Thread (78 messages) flat view 78 messages, 10 authors, 2h ago

Re: [Linux PPC] Disable PREEMPT

From: Christian Zigotzky <hidden>
Date: 2026-09-23 05:10:15

On 21/09/26 07:12, Shrikanth Hegde wrote:
quoted hunk ↗ jump to hunk
Hi Christian.

On 9/20/26 6:34 PM, Christian Zigotzky wrote:
quoted
On 09/20/2026 01:49 PM, Christian Zigotzky wrote:
quoted
On 09/18/2026 09:10 AM, Michal Suchánek wrote:
quoted
On Fri, Sep 18, 2026 at 04:39:57AM +0200, Christian Zigotzky wrote:
quoted
On 17/09/26 16:15, Shrikanth Hegde wrote:
quoted
Hi Michal,
quoted
quoted
Give it a try.

I am suspecting both lockup, stack corruptions are secondary
symptoms here.
While recreate, I have different type of traces.
Yes, it does work for the crash I see.
Thanks. Give me sometime to see further on that. Didn't get any time
today.
quoted
However, we hijacked the report
of KVM locking up on FSL Cyrus+, and that is likely a different 
problem
still not addressed.
Ah, that's what you meant. Yes, i still don't know where FSL 
Cyrus+ is
stuck.

I would need help from FSL Cyrus+ to get some logs or is there a 
way we
can spin of
a QEMU guest of it? We should have the log to start somewhere.
It's difficult to get a log because it freezes and can't write any 
error
messages to the log files. This bug is very old. We had this issue 
some
years ago but we was able to disable PREEMPT because we don't need 
it.
I would expect serial console and sysrq could give something, 
depending
on how much of the kernel is frozen.

Thanks

Michal
Here is a serial log: 
https://github.com/user-attachments/files/32433586/e5500_preempt_putty.log
Thanks Michal for helping to get the traces.
quoted
Another serial log with pcie_ports=compat: 
https://github.com/user-attachments/files/32434487/e5500_preempt_pcie_ports_compat_putty.log 
I see it is mostly busy with printing
WARNING: at .enable_kernel_fp+0x30/0x78, CPU#3: qemu-system-ppc/4884

The path seems to be same also.

[c000000085ca7700] [c00000000005de84] .kvmppc_load_guest_fp+0x30/0x80
[c000000085ca7780] [c00000000005f2a0] .kvmppc_handle_exit+0x5bc/0x5cc
[c000000085ca7830] [c00000000006204c] .kvmppc_resume_host+0xb8/0x10c


Does below help to boot or make it stable?
---
 arch/powerpc/kvm/booke.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)
diff --git a/arch/powerpc/kvm/booke.c b/arch/powerpc/kvm/booke.c
index 13ad4cf5fa71..d4be7a306bfc 100644
--- a/arch/powerpc/kvm/booke.c
+++ b/arch/powerpc/kvm/booke.c
@@ -1404,10 +1404,17 @@ int kvmppc_handle_exit(struct kvm_vcpu *vcpu, 
unsigned int exit_nr)
         if (s <= 0)
             r = (s << 2) | RESUME_HOST | (r & RESUME_FLAG_NV);
         else {
-            /* interrupts now hard-disabled */
+            /*
+             * interrupts now hard-disabled. i.e EE=0
+             * But irqs_disabled relies on PACA entries
+             * and fixup below enables it.
+             * So irqs_disabled == false
+             */
             kvmppc_fix_ee_before_entry();
+            preempt_disable();
             kvmppc_load_guest_fp(vcpu);
             kvmppc_load_guest_altivec(vcpu);
+            preempt_enable_no_resched();
         }
     }
Hi,

Thanks a lot for your patch!

I patched the 7.3-rc4 kernel sources with your new patch and compiled 
the kernel again yesterday. [1]

Unfortunately, the issue still exists.

I tested it with a guest kernel with and without preempt support.

Cheers,
Christian

[1] https://github.com/chzigotzky/kernels/releases/tag/v7.3.0-rc4-2-preempt

-- 
Sent with BrassMonkey 34.3.2.1 (https://github.com/chzigotzky/Web-Browsers-and-Suites-for-Linux-PPC/releases/tag/BrassMonkey_34.3.2.1)

Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help