mjhjmtu

I have months of kernel logs saved. Can I get any useful information from them about when my system started using KPTI, and possibly the ASID/PCID feature which mitigates its performance?

I am specifically interested in Fedora kernels 4.14 - 4.15.

My current kernel is 4.15.10-300.fc27.x86_64, and it has KPTI enabled.

$ cat /sys/devices/system/cpu/vulnerabilities/meltdown 
Mitigation: PTI

I have tried a naive search but there were no results.

$ journalctl --since=2018-01-01 _TRANSPORT=kernel | grep -iE "kpti|asid|pcid"

Context

I'm trying to debug a horrible Xwayland crash.

The crash seems to have first started for me on 2018-02-28. For others, it is more frequent and perhaps as early as 2018-02-25.

I was very suspicious about CPU microcode updates, a bug in which has been known to cause "unexpected page faults". However the timeline is not fully convincing. So it made me think, since I am investigating unexpected page faults, another place to look would be the kernel. And obviously there has been some big changes recently in the kernel page table support!

My kernel was upgraded from the 4.14 to 4.15 line (with whatever patches Fedora has applied) on the date 2018-02-18.

Feb 16 18:59:00 alan-laptop kernel: Linux version 4.14.18-300.fc27.x86_64 (mockbuild@bkernel01.phx2.fedoraproject.org) (gcc version 7.3.1 20180130 (Red Hat 7.3.1-2) (GCC)) #1 SMP Thu Feb 8 01:35:31 UTC 2018
Feb 18 12:50:42 alan-laptop kernel: Linux version 4.15.3-300.fc27.x86_64 (mockbuild@bkernel02.phx2.fedoraproject.org) (gcc version 7.3.1 20180130 (Red Hat 7.3.1-2) (GCC)) #1 SMP Tue Feb 13 17:02:01 UTC 2018

KPTI is shown by this line:

Jan 04 14:40:33 alan-laptop kernel: Kernel/User page tables isolation: enabled

If you disable PCID by passing nopcid, there's a log message.

+ pr_info("nopcid: PCID feature disabledn");

Otherwise, you need some deep knowledge to check the kernel versions, sorry.

PCID support was implemented in the upstream kernel version 4.14. It should be enabled if your cpu supports it (pcid and pge flags in lscpu), unless you are using Xen ("Xen PV would need some work") or booting a 32-bit kernel ("PCID can only be used in 64-bit mode").

This gets a bit confusing. KPTI was introduced subsequently in upstream Linux 4.15. However the log message above comes from the Fedora kernel 4.14.8-300, which must have included a backport of KPTI. It's good news considering how catastrophic the Meltdown vulnerability is. The code for KPTI, X86_BUG_CPU_INSECURE etc was backported to upstream stable 4.14.11, released Jan 2 2018.

The other most obvious messages, do not seem to provide an nice clear suspect for my investigation either :(.

CPU microcode updates:

May 18 18:42:52 alan-laptop kernel: microcode: sig=0x306d4, pf=0x40, revision=0x23
May 18 18:42:52 alan-laptop kernel: microcode: sig=0x306d4, pf=0x40, revision=0x24

May 19 09:03:58 alan-laptop kernel: microcode: microcode updated early to revision 0x24, date = 2016-04-29
May 19 09:03:58 alan-laptop kernel: microcode: sig=0x306d4, pf=0x40, revision=0x24
May 19 09:03:58 alan-laptop kernel: microcode: Microcode Update Driver: v2.2.

Jun 03 09:24:34 alan-laptop kernel: microcode: microcode updated early to revision 0x25, date = 2017-01-27
Jun 03 09:24:34 alan-laptop kernel: microcode: sig=0x306d4, pf=0x40, revision=0x25

Jan 11 08:48:40 alan-laptop kernel: microcode: microcode updated early to revision 0x25, date = 2017-01-27
Jan 11 08:48:40 alan-laptop kernel: microcode: sig=0x306d4, pf=0x40, revision=0x25

Feb 15 17:02:12 alan-laptop kernel: microcode: microcode updated early to revision 0x28, date = 2017-11-17
Feb 15 17:02:12 alan-laptop kernel: Intel Spectre v2 broken microcode detected; disabling Speculation Control
Feb 15 17:02:12 alan-laptop kernel: microcode: sig=0x306d4, pf=0x40, revision=0x28

Mar 21 12:21:28 alan-laptop kernel: microcode: microcode updated early to revision 0x2a, date = 2018-01-18
Mar 21 12:21:28 alan-laptop kernel: microcode: sig=0x306d4, pf=0x40, revision=0x2a

Spectre variant 2:

Jan 15 09:10:59 alan-laptop kernel: Spectre V2 mitigation: Vulnerable: Minimal generic ASM retpoline

Jan 25 10:59:57 alan-laptop kernel: Spectre V2 mitigation: Mitigation: Full generic retpoline

Feb 15 17:02:12 alan-laptop kernel: Intel Spectre v2 broken microcode detected; disabling Speculation Control
Feb 15 17:02:12 alan-laptop kernel: Spectre V2 : Mitigation: Full generic retpoline

Mar 21 12:21:28 alan-laptop kernel: Spectre V2 : Mitigation: Full generic retpoline
Mar 21 12:21:28 alan-laptop kernel: Spectre V2 : Spectre v2 mitigation: Enabling Indirect Branch Prediction Barrier
Mar 21 12:21:28 alan-laptop kernel: Spectre V2 : Enabling Restricted Speculation for firmware calls