xref: /freebsd/sys/dev/virtio/network/if_vtnetvar.h (revision 12e0d316644a4f80f5f1f78cf07bd93def43b1ca)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2011, Bryan Venteicher <bryanv@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice unmodified, this list of conditions, and the following
12  *    disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 #ifndef _IF_VTNETVAR_H
30 #define _IF_VTNETVAR_H
31 
32 #ifdef ALTQ
33 #define	VTNET_LEGACY_TX
34 #endif
35 
36 struct vtnet_softc;
37 
38 struct vtnet_statistics {
39 	uint64_t	mbuf_alloc_failed;
40 
41 	uint64_t	rx_frame_too_large;
42 	uint64_t	rx_enq_replacement_failed;
43 	uint64_t	rx_mergeable_failed;
44 	uint64_t	rx_csum_bad_ethtype;
45 	uint64_t	rx_csum_bad_ipproto;
46 	uint64_t	rx_csum_bad_offset;
47 	uint64_t	rx_csum_bad_proto;
48 	uint64_t	tx_csum_unknown_ethtype;
49 	uint64_t	tx_csum_proto_mismatch;
50 	uint64_t	tx_tso_not_tcp;
51 	uint64_t	tx_tso_without_csum;
52 	uint64_t	tx_defragged;
53 	uint64_t	tx_defrag_failed;
54 
55 	/*
56 	 * These are accumulated from each Rx/Tx queue.
57 	 */
58 	uint64_t	rx_csum_failed;
59 	uint64_t	rx_csum_offloaded;
60 	uint64_t	rx_task_rescheduled;
61 	uint64_t	tx_csum_offloaded;
62 	uint64_t	tx_tso_offloaded;
63 	uint64_t	tx_task_rescheduled;
64 };
65 
66 struct vtnet_rxq_stats {
67 	uint64_t	vrxs_ipackets;	/* if_ipackets */
68 	uint64_t	vrxs_ibytes;	/* if_ibytes */
69 	uint64_t	vrxs_iqdrops;	/* if_iqdrops */
70 	uint64_t	vrxs_ierrors;	/* if_ierrors */
71 	uint64_t	vrxs_csum;
72 	uint64_t	vrxs_csum_failed;
73 	uint64_t	vrxs_host_lro;
74 	uint64_t	vrxs_rescheduled;
75 };
76 
77 struct vtnet_rxq {
78 	struct mtx		 vtnrx_mtx;
79 	struct vtnet_softc	*vtnrx_sc;
80 	struct virtqueue	*vtnrx_vq;
81 	struct sglist		*vtnrx_sg;
82 	int			 vtnrx_id;
83 	struct vtnet_rxq_stats	 vtnrx_stats;
84 	struct taskqueue	*vtnrx_tq;
85 	struct task		 vtnrx_intrtask;
86 	struct lro_ctrl		 vtnrx_lro;
87 #ifdef DEV_NETMAP
88 	uint32_t		 vtnrx_nm_refill;
89 	struct virtio_net_hdr_mrg_rxbuf vtnrx_shrhdr;
90 #endif  /* DEV_NETMAP */
91 	char			 vtnrx_name[16];
92 } __aligned(CACHE_LINE_SIZE);
93 
94 #define VTNET_RXQ_LOCK(_rxq)	mtx_lock(&(_rxq)->vtnrx_mtx)
95 #define VTNET_RXQ_UNLOCK(_rxq)	mtx_unlock(&(_rxq)->vtnrx_mtx)
96 #define VTNET_RXQ_LOCK_ASSERT(_rxq)		\
97     mtx_assert(&(_rxq)->vtnrx_mtx, MA_OWNED)
98 #define VTNET_RXQ_LOCK_ASSERT_NOTOWNED(_rxq)	\
99     mtx_assert(&(_rxq)->vtnrx_mtx, MA_NOTOWNED)
100 
101 struct vtnet_txq_stats {
102 	uint64_t vtxs_opackets;	/* if_opackets */
103 	uint64_t vtxs_obytes;	/* if_obytes */
104 	uint64_t vtxs_omcasts;	/* if_omcasts */
105 	uint64_t vtxs_csum;
106 	uint64_t vtxs_tso;
107 	uint64_t vtxs_rescheduled;
108 };
109 
110 struct vtnet_txq {
111 	struct mtx		 vtntx_mtx;
112 	struct vtnet_softc	*vtntx_sc;
113 	struct virtqueue	*vtntx_vq;
114 	struct sglist		*vtntx_sg;
115 #ifndef VTNET_LEGACY_TX
116 	struct buf_ring		*vtntx_br;
117 #endif
118 	int			 vtntx_id;
119 	int			 vtntx_watchdog;
120 	int			 vtntx_intr_threshold;
121 	struct vtnet_txq_stats	 vtntx_stats;
122 	struct taskqueue	*vtntx_tq;
123 	struct task		 vtntx_intrtask;
124 #ifndef VTNET_LEGACY_TX
125 	struct task		 vtntx_defrtask;
126 #endif
127 #ifdef DEV_NETMAP
128 	struct virtio_net_hdr_mrg_rxbuf vtntx_shrhdr;
129 #endif  /* DEV_NETMAP */
130 	char			 vtntx_name[16];
131 } __aligned(CACHE_LINE_SIZE);
132 
133 #define VTNET_TXQ_LOCK(_txq)	mtx_lock(&(_txq)->vtntx_mtx)
134 #define VTNET_TXQ_TRYLOCK(_txq)	mtx_trylock(&(_txq)->vtntx_mtx)
135 #define VTNET_TXQ_UNLOCK(_txq)	mtx_unlock(&(_txq)->vtntx_mtx)
136 #define VTNET_TXQ_LOCK_ASSERT(_txq)		\
137     mtx_assert(&(_txq)->vtntx_mtx, MA_OWNED)
138 #define VTNET_TXQ_LOCK_ASSERT_NOTOWNED(_txq)	\
139     mtx_assert(&(_txq)->vtntx_mtx, MA_NOTOWNED)
140 
141 struct vtnet_softc {
142 	device_t		 vtnet_dev;
143 	if_t			 vtnet_ifp;
144 	struct vtnet_rxq	*vtnet_rxqs;
145 	struct vtnet_txq	*vtnet_txqs;
146 	pfil_head_t		 vtnet_pfil;
147 	uint64_t		 vtnet_features;
148 
149 	uint32_t		 vtnet_flags;
150 #define VTNET_FLAG_MODERN	 0x0001
151 #define VTNET_FLAG_MAC		 0x0002
152 #define VTNET_FLAG_CTRL_VQ	 0x0004
153 #define VTNET_FLAG_CTRL_RX	 0x0008
154 #define VTNET_FLAG_CTRL_MAC	 0x0010
155 #define VTNET_FLAG_VLAN_FILTER	 0x0020
156 #define VTNET_FLAG_TSO_ECN	 0x0040
157 #define VTNET_FLAG_MRG_RXBUFS	 0x0080
158 #define VTNET_FLAG_LRO_NOMRG	 0x0100
159 #define VTNET_FLAG_MQ		 0x0200
160 #define VTNET_FLAG_INDIRECT	 0x0400
161 #define VTNET_FLAG_EVENT_IDX	 0x0800
162 #define VTNET_FLAG_SUSPENDED	 0x1000
163 #define VTNET_FLAG_FIXUP_NEEDS_CSUM 0x2000
164 #define VTNET_FLAG_SW_LRO	 0x4000
165 
166 	u_int			 vtnet_hdr_size;
167 	int			 vtnet_rx_nmbufs;
168 	int			 vtnet_rx_clustersz;
169 	int			 vtnet_rx_nsegs;
170 	int			 vtnet_rx_process_limit;
171 	int			 vtnet_link_active;
172 	int			 vtnet_act_vq_pairs;
173 	int			 vtnet_req_vq_pairs;
174 	int			 vtnet_max_vq_pairs;
175 	int			 vtnet_tx_nsegs;
176 	int			 vtnet_if_flags;
177 	u_int			 vtnet_max_mtu;
178 	int			 vtnet_lro_entry_count;
179 	int			 vtnet_lro_mbufq_depth;
180 
181 	struct virtqueue	*vtnet_ctrl_vq;
182 	struct vtnet_mac_filter	*vtnet_mac_filter;
183 	uint32_t		*vtnet_vlan_filter;
184 
185 	uint64_t		 vtnet_negotiated_features;
186 	struct vtnet_statistics	 vtnet_stats;
187 	struct callout		 vtnet_tick_ch;
188 	struct ifmedia		 vtnet_media;
189 	eventhandler_tag	 vtnet_vlan_attach;
190 	eventhandler_tag	 vtnet_vlan_detach;
191 
192 	struct mtx		 vtnet_mtx;
193 	char			 vtnet_mtx_name[16];
194 	uint8_t			 vtnet_hwaddr[ETHER_ADDR_LEN];
195 };
196 
197 static bool
198 vtnet_modern(struct vtnet_softc *sc)
199 {
200 	return ((sc->vtnet_flags & VTNET_FLAG_MODERN) != 0);
201 }
202 
203 static bool
204 vtnet_software_lro(struct vtnet_softc *sc)
205 {
206 	return ((sc->vtnet_flags & VTNET_FLAG_SW_LRO) != 0);
207 }
208 
209 /*
210  * Maximum number of queue pairs we will autoconfigure to.
211  */
212 #define VTNET_MAX_QUEUE_PAIRS	32
213 
214 /*
215  * Additional completed entries can appear in a virtqueue before we can
216  * reenable interrupts. Number of times to retry before scheduling the
217  * taskqueue to process the completed entries.
218  */
219 #define VTNET_INTR_DISABLE_RETRIES	4
220 
221 /*
222  * Similarly, additional completed entries can appear in a virtqueue
223  * between when lasted checked and before notifying the host. Number
224  * of times to retry before scheduling the taskqueue to process the
225  * queue.
226  */
227 #define VTNET_NOTIFY_RETRIES		4
228 
229 /*
230  * Number of words to allocate for the VLAN shadow table. There is one
231  * bit for each VLAN.
232  */
233 #define VTNET_VLAN_FILTER_NWORDS	(4096 / 32)
234 
235 /*
236  * We depend on all of the hdr structures being even, and matching the standard
237  * length. As well, we depend on two being identally sized (with the same
238  * layout).
239  */
240 CTASSERT(sizeof(struct virtio_net_hdr_v1) == 12);
241 CTASSERT(sizeof(struct virtio_net_hdr) == 10);
242 CTASSERT(sizeof(struct virtio_net_hdr_mrg_rxbuf) ==
243     sizeof(struct virtio_net_hdr_v1));
244 
245 /*
246  * In legacy VirtIO when mergeable buffers are not negotiated, this structure
247  * is placed at the beginning of the mbuf data. Use 4 bytes of pad to keep
248  * both the VirtIO header and the data non-contiguous and the frame's payload
249  * 4 byte aligned. Note this padding would not be necessary if the
250  * VIRTIO_F_ANY_LAYOUT feature was negotiated (but we don't support that yet).
251  *
252  * In modern VirtIO or when mergeable buffers are negotiated, the host puts
253  * the VirtIO header in the beginning of the first mbuf's data.
254  */
255 #define VTNET_RX_HEADER_PAD	4
256 struct vtnet_rx_header {
257 	struct virtio_net_hdr	vrh_hdr;
258 	char			vrh_pad[VTNET_RX_HEADER_PAD];
259 } __packed;
260 
261 /*
262  * For each outgoing frame, the vtnet_tx_header below is allocated from
263  * the vtnet_tx_header_zone.
264  */
265 struct vtnet_tx_header {
266 	union {
267 		struct virtio_net_hdr		hdr;
268 		struct virtio_net_hdr_mrg_rxbuf	mhdr;
269 		struct virtio_net_hdr_v1	v1hdr;
270 	} vth_uhdr;
271 
272 	struct mbuf *vth_mbuf;
273 };
274 
275 /*
276  * The VirtIO specification does not place a limit on the number of MAC
277  * addresses the guest driver may request to be filtered. In practice,
278  * the host is constrained by available resources. To simplify this driver,
279  * impose a reasonably high limit of MAC addresses we will filter before
280  * falling back to promiscuous or all-multicast modes.
281  */
282 #define VTNET_MAX_MAC_ENTRIES	128
283 
284 /*
285  * The driver version of struct virtio_net_ctrl_mac but with our predefined
286  * number of MAC addresses allocated. This structure is shared with the host,
287  * so nentries field is in the correct VirtIO endianness.
288  */
289 struct vtnet_mac_table {
290 	uint32_t	nentries;
291 	uint8_t		macs[VTNET_MAX_MAC_ENTRIES][ETHER_ADDR_LEN];
292 } __packed;
293 
294 struct vtnet_mac_filter {
295 	struct vtnet_mac_table	vmf_unicast;
296 	uint32_t		vmf_pad; /* Make tables non-contiguous. */
297 	struct vtnet_mac_table	vmf_multicast;
298 };
299 
300 /*
301  * The MAC filter table is malloc(9)'d when needed. Ensure it will
302  * always fit in one segment.
303  */
304 CTASSERT(sizeof(struct vtnet_mac_filter) <= PAGE_SIZE);
305 
306 #define VTNET_TX_TIMEOUT	5
307 #define VTNET_CSUM_OFFLOAD	(CSUM_TCP | CSUM_UDP)
308 #define VTNET_CSUM_OFFLOAD_IPV6	(CSUM_TCP_IPV6 | CSUM_UDP_IPV6)
309 
310 #define VTNET_CSUM_ALL_OFFLOAD	\
311     (VTNET_CSUM_OFFLOAD | VTNET_CSUM_OFFLOAD_IPV6 | CSUM_TSO)
312 
313 #define VTNET_COMMON_FEATURES \
314     (VIRTIO_NET_F_MAC			| \
315      VIRTIO_NET_F_STATUS		| \
316      VIRTIO_NET_F_CTRL_GUEST_OFFLOADS	| \
317      VIRTIO_NET_F_MTU			| \
318      VIRTIO_NET_F_CTRL_VQ		| \
319      VIRTIO_NET_F_CTRL_RX		| \
320      VIRTIO_NET_F_CTRL_MAC_ADDR		| \
321      VIRTIO_NET_F_CTRL_VLAN		| \
322      VIRTIO_NET_F_CSUM			| \
323      VIRTIO_NET_F_HOST_TSO4		| \
324      VIRTIO_NET_F_HOST_TSO6		| \
325      VIRTIO_NET_F_HOST_ECN		| \
326      VIRTIO_NET_F_GUEST_CSUM		| \
327      VIRTIO_NET_F_GUEST_TSO4		| \
328      VIRTIO_NET_F_GUEST_TSO6		| \
329      VIRTIO_NET_F_GUEST_ECN		| \
330      VIRTIO_NET_F_MRG_RXBUF		| \
331      VIRTIO_NET_F_MQ			| \
332      VIRTIO_NET_F_SPEED_DUPLEX		| \
333      VIRTIO_RING_F_EVENT_IDX		| \
334      VIRTIO_RING_F_INDIRECT_DESC)
335 
336 #define VTNET_MODERN_FEATURES (VTNET_COMMON_FEATURES)
337 #define VTNET_LEGACY_FEATURES (VTNET_COMMON_FEATURES | VIRTIO_NET_F_GSO)
338 
339 /*
340  * The VIRTIO_NET_F_HOST_TSO[46] features permit us to send the host
341  * frames larger than 1514 bytes.
342  */
343 #define VTNET_TSO_FEATURES (VIRTIO_NET_F_GSO | VIRTIO_NET_F_HOST_TSO4 | \
344     VIRTIO_NET_F_HOST_TSO6 | VIRTIO_NET_F_HOST_ECN)
345 
346 /*
347  * The VIRTIO_NET_F_GUEST_TSO[46] features permit the host to send us
348  * frames larger than 1514 bytes.
349  */
350 #define VTNET_LRO_FEATURES (VIRTIO_NET_F_GUEST_TSO4 | \
351     VIRTIO_NET_F_GUEST_TSO6 | VIRTIO_NET_F_GUEST_ECN)
352 
353 #define VTNET_MIN_MTU		68
354 #define VTNET_MAX_MTU		65536
355 #define VTNET_MAX_RX_SIZE	65550
356 
357 /*
358  * Used to preallocate the VQ indirect descriptors. Modern and mergeable
359  * buffers do not required one segment for the VirtIO header since it is
360  * placed inline at the beginning of the receive buffer.
361  */
362 #define VTNET_RX_SEGS_HDR_INLINE	1
363 #define VTNET_RX_SEGS_HDR_SEPARATE	2
364 #define VTNET_RX_SEGS_LRO_NOMRG		34
365 #define VTNET_TX_SEGS_MIN		32
366 #define VTNET_TX_SEGS_MAX		64
367 
368 CTASSERT(((VTNET_RX_SEGS_LRO_NOMRG - 1) * MCLBYTES) >= VTNET_MAX_RX_SIZE);
369 CTASSERT(((VTNET_TX_SEGS_MAX - 1) * MCLBYTES) >= VTNET_MAX_MTU);
370 
371 /*
372  * Number of slots in the Tx bufrings. This value matches most other
373  * multiqueue drivers.
374  */
375 #define VTNET_DEFAULT_BUFRING_SIZE	4096
376 
377 #define VTNET_CORE_MTX(_sc)		&(_sc)->vtnet_mtx
378 #define VTNET_CORE_LOCK(_sc)		mtx_lock(VTNET_CORE_MTX((_sc)))
379 #define VTNET_CORE_UNLOCK(_sc)		mtx_unlock(VTNET_CORE_MTX((_sc)))
380 #define VTNET_CORE_LOCK_DESTROY(_sc)	mtx_destroy(VTNET_CORE_MTX((_sc)))
381 #define VTNET_CORE_LOCK_ASSERT(_sc)		\
382     mtx_assert(VTNET_CORE_MTX((_sc)), MA_OWNED)
383 #define VTNET_CORE_LOCK_ASSERT_NOTOWNED(_sc)	\
384     mtx_assert(VTNET_CORE_MTX((_sc)), MA_NOTOWNED)
385 
386 #define VTNET_CORE_LOCK_INIT(_sc) do {					\
387     snprintf((_sc)->vtnet_mtx_name, sizeof((_sc)->vtnet_mtx_name),	\
388         "%s", device_get_nameunit((_sc)->vtnet_dev));			\
389     mtx_init(VTNET_CORE_MTX((_sc)), (_sc)->vtnet_mtx_name,		\
390         "VTNET Core Lock", MTX_DEF);					\
391 } while (0)
392 
393 /*
394  * Values for the init_mode argument of vtnet_init_locked().
395  */
396 #define VTNET_INIT_NETMAP_ENTER		1
397 #define VTNET_INIT_NETMAP_EXIT		2
398 
399 #endif /* _IF_VTNETVAR_H */
400