This series aims to close the loop on a prior conversation on this matter. I have picked this up from Michal Wilczynski who had proposed different fixes (v1 and v2). v2: https://lore.kernel.org/all/20240123001555.4168188-1-michal.wilczynski@xxxxxxxxx/ v1: https://lore.kernel.org/all/20231222164543.918037-1-michal.wilczynski@xxxxxxxxx/ The issue was initially reported here: https://lore.kernel.org/all/CAMhUBjmXMYsEoVYw_M8hSZjBMHh24i88QYm-RY6HDta5YZ7Wgw@xxxxxxxxxxxxxx/ It is caused by setting nested_run_pending in the vendor-specific leave_smm() callback from the RSM emulation. The syzkaller test produced a triple fault in rsm_load_state_64() resulting in a nested VM-Exit with nested_run_pending being set and triggered the warnings. The commit message for patch 2 has a detailed description of the flow. The patches do the following: a) Move nested_run_pending out of vendor structs and into the x86 kvm_vcpu_arch so it can be accessed by common x86 code (e.g., the SMM emulation). The usage and semantics of this flag are common between SVM and VMX. b) Set nested_run_pending only after a successful RSM emulation. This evidently resolves the issue, but I would appreciate feedback (if the patches are acceptable) and/or suggestions. Kishen Maloor (2): KVM: x86: nSVM/nVMX: Move nested_run_pending to kvm_vcpu_arch KVM: x86: nSVM/nVMX: Fix RSM logic leading to L2 VM-Entries arch/x86/include/asm/kvm_host.h | 1 + arch/x86/kvm/smm.c | 12 ++++++++-- arch/x86/kvm/svm/nested.c | 14 +++++------ arch/x86/kvm/svm/svm.c | 12 ++++------ arch/x86/kvm/svm/svm.h | 4 ---- arch/x86/kvm/vmx/nested.c | 42 ++++++++++++++++----------------- arch/x86/kvm/vmx/vmx.c | 13 +++++----- arch/x86/kvm/vmx/vmx.h | 3 --- 8 files changed, 50 insertions(+), 51 deletions(-) -- 2.31.1
