gim error:(set_new_adapter:617) asic does not support SRIOV #3
Comments
|
Hi James, Would you be able to share the output of: lspci -vvv -d 1002:6829 Thanks, Emil |
|
Hi Emil, Thanks for reaching out! Here is the output for lspci -vvv -d 1002:6829: Nothing appeared to output. Our AMD FirePro is 1002:6929. Here is the output for lspci -vvv -d 1002:6929: |
|
Hi James, Can you confirm that you have patch 0001 built and installed? For the patch you can actually completely remove the check for end point type. See the same file in the v4.6 kernel for reference. This fix has been upstreamed as of v4.6 Thanks, |
|
Hi Kelly, I'll verify right now. Thanks, |
|
Hi Kelly, Everytime I run the apt-get source linux-image-$(uname -r), it disregards my 4.4.0-75-generic and installs a new 4.4.0 kernel. This is a step in the docs/environment_setup file. Any advice? Thanks! |
|
Hi James, I have never used apt-get to get the kernel. I always download the .tar.gz from kernel.org (https://mirrors.edge.kernel.org/pub/linux/kernel/v4.x/). gunzip the .gz file then 'tar -xf' the results. How comfortable are you with rebuilding the kernel? Thanks, |
|
Hi Kelly, I don't have much experience in building custom kernels. However, I followed an article and learned more about it! :) I'm running Thanks again for your assistance! :) Sincerely, |
|
Hi Kelly, I think I got the kernel going. I installed the gim module and added it to /etc/modules. There were no kernel panics in dmesg: When I do a I guess I'm wondering where I go from here. Thanks, |
|
Hi Kelly, Good news! I think we're making progress :) Here is my dmesg output, once we booted with the FirePro, gim, and updated kernel: Now that it's enumerating VF's, how do you define them in a virsh XML file? Thanks, |
|
hi，falcon78921 ，i also do the work as u do, here is my kernel logs, there are some differences **
May 2 09:17:35 vgpu kernel: gim info:(set_new_adapter:802) init work **
** **
** this a more complete log content |
|
Hi flintcq, As of this morning, I was able to see AMD MxGPU as a device in Device Manager (Windows guest VM). The AMD Pro guest driver installed successfully (https://support.amd.com/en-us/download/workstation?os=Guest%20OS#pro-driver). I used the Radeon™ Pro Software Adrenalin Edition for Windows® 10 (64-bit) driver. This driver was stated specifically for KVM. Unfortunately, we are having stability issues with it. Every time we pass a VF to the guest, Windows will boot but crashes shortly after getting to the desktop. It might be a misconfiguration on our end, still investigating. Your dmesg is a bit different than mine, are you running a different Linux distro? I would love to help other people who are trying to use the AMD FirePro GPU for virtualization. Looking at your dmesg, it appears you're at the same place as me. The gim module is loading successfully and enumerating VF's. I'll keep you posted on my progress. Thanks! |
|
hi falcon78921, in this environment, all my vms can see vgpu, but all are errored with code 43 then i downgrade my qemu to qemy-kvm-1.5.3(yum installed), vms (win 7/10) can recognize vgpus, but there are still some issues exists: [ 1372.644436] gim info:(handle_fullaccess_timeout:1333) begin to check full access timeout for VF7
|
|
@flintcq Would you dump an xml of your windows 10 vm and post that here? I'm working with @falcon78921 on this adventure, and we're experiencing instability with windows 10 - kind of like what you've mentioned for windows 7. We haven't even tried windows 7. We were able to boot the Windows 10 VM after installing the Radeon Driver and it seems to run for 5 minutes, then it locks up the server to the point where we have to power it off to recover. Here's what's on the console when this happens: |
|
hi all,
some tricks we found: some abnormals we also found: the gim error you mentioned didn't show up in our environment, but in our environment, one of my s7150x2 card failed at "init_register_init_state" when load gim driver, we think it maybe a hardware problem or a bug mentioned in gim code [ 588.503468] gim error:(wait_cmd_complete:1648) wait_cmd_complete -- time out after 0.100013123 sec
|
|
additional, each time we start or reboot a new vm, gim would do "check full access" to its vgpu, and each time do a loop switch, which cause all started vms halt for a while |
|
Hi @flintcq @falcon78921, I have the same virtual desktop environment using a S7150 with qemu/kvm. I'm no expert, but I'd love to share my experience and hopefully learn something along the way.
This is the VIRSH XML output for the Win10 VM. |
|
@Tungsten-lawrenceb can u paste the kernel log about gim module, because the errors showed in logs are not always affect the use of vm. |
|
@flintcq kernel logs are as follows: INITIALIZATION [ 6061.613321] [drm] amdgpu kernel modesetting enabled. BAR SETUP Mar 26 12:24:14 vdi kernel: pci 0000:03:00.0: [1002:6929] type 00 class 0x030000 Apr 19 21:16:28 vdi kernel: [ 2.878415] gim: loading out-of-tree module taints kernel. I hope you glean something useful from this data dump! |
|
@flintcq I have used RDP/VNC/SPICE to connect to the VM without any issues. 03:00.0 VGA compatible controller: Advanced Micro Devices, Inc. [AMD/ATI] Tonga XT GL [FirePro S7150] (prog-if 00 [VGA controller]) modinfo gim |
|
@Tungsten-lawrenceb, thanks for your logs and information, we have checked , all of them almost are the same, but the logs you provided is only the initializing phase of gim, not gim output while booting/running/rebooting/shutdowning the vms, most issues we met are occurred during these periods and we wonder why you can use vnc/spice, while black for us, and when using vnc(server)/spice/rdp, the featues of directx will not be opened for vgpu, how you work around, and we use gpu-z for checking vgpu, here is the result, how about yours? |
|
Thanks for the reference file @Tungsten-lawrenceb! I've been busy with other projects. I'm hoping to continue soon. |
|
Hi @falcon78921 Did you solve it? |
|
Hi @fanghuafan, What kernel are you using? Could you output I had to rebuild a Linux kernel with SRIOV support. The patches supplied in the git repo allow you to do this. According to @kzytaruk, the patches have been upstream'd as of kernel 4.6. Here is a good reference article, if you need assistance building kernels: If you're using Ubuntu, you can download kernels from http://kernel.ubuntu.com/~kernel-ppa/mainline/. You can also download the latest kernels from kernel.org. |
|
Hi @falcon78921 I'll try your method. |
Hello!
We are trying to use an AMD FirePro S7150x2 in a Ubuntu 16.04.2 system. Here is some information regarding the setup:
OS:
Ubuntu 16.04.2Hypervisor:
qemu-kvm 2.5.0 with libvirt 1.3.1Kernel:
4.4.0-75-genericI followed the README.md on the gim repo. I was able to load the kernel module:
I was able to get one of the patches installed (0001), but not the other (0002). We're trying to utilize the FirePro for VDI virtualization, exactly as stated on the AMD website (https://pro.radeon.com/en/solutions/vdi/).
However, we are having problems fully comprehending the documentation. I guess I'm just seeing if anyone could enlighten me on how I can utilize this hardware for a VDI infrastructure. I know for 100% certainty our hardware supports SRIOV.
I am more than happy to provide information, if requested.
There were some errors that I found in our dmesg:
Any help is much appreciated!
The text was updated successfully, but these errors were encountered: