** Description changed: Note: Updates are being staged at ppa:dannf/arm64-kdump. [Impact] It is not possible to collect a kernel crash dump from a crashed arm64 server for later debugging. [Test Case] sudo apt install kdump-tools (reboot, so crashkernel= is added to the kernel commandline) echo c | sudo tee /proc/sysrq-trigger Crash dump should occur, with artifacts collected in /var/crash. If you want to verify that the dump is usable, install the corresponding linux-image-<ver>-dbgsym package and run: sudo crash /usr/lib/debug/boot/vmlinux-<ver> /var/crash/<crash>/dump.<crash> - crash should successfully load, placing you at a "crash>" prompt. At - that prompt, you can issue the 'bt' command to see a backtrace. + crash should successfully load, placing you at a "crash>" prompt. At that prompt, you can issue the 'bt' command to see a backtrace. + Note: you need crash from zesty (7.1.8-1ubuntu1) or later. [Regression Risk] = Kernel = 3 patches here touch code outside of arch/arm64/: memblock: add memblock_clear_nomap() This adds a new function with no callers, so regression risk is negligible. (A later patch adds a call to it under arch/arm64/). memblock: add memblock_cap_memory_range() This refactors some of the code in memblock_mem_limit_remove_map() into a new function. The only existing caller of memblock_mem_limit_remove_map() is under arch/arm64/, so the regression risk outside of arm64 is negligible. efi/libstub/arm*: Set default address and size cells values for an empty dtb Because this code is for EFI platforms that support device-tree, it is de-facto ARM-specific (as noted in the patch title). For arm64, we have mitigated the risk by explicit regression testing on several platforms: - Qualcomm QDF2400 - Cavium ThunderX CRB1S - HP m400 (X-Gene) - HiSilicon D05 (Hi07) = kexec-tools = == zesty == For zesty, 10 patches are required to add kdump support. 0001-kexec-extend-the-semantics-of-kexec_iomem_for_each_l.patch: This modifies a function used on armhf & x86. The description explains the change, and why it does not impact those archs: ----- The current users of kexec_iomem_for_each_line(), arm, sh and x86, will not be affected by this change because * arm The callback function only returns -1 or 0, and the return value of kexec_iomem_for_each_line() will never be used. * sh, x86 The callback function may return (-1 for sh,) 0 or 1, but always returns 1 once we have reached the maximum number of entries allowed. Even so the current kexec_iomem_for_each_line() counts them up. This change actually fixes this bug. ----- 0002-kexec-generalize-and-rename-get_kernel_stext_sym.patch: This generalizes a function that was duplicated by arm & x86 and makes it common so arm64 can use it. The remaining 8 of these only touch code in kexec/arch/arm64, so regression risk for other architectures is negligible. Finally, I have tested this update on both i386 and amd64 VMs. i386 crashes do not currently work in zesty (filed LP: #1699874), and my test results show no change there. amd64 worked before, and continues to work with these changes. == yakkety == Since yakkety is based on an older upstream, 6 additional patches are required: arm-fix-get_kernel_stext_sym-to-close-its-file.patch: This is a cleanup patch, cherry-picked because it allows later patches to apply w/o backporting. ARM-specific. kexec-add-max_size-to-memory_ranges.patch: Adds a new element to struct, to be used by later commits. kexec-add-generic-helper-to-add-to-memoryq_regions.patch, kexec-add-mem_regions-sorting-implementation.patch, kexec-add-helper-to-exlude-a-region-from-a-set-of-me.patch, kexec-fix-mem_regions_sort.patch: These patches only add new functions, which will be used by later patches. kexec-arch-i386-Add-support-for-KASLR-memory-randomi.patch: This is a bug fix or i386 that allows later patches to apply w/o backporting. kdump support for i386 is apparently broken for the yakkety kernel (see LP: #1699874) so, if this introduces a regression, it won't be detectable. (I checked to see if this *fixes* i386 crashdumps - it does not). Note that, while makedumpfile in >= zesty is new enough to work on arm64, the yakkety version does not. kdump-tools falls back to copying the entire vmcore, which is what I tested. As with zesty, I have tested this update on both i386 and amd64 VMs. i386 crashes do not currently work in yakkety, and my test results show no change there. amd64 worked before, and continues to work with these changes.
-- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/1694859 Title: arm64 kernel crashdump support To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/kexec-tools/+bug/1694859/+subscriptions -- ubuntu-bugs mailing list ubuntu-bugs@lists.ubuntu.com https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
