if_em_netmap.h revision 231594 
1/*
2 * Copyright (C) 2011 Matteo Landi, Luigi Rizzo. All rights reserved.
3 *
4 * Redistribution and use in source and binary forms, with or without
5 * modification, are permitted provided that the following conditions
6 * are met:
7 * 1. Redistributions of source code must retain the above copyright
8 *    notice, this list of conditions and the following disclaimer.
9 * 2. Redistributions in binary form must reproduce the above copyright
10 *    notice, this list of conditions and the following disclaimer in the
11 *    documentation and/or other materials provided with the distribution.
12 *
13 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
14 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
15 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
16 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
17 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
18 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
19 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
20 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
21 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
22 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
23 * SUCH DAMAGE.
24 */
25
26/*
27 * $FreeBSD: head/sys/dev/netmap/if_em_netmap.h 231594 2012-02-13 18:56:34Z luigi $
28 * $Id: if_em_netmap.h 9802 2011-12-02 18:42:37Z luigi $
29 *
30 * netmap changes for if_em.
31 *
32 * For structure and details on the individual functions please see
33 * ixgbe_netmap.h
34 */
35
36#include <net/netmap.h>
37#include <sys/selinfo.h>
38#include <vm/vm.h>
39#include <vm/pmap.h>    /* vtophys ? */
40#include <dev/netmap/netmap_kern.h>
41
42static void	em_netmap_block_tasks(struct adapter *);
43static void	em_netmap_unblock_tasks(struct adapter *);
44static int	em_netmap_reg(struct ifnet *, int onoff);
45static int	em_netmap_txsync(struct ifnet *, u_int, int);
46static int	em_netmap_rxsync(struct ifnet *, u_int, int);
47static void	em_netmap_lock_wrapper(struct ifnet *, int, u_int);
48
49static void
50em_netmap_attach(struct adapter *adapter)
51{
52	struct netmap_adapter na;
53
54	bzero(&na, sizeof(na));
55
56	na.ifp = adapter->ifp;
57	na.separate_locks = 1;
58	na.num_tx_desc = adapter->num_tx_desc;
59	na.num_rx_desc = adapter->num_rx_desc;
60	na.nm_txsync = em_netmap_txsync;
61	na.nm_rxsync = em_netmap_rxsync;
62	na.nm_lock = em_netmap_lock_wrapper;
63	na.nm_register = em_netmap_reg;
64	netmap_attach(&na, adapter->num_queues);
65}
66
67
68/*
69 * wrapper to export locks to the generic code
70 */
71static void
72em_netmap_lock_wrapper(struct ifnet *ifp, int what, u_int queueid)
73{
74	struct adapter *adapter = ifp->if_softc;
75
76	ASSERT(queueid < adapter->num_queues);
77	switch (what) {
78	case NETMAP_CORE_LOCK:
79		EM_CORE_LOCK(adapter);
80		break;
81	case NETMAP_CORE_UNLOCK:
82		EM_CORE_UNLOCK(adapter);
83		break;
84	case NETMAP_TX_LOCK:
85		EM_TX_LOCK(&adapter->tx_rings[queueid]);
86		break;
87	case NETMAP_TX_UNLOCK:
88		EM_TX_UNLOCK(&adapter->tx_rings[queueid]);
89		break;
90	case NETMAP_RX_LOCK:
91		EM_RX_LOCK(&adapter->rx_rings[queueid]);
92		break;
93	case NETMAP_RX_UNLOCK:
94		EM_RX_UNLOCK(&adapter->rx_rings[queueid]);
95		break;
96	}
97}
98
99
100// XXX do we need to block/unblock the tasks ?
101static void
102em_netmap_block_tasks(struct adapter *adapter)
103{
104	if (adapter->msix > 1) { /* MSIX */
105		int i;
106		struct tx_ring *txr = adapter->tx_rings;
107		struct rx_ring *rxr = adapter->rx_rings;
108
109		for (i = 0; i < adapter->num_queues; i++, txr++, rxr++) {
110			taskqueue_block(txr->tq);
111			taskqueue_drain(txr->tq, &txr->tx_task);
112			taskqueue_block(rxr->tq);
113			taskqueue_drain(rxr->tq, &rxr->rx_task);
114		}
115	} else {	/* legacy */
116		taskqueue_block(adapter->tq);
117		taskqueue_drain(adapter->tq, &adapter->link_task);
118		taskqueue_drain(adapter->tq, &adapter->que_task);
119	}
120}
121
122
123static void
124em_netmap_unblock_tasks(struct adapter *adapter)
125{
126	if (adapter->msix > 1) {
127		struct tx_ring *txr = adapter->tx_rings;
128		struct rx_ring *rxr = adapter->rx_rings;
129		int i;
130
131		for (i = 0; i < adapter->num_queues; i++) {
132			taskqueue_unblock(txr->tq);
133			taskqueue_unblock(rxr->tq);
134		}
135	} else { /* legacy */
136		taskqueue_unblock(adapter->tq);
137	}
138}
139
140/*
141 * register-unregister routine
142 */
143static int
144em_netmap_reg(struct ifnet *ifp, int onoff)
145{
146	struct adapter *adapter = ifp->if_softc;
147	struct netmap_adapter *na = NA(ifp);
148	int error = 0;
149
150	if (na == NULL)
151		return EINVAL;	/* no netmap support here */
152
153	em_disable_intr(adapter);
154
155	/* Tell the stack that the interface is no longer active */
156	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
157
158	em_netmap_block_tasks(adapter);
159
160	if (onoff) {
161		ifp->if_capenable |= IFCAP_NETMAP;
162
163		na->if_transmit = ifp->if_transmit;
164		ifp->if_transmit = netmap_start;
165
166		em_init_locked(adapter);
167		if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) == 0) {
168			error = ENOMEM;
169			goto fail;
170		}
171	} else {
172fail:
173		/* restore if_transmit */
174		ifp->if_transmit = na->if_transmit;
175		ifp->if_capenable &= ~IFCAP_NETMAP;
176		em_init_locked(adapter);	/* also enable intr */
177	}
178	em_netmap_unblock_tasks(adapter);
179	return (error);
180}
181
182/*
183 * Reconcile hardware and user view of the transmit ring.
184 */
185static int
186em_netmap_txsync(struct ifnet *ifp, u_int ring_nr, int do_lock)
187{
188	struct adapter *adapter = ifp->if_softc;
189	struct tx_ring *txr = &adapter->tx_rings[ring_nr];
190	struct netmap_adapter *na = NA(adapter->ifp);
191	struct netmap_kring *kring = &na->tx_rings[ring_nr];
192	struct netmap_ring *ring = kring->ring;
193	int j, k, l, n = 0, lim = kring->nkr_num_slots - 1;
194
195	/* generate an interrupt approximately every half ring */
196	int report_frequency = kring->nkr_num_slots >> 1;
197
198	k = ring->cur;
199	if (k > lim)
200		return netmap_ring_reinit(kring);
201
202	if (do_lock)
203		EM_TX_LOCK(txr);
204	bus_dmamap_sync(txr->txdma.dma_tag, txr->txdma.dma_map,
205			BUS_DMASYNC_POSTREAD);
206
207	/* check for new packets to send.
208	 * j indexes the netmap ring, l indexes the nic ring, and
209	 *	j = kring->nr_hwcur, l = E1000_TDT (not tracked),
210	 *	j == (l + kring->nkr_hwofs) % ring_size
211	 */
212	j = kring->nr_hwcur;
213	if (j != k) {	/* we have packets to send */
214		l = j - kring->nkr_hwofs;
215		if (l < 0)
216			l += lim + 1;
217		while (j != k) {
218			struct netmap_slot *slot = &ring->slot[j];
219			struct e1000_tx_desc *curr = &txr->tx_base[l];
220			struct em_buffer *txbuf = &txr->tx_buffers[l];
221			int flags = ((slot->flags & NS_REPORT) ||
222				j == 0 || j == report_frequency) ?
223					E1000_TXD_CMD_RS : 0;
224			uint64_t paddr;
225			void *addr = PNMB(slot, &paddr);
226			int len = slot->len;
227			if (addr == netmap_buffer_base || len > NETMAP_BUF_SIZE) {
228				if (do_lock)
229					EM_TX_UNLOCK(txr);
230				return netmap_ring_reinit(kring);
231			}
232
233			slot->flags &= ~NS_REPORT;
234			curr->upper.data = 0;
235			curr->lower.data =
236			    htole32(adapter->txd_cmd | len |
237				(E1000_TXD_CMD_EOP | flags) );
238			if (slot->flags & NS_BUF_CHANGED) {
239				curr->buffer_addr = htole64(paddr);
240				/* buffer has changed, reload map */
241				netmap_reload_map(txr->txtag, txbuf->map, addr);
242				slot->flags &= ~NS_BUF_CHANGED;
243			}
244
245			bus_dmamap_sync(txr->txtag, txbuf->map,
246				BUS_DMASYNC_PREWRITE);
247			j = (j == lim) ? 0 : j + 1;
248			l = (l == lim) ? 0 : l + 1;
249			n++;
250		}
251		kring->nr_hwcur = k;
252
253		/* decrease avail by number of sent packets */
254		kring->nr_hwavail -= n;
255
256		bus_dmamap_sync(txr->txdma.dma_tag, txr->txdma.dma_map,
257		    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
258
259		E1000_WRITE_REG(&adapter->hw, E1000_TDT(txr->me), l);
260	}
261
262	if (n == 0 || kring->nr_hwavail < 1) {
263		int delta;
264
265		/* record completed transmissions using THD. */
266		l = E1000_READ_REG(&adapter->hw, E1000_TDH(ring_nr));
267		if (l >= kring->nkr_num_slots) { /* XXX can happen */
268			D("TDH wrap %d", l);
269			l -= kring->nkr_num_slots;
270		}
271		delta = l - txr->next_to_clean;
272		if (delta) {
273			/* some completed, increment hwavail. */
274			if (delta < 0)
275				delta += kring->nkr_num_slots;
276			txr->next_to_clean = l;
277			kring->nr_hwavail += delta;
278		}
279	}
280	/* update avail to what the hardware knows */
281	ring->avail = kring->nr_hwavail;
282
283	if (do_lock)
284		EM_TX_UNLOCK(txr);
285	return 0;
286}
287
288/*
289 * Reconcile kernel and user view of the receive ring.
290 */
291static int
292em_netmap_rxsync(struct ifnet *ifp, u_int ring_nr, int do_lock)
293{
294	struct adapter *adapter = ifp->if_softc;
295	struct rx_ring *rxr = &adapter->rx_rings[ring_nr];
296	struct netmap_adapter *na = NA(adapter->ifp);
297	struct netmap_kring *kring = &na->rx_rings[ring_nr];
298	struct netmap_ring *ring = kring->ring;
299	int j, k, l, n, lim = kring->nkr_num_slots - 1;
300
301	k = ring->cur;
302	if (k > lim)
303		return netmap_ring_reinit(kring);
304
305	if (do_lock)
306		EM_RX_LOCK(rxr);
307	/* XXX check sync modes */
308	bus_dmamap_sync(rxr->rxdma.dma_tag, rxr->rxdma.dma_map,
309			BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
310
311	/* import newly received packets into the netmap ring.
312	 * j is an index in the netmap ring, l in the NIC ring, and
313	 *	j = (kring->nr_hwcur + kring->nr_hwavail) % ring_size
314	 *	l = rxr->next_to_check;
315	 * and
316	 *	j == (l + kring->nkr_hwofs) % ring_size
317	 */
318	l = rxr->next_to_check;
319	j = l + kring->nkr_hwofs;
320	/* here nkr_hwofs can be negative so must check for j < 0 */
321	if (j < 0)
322		j += lim + 1;
323	else if (j > lim)
324		j -= lim + 1;
325	for (n = 0; ; n++) {
326		struct e1000_rx_desc *curr = &rxr->rx_base[l];
327
328		if ((curr->status & E1000_RXD_STAT_DD) == 0)
329			break;
330		ring->slot[j].len = le16toh(curr->length);
331		bus_dmamap_sync(rxr->rxtag, rxr->rx_buffers[l].map,
332			BUS_DMASYNC_POSTREAD);
333		j = (j == lim) ? 0 : j + 1;
334		/* make sure next_to_refresh follows next_to_check */
335		rxr->next_to_refresh = l;	// XXX
336		l = (l == lim) ? 0 : l + 1;
337	}
338	if (n) {
339		rxr->next_to_check = l;
340		kring->nr_hwavail += n;
341	}
342
343	/* skip past packets that userspace has already processed */
344	j = kring->nr_hwcur;
345	if (j != k) { /* userspace has read some packets. */
346		n = 0;
347		l = j - kring->nkr_hwofs; /* NIC ring index */
348		/* here nkr_hwofs can be negative so check for l > lim */
349		if (l < 0)
350			l += lim + 1;
351		else if (l > lim)
352			l -= lim + 1;
353		while (j != k) {
354			struct netmap_slot *slot = &ring->slot[j];
355			struct e1000_rx_desc *curr = &rxr->rx_base[l];
356			struct em_buffer *rxbuf = &rxr->rx_buffers[l];
357			uint64_t paddr;
358			void *addr = PNMB(slot, &paddr);
359
360			if (addr == netmap_buffer_base) { /* bad buf */
361				if (do_lock)
362					EM_RX_UNLOCK(rxr);
363				return netmap_ring_reinit(kring);
364			}
365
366			curr->status = 0;
367			if (slot->flags & NS_BUF_CHANGED) {
368				curr->buffer_addr = htole64(paddr);
369				/* buffer has changed, reload map */
370				netmap_reload_map(rxr->rxtag, rxbuf->map, addr);
371				slot->flags &= ~NS_BUF_CHANGED;
372			}
373
374			bus_dmamap_sync(rxr->rxtag, rxbuf->map,
375			    BUS_DMASYNC_PREREAD);
376
377			j = (j == lim) ? 0 : j + 1;
378			l = (l == lim) ? 0 : l + 1;
379			n++;
380		}
381		kring->nr_hwavail -= n;
382		kring->nr_hwcur = k;
383		bus_dmamap_sync(rxr->rxdma.dma_tag, rxr->rxdma.dma_map,
384		    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
385		/*
386		 * IMPORTANT: we must leave one free slot in the ring,
387		 * so move l back by one unit
388		 */
389		l = (l == 0) ? lim : l - 1;
390		E1000_WRITE_REG(&adapter->hw, E1000_RDT(rxr->me), l);
391	}
392	/* tell userspace that there are new packets */
393	ring->avail = kring->nr_hwavail ;
394	if (do_lock)
395		EM_RX_UNLOCK(rxr);
396	return 0;
397}
398