Re: Very unreliable booting (WARNING: CPU: 0 PID: 1 at kernel/context_tracking.c:215 ct_nmi_exit+0xa0/0xc0) with PPC_EARLY_DEBUG_G5 set on a PowerMac G5, kernel 6.7-rc6

On Tue, 9 Jan 2024 16:40:13 +1100
Rohan McLure [off-list ref] wrote:

On 12/21/23 09:20, Erhard Furtner wrote:

quoted

Greetings!

I wanted to check whether there are any changes on issue https://lore.kernel.org/all/20231114003721.4a9bfd37@yea/T/ (local) on kernel 6.7-rc. KCSAN enabled kernels still won't boot on this G5 it seems.  

Hi Erhard,

Apologies for the late response on both of your queries. I believe that
this patch should allow you to boot with KCSAN:
Link:
https://lore.kernel.org/all/20231127054648.1205221-4-rmclure@linux.ibm.com/ (local)

I wasn't able to replicate the Floating Point unavailable panic in qemu,
which puts into question whether this patches addresses the problem that
is presenting for you, or another probabilistic bug impeding startup.

Hi Rohan,

thanks for looking into this!

The patch applies on top of v6.7 but unfortunately it does not work out. The machine still only very rarely boots up with PPC_EARLY_DEBUG_G5 enabled. Does not seem to matter much whether KSCAN, KUEP, KUAP are enabled or disabled.

I got one successful boot with KSCAN, KUEP, KUAP all disabled. Here again dmesg of the booted machine shows this "WARNING: CPU: 0 PID: 1 at kernel/context_tracking.c:215 ct_nmi_exit+0xa0/0xc0" and

NIP [c00000000001d7fc] real_readb+0x44/0x68
LR [c000000000070c58] udbg_real_scc_putc+0x38/0x80

--- interrupt: 900

[c0000000031038f0] [c000000002641294] 0xc000000002641294 (unreliable)
[c000000003103920] [c000000000070dc0] udbg_adb_putc+0x30/0x50
[c000000003103940] [c00000000001cd84] udbg_puts+0x64/0xb0
[c000000003103970] [c000000000e6011c] udbg_progress+0x18/0x30
[c000000003103990] [c000000000072420] smp_core99_kick_cpu+0x140/0x180
[c000000003103a10] [c000000000030d1c] __cpu_up+0x12c/0x3c0
[c000000003103ad0] [c00000000009f2d8] bringup_cpu+0x68/0x200
[c000000003103b30] [c00000000009de40] cpuhp_invoke_callback+0x170/0x300
[c000000003103b80] [c0000000000a0808] _cpu_up.constprop.0+0x308/0x730
[c000000003103c10] [c0000000000a0d5c] cpu_up+0x12c/0x180
[c000000003103ca0] [c000000000e6d368] bringup_nonboot_cpus+0x7c/0xec
[c000000003103cf0] [c000000000e73f58] smp_init+0x40/0xa4
[c000000003103d50] [c000000000e5b300] kernel_init_freeable+0x188/0x358
[c000000003103de0] [c00000000000d138] kernel_init+0x30/0x158
[c000000003103e50] [c00000000000b4c4] ret_from_kernel_user_thread+0x14/0x1c

--- interrupt: 0 at 0x0

Code: 39200000 38600004 f9280010 4bfffd4d e92d0128 38210020 81290000 e8010010 7c0803a6 4e800020 60000000 60000000 <0fe00000> 4bffff9c 60000000 60000000


Looks like this is one of the hard-to-track down bugs... ;)

Kernel .config and full dmesg attached.

Regards,
Erhard