xref: /freebsd/sys/dev/virtio/network/if_vtnetvar.h (revision 3110d4ebd6c0848cf5e25890d01791bb407e2a9b)
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 struct vtnet_softc;
35 
36 struct vtnet_statistics {
37 	uint64_t	mbuf_alloc_failed;
38 
39 	uint64_t	rx_frame_too_large;
40 	uint64_t	rx_enq_replacement_failed;
41 	uint64_t	rx_mergeable_failed;
42 	uint64_t	rx_csum_bad_ethtype;
43 	uint64_t	rx_csum_bad_ipproto;
44 	uint64_t	rx_csum_bad_offset;
45 	uint64_t	rx_csum_bad_proto;
46 	uint64_t	tx_csum_unknown_ethtype;
47 	uint64_t	tx_csum_proto_mismatch;
48 	uint64_t	tx_tso_not_tcp;
49 	uint64_t	tx_tso_without_csum;
50 	uint64_t	tx_defragged;
51 	uint64_t	tx_defrag_failed;
52 
53 	/*
54 	 * These are accumulated from each Rx/Tx queue.
55 	 */
56 	uint64_t	rx_csum_failed;
57 	uint64_t	rx_csum_offloaded;
58 	uint64_t	rx_task_rescheduled;
59 	uint64_t	tx_csum_offloaded;
60 	uint64_t	tx_tso_offloaded;
61 	uint64_t	tx_task_rescheduled;
62 };
63 
64 struct vtnet_rxq_stats {
65 	uint64_t	vrxs_ipackets;	/* if_ipackets */
66 	uint64_t	vrxs_ibytes;	/* if_ibytes */
67 	uint64_t	vrxs_iqdrops;	/* if_iqdrops */
68 	uint64_t	vrxs_ierrors;	/* if_ierrors */
69 	uint64_t	vrxs_csum;
70 	uint64_t	vrxs_csum_failed;
71 	uint64_t	vrxs_host_lro;
72 	uint64_t	vrxs_rescheduled;
73 };
74 
75 struct vtnet_rxq {
76 	struct mtx		 vtnrx_mtx;
77 	struct vtnet_softc	*vtnrx_sc;
78 	struct virtqueue	*vtnrx_vq;
79 	struct sglist		*vtnrx_sg;
80 	int			 vtnrx_id;
81 	struct vtnet_rxq_stats	 vtnrx_stats;
82 	struct taskqueue	*vtnrx_tq;
83 	struct task		 vtnrx_intrtask;
84 	struct lro_ctrl		 vtnrx_lro;
85 #ifdef DEV_NETMAP
86 	uint32_t		 vtnrx_nm_refill;
87 	struct virtio_net_hdr_mrg_rxbuf vtnrx_shrhdr;
88 #endif  /* DEV_NETMAP */
89 	char			 vtnrx_name[16];
90 } __aligned(CACHE_LINE_SIZE);
91 
92 #define VTNET_RXQ_LOCK(_rxq)	mtx_lock(&(_rxq)->vtnrx_mtx)
93 #define VTNET_RXQ_UNLOCK(_rxq)	mtx_unlock(&(_rxq)->vtnrx_mtx)
94 #define VTNET_RXQ_LOCK_ASSERT(_rxq)		\
95     mtx_assert(&(_rxq)->vtnrx_mtx, MA_OWNED)
96 #define VTNET_RXQ_LOCK_ASSERT_NOTOWNED(_rxq)	\
97     mtx_assert(&(_rxq)->vtnrx_mtx, MA_NOTOWNED)
98 
99 struct vtnet_txq_stats {
100 	uint64_t vtxs_opackets;	/* if_opackets */
101 	uint64_t vtxs_obytes;	/* if_obytes */
102 	uint64_t vtxs_omcasts;	/* if_omcasts */
103 	uint64_t vtxs_csum;
104 	uint64_t vtxs_tso;
105 	uint64_t vtxs_rescheduled;
106 };
107 
108 struct vtnet_txq {
109 	struct mtx		 vtntx_mtx;
110 	struct vtnet_softc	*vtntx_sc;
111 	struct virtqueue	*vtntx_vq;
112 	struct sglist		*vtntx_sg;
113 #ifndef VTNET_LEGACY_TX
114 	struct buf_ring		*vtntx_br;
115 #endif
116 	int			 vtntx_id;
117 	int			 vtntx_watchdog;
118 	int			 vtntx_intr_threshold;
119 	struct vtnet_txq_stats	 vtntx_stats;
120 	struct taskqueue	*vtntx_tq;
121 	struct task		 vtntx_intrtask;
122 #ifndef VTNET_LEGACY_TX
123 	struct task		 vtntx_defrtask;
124 #endif
125 #ifdef DEV_NETMAP
126 	struct virtio_net_hdr_mrg_rxbuf vtntx_shrhdr;
127 #endif  /* DEV_NETMAP */
128 	char			 vtntx_name[16];
129 } __aligned(CACHE_LINE_SIZE);
130 
131 #define VTNET_TXQ_LOCK(_txq)	mtx_lock(&(_txq)->vtntx_mtx)
132 #define VTNET_TXQ_TRYLOCK(_txq)	mtx_trylock(&(_txq)->vtntx_mtx)
133 #define VTNET_TXQ_UNLOCK(_txq)	mtx_unlock(&(_txq)->vtntx_mtx)
134 #define VTNET_TXQ_LOCK_ASSERT(_txq)		\
135     mtx_assert(&(_txq)->vtntx_mtx, MA_OWNED)
136 #define VTNET_TXQ_LOCK_ASSERT_NOTOWNED(_txq)	\
137     mtx_assert(&(_txq)->vtntx_mtx, MA_NOTOWNED)
138 
139 struct vtnet_softc {
140 	device_t		 vtnet_dev;
141 	struct ifnet		*vtnet_ifp;
142 	struct vtnet_rxq	*vtnet_rxqs;
143 	struct vtnet_txq	*vtnet_txqs;
144 	pfil_head_t		 vtnet_pfil;
145 	uint64_t		 vtnet_features;
146 
147 	uint32_t		 vtnet_flags;
148 #define VTNET_FLAG_MODERN	 0x0001
149 #define VTNET_FLAG_MAC		 0x0002
150 #define VTNET_FLAG_CTRL_VQ	 0x0004
151 #define VTNET_FLAG_CTRL_RX	 0x0008
152 #define VTNET_FLAG_CTRL_MAC	 0x0010
153 #define VTNET_FLAG_VLAN_FILTER	 0x0020
154 #define VTNET_FLAG_TSO_ECN	 0x0040
155 #define VTNET_FLAG_MRG_RXBUFS	 0x0080
156 #define VTNET_FLAG_LRO_NOMRG	 0x0100
157 #define VTNET_FLAG_MQ		 0x0200
158 #define VTNET_FLAG_INDIRECT	 0x0400
159 #define VTNET_FLAG_EVENT_IDX	 0x0800
160 #define VTNET_FLAG_SUSPENDED	 0x1000
161 #define VTNET_FLAG_FIXUP_NEEDS_CSUM 0x2000
162 #define VTNET_FLAG_SW_LRO	 0x4000
163 
164 	int			 vtnet_hdr_size;
165 	int			 vtnet_rx_nmbufs;
166 	int			 vtnet_rx_clustersz;
167 	int			 vtnet_rx_nsegs;
168 	int			 vtnet_rx_process_limit;
169 	int			 vtnet_link_active;
170 	int			 vtnet_act_vq_pairs;
171 	int			 vtnet_req_vq_pairs;
172 	int			 vtnet_max_vq_pairs;
173 	int			 vtnet_tx_nsegs;
174 	int			 vtnet_if_flags;
175 	int			 vtnet_max_mtu;
176 	int			 vtnet_lro_entry_count;
177 	int			 vtnet_lro_mbufq_depth;
178 
179 	struct virtqueue	*vtnet_ctrl_vq;
180 	struct vtnet_mac_filter	*vtnet_mac_filter;
181 	uint32_t		*vtnet_vlan_filter;
182 
183 	uint64_t		 vtnet_negotiated_features;
184 	struct vtnet_statistics	 vtnet_stats;
185 	struct callout		 vtnet_tick_ch;
186 	struct ifmedia		 vtnet_media;
187 	eventhandler_tag	 vtnet_vlan_attach;
188 	eventhandler_tag	 vtnet_vlan_detach;
189 
190 	struct mtx		 vtnet_mtx;
191 	char			 vtnet_mtx_name[16];
192 	char			 vtnet_hwaddr[ETHER_ADDR_LEN];
193 };
194 
195 static bool
196 vtnet_modern(struct vtnet_softc *sc)
197 {
198 	return ((sc->vtnet_flags & VTNET_FLAG_MODERN) != 0);
199 }
200 
201 static bool
202 vtnet_software_lro(struct vtnet_softc *sc)
203 {
204 	return ((sc->vtnet_flags & VTNET_FLAG_SW_LRO) != 0);
205 }
206 
207 /*
208  * Maximum number of queue pairs we will autoconfigure to.
209  */
210 #define VTNET_MAX_QUEUE_PAIRS	32
211 
212 /*
213  * Additional completed entries can appear in a virtqueue before we can
214  * reenable interrupts. Number of times to retry before scheduling the
215  * taskqueue to process the completed entries.
216  */
217 #define VTNET_INTR_DISABLE_RETRIES	4
218 
219 /*
220  * Similarly, additional completed entries can appear in a virtqueue
221  * between when lasted checked and before notifying the host. Number
222  * of times to retry before scheduling the taskqueue to process the
223  * queue.
224  */
225 #define VTNET_NOTIFY_RETRIES		4
226 
227 /*
228  * Number of words to allocate for the VLAN shadow table. There is one
229  * bit for each VLAN.
230  */
231 #define VTNET_VLAN_FILTER_NWORDS	(4096 / 32)
232 
233 /* We depend on these being the same size (and same layout). */
234 CTASSERT(sizeof(struct virtio_net_hdr_mrg_rxbuf) ==
235     sizeof(struct virtio_net_hdr_v1));
236 
237 /*
238  * In legacy VirtIO when mergeable buffers are not negotiated, this structure
239  * is placed at the beginning of the mbuf data. Use 4 bytes of pad to keep
240  * both the VirtIO header and the data non-contiguous and the frame's payload
241  * 4 byte aligned. Note this padding would not be necessary if the
242  * VIRTIO_F_ANY_LAYOUT feature was negotiated (but we don't support that yet).
243  *
244  * In modern VirtIO or when mergeable buffers are negotiated, the host puts
245  * the VirtIO header in the beginning of the first mbuf's data.
246  */
247 #define VTNET_RX_HEADER_PAD	4
248 struct vtnet_rx_header {
249 	struct virtio_net_hdr	vrh_hdr;
250 	char			vrh_pad[VTNET_RX_HEADER_PAD];
251 } __packed;
252 
253 /*
254  * For each outgoing frame, the vtnet_tx_header below is allocated from
255  * the vtnet_tx_header_zone.
256  */
257 struct vtnet_tx_header {
258 	union {
259 		struct virtio_net_hdr		hdr;
260 		struct virtio_net_hdr_mrg_rxbuf	mhdr;
261 		struct virtio_net_hdr_v1	v1hdr;
262 	} vth_uhdr;
263 
264 	struct mbuf *vth_mbuf;
265 };
266 
267 /*
268  * The VirtIO specification does not place a limit on the number of MAC
269  * addresses the guest driver may request to be filtered. In practice,
270  * the host is constrained by available resources. To simplify this driver,
271  * impose a reasonably high limit of MAC addresses we will filter before
272  * falling back to promiscuous or all-multicast modes.
273  */
274 #define VTNET_MAX_MAC_ENTRIES	128
275 
276 /*
277  * The driver version of struct virtio_net_ctrl_mac but with our predefined
278  * number of MAC addresses allocated. This structure is shared with the host,
279  * so nentries field is in the correct VirtIO endianness.
280  */
281 struct vtnet_mac_table {
282 	uint32_t	nentries;
283 	uint8_t		macs[VTNET_MAX_MAC_ENTRIES][ETHER_ADDR_LEN];
284 } __packed;
285 
286 struct vtnet_mac_filter {
287 	struct vtnet_mac_table	vmf_unicast;
288 	uint32_t		vmf_pad; /* Make tables non-contiguous. */
289 	struct vtnet_mac_table	vmf_multicast;
290 };
291 
292 /*
293  * The MAC filter table is malloc(9)'d when needed. Ensure it will
294  * always fit in one segment.
295  */
296 CTASSERT(sizeof(struct vtnet_mac_filter) <= PAGE_SIZE);
297 
298 #define VTNET_TX_TIMEOUT	5
299 #define VTNET_CSUM_OFFLOAD	(CSUM_TCP | CSUM_UDP)
300 #define VTNET_CSUM_OFFLOAD_IPV6	(CSUM_TCP_IPV6 | CSUM_UDP_IPV6)
301 
302 #define VTNET_CSUM_ALL_OFFLOAD	\
303     (VTNET_CSUM_OFFLOAD | VTNET_CSUM_OFFLOAD_IPV6 | CSUM_TSO)
304 
305 #define VTNET_COMMON_FEATURES \
306     (VIRTIO_NET_F_MAC			| \
307      VIRTIO_NET_F_STATUS		| \
308      VIRTIO_NET_F_CTRL_GUEST_OFFLOADS	| \
309      VIRTIO_NET_F_MTU			| \
310      VIRTIO_NET_F_CTRL_VQ		| \
311      VIRTIO_NET_F_CTRL_RX		| \
312      VIRTIO_NET_F_CTRL_MAC_ADDR		| \
313      VIRTIO_NET_F_CTRL_VLAN		| \
314      VIRTIO_NET_F_CSUM			| \
315      VIRTIO_NET_F_HOST_TSO4		| \
316      VIRTIO_NET_F_HOST_TSO6		| \
317      VIRTIO_NET_F_HOST_ECN		| \
318      VIRTIO_NET_F_GUEST_CSUM		| \
319      VIRTIO_NET_F_GUEST_TSO4		| \
320      VIRTIO_NET_F_GUEST_TSO6		| \
321      VIRTIO_NET_F_GUEST_ECN		| \
322      VIRTIO_NET_F_MRG_RXBUF		| \
323      VIRTIO_NET_F_MQ			| \
324      VIRTIO_NET_F_SPEED_DUPLEX		| \
325      VIRTIO_RING_F_EVENT_IDX		| \
326      VIRTIO_RING_F_INDIRECT_DESC)
327 
328 #define VTNET_MODERN_FEATURES (VTNET_COMMON_FEATURES)
329 #define VTNET_LEGACY_FEATURES (VTNET_COMMON_FEATURES | VIRTIO_NET_F_GSO)
330 
331 /*
332  * The VIRTIO_NET_F_HOST_TSO[46] features permit us to send the host
333  * frames larger than 1514 bytes.
334  */
335 #define VTNET_TSO_FEATURES (VIRTIO_NET_F_GSO | VIRTIO_NET_F_HOST_TSO4 | \
336     VIRTIO_NET_F_HOST_TSO6 | VIRTIO_NET_F_HOST_ECN)
337 
338 /*
339  * The VIRTIO_NET_F_GUEST_TSO[46] features permit the host to send us
340  * frames larger than 1514 bytes.
341  */
342 #define VTNET_LRO_FEATURES (VIRTIO_NET_F_GUEST_TSO4 | \
343     VIRTIO_NET_F_GUEST_TSO6 | VIRTIO_NET_F_GUEST_ECN)
344 
345 #define VTNET_MIN_MTU		68
346 #define VTNET_MAX_MTU		65536
347 #define VTNET_MAX_RX_SIZE	65550
348 
349 /*
350  * Used to preallocate the VQ indirect descriptors. Modern and mergeable
351  * buffers do not required one segment for the VirtIO header since it is
352  * placed inline at the beginning of the receive buffer.
353  */
354 #define VTNET_RX_SEGS_HDR_INLINE	1
355 #define VTNET_RX_SEGS_HDR_SEPARATE	2
356 #define VTNET_RX_SEGS_LRO_NOMRG		34
357 #define VTNET_TX_SEGS_MIN		32
358 #define VTNET_TX_SEGS_MAX		64
359 
360 CTASSERT(((VTNET_RX_SEGS_LRO_NOMRG - 1) * MCLBYTES) >= VTNET_MAX_RX_SIZE);
361 CTASSERT(((VTNET_TX_SEGS_MAX - 1) * MCLBYTES) >= VTNET_MAX_MTU);
362 
363 /*
364  * Number of slots in the Tx bufrings. This value matches most other
365  * multiqueue drivers.
366  */
367 #define VTNET_DEFAULT_BUFRING_SIZE	4096
368 
369 #define VTNET_CORE_MTX(_sc)		&(_sc)->vtnet_mtx
370 #define VTNET_CORE_LOCK(_sc)		mtx_lock(VTNET_CORE_MTX((_sc)))
371 #define VTNET_CORE_UNLOCK(_sc)		mtx_unlock(VTNET_CORE_MTX((_sc)))
372 #define VTNET_CORE_LOCK_DESTROY(_sc)	mtx_destroy(VTNET_CORE_MTX((_sc)))
373 #define VTNET_CORE_LOCK_ASSERT(_sc)		\
374     mtx_assert(VTNET_CORE_MTX((_sc)), MA_OWNED)
375 #define VTNET_CORE_LOCK_ASSERT_NOTOWNED(_sc)	\
376     mtx_assert(VTNET_CORE_MTX((_sc)), MA_NOTOWNED)
377 
378 #define VTNET_CORE_LOCK_INIT(_sc) do {					\
379     snprintf((_sc)->vtnet_mtx_name, sizeof((_sc)->vtnet_mtx_name),	\
380         "%s", device_get_nameunit((_sc)->vtnet_dev));			\
381     mtx_init(VTNET_CORE_MTX((_sc)), (_sc)->vtnet_mtx_name,		\
382         "VTNET Core Lock", MTX_DEF);					\
383 } while (0)
384 
385 /*
386  * Values for the init_mode argument of vtnet_init_locked().
387  */
388 #define VTNET_INIT_NETMAP_ENTER		1
389 #define VTNET_INIT_NETMAP_EXIT		2
390 
391 #endif /* _IF_VTNETVAR_H */
392