IRQs from KVM use CPU hypervisor system
by Alex_Shved from LinuxQuestions.org on (#5DTB3)
The VM has debian installed. NIC SR-IOV VF. VF on VM has 4 network queues. At the moment when more 4Gbps comes to the server, the IRQ from the NIC (VF) to the VM utilizes the expected +/- 5-15% of the kernel by each thread. On the hypervisor, at the same time, I see the utilization of 4 cores at 100% on space system. Additionally, the server tries to solve this problem and the load goes in turn across all cores, which causes additional costs. Only 1 VM is running on the server
from grub: intel_iommu = on vfio_iommu_type1.allow_unsafe_interrupts = 1
If anyone has any ideas how to solve or in which direction to look, please tell me.
If you have any questions, I will be glad to answer. I have no ideas anymore
hypervisor: Debian 10 | 5.4.0-0.bpo.4-amd64 | libvirt 5.0.0-4+deb10u1 | qemu 3.1+dfsg-8+deb10u8
VM: Debian 10 | 4.19.0.-12


from grub: intel_iommu = on vfio_iommu_type1.allow_unsafe_interrupts = 1
If anyone has any ideas how to solve or in which direction to look, please tell me.
If you have any questions, I will be glad to answer. I have no ideas anymore
hypervisor: Debian 10 | 5.4.0-0.bpo.4-amd64 | libvirt 5.0.0-4+deb10u1 | qemu 3.1+dfsg-8+deb10u8
VM: Debian 10 | 4.19.0.-12