xref: /titanic_41/usr/src/uts/common/io/nxge/nxge_rxdma.c (revision 1bdd6c0e3710e91cb1f31aa78de33cb638494480)
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  * Globals: tunable parameters (/etc/system or adb)
43  *
44  */
45 extern uint32_t nxge_rbr_size;
46 extern uint32_t nxge_rcr_size;
47 extern uint32_t	nxge_rbr_spare_size;
48 
49 extern uint32_t nxge_mblks_pending;
50 
51 /*
52  * Tunable to reduce the amount of time spent in the
53  * ISR doing Rx Processing.
54  */
55 extern uint32_t nxge_max_rx_pkts;
56 boolean_t nxge_jumbo_enable;
57 
58 /*
59  * Tunables to manage the receive buffer blocks.
60  *
61  * nxge_rx_threshold_hi: copy all buffers.
62  * nxge_rx_bcopy_size_type: receive buffer block size type.
63  * nxge_rx_threshold_lo: copy only up to tunable block size type.
64  */
65 extern nxge_rxbuf_threshold_t nxge_rx_threshold_hi;
66 extern nxge_rxbuf_type_t nxge_rx_buf_size_type;
67 extern nxge_rxbuf_threshold_t nxge_rx_threshold_lo;
68 
69 extern uint32_t	nxge_cksum_offload;
70 
71 static nxge_status_t nxge_map_rxdma(p_nxge_t, int);
72 static void nxge_unmap_rxdma(p_nxge_t, int);
73 
74 static nxge_status_t nxge_rxdma_hw_start_common(p_nxge_t);
75 
76 static nxge_status_t nxge_rxdma_hw_start(p_nxge_t, int);
77 static void nxge_rxdma_hw_stop(p_nxge_t, int);
78 
79 static nxge_status_t nxge_map_rxdma_channel(p_nxge_t, uint16_t,
80     p_nxge_dma_common_t *,  p_rx_rbr_ring_t *,
81     uint32_t,
82     p_nxge_dma_common_t *, p_rx_rcr_ring_t *,
83     p_rx_mbox_t *);
84 static void nxge_unmap_rxdma_channel(p_nxge_t, uint16_t,
85     p_rx_rbr_ring_t, p_rx_rcr_ring_t, p_rx_mbox_t);
86 
87 static nxge_status_t nxge_map_rxdma_channel_cfg_ring(p_nxge_t,
88     uint16_t,
89     p_nxge_dma_common_t *, p_rx_rbr_ring_t *,
90     p_rx_rcr_ring_t *, p_rx_mbox_t *);
91 static void nxge_unmap_rxdma_channel_cfg_ring(p_nxge_t,
92     p_rx_rcr_ring_t, p_rx_mbox_t);
93 
94 static nxge_status_t nxge_map_rxdma_channel_buf_ring(p_nxge_t,
95     uint16_t,
96     p_nxge_dma_common_t *,
97     p_rx_rbr_ring_t *, uint32_t);
98 static void nxge_unmap_rxdma_channel_buf_ring(p_nxge_t,
99     p_rx_rbr_ring_t);
100 
101 static nxge_status_t nxge_rxdma_start_channel(p_nxge_t, uint16_t,
102     p_rx_rbr_ring_t, p_rx_rcr_ring_t, p_rx_mbox_t);
103 static nxge_status_t nxge_rxdma_stop_channel(p_nxge_t, uint16_t);
104 
105 static mblk_t *
106 nxge_rx_pkts(p_nxge_t, p_rx_rcr_ring_t, rx_dma_ctl_stat_t, int);
107 
108 static void nxge_receive_packet(p_nxge_t,
109 	p_rx_rcr_ring_t,
110 	p_rcr_entry_t,
111 	boolean_t *,
112 	mblk_t **, mblk_t **);
113 
114 nxge_status_t nxge_disable_rxdma_channel(p_nxge_t, uint16_t);
115 
116 static p_rx_msg_t nxge_allocb(size_t, uint32_t, p_nxge_dma_common_t);
117 static void nxge_freeb(p_rx_msg_t);
118 static mblk_t *nxge_rx_pkts_vring(p_nxge_t, uint_t, rx_dma_ctl_stat_t);
119 static nxge_status_t nxge_rx_err_evnts(p_nxge_t, int, rx_dma_ctl_stat_t);
120 
121 static nxge_status_t nxge_rxdma_handle_port_errors(p_nxge_t,
122 				uint32_t, uint32_t);
123 
124 static nxge_status_t nxge_rxbuf_index_info_init(p_nxge_t,
125     p_rx_rbr_ring_t);
126 
127 
128 static nxge_status_t
129 nxge_rxdma_fatal_err_recover(p_nxge_t, uint16_t);
130 
131 nxge_status_t
132 nxge_rx_port_fatal_err_recover(p_nxge_t);
133 
134 static void nxge_rxdma_databuf_free(p_rx_rbr_ring_t);
135 
136 nxge_status_t
137 nxge_init_rxdma_channels(p_nxge_t nxgep)
138 {
139 	nxge_grp_set_t	*set = &nxgep->rx_set;
140 	int		i, count, channel;
141 	nxge_grp_t	*group;
142 	dc_map_t	map;
143 	int		dev_gindex;
144 
145 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_init_rxdma_channels"));
146 
147 	if (!isLDOMguest(nxgep)) {
148 		if (nxge_rxdma_hw_start_common(nxgep) != NXGE_OK) {
149 			cmn_err(CE_NOTE, "hw_start_common");
150 			return (NXGE_ERROR);
151 		}
152 	}
153 
154 	/*
155 	 * NXGE_LOGICAL_GROUP_MAX > NXGE_MAX_RDC_GROUPS (8)
156 	 * We only have 8 hardware RDC tables, but we may have
157 	 * up to 16 logical (software-defined) groups of RDCS,
158 	 * if we make use of layer 3 & 4 hardware classification.
159 	 */
160 	for (i = 0, count = 0; i < NXGE_LOGICAL_GROUP_MAX; i++) {
161 		if ((1 << i) & set->lg.map) {
162 			group = set->group[i];
163 			dev_gindex =
164 			    nxgep->pt_config.hw_config.def_mac_rxdma_grpid + i;
165 			map = nxgep->pt_config.rdc_grps[dev_gindex].map;
166 			for (channel = 0; channel < NXGE_MAX_RDCS; channel++) {
167 				if ((1 << channel) & map) {
168 					if ((nxge_grp_dc_add(nxgep,
169 					    group, VP_BOUND_RX, channel)))
170 						goto init_rxdma_channels_exit;
171 				}
172 			}
173 		}
174 		if (++count == set->lg.count)
175 			break;
176 	}
177 
178 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "<== nxge_init_rxdma_channels"));
179 	return (NXGE_OK);
180 
181 init_rxdma_channels_exit:
182 	for (i = 0, count = 0; i < NXGE_LOGICAL_GROUP_MAX; i++) {
183 		if ((1 << i) & set->lg.map) {
184 			group = set->group[i];
185 			dev_gindex =
186 			    nxgep->pt_config.hw_config.def_mac_rxdma_grpid + i;
187 			map = nxgep->pt_config.rdc_grps[dev_gindex].map;
188 			for (channel = 0; channel < NXGE_MAX_RDCS; channel++) {
189 				if ((1 << channel) & map) {
190 					nxge_grp_dc_remove(nxgep,
191 					    VP_BOUND_RX, channel);
192 				}
193 			}
194 		}
195 		if (++count == set->lg.count)
196 			break;
197 	}
198 
199 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "<== nxge_init_rxdma_channels"));
200 	return (NXGE_ERROR);
201 }
202 
203 nxge_status_t
204 nxge_init_rxdma_channel(p_nxge_t nxge, int channel)
205 {
206 	nxge_status_t	status;
207 
208 	NXGE_DEBUG_MSG((nxge, MEM2_CTL, "==> nxge_init_rxdma_channel"));
209 
210 	status = nxge_map_rxdma(nxge, channel);
211 	if (status != NXGE_OK) {
212 		NXGE_ERROR_MSG((nxge, NXGE_ERR_CTL,
213 		    "<== nxge_init_rxdma: status 0x%x", status));
214 		return (status);
215 	}
216 
217 #if defined(sun4v)
218 	if (isLDOMguest(nxge)) {
219 		/* set rcr_ring */
220 		p_rx_rcr_ring_t ring = nxge->rx_rcr_rings->rcr_rings[channel];
221 
222 		status = nxge_hio_rxdma_bind_intr(nxge, ring, channel);
223 		if (status != NXGE_OK) {
224 			nxge_unmap_rxdma(nxge, channel);
225 			return (status);
226 		}
227 	}
228 #endif
229 
230 	status = nxge_rxdma_hw_start(nxge, channel);
231 	if (status != NXGE_OK) {
232 		nxge_unmap_rxdma(nxge, channel);
233 	}
234 
235 	if (!nxge->statsp->rdc_ksp[channel])
236 		nxge_setup_rdc_kstats(nxge, channel);
237 
238 	NXGE_DEBUG_MSG((nxge, MEM2_CTL,
239 	    "<== nxge_init_rxdma_channel: status 0x%x", status));
240 
241 	return (status);
242 }
243 
244 void
245 nxge_uninit_rxdma_channels(p_nxge_t nxgep)
246 {
247 	nxge_grp_set_t *set = &nxgep->rx_set;
248 	int rdc;
249 
250 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_uninit_rxdma_channels"));
251 
252 	if (set->owned.map == 0) {
253 		NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
254 		    "nxge_uninit_rxdma_channels: no channels"));
255 		return;
256 	}
257 
258 	for (rdc = 0; rdc < NXGE_MAX_RDCS; rdc++) {
259 		if ((1 << rdc) & set->owned.map) {
260 			nxge_grp_dc_remove(nxgep, VP_BOUND_RX, rdc);
261 		}
262 	}
263 
264 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "<== nxge_uninit_rxdma_channels"));
265 }
266 
267 void
268 nxge_uninit_rxdma_channel(p_nxge_t nxgep, int channel)
269 {
270 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_uninit_rxdma_channel"));
271 
272 	if (nxgep->statsp->rdc_ksp[channel]) {
273 		kstat_delete(nxgep->statsp->rdc_ksp[channel]);
274 		nxgep->statsp->rdc_ksp[channel] = 0;
275 	}
276 
277 	nxge_rxdma_hw_stop(nxgep, channel);
278 	nxge_unmap_rxdma(nxgep, channel);
279 
280 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "<== nxge_uinit_rxdma_channel"));
281 }
282 
283 nxge_status_t
284 nxge_reset_rxdma_channel(p_nxge_t nxgep, uint16_t channel)
285 {
286 	npi_handle_t		handle;
287 	npi_status_t		rs = NPI_SUCCESS;
288 	nxge_status_t		status = NXGE_OK;
289 
290 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "==> nxge_reset_rxdma_channel"));
291 
292 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
293 	rs = npi_rxdma_cfg_rdc_reset(handle, channel);
294 
295 	if (rs != NPI_SUCCESS) {
296 		status = NXGE_ERROR | rs;
297 	}
298 
299 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "<== nxge_reset_rxdma_channel"));
300 
301 	return (status);
302 }
303 
304 void
305 nxge_rxdma_regs_dump_channels(p_nxge_t nxgep)
306 {
307 	nxge_grp_set_t *set = &nxgep->rx_set;
308 	int rdc;
309 
310 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "==> nxge_rxdma_regs_dump_channels"));
311 
312 	if (!isLDOMguest(nxgep)) {
313 		npi_handle_t handle = NXGE_DEV_NPI_HANDLE(nxgep);
314 		(void) npi_rxdma_dump_fzc_regs(handle);
315 	}
316 
317 	if (nxgep->rx_rbr_rings == 0 || nxgep->rx_rbr_rings->rbr_rings == 0) {
318 		NXGE_DEBUG_MSG((nxgep, TX_CTL,
319 		    "nxge_rxdma_regs_dump_channels: "
320 		    "NULL ring pointer(s)"));
321 		return;
322 	}
323 
324 	if (set->owned.map == 0) {
325 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
326 		    "nxge_rxdma_regs_dump_channels: no channels"));
327 		return;
328 	}
329 
330 	for (rdc = 0; rdc < NXGE_MAX_RDCS; rdc++) {
331 		if ((1 << rdc) & set->owned.map) {
332 			rx_rbr_ring_t *ring =
333 			    nxgep->rx_rbr_rings->rbr_rings[rdc];
334 			if (ring) {
335 				(void) nxge_dump_rxdma_channel(nxgep, rdc);
336 			}
337 		}
338 	}
339 
340 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "<== nxge_rxdma_regs_dump"));
341 }
342 
343 nxge_status_t
344 nxge_dump_rxdma_channel(p_nxge_t nxgep, uint8_t channel)
345 {
346 	npi_handle_t		handle;
347 	npi_status_t		rs = NPI_SUCCESS;
348 	nxge_status_t		status = NXGE_OK;
349 
350 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "==> nxge_dump_rxdma_channel"));
351 
352 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
353 	rs = npi_rxdma_dump_rdc_regs(handle, channel);
354 
355 	if (rs != NPI_SUCCESS) {
356 		status = NXGE_ERROR | rs;
357 	}
358 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "<== nxge_dump_rxdma_channel"));
359 	return (status);
360 }
361 
362 nxge_status_t
363 nxge_init_rxdma_channel_event_mask(p_nxge_t nxgep, uint16_t channel,
364     p_rx_dma_ent_msk_t mask_p)
365 {
366 	npi_handle_t		handle;
367 	npi_status_t		rs = NPI_SUCCESS;
368 	nxge_status_t		status = NXGE_OK;
369 
370 	NXGE_DEBUG_MSG((nxgep, DMA_CTL,
371 	    "<== nxge_init_rxdma_channel_event_mask"));
372 
373 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
374 	rs = npi_rxdma_event_mask(handle, OP_SET, channel, mask_p);
375 	if (rs != NPI_SUCCESS) {
376 		status = NXGE_ERROR | rs;
377 	}
378 
379 	return (status);
380 }
381 
382 nxge_status_t
383 nxge_init_rxdma_channel_cntl_stat(p_nxge_t nxgep, uint16_t channel,
384     p_rx_dma_ctl_stat_t cs_p)
385 {
386 	npi_handle_t		handle;
387 	npi_status_t		rs = NPI_SUCCESS;
388 	nxge_status_t		status = NXGE_OK;
389 
390 	NXGE_DEBUG_MSG((nxgep, DMA_CTL,
391 	    "<== nxge_init_rxdma_channel_cntl_stat"));
392 
393 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
394 	rs = npi_rxdma_control_status(handle, OP_SET, channel, cs_p);
395 
396 	if (rs != NPI_SUCCESS) {
397 		status = NXGE_ERROR | rs;
398 	}
399 
400 	return (status);
401 }
402 
403 /*
404  * nxge_rxdma_cfg_rdcgrp_default_rdc
405  *
406  *	Set the default RDC for an RDC Group (Table)
407  *
408  * Arguments:
409  * 	nxgep
410  *	rdcgrp	The group to modify
411  *	rdc	The new default RDC.
412  *
413  * Notes:
414  *
415  * NPI/NXGE function calls:
416  *	npi_rxdma_cfg_rdc_table_default_rdc()
417  *
418  * Registers accessed:
419  *	RDC_TBL_REG: FZC_ZCP + 0x10000
420  *
421  * Context:
422  *	Service domain
423  */
424 nxge_status_t
425 nxge_rxdma_cfg_rdcgrp_default_rdc(
426 	p_nxge_t nxgep,
427 	uint8_t rdcgrp,
428 	uint8_t rdc)
429 {
430 	npi_handle_t		handle;
431 	npi_status_t		rs = NPI_SUCCESS;
432 	p_nxge_dma_pt_cfg_t	p_dma_cfgp;
433 	p_nxge_rdc_grp_t	rdc_grp_p;
434 	uint8_t actual_rdcgrp, actual_rdc;
435 
436 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
437 	    " ==> nxge_rxdma_cfg_rdcgrp_default_rdc"));
438 	p_dma_cfgp = (p_nxge_dma_pt_cfg_t)&nxgep->pt_config;
439 
440 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
441 
442 	/*
443 	 * This has to be rewritten.  Do we even allow this anymore?
444 	 */
445 	rdc_grp_p = &p_dma_cfgp->rdc_grps[rdcgrp];
446 	RDC_MAP_IN(rdc_grp_p->map, rdc);
447 	rdc_grp_p->def_rdc = rdc;
448 
449 	actual_rdcgrp = NXGE_ACTUAL_RDCGRP(nxgep, rdcgrp);
450 	actual_rdc = NXGE_ACTUAL_RDC(nxgep, rdc);
451 
452 	rs = npi_rxdma_cfg_rdc_table_default_rdc(
453 	    handle, actual_rdcgrp, actual_rdc);
454 
455 	if (rs != NPI_SUCCESS) {
456 		return (NXGE_ERROR | rs);
457 	}
458 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
459 	    " <== nxge_rxdma_cfg_rdcgrp_default_rdc"));
460 	return (NXGE_OK);
461 }
462 
463 nxge_status_t
464 nxge_rxdma_cfg_port_default_rdc(p_nxge_t nxgep, uint8_t port, uint8_t rdc)
465 {
466 	npi_handle_t		handle;
467 
468 	uint8_t actual_rdc;
469 	npi_status_t		rs = NPI_SUCCESS;
470 
471 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
472 	    " ==> nxge_rxdma_cfg_port_default_rdc"));
473 
474 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
475 	actual_rdc = rdc;	/* XXX Hack! */
476 	rs = npi_rxdma_cfg_default_port_rdc(handle, port, actual_rdc);
477 
478 
479 	if (rs != NPI_SUCCESS) {
480 		return (NXGE_ERROR | rs);
481 	}
482 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
483 	    " <== nxge_rxdma_cfg_port_default_rdc"));
484 
485 	return (NXGE_OK);
486 }
487 
488 nxge_status_t
489 nxge_rxdma_cfg_rcr_threshold(p_nxge_t nxgep, uint8_t channel,
490 				    uint16_t pkts)
491 {
492 	npi_status_t	rs = NPI_SUCCESS;
493 	npi_handle_t	handle;
494 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
495 	    " ==> nxge_rxdma_cfg_rcr_threshold"));
496 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
497 
498 	rs = npi_rxdma_cfg_rdc_rcr_threshold(handle, channel, pkts);
499 
500 	if (rs != NPI_SUCCESS) {
501 		return (NXGE_ERROR | rs);
502 	}
503 	NXGE_DEBUG_MSG((nxgep, RX2_CTL, " <== nxge_rxdma_cfg_rcr_threshold"));
504 	return (NXGE_OK);
505 }
506 
507 nxge_status_t
508 nxge_rxdma_cfg_rcr_timeout(p_nxge_t nxgep, uint8_t channel,
509 			    uint16_t tout, uint8_t enable)
510 {
511 	npi_status_t	rs = NPI_SUCCESS;
512 	npi_handle_t	handle;
513 	NXGE_DEBUG_MSG((nxgep, RX2_CTL, " ==> nxge_rxdma_cfg_rcr_timeout"));
514 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
515 	if (enable == 0) {
516 		rs = npi_rxdma_cfg_rdc_rcr_timeout_disable(handle, channel);
517 	} else {
518 		rs = npi_rxdma_cfg_rdc_rcr_timeout(handle, channel,
519 		    tout);
520 	}
521 
522 	if (rs != NPI_SUCCESS) {
523 		return (NXGE_ERROR | rs);
524 	}
525 	NXGE_DEBUG_MSG((nxgep, RX2_CTL, " <== nxge_rxdma_cfg_rcr_timeout"));
526 	return (NXGE_OK);
527 }
528 
529 nxge_status_t
530 nxge_enable_rxdma_channel(p_nxge_t nxgep, uint16_t channel,
531     p_rx_rbr_ring_t rbr_p, p_rx_rcr_ring_t rcr_p, p_rx_mbox_t mbox_p)
532 {
533 	npi_handle_t		handle;
534 	rdc_desc_cfg_t 		rdc_desc;
535 	p_rcrcfig_b_t		cfgb_p;
536 	npi_status_t		rs = NPI_SUCCESS;
537 
538 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "==> nxge_enable_rxdma_channel"));
539 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
540 	/*
541 	 * Use configuration data composed at init time.
542 	 * Write to hardware the receive ring configurations.
543 	 */
544 	rdc_desc.mbox_enable = 1;
545 	rdc_desc.mbox_addr = mbox_p->mbox_addr;
546 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
547 	    "==> nxge_enable_rxdma_channel: mboxp $%p($%p)",
548 	    mbox_p->mbox_addr, rdc_desc.mbox_addr));
549 
550 	rdc_desc.rbr_len = rbr_p->rbb_max;
551 	rdc_desc.rbr_addr = rbr_p->rbr_addr;
552 
553 	switch (nxgep->rx_bksize_code) {
554 	case RBR_BKSIZE_4K:
555 		rdc_desc.page_size = SIZE_4KB;
556 		break;
557 	case RBR_BKSIZE_8K:
558 		rdc_desc.page_size = SIZE_8KB;
559 		break;
560 	case RBR_BKSIZE_16K:
561 		rdc_desc.page_size = SIZE_16KB;
562 		break;
563 	case RBR_BKSIZE_32K:
564 		rdc_desc.page_size = SIZE_32KB;
565 		break;
566 	}
567 
568 	rdc_desc.size0 = rbr_p->npi_pkt_buf_size0;
569 	rdc_desc.valid0 = 1;
570 
571 	rdc_desc.size1 = rbr_p->npi_pkt_buf_size1;
572 	rdc_desc.valid1 = 1;
573 
574 	rdc_desc.size2 = rbr_p->npi_pkt_buf_size2;
575 	rdc_desc.valid2 = 1;
576 
577 	rdc_desc.full_hdr = rcr_p->full_hdr_flag;
578 	rdc_desc.offset = rcr_p->sw_priv_hdr_len;
579 
580 	rdc_desc.rcr_len = rcr_p->comp_size;
581 	rdc_desc.rcr_addr = rcr_p->rcr_addr;
582 
583 	cfgb_p = &(rcr_p->rcr_cfgb);
584 	rdc_desc.rcr_threshold = cfgb_p->bits.ldw.pthres;
585 	/* For now, disable this timeout in a guest domain. */
586 	if (isLDOMguest(nxgep)) {
587 		rdc_desc.rcr_timeout = 0;
588 		rdc_desc.rcr_timeout_enable = 0;
589 	} else {
590 		rdc_desc.rcr_timeout = cfgb_p->bits.ldw.timeout;
591 		rdc_desc.rcr_timeout_enable = cfgb_p->bits.ldw.entout;
592 	}
593 
594 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "==> nxge_enable_rxdma_channel: "
595 	    "rbr_len qlen %d pagesize code %d rcr_len %d",
596 	    rdc_desc.rbr_len, rdc_desc.page_size, rdc_desc.rcr_len));
597 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "==> nxge_enable_rxdma_channel: "
598 	    "size 0 %d size 1 %d size 2 %d",
599 	    rbr_p->npi_pkt_buf_size0, rbr_p->npi_pkt_buf_size1,
600 	    rbr_p->npi_pkt_buf_size2));
601 
602 	rs = npi_rxdma_cfg_rdc_ring(handle, rbr_p->rdc, &rdc_desc);
603 	if (rs != NPI_SUCCESS) {
604 		return (NXGE_ERROR | rs);
605 	}
606 
607 	/*
608 	 * Enable the timeout and threshold.
609 	 */
610 	rs = npi_rxdma_cfg_rdc_rcr_threshold(handle, channel,
611 	    rdc_desc.rcr_threshold);
612 	if (rs != NPI_SUCCESS) {
613 		return (NXGE_ERROR | rs);
614 	}
615 
616 	rs = npi_rxdma_cfg_rdc_rcr_timeout(handle, channel,
617 	    rdc_desc.rcr_timeout);
618 	if (rs != NPI_SUCCESS) {
619 		return (NXGE_ERROR | rs);
620 	}
621 
622 	if (!isLDOMguest(nxgep)) {
623 		/* Enable the DMA */
624 		rs = npi_rxdma_cfg_rdc_enable(handle, channel);
625 		if (rs != NPI_SUCCESS) {
626 			return (NXGE_ERROR | rs);
627 		}
628 	}
629 
630 	/* Kick the DMA engine. */
631 	npi_rxdma_rdc_rbr_kick(handle, channel, rbr_p->rbb_max);
632 
633 	if (!isLDOMguest(nxgep)) {
634 		/* Clear the rbr empty bit */
635 		(void) npi_rxdma_channel_rbr_empty_clear(handle, channel);
636 	}
637 
638 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "<== nxge_enable_rxdma_channel"));
639 
640 	return (NXGE_OK);
641 }
642 
643 nxge_status_t
644 nxge_disable_rxdma_channel(p_nxge_t nxgep, uint16_t channel)
645 {
646 	npi_handle_t		handle;
647 	npi_status_t		rs = NPI_SUCCESS;
648 
649 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "==> nxge_disable_rxdma_channel"));
650 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
651 
652 	/* disable the DMA */
653 	rs = npi_rxdma_cfg_rdc_disable(handle, channel);
654 	if (rs != NPI_SUCCESS) {
655 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
656 		    "<== nxge_disable_rxdma_channel:failed (0x%x)",
657 		    rs));
658 		return (NXGE_ERROR | rs);
659 	}
660 
661 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "<== nxge_disable_rxdma_channel"));
662 	return (NXGE_OK);
663 }
664 
665 nxge_status_t
666 nxge_rxdma_channel_rcrflush(p_nxge_t nxgep, uint8_t channel)
667 {
668 	npi_handle_t		handle;
669 	nxge_status_t		status = NXGE_OK;
670 
671 	NXGE_DEBUG_MSG((nxgep, DMA_CTL,
672 	    "<== nxge_init_rxdma_channel_rcrflush"));
673 
674 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
675 	npi_rxdma_rdc_rcr_flush(handle, channel);
676 
677 	NXGE_DEBUG_MSG((nxgep, DMA_CTL,
678 	    "<== nxge_init_rxdma_channel_rcrflsh"));
679 	return (status);
680 
681 }
682 
683 #define	MID_INDEX(l, r) ((r + l + 1) >> 1)
684 
685 #define	TO_LEFT -1
686 #define	TO_RIGHT 1
687 #define	BOTH_RIGHT (TO_RIGHT + TO_RIGHT)
688 #define	BOTH_LEFT (TO_LEFT + TO_LEFT)
689 #define	IN_MIDDLE (TO_RIGHT + TO_LEFT)
690 #define	NO_HINT 0xffffffff
691 
692 /*ARGSUSED*/
693 nxge_status_t
694 nxge_rxbuf_pp_to_vp(p_nxge_t nxgep, p_rx_rbr_ring_t rbr_p,
695 	uint8_t pktbufsz_type, uint64_t *pkt_buf_addr_pp,
696 	uint64_t **pkt_buf_addr_p, uint32_t *bufoffset, uint32_t *msg_index)
697 {
698 	int			bufsize;
699 	uint64_t		pktbuf_pp;
700 	uint64_t 		dvma_addr;
701 	rxring_info_t 		*ring_info;
702 	int 			base_side, end_side;
703 	int 			r_index, l_index, anchor_index;
704 	int 			found, search_done;
705 	uint32_t offset, chunk_size, block_size, page_size_mask;
706 	uint32_t chunk_index, block_index, total_index;
707 	int 			max_iterations, iteration;
708 	rxbuf_index_info_t 	*bufinfo;
709 
710 	NXGE_DEBUG_MSG((nxgep, RX2_CTL, "==> nxge_rxbuf_pp_to_vp"));
711 
712 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
713 	    "==> nxge_rxbuf_pp_to_vp: buf_pp $%p btype %d",
714 	    pkt_buf_addr_pp,
715 	    pktbufsz_type));
716 #if defined(__i386)
717 	pktbuf_pp = (uint64_t)(uint32_t)pkt_buf_addr_pp;
718 #else
719 	pktbuf_pp = (uint64_t)pkt_buf_addr_pp;
720 #endif
721 
722 	switch (pktbufsz_type) {
723 	case 0:
724 		bufsize = rbr_p->pkt_buf_size0;
725 		break;
726 	case 1:
727 		bufsize = rbr_p->pkt_buf_size1;
728 		break;
729 	case 2:
730 		bufsize = rbr_p->pkt_buf_size2;
731 		break;
732 	case RCR_SINGLE_BLOCK:
733 		bufsize = 0;
734 		anchor_index = 0;
735 		break;
736 	default:
737 		return (NXGE_ERROR);
738 	}
739 
740 	if (rbr_p->num_blocks == 1) {
741 		anchor_index = 0;
742 		ring_info = rbr_p->ring_info;
743 		bufinfo = (rxbuf_index_info_t *)ring_info->buffer;
744 		NXGE_DEBUG_MSG((nxgep, RX2_CTL,
745 		    "==> nxge_rxbuf_pp_to_vp: (found, 1 block) "
746 		    "buf_pp $%p btype %d anchor_index %d "
747 		    "bufinfo $%p",
748 		    pkt_buf_addr_pp,
749 		    pktbufsz_type,
750 		    anchor_index,
751 		    bufinfo));
752 
753 		goto found_index;
754 	}
755 
756 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
757 	    "==> nxge_rxbuf_pp_to_vp: "
758 	    "buf_pp $%p btype %d  anchor_index %d",
759 	    pkt_buf_addr_pp,
760 	    pktbufsz_type,
761 	    anchor_index));
762 
763 	ring_info = rbr_p->ring_info;
764 	found = B_FALSE;
765 	bufinfo = (rxbuf_index_info_t *)ring_info->buffer;
766 	iteration = 0;
767 	max_iterations = ring_info->max_iterations;
768 		/*
769 		 * First check if this block has been seen
770 		 * recently. This is indicated by a hint which
771 		 * is initialized when the first buffer of the block
772 		 * is seen. The hint is reset when the last buffer of
773 		 * the block has been processed.
774 		 * As three block sizes are supported, three hints
775 		 * are kept. The idea behind the hints is that once
776 		 * the hardware  uses a block for a buffer  of that
777 		 * size, it will use it exclusively for that size
778 		 * and will use it until it is exhausted. It is assumed
779 		 * that there would a single block being used for the same
780 		 * buffer sizes at any given time.
781 		 */
782 	if (ring_info->hint[pktbufsz_type] != NO_HINT) {
783 		anchor_index = ring_info->hint[pktbufsz_type];
784 		dvma_addr =  bufinfo[anchor_index].dvma_addr;
785 		chunk_size = bufinfo[anchor_index].buf_size;
786 		if ((pktbuf_pp >= dvma_addr) &&
787 		    (pktbuf_pp < (dvma_addr + chunk_size))) {
788 			found = B_TRUE;
789 				/*
790 				 * check if this is the last buffer in the block
791 				 * If so, then reset the hint for the size;
792 				 */
793 
794 			if ((pktbuf_pp + bufsize) >= (dvma_addr + chunk_size))
795 				ring_info->hint[pktbufsz_type] = NO_HINT;
796 		}
797 	}
798 
799 	if (found == B_FALSE) {
800 		NXGE_DEBUG_MSG((nxgep, RX2_CTL,
801 		    "==> nxge_rxbuf_pp_to_vp: (!found)"
802 		    "buf_pp $%p btype %d anchor_index %d",
803 		    pkt_buf_addr_pp,
804 		    pktbufsz_type,
805 		    anchor_index));
806 
807 			/*
808 			 * This is the first buffer of the block of this
809 			 * size. Need to search the whole information
810 			 * array.
811 			 * the search algorithm uses a binary tree search
812 			 * algorithm. It assumes that the information is
813 			 * already sorted with increasing order
814 			 * info[0] < info[1] < info[2]  .... < info[n-1]
815 			 * where n is the size of the information array
816 			 */
817 		r_index = rbr_p->num_blocks - 1;
818 		l_index = 0;
819 		search_done = B_FALSE;
820 		anchor_index = MID_INDEX(r_index, l_index);
821 		while (search_done == B_FALSE) {
822 			if ((r_index == l_index) ||
823 			    (iteration >= max_iterations))
824 				search_done = B_TRUE;
825 			end_side = TO_RIGHT; /* to the right */
826 			base_side = TO_LEFT; /* to the left */
827 			/* read the DVMA address information and sort it */
828 			dvma_addr =  bufinfo[anchor_index].dvma_addr;
829 			chunk_size = bufinfo[anchor_index].buf_size;
830 			NXGE_DEBUG_MSG((nxgep, RX2_CTL,
831 			    "==> nxge_rxbuf_pp_to_vp: (searching)"
832 			    "buf_pp $%p btype %d "
833 			    "anchor_index %d chunk_size %d dvmaaddr $%p",
834 			    pkt_buf_addr_pp,
835 			    pktbufsz_type,
836 			    anchor_index,
837 			    chunk_size,
838 			    dvma_addr));
839 
840 			if (pktbuf_pp >= dvma_addr)
841 				base_side = TO_RIGHT; /* to the right */
842 			if (pktbuf_pp < (dvma_addr + chunk_size))
843 				end_side = TO_LEFT; /* to the left */
844 
845 			switch (base_side + end_side) {
846 			case IN_MIDDLE:
847 				/* found */
848 				found = B_TRUE;
849 				search_done = B_TRUE;
850 				if ((pktbuf_pp + bufsize) <
851 				    (dvma_addr + chunk_size))
852 					ring_info->hint[pktbufsz_type] =
853 					    bufinfo[anchor_index].buf_index;
854 				break;
855 			case BOTH_RIGHT:
856 				/* not found: go to the right */
857 				l_index = anchor_index + 1;
858 				anchor_index = MID_INDEX(r_index, l_index);
859 				break;
860 
861 			case BOTH_LEFT:
862 				/* not found: go to the left */
863 				r_index = anchor_index - 1;
864 				anchor_index = MID_INDEX(r_index, l_index);
865 				break;
866 			default: /* should not come here */
867 				return (NXGE_ERROR);
868 			}
869 			iteration++;
870 		}
871 
872 		NXGE_DEBUG_MSG((nxgep, RX2_CTL,
873 		    "==> nxge_rxbuf_pp_to_vp: (search done)"
874 		    "buf_pp $%p btype %d anchor_index %d",
875 		    pkt_buf_addr_pp,
876 		    pktbufsz_type,
877 		    anchor_index));
878 	}
879 
880 	if (found == B_FALSE) {
881 		NXGE_DEBUG_MSG((nxgep, RX2_CTL,
882 		    "==> nxge_rxbuf_pp_to_vp: (search failed)"
883 		    "buf_pp $%p btype %d anchor_index %d",
884 		    pkt_buf_addr_pp,
885 		    pktbufsz_type,
886 		    anchor_index));
887 		return (NXGE_ERROR);
888 	}
889 
890 found_index:
891 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
892 	    "==> nxge_rxbuf_pp_to_vp: (FOUND1)"
893 	    "buf_pp $%p btype %d bufsize %d anchor_index %d",
894 	    pkt_buf_addr_pp,
895 	    pktbufsz_type,
896 	    bufsize,
897 	    anchor_index));
898 
899 	/* index of the first block in this chunk */
900 	chunk_index = bufinfo[anchor_index].start_index;
901 	dvma_addr =  bufinfo[anchor_index].dvma_addr;
902 	page_size_mask = ring_info->block_size_mask;
903 
904 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
905 	    "==> nxge_rxbuf_pp_to_vp: (FOUND3), get chunk)"
906 	    "buf_pp $%p btype %d bufsize %d "
907 	    "anchor_index %d chunk_index %d dvma $%p",
908 	    pkt_buf_addr_pp,
909 	    pktbufsz_type,
910 	    bufsize,
911 	    anchor_index,
912 	    chunk_index,
913 	    dvma_addr));
914 
915 	offset = pktbuf_pp - dvma_addr; /* offset within the chunk */
916 	block_size = rbr_p->block_size; /* System  block(page) size */
917 
918 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
919 	    "==> nxge_rxbuf_pp_to_vp: (FOUND4), get chunk)"
920 	    "buf_pp $%p btype %d bufsize %d "
921 	    "anchor_index %d chunk_index %d dvma $%p "
922 	    "offset %d block_size %d",
923 	    pkt_buf_addr_pp,
924 	    pktbufsz_type,
925 	    bufsize,
926 	    anchor_index,
927 	    chunk_index,
928 	    dvma_addr,
929 	    offset,
930 	    block_size));
931 
932 	NXGE_DEBUG_MSG((nxgep, RX2_CTL, "==> getting total index"));
933 
934 	block_index = (offset / block_size); /* index within chunk */
935 	total_index = chunk_index + block_index;
936 
937 
938 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
939 	    "==> nxge_rxbuf_pp_to_vp: "
940 	    "total_index %d dvma_addr $%p "
941 	    "offset %d block_size %d "
942 	    "block_index %d ",
943 	    total_index, dvma_addr,
944 	    offset, block_size,
945 	    block_index));
946 #if defined(__i386)
947 	*pkt_buf_addr_p = (uint64_t *)((uint32_t)bufinfo[anchor_index].kaddr +
948 	    (uint32_t)offset);
949 #else
950 	*pkt_buf_addr_p = (uint64_t *)((uint64_t)bufinfo[anchor_index].kaddr +
951 	    (uint64_t)offset);
952 #endif
953 
954 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
955 	    "==> nxge_rxbuf_pp_to_vp: "
956 	    "total_index %d dvma_addr $%p "
957 	    "offset %d block_size %d "
958 	    "block_index %d "
959 	    "*pkt_buf_addr_p $%p",
960 	    total_index, dvma_addr,
961 	    offset, block_size,
962 	    block_index,
963 	    *pkt_buf_addr_p));
964 
965 
966 	*msg_index = total_index;
967 	*bufoffset =  (offset & page_size_mask);
968 
969 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
970 	    "==> nxge_rxbuf_pp_to_vp: get msg index: "
971 	    "msg_index %d bufoffset_index %d",
972 	    *msg_index,
973 	    *bufoffset));
974 
975 	NXGE_DEBUG_MSG((nxgep, RX2_CTL, "<== nxge_rxbuf_pp_to_vp"));
976 
977 	return (NXGE_OK);
978 }
979 
980 /*
981  * used by quick sort (qsort) function
982  * to perform comparison
983  */
984 static int
985 nxge_sort_compare(const void *p1, const void *p2)
986 {
987 
988 	rxbuf_index_info_t *a, *b;
989 
990 	a = (rxbuf_index_info_t *)p1;
991 	b = (rxbuf_index_info_t *)p2;
992 
993 	if (a->dvma_addr > b->dvma_addr)
994 		return (1);
995 	if (a->dvma_addr < b->dvma_addr)
996 		return (-1);
997 	return (0);
998 }
999 
1000 
1001 
1002 /*
1003  * grabbed this sort implementation from common/syscall/avl.c
1004  *
1005  */
1006 /*
1007  * Generic shellsort, from K&R (1st ed, p 58.), somewhat modified.
1008  * v = Ptr to array/vector of objs
1009  * n = # objs in the array
1010  * s = size of each obj (must be multiples of a word size)
1011  * f = ptr to function to compare two objs
1012  *	returns (-1 = less than, 0 = equal, 1 = greater than
1013  */
1014 void
1015 nxge_ksort(caddr_t v, int n, int s, int (*f)())
1016 {
1017 	int g, i, j, ii;
1018 	unsigned int *p1, *p2;
1019 	unsigned int tmp;
1020 
1021 	/* No work to do */
1022 	if (v == NULL || n <= 1)
1023 		return;
1024 	/* Sanity check on arguments */
1025 	ASSERT(((uintptr_t)v & 0x3) == 0 && (s & 0x3) == 0);
1026 	ASSERT(s > 0);
1027 
1028 	for (g = n / 2; g > 0; g /= 2) {
1029 		for (i = g; i < n; i++) {
1030 			for (j = i - g; j >= 0 &&
1031 			    (*f)(v + j * s, v + (j + g) * s) == 1;
1032 			    j -= g) {
1033 				p1 = (unsigned *)(v + j * s);
1034 				p2 = (unsigned *)(v + (j + g) * s);
1035 				for (ii = 0; ii < s / 4; ii++) {
1036 					tmp = *p1;
1037 					*p1++ = *p2;
1038 					*p2++ = tmp;
1039 				}
1040 			}
1041 		}
1042 	}
1043 }
1044 
1045 /*
1046  * Initialize data structures required for rxdma
1047  * buffer dvma->vmem address lookup
1048  */
1049 /*ARGSUSED*/
1050 static nxge_status_t
1051 nxge_rxbuf_index_info_init(p_nxge_t nxgep, p_rx_rbr_ring_t rbrp)
1052 {
1053 
1054 	int index;
1055 	rxring_info_t *ring_info;
1056 	int max_iteration = 0, max_index = 0;
1057 
1058 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "==> nxge_rxbuf_index_info_init"));
1059 
1060 	ring_info = rbrp->ring_info;
1061 	ring_info->hint[0] = NO_HINT;
1062 	ring_info->hint[1] = NO_HINT;
1063 	ring_info->hint[2] = NO_HINT;
1064 	max_index = rbrp->num_blocks;
1065 
1066 		/* read the DVMA address information and sort it */
1067 		/* do init of the information array */
1068 
1069 
1070 	NXGE_DEBUG_MSG((nxgep, DMA2_CTL,
1071 	    " nxge_rxbuf_index_info_init Sort ptrs"));
1072 
1073 		/* sort the array */
1074 	nxge_ksort((void *)ring_info->buffer, max_index,
1075 	    sizeof (rxbuf_index_info_t), nxge_sort_compare);
1076 
1077 
1078 
1079 	for (index = 0; index < max_index; index++) {
1080 		NXGE_DEBUG_MSG((nxgep, DMA2_CTL,
1081 		    " nxge_rxbuf_index_info_init: sorted chunk %d "
1082 		    " ioaddr $%p kaddr $%p size %x",
1083 		    index, ring_info->buffer[index].dvma_addr,
1084 		    ring_info->buffer[index].kaddr,
1085 		    ring_info->buffer[index].buf_size));
1086 	}
1087 
1088 	max_iteration = 0;
1089 	while (max_index >= (1ULL << max_iteration))
1090 		max_iteration++;
1091 	ring_info->max_iterations = max_iteration + 1;
1092 	NXGE_DEBUG_MSG((nxgep, DMA2_CTL,
1093 	    " nxge_rxbuf_index_info_init Find max iter %d",
1094 	    ring_info->max_iterations));
1095 
1096 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "<== nxge_rxbuf_index_info_init"));
1097 	return (NXGE_OK);
1098 }
1099 
1100 /* ARGSUSED */
1101 void
1102 nxge_dump_rcr_entry(p_nxge_t nxgep, p_rcr_entry_t entry_p)
1103 {
1104 #ifdef	NXGE_DEBUG
1105 
1106 	uint32_t bptr;
1107 	uint64_t pp;
1108 
1109 	bptr = entry_p->bits.hdw.pkt_buf_addr;
1110 
1111 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
1112 	    "\trcr entry $%p "
1113 	    "\trcr entry 0x%0llx "
1114 	    "\trcr entry 0x%08x "
1115 	    "\trcr entry 0x%08x "
1116 	    "\tvalue 0x%0llx\n"
1117 	    "\tmulti = %d\n"
1118 	    "\tpkt_type = 0x%x\n"
1119 	    "\tzero_copy = %d\n"
1120 	    "\tnoport = %d\n"
1121 	    "\tpromis = %d\n"
1122 	    "\terror = 0x%04x\n"
1123 	    "\tdcf_err = 0x%01x\n"
1124 	    "\tl2_len = %d\n"
1125 	    "\tpktbufsize = %d\n"
1126 	    "\tpkt_buf_addr = $%p\n"
1127 	    "\tpkt_buf_addr (<< 6) = $%p\n",
1128 	    entry_p,
1129 	    *(int64_t *)entry_p,
1130 	    *(int32_t *)entry_p,
1131 	    *(int32_t *)((char *)entry_p + 32),
1132 	    entry_p->value,
1133 	    entry_p->bits.hdw.multi,
1134 	    entry_p->bits.hdw.pkt_type,
1135 	    entry_p->bits.hdw.zero_copy,
1136 	    entry_p->bits.hdw.noport,
1137 	    entry_p->bits.hdw.promis,
1138 	    entry_p->bits.hdw.error,
1139 	    entry_p->bits.hdw.dcf_err,
1140 	    entry_p->bits.hdw.l2_len,
1141 	    entry_p->bits.hdw.pktbufsz,
1142 	    bptr,
1143 	    entry_p->bits.ldw.pkt_buf_addr));
1144 
1145 	pp = (entry_p->value & RCR_PKT_BUF_ADDR_MASK) <<
1146 	    RCR_PKT_BUF_ADDR_SHIFT;
1147 
1148 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "rcr pp 0x%llx l2 len %d",
1149 	    pp, (*(int64_t *)entry_p >> 40) & 0x3fff));
1150 #endif
1151 }
1152 
1153 void
1154 nxge_rxdma_regs_dump(p_nxge_t nxgep, int rdc)
1155 {
1156 	npi_handle_t		handle;
1157 	rbr_stat_t 		rbr_stat;
1158 	addr44_t 		hd_addr;
1159 	addr44_t 		tail_addr;
1160 	uint16_t 		qlen;
1161 
1162 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
1163 	    "==> nxge_rxdma_regs_dump: rdc channel %d", rdc));
1164 
1165 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
1166 
1167 	/* RBR head */
1168 	hd_addr.addr = 0;
1169 	(void) npi_rxdma_rdc_rbr_head_get(handle, rdc, &hd_addr);
1170 #if defined(__i386)
1171 	printf("nxge_rxdma_regs_dump: got hdptr $%p \n",
1172 	    (void *)(uint32_t)hd_addr.addr);
1173 #else
1174 	printf("nxge_rxdma_regs_dump: got hdptr $%p \n",
1175 	    (void *)hd_addr.addr);
1176 #endif
1177 
1178 	/* RBR stats */
1179 	(void) npi_rxdma_rdc_rbr_stat_get(handle, rdc, &rbr_stat);
1180 	printf("nxge_rxdma_regs_dump: rbr len %d \n", rbr_stat.bits.ldw.qlen);
1181 
1182 	/* RCR tail */
1183 	tail_addr.addr = 0;
1184 	(void) npi_rxdma_rdc_rcr_tail_get(handle, rdc, &tail_addr);
1185 #if defined(__i386)
1186 	printf("nxge_rxdma_regs_dump: got tail ptr $%p \n",
1187 	    (void *)(uint32_t)tail_addr.addr);
1188 #else
1189 	printf("nxge_rxdma_regs_dump: got tail ptr $%p \n",
1190 	    (void *)tail_addr.addr);
1191 #endif
1192 
1193 	/* RCR qlen */
1194 	(void) npi_rxdma_rdc_rcr_qlen_get(handle, rdc, &qlen);
1195 	printf("nxge_rxdma_regs_dump: rcr len %x \n", qlen);
1196 
1197 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
1198 	    "<== nxge_rxdma_regs_dump: rdc rdc %d", rdc));
1199 }
1200 
1201 nxge_status_t
1202 nxge_rxdma_hw_mode(p_nxge_t nxgep, boolean_t enable)
1203 {
1204 	nxge_grp_set_t *set = &nxgep->rx_set;
1205 	nxge_status_t status;
1206 	npi_status_t rs;
1207 	int rdc;
1208 
1209 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
1210 	    "==> nxge_rxdma_hw_mode: mode %d", enable));
1211 
1212 	if (!(nxgep->drv_state & STATE_HW_INITIALIZED)) {
1213 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
1214 		    "<== nxge_rxdma_mode: not initialized"));
1215 		return (NXGE_ERROR);
1216 	}
1217 
1218 	if (nxgep->rx_rbr_rings == 0 || nxgep->rx_rbr_rings->rbr_rings == 0) {
1219 		NXGE_DEBUG_MSG((nxgep, TX_CTL,
1220 		    "<== nxge_tx_port_fatal_err_recover: "
1221 		    "NULL ring pointer(s)"));
1222 		return (NXGE_ERROR);
1223 	}
1224 
1225 	if (set->owned.map == 0) {
1226 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
1227 		    "nxge_rxdma_regs_dump_channels: no channels"));
1228 		return (NULL);
1229 	}
1230 
1231 	for (rdc = 0; rdc < NXGE_MAX_RDCS; rdc++) {
1232 		if ((1 << rdc) & set->owned.map) {
1233 			rx_rbr_ring_t *ring =
1234 			    nxgep->rx_rbr_rings->rbr_rings[rdc];
1235 			npi_handle_t handle = NXGE_DEV_NPI_HANDLE(nxgep);
1236 			if (ring) {
1237 				if (enable) {
1238 					NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
1239 					    "==> nxge_rxdma_hw_mode: "
1240 					    "channel %d (enable)", rdc));
1241 					rs = npi_rxdma_cfg_rdc_enable
1242 					    (handle, rdc);
1243 				} else {
1244 					NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
1245 					    "==> nxge_rxdma_hw_mode: "
1246 					    "channel %d disable)", rdc));
1247 					rs = npi_rxdma_cfg_rdc_disable
1248 					    (handle, rdc);
1249 				}
1250 			}
1251 		}
1252 	}
1253 
1254 	status = ((rs == NPI_SUCCESS) ? NXGE_OK : NXGE_ERROR | rs);
1255 
1256 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
1257 	    "<== nxge_rxdma_hw_mode: status 0x%x", status));
1258 
1259 	return (status);
1260 }
1261 
1262 void
1263 nxge_rxdma_enable_channel(p_nxge_t nxgep, uint16_t channel)
1264 {
1265 	npi_handle_t		handle;
1266 
1267 	NXGE_DEBUG_MSG((nxgep, DMA_CTL,
1268 	    "==> nxge_rxdma_enable_channel: channel %d", channel));
1269 
1270 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
1271 	(void) npi_rxdma_cfg_rdc_enable(handle, channel);
1272 
1273 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "<== nxge_rxdma_enable_channel"));
1274 }
1275 
1276 void
1277 nxge_rxdma_disable_channel(p_nxge_t nxgep, uint16_t channel)
1278 {
1279 	npi_handle_t		handle;
1280 
1281 	NXGE_DEBUG_MSG((nxgep, DMA_CTL,
1282 	    "==> nxge_rxdma_disable_channel: channel %d", channel));
1283 
1284 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
1285 	(void) npi_rxdma_cfg_rdc_disable(handle, channel);
1286 
1287 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "<== nxge_rxdma_disable_channel"));
1288 }
1289 
1290 void
1291 nxge_hw_start_rx(p_nxge_t nxgep)
1292 {
1293 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "==> nxge_hw_start_rx"));
1294 
1295 	(void) nxge_rxdma_hw_mode(nxgep, NXGE_DMA_START);
1296 	(void) nxge_rx_mac_enable(nxgep);
1297 
1298 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "<== nxge_hw_start_rx"));
1299 }
1300 
1301 /*ARGSUSED*/
1302 void
1303 nxge_fixup_rxdma_rings(p_nxge_t nxgep)
1304 {
1305 	nxge_grp_set_t *set = &nxgep->rx_set;
1306 	int rdc;
1307 
1308 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "==> nxge_fixup_rxdma_rings"));
1309 
1310 	if (nxgep->rx_rbr_rings == 0 || nxgep->rx_rbr_rings->rbr_rings == 0) {
1311 		NXGE_DEBUG_MSG((nxgep, TX_CTL,
1312 		    "<== nxge_tx_port_fatal_err_recover: "
1313 		    "NULL ring pointer(s)"));
1314 		return;
1315 	}
1316 
1317 	if (set->owned.map == 0) {
1318 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
1319 		    "nxge_rxdma_regs_dump_channels: no channels"));
1320 		return;
1321 	}
1322 
1323 	for (rdc = 0; rdc < NXGE_MAX_RDCS; rdc++) {
1324 		if ((1 << rdc) & set->owned.map) {
1325 			rx_rbr_ring_t *ring =
1326 			    nxgep->rx_rbr_rings->rbr_rings[rdc];
1327 			if (ring) {
1328 				nxge_rxdma_hw_stop(nxgep, rdc);
1329 				NXGE_DEBUG_MSG((nxgep, RX_CTL,
1330 				    "==> nxge_fixup_rxdma_rings: "
1331 				    "channel %d ring $%px",
1332 				    rdc, ring));
1333 				(void) nxge_rxdma_fixup_channel
1334 				    (nxgep, rdc, rdc);
1335 			}
1336 		}
1337 	}
1338 
1339 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "<== nxge_fixup_rxdma_rings"));
1340 }
1341 
1342 void
1343 nxge_rxdma_fix_channel(p_nxge_t nxgep, uint16_t channel)
1344 {
1345 	int		i;
1346 
1347 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "==> nxge_rxdma_fix_channel"));
1348 	i = nxge_rxdma_get_ring_index(nxgep, channel);
1349 	if (i < 0) {
1350 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
1351 		    "<== nxge_rxdma_fix_channel: no entry found"));
1352 		return;
1353 	}
1354 
1355 	nxge_rxdma_fixup_channel(nxgep, channel, i);
1356 
1357 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "<== nxge_rxdma_fix_channel"));
1358 }
1359 
1360 void
1361 nxge_rxdma_fixup_channel(p_nxge_t nxgep, uint16_t channel, int entry)
1362 {
1363 	int			ndmas;
1364 	p_rx_rbr_rings_t 	rx_rbr_rings;
1365 	p_rx_rbr_ring_t		*rbr_rings;
1366 	p_rx_rcr_rings_t 	rx_rcr_rings;
1367 	p_rx_rcr_ring_t		*rcr_rings;
1368 	p_rx_mbox_areas_t 	rx_mbox_areas_p;
1369 	p_rx_mbox_t		*rx_mbox_p;
1370 	p_nxge_dma_pool_t	dma_buf_poolp;
1371 	p_nxge_dma_pool_t	dma_cntl_poolp;
1372 	p_rx_rbr_ring_t 	rbrp;
1373 	p_rx_rcr_ring_t 	rcrp;
1374 	p_rx_mbox_t 		mboxp;
1375 	p_nxge_dma_common_t 	dmap;
1376 	nxge_status_t		status = NXGE_OK;
1377 
1378 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "==> nxge_rxdma_fixup_channel"));
1379 
1380 	(void) nxge_rxdma_stop_channel(nxgep, channel);
1381 
1382 	dma_buf_poolp = nxgep->rx_buf_pool_p;
1383 	dma_cntl_poolp = nxgep->rx_cntl_pool_p;
1384 
1385 	if (!dma_buf_poolp->buf_allocated || !dma_cntl_poolp->buf_allocated) {
1386 		NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
1387 		    "<== nxge_rxdma_fixup_channel: buf not allocated"));
1388 		return;
1389 	}
1390 
1391 	ndmas = dma_buf_poolp->ndmas;
1392 	if (!ndmas) {
1393 		NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
1394 		    "<== nxge_rxdma_fixup_channel: no dma allocated"));
1395 		return;
1396 	}
1397 
1398 	rx_rbr_rings = nxgep->rx_rbr_rings;
1399 	rx_rcr_rings = nxgep->rx_rcr_rings;
1400 	rbr_rings = rx_rbr_rings->rbr_rings;
1401 	rcr_rings = rx_rcr_rings->rcr_rings;
1402 	rx_mbox_areas_p = nxgep->rx_mbox_areas_p;
1403 	rx_mbox_p = rx_mbox_areas_p->rxmbox_areas;
1404 
1405 	/* Reinitialize the receive block and completion rings */
1406 	rbrp = (p_rx_rbr_ring_t)rbr_rings[entry],
1407 	    rcrp = (p_rx_rcr_ring_t)rcr_rings[entry],
1408 	    mboxp = (p_rx_mbox_t)rx_mbox_p[entry];
1409 
1410 
1411 	rbrp->rbr_wr_index = (rbrp->rbb_max - 1);
1412 	rbrp->rbr_rd_index = 0;
1413 	rcrp->comp_rd_index = 0;
1414 	rcrp->comp_wt_index = 0;
1415 
1416 	dmap = (p_nxge_dma_common_t)&rcrp->rcr_desc;
1417 	bzero((caddr_t)dmap->kaddrp, dmap->alength);
1418 
1419 	status = nxge_rxdma_start_channel(nxgep, channel,
1420 	    rbrp, rcrp, mboxp);
1421 	if (status != NXGE_OK) {
1422 		goto nxge_rxdma_fixup_channel_fail;
1423 	}
1424 	if (status != NXGE_OK) {
1425 		goto nxge_rxdma_fixup_channel_fail;
1426 	}
1427 
1428 nxge_rxdma_fixup_channel_fail:
1429 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
1430 	    "==> nxge_rxdma_fixup_channel: failed (0x%08x)", status));
1431 
1432 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "<== nxge_rxdma_fixup_channel"));
1433 }
1434 
1435 /*
1436  * Convert an absolute RDC number to a Receive Buffer Ring index.  That is,
1437  * map <channel> to an index into nxgep->rx_rbr_rings.
1438  * (device ring index -> port ring index)
1439  */
1440 int
1441 nxge_rxdma_get_ring_index(p_nxge_t nxgep, uint16_t channel)
1442 {
1443 	int			i, ndmas;
1444 	uint16_t		rdc;
1445 	p_rx_rbr_rings_t	rx_rbr_rings;
1446 	p_rx_rbr_ring_t		*rbr_rings;
1447 
1448 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
1449 	    "==> nxge_rxdma_get_ring_index: channel %d", channel));
1450 
1451 	rx_rbr_rings = nxgep->rx_rbr_rings;
1452 	if (rx_rbr_rings == NULL) {
1453 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
1454 		    "<== nxge_rxdma_get_ring_index: NULL ring pointer"));
1455 		return (-1);
1456 	}
1457 	ndmas = rx_rbr_rings->ndmas;
1458 	if (!ndmas) {
1459 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
1460 		    "<== nxge_rxdma_get_ring_index: no channel"));
1461 		return (-1);
1462 	}
1463 
1464 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
1465 	    "==> nxge_rxdma_get_ring_index (ndmas %d)", ndmas));
1466 
1467 	rbr_rings = rx_rbr_rings->rbr_rings;
1468 	for (i = 0; i < ndmas; i++) {
1469 		rdc = rbr_rings[i]->rdc;
1470 		if (channel == rdc) {
1471 			NXGE_DEBUG_MSG((nxgep, RX_CTL,
1472 			    "==> nxge_rxdma_get_rbr_ring: channel %d "
1473 			    "(index %d) ring %d", channel, i, rbr_rings[i]));
1474 			return (i);
1475 		}
1476 	}
1477 
1478 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
1479 	    "<== nxge_rxdma_get_rbr_ring_index: not found"));
1480 
1481 	return (-1);
1482 }
1483 
1484 p_rx_rbr_ring_t
1485 nxge_rxdma_get_rbr_ring(p_nxge_t nxgep, uint16_t channel)
1486 {
1487 	nxge_grp_set_t *set = &nxgep->rx_set;
1488 	nxge_channel_t rdc;
1489 
1490 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
1491 	    "==> nxge_rxdma_get_rbr_ring: channel %d", channel));
1492 
1493 	if (nxgep->rx_rbr_rings == 0 || nxgep->rx_rbr_rings->rbr_rings == 0) {
1494 		NXGE_DEBUG_MSG((nxgep, TX_CTL,
1495 		    "<== nxge_rxdma_get_rbr_ring: "
1496 		    "NULL ring pointer(s)"));
1497 		return (NULL);
1498 	}
1499 
1500 	if (set->owned.map == 0) {
1501 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
1502 		    "<== nxge_rxdma_get_rbr_ring: no channels"));
1503 		return (NULL);
1504 	}
1505 
1506 	for (rdc = 0; rdc < NXGE_MAX_RDCS; rdc++) {
1507 		if ((1 << rdc) & set->owned.map) {
1508 			rx_rbr_ring_t *ring =
1509 			    nxgep->rx_rbr_rings->rbr_rings[rdc];
1510 			if (ring) {
1511 				if (channel == ring->rdc) {
1512 					NXGE_DEBUG_MSG((nxgep, RX_CTL,
1513 					    "==> nxge_rxdma_get_rbr_ring: "
1514 					    "channel %d ring $%p", rdc, ring));
1515 					return (ring);
1516 				}
1517 			}
1518 		}
1519 	}
1520 
1521 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
1522 	    "<== nxge_rxdma_get_rbr_ring: not found"));
1523 
1524 	return (NULL);
1525 }
1526 
1527 p_rx_rcr_ring_t
1528 nxge_rxdma_get_rcr_ring(p_nxge_t nxgep, uint16_t channel)
1529 {
1530 	nxge_grp_set_t *set = &nxgep->rx_set;
1531 	nxge_channel_t rdc;
1532 
1533 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
1534 	    "==> nxge_rxdma_get_rcr_ring: channel %d", channel));
1535 
1536 	if (nxgep->rx_rcr_rings == 0 || nxgep->rx_rcr_rings->rcr_rings == 0) {
1537 		NXGE_DEBUG_MSG((nxgep, TX_CTL,
1538 		    "<== nxge_rxdma_get_rcr_ring: "
1539 		    "NULL ring pointer(s)"));
1540 		return (NULL);
1541 	}
1542 
1543 	if (set->owned.map == 0) {
1544 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
1545 		    "<== nxge_rxdma_get_rbr_ring: no channels"));
1546 		return (NULL);
1547 	}
1548 
1549 	for (rdc = 0; rdc < NXGE_MAX_RDCS; rdc++) {
1550 		if ((1 << rdc) & set->owned.map) {
1551 			rx_rcr_ring_t *ring =
1552 			    nxgep->rx_rcr_rings->rcr_rings[rdc];
1553 			if (ring) {
1554 				if (channel == ring->rdc) {
1555 					NXGE_DEBUG_MSG((nxgep, RX_CTL,
1556 					    "==> nxge_rxdma_get_rcr_ring: "
1557 					    "channel %d ring $%p", rdc, ring));
1558 					return (ring);
1559 				}
1560 			}
1561 		}
1562 	}
1563 
1564 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
1565 	    "<== nxge_rxdma_get_rcr_ring: not found"));
1566 
1567 	return (NULL);
1568 }
1569 
1570 /*
1571  * Static functions start here.
1572  */
1573 static p_rx_msg_t
1574 nxge_allocb(size_t size, uint32_t pri, p_nxge_dma_common_t dmabuf_p)
1575 {
1576 	p_rx_msg_t nxge_mp 		= NULL;
1577 	p_nxge_dma_common_t		dmamsg_p;
1578 	uchar_t 			*buffer;
1579 
1580 	nxge_mp = KMEM_ZALLOC(sizeof (rx_msg_t), KM_NOSLEEP);
1581 	if (nxge_mp == NULL) {
1582 		NXGE_ERROR_MSG((NULL, NXGE_ERR_CTL,
1583 		    "Allocation of a rx msg failed."));
1584 		goto nxge_allocb_exit;
1585 	}
1586 
1587 	nxge_mp->use_buf_pool = B_FALSE;
1588 	if (dmabuf_p) {
1589 		nxge_mp->use_buf_pool = B_TRUE;
1590 		dmamsg_p = (p_nxge_dma_common_t)&nxge_mp->buf_dma;
1591 		*dmamsg_p = *dmabuf_p;
1592 		dmamsg_p->nblocks = 1;
1593 		dmamsg_p->block_size = size;
1594 		dmamsg_p->alength = size;
1595 		buffer = (uchar_t *)dmabuf_p->kaddrp;
1596 
1597 		dmabuf_p->kaddrp = (void *)
1598 		    ((char *)dmabuf_p->kaddrp + size);
1599 		dmabuf_p->ioaddr_pp = (void *)
1600 		    ((char *)dmabuf_p->ioaddr_pp + size);
1601 		dmabuf_p->alength -= size;
1602 		dmabuf_p->offset += size;
1603 		dmabuf_p->dma_cookie.dmac_laddress += size;
1604 		dmabuf_p->dma_cookie.dmac_size -= size;
1605 
1606 	} else {
1607 		buffer = KMEM_ALLOC(size, KM_NOSLEEP);
1608 		if (buffer == NULL) {
1609 			NXGE_ERROR_MSG((NULL, NXGE_ERR_CTL,
1610 			    "Allocation of a receive page failed."));
1611 			goto nxge_allocb_fail1;
1612 		}
1613 	}
1614 
1615 	nxge_mp->rx_mblk_p = desballoc(buffer, size, pri, &nxge_mp->freeb);
1616 	if (nxge_mp->rx_mblk_p == NULL) {
1617 		NXGE_ERROR_MSG((NULL, NXGE_ERR_CTL, "desballoc failed."));
1618 		goto nxge_allocb_fail2;
1619 	}
1620 
1621 	nxge_mp->buffer = buffer;
1622 	nxge_mp->block_size = size;
1623 	nxge_mp->freeb.free_func = (void (*)())nxge_freeb;
1624 	nxge_mp->freeb.free_arg = (caddr_t)nxge_mp;
1625 	nxge_mp->ref_cnt = 1;
1626 	nxge_mp->free = B_TRUE;
1627 	nxge_mp->rx_use_bcopy = B_FALSE;
1628 
1629 	atomic_inc_32(&nxge_mblks_pending);
1630 
1631 	goto nxge_allocb_exit;
1632 
1633 nxge_allocb_fail2:
1634 	if (!nxge_mp->use_buf_pool) {
1635 		KMEM_FREE(buffer, size);
1636 	}
1637 
1638 nxge_allocb_fail1:
1639 	KMEM_FREE(nxge_mp, sizeof (rx_msg_t));
1640 	nxge_mp = NULL;
1641 
1642 nxge_allocb_exit:
1643 	return (nxge_mp);
1644 }
1645 
1646 p_mblk_t
1647 nxge_dupb(p_rx_msg_t nxge_mp, uint_t offset, size_t size)
1648 {
1649 	p_mblk_t mp;
1650 
1651 	NXGE_DEBUG_MSG((NULL, MEM_CTL, "==> nxge_dupb"));
1652 	NXGE_DEBUG_MSG((NULL, MEM_CTL, "nxge_mp = $%p "
1653 	    "offset = 0x%08X "
1654 	    "size = 0x%08X",
1655 	    nxge_mp, offset, size));
1656 
1657 	mp = desballoc(&nxge_mp->buffer[offset], size,
1658 	    0, &nxge_mp->freeb);
1659 	if (mp == NULL) {
1660 		NXGE_DEBUG_MSG((NULL, RX_CTL, "desballoc failed"));
1661 		goto nxge_dupb_exit;
1662 	}
1663 	atomic_inc_32(&nxge_mp->ref_cnt);
1664 
1665 
1666 nxge_dupb_exit:
1667 	NXGE_DEBUG_MSG((NULL, MEM_CTL, "<== nxge_dupb mp = $%p",
1668 	    nxge_mp));
1669 	return (mp);
1670 }
1671 
1672 p_mblk_t
1673 nxge_dupb_bcopy(p_rx_msg_t nxge_mp, uint_t offset, size_t size)
1674 {
1675 	p_mblk_t mp;
1676 	uchar_t *dp;
1677 
1678 	mp = allocb(size + NXGE_RXBUF_EXTRA, 0);
1679 	if (mp == NULL) {
1680 		NXGE_DEBUG_MSG((NULL, RX_CTL, "desballoc failed"));
1681 		goto nxge_dupb_bcopy_exit;
1682 	}
1683 	dp = mp->b_rptr = mp->b_rptr + NXGE_RXBUF_EXTRA;
1684 	bcopy((void *)&nxge_mp->buffer[offset], dp, size);
1685 	mp->b_wptr = dp + size;
1686 
1687 nxge_dupb_bcopy_exit:
1688 	NXGE_DEBUG_MSG((NULL, MEM_CTL, "<== nxge_dupb mp = $%p",
1689 	    nxge_mp));
1690 	return (mp);
1691 }
1692 
1693 void nxge_post_page(p_nxge_t nxgep, p_rx_rbr_ring_t rx_rbr_p,
1694 	p_rx_msg_t rx_msg_p);
1695 
1696 void
1697 nxge_post_page(p_nxge_t nxgep, p_rx_rbr_ring_t rx_rbr_p, p_rx_msg_t rx_msg_p)
1698 {
1699 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "==> nxge_post_page"));
1700 
1701 	/* Reuse this buffer */
1702 	rx_msg_p->free = B_FALSE;
1703 	rx_msg_p->cur_usage_cnt = 0;
1704 	rx_msg_p->max_usage_cnt = 0;
1705 	rx_msg_p->pkt_buf_size = 0;
1706 
1707 	if (rx_rbr_p->rbr_use_bcopy) {
1708 		rx_msg_p->rx_use_bcopy = B_FALSE;
1709 		atomic_dec_32(&rx_rbr_p->rbr_consumed);
1710 	}
1711 
1712 	/*
1713 	 * Get the rbr header pointer and its offset index.
1714 	 */
1715 	MUTEX_ENTER(&rx_rbr_p->post_lock);
1716 	rx_rbr_p->rbr_wr_index =  ((rx_rbr_p->rbr_wr_index + 1) &
1717 	    rx_rbr_p->rbr_wrap_mask);
1718 	rx_rbr_p->rbr_desc_vp[rx_rbr_p->rbr_wr_index] = rx_msg_p->shifted_addr;
1719 	MUTEX_EXIT(&rx_rbr_p->post_lock);
1720 	npi_rxdma_rdc_rbr_kick(NXGE_DEV_NPI_HANDLE(nxgep),
1721 	    rx_rbr_p->rdc, 1);
1722 
1723 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
1724 	    "<== nxge_post_page (channel %d post_next_index %d)",
1725 	    rx_rbr_p->rdc, rx_rbr_p->rbr_wr_index));
1726 
1727 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "<== nxge_post_page"));
1728 }
1729 
1730 void
1731 nxge_freeb(p_rx_msg_t rx_msg_p)
1732 {
1733 	size_t size;
1734 	uchar_t *buffer = NULL;
1735 	int ref_cnt;
1736 	boolean_t free_state = B_FALSE;
1737 
1738 	rx_rbr_ring_t *ring = rx_msg_p->rx_rbr_p;
1739 
1740 	NXGE_DEBUG_MSG((NULL, MEM2_CTL, "==> nxge_freeb"));
1741 	NXGE_DEBUG_MSG((NULL, MEM2_CTL,
1742 	    "nxge_freeb:rx_msg_p = $%p (block pending %d)",
1743 	    rx_msg_p, nxge_mblks_pending));
1744 
1745 	/*
1746 	 * First we need to get the free state, then
1747 	 * atomic decrement the reference count to prevent
1748 	 * the race condition with the interrupt thread that
1749 	 * is processing a loaned up buffer block.
1750 	 */
1751 	free_state = rx_msg_p->free;
1752 	ref_cnt = atomic_add_32_nv(&rx_msg_p->ref_cnt, -1);
1753 	if (!ref_cnt) {
1754 		atomic_dec_32(&nxge_mblks_pending);
1755 		buffer = rx_msg_p->buffer;
1756 		size = rx_msg_p->block_size;
1757 		NXGE_DEBUG_MSG((NULL, MEM2_CTL, "nxge_freeb: "
1758 		    "will free: rx_msg_p = $%p (block pending %d)",
1759 		    rx_msg_p, nxge_mblks_pending));
1760 
1761 		if (!rx_msg_p->use_buf_pool) {
1762 			KMEM_FREE(buffer, size);
1763 		}
1764 
1765 		KMEM_FREE(rx_msg_p, sizeof (rx_msg_t));
1766 
1767 		if (ring) {
1768 			/*
1769 			 * Decrement the receive buffer ring's reference
1770 			 * count, too.
1771 			 */
1772 			atomic_dec_32(&ring->rbr_ref_cnt);
1773 
1774 			/*
1775 			 * Free the receive buffer ring, if
1776 			 * 1. all the receive buffers have been freed
1777 			 * 2. and we are in the proper state (that is,
1778 			 *    we are not UNMAPPING).
1779 			 */
1780 			if (ring->rbr_ref_cnt == 0 &&
1781 			    ring->rbr_state == RBR_UNMAPPED) {
1782 				/*
1783 				 * Free receive data buffers,
1784 				 * buffer index information
1785 				 * (rxring_info) and
1786 				 * the message block ring.
1787 				 */
1788 				NXGE_DEBUG_MSG((NULL, RX_CTL,
1789 				    "nxge_freeb:rx_msg_p = $%p "
1790 				    "(block pending %d) free buffers",
1791 				    rx_msg_p, nxge_mblks_pending));
1792 				nxge_rxdma_databuf_free(ring);
1793 				if (ring->ring_info) {
1794 					KMEM_FREE(ring->ring_info,
1795 					    sizeof (rxring_info_t));
1796 				}
1797 
1798 				if (ring->rx_msg_ring) {
1799 					KMEM_FREE(ring->rx_msg_ring,
1800 					    ring->tnblocks *
1801 					    sizeof (p_rx_msg_t));
1802 				}
1803 				KMEM_FREE(ring, sizeof (*ring));
1804 			}
1805 		}
1806 		return;
1807 	}
1808 
1809 	/*
1810 	 * Repost buffer.
1811 	 */
1812 	if (free_state && (ref_cnt == 1) && ring) {
1813 		NXGE_DEBUG_MSG((NULL, RX_CTL,
1814 		    "nxge_freeb: post page $%p:", rx_msg_p));
1815 		if (ring->rbr_state == RBR_POSTING)
1816 			nxge_post_page(rx_msg_p->nxgep, ring, rx_msg_p);
1817 	}
1818 
1819 	NXGE_DEBUG_MSG((NULL, MEM2_CTL, "<== nxge_freeb"));
1820 }
1821 
1822 uint_t
1823 nxge_rx_intr(void *arg1, void *arg2)
1824 {
1825 	p_nxge_ldv_t		ldvp = (p_nxge_ldv_t)arg1;
1826 	p_nxge_t		nxgep = (p_nxge_t)arg2;
1827 	p_nxge_ldg_t		ldgp;
1828 	uint8_t			channel;
1829 	npi_handle_t		handle;
1830 	rx_dma_ctl_stat_t	cs;
1831 	p_rx_rcr_ring_t		rcr_ring;
1832 	mblk_t			*mp;
1833 
1834 	if (ldvp == NULL) {
1835 		NXGE_DEBUG_MSG((NULL, INT_CTL,
1836 		    "<== nxge_rx_intr: arg2 $%p arg1 $%p",
1837 		    nxgep, ldvp));
1838 		return (DDI_INTR_CLAIMED);
1839 	}
1840 
1841 	if (arg2 == NULL || (void *)ldvp->nxgep != arg2) {
1842 		nxgep = ldvp->nxgep;
1843 	}
1844 
1845 	if ((!(nxgep->drv_state & STATE_HW_INITIALIZED)) ||
1846 	    (nxgep->nxge_mac_state != NXGE_MAC_STARTED)) {
1847 		NXGE_DEBUG_MSG((nxgep, INT_CTL,
1848 		    "<== nxge_rx_intr: interface not started or intialized"));
1849 		return (DDI_INTR_CLAIMED);
1850 	}
1851 
1852 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
1853 	    "==> nxge_rx_intr: arg2 $%p arg1 $%p",
1854 	    nxgep, ldvp));
1855 
1856 	/*
1857 	 * Get the PIO handle.
1858 	 */
1859 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
1860 
1861 	/*
1862 	 * Get the ring to enable us to process packets.
1863 	 */
1864 	rcr_ring = nxgep->rx_rcr_rings->rcr_rings[ldvp->vdma_index];
1865 
1866 	/*
1867 	 * The RCR ring lock must be held when packets
1868 	 * are being processed and the hardware registers are
1869 	 * being read or written to prevent race condition
1870 	 * among the interrupt thread, the polling thread
1871 	 * (will cause fatal errors such as rcrincon bit set)
1872 	 * and the setting of the poll_flag.
1873 	 */
1874 	MUTEX_ENTER(&rcr_ring->lock);
1875 
1876 	/*
1877 	 * Get the control and status for this channel.
1878 	 */
1879 	channel = ldvp->channel;
1880 	ldgp = ldvp->ldgp;
1881 
1882 	if (!isLDOMguest(nxgep) && (!nxgep->rx_channel_started[channel])) {
1883 		NXGE_DEBUG_MSG((nxgep, INT_CTL,
1884 		    "<== nxge_rx_intr: channel is not started"));
1885 
1886 		/*
1887 		 * We received an interrupt before the ring is started.
1888 		 */
1889 		RXDMA_REG_READ64(handle, RX_DMA_CTL_STAT_REG, channel,
1890 		    &cs.value);
1891 		cs.value &= RX_DMA_CTL_STAT_WR1C;
1892 		cs.bits.hdw.mex = 1;
1893 		RXDMA_REG_WRITE64(handle, RX_DMA_CTL_STAT_REG, channel,
1894 		    cs.value);
1895 
1896 		/*
1897 		 * Rearm this logical group if this is a single device
1898 		 * group.
1899 		 */
1900 		if (ldgp->nldvs == 1) {
1901 			if (isLDOMguest(nxgep)) {
1902 				nxge_hio_ldgimgn(nxgep, ldgp);
1903 			} else {
1904 				ldgimgm_t mgm;
1905 
1906 				mgm.value = 0;
1907 				mgm.bits.ldw.arm = 1;
1908 				mgm.bits.ldw.timer = ldgp->ldg_timer;
1909 
1910 				NXGE_REG_WR64(handle,
1911 				    LDGIMGN_REG + LDSV_OFFSET(ldgp->ldg),
1912 				    mgm.value);
1913 			}
1914 		}
1915 		MUTEX_EXIT(&rcr_ring->lock);
1916 		return (DDI_INTR_CLAIMED);
1917 	}
1918 
1919 	ASSERT(rcr_ring->ldgp == ldgp);
1920 	ASSERT(rcr_ring->ldvp == ldvp);
1921 
1922 	RXDMA_REG_READ64(handle, RX_DMA_CTL_STAT_REG, channel, &cs.value);
1923 
1924 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "==> nxge_rx_intr:channel %d "
1925 	    "cs 0x%016llx rcrto 0x%x rcrthres %x",
1926 	    channel,
1927 	    cs.value,
1928 	    cs.bits.hdw.rcrto,
1929 	    cs.bits.hdw.rcrthres));
1930 
1931 	mp = nxge_rx_pkts_vring(nxgep, ldvp->vdma_index, cs);
1932 
1933 	/* error events. */
1934 	if (cs.value & RX_DMA_CTL_STAT_ERROR) {
1935 		(void) nxge_rx_err_evnts(nxgep, channel, cs);
1936 	}
1937 
1938 	/*
1939 	 * Enable the mailbox update interrupt if we want
1940 	 * to use mailbox. We probably don't need to use
1941 	 * mailbox as it only saves us one pio read.
1942 	 * Also write 1 to rcrthres and rcrto to clear
1943 	 * these two edge triggered bits.
1944 	 */
1945 	cs.value &= RX_DMA_CTL_STAT_WR1C;
1946 	cs.bits.hdw.mex = rcr_ring->poll_flag ? 0 : 1;
1947 	RXDMA_REG_WRITE64(handle, RX_DMA_CTL_STAT_REG, channel,
1948 	    cs.value);
1949 
1950 	/*
1951 	 * If the polling mode is enabled, disable the interrupt.
1952 	 */
1953 	if (rcr_ring->poll_flag) {
1954 		NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
1955 		    "==> nxge_rx_intr: rdc %d ldgp $%p ldvp $%p "
1956 		    "(disabling interrupts)", channel, ldgp, ldvp));
1957 		/*
1958 		 * Disarm this logical group if this is a single device
1959 		 * group.
1960 		 */
1961 		if (ldgp->nldvs == 1) {
1962 			ldgimgm_t mgm;
1963 			mgm.value = 0;
1964 			mgm.bits.ldw.arm = 0;
1965 			NXGE_REG_WR64(handle,
1966 			    LDGIMGN_REG + LDSV_OFFSET(ldgp->ldg), mgm.value);
1967 		}
1968 	} else {
1969 		/*
1970 		 * Rearm this logical group if this is a single device
1971 		 * group.
1972 		 */
1973 		if (ldgp->nldvs == 1) {
1974 			if (isLDOMguest(nxgep)) {
1975 				nxge_hio_ldgimgn(nxgep, ldgp);
1976 			} else {
1977 				ldgimgm_t mgm;
1978 
1979 				mgm.value = 0;
1980 				mgm.bits.ldw.arm = 1;
1981 				mgm.bits.ldw.timer = ldgp->ldg_timer;
1982 
1983 				NXGE_REG_WR64(handle,
1984 				    LDGIMGN_REG + LDSV_OFFSET(ldgp->ldg),
1985 				    mgm.value);
1986 			}
1987 		}
1988 
1989 		NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
1990 		    "==> nxge_rx_intr: rdc %d ldgp $%p "
1991 		    "exiting ISR (and call mac_rx_ring)", channel, ldgp));
1992 	}
1993 	MUTEX_EXIT(&rcr_ring->lock);
1994 
1995 	if (mp) {
1996 		if (!isLDOMguest(nxgep))
1997 			mac_rx_ring(nxgep->mach, rcr_ring->rcr_mac_handle, mp,
1998 			    rcr_ring->rcr_gen_num);
1999 #if defined(sun4v)
2000 		else {			/* isLDOMguest(nxgep) */
2001 			nxge_hio_data_t *nhd = (nxge_hio_data_t *)
2002 			    nxgep->nxge_hw_p->hio;
2003 			nx_vio_fp_t *vio = &nhd->hio.vio;
2004 
2005 			if (vio->cb.vio_net_rx_cb) {
2006 				(*vio->cb.vio_net_rx_cb)
2007 				    (nxgep->hio_vr->vhp, mp);
2008 			}
2009 		}
2010 #endif
2011 	}
2012 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "<== nxge_rx_intr: DDI_INTR_CLAIMED"));
2013 	return (DDI_INTR_CLAIMED);
2014 }
2015 
2016 /*
2017  * Process the packets received in the specified logical device
2018  * and pass up a chain of message blocks to the upper layer.
2019  * The RCR ring lock must be held before calling this function.
2020  */
2021 static mblk_t *
2022 nxge_rx_pkts_vring(p_nxge_t nxgep, uint_t vindex, rx_dma_ctl_stat_t cs)
2023 {
2024 	p_mblk_t		mp;
2025 	p_rx_rcr_ring_t		rcrp;
2026 
2027 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "==> nxge_rx_pkts_vring"));
2028 	rcrp = nxgep->rx_rcr_rings->rcr_rings[vindex];
2029 
2030 	NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
2031 	    "==> nxge_rx_pkts_vring: (calling nxge_rx_pkts)rdc %d "
2032 	    "rcr_mac_handle $%p ", rcrp->rdc, rcrp->rcr_mac_handle));
2033 	if ((mp = nxge_rx_pkts(nxgep, rcrp, cs, -1)) == NULL) {
2034 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
2035 		    "<== nxge_rx_pkts_vring: no mp"));
2036 		return (NULL);
2037 	}
2038 
2039 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "==> nxge_rx_pkts_vring: $%p",
2040 	    mp));
2041 
2042 #ifdef  NXGE_DEBUG
2043 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
2044 		    "==> nxge_rx_pkts_vring:calling mac_rx "
2045 		    "LEN %d mp $%p mp->b_cont $%p mp->b_next $%p rcrp $%p "
2046 		    "mac_handle $%p",
2047 		    mp->b_wptr - mp->b_rptr,
2048 		    mp, mp->b_cont, mp->b_next,
2049 		    rcrp, rcrp->rcr_mac_handle));
2050 
2051 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
2052 		    "==> nxge_rx_pkts_vring: dump packets "
2053 		    "(mp $%p b_rptr $%p b_wptr $%p):\n %s",
2054 		    mp,
2055 		    mp->b_rptr,
2056 		    mp->b_wptr,
2057 		    nxge_dump_packet((char *)mp->b_rptr,
2058 		    mp->b_wptr - mp->b_rptr)));
2059 		if (mp->b_cont) {
2060 			NXGE_DEBUG_MSG((nxgep, RX_CTL,
2061 			    "==> nxge_rx_pkts_vring: dump b_cont packets "
2062 			    "(mp->b_cont $%p b_rptr $%p b_wptr $%p):\n %s",
2063 			    mp->b_cont,
2064 			    mp->b_cont->b_rptr,
2065 			    mp->b_cont->b_wptr,
2066 			    nxge_dump_packet((char *)mp->b_cont->b_rptr,
2067 			    mp->b_cont->b_wptr - mp->b_cont->b_rptr)));
2068 		}
2069 		if (mp->b_next) {
2070 			NXGE_DEBUG_MSG((nxgep, RX_CTL,
2071 			    "==> nxge_rx_pkts_vring: dump next packets "
2072 			    "(b_rptr $%p): %s",
2073 			    mp->b_next->b_rptr,
2074 			    nxge_dump_packet((char *)mp->b_next->b_rptr,
2075 			    mp->b_next->b_wptr - mp->b_next->b_rptr)));
2076 		}
2077 #endif
2078 	NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
2079 	    "<== nxge_rx_pkts_vring: returning rdc %d rcr_mac_handle $%p ",
2080 	    rcrp->rdc, rcrp->rcr_mac_handle));
2081 
2082 	return (mp);
2083 }
2084 
2085 
2086 /*
2087  * This routine is the main packet receive processing function.
2088  * It gets the packet type, error code, and buffer related
2089  * information from the receive completion entry.
2090  * How many completion entries to process is based on the number of packets
2091  * queued by the hardware, a hardware maintained tail pointer
2092  * and a configurable receive packet count.
2093  *
2094  * A chain of message blocks will be created as result of processing
2095  * the completion entries. This chain of message blocks will be returned and
2096  * a hardware control status register will be updated with the number of
2097  * packets were removed from the hardware queue.
2098  *
2099  * The RCR ring lock is held when entering this function.
2100  */
2101 static mblk_t *
2102 nxge_rx_pkts(p_nxge_t nxgep, p_rx_rcr_ring_t rcr_p, rx_dma_ctl_stat_t cs,
2103     int bytes_to_pickup)
2104 {
2105 	npi_handle_t		handle;
2106 	uint8_t			channel;
2107 	uint32_t		comp_rd_index;
2108 	p_rcr_entry_t		rcr_desc_rd_head_p;
2109 	p_rcr_entry_t		rcr_desc_rd_head_pp;
2110 	p_mblk_t		nmp, mp_cont, head_mp, *tail_mp;
2111 	uint16_t		qlen, nrcr_read, npkt_read;
2112 	uint32_t		qlen_hw;
2113 	boolean_t		multi;
2114 	rcrcfig_b_t		rcr_cfg_b;
2115 	int			totallen = 0;
2116 #if defined(_BIG_ENDIAN)
2117 	npi_status_t		rs = NPI_SUCCESS;
2118 #endif
2119 
2120 	NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL, "==> nxge_rx_pkts: "
2121 	    "channel %d", rcr_p->rdc));
2122 
2123 	if (!(nxgep->drv_state & STATE_HW_INITIALIZED)) {
2124 		return (NULL);
2125 	}
2126 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
2127 	channel = rcr_p->rdc;
2128 
2129 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
2130 	    "==> nxge_rx_pkts: START: rcr channel %d "
2131 	    "head_p $%p head_pp $%p  index %d ",
2132 	    channel, rcr_p->rcr_desc_rd_head_p,
2133 	    rcr_p->rcr_desc_rd_head_pp,
2134 	    rcr_p->comp_rd_index));
2135 
2136 
2137 #if !defined(_BIG_ENDIAN)
2138 	qlen = RXDMA_REG_READ32(handle, RCRSTAT_A_REG, channel) & 0xffff;
2139 #else
2140 	rs = npi_rxdma_rdc_rcr_qlen_get(handle, channel, &qlen);
2141 	if (rs != NPI_SUCCESS) {
2142 		NXGE_DEBUG_MSG((nxgep, RX_CTL, "==> nxge_rx_pkts: "
2143 		"channel %d, get qlen failed 0x%08x",
2144 		    channel, rs));
2145 		return (NULL);
2146 	}
2147 #endif
2148 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "==> nxge_rx_pkts:rcr channel %d "
2149 	    "qlen %d", channel, qlen));
2150 
2151 
2152 
2153 	if (!qlen) {
2154 		NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
2155 		    "==> nxge_rx_pkts:rcr channel %d "
2156 		    "qlen %d (no pkts)", channel, qlen));
2157 
2158 		return (NULL);
2159 	}
2160 
2161 	comp_rd_index = rcr_p->comp_rd_index;
2162 
2163 	rcr_desc_rd_head_p = rcr_p->rcr_desc_rd_head_p;
2164 	rcr_desc_rd_head_pp = rcr_p->rcr_desc_rd_head_pp;
2165 	nrcr_read = npkt_read = 0;
2166 
2167 	/*
2168 	 * Number of packets queued
2169 	 * (The jumbo or multi packet will be counted as only one
2170 	 *  packets and it may take up more than one completion entry).
2171 	 */
2172 	qlen_hw = (qlen < nxge_max_rx_pkts) ?
2173 	    qlen : nxge_max_rx_pkts;
2174 	head_mp = NULL;
2175 	tail_mp = &head_mp;
2176 	nmp = mp_cont = NULL;
2177 	multi = B_FALSE;
2178 
2179 	while (qlen_hw) {
2180 
2181 #ifdef NXGE_DEBUG
2182 		nxge_dump_rcr_entry(nxgep, rcr_desc_rd_head_p);
2183 #endif
2184 		/*
2185 		 * Process one completion ring entry.
2186 		 */
2187 		nxge_receive_packet(nxgep,
2188 		    rcr_p, rcr_desc_rd_head_p, &multi, &nmp, &mp_cont);
2189 
2190 		/*
2191 		 * message chaining modes
2192 		 */
2193 		if (nmp) {
2194 			nmp->b_next = NULL;
2195 			if (!multi && !mp_cont) { /* frame fits a partition */
2196 				*tail_mp = nmp;
2197 				tail_mp = &nmp->b_next;
2198 				totallen += MBLKL(nmp);
2199 				nmp = NULL;
2200 			} else if (multi && !mp_cont) { /* first segment */
2201 				*tail_mp = nmp;
2202 				tail_mp = &nmp->b_cont;
2203 				totallen += MBLKL(nmp);
2204 			} else if (multi && mp_cont) {	/* mid of multi segs */
2205 				*tail_mp = mp_cont;
2206 				tail_mp = &mp_cont->b_cont;
2207 				totallen += MBLKL(mp_cont);
2208 			} else if (!multi && mp_cont) { /* last segment */
2209 				*tail_mp = mp_cont;
2210 				tail_mp = &nmp->b_next;
2211 				totallen += MBLKL(mp_cont);
2212 				nmp = NULL;
2213 			}
2214 		}
2215 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
2216 		    "==> nxge_rx_pkts: loop: rcr channel %d "
2217 		    "before updating: multi %d "
2218 		    "nrcr_read %d "
2219 		    "npk read %d "
2220 		    "head_pp $%p  index %d ",
2221 		    channel,
2222 		    multi,
2223 		    nrcr_read, npkt_read, rcr_desc_rd_head_pp,
2224 		    comp_rd_index));
2225 
2226 		if (!multi) {
2227 			qlen_hw--;
2228 			npkt_read++;
2229 		}
2230 
2231 		/*
2232 		 * Update the next read entry.
2233 		 */
2234 		comp_rd_index = NEXT_ENTRY(comp_rd_index,
2235 		    rcr_p->comp_wrap_mask);
2236 
2237 		rcr_desc_rd_head_p = NEXT_ENTRY_PTR(rcr_desc_rd_head_p,
2238 		    rcr_p->rcr_desc_first_p,
2239 		    rcr_p->rcr_desc_last_p);
2240 
2241 		nrcr_read++;
2242 
2243 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
2244 		    "<== nxge_rx_pkts: (SAM, process one packet) "
2245 		    "nrcr_read %d",
2246 		    nrcr_read));
2247 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
2248 		    "==> nxge_rx_pkts: loop: rcr channel %d "
2249 		    "multi %d "
2250 		    "nrcr_read %d "
2251 		    "npk read %d "
2252 		    "head_pp $%p  index %d ",
2253 		    channel,
2254 		    multi,
2255 		    nrcr_read, npkt_read, rcr_desc_rd_head_pp,
2256 		    comp_rd_index));
2257 
2258 		if ((bytes_to_pickup != -1) &&
2259 		    (totallen >= bytes_to_pickup)) {
2260 			break;
2261 		}
2262 	}
2263 
2264 	rcr_p->rcr_desc_rd_head_pp = rcr_desc_rd_head_pp;
2265 	rcr_p->comp_rd_index = comp_rd_index;
2266 	rcr_p->rcr_desc_rd_head_p = rcr_desc_rd_head_p;
2267 	if ((nxgep->intr_timeout != rcr_p->intr_timeout) ||
2268 	    (nxgep->intr_threshold != rcr_p->intr_threshold)) {
2269 
2270 		rcr_p->intr_timeout = (nxgep->intr_timeout <
2271 		    NXGE_RDC_RCR_TIMEOUT_MIN) ? NXGE_RDC_RCR_TIMEOUT_MIN :
2272 		    nxgep->intr_timeout;
2273 
2274 		rcr_p->intr_threshold = (nxgep->intr_threshold <
2275 		    NXGE_RDC_RCR_THRESHOLD_MIN) ? NXGE_RDC_RCR_THRESHOLD_MIN :
2276 		    nxgep->intr_threshold;
2277 
2278 		rcr_cfg_b.value = 0x0ULL;
2279 		rcr_cfg_b.bits.ldw.entout = 1;
2280 		rcr_cfg_b.bits.ldw.timeout = rcr_p->intr_timeout;
2281 		rcr_cfg_b.bits.ldw.pthres = rcr_p->intr_threshold;
2282 
2283 		RXDMA_REG_WRITE64(handle, RCRCFIG_B_REG,
2284 		    channel, rcr_cfg_b.value);
2285 	}
2286 
2287 	cs.bits.ldw.pktread = npkt_read;
2288 	cs.bits.ldw.ptrread = nrcr_read;
2289 	RXDMA_REG_WRITE64(handle, RX_DMA_CTL_STAT_REG,
2290 	    channel, cs.value);
2291 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
2292 	    "==> nxge_rx_pkts: EXIT: rcr channel %d "
2293 	    "head_pp $%p  index %016llx ",
2294 	    channel,
2295 	    rcr_p->rcr_desc_rd_head_pp,
2296 	    rcr_p->comp_rd_index));
2297 	/*
2298 	 * Update RCR buffer pointer read and number of packets
2299 	 * read.
2300 	 */
2301 
2302 	NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL, "<== nxge_rx_pkts: return"
2303 	    "channel %d", rcr_p->rdc));
2304 
2305 	return (head_mp);
2306 }
2307 
2308 void
2309 nxge_receive_packet(p_nxge_t nxgep,
2310     p_rx_rcr_ring_t rcr_p, p_rcr_entry_t rcr_desc_rd_head_p,
2311     boolean_t *multi_p, mblk_t **mp, mblk_t **mp_cont)
2312 {
2313 	p_mblk_t		nmp = NULL;
2314 	uint64_t		multi;
2315 	uint64_t		dcf_err;
2316 	uint8_t			channel;
2317 
2318 	boolean_t		first_entry = B_TRUE;
2319 	boolean_t		is_tcp_udp = B_FALSE;
2320 	boolean_t		buffer_free = B_FALSE;
2321 	boolean_t		error_send_up = B_FALSE;
2322 	uint8_t			error_type;
2323 	uint16_t		l2_len;
2324 	uint16_t		skip_len;
2325 	uint8_t			pktbufsz_type;
2326 	uint64_t		rcr_entry;
2327 	uint64_t		*pkt_buf_addr_pp;
2328 	uint64_t		*pkt_buf_addr_p;
2329 	uint32_t		buf_offset;
2330 	uint32_t		bsize;
2331 	uint32_t		error_disp_cnt;
2332 	uint32_t		msg_index;
2333 	p_rx_rbr_ring_t		rx_rbr_p;
2334 	p_rx_msg_t 		*rx_msg_ring_p;
2335 	p_rx_msg_t		rx_msg_p;
2336 	uint16_t		sw_offset_bytes = 0, hdr_size = 0;
2337 	nxge_status_t		status = NXGE_OK;
2338 	boolean_t		is_valid = B_FALSE;
2339 	p_nxge_rx_ring_stats_t	rdc_stats;
2340 	uint32_t		bytes_read;
2341 	uint64_t		pkt_type;
2342 	uint64_t		frag;
2343 	boolean_t		pkt_too_long_err = B_FALSE;
2344 #ifdef	NXGE_DEBUG
2345 	int			dump_len;
2346 #endif
2347 	NXGE_DEBUG_MSG((nxgep, RX2_CTL, "==> nxge_receive_packet"));
2348 	first_entry = (*mp == NULL) ? B_TRUE : B_FALSE;
2349 
2350 	rcr_entry = *((uint64_t *)rcr_desc_rd_head_p);
2351 
2352 	multi = (rcr_entry & RCR_MULTI_MASK);
2353 	dcf_err = (rcr_entry & RCR_DCF_ERROR_MASK);
2354 	pkt_type = (rcr_entry & RCR_PKT_TYPE_MASK);
2355 
2356 	error_type = ((rcr_entry & RCR_ERROR_MASK) >> RCR_ERROR_SHIFT);
2357 	frag = (rcr_entry & RCR_FRAG_MASK);
2358 
2359 	l2_len = ((rcr_entry & RCR_L2_LEN_MASK) >> RCR_L2_LEN_SHIFT);
2360 
2361 	pktbufsz_type = ((rcr_entry & RCR_PKTBUFSZ_MASK) >>
2362 	    RCR_PKTBUFSZ_SHIFT);
2363 #if defined(__i386)
2364 	pkt_buf_addr_pp = (uint64_t *)(uint32_t)((rcr_entry &
2365 	    RCR_PKT_BUF_ADDR_MASK) << RCR_PKT_BUF_ADDR_SHIFT);
2366 #else
2367 	pkt_buf_addr_pp = (uint64_t *)((rcr_entry & RCR_PKT_BUF_ADDR_MASK) <<
2368 	    RCR_PKT_BUF_ADDR_SHIFT);
2369 #endif
2370 
2371 	channel = rcr_p->rdc;
2372 
2373 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
2374 	    "==> nxge_receive_packet: entryp $%p entry 0x%0llx "
2375 	    "pkt_buf_addr_pp $%p l2_len %d multi 0x%llx "
2376 	    "error_type 0x%x pkt_type 0x%x  "
2377 	    "pktbufsz_type %d ",
2378 	    rcr_desc_rd_head_p,
2379 	    rcr_entry, pkt_buf_addr_pp, l2_len,
2380 	    multi,
2381 	    error_type,
2382 	    pkt_type,
2383 	    pktbufsz_type));
2384 
2385 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
2386 	    "==> nxge_receive_packet: entryp $%p entry 0x%0llx "
2387 	    "pkt_buf_addr_pp $%p l2_len %d multi 0x%llx "
2388 	    "error_type 0x%x pkt_type 0x%x ", rcr_desc_rd_head_p,
2389 	    rcr_entry, pkt_buf_addr_pp, l2_len,
2390 	    multi,
2391 	    error_type,
2392 	    pkt_type));
2393 
2394 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
2395 	    "==> (rbr) nxge_receive_packet: entry 0x%0llx "
2396 	    "full pkt_buf_addr_pp $%p l2_len %d",
2397 	    rcr_entry, pkt_buf_addr_pp, l2_len));
2398 
2399 	/* get the stats ptr */
2400 	rdc_stats = rcr_p->rdc_stats;
2401 
2402 	if (!l2_len) {
2403 
2404 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
2405 		    "<== nxge_receive_packet: failed: l2 length is 0."));
2406 		return;
2407 	}
2408 
2409 	/*
2410 	 * Software workaround for BMAC hardware limitation that allows
2411 	 * maxframe size of 1526, instead of 1522 for non-jumbo and 0x2406
2412 	 * instead of 0x2400 for jumbo.
2413 	 */
2414 	if (l2_len > nxgep->mac.maxframesize) {
2415 		pkt_too_long_err = B_TRUE;
2416 	}
2417 
2418 	/* Hardware sends us 4 bytes of CRC as no stripping is done.  */
2419 	l2_len -= ETHERFCSL;
2420 
2421 	/* shift 6 bits to get the full io address */
2422 #if defined(__i386)
2423 	pkt_buf_addr_pp = (uint64_t *)((uint32_t)pkt_buf_addr_pp <<
2424 	    RCR_PKT_BUF_ADDR_SHIFT_FULL);
2425 #else
2426 	pkt_buf_addr_pp = (uint64_t *)((uint64_t)pkt_buf_addr_pp <<
2427 	    RCR_PKT_BUF_ADDR_SHIFT_FULL);
2428 #endif
2429 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
2430 	    "==> (rbr) nxge_receive_packet: entry 0x%0llx "
2431 	    "full pkt_buf_addr_pp $%p l2_len %d",
2432 	    rcr_entry, pkt_buf_addr_pp, l2_len));
2433 
2434 	rx_rbr_p = rcr_p->rx_rbr_p;
2435 	rx_msg_ring_p = rx_rbr_p->rx_msg_ring;
2436 
2437 	if (first_entry) {
2438 		hdr_size = (rcr_p->full_hdr_flag ? RXDMA_HDR_SIZE_FULL :
2439 		    RXDMA_HDR_SIZE_DEFAULT);
2440 
2441 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
2442 		    "==> nxge_receive_packet: first entry 0x%016llx "
2443 		    "pkt_buf_addr_pp $%p l2_len %d hdr %d",
2444 		    rcr_entry, pkt_buf_addr_pp, l2_len,
2445 		    hdr_size));
2446 	}
2447 
2448 	MUTEX_ENTER(&rx_rbr_p->lock);
2449 
2450 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
2451 	    "==> (rbr 1) nxge_receive_packet: entry 0x%0llx "
2452 	    "full pkt_buf_addr_pp $%p l2_len %d",
2453 	    rcr_entry, pkt_buf_addr_pp, l2_len));
2454 
2455 	/*
2456 	 * Packet buffer address in the completion entry points
2457 	 * to the starting buffer address (offset 0).
2458 	 * Use the starting buffer address to locate the corresponding
2459 	 * kernel address.
2460 	 */
2461 	status = nxge_rxbuf_pp_to_vp(nxgep, rx_rbr_p,
2462 	    pktbufsz_type, pkt_buf_addr_pp, &pkt_buf_addr_p,
2463 	    &buf_offset,
2464 	    &msg_index);
2465 
2466 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
2467 	    "==> (rbr 2) 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 	if (status != NXGE_OK) {
2472 		MUTEX_EXIT(&rx_rbr_p->lock);
2473 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
2474 		    "<== nxge_receive_packet: found vaddr failed %d",
2475 		    status));
2476 		return;
2477 	}
2478 
2479 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
2480 	    "==> (rbr 3) nxge_receive_packet: entry 0x%0llx "
2481 	    "full pkt_buf_addr_pp $%p l2_len %d",
2482 	    rcr_entry, pkt_buf_addr_pp, l2_len));
2483 
2484 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
2485 	    "==> (rbr 4 msgindex %d) nxge_receive_packet: entry 0x%0llx "
2486 	    "full pkt_buf_addr_pp $%p l2_len %d",
2487 	    msg_index, rcr_entry, pkt_buf_addr_pp, l2_len));
2488 
2489 	rx_msg_p = rx_msg_ring_p[msg_index];
2490 
2491 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
2492 	    "==> (rbr 4 msgindex %d) nxge_receive_packet: entry 0x%0llx "
2493 	    "full pkt_buf_addr_pp $%p l2_len %d",
2494 	    msg_index, rcr_entry, pkt_buf_addr_pp, l2_len));
2495 
2496 	switch (pktbufsz_type) {
2497 	case RCR_PKTBUFSZ_0:
2498 		bsize = rx_rbr_p->pkt_buf_size0_bytes;
2499 		NXGE_DEBUG_MSG((nxgep, RX2_CTL,
2500 		    "==> nxge_receive_packet: 0 buf %d", bsize));
2501 		break;
2502 	case RCR_PKTBUFSZ_1:
2503 		bsize = rx_rbr_p->pkt_buf_size1_bytes;
2504 		NXGE_DEBUG_MSG((nxgep, RX2_CTL,
2505 		    "==> nxge_receive_packet: 1 buf %d", bsize));
2506 		break;
2507 	case RCR_PKTBUFSZ_2:
2508 		bsize = rx_rbr_p->pkt_buf_size2_bytes;
2509 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
2510 		    "==> nxge_receive_packet: 2 buf %d", bsize));
2511 		break;
2512 	case RCR_SINGLE_BLOCK:
2513 		bsize = rx_msg_p->block_size;
2514 		NXGE_DEBUG_MSG((nxgep, RX2_CTL,
2515 		    "==> nxge_receive_packet: single %d", bsize));
2516 
2517 		break;
2518 	default:
2519 		MUTEX_EXIT(&rx_rbr_p->lock);
2520 		return;
2521 	}
2522 
2523 	DMA_COMMON_SYNC_OFFSET(rx_msg_p->buf_dma,
2524 	    (buf_offset + sw_offset_bytes),
2525 	    (hdr_size + l2_len),
2526 	    DDI_DMA_SYNC_FORCPU);
2527 
2528 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
2529 	    "==> nxge_receive_packet: after first dump:usage count"));
2530 
2531 	if (rx_msg_p->cur_usage_cnt == 0) {
2532 		if (rx_rbr_p->rbr_use_bcopy) {
2533 			atomic_inc_32(&rx_rbr_p->rbr_consumed);
2534 			if (rx_rbr_p->rbr_consumed <
2535 			    rx_rbr_p->rbr_threshold_hi) {
2536 				if (rx_rbr_p->rbr_threshold_lo == 0 ||
2537 				    ((rx_rbr_p->rbr_consumed >=
2538 				    rx_rbr_p->rbr_threshold_lo) &&
2539 				    (rx_rbr_p->rbr_bufsize_type >=
2540 				    pktbufsz_type))) {
2541 					rx_msg_p->rx_use_bcopy = B_TRUE;
2542 				}
2543 			} else {
2544 				rx_msg_p->rx_use_bcopy = B_TRUE;
2545 			}
2546 		}
2547 		NXGE_DEBUG_MSG((nxgep, RX2_CTL,
2548 		    "==> nxge_receive_packet: buf %d (new block) ",
2549 		    bsize));
2550 
2551 		rx_msg_p->pkt_buf_size_code = pktbufsz_type;
2552 		rx_msg_p->pkt_buf_size = bsize;
2553 		rx_msg_p->cur_usage_cnt = 1;
2554 		if (pktbufsz_type == RCR_SINGLE_BLOCK) {
2555 			NXGE_DEBUG_MSG((nxgep, RX2_CTL,
2556 			    "==> nxge_receive_packet: buf %d "
2557 			    "(single block) ",
2558 			    bsize));
2559 			/*
2560 			 * Buffer can be reused once the free function
2561 			 * is called.
2562 			 */
2563 			rx_msg_p->max_usage_cnt = 1;
2564 			buffer_free = B_TRUE;
2565 		} else {
2566 			rx_msg_p->max_usage_cnt = rx_msg_p->block_size/bsize;
2567 			if (rx_msg_p->max_usage_cnt == 1) {
2568 				buffer_free = B_TRUE;
2569 			}
2570 		}
2571 	} else {
2572 		rx_msg_p->cur_usage_cnt++;
2573 		if (rx_msg_p->cur_usage_cnt == rx_msg_p->max_usage_cnt) {
2574 			buffer_free = B_TRUE;
2575 		}
2576 	}
2577 
2578 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
2579 	    "msgbuf index = %d l2len %d bytes usage %d max_usage %d ",
2580 	    msg_index, l2_len,
2581 	    rx_msg_p->cur_usage_cnt, rx_msg_p->max_usage_cnt));
2582 
2583 	if ((error_type) || (dcf_err) || (pkt_too_long_err)) {
2584 		rdc_stats->ierrors++;
2585 		if (dcf_err) {
2586 			rdc_stats->dcf_err++;
2587 #ifdef	NXGE_DEBUG
2588 			if (!rdc_stats->dcf_err) {
2589 				NXGE_DEBUG_MSG((nxgep, RX_CTL,
2590 				"nxge_receive_packet: channel %d dcf_err rcr"
2591 				" 0x%llx", channel, rcr_entry));
2592 			}
2593 #endif
2594 			NXGE_FM_REPORT_ERROR(nxgep, nxgep->mac.portnum, NULL,
2595 			    NXGE_FM_EREPORT_RDMC_DCF_ERR);
2596 		} else if (pkt_too_long_err) {
2597 			rdc_stats->pkt_too_long_err++;
2598 			NXGE_DEBUG_MSG((nxgep, RX_CTL, " nxge_receive_packet:"
2599 			    " channel %d packet length [%d] > "
2600 			    "maxframesize [%d]", channel, l2_len + ETHERFCSL,
2601 			    nxgep->mac.maxframesize));
2602 		} else {
2603 				/* Update error stats */
2604 			error_disp_cnt = NXGE_ERROR_SHOW_MAX;
2605 			rdc_stats->errlog.compl_err_type = error_type;
2606 
2607 			switch (error_type) {
2608 			/*
2609 			 * Do not send FMA ereport for RCR_L2_ERROR and
2610 			 * RCR_L4_CSUM_ERROR because most likely they indicate
2611 			 * back pressure rather than HW failures.
2612 			 */
2613 			case RCR_L2_ERROR:
2614 				rdc_stats->l2_err++;
2615 				if (rdc_stats->l2_err <
2616 				    error_disp_cnt) {
2617 					NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
2618 					    " nxge_receive_packet:"
2619 					    " channel %d RCR L2_ERROR",
2620 					    channel));
2621 				}
2622 				break;
2623 			case RCR_L4_CSUM_ERROR:
2624 				error_send_up = B_TRUE;
2625 				rdc_stats->l4_cksum_err++;
2626 				if (rdc_stats->l4_cksum_err <
2627 				    error_disp_cnt) {
2628 					NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
2629 					    " nxge_receive_packet:"
2630 					    " channel %d"
2631 					    " RCR L4_CSUM_ERROR", channel));
2632 				}
2633 				break;
2634 			/*
2635 			 * Do not send FMA ereport for RCR_FFLP_SOFT_ERROR and
2636 			 * RCR_ZCP_SOFT_ERROR because they reflect the same
2637 			 * FFLP and ZCP errors that have been reported by
2638 			 * nxge_fflp.c and nxge_zcp.c.
2639 			 */
2640 			case RCR_FFLP_SOFT_ERROR:
2641 				error_send_up = B_TRUE;
2642 				rdc_stats->fflp_soft_err++;
2643 				if (rdc_stats->fflp_soft_err <
2644 				    error_disp_cnt) {
2645 					NXGE_ERROR_MSG((nxgep,
2646 					    NXGE_ERR_CTL,
2647 					    " nxge_receive_packet:"
2648 					    " channel %d"
2649 					    " RCR FFLP_SOFT_ERROR", channel));
2650 				}
2651 				break;
2652 			case RCR_ZCP_SOFT_ERROR:
2653 				error_send_up = B_TRUE;
2654 				rdc_stats->fflp_soft_err++;
2655 				if (rdc_stats->zcp_soft_err <
2656 				    error_disp_cnt)
2657 					NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
2658 					    " nxge_receive_packet: Channel %d"
2659 					    " RCR ZCP_SOFT_ERROR", channel));
2660 				break;
2661 			default:
2662 				rdc_stats->rcr_unknown_err++;
2663 				if (rdc_stats->rcr_unknown_err
2664 				    < error_disp_cnt) {
2665 					NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
2666 					    " nxge_receive_packet: Channel %d"
2667 					    " RCR entry 0x%llx error 0x%x",
2668 					    rcr_entry, channel, error_type));
2669 				}
2670 				break;
2671 			}
2672 		}
2673 
2674 		/*
2675 		 * Update and repost buffer block if max usage
2676 		 * count is reached.
2677 		 */
2678 		if (error_send_up == B_FALSE) {
2679 			atomic_inc_32(&rx_msg_p->ref_cnt);
2680 			if (buffer_free == B_TRUE) {
2681 				rx_msg_p->free = B_TRUE;
2682 			}
2683 
2684 			MUTEX_EXIT(&rx_rbr_p->lock);
2685 			nxge_freeb(rx_msg_p);
2686 			return;
2687 		}
2688 	}
2689 
2690 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
2691 	    "==> nxge_receive_packet: DMA sync second "));
2692 
2693 	bytes_read = rcr_p->rcvd_pkt_bytes;
2694 	skip_len = sw_offset_bytes + hdr_size;
2695 	if (!rx_msg_p->rx_use_bcopy) {
2696 		/*
2697 		 * For loaned up buffers, the driver reference count
2698 		 * will be incremented first and then the free state.
2699 		 */
2700 		if ((nmp = nxge_dupb(rx_msg_p, buf_offset, bsize)) != NULL) {
2701 			if (first_entry) {
2702 				nmp->b_rptr = &nmp->b_rptr[skip_len];
2703 				if (l2_len < bsize - skip_len) {
2704 					nmp->b_wptr = &nmp->b_rptr[l2_len];
2705 				} else {
2706 					nmp->b_wptr = &nmp->b_rptr[bsize
2707 					    - skip_len];
2708 				}
2709 			} else {
2710 				if (l2_len - bytes_read < bsize) {
2711 					nmp->b_wptr =
2712 					    &nmp->b_rptr[l2_len - bytes_read];
2713 				} else {
2714 					nmp->b_wptr = &nmp->b_rptr[bsize];
2715 				}
2716 			}
2717 		}
2718 	} else {
2719 		if (first_entry) {
2720 			nmp = nxge_dupb_bcopy(rx_msg_p, buf_offset + skip_len,
2721 			    l2_len < bsize - skip_len ?
2722 			    l2_len : bsize - skip_len);
2723 		} else {
2724 			nmp = nxge_dupb_bcopy(rx_msg_p, buf_offset,
2725 			    l2_len - bytes_read < bsize ?
2726 			    l2_len - bytes_read : bsize);
2727 		}
2728 	}
2729 	if (nmp != NULL) {
2730 		if (first_entry) {
2731 			/*
2732 			 * Jumbo packets may be received with more than one
2733 			 * buffer, increment ipackets for the first entry only.
2734 			 */
2735 			rdc_stats->ipackets++;
2736 
2737 			/* Update ibytes for kstat. */
2738 			rdc_stats->ibytes += skip_len
2739 			    + l2_len < bsize ? l2_len : bsize;
2740 			/*
2741 			 * Update the number of bytes read so far for the
2742 			 * current frame.
2743 			 */
2744 			bytes_read  = nmp->b_wptr - nmp->b_rptr;
2745 		} else {
2746 			rdc_stats->ibytes += l2_len - bytes_read < bsize ?
2747 			    l2_len - bytes_read : bsize;
2748 			bytes_read += nmp->b_wptr - nmp->b_rptr;
2749 		}
2750 
2751 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
2752 		    "==> nxge_receive_packet after dupb: "
2753 		    "rbr consumed %d "
2754 		    "pktbufsz_type %d "
2755 		    "nmp $%p rptr $%p wptr $%p "
2756 		    "buf_offset %d bzise %d l2_len %d skip_len %d",
2757 		    rx_rbr_p->rbr_consumed,
2758 		    pktbufsz_type,
2759 		    nmp, nmp->b_rptr, nmp->b_wptr,
2760 		    buf_offset, bsize, l2_len, skip_len));
2761 	} else {
2762 		cmn_err(CE_WARN, "!nxge_receive_packet: "
2763 		    "update stats (error)");
2764 		atomic_inc_32(&rx_msg_p->ref_cnt);
2765 		if (buffer_free == B_TRUE) {
2766 			rx_msg_p->free = B_TRUE;
2767 		}
2768 		MUTEX_EXIT(&rx_rbr_p->lock);
2769 		nxge_freeb(rx_msg_p);
2770 		return;
2771 	}
2772 
2773 	if (buffer_free == B_TRUE) {
2774 		rx_msg_p->free = B_TRUE;
2775 	}
2776 
2777 	is_valid = (nmp != NULL);
2778 
2779 	rcr_p->rcvd_pkt_bytes = bytes_read;
2780 
2781 	MUTEX_EXIT(&rx_rbr_p->lock);
2782 
2783 	if (rx_msg_p->free && rx_msg_p->rx_use_bcopy) {
2784 		atomic_inc_32(&rx_msg_p->ref_cnt);
2785 		nxge_freeb(rx_msg_p);
2786 	}
2787 
2788 	if (is_valid) {
2789 		nmp->b_cont = NULL;
2790 		if (first_entry) {
2791 			*mp = nmp;
2792 			*mp_cont = NULL;
2793 		} else {
2794 			*mp_cont = nmp;
2795 		}
2796 	}
2797 
2798 	/*
2799 	 * ERROR, FRAG and PKT_TYPE are only reported in the first entry.
2800 	 * If a packet is not fragmented and no error bit is set, then
2801 	 * L4 checksum is OK.
2802 	 */
2803 
2804 	if (is_valid && !multi) {
2805 		/*
2806 		 * If the checksum flag nxge_chksum_offload
2807 		 * is 1, TCP and UDP packets can be sent
2808 		 * up with good checksum. If the checksum flag
2809 		 * is set to 0, checksum reporting will apply to
2810 		 * TCP packets only (workaround for a hardware bug).
2811 		 * If the checksum flag nxge_cksum_offload is
2812 		 * greater than 1, both TCP and UDP packets
2813 		 * will not be reported its hardware checksum results.
2814 		 */
2815 		if (nxge_cksum_offload == 1) {
2816 			is_tcp_udp = ((pkt_type == RCR_PKT_IS_TCP ||
2817 			    pkt_type == RCR_PKT_IS_UDP) ?
2818 			    B_TRUE: B_FALSE);
2819 		} else if (!nxge_cksum_offload) {
2820 			/* TCP checksum only. */
2821 			is_tcp_udp = ((pkt_type == RCR_PKT_IS_TCP) ?
2822 			    B_TRUE: B_FALSE);
2823 		}
2824 
2825 		NXGE_DEBUG_MSG((nxgep, RX_CTL, "==> nxge_receive_packet: "
2826 		    "is_valid 0x%x multi 0x%llx pkt %d frag %d error %d",
2827 		    is_valid, multi, is_tcp_udp, frag, error_type));
2828 
2829 		if (is_tcp_udp && !frag && !error_type) {
2830 			(void) hcksum_assoc(nmp, NULL, NULL, 0, 0, 0, 0,
2831 			    HCK_FULLCKSUM_OK | HCK_FULLCKSUM, 0);
2832 			NXGE_DEBUG_MSG((nxgep, RX_CTL,
2833 			    "==> nxge_receive_packet: Full tcp/udp cksum "
2834 			    "is_valid 0x%x multi 0x%llx pkt %d frag %d "
2835 			    "error %d",
2836 			    is_valid, multi, is_tcp_udp, frag, error_type));
2837 		}
2838 	}
2839 
2840 	NXGE_DEBUG_MSG((nxgep, RX2_CTL,
2841 	    "==> nxge_receive_packet: *mp 0x%016llx", *mp));
2842 
2843 	*multi_p = (multi == RCR_MULTI_MASK);
2844 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "<== nxge_receive_packet: "
2845 	    "multi %d nmp 0x%016llx *mp 0x%016llx *mp_cont 0x%016llx",
2846 	    *multi_p, nmp, *mp, *mp_cont));
2847 }
2848 
2849 /*
2850  * Enable polling for a ring. Interrupt for the ring is disabled when
2851  * the nxge interrupt comes (see nxge_rx_intr).
2852  */
2853 int
2854 nxge_enable_poll(void *arg)
2855 {
2856 	p_nxge_ring_handle_t	ring_handle = (p_nxge_ring_handle_t)arg;
2857 	p_rx_rcr_ring_t		ringp;
2858 	p_nxge_t		nxgep;
2859 	p_nxge_ldg_t		ldgp;
2860 	uint32_t		channel;
2861 
2862 	if (ring_handle == NULL) {
2863 		return (0);
2864 	}
2865 
2866 	nxgep = ring_handle->nxgep;
2867 	channel = nxgep->pt_config.hw_config.start_rdc + ring_handle->index;
2868 	ringp = nxgep->rx_rcr_rings->rcr_rings[channel];
2869 	NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
2870 	    "==> nxge_enable_poll: rdc %d ", ringp->rdc));
2871 	ldgp = ringp->ldgp;
2872 	if (ldgp == NULL) {
2873 		NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
2874 		    "==> nxge_enable_poll: rdc %d NULL ldgp: no change",
2875 		    ringp->rdc));
2876 		return (0);
2877 	}
2878 
2879 	MUTEX_ENTER(&ringp->lock);
2880 	/* enable polling */
2881 	if (ringp->poll_flag == 0) {
2882 		ringp->poll_flag = 1;
2883 		NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
2884 		    "==> nxge_enable_poll: rdc %d set poll flag to 1",
2885 		    ringp->rdc));
2886 	}
2887 
2888 	MUTEX_EXIT(&ringp->lock);
2889 	return (0);
2890 }
2891 /*
2892  * Disable polling for a ring and enable its interrupt.
2893  */
2894 int
2895 nxge_disable_poll(void *arg)
2896 {
2897 	p_nxge_ring_handle_t	ring_handle = (p_nxge_ring_handle_t)arg;
2898 	p_rx_rcr_ring_t		ringp;
2899 	p_nxge_t		nxgep;
2900 	uint32_t		channel;
2901 
2902 	if (ring_handle == NULL) {
2903 		return (0);
2904 	}
2905 
2906 	nxgep = ring_handle->nxgep;
2907 	channel = nxgep->pt_config.hw_config.start_rdc + ring_handle->index;
2908 	ringp = nxgep->rx_rcr_rings->rcr_rings[channel];
2909 
2910 	NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
2911 	    "==> nxge_disable_poll: rdc %d poll_flag %d", ringp->rdc));
2912 
2913 	MUTEX_ENTER(&ringp->lock);
2914 
2915 	/* disable polling: enable interrupt */
2916 	if (ringp->poll_flag) {
2917 		npi_handle_t		handle;
2918 		rx_dma_ctl_stat_t	cs;
2919 		uint8_t			channel;
2920 		p_nxge_ldg_t		ldgp;
2921 
2922 		/*
2923 		 * Get the control and status for this channel.
2924 		 */
2925 		handle = NXGE_DEV_NPI_HANDLE(nxgep);
2926 		channel = ringp->rdc;
2927 		RXDMA_REG_READ64(handle, RX_DMA_CTL_STAT_REG,
2928 		    channel, &cs.value);
2929 
2930 		/*
2931 		 * Enable mailbox update
2932 		 * Since packets were not read and the hardware uses
2933 		 * bits pktread and ptrread to update the queue
2934 		 * length, we need to set both bits to 0.
2935 		 */
2936 		cs.bits.ldw.pktread = 0;
2937 		cs.bits.ldw.ptrread = 0;
2938 		cs.bits.hdw.mex = 1;
2939 		RXDMA_REG_WRITE64(handle, RX_DMA_CTL_STAT_REG, channel,
2940 		    cs.value);
2941 
2942 		/*
2943 		 * Rearm this logical group if this is a single device
2944 		 * group.
2945 		 */
2946 		ldgp = ringp->ldgp;
2947 		if (ldgp == NULL) {
2948 			ringp->poll_flag = 0;
2949 			MUTEX_EXIT(&ringp->lock);
2950 			NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
2951 			    "==> nxge_disable_poll: no ldgp rdc %d "
2952 			    "(still set poll to 0", ringp->rdc));
2953 			return (0);
2954 		}
2955 		NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
2956 		    "==> nxge_disable_poll: rdc %d ldgp $%p (enable intr)",
2957 		    ringp->rdc, ldgp));
2958 		if (ldgp->nldvs == 1) {
2959 			ldgimgm_t	mgm;
2960 			mgm.value = 0;
2961 			mgm.bits.ldw.arm = 1;
2962 			mgm.bits.ldw.timer = ldgp->ldg_timer;
2963 			NXGE_REG_WR64(handle,
2964 			    LDGIMGN_REG + LDSV_OFFSET(ldgp->ldg), mgm.value);
2965 		}
2966 		ringp->poll_flag = 0;
2967 	}
2968 
2969 	MUTEX_EXIT(&ringp->lock);
2970 	return (0);
2971 }
2972 
2973 /*
2974  * Poll 'bytes_to_pickup' bytes of message from the rx ring.
2975  */
2976 mblk_t *
2977 nxge_rx_poll(void *arg, int bytes_to_pickup)
2978 {
2979 	p_nxge_ring_handle_t	ring_handle = (p_nxge_ring_handle_t)arg;
2980 	p_rx_rcr_ring_t		rcr_p;
2981 	p_nxge_t		nxgep;
2982 	npi_handle_t		handle;
2983 	rx_dma_ctl_stat_t	cs;
2984 	mblk_t			*mblk;
2985 	p_nxge_ldv_t		ldvp;
2986 	uint32_t		channel;
2987 
2988 	nxgep = ring_handle->nxgep;
2989 
2990 	/*
2991 	 * Get the control and status for this channel.
2992 	 */
2993 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
2994 	channel = nxgep->pt_config.hw_config.start_rdc + ring_handle->index;
2995 	rcr_p = nxgep->rx_rcr_rings->rcr_rings[channel];
2996 	MUTEX_ENTER(&rcr_p->lock);
2997 	ASSERT(rcr_p->poll_flag == 1);
2998 
2999 	RXDMA_REG_READ64(handle, RX_DMA_CTL_STAT_REG, rcr_p->rdc, &cs.value);
3000 
3001 	NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
3002 	    "==> nxge_rx_poll: calling nxge_rx_pkts: rdc %d poll_flag %d",
3003 	    rcr_p->rdc, rcr_p->poll_flag));
3004 	mblk = nxge_rx_pkts(nxgep, rcr_p, cs, bytes_to_pickup);
3005 
3006 	ldvp = rcr_p->ldvp;
3007 	/* error events. */
3008 	if (ldvp && (cs.value & RX_DMA_CTL_STAT_ERROR)) {
3009 		(void) nxge_rx_err_evnts(nxgep, ldvp->vdma_index, cs);
3010 	}
3011 
3012 	MUTEX_EXIT(&rcr_p->lock);
3013 
3014 	NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
3015 	    "<== nxge_rx_poll: rdc %d mblk $%p", rcr_p->rdc, mblk));
3016 	return (mblk);
3017 }
3018 
3019 
3020 /*ARGSUSED*/
3021 static nxge_status_t
3022 nxge_rx_err_evnts(p_nxge_t nxgep, int channel, rx_dma_ctl_stat_t cs)
3023 {
3024 	p_nxge_rx_ring_stats_t	rdc_stats;
3025 	npi_handle_t		handle;
3026 	npi_status_t		rs;
3027 	boolean_t		rxchan_fatal = B_FALSE;
3028 	boolean_t		rxport_fatal = B_FALSE;
3029 	uint8_t			portn;
3030 	nxge_status_t		status = NXGE_OK;
3031 	uint32_t		error_disp_cnt = NXGE_ERROR_SHOW_MAX;
3032 	NXGE_DEBUG_MSG((nxgep, INT_CTL, "==> nxge_rx_err_evnts"));
3033 
3034 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
3035 	portn = nxgep->mac.portnum;
3036 	rdc_stats = &nxgep->statsp->rdc_stats[channel];
3037 
3038 	if (cs.bits.hdw.rbr_tmout) {
3039 		rdc_stats->rx_rbr_tmout++;
3040 		NXGE_FM_REPORT_ERROR(nxgep, portn, channel,
3041 		    NXGE_FM_EREPORT_RDMC_RBR_TMOUT);
3042 		rxchan_fatal = B_TRUE;
3043 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3044 		    "==> nxge_rx_err_evnts: rx_rbr_timeout"));
3045 	}
3046 	if (cs.bits.hdw.rsp_cnt_err) {
3047 		rdc_stats->rsp_cnt_err++;
3048 		NXGE_FM_REPORT_ERROR(nxgep, portn, channel,
3049 		    NXGE_FM_EREPORT_RDMC_RSP_CNT_ERR);
3050 		rxchan_fatal = B_TRUE;
3051 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3052 		    "==> nxge_rx_err_evnts(channel %d): "
3053 		    "rsp_cnt_err", channel));
3054 	}
3055 	if (cs.bits.hdw.byte_en_bus) {
3056 		rdc_stats->byte_en_bus++;
3057 		NXGE_FM_REPORT_ERROR(nxgep, portn, channel,
3058 		    NXGE_FM_EREPORT_RDMC_BYTE_EN_BUS);
3059 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3060 		    "==> nxge_rx_err_evnts(channel %d): "
3061 		    "fatal error: byte_en_bus", channel));
3062 		rxchan_fatal = B_TRUE;
3063 	}
3064 	if (cs.bits.hdw.rsp_dat_err) {
3065 		rdc_stats->rsp_dat_err++;
3066 		NXGE_FM_REPORT_ERROR(nxgep, portn, channel,
3067 		    NXGE_FM_EREPORT_RDMC_RSP_DAT_ERR);
3068 		rxchan_fatal = B_TRUE;
3069 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3070 		    "==> nxge_rx_err_evnts(channel %d): "
3071 		    "fatal error: rsp_dat_err", channel));
3072 	}
3073 	if (cs.bits.hdw.rcr_ack_err) {
3074 		rdc_stats->rcr_ack_err++;
3075 		NXGE_FM_REPORT_ERROR(nxgep, portn, channel,
3076 		    NXGE_FM_EREPORT_RDMC_RCR_ACK_ERR);
3077 		rxchan_fatal = B_TRUE;
3078 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3079 		    "==> nxge_rx_err_evnts(channel %d): "
3080 		    "fatal error: rcr_ack_err", channel));
3081 	}
3082 	if (cs.bits.hdw.dc_fifo_err) {
3083 		rdc_stats->dc_fifo_err++;
3084 		NXGE_FM_REPORT_ERROR(nxgep, portn, channel,
3085 		    NXGE_FM_EREPORT_RDMC_DC_FIFO_ERR);
3086 		/* This is not a fatal error! */
3087 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3088 		    "==> nxge_rx_err_evnts(channel %d): "
3089 		    "dc_fifo_err", channel));
3090 		rxport_fatal = B_TRUE;
3091 	}
3092 	if ((cs.bits.hdw.rcr_sha_par) || (cs.bits.hdw.rbr_pre_par)) {
3093 		if ((rs = npi_rxdma_ring_perr_stat_get(handle,
3094 		    &rdc_stats->errlog.pre_par,
3095 		    &rdc_stats->errlog.sha_par))
3096 		    != NPI_SUCCESS) {
3097 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3098 			    "==> nxge_rx_err_evnts(channel %d): "
3099 			    "rcr_sha_par: get perr", channel));
3100 			return (NXGE_ERROR | rs);
3101 		}
3102 		if (cs.bits.hdw.rcr_sha_par) {
3103 			rdc_stats->rcr_sha_par++;
3104 			NXGE_FM_REPORT_ERROR(nxgep, portn, channel,
3105 			    NXGE_FM_EREPORT_RDMC_RCR_SHA_PAR);
3106 			rxchan_fatal = B_TRUE;
3107 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3108 			    "==> nxge_rx_err_evnts(channel %d): "
3109 			    "fatal error: rcr_sha_par", channel));
3110 		}
3111 		if (cs.bits.hdw.rbr_pre_par) {
3112 			rdc_stats->rbr_pre_par++;
3113 			NXGE_FM_REPORT_ERROR(nxgep, portn, channel,
3114 			    NXGE_FM_EREPORT_RDMC_RBR_PRE_PAR);
3115 			rxchan_fatal = B_TRUE;
3116 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3117 			    "==> nxge_rx_err_evnts(channel %d): "
3118 			    "fatal error: rbr_pre_par", channel));
3119 		}
3120 	}
3121 	/*
3122 	 * The Following 4 status bits are for information, the system
3123 	 * is running fine. There is no need to send FMA ereports or
3124 	 * log messages.
3125 	 */
3126 	if (cs.bits.hdw.port_drop_pkt) {
3127 		rdc_stats->port_drop_pkt++;
3128 	}
3129 	if (cs.bits.hdw.wred_drop) {
3130 		rdc_stats->wred_drop++;
3131 	}
3132 	if (cs.bits.hdw.rbr_pre_empty) {
3133 		rdc_stats->rbr_pre_empty++;
3134 	}
3135 	if (cs.bits.hdw.rcr_shadow_full) {
3136 		rdc_stats->rcr_shadow_full++;
3137 	}
3138 	if (cs.bits.hdw.config_err) {
3139 		rdc_stats->config_err++;
3140 		NXGE_FM_REPORT_ERROR(nxgep, portn, channel,
3141 		    NXGE_FM_EREPORT_RDMC_CONFIG_ERR);
3142 		rxchan_fatal = B_TRUE;
3143 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3144 		    "==> nxge_rx_err_evnts(channel %d): "
3145 		    "config error", channel));
3146 	}
3147 	if (cs.bits.hdw.rcrincon) {
3148 		rdc_stats->rcrincon++;
3149 		NXGE_FM_REPORT_ERROR(nxgep, portn, channel,
3150 		    NXGE_FM_EREPORT_RDMC_RCRINCON);
3151 		rxchan_fatal = B_TRUE;
3152 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3153 		    "==> nxge_rx_err_evnts(channel %d): "
3154 		    "fatal error: rcrincon error", channel));
3155 	}
3156 	if (cs.bits.hdw.rcrfull) {
3157 		rdc_stats->rcrfull++;
3158 		NXGE_FM_REPORT_ERROR(nxgep, portn, channel,
3159 		    NXGE_FM_EREPORT_RDMC_RCRFULL);
3160 		rxchan_fatal = B_TRUE;
3161 		if (rdc_stats->rcrfull < error_disp_cnt)
3162 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3163 		    "==> nxge_rx_err_evnts(channel %d): "
3164 		    "fatal error: rcrfull error", channel));
3165 	}
3166 	if (cs.bits.hdw.rbr_empty) {
3167 		/*
3168 		 * This bit is for information, there is no need
3169 		 * send FMA ereport or log a message.
3170 		 */
3171 		rdc_stats->rbr_empty++;
3172 	}
3173 	if (cs.bits.hdw.rbrfull) {
3174 		rdc_stats->rbrfull++;
3175 		NXGE_FM_REPORT_ERROR(nxgep, portn, channel,
3176 		    NXGE_FM_EREPORT_RDMC_RBRFULL);
3177 		rxchan_fatal = B_TRUE;
3178 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3179 		    "==> nxge_rx_err_evnts(channel %d): "
3180 		    "fatal error: rbr_full error", channel));
3181 	}
3182 	if (cs.bits.hdw.rbrlogpage) {
3183 		rdc_stats->rbrlogpage++;
3184 		NXGE_FM_REPORT_ERROR(nxgep, portn, channel,
3185 		    NXGE_FM_EREPORT_RDMC_RBRLOGPAGE);
3186 		rxchan_fatal = B_TRUE;
3187 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3188 		    "==> nxge_rx_err_evnts(channel %d): "
3189 		    "fatal error: rbr logical page error", channel));
3190 	}
3191 	if (cs.bits.hdw.cfiglogpage) {
3192 		rdc_stats->cfiglogpage++;
3193 		NXGE_FM_REPORT_ERROR(nxgep, portn, channel,
3194 		    NXGE_FM_EREPORT_RDMC_CFIGLOGPAGE);
3195 		rxchan_fatal = B_TRUE;
3196 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3197 		    "==> nxge_rx_err_evnts(channel %d): "
3198 		    "fatal error: cfig logical page error", channel));
3199 	}
3200 
3201 	if (rxport_fatal)  {
3202 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3203 		    " nxge_rx_err_evnts: fatal error on Port #%d\n",
3204 		    portn));
3205 		if (isLDOMguest(nxgep)) {
3206 			status = NXGE_ERROR;
3207 		} else {
3208 			status = nxge_ipp_fatal_err_recover(nxgep);
3209 			if (status == NXGE_OK) {
3210 				FM_SERVICE_RESTORED(nxgep);
3211 			}
3212 		}
3213 	}
3214 
3215 	if (rxchan_fatal) {
3216 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3217 		    " nxge_rx_err_evnts: fatal error on Channel #%d\n",
3218 		    channel));
3219 		if (isLDOMguest(nxgep)) {
3220 			status = NXGE_ERROR;
3221 		} else {
3222 			status = nxge_rxdma_fatal_err_recover(nxgep, channel);
3223 			if (status == NXGE_OK) {
3224 				FM_SERVICE_RESTORED(nxgep);
3225 			}
3226 		}
3227 	}
3228 
3229 	NXGE_DEBUG_MSG((nxgep, RX2_CTL, "<== nxge_rx_err_evnts"));
3230 
3231 	return (status);
3232 }
3233 
3234 /*
3235  * nxge_rdc_hvio_setup
3236  *
3237  *	This code appears to setup some Hypervisor variables.
3238  *
3239  * Arguments:
3240  * 	nxgep
3241  * 	channel
3242  *
3243  * Notes:
3244  *	What does NIU_LP_WORKAROUND mean?
3245  *
3246  * NPI/NXGE function calls:
3247  *	na
3248  *
3249  * Context:
3250  *	Any domain
3251  */
3252 #if defined(sun4v) && defined(NIU_LP_WORKAROUND)
3253 static void
3254 nxge_rdc_hvio_setup(
3255 	nxge_t *nxgep, int channel)
3256 {
3257 	nxge_dma_common_t	*dma_common;
3258 	nxge_dma_common_t	*dma_control;
3259 	rx_rbr_ring_t		*ring;
3260 
3261 	ring = nxgep->rx_rbr_rings->rbr_rings[channel];
3262 	dma_common = nxgep->rx_buf_pool_p->dma_buf_pool_p[channel];
3263 
3264 	ring->hv_set = B_FALSE;
3265 
3266 	ring->hv_rx_buf_base_ioaddr_pp = (uint64_t)
3267 	    dma_common->orig_ioaddr_pp;
3268 	ring->hv_rx_buf_ioaddr_size = (uint64_t)
3269 	    dma_common->orig_alength;
3270 
3271 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_map_rxdma_channel: "
3272 	    "channel %d data buf base io $%lx ($%p) size 0x%lx (%ld 0x%lx)",
3273 	    channel, ring->hv_rx_buf_base_ioaddr_pp,
3274 	    dma_common->ioaddr_pp, ring->hv_rx_buf_ioaddr_size,
3275 	    dma_common->orig_alength, dma_common->orig_alength));
3276 
3277 	dma_control = nxgep->rx_cntl_pool_p->dma_buf_pool_p[channel];
3278 
3279 	ring->hv_rx_cntl_base_ioaddr_pp =
3280 	    (uint64_t)dma_control->orig_ioaddr_pp;
3281 	ring->hv_rx_cntl_ioaddr_size =
3282 	    (uint64_t)dma_control->orig_alength;
3283 
3284 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_map_rxdma_channel: "
3285 	    "channel %d cntl base io $%p ($%p) size 0x%llx (%d 0x%x)",
3286 	    channel, ring->hv_rx_cntl_base_ioaddr_pp,
3287 	    dma_control->ioaddr_pp, ring->hv_rx_cntl_ioaddr_size,
3288 	    dma_control->orig_alength, dma_control->orig_alength));
3289 }
3290 #endif
3291 
3292 /*
3293  * nxge_map_rxdma
3294  *
3295  *	Map an RDC into our kernel space.
3296  *
3297  * Arguments:
3298  * 	nxgep
3299  * 	channel	The channel to map.
3300  *
3301  * Notes:
3302  *	1. Allocate & initialise a memory pool, if necessary.
3303  *	2. Allocate however many receive buffers are required.
3304  *	3. Setup buffers, descriptors, and mailbox.
3305  *
3306  * NPI/NXGE function calls:
3307  *	nxge_alloc_rx_mem_pool()
3308  *	nxge_alloc_rbb()
3309  *	nxge_map_rxdma_channel()
3310  *
3311  * Registers accessed:
3312  *
3313  * Context:
3314  *	Any domain
3315  */
3316 static nxge_status_t
3317 nxge_map_rxdma(p_nxge_t nxgep, int channel)
3318 {
3319 	nxge_dma_common_t	**data;
3320 	nxge_dma_common_t	**control;
3321 	rx_rbr_ring_t		**rbr_ring;
3322 	rx_rcr_ring_t		**rcr_ring;
3323 	rx_mbox_t		**mailbox;
3324 	uint32_t		chunks;
3325 
3326 	nxge_status_t		status;
3327 
3328 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_map_rxdma"));
3329 
3330 	if (!nxgep->rx_buf_pool_p) {
3331 		if (nxge_alloc_rx_mem_pool(nxgep) != NXGE_OK) {
3332 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3333 			    "<== nxge_map_rxdma: buf not allocated"));
3334 			return (NXGE_ERROR);
3335 		}
3336 	}
3337 
3338 	if (nxge_alloc_rxb(nxgep, channel) != NXGE_OK)
3339 		return (NXGE_ERROR);
3340 
3341 	/*
3342 	 * Map descriptors from the buffer polls for each dma channel.
3343 	 */
3344 
3345 	/*
3346 	 * Set up and prepare buffer blocks, descriptors
3347 	 * and mailbox.
3348 	 */
3349 	data = &nxgep->rx_buf_pool_p->dma_buf_pool_p[channel];
3350 	rbr_ring = &nxgep->rx_rbr_rings->rbr_rings[channel];
3351 	chunks = nxgep->rx_buf_pool_p->num_chunks[channel];
3352 
3353 	control = &nxgep->rx_cntl_pool_p->dma_buf_pool_p[channel];
3354 	rcr_ring = &nxgep->rx_rcr_rings->rcr_rings[channel];
3355 
3356 	mailbox = &nxgep->rx_mbox_areas_p->rxmbox_areas[channel];
3357 
3358 	status = nxge_map_rxdma_channel(nxgep, channel, data, rbr_ring,
3359 	    chunks, control, rcr_ring, mailbox);
3360 	if (status != NXGE_OK) {
3361 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3362 		    "==> nxge_map_rxdma: nxge_map_rxdma_channel(%d) "
3363 		    "returned 0x%x",
3364 		    channel, status));
3365 		return (status);
3366 	}
3367 	nxgep->rx_rbr_rings->rbr_rings[channel]->index = (uint16_t)channel;
3368 	nxgep->rx_rcr_rings->rcr_rings[channel]->index = (uint16_t)channel;
3369 	nxgep->rx_rcr_rings->rcr_rings[channel]->rdc_stats =
3370 	    &nxgep->statsp->rdc_stats[channel];
3371 
3372 #if	defined(sun4v) && defined(NIU_LP_WORKAROUND)
3373 	if (!isLDOMguest(nxgep))
3374 		nxge_rdc_hvio_setup(nxgep, channel);
3375 #endif
3376 
3377 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3378 	    "<== nxge_map_rxdma: (status 0x%x channel %d)", status, channel));
3379 
3380 	return (status);
3381 }
3382 
3383 static void
3384 nxge_unmap_rxdma(p_nxge_t nxgep, int channel)
3385 {
3386 	rx_rbr_ring_t	*rbr_ring;
3387 	rx_rcr_ring_t	*rcr_ring;
3388 	rx_mbox_t	*mailbox;
3389 
3390 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_unmap_rxdma(%d)", channel));
3391 
3392 	if (!nxgep->rx_rbr_rings || !nxgep->rx_rcr_rings ||
3393 	    !nxgep->rx_mbox_areas_p)
3394 		return;
3395 
3396 	rbr_ring = nxgep->rx_rbr_rings->rbr_rings[channel];
3397 	rcr_ring = nxgep->rx_rcr_rings->rcr_rings[channel];
3398 	mailbox = nxgep->rx_mbox_areas_p->rxmbox_areas[channel];
3399 
3400 	if (!rbr_ring || !rcr_ring || !mailbox)
3401 		return;
3402 
3403 	(void) nxge_unmap_rxdma_channel(
3404 	    nxgep, channel, rbr_ring, rcr_ring, mailbox);
3405 
3406 	nxge_free_rxb(nxgep, channel);
3407 
3408 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "<== nxge_unmap_rxdma"));
3409 }
3410 
3411 nxge_status_t
3412 nxge_map_rxdma_channel(p_nxge_t nxgep, uint16_t channel,
3413     p_nxge_dma_common_t *dma_buf_p,  p_rx_rbr_ring_t *rbr_p,
3414     uint32_t num_chunks,
3415     p_nxge_dma_common_t *dma_cntl_p, p_rx_rcr_ring_t *rcr_p,
3416     p_rx_mbox_t *rx_mbox_p)
3417 {
3418 	int	status = NXGE_OK;
3419 
3420 	/*
3421 	 * Set up and prepare buffer blocks, descriptors
3422 	 * and mailbox.
3423 	 */
3424 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3425 	    "==> nxge_map_rxdma_channel (channel %d)", channel));
3426 	/*
3427 	 * Receive buffer blocks
3428 	 */
3429 	status = nxge_map_rxdma_channel_buf_ring(nxgep, channel,
3430 	    dma_buf_p, rbr_p, num_chunks);
3431 	if (status != NXGE_OK) {
3432 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3433 		    "==> nxge_map_rxdma_channel (channel %d): "
3434 		    "map buffer failed 0x%x", channel, status));
3435 		goto nxge_map_rxdma_channel_exit;
3436 	}
3437 
3438 	/*
3439 	 * Receive block ring, completion ring and mailbox.
3440 	 */
3441 	status = nxge_map_rxdma_channel_cfg_ring(nxgep, channel,
3442 	    dma_cntl_p, rbr_p, rcr_p, rx_mbox_p);
3443 	if (status != NXGE_OK) {
3444 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3445 		    "==> nxge_map_rxdma_channel (channel %d): "
3446 		    "map config failed 0x%x", channel, status));
3447 		goto nxge_map_rxdma_channel_fail2;
3448 	}
3449 
3450 	goto nxge_map_rxdma_channel_exit;
3451 
3452 nxge_map_rxdma_channel_fail3:
3453 	/* Free rbr, rcr */
3454 	NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3455 	    "==> nxge_map_rxdma_channel: free rbr/rcr "
3456 	    "(status 0x%x channel %d)",
3457 	    status, channel));
3458 	nxge_unmap_rxdma_channel_cfg_ring(nxgep,
3459 	    *rcr_p, *rx_mbox_p);
3460 
3461 nxge_map_rxdma_channel_fail2:
3462 	/* Free buffer blocks */
3463 	NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3464 	    "==> nxge_map_rxdma_channel: free rx buffers"
3465 	    "(nxgep 0x%x status 0x%x channel %d)",
3466 	    nxgep, status, channel));
3467 	nxge_unmap_rxdma_channel_buf_ring(nxgep, *rbr_p);
3468 
3469 	status = NXGE_ERROR;
3470 
3471 nxge_map_rxdma_channel_exit:
3472 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3473 	    "<== nxge_map_rxdma_channel: "
3474 	    "(nxgep 0x%x status 0x%x channel %d)",
3475 	    nxgep, status, channel));
3476 
3477 	return (status);
3478 }
3479 
3480 /*ARGSUSED*/
3481 static void
3482 nxge_unmap_rxdma_channel(p_nxge_t nxgep, uint16_t channel,
3483     p_rx_rbr_ring_t rbr_p, p_rx_rcr_ring_t rcr_p, p_rx_mbox_t rx_mbox_p)
3484 {
3485 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3486 	    "==> nxge_unmap_rxdma_channel (channel %d)", channel));
3487 
3488 	/*
3489 	 * unmap receive block ring, completion ring and mailbox.
3490 	 */
3491 	(void) nxge_unmap_rxdma_channel_cfg_ring(nxgep,
3492 	    rcr_p, rx_mbox_p);
3493 
3494 	/* unmap buffer blocks */
3495 	(void) nxge_unmap_rxdma_channel_buf_ring(nxgep, rbr_p);
3496 
3497 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "<== nxge_unmap_rxdma_channel"));
3498 }
3499 
3500 /*ARGSUSED*/
3501 static nxge_status_t
3502 nxge_map_rxdma_channel_cfg_ring(p_nxge_t nxgep, uint16_t dma_channel,
3503     p_nxge_dma_common_t *dma_cntl_p, p_rx_rbr_ring_t *rbr_p,
3504     p_rx_rcr_ring_t *rcr_p, p_rx_mbox_t *rx_mbox_p)
3505 {
3506 	p_rx_rbr_ring_t 	rbrp;
3507 	p_rx_rcr_ring_t 	rcrp;
3508 	p_rx_mbox_t 		mboxp;
3509 	p_nxge_dma_common_t 	cntl_dmap;
3510 	p_nxge_dma_common_t 	dmap;
3511 	p_rx_msg_t 		*rx_msg_ring;
3512 	p_rx_msg_t 		rx_msg_p;
3513 	p_rbr_cfig_a_t		rcfga_p;
3514 	p_rbr_cfig_b_t		rcfgb_p;
3515 	p_rcrcfig_a_t		cfga_p;
3516 	p_rcrcfig_b_t		cfgb_p;
3517 	p_rxdma_cfig1_t		cfig1_p;
3518 	p_rxdma_cfig2_t		cfig2_p;
3519 	p_rbr_kick_t		kick_p;
3520 	uint32_t		dmaaddrp;
3521 	uint32_t		*rbr_vaddrp;
3522 	uint32_t		bkaddr;
3523 	nxge_status_t		status = NXGE_OK;
3524 	int			i;
3525 	uint32_t 		nxge_port_rcr_size;
3526 
3527 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3528 	    "==> nxge_map_rxdma_channel_cfg_ring"));
3529 
3530 	cntl_dmap = *dma_cntl_p;
3531 
3532 	/* Map in the receive block ring */
3533 	rbrp = *rbr_p;
3534 	dmap = (p_nxge_dma_common_t)&rbrp->rbr_desc;
3535 	nxge_setup_dma_common(dmap, cntl_dmap, rbrp->rbb_max, 4);
3536 	/*
3537 	 * Zero out buffer block ring descriptors.
3538 	 */
3539 	bzero((caddr_t)dmap->kaddrp, dmap->alength);
3540 
3541 	rcfga_p = &(rbrp->rbr_cfga);
3542 	rcfgb_p = &(rbrp->rbr_cfgb);
3543 	kick_p = &(rbrp->rbr_kick);
3544 	rcfga_p->value = 0;
3545 	rcfgb_p->value = 0;
3546 	kick_p->value = 0;
3547 	rbrp->rbr_addr = dmap->dma_cookie.dmac_laddress;
3548 	rcfga_p->value = (rbrp->rbr_addr &
3549 	    (RBR_CFIG_A_STDADDR_MASK |
3550 	    RBR_CFIG_A_STDADDR_BASE_MASK));
3551 	rcfga_p->value |= ((uint64_t)rbrp->rbb_max << RBR_CFIG_A_LEN_SHIFT);
3552 
3553 	rcfgb_p->bits.ldw.bufsz0 = rbrp->pkt_buf_size0;
3554 	rcfgb_p->bits.ldw.vld0 = 1;
3555 	rcfgb_p->bits.ldw.bufsz1 = rbrp->pkt_buf_size1;
3556 	rcfgb_p->bits.ldw.vld1 = 1;
3557 	rcfgb_p->bits.ldw.bufsz2 = rbrp->pkt_buf_size2;
3558 	rcfgb_p->bits.ldw.vld2 = 1;
3559 	rcfgb_p->bits.ldw.bksize = nxgep->rx_bksize_code;
3560 
3561 	/*
3562 	 * For each buffer block, enter receive block address to the ring.
3563 	 */
3564 	rbr_vaddrp = (uint32_t *)dmap->kaddrp;
3565 	rbrp->rbr_desc_vp = (uint32_t *)dmap->kaddrp;
3566 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3567 	    "==> nxge_map_rxdma_channel_cfg_ring: channel %d "
3568 	    "rbr_vaddrp $%p", dma_channel, rbr_vaddrp));
3569 
3570 	rx_msg_ring = rbrp->rx_msg_ring;
3571 	for (i = 0; i < rbrp->tnblocks; i++) {
3572 		rx_msg_p = rx_msg_ring[i];
3573 		rx_msg_p->nxgep = nxgep;
3574 		rx_msg_p->rx_rbr_p = rbrp;
3575 		bkaddr = (uint32_t)
3576 		    ((rx_msg_p->buf_dma.dma_cookie.dmac_laddress
3577 		    >> RBR_BKADDR_SHIFT));
3578 		rx_msg_p->free = B_FALSE;
3579 		rx_msg_p->max_usage_cnt = 0xbaddcafe;
3580 
3581 		*rbr_vaddrp++ = bkaddr;
3582 	}
3583 
3584 	kick_p->bits.ldw.bkadd = rbrp->rbb_max;
3585 	rbrp->rbr_wr_index = (rbrp->rbb_max - 1);
3586 
3587 	rbrp->rbr_rd_index = 0;
3588 
3589 	rbrp->rbr_consumed = 0;
3590 	rbrp->rbr_use_bcopy = B_TRUE;
3591 	rbrp->rbr_bufsize_type = RCR_PKTBUFSZ_0;
3592 	/*
3593 	 * Do bcopy on packets greater than bcopy size once
3594 	 * the lo threshold is reached.
3595 	 * This lo threshold should be less than the hi threshold.
3596 	 *
3597 	 * Do bcopy on every packet once the hi threshold is reached.
3598 	 */
3599 	if (nxge_rx_threshold_lo >= nxge_rx_threshold_hi) {
3600 		/* default it to use hi */
3601 		nxge_rx_threshold_lo = nxge_rx_threshold_hi;
3602 	}
3603 
3604 	if (nxge_rx_buf_size_type > NXGE_RBR_TYPE2) {
3605 		nxge_rx_buf_size_type = NXGE_RBR_TYPE2;
3606 	}
3607 	rbrp->rbr_bufsize_type = nxge_rx_buf_size_type;
3608 
3609 	switch (nxge_rx_threshold_hi) {
3610 	default:
3611 	case	NXGE_RX_COPY_NONE:
3612 		/* Do not do bcopy at all */
3613 		rbrp->rbr_use_bcopy = B_FALSE;
3614 		rbrp->rbr_threshold_hi = rbrp->rbb_max;
3615 		break;
3616 
3617 	case NXGE_RX_COPY_1:
3618 	case NXGE_RX_COPY_2:
3619 	case NXGE_RX_COPY_3:
3620 	case NXGE_RX_COPY_4:
3621 	case NXGE_RX_COPY_5:
3622 	case NXGE_RX_COPY_6:
3623 	case NXGE_RX_COPY_7:
3624 		rbrp->rbr_threshold_hi =
3625 		    rbrp->rbb_max *
3626 		    (nxge_rx_threshold_hi)/NXGE_RX_BCOPY_SCALE;
3627 		break;
3628 
3629 	case NXGE_RX_COPY_ALL:
3630 		rbrp->rbr_threshold_hi = 0;
3631 		break;
3632 	}
3633 
3634 	switch (nxge_rx_threshold_lo) {
3635 	default:
3636 	case	NXGE_RX_COPY_NONE:
3637 		/* Do not do bcopy at all */
3638 		if (rbrp->rbr_use_bcopy) {
3639 			rbrp->rbr_use_bcopy = B_FALSE;
3640 		}
3641 		rbrp->rbr_threshold_lo = rbrp->rbb_max;
3642 		break;
3643 
3644 	case NXGE_RX_COPY_1:
3645 	case NXGE_RX_COPY_2:
3646 	case NXGE_RX_COPY_3:
3647 	case NXGE_RX_COPY_4:
3648 	case NXGE_RX_COPY_5:
3649 	case NXGE_RX_COPY_6:
3650 	case NXGE_RX_COPY_7:
3651 		rbrp->rbr_threshold_lo =
3652 		    rbrp->rbb_max *
3653 		    (nxge_rx_threshold_lo)/NXGE_RX_BCOPY_SCALE;
3654 		break;
3655 
3656 	case NXGE_RX_COPY_ALL:
3657 		rbrp->rbr_threshold_lo = 0;
3658 		break;
3659 	}
3660 
3661 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
3662 	    "nxge_map_rxdma_channel_cfg_ring: channel %d "
3663 	    "rbb_max %d "
3664 	    "rbrp->rbr_bufsize_type %d "
3665 	    "rbb_threshold_hi %d "
3666 	    "rbb_threshold_lo %d",
3667 	    dma_channel,
3668 	    rbrp->rbb_max,
3669 	    rbrp->rbr_bufsize_type,
3670 	    rbrp->rbr_threshold_hi,
3671 	    rbrp->rbr_threshold_lo));
3672 
3673 	rbrp->page_valid.value = 0;
3674 	rbrp->page_mask_1.value = rbrp->page_mask_2.value = 0;
3675 	rbrp->page_value_1.value = rbrp->page_value_2.value = 0;
3676 	rbrp->page_reloc_1.value = rbrp->page_reloc_2.value = 0;
3677 	rbrp->page_hdl.value = 0;
3678 
3679 	rbrp->page_valid.bits.ldw.page0 = 1;
3680 	rbrp->page_valid.bits.ldw.page1 = 1;
3681 
3682 	/* Map in the receive completion ring */
3683 	rcrp = (p_rx_rcr_ring_t)
3684 	    KMEM_ZALLOC(sizeof (rx_rcr_ring_t), KM_SLEEP);
3685 	rcrp->rdc = dma_channel;
3686 
3687 	nxge_port_rcr_size = nxgep->nxge_port_rcr_size;
3688 	rcrp->comp_size = nxge_port_rcr_size;
3689 	rcrp->comp_wrap_mask = nxge_port_rcr_size - 1;
3690 
3691 	rcrp->max_receive_pkts = nxge_max_rx_pkts;
3692 
3693 	dmap = (p_nxge_dma_common_t)&rcrp->rcr_desc;
3694 	nxge_setup_dma_common(dmap, cntl_dmap, rcrp->comp_size,
3695 	    sizeof (rcr_entry_t));
3696 	rcrp->comp_rd_index = 0;
3697 	rcrp->comp_wt_index = 0;
3698 	rcrp->rcr_desc_rd_head_p = rcrp->rcr_desc_first_p =
3699 	    (p_rcr_entry_t)DMA_COMMON_VPTR(rcrp->rcr_desc);
3700 #if defined(__i386)
3701 	rcrp->rcr_desc_rd_head_pp = rcrp->rcr_desc_first_pp =
3702 	    (p_rcr_entry_t)(uint32_t)DMA_COMMON_IOADDR(rcrp->rcr_desc);
3703 #else
3704 	rcrp->rcr_desc_rd_head_pp = rcrp->rcr_desc_first_pp =
3705 	    (p_rcr_entry_t)DMA_COMMON_IOADDR(rcrp->rcr_desc);
3706 #endif
3707 
3708 	rcrp->rcr_desc_last_p = rcrp->rcr_desc_rd_head_p +
3709 	    (nxge_port_rcr_size - 1);
3710 	rcrp->rcr_desc_last_pp = rcrp->rcr_desc_rd_head_pp +
3711 	    (nxge_port_rcr_size - 1);
3712 
3713 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3714 	    "==> nxge_map_rxdma_channel_cfg_ring: "
3715 	    "channel %d "
3716 	    "rbr_vaddrp $%p "
3717 	    "rcr_desc_rd_head_p $%p "
3718 	    "rcr_desc_rd_head_pp $%p "
3719 	    "rcr_desc_rd_last_p $%p "
3720 	    "rcr_desc_rd_last_pp $%p ",
3721 	    dma_channel,
3722 	    rbr_vaddrp,
3723 	    rcrp->rcr_desc_rd_head_p,
3724 	    rcrp->rcr_desc_rd_head_pp,
3725 	    rcrp->rcr_desc_last_p,
3726 	    rcrp->rcr_desc_last_pp));
3727 
3728 	/*
3729 	 * Zero out buffer block ring descriptors.
3730 	 */
3731 	bzero((caddr_t)dmap->kaddrp, dmap->alength);
3732 
3733 	rcrp->intr_timeout = (nxgep->intr_timeout <
3734 	    NXGE_RDC_RCR_TIMEOUT_MIN) ? NXGE_RDC_RCR_TIMEOUT_MIN :
3735 	    nxgep->intr_timeout;
3736 
3737 	rcrp->intr_threshold = (nxgep->intr_threshold <
3738 	    NXGE_RDC_RCR_THRESHOLD_MIN) ? NXGE_RDC_RCR_THRESHOLD_MIN :
3739 	    nxgep->intr_threshold;
3740 
3741 	rcrp->full_hdr_flag = B_FALSE;
3742 	rcrp->sw_priv_hdr_len = 0;
3743 
3744 	cfga_p = &(rcrp->rcr_cfga);
3745 	cfgb_p = &(rcrp->rcr_cfgb);
3746 	cfga_p->value = 0;
3747 	cfgb_p->value = 0;
3748 	rcrp->rcr_addr = dmap->dma_cookie.dmac_laddress;
3749 	cfga_p->value = (rcrp->rcr_addr &
3750 	    (RCRCFIG_A_STADDR_MASK |
3751 	    RCRCFIG_A_STADDR_BASE_MASK));
3752 
3753 	rcfga_p->value |= ((uint64_t)rcrp->comp_size <<
3754 	    RCRCFIG_A_LEN_SHIF);
3755 
3756 	/*
3757 	 * Timeout should be set based on the system clock divider.
3758 	 * A timeout value of 1 assumes that the
3759 	 * granularity (1000) is 3 microseconds running at 300MHz.
3760 	 */
3761 	cfgb_p->bits.ldw.pthres = rcrp->intr_threshold;
3762 	cfgb_p->bits.ldw.timeout = rcrp->intr_timeout;
3763 	cfgb_p->bits.ldw.entout = 1;
3764 
3765 	/* Map in the mailbox */
3766 	mboxp = (p_rx_mbox_t)
3767 	    KMEM_ZALLOC(sizeof (rx_mbox_t), KM_SLEEP);
3768 	dmap = (p_nxge_dma_common_t)&mboxp->rx_mbox;
3769 	nxge_setup_dma_common(dmap, cntl_dmap, 1, sizeof (rxdma_mailbox_t));
3770 	cfig1_p = (p_rxdma_cfig1_t)&mboxp->rx_cfg1;
3771 	cfig2_p = (p_rxdma_cfig2_t)&mboxp->rx_cfg2;
3772 	cfig1_p->value = cfig2_p->value = 0;
3773 
3774 	mboxp->mbox_addr = dmap->dma_cookie.dmac_laddress;
3775 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3776 	    "==> nxge_map_rxdma_channel_cfg_ring: "
3777 	    "channel %d cfg1 0x%016llx cfig2 0x%016llx cookie 0x%016llx",
3778 	    dma_channel, cfig1_p->value, cfig2_p->value,
3779 	    mboxp->mbox_addr));
3780 
3781 	dmaaddrp = (uint32_t)(dmap->dma_cookie.dmac_laddress >> 32
3782 	    & 0xfff);
3783 	cfig1_p->bits.ldw.mbaddr_h = dmaaddrp;
3784 
3785 
3786 	dmaaddrp = (uint32_t)(dmap->dma_cookie.dmac_laddress & 0xffffffff);
3787 	dmaaddrp = (uint32_t)(dmap->dma_cookie.dmac_laddress &
3788 	    RXDMA_CFIG2_MBADDR_L_MASK);
3789 
3790 	cfig2_p->bits.ldw.mbaddr = (dmaaddrp >> RXDMA_CFIG2_MBADDR_L_SHIFT);
3791 
3792 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3793 	    "==> nxge_map_rxdma_channel_cfg_ring: "
3794 	    "channel %d damaddrp $%p "
3795 	    "cfg1 0x%016llx cfig2 0x%016llx",
3796 	    dma_channel, dmaaddrp,
3797 	    cfig1_p->value, cfig2_p->value));
3798 
3799 	cfig2_p->bits.ldw.full_hdr = rcrp->full_hdr_flag;
3800 	cfig2_p->bits.ldw.offset = rcrp->sw_priv_hdr_len;
3801 
3802 	rbrp->rx_rcr_p = rcrp;
3803 	rcrp->rx_rbr_p = rbrp;
3804 	*rcr_p = rcrp;
3805 	*rx_mbox_p = mboxp;
3806 
3807 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3808 	    "<== nxge_map_rxdma_channel_cfg_ring status 0x%08x", status));
3809 
3810 	return (status);
3811 }
3812 
3813 /*ARGSUSED*/
3814 static void
3815 nxge_unmap_rxdma_channel_cfg_ring(p_nxge_t nxgep,
3816     p_rx_rcr_ring_t rcr_p, p_rx_mbox_t rx_mbox_p)
3817 {
3818 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3819 	    "==> nxge_unmap_rxdma_channel_cfg_ring: channel %d",
3820 	    rcr_p->rdc));
3821 
3822 	KMEM_FREE(rcr_p, sizeof (rx_rcr_ring_t));
3823 	KMEM_FREE(rx_mbox_p, sizeof (rx_mbox_t));
3824 
3825 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3826 	    "<== nxge_unmap_rxdma_channel_cfg_ring"));
3827 }
3828 
3829 static nxge_status_t
3830 nxge_map_rxdma_channel_buf_ring(p_nxge_t nxgep, uint16_t channel,
3831     p_nxge_dma_common_t *dma_buf_p,
3832     p_rx_rbr_ring_t *rbr_p, uint32_t num_chunks)
3833 {
3834 	p_rx_rbr_ring_t 	rbrp;
3835 	p_nxge_dma_common_t 	dma_bufp, tmp_bufp;
3836 	p_rx_msg_t 		*rx_msg_ring;
3837 	p_rx_msg_t 		rx_msg_p;
3838 	p_mblk_t 		mblk_p;
3839 
3840 	rxring_info_t *ring_info;
3841 	nxge_status_t		status = NXGE_OK;
3842 	int			i, j, index;
3843 	uint32_t		size, bsize, nblocks, nmsgs;
3844 
3845 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3846 	    "==> nxge_map_rxdma_channel_buf_ring: channel %d",
3847 	    channel));
3848 
3849 	dma_bufp = tmp_bufp = *dma_buf_p;
3850 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3851 	    " nxge_map_rxdma_channel_buf_ring: channel %d to map %d "
3852 	    "chunks bufp 0x%016llx",
3853 	    channel, num_chunks, dma_bufp));
3854 
3855 	nmsgs = 0;
3856 	for (i = 0; i < num_chunks; i++, tmp_bufp++) {
3857 		NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3858 		    "==> nxge_map_rxdma_channel_buf_ring: channel %d "
3859 		    "bufp 0x%016llx nblocks %d nmsgs %d",
3860 		    channel, tmp_bufp, tmp_bufp->nblocks, nmsgs));
3861 		nmsgs += tmp_bufp->nblocks;
3862 	}
3863 	if (!nmsgs) {
3864 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3865 		    "<== nxge_map_rxdma_channel_buf_ring: channel %d "
3866 		    "no msg blocks",
3867 		    channel));
3868 		status = NXGE_ERROR;
3869 		goto nxge_map_rxdma_channel_buf_ring_exit;
3870 	}
3871 
3872 	rbrp = (p_rx_rbr_ring_t)KMEM_ZALLOC(sizeof (*rbrp), KM_SLEEP);
3873 
3874 	size = nmsgs * sizeof (p_rx_msg_t);
3875 	rx_msg_ring = KMEM_ZALLOC(size, KM_SLEEP);
3876 	ring_info = (rxring_info_t *)KMEM_ZALLOC(sizeof (rxring_info_t),
3877 	    KM_SLEEP);
3878 
3879 	MUTEX_INIT(&rbrp->lock, NULL, MUTEX_DRIVER,
3880 	    (void *)nxgep->interrupt_cookie);
3881 	MUTEX_INIT(&rbrp->post_lock, NULL, MUTEX_DRIVER,
3882 	    (void *)nxgep->interrupt_cookie);
3883 	rbrp->rdc = channel;
3884 	rbrp->num_blocks = num_chunks;
3885 	rbrp->tnblocks = nmsgs;
3886 	rbrp->rbb_max = nmsgs;
3887 	rbrp->rbr_max_size = nmsgs;
3888 	rbrp->rbr_wrap_mask = (rbrp->rbb_max - 1);
3889 
3890 	/*
3891 	 * Buffer sizes suggested by NIU architect.
3892 	 * 256, 512 and 2K.
3893 	 */
3894 
3895 	rbrp->pkt_buf_size0 = RBR_BUFSZ0_256B;
3896 	rbrp->pkt_buf_size0_bytes = RBR_BUFSZ0_256_BYTES;
3897 	rbrp->npi_pkt_buf_size0 = SIZE_256B;
3898 
3899 	rbrp->pkt_buf_size1 = RBR_BUFSZ1_1K;
3900 	rbrp->pkt_buf_size1_bytes = RBR_BUFSZ1_1K_BYTES;
3901 	rbrp->npi_pkt_buf_size1 = SIZE_1KB;
3902 
3903 	rbrp->block_size = nxgep->rx_default_block_size;
3904 
3905 	if (!nxge_jumbo_enable && !nxgep->param_arr[param_accept_jumbo].value) {
3906 		rbrp->pkt_buf_size2 = RBR_BUFSZ2_2K;
3907 		rbrp->pkt_buf_size2_bytes = RBR_BUFSZ2_2K_BYTES;
3908 		rbrp->npi_pkt_buf_size2 = SIZE_2KB;
3909 	} else {
3910 		if (rbrp->block_size >= 0x2000) {
3911 			rbrp->pkt_buf_size2 = RBR_BUFSZ2_8K;
3912 			rbrp->pkt_buf_size2_bytes = RBR_BUFSZ2_8K_BYTES;
3913 			rbrp->npi_pkt_buf_size2 = SIZE_8KB;
3914 		} else {
3915 			rbrp->pkt_buf_size2 = RBR_BUFSZ2_4K;
3916 			rbrp->pkt_buf_size2_bytes = RBR_BUFSZ2_4K_BYTES;
3917 			rbrp->npi_pkt_buf_size2 = SIZE_4KB;
3918 		}
3919 	}
3920 
3921 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3922 	    "==> nxge_map_rxdma_channel_buf_ring: channel %d "
3923 	    "actual rbr max %d rbb_max %d nmsgs %d "
3924 	    "rbrp->block_size %d default_block_size %d "
3925 	    "(config nxge_rbr_size %d nxge_rbr_spare_size %d)",
3926 	    channel, rbrp->rbr_max_size, rbrp->rbb_max, nmsgs,
3927 	    rbrp->block_size, nxgep->rx_default_block_size,
3928 	    nxge_rbr_size, nxge_rbr_spare_size));
3929 
3930 	/* Map in buffers from the buffer pool.  */
3931 	index = 0;
3932 	for (i = 0; i < rbrp->num_blocks; i++, dma_bufp++) {
3933 		bsize = dma_bufp->block_size;
3934 		nblocks = dma_bufp->nblocks;
3935 #if defined(__i386)
3936 		ring_info->buffer[i].dvma_addr = (uint32_t)dma_bufp->ioaddr_pp;
3937 #else
3938 		ring_info->buffer[i].dvma_addr = (uint64_t)dma_bufp->ioaddr_pp;
3939 #endif
3940 		ring_info->buffer[i].buf_index = i;
3941 		ring_info->buffer[i].buf_size = dma_bufp->alength;
3942 		ring_info->buffer[i].start_index = index;
3943 #if defined(__i386)
3944 		ring_info->buffer[i].kaddr = (uint32_t)dma_bufp->kaddrp;
3945 #else
3946 		ring_info->buffer[i].kaddr = (uint64_t)dma_bufp->kaddrp;
3947 #endif
3948 
3949 		NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3950 		    " nxge_map_rxdma_channel_buf_ring: map channel %d "
3951 		    "chunk %d"
3952 		    " nblocks %d chunk_size %x block_size 0x%x "
3953 		    "dma_bufp $%p", channel, i,
3954 		    dma_bufp->nblocks, ring_info->buffer[i].buf_size, bsize,
3955 		    dma_bufp));
3956 
3957 		for (j = 0; j < nblocks; j++) {
3958 			if ((rx_msg_p = nxge_allocb(bsize, BPRI_LO,
3959 			    dma_bufp)) == NULL) {
3960 				NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3961 				    "allocb failed (index %d i %d j %d)",
3962 				    index, i, j));
3963 				goto nxge_map_rxdma_channel_buf_ring_fail1;
3964 			}
3965 			rx_msg_ring[index] = rx_msg_p;
3966 			rx_msg_p->block_index = index;
3967 			rx_msg_p->shifted_addr = (uint32_t)
3968 			    ((rx_msg_p->buf_dma.dma_cookie.dmac_laddress >>
3969 			    RBR_BKADDR_SHIFT));
3970 
3971 			NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3972 			    "index %d j %d rx_msg_p $%p mblk %p",
3973 			    index, j, rx_msg_p, rx_msg_p->rx_mblk_p));
3974 
3975 			mblk_p = rx_msg_p->rx_mblk_p;
3976 			mblk_p->b_wptr = mblk_p->b_rptr + bsize;
3977 
3978 			rbrp->rbr_ref_cnt++;
3979 			index++;
3980 			rx_msg_p->buf_dma.dma_channel = channel;
3981 		}
3982 
3983 		rbrp->rbr_alloc_type = DDI_MEM_ALLOC;
3984 		if (dma_bufp->contig_alloc_type) {
3985 			rbrp->rbr_alloc_type = CONTIG_MEM_ALLOC;
3986 		}
3987 
3988 		if (dma_bufp->kmem_alloc_type) {
3989 			rbrp->rbr_alloc_type = KMEM_ALLOC;
3990 		}
3991 
3992 		NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3993 		    " nxge_map_rxdma_channel_buf_ring: map channel %d "
3994 		    "chunk %d"
3995 		    " nblocks %d chunk_size %x block_size 0x%x "
3996 		    "dma_bufp $%p",
3997 		    channel, i,
3998 		    dma_bufp->nblocks, ring_info->buffer[i].buf_size, bsize,
3999 		    dma_bufp));
4000 	}
4001 	if (i < rbrp->num_blocks) {
4002 		goto nxge_map_rxdma_channel_buf_ring_fail1;
4003 	}
4004 
4005 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4006 	    "nxge_map_rxdma_channel_buf_ring: done buf init "
4007 	    "channel %d msg block entries %d",
4008 	    channel, index));
4009 	ring_info->block_size_mask = bsize - 1;
4010 	rbrp->rx_msg_ring = rx_msg_ring;
4011 	rbrp->dma_bufp = dma_buf_p;
4012 	rbrp->ring_info = ring_info;
4013 
4014 	status = nxge_rxbuf_index_info_init(nxgep, rbrp);
4015 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4016 	    " nxge_map_rxdma_channel_buf_ring: "
4017 	    "channel %d done buf info init", channel));
4018 
4019 	/*
4020 	 * Finally, permit nxge_freeb() to call nxge_post_page().
4021 	 */
4022 	rbrp->rbr_state = RBR_POSTING;
4023 
4024 	*rbr_p = rbrp;
4025 	goto nxge_map_rxdma_channel_buf_ring_exit;
4026 
4027 nxge_map_rxdma_channel_buf_ring_fail1:
4028 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4029 	    " nxge_map_rxdma_channel_buf_ring: failed channel (0x%x)",
4030 	    channel, status));
4031 
4032 	index--;
4033 	for (; index >= 0; index--) {
4034 		rx_msg_p = rx_msg_ring[index];
4035 		if (rx_msg_p != NULL) {
4036 			freeb(rx_msg_p->rx_mblk_p);
4037 			rx_msg_ring[index] = NULL;
4038 		}
4039 	}
4040 nxge_map_rxdma_channel_buf_ring_fail:
4041 	MUTEX_DESTROY(&rbrp->post_lock);
4042 	MUTEX_DESTROY(&rbrp->lock);
4043 	KMEM_FREE(ring_info, sizeof (rxring_info_t));
4044 	KMEM_FREE(rx_msg_ring, size);
4045 	KMEM_FREE(rbrp, sizeof (rx_rbr_ring_t));
4046 
4047 	status = NXGE_ERROR;
4048 
4049 nxge_map_rxdma_channel_buf_ring_exit:
4050 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4051 	    "<== nxge_map_rxdma_channel_buf_ring status 0x%08x", status));
4052 
4053 	return (status);
4054 }
4055 
4056 /*ARGSUSED*/
4057 static void
4058 nxge_unmap_rxdma_channel_buf_ring(p_nxge_t nxgep,
4059     p_rx_rbr_ring_t rbr_p)
4060 {
4061 	p_rx_msg_t 		*rx_msg_ring;
4062 	p_rx_msg_t 		rx_msg_p;
4063 	rxring_info_t 		*ring_info;
4064 	int			i;
4065 	uint32_t		size;
4066 #ifdef	NXGE_DEBUG
4067 	int			num_chunks;
4068 #endif
4069 
4070 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4071 	    "==> nxge_unmap_rxdma_channel_buf_ring"));
4072 	if (rbr_p == NULL) {
4073 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
4074 		    "<== nxge_unmap_rxdma_channel_buf_ring: NULL rbrp"));
4075 		return;
4076 	}
4077 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4078 	    "==> nxge_unmap_rxdma_channel_buf_ring: channel %d",
4079 	    rbr_p->rdc));
4080 
4081 	rx_msg_ring = rbr_p->rx_msg_ring;
4082 	ring_info = rbr_p->ring_info;
4083 
4084 	if (rx_msg_ring == NULL || ring_info == NULL) {
4085 		NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4086 		    "<== nxge_unmap_rxdma_channel_buf_ring: "
4087 		    "rx_msg_ring $%p ring_info $%p",
4088 		    rx_msg_p, ring_info));
4089 		return;
4090 	}
4091 
4092 #ifdef	NXGE_DEBUG
4093 	num_chunks = rbr_p->num_blocks;
4094 #endif
4095 	size = rbr_p->tnblocks * sizeof (p_rx_msg_t);
4096 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4097 	    " nxge_unmap_rxdma_channel_buf_ring: channel %d chunks %d "
4098 	    "tnblocks %d (max %d) size ptrs %d ",
4099 	    rbr_p->rdc, num_chunks,
4100 	    rbr_p->tnblocks, rbr_p->rbr_max_size, size));
4101 
4102 	for (i = 0; i < rbr_p->tnblocks; i++) {
4103 		rx_msg_p = rx_msg_ring[i];
4104 		NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4105 		    " nxge_unmap_rxdma_channel_buf_ring: "
4106 		    "rx_msg_p $%p",
4107 		    rx_msg_p));
4108 		if (rx_msg_p != NULL) {
4109 			freeb(rx_msg_p->rx_mblk_p);
4110 			rx_msg_ring[i] = NULL;
4111 		}
4112 	}
4113 
4114 	/*
4115 	 * We no longer may use the mutex <post_lock>. By setting
4116 	 * <rbr_state> to anything but POSTING, we prevent
4117 	 * nxge_post_page() from accessing a dead mutex.
4118 	 */
4119 	rbr_p->rbr_state = RBR_UNMAPPING;
4120 	MUTEX_DESTROY(&rbr_p->post_lock);
4121 
4122 	MUTEX_DESTROY(&rbr_p->lock);
4123 
4124 	if (rbr_p->rbr_ref_cnt == 0) {
4125 		/*
4126 		 * This is the normal state of affairs.
4127 		 * Need to free the following buffers:
4128 		 *  - data buffers
4129 		 *  - rx_msg ring
4130 		 *  - ring_info
4131 		 *  - rbr ring
4132 		 */
4133 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
4134 		    "unmap_rxdma_buf_ring: No outstanding - freeing "));
4135 		nxge_rxdma_databuf_free(rbr_p);
4136 		KMEM_FREE(ring_info, sizeof (rxring_info_t));
4137 		KMEM_FREE(rx_msg_ring, size);
4138 		KMEM_FREE(rbr_p, sizeof (*rbr_p));
4139 	} else {
4140 		/*
4141 		 * Some of our buffers are still being used.
4142 		 * Therefore, tell nxge_freeb() this ring is
4143 		 * unmapped, so it may free <rbr_p> for us.
4144 		 */
4145 		rbr_p->rbr_state = RBR_UNMAPPED;
4146 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4147 		    "unmap_rxdma_buf_ring: %d %s outstanding.",
4148 		    rbr_p->rbr_ref_cnt,
4149 		    rbr_p->rbr_ref_cnt == 1 ? "msg" : "msgs"));
4150 	}
4151 
4152 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4153 	    "<== nxge_unmap_rxdma_channel_buf_ring"));
4154 }
4155 
4156 /*
4157  * nxge_rxdma_hw_start_common
4158  *
4159  * Arguments:
4160  * 	nxgep
4161  *
4162  * Notes:
4163  *
4164  * NPI/NXGE function calls:
4165  *	nxge_init_fzc_rx_common();
4166  *	nxge_init_fzc_rxdma_port();
4167  *
4168  * Registers accessed:
4169  *
4170  * Context:
4171  *	Service domain
4172  */
4173 static nxge_status_t
4174 nxge_rxdma_hw_start_common(p_nxge_t nxgep)
4175 {
4176 	nxge_status_t		status = NXGE_OK;
4177 
4178 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_rxdma_hw_start_common"));
4179 
4180 	/*
4181 	 * Load the sharable parameters by writing to the
4182 	 * function zero control registers. These FZC registers
4183 	 * should be initialized only once for the entire chip.
4184 	 */
4185 	(void) nxge_init_fzc_rx_common(nxgep);
4186 
4187 	/*
4188 	 * Initialize the RXDMA port specific FZC control configurations.
4189 	 * These FZC registers are pertaining to each port.
4190 	 */
4191 	(void) nxge_init_fzc_rxdma_port(nxgep);
4192 
4193 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_rxdma_hw_start_common"));
4194 
4195 	return (status);
4196 }
4197 
4198 static nxge_status_t
4199 nxge_rxdma_hw_start(p_nxge_t nxgep, int channel)
4200 {
4201 	int			i, ndmas;
4202 	p_rx_rbr_rings_t 	rx_rbr_rings;
4203 	p_rx_rbr_ring_t		*rbr_rings;
4204 	p_rx_rcr_rings_t 	rx_rcr_rings;
4205 	p_rx_rcr_ring_t		*rcr_rings;
4206 	p_rx_mbox_areas_t 	rx_mbox_areas_p;
4207 	p_rx_mbox_t		*rx_mbox_p;
4208 	nxge_status_t		status = NXGE_OK;
4209 
4210 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_rxdma_hw_start"));
4211 
4212 	rx_rbr_rings = nxgep->rx_rbr_rings;
4213 	rx_rcr_rings = nxgep->rx_rcr_rings;
4214 	if (rx_rbr_rings == NULL || rx_rcr_rings == NULL) {
4215 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
4216 		    "<== nxge_rxdma_hw_start: NULL ring pointers"));
4217 		return (NXGE_ERROR);
4218 	}
4219 	ndmas = rx_rbr_rings->ndmas;
4220 	if (ndmas == 0) {
4221 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
4222 		    "<== nxge_rxdma_hw_start: no dma channel allocated"));
4223 		return (NXGE_ERROR);
4224 	}
4225 
4226 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4227 	    "==> nxge_rxdma_hw_start (ndmas %d)", ndmas));
4228 
4229 	rbr_rings = rx_rbr_rings->rbr_rings;
4230 	rcr_rings = rx_rcr_rings->rcr_rings;
4231 	rx_mbox_areas_p = nxgep->rx_mbox_areas_p;
4232 	if (rx_mbox_areas_p) {
4233 		rx_mbox_p = rx_mbox_areas_p->rxmbox_areas;
4234 	}
4235 
4236 	i = channel;
4237 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4238 	    "==> nxge_rxdma_hw_start (ndmas %d) channel %d",
4239 	    ndmas, channel));
4240 	status = nxge_rxdma_start_channel(nxgep, channel,
4241 	    (p_rx_rbr_ring_t)rbr_rings[i],
4242 	    (p_rx_rcr_ring_t)rcr_rings[i],
4243 	    (p_rx_mbox_t)rx_mbox_p[i]);
4244 	if (status != NXGE_OK) {
4245 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4246 		    "==> nxge_rxdma_hw_start: disable "
4247 		    "(status 0x%x channel %d)", status, channel));
4248 		return (status);
4249 	}
4250 
4251 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_rxdma_hw_start: "
4252 	    "rx_rbr_rings 0x%016llx rings 0x%016llx",
4253 	    rx_rbr_rings, rx_rcr_rings));
4254 
4255 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4256 	    "==> nxge_rxdma_hw_start: (status 0x%x)", status));
4257 
4258 	return (status);
4259 }
4260 
4261 static void
4262 nxge_rxdma_hw_stop(p_nxge_t nxgep, int channel)
4263 {
4264 	p_rx_rbr_rings_t 	rx_rbr_rings;
4265 	p_rx_rcr_rings_t 	rx_rcr_rings;
4266 
4267 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_rxdma_hw_stop"));
4268 
4269 	rx_rbr_rings = nxgep->rx_rbr_rings;
4270 	rx_rcr_rings = nxgep->rx_rcr_rings;
4271 	if (rx_rbr_rings == NULL || rx_rcr_rings == NULL) {
4272 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
4273 		    "<== nxge_rxdma_hw_stop: NULL ring pointers"));
4274 		return;
4275 	}
4276 
4277 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4278 	    "==> nxge_rxdma_hw_stop(channel %d)",
4279 	    channel));
4280 	(void) nxge_rxdma_stop_channel(nxgep, channel);
4281 
4282 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_rxdma_hw_stop: "
4283 	    "rx_rbr_rings 0x%016llx rings 0x%016llx",
4284 	    rx_rbr_rings, rx_rcr_rings));
4285 
4286 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "<== nxge_rxdma_hw_stop"));
4287 }
4288 
4289 
4290 static nxge_status_t
4291 nxge_rxdma_start_channel(p_nxge_t nxgep, uint16_t channel,
4292     p_rx_rbr_ring_t rbr_p, p_rx_rcr_ring_t rcr_p, p_rx_mbox_t mbox_p)
4293 
4294 {
4295 	npi_handle_t		handle;
4296 	npi_status_t		rs = NPI_SUCCESS;
4297 	rx_dma_ctl_stat_t	cs;
4298 	rx_dma_ent_msk_t	ent_mask;
4299 	nxge_status_t		status = NXGE_OK;
4300 
4301 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_rxdma_start_channel"));
4302 
4303 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
4304 
4305 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "nxge_rxdma_start_channel: "
4306 		"npi handle addr $%p acc $%p",
4307 		nxgep->npi_handle.regp, nxgep->npi_handle.regh));
4308 
4309 	/* Reset RXDMA channel, but not if you're a guest. */
4310 	if (!isLDOMguest(nxgep)) {
4311 		rs = npi_rxdma_cfg_rdc_reset(handle, channel);
4312 		if (rs != NPI_SUCCESS) {
4313 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4314 			    "==> nxge_init_fzc_rdc: "
4315 			    "npi_rxdma_cfg_rdc_reset(%d) returned 0x%08x",
4316 			    channel, rs));
4317 			return (NXGE_ERROR | rs);
4318 		}
4319 
4320 		NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4321 		    "==> nxge_rxdma_start_channel: reset done: channel %d",
4322 		    channel));
4323 	}
4324 
4325 #if defined(sun4v) && defined(NIU_LP_WORKAROUND)
4326 	if (isLDOMguest(nxgep))
4327 		(void) nxge_rdc_lp_conf(nxgep, channel);
4328 #endif
4329 
4330 	/*
4331 	 * Initialize the RXDMA channel specific FZC control
4332 	 * configurations. These FZC registers are pertaining
4333 	 * to each RX channel (logical pages).
4334 	 */
4335 	if (!isLDOMguest(nxgep)) {
4336 		status = nxge_init_fzc_rxdma_channel(nxgep, channel);
4337 		if (status != NXGE_OK) {
4338 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4339 				"==> nxge_rxdma_start_channel: "
4340 				"init fzc rxdma failed (0x%08x channel %d)",
4341 				status, channel));
4342 			return (status);
4343 		}
4344 
4345 		NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4346 			"==> nxge_rxdma_start_channel: fzc done"));
4347 	}
4348 
4349 	/* Set up the interrupt event masks. */
4350 	ent_mask.value = 0;
4351 	ent_mask.value |= RX_DMA_ENT_MSK_RBREMPTY_MASK;
4352 	rs = npi_rxdma_event_mask(handle, OP_SET, channel,
4353 	    &ent_mask);
4354 	if (rs != NPI_SUCCESS) {
4355 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4356 			"==> nxge_rxdma_start_channel: "
4357 			"init rxdma event masks failed "
4358 			"(0x%08x channel %d)",
4359 			status, channel));
4360 		return (NXGE_ERROR | rs);
4361 	}
4362 
4363 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4364 		"==> nxge_rxdma_start_channel: "
4365 		"event done: channel %d (mask 0x%016llx)",
4366 		channel, ent_mask.value));
4367 
4368 	/* Initialize the receive DMA control and status register */
4369 	cs.value = 0;
4370 	cs.bits.hdw.mex = 1;
4371 	cs.bits.hdw.rcrthres = 1;
4372 	cs.bits.hdw.rcrto = 1;
4373 	cs.bits.hdw.rbr_empty = 1;
4374 	status = nxge_init_rxdma_channel_cntl_stat(nxgep, channel, &cs);
4375 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_rxdma_start_channel: "
4376 		"channel %d rx_dma_cntl_stat 0x%0016llx", channel, cs.value));
4377 	if (status != NXGE_OK) {
4378 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4379 			"==> nxge_rxdma_start_channel: "
4380 			"init rxdma control register failed (0x%08x channel %d",
4381 			status, channel));
4382 		return (status);
4383 	}
4384 
4385 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_rxdma_start_channel: "
4386 		"control done - channel %d cs 0x%016llx", channel, cs.value));
4387 
4388 	/*
4389 	 * Load RXDMA descriptors, buffers, mailbox,
4390 	 * initialise the receive DMA channels and
4391 	 * enable each DMA channel.
4392 	 */
4393 	status = nxge_enable_rxdma_channel(nxgep,
4394 	    channel, rbr_p, rcr_p, mbox_p);
4395 
4396 	if (status != NXGE_OK) {
4397 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4398 		    " nxge_rxdma_start_channel: "
4399 		    " enable rxdma failed (0x%08x channel %d)",
4400 		    status, channel));
4401 		return (status);
4402 	}
4403 
4404 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4405 	    "==> nxge_rxdma_start_channel: enabled channel %d"));
4406 
4407 	if (isLDOMguest(nxgep)) {
4408 		/* Add interrupt handler for this channel. */
4409 		if (nxge_hio_intr_add(nxgep, VP_BOUND_RX, channel)
4410 		    != NXGE_OK) {
4411 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4412 			    " nxge_rxdma_start_channel: "
4413 			    " nxge_hio_intr_add failed (0x%08x channel %d)",
4414 		    status, channel));
4415 		}
4416 	}
4417 
4418 	ent_mask.value = 0;
4419 	ent_mask.value |= (RX_DMA_ENT_MSK_WRED_DROP_MASK |
4420 				RX_DMA_ENT_MSK_PTDROP_PKT_MASK);
4421 	rs = npi_rxdma_event_mask(handle, OP_SET, channel,
4422 			&ent_mask);
4423 	if (rs != NPI_SUCCESS) {
4424 		NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
4425 			"==> nxge_rxdma_start_channel: "
4426 			"init rxdma event masks failed (0x%08x channel %d)",
4427 			status, channel));
4428 		return (NXGE_ERROR | rs);
4429 	}
4430 
4431 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_rxdma_start_channel: "
4432 		"control done - channel %d cs 0x%016llx", channel, cs.value));
4433 
4434 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "<== nxge_rxdma_start_channel"));
4435 
4436 	return (NXGE_OK);
4437 }
4438 
4439 static nxge_status_t
4440 nxge_rxdma_stop_channel(p_nxge_t nxgep, uint16_t channel)
4441 {
4442 	npi_handle_t		handle;
4443 	npi_status_t		rs = NPI_SUCCESS;
4444 	rx_dma_ctl_stat_t	cs;
4445 	rx_dma_ent_msk_t	ent_mask;
4446 	nxge_status_t		status = NXGE_OK;
4447 
4448 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "==> nxge_rxdma_stop_channel"));
4449 
4450 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
4451 
4452 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "nxge_rxdma_stop_channel: "
4453 	    "npi handle addr $%p acc $%p",
4454 	    nxgep->npi_handle.regp, nxgep->npi_handle.regh));
4455 
4456 	if (!isLDOMguest(nxgep)) {
4457 		/*
4458 		 * Stop RxMAC = A.9.2.6
4459 		 */
4460 		if (nxge_rx_mac_disable(nxgep) != NXGE_OK) {
4461 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4462 			    "nxge_rxdma_stop_channel: "
4463 			    "Failed to disable RxMAC"));
4464 		}
4465 
4466 		/*
4467 		 * Drain IPP Port = A.9.3.6
4468 		 */
4469 		(void) nxge_ipp_drain(nxgep);
4470 	}
4471 
4472 	/* Reset RXDMA channel */
4473 	rs = npi_rxdma_cfg_rdc_reset(handle, channel);
4474 	if (rs != NPI_SUCCESS) {
4475 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4476 		    " nxge_rxdma_stop_channel: "
4477 		    " reset rxdma failed (0x%08x channel %d)",
4478 		    rs, channel));
4479 		return (NXGE_ERROR | rs);
4480 	}
4481 
4482 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
4483 	    "==> nxge_rxdma_stop_channel: reset done"));
4484 
4485 	/* Set up the interrupt event masks. */
4486 	ent_mask.value = RX_DMA_ENT_MSK_ALL;
4487 	rs = npi_rxdma_event_mask(handle, OP_SET, channel,
4488 	    &ent_mask);
4489 	if (rs != NPI_SUCCESS) {
4490 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4491 		    "==> nxge_rxdma_stop_channel: "
4492 		    "set rxdma event masks failed (0x%08x channel %d)",
4493 		    rs, channel));
4494 		return (NXGE_ERROR | rs);
4495 	}
4496 
4497 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
4498 	    "==> nxge_rxdma_stop_channel: event done"));
4499 
4500 	/*
4501 	 * Initialize the receive DMA control and status register
4502 	 */
4503 	cs.value = 0;
4504 	status = nxge_init_rxdma_channel_cntl_stat(nxgep, channel, &cs);
4505 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "==> nxge_rxdma_stop_channel: control "
4506 	    " to default (all 0s) 0x%08x", cs.value));
4507 	if (status != NXGE_OK) {
4508 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4509 		    " nxge_rxdma_stop_channel: init rxdma"
4510 		    " control register failed (0x%08x channel %d",
4511 		    status, channel));
4512 		return (status);
4513 	}
4514 
4515 	NXGE_DEBUG_MSG((nxgep, RX_CTL,
4516 	    "==> nxge_rxdma_stop_channel: control done"));
4517 
4518 	/*
4519 	 * Make sure channel is disabled.
4520 	 */
4521 	status = nxge_disable_rxdma_channel(nxgep, channel);
4522 
4523 	if (status != NXGE_OK) {
4524 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4525 		    " nxge_rxdma_stop_channel: "
4526 		    " init enable rxdma failed (0x%08x channel %d)",
4527 		    status, channel));
4528 		return (status);
4529 	}
4530 
4531 	if (!isLDOMguest(nxgep)) {
4532 		/*
4533 		 * Enable RxMAC = A.9.2.10
4534 		 */
4535 		if (nxge_rx_mac_enable(nxgep) != NXGE_OK) {
4536 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4537 			    "nxge_rxdma_stop_channel: Rx MAC still disabled"));
4538 		}
4539 	}
4540 
4541 	NXGE_DEBUG_MSG((nxgep,
4542 	    RX_CTL, "==> nxge_rxdma_stop_channel: disable done"));
4543 
4544 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "<== nxge_rxdma_stop_channel"));
4545 
4546 	return (NXGE_OK);
4547 }
4548 
4549 nxge_status_t
4550 nxge_rxdma_handle_sys_errors(p_nxge_t nxgep)
4551 {
4552 	npi_handle_t		handle;
4553 	p_nxge_rdc_sys_stats_t	statsp;
4554 	rx_ctl_dat_fifo_stat_t	stat;
4555 	uint32_t		zcp_err_status;
4556 	uint32_t		ipp_err_status;
4557 	nxge_status_t		status = NXGE_OK;
4558 	npi_status_t		rs = NPI_SUCCESS;
4559 	boolean_t		my_err = B_FALSE;
4560 
4561 	handle = nxgep->npi_handle;
4562 	statsp = (p_nxge_rdc_sys_stats_t)&nxgep->statsp->rdc_sys_stats;
4563 
4564 	rs = npi_rxdma_rxctl_fifo_error_intr_get(handle, &stat);
4565 
4566 	if (rs != NPI_SUCCESS)
4567 		return (NXGE_ERROR | rs);
4568 
4569 	if (stat.bits.ldw.id_mismatch) {
4570 		statsp->id_mismatch++;
4571 		NXGE_FM_REPORT_ERROR(nxgep, nxgep->mac.portnum, NULL,
4572 		    NXGE_FM_EREPORT_RDMC_ID_MISMATCH);
4573 		/* Global fatal error encountered */
4574 	}
4575 
4576 	if ((stat.bits.ldw.zcp_eop_err) || (stat.bits.ldw.ipp_eop_err)) {
4577 		switch (nxgep->mac.portnum) {
4578 		case 0:
4579 			if ((stat.bits.ldw.zcp_eop_err & FIFO_EOP_PORT0) ||
4580 			    (stat.bits.ldw.ipp_eop_err & FIFO_EOP_PORT0)) {
4581 				my_err = B_TRUE;
4582 				zcp_err_status = stat.bits.ldw.zcp_eop_err;
4583 				ipp_err_status = stat.bits.ldw.ipp_eop_err;
4584 			}
4585 			break;
4586 		case 1:
4587 			if ((stat.bits.ldw.zcp_eop_err & FIFO_EOP_PORT1) ||
4588 			    (stat.bits.ldw.ipp_eop_err & FIFO_EOP_PORT1)) {
4589 				my_err = B_TRUE;
4590 				zcp_err_status = stat.bits.ldw.zcp_eop_err;
4591 				ipp_err_status = stat.bits.ldw.ipp_eop_err;
4592 			}
4593 			break;
4594 		case 2:
4595 			if ((stat.bits.ldw.zcp_eop_err & FIFO_EOP_PORT2) ||
4596 			    (stat.bits.ldw.ipp_eop_err & FIFO_EOP_PORT2)) {
4597 				my_err = B_TRUE;
4598 				zcp_err_status = stat.bits.ldw.zcp_eop_err;
4599 				ipp_err_status = stat.bits.ldw.ipp_eop_err;
4600 			}
4601 			break;
4602 		case 3:
4603 			if ((stat.bits.ldw.zcp_eop_err & FIFO_EOP_PORT3) ||
4604 			    (stat.bits.ldw.ipp_eop_err & FIFO_EOP_PORT3)) {
4605 				my_err = B_TRUE;
4606 				zcp_err_status = stat.bits.ldw.zcp_eop_err;
4607 				ipp_err_status = stat.bits.ldw.ipp_eop_err;
4608 			}
4609 			break;
4610 		default:
4611 			return (NXGE_ERROR);
4612 		}
4613 	}
4614 
4615 	if (my_err) {
4616 		status = nxge_rxdma_handle_port_errors(nxgep, ipp_err_status,
4617 		    zcp_err_status);
4618 		if (status != NXGE_OK)
4619 			return (status);
4620 	}
4621 
4622 	return (NXGE_OK);
4623 }
4624 
4625 static nxge_status_t
4626 nxge_rxdma_handle_port_errors(p_nxge_t nxgep, uint32_t ipp_status,
4627 							uint32_t zcp_status)
4628 {
4629 	boolean_t		rxport_fatal = B_FALSE;
4630 	p_nxge_rdc_sys_stats_t	statsp;
4631 	nxge_status_t		status = NXGE_OK;
4632 	uint8_t			portn;
4633 
4634 	portn = nxgep->mac.portnum;
4635 	statsp = (p_nxge_rdc_sys_stats_t)&nxgep->statsp->rdc_sys_stats;
4636 
4637 	if (ipp_status & (0x1 << portn)) {
4638 		statsp->ipp_eop_err++;
4639 		NXGE_FM_REPORT_ERROR(nxgep, portn, NULL,
4640 		    NXGE_FM_EREPORT_RDMC_IPP_EOP_ERR);
4641 		rxport_fatal = B_TRUE;
4642 	}
4643 
4644 	if (zcp_status & (0x1 << portn)) {
4645 		statsp->zcp_eop_err++;
4646 		NXGE_FM_REPORT_ERROR(nxgep, portn, NULL,
4647 		    NXGE_FM_EREPORT_RDMC_ZCP_EOP_ERR);
4648 		rxport_fatal = B_TRUE;
4649 	}
4650 
4651 	if (rxport_fatal) {
4652 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4653 		    " nxge_rxdma_handle_port_error: "
4654 		    " fatal error on Port #%d\n",
4655 		    portn));
4656 		status = nxge_rx_port_fatal_err_recover(nxgep);
4657 		if (status == NXGE_OK) {
4658 			FM_SERVICE_RESTORED(nxgep);
4659 		}
4660 	}
4661 
4662 	return (status);
4663 }
4664 
4665 static nxge_status_t
4666 nxge_rxdma_fatal_err_recover(p_nxge_t nxgep, uint16_t channel)
4667 {
4668 	npi_handle_t		handle;
4669 	npi_status_t		rs = NPI_SUCCESS;
4670 	nxge_status_t		status = NXGE_OK;
4671 	p_rx_rbr_ring_t		rbrp;
4672 	p_rx_rcr_ring_t		rcrp;
4673 	p_rx_mbox_t		mboxp;
4674 	rx_dma_ent_msk_t	ent_mask;
4675 	p_nxge_dma_common_t	dmap;
4676 	int			ring_idx;
4677 	uint32_t		ref_cnt;
4678 	p_rx_msg_t		rx_msg_p;
4679 	int			i;
4680 	uint32_t		nxge_port_rcr_size;
4681 
4682 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "<== nxge_rxdma_fatal_err_recover"));
4683 	NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4684 	    "Recovering from RxDMAChannel#%d error...", channel));
4685 
4686 	/*
4687 	 * Stop the dma channel waits for the stop done.
4688 	 * If the stop done bit is not set, then create
4689 	 * an error.
4690 	 */
4691 
4692 	handle = NXGE_DEV_NPI_HANDLE(nxgep);
4693 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "Rx DMA stop..."));
4694 
4695 	ring_idx = nxge_rxdma_get_ring_index(nxgep, channel);
4696 	rbrp = (p_rx_rbr_ring_t)nxgep->rx_rbr_rings->rbr_rings[ring_idx];
4697 	rcrp = (p_rx_rcr_ring_t)nxgep->rx_rcr_rings->rcr_rings[ring_idx];
4698 
4699 	MUTEX_ENTER(&rcrp->lock);
4700 	MUTEX_ENTER(&rbrp->lock);
4701 	MUTEX_ENTER(&rbrp->post_lock);
4702 
4703 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "Disable RxDMA channel..."));
4704 
4705 	rs = npi_rxdma_cfg_rdc_disable(handle, channel);
4706 	if (rs != NPI_SUCCESS) {
4707 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4708 		    "nxge_disable_rxdma_channel:failed"));
4709 		goto fail;
4710 	}
4711 
4712 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "Disable RxDMA interrupt..."));
4713 
4714 	/* Disable interrupt */
4715 	ent_mask.value = RX_DMA_ENT_MSK_ALL;
4716 	rs = npi_rxdma_event_mask(handle, OP_SET, channel, &ent_mask);
4717 	if (rs != NPI_SUCCESS) {
4718 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4719 		    "nxge_rxdma_stop_channel: "
4720 		    "set rxdma event masks failed (channel %d)",
4721 		    channel));
4722 	}
4723 
4724 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "RxDMA channel reset..."));
4725 
4726 	/* Reset RXDMA channel */
4727 	rs = npi_rxdma_cfg_rdc_reset(handle, channel);
4728 	if (rs != NPI_SUCCESS) {
4729 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4730 		    "nxge_rxdma_fatal_err_recover: "
4731 		    " reset rxdma failed (channel %d)", channel));
4732 		goto fail;
4733 	}
4734 
4735 	nxge_port_rcr_size = nxgep->nxge_port_rcr_size;
4736 
4737 	mboxp =
4738 	    (p_rx_mbox_t)nxgep->rx_mbox_areas_p->rxmbox_areas[ring_idx];
4739 
4740 	rbrp->rbr_wr_index = (rbrp->rbb_max - 1);
4741 	rbrp->rbr_rd_index = 0;
4742 
4743 	rcrp->comp_rd_index = 0;
4744 	rcrp->comp_wt_index = 0;
4745 	rcrp->rcr_desc_rd_head_p = rcrp->rcr_desc_first_p =
4746 	    (p_rcr_entry_t)DMA_COMMON_VPTR(rcrp->rcr_desc);
4747 #if defined(__i386)
4748 	rcrp->rcr_desc_rd_head_pp = rcrp->rcr_desc_first_pp =
4749 	    (p_rcr_entry_t)(uint32_t)DMA_COMMON_IOADDR(rcrp->rcr_desc);
4750 #else
4751 	rcrp->rcr_desc_rd_head_pp = rcrp->rcr_desc_first_pp =
4752 	    (p_rcr_entry_t)DMA_COMMON_IOADDR(rcrp->rcr_desc);
4753 #endif
4754 
4755 	rcrp->rcr_desc_last_p = rcrp->rcr_desc_rd_head_p +
4756 	    (nxge_port_rcr_size - 1);
4757 	rcrp->rcr_desc_last_pp = rcrp->rcr_desc_rd_head_pp +
4758 	    (nxge_port_rcr_size - 1);
4759 
4760 	dmap = (p_nxge_dma_common_t)&rcrp->rcr_desc;
4761 	bzero((caddr_t)dmap->kaddrp, dmap->alength);
4762 
4763 	cmn_err(CE_NOTE, "!rbr entries = %d\n", rbrp->rbr_max_size);
4764 
4765 	for (i = 0; i < rbrp->rbr_max_size; i++) {
4766 		rx_msg_p = rbrp->rx_msg_ring[i];
4767 		ref_cnt = rx_msg_p->ref_cnt;
4768 		if (ref_cnt != 1) {
4769 			if (rx_msg_p->cur_usage_cnt !=
4770 			    rx_msg_p->max_usage_cnt) {
4771 				NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4772 				    "buf[%d]: cur_usage_cnt = %d "
4773 				    "max_usage_cnt = %d\n", i,
4774 				    rx_msg_p->cur_usage_cnt,
4775 				    rx_msg_p->max_usage_cnt));
4776 			} else {
4777 				/* Buffer can be re-posted */
4778 				rx_msg_p->free = B_TRUE;
4779 				rx_msg_p->cur_usage_cnt = 0;
4780 				rx_msg_p->max_usage_cnt = 0xbaddcafe;
4781 				rx_msg_p->pkt_buf_size = 0;
4782 			}
4783 		}
4784 	}
4785 
4786 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "RxDMA channel re-start..."));
4787 
4788 	status = nxge_rxdma_start_channel(nxgep, channel, rbrp, rcrp, mboxp);
4789 	if (status != NXGE_OK) {
4790 		goto fail;
4791 	}
4792 
4793 	MUTEX_EXIT(&rbrp->post_lock);
4794 	MUTEX_EXIT(&rbrp->lock);
4795 	MUTEX_EXIT(&rcrp->lock);
4796 
4797 	NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4798 	    "Recovery Successful, RxDMAChannel#%d Restored",
4799 	    channel));
4800 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "==> nxge_rxdma_fatal_err_recover"));
4801 
4802 	return (NXGE_OK);
4803 fail:
4804 	MUTEX_EXIT(&rbrp->post_lock);
4805 	MUTEX_EXIT(&rbrp->lock);
4806 	MUTEX_EXIT(&rcrp->lock);
4807 	NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL, "Recovery failed"));
4808 
4809 	return (NXGE_ERROR | rs);
4810 }
4811 
4812 nxge_status_t
4813 nxge_rx_port_fatal_err_recover(p_nxge_t nxgep)
4814 {
4815 	nxge_grp_set_t *set = &nxgep->rx_set;
4816 	nxge_status_t status = NXGE_OK;
4817 	int rdc;
4818 
4819 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "<== nxge_rx_port_fatal_err_recover"));
4820 	NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4821 	    "Recovering from RxPort error..."));
4822 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "Disabling RxMAC...\n"));
4823 
4824 	if (nxge_rx_mac_disable(nxgep) != NXGE_OK)
4825 		goto fail;
4826 
4827 	NXGE_DELAY(1000);
4828 
4829 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "Stopping all RxDMA channels..."));
4830 
4831 	for (rdc = 0; rdc < NXGE_MAX_RDCS; rdc++) {
4832 		if ((1 << rdc) & set->owned.map) {
4833 			if (nxge_rxdma_fatal_err_recover(nxgep, rdc)
4834 			    != NXGE_OK) {
4835 				NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4836 				    "Could not recover channel %d", rdc));
4837 			}
4838 		}
4839 	}
4840 
4841 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "Resetting IPP..."));
4842 
4843 	/* Reset IPP */
4844 	if (nxge_ipp_reset(nxgep) != NXGE_OK) {
4845 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4846 		    "nxge_rx_port_fatal_err_recover: "
4847 		    "Failed to reset IPP"));
4848 		goto fail;
4849 	}
4850 
4851 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "Reset RxMAC..."));
4852 
4853 	/* Reset RxMAC */
4854 	if (nxge_rx_mac_reset(nxgep) != NXGE_OK) {
4855 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4856 		    "nxge_rx_port_fatal_err_recover: "
4857 		    "Failed to reset RxMAC"));
4858 		goto fail;
4859 	}
4860 
4861 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "Re-initialize IPP..."));
4862 
4863 	/* Re-Initialize IPP */
4864 	if (nxge_ipp_init(nxgep) != NXGE_OK) {
4865 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4866 		    "nxge_rx_port_fatal_err_recover: "
4867 		    "Failed to init IPP"));
4868 		goto fail;
4869 	}
4870 
4871 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "Re-initialize RxMAC..."));
4872 
4873 	/* Re-Initialize RxMAC */
4874 	if ((status = nxge_rx_mac_init(nxgep)) != NXGE_OK) {
4875 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4876 		    "nxge_rx_port_fatal_err_recover: "
4877 		    "Failed to reset RxMAC"));
4878 		goto fail;
4879 	}
4880 
4881 	NXGE_DEBUG_MSG((nxgep, RX_CTL, "Re-enable RxMAC..."));
4882 
4883 	/* Re-enable RxMAC */
4884 	if ((status = nxge_rx_mac_enable(nxgep)) != NXGE_OK) {
4885 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4886 		    "nxge_rx_port_fatal_err_recover: "
4887 		    "Failed to enable RxMAC"));
4888 		goto fail;
4889 	}
4890 
4891 	NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4892 	    "Recovery Successful, RxPort Restored"));
4893 
4894 	return (NXGE_OK);
4895 fail:
4896 	NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL, "Recovery failed"));
4897 	return (status);
4898 }
4899 
4900 void
4901 nxge_rxdma_inject_err(p_nxge_t nxgep, uint32_t err_id, uint8_t chan)
4902 {
4903 	rx_dma_ctl_stat_t	cs;
4904 	rx_ctl_dat_fifo_stat_t	cdfs;
4905 
4906 	switch (err_id) {
4907 	case NXGE_FM_EREPORT_RDMC_RCR_ACK_ERR:
4908 	case NXGE_FM_EREPORT_RDMC_DC_FIFO_ERR:
4909 	case NXGE_FM_EREPORT_RDMC_RCR_SHA_PAR:
4910 	case NXGE_FM_EREPORT_RDMC_RBR_PRE_PAR:
4911 	case NXGE_FM_EREPORT_RDMC_RBR_TMOUT:
4912 	case NXGE_FM_EREPORT_RDMC_RSP_CNT_ERR:
4913 	case NXGE_FM_EREPORT_RDMC_BYTE_EN_BUS:
4914 	case NXGE_FM_EREPORT_RDMC_RSP_DAT_ERR:
4915 	case NXGE_FM_EREPORT_RDMC_RCRINCON:
4916 	case NXGE_FM_EREPORT_RDMC_RCRFULL:
4917 	case NXGE_FM_EREPORT_RDMC_RBRFULL:
4918 	case NXGE_FM_EREPORT_RDMC_RBRLOGPAGE:
4919 	case NXGE_FM_EREPORT_RDMC_CFIGLOGPAGE:
4920 	case NXGE_FM_EREPORT_RDMC_CONFIG_ERR:
4921 		RXDMA_REG_READ64(nxgep->npi_handle, RX_DMA_CTL_STAT_DBG_REG,
4922 		    chan, &cs.value);
4923 		if (err_id == NXGE_FM_EREPORT_RDMC_RCR_ACK_ERR)
4924 			cs.bits.hdw.rcr_ack_err = 1;
4925 		else if (err_id == NXGE_FM_EREPORT_RDMC_DC_FIFO_ERR)
4926 			cs.bits.hdw.dc_fifo_err = 1;
4927 		else if (err_id == NXGE_FM_EREPORT_RDMC_RCR_SHA_PAR)
4928 			cs.bits.hdw.rcr_sha_par = 1;
4929 		else if (err_id == NXGE_FM_EREPORT_RDMC_RBR_PRE_PAR)
4930 			cs.bits.hdw.rbr_pre_par = 1;
4931 		else if (err_id == NXGE_FM_EREPORT_RDMC_RBR_TMOUT)
4932 			cs.bits.hdw.rbr_tmout = 1;
4933 		else if (err_id == NXGE_FM_EREPORT_RDMC_RSP_CNT_ERR)
4934 			cs.bits.hdw.rsp_cnt_err = 1;
4935 		else if (err_id == NXGE_FM_EREPORT_RDMC_BYTE_EN_BUS)
4936 			cs.bits.hdw.byte_en_bus = 1;
4937 		else if (err_id == NXGE_FM_EREPORT_RDMC_RSP_DAT_ERR)
4938 			cs.bits.hdw.rsp_dat_err = 1;
4939 		else if (err_id == NXGE_FM_EREPORT_RDMC_CONFIG_ERR)
4940 			cs.bits.hdw.config_err = 1;
4941 		else if (err_id == NXGE_FM_EREPORT_RDMC_RCRINCON)
4942 			cs.bits.hdw.rcrincon = 1;
4943 		else if (err_id == NXGE_FM_EREPORT_RDMC_RCRFULL)
4944 			cs.bits.hdw.rcrfull = 1;
4945 		else if (err_id == NXGE_FM_EREPORT_RDMC_RBRFULL)
4946 			cs.bits.hdw.rbrfull = 1;
4947 		else if (err_id == NXGE_FM_EREPORT_RDMC_RBRLOGPAGE)
4948 			cs.bits.hdw.rbrlogpage = 1;
4949 		else if (err_id == NXGE_FM_EREPORT_RDMC_CFIGLOGPAGE)
4950 			cs.bits.hdw.cfiglogpage = 1;
4951 #if defined(__i386)
4952 		cmn_err(CE_NOTE, "!Write 0x%llx to RX_DMA_CTL_STAT_DBG_REG\n",
4953 		    cs.value);
4954 #else
4955 		cmn_err(CE_NOTE, "!Write 0x%lx to RX_DMA_CTL_STAT_DBG_REG\n",
4956 		    cs.value);
4957 #endif
4958 		RXDMA_REG_WRITE64(nxgep->npi_handle, RX_DMA_CTL_STAT_DBG_REG,
4959 		    chan, cs.value);
4960 		break;
4961 	case NXGE_FM_EREPORT_RDMC_ID_MISMATCH:
4962 	case NXGE_FM_EREPORT_RDMC_ZCP_EOP_ERR:
4963 	case NXGE_FM_EREPORT_RDMC_IPP_EOP_ERR:
4964 		cdfs.value = 0;
4965 		if (err_id ==  NXGE_FM_EREPORT_RDMC_ID_MISMATCH)
4966 			cdfs.bits.ldw.id_mismatch = (1 << nxgep->mac.portnum);
4967 		else if (err_id == NXGE_FM_EREPORT_RDMC_ZCP_EOP_ERR)
4968 			cdfs.bits.ldw.zcp_eop_err = (1 << nxgep->mac.portnum);
4969 		else if (err_id == NXGE_FM_EREPORT_RDMC_IPP_EOP_ERR)
4970 			cdfs.bits.ldw.ipp_eop_err = (1 << nxgep->mac.portnum);
4971 #if defined(__i386)
4972 		cmn_err(CE_NOTE,
4973 		    "!Write 0x%llx to RX_CTL_DAT_FIFO_STAT_DBG_REG\n",
4974 		    cdfs.value);
4975 #else
4976 		cmn_err(CE_NOTE,
4977 		    "!Write 0x%lx to RX_CTL_DAT_FIFO_STAT_DBG_REG\n",
4978 		    cdfs.value);
4979 #endif
4980 		NXGE_REG_WR64(nxgep->npi_handle,
4981 		    RX_CTL_DAT_FIFO_STAT_DBG_REG, cdfs.value);
4982 		break;
4983 	case NXGE_FM_EREPORT_RDMC_DCF_ERR:
4984 		break;
4985 	case NXGE_FM_EREPORT_RDMC_RCR_ERR:
4986 		break;
4987 	}
4988 }
4989 
4990 static void
4991 nxge_rxdma_databuf_free(p_rx_rbr_ring_t rbr_p)
4992 {
4993 	rxring_info_t 		*ring_info;
4994 	int			index;
4995 	uint32_t		chunk_size;
4996 	uint64_t		kaddr;
4997 	uint_t			num_blocks;
4998 
4999 	NXGE_DEBUG_MSG((NULL, DMA_CTL, "==> nxge_rxdma_databuf_free"));
5000 
5001 	if (rbr_p == NULL) {
5002 		NXGE_ERROR_MSG((NULL, NXGE_ERR_CTL,
5003 		    "==> nxge_rxdma_databuf_free: NULL rbr pointer"));
5004 		return;
5005 	}
5006 
5007 	if (rbr_p->rbr_alloc_type == DDI_MEM_ALLOC) {
5008 		NXGE_DEBUG_MSG((NULL, DMA_CTL,
5009 		    "<== nxge_rxdma_databuf_free: DDI"));
5010 		return;
5011 	}
5012 
5013 	ring_info = rbr_p->ring_info;
5014 	if (ring_info == NULL) {
5015 		NXGE_ERROR_MSG((NULL, NXGE_ERR_CTL,
5016 		    "==> nxge_rxdma_databuf_free: NULL ring info"));
5017 		return;
5018 	}
5019 	num_blocks = rbr_p->num_blocks;
5020 	for (index = 0; index < num_blocks; index++) {
5021 		kaddr = ring_info->buffer[index].kaddr;
5022 		chunk_size = ring_info->buffer[index].buf_size;
5023 		NXGE_DEBUG_MSG((NULL, DMA_CTL,
5024 		    "==> nxge_rxdma_databuf_free: free chunk %d "
5025 		    "kaddrp $%p chunk size %d",
5026 		    index, kaddr, chunk_size));
5027 		if (kaddr == NULL) continue;
5028 		nxge_free_buf(rbr_p->rbr_alloc_type, kaddr, chunk_size);
5029 		ring_info->buffer[index].kaddr = NULL;
5030 	}
5031 
5032 	NXGE_DEBUG_MSG((NULL, DMA_CTL, "<== nxge_rxdma_databuf_free"));
5033 }
5034 
5035 #if	defined(sun4v) && defined(NIU_LP_WORKAROUND)
5036 extern void contig_mem_free(void *, size_t);
5037 #endif
5038 
5039 void
5040 nxge_free_buf(buf_alloc_type_t alloc_type, uint64_t kaddr, uint32_t buf_size)
5041 {
5042 	NXGE_DEBUG_MSG((NULL, DMA_CTL, "==> nxge_free_buf"));
5043 
5044 	if (kaddr == NULL || !buf_size) {
5045 		NXGE_ERROR_MSG((NULL, NXGE_ERR_CTL,
5046 		    "==> nxge_free_buf: invalid kaddr $%p size to free %d",
5047 		    kaddr, buf_size));
5048 		return;
5049 	}
5050 
5051 	switch (alloc_type) {
5052 	case KMEM_ALLOC:
5053 		NXGE_DEBUG_MSG((NULL, DMA_CTL,
5054 		    "==> nxge_free_buf: freeing kmem $%p size %d",
5055 		    kaddr, buf_size));
5056 #if defined(__i386)
5057 		KMEM_FREE((void *)(uint32_t)kaddr, buf_size);
5058 #else
5059 		KMEM_FREE((void *)kaddr, buf_size);
5060 #endif
5061 		break;
5062 
5063 #if	defined(sun4v) && defined(NIU_LP_WORKAROUND)
5064 	case CONTIG_MEM_ALLOC:
5065 		NXGE_DEBUG_MSG((NULL, DMA_CTL,
5066 		    "==> nxge_free_buf: freeing contig_mem kaddr $%p size %d",
5067 		    kaddr, buf_size));
5068 		contig_mem_free((void *)kaddr, buf_size);
5069 		break;
5070 #endif
5071 
5072 	default:
5073 		NXGE_ERROR_MSG((NULL, NXGE_ERR_CTL,
5074 		    "<== nxge_free_buf: unsupported alloc type %d",
5075 		    alloc_type));
5076 		return;
5077 	}
5078 
5079 	NXGE_DEBUG_MSG((NULL, DMA_CTL, "<== nxge_free_buf"));
5080 }
5081