Test F) only "dom0_max_vcpus=6" , update-grub , reboot @ 5900x Zen3 system * boot to XEN, works got in via SSH (amdgpu problem still on this system)
$ cat /sys/devices/system/cpu/possible 0-5 $ nproc 6 xl info & xentop commands work Previous test A2 "dom0_max_vcpus=12" * nproc result: 12 * xl info - functional Test G) only "nvme.poll_queues=24" , update-grub , reboot @ 5900x Zen3 system * boot to XEN, works got in via SSH (amdgpu problem still on this system) Yea the bug is only the combination of using both same time as in title. Test H) "dom0_max_vcpus=6" and "nvme.poll_queues=6" , update-grub , reboot @ 5900x Zen3 system * boot to XEN, works (huh??) got in via SSH (amdgpu problem still on this system) So the issue for XEN kernel specifically is setting more queues than vcpus/threads ? EVEN if nvme.io_poll=1 is NOT defined. Oooh ok ok. I think i landed on this from some tutrial like: https://xinnor.io/blog/pcie-gen-5-storage-array-optimization-overcoming-the-obstacles-to-maximize-performance/ There is no relation CPU core count for this parameter and i tested NVME performance without XEN, and landed on the best performing count which for the server was indeed the actual core count. https://www.phoronix.com/news/NVMe-IO-Polling-Queues-Test Aha this mentions relation to cores, but no warning that you can not add more. I could not find any documentation nvme.poll_queues ,and AI failed to find working link. And https://docs.kernel.org/admin-guide/kernel-parameters.html has nothing on it. Thank you for the suggestion. Test J) "dom0_max_vcpus=6" and "nvme.poll_queues=6" and "nvme.io_poll=1" , update-grub , reboot @ 5900x Zen3 system * boot to XEN, works :-) got in via SSH (amdgpu problem still on this system) $ cat /sys/devices/system/cpu/possible 0-5 $ sudo xl info | grep xen_v xen_version : 4.20.3-pre $ cat /sys/module/nvme/parameters/poll_queues 6 $ cat /proc/cmdline placeholder root=UUID=*** ro nvme.poll_queues=6 nvme.io_poll=1 $ sudo dmesg | grep -iE 'io_poll' ... [ 2.982008] nvme: unknown parameter 'io_poll' ignored $ sudo dmesg | grep -E 'queues' ... [ 3.035362] nvme nvme0: 6/0/6 default/read/poll queues $ cat /sys/block/nvme0n1/queue/io_poll 1 I guess "nvme.io_poll=1" is redundant in new 7.x kernels now, queues seem to be set up on nvme. -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/2170270 Title: Xen Dom0 fails to boot with "nvme.poll_queues=24" and "dom0_max_vcpus=6 dom0_vcpus_pin" To manage notifications about this bug go to: https://bugs.launchpad.net/linux/+bug/2170270/+subscriptions -- ubuntu-bugs mailing list [email protected] https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
