Kernel Crash Dump (kdump) broken with 6.5

[Michael Ellerman]

Mahesh J Salgaonkar <mahesh at linux.ibm.com> writes:
> On 2023-07-17 20:15:53 Mon, Sachin Sant wrote:
>> Kdump seems to be broken with 6.5 for ppc64le.
>> 
>> [ 14.200412] systemd[1]: Starting dracut pre-pivot and cleanup hook...
>> [[0;32m OK [0m] Started dracut pre-pivot and cleanup hook.
>> Starting Kdump Vmcore Save Service...
>> [ 14.231669] systemd[1]: Started dracut pre-pivot and cleanup hook.
>> [ 14.231801] systemd[1]: Starting Kdump Vmcore Save Service...
>> [ 14.341035] kdump.sh[297]: kdump: saving to /sysroot//var/crash//127.0.0.1-2023-07-14-13:32:34/
>> [ 14.350053] EXT4-fs (sda2): re-mounted e971a335-1ef8-4295-ab4e-3940f28e53fc r/w. Quota mode: none.
>> [ 14.345979] kdump.sh[297]: kdump: saving vmcore-dmesg.txt to /sysroot//var/crash//127.0.0.1-2023-07-14-13:32:34/
>> [ 14.348742] kdump.sh[331]: Cannot open /proc/vmcore: No such file or directory
>> [ 14.348845] kdump.sh[297]: kdump: saving vmcore-dmesg.txt failed
>> [ 14.349014] kdump.sh[297]: kdump: saving vmcore
>> [ 14.443422] kdump.sh[332]: open_dump_memory: Can't open the dump memory(/proc/vmcore). No such file or directory
>> [ 14.456413] kdump.sh[332]: makedumpfile Failed.
>> [ 14.456662] kdump.sh[297]: kdump: saving vmcore failed, _exitcode:1
>> [ 14.456822] kdump.sh[297]: kdump: saving the /run/initramfs/kexec-dmesg.log to /sysroot//var/crash//127.0.0.1-2023-07-14-13:32:34/
>> [ 14.487002] kdump.sh[297]: kdump: saving vmcore failed
>> [[0;1;31mFAILED[0m] Failed to start Kdump Vmcore Save Service.
>
> Thanks Sachin for catching this.
>
>> 
>> 6.4 was good. Git bisect points to following patch
>> 
>> commit 606787fed7268feb256957872586370b56af697a
>>     powerpc/64s: Remove support for ELFv1 little endian userspace
>> 
>> Reverting this patch allows a successful capture of vmcore.
>> 
>> Does this change require any corresponding change to kdump
>> and/or kexec tools?
>
> Need to investigate that. It looks like vmcore_elf64_check_arch(&ehdr)
> check from fs/proc/vmcore.c is failing after above commit.
>
> static int __init parse_crash_elf64_headers(void)
> {
> [...]
>
>         /* Do some basic Verification. */
>         if (memcmp(ehdr.e_ident, ELFMAG, SELFMAG) != 0 ||
>                 (ehdr.e_type != ET_CORE) ||
>                 !vmcore_elf64_check_arch(&ehdr) ||
> [...]

Where vmcore_elf64_check_arch() calls elf_check_arch(), which was
modified by the commit, so that makes sense.

> It looks like ehdr->e_flags are not set properly while generating vmcore
> ELF header. I see that in kexec_file_load, ehdr->e_flags left set to 0
> irrespective of IS_ENABLED(CONFIG_PPC64_ELF_ABI_V2) is true or false.

Does initialising it in crash_prepare_elf64_headers() fix the issue?

cheers