Bug 644202

Summary:	=gentoo-sources-4.14.13 causes mdadm to segfault at the initramfs stage
Product:	Gentoo Linux	Reporter:	calimeroteknik
Component:	Current packages	Assignee:	Gentoo Linux bug wranglers <bug-wranglers>
Status:	RESOLVED INVALID
Severity:	normal	CC:	bkohler, hydrapolic
Priority:	Normal
Version:	unspecified
Hardware:	All
OS:	Linux
Whiteboard:
Package list:		Runtime testing required:	---
Attachments:	mdadm segfaults on 4.14.13 in the initramfs emerge --info

Description calimeroteknik 2018-01-11 15:15:53 UTC

Created attachment 514226 [details]
mdadm segfaults on 4.14.13 in the initramfs

I observe a segfault of mdadm at the ramfs stage.
Screenshot provided in qemu (running without KVM).

As an aside, it seems to be known that python 3.5 also segfaults on 4.14.13.


Profile: default/linux/amd64/17.0/no-multilib/hardened

Notable USE: lzma for sys-apps/kmod, required for ko.xz modules

Kernel config:
https://git.archlinux.org/svntogit/packages.git/plain/trunk/config?h=packages/linux&id=dc615c7e4fc98551f6b2df9d0e97743350ba94bd with FIRMWARE_IN_KERNEL=y so genkernel doesn't fail when trying to `make firmware_install`

Kernel/ramfs generated with:
genkernel --kernel-cc=/usr/lib/ccache/bin/gcc --kernel-config=config.seeabove --makeopts=2 all

Comment 1 Ben Kohler gentoo-dev

2018-01-11 15:45:16 UTC

emerge --info output and genkernel version will definitely help investigate this

Comment 2 Tomáš Mózes 2018-01-11 16:04:01 UTC

Maybe hardened related as I've some machines on 4.14.13 with mdadm.

Comment 3 Tomáš Mózes 2018-01-11 16:06:26 UTC

Have you tried booting with pti=off?

Comment 4 calimeroteknik 2018-01-11 21:47:45 UTC

This might be a false alert, since it boots on the real machine, just not in QEMU.
I have this in /etc/portage/make.conf:
CFLAGS="-march=sandybridge -O2 -pipe"

Maybe emulation doesn't support some opcodes of sandybridge (instead of say, generic x86_64) and it's just that.

As requested:
$ genkernel --version
3.4.52.4

Attaching output of emerge --info

Trying pti=off did not change that, and it actually happens on 4.14.12 as well, which has me say sorry again, it suspciously looks to be all a fluke with QEMU on non-generic x86_64.

Comment 5 calimeroteknik 2018-01-11 21:48:44 UTC

Created attachment 514424 [details]
emerge --info

Comment 6 Ben Kohler gentoo-dev

2018-01-11 21:50:47 UTC

Yeah I wouldn't 100% trust qemu to pass through all those sandy-specific instructions, doubly so if you have hwvirt off.