Cannot boot with retbleed=stuff added as a kernel parameter

[aghpub]

I'm running Arch Linux. With the new 6.2.1 kernel call depth tracking can be enabled by adding retbleed=stuff as a kernel paramter. If I do this the system is unable to boot and hangs with the following messages:

[......] FAT-fs (sdb1): IO charset iso8859-1 not found.
[FAILED] Failed to mount /efi.
[DEPEND] Dependency failed for Local File Systems.

There is a ~80s long pause before the "FAT-fs (sbd1)..." message above. Without adding the parameter the system boots and runs perfectly fine.

I tested the same on the vanilla 6.2.1 kernel and had no problems there.

In both cases /proc/config.gz contains CONFIG_CALL_DEPTH_TRACKING=y.

Additional info:

CPU: i5-6600 (Skylake)
GPU: GTX 1070 (Nvidia)
Arch package: linux-zen 6.2.1.zen1-1

Steps to reproduce:

Install the Zen kernel. Add retbleed=stuff to the kernel parameters. Reboot.

[heftig]

A patch for call depth tracking was added to the Arch and ZEN kernels as part of https://bugs.archlinux.org/task/77601.

Does reverting 7e4b3d7 help?

[aghpub]

Wow, compiling the kernel literally took hours on my crappy machine. I'm definitely not the right guy· for this job.

Anyway. reverting didn't help. It crashed the machine during boot even harder than with the patch applied.

I'll just do without this retbleed=stuff thing. No problem.

[aghpub]

There is a bug visible in journalctl if I boot with retbleed=stuff:

BUG: unable to handle page fault for address: ffffffffc4b7fff6

See: https://pastebin.com/Gy5Xrjqn

(Line: 1002ff)

[heftig]

Is it crashing loading the nvidia module?

[aghpub]

Is it crashing loading the nvidia module?

I'm a complete noob when it comes to debugging kernel problems. It seems it's crashing loading a module but I see no hint of Nvidia in that region of the log and the Nvidia module seems to be loaded before (around line 717). But as I said: I have no real clue.when it comes to the kernel.

[aghpub]

The problem persists with linux-zen 6.2.2.zen1-1

[damentz]

Per reference above, the missing patch was applied and is in the main branch. Next release of Zen Kernel and Liquorix will include it.

[An4cl3t0]

Hi everyone, I still have Kernel Panic booting from Arch with 6.3.5-zen1-1-zen with retbleed=stuff, using i5-6200u, no nvidia or proprietary drivers loaded.

My boot config:

"root="UUID=XXXXXXXXXXXXXXXXXX" rw resume="UUID=XXXXXXXXXXXXXXXXXXX" mem_sleep_default=s2idle libata.force=2:3.0G apparmor=1 retbleed=stuff lsm=landlock,lockdown,yama,integrity,apparmor,bpf intel_iommu=on intel_pstate=hwp_only randomize_kstack_offset=1 audit=1 audit_backlog_limit=32768 tsc=reliable clocksource=tsc nowatchdog nmi_watchdog=0"

In journalctl I have:

kernel:#PF: error_code(0x0000) - not-present page
kernel: #PF: supervisor read access in kernel mode
kernel: BUG: unable to handle page fault for address: ffffffffc0bd6ff6

Works without retbleed=stuff obv. Thanks for your help

[heftig]

I think that matches what I've seen with the previous Arch kernel. The patch isn't enough to get retbleed=stuff working and since it didn't get accepted upstream, I dropped it again.

[damentz]

@heftig I'm thinking at this point retbleed=stuff is not a supportable option. I'm ok just closing out this issue, or we can put in a zen-sauce tweak to remove support for it through code. What do you think?

[heftig]

I don't think a patch removing it is worth it. We can just drop the incomplete fix again in the 6.4 cycle.

[damentz]

This issue is stale and I believe no longer relevant, closing.