LinuxKPI: 802.11: disable ic_headroom for the moment

There is a problem with some drivers, such as rtw88, asking for more
headroom than we currently can handle throughout the stack (we have
other legacy wireless driver in the tree with similar problems).
This may trigger an assertion in the TCP syncache where we are checking
for a reply to fit in MHLEN.
While for the moment we still copy data from mbufs to skbs,
we can simply disable the extra headroom request in ic_headroom and
deal with it ourselves (which we already did anyway).
Leave a link to the thread on freebsd-transport detailing more of the
problem so we can find it again and solve it here or there.

MFC after:	3 days

sys/compat/linuxkpi/common/src/linux_80211.c

@@ -2423,6 +2423,8 @@ lkpi_80211_txq_tx_one(struct lkpi_sta *lsta, struct mbuf *m)
 	/*
 	 * net80211 should handle hw->extra_tx_headroom.
 	 * Though for as long as we are copying we don't mind.
+	 * XXX-BZ rtw88 asks for too much headroom for ipv6+tcp:
+	 * https://lists.freebsd.org/archives/freebsd-transport/2022-February/000012.html
 	 */
 	skb = dev_alloc_skb(hw->extra_tx_headroom + m->m_pkthdr.len);
 	if (skb == NULL) {
@@ -2852,7 +2854,10 @@ linuxkpi_ieee80211_ifattach(struct ieee80211_hw *hw)
 		ic_printf(ic, "%s: warning, no hardware address!\n", __func__);
 	}
 
+#ifdef __not_yet__
+	/* See comment in lkpi_80211_txq_tx_one(). */
 	ic->ic_headroom = hw->extra_tx_headroom;
+#endif
 
 	ic->ic_phytype = IEEE80211_T_OFDM;	/* not only, but not used */
 	ic->ic_opmode = IEEE80211_M_STA;