Microcode update causes hard lockups in Intel N4200 CPU/SoC

Months ago, my Acer Swift SF113-31-P5CK laptop (running KDE Neon or Kubuntu) began to hard lock at seemingly random intervals, ignoring sysrq+REISUB and requiring a force-off. After some trial and error, I tried removing the pertinent microcode file (06-5c-09) from /lib/firmware/intel-ucode and updating the initramfs, and the lockup never happened again. After I discovered that, I checked, and indeed, the onset of the lockups coincided with the delivery of the latest Intel microcode by Ubuntu.

I then tried inserting the second newest microcode file into /lib/firmware/intel-ucode and updated the initramfs, and again, no lockups. They previously happened anywhere from 5 minutes to a few hours into a session, and often multiple times a day.

By now it has been several months since the last lockup, and I use the laptop hours each day. I could not get through a day without a lockup with the newest microcode file (0x38) in use, and often there were multiples per day.

The faulty microcode for the Intel Pentium Silver N4200 SoC that my Swift uses is 0x38, which is the newest, as far as I know, so that's the one the Ubuntu update delivers. The one I am using now is 0x32, the previous one before 0x38, and it's been in use for over a thousand hours with no lockup.

I saw a recent Microsoft info bulletin describing the newest Windows microcode update at the time, and I noticed that the microcode listed for my CPU in that update was 0x32. The prior Microsoft microcode update showed 0x38, and there was a footnote for the bulletin that mentioned it had been reverted (though it didn't say why). I can only conclude that I was not the only one having problems with the 0x38 release of 06-5c-09 (SHA256 9f6d3bef49d33dfc556aedb94670c6c347662d9c3cb9ececd58e23cd636512fc).

Not surprisingly, I had the same experience with non-Ubuntu distros (OpenSUSE and Fedora). If 0x38 was loaded, it was going to lock up frequently.

I think 0x38 was probably one of the Spectre or Meltdown ucode updates, but it's no good if the thing causes PCs to lock up in 5 minutes of use. Until/unless Intel decides to release a version that actually works as intended, I think 0x32 is the lesser of the two evils.