xref: /freebsd/sys/dev/e1000/em_txrx.c (revision 74ca7bf1d4c7173d5575ba168bc4b5f6d181ff5a)
1 /*-
2  * Copyright (c) 2016 Nicole Graziano <nicole@nextbsd.org>
3  * Copyright (c) 2017 Matthew Macy <mmacy@mattmacy.io>
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
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  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  */
27 
28 /* $FreeBSD$ */
29 #include "if_em.h"
30 
31 #ifdef RSS
32 #include <net/rss_config.h>
33 #include <netinet/in_rss.h>
34 #endif
35 
36 #ifdef VERBOSE_DEBUG
37 #define DPRINTF device_printf
38 #else
39 #define DPRINTF(...)
40 #endif
41 
42 /*********************************************************************
43  *  Local Function prototypes
44  *********************************************************************/
45 static int em_tso_setup(struct adapter *adapter, if_pkt_info_t pi, u32 *txd_upper,
46     u32 *txd_lower);
47 static int em_transmit_checksum_setup(struct adapter *adapter, if_pkt_info_t pi,
48     u32 *txd_upper, u32 *txd_lower);
49 static int em_isc_txd_encap(void *arg, if_pkt_info_t pi);
50 static void em_isc_txd_flush(void *arg, uint16_t txqid, qidx_t pidx);
51 static int em_isc_txd_credits_update(void *arg, uint16_t txqid, bool clear);
52 static void em_isc_rxd_refill(void *arg, if_rxd_update_t iru);
53 static void em_isc_rxd_flush(void *arg, uint16_t rxqid, uint8_t flid __unused,
54     qidx_t pidx);
55 static int em_isc_rxd_available(void *arg, uint16_t rxqid, qidx_t idx,
56     qidx_t budget);
57 static int em_isc_rxd_pkt_get(void *arg, if_rxd_info_t ri);
58 
59 static void lem_isc_rxd_refill(void *arg, if_rxd_update_t iru);
60 
61 static int lem_isc_rxd_available(void *arg, uint16_t rxqid, qidx_t idx,
62    qidx_t budget);
63 static int lem_isc_rxd_pkt_get(void *arg, if_rxd_info_t ri);
64 
65 static void lem_receive_checksum(int status, int errors, if_rxd_info_t ri);
66 static void em_receive_checksum(uint32_t status, if_rxd_info_t ri);
67 static int em_determine_rsstype(u32 pkt_info);
68 extern int em_intr(void *arg);
69 
70 struct if_txrx em_txrx = {
71 	em_isc_txd_encap,
72 	em_isc_txd_flush,
73 	em_isc_txd_credits_update,
74 	em_isc_rxd_available,
75 	em_isc_rxd_pkt_get,
76 	em_isc_rxd_refill,
77 	em_isc_rxd_flush,
78 	em_intr
79 };
80 
81 struct if_txrx lem_txrx = {
82 	em_isc_txd_encap,
83 	em_isc_txd_flush,
84 	em_isc_txd_credits_update,
85 	lem_isc_rxd_available,
86 	lem_isc_rxd_pkt_get,
87 	lem_isc_rxd_refill,
88 	em_isc_rxd_flush,
89 	em_intr
90 };
91 
92 extern if_shared_ctx_t em_sctx;
93 
94 void
95 em_dump_rs(struct adapter *adapter)
96 {
97 	if_softc_ctx_t scctx = adapter->shared;
98 	struct em_tx_queue *que;
99 	struct tx_ring *txr;
100 	qidx_t i, ntxd, qid, cur;
101 	int16_t rs_cidx;
102 	uint8_t status;
103 
104 	printf("\n");
105 	ntxd = scctx->isc_ntxd[0];
106 	for (qid = 0; qid < adapter->tx_num_queues; qid++) {
107 		que = &adapter->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 ", qid, rs_cidx, cur);
115 		} else {
116 			rs_cidx = (rs_cidx-1)&(ntxd-1);
117 			cur = txr->tx_rsq[rs_cidx];
118 			printf("qid[%d]->tx_rsq[rs_cidx-1=%d]: %d  ", qid, rs_cidx, cur);
119 		}
120 		printf("cidx_prev=%d rs_pidx=%d ",txr->tx_cidx_processed, txr->tx_rs_pidx);
121 		for (i = 0; i < ntxd; i++) {
122 			if (txr->tx_base[i].upper.fields.status & E1000_TXD_STAT_DD)
123 				printf("%d set ", i);
124 		}
125 		printf("\n");
126 	}
127 }
128 
129 /**********************************************************************
130  *
131  *  Setup work for hardware segmentation offload (TSO) on
132  *  adapters using advanced tx descriptors
133  *
134  **********************************************************************/
135 static int
136 em_tso_setup(struct adapter *adapter, if_pkt_info_t pi, u32 *txd_upper, u32 *txd_lower)
137 {
138 	if_softc_ctx_t scctx = adapter->shared;
139 	struct em_tx_queue *que = &adapter->tx_queues[pi->ipi_qsidx];
140 	struct tx_ring *txr = &que->txr;
141 	struct e1000_context_desc *TXD;
142 	int cur, hdr_len;
143 
144 	hdr_len = pi->ipi_ehdrlen + pi->ipi_ip_hlen + pi->ipi_tcp_hlen;
145 	*txd_lower = (E1000_TXD_CMD_DEXT |	/* Extended descr type */
146 		      E1000_TXD_DTYP_D |	/* Data descr type */
147 		      E1000_TXD_CMD_TSE);	/* Do TSE on this packet */
148 
149 	/* IP and/or TCP header checksum calculation and insertion. */
150 	*txd_upper = (E1000_TXD_POPTS_IXSM | E1000_TXD_POPTS_TXSM) << 8;
151 
152 	cur = pi->ipi_pidx;
153 	TXD = (struct e1000_context_desc *)&txr->tx_base[cur];
154 
155 	/*
156 	 * Start offset for header checksum calculation.
157 	 * End offset for header checksum calculation.
158 	 * Offset of place put the checksum.
159 	 */
160 	TXD->lower_setup.ip_fields.ipcss = pi->ipi_ehdrlen;
161 	TXD->lower_setup.ip_fields.ipcse =
162 	    htole16(pi->ipi_ehdrlen + pi->ipi_ip_hlen - 1);
163 	TXD->lower_setup.ip_fields.ipcso = pi->ipi_ehdrlen + offsetof(struct ip, ip_sum);
164 
165 	/*
166 	 * Start offset for payload checksum calculation.
167 	 * End offset for payload checksum calculation.
168 	 * Offset of place to put the checksum.
169 	 */
170 	TXD->upper_setup.tcp_fields.tucss = pi->ipi_ehdrlen + pi->ipi_ip_hlen;
171 	TXD->upper_setup.tcp_fields.tucse = 0;
172 	TXD->upper_setup.tcp_fields.tucso =
173 	    pi->ipi_ehdrlen + pi->ipi_ip_hlen + offsetof(struct tcphdr, th_sum);
174 
175 	/*
176 	 * Payload size per packet w/o any headers.
177 	 * Length of all headers up to payload.
178 	 */
179 	TXD->tcp_seg_setup.fields.mss = htole16(pi->ipi_tso_segsz);
180 	TXD->tcp_seg_setup.fields.hdr_len = hdr_len;
181 
182 	TXD->cmd_and_length = htole32(adapter->txd_cmd |
183 				E1000_TXD_CMD_DEXT |	/* Extended descr */
184 				E1000_TXD_CMD_TSE |	/* TSE context */
185 				E1000_TXD_CMD_IP |	/* Do IP csum */
186 				E1000_TXD_CMD_TCP |	/* Do TCP checksum */
187 				      (pi->ipi_len - hdr_len)); /* Total len */
188 	txr->tx_tso = TRUE;
189 
190 	if (++cur == scctx->isc_ntxd[0]) {
191 		cur = 0;
192 	}
193 	DPRINTF(iflib_get_dev(adapter->ctx), "%s: pidx: %d cur: %d\n", __FUNCTION__, pi->ipi_pidx, cur);
194 	return (cur);
195 }
196 
197 #define TSO_WORKAROUND 4
198 #define DONT_FORCE_CTX 1
199 
200 
201 /*********************************************************************
202  *  The offload context is protocol specific (TCP/UDP) and thus
203  *  only needs to be set when the protocol changes. The occasion
204  *  of a context change can be a performance detriment, and
205  *  might be better just disabled. The reason arises in the way
206  *  in which the controller supports pipelined requests from the
207  *  Tx data DMA. Up to four requests can be pipelined, and they may
208  *  belong to the same packet or to multiple packets. However all
209  *  requests for one packet are issued before a request is issued
210  *  for a subsequent packet and if a request for the next packet
211  *  requires a context change, that request will be stalled
212  *  until the previous request completes. This means setting up
213  *  a new context effectively disables pipelined Tx data DMA which
214  *  in turn greatly slow down performance to send small sized
215  *  frames.
216  **********************************************************************/
217 
218 static int
219 em_transmit_checksum_setup(struct adapter *adapter, if_pkt_info_t pi, u32 *txd_upper, u32 *txd_lower)
220 {
221 	 struct e1000_context_desc *TXD = NULL;
222 	if_softc_ctx_t scctx = adapter->shared;
223 	struct em_tx_queue *que = &adapter->tx_queues[pi->ipi_qsidx];
224 	struct tx_ring *txr = &que->txr;
225 	int csum_flags = pi->ipi_csum_flags;
226 	int cur, hdr_len;
227 	u32 cmd;
228 
229 	cur = pi->ipi_pidx;
230 	hdr_len = pi->ipi_ehdrlen + pi->ipi_ip_hlen;
231 	cmd = adapter->txd_cmd;
232 
233 	/*
234 	 * The 82574L can only remember the *last* context used
235 	 * regardless of queue that it was use for.  We cannot reuse
236 	 * contexts on this hardware platform and must generate a new
237 	 * context every time.  82574L hardware spec, section 7.2.6,
238 	 * second note.
239 	 */
240 	if (DONT_FORCE_CTX &&
241 	    adapter->tx_num_queues == 1 &&
242 	    txr->csum_lhlen == pi->ipi_ehdrlen &&
243 	    txr->csum_iphlen == pi->ipi_ip_hlen &&
244 	    txr->csum_flags == csum_flags) {
245 		/*
246 		 * Same csum offload context as the previous packets;
247 		 * just return.
248 		 */
249 		*txd_upper = txr->csum_txd_upper;
250 		*txd_lower = txr->csum_txd_lower;
251 		return (cur);
252 	}
253 
254 	TXD = (struct e1000_context_desc *)&txr->tx_base[cur];
255 	if (csum_flags & CSUM_IP) {
256 		*txd_upper |= E1000_TXD_POPTS_IXSM << 8;
257 		/*
258 		 * Start offset for header checksum calculation.
259 		 * End offset for header checksum calculation.
260 		 * Offset of place to put the checksum.
261 		 */
262 		TXD->lower_setup.ip_fields.ipcss = pi->ipi_ehdrlen;
263 		TXD->lower_setup.ip_fields.ipcse = htole16(hdr_len);
264 		TXD->lower_setup.ip_fields.ipcso = pi->ipi_ehdrlen + offsetof(struct ip, ip_sum);
265 		cmd |= E1000_TXD_CMD_IP;
266 	}
267 
268 	if (csum_flags & (CSUM_TCP|CSUM_UDP)) {
269 		uint8_t tucso;
270 
271 		*txd_upper |= E1000_TXD_POPTS_TXSM << 8;
272 		*txd_lower = E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
273 
274 		if (csum_flags & CSUM_TCP) {
275 			tucso = hdr_len + offsetof(struct tcphdr, th_sum);
276 			cmd |= E1000_TXD_CMD_TCP;
277 		} else
278 			tucso = hdr_len + offsetof(struct udphdr, uh_sum);
279 		TXD->upper_setup.tcp_fields.tucss = hdr_len;
280 		TXD->upper_setup.tcp_fields.tucse = htole16(0);
281 		TXD->upper_setup.tcp_fields.tucso = tucso;
282 	}
283 
284 	txr->csum_lhlen = pi->ipi_ehdrlen;
285 	txr->csum_iphlen = pi->ipi_ip_hlen;
286 	txr->csum_flags = csum_flags;
287 	txr->csum_txd_upper = *txd_upper;
288 	txr->csum_txd_lower = *txd_lower;
289 
290 	TXD->tcp_seg_setup.data = htole32(0);
291 	TXD->cmd_and_length =
292 		htole32(E1000_TXD_CMD_IFCS | E1000_TXD_CMD_DEXT | cmd);
293 
294 	if (++cur == scctx->isc_ntxd[0]) {
295 		cur = 0;
296 	}
297 	DPRINTF(iflib_get_dev(adapter->ctx), "checksum_setup csum_flags=%x txd_upper=%x txd_lower=%x hdr_len=%d cmd=%x\n",
298 		      csum_flags, *txd_upper, *txd_lower, hdr_len, cmd);
299 	return (cur);
300 }
301 
302 static int
303 em_isc_txd_encap(void *arg, if_pkt_info_t pi)
304 {
305 	struct adapter *sc = arg;
306 	if_softc_ctx_t scctx = sc->shared;
307 	struct em_tx_queue *que = &sc->tx_queues[pi->ipi_qsidx];
308 	struct tx_ring *txr = &que->txr;
309 	bus_dma_segment_t *segs = pi->ipi_segs;
310 	int nsegs = pi->ipi_nsegs;
311 	int csum_flags = pi->ipi_csum_flags;
312 	int i, j, first, pidx_last;
313 	u32 txd_flags, txd_upper = 0, txd_lower = 0;
314 
315 	struct e1000_tx_desc *ctxd = NULL;
316 	bool do_tso, tso_desc;
317 	qidx_t ntxd;
318 
319 	txd_flags = pi->ipi_flags & IPI_TX_INTR ? E1000_TXD_CMD_RS : 0;
320 	i = first = pi->ipi_pidx;
321 	do_tso = (csum_flags & CSUM_TSO);
322 	tso_desc = FALSE;
323 	ntxd = scctx->isc_ntxd[0];
324 	/*
325 	 * TSO Hardware workaround, if this packet is not
326 	 * TSO, and is only a single descriptor long, and
327 	 * it follows a TSO burst, then we need to add a
328 	 * sentinel descriptor to prevent premature writeback.
329 	 */
330 	if ((!do_tso) && (txr->tx_tso == TRUE)) {
331 		if (nsegs == 1)
332 			tso_desc = TRUE;
333 		txr->tx_tso = FALSE;
334 	}
335 
336 	/* Do hardware assists */
337 	if (do_tso) {
338 		i = em_tso_setup(sc, pi, &txd_upper, &txd_lower);
339 		tso_desc = TRUE;
340 	} else if (csum_flags & EM_CSUM_OFFLOAD) {
341 		i = em_transmit_checksum_setup(sc, pi, &txd_upper, &txd_lower);
342 	}
343 
344 	if (pi->ipi_mflags & M_VLANTAG) {
345 		/* Set the vlan id. */
346 		txd_upper |= htole16(pi->ipi_vtag) << 16;
347 		/* Tell hardware to add tag */
348 		txd_lower |= htole32(E1000_TXD_CMD_VLE);
349 	}
350 
351 	DPRINTF(iflib_get_dev(sc->ctx), "encap: set up tx: nsegs=%d first=%d i=%d\n", nsegs, first, i);
352 	/* XXX adapter->pcix_82544 -- lem_fill_descriptors */
353 
354 	/* Set up our transmit descriptors */
355 	for (j = 0; j < nsegs; j++) {
356 		bus_size_t seg_len;
357 		bus_addr_t seg_addr;
358 		uint32_t cmd;
359 
360 		ctxd = &txr->tx_base[i];
361 		seg_addr = segs[j].ds_addr;
362 		seg_len = segs[j].ds_len;
363 		cmd = E1000_TXD_CMD_IFCS | sc->txd_cmd;
364 
365 		/*
366 		 * TSO Workaround:
367 		 * If this is the last descriptor, we want to
368 		 * split it so we have a small final sentinel
369 		 */
370 		if (tso_desc && (j == (nsegs - 1)) && (seg_len > 8)) {
371 			seg_len -= TSO_WORKAROUND;
372 			ctxd->buffer_addr = htole64(seg_addr);
373 			ctxd->lower.data = htole32(cmd | txd_lower | seg_len);
374 			ctxd->upper.data = htole32(txd_upper);
375 
376 			if (++i == scctx->isc_ntxd[0])
377 				i = 0;
378 
379 			/* Now make the sentinel */
380 			ctxd = &txr->tx_base[i];
381 			ctxd->buffer_addr = htole64(seg_addr + seg_len);
382 			ctxd->lower.data = htole32(cmd | txd_lower | TSO_WORKAROUND);
383 			ctxd->upper.data = htole32(txd_upper);
384 			pidx_last = i;
385 			if (++i == scctx->isc_ntxd[0])
386 				i = 0;
387 			DPRINTF(iflib_get_dev(sc->ctx), "TSO path pidx_last=%d i=%d ntxd[0]=%d\n", pidx_last, i, scctx->isc_ntxd[0]);
388 		} else {
389 			ctxd->buffer_addr = htole64(seg_addr);
390 			ctxd->lower.data = htole32(cmd | txd_lower | seg_len);
391 			ctxd->upper.data = htole32(txd_upper);
392 			pidx_last = i;
393 			if (++i == scctx->isc_ntxd[0])
394 				i = 0;
395 			DPRINTF(iflib_get_dev(sc->ctx), "pidx_last=%d i=%d ntxd[0]=%d\n", pidx_last, i, scctx->isc_ntxd[0]);
396 		}
397 	}
398 
399 	/*
400 	 * Last Descriptor of Packet
401 	 * needs End Of Packet (EOP)
402 	 * and Report Status (RS)
403 	 */
404 	if (txd_flags) {
405 		txr->tx_rsq[txr->tx_rs_pidx] = pidx_last;
406 		DPRINTF(iflib_get_dev(sc->ctx), "setting to RS on %d rs_pidx %d first: %d\n", pidx_last, txr->tx_rs_pidx, first);
407 		txr->tx_rs_pidx = (txr->tx_rs_pidx+1) & (ntxd-1);
408 		MPASS(txr->tx_rs_pidx != txr->tx_rs_cidx);
409 	}
410 	ctxd->lower.data |= htole32(E1000_TXD_CMD_EOP | txd_flags);
411 	DPRINTF(iflib_get_dev(sc->ctx), "tx_buffers[%d]->eop = %d ipi_new_pidx=%d\n", first, pidx_last, i);
412 	pi->ipi_new_pidx = i;
413 
414 	return (0);
415 }
416 
417 static void
418 em_isc_txd_flush(void *arg, uint16_t txqid, qidx_t pidx)
419 {
420 	struct adapter *adapter = arg;
421 	struct em_tx_queue *que = &adapter->tx_queues[txqid];
422 	struct tx_ring *txr = &que->txr;
423 
424 	E1000_WRITE_REG(&adapter->hw, E1000_TDT(txr->me), pidx);
425 }
426 
427 static int
428 em_isc_txd_credits_update(void *arg, uint16_t txqid, bool clear)
429 {
430 	struct adapter *adapter = arg;
431 	if_softc_ctx_t scctx = adapter->shared;
432 	struct em_tx_queue *que = &adapter->tx_queues[txqid];
433 	struct tx_ring *txr = &que->txr;
434 
435 	qidx_t processed = 0;
436 	int updated;
437 	qidx_t cur, prev, ntxd, rs_cidx;
438 	int32_t delta;
439 	uint8_t status;
440 
441 	rs_cidx = txr->tx_rs_cidx;
442 	if (rs_cidx == txr->tx_rs_pidx)
443 		return (0);
444 	cur = txr->tx_rsq[rs_cidx];
445 	MPASS(cur != QIDX_INVALID);
446 	status = txr->tx_base[cur].upper.fields.status;
447 	updated = !!(status & E1000_TXD_STAT_DD);
448 
449 	if (clear == false || updated == 0)
450 		return (updated);
451 
452 	prev = txr->tx_cidx_processed;
453 	ntxd = scctx->isc_ntxd[0];
454 	do {
455 		delta = (int32_t)cur - (int32_t)prev;
456 		MPASS(prev == 0 || delta != 0);
457 		if (delta < 0)
458 			delta += ntxd;
459 		DPRINTF(iflib_get_dev(adapter->ctx),
460 			      "%s: cidx_processed=%u cur=%u clear=%d delta=%d\n",
461 			      __FUNCTION__, prev, cur, clear, delta);
462 
463 		processed += delta;
464 		prev  = cur;
465 		rs_cidx = (rs_cidx + 1) & (ntxd-1);
466 		if (rs_cidx  == txr->tx_rs_pidx)
467 			break;
468 		cur = txr->tx_rsq[rs_cidx];
469 		MPASS(cur != QIDX_INVALID);
470 		status = txr->tx_base[cur].upper.fields.status;
471 	} while ((status & E1000_TXD_STAT_DD));
472 
473 	txr->tx_rs_cidx = rs_cidx;
474 	txr->tx_cidx_processed = prev;
475 	return(processed);
476 }
477 
478 static void
479 lem_isc_rxd_refill(void *arg, if_rxd_update_t iru)
480 {
481 	struct adapter *sc = arg;
482 	if_softc_ctx_t scctx = sc->shared;
483 	struct em_rx_queue *que = &sc->rx_queues[iru->iru_qsidx];
484 	struct rx_ring *rxr = &que->rxr;
485 	struct e1000_rx_desc *rxd;
486 	uint64_t *paddrs;
487 	uint32_t next_pidx, pidx;
488 	uint16_t count;
489 	int i;
490 
491 	paddrs = iru->iru_paddrs;
492 	pidx = iru->iru_pidx;
493 	count = iru->iru_count;
494 
495 	for (i = 0, next_pidx = pidx; i < count; i++) {
496 		rxd = (struct e1000_rx_desc *)&rxr->rx_base[next_pidx];
497 		rxd->buffer_addr = htole64(paddrs[i]);
498 		/* status bits must be cleared */
499 		rxd->status = 0;
500 
501 		if (++next_pidx == scctx->isc_nrxd[0])
502 			next_pidx = 0;
503 	}
504 }
505 
506 static void
507 em_isc_rxd_refill(void *arg, if_rxd_update_t iru)
508 {
509 	struct adapter *sc = arg;
510 	if_softc_ctx_t scctx = sc->shared;
511 	uint16_t rxqid = iru->iru_qsidx;
512 	struct em_rx_queue *que = &sc->rx_queues[rxqid];
513 	struct rx_ring *rxr = &que->rxr;
514 	union e1000_rx_desc_extended *rxd;
515 	uint64_t *paddrs;
516 	uint32_t next_pidx, pidx;
517 	uint16_t count;
518 	int i;
519 
520 	paddrs = iru->iru_paddrs;
521 	pidx = iru->iru_pidx;
522 	count = iru->iru_count;
523 
524 	for (i = 0, next_pidx = pidx; i < count; i++) {
525 		rxd = &rxr->rx_base[next_pidx];
526 		rxd->read.buffer_addr = htole64(paddrs[i]);
527 		/* DD bits must be cleared */
528 		rxd->wb.upper.status_error = 0;
529 
530 		if (++next_pidx == scctx->isc_nrxd[0])
531 			next_pidx = 0;
532 	}
533 }
534 
535 static void
536 em_isc_rxd_flush(void *arg, uint16_t rxqid, uint8_t flid __unused, qidx_t pidx)
537 {
538 	struct adapter *sc = arg;
539 	struct em_rx_queue *que = &sc->rx_queues[rxqid];
540 	struct rx_ring *rxr = &que->rxr;
541 
542 	E1000_WRITE_REG(&sc->hw, E1000_RDT(rxr->me), pidx);
543 }
544 
545 static int
546 lem_isc_rxd_available(void *arg, uint16_t rxqid, qidx_t idx, qidx_t budget)
547 {
548 	struct adapter *sc = arg;
549 	if_softc_ctx_t scctx = sc->shared;
550 	struct em_rx_queue *que = &sc->rx_queues[rxqid];
551 	struct rx_ring *rxr = &que->rxr;
552 	struct e1000_rx_desc *rxd;
553 	u32 staterr = 0;
554 	int cnt, i;
555 
556 	if (budget == 1) {
557 		rxd = (struct e1000_rx_desc *)&rxr->rx_base[idx];
558 		staterr = rxd->status;
559 		return (staterr & E1000_RXD_STAT_DD);
560 	}
561 
562 	for (cnt = 0, i = idx; cnt < scctx->isc_nrxd[0] && cnt <= budget;) {
563 		rxd = (struct e1000_rx_desc *)&rxr->rx_base[i];
564 		staterr = rxd->status;
565 
566 		if ((staterr & E1000_RXD_STAT_DD) == 0)
567 			break;
568 
569 		if (++i == scctx->isc_nrxd[0])
570 			i = 0;
571 
572 		if (staterr & E1000_RXD_STAT_EOP)
573 			cnt++;
574 	}
575 	return (cnt);
576 }
577 
578 static int
579 em_isc_rxd_available(void *arg, uint16_t rxqid, qidx_t idx, qidx_t budget)
580 {
581 	struct adapter *sc = arg;
582 	if_softc_ctx_t scctx = sc->shared;
583 	struct em_rx_queue *que = &sc->rx_queues[rxqid];
584 	struct rx_ring *rxr = &que->rxr;
585 	union e1000_rx_desc_extended *rxd;
586 	u32 staterr = 0;
587 	int cnt, i;
588 
589 	if (budget == 1) {
590 		rxd = &rxr->rx_base[idx];
591 		staterr = le32toh(rxd->wb.upper.status_error);
592 		return (staterr & E1000_RXD_STAT_DD);
593 	}
594 
595 	for (cnt = 0, i = idx; cnt < scctx->isc_nrxd[0] && cnt <= budget;) {
596 		rxd = &rxr->rx_base[i];
597 		staterr = le32toh(rxd->wb.upper.status_error);
598 
599 		if ((staterr & E1000_RXD_STAT_DD) == 0)
600 			break;
601 
602 		if (++i == scctx->isc_nrxd[0]) {
603 			i = 0;
604 		}
605 
606 		if (staterr & E1000_RXD_STAT_EOP)
607 			cnt++;
608 
609 	}
610 	return (cnt);
611 }
612 
613 static int
614 lem_isc_rxd_pkt_get(void *arg, if_rxd_info_t ri)
615 {
616 	struct adapter *adapter = arg;
617 	if_softc_ctx_t scctx = adapter->shared;
618 	struct em_rx_queue *que = &adapter->rx_queues[ri->iri_qsidx];
619 	struct rx_ring *rxr = &que->rxr;
620 	struct e1000_rx_desc *rxd;
621 	u16 len;
622 	u32 status, errors;
623 	bool eop;
624 	int i, cidx;
625 
626 	status = errors = i = 0;
627 	cidx = ri->iri_cidx;
628 
629 	do {
630 		rxd = (struct e1000_rx_desc *)&rxr->rx_base[cidx];
631 		status = rxd->status;
632 		errors = rxd->errors;
633 
634 		/* Error Checking then decrement count */
635 		MPASS ((status & E1000_RXD_STAT_DD) != 0);
636 
637 		len = le16toh(rxd->length);
638 		ri->iri_len += len;
639 
640 		eop = (status & E1000_RXD_STAT_EOP) != 0;
641 
642 		/* Make sure bad packets are discarded */
643 		if (errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
644 			adapter->dropped_pkts++;
645 			/* XXX fixup if common */
646 			return (EBADMSG);
647 		}
648 
649 		ri->iri_frags[i].irf_flid = 0;
650 		ri->iri_frags[i].irf_idx = cidx;
651 		ri->iri_frags[i].irf_len = len;
652 		/* Zero out the receive descriptors status. */
653 		rxd->status = 0;
654 
655 		if (++cidx == scctx->isc_nrxd[0])
656 			cidx = 0;
657 		i++;
658 	} while (!eop);
659 
660 	/* XXX add a faster way to look this up */
661 	if (adapter->hw.mac.type >= e1000_82543 && !(status & E1000_RXD_STAT_IXSM))
662 		lem_receive_checksum(status, errors, ri);
663 
664 	if (status & E1000_RXD_STAT_VP) {
665 		ri->iri_vtag = le16toh(rxd->special);
666 		ri->iri_flags |= M_VLANTAG;
667 	}
668 
669 	ri->iri_nfrags = i;
670 
671 	return (0);
672 }
673 
674 static int
675 em_isc_rxd_pkt_get(void *arg, if_rxd_info_t ri)
676 {
677 	struct adapter *adapter = arg;
678 	if_softc_ctx_t scctx = adapter->shared;
679 	struct em_rx_queue *que = &adapter->rx_queues[ri->iri_qsidx];
680 	struct rx_ring *rxr = &que->rxr;
681 	union e1000_rx_desc_extended *rxd;
682 
683 	u16 len;
684 	u32 pkt_info;
685 	u32 staterr = 0;
686 	bool eop;
687 	int i, cidx, vtag;
688 
689 	i = vtag = 0;
690 	cidx = ri->iri_cidx;
691 
692 	do {
693 		rxd = &rxr->rx_base[cidx];
694 		staterr = le32toh(rxd->wb.upper.status_error);
695 		pkt_info = le32toh(rxd->wb.lower.mrq);
696 
697 		/* Error Checking then decrement count */
698 		MPASS ((staterr & E1000_RXD_STAT_DD) != 0);
699 
700 		len = le16toh(rxd->wb.upper.length);
701 		ri->iri_len += len;
702 
703 		eop = (staterr & E1000_RXD_STAT_EOP) != 0;
704 
705 		/* Make sure bad packets are discarded */
706 		if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
707 			adapter->dropped_pkts++;
708 			return EBADMSG;
709 		}
710 
711 		ri->iri_frags[i].irf_flid = 0;
712 		ri->iri_frags[i].irf_idx = cidx;
713 		ri->iri_frags[i].irf_len = len;
714 		/* Zero out the receive descriptors status. */
715 		rxd->wb.upper.status_error &= htole32(~0xFF);
716 
717 		if (++cidx == scctx->isc_nrxd[0])
718 			cidx = 0;
719 		i++;
720 	} while (!eop);
721 
722 	/* XXX add a faster way to look this up */
723 	if (adapter->hw.mac.type >= e1000_82543)
724 		em_receive_checksum(staterr, ri);
725 
726 	if (staterr & E1000_RXD_STAT_VP) {
727 		vtag = le16toh(rxd->wb.upper.vlan);
728 	}
729 
730 	ri->iri_vtag = vtag;
731 	if (vtag)
732 		ri->iri_flags |= M_VLANTAG;
733 
734 	ri->iri_flowid = le32toh(rxd->wb.lower.hi_dword.rss);
735 	ri->iri_rsstype = em_determine_rsstype(pkt_info);
736 
737 	ri->iri_nfrags = i;
738 	return (0);
739 }
740 
741 /*********************************************************************
742  *
743  *  Verify that the hardware indicated that the checksum is valid.
744  *  Inform the stack about the status of checksum so that stack
745  *  doesn't spend time verifying the checksum.
746  *
747  *********************************************************************/
748 static void
749 lem_receive_checksum(int status, int errors, if_rxd_info_t ri)
750 {
751 	/* Did it pass? */
752 	if (status & E1000_RXD_STAT_IPCS && !(errors & E1000_RXD_ERR_IPE))
753 		ri->iri_csum_flags = (CSUM_IP_CHECKED|CSUM_IP_VALID);
754 
755 	if (status & E1000_RXD_STAT_TCPCS) {
756 		/* Did it pass? */
757 		if (!(errors & E1000_RXD_ERR_TCPE)) {
758 			ri->iri_csum_flags |=
759 			(CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
760 			ri->iri_csum_data = htons(0xffff);
761 		}
762 	}
763 }
764 
765 /********************************************************************
766  *
767  *  Parse the packet type to determine the appropriate hash
768  *
769  ******************************************************************/
770 static int
771 em_determine_rsstype(u32 pkt_info)
772 {
773 	switch (pkt_info & E1000_RXDADV_RSSTYPE_MASK) {
774 	case E1000_RXDADV_RSSTYPE_IPV4_TCP:
775 		return M_HASHTYPE_RSS_TCP_IPV4;
776 	case E1000_RXDADV_RSSTYPE_IPV4:
777 		return M_HASHTYPE_RSS_IPV4;
778 	case E1000_RXDADV_RSSTYPE_IPV6_TCP:
779 		return M_HASHTYPE_RSS_TCP_IPV6;
780 	case E1000_RXDADV_RSSTYPE_IPV6_EX:
781 		return M_HASHTYPE_RSS_IPV6_EX;
782 	case E1000_RXDADV_RSSTYPE_IPV6:
783 		return M_HASHTYPE_RSS_IPV6;
784 	case E1000_RXDADV_RSSTYPE_IPV6_TCP_EX:
785 		return M_HASHTYPE_RSS_TCP_IPV6_EX;
786 	default:
787 		return M_HASHTYPE_OPAQUE;
788 	}
789 }
790 
791 static void
792 em_receive_checksum(uint32_t status, if_rxd_info_t ri)
793 {
794 	ri->iri_csum_flags = 0;
795 
796 	/* Ignore Checksum bit is set */
797 	if (status & E1000_RXD_STAT_IXSM)
798 		return;
799 
800 	/* If the IP checksum exists and there is no IP Checksum error */
801 	if ((status & (E1000_RXD_STAT_IPCS | E1000_RXDEXT_STATERR_IPE)) ==
802 	    E1000_RXD_STAT_IPCS) {
803 		ri->iri_csum_flags = (CSUM_IP_CHECKED | CSUM_IP_VALID);
804 	}
805 
806 	/* TCP or UDP checksum */
807 	if ((status & (E1000_RXD_STAT_TCPCS | E1000_RXDEXT_STATERR_TCPE)) ==
808 	    E1000_RXD_STAT_TCPCS) {
809 		ri->iri_csum_flags |= (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
810 		ri->iri_csum_data = htons(0xffff);
811 	}
812 	if (status & E1000_RXD_STAT_UDPCS) {
813 		ri->iri_csum_flags |= (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
814 		ri->iri_csum_data = htons(0xffff);
815 	}
816 }
817