xref: /freebsd/sys/dev/e1000/em_txrx.c (revision 2a7af8ebcf90af83f01f9523b25bfc6e58c7b809)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2016 Nicole Graziano <nicole@nextbsd.org>
5  * Copyright (c) 2017 Matthew Macy <mmacy@mattmacy.io>
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 
30 #include "if_em.h"
31 
32 #include <net/rss_config.h>
33 #include <netinet/in_rss.h>
34 
35 #ifdef VERBOSE_DEBUG
36 #define DPRINTF device_printf
37 #else
38 #define DPRINTF(...)
39 #endif
40 
41 /*********************************************************************
42  *  Local Function prototypes
43  *********************************************************************/
44 static int em_tso_setup(struct e1000_softc *, if_pkt_info_t, uint32_t *,
45     uint32_t *);
46 static int em_transmit_checksum_setup(struct e1000_softc *, if_pkt_info_t,
47     uint32_t *, uint32_t *);
48 static int em_isc_txd_encap(void *, if_pkt_info_t);
49 static void em_isc_txd_flush(void *, uint16_t, qidx_t);
50 static int em_isc_txd_credits_update(void *, uint16_t, bool);
51 static void em_isc_rxd_refill(void *, if_rxd_update_t);
52 static void em_isc_rxd_flush(void *, uint16_t, uint8_t, qidx_t);
53 static int em_isc_rxd_available(void *, uint16_t, qidx_t, qidx_t);
54 static int em_isc_rxd_pkt_get(void *, if_rxd_info_t);
55 
56 static void lem_isc_rxd_refill(void *, if_rxd_update_t);
57 
58 static int lem_isc_rxd_available(void *, uint16_t, qidx_t, qidx_t);
59 static int lem_isc_rxd_pkt_get(void *, if_rxd_info_t);
60 
61 static void em_receive_checksum(uint16_t, uint8_t, if_rxd_info_t);
62 static int em_determine_rsstype(uint32_t);
63 extern int em_intr(void *);
64 
65 struct if_txrx em_txrx = {
66 	.ift_txd_encap = em_isc_txd_encap,
67 	.ift_txd_flush = em_isc_txd_flush,
68 	.ift_txd_credits_update = em_isc_txd_credits_update,
69 	.ift_rxd_available = em_isc_rxd_available,
70 	.ift_rxd_pkt_get = em_isc_rxd_pkt_get,
71 	.ift_rxd_refill = em_isc_rxd_refill,
72 	.ift_rxd_flush = em_isc_rxd_flush,
73 	.ift_legacy_intr = em_intr
74 };
75 
76 struct if_txrx lem_txrx = {
77 	.ift_txd_encap = em_isc_txd_encap,
78 	.ift_txd_flush = em_isc_txd_flush,
79 	.ift_txd_credits_update = em_isc_txd_credits_update,
80 	.ift_rxd_available = lem_isc_rxd_available,
81 	.ift_rxd_pkt_get = lem_isc_rxd_pkt_get,
82 	.ift_rxd_refill = lem_isc_rxd_refill,
83 	.ift_rxd_flush = em_isc_rxd_flush,
84 	.ift_legacy_intr = em_intr
85 };
86 
87 extern if_shared_ctx_t em_sctx;
88 
89 void
em_dump_rs(struct e1000_softc * sc)90 em_dump_rs(struct e1000_softc *sc)
91 {
92 	if_softc_ctx_t scctx = sc->shared;
93 	struct em_tx_queue *que;
94 	struct tx_ring *txr;
95 	qidx_t i, ntxd, qid, cur;
96 	int16_t rs_cidx;
97 	uint8_t status;
98 
99 	if (sc->tx_queues == NULL) {
100 		device_printf(sc->dev, "queue state is unavailable\n");
101 		return;
102 	}
103 
104 	printf("\n");
105 	ntxd = scctx->isc_ntxd[0];
106 	for (qid = 0; qid < sc->tx_num_queues; qid++) {
107 		que = &sc->tx_queues[qid];
108 		txr =  &que->txr;
109 		rs_cidx = txr->tx_rs_cidx;
110 		if (rs_cidx != txr->tx_rs_pidx) {
111 			cur = txr->tx_rsq[rs_cidx];
112 			status = txr->tx_base[cur].upper.fields.status;
113 			if (!(status & E1000_TXD_STAT_DD))
114 				printf("qid[%d]->tx_rsq[%d]: %d clear ",
115 				    qid, rs_cidx, cur);
116 		} else {
117 			rs_cidx = (rs_cidx-1)&(ntxd-1);
118 			cur = txr->tx_rsq[rs_cidx];
119 			printf("qid[%d]->tx_rsq[rs_cidx-1=%d]: %d  ",
120 			    qid, rs_cidx, cur);
121 		}
122 		printf("cidx_prev=%d rs_pidx=%d ",txr->tx_cidx_processed,
123 		    txr->tx_rs_pidx);
124 		for (i = 0; i < ntxd; i++) {
125 			if (txr->tx_base[i].upper.fields.status &
126 			    E1000_TXD_STAT_DD)
127 				printf("%d set ", i);
128 		}
129 		printf("\n");
130 	}
131 }
132 
133 /**********************************************************************
134  *
135  *  Setup work for hardware segmentation offload (TSO) on
136  *  adapters using advanced tx descriptors
137  *
138  **********************************************************************/
139 static int
em_tso_setup(struct e1000_softc * sc,if_pkt_info_t pi,uint32_t * txd_upper,uint32_t * txd_lower)140 em_tso_setup(struct e1000_softc *sc, if_pkt_info_t pi, uint32_t *txd_upper,
141     uint32_t *txd_lower)
142 {
143 	if_softc_ctx_t scctx = sc->shared;
144 	struct em_tx_queue *que = &sc->tx_queues[pi->ipi_qsidx];
145 	struct tx_ring *txr = &que->txr;
146 	struct e1000_context_desc *TXD;
147 	int cur, hdr_len;
148 	uint32_t cmd_type_len;
149 
150 	hdr_len = pi->ipi_ehdrlen + pi->ipi_ip_hlen + pi->ipi_tcp_hlen;
151 	*txd_lower = (E1000_TXD_CMD_DEXT |	/* Extended descr type */
152 	    E1000_TXD_DTYP_D |			/* Data descr type */
153 	    E1000_TXD_CMD_TSE);			/* Do TSE on this packet */
154 
155 	cur = pi->ipi_pidx;
156 	TXD = (struct e1000_context_desc *)&txr->tx_base[cur];
157 
158 	/*
159 	 * ipcss - Start offset for header checksum calculation.
160 	 * ipcse - End offset for header checksum calculation.
161 	 * ipcso - Offset of place to put the checksum.
162 	 */
163 	switch(pi->ipi_etype) {
164 	case ETHERTYPE_IP:
165 		/* IP and/or TCP header checksum calculation and insertion. */
166 		*txd_upper =
167 		    (E1000_TXD_POPTS_IXSM | E1000_TXD_POPTS_TXSM) << 8;
168 
169 		TXD->lower_setup.ip_fields.ipcse =
170 		    htole16(pi->ipi_ehdrlen + pi->ipi_ip_hlen - 1);
171 		break;
172 	case ETHERTYPE_IPV6:
173 		/* TCP header checksum calculation and insertion. */
174 		*txd_upper = E1000_TXD_POPTS_TXSM << 8;
175 
176 		TXD->lower_setup.ip_fields.ipcse = htole16(0);
177 		break;
178 	default:
179 		break;
180 	}
181 	TXD->lower_setup.ip_fields.ipcss = pi->ipi_ehdrlen;
182 	TXD->lower_setup.ip_fields.ipcso =
183 	    pi->ipi_ehdrlen + offsetof(struct ip, ip_sum);
184 
185 	/*
186 	 * tucss - Start offset for payload checksum calculation.
187 	 * tucse - End offset for payload checksum calculation.
188 	 * tucso - Offset of place to put the checksum.
189 	 */
190 	TXD->upper_setup.tcp_fields.tucss = pi->ipi_ehdrlen + pi->ipi_ip_hlen;
191 	TXD->upper_setup.tcp_fields.tucse = 0;
192 	TXD->upper_setup.tcp_fields.tucso =
193 	    pi->ipi_ehdrlen + pi->ipi_ip_hlen +
194 	    offsetof(struct tcphdr, th_sum);
195 
196 	/*
197 	 * Payload size per packet w/o any headers.
198 	 * Length of all headers up to payload.
199 	 */
200 	TXD->tcp_seg_setup.fields.mss = htole16(pi->ipi_tso_segsz);
201 	TXD->tcp_seg_setup.fields.hdr_len = hdr_len;
202 
203 	/*
204 	 * "PCI/PCI-X SDM 4.0" page 45, and "PCIe GbE SDM 2.5" page 63
205 	 * - Set up basic TUCMDs
206 	 * - For others IP bit on indicates IPv4, while off indicates IPv6
207 	*/
208 	cmd_type_len = sc->txd_cmd |
209 	    E1000_TXD_CMD_DEXT | /* Extended descr */
210 	    E1000_TXD_CMD_TSE |  /* TSE context */
211 	    E1000_TXD_CMD_TCP;   /* Do TCP checksum */
212 	if (pi->ipi_etype == ETHERTYPE_IP)
213 		cmd_type_len |= E1000_TXD_CMD_IP;
214 	TXD->cmd_and_length = htole32(cmd_type_len |
215 	    (pi->ipi_len - hdr_len)); /* Total len */
216 
217 	/*
218 	 * 82540EP and 82545GM erratum 3 and 82546GB erratum 1 require
219 	 * a second descriptor for the final four bytes of an LSO packet.
220 	 */
221 	txr->tx_tso = sc->hw.mac.type < e1000_82571;
222 
223 	if (++cur == scctx->isc_ntxd[0]) {
224 		cur = 0;
225 	}
226 	DPRINTF(iflib_get_dev(sc->ctx), "%s: pidx: %d cur: %d\n",
227 	    __FUNCTION__, pi->ipi_pidx, cur);
228 	return (cur);
229 }
230 
231 /*********************************************************************
232  *  The offload context is protocol specific (TCP/UDP) and thus
233  *  only needs to be set when the protocol changes. The occasion
234  *  of a context change can be a performance detriment, and
235  *  might be better just disabled. The reason arises in the way
236  *  in which the controller supports pipelined requests from the
237  *  Tx data DMA. Up to four requests can be pipelined, and they may
238  *  belong to the same packet or to multiple packets. However all
239  *  requests for one packet are issued before a request is issued
240  *  for a subsequent packet and if a request for the next packet
241  *  requires a context change, that request will be stalled
242  *  until the previous request completes. This means setting up
243  *  a new context effectively disables pipelined Tx data DMA which
244  *  in turn greatly slow down performance to send small sized
245  *  frames.
246  **********************************************************************/
247 #define DONT_FORCE_CTX 1
248 
249 static int
em_transmit_checksum_setup(struct e1000_softc * sc,if_pkt_info_t pi,uint32_t * txd_upper,uint32_t * txd_lower)250 em_transmit_checksum_setup(struct e1000_softc *sc, if_pkt_info_t pi,
251     uint32_t *txd_upper, uint32_t *txd_lower)
252 {
253 	struct e1000_context_desc *TXD = NULL;
254 	if_softc_ctx_t scctx = sc->shared;
255 	struct em_tx_queue *que = &sc->tx_queues[pi->ipi_qsidx];
256 	struct tx_ring *txr = &que->txr;
257 	int csum_flags = pi->ipi_csum_flags;
258 	int cur, hdr_len;
259 	uint32_t cmd;
260 
261 	cur = pi->ipi_pidx;
262 	hdr_len = pi->ipi_ehdrlen + pi->ipi_ip_hlen;
263 	cmd = sc->txd_cmd;
264 
265 	/*
266 	 * The 82574L can only remember the *last* context used
267 	 * regardless of queue that it was use for.  We cannot reuse
268 	 * contexts on this hardware platform and must generate a new
269 	 * context every time.  82574L hardware spec, section 7.2.6,
270 	 * second note.
271 	 */
272 	if (DONT_FORCE_CTX &&
273 	    sc->tx_num_queues == 1 &&
274 	    txr->csum_lhlen == pi->ipi_ehdrlen &&
275 	    txr->csum_iphlen == pi->ipi_ip_hlen &&
276 	    txr->csum_flags == csum_flags) {
277 		/*
278 		 * Same csum offload context as the previous packets;
279 		 * just return.
280 		 */
281 		*txd_upper = txr->csum_txd_upper;
282 		*txd_lower = txr->csum_txd_lower;
283 		return (cur);
284 	}
285 
286 	TXD = (struct e1000_context_desc *)&txr->tx_base[cur];
287 	/*
288 	 * ipcss - Start offset for header checksum calculation.
289 	 * ipcse - End offset for header checksum calculation.
290 	 * ipcso - Offset of place to put the checksum.
291 	 *
292 	 * We set ipcsX values regardless of IP version to work around HW
293 	 * issues and ipcse must be 0 for IPv6 per "PCIe GbE SDM 2.5" page 61.
294 	 * IXSM controls whether it's inserted.
295 	 */
296 	TXD->lower_setup.ip_fields.ipcss = pi->ipi_ehdrlen;
297 	TXD->lower_setup.ip_fields.ipcso = pi->ipi_ehdrlen +
298 	    offsetof(struct ip, ip_sum);
299 	if (csum_flags & CSUM_IP) {
300 		*txd_upper |= E1000_TXD_POPTS_IXSM << 8;
301 		TXD->lower_setup.ip_fields.ipcse = htole16(hdr_len - 1);
302 		cmd |= E1000_TXD_CMD_IP;
303 	} else if (csum_flags & (CSUM_IP6_TCP | CSUM_IP6_UDP))
304 		TXD->lower_setup.ip_fields.ipcse = htole16(0);
305 
306 	/*
307 	 * tucss - Start offset for payload checksum calculation.
308 	 * tucse - End offset for payload checksum calculation.
309 	 * tucso - Offset of place to put the checksum.
310 	 */
311 	if (csum_flags & (CSUM_TCP | CSUM_UDP | CSUM_IP6_TCP |
312 	    CSUM_IP6_UDP)) {
313 		uint8_t tucso;
314 
315 		*txd_upper |= E1000_TXD_POPTS_TXSM << 8;
316 		*txd_lower = E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
317 
318 		if (csum_flags & (CSUM_TCP | CSUM_IP6_TCP)) {
319 			tucso = hdr_len + offsetof(struct tcphdr, th_sum);
320 			cmd |= E1000_TXD_CMD_TCP;
321 		} else
322 			tucso = hdr_len + offsetof(struct udphdr, uh_sum);
323 		TXD->upper_setup.tcp_fields.tucss = hdr_len;
324 		TXD->upper_setup.tcp_fields.tucse = htole16(0);
325 		TXD->upper_setup.tcp_fields.tucso = tucso;
326 	}
327 
328 	txr->csum_lhlen = pi->ipi_ehdrlen;
329 	txr->csum_iphlen = pi->ipi_ip_hlen;
330 	txr->csum_flags = csum_flags;
331 	txr->csum_txd_upper = *txd_upper;
332 	txr->csum_txd_lower = *txd_lower;
333 
334 	TXD->tcp_seg_setup.data = htole32(0);
335 	TXD->cmd_and_length =
336 		htole32(E1000_TXD_CMD_IFCS | E1000_TXD_CMD_DEXT | cmd);
337 
338 	if (++cur == scctx->isc_ntxd[0]) {
339 		cur = 0;
340 	}
341 	DPRINTF(iflib_get_dev(sc->ctx),
342 	    "checksum_setup csum_flags=%x txd_upper=%x txd_lower=%x"
343 	    " hdr_len=%d cmd=%x\n",
344 	    csum_flags, *txd_upper, *txd_lower, hdr_len, cmd);
345 	return (cur);
346 }
347 
348 #define TSO_WORKAROUND 4 /* TSO sentinel descriptor */
349 
350 static int
em_isc_txd_encap(void * arg,if_pkt_info_t pi)351 em_isc_txd_encap(void *arg, if_pkt_info_t pi)
352 {
353 	struct e1000_softc *sc = arg;
354 	if_softc_ctx_t scctx = sc->shared;
355 	struct em_tx_queue *que = &sc->tx_queues[pi->ipi_qsidx];
356 	struct tx_ring *txr = &que->txr;
357 	bus_dma_segment_t *segs = pi->ipi_segs;
358 	int nsegs = pi->ipi_nsegs;
359 	int csum_flags = pi->ipi_csum_flags;
360 	int i, j, first, pidx_last;
361 	uint32_t txd_flags, txd_upper = 0, txd_lower = 0;
362 
363 	struct e1000_tx_desc *ctxd = NULL;
364 	bool do_tso, tso_desc;
365 	qidx_t ntxd;
366 
367 	txd_flags = pi->ipi_flags & IPI_TX_INTR ? E1000_TXD_CMD_RS : 0;
368 	i = first = pi->ipi_pidx;
369 	do_tso = (csum_flags & CSUM_TSO);
370 	tso_desc = false;
371 	ntxd = scctx->isc_ntxd[0];
372 	/*
373 	 * TSO Hardware workaround, if this packet is not
374 	 * TSO, and is only a single descriptor long, and
375 	 * it follows a TSO burst, then we need to add a
376 	 * sentinel descriptor to prevent premature writeback.
377 	 */
378 	if ((!do_tso) && (txr->tx_tso == true)) {
379 		if (nsegs == 1)
380 			tso_desc = true;
381 		txr->tx_tso = false;
382 	}
383 
384 	/* Do hardware assists */
385 	if (do_tso) {
386 		i = em_tso_setup(sc, pi, &txd_upper, &txd_lower);
387 		tso_desc = txr->tx_tso;
388 	} else if (csum_flags & EM_CSUM_OFFLOAD) {
389 		i = em_transmit_checksum_setup(sc, pi, &txd_upper,
390 		    &txd_lower);
391 	}
392 
393 	if (pi->ipi_mflags & M_VLANTAG) {
394 		/* Set the vlan id. */
395 		txd_upper |= htole16(pi->ipi_vtag) << 16;
396 		/* Tell hardware to add tag */
397 		txd_lower |= htole32(E1000_TXD_CMD_VLE);
398 	}
399 
400 	DPRINTF(iflib_get_dev(sc->ctx),
401 	    "encap: set up tx: nsegs=%d first=%d i=%d\n", nsegs, first, i);
402 	/* XXX sc->pcix_82544 -- lem_fill_descriptors */
403 
404 	/* Set up our transmit descriptors */
405 	for (j = 0; j < nsegs; j++) {
406 		bus_size_t seg_len;
407 		bus_addr_t seg_addr;
408 		uint32_t cmd;
409 
410 		ctxd = &txr->tx_base[i];
411 		seg_addr = segs[j].ds_addr;
412 		seg_len = segs[j].ds_len;
413 		cmd = E1000_TXD_CMD_IFCS | sc->txd_cmd;
414 
415 		/*
416 		 * TSO Workaround:
417 		 * If this is the last descriptor, we want to
418 		 * split it so we have a small final sentinel
419 		 */
420 		if (tso_desc && (j == (nsegs - 1)) && (seg_len > 8)) {
421 			seg_len -= TSO_WORKAROUND;
422 			ctxd->buffer_addr = htole64(seg_addr);
423 			ctxd->lower.data = htole32(cmd | txd_lower | seg_len);
424 			ctxd->upper.data = htole32(txd_upper);
425 
426 			if (++i == scctx->isc_ntxd[0])
427 				i = 0;
428 
429 			/* Now make the sentinel */
430 			ctxd = &txr->tx_base[i];
431 			ctxd->buffer_addr = htole64(seg_addr + seg_len);
432 			ctxd->lower.data =
433 			    htole32(cmd | txd_lower | TSO_WORKAROUND);
434 			ctxd->upper.data = htole32(txd_upper);
435 			pidx_last = i;
436 			if (++i == scctx->isc_ntxd[0])
437 				i = 0;
438 			DPRINTF(iflib_get_dev(sc->ctx),
439 			    "TSO path pidx_last=%d i=%d ntxd[0]=%d\n",
440 			    pidx_last, i, scctx->isc_ntxd[0]);
441 		} else {
442 			ctxd->buffer_addr = htole64(seg_addr);
443 			ctxd->lower.data = htole32(cmd | txd_lower | seg_len);
444 			ctxd->upper.data = htole32(txd_upper);
445 			pidx_last = i;
446 			if (++i == scctx->isc_ntxd[0])
447 				i = 0;
448 			DPRINTF(iflib_get_dev(sc->ctx),
449 			    "pidx_last=%d i=%d ntxd[0]=%d\n",
450 			    pidx_last, i, scctx->isc_ntxd[0]);
451 		}
452 	}
453 
454 	/*
455 	 * Last Descriptor of Packet
456 	 * needs End Of Packet (EOP)
457 	 * and Report Status (RS)
458 	 */
459 	if (txd_flags && nsegs) {
460 		txr->tx_rsq[txr->tx_rs_pidx] = pidx_last;
461 		DPRINTF(iflib_get_dev(sc->ctx),
462 		    "setting to RS on %d rs_pidx %d first: %d\n",
463 		    pidx_last, txr->tx_rs_pidx, first);
464 		txr->tx_rs_pidx = (txr->tx_rs_pidx+1) & (ntxd-1);
465 		MPASS(txr->tx_rs_pidx != txr->tx_rs_cidx);
466 	}
467 	ctxd->lower.data |= htole32(E1000_TXD_CMD_EOP | txd_flags);
468 	DPRINTF(iflib_get_dev(sc->ctx),
469 	    "tx_buffers[%d]->eop = %d ipi_new_pidx=%d\n",
470 	    first, pidx_last, i);
471 	pi->ipi_new_pidx = i;
472 
473 	/*
474 	 * Sent data accounting for AIM.  For TSO, ipi_len is the whole
475 	 * unsegmented payload, which is not a size the moderation
476 	 * calculation can use.  Count the segments the hardware will put on
477 	 * the wire and the header each of them carries, so that the average
478 	 * it sees is a wire packet.
479 	 */
480 	if (do_tso && pi->ipi_tso_segsz != 0) {
481 		u32 hdrlen, segs;
482 
483 		hdrlen = pi->ipi_ehdrlen + pi->ipi_ip_hlen + pi->ipi_tcp_hlen;
484 		if (pi->ipi_len > hdrlen) {
485 			segs = howmany(pi->ipi_len - hdrlen, pi->ipi_tso_segsz);
486 			txr->tx_bytes += pi->ipi_len + (segs - 1) * hdrlen;
487 			txr->tx_packets += segs;
488 			return (0);
489 		}
490 	}
491 
492 	txr->tx_bytes += pi->ipi_len;
493 	++txr->tx_packets;
494 	return (0);
495 }
496 
497 static void
em_isc_txd_flush(void * arg,uint16_t txqid,qidx_t pidx)498 em_isc_txd_flush(void *arg, uint16_t txqid, qidx_t pidx)
499 {
500 	struct e1000_softc *sc = arg;
501 	struct em_tx_queue *que = &sc->tx_queues[txqid];
502 	struct tx_ring *txr = &que->txr;
503 
504 	E1000_WRITE_REG(&sc->hw, E1000_TDT(txr->me), pidx);
505 	if (sc->hw.mac.type >= e1000_82540)
506 		em_aim_publish(txr);
507 }
508 
509 static int
em_isc_txd_credits_update(void * arg,uint16_t txqid,bool clear)510 em_isc_txd_credits_update(void *arg, uint16_t txqid, bool clear)
511 {
512 	struct e1000_softc *sc = arg;
513 	if_softc_ctx_t scctx = sc->shared;
514 	struct em_tx_queue *que = &sc->tx_queues[txqid];
515 	struct tx_ring *txr = &que->txr;
516 
517 	qidx_t processed = 0;
518 	int updated;
519 	qidx_t cur, prev, ntxd, rs_cidx;
520 	int32_t delta;
521 	uint8_t status;
522 
523 	rs_cidx = txr->tx_rs_cidx;
524 	if (rs_cidx == txr->tx_rs_pidx)
525 		return (0);
526 	cur = txr->tx_rsq[rs_cidx];
527 	MPASS(cur != QIDX_INVALID);
528 	status = txr->tx_base[cur].upper.fields.status;
529 	updated = !!(status & E1000_TXD_STAT_DD);
530 
531 	if (!updated)
532 		return (0);
533 
534 	/* If clear is false just let caller know that there
535 	 * are descriptors to reclaim */
536 	if (!clear)
537 		return (1);
538 
539 	prev = txr->tx_cidx_processed;
540 	ntxd = scctx->isc_ntxd[0];
541 	do {
542 		MPASS(prev != cur);
543 		delta = (int32_t)cur - (int32_t)prev;
544 		if (delta < 0)
545 			delta += ntxd;
546 		MPASS(delta > 0);
547 		DPRINTF(iflib_get_dev(sc->ctx),
548 		    "%s: cidx_processed=%u cur=%u clear=%d delta=%d\n",
549 		    __FUNCTION__, prev, cur, clear, delta);
550 
551 		processed += delta;
552 		prev  = cur;
553 		rs_cidx = (rs_cidx + 1) & (ntxd-1);
554 		if (rs_cidx  == txr->tx_rs_pidx)
555 			break;
556 		cur = txr->tx_rsq[rs_cidx];
557 		MPASS(cur != QIDX_INVALID);
558 		status = txr->tx_base[cur].upper.fields.status;
559 	} while ((status & E1000_TXD_STAT_DD));
560 
561 	txr->tx_rs_cidx = rs_cidx;
562 	txr->tx_cidx_processed = prev;
563 	return(processed);
564 }
565 
566 static void
lem_isc_rxd_refill(void * arg,if_rxd_update_t iru)567 lem_isc_rxd_refill(void *arg, if_rxd_update_t iru)
568 {
569 	struct e1000_softc *sc = arg;
570 	if_softc_ctx_t scctx = sc->shared;
571 	struct em_rx_queue *que = &sc->rx_queues[iru->iru_qsidx];
572 	struct rx_ring *rxr = &que->rxr;
573 	struct e1000_rx_desc *rxd;
574 	uint64_t *paddrs;
575 	uint32_t next_pidx, pidx;
576 	uint16_t count;
577 	int i;
578 
579 	paddrs = iru->iru_paddrs;
580 	pidx = iru->iru_pidx;
581 	count = iru->iru_count;
582 
583 	for (i = 0, next_pidx = pidx; i < count; i++) {
584 		rxd = (struct e1000_rx_desc *)&rxr->rx_base[next_pidx];
585 		rxd->buffer_addr = htole64(paddrs[i]);
586 		/* status bits must be cleared */
587 		rxd->status = 0;
588 
589 		if (++next_pidx == scctx->isc_nrxd[0])
590 			next_pidx = 0;
591 	}
592 }
593 
594 static void
em_isc_rxd_refill(void * arg,if_rxd_update_t iru)595 em_isc_rxd_refill(void *arg, if_rxd_update_t iru)
596 {
597 	struct e1000_softc *sc = arg;
598 	if_softc_ctx_t scctx = sc->shared;
599 	uint16_t rxqid = iru->iru_qsidx;
600 	struct em_rx_queue *que = &sc->rx_queues[rxqid];
601 	struct rx_ring *rxr = &que->rxr;
602 	union e1000_rx_desc_extended *rxd;
603 	uint64_t *paddrs;
604 	uint32_t next_pidx, pidx;
605 	uint16_t count;
606 	int i;
607 
608 	paddrs = iru->iru_paddrs;
609 	pidx = iru->iru_pidx;
610 	count = iru->iru_count;
611 
612 	for (i = 0, next_pidx = pidx; i < count; i++) {
613 		rxd = &rxr->rx_base[next_pidx];
614 		rxd->read.buffer_addr = htole64(paddrs[i]);
615 		/* DD bits must be cleared */
616 		rxd->wb.upper.status_error = 0;
617 
618 		if (++next_pidx == scctx->isc_nrxd[0])
619 			next_pidx = 0;
620 	}
621 }
622 
623 static void
em_isc_rxd_flush(void * arg,uint16_t rxqid,uint8_t flid __unused,qidx_t pidx)624 em_isc_rxd_flush(void *arg, uint16_t rxqid, uint8_t flid __unused,
625     qidx_t pidx)
626 {
627 	struct e1000_softc *sc = arg;
628 	struct em_rx_queue *que = &sc->rx_queues[rxqid];
629 	struct rx_ring *rxr = &que->rxr;
630 
631 	E1000_WRITE_REG(&sc->hw, E1000_RDT(rxr->me), pidx);
632 	if (sc->hw.mac.type >= e1000_82540)
633 		em_aim_publish_rx(rxr);
634 }
635 
636 static int
lem_isc_rxd_available(void * arg,uint16_t rxqid,qidx_t idx,qidx_t budget)637 lem_isc_rxd_available(void *arg, uint16_t rxqid, qidx_t idx, qidx_t budget)
638 {
639 	struct e1000_softc *sc = arg;
640 	if_softc_ctx_t scctx = sc->shared;
641 	struct em_rx_queue *que = &sc->rx_queues[rxqid];
642 	struct rx_ring *rxr = &que->rxr;
643 	struct e1000_rx_desc *rxd;
644 	uint32_t staterr = 0;
645 	int cnt, i;
646 
647 	for (cnt = 0, i = idx; cnt < scctx->isc_nrxd[0] && cnt <= budget;) {
648 		rxd = (struct e1000_rx_desc *)&rxr->rx_base[i];
649 		staterr = rxd->status;
650 
651 		if ((staterr & E1000_RXD_STAT_DD) == 0)
652 			break;
653 		if (++i == scctx->isc_nrxd[0])
654 			i = 0;
655 		if (staterr & E1000_RXD_STAT_EOP)
656 			cnt++;
657 	}
658 	return (cnt);
659 }
660 
661 static int
em_isc_rxd_available(void * arg,uint16_t rxqid,qidx_t idx,qidx_t budget)662 em_isc_rxd_available(void *arg, uint16_t rxqid, qidx_t idx, qidx_t budget)
663 {
664 	struct e1000_softc *sc = arg;
665 	if_softc_ctx_t scctx = sc->shared;
666 	struct em_rx_queue *que = &sc->rx_queues[rxqid];
667 	struct rx_ring *rxr = &que->rxr;
668 	union e1000_rx_desc_extended *rxd;
669 	uint32_t staterr = 0;
670 	int cnt, i;
671 
672 	for (cnt = 0, i = idx; cnt < scctx->isc_nrxd[0] && cnt <= budget;) {
673 		rxd = &rxr->rx_base[i];
674 		staterr = le32toh(rxd->wb.upper.status_error);
675 
676 		if ((staterr & E1000_RXD_STAT_DD) == 0)
677 			break;
678 		if (++i == scctx->isc_nrxd[0])
679 			i = 0;
680 		if (staterr & E1000_RXD_STAT_EOP)
681 			cnt++;
682 	}
683 	return (cnt);
684 }
685 
686 static int
lem_isc_rxd_pkt_get(void * arg,if_rxd_info_t ri)687 lem_isc_rxd_pkt_get(void *arg, if_rxd_info_t ri)
688 {
689 	struct e1000_softc *sc = arg;
690 	if_softc_ctx_t scctx = sc->shared;
691 	struct em_rx_queue *que = &sc->rx_queues[ri->iri_qsidx];
692 	struct rx_ring *rxr = &que->rxr;
693 	struct e1000_rx_desc *rxd;
694 	uint16_t len;
695 	uint32_t status, errors;
696 	bool eop;
697 	int i, cidx;
698 
699 	status = errors = i = 0;
700 	cidx = ri->iri_cidx;
701 
702 	do {
703 		rxd = (struct e1000_rx_desc *)&rxr->rx_base[cidx];
704 		status = rxd->status;
705 		errors = rxd->errors;
706 
707 		/* Error Checking then decrement count */
708 		MPASS ((status & E1000_RXD_STAT_DD) != 0);
709 
710 		len = le16toh(rxd->length);
711 		ri->iri_len += len;
712 
713 		eop = (status & E1000_RXD_STAT_EOP) != 0;
714 
715 		/* Make sure bad packets are discarded */
716 		if (errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
717 			sc->dropped_pkts++;
718 			/* XXX fixup if common */
719 			return (EBADMSG);
720 		}
721 
722 		ri->iri_frags[i].irf_flid = 0;
723 		ri->iri_frags[i].irf_idx = cidx;
724 		ri->iri_frags[i].irf_len = len;
725 		/* Zero out the receive descriptors status. */
726 		rxd->status = 0;
727 
728 		if (++cidx == scctx->isc_nrxd[0])
729 			cidx = 0;
730 		i++;
731 	} while (!eop);
732 
733 	rxr->rx_bytes += ri->iri_len;
734 	rxr->rx_packets++;
735 
736 	if (scctx->isc_capenable & IFCAP_RXCSUM)
737 		em_receive_checksum(status, errors, ri);
738 
739 	if (scctx->isc_capenable & IFCAP_VLAN_HWTAGGING &&
740 	    status & E1000_RXD_STAT_VP) {
741 		ri->iri_vtag = le16toh(rxd->special);
742 		ri->iri_flags |= M_VLANTAG;
743 	}
744 
745 	ri->iri_nfrags = i;
746 
747 	return (0);
748 }
749 
750 static int
em_isc_rxd_pkt_get(void * arg,if_rxd_info_t ri)751 em_isc_rxd_pkt_get(void *arg, if_rxd_info_t ri)
752 {
753 	struct e1000_softc *sc = arg;
754 	if_softc_ctx_t scctx = sc->shared;
755 	struct em_rx_queue *que = &sc->rx_queues[ri->iri_qsidx];
756 	struct rx_ring *rxr = &que->rxr;
757 	union e1000_rx_desc_extended *rxd;
758 
759 	uint16_t len;
760 	uint32_t pkt_info;
761 	uint32_t staterr;
762 	bool eop;
763 	int i, cidx;
764 
765 	staterr = i = 0;
766 	cidx = ri->iri_cidx;
767 
768 	do {
769 		rxd = &rxr->rx_base[cidx];
770 		staterr = le32toh(rxd->wb.upper.status_error);
771 		pkt_info = le32toh(rxd->wb.lower.mrq);
772 
773 		/* Error Checking then decrement count */
774 		MPASS ((staterr & E1000_RXD_STAT_DD) != 0);
775 
776 		len = le16toh(rxd->wb.upper.length);
777 		ri->iri_len += len;
778 
779 		eop = (staterr & E1000_RXD_STAT_EOP) != 0;
780 
781 		/* Make sure bad packets are discarded */
782 		if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
783 			sc->dropped_pkts++;
784 			return EBADMSG;
785 		}
786 
787 		ri->iri_frags[i].irf_flid = 0;
788 		ri->iri_frags[i].irf_idx = cidx;
789 		ri->iri_frags[i].irf_len = len;
790 		/* Zero out the receive descriptors status. */
791 		rxd->wb.upper.status_error &= htole32(~0xFF);
792 
793 		if (++cidx == scctx->isc_nrxd[0])
794 			cidx = 0;
795 		i++;
796 	} while (!eop);
797 
798 	rxr->rx_bytes += ri->iri_len;
799 	rxr->rx_packets++;
800 
801 	if (scctx->isc_capenable & IFCAP_RXCSUM)
802 		em_receive_checksum(staterr, staterr >> 24, ri);
803 
804 	if (scctx->isc_capenable & IFCAP_VLAN_HWTAGGING &&
805 	    staterr & E1000_RXD_STAT_VP) {
806 		ri->iri_vtag = le16toh(rxd->wb.upper.vlan);
807 		ri->iri_flags |= M_VLANTAG;
808 	}
809 
810 	ri->iri_flowid = le32toh(rxd->wb.lower.hi_dword.rss);
811 	ri->iri_rsstype = em_determine_rsstype(pkt_info);
812 
813 	ri->iri_nfrags = i;
814 	return (0);
815 }
816 
817 /*********************************************************************
818  *
819  *  Verify that the hardware indicated that the checksum is valid.
820  *  Inform the stack about the status of checksum so that stack
821  *  doesn't spend time verifying the checksum.
822  *
823  *********************************************************************/
824 static void
em_receive_checksum(uint16_t status,uint8_t errors,if_rxd_info_t ri)825 em_receive_checksum(uint16_t status, uint8_t errors, if_rxd_info_t ri)
826 {
827 	if (__predict_false(status & E1000_RXD_STAT_IXSM))
828 		return;
829 
830 	/* If there is a layer 3 or 4 error we are done */
831 	if (__predict_false(errors & (E1000_RXD_ERR_IPE |
832 	    E1000_RXD_ERR_TCPE)))
833 		return;
834 
835 	/* IP Checksum Good */
836 	if (status & E1000_RXD_STAT_IPCS)
837 		ri->iri_csum_flags = (CSUM_IP_CHECKED | CSUM_IP_VALID);
838 
839 	/* Valid L4E checksum */
840 	if (__predict_true(status &
841 	    (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))) {
842 		ri->iri_csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
843 		ri->iri_csum_data = htons(0xffff);
844 	}
845 }
846 
847 /********************************************************************
848  *
849  *  Parse the packet type to determine the appropriate hash
850  *
851  ******************************************************************/
852 static int
em_determine_rsstype(uint32_t pkt_info)853 em_determine_rsstype(uint32_t pkt_info)
854 {
855 	switch (pkt_info & E1000_RXDADV_RSSTYPE_MASK) {
856 	case E1000_RXDADV_RSSTYPE_IPV4_TCP:
857 		return M_HASHTYPE_RSS_TCP_IPV4;
858 	case E1000_RXDADV_RSSTYPE_IPV4:
859 		return M_HASHTYPE_RSS_IPV4;
860 	case E1000_RXDADV_RSSTYPE_IPV6_TCP:
861 		return M_HASHTYPE_RSS_TCP_IPV6;
862 	case E1000_RXDADV_RSSTYPE_IPV6_EX:
863 		return M_HASHTYPE_RSS_IPV6_EX;
864 	case E1000_RXDADV_RSSTYPE_IPV6:
865 		return M_HASHTYPE_RSS_IPV6;
866 	case E1000_RXDADV_RSSTYPE_IPV6_TCP_EX:
867 		return M_HASHTYPE_RSS_TCP_IPV6_EX;
868 	case E1000_RXDADV_RSSTYPE_IPV4_UDP:
869 		return M_HASHTYPE_RSS_UDP_IPV4;
870 	case E1000_RXDADV_RSSTYPE_IPV6_UDP:
871 		return M_HASHTYPE_RSS_UDP_IPV6;
872 	case E1000_RXDADV_RSSTYPE_IPV6_UDP_EX:
873 		return M_HASHTYPE_RSS_UDP_IPV6_EX;
874 	default:
875 		return M_HASHTYPE_NONE;
876 	}
877 }
878