System crash when SRIOV VF is assigned to a guest VM and VM is powered on

This document (000019898) is provided subject to the disclaimer at the end of this document.

Environment

SUSE Linux Enterprise Server 15 SP2

Situation

This issue can occur on an AMD CPU based system with a NIC with SRIOV capability.
With SLES 15 SP2 installed and a guest VM with SRIOV VF assigned, the host system may crash when VM is powered on.

The system will have had IOMMU enabled in system BIOS and 'amd_iommu=on' parameter passed to the kernel command line.

The crash will show a call trace like this:

[ 302.402993] Call Trace:
[ 302.403001] svm_update_pi_irte.part.60+0x3ea/0x550 [kvm_amd]
[ 302.403009] ? irq_setup_affinity+0xc4/0x150
[ 302.403014] __connect+0x82/0xa0 [irqbypass]
[ 302.403017] irq_bypass_register_producer+0xeb/0x130 [irqbypass]
[ 302.403021] vfio_msi_set_vector_signal+0x1ab/0x290 [vfio_pci]
[ 302.403023] vfio_msi_set_block+0x75/0xd0 [vfio_pci]
[ 302.403025] vfio_pci_ioctl+0x253/0xc10 [vfio_pci]
[ 302.403029] ? __seccomp_filter+0x7a/0x6a0
[ 302.403036] do_vfs_ioctl+0xa0/0x680
[ 302.403043] ? syscall_trace_enter+0xc2/0x2b0
[ 302.403045] ksys_ioctl+0x70/0x80
[ 302.403047] __x64_sys_ioctl+0x16/0x20
[ 302.403048] do_syscall_64+0x5b/0x1e0
[ 302.403055] entry_SYSCALL_64_after_hwframe+0x44/0xa9

Resolution

A fix for this is included in the maintenance update kernel version 5.3.18-24.49.2 (released Feb/2021) or later.

Cause

This issue is related to the use of an uninitialized structure member, namely amd_iommu_pi_data.prev_ga_tag, that leads to an unintended code path that triggers NULL pointer deference.

Disclaimer

This Support Knowledgebase provides a valuable tool for SUSE customers and parties interested in our products and solutions to acquire information, ideas and learn from one another. Materials are provided for informational, personal or non-commercial use within your organization and are presented "AS IS" WITHOUT WARRANTY OF ANY KIND.