xref: /linux/drivers/net/ethernet/chelsio/inline_crypto/ch_ktls/chcr_ktls.c (revision 621cde16e49b3ecf7d59a8106a20aaebfb4a59a9)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright (C) 2020 Chelsio Communications.  All rights reserved. */
3 
4 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
5 
6 #include <linux/skbuff.h>
7 #include <linux/module.h>
8 #include <linux/highmem.h>
9 #include <linux/ip.h>
10 #include <net/ipv6.h>
11 #include <linux/netdevice.h>
12 #include <crypto/aes.h>
13 #include <linux/skbuff_ref.h>
14 #include "chcr_ktls.h"
15 
16 static LIST_HEAD(uld_ctx_list);
17 static DEFINE_MUTEX(dev_mutex);
18 
19 /* chcr_get_nfrags_to_send: get the remaining nfrags after start offset
20  * @skb: skb
21  * @start: start offset.
22  * @len: how much data to send after @start
23  */
chcr_get_nfrags_to_send(struct sk_buff * skb,u32 start,u32 len)24 static int chcr_get_nfrags_to_send(struct sk_buff *skb, u32 start, u32 len)
25 {
26 	struct skb_shared_info *si = skb_shinfo(skb);
27 	u32 frag_size, skb_linear_data_len = skb_headlen(skb);
28 	u8 nfrags = 0, frag_idx = 0;
29 	skb_frag_t *frag;
30 
31 	/* if its a linear skb then return 1 */
32 	if (!skb_is_nonlinear(skb))
33 		return 1;
34 
35 	if (unlikely(start < skb_linear_data_len)) {
36 		frag_size = min(len, skb_linear_data_len - start);
37 	} else {
38 		start -= skb_linear_data_len;
39 
40 		frag = &si->frags[frag_idx];
41 		frag_size = skb_frag_size(frag);
42 		while (start >= frag_size) {
43 			start -= frag_size;
44 			frag_idx++;
45 			frag = &si->frags[frag_idx];
46 			frag_size = skb_frag_size(frag);
47 		}
48 		frag_size = min(len, skb_frag_size(frag) - start);
49 	}
50 	len -= frag_size;
51 	nfrags++;
52 
53 	while (len) {
54 		frag_size = min(len, skb_frag_size(&si->frags[frag_idx]));
55 		len -= frag_size;
56 		nfrags++;
57 		frag_idx++;
58 	}
59 	return nfrags;
60 }
61 
62 static int chcr_init_tcb_fields(struct chcr_ktls_info *tx_info);
63 static void clear_conn_resources(struct chcr_ktls_info *tx_info);
64 /*
65  * chcr_ktls_save_keys: calculate and save crypto keys.
66  * @tx_info - driver specific tls info.
67  * @crypto_info - tls crypto information.
68  * @direction - TX/RX direction.
69  * return - SUCCESS/FAILURE.
70  */
chcr_ktls_save_keys(struct chcr_ktls_info * tx_info,struct tls_crypto_info * crypto_info,enum tls_offload_ctx_dir direction)71 static int chcr_ktls_save_keys(struct chcr_ktls_info *tx_info,
72 			       struct tls_crypto_info *crypto_info,
73 			       enum tls_offload_ctx_dir direction)
74 {
75 	int ck_size, key_ctx_size, mac_key_size, keylen, ghash_size, ret;
76 	unsigned char ghash_h[TLS_CIPHER_AES_GCM_256_TAG_SIZE];
77 	struct tls12_crypto_info_aes_gcm_128 *info_128_gcm;
78 	struct ktls_key_ctx *kctx = &tx_info->key_ctx;
79 	struct crypto_aes_ctx aes_ctx;
80 	unsigned char *key, *salt;
81 
82 	switch (crypto_info->cipher_type) {
83 	case TLS_CIPHER_AES_GCM_128:
84 		info_128_gcm =
85 			(struct tls12_crypto_info_aes_gcm_128 *)crypto_info;
86 		keylen = TLS_CIPHER_AES_GCM_128_KEY_SIZE;
87 		ck_size = CHCR_KEYCTX_CIPHER_KEY_SIZE_128;
88 		tx_info->salt_size = TLS_CIPHER_AES_GCM_128_SALT_SIZE;
89 		mac_key_size = CHCR_KEYCTX_MAC_KEY_SIZE_128;
90 		tx_info->iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
91 		tx_info->iv = be64_to_cpu(*(__be64 *)info_128_gcm->iv);
92 
93 		ghash_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
94 		key = info_128_gcm->key;
95 		salt = info_128_gcm->salt;
96 		tx_info->record_no = *(u64 *)info_128_gcm->rec_seq;
97 
98 		/* The SCMD fields used when encrypting a full TLS
99 		 * record. Its a one time calculation till the
100 		 * connection exists.
101 		 */
102 		tx_info->scmd0_seqno_numivs =
103 			SCMD_SEQ_NO_CTRL_V(CHCR_SCMD_SEQ_NO_CTRL_64BIT) |
104 			SCMD_CIPH_AUTH_SEQ_CTRL_F |
105 			SCMD_PROTO_VERSION_V(CHCR_SCMD_PROTO_VERSION_TLS) |
106 			SCMD_CIPH_MODE_V(CHCR_SCMD_CIPHER_MODE_AES_GCM) |
107 			SCMD_AUTH_MODE_V(CHCR_SCMD_AUTH_MODE_GHASH) |
108 			SCMD_IV_SIZE_V(TLS_CIPHER_AES_GCM_128_IV_SIZE >> 1) |
109 			SCMD_NUM_IVS_V(1);
110 
111 		/* keys will be sent inline. */
112 		tx_info->scmd0_ivgen_hdrlen = SCMD_KEY_CTX_INLINE_F;
113 
114 		/* The SCMD fields used when encrypting a partial TLS
115 		 * record (no trailer and possibly a truncated payload).
116 		 */
117 		tx_info->scmd0_short_seqno_numivs =
118 			SCMD_CIPH_AUTH_SEQ_CTRL_F |
119 			SCMD_PROTO_VERSION_V(CHCR_SCMD_PROTO_VERSION_GENERIC) |
120 			SCMD_CIPH_MODE_V(CHCR_SCMD_CIPHER_MODE_AES_CTR) |
121 			SCMD_IV_SIZE_V(AES_BLOCK_LEN >> 1);
122 
123 		tx_info->scmd0_short_ivgen_hdrlen =
124 			tx_info->scmd0_ivgen_hdrlen | SCMD_AADIVDROP_F;
125 
126 		break;
127 
128 	default:
129 		pr_err("GCM: cipher type 0x%x not supported\n",
130 		       crypto_info->cipher_type);
131 		ret = -EINVAL;
132 		goto out;
133 	}
134 
135 	key_ctx_size = CHCR_KTLS_KEY_CTX_LEN +
136 		       roundup(keylen, 16) + ghash_size;
137 	/* Calculate the H = CIPH(K, 0 repeated 16 times).
138 	 * It will go in key context
139 	 */
140 
141 	ret = aes_expandkey(&aes_ctx, key, keylen);
142 	if (ret)
143 		goto out;
144 
145 	memset(ghash_h, 0, ghash_size);
146 	aes_encrypt(&aes_ctx, ghash_h, ghash_h);
147 	memzero_explicit(&aes_ctx, sizeof(aes_ctx));
148 
149 	/* fill the Key context */
150 	if (direction == TLS_OFFLOAD_CTX_DIR_TX) {
151 		kctx->ctx_hdr = FILL_KEY_CTX_HDR(ck_size,
152 						 mac_key_size,
153 						 key_ctx_size >> 4);
154 	} else {
155 		ret = -EINVAL;
156 		goto out;
157 	}
158 
159 	memcpy(kctx->salt, salt, tx_info->salt_size);
160 	memcpy(kctx->key, key, keylen);
161 	memcpy(kctx->key + keylen, ghash_h, ghash_size);
162 	tx_info->key_ctx_len = key_ctx_size;
163 
164 out:
165 	return ret;
166 }
167 
168 /*
169  * chcr_ktls_act_open_req: creates TCB entry for ipv4 connection.
170  * @sk - tcp socket.
171  * @tx_info - driver specific tls info.
172  * @atid - connection active tid.
173  * return - send success/failure.
174  */
chcr_ktls_act_open_req(struct sock * sk,struct chcr_ktls_info * tx_info,int atid)175 static int chcr_ktls_act_open_req(struct sock *sk,
176 				  struct chcr_ktls_info *tx_info,
177 				  int atid)
178 {
179 	struct inet_sock *inet = inet_sk(sk);
180 	struct cpl_t6_act_open_req *cpl6;
181 	struct cpl_act_open_req *cpl;
182 	struct sk_buff *skb;
183 	unsigned int len;
184 	int qid_atid;
185 	u64 options;
186 
187 	len = sizeof(*cpl6);
188 	skb = alloc_skb(len, GFP_KERNEL);
189 	if (unlikely(!skb))
190 		return -ENOMEM;
191 	/* mark it a control pkt */
192 	set_wr_txq(skb, CPL_PRIORITY_CONTROL, tx_info->port_id);
193 
194 	cpl6 = __skb_put_zero(skb, len);
195 	cpl = (struct cpl_act_open_req *)cpl6;
196 	INIT_TP_WR(cpl6, 0);
197 	qid_atid = TID_QID_V(tx_info->rx_qid) |
198 		   TID_TID_V(atid);
199 	OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_ACT_OPEN_REQ, qid_atid));
200 	cpl->local_port = inet->inet_sport;
201 	cpl->peer_port = inet->inet_dport;
202 	cpl->local_ip = inet->inet_rcv_saddr;
203 	cpl->peer_ip = inet->inet_daddr;
204 
205 	/* fill first 64 bit option field. */
206 	options = TCAM_BYPASS_F | ULP_MODE_V(ULP_MODE_NONE) | NON_OFFLOAD_F |
207 		  SMAC_SEL_V(tx_info->smt_idx) | TX_CHAN_V(tx_info->tx_chan);
208 	cpl->opt0 = cpu_to_be64(options);
209 
210 	/* next 64 bit option field. */
211 	options =
212 		TX_QUEUE_V(tx_info->adap->params.tp.tx_modq[tx_info->tx_chan]);
213 	cpl->opt2 = htonl(options);
214 
215 	return cxgb4_l2t_send(tx_info->netdev, skb, tx_info->l2te);
216 }
217 
218 #if IS_ENABLED(CONFIG_IPV6)
219 /*
220  * chcr_ktls_act_open_req6: creates TCB entry for ipv6 connection.
221  * @sk - tcp socket.
222  * @tx_info - driver specific tls info.
223  * @atid - connection active tid.
224  * return - send success/failure.
225  */
chcr_ktls_act_open_req6(struct sock * sk,struct chcr_ktls_info * tx_info,int atid)226 static int chcr_ktls_act_open_req6(struct sock *sk,
227 				   struct chcr_ktls_info *tx_info,
228 				   int atid)
229 {
230 	struct inet_sock *inet = inet_sk(sk);
231 	struct cpl_t6_act_open_req6 *cpl6;
232 	struct cpl_act_open_req6 *cpl;
233 	struct sk_buff *skb;
234 	unsigned int len;
235 	int qid_atid;
236 	u64 options;
237 
238 	len = sizeof(*cpl6);
239 	skb = alloc_skb(len, GFP_KERNEL);
240 	if (unlikely(!skb))
241 		return -ENOMEM;
242 	/* mark it a control pkt */
243 	set_wr_txq(skb, CPL_PRIORITY_CONTROL, tx_info->port_id);
244 
245 	cpl6 = __skb_put_zero(skb, len);
246 	cpl = (struct cpl_act_open_req6 *)cpl6;
247 	INIT_TP_WR(cpl6, 0);
248 	qid_atid = TID_QID_V(tx_info->rx_qid) | TID_TID_V(atid);
249 	OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6, qid_atid));
250 	cpl->local_port = inet->inet_sport;
251 	cpl->peer_port = inet->inet_dport;
252 	cpl->local_ip_hi = *(__be64 *)&sk->sk_v6_rcv_saddr.in6_u.u6_addr8[0];
253 	cpl->local_ip_lo = *(__be64 *)&sk->sk_v6_rcv_saddr.in6_u.u6_addr8[8];
254 	cpl->peer_ip_hi = *(__be64 *)&sk->sk_v6_daddr.in6_u.u6_addr8[0];
255 	cpl->peer_ip_lo = *(__be64 *)&sk->sk_v6_daddr.in6_u.u6_addr8[8];
256 
257 	/* first 64 bit option field. */
258 	options = TCAM_BYPASS_F | ULP_MODE_V(ULP_MODE_NONE) | NON_OFFLOAD_F |
259 		  SMAC_SEL_V(tx_info->smt_idx) | TX_CHAN_V(tx_info->tx_chan);
260 	cpl->opt0 = cpu_to_be64(options);
261 	/* next 64 bit option field. */
262 	options =
263 		TX_QUEUE_V(tx_info->adap->params.tp.tx_modq[tx_info->tx_chan]);
264 	cpl->opt2 = htonl(options);
265 
266 	return cxgb4_l2t_send(tx_info->netdev, skb, tx_info->l2te);
267 }
268 #endif /* #if IS_ENABLED(CONFIG_IPV6) */
269 
270 /*
271  * chcr_setup_connection:  create a TCB entry so that TP will form tcp packets.
272  * @sk - tcp socket.
273  * @tx_info - driver specific tls info.
274  * return: NET_TX_OK/NET_XMIT_DROP
275  */
chcr_setup_connection(struct sock * sk,struct chcr_ktls_info * tx_info)276 static int chcr_setup_connection(struct sock *sk,
277 				 struct chcr_ktls_info *tx_info)
278 {
279 	struct tid_info *t = &tx_info->adap->tids;
280 	int atid, ret = 0;
281 
282 	atid = cxgb4_alloc_atid(t, tx_info);
283 	if (atid == -1)
284 		return -EINVAL;
285 
286 	tx_info->atid = atid;
287 
288 	if (tx_info->ip_family == AF_INET) {
289 		ret = chcr_ktls_act_open_req(sk, tx_info, atid);
290 #if IS_ENABLED(CONFIG_IPV6)
291 	} else {
292 		ret = cxgb4_clip_get(tx_info->netdev, (const u32 *)
293 				     &sk->sk_v6_rcv_saddr,
294 				     1);
295 		if (ret)
296 			return ret;
297 		ret = chcr_ktls_act_open_req6(sk, tx_info, atid);
298 #endif
299 	}
300 
301 	/* if return type is NET_XMIT_CN, msg will be sent but delayed, mark ret
302 	 * success, if any other return type clear atid and return that failure.
303 	 */
304 	if (ret) {
305 		if (ret == NET_XMIT_CN) {
306 			ret = 0;
307 		} else {
308 #if IS_ENABLED(CONFIG_IPV6)
309 			/* clear clip entry */
310 			if (tx_info->ip_family == AF_INET6)
311 				cxgb4_clip_release(tx_info->netdev,
312 						   (const u32 *)
313 						   &sk->sk_v6_rcv_saddr,
314 						   1);
315 #endif
316 			cxgb4_free_atid(t, atid);
317 		}
318 	}
319 
320 	return ret;
321 }
322 
323 /*
324  * chcr_set_tcb_field: update tcb fields.
325  * @tx_info - driver specific tls info.
326  * @word - TCB word.
327  * @mask - TCB word related mask.
328  * @val - TCB word related value.
329  * @no_reply - set 1 if not looking for TP response.
330  */
chcr_set_tcb_field(struct chcr_ktls_info * tx_info,u16 word,u64 mask,u64 val,int no_reply)331 static int chcr_set_tcb_field(struct chcr_ktls_info *tx_info, u16 word,
332 			      u64 mask, u64 val, int no_reply)
333 {
334 	struct cpl_set_tcb_field *req;
335 	struct sk_buff *skb;
336 
337 	skb = alloc_skb(sizeof(struct cpl_set_tcb_field), GFP_ATOMIC);
338 	if (!skb)
339 		return -ENOMEM;
340 
341 	req = (struct cpl_set_tcb_field *)__skb_put_zero(skb, sizeof(*req));
342 	INIT_TP_WR_CPL(req, CPL_SET_TCB_FIELD, tx_info->tid);
343 	req->reply_ctrl = htons(QUEUENO_V(tx_info->rx_qid) |
344 				NO_REPLY_V(no_reply));
345 	req->word_cookie = htons(TCB_WORD_V(word));
346 	req->mask = cpu_to_be64(mask);
347 	req->val = cpu_to_be64(val);
348 
349 	set_wr_txq(skb, CPL_PRIORITY_CONTROL, tx_info->port_id);
350 	return cxgb4_ofld_send(tx_info->netdev, skb);
351 }
352 
353 /*
354  * chcr_ktls_dev_del:  call back for tls_dev_del.
355  * Remove the tid and l2t entry and close the connection.
356  * it per connection basis.
357  * @netdev - net device.
358  * @tls_cts - tls context.
359  * @direction - TX/RX crypto direction
360  */
chcr_ktls_dev_del(struct net_device * netdev,struct tls_context * tls_ctx,enum tls_offload_ctx_dir direction)361 static void chcr_ktls_dev_del(struct net_device *netdev,
362 			      struct tls_context *tls_ctx,
363 			      enum tls_offload_ctx_dir direction)
364 {
365 	struct chcr_ktls_info *tx_info = chcr_get_ktls_tx_info(tls_ctx);
366 	struct ch_ktls_port_stats_debug *port_stats;
367 	struct chcr_ktls_uld_ctx *u_ctx;
368 
369 	if (!tx_info)
370 		return;
371 
372 	u_ctx = tx_info->adap->uld[CXGB4_ULD_KTLS].handle;
373 	if (u_ctx && u_ctx->detach)
374 		return;
375 	/* clear l2t entry */
376 	if (tx_info->l2te)
377 		cxgb4_l2t_release(tx_info->l2te);
378 
379 #if IS_ENABLED(CONFIG_IPV6)
380 	/* clear clip entry */
381 	if (tx_info->ip_family == AF_INET6)
382 		cxgb4_clip_release(netdev, (const u32 *)
383 				   &tx_info->sk->sk_v6_rcv_saddr,
384 				   1);
385 #endif
386 
387 	/* clear tid */
388 	if (tx_info->tid != -1) {
389 		cxgb4_remove_tid(&tx_info->adap->tids, tx_info->tx_chan,
390 				 tx_info->tid, tx_info->ip_family);
391 
392 		xa_erase(&u_ctx->tid_list, tx_info->tid);
393 	}
394 
395 	port_stats = &tx_info->adap->ch_ktls_stats.ktls_port[tx_info->port_id];
396 	atomic64_inc(&port_stats->ktls_tx_connection_close);
397 	kvfree(tx_info);
398 	chcr_set_ktls_tx_info(tls_ctx, NULL);
399 	/* release module refcount */
400 	module_put(THIS_MODULE);
401 }
402 
403 /*
404  * chcr_ktls_dev_add:  call back for tls_dev_add.
405  * Create a tcb entry for TP. Also add l2t entry for the connection. And
406  * generate keys & save those keys locally.
407  * @netdev - net device.
408  * @tls_cts - tls context.
409  * @direction - TX/RX crypto direction
410  * return: SUCCESS/FAILURE.
411  */
chcr_ktls_dev_add(struct net_device * netdev,struct sock * sk,enum tls_offload_ctx_dir direction,struct tls_crypto_info * crypto_info,u32 start_offload_tcp_sn)412 static int chcr_ktls_dev_add(struct net_device *netdev, struct sock *sk,
413 			     enum tls_offload_ctx_dir direction,
414 			     struct tls_crypto_info *crypto_info,
415 			     u32 start_offload_tcp_sn)
416 {
417 	struct tls_context *tls_ctx = tls_get_ctx(sk);
418 	struct ch_ktls_port_stats_debug *port_stats;
419 	struct chcr_ktls_uld_ctx *u_ctx;
420 	struct chcr_ktls_info *tx_info;
421 	struct dst_entry *dst;
422 	struct adapter *adap;
423 	struct port_info *pi;
424 	struct neighbour *n;
425 	u8 daaddr[16];
426 	int ret = -1;
427 
428 	pi = netdev_priv(netdev);
429 	adap = pi->adapter;
430 	port_stats = &adap->ch_ktls_stats.ktls_port[pi->port_id];
431 	atomic64_inc(&port_stats->ktls_tx_connection_open);
432 	u_ctx = adap->uld[CXGB4_ULD_KTLS].handle;
433 
434 	if (direction == TLS_OFFLOAD_CTX_DIR_RX) {
435 		pr_err("not expecting for RX direction\n");
436 		goto out;
437 	}
438 
439 	if (chcr_get_ktls_tx_info(tls_ctx))
440 		goto out;
441 
442 	if (u_ctx && u_ctx->detach)
443 		goto out;
444 
445 	tx_info = kvzalloc(sizeof(*tx_info), GFP_KERNEL);
446 	if (!tx_info)
447 		goto out;
448 
449 	tx_info->sk = sk;
450 	spin_lock_init(&tx_info->lock);
451 	/* initialize tid and atid to -1, 0 is a also a valid id. */
452 	tx_info->tid = -1;
453 	tx_info->atid = -1;
454 
455 	tx_info->adap = adap;
456 	tx_info->netdev = netdev;
457 	tx_info->first_qset = pi->first_qset;
458 	tx_info->tx_chan = pi->tx_chan;
459 	tx_info->smt_idx = pi->smt_idx;
460 	tx_info->port_id = pi->port_id;
461 	tx_info->prev_ack = 0;
462 	tx_info->prev_win = 0;
463 
464 	tx_info->rx_qid = chcr_get_first_rx_qid(adap);
465 	if (unlikely(tx_info->rx_qid < 0))
466 		goto free_tx_info;
467 
468 	tx_info->prev_seq = start_offload_tcp_sn;
469 	tx_info->tcp_start_seq_number = start_offload_tcp_sn;
470 
471 	/* save crypto keys */
472 	ret = chcr_ktls_save_keys(tx_info, crypto_info, direction);
473 	if (ret < 0)
474 		goto free_tx_info;
475 
476 	/* get peer ip */
477 	if (sk->sk_family == AF_INET) {
478 		memcpy(daaddr, &sk->sk_daddr, 4);
479 		tx_info->ip_family = AF_INET;
480 #if IS_ENABLED(CONFIG_IPV6)
481 	} else {
482 		if (!ipv6_only_sock(sk) &&
483 		    ipv6_addr_type(&sk->sk_v6_daddr) == IPV6_ADDR_MAPPED) {
484 			memcpy(daaddr, &sk->sk_daddr, 4);
485 			tx_info->ip_family = AF_INET;
486 		} else {
487 			memcpy(daaddr, sk->sk_v6_daddr.in6_u.u6_addr8, 16);
488 			tx_info->ip_family = AF_INET6;
489 		}
490 #endif
491 	}
492 
493 	/* get the l2t index */
494 	dst = sk_dst_get(sk);
495 	if (!dst) {
496 		pr_err("DST entry not found\n");
497 		goto free_tx_info;
498 	}
499 	n = dst_neigh_lookup(dst, daaddr);
500 	if (!n || !n->dev) {
501 		pr_err("neighbour not found\n");
502 		dst_release(dst);
503 		goto free_tx_info;
504 	}
505 	tx_info->l2te  = cxgb4_l2t_get(adap->l2t, n, n->dev, 0);
506 
507 	neigh_release(n);
508 	dst_release(dst);
509 
510 	if (!tx_info->l2te) {
511 		pr_err("l2t entry not found\n");
512 		goto free_tx_info;
513 	}
514 
515 	/* Driver shouldn't be removed until any single connection exists */
516 	if (!try_module_get(THIS_MODULE))
517 		goto free_l2t;
518 
519 	init_completion(&tx_info->completion);
520 	/* create a filter and call cxgb4_l2t_send to send the packet out, which
521 	 * will take care of updating l2t entry in hw if not already done.
522 	 */
523 	tx_info->open_state = CH_KTLS_OPEN_PENDING;
524 
525 	if (chcr_setup_connection(sk, tx_info))
526 		goto put_module;
527 
528 	/* Wait for reply */
529 	wait_for_completion_timeout(&tx_info->completion, 30 * HZ);
530 	spin_lock_bh(&tx_info->lock);
531 	if (tx_info->open_state) {
532 		/* need to wait for hw response, can't free tx_info yet. */
533 		if (tx_info->open_state == CH_KTLS_OPEN_PENDING)
534 			tx_info->pending_close = true;
535 		else
536 			spin_unlock_bh(&tx_info->lock);
537 		/* if in pending close, free the lock after the cleanup */
538 		goto put_module;
539 	}
540 	spin_unlock_bh(&tx_info->lock);
541 
542 	/* initialize tcb */
543 	reinit_completion(&tx_info->completion);
544 	/* mark it pending for hw response */
545 	tx_info->open_state = CH_KTLS_OPEN_PENDING;
546 
547 	if (chcr_init_tcb_fields(tx_info))
548 		goto free_tid;
549 
550 	/* Wait for reply */
551 	wait_for_completion_timeout(&tx_info->completion, 30 * HZ);
552 	spin_lock_bh(&tx_info->lock);
553 	if (tx_info->open_state) {
554 		/* need to wait for hw response, can't free tx_info yet. */
555 		tx_info->pending_close = true;
556 		/* free the lock after cleanup */
557 		goto free_tid;
558 	}
559 	spin_unlock_bh(&tx_info->lock);
560 
561 	if (!cxgb4_check_l2t_valid(tx_info->l2te))
562 		goto free_tid;
563 
564 	atomic64_inc(&port_stats->ktls_tx_ctx);
565 	chcr_set_ktls_tx_info(tls_ctx, tx_info);
566 
567 	return 0;
568 
569 free_tid:
570 #if IS_ENABLED(CONFIG_IPV6)
571 	/* clear clip entry */
572 	if (tx_info->ip_family == AF_INET6)
573 		cxgb4_clip_release(netdev, (const u32 *)
574 				   &sk->sk_v6_rcv_saddr,
575 				   1);
576 #endif
577 	cxgb4_remove_tid(&tx_info->adap->tids, tx_info->tx_chan,
578 			 tx_info->tid, tx_info->ip_family);
579 
580 	xa_erase(&u_ctx->tid_list, tx_info->tid);
581 
582 put_module:
583 	/* release module refcount */
584 	module_put(THIS_MODULE);
585 free_l2t:
586 	cxgb4_l2t_release(tx_info->l2te);
587 free_tx_info:
588 	if (tx_info->pending_close)
589 		spin_unlock_bh(&tx_info->lock);
590 	else
591 		kvfree(tx_info);
592 out:
593 	atomic64_inc(&port_stats->ktls_tx_connection_fail);
594 	return -1;
595 }
596 
597 /*
598  * chcr_init_tcb_fields:  Initialize tcb fields to handle TCP seq number
599  *			  handling.
600  * @tx_info - driver specific tls info.
601  * return: NET_TX_OK/NET_XMIT_DROP
602  */
chcr_init_tcb_fields(struct chcr_ktls_info * tx_info)603 static int chcr_init_tcb_fields(struct chcr_ktls_info *tx_info)
604 {
605 	int  ret = 0;
606 
607 	/* set tcb in offload and bypass */
608 	ret =
609 	chcr_set_tcb_field(tx_info, TCB_T_FLAGS_W,
610 			   TCB_T_FLAGS_V(TF_CORE_BYPASS_F | TF_NON_OFFLOAD_F),
611 			   TCB_T_FLAGS_V(TF_CORE_BYPASS_F), 1);
612 	if (ret)
613 		return ret;
614 	/* reset snd_una and snd_next fields in tcb */
615 	ret = chcr_set_tcb_field(tx_info, TCB_SND_UNA_RAW_W,
616 				 TCB_SND_NXT_RAW_V(TCB_SND_NXT_RAW_M) |
617 				 TCB_SND_UNA_RAW_V(TCB_SND_UNA_RAW_M),
618 				 0, 1);
619 	if (ret)
620 		return ret;
621 
622 	/* reset send max */
623 	ret = chcr_set_tcb_field(tx_info, TCB_SND_MAX_RAW_W,
624 				 TCB_SND_MAX_RAW_V(TCB_SND_MAX_RAW_M),
625 				 0, 1);
626 	if (ret)
627 		return ret;
628 
629 	/* update l2t index and request for tp reply to confirm tcb is
630 	 * initialised to handle tx traffic.
631 	 */
632 	ret = chcr_set_tcb_field(tx_info, TCB_L2T_IX_W,
633 				 TCB_L2T_IX_V(TCB_L2T_IX_M),
634 				 TCB_L2T_IX_V(tx_info->l2te->idx), 0);
635 	return ret;
636 }
637 
638 /*
639  * chcr_ktls_cpl_act_open_rpl: connection reply received from TP.
640  */
chcr_ktls_cpl_act_open_rpl(struct adapter * adap,unsigned char * input)641 static int chcr_ktls_cpl_act_open_rpl(struct adapter *adap,
642 				      unsigned char *input)
643 {
644 	const struct cpl_act_open_rpl *p = (void *)input;
645 	struct chcr_ktls_info *tx_info = NULL;
646 	struct tls_offload_context_tx *tx_ctx;
647 	struct chcr_ktls_uld_ctx *u_ctx;
648 	unsigned int atid, tid, status;
649 	struct tls_context *tls_ctx;
650 	struct tid_info *t;
651 	int ret = 0;
652 
653 	tid = GET_TID(p);
654 	status = AOPEN_STATUS_G(ntohl(p->atid_status));
655 	atid = TID_TID_G(AOPEN_ATID_G(ntohl(p->atid_status)));
656 
657 	t = &adap->tids;
658 	tx_info = lookup_atid(t, atid);
659 
660 	if (!tx_info || tx_info->atid != atid) {
661 		pr_err("%s: incorrect tx_info or atid\n", __func__);
662 		return -1;
663 	}
664 
665 	cxgb4_free_atid(t, atid);
666 	tx_info->atid = -1;
667 
668 	spin_lock(&tx_info->lock);
669 	/* HW response is very close, finish pending cleanup */
670 	if (tx_info->pending_close) {
671 		spin_unlock(&tx_info->lock);
672 		if (!status) {
673 			cxgb4_remove_tid(&tx_info->adap->tids, tx_info->tx_chan,
674 					 tid, tx_info->ip_family);
675 		}
676 		kvfree(tx_info);
677 		return 0;
678 	}
679 
680 	if (!status) {
681 		tx_info->tid = tid;
682 		cxgb4_insert_tid(t, tx_info, tx_info->tid, tx_info->ip_family);
683 		/* Adding tid */
684 		tls_ctx = tls_get_ctx(tx_info->sk);
685 		tx_ctx = tls_offload_ctx_tx(tls_ctx);
686 		u_ctx = adap->uld[CXGB4_ULD_KTLS].handle;
687 		if (u_ctx) {
688 			ret = xa_insert_bh(&u_ctx->tid_list, tid, tx_ctx,
689 					   GFP_NOWAIT);
690 			if (ret < 0) {
691 				pr_err("%s: Failed to allocate tid XA entry = %d\n",
692 				       __func__, tx_info->tid);
693 				tx_info->open_state = CH_KTLS_OPEN_FAILURE;
694 				goto out;
695 			}
696 		}
697 		tx_info->open_state = CH_KTLS_OPEN_SUCCESS;
698 	} else {
699 		tx_info->open_state = CH_KTLS_OPEN_FAILURE;
700 	}
701 out:
702 	spin_unlock(&tx_info->lock);
703 
704 	complete(&tx_info->completion);
705 	return ret;
706 }
707 
708 /*
709  * chcr_ktls_cpl_set_tcb_rpl: TCB reply received from TP.
710  */
chcr_ktls_cpl_set_tcb_rpl(struct adapter * adap,unsigned char * input)711 static int chcr_ktls_cpl_set_tcb_rpl(struct adapter *adap, unsigned char *input)
712 {
713 	const struct cpl_set_tcb_rpl *p = (void *)input;
714 	struct chcr_ktls_info *tx_info = NULL;
715 	struct tid_info *t;
716 	u32 tid;
717 
718 	tid = GET_TID(p);
719 
720 	t = &adap->tids;
721 	tx_info = lookup_tid(t, tid);
722 
723 	if (!tx_info || tx_info->tid != tid) {
724 		pr_err("%s: incorrect tx_info or tid\n", __func__);
725 		return -1;
726 	}
727 
728 	spin_lock(&tx_info->lock);
729 	if (tx_info->pending_close) {
730 		spin_unlock(&tx_info->lock);
731 		kvfree(tx_info);
732 		return 0;
733 	}
734 	tx_info->open_state = CH_KTLS_OPEN_SUCCESS;
735 	spin_unlock(&tx_info->lock);
736 
737 	complete(&tx_info->completion);
738 	return 0;
739 }
740 
__chcr_write_cpl_set_tcb_ulp(struct chcr_ktls_info * tx_info,u32 tid,void * pos,u16 word,struct sge_eth_txq * q,u64 mask,u64 val,u32 reply)741 static void *__chcr_write_cpl_set_tcb_ulp(struct chcr_ktls_info *tx_info,
742 					u32 tid, void *pos, u16 word,
743 					struct sge_eth_txq *q, u64 mask,
744 					u64 val, u32 reply)
745 {
746 	struct cpl_set_tcb_field_core *cpl;
747 	struct ulptx_idata *idata;
748 	struct ulp_txpkt *txpkt;
749 
750 	/* ULP_TXPKT */
751 	txpkt = pos;
752 	txpkt->cmd_dest = htonl(ULPTX_CMD_V(ULP_TX_PKT) |
753 				ULP_TXPKT_CHANNELID_V(tx_info->port_id) |
754 				ULP_TXPKT_FID_V(q->q.cntxt_id) |
755 				ULP_TXPKT_RO_F);
756 	txpkt->len = htonl(DIV_ROUND_UP(CHCR_SET_TCB_FIELD_LEN, 16));
757 
758 	/* ULPTX_IDATA sub-command */
759 	idata = (struct ulptx_idata *)(txpkt + 1);
760 	idata->cmd_more = htonl(ULPTX_CMD_V(ULP_TX_SC_IMM));
761 	idata->len = htonl(sizeof(*cpl));
762 	pos = idata + 1;
763 
764 	cpl = pos;
765 	/* CPL_SET_TCB_FIELD */
766 	OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, tid));
767 	cpl->reply_ctrl = htons(QUEUENO_V(tx_info->rx_qid) |
768 			NO_REPLY_V(!reply));
769 	cpl->word_cookie = htons(TCB_WORD_V(word));
770 	cpl->mask = cpu_to_be64(mask);
771 	cpl->val = cpu_to_be64(val);
772 
773 	/* ULPTX_NOOP */
774 	idata = (struct ulptx_idata *)(cpl + 1);
775 	idata->cmd_more = htonl(ULPTX_CMD_V(ULP_TX_SC_NOOP));
776 	idata->len = htonl(0);
777 	pos = idata + 1;
778 
779 	return pos;
780 }
781 
782 
783 /*
784  * chcr_write_cpl_set_tcb_ulp: update tcb values.
785  * TCB is responsible to create tcp headers, so all the related values
786  * should be correctly updated.
787  * @tx_info - driver specific tls info.
788  * @q - tx queue on which packet is going out.
789  * @tid - TCB identifier.
790  * @pos - current index where should we start writing.
791  * @word - TCB word.
792  * @mask - TCB word related mask.
793  * @val - TCB word related value.
794  * @reply - set 1 if looking for TP response.
795  * return - next position to write.
796  */
chcr_write_cpl_set_tcb_ulp(struct chcr_ktls_info * tx_info,struct sge_eth_txq * q,u32 tid,void * pos,u16 word,u64 mask,u64 val,u32 reply)797 static void *chcr_write_cpl_set_tcb_ulp(struct chcr_ktls_info *tx_info,
798 					struct sge_eth_txq *q, u32 tid,
799 					void *pos, u16 word, u64 mask,
800 					u64 val, u32 reply)
801 {
802 	int left = (void *)q->q.stat - pos;
803 
804 	if (unlikely(left < CHCR_SET_TCB_FIELD_LEN)) {
805 		if (!left) {
806 			pos = q->q.desc;
807 		} else {
808 			u8 buf[48] = {0};
809 
810 			__chcr_write_cpl_set_tcb_ulp(tx_info, tid, buf, word, q,
811 						     mask, val, reply);
812 
813 			return chcr_copy_to_txd(buf, &q->q, pos,
814 						CHCR_SET_TCB_FIELD_LEN);
815 		}
816 	}
817 
818 	pos = __chcr_write_cpl_set_tcb_ulp(tx_info, tid, pos, word, q,
819 					   mask, val, reply);
820 
821 	/* check again if we are at the end of the queue */
822 	if (left == CHCR_SET_TCB_FIELD_LEN)
823 		pos = q->q.desc;
824 
825 	return pos;
826 }
827 
828 /*
829  * chcr_ktls_xmit_tcb_cpls: update tcb entry so that TP will create the header
830  * with updated values like tcp seq, ack, window etc.
831  * @tx_info - driver specific tls info.
832  * @q - TX queue.
833  * @tcp_seq
834  * @tcp_ack
835  * @tcp_win
836  * return: NETDEV_TX_BUSY/NET_TX_OK.
837  */
chcr_ktls_xmit_tcb_cpls(struct chcr_ktls_info * tx_info,struct sge_eth_txq * q,u64 tcp_seq,u64 tcp_ack,u64 tcp_win,bool offset)838 static int chcr_ktls_xmit_tcb_cpls(struct chcr_ktls_info *tx_info,
839 				   struct sge_eth_txq *q, u64 tcp_seq,
840 				   u64 tcp_ack, u64 tcp_win, bool offset)
841 {
842 	bool first_wr = ((tx_info->prev_ack == 0) && (tx_info->prev_win == 0));
843 	struct ch_ktls_port_stats_debug *port_stats;
844 	u32 len, cpl = 0, ndesc, wr_len, wr_mid = 0;
845 	struct fw_ulptx_wr *wr;
846 	int credits;
847 	void *pos;
848 
849 	wr_len = sizeof(*wr);
850 	/* there can be max 4 cpls, check if we have enough credits */
851 	len = wr_len + 4 * roundup(CHCR_SET_TCB_FIELD_LEN, 16);
852 	ndesc = DIV_ROUND_UP(len, 64);
853 
854 	credits = chcr_txq_avail(&q->q) - ndesc;
855 	if (unlikely(credits < 0)) {
856 		chcr_eth_txq_stop(q);
857 		return NETDEV_TX_BUSY;
858 	}
859 
860 	if (unlikely(credits < ETHTXQ_STOP_THRES)) {
861 		chcr_eth_txq_stop(q);
862 		wr_mid |= FW_WR_EQUEQ_F | FW_WR_EQUIQ_F;
863 	}
864 
865 	pos = &q->q.desc[q->q.pidx];
866 	/* make space for WR, we'll fill it later when we know all the cpls
867 	 * being sent out and have complete length.
868 	 */
869 	wr = pos;
870 	pos += wr_len;
871 	/* update tx_max if its a re-transmit or the first wr */
872 	if (first_wr || tcp_seq != tx_info->prev_seq) {
873 		pos = chcr_write_cpl_set_tcb_ulp(tx_info, q, tx_info->tid, pos,
874 						 TCB_TX_MAX_W,
875 						 TCB_TX_MAX_V(TCB_TX_MAX_M),
876 						 TCB_TX_MAX_V(tcp_seq), 0);
877 		cpl++;
878 	}
879 	/* reset snd una if it's a re-transmit pkt */
880 	if (tcp_seq != tx_info->prev_seq || offset) {
881 		/* reset snd_una */
882 		port_stats =
883 			&tx_info->adap->ch_ktls_stats.ktls_port[tx_info->port_id];
884 		pos = chcr_write_cpl_set_tcb_ulp(tx_info, q, tx_info->tid, pos,
885 						 TCB_SND_UNA_RAW_W,
886 						 TCB_SND_UNA_RAW_V
887 						 (TCB_SND_UNA_RAW_M),
888 						 TCB_SND_UNA_RAW_V(0), 0);
889 		if (tcp_seq != tx_info->prev_seq)
890 			atomic64_inc(&port_stats->ktls_tx_ooo);
891 		cpl++;
892 	}
893 	/* update ack */
894 	if (first_wr || tx_info->prev_ack != tcp_ack) {
895 		pos = chcr_write_cpl_set_tcb_ulp(tx_info, q, tx_info->tid, pos,
896 						 TCB_RCV_NXT_W,
897 						 TCB_RCV_NXT_V(TCB_RCV_NXT_M),
898 						 TCB_RCV_NXT_V(tcp_ack), 0);
899 		tx_info->prev_ack = tcp_ack;
900 		cpl++;
901 	}
902 	/* update receive window */
903 	if (first_wr || tx_info->prev_win != tcp_win) {
904 		chcr_write_cpl_set_tcb_ulp(tx_info, q, tx_info->tid, pos,
905 					   TCB_RCV_WND_W,
906 					   TCB_RCV_WND_V(TCB_RCV_WND_M),
907 					   TCB_RCV_WND_V(tcp_win), 0);
908 		tx_info->prev_win = tcp_win;
909 		cpl++;
910 	}
911 
912 	if (cpl) {
913 		/* get the actual length */
914 		len = wr_len + cpl * roundup(CHCR_SET_TCB_FIELD_LEN, 16);
915 		/* ULPTX wr */
916 		wr->op_to_compl = htonl(FW_WR_OP_V(FW_ULPTX_WR));
917 		wr->cookie = 0;
918 		/* fill len in wr field */
919 		wr->flowid_len16 = htonl(wr_mid |
920 					 FW_WR_LEN16_V(DIV_ROUND_UP(len, 16)));
921 
922 		ndesc = DIV_ROUND_UP(len, 64);
923 		chcr_txq_advance(&q->q, ndesc);
924 		cxgb4_ring_tx_db(tx_info->adap, &q->q, ndesc);
925 	}
926 	return 0;
927 }
928 
929 /*
930  * chcr_ktls_get_tx_flits
931  * returns number of flits to be sent out, it includes key context length, WR
932  * size and skb fragments.
933  */
934 static unsigned int
chcr_ktls_get_tx_flits(u32 nr_frags,unsigned int key_ctx_len)935 chcr_ktls_get_tx_flits(u32 nr_frags, unsigned int key_ctx_len)
936 {
937 	return chcr_sgl_len(nr_frags) +
938 	       DIV_ROUND_UP(key_ctx_len + CHCR_KTLS_WR_SIZE, 8);
939 }
940 
941 /*
942  * chcr_ktls_check_tcp_options: To check if there is any TCP option available
943  * other than timestamp.
944  * @skb - skb contains partial record..
945  * return: 1 / 0
946  */
947 static int
chcr_ktls_check_tcp_options(struct tcphdr * tcp)948 chcr_ktls_check_tcp_options(struct tcphdr *tcp)
949 {
950 	int cnt, opt, optlen;
951 	u_char *cp;
952 
953 	cp = (u_char *)(tcp + 1);
954 	cnt = (tcp->doff << 2) - sizeof(struct tcphdr);
955 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
956 		opt = cp[0];
957 		if (opt == TCPOPT_EOL)
958 			break;
959 		if (opt == TCPOPT_NOP) {
960 			optlen = 1;
961 		} else {
962 			if (cnt < 2)
963 				break;
964 			optlen = cp[1];
965 			if (optlen < 2 || optlen > cnt)
966 				break;
967 		}
968 		switch (opt) {
969 		case TCPOPT_NOP:
970 			break;
971 		default:
972 			return 1;
973 		}
974 	}
975 	return 0;
976 }
977 
978 /*
979  * chcr_ktls_write_tcp_options : TP can't send out all the options, we need to
980  * send out separately.
981  * @tx_info - driver specific tls info.
982  * @skb - skb contains partial record..
983  * @q - TX queue.
984  * @tx_chan - channel number.
985  * return: NETDEV_TX_OK/NETDEV_TX_BUSY.
986  */
987 static int
chcr_ktls_write_tcp_options(struct chcr_ktls_info * tx_info,struct sk_buff * skb,struct sge_eth_txq * q,uint32_t tx_chan)988 chcr_ktls_write_tcp_options(struct chcr_ktls_info *tx_info, struct sk_buff *skb,
989 			    struct sge_eth_txq *q, uint32_t tx_chan)
990 {
991 	struct fw_eth_tx_pkt_wr *wr;
992 	struct cpl_tx_pkt_core *cpl;
993 	u32 ctrl, iplen, maclen;
994 	struct ipv6hdr *ip6;
995 	unsigned int ndesc;
996 	struct tcphdr *tcp;
997 	int len16, pktlen;
998 	struct iphdr *ip;
999 	u32 wr_mid = 0;
1000 	int credits;
1001 	u8 buf[150];
1002 	u64 cntrl1;
1003 	void *pos;
1004 
1005 	iplen = skb_network_header_len(skb);
1006 	maclen = skb_mac_header_len(skb);
1007 
1008 	/* packet length = eth hdr len + ip hdr len + tcp hdr len
1009 	 * (including options).
1010 	 */
1011 	pktlen = skb_tcp_all_headers(skb);
1012 
1013 	ctrl = sizeof(*cpl) + pktlen;
1014 	len16 = DIV_ROUND_UP(sizeof(*wr) + ctrl, 16);
1015 	/* check how many descriptors needed */
1016 	ndesc = DIV_ROUND_UP(len16, 4);
1017 
1018 	credits = chcr_txq_avail(&q->q) - ndesc;
1019 	if (unlikely(credits < 0)) {
1020 		chcr_eth_txq_stop(q);
1021 		return NETDEV_TX_BUSY;
1022 	}
1023 
1024 	if (unlikely(credits < ETHTXQ_STOP_THRES)) {
1025 		chcr_eth_txq_stop(q);
1026 		wr_mid |= FW_WR_EQUEQ_F | FW_WR_EQUIQ_F;
1027 	}
1028 
1029 	pos = &q->q.desc[q->q.pidx];
1030 	wr = pos;
1031 
1032 	/* Firmware work request header */
1033 	wr->op_immdlen = htonl(FW_WR_OP_V(FW_ETH_TX_PKT_WR) |
1034 			       FW_WR_IMMDLEN_V(ctrl));
1035 
1036 	wr->equiq_to_len16 = htonl(wr_mid | FW_WR_LEN16_V(len16));
1037 	wr->r3 = 0;
1038 
1039 	cpl = (void *)(wr + 1);
1040 
1041 	/* CPL header */
1042 	cpl->ctrl0 = htonl(TXPKT_OPCODE_V(CPL_TX_PKT) | TXPKT_INTF_V(tx_chan) |
1043 			   TXPKT_PF_V(tx_info->adap->pf));
1044 	cpl->pack = 0;
1045 	cpl->len = htons(pktlen);
1046 
1047 	memcpy(buf, skb->data, pktlen);
1048 	if (!IS_ENABLED(CONFIG_IPV6) || tx_info->ip_family == AF_INET) {
1049 		/* we need to correct ip header len */
1050 		ip = (struct iphdr *)(buf + maclen);
1051 		ip->tot_len = htons(pktlen - maclen);
1052 		cntrl1 = TXPKT_CSUM_TYPE_V(TX_CSUM_TCPIP);
1053 	} else {
1054 		ip6 = (struct ipv6hdr *)(buf + maclen);
1055 		ip6->payload_len = htons(pktlen - maclen - iplen);
1056 		cntrl1 = TXPKT_CSUM_TYPE_V(TX_CSUM_TCPIP6);
1057 	}
1058 
1059 	cntrl1 |= T6_TXPKT_ETHHDR_LEN_V(maclen - ETH_HLEN) |
1060 		  TXPKT_IPHDR_LEN_V(iplen);
1061 	/* checksum offload */
1062 	cpl->ctrl1 = cpu_to_be64(cntrl1);
1063 
1064 	pos = cpl + 1;
1065 
1066 	/* now take care of the tcp header, if fin is not set then clear push
1067 	 * bit as well, and if fin is set, it will be sent at the last so we
1068 	 * need to update the tcp sequence number as per the last packet.
1069 	 */
1070 	tcp = (struct tcphdr *)(buf + maclen + iplen);
1071 
1072 	if (!tcp->fin)
1073 		tcp->psh = 0;
1074 	else
1075 		tcp->seq = htonl(tx_info->prev_seq);
1076 
1077 	chcr_copy_to_txd(buf, &q->q, pos, pktlen);
1078 
1079 	chcr_txq_advance(&q->q, ndesc);
1080 	cxgb4_ring_tx_db(tx_info->adap, &q->q, ndesc);
1081 	return 0;
1082 }
1083 
1084 /*
1085  * chcr_ktls_xmit_wr_complete: This sends out the complete record. If an skb
1086  * received has partial end part of the record, send out the complete record, so
1087  * that crypto block will be able to generate TAG/HASH.
1088  * @skb - segment which has complete or partial end part.
1089  * @tx_info - driver specific tls info.
1090  * @q - TX queue.
1091  * @tcp_seq
1092  * @tcp_push - tcp push bit.
1093  * @mss - segment size.
1094  * return: NETDEV_TX_BUSY/NET_TX_OK.
1095  */
chcr_ktls_xmit_wr_complete(struct sk_buff * skb,struct chcr_ktls_info * tx_info,struct sge_eth_txq * q,u32 tcp_seq,bool is_last_wr,u32 data_len,u32 skb_offset,u32 nfrags,bool tcp_push,u32 mss)1096 static int chcr_ktls_xmit_wr_complete(struct sk_buff *skb,
1097 				      struct chcr_ktls_info *tx_info,
1098 				      struct sge_eth_txq *q, u32 tcp_seq,
1099 				      bool is_last_wr, u32 data_len,
1100 				      u32 skb_offset, u32 nfrags,
1101 				      bool tcp_push, u32 mss)
1102 {
1103 	u32 len16, wr_mid = 0, flits = 0, ndesc, cipher_start;
1104 	struct adapter *adap = tx_info->adap;
1105 	int credits, left, last_desc;
1106 	struct tx_sw_desc *sgl_sdesc;
1107 	struct cpl_tx_data *tx_data;
1108 	struct cpl_tx_sec_pdu *cpl;
1109 	struct ulptx_idata *idata;
1110 	struct ulp_txpkt *ulptx;
1111 	struct fw_ulptx_wr *wr;
1112 	void *pos;
1113 	u64 *end;
1114 
1115 	/* get the number of flits required */
1116 	flits = chcr_ktls_get_tx_flits(nfrags, tx_info->key_ctx_len);
1117 	/* number of descriptors */
1118 	ndesc = chcr_flits_to_desc(flits);
1119 	/* check if enough credits available */
1120 	credits = chcr_txq_avail(&q->q) - ndesc;
1121 	if (unlikely(credits < 0)) {
1122 		chcr_eth_txq_stop(q);
1123 		return NETDEV_TX_BUSY;
1124 	}
1125 
1126 	if (unlikely(credits < ETHTXQ_STOP_THRES)) {
1127 		/* Credits are below the threshold values, stop the queue after
1128 		 * injecting the Work Request for this packet.
1129 		 */
1130 		chcr_eth_txq_stop(q);
1131 		wr_mid |= FW_WR_EQUEQ_F | FW_WR_EQUIQ_F;
1132 	}
1133 
1134 	last_desc = q->q.pidx + ndesc - 1;
1135 	if (last_desc >= q->q.size)
1136 		last_desc -= q->q.size;
1137 	sgl_sdesc = &q->q.sdesc[last_desc];
1138 
1139 	if (unlikely(cxgb4_map_skb(adap->pdev_dev, skb, sgl_sdesc->addr) < 0)) {
1140 		memset(sgl_sdesc->addr, 0, sizeof(sgl_sdesc->addr));
1141 		q->mapping_err++;
1142 		return NETDEV_TX_BUSY;
1143 	}
1144 
1145 	if (!is_last_wr)
1146 		skb_get(skb);
1147 
1148 	pos = &q->q.desc[q->q.pidx];
1149 	end = (u64 *)pos + flits;
1150 	/* FW_ULPTX_WR */
1151 	wr = pos;
1152 	/* WR will need len16 */
1153 	len16 = DIV_ROUND_UP(flits, 2);
1154 	wr->op_to_compl = htonl(FW_WR_OP_V(FW_ULPTX_WR));
1155 	wr->flowid_len16 = htonl(wr_mid | FW_WR_LEN16_V(len16));
1156 	wr->cookie = 0;
1157 	pos += sizeof(*wr);
1158 	/* ULP_TXPKT */
1159 	ulptx = pos;
1160 	ulptx->cmd_dest = htonl(ULPTX_CMD_V(ULP_TX_PKT) |
1161 				ULP_TXPKT_CHANNELID_V(tx_info->port_id) |
1162 				ULP_TXPKT_FID_V(q->q.cntxt_id) |
1163 				ULP_TXPKT_RO_F);
1164 	ulptx->len = htonl(len16 - 1);
1165 	/* ULPTX_IDATA sub-command */
1166 	idata = (struct ulptx_idata *)(ulptx + 1);
1167 	idata->cmd_more = htonl(ULPTX_CMD_V(ULP_TX_SC_IMM) | ULP_TX_SC_MORE_F);
1168 	/* idata length will include cpl_tx_sec_pdu + key context size +
1169 	 * cpl_tx_data header.
1170 	 */
1171 	idata->len = htonl(sizeof(*cpl) + tx_info->key_ctx_len +
1172 			   sizeof(*tx_data));
1173 	/* SEC CPL */
1174 	cpl = (struct cpl_tx_sec_pdu *)(idata + 1);
1175 	cpl->op_ivinsrtofst =
1176 		htonl(CPL_TX_SEC_PDU_OPCODE_V(CPL_TX_SEC_PDU) |
1177 		      CPL_TX_SEC_PDU_CPLLEN_V(CHCR_CPL_TX_SEC_PDU_LEN_64BIT) |
1178 		      CPL_TX_SEC_PDU_PLACEHOLDER_V(1) |
1179 		      CPL_TX_SEC_PDU_IVINSRTOFST_V(TLS_HEADER_SIZE + 1));
1180 	cpl->pldlen = htonl(data_len);
1181 
1182 	/* encryption should start after tls header size + iv size */
1183 	cipher_start = TLS_HEADER_SIZE + tx_info->iv_size + 1;
1184 
1185 	cpl->aadstart_cipherstop_hi =
1186 		htonl(CPL_TX_SEC_PDU_AADSTART_V(1) |
1187 		      CPL_TX_SEC_PDU_AADSTOP_V(TLS_HEADER_SIZE) |
1188 		      CPL_TX_SEC_PDU_CIPHERSTART_V(cipher_start));
1189 
1190 	/* authentication will also start after tls header + iv size */
1191 	cpl->cipherstop_lo_authinsert =
1192 	htonl(CPL_TX_SEC_PDU_AUTHSTART_V(cipher_start) |
1193 	      CPL_TX_SEC_PDU_AUTHSTOP_V(TLS_CIPHER_AES_GCM_128_TAG_SIZE) |
1194 	      CPL_TX_SEC_PDU_AUTHINSERT_V(TLS_CIPHER_AES_GCM_128_TAG_SIZE));
1195 
1196 	/* These two flits are actually a CPL_TLS_TX_SCMD_FMT. */
1197 	cpl->seqno_numivs = htonl(tx_info->scmd0_seqno_numivs);
1198 	cpl->ivgen_hdrlen = htonl(tx_info->scmd0_ivgen_hdrlen);
1199 	cpl->scmd1 = cpu_to_be64(tx_info->record_no);
1200 
1201 	pos = cpl + 1;
1202 	/* check if space left to fill the keys */
1203 	left = (void *)q->q.stat - pos;
1204 	if (!left) {
1205 		left = (void *)end - (void *)q->q.stat;
1206 		pos = q->q.desc;
1207 		end = pos + left;
1208 	}
1209 
1210 	pos = chcr_copy_to_txd(&tx_info->key_ctx, &q->q, pos,
1211 			       tx_info->key_ctx_len);
1212 	left = (void *)q->q.stat - pos;
1213 
1214 	if (!left) {
1215 		left = (void *)end - (void *)q->q.stat;
1216 		pos = q->q.desc;
1217 		end = pos + left;
1218 	}
1219 	/* CPL_TX_DATA */
1220 	tx_data = (void *)pos;
1221 	OPCODE_TID(tx_data) = htonl(MK_OPCODE_TID(CPL_TX_DATA, tx_info->tid));
1222 	tx_data->len = htonl(TX_DATA_MSS_V(mss) | TX_LENGTH_V(data_len));
1223 
1224 	tx_data->rsvd = htonl(tcp_seq);
1225 
1226 	tx_data->flags = htonl(TX_BYPASS_F);
1227 	if (tcp_push)
1228 		tx_data->flags |= htonl(TX_PUSH_F | TX_SHOVE_F);
1229 
1230 	/* check left again, it might go beyond queue limit */
1231 	pos = tx_data + 1;
1232 	left = (void *)q->q.stat - pos;
1233 
1234 	/* check the position again */
1235 	if (!left) {
1236 		left = (void *)end - (void *)q->q.stat;
1237 		pos = q->q.desc;
1238 		end = pos + left;
1239 	}
1240 
1241 	/* send the complete packet except the header */
1242 	cxgb4_write_partial_sgl(skb, &q->q, pos, end, sgl_sdesc->addr,
1243 				skb_offset, data_len);
1244 	sgl_sdesc->skb = skb;
1245 
1246 	chcr_txq_advance(&q->q, ndesc);
1247 	cxgb4_ring_tx_db(adap, &q->q, ndesc);
1248 	atomic64_inc(&adap->ch_ktls_stats.ktls_tx_send_records);
1249 
1250 	return 0;
1251 }
1252 
1253 /*
1254  * chcr_ktls_xmit_wr_short: This is to send out partial records. If its
1255  * a middle part of a record, fetch the prior data to make it 16 byte aligned
1256  * and then only send it out.
1257  *
1258  * @skb - skb contains partial record..
1259  * @tx_info - driver specific tls info.
1260  * @q - TX queue.
1261  * @tcp_seq
1262  * @tcp_push - tcp push bit.
1263  * @mss - segment size.
1264  * @tls_rec_offset - offset from start of the tls record.
1265  * @perior_data - data before the current segment, required to make this record
1266  *		  16 byte aligned.
1267  * @prior_data_len - prior_data length (less than 16)
1268  * return: NETDEV_TX_BUSY/NET_TX_OK.
1269  */
chcr_ktls_xmit_wr_short(struct sk_buff * skb,struct chcr_ktls_info * tx_info,struct sge_eth_txq * q,u32 tcp_seq,bool tcp_push,u32 mss,u32 tls_rec_offset,u8 * prior_data,u32 prior_data_len,u32 data_len,u32 skb_offset)1270 static int chcr_ktls_xmit_wr_short(struct sk_buff *skb,
1271 				   struct chcr_ktls_info *tx_info,
1272 				   struct sge_eth_txq *q,
1273 				   u32 tcp_seq, bool tcp_push, u32 mss,
1274 				   u32 tls_rec_offset, u8 *prior_data,
1275 				   u32 prior_data_len, u32 data_len,
1276 				   u32 skb_offset)
1277 {
1278 	u32 len16, wr_mid = 0, cipher_start, nfrags;
1279 	struct adapter *adap = tx_info->adap;
1280 	unsigned int flits = 0, ndesc;
1281 	int credits, left, last_desc;
1282 	struct tx_sw_desc *sgl_sdesc;
1283 	struct cpl_tx_data *tx_data;
1284 	struct cpl_tx_sec_pdu *cpl;
1285 	struct ulptx_idata *idata;
1286 	struct ulp_txpkt *ulptx;
1287 	struct fw_ulptx_wr *wr;
1288 	__be64 iv_record;
1289 	void *pos;
1290 	u64 *end;
1291 
1292 	nfrags = chcr_get_nfrags_to_send(skb, skb_offset, data_len);
1293 	/* get the number of flits required, it's a partial record so 2 flits
1294 	 * (AES_BLOCK_SIZE) will be added.
1295 	 */
1296 	flits = chcr_ktls_get_tx_flits(nfrags, tx_info->key_ctx_len) + 2;
1297 	/* get the correct 8 byte IV of this record */
1298 	iv_record = cpu_to_be64(tx_info->iv + tx_info->record_no);
1299 	/* If it's a middle record and not 16 byte aligned to run AES CTR, need
1300 	 * to make it 16 byte aligned. So atleadt 2 extra flits of immediate
1301 	 * data will be added.
1302 	 */
1303 	if (prior_data_len)
1304 		flits += 2;
1305 	/* number of descriptors */
1306 	ndesc = chcr_flits_to_desc(flits);
1307 	/* check if enough credits available */
1308 	credits = chcr_txq_avail(&q->q) - ndesc;
1309 	if (unlikely(credits < 0)) {
1310 		chcr_eth_txq_stop(q);
1311 		return NETDEV_TX_BUSY;
1312 	}
1313 
1314 	if (unlikely(credits < ETHTXQ_STOP_THRES)) {
1315 		chcr_eth_txq_stop(q);
1316 		wr_mid |= FW_WR_EQUEQ_F | FW_WR_EQUIQ_F;
1317 	}
1318 
1319 	last_desc = q->q.pidx + ndesc - 1;
1320 	if (last_desc >= q->q.size)
1321 		last_desc -= q->q.size;
1322 	sgl_sdesc = &q->q.sdesc[last_desc];
1323 
1324 	if (unlikely(cxgb4_map_skb(adap->pdev_dev, skb, sgl_sdesc->addr) < 0)) {
1325 		memset(sgl_sdesc->addr, 0, sizeof(sgl_sdesc->addr));
1326 		q->mapping_err++;
1327 		return NETDEV_TX_BUSY;
1328 	}
1329 
1330 	pos = &q->q.desc[q->q.pidx];
1331 	end = (u64 *)pos + flits;
1332 	/* FW_ULPTX_WR */
1333 	wr = pos;
1334 	/* WR will need len16 */
1335 	len16 = DIV_ROUND_UP(flits, 2);
1336 	wr->op_to_compl = htonl(FW_WR_OP_V(FW_ULPTX_WR));
1337 	wr->flowid_len16 = htonl(wr_mid | FW_WR_LEN16_V(len16));
1338 	wr->cookie = 0;
1339 	pos += sizeof(*wr);
1340 	/* ULP_TXPKT */
1341 	ulptx = pos;
1342 	ulptx->cmd_dest = htonl(ULPTX_CMD_V(ULP_TX_PKT) |
1343 				ULP_TXPKT_CHANNELID_V(tx_info->port_id) |
1344 				ULP_TXPKT_FID_V(q->q.cntxt_id) |
1345 				ULP_TXPKT_RO_F);
1346 	ulptx->len = htonl(len16 - 1);
1347 	/* ULPTX_IDATA sub-command */
1348 	idata = (struct ulptx_idata *)(ulptx + 1);
1349 	idata->cmd_more = htonl(ULPTX_CMD_V(ULP_TX_SC_IMM) | ULP_TX_SC_MORE_F);
1350 	/* idata length will include cpl_tx_sec_pdu + key context size +
1351 	 * cpl_tx_data header.
1352 	 */
1353 	idata->len = htonl(sizeof(*cpl) + tx_info->key_ctx_len +
1354 			   sizeof(*tx_data) + AES_BLOCK_LEN + prior_data_len);
1355 	/* SEC CPL */
1356 	cpl = (struct cpl_tx_sec_pdu *)(idata + 1);
1357 	/* cipher start will have tls header + iv size extra if its a header
1358 	 * part of tls record. else only 16 byte IV will be added.
1359 	 */
1360 	cipher_start =
1361 		AES_BLOCK_LEN + 1 +
1362 		(!tls_rec_offset ? TLS_HEADER_SIZE + tx_info->iv_size : 0);
1363 
1364 	cpl->op_ivinsrtofst =
1365 		htonl(CPL_TX_SEC_PDU_OPCODE_V(CPL_TX_SEC_PDU) |
1366 		      CPL_TX_SEC_PDU_CPLLEN_V(CHCR_CPL_TX_SEC_PDU_LEN_64BIT) |
1367 		      CPL_TX_SEC_PDU_IVINSRTOFST_V(1));
1368 	cpl->pldlen = htonl(data_len + AES_BLOCK_LEN + prior_data_len);
1369 	cpl->aadstart_cipherstop_hi =
1370 		htonl(CPL_TX_SEC_PDU_CIPHERSTART_V(cipher_start));
1371 	cpl->cipherstop_lo_authinsert = 0;
1372 	/* These two flits are actually a CPL_TLS_TX_SCMD_FMT. */
1373 	cpl->seqno_numivs = htonl(tx_info->scmd0_short_seqno_numivs);
1374 	cpl->ivgen_hdrlen = htonl(tx_info->scmd0_short_ivgen_hdrlen);
1375 	cpl->scmd1 = 0;
1376 
1377 	pos = cpl + 1;
1378 	/* check if space left to fill the keys */
1379 	left = (void *)q->q.stat - pos;
1380 	if (!left) {
1381 		left = (void *)end - (void *)q->q.stat;
1382 		pos = q->q.desc;
1383 		end = pos + left;
1384 	}
1385 
1386 	pos = chcr_copy_to_txd(&tx_info->key_ctx, &q->q, pos,
1387 			       tx_info->key_ctx_len);
1388 	left = (void *)q->q.stat - pos;
1389 
1390 	if (!left) {
1391 		left = (void *)end - (void *)q->q.stat;
1392 		pos = q->q.desc;
1393 		end = pos + left;
1394 	}
1395 	/* CPL_TX_DATA */
1396 	tx_data = (void *)pos;
1397 	OPCODE_TID(tx_data) = htonl(MK_OPCODE_TID(CPL_TX_DATA, tx_info->tid));
1398 	tx_data->len = htonl(TX_DATA_MSS_V(mss) |
1399 			     TX_LENGTH_V(data_len + prior_data_len));
1400 	tx_data->rsvd = htonl(tcp_seq);
1401 	tx_data->flags = htonl(TX_BYPASS_F);
1402 	if (tcp_push)
1403 		tx_data->flags |= htonl(TX_PUSH_F | TX_SHOVE_F);
1404 
1405 	/* check left again, it might go beyond queue limit */
1406 	pos = tx_data + 1;
1407 	left = (void *)q->q.stat - pos;
1408 
1409 	/* check the position again */
1410 	if (!left) {
1411 		left = (void *)end - (void *)q->q.stat;
1412 		pos = q->q.desc;
1413 		end = pos + left;
1414 	}
1415 	/* copy the 16 byte IV for AES-CTR, which includes 4 bytes of salt, 8
1416 	 * bytes of actual IV and 4 bytes of 16 byte-sequence.
1417 	 */
1418 	memcpy(pos, tx_info->key_ctx.salt, tx_info->salt_size);
1419 	memcpy(pos + tx_info->salt_size, &iv_record, tx_info->iv_size);
1420 	*(__be32 *)(pos + tx_info->salt_size + tx_info->iv_size) =
1421 		htonl(2 + (tls_rec_offset ? ((tls_rec_offset -
1422 		(TLS_HEADER_SIZE + tx_info->iv_size)) / AES_BLOCK_LEN) : 0));
1423 
1424 	pos += 16;
1425 	/* Prior_data_len will always be less than 16 bytes, fill the
1426 	 * prio_data_len after AES_CTRL_BLOCK and clear the remaining length
1427 	 * to 0.
1428 	 */
1429 	if (prior_data_len)
1430 		pos = chcr_copy_to_txd(prior_data, &q->q, pos, 16);
1431 	/* send the complete packet except the header */
1432 	cxgb4_write_partial_sgl(skb, &q->q, pos, end, sgl_sdesc->addr,
1433 				skb_offset, data_len);
1434 	sgl_sdesc->skb = skb;
1435 
1436 	chcr_txq_advance(&q->q, ndesc);
1437 	cxgb4_ring_tx_db(adap, &q->q, ndesc);
1438 
1439 	return 0;
1440 }
1441 
1442 /*
1443  * chcr_ktls_tx_plaintxt: This handler will take care of the records which has
1444  * only plain text (only tls header and iv)
1445  * @tx_info - driver specific tls info.
1446  * @skb - skb contains partial record..
1447  * @tcp_seq
1448  * @mss - segment size.
1449  * @tcp_push - tcp push bit.
1450  * @q - TX queue.
1451  * @port_id : port number
1452  * @perior_data - data before the current segment, required to make this record
1453  *		 16 byte aligned.
1454  * @prior_data_len - prior_data length (less than 16)
1455  * return: NETDEV_TX_BUSY/NET_TX_OK.
1456  */
chcr_ktls_tx_plaintxt(struct chcr_ktls_info * tx_info,struct sk_buff * skb,u32 tcp_seq,u32 mss,bool tcp_push,struct sge_eth_txq * q,u32 port_id,u8 * prior_data,u32 data_len,u32 skb_offset,u32 prior_data_len)1457 static int chcr_ktls_tx_plaintxt(struct chcr_ktls_info *tx_info,
1458 				 struct sk_buff *skb, u32 tcp_seq, u32 mss,
1459 				 bool tcp_push, struct sge_eth_txq *q,
1460 				 u32 port_id, u8 *prior_data,
1461 				 u32 data_len, u32 skb_offset,
1462 				 u32 prior_data_len)
1463 {
1464 	int credits, left, len16, last_desc;
1465 	unsigned int flits = 0, ndesc;
1466 	struct tx_sw_desc *sgl_sdesc;
1467 	struct cpl_tx_data *tx_data;
1468 	struct ulptx_idata *idata;
1469 	struct ulp_txpkt *ulptx;
1470 	struct fw_ulptx_wr *wr;
1471 	u32 wr_mid = 0, nfrags;
1472 	void *pos;
1473 	u64 *end;
1474 
1475 	flits = DIV_ROUND_UP(CHCR_PLAIN_TX_DATA_LEN, 8);
1476 	nfrags = chcr_get_nfrags_to_send(skb, skb_offset, data_len);
1477 	flits += chcr_sgl_len(nfrags);
1478 	if (prior_data_len)
1479 		flits += 2;
1480 
1481 	/* WR will need len16 */
1482 	len16 = DIV_ROUND_UP(flits, 2);
1483 	/* check how many descriptors needed */
1484 	ndesc = DIV_ROUND_UP(flits, 8);
1485 
1486 	credits = chcr_txq_avail(&q->q) - ndesc;
1487 	if (unlikely(credits < 0)) {
1488 		chcr_eth_txq_stop(q);
1489 		return NETDEV_TX_BUSY;
1490 	}
1491 
1492 	if (unlikely(credits < ETHTXQ_STOP_THRES)) {
1493 		chcr_eth_txq_stop(q);
1494 		wr_mid |= FW_WR_EQUEQ_F | FW_WR_EQUIQ_F;
1495 	}
1496 
1497 	last_desc = q->q.pidx + ndesc - 1;
1498 	if (last_desc >= q->q.size)
1499 		last_desc -= q->q.size;
1500 	sgl_sdesc = &q->q.sdesc[last_desc];
1501 
1502 	if (unlikely(cxgb4_map_skb(tx_info->adap->pdev_dev, skb,
1503 				   sgl_sdesc->addr) < 0)) {
1504 		memset(sgl_sdesc->addr, 0, sizeof(sgl_sdesc->addr));
1505 		q->mapping_err++;
1506 		return NETDEV_TX_BUSY;
1507 	}
1508 
1509 	pos = &q->q.desc[q->q.pidx];
1510 	end = (u64 *)pos + flits;
1511 	/* FW_ULPTX_WR */
1512 	wr = pos;
1513 	wr->op_to_compl = htonl(FW_WR_OP_V(FW_ULPTX_WR));
1514 	wr->flowid_len16 = htonl(wr_mid | FW_WR_LEN16_V(len16));
1515 	wr->cookie = 0;
1516 	/* ULP_TXPKT */
1517 	ulptx = (struct ulp_txpkt *)(wr + 1);
1518 	ulptx->cmd_dest = htonl(ULPTX_CMD_V(ULP_TX_PKT) |
1519 			ULP_TXPKT_DATAMODIFY_V(0) |
1520 			ULP_TXPKT_CHANNELID_V(tx_info->port_id) |
1521 			ULP_TXPKT_DEST_V(0) |
1522 			ULP_TXPKT_FID_V(q->q.cntxt_id) | ULP_TXPKT_RO_V(1));
1523 	ulptx->len = htonl(len16 - 1);
1524 	/* ULPTX_IDATA sub-command */
1525 	idata = (struct ulptx_idata *)(ulptx + 1);
1526 	idata->cmd_more = htonl(ULPTX_CMD_V(ULP_TX_SC_IMM) | ULP_TX_SC_MORE_F);
1527 	idata->len = htonl(sizeof(*tx_data) + prior_data_len);
1528 	/* CPL_TX_DATA */
1529 	tx_data = (struct cpl_tx_data *)(idata + 1);
1530 	OPCODE_TID(tx_data) = htonl(MK_OPCODE_TID(CPL_TX_DATA, tx_info->tid));
1531 	tx_data->len = htonl(TX_DATA_MSS_V(mss) |
1532 			     TX_LENGTH_V(data_len + prior_data_len));
1533 	/* set tcp seq number */
1534 	tx_data->rsvd = htonl(tcp_seq);
1535 	tx_data->flags = htonl(TX_BYPASS_F);
1536 	if (tcp_push)
1537 		tx_data->flags |= htonl(TX_PUSH_F | TX_SHOVE_F);
1538 
1539 	pos = tx_data + 1;
1540 	/* apart from prior_data_len, we should set remaining part of 16 bytes
1541 	 * to be zero.
1542 	 */
1543 	if (prior_data_len)
1544 		pos = chcr_copy_to_txd(prior_data, &q->q, pos, 16);
1545 
1546 	/* check left again, it might go beyond queue limit */
1547 	left = (void *)q->q.stat - pos;
1548 
1549 	/* check the position again */
1550 	if (!left) {
1551 		left = (void *)end - (void *)q->q.stat;
1552 		pos = q->q.desc;
1553 		end = pos + left;
1554 	}
1555 	/* send the complete packet including the header */
1556 	cxgb4_write_partial_sgl(skb, &q->q, pos, end, sgl_sdesc->addr,
1557 				skb_offset, data_len);
1558 	sgl_sdesc->skb = skb;
1559 
1560 	chcr_txq_advance(&q->q, ndesc);
1561 	cxgb4_ring_tx_db(tx_info->adap, &q->q, ndesc);
1562 	return 0;
1563 }
1564 
chcr_ktls_tunnel_pkt(struct chcr_ktls_info * tx_info,struct sk_buff * skb,struct sge_eth_txq * q)1565 static int chcr_ktls_tunnel_pkt(struct chcr_ktls_info *tx_info,
1566 				struct sk_buff *skb,
1567 				struct sge_eth_txq *q)
1568 {
1569 	u32 ctrl, iplen, maclen, wr_mid = 0, len16;
1570 	struct tx_sw_desc *sgl_sdesc;
1571 	struct fw_eth_tx_pkt_wr *wr;
1572 	struct cpl_tx_pkt_core *cpl;
1573 	unsigned int flits, ndesc;
1574 	int credits, last_desc;
1575 	u64 cntrl1, *end;
1576 	void *pos;
1577 
1578 	ctrl = sizeof(*cpl);
1579 	flits = DIV_ROUND_UP(sizeof(*wr) + ctrl, 8);
1580 
1581 	flits += chcr_sgl_len(skb_shinfo(skb)->nr_frags + 1);
1582 	len16 = DIV_ROUND_UP(flits, 2);
1583 	/* check how many descriptors needed */
1584 	ndesc = DIV_ROUND_UP(flits, 8);
1585 
1586 	credits = chcr_txq_avail(&q->q) - ndesc;
1587 	if (unlikely(credits < 0)) {
1588 		chcr_eth_txq_stop(q);
1589 		return -ENOMEM;
1590 	}
1591 
1592 	if (unlikely(credits < ETHTXQ_STOP_THRES)) {
1593 		chcr_eth_txq_stop(q);
1594 		wr_mid |= FW_WR_EQUEQ_F | FW_WR_EQUIQ_F;
1595 	}
1596 
1597 	last_desc = q->q.pidx + ndesc - 1;
1598 	if (last_desc >= q->q.size)
1599 		last_desc -= q->q.size;
1600 	sgl_sdesc = &q->q.sdesc[last_desc];
1601 
1602 	if (unlikely(cxgb4_map_skb(tx_info->adap->pdev_dev, skb,
1603 				   sgl_sdesc->addr) < 0)) {
1604 		memset(sgl_sdesc->addr, 0, sizeof(sgl_sdesc->addr));
1605 		q->mapping_err++;
1606 		return -ENOMEM;
1607 	}
1608 
1609 	iplen = skb_network_header_len(skb);
1610 	maclen = skb_mac_header_len(skb);
1611 
1612 	pos = &q->q.desc[q->q.pidx];
1613 	end = (u64 *)pos + flits;
1614 	wr = pos;
1615 
1616 	/* Firmware work request header */
1617 	wr->op_immdlen = htonl(FW_WR_OP_V(FW_ETH_TX_PKT_WR) |
1618 			       FW_WR_IMMDLEN_V(ctrl));
1619 
1620 	wr->equiq_to_len16 = htonl(wr_mid | FW_WR_LEN16_V(len16));
1621 	wr->r3 = 0;
1622 
1623 	cpl = (void *)(wr + 1);
1624 
1625 	/* CPL header */
1626 	cpl->ctrl0 = htonl(TXPKT_OPCODE_V(CPL_TX_PKT) |
1627 			   TXPKT_INTF_V(tx_info->tx_chan) |
1628 			   TXPKT_PF_V(tx_info->adap->pf));
1629 	cpl->pack = 0;
1630 	cntrl1 = TXPKT_CSUM_TYPE_V(tx_info->ip_family == AF_INET ?
1631 				   TX_CSUM_TCPIP : TX_CSUM_TCPIP6);
1632 	cntrl1 |= T6_TXPKT_ETHHDR_LEN_V(maclen - ETH_HLEN) |
1633 		  TXPKT_IPHDR_LEN_V(iplen);
1634 	/* checksum offload */
1635 	cpl->ctrl1 = cpu_to_be64(cntrl1);
1636 	cpl->len = htons(skb->len);
1637 
1638 	pos = cpl + 1;
1639 
1640 	cxgb4_write_sgl(skb, &q->q, pos, end, 0, sgl_sdesc->addr);
1641 	sgl_sdesc->skb = skb;
1642 	chcr_txq_advance(&q->q, ndesc);
1643 	cxgb4_ring_tx_db(tx_info->adap, &q->q, ndesc);
1644 	return 0;
1645 }
1646 
1647 /*
1648  * chcr_ktls_copy_record_in_skb
1649  * @nskb - new skb where the frags to be added.
1650  * @skb - old skb, to copy socket and destructor details.
1651  * @record - specific record which has complete 16k record in frags.
1652  */
chcr_ktls_copy_record_in_skb(struct sk_buff * nskb,struct sk_buff * skb,struct tls_record_info * record)1653 static void chcr_ktls_copy_record_in_skb(struct sk_buff *nskb,
1654 					 struct sk_buff *skb,
1655 					 struct tls_record_info *record)
1656 {
1657 	int i = 0;
1658 
1659 	for (i = 0; i < record->num_frags; i++) {
1660 		skb_shinfo(nskb)->frags[i] = record->frags[i];
1661 		/* increase the frag ref count */
1662 		__skb_frag_ref(&skb_shinfo(nskb)->frags[i]);
1663 	}
1664 
1665 	skb_shinfo(nskb)->nr_frags = record->num_frags;
1666 	nskb->data_len = record->len;
1667 	nskb->len += record->len;
1668 	nskb->truesize += record->len;
1669 	nskb->sk = skb->sk;
1670 	nskb->destructor = skb->destructor;
1671 	refcount_add(nskb->truesize, &nskb->sk->sk_wmem_alloc);
1672 }
1673 
1674 /*
1675  * chcr_end_part_handler: This handler will handle the record which
1676  * is complete or if record's end part is received. T6 adapter has a issue that
1677  * it can't send out TAG with partial record so if its an end part then we have
1678  * to send TAG as well and for which we need to fetch the complete record and
1679  * send it to crypto module.
1680  * @tx_info - driver specific tls info.
1681  * @skb - skb contains partial record.
1682  * @record - complete record of 16K size.
1683  * @tcp_seq
1684  * @mss - segment size in which TP needs to chop a packet.
1685  * @tcp_push_no_fin - tcp push if fin is not set.
1686  * @q - TX queue.
1687  * @tls_end_offset - offset from end of the record.
1688  * @last wr : check if this is the last part of the skb going out.
1689  * return: NETDEV_TX_OK/NETDEV_TX_BUSY.
1690  */
chcr_end_part_handler(struct chcr_ktls_info * tx_info,struct sk_buff * skb,struct tls_record_info * record,u32 tcp_seq,int mss,bool tcp_push_no_fin,struct sge_eth_txq * q,u32 skb_offset,u32 tls_end_offset,bool last_wr)1691 static int chcr_end_part_handler(struct chcr_ktls_info *tx_info,
1692 				 struct sk_buff *skb,
1693 				 struct tls_record_info *record,
1694 				 u32 tcp_seq, int mss, bool tcp_push_no_fin,
1695 				 struct sge_eth_txq *q, u32 skb_offset,
1696 				 u32 tls_end_offset, bool last_wr)
1697 {
1698 	bool free_skb_if_tx_fails = false;
1699 	struct sk_buff *nskb = NULL;
1700 
1701 	/* check if it is a complete record */
1702 	if (tls_end_offset == record->len) {
1703 		nskb = skb;
1704 		atomic64_inc(&tx_info->adap->ch_ktls_stats.ktls_tx_complete_pkts);
1705 	} else {
1706 		nskb = alloc_skb(0, GFP_ATOMIC);
1707 		if (!nskb) {
1708 			dev_kfree_skb_any(skb);
1709 			return NETDEV_TX_BUSY;
1710 		}
1711 
1712 		/* copy complete record in skb */
1713 		chcr_ktls_copy_record_in_skb(nskb, skb, record);
1714 		/* packet is being sent from the beginning, update the tcp_seq
1715 		 * accordingly.
1716 		 */
1717 		tcp_seq = tls_record_start_seq(record);
1718 		/* reset skb offset */
1719 		skb_offset = 0;
1720 
1721 		if (last_wr)
1722 			dev_kfree_skb_any(skb);
1723 		else
1724 			free_skb_if_tx_fails = true;
1725 
1726 		last_wr = true;
1727 
1728 		atomic64_inc(&tx_info->adap->ch_ktls_stats.ktls_tx_end_pkts);
1729 	}
1730 
1731 	if (chcr_ktls_xmit_wr_complete(nskb, tx_info, q, tcp_seq,
1732 				       last_wr, record->len, skb_offset,
1733 				       record->num_frags,
1734 				       (last_wr && tcp_push_no_fin),
1735 				       mss)) {
1736 		if (free_skb_if_tx_fails)
1737 			dev_kfree_skb_any(skb);
1738 		goto out;
1739 	}
1740 	tx_info->prev_seq = record->end_seq;
1741 	return 0;
1742 out:
1743 	dev_kfree_skb_any(nskb);
1744 	return NETDEV_TX_BUSY;
1745 }
1746 
1747 /*
1748  * chcr_short_record_handler: This handler will take care of the records which
1749  * doesn't have end part (1st part or the middle part(/s) of a record). In such
1750  * cases, AES CTR will be used in place of AES GCM to send out partial packet.
1751  * This partial record might be the first part of the record, or the middle
1752  * part. In case of middle record we should fetch the prior data to make it 16
1753  * byte aligned. If it has a partial tls header or iv then get to the start of
1754  * tls header. And if it has partial TAG, then remove the complete TAG and send
1755  * only the payload.
1756  * There is one more possibility that it gets a partial header, send that
1757  * portion as a plaintext.
1758  * @tx_info - driver specific tls info.
1759  * @skb - skb contains partial record..
1760  * @record - complete record of 16K size.
1761  * @tcp_seq
1762  * @mss - segment size in which TP needs to chop a packet.
1763  * @tcp_push_no_fin - tcp push if fin is not set.
1764  * @q - TX queue.
1765  * @tls_end_offset - offset from end of the record.
1766  * return: NETDEV_TX_OK/NETDEV_TX_BUSY.
1767  */
chcr_short_record_handler(struct chcr_ktls_info * tx_info,struct sk_buff * skb,struct tls_record_info * record,u32 tcp_seq,int mss,bool tcp_push_no_fin,u32 data_len,u32 skb_offset,struct sge_eth_txq * q,u32 tls_end_offset)1768 static int chcr_short_record_handler(struct chcr_ktls_info *tx_info,
1769 				     struct sk_buff *skb,
1770 				     struct tls_record_info *record,
1771 				     u32 tcp_seq, int mss, bool tcp_push_no_fin,
1772 				     u32 data_len, u32 skb_offset,
1773 				     struct sge_eth_txq *q, u32 tls_end_offset)
1774 {
1775 	u32 tls_rec_offset = tcp_seq - tls_record_start_seq(record);
1776 	u8 prior_data[16] = {0};
1777 	u32 prior_data_len = 0;
1778 
1779 	/* check if the skb is ending in middle of tag/HASH, its a big
1780 	 * trouble, send the packet before the HASH.
1781 	 */
1782 	int remaining_record = tls_end_offset - data_len;
1783 
1784 	if (remaining_record > 0 &&
1785 	    remaining_record < TLS_CIPHER_AES_GCM_128_TAG_SIZE) {
1786 		int trimmed_len = 0;
1787 
1788 		if (tls_end_offset > TLS_CIPHER_AES_GCM_128_TAG_SIZE)
1789 			trimmed_len = data_len -
1790 				      (TLS_CIPHER_AES_GCM_128_TAG_SIZE -
1791 				       remaining_record);
1792 		if (!trimmed_len)
1793 			return FALLBACK;
1794 
1795 		WARN_ON(trimmed_len > data_len);
1796 
1797 		data_len = trimmed_len;
1798 		atomic64_inc(&tx_info->adap->ch_ktls_stats.ktls_tx_trimmed_pkts);
1799 	}
1800 
1801 	/* check if it is only the header part. */
1802 	if (tls_rec_offset + data_len <= (TLS_HEADER_SIZE + tx_info->iv_size)) {
1803 		if (chcr_ktls_tx_plaintxt(tx_info, skb, tcp_seq, mss,
1804 					  tcp_push_no_fin, q,
1805 					  tx_info->port_id, prior_data,
1806 					  data_len, skb_offset, prior_data_len))
1807 			goto out;
1808 
1809 		tx_info->prev_seq = tcp_seq + data_len;
1810 		return 0;
1811 	}
1812 
1813 	/* check if the middle record's start point is 16 byte aligned. CTR
1814 	 * needs 16 byte aligned start point to start encryption.
1815 	 */
1816 	if (tls_rec_offset) {
1817 		/* there is an offset from start, means its a middle record */
1818 		int remaining = 0;
1819 
1820 		if (tls_rec_offset < (TLS_HEADER_SIZE + tx_info->iv_size)) {
1821 			prior_data_len = tls_rec_offset;
1822 			tls_rec_offset = 0;
1823 			remaining = 0;
1824 		} else {
1825 			prior_data_len =
1826 				(tls_rec_offset -
1827 				(TLS_HEADER_SIZE + tx_info->iv_size))
1828 				% AES_BLOCK_LEN;
1829 			remaining = tls_rec_offset - prior_data_len;
1830 		}
1831 
1832 		/* if prior_data_len is not zero, means we need to fetch prior
1833 		 * data to make this record 16 byte aligned, or we need to reach
1834 		 * to start offset.
1835 		 */
1836 		if (prior_data_len) {
1837 			int i = 0;
1838 			skb_frag_t *f;
1839 			int frag_size = 0, frag_delta = 0;
1840 
1841 			while (remaining > 0) {
1842 				frag_size = skb_frag_size(&record->frags[i]);
1843 				if (remaining < frag_size)
1844 					break;
1845 
1846 				remaining -= frag_size;
1847 				i++;
1848 			}
1849 			f = &record->frags[i];
1850 			frag_delta = skb_frag_size(f) - remaining;
1851 
1852 			if (frag_delta >= prior_data_len) {
1853 				memcpy_from_page(prior_data, skb_frag_page(f),
1854 						 skb_frag_off(f) + remaining,
1855 						 prior_data_len);
1856 			} else {
1857 				memcpy_from_page(prior_data, skb_frag_page(f),
1858 						 skb_frag_off(f) + remaining,
1859 						 frag_delta);
1860 
1861 				/* get the next page */
1862 				f = &record->frags[i + 1];
1863 
1864 				memcpy_from_page(prior_data + frag_delta,
1865 						 skb_frag_page(f),
1866 						 skb_frag_off(f),
1867 						 prior_data_len - frag_delta);
1868 			}
1869 			/* reset tcp_seq as per the prior_data_required len */
1870 			tcp_seq -= prior_data_len;
1871 		}
1872 		atomic64_inc(&tx_info->adap->ch_ktls_stats.ktls_tx_middle_pkts);
1873 	} else {
1874 		atomic64_inc(&tx_info->adap->ch_ktls_stats.ktls_tx_start_pkts);
1875 	}
1876 
1877 	if (chcr_ktls_xmit_wr_short(skb, tx_info, q, tcp_seq, tcp_push_no_fin,
1878 				    mss, tls_rec_offset, prior_data,
1879 				    prior_data_len, data_len, skb_offset)) {
1880 		goto out;
1881 	}
1882 
1883 	tx_info->prev_seq = tcp_seq + data_len + prior_data_len;
1884 	return 0;
1885 out:
1886 	dev_kfree_skb_any(skb);
1887 	return NETDEV_TX_BUSY;
1888 }
1889 
chcr_ktls_sw_fallback(struct sk_buff * skb,struct chcr_ktls_info * tx_info,struct sge_eth_txq * q)1890 static int chcr_ktls_sw_fallback(struct sk_buff *skb,
1891 				 struct chcr_ktls_info *tx_info,
1892 				 struct sge_eth_txq *q)
1893 {
1894 	u32 data_len, skb_offset;
1895 	struct sk_buff *nskb;
1896 	struct tcphdr *th;
1897 
1898 	nskb = tls_encrypt_skb(skb);
1899 
1900 	if (!nskb)
1901 		return 0;
1902 
1903 	th = tcp_hdr(nskb);
1904 	skb_offset = skb_tcp_all_headers(nskb);
1905 	data_len = nskb->len - skb_offset;
1906 	skb_tx_timestamp(nskb);
1907 
1908 	if (chcr_ktls_tunnel_pkt(tx_info, nskb, q))
1909 		goto out;
1910 
1911 	tx_info->prev_seq = ntohl(th->seq) + data_len;
1912 	atomic64_inc(&tx_info->adap->ch_ktls_stats.ktls_tx_fallback);
1913 	return 0;
1914 out:
1915 	dev_kfree_skb_any(nskb);
1916 	return 0;
1917 }
1918 /* nic tls TX handler */
chcr_ktls_xmit(struct sk_buff * skb,struct net_device * dev)1919 static int chcr_ktls_xmit(struct sk_buff *skb, struct net_device *dev)
1920 {
1921 	u32 tls_end_offset, tcp_seq, skb_data_len, skb_offset;
1922 	struct ch_ktls_port_stats_debug *port_stats;
1923 	struct tls_offload_context_tx *tx_ctx;
1924 	struct ch_ktls_stats_debug *stats;
1925 	struct tcphdr *th = tcp_hdr(skb);
1926 	int data_len, qidx, ret = 0, mss;
1927 	struct tls_record_info *record;
1928 	struct chcr_ktls_info *tx_info;
1929 	struct net_device *tls_netdev;
1930 	struct tls_context *tls_ctx;
1931 	struct sge_eth_txq *q;
1932 	struct adapter *adap;
1933 	unsigned long flags;
1934 
1935 	tcp_seq = ntohl(th->seq);
1936 	skb_offset = skb_tcp_all_headers(skb);
1937 	skb_data_len = skb->len - skb_offset;
1938 	data_len = skb_data_len;
1939 
1940 	mss = skb_is_gso(skb) ? skb_shinfo(skb)->gso_size : data_len;
1941 
1942 	tls_ctx = tls_get_ctx(skb->sk);
1943 	tx_ctx = tls_offload_ctx_tx(tls_ctx);
1944 	tls_netdev = rcu_dereference_bh(tls_ctx->netdev);
1945 	/* Don't quit on NULL: if tls_device_down is running in parallel,
1946 	 * netdev might become NULL, even if tls_is_skb_tx_device_offloaded was
1947 	 * true. Rather continue processing this packet.
1948 	 */
1949 	if (unlikely(tls_netdev && tls_netdev != dev))
1950 		goto out;
1951 
1952 	tx_info = chcr_get_ktls_tx_info(tls_ctx);
1953 
1954 	if (unlikely(!tx_info))
1955 		goto out;
1956 
1957 	adap = tx_info->adap;
1958 	stats = &adap->ch_ktls_stats;
1959 	port_stats = &stats->ktls_port[tx_info->port_id];
1960 
1961 	qidx = skb->queue_mapping;
1962 	q = &adap->sge.ethtxq[qidx + tx_info->first_qset];
1963 	cxgb4_reclaim_completed_tx(adap, &q->q, true);
1964 	/* if tcp options are set but finish is not send the options first */
1965 	if (!th->fin && chcr_ktls_check_tcp_options(th)) {
1966 		ret = chcr_ktls_write_tcp_options(tx_info, skb, q,
1967 						  tx_info->tx_chan);
1968 		if (ret)
1969 			return NETDEV_TX_BUSY;
1970 	}
1971 
1972 	/* TCP segments can be in received either complete or partial.
1973 	 * chcr_end_part_handler will handle cases if complete record or end
1974 	 * part of the record is received. In case of partial end part of record,
1975 	 * we will send the complete record again.
1976 	 */
1977 
1978 	spin_lock_irqsave(&tx_ctx->lock, flags);
1979 
1980 	do {
1981 
1982 		cxgb4_reclaim_completed_tx(adap, &q->q, true);
1983 		/* fetch the tls record */
1984 		record = tls_get_record(tx_ctx, tcp_seq,
1985 					&tx_info->record_no);
1986 		/* By the time packet reached to us, ACK is received, and record
1987 		 * won't be found in that case, handle it gracefully.
1988 		 */
1989 		if (unlikely(!record)) {
1990 			spin_unlock_irqrestore(&tx_ctx->lock, flags);
1991 			atomic64_inc(&port_stats->ktls_tx_drop_no_sync_data);
1992 			goto out;
1993 		}
1994 
1995 		tls_end_offset = record->end_seq - tcp_seq;
1996 
1997 		pr_debug("seq 0x%x, end_seq 0x%x prev_seq 0x%x, datalen 0x%x\n",
1998 			 tcp_seq, record->end_seq, tx_info->prev_seq, data_len);
1999 		/* update tcb for the skb */
2000 		if (skb_data_len == data_len) {
2001 			u32 tx_max = tcp_seq;
2002 
2003 			if (!tls_record_is_start_marker(record) &&
2004 			    tls_end_offset < TLS_CIPHER_AES_GCM_128_TAG_SIZE)
2005 				tx_max = record->end_seq -
2006 					TLS_CIPHER_AES_GCM_128_TAG_SIZE;
2007 
2008 			ret = chcr_ktls_xmit_tcb_cpls(tx_info, q, tx_max,
2009 						      ntohl(th->ack_seq),
2010 						      ntohs(th->window),
2011 						      tls_end_offset !=
2012 						      record->len);
2013 			if (ret) {
2014 				spin_unlock_irqrestore(&tx_ctx->lock,
2015 						       flags);
2016 				goto out;
2017 			}
2018 
2019 			if (th->fin)
2020 				skb_get(skb);
2021 		}
2022 
2023 		if (unlikely(tls_record_is_start_marker(record))) {
2024 			atomic64_inc(&port_stats->ktls_tx_skip_no_sync_data);
2025 			/* If tls_end_offset < data_len, means there is some
2026 			 * data after start marker, which needs encryption, send
2027 			 * plaintext first and take skb refcount. else send out
2028 			 * complete pkt as plaintext.
2029 			 */
2030 			if (tls_end_offset < data_len)
2031 				skb_get(skb);
2032 			else
2033 				tls_end_offset = data_len;
2034 
2035 			ret = chcr_ktls_tx_plaintxt(tx_info, skb, tcp_seq, mss,
2036 						    (!th->fin && th->psh), q,
2037 						    tx_info->port_id, NULL,
2038 						    tls_end_offset, skb_offset,
2039 						    0);
2040 
2041 			if (ret) {
2042 				/* free the refcount taken earlier */
2043 				if (tls_end_offset < data_len)
2044 					dev_kfree_skb_any(skb);
2045 				spin_unlock_irqrestore(&tx_ctx->lock, flags);
2046 				goto out;
2047 			}
2048 
2049 			data_len -= tls_end_offset;
2050 			tcp_seq = record->end_seq;
2051 			skb_offset += tls_end_offset;
2052 			continue;
2053 		}
2054 
2055 		/* if a tls record is finishing in this SKB */
2056 		if (tls_end_offset <= data_len) {
2057 			ret = chcr_end_part_handler(tx_info, skb, record,
2058 						    tcp_seq, mss,
2059 						    (!th->fin && th->psh), q,
2060 						    skb_offset,
2061 						    tls_end_offset,
2062 						    skb_offset +
2063 						    tls_end_offset == skb->len);
2064 
2065 			data_len -= tls_end_offset;
2066 			/* tcp_seq increment is required to handle next record.
2067 			 */
2068 			tcp_seq += tls_end_offset;
2069 			skb_offset += tls_end_offset;
2070 		} else {
2071 			ret = chcr_short_record_handler(tx_info, skb,
2072 							record, tcp_seq, mss,
2073 							(!th->fin && th->psh),
2074 							data_len, skb_offset,
2075 							q, tls_end_offset);
2076 			data_len = 0;
2077 		}
2078 
2079 		/* if any failure, come out from the loop. */
2080 		if (ret) {
2081 			spin_unlock_irqrestore(&tx_ctx->lock, flags);
2082 			if (th->fin)
2083 				dev_kfree_skb_any(skb);
2084 
2085 			if (ret == FALLBACK)
2086 				return chcr_ktls_sw_fallback(skb, tx_info, q);
2087 
2088 			return NETDEV_TX_OK;
2089 		}
2090 
2091 		/* length should never be less than 0 */
2092 		WARN_ON(data_len < 0);
2093 
2094 	} while (data_len > 0);
2095 
2096 	spin_unlock_irqrestore(&tx_ctx->lock, flags);
2097 	atomic64_inc(&port_stats->ktls_tx_encrypted_packets);
2098 	atomic64_add(skb_data_len, &port_stats->ktls_tx_encrypted_bytes);
2099 
2100 	/* tcp finish is set, send a separate tcp msg including all the options
2101 	 * as well.
2102 	 */
2103 	if (th->fin) {
2104 		chcr_ktls_write_tcp_options(tx_info, skb, q, tx_info->tx_chan);
2105 		dev_kfree_skb_any(skb);
2106 	}
2107 
2108 	return NETDEV_TX_OK;
2109 out:
2110 	dev_kfree_skb_any(skb);
2111 	return NETDEV_TX_OK;
2112 }
2113 
chcr_ktls_uld_add(const struct cxgb4_lld_info * lldi)2114 static void *chcr_ktls_uld_add(const struct cxgb4_lld_info *lldi)
2115 {
2116 	struct chcr_ktls_uld_ctx *u_ctx;
2117 
2118 	pr_info_once("%s - version %s\n", CHCR_KTLS_DRV_DESC,
2119 		     CHCR_KTLS_DRV_VERSION);
2120 	u_ctx = kzalloc(sizeof(*u_ctx), GFP_KERNEL);
2121 	if (!u_ctx) {
2122 		u_ctx = ERR_PTR(-ENOMEM);
2123 		goto out;
2124 	}
2125 	u_ctx->lldi = *lldi;
2126 	u_ctx->detach = false;
2127 	xa_init_flags(&u_ctx->tid_list, XA_FLAGS_LOCK_BH);
2128 out:
2129 	return u_ctx;
2130 }
2131 
2132 static const struct tlsdev_ops chcr_ktls_ops = {
2133 	.tls_dev_add = chcr_ktls_dev_add,
2134 	.tls_dev_del = chcr_ktls_dev_del,
2135 };
2136 
2137 static chcr_handler_func work_handlers[NUM_CPL_CMDS] = {
2138 	[CPL_ACT_OPEN_RPL] = chcr_ktls_cpl_act_open_rpl,
2139 	[CPL_SET_TCB_RPL] = chcr_ktls_cpl_set_tcb_rpl,
2140 };
2141 
chcr_ktls_uld_rx_handler(void * handle,const __be64 * rsp,const struct pkt_gl * pgl)2142 static int chcr_ktls_uld_rx_handler(void *handle, const __be64 *rsp,
2143 				    const struct pkt_gl *pgl)
2144 {
2145 	const struct cpl_act_open_rpl *rpl = (struct cpl_act_open_rpl *)rsp;
2146 	struct chcr_ktls_uld_ctx *u_ctx = handle;
2147 	u8 opcode = rpl->ot.opcode;
2148 	struct adapter *adap;
2149 
2150 	adap = pci_get_drvdata(u_ctx->lldi.pdev);
2151 
2152 	if (!work_handlers[opcode]) {
2153 		pr_err("Unsupported opcode %d received\n", opcode);
2154 		return 0;
2155 	}
2156 
2157 	work_handlers[opcode](adap, (unsigned char *)&rsp[1]);
2158 	return 0;
2159 }
2160 
clear_conn_resources(struct chcr_ktls_info * tx_info)2161 static void clear_conn_resources(struct chcr_ktls_info *tx_info)
2162 {
2163 	/* clear l2t entry */
2164 	if (tx_info->l2te)
2165 		cxgb4_l2t_release(tx_info->l2te);
2166 
2167 #if IS_ENABLED(CONFIG_IPV6)
2168 	/* clear clip entry */
2169 	if (tx_info->ip_family == AF_INET6)
2170 		cxgb4_clip_release(tx_info->netdev, (const u32 *)
2171 				   &tx_info->sk->sk_v6_rcv_saddr,
2172 				   1);
2173 #endif
2174 
2175 	/* clear tid */
2176 	if (tx_info->tid != -1)
2177 		cxgb4_remove_tid(&tx_info->adap->tids, tx_info->tx_chan,
2178 				 tx_info->tid, tx_info->ip_family);
2179 }
2180 
ch_ktls_reset_all_conn(struct chcr_ktls_uld_ctx * u_ctx)2181 static void ch_ktls_reset_all_conn(struct chcr_ktls_uld_ctx *u_ctx)
2182 {
2183 	struct ch_ktls_port_stats_debug *port_stats;
2184 	struct tls_offload_context_tx *tx_ctx;
2185 	struct chcr_ktls_info *tx_info;
2186 	unsigned long index;
2187 
2188 	xa_for_each(&u_ctx->tid_list, index, tx_ctx) {
2189 		tx_info = __chcr_get_ktls_tx_info(tx_ctx);
2190 		clear_conn_resources(tx_info);
2191 		port_stats = &tx_info->adap->ch_ktls_stats.ktls_port[tx_info->port_id];
2192 		atomic64_inc(&port_stats->ktls_tx_connection_close);
2193 		kvfree(tx_info);
2194 		memset(tx_ctx->driver_state, 0, TLS_DRIVER_STATE_SIZE_TX);
2195 		/* release module refcount */
2196 		module_put(THIS_MODULE);
2197 	}
2198 }
2199 
chcr_ktls_uld_state_change(void * handle,enum cxgb4_state new_state)2200 static int chcr_ktls_uld_state_change(void *handle, enum cxgb4_state new_state)
2201 {
2202 	struct chcr_ktls_uld_ctx *u_ctx = handle;
2203 
2204 	switch (new_state) {
2205 	case CXGB4_STATE_UP:
2206 		pr_info("%s: Up\n", pci_name(u_ctx->lldi.pdev));
2207 		mutex_lock(&dev_mutex);
2208 		list_add_tail(&u_ctx->entry, &uld_ctx_list);
2209 		mutex_unlock(&dev_mutex);
2210 		break;
2211 	case CXGB4_STATE_START_RECOVERY:
2212 	case CXGB4_STATE_DOWN:
2213 	case CXGB4_STATE_DETACH:
2214 		pr_info("%s: Down\n", pci_name(u_ctx->lldi.pdev));
2215 		mutex_lock(&dev_mutex);
2216 		u_ctx->detach = true;
2217 		list_del(&u_ctx->entry);
2218 		ch_ktls_reset_all_conn(u_ctx);
2219 		xa_destroy(&u_ctx->tid_list);
2220 		mutex_unlock(&dev_mutex);
2221 		break;
2222 	default:
2223 		break;
2224 	}
2225 
2226 	return 0;
2227 }
2228 
2229 static struct cxgb4_uld_info chcr_ktls_uld_info = {
2230 	.name = CHCR_KTLS_DRV_MODULE_NAME,
2231 	.nrxq = 1,
2232 	.rxq_size = 1024,
2233 	.add = chcr_ktls_uld_add,
2234 	.tx_handler = chcr_ktls_xmit,
2235 	.rx_handler = chcr_ktls_uld_rx_handler,
2236 	.state_change = chcr_ktls_uld_state_change,
2237 	.tlsdev_ops = &chcr_ktls_ops,
2238 };
2239 
chcr_ktls_init(void)2240 static int __init chcr_ktls_init(void)
2241 {
2242 	cxgb4_register_uld(CXGB4_ULD_KTLS, &chcr_ktls_uld_info);
2243 	return 0;
2244 }
2245 
chcr_ktls_exit(void)2246 static void __exit chcr_ktls_exit(void)
2247 {
2248 	struct chcr_ktls_uld_ctx *u_ctx, *tmp;
2249 	struct adapter *adap;
2250 
2251 	pr_info("driver unloaded\n");
2252 
2253 	mutex_lock(&dev_mutex);
2254 	list_for_each_entry_safe(u_ctx, tmp, &uld_ctx_list, entry) {
2255 		adap = pci_get_drvdata(u_ctx->lldi.pdev);
2256 		memset(&adap->ch_ktls_stats, 0, sizeof(adap->ch_ktls_stats));
2257 		list_del(&u_ctx->entry);
2258 		xa_destroy(&u_ctx->tid_list);
2259 		kfree(u_ctx);
2260 	}
2261 	mutex_unlock(&dev_mutex);
2262 	cxgb4_unregister_uld(CXGB4_ULD_KTLS);
2263 }
2264 
2265 module_init(chcr_ktls_init);
2266 module_exit(chcr_ktls_exit);
2267 
2268 MODULE_DESCRIPTION("Chelsio NIC TLS ULD driver");
2269 MODULE_LICENSE("GPL");
2270 MODULE_AUTHOR("Chelsio Communications");
2271 MODULE_VERSION(CHCR_KTLS_DRV_VERSION);
2272