GPU hangs when setting sriov_numvfs #279
Comments
|
Also, if reboot the desktop after Unfortunately the 2 NUCs with same model are the only 2 gadgets I owned which support SR-IOV based Intel vGPU. (not the now-deprecated GVT-g) Thus I have no option to try to reproduce it on another platform/CPU. |
|
I think this might be related to gnome/mutter not handling hotplug of VFs correctly. A possible solution is to delay the startup of gdm or load the i915 module as early as possible and enable VF. |
|
@bbaa-bbaa Thanks for your instruction! I didnot try the first approach (delay gdm after sysfs
#!/bin/bash
build() {
add_module i915
add_runscript
}
help() {
echo i915 with SR-IOV support
}
*: acknowledgement for step 2 & 3: https://mop.koeln/blog/custom-mkinitcpio-hooks/ This issue persists. |
|
Sorry, maybe I did not describe the issue clearly. There are basically 2 fail patterns I noticed. The first one occurs when The related logs looks like: Maybe @bbaa-bbaa proposed suggestions for this fail pattern, as the error log contains something related to GPU hotplugging. The second fail pattern occurs when The related logs looks like: |
|
Perhaps we need to set sriov_numvfs after gdm started and before the gnome session launch. |
|
@bbaa-bbaa Thanks (again) for your suggestion. I actually tried this by creating a shell script to set Some further investigation shows that if gdm is started but no user is logged in (when gdm stays at the login page), and Seems the only way is to set |
|
Probably it's time to give up intel GPU virtualization on my workstation and just move the stacks to my server build, which works like a charm following the README of this repo. :-) Besides, I also tried to set |
I had experienced an issue with i915-sriov-dkms which result in GPU hangs.
My system informations are listed as below:
My system is Intel NUC11PAHi5, which has an i5-1135G7 CPU and Xe iGPU, and OS is Arch Linux, with Gnome as DE.
The procedure below is committed to enable SR-IOV for iGPU:
i915-sriov-dkmsfrom AURintel_iommu=on iommu=ptto kernel command line/etc/modprobe.d/i915.conf:/etc/tmpfiles.d/i915-set-sriov-numvfs.conf:After rebooting, the gdm hangs while switching to another tty with
Control+Alt+Fxis possible.I also tried comment out the line in
/etc/tmpfiles.d/i915-set-sriov-numvfs.confwhich setssriov_numvfs, after reboot, the system worked fine.I tried to manually set this by invoking
echo 7 | sudo tee /sys/devices/pci0000:00/0000:00:02.0/sriov_numvfs.After doing so, gnome crashed, but everything worked like a charm once gnome restarted.
I own another NUC unit with identical model, which runs as a server without any DE. (it ends up
logintty after booted up) I also tried the procedure above with this system, everything seems worked fine, no hangs, no freeze.I also swap the SSDs (which Arch Linux installations reside on) of the 2 systems, The server (now with the desktop's SSD) hangs when launching gdm, the desktop (now with the server's SSD) works perfectly. The purpose doing so is check potential hardware issue.
The text was updated successfully, but these errors were encountered: