Re: [mainline] Kernel OOPs while running powerpc/mce/ selftest

[Venkat Rao Bagalkote <venkat88@linux.ibm.com>]

On 05/11/25 7:15 pm, Christophe Leroy wrote:
>
>
> Le 23/10/2025 à 06:54, Venkat Rao Bagalkote a écrit :
>> Greetings!!!
>>
>>
>> IBM CI has reported a kernel crash while running mce selftests on 
>> mainline kernel, from tools/testing/selftests/powerpc/mce/.
>>
>>
>> This issue is hit when CONFIG_KASAN is enabled. If its disabled, test 
>> passes.
>>
>>
>> Traces:
>>
>>
>> [ 8041.225432] BUG: Unable to handle kernel data access on read at 
>> 0xc00e0001a1ad6103
>> [ 8041.225453] Faulting instruction address: 0xc0000000008c54d8
>> [ 8041.225461] Oops: Kernel access of bad area, sig: 11 [#1]
>> [ 8041.225467] LE PAGE_SIZE=64K MMU=Radix  SMP NR_CPUS=8192 NUMA pSeries
>> [ 8041.225475] Modules linked in: nft_fib_inet nft_fib_ipv4 
>> nft_fib_ipv6 nft_fib nft_reject_inet nf_reject_ipv4 nf_reject_ipv6 
>> nft_reject nft_ct nft_chain_nat nf_nat nf_conntrack bonding tls 
>> nf_defrag_ipv6 nf_defrag_ipv4 rfkill ip_set nf_tables nfnetlink 
>> pseries_rng vmx_crypto dax_pmem fuse ext4 crc16 mbcache jbd2 nd_pmem 
>> papr_scm sd_mod libnvdimm sg ibmvscsi ibmveth scsi_transport_srp 
>> pseries_wdt
>> [ 8041.225558] CPU: 17 UID: 0 PID: 877869 Comm: inject-ra-err Kdump: 
>> loaded Not tainted 6.18.0-rc2+ #1 VOLUNTARY
>> [ 8041.225569] Hardware name: IBM,9080-HEX Power11 (architected) 
>> 0x820200 0xf000007 of:IBM,FW1110.01 (NH1110_069) hv:phyp pSeries
>> [ 8041.225576] NIP:  c0000000008c54d8 LR: c00000000004e464 CTR: 
>> 0000000000000000
>> [ 8041.225583] REGS: c0000000fff778d0 TRAP: 0300   Not tainted 
>> (6.18.0- rc2+)
>> [ 8041.225590] MSR:  8000000000001003 <SF,ME,RI,LE>  CR: 48002828  
>> XER: 00000000
>> [ 8041.225607] CFAR: c00000000004e460 DAR: c00e0001a1ad6103 DSISR: 
>> 40000000 IRQMASK: 3
>> [ 8041.225607] GPR00: c0000000019d0598 c0000000fff77b70 
>> c00000000244a400 c000000d0d6b0818
>> [ 8041.225607] GPR04: 0000000000004d43 0000000000000008 
>> c00000000004e464 004d424900000000
>> [ 8041.225607] GPR08: 0000000000000001 18000001a1ad6103 
>> a80e000000000000 0000000003000048
>> [ 8041.225607] GPR12: 0000000000000000 c000000d0ddf3300 
>> 0000000000000000 0000000000000000
>> [ 8041.225607] GPR16: 0000000000000000 0000000000000000 
>> 0000000000000000 0000000000000000
>> [ 8041.225607] GPR20: 0000000000000000 0000000000000000 
>> 0000000000000000 0000000000000000
>> [ 8041.225607] GPR24: 0000000000000000 0000000000000000 
>> 0000000000000000 0000000000000000
>> [ 8041.225607] GPR28: c000000d0d6b0888 c000000d0d6b0800 
>> 0000000000004d43 c000000d0d6b0818
>> [ 8041.225701] NIP [c0000000008c54d8] __asan_load2+0x54/0xd8
>> [ 8041.225712] LR [c00000000004e464] pseries_errorlog_id+0x20/0x3c
>> [ 8041.225722] Call Trace:
>> [ 8041.225726] [c0000000fff77b90] [c0000000001f8748] 
>> fwnmi_get_errinfo+0xd4/0x104
>> [ 8041.225738] [c0000000fff77bc0] [c0000000019d0598] 
>> get_pseries_errorlog+0xa8/0x110
>> [ 8041.225750] [c0000000fff77c00] [c0000000001f8f68] 
>> pseries_machine_check_realmode+0x11c/0x214
>> [ 8041.225762] [c0000000fff77ce0] [c000000000049ca4] 
>> machine_check_early+0x74/0xc0
>> [ 8041.225771] [c0000000fff77d30] [c0000000000084a4] 
>> machine_check_early_common+0x1b4/0x2c0
>
> Is it a new problem or has it always been there ?


Its not a new problem. I have enabled KASAN recently in the config, and 
then I started seeing this issue.

I have tested on 6.17, 6.16 and 6.15 kernels and issues is there all along.


Regards,

Venkat.

>
> The problem is because KASAN is not compatible with realmode (MMU 
> translation is OFF).
>
> pseries_machine_check_realmode() is located in 
> arch/powerpc/platforms/pseries/ras.c built with KASAN_SANITIZE_ras.o := n
>
> But pseries_machine_check_realmode() calls mce_handle_error() which 
> calls get_pseries_errorlog().
>
> get_pseries_errorlog() is in arch/powerpc/kernel/rtas.c which is _not_ 
> built with KASAN_SANITIZE disabled hence the Oops.
>
> Unrelated, but it looks like there is also a problem with commit 
> cc15ff327569 ("powerpc/mce: Avoid using irq_work_queue() in 
> realmode"), which removed the re-enabling of translation but left the 
> call to mce_handle_err_virtmode().
>
> Christophe
>