xref: /linux/drivers/infiniband/ulp/opa_vnic/opa_vnic_encap.c (revision 3eb66e91a25497065c5322b1268cbc3953642227)
1 /*
2  * Copyright(c) 2017 Intel Corporation.
3  *
4  * This file is provided under a dual BSD/GPLv2 license.  When using or
5  * redistributing this file, you may do so under either license.
6  *
7  * GPL LICENSE SUMMARY
8  *
9  * This program is free software; you can redistribute it and/or modify
10  * it under the terms of version 2 of the GNU General Public License as
11  * published by the Free Software Foundation.
12  *
13  * This program is distributed in the hope that it will be useful, but
14  * WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
16  * General Public License for more details.
17  *
18  * BSD LICENSE
19  *
20  * Redistribution and use in source and binary forms, with or without
21  * modification, are permitted provided that the following conditions
22  * are met:
23  *
24  *  - Redistributions of source code must retain the above copyright
25  *    notice, this list of conditions and the following disclaimer.
26  *  - Redistributions in binary form must reproduce the above copyright
27  *    notice, this list of conditions and the following disclaimer in
28  *    the documentation and/or other materials provided with the
29  *    distribution.
30  *  - Neither the name of Intel Corporation nor the names of its
31  *    contributors may be used to endorse or promote products derived
32  *    from this software without specific prior written permission.
33  *
34  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
35  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
36  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
37  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
38  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
39  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
40  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
41  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
42  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
43  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
44  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
45  *
46  */
47 
48 /*
49  * This file contains OPA VNIC encapsulation/decapsulation function.
50  */
51 
52 #include <linux/if_ether.h>
53 #include <linux/if_vlan.h>
54 
55 #include "opa_vnic_internal.h"
56 
57 /* OPA 16B Header fields */
58 #define OPA_16B_LID_MASK        0xFFFFFull
59 #define OPA_16B_SLID_HIGH_SHFT  8
60 #define OPA_16B_SLID_MASK       0xF00ull
61 #define OPA_16B_DLID_MASK       0xF000ull
62 #define OPA_16B_DLID_HIGH_SHFT  12
63 #define OPA_16B_LEN_SHFT        20
64 #define OPA_16B_SC_SHFT         20
65 #define OPA_16B_RC_SHFT         25
66 #define OPA_16B_PKEY_SHFT       16
67 
68 #define OPA_VNIC_L4_HDR_SHFT    16
69 
70 /* L2+L4 hdr len is 20 bytes (5 quad words) */
71 #define OPA_VNIC_HDR_QW_LEN   5
72 
opa_vnic_make_header(u8 * hdr,u32 slid,u32 dlid,u16 len,u16 pkey,u16 entropy,u8 sc,u8 rc,u8 l4_type,u16 l4_hdr)73 static inline void opa_vnic_make_header(u8 *hdr, u32 slid, u32 dlid, u16 len,
74 					u16 pkey, u16 entropy, u8 sc, u8 rc,
75 					u8 l4_type, u16 l4_hdr)
76 {
77 	/* h[1]: LT=1, 16B L2=10 */
78 	u32 h[OPA_VNIC_HDR_QW_LEN] = {0, 0xc0000000, 0, 0, 0};
79 
80 	h[2] = l4_type;
81 	h[3] = entropy;
82 	h[4] = l4_hdr << OPA_VNIC_L4_HDR_SHFT;
83 
84 	/* Extract and set 4 upper bits and 20 lower bits of the lids */
85 	h[0] |= (slid & OPA_16B_LID_MASK);
86 	h[2] |= ((slid >> (20 - OPA_16B_SLID_HIGH_SHFT)) & OPA_16B_SLID_MASK);
87 
88 	h[1] |= (dlid & OPA_16B_LID_MASK);
89 	h[2] |= ((dlid >> (20 - OPA_16B_DLID_HIGH_SHFT)) & OPA_16B_DLID_MASK);
90 
91 	h[0] |= (len << OPA_16B_LEN_SHFT);
92 	h[1] |= (rc << OPA_16B_RC_SHFT);
93 	h[1] |= (sc << OPA_16B_SC_SHFT);
94 	h[2] |= ((u32)pkey << OPA_16B_PKEY_SHFT);
95 
96 	memcpy(hdr, h, OPA_VNIC_HDR_LEN);
97 }
98 
99 /*
100  * Using a simple hash table for mac table implementation with the last octet
101  * of mac address as a key.
102  */
opa_vnic_free_mac_tbl(struct hlist_head * mactbl)103 static void opa_vnic_free_mac_tbl(struct hlist_head *mactbl)
104 {
105 	struct opa_vnic_mac_tbl_node *node;
106 	struct hlist_node *tmp;
107 	int bkt;
108 
109 	if (!mactbl)
110 		return;
111 
112 	vnic_hash_for_each_safe(mactbl, bkt, tmp, node, hlist) {
113 		hash_del(&node->hlist);
114 		kfree(node);
115 	}
116 	kfree(mactbl);
117 }
118 
opa_vnic_alloc_mac_tbl(void)119 static struct hlist_head *opa_vnic_alloc_mac_tbl(void)
120 {
121 	u32 size = sizeof(struct hlist_head) * OPA_VNIC_MAC_TBL_SIZE;
122 	struct hlist_head *mactbl;
123 
124 	mactbl = kzalloc(size, GFP_KERNEL);
125 	if (!mactbl)
126 		return ERR_PTR(-ENOMEM);
127 
128 	vnic_hash_init(mactbl);
129 	return mactbl;
130 }
131 
132 /* opa_vnic_release_mac_tbl - empty and free the mac table */
opa_vnic_release_mac_tbl(struct opa_vnic_adapter * adapter)133 void opa_vnic_release_mac_tbl(struct opa_vnic_adapter *adapter)
134 {
135 	struct hlist_head *mactbl;
136 
137 	mutex_lock(&adapter->mactbl_lock);
138 	mactbl = rcu_access_pointer(adapter->mactbl);
139 	rcu_assign_pointer(adapter->mactbl, NULL);
140 	synchronize_rcu();
141 	opa_vnic_free_mac_tbl(mactbl);
142 	adapter->info.vport.mac_tbl_digest = 0;
143 	mutex_unlock(&adapter->mactbl_lock);
144 }
145 
146 /*
147  * opa_vnic_query_mac_tbl - query the mac table for a section
148  *
149  * This function implements query of specific function of the mac table.
150  * The function also expects the requested range to be valid.
151  */
opa_vnic_query_mac_tbl(struct opa_vnic_adapter * adapter,struct opa_veswport_mactable * tbl)152 void opa_vnic_query_mac_tbl(struct opa_vnic_adapter *adapter,
153 			    struct opa_veswport_mactable *tbl)
154 {
155 	struct opa_vnic_mac_tbl_node *node;
156 	struct hlist_head *mactbl;
157 	int bkt;
158 	u16 loffset, lnum_entries;
159 
160 	rcu_read_lock();
161 	mactbl = rcu_dereference(adapter->mactbl);
162 	if (!mactbl)
163 		goto get_mac_done;
164 
165 	loffset = be16_to_cpu(tbl->offset);
166 	lnum_entries = be16_to_cpu(tbl->num_entries);
167 
168 	vnic_hash_for_each(mactbl, bkt, node, hlist) {
169 		struct __opa_vnic_mactable_entry *nentry = &node->entry;
170 		struct opa_veswport_mactable_entry *entry;
171 
172 		if ((node->index < loffset) ||
173 		    (node->index >= (loffset + lnum_entries)))
174 			continue;
175 
176 		/* populate entry in the tbl corresponding to the index */
177 		entry = &tbl->tbl_entries[node->index - loffset];
178 		memcpy(entry->mac_addr, nentry->mac_addr,
179 		       ARRAY_SIZE(entry->mac_addr));
180 		memcpy(entry->mac_addr_mask, nentry->mac_addr_mask,
181 		       ARRAY_SIZE(entry->mac_addr_mask));
182 		entry->dlid_sd = cpu_to_be32(nentry->dlid_sd);
183 	}
184 	tbl->mac_tbl_digest = cpu_to_be32(adapter->info.vport.mac_tbl_digest);
185 get_mac_done:
186 	rcu_read_unlock();
187 }
188 
189 /*
190  * opa_vnic_update_mac_tbl - update mac table section
191  *
192  * This function updates the specified section of the mac table.
193  * The procedure includes following steps.
194  *  - Allocate a new mac (hash) table.
195  *  - Add the specified entries to the new table.
196  *    (except the ones that are requested to be deleted).
197  *  - Add all the other entries from the old mac table.
198  *  - If there is a failure, free the new table and return.
199  *  - Switch to the new table.
200  *  - Free the old table and return.
201  *
202  * The function also expects the requested range to be valid.
203  */
opa_vnic_update_mac_tbl(struct opa_vnic_adapter * adapter,struct opa_veswport_mactable * tbl)204 int opa_vnic_update_mac_tbl(struct opa_vnic_adapter *adapter,
205 			    struct opa_veswport_mactable *tbl)
206 {
207 	struct opa_vnic_mac_tbl_node *node, *new_node;
208 	struct hlist_head *new_mactbl, *old_mactbl;
209 	int i, bkt, rc = 0;
210 	u8 key;
211 	u16 loffset, lnum_entries;
212 
213 	mutex_lock(&adapter->mactbl_lock);
214 	/* allocate new mac table */
215 	new_mactbl = opa_vnic_alloc_mac_tbl();
216 	if (IS_ERR(new_mactbl)) {
217 		mutex_unlock(&adapter->mactbl_lock);
218 		return PTR_ERR(new_mactbl);
219 	}
220 
221 	loffset = be16_to_cpu(tbl->offset);
222 	lnum_entries = be16_to_cpu(tbl->num_entries);
223 
224 	/* add updated entries to the new mac table */
225 	for (i = 0; i < lnum_entries; i++) {
226 		struct __opa_vnic_mactable_entry *nentry;
227 		struct opa_veswport_mactable_entry *entry =
228 							&tbl->tbl_entries[i];
229 		u8 *mac_addr = entry->mac_addr;
230 		u8 empty_mac[ETH_ALEN] = { 0 };
231 
232 		v_dbg("new mac entry %4d: %02x:%02x:%02x:%02x:%02x:%02x %x\n",
233 		      loffset + i, mac_addr[0], mac_addr[1], mac_addr[2],
234 		      mac_addr[3], mac_addr[4], mac_addr[5],
235 		      entry->dlid_sd);
236 
237 		/* if the entry is being removed, do not add it */
238 		if (!memcmp(mac_addr, empty_mac, ARRAY_SIZE(empty_mac)))
239 			continue;
240 
241 		node = kzalloc(sizeof(*node), GFP_KERNEL);
242 		if (!node) {
243 			rc = -ENOMEM;
244 			goto updt_done;
245 		}
246 
247 		node->index = loffset + i;
248 		nentry = &node->entry;
249 		memcpy(nentry->mac_addr, entry->mac_addr,
250 		       ARRAY_SIZE(nentry->mac_addr));
251 		memcpy(nentry->mac_addr_mask, entry->mac_addr_mask,
252 		       ARRAY_SIZE(nentry->mac_addr_mask));
253 		nentry->dlid_sd = be32_to_cpu(entry->dlid_sd);
254 		key = node->entry.mac_addr[OPA_VNIC_MAC_HASH_IDX];
255 		vnic_hash_add(new_mactbl, &node->hlist, key);
256 	}
257 
258 	/* add other entries from current mac table to new mac table */
259 	old_mactbl = rcu_access_pointer(adapter->mactbl);
260 	if (!old_mactbl)
261 		goto switch_tbl;
262 
263 	vnic_hash_for_each(old_mactbl, bkt, node, hlist) {
264 		if ((node->index >= loffset) &&
265 		    (node->index < (loffset + lnum_entries)))
266 			continue;
267 
268 		new_node = kzalloc(sizeof(*new_node), GFP_KERNEL);
269 		if (!new_node) {
270 			rc = -ENOMEM;
271 			goto updt_done;
272 		}
273 
274 		new_node->index = node->index;
275 		memcpy(&new_node->entry, &node->entry, sizeof(node->entry));
276 		key = new_node->entry.mac_addr[OPA_VNIC_MAC_HASH_IDX];
277 		vnic_hash_add(new_mactbl, &new_node->hlist, key);
278 	}
279 
280 switch_tbl:
281 	/* switch to new table */
282 	rcu_assign_pointer(adapter->mactbl, new_mactbl);
283 	synchronize_rcu();
284 
285 	adapter->info.vport.mac_tbl_digest = be32_to_cpu(tbl->mac_tbl_digest);
286 updt_done:
287 	/* upon failure, free the new table; otherwise, free the old table */
288 	if (rc)
289 		opa_vnic_free_mac_tbl(new_mactbl);
290 	else
291 		opa_vnic_free_mac_tbl(old_mactbl);
292 
293 	mutex_unlock(&adapter->mactbl_lock);
294 	return rc;
295 }
296 
297 /* opa_vnic_chk_mac_tbl - check mac table for dlid */
opa_vnic_chk_mac_tbl(struct opa_vnic_adapter * adapter,struct ethhdr * mac_hdr)298 static uint32_t opa_vnic_chk_mac_tbl(struct opa_vnic_adapter *adapter,
299 				     struct ethhdr *mac_hdr)
300 {
301 	struct opa_vnic_mac_tbl_node *node;
302 	struct hlist_head *mactbl;
303 	u32 dlid = 0;
304 	u8 key;
305 
306 	rcu_read_lock();
307 	mactbl = rcu_dereference(adapter->mactbl);
308 	if (unlikely(!mactbl))
309 		goto chk_done;
310 
311 	key = mac_hdr->h_dest[OPA_VNIC_MAC_HASH_IDX];
312 	vnic_hash_for_each_possible(mactbl, node, hlist, key) {
313 		struct __opa_vnic_mactable_entry *entry = &node->entry;
314 
315 		/* if related to source mac, skip */
316 		if (unlikely(OPA_VNIC_DLID_SD_IS_SRC_MAC(entry->dlid_sd)))
317 			continue;
318 
319 		if (!memcmp(node->entry.mac_addr, mac_hdr->h_dest,
320 			    ARRAY_SIZE(node->entry.mac_addr))) {
321 			/* mac address found */
322 			dlid = OPA_VNIC_DLID_SD_GET_DLID(node->entry.dlid_sd);
323 			break;
324 		}
325 	}
326 
327 chk_done:
328 	rcu_read_unlock();
329 	return dlid;
330 }
331 
332 /* opa_vnic_get_dlid - find and return the DLID */
opa_vnic_get_dlid(struct opa_vnic_adapter * adapter,struct sk_buff * skb,u8 def_port)333 static uint32_t opa_vnic_get_dlid(struct opa_vnic_adapter *adapter,
334 				  struct sk_buff *skb, u8 def_port)
335 {
336 	struct __opa_veswport_info *info = &adapter->info;
337 	struct ethhdr *mac_hdr = (struct ethhdr *)skb_mac_header(skb);
338 	u32 dlid;
339 
340 	dlid = opa_vnic_chk_mac_tbl(adapter, mac_hdr);
341 	if (dlid)
342 		return dlid;
343 
344 	if (is_multicast_ether_addr(mac_hdr->h_dest)) {
345 		dlid = info->vesw.u_mcast_dlid;
346 	} else {
347 		if (is_local_ether_addr(mac_hdr->h_dest)) {
348 			dlid = ((uint32_t)mac_hdr->h_dest[5] << 16) |
349 				((uint32_t)mac_hdr->h_dest[4] << 8)  |
350 				mac_hdr->h_dest[3];
351 			if (unlikely(!dlid))
352 				v_warn("Null dlid in MAC address\n");
353 		} else if (def_port != OPA_VNIC_INVALID_PORT) {
354 			if (def_port < OPA_VESW_MAX_NUM_DEF_PORT)
355 				dlid = info->vesw.u_ucast_dlid[def_port];
356 		}
357 	}
358 
359 	return dlid;
360 }
361 
362 /* opa_vnic_get_sc - return the service class */
opa_vnic_get_sc(struct __opa_veswport_info * info,struct sk_buff * skb)363 static u8 opa_vnic_get_sc(struct __opa_veswport_info *info,
364 			  struct sk_buff *skb)
365 {
366 	struct ethhdr *mac_hdr = (struct ethhdr *)skb_mac_header(skb);
367 	u16 vlan_tci;
368 	u8 sc;
369 
370 	if (!__vlan_get_tag(skb, &vlan_tci)) {
371 		u8 pcp = OPA_VNIC_VLAN_PCP(vlan_tci);
372 
373 		if (is_multicast_ether_addr(mac_hdr->h_dest))
374 			sc = info->vport.pcp_to_sc_mc[pcp];
375 		else
376 			sc = info->vport.pcp_to_sc_uc[pcp];
377 	} else {
378 		if (is_multicast_ether_addr(mac_hdr->h_dest))
379 			sc = info->vport.non_vlan_sc_mc;
380 		else
381 			sc = info->vport.non_vlan_sc_uc;
382 	}
383 
384 	return sc;
385 }
386 
opa_vnic_get_vl(struct opa_vnic_adapter * adapter,struct sk_buff * skb)387 u8 opa_vnic_get_vl(struct opa_vnic_adapter *adapter, struct sk_buff *skb)
388 {
389 	struct ethhdr *mac_hdr = (struct ethhdr *)skb_mac_header(skb);
390 	struct __opa_veswport_info *info = &adapter->info;
391 	u8 vl;
392 
393 	if (skb_vlan_tag_present(skb)) {
394 		u8 pcp = skb_vlan_tag_get(skb) >> VLAN_PRIO_SHIFT;
395 
396 		if (is_multicast_ether_addr(mac_hdr->h_dest))
397 			vl = info->vport.pcp_to_vl_mc[pcp];
398 		else
399 			vl = info->vport.pcp_to_vl_uc[pcp];
400 	} else {
401 		if (is_multicast_ether_addr(mac_hdr->h_dest))
402 			vl = info->vport.non_vlan_vl_mc;
403 		else
404 			vl = info->vport.non_vlan_vl_uc;
405 	}
406 
407 	return vl;
408 }
409 
410 /* opa_vnic_get_rc - return the routing control */
opa_vnic_get_rc(struct __opa_veswport_info * info,struct sk_buff * skb)411 static u8 opa_vnic_get_rc(struct __opa_veswport_info *info,
412 			  struct sk_buff *skb)
413 {
414 	u8 proto, rout_ctrl;
415 
416 	switch (vlan_get_protocol(skb)) {
417 	case htons(ETH_P_IPV6):
418 		proto = ipv6_hdr(skb)->nexthdr;
419 		if (proto == IPPROTO_TCP)
420 			rout_ctrl = OPA_VNIC_ENCAP_RC_EXT(info->vesw.rc,
421 							  IPV6_TCP);
422 		else if (proto == IPPROTO_UDP)
423 			rout_ctrl = OPA_VNIC_ENCAP_RC_EXT(info->vesw.rc,
424 							  IPV6_UDP);
425 		else
426 			rout_ctrl = OPA_VNIC_ENCAP_RC_EXT(info->vesw.rc, IPV6);
427 		break;
428 	case htons(ETH_P_IP):
429 		proto = ip_hdr(skb)->protocol;
430 		if (proto == IPPROTO_TCP)
431 			rout_ctrl = OPA_VNIC_ENCAP_RC_EXT(info->vesw.rc,
432 							  IPV4_TCP);
433 		else if (proto == IPPROTO_UDP)
434 			rout_ctrl = OPA_VNIC_ENCAP_RC_EXT(info->vesw.rc,
435 							  IPV4_UDP);
436 		else
437 			rout_ctrl = OPA_VNIC_ENCAP_RC_EXT(info->vesw.rc, IPV4);
438 		break;
439 	default:
440 		rout_ctrl = OPA_VNIC_ENCAP_RC_EXT(info->vesw.rc, DEFAULT);
441 	}
442 
443 	return rout_ctrl;
444 }
445 
446 /* opa_vnic_calc_entropy - calculate the packet entropy */
opa_vnic_calc_entropy(struct sk_buff * skb)447 u8 opa_vnic_calc_entropy(struct sk_buff *skb)
448 {
449 	u32 hash = skb_get_hash(skb);
450 
451 	/* store XOR of all bytes in lower 8 bits */
452 	hash ^= hash >> 8;
453 	hash ^= hash >> 16;
454 
455 	/* return lower 8 bits as entropy */
456 	return (u8)(hash & 0xFF);
457 }
458 
459 /* opa_vnic_get_def_port - get default port based on entropy */
opa_vnic_get_def_port(struct opa_vnic_adapter * adapter,u8 entropy)460 static inline u8 opa_vnic_get_def_port(struct opa_vnic_adapter *adapter,
461 				       u8 entropy)
462 {
463 	u8 flow_id;
464 
465 	/* Add the upper and lower 4-bits of entropy to get the flow id */
466 	flow_id = ((entropy & 0xf) + (entropy >> 4));
467 	return adapter->flow_tbl[flow_id & (OPA_VNIC_FLOW_TBL_SIZE - 1)];
468 }
469 
470 /* Calculate packet length including OPA header, crc and padding */
opa_vnic_wire_length(struct sk_buff * skb)471 static inline int opa_vnic_wire_length(struct sk_buff *skb)
472 {
473 	u32 pad_len;
474 
475 	/* padding for 8 bytes size alignment */
476 	pad_len = -(skb->len + OPA_VNIC_ICRC_TAIL_LEN) & 0x7;
477 	pad_len += OPA_VNIC_ICRC_TAIL_LEN;
478 
479 	return (skb->len + pad_len) >> 3;
480 }
481 
482 /* opa_vnic_encap_skb - encapsulate skb packet with OPA header and meta data */
opa_vnic_encap_skb(struct opa_vnic_adapter * adapter,struct sk_buff * skb)483 void opa_vnic_encap_skb(struct opa_vnic_adapter *adapter, struct sk_buff *skb)
484 {
485 	struct __opa_veswport_info *info = &adapter->info;
486 	struct opa_vnic_skb_mdata *mdata;
487 	u8 def_port, sc, rc, entropy, *hdr;
488 	u16 len, l4_hdr;
489 	u32 dlid;
490 
491 	hdr = skb_push(skb, OPA_VNIC_HDR_LEN);
492 
493 	entropy = opa_vnic_calc_entropy(skb);
494 	def_port = opa_vnic_get_def_port(adapter, entropy);
495 	len = opa_vnic_wire_length(skb);
496 	dlid = opa_vnic_get_dlid(adapter, skb, def_port);
497 	sc = opa_vnic_get_sc(info, skb);
498 	rc = opa_vnic_get_rc(info, skb);
499 	l4_hdr = info->vesw.vesw_id;
500 
501 	mdata = skb_push(skb, sizeof(*mdata));
502 	mdata->vl = opa_vnic_get_vl(adapter, skb);
503 	mdata->entropy = entropy;
504 	mdata->flags = 0;
505 	if (unlikely(!dlid)) {
506 		mdata->flags = OPA_VNIC_SKB_MDATA_ENCAP_ERR;
507 		return;
508 	}
509 
510 	opa_vnic_make_header(hdr, info->vport.encap_slid, dlid, len,
511 			     info->vesw.pkey, entropy, sc, rc,
512 			     OPA_VNIC_L4_ETHR, l4_hdr);
513 }
514