xref: /freebsd/sys/dev/virtio/network/if_vtnetvar.h (revision fa7ca1e332414a8c5e05aff3e6bb17854b91cec6)
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_bad_ethtype;
47 	uint64_t	tx_tso_bad_ethtype;
48 	uint64_t	tx_tso_not_tcp;
49 	uint64_t	tx_defragged;
50 	uint64_t	tx_defrag_failed;
51 
52 	/*
53 	 * These are accumulated from each Rx/Tx queue.
54 	 */
55 	uint64_t	rx_csum_failed;
56 	uint64_t	rx_csum_offloaded;
57 	uint64_t	rx_task_rescheduled;
58 	uint64_t	tx_csum_offloaded;
59 	uint64_t	tx_tso_offloaded;
60 	uint64_t	tx_task_rescheduled;
61 };
62 
63 struct vtnet_rxq_stats {
64 	uint64_t	vrxs_ipackets;	/* if_ipackets */
65 	uint64_t	vrxs_ibytes;	/* if_ibytes */
66 	uint64_t	vrxs_iqdrops;	/* if_iqdrops */
67 	uint64_t	vrxs_ierrors;	/* if_ierrors */
68 	uint64_t	vrxs_csum;
69 	uint64_t	vrxs_csum_failed;
70 	uint64_t	vrxs_rescheduled;
71 };
72 
73 struct vtnet_rxq {
74 	struct mtx		 vtnrx_mtx;
75 	struct vtnet_softc	*vtnrx_sc;
76 	struct virtqueue	*vtnrx_vq;
77 	struct sglist		*vtnrx_sg;
78 	int			 vtnrx_id;
79 	struct vtnet_rxq_stats	 vtnrx_stats;
80 	struct taskqueue	*vtnrx_tq;
81 	struct task		 vtnrx_intrtask;
82 #ifdef DEV_NETMAP
83 	uint32_t		 vtnrx_nm_refill;
84 	struct virtio_net_hdr_mrg_rxbuf vtnrx_shrhdr;
85 #endif  /* DEV_NETMAP */
86 	char			 vtnrx_name[16];
87 } __aligned(CACHE_LINE_SIZE);
88 
89 #define VTNET_RXQ_LOCK(_rxq)	mtx_lock(&(_rxq)->vtnrx_mtx)
90 #define VTNET_RXQ_UNLOCK(_rxq)	mtx_unlock(&(_rxq)->vtnrx_mtx)
91 #define VTNET_RXQ_LOCK_ASSERT(_rxq)		\
92     mtx_assert(&(_rxq)->vtnrx_mtx, MA_OWNED)
93 #define VTNET_RXQ_LOCK_ASSERT_NOTOWNED(_rxq)	\
94     mtx_assert(&(_rxq)->vtnrx_mtx, MA_NOTOWNED)
95 
96 struct vtnet_txq_stats {
97 	uint64_t vtxs_opackets;	/* if_opackets */
98 	uint64_t vtxs_obytes;	/* if_obytes */
99 	uint64_t vtxs_omcasts;	/* if_omcasts */
100 	uint64_t vtxs_csum;
101 	uint64_t vtxs_tso;
102 	uint64_t vtxs_rescheduled;
103 };
104 
105 struct vtnet_txq {
106 	struct mtx		 vtntx_mtx;
107 	struct vtnet_softc	*vtntx_sc;
108 	struct virtqueue	*vtntx_vq;
109 	struct sglist		*vtntx_sg;
110 #ifndef VTNET_LEGACY_TX
111 	struct buf_ring		*vtntx_br;
112 #endif
113 	int			 vtntx_id;
114 	int			 vtntx_watchdog;
115 	struct vtnet_txq_stats	 vtntx_stats;
116 	struct taskqueue	*vtntx_tq;
117 	struct task		 vtntx_intrtask;
118 #ifndef VTNET_LEGACY_TX
119 	struct task		 vtntx_defrtask;
120 #endif
121 #ifdef DEV_NETMAP
122 	struct virtio_net_hdr_mrg_rxbuf vtntx_shrhdr;
123 #endif  /* DEV_NETMAP */
124 	char			 vtntx_name[16];
125 } __aligned(CACHE_LINE_SIZE);
126 
127 #define VTNET_TXQ_LOCK(_txq)	mtx_lock(&(_txq)->vtntx_mtx)
128 #define VTNET_TXQ_TRYLOCK(_txq)	mtx_trylock(&(_txq)->vtntx_mtx)
129 #define VTNET_TXQ_UNLOCK(_txq)	mtx_unlock(&(_txq)->vtntx_mtx)
130 #define VTNET_TXQ_LOCK_ASSERT(_txq)		\
131     mtx_assert(&(_txq)->vtntx_mtx, MA_OWNED)
132 #define VTNET_TXQ_LOCK_ASSERT_NOTOWNED(_txq)	\
133     mtx_assert(&(_txq)->vtntx_mtx, MA_NOTOWNED)
134 
135 struct vtnet_softc {
136 	device_t		 vtnet_dev;
137 	struct ifnet		*vtnet_ifp;
138 	struct vtnet_rxq	*vtnet_rxqs;
139 	struct vtnet_txq	*vtnet_txqs;
140 	pfil_head_t		 vtnet_pfil;
141 
142 	uint32_t		 vtnet_flags;
143 #define VTNET_FLAG_MODERN	 0x0001
144 #define VTNET_FLAG_MAC		 0x0002
145 #define VTNET_FLAG_CTRL_VQ	 0x0004
146 #define VTNET_FLAG_CTRL_RX	 0x0008
147 #define VTNET_FLAG_CTRL_MAC	 0x0010
148 #define VTNET_FLAG_VLAN_FILTER	 0x0020
149 #define VTNET_FLAG_TSO_ECN	 0x0040
150 #define VTNET_FLAG_MRG_RXBUFS	 0x0080
151 #define VTNET_FLAG_LRO_NOMRG	 0x0100
152 #define VTNET_FLAG_MQ		 0x0200
153 #define VTNET_FLAG_INDIRECT	 0x0400
154 #define VTNET_FLAG_EVENT_IDX	 0x0800
155 #define VTNET_FLAG_SUSPENDED	 0x1000
156 #define VTNET_FLAG_FIXUP_NEEDS_CSUM 0x2000
157 
158 	int			 vtnet_link_active;
159 	int			 vtnet_hdr_size;
160 	int			 vtnet_rx_process_limit;
161 	int			 vtnet_rx_nsegs;
162 	int			 vtnet_rx_nmbufs;
163 	int			 vtnet_rx_clustersz;
164 	int			 vtnet_tx_intr_thresh;
165 	int			 vtnet_tx_nsegs;
166 	int			 vtnet_if_flags;
167 	int			 vtnet_act_vq_pairs;
168 	int			 vtnet_max_vq_pairs;
169 	int			 vtnet_requested_vq_pairs;
170 
171 	struct virtqueue	*vtnet_ctrl_vq;
172 	struct vtnet_mac_filter	*vtnet_mac_filter;
173 	uint32_t		*vtnet_vlan_filter;
174 
175 	uint64_t		 vtnet_features;
176 	struct vtnet_statistics	 vtnet_stats;
177 	struct callout		 vtnet_tick_ch;
178 	struct ifmedia		 vtnet_media;
179 	eventhandler_tag	 vtnet_vlan_attach;
180 	eventhandler_tag	 vtnet_vlan_detach;
181 
182 	struct mtx		 vtnet_mtx;
183 	char			 vtnet_mtx_name[16];
184 	char			 vtnet_hwaddr[ETHER_ADDR_LEN];
185 };
186 
187 static bool
188 vtnet_modern(struct vtnet_softc *sc)
189 {
190 	return ((sc->vtnet_flags & VTNET_FLAG_MODERN) != 0);
191 }
192 
193 /*
194  * Maximum number of queue pairs we will autoconfigure to.
195  */
196 #define VTNET_MAX_QUEUE_PAIRS	32
197 
198 /*
199  * Additional completed entries can appear in a virtqueue before we can
200  * reenable interrupts. Number of times to retry before scheduling the
201  * taskqueue to process the completed entries.
202  */
203 #define VTNET_INTR_DISABLE_RETRIES	4
204 
205 /*
206  * Similarly, additional completed entries can appear in a virtqueue
207  * between when lasted checked and before notifying the host. Number
208  * of times to retry before scheduling the taskqueue to process the
209  * queue.
210  */
211 #define VTNET_NOTIFY_RETRIES		4
212 
213 /*
214  * Fake the media type. The host does not provide us with any real media
215  * information.
216  */
217 #define VTNET_MEDIATYPE		 (IFM_ETHER | IFM_10G_T | IFM_FDX)
218 
219 /*
220  * Number of words to allocate for the VLAN shadow table. There is one
221  * bit for each VLAN.
222  */
223 #define VTNET_VLAN_FILTER_NWORDS	(4096 / 32)
224 
225 /* We depend on these being the same size (and same layout). */
226 CTASSERT(sizeof(struct virtio_net_hdr_mrg_rxbuf) ==
227     sizeof(struct virtio_net_hdr_v1));
228 
229 /*
230  * In legacy VirtIO when mergeable buffers are not negotiated, this structure
231  * is placed at the beginning of the mbuf data. Use 4 bytes of pad to keep
232  * both the VirtIO header and the data non-contiguous and the frame's payload
233  * 4 byte aligned. Note this padding would not be necessary if the
234  * VIRTIO_F_ANY_LAYOUT feature was negotiated (but we don't support that yet).
235  *
236  * In modern VirtIO or when mergeable buffers are negotiated, the host puts
237  * the VirtIO header in the beginning of the first mbuf's data.
238  */
239 #define VTNET_RX_HEADER_PAD	4
240 struct vtnet_rx_header {
241 	struct virtio_net_hdr	vrh_hdr;
242 	char			vrh_pad[VTNET_RX_HEADER_PAD];
243 } __packed;
244 
245 /*
246  * For each outgoing frame, the vtnet_tx_header below is allocated from
247  * the vtnet_tx_header_zone.
248  */
249 struct vtnet_tx_header {
250 	union {
251 		struct virtio_net_hdr		hdr;
252 		struct virtio_net_hdr_mrg_rxbuf	mhdr;
253 		struct virtio_net_hdr_v1	v1hdr;
254 	} vth_uhdr;
255 
256 	struct mbuf *vth_mbuf;
257 };
258 
259 /*
260  * The VirtIO specification does not place a limit on the number of MAC
261  * addresses the guest driver may request to be filtered. In practice,
262  * the host is constrained by available resources. To simplify this driver,
263  * impose a reasonably high limit of MAC addresses we will filter before
264  * falling back to promiscuous or all-multicast modes.
265  */
266 #define VTNET_MAX_MAC_ENTRIES	128
267 
268 /*
269  * The driver version of struct virtio_net_ctrl_mac but with our predefined
270  * number of MAC addresses allocated. This structure is shared with the host,
271  * so nentries field is in the correct VirtIO endianness.
272  */
273 struct vtnet_mac_table {
274 	uint32_t	nentries;
275 	uint8_t		macs[VTNET_MAX_MAC_ENTRIES][ETHER_ADDR_LEN];
276 } __packed;
277 
278 struct vtnet_mac_filter {
279 	struct vtnet_mac_table	vmf_unicast;
280 	uint32_t		vmf_pad; /* Make tables non-contiguous. */
281 	struct vtnet_mac_table	vmf_multicast;
282 };
283 
284 /*
285  * The MAC filter table is malloc(9)'d when needed. Ensure it will
286  * always fit in one segment.
287  */
288 CTASSERT(sizeof(struct vtnet_mac_filter) <= PAGE_SIZE);
289 
290 #define VTNET_TX_TIMEOUT	5
291 #define VTNET_CSUM_OFFLOAD	(CSUM_TCP | CSUM_UDP)
292 #define VTNET_CSUM_OFFLOAD_IPV6	(CSUM_TCP_IPV6 | CSUM_UDP_IPV6)
293 
294 #define VTNET_CSUM_ALL_OFFLOAD	\
295     (VTNET_CSUM_OFFLOAD | VTNET_CSUM_OFFLOAD_IPV6 | CSUM_TSO)
296 
297 #define VTNET_COMMON_FEATURES \
298     (VIRTIO_NET_F_MAC			| \
299      VIRTIO_NET_F_STATUS		| \
300      VIRTIO_NET_F_CTRL_VQ		| \
301      VIRTIO_NET_F_CTRL_RX		| \
302      VIRTIO_NET_F_CTRL_MAC_ADDR		| \
303      VIRTIO_NET_F_CTRL_VLAN		| \
304      VIRTIO_NET_F_CSUM			| \
305      VIRTIO_NET_F_HOST_TSO4		| \
306      VIRTIO_NET_F_HOST_TSO6		| \
307      VIRTIO_NET_F_HOST_ECN		| \
308      VIRTIO_NET_F_GUEST_CSUM		| \
309      VIRTIO_NET_F_GUEST_TSO4		| \
310      VIRTIO_NET_F_GUEST_TSO6		| \
311      VIRTIO_NET_F_GUEST_ECN		| \
312      VIRTIO_NET_F_MRG_RXBUF		| \
313      VIRTIO_NET_F_MQ			| \
314      VIRTIO_RING_F_EVENT_IDX		| \
315      VIRTIO_RING_F_INDIRECT_DESC)
316 
317 #define VTNET_MODERN_FEATURES (VTNET_COMMON_FEATURES)
318 #define VTNET_LEGACY_FEATURES (VTNET_COMMON_FEATURES | VIRTIO_NET_F_GSO)
319 
320 /*
321  * The VIRTIO_NET_F_HOST_TSO[46] features permit us to send the host
322  * frames larger than 1514 bytes.
323  */
324 #define VTNET_TSO_FEATURES (VIRTIO_NET_F_GSO | VIRTIO_NET_F_HOST_TSO4 | \
325     VIRTIO_NET_F_HOST_TSO6 | VIRTIO_NET_F_HOST_ECN)
326 
327 /*
328  * The VIRTIO_NET_F_GUEST_TSO[46] features permit the host to send us
329  * frames larger than 1514 bytes. We do not yet support software LRO
330  * via tcp_lro_rx().
331  */
332 #define VTNET_LRO_FEATURES (VIRTIO_NET_F_GUEST_TSO4 | \
333     VIRTIO_NET_F_GUEST_TSO6 | VIRTIO_NET_F_GUEST_ECN)
334 
335 #define VTNET_MAX_MTU		65536
336 #define VTNET_MAX_RX_SIZE	65550
337 
338 /*
339  * Used to preallocate the VQ indirect descriptors. Modern and mergeable
340  * buffers do not required one segment for the VirtIO header since it is
341  * placed inline at the beginning of the receive buffer.
342  */
343 #define VTNET_RX_SEGS_HDR_INLINE	1
344 #define VTNET_RX_SEGS_HDR_SEPARATE	2
345 #define VTNET_RX_SEGS_LRO_NOMRG		34
346 #define VTNET_TX_SEGS_MIN		32
347 #define VTNET_TX_SEGS_MAX		64
348 
349 CTASSERT(((VTNET_RX_SEGS_LRO_NOMRG - 1) * MCLBYTES) >= VTNET_MAX_RX_SIZE);
350 CTASSERT(((VTNET_TX_SEGS_MAX - 1) * MCLBYTES) >= VTNET_MAX_MTU);
351 
352 /*
353  * Number of slots in the Tx bufrings. This value matches most other
354  * multiqueue drivers.
355  */
356 #define VTNET_DEFAULT_BUFRING_SIZE	4096
357 
358 #define VTNET_CORE_MTX(_sc)		&(_sc)->vtnet_mtx
359 #define VTNET_CORE_LOCK(_sc)		mtx_lock(VTNET_CORE_MTX((_sc)))
360 #define VTNET_CORE_UNLOCK(_sc)		mtx_unlock(VTNET_CORE_MTX((_sc)))
361 #define VTNET_CORE_LOCK_DESTROY(_sc)	mtx_destroy(VTNET_CORE_MTX((_sc)))
362 #define VTNET_CORE_LOCK_ASSERT(_sc)		\
363     mtx_assert(VTNET_CORE_MTX((_sc)), MA_OWNED)
364 #define VTNET_CORE_LOCK_ASSERT_NOTOWNED(_sc)	\
365     mtx_assert(VTNET_CORE_MTX((_sc)), MA_NOTOWNED)
366 
367 #define VTNET_CORE_LOCK_INIT(_sc) do {					\
368     snprintf((_sc)->vtnet_mtx_name, sizeof((_sc)->vtnet_mtx_name),	\
369         "%s", device_get_nameunit((_sc)->vtnet_dev));			\
370     mtx_init(VTNET_CORE_MTX((_sc)), (_sc)->vtnet_mtx_name,		\
371         "VTNET Core Lock", MTX_DEF);					\
372 } while (0)
373 
374 /*
375  * Values for the init_mode argument of vtnet_init_locked().
376  */
377 #define VTNET_INIT_NETMAP_ENTER		1
378 #define VTNET_INIT_NETMAP_EXIT		2
379 
380 #endif /* _IF_VTNETVAR_H */
381