Bug 199613

Summary:	Black screen every 1/3 times on boot
Product:	Memory Management	Reporter:	Francisco Pina Martins (f.pinamartins)
Component:	Other	Assignee:	Andrew Morton (akpm)
Status:	NEW ---
Severity:	normal	CC:	ae3nerdgod, michel, ollybee, zapduke
Priority:	P1
Hardware:	x86-64
OS:	Linux
Kernel Version:	4.16.3	Subsystem:
Regression:	No	Bisected commit-id:
Attachments:	journalctl log with KASAN_OUTLINE and kasan_multi_shot using a kernel compiled with debug info journalctl log with using a kernel compiled with debug info

Description Francisco Pina Martins 2018-05-03 21:33:27 UTC

Created attachment 275759 [details]
journalctl log with KASAN_OUTLINE and kasan_multi_shot using a kernel compiled with debug info

Occurs on a new Ryzen 5 2400G build, using linux-4.16.3 (Arch Linux).
On some boots (~1 out of every 3) I get a black screen after modesetting occurs.
I have submitted the bug to freedesktop.org at first (https://bugs.freedesktop.org/show_bug.cgi?id=106225) since I suspected AMDGPU to be the cause.

However, after some diagnosing (with the help of Michel Dänzer) with KASAN it seems we have uncovered a possible bug in `rcu_cpu_kthread`, as indicated by `faddr2line`.

I have attached the relevant systemd logfile (for the entire boot process, with KASAN enabled and `kasan_multi_shot=1` as a kernel option).
If it matters, the mainboard is a Gigabyte "AB350N-Gaminig Wifi", with the latest available BIOS as of writing (BIOS F23d 04/17/2018).

Please advise if more information is required.

Comment 1 Oliver 2018-05-16 21:08:31 UTC

Created attachment 276011 [details]
journalctl log with using a kernel compiled with debug info

Comment 2 Oliver 2018-05-16 21:09:19 UTC

I can confirm the same issue with a Ryzen 2200G an MSI a320 motherboard. Boot fails when mode setting about half the time with linux-4.16.3 Arch and the current 4.16 fedora 28 kernel and the current 4.15 Ubuntu 18.04 kernel. Using the KASAN enabled kernel provided by Francisco I am able to boot everytime. The only difference being this only works for me without setting kasan_multi_shot=1, with that option I still fail to boot. Logs attached.

Comment 3 Luca 2018-06-15 18:13:22 UTC

I confirm this bug, it happens in the first phase of the boot after modesetting, the last line I can read is "amdgpudrmfb switching from efi fb". I'm usually able to reboot the system with a Sysrq and after that it almost always boot fine. 

Kernel 4.17
Asrock AB350M PRO4 - firmware 4.73B (AGESA 1.0.0.3b)
CPU ryzen 2200g

Comment 4 Luca 2018-08-28 11:18:22 UTC

Seems to be fixed in 4.19 rc1.

Comment 5 Francisco Pina Martins 2018-08-30 23:26:41 UTC

I can confirm that after 7-8 reboots with linux-4.19-rc1 I have gotten no crashes at all.

Comment 6 ae3nerdgod 2018-09-22 01:37:32 UTC

I can confirm any kernel After 4.14 does this for me, I have two monitors and the primary one boots blank every once an a while. Unplugging and replugging the affected monitor fixes the problem for me. (PS the first two monitors are connected to an and you, the 3rd is connected to an nvdia 1070 and I haven't noticed it happening.) Will try to get log when it happens again