Re: [kbuild] arch/x86/kvm/x86.c:4988 kvm_arch_tsc_set_attr() warn: check for integer overflow 'offset'

[Sean Christopherson]

On Mon, Oct 10, 2022, Dan Carpenter wrote:
> tree:   https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git   master
> head:   493ffd6605b2d3d4dc7008ab927dba319f36671f
> commit: 828ca89628bfcb1b8f27535025f69dd00eb55207 KVM: x86: Expose TSC offset controls to userspace
> config: x86_64-randconfig-m001-20221010
> compiler: gcc-11 (Debian 11.3.0-5) 11.3.0
> 
> If you fix the issue, kindly add following tag where applicable
> | Reported-by: kernel test robot <lkp@xxxxxxxxx>
> | Reported-by: Dan Carpenter <dan.carpenter@xxxxxxxxxx>
> 
> smatch warnings:
> arch/x86/kvm/x86.c:4988 kvm_arch_tsc_set_attr() warn: check for integer overflow 'offset'
> 
> vim +/offset +4988 arch/x86/kvm/x86.c
> 
> 828ca89628bfcb Oliver Upton 2021-09-16  4962  static int kvm_arch_tsc_set_attr(struct kvm_vcpu *vcpu,
> 828ca89628bfcb Oliver Upton 2021-09-16  4963  				 struct kvm_device_attr *attr)
> 828ca89628bfcb Oliver Upton 2021-09-16  4964  {
> 828ca89628bfcb Oliver Upton 2021-09-16  4965  	u64 __user *uaddr = (u64 __user *)(unsigned long)attr->addr;
> 828ca89628bfcb Oliver Upton 2021-09-16  4966  	struct kvm *kvm = vcpu->kvm;
> 828ca89628bfcb Oliver Upton 2021-09-16  4967  	int r;
> 828ca89628bfcb Oliver Upton 2021-09-16  4968  
> 828ca89628bfcb Oliver Upton 2021-09-16  4969  	if ((u64)(unsigned long)uaddr != attr->addr)
> 828ca89628bfcb Oliver Upton 2021-09-16  4970  		return -EFAULT;
> 828ca89628bfcb Oliver Upton 2021-09-16  4971  
> 828ca89628bfcb Oliver Upton 2021-09-16  4972  	switch (attr->attr) {
> 828ca89628bfcb Oliver Upton 2021-09-16  4973  	case KVM_VCPU_TSC_OFFSET: {
> 828ca89628bfcb Oliver Upton 2021-09-16  4974  		u64 offset, tsc, ns;
> 828ca89628bfcb Oliver Upton 2021-09-16  4975  		unsigned long flags;
> 828ca89628bfcb Oliver Upton 2021-09-16  4976  		bool matched;
> 828ca89628bfcb Oliver Upton 2021-09-16  4977  
> 828ca89628bfcb Oliver Upton 2021-09-16  4978  		r = -EFAULT;
> 828ca89628bfcb Oliver Upton 2021-09-16  4979  		if (get_user(offset, uaddr))
> 828ca89628bfcb Oliver Upton 2021-09-16  4980  			break;
> 828ca89628bfcb Oliver Upton 2021-09-16  4981  
> 828ca89628bfcb Oliver Upton 2021-09-16  4982  		raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
> 828ca89628bfcb Oliver Upton 2021-09-16  4983  
> 828ca89628bfcb Oliver Upton 2021-09-16  4984  		matched = (vcpu->arch.virtual_tsc_khz &&
> 828ca89628bfcb Oliver Upton 2021-09-16  4985  			   kvm->arch.last_tsc_khz == vcpu->arch.virtual_tsc_khz &&
> 828ca89628bfcb Oliver Upton 2021-09-16  4986  			   kvm->arch.last_tsc_offset == offset);
> 828ca89628bfcb Oliver Upton 2021-09-16  4987  
> 828ca89628bfcb Oliver Upton 2021-09-16 @4988  		tsc = kvm_scale_tsc(vcpu, rdtsc(), vcpu->arch.l1_tsc_scaling_ratio) + offset;
> 
> Smatch hates obvious user triggerable integer overflows...  No checking
> on offset.

This is ok, and even necessary, e.g. if the host TSC > guest TSC.  Is there anything
we can do in KVM to help Smatch avoid false positives?  Or do you/Smatch already
maintain a list of known false positives?