Re: [PATCH v2] Receive Packet Steering

[Eric Dumazet <dada1@cosmosbay.com>]

Eric Dumazet a écrit :
> Tom Herbert a écrit :

>> +static ssize_t store_rps_cpus(struct device *dev,
>> +			      struct device_attribute *attr,
>> +			      const char *buf, size_t len)
>> +{
>> +	struct net_device *net = to_net_dev(dev);
>> +	cpumask_t mask;
>> +	int err, cpu;
>> +	int i = 0;
>> +
>> +	if (!capable(CAP_NET_ADMIN))
>> +		return -EPERM;
>> +
>> +	err = bitmap_parse(buf, len, cpumask_bits(&mask), nr_cpumask_bits);
>> +	if (err)
>> +		return err;
>> +
>> +	rtnl_lock();
>> +	if (dev_isalive(net)) {
>> +		if (!net->rps_map) {
>> +			net->rps_map = kzalloc(sizeof(u16) *
>> +			    num_possible_cpus(), GFP_KERNEL);
> 
> num_possible_cpus() is not the max index of a cpu, but the number of possible cpus.
> it can be for example 2, but cpu0 being index 0, and second cpu at index 511
> 
> So I believe you want nr_cpu_ids here
> 
> (num_possible_cpus() <= nr_cpu_ids), not necessarly equal. 
> 
> 
>> +			if (!net->rps_map)
>> +				return -ENOMEM;
>> +		}
>> +		cpus_and(mask, mask, cpu_online_map);
>> +		for_each_cpu_mask(cpu, mask)
>> +			net->rps_map[i++] =  cpu;
>> +		net->rps_map_len = i;
>> +	}
>> +	rtnl_unlock();
>> +
>> +	return len;

I misread patch and my previous remark was wrong, as you put in rps_map
all possible cpu indexes, there is no hole in the array, so its size
is num_possible_cpus() indeed, sorry for the false alarm.

Now I wonder how this can handle cpu hotplug, I see nothing in your patch to transfert
packets ownership or  rps_map[] rebuilding in case a cpu is offlined...