Hard lockups using microcode releases 20191115 and 20191112

[jariruusu]

Laptop computer: Dell Latitude 5580, i5-7200U CPU @ 2.50GHz
Kaby Lake Mobile, id 806E9, family 6 model 142 stepping 9 pf 0x80
Latest BIOS from Dell, 1.16.0 07/03/2019
Linux kernel version 4.14.164 (stable mainline + local patches)
BIOS microcode revision 0xb4 is stable, no issues observed.
Microcode release 20191112, revision 0xc6 is unstable, hard lockups.
Microcode release 20191115, revision 0xca is unstable, hard lockups.

[hmh]

Hard lockups when?

• boot, when cold booting ? (from power-off state)
• boot, when rebooting ?
• during normal system operation ?

[jariruusu]

No issues on booting, cold or warm.
Hard lockups after 1-2 days of normal use.

[ashok-raj]

Could you also add /proc/cpuinfo from the system?
Is it possible to try with a more upstream kernel?
Do you suspend/resume atall during the 2 days? or its operating non-stop. Just want to understand if there is something else that could be triggering it.

[jariruusu]

Could you also add /proc/cpuinfo from the system?

More information here:
http://www.elisanet.fi/jariruusu/916/issue23-1.tar.gz

Is it possible to try with a more upstream kernel?

That is possible, but the problem here is unstable microcode.

Do you suspend/resume atall during the 2 days?

No suspend/resume, it is disabled in my kernel configuration.

Dell BIOS folks may be able provide more information. Dell being
significant customer of Intel, it may be that Intel employees
have contacts there.

I have learned that Dell removes BIOS versions from their
download pages if Dell learns that a version of BIOS has severe
stability issues. They don't remove those that have exploitable
vulnerabilities, just those that crash or lock up unexpectedly.

I have archived download links to old BIOS versions of this
computer. I just checked if the links still work. Yes, old
stable ones are still available, unstable ones are gone. One
removed unstable example is BIOS version 1.8.1 (microcode
revision 0x7C) that attempted spectre_v2 mitigations, but
resulted in programs crashing. Later microcode versions were
better.

Dell has released 3 BIOS versions using same version 1.16.0
number. The dates below are "published date" not BIOS build
date.

2019-07-01 BIOS 1.15.1 has microcode 0xb4, stable, available
2019-07-24 BIOS 1.16.0 has unknown microcode version, NOT AVALABLE
2019-09-20 BIOS 1.16.0 has unknown microcode version, NOT AVALABLE
2019-10-22 BIOS 1.16.0 has microcode 0xb4, stable, available

So, Dell pulled the plug on two of them, presumably because they
were unstable. Is it cosmic coincidence that there are two newer
microcode revisions (0xc6 and 0xca) released after that stable
revision 0xb4 for this processor and that I have tested and
found both of them unstable?

I have archived these BIOS updates. Here are the MD5 sums for them:

d9218217b691872852a65fd37d55406e Latitude_5X80_Precision_3520_1.16.0.exe
33b161caa01e70863c74f5af64947647 Latitude_5X80_Precision_3520_1.16.0.exe
1e28074ecdb5863df309ad9feadaeae1 Latitude_5X80_Precision_3520_1.16.0.exe

I only installed the last of these, the one which has microcode 0xb4.

Would it be possible for some Intel employee to contact Dell
BIOS folks and ask them why Dell decided to pull the plug on
those two BIOS versions.

[jariruusu]

For few weeks I have been using microcode revision 0xd2 from
https://github.com/platomav/CPUMicrocodes on my laptop
computer (sig 0x806E9 fam 6 model 142 step 9 pf 0x80). That
microcode has stable, no issues observed. Now that Intel has
released newer microcode 0xd6 for that processor, this
stability issue with older microcodes 0xc6 and 0xca can be
closed.