Commit 7636c13 ("vmcoreinfo: read vmcoreinfo using
'vmcoreinfo_data'
when unavailable in elf note") moved the vmcoreinfo reading to always
read from memory instead of relying on diskdump/netdump's local
handlers. This was later reverted to fix regression in X86_64 kslar
images.
Reintroduce the `vmcoreinfo_read_from_memory` as fallback to
diskdump/netdump vmcores.
This reverts commit 72e2776caf1ca41dffcc8aba11c55c636565725b.
Cc: Aditya Gupta <adityag(a)linux.ibm.com>
Cc: Tao Liu <ltao(a)redhat.com>
Signed-off-by: Shivang Upadhyay <shivangu(a)linux.ibm.com>
Tested-by: Misbah Anjum N <misanjum(a)linux.ibm.com>
Hi,
I have tested this patch successfully on the following platform:
Platform: IBM PowerPC 64LE (PPC64LE)
Architecture: ppc64le
Host Kernel: 6.18.0-8-generic
Guest Kernel: 6.18.0-8-generic
OS: Ubuntu 26.04 (Resolute Raccoon)
Crash version: 9.0.1 (patched)
Compiler: gcc 15.2.0
GDB version: 16.2 (embedded)
Test cases verified:
--------------------
1. virsh dump --memory-only (elf, kdump-zlib compression)
2. virsh dump --memory-only --bypass-cache (elf, kdump-zlib compression)
Test results:
------------
- Patch applied cleanly to crash-9.0.1 (with offsets)
- Build completed successfully with all dependencies
- vmcoreinfo_read_from_memory fallback is active in diskdump.c and netdump.c
- Successfully reads vmcoreinfo from memory when ELF note method is unavailable
- Resolves the X86_64 kslar image regression
- No regressions detected
Test logs (trimmed):
-------------------
# virsh dump ubuntu2604 vmcore-mem --memory-only
Domain 'ubuntu2604' dumped to vmcore-mem
# crash /usr/lib/debug/boot/vmlinux-6.18.0-8-generic vmcore-mem
crash 9.0.1
...
KERNEL: /usr/lib/debug/boot/vmlinux-6.18.0-8-generic
DUMPFILE: vmcore-mem
...
crash> bt
PID: 0 TASK: c0000000038b1580 CPU: 0 COMMAND: "swapper/0"
[NIP : plpar_hcall_norets_notrace+24]
[LR : check_and_cede_processor+72]
...
The patch works as intended and fixes the vmcoreinfo reading issue.
Thanks,
Misbah Anjum N