xref: /freebsd/sys/dev/virtio/network/if_vtnetvar.h (revision 0572ccaa4543b0abef8ef81e384c1d04de9f3da1)
1 /*-
2  * Copyright (c) 2011, Bryan Venteicher <bryanv@FreeBSD.org>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice unmodified, this list of conditions, and the following
10  *    disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25  *
26  * $FreeBSD$
27  */
28 
29 #ifndef _IF_VTNETVAR_H
30 #define _IF_VTNETVAR_H
31 
32 struct vtnet_softc;
33 
34 struct vtnet_statistics {
35 	uint64_t	mbuf_alloc_failed;
36 
37 	uint64_t	rx_frame_too_large;
38 	uint64_t	rx_enq_replacement_failed;
39 	uint64_t	rx_mergeable_failed;
40 	uint64_t	rx_csum_bad_ethtype;
41 	uint64_t	rx_csum_bad_ipproto;
42 	uint64_t	rx_csum_bad_offset;
43 	uint64_t	rx_csum_bad_proto;
44 	uint64_t	tx_csum_bad_ethtype;
45 	uint64_t	tx_tso_bad_ethtype;
46 	uint64_t	tx_tso_not_tcp;
47 	uint64_t	tx_defragged;
48 	uint64_t	tx_defrag_failed;
49 
50 	/*
51 	 * These are accumulated from each Rx/Tx queue.
52 	 */
53 	uint64_t	rx_csum_failed;
54 	uint64_t	rx_csum_offloaded;
55 	uint64_t	rx_task_rescheduled;
56 	uint64_t	tx_csum_offloaded;
57 	uint64_t	tx_tso_offloaded;
58 	uint64_t	tx_task_rescheduled;
59 };
60 
61 struct vtnet_rxq_stats {
62 	uint64_t	vrxs_ipackets;	/* if_ipackets */
63 	uint64_t	vrxs_ibytes;	/* if_ibytes */
64 	uint64_t	vrxs_iqdrops;	/* if_iqdrops */
65 	uint64_t	vrxs_ierrors;	/* if_ierrors */
66 	uint64_t	vrxs_csum;
67 	uint64_t	vrxs_csum_failed;
68 	uint64_t	vrxs_rescheduled;
69 };
70 
71 struct vtnet_rxq {
72 	struct mtx		 vtnrx_mtx;
73 	struct vtnet_softc	*vtnrx_sc;
74 	struct virtqueue	*vtnrx_vq;
75 	struct sglist		*vtnrx_sg;
76 	int			 vtnrx_id;
77 	int			 vtnrx_process_limit;
78 	struct vtnet_rxq_stats	 vtnrx_stats;
79 	struct taskqueue	*vtnrx_tq;
80 	struct task		 vtnrx_intrtask;
81 	char			 vtnrx_name[16];
82 } __aligned(CACHE_LINE_SIZE);
83 
84 #define VTNET_RXQ_LOCK(_rxq)	mtx_lock(&(_rxq)->vtnrx_mtx)
85 #define VTNET_RXQ_UNLOCK(_rxq)	mtx_unlock(&(_rxq)->vtnrx_mtx)
86 #define VTNET_RXQ_LOCK_ASSERT(_rxq)		\
87     mtx_assert(&(_rxq)->vtnrx_mtx, MA_OWNED)
88 #define VTNET_RXQ_LOCK_ASSERT_NOTOWNED(_rxq)	\
89     mtx_assert(&(_rxq)->vtnrx_mtx, MA_NOTOWNED)
90 
91 struct vtnet_txq_stats {
92 	uint64_t vtxs_opackets;	/* if_opackets */
93 	uint64_t vtxs_obytes;	/* if_obytes */
94 	uint64_t vtxs_omcasts;	/* if_omcasts */
95 	uint64_t vtxs_csum;
96 	uint64_t vtxs_tso;
97 	uint64_t vtxs_rescheduled;
98 };
99 
100 struct vtnet_txq {
101 	struct mtx		 vtntx_mtx;
102 	struct vtnet_softc	*vtntx_sc;
103 	struct virtqueue	*vtntx_vq;
104 	struct sglist		*vtntx_sg;
105 #ifndef VTNET_LEGACY_TX
106 	struct buf_ring		*vtntx_br;
107 #endif
108 	int			 vtntx_id;
109 	int			 vtntx_watchdog;
110 	struct vtnet_txq_stats	 vtntx_stats;
111 	struct taskqueue	*vtntx_tq;
112 	struct task		 vtntx_intrtask;
113 #ifndef VTNET_LEGACY_TX
114 	struct task		 vtntx_defrtask;
115 #endif
116 	char			 vtntx_name[16];
117 } __aligned(CACHE_LINE_SIZE);
118 
119 #define VTNET_TXQ_LOCK(_txq)	mtx_lock(&(_txq)->vtntx_mtx)
120 #define VTNET_TXQ_TRYLOCK(_txq)	mtx_trylock(&(_txq)->vtntx_mtx)
121 #define VTNET_TXQ_UNLOCK(_txq)	mtx_unlock(&(_txq)->vtntx_mtx)
122 #define VTNET_TXQ_LOCK_ASSERT(_txq)		\
123     mtx_assert(&(_txq)->vtntx_mtx, MA_OWNED)
124 #define VTNET_TXQ_LOCK_ASSERT_NOTOWNED(_txq)	\
125     mtx_assert(&(_txq)->vtntx_mtx, MA_NOTOWNED)
126 
127 struct vtnet_softc {
128 	device_t		 vtnet_dev;
129 	struct ifnet		*vtnet_ifp;
130 	struct vtnet_rxq	*vtnet_rxqs;
131 	struct vtnet_txq	*vtnet_txqs;
132 
133 	uint32_t		 vtnet_flags;
134 #define VTNET_FLAG_SUSPENDED	 0x0001
135 #define VTNET_FLAG_MAC		 0x0002
136 #define VTNET_FLAG_CTRL_VQ	 0x0004
137 #define VTNET_FLAG_CTRL_RX	 0x0008
138 #define VTNET_FLAG_CTRL_MAC	 0x0010
139 #define VTNET_FLAG_VLAN_FILTER	 0x0020
140 #define VTNET_FLAG_TSO_ECN	 0x0040
141 #define VTNET_FLAG_MRG_RXBUFS	 0x0080
142 #define VTNET_FLAG_LRO_NOMRG	 0x0100
143 #define VTNET_FLAG_MULTIQ	 0x0200
144 #define VTNET_FLAG_EVENT_IDX	 0x0400
145 
146 	int			 vtnet_link_active;
147 	int			 vtnet_hdr_size;
148 	int			 vtnet_rx_process_limit;
149 	int			 vtnet_rx_nsegs;
150 	int			 vtnet_rx_nmbufs;
151 	int			 vtnet_rx_clsize;
152 	int			 vtnet_rx_new_clsize;
153 	int			 vtnet_tx_nsegs;
154 	int			 vtnet_if_flags;
155 	int			 vtnet_act_vq_pairs;
156 	int			 vtnet_max_vq_pairs;
157 
158 	struct virtqueue	*vtnet_ctrl_vq;
159 	struct vtnet_mac_filter	*vtnet_mac_filter;
160 	uint32_t		*vtnet_vlan_filter;
161 
162 	uint64_t		 vtnet_features;
163 	struct vtnet_statistics	 vtnet_stats;
164 	struct callout		 vtnet_tick_ch;
165 	struct ifmedia		 vtnet_media;
166 	eventhandler_tag	 vtnet_vlan_attach;
167 	eventhandler_tag	 vtnet_vlan_detach;
168 
169 	struct mtx		 vtnet_mtx;
170 	char			 vtnet_mtx_name[16];
171 	char			 vtnet_hwaddr[ETHER_ADDR_LEN];
172 };
173 
174 /*
175  * Maximum number of queue pairs we will autoconfigure to.
176  */
177 #define VTNET_MAX_QUEUE_PAIRS	8
178 
179 /*
180  * Additional completed entries can appear in a virtqueue before we can
181  * reenable interrupts. Number of times to retry before scheduling the
182  * taskqueue to process the completed entries.
183  */
184 #define VTNET_INTR_DISABLE_RETRIES	4
185 
186 /*
187  * Fake the media type. The host does not provide us with any real media
188  * information.
189  */
190 #define VTNET_MEDIATYPE		 (IFM_ETHER | IFM_10G_T | IFM_FDX)
191 
192 /*
193  * Number of words to allocate for the VLAN shadow table. There is one
194  * bit for each VLAN.
195  */
196 #define VTNET_VLAN_FILTER_NWORDS	(4096 / 32)
197 
198 /*
199  * When mergeable buffers are not negotiated, the vtnet_rx_header structure
200  * below is placed at the beginning of the mbuf data. Use 4 bytes of pad to
201  * both keep the VirtIO header and the data non-contiguous and to keep the
202  * frame's payload 4 byte aligned.
203  *
204  * When mergeable buffers are negotiated, the host puts the VirtIO header in
205  * the beginning of the first mbuf's data.
206  */
207 #define VTNET_RX_HEADER_PAD	4
208 struct vtnet_rx_header {
209 	struct virtio_net_hdr	vrh_hdr;
210 	char			vrh_pad[VTNET_RX_HEADER_PAD];
211 } __packed;
212 
213 /*
214  * For each outgoing frame, the vtnet_tx_header below is allocated from
215  * the vtnet_tx_header_zone.
216  */
217 struct vtnet_tx_header {
218 	union {
219 		struct virtio_net_hdr		hdr;
220 		struct virtio_net_hdr_mrg_rxbuf	mhdr;
221 	} vth_uhdr;
222 
223 	struct mbuf *vth_mbuf;
224 };
225 
226 /*
227  * The VirtIO specification does not place a limit on the number of MAC
228  * addresses the guest driver may request to be filtered. In practice,
229  * the host is constrained by available resources. To simplify this driver,
230  * impose a reasonably high limit of MAC addresses we will filter before
231  * falling back to promiscuous or all-multicast modes.
232  */
233 #define VTNET_MAX_MAC_ENTRIES	128
234 
235 struct vtnet_mac_table {
236 	uint32_t	nentries;
237 	uint8_t		macs[VTNET_MAX_MAC_ENTRIES][ETHER_ADDR_LEN];
238 } __packed;
239 
240 struct vtnet_mac_filter {
241 	struct vtnet_mac_table	vmf_unicast;
242 	uint32_t		vmf_pad; /* Make tables non-contiguous. */
243 	struct vtnet_mac_table	vmf_multicast;
244 };
245 
246 /*
247  * The MAC filter table is malloc(9)'d when needed. Ensure it will
248  * always fit in one segment.
249  */
250 CTASSERT(sizeof(struct vtnet_mac_filter) <= PAGE_SIZE);
251 
252 #define VTNET_TX_TIMEOUT	5
253 #define VTNET_CSUM_OFFLOAD	(CSUM_TCP | CSUM_UDP | CSUM_SCTP)
254 #define VTNET_CSUM_OFFLOAD_IPV6	(CSUM_TCP_IPV6 | CSUM_UDP_IPV6 | CSUM_SCTP_IPV6)
255 
256 #define VTNET_CSUM_ALL_OFFLOAD	\
257     (VTNET_CSUM_OFFLOAD | VTNET_CSUM_OFFLOAD_IPV6 | CSUM_TSO)
258 
259 /* Features desired/implemented by this driver. */
260 #define VTNET_FEATURES \
261     (VIRTIO_NET_F_MAC			| \
262      VIRTIO_NET_F_STATUS		| \
263      VIRTIO_NET_F_CTRL_VQ		| \
264      VIRTIO_NET_F_CTRL_RX		| \
265      VIRTIO_NET_F_CTRL_MAC_ADDR		| \
266      VIRTIO_NET_F_CTRL_VLAN		| \
267      VIRTIO_NET_F_CSUM			| \
268      VIRTIO_NET_F_GSO			| \
269      VIRTIO_NET_F_HOST_TSO4		| \
270      VIRTIO_NET_F_HOST_TSO6		| \
271      VIRTIO_NET_F_HOST_ECN		| \
272      VIRTIO_NET_F_GUEST_CSUM		| \
273      VIRTIO_NET_F_GUEST_TSO4		| \
274      VIRTIO_NET_F_GUEST_TSO6		| \
275      VIRTIO_NET_F_GUEST_ECN		| \
276      VIRTIO_NET_F_MRG_RXBUF		| \
277      VIRTIO_NET_F_MQ			| \
278      VIRTIO_RING_F_EVENT_IDX		| \
279      VIRTIO_RING_F_INDIRECT_DESC)
280 
281 /*
282  * The VIRTIO_NET_F_HOST_TSO[46] features permit us to send the host
283  * frames larger than 1514 bytes.
284  */
285 #define VTNET_TSO_FEATURES (VIRTIO_NET_F_GSO | VIRTIO_NET_F_HOST_TSO4 | \
286     VIRTIO_NET_F_HOST_TSO6 | VIRTIO_NET_F_HOST_ECN)
287 
288 /*
289  * The VIRTIO_NET_F_GUEST_TSO[46] features permit the host to send us
290  * frames larger than 1514 bytes. We do not yet support software LRO
291  * via tcp_lro_rx().
292  */
293 #define VTNET_LRO_FEATURES (VIRTIO_NET_F_GUEST_TSO4 | \
294     VIRTIO_NET_F_GUEST_TSO6 | VIRTIO_NET_F_GUEST_ECN)
295 
296 #define VTNET_MAX_MTU		65536
297 #define VTNET_MAX_RX_SIZE	65550
298 
299 /*
300  * Used to preallocate the Vq indirect descriptors. The first segment
301  * is reserved for the header, except for mergeable buffers since the
302  * header is placed inline with the data.
303  */
304 #define VTNET_MRG_RX_SEGS	1
305 #define VTNET_MIN_RX_SEGS	2
306 #define VTNET_MAX_RX_SEGS	34
307 #define VTNET_MIN_TX_SEGS	4
308 #define VTNET_MAX_TX_SEGS	64
309 
310 /*
311  * Assert we can receive and transmit the maximum with regular
312  * size clusters.
313  */
314 CTASSERT(((VTNET_MAX_RX_SEGS - 1) * MCLBYTES) >= VTNET_MAX_RX_SIZE);
315 CTASSERT(((VTNET_MAX_TX_SEGS - 1) * MCLBYTES) >= VTNET_MAX_MTU);
316 
317 /*
318  * Number of slots in the Tx bufrings. This value matches most other
319  * multiqueue drivers.
320  */
321 #define VTNET_DEFAULT_BUFRING_SIZE	4096
322 
323 /*
324  * Determine how many mbufs are in each receive buffer. For LRO without
325  * mergeable buffers, we must allocate an mbuf chain large enough to
326  * hold both the vtnet_rx_header and the maximum receivable data.
327  */
328 #define VTNET_NEEDED_RX_MBUFS(_sc, _clsize)				\
329 	((_sc)->vtnet_flags & VTNET_FLAG_LRO_NOMRG) == 0 ? 1 :		\
330 	    howmany(sizeof(struct vtnet_rx_header) + VTNET_MAX_RX_SIZE,	\
331 	        (_clsize))
332 
333 #define VTNET_CORE_MTX(_sc)		&(_sc)->vtnet_mtx
334 #define VTNET_CORE_LOCK(_sc)		mtx_lock(VTNET_CORE_MTX((_sc)))
335 #define VTNET_CORE_UNLOCK(_sc)		mtx_unlock(VTNET_CORE_MTX((_sc)))
336 #define VTNET_CORE_LOCK_DESTROY(_sc)	mtx_destroy(VTNET_CORE_MTX((_sc)))
337 #define VTNET_CORE_LOCK_ASSERT(_sc)		\
338     mtx_assert(VTNET_CORE_MTX((_sc)), MA_OWNED)
339 #define VTNET_CORE_LOCK_ASSERT_NOTOWNED(_sc)	\
340     mtx_assert(VTNET_CORE_MTX((_sc)), MA_NOTOWNED)
341 
342 #define VTNET_CORE_LOCK_INIT(_sc) do {					\
343     snprintf((_sc)->vtnet_mtx_name, sizeof((_sc)->vtnet_mtx_name),	\
344         "%s", device_get_nameunit((_sc)->vtnet_dev));			\
345     mtx_init(VTNET_CORE_MTX((_sc)), (_sc)->vtnet_mtx_name,		\
346         "VTNET Core Lock", MTX_DEF);					\
347 } while (0)
348 
349 #endif /* _IF_VTNETVAR_H */
350