Re: [Linux PPC] Disable PREEMPT

On Thu, Sep 03, 2026 at 10:52:59PM +0530, Shrikanth Hegde wrote:

quoted

quoted

If possible run against current upstream and share the results.

https://github.com/openSUSE/kernel-source/blob/6824496d1801f73def615dca8794202eeb7b0d86/config/ppc64le/default

[  472.091531][ T6181] Kernel panic - not syncing: stack-protector: Kernel stack is corrupted in: kvmhv_run_single_vcpu+0x19d4/0x1b50 [kvm_hv]
[  472.091598][ T6181] CPU: 29 UID: 107 PID: 6181 Comm: CPU 112/KVM Not tainted 7.2.2-5.g6824496-default #1 PREEMPT(full) openSUSE Tumbleweed (unreleased)  61871a5f06863b4006f5ef27cd9f18e8a7a2edad
[  472.091612][ T6181] Hardware name: IBM,9824-42A Power11 (architected) 0x820200 0xf000007 of:IBM,FW1110.20 (OB1110_130) hv:phyp pSeries
[  472.091624][ T6181] Call Trace:
[  472.091630][ T6181] [c00000000fdfb680] [c00000000134ce90] dump_stack_lvl+0x84/0xc0 (unreliable)
[  472.091653][ T6181] [c00000000fdfb6b0] [c00000000022e7c8] vpanic+0x324/0x5e4
[  472.091666][ T6181] [c00000000fdfb760] [c00000000022eac4] do_panic_on_target_cpu+0x0/0x2c
[  472.091677][ T6181] [c00000000fdfb780] [c0000000013c1ff8] __stack_chk_fail+0x48/0x60
[  472.091689][ T6181] [c00000000fdfb7f0] [c00800001aae219c] kvmhv_run_single_vcpu+0x19d4/0x1b50 [kvm_hv]
[  472.091712][ T6181] [c00000000fdfb940] [c00800001aae24b4] kvmppc_vcpu_run_hv+0x19c/0x12f0 [kvm_hv]
[  472.091732][ T6181] [c00000000fdfba10] [c00800001aeeed18] kvmppc_vcpu_run+0x30/0x48 [kvm]
[  472.091779][ T6181] [c00000000fdfba30] [c00800001aee9ef4] kvm_arch_vcpu_ioctl_run+0x35c/0x4a0 [kvm]
[  472.091813][ T6181] [c00000000fdfbac0] [c00800001aedaac4] kvm_vcpu_ioctl+0x1ac/0xad8 [kvm]
[  472.091844][ T6181] [c00000000fdfbca0] [c0000000007f1244] sys_ioctl+0x374/0x1060
[  472.091857][ T6181] [c00000000fdfbdb0] [c00000000002f7f8] system_call_exception+0x188/0x430
[  472.091871][ T6181] [c00000000fdfbe50] [c00000000000cfdc] system_call_vectored_common+0x15c/0x2ec
[  472.091886][ T6181] ---- interrupt: 3000 at 0x7fffb5565fac
[  472.091896][ T6181] NIP:  00007fffb5565fac LR: 00007fffb5565fac CTR: 0000000000000000
[  472.091904][ T6181] REGS: c00000000fdfbe80 TRAP: 3000   Not tainted  (7.2.2-5.g6824496-default)
[  472.091911][ T6181] MSR:  800000000280f033 <SF,VEC,VSX,EE,PR,FP,ME,IR,DR,RI,LE>  CR: 42044402  XER: 00000000
[  472.091938][ T6181] IRQMASK: 0
[  472.091938][ T6181] GPR00: 0000000000000036 00007fbfa77ed7a0 00007fffb5677100 00000000000000fa
[  472.091938][ T6181] GPR04: 000000002000ae80 0000000000000000 0000000000000000 0000000000000000
[  472.091938][ T6181] GPR08: 00000000000000fa 0000000000000000 0000000000000000 0000000000000000
[  472.091938][ T6181] GPR12: 0000000000000000 00007fbfa77f5ec0 000000014676f000 00007fbfa77ee7c0
[  472.091938][ T6181] GPR16: 000000014674e8d0 00007fbfa77eeec0 00007fbfa77eeec0 fffffffffffffff7
[  472.091938][ T6181] GPR20: 00007fffb71210d0 0000000000000001 00007fbfa77eeec0 0000000000000000
[  472.091938][ T6181] GPR24: 00007fbfa77ed8e8 0000000105971428 000000002000ae80 0000000105f77a70
[  472.091938][ T6181] GPR28: 0000000000000000 0000000000000000 000000002000ae80 000000014674f000
[  472.092020][ T6181] NIP [00007fffb5565fac] 0x7fffb5565fac
[  472.092027][ T6181] LR [00007fffb5565fac] 0x7fffb5565fac
[  472.092033][ T6181] ---- interrupt: 3000
[  472.098256][ T6181] pstore: backend (nvram) writing error (-1)

This is the host, cannot run the kernel as guest because it fails to boot most
of the time inside KVM.

quoted

quoted

Nonethless, there are quite a few platforms. Originally no preemption
was the only option, and that's the reason why many people run that.
It's the conservative, known working option. And that's the reason a lot
of platfrom code does not get tested with more aggressive preemtion
models, and never gets fixed to work with them.

Full preemption has been there for many years!.

Possible for years, forced only recently.

quoted

Lazy is not that aggressive compared to that.

quoted

Simply disabling the no preemtion option does not make the platform code
ready.

Let's understand your crash case. Let's see where it is going wrong. I am suspecting
it is some wrong usage of preemption api rather than arch can't support preemption.

Very likely some wrong use of the preemption API by the arch code, or no
use where it should have been used. It did not matter so long as people
could run their no preempt configs and ignore the problem.

Thanks

Friendly LLM analysis says preemption is enabled too early and before
completing kvmppc_handle_exit_hv() / kvmppc_handle_nested_exit()

Below is ONLY a speculation and completely UNTESTED.
Maybe worth a try.

The patch is munged by the e-mail client, and it causes immediate
voluntary preemprion in rcu critical section and hard lockup on starting
a KVM VM.

Also it would be sort of bad news if it worked because that would be
specific to book3s KVM HV and would not help with the KVM HV from the
original report which likely is not book3s, nor with KVM PR.

Thanks

Michal