On 8/17/23 15:01, Nuno Das Neves wrote:
> +static inline union hv_proximity_domain_info
> +numa_node_to_proximity_domain_info(int node)
> +{
> + union hv_proximity_domain_info proximity_domain_info;
> +
> + if (node != NUMA_NO_NODE) {
> + proximity_domain_info.domain_id = node_to_pxm(node);
> + proximity_domain_info.flags.reserved = 0;
> + proximity_domain_info.flags.proximity_info_valid = 1;
> + proximity_domain_info.flags.proximity_preferred = 1;
> + } else {
> + proximity_domain_info.as_uint64 = 0;
> + }
> +
> + return proximity_domain_info;
> +}
Pop quiz: What are the rules for the 30 bits of uninitialized data of
proximity_domain_info.flags in the (node != NUMA_NO_NODE) case?
I actually don't know off the top of my head. I generally avoid
bitfields, but if they were normal stack-allocated variable space,
they'd be garbage.
I'd also *much* rather see the "as_uint64 = 0" coded up as a memset() or
even explicitly zeroing all the same variables as the other half of the
if(). As it stands, it's not 100% obvious that proximity_domain_info is
64 bits and that .as_uint64=0 zeroes the whole thing. It *WOULD* be
totally obvious if it were a memset().
