AM4 Mobos used for Ryzen

[Anon-E-moose]

If I build amdgpu into the kernel, I get a panic because it tries to kill init.
When its a module, its just an Oops but it works.

What video card? Any messages in dmesg?
With it built in are you providing all the firmware in the kernel?

[NeddySeagoon]

Anon-E-moose,

I've tried an RX450 and an RX550. That's Polaris11 and Polaris12 respectively.
Both work in my Phenom II. Indeed, I migrated the kernel from the old box. The kernel has both sets of firmware built in.

Here's dmesg after

Code: Select all

modprobe amdgpu 

The address in

Code: Select all

[   73.145545] BUG: kernel NULL pointer dereference, address: 0000000000000056

varies from kernel to kernel and from built in to modular.
I don't think that means anything as I would expect things to move around in RAM.
Here's the kernel .config

It may look a bit 'odd' as its for a static /dev install. No DEVTMPFS, no EVDEV, no autoblackmagic at all.
That's the same as the Phenom box though.

Its unlikely to be RAM. I've tried 128GB and two different 32GB sticks on their own.

[Anon-E-moose]

I wouldn't think it would be memory either, since it works as a module.

I have an rx550 I couldn't get it to work as builtin either, so that's why I've stuck with a module.

You have a few odd selections in your config but I wouldn't think it would cause problems. (stuff dealing with soc's)

You mentioned having an oops even as a module, was it generated by the video card? If so what was it,

[NeddySeagoon]

Anon-E-moose,

Its in my dmesg linked above.

[Anon-E-moose]

Sorry I missed the link to dmesg (I did look for it, just overlooked it)

I would be tempted to turn the acpi enforce from lax to no. ETA. I used to run lax, but have it set to no at least since I've been running 5.11 series

This looks a little funny

Code: Select all

[   73.145545] BUG: kernel NULL pointer dereference, address: 0000000000000056
[   73.145566] #PF: supervisor read access in kernel mode
[   73.145582] #PF: error_code(0x0000) - not-present page

and I think they may be related.

I'm not sure if the address is a real addy or just from the start of something within the kernel itself.

I'd also go back and turn off the few things that you shouldn't need
CONFIG_I2C_AMD756=y
CONFIG_DRM_AMD_ACP=y

I have these set though they're not supposed to be necessary for things like the 550, not sure about the 450 though.
CONFIG_DRM_AMDGPU_SI=y
CONFIG_DRM_AMDGPU_CIK=y

[Marlo]

@Hello NeddySeagoon,

I also have a Ryzen and a RX 550 for Polaris 11. I switched my system to buid-in last week.

Code: Select all

Graphics:  Device-1: Advanced Micro Devices [AMD/ATI] Baffin [Radeon RX 550 640SP / RX 560/560X] vendor: Micro-Star MSI 
           driver: amdgpu v: kernel bus ID: 08:00.0 chip ID: 1002:67ff 
           Display: server: X.Org 1.20.13 compositor: kwin_x11 driver: amdgpu FAILED: ati unloaded: modesetting 
           alternate: fbdev,vesa resolution: 1: 3840x2160~60Hz 2: 1920x1080~60Hz s-dpi: 96 
           OpenGL: renderer: Radeon RX 560 Series (POLARIS11 DRM 3.41.0 5.13.16-gentoo-dist-hardened LLVM 12.0.1) 
           v: 4.6 Mesa 21.1.7 direct render: Yes 

I read in your .config the line CONFIG_EXTRA_FIRMWARE. I am sure you have too many specs built in there.
If you execute:

Code: Select all

$ cat /proc/cpuinfo | grep "cpu family"

you will see:

Code: Select all

cpu family : 23....

Then go to https://wiki.gentoo.org/wiki/AMD_microcode
and look for the number 23, which is the line for:

Code: Select all

 "Zen, Zen+, Zen 2 2017 Ryzen 3|5|7|9, Threadripper, EPYC".

Here you can see that you only need

Code: Select all

amd-ucode/microcode_amd_fam17h.bin

No others.

Likewise, too many elements are included for polaris11.
You only need this line:

Code: Select all

  amdgpu/polaris11_k_mc.bin amdgpu/polaris11_pfp_2.bin amdgpu/polaris11_me_2.bin amdgpu/polaris11_ce_2.bin amdgpu/polaris11_rlc.bin amdgpu/polaris11_mec_2.bin amdgpu/polaris11_mec2_2.bin amdgpu/polaris11_sdma.bin amdgpu/polaris11_sdma1.bin amdgpu/polaris11_uvd.bin amdgpu/polaris11_vce.bin amdgpu/polaris11_k_smc.bin

Last, with polaris12 I have no experience.

Greetings
Ma

[nikolis]

@Tony0945
I hope you'll get better soon. (You did not drink water?)

Thank you. I hate water. Never have voluntarily drank it (srunk it?). Coffee, milk, juice. Yes

Βad idea ! You should drink water, If I do not convince you, take a walk to a hospital with kidney patients.

[NeddySeagoon]

Anon-E-moose,

I've tried turning off

Code: Select all

CONFIG_I2C_AMD756=y
CONFIG_DRM_AMD_ACP=y 

I suspect I left them on by accident after I was poking about in the kernel to help here or on IRC.

I also had AGP support on but its been 15 years since I had a system with an AGP slot, so I turned that off too.

I can confirm that

Code: Select all

CONFIG_DRM_AMDGPU_SI=y
CONFIG_DRM_AMDGPU_CIK=y

are not needed for Polaris* chipsets, which both the RX450 and RX550 are.

I also changed lax to no.

With those changes and the bump from 5.14.6 to 5.14.7, no only does amdgpu not painc, but it switches off on shutdown too. :)
Thank you.

For completness, all that's left in dmesg now is

Code: Select all

[   58.627285] [drm:amdgpu_debugfs_init [amdgpu]] *ERROR* unable to create amdgpu_preempt_ib debugsfs file
[   58.627410] [drm:amdgpu_pci_probe [amdgpu]] *ERROR* Creating debugfs files failed (-2)

but thats harmless as I don't have debugfs support.

When 5.14.8 comes along, I'll try amdgpu built in. Baby steps.
Abnormality is nearly restored.

[Anon-E-moose]

Neddy, I think the real key was telling acpi to leave things alone, even if it's in it's "range".
I think there is a difference between lax and no, even though the kernel doc says they should be effectively the same.
I suppose that it could be just something that affects a minor kernel version or two, but I've never gotten ambitious enough to chase the problem down.

As far as the other options, they were probably ok, but when I get odd conflicts, especially reboots, oops, etc I tend to shut off options I know I don't need or add options that I might need.

I really need to go through my config and shut off options that are no longer relevant, I just got lazy because things are working well.

[NeddySeagoon]

Anon-E-moose,

That was what bit me.

I copied the kernel .config from a Phenom II to a Zen3 and turned on the extra bits for the new motherboard.
That mostly worked :)

There is some junk there that I will get rid of as kernel updates happen.
That way, when, not if, I build a dud kernel, the last working kernel is there as a get out of jail free card.

[greyspoke]

Just built a new system with a MSI B-550A PRO. It wasn't clear that it needed a firmware update before it would recognize my 5600G processor, but once I sorted that it seems to work ok. Still to fully complete migration from Intel though.

Using 32G RAM in 4 sticks of 2400 MHz so I can re-use the 16G I already had.

There wasn't enough room on the screen for all the penguins until I adjusted the resolution!

[greyspoke]

While I am here a question about processor firmware. My processor appears to be series 25, whereas Linux a firmware stops at 23. Would there be any point in using version 23?

[Tony0945]

While I am here a question about processor firmware. My processor appears to be series 25, whereas Linux a firmware stops at 23. Would there be any point in using version 23?

Best asked on the MSI forum. Or MSI tech support for an authoritative answer.

[mike155]

While I am here a question about processor firmware. My processor appears to be series 25, whereas Linux a firmware stops at 23. Would there be any point in using version 23?

Be careful with those version numbers. Sometimes they are given in decimal notation (25) and sometimes in hexadecimal (19)...

As far as I know, AMD has not released firmware updates for the 5600G to the public. AMD distributes firmware to mainboard manufacturers and those distribute firmware updates via BIOS updates. Update your BIOS if you want the latest firmware. I expect that AMD will release firmware updates to the public when the first mainboard manufacturers stop support for their mainboards.

[greyspoke]

While I am here a question about processor firmware. My processor appears to be series 25, whereas Linux a firmware stops at 23. Would there be any point in using version 23?

Be careful with those version numbers. Sometimes they are given in decimal notation (25) and sometimes in hexadecimal (19)...
d
As far as I know, AMD has not released firmware updates for the 5600G to the public. AMD distributes firmware to mainboard manufacturers and those distribute firmware updates via BIOS updates. Update your BIOS if you want the latest firmware. I expect that AMD will release firmware updates to the public when the first mainboard manufacturers stop support for their mainboards.

Thanks @mike155, I think it is 25 (decimal). I am not sure if the AMD motherboard firmware that is part of the BIOS is/contains the same thing as the cpu microcode that is contained in the blobs in /lib/firmware/. It appears to be a different thing. But I have the latest non-beta version for my board, it didn't recognise the processor at all with the BIOS as shipped.

[NeddySeagoon]

greyspoke,

AGESA is a part of the BIOS. It executes on the CPU like the rest of the BIOS.
The microcode is the hard wiring (it used to be hard wired) the tells the CPU how to execute each instruction.

There is a version of the microcode in slow permanent memory inside the CPU. This microcode in transferred to the microcode RAM inside the CPU for execution.
It executes at the full CPU core speed there.
As its in RAM, it can be updated.

The BIOS often contains a copy of a later microcode that that shipped inside the CPU, so the BIOS can update the microcode.
/lib/firmware/ can contain a later still version of the microcode, so the kernel can update it too.
The microcode updater does version checking. It won't load the wrong microcode for your CPU and it won't overwrite a new one with an old one.

Intel have recalled CPUs for a microcode error, in the days when it could not be updated.
FDIV bug

[greyspoke]

Thanks Neddy, that makes sense of it for me.

[shimbob]

Just wanted to mention the new AMD P-State driver in the 5.17 kernel seems to be working here. Can't say I've noticed any difference but I'm not looking very hard.

[Anon-E-moose]

So, interesting problem showed up when I did a bios update (asus 1405 -> 4204)

After the update, system started, but for the next week, I spent many hours trying to figure out why it was unstable, spurious reboots, etc.

Among some of the things that the update "did for me" was to leave garbage/bad values spread throughout the mem setting area,
most things were originally set auto, but the update set some strange values. Got that straightened out. Still unstable.

Mem is 3600Ghz and has reliable run at that for well over a year now, but I had to lower it to 3200 level to get the system to quit rebooting.

So playing around with some of the knobs in the bios and clicked on TPU tab, which allows 2 values TPU1/TPU2, one being air cooled oc, the other geared toward water cooling.

It set the speed to be 4050Ghz, which is a nice bump, by setting the CPU core ratio option (40.5 in this case) but I was still seeking stablility, so I set the value to 36 which corresponds to 3600Ghz which is the nominal rated speed. No problem. But in setting the value to 36 I noticed that that adjusted the voltage for things like VDDCR Cpu Voltage, left me at 1.105 volts. This got me to thinking, I know that my original setting for core ratio was auto, which gave me a running high 900 mv range, appx .1 to .2 off the voltage set when 36 was selected. But the system had been running fine this way for over a year, until the update.

So either amd is doing something different (asus and other use their reference) or asus is doing something different.

And since I'm running a slightly higher voltage, I notice the temps are slight higher (not surprising) but still well within tolerance (38 idle instead of 34)

Anyway, bottom line if you have stability problems after an update on your ryzen mb, you might check things like the cpu voltage.

Edit to add: These are the messages I started seeing (in linux) after the bios update (before I figured out what was going on)

Code: Select all

mce: [Hardware Error]: Machine check events logged
 [Hardware Error]: Corrected error, no action required.
 [Hardware Error]: CPU:0 (17:71:0) MC25_STATUS[-|CE|MiscV|-|-|-|-|CECC|-|-|-]: 0x98004000003e0000
 [Hardware Error]: IPID: 0x000100ff03830400
 [Hardware Error]: Platform Security Processor Ext. Error Code: 62
 [Hardware Error]: cache level: RESV, tx: INSN

[saellaven]

So, interesting problem showed up when I did a bios update (asus 1405 -> 4204)

After the update, system started, but for the next week, I spent many hours trying to figure out why it was unstable, spurious reboots, etc.

I've been hanging out on 2802 for a while now... I get tempted to update, but everything is just working and I don't want to risk stability issues. It seems like many people have been experiencing USB and stability issues on higher versions. I figured most of the stability issues were related to overclocking, voltages, etc... but I don't want random weird USB hangs/resets, so I've just kept where I am. I do download the newer BIOSes with a plan to try them, I just never get the time to invest in having to fix potential instability.

[nvaert1986]

I'm actually looking for a new AMD X570 based motherboard. Does anybody have a recommendation?

I had either the ASUS TUF Gaming X570 Pro WiFi II or the Gigabyte X570S Aero G in mind

- From what I figured neither of them has the annoying chipset fan that the X570 based models had (ASUS just doesn't seem to call them X570S motherboards?)
- Both of them have an Intel 2,5Gbps NIC onboard (essential for any stable network connection in my opinion)
- I don't care about the WiFi / Bluetooth that they offer (as I'll not be using it anyway; The ASUS motherboard has a Mediatek based chipset and the Gigabyte motherboard has a ADMTek based chipset)
- Both of them offer Realtek High Definition Audio (which is fine by me, as I'll be using my AudioQuest Dragonfly USB device)

Does anybody have any experience with either of these motherboards on Gentoo? How about sensor monitoring? I read that the Pro WiFi is supported, but I couldn't find anything anywhere about the Pro WiFi II (ASUS basically markets them as the same motherboard, as they even point to the same reviews on the website, while they're actually quite different). Couldn't find anybody that had any experience with Linux on either of these boards either.

[Thulle]

Just wanted to mention the new AMD P-State driver in the 5.17 kernel seems to be working here. Can't say I've noticed any difference but I'm not looking very hard.

Tried it out, was worse in every way, which from the conversations around it seemed almost intentional, just wanted the code merged so haven't tuned it yet.

[Thulle]

Just wanted to add that things have improved with the P-State driver, happily using it nowadays.

[nikolis]

Is there anyone with am5 ?

[thumper]

Just wanted to add that things have improved with the P-State driver, happily using it nowadays.

Do you have a link to the how-to, or comparable document?

TIA

George