Re: [PATCH] KVM: s390: Fix potential spectre warnings

[Eric Farman <farman@linux.ibm.com>]

On 4/17/19 3:49 AM, David Hildenbrand wrote:
> On 17.04.19 02:54, Eric Farman wrote:
>> Fix some warnings from smatch:
>>
>> arch/s390/kvm/interrupt.c:2310 get_io_adapter() warn: potential spectre issue 'kvm->arch.adapters' [r] (local cap)
>> arch/s390/kvm/interrupt.c:2341 register_io_adapter() warn: potential spectre issue 'dev->kvm->arch.adapters' [w]
>>
>> Signed-off-by: Eric Farman <farman@linux.ibm.com>
>> ---
>> A recent patch from Paolo [1] acted as a reminder (thanks, Christian!)
>> that I had one for the s390 KVM code after some code reviews [2].
>> Let's clean that up.
>>
>> [1] https://patchwork.kernel.org/patch/10895463/
>> [2] https://patchwork.kernel.org/patch/10788565/#22484223
>> ---
>>   arch/s390/kvm/interrupt.c | 11 +++++++++--
>>   1 file changed, 9 insertions(+), 2 deletions(-)
>>
>> diff --git a/arch/s390/kvm/interrupt.c b/arch/s390/kvm/interrupt.c
>> index 82162867f378..bfd55ad34a3e 100644
>> --- a/arch/s390/kvm/interrupt.c
>> +++ b/arch/s390/kvm/interrupt.c
>> @@ -14,6 +14,7 @@
>>   #include <linux/kvm_host.h>
>>   #include <linux/hrtimer.h>
>>   #include <linux/mmu_context.h>
>> +#include <linux/nospec.h>
>>   #include <linux/signal.h>
>>   #include <linux/slab.h>
>>   #include <linux/bitmap.h>
>> @@ -2307,6 +2308,7 @@ static struct s390_io_adapter *get_io_adapter(struct kvm *kvm, unsigned int id)
>>   {
>>   	if (id >= MAX_S390_IO_ADAPTERS)
>>   		return NULL;
>> +	id = array_index_nospec(id, MAX_S390_IO_ADAPTERS);
>>   	return kvm->arch.adapters[id];
> 
> return kvm->arch.adapters[array_index_nospec(id, MAX_S390_IO_ADAPTERS)];
> 
> should exactly fit into a single line if I am not wrong.

Yeah, just.  As Paolo pointed out, that's not common usage.  Though of 
the four other hits I see, only one of them is the same as this 
instance, in that "id" is passed as a variable and then we immediately 
return with an array entry (even if NULL) rather than doing something 
else in that function.  So maybe all-in-one-line here is a little cleaner.

> 
>>   }
>>   
>> @@ -2320,8 +2322,13 @@ static int register_io_adapter(struct kvm_device *dev,
>>   			   (void __user *)attr->addr, sizeof(adapter_info)))
>>   		return -EFAULT;
>>   
>> -	if ((adapter_info.id >= MAX_S390_IO_ADAPTERS) ||
>> -	    (dev->kvm->arch.adapters[adapter_info.id] != NULL))
>> +	if (adapter_info.id >= MAX_S390_IO_ADAPTERS)
>> +		return -EINVAL;
>> +
>> +	adapter_info.id = array_index_nospec(adapter_info.id,
>> +					     MAX_S390_IO_ADAPTERS);
> 
> I dislike that we are modifying adapter_info here. Can you use a local
> variable instead?

I guess, but adapter_info is a local variable too.  So sanitization this 
way seems fine to me.  But if you dislike it more than I don't care, 
I'll add another local variable.  :)

> 
>> +
>> +	if (dev->kvm->arch.adapters[adapter_info.id] != NULL)
>>   		return -EINVAL;
>>   
>>   	adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
>>
>