 1, 2026
…dating udevd rules (sonic-net#26343) - Why I did it On SONiC SmartSwitch platforms with DPUs, systemd-udevd crashes with SIGABRT on every reboot when DPU firmware initialization is slow. During the initramfs boot phase, a standalone systemd-udevd daemon is started to handle device discovery. If DPU firmware takes longer than the 60-second udevadm settle timeout (BlueField-3 DPUs can take 120 seconds each in the failure case when they are stuck), the initramfs cannot stop this udevd before switch_root. The stale process survives into the real system but is never chrooted into the overlayfs root, leaving it with a broken filesystem view. When dpu-udev-manager.sh writes udev rules, the stale udevd detects the change and crashes on an assertion in systemd's chase() path resolution (assert(path_is_absolute(p)) at chase.c:648), because dir_fd_is_root() returns false for a process whose root still points to the initramfs rootfs rather than the overlayfs. This triggers a systemd issue : systemd/systemd#29559 which maintainers doesn't consider as a bug from systemd side. Raising this fix for our usecase. Core was generated by `/usr/lib/systemd/systemd-udevd --daemon --resolve-names=never'. Program terminated with signal SIGABRT, Aborted. #0 0x00007f29fe7f695c in ?? () from /lib/x86_64-linux-gnu/libc.so.6 (gdb) bt #0 0x00007f29fe7f695c in ?? () from /lib/x86_64-linux-gnu/libc.so.6 sonic-net#1 0x00007f29fe7a1cc2 in raise () from /lib/x86_64-linux-gnu/libc.so.6 sonic-net#2 0x00007f29fe78a4ac in abort () from /lib/x86_64-linux-gnu/libc.so.6 sonic-net#3 0x00007f29fea50c11 in ?? () from /usr/lib/x86_64-linux-gnu/systemd/libsystemd-shared-257.so sonic-net#4 0x00007f29feb94a8b in chase () from /usr/lib/x86_64-linux-gnu/systemd/libsystemd-shared-257.so sonic-net#5 0x00007f29feb956e2 in chase_and_opendir () from /usr/lib/x86_64-linux-gnu/systemd/libsystemd-shared-257.so sonic-net#6 0x00007f29feb9a609 in conf_files_list_strv () from /usr/lib/x86_64-linux-gnu/systemd/libsystemd-shared-257.so sonic-net#7 0x00007f29fea913e8 in config_get_stats_by_path () from /usr/lib/x86_64-linux-gnu/systemd/libsystemd-shared-257.so sonic-net#8 0x0000559f295519cf in ?? () sonic-net#9 0x0000559f29553a77 in ?? () sonic-net#10 0x00007f29fec36055 in ?? () from /usr/lib/x86_64-linux-gnu/systemd/libsystemd-shared-257.so sonic-net#11 0x00007f29fec3668d in sd_event_dispatch () from /usr/lib/x86_64-linux-gnu/systemd/libsystemd-shared-257.so sonic-net#12 0x00007f29fec394a8 in sd_event_run () from /usr/lib/x86_64-linux-gnu/systemd/libsystemd-shared-257.so sonic-net#13 0x00007f29fec396c7 in sd_event_loop () from /usr/lib/x86_64-linux-gnu/systemd/libsystemd-shared-257.so sonic-net#14 0x0000559f29545820 in ?? () sonic-net#15 0x00007f29fe78bca8 in ?? () from /lib/x86_64-linux-gnu/libc.so.6 sonic-net#16 0x00007f29fe78bd65 in __libc_start_main () from /lib/x86_64-linux-gnu/libc.so.6 sonic-net#17 0x0000559f29545c51 in ?? () - How I did it Added a kill_stale_udevd() function to dpu-udev-manager.sh that runs before writing the udev rules. It identifies the systemd-managed udevd PID via systemctl show, then kills any other systemd-udevd --daemon process that doesn't match -- these are leftover initramfs instances. If no stale process exists (e.g. DPUs are healthy and the initramfs udevd exited cleanly), the function is a no-op. - How to verify it Deploy the image on a SmartSwitch with DPUs in a state where firmware initialization times out (>60s per DPU) by stopping image installation before firmware install step Reboot the switch Verify no new systemd-udevd coredumps in /var/core/ Verify the stale process was killed: journalctl -b 0 | grep dpu-udev-manager should show killing stale initramfs udevd PID (systemd udevd is PID ) Verify systemd-udevd.service is healthy: systemctl status systemd-udevd should show active (running) Verify DPU udev rules were written: cat /etc/udev/rules.d/92-midplane-intf.rules should contain the DPU interface naming rules Signed-off-by: Hemanth Kumar Tirupati <tirupatihemanthkumar@gmail.com> Signed-off-by: arlakshm <arlakshm@microsoft.com>
Collaborator
|
Cherry-pick PR to 202511: #26573 |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Why I did it
On SONiC SmartSwitch platforms with DPUs, systemd-udevd crashes with SIGABRT on every reboot when DPU firmware initialization is slow. During the initramfs boot phase, a standalone systemd-udevd daemon is started to handle device discovery. If DPU firmware takes longer than the 60-second udevadm settle timeout (BlueField-3 DPUs can take 120 seconds each in the failure case when they are stuck), the initramfs cannot stop this udevd before switch_root. The stale process survives into the real system but is never chrooted into the overlayfs root, leaving it with a broken filesystem view. When dpu-udev-manager.sh writes udev rules, the stale udevd detects the change and crashes on an assertion in systemd's chase() path resolution (assert(path_is_absolute(p)) at chase.c:648), because dir_fd_is_root() returns false for a process whose root still points to the initramfs rootfs rather than the overlayfs.
This triggers a systemd issue : systemd/systemd#29559 which maintainers doesn't consider as a bug from systemd side. Raising this fix for our usecase.
How I did it
Added a kill_stale_udevd() function to dpu-udev-manager.sh that runs before writing the udev rules. It identifies the systemd-managed udevd PID via systemctl show, then kills any other systemd-udevd --daemon process that doesn't match -- these are leftover initramfs instances. If no stale process exists (e.g. DPUs are healthy and the initramfs udevd exited cleanly), the function is a no-op.
How to verify it