Update — 7+ days stable with retbleed=off

  Confirming that adding retbleed=off to boot parameters has completely
resolved the crashes.

  Current status:

  - Uptime: 7 days, 16 hours (and counting)
  - Kernel: 7.0.0-27-generic (Ubuntu 26.04 LTS)
  - CPU: AMD Ryzen 7 4800U (Zen2)
  - Zero crashes, zero dmesg errors related to srso/retbleed/rethunk

  Boot parameters that stabilized the system:

  processor.max_cstate=1 spec_rstack_overflow=off retbleed=off
nmi_watchdog=1

  Timeline of fixes attempted:

  1. processor.max_cstate=1 alone → still crashed (srso_safe_ret)
  2. Added spec_rstack_overflow=off → crash moved to retbleed_return_thunk
  3. Added retbleed=off → stable since Jul 15 (7+ days, previously crashed 
every 12-36h)

  Conclusion: The return thunk mitigations (both SRSO and RETBleed)
interact badly with cpuidle on Zen2 (Renoir/4800U), specifically in the
swapper/0 → cpuidle_enter_state → sched_tick → timer interrupt path.
Disabling both mitigations eliminates the crashes entirely.

  This is on a desktop/home-server workload (no VMs, no containers
requiring mitigation isolation), so the security trade-off is acceptable
for this user. However, the kernel should not crash regardless of
mitigation state — this appears to be a code-generation or thunk-
alignment bug in the mitigation implementation for this CPU family.

-- 
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2160457

Title:
   Kernel panic in srso_safe_ret / kick_ilb during sched_tick on AMD
  Ryzen 7 4800U (Zen2)

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2160457/+subscriptions


-- 
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs

Reply via email to