Re: [linux-next-20250212] syscall kexec_file_load not available

[Sourabh Jain <sourabhjain@linux.ibm.com>]

On 14/02/25 12:38, Hari Bathini wrote:
>
>
> On 14/02/25 12:15 pm, Sourabh Jain wrote:
>> Hello Hari,
>>
>>
>> On 14/02/25 12:02, Hari Bathini wrote:
>>>
>>>
>>> On 13/02/25 8:34 pm, Venkat Rao Bagalkote wrote:
>>>> Greetings!!!
>>>>
>>>>  From kernel next-20250210, I am observing syscall kexec_file_load 
>>>> not available, there by kdump service is failing to start.
>>>>
>>>>
>>>> Logs:
>>>>
>>>> [root@ltc-zzci-1 ~]# kexec -p --initrd=/boot/initramfs-6.14.0-rc2- 
>>>> next-20250212kdump.img /boot/vmlinuz-6.14.0-rc2-next-20250212 -c
>>>> Warning: append= option is not passed. Using the first kernel root 
>>>> partition
>>>> Modified cmdline: elfcorehdr=311424K root=UUID=b5b1f89c- 
>>>> d479-48b3-90e2-744a2fd05667
>>>> [root@ltc-zzci-1 ~]# kexec -p --initrd=/boot/initramfs-6.14.0-rc2- 
>>>> next-20250212kdump.img /boot/vmlinuz-6.14.0-rc2-next-20250212 -s
>>>> syscall kexec_file_load not available.
>>>> [root@ltc-zzci-1 ~]# kexec -v
>>>> kexec-tools 2.0.27
>>>> [root@ltc-zzci-1 ~]# uname -r
>>>> 6.14.0-rc2-next-20250212
>>>>
>>>
>>> Is the kernel built with CONFIG_KEXEC_FILE ?
>>
>> I am able to reproduce it with CONFIG_KEXEC_FILE enabled.
>>
>> Seems like there is something went wrong in next-20250210 and 
>> next-20250212.
>>
>> kexec -p --initrd=/boot/initramfs-6.14.0-rc2-next-20250210kdump.img / 
>> boot/vmlinuz-6.14.0-rc2-next-20250210 -d -s
>>
>> Try gzip decompression.
>> Try LZMA decompression.
>> [ 3375.712319] kexec_file: kernel: 00000000e539303c kernel_size: 
>> 0x2cdacf0
>> [ 3375.717022] ima: kexec measurement buffer for the loaded kernel at 
>> 0x0.
>> [ 3375.717076] kexec_elf: Loaded the kernel at 0x0
>> [ 3375.717094] kexec_elf: Loaded purgatory at 0x0
>> [ 3375.717104] Loaded the backup region at 0x0
>> [ 3375.717130] crash_core: Crash PT_LOAD ELF header. 
>> phdr=000000004720e656 vaddr=0xc000000000000000, paddr=0x0, sz=0x10000 
>> e_phnum=18 p_offset=0x0
>> [ 3375.717156] crash_core: Crash PT_LOAD ELF header. 
>> phdr=0000000005eb3f14 vaddr=0xc000000000010000, paddr=0x10000, 
>> sz=0xfff0000 e_phnum=19 p_offset=0x10000
>> [ 3375.717174] crash_core: Crash PT_LOAD ELF header. 
>> phdr=000000000ec70071 vaddr=0xc00000001ec20000, paddr=0x1ec20000, 
>> sz=0x13e0000 e_phnum=20 p_offset=0x1ec20000
>> [ 3375.717192] crash_core: Crash PT_LOAD ELF header. 
>> phdr=00000000b66c9c25 vaddr=0xc000000050000000, paddr=0x50000000, 
>> sz=0x3b0000000 e_phnum=21 p_offset=0x50000000
>> [ 3375.717215] Loaded elf core header at 0x0, bufsz=0x1000 memsz=0x80000
>> [ 3375.717229] kexec_elf: Loaded initrd at 0x0
>> [ 3375.718043] Memory node path: /memory@0
>> [ 3375.722854] kexec_elf: Loaded device tree at 0x0
>> syscall kexec_file_load not available.
>>
>> Kernel is reporting that all kexec segments are getting loaded at 0x0.
>>
>> Running kexec with strace shows that kexec_file_load system return -1 
>> EINVAL.
>>
>> kexec_file_load(3, 4, 1, "\0", KEXEC_FILE_ON_CRASH) = -1 EINVAL 
>> (Invalid argument)
>>
>> Based on the logs printed on the console and kexec_file_load return 
>> value. I am suspecting
>> kexec_file_load returned early form sanity_check_segment_list() 
>> because the segment is 0x0.
>>
>> I am investigating further to find how segment.mem for all segment is 
>> 0x0.
>>
>
> Interesting. Thanks for the update, Sourabh.
> I believe the error "syscall kexec_file_load not available." is
> inappropriate and misleading. That should be fixed too.

Agree. It is happening because kexec is handling ENOSYS, EINVAL, 
ENOEXEC, and ENOTSUP
as a single case and printing "syscall kexec_file_load not available".

I will send a fix for kexec tool to avoid this confusing error message.

Thanks,
Sourabh Jain