1 /*
2  * Copyright (c) 2013  Chris Torek <torek @ torek net>
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, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  */
26 /*
27  * This file and its contents are supplied under the terms of the
28  * Common Development and Distribution License ("CDDL"), version 1.0.
29  * You may only use this file in accordance with the terms of version
30  * 1.0 of the CDDL.
31  *
32  * A full copy of the text of the CDDL should have accompanied this
33  * source.  A copy of the CDDL is also available via the Internet at
34  * http://www.illumos.org/license/CDDL.
35  *
36  * Copyright 2015 Pluribus Networks Inc.
37  * Copyright 2019 Joyent, Inc.
38  */
39 
40 #ifndef	_VIONA_IMPL_H
41 #define	_VIONA_IMPL_H
42 
43 #include <sys/ddi.h>
44 #include <sys/list.h>
45 #include <sys/sunddi.h>
46 #include <sys/sunndi.h>
47 #include <sys/strsun.h>
48 #include <sys/sysmacros.h>
49 #include <sys/uio.h>
50 
51 #include <sys/mac_client.h>
52 #include <sys/mac_provider.h>
53 #include <sys/mac_client_priv.h>
54 #include <sys/neti.h>
55 #include <inet/ip.h>
56 #include <inet/tcp.h>
57 
58 #include <sys/vmm_drv.h>
59 #include <sys/viona_io.h>
60 
61 struct viona_link;
62 typedef struct viona_link viona_link_t;
63 struct viona_desb;
64 typedef struct viona_desb viona_desb_t;
65 struct viona_net;
66 typedef struct viona_neti viona_neti_t;
67 
68 enum viona_ring_state {
69 	VRS_RESET	= 0x0,	/* just allocated or reset */
70 	VRS_SETUP	= 0x1,	/* addrs setup and starting worker thread */
71 	VRS_INIT	= 0x2,	/* worker thread started & waiting to run */
72 	VRS_RUN		= 0x3,	/* running work routine */
73 	VRS_STOP	= 0x4,	/* worker is exiting */
74 };
75 enum viona_ring_state_flags {
76 	VRSF_REQ_START	= 0x1,	/* start running from INIT state */
77 	VRSF_REQ_STOP	= 0x2,	/* stop running, clean up, goto RESET state */
78 	VRSF_RENEW	= 0x4,	/* ring renewing lease */
79 };
80 
81 typedef struct viona_vring {
82 	viona_link_t	*vr_link;
83 
84 	kmutex_t	vr_lock;
85 	kcondvar_t	vr_cv;
86 	uint16_t	vr_state;
87 	uint16_t	vr_state_flags;
88 	uint_t		vr_xfer_outstanding;
89 	kthread_t	*vr_worker_thread;
90 	vmm_lease_t	*vr_lease;
91 
92 	/* ring-sized resources for TX activity */
93 	viona_desb_t	*vr_txdesb;
94 	struct iovec	*vr_txiov;
95 
96 	uint_t		vr_intr_enabled;
97 	uint64_t	vr_msi_addr;
98 	uint64_t	vr_msi_msg;
99 
100 	/* Internal ring-related state */
101 	kmutex_t	vr_a_mutex;	/* sync consumers of 'avail' */
102 	kmutex_t	vr_u_mutex;	/* sync consumers of 'used' */
103 	uint64_t	vr_pa;
104 	uint16_t	vr_size;
105 	uint16_t	vr_mask;	/* cached from vr_size */
106 	uint16_t	vr_cur_aidx;	/* trails behind 'avail_idx' */
107 
108 	/* Host-context pointers to the queue */
109 	volatile struct virtio_desc	*vr_descr;
110 
111 	volatile uint16_t		*vr_avail_flags;
112 	volatile uint16_t		*vr_avail_idx;
113 	volatile uint16_t		*vr_avail_ring;
114 	volatile uint16_t		*vr_avail_used_event;
115 
116 	volatile uint16_t		*vr_used_flags;
117 	volatile uint16_t		*vr_used_idx;
118 	volatile struct virtio_used	*vr_used_ring;
119 	volatile uint16_t		*vr_used_avail_event;
120 
121 	/* Per-ring error condition statistics */
122 	struct viona_ring_stats {
123 		uint64_t	rs_ndesc_too_high;
124 		uint64_t	rs_bad_idx;
125 		uint64_t	rs_indir_bad_len;
126 		uint64_t	rs_indir_bad_nest;
127 		uint64_t	rs_indir_bad_next;
128 		uint64_t	rs_no_space;
129 		uint64_t	rs_too_many_desc;
130 		uint64_t	rs_desc_bad_len;
131 
132 		uint64_t	rs_bad_ring_addr;
133 
134 		uint64_t	rs_fail_hcksum;
135 		uint64_t	rs_fail_hcksum6;
136 		uint64_t	rs_fail_hcksum_proto;
137 
138 		uint64_t	rs_bad_rx_frame;
139 		uint64_t	rs_rx_merge_overrun;
140 		uint64_t	rs_rx_merge_underrun;
141 		uint64_t	rs_rx_pad_short;
142 		uint64_t	rs_rx_mcast_check;
143 		uint64_t	rs_too_short;
144 		uint64_t	rs_tx_absent;
145 
146 		uint64_t	rs_rx_hookdrop;
147 		uint64_t	rs_tx_hookdrop;
148 	} vr_stats;
149 } viona_vring_t;
150 
151 struct viona_link {
152 	vmm_hold_t		*l_vm_hold;
153 	boolean_t		l_destroyed;
154 
155 	viona_vring_t		l_vrings[VIONA_VQ_MAX];
156 
157 	uint32_t		l_features;
158 	uint32_t		l_features_hw;
159 	uint32_t		l_cap_csum;
160 
161 	uintptr_t		l_notify_ioport;
162 	void			*l_notify_cookie;
163 
164 	datalink_id_t		l_linkid;
165 	mac_handle_t		l_mh;
166 	mac_client_handle_t	l_mch;
167 	mac_promisc_handle_t	l_mph;
168 
169 	pollhead_t		l_pollhead;
170 
171 	viona_neti_t		*l_neti;
172 };
173 
174 typedef struct viona_nethook {
175 	net_handle_t		vnh_neti;
176 	hook_family_t		vnh_family;
177 	hook_event_t		vnh_event_in;
178 	hook_event_t		vnh_event_out;
179 	hook_event_token_t	vnh_token_in;
180 	hook_event_token_t	vnh_token_out;
181 	boolean_t		vnh_hooked;
182 } viona_nethook_t;
183 
184 struct viona_neti {
185 	list_node_t		vni_node;
186 
187 	netid_t			vni_netid;
188 	zoneid_t		vni_zid;
189 
190 	viona_nethook_t		vni_nethook;
191 
192 	kmutex_t		vni_lock;	/* Protects remaining members */
193 	kcondvar_t		vni_ref_change; /* Protected by vni_lock */
194 	uint_t			vni_ref;	/* Protected by vni_lock */
195 	list_t			vni_dev_list;	/* Protected by vni_lock */
196 };
197 
198 typedef struct used_elem {
199 	uint16_t	id;
200 	uint32_t	len;
201 } used_elem_t;
202 
203 typedef struct viona_soft_state {
204 	kmutex_t		ss_lock;
205 	viona_link_t		*ss_link;
206 	list_node_t		ss_node;
207 } viona_soft_state_t;
208 
209 #pragma pack(1)
210 struct virtio_desc {
211 	uint64_t	vd_addr;
212 	uint32_t	vd_len;
213 	uint16_t	vd_flags;
214 	uint16_t	vd_next;
215 };
216 
217 struct virtio_used {
218 	uint32_t	vu_idx;
219 	uint32_t	vu_tlen;
220 };
221 
222 struct virtio_net_mrgrxhdr {
223 	uint8_t		vrh_flags;
224 	uint8_t		vrh_gso_type;
225 	uint16_t	vrh_hdr_len;
226 	uint16_t	vrh_gso_size;
227 	uint16_t	vrh_csum_start;
228 	uint16_t	vrh_csum_offset;
229 	uint16_t	vrh_bufs;
230 };
231 
232 struct virtio_net_hdr {
233 	uint8_t		vrh_flags;
234 	uint8_t		vrh_gso_type;
235 	uint16_t	vrh_hdr_len;
236 	uint16_t	vrh_gso_size;
237 	uint16_t	vrh_csum_start;
238 	uint16_t	vrh_csum_offset;
239 };
240 #pragma pack()
241 
242 #define	VRING_NEED_BAIL(ring, proc)					\
243 		(((ring)->vr_state_flags & VRSF_REQ_STOP) != 0 ||	\
244 		((proc)->p_flag & SEXITING) != 0)
245 
246 
247 #define	VNETHOOK_INTERESTED_IN(neti) \
248 	(neti)->vni_nethook.vnh_event_in.he_interested
249 #define	VNETHOOK_INTERESTED_OUT(neti) \
250 	(neti)->vni_nethook.vnh_event_out.he_interested
251 
252 
253 #define	VIONA_PROBE(name)	DTRACE_PROBE(viona__##name)
254 #define	VIONA_PROBE1(name, arg1, arg2)	\
255 	DTRACE_PROBE1(viona__##name, arg1, arg2)
256 #define	VIONA_PROBE2(name, arg1, arg2, arg3, arg4)	\
257 	DTRACE_PROBE2(viona__##name, arg1, arg2, arg3, arg4)
258 #define	VIONA_PROBE3(name, arg1, arg2, arg3, arg4, arg5, arg6)	\
259 	DTRACE_PROBE3(viona__##name, arg1, arg2, arg3, arg4, arg5, arg6)
260 #define	VIONA_PROBE5(name, arg1, arg2, arg3, arg4, arg5, arg6, arg7, arg8, \
261 	arg9, arg10) \
262 	DTRACE_PROBE5(viona__##name, arg1, arg2, arg3, arg4, arg5, arg6, arg7, \
263 	arg8, arg9, arg10)
264 #define	VIONA_PROBE_BAD_RING_ADDR(r, a)		\
265 	VIONA_PROBE2(bad_ring_addr, viona_vring_t *, r, void *, (void *)(a))
266 
267 #define	VIONA_RING_STAT_INCR(r, name)	\
268 	(((r)->vr_stats.rs_ ## name)++)
269 
270 
271 #define	VIONA_MAX_HDRS_LEN	(sizeof (struct ether_vlan_header) + \
272 	IP_MAX_HDR_LENGTH + TCP_MAX_HDR_LENGTH)
273 
274 #define	VRING_AVAIL_F_NO_INTERRUPT	1
275 #define	VRING_USED_F_NO_NOTIFY		1
276 
277 #define	VRING_DESC_F_NEXT	(1 << 0)
278 #define	VRING_DESC_F_WRITE	(1 << 1)
279 #define	VRING_DESC_F_INDIRECT	(1 << 2)
280 
281 #define	VIRTIO_NET_HDR_F_NEEDS_CSUM	(1 << 0)
282 #define	VIRTIO_NET_HDR_F_DATA_VALID	(1 << 1)
283 
284 #define	VIRTIO_NET_HDR_GSO_NONE		0
285 #define	VIRTIO_NET_HDR_GSO_TCPV4	1
286 
287 #define	VIRTIO_NET_F_CSUM		(1 << 0)
288 #define	VIRTIO_NET_F_GUEST_CSUM		(1 << 1)
289 #define	VIRTIO_NET_F_MAC		(1 << 5) /* host supplies MAC */
290 #define	VIRTIO_NET_F_GUEST_TSO4		(1 << 7) /* guest can accept TSO */
291 #define	VIRTIO_NET_F_HOST_TSO4		(1 << 11) /* host can accept TSO */
292 #define	VIRTIO_NET_F_MRG_RXBUF		(1 << 15) /* host can merge RX bufs */
293 #define	VIRTIO_NET_F_STATUS		(1 << 16) /* cfg status field present */
294 #define	VIRTIO_F_RING_NOTIFY_ON_EMPTY	(1 << 24)
295 #define	VIRTIO_F_RING_INDIRECT_DESC	(1 << 28)
296 #define	VIRTIO_F_RING_EVENT_IDX		(1 << 29)
297 
298 
299 void viona_ring_alloc(viona_link_t *, viona_vring_t *);
300 void viona_ring_free(viona_vring_t *);
301 int viona_ring_reset(viona_vring_t *, boolean_t);
302 int viona_ring_init(viona_link_t *, uint16_t, uint16_t, uint64_t);
303 boolean_t viona_ring_lease_renew(viona_vring_t *);
304 int vq_popchain(viona_vring_t *, struct iovec *, uint_t, uint16_t *);
305 void vq_pushchain(viona_vring_t *, uint32_t, uint16_t);
306 void vq_pushchain_many(viona_vring_t *, uint_t, used_elem_t *);
307 void viona_intr_ring(viona_vring_t *ring);
308 
309 void viona_rx_init(void);
310 void viona_rx_fini(void);
311 int viona_rx_set(viona_link_t *);
312 void viona_rx_clear(viona_link_t *);
313 void viona_worker_rx(viona_vring_t *, viona_link_t *);
314 
315 extern kmutex_t viona_force_copy_lock;
316 void viona_worker_tx(viona_vring_t *, viona_link_t *);
317 void viona_tx_ring_alloc(viona_vring_t *, const uint16_t);
318 void viona_tx_ring_free(viona_vring_t *, const uint16_t);
319 
320 void viona_neti_attach(void);
321 void viona_neti_detach(void);
322 viona_neti_t *viona_neti_lookup_by_zid(zoneid_t);
323 void viona_neti_rele(viona_neti_t *);
324 int viona_hook(viona_link_t *, viona_vring_t *, mblk_t **, boolean_t);
325 
326 #endif	/* _VIONA_IMPL_H */
327