xref: /titanic_41/usr/src/uts/common/io/chxge/pe.c (revision 40db2e2b777b79f3dd0d6d9629593a07f86b9c0a)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 /*
28  * This file is part of the Chelsio T1 Ethernet driver.
29  *
30  * Copyright (C) 2003-2005 Chelsio Communications.  All rights reserved.
31  */
32 
33 /*
34  * Solaris Multithreaded STREAMS Chelsio PCI Ethernet Driver.
35  * Interface code
36  */
37 
38 #pragma ident	"%Z%%M%	%I%	%E% SMI"
39 
40 #include <sys/types.h>
41 #include <sys/systm.h>
42 #include <sys/cmn_err.h>
43 #include <sys/ddi.h>
44 #include <sys/sunddi.h>
45 #include <sys/byteorder.h>
46 #include <sys/atomic.h>
47 #include <sys/ethernet.h>
48 #if PE_PROFILING_ENABLED
49 #include <sys/time.h>
50 #endif
51 #include <sys/gld.h>
52 #include "ostypes.h"
53 #include "common.h"
54 #include "oschtoe.h"
55 #ifdef CONFIG_CHELSIO_T1_1G
56 #include "fpga_defs.h"
57 #endif
58 #include "regs.h"
59 #ifdef CONFIG_CHELSIO_T1_OFFLOAD
60 #include "mc3.h"
61 #include "mc4.h"
62 #endif
63 #include "sge.h"
64 #include "tp.h"
65 #ifdef CONFIG_CHELSIO_T1_OFFLOAD
66 #include "ulp.h"
67 #endif
68 #include "espi.h"
69 #include "elmer0.h"
70 #include "gmac.h"
71 #include "cphy.h"
72 #include "suni1x10gexp_regs.h"
73 #include "ch.h"
74 
75 #define	MLEN(mp) ((mp)->b_wptr - (mp)->b_rptr)
76 
77 extern uint32_t buffers_in_use[];
78 extern kmutex_t in_use_l;
79 extern uint32_t in_use_index;
80 
81 static void link_start(ch_t *sa, struct pe_port_t *pp);
82 static ch_esb_t *ch_alloc_small_esbbuf(ch_t *sa, uint32_t i);
83 static ch_esb_t *ch_alloc_big_esbbuf(ch_t *sa, uint32_t i);
84 void ch_big_rbuf_recycle(ch_esb_t *rbp);
85 void ch_small_rbuf_recycle(ch_esb_t *rbp);
86 static const struct board_info *pe_sa_init(ch_t *sa);
87 static int ch_set_config_data(ch_t *chp);
88 void pe_rbuf_pool_free(ch_t *chp);
89 static void pe_free_driver_resources(ch_t *sa);
90 static void update_mtu_tab(ch_t *adapter);
91 static int pe_change_mtu(ch_t *chp);
92 
93 /*
94  * CPL5 Defines (from netinet/cpl5_commands.h)
95  */
96 #define	FLITSTOBYTES	8
97 
98 #define	CPL_FORMAT_0_SIZE 8
99 #define	CPL_FORMAT_1_SIZE 16
100 #define	CPL_FORMAT_2_SIZE 24
101 #define	CPL_FORMAT_3_SIZE 32
102 #define	CPL_FORMAT_4_SIZE 40
103 #define	CPL_FORMAT_5_SIZE 48
104 
105 #define	TID_MASK 0xffffff
106 
107 #define	PE_LINK_SPEED_AUTONEG	5
108 
109 static int pe_small_rbuf_pool_init(ch_t *sa);
110 static int pe_big_rbuf_pool_init(ch_t *sa);
111 static int pe_make_fake_arp(ch_t *chp, unsigned char *arpp);
112 static uint32_t pe_get_ip(unsigned char *arpp);
113 
114 /*
115  * May be set in /etc/system to 0 to use default latency timer for 10G.
116  * See PCI register 0xc definition.
117  */
118 int enable_latency_timer = 1;
119 
120 /*
121  * May be set in /etc/system to 0 to disable hardware checksum for
122  * TCP and UDP.
123  */
124 int enable_checksum_offload = 1;
125 
126 /*
127  * Multiplier for freelist pool.
128  */
129 int fl_sz_multiplier = 6;
130 
131 uint_t
132 pe_intr(ch_t *sa)
133 {
134 	mutex_enter(&sa->ch_intr);
135 
136 	if (sge_data_in(sa->sge)) {
137 		sa->isr_intr++;
138 		mutex_exit(&sa->ch_intr);
139 		return (DDI_INTR_CLAIMED);
140 	}
141 
142 	mutex_exit(&sa->ch_intr);
143 
144 	return (DDI_INTR_UNCLAIMED);
145 }
146 
147 /*
148  * Each setup struct will call this function to
149  * initialize.
150  */
151 void
152 pe_init(void* xsa)
153 {
154 	ch_t *sa = NULL;
155 	int i = 0;
156 
157 	sa = (ch_t *)xsa;
158 
159 	/*
160 	 * Need to count the number of times this routine is called
161 	 * because we only want the resources to be allocated once.
162 	 * The 7500 has four ports and so this routine can be called
163 	 * once for each port.
164 	 */
165 	if (sa->init_counter == 0) {
166 		for_each_port(sa, i) {
167 
168 			/*
169 			 * We only want to initialize the line if it is down.
170 			 */
171 			if (sa->port[i].line_up == 0) {
172 				link_start(sa, &sa->port[i]);
173 				sa->port[i].line_up = 1;
174 			}
175 		}
176 
177 		(void) t1_init_hw_modules(sa);
178 
179 		/*
180 		 * Enable/Disable checksum offloading.
181 		 */
182 		if (sa->ch_config.cksum_enabled) {
183 			if (sa->config_data.offload_ip_cksum) {
184 				/* Notify that HW will do the checksum. */
185 				t1_tp_set_ip_checksum_offload(sa->tp, 1);
186 			}
187 
188 			if (sa->config_data.offload_tcp_cksum) {
189 				/* Notify that HW will do the checksum. */
190 				t1_tp_set_tcp_checksum_offload(sa->tp, 1);
191 			}
192 
193 			if (sa->config_data.offload_udp_cksum) {
194 				/* Notify that HW will do the checksum. */
195 				t1_tp_set_udp_checksum_offload(sa->tp, 1);
196 			}
197 		}
198 
199 		sa->ch_flags |= PEINITDONE;
200 
201 		sa->init_counter++;
202 	}
203 
204 	/*
205 	 * Enable interrupts after starting the SGE so
206 	 * that the SGE is ready to handle interrupts.
207 	 */
208 	(void) sge_start(sa->sge);
209 	t1_interrupts_enable(sa);
210 
211 	/*
212 	 * set mtu (either 1500 or bigger)
213 	 */
214 	(void) pe_change_mtu(sa);
215 #ifdef HOST_PAUSE
216 	/*
217 	 * get the configured value of the MAC.
218 	 */
219 	(void) t1_tpi_read(sa, SUNI1x10GEXP_REG_TXXG_CONFIG_1 << 2,
220 	    &sa->txxg_cfg1);
221 #endif
222 }
223 
224 /* ARGSUSED */
225 static void
226 link_start(ch_t *sa, struct pe_port_t *p)
227 {
228 	struct cmac *mac = p->mac;
229 
230 	mac->ops->reset(mac);
231 	if (mac->ops->macaddress_set)
232 		mac->ops->macaddress_set(mac, p->enaddr);
233 	(void) t1_link_start(p->phy, mac, &p->link_config);
234 	mac->ops->enable(mac, MAC_DIRECTION_RX | MAC_DIRECTION_TX);
235 }
236 
237 /*
238  * turn off interrupts...
239  */
240 void
241 pe_stop(ch_t *sa)
242 {
243 	t1_interrupts_disable(sa);
244 	(void) sge_stop(sa->sge);
245 
246 	/*
247 	 * we can still be running an interrupt thread in sge_data_in().
248 	 * If we are, we'll block on the ch_intr lock
249 	 */
250 	mutex_enter(&sa->ch_intr);
251 	mutex_exit(&sa->ch_intr);
252 }
253 
254 /*
255  * output mblk to SGE level and out to the wire.
256  */
257 
258 int
259 pe_start(ch_t *sa, mblk_t *mp, uint32_t flg)
260 {
261 	mblk_t *m0 = mp;
262 	cmdQ_ce_t cm[16];
263 	cmdQ_ce_t *cmp;
264 	cmdQ_ce_t *hmp = &cm[0]; /* head of cm table (may be kmem_alloed) */
265 	int cm_flg = 0;		/* flag (1 - if kmem-alloced) */
266 	int nseg = 0;		/* number cmdQ_ce entries created */
267 	int mseg = 16;		/* maximum entries in hmp arrary */
268 	int freeme = 0;		/* we have an mblk to free in case of error */
269 	uint32_t ch_bind_dma_handle(ch_t *, int, caddr_t, cmdQ_ce_t *,
270 	    uint32_t);
271 #if defined(__sparc)
272 	uint32_t ch_bind_dvma_handle(ch_t *, int, caddr_t, cmdQ_ce_t *,
273 	    uint32_t);
274 #endif
275 	int rv;			/* return value on error */
276 
277 #ifdef CONFIG_CHELSIO_T1_OFFLOAD
278 	if (flg & CH_OFFLOAD) {
279 		hmp->ce_pa = ((tbuf_t *)mp)->tb_pa;
280 		hmp->ce_dh = NULL;
281 		hmp->ce_flg = DH_TOE;
282 		hmp->ce_len = ((tbuf_t *)mp)->tb_len;
283 		hmp->ce_mp = mp;
284 
285 		/* make sure data is flushed to physical memory */
286 		(void) ddi_dma_sync((ddi_dma_handle_t)((tbuf_t *)mp)->tb_dh,
287 		    (off_t)0, hmp->ce_len, DDI_DMA_SYNC_FORDEV);
288 
289 		if (sge_data_out(sa->sge, 0, mp, hmp, 1, flg) == 0) {
290 			return (0);
291 		}
292 
293 		/*
294 		 * set a flag so we'll restart upper layer when
295 		 * resources become available.
296 		 */
297 		sa->ch_blked = 1;
298 		return (1);
299 	}
300 #endif	/* CONFIG_CHELSIO_T1_OFFLOAD */
301 
302 	/* writes from toe will always have CPL header in place */
303 	if (flg & CH_NO_CPL) {
304 		struct cpl_tx_pkt *cpl;
305 
306 		/* PR2928 & PR3309 */
307 		if (sa->ch_ip == NULL) {
308 			ushort_t ethertype = ntohs(*(short *)&mp->b_rptr[12]);
309 			if (ethertype == ETHERTYPE_ARP) {
310 				if (is_T2(sa)) {
311 					/*
312 					 * We assume here that the arp will be
313 					 * contained in one mblk.
314 					 */
315 					if (pe_make_fake_arp(sa, mp->b_rptr)) {
316 						freemsg(mp);
317 						sa->oerr++;
318 						return (0);
319 					}
320 				} else {
321 					sa->ch_ip = pe_get_ip(mp->b_rptr);
322 				}
323 			}
324 		}
325 
326 		/*
327 		 * if space in front of packet big enough for CPL
328 		 * header, then use it. We'll allocate an mblk
329 		 * otherwise.
330 		 */
331 		if ((mp->b_rptr - mp->b_datap->db_base) >= SZ_CPL_TX_PKT) {
332 
333 			mp->b_rptr -= SZ_CPL_TX_PKT;
334 
335 		} else {
336 
337 #ifdef SUN_KSTATS
338 			sa->sge->intr_cnt.tx_need_cpl_space++;
339 #endif
340 			m0 = allocb(SZ_CPL_TX_PKT, BPRI_HI);
341 			if (m0 == NULL) {
342 				freemsg(mp);
343 				sa->oerr++;
344 				return (0);
345 			}
346 
347 			m0->b_wptr = m0->b_rptr + SZ_CPL_TX_PKT;
348 			m0->b_cont = mp;
349 			freeme = 1;
350 
351 			mp = m0;
352 		}
353 
354 		/* fill in cpl header */
355 		cpl = (struct cpl_tx_pkt *)mp->b_rptr;
356 		cpl->opcode = CPL_TX_PKT;
357 		cpl->iff = 0;		/* XXX port 0 needs fixing with NEMO */
358 		cpl->ip_csum_dis = 1;	/* no IP header cksum */
359 		cpl->l4_csum_dis =
360 		    flg & CH_NO_HWCKSUM;	/* CH_NO_HWCKSUM == 1 */
361 		cpl->vlan_valid = 0;		/* no vlan */
362 	}
363 
364 	if (m0->b_cont) {
365 
366 #ifdef SUN_KSTATS
367 			sa->sge->intr_cnt.tx_multi_mblks++;
368 #endif
369 
370 		while (mp) {
371 			int lseg;	/* added by ch_bind_dma_handle() */
372 			int len;
373 
374 			len = MLEN(mp);
375 			/* skip mlks with no data */
376 			if (len == 0) {
377 				mp = mp->b_cont;
378 				continue;
379 			}
380 
381 			/*
382 			 * if we've run out of space on stack, then we
383 			 * allocate a temporary buffer to hold the
384 			 * information. This will kill the the performance,
385 			 * but since it shouldn't really occur, we can live
386 			 * with it. Since jumbo frames may map multiple
387 			 * descriptors, we reallocate the hmp[] array before
388 			 * we reach the end.
389 			 */
390 			if (nseg >= (mseg-4)) {
391 				cmdQ_ce_t *buf;
392 				int j;
393 
394 				buf = kmem_alloc(sizeof (cmdQ_ce_t) * 2 * mseg,
395 				    KM_SLEEP);
396 
397 				for (j = 0; j < nseg; j++)
398 					buf[j] = hmp[j];
399 
400 				if (cm_flg) {
401 					kmem_free(hmp,
402 					    mseg * sizeof (cmdQ_ce_t));
403 				} else
404 					cm_flg = 1;
405 
406 				hmp = buf;
407 				mseg = 2*mseg;
408 
409 				/*
410 				 * We've used up ch table on stack
411 				 */
412 			}
413 
414 #if defined(__sparc)
415 			if (sa->ch_config.enable_dvma) {
416 				lseg = ch_bind_dvma_handle(sa, len,
417 				    (void *)mp->b_rptr,
418 				    &hmp[nseg], mseg - nseg);
419 				if (lseg == NULL) {
420 					sa->sge->intr_cnt.tx_no_dvma1++;
421 					if ((lseg = ch_bind_dma_handle(sa, len,
422 					    (void *)mp->b_rptr,
423 					    &hmp[nseg],
424 					    mseg - nseg)) == NULL) {
425 						sa->sge->intr_cnt.tx_no_dma1++;
426 
427 						/*
428 						 * ran out of space. Gonna bale
429 						 */
430 						rv = 0;
431 
432 						/*
433 						 * we may have processed
434 						 * previous mblks and have
435 						 * descriptors. If so, we need
436 						 * to free the meta struct
437 						 * entries before freeing
438 						 * the mblk.
439 						 */
440 						if (nseg)
441 							goto error;
442 						goto error1;
443 					}
444 				}
445 			} else {
446 				lseg = ch_bind_dma_handle(sa, len,
447 				    (void *)mp->b_rptr, &hmp[nseg],
448 				    mseg - nseg);
449 				if (lseg == NULL) {
450 					sa->sge->intr_cnt.tx_no_dma1++;
451 
452 					/*
453 					 * ran out of space. Gona bale
454 					 */
455 					rv = 0;
456 
457 					/*
458 					 * we may have processed previous
459 					 * mblks and have descriptors. If so,
460 					 * we need to free the meta struct
461 					 * entries before freeing the mblk.
462 					 */
463 					if (nseg)
464 						goto error;
465 					goto error1;
466 				}
467 			}
468 #else	/* defined(__sparc) */
469 			lseg = ch_bind_dma_handle(sa, len,
470 			    (void *)mp->b_rptr, &hmp[nseg],
471 			    mseg - nseg);
472 			if (lseg == NULL) {
473 				sa->sge->intr_cnt.tx_no_dma1++;
474 
475 				/*
476 				 * ran out of space. Gona bale
477 				 */
478 				rv = 0;
479 
480 				/*
481 				 * we may have processed previous mblks and
482 				 * have descriptors. If so, we need to free
483 				 * the meta struct entries before freeing
484 				 * the mblk.
485 				 */
486 				if (nseg)
487 					goto error;
488 				goto error1;
489 			}
490 #endif	/* defined(__sparc) */
491 			nseg += lseg;
492 			mp = mp->b_cont;
493 		}
494 
495 		/*
496 		 * SHOULD NEVER OCCUR, BUT...
497 		 * no data if nseg 0 or
498 		 * nseg 1 and a CPL mblk (CPL mblk only with offload mode)
499 		 * and no data
500 		 */
501 		if ((nseg == 0) || (freeme && (nseg == 1))) {
502 			rv = 0;
503 			goto error1;
504 		}
505 
506 	} else {
507 		int len;
508 
509 		/* we assume that we always have data with one packet */
510 		len = MLEN(mp);
511 
512 #if defined(__sparc)
513 		if (sa->ch_config.enable_dvma) {
514 			nseg = ch_bind_dvma_handle(sa, len,
515 			    (void *)mp->b_rptr,
516 			    &hmp[0], 16);
517 			if (nseg == NULL) {
518 				sa->sge->intr_cnt.tx_no_dvma2++;
519 				nseg = ch_bind_dma_handle(sa, len,
520 				    (void *)mp->b_rptr,
521 				    &hmp[0], 16);
522 				if (nseg == NULL) {
523 					sa->sge->intr_cnt.tx_no_dma2++;
524 
525 					/*
526 					 * ran out of space. Gona bale
527 					 */
528 					rv = 0;
529 					goto error1;
530 				}
531 			}
532 		} else {
533 			nseg = ch_bind_dma_handle(sa, len,
534 			    (void *)mp->b_rptr, &hmp[0], 16);
535 			if (nseg == NULL) {
536 				sa->sge->intr_cnt.tx_no_dma2++;
537 
538 				/*
539 				 * ran out of space. Gona bale
540 				 */
541 				rv = 0;
542 				goto error1;
543 			}
544 		}
545 #else	/* defined(__sparc) */
546 		nseg = ch_bind_dma_handle(sa, len,
547 		    (void *)mp->b_rptr, &hmp[0], 16);
548 		if (nseg == NULL) {
549 			sa->sge->intr_cnt.tx_no_dma2++;
550 
551 			/*
552 			 * ran out of space. Gona bale
553 			 */
554 			rv = 0;
555 			goto error1;
556 		}
557 #endif	/* defined(__sparc) */
558 
559 		/*
560 		 * dummy arp message to handle PR3309 & PR2928
561 		 */
562 		if (flg & CH_ARP)
563 			hmp->ce_flg |= DH_ARP;
564 	}
565 
566 	if (sge_data_out(sa->sge, 0, m0, hmp, nseg, flg) == 0) {
567 		if (cm_flg)
568 			kmem_free(hmp, mseg * sizeof (cmdQ_ce_t));
569 		return (0);
570 	}
571 
572 	/*
573 	 * set a flag so we'll restart upper layer when
574 	 * resources become available.
575 	 */
576 	if ((flg & CH_ARP) == 0)
577 		sa->ch_blked = 1;
578 	rv = 1;
579 
580 error:
581 	/*
582 	 * unmap the physical addresses allocated earlier.
583 	 */
584 	cmp = hmp;
585 	for (--nseg; nseg >= 0; nseg--) {
586 		if (cmp->ce_dh) {
587 			if (cmp->ce_flg == DH_DMA)
588 				ch_unbind_dma_handle(sa, cmp->ce_dh);
589 #if defined(__sparc)
590 			else
591 				ch_unbind_dvma_handle(sa, cmp->ce_dh);
592 #endif
593 		}
594 		cmp++;
595 	}
596 
597 error1:
598 
599 	/* free the temporary array */
600 	if (cm_flg)
601 		kmem_free(hmp, mseg * sizeof (cmdQ_ce_t));
602 
603 	/*
604 	 * if we've allocated an mblk above, then we need to free it
605 	 * before returning. This is safe since we haven't done anything to
606 	 * the original message. The caller, gld, will still have a pointer
607 	 * to the original mblk.
608 	 */
609 	if (rv == 1) {
610 		if (freeme) {
611 			/* we had to allocate an mblk. Free it. */
612 			freeb(m0);
613 		} else {
614 			/* adjust the mblk back to original start */
615 			if (flg & CH_NO_CPL)
616 				m0->b_rptr += SZ_CPL_TX_PKT;
617 		}
618 	} else {
619 		freemsg(m0);
620 		sa->oerr++;
621 	}
622 
623 	return (rv);
624 }
625 
626 /* KLUDGE ALERT. HARD WIRED TO PORT ZERO */
627 void
628 pe_set_mac(ch_t *sa, unsigned char *ac_enaddr)
629 {
630 	sa->port[0].mac->ops->macaddress_set(sa->port[0].mac, ac_enaddr);
631 }
632 
633 /* KLUDGE ALERT. HARD WIRED TO PORT ZERO */
634 unsigned char *
635 pe_get_mac(ch_t *sa)
636 {
637 	return (sa->port[0].enaddr);
638 }
639 
640 /* KLUDGE ALERT. HARD WIRED TO ONE PORT */
641 void
642 pe_set_promiscuous(ch_t *sa, int flag)
643 {
644 	struct cmac *mac = sa->port[0].mac;
645 	struct t1_rx_mode rm;
646 
647 	switch (flag) {
648 	case 0:		/* turn off promiscuous mode */
649 		sa->ch_flags &= ~(PEPROMISC|PEALLMULTI);
650 		break;
651 
652 	case 1:		/* turn on promiscuous mode */
653 		sa->ch_flags |= PEPROMISC;
654 		break;
655 
656 	case 2:		/* turn on multicast reception */
657 		sa->ch_flags |= PEALLMULTI;
658 		break;
659 	}
660 
661 	mutex_enter(&sa->ch_mc_lck);
662 	rm.chp = sa;
663 	rm.mc = sa->ch_mc;
664 
665 	mac->ops->set_rx_mode(mac, &rm);
666 	mutex_exit(&sa->ch_mc_lck);
667 }
668 
669 int
670 pe_set_mc(ch_t *sa, uint8_t *ep, int flg)
671 {
672 	struct cmac *mac = sa->port[0].mac;
673 	struct t1_rx_mode rm;
674 
675 	if (flg == GLD_MULTI_ENABLE) {
676 		ch_mc_t *mcp;
677 
678 		mcp = (ch_mc_t *)kmem_zalloc(sizeof (struct ch_mc),
679 		    KM_NOSLEEP);
680 		if (mcp == NULL)
681 			return (GLD_NORESOURCES);
682 
683 		bcopy(ep, &mcp->cmc_mca, 6);
684 
685 		mutex_enter(&sa->ch_mc_lck);
686 		mcp->cmc_next = sa->ch_mc;
687 		sa->ch_mc = mcp;
688 		sa->ch_mc_cnt++;
689 		mutex_exit(&sa->ch_mc_lck);
690 
691 	} else if (flg == GLD_MULTI_DISABLE) {
692 		ch_mc_t **p = &sa->ch_mc;
693 		ch_mc_t *q = NULL;
694 
695 		mutex_enter(&sa->ch_mc_lck);
696 		p = &sa->ch_mc;
697 		while (*p) {
698 			if (bcmp(ep, (*p)->cmc_mca, 6) == 0) {
699 				q = *p;
700 				*p = (*p)->cmc_next;
701 				kmem_free(q, sizeof (*q));
702 				sa->ch_mc_cnt--;
703 				break;
704 			}
705 
706 			p = &(*p)->cmc_next;
707 		}
708 		mutex_exit(&sa->ch_mc_lck);
709 
710 		if (q == NULL)
711 			return (GLD_BADARG);
712 	} else
713 		return (GLD_BADARG);
714 
715 	mutex_enter(&sa->ch_mc_lck);
716 	rm.chp = sa;
717 	rm.mc = sa->ch_mc;
718 
719 	mac->ops->set_rx_mode(mac, &rm);
720 	mutex_exit(&sa->ch_mc_lck);
721 
722 	return (GLD_SUCCESS);
723 }
724 
725 /*
726  * return: speed       - bandwidth of interface
727  * return: intrcnt     - # interrupts
728  * return: norcvbuf    - # recedived packets dropped by driver
729  * return: oerrors     - # bad send packets
730  * return: ierrors     - # bad receive packets
731  * return: underrun    - # bad underrun xmit packets
732  * return: overrun     - # bad overrun recv packets
733  * return: framing     - # bad aligned recv packets
734  * return: crc         - # bad FCS (crc) recv packets
735  * return: carrier     - times carrier was lost
736  * return: collisions  - # xmit collisions
737  * return: xcollisions - # xmit pkts dropped due to collisions
738  * return: late        - # late xmit collisions
739  * return: defer       - # deferred xmit packets
740  * return: xerrs       - # xmit dropped packets
741  * return: rerrs       - # recv dropped packets
742  * return: toolong     - # recv pkts too long
743  * return: runt        - # recv runt pkts
744  * return: multixmt    - # multicast pkts xmitted
745  * return: multircv    - # multicast pkts recved
746  * return: brdcstxmt   - # broadcast pkts xmitted
747  * return: brdcstrcv   - # broadcast pkts rcv
748  */
749 
750 int
751 pe_get_stats(ch_t *sa, uint64_t *speed, uint32_t *intrcnt, uint32_t *norcvbuf,
752     uint32_t *oerrors, uint32_t *ierrors, uint32_t *underrun,
753     uint32_t *overrun, uint32_t *framing, uint32_t *crc,
754     uint32_t *carrier, uint32_t *collisions, uint32_t *xcollisions,
755     uint32_t *late, uint32_t *defer, uint32_t *xerrs, uint32_t *rerrs,
756     uint32_t *toolong, uint32_t *runt, ulong_t  *multixmt, ulong_t  *multircv,
757     ulong_t  *brdcstxmt, ulong_t  *brdcstrcv)
758 {
759 	struct pe_port_t *pt;
760 	int line_speed;
761 	int line_duplex;
762 	int line_is_active;
763 	uint64_t v;
764 	const struct cmac_statistics *sp;
765 
766 	pt = &(sa->port[0]);
767 	(void) pt->phy->ops->get_link_status(pt->phy,
768 	    &line_is_active, &line_speed, &line_duplex, NULL);
769 
770 	switch (line_speed) {
771 	case SPEED_10:
772 		*speed = 10000000;
773 		break;
774 	case SPEED_100:
775 		*speed = 100000000;
776 		break;
777 	case SPEED_1000:
778 		*speed = 1000000000;
779 		break;
780 	case SPEED_10000:
781 		/*
782 		 * kludge to get 10,000,000,000 constant (and keep
783 		 * compiler happy).
784 		 */
785 		v = 10000000;
786 		v *= 1000;
787 		*speed = v;
788 		break;
789 	default:
790 		goto error;
791 	}
792 
793 	*intrcnt = sa->isr_intr;
794 	*norcvbuf = sa->norcvbuf;
795 
796 	sp = sa->port[0].mac->ops->statistics_update(sa->port[0].mac,
797 	    MAC_STATS_UPDATE_FULL);
798 
799 	*ierrors = sp->RxOctetsBad;
800 
801 	/*
802 	 * not sure this is correct. # aborted at driver level +
803 	 * # at hardware level
804 	 */
805 	*oerrors = sa->oerr + sp->TxFramesAbortedDueToXSCollisions +
806 	    sp->TxUnderrun + sp->TxLengthErrors +
807 	    sp->TxInternalMACXmitError +
808 	    sp->TxFramesWithExcessiveDeferral +
809 	    sp->TxFCSErrors;
810 
811 	*underrun = sp->TxUnderrun;
812 	*overrun = sp->RxFrameTooLongErrors;
813 	*framing = sp->RxAlignErrors;
814 	*crc = sp->RxFCSErrors;
815 	*carrier = 0;		/* need to find this */
816 	*collisions = sp->TxTotalCollisions;
817 	*xcollisions = sp->TxFramesAbortedDueToXSCollisions;
818 	*late = sp->TxLateCollisions;
819 	*defer = sp->TxFramesWithDeferredXmissions;
820 	*xerrs = sp->TxUnderrun + sp->TxLengthErrors +
821 	    sp->TxInternalMACXmitError + sp->TxFCSErrors;
822 	*rerrs = sp->RxSymbolErrors + sp->RxSequenceErrors + sp->RxRuntErrors +
823 	    sp->RxJabberErrors + sp->RxInternalMACRcvError +
824 	    sp->RxInRangeLengthErrors + sp->RxOutOfRangeLengthField;
825 	*toolong = sp->RxFrameTooLongErrors;
826 	*runt = sp->RxRuntErrors;
827 
828 	*multixmt = sp->TxMulticastFramesOK;
829 	*multircv = sp->RxMulticastFramesOK;
830 	*brdcstxmt = sp->TxBroadcastFramesOK;
831 	*brdcstrcv = sp->RxBroadcastFramesOK;
832 
833 	return (0);
834 
835 error:
836 	*speed = 0;
837 	*intrcnt = 0;
838 	*norcvbuf = 0;
839 	*norcvbuf = 0;
840 	*oerrors = 0;
841 	*ierrors = 0;
842 	*underrun = 0;
843 	*overrun = 0;
844 	*framing = 0;
845 	*crc = 0;
846 	*carrier = 0;
847 	*collisions = 0;
848 	*xcollisions = 0;
849 	*late = 0;
850 	*defer = 0;
851 	*xerrs = 0;
852 	*rerrs = 0;
853 	*toolong = 0;
854 	*runt = 0;
855 	*multixmt = 0;
856 	*multircv = 0;
857 	*brdcstxmt = 0;
858 	*brdcstrcv = 0;
859 
860 	return (1);
861 }
862 
863 uint32_t ch_gtm = 0;		/* Default: Global Tunnel Mode off */
864 uint32_t ch_global_config = 0x07000000;	/* Default: errors, warnings, status */
865 uint32_t ch_is_asic = 0;	/* Default: non-ASIC */
866 uint32_t ch_link_speed = PE_LINK_SPEED_AUTONEG;	/* Default: auto-negoiate */
867 uint32_t ch_num_of_ports = 1;	/* Default: 1 port */
868 uint32_t ch_tp_reset_cm = 1;	/* Default: reset CM memory map */
869 uint32_t ch_phy_tx_fifo = 0;	/* Default: 0 phy tx fifo depth */
870 uint32_t ch_phy_rx_fifo = 0;	/* Default: 0 phy rx fifo depth */
871 uint32_t ch_phy_force_master = 1;	/* Default: link always master mode */
872 uint32_t ch_mc5_rtbl_size = 2048;	/* Default: TCAM routing table size */
873 uint32_t ch_mc5_dbsvr_size = 128;	/* Default: TCAM server size */
874 uint32_t ch_mc5_parity = 1;	/* Default: parity error checking */
875 uint32_t ch_mc5_issue_syn = 0;	/* Default: Allow transaction overlap */
876 uint32_t ch_packet_tracing = 0;		/* Default: no packet tracing */
877 uint32_t ch_server_region_len =
878 	DEFAULT_SERVER_REGION_LEN;
879 uint32_t ch_rt_region_len =
880 	DEFAULT_RT_REGION_LEN;
881 uint32_t ch_offload_ip_cksum = 0;	/* Default: no checksum offloading */
882 uint32_t ch_offload_udp_cksum = 1;	/* Default: offload UDP ckecksum */
883 uint32_t ch_offload_tcp_cksum = 1;	/* Default: offload TCP checksum */
884 uint32_t ch_sge_cmdq_threshold = 0;	/* Default: threshold 0 */
885 uint32_t ch_sge_flq_threshold = 0;	/* Default: SGE flq threshold */
886 uint32_t ch_sge_cmdq0_cnt =	/* Default: cmd queue 0 size */
887 	SGE_CMDQ0_CNT;
888 uint32_t ch_sge_cmdq1_cnt =	/* Default: cmd queue 1 size */
889 	SGE_CMDQ0_CNT;
890 uint32_t ch_sge_flq0_cnt =	/* Default: free list queue-0 length */
891 	SGE_FLQ0_CNT;
892 uint32_t ch_sge_flq1_cnt =	/* Default: free list queue-1 length */
893 	SGE_FLQ0_CNT;
894 uint32_t ch_sge_respq_cnt =	/* Default: reqsponse queue size */
895 	SGE_RESPQ_CNT;
896 uint32_t ch_stats = 1;		/* Default: Automatic Update MAC stats */
897 uint32_t ch_tx_delay_us = 0;	/* Default: No Msec delay to Tx pkts */
898 int32_t ch_chip = -1;		/* Default: use hardware lookup tbl */
899 uint32_t ch_exit_early = 0;	/* Default: complete initialization */
900 uint32_t ch_rb_num_of_entries = 1000; /* Default: number ring buffer entries */
901 uint32_t ch_rb_size_of_entries = 64;	/* Default: ring buffer entry size */
902 uint32_t ch_rb_flag = 1;	/* Default: ring buffer flag */
903 uint32_t ch_type;
904 uint64_t ch_cat_opt0 = 0;
905 uint64_t ch_cat_opt1 = 0;
906 uint32_t ch_timer_delay = 0;	/* Default: use value from board entry */
907 
908 int
909 pe_attach(ch_t *chp)
910 {
911 	int return_val = 1;
912 	const struct board_info *bi;
913 	uint32_t pcix_cmd;
914 
915 	(void) ch_set_config_data(chp);
916 
917 	bi = pe_sa_init(chp);
918 	if (bi == 0)
919 		return (1);
920 
921 	if (t1_init_sw_modules(chp, bi) < 0)
922 		return (1);
923 
924 	if (pe_small_rbuf_pool_init(chp) == NULL)
925 		return (1);
926 
927 	if (pe_big_rbuf_pool_init(chp) == NULL)
928 		return (1);
929 
930 	/*
931 	 * We gain significaint performance improvements when we
932 	 * increase the PCI's maximum memory read byte count to
933 	 * 2K(HW doesn't support 4K at this time) and set the PCI's
934 	 * maximum outstanding split transactions to 4. We want to do
935 	 * this for 10G. Done by software utility.
936 	 */
937 
938 	if (board_info(chp)->caps & SUPPORTED_10000baseT_Full) {
939 		(void) t1_os_pci_read_config_4(chp, A_PCICFG_PCIX_CMD,
940 		    &pcix_cmd);
941 		/*
942 		 * if the burstsize is set, then use it instead of default
943 		 */
944 		if (chp->ch_config.burstsize_set) {
945 			pcix_cmd &= ~0xc0000;
946 			pcix_cmd |= (chp->ch_config.burstsize << 18);
947 		}
948 		/*
949 		 * if the split transaction count is set, then use it.
950 		 */
951 		if (chp->ch_config.transaction_cnt_set) {
952 			pcix_cmd &= ~ 0x700000;
953 			pcix_cmd |= (chp->ch_config.transaction_cnt << 20);
954 		}
955 
956 		/*
957 		 * set ralaxed ordering flag as configured in chxge.conf
958 		 */
959 		pcix_cmd |= (chp->ch_config.relaxed_ordering << 17);
960 
961 		(void) t1_os_pci_write_config_4(chp, A_PCICFG_PCIX_CMD,
962 		    pcix_cmd);
963 	}
964 
965 	/*
966 	 * set the latency time to F8 for 10G cards.
967 	 * Done by software utiltiy.
968 	 */
969 	if (enable_latency_timer) {
970 		if (board_info(chp)->caps & SUPPORTED_10000baseT_Full) {
971 			(void) t1_os_pci_write_config_4(chp, 0xc, 0xf800);
972 		}
973 	}
974 
975 	/*
976 	 * update mtu table (regs: 0x404 - 0x420) with bigger values than
977 	 * default.
978 	 */
979 	update_mtu_tab(chp);
980 
981 	/*
982 	 * Clear all interrupts now.  Don't enable
983 	 * them until later.
984 	 */
985 	t1_interrupts_clear(chp);
986 
987 	/*
988 	 * Function succeeded.
989 	 */
990 	return_val = 0;
991 
992 	return (return_val);
993 }
994 
995 /*
996  * DESC: Read variables set in /boot/loader.conf and save
997  *       them internally. These internal values are then
998  *       used to make decisions at run-time on behavior thus
999  *       allowing a certain level of customization.
1000  * OUT:  p_config - pointer to config structure that
1001  *                  contains all of the new values.
1002  * RTN:  0 - Success;
1003  */
1004 static int
1005 ch_set_config_data(ch_t *chp)
1006 {
1007 	pe_config_data_t *p_config = (pe_config_data_t *)&chp->config_data;
1008 
1009 	bzero(p_config, sizeof (pe_config_data_t));
1010 
1011 	/*
1012 	 * Global Tunnel Mode configuration
1013 	 */
1014 	p_config->gtm = ch_gtm;
1015 
1016 	p_config->global_config = ch_global_config;
1017 
1018 	if (p_config->gtm)
1019 		p_config->global_config |= CFGMD_TUNNEL;
1020 
1021 	p_config->tp_reset_cm = ch_tp_reset_cm;
1022 	p_config->is_asic = ch_is_asic;
1023 
1024 	/*
1025 	 * MC5 configuration.
1026 	 */
1027 	p_config->mc5_rtbl_size = ch_mc5_rtbl_size;
1028 	p_config->mc5_dbsvr_size = ch_mc5_dbsvr_size;
1029 	p_config->mc5_parity = ch_mc5_parity;
1030 	p_config->mc5_issue_syn = ch_mc5_issue_syn;
1031 
1032 	p_config->offload_ip_cksum = ch_offload_ip_cksum;
1033 	p_config->offload_udp_cksum = ch_offload_udp_cksum;
1034 	p_config->offload_tcp_cksum = ch_offload_tcp_cksum;
1035 
1036 	p_config->packet_tracing = ch_packet_tracing;
1037 
1038 	p_config->server_region_len = ch_server_region_len;
1039 	p_config->rt_region_len = ch_rt_region_len;
1040 
1041 	/*
1042 	 * Link configuration.
1043 	 *
1044 	 * 5-auto-neg 2-1000Gbps; 1-100Gbps; 0-10Gbps
1045 	 */
1046 	p_config->link_speed = ch_link_speed;
1047 	p_config->num_of_ports = ch_num_of_ports;
1048 
1049 	/*
1050 	 * Catp options
1051 	 */
1052 	p_config->cat_opt0 = ch_cat_opt0;
1053 	p_config->cat_opt1 = ch_cat_opt1;
1054 
1055 	/*
1056 	 * SGE configuration.
1057 	 */
1058 	p_config->sge_cmdq0_cnt = ch_sge_cmdq0_cnt;
1059 	p_config->sge_cmdq1_cnt = ch_sge_cmdq1_cnt;
1060 	p_config->sge_flq0_cnt = ch_sge_flq0_cnt;
1061 	p_config->sge_flq1_cnt = ch_sge_flq1_cnt;
1062 	p_config->sge_respq_cnt = ch_sge_respq_cnt;
1063 
1064 	p_config->phy_rx_fifo = ch_phy_rx_fifo;
1065 	p_config->phy_tx_fifo = ch_phy_tx_fifo;
1066 
1067 	p_config->sge_cmdq_threshold = ch_sge_cmdq_threshold;
1068 
1069 	p_config->sge_flq_threshold = ch_sge_flq_threshold;
1070 
1071 	p_config->phy_force_master = ch_phy_force_master;
1072 
1073 	p_config->rb_num_of_entries = ch_rb_num_of_entries;
1074 
1075 	p_config->rb_size_of_entries = ch_rb_size_of_entries;
1076 
1077 	p_config->rb_flag = ch_rb_flag;
1078 
1079 	p_config->exit_early = ch_exit_early;
1080 
1081 	p_config->chip = ch_chip;
1082 
1083 	p_config->stats = ch_stats;
1084 
1085 	p_config->tx_delay_us = ch_tx_delay_us;
1086 
1087 	return (0);
1088 }
1089 
1090 static const struct board_info *
1091 pe_sa_init(ch_t *sa)
1092 {
1093 	uint16_t device_id;
1094 	uint16_t device_subid;
1095 	const struct board_info *bi;
1096 
1097 	sa->config = sa->config_data.global_config;
1098 	device_id = pci_config_get16(sa->ch_hpci, 2);
1099 	device_subid = pci_config_get16(sa->ch_hpci, 0x2e);
1100 
1101 	bi = t1_get_board_info_from_ids(device_id, device_subid);
1102 	if (bi == NULL) {
1103 		cmn_err(CE_NOTE,
1104 		    "The adapter with device_id %d %d is not supported.\n",
1105 		    device_id, device_subid);
1106 		return (NULL);
1107 	}
1108 
1109 	if (t1_get_board_rev(sa, bi, &sa->params)) {
1110 		cmn_err(CE_NOTE, "unknown device_id %d %d\n",
1111 		    device_id, device_subid);
1112 		return ((const struct board_info *)NULL);
1113 	}
1114 
1115 	return (bi);
1116 }
1117 
1118 /*
1119  * allocate pool of small receive buffers (with vaddr & paddr) and
1120  * receiver buffer control structure (ch_esb_t *rbp).
1121  * XXX we should allow better tuning of the # of preallocated
1122  * free buffers against the # of freelist entries.
1123  */
1124 static int
1125 pe_small_rbuf_pool_init(ch_t *sa)
1126 {
1127 	int i;
1128 	ch_esb_t *rbp;
1129 	extern uint32_t sge_flq0_cnt;
1130 	extern uint32_t sge_flq1_cnt;
1131 	int size;
1132 	uint32_t j;
1133 
1134 	if (is_T2(sa))
1135 		size = sge_flq1_cnt * fl_sz_multiplier;
1136 	else
1137 		size = sge_flq0_cnt * fl_sz_multiplier;
1138 
1139 	mutex_init(&sa->ch_small_esbl, NULL, MUTEX_DRIVER, sa->ch_icookp);
1140 
1141 	mutex_enter(&in_use_l);
1142 	j = in_use_index++;
1143 	if (in_use_index >= SZ_INUSE)
1144 		in_use_index = 0;
1145 	mutex_exit(&in_use_l);
1146 
1147 	sa->ch_small_owner = NULL;
1148 	sa->ch_sm_index = j;
1149 	sa->ch_small_esb_free = NULL;
1150 	for (i = 0; i < size; i++) {
1151 		rbp = ch_alloc_small_esbbuf(sa, j);
1152 		if (rbp == NULL)
1153 			goto error;
1154 		/*
1155 		 * add entry to free list
1156 		 */
1157 		rbp->cs_next = sa->ch_small_esb_free;
1158 		sa->ch_small_esb_free = rbp;
1159 
1160 		/*
1161 		 * add entry to owned list
1162 		 */
1163 		rbp->cs_owner = sa->ch_small_owner;
1164 		sa->ch_small_owner = rbp;
1165 	}
1166 	return (1);
1167 
1168 error:
1169 	sa->ch_small_owner = NULL;
1170 
1171 	/* free whatever we've already allocated */
1172 	pe_rbuf_pool_free(sa);
1173 
1174 	return (0);
1175 }
1176 
1177 /*
1178  * allocate pool of receive buffers (with vaddr & paddr) and
1179  * receiver buffer control structure (ch_esb_t *rbp).
1180  * XXX we should allow better tuning of the # of preallocated
1181  * free buffers against the # of freelist entries.
1182  */
1183 static int
1184 pe_big_rbuf_pool_init(ch_t *sa)
1185 {
1186 	int i;
1187 	ch_esb_t *rbp;
1188 	extern uint32_t sge_flq0_cnt;
1189 	extern uint32_t sge_flq1_cnt;
1190 	int size;
1191 	uint32_t j;
1192 
1193 	if (is_T2(sa))
1194 		size = sge_flq0_cnt * fl_sz_multiplier;
1195 	else
1196 		size = sge_flq1_cnt * fl_sz_multiplier;
1197 
1198 	mutex_init(&sa->ch_big_esbl, NULL, MUTEX_DRIVER, sa->ch_icookp);
1199 
1200 	mutex_enter(&in_use_l);
1201 	j = in_use_index++;
1202 	if (in_use_index >= SZ_INUSE)
1203 		in_use_index = 0;
1204 	mutex_exit(&in_use_l);
1205 
1206 	sa->ch_big_owner = NULL;
1207 	sa->ch_big_index = j;
1208 	sa->ch_big_esb_free = NULL;
1209 	for (i = 0; i < size; i++) {
1210 		rbp = ch_alloc_big_esbbuf(sa, j);
1211 		if (rbp == NULL)
1212 			goto error;
1213 		rbp->cs_next = sa->ch_big_esb_free;
1214 		sa->ch_big_esb_free = rbp;
1215 
1216 		/*
1217 		 * add entry to owned list
1218 		 */
1219 		rbp->cs_owner = sa->ch_big_owner;
1220 		sa->ch_big_owner = rbp;
1221 	}
1222 	return (1);
1223 
1224 error:
1225 	sa->ch_big_owner = NULL;
1226 
1227 	/* free whatever we've already allocated */
1228 	pe_rbuf_pool_free(sa);
1229 
1230 	return (0);
1231 }
1232 
1233 /*
1234  * allocate receive buffer structure and dma mapped buffer (SGE_SM_BUF_SZ bytes)
1235  * note that we will DMA at a 2 byte offset for Solaris when checksum offload
1236  * is enabled.
1237  */
1238 static ch_esb_t *
1239 ch_alloc_small_esbbuf(ch_t *sa, uint32_t i)
1240 {
1241 	ch_esb_t *rbp;
1242 
1243 	rbp = (ch_esb_t *)kmem_zalloc(sizeof (ch_esb_t), KM_SLEEP);
1244 	if (rbp == NULL) {
1245 		return ((ch_esb_t *)0);
1246 	}
1247 
1248 #if BYTE_ORDER == BIG_ENDIAN
1249 	rbp->cs_buf = (caddr_t)ch_alloc_dma_mem(sa, 1, DMA_STREAM|DMA_SMALN,
1250 	    SGE_SM_BUF_SZ(sa), &rbp->cs_pa, &rbp->cs_dh, &rbp->cs_ah);
1251 #else
1252 	rbp->cs_buf = (caddr_t)ch_alloc_dma_mem(sa, 0, DMA_STREAM|DMA_SMALN,
1253 	    SGE_SM_BUF_SZ(sa), &rbp->cs_pa, &rbp->cs_dh, &rbp->cs_ah);
1254 #endif
1255 
1256 	if (rbp->cs_buf == NULL) {
1257 		kmem_free(rbp, sizeof (ch_esb_t));
1258 		return ((ch_esb_t *)0);
1259 	}
1260 
1261 	rbp->cs_sa = sa;
1262 	rbp->cs_index = i;
1263 
1264 	rbp->cs_frtn.free_func = (void (*)())&ch_small_rbuf_recycle;
1265 	rbp->cs_frtn.free_arg  = (caddr_t)rbp;
1266 
1267 	return (rbp);
1268 }
1269 
1270 /*
1271  * allocate receive buffer structure and dma mapped buffer (SGE_BG_BUF_SZ bytes)
1272  * note that we will DMA at a 2 byte offset for Solaris when checksum offload
1273  * is enabled.
1274  */
1275 static ch_esb_t *
1276 ch_alloc_big_esbbuf(ch_t *sa, uint32_t i)
1277 {
1278 	ch_esb_t *rbp;
1279 
1280 	rbp = (ch_esb_t *)kmem_zalloc(sizeof (ch_esb_t), KM_SLEEP);
1281 	if (rbp == NULL) {
1282 		return ((ch_esb_t *)0);
1283 	}
1284 
1285 #if BYTE_ORDER == BIG_ENDIAN
1286 	rbp->cs_buf = (caddr_t)ch_alloc_dma_mem(sa, 1, DMA_STREAM|DMA_BGALN,
1287 	    SGE_BG_BUF_SZ(sa), &rbp->cs_pa, &rbp->cs_dh, &rbp->cs_ah);
1288 #else
1289 	rbp->cs_buf = (caddr_t)ch_alloc_dma_mem(sa, 0, DMA_STREAM|DMA_BGALN,
1290 	    SGE_BG_BUF_SZ(sa), &rbp->cs_pa, &rbp->cs_dh, &rbp->cs_ah);
1291 #endif
1292 
1293 	if (rbp->cs_buf == NULL) {
1294 		kmem_free(rbp, sizeof (ch_esb_t));
1295 		return ((ch_esb_t *)0);
1296 	}
1297 
1298 	rbp->cs_sa = sa;
1299 	rbp->cs_index = i;
1300 
1301 	rbp->cs_frtn.free_func = (void (*)())&ch_big_rbuf_recycle;
1302 	rbp->cs_frtn.free_arg  = (caddr_t)rbp;
1303 
1304 	return (rbp);
1305 }
1306 
1307 /*
1308  * free entries on the receive buffer list.
1309  */
1310 void
1311 pe_rbuf_pool_free(ch_t *sa)
1312 {
1313 	ch_esb_t *rbp;
1314 
1315 	mutex_enter(&sa->ch_small_esbl);
1316 
1317 	/*
1318 	 * Now set-up the rest to commit suicide.
1319 	 */
1320 	while (sa->ch_small_owner) {
1321 		rbp = sa->ch_small_owner;
1322 		sa->ch_small_owner = rbp->cs_owner;
1323 		rbp->cs_owner = NULL;
1324 		rbp->cs_flag = 1;
1325 	}
1326 
1327 	while ((rbp = sa->ch_small_esb_free) != NULL) {
1328 		/* advance head ptr to next entry */
1329 		sa->ch_small_esb_free = rbp->cs_next;
1330 		/* free private buffer allocated in ch_alloc_esbbuf() */
1331 		ch_free_dma_mem(rbp->cs_dh, rbp->cs_ah);
1332 		/* free descripter buffer */
1333 		kmem_free(rbp, sizeof (ch_esb_t));
1334 	}
1335 
1336 	mutex_exit(&sa->ch_small_esbl);
1337 
1338 	/* destroy ch_esbl lock */
1339 	mutex_destroy(&sa->ch_small_esbl);
1340 
1341 
1342 	mutex_enter(&sa->ch_big_esbl);
1343 
1344 	/*
1345 	 * Now set-up the rest to commit suicide.
1346 	 */
1347 	while (sa->ch_big_owner) {
1348 		rbp = sa->ch_big_owner;
1349 		sa->ch_big_owner = rbp->cs_owner;
1350 		rbp->cs_owner = NULL;
1351 		rbp->cs_flag = 1;
1352 	}
1353 
1354 	while ((rbp = sa->ch_big_esb_free) != NULL) {
1355 		/* advance head ptr to next entry */
1356 		sa->ch_big_esb_free = rbp->cs_next;
1357 		/* free private buffer allocated in ch_alloc_esbbuf() */
1358 		ch_free_dma_mem(rbp->cs_dh, rbp->cs_ah);
1359 		/* free descripter buffer */
1360 		kmem_free(rbp, sizeof (ch_esb_t));
1361 	}
1362 
1363 	mutex_exit(&sa->ch_big_esbl);
1364 
1365 	/* destroy ch_esbl lock */
1366 	mutex_destroy(&sa->ch_big_esbl);
1367 }
1368 
1369 void
1370 ch_small_rbuf_recycle(ch_esb_t *rbp)
1371 {
1372 	ch_t *sa = rbp->cs_sa;
1373 
1374 	if (rbp->cs_flag) {
1375 		uint32_t i;
1376 		/*
1377 		 * free private buffer allocated in ch_alloc_esbbuf()
1378 		 */
1379 		ch_free_dma_mem(rbp->cs_dh, rbp->cs_ah);
1380 
1381 		i = rbp->cs_index;
1382 
1383 		/*
1384 		 * free descripter buffer
1385 		 */
1386 		kmem_free(rbp, sizeof (ch_esb_t));
1387 
1388 		/*
1389 		 * decrement count of receive buffers freed by callback
1390 		 * We decrement here so anyone trying to do fini will
1391 		 * only remove the driver once the counts go to 0.
1392 		 */
1393 		atomic_add_32(&buffers_in_use[i], -1);
1394 
1395 		return;
1396 	}
1397 
1398 	mutex_enter(&sa->ch_small_esbl);
1399 	rbp->cs_next = sa->ch_small_esb_free;
1400 	sa->ch_small_esb_free = rbp;
1401 	mutex_exit(&sa->ch_small_esbl);
1402 
1403 	/*
1404 	 * decrement count of receive buffers freed by callback
1405 	 */
1406 	atomic_add_32(&buffers_in_use[rbp->cs_index], -1);
1407 }
1408 
1409 /*
1410  * callback function from freeb() when esballoced mblk freed.
1411  */
1412 void
1413 ch_big_rbuf_recycle(ch_esb_t *rbp)
1414 {
1415 	ch_t *sa = rbp->cs_sa;
1416 
1417 	if (rbp->cs_flag) {
1418 		uint32_t i;
1419 		/*
1420 		 * free private buffer allocated in ch_alloc_esbbuf()
1421 		 */
1422 		ch_free_dma_mem(rbp->cs_dh, rbp->cs_ah);
1423 
1424 		i = rbp->cs_index;
1425 
1426 		/*
1427 		 * free descripter buffer
1428 		 */
1429 		kmem_free(rbp, sizeof (ch_esb_t));
1430 
1431 		/*
1432 		 * decrement count of receive buffers freed by callback
1433 		 * We decrement here so anyone trying to do fini will
1434 		 * only remove the driver once the counts go to 0.
1435 		 */
1436 		atomic_add_32(&buffers_in_use[i], -1);
1437 
1438 		return;
1439 	}
1440 
1441 	mutex_enter(&sa->ch_big_esbl);
1442 	rbp->cs_next = sa->ch_big_esb_free;
1443 	sa->ch_big_esb_free = rbp;
1444 	mutex_exit(&sa->ch_big_esbl);
1445 
1446 	/*
1447 	 * decrement count of receive buffers freed by callback
1448 	 */
1449 	atomic_add_32(&buffers_in_use[rbp->cs_index], -1);
1450 }
1451 
1452 /*
1453  * get a pre-allocated, pre-mapped receive buffer from free list.
1454  * (used sge.c)
1455  */
1456 ch_esb_t *
1457 ch_get_small_rbuf(ch_t *sa)
1458 {
1459 	ch_esb_t *rbp;
1460 
1461 	mutex_enter(&sa->ch_small_esbl);
1462 	rbp = sa->ch_small_esb_free;
1463 	if (rbp) {
1464 		sa->ch_small_esb_free = rbp->cs_next;
1465 	}
1466 	mutex_exit(&sa->ch_small_esbl);
1467 
1468 	return (rbp);
1469 }
1470 
1471 /*
1472  * get a pre-allocated, pre-mapped receive buffer from free list.
1473  * (used sge.c)
1474  */
1475 
1476 ch_esb_t *
1477 ch_get_big_rbuf(ch_t *sa)
1478 {
1479 	ch_esb_t *rbp;
1480 
1481 	mutex_enter(&sa->ch_big_esbl);
1482 	rbp = sa->ch_big_esb_free;
1483 	if (rbp) {
1484 		sa->ch_big_esb_free = rbp->cs_next;
1485 	}
1486 	mutex_exit(&sa->ch_big_esbl);
1487 
1488 	return (rbp);
1489 }
1490 
1491 void
1492 pe_detach(ch_t *sa)
1493 {
1494 	(void) sge_stop(sa->sge);
1495 
1496 	pe_free_driver_resources(sa);
1497 }
1498 
1499 static void
1500 pe_free_driver_resources(ch_t *sa)
1501 {
1502 	if (sa) {
1503 		t1_free_sw_modules(sa);
1504 
1505 		/* free pool of receive buffers */
1506 		pe_rbuf_pool_free(sa);
1507 	}
1508 }
1509 
1510 /*
1511  * Processes elmer0 external interrupts in process context.
1512  */
1513 static void
1514 ext_intr_task(ch_t *adapter)
1515 {
1516 	u32 enable;
1517 
1518 	(void) elmer0_ext_intr_handler(adapter);
1519 
1520 	/* Now reenable external interrupts */
1521 	t1_write_reg_4(adapter, A_PL_CAUSE, F_PL_INTR_EXT);
1522 	enable = t1_read_reg_4(adapter, A_PL_ENABLE);
1523 	t1_write_reg_4(adapter, A_PL_ENABLE, enable | F_PL_INTR_EXT);
1524 	adapter->slow_intr_mask |= F_PL_INTR_EXT;
1525 }
1526 
1527 /*
1528  * Interrupt-context handler for elmer0 external interrupts.
1529  */
1530 void
1531 t1_os_elmer0_ext_intr(ch_t *adapter)
1532 {
1533 	u32 enable = t1_read_reg_4(adapter, A_PL_ENABLE);
1534 
1535 	adapter->slow_intr_mask &= ~F_PL_INTR_EXT;
1536 	t1_write_reg_4(adapter, A_PL_ENABLE, enable & ~F_PL_INTR_EXT);
1537 #ifdef NOTYET
1538 	schedule_work(&adapter->ext_intr_handler_task);
1539 #else
1540 	ext_intr_task(adapter);
1541 #endif
1542 }
1543 
1544 uint8_t *
1545 t1_get_next_mcaddr(struct t1_rx_mode *rmp)
1546 {
1547 	uint8_t *addr = 0;
1548 	if (rmp->mc) {
1549 		addr = rmp->mc->cmc_mca;
1550 		rmp->mc = rmp->mc->cmc_next;
1551 	}
1552 	return (addr);
1553 }
1554 
1555 void
1556 pe_dma_handle_init(ch_t *chp, int cnt)
1557 {
1558 	free_dh_t *dhe;
1559 #if defined(__sparc)
1560 	int tcnt = cnt/2;
1561 
1562 	for (; cnt; cnt--) {
1563 		dhe = ch_get_dvma_handle(chp);
1564 		if (dhe == NULL)
1565 			break;
1566 		mutex_enter(&chp->ch_dh_lck);
1567 		dhe->dhe_next = chp->ch_vdh;
1568 		chp->ch_vdh = dhe;
1569 		mutex_exit(&chp->ch_dh_lck);
1570 	}
1571 
1572 	cnt += tcnt;
1573 #endif
1574 	while (cnt--) {
1575 		dhe = ch_get_dma_handle(chp);
1576 		if (dhe == NULL)
1577 			return;
1578 		mutex_enter(&chp->ch_dh_lck);
1579 		dhe->dhe_next = chp->ch_dh;
1580 		chp->ch_dh = dhe;
1581 		mutex_exit(&chp->ch_dh_lck);
1582 	}
1583 }
1584 
1585 /*
1586  * Write new values to the MTU table.  Caller must validate that the new MTUs
1587  * are in ascending order. params.mtus[] is initialized by init_mtus()
1588  * called in t1_init_sw_modules().
1589  */
1590 #define	MTUREG(idx) (A_TP_MTU_REG0 + (idx) * 4)
1591 
1592 static void
1593 update_mtu_tab(ch_t *adapter)
1594 {
1595 	int i;
1596 
1597 	for (i = 0; i < NMTUS; ++i) {
1598 		int mtu = (unsigned int)adapter->params.mtus[i];
1599 
1600 		t1_write_reg_4(adapter, MTUREG(i), mtu);
1601 	}
1602 }
1603 
1604 static int
1605 pe_change_mtu(ch_t *chp)
1606 {
1607 	struct cmac *mac = chp->port[0].mac;
1608 	int ret;
1609 
1610 	if (!mac->ops->set_mtu) {
1611 		return (EOPNOTSUPP);
1612 	}
1613 	if (chp->ch_mtu < 68) {
1614 		return (EINVAL);
1615 	}
1616 	if (ret = mac->ops->set_mtu(mac, chp->ch_mtu)) {
1617 		return (ret);
1618 	}
1619 
1620 	return (0);
1621 }
1622 
1623 typedef struct fake_arp {
1624 	char fa_dst[6];		/* ethernet header */
1625 	char fa_src[6];		/* ethernet header */
1626 	ushort_t fa_typ;		/* ethernet header */
1627 
1628 	ushort_t fa_hrd;		/* arp */
1629 	ushort_t fa_pro;
1630 	char fa_hln;
1631 	char fa_pln;
1632 	ushort_t fa_op;
1633 	char fa_src_mac[6];
1634 	uint_t fa_src_ip;
1635 	char fa_dst_mac[6];
1636 	char fa_dst_ip[4];
1637 } fake_arp_t;
1638 
1639 /*
1640  * PR2928 & PR3309
1641  * construct packet in mblk and attach it to sge structure.
1642  */
1643 static int
1644 pe_make_fake_arp(ch_t *chp, unsigned char *arpp)
1645 {
1646 	pesge *sge = chp->sge;
1647 	mblk_t *bp;
1648 	fake_arp_t *fap;
1649 	static char buf[6] = {0, 7, 0x43, 0, 0, 0};
1650 	struct cpl_tx_pkt *cpl;
1651 
1652 	bp = allocb(sizeof (struct fake_arp) + SZ_CPL_TX_PKT, BPRI_HI);
1653 	if (bp == NULL) {
1654 		return (1);
1655 	}
1656 	bzero(bp->b_rptr, sizeof (struct fake_arp) + SZ_CPL_TX_PKT);
1657 
1658 	/* fill in cpl header */
1659 	cpl = (struct cpl_tx_pkt *)bp->b_rptr;
1660 	cpl->opcode = CPL_TX_PKT;
1661 	cpl->iff = 0;			/* XXX port 0 needs fixing with NEMO */
1662 	cpl->ip_csum_dis = 1;		/* no IP header cksum */
1663 	cpl->l4_csum_dis = 1;		/* no tcp/udp cksum */
1664 	cpl->vlan_valid = 0;		/* no vlan */
1665 
1666 	fap = (fake_arp_t *)&bp->b_rptr[SZ_CPL_TX_PKT];
1667 
1668 	bcopy(arpp, fap, sizeof (*fap));	/* copy first arp to mblk */
1669 
1670 	bcopy(buf, fap->fa_dst, 6);		/* overwrite dst mac */
1671 	chp->ch_ip = fap->fa_src_ip;		/* not used yet */
1672 	bcopy(buf, fap->fa_dst_mac, 6);		/* overwrite dst mac */
1673 
1674 	bp->b_wptr = bp->b_rptr + sizeof (struct fake_arp)+SZ_CPL_TX_PKT;
1675 
1676 	sge_add_fake_arp(sge, (void *)bp);
1677 
1678 	return (0);
1679 }
1680 
1681 /*
1682  * PR2928 & PR3309
1683  * free the fake arp's mblk on sge structure.
1684  */
1685 void
1686 pe_free_fake_arp(void *arp)
1687 {
1688 	mblk_t *bp = (mblk_t *)(arp);
1689 
1690 	freemsg(bp);
1691 }
1692 
1693 /*
1694  * extract ip address of nic from first outgoing arp.
1695  */
1696 static uint32_t
1697 pe_get_ip(unsigned char *arpp)
1698 {
1699 	fake_arp_t fap;
1700 
1701 	/*
1702 	 * first copy packet to buffer so we know
1703 	 * it will be properly aligned.
1704 	 */
1705 	bcopy(arpp, &fap, sizeof (fap));	/* copy first arp to buffer */
1706 	return (fap.fa_src_ip);
1707 }
1708 
1709 /* ARGSUSED */
1710 void
1711 t1_os_link_changed(ch_t *obj, int port_id, int link_status,
1712     int speed, int duplex, int fc)
1713 {
1714 	gld_mac_info_t *macinfo = obj->ch_macp;
1715 	if (link_status) {
1716 		gld_linkstate(macinfo, GLD_LINKSTATE_UP);
1717 		/*
1718 		 * Link states should be reported to user
1719 		 * whenever it changes
1720 		 */
1721 		cmn_err(CE_NOTE, "%s: link is up", adapter_name(obj));
1722 	} else {
1723 		gld_linkstate(macinfo, GLD_LINKSTATE_DOWN);
1724 		/*
1725 		 * Link states should be reported to user
1726 		 * whenever it changes
1727 		 */
1728 		cmn_err(CE_NOTE, "%s: link is down", adapter_name(obj));
1729 	}
1730 }
1731