Navi "divide error" hang

Submitted by Doug Tyrrell @DougTy

Assigned to Default DRI bug account

Link to original bug (#111869)

Description

Created attachment 145593
divide-error_09-20.txt

Occasionally, usually while watching videos in Firefox, my GPU will hang and the screen will freeze -- sound and keyboard input still work in the background, and I need to use REISUB hotkeys to reboot. This is separate, in addition to the sdma and ring gfx_0.0.0 hangs.

Upon rebooting, journalctl shows the attached "divide error". I've included logs from 3 instances of it happening. I'm currently using the Jul 14 firmware from Fedora's linux-firmware package as the hang appears to occur more often on the newer firmware from https://people.freedesktop.org/~agd5f/radeon_ucode/navi10/ however this may just be placebo.

It occurs with or without the "0-sized IBs" kernel patch from https://bugs.freedesktop.org/show_bug.cgi?id=111481#c33 and on both PCIe 3.0 and 4.0. I'm not using a PCIe riser and the card works without issue on Windows 10 dual boot.

CPU: 3700X
GPU: Sapphire 5700XT (reference)
Motherboard: Gigabyte X570-I (BIOS F4)
Kernel: 5.3.0
Mesa: mesa-git 1:19.3.0_devel.115682.3c966fd688c-1
LLVM: llvm-git 10.0.0_r327425.63f6066b53d-1

Please let me know if any more information would be helpful, or if there's anything I can do to troubleshoot. Thanks.

Attachment 145593, "divide-error_09-20.txt":
divide-error_09-20.txt