[Support] ich777 - AMD Vendor Reset, CoralTPU, hpsahba,...

[ich777]

1 hour ago, Lebowski89 said:

That's what I assumed as well. But I get this:

I'm not sure if it is a good idea to change the PSID since usually rebranded cards use IBM or HP instead of MT.

 

You can however try but I don't recommend that because what can happen is that you end up with a bricked card.

 

I can't find much info about that card either, sorry... can't help here.

[Lebowski89]

12 hours ago, ich777 said:

I can't find much info about that card either, sorry... can't help here.

 

Np, ich, thank you for trying to help. The tool itself is well made and easy to navigate.

 

I'll ask around if anyone has had something similar with my cards. But in the meantime, the NICs are working and doing what they're designed to do. iperf3 results show they're good to go. It's just a mystery of what the heck is going on with the firmware and can it be updated, if it even needs to be updated.

[edog305]

On 8/23/2023 at 3:05 AM, ich777 said:

Please remove the sound driver plugin, reboot and see if that makes a difference.

It is nice to see that you are using this plugin but I'm currently not supporting it officially since I have not enough free spare time to do so.

 

I think it might be the cause of the issue. If it's the cause of the issue I will maybe fix this for the next Unraid version.

 

Hey ich777, I actually just ran into this same issue myself, trying to get sound out of a steam-headless container. The sound plugin breaks the Radeon top plugin. I realize the plugin isn't officially supported, but can you maybe suggest what might be the problem? If its small enough, maybe I can open a PR in Github.

Thanks!

[ich777]

5 hours ago, edog305 said:

I realize the plugin isn't officially supported, but can you maybe suggest what might be the problem? If its small enough, maybe I can open a PR in Github.

It is already fixed but not for 6.12.4, the next release will have proper sound drivers, but keep in mind only for dedicated USB devcies and dedicated sound chips, not integrated audio solutions over HDMI or via the SoC.

[Skatman]

Apologies if this has been covered numerous times, but I cannot find the Intel-gpu-TOP plugin anywhere on CA anymore. Has it been removed or something?

[ich777]

12 minutes ago, Skatman said:

but I cannot find the Intel-gpu-TOP plugin anywhere on CA anymore. Has it been removed or something?

Right there:

 

 

On what Unraid version are you? Maybe it's time for a upgrade.

[Skatman]

6 minutes ago, ich777 said:

Right there:

 

 

On what Unraid version are you? Maybe it's time for a upgrade.

 

That is so strange as I cannot find it at all.

Unraid version (apols for the blown-out colour. Monitor is HDR):


CA Version:

 

Search:

[Skatman]

*Facepalm* Just seen we're at 6.12.4 now. I'm so used to Windows updates with a pop-up.

[Skatman]

55 minutes ago, ich777 said:

On what Unraid version are you? Maybe it's time for a upgrade.

Got it now! Cheers. Can't believe how out-of-date I was.

[Envie]

Hi, 
I bought a Intel A380 but Intel TOP GPU doesnt give out any data and im generally not sure if it works now for my jellyfin server.

[ich777]

6 hours ago, Envie said:

I bought a Intel A380 but Intel TOP GPU doesnt give out any data and im generally not sure if it works now for my jellyfin server.

ARC cards dont work on Unraid 6.12.x, you have to wait for 6.13

[Envie]

4 hours ago, ich777 said:

ARC cards dont work on Unraid 6.12.x, you have to wait for 6.13

So right now i cant use my gpu? Do we know when v 6.13. Will come? 

[ich777]

13 minutes ago, Envie said:

So right now i cant use my gpu?

Exactly, you actually can with custom Kernels but you would be better by waiting until 6.13 is released since I don't recommend using custom Kernels anymore.

 

15 minutes ago, Envie said:

Do we know when v 6.13. Will come? 

SOON™

 

Sorry but can't tell you when exactly but maybe in about a month or so... But that is just a guess.

[dhstsw]

Hi there,

Since updating to 6.12.4, every VM i create fails to start, stating:
 

char device redirected to /dev/pts/1 (label charserial0)
2023-11-19T10:27:44.614155Z qemu-system-x86_64: vfio: Cannot reset device 0000:01:00.1, no available reset mechanism.
2023-11-19T10:27:44.618203Z qemu-system-x86_64: vfio: Cannot reset device 0000:01:00.1, no available reset mechanism.

VMd created before updating do work.

So, i've been digging a bit into this.

I uninstalled the AMD Vendor Reset plugin, reboot, started one of those new VM and it went on without any problem.

Reinstalled the AMD Vendor Reset plugin, reboot, restarted the very same VM and it keeps working.

Checking one of the "non working VMs" (one wich was NOT started previously without the plugin), i find that in the XML view of the definition of the VM, in the "os" part, there's this (from "Lakka" VM):
 

<os>
    <type arch='x86_64' machine='pc-q35-7.1'>hvm</type>
  </os>

"Libreelec" VM, wich was experiencing the same problem, starting it without the AMD Vendor Reset plugin, is now showing this:
 

<os>
    <type arch='x86_64' machine='pc-q35-7.1'>hvm</type>
    <loader readonly='yes' type='pflash'>/usr/share/qemu/ovmf-x64/OVMF_CODE-pure-efi.fd</loader>
    <nvram>/etc/libvirt/qemu/nvram/e56733f0-82df-31c5-ddfd-21b884fb67c7_VARS-pure-efi.fd</nvram>
  </os>

Basically, with the plugin installed, if i create a new VM and start/it, there are no <loader> and <nvram> lines added, and also, the respective nvram file isn't created at all (if i look into /etc/libvirt/qemu/nvram/ ).

Not sure if this is a problem with the plugin itself, but the fact that everything goes as it should with it uninstalled leads me to think so.

Any ideas?

Diagnostics attacched, problematic VM is "Lakka".

Thanks.

incubus-diagnostics-20231119-1401.zip

Edited November 19, 2023 by dhstsw

[ich777]

2 hours ago, dhstsw said:

VMd created before updating do work.

Please unbind the card from VFIO because otherwise the plugin win‘t work properly anyways.

 

2 hours ago, dhstsw said:

Not sure if this is a problem with the plugin itself, but the fact that everything goes as it should with it uninstalled leads me to think so.

The plugin does nothing to the VM XML it does only install a Kernel module that watches the specific AMD cards.

 

What card are you using?

 

2 hours ago, dhstsw said:

Any ideas?

Again, I can only assume that the VM templates where you see other values are created differently. Are you sure that you where using OVMF before or where you using SeaBIOS?

[dhstsw]

5 minutes ago, ich777 said:

Please unbind the card from VFIO because otherwise the plugin win‘t work properly anyways.

I will try that now.
 

Quote

What card are you using?

RX560.
 

Quote

Again, I can only assume that the VM templates where you see other values are created differently. Are you sure that you where using OVMF before or where you using SeaBIOS?

Problem shows up with both SeaBIOS and OVMF (even if it manifests in different ways).

Trying to unbinding and will let you know.

THX

[dhstsw]

So, card unbind.

1st thing i can notice is the fans on the card spinning up like crazy.

Seabios:

char device redirected to /dev/pts/1 (label charserial0)
2023-11-19T14:03:50.200337Z qemu-system-x86_64: vfio: Cannot reset device 0000:01:00.1, no available reset mechanism.
2023-11-19T14:03:50.205360Z qemu-system-x86_64: vfio: Cannot reset device 0000:01:00.1, no available reset mechanism.
2023-11-19T14:03:50.991547Z qemu-system-x86_64: VFIO_MAP_DMA failed: Bad address
2023-11-19T14:03:50.991592Z qemu-system-x86_64: vfio_dma_map(0x147f946c2400, 0xd0000000, 0x10000000, 0x147f84200000) = -14 (Bad address)
2023-11-19T14:03:50.992033Z qemu-system-x86_64: VFIO_MAP_DMA failed: Bad address
2023-11-19T14:03:50.992051Z qemu-system-x86_64: vfio_dma_map(0x147f946c2400, 0xe0000000, 0x200000, 0x147f84000000) = -14 (Bad address)
2023-11-19T14:03:50.992743Z qemu-system-x86_64: VFIO_MAP_DMA failed: Bad address
2023-11-19T14:03:50.992761Z qemu-system-x86_64: vfio_dma_map(0x147f946c2400, 0xfe400000, 0x40000, 0x14809928c000) = -14 (Bad address)
2023-11-19T14:03:50.993306Z qemu-system-x86_64: VFIO_MAP_DMA failed: Bad address
2023-11-19T14:03:50.993323Z qemu-system-x86_64: vfio_dma_map(0x147f946c2400, 0xfe200000, 0x4000, 0x148099288000) = -14 (Bad address)
 

OVMF.

 

-msg timestamp=on
char device redirected to /dev/pts/1 (label charserial0)
2023-11-19T15:32:14.346058Z qemu-system-x86_64: vfio: Cannot reset device 0000:01:00.1, no available reset mechanism.
2023-11-19T15:32:14.362085Z qemu-system-x86_64: vfio: Cannot reset device 0000:01:00.1, no available reset mechanism.

Thx.

Edited November 19, 2023 by dhstsw

[dhstsw]

So, after hours of fiddling, what comes out (at least, in my system) is:

-you have to create and _run_ at least once every new VM *without* the AMD Vendor Reset plugin installed;
-the you can install the plugin, and everything works.

[ich777]

8 minutes ago, dhstsw said:

-you have to create and _run_ at least once every new VM *without* the AMD Vendor Reset plugin installed;
-the you can install the plugin, and everything works.

Sorry but I never heard of that issue because all the plugin does is it installs a Kernel module and loads it on boot.

 

This should not affect any VM. However I'm happy to hear that you've solved your issue.

[dhstsw]

6 minutes ago, ich777 said:

Sorry but I never heard of that issue because all the plugin does is it installs a Kernel module and loads it on boot.

 

This should not affect any VM. However I'm happy to hear that you've solved your issue.

Me being me, it's quite normal things don't go the way they're supposed to

Thanks.

C.

[dhstsw]

Little addition:

Removing radeontop plugin solved the issue with the fans running high.

Cheers.

[ich777]

13 minutes ago, dhstsw said:

Removing radeontop plugin solved the issue with the fans running high.

This is really strange radeontop is just a binary and eventually loads the module on boot if it‘s not already loaded (which it should be).

 

Do the fans start to spin when you visit the dashboard or when you call radeontop? Or when do the fans start spinning?

The Dashboard also calls radeontop if you got GPUStatistics installed.

[dhstsw]

29 minutes ago, ich777 said:

This is really strange radeontop is just a binary and eventually loads the module on boot if it‘s not already loaded (which it should be).

 

Do the fans start to spin when you visit the dashboard or when you call radeontop? Or when do the fans start spinning?

The Dashboard also calls radeontop if you got GPUStatistics installed.

starting with the computer turned on:

-No fans.
-Unraid prompt for selection (with or withour guy/safe mode).
-After a while, during loading, the fans spin to max then stop immediately;
-text on video switched from low res to high res;
-Fans tun on after a while, and like, a minute after turn off.
-Just before Unraid finishes loading (and showing prompt to login) fans start again at mid level;

after this, they may turn off for a while and then start again (and stay on).

(a little clarification: the internal of the computer isn't hot and the gfx card is doing basically nothing).

-Loading a VM (with gfx card passthrough) ends this behaviour, until its stopped. Then the fan starts again.
-Removing the radeontop, this behaviour ceases.

Also, until yesterday (when i was stubbing the card in vfio) all of that wasn't happening (no fans spinning), so i guess it has also something to do with Unraid loading AMD gfx drivers.

C.
 

[ich777]

58 minutes ago, dhstsw said:

Also, until yesterday (when i was stubbing the card in vfio) all of that wasn't happening (no fans spinning), so i guess it has also something to do with Unraid loading AMD gfx drivers.

Do that from a terminal:

mkdir -p /boot/config/modprobe.d
echo "blacklist amdgpu" > /boot/config/modprobe.d/amdgpu.conf

 

This should prevent the drivers from loading, as long as you don't have radeontop installed.

[dhstsw]

Thanks,

in the while i solved with Fan Auto Control.

C.