Re: [PATCH] KVM: X86: Fix dereference null cpufreq policy

[Viresh Kumar]

On 02-03-20, 09:39, Paolo Bonzini wrote:
> On 02/03/20 09:12, Viresh Kumar wrote:
> > On 02-03-20, 08:55, Paolo Bonzini wrote:
> >> On 02/03/20 08:15, Wanpeng Li wrote:
> >>> From: Wanpeng Li <wanpengli@xxxxxxxxxxx>
> >>>
> >>> cpufreq policy which is get by cpufreq_cpu_get() can be NULL if it is failure,
> >>> this patch takes care of it.
> >>>
> >>> Fixes: aaec7c03de (KVM: x86: avoid useless copy of cpufreq policy)
> >>> Reported-by: Naresh Kamboju <naresh.kamboju@xxxxxxxxxx>
> >>> Cc: Naresh Kamboju <naresh.kamboju@xxxxxxxxxx>
> >>> Signed-off-by: Wanpeng Li <wanpengli@xxxxxxxxxxx>
> >>
> >> My bad, I checked kobject_put but didn't check that kobj is first in
> >> struct cpufreq_policy.
> >>
> >> I think we should do this in cpufreq_cpu_put or, even better, move the
> >> kobject struct first in struct cpufreq_policy.  Rafael, Viresh, any
> >> objection?
> >>
> >> Paolo
> >>
> >>>  		policy = cpufreq_cpu_get(cpu);
> >>> -		if (policy && policy->cpuinfo.max_freq)
> >>> -			max_tsc_khz = policy->cpuinfo.max_freq;
> >>> +		if (policy) {
> >>> +			if (policy->cpuinfo.max_freq)
> >>> +				max_tsc_khz = policy->cpuinfo.max_freq;
> >>> +			cpufreq_cpu_put(policy);
> >>> +		}
> > 
> > I think this change makes sense and I am not sure why should we even
> > try to support cpufreq_cpu_put(NULL).
> 
> For the same reason why we support kfree(NULL) and kobject_put(NULL)?

These two helpers are used widely within kernel and many a times the
resource is taken by one routine and dropped by another, and so
someone needed to check if it can call the resource-free helper safely
or not. IMO, that's not the case with cpufreq_cpu_put(). It is used
mostly by the cpufreq core only and not too often by external
entities. And even in that case we don't need to call
cpufreq_cpu_put() from a different routine than the one which called
cpufreq_cpu_get(). Like in your case. And so there is no need of an
extra check to be made.

I don't think we need to support cpufreq_cpu_put(NULL), but if Rafael
wants it to be supported, I won't object to it.

-- 
viresh