xref: /titanic_41/usr/src/uts/common/io/hxge/hxge_rxdma.c (revision b7d3956b92a285d8dac2c7f5f7e28d2ef5347ef8)
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 2008 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 #include <hxge_impl.h>
29 #include <hxge_rxdma.h>
30 
31 /*
32  * Globals: tunable parameters (/etc/system or adb)
33  *
34  */
35 extern uint32_t hxge_rbr_size;
36 extern uint32_t hxge_rcr_size;
37 extern uint32_t hxge_rbr_spare_size;
38 
39 extern uint32_t hxge_mblks_pending;
40 
41 /*
42  * Tunable to reduce the amount of time spent in the
43  * ISR doing Rx Processing.
44  */
45 extern uint32_t hxge_max_rx_pkts;
46 boolean_t hxge_jumbo_enable;
47 
48 /*
49  * Tunables to manage the receive buffer blocks.
50  *
51  * hxge_rx_threshold_hi: copy all buffers.
52  * hxge_rx_bcopy_size_type: receive buffer block size type.
53  * hxge_rx_threshold_lo: copy only up to tunable block size type.
54  */
55 extern hxge_rxbuf_threshold_t hxge_rx_threshold_hi;
56 extern hxge_rxbuf_type_t hxge_rx_buf_size_type;
57 extern hxge_rxbuf_threshold_t hxge_rx_threshold_lo;
58 
59 static hxge_status_t hxge_map_rxdma(p_hxge_t hxgep);
60 static void hxge_unmap_rxdma(p_hxge_t hxgep);
61 static hxge_status_t hxge_rxdma_hw_start_common(p_hxge_t hxgep);
62 static hxge_status_t hxge_rxdma_hw_start(p_hxge_t hxgep);
63 static void hxge_rxdma_hw_stop(p_hxge_t hxgep);
64 static hxge_status_t hxge_map_rxdma_channel(p_hxge_t hxgep, uint16_t channel,
65 	p_hxge_dma_common_t *dma_buf_p, p_rx_rbr_ring_t *rbr_p,
66 	uint32_t num_chunks, p_hxge_dma_common_t *dma_cntl_p,
67 	p_rx_rcr_ring_t *rcr_p, p_rx_mbox_t *rx_mbox_p);
68 static void hxge_unmap_rxdma_channel(p_hxge_t hxgep, uint16_t channel,
69 	p_rx_rbr_ring_t rbr_p, p_rx_rcr_ring_t rcr_p, p_rx_mbox_t rx_mbox_p);
70 static hxge_status_t hxge_map_rxdma_channel_cfg_ring(p_hxge_t hxgep,
71 	uint16_t dma_channel, p_hxge_dma_common_t *dma_cntl_p,
72 	p_rx_rbr_ring_t *rbr_p, p_rx_rcr_ring_t *rcr_p, p_rx_mbox_t *rx_mbox_p);
73 static void hxge_unmap_rxdma_channel_cfg_ring(p_hxge_t hxgep,
74 	p_rx_rcr_ring_t rcr_p, p_rx_mbox_t rx_mbox_p);
75 static hxge_status_t hxge_map_rxdma_channel_buf_ring(p_hxge_t hxgep,
76 	uint16_t channel, p_hxge_dma_common_t *dma_buf_p,
77 	p_rx_rbr_ring_t *rbr_p, uint32_t num_chunks);
78 static void hxge_unmap_rxdma_channel_buf_ring(p_hxge_t hxgep,
79 	p_rx_rbr_ring_t rbr_p);
80 static hxge_status_t hxge_rxdma_start_channel(p_hxge_t hxgep, uint16_t channel,
81 	p_rx_rbr_ring_t rbr_p, p_rx_rcr_ring_t rcr_p, p_rx_mbox_t mbox_p);
82 static hxge_status_t hxge_rxdma_stop_channel(p_hxge_t hxgep, uint16_t channel);
83 static mblk_t *hxge_rx_pkts(p_hxge_t hxgep, uint_t vindex, p_hxge_ldv_t ldvp,
84 	p_rx_rcr_ring_t	*rcr_p, rdc_stat_t cs);
85 static void hxge_receive_packet(p_hxge_t hxgep, p_rx_rcr_ring_t rcr_p,
86 	p_rcr_entry_t rcr_desc_rd_head_p, boolean_t *multi_p,
87 	mblk_t ** mp, mblk_t ** mp_cont, uint32_t *invalid_rcr_entry);
88 static hxge_status_t hxge_disable_rxdma_channel(p_hxge_t hxgep,
89 	uint16_t channel);
90 static p_rx_msg_t hxge_allocb(size_t, uint32_t, p_hxge_dma_common_t);
91 static void hxge_freeb(p_rx_msg_t);
92 static void hxge_rx_pkts_vring(p_hxge_t hxgep, uint_t vindex,
93     p_hxge_ldv_t ldvp, rdc_stat_t cs);
94 static hxge_status_t hxge_rx_err_evnts(p_hxge_t hxgep, uint_t index,
95 	p_hxge_ldv_t ldvp, rdc_stat_t cs);
96 static hxge_status_t hxge_rxbuf_index_info_init(p_hxge_t hxgep,
97 	p_rx_rbr_ring_t rx_dmap);
98 static hxge_status_t hxge_rxdma_fatal_err_recover(p_hxge_t hxgep,
99 	uint16_t channel);
100 static hxge_status_t hxge_rx_port_fatal_err_recover(p_hxge_t hxgep);
101 
102 hxge_status_t
103 hxge_init_rxdma_channels(p_hxge_t hxgep)
104 {
105 	hxge_status_t		status = HXGE_OK;
106 	block_reset_t		reset_reg;
107 
108 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_init_rxdma_channels"));
109 
110 	/* Reset RDC block from PEU to clear any previous state */
111 	reset_reg.value = 0;
112 	reset_reg.bits.rdc_rst = 1;
113 	HXGE_REG_WR32(hxgep->hpi_handle, BLOCK_RESET, reset_reg.value);
114 	HXGE_DELAY(1000);
115 
116 	status = hxge_map_rxdma(hxgep);
117 	if (status != HXGE_OK) {
118 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
119 		    "<== hxge_init_rxdma: status 0x%x", status));
120 		return (status);
121 	}
122 
123 	status = hxge_rxdma_hw_start_common(hxgep);
124 	if (status != HXGE_OK) {
125 		hxge_unmap_rxdma(hxgep);
126 	}
127 
128 	status = hxge_rxdma_hw_start(hxgep);
129 	if (status != HXGE_OK) {
130 		hxge_unmap_rxdma(hxgep);
131 	}
132 
133 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
134 	    "<== hxge_init_rxdma_channels: status 0x%x", status));
135 	return (status);
136 }
137 
138 void
139 hxge_uninit_rxdma_channels(p_hxge_t hxgep)
140 {
141 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_uninit_rxdma_channels"));
142 
143 	hxge_rxdma_hw_stop(hxgep);
144 	hxge_unmap_rxdma(hxgep);
145 
146 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "<== hxge_uinit_rxdma_channels"));
147 }
148 
149 hxge_status_t
150 hxge_init_rxdma_channel_cntl_stat(p_hxge_t hxgep, uint16_t channel,
151     rdc_stat_t *cs_p)
152 {
153 	hpi_handle_t	handle;
154 	hpi_status_t	rs = HPI_SUCCESS;
155 	hxge_status_t	status = HXGE_OK;
156 
157 	HXGE_DEBUG_MSG((hxgep, DMA_CTL,
158 	    "<== hxge_init_rxdma_channel_cntl_stat"));
159 
160 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
161 	rs = hpi_rxdma_control_status(handle, OP_SET, channel, cs_p);
162 
163 	if (rs != HPI_SUCCESS) {
164 		status = HXGE_ERROR | rs;
165 	}
166 	return (status);
167 }
168 
169 
170 hxge_status_t
171 hxge_enable_rxdma_channel(p_hxge_t hxgep, uint16_t channel,
172     p_rx_rbr_ring_t rbr_p, p_rx_rcr_ring_t rcr_p, p_rx_mbox_t mbox_p)
173 {
174 	hpi_handle_t		handle;
175 	rdc_desc_cfg_t 		rdc_desc;
176 	rdc_rcr_cfg_b_t		*cfgb_p;
177 	hpi_status_t		rs = HPI_SUCCESS;
178 
179 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "==> hxge_enable_rxdma_channel"));
180 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
181 
182 	/*
183 	 * Use configuration data composed at init time. Write to hardware the
184 	 * receive ring configurations.
185 	 */
186 	rdc_desc.mbox_enable = 1;
187 	rdc_desc.mbox_addr = mbox_p->mbox_addr;
188 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
189 	    "==> hxge_enable_rxdma_channel: mboxp $%p($%p)",
190 	    mbox_p->mbox_addr, rdc_desc.mbox_addr));
191 
192 	rdc_desc.rbr_len = rbr_p->rbb_max;
193 	rdc_desc.rbr_addr = rbr_p->rbr_addr;
194 
195 	switch (hxgep->rx_bksize_code) {
196 	case RBR_BKSIZE_4K:
197 		rdc_desc.page_size = SIZE_4KB;
198 		break;
199 	case RBR_BKSIZE_8K:
200 		rdc_desc.page_size = SIZE_8KB;
201 		break;
202 	}
203 
204 	rdc_desc.size0 = rbr_p->hpi_pkt_buf_size0;
205 	rdc_desc.valid0 = 1;
206 
207 	rdc_desc.size1 = rbr_p->hpi_pkt_buf_size1;
208 	rdc_desc.valid1 = 1;
209 
210 	rdc_desc.size2 = rbr_p->hpi_pkt_buf_size2;
211 	rdc_desc.valid2 = 1;
212 
213 	rdc_desc.full_hdr = rcr_p->full_hdr_flag;
214 	rdc_desc.offset = rcr_p->sw_priv_hdr_len;
215 
216 	rdc_desc.rcr_len = rcr_p->comp_size;
217 	rdc_desc.rcr_addr = rcr_p->rcr_addr;
218 
219 	cfgb_p = &(rcr_p->rcr_cfgb);
220 	rdc_desc.rcr_threshold = cfgb_p->bits.pthres;
221 	rdc_desc.rcr_timeout = cfgb_p->bits.timeout;
222 	rdc_desc.rcr_timeout_enable = cfgb_p->bits.entout;
223 
224 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "==> hxge_enable_rxdma_channel: "
225 	    "rbr_len qlen %d pagesize code %d rcr_len %d",
226 	    rdc_desc.rbr_len, rdc_desc.page_size, rdc_desc.rcr_len));
227 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "==> hxge_enable_rxdma_channel: "
228 	    "size 0 %d size 1 %d size 2 %d",
229 	    rbr_p->hpi_pkt_buf_size0, rbr_p->hpi_pkt_buf_size1,
230 	    rbr_p->hpi_pkt_buf_size2));
231 
232 	rs = hpi_rxdma_cfg_rdc_ring(handle, rbr_p->rdc, &rdc_desc);
233 	if (rs != HPI_SUCCESS) {
234 		return (HXGE_ERROR | rs);
235 	}
236 
237 	/*
238 	 * Enable the timeout and threshold.
239 	 */
240 	rs = hpi_rxdma_cfg_rdc_rcr_threshold(handle, channel,
241 	    rdc_desc.rcr_threshold);
242 	if (rs != HPI_SUCCESS) {
243 		return (HXGE_ERROR | rs);
244 	}
245 
246 	rs = hpi_rxdma_cfg_rdc_rcr_timeout(handle, channel,
247 	    rdc_desc.rcr_timeout);
248 	if (rs != HPI_SUCCESS) {
249 		return (HXGE_ERROR | rs);
250 	}
251 
252 	/* Enable the DMA */
253 	rs = hpi_rxdma_cfg_rdc_enable(handle, channel);
254 	if (rs != HPI_SUCCESS) {
255 		return (HXGE_ERROR | rs);
256 	}
257 
258 	/* Kick the DMA engine. */
259 	hpi_rxdma_rdc_rbr_kick(handle, channel, rbr_p->rbb_max);
260 	/* Clear the rbr empty bit */
261 	(void) hpi_rxdma_channel_rbr_empty_clear(handle, channel);
262 
263 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "<== hxge_enable_rxdma_channel"));
264 
265 	return (HXGE_OK);
266 }
267 
268 static hxge_status_t
269 hxge_disable_rxdma_channel(p_hxge_t hxgep, uint16_t channel)
270 {
271 	hpi_handle_t handle;
272 	hpi_status_t rs = HPI_SUCCESS;
273 
274 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "==> hxge_disable_rxdma_channel"));
275 
276 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
277 
278 	/* disable the DMA */
279 	rs = hpi_rxdma_cfg_rdc_disable(handle, channel);
280 	if (rs != HPI_SUCCESS) {
281 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
282 		    "<== hxge_disable_rxdma_channel:failed (0x%x)", rs));
283 		return (HXGE_ERROR | rs);
284 	}
285 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "<== hxge_disable_rxdma_channel"));
286 	return (HXGE_OK);
287 }
288 
289 hxge_status_t
290 hxge_rxdma_channel_rcrflush(p_hxge_t hxgep, uint8_t channel)
291 {
292 	hpi_handle_t	handle;
293 	hxge_status_t	status = HXGE_OK;
294 
295 	HXGE_DEBUG_MSG((hxgep, DMA_CTL,
296 	    "==> hxge_rxdma_channel_rcrflush"));
297 
298 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
299 	hpi_rxdma_rdc_rcr_flush(handle, channel);
300 
301 	HXGE_DEBUG_MSG((hxgep, DMA_CTL,
302 	    "<== hxge_rxdma_channel_rcrflush"));
303 	return (status);
304 
305 }
306 
307 #define	MID_INDEX(l, r) ((r + l + 1) >> 1)
308 
309 #define	TO_LEFT -1
310 #define	TO_RIGHT 1
311 #define	BOTH_RIGHT (TO_RIGHT + TO_RIGHT)
312 #define	BOTH_LEFT (TO_LEFT + TO_LEFT)
313 #define	IN_MIDDLE (TO_RIGHT + TO_LEFT)
314 #define	NO_HINT 0xffffffff
315 
316 /*ARGSUSED*/
317 hxge_status_t
318 hxge_rxbuf_pp_to_vp(p_hxge_t hxgep, p_rx_rbr_ring_t rbr_p,
319     uint8_t pktbufsz_type, uint64_t *pkt_buf_addr_pp,
320     uint64_t **pkt_buf_addr_p, uint32_t *bufoffset, uint32_t *msg_index)
321 {
322 	int			bufsize;
323 	uint64_t		pktbuf_pp;
324 	uint64_t		dvma_addr;
325 	rxring_info_t		*ring_info;
326 	int			base_side, end_side;
327 	int			r_index, l_index, anchor_index;
328 	int			found, search_done;
329 	uint32_t		offset, chunk_size, block_size, page_size_mask;
330 	uint32_t		chunk_index, block_index, total_index;
331 	int			max_iterations, iteration;
332 	rxbuf_index_info_t	*bufinfo;
333 
334 	HXGE_DEBUG_MSG((hxgep, RX2_CTL, "==> hxge_rxbuf_pp_to_vp"));
335 
336 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
337 	    "==> hxge_rxbuf_pp_to_vp: buf_pp $%p btype %d",
338 	    pkt_buf_addr_pp, pktbufsz_type));
339 
340 #if defined(__i386)
341 	pktbuf_pp = (uint64_t)(uint32_t)pkt_buf_addr_pp;
342 #else
343 	pktbuf_pp = (uint64_t)pkt_buf_addr_pp;
344 #endif
345 
346 	switch (pktbufsz_type) {
347 	case 0:
348 		bufsize = rbr_p->pkt_buf_size0;
349 		break;
350 	case 1:
351 		bufsize = rbr_p->pkt_buf_size1;
352 		break;
353 	case 2:
354 		bufsize = rbr_p->pkt_buf_size2;
355 		break;
356 	case RCR_SINGLE_BLOCK:
357 		bufsize = 0;
358 		anchor_index = 0;
359 		break;
360 	default:
361 		return (HXGE_ERROR);
362 	}
363 
364 	if (rbr_p->num_blocks == 1) {
365 		anchor_index = 0;
366 		ring_info = rbr_p->ring_info;
367 		bufinfo = (rxbuf_index_info_t *)ring_info->buffer;
368 
369 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
370 		    "==> hxge_rxbuf_pp_to_vp: (found, 1 block) "
371 		    "buf_pp $%p btype %d anchor_index %d bufinfo $%p",
372 		    pkt_buf_addr_pp, pktbufsz_type, anchor_index, bufinfo));
373 
374 		goto found_index;
375 	}
376 
377 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
378 	    "==> hxge_rxbuf_pp_to_vp: buf_pp $%p btype %d anchor_index %d",
379 	    pkt_buf_addr_pp, pktbufsz_type, anchor_index));
380 
381 	ring_info = rbr_p->ring_info;
382 	found = B_FALSE;
383 	bufinfo = (rxbuf_index_info_t *)ring_info->buffer;
384 	iteration = 0;
385 	max_iterations = ring_info->max_iterations;
386 
387 	/*
388 	 * First check if this block have been seen recently. This is indicated
389 	 * by a hint which is initialized when the first buffer of the block is
390 	 * seen. The hint is reset when the last buffer of the block has been
391 	 * processed. As three block sizes are supported, three hints are kept.
392 	 * The idea behind the hints is that once the hardware  uses a block
393 	 * for a buffer  of that size, it will use it exclusively for that size
394 	 * and will use it until it is exhausted. It is assumed that there
395 	 * would a single block being used for the same buffer sizes at any
396 	 * given time.
397 	 */
398 	if (ring_info->hint[pktbufsz_type] != NO_HINT) {
399 		anchor_index = ring_info->hint[pktbufsz_type];
400 		dvma_addr = bufinfo[anchor_index].dvma_addr;
401 		chunk_size = bufinfo[anchor_index].buf_size;
402 		if ((pktbuf_pp >= dvma_addr) &&
403 		    (pktbuf_pp < (dvma_addr + chunk_size))) {
404 			found = B_TRUE;
405 			/*
406 			 * check if this is the last buffer in the block If so,
407 			 * then reset the hint for the size;
408 			 */
409 
410 			if ((pktbuf_pp + bufsize) >= (dvma_addr + chunk_size))
411 				ring_info->hint[pktbufsz_type] = NO_HINT;
412 		}
413 	}
414 
415 	if (found == B_FALSE) {
416 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
417 		    "==> hxge_rxbuf_pp_to_vp: (!found)"
418 		    "buf_pp $%p btype %d anchor_index %d",
419 		    pkt_buf_addr_pp, pktbufsz_type, anchor_index));
420 
421 		/*
422 		 * This is the first buffer of the block of this size. Need to
423 		 * search the whole information array. the search algorithm
424 		 * uses a binary tree search algorithm. It assumes that the
425 		 * information is already sorted with increasing order info[0]
426 		 * < info[1] < info[2]  .... < info[n-1] where n is the size of
427 		 * the information array
428 		 */
429 		r_index = rbr_p->num_blocks - 1;
430 		l_index = 0;
431 		search_done = B_FALSE;
432 		anchor_index = MID_INDEX(r_index, l_index);
433 		while (search_done == B_FALSE) {
434 			if ((r_index == l_index) ||
435 			    (iteration >= max_iterations))
436 				search_done = B_TRUE;
437 
438 			end_side = TO_RIGHT;	/* to the right */
439 			base_side = TO_LEFT;	/* to the left */
440 			/* read the DVMA address information and sort it */
441 			dvma_addr = bufinfo[anchor_index].dvma_addr;
442 			chunk_size = bufinfo[anchor_index].buf_size;
443 
444 			HXGE_DEBUG_MSG((hxgep, RX2_CTL,
445 			    "==> hxge_rxbuf_pp_to_vp: (searching)"
446 			    "buf_pp $%p btype %d "
447 			    "anchor_index %d chunk_size %d dvmaaddr $%p",
448 			    pkt_buf_addr_pp, pktbufsz_type, anchor_index,
449 			    chunk_size, dvma_addr));
450 
451 			if (pktbuf_pp >= dvma_addr)
452 				base_side = TO_RIGHT;	/* to the right */
453 			if (pktbuf_pp < (dvma_addr + chunk_size))
454 				end_side = TO_LEFT;	/* to the left */
455 
456 			switch (base_side + end_side) {
457 			case IN_MIDDLE:
458 				/* found */
459 				found = B_TRUE;
460 				search_done = B_TRUE;
461 				if ((pktbuf_pp + bufsize) <
462 				    (dvma_addr + chunk_size))
463 					ring_info->hint[pktbufsz_type] =
464 					    bufinfo[anchor_index].buf_index;
465 				break;
466 			case BOTH_RIGHT:
467 				/* not found: go to the right */
468 				l_index = anchor_index + 1;
469 				anchor_index = MID_INDEX(r_index, l_index);
470 				break;
471 
472 			case BOTH_LEFT:
473 				/* not found: go to the left */
474 				r_index = anchor_index - 1;
475 				anchor_index = MID_INDEX(r_index, l_index);
476 				break;
477 			default:	/* should not come here */
478 				return (HXGE_ERROR);
479 			}
480 			iteration++;
481 		}
482 
483 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
484 		    "==> hxge_rxbuf_pp_to_vp: (search done)"
485 		    "buf_pp $%p btype %d anchor_index %d",
486 		    pkt_buf_addr_pp, pktbufsz_type, anchor_index));
487 	}
488 
489 	if (found == B_FALSE) {
490 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
491 		    "==> hxge_rxbuf_pp_to_vp: (search failed)"
492 		    "buf_pp $%p btype %d anchor_index %d",
493 		    pkt_buf_addr_pp, pktbufsz_type, anchor_index));
494 		return (HXGE_ERROR);
495 	}
496 
497 found_index:
498 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
499 	    "==> hxge_rxbuf_pp_to_vp: (FOUND1)"
500 	    "buf_pp $%p btype %d bufsize %d anchor_index %d",
501 	    pkt_buf_addr_pp, pktbufsz_type, bufsize, anchor_index));
502 
503 	/* index of the first block in this chunk */
504 	chunk_index = bufinfo[anchor_index].start_index;
505 	dvma_addr = bufinfo[anchor_index].dvma_addr;
506 	page_size_mask = ring_info->block_size_mask;
507 
508 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
509 	    "==> hxge_rxbuf_pp_to_vp: (FOUND3), get chunk)"
510 	    "buf_pp $%p btype %d bufsize %d "
511 	    "anchor_index %d chunk_index %d dvma $%p",
512 	    pkt_buf_addr_pp, pktbufsz_type, bufsize,
513 	    anchor_index, chunk_index, dvma_addr));
514 
515 	offset = pktbuf_pp - dvma_addr;	/* offset within the chunk */
516 	block_size = rbr_p->block_size;	/* System  block(page) size */
517 
518 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
519 	    "==> hxge_rxbuf_pp_to_vp: (FOUND4), get chunk)"
520 	    "buf_pp $%p btype %d bufsize %d "
521 	    "anchor_index %d chunk_index %d dvma $%p "
522 	    "offset %d block_size %d",
523 	    pkt_buf_addr_pp, pktbufsz_type, bufsize, anchor_index,
524 	    chunk_index, dvma_addr, offset, block_size));
525 	HXGE_DEBUG_MSG((hxgep, RX2_CTL, "==> getting total index"));
526 
527 	block_index = (offset / block_size);	/* index within chunk */
528 	total_index = chunk_index + block_index;
529 
530 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
531 	    "==> hxge_rxbuf_pp_to_vp: "
532 	    "total_index %d dvma_addr $%p "
533 	    "offset %d block_size %d "
534 	    "block_index %d ",
535 	    total_index, dvma_addr, offset, block_size, block_index));
536 
537 #if defined(__i386)
538 	*pkt_buf_addr_p = (uint64_t *)((uint32_t)bufinfo[anchor_index].kaddr +
539 	    (uint32_t)offset);
540 #else
541 	*pkt_buf_addr_p = (uint64_t *)((uint64_t)bufinfo[anchor_index].kaddr +
542 	    offset);
543 #endif
544 
545 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
546 	    "==> hxge_rxbuf_pp_to_vp: "
547 	    "total_index %d dvma_addr $%p "
548 	    "offset %d block_size %d "
549 	    "block_index %d "
550 	    "*pkt_buf_addr_p $%p",
551 	    total_index, dvma_addr, offset, block_size,
552 	    block_index, *pkt_buf_addr_p));
553 
554 	*msg_index = total_index;
555 	*bufoffset = (offset & page_size_mask);
556 
557 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
558 	    "==> hxge_rxbuf_pp_to_vp: get msg index: "
559 	    "msg_index %d bufoffset_index %d",
560 	    *msg_index, *bufoffset));
561 	HXGE_DEBUG_MSG((hxgep, RX2_CTL, "<== hxge_rxbuf_pp_to_vp"));
562 
563 	return (HXGE_OK);
564 }
565 
566 
567 /*
568  * used by quick sort (qsort) function
569  * to perform comparison
570  */
571 static int
572 hxge_sort_compare(const void *p1, const void *p2)
573 {
574 
575 	rxbuf_index_info_t *a, *b;
576 
577 	a = (rxbuf_index_info_t *)p1;
578 	b = (rxbuf_index_info_t *)p2;
579 
580 	if (a->dvma_addr > b->dvma_addr)
581 		return (1);
582 	if (a->dvma_addr < b->dvma_addr)
583 		return (-1);
584 	return (0);
585 }
586 
587 /*
588  * Grabbed this sort implementation from common/syscall/avl.c
589  *
590  * Generic shellsort, from K&R (1st ed, p 58.), somewhat modified.
591  * v = Ptr to array/vector of objs
592  * n = # objs in the array
593  * s = size of each obj (must be multiples of a word size)
594  * f = ptr to function to compare two objs
595  *	returns (-1 = less than, 0 = equal, 1 = greater than
596  */
597 void
598 hxge_ksort(caddr_t v, int n, int s, int (*f) ())
599 {
600 	int		g, i, j, ii;
601 	unsigned int	*p1, *p2;
602 	unsigned int	tmp;
603 
604 	/* No work to do */
605 	if (v == NULL || n <= 1)
606 		return;
607 	/* Sanity check on arguments */
608 	ASSERT(((uintptr_t)v & 0x3) == 0 && (s & 0x3) == 0);
609 	ASSERT(s > 0);
610 
611 	for (g = n / 2; g > 0; g /= 2) {
612 		for (i = g; i < n; i++) {
613 			for (j = i - g; j >= 0 &&
614 			    (*f) (v + j * s, v + (j + g) * s) == 1; j -= g) {
615 				p1 = (unsigned *)(v + j * s);
616 				p2 = (unsigned *)(v + (j + g) * s);
617 				for (ii = 0; ii < s / 4; ii++) {
618 					tmp = *p1;
619 					*p1++ = *p2;
620 					*p2++ = tmp;
621 				}
622 			}
623 		}
624 	}
625 }
626 
627 /*
628  * Initialize data structures required for rxdma
629  * buffer dvma->vmem address lookup
630  */
631 /*ARGSUSED*/
632 static hxge_status_t
633 hxge_rxbuf_index_info_init(p_hxge_t hxgep, p_rx_rbr_ring_t rbrp)
634 {
635 	int		index;
636 	rxring_info_t	*ring_info;
637 	int		max_iteration = 0, max_index = 0;
638 
639 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "==> hxge_rxbuf_index_info_init"));
640 
641 	ring_info = rbrp->ring_info;
642 	ring_info->hint[0] = NO_HINT;
643 	ring_info->hint[1] = NO_HINT;
644 	ring_info->hint[2] = NO_HINT;
645 	max_index = rbrp->num_blocks;
646 
647 	/* read the DVMA address information and sort it */
648 	/* do init of the information array */
649 
650 	HXGE_DEBUG_MSG((hxgep, DMA2_CTL,
651 	    " hxge_rxbuf_index_info_init Sort ptrs"));
652 
653 	/* sort the array */
654 	hxge_ksort((void *) ring_info->buffer, max_index,
655 	    sizeof (rxbuf_index_info_t), hxge_sort_compare);
656 
657 	for (index = 0; index < max_index; index++) {
658 		HXGE_DEBUG_MSG((hxgep, DMA2_CTL,
659 		    " hxge_rxbuf_index_info_init: sorted chunk %d "
660 		    " ioaddr $%p kaddr $%p size %x",
661 		    index, ring_info->buffer[index].dvma_addr,
662 		    ring_info->buffer[index].kaddr,
663 		    ring_info->buffer[index].buf_size));
664 	}
665 
666 	max_iteration = 0;
667 	while (max_index >= (1ULL << max_iteration))
668 		max_iteration++;
669 	ring_info->max_iterations = max_iteration + 1;
670 
671 	HXGE_DEBUG_MSG((hxgep, DMA2_CTL,
672 	    " hxge_rxbuf_index_info_init Find max iter %d",
673 	    ring_info->max_iterations));
674 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "<== hxge_rxbuf_index_info_init"));
675 
676 	return (HXGE_OK);
677 }
678 
679 /*ARGSUSED*/
680 void
681 hxge_dump_rcr_entry(p_hxge_t hxgep, p_rcr_entry_t entry_p)
682 {
683 #ifdef	HXGE_DEBUG
684 
685 	uint32_t bptr;
686 	uint64_t pp;
687 
688 	bptr = entry_p->bits.pkt_buf_addr;
689 
690 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
691 	    "\trcr entry $%p "
692 	    "\trcr entry 0x%0llx "
693 	    "\trcr entry 0x%08x "
694 	    "\trcr entry 0x%08x "
695 	    "\tvalue 0x%0llx\n"
696 	    "\tmulti = %d\n"
697 	    "\tpkt_type = 0x%x\n"
698 	    "\terror = 0x%04x\n"
699 	    "\tl2_len = %d\n"
700 	    "\tpktbufsize = %d\n"
701 	    "\tpkt_buf_addr = $%p\n"
702 	    "\tpkt_buf_addr (<< 6) = $%p\n",
703 	    entry_p,
704 	    *(int64_t *)entry_p,
705 	    *(int32_t *)entry_p,
706 	    *(int32_t *)((char *)entry_p + 32),
707 	    entry_p->value,
708 	    entry_p->bits.multi,
709 	    entry_p->bits.pkt_type,
710 	    entry_p->bits.error,
711 	    entry_p->bits.l2_len,
712 	    entry_p->bits.pktbufsz,
713 	    bptr,
714 	    entry_p->bits.pkt_buf_addr_l));
715 
716 	pp = (entry_p->value & RCR_PKT_BUF_ADDR_MASK) <<
717 	    RCR_PKT_BUF_ADDR_SHIFT;
718 
719 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "rcr pp 0x%llx l2 len %d",
720 	    pp, (*(int64_t *)entry_p >> 40) & 0x3fff));
721 #endif
722 }
723 
724 /*ARGSUSED*/
725 void
726 hxge_rxdma_stop(p_hxge_t hxgep)
727 {
728 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rxdma_stop"));
729 
730 	(void) hxge_rx_vmac_disable(hxgep);
731 	(void) hxge_rxdma_hw_mode(hxgep, HXGE_DMA_STOP);
732 
733 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_rxdma_stop"));
734 }
735 
736 void
737 hxge_rxdma_stop_reinit(p_hxge_t hxgep)
738 {
739 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rxdma_stop_reinit"));
740 
741 	(void) hxge_rxdma_stop(hxgep);
742 	(void) hxge_uninit_rxdma_channels(hxgep);
743 	(void) hxge_init_rxdma_channels(hxgep);
744 
745 	(void) hxge_rx_vmac_enable(hxgep);
746 
747 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_rxdma_stop_reinit"));
748 }
749 
750 hxge_status_t
751 hxge_rxdma_hw_mode(p_hxge_t hxgep, boolean_t enable)
752 {
753 	int			i, ndmas;
754 	uint16_t		channel;
755 	p_rx_rbr_rings_t	rx_rbr_rings;
756 	p_rx_rbr_ring_t		*rbr_rings;
757 	hpi_handle_t		handle;
758 	hpi_status_t		rs = HPI_SUCCESS;
759 	hxge_status_t		status = HXGE_OK;
760 
761 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
762 	    "==> hxge_rxdma_hw_mode: mode %d", enable));
763 
764 	if (!(hxgep->drv_state & STATE_HW_INITIALIZED)) {
765 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
766 		    "<== hxge_rxdma_mode: not initialized"));
767 		return (HXGE_ERROR);
768 	}
769 
770 	rx_rbr_rings = hxgep->rx_rbr_rings;
771 	if (rx_rbr_rings == NULL) {
772 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
773 		    "<== hxge_rxdma_mode: NULL ring pointer"));
774 		return (HXGE_ERROR);
775 	}
776 
777 	if (rx_rbr_rings->rbr_rings == NULL) {
778 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
779 		    "<== hxge_rxdma_mode: NULL rbr rings pointer"));
780 		return (HXGE_ERROR);
781 	}
782 
783 	ndmas = rx_rbr_rings->ndmas;
784 	if (!ndmas) {
785 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
786 		    "<== hxge_rxdma_mode: no channel"));
787 		return (HXGE_ERROR);
788 	}
789 
790 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
791 	    "==> hxge_rxdma_mode (ndmas %d)", ndmas));
792 
793 	rbr_rings = rx_rbr_rings->rbr_rings;
794 
795 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
796 
797 	for (i = 0; i < ndmas; i++) {
798 		if (rbr_rings == NULL || rbr_rings[i] == NULL) {
799 			continue;
800 		}
801 		channel = rbr_rings[i]->rdc;
802 		if (enable) {
803 			HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
804 			    "==> hxge_rxdma_hw_mode: channel %d (enable)",
805 			    channel));
806 			rs = hpi_rxdma_cfg_rdc_enable(handle, channel);
807 		} else {
808 			HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
809 			    "==> hxge_rxdma_hw_mode: channel %d (disable)",
810 			    channel));
811 			rs = hpi_rxdma_cfg_rdc_disable(handle, channel);
812 		}
813 	}
814 
815 	status = ((rs == HPI_SUCCESS) ? HXGE_OK : HXGE_ERROR | rs);
816 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
817 	    "<== hxge_rxdma_hw_mode: status 0x%x", status));
818 
819 	return (status);
820 }
821 
822 int
823 hxge_rxdma_get_ring_index(p_hxge_t hxgep, uint16_t channel)
824 {
825 	int			i, ndmas;
826 	uint16_t		rdc;
827 	p_rx_rbr_rings_t 	rx_rbr_rings;
828 	p_rx_rbr_ring_t		*rbr_rings;
829 
830 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
831 	    "==> hxge_rxdma_get_ring_index: channel %d", channel));
832 
833 	rx_rbr_rings = hxgep->rx_rbr_rings;
834 	if (rx_rbr_rings == NULL) {
835 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
836 		    "<== hxge_rxdma_get_ring_index: NULL ring pointer"));
837 		return (-1);
838 	}
839 
840 	ndmas = rx_rbr_rings->ndmas;
841 	if (!ndmas) {
842 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
843 		    "<== hxge_rxdma_get_ring_index: no channel"));
844 		return (-1);
845 	}
846 
847 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
848 	    "==> hxge_rxdma_get_ring_index (ndmas %d)", ndmas));
849 
850 	rbr_rings = rx_rbr_rings->rbr_rings;
851 	for (i = 0; i < ndmas; i++) {
852 		rdc = rbr_rings[i]->rdc;
853 		if (channel == rdc) {
854 			HXGE_DEBUG_MSG((hxgep, RX_CTL,
855 			    "==> hxge_rxdma_get_rbr_ring: "
856 			    "channel %d (index %d) "
857 			    "ring %d", channel, i, rbr_rings[i]));
858 
859 			return (i);
860 		}
861 	}
862 
863 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
864 	    "<== hxge_rxdma_get_rbr_ring_index: not found"));
865 
866 	return (-1);
867 }
868 
869 /*
870  * Static functions start here.
871  */
872 static p_rx_msg_t
873 hxge_allocb(size_t size, uint32_t pri, p_hxge_dma_common_t dmabuf_p)
874 {
875 	p_rx_msg_t		hxge_mp = NULL;
876 	p_hxge_dma_common_t	dmamsg_p;
877 	uchar_t			*buffer;
878 
879 	hxge_mp = KMEM_ZALLOC(sizeof (rx_msg_t), KM_NOSLEEP);
880 	if (hxge_mp == NULL) {
881 		HXGE_ERROR_MSG((NULL, HXGE_ERR_CTL,
882 		    "Allocation of a rx msg failed."));
883 		goto hxge_allocb_exit;
884 	}
885 
886 	hxge_mp->use_buf_pool = B_FALSE;
887 	if (dmabuf_p) {
888 		hxge_mp->use_buf_pool = B_TRUE;
889 
890 		dmamsg_p = (p_hxge_dma_common_t)&hxge_mp->buf_dma;
891 		*dmamsg_p = *dmabuf_p;
892 		dmamsg_p->nblocks = 1;
893 		dmamsg_p->block_size = size;
894 		dmamsg_p->alength = size;
895 		buffer = (uchar_t *)dmabuf_p->kaddrp;
896 
897 		dmabuf_p->kaddrp = (void *)((char *)dmabuf_p->kaddrp + size);
898 		dmabuf_p->ioaddr_pp = (void *)
899 		    ((char *)dmabuf_p->ioaddr_pp + size);
900 
901 		dmabuf_p->alength -= size;
902 		dmabuf_p->offset += size;
903 		dmabuf_p->dma_cookie.dmac_laddress += size;
904 		dmabuf_p->dma_cookie.dmac_size -= size;
905 	} else {
906 		buffer = KMEM_ALLOC(size, KM_NOSLEEP);
907 		if (buffer == NULL) {
908 			HXGE_ERROR_MSG((NULL, HXGE_ERR_CTL,
909 			    "Allocation of a receive page failed."));
910 			goto hxge_allocb_fail1;
911 		}
912 	}
913 
914 	hxge_mp->rx_mblk_p = desballoc(buffer, size, pri, &hxge_mp->freeb);
915 	if (hxge_mp->rx_mblk_p == NULL) {
916 		HXGE_ERROR_MSG((NULL, HXGE_ERR_CTL, "desballoc failed."));
917 		goto hxge_allocb_fail2;
918 	}
919 	hxge_mp->buffer = buffer;
920 	hxge_mp->block_size = size;
921 	hxge_mp->freeb.free_func = (void (*) ()) hxge_freeb;
922 	hxge_mp->freeb.free_arg = (caddr_t)hxge_mp;
923 	hxge_mp->ref_cnt = 1;
924 	hxge_mp->free = B_TRUE;
925 	hxge_mp->rx_use_bcopy = B_FALSE;
926 
927 	atomic_add_32(&hxge_mblks_pending, 1);
928 
929 	goto hxge_allocb_exit;
930 
931 hxge_allocb_fail2:
932 	if (!hxge_mp->use_buf_pool) {
933 		KMEM_FREE(buffer, size);
934 	}
935 hxge_allocb_fail1:
936 	KMEM_FREE(hxge_mp, sizeof (rx_msg_t));
937 	hxge_mp = NULL;
938 
939 hxge_allocb_exit:
940 	return (hxge_mp);
941 }
942 
943 p_mblk_t
944 hxge_dupb(p_rx_msg_t hxge_mp, uint_t offset, size_t size)
945 {
946 	p_mblk_t mp;
947 
948 	HXGE_DEBUG_MSG((NULL, MEM_CTL, "==> hxge_dupb"));
949 	HXGE_DEBUG_MSG((NULL, MEM_CTL, "hxge_mp = $%p "
950 	    "offset = 0x%08X " "size = 0x%08X", hxge_mp, offset, size));
951 
952 	mp = desballoc(&hxge_mp->buffer[offset], size, 0, &hxge_mp->freeb);
953 	if (mp == NULL) {
954 		HXGE_DEBUG_MSG((NULL, RX_CTL, "desballoc failed"));
955 		goto hxge_dupb_exit;
956 	}
957 
958 	atomic_inc_32(&hxge_mp->ref_cnt);
959 	atomic_inc_32(&hxge_mblks_pending);
960 
961 hxge_dupb_exit:
962 	HXGE_DEBUG_MSG((NULL, MEM_CTL, "<== hxge_dupb mp = $%p", hxge_mp));
963 	return (mp);
964 }
965 
966 p_mblk_t
967 hxge_dupb_bcopy(p_rx_msg_t hxge_mp, uint_t offset, size_t size)
968 {
969 	p_mblk_t	mp;
970 	uchar_t		*dp;
971 
972 	mp = allocb(size + HXGE_RXBUF_EXTRA, 0);
973 	if (mp == NULL) {
974 		HXGE_DEBUG_MSG((NULL, RX_CTL, "desballoc failed"));
975 		goto hxge_dupb_bcopy_exit;
976 	}
977 	dp = mp->b_rptr = mp->b_rptr + HXGE_RXBUF_EXTRA;
978 	bcopy((void *) &hxge_mp->buffer[offset], dp, size);
979 	mp->b_wptr = dp + size;
980 
981 hxge_dupb_bcopy_exit:
982 
983 	HXGE_DEBUG_MSG((NULL, MEM_CTL, "<== hxge_dupb mp = $%p", hxge_mp));
984 
985 	return (mp);
986 }
987 
988 void hxge_post_page(p_hxge_t hxgep, p_rx_rbr_ring_t rx_rbr_p,
989     p_rx_msg_t rx_msg_p);
990 
991 void
992 hxge_post_page(p_hxge_t hxgep, p_rx_rbr_ring_t rx_rbr_p, p_rx_msg_t rx_msg_p)
993 {
994 	hpi_handle_t handle;
995 
996 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_post_page"));
997 
998 	/* Reuse this buffer */
999 	rx_msg_p->free = B_FALSE;
1000 	rx_msg_p->cur_usage_cnt = 0;
1001 	rx_msg_p->max_usage_cnt = 0;
1002 	rx_msg_p->pkt_buf_size = 0;
1003 
1004 	if (rx_rbr_p->rbr_use_bcopy) {
1005 		rx_msg_p->rx_use_bcopy = B_FALSE;
1006 		atomic_dec_32(&rx_rbr_p->rbr_consumed);
1007 	}
1008 
1009 	/*
1010 	 * Get the rbr header pointer and its offset index.
1011 	 */
1012 	MUTEX_ENTER(&rx_rbr_p->post_lock);
1013 
1014 	rx_rbr_p->rbr_wr_index = ((rx_rbr_p->rbr_wr_index + 1) &
1015 	    rx_rbr_p->rbr_wrap_mask);
1016 	rx_rbr_p->rbr_desc_vp[rx_rbr_p->rbr_wr_index] = rx_msg_p->shifted_addr;
1017 
1018 	/*
1019 	 * Don't post when index is close to 0 or near the max to reduce the
1020 	 * number rbr_emepty errors
1021 	 */
1022 	rx_rbr_p->pages_to_post++;
1023 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
1024 	if (rx_rbr_p->rbr_wr_index > (rx_rbr_p->pages_to_skip / 2) &&
1025 	    rx_rbr_p->rbr_wr_index < rx_rbr_p->pages_to_post_threshold) {
1026 		hpi_rxdma_rdc_rbr_kick(handle, rx_rbr_p->rdc,
1027 		    rx_rbr_p->pages_to_post);
1028 		rx_rbr_p->pages_to_post = 0;
1029 	}
1030 
1031 	MUTEX_EXIT(&rx_rbr_p->post_lock);
1032 
1033 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
1034 	    "<== hxge_post_page (channel %d post_next_index %d)",
1035 	    rx_rbr_p->rdc, rx_rbr_p->rbr_wr_index));
1036 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_post_page"));
1037 }
1038 
1039 void
1040 hxge_freeb(p_rx_msg_t rx_msg_p)
1041 {
1042 	size_t		size;
1043 	uchar_t		*buffer = NULL;
1044 	int		ref_cnt;
1045 	boolean_t	free_state = B_FALSE;
1046 	rx_rbr_ring_t	*ring = rx_msg_p->rx_rbr_p;
1047 
1048 	HXGE_DEBUG_MSG((NULL, MEM2_CTL, "==> hxge_freeb"));
1049 	HXGE_DEBUG_MSG((NULL, MEM2_CTL,
1050 	    "hxge_freeb:rx_msg_p = $%p (block pending %d)",
1051 	    rx_msg_p, hxge_mblks_pending));
1052 
1053 	atomic_dec_32(&hxge_mblks_pending);
1054 
1055 	/*
1056 	 * First we need to get the free state, then
1057 	 * atomic decrement the reference count to prevent
1058 	 * the race condition with the interrupt thread that
1059 	 * is processing a loaned up buffer block.
1060 	 */
1061 	free_state = rx_msg_p->free;
1062 
1063 	ref_cnt = atomic_add_32_nv(&rx_msg_p->ref_cnt, -1);
1064 	if (!ref_cnt) {
1065 		buffer = rx_msg_p->buffer;
1066 		size = rx_msg_p->block_size;
1067 
1068 		HXGE_DEBUG_MSG((NULL, MEM2_CTL, "hxge_freeb: "
1069 		    "will free: rx_msg_p = $%p (block pending %d)",
1070 		    rx_msg_p, hxge_mblks_pending));
1071 
1072 		if (!rx_msg_p->use_buf_pool) {
1073 			KMEM_FREE(buffer, size);
1074 		}
1075 
1076 		KMEM_FREE(rx_msg_p, sizeof (rx_msg_t));
1077 		/* Decrement the receive buffer ring's reference count, too. */
1078 		atomic_dec_32(&ring->rbr_ref_cnt);
1079 
1080 		/*
1081 		 * Free the receive buffer ring, iff
1082 		 * 1. all the receive buffers have been freed
1083 		 * 2. and we are in the proper state (that is,
1084 		 *    we are not UNMAPPING).
1085 		 */
1086 		if (ring->rbr_ref_cnt == 0 && ring->rbr_state == RBR_UNMAPPED) {
1087 			KMEM_FREE(ring, sizeof (*ring));
1088 		}
1089 		return;
1090 	}
1091 
1092 	/*
1093 	 * Repost buffer.
1094 	 */
1095 	if (free_state && (ref_cnt == 1)) {
1096 		HXGE_DEBUG_MSG((NULL, RX_CTL,
1097 		    "hxge_freeb: post page $%p:", rx_msg_p));
1098 		if (ring->rbr_state == RBR_POSTING)
1099 			hxge_post_page(rx_msg_p->hxgep, ring, rx_msg_p);
1100 	}
1101 
1102 	HXGE_DEBUG_MSG((NULL, MEM2_CTL, "<== hxge_freeb"));
1103 }
1104 
1105 uint_t
1106 hxge_rx_intr(caddr_t arg1, caddr_t arg2)
1107 {
1108 	p_hxge_ldv_t		ldvp = (p_hxge_ldv_t)arg1;
1109 	p_hxge_t		hxgep = (p_hxge_t)arg2;
1110 	p_hxge_ldg_t		ldgp;
1111 	uint8_t			channel;
1112 	hpi_handle_t		handle;
1113 	rdc_stat_t		cs;
1114 	uint_t			serviced = DDI_INTR_UNCLAIMED;
1115 
1116 	if (ldvp == NULL) {
1117 		HXGE_DEBUG_MSG((NULL, RX_INT_CTL,
1118 		    "<== hxge_rx_intr: arg2 $%p arg1 $%p", hxgep, ldvp));
1119 		return (DDI_INTR_CLAIMED);
1120 	}
1121 
1122 	if (arg2 == NULL || (void *) ldvp->hxgep != arg2) {
1123 		hxgep = ldvp->hxgep;
1124 	}
1125 
1126 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1127 	    "==> hxge_rx_intr: arg2 $%p arg1 $%p", hxgep, ldvp));
1128 
1129 	/*
1130 	 * This interrupt handler is for a specific receive dma channel.
1131 	 */
1132 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
1133 
1134 	/*
1135 	 * Get the control and status for this channel.
1136 	 */
1137 	channel = ldvp->channel;
1138 	ldgp = ldvp->ldgp;
1139 	RXDMA_REG_READ64(handle, RDC_STAT, channel, &cs.value);
1140 
1141 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL, "==> hxge_rx_intr:channel %d "
1142 	    "cs 0x%016llx rcrto 0x%x rcrthres %x",
1143 	    channel, cs.value, cs.bits.rcr_to, cs.bits.rcr_thres));
1144 
1145 	hxge_rx_pkts_vring(hxgep, ldvp->vdma_index, ldvp, cs);
1146 	serviced = DDI_INTR_CLAIMED;
1147 
1148 	/* error events. */
1149 	if (cs.value & RDC_STAT_ERROR) {
1150 		(void) hxge_rx_err_evnts(hxgep, ldvp->vdma_index, ldvp, cs);
1151 	}
1152 
1153 hxge_intr_exit:
1154 	/*
1155 	 * Enable the mailbox update interrupt if we want to use mailbox. We
1156 	 * probably don't need to use mailbox as it only saves us one pio read.
1157 	 * Also write 1 to rcrthres and rcrto to clear these two edge triggered
1158 	 * bits.
1159 	 */
1160 	cs.value &= RDC_STAT_WR1C;
1161 	cs.bits.mex = 1;
1162 	RXDMA_REG_WRITE64(handle, RDC_STAT, channel, cs.value);
1163 
1164 	/*
1165 	 * Rearm this logical group if this is a single device group.
1166 	 */
1167 	if (ldgp->nldvs == 1) {
1168 		ld_intr_mgmt_t mgm;
1169 
1170 		mgm.value = 0;
1171 		mgm.bits.arm = 1;
1172 		mgm.bits.timer = ldgp->ldg_timer;
1173 		HXGE_REG_WR32(handle,
1174 		    LD_INTR_MGMT + LDSV_OFFSET(ldgp->ldg), mgm.value);
1175 	}
1176 
1177 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1178 	    "<== hxge_rx_intr: serviced %d", serviced));
1179 
1180 	return (serviced);
1181 }
1182 
1183 static void
1184 hxge_rx_pkts_vring(p_hxge_t hxgep, uint_t vindex, p_hxge_ldv_t ldvp,
1185     rdc_stat_t cs)
1186 {
1187 	p_mblk_t		mp;
1188 	p_rx_rcr_ring_t		rcrp;
1189 
1190 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL, "==> hxge_rx_pkts_vring"));
1191 	if ((mp = hxge_rx_pkts(hxgep, vindex, ldvp, &rcrp, cs)) == NULL) {
1192 		HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1193 		    "<== hxge_rx_pkts_vring: no mp"));
1194 		return;
1195 	}
1196 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rx_pkts_vring: $%p", mp));
1197 
1198 #ifdef  HXGE_DEBUG
1199 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
1200 	    "==> hxge_rx_pkts_vring:calling mac_rx (NEMO) "
1201 	    "LEN %d mp $%p mp->b_next $%p rcrp $%p "
1202 	    "mac_handle $%p",
1203 	    (mp->b_wptr - mp->b_rptr), mp, mp->b_next,
1204 	    rcrp, rcrp->rcr_mac_handle));
1205 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
1206 	    "==> hxge_rx_pkts_vring: dump packets "
1207 	    "(mp $%p b_rptr $%p b_wptr $%p):\n %s",
1208 	    mp, mp->b_rptr, mp->b_wptr,
1209 	    hxge_dump_packet((char *)mp->b_rptr, 64)));
1210 
1211 	if (mp->b_cont) {
1212 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
1213 		    "==> hxge_rx_pkts_vring: dump b_cont packets "
1214 		    "(mp->b_cont $%p b_rptr $%p b_wptr $%p):\n %s",
1215 		    mp->b_cont, mp->b_cont->b_rptr, mp->b_cont->b_wptr,
1216 		    hxge_dump_packet((char *)mp->b_cont->b_rptr,
1217 		    mp->b_cont->b_wptr - mp->b_cont->b_rptr)));
1218 		}
1219 	if (mp->b_next) {
1220 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
1221 		    "==> hxge_rx_pkts_vring: dump next packets "
1222 		    "(b_rptr $%p): %s",
1223 		    mp->b_next->b_rptr,
1224 		    hxge_dump_packet((char *)mp->b_next->b_rptr, 64)));
1225 	}
1226 #endif
1227 
1228 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
1229 	    "==> hxge_rx_pkts_vring: send packet to stack"));
1230 	mac_rx(hxgep->mach, rcrp->rcr_mac_handle, mp);
1231 
1232 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_rx_pkts_vring"));
1233 }
1234 
1235 /*ARGSUSED*/
1236 mblk_t *
1237 hxge_rx_pkts(p_hxge_t hxgep, uint_t vindex, p_hxge_ldv_t ldvp,
1238     p_rx_rcr_ring_t *rcrp, rdc_stat_t cs)
1239 {
1240 	hpi_handle_t		handle;
1241 	uint8_t			channel;
1242 	p_rx_rcr_rings_t	rx_rcr_rings;
1243 	p_rx_rcr_ring_t		rcr_p;
1244 	uint32_t		comp_rd_index;
1245 	p_rcr_entry_t		rcr_desc_rd_head_p;
1246 	p_rcr_entry_t		rcr_desc_rd_head_pp;
1247 	p_mblk_t		nmp, mp_cont, head_mp, *tail_mp;
1248 	uint16_t		qlen, nrcr_read, npkt_read;
1249 	uint32_t		qlen_hw;
1250 	uint32_t		invalid_rcr_entry;
1251 	boolean_t		multi;
1252 	rdc_rcr_cfg_b_t		rcr_cfg_b;
1253 	p_rx_mbox_t		rx_mboxp;
1254 	p_rxdma_mailbox_t	mboxp;
1255 #if defined(_BIG_ENDIAN)
1256 	hpi_status_t		rs = HPI_SUCCESS;
1257 #endif
1258 
1259 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL, "==> hxge_rx_pkts:vindex %d "
1260 	    "channel %d", vindex, ldvp->channel));
1261 
1262 	if (!(hxgep->drv_state & STATE_HW_INITIALIZED)) {
1263 		return (NULL);
1264 	}
1265 
1266 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
1267 	rx_rcr_rings = hxgep->rx_rcr_rings;
1268 	rcr_p = rx_rcr_rings->rcr_rings[vindex];
1269 	channel = rcr_p->rdc;
1270 	if (channel != ldvp->channel) {
1271 		HXGE_DEBUG_MSG((hxgep, RX_INT_CTL, "==> hxge_rx_pkts:index %d "
1272 		    "channel %d, and rcr channel %d not matched.",
1273 		    vindex, ldvp->channel, channel));
1274 		return (NULL);
1275 	}
1276 
1277 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1278 	    "==> hxge_rx_pkts: START: rcr channel %d "
1279 	    "head_p $%p head_pp $%p  index %d ",
1280 	    channel, rcr_p->rcr_desc_rd_head_p,
1281 	    rcr_p->rcr_desc_rd_head_pp, rcr_p->comp_rd_index));
1282 
1283 	rx_mboxp = hxgep->rx_mbox_areas_p->rxmbox_areas[channel];
1284 	mboxp = (p_rxdma_mailbox_t)rx_mboxp->rx_mbox.kaddrp;
1285 	qlen = mboxp->rcrstat_a.bits.qlen;
1286 
1287 	if (!qlen) {
1288 		HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1289 		    "<== hxge_rx_pkts:rcr channel %d qlen %d (no pkts)",
1290 		    channel, qlen));
1291 		return (NULL);
1292 	}
1293 
1294 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rx_pkts:rcr channel %d "
1295 	    "qlen %d", channel, qlen));
1296 
1297 	comp_rd_index = rcr_p->comp_rd_index;
1298 
1299 	rcr_desc_rd_head_p = rcr_p->rcr_desc_rd_head_p;
1300 	rcr_desc_rd_head_pp = rcr_p->rcr_desc_rd_head_pp;
1301 	nrcr_read = npkt_read = 0;
1302 
1303 	/*
1304 	 * Number of packets queued (The jumbo or multi packet will be counted
1305 	 * as only one paccket and it may take up more than one completion
1306 	 * entry).
1307 	 */
1308 	qlen_hw = (qlen < hxge_max_rx_pkts) ? qlen : hxge_max_rx_pkts;
1309 	head_mp = NULL;
1310 	tail_mp = &head_mp;
1311 	nmp = mp_cont = NULL;
1312 	multi = B_FALSE;
1313 
1314 	while (qlen_hw) {
1315 #ifdef HXGE_DEBUG
1316 		hxge_dump_rcr_entry(hxgep, rcr_desc_rd_head_p);
1317 #endif
1318 		/*
1319 		 * Process one completion ring entry.
1320 		 */
1321 		invalid_rcr_entry = 0;
1322 		hxge_receive_packet(hxgep,
1323 		    rcr_p, rcr_desc_rd_head_p, &multi, &nmp, &mp_cont,
1324 		    &invalid_rcr_entry);
1325 		if (invalid_rcr_entry != 0) {
1326 			HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1327 			    "Channel %d could only read 0x%x packets, "
1328 			    "but 0x%x pending\n", channel, npkt_read, qlen_hw));
1329 			break;
1330 		}
1331 
1332 		/*
1333 		 * message chaining modes (nemo msg chaining)
1334 		 */
1335 		if (nmp) {
1336 			nmp->b_next = NULL;
1337 			if (!multi && !mp_cont) { /* frame fits a partition */
1338 				*tail_mp = nmp;
1339 				tail_mp = &nmp->b_next;
1340 				nmp = NULL;
1341 			} else if (multi && !mp_cont) { /* first segment */
1342 				*tail_mp = nmp;
1343 				tail_mp = &nmp->b_cont;
1344 			} else if (multi && mp_cont) {	/* mid of multi segs */
1345 				*tail_mp = mp_cont;
1346 				tail_mp = &mp_cont->b_cont;
1347 			} else if (!multi && mp_cont) { /* last segment */
1348 				*tail_mp = mp_cont;
1349 				tail_mp = &nmp->b_next;
1350 				nmp = NULL;
1351 			}
1352 		}
1353 
1354 		HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1355 		    "==> hxge_rx_pkts: loop: rcr channel %d "
1356 		    "before updating: multi %d "
1357 		    "nrcr_read %d "
1358 		    "npk read %d "
1359 		    "head_pp $%p  index %d ",
1360 		    channel, multi,
1361 		    nrcr_read, npkt_read, rcr_desc_rd_head_pp, comp_rd_index));
1362 
1363 		if (!multi) {
1364 			qlen_hw--;
1365 			npkt_read++;
1366 		}
1367 
1368 		/*
1369 		 * Update the next read entry.
1370 		 */
1371 		comp_rd_index = NEXT_ENTRY(comp_rd_index,
1372 		    rcr_p->comp_wrap_mask);
1373 
1374 		rcr_desc_rd_head_p = NEXT_ENTRY_PTR(rcr_desc_rd_head_p,
1375 		    rcr_p->rcr_desc_first_p, rcr_p->rcr_desc_last_p);
1376 
1377 		nrcr_read++;
1378 
1379 		HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1380 		    "<== hxge_rx_pkts: (SAM, process one packet) "
1381 		    "nrcr_read %d", nrcr_read));
1382 		HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1383 		    "==> hxge_rx_pkts: loop: rcr channel %d "
1384 		    "multi %d nrcr_read %d npk read %d head_pp $%p  index %d ",
1385 		    channel, multi, nrcr_read, npkt_read, rcr_desc_rd_head_pp,
1386 		    comp_rd_index));
1387 	}
1388 
1389 	rcr_p->rcr_desc_rd_head_pp = rcr_desc_rd_head_pp;
1390 	rcr_p->comp_rd_index = comp_rd_index;
1391 	rcr_p->rcr_desc_rd_head_p = rcr_desc_rd_head_p;
1392 
1393 	/* Adjust the mailbox queue length for a hardware bug workaround */
1394 	mboxp->rcrstat_a.bits.qlen -= npkt_read;
1395 
1396 	if ((hxgep->intr_timeout != rcr_p->intr_timeout) ||
1397 	    (hxgep->intr_threshold != rcr_p->intr_threshold)) {
1398 		rcr_p->intr_timeout = hxgep->intr_timeout;
1399 		rcr_p->intr_threshold = hxgep->intr_threshold;
1400 		rcr_cfg_b.value = 0x0ULL;
1401 		if (rcr_p->intr_timeout)
1402 			rcr_cfg_b.bits.entout = 1;
1403 		rcr_cfg_b.bits.timeout = rcr_p->intr_timeout;
1404 		rcr_cfg_b.bits.pthres = rcr_p->intr_threshold;
1405 		RXDMA_REG_WRITE64(handle, RDC_RCR_CFG_B,
1406 		    channel, rcr_cfg_b.value);
1407 	}
1408 
1409 	cs.bits.pktread = npkt_read;
1410 	cs.bits.ptrread = nrcr_read;
1411 	RXDMA_REG_WRITE64(handle, RDC_STAT, channel, cs.value);
1412 
1413 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1414 	    "==> hxge_rx_pkts: EXIT: rcr channel %d "
1415 	    "head_pp $%p  index %016llx ",
1416 	    channel, rcr_p->rcr_desc_rd_head_pp, rcr_p->comp_rd_index));
1417 
1418 	/*
1419 	 * Update RCR buffer pointer read and number of packets read.
1420 	 */
1421 
1422 	*rcrp = rcr_p;
1423 
1424 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL, "<== hxge_rx_pkts"));
1425 
1426 	return (head_mp);
1427 }
1428 
1429 #define	RCR_ENTRY_PATTERN	0x5a5a6b6b7c7c8d8dULL
1430 
1431 /*ARGSUSED*/
1432 void
1433 hxge_receive_packet(p_hxge_t hxgep,
1434     p_rx_rcr_ring_t rcr_p, p_rcr_entry_t rcr_desc_rd_head_p,
1435     boolean_t *multi_p, mblk_t **mp, mblk_t **mp_cont,
1436     uint32_t *invalid_rcr_entry)
1437 {
1438 	p_mblk_t		nmp = NULL;
1439 	uint64_t		multi;
1440 	uint8_t			channel;
1441 
1442 	boolean_t first_entry = B_TRUE;
1443 	boolean_t is_tcp_udp = B_FALSE;
1444 	boolean_t buffer_free = B_FALSE;
1445 	boolean_t error_send_up = B_FALSE;
1446 	uint8_t error_type;
1447 	uint16_t l2_len;
1448 	uint16_t skip_len;
1449 	uint8_t pktbufsz_type;
1450 	uint64_t rcr_entry;
1451 	uint64_t *pkt_buf_addr_pp;
1452 	uint64_t *pkt_buf_addr_p;
1453 	uint32_t buf_offset;
1454 	uint32_t bsize;
1455 	uint32_t msg_index;
1456 	p_rx_rbr_ring_t rx_rbr_p;
1457 	p_rx_msg_t *rx_msg_ring_p;
1458 	p_rx_msg_t rx_msg_p;
1459 
1460 	uint16_t sw_offset_bytes = 0, hdr_size = 0;
1461 	hxge_status_t status = HXGE_OK;
1462 	boolean_t is_valid = B_FALSE;
1463 	p_hxge_rx_ring_stats_t rdc_stats;
1464 	uint32_t bytes_read;
1465 
1466 	uint64_t pkt_type;
1467 
1468 	channel = rcr_p->rdc;
1469 
1470 	HXGE_DEBUG_MSG((hxgep, RX2_CTL, "==> hxge_receive_packet"));
1471 
1472 	first_entry = (*mp == NULL) ? B_TRUE : B_FALSE;
1473 	rcr_entry = *((uint64_t *)rcr_desc_rd_head_p);
1474 
1475 	/* Verify the content of the rcr_entry for a hardware bug workaround */
1476 	if ((rcr_entry == 0x0) || (rcr_entry == RCR_ENTRY_PATTERN)) {
1477 		*invalid_rcr_entry = 1;
1478 		HXGE_DEBUG_MSG((hxgep, RX2_CTL, "hxge_receive_packet "
1479 		    "Channel %d invalid RCR entry 0x%llx found, returning\n",
1480 		    channel, (long long) rcr_entry));
1481 		return;
1482 	}
1483 	*((uint64_t *)rcr_desc_rd_head_p) = RCR_ENTRY_PATTERN;
1484 
1485 	multi = (rcr_entry & RCR_MULTI_MASK);
1486 	pkt_type = (rcr_entry & RCR_PKT_TYPE_MASK);
1487 
1488 	error_type = ((rcr_entry & RCR_ERROR_MASK) >> RCR_ERROR_SHIFT);
1489 	l2_len = ((rcr_entry & RCR_L2_LEN_MASK) >> RCR_L2_LEN_SHIFT);
1490 
1491 	/*
1492 	 * Hardware does not strip the CRC due bug ID 11451 where
1493 	 * the hardware mis handles minimum size packets.
1494 	 */
1495 	l2_len -= ETHERFCSL;
1496 
1497 	pktbufsz_type = ((rcr_entry & RCR_PKTBUFSZ_MASK) >>
1498 	    RCR_PKTBUFSZ_SHIFT);
1499 #if defined(__i386)
1500 	pkt_buf_addr_pp = (uint64_t *)(uint32_t)((rcr_entry &
1501 	    RCR_PKT_BUF_ADDR_MASK) << RCR_PKT_BUF_ADDR_SHIFT);
1502 #else
1503 	pkt_buf_addr_pp = (uint64_t *)((rcr_entry & RCR_PKT_BUF_ADDR_MASK) <<
1504 	    RCR_PKT_BUF_ADDR_SHIFT);
1505 #endif
1506 
1507 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1508 	    "==> hxge_receive_packet: entryp $%p entry 0x%0llx "
1509 	    "pkt_buf_addr_pp $%p l2_len %d multi %d "
1510 	    "error_type 0x%x pkt_type 0x%x  "
1511 	    "pktbufsz_type %d ",
1512 	    rcr_desc_rd_head_p, rcr_entry, pkt_buf_addr_pp, l2_len,
1513 	    multi, error_type, pkt_type, pktbufsz_type));
1514 
1515 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1516 	    "==> hxge_receive_packet: entryp $%p entry 0x%0llx "
1517 	    "pkt_buf_addr_pp $%p l2_len %d multi %d "
1518 	    "error_type 0x%x pkt_type 0x%x ", rcr_desc_rd_head_p,
1519 	    rcr_entry, pkt_buf_addr_pp, l2_len, multi, error_type, pkt_type));
1520 
1521 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1522 	    "==> (rbr) hxge_receive_packet: entry 0x%0llx "
1523 	    "full pkt_buf_addr_pp $%p l2_len %d",
1524 	    rcr_entry, pkt_buf_addr_pp, l2_len));
1525 
1526 	/* get the stats ptr */
1527 	rdc_stats = rcr_p->rdc_stats;
1528 	if (l2_len > (STD_FRAME_SIZE - ETHERFCSL))
1529 		rdc_stats->jumbo_pkts++;
1530 
1531 	if (!l2_len) {
1532 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
1533 		    "<== hxge_receive_packet: failed: l2 length is 0."));
1534 		return;
1535 	}
1536 
1537 	/* shift 6 bits to get the full io address */
1538 #if defined(__i386)
1539 	pkt_buf_addr_pp = (uint64_t *)((uint32_t)pkt_buf_addr_pp <<
1540 	    RCR_PKT_BUF_ADDR_SHIFT_FULL);
1541 #else
1542 	pkt_buf_addr_pp = (uint64_t *)((uint64_t)pkt_buf_addr_pp <<
1543 	    RCR_PKT_BUF_ADDR_SHIFT_FULL);
1544 #endif
1545 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1546 	    "==> (rbr) hxge_receive_packet: entry 0x%0llx "
1547 	    "full pkt_buf_addr_pp $%p l2_len %d",
1548 	    rcr_entry, pkt_buf_addr_pp, l2_len));
1549 
1550 	rx_rbr_p = rcr_p->rx_rbr_p;
1551 	rx_msg_ring_p = rx_rbr_p->rx_msg_ring;
1552 
1553 	if (first_entry) {
1554 		hdr_size = (rcr_p->full_hdr_flag ? RXDMA_HDR_SIZE_FULL :
1555 		    RXDMA_HDR_SIZE_DEFAULT);
1556 
1557 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
1558 		    "==> hxge_receive_packet: first entry 0x%016llx "
1559 		    "pkt_buf_addr_pp $%p l2_len %d hdr %d",
1560 		    rcr_entry, pkt_buf_addr_pp, l2_len, hdr_size));
1561 	}
1562 
1563 	MUTEX_ENTER(&rcr_p->lock);
1564 	MUTEX_ENTER(&rx_rbr_p->lock);
1565 
1566 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
1567 	    "==> (rbr 1) hxge_receive_packet: entry 0x%0llx "
1568 	    "full pkt_buf_addr_pp $%p l2_len %d",
1569 	    rcr_entry, pkt_buf_addr_pp, l2_len));
1570 
1571 	/*
1572 	 * Packet buffer address in the completion entry points to the starting
1573 	 * buffer address (offset 0). Use the starting buffer address to locate
1574 	 * the corresponding kernel address.
1575 	 */
1576 	status = hxge_rxbuf_pp_to_vp(hxgep, rx_rbr_p,
1577 	    pktbufsz_type, pkt_buf_addr_pp, &pkt_buf_addr_p,
1578 	    &buf_offset, &msg_index);
1579 
1580 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
1581 	    "==> (rbr 2) hxge_receive_packet: entry 0x%0llx "
1582 	    "full pkt_buf_addr_pp $%p l2_len %d",
1583 	    rcr_entry, pkt_buf_addr_pp, l2_len));
1584 
1585 	if (status != HXGE_OK) {
1586 		MUTEX_EXIT(&rx_rbr_p->lock);
1587 		MUTEX_EXIT(&rcr_p->lock);
1588 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
1589 		    "<== hxge_receive_packet: found vaddr failed %d", status));
1590 		return;
1591 	}
1592 
1593 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1594 	    "==> (rbr 3) hxge_receive_packet: entry 0x%0llx "
1595 	    "full pkt_buf_addr_pp $%p l2_len %d",
1596 	    rcr_entry, pkt_buf_addr_pp, l2_len));
1597 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1598 	    "==> (rbr 4 msgindex %d) hxge_receive_packet: entry 0x%0llx "
1599 	    "full pkt_buf_addr_pp $%p l2_len %d",
1600 	    msg_index, rcr_entry, pkt_buf_addr_pp, l2_len));
1601 
1602 	if (msg_index >= rx_rbr_p->tnblocks) {
1603 		MUTEX_EXIT(&rx_rbr_p->lock);
1604 		MUTEX_EXIT(&rcr_p->lock);
1605 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1606 		    "==> hxge_receive_packet: FATAL msg_index (%d) "
1607 		    "should be smaller than tnblocks (%d)\n",
1608 		    msg_index, rx_rbr_p->tnblocks));
1609 		return;
1610 	}
1611 
1612 	rx_msg_p = rx_msg_ring_p[msg_index];
1613 
1614 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1615 	    "==> (rbr 4 msgindex %d) hxge_receive_packet: entry 0x%0llx "
1616 	    "full pkt_buf_addr_pp $%p l2_len %d",
1617 	    msg_index, rcr_entry, pkt_buf_addr_pp, l2_len));
1618 
1619 	switch (pktbufsz_type) {
1620 	case RCR_PKTBUFSZ_0:
1621 		bsize = rx_rbr_p->pkt_buf_size0_bytes;
1622 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1623 		    "==> hxge_receive_packet: 0 buf %d", bsize));
1624 		break;
1625 	case RCR_PKTBUFSZ_1:
1626 		bsize = rx_rbr_p->pkt_buf_size1_bytes;
1627 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1628 		    "==> hxge_receive_packet: 1 buf %d", bsize));
1629 		break;
1630 	case RCR_PKTBUFSZ_2:
1631 		bsize = rx_rbr_p->pkt_buf_size2_bytes;
1632 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
1633 		    "==> hxge_receive_packet: 2 buf %d", bsize));
1634 		break;
1635 	case RCR_SINGLE_BLOCK:
1636 		bsize = rx_msg_p->block_size;
1637 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1638 		    "==> hxge_receive_packet: single %d", bsize));
1639 
1640 		break;
1641 	default:
1642 		MUTEX_EXIT(&rx_rbr_p->lock);
1643 		MUTEX_EXIT(&rcr_p->lock);
1644 		return;
1645 	}
1646 
1647 	DMA_COMMON_SYNC_OFFSET(rx_msg_p->buf_dma,
1648 	    (buf_offset + sw_offset_bytes), (hdr_size + l2_len),
1649 	    DDI_DMA_SYNC_FORCPU);
1650 
1651 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1652 	    "==> hxge_receive_packet: after first dump:usage count"));
1653 
1654 	if (rx_msg_p->cur_usage_cnt == 0) {
1655 		if (rx_rbr_p->rbr_use_bcopy) {
1656 			atomic_inc_32(&rx_rbr_p->rbr_consumed);
1657 			if (rx_rbr_p->rbr_consumed <
1658 			    rx_rbr_p->rbr_threshold_hi) {
1659 				if (rx_rbr_p->rbr_threshold_lo == 0 ||
1660 				    ((rx_rbr_p->rbr_consumed >=
1661 				    rx_rbr_p->rbr_threshold_lo) &&
1662 				    (rx_rbr_p->rbr_bufsize_type >=
1663 				    pktbufsz_type))) {
1664 					rx_msg_p->rx_use_bcopy = B_TRUE;
1665 				}
1666 			} else {
1667 				rx_msg_p->rx_use_bcopy = B_TRUE;
1668 			}
1669 		}
1670 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1671 		    "==> hxge_receive_packet: buf %d (new block) ", bsize));
1672 
1673 		rx_msg_p->pkt_buf_size_code = pktbufsz_type;
1674 		rx_msg_p->pkt_buf_size = bsize;
1675 		rx_msg_p->cur_usage_cnt = 1;
1676 		if (pktbufsz_type == RCR_SINGLE_BLOCK) {
1677 			HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1678 			    "==> hxge_receive_packet: buf %d (single block) ",
1679 			    bsize));
1680 			/*
1681 			 * Buffer can be reused once the free function is
1682 			 * called.
1683 			 */
1684 			rx_msg_p->max_usage_cnt = 1;
1685 			buffer_free = B_TRUE;
1686 		} else {
1687 			rx_msg_p->max_usage_cnt = rx_msg_p->block_size / bsize;
1688 			if (rx_msg_p->max_usage_cnt == 1) {
1689 				buffer_free = B_TRUE;
1690 			}
1691 		}
1692 	} else {
1693 		rx_msg_p->cur_usage_cnt++;
1694 		if (rx_msg_p->cur_usage_cnt == rx_msg_p->max_usage_cnt) {
1695 			buffer_free = B_TRUE;
1696 		}
1697 	}
1698 
1699 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
1700 	    "msgbuf index = %d l2len %d bytes usage %d max_usage %d ",
1701 	    msg_index, l2_len,
1702 	    rx_msg_p->cur_usage_cnt, rx_msg_p->max_usage_cnt));
1703 
1704 	if (error_type) {
1705 		rdc_stats->ierrors++;
1706 		/* Update error stats */
1707 		rdc_stats->errlog.compl_err_type = error_type;
1708 		HXGE_FM_REPORT_ERROR(hxgep, NULL, HXGE_FM_EREPORT_RDMC_RCR_ERR);
1709 
1710 		if (error_type & RCR_CTRL_FIFO_DED) {
1711 			rdc_stats->ctrl_fifo_ecc_err++;
1712 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
1713 			    " hxge_receive_packet: "
1714 			    " channel %d RCR ctrl_fifo_ded error", channel));
1715 		} else if (error_type & RCR_DATA_FIFO_DED) {
1716 			rdc_stats->data_fifo_ecc_err++;
1717 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
1718 			    " hxge_receive_packet: channel %d"
1719 			    " RCR data_fifo_ded error", channel));
1720 		}
1721 
1722 		/*
1723 		 * Update and repost buffer block if max usage count is
1724 		 * reached.
1725 		 */
1726 		if (error_send_up == B_FALSE) {
1727 			atomic_inc_32(&rx_msg_p->ref_cnt);
1728 			atomic_inc_32(&hxge_mblks_pending);
1729 			if (buffer_free == B_TRUE) {
1730 				rx_msg_p->free = B_TRUE;
1731 			}
1732 
1733 			MUTEX_EXIT(&rx_rbr_p->lock);
1734 			MUTEX_EXIT(&rcr_p->lock);
1735 			hxge_freeb(rx_msg_p);
1736 			return;
1737 		}
1738 	}
1739 
1740 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1741 	    "==> hxge_receive_packet: DMA sync second "));
1742 
1743 	bytes_read = rcr_p->rcvd_pkt_bytes;
1744 	skip_len = sw_offset_bytes + hdr_size;
1745 	if (!rx_msg_p->rx_use_bcopy) {
1746 		/*
1747 		 * For loaned up buffers, the driver reference count
1748 		 * will be incremented first and then the free state.
1749 		 */
1750 		if ((nmp = hxge_dupb(rx_msg_p, buf_offset, bsize)) != NULL) {
1751 			if (first_entry) {
1752 				nmp->b_rptr = &nmp->b_rptr[skip_len];
1753 				if (l2_len < bsize - skip_len) {
1754 					nmp->b_wptr = &nmp->b_rptr[l2_len];
1755 				} else {
1756 					nmp->b_wptr = &nmp->b_rptr[bsize
1757 					    - skip_len];
1758 				}
1759 			} else {
1760 				if (l2_len - bytes_read < bsize) {
1761 					nmp->b_wptr =
1762 					    &nmp->b_rptr[l2_len - bytes_read];
1763 				} else {
1764 					nmp->b_wptr = &nmp->b_rptr[bsize];
1765 				}
1766 			}
1767 		}
1768 	} else {
1769 		if (first_entry) {
1770 			nmp = hxge_dupb_bcopy(rx_msg_p, buf_offset + skip_len,
1771 			    l2_len < bsize - skip_len ?
1772 			    l2_len : bsize - skip_len);
1773 		} else {
1774 			nmp = hxge_dupb_bcopy(rx_msg_p, buf_offset,
1775 			    l2_len - bytes_read < bsize ?
1776 			    l2_len - bytes_read : bsize);
1777 		}
1778 	}
1779 
1780 	if (nmp != NULL) {
1781 		if (first_entry)
1782 			bytes_read  = nmp->b_wptr - nmp->b_rptr;
1783 		else
1784 			bytes_read += nmp->b_wptr - nmp->b_rptr;
1785 
1786 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
1787 		    "==> hxge_receive_packet after dupb: "
1788 		    "rbr consumed %d "
1789 		    "pktbufsz_type %d "
1790 		    "nmp $%p rptr $%p wptr $%p "
1791 		    "buf_offset %d bzise %d l2_len %d skip_len %d",
1792 		    rx_rbr_p->rbr_consumed,
1793 		    pktbufsz_type,
1794 		    nmp, nmp->b_rptr, nmp->b_wptr,
1795 		    buf_offset, bsize, l2_len, skip_len));
1796 	} else {
1797 		cmn_err(CE_WARN, "!hxge_receive_packet: update stats (error)");
1798 
1799 		atomic_inc_32(&rx_msg_p->ref_cnt);
1800 		atomic_inc_32(&hxge_mblks_pending);
1801 		if (buffer_free == B_TRUE) {
1802 			rx_msg_p->free = B_TRUE;
1803 		}
1804 
1805 		MUTEX_EXIT(&rx_rbr_p->lock);
1806 		MUTEX_EXIT(&rcr_p->lock);
1807 		hxge_freeb(rx_msg_p);
1808 		return;
1809 	}
1810 
1811 	if (buffer_free == B_TRUE) {
1812 		rx_msg_p->free = B_TRUE;
1813 	}
1814 
1815 	/*
1816 	 * ERROR, FRAG and PKT_TYPE are only reported in the first entry. If a
1817 	 * packet is not fragmented and no error bit is set, then L4 checksum
1818 	 * is OK.
1819 	 */
1820 	is_valid = (nmp != NULL);
1821 	if (first_entry) {
1822 		rdc_stats->ipackets++; /* count only 1st seg for jumbo */
1823 		rdc_stats->ibytes += skip_len + l2_len < bsize ?
1824 		    l2_len : bsize;
1825 	} else {
1826 		rdc_stats->ibytes += l2_len - bytes_read < bsize ?
1827 		    l2_len - bytes_read : bsize;
1828 	}
1829 
1830 	rcr_p->rcvd_pkt_bytes = bytes_read;
1831 
1832 	MUTEX_EXIT(&rx_rbr_p->lock);
1833 	MUTEX_EXIT(&rcr_p->lock);
1834 
1835 	if (rx_msg_p->free && rx_msg_p->rx_use_bcopy) {
1836 		atomic_inc_32(&rx_msg_p->ref_cnt);
1837 		atomic_inc_32(&hxge_mblks_pending);
1838 		hxge_freeb(rx_msg_p);
1839 	}
1840 
1841 	if (is_valid) {
1842 		nmp->b_cont = NULL;
1843 		if (first_entry) {
1844 			*mp = nmp;
1845 			*mp_cont = NULL;
1846 		} else {
1847 			*mp_cont = nmp;
1848 		}
1849 	}
1850 
1851 	/*
1852 	 * Update stats and hardware checksuming.
1853 	 */
1854 	if (is_valid && !multi) {
1855 
1856 		is_tcp_udp = ((pkt_type == RCR_PKT_IS_TCP ||
1857 		    pkt_type == RCR_PKT_IS_UDP) ? B_TRUE : B_FALSE);
1858 
1859 		HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_receive_packet: "
1860 		    "is_valid 0x%x multi %d pkt %d d error %d",
1861 		    is_valid, multi, is_tcp_udp, error_type));
1862 
1863 		if (is_tcp_udp && !error_type) {
1864 			(void) hcksum_assoc(nmp, NULL, NULL, 0, 0, 0, 0,
1865 			    HCK_FULLCKSUM_OK | HCK_FULLCKSUM, 0);
1866 
1867 			HXGE_DEBUG_MSG((hxgep, RX_CTL,
1868 			    "==> hxge_receive_packet: Full tcp/udp cksum "
1869 			    "is_valid 0x%x multi %d pkt %d "
1870 			    "error %d",
1871 			    is_valid, multi, is_tcp_udp, error_type));
1872 		}
1873 	}
1874 
1875 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1876 	    "==> hxge_receive_packet: *mp 0x%016llx", *mp));
1877 
1878 	*multi_p = (multi == RCR_MULTI_MASK);
1879 
1880 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_receive_packet: "
1881 	    "multi %d nmp 0x%016llx *mp 0x%016llx *mp_cont 0x%016llx",
1882 	    *multi_p, nmp, *mp, *mp_cont));
1883 }
1884 
1885 /*ARGSUSED*/
1886 static hxge_status_t
1887 hxge_rx_err_evnts(p_hxge_t hxgep, uint_t index, p_hxge_ldv_t ldvp,
1888     rdc_stat_t cs)
1889 {
1890 	p_hxge_rx_ring_stats_t	rdc_stats;
1891 	hpi_handle_t		handle;
1892 	boolean_t		rxchan_fatal = B_FALSE;
1893 	uint8_t			channel;
1894 	hxge_status_t		status = HXGE_OK;
1895 	uint64_t		cs_val;
1896 
1897 	HXGE_DEBUG_MSG((hxgep, INT_CTL, "==> hxge_rx_err_evnts"));
1898 
1899 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
1900 	channel = ldvp->channel;
1901 
1902 	/* Clear the interrupts */
1903 	cs_val = cs.value & RDC_STAT_WR1C;
1904 	RXDMA_REG_WRITE64(handle, RDC_STAT, channel, cs_val);
1905 
1906 	rdc_stats = &hxgep->statsp->rdc_stats[ldvp->vdma_index];
1907 
1908 	if (cs.bits.rbr_cpl_to) {
1909 		rdc_stats->rbr_tmout++;
1910 		HXGE_FM_REPORT_ERROR(hxgep, channel,
1911 		    HXGE_FM_EREPORT_RDMC_RBR_CPL_TO);
1912 		rxchan_fatal = B_TRUE;
1913 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
1914 		    "==> hxge_rx_err_evnts(channel %d): "
1915 		    "fatal error: rx_rbr_timeout", channel));
1916 	}
1917 
1918 	if ((cs.bits.rcr_shadow_par_err) || (cs.bits.rbr_prefetch_par_err)) {
1919 		(void) hpi_rxdma_ring_perr_stat_get(handle,
1920 		    &rdc_stats->errlog.pre_par, &rdc_stats->errlog.sha_par);
1921 	}
1922 
1923 	if (cs.bits.rcr_shadow_par_err) {
1924 		rdc_stats->rcr_sha_par++;
1925 		HXGE_FM_REPORT_ERROR(hxgep, channel,
1926 		    HXGE_FM_EREPORT_RDMC_RCR_SHA_PAR);
1927 		rxchan_fatal = B_TRUE;
1928 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
1929 		    "==> hxge_rx_err_evnts(channel %d): "
1930 		    "fatal error: rcr_shadow_par_err", channel));
1931 	}
1932 
1933 	if (cs.bits.rbr_prefetch_par_err) {
1934 		rdc_stats->rbr_pre_par++;
1935 		HXGE_FM_REPORT_ERROR(hxgep, channel,
1936 		    HXGE_FM_EREPORT_RDMC_RBR_PRE_PAR);
1937 		rxchan_fatal = B_TRUE;
1938 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
1939 		    "==> hxge_rx_err_evnts(channel %d): "
1940 		    "fatal error: rbr_prefetch_par_err", channel));
1941 	}
1942 
1943 	if (cs.bits.rbr_pre_empty) {
1944 		rdc_stats->rbr_pre_empty++;
1945 		HXGE_FM_REPORT_ERROR(hxgep, channel,
1946 		    HXGE_FM_EREPORT_RDMC_RBR_PRE_EMPTY);
1947 		rxchan_fatal = B_TRUE;
1948 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
1949 		    "==> hxge_rx_err_evnts(channel %d): "
1950 		    "fatal error: rbr_pre_empty", channel));
1951 	}
1952 
1953 	if (cs.bits.peu_resp_err) {
1954 		rdc_stats->peu_resp_err++;
1955 		HXGE_FM_REPORT_ERROR(hxgep, channel,
1956 		    HXGE_FM_EREPORT_RDMC_PEU_RESP_ERR);
1957 		rxchan_fatal = B_TRUE;
1958 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
1959 		    "==> hxge_rx_err_evnts(channel %d): "
1960 		    "fatal error: peu_resp_err", channel));
1961 	}
1962 
1963 	if (cs.bits.rcr_thres) {
1964 		rdc_stats->rcr_thres++;
1965 	}
1966 
1967 	if (cs.bits.rcr_to) {
1968 		rdc_stats->rcr_to++;
1969 	}
1970 
1971 	if (cs.bits.rcr_shadow_full) {
1972 		rdc_stats->rcr_shadow_full++;
1973 		HXGE_FM_REPORT_ERROR(hxgep, channel,
1974 		    HXGE_FM_EREPORT_RDMC_RCR_SHA_FULL);
1975 		rxchan_fatal = B_TRUE;
1976 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
1977 		    "==> hxge_rx_err_evnts(channel %d): "
1978 		    "fatal error: rcr_shadow_full", channel));
1979 	}
1980 
1981 	if (cs.bits.rcr_full) {
1982 		rdc_stats->rcrfull++;
1983 		HXGE_FM_REPORT_ERROR(hxgep, channel,
1984 		    HXGE_FM_EREPORT_RDMC_RCRFULL);
1985 		rxchan_fatal = B_TRUE;
1986 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
1987 		    "==> hxge_rx_err_evnts(channel %d): "
1988 		    "fatal error: rcrfull error", channel));
1989 	}
1990 
1991 	if (cs.bits.rbr_empty) {
1992 		rdc_stats->rbr_empty++;
1993 		if (rdc_stats->rbr_empty == 1)
1994 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
1995 			    "==> hxge_rx_err_evnts(channel %d): "
1996 			    "rbr empty error", channel));
1997 		/*
1998 		 * DMA channel is disabled due to rbr_empty bit is set
1999 		 * although it is not fatal. Enable the DMA channel here
2000 		 * to work-around the hardware bug.
2001 		 */
2002 		(void) hpi_rxdma_cfg_rdc_enable(handle, channel);
2003 	}
2004 
2005 	if (cs.bits.rbr_full) {
2006 		rdc_stats->rbrfull++;
2007 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2008 		    HXGE_FM_EREPORT_RDMC_RBRFULL);
2009 		rxchan_fatal = B_TRUE;
2010 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2011 		    "==> hxge_rx_err_evnts(channel %d): "
2012 		    "fatal error: rbr_full error", channel));
2013 	}
2014 
2015 	if (rxchan_fatal) {
2016 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2017 		    " hxge_rx_err_evnts: fatal error on Channel #%d\n",
2018 		    channel));
2019 		status = hxge_rxdma_fatal_err_recover(hxgep, channel);
2020 		if (status == HXGE_OK) {
2021 			FM_SERVICE_RESTORED(hxgep);
2022 		}
2023 	}
2024 	HXGE_DEBUG_MSG((hxgep, RX2_CTL, "<== hxge_rx_err_evnts"));
2025 
2026 	return (status);
2027 }
2028 
2029 static hxge_status_t
2030 hxge_map_rxdma(p_hxge_t hxgep)
2031 {
2032 	int			i, ndmas;
2033 	uint16_t		channel;
2034 	p_rx_rbr_rings_t	rx_rbr_rings;
2035 	p_rx_rbr_ring_t		*rbr_rings;
2036 	p_rx_rcr_rings_t	rx_rcr_rings;
2037 	p_rx_rcr_ring_t		*rcr_rings;
2038 	p_rx_mbox_areas_t	rx_mbox_areas_p;
2039 	p_rx_mbox_t		*rx_mbox_p;
2040 	p_hxge_dma_pool_t	dma_buf_poolp;
2041 	p_hxge_dma_pool_t	dma_cntl_poolp;
2042 	p_hxge_dma_common_t	*dma_buf_p;
2043 	p_hxge_dma_common_t	*dma_cntl_p;
2044 	uint32_t		*num_chunks;
2045 	hxge_status_t		status = HXGE_OK;
2046 
2047 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_map_rxdma"));
2048 
2049 	dma_buf_poolp = hxgep->rx_buf_pool_p;
2050 	dma_cntl_poolp = hxgep->rx_cntl_pool_p;
2051 
2052 	if (!dma_buf_poolp->buf_allocated || !dma_cntl_poolp->buf_allocated) {
2053 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2054 		    "<== hxge_map_rxdma: buf not allocated"));
2055 		return (HXGE_ERROR);
2056 	}
2057 
2058 	ndmas = dma_buf_poolp->ndmas;
2059 	if (!ndmas) {
2060 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
2061 		    "<== hxge_map_rxdma: no dma allocated"));
2062 		return (HXGE_ERROR);
2063 	}
2064 
2065 	num_chunks = dma_buf_poolp->num_chunks;
2066 	dma_buf_p = dma_buf_poolp->dma_buf_pool_p;
2067 	dma_cntl_p = dma_cntl_poolp->dma_buf_pool_p;
2068 	rx_rbr_rings = (p_rx_rbr_rings_t)
2069 	    KMEM_ZALLOC(sizeof (rx_rbr_rings_t), KM_SLEEP);
2070 	rbr_rings = (p_rx_rbr_ring_t *)KMEM_ZALLOC(
2071 	    sizeof (p_rx_rbr_ring_t) * ndmas, KM_SLEEP);
2072 
2073 	rx_rcr_rings = (p_rx_rcr_rings_t)
2074 	    KMEM_ZALLOC(sizeof (rx_rcr_rings_t), KM_SLEEP);
2075 	rcr_rings = (p_rx_rcr_ring_t *)KMEM_ZALLOC(
2076 	    sizeof (p_rx_rcr_ring_t) * ndmas, KM_SLEEP);
2077 
2078 	rx_mbox_areas_p = (p_rx_mbox_areas_t)
2079 	    KMEM_ZALLOC(sizeof (rx_mbox_areas_t), KM_SLEEP);
2080 	rx_mbox_p = (p_rx_mbox_t *)KMEM_ZALLOC(
2081 	    sizeof (p_rx_mbox_t) * ndmas, KM_SLEEP);
2082 
2083 	/*
2084 	 * Timeout should be set based on the system clock divider.
2085 	 * The following timeout value of 1 assumes that the
2086 	 * granularity (1000) is 3 microseconds running at 300MHz.
2087 	 */
2088 
2089 	hxgep->intr_threshold = RXDMA_RCR_PTHRES_DEFAULT;
2090 	hxgep->intr_timeout = RXDMA_RCR_TO_DEFAULT;
2091 
2092 	/*
2093 	 * Map descriptors from the buffer polls for each dam channel.
2094 	 */
2095 	for (i = 0; i < ndmas; i++) {
2096 		/*
2097 		 * Set up and prepare buffer blocks, descriptors and mailbox.
2098 		 */
2099 		channel = ((p_hxge_dma_common_t)dma_buf_p[i])->dma_channel;
2100 		status = hxge_map_rxdma_channel(hxgep, channel,
2101 		    (p_hxge_dma_common_t *)&dma_buf_p[i],
2102 		    (p_rx_rbr_ring_t *)&rbr_rings[i],
2103 		    num_chunks[i], (p_hxge_dma_common_t *)&dma_cntl_p[i],
2104 		    (p_rx_rcr_ring_t *)&rcr_rings[i],
2105 		    (p_rx_mbox_t *)&rx_mbox_p[i]);
2106 		if (status != HXGE_OK) {
2107 			goto hxge_map_rxdma_fail1;
2108 		}
2109 		rbr_rings[i]->index = (uint16_t)i;
2110 		rcr_rings[i]->index = (uint16_t)i;
2111 		rcr_rings[i]->rdc_stats = &hxgep->statsp->rdc_stats[i];
2112 	}
2113 
2114 	rx_rbr_rings->ndmas = rx_rcr_rings->ndmas = ndmas;
2115 	rx_rbr_rings->rbr_rings = rbr_rings;
2116 	hxgep->rx_rbr_rings = rx_rbr_rings;
2117 	rx_rcr_rings->rcr_rings = rcr_rings;
2118 	hxgep->rx_rcr_rings = rx_rcr_rings;
2119 
2120 	rx_mbox_areas_p->rxmbox_areas = rx_mbox_p;
2121 	hxgep->rx_mbox_areas_p = rx_mbox_areas_p;
2122 
2123 	goto hxge_map_rxdma_exit;
2124 
2125 hxge_map_rxdma_fail1:
2126 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2127 	    "==> hxge_map_rxdma: unmap rbr,rcr (status 0x%x channel %d i %d)",
2128 	    status, channel, i));
2129 	i--;
2130 	for (; i >= 0; i--) {
2131 		channel = ((p_hxge_dma_common_t)dma_buf_p[i])->dma_channel;
2132 		hxge_unmap_rxdma_channel(hxgep, channel,
2133 		    rbr_rings[i], rcr_rings[i], rx_mbox_p[i]);
2134 	}
2135 
2136 	KMEM_FREE(rbr_rings, sizeof (p_rx_rbr_ring_t) * ndmas);
2137 	KMEM_FREE(rx_rbr_rings, sizeof (rx_rbr_rings_t));
2138 	KMEM_FREE(rcr_rings, sizeof (p_rx_rcr_ring_t) * ndmas);
2139 	KMEM_FREE(rx_rcr_rings, sizeof (rx_rcr_rings_t));
2140 	KMEM_FREE(rx_mbox_p, sizeof (p_rx_mbox_t) * ndmas);
2141 	KMEM_FREE(rx_mbox_areas_p, sizeof (rx_mbox_areas_t));
2142 
2143 hxge_map_rxdma_exit:
2144 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2145 	    "<== hxge_map_rxdma: (status 0x%x channel %d)", status, channel));
2146 
2147 	return (status);
2148 }
2149 
2150 static void
2151 hxge_unmap_rxdma(p_hxge_t hxgep)
2152 {
2153 	int			i, ndmas;
2154 	uint16_t		channel;
2155 	p_rx_rbr_rings_t	rx_rbr_rings;
2156 	p_rx_rbr_ring_t		*rbr_rings;
2157 	p_rx_rcr_rings_t	rx_rcr_rings;
2158 	p_rx_rcr_ring_t		*rcr_rings;
2159 	p_rx_mbox_areas_t	rx_mbox_areas_p;
2160 	p_rx_mbox_t		*rx_mbox_p;
2161 	p_hxge_dma_pool_t	dma_buf_poolp;
2162 	p_hxge_dma_pool_t	dma_cntl_poolp;
2163 	p_hxge_dma_common_t	*dma_buf_p;
2164 
2165 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_unmap_rxdma"));
2166 
2167 	dma_buf_poolp = hxgep->rx_buf_pool_p;
2168 	dma_cntl_poolp = hxgep->rx_cntl_pool_p;
2169 
2170 	if (!dma_buf_poolp->buf_allocated || !dma_cntl_poolp->buf_allocated) {
2171 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2172 		    "<== hxge_unmap_rxdma: NULL buf pointers"));
2173 		return;
2174 	}
2175 
2176 	rx_rbr_rings = hxgep->rx_rbr_rings;
2177 	rx_rcr_rings = hxgep->rx_rcr_rings;
2178 	if (rx_rbr_rings == NULL || rx_rcr_rings == NULL) {
2179 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2180 		    "<== hxge_unmap_rxdma: NULL ring pointers"));
2181 		return;
2182 	}
2183 
2184 	ndmas = rx_rbr_rings->ndmas;
2185 	if (!ndmas) {
2186 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2187 		    "<== hxge_unmap_rxdma: no channel"));
2188 		return;
2189 	}
2190 
2191 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2192 	    "==> hxge_unmap_rxdma (ndmas %d)", ndmas));
2193 
2194 	rbr_rings = rx_rbr_rings->rbr_rings;
2195 	rcr_rings = rx_rcr_rings->rcr_rings;
2196 	rx_mbox_areas_p = hxgep->rx_mbox_areas_p;
2197 	rx_mbox_p = rx_mbox_areas_p->rxmbox_areas;
2198 	dma_buf_p = dma_buf_poolp->dma_buf_pool_p;
2199 
2200 	for (i = 0; i < ndmas; i++) {
2201 		channel = ((p_hxge_dma_common_t)dma_buf_p[i])->dma_channel;
2202 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2203 		    "==> hxge_unmap_rxdma (ndmas %d) channel %d",
2204 		    ndmas, channel));
2205 		(void) hxge_unmap_rxdma_channel(hxgep, channel,
2206 		    (p_rx_rbr_ring_t)rbr_rings[i],
2207 		    (p_rx_rcr_ring_t)rcr_rings[i],
2208 		    (p_rx_mbox_t)rx_mbox_p[i]);
2209 	}
2210 
2211 	KMEM_FREE(rx_rbr_rings, sizeof (rx_rbr_rings_t));
2212 	KMEM_FREE(rbr_rings, sizeof (p_rx_rbr_ring_t) * ndmas);
2213 	KMEM_FREE(rx_rcr_rings, sizeof (rx_rcr_rings_t));
2214 	KMEM_FREE(rcr_rings, sizeof (p_rx_rcr_ring_t) * ndmas);
2215 	KMEM_FREE(rx_mbox_areas_p, sizeof (rx_mbox_areas_t));
2216 	KMEM_FREE(rx_mbox_p, sizeof (p_rx_mbox_t) * ndmas);
2217 
2218 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "<== hxge_unmap_rxdma"));
2219 }
2220 
2221 hxge_status_t
2222 hxge_map_rxdma_channel(p_hxge_t hxgep, uint16_t channel,
2223     p_hxge_dma_common_t *dma_buf_p, p_rx_rbr_ring_t *rbr_p,
2224     uint32_t num_chunks, p_hxge_dma_common_t *dma_cntl_p,
2225     p_rx_rcr_ring_t *rcr_p, p_rx_mbox_t *rx_mbox_p)
2226 {
2227 	int status = HXGE_OK;
2228 
2229 	/*
2230 	 * Set up and prepare buffer blocks, descriptors and mailbox.
2231 	 */
2232 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2233 	    "==> hxge_map_rxdma_channel (channel %d)", channel));
2234 
2235 	/*
2236 	 * Receive buffer blocks
2237 	 */
2238 	status = hxge_map_rxdma_channel_buf_ring(hxgep, channel,
2239 	    dma_buf_p, rbr_p, num_chunks);
2240 	if (status != HXGE_OK) {
2241 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2242 		    "==> hxge_map_rxdma_channel (channel %d): "
2243 		    "map buffer failed 0x%x", channel, status));
2244 		goto hxge_map_rxdma_channel_exit;
2245 	}
2246 
2247 	/*
2248 	 * Receive block ring, completion ring and mailbox.
2249 	 */
2250 	status = hxge_map_rxdma_channel_cfg_ring(hxgep, channel,
2251 	    dma_cntl_p, rbr_p, rcr_p, rx_mbox_p);
2252 	if (status != HXGE_OK) {
2253 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2254 		    "==> hxge_map_rxdma_channel (channel %d): "
2255 		    "map config failed 0x%x", channel, status));
2256 		goto hxge_map_rxdma_channel_fail2;
2257 	}
2258 	goto hxge_map_rxdma_channel_exit;
2259 
2260 hxge_map_rxdma_channel_fail3:
2261 	/* Free rbr, rcr */
2262 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2263 	    "==> hxge_map_rxdma_channel: free rbr/rcr (status 0x%x channel %d)",
2264 	    status, channel));
2265 	hxge_unmap_rxdma_channel_cfg_ring(hxgep, *rcr_p, *rx_mbox_p);
2266 
2267 hxge_map_rxdma_channel_fail2:
2268 	/* Free buffer blocks */
2269 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2270 	    "==> hxge_map_rxdma_channel: free rx buffers"
2271 	    "(hxgep 0x%x status 0x%x channel %d)",
2272 	    hxgep, status, channel));
2273 	hxge_unmap_rxdma_channel_buf_ring(hxgep, *rbr_p);
2274 
2275 	status = HXGE_ERROR;
2276 
2277 hxge_map_rxdma_channel_exit:
2278 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2279 	    "<== hxge_map_rxdma_channel: (hxgep 0x%x status 0x%x channel %d)",
2280 	    hxgep, status, channel));
2281 
2282 	return (status);
2283 }
2284 
2285 /*ARGSUSED*/
2286 static void
2287 hxge_unmap_rxdma_channel(p_hxge_t hxgep, uint16_t channel,
2288     p_rx_rbr_ring_t rbr_p, p_rx_rcr_ring_t rcr_p, p_rx_mbox_t rx_mbox_p)
2289 {
2290 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2291 	    "==> hxge_unmap_rxdma_channel (channel %d)", channel));
2292 
2293 	/*
2294 	 * unmap receive block ring, completion ring and mailbox.
2295 	 */
2296 	(void) hxge_unmap_rxdma_channel_cfg_ring(hxgep, rcr_p, rx_mbox_p);
2297 
2298 	/* unmap buffer blocks */
2299 	(void) hxge_unmap_rxdma_channel_buf_ring(hxgep, rbr_p);
2300 
2301 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "<== hxge_unmap_rxdma_channel"));
2302 }
2303 
2304 /*ARGSUSED*/
2305 static hxge_status_t
2306 hxge_map_rxdma_channel_cfg_ring(p_hxge_t hxgep, uint16_t dma_channel,
2307     p_hxge_dma_common_t *dma_cntl_p, p_rx_rbr_ring_t *rbr_p,
2308     p_rx_rcr_ring_t *rcr_p, p_rx_mbox_t *rx_mbox_p)
2309 {
2310 	p_rx_rbr_ring_t 	rbrp;
2311 	p_rx_rcr_ring_t 	rcrp;
2312 	p_rx_mbox_t 		mboxp;
2313 	p_hxge_dma_common_t 	cntl_dmap;
2314 	p_hxge_dma_common_t 	dmap;
2315 	p_rx_msg_t 		*rx_msg_ring;
2316 	p_rx_msg_t 		rx_msg_p;
2317 	rdc_rbr_cfg_a_t		*rcfga_p;
2318 	rdc_rbr_cfg_b_t		*rcfgb_p;
2319 	rdc_rcr_cfg_a_t		*cfga_p;
2320 	rdc_rcr_cfg_b_t		*cfgb_p;
2321 	rdc_rx_cfg1_t		*cfig1_p;
2322 	rdc_rx_cfg2_t		*cfig2_p;
2323 	rdc_rbr_kick_t		*kick_p;
2324 	uint32_t		dmaaddrp;
2325 	uint32_t		*rbr_vaddrp;
2326 	uint32_t		bkaddr;
2327 	hxge_status_t		status = HXGE_OK;
2328 	int			i;
2329 	uint32_t 		hxge_port_rcr_size;
2330 
2331 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2332 	    "==> hxge_map_rxdma_channel_cfg_ring"));
2333 
2334 	cntl_dmap = *dma_cntl_p;
2335 
2336 	/* Map in the receive block ring */
2337 	rbrp = *rbr_p;
2338 	dmap = (p_hxge_dma_common_t)&rbrp->rbr_desc;
2339 	hxge_setup_dma_common(dmap, cntl_dmap, rbrp->rbb_max, 4);
2340 
2341 	/*
2342 	 * Zero out buffer block ring descriptors.
2343 	 */
2344 	bzero((caddr_t)dmap->kaddrp, dmap->alength);
2345 
2346 	rcfga_p = &(rbrp->rbr_cfga);
2347 	rcfgb_p = &(rbrp->rbr_cfgb);
2348 	kick_p = &(rbrp->rbr_kick);
2349 	rcfga_p->value = 0;
2350 	rcfgb_p->value = 0;
2351 	kick_p->value = 0;
2352 	rbrp->rbr_addr = dmap->dma_cookie.dmac_laddress;
2353 	rcfga_p->value = (rbrp->rbr_addr &
2354 	    (RBR_CFIG_A_STDADDR_MASK | RBR_CFIG_A_STDADDR_BASE_MASK));
2355 	rcfga_p->value |= ((uint64_t)rbrp->rbb_max << RBR_CFIG_A_LEN_SHIFT);
2356 
2357 	/* XXXX: how to choose packet buffer sizes */
2358 	rcfgb_p->bits.bufsz0 = rbrp->pkt_buf_size0;
2359 	rcfgb_p->bits.vld0 = 1;
2360 	rcfgb_p->bits.bufsz1 = rbrp->pkt_buf_size1;
2361 	rcfgb_p->bits.vld1 = 1;
2362 	rcfgb_p->bits.bufsz2 = rbrp->pkt_buf_size2;
2363 	rcfgb_p->bits.vld2 = 1;
2364 	rcfgb_p->bits.bksize = hxgep->rx_bksize_code;
2365 
2366 	/*
2367 	 * For each buffer block, enter receive block address to the ring.
2368 	 */
2369 	rbr_vaddrp = (uint32_t *)dmap->kaddrp;
2370 	rbrp->rbr_desc_vp = (uint32_t *)dmap->kaddrp;
2371 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2372 	    "==> hxge_map_rxdma_channel_cfg_ring: channel %d "
2373 	    "rbr_vaddrp $%p", dma_channel, rbr_vaddrp));
2374 
2375 	rx_msg_ring = rbrp->rx_msg_ring;
2376 	for (i = 0; i < rbrp->tnblocks; i++) {
2377 		rx_msg_p = rx_msg_ring[i];
2378 		rx_msg_p->hxgep = hxgep;
2379 		rx_msg_p->rx_rbr_p = rbrp;
2380 		bkaddr = (uint32_t)
2381 		    ((rx_msg_p->buf_dma.dma_cookie.dmac_laddress >>
2382 		    RBR_BKADDR_SHIFT));
2383 		rx_msg_p->free = B_FALSE;
2384 		rx_msg_p->max_usage_cnt = 0xbaddcafe;
2385 
2386 		*rbr_vaddrp++ = bkaddr;
2387 	}
2388 
2389 	kick_p->bits.bkadd = rbrp->rbb_max;
2390 	rbrp->rbr_wr_index = (rbrp->rbb_max - 1);
2391 
2392 	rbrp->rbr_rd_index = 0;
2393 
2394 	rbrp->rbr_consumed = 0;
2395 	rbrp->rbr_use_bcopy = B_TRUE;
2396 	rbrp->rbr_bufsize_type = RCR_PKTBUFSZ_0;
2397 
2398 	/*
2399 	 * Do bcopy on packets greater than bcopy size once the lo threshold is
2400 	 * reached. This lo threshold should be less than the hi threshold.
2401 	 *
2402 	 * Do bcopy on every packet once the hi threshold is reached.
2403 	 */
2404 	if (hxge_rx_threshold_lo >= hxge_rx_threshold_hi) {
2405 		/* default it to use hi */
2406 		hxge_rx_threshold_lo = hxge_rx_threshold_hi;
2407 	}
2408 	if (hxge_rx_buf_size_type > HXGE_RBR_TYPE2) {
2409 		hxge_rx_buf_size_type = HXGE_RBR_TYPE2;
2410 	}
2411 	rbrp->rbr_bufsize_type = hxge_rx_buf_size_type;
2412 
2413 	switch (hxge_rx_threshold_hi) {
2414 	default:
2415 	case HXGE_RX_COPY_NONE:
2416 		/* Do not do bcopy at all */
2417 		rbrp->rbr_use_bcopy = B_FALSE;
2418 		rbrp->rbr_threshold_hi = rbrp->rbb_max;
2419 		break;
2420 
2421 	case HXGE_RX_COPY_1:
2422 	case HXGE_RX_COPY_2:
2423 	case HXGE_RX_COPY_3:
2424 	case HXGE_RX_COPY_4:
2425 	case HXGE_RX_COPY_5:
2426 	case HXGE_RX_COPY_6:
2427 	case HXGE_RX_COPY_7:
2428 		rbrp->rbr_threshold_hi =
2429 		    rbrp->rbb_max * (hxge_rx_threshold_hi) /
2430 		    HXGE_RX_BCOPY_SCALE;
2431 		break;
2432 
2433 	case HXGE_RX_COPY_ALL:
2434 		rbrp->rbr_threshold_hi = 0;
2435 		break;
2436 	}
2437 
2438 	switch (hxge_rx_threshold_lo) {
2439 	default:
2440 	case HXGE_RX_COPY_NONE:
2441 		/* Do not do bcopy at all */
2442 		if (rbrp->rbr_use_bcopy) {
2443 			rbrp->rbr_use_bcopy = B_FALSE;
2444 		}
2445 		rbrp->rbr_threshold_lo = rbrp->rbb_max;
2446 		break;
2447 
2448 	case HXGE_RX_COPY_1:
2449 	case HXGE_RX_COPY_2:
2450 	case HXGE_RX_COPY_3:
2451 	case HXGE_RX_COPY_4:
2452 	case HXGE_RX_COPY_5:
2453 	case HXGE_RX_COPY_6:
2454 	case HXGE_RX_COPY_7:
2455 		rbrp->rbr_threshold_lo =
2456 		    rbrp->rbb_max * (hxge_rx_threshold_lo) /
2457 		    HXGE_RX_BCOPY_SCALE;
2458 		break;
2459 
2460 	case HXGE_RX_COPY_ALL:
2461 		rbrp->rbr_threshold_lo = 0;
2462 		break;
2463 	}
2464 
2465 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
2466 	    "hxge_map_rxdma_channel_cfg_ring: channel %d rbb_max %d "
2467 	    "rbrp->rbr_bufsize_type %d rbb_threshold_hi %d "
2468 	    "rbb_threshold_lo %d",
2469 	    dma_channel, rbrp->rbb_max, rbrp->rbr_bufsize_type,
2470 	    rbrp->rbr_threshold_hi, rbrp->rbr_threshold_lo));
2471 
2472 	/* Map in the receive completion ring */
2473 	rcrp = (p_rx_rcr_ring_t)KMEM_ZALLOC(sizeof (rx_rcr_ring_t), KM_SLEEP);
2474 	rcrp->rdc = dma_channel;
2475 
2476 	hxge_port_rcr_size = hxgep->hxge_port_rcr_size;
2477 	rcrp->comp_size = hxge_port_rcr_size;
2478 	rcrp->comp_wrap_mask = hxge_port_rcr_size - 1;
2479 
2480 	rcrp->max_receive_pkts = hxge_max_rx_pkts;
2481 
2482 	dmap = (p_hxge_dma_common_t)&rcrp->rcr_desc;
2483 	hxge_setup_dma_common(dmap, cntl_dmap, rcrp->comp_size,
2484 	    sizeof (rcr_entry_t));
2485 	rcrp->comp_rd_index = 0;
2486 	rcrp->comp_wt_index = 0;
2487 	rcrp->rcr_desc_rd_head_p = rcrp->rcr_desc_first_p =
2488 	    (p_rcr_entry_t)DMA_COMMON_VPTR(rcrp->rcr_desc);
2489 #if defined(__i386)
2490 	rcrp->rcr_desc_rd_head_pp = rcrp->rcr_desc_first_pp =
2491 	    (p_rcr_entry_t)(uint32_t)DMA_COMMON_IOADDR(rcrp->rcr_desc);
2492 #else
2493 	rcrp->rcr_desc_rd_head_pp = rcrp->rcr_desc_first_pp =
2494 	    (p_rcr_entry_t)DMA_COMMON_IOADDR(rcrp->rcr_desc);
2495 #endif
2496 	rcrp->rcr_desc_last_p = rcrp->rcr_desc_rd_head_p +
2497 	    (hxge_port_rcr_size - 1);
2498 	rcrp->rcr_desc_last_pp = rcrp->rcr_desc_rd_head_pp +
2499 	    (hxge_port_rcr_size - 1);
2500 
2501 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2502 	    "==> hxge_map_rxdma_channel_cfg_ring: channel %d "
2503 	    "rbr_vaddrp $%p rcr_desc_rd_head_p $%p "
2504 	    "rcr_desc_rd_head_pp $%p rcr_desc_rd_last_p $%p "
2505 	    "rcr_desc_rd_last_pp $%p ",
2506 	    dma_channel, rbr_vaddrp, rcrp->rcr_desc_rd_head_p,
2507 	    rcrp->rcr_desc_rd_head_pp, rcrp->rcr_desc_last_p,
2508 	    rcrp->rcr_desc_last_pp));
2509 
2510 	/*
2511 	 * Zero out buffer block ring descriptors.
2512 	 */
2513 	bzero((caddr_t)dmap->kaddrp, dmap->alength);
2514 	rcrp->intr_timeout = hxgep->intr_timeout;
2515 	rcrp->intr_threshold = hxgep->intr_threshold;
2516 	rcrp->full_hdr_flag = B_FALSE;
2517 	rcrp->sw_priv_hdr_len = 0;
2518 
2519 	cfga_p = &(rcrp->rcr_cfga);
2520 	cfgb_p = &(rcrp->rcr_cfgb);
2521 	cfga_p->value = 0;
2522 	cfgb_p->value = 0;
2523 	rcrp->rcr_addr = dmap->dma_cookie.dmac_laddress;
2524 
2525 	cfga_p->value = (rcrp->rcr_addr &
2526 	    (RCRCFIG_A_STADDR_MASK | RCRCFIG_A_STADDR_BASE_MASK));
2527 
2528 	cfga_p->value |= ((uint64_t)rcrp->comp_size << RCRCFIG_A_LEN_SHIF);
2529 
2530 	/*
2531 	 * Timeout should be set based on the system clock divider. The
2532 	 * following timeout value of 1 assumes that the granularity (1000) is
2533 	 * 3 microseconds running at 300MHz.
2534 	 */
2535 	cfgb_p->bits.pthres = rcrp->intr_threshold;
2536 	cfgb_p->bits.timeout = rcrp->intr_timeout;
2537 	cfgb_p->bits.entout = 1;
2538 
2539 	/* Map in the mailbox */
2540 	mboxp = (p_rx_mbox_t)KMEM_ZALLOC(sizeof (rx_mbox_t), KM_SLEEP);
2541 	dmap = (p_hxge_dma_common_t)&mboxp->rx_mbox;
2542 	hxge_setup_dma_common(dmap, cntl_dmap, 1, sizeof (rxdma_mailbox_t));
2543 	cfig1_p = (rdc_rx_cfg1_t *)&mboxp->rx_cfg1;
2544 	cfig2_p = (rdc_rx_cfg2_t *)&mboxp->rx_cfg2;
2545 	cfig1_p->value = cfig2_p->value = 0;
2546 
2547 	mboxp->mbox_addr = dmap->dma_cookie.dmac_laddress;
2548 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2549 	    "==> hxge_map_rxdma_channel_cfg_ring: "
2550 	    "channel %d cfg1 0x%016llx cfig2 0x%016llx cookie 0x%016llx",
2551 	    dma_channel, cfig1_p->value, cfig2_p->value,
2552 	    mboxp->mbox_addr));
2553 
2554 	dmaaddrp = (uint32_t)((dmap->dma_cookie.dmac_laddress >> 32) & 0xfff);
2555 	cfig1_p->bits.mbaddr_h = dmaaddrp;
2556 
2557 	dmaaddrp = (uint32_t)(dmap->dma_cookie.dmac_laddress & 0xffffffff);
2558 	dmaaddrp = (uint32_t)(dmap->dma_cookie.dmac_laddress &
2559 	    RXDMA_CFIG2_MBADDR_L_MASK);
2560 
2561 	cfig2_p->bits.mbaddr_l = (dmaaddrp >> RXDMA_CFIG2_MBADDR_L_SHIFT);
2562 
2563 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2564 	    "==> hxge_map_rxdma_channel_cfg_ring: channel %d damaddrp $%p "
2565 	    "cfg1 0x%016llx cfig2 0x%016llx",
2566 	    dma_channel, dmaaddrp, cfig1_p->value, cfig2_p->value));
2567 
2568 	cfig2_p->bits.full_hdr = rcrp->full_hdr_flag;
2569 	cfig2_p->bits.offset = rcrp->sw_priv_hdr_len;
2570 
2571 	rbrp->rx_rcr_p = rcrp;
2572 	rcrp->rx_rbr_p = rbrp;
2573 	*rcr_p = rcrp;
2574 	*rx_mbox_p = mboxp;
2575 
2576 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2577 	    "<== hxge_map_rxdma_channel_cfg_ring status 0x%08x", status));
2578 	return (status);
2579 }
2580 
2581 /*ARGSUSED*/
2582 static void
2583 hxge_unmap_rxdma_channel_cfg_ring(p_hxge_t hxgep,
2584     p_rx_rcr_ring_t rcr_p, p_rx_mbox_t rx_mbox_p)
2585 {
2586 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2587 	    "==> hxge_unmap_rxdma_channel_cfg_ring: channel %d", rcr_p->rdc));
2588 
2589 	KMEM_FREE(rcr_p, sizeof (rx_rcr_ring_t));
2590 	KMEM_FREE(rx_mbox_p, sizeof (rx_mbox_t));
2591 
2592 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2593 	    "<== hxge_unmap_rxdma_channel_cfg_ring"));
2594 }
2595 
2596 static hxge_status_t
2597 hxge_map_rxdma_channel_buf_ring(p_hxge_t hxgep, uint16_t channel,
2598     p_hxge_dma_common_t *dma_buf_p,
2599     p_rx_rbr_ring_t *rbr_p, uint32_t num_chunks)
2600 {
2601 	p_rx_rbr_ring_t		rbrp;
2602 	p_hxge_dma_common_t	dma_bufp, tmp_bufp;
2603 	p_rx_msg_t		*rx_msg_ring;
2604 	p_rx_msg_t		rx_msg_p;
2605 	p_mblk_t		mblk_p;
2606 
2607 	rxring_info_t *ring_info;
2608 	hxge_status_t status = HXGE_OK;
2609 	int i, j, index;
2610 	uint32_t size, bsize, nblocks, nmsgs;
2611 
2612 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2613 	    "==> hxge_map_rxdma_channel_buf_ring: channel %d", channel));
2614 
2615 	dma_bufp = tmp_bufp = *dma_buf_p;
2616 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2617 	    " hxge_map_rxdma_channel_buf_ring: channel %d to map %d "
2618 	    "chunks bufp 0x%016llx", channel, num_chunks, dma_bufp));
2619 
2620 	nmsgs = 0;
2621 	for (i = 0; i < num_chunks; i++, tmp_bufp++) {
2622 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2623 		    "==> hxge_map_rxdma_channel_buf_ring: channel %d "
2624 		    "bufp 0x%016llx nblocks %d nmsgs %d",
2625 		    channel, tmp_bufp, tmp_bufp->nblocks, nmsgs));
2626 		nmsgs += tmp_bufp->nblocks;
2627 	}
2628 	if (!nmsgs) {
2629 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2630 		    "<== hxge_map_rxdma_channel_buf_ring: channel %d "
2631 		    "no msg blocks", channel));
2632 		status = HXGE_ERROR;
2633 		goto hxge_map_rxdma_channel_buf_ring_exit;
2634 	}
2635 	rbrp = (p_rx_rbr_ring_t)KMEM_ZALLOC(sizeof (rx_rbr_ring_t), KM_SLEEP);
2636 
2637 	size = nmsgs * sizeof (p_rx_msg_t);
2638 	rx_msg_ring = KMEM_ZALLOC(size, KM_SLEEP);
2639 	ring_info = (rxring_info_t *)KMEM_ZALLOC(sizeof (rxring_info_t),
2640 	    KM_SLEEP);
2641 
2642 	MUTEX_INIT(&rbrp->lock, NULL, MUTEX_DRIVER,
2643 	    (void *) hxgep->interrupt_cookie);
2644 	MUTEX_INIT(&rbrp->post_lock, NULL, MUTEX_DRIVER,
2645 	    (void *) hxgep->interrupt_cookie);
2646 	rbrp->rdc = channel;
2647 	rbrp->num_blocks = num_chunks;
2648 	rbrp->tnblocks = nmsgs;
2649 	rbrp->rbb_max = nmsgs;
2650 	rbrp->rbr_max_size = nmsgs;
2651 	rbrp->rbr_wrap_mask = (rbrp->rbb_max - 1);
2652 
2653 	rbrp->pages_to_post = 0;
2654 	rbrp->pages_to_skip = 20;
2655 	rbrp->pages_to_post_threshold = rbrp->rbb_max - rbrp->pages_to_skip / 2;
2656 
2657 	/*
2658 	 * Buffer sizes suggested by NIU architect. 256, 512 and 2K.
2659 	 */
2660 
2661 	rbrp->pkt_buf_size0 = RBR_BUFSZ0_256B;
2662 	rbrp->pkt_buf_size0_bytes = RBR_BUFSZ0_256_BYTES;
2663 	rbrp->hpi_pkt_buf_size0 = SIZE_256B;
2664 
2665 	rbrp->pkt_buf_size1 = RBR_BUFSZ1_1K;
2666 	rbrp->pkt_buf_size1_bytes = RBR_BUFSZ1_1K_BYTES;
2667 	rbrp->hpi_pkt_buf_size1 = SIZE_1KB;
2668 
2669 	rbrp->block_size = hxgep->rx_default_block_size;
2670 
2671 	if (!hxge_jumbo_enable && !hxgep->param_arr[param_accept_jumbo].value) {
2672 		rbrp->pkt_buf_size2 = RBR_BUFSZ2_2K;
2673 		rbrp->pkt_buf_size2_bytes = RBR_BUFSZ2_2K_BYTES;
2674 		rbrp->hpi_pkt_buf_size2 = SIZE_2KB;
2675 	} else {
2676 		if (rbrp->block_size >= 0x2000) {
2677 			HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2678 			    "<== hxge_map_rxdma_channel_buf_ring: channel %d "
2679 			    "no msg blocks", channel));
2680 			status = HXGE_ERROR;
2681 			goto hxge_map_rxdma_channel_buf_ring_fail1;
2682 		} else {
2683 			rbrp->pkt_buf_size2 = RBR_BUFSZ2_4K;
2684 			rbrp->pkt_buf_size2_bytes = RBR_BUFSZ2_4K_BYTES;
2685 			rbrp->hpi_pkt_buf_size2 = SIZE_4KB;
2686 		}
2687 	}
2688 
2689 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2690 	    "==> hxge_map_rxdma_channel_buf_ring: channel %d "
2691 	    "actual rbr max %d rbb_max %d nmsgs %d "
2692 	    "rbrp->block_size %d default_block_size %d "
2693 	    "(config hxge_rbr_size %d hxge_rbr_spare_size %d)",
2694 	    channel, rbrp->rbr_max_size, rbrp->rbb_max, nmsgs,
2695 	    rbrp->block_size, hxgep->rx_default_block_size,
2696 	    hxge_rbr_size, hxge_rbr_spare_size));
2697 
2698 	/*
2699 	 * Map in buffers from the buffer pool.
2700 	 * Note that num_blocks is the num_chunks. For Sparc, there is likely
2701 	 * only one chunk. For x86, there will be many chunks.
2702 	 * Loop over chunks.
2703 	 */
2704 	index = 0;
2705 	for (i = 0; i < rbrp->num_blocks; i++, dma_bufp++) {
2706 		bsize = dma_bufp->block_size;
2707 		nblocks = dma_bufp->nblocks;
2708 #if defined(__i386)
2709 		ring_info->buffer[i].dvma_addr = (uint32_t)dma_bufp->ioaddr_pp;
2710 #else
2711 		ring_info->buffer[i].dvma_addr = (uint64_t)dma_bufp->ioaddr_pp;
2712 #endif
2713 		ring_info->buffer[i].buf_index = i;
2714 		ring_info->buffer[i].buf_size = dma_bufp->alength;
2715 		ring_info->buffer[i].start_index = index;
2716 #if defined(__i386)
2717 		ring_info->buffer[i].kaddr = (uint32_t)dma_bufp->kaddrp;
2718 #else
2719 		ring_info->buffer[i].kaddr = (uint64_t)dma_bufp->kaddrp;
2720 #endif
2721 
2722 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2723 		    " hxge_map_rxdma_channel_buf_ring: map channel %d "
2724 		    "chunk %d nblocks %d chunk_size %x block_size 0x%x "
2725 		    "dma_bufp $%p dvma_addr $%p", channel, i,
2726 		    dma_bufp->nblocks,
2727 		    ring_info->buffer[i].buf_size, bsize, dma_bufp,
2728 		    ring_info->buffer[i].dvma_addr));
2729 
2730 		/* loop over blocks within a chunk */
2731 		for (j = 0; j < nblocks; j++) {
2732 			if ((rx_msg_p = hxge_allocb(bsize, BPRI_LO,
2733 			    dma_bufp)) == NULL) {
2734 				HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2735 				    "allocb failed (index %d i %d j %d)",
2736 				    index, i, j));
2737 				goto hxge_map_rxdma_channel_buf_ring_fail1;
2738 			}
2739 			rx_msg_ring[index] = rx_msg_p;
2740 			rx_msg_p->block_index = index;
2741 			rx_msg_p->shifted_addr = (uint32_t)
2742 			    ((rx_msg_p->buf_dma.dma_cookie.dmac_laddress >>
2743 			    RBR_BKADDR_SHIFT));
2744 			/*
2745 			 * Too much output
2746 			 * HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2747 			 *	"index %d j %d rx_msg_p $%p mblk %p",
2748 			 *	index, j, rx_msg_p, rx_msg_p->rx_mblk_p));
2749 			 */
2750 			mblk_p = rx_msg_p->rx_mblk_p;
2751 			mblk_p->b_wptr = mblk_p->b_rptr + bsize;
2752 
2753 			rbrp->rbr_ref_cnt++;
2754 			index++;
2755 			rx_msg_p->buf_dma.dma_channel = channel;
2756 		}
2757 	}
2758 	if (i < rbrp->num_blocks) {
2759 		goto hxge_map_rxdma_channel_buf_ring_fail1;
2760 	}
2761 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2762 	    "hxge_map_rxdma_channel_buf_ring: done buf init "
2763 	    "channel %d msg block entries %d", channel, index));
2764 	ring_info->block_size_mask = bsize - 1;
2765 	rbrp->rx_msg_ring = rx_msg_ring;
2766 	rbrp->dma_bufp = dma_buf_p;
2767 	rbrp->ring_info = ring_info;
2768 
2769 	status = hxge_rxbuf_index_info_init(hxgep, rbrp);
2770 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, " hxge_map_rxdma_channel_buf_ring: "
2771 	    "channel %d done buf info init", channel));
2772 
2773 	/*
2774 	 * Finally, permit hxge_freeb() to call hxge_post_page().
2775 	 */
2776 	rbrp->rbr_state = RBR_POSTING;
2777 
2778 	*rbr_p = rbrp;
2779 
2780 	goto hxge_map_rxdma_channel_buf_ring_exit;
2781 
2782 hxge_map_rxdma_channel_buf_ring_fail1:
2783 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2784 	    " hxge_map_rxdma_channel_buf_ring: failed channel (0x%x)",
2785 	    channel, status));
2786 
2787 	index--;
2788 	for (; index >= 0; index--) {
2789 		rx_msg_p = rx_msg_ring[index];
2790 		if (rx_msg_p != NULL) {
2791 			hxge_freeb(rx_msg_p);
2792 			rx_msg_ring[index] = NULL;
2793 		}
2794 	}
2795 
2796 hxge_map_rxdma_channel_buf_ring_fail:
2797 	MUTEX_DESTROY(&rbrp->post_lock);
2798 	MUTEX_DESTROY(&rbrp->lock);
2799 	KMEM_FREE(ring_info, sizeof (rxring_info_t));
2800 	KMEM_FREE(rx_msg_ring, size);
2801 	KMEM_FREE(rbrp, sizeof (rx_rbr_ring_t));
2802 
2803 	status = HXGE_ERROR;
2804 
2805 hxge_map_rxdma_channel_buf_ring_exit:
2806 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2807 	    "<== hxge_map_rxdma_channel_buf_ring status 0x%08x", status));
2808 
2809 	return (status);
2810 }
2811 
2812 /*ARGSUSED*/
2813 static void
2814 hxge_unmap_rxdma_channel_buf_ring(p_hxge_t hxgep,
2815     p_rx_rbr_ring_t rbr_p)
2816 {
2817 	p_rx_msg_t	*rx_msg_ring;
2818 	p_rx_msg_t	rx_msg_p;
2819 	rxring_info_t	*ring_info;
2820 	int		i;
2821 	uint32_t	size;
2822 
2823 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2824 	    "==> hxge_unmap_rxdma_channel_buf_ring"));
2825 	if (rbr_p == NULL) {
2826 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
2827 		    "<== hxge_unmap_rxdma_channel_buf_ring: NULL rbrp"));
2828 		return;
2829 	}
2830 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2831 	    "==> hxge_unmap_rxdma_channel_buf_ring: channel %d", rbr_p->rdc));
2832 
2833 	rx_msg_ring = rbr_p->rx_msg_ring;
2834 	ring_info = rbr_p->ring_info;
2835 
2836 	if (rx_msg_ring == NULL || ring_info == NULL) {
2837 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2838 		    "<== hxge_unmap_rxdma_channel_buf_ring: "
2839 		    "rx_msg_ring $%p ring_info $%p", rx_msg_p, ring_info));
2840 		return;
2841 	}
2842 
2843 	size = rbr_p->tnblocks * sizeof (p_rx_msg_t);
2844 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2845 	    " hxge_unmap_rxdma_channel_buf_ring: channel %d chunks %d "
2846 	    "tnblocks %d (max %d) size ptrs %d ", rbr_p->rdc, rbr_p->num_blocks,
2847 	    rbr_p->tnblocks, rbr_p->rbr_max_size, size));
2848 
2849 	for (i = 0; i < rbr_p->tnblocks; i++) {
2850 		rx_msg_p = rx_msg_ring[i];
2851 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2852 		    " hxge_unmap_rxdma_channel_buf_ring: "
2853 		    "rx_msg_p $%p", rx_msg_p));
2854 		if (rx_msg_p != NULL) {
2855 			hxge_freeb(rx_msg_p);
2856 			rx_msg_ring[i] = NULL;
2857 		}
2858 	}
2859 
2860 	/*
2861 	 * We no longer may use the mutex <post_lock>. By setting
2862 	 * <rbr_state> to anything but POSTING, we prevent
2863 	 * hxge_post_page() from accessing a dead mutex.
2864 	 */
2865 	rbr_p->rbr_state = RBR_UNMAPPING;
2866 	MUTEX_DESTROY(&rbr_p->post_lock);
2867 
2868 	MUTEX_DESTROY(&rbr_p->lock);
2869 	KMEM_FREE(ring_info, sizeof (rxring_info_t));
2870 	KMEM_FREE(rx_msg_ring, size);
2871 
2872 	if (rbr_p->rbr_ref_cnt == 0) {
2873 		/* This is the normal state of affairs. */
2874 		KMEM_FREE(rbr_p, sizeof (*rbr_p));
2875 	} else {
2876 		/*
2877 		 * Some of our buffers are still being used.
2878 		 * Therefore, tell hxge_freeb() this ring is
2879 		 * unmapped, so it may free <rbr_p> for us.
2880 		 */
2881 		rbr_p->rbr_state = RBR_UNMAPPED;
2882 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2883 		    "unmap_rxdma_buf_ring: %d %s outstanding.",
2884 		    rbr_p->rbr_ref_cnt,
2885 		    rbr_p->rbr_ref_cnt == 1 ? "msg" : "msgs"));
2886 	}
2887 
2888 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2889 	    "<== hxge_unmap_rxdma_channel_buf_ring"));
2890 }
2891 
2892 static hxge_status_t
2893 hxge_rxdma_hw_start_common(p_hxge_t hxgep)
2894 {
2895 	hxge_status_t status = HXGE_OK;
2896 
2897 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_hw_start_common"));
2898 
2899 	/*
2900 	 * Load the sharable parameters by writing to the function zero control
2901 	 * registers. These FZC registers should be initialized only once for
2902 	 * the entire chip.
2903 	 */
2904 	(void) hxge_init_fzc_rx_common(hxgep);
2905 
2906 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_hw_start_common"));
2907 
2908 	return (status);
2909 }
2910 
2911 static hxge_status_t
2912 hxge_rxdma_hw_start(p_hxge_t hxgep)
2913 {
2914 	int			i, ndmas;
2915 	uint16_t		channel;
2916 	p_rx_rbr_rings_t	rx_rbr_rings;
2917 	p_rx_rbr_ring_t		*rbr_rings;
2918 	p_rx_rcr_rings_t	rx_rcr_rings;
2919 	p_rx_rcr_ring_t		*rcr_rings;
2920 	p_rx_mbox_areas_t	rx_mbox_areas_p;
2921 	p_rx_mbox_t		*rx_mbox_p;
2922 	hxge_status_t		status = HXGE_OK;
2923 
2924 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_hw_start"));
2925 
2926 	rx_rbr_rings = hxgep->rx_rbr_rings;
2927 	rx_rcr_rings = hxgep->rx_rcr_rings;
2928 	if (rx_rbr_rings == NULL || rx_rcr_rings == NULL) {
2929 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
2930 		    "<== hxge_rxdma_hw_start: NULL ring pointers"));
2931 		return (HXGE_ERROR);
2932 	}
2933 
2934 	ndmas = rx_rbr_rings->ndmas;
2935 	if (ndmas == 0) {
2936 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
2937 		    "<== hxge_rxdma_hw_start: no dma channel allocated"));
2938 		return (HXGE_ERROR);
2939 	}
2940 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2941 	    "==> hxge_rxdma_hw_start (ndmas %d)", ndmas));
2942 
2943 	/*
2944 	 * Scrub the RDC Rx DMA Prefetch Buffer Command.
2945 	 */
2946 	for (i = 0; i < 128; i++) {
2947 		HXGE_REG_WR64(hxgep->hpi_handle, RDC_PREF_CMD, i);
2948 	}
2949 
2950 	/*
2951 	 * Scrub Rx DMA Shadow Tail Command.
2952 	 */
2953 	for (i = 0; i < 64; i++) {
2954 		HXGE_REG_WR64(hxgep->hpi_handle, RDC_SHADOW_CMD, i);
2955 	}
2956 
2957 	/*
2958 	 * Scrub Rx DMA Control Fifo Command.
2959 	 */
2960 	for (i = 0; i < 512; i++) {
2961 		HXGE_REG_WR64(hxgep->hpi_handle, RDC_CTRL_FIFO_CMD, i);
2962 	}
2963 
2964 	/*
2965 	 * Scrub Rx DMA Data Fifo Command.
2966 	 */
2967 	for (i = 0; i < 1536; i++) {
2968 		HXGE_REG_WR64(hxgep->hpi_handle, RDC_DATA_FIFO_CMD, i);
2969 	}
2970 
2971 	/*
2972 	 * Reset the FIFO Error Stat.
2973 	 */
2974 	HXGE_REG_WR64(hxgep->hpi_handle, RDC_FIFO_ERR_STAT, 0xFF);
2975 
2976 	/* Set the error mask to receive interrupts */
2977 	HXGE_REG_WR64(hxgep->hpi_handle, RDC_FIFO_ERR_INT_MASK, 0x0);
2978 
2979 	rbr_rings = rx_rbr_rings->rbr_rings;
2980 	rcr_rings = rx_rcr_rings->rcr_rings;
2981 	rx_mbox_areas_p = hxgep->rx_mbox_areas_p;
2982 	if (rx_mbox_areas_p) {
2983 		rx_mbox_p = rx_mbox_areas_p->rxmbox_areas;
2984 	}
2985 
2986 	for (i = 0; i < ndmas; i++) {
2987 		channel = rbr_rings[i]->rdc;
2988 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2989 		    "==> hxge_rxdma_hw_start (ndmas %d) channel %d",
2990 		    ndmas, channel));
2991 		status = hxge_rxdma_start_channel(hxgep, channel,
2992 		    (p_rx_rbr_ring_t)rbr_rings[i],
2993 		    (p_rx_rcr_ring_t)rcr_rings[i],
2994 		    (p_rx_mbox_t)rx_mbox_p[i]);
2995 		if (status != HXGE_OK) {
2996 			goto hxge_rxdma_hw_start_fail1;
2997 		}
2998 	}
2999 
3000 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_hw_start: "
3001 	    "rx_rbr_rings 0x%016llx rings 0x%016llx",
3002 	    rx_rbr_rings, rx_rcr_rings));
3003 	goto hxge_rxdma_hw_start_exit;
3004 
3005 hxge_rxdma_hw_start_fail1:
3006 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3007 	    "==> hxge_rxdma_hw_start: disable "
3008 	    "(status 0x%x channel %d i %d)", status, channel, i));
3009 	for (; i >= 0; i--) {
3010 		channel = rbr_rings[i]->rdc;
3011 		(void) hxge_rxdma_stop_channel(hxgep, channel);
3012 	}
3013 
3014 hxge_rxdma_hw_start_exit:
3015 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3016 	    "==> hxge_rxdma_hw_start: (status 0x%x)", status));
3017 	return (status);
3018 }
3019 
3020 static void
3021 hxge_rxdma_hw_stop(p_hxge_t hxgep)
3022 {
3023 	int			i, ndmas;
3024 	uint16_t		channel;
3025 	p_rx_rbr_rings_t	rx_rbr_rings;
3026 	p_rx_rbr_ring_t		*rbr_rings;
3027 	p_rx_rcr_rings_t	rx_rcr_rings;
3028 
3029 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_hw_stop"));
3030 
3031 	rx_rbr_rings = hxgep->rx_rbr_rings;
3032 	rx_rcr_rings = hxgep->rx_rcr_rings;
3033 
3034 	if (rx_rbr_rings == NULL || rx_rcr_rings == NULL) {
3035 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
3036 		    "<== hxge_rxdma_hw_stop: NULL ring pointers"));
3037 		return;
3038 	}
3039 
3040 	ndmas = rx_rbr_rings->ndmas;
3041 	if (!ndmas) {
3042 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
3043 		    "<== hxge_rxdma_hw_stop: no dma channel allocated"));
3044 		return;
3045 	}
3046 
3047 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3048 	    "==> hxge_rxdma_hw_stop (ndmas %d)", ndmas));
3049 
3050 	rbr_rings = rx_rbr_rings->rbr_rings;
3051 	for (i = 0; i < ndmas; i++) {
3052 		channel = rbr_rings[i]->rdc;
3053 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3054 		    "==> hxge_rxdma_hw_stop (ndmas %d) channel %d",
3055 		    ndmas, channel));
3056 		(void) hxge_rxdma_stop_channel(hxgep, channel);
3057 	}
3058 
3059 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_hw_stop: "
3060 	    "rx_rbr_rings 0x%016llx rings 0x%016llx",
3061 	    rx_rbr_rings, rx_rcr_rings));
3062 
3063 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "<== hxge_rxdma_hw_stop"));
3064 }
3065 
3066 static hxge_status_t
3067 hxge_rxdma_start_channel(p_hxge_t hxgep, uint16_t channel,
3068     p_rx_rbr_ring_t rbr_p, p_rx_rcr_ring_t rcr_p, p_rx_mbox_t mbox_p)
3069 {
3070 	hpi_handle_t		handle;
3071 	hpi_status_t		rs = HPI_SUCCESS;
3072 	rdc_stat_t		cs;
3073 	rdc_int_mask_t		ent_mask;
3074 	hxge_status_t		status = HXGE_OK;
3075 
3076 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_start_channel"));
3077 
3078 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
3079 
3080 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "hxge_rxdma_start_channel: "
3081 	    "hpi handle addr $%p acc $%p",
3082 	    hxgep->hpi_handle.regp, hxgep->hpi_handle.regh));
3083 
3084 	/* Reset RXDMA channel */
3085 	rs = hpi_rxdma_cfg_rdc_reset(handle, channel);
3086 	if (rs != HPI_SUCCESS) {
3087 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3088 		    "==> hxge_rxdma_start_channel: "
3089 		    "reset rxdma failed (0x%08x channel %d)",
3090 		    status, channel));
3091 		return (HXGE_ERROR | rs);
3092 	}
3093 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3094 	    "==> hxge_rxdma_start_channel: reset done: channel %d", channel));
3095 
3096 	/*
3097 	 * Initialize the RXDMA channel specific FZC control configurations.
3098 	 * These FZC registers are pertaining to each RX channel (logical
3099 	 * pages).
3100 	 */
3101 	status = hxge_init_fzc_rxdma_channel(hxgep,
3102 	    channel, rbr_p, rcr_p, mbox_p);
3103 	if (status != HXGE_OK) {
3104 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3105 		    "==> hxge_rxdma_start_channel: "
3106 		    "init fzc rxdma failed (0x%08x channel %d)",
3107 		    status, channel));
3108 		return (status);
3109 	}
3110 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3111 	    "==> hxge_rxdma_start_channel: fzc done"));
3112 
3113 	/*
3114 	 * Zero out the shadow  and prefetch ram.
3115 	 */
3116 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3117 	    "==> hxge_rxdma_start_channel: ram done"));
3118 
3119 	/* Set up the interrupt event masks. */
3120 	ent_mask.value = 0;
3121 	rs = hpi_rxdma_event_mask(handle, OP_SET, channel, &ent_mask);
3122 	if (rs != HPI_SUCCESS) {
3123 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3124 		    "==> hxge_rxdma_start_channel: "
3125 		    "init rxdma event masks failed (0x%08x channel %d)",
3126 		    status, channel));
3127 		return (HXGE_ERROR | rs);
3128 	}
3129 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_start_channel: "
3130 	    "event done: channel %d (mask 0x%016llx)",
3131 	    channel, ent_mask.value));
3132 
3133 	/*
3134 	 * Load RXDMA descriptors, buffers, mailbox, initialise the receive DMA
3135 	 * channels and enable each DMA channel.
3136 	 */
3137 	status = hxge_enable_rxdma_channel(hxgep,
3138 	    channel, rbr_p, rcr_p, mbox_p);
3139 	if (status != HXGE_OK) {
3140 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3141 		    " hxge_rxdma_start_channel: "
3142 		    " init enable rxdma failed (0x%08x channel %d)",
3143 		    status, channel));
3144 		return (status);
3145 	}
3146 
3147 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_start_channel: "
3148 	    "control done - channel %d cs 0x%016llx", channel, cs.value));
3149 
3150 	/*
3151 	 * Initialize the receive DMA control and status register
3152 	 * Note that rdc_stat HAS to be set after RBR and RCR rings are set
3153 	 */
3154 	cs.value = 0;
3155 	cs.bits.mex = 1;
3156 	cs.bits.rcr_thres = 1;
3157 	cs.bits.rcr_to = 1;
3158 	cs.bits.rbr_empty = 1;
3159 	status = hxge_init_rxdma_channel_cntl_stat(hxgep, channel, &cs);
3160 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_start_channel: "
3161 	    "channel %d rx_dma_cntl_stat 0x%0016llx", channel, cs.value));
3162 	if (status != HXGE_OK) {
3163 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3164 		    "==> hxge_rxdma_start_channel: "
3165 		    "init rxdma control register failed (0x%08x channel %d",
3166 		    status, channel));
3167 		return (status);
3168 	}
3169 
3170 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_start_channel: "
3171 	    "control done - channel %d cs 0x%016llx", channel, cs.value));
3172 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3173 	    "==> hxge_rxdma_start_channel: enable done"));
3174 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "<== hxge_rxdma_start_channel"));
3175 
3176 	return (HXGE_OK);
3177 }
3178 
3179 static hxge_status_t
3180 hxge_rxdma_stop_channel(p_hxge_t hxgep, uint16_t channel)
3181 {
3182 	hpi_handle_t		handle;
3183 	hpi_status_t		rs = HPI_SUCCESS;
3184 	rdc_stat_t		cs;
3185 	rdc_int_mask_t		ent_mask;
3186 	hxge_status_t		status = HXGE_OK;
3187 
3188 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rxdma_stop_channel"));
3189 
3190 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
3191 
3192 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "hxge_rxdma_stop_channel: "
3193 	    "hpi handle addr $%p acc $%p",
3194 	    hxgep->hpi_handle.regp, hxgep->hpi_handle.regh));
3195 
3196 	/* Reset RXDMA channel */
3197 	rs = hpi_rxdma_cfg_rdc_reset(handle, channel);
3198 	if (rs != HPI_SUCCESS) {
3199 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3200 		    " hxge_rxdma_stop_channel: "
3201 		    " reset rxdma failed (0x%08x channel %d)",
3202 		    rs, channel));
3203 		return (HXGE_ERROR | rs);
3204 	}
3205 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
3206 	    "==> hxge_rxdma_stop_channel: reset done"));
3207 
3208 	/* Set up the interrupt event masks. */
3209 	ent_mask.value = RDC_INT_MASK_ALL;
3210 	rs = hpi_rxdma_event_mask(handle, OP_SET, channel, &ent_mask);
3211 	if (rs != HPI_SUCCESS) {
3212 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3213 		    "==> hxge_rxdma_stop_channel: "
3214 		    "set rxdma event masks failed (0x%08x channel %d)",
3215 		    rs, channel));
3216 		return (HXGE_ERROR | rs);
3217 	}
3218 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
3219 	    "==> hxge_rxdma_stop_channel: event done"));
3220 
3221 	/* Initialize the receive DMA control and status register */
3222 	cs.value = 0;
3223 	status = hxge_init_rxdma_channel_cntl_stat(hxgep, channel, &cs);
3224 
3225 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rxdma_stop_channel: control "
3226 	    " to default (all 0s) 0x%08x", cs.value));
3227 
3228 	if (status != HXGE_OK) {
3229 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3230 		    " hxge_rxdma_stop_channel: init rxdma"
3231 		    " control register failed (0x%08x channel %d",
3232 		    status, channel));
3233 		return (status);
3234 	}
3235 
3236 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
3237 	    "==> hxge_rxdma_stop_channel: control done"));
3238 
3239 	/* disable dma channel */
3240 	status = hxge_disable_rxdma_channel(hxgep, channel);
3241 
3242 	if (status != HXGE_OK) {
3243 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3244 		    " hxge_rxdma_stop_channel: "
3245 		    " init enable rxdma failed (0x%08x channel %d)",
3246 		    status, channel));
3247 		return (status);
3248 	}
3249 
3250 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
3251 	    "==> hxge_rxdma_stop_channel: disable done"));
3252 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_rxdma_stop_channel"));
3253 
3254 	return (HXGE_OK);
3255 }
3256 
3257 hxge_status_t
3258 hxge_rxdma_handle_sys_errors(p_hxge_t hxgep)
3259 {
3260 	hpi_handle_t		handle;
3261 	p_hxge_rdc_sys_stats_t	statsp;
3262 	rdc_fifo_err_stat_t	stat;
3263 	hxge_status_t		status = HXGE_OK;
3264 
3265 	handle = hxgep->hpi_handle;
3266 	statsp = (p_hxge_rdc_sys_stats_t)&hxgep->statsp->rdc_sys_stats;
3267 
3268 	/* Clear the int_dbg register in case it is an injected err */
3269 	HXGE_REG_WR64(handle, RDC_FIFO_ERR_INT_DBG, 0x0);
3270 
3271 	/* Get the error status and clear the register */
3272 	HXGE_REG_RD64(handle, RDC_FIFO_ERR_STAT, &stat.value);
3273 	HXGE_REG_WR64(handle, RDC_FIFO_ERR_STAT, stat.value);
3274 
3275 	if (stat.bits.rx_ctrl_fifo_sec) {
3276 		statsp->ctrl_fifo_sec++;
3277 		if (statsp->ctrl_fifo_sec == 1)
3278 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3279 			    "==> hxge_rxdma_handle_sys_errors: "
3280 			    "rx_ctrl_fifo_sec"));
3281 	}
3282 
3283 	if (stat.bits.rx_ctrl_fifo_ded) {
3284 		/* Global fatal error encountered */
3285 		statsp->ctrl_fifo_ded++;
3286 		HXGE_FM_REPORT_ERROR(hxgep, NULL,
3287 		    HXGE_FM_EREPORT_RDMC_CTRL_FIFO_DED);
3288 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3289 		    "==> hxge_rxdma_handle_sys_errors: "
3290 		    "fatal error: rx_ctrl_fifo_ded error"));
3291 	}
3292 
3293 	if (stat.bits.rx_data_fifo_sec) {
3294 		statsp->data_fifo_sec++;
3295 		if (statsp->data_fifo_sec == 1)
3296 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3297 			    "==> hxge_rxdma_handle_sys_errors: "
3298 			    "rx_data_fifo_sec"));
3299 	}
3300 
3301 	if (stat.bits.rx_data_fifo_ded) {
3302 		/* Global fatal error encountered */
3303 		statsp->data_fifo_ded++;
3304 		HXGE_FM_REPORT_ERROR(hxgep, NULL,
3305 		    HXGE_FM_EREPORT_RDMC_DATA_FIFO_DED);
3306 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3307 		    "==> hxge_rxdma_handle_sys_errors: "
3308 		    "fatal error: rx_data_fifo_ded error"));
3309 	}
3310 
3311 	if (stat.bits.rx_ctrl_fifo_ded || stat.bits.rx_data_fifo_ded) {
3312 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3313 		    " hxge_rxdma_handle_sys_errors: fatal error\n"));
3314 		status = hxge_rx_port_fatal_err_recover(hxgep);
3315 		if (status == HXGE_OK) {
3316 			FM_SERVICE_RESTORED(hxgep);
3317 		}
3318 	}
3319 
3320 	return (HXGE_OK);
3321 }
3322 
3323 static hxge_status_t
3324 hxge_rxdma_fatal_err_recover(p_hxge_t hxgep, uint16_t channel)
3325 {
3326 	hpi_handle_t		handle;
3327 	hpi_status_t 		rs = HPI_SUCCESS;
3328 	hxge_status_t 		status = HXGE_OK;
3329 	p_rx_rbr_ring_t		rbrp;
3330 	p_rx_rcr_ring_t		rcrp;
3331 	p_rx_mbox_t		mboxp;
3332 	rdc_int_mask_t		ent_mask;
3333 	p_hxge_dma_common_t	dmap;
3334 	int			ring_idx;
3335 	uint32_t		ref_cnt;
3336 	p_rx_msg_t		rx_msg_p;
3337 	int			i;
3338 	uint32_t		hxge_port_rcr_size;
3339 	uint64_t		tmp;
3340 
3341 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rxdma_fatal_err_recover"));
3342 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3343 	    "Recovering from RxDMAChannel#%d error...", channel));
3344 
3345 	/*
3346 	 * Stop the dma channel waits for the stop done. If the stop done bit
3347 	 * is not set, then create an error.
3348 	 */
3349 
3350 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
3351 
3352 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Rx DMA stop..."));
3353 
3354 	ring_idx = hxge_rxdma_get_ring_index(hxgep, channel);
3355 	rbrp = (p_rx_rbr_ring_t)hxgep->rx_rbr_rings->rbr_rings[ring_idx];
3356 	rcrp = (p_rx_rcr_ring_t)hxgep->rx_rcr_rings->rcr_rings[ring_idx];
3357 
3358 	MUTEX_ENTER(&rcrp->lock);
3359 	MUTEX_ENTER(&rbrp->lock);
3360 	MUTEX_ENTER(&rbrp->post_lock);
3361 
3362 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Disable RxDMA channel..."));
3363 
3364 	rs = hpi_rxdma_cfg_rdc_disable(handle, channel);
3365 	if (rs != HPI_SUCCESS) {
3366 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3367 		    "hxge_disable_rxdma_channel:failed"));
3368 		goto fail;
3369 	}
3370 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Disable RxDMA interrupt..."));
3371 
3372 	/* Disable interrupt */
3373 	ent_mask.value = RDC_INT_MASK_ALL;
3374 	rs = hpi_rxdma_event_mask(handle, OP_SET, channel, &ent_mask);
3375 	if (rs != HPI_SUCCESS) {
3376 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3377 		    "Set rxdma event masks failed (channel %d)", channel));
3378 	}
3379 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "RxDMA channel reset..."));
3380 
3381 	/* Reset RXDMA channel */
3382 	rs = hpi_rxdma_cfg_rdc_reset(handle, channel);
3383 	if (rs != HPI_SUCCESS) {
3384 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3385 		    "Reset rxdma failed (channel %d)", channel));
3386 		goto fail;
3387 	}
3388 	hxge_port_rcr_size = hxgep->hxge_port_rcr_size;
3389 	mboxp = (p_rx_mbox_t)hxgep->rx_mbox_areas_p->rxmbox_areas[ring_idx];
3390 
3391 	rbrp->rbr_wr_index = (rbrp->rbb_max - 1);
3392 	rbrp->rbr_rd_index = 0;
3393 	rbrp->pages_to_post = 0;
3394 
3395 	rcrp->comp_rd_index = 0;
3396 	rcrp->comp_wt_index = 0;
3397 	rcrp->rcr_desc_rd_head_p = rcrp->rcr_desc_first_p =
3398 	    (p_rcr_entry_t)DMA_COMMON_VPTR(rcrp->rcr_desc);
3399 #if defined(__i386)
3400 	rcrp->rcr_desc_rd_head_pp = rcrp->rcr_desc_first_pp =
3401 	    (p_rcr_entry_t)(uint32_t)DMA_COMMON_IOADDR(rcrp->rcr_desc);
3402 #else
3403 	rcrp->rcr_desc_rd_head_pp = rcrp->rcr_desc_first_pp =
3404 	    (p_rcr_entry_t)DMA_COMMON_IOADDR(rcrp->rcr_desc);
3405 #endif
3406 
3407 	rcrp->rcr_desc_last_p = rcrp->rcr_desc_rd_head_p +
3408 	    (hxge_port_rcr_size - 1);
3409 	rcrp->rcr_desc_last_pp = rcrp->rcr_desc_rd_head_pp +
3410 	    (hxge_port_rcr_size - 1);
3411 
3412 	dmap = (p_hxge_dma_common_t)&rcrp->rcr_desc;
3413 	bzero((caddr_t)dmap->kaddrp, dmap->alength);
3414 
3415 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "rbr entries = %d\n",
3416 	    rbrp->rbr_max_size));
3417 
3418 	for (i = 0; i < rbrp->rbr_max_size; i++) {
3419 		/* Reset all the buffers */
3420 		rx_msg_p = rbrp->rx_msg_ring[i];
3421 		ref_cnt = rx_msg_p->ref_cnt;
3422 
3423 		rx_msg_p->ref_cnt = 1;
3424 		rx_msg_p->free = B_TRUE;
3425 		rx_msg_p->cur_usage_cnt = 0;
3426 		rx_msg_p->max_usage_cnt = 0;
3427 		rx_msg_p->pkt_buf_size = 0;
3428 
3429 		if (ref_cnt > 1)
3430 			atomic_add_32(&hxge_mblks_pending, 1 - ref_cnt);
3431 	}
3432 
3433 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "RxDMA channel re-start..."));
3434 
3435 	status = hxge_rxdma_start_channel(hxgep, channel, rbrp, rcrp, mboxp);
3436 	if (status != HXGE_OK) {
3437 		goto fail;
3438 	}
3439 
3440 	/*
3441 	 * The DMA channel may disable itself automatically.
3442 	 * The following is a work-around.
3443 	 */
3444 	HXGE_REG_RD64(handle, RDC_RX_CFG1, &tmp);
3445 	rs = hpi_rxdma_cfg_rdc_enable(handle, channel);
3446 	if (rs != HPI_SUCCESS) {
3447 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3448 		    "hpi_rxdma_cfg_rdc_enable (channel %d)", channel));
3449 	}
3450 
3451 	MUTEX_EXIT(&rbrp->post_lock);
3452 	MUTEX_EXIT(&rbrp->lock);
3453 	MUTEX_EXIT(&rcrp->lock);
3454 
3455 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3456 	    "Recovery Successful, RxDMAChannel#%d Restored", channel));
3457 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_rxdma_fatal_err_recover"));
3458 
3459 	return (HXGE_OK);
3460 
3461 fail:
3462 	MUTEX_EXIT(&rbrp->post_lock);
3463 	MUTEX_EXIT(&rbrp->lock);
3464 	MUTEX_EXIT(&rcrp->lock);
3465 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL, "Recovery failed"));
3466 
3467 	return (HXGE_ERROR | rs);
3468 }
3469 
3470 static hxge_status_t
3471 hxge_rx_port_fatal_err_recover(p_hxge_t hxgep)
3472 {
3473 	hxge_status_t		status = HXGE_OK;
3474 	p_hxge_dma_common_t	*dma_buf_p;
3475 	uint16_t		channel;
3476 	int			ndmas;
3477 	int			i;
3478 	block_reset_t		reset_reg;
3479 
3480 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rx_port_fatal_err_recover"));
3481 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL, "Recovering from RDC error ..."));
3482 
3483 	/* Reset RDC block from PEU for this fatal error */
3484 	reset_reg.value = 0;
3485 	reset_reg.bits.rdc_rst = 1;
3486 	HXGE_REG_WR32(hxgep->hpi_handle, BLOCK_RESET, reset_reg.value);
3487 
3488 	/* Disable RxMAC */
3489 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Disable RxMAC...\n"));
3490 	if (hxge_rx_vmac_disable(hxgep) != HXGE_OK)
3491 		goto fail;
3492 
3493 	HXGE_DELAY(1000);
3494 
3495 	/* Restore any common settings after PEU reset */
3496 	if (hxge_rxdma_hw_start_common(hxgep) != HXGE_OK)
3497 		goto fail;
3498 
3499 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Stop all RxDMA channels..."));
3500 
3501 	ndmas = hxgep->rx_buf_pool_p->ndmas;
3502 	dma_buf_p = hxgep->rx_buf_pool_p->dma_buf_pool_p;
3503 
3504 	for (i = 0; i < ndmas; i++) {
3505 		channel = ((p_hxge_dma_common_t)dma_buf_p[i])->dma_channel;
3506 		if (hxge_rxdma_fatal_err_recover(hxgep, channel) != HXGE_OK) {
3507 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3508 			    "Could not recover channel %d", channel));
3509 		}
3510 	}
3511 
3512 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Reset RxMAC..."));
3513 
3514 	/* Reset RxMAC */
3515 	if (hxge_rx_vmac_reset(hxgep) != HXGE_OK) {
3516 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3517 		    "hxge_rx_port_fatal_err_recover: Failed to reset RxMAC"));
3518 		goto fail;
3519 	}
3520 
3521 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Re-initialize RxMAC..."));
3522 
3523 	/* Re-Initialize RxMAC */
3524 	if ((status = hxge_rx_vmac_init(hxgep)) != HXGE_OK) {
3525 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3526 		    "hxge_rx_port_fatal_err_recover: Failed to reset RxMAC"));
3527 		goto fail;
3528 	}
3529 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Re-enable RxMAC..."));
3530 
3531 	/* Re-enable RxMAC */
3532 	if ((status = hxge_rx_vmac_enable(hxgep)) != HXGE_OK) {
3533 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3534 		    "hxge_rx_port_fatal_err_recover: Failed to enable RxMAC"));
3535 		goto fail;
3536 	}
3537 
3538 	/* Reset the error mask since PEU reset cleared it */
3539 	HXGE_REG_WR64(hxgep->hpi_handle, RDC_FIFO_ERR_INT_MASK, 0x0);
3540 
3541 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3542 	    "Recovery Successful, RxPort Restored"));
3543 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_rx_port_fatal_err_recover"));
3544 
3545 	return (HXGE_OK);
3546 fail:
3547 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL, "Recovery failed"));
3548 	return (status);
3549 }
3550