Re: [RFC 4/5] wifi: mac80211: mesh fast xmit support

[Felix Fietkau <nbd@nbd.name>]

On 27.02.23 14:14, Johannes Berg wrote:
> On Mon, 2023-02-27 at 12:07 +0100, Felix Fietkau wrote:
>> 
>> +	/* If the skb is shared we need to obtain our own copy */
>> +	if (skb_shared(skb)) {
>> +		struct sk_buff *oskb = skb;
>> +
>> +		skb = skb_clone(skb, GFP_ATOMIC);
>> +		if (!skb)
>> +			return false;
>> +
>> +		kfree_skb(oskb);
>> +	}
> 
> Use skb_share_check()?
Will do.

>>  	next_hop = rcu_dereference(mpath->next_hop);
>>  	if (next_hop) {
>>  		memcpy(hdr->addr1, next_hop->sta.addr, ETH_ALEN);
>>  		memcpy(hdr->addr2, sdata->vif.addr, ETH_ALEN);
>>  		ieee80211_mps_set_frame_flags(sdata, next_hop, hdr);
>> +		if (ieee80211_hw_check(&sdata->local->hw, SUPPORT_FAST_XMIT))
>> +			mesh_fast_tx_cache(sdata, skb, mpath);
> 
> 
> I wondered briefly if it's worth moving that check into the function,
> but not sure.
I put it there to avoid an extra function call for the non-fast-xmit case.

>> +	/* rate limit, in case fast xmit can't be enabled */
>> +	if (mppath->fast_tx_check == jiffies)
>> +		return;
>> +
>> +	mppath->fast_tx_check = jiffies;
> 
> once every jiffies seems pretty frequent though?
It should never happen under normal conditions anyway. It's just a bit 
of extra safety to avoid doing this per packet under high load if a 
corner case hits.

>> +	spin_lock_bh(&cache->walk_lock);
> 
> you could just spin_lock() the inner lock, _bh already taken care of by
> the outer?
Will do.

>> @@ -3686,7 +3690,7 @@ static void __ieee80211_xmit_fast(struct ieee80211_sub_if_data *sdata,
>>  #endif
>>  
>>  	if (hdr->frame_control & cpu_to_le16(IEEE80211_STYPE_QOS_DATA)) {
>> -		tid = skb->priority & IEEE80211_QOS_CTL_TAG1D_MASK;
>> +		u8 tid = skb->priority & IEEE80211_QOS_CTL_TAG1D_MASK;
>>  		*ieee80211_get_qos_ctl(hdr) = tid;
> 
> That's ... interesting, why wss there an argument in the first place?
I think last time I changed the code, I intended to use the tid value, 
but forgot to drop this line that assigns it.

> but maybe add a blank line now :)
Sure.

Thanks,

- Felix