Bug 27833

Created attachment 12086 [details]
DMESG file

Created attachment 12087 [details]
Journal File

no references what so ever of even booting 5.10.1 in the logs.

but I see it had some problems already with 5.9.12 with amdgpu

 RIP: 0010:dc_link_set_backlight_level+0x8a/0xf0 [amdgpu]


can you remove "splash silent" from kernel command line when you boot 5.10 to maybe catch some output

tried it.  No luck, flashes by way to fast. 

I tried the parameter amdgpu.noretry=0  that didn't fix it either.

ok, so then it does not seem to crash atleast on boot.

can you after it has booted to black screen, either try to ssh into the machine ?
(remember to install ssh server and open firewall port (or disable firewall temporarily))

or after it reached the black screen, wait a bit, reboot into working kernel and grab the last boot journal with:

journalctl -b -1 > bootlog 

and attach it here...

Nothing I can try here, Grubless! My own EFI M8 system with 5.10.1-desktop-1.mga8 boots fine via rEFInd (kernel stub).

CC'ing Aurélien, who I think uses Grub2.

CC: (none) => ouaurelien

@Brian, 

Please edit kernel 5.10.1 command line and add "systemd.unit=multi-user.target", remove "splash quiet". Press F100 to boot. Don't add the quote.

This will boot your system to console mode, without graphical session.
We must know when the process explode on your system.

Feel free to take a picture of your monitor if boot freezes.
Add here or send me by mail.

Hi,
Tried this with 5.10.1 and it did not work, same crash out to blank screen.

I change the command line in 5.9.12 and get to terminal.  If I really really screw up the command line I get to a Kernel panic.  I'll attach the photo, but I think it is just me flummoxing up the system so bad it doesn't know what is left and right.

Looking up notes on Kernel 5.10.1, it does seem that AMD slipped in some APU/GPU changes into the kernel that might not be fully vetted, at least that's what I interpreted in it.

I'll try and attach the complete flummoxed screen print.

Created attachment 12089 [details]
screen shot

(In reply to Brian Rockwell from comment #8)
> Hi,
> Tried this with 5.10.1 and it did not work, same crash out to blank screen.
> 
> I change the command line in 5.9.12 and get to terminal.  If I really really
> screw up the command line I get to a Kernel panic.  I'll attach the photo,
> but I think it is just me flummoxing up the system so bad it doesn't know
> what is left and right.
> 
> Looking up notes on Kernel 5.10.1, it does seem that AMD slipped in some
> APU/GPU changes into the kernel that might not be fully vetted, at least
> that's what I interpreted in it.
> 
> I'll try and attach the complete flummoxed screen print.

All I see in this kernel panic is that it can't find the root device.
This is really strange and I'd rather think a bad initrd image.

*** Bug 27845 has been marked as a duplicate of this bug. ***

CC: (none) => joequant

Aurelian,
As I said, your originally provided command went to a blank screen as well.  So I messed with the initd settings seeing if I could convince it to run.

Probably not valuable, but noted it anyhow.

 
ok, so atleast one R4 and one R5 APU fails

Created attachment 12093 [details]
boot stage

Same problem on my dual GPU laptop, both are R7 (Carizzo + M340). Using the "nomodeset" loading option brings me to this stage.

There is now a few packages to test:

kernel-5.10.1-2.mga8
x11-driver-video-amdgpu-19.1.0-6.mga8
mesa-20.3.1-1.mga8 (currently building)

(In reply to Cristian Pîrîu from comment #15)
> Same problem on my dual GPU laptop, both are R7 (Carizzo + M340). Using the
> "nomodeset" loading option brings me to this stage.

yeah, that happends because "nomodeset" blocks modesetting drivers, and amdgpu is a modeset-only driver, so its prevented to load properly

(In reply to Thomas Backlund from comment #17)
> (In reply to Cristian Pîrîu from comment #15)
> > Same problem on my dual GPU laptop, both are R7 (Carizzo + M340). Using the
> > "nomodeset" loading option brings me to this stage.
> 
> yeah, that happends because "nomodeset" blocks modesetting drivers, and
> amdgpu is a modeset-only driver, so its prevented to load properly

I'm sorry, my message was not clear. Using the "nomodeset" option allows login as root in "safe mode", without it is impossible. The problem persists with kernel-5.10.1-2.mga8. It seems to me that the amdgpu chooses the wrong gpu/output in the dual GPU configuration.

(In reply to Cristian Pîrîu from comment #18)

> I'm sorry, my message was not clear. Using the "nomodeset" option allows
> login as root in "safe mode", without it is impossible. The problem persists
> with kernel-5.10.1-2.mga8. It seems to me that the amdgpu chooses the wrong
> gpu/output in the dual GPU configuration.


can you ssh into it to see if it otherwise seems to work ?
anything special in dmesg or journal ?

Created attachment 12103 [details]
boot log with nomodeset and vga=3D4 options

Bootlog attached, I didn't see anything strange in it. In another system (desktop), with integrated gpu (disabled in the BIOS) and a separate RX550 GPU, everything works as expected.

can you set up so you have ssh access to the system and then boot without adding the "nomodeset" part so we maybe can see how the system actually gets and get the logs from it

Good news, everyone! Adding "amdgpu.dc=0" to kernel boot parameters is a temporary solution, everything works ok, for now. I have not been able to access my laptop using ssh.

(In reply to Cristian Pîrîu from comment #22)
> Good news, everyone! Adding "amdgpu.dc=0" to kernel boot parameters is a
> temporary solution, everything works ok, for now. I have not been able to
> access my laptop using ssh.

Nice, that could help tracking changes causing this

Created attachment 12119 [details]
Journal file with amdgpu.dc=0 set

HI All,
The kernel setting amdgpu.dc=0 worked on my APU as well.

I've attached the journal.  You'll find me boot to 5.9.12 then apply latest patches and then try again, but locked.  Go back to 5.9.12 lookup command and then try again with 5.10.1.2 with settings. 

They worked.

Let me know if you need anything else.

There is now a kernel-5.10.2-1.mga8 building where I've backported another fix from upstream that might fix this...

I can confirm that the kernel-5.10.2-1.mga8 works ok, as intended.

(In reply to Thomas Backlund from comment #25)
> There is now a kernel-5.10.2-1.mga8 building where I've backported another
> fix from upstream that might fix this...

(In reply to Cristian Pîrîu from comment #26)
> I can confirm that the kernel-5.10.2-1.mga8 works ok, as intended.

It seems we have a good candidate fix on this issue.
I pinged French forum IRC to relay test on this.

Installed update and removed amdgpu.dc=0 from kernel parms.

System booted and is working fine.

$ uname -a
Linux linux.local 5.10.2-desktop-1.mga8 #1 SMP Mon Dec 21 13:01:59 UTC 2020 x86_64 x86_64 x86_64 GNU/Linux

Need anything from me?

(In reply to Brian Rockwell from comment #28)
> Installed update and removed amdgpu.dc=0 from kernel parms.
> 
> System booted and is working fine.
> 
> $ uname -a
> Linux linux.local 5.10.2-desktop-1.mga8 #1 SMP Mon Dec 21 13:01:59 UTC 2020
> x86_64 x86_64 x86_64 GNU/Linux
> 
> Need anything from me?

Thanks testing this.

Closing this.

Status comment: (none) => fixed in 5.10.2 and updated nonfree firmware
Summary: Kernel 5.10.1 X86_64 (desktop) - unable to see login after update => Some AMD Radeon cards with kernel 5.10.1 can result of non working display

Really.

Status: NEW => RESOLVED
Resolution: (none) => FIXED