xref: /freebsd/sys/dev/virtio/network/if_vtnetvar.h (revision 9768746b)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
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  * $FreeBSD$
29  */
30 
31 #ifndef _IF_VTNETVAR_H
32 #define _IF_VTNETVAR_H
33 
34 #ifdef ALTQ
35 #define	VTNET_LEGACY_TX
36 #endif
37 
38 struct vtnet_softc;
39 
40 struct vtnet_statistics {
41 	uint64_t	mbuf_alloc_failed;
42 
43 	uint64_t	rx_frame_too_large;
44 	uint64_t	rx_enq_replacement_failed;
45 	uint64_t	rx_mergeable_failed;
46 	uint64_t	rx_csum_bad_ethtype;
47 	uint64_t	rx_csum_bad_ipproto;
48 	uint64_t	rx_csum_bad_offset;
49 	uint64_t	rx_csum_bad_proto;
50 	uint64_t	tx_csum_unknown_ethtype;
51 	uint64_t	tx_csum_proto_mismatch;
52 	uint64_t	tx_tso_not_tcp;
53 	uint64_t	tx_tso_without_csum;
54 	uint64_t	tx_defragged;
55 	uint64_t	tx_defrag_failed;
56 
57 	/*
58 	 * These are accumulated from each Rx/Tx queue.
59 	 */
60 	uint64_t	rx_csum_failed;
61 	uint64_t	rx_csum_offloaded;
62 	uint64_t	rx_task_rescheduled;
63 	uint64_t	tx_csum_offloaded;
64 	uint64_t	tx_tso_offloaded;
65 	uint64_t	tx_task_rescheduled;
66 };
67 
68 struct vtnet_rxq_stats {
69 	uint64_t	vrxs_ipackets;	/* if_ipackets */
70 	uint64_t	vrxs_ibytes;	/* if_ibytes */
71 	uint64_t	vrxs_iqdrops;	/* if_iqdrops */
72 	uint64_t	vrxs_ierrors;	/* if_ierrors */
73 	uint64_t	vrxs_csum;
74 	uint64_t	vrxs_csum_failed;
75 	uint64_t	vrxs_host_lro;
76 	uint64_t	vrxs_rescheduled;
77 };
78 
79 struct vtnet_rxq {
80 	struct mtx		 vtnrx_mtx;
81 	struct vtnet_softc	*vtnrx_sc;
82 	struct virtqueue	*vtnrx_vq;
83 	struct sglist		*vtnrx_sg;
84 	int			 vtnrx_id;
85 	struct vtnet_rxq_stats	 vtnrx_stats;
86 	struct taskqueue	*vtnrx_tq;
87 	struct task		 vtnrx_intrtask;
88 	struct lro_ctrl		 vtnrx_lro;
89 #ifdef DEV_NETMAP
90 	uint32_t		 vtnrx_nm_refill;
91 	struct virtio_net_hdr_mrg_rxbuf vtnrx_shrhdr;
92 #endif  /* DEV_NETMAP */
93 	char			 vtnrx_name[16];
94 } __aligned(CACHE_LINE_SIZE);
95 
96 #define VTNET_RXQ_LOCK(_rxq)	mtx_lock(&(_rxq)->vtnrx_mtx)
97 #define VTNET_RXQ_UNLOCK(_rxq)	mtx_unlock(&(_rxq)->vtnrx_mtx)
98 #define VTNET_RXQ_LOCK_ASSERT(_rxq)		\
99     mtx_assert(&(_rxq)->vtnrx_mtx, MA_OWNED)
100 #define VTNET_RXQ_LOCK_ASSERT_NOTOWNED(_rxq)	\
101     mtx_assert(&(_rxq)->vtnrx_mtx, MA_NOTOWNED)
102 
103 struct vtnet_txq_stats {
104 	uint64_t vtxs_opackets;	/* if_opackets */
105 	uint64_t vtxs_obytes;	/* if_obytes */
106 	uint64_t vtxs_omcasts;	/* if_omcasts */
107 	uint64_t vtxs_csum;
108 	uint64_t vtxs_tso;
109 	uint64_t vtxs_rescheduled;
110 };
111 
112 struct vtnet_txq {
113 	struct mtx		 vtntx_mtx;
114 	struct vtnet_softc	*vtntx_sc;
115 	struct virtqueue	*vtntx_vq;
116 	struct sglist		*vtntx_sg;
117 #ifndef VTNET_LEGACY_TX
118 	struct buf_ring		*vtntx_br;
119 #endif
120 	int			 vtntx_id;
121 	int			 vtntx_watchdog;
122 	int			 vtntx_intr_threshold;
123 	struct vtnet_txq_stats	 vtntx_stats;
124 	struct taskqueue	*vtntx_tq;
125 	struct task		 vtntx_intrtask;
126 #ifndef VTNET_LEGACY_TX
127 	struct task		 vtntx_defrtask;
128 #endif
129 #ifdef DEV_NETMAP
130 	struct virtio_net_hdr_mrg_rxbuf vtntx_shrhdr;
131 #endif  /* DEV_NETMAP */
132 	char			 vtntx_name[16];
133 } __aligned(CACHE_LINE_SIZE);
134 
135 #define VTNET_TXQ_LOCK(_txq)	mtx_lock(&(_txq)->vtntx_mtx)
136 #define VTNET_TXQ_TRYLOCK(_txq)	mtx_trylock(&(_txq)->vtntx_mtx)
137 #define VTNET_TXQ_UNLOCK(_txq)	mtx_unlock(&(_txq)->vtntx_mtx)
138 #define VTNET_TXQ_LOCK_ASSERT(_txq)		\
139     mtx_assert(&(_txq)->vtntx_mtx, MA_OWNED)
140 #define VTNET_TXQ_LOCK_ASSERT_NOTOWNED(_txq)	\
141     mtx_assert(&(_txq)->vtntx_mtx, MA_NOTOWNED)
142 
143 struct vtnet_softc {
144 	device_t		 vtnet_dev;
145 	if_t			 vtnet_ifp;
146 	struct vtnet_rxq	*vtnet_rxqs;
147 	struct vtnet_txq	*vtnet_txqs;
148 	pfil_head_t		 vtnet_pfil;
149 	uint64_t		 vtnet_features;
150 
151 	uint32_t		 vtnet_flags;
152 #define VTNET_FLAG_MODERN	 0x0001
153 #define VTNET_FLAG_MAC		 0x0002
154 #define VTNET_FLAG_CTRL_VQ	 0x0004
155 #define VTNET_FLAG_CTRL_RX	 0x0008
156 #define VTNET_FLAG_CTRL_MAC	 0x0010
157 #define VTNET_FLAG_VLAN_FILTER	 0x0020
158 #define VTNET_FLAG_TSO_ECN	 0x0040
159 #define VTNET_FLAG_MRG_RXBUFS	 0x0080
160 #define VTNET_FLAG_LRO_NOMRG	 0x0100
161 #define VTNET_FLAG_MQ		 0x0200
162 #define VTNET_FLAG_INDIRECT	 0x0400
163 #define VTNET_FLAG_EVENT_IDX	 0x0800
164 #define VTNET_FLAG_SUSPENDED	 0x1000
165 #define VTNET_FLAG_FIXUP_NEEDS_CSUM 0x2000
166 #define VTNET_FLAG_SW_LRO	 0x4000
167 
168 	u_int			 vtnet_hdr_size;
169 	int			 vtnet_rx_nmbufs;
170 	int			 vtnet_rx_clustersz;
171 	int			 vtnet_rx_nsegs;
172 	int			 vtnet_rx_process_limit;
173 	int			 vtnet_link_active;
174 	int			 vtnet_act_vq_pairs;
175 	int			 vtnet_req_vq_pairs;
176 	int			 vtnet_max_vq_pairs;
177 	int			 vtnet_tx_nsegs;
178 	int			 vtnet_if_flags;
179 	u_int			 vtnet_max_mtu;
180 	int			 vtnet_lro_entry_count;
181 	int			 vtnet_lro_mbufq_depth;
182 
183 	struct virtqueue	*vtnet_ctrl_vq;
184 	struct vtnet_mac_filter	*vtnet_mac_filter;
185 	uint32_t		*vtnet_vlan_filter;
186 
187 	uint64_t		 vtnet_negotiated_features;
188 	struct vtnet_statistics	 vtnet_stats;
189 	struct callout		 vtnet_tick_ch;
190 	struct ifmedia		 vtnet_media;
191 	eventhandler_tag	 vtnet_vlan_attach;
192 	eventhandler_tag	 vtnet_vlan_detach;
193 
194 	struct mtx		 vtnet_mtx;
195 	char			 vtnet_mtx_name[16];
196 	uint8_t			 vtnet_hwaddr[ETHER_ADDR_LEN];
197 };
198 
199 static bool
200 vtnet_modern(struct vtnet_softc *sc)
201 {
202 	return ((sc->vtnet_flags & VTNET_FLAG_MODERN) != 0);
203 }
204 
205 static bool
206 vtnet_software_lro(struct vtnet_softc *sc)
207 {
208 	return ((sc->vtnet_flags & VTNET_FLAG_SW_LRO) != 0);
209 }
210 
211 /*
212  * Maximum number of queue pairs we will autoconfigure to.
213  */
214 #define VTNET_MAX_QUEUE_PAIRS	32
215 
216 /*
217  * Additional completed entries can appear in a virtqueue before we can
218  * reenable interrupts. Number of times to retry before scheduling the
219  * taskqueue to process the completed entries.
220  */
221 #define VTNET_INTR_DISABLE_RETRIES	4
222 
223 /*
224  * Similarly, additional completed entries can appear in a virtqueue
225  * between when lasted checked and before notifying the host. Number
226  * of times to retry before scheduling the taskqueue to process the
227  * queue.
228  */
229 #define VTNET_NOTIFY_RETRIES		4
230 
231 /*
232  * Number of words to allocate for the VLAN shadow table. There is one
233  * bit for each VLAN.
234  */
235 #define VTNET_VLAN_FILTER_NWORDS	(4096 / 32)
236 
237 /* We depend on these being the same size (and same layout). */
238 CTASSERT(sizeof(struct virtio_net_hdr_mrg_rxbuf) ==
239     sizeof(struct virtio_net_hdr_v1));
240 
241 /*
242  * In legacy VirtIO when mergeable buffers are not negotiated, this structure
243  * is placed at the beginning of the mbuf data. Use 4 bytes of pad to keep
244  * both the VirtIO header and the data non-contiguous and the frame's payload
245  * 4 byte aligned. Note this padding would not be necessary if the
246  * VIRTIO_F_ANY_LAYOUT feature was negotiated (but we don't support that yet).
247  *
248  * In modern VirtIO or when mergeable buffers are negotiated, the host puts
249  * the VirtIO header in the beginning of the first mbuf's data.
250  */
251 #define VTNET_RX_HEADER_PAD	4
252 struct vtnet_rx_header {
253 	struct virtio_net_hdr	vrh_hdr;
254 	char			vrh_pad[VTNET_RX_HEADER_PAD];
255 } __packed;
256 
257 /*
258  * For each outgoing frame, the vtnet_tx_header below is allocated from
259  * the vtnet_tx_header_zone.
260  */
261 struct vtnet_tx_header {
262 	union {
263 		struct virtio_net_hdr		hdr;
264 		struct virtio_net_hdr_mrg_rxbuf	mhdr;
265 		struct virtio_net_hdr_v1	v1hdr;
266 	} vth_uhdr;
267 
268 	struct mbuf *vth_mbuf;
269 };
270 
271 /*
272  * The VirtIO specification does not place a limit on the number of MAC
273  * addresses the guest driver may request to be filtered. In practice,
274  * the host is constrained by available resources. To simplify this driver,
275  * impose a reasonably high limit of MAC addresses we will filter before
276  * falling back to promiscuous or all-multicast modes.
277  */
278 #define VTNET_MAX_MAC_ENTRIES	128
279 
280 /*
281  * The driver version of struct virtio_net_ctrl_mac but with our predefined
282  * number of MAC addresses allocated. This structure is shared with the host,
283  * so nentries field is in the correct VirtIO endianness.
284  */
285 struct vtnet_mac_table {
286 	uint32_t	nentries;
287 	uint8_t		macs[VTNET_MAX_MAC_ENTRIES][ETHER_ADDR_LEN];
288 } __packed;
289 
290 struct vtnet_mac_filter {
291 	struct vtnet_mac_table	vmf_unicast;
292 	uint32_t		vmf_pad; /* Make tables non-contiguous. */
293 	struct vtnet_mac_table	vmf_multicast;
294 };
295 
296 /*
297  * The MAC filter table is malloc(9)'d when needed. Ensure it will
298  * always fit in one segment.
299  */
300 CTASSERT(sizeof(struct vtnet_mac_filter) <= PAGE_SIZE);
301 
302 #define VTNET_TX_TIMEOUT	5
303 #define VTNET_CSUM_OFFLOAD	(CSUM_TCP | CSUM_UDP)
304 #define VTNET_CSUM_OFFLOAD_IPV6	(CSUM_TCP_IPV6 | CSUM_UDP_IPV6)
305 
306 #define VTNET_CSUM_ALL_OFFLOAD	\
307     (VTNET_CSUM_OFFLOAD | VTNET_CSUM_OFFLOAD_IPV6 | CSUM_TSO)
308 
309 #define VTNET_COMMON_FEATURES \
310     (VIRTIO_NET_F_MAC			| \
311      VIRTIO_NET_F_STATUS		| \
312      VIRTIO_NET_F_CTRL_GUEST_OFFLOADS	| \
313      VIRTIO_NET_F_MTU			| \
314      VIRTIO_NET_F_CTRL_VQ		| \
315      VIRTIO_NET_F_CTRL_RX		| \
316      VIRTIO_NET_F_CTRL_MAC_ADDR		| \
317      VIRTIO_NET_F_CTRL_VLAN		| \
318      VIRTIO_NET_F_CSUM			| \
319      VIRTIO_NET_F_HOST_TSO4		| \
320      VIRTIO_NET_F_HOST_TSO6		| \
321      VIRTIO_NET_F_HOST_ECN		| \
322      VIRTIO_NET_F_GUEST_CSUM		| \
323      VIRTIO_NET_F_GUEST_TSO4		| \
324      VIRTIO_NET_F_GUEST_TSO6		| \
325      VIRTIO_NET_F_GUEST_ECN		| \
326      VIRTIO_NET_F_MRG_RXBUF		| \
327      VIRTIO_NET_F_MQ			| \
328      VIRTIO_NET_F_SPEED_DUPLEX		| \
329      VIRTIO_RING_F_EVENT_IDX		| \
330      VIRTIO_RING_F_INDIRECT_DESC)
331 
332 #define VTNET_MODERN_FEATURES (VTNET_COMMON_FEATURES)
333 #define VTNET_LEGACY_FEATURES (VTNET_COMMON_FEATURES | VIRTIO_NET_F_GSO)
334 
335 /*
336  * The VIRTIO_NET_F_HOST_TSO[46] features permit us to send the host
337  * frames larger than 1514 bytes.
338  */
339 #define VTNET_TSO_FEATURES (VIRTIO_NET_F_GSO | VIRTIO_NET_F_HOST_TSO4 | \
340     VIRTIO_NET_F_HOST_TSO6 | VIRTIO_NET_F_HOST_ECN)
341 
342 /*
343  * The VIRTIO_NET_F_GUEST_TSO[46] features permit the host to send us
344  * frames larger than 1514 bytes.
345  */
346 #define VTNET_LRO_FEATURES (VIRTIO_NET_F_GUEST_TSO4 | \
347     VIRTIO_NET_F_GUEST_TSO6 | VIRTIO_NET_F_GUEST_ECN)
348 
349 #define VTNET_MIN_MTU		68
350 #define VTNET_MAX_MTU		65536
351 #define VTNET_MAX_RX_SIZE	65550
352 
353 /*
354  * Used to preallocate the VQ indirect descriptors. Modern and mergeable
355  * buffers do not required one segment for the VirtIO header since it is
356  * placed inline at the beginning of the receive buffer.
357  */
358 #define VTNET_RX_SEGS_HDR_INLINE	1
359 #define VTNET_RX_SEGS_HDR_SEPARATE	2
360 #define VTNET_RX_SEGS_LRO_NOMRG		34
361 #define VTNET_TX_SEGS_MIN		32
362 #define VTNET_TX_SEGS_MAX		64
363 
364 CTASSERT(((VTNET_RX_SEGS_LRO_NOMRG - 1) * MCLBYTES) >= VTNET_MAX_RX_SIZE);
365 CTASSERT(((VTNET_TX_SEGS_MAX - 1) * MCLBYTES) >= VTNET_MAX_MTU);
366 
367 /*
368  * Number of slots in the Tx bufrings. This value matches most other
369  * multiqueue drivers.
370  */
371 #define VTNET_DEFAULT_BUFRING_SIZE	4096
372 
373 #define VTNET_CORE_MTX(_sc)		&(_sc)->vtnet_mtx
374 #define VTNET_CORE_LOCK(_sc)		mtx_lock(VTNET_CORE_MTX((_sc)))
375 #define VTNET_CORE_UNLOCK(_sc)		mtx_unlock(VTNET_CORE_MTX((_sc)))
376 #define VTNET_CORE_LOCK_DESTROY(_sc)	mtx_destroy(VTNET_CORE_MTX((_sc)))
377 #define VTNET_CORE_LOCK_ASSERT(_sc)		\
378     mtx_assert(VTNET_CORE_MTX((_sc)), MA_OWNED)
379 #define VTNET_CORE_LOCK_ASSERT_NOTOWNED(_sc)	\
380     mtx_assert(VTNET_CORE_MTX((_sc)), MA_NOTOWNED)
381 
382 #define VTNET_CORE_LOCK_INIT(_sc) do {					\
383     snprintf((_sc)->vtnet_mtx_name, sizeof((_sc)->vtnet_mtx_name),	\
384         "%s", device_get_nameunit((_sc)->vtnet_dev));			\
385     mtx_init(VTNET_CORE_MTX((_sc)), (_sc)->vtnet_mtx_name,		\
386         "VTNET Core Lock", MTX_DEF);					\
387 } while (0)
388 
389 /*
390  * Values for the init_mode argument of vtnet_init_locked().
391  */
392 #define VTNET_INIT_NETMAP_ENTER		1
393 #define VTNET_INIT_NETMAP_EXIT		2
394 
395 #endif /* _IF_VTNETVAR_H */
396