Re: [PATCH 00/43] KVM: x86: vCPU RESET/INIT fixes and consolidation

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On 24/04/21 02:46, Sean Christopherson wrote:
For the record, I went into this thinking it was going to be a simple code
shuffle between {svm,vmx}_vcpu_reset() and kvm_vcpu_reset().  The actual
goal is to consolidate the RESET/INIT code, both to deduplicate code and
to try to avoid divergent behavior/bugs, e.g. SVM only recently started
updating vcpu->arch.cr4 on INIT.

The TL;DR of why it takes 40+ patches to get there is that the RESET/INIT
flows have multiple latent bugs and hidden dependencies, but "work"
because they're rarely touched, are mostly fixed flows in both KVM and the
guest, and because guests don't sanity check state after INIT.

While several of the patches have Fixes tags, I am absolutely terrified of
backporting most of them due to the likelihood of breaking a different
version of KVM.  And, for the most part the bugs are benign in the sense
no guest has actually encountered any of these bugs.  For that reason, I
intentionally omitted stable@ entirely.  The only patches I would consider
even remotely safe for backporting are the first two patches in the series.


Sean Christopherson (43):
   KVM: nVMX: Set LDTR to its architecturally defined value on nested
     VM-Exit
   KVM: VMX: Set EDX at INIT with CPUID.0x1, Family-Model-Stepping
   KVM: SVM: Require exact CPUID.0x1 match when stuffing EDX at INIT
   KVM: SVM: Fall back to KVM's hardcoded value for EDX at RESET/INIT
   KVM: x86: Split out CR0/CR4 MMU role change detectors to separate
     helpers
   KVM: x86: Properly reset MMU context at vCPU RESET/INIT
   KVM: VMX: Remove explicit MMU reset in enter_rmode()
   KVM: SVM: Drop explicit MMU reset at RESET/INIT
   KVM: SVM: Drop a redundant init_vmcb() from svm_create_vcpu()
   KVM: VMX: Move init_vmcs() invocation to vmx_vcpu_reset()
   KVM: x86: WARN if the APIC map is dirty without an in-kernel local
     APIC
   KVM: x86: Remove defunct BSP "update" in local APIC reset
   KVM: x86: Migrate the PIT only if vcpu0 is migrated, not any BSP
   KVM: x86: Don't force set BSP bit when local APIC is managed by
     userspace
   KVM: x86: Set BSP bit in reset BSP vCPU's APIC base by default
   KVM: VMX: Stuff vcpu->arch.apic_base directly at vCPU RESET
   KVM: x86: Open code necessary bits of kvm_lapic_set_base() at vCPU
     RESET
   KVM: x86: Consolidate APIC base RESET initialization code
   KVM: x86: Move EDX initialization at vCPU RESET to common code
   KVM: SVM: Don't bother writing vmcb->save.rip at vCPU RESET/INIT
   KVM: VMX: Invert handling of CR0.WP for EPT without unrestricted guest
   KVM: VMX: Remove direct write to vcpu->arch.cr0 during vCPU RESET/INIT
   KVM: VMX: Fold ept_update_paging_mode_cr0() back into vmx_set_cr0()
   KVM: nVMX: Do not clear CR3 load/store exiting bits if L1 wants 'em
   KVM: VMX: Pull GUEST_CR3 from the VMCS iff CR3 load exiting is
     disabled
   KVM: VMX: Process CR0.PG side effects after setting CR0 assets
   KVM: VMX: Skip emulation required checks during pmode/rmode
     transitions
   KVM: nVMX: Don't evaluate "emulation required" on VM-Exit
   KVM: SVM: Tweak order of cr0/cr4/efer writes at RESET/INIT
   KVM: SVM: Drop redundant writes to vmcb->save.cr4 at RESET/INIT
   KVM: SVM: Stuff save->dr6 at during VMSA sync, not at RESET/INIT
   KVM: VMX: Skip pointless MSR bitmap update when setting EFER
   KVM: VMX: Refresh list of user return MSRs after setting guest CPUID
   KVM: VMX: Don't _explicitly_ reconfigure user return MSRs on vCPU INIT
   KVM: x86: Move setting of sregs during vCPU RESET/INIT to common x86
   KVM: VMX: Remove obsolete MSR bitmap refresh at vCPU RESET/INIT
   KVM: nVMX: Remove obsolete MSR bitmap refresh at nested transitions
   KVM: VMX: Don't redo x2APIC MSR bitmaps when userspace filter is
     changed
   KVM: VMX: Remove unnecessary initialization of msr_bitmap_mode
   KVM: VMX: Smush x2APIC MSR bitmap adjustments into single function
   KVM: VMX: Remove redundant write to set vCPU as active at RESET/INIT
   KVM: VMX: Drop VMWRITEs to zero fields at vCPU RESET
   KVM: x86: Drop pointless @reset_roots from kvm_init_mmu()

  arch/x86/include/asm/kvm_host.h |   5 -
  arch/x86/kvm/i8254.c            |   3 +-
  arch/x86/kvm/lapic.c            |  26 +--
  arch/x86/kvm/mmu.h              |   2 +-
  arch/x86/kvm/mmu/mmu.c          |  13 +-
  arch/x86/kvm/svm/nested.c       |   2 +-
  arch/x86/kvm/svm/sev.c          |   1 +
  arch/x86/kvm/svm/svm.c          |  33 +---
  arch/x86/kvm/vmx/nested.c       |  26 ++-
  arch/x86/kvm/vmx/vmx.c          | 271 +++++++++++++-------------------
  arch/x86/kvm/vmx/vmx.h          |   5 +-
  arch/x86/kvm/x86.c              |  51 +++++-
  12 files changed, 189 insertions(+), 249 deletions(-)


I'm waiting for a v2 of this; it applies with relatively few conflicts, but there were some comments so it's better if you take care of updating it.

Paolo




[Index of Archives]     [KVM ARM]     [KVM ia64]     [KVM ppc]     [Virtualization Tools]     [Spice Development]     [Libvirt]     [Libvirt Users]     [Linux USB Devel]     [Linux Audio Users]     [Yosemite Questions]     [Linux Kernel]     [Linux SCSI]     [XFree86]

  Powered by Linux