xref: /freebsd/sys/dev/ice/ice_common_txrx.h (revision 094517119c62c23369d545a7475ae982d86330a3)
1 /* SPDX-License-Identifier: BSD-3-Clause */
2 /*  Copyright (c) 2021, Intel Corporation
3  *  All rights reserved.
4  *
5  *  Redistribution and use in source and binary forms, with or without
6  *  modification, are permitted provided that the following conditions are met:
7  *
8  *   1. Redistributions of source code must retain the above copyright notice,
9  *      this list of conditions and the following disclaimer.
10  *
11  *   2. Redistributions in binary form must reproduce the above copyright
12  *      notice, this list of conditions and the following disclaimer in the
13  *      documentation and/or other materials provided with the distribution.
14  *
15  *   3. Neither the name of the Intel Corporation nor the names of its
16  *      contributors may be used to endorse or promote products derived from
17  *      this software without specific prior written permission.
18  *
19  *  THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
20  *  AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  *  IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  *  ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
23  *  LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  *  INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  *  CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  *  ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  *  POSSIBILITY OF SUCH DAMAGE.
30  */
31 /*$FreeBSD$*/
32 
33 /**
34  * @file ice_common_txrx.h
35  * @brief common Tx/Rx utility functions
36  *
37  * Contains common utility functions for the Tx/Rx hot path.
38  *
39  * The functions do depend on the if_pkt_info_t structure. A suitable
40  * implementation of this structure must be provided if these functions are to
41  * be used without the iflib networking stack.
42  */
43 
44 #ifndef _ICE_COMMON_TXRX_H_
45 #define _ICE_COMMON_TXRX_H_
46 
47 #include <netinet/udp.h>
48 #include <netinet/sctp.h>
49 
50 /**
51  * ice_tso_detect_sparse - detect TSO packets with too many segments
52  * @pi: packet information
53  *
54  * Hardware only transmits packets with a maximum of 8 descriptors. For TSO
55  * packets, hardware needs to be able to build the split packets using 8 or
56  * fewer descriptors. Additionally, the header must be contained within at
57  * most 3 descriptors.
58  *
59  * To verify this, we walk the headers to find out how many descriptors the
60  * headers require (usually 1). Then we ensure that, for each TSO segment, its
61  * data plus the headers are contained within 8 or fewer descriptors.
62  */
63 static inline int
64 ice_tso_detect_sparse(if_pkt_info_t pi)
65 {
66 	int count, curseg, i, hlen, segsz, seglen, tsolen, hdrs, maxsegs;
67 	bus_dma_segment_t *segs = pi->ipi_segs;
68 	int nsegs = pi->ipi_nsegs;
69 
70 	curseg = hdrs = 0;
71 
72 	hlen = pi->ipi_ehdrlen + pi->ipi_ip_hlen + pi->ipi_tcp_hlen;
73 	tsolen = pi->ipi_len - hlen;
74 
75 	/* First, count the number of descriptors for the header.
76 	 * Additionally, make sure it does not span more than 3 segments.
77 	 */
78 	i = 0;
79 	curseg = segs[0].ds_len;
80 	while (hlen > 0) {
81 		hdrs++;
82 		if (hdrs > ICE_MAX_TSO_HDR_SEGS)
83 			return (1);
84 		if (curseg == 0) {
85 			i++;
86 			if (__predict_false(i == nsegs))
87 				return (1);
88 
89 			curseg = segs[i].ds_len;
90 		}
91 		seglen = min(curseg, hlen);
92 		curseg -= seglen;
93 		hlen -= seglen;
94 	}
95 
96 	maxsegs = ICE_MAX_TX_SEGS - hdrs;
97 
98 	/* We must count the headers, in order to verify that they take up
99 	 * 3 or fewer descriptors. However, we don't need to check the data
100 	 * if the total segments is small.
101 	 */
102 	if (nsegs <= maxsegs)
103 		return (0);
104 
105 	count = 0;
106 
107 	/* Now check the data to make sure that each TSO segment is made up of
108 	 * no more than maxsegs descriptors. This ensures that hardware will
109 	 * be capable of performing TSO offload.
110 	 */
111 	while (tsolen > 0) {
112 		segsz = pi->ipi_tso_segsz;
113 		while (segsz > 0 && tsolen != 0) {
114 			count++;
115 			if (count > maxsegs) {
116 				return (1);
117 			}
118 			if (curseg == 0) {
119 				i++;
120 				if (__predict_false(i == nsegs)) {
121 					return (1);
122 				}
123 				curseg = segs[i].ds_len;
124 			}
125 			seglen = min(curseg, segsz);
126 			segsz -= seglen;
127 			curseg -= seglen;
128 			tsolen -= seglen;
129 		}
130 		count = 0;
131 	}
132 
133 	return (0);
134 }
135 
136 /**
137  * ice_tso_setup - Setup a context descriptor to prepare for a TSO packet
138  * @txq: the Tx queue to use
139  * @pi: the packet info to prepare for
140  *
141  * Setup a context descriptor in preparation for sending a Tx packet that
142  * requires the TSO offload. Returns the index of the descriptor to use when
143  * encapsulating the Tx packet data into descriptors.
144  */
145 static inline int
146 ice_tso_setup(struct ice_tx_queue *txq, if_pkt_info_t pi)
147 {
148 	struct ice_tx_ctx_desc		*txd;
149 	u32				cmd, mss, type, tsolen;
150 	int				idx;
151 	u64				type_cmd_tso_mss;
152 
153 	idx = pi->ipi_pidx;
154 	txd = (struct ice_tx_ctx_desc *)&txq->tx_base[idx];
155 	tsolen = pi->ipi_len - (pi->ipi_ehdrlen + pi->ipi_ip_hlen + pi->ipi_tcp_hlen);
156 
157 	type = ICE_TX_DESC_DTYPE_CTX;
158 	cmd = ICE_TX_CTX_DESC_TSO;
159 	/* TSO MSS must not be less than 64 */
160 	if (pi->ipi_tso_segsz < ICE_MIN_TSO_MSS) {
161 		txq->stats.mss_too_small++;
162 		pi->ipi_tso_segsz = ICE_MIN_TSO_MSS;
163 	}
164 	mss = pi->ipi_tso_segsz;
165 
166 	type_cmd_tso_mss = ((u64)type << ICE_TXD_CTX_QW1_DTYPE_S) |
167 	    ((u64)cmd << ICE_TXD_CTX_QW1_CMD_S) |
168 	    ((u64)tsolen << ICE_TXD_CTX_QW1_TSO_LEN_S) |
169 	    ((u64)mss << ICE_TXD_CTX_QW1_MSS_S);
170 	txd->qw1 = htole64(type_cmd_tso_mss);
171 
172 	txd->tunneling_params = htole32(0);
173 	txq->tso++;
174 
175 	return ((idx + 1) & (txq->desc_count-1));
176 }
177 
178 /**
179  * ice_tx_setup_offload - Setup register values for performing a Tx offload
180  * @txq: The Tx queue, used to track checksum offload stats
181  * @pi: the packet info to program for
182  * @cmd: the cmd register value to update
183  * @off: the off register value to update
184  *
185  * Based on the packet info provided, update the cmd and off values for
186  * enabling Tx offloads. This depends on the packet type and which offloads
187  * have been requested.
188  *
189  * We also track the total number of times that we've requested hardware
190  * offload a particular type of checksum for debugging purposes.
191  */
192 static inline void
193 ice_tx_setup_offload(struct ice_tx_queue *txq, if_pkt_info_t pi, u32 *cmd, u32 *off)
194 {
195 	u32 remaining_csum_flags = pi->ipi_csum_flags;
196 
197 	switch (pi->ipi_etype) {
198 #ifdef INET
199 		case ETHERTYPE_IP:
200 			if (pi->ipi_csum_flags & ICE_CSUM_IP) {
201 				*cmd |= ICE_TX_DESC_CMD_IIPT_IPV4_CSUM;
202 				txq->stats.cso[ICE_CSO_STAT_TX_IP4]++;
203 				remaining_csum_flags &= ~CSUM_IP;
204 			} else
205 				*cmd |= ICE_TX_DESC_CMD_IIPT_IPV4;
206 			break;
207 #endif
208 #ifdef INET6
209 		case ETHERTYPE_IPV6:
210 			*cmd |= ICE_TX_DESC_CMD_IIPT_IPV6;
211 			/*
212 			 * This indicates that the IIPT flag was set to the IPV6 value;
213 			 * there's no checksum for IPv6 packets.
214 			 */
215 			txq->stats.cso[ICE_CSO_STAT_TX_IP6]++;
216 			break;
217 #endif
218 		default:
219 			txq->stats.cso[ICE_CSO_STAT_TX_L3_ERR]++;
220 			break;
221 	}
222 
223 	*off |= (pi->ipi_ehdrlen >> 1) << ICE_TX_DESC_LEN_MACLEN_S;
224 	*off |= (pi->ipi_ip_hlen >> 2) << ICE_TX_DESC_LEN_IPLEN_S;
225 
226 	if (!(remaining_csum_flags & ~ICE_RX_CSUM_FLAGS))
227 		return;
228 
229 	switch (pi->ipi_ipproto) {
230 		case IPPROTO_TCP:
231 			if (pi->ipi_csum_flags & ICE_CSUM_TCP) {
232 				*cmd |= ICE_TX_DESC_CMD_L4T_EOFT_TCP;
233 				*off |= (pi->ipi_tcp_hlen >> 2) <<
234 				    ICE_TX_DESC_LEN_L4_LEN_S;
235 				txq->stats.cso[ICE_CSO_STAT_TX_TCP]++;
236 			}
237 			break;
238 		case IPPROTO_UDP:
239 			if (pi->ipi_csum_flags & ICE_CSUM_UDP) {
240 				*cmd |= ICE_TX_DESC_CMD_L4T_EOFT_UDP;
241 				*off |= (sizeof(struct udphdr) >> 2) <<
242 				    ICE_TX_DESC_LEN_L4_LEN_S;
243 				txq->stats.cso[ICE_CSO_STAT_TX_UDP]++;
244 			}
245 			break;
246 		case IPPROTO_SCTP:
247 			if (pi->ipi_csum_flags & ICE_CSUM_SCTP) {
248 				*cmd |= ICE_TX_DESC_CMD_L4T_EOFT_SCTP;
249 				*off |= (sizeof(struct sctphdr) >> 2) <<
250 				    ICE_TX_DESC_LEN_L4_LEN_S;
251 				txq->stats.cso[ICE_CSO_STAT_TX_SCTP]++;
252 			}
253 			break;
254 		default:
255 			txq->stats.cso[ICE_CSO_STAT_TX_L4_ERR]++;
256 			break;
257 	}
258 }
259 
260 /**
261  * ice_rx_checksum - verify hardware checksum is valid or not
262  * @rxq: the Rx queue structure
263  * @flags: checksum flags to update
264  * @data: checksum data to update
265  * @status0: descriptor status data
266  * @ptype: packet type
267  *
268  * Determine whether the hardware indicated that the Rx checksum is valid. If
269  * so, update the checksum flags and data, informing the stack of the status
270  * of the checksum so that it does not spend time verifying it manually.
271  */
272 static void
273 ice_rx_checksum(struct ice_rx_queue *rxq, uint32_t *flags, uint32_t *data,
274 		u16 status0, u16 ptype)
275 {
276 	const u16 l3_error = (BIT(ICE_RX_FLEX_DESC_STATUS0_XSUM_IPE_S) |
277 			      BIT(ICE_RX_FLEX_DESC_STATUS0_XSUM_EIPE_S));
278 	const u16 l4_error = (BIT(ICE_RX_FLEX_DESC_STATUS0_XSUM_L4E_S) |
279 			      BIT(ICE_RX_FLEX_DESC_STATUS0_XSUM_EUDPE_S));
280 	const u16 xsum_errors = (l3_error | l4_error |
281 				 BIT(ICE_RX_FLEX_DESC_STATUS0_IPV6EXADD_S));
282 	struct ice_rx_ptype_decoded decoded;
283 	bool is_ipv4, is_ipv6;
284 
285 	/* No L3 or L4 checksum was calculated */
286 	if (!(status0 & BIT(ICE_RX_FLEX_DESC_STATUS0_L3L4P_S))) {
287 		return;
288 	}
289 
290 	decoded = ice_decode_rx_desc_ptype(ptype);
291 	*flags = 0;
292 
293 	if (!(decoded.known && decoded.outer_ip))
294 		return;
295 
296 	is_ipv4 = (decoded.outer_ip == ICE_RX_PTYPE_OUTER_IP) &&
297 	    (decoded.outer_ip_ver == ICE_RX_PTYPE_OUTER_IPV4);
298 	is_ipv6 = (decoded.outer_ip == ICE_RX_PTYPE_OUTER_IP) &&
299 	    (decoded.outer_ip_ver == ICE_RX_PTYPE_OUTER_IPV6);
300 
301 	/* No checksum errors were reported */
302 	if (!(status0 & xsum_errors)) {
303 		if (is_ipv4)
304 			*flags |= CSUM_L3_CALC | CSUM_L3_VALID;
305 
306 		switch (decoded.inner_prot) {
307 		case ICE_RX_PTYPE_INNER_PROT_TCP:
308 		case ICE_RX_PTYPE_INNER_PROT_UDP:
309 		case ICE_RX_PTYPE_INNER_PROT_SCTP:
310 			*flags |= CSUM_L4_CALC | CSUM_L4_VALID;
311 			*data |= htons(0xffff);
312 			break;
313 		default:
314 			break;
315 		}
316 
317 		return;
318 	}
319 
320 	/*
321 	 * Certain IPv6 extension headers impact the validity of L4 checksums.
322 	 * If one of these headers exist, hardware will set the IPV6EXADD bit
323 	 * in the descriptor. If the bit is set then pretend like hardware
324 	 * didn't checksum this packet.
325 	 */
326 	if (is_ipv6 && (status0 & BIT(ICE_RX_FLEX_DESC_STATUS0_IPV6EXADD_S))) {
327 		rxq->stats.cso[ICE_CSO_STAT_RX_IP6_ERR]++;
328 		return;
329 	}
330 
331 	/*
332 	 * At this point, status0 must have at least one of the l3_error or
333 	 * l4_error bits set.
334 	 */
335 
336 	if (status0 & l3_error) {
337 		if (is_ipv4) {
338 			rxq->stats.cso[ICE_CSO_STAT_RX_IP4_ERR]++;
339 			*flags |= CSUM_L3_CALC;
340 		} else {
341 			/* Hardware indicated L3 error but this isn't IPv4? */
342 			rxq->stats.cso[ICE_CSO_STAT_RX_L3_ERR]++;
343 		}
344 		/* don't bother reporting L4 errors if we got an L3 error */
345 		return;
346 	} else if (is_ipv4) {
347 		*flags |= CSUM_L3_CALC | CSUM_L3_VALID;
348 	}
349 
350 	if (status0 & l4_error) {
351 		switch (decoded.inner_prot) {
352 		case ICE_RX_PTYPE_INNER_PROT_TCP:
353 			rxq->stats.cso[ICE_CSO_STAT_RX_TCP_ERR]++;
354 			*flags |= CSUM_L4_CALC;
355 			break;
356 		case ICE_RX_PTYPE_INNER_PROT_UDP:
357 			rxq->stats.cso[ICE_CSO_STAT_RX_UDP_ERR]++;
358 			*flags |= CSUM_L4_CALC;
359 			break;
360 		case ICE_RX_PTYPE_INNER_PROT_SCTP:
361 			rxq->stats.cso[ICE_CSO_STAT_RX_SCTP_ERR]++;
362 			*flags |= CSUM_L4_CALC;
363 			break;
364 		default:
365 			/*
366 			 * Hardware indicated L4 error, but this isn't one of
367 			 * the expected protocols.
368 			 */
369 			rxq->stats.cso[ICE_CSO_STAT_RX_L4_ERR]++;
370 		}
371 	}
372 }
373 
374 /**
375  * ice_ptype_to_hash - Convert packet type to a hash value
376  * @ptype: the packet type to convert
377  *
378  * Given the packet type, convert to a suitable hashtype to report to the
379  * upper stack via the iri_rsstype value of the if_rxd_info_t structure.
380  *
381  * If the hash type is unknown we'll report M_HASHTYPE_OPAQUE.
382  */
383 static inline int
384 ice_ptype_to_hash(u16 ptype)
385 {
386 	struct ice_rx_ptype_decoded decoded;
387 
388 	if (ptype >= ARRAY_SIZE(ice_ptype_lkup))
389 		return M_HASHTYPE_OPAQUE;
390 
391 	decoded = ice_decode_rx_desc_ptype(ptype);
392 
393 	if (!decoded.known)
394 		return M_HASHTYPE_OPAQUE;
395 
396 	if (decoded.outer_ip == ICE_RX_PTYPE_OUTER_L2)
397 		return M_HASHTYPE_OPAQUE;
398 
399 	/* Note: anything that gets to this point is IP */
400 	if (decoded.outer_ip_ver == ICE_RX_PTYPE_OUTER_IPV6) {
401 		switch (decoded.inner_prot) {
402 		case ICE_RX_PTYPE_INNER_PROT_TCP:
403 			return M_HASHTYPE_RSS_TCP_IPV6;
404 		case ICE_RX_PTYPE_INNER_PROT_UDP:
405 			return M_HASHTYPE_RSS_UDP_IPV6;
406 		default:
407 			return M_HASHTYPE_RSS_IPV6;
408 		}
409 	}
410 	if (decoded.outer_ip_ver == ICE_RX_PTYPE_OUTER_IPV4) {
411 		switch (decoded.inner_prot) {
412 		case ICE_RX_PTYPE_INNER_PROT_TCP:
413 			return M_HASHTYPE_RSS_TCP_IPV4;
414 		case ICE_RX_PTYPE_INNER_PROT_UDP:
415 			return M_HASHTYPE_RSS_UDP_IPV4;
416 		default:
417 			return M_HASHTYPE_RSS_IPV4;
418 		}
419 	}
420 
421 	/* We should never get here!! */
422 	return M_HASHTYPE_OPAQUE;
423 }
424 #endif
425