Dan Carpenter <dan.carpenter@xxxxxxxxxx> writes:
> Hello Toke Høiland-Jørgensen,
>
> This is a semi-automatic email about new static checker warnings.
>
> The patch bb42f2d13ffc: "mac80211: Move reorder-sensitive TX handlers
> to after TXQ dequeue" from Sep 22, 2016, leads to the following
> Smatch complaint:
>
> net/mac80211/tx.c:3242 ieee80211_xmit_fast_finish()
> error: we previously assumed 'key' could be null (see line 3209)
>
> net/mac80211/tx.c
> 3208
> 3209 if (key)
> ^^^
> Check.
>
> 3210 info->control.hw_key = &key->conf;
> 3211
> 3212 ieee80211_tx_stats(skb->dev, skb->len);
> 3213
> 3214 if (hdr->frame_control & cpu_to_le16(IEEE80211_STYPE_QOS_DATA)) {
> 3215 tid = skb->priority & IEEE80211_QOS_CTL_TAG1D_MASK;
> 3216 *ieee80211_get_qos_ctl(hdr) = tid;
> 3217 hdr->seq_ctrl = ieee80211_tx_next_seq(sta, tid);
> 3218 } else {
> 3219 info->flags |= IEEE80211_TX_CTL_ASSIGN_SEQ;
> 3220 hdr->seq_ctrl = cpu_to_le16(sdata->sequence_number);
> 3221 sdata->sequence_number += 0x10;
> 3222 }
> 3223
> 3224 if (skb_shinfo(skb)->gso_size)
> 3225 sta->tx_stats.msdu[tid] +=
> 3226 DIV_ROUND_UP(skb->len, skb_shinfo(skb)->gso_size);
> 3227 else
> 3228 sta->tx_stats.msdu[tid]++;
> 3229
> 3230 info->hw_queue = sdata->vif.hw_queue[skb_get_queue_mapping(skb)];
> 3231
> 3232 /* statistics normally done by ieee80211_tx_h_stats (but that
> 3233 * has to consider fragmentation, so is more complex)
> 3234 */
> 3235 sta->tx_stats.bytes[skb_get_queue_mapping(skb)] += skb->len;
> 3236 sta->tx_stats.packets[skb_get_queue_mapping(skb)]++;
> 3237
> 3238 if (pn_offs) {
> ^^^^^^^
> Maybe when pn_offs is non-zero that implies key is non-NULL?
Yes, it does. fast_tx->pn_offs is set in ieee80211_check_fast_xmit()
which only sets it if fast_tx->key is set. The other call to
ieee80211_xmit_fast_finish() is in ieee80211_tx_dequeue() which also
only sets pn_offs if the key is set.
-Toke
