Re: [RFC PATCH v3 05/12] mac80211: Mark A-MPDU keyid borders for drivers

[Johannes Berg <johannes@sipsolutions.net>]

> The intent of the wording was probably written without considering
> Extended Key IDs. At least it makes no sense for me to forbid mixing
> MPDUs using keyid 0 and 1 in one A-MPDU. 

I think it may make sense - reprogramming the hardware engines may take
some time, and doing that in the middle of the A-MPDU may not be
feasible? You don't just have to load the key (that you need to do
anyway) but also extract the status? I dunno, I'm more handwaving, but
it doesn't make sense to add such a requirement when only one key index
can be used to start with.

> The code is assuming that the driver is not aggregating MPDUs more than
> 5s apart. We probably don't have wait nearly so long but I'm not sure
> what is the minimum time.

OTOH, if you have a lot of BE/VI/VO traffic BK might be starved even
longer than that, technically indefinitely.

> +static struct ieee80211_key debug_noinline
> +*ieee80211_select_sta_key(struct ieee80211_tx_data *tx)
> +{
> +	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)tx->skb->data;
> +	struct ieee80211_tx_info *info = IEEE80211_SKB_CB(tx->skb);
> +	struct sta_info *sta = tx->sta;
> +	struct ieee80211_key *key;
> +	struct ieee80211_key *next_key;
> +
> +	key = rcu_dereference(tx->sta->ptk[tx->sta->ptk_idx]);
> +
> +	if (likely(sta->ptk_idx_next == INVALID_PTK_KEYIDX))
> +		return key;
> +
> +	/* Only when using Extended Key ID the code below can be executed */
> +
> +	if (!ieee80211_is_data_present(hdr->frame_control))
> +		return key;
> +
> +	if (sta->ptk_idx_next == sta->ptk_idx) {
> +		/* First packet using new key with A-MPDU active*/
> +		sta->ptk_idx_next = INVALID_PTK_KEYIDX;
> +		ieee80211_check_fast_xmit(tx->sta);

I'm not convinced you can call this from this context? It looks safe
though, but it's really strange in a way.

> +	info->flags &= ~IEEE80211_TX_CTL_AMPDU;

Like you say above, I don't think this really makes a lot of sense. If
we don't have any free bits I guess we should try to find some ...

johannes