Re: [bug report] SVM: nSVM: setup nested msr permission bitmap on nested state load

[Dan Carpenter <dan.carpenter@oracle.com>]

On Wed, Sep 23, 2020 at 04:50:58PM +0300, Maxim Levitsky wrote:
> On Wed, 2020-09-23 at 16:44 +0300, Dan Carpenter wrote:
> > Hello Maxim Levitsky,
> > 
> > The patch 772b81bb2f9b: "SVM: nSVM: setup nested msr permission
> > bitmap on nested state load" from Aug 27, 2020, leads to the
> > following static checker warning:
> > 
> > 	arch/x86/kvm/svm/nested.c:1161 svm_set_nested_state()
> > 	warn: 'ctl' not released on lines: 1152.
> > 
> > arch/x86/kvm/svm/nested.c
> >   1135          if (!(save->cr0 & X86_CR0_PG))
> >   1136                  goto out_free;
> >   1137  
> >   1138          /*
> >   1139           * All checks done, we can enter guest mode.  L1 control fields
> >   1140           * come from the nested save state.  Guest state is already
> >   1141           * in the registers, the save area of the nested state instead
> >   1142           * contains saved L1 state.
> >   1143           */
> >   1144          copy_vmcb_control_area(&hsave->control, &svm->vmcb->control);
> >   1145          hsave->save = *save;
> >   1146  
> >   1147          svm->nested.vmcb = kvm_state->hdr.svm.vmcb_pa;
> >   1148          load_nested_vmcb_control(svm, ctl);
> >   1149          nested_prepare_vmcb_control(svm);
> >   1150  
> >   1151          if (!nested_svm_vmrun_msrpm(svm))
> >   1152                  return -EINVAL;
> > 
> > goto out_free?
> > 
> >   1153  
> >   1154          svm_set_gif(svm, !!(kvm_state->flags & KVM_STATE_NESTED_GIF_SET));
> >   1155  
> >   1156          ret = 0;
> >   1157  out_free:
> >   1158          kfree(save);
> >   1159          kfree(ctl);
> >   1160  
> >   1161          return ret;
> >   1162  }
> > 
> > regards,
> > dan carpenter
> > 
> Which kernel tree is this? 
> 
> This again seems to be the result of other commit
> that made save, ctl to be dynamically allocated. I based my patch on the version
> that allocates both on the stack so no freeing is needed.
> 
> As far as I know from a check I did about week ago, none of branches on 
> git://git.kernel.org/pub/scm/virt/kvm/kvm.git had that patch (that made save/ctr
> be allocated dynamically).

This is from yesterday's linux-next.  Look like a merge issue with
commit 6ccbd29ade0d ("KVM: SVM: nested: Don't allocate VMCB structures
on stack").

regards,
dan carpenter