CVE-2024-39483

In the Linux kernel, the following vulnerability has been resolved:

KVM: SVM: WARN on vNMI + NMI window iff NMIs are outright masked

When requesting an NMI window, WARN on vNMI support being enabled if and only if NMIs are actually masked, i.e. if the vCPU is already handling an NMI. KVMs ABI for NMIs that arrive simultanesouly (from KVMs point of view) is to inject one NMI and pend the other. When using vNMI, KVM pends the second NMI simply by setting V_NMI_PENDING, and lets the CPU do the rest (hardware automatically sets V_NMI_BLOCKING when an NMI is injected).

However, if KVM cant immediately inject an NMI, e.g. because the vCPU is in an STI shadow or is running with GIF=0, then KVM will request an NMI window and trigger the WARN (but still function correctly).

Whether or not the GIF=0 case makes sense is debatable, as the intent of KVMs behavior is to provide functionality that is as close to real hardware as possible. E.g. if two NMIs are sent in quick succession, the probability of both NMIs arriving in an STI shadow is infinitesimally low on real hardware, but significantly larger in a virtual environment, e.g. if the vCPU is preempted in the STI shadow. For GIF=0, the argument isnt as clear cut, because the window where two NMIs can collide is much larger in bare metal (though still small).

That said, KVM should not have divergent behavior for the GIF=0 case based on whether or not vNMI support is enabled. And KVM has allowed simultaneous NMIs with GIF=0 for over a decade, since commit 7460fb4a3400 (KVM: Fix simultaneous NMIs). I.e. KVMs GIF=0 handling shouldnt be modified without a really good reason to do so, and if KVMs behavior were to be modified, it should be done irrespective of vNMI support.

Affected Software

References

• https://git.kernel.org/stable/c/1d87cf2eba46deaff6142366127f2323de9f84d1
• https://git.kernel.org/stable/c/b4bd556467477420ee3a91fbcba73c579669edc6
• https://git.kernel.org/stable/c/f79edaf7370986d73d204b36c50cc563a4c0f356