On 29/07/21 01:08, Zubin Mithra wrote:

From: Maxim Levitsky mlevitsk@redhat.com

commit b97f074583736c42fb36f2da1164e28c73758912 upstream.

A page fault can be queued while vCPU is in real paged mode on AMD, and AMD manual asks the user to always intercept it (otherwise result is undefined). The resulting VM exit, does have an error code.

Signed-off-by: Maxim Levitsky mlevitsk@redhat.com Message-Id: 20210225154135.405125-2-mlevitsk@redhat.com Signed-off-by: Paolo Bonzini pbonzini@redhat.com Signed-off-by: Zubin Mithra zsm@chromium.org

---

Backport Note:

• Syzkaller triggered a WARNING with the following stacktrace: WARNING: CPU: 3 PID: 3489 at arch/x86/kvm/x86.c:8097 kvm_arch_vcpu_ioctl_run+0x42b/0x33d7 Kernel panic - not syncing: panic_on_warn set ...

  CPU: 3 PID: 3489 Comm: poc Not tainted 4.19.199-00120-ga89b48fe9308 #5 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.14.0-2 04/01/2014 Call Trace: dump_stack+0x11b/0x19a panic+0x188/0x2f1 ? __warn_printk+0xee/0xee ? __probe_kernel_read+0xad/0xc8 ? kvm_arch_vcpu_ioctl_run+0x42b/0x33d7 ? kvm_arch_vcpu_ioctl_run+0x42b/0x33d7 __warn+0xe9/0x12a ? kvm_arch_vcpu_ioctl_run+0x42b/0x33d7 report_bug+0x93/0xdd fixup_bug+0x28/0x4b do_error_trap+0xd1/0x1f5 ? fixup_bug+0x4b/0x4b ? tick_nohz_tick_stopped+0x2e/0x39 ? __irq_work_queue_local+0x50/0x96 ? preempt_count_sub+0xf/0xbf ? lockdep_hardirqs_off+0x10c/0x115 ? error_entry+0x72/0xd0 ? trace_hardirqs_off_caller+0x56/0x116 ? trace_hardirqs_off_thunk+0x1a/0x1c invalid_op+0x14/0x20 RIP: 0010:kvm_arch_vcpu_ioctl_run+0x42b/0x33d7

• This commit is present in linux-5.13.y.

• Conflict arises as the following commit is is not present in

linux-5.10.y and older.

• b3646477d458 ("KVM: x86: use static calls to reduce kvm_x86_ops overhead")

• Tests run: syzkaller reproducer, Chrome OS tryjobs

  arch/x86/kvm/x86.c | 13 +++++++++---- 1 file changed, 9 insertions(+), 4 deletions(-)

diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c index 43fb4e296d8d..9cfc669b4a24 100644 --- a/arch/x86/kvm/x86.c +++ b/arch/x86/kvm/x86.c @@ -416,8 +416,6 @@ static void kvm_multiple_exception(struct kvm_vcpu *vcpu, if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) { queue:

• if (has_error && !is_protmode(vcpu))
  

• 	has_error = false;
  

  if (reinject) { /* * On vmentry, vcpu->arch.exception.pending is only

@@ -7114,6 +7112,13 @@ static void update_cr8_intercept(struct kvm_vcpu *vcpu) kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr); } +static void kvm_inject_exception(struct kvm_vcpu *vcpu) +{

•   if (vcpu->arch.exception.error_code && !is_protmode(vcpu))
  

•           vcpu->arch.exception.error_code = false;
  

•   kvm_x86_ops->queue_exception(vcpu);
  

+}

• static int inject_pending_event(struct kvm_vcpu *vcpu) { int r;

@@ -7121,7 +7126,7 @@ static int inject_pending_event(struct kvm_vcpu *vcpu) /* try to reinject previous events if any */ if (vcpu->arch.exception.injected)

• kvm_x86_ops->queue_exception(vcpu);
  

• kvm_inject_exception(vcpu);
  

  /*
  • Do not inject an NMI or interrupt if there is a pending
  • exception. Exceptions and interrupts are recognized at

@@ -7175,7 +7180,7 @@ static int inject_pending_event(struct kvm_vcpu *vcpu) kvm_update_dr7(vcpu); }

• kvm_x86_ops->queue_exception(vcpu);
  

• kvm_inject_exception(vcpu);
  

  }

/* Don't consider new event if we re-injected an event */

Acked-by: Paolo Bonzini pbonzini@redhat.com