xref: /freebsd/sys/dev/ice/ice_iflib_txrx.c (revision 7ee6b0f125a092ed99d327bb8d608dd2ff77b7aa)
1 /* SPDX-License-Identifier: BSD-3-Clause */
2 /*  Copyright (c) 2023, 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_iflib_txrx.c
35  * @brief iflib Tx/Rx hotpath
36  *
37  * Main location for the iflib Tx/Rx hotpath implementation.
38  *
39  * Contains the implementation for the iflib function callbacks and the
40  * if_txrx ops structure.
41  */
42 
43 #include "ice_iflib.h"
44 
45 /* Tx/Rx hotpath utility functions */
46 #include "ice_common_txrx.h"
47 
48 /*
49  * iflib txrx method declarations
50  */
51 static int ice_ift_txd_encap(void *arg, if_pkt_info_t pi);
52 static int ice_ift_rxd_pkt_get(void *arg, if_rxd_info_t ri);
53 static void ice_ift_txd_flush(void *arg, uint16_t txqid, qidx_t pidx);
54 static int ice_ift_txd_credits_update(void *arg, uint16_t txqid, bool clear);
55 static int ice_ift_rxd_available(void *arg, uint16_t rxqid, qidx_t pidx, qidx_t budget);
56 static void ice_ift_rxd_flush(void *arg, uint16_t rxqid, uint8_t flidx, qidx_t pidx);
57 static void ice_ift_rxd_refill(void *arg, if_rxd_update_t iru);
58 static qidx_t ice_ift_queue_select(void *arg, struct mbuf *m, if_pkt_info_t pi);
59 
60 /* Macro to help extract the NIC mode flexible Rx descriptor fields from the
61  * advanced 32byte Rx descriptors.
62  */
63 #define RX_FLEX_NIC(desc, field) \
64 	(((struct ice_32b_rx_flex_desc_nic *)desc)->field)
65 
66 /**
67  * @var ice_txrx
68  * @brief Tx/Rx operations for the iflib stack
69  *
70  * Structure defining the Tx and Rx related operations that iflib can request
71  * the driver to perform. These are the main entry points for the hot path of
72  * the transmit and receive paths in the iflib driver.
73  */
74 struct if_txrx ice_txrx = {
75 	.ift_txd_encap = ice_ift_txd_encap,
76 	.ift_txd_flush = ice_ift_txd_flush,
77 	.ift_txd_credits_update = ice_ift_txd_credits_update,
78 	.ift_rxd_available = ice_ift_rxd_available,
79 	.ift_rxd_pkt_get = ice_ift_rxd_pkt_get,
80 	.ift_rxd_refill = ice_ift_rxd_refill,
81 	.ift_rxd_flush = ice_ift_rxd_flush,
82 	.ift_txq_select_v2 = ice_ift_queue_select,
83 };
84 
85 /**
86  * ice_ift_txd_encap - prepare Tx descriptors for a packet
87  * @arg: the iflib softc structure pointer
88  * @pi: packet info
89  *
90  * Prepares and encapsulates the given packet into into Tx descriptors, in
91  * preparation for sending to the transmit engine. Sets the necessary context
92  * descriptors for TSO and other offloads, and prepares the last descriptor
93  * for the writeback status.
94  *
95  * Return 0 on success, non-zero error code on failure.
96  */
97 static int
98 ice_ift_txd_encap(void *arg, if_pkt_info_t pi)
99 {
100 	struct ice_softc *sc = (struct ice_softc *)arg;
101 	struct ice_tx_queue *txq = &sc->pf_vsi.tx_queues[pi->ipi_qsidx];
102 	int nsegs = pi->ipi_nsegs;
103 	bus_dma_segment_t *segs = pi->ipi_segs;
104 	struct ice_tx_desc *txd = NULL;
105 	int i, j, mask, pidx_last;
106 	u32 cmd, off;
107 
108 	cmd = off = 0;
109 	i = pi->ipi_pidx;
110 
111 	/* Set up the TSO/CSUM offload */
112 	if (pi->ipi_csum_flags & ICE_CSUM_OFFLOAD) {
113 		/* Set up the TSO context descriptor if required */
114 		if (pi->ipi_csum_flags & CSUM_TSO) {
115 			if (ice_tso_detect_sparse(pi))
116 				return (EFBIG);
117 			i = ice_tso_setup(txq, pi);
118 		}
119 		ice_tx_setup_offload(txq, pi, &cmd, &off);
120 	}
121 	if (pi->ipi_mflags & M_VLANTAG)
122 		cmd |= ICE_TX_DESC_CMD_IL2TAG1;
123 
124 	mask = txq->desc_count - 1;
125 	for (j = 0; j < nsegs; j++) {
126 		bus_size_t seglen;
127 
128 		txd = &txq->tx_base[i];
129 		seglen = segs[j].ds_len;
130 
131 		txd->buf_addr = htole64(segs[j].ds_addr);
132 		txd->cmd_type_offset_bsz =
133 		    htole64(ICE_TX_DESC_DTYPE_DATA
134 		    | ((u64)cmd  << ICE_TXD_QW1_CMD_S)
135 		    | ((u64)off << ICE_TXD_QW1_OFFSET_S)
136 		    | ((u64)seglen  << ICE_TXD_QW1_TX_BUF_SZ_S)
137 		    | ((u64)htole16(pi->ipi_vtag) << ICE_TXD_QW1_L2TAG1_S));
138 
139 		txq->stats.tx_bytes += seglen;
140 		pidx_last = i;
141 		i = (i+1) & mask;
142 	}
143 
144 	/* Set the last descriptor for report */
145 #define ICE_TXD_CMD (ICE_TX_DESC_CMD_EOP | ICE_TX_DESC_CMD_RS)
146 	txd->cmd_type_offset_bsz |=
147 	    htole64(((u64)ICE_TXD_CMD << ICE_TXD_QW1_CMD_S));
148 
149 	/* Add to report status array */
150 	txq->tx_rsq[txq->tx_rs_pidx] = pidx_last;
151 	txq->tx_rs_pidx = (txq->tx_rs_pidx+1) & mask;
152 	MPASS(txq->tx_rs_pidx != txq->tx_rs_cidx);
153 
154 	pi->ipi_new_pidx = i;
155 
156 	++txq->stats.tx_packets;
157 	return (0);
158 }
159 
160 /**
161  * ice_ift_txd_flush - Flush Tx descriptors to hardware
162  * @arg: device specific softc pointer
163  * @txqid: the Tx queue to flush
164  * @pidx: descriptor index to advance tail to
165  *
166  * Advance the Transmit Descriptor Tail (TDT). This indicates to hardware that
167  * frames are available for transmit.
168  */
169 static void
170 ice_ift_txd_flush(void *arg, uint16_t txqid, qidx_t pidx)
171 {
172 	struct ice_softc *sc = (struct ice_softc *)arg;
173 	struct ice_tx_queue *txq = &sc->pf_vsi.tx_queues[txqid];
174 	struct ice_hw *hw = &sc->hw;
175 
176 	wr32(hw, txq->tail, pidx);
177 }
178 
179 /**
180  * ice_ift_txd_credits_update - cleanup Tx descriptors
181  * @arg: device private softc
182  * @txqid: the Tx queue to update
183  * @clear: if false, only report, do not actually clean
184  *
185  * If clear is false, iflib is asking if we *could* clean up any Tx
186  * descriptors.
187  *
188  * If clear is true, iflib is requesting to cleanup and reclaim used Tx
189  * descriptors.
190  */
191 static int
192 ice_ift_txd_credits_update(void *arg, uint16_t txqid, bool clear)
193 {
194 	struct ice_softc *sc = (struct ice_softc *)arg;
195 	struct ice_tx_queue *txq = &sc->pf_vsi.tx_queues[txqid];
196 
197 	qidx_t processed = 0;
198 	qidx_t cur, prev, ntxd, rs_cidx;
199 	int32_t delta;
200 	bool is_done;
201 
202 	rs_cidx = txq->tx_rs_cidx;
203 	if (rs_cidx == txq->tx_rs_pidx)
204 		return (0);
205 	cur = txq->tx_rsq[rs_cidx];
206 	MPASS(cur != QIDX_INVALID);
207 	is_done = ice_is_tx_desc_done(&txq->tx_base[cur]);
208 
209 	if (!is_done)
210 		return (0);
211 	else if (clear == false)
212 		return (1);
213 
214 	prev = txq->tx_cidx_processed;
215 	ntxd = txq->desc_count;
216 	do {
217 		MPASS(prev != cur);
218 		delta = (int32_t)cur - (int32_t)prev;
219 		if (delta < 0)
220 			delta += ntxd;
221 		MPASS(delta > 0);
222 		processed += delta;
223 		prev = cur;
224 		rs_cidx = (rs_cidx + 1) & (ntxd-1);
225 		if (rs_cidx == txq->tx_rs_pidx)
226 			break;
227 		cur = txq->tx_rsq[rs_cidx];
228 		MPASS(cur != QIDX_INVALID);
229 		is_done = ice_is_tx_desc_done(&txq->tx_base[cur]);
230 	} while (is_done);
231 
232 	txq->tx_rs_cidx = rs_cidx;
233 	txq->tx_cidx_processed = prev;
234 
235 	return (processed);
236 }
237 
238 /**
239  * ice_ift_rxd_available - Return number of available Rx packets
240  * @arg: device private softc
241  * @rxqid: the Rx queue id
242  * @pidx: descriptor start point
243  * @budget: maximum Rx budget
244  *
245  * Determines how many Rx packets are available on the queue, up to a maximum
246  * of the given budget.
247  */
248 static int
249 ice_ift_rxd_available(void *arg, uint16_t rxqid, qidx_t pidx, qidx_t budget)
250 {
251 	struct ice_softc *sc = (struct ice_softc *)arg;
252 	struct ice_rx_queue *rxq = &sc->pf_vsi.rx_queues[rxqid];
253 	union ice_32b_rx_flex_desc *rxd;
254 	uint16_t status0;
255 	int cnt, i, nrxd;
256 
257 	nrxd = rxq->desc_count;
258 
259 	for (cnt = 0, i = pidx; cnt < nrxd - 1 && cnt < budget;) {
260 		rxd = &rxq->rx_base[i];
261 		status0 = le16toh(rxd->wb.status_error0);
262 
263 		if ((status0 & BIT(ICE_RX_FLEX_DESC_STATUS0_DD_S)) == 0)
264 			break;
265 		if (++i == nrxd)
266 			i = 0;
267 		if (status0 & BIT(ICE_RX_FLEX_DESC_STATUS0_EOF_S))
268 			cnt++;
269 	}
270 
271 	return (cnt);
272 }
273 
274 /**
275  * ice_ift_rxd_pkt_get - Called by iflib to send data to upper layer
276  * @arg: device specific softc
277  * @ri: receive packet info
278  *
279  * This function is called by iflib, and executes in ithread context. It is
280  * called by iflib to obtain data which has been DMA'ed into host memory.
281  * Returns zero on success, and EBADMSG on failure.
282  */
283 static int
284 ice_ift_rxd_pkt_get(void *arg, if_rxd_info_t ri)
285 {
286 	struct ice_softc *sc = (struct ice_softc *)arg;
287 	if_softc_ctx_t scctx = sc->scctx;
288 	struct ice_rx_queue *rxq = &sc->pf_vsi.rx_queues[ri->iri_qsidx];
289 	union ice_32b_rx_flex_desc *cur;
290 	u16 status0, plen, ptype;
291 	bool eop;
292 	size_t cidx;
293 	int i;
294 
295 	cidx = ri->iri_cidx;
296 	i = 0;
297 	do {
298 		/* 5 descriptor receive limit */
299 		MPASS(i < ICE_MAX_RX_SEGS);
300 
301 		cur = &rxq->rx_base[cidx];
302 		status0 = le16toh(cur->wb.status_error0);
303 		plen = le16toh(cur->wb.pkt_len) &
304 			ICE_RX_FLX_DESC_PKT_LEN_M;
305 
306 		/* we should never be called without a valid descriptor */
307 		MPASS((status0 & BIT(ICE_RX_FLEX_DESC_STATUS0_DD_S)) != 0);
308 
309 		ri->iri_len += plen;
310 
311 		cur->wb.status_error0 = 0;
312 		eop = (status0 & BIT(ICE_RX_FLEX_DESC_STATUS0_EOF_S));
313 
314 		ri->iri_frags[i].irf_flid = 0;
315 		ri->iri_frags[i].irf_idx = cidx;
316 		ri->iri_frags[i].irf_len = plen;
317 		if (++cidx == rxq->desc_count)
318 			cidx = 0;
319 		i++;
320 	} while (!eop);
321 
322 	/* End of Packet reached; cur is eop/last descriptor */
323 
324 	/* Make sure packets with bad L2 values are discarded.
325 	 * This bit is only valid in the last descriptor.
326 	 */
327 	if (status0 & BIT(ICE_RX_FLEX_DESC_STATUS0_RXE_S)) {
328 		rxq->stats.desc_errs++;
329 		return (EBADMSG);
330 	}
331 
332 	/* Get VLAN tag information if one is in descriptor */
333 	if (status0 & BIT(ICE_RX_FLEX_DESC_STATUS0_L2TAG1P_S)) {
334 		ri->iri_vtag = le16toh(cur->wb.l2tag1);
335 		ri->iri_flags |= M_VLANTAG;
336 	}
337 
338 	/* Capture soft statistics for this Rx queue */
339 	rxq->stats.rx_packets++;
340 	rxq->stats.rx_bytes += ri->iri_len;
341 
342 	/* Get packet type and set checksum flags */
343 	ptype = le16toh(cur->wb.ptype_flex_flags0) &
344 		ICE_RX_FLEX_DESC_PTYPE_M;
345 	if ((scctx->isc_capenable & IFCAP_RXCSUM) != 0)
346 		ice_rx_checksum(rxq, &ri->iri_csum_flags,
347 				&ri->iri_csum_data, status0, ptype);
348 
349 	/* Set remaining iflib RX descriptor info fields */
350 	ri->iri_flowid = le32toh(RX_FLEX_NIC(&cur->wb, rss_hash));
351 	ri->iri_rsstype = ice_ptype_to_hash(ptype);
352 	ri->iri_nfrags = i;
353 	return (0);
354 }
355 
356 /**
357  * ice_ift_rxd_refill - Prepare Rx descriptors for re-use by hardware
358  * @arg: device specific softc structure
359  * @iru: the Rx descriptor update structure
360  *
361  * Update the Rx descriptor indices for a given queue, assigning new physical
362  * addresses to the descriptors, preparing them for re-use by the hardware.
363  */
364 static void
365 ice_ift_rxd_refill(void *arg, if_rxd_update_t iru)
366 {
367 	struct ice_softc *sc = (struct ice_softc *)arg;
368 	struct ice_rx_queue *rxq;
369 	uint32_t next_pidx;
370 	int i;
371 	uint64_t *paddrs;
372 	uint32_t pidx;
373 	uint16_t qsidx, count;
374 
375 	paddrs = iru->iru_paddrs;
376 	pidx = iru->iru_pidx;
377 	qsidx = iru->iru_qsidx;
378 	count = iru->iru_count;
379 
380 	rxq = &(sc->pf_vsi.rx_queues[qsidx]);
381 
382 	for (i = 0, next_pidx = pidx; i < count; i++) {
383 		rxq->rx_base[next_pidx].read.pkt_addr = htole64(paddrs[i]);
384 		if (++next_pidx == (uint32_t)rxq->desc_count)
385 			next_pidx = 0;
386 	}
387 }
388 
389 /**
390  * ice_ift_rxd_flush - Flush Rx descriptors to hardware
391  * @arg: device specific softc pointer
392  * @rxqid: the Rx queue to flush
393  * @flidx: unused parameter
394  * @pidx: descriptor index to advance tail to
395  *
396  * Advance the Receive Descriptor Tail (RDT). This indicates to hardware that
397  * software is done with the descriptor and it can be recycled.
398  */
399 static void
400 ice_ift_rxd_flush(void *arg, uint16_t rxqid, uint8_t flidx __unused,
401 		  qidx_t pidx)
402 {
403 	struct ice_softc *sc = (struct ice_softc *)arg;
404 	struct ice_rx_queue *rxq = &sc->pf_vsi.rx_queues[rxqid];
405 	struct ice_hw *hw = &sc->hw;
406 
407 	wr32(hw, rxq->tail, pidx);
408 }
409 
410 static qidx_t
411 ice_ift_queue_select(void *arg, struct mbuf *m, if_pkt_info_t pi)
412 {
413 	struct ice_softc *sc = (struct ice_softc *)arg;
414 	struct ice_dcbx_cfg *local_dcbx_cfg;
415 	struct ice_vsi *vsi = &sc->pf_vsi;
416 	u16 tc_base_queue, tc_qcount;
417 	u8 up, tc;
418 
419 #ifdef ALTQ
420 	/* Included to match default iflib behavior */
421 	/* Only go out on default queue if ALTQ is enabled */
422 	struct ifnet *ifp = (struct ifnet *)iflib_get_ifp(sc->ctx);
423 	if (if_altq_is_enabled(ifp))
424 		return (0);
425 #endif
426 
427 	if (!ice_test_state(&sc->state, ICE_STATE_MULTIPLE_TCS)) {
428 		if (M_HASHTYPE_GET(m)) {
429 			/* Default iflib queue selection method */
430 			return (m->m_pkthdr.flowid % sc->pf_vsi.num_tx_queues);
431 		} else
432 			return (0);
433 	}
434 
435 	/* Use default TC unless overridden later */
436 	tc = 0; /* XXX: Get default TC for traffic if >1 TC? */
437 
438 	local_dcbx_cfg = &sc->hw.port_info->qos_cfg.local_dcbx_cfg;
439 
440 #if defined(INET) || defined(INET6)
441 	if ((local_dcbx_cfg->pfc_mode == ICE_QOS_MODE_DSCP) &&
442 	    (pi->ipi_flags & (IPI_TX_IPV4 | IPI_TX_IPV6))) {
443 		u8 dscp_val = pi->ipi_ip_tos >> 2;
444 		tc = local_dcbx_cfg->dscp_map[dscp_val];
445 	} else
446 #endif /* defined(INET) || defined(INET6) */
447 	if (m->m_flags & M_VLANTAG) { /* ICE_QOS_MODE_VLAN */
448 		up = EVL_PRIOFTAG(m->m_pkthdr.ether_vtag);
449 		tc = local_dcbx_cfg->etscfg.prio_table[up];
450 	}
451 
452 	tc_base_queue = vsi->tc_info[tc].qoffset;
453 	tc_qcount = vsi->tc_info[tc].qcount_tx;
454 
455 	if (M_HASHTYPE_GET(m))
456 		return ((m->m_pkthdr.flowid % tc_qcount) + tc_base_queue);
457 	else
458 		return (tc_base_queue);
459 }
460