On Mon, 24 May 2021 at 23:32, Sean Christopherson <seanjc@xxxxxxxxxx> wrote: > > On Sun, May 23, 2021, Wanpeng Li wrote: > > From: Wanpeng Li <wanpengli@xxxxxxxxxxx> > > > > Reported by syzkaller: > > > > WARNING: CPU: 7 PID: 10526 at /home/kernel/ssd/linux/arch/x86/kvm//x86.c:7621 x86_emulate_instruction+0x41b/0x510 [kvm] > > RIP: 0010:x86_emulate_instruction+0x41b/0x510 [kvm] > > Call Trace: > > kvm_mmu_page_fault+0x126/0x8f0 [kvm] > > vmx_handle_exit+0x11e/0x680 [kvm_intel] > > vcpu_enter_guest+0xd95/0x1b40 [kvm] > > kvm_arch_vcpu_ioctl_run+0x377/0x6a0 [kvm] > > kvm_vcpu_ioctl+0x389/0x630 [kvm] > > __x64_sys_ioctl+0x8e/0xd0 > > do_syscall_64+0x3c/0xb0 > > entry_SYSCALL_64_after_hwframe+0x44/0xae > > > > Commit 4a1e10d5b5d8c (KVM: x86: handle hardware breakpoints during emulation()) > > adds hardware breakpoints check before emulation the instruction and parts of > > emulation context initialization, actually we don't have EMULTYPE_NO_DECODE flag > > here and the emulation context will not be reused. Commit c8848cee74ff (KVM: x86: > > set ctxt->have_exception in x86_decode_insn()) triggers the warning because it > > catches the stale emulation context has #UD, however, it is not during instruction > > decoding which should result in EMULATION_FAILED. This patch fixes it by moving > > the second part emulation context initialization before hardware breakpoints check. > > > > syzkaller source: https://syzkaller.appspot.com/x/repro.c?x=134683fdd00000 > > > > Reported-by: syzbot+71271244f206d17f6441@xxxxxxxxxxxxxxxxxxxxxxxxx > > Fixes: 4a1e10d5b5d8 (KVM: x86: handle hardware breakpoints during emulation) > > Signed-off-by: Wanpeng Li <wanpengli@xxxxxxxxxxx> > > --- > > arch/x86/kvm/x86.c | 14 +++++++------- > > 1 file changed, 7 insertions(+), 7 deletions(-) > > > > diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c > > index bbc4e04..eca69f9 100644 > > --- a/arch/x86/kvm/x86.c > > +++ b/arch/x86/kvm/x86.c > > @@ -7552,6 +7552,13 @@ int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type, > > > > init_emulate_ctxt(vcpu); > > > > + ctxt->interruptibility = 0; > > + ctxt->have_exception = false; > > + ctxt->exception.vector = -1; > > + ctxt->perm_ok = false; > > What about moving this block all the way into init_emulate_ctxt()? > > > + ctxt->ud = emulation_type & EMULTYPE_TRAP_UD; > > This can be left where it is since ctxt->ud is consumed only by x86_decode_insn(). > I don't have a strong preference as it really only matters for the backport. For > upstream, we can kill it off in a follow-up patch by passing emulation_type to > x86_decode_insn() and dropping ctxt->ud altogether. Tracking that info in ctxt > for literally one call is silly. Good suggestion, will do in v2. Wanpeng
