xref: /freebsd/sys/dev/hyperv/netvsc/hn_rndis.c (revision a5d223e641705cbe537d23e5c023395a929ab8da)
1 /*-
2  * Copyright (c) 2009-2012,2016 Microsoft Corp.
3  * Copyright (c) 2010-2012 Citrix Inc.
4  * Copyright (c) 2012 NetApp Inc.
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 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include "opt_inet6.h"
33 #include "opt_inet.h"
34 
35 #include <sys/param.h>
36 #include <sys/socket.h>
37 #include <sys/systm.h>
38 #include <sys/taskqueue.h>
39 
40 #include <machine/atomic.h>
41 
42 #include <net/ethernet.h>
43 #include <net/if.h>
44 #include <net/if_var.h>
45 #include <net/if_media.h>
46 #include <net/rndis.h>
47 
48 #include <netinet/in.h>
49 #include <netinet/ip.h>
50 #include <netinet/tcp_lro.h>
51 
52 #include <dev/hyperv/include/hyperv.h>
53 #include <dev/hyperv/include/hyperv_busdma.h>
54 #include <dev/hyperv/include/vmbus.h>
55 #include <dev/hyperv/include/vmbus_xact.h>
56 
57 #include <dev/hyperv/netvsc/ndis.h>
58 #include <dev/hyperv/netvsc/if_hnreg.h>
59 #include <dev/hyperv/netvsc/if_hnvar.h>
60 #include <dev/hyperv/netvsc/hn_nvs.h>
61 #include <dev/hyperv/netvsc/hn_rndis.h>
62 
63 #define HN_RNDIS_RID_COMPAT_MASK	0xffff
64 #define HN_RNDIS_RID_COMPAT_MAX		HN_RNDIS_RID_COMPAT_MASK
65 
66 #define HN_RNDIS_XFER_SIZE		2048
67 
68 #define HN_NDIS_TXCSUM_CAP_IP4		\
69 	(NDIS_TXCSUM_CAP_IP4 | NDIS_TXCSUM_CAP_IP4OPT)
70 #define HN_NDIS_TXCSUM_CAP_TCP4		\
71 	(NDIS_TXCSUM_CAP_TCP4 | NDIS_TXCSUM_CAP_TCP4OPT)
72 #define HN_NDIS_TXCSUM_CAP_TCP6		\
73 	(NDIS_TXCSUM_CAP_TCP6 | NDIS_TXCSUM_CAP_TCP6OPT | \
74 	 NDIS_TXCSUM_CAP_IP6EXT)
75 #define HN_NDIS_TXCSUM_CAP_UDP6		\
76 	(NDIS_TXCSUM_CAP_UDP6 | NDIS_TXCSUM_CAP_IP6EXT)
77 #define HN_NDIS_LSOV2_CAP_IP6		\
78 	(NDIS_LSOV2_CAP_IP6EXT | NDIS_LSOV2_CAP_TCP6OPT)
79 
80 static const void	*hn_rndis_xact_exec1(struct hn_softc *,
81 			    struct vmbus_xact *, size_t,
82 			    struct hn_nvs_sendctx *, size_t *);
83 static const void	*hn_rndis_xact_execute(struct hn_softc *,
84 			    struct vmbus_xact *, uint32_t, size_t, size_t *,
85 			    uint32_t);
86 static int		hn_rndis_query(struct hn_softc *, uint32_t,
87 			    const void *, size_t, void *, size_t *);
88 static int		hn_rndis_query2(struct hn_softc *, uint32_t,
89 			    const void *, size_t, void *, size_t *, size_t);
90 static int		hn_rndis_set(struct hn_softc *, uint32_t,
91 			    const void *, size_t);
92 static int		hn_rndis_init(struct hn_softc *);
93 static int		hn_rndis_halt(struct hn_softc *);
94 static int		hn_rndis_conf_offload(struct hn_softc *, int);
95 static int		hn_rndis_query_hwcaps(struct hn_softc *,
96 			    struct ndis_offload *);
97 
98 static __inline uint32_t
99 hn_rndis_rid(struct hn_softc *sc)
100 {
101 	uint32_t rid;
102 
103 again:
104 	rid = atomic_fetchadd_int(&sc->hn_rndis_rid, 1);
105 	if (rid == 0)
106 		goto again;
107 
108 	/* Use upper 16 bits for non-compat RNDIS messages. */
109 	return ((rid & 0xffff) << 16);
110 }
111 
112 void
113 hn_rndis_rx_ctrl(struct hn_softc *sc, const void *data, int dlen)
114 {
115 	const struct rndis_comp_hdr *comp;
116 	const struct rndis_msghdr *hdr;
117 
118 	KASSERT(dlen >= sizeof(*hdr), ("invalid RNDIS msg\n"));
119 	hdr = data;
120 
121 	switch (hdr->rm_type) {
122 	case REMOTE_NDIS_INITIALIZE_CMPLT:
123 	case REMOTE_NDIS_QUERY_CMPLT:
124 	case REMOTE_NDIS_SET_CMPLT:
125 	case REMOTE_NDIS_KEEPALIVE_CMPLT:	/* unused */
126 		if (dlen < sizeof(*comp)) {
127 			if_printf(sc->hn_ifp, "invalid RNDIS cmplt\n");
128 			return;
129 		}
130 		comp = data;
131 
132 		KASSERT(comp->rm_rid > HN_RNDIS_RID_COMPAT_MAX,
133 		    ("invalid RNDIS rid 0x%08x\n", comp->rm_rid));
134 		vmbus_xact_ctx_wakeup(sc->hn_xact, comp, dlen);
135 		break;
136 
137 	case REMOTE_NDIS_RESET_CMPLT:
138 		/*
139 		 * Reset completed, no rid.
140 		 *
141 		 * NOTE:
142 		 * RESET is not issued by hn(4), so this message should
143 		 * _not_ be observed.
144 		 */
145 		if_printf(sc->hn_ifp, "RESET cmplt received\n");
146 		break;
147 
148 	default:
149 		if_printf(sc->hn_ifp, "unknown RNDIS msg 0x%x\n",
150 		    hdr->rm_type);
151 		break;
152 	}
153 }
154 
155 int
156 hn_rndis_get_eaddr(struct hn_softc *sc, uint8_t *eaddr)
157 {
158 	size_t eaddr_len;
159 	int error;
160 
161 	eaddr_len = ETHER_ADDR_LEN;
162 	error = hn_rndis_query(sc, OID_802_3_PERMANENT_ADDRESS, NULL, 0,
163 	    eaddr, &eaddr_len);
164 	if (error)
165 		return (error);
166 	if (eaddr_len != ETHER_ADDR_LEN) {
167 		if_printf(sc->hn_ifp, "invalid eaddr len %zu\n", eaddr_len);
168 		return (EINVAL);
169 	}
170 	return (0);
171 }
172 
173 int
174 hn_rndis_get_linkstatus(struct hn_softc *sc, uint32_t *link_status)
175 {
176 	size_t size;
177 	int error;
178 
179 	size = sizeof(*link_status);
180 	error = hn_rndis_query(sc, OID_GEN_MEDIA_CONNECT_STATUS, NULL, 0,
181 	    link_status, &size);
182 	if (error)
183 		return (error);
184 	if (size != sizeof(uint32_t)) {
185 		if_printf(sc->hn_ifp, "invalid link status len %zu\n", size);
186 		return (EINVAL);
187 	}
188 	return (0);
189 }
190 
191 static const void *
192 hn_rndis_xact_exec1(struct hn_softc *sc, struct vmbus_xact *xact, size_t reqlen,
193     struct hn_nvs_sendctx *sndc, size_t *comp_len)
194 {
195 	struct vmbus_gpa gpa[HN_XACT_REQ_PGCNT];
196 	int gpa_cnt, error;
197 	bus_addr_t paddr;
198 
199 	KASSERT(reqlen <= HN_XACT_REQ_SIZE && reqlen > 0,
200 	    ("invalid request length %zu", reqlen));
201 
202 	/*
203 	 * Setup the SG list.
204 	 */
205 	paddr = vmbus_xact_req_paddr(xact);
206 	KASSERT((paddr & PAGE_MASK) == 0,
207 	    ("vmbus xact request is not page aligned 0x%jx", (uintmax_t)paddr));
208 	for (gpa_cnt = 0; gpa_cnt < HN_XACT_REQ_PGCNT; ++gpa_cnt) {
209 		int len = PAGE_SIZE;
210 
211 		if (reqlen == 0)
212 			break;
213 		if (reqlen < len)
214 			len = reqlen;
215 
216 		gpa[gpa_cnt].gpa_page = atop(paddr) + gpa_cnt;
217 		gpa[gpa_cnt].gpa_len = len;
218 		gpa[gpa_cnt].gpa_ofs = 0;
219 
220 		reqlen -= len;
221 	}
222 	KASSERT(reqlen == 0, ("still have %zu request data left", reqlen));
223 
224 	/*
225 	 * Send this RNDIS control message and wait for its completion
226 	 * message.
227 	 */
228 	vmbus_xact_activate(xact);
229 	error = hn_nvs_send_rndis_ctrl(sc->hn_prichan, sndc, gpa, gpa_cnt);
230 	if (error) {
231 		vmbus_xact_deactivate(xact);
232 		if_printf(sc->hn_ifp, "RNDIS ctrl send failed: %d\n", error);
233 		return (NULL);
234 	}
235 	return (vmbus_xact_wait(xact, comp_len));
236 }
237 
238 static const void *
239 hn_rndis_xact_execute(struct hn_softc *sc, struct vmbus_xact *xact, uint32_t rid,
240     size_t reqlen, size_t *comp_len0, uint32_t comp_type)
241 {
242 	const struct rndis_comp_hdr *comp;
243 	size_t comp_len, min_complen = *comp_len0;
244 
245 	KASSERT(rid > HN_RNDIS_RID_COMPAT_MAX, ("invalid rid %u\n", rid));
246 	KASSERT(min_complen >= sizeof(*comp),
247 	    ("invalid minimum complete len %zu", min_complen));
248 
249 	/*
250 	 * Execute the xact setup by the caller.
251 	 */
252 	comp = hn_rndis_xact_exec1(sc, xact, reqlen, &hn_nvs_sendctx_none,
253 	    &comp_len);
254 	if (comp == NULL)
255 		return (NULL);
256 
257 	/*
258 	 * Check this RNDIS complete message.
259 	 */
260 	if (comp_len < min_complen) {
261 		if (comp_len >= sizeof(*comp)) {
262 			/* rm_status field is valid */
263 			if_printf(sc->hn_ifp, "invalid RNDIS comp len %zu, "
264 			    "status 0x%08x\n", comp_len, comp->rm_status);
265 		} else {
266 			if_printf(sc->hn_ifp, "invalid RNDIS comp len %zu\n",
267 			    comp_len);
268 		}
269 		return (NULL);
270 	}
271 	if (comp->rm_len < min_complen) {
272 		if_printf(sc->hn_ifp, "invalid RNDIS comp msglen %u\n",
273 		    comp->rm_len);
274 		return (NULL);
275 	}
276 	if (comp->rm_type != comp_type) {
277 		if_printf(sc->hn_ifp, "unexpected RNDIS comp 0x%08x, "
278 		    "expect 0x%08x\n", comp->rm_type, comp_type);
279 		return (NULL);
280 	}
281 	if (comp->rm_rid != rid) {
282 		if_printf(sc->hn_ifp, "RNDIS comp rid mismatch %u, "
283 		    "expect %u\n", comp->rm_rid, rid);
284 		return (NULL);
285 	}
286 	/* All pass! */
287 	*comp_len0 = comp_len;
288 	return (comp);
289 }
290 
291 static int
292 hn_rndis_query(struct hn_softc *sc, uint32_t oid,
293     const void *idata, size_t idlen, void *odata, size_t *odlen0)
294 {
295 
296 	return (hn_rndis_query2(sc, oid, idata, idlen, odata, odlen0, *odlen0));
297 }
298 
299 static int
300 hn_rndis_query2(struct hn_softc *sc, uint32_t oid,
301     const void *idata, size_t idlen, void *odata, size_t *odlen0,
302     size_t min_odlen)
303 {
304 	struct rndis_query_req *req;
305 	const struct rndis_query_comp *comp;
306 	struct vmbus_xact *xact;
307 	size_t reqlen, odlen = *odlen0, comp_len;
308 	int error, ofs;
309 	uint32_t rid;
310 
311 	reqlen = sizeof(*req) + idlen;
312 	xact = vmbus_xact_get(sc->hn_xact, reqlen);
313 	if (xact == NULL) {
314 		if_printf(sc->hn_ifp, "no xact for RNDIS query 0x%08x\n", oid);
315 		return (ENXIO);
316 	}
317 	rid = hn_rndis_rid(sc);
318 	req = vmbus_xact_req_data(xact);
319 	req->rm_type = REMOTE_NDIS_QUERY_MSG;
320 	req->rm_len = reqlen;
321 	req->rm_rid = rid;
322 	req->rm_oid = oid;
323 	/*
324 	 * XXX
325 	 * This is _not_ RNDIS Spec conforming:
326 	 * "This MUST be set to 0 when there is no input data
327 	 *  associated with the OID."
328 	 *
329 	 * If this field was set to 0 according to the RNDIS Spec,
330 	 * Hyper-V would set non-SUCCESS status in the query
331 	 * completion.
332 	 */
333 	req->rm_infobufoffset = RNDIS_QUERY_REQ_INFOBUFOFFSET;
334 
335 	if (idlen > 0) {
336 		req->rm_infobuflen = idlen;
337 		/* Input data immediately follows RNDIS query. */
338 		memcpy(req + 1, idata, idlen);
339 	}
340 
341 	comp_len = sizeof(*comp) + min_odlen;
342 	comp = hn_rndis_xact_execute(sc, xact, rid, reqlen, &comp_len,
343 	    REMOTE_NDIS_QUERY_CMPLT);
344 	if (comp == NULL) {
345 		if_printf(sc->hn_ifp, "exec RNDIS query 0x%08x failed\n", oid);
346 		error = EIO;
347 		goto done;
348 	}
349 
350 	if (comp->rm_status != RNDIS_STATUS_SUCCESS) {
351 		if_printf(sc->hn_ifp, "RNDIS query 0x%08x failed: "
352 		    "status 0x%08x\n", oid, comp->rm_status);
353 		error = EIO;
354 		goto done;
355 	}
356 	if (comp->rm_infobuflen == 0 || comp->rm_infobufoffset == 0) {
357 		/* No output data! */
358 		if_printf(sc->hn_ifp, "RNDIS query 0x%08x, no data\n", oid);
359 		*odlen0 = 0;
360 		error = 0;
361 		goto done;
362 	}
363 
364 	/*
365 	 * Check output data length and offset.
366 	 */
367 	/* ofs is the offset from the beginning of comp. */
368 	ofs = RNDIS_QUERY_COMP_INFOBUFOFFSET_ABS(comp->rm_infobufoffset);
369 	if (ofs < sizeof(*comp) || ofs + comp->rm_infobuflen > comp_len) {
370 		if_printf(sc->hn_ifp, "RNDIS query invalid comp ib off/len, "
371 		    "%u/%u\n", comp->rm_infobufoffset, comp->rm_infobuflen);
372 		error = EINVAL;
373 		goto done;
374 	}
375 
376 	/*
377 	 * Save output data.
378 	 */
379 	if (comp->rm_infobuflen < odlen)
380 		odlen = comp->rm_infobuflen;
381 	memcpy(odata, ((const uint8_t *)comp) + ofs, odlen);
382 	*odlen0 = odlen;
383 
384 	error = 0;
385 done:
386 	vmbus_xact_put(xact);
387 	return (error);
388 }
389 
390 int
391 hn_rndis_query_rsscaps(struct hn_softc *sc, int *rxr_cnt0)
392 {
393 	struct ndis_rss_caps in, caps;
394 	size_t caps_len;
395 	int error, indsz, rxr_cnt, hash_fnidx;
396 	uint32_t hash_func = 0, hash_types = 0;
397 
398 	*rxr_cnt0 = 0;
399 
400 	if (sc->hn_ndis_ver < HN_NDIS_VERSION_6_20)
401 		return (EOPNOTSUPP);
402 
403 	memset(&in, 0, sizeof(in));
404 	in.ndis_hdr.ndis_type = NDIS_OBJTYPE_RSS_CAPS;
405 	in.ndis_hdr.ndis_rev = NDIS_RSS_CAPS_REV_2;
406 	in.ndis_hdr.ndis_size = NDIS_RSS_CAPS_SIZE;
407 
408 	caps_len = NDIS_RSS_CAPS_SIZE;
409 	error = hn_rndis_query2(sc, OID_GEN_RECEIVE_SCALE_CAPABILITIES,
410 	    &in, NDIS_RSS_CAPS_SIZE, &caps, &caps_len, NDIS_RSS_CAPS_SIZE_6_0);
411 	if (error)
412 		return (error);
413 
414 	/*
415 	 * Preliminary verification.
416 	 */
417 	if (caps.ndis_hdr.ndis_type != NDIS_OBJTYPE_RSS_CAPS) {
418 		if_printf(sc->hn_ifp, "invalid NDIS objtype 0x%02x\n",
419 		    caps.ndis_hdr.ndis_type);
420 		return (EINVAL);
421 	}
422 	if (caps.ndis_hdr.ndis_rev < NDIS_RSS_CAPS_REV_1) {
423 		if_printf(sc->hn_ifp, "invalid NDIS objrev 0x%02x\n",
424 		    caps.ndis_hdr.ndis_rev);
425 		return (EINVAL);
426 	}
427 	if (caps.ndis_hdr.ndis_size > caps_len) {
428 		if_printf(sc->hn_ifp, "invalid NDIS objsize %u, "
429 		    "data size %zu\n", caps.ndis_hdr.ndis_size, caps_len);
430 		return (EINVAL);
431 	} else if (caps.ndis_hdr.ndis_size < NDIS_RSS_CAPS_SIZE_6_0) {
432 		if_printf(sc->hn_ifp, "invalid NDIS objsize %u\n",
433 		    caps.ndis_hdr.ndis_size);
434 		return (EINVAL);
435 	}
436 
437 	/*
438 	 * Save information for later RSS configuration.
439 	 */
440 	if (caps.ndis_nrxr == 0) {
441 		if_printf(sc->hn_ifp, "0 RX rings!?\n");
442 		return (EINVAL);
443 	}
444 	if (bootverbose)
445 		if_printf(sc->hn_ifp, "%u RX rings\n", caps.ndis_nrxr);
446 	rxr_cnt = caps.ndis_nrxr;
447 
448 	if (caps.ndis_hdr.ndis_size == NDIS_RSS_CAPS_SIZE &&
449 	    caps.ndis_hdr.ndis_rev >= NDIS_RSS_CAPS_REV_2) {
450 		if (caps.ndis_nind > NDIS_HASH_INDCNT) {
451 			if_printf(sc->hn_ifp,
452 			    "too many RSS indirect table entries %u\n",
453 			    caps.ndis_nind);
454 			return (EOPNOTSUPP);
455 		}
456 		if (!powerof2(caps.ndis_nind)) {
457 			if_printf(sc->hn_ifp, "RSS indirect table size is not "
458 			    "power-of-2 %u\n", caps.ndis_nind);
459 		}
460 
461 		if (bootverbose) {
462 			if_printf(sc->hn_ifp, "RSS indirect table size %u\n",
463 			    caps.ndis_nind);
464 		}
465 		indsz = caps.ndis_nind;
466 	} else {
467 		indsz = NDIS_HASH_INDCNT;
468 	}
469 	if (indsz < rxr_cnt) {
470 		if_printf(sc->hn_ifp, "# of RX rings (%d) > "
471 		    "RSS indirect table size %d\n", rxr_cnt, indsz);
472 		rxr_cnt = indsz;
473 	}
474 
475 	/*
476 	 * NOTE:
477 	 * Toeplitz is at the lowest bit, and it is prefered; so ffs(),
478 	 * instead of fls(), is used here.
479 	 */
480 	hash_fnidx = ffs(caps.ndis_caps & NDIS_RSS_CAP_HASHFUNC_MASK);
481 	if (hash_fnidx == 0) {
482 		if_printf(sc->hn_ifp, "no hash functions, caps 0x%08x\n",
483 		    caps.ndis_caps);
484 		return (EOPNOTSUPP);
485 	}
486 	hash_func = 1 << (hash_fnidx - 1); /* ffs is 1-based */
487 
488 	if (caps.ndis_caps & NDIS_RSS_CAP_IPV4)
489 		hash_types |= NDIS_HASH_IPV4 | NDIS_HASH_TCP_IPV4;
490 	if (caps.ndis_caps & NDIS_RSS_CAP_IPV6)
491 		hash_types |= NDIS_HASH_IPV6 | NDIS_HASH_TCP_IPV6;
492 	if (caps.ndis_caps & NDIS_RSS_CAP_IPV6_EX)
493 		hash_types |= NDIS_HASH_IPV6_EX | NDIS_HASH_TCP_IPV6_EX;
494 	if (hash_types == 0) {
495 		if_printf(sc->hn_ifp, "no hash types, caps 0x%08x\n",
496 		    caps.ndis_caps);
497 		return (EOPNOTSUPP);
498 	}
499 
500 	/* Commit! */
501 	sc->hn_rss_ind_size = indsz;
502 	sc->hn_rss_hash = hash_func | hash_types;
503 	*rxr_cnt0 = rxr_cnt;
504 	return (0);
505 }
506 
507 static int
508 hn_rndis_set(struct hn_softc *sc, uint32_t oid, const void *data, size_t dlen)
509 {
510 	struct rndis_set_req *req;
511 	const struct rndis_set_comp *comp;
512 	struct vmbus_xact *xact;
513 	size_t reqlen, comp_len;
514 	uint32_t rid;
515 	int error;
516 
517 	KASSERT(dlen > 0, ("invalid dlen %zu", dlen));
518 
519 	reqlen = sizeof(*req) + dlen;
520 	xact = vmbus_xact_get(sc->hn_xact, reqlen);
521 	if (xact == NULL) {
522 		if_printf(sc->hn_ifp, "no xact for RNDIS set 0x%08x\n", oid);
523 		return (ENXIO);
524 	}
525 	rid = hn_rndis_rid(sc);
526 	req = vmbus_xact_req_data(xact);
527 	req->rm_type = REMOTE_NDIS_SET_MSG;
528 	req->rm_len = reqlen;
529 	req->rm_rid = rid;
530 	req->rm_oid = oid;
531 	req->rm_infobuflen = dlen;
532 	req->rm_infobufoffset = RNDIS_SET_REQ_INFOBUFOFFSET;
533 	/* Data immediately follows RNDIS set. */
534 	memcpy(req + 1, data, dlen);
535 
536 	comp_len = sizeof(*comp);
537 	comp = hn_rndis_xact_execute(sc, xact, rid, reqlen, &comp_len,
538 	    REMOTE_NDIS_SET_CMPLT);
539 	if (comp == NULL) {
540 		if_printf(sc->hn_ifp, "exec RNDIS set 0x%08x failed\n", oid);
541 		error = EIO;
542 		goto done;
543 	}
544 
545 	if (comp->rm_status != RNDIS_STATUS_SUCCESS) {
546 		if_printf(sc->hn_ifp, "RNDIS set 0x%08x failed: "
547 		    "status 0x%08x\n", oid, comp->rm_status);
548 		error = EIO;
549 		goto done;
550 	}
551 	error = 0;
552 done:
553 	vmbus_xact_put(xact);
554 	return (error);
555 }
556 
557 static int
558 hn_rndis_conf_offload(struct hn_softc *sc, int mtu)
559 {
560 	struct ndis_offload hwcaps;
561 	struct ndis_offload_params params;
562 	uint32_t caps = 0;
563 	size_t paramsz;
564 	int error, tso_maxsz, tso_minsg;
565 
566 	error = hn_rndis_query_hwcaps(sc, &hwcaps);
567 	if (error) {
568 		if_printf(sc->hn_ifp, "hwcaps query failed: %d\n", error);
569 		return (error);
570 	}
571 
572 	/* NOTE: 0 means "no change" */
573 	memset(&params, 0, sizeof(params));
574 
575 	params.ndis_hdr.ndis_type = NDIS_OBJTYPE_DEFAULT;
576 	if (sc->hn_ndis_ver < HN_NDIS_VERSION_6_30) {
577 		params.ndis_hdr.ndis_rev = NDIS_OFFLOAD_PARAMS_REV_2;
578 		paramsz = NDIS_OFFLOAD_PARAMS_SIZE_6_1;
579 	} else {
580 		params.ndis_hdr.ndis_rev = NDIS_OFFLOAD_PARAMS_REV_3;
581 		paramsz = NDIS_OFFLOAD_PARAMS_SIZE;
582 	}
583 	params.ndis_hdr.ndis_size = paramsz;
584 
585 	/*
586 	 * TSO4/TSO6 setup.
587 	 */
588 	tso_maxsz = IP_MAXPACKET;
589 	tso_minsg = 2;
590 	if (hwcaps.ndis_lsov2.ndis_ip4_encap & NDIS_OFFLOAD_ENCAP_8023) {
591 		caps |= HN_CAP_TSO4;
592 		params.ndis_lsov2_ip4 = NDIS_OFFLOAD_LSOV2_ON;
593 
594 		if (hwcaps.ndis_lsov2.ndis_ip4_maxsz < tso_maxsz)
595 			tso_maxsz = hwcaps.ndis_lsov2.ndis_ip4_maxsz;
596 		if (hwcaps.ndis_lsov2.ndis_ip4_minsg > tso_minsg)
597 			tso_minsg = hwcaps.ndis_lsov2.ndis_ip4_minsg;
598 	}
599 	if ((hwcaps.ndis_lsov2.ndis_ip6_encap & NDIS_OFFLOAD_ENCAP_8023) &&
600 	    (hwcaps.ndis_lsov2.ndis_ip6_opts & HN_NDIS_LSOV2_CAP_IP6) ==
601 	    HN_NDIS_LSOV2_CAP_IP6) {
602 #ifdef notyet
603 		caps |= HN_CAP_TSO6;
604 		params.ndis_lsov2_ip6 = NDIS_OFFLOAD_LSOV2_ON;
605 
606 		if (hwcaps.ndis_lsov2.ndis_ip6_maxsz < tso_maxsz)
607 			tso_maxsz = hwcaps.ndis_lsov2.ndis_ip6_maxsz;
608 		if (hwcaps.ndis_lsov2.ndis_ip6_minsg > tso_minsg)
609 			tso_minsg = hwcaps.ndis_lsov2.ndis_ip6_minsg;
610 #endif
611 	}
612 	sc->hn_ndis_tso_szmax = 0;
613 	sc->hn_ndis_tso_sgmin = 0;
614 	if (caps & (HN_CAP_TSO4 | HN_CAP_TSO6)) {
615 		KASSERT(tso_maxsz <= IP_MAXPACKET,
616 		    ("invalid NDIS TSO maxsz %d", tso_maxsz));
617 		KASSERT(tso_minsg >= 2,
618 		    ("invalid NDIS TSO minsg %d", tso_minsg));
619 		if (tso_maxsz < tso_minsg * mtu) {
620 			if_printf(sc->hn_ifp, "invalid NDIS TSO config: "
621 			    "maxsz %d, minsg %d, mtu %d; "
622 			    "disable TSO4 and TSO6\n",
623 			    tso_maxsz, tso_minsg, mtu);
624 			caps &= ~(HN_CAP_TSO4 | HN_CAP_TSO6);
625 			params.ndis_lsov2_ip4 = NDIS_OFFLOAD_LSOV2_OFF;
626 			params.ndis_lsov2_ip6 = NDIS_OFFLOAD_LSOV2_OFF;
627 		} else {
628 			sc->hn_ndis_tso_szmax = tso_maxsz;
629 			sc->hn_ndis_tso_sgmin = tso_minsg;
630 			if (bootverbose) {
631 				if_printf(sc->hn_ifp, "NDIS TSO "
632 				    "szmax %d sgmin %d\n",
633 				    sc->hn_ndis_tso_szmax,
634 				    sc->hn_ndis_tso_sgmin);
635 			}
636 		}
637 	}
638 
639 	/* IPv4 checksum */
640 	if ((hwcaps.ndis_csum.ndis_ip4_txcsum & HN_NDIS_TXCSUM_CAP_IP4) ==
641 	    HN_NDIS_TXCSUM_CAP_IP4) {
642 		caps |= HN_CAP_IPCS;
643 		params.ndis_ip4csum = NDIS_OFFLOAD_PARAM_TX;
644 	}
645 	if (hwcaps.ndis_csum.ndis_ip4_rxcsum & NDIS_RXCSUM_CAP_IP4) {
646 		if (params.ndis_ip4csum == NDIS_OFFLOAD_PARAM_TX)
647 			params.ndis_ip4csum = NDIS_OFFLOAD_PARAM_TXRX;
648 		else
649 			params.ndis_ip4csum = NDIS_OFFLOAD_PARAM_RX;
650 	}
651 
652 	/* TCP4 checksum */
653 	if ((hwcaps.ndis_csum.ndis_ip4_txcsum & HN_NDIS_TXCSUM_CAP_TCP4) ==
654 	    HN_NDIS_TXCSUM_CAP_TCP4) {
655 		caps |= HN_CAP_TCP4CS;
656 		params.ndis_tcp4csum = NDIS_OFFLOAD_PARAM_TX;
657 	}
658 	if (hwcaps.ndis_csum.ndis_ip4_rxcsum & NDIS_RXCSUM_CAP_TCP4) {
659 		if (params.ndis_tcp4csum == NDIS_OFFLOAD_PARAM_TX)
660 			params.ndis_tcp4csum = NDIS_OFFLOAD_PARAM_TXRX;
661 		else
662 			params.ndis_tcp4csum = NDIS_OFFLOAD_PARAM_RX;
663 	}
664 
665 	/* UDP4 checksum */
666 	if (hwcaps.ndis_csum.ndis_ip4_txcsum & NDIS_TXCSUM_CAP_UDP4) {
667 		caps |= HN_CAP_UDP4CS;
668 		params.ndis_udp4csum = NDIS_OFFLOAD_PARAM_TX;
669 	}
670 	if (hwcaps.ndis_csum.ndis_ip4_rxcsum & NDIS_RXCSUM_CAP_UDP4) {
671 		if (params.ndis_udp4csum == NDIS_OFFLOAD_PARAM_TX)
672 			params.ndis_udp4csum = NDIS_OFFLOAD_PARAM_TXRX;
673 		else
674 			params.ndis_udp4csum = NDIS_OFFLOAD_PARAM_RX;
675 	}
676 
677 	/* TCP6 checksum */
678 	if ((hwcaps.ndis_csum.ndis_ip6_txcsum & HN_NDIS_TXCSUM_CAP_TCP6) ==
679 	    HN_NDIS_TXCSUM_CAP_TCP6) {
680 		caps |= HN_CAP_TCP6CS;
681 		params.ndis_tcp6csum = NDIS_OFFLOAD_PARAM_TX;
682 	}
683 	if (hwcaps.ndis_csum.ndis_ip6_rxcsum & NDIS_RXCSUM_CAP_TCP6) {
684 		if (params.ndis_tcp6csum == NDIS_OFFLOAD_PARAM_TX)
685 			params.ndis_tcp6csum = NDIS_OFFLOAD_PARAM_TXRX;
686 		else
687 			params.ndis_tcp6csum = NDIS_OFFLOAD_PARAM_RX;
688 	}
689 
690 	/* UDP6 checksum */
691 	if ((hwcaps.ndis_csum.ndis_ip6_txcsum & HN_NDIS_TXCSUM_CAP_UDP6) ==
692 	    HN_NDIS_TXCSUM_CAP_UDP6) {
693 		caps |= HN_CAP_UDP6CS;
694 		params.ndis_udp6csum = NDIS_OFFLOAD_PARAM_TX;
695 	}
696 	if (hwcaps.ndis_csum.ndis_ip6_rxcsum & NDIS_RXCSUM_CAP_UDP6) {
697 		if (params.ndis_udp6csum == NDIS_OFFLOAD_PARAM_TX)
698 			params.ndis_udp6csum = NDIS_OFFLOAD_PARAM_TXRX;
699 		else
700 			params.ndis_udp6csum = NDIS_OFFLOAD_PARAM_RX;
701 	}
702 
703 	if (bootverbose) {
704 		if_printf(sc->hn_ifp, "offload csum: "
705 		    "ip4 %u, tcp4 %u, udp4 %u, tcp6 %u, udp6 %u\n",
706 		    params.ndis_ip4csum,
707 		    params.ndis_tcp4csum,
708 		    params.ndis_udp4csum,
709 		    params.ndis_tcp6csum,
710 		    params.ndis_udp6csum);
711 		if_printf(sc->hn_ifp, "offload lsov2: ip4 %u, ip6 %u\n",
712 		    params.ndis_lsov2_ip4,
713 		    params.ndis_lsov2_ip6);
714 	}
715 
716 	error = hn_rndis_set(sc, OID_TCP_OFFLOAD_PARAMETERS, &params, paramsz);
717 	if (error) {
718 		if_printf(sc->hn_ifp, "offload config failed: %d\n", error);
719 		return (error);
720 	}
721 
722 	if (bootverbose)
723 		if_printf(sc->hn_ifp, "offload config done\n");
724 	sc->hn_caps |= caps;
725 	return (0);
726 }
727 
728 int
729 hn_rndis_conf_rss(struct hn_softc *sc, uint16_t flags)
730 {
731 	struct ndis_rssprm_toeplitz *rss = &sc->hn_rss;
732 	struct ndis_rss_params *prm = &rss->rss_params;
733 	int error, rss_size;
734 
735 	/*
736 	 * Only NDIS 6.20+ is supported:
737 	 * We only support 4bytes element in indirect table, which has been
738 	 * adopted since NDIS 6.20.
739 	 */
740 	KASSERT(sc->hn_ndis_ver >= HN_NDIS_VERSION_6_20,
741 	    ("NDIS 6.20+ is required, NDIS version 0x%08x", sc->hn_ndis_ver));
742 
743 	/* XXX only one can be specified through, popcnt? */
744 	KASSERT((sc->hn_rss_hash & NDIS_HASH_FUNCTION_MASK), ("no hash func"));
745 	KASSERT((sc->hn_rss_hash & NDIS_HASH_TYPE_MASK), ("no hash types"));
746 	KASSERT(sc->hn_rss_ind_size > 0, ("no indirect table size"));
747 
748 	if (bootverbose) {
749 		if_printf(sc->hn_ifp, "RSS indirect table size %d, "
750 		    "hash 0x%08x\n", sc->hn_rss_ind_size, sc->hn_rss_hash);
751 	}
752 
753 	/*
754 	 * NOTE:
755 	 * DO NOT whack rss_key and rss_ind, which are setup by the caller.
756 	 */
757 	memset(prm, 0, sizeof(*prm));
758 	rss_size = NDIS_RSSPRM_TOEPLITZ_SIZE(sc->hn_rss_ind_size);
759 
760 	prm->ndis_hdr.ndis_type = NDIS_OBJTYPE_RSS_PARAMS;
761 	prm->ndis_hdr.ndis_rev = NDIS_RSS_PARAMS_REV_2;
762 	prm->ndis_hdr.ndis_size = rss_size;
763 	prm->ndis_flags = flags;
764 	prm->ndis_hash = sc->hn_rss_hash;
765 	prm->ndis_indsize = sizeof(rss->rss_ind[0]) * sc->hn_rss_ind_size;
766 	prm->ndis_indoffset =
767 	    __offsetof(struct ndis_rssprm_toeplitz, rss_ind[0]);
768 	prm->ndis_keysize = sizeof(rss->rss_key);
769 	prm->ndis_keyoffset =
770 	    __offsetof(struct ndis_rssprm_toeplitz, rss_key[0]);
771 
772 	error = hn_rndis_set(sc, OID_GEN_RECEIVE_SCALE_PARAMETERS,
773 	    rss, rss_size);
774 	if (error) {
775 		if_printf(sc->hn_ifp, "RSS config failed: %d\n", error);
776 	} else {
777 		if (bootverbose)
778 			if_printf(sc->hn_ifp, "RSS config done\n");
779 	}
780 	return (error);
781 }
782 
783 int
784 hn_rndis_set_rxfilter(struct hn_softc *sc, uint32_t filter)
785 {
786 	int error;
787 
788 	error = hn_rndis_set(sc, OID_GEN_CURRENT_PACKET_FILTER,
789 	    &filter, sizeof(filter));
790 	if (error) {
791 		if_printf(sc->hn_ifp, "set RX filter 0x%08x failed: %d\n",
792 		    filter, error);
793 	} else {
794 		if (bootverbose) {
795 			if_printf(sc->hn_ifp, "set RX filter 0x%08x done\n",
796 			    filter);
797 		}
798 	}
799 	return (error);
800 }
801 
802 static int
803 hn_rndis_init(struct hn_softc *sc)
804 {
805 	struct rndis_init_req *req;
806 	const struct rndis_init_comp *comp;
807 	struct vmbus_xact *xact;
808 	size_t comp_len;
809 	uint32_t rid;
810 	int error;
811 
812 	xact = vmbus_xact_get(sc->hn_xact, sizeof(*req));
813 	if (xact == NULL) {
814 		if_printf(sc->hn_ifp, "no xact for RNDIS init\n");
815 		return (ENXIO);
816 	}
817 	rid = hn_rndis_rid(sc);
818 	req = vmbus_xact_req_data(xact);
819 	req->rm_type = REMOTE_NDIS_INITIALIZE_MSG;
820 	req->rm_len = sizeof(*req);
821 	req->rm_rid = rid;
822 	req->rm_ver_major = RNDIS_VERSION_MAJOR;
823 	req->rm_ver_minor = RNDIS_VERSION_MINOR;
824 	req->rm_max_xfersz = HN_RNDIS_XFER_SIZE;
825 
826 	comp_len = RNDIS_INIT_COMP_SIZE_MIN;
827 	comp = hn_rndis_xact_execute(sc, xact, rid, sizeof(*req), &comp_len,
828 	    REMOTE_NDIS_INITIALIZE_CMPLT);
829 	if (comp == NULL) {
830 		if_printf(sc->hn_ifp, "exec RNDIS init failed\n");
831 		error = EIO;
832 		goto done;
833 	}
834 
835 	if (comp->rm_status != RNDIS_STATUS_SUCCESS) {
836 		if_printf(sc->hn_ifp, "RNDIS init failed: status 0x%08x\n",
837 		    comp->rm_status);
838 		error = EIO;
839 		goto done;
840 	}
841 	if (bootverbose) {
842 		if_printf(sc->hn_ifp, "RNDIS ver %u.%u, pktsz %u, pktcnt %u, "
843 		    "align %u\n", comp->rm_ver_major, comp->rm_ver_minor,
844 		    comp->rm_pktmaxsz, comp->rm_pktmaxcnt,
845 		    1U << comp->rm_align);
846 	}
847 	error = 0;
848 done:
849 	vmbus_xact_put(xact);
850 	return (error);
851 }
852 
853 static int
854 hn_rndis_halt(struct hn_softc *sc)
855 {
856 	struct vmbus_xact *xact;
857 	struct rndis_halt_req *halt;
858 	struct hn_nvs_sendctx sndc;
859 	size_t comp_len;
860 
861 	xact = vmbus_xact_get(sc->hn_xact, sizeof(*halt));
862 	if (xact == NULL) {
863 		if_printf(sc->hn_ifp, "no xact for RNDIS halt\n");
864 		return (ENXIO);
865 	}
866 	halt = vmbus_xact_req_data(xact);
867 	halt->rm_type = REMOTE_NDIS_HALT_MSG;
868 	halt->rm_len = sizeof(*halt);
869 	halt->rm_rid = hn_rndis_rid(sc);
870 
871 	/* No RNDIS completion; rely on NVS message send completion */
872 	hn_nvs_sendctx_init(&sndc, hn_nvs_sent_xact, xact);
873 	hn_rndis_xact_exec1(sc, xact, sizeof(*halt), &sndc, &comp_len);
874 
875 	vmbus_xact_put(xact);
876 	if (bootverbose)
877 		if_printf(sc->hn_ifp, "RNDIS halt done\n");
878 	return (0);
879 }
880 
881 static int
882 hn_rndis_query_hwcaps(struct hn_softc *sc, struct ndis_offload *caps)
883 {
884 	struct ndis_offload in;
885 	size_t caps_len, size;
886 	int error;
887 
888 	memset(&in, 0, sizeof(in));
889 	in.ndis_hdr.ndis_type = NDIS_OBJTYPE_OFFLOAD;
890 	if (sc->hn_ndis_ver >= HN_NDIS_VERSION_6_30) {
891 		in.ndis_hdr.ndis_rev = NDIS_OFFLOAD_REV_3;
892 		size = NDIS_OFFLOAD_SIZE;
893 	} else if (sc->hn_ndis_ver >= HN_NDIS_VERSION_6_1) {
894 		in.ndis_hdr.ndis_rev = NDIS_OFFLOAD_REV_2;
895 		size = NDIS_OFFLOAD_SIZE_6_1;
896 	} else {
897 		in.ndis_hdr.ndis_rev = NDIS_OFFLOAD_REV_1;
898 		size = NDIS_OFFLOAD_SIZE_6_0;
899 	}
900 	in.ndis_hdr.ndis_size = size;
901 
902 	caps_len = NDIS_OFFLOAD_SIZE;
903 	error = hn_rndis_query2(sc, OID_TCP_OFFLOAD_HARDWARE_CAPABILITIES,
904 	    &in, size, caps, &caps_len, NDIS_OFFLOAD_SIZE_6_0);
905 	if (error)
906 		return (error);
907 
908 	/*
909 	 * Preliminary verification.
910 	 */
911 	if (caps->ndis_hdr.ndis_type != NDIS_OBJTYPE_OFFLOAD) {
912 		if_printf(sc->hn_ifp, "invalid NDIS objtype 0x%02x\n",
913 		    caps->ndis_hdr.ndis_type);
914 		return (EINVAL);
915 	}
916 	if (caps->ndis_hdr.ndis_rev < NDIS_OFFLOAD_REV_1) {
917 		if_printf(sc->hn_ifp, "invalid NDIS objrev 0x%02x\n",
918 		    caps->ndis_hdr.ndis_rev);
919 		return (EINVAL);
920 	}
921 	if (caps->ndis_hdr.ndis_size > caps_len) {
922 		if_printf(sc->hn_ifp, "invalid NDIS objsize %u, "
923 		    "data size %zu\n", caps->ndis_hdr.ndis_size, caps_len);
924 		return (EINVAL);
925 	} else if (caps->ndis_hdr.ndis_size < NDIS_OFFLOAD_SIZE_6_0) {
926 		if_printf(sc->hn_ifp, "invalid NDIS objsize %u\n",
927 		    caps->ndis_hdr.ndis_size);
928 		return (EINVAL);
929 	}
930 
931 	if (bootverbose) {
932 		/*
933 		 * NOTE:
934 		 * caps->ndis_hdr.ndis_size MUST be checked before accessing
935 		 * NDIS 6.1+ specific fields.
936 		 */
937 		if_printf(sc->hn_ifp, "hwcaps rev %u\n",
938 		    caps->ndis_hdr.ndis_rev);
939 
940 		if_printf(sc->hn_ifp, "hwcaps csum: "
941 		    "ip4 tx 0x%x/0x%x rx 0x%x/0x%x, "
942 		    "ip6 tx 0x%x/0x%x rx 0x%x/0x%x\n",
943 		    caps->ndis_csum.ndis_ip4_txcsum,
944 		    caps->ndis_csum.ndis_ip4_txenc,
945 		    caps->ndis_csum.ndis_ip4_rxcsum,
946 		    caps->ndis_csum.ndis_ip4_rxenc,
947 		    caps->ndis_csum.ndis_ip6_txcsum,
948 		    caps->ndis_csum.ndis_ip6_txenc,
949 		    caps->ndis_csum.ndis_ip6_rxcsum,
950 		    caps->ndis_csum.ndis_ip6_rxenc);
951 		if_printf(sc->hn_ifp, "hwcaps lsov2: "
952 		    "ip4 maxsz %u minsg %u encap 0x%x, "
953 		    "ip6 maxsz %u minsg %u encap 0x%x opts 0x%x\n",
954 		    caps->ndis_lsov2.ndis_ip4_maxsz,
955 		    caps->ndis_lsov2.ndis_ip4_minsg,
956 		    caps->ndis_lsov2.ndis_ip4_encap,
957 		    caps->ndis_lsov2.ndis_ip6_maxsz,
958 		    caps->ndis_lsov2.ndis_ip6_minsg,
959 		    caps->ndis_lsov2.ndis_ip6_encap,
960 		    caps->ndis_lsov2.ndis_ip6_opts);
961 	}
962 	return (0);
963 }
964 
965 int
966 hn_rndis_attach(struct hn_softc *sc, int mtu)
967 {
968 	int error;
969 
970 	/*
971 	 * Initialize RNDIS.
972 	 */
973 	error = hn_rndis_init(sc);
974 	if (error)
975 		return (error);
976 
977 	/*
978 	 * Configure NDIS offload settings.
979 	 * XXX no offloading, if error happened?
980 	 */
981 	hn_rndis_conf_offload(sc, mtu);
982 	return (0);
983 }
984 
985 void
986 hn_rndis_detach(struct hn_softc *sc)
987 {
988 
989 	/* Halt the RNDIS. */
990 	hn_rndis_halt(sc);
991 }
992