xref: /titanic_41/usr/src/uts/common/io/hxge/hxge_rxdma.c (revision 3afe87ebb25691cb6d158edaa34a6fb9b703a691)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #include <hxge_impl.h>
27 #include <hxge_rxdma.h>
28 
29 /*
30  * Number of blocks to accumulate before re-enabling DMA
31  * when we get RBR empty.
32  */
33 #define	HXGE_RBR_EMPTY_THRESHOLD	64
34 
35 /*
36  * Globals: tunable parameters (/etc/system or adb)
37  *
38  */
39 extern uint32_t hxge_rbr_size;
40 extern uint32_t hxge_rcr_size;
41 extern uint32_t hxge_rbr_spare_size;
42 extern uint32_t hxge_mblks_pending;
43 
44 /*
45  * Tunable to reduce the amount of time spent in the
46  * ISR doing Rx Processing.
47  */
48 extern uint32_t hxge_max_rx_pkts;
49 
50 /*
51  * Tunables to manage the receive buffer blocks.
52  *
53  * hxge_rx_threshold_hi: copy all buffers.
54  * hxge_rx_bcopy_size_type: receive buffer block size type.
55  * hxge_rx_threshold_lo: copy only up to tunable block size type.
56  */
57 extern hxge_rxbuf_threshold_t hxge_rx_threshold_hi;
58 extern hxge_rxbuf_type_t hxge_rx_buf_size_type;
59 extern hxge_rxbuf_threshold_t hxge_rx_threshold_lo;
60 
61 /*
62  * Static local functions.
63  */
64 static hxge_status_t hxge_map_rxdma(p_hxge_t hxgep);
65 static void hxge_unmap_rxdma(p_hxge_t hxgep);
66 static hxge_status_t hxge_rxdma_hw_start_common(p_hxge_t hxgep);
67 static hxge_status_t hxge_rxdma_hw_start(p_hxge_t hxgep);
68 static void hxge_rxdma_hw_stop(p_hxge_t hxgep);
69 static hxge_status_t hxge_map_rxdma_channel(p_hxge_t hxgep, uint16_t channel,
70     p_hxge_dma_common_t *dma_buf_p, p_rx_rbr_ring_t *rbr_p,
71     uint32_t num_chunks, p_hxge_dma_common_t *dma_rbr_cntl_p,
72     p_hxge_dma_common_t *dma_rcr_cntl_p, p_hxge_dma_common_t *dma_mbox_cntl_p,
73     p_rx_rcr_ring_t *rcr_p, p_rx_mbox_t *rx_mbox_p);
74 static void hxge_unmap_rxdma_channel(p_hxge_t hxgep, uint16_t channel,
75 	p_rx_rbr_ring_t rbr_p, p_rx_rcr_ring_t rcr_p, p_rx_mbox_t rx_mbox_p);
76 static hxge_status_t hxge_map_rxdma_channel_cfg_ring(p_hxge_t hxgep,
77     uint16_t dma_channel, p_hxge_dma_common_t *dma_rbr_cntl_p,
78     p_hxge_dma_common_t *dma_rcr_cntl_p, p_hxge_dma_common_t *dma_mbox_cntl_p,
79     p_rx_rbr_ring_t *rbr_p, p_rx_rcr_ring_t *rcr_p, p_rx_mbox_t *rx_mbox_p);
80 static void hxge_unmap_rxdma_channel_cfg_ring(p_hxge_t hxgep,
81 	p_rx_rcr_ring_t rcr_p, p_rx_mbox_t rx_mbox_p);
82 static hxge_status_t hxge_map_rxdma_channel_buf_ring(p_hxge_t hxgep,
83 	uint16_t channel, p_hxge_dma_common_t *dma_buf_p,
84 	p_rx_rbr_ring_t *rbr_p, uint32_t num_chunks);
85 static void hxge_unmap_rxdma_channel_buf_ring(p_hxge_t hxgep,
86 	p_rx_rbr_ring_t rbr_p);
87 static hxge_status_t hxge_rxdma_start_channel(p_hxge_t hxgep, uint16_t channel,
88 	p_rx_rbr_ring_t rbr_p, p_rx_rcr_ring_t rcr_p, p_rx_mbox_t mbox_p,
89 	int n_init_kick);
90 static hxge_status_t hxge_rxdma_stop_channel(p_hxge_t hxgep, uint16_t channel);
91 static mblk_t *hxge_rx_pkts(p_hxge_t hxgep, uint_t vindex, p_hxge_ldv_t ldvp,
92 	p_rx_rcr_ring_t	*rcr_p, rdc_stat_t cs);
93 static uint32_t hxge_scan_for_last_eop(p_rx_rcr_ring_t rcr_p,
94     p_rcr_entry_t rcr_desc_rd_head_p, uint32_t num_rcrs);
95 static void hxge_receive_packet(p_hxge_t hxgep, p_rx_rcr_ring_t rcr_p,
96 	p_rcr_entry_t rcr_desc_rd_head_p, boolean_t *multi_p,
97 	mblk_t ** mp, mblk_t ** mp_cont, uint32_t *invalid_rcr_entry);
98 static hxge_status_t hxge_disable_rxdma_channel(p_hxge_t hxgep,
99 	uint16_t channel);
100 static p_rx_msg_t hxge_allocb(size_t, uint32_t, p_hxge_dma_common_t);
101 static void hxge_freeb(p_rx_msg_t);
102 static void hxge_rx_pkts_vring(p_hxge_t hxgep, uint_t vindex,
103 	p_hxge_ldv_t ldvp, rdc_stat_t cs);
104 static hxge_status_t hxge_rx_err_evnts(p_hxge_t hxgep, uint_t index,
105 	p_hxge_ldv_t ldvp, rdc_stat_t cs);
106 static hxge_status_t hxge_rxbuf_index_info_init(p_hxge_t hxgep,
107 	p_rx_rbr_ring_t rx_dmap);
108 static hxge_status_t hxge_rxdma_fatal_err_recover(p_hxge_t hxgep,
109 	uint16_t channel);
110 static hxge_status_t hxge_rx_port_fatal_err_recover(p_hxge_t hxgep);
111 static void hxge_rbr_empty_restore(p_hxge_t hxgep,
112 	p_rx_rbr_ring_t rx_rbr_p);
113 
114 hxge_status_t
115 hxge_init_rxdma_channels(p_hxge_t hxgep)
116 {
117 	hxge_status_t		status = HXGE_OK;
118 	block_reset_t		reset_reg;
119 	int			i;
120 
121 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_init_rxdma_channels"));
122 
123 	for (i = 0; i < HXGE_MAX_RDCS; i++)
124 		hxgep->rdc_first_intr[i] = B_TRUE;
125 
126 	/* Reset RDC block from PEU to clear any previous state */
127 	reset_reg.value = 0;
128 	reset_reg.bits.rdc_rst = 1;
129 	HXGE_REG_WR32(hxgep->hpi_handle, BLOCK_RESET, reset_reg.value);
130 	HXGE_DELAY(1000);
131 
132 	status = hxge_map_rxdma(hxgep);
133 	if (status != HXGE_OK) {
134 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
135 		    "<== hxge_init_rxdma: status 0x%x", status));
136 		return (status);
137 	}
138 
139 	status = hxge_rxdma_hw_start_common(hxgep);
140 	if (status != HXGE_OK) {
141 		hxge_unmap_rxdma(hxgep);
142 	}
143 
144 	status = hxge_rxdma_hw_start(hxgep);
145 	if (status != HXGE_OK) {
146 		hxge_unmap_rxdma(hxgep);
147 	}
148 
149 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
150 	    "<== hxge_init_rxdma_channels: status 0x%x", status));
151 	return (status);
152 }
153 
154 void
155 hxge_uninit_rxdma_channels(p_hxge_t hxgep)
156 {
157 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_uninit_rxdma_channels"));
158 
159 	hxge_rxdma_hw_stop(hxgep);
160 	hxge_unmap_rxdma(hxgep);
161 
162 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "<== hxge_uinit_rxdma_channels"));
163 }
164 
165 hxge_status_t
166 hxge_init_rxdma_channel_cntl_stat(p_hxge_t hxgep, uint16_t channel,
167     rdc_stat_t *cs_p)
168 {
169 	hpi_handle_t	handle;
170 	hpi_status_t	rs = HPI_SUCCESS;
171 	hxge_status_t	status = HXGE_OK;
172 
173 	HXGE_DEBUG_MSG((hxgep, DMA_CTL,
174 	    "<== hxge_init_rxdma_channel_cntl_stat"));
175 
176 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
177 	rs = hpi_rxdma_control_status(handle, OP_SET, channel, cs_p);
178 
179 	if (rs != HPI_SUCCESS) {
180 		status = HXGE_ERROR | rs;
181 	}
182 	return (status);
183 }
184 
185 
186 hxge_status_t
187 hxge_enable_rxdma_channel(p_hxge_t hxgep, uint16_t channel,
188     p_rx_rbr_ring_t rbr_p, p_rx_rcr_ring_t rcr_p, p_rx_mbox_t mbox_p,
189     int n_init_kick)
190 {
191 	hpi_handle_t		handle;
192 	rdc_desc_cfg_t 		rdc_desc;
193 	rdc_rcr_cfg_b_t		*cfgb_p;
194 	hpi_status_t		rs = HPI_SUCCESS;
195 
196 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "==> hxge_enable_rxdma_channel"));
197 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
198 
199 	/*
200 	 * Use configuration data composed at init time. Write to hardware the
201 	 * receive ring configurations.
202 	 */
203 	rdc_desc.mbox_enable = 1;
204 	rdc_desc.mbox_addr = mbox_p->mbox_addr;
205 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
206 	    "==> hxge_enable_rxdma_channel: mboxp $%p($%p)",
207 	    mbox_p->mbox_addr, rdc_desc.mbox_addr));
208 
209 	rdc_desc.rbr_len = rbr_p->rbb_max;
210 	rdc_desc.rbr_addr = rbr_p->rbr_addr;
211 
212 	switch (hxgep->rx_bksize_code) {
213 	case RBR_BKSIZE_4K:
214 		rdc_desc.page_size = SIZE_4KB;
215 		break;
216 	case RBR_BKSIZE_8K:
217 		rdc_desc.page_size = SIZE_8KB;
218 		break;
219 	}
220 
221 	rdc_desc.size0 = rbr_p->hpi_pkt_buf_size0;
222 	rdc_desc.valid0 = 1;
223 
224 	rdc_desc.size1 = rbr_p->hpi_pkt_buf_size1;
225 	rdc_desc.valid1 = 1;
226 
227 	rdc_desc.size2 = rbr_p->hpi_pkt_buf_size2;
228 	rdc_desc.valid2 = 1;
229 
230 	rdc_desc.full_hdr = rcr_p->full_hdr_flag;
231 	rdc_desc.offset = rcr_p->sw_priv_hdr_len;
232 
233 	rdc_desc.rcr_len = rcr_p->comp_size;
234 	rdc_desc.rcr_addr = rcr_p->rcr_addr;
235 
236 	cfgb_p = &(rcr_p->rcr_cfgb);
237 	rdc_desc.rcr_threshold = cfgb_p->bits.pthres;
238 	rdc_desc.rcr_timeout = cfgb_p->bits.timeout;
239 	rdc_desc.rcr_timeout_enable = cfgb_p->bits.entout;
240 
241 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "==> hxge_enable_rxdma_channel: "
242 	    "rbr_len qlen %d pagesize code %d rcr_len %d",
243 	    rdc_desc.rbr_len, rdc_desc.page_size, rdc_desc.rcr_len));
244 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "==> hxge_enable_rxdma_channel: "
245 	    "size 0 %d size 1 %d size 2 %d",
246 	    rbr_p->hpi_pkt_buf_size0, rbr_p->hpi_pkt_buf_size1,
247 	    rbr_p->hpi_pkt_buf_size2));
248 
249 	rs = hpi_rxdma_cfg_rdc_ring(handle, rbr_p->rdc, &rdc_desc);
250 	if (rs != HPI_SUCCESS) {
251 		return (HXGE_ERROR | rs);
252 	}
253 
254 	/*
255 	 * Enable the timeout and threshold.
256 	 */
257 	rs = hpi_rxdma_cfg_rdc_rcr_threshold(handle, channel,
258 	    rdc_desc.rcr_threshold);
259 	if (rs != HPI_SUCCESS) {
260 		return (HXGE_ERROR | rs);
261 	}
262 
263 	rs = hpi_rxdma_cfg_rdc_rcr_timeout(handle, channel,
264 	    rdc_desc.rcr_timeout);
265 	if (rs != HPI_SUCCESS) {
266 		return (HXGE_ERROR | rs);
267 	}
268 
269 	/* Enable the DMA */
270 	rs = hpi_rxdma_cfg_rdc_enable(handle, channel);
271 	if (rs != HPI_SUCCESS) {
272 		return (HXGE_ERROR | rs);
273 	}
274 
275 	/* Kick the DMA engine */
276 	hpi_rxdma_rdc_rbr_kick(handle, channel, n_init_kick);
277 
278 	/* Clear the rbr empty bit */
279 	(void) hpi_rxdma_channel_rbr_empty_clear(handle, channel);
280 
281 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "<== hxge_enable_rxdma_channel"));
282 
283 	return (HXGE_OK);
284 }
285 
286 static hxge_status_t
287 hxge_disable_rxdma_channel(p_hxge_t hxgep, uint16_t channel)
288 {
289 	hpi_handle_t handle;
290 	hpi_status_t rs = HPI_SUCCESS;
291 
292 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "==> hxge_disable_rxdma_channel"));
293 
294 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
295 
296 	/* disable the DMA */
297 	rs = hpi_rxdma_cfg_rdc_disable(handle, channel);
298 	if (rs != HPI_SUCCESS) {
299 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
300 		    "<== hxge_disable_rxdma_channel:failed (0x%x)", rs));
301 		return (HXGE_ERROR | rs);
302 	}
303 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "<== hxge_disable_rxdma_channel"));
304 	return (HXGE_OK);
305 }
306 
307 hxge_status_t
308 hxge_rxdma_channel_rcrflush(p_hxge_t hxgep, uint8_t channel)
309 {
310 	hpi_handle_t	handle;
311 	hxge_status_t	status = HXGE_OK;
312 
313 	HXGE_DEBUG_MSG((hxgep, DMA_CTL,
314 	    "==> hxge_rxdma_channel_rcrflush"));
315 
316 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
317 	hpi_rxdma_rdc_rcr_flush(handle, channel);
318 
319 	HXGE_DEBUG_MSG((hxgep, DMA_CTL,
320 	    "<== hxge_rxdma_channel_rcrflush"));
321 	return (status);
322 
323 }
324 
325 #define	MID_INDEX(l, r) ((r + l + 1) >> 1)
326 
327 #define	TO_LEFT -1
328 #define	TO_RIGHT 1
329 #define	BOTH_RIGHT (TO_RIGHT + TO_RIGHT)
330 #define	BOTH_LEFT (TO_LEFT + TO_LEFT)
331 #define	IN_MIDDLE (TO_RIGHT + TO_LEFT)
332 #define	NO_HINT 0xffffffff
333 
334 /*ARGSUSED*/
335 hxge_status_t
336 hxge_rxbuf_pp_to_vp(p_hxge_t hxgep, p_rx_rbr_ring_t rbr_p,
337     uint8_t pktbufsz_type, uint64_t *pkt_buf_addr_pp,
338     uint64_t **pkt_buf_addr_p, uint32_t *bufoffset, uint32_t *msg_index)
339 {
340 	int			bufsize;
341 	uint64_t		pktbuf_pp;
342 	uint64_t		dvma_addr;
343 	rxring_info_t		*ring_info;
344 	int			base_side, end_side;
345 	int			r_index, l_index, anchor_index;
346 	int			found, search_done;
347 	uint32_t		offset, chunk_size, block_size, page_size_mask;
348 	uint32_t		chunk_index, block_index, total_index;
349 	int			max_iterations, iteration;
350 	rxbuf_index_info_t	*bufinfo;
351 
352 	HXGE_DEBUG_MSG((hxgep, RX2_CTL, "==> hxge_rxbuf_pp_to_vp"));
353 
354 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
355 	    "==> hxge_rxbuf_pp_to_vp: buf_pp $%p btype %d",
356 	    pkt_buf_addr_pp, pktbufsz_type));
357 
358 #if defined(__i386)
359 	pktbuf_pp = (uint64_t)(uint32_t)pkt_buf_addr_pp;
360 #else
361 	pktbuf_pp = (uint64_t)pkt_buf_addr_pp;
362 #endif
363 
364 	switch (pktbufsz_type) {
365 	case 0:
366 		bufsize = rbr_p->pkt_buf_size0;
367 		break;
368 	case 1:
369 		bufsize = rbr_p->pkt_buf_size1;
370 		break;
371 	case 2:
372 		bufsize = rbr_p->pkt_buf_size2;
373 		break;
374 	case RCR_SINGLE_BLOCK:
375 		bufsize = 0;
376 		anchor_index = 0;
377 		break;
378 	default:
379 		return (HXGE_ERROR);
380 	}
381 
382 	if (rbr_p->num_blocks == 1) {
383 		anchor_index = 0;
384 		ring_info = rbr_p->ring_info;
385 		bufinfo = (rxbuf_index_info_t *)ring_info->buffer;
386 
387 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
388 		    "==> hxge_rxbuf_pp_to_vp: (found, 1 block) "
389 		    "buf_pp $%p btype %d anchor_index %d bufinfo $%p",
390 		    pkt_buf_addr_pp, pktbufsz_type, anchor_index, bufinfo));
391 
392 		goto found_index;
393 	}
394 
395 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
396 	    "==> hxge_rxbuf_pp_to_vp: buf_pp $%p btype %d anchor_index %d",
397 	    pkt_buf_addr_pp, pktbufsz_type, anchor_index));
398 
399 	ring_info = rbr_p->ring_info;
400 	found = B_FALSE;
401 	bufinfo = (rxbuf_index_info_t *)ring_info->buffer;
402 	iteration = 0;
403 	max_iterations = ring_info->max_iterations;
404 
405 	/*
406 	 * First check if this block have been seen recently. This is indicated
407 	 * by a hint which is initialized when the first buffer of the block is
408 	 * seen. The hint is reset when the last buffer of the block has been
409 	 * processed. As three block sizes are supported, three hints are kept.
410 	 * The idea behind the hints is that once the hardware  uses a block
411 	 * for a buffer  of that size, it will use it exclusively for that size
412 	 * and will use it until it is exhausted. It is assumed that there
413 	 * would a single block being used for the same buffer sizes at any
414 	 * given time.
415 	 */
416 	if (ring_info->hint[pktbufsz_type] != NO_HINT) {
417 		anchor_index = ring_info->hint[pktbufsz_type];
418 		dvma_addr = bufinfo[anchor_index].dvma_addr;
419 		chunk_size = bufinfo[anchor_index].buf_size;
420 		if ((pktbuf_pp >= dvma_addr) &&
421 		    (pktbuf_pp < (dvma_addr + chunk_size))) {
422 			found = B_TRUE;
423 			/*
424 			 * check if this is the last buffer in the block If so,
425 			 * then reset the hint for the size;
426 			 */
427 
428 			if ((pktbuf_pp + bufsize) >= (dvma_addr + chunk_size))
429 				ring_info->hint[pktbufsz_type] = NO_HINT;
430 		}
431 	}
432 
433 	if (found == B_FALSE) {
434 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
435 		    "==> hxge_rxbuf_pp_to_vp: (!found)"
436 		    "buf_pp $%p btype %d anchor_index %d",
437 		    pkt_buf_addr_pp, pktbufsz_type, anchor_index));
438 
439 		/*
440 		 * This is the first buffer of the block of this size. Need to
441 		 * search the whole information array. the search algorithm
442 		 * uses a binary tree search algorithm. It assumes that the
443 		 * information is already sorted with increasing order info[0]
444 		 * < info[1] < info[2]  .... < info[n-1] where n is the size of
445 		 * the information array
446 		 */
447 		r_index = rbr_p->num_blocks - 1;
448 		l_index = 0;
449 		search_done = B_FALSE;
450 		anchor_index = MID_INDEX(r_index, l_index);
451 		while (search_done == B_FALSE) {
452 			if ((r_index == l_index) ||
453 			    (iteration >= max_iterations))
454 				search_done = B_TRUE;
455 
456 			end_side = TO_RIGHT;	/* to the right */
457 			base_side = TO_LEFT;	/* to the left */
458 			/* read the DVMA address information and sort it */
459 			dvma_addr = bufinfo[anchor_index].dvma_addr;
460 			chunk_size = bufinfo[anchor_index].buf_size;
461 
462 			HXGE_DEBUG_MSG((hxgep, RX2_CTL,
463 			    "==> hxge_rxbuf_pp_to_vp: (searching)"
464 			    "buf_pp $%p btype %d "
465 			    "anchor_index %d chunk_size %d dvmaaddr $%p",
466 			    pkt_buf_addr_pp, pktbufsz_type, anchor_index,
467 			    chunk_size, dvma_addr));
468 
469 			if (pktbuf_pp >= dvma_addr)
470 				base_side = TO_RIGHT;	/* to the right */
471 			if (pktbuf_pp < (dvma_addr + chunk_size))
472 				end_side = TO_LEFT;	/* to the left */
473 
474 			switch (base_side + end_side) {
475 			case IN_MIDDLE:
476 				/* found */
477 				found = B_TRUE;
478 				search_done = B_TRUE;
479 				if ((pktbuf_pp + bufsize) <
480 				    (dvma_addr + chunk_size))
481 					ring_info->hint[pktbufsz_type] =
482 					    bufinfo[anchor_index].buf_index;
483 				break;
484 			case BOTH_RIGHT:
485 				/* not found: go to the right */
486 				l_index = anchor_index + 1;
487 				anchor_index = MID_INDEX(r_index, l_index);
488 				break;
489 
490 			case BOTH_LEFT:
491 				/* not found: go to the left */
492 				r_index = anchor_index - 1;
493 				anchor_index = MID_INDEX(r_index, l_index);
494 				break;
495 			default:	/* should not come here */
496 				return (HXGE_ERROR);
497 			}
498 			iteration++;
499 		}
500 
501 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
502 		    "==> hxge_rxbuf_pp_to_vp: (search done)"
503 		    "buf_pp $%p btype %d anchor_index %d",
504 		    pkt_buf_addr_pp, pktbufsz_type, anchor_index));
505 	}
506 
507 	if (found == B_FALSE) {
508 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
509 		    "==> hxge_rxbuf_pp_to_vp: (search failed)"
510 		    "buf_pp $%p btype %d anchor_index %d",
511 		    pkt_buf_addr_pp, pktbufsz_type, anchor_index));
512 		return (HXGE_ERROR);
513 	}
514 
515 found_index:
516 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
517 	    "==> hxge_rxbuf_pp_to_vp: (FOUND1)"
518 	    "buf_pp $%p btype %d bufsize %d anchor_index %d",
519 	    pkt_buf_addr_pp, pktbufsz_type, bufsize, anchor_index));
520 
521 	/* index of the first block in this chunk */
522 	chunk_index = bufinfo[anchor_index].start_index;
523 	dvma_addr = bufinfo[anchor_index].dvma_addr;
524 	page_size_mask = ring_info->block_size_mask;
525 
526 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
527 	    "==> hxge_rxbuf_pp_to_vp: (FOUND3), get chunk)"
528 	    "buf_pp $%p btype %d bufsize %d "
529 	    "anchor_index %d chunk_index %d dvma $%p",
530 	    pkt_buf_addr_pp, pktbufsz_type, bufsize,
531 	    anchor_index, chunk_index, dvma_addr));
532 
533 	offset = pktbuf_pp - dvma_addr;	/* offset within the chunk */
534 	block_size = rbr_p->block_size;	/* System  block(page) size */
535 
536 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
537 	    "==> hxge_rxbuf_pp_to_vp: (FOUND4), get chunk)"
538 	    "buf_pp $%p btype %d bufsize %d "
539 	    "anchor_index %d chunk_index %d dvma $%p "
540 	    "offset %d block_size %d",
541 	    pkt_buf_addr_pp, pktbufsz_type, bufsize, anchor_index,
542 	    chunk_index, dvma_addr, offset, block_size));
543 	HXGE_DEBUG_MSG((hxgep, RX2_CTL, "==> getting total index"));
544 
545 	block_index = (offset / block_size);	/* index within chunk */
546 	total_index = chunk_index + block_index;
547 
548 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
549 	    "==> hxge_rxbuf_pp_to_vp: "
550 	    "total_index %d dvma_addr $%p "
551 	    "offset %d block_size %d "
552 	    "block_index %d ",
553 	    total_index, dvma_addr, offset, block_size, block_index));
554 
555 #if defined(__i386)
556 	*pkt_buf_addr_p = (uint64_t *)((uint32_t)bufinfo[anchor_index].kaddr +
557 	    (uint32_t)offset);
558 #else
559 	*pkt_buf_addr_p = (uint64_t *)((uint64_t)bufinfo[anchor_index].kaddr +
560 	    offset);
561 #endif
562 
563 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
564 	    "==> hxge_rxbuf_pp_to_vp: "
565 	    "total_index %d dvma_addr $%p "
566 	    "offset %d block_size %d "
567 	    "block_index %d "
568 	    "*pkt_buf_addr_p $%p",
569 	    total_index, dvma_addr, offset, block_size,
570 	    block_index, *pkt_buf_addr_p));
571 
572 	*msg_index = total_index;
573 	*bufoffset = (offset & page_size_mask);
574 
575 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
576 	    "==> hxge_rxbuf_pp_to_vp: get msg index: "
577 	    "msg_index %d bufoffset_index %d",
578 	    *msg_index, *bufoffset));
579 	HXGE_DEBUG_MSG((hxgep, RX2_CTL, "<== hxge_rxbuf_pp_to_vp"));
580 
581 	return (HXGE_OK);
582 }
583 
584 
585 /*
586  * used by quick sort (qsort) function
587  * to perform comparison
588  */
589 static int
590 hxge_sort_compare(const void *p1, const void *p2)
591 {
592 
593 	rxbuf_index_info_t *a, *b;
594 
595 	a = (rxbuf_index_info_t *)p1;
596 	b = (rxbuf_index_info_t *)p2;
597 
598 	if (a->dvma_addr > b->dvma_addr)
599 		return (1);
600 	if (a->dvma_addr < b->dvma_addr)
601 		return (-1);
602 	return (0);
603 }
604 
605 /*
606  * Grabbed this sort implementation from common/syscall/avl.c
607  *
608  * Generic shellsort, from K&R (1st ed, p 58.), somewhat modified.
609  * v = Ptr to array/vector of objs
610  * n = # objs in the array
611  * s = size of each obj (must be multiples of a word size)
612  * f = ptr to function to compare two objs
613  *	returns (-1 = less than, 0 = equal, 1 = greater than
614  */
615 void
616 hxge_ksort(caddr_t v, int n, int s, int (*f) ())
617 {
618 	int		g, i, j, ii;
619 	unsigned int	*p1, *p2;
620 	unsigned int	tmp;
621 
622 	/* No work to do */
623 	if (v == NULL || n <= 1)
624 		return;
625 	/* Sanity check on arguments */
626 	ASSERT(((uintptr_t)v & 0x3) == 0 && (s & 0x3) == 0);
627 	ASSERT(s > 0);
628 
629 	for (g = n / 2; g > 0; g /= 2) {
630 		for (i = g; i < n; i++) {
631 			for (j = i - g; j >= 0 &&
632 			    (*f) (v + j * s, v + (j + g) * s) == 1; j -= g) {
633 				p1 = (unsigned *)(v + j * s);
634 				p2 = (unsigned *)(v + (j + g) * s);
635 				for (ii = 0; ii < s / 4; ii++) {
636 					tmp = *p1;
637 					*p1++ = *p2;
638 					*p2++ = tmp;
639 				}
640 			}
641 		}
642 	}
643 }
644 
645 /*
646  * Initialize data structures required for rxdma
647  * buffer dvma->vmem address lookup
648  */
649 /*ARGSUSED*/
650 static hxge_status_t
651 hxge_rxbuf_index_info_init(p_hxge_t hxgep, p_rx_rbr_ring_t rbrp)
652 {
653 	int		index;
654 	rxring_info_t	*ring_info;
655 	int		max_iteration = 0, max_index = 0;
656 
657 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "==> hxge_rxbuf_index_info_init"));
658 
659 	ring_info = rbrp->ring_info;
660 	ring_info->hint[0] = NO_HINT;
661 	ring_info->hint[1] = NO_HINT;
662 	ring_info->hint[2] = NO_HINT;
663 	max_index = rbrp->num_blocks;
664 
665 	/* read the DVMA address information and sort it */
666 	/* do init of the information array */
667 
668 	HXGE_DEBUG_MSG((hxgep, DMA2_CTL,
669 	    " hxge_rxbuf_index_info_init Sort ptrs"));
670 
671 	/* sort the array */
672 	hxge_ksort((void *) ring_info->buffer, max_index,
673 	    sizeof (rxbuf_index_info_t), hxge_sort_compare);
674 
675 	for (index = 0; index < max_index; index++) {
676 		HXGE_DEBUG_MSG((hxgep, DMA2_CTL,
677 		    " hxge_rxbuf_index_info_init: sorted chunk %d "
678 		    " ioaddr $%p kaddr $%p size %x",
679 		    index, ring_info->buffer[index].dvma_addr,
680 		    ring_info->buffer[index].kaddr,
681 		    ring_info->buffer[index].buf_size));
682 	}
683 
684 	max_iteration = 0;
685 	while (max_index >= (1ULL << max_iteration))
686 		max_iteration++;
687 	ring_info->max_iterations = max_iteration + 1;
688 
689 	HXGE_DEBUG_MSG((hxgep, DMA2_CTL,
690 	    " hxge_rxbuf_index_info_init Find max iter %d",
691 	    ring_info->max_iterations));
692 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "<== hxge_rxbuf_index_info_init"));
693 
694 	return (HXGE_OK);
695 }
696 
697 /*ARGSUSED*/
698 void
699 hxge_dump_rcr_entry(p_hxge_t hxgep, p_rcr_entry_t entry_p)
700 {
701 #ifdef	HXGE_DEBUG
702 
703 	uint32_t bptr;
704 	uint64_t pp;
705 
706 	bptr = entry_p->bits.pkt_buf_addr;
707 
708 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
709 	    "\trcr entry $%p "
710 	    "\trcr entry 0x%0llx "
711 	    "\trcr entry 0x%08x "
712 	    "\trcr entry 0x%08x "
713 	    "\tvalue 0x%0llx\n"
714 	    "\tmulti = %d\n"
715 	    "\tpkt_type = 0x%x\n"
716 	    "\terror = 0x%04x\n"
717 	    "\tl2_len = %d\n"
718 	    "\tpktbufsize = %d\n"
719 	    "\tpkt_buf_addr = $%p\n"
720 	    "\tpkt_buf_addr (<< 6) = $%p\n",
721 	    entry_p,
722 	    *(int64_t *)entry_p,
723 	    *(int32_t *)entry_p,
724 	    *(int32_t *)((char *)entry_p + 32),
725 	    entry_p->value,
726 	    entry_p->bits.multi,
727 	    entry_p->bits.pkt_type,
728 	    entry_p->bits.error,
729 	    entry_p->bits.l2_len,
730 	    entry_p->bits.pktbufsz,
731 	    bptr,
732 	    entry_p->bits.pkt_buf_addr_l));
733 
734 	pp = (entry_p->value & RCR_PKT_BUF_ADDR_MASK) <<
735 	    RCR_PKT_BUF_ADDR_SHIFT;
736 
737 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "rcr pp 0x%llx l2 len %d",
738 	    pp, (*(int64_t *)entry_p >> 40) & 0x3fff));
739 #endif
740 }
741 
742 /*ARGSUSED*/
743 void
744 hxge_rxdma_stop(p_hxge_t hxgep)
745 {
746 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rxdma_stop"));
747 
748 	(void) hxge_rx_vmac_disable(hxgep);
749 	(void) hxge_rxdma_hw_mode(hxgep, HXGE_DMA_STOP);
750 
751 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_rxdma_stop"));
752 }
753 
754 void
755 hxge_rxdma_stop_reinit(p_hxge_t hxgep)
756 {
757 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rxdma_stop_reinit"));
758 
759 	(void) hxge_rxdma_stop(hxgep);
760 	(void) hxge_uninit_rxdma_channels(hxgep);
761 	(void) hxge_init_rxdma_channels(hxgep);
762 
763 	(void) hxge_rx_vmac_enable(hxgep);
764 
765 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_rxdma_stop_reinit"));
766 }
767 
768 hxge_status_t
769 hxge_rxdma_hw_mode(p_hxge_t hxgep, boolean_t enable)
770 {
771 	int			i, ndmas;
772 	uint16_t		channel;
773 	p_rx_rbr_rings_t	rx_rbr_rings;
774 	p_rx_rbr_ring_t		*rbr_rings;
775 	hpi_handle_t		handle;
776 	hpi_status_t		rs = HPI_SUCCESS;
777 	hxge_status_t		status = HXGE_OK;
778 
779 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
780 	    "==> hxge_rxdma_hw_mode: mode %d", enable));
781 
782 	if (!(hxgep->drv_state & STATE_HW_INITIALIZED)) {
783 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
784 		    "<== hxge_rxdma_mode: not initialized"));
785 		return (HXGE_ERROR);
786 	}
787 
788 	rx_rbr_rings = hxgep->rx_rbr_rings;
789 	if (rx_rbr_rings == NULL) {
790 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
791 		    "<== hxge_rxdma_mode: NULL ring pointer"));
792 		return (HXGE_ERROR);
793 	}
794 
795 	if (rx_rbr_rings->rbr_rings == NULL) {
796 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
797 		    "<== hxge_rxdma_mode: NULL rbr rings pointer"));
798 		return (HXGE_ERROR);
799 	}
800 
801 	ndmas = rx_rbr_rings->ndmas;
802 	if (!ndmas) {
803 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
804 		    "<== hxge_rxdma_mode: no channel"));
805 		return (HXGE_ERROR);
806 	}
807 
808 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
809 	    "==> hxge_rxdma_mode (ndmas %d)", ndmas));
810 
811 	rbr_rings = rx_rbr_rings->rbr_rings;
812 
813 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
814 
815 	for (i = 0; i < ndmas; i++) {
816 		if (rbr_rings == NULL || rbr_rings[i] == NULL) {
817 			continue;
818 		}
819 		channel = rbr_rings[i]->rdc;
820 		if (enable) {
821 			HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
822 			    "==> hxge_rxdma_hw_mode: channel %d (enable)",
823 			    channel));
824 			rs = hpi_rxdma_cfg_rdc_enable(handle, channel);
825 		} else {
826 			HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
827 			    "==> hxge_rxdma_hw_mode: channel %d (disable)",
828 			    channel));
829 			rs = hpi_rxdma_cfg_rdc_disable(handle, channel);
830 		}
831 	}
832 
833 	status = ((rs == HPI_SUCCESS) ? HXGE_OK : HXGE_ERROR | rs);
834 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
835 	    "<== hxge_rxdma_hw_mode: status 0x%x", status));
836 
837 	return (status);
838 }
839 
840 int
841 hxge_rxdma_get_ring_index(p_hxge_t hxgep, uint16_t channel)
842 {
843 	int			i, ndmas;
844 	uint16_t		rdc;
845 	p_rx_rbr_rings_t 	rx_rbr_rings;
846 	p_rx_rbr_ring_t		*rbr_rings;
847 
848 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
849 	    "==> hxge_rxdma_get_ring_index: channel %d", channel));
850 
851 	rx_rbr_rings = hxgep->rx_rbr_rings;
852 	if (rx_rbr_rings == NULL) {
853 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
854 		    "<== hxge_rxdma_get_ring_index: NULL ring pointer"));
855 		return (-1);
856 	}
857 
858 	ndmas = rx_rbr_rings->ndmas;
859 	if (!ndmas) {
860 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
861 		    "<== hxge_rxdma_get_ring_index: no channel"));
862 		return (-1);
863 	}
864 
865 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
866 	    "==> hxge_rxdma_get_ring_index (ndmas %d)", ndmas));
867 
868 	rbr_rings = rx_rbr_rings->rbr_rings;
869 	for (i = 0; i < ndmas; i++) {
870 		rdc = rbr_rings[i]->rdc;
871 		if (channel == rdc) {
872 			HXGE_DEBUG_MSG((hxgep, RX_CTL,
873 			    "==> hxge_rxdma_get_rbr_ring: "
874 			    "channel %d (index %d) "
875 			    "ring %d", channel, i, rbr_rings[i]));
876 
877 			return (i);
878 		}
879 	}
880 
881 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
882 	    "<== hxge_rxdma_get_rbr_ring_index: not found"));
883 
884 	return (-1);
885 }
886 
887 /*
888  * Static functions start here.
889  */
890 static p_rx_msg_t
891 hxge_allocb(size_t size, uint32_t pri, p_hxge_dma_common_t dmabuf_p)
892 {
893 	p_rx_msg_t		hxge_mp = NULL;
894 	p_hxge_dma_common_t	dmamsg_p;
895 	uchar_t			*buffer;
896 
897 	hxge_mp = KMEM_ZALLOC(sizeof (rx_msg_t), KM_NOSLEEP);
898 	if (hxge_mp == NULL) {
899 		HXGE_ERROR_MSG((NULL, HXGE_ERR_CTL,
900 		    "Allocation of a rx msg failed."));
901 		goto hxge_allocb_exit;
902 	}
903 
904 	hxge_mp->use_buf_pool = B_FALSE;
905 	if (dmabuf_p) {
906 		hxge_mp->use_buf_pool = B_TRUE;
907 
908 		dmamsg_p = (p_hxge_dma_common_t)&hxge_mp->buf_dma;
909 		*dmamsg_p = *dmabuf_p;
910 		dmamsg_p->nblocks = 1;
911 		dmamsg_p->block_size = size;
912 		dmamsg_p->alength = size;
913 		buffer = (uchar_t *)dmabuf_p->kaddrp;
914 
915 		dmabuf_p->kaddrp = (void *)((char *)dmabuf_p->kaddrp + size);
916 		dmabuf_p->ioaddr_pp = (void *)
917 		    ((char *)dmabuf_p->ioaddr_pp + size);
918 
919 		dmabuf_p->alength -= size;
920 		dmabuf_p->offset += size;
921 		dmabuf_p->dma_cookie.dmac_laddress += size;
922 		dmabuf_p->dma_cookie.dmac_size -= size;
923 	} else {
924 		buffer = KMEM_ALLOC(size, KM_NOSLEEP);
925 		if (buffer == NULL) {
926 			HXGE_ERROR_MSG((NULL, HXGE_ERR_CTL,
927 			    "Allocation of a receive page failed."));
928 			goto hxge_allocb_fail1;
929 		}
930 	}
931 
932 	hxge_mp->rx_mblk_p = desballoc(buffer, size, pri, &hxge_mp->freeb);
933 	if (hxge_mp->rx_mblk_p == NULL) {
934 		HXGE_ERROR_MSG((NULL, HXGE_ERR_CTL, "desballoc failed."));
935 		goto hxge_allocb_fail2;
936 	}
937 	hxge_mp->buffer = buffer;
938 	hxge_mp->block_size = size;
939 	hxge_mp->freeb.free_func = (void (*) ()) hxge_freeb;
940 	hxge_mp->freeb.free_arg = (caddr_t)hxge_mp;
941 	hxge_mp->ref_cnt = 1;
942 	hxge_mp->free = B_TRUE;
943 	hxge_mp->rx_use_bcopy = B_FALSE;
944 
945 	atomic_inc_32(&hxge_mblks_pending);
946 
947 	goto hxge_allocb_exit;
948 
949 hxge_allocb_fail2:
950 	if (!hxge_mp->use_buf_pool) {
951 		KMEM_FREE(buffer, size);
952 	}
953 hxge_allocb_fail1:
954 	KMEM_FREE(hxge_mp, sizeof (rx_msg_t));
955 	hxge_mp = NULL;
956 
957 hxge_allocb_exit:
958 	return (hxge_mp);
959 }
960 
961 p_mblk_t
962 hxge_dupb(p_rx_msg_t hxge_mp, uint_t offset, size_t size)
963 {
964 	p_mblk_t mp;
965 
966 	HXGE_DEBUG_MSG((NULL, MEM_CTL, "==> hxge_dupb"));
967 	HXGE_DEBUG_MSG((NULL, MEM_CTL, "hxge_mp = $%p "
968 	    "offset = 0x%08X " "size = 0x%08X", hxge_mp, offset, size));
969 
970 	mp = desballoc(&hxge_mp->buffer[offset], size, 0, &hxge_mp->freeb);
971 	if (mp == NULL) {
972 		HXGE_DEBUG_MSG((NULL, RX_CTL, "desballoc failed"));
973 		goto hxge_dupb_exit;
974 	}
975 
976 	atomic_inc_32(&hxge_mp->ref_cnt);
977 
978 hxge_dupb_exit:
979 	HXGE_DEBUG_MSG((NULL, MEM_CTL, "<== hxge_dupb mp = $%p", hxge_mp));
980 	return (mp);
981 }
982 
983 p_mblk_t
984 hxge_dupb_bcopy(p_rx_msg_t hxge_mp, uint_t offset, size_t size)
985 {
986 	p_mblk_t	mp;
987 	uchar_t		*dp;
988 
989 	mp = allocb(size + HXGE_RXBUF_EXTRA, 0);
990 	if (mp == NULL) {
991 		HXGE_DEBUG_MSG((NULL, RX_CTL, "desballoc failed"));
992 		goto hxge_dupb_bcopy_exit;
993 	}
994 	dp = mp->b_rptr = mp->b_rptr + HXGE_RXBUF_EXTRA;
995 	bcopy((void *) &hxge_mp->buffer[offset], dp, size);
996 	mp->b_wptr = dp + size;
997 
998 hxge_dupb_bcopy_exit:
999 
1000 	HXGE_DEBUG_MSG((NULL, MEM_CTL, "<== hxge_dupb mp = $%p", hxge_mp));
1001 
1002 	return (mp);
1003 }
1004 
1005 void hxge_post_page(p_hxge_t hxgep, p_rx_rbr_ring_t rx_rbr_p,
1006     p_rx_msg_t rx_msg_p);
1007 
1008 void
1009 hxge_post_page(p_hxge_t hxgep, p_rx_rbr_ring_t rx_rbr_p, p_rx_msg_t rx_msg_p)
1010 {
1011 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_post_page"));
1012 
1013 	/* Reuse this buffer */
1014 	rx_msg_p->free = B_FALSE;
1015 	rx_msg_p->cur_usage_cnt = 0;
1016 	rx_msg_p->max_usage_cnt = 0;
1017 	rx_msg_p->pkt_buf_size = 0;
1018 
1019 	if (rx_rbr_p->rbr_use_bcopy) {
1020 		rx_msg_p->rx_use_bcopy = B_FALSE;
1021 		atomic_dec_32(&rx_rbr_p->rbr_consumed);
1022 	}
1023 	atomic_dec_32(&rx_rbr_p->rbr_used);
1024 
1025 	/*
1026 	 * Get the rbr header pointer and its offset index.
1027 	 */
1028 	rx_rbr_p->rbr_wr_index = ((rx_rbr_p->rbr_wr_index + 1) &
1029 	    rx_rbr_p->rbr_wrap_mask);
1030 	rx_rbr_p->rbr_desc_vp[rx_rbr_p->rbr_wr_index] = rx_msg_p->shifted_addr;
1031 
1032 	/*
1033 	 * Accumulate some buffers in the ring before re-enabling the
1034 	 * DMA channel, if rbr empty was signaled.
1035 	 */
1036 	hpi_rxdma_rdc_rbr_kick(HXGE_DEV_HPI_HANDLE(hxgep), rx_rbr_p->rdc, 1);
1037 	if (rx_rbr_p->rbr_is_empty &&
1038 	    (rx_rbr_p->rbb_max - rx_rbr_p->rbr_used) >=
1039 	    HXGE_RBR_EMPTY_THRESHOLD) {
1040 		hxge_rbr_empty_restore(hxgep, rx_rbr_p);
1041 	}
1042 
1043 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
1044 	    "<== hxge_post_page (channel %d post_next_index %d)",
1045 	    rx_rbr_p->rdc, rx_rbr_p->rbr_wr_index));
1046 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_post_page"));
1047 }
1048 
1049 void
1050 hxge_freeb(p_rx_msg_t rx_msg_p)
1051 {
1052 	size_t		size;
1053 	uchar_t		*buffer = NULL;
1054 	int		ref_cnt;
1055 	boolean_t	free_state = B_FALSE;
1056 	rx_rbr_ring_t	*ring = rx_msg_p->rx_rbr_p;
1057 
1058 	HXGE_DEBUG_MSG((NULL, MEM2_CTL, "==> hxge_freeb"));
1059 	HXGE_DEBUG_MSG((NULL, MEM2_CTL,
1060 	    "hxge_freeb:rx_msg_p = $%p (block pending %d)",
1061 	    rx_msg_p, hxge_mblks_pending));
1062 
1063 	if (ring == NULL)
1064 		return;
1065 
1066 	/*
1067 	 * This is to prevent posting activities while we are recovering
1068 	 * from fatal errors. This should not be a performance drag since
1069 	 * ref_cnt != 0 most times.
1070 	 */
1071 	if (ring->rbr_state == RBR_POSTING)
1072 		MUTEX_ENTER(&ring->post_lock);
1073 
1074 	/*
1075 	 * First we need to get the free state, then
1076 	 * atomic decrement the reference count to prevent
1077 	 * the race condition with the interrupt thread that
1078 	 * is processing a loaned up buffer block.
1079 	 */
1080 	free_state = rx_msg_p->free;
1081 	ref_cnt = atomic_add_32_nv(&rx_msg_p->ref_cnt, -1);
1082 	if (!ref_cnt) {
1083 		atomic_dec_32(&hxge_mblks_pending);
1084 
1085 		buffer = rx_msg_p->buffer;
1086 		size = rx_msg_p->block_size;
1087 
1088 		HXGE_DEBUG_MSG((NULL, MEM2_CTL, "hxge_freeb: "
1089 		    "will free: rx_msg_p = $%p (block pending %d)",
1090 		    rx_msg_p, hxge_mblks_pending));
1091 
1092 		if (!rx_msg_p->use_buf_pool) {
1093 			KMEM_FREE(buffer, size);
1094 		}
1095 
1096 		KMEM_FREE(rx_msg_p, sizeof (rx_msg_t));
1097 		/*
1098 		 * Decrement the receive buffer ring's reference
1099 		 * count, too.
1100 		 */
1101 		atomic_dec_32(&ring->rbr_ref_cnt);
1102 
1103 		/*
1104 		 * Free the receive buffer ring, iff
1105 		 * 1. all the receive buffers have been freed
1106 		 * 2. and we are in the proper state (that is,
1107 		 *    we are not UNMAPPING).
1108 		 */
1109 		if (ring->rbr_ref_cnt == 0 &&
1110 		    ring->rbr_state == RBR_UNMAPPED) {
1111 			KMEM_FREE(ring, sizeof (*ring));
1112 			/* post_lock has been destroyed already */
1113 			return;
1114 		}
1115 	}
1116 
1117 	/*
1118 	 * Repost buffer.
1119 	 */
1120 	if (free_state && (ref_cnt == 1)) {
1121 		HXGE_DEBUG_MSG((NULL, RX_CTL,
1122 		    "hxge_freeb: post page $%p:", rx_msg_p));
1123 		if (ring->rbr_state == RBR_POSTING)
1124 			hxge_post_page(rx_msg_p->hxgep, ring, rx_msg_p);
1125 	}
1126 
1127 	if (ring->rbr_state == RBR_POSTING)
1128 		MUTEX_EXIT(&ring->post_lock);
1129 
1130 	HXGE_DEBUG_MSG((NULL, MEM2_CTL, "<== hxge_freeb"));
1131 }
1132 
1133 uint_t
1134 hxge_rx_intr(caddr_t arg1, caddr_t arg2)
1135 {
1136 	p_hxge_ldv_t		ldvp = (p_hxge_ldv_t)arg1;
1137 	p_hxge_t		hxgep = (p_hxge_t)arg2;
1138 	p_hxge_ldg_t		ldgp;
1139 	uint8_t			channel;
1140 	hpi_handle_t		handle;
1141 	rdc_stat_t		cs;
1142 	uint_t			serviced = DDI_INTR_UNCLAIMED;
1143 
1144 	if (ldvp == NULL) {
1145 		HXGE_DEBUG_MSG((NULL, RX_INT_CTL,
1146 		    "<== hxge_rx_intr: arg2 $%p arg1 $%p", hxgep, ldvp));
1147 		return (DDI_INTR_UNCLAIMED);
1148 	}
1149 
1150 	if (arg2 == NULL || (void *) ldvp->hxgep != arg2) {
1151 		hxgep = ldvp->hxgep;
1152 	}
1153 
1154 	/*
1155 	 * If the interface is not started, just swallow the interrupt
1156 	 * for the logical device and don't rearm it.
1157 	 */
1158 	if (hxgep->hxge_mac_state != HXGE_MAC_STARTED)
1159 		return (DDI_INTR_CLAIMED);
1160 
1161 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1162 	    "==> hxge_rx_intr: arg2 $%p arg1 $%p", hxgep, ldvp));
1163 
1164 	/*
1165 	 * This interrupt handler is for a specific receive dma channel.
1166 	 */
1167 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
1168 
1169 	/*
1170 	 * Get the control and status for this channel.
1171 	 */
1172 	channel = ldvp->channel;
1173 	ldgp = ldvp->ldgp;
1174 	RXDMA_REG_READ64(handle, RDC_STAT, channel, &cs.value);
1175 	cs.bits.ptrread = 0;
1176 	cs.bits.pktread = 0;
1177 	RXDMA_REG_WRITE64(handle, RDC_STAT, channel, cs.value);
1178 
1179 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL, "==> hxge_rx_intr:channel %d "
1180 	    "cs 0x%016llx rcrto 0x%x rcrthres %x",
1181 	    channel, cs.value, cs.bits.rcr_to, cs.bits.rcr_thres));
1182 
1183 	hxge_rx_pkts_vring(hxgep, ldvp->vdma_index, ldvp, cs);
1184 	serviced = DDI_INTR_CLAIMED;
1185 
1186 	/* error events. */
1187 	if (cs.value & RDC_STAT_ERROR) {
1188 		(void) hxge_rx_err_evnts(hxgep, ldvp->vdma_index, ldvp, cs);
1189 	}
1190 
1191 hxge_intr_exit:
1192 	/*
1193 	 * Enable the mailbox update interrupt if we want to use mailbox. We
1194 	 * probably don't need to use mailbox as it only saves us one pio read.
1195 	 * Also write 1 to rcrthres and rcrto to clear these two edge triggered
1196 	 * bits.
1197 	 */
1198 	cs.value &= RDC_STAT_WR1C;
1199 	cs.bits.mex = 1;
1200 	cs.bits.ptrread = 0;
1201 	cs.bits.pktread = 0;
1202 	RXDMA_REG_WRITE64(handle, RDC_STAT, channel, cs.value);
1203 
1204 	/*
1205 	 * Rearm this logical group if this is a single device group.
1206 	 */
1207 	if (ldgp->nldvs == 1) {
1208 		ld_intr_mgmt_t mgm;
1209 
1210 		mgm.value = 0;
1211 		mgm.bits.arm = 1;
1212 		mgm.bits.timer = ldgp->ldg_timer;
1213 		HXGE_REG_WR32(handle,
1214 		    LD_INTR_MGMT + LDSV_OFFSET(ldgp->ldg), mgm.value);
1215 	}
1216 
1217 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1218 	    "<== hxge_rx_intr: serviced %d", serviced));
1219 	return (serviced);
1220 }
1221 
1222 static void
1223 hxge_rx_pkts_vring(p_hxge_t hxgep, uint_t vindex, p_hxge_ldv_t ldvp,
1224     rdc_stat_t cs)
1225 {
1226 	p_mblk_t		mp;
1227 	p_rx_rcr_ring_t		rcrp;
1228 
1229 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL, "==> hxge_rx_pkts_vring"));
1230 	if ((mp = hxge_rx_pkts(hxgep, vindex, ldvp, &rcrp, cs)) == NULL) {
1231 		HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1232 		    "<== hxge_rx_pkts_vring: no mp"));
1233 		return;
1234 	}
1235 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rx_pkts_vring: $%p", mp));
1236 
1237 #ifdef  HXGE_DEBUG
1238 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
1239 	    "==> hxge_rx_pkts_vring:calling mac_rx (NEMO) "
1240 	    "LEN %d mp $%p mp->b_next $%p rcrp $%p",
1241 	    (mp->b_wptr - mp->b_rptr), mp, mp->b_next, rcrp));
1242 
1243 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
1244 	    "==> hxge_rx_pkts_vring: dump packets "
1245 	    "(mp $%p b_rptr $%p b_wptr $%p):\n %s",
1246 	    mp, mp->b_rptr, mp->b_wptr,
1247 	    hxge_dump_packet((char *)mp->b_rptr, 64)));
1248 
1249 	if (mp->b_cont) {
1250 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
1251 		    "==> hxge_rx_pkts_vring: dump b_cont packets "
1252 		    "(mp->b_cont $%p b_rptr $%p b_wptr $%p):\n %s",
1253 		    mp->b_cont, mp->b_cont->b_rptr, mp->b_cont->b_wptr,
1254 		    hxge_dump_packet((char *)mp->b_cont->b_rptr,
1255 		    mp->b_cont->b_wptr - mp->b_cont->b_rptr)));
1256 		}
1257 	if (mp->b_next) {
1258 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
1259 		    "==> hxge_rx_pkts_vring: dump next packets "
1260 		    "(b_rptr $%p): %s",
1261 		    mp->b_next->b_rptr,
1262 		    hxge_dump_packet((char *)mp->b_next->b_rptr, 64)));
1263 	}
1264 #endif
1265 
1266 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
1267 	    "==> hxge_rx_pkts_vring: send packet to stack"));
1268 	mac_rx(hxgep->mach, NULL, mp);
1269 
1270 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_rx_pkts_vring"));
1271 }
1272 
1273 /*ARGSUSED*/
1274 mblk_t *
1275 hxge_rx_pkts(p_hxge_t hxgep, uint_t vindex, p_hxge_ldv_t ldvp,
1276     p_rx_rcr_ring_t *rcrp, rdc_stat_t cs)
1277 {
1278 	hpi_handle_t		handle;
1279 	uint8_t			channel;
1280 	p_rx_rcr_rings_t	rx_rcr_rings;
1281 	p_rx_rcr_ring_t		rcr_p;
1282 	uint32_t		comp_rd_index;
1283 	p_rcr_entry_t		rcr_desc_rd_head_p;
1284 	p_rcr_entry_t		rcr_desc_rd_head_pp;
1285 	p_mblk_t		nmp, mp_cont, head_mp, *tail_mp;
1286 	uint16_t		qlen, nrcr_read, npkt_read;
1287 	uint32_t		qlen_hw, qlen_sw, num_rcrs;
1288 	uint32_t		invalid_rcr_entry;
1289 	boolean_t		multi;
1290 	rdc_rcr_cfg_b_t		rcr_cfg_b;
1291 	uint64_t		rcr_head_index, rcr_tail_index;
1292 	uint64_t		rcr_tail;
1293 	rdc_rcr_tail_t		rcr_tail_reg;
1294 	p_hxge_rx_ring_stats_t	rdc_stats;
1295 
1296 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL, "==> hxge_rx_pkts:vindex %d "
1297 	    "channel %d", vindex, ldvp->channel));
1298 
1299 	if (!(hxgep->drv_state & STATE_HW_INITIALIZED)) {
1300 		return (NULL);
1301 	}
1302 
1303 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
1304 	rx_rcr_rings = hxgep->rx_rcr_rings;
1305 	rcr_p = rx_rcr_rings->rcr_rings[vindex];
1306 	channel = rcr_p->rdc;
1307 	if (channel != ldvp->channel) {
1308 		HXGE_DEBUG_MSG((hxgep, RX_INT_CTL, "==> hxge_rx_pkts:index %d "
1309 		    "channel %d, and rcr channel %d not matched.",
1310 		    vindex, ldvp->channel, channel));
1311 		return (NULL);
1312 	}
1313 
1314 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1315 	    "==> hxge_rx_pkts: START: rcr channel %d "
1316 	    "head_p $%p head_pp $%p  index %d ",
1317 	    channel, rcr_p->rcr_desc_rd_head_p,
1318 	    rcr_p->rcr_desc_rd_head_pp, rcr_p->comp_rd_index));
1319 
1320 	(void) hpi_rxdma_rdc_rcr_qlen_get(handle, channel, &qlen);
1321 	RXDMA_REG_READ64(handle, RDC_RCR_TAIL, channel, &rcr_tail_reg.value);
1322 	rcr_tail = rcr_tail_reg.bits.tail;
1323 
1324 	if (!qlen) {
1325 		HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1326 		    "<== hxge_rx_pkts:rcr channel %d qlen %d (no pkts)",
1327 		    channel, qlen));
1328 		return (NULL);
1329 	}
1330 
1331 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rx_pkts:rcr channel %d "
1332 	    "qlen %d", channel, qlen));
1333 
1334 	comp_rd_index = rcr_p->comp_rd_index;
1335 
1336 	rcr_desc_rd_head_p = rcr_p->rcr_desc_rd_head_p;
1337 	rcr_desc_rd_head_pp = rcr_p->rcr_desc_rd_head_pp;
1338 	nrcr_read = npkt_read = 0;
1339 
1340 	if (hxgep->rdc_first_intr[channel])
1341 		qlen_hw = qlen;
1342 	else
1343 		qlen_hw = qlen - 1;
1344 
1345 	head_mp = NULL;
1346 	tail_mp = &head_mp;
1347 	nmp = mp_cont = NULL;
1348 	multi = B_FALSE;
1349 
1350 	rcr_head_index = rcr_p->rcr_desc_rd_head_p - rcr_p->rcr_desc_first_p;
1351 	rcr_tail_index = rcr_tail - rcr_p->rcr_tail_begin;
1352 
1353 	if (rcr_tail_index >= rcr_head_index) {
1354 		num_rcrs = rcr_tail_index - rcr_head_index;
1355 	} else {
1356 		/* rcr_tail has wrapped around */
1357 		num_rcrs = (rcr_p->comp_size - rcr_head_index) + rcr_tail_index;
1358 	}
1359 
1360 	qlen_sw = hxge_scan_for_last_eop(rcr_p, rcr_desc_rd_head_p, num_rcrs);
1361 	if (!qlen_sw)
1362 		return (NULL);
1363 
1364 	if (qlen_hw > qlen_sw) {
1365 		HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1366 		    "Channel %d, rcr_qlen from reg %d and from rcr_tail %d\n",
1367 		    channel, qlen_hw, qlen_sw));
1368 		qlen_hw = qlen_sw;
1369 	}
1370 
1371 	while (qlen_hw) {
1372 #ifdef HXGE_DEBUG
1373 		hxge_dump_rcr_entry(hxgep, rcr_desc_rd_head_p);
1374 #endif
1375 		/*
1376 		 * Process one completion ring entry.
1377 		 */
1378 		invalid_rcr_entry = 0;
1379 		hxge_receive_packet(hxgep,
1380 		    rcr_p, rcr_desc_rd_head_p, &multi, &nmp, &mp_cont,
1381 		    &invalid_rcr_entry);
1382 		if (invalid_rcr_entry != 0) {
1383 			rdc_stats = rcr_p->rdc_stats;
1384 			rdc_stats->rcr_invalids++;
1385 			HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1386 			    "Channel %d could only read 0x%x packets, "
1387 			    "but 0x%x pending\n", channel, npkt_read, qlen_hw));
1388 			break;
1389 		}
1390 
1391 		/*
1392 		 * message chaining modes (nemo msg chaining)
1393 		 */
1394 		if (nmp) {
1395 			nmp->b_next = NULL;
1396 			if (!multi && !mp_cont) { /* frame fits a partition */
1397 				*tail_mp = nmp;
1398 				tail_mp = &nmp->b_next;
1399 				nmp = NULL;
1400 			} else if (multi && !mp_cont) { /* first segment */
1401 				*tail_mp = nmp;
1402 				tail_mp = &nmp->b_cont;
1403 			} else if (multi && mp_cont) {	/* mid of multi segs */
1404 				*tail_mp = mp_cont;
1405 				tail_mp = &mp_cont->b_cont;
1406 			} else if (!multi && mp_cont) { /* last segment */
1407 				*tail_mp = mp_cont;
1408 				tail_mp = &nmp->b_next;
1409 				nmp = NULL;
1410 			}
1411 		}
1412 
1413 		HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1414 		    "==> hxge_rx_pkts: loop: rcr channel %d "
1415 		    "before updating: multi %d "
1416 		    "nrcr_read %d "
1417 		    "npk read %d "
1418 		    "head_pp $%p  index %d ",
1419 		    channel, multi,
1420 		    nrcr_read, npkt_read, rcr_desc_rd_head_pp, comp_rd_index));
1421 
1422 		if (!multi) {
1423 			qlen_hw--;
1424 			npkt_read++;
1425 		}
1426 
1427 		/*
1428 		 * Update the next read entry.
1429 		 */
1430 		comp_rd_index = NEXT_ENTRY(comp_rd_index,
1431 		    rcr_p->comp_wrap_mask);
1432 
1433 		rcr_desc_rd_head_p = NEXT_ENTRY_PTR(rcr_desc_rd_head_p,
1434 		    rcr_p->rcr_desc_first_p, rcr_p->rcr_desc_last_p);
1435 
1436 		nrcr_read++;
1437 
1438 		HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1439 		    "<== hxge_rx_pkts: (SAM, process one packet) "
1440 		    "nrcr_read %d", nrcr_read));
1441 		HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1442 		    "==> hxge_rx_pkts: loop: rcr channel %d "
1443 		    "multi %d nrcr_read %d npk read %d head_pp $%p  index %d ",
1444 		    channel, multi, nrcr_read, npkt_read, rcr_desc_rd_head_pp,
1445 		    comp_rd_index));
1446 	}
1447 
1448 	rcr_p->rcr_desc_rd_head_pp = rcr_desc_rd_head_pp;
1449 	rcr_p->comp_rd_index = comp_rd_index;
1450 	rcr_p->rcr_desc_rd_head_p = rcr_desc_rd_head_p;
1451 
1452 	if ((hxgep->intr_timeout != rcr_p->intr_timeout) ||
1453 	    (hxgep->intr_threshold != rcr_p->intr_threshold)) {
1454 		rcr_p->intr_timeout = hxgep->intr_timeout;
1455 		rcr_p->intr_threshold = hxgep->intr_threshold;
1456 		rcr_cfg_b.value = 0x0ULL;
1457 		if (rcr_p->intr_timeout)
1458 			rcr_cfg_b.bits.entout = 1;
1459 		rcr_cfg_b.bits.timeout = rcr_p->intr_timeout;
1460 		rcr_cfg_b.bits.pthres = rcr_p->intr_threshold;
1461 		RXDMA_REG_WRITE64(handle, RDC_RCR_CFG_B,
1462 		    channel, rcr_cfg_b.value);
1463 	}
1464 
1465 	if (hxgep->rdc_first_intr[channel] && (npkt_read > 0)) {
1466 		hxgep->rdc_first_intr[channel] = B_FALSE;
1467 		cs.bits.pktread = npkt_read - 1;
1468 	} else
1469 		cs.bits.pktread = npkt_read;
1470 	cs.bits.ptrread = nrcr_read;
1471 	cs.value &= 0xffffffffULL;
1472 	RXDMA_REG_WRITE64(handle, RDC_STAT, channel, cs.value);
1473 
1474 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1475 	    "==> hxge_rx_pkts: EXIT: rcr channel %d "
1476 	    "head_pp $%p  index %016llx ",
1477 	    channel, rcr_p->rcr_desc_rd_head_pp, rcr_p->comp_rd_index));
1478 
1479 	/*
1480 	 * Update RCR buffer pointer read and number of packets read.
1481 	 */
1482 	*rcrp = rcr_p;
1483 
1484 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL, "<== hxge_rx_pkts"));
1485 
1486 	return (head_mp);
1487 }
1488 
1489 #define	RCR_ENTRY_PATTERN	0x5a5a6b6b7c7c8d8dULL
1490 #define	NO_PORT_BIT		0x20
1491 #define	L4_CS_EQ_BIT		0x40
1492 
1493 static uint32_t hxge_scan_for_last_eop(p_rx_rcr_ring_t rcr_p,
1494     p_rcr_entry_t rcr_desc_rd_head_p, uint32_t num_rcrs)
1495 {
1496 	uint64_t	rcr_entry;
1497 	uint32_t	rcrs = 0;
1498 	uint32_t	pkts = 0;
1499 
1500 	while (rcrs++ < num_rcrs) {
1501 		rcr_entry = *((uint64_t *)rcr_desc_rd_head_p);
1502 
1503 		if ((rcr_entry == 0x0) || (rcr_entry == RCR_ENTRY_PATTERN))
1504 			break;
1505 
1506 		if (!(rcr_entry & RCR_MULTI_MASK))
1507 			pkts++;
1508 
1509 		rcr_desc_rd_head_p = NEXT_ENTRY_PTR(rcr_desc_rd_head_p,
1510 		    rcr_p->rcr_desc_first_p, rcr_p->rcr_desc_last_p);
1511 	}
1512 
1513 	return (pkts);
1514 }
1515 
1516 /*ARGSUSED*/
1517 void
1518 hxge_receive_packet(p_hxge_t hxgep,
1519     p_rx_rcr_ring_t rcr_p, p_rcr_entry_t rcr_desc_rd_head_p,
1520     boolean_t *multi_p, mblk_t **mp, mblk_t **mp_cont,
1521     uint32_t *invalid_rcr_entry)
1522 {
1523 	p_mblk_t		nmp = NULL;
1524 	uint64_t		multi;
1525 	uint8_t			channel;
1526 
1527 	boolean_t first_entry = B_TRUE;
1528 	boolean_t is_tcp_udp = B_FALSE;
1529 	boolean_t buffer_free = B_FALSE;
1530 	boolean_t error_send_up = B_FALSE;
1531 	uint8_t error_type;
1532 	uint16_t l2_len;
1533 	uint16_t skip_len;
1534 	uint8_t pktbufsz_type;
1535 	uint64_t rcr_entry;
1536 	uint64_t *pkt_buf_addr_pp;
1537 	uint64_t *pkt_buf_addr_p;
1538 	uint32_t buf_offset;
1539 	uint32_t bsize;
1540 	uint32_t msg_index;
1541 	p_rx_rbr_ring_t rx_rbr_p;
1542 	p_rx_msg_t *rx_msg_ring_p;
1543 	p_rx_msg_t rx_msg_p;
1544 
1545 	uint16_t sw_offset_bytes = 0, hdr_size = 0;
1546 	hxge_status_t status = HXGE_OK;
1547 	boolean_t is_valid = B_FALSE;
1548 	p_hxge_rx_ring_stats_t rdc_stats;
1549 	uint32_t bytes_read;
1550 	uint8_t header0 = 0;
1551 	uint8_t header1 = 0;
1552 	uint64_t pkt_type;
1553 	uint8_t no_port_bit = 0;
1554 	uint8_t l4_cs_eq_bit = 0;
1555 
1556 	channel = rcr_p->rdc;
1557 
1558 	HXGE_DEBUG_MSG((hxgep, RX2_CTL, "==> hxge_receive_packet"));
1559 
1560 	first_entry = (*mp == NULL) ? B_TRUE : B_FALSE;
1561 	rcr_entry = *((uint64_t *)rcr_desc_rd_head_p);
1562 
1563 	/* Verify the content of the rcr_entry for a hardware bug workaround */
1564 	if ((rcr_entry == 0x0) || (rcr_entry == RCR_ENTRY_PATTERN)) {
1565 		*invalid_rcr_entry = 1;
1566 		HXGE_DEBUG_MSG((hxgep, RX2_CTL, "hxge_receive_packet "
1567 		    "Channel %d invalid RCR entry 0x%llx found, returning\n",
1568 		    channel, (long long) rcr_entry));
1569 		return;
1570 	}
1571 	*((uint64_t *)rcr_desc_rd_head_p) = RCR_ENTRY_PATTERN;
1572 
1573 	multi = (rcr_entry & RCR_MULTI_MASK);
1574 	pkt_type = (rcr_entry & RCR_PKT_TYPE_MASK);
1575 
1576 	error_type = ((rcr_entry & RCR_ERROR_MASK) >> RCR_ERROR_SHIFT);
1577 	l2_len = ((rcr_entry & RCR_L2_LEN_MASK) >> RCR_L2_LEN_SHIFT);
1578 
1579 	/*
1580 	 * Hardware does not strip the CRC due bug ID 11451 where
1581 	 * the hardware mis handles minimum size packets.
1582 	 */
1583 	l2_len -= ETHERFCSL;
1584 
1585 	pktbufsz_type = ((rcr_entry & RCR_PKTBUFSZ_MASK) >>
1586 	    RCR_PKTBUFSZ_SHIFT);
1587 #if defined(__i386)
1588 	pkt_buf_addr_pp = (uint64_t *)(uint32_t)((rcr_entry &
1589 	    RCR_PKT_BUF_ADDR_MASK) << RCR_PKT_BUF_ADDR_SHIFT);
1590 #else
1591 	pkt_buf_addr_pp = (uint64_t *)((rcr_entry & RCR_PKT_BUF_ADDR_MASK) <<
1592 	    RCR_PKT_BUF_ADDR_SHIFT);
1593 #endif
1594 
1595 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1596 	    "==> hxge_receive_packet: entryp $%p entry 0x%0llx "
1597 	    "pkt_buf_addr_pp $%p l2_len %d multi %d "
1598 	    "error_type 0x%x pktbufsz_type %d ",
1599 	    rcr_desc_rd_head_p, rcr_entry, pkt_buf_addr_pp, l2_len,
1600 	    multi, error_type, pktbufsz_type));
1601 
1602 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1603 	    "==> hxge_receive_packet: entryp $%p entry 0x%0llx "
1604 	    "pkt_buf_addr_pp $%p l2_len %d multi %d "
1605 	    "error_type 0x%x ", rcr_desc_rd_head_p,
1606 	    rcr_entry, pkt_buf_addr_pp, l2_len, multi, error_type));
1607 
1608 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1609 	    "==> (rbr) hxge_receive_packet: entry 0x%0llx "
1610 	    "full pkt_buf_addr_pp $%p l2_len %d",
1611 	    rcr_entry, pkt_buf_addr_pp, l2_len));
1612 
1613 	/* get the stats ptr */
1614 	rdc_stats = rcr_p->rdc_stats;
1615 
1616 	if (!l2_len) {
1617 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
1618 		    "<== hxge_receive_packet: failed: l2 length is 0."));
1619 		return;
1620 	}
1621 
1622 	/* shift 6 bits to get the full io address */
1623 #if defined(__i386)
1624 	pkt_buf_addr_pp = (uint64_t *)((uint32_t)pkt_buf_addr_pp <<
1625 	    RCR_PKT_BUF_ADDR_SHIFT_FULL);
1626 #else
1627 	pkt_buf_addr_pp = (uint64_t *)((uint64_t)pkt_buf_addr_pp <<
1628 	    RCR_PKT_BUF_ADDR_SHIFT_FULL);
1629 #endif
1630 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1631 	    "==> (rbr) hxge_receive_packet: entry 0x%0llx "
1632 	    "full pkt_buf_addr_pp $%p l2_len %d",
1633 	    rcr_entry, pkt_buf_addr_pp, l2_len));
1634 
1635 	rx_rbr_p = rcr_p->rx_rbr_p;
1636 	rx_msg_ring_p = rx_rbr_p->rx_msg_ring;
1637 
1638 	if (first_entry) {
1639 		hdr_size = (rcr_p->full_hdr_flag ? RXDMA_HDR_SIZE_FULL :
1640 		    RXDMA_HDR_SIZE_DEFAULT);
1641 
1642 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
1643 		    "==> hxge_receive_packet: first entry 0x%016llx "
1644 		    "pkt_buf_addr_pp $%p l2_len %d hdr %d",
1645 		    rcr_entry, pkt_buf_addr_pp, l2_len, hdr_size));
1646 	}
1647 
1648 	MUTEX_ENTER(&rcr_p->lock);
1649 	MUTEX_ENTER(&rx_rbr_p->lock);
1650 
1651 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
1652 	    "==> (rbr 1) hxge_receive_packet: entry 0x%0llx "
1653 	    "full pkt_buf_addr_pp $%p l2_len %d",
1654 	    rcr_entry, pkt_buf_addr_pp, l2_len));
1655 
1656 	/*
1657 	 * Packet buffer address in the completion entry points to the starting
1658 	 * buffer address (offset 0). Use the starting buffer address to locate
1659 	 * the corresponding kernel address.
1660 	 */
1661 	status = hxge_rxbuf_pp_to_vp(hxgep, rx_rbr_p,
1662 	    pktbufsz_type, pkt_buf_addr_pp, &pkt_buf_addr_p,
1663 	    &buf_offset, &msg_index);
1664 
1665 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
1666 	    "==> (rbr 2) hxge_receive_packet: entry 0x%0llx "
1667 	    "full pkt_buf_addr_pp $%p l2_len %d",
1668 	    rcr_entry, pkt_buf_addr_pp, l2_len));
1669 
1670 	if (status != HXGE_OK) {
1671 		MUTEX_EXIT(&rx_rbr_p->lock);
1672 		MUTEX_EXIT(&rcr_p->lock);
1673 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
1674 		    "<== hxge_receive_packet: found vaddr failed %d", status));
1675 		return;
1676 	}
1677 
1678 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1679 	    "==> (rbr 3) hxge_receive_packet: entry 0x%0llx "
1680 	    "full pkt_buf_addr_pp $%p l2_len %d",
1681 	    rcr_entry, pkt_buf_addr_pp, l2_len));
1682 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1683 	    "==> (rbr 4 msgindex %d) hxge_receive_packet: entry 0x%0llx "
1684 	    "full pkt_buf_addr_pp $%p l2_len %d",
1685 	    msg_index, rcr_entry, pkt_buf_addr_pp, l2_len));
1686 
1687 	if (msg_index >= rx_rbr_p->tnblocks) {
1688 		MUTEX_EXIT(&rx_rbr_p->lock);
1689 		MUTEX_EXIT(&rcr_p->lock);
1690 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1691 		    "==> hxge_receive_packet: FATAL msg_index (%d) "
1692 		    "should be smaller than tnblocks (%d)\n",
1693 		    msg_index, rx_rbr_p->tnblocks));
1694 		return;
1695 	}
1696 
1697 	rx_msg_p = rx_msg_ring_p[msg_index];
1698 
1699 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1700 	    "==> (rbr 4 msgindex %d) hxge_receive_packet: entry 0x%0llx "
1701 	    "full pkt_buf_addr_pp $%p l2_len %d",
1702 	    msg_index, rcr_entry, pkt_buf_addr_pp, l2_len));
1703 
1704 	switch (pktbufsz_type) {
1705 	case RCR_PKTBUFSZ_0:
1706 		bsize = rx_rbr_p->pkt_buf_size0_bytes;
1707 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1708 		    "==> hxge_receive_packet: 0 buf %d", bsize));
1709 		break;
1710 	case RCR_PKTBUFSZ_1:
1711 		bsize = rx_rbr_p->pkt_buf_size1_bytes;
1712 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1713 		    "==> hxge_receive_packet: 1 buf %d", bsize));
1714 		break;
1715 	case RCR_PKTBUFSZ_2:
1716 		bsize = rx_rbr_p->pkt_buf_size2_bytes;
1717 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
1718 		    "==> hxge_receive_packet: 2 buf %d", bsize));
1719 		break;
1720 	case RCR_SINGLE_BLOCK:
1721 		bsize = rx_msg_p->block_size;
1722 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1723 		    "==> hxge_receive_packet: single %d", bsize));
1724 
1725 		break;
1726 	default:
1727 		MUTEX_EXIT(&rx_rbr_p->lock);
1728 		MUTEX_EXIT(&rcr_p->lock);
1729 		return;
1730 	}
1731 
1732 	DMA_COMMON_SYNC_OFFSET(rx_msg_p->buf_dma,
1733 	    (buf_offset + sw_offset_bytes), (hdr_size + l2_len),
1734 	    DDI_DMA_SYNC_FORCPU);
1735 
1736 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1737 	    "==> hxge_receive_packet: after first dump:usage count"));
1738 
1739 	if (rx_msg_p->cur_usage_cnt == 0) {
1740 		atomic_inc_32(&rx_rbr_p->rbr_used);
1741 		if (rx_rbr_p->rbr_use_bcopy) {
1742 			atomic_inc_32(&rx_rbr_p->rbr_consumed);
1743 			if (rx_rbr_p->rbr_consumed <
1744 			    rx_rbr_p->rbr_threshold_hi) {
1745 				if (rx_rbr_p->rbr_threshold_lo == 0 ||
1746 				    ((rx_rbr_p->rbr_consumed >=
1747 				    rx_rbr_p->rbr_threshold_lo) &&
1748 				    (rx_rbr_p->rbr_bufsize_type >=
1749 				    pktbufsz_type))) {
1750 					rx_msg_p->rx_use_bcopy = B_TRUE;
1751 				}
1752 			} else {
1753 				rx_msg_p->rx_use_bcopy = B_TRUE;
1754 			}
1755 		}
1756 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1757 		    "==> hxge_receive_packet: buf %d (new block) ", bsize));
1758 
1759 		rx_msg_p->pkt_buf_size_code = pktbufsz_type;
1760 		rx_msg_p->pkt_buf_size = bsize;
1761 		rx_msg_p->cur_usage_cnt = 1;
1762 		if (pktbufsz_type == RCR_SINGLE_BLOCK) {
1763 			HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1764 			    "==> hxge_receive_packet: buf %d (single block) ",
1765 			    bsize));
1766 			/*
1767 			 * Buffer can be reused once the free function is
1768 			 * called.
1769 			 */
1770 			rx_msg_p->max_usage_cnt = 1;
1771 			buffer_free = B_TRUE;
1772 		} else {
1773 			rx_msg_p->max_usage_cnt = rx_msg_p->block_size / bsize;
1774 			if (rx_msg_p->max_usage_cnt == 1) {
1775 				buffer_free = B_TRUE;
1776 			}
1777 		}
1778 	} else {
1779 		rx_msg_p->cur_usage_cnt++;
1780 		if (rx_msg_p->cur_usage_cnt == rx_msg_p->max_usage_cnt) {
1781 			buffer_free = B_TRUE;
1782 		}
1783 	}
1784 
1785 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
1786 	    "msgbuf index = %d l2len %d bytes usage %d max_usage %d ",
1787 	    msg_index, l2_len,
1788 	    rx_msg_p->cur_usage_cnt, rx_msg_p->max_usage_cnt));
1789 
1790 	if (error_type) {
1791 		rdc_stats->ierrors++;
1792 		/* Update error stats */
1793 		rdc_stats->errlog.compl_err_type = error_type;
1794 		HXGE_FM_REPORT_ERROR(hxgep, NULL, HXGE_FM_EREPORT_RDMC_RCR_ERR);
1795 
1796 		if (error_type & RCR_CTRL_FIFO_DED) {
1797 			rdc_stats->ctrl_fifo_ecc_err++;
1798 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
1799 			    " hxge_receive_packet: "
1800 			    " channel %d RCR ctrl_fifo_ded error", channel));
1801 		} else if (error_type & RCR_DATA_FIFO_DED) {
1802 			rdc_stats->data_fifo_ecc_err++;
1803 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
1804 			    " hxge_receive_packet: channel %d"
1805 			    " RCR data_fifo_ded error", channel));
1806 		}
1807 
1808 		/*
1809 		 * Update and repost buffer block if max usage count is
1810 		 * reached.
1811 		 */
1812 		if (error_send_up == B_FALSE) {
1813 			atomic_inc_32(&rx_msg_p->ref_cnt);
1814 			if (buffer_free == B_TRUE) {
1815 				rx_msg_p->free = B_TRUE;
1816 			}
1817 
1818 			MUTEX_EXIT(&rx_rbr_p->lock);
1819 			MUTEX_EXIT(&rcr_p->lock);
1820 			hxge_freeb(rx_msg_p);
1821 			return;
1822 		}
1823 	}
1824 
1825 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1826 	    "==> hxge_receive_packet: DMA sync second "));
1827 
1828 	bytes_read = rcr_p->rcvd_pkt_bytes;
1829 	skip_len = sw_offset_bytes + hdr_size;
1830 
1831 	if (first_entry) {
1832 		header0 = rx_msg_p->buffer[buf_offset];
1833 		no_port_bit = header0 & NO_PORT_BIT;
1834 
1835 		header1 = rx_msg_p->buffer[buf_offset + 1];
1836 		l4_cs_eq_bit = header1 & L4_CS_EQ_BIT;
1837 	}
1838 
1839 	if (!rx_msg_p->rx_use_bcopy) {
1840 		/*
1841 		 * For loaned up buffers, the driver reference count
1842 		 * will be incremented first and then the free state.
1843 		 */
1844 		if ((nmp = hxge_dupb(rx_msg_p, buf_offset, bsize)) != NULL) {
1845 			if (first_entry) {
1846 				nmp->b_rptr = &nmp->b_rptr[skip_len];
1847 				if (l2_len < bsize - skip_len) {
1848 					nmp->b_wptr = &nmp->b_rptr[l2_len];
1849 				} else {
1850 					nmp->b_wptr = &nmp->b_rptr[bsize
1851 					    - skip_len];
1852 				}
1853 			} else {
1854 				if (l2_len - bytes_read < bsize) {
1855 					nmp->b_wptr =
1856 					    &nmp->b_rptr[l2_len - bytes_read];
1857 				} else {
1858 					nmp->b_wptr = &nmp->b_rptr[bsize];
1859 				}
1860 			}
1861 		}
1862 	} else {
1863 		if (first_entry) {
1864 			nmp = hxge_dupb_bcopy(rx_msg_p, buf_offset + skip_len,
1865 			    l2_len < bsize - skip_len ?
1866 			    l2_len : bsize - skip_len);
1867 		} else {
1868 			nmp = hxge_dupb_bcopy(rx_msg_p, buf_offset,
1869 			    l2_len - bytes_read < bsize ?
1870 			    l2_len - bytes_read : bsize);
1871 		}
1872 	}
1873 
1874 	if (nmp != NULL) {
1875 		if (first_entry)
1876 			bytes_read  = nmp->b_wptr - nmp->b_rptr;
1877 		else
1878 			bytes_read += nmp->b_wptr - nmp->b_rptr;
1879 
1880 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
1881 		    "==> hxge_receive_packet after dupb: "
1882 		    "rbr consumed %d "
1883 		    "pktbufsz_type %d "
1884 		    "nmp $%p rptr $%p wptr $%p "
1885 		    "buf_offset %d bzise %d l2_len %d skip_len %d",
1886 		    rx_rbr_p->rbr_consumed,
1887 		    pktbufsz_type,
1888 		    nmp, nmp->b_rptr, nmp->b_wptr,
1889 		    buf_offset, bsize, l2_len, skip_len));
1890 	} else {
1891 		cmn_err(CE_WARN, "!hxge_receive_packet: update stats (error)");
1892 
1893 		atomic_inc_32(&rx_msg_p->ref_cnt);
1894 		if (buffer_free == B_TRUE) {
1895 			rx_msg_p->free = B_TRUE;
1896 		}
1897 
1898 		MUTEX_EXIT(&rx_rbr_p->lock);
1899 		MUTEX_EXIT(&rcr_p->lock);
1900 		hxge_freeb(rx_msg_p);
1901 		return;
1902 	}
1903 
1904 	if (buffer_free == B_TRUE) {
1905 		rx_msg_p->free = B_TRUE;
1906 	}
1907 
1908 	/*
1909 	 * ERROR, FRAG and PKT_TYPE are only reported in the first entry. If a
1910 	 * packet is not fragmented and no error bit is set, then L4 checksum
1911 	 * is OK.
1912 	 */
1913 	is_valid = (nmp != NULL);
1914 	if (first_entry) {
1915 		rdc_stats->ipackets++; /* count only 1st seg for jumbo */
1916 		if (l2_len > (STD_FRAME_SIZE - ETHERFCSL))
1917 			rdc_stats->jumbo_pkts++;
1918 		rdc_stats->ibytes += skip_len + l2_len < bsize ?
1919 		    l2_len : bsize;
1920 	} else {
1921 		/*
1922 		 * Add the current portion of the packet to the kstats.
1923 		 * The current portion of the packet is calculated by using
1924 		 * length of the packet and the previously received portion.
1925 		 */
1926 		rdc_stats->ibytes += l2_len - rcr_p->rcvd_pkt_bytes < bsize ?
1927 		    l2_len - rcr_p->rcvd_pkt_bytes : bsize;
1928 	}
1929 
1930 	rcr_p->rcvd_pkt_bytes = bytes_read;
1931 
1932 	if (rx_msg_p->free && rx_msg_p->rx_use_bcopy) {
1933 		atomic_inc_32(&rx_msg_p->ref_cnt);
1934 		MUTEX_EXIT(&rx_rbr_p->lock);
1935 		MUTEX_EXIT(&rcr_p->lock);
1936 		hxge_freeb(rx_msg_p);
1937 	} else {
1938 		MUTEX_EXIT(&rx_rbr_p->lock);
1939 		MUTEX_EXIT(&rcr_p->lock);
1940 	}
1941 
1942 	if (is_valid) {
1943 		nmp->b_cont = NULL;
1944 		if (first_entry) {
1945 			*mp = nmp;
1946 			*mp_cont = NULL;
1947 		} else {
1948 			*mp_cont = nmp;
1949 		}
1950 	}
1951 
1952 	/*
1953 	 * Update stats and hardware checksuming.
1954 	 */
1955 	if (is_valid && !multi) {
1956 		is_tcp_udp = ((pkt_type == RCR_PKT_IS_TCP ||
1957 		    pkt_type == RCR_PKT_IS_UDP) ? B_TRUE : B_FALSE);
1958 
1959 		if (!no_port_bit && l4_cs_eq_bit && is_tcp_udp && !error_type) {
1960 			(void) hcksum_assoc(nmp, NULL, NULL, 0, 0, 0, 0,
1961 			    HCK_FULLCKSUM_OK | HCK_FULLCKSUM, 0);
1962 
1963 			HXGE_DEBUG_MSG((hxgep, RX_CTL,
1964 			    "==> hxge_receive_packet: Full tcp/udp cksum "
1965 			    "is_valid 0x%x multi %d error %d",
1966 			    is_valid, multi, error_type));
1967 		}
1968 	}
1969 
1970 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1971 	    "==> hxge_receive_packet: *mp 0x%016llx", *mp));
1972 
1973 	*multi_p = (multi == RCR_MULTI_MASK);
1974 
1975 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_receive_packet: "
1976 	    "multi %d nmp 0x%016llx *mp 0x%016llx *mp_cont 0x%016llx",
1977 	    *multi_p, nmp, *mp, *mp_cont));
1978 }
1979 
1980 static void
1981 hxge_rx_rbr_empty_recover(p_hxge_t hxgep, uint8_t channel)
1982 {
1983 	hpi_handle_t	handle;
1984 	p_rx_rcr_ring_t	rcrp;
1985 	p_rx_rbr_ring_t	rbrp;
1986 
1987 	rcrp = hxgep->rx_rcr_rings->rcr_rings[channel];
1988 	rbrp = rcrp->rx_rbr_p;
1989 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
1990 
1991 	/*
1992 	 * Wait for the channel to be quiet
1993 	 */
1994 	(void) hpi_rxdma_cfg_rdc_wait_for_qst(handle, channel);
1995 
1996 	/*
1997 	 * Post page will accumulate some buffers before re-enabling
1998 	 * the DMA channel.
1999 	 */
2000 
2001 	MUTEX_ENTER(&rbrp->post_lock);
2002 	if ((rbrp->rbb_max - rbrp->rbr_used) >= HXGE_RBR_EMPTY_THRESHOLD) {
2003 		hxge_rbr_empty_restore(hxgep, rbrp);
2004 	} else {
2005 		rbrp->rbr_is_empty = B_TRUE;
2006 	}
2007 	MUTEX_EXIT(&rbrp->post_lock);
2008 }
2009 
2010 /*ARGSUSED*/
2011 static hxge_status_t
2012 hxge_rx_err_evnts(p_hxge_t hxgep, uint_t index, p_hxge_ldv_t ldvp,
2013     rdc_stat_t cs)
2014 {
2015 	p_hxge_rx_ring_stats_t	rdc_stats;
2016 	hpi_handle_t		handle;
2017 	boolean_t		rxchan_fatal = B_FALSE;
2018 	uint8_t			channel;
2019 	hxge_status_t		status = HXGE_OK;
2020 	uint64_t		cs_val;
2021 
2022 	HXGE_DEBUG_MSG((hxgep, INT_CTL, "==> hxge_rx_err_evnts"));
2023 
2024 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
2025 	channel = ldvp->channel;
2026 
2027 	/* Clear the interrupts */
2028 	cs.bits.pktread = 0;
2029 	cs.bits.ptrread = 0;
2030 	cs_val = cs.value & RDC_STAT_WR1C;
2031 	RXDMA_REG_WRITE64(handle, RDC_STAT, channel, cs_val);
2032 
2033 	rdc_stats = &hxgep->statsp->rdc_stats[ldvp->vdma_index];
2034 
2035 	if (cs.bits.rbr_cpl_to) {
2036 		rdc_stats->rbr_tmout++;
2037 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2038 		    HXGE_FM_EREPORT_RDMC_RBR_CPL_TO);
2039 		rxchan_fatal = B_TRUE;
2040 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2041 		    "==> hxge_rx_err_evnts(channel %d): "
2042 		    "fatal error: rx_rbr_timeout", channel));
2043 	}
2044 
2045 	if ((cs.bits.rcr_shadow_par_err) || (cs.bits.rbr_prefetch_par_err)) {
2046 		(void) hpi_rxdma_ring_perr_stat_get(handle,
2047 		    &rdc_stats->errlog.pre_par, &rdc_stats->errlog.sha_par);
2048 	}
2049 
2050 	if (cs.bits.rcr_shadow_par_err) {
2051 		rdc_stats->rcr_sha_par++;
2052 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2053 		    HXGE_FM_EREPORT_RDMC_RCR_SHA_PAR);
2054 		rxchan_fatal = B_TRUE;
2055 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2056 		    "==> hxge_rx_err_evnts(channel %d): "
2057 		    "fatal error: rcr_shadow_par_err", channel));
2058 	}
2059 
2060 	if (cs.bits.rbr_prefetch_par_err) {
2061 		rdc_stats->rbr_pre_par++;
2062 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2063 		    HXGE_FM_EREPORT_RDMC_RBR_PRE_PAR);
2064 		rxchan_fatal = B_TRUE;
2065 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2066 		    "==> hxge_rx_err_evnts(channel %d): "
2067 		    "fatal error: rbr_prefetch_par_err", channel));
2068 	}
2069 
2070 	if (cs.bits.rbr_pre_empty) {
2071 		rdc_stats->rbr_pre_empty++;
2072 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2073 		    HXGE_FM_EREPORT_RDMC_RBR_PRE_EMPTY);
2074 		rxchan_fatal = B_TRUE;
2075 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2076 		    "==> hxge_rx_err_evnts(channel %d): "
2077 		    "fatal error: rbr_pre_empty", channel));
2078 	}
2079 
2080 	if (cs.bits.peu_resp_err) {
2081 		rdc_stats->peu_resp_err++;
2082 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2083 		    HXGE_FM_EREPORT_RDMC_PEU_RESP_ERR);
2084 		rxchan_fatal = B_TRUE;
2085 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2086 		    "==> hxge_rx_err_evnts(channel %d): "
2087 		    "fatal error: peu_resp_err", channel));
2088 	}
2089 
2090 	if (cs.bits.rcr_thres) {
2091 		rdc_stats->rcr_thres++;
2092 	}
2093 
2094 	if (cs.bits.rcr_to) {
2095 		rdc_stats->rcr_to++;
2096 	}
2097 
2098 	if (cs.bits.rcr_shadow_full) {
2099 		rdc_stats->rcr_shadow_full++;
2100 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2101 		    HXGE_FM_EREPORT_RDMC_RCR_SHA_FULL);
2102 		rxchan_fatal = B_TRUE;
2103 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2104 		    "==> hxge_rx_err_evnts(channel %d): "
2105 		    "fatal error: rcr_shadow_full", channel));
2106 	}
2107 
2108 	if (cs.bits.rcr_full) {
2109 		rdc_stats->rcrfull++;
2110 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2111 		    HXGE_FM_EREPORT_RDMC_RCRFULL);
2112 		rxchan_fatal = B_TRUE;
2113 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2114 		    "==> hxge_rx_err_evnts(channel %d): "
2115 		    "fatal error: rcrfull error", channel));
2116 	}
2117 
2118 	if (cs.bits.rbr_empty) {
2119 		rdc_stats->rbr_empty++;
2120 		hxge_rx_rbr_empty_recover(hxgep, channel);
2121 	}
2122 
2123 	if (cs.bits.rbr_full) {
2124 		rdc_stats->rbrfull++;
2125 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2126 		    HXGE_FM_EREPORT_RDMC_RBRFULL);
2127 		rxchan_fatal = B_TRUE;
2128 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2129 		    "==> hxge_rx_err_evnts(channel %d): "
2130 		    "fatal error: rbr_full error", channel));
2131 	}
2132 
2133 	if (rxchan_fatal) {
2134 		p_rx_rcr_ring_t	rcrp;
2135 		p_rx_rbr_ring_t rbrp;
2136 
2137 		rcrp = hxgep->rx_rcr_rings->rcr_rings[channel];
2138 		rbrp = rcrp->rx_rbr_p;
2139 
2140 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2141 		    " hxge_rx_err_evnts: fatal error on Channel #%d\n",
2142 		    channel));
2143 		MUTEX_ENTER(&rbrp->post_lock);
2144 		/* This function needs to be inside the post_lock */
2145 		status = hxge_rxdma_fatal_err_recover(hxgep, channel);
2146 		MUTEX_EXIT(&rbrp->post_lock);
2147 		if (status == HXGE_OK) {
2148 			FM_SERVICE_RESTORED(hxgep);
2149 		}
2150 	}
2151 	HXGE_DEBUG_MSG((hxgep, INT_CTL, "<== hxge_rx_err_evnts"));
2152 
2153 	return (status);
2154 }
2155 
2156 static hxge_status_t
2157 hxge_map_rxdma(p_hxge_t hxgep)
2158 {
2159 	int			i, ndmas;
2160 	uint16_t		channel;
2161 	p_rx_rbr_rings_t	rx_rbr_rings;
2162 	p_rx_rbr_ring_t		*rbr_rings;
2163 	p_rx_rcr_rings_t	rx_rcr_rings;
2164 	p_rx_rcr_ring_t		*rcr_rings;
2165 	p_rx_mbox_areas_t	rx_mbox_areas_p;
2166 	p_rx_mbox_t		*rx_mbox_p;
2167 	p_hxge_dma_pool_t	dma_buf_poolp;
2168 	p_hxge_dma_common_t	*dma_buf_p;
2169 	p_hxge_dma_pool_t	dma_rbr_cntl_poolp;
2170 	p_hxge_dma_common_t	*dma_rbr_cntl_p;
2171 	p_hxge_dma_pool_t	dma_rcr_cntl_poolp;
2172 	p_hxge_dma_common_t	*dma_rcr_cntl_p;
2173 	p_hxge_dma_pool_t	dma_mbox_cntl_poolp;
2174 	p_hxge_dma_common_t	*dma_mbox_cntl_p;
2175 	uint32_t		*num_chunks;
2176 	hxge_status_t		status = HXGE_OK;
2177 
2178 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_map_rxdma"));
2179 
2180 	dma_buf_poolp = hxgep->rx_buf_pool_p;
2181 	dma_rbr_cntl_poolp = hxgep->rx_rbr_cntl_pool_p;
2182 	dma_rcr_cntl_poolp = hxgep->rx_rcr_cntl_pool_p;
2183 	dma_mbox_cntl_poolp = hxgep->rx_mbox_cntl_pool_p;
2184 
2185 	if (!dma_buf_poolp->buf_allocated ||
2186 	    !dma_rbr_cntl_poolp->buf_allocated ||
2187 	    !dma_rcr_cntl_poolp->buf_allocated ||
2188 	    !dma_mbox_cntl_poolp->buf_allocated) {
2189 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2190 		    "<== hxge_map_rxdma: buf not allocated"));
2191 		return (HXGE_ERROR);
2192 	}
2193 
2194 	ndmas = dma_buf_poolp->ndmas;
2195 	if (!ndmas) {
2196 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
2197 		    "<== hxge_map_rxdma: no dma allocated"));
2198 		return (HXGE_ERROR);
2199 	}
2200 
2201 	num_chunks = dma_buf_poolp->num_chunks;
2202 	dma_buf_p = dma_buf_poolp->dma_buf_pool_p;
2203 	dma_rbr_cntl_p = dma_rbr_cntl_poolp->dma_buf_pool_p;
2204 	dma_rcr_cntl_p = dma_rcr_cntl_poolp->dma_buf_pool_p;
2205 	dma_mbox_cntl_p = dma_mbox_cntl_poolp->dma_buf_pool_p;
2206 
2207 	rx_rbr_rings = (p_rx_rbr_rings_t)
2208 	    KMEM_ZALLOC(sizeof (rx_rbr_rings_t), KM_SLEEP);
2209 	rbr_rings = (p_rx_rbr_ring_t *)KMEM_ZALLOC(
2210 	    sizeof (p_rx_rbr_ring_t) * ndmas, KM_SLEEP);
2211 
2212 	rx_rcr_rings = (p_rx_rcr_rings_t)
2213 	    KMEM_ZALLOC(sizeof (rx_rcr_rings_t), KM_SLEEP);
2214 	rcr_rings = (p_rx_rcr_ring_t *)KMEM_ZALLOC(
2215 	    sizeof (p_rx_rcr_ring_t) * ndmas, KM_SLEEP);
2216 
2217 	rx_mbox_areas_p = (p_rx_mbox_areas_t)
2218 	    KMEM_ZALLOC(sizeof (rx_mbox_areas_t), KM_SLEEP);
2219 	rx_mbox_p = (p_rx_mbox_t *)KMEM_ZALLOC(
2220 	    sizeof (p_rx_mbox_t) * ndmas, KM_SLEEP);
2221 
2222 	/*
2223 	 * Timeout should be set based on the system clock divider.
2224 	 * The following timeout value of 1 assumes that the
2225 	 * granularity (1000) is 3 microseconds running at 300MHz.
2226 	 */
2227 
2228 	hxgep->intr_threshold = RXDMA_RCR_PTHRES_DEFAULT;
2229 	hxgep->intr_timeout = RXDMA_RCR_TO_DEFAULT;
2230 
2231 	/*
2232 	 * Map descriptors from the buffer polls for each dam channel.
2233 	 */
2234 	for (i = 0; i < ndmas; i++) {
2235 		/*
2236 		 * Set up and prepare buffer blocks, descriptors and mailbox.
2237 		 */
2238 		channel = ((p_hxge_dma_common_t)dma_buf_p[i])->dma_channel;
2239 		status = hxge_map_rxdma_channel(hxgep, channel,
2240 		    (p_hxge_dma_common_t *)&dma_buf_p[i],
2241 		    (p_rx_rbr_ring_t *)&rbr_rings[i],
2242 		    num_chunks[i],
2243 		    (p_hxge_dma_common_t *)&dma_rbr_cntl_p[i],
2244 		    (p_hxge_dma_common_t *)&dma_rcr_cntl_p[i],
2245 		    (p_hxge_dma_common_t *)&dma_mbox_cntl_p[i],
2246 		    (p_rx_rcr_ring_t *)&rcr_rings[i],
2247 		    (p_rx_mbox_t *)&rx_mbox_p[i]);
2248 		if (status != HXGE_OK) {
2249 			goto hxge_map_rxdma_fail1;
2250 		}
2251 		rbr_rings[i]->index = (uint16_t)i;
2252 		rcr_rings[i]->index = (uint16_t)i;
2253 		rcr_rings[i]->rdc_stats = &hxgep->statsp->rdc_stats[i];
2254 	}
2255 
2256 	rx_rbr_rings->ndmas = rx_rcr_rings->ndmas = ndmas;
2257 	rx_rbr_rings->rbr_rings = rbr_rings;
2258 	hxgep->rx_rbr_rings = rx_rbr_rings;
2259 	rx_rcr_rings->rcr_rings = rcr_rings;
2260 	hxgep->rx_rcr_rings = rx_rcr_rings;
2261 
2262 	rx_mbox_areas_p->rxmbox_areas = rx_mbox_p;
2263 	hxgep->rx_mbox_areas_p = rx_mbox_areas_p;
2264 
2265 	goto hxge_map_rxdma_exit;
2266 
2267 hxge_map_rxdma_fail1:
2268 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2269 	    "==> hxge_map_rxdma: unmap rbr,rcr (status 0x%x channel %d i %d)",
2270 	    status, channel, i));
2271 	i--;
2272 	for (; i >= 0; i--) {
2273 		channel = ((p_hxge_dma_common_t)dma_buf_p[i])->dma_channel;
2274 		hxge_unmap_rxdma_channel(hxgep, channel,
2275 		    rbr_rings[i], rcr_rings[i], rx_mbox_p[i]);
2276 	}
2277 
2278 	KMEM_FREE(rbr_rings, sizeof (p_rx_rbr_ring_t) * ndmas);
2279 	KMEM_FREE(rx_rbr_rings, sizeof (rx_rbr_rings_t));
2280 	KMEM_FREE(rcr_rings, sizeof (p_rx_rcr_ring_t) * ndmas);
2281 	KMEM_FREE(rx_rcr_rings, sizeof (rx_rcr_rings_t));
2282 	KMEM_FREE(rx_mbox_p, sizeof (p_rx_mbox_t) * ndmas);
2283 	KMEM_FREE(rx_mbox_areas_p, sizeof (rx_mbox_areas_t));
2284 
2285 hxge_map_rxdma_exit:
2286 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2287 	    "<== hxge_map_rxdma: (status 0x%x channel %d)", status, channel));
2288 
2289 	return (status);
2290 }
2291 
2292 static void
2293 hxge_unmap_rxdma(p_hxge_t hxgep)
2294 {
2295 	int			i, ndmas;
2296 	uint16_t		channel;
2297 	p_rx_rbr_rings_t	rx_rbr_rings;
2298 	p_rx_rbr_ring_t		*rbr_rings;
2299 	p_rx_rcr_rings_t	rx_rcr_rings;
2300 	p_rx_rcr_ring_t		*rcr_rings;
2301 	p_rx_mbox_areas_t	rx_mbox_areas_p;
2302 	p_rx_mbox_t		*rx_mbox_p;
2303 	p_hxge_dma_pool_t	dma_buf_poolp;
2304 	p_hxge_dma_pool_t	dma_rbr_cntl_poolp;
2305 	p_hxge_dma_pool_t	dma_rcr_cntl_poolp;
2306 	p_hxge_dma_pool_t	dma_mbox_cntl_poolp;
2307 	p_hxge_dma_common_t	*dma_buf_p;
2308 
2309 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_unmap_rxdma"));
2310 
2311 	dma_buf_poolp = hxgep->rx_buf_pool_p;
2312 	dma_rbr_cntl_poolp = hxgep->rx_rbr_cntl_pool_p;
2313 	dma_rcr_cntl_poolp = hxgep->rx_rcr_cntl_pool_p;
2314 	dma_mbox_cntl_poolp = hxgep->rx_mbox_cntl_pool_p;
2315 
2316 	if (!dma_buf_poolp->buf_allocated ||
2317 	    !dma_rbr_cntl_poolp->buf_allocated ||
2318 	    !dma_rcr_cntl_poolp->buf_allocated ||
2319 	    !dma_mbox_cntl_poolp->buf_allocated) {
2320 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2321 		    "<== hxge_unmap_rxdma: NULL buf pointers"));
2322 		return;
2323 	}
2324 
2325 	rx_rbr_rings = hxgep->rx_rbr_rings;
2326 	rx_rcr_rings = hxgep->rx_rcr_rings;
2327 	if (rx_rbr_rings == NULL || rx_rcr_rings == NULL) {
2328 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2329 		    "<== hxge_unmap_rxdma: NULL pointers"));
2330 		return;
2331 	}
2332 
2333 	ndmas = rx_rbr_rings->ndmas;
2334 	if (!ndmas) {
2335 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2336 		    "<== hxge_unmap_rxdma: no channel"));
2337 		return;
2338 	}
2339 
2340 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2341 	    "==> hxge_unmap_rxdma (ndmas %d)", ndmas));
2342 
2343 	rbr_rings = rx_rbr_rings->rbr_rings;
2344 	rcr_rings = rx_rcr_rings->rcr_rings;
2345 	rx_mbox_areas_p = hxgep->rx_mbox_areas_p;
2346 	rx_mbox_p = rx_mbox_areas_p->rxmbox_areas;
2347 	dma_buf_p = dma_buf_poolp->dma_buf_pool_p;
2348 
2349 	for (i = 0; i < ndmas; i++) {
2350 		channel = ((p_hxge_dma_common_t)dma_buf_p[i])->dma_channel;
2351 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2352 		    "==> hxge_unmap_rxdma (ndmas %d) channel %d",
2353 		    ndmas, channel));
2354 		(void) hxge_unmap_rxdma_channel(hxgep, channel,
2355 		    (p_rx_rbr_ring_t)rbr_rings[i],
2356 		    (p_rx_rcr_ring_t)rcr_rings[i],
2357 		    (p_rx_mbox_t)rx_mbox_p[i]);
2358 	}
2359 
2360 	KMEM_FREE(rx_rbr_rings, sizeof (rx_rbr_rings_t));
2361 	KMEM_FREE(rbr_rings, sizeof (p_rx_rbr_ring_t) * ndmas);
2362 	KMEM_FREE(rx_rcr_rings, sizeof (rx_rcr_rings_t));
2363 	KMEM_FREE(rcr_rings, sizeof (p_rx_rcr_ring_t) * ndmas);
2364 	KMEM_FREE(rx_mbox_areas_p, sizeof (rx_mbox_areas_t));
2365 	KMEM_FREE(rx_mbox_p, sizeof (p_rx_mbox_t) * ndmas);
2366 
2367 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "<== hxge_unmap_rxdma"));
2368 }
2369 
2370 hxge_status_t
2371 hxge_map_rxdma_channel(p_hxge_t hxgep, uint16_t channel,
2372     p_hxge_dma_common_t *dma_buf_p, p_rx_rbr_ring_t *rbr_p,
2373     uint32_t num_chunks, p_hxge_dma_common_t *dma_rbr_cntl_p,
2374     p_hxge_dma_common_t *dma_rcr_cntl_p, p_hxge_dma_common_t *dma_mbox_cntl_p,
2375     p_rx_rcr_ring_t *rcr_p, p_rx_mbox_t *rx_mbox_p)
2376 {
2377 	int status = HXGE_OK;
2378 
2379 	/*
2380 	 * Set up and prepare buffer blocks, descriptors and mailbox.
2381 	 */
2382 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2383 	    "==> hxge_map_rxdma_channel (channel %d)", channel));
2384 
2385 	/*
2386 	 * Receive buffer blocks
2387 	 */
2388 	status = hxge_map_rxdma_channel_buf_ring(hxgep, channel,
2389 	    dma_buf_p, rbr_p, num_chunks);
2390 	if (status != HXGE_OK) {
2391 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2392 		    "==> hxge_map_rxdma_channel (channel %d): "
2393 		    "map buffer failed 0x%x", channel, status));
2394 		goto hxge_map_rxdma_channel_exit;
2395 	}
2396 
2397 	/*
2398 	 * Receive block ring, completion ring and mailbox.
2399 	 */
2400 	status = hxge_map_rxdma_channel_cfg_ring(hxgep, channel,
2401 	    dma_rbr_cntl_p, dma_rcr_cntl_p, dma_mbox_cntl_p,
2402 	    rbr_p, rcr_p, rx_mbox_p);
2403 	if (status != HXGE_OK) {
2404 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2405 		    "==> hxge_map_rxdma_channel (channel %d): "
2406 		    "map config failed 0x%x", channel, status));
2407 		goto hxge_map_rxdma_channel_fail2;
2408 	}
2409 	goto hxge_map_rxdma_channel_exit;
2410 
2411 hxge_map_rxdma_channel_fail3:
2412 	/* Free rbr, rcr */
2413 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2414 	    "==> hxge_map_rxdma_channel: free rbr/rcr (status 0x%x channel %d)",
2415 	    status, channel));
2416 	hxge_unmap_rxdma_channel_cfg_ring(hxgep, *rcr_p, *rx_mbox_p);
2417 
2418 hxge_map_rxdma_channel_fail2:
2419 	/* Free buffer blocks */
2420 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2421 	    "==> hxge_map_rxdma_channel: free rx buffers"
2422 	    "(hxgep 0x%x status 0x%x channel %d)",
2423 	    hxgep, status, channel));
2424 	hxge_unmap_rxdma_channel_buf_ring(hxgep, *rbr_p);
2425 
2426 	status = HXGE_ERROR;
2427 
2428 hxge_map_rxdma_channel_exit:
2429 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2430 	    "<== hxge_map_rxdma_channel: (hxgep 0x%x status 0x%x channel %d)",
2431 	    hxgep, status, channel));
2432 
2433 	return (status);
2434 }
2435 
2436 /*ARGSUSED*/
2437 static void
2438 hxge_unmap_rxdma_channel(p_hxge_t hxgep, uint16_t channel,
2439     p_rx_rbr_ring_t rbr_p, p_rx_rcr_ring_t rcr_p, p_rx_mbox_t rx_mbox_p)
2440 {
2441 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2442 	    "==> hxge_unmap_rxdma_channel (channel %d)", channel));
2443 
2444 	/*
2445 	 * unmap receive block ring, completion ring and mailbox.
2446 	 */
2447 	(void) hxge_unmap_rxdma_channel_cfg_ring(hxgep, rcr_p, rx_mbox_p);
2448 
2449 	/* unmap buffer blocks */
2450 	(void) hxge_unmap_rxdma_channel_buf_ring(hxgep, rbr_p);
2451 
2452 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "<== hxge_unmap_rxdma_channel"));
2453 }
2454 
2455 /*ARGSUSED*/
2456 static hxge_status_t
2457 hxge_map_rxdma_channel_cfg_ring(p_hxge_t hxgep, uint16_t dma_channel,
2458     p_hxge_dma_common_t *dma_rbr_cntl_p, p_hxge_dma_common_t *dma_rcr_cntl_p,
2459     p_hxge_dma_common_t *dma_mbox_cntl_p, p_rx_rbr_ring_t *rbr_p,
2460     p_rx_rcr_ring_t *rcr_p, p_rx_mbox_t *rx_mbox_p)
2461 {
2462 	p_rx_rbr_ring_t 	rbrp;
2463 	p_rx_rcr_ring_t 	rcrp;
2464 	p_rx_mbox_t 		mboxp;
2465 	p_hxge_dma_common_t 	cntl_dmap;
2466 	p_hxge_dma_common_t 	dmap;
2467 	p_rx_msg_t 		*rx_msg_ring;
2468 	p_rx_msg_t 		rx_msg_p;
2469 	rdc_rbr_cfg_a_t		*rcfga_p;
2470 	rdc_rbr_cfg_b_t		*rcfgb_p;
2471 	rdc_rcr_cfg_a_t		*cfga_p;
2472 	rdc_rcr_cfg_b_t		*cfgb_p;
2473 	rdc_rx_cfg1_t		*cfig1_p;
2474 	rdc_rx_cfg2_t		*cfig2_p;
2475 	rdc_rbr_kick_t		*kick_p;
2476 	uint32_t		dmaaddrp;
2477 	uint32_t		*rbr_vaddrp;
2478 	uint32_t		bkaddr;
2479 	hxge_status_t		status = HXGE_OK;
2480 	int			i;
2481 	uint32_t 		hxge_port_rcr_size;
2482 
2483 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2484 	    "==> hxge_map_rxdma_channel_cfg_ring"));
2485 
2486 	cntl_dmap = *dma_rbr_cntl_p;
2487 
2488 	/*
2489 	 * Map in the receive block ring
2490 	 */
2491 	rbrp = *rbr_p;
2492 	dmap = (p_hxge_dma_common_t)&rbrp->rbr_desc;
2493 	hxge_setup_dma_common(dmap, cntl_dmap, rbrp->rbb_max, 4);
2494 
2495 	/*
2496 	 * Zero out buffer block ring descriptors.
2497 	 */
2498 	bzero((caddr_t)dmap->kaddrp, dmap->alength);
2499 
2500 	rcfga_p = &(rbrp->rbr_cfga);
2501 	rcfgb_p = &(rbrp->rbr_cfgb);
2502 	kick_p = &(rbrp->rbr_kick);
2503 	rcfga_p->value = 0;
2504 	rcfgb_p->value = 0;
2505 	kick_p->value = 0;
2506 	rbrp->rbr_addr = dmap->dma_cookie.dmac_laddress;
2507 	rcfga_p->value = (rbrp->rbr_addr &
2508 	    (RBR_CFIG_A_STDADDR_MASK | RBR_CFIG_A_STDADDR_BASE_MASK));
2509 	rcfga_p->value |= ((uint64_t)rbrp->rbb_max << RBR_CFIG_A_LEN_SHIFT);
2510 
2511 	/* XXXX: how to choose packet buffer sizes */
2512 	rcfgb_p->bits.bufsz0 = rbrp->pkt_buf_size0;
2513 	rcfgb_p->bits.vld0 = 1;
2514 	rcfgb_p->bits.bufsz1 = rbrp->pkt_buf_size1;
2515 	rcfgb_p->bits.vld1 = 1;
2516 	rcfgb_p->bits.bufsz2 = rbrp->pkt_buf_size2;
2517 	rcfgb_p->bits.vld2 = 1;
2518 	rcfgb_p->bits.bksize = hxgep->rx_bksize_code;
2519 
2520 	/*
2521 	 * For each buffer block, enter receive block address to the ring.
2522 	 */
2523 	rbr_vaddrp = (uint32_t *)dmap->kaddrp;
2524 	rbrp->rbr_desc_vp = (uint32_t *)dmap->kaddrp;
2525 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2526 	    "==> hxge_map_rxdma_channel_cfg_ring: channel %d "
2527 	    "rbr_vaddrp $%p", dma_channel, rbr_vaddrp));
2528 
2529 	rx_msg_ring = rbrp->rx_msg_ring;
2530 	for (i = 0; i < rbrp->tnblocks; i++) {
2531 		rx_msg_p = rx_msg_ring[i];
2532 		rx_msg_p->hxgep = hxgep;
2533 		rx_msg_p->rx_rbr_p = rbrp;
2534 		bkaddr = (uint32_t)
2535 		    ((rx_msg_p->buf_dma.dma_cookie.dmac_laddress >>
2536 		    RBR_BKADDR_SHIFT));
2537 		rx_msg_p->free = B_FALSE;
2538 		rx_msg_p->max_usage_cnt = 0xbaddcafe;
2539 
2540 		*rbr_vaddrp++ = bkaddr;
2541 	}
2542 
2543 	kick_p->bits.bkadd = rbrp->rbb_max;
2544 	rbrp->rbr_wr_index = (rbrp->rbb_max - 1);
2545 
2546 	rbrp->rbr_rd_index = 0;
2547 
2548 	rbrp->rbr_consumed = 0;
2549 	rbrp->rbr_used = 0;
2550 	rbrp->rbr_use_bcopy = B_TRUE;
2551 	rbrp->rbr_bufsize_type = RCR_PKTBUFSZ_0;
2552 
2553 	/*
2554 	 * Do bcopy on packets greater than bcopy size once the lo threshold is
2555 	 * reached. This lo threshold should be less than the hi threshold.
2556 	 *
2557 	 * Do bcopy on every packet once the hi threshold is reached.
2558 	 */
2559 	if (hxge_rx_threshold_lo >= hxge_rx_threshold_hi) {
2560 		/* default it to use hi */
2561 		hxge_rx_threshold_lo = hxge_rx_threshold_hi;
2562 	}
2563 	if (hxge_rx_buf_size_type > HXGE_RBR_TYPE2) {
2564 		hxge_rx_buf_size_type = HXGE_RBR_TYPE2;
2565 	}
2566 	rbrp->rbr_bufsize_type = hxge_rx_buf_size_type;
2567 
2568 	switch (hxge_rx_threshold_hi) {
2569 	default:
2570 	case HXGE_RX_COPY_NONE:
2571 		/* Do not do bcopy at all */
2572 		rbrp->rbr_use_bcopy = B_FALSE;
2573 		rbrp->rbr_threshold_hi = rbrp->rbb_max;
2574 		break;
2575 
2576 	case HXGE_RX_COPY_1:
2577 	case HXGE_RX_COPY_2:
2578 	case HXGE_RX_COPY_3:
2579 	case HXGE_RX_COPY_4:
2580 	case HXGE_RX_COPY_5:
2581 	case HXGE_RX_COPY_6:
2582 	case HXGE_RX_COPY_7:
2583 		rbrp->rbr_threshold_hi =
2584 		    rbrp->rbb_max * (hxge_rx_threshold_hi) /
2585 		    HXGE_RX_BCOPY_SCALE;
2586 		break;
2587 
2588 	case HXGE_RX_COPY_ALL:
2589 		rbrp->rbr_threshold_hi = 0;
2590 		break;
2591 	}
2592 
2593 	switch (hxge_rx_threshold_lo) {
2594 	default:
2595 	case HXGE_RX_COPY_NONE:
2596 		/* Do not do bcopy at all */
2597 		if (rbrp->rbr_use_bcopy) {
2598 			rbrp->rbr_use_bcopy = B_FALSE;
2599 		}
2600 		rbrp->rbr_threshold_lo = rbrp->rbb_max;
2601 		break;
2602 
2603 	case HXGE_RX_COPY_1:
2604 	case HXGE_RX_COPY_2:
2605 	case HXGE_RX_COPY_3:
2606 	case HXGE_RX_COPY_4:
2607 	case HXGE_RX_COPY_5:
2608 	case HXGE_RX_COPY_6:
2609 	case HXGE_RX_COPY_7:
2610 		rbrp->rbr_threshold_lo =
2611 		    rbrp->rbb_max * (hxge_rx_threshold_lo) /
2612 		    HXGE_RX_BCOPY_SCALE;
2613 		break;
2614 
2615 	case HXGE_RX_COPY_ALL:
2616 		rbrp->rbr_threshold_lo = 0;
2617 		break;
2618 	}
2619 
2620 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
2621 	    "hxge_map_rxdma_channel_cfg_ring: channel %d rbb_max %d "
2622 	    "rbrp->rbr_bufsize_type %d rbb_threshold_hi %d "
2623 	    "rbb_threshold_lo %d",
2624 	    dma_channel, rbrp->rbb_max, rbrp->rbr_bufsize_type,
2625 	    rbrp->rbr_threshold_hi, rbrp->rbr_threshold_lo));
2626 
2627 	/* Map in the receive completion ring */
2628 	rcrp = (p_rx_rcr_ring_t)KMEM_ZALLOC(sizeof (rx_rcr_ring_t), KM_SLEEP);
2629 	rcrp->rdc = dma_channel;
2630 	rcrp->hxgep = hxgep;
2631 
2632 	hxge_port_rcr_size = hxgep->hxge_port_rcr_size;
2633 	rcrp->comp_size = hxge_port_rcr_size;
2634 	rcrp->comp_wrap_mask = hxge_port_rcr_size - 1;
2635 
2636 	rcrp->max_receive_pkts = hxge_max_rx_pkts;
2637 
2638 	cntl_dmap = *dma_rcr_cntl_p;
2639 
2640 	dmap = (p_hxge_dma_common_t)&rcrp->rcr_desc;
2641 	hxge_setup_dma_common(dmap, cntl_dmap, rcrp->comp_size,
2642 	    sizeof (rcr_entry_t));
2643 	rcrp->comp_rd_index = 0;
2644 	rcrp->comp_wt_index = 0;
2645 	rcrp->rcr_desc_rd_head_p = rcrp->rcr_desc_first_p =
2646 	    (p_rcr_entry_t)DMA_COMMON_VPTR(rcrp->rcr_desc);
2647 #if defined(__i386)
2648 	rcrp->rcr_desc_rd_head_pp = rcrp->rcr_desc_first_pp =
2649 	    (p_rcr_entry_t)(uint32_t)DMA_COMMON_IOADDR(rcrp->rcr_desc);
2650 #else
2651 	rcrp->rcr_desc_rd_head_pp = rcrp->rcr_desc_first_pp =
2652 	    (p_rcr_entry_t)DMA_COMMON_IOADDR(rcrp->rcr_desc);
2653 #endif
2654 	rcrp->rcr_desc_last_p = rcrp->rcr_desc_rd_head_p +
2655 	    (hxge_port_rcr_size - 1);
2656 	rcrp->rcr_desc_last_pp = rcrp->rcr_desc_rd_head_pp +
2657 	    (hxge_port_rcr_size - 1);
2658 
2659 	rcrp->rcr_tail_begin = DMA_COMMON_IOADDR(rcrp->rcr_desc);
2660 	rcrp->rcr_tail_begin = (rcrp->rcr_tail_begin & 0x7ffffULL) >> 3;
2661 
2662 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2663 	    "==> hxge_map_rxdma_channel_cfg_ring: channel %d "
2664 	    "rbr_vaddrp $%p rcr_desc_rd_head_p $%p "
2665 	    "rcr_desc_rd_head_pp $%p rcr_desc_rd_last_p $%p "
2666 	    "rcr_desc_rd_last_pp $%p ",
2667 	    dma_channel, rbr_vaddrp, rcrp->rcr_desc_rd_head_p,
2668 	    rcrp->rcr_desc_rd_head_pp, rcrp->rcr_desc_last_p,
2669 	    rcrp->rcr_desc_last_pp));
2670 
2671 	/*
2672 	 * Zero out buffer block ring descriptors.
2673 	 */
2674 	bzero((caddr_t)dmap->kaddrp, dmap->alength);
2675 	rcrp->intr_timeout = hxgep->intr_timeout;
2676 	rcrp->intr_threshold = hxgep->intr_threshold;
2677 	rcrp->full_hdr_flag = B_FALSE;
2678 	rcrp->sw_priv_hdr_len = 0;
2679 
2680 	cfga_p = &(rcrp->rcr_cfga);
2681 	cfgb_p = &(rcrp->rcr_cfgb);
2682 	cfga_p->value = 0;
2683 	cfgb_p->value = 0;
2684 	rcrp->rcr_addr = dmap->dma_cookie.dmac_laddress;
2685 
2686 	cfga_p->value = (rcrp->rcr_addr &
2687 	    (RCRCFIG_A_STADDR_MASK | RCRCFIG_A_STADDR_BASE_MASK));
2688 
2689 	cfga_p->value |= ((uint64_t)rcrp->comp_size << RCRCFIG_A_LEN_SHIF);
2690 
2691 	/*
2692 	 * Timeout should be set based on the system clock divider. The
2693 	 * following timeout value of 1 assumes that the granularity (1000) is
2694 	 * 3 microseconds running at 300MHz.
2695 	 */
2696 	cfgb_p->bits.pthres = rcrp->intr_threshold;
2697 	cfgb_p->bits.timeout = rcrp->intr_timeout;
2698 	cfgb_p->bits.entout = 1;
2699 
2700 	/* Map in the mailbox */
2701 	cntl_dmap = *dma_mbox_cntl_p;
2702 	mboxp = (p_rx_mbox_t)KMEM_ZALLOC(sizeof (rx_mbox_t), KM_SLEEP);
2703 	dmap = (p_hxge_dma_common_t)&mboxp->rx_mbox;
2704 	hxge_setup_dma_common(dmap, cntl_dmap, 1, sizeof (rxdma_mailbox_t));
2705 	cfig1_p = (rdc_rx_cfg1_t *)&mboxp->rx_cfg1;
2706 	cfig2_p = (rdc_rx_cfg2_t *)&mboxp->rx_cfg2;
2707 	cfig1_p->value = cfig2_p->value = 0;
2708 
2709 	mboxp->mbox_addr = dmap->dma_cookie.dmac_laddress;
2710 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2711 	    "==> hxge_map_rxdma_channel_cfg_ring: "
2712 	    "channel %d cfg1 0x%016llx cfig2 0x%016llx cookie 0x%016llx",
2713 	    dma_channel, cfig1_p->value, cfig2_p->value,
2714 	    mboxp->mbox_addr));
2715 
2716 	dmaaddrp = (uint32_t)((dmap->dma_cookie.dmac_laddress >> 32) & 0xfff);
2717 	cfig1_p->bits.mbaddr_h = dmaaddrp;
2718 
2719 	dmaaddrp = (uint32_t)(dmap->dma_cookie.dmac_laddress & 0xffffffff);
2720 	dmaaddrp = (uint32_t)(dmap->dma_cookie.dmac_laddress &
2721 	    RXDMA_CFIG2_MBADDR_L_MASK);
2722 
2723 	cfig2_p->bits.mbaddr_l = (dmaaddrp >> RXDMA_CFIG2_MBADDR_L_SHIFT);
2724 
2725 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2726 	    "==> hxge_map_rxdma_channel_cfg_ring: channel %d damaddrp $%p "
2727 	    "cfg1 0x%016llx cfig2 0x%016llx",
2728 	    dma_channel, dmaaddrp, cfig1_p->value, cfig2_p->value));
2729 
2730 	cfig2_p->bits.full_hdr = rcrp->full_hdr_flag;
2731 	cfig2_p->bits.offset = rcrp->sw_priv_hdr_len;
2732 
2733 	rbrp->rx_rcr_p = rcrp;
2734 	rcrp->rx_rbr_p = rbrp;
2735 	*rcr_p = rcrp;
2736 	*rx_mbox_p = mboxp;
2737 
2738 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2739 	    "<== hxge_map_rxdma_channel_cfg_ring status 0x%08x", status));
2740 	return (status);
2741 }
2742 
2743 /*ARGSUSED*/
2744 static void
2745 hxge_unmap_rxdma_channel_cfg_ring(p_hxge_t hxgep,
2746     p_rx_rcr_ring_t rcr_p, p_rx_mbox_t rx_mbox_p)
2747 {
2748 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2749 	    "==> hxge_unmap_rxdma_channel_cfg_ring: channel %d", rcr_p->rdc));
2750 
2751 	KMEM_FREE(rcr_p, sizeof (rx_rcr_ring_t));
2752 	KMEM_FREE(rx_mbox_p, sizeof (rx_mbox_t));
2753 
2754 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2755 	    "<== hxge_unmap_rxdma_channel_cfg_ring"));
2756 }
2757 
2758 static hxge_status_t
2759 hxge_map_rxdma_channel_buf_ring(p_hxge_t hxgep, uint16_t channel,
2760     p_hxge_dma_common_t *dma_buf_p,
2761     p_rx_rbr_ring_t *rbr_p, uint32_t num_chunks)
2762 {
2763 	p_rx_rbr_ring_t		rbrp;
2764 	p_hxge_dma_common_t	dma_bufp, tmp_bufp;
2765 	p_rx_msg_t		*rx_msg_ring;
2766 	p_rx_msg_t		rx_msg_p;
2767 	p_mblk_t		mblk_p;
2768 
2769 	rxring_info_t *ring_info;
2770 	hxge_status_t status = HXGE_OK;
2771 	int i, j, index;
2772 	uint32_t size, bsize, nblocks, nmsgs;
2773 
2774 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2775 	    "==> hxge_map_rxdma_channel_buf_ring: channel %d", channel));
2776 
2777 	dma_bufp = tmp_bufp = *dma_buf_p;
2778 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2779 	    " hxge_map_rxdma_channel_buf_ring: channel %d to map %d "
2780 	    "chunks bufp 0x%016llx", channel, num_chunks, dma_bufp));
2781 
2782 	nmsgs = 0;
2783 	for (i = 0; i < num_chunks; i++, tmp_bufp++) {
2784 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2785 		    "==> hxge_map_rxdma_channel_buf_ring: channel %d "
2786 		    "bufp 0x%016llx nblocks %d nmsgs %d",
2787 		    channel, tmp_bufp, tmp_bufp->nblocks, nmsgs));
2788 		nmsgs += tmp_bufp->nblocks;
2789 	}
2790 	if (!nmsgs) {
2791 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2792 		    "<== hxge_map_rxdma_channel_buf_ring: channel %d "
2793 		    "no msg blocks", channel));
2794 		status = HXGE_ERROR;
2795 		goto hxge_map_rxdma_channel_buf_ring_exit;
2796 	}
2797 	rbrp = (p_rx_rbr_ring_t)KMEM_ZALLOC(sizeof (rx_rbr_ring_t), KM_SLEEP);
2798 
2799 	size = nmsgs * sizeof (p_rx_msg_t);
2800 	rx_msg_ring = KMEM_ZALLOC(size, KM_SLEEP);
2801 	ring_info = (rxring_info_t *)KMEM_ZALLOC(sizeof (rxring_info_t),
2802 	    KM_SLEEP);
2803 
2804 	MUTEX_INIT(&rbrp->lock, NULL, MUTEX_DRIVER,
2805 	    (void *) hxgep->interrupt_cookie);
2806 	MUTEX_INIT(&rbrp->post_lock, NULL, MUTEX_DRIVER,
2807 	    (void *) hxgep->interrupt_cookie);
2808 
2809 	rbrp->rdc = channel;
2810 	rbrp->num_blocks = num_chunks;
2811 	rbrp->tnblocks = nmsgs;
2812 	rbrp->rbb_max = nmsgs;
2813 	rbrp->rbr_max_size = nmsgs;
2814 	rbrp->rbr_wrap_mask = (rbrp->rbb_max - 1);
2815 
2816 	/*
2817 	 * Buffer sizes suggested by NIU architect. 256, 512 and 2K.
2818 	 */
2819 
2820 	switch (hxgep->rx_bksize_code) {
2821 	case RBR_BKSIZE_4K:
2822 		rbrp->pkt_buf_size0 = RBR_BUFSZ0_256B;
2823 		rbrp->pkt_buf_size0_bytes = RBR_BUFSZ0_256_BYTES;
2824 		rbrp->hpi_pkt_buf_size0 = SIZE_256B;
2825 		break;
2826 	case RBR_BKSIZE_8K:
2827 		/* Use 512 to avoid possible rcr_full condition */
2828 		rbrp->pkt_buf_size0 = RBR_BUFSZ0_512B;
2829 		rbrp->pkt_buf_size0_bytes = RBR_BUFSZ0_512_BYTES;
2830 		rbrp->hpi_pkt_buf_size0 = SIZE_512B;
2831 		break;
2832 	}
2833 
2834 	rbrp->pkt_buf_size1 = RBR_BUFSZ1_1K;
2835 	rbrp->pkt_buf_size1_bytes = RBR_BUFSZ1_1K_BYTES;
2836 	rbrp->hpi_pkt_buf_size1 = SIZE_1KB;
2837 
2838 	rbrp->block_size = hxgep->rx_default_block_size;
2839 
2840 	if (!hxgep->param_arr[param_accept_jumbo].value) {
2841 		rbrp->pkt_buf_size2 = RBR_BUFSZ2_2K;
2842 		rbrp->pkt_buf_size2_bytes = RBR_BUFSZ2_2K_BYTES;
2843 		rbrp->hpi_pkt_buf_size2 = SIZE_2KB;
2844 	} else {
2845 		rbrp->hpi_pkt_buf_size2 = SIZE_4KB;
2846 		rbrp->pkt_buf_size2 = RBR_BUFSZ2_4K;
2847 		rbrp->pkt_buf_size2_bytes = RBR_BUFSZ2_4K_BYTES;
2848 	}
2849 
2850 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2851 	    "==> hxge_map_rxdma_channel_buf_ring: channel %d "
2852 	    "actual rbr max %d rbb_max %d nmsgs %d "
2853 	    "rbrp->block_size %d default_block_size %d "
2854 	    "(config hxge_rbr_size %d hxge_rbr_spare_size %d)",
2855 	    channel, rbrp->rbr_max_size, rbrp->rbb_max, nmsgs,
2856 	    rbrp->block_size, hxgep->rx_default_block_size,
2857 	    hxge_rbr_size, hxge_rbr_spare_size));
2858 
2859 	/*
2860 	 * Map in buffers from the buffer pool.
2861 	 * Note that num_blocks is the num_chunks. For Sparc, there is likely
2862 	 * only one chunk. For x86, there will be many chunks.
2863 	 * Loop over chunks.
2864 	 */
2865 	index = 0;
2866 	for (i = 0; i < rbrp->num_blocks; i++, dma_bufp++) {
2867 		bsize = dma_bufp->block_size;
2868 		nblocks = dma_bufp->nblocks;
2869 #if defined(__i386)
2870 		ring_info->buffer[i].dvma_addr = (uint32_t)dma_bufp->ioaddr_pp;
2871 #else
2872 		ring_info->buffer[i].dvma_addr = (uint64_t)dma_bufp->ioaddr_pp;
2873 #endif
2874 		ring_info->buffer[i].buf_index = i;
2875 		ring_info->buffer[i].buf_size = dma_bufp->alength;
2876 		ring_info->buffer[i].start_index = index;
2877 #if defined(__i386)
2878 		ring_info->buffer[i].kaddr = (uint32_t)dma_bufp->kaddrp;
2879 #else
2880 		ring_info->buffer[i].kaddr = (uint64_t)dma_bufp->kaddrp;
2881 #endif
2882 
2883 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2884 		    " hxge_map_rxdma_channel_buf_ring: map channel %d "
2885 		    "chunk %d nblocks %d chunk_size %x block_size 0x%x "
2886 		    "dma_bufp $%p dvma_addr $%p", channel, i,
2887 		    dma_bufp->nblocks,
2888 		    ring_info->buffer[i].buf_size, bsize, dma_bufp,
2889 		    ring_info->buffer[i].dvma_addr));
2890 
2891 		/* loop over blocks within a chunk */
2892 		for (j = 0; j < nblocks; j++) {
2893 			if ((rx_msg_p = hxge_allocb(bsize, BPRI_LO,
2894 			    dma_bufp)) == NULL) {
2895 				HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2896 				    "allocb failed (index %d i %d j %d)",
2897 				    index, i, j));
2898 				goto hxge_map_rxdma_channel_buf_ring_fail1;
2899 			}
2900 			rx_msg_ring[index] = rx_msg_p;
2901 			rx_msg_p->block_index = index;
2902 			rx_msg_p->shifted_addr = (uint32_t)
2903 			    ((rx_msg_p->buf_dma.dma_cookie.dmac_laddress >>
2904 			    RBR_BKADDR_SHIFT));
2905 			/*
2906 			 * Too much output
2907 			 * HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2908 			 *	"index %d j %d rx_msg_p $%p mblk %p",
2909 			 *	index, j, rx_msg_p, rx_msg_p->rx_mblk_p));
2910 			 */
2911 			mblk_p = rx_msg_p->rx_mblk_p;
2912 			mblk_p->b_wptr = mblk_p->b_rptr + bsize;
2913 
2914 			rbrp->rbr_ref_cnt++;
2915 			index++;
2916 			rx_msg_p->buf_dma.dma_channel = channel;
2917 		}
2918 	}
2919 	if (i < rbrp->num_blocks) {
2920 		goto hxge_map_rxdma_channel_buf_ring_fail1;
2921 	}
2922 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2923 	    "hxge_map_rxdma_channel_buf_ring: done buf init "
2924 	    "channel %d msg block entries %d", channel, index));
2925 	ring_info->block_size_mask = bsize - 1;
2926 	rbrp->rx_msg_ring = rx_msg_ring;
2927 	rbrp->dma_bufp = dma_buf_p;
2928 	rbrp->ring_info = ring_info;
2929 
2930 	status = hxge_rxbuf_index_info_init(hxgep, rbrp);
2931 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, " hxge_map_rxdma_channel_buf_ring: "
2932 	    "channel %d done buf info init", channel));
2933 
2934 	/*
2935 	 * Finally, permit hxge_freeb() to call hxge_post_page().
2936 	 */
2937 	rbrp->rbr_state = RBR_POSTING;
2938 
2939 	*rbr_p = rbrp;
2940 
2941 	goto hxge_map_rxdma_channel_buf_ring_exit;
2942 
2943 hxge_map_rxdma_channel_buf_ring_fail1:
2944 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2945 	    " hxge_map_rxdma_channel_buf_ring: failed channel (0x%x)",
2946 	    channel, status));
2947 
2948 	index--;
2949 	for (; index >= 0; index--) {
2950 		rx_msg_p = rx_msg_ring[index];
2951 		if (rx_msg_p != NULL) {
2952 			freeb(rx_msg_p->rx_mblk_p);
2953 			rx_msg_ring[index] = NULL;
2954 		}
2955 	}
2956 
2957 hxge_map_rxdma_channel_buf_ring_fail:
2958 	MUTEX_DESTROY(&rbrp->post_lock);
2959 	MUTEX_DESTROY(&rbrp->lock);
2960 	KMEM_FREE(ring_info, sizeof (rxring_info_t));
2961 	KMEM_FREE(rx_msg_ring, size);
2962 	KMEM_FREE(rbrp, sizeof (rx_rbr_ring_t));
2963 
2964 	status = HXGE_ERROR;
2965 
2966 hxge_map_rxdma_channel_buf_ring_exit:
2967 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2968 	    "<== hxge_map_rxdma_channel_buf_ring status 0x%08x", status));
2969 
2970 	return (status);
2971 }
2972 
2973 /*ARGSUSED*/
2974 static void
2975 hxge_unmap_rxdma_channel_buf_ring(p_hxge_t hxgep,
2976     p_rx_rbr_ring_t rbr_p)
2977 {
2978 	p_rx_msg_t	*rx_msg_ring;
2979 	p_rx_msg_t	rx_msg_p;
2980 	rxring_info_t	*ring_info;
2981 	int		i;
2982 	uint32_t	size;
2983 
2984 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2985 	    "==> hxge_unmap_rxdma_channel_buf_ring"));
2986 	if (rbr_p == NULL) {
2987 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
2988 		    "<== hxge_unmap_rxdma_channel_buf_ring: NULL rbrp"));
2989 		return;
2990 	}
2991 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2992 	    "==> hxge_unmap_rxdma_channel_buf_ring: channel %d", rbr_p->rdc));
2993 
2994 	rx_msg_ring = rbr_p->rx_msg_ring;
2995 	ring_info = rbr_p->ring_info;
2996 
2997 	if (rx_msg_ring == NULL || ring_info == NULL) {
2998 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2999 		    "<== hxge_unmap_rxdma_channel_buf_ring: "
3000 		    "rx_msg_ring $%p ring_info $%p", rx_msg_p, ring_info));
3001 		return;
3002 	}
3003 
3004 	size = rbr_p->tnblocks * sizeof (p_rx_msg_t);
3005 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3006 	    " hxge_unmap_rxdma_channel_buf_ring: channel %d chunks %d "
3007 	    "tnblocks %d (max %d) size ptrs %d ", rbr_p->rdc, rbr_p->num_blocks,
3008 	    rbr_p->tnblocks, rbr_p->rbr_max_size, size));
3009 
3010 	for (i = 0; i < rbr_p->tnblocks; i++) {
3011 		rx_msg_p = rx_msg_ring[i];
3012 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3013 		    " hxge_unmap_rxdma_channel_buf_ring: "
3014 		    "rx_msg_p $%p", rx_msg_p));
3015 		if (rx_msg_p != NULL) {
3016 			freeb(rx_msg_p->rx_mblk_p);
3017 			rx_msg_ring[i] = NULL;
3018 		}
3019 	}
3020 
3021 	/*
3022 	 * We no longer may use the mutex <post_lock>. By setting
3023 	 * <rbr_state> to anything but POSTING, we prevent
3024 	 * hxge_post_page() from accessing a dead mutex.
3025 	 */
3026 	rbr_p->rbr_state = RBR_UNMAPPING;
3027 	MUTEX_DESTROY(&rbr_p->post_lock);
3028 
3029 	MUTEX_DESTROY(&rbr_p->lock);
3030 	KMEM_FREE(ring_info, sizeof (rxring_info_t));
3031 	KMEM_FREE(rx_msg_ring, size);
3032 
3033 	if (rbr_p->rbr_ref_cnt == 0) {
3034 		/* This is the normal state of affairs. */
3035 		KMEM_FREE(rbr_p, sizeof (*rbr_p));
3036 	} else {
3037 		/*
3038 		 * Some of our buffers are still being used.
3039 		 * Therefore, tell hxge_freeb() this ring is
3040 		 * unmapped, so it may free <rbr_p> for us.
3041 		 */
3042 		rbr_p->rbr_state = RBR_UNMAPPED;
3043 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3044 		    "unmap_rxdma_buf_ring: %d %s outstanding.",
3045 		    rbr_p->rbr_ref_cnt,
3046 		    rbr_p->rbr_ref_cnt == 1 ? "msg" : "msgs"));
3047 	}
3048 
3049 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3050 	    "<== hxge_unmap_rxdma_channel_buf_ring"));
3051 }
3052 
3053 static hxge_status_t
3054 hxge_rxdma_hw_start_common(p_hxge_t hxgep)
3055 {
3056 	hxge_status_t status = HXGE_OK;
3057 
3058 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_hw_start_common"));
3059 
3060 	/*
3061 	 * Load the sharable parameters by writing to the function zero control
3062 	 * registers. These FZC registers should be initialized only once for
3063 	 * the entire chip.
3064 	 */
3065 	(void) hxge_init_fzc_rx_common(hxgep);
3066 
3067 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_hw_start_common"));
3068 
3069 	return (status);
3070 }
3071 
3072 static hxge_status_t
3073 hxge_rxdma_hw_start(p_hxge_t hxgep)
3074 {
3075 	int			i, ndmas;
3076 	uint16_t		channel;
3077 	p_rx_rbr_rings_t	rx_rbr_rings;
3078 	p_rx_rbr_ring_t		*rbr_rings;
3079 	p_rx_rcr_rings_t	rx_rcr_rings;
3080 	p_rx_rcr_ring_t		*rcr_rings;
3081 	p_rx_mbox_areas_t	rx_mbox_areas_p;
3082 	p_rx_mbox_t		*rx_mbox_p;
3083 	hxge_status_t		status = HXGE_OK;
3084 
3085 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_hw_start"));
3086 
3087 	rx_rbr_rings = hxgep->rx_rbr_rings;
3088 	rx_rcr_rings = hxgep->rx_rcr_rings;
3089 	if (rx_rbr_rings == NULL || rx_rcr_rings == NULL) {
3090 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
3091 		    "<== hxge_rxdma_hw_start: NULL ring pointers"));
3092 		return (HXGE_ERROR);
3093 	}
3094 
3095 	ndmas = rx_rbr_rings->ndmas;
3096 	if (ndmas == 0) {
3097 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
3098 		    "<== hxge_rxdma_hw_start: no dma channel allocated"));
3099 		return (HXGE_ERROR);
3100 	}
3101 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3102 	    "==> hxge_rxdma_hw_start (ndmas %d)", ndmas));
3103 
3104 	/*
3105 	 * Scrub the RDC Rx DMA Prefetch Buffer Command.
3106 	 */
3107 	for (i = 0; i < 128; i++) {
3108 		HXGE_REG_WR64(hxgep->hpi_handle, RDC_PREF_CMD, i);
3109 	}
3110 
3111 	/*
3112 	 * Scrub Rx DMA Shadow Tail Command.
3113 	 */
3114 	for (i = 0; i < 64; i++) {
3115 		HXGE_REG_WR64(hxgep->hpi_handle, RDC_SHADOW_CMD, i);
3116 	}
3117 
3118 	/*
3119 	 * Scrub Rx DMA Control Fifo Command.
3120 	 */
3121 	for (i = 0; i < 512; i++) {
3122 		HXGE_REG_WR64(hxgep->hpi_handle, RDC_CTRL_FIFO_CMD, i);
3123 	}
3124 
3125 	/*
3126 	 * Scrub Rx DMA Data Fifo Command.
3127 	 */
3128 	for (i = 0; i < 1536; i++) {
3129 		HXGE_REG_WR64(hxgep->hpi_handle, RDC_DATA_FIFO_CMD, i);
3130 	}
3131 
3132 	/*
3133 	 * Reset the FIFO Error Stat.
3134 	 */
3135 	HXGE_REG_WR64(hxgep->hpi_handle, RDC_FIFO_ERR_STAT, 0xFF);
3136 
3137 	/* Set the error mask to receive interrupts */
3138 	HXGE_REG_WR64(hxgep->hpi_handle, RDC_FIFO_ERR_INT_MASK, 0x0);
3139 
3140 	rbr_rings = rx_rbr_rings->rbr_rings;
3141 	rcr_rings = rx_rcr_rings->rcr_rings;
3142 	rx_mbox_areas_p = hxgep->rx_mbox_areas_p;
3143 	if (rx_mbox_areas_p) {
3144 		rx_mbox_p = rx_mbox_areas_p->rxmbox_areas;
3145 	}
3146 
3147 	for (i = 0; i < ndmas; i++) {
3148 		channel = rbr_rings[i]->rdc;
3149 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3150 		    "==> hxge_rxdma_hw_start (ndmas %d) channel %d",
3151 		    ndmas, channel));
3152 		status = hxge_rxdma_start_channel(hxgep, channel,
3153 		    (p_rx_rbr_ring_t)rbr_rings[i],
3154 		    (p_rx_rcr_ring_t)rcr_rings[i],
3155 		    (p_rx_mbox_t)rx_mbox_p[i], rbr_rings[i]->rbb_max);
3156 		if (status != HXGE_OK) {
3157 			goto hxge_rxdma_hw_start_fail1;
3158 		}
3159 	}
3160 
3161 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_hw_start: "
3162 	    "rx_rbr_rings 0x%016llx rings 0x%016llx",
3163 	    rx_rbr_rings, rx_rcr_rings));
3164 	goto hxge_rxdma_hw_start_exit;
3165 
3166 hxge_rxdma_hw_start_fail1:
3167 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3168 	    "==> hxge_rxdma_hw_start: disable "
3169 	    "(status 0x%x channel %d i %d)", status, channel, i));
3170 	for (; i >= 0; i--) {
3171 		channel = rbr_rings[i]->rdc;
3172 		(void) hxge_rxdma_stop_channel(hxgep, channel);
3173 	}
3174 
3175 hxge_rxdma_hw_start_exit:
3176 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3177 	    "==> hxge_rxdma_hw_start: (status 0x%x)", status));
3178 	return (status);
3179 }
3180 
3181 static void
3182 hxge_rxdma_hw_stop(p_hxge_t hxgep)
3183 {
3184 	int			i, ndmas;
3185 	uint16_t		channel;
3186 	p_rx_rbr_rings_t	rx_rbr_rings;
3187 	p_rx_rbr_ring_t		*rbr_rings;
3188 	p_rx_rcr_rings_t	rx_rcr_rings;
3189 
3190 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_hw_stop"));
3191 
3192 	rx_rbr_rings = hxgep->rx_rbr_rings;
3193 	rx_rcr_rings = hxgep->rx_rcr_rings;
3194 
3195 	if (rx_rbr_rings == NULL || rx_rcr_rings == NULL) {
3196 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
3197 		    "<== hxge_rxdma_hw_stop: NULL ring pointers"));
3198 		return;
3199 	}
3200 
3201 	ndmas = rx_rbr_rings->ndmas;
3202 	if (!ndmas) {
3203 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
3204 		    "<== hxge_rxdma_hw_stop: no dma channel allocated"));
3205 		return;
3206 	}
3207 
3208 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3209 	    "==> hxge_rxdma_hw_stop (ndmas %d)", ndmas));
3210 
3211 	rbr_rings = rx_rbr_rings->rbr_rings;
3212 	for (i = 0; i < ndmas; i++) {
3213 		channel = rbr_rings[i]->rdc;
3214 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3215 		    "==> hxge_rxdma_hw_stop (ndmas %d) channel %d",
3216 		    ndmas, channel));
3217 		(void) hxge_rxdma_stop_channel(hxgep, channel);
3218 	}
3219 
3220 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_hw_stop: "
3221 	    "rx_rbr_rings 0x%016llx rings 0x%016llx",
3222 	    rx_rbr_rings, rx_rcr_rings));
3223 
3224 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "<== hxge_rxdma_hw_stop"));
3225 }
3226 
3227 static hxge_status_t
3228 hxge_rxdma_start_channel(p_hxge_t hxgep, uint16_t channel,
3229     p_rx_rbr_ring_t rbr_p, p_rx_rcr_ring_t rcr_p, p_rx_mbox_t mbox_p,
3230     int n_init_kick)
3231 {
3232 	hpi_handle_t		handle;
3233 	hpi_status_t		rs = HPI_SUCCESS;
3234 	rdc_stat_t		cs;
3235 	rdc_int_mask_t		ent_mask;
3236 	hxge_status_t		status = HXGE_OK;
3237 
3238 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_start_channel"));
3239 
3240 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
3241 
3242 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "hxge_rxdma_start_channel: "
3243 	    "hpi handle addr $%p acc $%p",
3244 	    hxgep->hpi_handle.regp, hxgep->hpi_handle.regh));
3245 
3246 	/* Reset RXDMA channel */
3247 	rs = hpi_rxdma_cfg_rdc_reset(handle, channel);
3248 	if (rs != HPI_SUCCESS) {
3249 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3250 		    "==> hxge_rxdma_start_channel: "
3251 		    "reset rxdma failed (0x%08x channel %d)",
3252 		    status, channel));
3253 		return (HXGE_ERROR | rs);
3254 	}
3255 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3256 	    "==> hxge_rxdma_start_channel: reset done: channel %d", channel));
3257 
3258 	/*
3259 	 * Initialize the RXDMA channel specific FZC control configurations.
3260 	 * These FZC registers are pertaining to each RX channel (logical
3261 	 * pages).
3262 	 */
3263 	status = hxge_init_fzc_rxdma_channel(hxgep,
3264 	    channel, rbr_p, rcr_p, mbox_p);
3265 	if (status != HXGE_OK) {
3266 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3267 		    "==> hxge_rxdma_start_channel: "
3268 		    "init fzc rxdma failed (0x%08x channel %d)",
3269 		    status, channel));
3270 		return (status);
3271 	}
3272 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3273 	    "==> hxge_rxdma_start_channel: fzc done"));
3274 
3275 	/*
3276 	 * Zero out the shadow  and prefetch ram.
3277 	 */
3278 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3279 	    "==> hxge_rxdma_start_channel: ram done"));
3280 
3281 	/* Set up the interrupt event masks. */
3282 	ent_mask.value = 0;
3283 	rs = hpi_rxdma_event_mask(handle, OP_SET, channel, &ent_mask);
3284 	if (rs != HPI_SUCCESS) {
3285 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3286 		    "==> hxge_rxdma_start_channel: "
3287 		    "init rxdma event masks failed (0x%08x channel %d)",
3288 		    status, channel));
3289 		return (HXGE_ERROR | rs);
3290 	}
3291 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_start_channel: "
3292 	    "event done: channel %d (mask 0x%016llx)",
3293 	    channel, ent_mask.value));
3294 
3295 	/*
3296 	 * Load RXDMA descriptors, buffers, mailbox, initialise the receive DMA
3297 	 * channels and enable each DMA channel.
3298 	 */
3299 	status = hxge_enable_rxdma_channel(hxgep,
3300 	    channel, rbr_p, rcr_p, mbox_p, n_init_kick);
3301 	if (status != HXGE_OK) {
3302 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3303 		    " hxge_rxdma_start_channel: "
3304 		    " init enable rxdma failed (0x%08x channel %d)",
3305 		    status, channel));
3306 		return (status);
3307 	}
3308 
3309 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_start_channel: "
3310 	    "control done - channel %d cs 0x%016llx", channel, cs.value));
3311 
3312 	/*
3313 	 * Initialize the receive DMA control and status register
3314 	 * Note that rdc_stat HAS to be set after RBR and RCR rings are set
3315 	 */
3316 	cs.value = 0;
3317 	cs.bits.mex = 1;
3318 	cs.bits.rcr_thres = 1;
3319 	cs.bits.rcr_to = 1;
3320 	cs.bits.rbr_empty = 1;
3321 	status = hxge_init_rxdma_channel_cntl_stat(hxgep, channel, &cs);
3322 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_start_channel: "
3323 	    "channel %d rx_dma_cntl_stat 0x%0016llx", channel, cs.value));
3324 	if (status != HXGE_OK) {
3325 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3326 		    "==> hxge_rxdma_start_channel: "
3327 		    "init rxdma control register failed (0x%08x channel %d",
3328 		    status, channel));
3329 		return (status);
3330 	}
3331 
3332 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_start_channel: "
3333 	    "control done - channel %d cs 0x%016llx", channel, cs.value));
3334 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3335 	    "==> hxge_rxdma_start_channel: enable done"));
3336 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "<== hxge_rxdma_start_channel"));
3337 
3338 	return (HXGE_OK);
3339 }
3340 
3341 static hxge_status_t
3342 hxge_rxdma_stop_channel(p_hxge_t hxgep, uint16_t channel)
3343 {
3344 	hpi_handle_t		handle;
3345 	hpi_status_t		rs = HPI_SUCCESS;
3346 	rdc_stat_t		cs;
3347 	rdc_int_mask_t		ent_mask;
3348 	hxge_status_t		status = HXGE_OK;
3349 
3350 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rxdma_stop_channel"));
3351 
3352 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
3353 
3354 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "hxge_rxdma_stop_channel: "
3355 	    "hpi handle addr $%p acc $%p",
3356 	    hxgep->hpi_handle.regp, hxgep->hpi_handle.regh));
3357 
3358 	/* Reset RXDMA channel */
3359 	rs = hpi_rxdma_cfg_rdc_reset(handle, channel);
3360 	if (rs != HPI_SUCCESS) {
3361 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3362 		    " hxge_rxdma_stop_channel: "
3363 		    " reset rxdma failed (0x%08x channel %d)",
3364 		    rs, channel));
3365 		return (HXGE_ERROR | rs);
3366 	}
3367 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
3368 	    "==> hxge_rxdma_stop_channel: reset done"));
3369 
3370 	/* Set up the interrupt event masks. */
3371 	ent_mask.value = RDC_INT_MASK_ALL;
3372 	rs = hpi_rxdma_event_mask(handle, OP_SET, channel, &ent_mask);
3373 	if (rs != HPI_SUCCESS) {
3374 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3375 		    "==> hxge_rxdma_stop_channel: "
3376 		    "set rxdma event masks failed (0x%08x channel %d)",
3377 		    rs, channel));
3378 		return (HXGE_ERROR | rs);
3379 	}
3380 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
3381 	    "==> hxge_rxdma_stop_channel: event done"));
3382 
3383 	/* Initialize the receive DMA control and status register */
3384 	cs.value = 0;
3385 	status = hxge_init_rxdma_channel_cntl_stat(hxgep, channel, &cs);
3386 
3387 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rxdma_stop_channel: control "
3388 	    " to default (all 0s) 0x%08x", cs.value));
3389 
3390 	if (status != HXGE_OK) {
3391 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3392 		    " hxge_rxdma_stop_channel: init rxdma"
3393 		    " control register failed (0x%08x channel %d",
3394 		    status, channel));
3395 		return (status);
3396 	}
3397 
3398 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
3399 	    "==> hxge_rxdma_stop_channel: control done"));
3400 
3401 	/* disable dma channel */
3402 	status = hxge_disable_rxdma_channel(hxgep, channel);
3403 
3404 	if (status != HXGE_OK) {
3405 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3406 		    " hxge_rxdma_stop_channel: "
3407 		    " init enable rxdma failed (0x%08x channel %d)",
3408 		    status, channel));
3409 		return (status);
3410 	}
3411 
3412 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
3413 	    "==> hxge_rxdma_stop_channel: disable done"));
3414 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_rxdma_stop_channel"));
3415 
3416 	return (HXGE_OK);
3417 }
3418 
3419 hxge_status_t
3420 hxge_rxdma_handle_sys_errors(p_hxge_t hxgep)
3421 {
3422 	hpi_handle_t		handle;
3423 	p_hxge_rdc_sys_stats_t	statsp;
3424 	rdc_fifo_err_stat_t	stat;
3425 	hxge_status_t		status = HXGE_OK;
3426 
3427 	handle = hxgep->hpi_handle;
3428 	statsp = (p_hxge_rdc_sys_stats_t)&hxgep->statsp->rdc_sys_stats;
3429 
3430 	/* Clear the int_dbg register in case it is an injected err */
3431 	HXGE_REG_WR64(handle, RDC_FIFO_ERR_INT_DBG, 0x0);
3432 
3433 	/* Get the error status and clear the register */
3434 	HXGE_REG_RD64(handle, RDC_FIFO_ERR_STAT, &stat.value);
3435 	HXGE_REG_WR64(handle, RDC_FIFO_ERR_STAT, stat.value);
3436 
3437 	if (stat.bits.rx_ctrl_fifo_sec) {
3438 		statsp->ctrl_fifo_sec++;
3439 		if (statsp->ctrl_fifo_sec == 1)
3440 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3441 			    "==> hxge_rxdma_handle_sys_errors: "
3442 			    "rx_ctrl_fifo_sec"));
3443 	}
3444 
3445 	if (stat.bits.rx_ctrl_fifo_ded) {
3446 		/* Global fatal error encountered */
3447 		statsp->ctrl_fifo_ded++;
3448 		HXGE_FM_REPORT_ERROR(hxgep, NULL,
3449 		    HXGE_FM_EREPORT_RDMC_CTRL_FIFO_DED);
3450 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3451 		    "==> hxge_rxdma_handle_sys_errors: "
3452 		    "fatal error: rx_ctrl_fifo_ded error"));
3453 	}
3454 
3455 	if (stat.bits.rx_data_fifo_sec) {
3456 		statsp->data_fifo_sec++;
3457 		if (statsp->data_fifo_sec == 1)
3458 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3459 			    "==> hxge_rxdma_handle_sys_errors: "
3460 			    "rx_data_fifo_sec"));
3461 	}
3462 
3463 	if (stat.bits.rx_data_fifo_ded) {
3464 		/* Global fatal error encountered */
3465 		statsp->data_fifo_ded++;
3466 		HXGE_FM_REPORT_ERROR(hxgep, NULL,
3467 		    HXGE_FM_EREPORT_RDMC_DATA_FIFO_DED);
3468 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3469 		    "==> hxge_rxdma_handle_sys_errors: "
3470 		    "fatal error: rx_data_fifo_ded error"));
3471 	}
3472 
3473 	if (stat.bits.rx_ctrl_fifo_ded || stat.bits.rx_data_fifo_ded) {
3474 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3475 		    " hxge_rxdma_handle_sys_errors: fatal error\n"));
3476 		status = hxge_rx_port_fatal_err_recover(hxgep);
3477 		if (status == HXGE_OK) {
3478 			FM_SERVICE_RESTORED(hxgep);
3479 		}
3480 	}
3481 
3482 	return (HXGE_OK);
3483 }
3484 
3485 static hxge_status_t
3486 hxge_rxdma_fatal_err_recover(p_hxge_t hxgep, uint16_t channel)
3487 {
3488 	hpi_handle_t		handle;
3489 	hpi_status_t 		rs = HPI_SUCCESS;
3490 	hxge_status_t 		status = HXGE_OK;
3491 	p_rx_rbr_ring_t		rbrp;
3492 	p_rx_rcr_ring_t		rcrp;
3493 	p_rx_mbox_t		mboxp;
3494 	rdc_int_mask_t		ent_mask;
3495 	p_hxge_dma_common_t	dmap;
3496 	int			ring_idx;
3497 	p_rx_msg_t		rx_msg_p;
3498 	int			i;
3499 	uint32_t		hxge_port_rcr_size;
3500 	uint64_t		tmp;
3501 	int			n_init_kick = 0;
3502 
3503 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rxdma_fatal_err_recover"));
3504 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3505 	    "Recovering from RxDMAChannel#%d error...", channel));
3506 
3507 	/*
3508 	 * Stop the dma channel waits for the stop done. If the stop done bit
3509 	 * is not set, then create an error.
3510 	 */
3511 
3512 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
3513 
3514 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Rx DMA stop..."));
3515 
3516 	ring_idx = hxge_rxdma_get_ring_index(hxgep, channel);
3517 	rbrp = (p_rx_rbr_ring_t)hxgep->rx_rbr_rings->rbr_rings[ring_idx];
3518 	rcrp = (p_rx_rcr_ring_t)hxgep->rx_rcr_rings->rcr_rings[ring_idx];
3519 
3520 	MUTEX_ENTER(&rcrp->lock);
3521 	MUTEX_ENTER(&rbrp->lock);
3522 
3523 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Disable RxDMA channel..."));
3524 
3525 	rs = hpi_rxdma_cfg_rdc_disable(handle, channel);
3526 	if (rs != HPI_SUCCESS) {
3527 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3528 		    "hxge_disable_rxdma_channel:failed"));
3529 		goto fail;
3530 	}
3531 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Disable RxDMA interrupt..."));
3532 
3533 	/* Disable interrupt */
3534 	ent_mask.value = RDC_INT_MASK_ALL;
3535 	rs = hpi_rxdma_event_mask(handle, OP_SET, channel, &ent_mask);
3536 	if (rs != HPI_SUCCESS) {
3537 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3538 		    "Set rxdma event masks failed (channel %d)", channel));
3539 	}
3540 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "RxDMA channel reset..."));
3541 
3542 	/* Reset RXDMA channel */
3543 	rs = hpi_rxdma_cfg_rdc_reset(handle, channel);
3544 	if (rs != HPI_SUCCESS) {
3545 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3546 		    "Reset rxdma failed (channel %d)", channel));
3547 		goto fail;
3548 	}
3549 	hxge_port_rcr_size = hxgep->hxge_port_rcr_size;
3550 	mboxp = (p_rx_mbox_t)hxgep->rx_mbox_areas_p->rxmbox_areas[ring_idx];
3551 
3552 	rbrp->rbr_wr_index = (rbrp->rbb_max - 1);
3553 	rbrp->rbr_rd_index = 0;
3554 
3555 	rcrp->comp_rd_index = 0;
3556 	rcrp->comp_wt_index = 0;
3557 	rcrp->rcr_desc_rd_head_p = rcrp->rcr_desc_first_p =
3558 	    (p_rcr_entry_t)DMA_COMMON_VPTR(rcrp->rcr_desc);
3559 #if defined(__i386)
3560 	rcrp->rcr_desc_rd_head_pp = rcrp->rcr_desc_first_pp =
3561 	    (p_rcr_entry_t)(uint32_t)DMA_COMMON_IOADDR(rcrp->rcr_desc);
3562 #else
3563 	rcrp->rcr_desc_rd_head_pp = rcrp->rcr_desc_first_pp =
3564 	    (p_rcr_entry_t)DMA_COMMON_IOADDR(rcrp->rcr_desc);
3565 #endif
3566 
3567 	rcrp->rcr_desc_last_p = rcrp->rcr_desc_rd_head_p +
3568 	    (hxge_port_rcr_size - 1);
3569 	rcrp->rcr_desc_last_pp = rcrp->rcr_desc_rd_head_pp +
3570 	    (hxge_port_rcr_size - 1);
3571 
3572 	rcrp->rcr_tail_begin = DMA_COMMON_IOADDR(rcrp->rcr_desc);
3573 	rcrp->rcr_tail_begin = (rcrp->rcr_tail_begin & 0x7ffffULL) >> 3;
3574 
3575 	dmap = (p_hxge_dma_common_t)&rcrp->rcr_desc;
3576 	bzero((caddr_t)dmap->kaddrp, dmap->alength);
3577 
3578 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "rbr entries = %d\n",
3579 	    rbrp->rbr_max_size));
3580 
3581 	/* Count the number of buffers owned by the hardware at this moment */
3582 	for (i = 0; i < rbrp->rbr_max_size; i++) {
3583 		rx_msg_p = rbrp->rx_msg_ring[i];
3584 		if (rx_msg_p->ref_cnt == 1) {
3585 			n_init_kick++;
3586 		}
3587 	}
3588 
3589 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "RxDMA channel re-start..."));
3590 
3591 	/*
3592 	 * This is error recover! Some buffers are owned by the hardware and
3593 	 * the rest are owned by the apps. We should only kick in those
3594 	 * owned by the hardware initially. The apps will post theirs
3595 	 * eventually.
3596 	 */
3597 	status = hxge_rxdma_start_channel(hxgep, channel, rbrp, rcrp, mboxp,
3598 	    n_init_kick);
3599 	if (status != HXGE_OK) {
3600 		goto fail;
3601 	}
3602 
3603 	/*
3604 	 * The DMA channel may disable itself automatically.
3605 	 * The following is a work-around.
3606 	 */
3607 	HXGE_REG_RD64(handle, RDC_RX_CFG1, &tmp);
3608 	rs = hpi_rxdma_cfg_rdc_enable(handle, channel);
3609 	if (rs != HPI_SUCCESS) {
3610 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3611 		    "hpi_rxdma_cfg_rdc_enable (channel %d)", channel));
3612 	}
3613 
3614 	MUTEX_EXIT(&rbrp->lock);
3615 	MUTEX_EXIT(&rcrp->lock);
3616 
3617 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3618 	    "Recovery Successful, RxDMAChannel#%d Restored", channel));
3619 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_rxdma_fatal_err_recover"));
3620 
3621 	return (HXGE_OK);
3622 
3623 fail:
3624 	MUTEX_EXIT(&rbrp->lock);
3625 	MUTEX_EXIT(&rcrp->lock);
3626 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL, "Recovery failed"));
3627 
3628 	return (HXGE_ERROR | rs);
3629 }
3630 
3631 static hxge_status_t
3632 hxge_rx_port_fatal_err_recover(p_hxge_t hxgep)
3633 {
3634 	hxge_status_t		status = HXGE_OK;
3635 	p_hxge_dma_common_t	*dma_buf_p;
3636 	uint16_t		channel;
3637 	int			ndmas;
3638 	int			i;
3639 	block_reset_t		reset_reg;
3640 	p_rx_rcr_ring_t	rcrp;
3641 	p_rx_rbr_ring_t rbrp;
3642 
3643 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rx_port_fatal_err_recover"));
3644 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL, "Recovering from RDC error ..."));
3645 
3646 	/* Reset RDC block from PEU for this fatal error */
3647 	reset_reg.value = 0;
3648 	reset_reg.bits.rdc_rst = 1;
3649 	HXGE_REG_WR32(hxgep->hpi_handle, BLOCK_RESET, reset_reg.value);
3650 
3651 	/* Disable RxMAC */
3652 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Disable RxMAC...\n"));
3653 	if (hxge_rx_vmac_disable(hxgep) != HXGE_OK)
3654 		goto fail;
3655 
3656 	HXGE_DELAY(1000);
3657 
3658 	/* Restore any common settings after PEU reset */
3659 	if (hxge_rxdma_hw_start_common(hxgep) != HXGE_OK)
3660 		goto fail;
3661 
3662 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Stop all RxDMA channels..."));
3663 
3664 	ndmas = hxgep->rx_buf_pool_p->ndmas;
3665 	dma_buf_p = hxgep->rx_buf_pool_p->dma_buf_pool_p;
3666 
3667 	for (i = 0; i < ndmas; i++) {
3668 		channel = ((p_hxge_dma_common_t)dma_buf_p[i])->dma_channel;
3669 		rcrp = hxgep->rx_rcr_rings->rcr_rings[channel];
3670 		rbrp = rcrp->rx_rbr_p;
3671 
3672 		MUTEX_ENTER(&rbrp->post_lock);
3673 		/* This function needs to be inside the post_lock */
3674 		if (hxge_rxdma_fatal_err_recover(hxgep, channel) != HXGE_OK) {
3675 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3676 			    "Could not recover channel %d", channel));
3677 		}
3678 		MUTEX_EXIT(&rbrp->post_lock);
3679 	}
3680 
3681 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Reset RxMAC..."));
3682 
3683 	/* Reset RxMAC */
3684 	if (hxge_rx_vmac_reset(hxgep) != HXGE_OK) {
3685 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3686 		    "hxge_rx_port_fatal_err_recover: Failed to reset RxMAC"));
3687 		goto fail;
3688 	}
3689 
3690 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Re-initialize RxMAC..."));
3691 
3692 	/* Re-Initialize RxMAC */
3693 	if ((status = hxge_rx_vmac_init(hxgep)) != HXGE_OK) {
3694 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3695 		    "hxge_rx_port_fatal_err_recover: Failed to reset RxMAC"));
3696 		goto fail;
3697 	}
3698 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Re-enable RxMAC..."));
3699 
3700 	/* Re-enable RxMAC */
3701 	if ((status = hxge_rx_vmac_enable(hxgep)) != HXGE_OK) {
3702 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3703 		    "hxge_rx_port_fatal_err_recover: Failed to enable RxMAC"));
3704 		goto fail;
3705 	}
3706 
3707 	/* Reset the error mask since PEU reset cleared it */
3708 	HXGE_REG_WR64(hxgep->hpi_handle, RDC_FIFO_ERR_INT_MASK, 0x0);
3709 
3710 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3711 	    "Recovery Successful, RxPort Restored"));
3712 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_rx_port_fatal_err_recover"));
3713 
3714 	return (HXGE_OK);
3715 fail:
3716 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL, "Recovery failed"));
3717 	return (status);
3718 }
3719 
3720 static void
3721 hxge_rbr_empty_restore(p_hxge_t hxgep, p_rx_rbr_ring_t rx_rbr_p)
3722 {
3723 	hpi_status_t		hpi_status;
3724 	hxge_status_t		status;
3725 	int			i;
3726 	p_hxge_rx_ring_stats_t	rdc_stats;
3727 
3728 	rdc_stats = &hxgep->statsp->rdc_stats[rx_rbr_p->rdc];
3729 	rdc_stats->rbr_empty_restore++;
3730 	rx_rbr_p->rbr_is_empty = B_FALSE;
3731 
3732 	/*
3733 	 * Complete the processing for the RBR Empty by:
3734 	 *	0) kicking back HXGE_RBR_EMPTY_THRESHOLD
3735 	 *	   packets.
3736 	 *	1) Disable the RX vmac.
3737 	 *	2) Re-enable the affected DMA channel.
3738 	 *	3) Re-enable the RX vmac.
3739 	 */
3740 
3741 	/*
3742 	 * Disable the RX VMAC, but setting the framelength
3743 	 * to 0, since there is a hardware bug when disabling
3744 	 * the vmac.
3745 	 */
3746 	MUTEX_ENTER(hxgep->genlock);
3747 	(void) hpi_vmac_rx_set_framesize(
3748 	    HXGE_DEV_HPI_HANDLE(hxgep), (uint16_t)0);
3749 
3750 	hpi_status = hpi_rxdma_cfg_rdc_enable(
3751 	    HXGE_DEV_HPI_HANDLE(hxgep), rx_rbr_p->rdc);
3752 	if (hpi_status != HPI_SUCCESS) {
3753 		rdc_stats->rbr_empty_fail++;
3754 
3755 		/* Assume we are already inside the post_lock */
3756 		status = hxge_rxdma_fatal_err_recover(hxgep, rx_rbr_p->rdc);
3757 		if (status != HXGE_OK) {
3758 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3759 			    "hxge(%d): channel(%d) is empty.",
3760 			    hxgep->instance, rx_rbr_p->rdc));
3761 		}
3762 	}
3763 
3764 	for (i = 0; i < 1024; i++) {
3765 		uint64_t value;
3766 		RXDMA_REG_READ64(HXGE_DEV_HPI_HANDLE(hxgep),
3767 		    RDC_STAT, i & 3, &value);
3768 	}
3769 
3770 	/*
3771 	 * Re-enable the RX VMAC.
3772 	 */
3773 	(void) hpi_vmac_rx_set_framesize(HXGE_DEV_HPI_HANDLE(hxgep),
3774 	    (uint16_t)hxgep->vmac.maxframesize);
3775 	MUTEX_EXIT(hxgep->genlock);
3776 }
3777