xref: /titanic_52/usr/src/uts/common/io/nxge/nxge_rxdma.c (revision 2d6b5ea734bb47d251c82670646fde46af15fd69)
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  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #include <sys/nxge/nxge_impl.h>
27 #include <sys/nxge/nxge_rxdma.h>
28 #include <sys/nxge/nxge_hio.h>
29 
30 #if !defined(_BIG_ENDIAN)
31 #include <npi_rx_rd32.h>
32 #endif
33 #include <npi_rx_rd64.h>
34 #include <npi_rx_wr64.h>
35 
36 #define	NXGE_ACTUAL_RDCGRP(nxgep, rdcgrp)	\
37 	(rdcgrp + nxgep->pt_config.hw_config.def_mac_rxdma_grpid)
38 #define	NXGE_ACTUAL_RDC(nxgep, rdc)	\
39 	(rdc + nxgep->pt_config.hw_config.start_rdc)
40 
41 /*
42  * XXX: This is a tunable to limit the number of packets each interrupt
43  * handles.  0 (default) means that each interrupt takes as much packets
44  * as it finds.
45  */
46 extern int	nxge_max_intr_pkts;
47 
48 /*
49  * Globals: tunable parameters (/etc/system or adb)
50  *
51  */
52 extern uint32_t nxge_rbr_size;
53 extern uint32_t nxge_rcr_size;
54 extern uint32_t	nxge_rbr_spare_size;
55 
56 extern uint32_t nxge_mblks_pending;
57 
58 /*
59  * Tunable to reduce the amount of time spent in the
60  * ISR doing Rx Processing.
61  */
62 extern uint32_t nxge_max_rx_pkts;
63 boolean_t nxge_jumbo_enable;
64 
65 extern uint16_t nxge_rcr_timeout;
66 extern uint16_t nxge_rcr_threshold;
67 
68 /*
69  * Tunables to manage the receive buffer blocks.
70  *
71  * nxge_rx_threshold_hi: copy all buffers.
72  * nxge_rx_bcopy_size_type: receive buffer block size type.
73  * nxge_rx_threshold_lo: copy only up to tunable block size type.
74  */
75 extern nxge_rxbuf_threshold_t nxge_rx_threshold_hi;
76 extern nxge_rxbuf_type_t nxge_rx_buf_size_type;
77 extern nxge_rxbuf_threshold_t nxge_rx_threshold_lo;
78 
79 extern uint32_t	nxge_cksum_offload;
80 
81 static nxge_status_t nxge_map_rxdma(p_nxge_t, int);
82 static void nxge_unmap_rxdma(p_nxge_t, int);
83 
84 static nxge_status_t nxge_rxdma_hw_start_common(p_nxge_t);
85 
86 static nxge_status_t nxge_rxdma_hw_start(p_nxge_t, int);
87 static void nxge_rxdma_hw_stop(p_nxge_t, int);
88 
89 static nxge_status_t nxge_map_rxdma_channel(p_nxge_t, uint16_t,
90     p_nxge_dma_common_t *,  p_rx_rbr_ring_t *,
91     uint32_t,
92     p_nxge_dma_common_t *, p_rx_rcr_ring_t *,
93     p_rx_mbox_t *);
94 static void nxge_unmap_rxdma_channel(p_nxge_t, uint16_t,
95     p_rx_rbr_ring_t, p_rx_rcr_ring_t, p_rx_mbox_t);
96 
97 static nxge_status_t nxge_map_rxdma_channel_cfg_ring(p_nxge_t,
98     uint16_t,
99     p_nxge_dma_common_t *, p_rx_rbr_ring_t *,
100     p_rx_rcr_ring_t *, p_rx_mbox_t *);
101 static void nxge_unmap_rxdma_channel_cfg_ring(p_nxge_t,
102     p_rx_rcr_ring_t, p_rx_mbox_t);
103 
104 static nxge_status_t nxge_map_rxdma_channel_buf_ring(p_nxge_t,
105     uint16_t,
106     p_nxge_dma_common_t *,
107     p_rx_rbr_ring_t *, uint32_t);
108 static void nxge_unmap_rxdma_channel_buf_ring(p_nxge_t,
109     p_rx_rbr_ring_t);
110 
111 static nxge_status_t nxge_rxdma_start_channel(p_nxge_t, uint16_t,
112     p_rx_rbr_ring_t, p_rx_rcr_ring_t, p_rx_mbox_t);
113 static nxge_status_t nxge_rxdma_stop_channel(p_nxge_t, uint16_t);
114 
115 static mblk_t *
116 nxge_rx_pkts(p_nxge_t, p_rx_rcr_ring_t, rx_dma_ctl_stat_t, int);
117 
118 static void nxge_receive_packet(p_nxge_t,
119 	p_rx_rcr_ring_t,
120 	p_rcr_entry_t,
121 	boolean_t *,
122 	mblk_t **, mblk_t **);
123 
124 nxge_status_t nxge_disable_rxdma_channel(p_nxge_t, uint16_t);
125 
126 static p_rx_msg_t nxge_allocb(size_t, uint32_t, p_nxge_dma_common_t);
127 static void nxge_freeb(p_rx_msg_t);
128 static mblk_t *nxge_rx_pkts_vring(p_nxge_t, uint_t, rx_dma_ctl_stat_t);
129 static nxge_status_t nxge_rx_err_evnts(p_nxge_t, int, rx_dma_ctl_stat_t);
130 
131 static nxge_status_t nxge_rxdma_handle_port_errors(p_nxge_t,
132 				uint32_t, uint32_t);
133 
134 static nxge_status_t nxge_rxbuf_index_info_init(p_nxge_t,
135     p_rx_rbr_ring_t);
136 
137 
138 static nxge_status_t
139 nxge_rxdma_fatal_err_recover(p_nxge_t, uint16_t);
140 
141 nxge_status_t
142 nxge_rx_port_fatal_err_recover(p_nxge_t);
143 
144 static void nxge_rxdma_databuf_free(p_rx_rbr_ring_t);
145 
146 nxge_status_t
147 nxge_init_rxdma_channels(p_nxge_t nxgep)
148 {
149 	nxge_grp_set_t	*set = &nxgep->rx_set;
150 	int		i, count, channel;
151 	nxge_grp_t	*group;
152 	dc_map_t	map;
153 	int		dev_gindex;
154 
155 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_init_rxdma_channels"));
156 
157 	if (!isLDOMguest(nxgep)) {
158 		if (nxge_rxdma_hw_start_common(nxgep) != NXGE_OK) {
159 			cmn_err(CE_NOTE, "hw_start_common");
160 			return (NXGE_ERROR);
161 		}
162 	}
163 
164 	/*
165 	 * NXGE_LOGICAL_GROUP_MAX > NXGE_MAX_RDC_GROUPS (8)
166 	 * We only have 8 hardware RDC tables, but we may have
167 	 * up to 16 logical (software-defined) groups of RDCS,
168 	 * if we make use of layer 3 & 4 hardware classification.
169 	 */
170 	for (i = 0, count = 0; i < NXGE_LOGICAL_GROUP_MAX; i++) {
171 		if ((1 << i) & set->lg.map) {
172 			group = set->group[i];
173 			dev_gindex =
174 			    nxgep->pt_config.hw_config.def_mac_rxdma_grpid + i;
175 			map = nxgep->pt_config.rdc_grps[dev_gindex].map;
176 			for (channel = 0; channel < NXGE_MAX_RDCS; channel++) {
177 				if ((1 << channel) & map) {
178 					if ((nxge_grp_dc_add(nxgep,
179 					    group, VP_BOUND_RX, channel)))
180 						goto init_rxdma_channels_exit;
181 				}
182 			}
183 		}
184 		if (++count == set->lg.count)
185 			break;
186 	}
187 
188 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "<== nxge_init_rxdma_channels"));
189 	return (NXGE_OK);
190 
191 init_rxdma_channels_exit:
192 	for (i = 0, count = 0; i < NXGE_LOGICAL_GROUP_MAX; i++) {
193 		if ((1 << i) & set->lg.map) {
194 			group = set->group[i];
195 			dev_gindex =
196 			    nxgep->pt_config.hw_config.def_mac_rxdma_grpid + i;
197 			map = nxgep->pt_config.rdc_grps[dev_gindex].map;
198 			for (channel = 0; channel < NXGE_MAX_RDCS; channel++) {
199 				if ((1 << channel) & map) {
200 					nxge_grp_dc_remove(nxgep,
201 					    VP_BOUND_RX, channel);
202 				}
203 			}
204 		}
205 		if (++count == set->lg.count)
206 			break;
207 	}
208 
209 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "<== nxge_init_rxdma_channels"));
210 	return (NXGE_ERROR);
211 }
212 
213 nxge_status_t
214 nxge_init_rxdma_channel(p_nxge_t nxge, int channel)
215 {
216 	nxge_status_t	status;
217 
218 	NXGE_DEBUG_MSG((nxge, MEM2_CTL, "==> nxge_init_rxdma_channel"));
219 
220 	status = nxge_map_rxdma(nxge, channel);
221 	if (status != NXGE_OK) {
222 		NXGE_ERROR_MSG((nxge, NXGE_ERR_CTL,
223 		    "<== nxge_init_rxdma: status 0x%x", status));
224 		return (status);
225 	}
226 
227 #if defined(sun4v)
228 	if (isLDOMguest(nxge)) {
229 		/* set rcr_ring */
230 		p_rx_rcr_ring_t ring = nxge->rx_rcr_rings->rcr_rings[channel];
231 
232 		status = nxge_hio_rxdma_bind_intr(nxge, ring, channel);
233 		if (status != NXGE_OK) {
234 			nxge_unmap_rxdma(nxge, channel);
235 			return (status);
236 		}
237 	}
238 #endif
239 
240 	status = nxge_rxdma_hw_start(nxge, channel);
241 	if (status != NXGE_OK) {
242 		nxge_unmap_rxdma(nxge, channel);
243 	}
244 
245 	if (!nxge->statsp->rdc_ksp[channel])
246 		nxge_setup_rdc_kstats(nxge, channel);
247 
248 	NXGE_DEBUG_MSG((nxge, MEM2_CTL,
249 	    "<== nxge_init_rxdma_channel: status 0x%x", status));
250 
251 	return (status);
252 }
253 
254 void
255 nxge_uninit_rxdma_channels(p_nxge_t nxgep)
256 {
257 	nxge_grp_set_t *set = &nxgep->rx_set;
258 	int rdc;
259 
260 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_uninit_rxdma_channels"));
261 
262 	if (set->owned.map == 0) {
263 		NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
264 		    "nxge_uninit_rxdma_channels: no channels"));
265 		return;
266 	}
267 
268 	for (rdc = 0; rdc < NXGE_MAX_RDCS; rdc++) {
269 		if ((1 << rdc) & set->owned.map) {
270 			nxge_grp_dc_remove(nxgep, VP_BOUND_RX, rdc);
271 		}
272 	}
273 
274 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "<== nxge_uninit_rxdma_channels"));
275 }
276 
277 void
278 nxge_uninit_rxdma_channel(p_nxge_t nxgep, int channel)
279 {
280 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_uninit_rxdma_channel"));
281 
282 	if (nxgep->statsp->rdc_ksp[channel]) {
283 		kstat_delete(nxgep->statsp->rdc_ksp[channel]);
284 		nxgep->statsp->rdc_ksp[channel] = 0;
285 	}
286 
287 	nxge_rxdma_hw_stop(nxgep, channel);
288 	nxge_unmap_rxdma(nxgep, channel);
289 
290 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "<== nxge_uinit_rxdma_channel"));
291 }
292 
293 nxge_status_t
294 nxge_reset_rxdma_channel(p_nxge_t nxgep, uint16_t channel)
295 {
296 	npi_handle_t		handle;
297 	npi_status_t		rs = NPI_SUCCESS;
298 	nxge_status_t		status = NXGE_OK;
299 
300 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "==> nxge_reset_rxdma_channel"));
301 
302 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
303 	rs = npi_rxdma_cfg_rdc_reset(handle, channel);
304 
305 	if (rs != NPI_SUCCESS) {
306 		status = NXGE_ERROR | rs;
307 	}
308 
309 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "<== nxge_reset_rxdma_channel"));
310 
311 	return (status);
312 }
313 
314 void
315 nxge_rxdma_regs_dump_channels(p_nxge_t nxgep)
316 {
317 	nxge_grp_set_t *set = &nxgep->rx_set;
318 	int rdc;
319 
320 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "==> nxge_rxdma_regs_dump_channels"));
321 
322 	if (!isLDOMguest(nxgep)) {
323 		npi_handle_t handle = NXGE_DEV_NPI_HANDLE(nxgep);
324 		(void) npi_rxdma_dump_fzc_regs(handle);
325 	}
326 
327 	if (nxgep->rx_rbr_rings == 0 || nxgep->rx_rbr_rings->rbr_rings == 0) {
328 		NXGE_DEBUG_MSG((nxgep, TX_CTL,
329 		    "nxge_rxdma_regs_dump_channels: "
330 		    "NULL ring pointer(s)"));
331 		return;
332 	}
333 
334 	if (set->owned.map == 0) {
335 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
336 		    "nxge_rxdma_regs_dump_channels: no channels"));
337 		return;
338 	}
339 
340 	for (rdc = 0; rdc < NXGE_MAX_RDCS; rdc++) {
341 		if ((1 << rdc) & set->owned.map) {
342 			rx_rbr_ring_t *ring =
343 			    nxgep->rx_rbr_rings->rbr_rings[rdc];
344 			if (ring) {
345 				(void) nxge_dump_rxdma_channel(nxgep, rdc);
346 			}
347 		}
348 	}
349 
350 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "<== nxge_rxdma_regs_dump"));
351 }
352 
353 nxge_status_t
354 nxge_dump_rxdma_channel(p_nxge_t nxgep, uint8_t channel)
355 {
356 	npi_handle_t		handle;
357 	npi_status_t		rs = NPI_SUCCESS;
358 	nxge_status_t		status = NXGE_OK;
359 
360 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "==> nxge_dump_rxdma_channel"));
361 
362 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
363 	rs = npi_rxdma_dump_rdc_regs(handle, channel);
364 
365 	if (rs != NPI_SUCCESS) {
366 		status = NXGE_ERROR | rs;
367 	}
368 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "<== nxge_dump_rxdma_channel"));
369 	return (status);
370 }
371 
372 nxge_status_t
373 nxge_init_rxdma_channel_event_mask(p_nxge_t nxgep, uint16_t channel,
374     p_rx_dma_ent_msk_t mask_p)
375 {
376 	npi_handle_t		handle;
377 	npi_status_t		rs = NPI_SUCCESS;
378 	nxge_status_t		status = NXGE_OK;
379 
380 	NXGE_DEBUG_MSG((nxgep, DMA_CTL,
381 	    "<== nxge_init_rxdma_channel_event_mask"));
382 
383 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
384 	rs = npi_rxdma_event_mask(handle, OP_SET, channel, mask_p);
385 	if (rs != NPI_SUCCESS) {
386 		status = NXGE_ERROR | rs;
387 	}
388 
389 	return (status);
390 }
391 
392 nxge_status_t
393 nxge_init_rxdma_channel_cntl_stat(p_nxge_t nxgep, uint16_t channel,
394     p_rx_dma_ctl_stat_t cs_p)
395 {
396 	npi_handle_t		handle;
397 	npi_status_t		rs = NPI_SUCCESS;
398 	nxge_status_t		status = NXGE_OK;
399 
400 	NXGE_DEBUG_MSG((nxgep, DMA_CTL,
401 	    "<== nxge_init_rxdma_channel_cntl_stat"));
402 
403 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
404 	rs = npi_rxdma_control_status(handle, OP_SET, channel, cs_p);
405 
406 	if (rs != NPI_SUCCESS) {
407 		status = NXGE_ERROR | rs;
408 	}
409 
410 	return (status);
411 }
412 
413 /*
414  * nxge_rxdma_cfg_rdcgrp_default_rdc
415  *
416  *	Set the default RDC for an RDC Group (Table)
417  *
418  * Arguments:
419  * 	nxgep
420  *	rdcgrp	The group to modify
421  *	rdc	The new default RDC.
422  *
423  * Notes:
424  *
425  * NPI/NXGE function calls:
426  *	npi_rxdma_cfg_rdc_table_default_rdc()
427  *
428  * Registers accessed:
429  *	RDC_TBL_REG: FZC_ZCP + 0x10000
430  *
431  * Context:
432  *	Service domain
433  */
434 nxge_status_t
435 nxge_rxdma_cfg_rdcgrp_default_rdc(
436 	p_nxge_t nxgep,
437 	uint8_t rdcgrp,
438 	uint8_t rdc)
439 {
440 	npi_handle_t		handle;
441 	npi_status_t		rs = NPI_SUCCESS;
442 	p_nxge_dma_pt_cfg_t	p_dma_cfgp;
443 	p_nxge_rdc_grp_t	rdc_grp_p;
444 	uint8_t actual_rdcgrp, actual_rdc;
445 
446 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
447 	    " ==> nxge_rxdma_cfg_rdcgrp_default_rdc"));
448 	p_dma_cfgp = (p_nxge_dma_pt_cfg_t)&nxgep->pt_config;
449 
450 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
451 
452 	/*
453 	 * This has to be rewritten.  Do we even allow this anymore?
454 	 */
455 	rdc_grp_p = &p_dma_cfgp->rdc_grps[rdcgrp];
456 	RDC_MAP_IN(rdc_grp_p->map, rdc);
457 	rdc_grp_p->def_rdc = rdc;
458 
459 	actual_rdcgrp = NXGE_ACTUAL_RDCGRP(nxgep, rdcgrp);
460 	actual_rdc = NXGE_ACTUAL_RDC(nxgep, rdc);
461 
462 	rs = npi_rxdma_cfg_rdc_table_default_rdc(
463 	    handle, actual_rdcgrp, actual_rdc);
464 
465 	if (rs != NPI_SUCCESS) {
466 		return (NXGE_ERROR | rs);
467 	}
468 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
469 	    " <== nxge_rxdma_cfg_rdcgrp_default_rdc"));
470 	return (NXGE_OK);
471 }
472 
473 nxge_status_t
474 nxge_rxdma_cfg_port_default_rdc(p_nxge_t nxgep, uint8_t port, uint8_t rdc)
475 {
476 	npi_handle_t		handle;
477 
478 	uint8_t actual_rdc;
479 	npi_status_t		rs = NPI_SUCCESS;
480 
481 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
482 	    " ==> nxge_rxdma_cfg_port_default_rdc"));
483 
484 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
485 	actual_rdc = rdc;	/* XXX Hack! */
486 	rs = npi_rxdma_cfg_default_port_rdc(handle, port, actual_rdc);
487 
488 
489 	if (rs != NPI_SUCCESS) {
490 		return (NXGE_ERROR | rs);
491 	}
492 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
493 	    " <== nxge_rxdma_cfg_port_default_rdc"));
494 
495 	return (NXGE_OK);
496 }
497 
498 nxge_status_t
499 nxge_rxdma_cfg_rcr_threshold(p_nxge_t nxgep, uint8_t channel,
500 				    uint16_t pkts)
501 {
502 	npi_status_t	rs = NPI_SUCCESS;
503 	npi_handle_t	handle;
504 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
505 	    " ==> nxge_rxdma_cfg_rcr_threshold"));
506 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
507 
508 	rs = npi_rxdma_cfg_rdc_rcr_threshold(handle, channel, pkts);
509 
510 	if (rs != NPI_SUCCESS) {
511 		return (NXGE_ERROR | rs);
512 	}
513 	NXGE_DEBUG_MSG((nxgep, RX2_CTL, " <== nxge_rxdma_cfg_rcr_threshold"));
514 	return (NXGE_OK);
515 }
516 
517 nxge_status_t
518 nxge_rxdma_cfg_rcr_timeout(p_nxge_t nxgep, uint8_t channel,
519 			    uint16_t tout, uint8_t enable)
520 {
521 	npi_status_t	rs = NPI_SUCCESS;
522 	npi_handle_t	handle;
523 	NXGE_DEBUG_MSG((nxgep, RX2_CTL, " ==> nxge_rxdma_cfg_rcr_timeout"));
524 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
525 	if (enable == 0) {
526 		rs = npi_rxdma_cfg_rdc_rcr_timeout_disable(handle, channel);
527 	} else {
528 		rs = npi_rxdma_cfg_rdc_rcr_timeout(handle, channel,
529 		    tout);
530 	}
531 
532 	if (rs != NPI_SUCCESS) {
533 		return (NXGE_ERROR | rs);
534 	}
535 	NXGE_DEBUG_MSG((nxgep, RX2_CTL, " <== nxge_rxdma_cfg_rcr_timeout"));
536 	return (NXGE_OK);
537 }
538 
539 nxge_status_t
540 nxge_enable_rxdma_channel(p_nxge_t nxgep, uint16_t channel,
541     p_rx_rbr_ring_t rbr_p, p_rx_rcr_ring_t rcr_p, p_rx_mbox_t mbox_p)
542 {
543 	npi_handle_t		handle;
544 	rdc_desc_cfg_t 		rdc_desc;
545 	p_rcrcfig_b_t		cfgb_p;
546 	npi_status_t		rs = NPI_SUCCESS;
547 
548 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "==> nxge_enable_rxdma_channel"));
549 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
550 	/*
551 	 * Use configuration data composed at init time.
552 	 * Write to hardware the receive ring configurations.
553 	 */
554 	rdc_desc.mbox_enable = 1;
555 	rdc_desc.mbox_addr = mbox_p->mbox_addr;
556 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
557 	    "==> nxge_enable_rxdma_channel: mboxp $%p($%p)",
558 	    mbox_p->mbox_addr, rdc_desc.mbox_addr));
559 
560 	rdc_desc.rbr_len = rbr_p->rbb_max;
561 	rdc_desc.rbr_addr = rbr_p->rbr_addr;
562 
563 	switch (nxgep->rx_bksize_code) {
564 	case RBR_BKSIZE_4K:
565 		rdc_desc.page_size = SIZE_4KB;
566 		break;
567 	case RBR_BKSIZE_8K:
568 		rdc_desc.page_size = SIZE_8KB;
569 		break;
570 	case RBR_BKSIZE_16K:
571 		rdc_desc.page_size = SIZE_16KB;
572 		break;
573 	case RBR_BKSIZE_32K:
574 		rdc_desc.page_size = SIZE_32KB;
575 		break;
576 	}
577 
578 	rdc_desc.size0 = rbr_p->npi_pkt_buf_size0;
579 	rdc_desc.valid0 = 1;
580 
581 	rdc_desc.size1 = rbr_p->npi_pkt_buf_size1;
582 	rdc_desc.valid1 = 1;
583 
584 	rdc_desc.size2 = rbr_p->npi_pkt_buf_size2;
585 	rdc_desc.valid2 = 1;
586 
587 	rdc_desc.full_hdr = rcr_p->full_hdr_flag;
588 	rdc_desc.offset = rcr_p->sw_priv_hdr_len;
589 
590 	rdc_desc.rcr_len = rcr_p->comp_size;
591 	rdc_desc.rcr_addr = rcr_p->rcr_addr;
592 
593 	cfgb_p = &(rcr_p->rcr_cfgb);
594 	rdc_desc.rcr_threshold = cfgb_p->bits.ldw.pthres;
595 	/* For now, disable this timeout in a guest domain. */
596 	if (isLDOMguest(nxgep)) {
597 		rdc_desc.rcr_timeout = 0;
598 		rdc_desc.rcr_timeout_enable = 0;
599 	} else {
600 		rdc_desc.rcr_timeout = cfgb_p->bits.ldw.timeout;
601 		rdc_desc.rcr_timeout_enable = cfgb_p->bits.ldw.entout;
602 	}
603 
604 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "==> nxge_enable_rxdma_channel: "
605 	    "rbr_len qlen %d pagesize code %d rcr_len %d",
606 	    rdc_desc.rbr_len, rdc_desc.page_size, rdc_desc.rcr_len));
607 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "==> nxge_enable_rxdma_channel: "
608 	    "size 0 %d size 1 %d size 2 %d",
609 	    rbr_p->npi_pkt_buf_size0, rbr_p->npi_pkt_buf_size1,
610 	    rbr_p->npi_pkt_buf_size2));
611 
612 	rs = npi_rxdma_cfg_rdc_ring(handle, rbr_p->rdc, &rdc_desc);
613 	if (rs != NPI_SUCCESS) {
614 		return (NXGE_ERROR | rs);
615 	}
616 
617 	/*
618 	 * Enable the timeout and threshold.
619 	 */
620 	rs = npi_rxdma_cfg_rdc_rcr_threshold(handle, channel,
621 	    rdc_desc.rcr_threshold);
622 	if (rs != NPI_SUCCESS) {
623 		return (NXGE_ERROR | rs);
624 	}
625 
626 	rs = npi_rxdma_cfg_rdc_rcr_timeout(handle, channel,
627 	    rdc_desc.rcr_timeout);
628 	if (rs != NPI_SUCCESS) {
629 		return (NXGE_ERROR | rs);
630 	}
631 
632 	/* Enable the DMA */
633 	rs = npi_rxdma_cfg_rdc_enable(handle, channel);
634 	if (rs != NPI_SUCCESS) {
635 		return (NXGE_ERROR | rs);
636 	}
637 
638 	/* Kick the DMA engine. */
639 	npi_rxdma_rdc_rbr_kick(handle, channel, rbr_p->rbb_max);
640 	/* Clear the rbr empty bit */
641 	(void) npi_rxdma_channel_rbr_empty_clear(handle, channel);
642 
643 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "<== nxge_enable_rxdma_channel"));
644 
645 	return (NXGE_OK);
646 }
647 
648 nxge_status_t
649 nxge_disable_rxdma_channel(p_nxge_t nxgep, uint16_t channel)
650 {
651 	npi_handle_t		handle;
652 	npi_status_t		rs = NPI_SUCCESS;
653 
654 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "==> nxge_disable_rxdma_channel"));
655 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
656 
657 	/* disable the DMA */
658 	rs = npi_rxdma_cfg_rdc_disable(handle, channel);
659 	if (rs != NPI_SUCCESS) {
660 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
661 		    "<== nxge_disable_rxdma_channel:failed (0x%x)",
662 		    rs));
663 		return (NXGE_ERROR | rs);
664 	}
665 
666 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "<== nxge_disable_rxdma_channel"));
667 	return (NXGE_OK);
668 }
669 
670 nxge_status_t
671 nxge_rxdma_channel_rcrflush(p_nxge_t nxgep, uint8_t channel)
672 {
673 	npi_handle_t		handle;
674 	nxge_status_t		status = NXGE_OK;
675 
676 	NXGE_DEBUG_MSG((nxgep, DMA_CTL,
677 	    "<== nxge_init_rxdma_channel_rcrflush"));
678 
679 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
680 	npi_rxdma_rdc_rcr_flush(handle, channel);
681 
682 	NXGE_DEBUG_MSG((nxgep, DMA_CTL,
683 	    "<== nxge_init_rxdma_channel_rcrflsh"));
684 	return (status);
685 
686 }
687 
688 #define	MID_INDEX(l, r) ((r + l + 1) >> 1)
689 
690 #define	TO_LEFT -1
691 #define	TO_RIGHT 1
692 #define	BOTH_RIGHT (TO_RIGHT + TO_RIGHT)
693 #define	BOTH_LEFT (TO_LEFT + TO_LEFT)
694 #define	IN_MIDDLE (TO_RIGHT + TO_LEFT)
695 #define	NO_HINT 0xffffffff
696 
697 /*ARGSUSED*/
698 nxge_status_t
699 nxge_rxbuf_pp_to_vp(p_nxge_t nxgep, p_rx_rbr_ring_t rbr_p,
700 	uint8_t pktbufsz_type, uint64_t *pkt_buf_addr_pp,
701 	uint64_t **pkt_buf_addr_p, uint32_t *bufoffset, uint32_t *msg_index)
702 {
703 	int			bufsize;
704 	uint64_t		pktbuf_pp;
705 	uint64_t 		dvma_addr;
706 	rxring_info_t 		*ring_info;
707 	int 			base_side, end_side;
708 	int 			r_index, l_index, anchor_index;
709 	int 			found, search_done;
710 	uint32_t offset, chunk_size, block_size, page_size_mask;
711 	uint32_t chunk_index, block_index, total_index;
712 	int 			max_iterations, iteration;
713 	rxbuf_index_info_t 	*bufinfo;
714 
715 	NXGE_DEBUG_MSG((nxgep, RX2_CTL, "==> nxge_rxbuf_pp_to_vp"));
716 
717 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
718 	    "==> nxge_rxbuf_pp_to_vp: buf_pp $%p btype %d",
719 	    pkt_buf_addr_pp,
720 	    pktbufsz_type));
721 #if defined(__i386)
722 	pktbuf_pp = (uint64_t)(uint32_t)pkt_buf_addr_pp;
723 #else
724 	pktbuf_pp = (uint64_t)pkt_buf_addr_pp;
725 #endif
726 
727 	switch (pktbufsz_type) {
728 	case 0:
729 		bufsize = rbr_p->pkt_buf_size0;
730 		break;
731 	case 1:
732 		bufsize = rbr_p->pkt_buf_size1;
733 		break;
734 	case 2:
735 		bufsize = rbr_p->pkt_buf_size2;
736 		break;
737 	case RCR_SINGLE_BLOCK:
738 		bufsize = 0;
739 		anchor_index = 0;
740 		break;
741 	default:
742 		return (NXGE_ERROR);
743 	}
744 
745 	if (rbr_p->num_blocks == 1) {
746 		anchor_index = 0;
747 		ring_info = rbr_p->ring_info;
748 		bufinfo = (rxbuf_index_info_t *)ring_info->buffer;
749 		NXGE_DEBUG_MSG((nxgep, RX2_CTL,
750 		    "==> nxge_rxbuf_pp_to_vp: (found, 1 block) "
751 		    "buf_pp $%p btype %d anchor_index %d "
752 		    "bufinfo $%p",
753 		    pkt_buf_addr_pp,
754 		    pktbufsz_type,
755 		    anchor_index,
756 		    bufinfo));
757 
758 		goto found_index;
759 	}
760 
761 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
762 	    "==> nxge_rxbuf_pp_to_vp: "
763 	    "buf_pp $%p btype %d  anchor_index %d",
764 	    pkt_buf_addr_pp,
765 	    pktbufsz_type,
766 	    anchor_index));
767 
768 	ring_info = rbr_p->ring_info;
769 	found = B_FALSE;
770 	bufinfo = (rxbuf_index_info_t *)ring_info->buffer;
771 	iteration = 0;
772 	max_iterations = ring_info->max_iterations;
773 		/*
774 		 * First check if this block has been seen
775 		 * recently. This is indicated by a hint which
776 		 * is initialized when the first buffer of the block
777 		 * is seen. The hint is reset when the last buffer of
778 		 * the block has been processed.
779 		 * As three block sizes are supported, three hints
780 		 * are kept. The idea behind the hints is that once
781 		 * the hardware  uses a block for a buffer  of that
782 		 * size, it will use it exclusively for that size
783 		 * and will use it until it is exhausted. It is assumed
784 		 * that there would a single block being used for the same
785 		 * buffer sizes at any given time.
786 		 */
787 	if (ring_info->hint[pktbufsz_type] != NO_HINT) {
788 		anchor_index = ring_info->hint[pktbufsz_type];
789 		dvma_addr =  bufinfo[anchor_index].dvma_addr;
790 		chunk_size = bufinfo[anchor_index].buf_size;
791 		if ((pktbuf_pp >= dvma_addr) &&
792 		    (pktbuf_pp < (dvma_addr + chunk_size))) {
793 			found = B_TRUE;
794 				/*
795 				 * check if this is the last buffer in the block
796 				 * If so, then reset the hint for the size;
797 				 */
798 
799 			if ((pktbuf_pp + bufsize) >= (dvma_addr + chunk_size))
800 				ring_info->hint[pktbufsz_type] = NO_HINT;
801 		}
802 	}
803 
804 	if (found == B_FALSE) {
805 		NXGE_DEBUG_MSG((nxgep, RX2_CTL,
806 		    "==> nxge_rxbuf_pp_to_vp: (!found)"
807 		    "buf_pp $%p btype %d anchor_index %d",
808 		    pkt_buf_addr_pp,
809 		    pktbufsz_type,
810 		    anchor_index));
811 
812 			/*
813 			 * This is the first buffer of the block of this
814 			 * size. Need to search the whole information
815 			 * array.
816 			 * the search algorithm uses a binary tree search
817 			 * algorithm. It assumes that the information is
818 			 * already sorted with increasing order
819 			 * info[0] < info[1] < info[2]  .... < info[n-1]
820 			 * where n is the size of the information array
821 			 */
822 		r_index = rbr_p->num_blocks - 1;
823 		l_index = 0;
824 		search_done = B_FALSE;
825 		anchor_index = MID_INDEX(r_index, l_index);
826 		while (search_done == B_FALSE) {
827 			if ((r_index == l_index) ||
828 			    (iteration >= max_iterations))
829 				search_done = B_TRUE;
830 			end_side = TO_RIGHT; /* to the right */
831 			base_side = TO_LEFT; /* to the left */
832 			/* read the DVMA address information and sort it */
833 			dvma_addr =  bufinfo[anchor_index].dvma_addr;
834 			chunk_size = bufinfo[anchor_index].buf_size;
835 			NXGE_DEBUG_MSG((nxgep, RX2_CTL,
836 			    "==> nxge_rxbuf_pp_to_vp: (searching)"
837 			    "buf_pp $%p btype %d "
838 			    "anchor_index %d chunk_size %d dvmaaddr $%p",
839 			    pkt_buf_addr_pp,
840 			    pktbufsz_type,
841 			    anchor_index,
842 			    chunk_size,
843 			    dvma_addr));
844 
845 			if (pktbuf_pp >= dvma_addr)
846 				base_side = TO_RIGHT; /* to the right */
847 			if (pktbuf_pp < (dvma_addr + chunk_size))
848 				end_side = TO_LEFT; /* to the left */
849 
850 			switch (base_side + end_side) {
851 			case IN_MIDDLE:
852 				/* found */
853 				found = B_TRUE;
854 				search_done = B_TRUE;
855 				if ((pktbuf_pp + bufsize) <
856 				    (dvma_addr + chunk_size))
857 					ring_info->hint[pktbufsz_type] =
858 					    bufinfo[anchor_index].buf_index;
859 				break;
860 			case BOTH_RIGHT:
861 				/* not found: go to the right */
862 				l_index = anchor_index + 1;
863 				anchor_index = MID_INDEX(r_index, l_index);
864 				break;
865 
866 			case BOTH_LEFT:
867 				/* not found: go to the left */
868 				r_index = anchor_index - 1;
869 				anchor_index = MID_INDEX(r_index, l_index);
870 				break;
871 			default: /* should not come here */
872 				return (NXGE_ERROR);
873 			}
874 			iteration++;
875 		}
876 
877 		NXGE_DEBUG_MSG((nxgep, RX2_CTL,
878 		    "==> nxge_rxbuf_pp_to_vp: (search done)"
879 		    "buf_pp $%p btype %d anchor_index %d",
880 		    pkt_buf_addr_pp,
881 		    pktbufsz_type,
882 		    anchor_index));
883 	}
884 
885 	if (found == B_FALSE) {
886 		NXGE_DEBUG_MSG((nxgep, RX2_CTL,
887 		    "==> nxge_rxbuf_pp_to_vp: (search failed)"
888 		    "buf_pp $%p btype %d anchor_index %d",
889 		    pkt_buf_addr_pp,
890 		    pktbufsz_type,
891 		    anchor_index));
892 		return (NXGE_ERROR);
893 	}
894 
895 found_index:
896 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
897 	    "==> nxge_rxbuf_pp_to_vp: (FOUND1)"
898 	    "buf_pp $%p btype %d bufsize %d anchor_index %d",
899 	    pkt_buf_addr_pp,
900 	    pktbufsz_type,
901 	    bufsize,
902 	    anchor_index));
903 
904 	/* index of the first block in this chunk */
905 	chunk_index = bufinfo[anchor_index].start_index;
906 	dvma_addr =  bufinfo[anchor_index].dvma_addr;
907 	page_size_mask = ring_info->block_size_mask;
908 
909 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
910 	    "==> nxge_rxbuf_pp_to_vp: (FOUND3), get chunk)"
911 	    "buf_pp $%p btype %d bufsize %d "
912 	    "anchor_index %d chunk_index %d dvma $%p",
913 	    pkt_buf_addr_pp,
914 	    pktbufsz_type,
915 	    bufsize,
916 	    anchor_index,
917 	    chunk_index,
918 	    dvma_addr));
919 
920 	offset = pktbuf_pp - dvma_addr; /* offset within the chunk */
921 	block_size = rbr_p->block_size; /* System  block(page) size */
922 
923 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
924 	    "==> nxge_rxbuf_pp_to_vp: (FOUND4), get chunk)"
925 	    "buf_pp $%p btype %d bufsize %d "
926 	    "anchor_index %d chunk_index %d dvma $%p "
927 	    "offset %d block_size %d",
928 	    pkt_buf_addr_pp,
929 	    pktbufsz_type,
930 	    bufsize,
931 	    anchor_index,
932 	    chunk_index,
933 	    dvma_addr,
934 	    offset,
935 	    block_size));
936 
937 	NXGE_DEBUG_MSG((nxgep, RX2_CTL, "==> getting total index"));
938 
939 	block_index = (offset / block_size); /* index within chunk */
940 	total_index = chunk_index + block_index;
941 
942 
943 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
944 	    "==> nxge_rxbuf_pp_to_vp: "
945 	    "total_index %d dvma_addr $%p "
946 	    "offset %d block_size %d "
947 	    "block_index %d ",
948 	    total_index, dvma_addr,
949 	    offset, block_size,
950 	    block_index));
951 #if defined(__i386)
952 	*pkt_buf_addr_p = (uint64_t *)((uint32_t)bufinfo[anchor_index].kaddr +
953 	    (uint32_t)offset);
954 #else
955 	*pkt_buf_addr_p = (uint64_t *)((uint64_t)bufinfo[anchor_index].kaddr +
956 	    (uint64_t)offset);
957 #endif
958 
959 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
960 	    "==> nxge_rxbuf_pp_to_vp: "
961 	    "total_index %d dvma_addr $%p "
962 	    "offset %d block_size %d "
963 	    "block_index %d "
964 	    "*pkt_buf_addr_p $%p",
965 	    total_index, dvma_addr,
966 	    offset, block_size,
967 	    block_index,
968 	    *pkt_buf_addr_p));
969 
970 
971 	*msg_index = total_index;
972 	*bufoffset =  (offset & page_size_mask);
973 
974 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
975 	    "==> nxge_rxbuf_pp_to_vp: get msg index: "
976 	    "msg_index %d bufoffset_index %d",
977 	    *msg_index,
978 	    *bufoffset));
979 
980 	NXGE_DEBUG_MSG((nxgep, RX2_CTL, "<== nxge_rxbuf_pp_to_vp"));
981 
982 	return (NXGE_OK);
983 }
984 
985 /*
986  * used by quick sort (qsort) function
987  * to perform comparison
988  */
989 static int
990 nxge_sort_compare(const void *p1, const void *p2)
991 {
992 
993 	rxbuf_index_info_t *a, *b;
994 
995 	a = (rxbuf_index_info_t *)p1;
996 	b = (rxbuf_index_info_t *)p2;
997 
998 	if (a->dvma_addr > b->dvma_addr)
999 		return (1);
1000 	if (a->dvma_addr < b->dvma_addr)
1001 		return (-1);
1002 	return (0);
1003 }
1004 
1005 
1006 
1007 /*
1008  * grabbed this sort implementation from common/syscall/avl.c
1009  *
1010  */
1011 /*
1012  * Generic shellsort, from K&R (1st ed, p 58.), somewhat modified.
1013  * v = Ptr to array/vector of objs
1014  * n = # objs in the array
1015  * s = size of each obj (must be multiples of a word size)
1016  * f = ptr to function to compare two objs
1017  *	returns (-1 = less than, 0 = equal, 1 = greater than
1018  */
1019 void
1020 nxge_ksort(caddr_t v, int n, int s, int (*f)())
1021 {
1022 	int g, i, j, ii;
1023 	unsigned int *p1, *p2;
1024 	unsigned int tmp;
1025 
1026 	/* No work to do */
1027 	if (v == NULL || n <= 1)
1028 		return;
1029 	/* Sanity check on arguments */
1030 	ASSERT(((uintptr_t)v & 0x3) == 0 && (s & 0x3) == 0);
1031 	ASSERT(s > 0);
1032 
1033 	for (g = n / 2; g > 0; g /= 2) {
1034 		for (i = g; i < n; i++) {
1035 			for (j = i - g; j >= 0 &&
1036 			    (*f)(v + j * s, v + (j + g) * s) == 1;
1037 			    j -= g) {
1038 				p1 = (unsigned *)(v + j * s);
1039 				p2 = (unsigned *)(v + (j + g) * s);
1040 				for (ii = 0; ii < s / 4; ii++) {
1041 					tmp = *p1;
1042 					*p1++ = *p2;
1043 					*p2++ = tmp;
1044 				}
1045 			}
1046 		}
1047 	}
1048 }
1049 
1050 /*
1051  * Initialize data structures required for rxdma
1052  * buffer dvma->vmem address lookup
1053  */
1054 /*ARGSUSED*/
1055 static nxge_status_t
1056 nxge_rxbuf_index_info_init(p_nxge_t nxgep, p_rx_rbr_ring_t rbrp)
1057 {
1058 
1059 	int index;
1060 	rxring_info_t *ring_info;
1061 	int max_iteration = 0, max_index = 0;
1062 
1063 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "==> nxge_rxbuf_index_info_init"));
1064 
1065 	ring_info = rbrp->ring_info;
1066 	ring_info->hint[0] = NO_HINT;
1067 	ring_info->hint[1] = NO_HINT;
1068 	ring_info->hint[2] = NO_HINT;
1069 	max_index = rbrp->num_blocks;
1070 
1071 		/* read the DVMA address information and sort it */
1072 		/* do init of the information array */
1073 
1074 
1075 	NXGE_DEBUG_MSG((nxgep, DMA2_CTL,
1076 	    " nxge_rxbuf_index_info_init Sort ptrs"));
1077 
1078 		/* sort the array */
1079 	nxge_ksort((void *)ring_info->buffer, max_index,
1080 	    sizeof (rxbuf_index_info_t), nxge_sort_compare);
1081 
1082 
1083 
1084 	for (index = 0; index < max_index; index++) {
1085 		NXGE_DEBUG_MSG((nxgep, DMA2_CTL,
1086 		    " nxge_rxbuf_index_info_init: sorted chunk %d "
1087 		    " ioaddr $%p kaddr $%p size %x",
1088 		    index, ring_info->buffer[index].dvma_addr,
1089 		    ring_info->buffer[index].kaddr,
1090 		    ring_info->buffer[index].buf_size));
1091 	}
1092 
1093 	max_iteration = 0;
1094 	while (max_index >= (1ULL << max_iteration))
1095 		max_iteration++;
1096 	ring_info->max_iterations = max_iteration + 1;
1097 	NXGE_DEBUG_MSG((nxgep, DMA2_CTL,
1098 	    " nxge_rxbuf_index_info_init Find max iter %d",
1099 	    ring_info->max_iterations));
1100 
1101 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "<== nxge_rxbuf_index_info_init"));
1102 	return (NXGE_OK);
1103 }
1104 
1105 /* ARGSUSED */
1106 void
1107 nxge_dump_rcr_entry(p_nxge_t nxgep, p_rcr_entry_t entry_p)
1108 {
1109 #ifdef	NXGE_DEBUG
1110 
1111 	uint32_t bptr;
1112 	uint64_t pp;
1113 
1114 	bptr = entry_p->bits.hdw.pkt_buf_addr;
1115 
1116 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
1117 	    "\trcr entry $%p "
1118 	    "\trcr entry 0x%0llx "
1119 	    "\trcr entry 0x%08x "
1120 	    "\trcr entry 0x%08x "
1121 	    "\tvalue 0x%0llx\n"
1122 	    "\tmulti = %d\n"
1123 	    "\tpkt_type = 0x%x\n"
1124 	    "\tzero_copy = %d\n"
1125 	    "\tnoport = %d\n"
1126 	    "\tpromis = %d\n"
1127 	    "\terror = 0x%04x\n"
1128 	    "\tdcf_err = 0x%01x\n"
1129 	    "\tl2_len = %d\n"
1130 	    "\tpktbufsize = %d\n"
1131 	    "\tpkt_buf_addr = $%p\n"
1132 	    "\tpkt_buf_addr (<< 6) = $%p\n",
1133 	    entry_p,
1134 	    *(int64_t *)entry_p,
1135 	    *(int32_t *)entry_p,
1136 	    *(int32_t *)((char *)entry_p + 32),
1137 	    entry_p->value,
1138 	    entry_p->bits.hdw.multi,
1139 	    entry_p->bits.hdw.pkt_type,
1140 	    entry_p->bits.hdw.zero_copy,
1141 	    entry_p->bits.hdw.noport,
1142 	    entry_p->bits.hdw.promis,
1143 	    entry_p->bits.hdw.error,
1144 	    entry_p->bits.hdw.dcf_err,
1145 	    entry_p->bits.hdw.l2_len,
1146 	    entry_p->bits.hdw.pktbufsz,
1147 	    bptr,
1148 	    entry_p->bits.ldw.pkt_buf_addr));
1149 
1150 	pp = (entry_p->value & RCR_PKT_BUF_ADDR_MASK) <<
1151 	    RCR_PKT_BUF_ADDR_SHIFT;
1152 
1153 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "rcr pp 0x%llx l2 len %d",
1154 	    pp, (*(int64_t *)entry_p >> 40) & 0x3fff));
1155 #endif
1156 }
1157 
1158 void
1159 nxge_rxdma_regs_dump(p_nxge_t nxgep, int rdc)
1160 {
1161 	npi_handle_t		handle;
1162 	rbr_stat_t 		rbr_stat;
1163 	addr44_t 		hd_addr;
1164 	addr44_t 		tail_addr;
1165 	uint16_t 		qlen;
1166 
1167 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
1168 	    "==> nxge_rxdma_regs_dump: rdc channel %d", rdc));
1169 
1170 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
1171 
1172 	/* RBR head */
1173 	hd_addr.addr = 0;
1174 	(void) npi_rxdma_rdc_rbr_head_get(handle, rdc, &hd_addr);
1175 #if defined(__i386)
1176 	printf("nxge_rxdma_regs_dump: got hdptr $%p \n",
1177 	    (void *)(uint32_t)hd_addr.addr);
1178 #else
1179 	printf("nxge_rxdma_regs_dump: got hdptr $%p \n",
1180 	    (void *)hd_addr.addr);
1181 #endif
1182 
1183 	/* RBR stats */
1184 	(void) npi_rxdma_rdc_rbr_stat_get(handle, rdc, &rbr_stat);
1185 	printf("nxge_rxdma_regs_dump: rbr len %d \n", rbr_stat.bits.ldw.qlen);
1186 
1187 	/* RCR tail */
1188 	tail_addr.addr = 0;
1189 	(void) npi_rxdma_rdc_rcr_tail_get(handle, rdc, &tail_addr);
1190 #if defined(__i386)
1191 	printf("nxge_rxdma_regs_dump: got tail ptr $%p \n",
1192 	    (void *)(uint32_t)tail_addr.addr);
1193 #else
1194 	printf("nxge_rxdma_regs_dump: got tail ptr $%p \n",
1195 	    (void *)tail_addr.addr);
1196 #endif
1197 
1198 	/* RCR qlen */
1199 	(void) npi_rxdma_rdc_rcr_qlen_get(handle, rdc, &qlen);
1200 	printf("nxge_rxdma_regs_dump: rcr len %x \n", qlen);
1201 
1202 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
1203 	    "<== nxge_rxdma_regs_dump: rdc rdc %d", rdc));
1204 }
1205 
1206 nxge_status_t
1207 nxge_rxdma_hw_mode(p_nxge_t nxgep, boolean_t enable)
1208 {
1209 	nxge_grp_set_t *set = &nxgep->rx_set;
1210 	nxge_status_t status;
1211 	npi_status_t rs;
1212 	int rdc;
1213 
1214 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
1215 	    "==> nxge_rxdma_hw_mode: mode %d", enable));
1216 
1217 	if (!(nxgep->drv_state & STATE_HW_INITIALIZED)) {
1218 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
1219 		    "<== nxge_rxdma_mode: not initialized"));
1220 		return (NXGE_ERROR);
1221 	}
1222 
1223 	if (nxgep->rx_rbr_rings == 0 || nxgep->rx_rbr_rings->rbr_rings == 0) {
1224 		NXGE_DEBUG_MSG((nxgep, TX_CTL,
1225 		    "<== nxge_tx_port_fatal_err_recover: "
1226 		    "NULL ring pointer(s)"));
1227 		return (NXGE_ERROR);
1228 	}
1229 
1230 	if (set->owned.map == 0) {
1231 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
1232 		    "nxge_rxdma_regs_dump_channels: no channels"));
1233 		return (NULL);
1234 	}
1235 
1236 	for (rdc = 0; rdc < NXGE_MAX_RDCS; rdc++) {
1237 		if ((1 << rdc) & set->owned.map) {
1238 			rx_rbr_ring_t *ring =
1239 			    nxgep->rx_rbr_rings->rbr_rings[rdc];
1240 			npi_handle_t handle = NXGE_DEV_NPI_HANDLE(nxgep);
1241 			if (ring) {
1242 				if (enable) {
1243 					NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
1244 					    "==> nxge_rxdma_hw_mode: "
1245 					    "channel %d (enable)", rdc));
1246 					rs = npi_rxdma_cfg_rdc_enable
1247 					    (handle, rdc);
1248 				} else {
1249 					NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
1250 					    "==> nxge_rxdma_hw_mode: "
1251 					    "channel %d disable)", rdc));
1252 					rs = npi_rxdma_cfg_rdc_disable
1253 					    (handle, rdc);
1254 				}
1255 			}
1256 		}
1257 	}
1258 
1259 	status = ((rs == NPI_SUCCESS) ? NXGE_OK : NXGE_ERROR | rs);
1260 
1261 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
1262 	    "<== nxge_rxdma_hw_mode: status 0x%x", status));
1263 
1264 	return (status);
1265 }
1266 
1267 void
1268 nxge_rxdma_enable_channel(p_nxge_t nxgep, uint16_t channel)
1269 {
1270 	npi_handle_t		handle;
1271 
1272 	NXGE_DEBUG_MSG((nxgep, DMA_CTL,
1273 	    "==> nxge_rxdma_enable_channel: channel %d", channel));
1274 
1275 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
1276 	(void) npi_rxdma_cfg_rdc_enable(handle, channel);
1277 
1278 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "<== nxge_rxdma_enable_channel"));
1279 }
1280 
1281 void
1282 nxge_rxdma_disable_channel(p_nxge_t nxgep, uint16_t channel)
1283 {
1284 	npi_handle_t		handle;
1285 
1286 	NXGE_DEBUG_MSG((nxgep, DMA_CTL,
1287 	    "==> nxge_rxdma_disable_channel: channel %d", channel));
1288 
1289 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
1290 	(void) npi_rxdma_cfg_rdc_disable(handle, channel);
1291 
1292 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "<== nxge_rxdma_disable_channel"));
1293 }
1294 
1295 void
1296 nxge_hw_start_rx(p_nxge_t nxgep)
1297 {
1298 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "==> nxge_hw_start_rx"));
1299 
1300 	(void) nxge_rxdma_hw_mode(nxgep, NXGE_DMA_START);
1301 	(void) nxge_rx_mac_enable(nxgep);
1302 
1303 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "<== nxge_hw_start_rx"));
1304 }
1305 
1306 /*ARGSUSED*/
1307 void
1308 nxge_fixup_rxdma_rings(p_nxge_t nxgep)
1309 {
1310 	nxge_grp_set_t *set = &nxgep->rx_set;
1311 	int rdc;
1312 
1313 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "==> nxge_fixup_rxdma_rings"));
1314 
1315 	if (nxgep->rx_rbr_rings == 0 || nxgep->rx_rbr_rings->rbr_rings == 0) {
1316 		NXGE_DEBUG_MSG((nxgep, TX_CTL,
1317 		    "<== nxge_tx_port_fatal_err_recover: "
1318 		    "NULL ring pointer(s)"));
1319 		return;
1320 	}
1321 
1322 	if (set->owned.map == 0) {
1323 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
1324 		    "nxge_rxdma_regs_dump_channels: no channels"));
1325 		return;
1326 	}
1327 
1328 	for (rdc = 0; rdc < NXGE_MAX_RDCS; rdc++) {
1329 		if ((1 << rdc) & set->owned.map) {
1330 			rx_rbr_ring_t *ring =
1331 			    nxgep->rx_rbr_rings->rbr_rings[rdc];
1332 			if (ring) {
1333 				nxge_rxdma_hw_stop(nxgep, rdc);
1334 				NXGE_DEBUG_MSG((nxgep, RX_CTL,
1335 				    "==> nxge_fixup_rxdma_rings: "
1336 				    "channel %d ring $%px",
1337 				    rdc, ring));
1338 				(void) nxge_rxdma_fixup_channel
1339 				    (nxgep, rdc, rdc);
1340 			}
1341 		}
1342 	}
1343 
1344 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "<== nxge_fixup_rxdma_rings"));
1345 }
1346 
1347 void
1348 nxge_rxdma_fix_channel(p_nxge_t nxgep, uint16_t channel)
1349 {
1350 	int		i;
1351 
1352 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "==> nxge_rxdma_fix_channel"));
1353 	i = nxge_rxdma_get_ring_index(nxgep, channel);
1354 	if (i < 0) {
1355 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
1356 		    "<== nxge_rxdma_fix_channel: no entry found"));
1357 		return;
1358 	}
1359 
1360 	nxge_rxdma_fixup_channel(nxgep, channel, i);
1361 
1362 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "<== nxge_rxdma_fix_channel"));
1363 }
1364 
1365 void
1366 nxge_rxdma_fixup_channel(p_nxge_t nxgep, uint16_t channel, int entry)
1367 {
1368 	int			ndmas;
1369 	p_rx_rbr_rings_t 	rx_rbr_rings;
1370 	p_rx_rbr_ring_t		*rbr_rings;
1371 	p_rx_rcr_rings_t 	rx_rcr_rings;
1372 	p_rx_rcr_ring_t		*rcr_rings;
1373 	p_rx_mbox_areas_t 	rx_mbox_areas_p;
1374 	p_rx_mbox_t		*rx_mbox_p;
1375 	p_nxge_dma_pool_t	dma_buf_poolp;
1376 	p_nxge_dma_pool_t	dma_cntl_poolp;
1377 	p_rx_rbr_ring_t 	rbrp;
1378 	p_rx_rcr_ring_t 	rcrp;
1379 	p_rx_mbox_t 		mboxp;
1380 	p_nxge_dma_common_t 	dmap;
1381 	nxge_status_t		status = NXGE_OK;
1382 
1383 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "==> nxge_rxdma_fixup_channel"));
1384 
1385 	(void) nxge_rxdma_stop_channel(nxgep, channel);
1386 
1387 	dma_buf_poolp = nxgep->rx_buf_pool_p;
1388 	dma_cntl_poolp = nxgep->rx_cntl_pool_p;
1389 
1390 	if (!dma_buf_poolp->buf_allocated || !dma_cntl_poolp->buf_allocated) {
1391 		NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
1392 		    "<== nxge_rxdma_fixup_channel: buf not allocated"));
1393 		return;
1394 	}
1395 
1396 	ndmas = dma_buf_poolp->ndmas;
1397 	if (!ndmas) {
1398 		NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
1399 		    "<== nxge_rxdma_fixup_channel: no dma allocated"));
1400 		return;
1401 	}
1402 
1403 	rx_rbr_rings = nxgep->rx_rbr_rings;
1404 	rx_rcr_rings = nxgep->rx_rcr_rings;
1405 	rbr_rings = rx_rbr_rings->rbr_rings;
1406 	rcr_rings = rx_rcr_rings->rcr_rings;
1407 	rx_mbox_areas_p = nxgep->rx_mbox_areas_p;
1408 	rx_mbox_p = rx_mbox_areas_p->rxmbox_areas;
1409 
1410 	/* Reinitialize the receive block and completion rings */
1411 	rbrp = (p_rx_rbr_ring_t)rbr_rings[entry],
1412 	    rcrp = (p_rx_rcr_ring_t)rcr_rings[entry],
1413 	    mboxp = (p_rx_mbox_t)rx_mbox_p[entry];
1414 
1415 
1416 	rbrp->rbr_wr_index = (rbrp->rbb_max - 1);
1417 	rbrp->rbr_rd_index = 0;
1418 	rcrp->comp_rd_index = 0;
1419 	rcrp->comp_wt_index = 0;
1420 
1421 	dmap = (p_nxge_dma_common_t)&rcrp->rcr_desc;
1422 	bzero((caddr_t)dmap->kaddrp, dmap->alength);
1423 
1424 	status = nxge_rxdma_start_channel(nxgep, channel,
1425 	    rbrp, rcrp, mboxp);
1426 	if (status != NXGE_OK) {
1427 		goto nxge_rxdma_fixup_channel_fail;
1428 	}
1429 	if (status != NXGE_OK) {
1430 		goto nxge_rxdma_fixup_channel_fail;
1431 	}
1432 
1433 nxge_rxdma_fixup_channel_fail:
1434 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
1435 	    "==> nxge_rxdma_fixup_channel: failed (0x%08x)", status));
1436 
1437 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "<== nxge_rxdma_fixup_channel"));
1438 }
1439 
1440 /*
1441  * Convert an absolute RDC number to a Receive Buffer Ring index.  That is,
1442  * map <channel> to an index into nxgep->rx_rbr_rings.
1443  * (device ring index -> port ring index)
1444  */
1445 int
1446 nxge_rxdma_get_ring_index(p_nxge_t nxgep, uint16_t channel)
1447 {
1448 	int			i, ndmas;
1449 	uint16_t		rdc;
1450 	p_rx_rbr_rings_t	rx_rbr_rings;
1451 	p_rx_rbr_ring_t		*rbr_rings;
1452 
1453 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
1454 	    "==> nxge_rxdma_get_ring_index: channel %d", channel));
1455 
1456 	rx_rbr_rings = nxgep->rx_rbr_rings;
1457 	if (rx_rbr_rings == NULL) {
1458 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
1459 		    "<== nxge_rxdma_get_ring_index: NULL ring pointer"));
1460 		return (-1);
1461 	}
1462 	ndmas = rx_rbr_rings->ndmas;
1463 	if (!ndmas) {
1464 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
1465 		    "<== nxge_rxdma_get_ring_index: no channel"));
1466 		return (-1);
1467 	}
1468 
1469 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
1470 	    "==> nxge_rxdma_get_ring_index (ndmas %d)", ndmas));
1471 
1472 	rbr_rings = rx_rbr_rings->rbr_rings;
1473 	for (i = 0; i < ndmas; i++) {
1474 		rdc = rbr_rings[i]->rdc;
1475 		if (channel == rdc) {
1476 			NXGE_DEBUG_MSG((nxgep, RX_CTL,
1477 			    "==> nxge_rxdma_get_rbr_ring: channel %d "
1478 			    "(index %d) ring %d", channel, i, rbr_rings[i]));
1479 			return (i);
1480 		}
1481 	}
1482 
1483 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
1484 	    "<== nxge_rxdma_get_rbr_ring_index: not found"));
1485 
1486 	return (-1);
1487 }
1488 
1489 p_rx_rbr_ring_t
1490 nxge_rxdma_get_rbr_ring(p_nxge_t nxgep, uint16_t channel)
1491 {
1492 	nxge_grp_set_t *set = &nxgep->rx_set;
1493 	nxge_channel_t rdc;
1494 
1495 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
1496 	    "==> nxge_rxdma_get_rbr_ring: channel %d", channel));
1497 
1498 	if (nxgep->rx_rbr_rings == 0 || nxgep->rx_rbr_rings->rbr_rings == 0) {
1499 		NXGE_DEBUG_MSG((nxgep, TX_CTL,
1500 		    "<== nxge_rxdma_get_rbr_ring: "
1501 		    "NULL ring pointer(s)"));
1502 		return (NULL);
1503 	}
1504 
1505 	if (set->owned.map == 0) {
1506 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
1507 		    "<== nxge_rxdma_get_rbr_ring: no channels"));
1508 		return (NULL);
1509 	}
1510 
1511 	for (rdc = 0; rdc < NXGE_MAX_RDCS; rdc++) {
1512 		if ((1 << rdc) & set->owned.map) {
1513 			rx_rbr_ring_t *ring =
1514 			    nxgep->rx_rbr_rings->rbr_rings[rdc];
1515 			if (ring) {
1516 				if (channel == ring->rdc) {
1517 					NXGE_DEBUG_MSG((nxgep, RX_CTL,
1518 					    "==> nxge_rxdma_get_rbr_ring: "
1519 					    "channel %d ring $%p", rdc, ring));
1520 					return (ring);
1521 				}
1522 			}
1523 		}
1524 	}
1525 
1526 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
1527 	    "<== nxge_rxdma_get_rbr_ring: not found"));
1528 
1529 	return (NULL);
1530 }
1531 
1532 p_rx_rcr_ring_t
1533 nxge_rxdma_get_rcr_ring(p_nxge_t nxgep, uint16_t channel)
1534 {
1535 	nxge_grp_set_t *set = &nxgep->rx_set;
1536 	nxge_channel_t rdc;
1537 
1538 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
1539 	    "==> nxge_rxdma_get_rcr_ring: channel %d", channel));
1540 
1541 	if (nxgep->rx_rcr_rings == 0 || nxgep->rx_rcr_rings->rcr_rings == 0) {
1542 		NXGE_DEBUG_MSG((nxgep, TX_CTL,
1543 		    "<== nxge_rxdma_get_rcr_ring: "
1544 		    "NULL ring pointer(s)"));
1545 		return (NULL);
1546 	}
1547 
1548 	if (set->owned.map == 0) {
1549 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
1550 		    "<== nxge_rxdma_get_rbr_ring: no channels"));
1551 		return (NULL);
1552 	}
1553 
1554 	for (rdc = 0; rdc < NXGE_MAX_RDCS; rdc++) {
1555 		if ((1 << rdc) & set->owned.map) {
1556 			rx_rcr_ring_t *ring =
1557 			    nxgep->rx_rcr_rings->rcr_rings[rdc];
1558 			if (ring) {
1559 				if (channel == ring->rdc) {
1560 					NXGE_DEBUG_MSG((nxgep, RX_CTL,
1561 					    "==> nxge_rxdma_get_rcr_ring: "
1562 					    "channel %d ring $%p", rdc, ring));
1563 					return (ring);
1564 				}
1565 			}
1566 		}
1567 	}
1568 
1569 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
1570 	    "<== nxge_rxdma_get_rcr_ring: not found"));
1571 
1572 	return (NULL);
1573 }
1574 
1575 /*
1576  * Static functions start here.
1577  */
1578 static p_rx_msg_t
1579 nxge_allocb(size_t size, uint32_t pri, p_nxge_dma_common_t dmabuf_p)
1580 {
1581 	p_rx_msg_t nxge_mp 		= NULL;
1582 	p_nxge_dma_common_t		dmamsg_p;
1583 	uchar_t 			*buffer;
1584 
1585 	nxge_mp = KMEM_ZALLOC(sizeof (rx_msg_t), KM_NOSLEEP);
1586 	if (nxge_mp == NULL) {
1587 		NXGE_ERROR_MSG((NULL, NXGE_ERR_CTL,
1588 		    "Allocation of a rx msg failed."));
1589 		goto nxge_allocb_exit;
1590 	}
1591 
1592 	nxge_mp->use_buf_pool = B_FALSE;
1593 	if (dmabuf_p) {
1594 		nxge_mp->use_buf_pool = B_TRUE;
1595 		dmamsg_p = (p_nxge_dma_common_t)&nxge_mp->buf_dma;
1596 		*dmamsg_p = *dmabuf_p;
1597 		dmamsg_p->nblocks = 1;
1598 		dmamsg_p->block_size = size;
1599 		dmamsg_p->alength = size;
1600 		buffer = (uchar_t *)dmabuf_p->kaddrp;
1601 
1602 		dmabuf_p->kaddrp = (void *)
1603 		    ((char *)dmabuf_p->kaddrp + size);
1604 		dmabuf_p->ioaddr_pp = (void *)
1605 		    ((char *)dmabuf_p->ioaddr_pp + size);
1606 		dmabuf_p->alength -= size;
1607 		dmabuf_p->offset += size;
1608 		dmabuf_p->dma_cookie.dmac_laddress += size;
1609 		dmabuf_p->dma_cookie.dmac_size -= size;
1610 
1611 	} else {
1612 		buffer = KMEM_ALLOC(size, KM_NOSLEEP);
1613 		if (buffer == NULL) {
1614 			NXGE_ERROR_MSG((NULL, NXGE_ERR_CTL,
1615 			    "Allocation of a receive page failed."));
1616 			goto nxge_allocb_fail1;
1617 		}
1618 	}
1619 
1620 	nxge_mp->rx_mblk_p = desballoc(buffer, size, pri, &nxge_mp->freeb);
1621 	if (nxge_mp->rx_mblk_p == NULL) {
1622 		NXGE_ERROR_MSG((NULL, NXGE_ERR_CTL, "desballoc failed."));
1623 		goto nxge_allocb_fail2;
1624 	}
1625 
1626 	nxge_mp->buffer = buffer;
1627 	nxge_mp->block_size = size;
1628 	nxge_mp->freeb.free_func = (void (*)())nxge_freeb;
1629 	nxge_mp->freeb.free_arg = (caddr_t)nxge_mp;
1630 	nxge_mp->ref_cnt = 1;
1631 	nxge_mp->free = B_TRUE;
1632 	nxge_mp->rx_use_bcopy = B_FALSE;
1633 
1634 	atomic_inc_32(&nxge_mblks_pending);
1635 
1636 	goto nxge_allocb_exit;
1637 
1638 nxge_allocb_fail2:
1639 	if (!nxge_mp->use_buf_pool) {
1640 		KMEM_FREE(buffer, size);
1641 	}
1642 
1643 nxge_allocb_fail1:
1644 	KMEM_FREE(nxge_mp, sizeof (rx_msg_t));
1645 	nxge_mp = NULL;
1646 
1647 nxge_allocb_exit:
1648 	return (nxge_mp);
1649 }
1650 
1651 p_mblk_t
1652 nxge_dupb(p_rx_msg_t nxge_mp, uint_t offset, size_t size)
1653 {
1654 	p_mblk_t mp;
1655 
1656 	NXGE_DEBUG_MSG((NULL, MEM_CTL, "==> nxge_dupb"));
1657 	NXGE_DEBUG_MSG((NULL, MEM_CTL, "nxge_mp = $%p "
1658 	    "offset = 0x%08X "
1659 	    "size = 0x%08X",
1660 	    nxge_mp, offset, size));
1661 
1662 	mp = desballoc(&nxge_mp->buffer[offset], size,
1663 	    0, &nxge_mp->freeb);
1664 	if (mp == NULL) {
1665 		NXGE_DEBUG_MSG((NULL, RX_CTL, "desballoc failed"));
1666 		goto nxge_dupb_exit;
1667 	}
1668 	atomic_inc_32(&nxge_mp->ref_cnt);
1669 
1670 
1671 nxge_dupb_exit:
1672 	NXGE_DEBUG_MSG((NULL, MEM_CTL, "<== nxge_dupb mp = $%p",
1673 	    nxge_mp));
1674 	return (mp);
1675 }
1676 
1677 p_mblk_t
1678 nxge_dupb_bcopy(p_rx_msg_t nxge_mp, uint_t offset, size_t size)
1679 {
1680 	p_mblk_t mp;
1681 	uchar_t *dp;
1682 
1683 	mp = allocb(size + NXGE_RXBUF_EXTRA, 0);
1684 	if (mp == NULL) {
1685 		NXGE_DEBUG_MSG((NULL, RX_CTL, "desballoc failed"));
1686 		goto nxge_dupb_bcopy_exit;
1687 	}
1688 	dp = mp->b_rptr = mp->b_rptr + NXGE_RXBUF_EXTRA;
1689 	bcopy((void *)&nxge_mp->buffer[offset], dp, size);
1690 	mp->b_wptr = dp + size;
1691 
1692 nxge_dupb_bcopy_exit:
1693 	NXGE_DEBUG_MSG((NULL, MEM_CTL, "<== nxge_dupb mp = $%p",
1694 	    nxge_mp));
1695 	return (mp);
1696 }
1697 
1698 void nxge_post_page(p_nxge_t nxgep, p_rx_rbr_ring_t rx_rbr_p,
1699 	p_rx_msg_t rx_msg_p);
1700 
1701 void
1702 nxge_post_page(p_nxge_t nxgep, p_rx_rbr_ring_t rx_rbr_p, p_rx_msg_t rx_msg_p)
1703 {
1704 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "==> nxge_post_page"));
1705 
1706 	/* Reuse this buffer */
1707 	rx_msg_p->free = B_FALSE;
1708 	rx_msg_p->cur_usage_cnt = 0;
1709 	rx_msg_p->max_usage_cnt = 0;
1710 	rx_msg_p->pkt_buf_size = 0;
1711 
1712 	if (rx_rbr_p->rbr_use_bcopy) {
1713 		rx_msg_p->rx_use_bcopy = B_FALSE;
1714 		atomic_dec_32(&rx_rbr_p->rbr_consumed);
1715 	}
1716 
1717 	/*
1718 	 * Get the rbr header pointer and its offset index.
1719 	 */
1720 	MUTEX_ENTER(&rx_rbr_p->post_lock);
1721 	rx_rbr_p->rbr_wr_index =  ((rx_rbr_p->rbr_wr_index + 1) &
1722 	    rx_rbr_p->rbr_wrap_mask);
1723 	rx_rbr_p->rbr_desc_vp[rx_rbr_p->rbr_wr_index] = rx_msg_p->shifted_addr;
1724 	MUTEX_EXIT(&rx_rbr_p->post_lock);
1725 	npi_rxdma_rdc_rbr_kick(NXGE_DEV_NPI_HANDLE(nxgep),
1726 	    rx_rbr_p->rdc, 1);
1727 
1728 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
1729 	    "<== nxge_post_page (channel %d post_next_index %d)",
1730 	    rx_rbr_p->rdc, rx_rbr_p->rbr_wr_index));
1731 
1732 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "<== nxge_post_page"));
1733 }
1734 
1735 void
1736 nxge_freeb(p_rx_msg_t rx_msg_p)
1737 {
1738 	size_t size;
1739 	uchar_t *buffer = NULL;
1740 	int ref_cnt;
1741 	boolean_t free_state = B_FALSE;
1742 
1743 	rx_rbr_ring_t *ring = rx_msg_p->rx_rbr_p;
1744 
1745 	NXGE_DEBUG_MSG((NULL, MEM2_CTL, "==> nxge_freeb"));
1746 	NXGE_DEBUG_MSG((NULL, MEM2_CTL,
1747 	    "nxge_freeb:rx_msg_p = $%p (block pending %d)",
1748 	    rx_msg_p, nxge_mblks_pending));
1749 
1750 	/*
1751 	 * First we need to get the free state, then
1752 	 * atomic decrement the reference count to prevent
1753 	 * the race condition with the interrupt thread that
1754 	 * is processing a loaned up buffer block.
1755 	 */
1756 	free_state = rx_msg_p->free;
1757 	ref_cnt = atomic_add_32_nv(&rx_msg_p->ref_cnt, -1);
1758 	if (!ref_cnt) {
1759 		atomic_dec_32(&nxge_mblks_pending);
1760 		buffer = rx_msg_p->buffer;
1761 		size = rx_msg_p->block_size;
1762 		NXGE_DEBUG_MSG((NULL, MEM2_CTL, "nxge_freeb: "
1763 		    "will free: rx_msg_p = $%p (block pending %d)",
1764 		    rx_msg_p, nxge_mblks_pending));
1765 
1766 		if (!rx_msg_p->use_buf_pool) {
1767 			KMEM_FREE(buffer, size);
1768 		}
1769 
1770 		KMEM_FREE(rx_msg_p, sizeof (rx_msg_t));
1771 
1772 		if (ring) {
1773 			/*
1774 			 * Decrement the receive buffer ring's reference
1775 			 * count, too.
1776 			 */
1777 			atomic_dec_32(&ring->rbr_ref_cnt);
1778 
1779 			/*
1780 			 * Free the receive buffer ring, if
1781 			 * 1. all the receive buffers have been freed
1782 			 * 2. and we are in the proper state (that is,
1783 			 *    we are not UNMAPPING).
1784 			 */
1785 			if (ring->rbr_ref_cnt == 0 &&
1786 			    ring->rbr_state == RBR_UNMAPPED) {
1787 				/*
1788 				 * Free receive data buffers,
1789 				 * buffer index information
1790 				 * (rxring_info) and
1791 				 * the message block ring.
1792 				 */
1793 				NXGE_DEBUG_MSG((NULL, RX_CTL,
1794 				    "nxge_freeb:rx_msg_p = $%p "
1795 				    "(block pending %d) free buffers",
1796 				    rx_msg_p, nxge_mblks_pending));
1797 				nxge_rxdma_databuf_free(ring);
1798 				if (ring->ring_info) {
1799 					KMEM_FREE(ring->ring_info,
1800 					    sizeof (rxring_info_t));
1801 				}
1802 
1803 				if (ring->rx_msg_ring) {
1804 					KMEM_FREE(ring->rx_msg_ring,
1805 					    ring->tnblocks *
1806 					    sizeof (p_rx_msg_t));
1807 				}
1808 				KMEM_FREE(ring, sizeof (*ring));
1809 			}
1810 		}
1811 		return;
1812 	}
1813 
1814 	/*
1815 	 * Repost buffer.
1816 	 */
1817 	if (free_state && (ref_cnt == 1) && ring) {
1818 		NXGE_DEBUG_MSG((NULL, RX_CTL,
1819 		    "nxge_freeb: post page $%p:", rx_msg_p));
1820 		if (ring->rbr_state == RBR_POSTING)
1821 			nxge_post_page(rx_msg_p->nxgep, ring, rx_msg_p);
1822 	}
1823 
1824 	NXGE_DEBUG_MSG((NULL, MEM2_CTL, "<== nxge_freeb"));
1825 }
1826 
1827 uint_t
1828 nxge_rx_intr(void *arg1, void *arg2)
1829 {
1830 	p_nxge_ldv_t		ldvp = (p_nxge_ldv_t)arg1;
1831 	p_nxge_t		nxgep = (p_nxge_t)arg2;
1832 	p_nxge_ldg_t		ldgp;
1833 	uint8_t			channel;
1834 	npi_handle_t		handle;
1835 	rx_dma_ctl_stat_t	cs;
1836 	p_rx_rcr_ring_t		rcr_ring;
1837 	mblk_t			*mp;
1838 
1839 #ifdef	NXGE_DEBUG
1840 	rxdma_cfig1_t		cfg;
1841 #endif
1842 
1843 	if (ldvp == NULL) {
1844 		NXGE_DEBUG_MSG((NULL, INT_CTL,
1845 		    "<== nxge_rx_intr: arg2 $%p arg1 $%p",
1846 		    nxgep, ldvp));
1847 
1848 		return (DDI_INTR_CLAIMED);
1849 	}
1850 
1851 	if (arg2 == NULL || (void *)ldvp->nxgep != arg2) {
1852 		nxgep = ldvp->nxgep;
1853 	}
1854 
1855 	if ((!(nxgep->drv_state & STATE_HW_INITIALIZED)) ||
1856 	    (nxgep->nxge_mac_state != NXGE_MAC_STARTED)) {
1857 		NXGE_DEBUG_MSG((nxgep, INT_CTL,
1858 		    "<== nxge_rx_intr: interface not started or intialized"));
1859 		return (DDI_INTR_CLAIMED);
1860 	}
1861 
1862 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
1863 	    "==> nxge_rx_intr: arg2 $%p arg1 $%p",
1864 	    nxgep, ldvp));
1865 
1866 	/*
1867 	 * This interrupt handler is for a specific
1868 	 * receive dma channel.
1869 	 */
1870 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
1871 
1872 	rcr_ring = nxgep->rx_rcr_rings->rcr_rings[ldvp->vdma_index];
1873 
1874 	/*
1875 	 * The RCR ring lock must be held when packets
1876 	 * are being processed and the hardware registers are
1877 	 * being read or written to prevent race condition
1878 	 * among the interrupt thread, the polling thread
1879 	 * (will cause fatal errors such as rcrincon bit set)
1880 	 * and the setting of the poll_flag.
1881 	 */
1882 	MUTEX_ENTER(&rcr_ring->lock);
1883 
1884 	/*
1885 	 * Get the control and status for this channel.
1886 	 */
1887 	channel = ldvp->channel;
1888 	ldgp = ldvp->ldgp;
1889 
1890 	if (!isLDOMguest(nxgep)) {
1891 		if (!nxgep->rx_channel_started[channel]) {
1892 			NXGE_DEBUG_MSG((nxgep, INT_CTL,
1893 			    "<== nxge_rx_intr: channel is not started"));
1894 			MUTEX_EXIT(&rcr_ring->lock);
1895 			return (DDI_INTR_CLAIMED);
1896 		}
1897 	}
1898 
1899 	ASSERT(rcr_ring->ldgp == ldgp);
1900 	ASSERT(rcr_ring->ldvp == ldvp);
1901 
1902 	RXDMA_REG_READ64(handle, RX_DMA_CTL_STAT_REG, channel, &cs.value);
1903 
1904 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "==> nxge_rx_intr:channel %d "
1905 	    "cs 0x%016llx rcrto 0x%x rcrthres %x",
1906 	    channel,
1907 	    cs.value,
1908 	    cs.bits.hdw.rcrto,
1909 	    cs.bits.hdw.rcrthres));
1910 
1911 	mp = nxge_rx_pkts_vring(nxgep, ldvp->vdma_index, cs);
1912 
1913 	/* error events. */
1914 	if (cs.value & RX_DMA_CTL_STAT_ERROR) {
1915 		(void) nxge_rx_err_evnts(nxgep, channel, cs);
1916 	}
1917 
1918 	/*
1919 	 * Enable the mailbox update interrupt if we want
1920 	 * to use mailbox. We probably don't need to use
1921 	 * mailbox as it only saves us one pio read.
1922 	 * Also write 1 to rcrthres and rcrto to clear
1923 	 * these two edge triggered bits.
1924 	 */
1925 	cs.value &= RX_DMA_CTL_STAT_WR1C;
1926 	cs.bits.hdw.mex = rcr_ring->poll_flag ? 0 : 1;
1927 	RXDMA_REG_WRITE64(handle, RX_DMA_CTL_STAT_REG, channel,
1928 	    cs.value);
1929 
1930 	/*
1931 	 * If the polling mode is enabled, disable the interrupt.
1932 	 */
1933 	if (rcr_ring->poll_flag) {
1934 		NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
1935 		    "==> nxge_rx_intr: rdc %d ldgp $%p ldvp $%p "
1936 		    "(disabling interrupts)", channel, ldgp, ldvp));
1937 		/*
1938 		 * Disarm this logical group if this is a single device
1939 		 * group.
1940 		 */
1941 		if (ldgp->nldvs == 1) {
1942 			ldgimgm_t mgm;
1943 			mgm.value = 0;
1944 			mgm.bits.ldw.arm = 0;
1945 			NXGE_REG_WR64(handle,
1946 			    LDGIMGN_REG + LDSV_OFFSET(ldgp->ldg), mgm.value);
1947 		}
1948 	} else {
1949 		/*
1950 		 * Rearm this logical group if this is a single device
1951 		 * group.
1952 		 */
1953 		if (ldgp->nldvs == 1) {
1954 			if (isLDOMguest(nxgep)) {
1955 				nxge_hio_ldgimgn(nxgep, ldgp);
1956 			} else {
1957 				ldgimgm_t mgm;
1958 
1959 				mgm.value = 0;
1960 				mgm.bits.ldw.arm = 1;
1961 				mgm.bits.ldw.timer = ldgp->ldg_timer;
1962 
1963 				NXGE_REG_WR64(handle,
1964 				    LDGIMGN_REG + LDSV_OFFSET(ldgp->ldg),
1965 				    mgm.value);
1966 			}
1967 		}
1968 
1969 		NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
1970 		    "==> nxge_rx_intr: rdc %d ldgp $%p "
1971 		    "exiting ISR (and call mac_rx_ring)", channel, ldgp));
1972 	}
1973 	MUTEX_EXIT(&rcr_ring->lock);
1974 
1975 	if (mp) {
1976 		if (!isLDOMguest(nxgep))
1977 			mac_rx_ring(nxgep->mach, rcr_ring->rcr_mac_handle, mp,
1978 			    rcr_ring->rcr_gen_num);
1979 #if defined(sun4v)
1980 		else {			/* isLDOMguest(nxgep) */
1981 			nxge_hio_data_t *nhd = (nxge_hio_data_t *)
1982 			    nxgep->nxge_hw_p->hio;
1983 			nx_vio_fp_t *vio = &nhd->hio.vio;
1984 
1985 			if (vio->cb.vio_net_rx_cb) {
1986 				(*vio->cb.vio_net_rx_cb)
1987 				    (nxgep->hio_vr->vhp, mp);
1988 			}
1989 		}
1990 #endif
1991 	}
1992 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "<== nxge_rx_intr: DDI_INTR_CLAIMED"));
1993 	return (DDI_INTR_CLAIMED);
1994 }
1995 
1996 /*
1997  * Process the packets received in the specified logical device
1998  * and pass up a chain of message blocks to the upper layer.
1999  * The RCR ring lock must be held before calling this function.
2000  */
2001 static mblk_t *
2002 nxge_rx_pkts_vring(p_nxge_t nxgep, uint_t vindex, rx_dma_ctl_stat_t cs)
2003 {
2004 	p_mblk_t		mp;
2005 	p_rx_rcr_ring_t		rcrp;
2006 
2007 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "==> nxge_rx_pkts_vring"));
2008 	rcrp = nxgep->rx_rcr_rings->rcr_rings[vindex];
2009 
2010 	NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
2011 	    "==> nxge_rx_pkts_vring: (calling nxge_rx_pkts)rdc %d "
2012 	    "rcr_mac_handle $%p ", rcrp->rdc, rcrp->rcr_mac_handle));
2013 	if ((mp = nxge_rx_pkts(nxgep, rcrp, cs, -1)) == NULL) {
2014 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
2015 		    "<== nxge_rx_pkts_vring: no mp"));
2016 		return (NULL);
2017 	}
2018 
2019 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "==> nxge_rx_pkts_vring: $%p",
2020 	    mp));
2021 
2022 #ifdef  NXGE_DEBUG
2023 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
2024 		    "==> nxge_rx_pkts_vring:calling mac_rx "
2025 		    "LEN %d mp $%p mp->b_cont $%p mp->b_next $%p rcrp $%p "
2026 		    "mac_handle $%p",
2027 		    mp->b_wptr - mp->b_rptr,
2028 		    mp, mp->b_cont, mp->b_next,
2029 		    rcrp, rcrp->rcr_mac_handle));
2030 
2031 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
2032 		    "==> nxge_rx_pkts_vring: dump packets "
2033 		    "(mp $%p b_rptr $%p b_wptr $%p):\n %s",
2034 		    mp,
2035 		    mp->b_rptr,
2036 		    mp->b_wptr,
2037 		    nxge_dump_packet((char *)mp->b_rptr,
2038 		    mp->b_wptr - mp->b_rptr)));
2039 		if (mp->b_cont) {
2040 			NXGE_DEBUG_MSG((nxgep, RX_CTL,
2041 			    "==> nxge_rx_pkts_vring: dump b_cont packets "
2042 			    "(mp->b_cont $%p b_rptr $%p b_wptr $%p):\n %s",
2043 			    mp->b_cont,
2044 			    mp->b_cont->b_rptr,
2045 			    mp->b_cont->b_wptr,
2046 			    nxge_dump_packet((char *)mp->b_cont->b_rptr,
2047 			    mp->b_cont->b_wptr - mp->b_cont->b_rptr)));
2048 		}
2049 		if (mp->b_next) {
2050 			NXGE_DEBUG_MSG((nxgep, RX_CTL,
2051 			    "==> nxge_rx_pkts_vring: dump next packets "
2052 			    "(b_rptr $%p): %s",
2053 			    mp->b_next->b_rptr,
2054 			    nxge_dump_packet((char *)mp->b_next->b_rptr,
2055 			    mp->b_next->b_wptr - mp->b_next->b_rptr)));
2056 		}
2057 #endif
2058 	NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
2059 	    "<== nxge_rx_pkts_vring: returning rdc %d rcr_mac_handle $%p ",
2060 	    rcrp->rdc, rcrp->rcr_mac_handle));
2061 
2062 	return (mp);
2063 }
2064 
2065 
2066 /*
2067  * This routine is the main packet receive processing function.
2068  * It gets the packet type, error code, and buffer related
2069  * information from the receive completion entry.
2070  * How many completion entries to process is based on the number of packets
2071  * queued by the hardware, a hardware maintained tail pointer
2072  * and a configurable receive packet count.
2073  *
2074  * A chain of message blocks will be created as result of processing
2075  * the completion entries. This chain of message blocks will be returned and
2076  * a hardware control status register will be updated with the number of
2077  * packets were removed from the hardware queue.
2078  *
2079  * The RCR ring lock is held when entering this function.
2080  */
2081 static mblk_t *
2082 nxge_rx_pkts(p_nxge_t nxgep, p_rx_rcr_ring_t rcr_p, rx_dma_ctl_stat_t cs,
2083     int bytes_to_pickup)
2084 {
2085 	npi_handle_t		handle;
2086 	uint8_t			channel;
2087 	uint32_t		comp_rd_index;
2088 	p_rcr_entry_t		rcr_desc_rd_head_p;
2089 	p_rcr_entry_t		rcr_desc_rd_head_pp;
2090 	p_mblk_t		nmp, mp_cont, head_mp, *tail_mp;
2091 	uint16_t		qlen, nrcr_read, npkt_read;
2092 	uint32_t		qlen_hw;
2093 	boolean_t		multi;
2094 	rcrcfig_b_t		rcr_cfg_b;
2095 	int			totallen = 0;
2096 #if defined(_BIG_ENDIAN)
2097 	npi_status_t		rs = NPI_SUCCESS;
2098 #endif
2099 
2100 	NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL, "==> nxge_rx_pkts: "
2101 	    "channel %d", rcr_p->rdc));
2102 
2103 	if (!(nxgep->drv_state & STATE_HW_INITIALIZED)) {
2104 		return (NULL);
2105 	}
2106 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
2107 	channel = rcr_p->rdc;
2108 
2109 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
2110 	    "==> nxge_rx_pkts: START: rcr channel %d "
2111 	    "head_p $%p head_pp $%p  index %d ",
2112 	    channel, rcr_p->rcr_desc_rd_head_p,
2113 	    rcr_p->rcr_desc_rd_head_pp,
2114 	    rcr_p->comp_rd_index));
2115 
2116 
2117 #if !defined(_BIG_ENDIAN)
2118 	qlen = RXDMA_REG_READ32(handle, RCRSTAT_A_REG, channel) & 0xffff;
2119 #else
2120 	rs = npi_rxdma_rdc_rcr_qlen_get(handle, channel, &qlen);
2121 	if (rs != NPI_SUCCESS) {
2122 		NXGE_DEBUG_MSG((nxgep, RX_CTL, "==> nxge_rx_pkts: "
2123 		"channel %d, get qlen failed 0x%08x",
2124 		    channel, rs));
2125 		return (NULL);
2126 	}
2127 #endif
2128 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "==> nxge_rx_pkts:rcr channel %d "
2129 	    "qlen %d", channel, qlen));
2130 
2131 
2132 
2133 	if (!qlen) {
2134 		NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
2135 		    "==> nxge_rx_pkts:rcr channel %d "
2136 		    "qlen %d (no pkts)", channel, qlen));
2137 
2138 		return (NULL);
2139 	}
2140 
2141 	comp_rd_index = rcr_p->comp_rd_index;
2142 
2143 	rcr_desc_rd_head_p = rcr_p->rcr_desc_rd_head_p;
2144 	rcr_desc_rd_head_pp = rcr_p->rcr_desc_rd_head_pp;
2145 	nrcr_read = npkt_read = 0;
2146 
2147 	/*
2148 	 * Number of packets queued
2149 	 * (The jumbo or multi packet will be counted as only one
2150 	 *  packets and it may take up more than one completion entry).
2151 	 */
2152 	qlen_hw = (qlen < nxge_max_rx_pkts) ?
2153 	    qlen : nxge_max_rx_pkts;
2154 	head_mp = NULL;
2155 	tail_mp = &head_mp;
2156 	nmp = mp_cont = NULL;
2157 	multi = B_FALSE;
2158 
2159 	while (qlen_hw) {
2160 
2161 #ifdef NXGE_DEBUG
2162 		nxge_dump_rcr_entry(nxgep, rcr_desc_rd_head_p);
2163 #endif
2164 		/*
2165 		 * Process one completion ring entry.
2166 		 */
2167 		nxge_receive_packet(nxgep,
2168 		    rcr_p, rcr_desc_rd_head_p, &multi, &nmp, &mp_cont);
2169 
2170 		/*
2171 		 * message chaining modes
2172 		 */
2173 		if (nmp) {
2174 			nmp->b_next = NULL;
2175 			if (!multi && !mp_cont) { /* frame fits a partition */
2176 				*tail_mp = nmp;
2177 				tail_mp = &nmp->b_next;
2178 				totallen += MBLKL(nmp);
2179 				nmp = NULL;
2180 			} else if (multi && !mp_cont) { /* first segment */
2181 				*tail_mp = nmp;
2182 				tail_mp = &nmp->b_cont;
2183 				totallen += MBLKL(nmp);
2184 			} else if (multi && mp_cont) {	/* mid of multi segs */
2185 				*tail_mp = mp_cont;
2186 				tail_mp = &mp_cont->b_cont;
2187 				totallen += MBLKL(mp_cont);
2188 			} else if (!multi && mp_cont) { /* last segment */
2189 				*tail_mp = mp_cont;
2190 				tail_mp = &nmp->b_next;
2191 				totallen += MBLKL(mp_cont);
2192 				nmp = NULL;
2193 			}
2194 		}
2195 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
2196 		    "==> nxge_rx_pkts: loop: rcr channel %d "
2197 		    "before updating: multi %d "
2198 		    "nrcr_read %d "
2199 		    "npk read %d "
2200 		    "head_pp $%p  index %d ",
2201 		    channel,
2202 		    multi,
2203 		    nrcr_read, npkt_read, rcr_desc_rd_head_pp,
2204 		    comp_rd_index));
2205 
2206 		if (!multi) {
2207 			qlen_hw--;
2208 			npkt_read++;
2209 		}
2210 
2211 		/*
2212 		 * Update the next read entry.
2213 		 */
2214 		comp_rd_index = NEXT_ENTRY(comp_rd_index,
2215 		    rcr_p->comp_wrap_mask);
2216 
2217 		rcr_desc_rd_head_p = NEXT_ENTRY_PTR(rcr_desc_rd_head_p,
2218 		    rcr_p->rcr_desc_first_p,
2219 		    rcr_p->rcr_desc_last_p);
2220 
2221 		nrcr_read++;
2222 
2223 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
2224 		    "<== nxge_rx_pkts: (SAM, process one packet) "
2225 		    "nrcr_read %d",
2226 		    nrcr_read));
2227 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
2228 		    "==> nxge_rx_pkts: loop: rcr channel %d "
2229 		    "multi %d "
2230 		    "nrcr_read %d "
2231 		    "npk read %d "
2232 		    "head_pp $%p  index %d ",
2233 		    channel,
2234 		    multi,
2235 		    nrcr_read, npkt_read, rcr_desc_rd_head_pp,
2236 		    comp_rd_index));
2237 
2238 		if ((bytes_to_pickup != -1) &&
2239 		    (totallen >= bytes_to_pickup)) {
2240 			break;
2241 		}
2242 
2243 		/* limit the number of packets for interrupt */
2244 		if (!(rcr_p->poll_flag)) {
2245 			if (npkt_read == nxge_max_intr_pkts) {
2246 				break;
2247 			}
2248 		}
2249 	}
2250 
2251 	rcr_p->rcr_desc_rd_head_pp = rcr_desc_rd_head_pp;
2252 	rcr_p->comp_rd_index = comp_rd_index;
2253 	rcr_p->rcr_desc_rd_head_p = rcr_desc_rd_head_p;
2254 	if ((nxgep->intr_timeout != rcr_p->intr_timeout) ||
2255 	    (nxgep->intr_threshold != rcr_p->intr_threshold)) {
2256 
2257 		rcr_p->intr_timeout = (nxgep->intr_timeout <
2258 		    NXGE_RDC_RCR_TIMEOUT_MIN) ? NXGE_RDC_RCR_TIMEOUT_MIN :
2259 		    nxgep->intr_timeout;
2260 
2261 		rcr_p->intr_threshold = (nxgep->intr_threshold <
2262 		    NXGE_RDC_RCR_THRESHOLD_MIN) ? NXGE_RDC_RCR_THRESHOLD_MIN :
2263 		    nxgep->intr_threshold;
2264 
2265 		rcr_cfg_b.value = 0x0ULL;
2266 		rcr_cfg_b.bits.ldw.entout = 1;
2267 		rcr_cfg_b.bits.ldw.timeout = rcr_p->intr_timeout;
2268 		rcr_cfg_b.bits.ldw.pthres = rcr_p->intr_threshold;
2269 
2270 		RXDMA_REG_WRITE64(handle, RCRCFIG_B_REG,
2271 		    channel, rcr_cfg_b.value);
2272 	}
2273 
2274 	cs.bits.ldw.pktread = npkt_read;
2275 	cs.bits.ldw.ptrread = nrcr_read;
2276 	RXDMA_REG_WRITE64(handle, RX_DMA_CTL_STAT_REG,
2277 	    channel, cs.value);
2278 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
2279 	    "==> nxge_rx_pkts: EXIT: rcr channel %d "
2280 	    "head_pp $%p  index %016llx ",
2281 	    channel,
2282 	    rcr_p->rcr_desc_rd_head_pp,
2283 	    rcr_p->comp_rd_index));
2284 	/*
2285 	 * Update RCR buffer pointer read and number of packets
2286 	 * read.
2287 	 */
2288 
2289 	NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL, "<== nxge_rx_pkts: return"
2290 	    "channel %d", rcr_p->rdc));
2291 
2292 	return (head_mp);
2293 }
2294 
2295 void
2296 nxge_receive_packet(p_nxge_t nxgep,
2297     p_rx_rcr_ring_t rcr_p, p_rcr_entry_t rcr_desc_rd_head_p,
2298     boolean_t *multi_p, mblk_t **mp, mblk_t **mp_cont)
2299 {
2300 	p_mblk_t		nmp = NULL;
2301 	uint64_t		multi;
2302 	uint64_t		dcf_err;
2303 	uint8_t			channel;
2304 
2305 	boolean_t		first_entry = B_TRUE;
2306 	boolean_t		is_tcp_udp = B_FALSE;
2307 	boolean_t		buffer_free = B_FALSE;
2308 	boolean_t		error_send_up = B_FALSE;
2309 	uint8_t			error_type;
2310 	uint16_t		l2_len;
2311 	uint16_t		skip_len;
2312 	uint8_t			pktbufsz_type;
2313 	uint64_t		rcr_entry;
2314 	uint64_t		*pkt_buf_addr_pp;
2315 	uint64_t		*pkt_buf_addr_p;
2316 	uint32_t		buf_offset;
2317 	uint32_t		bsize;
2318 	uint32_t		error_disp_cnt;
2319 	uint32_t		msg_index;
2320 	p_rx_rbr_ring_t		rx_rbr_p;
2321 	p_rx_msg_t 		*rx_msg_ring_p;
2322 	p_rx_msg_t		rx_msg_p;
2323 	uint16_t		sw_offset_bytes = 0, hdr_size = 0;
2324 	nxge_status_t		status = NXGE_OK;
2325 	boolean_t		is_valid = B_FALSE;
2326 	p_nxge_rx_ring_stats_t	rdc_stats;
2327 	uint32_t		bytes_read;
2328 	uint64_t		pkt_type;
2329 	uint64_t		frag;
2330 	boolean_t		pkt_too_long_err = B_FALSE;
2331 #ifdef	NXGE_DEBUG
2332 	int			dump_len;
2333 #endif
2334 	NXGE_DEBUG_MSG((nxgep, RX2_CTL, "==> nxge_receive_packet"));
2335 	first_entry = (*mp == NULL) ? B_TRUE : B_FALSE;
2336 
2337 	rcr_entry = *((uint64_t *)rcr_desc_rd_head_p);
2338 
2339 	multi = (rcr_entry & RCR_MULTI_MASK);
2340 	dcf_err = (rcr_entry & RCR_DCF_ERROR_MASK);
2341 	pkt_type = (rcr_entry & RCR_PKT_TYPE_MASK);
2342 
2343 	error_type = ((rcr_entry & RCR_ERROR_MASK) >> RCR_ERROR_SHIFT);
2344 	frag = (rcr_entry & RCR_FRAG_MASK);
2345 
2346 	l2_len = ((rcr_entry & RCR_L2_LEN_MASK) >> RCR_L2_LEN_SHIFT);
2347 
2348 	pktbufsz_type = ((rcr_entry & RCR_PKTBUFSZ_MASK) >>
2349 	    RCR_PKTBUFSZ_SHIFT);
2350 #if defined(__i386)
2351 	pkt_buf_addr_pp = (uint64_t *)(uint32_t)((rcr_entry &
2352 	    RCR_PKT_BUF_ADDR_MASK) << RCR_PKT_BUF_ADDR_SHIFT);
2353 #else
2354 	pkt_buf_addr_pp = (uint64_t *)((rcr_entry & RCR_PKT_BUF_ADDR_MASK) <<
2355 	    RCR_PKT_BUF_ADDR_SHIFT);
2356 #endif
2357 
2358 	channel = rcr_p->rdc;
2359 
2360 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
2361 	    "==> nxge_receive_packet: entryp $%p entry 0x%0llx "
2362 	    "pkt_buf_addr_pp $%p l2_len %d multi 0x%llx "
2363 	    "error_type 0x%x pkt_type 0x%x  "
2364 	    "pktbufsz_type %d ",
2365 	    rcr_desc_rd_head_p,
2366 	    rcr_entry, pkt_buf_addr_pp, l2_len,
2367 	    multi,
2368 	    error_type,
2369 	    pkt_type,
2370 	    pktbufsz_type));
2371 
2372 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
2373 	    "==> nxge_receive_packet: entryp $%p entry 0x%0llx "
2374 	    "pkt_buf_addr_pp $%p l2_len %d multi 0x%llx "
2375 	    "error_type 0x%x pkt_type 0x%x ", rcr_desc_rd_head_p,
2376 	    rcr_entry, pkt_buf_addr_pp, l2_len,
2377 	    multi,
2378 	    error_type,
2379 	    pkt_type));
2380 
2381 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
2382 	    "==> (rbr) nxge_receive_packet: entry 0x%0llx "
2383 	    "full pkt_buf_addr_pp $%p l2_len %d",
2384 	    rcr_entry, pkt_buf_addr_pp, l2_len));
2385 
2386 	/* get the stats ptr */
2387 	rdc_stats = rcr_p->rdc_stats;
2388 
2389 	if (!l2_len) {
2390 
2391 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
2392 		    "<== nxge_receive_packet: failed: l2 length is 0."));
2393 		return;
2394 	}
2395 
2396 	/*
2397 	 * Software workaround for BMAC hardware limitation that allows
2398 	 * maxframe size of 1526, instead of 1522 for non-jumbo and 0x2406
2399 	 * instead of 0x2400 for jumbo.
2400 	 */
2401 	if (l2_len > nxgep->mac.maxframesize) {
2402 		pkt_too_long_err = B_TRUE;
2403 	}
2404 
2405 	/* Hardware sends us 4 bytes of CRC as no stripping is done.  */
2406 	l2_len -= ETHERFCSL;
2407 
2408 	/* shift 6 bits to get the full io address */
2409 #if defined(__i386)
2410 	pkt_buf_addr_pp = (uint64_t *)((uint32_t)pkt_buf_addr_pp <<
2411 	    RCR_PKT_BUF_ADDR_SHIFT_FULL);
2412 #else
2413 	pkt_buf_addr_pp = (uint64_t *)((uint64_t)pkt_buf_addr_pp <<
2414 	    RCR_PKT_BUF_ADDR_SHIFT_FULL);
2415 #endif
2416 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
2417 	    "==> (rbr) nxge_receive_packet: entry 0x%0llx "
2418 	    "full pkt_buf_addr_pp $%p l2_len %d",
2419 	    rcr_entry, pkt_buf_addr_pp, l2_len));
2420 
2421 	rx_rbr_p = rcr_p->rx_rbr_p;
2422 	rx_msg_ring_p = rx_rbr_p->rx_msg_ring;
2423 
2424 	if (first_entry) {
2425 		hdr_size = (rcr_p->full_hdr_flag ? RXDMA_HDR_SIZE_FULL :
2426 		    RXDMA_HDR_SIZE_DEFAULT);
2427 
2428 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
2429 		    "==> nxge_receive_packet: first entry 0x%016llx "
2430 		    "pkt_buf_addr_pp $%p l2_len %d hdr %d",
2431 		    rcr_entry, pkt_buf_addr_pp, l2_len,
2432 		    hdr_size));
2433 	}
2434 
2435 	MUTEX_ENTER(&rx_rbr_p->lock);
2436 
2437 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
2438 	    "==> (rbr 1) nxge_receive_packet: entry 0x%0llx "
2439 	    "full pkt_buf_addr_pp $%p l2_len %d",
2440 	    rcr_entry, pkt_buf_addr_pp, l2_len));
2441 
2442 	/*
2443 	 * Packet buffer address in the completion entry points
2444 	 * to the starting buffer address (offset 0).
2445 	 * Use the starting buffer address to locate the corresponding
2446 	 * kernel address.
2447 	 */
2448 	status = nxge_rxbuf_pp_to_vp(nxgep, rx_rbr_p,
2449 	    pktbufsz_type, pkt_buf_addr_pp, &pkt_buf_addr_p,
2450 	    &buf_offset,
2451 	    &msg_index);
2452 
2453 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
2454 	    "==> (rbr 2) nxge_receive_packet: entry 0x%0llx "
2455 	    "full pkt_buf_addr_pp $%p l2_len %d",
2456 	    rcr_entry, pkt_buf_addr_pp, l2_len));
2457 
2458 	if (status != NXGE_OK) {
2459 		MUTEX_EXIT(&rx_rbr_p->lock);
2460 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
2461 		    "<== nxge_receive_packet: found vaddr failed %d",
2462 		    status));
2463 		return;
2464 	}
2465 
2466 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
2467 	    "==> (rbr 3) nxge_receive_packet: entry 0x%0llx "
2468 	    "full pkt_buf_addr_pp $%p l2_len %d",
2469 	    rcr_entry, pkt_buf_addr_pp, l2_len));
2470 
2471 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
2472 	    "==> (rbr 4 msgindex %d) nxge_receive_packet: entry 0x%0llx "
2473 	    "full pkt_buf_addr_pp $%p l2_len %d",
2474 	    msg_index, rcr_entry, pkt_buf_addr_pp, l2_len));
2475 
2476 	rx_msg_p = rx_msg_ring_p[msg_index];
2477 
2478 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
2479 	    "==> (rbr 4 msgindex %d) nxge_receive_packet: entry 0x%0llx "
2480 	    "full pkt_buf_addr_pp $%p l2_len %d",
2481 	    msg_index, rcr_entry, pkt_buf_addr_pp, l2_len));
2482 
2483 	switch (pktbufsz_type) {
2484 	case RCR_PKTBUFSZ_0:
2485 		bsize = rx_rbr_p->pkt_buf_size0_bytes;
2486 		NXGE_DEBUG_MSG((nxgep, RX2_CTL,
2487 		    "==> nxge_receive_packet: 0 buf %d", bsize));
2488 		break;
2489 	case RCR_PKTBUFSZ_1:
2490 		bsize = rx_rbr_p->pkt_buf_size1_bytes;
2491 		NXGE_DEBUG_MSG((nxgep, RX2_CTL,
2492 		    "==> nxge_receive_packet: 1 buf %d", bsize));
2493 		break;
2494 	case RCR_PKTBUFSZ_2:
2495 		bsize = rx_rbr_p->pkt_buf_size2_bytes;
2496 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
2497 		    "==> nxge_receive_packet: 2 buf %d", bsize));
2498 		break;
2499 	case RCR_SINGLE_BLOCK:
2500 		bsize = rx_msg_p->block_size;
2501 		NXGE_DEBUG_MSG((nxgep, RX2_CTL,
2502 		    "==> nxge_receive_packet: single %d", bsize));
2503 
2504 		break;
2505 	default:
2506 		MUTEX_EXIT(&rx_rbr_p->lock);
2507 		return;
2508 	}
2509 
2510 	DMA_COMMON_SYNC_OFFSET(rx_msg_p->buf_dma,
2511 	    (buf_offset + sw_offset_bytes),
2512 	    (hdr_size + l2_len),
2513 	    DDI_DMA_SYNC_FORCPU);
2514 
2515 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
2516 	    "==> nxge_receive_packet: after first dump:usage count"));
2517 
2518 	if (rx_msg_p->cur_usage_cnt == 0) {
2519 		if (rx_rbr_p->rbr_use_bcopy) {
2520 			atomic_inc_32(&rx_rbr_p->rbr_consumed);
2521 			if (rx_rbr_p->rbr_consumed <
2522 			    rx_rbr_p->rbr_threshold_hi) {
2523 				if (rx_rbr_p->rbr_threshold_lo == 0 ||
2524 				    ((rx_rbr_p->rbr_consumed >=
2525 				    rx_rbr_p->rbr_threshold_lo) &&
2526 				    (rx_rbr_p->rbr_bufsize_type >=
2527 				    pktbufsz_type))) {
2528 					rx_msg_p->rx_use_bcopy = B_TRUE;
2529 				}
2530 			} else {
2531 				rx_msg_p->rx_use_bcopy = B_TRUE;
2532 			}
2533 		}
2534 		NXGE_DEBUG_MSG((nxgep, RX2_CTL,
2535 		    "==> nxge_receive_packet: buf %d (new block) ",
2536 		    bsize));
2537 
2538 		rx_msg_p->pkt_buf_size_code = pktbufsz_type;
2539 		rx_msg_p->pkt_buf_size = bsize;
2540 		rx_msg_p->cur_usage_cnt = 1;
2541 		if (pktbufsz_type == RCR_SINGLE_BLOCK) {
2542 			NXGE_DEBUG_MSG((nxgep, RX2_CTL,
2543 			    "==> nxge_receive_packet: buf %d "
2544 			    "(single block) ",
2545 			    bsize));
2546 			/*
2547 			 * Buffer can be reused once the free function
2548 			 * is called.
2549 			 */
2550 			rx_msg_p->max_usage_cnt = 1;
2551 			buffer_free = B_TRUE;
2552 		} else {
2553 			rx_msg_p->max_usage_cnt = rx_msg_p->block_size/bsize;
2554 			if (rx_msg_p->max_usage_cnt == 1) {
2555 				buffer_free = B_TRUE;
2556 			}
2557 		}
2558 	} else {
2559 		rx_msg_p->cur_usage_cnt++;
2560 		if (rx_msg_p->cur_usage_cnt == rx_msg_p->max_usage_cnt) {
2561 			buffer_free = B_TRUE;
2562 		}
2563 	}
2564 
2565 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
2566 	    "msgbuf index = %d l2len %d bytes usage %d max_usage %d ",
2567 	    msg_index, l2_len,
2568 	    rx_msg_p->cur_usage_cnt, rx_msg_p->max_usage_cnt));
2569 
2570 	if ((error_type) || (dcf_err) || (pkt_too_long_err)) {
2571 		rdc_stats->ierrors++;
2572 		if (dcf_err) {
2573 			rdc_stats->dcf_err++;
2574 #ifdef	NXGE_DEBUG
2575 			if (!rdc_stats->dcf_err) {
2576 				NXGE_DEBUG_MSG((nxgep, RX_CTL,
2577 				"nxge_receive_packet: channel %d dcf_err rcr"
2578 				" 0x%llx", channel, rcr_entry));
2579 			}
2580 #endif
2581 			NXGE_FM_REPORT_ERROR(nxgep, nxgep->mac.portnum, NULL,
2582 			    NXGE_FM_EREPORT_RDMC_DCF_ERR);
2583 		} else if (pkt_too_long_err) {
2584 			rdc_stats->pkt_too_long_err++;
2585 			NXGE_DEBUG_MSG((nxgep, RX_CTL, " nxge_receive_packet:"
2586 			    " channel %d packet length [%d] > "
2587 			    "maxframesize [%d]", channel, l2_len + ETHERFCSL,
2588 			    nxgep->mac.maxframesize));
2589 		} else {
2590 				/* Update error stats */
2591 			error_disp_cnt = NXGE_ERROR_SHOW_MAX;
2592 			rdc_stats->errlog.compl_err_type = error_type;
2593 
2594 			switch (error_type) {
2595 			/*
2596 			 * Do not send FMA ereport for RCR_L2_ERROR and
2597 			 * RCR_L4_CSUM_ERROR because most likely they indicate
2598 			 * back pressure rather than HW failures.
2599 			 */
2600 			case RCR_L2_ERROR:
2601 				rdc_stats->l2_err++;
2602 				if (rdc_stats->l2_err <
2603 				    error_disp_cnt) {
2604 					NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
2605 					    " nxge_receive_packet:"
2606 					    " channel %d RCR L2_ERROR",
2607 					    channel));
2608 				}
2609 				break;
2610 			case RCR_L4_CSUM_ERROR:
2611 				error_send_up = B_TRUE;
2612 				rdc_stats->l4_cksum_err++;
2613 				if (rdc_stats->l4_cksum_err <
2614 				    error_disp_cnt) {
2615 					NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
2616 					    " nxge_receive_packet:"
2617 					    " channel %d"
2618 					    " RCR L4_CSUM_ERROR", channel));
2619 				}
2620 				break;
2621 			/*
2622 			 * Do not send FMA ereport for RCR_FFLP_SOFT_ERROR and
2623 			 * RCR_ZCP_SOFT_ERROR because they reflect the same
2624 			 * FFLP and ZCP errors that have been reported by
2625 			 * nxge_fflp.c and nxge_zcp.c.
2626 			 */
2627 			case RCR_FFLP_SOFT_ERROR:
2628 				error_send_up = B_TRUE;
2629 				rdc_stats->fflp_soft_err++;
2630 				if (rdc_stats->fflp_soft_err <
2631 				    error_disp_cnt) {
2632 					NXGE_ERROR_MSG((nxgep,
2633 					    NXGE_ERR_CTL,
2634 					    " nxge_receive_packet:"
2635 					    " channel %d"
2636 					    " RCR FFLP_SOFT_ERROR", channel));
2637 				}
2638 				break;
2639 			case RCR_ZCP_SOFT_ERROR:
2640 				error_send_up = B_TRUE;
2641 				rdc_stats->fflp_soft_err++;
2642 				if (rdc_stats->zcp_soft_err <
2643 				    error_disp_cnt)
2644 					NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
2645 					    " nxge_receive_packet: Channel %d"
2646 					    " RCR ZCP_SOFT_ERROR", channel));
2647 				break;
2648 			default:
2649 				rdc_stats->rcr_unknown_err++;
2650 				if (rdc_stats->rcr_unknown_err
2651 				    < error_disp_cnt) {
2652 					NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
2653 					    " nxge_receive_packet: Channel %d"
2654 					    " RCR entry 0x%llx error 0x%x",
2655 					    rcr_entry, channel, error_type));
2656 				}
2657 				break;
2658 			}
2659 		}
2660 
2661 		/*
2662 		 * Update and repost buffer block if max usage
2663 		 * count is reached.
2664 		 */
2665 		if (error_send_up == B_FALSE) {
2666 			atomic_inc_32(&rx_msg_p->ref_cnt);
2667 			if (buffer_free == B_TRUE) {
2668 				rx_msg_p->free = B_TRUE;
2669 			}
2670 
2671 			MUTEX_EXIT(&rx_rbr_p->lock);
2672 			nxge_freeb(rx_msg_p);
2673 			return;
2674 		}
2675 	}
2676 
2677 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
2678 	    "==> nxge_receive_packet: DMA sync second "));
2679 
2680 	bytes_read = rcr_p->rcvd_pkt_bytes;
2681 	skip_len = sw_offset_bytes + hdr_size;
2682 	if (!rx_msg_p->rx_use_bcopy) {
2683 		/*
2684 		 * For loaned up buffers, the driver reference count
2685 		 * will be incremented first and then the free state.
2686 		 */
2687 		if ((nmp = nxge_dupb(rx_msg_p, buf_offset, bsize)) != NULL) {
2688 			if (first_entry) {
2689 				nmp->b_rptr = &nmp->b_rptr[skip_len];
2690 				if (l2_len < bsize - skip_len) {
2691 					nmp->b_wptr = &nmp->b_rptr[l2_len];
2692 				} else {
2693 					nmp->b_wptr = &nmp->b_rptr[bsize
2694 					    - skip_len];
2695 				}
2696 			} else {
2697 				if (l2_len - bytes_read < bsize) {
2698 					nmp->b_wptr =
2699 					    &nmp->b_rptr[l2_len - bytes_read];
2700 				} else {
2701 					nmp->b_wptr = &nmp->b_rptr[bsize];
2702 				}
2703 			}
2704 		}
2705 	} else {
2706 		if (first_entry) {
2707 			nmp = nxge_dupb_bcopy(rx_msg_p, buf_offset + skip_len,
2708 			    l2_len < bsize - skip_len ?
2709 			    l2_len : bsize - skip_len);
2710 		} else {
2711 			nmp = nxge_dupb_bcopy(rx_msg_p, buf_offset,
2712 			    l2_len - bytes_read < bsize ?
2713 			    l2_len - bytes_read : bsize);
2714 		}
2715 	}
2716 	if (nmp != NULL) {
2717 		if (first_entry) {
2718 			/*
2719 			 * Jumbo packets may be received with more than one
2720 			 * buffer, increment ipackets for the first entry only.
2721 			 */
2722 			rdc_stats->ipackets++;
2723 
2724 			/* Update ibytes for kstat. */
2725 			rdc_stats->ibytes += skip_len
2726 			    + l2_len < bsize ? l2_len : bsize;
2727 			/*
2728 			 * Update the number of bytes read so far for the
2729 			 * current frame.
2730 			 */
2731 			bytes_read  = nmp->b_wptr - nmp->b_rptr;
2732 		} else {
2733 			rdc_stats->ibytes += l2_len - bytes_read < bsize ?
2734 			    l2_len - bytes_read : bsize;
2735 			bytes_read += nmp->b_wptr - nmp->b_rptr;
2736 		}
2737 
2738 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
2739 		    "==> nxge_receive_packet after dupb: "
2740 		    "rbr consumed %d "
2741 		    "pktbufsz_type %d "
2742 		    "nmp $%p rptr $%p wptr $%p "
2743 		    "buf_offset %d bzise %d l2_len %d skip_len %d",
2744 		    rx_rbr_p->rbr_consumed,
2745 		    pktbufsz_type,
2746 		    nmp, nmp->b_rptr, nmp->b_wptr,
2747 		    buf_offset, bsize, l2_len, skip_len));
2748 	} else {
2749 		cmn_err(CE_WARN, "!nxge_receive_packet: "
2750 		    "update stats (error)");
2751 		atomic_inc_32(&rx_msg_p->ref_cnt);
2752 		if (buffer_free == B_TRUE) {
2753 			rx_msg_p->free = B_TRUE;
2754 		}
2755 		MUTEX_EXIT(&rx_rbr_p->lock);
2756 		nxge_freeb(rx_msg_p);
2757 		return;
2758 	}
2759 
2760 	if (buffer_free == B_TRUE) {
2761 		rx_msg_p->free = B_TRUE;
2762 	}
2763 
2764 	is_valid = (nmp != NULL);
2765 
2766 	rcr_p->rcvd_pkt_bytes = bytes_read;
2767 
2768 	MUTEX_EXIT(&rx_rbr_p->lock);
2769 
2770 	if (rx_msg_p->free && rx_msg_p->rx_use_bcopy) {
2771 		atomic_inc_32(&rx_msg_p->ref_cnt);
2772 		nxge_freeb(rx_msg_p);
2773 	}
2774 
2775 	if (is_valid) {
2776 		nmp->b_cont = NULL;
2777 		if (first_entry) {
2778 			*mp = nmp;
2779 			*mp_cont = NULL;
2780 		} else {
2781 			*mp_cont = nmp;
2782 		}
2783 	}
2784 
2785 	/*
2786 	 * ERROR, FRAG and PKT_TYPE are only reported in the first entry.
2787 	 * If a packet is not fragmented and no error bit is set, then
2788 	 * L4 checksum is OK.
2789 	 */
2790 
2791 	if (is_valid && !multi) {
2792 		/*
2793 		 * If the checksum flag nxge_chksum_offload
2794 		 * is 1, TCP and UDP packets can be sent
2795 		 * up with good checksum. If the checksum flag
2796 		 * is set to 0, checksum reporting will apply to
2797 		 * TCP packets only (workaround for a hardware bug).
2798 		 * If the checksum flag nxge_cksum_offload is
2799 		 * greater than 1, both TCP and UDP packets
2800 		 * will not be reported its hardware checksum results.
2801 		 */
2802 		if (nxge_cksum_offload == 1) {
2803 			is_tcp_udp = ((pkt_type == RCR_PKT_IS_TCP ||
2804 			    pkt_type == RCR_PKT_IS_UDP) ?
2805 			    B_TRUE: B_FALSE);
2806 		} else if (!nxge_cksum_offload) {
2807 			/* TCP checksum only. */
2808 			is_tcp_udp = ((pkt_type == RCR_PKT_IS_TCP) ?
2809 			    B_TRUE: B_FALSE);
2810 		}
2811 
2812 		NXGE_DEBUG_MSG((nxgep, RX_CTL, "==> nxge_receive_packet: "
2813 		    "is_valid 0x%x multi 0x%llx pkt %d frag %d error %d",
2814 		    is_valid, multi, is_tcp_udp, frag, error_type));
2815 
2816 		if (is_tcp_udp && !frag && !error_type) {
2817 			(void) hcksum_assoc(nmp, NULL, NULL, 0, 0, 0, 0,
2818 			    HCK_FULLCKSUM_OK | HCK_FULLCKSUM, 0);
2819 			NXGE_DEBUG_MSG((nxgep, RX_CTL,
2820 			    "==> nxge_receive_packet: Full tcp/udp cksum "
2821 			    "is_valid 0x%x multi 0x%llx pkt %d frag %d "
2822 			    "error %d",
2823 			    is_valid, multi, is_tcp_udp, frag, error_type));
2824 		}
2825 	}
2826 
2827 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
2828 	    "==> nxge_receive_packet: *mp 0x%016llx", *mp));
2829 
2830 	*multi_p = (multi == RCR_MULTI_MASK);
2831 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "<== nxge_receive_packet: "
2832 	    "multi %d nmp 0x%016llx *mp 0x%016llx *mp_cont 0x%016llx",
2833 	    *multi_p, nmp, *mp, *mp_cont));
2834 }
2835 
2836 /*
2837  * Enable polling for a ring. Interrupt for the ring is disabled when
2838  * the nxge interrupt comes (see nxge_rx_intr).
2839  */
2840 int
2841 nxge_enable_poll(void *arg)
2842 {
2843 	p_nxge_ring_handle_t	ring_handle = (p_nxge_ring_handle_t)arg;
2844 	p_rx_rcr_ring_t		ringp;
2845 	p_nxge_t		nxgep;
2846 	p_nxge_ldg_t		ldgp;
2847 	uint32_t		channel;
2848 
2849 	if (ring_handle == NULL) {
2850 		return (0);
2851 	}
2852 
2853 	nxgep = ring_handle->nxgep;
2854 	channel = nxgep->pt_config.hw_config.start_rdc + ring_handle->index;
2855 	ringp = nxgep->rx_rcr_rings->rcr_rings[channel];
2856 	NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
2857 	    "==> nxge_enable_poll: rdc %d ", ringp->rdc));
2858 	ldgp = ringp->ldgp;
2859 	if (ldgp == NULL) {
2860 		NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
2861 		    "==> nxge_enable_poll: rdc %d NULL ldgp: no change",
2862 		    ringp->rdc));
2863 		return (0);
2864 	}
2865 
2866 	MUTEX_ENTER(&ringp->lock);
2867 	/* enable polling */
2868 	if (ringp->poll_flag == 0) {
2869 		ringp->poll_flag = 1;
2870 		NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
2871 		    "==> nxge_enable_poll: rdc %d set poll flag to 1",
2872 		    ringp->rdc));
2873 	}
2874 
2875 	MUTEX_EXIT(&ringp->lock);
2876 	return (0);
2877 }
2878 /*
2879  * Disable polling for a ring and enable its interrupt.
2880  */
2881 int
2882 nxge_disable_poll(void *arg)
2883 {
2884 	p_nxge_ring_handle_t	ring_handle = (p_nxge_ring_handle_t)arg;
2885 	p_rx_rcr_ring_t		ringp;
2886 	p_nxge_t		nxgep;
2887 	uint32_t		channel;
2888 
2889 	if (ring_handle == NULL) {
2890 		return (0);
2891 	}
2892 
2893 	nxgep = ring_handle->nxgep;
2894 	channel = nxgep->pt_config.hw_config.start_rdc + ring_handle->index;
2895 	ringp = nxgep->rx_rcr_rings->rcr_rings[channel];
2896 
2897 	NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
2898 	    "==> nxge_disable_poll: rdc %d poll_flag %d", ringp->rdc));
2899 
2900 	MUTEX_ENTER(&ringp->lock);
2901 
2902 	/* disable polling: enable interrupt */
2903 	if (ringp->poll_flag) {
2904 		npi_handle_t		handle;
2905 		rx_dma_ctl_stat_t	cs;
2906 		uint8_t			channel;
2907 		p_nxge_ldg_t		ldgp;
2908 
2909 		/*
2910 		 * Get the control and status for this channel.
2911 		 */
2912 		handle = NXGE_DEV_NPI_HANDLE(nxgep);
2913 		channel = ringp->rdc;
2914 		RXDMA_REG_READ64(handle, RX_DMA_CTL_STAT_REG,
2915 		    channel, &cs.value);
2916 
2917 		/*
2918 		 * Enable mailbox update
2919 		 * Since packets were not read and the hardware uses
2920 		 * bits pktread and ptrread to update the queue
2921 		 * length, we need to set both bits to 0.
2922 		 */
2923 		cs.bits.ldw.pktread = 0;
2924 		cs.bits.ldw.ptrread = 0;
2925 		cs.bits.hdw.mex = 1;
2926 		RXDMA_REG_WRITE64(handle, RX_DMA_CTL_STAT_REG, channel,
2927 		    cs.value);
2928 
2929 		/*
2930 		 * Rearm this logical group if this is a single device
2931 		 * group.
2932 		 */
2933 		ldgp = ringp->ldgp;
2934 		if (ldgp == NULL) {
2935 			ringp->poll_flag = 0;
2936 			MUTEX_EXIT(&ringp->lock);
2937 			NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
2938 			    "==> nxge_disable_poll: no ldgp rdc %d "
2939 			    "(still set poll to 0", ringp->rdc));
2940 			return (0);
2941 		}
2942 		NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
2943 		    "==> nxge_disable_poll: rdc %d ldgp $%p (enable intr)",
2944 		    ringp->rdc, ldgp));
2945 		if (ldgp->nldvs == 1) {
2946 			ldgimgm_t	mgm;
2947 			mgm.value = 0;
2948 			mgm.bits.ldw.arm = 1;
2949 			mgm.bits.ldw.timer = ldgp->ldg_timer;
2950 			NXGE_REG_WR64(handle,
2951 			    LDGIMGN_REG + LDSV_OFFSET(ldgp->ldg), mgm.value);
2952 		}
2953 		ringp->poll_flag = 0;
2954 	}
2955 
2956 	MUTEX_EXIT(&ringp->lock);
2957 	return (0);
2958 }
2959 
2960 /*
2961  * Poll 'bytes_to_pickup' bytes of message from the rx ring.
2962  */
2963 mblk_t *
2964 nxge_rx_poll(void *arg, int bytes_to_pickup)
2965 {
2966 	p_nxge_ring_handle_t	ring_handle = (p_nxge_ring_handle_t)arg;
2967 	p_rx_rcr_ring_t		rcr_p;
2968 	p_nxge_t		nxgep;
2969 	npi_handle_t		handle;
2970 	rx_dma_ctl_stat_t	cs;
2971 	mblk_t			*mblk;
2972 	p_nxge_ldv_t		ldvp;
2973 	uint32_t		channel;
2974 
2975 	nxgep = ring_handle->nxgep;
2976 
2977 	/*
2978 	 * Get the control and status for this channel.
2979 	 */
2980 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
2981 	channel = nxgep->pt_config.hw_config.start_rdc + ring_handle->index;
2982 	rcr_p = nxgep->rx_rcr_rings->rcr_rings[channel];
2983 	MUTEX_ENTER(&rcr_p->lock);
2984 	ASSERT(rcr_p->poll_flag == 1);
2985 
2986 	RXDMA_REG_READ64(handle, RX_DMA_CTL_STAT_REG, rcr_p->rdc, &cs.value);
2987 
2988 	NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
2989 	    "==> nxge_rx_poll: calling nxge_rx_pkts: rdc %d poll_flag %d",
2990 	    rcr_p->rdc, rcr_p->poll_flag));
2991 	mblk = nxge_rx_pkts(nxgep, rcr_p, cs, bytes_to_pickup);
2992 
2993 	ldvp = rcr_p->ldvp;
2994 	/* error events. */
2995 	if (ldvp && (cs.value & RX_DMA_CTL_STAT_ERROR)) {
2996 		(void) nxge_rx_err_evnts(nxgep, ldvp->vdma_index, cs);
2997 	}
2998 
2999 	MUTEX_EXIT(&rcr_p->lock);
3000 
3001 	NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
3002 	    "<== nxge_rx_poll: rdc %d mblk $%p", rcr_p->rdc, mblk));
3003 	return (mblk);
3004 }
3005 
3006 
3007 /*ARGSUSED*/
3008 static nxge_status_t
3009 nxge_rx_err_evnts(p_nxge_t nxgep, int channel, rx_dma_ctl_stat_t cs)
3010 {
3011 	p_nxge_rx_ring_stats_t	rdc_stats;
3012 	npi_handle_t		handle;
3013 	npi_status_t		rs;
3014 	boolean_t		rxchan_fatal = B_FALSE;
3015 	boolean_t		rxport_fatal = B_FALSE;
3016 	uint8_t			portn;
3017 	nxge_status_t		status = NXGE_OK;
3018 	uint32_t		error_disp_cnt = NXGE_ERROR_SHOW_MAX;
3019 	NXGE_DEBUG_MSG((nxgep, INT_CTL, "==> nxge_rx_err_evnts"));
3020 
3021 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
3022 	portn = nxgep->mac.portnum;
3023 	rdc_stats = &nxgep->statsp->rdc_stats[channel];
3024 
3025 	if (cs.bits.hdw.rbr_tmout) {
3026 		rdc_stats->rx_rbr_tmout++;
3027 		NXGE_FM_REPORT_ERROR(nxgep, portn, channel,
3028 		    NXGE_FM_EREPORT_RDMC_RBR_TMOUT);
3029 		rxchan_fatal = B_TRUE;
3030 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3031 		    "==> nxge_rx_err_evnts: rx_rbr_timeout"));
3032 	}
3033 	if (cs.bits.hdw.rsp_cnt_err) {
3034 		rdc_stats->rsp_cnt_err++;
3035 		NXGE_FM_REPORT_ERROR(nxgep, portn, channel,
3036 		    NXGE_FM_EREPORT_RDMC_RSP_CNT_ERR);
3037 		rxchan_fatal = B_TRUE;
3038 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3039 		    "==> nxge_rx_err_evnts(channel %d): "
3040 		    "rsp_cnt_err", channel));
3041 	}
3042 	if (cs.bits.hdw.byte_en_bus) {
3043 		rdc_stats->byte_en_bus++;
3044 		NXGE_FM_REPORT_ERROR(nxgep, portn, channel,
3045 		    NXGE_FM_EREPORT_RDMC_BYTE_EN_BUS);
3046 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3047 		    "==> nxge_rx_err_evnts(channel %d): "
3048 		    "fatal error: byte_en_bus", channel));
3049 		rxchan_fatal = B_TRUE;
3050 	}
3051 	if (cs.bits.hdw.rsp_dat_err) {
3052 		rdc_stats->rsp_dat_err++;
3053 		NXGE_FM_REPORT_ERROR(nxgep, portn, channel,
3054 		    NXGE_FM_EREPORT_RDMC_RSP_DAT_ERR);
3055 		rxchan_fatal = B_TRUE;
3056 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3057 		    "==> nxge_rx_err_evnts(channel %d): "
3058 		    "fatal error: rsp_dat_err", channel));
3059 	}
3060 	if (cs.bits.hdw.rcr_ack_err) {
3061 		rdc_stats->rcr_ack_err++;
3062 		NXGE_FM_REPORT_ERROR(nxgep, portn, channel,
3063 		    NXGE_FM_EREPORT_RDMC_RCR_ACK_ERR);
3064 		rxchan_fatal = B_TRUE;
3065 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3066 		    "==> nxge_rx_err_evnts(channel %d): "
3067 		    "fatal error: rcr_ack_err", channel));
3068 	}
3069 	if (cs.bits.hdw.dc_fifo_err) {
3070 		rdc_stats->dc_fifo_err++;
3071 		NXGE_FM_REPORT_ERROR(nxgep, portn, channel,
3072 		    NXGE_FM_EREPORT_RDMC_DC_FIFO_ERR);
3073 		/* This is not a fatal error! */
3074 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3075 		    "==> nxge_rx_err_evnts(channel %d): "
3076 		    "dc_fifo_err", channel));
3077 		rxport_fatal = B_TRUE;
3078 	}
3079 	if ((cs.bits.hdw.rcr_sha_par) || (cs.bits.hdw.rbr_pre_par)) {
3080 		if ((rs = npi_rxdma_ring_perr_stat_get(handle,
3081 		    &rdc_stats->errlog.pre_par,
3082 		    &rdc_stats->errlog.sha_par))
3083 		    != NPI_SUCCESS) {
3084 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3085 			    "==> nxge_rx_err_evnts(channel %d): "
3086 			    "rcr_sha_par: get perr", channel));
3087 			return (NXGE_ERROR | rs);
3088 		}
3089 		if (cs.bits.hdw.rcr_sha_par) {
3090 			rdc_stats->rcr_sha_par++;
3091 			NXGE_FM_REPORT_ERROR(nxgep, portn, channel,
3092 			    NXGE_FM_EREPORT_RDMC_RCR_SHA_PAR);
3093 			rxchan_fatal = B_TRUE;
3094 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3095 			    "==> nxge_rx_err_evnts(channel %d): "
3096 			    "fatal error: rcr_sha_par", channel));
3097 		}
3098 		if (cs.bits.hdw.rbr_pre_par) {
3099 			rdc_stats->rbr_pre_par++;
3100 			NXGE_FM_REPORT_ERROR(nxgep, portn, channel,
3101 			    NXGE_FM_EREPORT_RDMC_RBR_PRE_PAR);
3102 			rxchan_fatal = B_TRUE;
3103 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3104 			    "==> nxge_rx_err_evnts(channel %d): "
3105 			    "fatal error: rbr_pre_par", channel));
3106 		}
3107 	}
3108 	/*
3109 	 * The Following 4 status bits are for information, the system
3110 	 * is running fine. There is no need to send FMA ereports or
3111 	 * log messages.
3112 	 */
3113 	if (cs.bits.hdw.port_drop_pkt) {
3114 		rdc_stats->port_drop_pkt++;
3115 	}
3116 	if (cs.bits.hdw.wred_drop) {
3117 		rdc_stats->wred_drop++;
3118 	}
3119 	if (cs.bits.hdw.rbr_pre_empty) {
3120 		rdc_stats->rbr_pre_empty++;
3121 	}
3122 	if (cs.bits.hdw.rcr_shadow_full) {
3123 		rdc_stats->rcr_shadow_full++;
3124 	}
3125 	if (cs.bits.hdw.config_err) {
3126 		rdc_stats->config_err++;
3127 		NXGE_FM_REPORT_ERROR(nxgep, portn, channel,
3128 		    NXGE_FM_EREPORT_RDMC_CONFIG_ERR);
3129 		rxchan_fatal = B_TRUE;
3130 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3131 		    "==> nxge_rx_err_evnts(channel %d): "
3132 		    "config error", channel));
3133 	}
3134 	if (cs.bits.hdw.rcrincon) {
3135 		rdc_stats->rcrincon++;
3136 		NXGE_FM_REPORT_ERROR(nxgep, portn, channel,
3137 		    NXGE_FM_EREPORT_RDMC_RCRINCON);
3138 		rxchan_fatal = B_TRUE;
3139 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3140 		    "==> nxge_rx_err_evnts(channel %d): "
3141 		    "fatal error: rcrincon error", channel));
3142 	}
3143 	if (cs.bits.hdw.rcrfull) {
3144 		rdc_stats->rcrfull++;
3145 		NXGE_FM_REPORT_ERROR(nxgep, portn, channel,
3146 		    NXGE_FM_EREPORT_RDMC_RCRFULL);
3147 		rxchan_fatal = B_TRUE;
3148 		if (rdc_stats->rcrfull < error_disp_cnt)
3149 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3150 		    "==> nxge_rx_err_evnts(channel %d): "
3151 		    "fatal error: rcrfull error", channel));
3152 	}
3153 	if (cs.bits.hdw.rbr_empty) {
3154 		/*
3155 		 * This bit is for information, there is no need
3156 		 * send FMA ereport or log a message.
3157 		 */
3158 		rdc_stats->rbr_empty++;
3159 	}
3160 	if (cs.bits.hdw.rbrfull) {
3161 		rdc_stats->rbrfull++;
3162 		NXGE_FM_REPORT_ERROR(nxgep, portn, channel,
3163 		    NXGE_FM_EREPORT_RDMC_RBRFULL);
3164 		rxchan_fatal = B_TRUE;
3165 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3166 		    "==> nxge_rx_err_evnts(channel %d): "
3167 		    "fatal error: rbr_full error", channel));
3168 	}
3169 	if (cs.bits.hdw.rbrlogpage) {
3170 		rdc_stats->rbrlogpage++;
3171 		NXGE_FM_REPORT_ERROR(nxgep, portn, channel,
3172 		    NXGE_FM_EREPORT_RDMC_RBRLOGPAGE);
3173 		rxchan_fatal = B_TRUE;
3174 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3175 		    "==> nxge_rx_err_evnts(channel %d): "
3176 		    "fatal error: rbr logical page error", channel));
3177 	}
3178 	if (cs.bits.hdw.cfiglogpage) {
3179 		rdc_stats->cfiglogpage++;
3180 		NXGE_FM_REPORT_ERROR(nxgep, portn, channel,
3181 		    NXGE_FM_EREPORT_RDMC_CFIGLOGPAGE);
3182 		rxchan_fatal = B_TRUE;
3183 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3184 		    "==> nxge_rx_err_evnts(channel %d): "
3185 		    "fatal error: cfig logical page error", channel));
3186 	}
3187 
3188 	if (rxport_fatal)  {
3189 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3190 		    " nxge_rx_err_evnts: fatal error on Port #%d\n",
3191 		    portn));
3192 		if (isLDOMguest(nxgep)) {
3193 			status = NXGE_ERROR;
3194 		} else {
3195 			status = nxge_ipp_fatal_err_recover(nxgep);
3196 			if (status == NXGE_OK) {
3197 				FM_SERVICE_RESTORED(nxgep);
3198 			}
3199 		}
3200 	}
3201 
3202 	if (rxchan_fatal) {
3203 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3204 		    " nxge_rx_err_evnts: fatal error on Channel #%d\n",
3205 		    channel));
3206 		if (isLDOMguest(nxgep)) {
3207 			status = NXGE_ERROR;
3208 		} else {
3209 			status = nxge_rxdma_fatal_err_recover(nxgep, channel);
3210 			if (status == NXGE_OK) {
3211 				FM_SERVICE_RESTORED(nxgep);
3212 			}
3213 		}
3214 	}
3215 
3216 	NXGE_DEBUG_MSG((nxgep, RX2_CTL, "<== nxge_rx_err_evnts"));
3217 
3218 	return (status);
3219 }
3220 
3221 /*
3222  * nxge_rdc_hvio_setup
3223  *
3224  *	This code appears to setup some Hypervisor variables.
3225  *
3226  * Arguments:
3227  * 	nxgep
3228  * 	channel
3229  *
3230  * Notes:
3231  *	What does NIU_LP_WORKAROUND mean?
3232  *
3233  * NPI/NXGE function calls:
3234  *	na
3235  *
3236  * Context:
3237  *	Any domain
3238  */
3239 #if defined(sun4v) && defined(NIU_LP_WORKAROUND)
3240 static void
3241 nxge_rdc_hvio_setup(
3242 	nxge_t *nxgep, int channel)
3243 {
3244 	nxge_dma_common_t	*dma_common;
3245 	nxge_dma_common_t	*dma_control;
3246 	rx_rbr_ring_t		*ring;
3247 
3248 	ring = nxgep->rx_rbr_rings->rbr_rings[channel];
3249 	dma_common = nxgep->rx_buf_pool_p->dma_buf_pool_p[channel];
3250 
3251 	ring->hv_set = B_FALSE;
3252 
3253 	ring->hv_rx_buf_base_ioaddr_pp = (uint64_t)
3254 	    dma_common->orig_ioaddr_pp;
3255 	ring->hv_rx_buf_ioaddr_size = (uint64_t)
3256 	    dma_common->orig_alength;
3257 
3258 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_map_rxdma_channel: "
3259 	    "channel %d data buf base io $%lx ($%p) size 0x%lx (%ld 0x%lx)",
3260 	    channel, ring->hv_rx_buf_base_ioaddr_pp,
3261 	    dma_common->ioaddr_pp, ring->hv_rx_buf_ioaddr_size,
3262 	    dma_common->orig_alength, dma_common->orig_alength));
3263 
3264 	dma_control = nxgep->rx_cntl_pool_p->dma_buf_pool_p[channel];
3265 
3266 	ring->hv_rx_cntl_base_ioaddr_pp =
3267 	    (uint64_t)dma_control->orig_ioaddr_pp;
3268 	ring->hv_rx_cntl_ioaddr_size =
3269 	    (uint64_t)dma_control->orig_alength;
3270 
3271 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_map_rxdma_channel: "
3272 	    "channel %d cntl base io $%p ($%p) size 0x%llx (%d 0x%x)",
3273 	    channel, ring->hv_rx_cntl_base_ioaddr_pp,
3274 	    dma_control->ioaddr_pp, ring->hv_rx_cntl_ioaddr_size,
3275 	    dma_control->orig_alength, dma_control->orig_alength));
3276 }
3277 #endif
3278 
3279 /*
3280  * nxge_map_rxdma
3281  *
3282  *	Map an RDC into our kernel space.
3283  *
3284  * Arguments:
3285  * 	nxgep
3286  * 	channel	The channel to map.
3287  *
3288  * Notes:
3289  *	1. Allocate & initialise a memory pool, if necessary.
3290  *	2. Allocate however many receive buffers are required.
3291  *	3. Setup buffers, descriptors, and mailbox.
3292  *
3293  * NPI/NXGE function calls:
3294  *	nxge_alloc_rx_mem_pool()
3295  *	nxge_alloc_rbb()
3296  *	nxge_map_rxdma_channel()
3297  *
3298  * Registers accessed:
3299  *
3300  * Context:
3301  *	Any domain
3302  */
3303 static nxge_status_t
3304 nxge_map_rxdma(p_nxge_t nxgep, int channel)
3305 {
3306 	nxge_dma_common_t	**data;
3307 	nxge_dma_common_t	**control;
3308 	rx_rbr_ring_t		**rbr_ring;
3309 	rx_rcr_ring_t		**rcr_ring;
3310 	rx_mbox_t		**mailbox;
3311 	uint32_t		chunks;
3312 
3313 	nxge_status_t		status;
3314 
3315 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_map_rxdma"));
3316 
3317 	if (!nxgep->rx_buf_pool_p) {
3318 		if (nxge_alloc_rx_mem_pool(nxgep) != NXGE_OK) {
3319 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3320 			    "<== nxge_map_rxdma: buf not allocated"));
3321 			return (NXGE_ERROR);
3322 		}
3323 	}
3324 
3325 	if (nxge_alloc_rxb(nxgep, channel) != NXGE_OK)
3326 		return (NXGE_ERROR);
3327 
3328 	/*
3329 	 * Timeout should be set based on the system clock divider.
3330 	 * A timeout value of 1 assumes that the
3331 	 * granularity (1000) is 3 microseconds running at 300MHz.
3332 	 */
3333 
3334 	nxgep->intr_threshold = nxge_rcr_threshold;
3335 	nxgep->intr_timeout = nxge_rcr_timeout;
3336 
3337 	/*
3338 	 * Map descriptors from the buffer polls for each dma channel.
3339 	 */
3340 
3341 	/*
3342 	 * Set up and prepare buffer blocks, descriptors
3343 	 * and mailbox.
3344 	 */
3345 	data = &nxgep->rx_buf_pool_p->dma_buf_pool_p[channel];
3346 	rbr_ring = &nxgep->rx_rbr_rings->rbr_rings[channel];
3347 	chunks = nxgep->rx_buf_pool_p->num_chunks[channel];
3348 
3349 	control = &nxgep->rx_cntl_pool_p->dma_buf_pool_p[channel];
3350 	rcr_ring = &nxgep->rx_rcr_rings->rcr_rings[channel];
3351 
3352 	mailbox = &nxgep->rx_mbox_areas_p->rxmbox_areas[channel];
3353 
3354 	status = nxge_map_rxdma_channel(nxgep, channel, data, rbr_ring,
3355 	    chunks, control, rcr_ring, mailbox);
3356 	if (status != NXGE_OK) {
3357 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3358 		    "==> nxge_map_rxdma: nxge_map_rxdma_channel(%d) "
3359 		    "returned 0x%x",
3360 		    channel, status));
3361 		return (status);
3362 	}
3363 	nxgep->rx_rbr_rings->rbr_rings[channel]->index = (uint16_t)channel;
3364 	nxgep->rx_rcr_rings->rcr_rings[channel]->index = (uint16_t)channel;
3365 	nxgep->rx_rcr_rings->rcr_rings[channel]->rdc_stats =
3366 	    &nxgep->statsp->rdc_stats[channel];
3367 
3368 #if	defined(sun4v) && defined(NIU_LP_WORKAROUND)
3369 	if (!isLDOMguest(nxgep))
3370 		nxge_rdc_hvio_setup(nxgep, channel);
3371 #endif
3372 
3373 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3374 	    "<== nxge_map_rxdma: (status 0x%x channel %d)", status, channel));
3375 
3376 	return (status);
3377 }
3378 
3379 static void
3380 nxge_unmap_rxdma(p_nxge_t nxgep, int channel)
3381 {
3382 	rx_rbr_ring_t	*rbr_ring;
3383 	rx_rcr_ring_t	*rcr_ring;
3384 	rx_mbox_t	*mailbox;
3385 
3386 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_unmap_rxdma(%d)", channel));
3387 
3388 	if (!nxgep->rx_rbr_rings || !nxgep->rx_rcr_rings ||
3389 	    !nxgep->rx_mbox_areas_p)
3390 		return;
3391 
3392 	rbr_ring = nxgep->rx_rbr_rings->rbr_rings[channel];
3393 	rcr_ring = nxgep->rx_rcr_rings->rcr_rings[channel];
3394 	mailbox = nxgep->rx_mbox_areas_p->rxmbox_areas[channel];
3395 
3396 	if (!rbr_ring || !rcr_ring || !mailbox)
3397 		return;
3398 
3399 	(void) nxge_unmap_rxdma_channel(
3400 	    nxgep, channel, rbr_ring, rcr_ring, mailbox);
3401 
3402 	nxge_free_rxb(nxgep, channel);
3403 
3404 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "<== nxge_unmap_rxdma"));
3405 }
3406 
3407 nxge_status_t
3408 nxge_map_rxdma_channel(p_nxge_t nxgep, uint16_t channel,
3409     p_nxge_dma_common_t *dma_buf_p,  p_rx_rbr_ring_t *rbr_p,
3410     uint32_t num_chunks,
3411     p_nxge_dma_common_t *dma_cntl_p, p_rx_rcr_ring_t *rcr_p,
3412     p_rx_mbox_t *rx_mbox_p)
3413 {
3414 	int	status = NXGE_OK;
3415 
3416 	/*
3417 	 * Set up and prepare buffer blocks, descriptors
3418 	 * and mailbox.
3419 	 */
3420 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3421 	    "==> nxge_map_rxdma_channel (channel %d)", channel));
3422 	/*
3423 	 * Receive buffer blocks
3424 	 */
3425 	status = nxge_map_rxdma_channel_buf_ring(nxgep, channel,
3426 	    dma_buf_p, rbr_p, num_chunks);
3427 	if (status != NXGE_OK) {
3428 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3429 		    "==> nxge_map_rxdma_channel (channel %d): "
3430 		    "map buffer failed 0x%x", channel, status));
3431 		goto nxge_map_rxdma_channel_exit;
3432 	}
3433 
3434 	/*
3435 	 * Receive block ring, completion ring and mailbox.
3436 	 */
3437 	status = nxge_map_rxdma_channel_cfg_ring(nxgep, channel,
3438 	    dma_cntl_p, rbr_p, rcr_p, rx_mbox_p);
3439 	if (status != NXGE_OK) {
3440 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3441 		    "==> nxge_map_rxdma_channel (channel %d): "
3442 		    "map config failed 0x%x", channel, status));
3443 		goto nxge_map_rxdma_channel_fail2;
3444 	}
3445 
3446 	goto nxge_map_rxdma_channel_exit;
3447 
3448 nxge_map_rxdma_channel_fail3:
3449 	/* Free rbr, rcr */
3450 	NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3451 	    "==> nxge_map_rxdma_channel: free rbr/rcr "
3452 	    "(status 0x%x channel %d)",
3453 	    status, channel));
3454 	nxge_unmap_rxdma_channel_cfg_ring(nxgep,
3455 	    *rcr_p, *rx_mbox_p);
3456 
3457 nxge_map_rxdma_channel_fail2:
3458 	/* Free buffer blocks */
3459 	NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3460 	    "==> nxge_map_rxdma_channel: free rx buffers"
3461 	    "(nxgep 0x%x status 0x%x channel %d)",
3462 	    nxgep, status, channel));
3463 	nxge_unmap_rxdma_channel_buf_ring(nxgep, *rbr_p);
3464 
3465 	status = NXGE_ERROR;
3466 
3467 nxge_map_rxdma_channel_exit:
3468 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3469 	    "<== nxge_map_rxdma_channel: "
3470 	    "(nxgep 0x%x status 0x%x channel %d)",
3471 	    nxgep, status, channel));
3472 
3473 	return (status);
3474 }
3475 
3476 /*ARGSUSED*/
3477 static void
3478 nxge_unmap_rxdma_channel(p_nxge_t nxgep, uint16_t channel,
3479     p_rx_rbr_ring_t rbr_p, p_rx_rcr_ring_t rcr_p, p_rx_mbox_t rx_mbox_p)
3480 {
3481 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3482 	    "==> nxge_unmap_rxdma_channel (channel %d)", channel));
3483 
3484 	/*
3485 	 * unmap receive block ring, completion ring and mailbox.
3486 	 */
3487 	(void) nxge_unmap_rxdma_channel_cfg_ring(nxgep,
3488 	    rcr_p, rx_mbox_p);
3489 
3490 	/* unmap buffer blocks */
3491 	(void) nxge_unmap_rxdma_channel_buf_ring(nxgep, rbr_p);
3492 
3493 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "<== nxge_unmap_rxdma_channel"));
3494 }
3495 
3496 /*ARGSUSED*/
3497 static nxge_status_t
3498 nxge_map_rxdma_channel_cfg_ring(p_nxge_t nxgep, uint16_t dma_channel,
3499     p_nxge_dma_common_t *dma_cntl_p, p_rx_rbr_ring_t *rbr_p,
3500     p_rx_rcr_ring_t *rcr_p, p_rx_mbox_t *rx_mbox_p)
3501 {
3502 	p_rx_rbr_ring_t 	rbrp;
3503 	p_rx_rcr_ring_t 	rcrp;
3504 	p_rx_mbox_t 		mboxp;
3505 	p_nxge_dma_common_t 	cntl_dmap;
3506 	p_nxge_dma_common_t 	dmap;
3507 	p_rx_msg_t 		*rx_msg_ring;
3508 	p_rx_msg_t 		rx_msg_p;
3509 	p_rbr_cfig_a_t		rcfga_p;
3510 	p_rbr_cfig_b_t		rcfgb_p;
3511 	p_rcrcfig_a_t		cfga_p;
3512 	p_rcrcfig_b_t		cfgb_p;
3513 	p_rxdma_cfig1_t		cfig1_p;
3514 	p_rxdma_cfig2_t		cfig2_p;
3515 	p_rbr_kick_t		kick_p;
3516 	uint32_t		dmaaddrp;
3517 	uint32_t		*rbr_vaddrp;
3518 	uint32_t		bkaddr;
3519 	nxge_status_t		status = NXGE_OK;
3520 	int			i;
3521 	uint32_t 		nxge_port_rcr_size;
3522 
3523 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3524 	    "==> nxge_map_rxdma_channel_cfg_ring"));
3525 
3526 	cntl_dmap = *dma_cntl_p;
3527 
3528 	/* Map in the receive block ring */
3529 	rbrp = *rbr_p;
3530 	dmap = (p_nxge_dma_common_t)&rbrp->rbr_desc;
3531 	nxge_setup_dma_common(dmap, cntl_dmap, rbrp->rbb_max, 4);
3532 	/*
3533 	 * Zero out buffer block ring descriptors.
3534 	 */
3535 	bzero((caddr_t)dmap->kaddrp, dmap->alength);
3536 
3537 	rcfga_p = &(rbrp->rbr_cfga);
3538 	rcfgb_p = &(rbrp->rbr_cfgb);
3539 	kick_p = &(rbrp->rbr_kick);
3540 	rcfga_p->value = 0;
3541 	rcfgb_p->value = 0;
3542 	kick_p->value = 0;
3543 	rbrp->rbr_addr = dmap->dma_cookie.dmac_laddress;
3544 	rcfga_p->value = (rbrp->rbr_addr &
3545 	    (RBR_CFIG_A_STDADDR_MASK |
3546 	    RBR_CFIG_A_STDADDR_BASE_MASK));
3547 	rcfga_p->value |= ((uint64_t)rbrp->rbb_max << RBR_CFIG_A_LEN_SHIFT);
3548 
3549 	rcfgb_p->bits.ldw.bufsz0 = rbrp->pkt_buf_size0;
3550 	rcfgb_p->bits.ldw.vld0 = 1;
3551 	rcfgb_p->bits.ldw.bufsz1 = rbrp->pkt_buf_size1;
3552 	rcfgb_p->bits.ldw.vld1 = 1;
3553 	rcfgb_p->bits.ldw.bufsz2 = rbrp->pkt_buf_size2;
3554 	rcfgb_p->bits.ldw.vld2 = 1;
3555 	rcfgb_p->bits.ldw.bksize = nxgep->rx_bksize_code;
3556 
3557 	/*
3558 	 * For each buffer block, enter receive block address to the ring.
3559 	 */
3560 	rbr_vaddrp = (uint32_t *)dmap->kaddrp;
3561 	rbrp->rbr_desc_vp = (uint32_t *)dmap->kaddrp;
3562 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3563 	    "==> nxge_map_rxdma_channel_cfg_ring: channel %d "
3564 	    "rbr_vaddrp $%p", dma_channel, rbr_vaddrp));
3565 
3566 	rx_msg_ring = rbrp->rx_msg_ring;
3567 	for (i = 0; i < rbrp->tnblocks; i++) {
3568 		rx_msg_p = rx_msg_ring[i];
3569 		rx_msg_p->nxgep = nxgep;
3570 		rx_msg_p->rx_rbr_p = rbrp;
3571 		bkaddr = (uint32_t)
3572 		    ((rx_msg_p->buf_dma.dma_cookie.dmac_laddress
3573 		    >> RBR_BKADDR_SHIFT));
3574 		rx_msg_p->free = B_FALSE;
3575 		rx_msg_p->max_usage_cnt = 0xbaddcafe;
3576 
3577 		*rbr_vaddrp++ = bkaddr;
3578 	}
3579 
3580 	kick_p->bits.ldw.bkadd = rbrp->rbb_max;
3581 	rbrp->rbr_wr_index = (rbrp->rbb_max - 1);
3582 
3583 	rbrp->rbr_rd_index = 0;
3584 
3585 	rbrp->rbr_consumed = 0;
3586 	rbrp->rbr_use_bcopy = B_TRUE;
3587 	rbrp->rbr_bufsize_type = RCR_PKTBUFSZ_0;
3588 	/*
3589 	 * Do bcopy on packets greater than bcopy size once
3590 	 * the lo threshold is reached.
3591 	 * This lo threshold should be less than the hi threshold.
3592 	 *
3593 	 * Do bcopy on every packet once the hi threshold is reached.
3594 	 */
3595 	if (nxge_rx_threshold_lo >= nxge_rx_threshold_hi) {
3596 		/* default it to use hi */
3597 		nxge_rx_threshold_lo = nxge_rx_threshold_hi;
3598 	}
3599 
3600 	if (nxge_rx_buf_size_type > NXGE_RBR_TYPE2) {
3601 		nxge_rx_buf_size_type = NXGE_RBR_TYPE2;
3602 	}
3603 	rbrp->rbr_bufsize_type = nxge_rx_buf_size_type;
3604 
3605 	switch (nxge_rx_threshold_hi) {
3606 	default:
3607 	case	NXGE_RX_COPY_NONE:
3608 		/* Do not do bcopy at all */
3609 		rbrp->rbr_use_bcopy = B_FALSE;
3610 		rbrp->rbr_threshold_hi = rbrp->rbb_max;
3611 		break;
3612 
3613 	case NXGE_RX_COPY_1:
3614 	case NXGE_RX_COPY_2:
3615 	case NXGE_RX_COPY_3:
3616 	case NXGE_RX_COPY_4:
3617 	case NXGE_RX_COPY_5:
3618 	case NXGE_RX_COPY_6:
3619 	case NXGE_RX_COPY_7:
3620 		rbrp->rbr_threshold_hi =
3621 		    rbrp->rbb_max *
3622 		    (nxge_rx_threshold_hi)/NXGE_RX_BCOPY_SCALE;
3623 		break;
3624 
3625 	case NXGE_RX_COPY_ALL:
3626 		rbrp->rbr_threshold_hi = 0;
3627 		break;
3628 	}
3629 
3630 	switch (nxge_rx_threshold_lo) {
3631 	default:
3632 	case	NXGE_RX_COPY_NONE:
3633 		/* Do not do bcopy at all */
3634 		if (rbrp->rbr_use_bcopy) {
3635 			rbrp->rbr_use_bcopy = B_FALSE;
3636 		}
3637 		rbrp->rbr_threshold_lo = rbrp->rbb_max;
3638 		break;
3639 
3640 	case NXGE_RX_COPY_1:
3641 	case NXGE_RX_COPY_2:
3642 	case NXGE_RX_COPY_3:
3643 	case NXGE_RX_COPY_4:
3644 	case NXGE_RX_COPY_5:
3645 	case NXGE_RX_COPY_6:
3646 	case NXGE_RX_COPY_7:
3647 		rbrp->rbr_threshold_lo =
3648 		    rbrp->rbb_max *
3649 		    (nxge_rx_threshold_lo)/NXGE_RX_BCOPY_SCALE;
3650 		break;
3651 
3652 	case NXGE_RX_COPY_ALL:
3653 		rbrp->rbr_threshold_lo = 0;
3654 		break;
3655 	}
3656 
3657 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
3658 	    "nxge_map_rxdma_channel_cfg_ring: channel %d "
3659 	    "rbb_max %d "
3660 	    "rbrp->rbr_bufsize_type %d "
3661 	    "rbb_threshold_hi %d "
3662 	    "rbb_threshold_lo %d",
3663 	    dma_channel,
3664 	    rbrp->rbb_max,
3665 	    rbrp->rbr_bufsize_type,
3666 	    rbrp->rbr_threshold_hi,
3667 	    rbrp->rbr_threshold_lo));
3668 
3669 	rbrp->page_valid.value = 0;
3670 	rbrp->page_mask_1.value = rbrp->page_mask_2.value = 0;
3671 	rbrp->page_value_1.value = rbrp->page_value_2.value = 0;
3672 	rbrp->page_reloc_1.value = rbrp->page_reloc_2.value = 0;
3673 	rbrp->page_hdl.value = 0;
3674 
3675 	rbrp->page_valid.bits.ldw.page0 = 1;
3676 	rbrp->page_valid.bits.ldw.page1 = 1;
3677 
3678 	/* Map in the receive completion ring */
3679 	rcrp = (p_rx_rcr_ring_t)
3680 	    KMEM_ZALLOC(sizeof (rx_rcr_ring_t), KM_SLEEP);
3681 	rcrp->rdc = dma_channel;
3682 
3683 	nxge_port_rcr_size = nxgep->nxge_port_rcr_size;
3684 	rcrp->comp_size = nxge_port_rcr_size;
3685 	rcrp->comp_wrap_mask = nxge_port_rcr_size - 1;
3686 
3687 	rcrp->max_receive_pkts = nxge_max_rx_pkts;
3688 
3689 	dmap = (p_nxge_dma_common_t)&rcrp->rcr_desc;
3690 	nxge_setup_dma_common(dmap, cntl_dmap, rcrp->comp_size,
3691 	    sizeof (rcr_entry_t));
3692 	rcrp->comp_rd_index = 0;
3693 	rcrp->comp_wt_index = 0;
3694 	rcrp->rcr_desc_rd_head_p = rcrp->rcr_desc_first_p =
3695 	    (p_rcr_entry_t)DMA_COMMON_VPTR(rcrp->rcr_desc);
3696 #if defined(__i386)
3697 	rcrp->rcr_desc_rd_head_pp = rcrp->rcr_desc_first_pp =
3698 	    (p_rcr_entry_t)(uint32_t)DMA_COMMON_IOADDR(rcrp->rcr_desc);
3699 #else
3700 	rcrp->rcr_desc_rd_head_pp = rcrp->rcr_desc_first_pp =
3701 	    (p_rcr_entry_t)DMA_COMMON_IOADDR(rcrp->rcr_desc);
3702 #endif
3703 
3704 	rcrp->rcr_desc_last_p = rcrp->rcr_desc_rd_head_p +
3705 	    (nxge_port_rcr_size - 1);
3706 	rcrp->rcr_desc_last_pp = rcrp->rcr_desc_rd_head_pp +
3707 	    (nxge_port_rcr_size - 1);
3708 
3709 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3710 	    "==> nxge_map_rxdma_channel_cfg_ring: "
3711 	    "channel %d "
3712 	    "rbr_vaddrp $%p "
3713 	    "rcr_desc_rd_head_p $%p "
3714 	    "rcr_desc_rd_head_pp $%p "
3715 	    "rcr_desc_rd_last_p $%p "
3716 	    "rcr_desc_rd_last_pp $%p ",
3717 	    dma_channel,
3718 	    rbr_vaddrp,
3719 	    rcrp->rcr_desc_rd_head_p,
3720 	    rcrp->rcr_desc_rd_head_pp,
3721 	    rcrp->rcr_desc_last_p,
3722 	    rcrp->rcr_desc_last_pp));
3723 
3724 	/*
3725 	 * Zero out buffer block ring descriptors.
3726 	 */
3727 	bzero((caddr_t)dmap->kaddrp, dmap->alength);
3728 
3729 	rcrp->intr_timeout = (nxgep->intr_timeout <
3730 	    NXGE_RDC_RCR_TIMEOUT_MIN) ? NXGE_RDC_RCR_TIMEOUT_MIN :
3731 	    nxgep->intr_timeout;
3732 
3733 	rcrp->intr_threshold = (nxgep->intr_threshold <
3734 	    NXGE_RDC_RCR_THRESHOLD_MIN) ? NXGE_RDC_RCR_THRESHOLD_MIN :
3735 	    nxgep->intr_threshold;
3736 
3737 	rcrp->full_hdr_flag = B_FALSE;
3738 	rcrp->sw_priv_hdr_len = 0;
3739 
3740 	cfga_p = &(rcrp->rcr_cfga);
3741 	cfgb_p = &(rcrp->rcr_cfgb);
3742 	cfga_p->value = 0;
3743 	cfgb_p->value = 0;
3744 	rcrp->rcr_addr = dmap->dma_cookie.dmac_laddress;
3745 	cfga_p->value = (rcrp->rcr_addr &
3746 	    (RCRCFIG_A_STADDR_MASK |
3747 	    RCRCFIG_A_STADDR_BASE_MASK));
3748 
3749 	rcfga_p->value |= ((uint64_t)rcrp->comp_size <<
3750 	    RCRCFIG_A_LEN_SHIF);
3751 
3752 	/*
3753 	 * Timeout should be set based on the system clock divider.
3754 	 * A timeout value of 1 assumes that the
3755 	 * granularity (1000) is 3 microseconds running at 300MHz.
3756 	 */
3757 	cfgb_p->bits.ldw.pthres = rcrp->intr_threshold;
3758 	cfgb_p->bits.ldw.timeout = rcrp->intr_timeout;
3759 	cfgb_p->bits.ldw.entout = 1;
3760 
3761 	/* Map in the mailbox */
3762 	mboxp = (p_rx_mbox_t)
3763 	    KMEM_ZALLOC(sizeof (rx_mbox_t), KM_SLEEP);
3764 	dmap = (p_nxge_dma_common_t)&mboxp->rx_mbox;
3765 	nxge_setup_dma_common(dmap, cntl_dmap, 1, sizeof (rxdma_mailbox_t));
3766 	cfig1_p = (p_rxdma_cfig1_t)&mboxp->rx_cfg1;
3767 	cfig2_p = (p_rxdma_cfig2_t)&mboxp->rx_cfg2;
3768 	cfig1_p->value = cfig2_p->value = 0;
3769 
3770 	mboxp->mbox_addr = dmap->dma_cookie.dmac_laddress;
3771 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3772 	    "==> nxge_map_rxdma_channel_cfg_ring: "
3773 	    "channel %d cfg1 0x%016llx cfig2 0x%016llx cookie 0x%016llx",
3774 	    dma_channel, cfig1_p->value, cfig2_p->value,
3775 	    mboxp->mbox_addr));
3776 
3777 	dmaaddrp = (uint32_t)(dmap->dma_cookie.dmac_laddress >> 32
3778 	    & 0xfff);
3779 	cfig1_p->bits.ldw.mbaddr_h = dmaaddrp;
3780 
3781 
3782 	dmaaddrp = (uint32_t)(dmap->dma_cookie.dmac_laddress & 0xffffffff);
3783 	dmaaddrp = (uint32_t)(dmap->dma_cookie.dmac_laddress &
3784 	    RXDMA_CFIG2_MBADDR_L_MASK);
3785 
3786 	cfig2_p->bits.ldw.mbaddr = (dmaaddrp >> RXDMA_CFIG2_MBADDR_L_SHIFT);
3787 
3788 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3789 	    "==> nxge_map_rxdma_channel_cfg_ring: "
3790 	    "channel %d damaddrp $%p "
3791 	    "cfg1 0x%016llx cfig2 0x%016llx",
3792 	    dma_channel, dmaaddrp,
3793 	    cfig1_p->value, cfig2_p->value));
3794 
3795 	cfig2_p->bits.ldw.full_hdr = rcrp->full_hdr_flag;
3796 	cfig2_p->bits.ldw.offset = rcrp->sw_priv_hdr_len;
3797 
3798 	rbrp->rx_rcr_p = rcrp;
3799 	rcrp->rx_rbr_p = rbrp;
3800 	*rcr_p = rcrp;
3801 	*rx_mbox_p = mboxp;
3802 
3803 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3804 	    "<== nxge_map_rxdma_channel_cfg_ring status 0x%08x", status));
3805 
3806 	return (status);
3807 }
3808 
3809 /*ARGSUSED*/
3810 static void
3811 nxge_unmap_rxdma_channel_cfg_ring(p_nxge_t nxgep,
3812     p_rx_rcr_ring_t rcr_p, p_rx_mbox_t rx_mbox_p)
3813 {
3814 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3815 	    "==> nxge_unmap_rxdma_channel_cfg_ring: channel %d",
3816 	    rcr_p->rdc));
3817 
3818 	KMEM_FREE(rcr_p, sizeof (rx_rcr_ring_t));
3819 	KMEM_FREE(rx_mbox_p, sizeof (rx_mbox_t));
3820 
3821 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3822 	    "<== nxge_unmap_rxdma_channel_cfg_ring"));
3823 }
3824 
3825 static nxge_status_t
3826 nxge_map_rxdma_channel_buf_ring(p_nxge_t nxgep, uint16_t channel,
3827     p_nxge_dma_common_t *dma_buf_p,
3828     p_rx_rbr_ring_t *rbr_p, uint32_t num_chunks)
3829 {
3830 	p_rx_rbr_ring_t 	rbrp;
3831 	p_nxge_dma_common_t 	dma_bufp, tmp_bufp;
3832 	p_rx_msg_t 		*rx_msg_ring;
3833 	p_rx_msg_t 		rx_msg_p;
3834 	p_mblk_t 		mblk_p;
3835 
3836 	rxring_info_t *ring_info;
3837 	nxge_status_t		status = NXGE_OK;
3838 	int			i, j, index;
3839 	uint32_t		size, bsize, nblocks, nmsgs;
3840 
3841 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3842 	    "==> nxge_map_rxdma_channel_buf_ring: channel %d",
3843 	    channel));
3844 
3845 	dma_bufp = tmp_bufp = *dma_buf_p;
3846 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3847 	    " nxge_map_rxdma_channel_buf_ring: channel %d to map %d "
3848 	    "chunks bufp 0x%016llx",
3849 	    channel, num_chunks, dma_bufp));
3850 
3851 	nmsgs = 0;
3852 	for (i = 0; i < num_chunks; i++, tmp_bufp++) {
3853 		NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3854 		    "==> nxge_map_rxdma_channel_buf_ring: channel %d "
3855 		    "bufp 0x%016llx nblocks %d nmsgs %d",
3856 		    channel, tmp_bufp, tmp_bufp->nblocks, nmsgs));
3857 		nmsgs += tmp_bufp->nblocks;
3858 	}
3859 	if (!nmsgs) {
3860 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3861 		    "<== nxge_map_rxdma_channel_buf_ring: channel %d "
3862 		    "no msg blocks",
3863 		    channel));
3864 		status = NXGE_ERROR;
3865 		goto nxge_map_rxdma_channel_buf_ring_exit;
3866 	}
3867 
3868 	rbrp = (p_rx_rbr_ring_t)KMEM_ZALLOC(sizeof (*rbrp), KM_SLEEP);
3869 
3870 	size = nmsgs * sizeof (p_rx_msg_t);
3871 	rx_msg_ring = KMEM_ZALLOC(size, KM_SLEEP);
3872 	ring_info = (rxring_info_t *)KMEM_ZALLOC(sizeof (rxring_info_t),
3873 	    KM_SLEEP);
3874 
3875 	MUTEX_INIT(&rbrp->lock, NULL, MUTEX_DRIVER,
3876 	    (void *)nxgep->interrupt_cookie);
3877 	MUTEX_INIT(&rbrp->post_lock, NULL, MUTEX_DRIVER,
3878 	    (void *)nxgep->interrupt_cookie);
3879 	rbrp->rdc = channel;
3880 	rbrp->num_blocks = num_chunks;
3881 	rbrp->tnblocks = nmsgs;
3882 	rbrp->rbb_max = nmsgs;
3883 	rbrp->rbr_max_size = nmsgs;
3884 	rbrp->rbr_wrap_mask = (rbrp->rbb_max - 1);
3885 
3886 	/*
3887 	 * Buffer sizes suggested by NIU architect.
3888 	 * 256, 512 and 2K.
3889 	 */
3890 
3891 	rbrp->pkt_buf_size0 = RBR_BUFSZ0_256B;
3892 	rbrp->pkt_buf_size0_bytes = RBR_BUFSZ0_256_BYTES;
3893 	rbrp->npi_pkt_buf_size0 = SIZE_256B;
3894 
3895 	rbrp->pkt_buf_size1 = RBR_BUFSZ1_1K;
3896 	rbrp->pkt_buf_size1_bytes = RBR_BUFSZ1_1K_BYTES;
3897 	rbrp->npi_pkt_buf_size1 = SIZE_1KB;
3898 
3899 	rbrp->block_size = nxgep->rx_default_block_size;
3900 
3901 	if (!nxge_jumbo_enable && !nxgep->param_arr[param_accept_jumbo].value) {
3902 		rbrp->pkt_buf_size2 = RBR_BUFSZ2_2K;
3903 		rbrp->pkt_buf_size2_bytes = RBR_BUFSZ2_2K_BYTES;
3904 		rbrp->npi_pkt_buf_size2 = SIZE_2KB;
3905 	} else {
3906 		if (rbrp->block_size >= 0x2000) {
3907 			rbrp->pkt_buf_size2 = RBR_BUFSZ2_8K;
3908 			rbrp->pkt_buf_size2_bytes = RBR_BUFSZ2_8K_BYTES;
3909 			rbrp->npi_pkt_buf_size2 = SIZE_8KB;
3910 		} else {
3911 			rbrp->pkt_buf_size2 = RBR_BUFSZ2_4K;
3912 			rbrp->pkt_buf_size2_bytes = RBR_BUFSZ2_4K_BYTES;
3913 			rbrp->npi_pkt_buf_size2 = SIZE_4KB;
3914 		}
3915 	}
3916 
3917 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3918 	    "==> nxge_map_rxdma_channel_buf_ring: channel %d "
3919 	    "actual rbr max %d rbb_max %d nmsgs %d "
3920 	    "rbrp->block_size %d default_block_size %d "
3921 	    "(config nxge_rbr_size %d nxge_rbr_spare_size %d)",
3922 	    channel, rbrp->rbr_max_size, rbrp->rbb_max, nmsgs,
3923 	    rbrp->block_size, nxgep->rx_default_block_size,
3924 	    nxge_rbr_size, nxge_rbr_spare_size));
3925 
3926 	/* Map in buffers from the buffer pool.  */
3927 	index = 0;
3928 	for (i = 0; i < rbrp->num_blocks; i++, dma_bufp++) {
3929 		bsize = dma_bufp->block_size;
3930 		nblocks = dma_bufp->nblocks;
3931 #if defined(__i386)
3932 		ring_info->buffer[i].dvma_addr = (uint32_t)dma_bufp->ioaddr_pp;
3933 #else
3934 		ring_info->buffer[i].dvma_addr = (uint64_t)dma_bufp->ioaddr_pp;
3935 #endif
3936 		ring_info->buffer[i].buf_index = i;
3937 		ring_info->buffer[i].buf_size = dma_bufp->alength;
3938 		ring_info->buffer[i].start_index = index;
3939 #if defined(__i386)
3940 		ring_info->buffer[i].kaddr = (uint32_t)dma_bufp->kaddrp;
3941 #else
3942 		ring_info->buffer[i].kaddr = (uint64_t)dma_bufp->kaddrp;
3943 #endif
3944 
3945 		NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3946 		    " nxge_map_rxdma_channel_buf_ring: map channel %d "
3947 		    "chunk %d"
3948 		    " nblocks %d chunk_size %x block_size 0x%x "
3949 		    "dma_bufp $%p", channel, i,
3950 		    dma_bufp->nblocks, ring_info->buffer[i].buf_size, bsize,
3951 		    dma_bufp));
3952 
3953 		for (j = 0; j < nblocks; j++) {
3954 			if ((rx_msg_p = nxge_allocb(bsize, BPRI_LO,
3955 			    dma_bufp)) == NULL) {
3956 				NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3957 				    "allocb failed (index %d i %d j %d)",
3958 				    index, i, j));
3959 				goto nxge_map_rxdma_channel_buf_ring_fail1;
3960 			}
3961 			rx_msg_ring[index] = rx_msg_p;
3962 			rx_msg_p->block_index = index;
3963 			rx_msg_p->shifted_addr = (uint32_t)
3964 			    ((rx_msg_p->buf_dma.dma_cookie.dmac_laddress >>
3965 			    RBR_BKADDR_SHIFT));
3966 
3967 			NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3968 			    "index %d j %d rx_msg_p $%p mblk %p",
3969 			    index, j, rx_msg_p, rx_msg_p->rx_mblk_p));
3970 
3971 			mblk_p = rx_msg_p->rx_mblk_p;
3972 			mblk_p->b_wptr = mblk_p->b_rptr + bsize;
3973 
3974 			rbrp->rbr_ref_cnt++;
3975 			index++;
3976 			rx_msg_p->buf_dma.dma_channel = channel;
3977 		}
3978 
3979 		rbrp->rbr_alloc_type = DDI_MEM_ALLOC;
3980 		if (dma_bufp->contig_alloc_type) {
3981 			rbrp->rbr_alloc_type = CONTIG_MEM_ALLOC;
3982 		}
3983 
3984 		if (dma_bufp->kmem_alloc_type) {
3985 			rbrp->rbr_alloc_type = KMEM_ALLOC;
3986 		}
3987 
3988 		NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3989 		    " nxge_map_rxdma_channel_buf_ring: map channel %d "
3990 		    "chunk %d"
3991 		    " nblocks %d chunk_size %x block_size 0x%x "
3992 		    "dma_bufp $%p",
3993 		    channel, i,
3994 		    dma_bufp->nblocks, ring_info->buffer[i].buf_size, bsize,
3995 		    dma_bufp));
3996 	}
3997 	if (i < rbrp->num_blocks) {
3998 		goto nxge_map_rxdma_channel_buf_ring_fail1;
3999 	}
4000 
4001 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4002 	    "nxge_map_rxdma_channel_buf_ring: done buf init "
4003 	    "channel %d msg block entries %d",
4004 	    channel, index));
4005 	ring_info->block_size_mask = bsize - 1;
4006 	rbrp->rx_msg_ring = rx_msg_ring;
4007 	rbrp->dma_bufp = dma_buf_p;
4008 	rbrp->ring_info = ring_info;
4009 
4010 	status = nxge_rxbuf_index_info_init(nxgep, rbrp);
4011 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4012 	    " nxge_map_rxdma_channel_buf_ring: "
4013 	    "channel %d done buf info init", channel));
4014 
4015 	/*
4016 	 * Finally, permit nxge_freeb() to call nxge_post_page().
4017 	 */
4018 	rbrp->rbr_state = RBR_POSTING;
4019 
4020 	*rbr_p = rbrp;
4021 	goto nxge_map_rxdma_channel_buf_ring_exit;
4022 
4023 nxge_map_rxdma_channel_buf_ring_fail1:
4024 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4025 	    " nxge_map_rxdma_channel_buf_ring: failed channel (0x%x)",
4026 	    channel, status));
4027 
4028 	index--;
4029 	for (; index >= 0; index--) {
4030 		rx_msg_p = rx_msg_ring[index];
4031 		if (rx_msg_p != NULL) {
4032 			freeb(rx_msg_p->rx_mblk_p);
4033 			rx_msg_ring[index] = NULL;
4034 		}
4035 	}
4036 nxge_map_rxdma_channel_buf_ring_fail:
4037 	MUTEX_DESTROY(&rbrp->post_lock);
4038 	MUTEX_DESTROY(&rbrp->lock);
4039 	KMEM_FREE(ring_info, sizeof (rxring_info_t));
4040 	KMEM_FREE(rx_msg_ring, size);
4041 	KMEM_FREE(rbrp, sizeof (rx_rbr_ring_t));
4042 
4043 	status = NXGE_ERROR;
4044 
4045 nxge_map_rxdma_channel_buf_ring_exit:
4046 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4047 	    "<== nxge_map_rxdma_channel_buf_ring status 0x%08x", status));
4048 
4049 	return (status);
4050 }
4051 
4052 /*ARGSUSED*/
4053 static void
4054 nxge_unmap_rxdma_channel_buf_ring(p_nxge_t nxgep,
4055     p_rx_rbr_ring_t rbr_p)
4056 {
4057 	p_rx_msg_t 		*rx_msg_ring;
4058 	p_rx_msg_t 		rx_msg_p;
4059 	rxring_info_t 		*ring_info;
4060 	int			i;
4061 	uint32_t		size;
4062 #ifdef	NXGE_DEBUG
4063 	int			num_chunks;
4064 #endif
4065 
4066 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4067 	    "==> nxge_unmap_rxdma_channel_buf_ring"));
4068 	if (rbr_p == NULL) {
4069 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
4070 		    "<== nxge_unmap_rxdma_channel_buf_ring: NULL rbrp"));
4071 		return;
4072 	}
4073 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4074 	    "==> nxge_unmap_rxdma_channel_buf_ring: channel %d",
4075 	    rbr_p->rdc));
4076 
4077 	rx_msg_ring = rbr_p->rx_msg_ring;
4078 	ring_info = rbr_p->ring_info;
4079 
4080 	if (rx_msg_ring == NULL || ring_info == NULL) {
4081 		NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4082 		    "<== nxge_unmap_rxdma_channel_buf_ring: "
4083 		    "rx_msg_ring $%p ring_info $%p",
4084 		    rx_msg_p, ring_info));
4085 		return;
4086 	}
4087 
4088 #ifdef	NXGE_DEBUG
4089 	num_chunks = rbr_p->num_blocks;
4090 #endif
4091 	size = rbr_p->tnblocks * sizeof (p_rx_msg_t);
4092 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4093 	    " nxge_unmap_rxdma_channel_buf_ring: channel %d chunks %d "
4094 	    "tnblocks %d (max %d) size ptrs %d ",
4095 	    rbr_p->rdc, num_chunks,
4096 	    rbr_p->tnblocks, rbr_p->rbr_max_size, size));
4097 
4098 	for (i = 0; i < rbr_p->tnblocks; i++) {
4099 		rx_msg_p = rx_msg_ring[i];
4100 		NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4101 		    " nxge_unmap_rxdma_channel_buf_ring: "
4102 		    "rx_msg_p $%p",
4103 		    rx_msg_p));
4104 		if (rx_msg_p != NULL) {
4105 			freeb(rx_msg_p->rx_mblk_p);
4106 			rx_msg_ring[i] = NULL;
4107 		}
4108 	}
4109 
4110 	/*
4111 	 * We no longer may use the mutex <post_lock>. By setting
4112 	 * <rbr_state> to anything but POSTING, we prevent
4113 	 * nxge_post_page() from accessing a dead mutex.
4114 	 */
4115 	rbr_p->rbr_state = RBR_UNMAPPING;
4116 	MUTEX_DESTROY(&rbr_p->post_lock);
4117 
4118 	MUTEX_DESTROY(&rbr_p->lock);
4119 
4120 	if (rbr_p->rbr_ref_cnt == 0) {
4121 		/*
4122 		 * This is the normal state of affairs.
4123 		 * Need to free the following buffers:
4124 		 *  - data buffers
4125 		 *  - rx_msg ring
4126 		 *  - ring_info
4127 		 *  - rbr ring
4128 		 */
4129 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
4130 		    "unmap_rxdma_buf_ring: No outstanding - freeing "));
4131 		nxge_rxdma_databuf_free(rbr_p);
4132 		KMEM_FREE(ring_info, sizeof (rxring_info_t));
4133 		KMEM_FREE(rx_msg_ring, size);
4134 		KMEM_FREE(rbr_p, sizeof (*rbr_p));
4135 	} else {
4136 		/*
4137 		 * Some of our buffers are still being used.
4138 		 * Therefore, tell nxge_freeb() this ring is
4139 		 * unmapped, so it may free <rbr_p> for us.
4140 		 */
4141 		rbr_p->rbr_state = RBR_UNMAPPED;
4142 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4143 		    "unmap_rxdma_buf_ring: %d %s outstanding.",
4144 		    rbr_p->rbr_ref_cnt,
4145 		    rbr_p->rbr_ref_cnt == 1 ? "msg" : "msgs"));
4146 	}
4147 
4148 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4149 	    "<== nxge_unmap_rxdma_channel_buf_ring"));
4150 }
4151 
4152 /*
4153  * nxge_rxdma_hw_start_common
4154  *
4155  * Arguments:
4156  * 	nxgep
4157  *
4158  * Notes:
4159  *
4160  * NPI/NXGE function calls:
4161  *	nxge_init_fzc_rx_common();
4162  *	nxge_init_fzc_rxdma_port();
4163  *
4164  * Registers accessed:
4165  *
4166  * Context:
4167  *	Service domain
4168  */
4169 static nxge_status_t
4170 nxge_rxdma_hw_start_common(p_nxge_t nxgep)
4171 {
4172 	nxge_status_t		status = NXGE_OK;
4173 
4174 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_rxdma_hw_start_common"));
4175 
4176 	/*
4177 	 * Load the sharable parameters by writing to the
4178 	 * function zero control registers. These FZC registers
4179 	 * should be initialized only once for the entire chip.
4180 	 */
4181 	(void) nxge_init_fzc_rx_common(nxgep);
4182 
4183 	/*
4184 	 * Initialize the RXDMA port specific FZC control configurations.
4185 	 * These FZC registers are pertaining to each port.
4186 	 */
4187 	(void) nxge_init_fzc_rxdma_port(nxgep);
4188 
4189 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_rxdma_hw_start_common"));
4190 
4191 	return (status);
4192 }
4193 
4194 static nxge_status_t
4195 nxge_rxdma_hw_start(p_nxge_t nxgep, int channel)
4196 {
4197 	int			i, ndmas;
4198 	p_rx_rbr_rings_t 	rx_rbr_rings;
4199 	p_rx_rbr_ring_t		*rbr_rings;
4200 	p_rx_rcr_rings_t 	rx_rcr_rings;
4201 	p_rx_rcr_ring_t		*rcr_rings;
4202 	p_rx_mbox_areas_t 	rx_mbox_areas_p;
4203 	p_rx_mbox_t		*rx_mbox_p;
4204 	nxge_status_t		status = NXGE_OK;
4205 
4206 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_rxdma_hw_start"));
4207 
4208 	rx_rbr_rings = nxgep->rx_rbr_rings;
4209 	rx_rcr_rings = nxgep->rx_rcr_rings;
4210 	if (rx_rbr_rings == NULL || rx_rcr_rings == NULL) {
4211 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
4212 		    "<== nxge_rxdma_hw_start: NULL ring pointers"));
4213 		return (NXGE_ERROR);
4214 	}
4215 	ndmas = rx_rbr_rings->ndmas;
4216 	if (ndmas == 0) {
4217 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
4218 		    "<== nxge_rxdma_hw_start: no dma channel allocated"));
4219 		return (NXGE_ERROR);
4220 	}
4221 
4222 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4223 	    "==> nxge_rxdma_hw_start (ndmas %d)", ndmas));
4224 
4225 	rbr_rings = rx_rbr_rings->rbr_rings;
4226 	rcr_rings = rx_rcr_rings->rcr_rings;
4227 	rx_mbox_areas_p = nxgep->rx_mbox_areas_p;
4228 	if (rx_mbox_areas_p) {
4229 		rx_mbox_p = rx_mbox_areas_p->rxmbox_areas;
4230 	}
4231 
4232 	i = channel;
4233 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4234 	    "==> nxge_rxdma_hw_start (ndmas %d) channel %d",
4235 	    ndmas, channel));
4236 	status = nxge_rxdma_start_channel(nxgep, channel,
4237 	    (p_rx_rbr_ring_t)rbr_rings[i],
4238 	    (p_rx_rcr_ring_t)rcr_rings[i],
4239 	    (p_rx_mbox_t)rx_mbox_p[i]);
4240 	if (status != NXGE_OK) {
4241 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4242 		    "==> nxge_rxdma_hw_start: disable "
4243 		    "(status 0x%x channel %d)", status, channel));
4244 		return (status);
4245 	}
4246 
4247 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_rxdma_hw_start: "
4248 	    "rx_rbr_rings 0x%016llx rings 0x%016llx",
4249 	    rx_rbr_rings, rx_rcr_rings));
4250 
4251 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4252 	    "==> nxge_rxdma_hw_start: (status 0x%x)", status));
4253 
4254 	return (status);
4255 }
4256 
4257 static void
4258 nxge_rxdma_hw_stop(p_nxge_t nxgep, int channel)
4259 {
4260 	p_rx_rbr_rings_t 	rx_rbr_rings;
4261 	p_rx_rcr_rings_t 	rx_rcr_rings;
4262 
4263 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_rxdma_hw_stop"));
4264 
4265 	rx_rbr_rings = nxgep->rx_rbr_rings;
4266 	rx_rcr_rings = nxgep->rx_rcr_rings;
4267 	if (rx_rbr_rings == NULL || rx_rcr_rings == NULL) {
4268 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
4269 		    "<== nxge_rxdma_hw_stop: NULL ring pointers"));
4270 		return;
4271 	}
4272 
4273 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4274 	    "==> nxge_rxdma_hw_stop(channel %d)",
4275 	    channel));
4276 	(void) nxge_rxdma_stop_channel(nxgep, channel);
4277 
4278 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_rxdma_hw_stop: "
4279 	    "rx_rbr_rings 0x%016llx rings 0x%016llx",
4280 	    rx_rbr_rings, rx_rcr_rings));
4281 
4282 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "<== nxge_rxdma_hw_stop"));
4283 }
4284 
4285 
4286 static nxge_status_t
4287 nxge_rxdma_start_channel(p_nxge_t nxgep, uint16_t channel,
4288     p_rx_rbr_ring_t rbr_p, p_rx_rcr_ring_t rcr_p, p_rx_mbox_t mbox_p)
4289 
4290 {
4291 	npi_handle_t		handle;
4292 	npi_status_t		rs = NPI_SUCCESS;
4293 	rx_dma_ctl_stat_t	cs;
4294 	rx_dma_ent_msk_t	ent_mask;
4295 	nxge_status_t		status = NXGE_OK;
4296 
4297 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_rxdma_start_channel"));
4298 
4299 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
4300 
4301 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "nxge_rxdma_start_channel: "
4302 		"npi handle addr $%p acc $%p",
4303 		nxgep->npi_handle.regp, nxgep->npi_handle.regh));
4304 
4305 	/* Reset RXDMA channel, but not if you're a guest. */
4306 	if (!isLDOMguest(nxgep)) {
4307 		rs = npi_rxdma_cfg_rdc_reset(handle, channel);
4308 		if (rs != NPI_SUCCESS) {
4309 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4310 			    "==> nxge_init_fzc_rdc: "
4311 			    "npi_rxdma_cfg_rdc_reset(%d) returned 0x%08x",
4312 			    channel, rs));
4313 			return (NXGE_ERROR | rs);
4314 		}
4315 
4316 		NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4317 		    "==> nxge_rxdma_start_channel: reset done: channel %d",
4318 		    channel));
4319 	}
4320 
4321 #if defined(sun4v) && defined(NIU_LP_WORKAROUND)
4322 	if (isLDOMguest(nxgep))
4323 		(void) nxge_rdc_lp_conf(nxgep, channel);
4324 #endif
4325 
4326 	/*
4327 	 * Initialize the RXDMA channel specific FZC control
4328 	 * configurations. These FZC registers are pertaining
4329 	 * to each RX channel (logical pages).
4330 	 */
4331 	if (!isLDOMguest(nxgep)) {
4332 		status = nxge_init_fzc_rxdma_channel(nxgep, channel);
4333 		if (status != NXGE_OK) {
4334 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4335 				"==> nxge_rxdma_start_channel: "
4336 				"init fzc rxdma failed (0x%08x channel %d)",
4337 				status, channel));
4338 			return (status);
4339 		}
4340 
4341 		NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4342 			"==> nxge_rxdma_start_channel: fzc done"));
4343 	}
4344 
4345 	/* Set up the interrupt event masks. */
4346 	ent_mask.value = 0;
4347 	ent_mask.value |= RX_DMA_ENT_MSK_RBREMPTY_MASK;
4348 	rs = npi_rxdma_event_mask(handle, OP_SET, channel,
4349 	    &ent_mask);
4350 	if (rs != NPI_SUCCESS) {
4351 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4352 			"==> nxge_rxdma_start_channel: "
4353 			"init rxdma event masks failed "
4354 			"(0x%08x channel %d)",
4355 			status, channel));
4356 		return (NXGE_ERROR | rs);
4357 	}
4358 
4359 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4360 		"==> nxge_rxdma_start_channel: "
4361 		"event done: channel %d (mask 0x%016llx)",
4362 		channel, ent_mask.value));
4363 
4364 	/* Initialize the receive DMA control and status register */
4365 	cs.value = 0;
4366 	cs.bits.hdw.mex = 1;
4367 	cs.bits.hdw.rcrthres = 1;
4368 	cs.bits.hdw.rcrto = 1;
4369 	cs.bits.hdw.rbr_empty = 1;
4370 	status = nxge_init_rxdma_channel_cntl_stat(nxgep, channel, &cs);
4371 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_rxdma_start_channel: "
4372 		"channel %d rx_dma_cntl_stat 0x%0016llx", channel, cs.value));
4373 	if (status != NXGE_OK) {
4374 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4375 			"==> nxge_rxdma_start_channel: "
4376 			"init rxdma control register failed (0x%08x channel %d",
4377 			status, channel));
4378 		return (status);
4379 	}
4380 
4381 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_rxdma_start_channel: "
4382 		"control done - channel %d cs 0x%016llx", channel, cs.value));
4383 
4384 	/*
4385 	 * Load RXDMA descriptors, buffers, mailbox,
4386 	 * initialise the receive DMA channels and
4387 	 * enable each DMA channel.
4388 	 */
4389 	status = nxge_enable_rxdma_channel(nxgep,
4390 	    channel, rbr_p, rcr_p, mbox_p);
4391 
4392 	if (status != NXGE_OK) {
4393 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4394 		    " nxge_rxdma_start_channel: "
4395 		    " enable rxdma failed (0x%08x channel %d)",
4396 		    status, channel));
4397 		return (status);
4398 	}
4399 
4400 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4401 	    "==> nxge_rxdma_start_channel: enabled channel %d"));
4402 
4403 	if (isLDOMguest(nxgep)) {
4404 		/* Add interrupt handler for this channel. */
4405 		if (nxge_hio_intr_add(nxgep, VP_BOUND_RX, channel)
4406 		    != NXGE_OK) {
4407 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4408 			    " nxge_rxdma_start_channel: "
4409 			    " nxge_hio_intr_add failed (0x%08x channel %d)",
4410 		    status, channel));
4411 		}
4412 	}
4413 
4414 	ent_mask.value = 0;
4415 	ent_mask.value |= (RX_DMA_ENT_MSK_WRED_DROP_MASK |
4416 				RX_DMA_ENT_MSK_PTDROP_PKT_MASK);
4417 	rs = npi_rxdma_event_mask(handle, OP_SET, channel,
4418 			&ent_mask);
4419 	if (rs != NPI_SUCCESS) {
4420 		NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4421 			"==> nxge_rxdma_start_channel: "
4422 			"init rxdma event masks failed (0x%08x channel %d)",
4423 			status, channel));
4424 		return (NXGE_ERROR | rs);
4425 	}
4426 
4427 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_rxdma_start_channel: "
4428 		"control done - channel %d cs 0x%016llx", channel, cs.value));
4429 
4430 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "<== nxge_rxdma_start_channel"));
4431 
4432 	return (NXGE_OK);
4433 }
4434 
4435 static nxge_status_t
4436 nxge_rxdma_stop_channel(p_nxge_t nxgep, uint16_t channel)
4437 {
4438 	npi_handle_t		handle;
4439 	npi_status_t		rs = NPI_SUCCESS;
4440 	rx_dma_ctl_stat_t	cs;
4441 	rx_dma_ent_msk_t	ent_mask;
4442 	nxge_status_t		status = NXGE_OK;
4443 
4444 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "==> nxge_rxdma_stop_channel"));
4445 
4446 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
4447 
4448 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "nxge_rxdma_stop_channel: "
4449 	    "npi handle addr $%p acc $%p",
4450 	    nxgep->npi_handle.regp, nxgep->npi_handle.regh));
4451 
4452 	if (!isLDOMguest(nxgep)) {
4453 		/*
4454 		 * Stop RxMAC = A.9.2.6
4455 		 */
4456 		if (nxge_rx_mac_disable(nxgep) != NXGE_OK) {
4457 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4458 			    "nxge_rxdma_stop_channel: "
4459 			    "Failed to disable RxMAC"));
4460 		}
4461 
4462 		/*
4463 		 * Drain IPP Port = A.9.3.6
4464 		 */
4465 		(void) nxge_ipp_drain(nxgep);
4466 	}
4467 
4468 	/* Reset RXDMA channel */
4469 	rs = npi_rxdma_cfg_rdc_reset(handle, channel);
4470 	if (rs != NPI_SUCCESS) {
4471 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4472 		    " nxge_rxdma_stop_channel: "
4473 		    " reset rxdma failed (0x%08x channel %d)",
4474 		    rs, channel));
4475 		return (NXGE_ERROR | rs);
4476 	}
4477 
4478 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
4479 	    "==> nxge_rxdma_stop_channel: reset done"));
4480 
4481 	/* Set up the interrupt event masks. */
4482 	ent_mask.value = RX_DMA_ENT_MSK_ALL;
4483 	rs = npi_rxdma_event_mask(handle, OP_SET, channel,
4484 	    &ent_mask);
4485 	if (rs != NPI_SUCCESS) {
4486 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4487 		    "==> nxge_rxdma_stop_channel: "
4488 		    "set rxdma event masks failed (0x%08x channel %d)",
4489 		    rs, channel));
4490 		return (NXGE_ERROR | rs);
4491 	}
4492 
4493 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
4494 	    "==> nxge_rxdma_stop_channel: event done"));
4495 
4496 	/*
4497 	 * Initialize the receive DMA control and status register
4498 	 */
4499 	cs.value = 0;
4500 	status = nxge_init_rxdma_channel_cntl_stat(nxgep, channel, &cs);
4501 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "==> nxge_rxdma_stop_channel: control "
4502 	    " to default (all 0s) 0x%08x", cs.value));
4503 	if (status != NXGE_OK) {
4504 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4505 		    " nxge_rxdma_stop_channel: init rxdma"
4506 		    " control register failed (0x%08x channel %d",
4507 		    status, channel));
4508 		return (status);
4509 	}
4510 
4511 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
4512 	    "==> nxge_rxdma_stop_channel: control done"));
4513 
4514 	/*
4515 	 * Make sure channel is disabled.
4516 	 */
4517 	status = nxge_disable_rxdma_channel(nxgep, channel);
4518 
4519 	if (status != NXGE_OK) {
4520 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4521 		    " nxge_rxdma_stop_channel: "
4522 		    " init enable rxdma failed (0x%08x channel %d)",
4523 		    status, channel));
4524 		return (status);
4525 	}
4526 
4527 	if (!isLDOMguest(nxgep)) {
4528 		/*
4529 		 * Enable RxMAC = A.9.2.10
4530 		 */
4531 		if (nxge_rx_mac_enable(nxgep) != NXGE_OK) {
4532 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4533 			    "nxge_rxdma_stop_channel: Rx MAC still disabled"));
4534 		}
4535 	}
4536 
4537 	NXGE_DEBUG_MSG((nxgep,
4538 	    RX_CTL, "==> nxge_rxdma_stop_channel: disable done"));
4539 
4540 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "<== nxge_rxdma_stop_channel"));
4541 
4542 	return (NXGE_OK);
4543 }
4544 
4545 nxge_status_t
4546 nxge_rxdma_handle_sys_errors(p_nxge_t nxgep)
4547 {
4548 	npi_handle_t		handle;
4549 	p_nxge_rdc_sys_stats_t	statsp;
4550 	rx_ctl_dat_fifo_stat_t	stat;
4551 	uint32_t		zcp_err_status;
4552 	uint32_t		ipp_err_status;
4553 	nxge_status_t		status = NXGE_OK;
4554 	npi_status_t		rs = NPI_SUCCESS;
4555 	boolean_t		my_err = B_FALSE;
4556 
4557 	handle = nxgep->npi_handle;
4558 	statsp = (p_nxge_rdc_sys_stats_t)&nxgep->statsp->rdc_sys_stats;
4559 
4560 	rs = npi_rxdma_rxctl_fifo_error_intr_get(handle, &stat);
4561 
4562 	if (rs != NPI_SUCCESS)
4563 		return (NXGE_ERROR | rs);
4564 
4565 	if (stat.bits.ldw.id_mismatch) {
4566 		statsp->id_mismatch++;
4567 		NXGE_FM_REPORT_ERROR(nxgep, nxgep->mac.portnum, NULL,
4568 		    NXGE_FM_EREPORT_RDMC_ID_MISMATCH);
4569 		/* Global fatal error encountered */
4570 	}
4571 
4572 	if ((stat.bits.ldw.zcp_eop_err) || (stat.bits.ldw.ipp_eop_err)) {
4573 		switch (nxgep->mac.portnum) {
4574 		case 0:
4575 			if ((stat.bits.ldw.zcp_eop_err & FIFO_EOP_PORT0) ||
4576 			    (stat.bits.ldw.ipp_eop_err & FIFO_EOP_PORT0)) {
4577 				my_err = B_TRUE;
4578 				zcp_err_status = stat.bits.ldw.zcp_eop_err;
4579 				ipp_err_status = stat.bits.ldw.ipp_eop_err;
4580 			}
4581 			break;
4582 		case 1:
4583 			if ((stat.bits.ldw.zcp_eop_err & FIFO_EOP_PORT1) ||
4584 			    (stat.bits.ldw.ipp_eop_err & FIFO_EOP_PORT1)) {
4585 				my_err = B_TRUE;
4586 				zcp_err_status = stat.bits.ldw.zcp_eop_err;
4587 				ipp_err_status = stat.bits.ldw.ipp_eop_err;
4588 			}
4589 			break;
4590 		case 2:
4591 			if ((stat.bits.ldw.zcp_eop_err & FIFO_EOP_PORT2) ||
4592 			    (stat.bits.ldw.ipp_eop_err & FIFO_EOP_PORT2)) {
4593 				my_err = B_TRUE;
4594 				zcp_err_status = stat.bits.ldw.zcp_eop_err;
4595 				ipp_err_status = stat.bits.ldw.ipp_eop_err;
4596 			}
4597 			break;
4598 		case 3:
4599 			if ((stat.bits.ldw.zcp_eop_err & FIFO_EOP_PORT3) ||
4600 			    (stat.bits.ldw.ipp_eop_err & FIFO_EOP_PORT3)) {
4601 				my_err = B_TRUE;
4602 				zcp_err_status = stat.bits.ldw.zcp_eop_err;
4603 				ipp_err_status = stat.bits.ldw.ipp_eop_err;
4604 			}
4605 			break;
4606 		default:
4607 			return (NXGE_ERROR);
4608 		}
4609 	}
4610 
4611 	if (my_err) {
4612 		status = nxge_rxdma_handle_port_errors(nxgep, ipp_err_status,
4613 		    zcp_err_status);
4614 		if (status != NXGE_OK)
4615 			return (status);
4616 	}
4617 
4618 	return (NXGE_OK);
4619 }
4620 
4621 static nxge_status_t
4622 nxge_rxdma_handle_port_errors(p_nxge_t nxgep, uint32_t ipp_status,
4623 							uint32_t zcp_status)
4624 {
4625 	boolean_t		rxport_fatal = B_FALSE;
4626 	p_nxge_rdc_sys_stats_t	statsp;
4627 	nxge_status_t		status = NXGE_OK;
4628 	uint8_t			portn;
4629 
4630 	portn = nxgep->mac.portnum;
4631 	statsp = (p_nxge_rdc_sys_stats_t)&nxgep->statsp->rdc_sys_stats;
4632 
4633 	if (ipp_status & (0x1 << portn)) {
4634 		statsp->ipp_eop_err++;
4635 		NXGE_FM_REPORT_ERROR(nxgep, portn, NULL,
4636 		    NXGE_FM_EREPORT_RDMC_IPP_EOP_ERR);
4637 		rxport_fatal = B_TRUE;
4638 	}
4639 
4640 	if (zcp_status & (0x1 << portn)) {
4641 		statsp->zcp_eop_err++;
4642 		NXGE_FM_REPORT_ERROR(nxgep, portn, NULL,
4643 		    NXGE_FM_EREPORT_RDMC_ZCP_EOP_ERR);
4644 		rxport_fatal = B_TRUE;
4645 	}
4646 
4647 	if (rxport_fatal) {
4648 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4649 		    " nxge_rxdma_handle_port_error: "
4650 		    " fatal error on Port #%d\n",
4651 		    portn));
4652 		status = nxge_rx_port_fatal_err_recover(nxgep);
4653 		if (status == NXGE_OK) {
4654 			FM_SERVICE_RESTORED(nxgep);
4655 		}
4656 	}
4657 
4658 	return (status);
4659 }
4660 
4661 static nxge_status_t
4662 nxge_rxdma_fatal_err_recover(p_nxge_t nxgep, uint16_t channel)
4663 {
4664 	npi_handle_t		handle;
4665 	npi_status_t		rs = NPI_SUCCESS;
4666 	nxge_status_t		status = NXGE_OK;
4667 	p_rx_rbr_ring_t		rbrp;
4668 	p_rx_rcr_ring_t		rcrp;
4669 	p_rx_mbox_t		mboxp;
4670 	rx_dma_ent_msk_t	ent_mask;
4671 	p_nxge_dma_common_t	dmap;
4672 	int			ring_idx;
4673 	uint32_t		ref_cnt;
4674 	p_rx_msg_t		rx_msg_p;
4675 	int			i;
4676 	uint32_t		nxge_port_rcr_size;
4677 
4678 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "<== nxge_rxdma_fatal_err_recover"));
4679 	NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4680 	    "Recovering from RxDMAChannel#%d error...", channel));
4681 
4682 	/*
4683 	 * Stop the dma channel waits for the stop done.
4684 	 * If the stop done bit is not set, then create
4685 	 * an error.
4686 	 */
4687 
4688 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
4689 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "Rx DMA stop..."));
4690 
4691 	ring_idx = nxge_rxdma_get_ring_index(nxgep, channel);
4692 	rbrp = (p_rx_rbr_ring_t)nxgep->rx_rbr_rings->rbr_rings[ring_idx];
4693 	rcrp = (p_rx_rcr_ring_t)nxgep->rx_rcr_rings->rcr_rings[ring_idx];
4694 
4695 	MUTEX_ENTER(&rcrp->lock);
4696 	MUTEX_ENTER(&rbrp->lock);
4697 	MUTEX_ENTER(&rbrp->post_lock);
4698 
4699 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "Disable RxDMA channel..."));
4700 
4701 	rs = npi_rxdma_cfg_rdc_disable(handle, channel);
4702 	if (rs != NPI_SUCCESS) {
4703 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4704 		    "nxge_disable_rxdma_channel:failed"));
4705 		goto fail;
4706 	}
4707 
4708 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "Disable RxDMA interrupt..."));
4709 
4710 	/* Disable interrupt */
4711 	ent_mask.value = RX_DMA_ENT_MSK_ALL;
4712 	rs = npi_rxdma_event_mask(handle, OP_SET, channel, &ent_mask);
4713 	if (rs != NPI_SUCCESS) {
4714 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4715 		    "nxge_rxdma_stop_channel: "
4716 		    "set rxdma event masks failed (channel %d)",
4717 		    channel));
4718 	}
4719 
4720 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "RxDMA channel reset..."));
4721 
4722 	/* Reset RXDMA channel */
4723 	rs = npi_rxdma_cfg_rdc_reset(handle, channel);
4724 	if (rs != NPI_SUCCESS) {
4725 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4726 		    "nxge_rxdma_fatal_err_recover: "
4727 		    " reset rxdma failed (channel %d)", channel));
4728 		goto fail;
4729 	}
4730 
4731 	nxge_port_rcr_size = nxgep->nxge_port_rcr_size;
4732 
4733 	mboxp =
4734 	    (p_rx_mbox_t)nxgep->rx_mbox_areas_p->rxmbox_areas[ring_idx];
4735 
4736 	rbrp->rbr_wr_index = (rbrp->rbb_max - 1);
4737 	rbrp->rbr_rd_index = 0;
4738 
4739 	rcrp->comp_rd_index = 0;
4740 	rcrp->comp_wt_index = 0;
4741 	rcrp->rcr_desc_rd_head_p = rcrp->rcr_desc_first_p =
4742 	    (p_rcr_entry_t)DMA_COMMON_VPTR(rcrp->rcr_desc);
4743 #if defined(__i386)
4744 	rcrp->rcr_desc_rd_head_pp = rcrp->rcr_desc_first_pp =
4745 	    (p_rcr_entry_t)(uint32_t)DMA_COMMON_IOADDR(rcrp->rcr_desc);
4746 #else
4747 	rcrp->rcr_desc_rd_head_pp = rcrp->rcr_desc_first_pp =
4748 	    (p_rcr_entry_t)DMA_COMMON_IOADDR(rcrp->rcr_desc);
4749 #endif
4750 
4751 	rcrp->rcr_desc_last_p = rcrp->rcr_desc_rd_head_p +
4752 	    (nxge_port_rcr_size - 1);
4753 	rcrp->rcr_desc_last_pp = rcrp->rcr_desc_rd_head_pp +
4754 	    (nxge_port_rcr_size - 1);
4755 
4756 	dmap = (p_nxge_dma_common_t)&rcrp->rcr_desc;
4757 	bzero((caddr_t)dmap->kaddrp, dmap->alength);
4758 
4759 	cmn_err(CE_NOTE, "!rbr entries = %d\n", rbrp->rbr_max_size);
4760 
4761 	for (i = 0; i < rbrp->rbr_max_size; i++) {
4762 		rx_msg_p = rbrp->rx_msg_ring[i];
4763 		ref_cnt = rx_msg_p->ref_cnt;
4764 		if (ref_cnt != 1) {
4765 			if (rx_msg_p->cur_usage_cnt !=
4766 			    rx_msg_p->max_usage_cnt) {
4767 				NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4768 				    "buf[%d]: cur_usage_cnt = %d "
4769 				    "max_usage_cnt = %d\n", i,
4770 				    rx_msg_p->cur_usage_cnt,
4771 				    rx_msg_p->max_usage_cnt));
4772 			} else {
4773 				/* Buffer can be re-posted */
4774 				rx_msg_p->free = B_TRUE;
4775 				rx_msg_p->cur_usage_cnt = 0;
4776 				rx_msg_p->max_usage_cnt = 0xbaddcafe;
4777 				rx_msg_p->pkt_buf_size = 0;
4778 			}
4779 		}
4780 	}
4781 
4782 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "RxDMA channel re-start..."));
4783 
4784 	status = nxge_rxdma_start_channel(nxgep, channel, rbrp, rcrp, mboxp);
4785 	if (status != NXGE_OK) {
4786 		goto fail;
4787 	}
4788 
4789 	MUTEX_EXIT(&rbrp->post_lock);
4790 	MUTEX_EXIT(&rbrp->lock);
4791 	MUTEX_EXIT(&rcrp->lock);
4792 
4793 	NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4794 	    "Recovery Successful, RxDMAChannel#%d Restored",
4795 	    channel));
4796 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "==> nxge_rxdma_fatal_err_recover"));
4797 
4798 	return (NXGE_OK);
4799 fail:
4800 	MUTEX_EXIT(&rbrp->post_lock);
4801 	MUTEX_EXIT(&rbrp->lock);
4802 	MUTEX_EXIT(&rcrp->lock);
4803 	NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL, "Recovery failed"));
4804 
4805 	return (NXGE_ERROR | rs);
4806 }
4807 
4808 nxge_status_t
4809 nxge_rx_port_fatal_err_recover(p_nxge_t nxgep)
4810 {
4811 	nxge_grp_set_t *set = &nxgep->rx_set;
4812 	nxge_status_t status = NXGE_OK;
4813 	int rdc;
4814 
4815 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "<== nxge_rx_port_fatal_err_recover"));
4816 	NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4817 	    "Recovering from RxPort error..."));
4818 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "Disabling RxMAC...\n"));
4819 
4820 	if (nxge_rx_mac_disable(nxgep) != NXGE_OK)
4821 		goto fail;
4822 
4823 	NXGE_DELAY(1000);
4824 
4825 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "Stopping all RxDMA channels..."));
4826 
4827 	for (rdc = 0; rdc < NXGE_MAX_RDCS; rdc++) {
4828 		if ((1 << rdc) & set->owned.map) {
4829 			if (nxge_rxdma_fatal_err_recover(nxgep, rdc)
4830 			    != NXGE_OK) {
4831 				NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4832 				    "Could not recover channel %d", rdc));
4833 			}
4834 		}
4835 	}
4836 
4837 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "Resetting IPP..."));
4838 
4839 	/* Reset IPP */
4840 	if (nxge_ipp_reset(nxgep) != NXGE_OK) {
4841 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4842 		    "nxge_rx_port_fatal_err_recover: "
4843 		    "Failed to reset IPP"));
4844 		goto fail;
4845 	}
4846 
4847 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "Reset RxMAC..."));
4848 
4849 	/* Reset RxMAC */
4850 	if (nxge_rx_mac_reset(nxgep) != NXGE_OK) {
4851 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4852 		    "nxge_rx_port_fatal_err_recover: "
4853 		    "Failed to reset RxMAC"));
4854 		goto fail;
4855 	}
4856 
4857 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "Re-initialize IPP..."));
4858 
4859 	/* Re-Initialize IPP */
4860 	if (nxge_ipp_init(nxgep) != NXGE_OK) {
4861 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4862 		    "nxge_rx_port_fatal_err_recover: "
4863 		    "Failed to init IPP"));
4864 		goto fail;
4865 	}
4866 
4867 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "Re-initialize RxMAC..."));
4868 
4869 	/* Re-Initialize RxMAC */
4870 	if ((status = nxge_rx_mac_init(nxgep)) != NXGE_OK) {
4871 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4872 		    "nxge_rx_port_fatal_err_recover: "
4873 		    "Failed to reset RxMAC"));
4874 		goto fail;
4875 	}
4876 
4877 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "Re-enable RxMAC..."));
4878 
4879 	/* Re-enable RxMAC */
4880 	if ((status = nxge_rx_mac_enable(nxgep)) != NXGE_OK) {
4881 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4882 		    "nxge_rx_port_fatal_err_recover: "
4883 		    "Failed to enable RxMAC"));
4884 		goto fail;
4885 	}
4886 
4887 	NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4888 	    "Recovery Successful, RxPort Restored"));
4889 
4890 	return (NXGE_OK);
4891 fail:
4892 	NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL, "Recovery failed"));
4893 	return (status);
4894 }
4895 
4896 void
4897 nxge_rxdma_inject_err(p_nxge_t nxgep, uint32_t err_id, uint8_t chan)
4898 {
4899 	rx_dma_ctl_stat_t	cs;
4900 	rx_ctl_dat_fifo_stat_t	cdfs;
4901 
4902 	switch (err_id) {
4903 	case NXGE_FM_EREPORT_RDMC_RCR_ACK_ERR:
4904 	case NXGE_FM_EREPORT_RDMC_DC_FIFO_ERR:
4905 	case NXGE_FM_EREPORT_RDMC_RCR_SHA_PAR:
4906 	case NXGE_FM_EREPORT_RDMC_RBR_PRE_PAR:
4907 	case NXGE_FM_EREPORT_RDMC_RBR_TMOUT:
4908 	case NXGE_FM_EREPORT_RDMC_RSP_CNT_ERR:
4909 	case NXGE_FM_EREPORT_RDMC_BYTE_EN_BUS:
4910 	case NXGE_FM_EREPORT_RDMC_RSP_DAT_ERR:
4911 	case NXGE_FM_EREPORT_RDMC_RCRINCON:
4912 	case NXGE_FM_EREPORT_RDMC_RCRFULL:
4913 	case NXGE_FM_EREPORT_RDMC_RBRFULL:
4914 	case NXGE_FM_EREPORT_RDMC_RBRLOGPAGE:
4915 	case NXGE_FM_EREPORT_RDMC_CFIGLOGPAGE:
4916 	case NXGE_FM_EREPORT_RDMC_CONFIG_ERR:
4917 		RXDMA_REG_READ64(nxgep->npi_handle, RX_DMA_CTL_STAT_DBG_REG,
4918 		    chan, &cs.value);
4919 		if (err_id == NXGE_FM_EREPORT_RDMC_RCR_ACK_ERR)
4920 			cs.bits.hdw.rcr_ack_err = 1;
4921 		else if (err_id == NXGE_FM_EREPORT_RDMC_DC_FIFO_ERR)
4922 			cs.bits.hdw.dc_fifo_err = 1;
4923 		else if (err_id == NXGE_FM_EREPORT_RDMC_RCR_SHA_PAR)
4924 			cs.bits.hdw.rcr_sha_par = 1;
4925 		else if (err_id == NXGE_FM_EREPORT_RDMC_RBR_PRE_PAR)
4926 			cs.bits.hdw.rbr_pre_par = 1;
4927 		else if (err_id == NXGE_FM_EREPORT_RDMC_RBR_TMOUT)
4928 			cs.bits.hdw.rbr_tmout = 1;
4929 		else if (err_id == NXGE_FM_EREPORT_RDMC_RSP_CNT_ERR)
4930 			cs.bits.hdw.rsp_cnt_err = 1;
4931 		else if (err_id == NXGE_FM_EREPORT_RDMC_BYTE_EN_BUS)
4932 			cs.bits.hdw.byte_en_bus = 1;
4933 		else if (err_id == NXGE_FM_EREPORT_RDMC_RSP_DAT_ERR)
4934 			cs.bits.hdw.rsp_dat_err = 1;
4935 		else if (err_id == NXGE_FM_EREPORT_RDMC_CONFIG_ERR)
4936 			cs.bits.hdw.config_err = 1;
4937 		else if (err_id == NXGE_FM_EREPORT_RDMC_RCRINCON)
4938 			cs.bits.hdw.rcrincon = 1;
4939 		else if (err_id == NXGE_FM_EREPORT_RDMC_RCRFULL)
4940 			cs.bits.hdw.rcrfull = 1;
4941 		else if (err_id == NXGE_FM_EREPORT_RDMC_RBRFULL)
4942 			cs.bits.hdw.rbrfull = 1;
4943 		else if (err_id == NXGE_FM_EREPORT_RDMC_RBRLOGPAGE)
4944 			cs.bits.hdw.rbrlogpage = 1;
4945 		else if (err_id == NXGE_FM_EREPORT_RDMC_CFIGLOGPAGE)
4946 			cs.bits.hdw.cfiglogpage = 1;
4947 #if defined(__i386)
4948 		cmn_err(CE_NOTE, "!Write 0x%llx to RX_DMA_CTL_STAT_DBG_REG\n",
4949 		    cs.value);
4950 #else
4951 		cmn_err(CE_NOTE, "!Write 0x%lx to RX_DMA_CTL_STAT_DBG_REG\n",
4952 		    cs.value);
4953 #endif
4954 		RXDMA_REG_WRITE64(nxgep->npi_handle, RX_DMA_CTL_STAT_DBG_REG,
4955 		    chan, cs.value);
4956 		break;
4957 	case NXGE_FM_EREPORT_RDMC_ID_MISMATCH:
4958 	case NXGE_FM_EREPORT_RDMC_ZCP_EOP_ERR:
4959 	case NXGE_FM_EREPORT_RDMC_IPP_EOP_ERR:
4960 		cdfs.value = 0;
4961 		if (err_id ==  NXGE_FM_EREPORT_RDMC_ID_MISMATCH)
4962 			cdfs.bits.ldw.id_mismatch = (1 << nxgep->mac.portnum);
4963 		else if (err_id == NXGE_FM_EREPORT_RDMC_ZCP_EOP_ERR)
4964 			cdfs.bits.ldw.zcp_eop_err = (1 << nxgep->mac.portnum);
4965 		else if (err_id == NXGE_FM_EREPORT_RDMC_IPP_EOP_ERR)
4966 			cdfs.bits.ldw.ipp_eop_err = (1 << nxgep->mac.portnum);
4967 #if defined(__i386)
4968 		cmn_err(CE_NOTE,
4969 		    "!Write 0x%llx to RX_CTL_DAT_FIFO_STAT_DBG_REG\n",
4970 		    cdfs.value);
4971 #else
4972 		cmn_err(CE_NOTE,
4973 		    "!Write 0x%lx to RX_CTL_DAT_FIFO_STAT_DBG_REG\n",
4974 		    cdfs.value);
4975 #endif
4976 		NXGE_REG_WR64(nxgep->npi_handle,
4977 		    RX_CTL_DAT_FIFO_STAT_DBG_REG, cdfs.value);
4978 		break;
4979 	case NXGE_FM_EREPORT_RDMC_DCF_ERR:
4980 		break;
4981 	case NXGE_FM_EREPORT_RDMC_RCR_ERR:
4982 		break;
4983 	}
4984 }
4985 
4986 static void
4987 nxge_rxdma_databuf_free(p_rx_rbr_ring_t rbr_p)
4988 {
4989 	rxring_info_t 		*ring_info;
4990 	int			index;
4991 	uint32_t		chunk_size;
4992 	uint64_t		kaddr;
4993 	uint_t			num_blocks;
4994 
4995 	NXGE_DEBUG_MSG((NULL, DMA_CTL, "==> nxge_rxdma_databuf_free"));
4996 
4997 	if (rbr_p == NULL) {
4998 		NXGE_ERROR_MSG((NULL, NXGE_ERR_CTL,
4999 		    "==> nxge_rxdma_databuf_free: NULL rbr pointer"));
5000 		return;
5001 	}
5002 
5003 	if (rbr_p->rbr_alloc_type == DDI_MEM_ALLOC) {
5004 		NXGE_ERROR_MSG((NULL, NXGE_ERR_CTL,
5005 		    "==> nxge_rxdma_databuf_free: DDI"));
5006 		return;
5007 	}
5008 
5009 	ring_info = rbr_p->ring_info;
5010 	if (ring_info == NULL) {
5011 		NXGE_ERROR_MSG((NULL, NXGE_ERR_CTL,
5012 		    "==> nxge_rxdma_databuf_free: NULL ring info"));
5013 		return;
5014 	}
5015 	num_blocks = rbr_p->num_blocks;
5016 	for (index = 0; index < num_blocks; index++) {
5017 		kaddr = ring_info->buffer[index].kaddr;
5018 		chunk_size = ring_info->buffer[index].buf_size;
5019 		NXGE_DEBUG_MSG((NULL, DMA_CTL,
5020 		    "==> nxge_rxdma_databuf_free: free chunk %d "
5021 		    "kaddrp $%p chunk size %d",
5022 		    index, kaddr, chunk_size));
5023 		if (kaddr == NULL) continue;
5024 		nxge_free_buf(rbr_p->rbr_alloc_type, kaddr, chunk_size);
5025 		ring_info->buffer[index].kaddr = NULL;
5026 	}
5027 
5028 	NXGE_DEBUG_MSG((NULL, DMA_CTL, "<== nxge_rxdma_databuf_free"));
5029 }
5030 
5031 #if	defined(sun4v) && defined(NIU_LP_WORKAROUND)
5032 extern void contig_mem_free(void *, size_t);
5033 #endif
5034 
5035 void
5036 nxge_free_buf(buf_alloc_type_t alloc_type, uint64_t kaddr, uint32_t buf_size)
5037 {
5038 	NXGE_DEBUG_MSG((NULL, DMA_CTL, "==> nxge_free_buf"));
5039 
5040 	if (kaddr == NULL || !buf_size) {
5041 		NXGE_ERROR_MSG((NULL, NXGE_ERR_CTL,
5042 		    "==> nxge_free_buf: invalid kaddr $%p size to free %d",
5043 		    kaddr, buf_size));
5044 		return;
5045 	}
5046 
5047 	switch (alloc_type) {
5048 	case KMEM_ALLOC:
5049 		NXGE_DEBUG_MSG((NULL, DMA_CTL,
5050 		    "==> nxge_free_buf: freeing kmem $%p size %d",
5051 		    kaddr, buf_size));
5052 #if defined(__i386)
5053 		KMEM_FREE((void *)(uint32_t)kaddr, buf_size);
5054 #else
5055 		KMEM_FREE((void *)kaddr, buf_size);
5056 #endif
5057 		break;
5058 
5059 #if	defined(sun4v) && defined(NIU_LP_WORKAROUND)
5060 	case CONTIG_MEM_ALLOC:
5061 		NXGE_DEBUG_MSG((NULL, DMA_CTL,
5062 		    "==> nxge_free_buf: freeing contig_mem kaddr $%p size %d",
5063 		    kaddr, buf_size));
5064 		contig_mem_free((void *)kaddr, buf_size);
5065 		break;
5066 #endif
5067 
5068 	default:
5069 		NXGE_ERROR_MSG((NULL, NXGE_ERR_CTL,
5070 		    "<== nxge_free_buf: unsupported alloc type %d",
5071 		    alloc_type));
5072 		return;
5073 	}
5074 
5075 	NXGE_DEBUG_MSG((NULL, DMA_CTL, "<== nxge_free_buf"));
5076 }
5077