Re: [PATCH V3 1/9] nl80211: allow multiple active scheduled scan requests

[Arend van Spriel <arend.vanspriel@broadcom.com>]

On 4/25/2017 9:49 PM, Johannes Berg wrote:
> 
>>   		if (schedule_destroy_work) {
>> -			struct cfg80211_iface_destroy *destroy;
>> +			struct cfg80211_nlport_release *destroy;
>>   
>>   			destroy = kzalloc(sizeof(*destroy),
>> GFP_ATOMIC);
> 
> I was never really happy with this allocation and really want to get
> rid of it ...

Agree. I don't like the allocation overhead either for deferred 
processing in another context.

>> +		if (schedule_sched_stop_work) {
>> +			struct cfg80211_nlport_release *destroy;
>> +
>> +			destroy = kzalloc(sizeof(*destroy),
>> GFP_ATOMIC);
>> +			if (destroy) {
>> +				destroy->nlportid = notify->portid;
>> +				spin_lock(&rdev-
>>> sched_stop_list_lock);
>> +				list_add(&destroy->list, &rdev-
>>> sched_stop_list);
>> +				spin_unlock(&rdev-
>>> sched_stop_list_lock);
>> +				schedule_work(&rdev-
>>> sched_scan_stop_wk);
>> +			}
>> +		}
> 
> Can't we set some kind of flag in the struct
> cfg80211_sched_scan_request and iterate through them later again,
> checking which need to be destroyed?

Sure. In the worker we are already iterating so I guess we could. The 
number of internal fields in struct cfg80211_sched_scan_request with one 
of them being the list_head. Thinking about hiding those from the 
drivers. Is it worth it?

Regards,
Arend