Jean-Philippe Brucker <jean-philippe.brucker@xxxxxxx> writes:
> Hi Punit,
>
> On 05/03/18 15:11, Punit Agrawal wrote:
>>> + devs = realloc(kvm->cfg.vfio_devices, sizeof(*dev) * (idx + 1));
>>> + if (!devs) {
>>> + ret = -ENOMEM;
>>> + goto out_free_buf;
>>> + }
>>> +
>>> + kvm->cfg.vfio_devices = devs;
>>
>> I noticed that vfio_devices never gets freed. I'm not sure if this needs
>> fixing as I noticed a similar pattern in virtio as well.
>
> I think kvmtool is pretty lax when it comes to freeing permanent objects,
> or cleaning up before dying. It often relies on the kernel to clean up on
> process exit. The downside is that finding runtime memory leaks with
> valgrind becomes a bit more difficult (on my current setup for example, I
> can see 1.4M still reachable). So it's not worth doing tree-wide, but I
> don't mind fixing this one.
Just to be clear, I wasn't suggesting for you to fix this issue
tree-wide. :)
> I think it's safe to free in vfio__exit.
It would be great if you can fix it for the patch though - considering
that all the other allocations are dealt with symmetrically, it'd be a
shame to leave this one out.
>
>> The previous patches in the series look fine. I'll have a look at the
>> rest in the next couple of days.
>
> Thanks!
> Jean
