]> git.ipfire.org Git - thirdparty/kernel/stable.git/commit
KVM: SVM: Don't skip unrelated instruction if INT3/INTO is replaced
authorOmar Sandoval <osandov@fb.com>
Tue, 4 Nov 2025 17:55:26 +0000 (09:55 -0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Fri, 12 Dec 2025 17:42:45 +0000 (18:42 +0100)
commit53903ac9ca1abffa27327e85075ec496fa55ccf3
treee72f50d3f5e1812d89a11db17288a1fea6fb0a33
parent877adccfacb32687b90714a27cfb09f444fdfa16
KVM: SVM: Don't skip unrelated instruction if INT3/INTO is replaced

commit 4da3768e1820cf15cced390242d8789aed34f54d upstream.

When re-injecting a soft interrupt from an INT3, INT0, or (select) INTn
instruction, discard the exception and retry the instruction if the code
stream is changed (e.g. by a different vCPU) between when the CPU
executes the instruction and when KVM decodes the instruction to get the
next RIP.

As effectively predicted by commit 6ef88d6e36c2 ("KVM: SVM: Re-inject
INT3/INTO instead of retrying the instruction"), failure to verify that
the correct INTn instruction was decoded can effectively clobber guest
state due to decoding the wrong instruction and thus specifying the
wrong next RIP.

The bug most often manifests as "Oops: int3" panics on static branch
checks in Linux guests.  Enabling or disabling a static branch in Linux
uses the kernel's "text poke" code patching mechanism.  To modify code
while other CPUs may be executing that code, Linux (temporarily)
replaces the first byte of the original instruction with an int3 (opcode
0xcc), then patches in the new code stream except for the first byte,
and finally replaces the int3 with the first byte of the new code
stream.  If a CPU hits the int3, i.e. executes the code while it's being
modified, then the guest kernel must look up the RIP to determine how to
handle the #BP, e.g. by emulating the new instruction.  If the RIP is
incorrect, then this lookup fails and the guest kernel panics.

The bug reproduces almost instantly by hacking the guest kernel to
repeatedly check a static branch[1] while running a drgn script[2] on
the host to constantly swap out the memory containing the guest's TSS.

[1]: https://gist.github.com/osandov/44d17c51c28c0ac998ea0334edf90b5a
[2]: https://gist.github.com/osandov/10e45e45afa29b11e0c7209247afc00b

Fixes: 6ef88d6e36c2 ("KVM: SVM: Re-inject INT3/INTO instead of retrying the instruction")
Cc: stable@vger.kernel.org
Co-developed-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Link: https://patch.msgid.link/1cc6dcdf36e3add7ee7c8d90ad58414eeb6c3d34.1762278762.git.osandov@fb.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
arch/x86/include/asm/kvm_host.h
arch/x86/kvm/svm/svm.c
arch/x86/kvm/x86.c