xref: /titanic_51/usr/src/uts/common/io/hxge/hxge_rxdma.c (revision 63251bc7f1ca38259078c48e316fee4ed66d4e93)
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  * Tunables to manage the receive buffer blocks.
46  *
47  * hxge_rx_threshold_hi: copy all buffers.
48  * hxge_rx_bcopy_size_type: receive buffer block size type.
49  * hxge_rx_threshold_lo: copy only up to tunable block size type.
50  */
51 extern hxge_rxbuf_threshold_t hxge_rx_threshold_hi;
52 extern hxge_rxbuf_type_t hxge_rx_buf_size_type;
53 extern hxge_rxbuf_threshold_t hxge_rx_threshold_lo;
54 
55 /*
56  * Static local functions.
57  */
58 static hxge_status_t hxge_map_rxdma(p_hxge_t hxgep);
59 static void hxge_unmap_rxdma(p_hxge_t hxgep);
60 static hxge_status_t hxge_rxdma_hw_start_common(p_hxge_t hxgep);
61 static hxge_status_t hxge_rxdma_hw_start(p_hxge_t hxgep);
62 static void hxge_rxdma_hw_stop(p_hxge_t hxgep);
63 static hxge_status_t hxge_map_rxdma_channel(p_hxge_t hxgep, uint16_t channel,
64     p_hxge_dma_common_t *dma_buf_p, p_rx_rbr_ring_t *rbr_p,
65     uint32_t num_chunks, p_hxge_dma_common_t *dma_rbr_cntl_p,
66     p_hxge_dma_common_t *dma_rcr_cntl_p, p_hxge_dma_common_t *dma_mbox_cntl_p,
67     p_rx_rcr_ring_t *rcr_p, p_rx_mbox_t *rx_mbox_p);
68 static void hxge_unmap_rxdma_channel(p_hxge_t hxgep, uint16_t channel,
69 	p_rx_rbr_ring_t rbr_p, p_rx_rcr_ring_t rcr_p, p_rx_mbox_t rx_mbox_p);
70 static hxge_status_t hxge_map_rxdma_channel_cfg_ring(p_hxge_t hxgep,
71     uint16_t dma_channel, p_hxge_dma_common_t *dma_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_rbr_ring_t *rbr_p, p_rx_rcr_ring_t *rcr_p, p_rx_mbox_t *rx_mbox_p);
74 static void hxge_unmap_rxdma_channel_cfg_ring(p_hxge_t hxgep,
75 	p_rx_rcr_ring_t rcr_p, p_rx_mbox_t rx_mbox_p);
76 static hxge_status_t hxge_map_rxdma_channel_buf_ring(p_hxge_t hxgep,
77 	uint16_t channel, p_hxge_dma_common_t *dma_buf_p,
78 	p_rx_rbr_ring_t *rbr_p, uint32_t num_chunks);
79 static void hxge_unmap_rxdma_channel_buf_ring(p_hxge_t hxgep,
80 	p_rx_rbr_ring_t rbr_p);
81 static hxge_status_t hxge_rxdma_start_channel(p_hxge_t hxgep, uint16_t channel,
82 	p_rx_rbr_ring_t rbr_p, p_rx_rcr_ring_t rcr_p, p_rx_mbox_t mbox_p,
83 	int n_init_kick);
84 static hxge_status_t hxge_rxdma_stop_channel(p_hxge_t hxgep, uint16_t channel);
85 static mblk_t *hxge_rx_pkts(p_hxge_t hxgep, uint_t vindex, p_hxge_ldv_t ldvp,
86 	p_rx_rcr_ring_t	rcr_p, rdc_stat_t cs, int bytes_to_read);
87 static uint32_t hxge_scan_for_last_eop(p_rx_rcr_ring_t rcr_p,
88     p_rcr_entry_t rcr_desc_rd_head_p, uint32_t num_rcrs);
89 static void hxge_receive_packet(p_hxge_t hxgep, p_rx_rcr_ring_t rcr_p,
90 	p_rcr_entry_t rcr_desc_rd_head_p, boolean_t *multi_p,
91 	mblk_t ** mp, mblk_t ** mp_cont, uint32_t *invalid_rcr_entry);
92 static hxge_status_t hxge_disable_rxdma_channel(p_hxge_t hxgep,
93 	uint16_t channel);
94 static p_rx_msg_t hxge_allocb(size_t, uint32_t, p_hxge_dma_common_t);
95 static void hxge_freeb(p_rx_msg_t);
96 static hxge_status_t hxge_rx_err_evnts(p_hxge_t hxgep, uint_t index,
97 	p_hxge_ldv_t ldvp, rdc_stat_t cs);
98 static hxge_status_t hxge_rxbuf_index_info_init(p_hxge_t hxgep,
99 	p_rx_rbr_ring_t rx_dmap);
100 static hxge_status_t hxge_rxdma_fatal_err_recover(p_hxge_t hxgep,
101 	uint16_t channel);
102 static hxge_status_t hxge_rx_port_fatal_err_recover(p_hxge_t hxgep);
103 static void hxge_rbr_empty_restore(p_hxge_t hxgep,
104 	p_rx_rbr_ring_t rx_rbr_p);
105 
106 hxge_status_t
107 hxge_init_rxdma_channels(p_hxge_t hxgep)
108 {
109 	hxge_status_t		status = HXGE_OK;
110 	block_reset_t		reset_reg;
111 	int			i;
112 
113 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_init_rxdma_channels"));
114 
115 	for (i = 0; i < HXGE_MAX_RDCS; i++)
116 		hxgep->rdc_first_intr[i] = B_TRUE;
117 
118 	/* Reset RDC block from PEU to clear any previous state */
119 	reset_reg.value = 0;
120 	reset_reg.bits.rdc_rst = 1;
121 	HXGE_REG_WR32(hxgep->hpi_handle, BLOCK_RESET, reset_reg.value);
122 	HXGE_DELAY(1000);
123 
124 	status = hxge_map_rxdma(hxgep);
125 	if (status != HXGE_OK) {
126 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
127 		    "<== hxge_init_rxdma: status 0x%x", status));
128 		return (status);
129 	}
130 
131 	status = hxge_rxdma_hw_start_common(hxgep);
132 	if (status != HXGE_OK) {
133 		hxge_unmap_rxdma(hxgep);
134 	}
135 
136 	status = hxge_rxdma_hw_start(hxgep);
137 	if (status != HXGE_OK) {
138 		hxge_unmap_rxdma(hxgep);
139 	}
140 
141 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
142 	    "<== hxge_init_rxdma_channels: status 0x%x", status));
143 	return (status);
144 }
145 
146 void
147 hxge_uninit_rxdma_channels(p_hxge_t hxgep)
148 {
149 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_uninit_rxdma_channels"));
150 
151 	hxge_rxdma_hw_stop(hxgep);
152 	hxge_unmap_rxdma(hxgep);
153 
154 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "<== hxge_uinit_rxdma_channels"));
155 }
156 
157 hxge_status_t
158 hxge_init_rxdma_channel_cntl_stat(p_hxge_t hxgep, uint16_t channel,
159     rdc_stat_t *cs_p)
160 {
161 	hpi_handle_t	handle;
162 	hpi_status_t	rs = HPI_SUCCESS;
163 	hxge_status_t	status = HXGE_OK;
164 
165 	HXGE_DEBUG_MSG((hxgep, DMA_CTL,
166 	    "<== hxge_init_rxdma_channel_cntl_stat"));
167 
168 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
169 	rs = hpi_rxdma_control_status(handle, OP_SET, channel, cs_p);
170 
171 	if (rs != HPI_SUCCESS) {
172 		status = HXGE_ERROR | rs;
173 	}
174 	return (status);
175 }
176 
177 
178 hxge_status_t
179 hxge_enable_rxdma_channel(p_hxge_t hxgep, uint16_t channel,
180     p_rx_rbr_ring_t rbr_p, p_rx_rcr_ring_t rcr_p, p_rx_mbox_t mbox_p,
181     int n_init_kick)
182 {
183 	hpi_handle_t		handle;
184 	rdc_desc_cfg_t 		rdc_desc;
185 	rdc_rcr_cfg_b_t		*cfgb_p;
186 	hpi_status_t		rs = HPI_SUCCESS;
187 
188 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "==> hxge_enable_rxdma_channel"));
189 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
190 
191 	/*
192 	 * Use configuration data composed at init time. Write to hardware the
193 	 * receive ring configurations.
194 	 */
195 	rdc_desc.mbox_enable = 1;
196 	rdc_desc.mbox_addr = mbox_p->mbox_addr;
197 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
198 	    "==> hxge_enable_rxdma_channel: mboxp $%p($%p)",
199 	    mbox_p->mbox_addr, rdc_desc.mbox_addr));
200 
201 	rdc_desc.rbr_len = rbr_p->rbb_max;
202 	rdc_desc.rbr_addr = rbr_p->rbr_addr;
203 
204 	switch (hxgep->rx_bksize_code) {
205 	case RBR_BKSIZE_4K:
206 		rdc_desc.page_size = SIZE_4KB;
207 		break;
208 	case RBR_BKSIZE_8K:
209 		rdc_desc.page_size = SIZE_8KB;
210 		break;
211 	}
212 
213 	rdc_desc.size0 = rbr_p->hpi_pkt_buf_size0;
214 	rdc_desc.valid0 = 1;
215 
216 	rdc_desc.size1 = rbr_p->hpi_pkt_buf_size1;
217 	rdc_desc.valid1 = 1;
218 
219 	rdc_desc.size2 = rbr_p->hpi_pkt_buf_size2;
220 	rdc_desc.valid2 = 1;
221 
222 	rdc_desc.full_hdr = rcr_p->full_hdr_flag;
223 	rdc_desc.offset = rcr_p->sw_priv_hdr_len;
224 
225 	rdc_desc.rcr_len = rcr_p->comp_size;
226 	rdc_desc.rcr_addr = rcr_p->rcr_addr;
227 
228 	cfgb_p = &(rcr_p->rcr_cfgb);
229 	rdc_desc.rcr_threshold = cfgb_p->bits.pthres;
230 	rdc_desc.rcr_timeout = cfgb_p->bits.timeout;
231 	rdc_desc.rcr_timeout_enable = cfgb_p->bits.entout;
232 
233 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "==> hxge_enable_rxdma_channel: "
234 	    "rbr_len qlen %d pagesize code %d rcr_len %d",
235 	    rdc_desc.rbr_len, rdc_desc.page_size, rdc_desc.rcr_len));
236 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "==> hxge_enable_rxdma_channel: "
237 	    "size 0 %d size 1 %d size 2 %d",
238 	    rbr_p->hpi_pkt_buf_size0, rbr_p->hpi_pkt_buf_size1,
239 	    rbr_p->hpi_pkt_buf_size2));
240 
241 	rs = hpi_rxdma_cfg_rdc_ring(handle, rbr_p->rdc, &rdc_desc);
242 	if (rs != HPI_SUCCESS) {
243 		return (HXGE_ERROR | rs);
244 	}
245 
246 	/*
247 	 * Enable the timeout and threshold.
248 	 */
249 	rs = hpi_rxdma_cfg_rdc_rcr_threshold(handle, channel,
250 	    rdc_desc.rcr_threshold);
251 	if (rs != HPI_SUCCESS) {
252 		return (HXGE_ERROR | rs);
253 	}
254 
255 	rs = hpi_rxdma_cfg_rdc_rcr_timeout(handle, channel,
256 	    rdc_desc.rcr_timeout);
257 	if (rs != HPI_SUCCESS) {
258 		return (HXGE_ERROR | rs);
259 	}
260 
261 	/* Enable the DMA */
262 	rs = hpi_rxdma_cfg_rdc_enable(handle, channel);
263 	if (rs != HPI_SUCCESS) {
264 		return (HXGE_ERROR | rs);
265 	}
266 
267 	/* Kick the DMA engine */
268 	hpi_rxdma_rdc_rbr_kick(handle, channel, n_init_kick);
269 
270 	/* Clear the rbr empty bit */
271 	(void) hpi_rxdma_channel_rbr_empty_clear(handle, channel);
272 
273 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "<== hxge_enable_rxdma_channel"));
274 
275 	return (HXGE_OK);
276 }
277 
278 static hxge_status_t
279 hxge_disable_rxdma_channel(p_hxge_t hxgep, uint16_t channel)
280 {
281 	hpi_handle_t handle;
282 	hpi_status_t rs = HPI_SUCCESS;
283 
284 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "==> hxge_disable_rxdma_channel"));
285 
286 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
287 
288 	/* disable the DMA */
289 	rs = hpi_rxdma_cfg_rdc_disable(handle, channel);
290 	if (rs != HPI_SUCCESS) {
291 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
292 		    "<== hxge_disable_rxdma_channel:failed (0x%x)", rs));
293 		return (HXGE_ERROR | rs);
294 	}
295 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "<== hxge_disable_rxdma_channel"));
296 	return (HXGE_OK);
297 }
298 
299 hxge_status_t
300 hxge_rxdma_channel_rcrflush(p_hxge_t hxgep, uint8_t channel)
301 {
302 	hpi_handle_t	handle;
303 	hxge_status_t	status = HXGE_OK;
304 
305 	HXGE_DEBUG_MSG((hxgep, DMA_CTL,
306 	    "==> hxge_rxdma_channel_rcrflush"));
307 
308 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
309 	hpi_rxdma_rdc_rcr_flush(handle, channel);
310 
311 	HXGE_DEBUG_MSG((hxgep, DMA_CTL,
312 	    "<== hxge_rxdma_channel_rcrflush"));
313 	return (status);
314 
315 }
316 
317 #define	MID_INDEX(l, r) ((r + l + 1) >> 1)
318 
319 #define	TO_LEFT -1
320 #define	TO_RIGHT 1
321 #define	BOTH_RIGHT (TO_RIGHT + TO_RIGHT)
322 #define	BOTH_LEFT (TO_LEFT + TO_LEFT)
323 #define	IN_MIDDLE (TO_RIGHT + TO_LEFT)
324 #define	NO_HINT 0xffffffff
325 
326 /*ARGSUSED*/
327 hxge_status_t
328 hxge_rxbuf_pp_to_vp(p_hxge_t hxgep, p_rx_rbr_ring_t rbr_p,
329     uint8_t pktbufsz_type, uint64_t *pkt_buf_addr_pp,
330     uint64_t **pkt_buf_addr_p, uint32_t *bufoffset, uint32_t *msg_index)
331 {
332 	int			bufsize;
333 	uint64_t		pktbuf_pp;
334 	uint64_t		dvma_addr;
335 	rxring_info_t		*ring_info;
336 	int			base_side, end_side;
337 	int			r_index, l_index, anchor_index;
338 	int			found, search_done;
339 	uint32_t		offset, chunk_size, block_size, page_size_mask;
340 	uint32_t		chunk_index, block_index, total_index;
341 	int			max_iterations, iteration;
342 	rxbuf_index_info_t	*bufinfo;
343 
344 	HXGE_DEBUG_MSG((hxgep, RX2_CTL, "==> hxge_rxbuf_pp_to_vp"));
345 
346 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
347 	    "==> hxge_rxbuf_pp_to_vp: buf_pp $%p btype %d",
348 	    pkt_buf_addr_pp, pktbufsz_type));
349 
350 #if defined(__i386)
351 	pktbuf_pp = (uint64_t)(uint32_t)pkt_buf_addr_pp;
352 #else
353 	pktbuf_pp = (uint64_t)pkt_buf_addr_pp;
354 #endif
355 
356 	switch (pktbufsz_type) {
357 	case 0:
358 		bufsize = rbr_p->pkt_buf_size0;
359 		break;
360 	case 1:
361 		bufsize = rbr_p->pkt_buf_size1;
362 		break;
363 	case 2:
364 		bufsize = rbr_p->pkt_buf_size2;
365 		break;
366 	case RCR_SINGLE_BLOCK:
367 		bufsize = 0;
368 		anchor_index = 0;
369 		break;
370 	default:
371 		return (HXGE_ERROR);
372 	}
373 
374 	if (rbr_p->num_blocks == 1) {
375 		anchor_index = 0;
376 		ring_info = rbr_p->ring_info;
377 		bufinfo = (rxbuf_index_info_t *)ring_info->buffer;
378 
379 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
380 		    "==> hxge_rxbuf_pp_to_vp: (found, 1 block) "
381 		    "buf_pp $%p btype %d anchor_index %d bufinfo $%p",
382 		    pkt_buf_addr_pp, pktbufsz_type, anchor_index, bufinfo));
383 
384 		goto found_index;
385 	}
386 
387 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
388 	    "==> hxge_rxbuf_pp_to_vp: buf_pp $%p btype %d anchor_index %d",
389 	    pkt_buf_addr_pp, pktbufsz_type, anchor_index));
390 
391 	ring_info = rbr_p->ring_info;
392 	found = B_FALSE;
393 	bufinfo = (rxbuf_index_info_t *)ring_info->buffer;
394 	iteration = 0;
395 	max_iterations = ring_info->max_iterations;
396 
397 	/*
398 	 * First check if this block have been seen recently. This is indicated
399 	 * by a hint which is initialized when the first buffer of the block is
400 	 * seen. The hint is reset when the last buffer of the block has been
401 	 * processed. As three block sizes are supported, three hints are kept.
402 	 * The idea behind the hints is that once the hardware  uses a block
403 	 * for a buffer  of that size, it will use it exclusively for that size
404 	 * and will use it until it is exhausted. It is assumed that there
405 	 * would a single block being used for the same buffer sizes at any
406 	 * given time.
407 	 */
408 	if (ring_info->hint[pktbufsz_type] != NO_HINT) {
409 		anchor_index = ring_info->hint[pktbufsz_type];
410 		dvma_addr = bufinfo[anchor_index].dvma_addr;
411 		chunk_size = bufinfo[anchor_index].buf_size;
412 		if ((pktbuf_pp >= dvma_addr) &&
413 		    (pktbuf_pp < (dvma_addr + chunk_size))) {
414 			found = B_TRUE;
415 			/*
416 			 * check if this is the last buffer in the block If so,
417 			 * then reset the hint for the size;
418 			 */
419 
420 			if ((pktbuf_pp + bufsize) >= (dvma_addr + chunk_size))
421 				ring_info->hint[pktbufsz_type] = NO_HINT;
422 		}
423 	}
424 
425 	if (found == B_FALSE) {
426 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
427 		    "==> hxge_rxbuf_pp_to_vp: (!found)"
428 		    "buf_pp $%p btype %d anchor_index %d",
429 		    pkt_buf_addr_pp, pktbufsz_type, anchor_index));
430 
431 		/*
432 		 * This is the first buffer of the block of this size. Need to
433 		 * search the whole information array. the search algorithm
434 		 * uses a binary tree search algorithm. It assumes that the
435 		 * information is already sorted with increasing order info[0]
436 		 * < info[1] < info[2]  .... < info[n-1] where n is the size of
437 		 * the information array
438 		 */
439 		r_index = rbr_p->num_blocks - 1;
440 		l_index = 0;
441 		search_done = B_FALSE;
442 		anchor_index = MID_INDEX(r_index, l_index);
443 		while (search_done == B_FALSE) {
444 			if ((r_index == l_index) ||
445 			    (iteration >= max_iterations))
446 				search_done = B_TRUE;
447 
448 			end_side = TO_RIGHT;	/* to the right */
449 			base_side = TO_LEFT;	/* to the left */
450 			/* read the DVMA address information and sort it */
451 			dvma_addr = bufinfo[anchor_index].dvma_addr;
452 			chunk_size = bufinfo[anchor_index].buf_size;
453 
454 			HXGE_DEBUG_MSG((hxgep, RX2_CTL,
455 			    "==> hxge_rxbuf_pp_to_vp: (searching)"
456 			    "buf_pp $%p btype %d "
457 			    "anchor_index %d chunk_size %d dvmaaddr $%p",
458 			    pkt_buf_addr_pp, pktbufsz_type, anchor_index,
459 			    chunk_size, dvma_addr));
460 
461 			if (pktbuf_pp >= dvma_addr)
462 				base_side = TO_RIGHT;	/* to the right */
463 			if (pktbuf_pp < (dvma_addr + chunk_size))
464 				end_side = TO_LEFT;	/* to the left */
465 
466 			switch (base_side + end_side) {
467 			case IN_MIDDLE:
468 				/* found */
469 				found = B_TRUE;
470 				search_done = B_TRUE;
471 				if ((pktbuf_pp + bufsize) <
472 				    (dvma_addr + chunk_size))
473 					ring_info->hint[pktbufsz_type] =
474 					    bufinfo[anchor_index].buf_index;
475 				break;
476 			case BOTH_RIGHT:
477 				/* not found: go to the right */
478 				l_index = anchor_index + 1;
479 				anchor_index = MID_INDEX(r_index, l_index);
480 				break;
481 
482 			case BOTH_LEFT:
483 				/* not found: go to the left */
484 				r_index = anchor_index - 1;
485 				anchor_index = MID_INDEX(r_index, l_index);
486 				break;
487 			default:	/* should not come here */
488 				return (HXGE_ERROR);
489 			}
490 			iteration++;
491 		}
492 
493 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
494 		    "==> hxge_rxbuf_pp_to_vp: (search done)"
495 		    "buf_pp $%p btype %d anchor_index %d",
496 		    pkt_buf_addr_pp, pktbufsz_type, anchor_index));
497 	}
498 
499 	if (found == B_FALSE) {
500 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
501 		    "==> hxge_rxbuf_pp_to_vp: (search failed)"
502 		    "buf_pp $%p btype %d anchor_index %d",
503 		    pkt_buf_addr_pp, pktbufsz_type, anchor_index));
504 		return (HXGE_ERROR);
505 	}
506 
507 found_index:
508 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
509 	    "==> hxge_rxbuf_pp_to_vp: (FOUND1)"
510 	    "buf_pp $%p btype %d bufsize %d anchor_index %d",
511 	    pkt_buf_addr_pp, pktbufsz_type, bufsize, anchor_index));
512 
513 	/* index of the first block in this chunk */
514 	chunk_index = bufinfo[anchor_index].start_index;
515 	dvma_addr = bufinfo[anchor_index].dvma_addr;
516 	page_size_mask = ring_info->block_size_mask;
517 
518 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
519 	    "==> hxge_rxbuf_pp_to_vp: (FOUND3), get chunk)"
520 	    "buf_pp $%p btype %d bufsize %d "
521 	    "anchor_index %d chunk_index %d dvma $%p",
522 	    pkt_buf_addr_pp, pktbufsz_type, bufsize,
523 	    anchor_index, chunk_index, dvma_addr));
524 
525 	offset = pktbuf_pp - dvma_addr;	/* offset within the chunk */
526 	block_size = rbr_p->block_size;	/* System  block(page) size */
527 
528 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
529 	    "==> hxge_rxbuf_pp_to_vp: (FOUND4), get chunk)"
530 	    "buf_pp $%p btype %d bufsize %d "
531 	    "anchor_index %d chunk_index %d dvma $%p "
532 	    "offset %d block_size %d",
533 	    pkt_buf_addr_pp, pktbufsz_type, bufsize, anchor_index,
534 	    chunk_index, dvma_addr, offset, block_size));
535 	HXGE_DEBUG_MSG((hxgep, RX2_CTL, "==> getting total index"));
536 
537 	block_index = (offset / block_size);	/* index within chunk */
538 	total_index = chunk_index + block_index;
539 
540 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
541 	    "==> hxge_rxbuf_pp_to_vp: "
542 	    "total_index %d dvma_addr $%p "
543 	    "offset %d block_size %d "
544 	    "block_index %d ",
545 	    total_index, dvma_addr, offset, block_size, block_index));
546 
547 #if defined(__i386)
548 	*pkt_buf_addr_p = (uint64_t *)((uint32_t)bufinfo[anchor_index].kaddr +
549 	    (uint32_t)offset);
550 #else
551 	*pkt_buf_addr_p = (uint64_t *)((uint64_t)bufinfo[anchor_index].kaddr +
552 	    offset);
553 #endif
554 
555 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
556 	    "==> hxge_rxbuf_pp_to_vp: "
557 	    "total_index %d dvma_addr $%p "
558 	    "offset %d block_size %d "
559 	    "block_index %d "
560 	    "*pkt_buf_addr_p $%p",
561 	    total_index, dvma_addr, offset, block_size,
562 	    block_index, *pkt_buf_addr_p));
563 
564 	*msg_index = total_index;
565 	*bufoffset = (offset & page_size_mask);
566 
567 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
568 	    "==> hxge_rxbuf_pp_to_vp: get msg index: "
569 	    "msg_index %d bufoffset_index %d",
570 	    *msg_index, *bufoffset));
571 	HXGE_DEBUG_MSG((hxgep, RX2_CTL, "<== hxge_rxbuf_pp_to_vp"));
572 
573 	return (HXGE_OK);
574 }
575 
576 
577 /*
578  * used by quick sort (qsort) function
579  * to perform comparison
580  */
581 static int
582 hxge_sort_compare(const void *p1, const void *p2)
583 {
584 
585 	rxbuf_index_info_t *a, *b;
586 
587 	a = (rxbuf_index_info_t *)p1;
588 	b = (rxbuf_index_info_t *)p2;
589 
590 	if (a->dvma_addr > b->dvma_addr)
591 		return (1);
592 	if (a->dvma_addr < b->dvma_addr)
593 		return (-1);
594 	return (0);
595 }
596 
597 /*
598  * Grabbed this sort implementation from common/syscall/avl.c
599  *
600  * Generic shellsort, from K&R (1st ed, p 58.), somewhat modified.
601  * v = Ptr to array/vector of objs
602  * n = # objs in the array
603  * s = size of each obj (must be multiples of a word size)
604  * f = ptr to function to compare two objs
605  *	returns (-1 = less than, 0 = equal, 1 = greater than
606  */
607 void
608 hxge_ksort(caddr_t v, int n, int s, int (*f) ())
609 {
610 	int		g, i, j, ii;
611 	unsigned int	*p1, *p2;
612 	unsigned int	tmp;
613 
614 	/* No work to do */
615 	if (v == NULL || n <= 1)
616 		return;
617 	/* Sanity check on arguments */
618 	ASSERT(((uintptr_t)v & 0x3) == 0 && (s & 0x3) == 0);
619 	ASSERT(s > 0);
620 
621 	for (g = n / 2; g > 0; g /= 2) {
622 		for (i = g; i < n; i++) {
623 			for (j = i - g; j >= 0 &&
624 			    (*f) (v + j * s, v + (j + g) * s) == 1; j -= g) {
625 				p1 = (unsigned *)(v + j * s);
626 				p2 = (unsigned *)(v + (j + g) * s);
627 				for (ii = 0; ii < s / 4; ii++) {
628 					tmp = *p1;
629 					*p1++ = *p2;
630 					*p2++ = tmp;
631 				}
632 			}
633 		}
634 	}
635 }
636 
637 /*
638  * Initialize data structures required for rxdma
639  * buffer dvma->vmem address lookup
640  */
641 /*ARGSUSED*/
642 static hxge_status_t
643 hxge_rxbuf_index_info_init(p_hxge_t hxgep, p_rx_rbr_ring_t rbrp)
644 {
645 	int		index;
646 	rxring_info_t	*ring_info;
647 	int		max_iteration = 0, max_index = 0;
648 
649 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "==> hxge_rxbuf_index_info_init"));
650 
651 	ring_info = rbrp->ring_info;
652 	ring_info->hint[0] = NO_HINT;
653 	ring_info->hint[1] = NO_HINT;
654 	ring_info->hint[2] = NO_HINT;
655 	max_index = rbrp->num_blocks;
656 
657 	/* read the DVMA address information and sort it */
658 	/* do init of the information array */
659 
660 	HXGE_DEBUG_MSG((hxgep, DMA2_CTL,
661 	    " hxge_rxbuf_index_info_init Sort ptrs"));
662 
663 	/* sort the array */
664 	hxge_ksort((void *) ring_info->buffer, max_index,
665 	    sizeof (rxbuf_index_info_t), hxge_sort_compare);
666 
667 	for (index = 0; index < max_index; index++) {
668 		HXGE_DEBUG_MSG((hxgep, DMA2_CTL,
669 		    " hxge_rxbuf_index_info_init: sorted chunk %d "
670 		    " ioaddr $%p kaddr $%p size %x",
671 		    index, ring_info->buffer[index].dvma_addr,
672 		    ring_info->buffer[index].kaddr,
673 		    ring_info->buffer[index].buf_size));
674 	}
675 
676 	max_iteration = 0;
677 	while (max_index >= (1ULL << max_iteration))
678 		max_iteration++;
679 	ring_info->max_iterations = max_iteration + 1;
680 
681 	HXGE_DEBUG_MSG((hxgep, DMA2_CTL,
682 	    " hxge_rxbuf_index_info_init Find max iter %d",
683 	    ring_info->max_iterations));
684 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "<== hxge_rxbuf_index_info_init"));
685 
686 	return (HXGE_OK);
687 }
688 
689 /*ARGSUSED*/
690 void
691 hxge_dump_rcr_entry(p_hxge_t hxgep, p_rcr_entry_t entry_p)
692 {
693 #ifdef	HXGE_DEBUG
694 
695 	uint32_t bptr;
696 	uint64_t pp;
697 
698 	bptr = entry_p->bits.pkt_buf_addr;
699 
700 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
701 	    "\trcr entry $%p "
702 	    "\trcr entry 0x%0llx "
703 	    "\trcr entry 0x%08x "
704 	    "\trcr entry 0x%08x "
705 	    "\tvalue 0x%0llx\n"
706 	    "\tmulti = %d\n"
707 	    "\tpkt_type = 0x%x\n"
708 	    "\terror = 0x%04x\n"
709 	    "\tl2_len = %d\n"
710 	    "\tpktbufsize = %d\n"
711 	    "\tpkt_buf_addr = $%p\n"
712 	    "\tpkt_buf_addr (<< 6) = $%p\n",
713 	    entry_p,
714 	    *(int64_t *)entry_p,
715 	    *(int32_t *)entry_p,
716 	    *(int32_t *)((char *)entry_p + 32),
717 	    entry_p->value,
718 	    entry_p->bits.multi,
719 	    entry_p->bits.pkt_type,
720 	    entry_p->bits.error,
721 	    entry_p->bits.l2_len,
722 	    entry_p->bits.pktbufsz,
723 	    bptr,
724 	    entry_p->bits.pkt_buf_addr_l));
725 
726 	pp = (entry_p->value & RCR_PKT_BUF_ADDR_MASK) <<
727 	    RCR_PKT_BUF_ADDR_SHIFT;
728 
729 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "rcr pp 0x%llx l2 len %d",
730 	    pp, (*(int64_t *)entry_p >> 40) & 0x3fff));
731 #endif
732 }
733 
734 /*ARGSUSED*/
735 void
736 hxge_rxdma_stop(p_hxge_t hxgep)
737 {
738 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rxdma_stop"));
739 
740 	(void) hxge_rx_vmac_disable(hxgep);
741 	(void) hxge_rxdma_hw_mode(hxgep, HXGE_DMA_STOP);
742 
743 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_rxdma_stop"));
744 }
745 
746 void
747 hxge_rxdma_stop_reinit(p_hxge_t hxgep)
748 {
749 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rxdma_stop_reinit"));
750 
751 	(void) hxge_rxdma_stop(hxgep);
752 	(void) hxge_uninit_rxdma_channels(hxgep);
753 	(void) hxge_init_rxdma_channels(hxgep);
754 
755 	(void) hxge_rx_vmac_enable(hxgep);
756 
757 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_rxdma_stop_reinit"));
758 }
759 
760 hxge_status_t
761 hxge_rxdma_hw_mode(p_hxge_t hxgep, boolean_t enable)
762 {
763 	int			i, ndmas;
764 	uint16_t		channel;
765 	p_rx_rbr_rings_t	rx_rbr_rings;
766 	p_rx_rbr_ring_t		*rbr_rings;
767 	hpi_handle_t		handle;
768 	hpi_status_t		rs = HPI_SUCCESS;
769 	hxge_status_t		status = HXGE_OK;
770 
771 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
772 	    "==> hxge_rxdma_hw_mode: mode %d", enable));
773 
774 	if (!(hxgep->drv_state & STATE_HW_INITIALIZED)) {
775 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
776 		    "<== hxge_rxdma_mode: not initialized"));
777 		return (HXGE_ERROR);
778 	}
779 
780 	rx_rbr_rings = hxgep->rx_rbr_rings;
781 	if (rx_rbr_rings == NULL) {
782 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
783 		    "<== hxge_rxdma_mode: NULL ring pointer"));
784 		return (HXGE_ERROR);
785 	}
786 
787 	if (rx_rbr_rings->rbr_rings == NULL) {
788 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
789 		    "<== hxge_rxdma_mode: NULL rbr rings pointer"));
790 		return (HXGE_ERROR);
791 	}
792 
793 	ndmas = rx_rbr_rings->ndmas;
794 	if (!ndmas) {
795 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
796 		    "<== hxge_rxdma_mode: no channel"));
797 		return (HXGE_ERROR);
798 	}
799 
800 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
801 	    "==> hxge_rxdma_mode (ndmas %d)", ndmas));
802 
803 	rbr_rings = rx_rbr_rings->rbr_rings;
804 
805 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
806 
807 	for (i = 0; i < ndmas; i++) {
808 		if (rbr_rings == NULL || rbr_rings[i] == NULL) {
809 			continue;
810 		}
811 		channel = rbr_rings[i]->rdc;
812 		if (enable) {
813 			HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
814 			    "==> hxge_rxdma_hw_mode: channel %d (enable)",
815 			    channel));
816 			rs = hpi_rxdma_cfg_rdc_enable(handle, channel);
817 		} else {
818 			HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
819 			    "==> hxge_rxdma_hw_mode: channel %d (disable)",
820 			    channel));
821 			rs = hpi_rxdma_cfg_rdc_disable(handle, channel);
822 		}
823 	}
824 
825 	status = ((rs == HPI_SUCCESS) ? HXGE_OK : HXGE_ERROR | rs);
826 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
827 	    "<== hxge_rxdma_hw_mode: status 0x%x", status));
828 
829 	return (status);
830 }
831 
832 int
833 hxge_rxdma_get_ring_index(p_hxge_t hxgep, uint16_t channel)
834 {
835 	int			i, ndmas;
836 	uint16_t		rdc;
837 	p_rx_rbr_rings_t 	rx_rbr_rings;
838 	p_rx_rbr_ring_t		*rbr_rings;
839 
840 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
841 	    "==> hxge_rxdma_get_ring_index: channel %d", channel));
842 
843 	rx_rbr_rings = hxgep->rx_rbr_rings;
844 	if (rx_rbr_rings == NULL) {
845 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
846 		    "<== hxge_rxdma_get_ring_index: NULL ring pointer"));
847 		return (-1);
848 	}
849 
850 	ndmas = rx_rbr_rings->ndmas;
851 	if (!ndmas) {
852 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
853 		    "<== hxge_rxdma_get_ring_index: no channel"));
854 		return (-1);
855 	}
856 
857 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
858 	    "==> hxge_rxdma_get_ring_index (ndmas %d)", ndmas));
859 
860 	rbr_rings = rx_rbr_rings->rbr_rings;
861 	for (i = 0; i < ndmas; i++) {
862 		rdc = rbr_rings[i]->rdc;
863 		if (channel == rdc) {
864 			HXGE_DEBUG_MSG((hxgep, RX_CTL,
865 			    "==> hxge_rxdma_get_rbr_ring: "
866 			    "channel %d (index %d) "
867 			    "ring %d", channel, i, rbr_rings[i]));
868 
869 			return (i);
870 		}
871 	}
872 
873 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
874 	    "<== hxge_rxdma_get_rbr_ring_index: not found"));
875 
876 	return (-1);
877 }
878 
879 /*
880  * Static functions start here.
881  */
882 static p_rx_msg_t
883 hxge_allocb(size_t size, uint32_t pri, p_hxge_dma_common_t dmabuf_p)
884 {
885 	p_rx_msg_t		hxge_mp = NULL;
886 	p_hxge_dma_common_t	dmamsg_p;
887 	uchar_t			*buffer;
888 
889 	hxge_mp = KMEM_ZALLOC(sizeof (rx_msg_t), KM_NOSLEEP);
890 	if (hxge_mp == NULL) {
891 		HXGE_ERROR_MSG((NULL, HXGE_ERR_CTL,
892 		    "Allocation of a rx msg failed."));
893 		goto hxge_allocb_exit;
894 	}
895 
896 	hxge_mp->use_buf_pool = B_FALSE;
897 	if (dmabuf_p) {
898 		hxge_mp->use_buf_pool = B_TRUE;
899 
900 		dmamsg_p = (p_hxge_dma_common_t)&hxge_mp->buf_dma;
901 		*dmamsg_p = *dmabuf_p;
902 		dmamsg_p->nblocks = 1;
903 		dmamsg_p->block_size = size;
904 		dmamsg_p->alength = size;
905 		buffer = (uchar_t *)dmabuf_p->kaddrp;
906 
907 		dmabuf_p->kaddrp = (void *)((char *)dmabuf_p->kaddrp + size);
908 		dmabuf_p->ioaddr_pp = (void *)
909 		    ((char *)dmabuf_p->ioaddr_pp + size);
910 
911 		dmabuf_p->alength -= size;
912 		dmabuf_p->offset += size;
913 		dmabuf_p->dma_cookie.dmac_laddress += size;
914 		dmabuf_p->dma_cookie.dmac_size -= size;
915 	} else {
916 		buffer = KMEM_ALLOC(size, KM_NOSLEEP);
917 		if (buffer == NULL) {
918 			HXGE_ERROR_MSG((NULL, HXGE_ERR_CTL,
919 			    "Allocation of a receive page failed."));
920 			goto hxge_allocb_fail1;
921 		}
922 	}
923 
924 	hxge_mp->rx_mblk_p = desballoc(buffer, size, pri, &hxge_mp->freeb);
925 	if (hxge_mp->rx_mblk_p == NULL) {
926 		HXGE_ERROR_MSG((NULL, HXGE_ERR_CTL, "desballoc failed."));
927 		goto hxge_allocb_fail2;
928 	}
929 	hxge_mp->buffer = buffer;
930 	hxge_mp->block_size = size;
931 	hxge_mp->freeb.free_func = (void (*) ()) hxge_freeb;
932 	hxge_mp->freeb.free_arg = (caddr_t)hxge_mp;
933 	hxge_mp->ref_cnt = 1;
934 	hxge_mp->free = B_TRUE;
935 	hxge_mp->rx_use_bcopy = B_FALSE;
936 
937 	atomic_inc_32(&hxge_mblks_pending);
938 
939 	goto hxge_allocb_exit;
940 
941 hxge_allocb_fail2:
942 	if (!hxge_mp->use_buf_pool) {
943 		KMEM_FREE(buffer, size);
944 	}
945 hxge_allocb_fail1:
946 	KMEM_FREE(hxge_mp, sizeof (rx_msg_t));
947 	hxge_mp = NULL;
948 
949 hxge_allocb_exit:
950 	return (hxge_mp);
951 }
952 
953 p_mblk_t
954 hxge_dupb(p_rx_msg_t hxge_mp, uint_t offset, size_t size)
955 {
956 	p_mblk_t mp;
957 
958 	HXGE_DEBUG_MSG((NULL, MEM_CTL, "==> hxge_dupb"));
959 	HXGE_DEBUG_MSG((NULL, MEM_CTL, "hxge_mp = $%p "
960 	    "offset = 0x%08X " "size = 0x%08X", hxge_mp, offset, size));
961 
962 	mp = desballoc(&hxge_mp->buffer[offset], size, 0, &hxge_mp->freeb);
963 	if (mp == NULL) {
964 		HXGE_DEBUG_MSG((NULL, RX_CTL, "desballoc failed"));
965 		goto hxge_dupb_exit;
966 	}
967 
968 	atomic_inc_32(&hxge_mp->ref_cnt);
969 
970 hxge_dupb_exit:
971 	HXGE_DEBUG_MSG((NULL, MEM_CTL, "<== hxge_dupb mp = $%p", hxge_mp));
972 	return (mp);
973 }
974 
975 p_mblk_t
976 hxge_dupb_bcopy(p_rx_msg_t hxge_mp, uint_t offset, size_t size)
977 {
978 	p_mblk_t	mp;
979 	uchar_t		*dp;
980 
981 	mp = allocb(size + HXGE_RXBUF_EXTRA, 0);
982 	if (mp == NULL) {
983 		HXGE_DEBUG_MSG((NULL, RX_CTL, "desballoc failed"));
984 		goto hxge_dupb_bcopy_exit;
985 	}
986 	dp = mp->b_rptr = mp->b_rptr + HXGE_RXBUF_EXTRA;
987 	bcopy((void *) &hxge_mp->buffer[offset], dp, size);
988 	mp->b_wptr = dp + size;
989 
990 hxge_dupb_bcopy_exit:
991 
992 	HXGE_DEBUG_MSG((NULL, MEM_CTL, "<== hxge_dupb mp = $%p", hxge_mp));
993 
994 	return (mp);
995 }
996 
997 void hxge_post_page(p_hxge_t hxgep, p_rx_rbr_ring_t rx_rbr_p,
998     p_rx_msg_t rx_msg_p);
999 
1000 void
1001 hxge_post_page(p_hxge_t hxgep, p_rx_rbr_ring_t rx_rbr_p, p_rx_msg_t rx_msg_p)
1002 {
1003 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_post_page"));
1004 
1005 	/* Reuse this buffer */
1006 	rx_msg_p->free = B_FALSE;
1007 	rx_msg_p->cur_usage_cnt = 0;
1008 	rx_msg_p->max_usage_cnt = 0;
1009 	rx_msg_p->pkt_buf_size = 0;
1010 
1011 	if (rx_rbr_p->rbr_use_bcopy) {
1012 		rx_msg_p->rx_use_bcopy = B_FALSE;
1013 		atomic_dec_32(&rx_rbr_p->rbr_consumed);
1014 	}
1015 	atomic_dec_32(&rx_rbr_p->rbr_used);
1016 
1017 	/*
1018 	 * Get the rbr header pointer and its offset index.
1019 	 */
1020 	rx_rbr_p->rbr_wr_index = ((rx_rbr_p->rbr_wr_index + 1) &
1021 	    rx_rbr_p->rbr_wrap_mask);
1022 	rx_rbr_p->rbr_desc_vp[rx_rbr_p->rbr_wr_index] = rx_msg_p->shifted_addr;
1023 
1024 	/*
1025 	 * Accumulate some buffers in the ring before re-enabling the
1026 	 * DMA channel, if rbr empty was signaled.
1027 	 */
1028 	hpi_rxdma_rdc_rbr_kick(HXGE_DEV_HPI_HANDLE(hxgep), rx_rbr_p->rdc, 1);
1029 	if (rx_rbr_p->rbr_is_empty &&
1030 	    (rx_rbr_p->rbb_max - rx_rbr_p->rbr_used) >=
1031 	    HXGE_RBR_EMPTY_THRESHOLD) {
1032 		hxge_rbr_empty_restore(hxgep, rx_rbr_p);
1033 	}
1034 
1035 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
1036 	    "<== hxge_post_page (channel %d post_next_index %d)",
1037 	    rx_rbr_p->rdc, rx_rbr_p->rbr_wr_index));
1038 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_post_page"));
1039 }
1040 
1041 void
1042 hxge_freeb(p_rx_msg_t rx_msg_p)
1043 {
1044 	size_t		size;
1045 	uchar_t		*buffer = NULL;
1046 	int		ref_cnt;
1047 	boolean_t	free_state = B_FALSE;
1048 	rx_rbr_ring_t	*ring = rx_msg_p->rx_rbr_p;
1049 
1050 	HXGE_DEBUG_MSG((NULL, MEM2_CTL, "==> hxge_freeb"));
1051 	HXGE_DEBUG_MSG((NULL, MEM2_CTL,
1052 	    "hxge_freeb:rx_msg_p = $%p (block pending %d)",
1053 	    rx_msg_p, hxge_mblks_pending));
1054 
1055 	if (ring == NULL)
1056 		return;
1057 
1058 	/*
1059 	 * This is to prevent posting activities while we are recovering
1060 	 * from fatal errors. This should not be a performance drag since
1061 	 * ref_cnt != 0 most times.
1062 	 */
1063 	if (ring->rbr_state == RBR_POSTING)
1064 		MUTEX_ENTER(&ring->post_lock);
1065 
1066 	/*
1067 	 * First we need to get the free state, then
1068 	 * atomic decrement the reference count to prevent
1069 	 * the race condition with the interrupt thread that
1070 	 * is processing a loaned up buffer block.
1071 	 */
1072 	free_state = rx_msg_p->free;
1073 	ref_cnt = atomic_add_32_nv(&rx_msg_p->ref_cnt, -1);
1074 	if (!ref_cnt) {
1075 		atomic_dec_32(&hxge_mblks_pending);
1076 
1077 		buffer = rx_msg_p->buffer;
1078 		size = rx_msg_p->block_size;
1079 
1080 		HXGE_DEBUG_MSG((NULL, MEM2_CTL, "hxge_freeb: "
1081 		    "will free: rx_msg_p = $%p (block pending %d)",
1082 		    rx_msg_p, hxge_mblks_pending));
1083 
1084 		if (!rx_msg_p->use_buf_pool) {
1085 			KMEM_FREE(buffer, size);
1086 		}
1087 
1088 		KMEM_FREE(rx_msg_p, sizeof (rx_msg_t));
1089 		/*
1090 		 * Decrement the receive buffer ring's reference
1091 		 * count, too.
1092 		 */
1093 		atomic_dec_32(&ring->rbr_ref_cnt);
1094 
1095 		/*
1096 		 * Free the receive buffer ring, iff
1097 		 * 1. all the receive buffers have been freed
1098 		 * 2. and we are in the proper state (that is,
1099 		 *    we are not UNMAPPING).
1100 		 */
1101 		if (ring->rbr_ref_cnt == 0 &&
1102 		    ring->rbr_state == RBR_UNMAPPED) {
1103 			KMEM_FREE(ring, sizeof (*ring));
1104 			/* post_lock has been destroyed already */
1105 			return;
1106 		}
1107 	}
1108 
1109 	/*
1110 	 * Repost buffer.
1111 	 */
1112 	if (free_state && (ref_cnt == 1)) {
1113 		HXGE_DEBUG_MSG((NULL, RX_CTL,
1114 		    "hxge_freeb: post page $%p:", rx_msg_p));
1115 		if (ring->rbr_state == RBR_POSTING)
1116 			hxge_post_page(rx_msg_p->hxgep, ring, rx_msg_p);
1117 	}
1118 
1119 	if (ring->rbr_state == RBR_POSTING)
1120 		MUTEX_EXIT(&ring->post_lock);
1121 
1122 	HXGE_DEBUG_MSG((NULL, MEM2_CTL, "<== hxge_freeb"));
1123 }
1124 
1125 uint_t
1126 hxge_rx_intr(caddr_t arg1, caddr_t arg2)
1127 {
1128 	p_hxge_ring_handle_t	rhp;
1129 	p_hxge_ldv_t		ldvp = (p_hxge_ldv_t)arg1;
1130 	p_hxge_t		hxgep = (p_hxge_t)arg2;
1131 	p_hxge_ldg_t		ldgp;
1132 	uint8_t			channel;
1133 	hpi_handle_t		handle;
1134 	rdc_stat_t		cs;
1135 	uint_t			serviced = DDI_INTR_UNCLAIMED;
1136 	p_rx_rcr_ring_t		ring;
1137 	mblk_t			*mp;
1138 
1139 	if (ldvp == NULL) {
1140 		HXGE_DEBUG_MSG((NULL, RX_INT_CTL,
1141 		    "<== hxge_rx_intr: arg2 $%p arg1 $%p", hxgep, ldvp));
1142 		return (DDI_INTR_UNCLAIMED);
1143 	}
1144 
1145 	if (arg2 == NULL || (void *) ldvp->hxgep != arg2) {
1146 		hxgep = ldvp->hxgep;
1147 	}
1148 
1149 	/*
1150 	 * If the interface is not started, just swallow the interrupt
1151 	 * for the logical device and don't rearm it.
1152 	 */
1153 	if (hxgep->hxge_mac_state != HXGE_MAC_STARTED)
1154 		return (DDI_INTR_CLAIMED);
1155 
1156 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1157 	    "==> hxge_rx_intr: arg2 $%p arg1 $%p", hxgep, ldvp));
1158 
1159 	/*
1160 	 * This interrupt handler is for a specific receive dma channel.
1161 	 */
1162 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
1163 
1164 	/*
1165 	 * Get the control and status for this channel.
1166 	 */
1167 	channel = ldvp->vdma_index;
1168 	ring = hxgep->rx_rcr_rings->rcr_rings[channel];
1169 	rhp = &hxgep->rx_ring_handles[channel];
1170 
1171 	MUTEX_ENTER(&ring->lock);
1172 
1173 	/*
1174 	 * If the channel is not started, then we are not
1175 	 * ready to process packets.
1176 	 */
1177 	if (!rhp->started) {
1178 		MUTEX_EXIT(&ring->lock);
1179 		return (DDI_INTR_CLAIMED);
1180 	}
1181 
1182 	ldgp = ldvp->ldgp;
1183 	RXDMA_REG_READ64(handle, RDC_STAT, channel, &cs.value);
1184 	cs.bits.ptrread = 0;
1185 	cs.bits.pktread = 0;
1186 	RXDMA_REG_WRITE64(handle, RDC_STAT, channel, cs.value);
1187 
1188 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL, "==> hxge_rx_intr:channel %d "
1189 	    "cs 0x%016llx rcrto 0x%x rcrthres %x",
1190 	    channel, cs.value, cs.bits.rcr_to, cs.bits.rcr_thres));
1191 
1192 	mp = hxge_rx_pkts(hxgep, ldvp->vdma_index, ldvp, ring, cs, -1);
1193 	serviced = DDI_INTR_CLAIMED;
1194 	MUTEX_EXIT(&ring->lock);
1195 
1196 	/* error events. */
1197 	if (cs.value & RDC_STAT_ERROR) {
1198 		(void) hxge_rx_err_evnts(hxgep, ldvp->vdma_index, ldvp, cs);
1199 	}
1200 
1201 hxge_intr_exit:
1202 	/*
1203 	 * Enable the mailbox update interrupt if we want to use mailbox. We
1204 	 * probably don't need to use mailbox as it only saves us one pio read.
1205 	 * Also write 1 to rcrthres and rcrto to clear these two edge triggered
1206 	 * bits.
1207 	 */
1208 	cs.value &= RDC_STAT_WR1C;
1209 	cs.bits.mex = 1;
1210 	cs.bits.ptrread = 0;
1211 	cs.bits.pktread = 0;
1212 	RXDMA_REG_WRITE64(handle, RDC_STAT, channel, cs.value);
1213 
1214 	/*
1215 	 * Rearm this logical group if this is a single device group.
1216 	 */
1217 	MUTEX_ENTER(&ring->lock);
1218 	if (ring->poll_flag) {
1219 		if (ldgp->nldvs == 1) {
1220 			ld_intr_mgmt_t mgm;
1221 
1222 			mgm.value = 0;
1223 			mgm.bits.arm = 0;
1224 			HXGE_REG_WR32(handle,
1225 			    LD_INTR_MGMT + LDSV_OFFSET(ldgp->ldg), mgm.value);
1226 		}
1227 	} else if (ldgp->nldvs == 1) {
1228 		ld_intr_mgmt_t mgm;
1229 
1230 		mgm.value = 0;
1231 		mgm.bits.arm = 1;
1232 		mgm.bits.timer = ldgp->ldg_timer;
1233 		HXGE_REG_WR32(handle,
1234 		    LD_INTR_MGMT + LDSV_OFFSET(ldgp->ldg), mgm.value);
1235 	}
1236 	MUTEX_EXIT(&ring->lock);
1237 
1238 	/*
1239 	 * Send the packets up the stack.
1240 	 */
1241 	if (mp != NULL) {
1242 		mac_rx_ring(hxgep->mach, ring->rcr_mac_handle, mp,
1243 		    ring->rcr_gen_num);
1244 	}
1245 
1246 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1247 	    "<== hxge_rx_intr: serviced %d", serviced));
1248 	return (serviced);
1249 }
1250 
1251 /*
1252  * Enable polling for a ring. Interrupt for the ring is disabled when
1253  * the hxge interrupt comes (see hxge_rx_intr).
1254  */
1255 int
1256 hxge_enable_poll(void *arg)
1257 {
1258 	p_hxge_ring_handle_t	ring_handle = (p_hxge_ring_handle_t)arg;
1259 	p_rx_rcr_ring_t		ringp;
1260 	p_hxge_t		hxgep;
1261 	p_hxge_ldg_t		ldgp;
1262 
1263 	if (ring_handle == NULL) {
1264 		return (0);
1265 	}
1266 
1267 	hxgep = ring_handle->hxgep;
1268 	ringp = hxgep->rx_rcr_rings->rcr_rings[ring_handle->index];
1269 
1270 	MUTEX_ENTER(&ringp->lock);
1271 
1272 	ldgp = ringp->ldgp;
1273 	if (ldgp == NULL) {
1274 		MUTEX_EXIT(&ringp->lock);
1275 		return (0);
1276 	}
1277 
1278 	/*
1279 	 * Enable polling
1280 	 */
1281 	if (ringp->poll_flag == 0) {
1282 		ringp->poll_flag = 1;
1283 	}
1284 
1285 	MUTEX_EXIT(&ringp->lock);
1286 	return (0);
1287 }
1288 
1289 /*
1290  * Disable polling for a ring and enable its interrupt.
1291  */
1292 int
1293 hxge_disable_poll(void *arg)
1294 {
1295 	p_hxge_ring_handle_t	ring_handle = (p_hxge_ring_handle_t)arg;
1296 	p_rx_rcr_ring_t		ringp;
1297 	p_hxge_t		hxgep;
1298 
1299 	if (ring_handle == NULL) {
1300 		return (0);
1301 	}
1302 
1303 	hxgep = ring_handle->hxgep;
1304 	ringp = hxgep->rx_rcr_rings->rcr_rings[ring_handle->index];
1305 
1306 	MUTEX_ENTER(&ringp->lock);
1307 
1308 	/*
1309 	 * Disable polling: enable interrupt
1310 	 */
1311 	if (ringp->poll_flag) {
1312 		hpi_handle_t		handle;
1313 		rdc_stat_t		cs;
1314 		uint8_t			channel;
1315 		p_hxge_ldg_t		ldgp;
1316 
1317 		/*
1318 		 * Get the control and status for this channel.
1319 		 */
1320 		handle = HXGE_DEV_HPI_HANDLE(hxgep);
1321 		channel = ringp->rdc;
1322 		RXDMA_REG_READ64(handle, RDC_STAT, channel, &cs.value);
1323 
1324 		/*
1325 		 * Enable mailbox update
1326 		 * Since packets were not read and the hardware uses
1327 		 * bits pktread and ptrread to update the queue
1328 		 * length, we need to set both bits to 0.
1329 		 */
1330 		cs.bits.pktread = 0;
1331 		cs.bits.ptrread = 0;
1332 		cs.bits.mex = 1;
1333 		RXDMA_REG_WRITE64(handle, RDC_STAT, channel, cs.value);
1334 
1335 		/*
1336 		 * Rearm this logical group if this is a single device
1337 		 * group.
1338 		 */
1339 		ldgp = ringp->ldgp;
1340 		if (ldgp == NULL) {
1341 			ringp->poll_flag = 0;
1342 			MUTEX_EXIT(&ringp->lock);
1343 			return (0);
1344 		}
1345 
1346 		if (ldgp->nldvs == 1) {
1347 			ld_intr_mgmt_t mgm;
1348 
1349 			mgm.value = 0;
1350 			mgm.bits.arm = 1;
1351 			mgm.bits.timer = ldgp->ldg_timer;
1352 			HXGE_REG_WR32(handle,
1353 			    LD_INTR_MGMT + LDSV_OFFSET(ldgp->ldg), mgm.value);
1354 		}
1355 		ringp->poll_flag = 0;
1356 	}
1357 	MUTEX_EXIT(&ringp->lock);
1358 	return (0);
1359 }
1360 
1361 /*
1362  * Poll 'bytes_to_pickup' bytes of message from the rx ring.
1363  */
1364 mblk_t *
1365 hxge_rx_poll(void *arg, int bytes_to_pickup)
1366 {
1367 	p_hxge_ring_handle_t	rhp = (p_hxge_ring_handle_t)arg;
1368 	p_rx_rcr_ring_t		ring;
1369 	p_hxge_t		hxgep;
1370 	hpi_handle_t		handle;
1371 	rdc_stat_t		cs;
1372 	mblk_t			*mblk;
1373 	p_hxge_ldv_t		ldvp;
1374 
1375 	hxgep = rhp->hxgep;
1376 
1377 	/*
1378 	 * Get the control and status for this channel.
1379 	 */
1380 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
1381 	ring = hxgep->rx_rcr_rings->rcr_rings[rhp->index];
1382 
1383 	MUTEX_ENTER(&ring->lock);
1384 	ASSERT(ring->poll_flag == 1);
1385 	ASSERT(rhp->started);
1386 
1387 	/*
1388 	 * Make sure the ring is started and polling is
1389 	 * started before processing packets.
1390 	 */
1391 	if ((!rhp->started) || (ring->poll_flag == 0)) {
1392 		MUTEX_EXIT(&ring->lock);
1393 		return ((mblk_t *)NULL);
1394 	}
1395 
1396 	RXDMA_REG_READ64(handle, RDC_STAT, rhp->index, &cs.value);
1397 	cs.bits.ptrread = 0;
1398 	cs.bits.pktread = 0;
1399 	RXDMA_REG_WRITE64(handle, RDC_STAT, rhp->index, cs.value);
1400 
1401 	mblk = hxge_rx_pkts(hxgep, ring->ldvp->vdma_index,
1402 	    ring->ldvp, ring, cs, bytes_to_pickup);
1403 	ldvp = ring->ldvp;
1404 
1405 	/*
1406 	 * Process Error Events.
1407 	 */
1408 	if (ldvp && (cs.value & RDC_STAT_ERROR)) {
1409 		(void) hxge_rx_err_evnts(hxgep, ldvp->vdma_index, ldvp, cs);
1410 	}
1411 
1412 	MUTEX_EXIT(&ring->lock);
1413 	return (mblk);
1414 }
1415 
1416 /*ARGSUSED*/
1417 mblk_t *
1418 hxge_rx_pkts(p_hxge_t hxgep, uint_t vindex, p_hxge_ldv_t ldvp,
1419     p_rx_rcr_ring_t rcrp, rdc_stat_t cs, int bytes_to_read)
1420 {
1421 	hpi_handle_t		handle;
1422 	uint8_t			channel;
1423 	uint32_t		comp_rd_index;
1424 	p_rcr_entry_t		rcr_desc_rd_head_p;
1425 	p_rcr_entry_t		rcr_desc_rd_head_pp;
1426 	p_mblk_t		nmp, mp_cont, head_mp, *tail_mp;
1427 	uint16_t		qlen, nrcr_read, npkt_read;
1428 	uint32_t		qlen_hw, qlen_sw, num_rcrs;
1429 	uint32_t		invalid_rcr_entry;
1430 	boolean_t		multi;
1431 	rdc_rcr_cfg_b_t		rcr_cfg_b;
1432 	uint64_t		rcr_head_index, rcr_tail_index;
1433 	uint64_t		rcr_tail;
1434 	rdc_rcr_tail_t		rcr_tail_reg;
1435 	p_hxge_rx_ring_stats_t	rdc_stats;
1436 	int			totallen = 0;
1437 
1438 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL, "==> hxge_rx_pkts:vindex %d "
1439 	    "channel %d", vindex, ldvp->channel));
1440 
1441 	if (!(hxgep->drv_state & STATE_HW_INITIALIZED)) {
1442 		return (NULL);
1443 	}
1444 
1445 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
1446 	channel = rcrp->rdc;
1447 	if (channel != ldvp->channel) {
1448 		HXGE_DEBUG_MSG((hxgep, RX_INT_CTL, "==> hxge_rx_pkts:index %d "
1449 		    "channel %d, and rcr channel %d not matched.",
1450 		    vindex, ldvp->channel, channel));
1451 		return (NULL);
1452 	}
1453 
1454 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1455 	    "==> hxge_rx_pkts: START: rcr channel %d "
1456 	    "head_p $%p head_pp $%p  index %d ",
1457 	    channel, rcrp->rcr_desc_rd_head_p,
1458 	    rcrp->rcr_desc_rd_head_pp, rcrp->comp_rd_index));
1459 
1460 	(void) hpi_rxdma_rdc_rcr_qlen_get(handle, channel, &qlen);
1461 	RXDMA_REG_READ64(handle, RDC_RCR_TAIL, channel, &rcr_tail_reg.value);
1462 	rcr_tail = rcr_tail_reg.bits.tail;
1463 
1464 	if (!qlen) {
1465 		HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1466 		    "<== hxge_rx_pkts:rcr channel %d qlen %d (no pkts)",
1467 		    channel, qlen));
1468 		return (NULL);
1469 	}
1470 
1471 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rx_pkts:rcr channel %d "
1472 	    "qlen %d", channel, qlen));
1473 
1474 	comp_rd_index = rcrp->comp_rd_index;
1475 
1476 	rcr_desc_rd_head_p = rcrp->rcr_desc_rd_head_p;
1477 	rcr_desc_rd_head_pp = rcrp->rcr_desc_rd_head_pp;
1478 	nrcr_read = npkt_read = 0;
1479 
1480 	if (hxgep->rdc_first_intr[channel])
1481 		qlen_hw = qlen;
1482 	else
1483 		qlen_hw = qlen - 1;
1484 
1485 	head_mp = NULL;
1486 	tail_mp = &head_mp;
1487 	nmp = mp_cont = NULL;
1488 	multi = B_FALSE;
1489 
1490 	rcr_head_index = rcrp->rcr_desc_rd_head_p - rcrp->rcr_desc_first_p;
1491 	rcr_tail_index = rcr_tail - rcrp->rcr_tail_begin;
1492 
1493 	if (rcr_tail_index >= rcr_head_index) {
1494 		num_rcrs = rcr_tail_index - rcr_head_index;
1495 	} else {
1496 		/* rcr_tail has wrapped around */
1497 		num_rcrs = (rcrp->comp_size - rcr_head_index) + rcr_tail_index;
1498 	}
1499 
1500 	qlen_sw = hxge_scan_for_last_eop(rcrp, rcr_desc_rd_head_p, num_rcrs);
1501 	if (!qlen_sw)
1502 		return (NULL);
1503 
1504 	if (qlen_hw > qlen_sw) {
1505 		HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1506 		    "Channel %d, rcr_qlen from reg %d and from rcr_tail %d\n",
1507 		    channel, qlen_hw, qlen_sw));
1508 		qlen_hw = qlen_sw;
1509 	}
1510 
1511 	while (qlen_hw) {
1512 #ifdef HXGE_DEBUG
1513 		hxge_dump_rcr_entry(hxgep, rcr_desc_rd_head_p);
1514 #endif
1515 		/*
1516 		 * Process one completion ring entry.
1517 		 */
1518 		invalid_rcr_entry = 0;
1519 		hxge_receive_packet(hxgep,
1520 		    rcrp, rcr_desc_rd_head_p, &multi, &nmp, &mp_cont,
1521 		    &invalid_rcr_entry);
1522 		if (invalid_rcr_entry != 0) {
1523 			rdc_stats = rcrp->rdc_stats;
1524 			rdc_stats->rcr_invalids++;
1525 			HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1526 			    "Channel %d could only read 0x%x packets, "
1527 			    "but 0x%x pending\n", channel, npkt_read, qlen_hw));
1528 			break;
1529 		}
1530 
1531 		/*
1532 		 * message chaining modes (nemo msg chaining)
1533 		 */
1534 		if (nmp) {
1535 			nmp->b_next = NULL;
1536 			if (!multi && !mp_cont) { /* frame fits a partition */
1537 				*tail_mp = nmp;
1538 				tail_mp = &nmp->b_next;
1539 				nmp = NULL;
1540 			} else if (multi && !mp_cont) { /* first segment */
1541 				*tail_mp = nmp;
1542 				tail_mp = &nmp->b_cont;
1543 			} else if (multi && mp_cont) {	/* mid of multi segs */
1544 				*tail_mp = mp_cont;
1545 				tail_mp = &mp_cont->b_cont;
1546 			} else if (!multi && mp_cont) { /* last segment */
1547 				*tail_mp = mp_cont;
1548 				tail_mp = &nmp->b_next;
1549 				totallen += MBLKL(mp_cont);
1550 				nmp = NULL;
1551 			}
1552 		}
1553 
1554 		HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1555 		    "==> hxge_rx_pkts: loop: rcr channel %d "
1556 		    "before updating: multi %d "
1557 		    "nrcr_read %d "
1558 		    "npk read %d "
1559 		    "head_pp $%p  index %d ",
1560 		    channel, multi,
1561 		    nrcr_read, npkt_read, rcr_desc_rd_head_pp, comp_rd_index));
1562 
1563 		if (!multi) {
1564 			qlen_hw--;
1565 			npkt_read++;
1566 		}
1567 
1568 		/*
1569 		 * Update the next read entry.
1570 		 */
1571 		comp_rd_index = NEXT_ENTRY(comp_rd_index,
1572 		    rcrp->comp_wrap_mask);
1573 
1574 		rcr_desc_rd_head_p = NEXT_ENTRY_PTR(rcr_desc_rd_head_p,
1575 		    rcrp->rcr_desc_first_p, rcrp->rcr_desc_last_p);
1576 
1577 		nrcr_read++;
1578 
1579 		HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1580 		    "<== hxge_rx_pkts: (SAM, process one packet) "
1581 		    "nrcr_read %d", nrcr_read));
1582 		HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1583 		    "==> hxge_rx_pkts: loop: rcr channel %d "
1584 		    "multi %d nrcr_read %d npk read %d head_pp $%p  index %d ",
1585 		    channel, multi, nrcr_read, npkt_read, rcr_desc_rd_head_pp,
1586 		    comp_rd_index));
1587 
1588 		if ((bytes_to_read != -1) &&
1589 		    (totallen >= bytes_to_read)) {
1590 			break;
1591 		}
1592 	}
1593 
1594 	rcrp->rcr_desc_rd_head_pp = rcr_desc_rd_head_pp;
1595 	rcrp->comp_rd_index = comp_rd_index;
1596 	rcrp->rcr_desc_rd_head_p = rcr_desc_rd_head_p;
1597 
1598 	if ((hxgep->intr_timeout != rcrp->intr_timeout) ||
1599 	    (hxgep->intr_threshold != rcrp->intr_threshold)) {
1600 		rcrp->intr_timeout = hxgep->intr_timeout;
1601 		rcrp->intr_threshold = hxgep->intr_threshold;
1602 		rcr_cfg_b.value = 0x0ULL;
1603 		if (rcrp->intr_timeout)
1604 			rcr_cfg_b.bits.entout = 1;
1605 		rcr_cfg_b.bits.timeout = rcrp->intr_timeout;
1606 		rcr_cfg_b.bits.pthres = rcrp->intr_threshold;
1607 		RXDMA_REG_WRITE64(handle, RDC_RCR_CFG_B,
1608 		    channel, rcr_cfg_b.value);
1609 	}
1610 
1611 	if (hxgep->rdc_first_intr[channel] && (npkt_read > 0)) {
1612 		hxgep->rdc_first_intr[channel] = B_FALSE;
1613 		cs.bits.pktread = npkt_read - 1;
1614 	} else
1615 		cs.bits.pktread = npkt_read;
1616 	cs.bits.ptrread = nrcr_read;
1617 	cs.value &= 0xffffffffULL;
1618 	RXDMA_REG_WRITE64(handle, RDC_STAT, channel, cs.value);
1619 
1620 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1621 	    "==> hxge_rx_pkts: EXIT: rcr channel %d "
1622 	    "head_pp $%p  index %016llx ",
1623 	    channel, rcrp->rcr_desc_rd_head_pp, rcrp->comp_rd_index));
1624 
1625 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL, "<== hxge_rx_pkts"));
1626 
1627 	return (head_mp);
1628 }
1629 
1630 #define	RCR_ENTRY_PATTERN	0x5a5a6b6b7c7c8d8dULL
1631 #define	NO_PORT_BIT		0x20
1632 #define	L4_CS_EQ_BIT		0x40
1633 
1634 static uint32_t hxge_scan_for_last_eop(p_rx_rcr_ring_t rcrp,
1635     p_rcr_entry_t rcr_desc_rd_head_p, uint32_t num_rcrs)
1636 {
1637 	uint64_t	rcr_entry;
1638 	uint32_t	rcrs = 0;
1639 	uint32_t	pkts = 0;
1640 
1641 	while (rcrs++ < num_rcrs) {
1642 		rcr_entry = *((uint64_t *)rcr_desc_rd_head_p);
1643 
1644 		if ((rcr_entry == 0x0) || (rcr_entry == RCR_ENTRY_PATTERN))
1645 			break;
1646 
1647 		if (!(rcr_entry & RCR_MULTI_MASK))
1648 			pkts++;
1649 
1650 		rcr_desc_rd_head_p = NEXT_ENTRY_PTR(rcr_desc_rd_head_p,
1651 		    rcrp->rcr_desc_first_p, rcrp->rcr_desc_last_p);
1652 	}
1653 
1654 	return (pkts);
1655 }
1656 
1657 /*ARGSUSED*/
1658 void
1659 hxge_receive_packet(p_hxge_t hxgep, p_rx_rcr_ring_t rcr_p,
1660     p_rcr_entry_t rcr_desc_rd_head_p, boolean_t *multi_p, mblk_t **mp,
1661     mblk_t **mp_cont, uint32_t *invalid_rcr_entry)
1662 {
1663 	p_mblk_t nmp = NULL;
1664 	uint64_t multi;
1665 	uint8_t channel;
1666 	boolean_t first_entry = B_TRUE;
1667 	boolean_t is_tcp_udp = B_FALSE;
1668 	boolean_t buffer_free = B_FALSE;
1669 	boolean_t error_send_up = B_FALSE;
1670 	uint8_t error_type;
1671 	uint16_t l2_len;
1672 	uint16_t skip_len;
1673 	uint8_t pktbufsz_type;
1674 	uint64_t rcr_entry;
1675 	uint64_t *pkt_buf_addr_pp;
1676 	uint64_t *pkt_buf_addr_p;
1677 	uint32_t buf_offset;
1678 	uint32_t bsize;
1679 	uint32_t msg_index;
1680 	p_rx_rbr_ring_t rx_rbr_p;
1681 	p_rx_msg_t *rx_msg_ring_p;
1682 	p_rx_msg_t rx_msg_p;
1683 	uint16_t sw_offset_bytes = 0, hdr_size = 0;
1684 	hxge_status_t status = HXGE_OK;
1685 	boolean_t is_valid = B_FALSE;
1686 	p_hxge_rx_ring_stats_t rdc_stats;
1687 	uint32_t bytes_read;
1688 	uint8_t header0 = 0;
1689 	uint8_t header1 = 0;
1690 	uint64_t pkt_type;
1691 	uint8_t no_port_bit = 0;
1692 	uint8_t l4_cs_eq_bit = 0;
1693 
1694 	channel = rcr_p->rdc;
1695 
1696 	HXGE_DEBUG_MSG((hxgep, RX2_CTL, "==> hxge_receive_packet"));
1697 
1698 	first_entry = (*mp == NULL) ? B_TRUE : B_FALSE;
1699 	rcr_entry = *((uint64_t *)rcr_desc_rd_head_p);
1700 
1701 	/* Verify the content of the rcr_entry for a hardware bug workaround */
1702 	if ((rcr_entry == 0x0) || (rcr_entry == RCR_ENTRY_PATTERN)) {
1703 		*invalid_rcr_entry = 1;
1704 		HXGE_DEBUG_MSG((hxgep, RX2_CTL, "hxge_receive_packet "
1705 		    "Channel %d invalid RCR entry 0x%llx found, returning\n",
1706 		    channel, (long long) rcr_entry));
1707 		return;
1708 	}
1709 	*((uint64_t *)rcr_desc_rd_head_p) = RCR_ENTRY_PATTERN;
1710 
1711 	multi = (rcr_entry & RCR_MULTI_MASK);
1712 	pkt_type = (rcr_entry & RCR_PKT_TYPE_MASK);
1713 
1714 	error_type = ((rcr_entry & RCR_ERROR_MASK) >> RCR_ERROR_SHIFT);
1715 	l2_len = ((rcr_entry & RCR_L2_LEN_MASK) >> RCR_L2_LEN_SHIFT);
1716 
1717 	/*
1718 	 * Hardware does not strip the CRC due bug ID 11451 where
1719 	 * the hardware mis handles minimum size packets.
1720 	 */
1721 	l2_len -= ETHERFCSL;
1722 
1723 	pktbufsz_type = ((rcr_entry & RCR_PKTBUFSZ_MASK) >>
1724 	    RCR_PKTBUFSZ_SHIFT);
1725 #if defined(__i386)
1726 	pkt_buf_addr_pp = (uint64_t *)(uint32_t)((rcr_entry &
1727 	    RCR_PKT_BUF_ADDR_MASK) << RCR_PKT_BUF_ADDR_SHIFT);
1728 #else
1729 	pkt_buf_addr_pp = (uint64_t *)((rcr_entry & RCR_PKT_BUF_ADDR_MASK) <<
1730 	    RCR_PKT_BUF_ADDR_SHIFT);
1731 #endif
1732 
1733 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1734 	    "==> hxge_receive_packet: entryp $%p entry 0x%0llx "
1735 	    "pkt_buf_addr_pp $%p l2_len %d multi %d "
1736 	    "error_type 0x%x pktbufsz_type %d ",
1737 	    rcr_desc_rd_head_p, rcr_entry, pkt_buf_addr_pp, l2_len,
1738 	    multi, error_type, pktbufsz_type));
1739 
1740 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1741 	    "==> hxge_receive_packet: entryp $%p entry 0x%0llx "
1742 	    "pkt_buf_addr_pp $%p l2_len %d multi %d "
1743 	    "error_type 0x%x ", rcr_desc_rd_head_p,
1744 	    rcr_entry, pkt_buf_addr_pp, l2_len, multi, error_type));
1745 
1746 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1747 	    "==> (rbr) hxge_receive_packet: entry 0x%0llx "
1748 	    "full pkt_buf_addr_pp $%p l2_len %d",
1749 	    rcr_entry, pkt_buf_addr_pp, l2_len));
1750 
1751 	/* get the stats ptr */
1752 	rdc_stats = rcr_p->rdc_stats;
1753 
1754 	if (!l2_len) {
1755 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
1756 		    "<== hxge_receive_packet: failed: l2 length is 0."));
1757 		return;
1758 	}
1759 
1760 	/* shift 6 bits to get the full io address */
1761 #if defined(__i386)
1762 	pkt_buf_addr_pp = (uint64_t *)((uint32_t)pkt_buf_addr_pp <<
1763 	    RCR_PKT_BUF_ADDR_SHIFT_FULL);
1764 #else
1765 	pkt_buf_addr_pp = (uint64_t *)((uint64_t)pkt_buf_addr_pp <<
1766 	    RCR_PKT_BUF_ADDR_SHIFT_FULL);
1767 #endif
1768 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1769 	    "==> (rbr) hxge_receive_packet: entry 0x%0llx "
1770 	    "full pkt_buf_addr_pp $%p l2_len %d",
1771 	    rcr_entry, pkt_buf_addr_pp, l2_len));
1772 
1773 	rx_rbr_p = rcr_p->rx_rbr_p;
1774 	rx_msg_ring_p = rx_rbr_p->rx_msg_ring;
1775 
1776 	if (first_entry) {
1777 		hdr_size = (rcr_p->full_hdr_flag ? RXDMA_HDR_SIZE_FULL :
1778 		    RXDMA_HDR_SIZE_DEFAULT);
1779 
1780 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
1781 		    "==> hxge_receive_packet: first entry 0x%016llx "
1782 		    "pkt_buf_addr_pp $%p l2_len %d hdr %d",
1783 		    rcr_entry, pkt_buf_addr_pp, l2_len, hdr_size));
1784 	}
1785 
1786 	MUTEX_ENTER(&rx_rbr_p->lock);
1787 
1788 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
1789 	    "==> (rbr 1) hxge_receive_packet: entry 0x%0llx "
1790 	    "full pkt_buf_addr_pp $%p l2_len %d",
1791 	    rcr_entry, pkt_buf_addr_pp, l2_len));
1792 
1793 	/*
1794 	 * Packet buffer address in the completion entry points to the starting
1795 	 * buffer address (offset 0). Use the starting buffer address to locate
1796 	 * the corresponding kernel address.
1797 	 */
1798 	status = hxge_rxbuf_pp_to_vp(hxgep, rx_rbr_p,
1799 	    pktbufsz_type, pkt_buf_addr_pp, &pkt_buf_addr_p,
1800 	    &buf_offset, &msg_index);
1801 
1802 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
1803 	    "==> (rbr 2) hxge_receive_packet: entry 0x%0llx "
1804 	    "full pkt_buf_addr_pp $%p l2_len %d",
1805 	    rcr_entry, pkt_buf_addr_pp, l2_len));
1806 
1807 	if (status != HXGE_OK) {
1808 		MUTEX_EXIT(&rx_rbr_p->lock);
1809 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
1810 		    "<== hxge_receive_packet: found vaddr failed %d", status));
1811 		return;
1812 	}
1813 
1814 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1815 	    "==> (rbr 3) hxge_receive_packet: entry 0x%0llx "
1816 	    "full pkt_buf_addr_pp $%p l2_len %d",
1817 	    rcr_entry, pkt_buf_addr_pp, l2_len));
1818 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1819 	    "==> (rbr 4 msgindex %d) hxge_receive_packet: entry 0x%0llx "
1820 	    "full pkt_buf_addr_pp $%p l2_len %d",
1821 	    msg_index, rcr_entry, pkt_buf_addr_pp, l2_len));
1822 
1823 	if (msg_index >= rx_rbr_p->tnblocks) {
1824 		MUTEX_EXIT(&rx_rbr_p->lock);
1825 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1826 		    "==> hxge_receive_packet: FATAL msg_index (%d) "
1827 		    "should be smaller than tnblocks (%d)\n",
1828 		    msg_index, rx_rbr_p->tnblocks));
1829 		return;
1830 	}
1831 
1832 	rx_msg_p = rx_msg_ring_p[msg_index];
1833 
1834 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1835 	    "==> (rbr 4 msgindex %d) hxge_receive_packet: entry 0x%0llx "
1836 	    "full pkt_buf_addr_pp $%p l2_len %d",
1837 	    msg_index, rcr_entry, pkt_buf_addr_pp, l2_len));
1838 
1839 	switch (pktbufsz_type) {
1840 	case RCR_PKTBUFSZ_0:
1841 		bsize = rx_rbr_p->pkt_buf_size0_bytes;
1842 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1843 		    "==> hxge_receive_packet: 0 buf %d", bsize));
1844 		break;
1845 	case RCR_PKTBUFSZ_1:
1846 		bsize = rx_rbr_p->pkt_buf_size1_bytes;
1847 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1848 		    "==> hxge_receive_packet: 1 buf %d", bsize));
1849 		break;
1850 	case RCR_PKTBUFSZ_2:
1851 		bsize = rx_rbr_p->pkt_buf_size2_bytes;
1852 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
1853 		    "==> hxge_receive_packet: 2 buf %d", bsize));
1854 		break;
1855 	case RCR_SINGLE_BLOCK:
1856 		bsize = rx_msg_p->block_size;
1857 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1858 		    "==> hxge_receive_packet: single %d", bsize));
1859 
1860 		break;
1861 	default:
1862 		MUTEX_EXIT(&rx_rbr_p->lock);
1863 		return;
1864 	}
1865 
1866 	DMA_COMMON_SYNC_OFFSET(rx_msg_p->buf_dma,
1867 	    (buf_offset + sw_offset_bytes), (hdr_size + l2_len),
1868 	    DDI_DMA_SYNC_FORCPU);
1869 
1870 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1871 	    "==> hxge_receive_packet: after first dump:usage count"));
1872 
1873 	if (rx_msg_p->cur_usage_cnt == 0) {
1874 		atomic_inc_32(&rx_rbr_p->rbr_used);
1875 		if (rx_rbr_p->rbr_use_bcopy) {
1876 			atomic_inc_32(&rx_rbr_p->rbr_consumed);
1877 			if (rx_rbr_p->rbr_consumed <
1878 			    rx_rbr_p->rbr_threshold_hi) {
1879 				if (rx_rbr_p->rbr_threshold_lo == 0 ||
1880 				    ((rx_rbr_p->rbr_consumed >=
1881 				    rx_rbr_p->rbr_threshold_lo) &&
1882 				    (rx_rbr_p->rbr_bufsize_type >=
1883 				    pktbufsz_type))) {
1884 					rx_msg_p->rx_use_bcopy = B_TRUE;
1885 				}
1886 			} else {
1887 				rx_msg_p->rx_use_bcopy = B_TRUE;
1888 			}
1889 		}
1890 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1891 		    "==> hxge_receive_packet: buf %d (new block) ", bsize));
1892 
1893 		rx_msg_p->pkt_buf_size_code = pktbufsz_type;
1894 		rx_msg_p->pkt_buf_size = bsize;
1895 		rx_msg_p->cur_usage_cnt = 1;
1896 		if (pktbufsz_type == RCR_SINGLE_BLOCK) {
1897 			HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1898 			    "==> hxge_receive_packet: buf %d (single block) ",
1899 			    bsize));
1900 			/*
1901 			 * Buffer can be reused once the free function is
1902 			 * called.
1903 			 */
1904 			rx_msg_p->max_usage_cnt = 1;
1905 			buffer_free = B_TRUE;
1906 		} else {
1907 			rx_msg_p->max_usage_cnt = rx_msg_p->block_size / bsize;
1908 			if (rx_msg_p->max_usage_cnt == 1) {
1909 				buffer_free = B_TRUE;
1910 			}
1911 		}
1912 	} else {
1913 		rx_msg_p->cur_usage_cnt++;
1914 		if (rx_msg_p->cur_usage_cnt == rx_msg_p->max_usage_cnt) {
1915 			buffer_free = B_TRUE;
1916 		}
1917 	}
1918 
1919 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
1920 	    "msgbuf index = %d l2len %d bytes usage %d max_usage %d ",
1921 	    msg_index, l2_len,
1922 	    rx_msg_p->cur_usage_cnt, rx_msg_p->max_usage_cnt));
1923 
1924 	if (error_type) {
1925 		rdc_stats->ierrors++;
1926 		/* Update error stats */
1927 		rdc_stats->errlog.compl_err_type = error_type;
1928 		HXGE_FM_REPORT_ERROR(hxgep, NULL, HXGE_FM_EREPORT_RDMC_RCR_ERR);
1929 
1930 		if (error_type & RCR_CTRL_FIFO_DED) {
1931 			rdc_stats->ctrl_fifo_ecc_err++;
1932 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
1933 			    " hxge_receive_packet: "
1934 			    " channel %d RCR ctrl_fifo_ded error", channel));
1935 		} else if (error_type & RCR_DATA_FIFO_DED) {
1936 			rdc_stats->data_fifo_ecc_err++;
1937 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
1938 			    " hxge_receive_packet: channel %d"
1939 			    " RCR data_fifo_ded error", channel));
1940 		}
1941 
1942 		/*
1943 		 * Update and repost buffer block if max usage count is
1944 		 * reached.
1945 		 */
1946 		if (error_send_up == B_FALSE) {
1947 			atomic_inc_32(&rx_msg_p->ref_cnt);
1948 			if (buffer_free == B_TRUE) {
1949 				rx_msg_p->free = B_TRUE;
1950 			}
1951 
1952 			MUTEX_EXIT(&rx_rbr_p->lock);
1953 			hxge_freeb(rx_msg_p);
1954 			return;
1955 		}
1956 	}
1957 
1958 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1959 	    "==> hxge_receive_packet: DMA sync second "));
1960 
1961 	bytes_read = rcr_p->rcvd_pkt_bytes;
1962 	skip_len = sw_offset_bytes + hdr_size;
1963 
1964 	if (first_entry) {
1965 		header0 = rx_msg_p->buffer[buf_offset];
1966 		no_port_bit = header0 & NO_PORT_BIT;
1967 		header1 = rx_msg_p->buffer[buf_offset + 1];
1968 		l4_cs_eq_bit = header1 & L4_CS_EQ_BIT;
1969 	}
1970 
1971 	if (!rx_msg_p->rx_use_bcopy) {
1972 		/*
1973 		 * For loaned up buffers, the driver reference count
1974 		 * will be incremented first and then the free state.
1975 		 */
1976 		if ((nmp = hxge_dupb(rx_msg_p, buf_offset, bsize)) != NULL) {
1977 			if (first_entry) {
1978 				nmp->b_rptr = &nmp->b_rptr[skip_len];
1979 				if (l2_len < bsize - skip_len) {
1980 					nmp->b_wptr = &nmp->b_rptr[l2_len];
1981 				} else {
1982 					nmp->b_wptr = &nmp->b_rptr[bsize
1983 					    - skip_len];
1984 				}
1985 			} else {
1986 				if (l2_len - bytes_read < bsize) {
1987 					nmp->b_wptr =
1988 					    &nmp->b_rptr[l2_len - bytes_read];
1989 				} else {
1990 					nmp->b_wptr = &nmp->b_rptr[bsize];
1991 				}
1992 			}
1993 		}
1994 	} else {
1995 		if (first_entry) {
1996 			nmp = hxge_dupb_bcopy(rx_msg_p, buf_offset + skip_len,
1997 			    l2_len < bsize - skip_len ?
1998 			    l2_len : bsize - skip_len);
1999 		} else {
2000 			nmp = hxge_dupb_bcopy(rx_msg_p, buf_offset,
2001 			    l2_len - bytes_read < bsize ?
2002 			    l2_len - bytes_read : bsize);
2003 		}
2004 	}
2005 
2006 	if (nmp != NULL) {
2007 		if (first_entry)
2008 			bytes_read  = nmp->b_wptr - nmp->b_rptr;
2009 		else
2010 			bytes_read += nmp->b_wptr - nmp->b_rptr;
2011 
2012 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
2013 		    "==> hxge_receive_packet after dupb: "
2014 		    "rbr consumed %d "
2015 		    "pktbufsz_type %d "
2016 		    "nmp $%p rptr $%p wptr $%p "
2017 		    "buf_offset %d bzise %d l2_len %d skip_len %d",
2018 		    rx_rbr_p->rbr_consumed,
2019 		    pktbufsz_type,
2020 		    nmp, nmp->b_rptr, nmp->b_wptr,
2021 		    buf_offset, bsize, l2_len, skip_len));
2022 	} else {
2023 		cmn_err(CE_WARN, "!hxge_receive_packet: update stats (error)");
2024 
2025 		atomic_inc_32(&rx_msg_p->ref_cnt);
2026 		if (buffer_free == B_TRUE) {
2027 			rx_msg_p->free = B_TRUE;
2028 		}
2029 
2030 		MUTEX_EXIT(&rx_rbr_p->lock);
2031 		hxge_freeb(rx_msg_p);
2032 		return;
2033 	}
2034 
2035 	if (buffer_free == B_TRUE) {
2036 		rx_msg_p->free = B_TRUE;
2037 	}
2038 
2039 	/*
2040 	 * ERROR, FRAG and PKT_TYPE are only reported in the first entry. If a
2041 	 * packet is not fragmented and no error bit is set, then L4 checksum
2042 	 * is OK.
2043 	 */
2044 	is_valid = (nmp != NULL);
2045 	if (first_entry) {
2046 		rdc_stats->ipackets++; /* count only 1st seg for jumbo */
2047 		if (l2_len > (STD_FRAME_SIZE - ETHERFCSL))
2048 			rdc_stats->jumbo_pkts++;
2049 		rdc_stats->ibytes += skip_len + l2_len < bsize ?
2050 		    l2_len : bsize;
2051 	} else {
2052 		/*
2053 		 * Add the current portion of the packet to the kstats.
2054 		 * The current portion of the packet is calculated by using
2055 		 * length of the packet and the previously received portion.
2056 		 */
2057 		rdc_stats->ibytes += l2_len - rcr_p->rcvd_pkt_bytes < bsize ?
2058 		    l2_len - rcr_p->rcvd_pkt_bytes : bsize;
2059 	}
2060 
2061 	rcr_p->rcvd_pkt_bytes = bytes_read;
2062 
2063 	if (rx_msg_p->free && rx_msg_p->rx_use_bcopy) {
2064 		atomic_inc_32(&rx_msg_p->ref_cnt);
2065 		MUTEX_EXIT(&rx_rbr_p->lock);
2066 		hxge_freeb(rx_msg_p);
2067 	} else
2068 		MUTEX_EXIT(&rx_rbr_p->lock);
2069 
2070 	if (is_valid) {
2071 		nmp->b_cont = NULL;
2072 		if (first_entry) {
2073 			*mp = nmp;
2074 			*mp_cont = NULL;
2075 		} else {
2076 			*mp_cont = nmp;
2077 		}
2078 	}
2079 
2080 	/*
2081 	 * Update stats and hardware checksuming.
2082 	 */
2083 	if (is_valid && !multi) {
2084 		is_tcp_udp = ((pkt_type == RCR_PKT_IS_TCP ||
2085 		    pkt_type == RCR_PKT_IS_UDP) ? B_TRUE : B_FALSE);
2086 
2087 		if (!no_port_bit && l4_cs_eq_bit && is_tcp_udp && !error_type) {
2088 			(void) hcksum_assoc(nmp, NULL, NULL, 0, 0, 0, 0,
2089 			    HCK_FULLCKSUM_OK | HCK_FULLCKSUM, 0);
2090 
2091 			HXGE_DEBUG_MSG((hxgep, RX_CTL,
2092 			    "==> hxge_receive_packet: Full tcp/udp cksum "
2093 			    "is_valid 0x%x multi %d error %d",
2094 			    is_valid, multi, error_type));
2095 		}
2096 	}
2097 
2098 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
2099 	    "==> hxge_receive_packet: *mp 0x%016llx", *mp));
2100 
2101 	*multi_p = (multi == RCR_MULTI_MASK);
2102 
2103 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_receive_packet: "
2104 	    "multi %d nmp 0x%016llx *mp 0x%016llx *mp_cont 0x%016llx",
2105 	    *multi_p, nmp, *mp, *mp_cont));
2106 }
2107 
2108 static void
2109 hxge_rx_rbr_empty_recover(p_hxge_t hxgep, uint8_t channel)
2110 {
2111 	hpi_handle_t	handle;
2112 	p_rx_rcr_ring_t	rcrp;
2113 	p_rx_rbr_ring_t	rbrp;
2114 
2115 	rcrp = hxgep->rx_rcr_rings->rcr_rings[channel];
2116 	rbrp = rcrp->rx_rbr_p;
2117 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
2118 
2119 	/*
2120 	 * Wait for the channel to be quiet
2121 	 */
2122 	(void) hpi_rxdma_cfg_rdc_wait_for_qst(handle, channel);
2123 
2124 	/*
2125 	 * Post page will accumulate some buffers before re-enabling
2126 	 * the DMA channel.
2127 	 */
2128 
2129 	MUTEX_ENTER(&rbrp->post_lock);
2130 	if ((rbrp->rbb_max - rbrp->rbr_used) >= HXGE_RBR_EMPTY_THRESHOLD) {
2131 		hxge_rbr_empty_restore(hxgep, rbrp);
2132 	} else {
2133 		rbrp->rbr_is_empty = B_TRUE;
2134 	}
2135 	MUTEX_EXIT(&rbrp->post_lock);
2136 }
2137 
2138 
2139 /*ARGSUSED*/
2140 static hxge_status_t
2141 hxge_rx_err_evnts(p_hxge_t hxgep, uint_t index, p_hxge_ldv_t ldvp,
2142     rdc_stat_t cs)
2143 {
2144 	p_hxge_rx_ring_stats_t	rdc_stats;
2145 	hpi_handle_t		handle;
2146 	boolean_t		rxchan_fatal = B_FALSE;
2147 	uint8_t			channel;
2148 	hxge_status_t		status = HXGE_OK;
2149 	uint64_t		cs_val;
2150 
2151 	HXGE_DEBUG_MSG((hxgep, INT_CTL, "==> hxge_rx_err_evnts"));
2152 
2153 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
2154 	channel = ldvp->channel;
2155 
2156 	/* Clear the interrupts */
2157 	cs.bits.pktread = 0;
2158 	cs.bits.ptrread = 0;
2159 	cs_val = cs.value & RDC_STAT_WR1C;
2160 	RXDMA_REG_WRITE64(handle, RDC_STAT, channel, cs_val);
2161 
2162 	rdc_stats = &hxgep->statsp->rdc_stats[ldvp->vdma_index];
2163 
2164 	if (cs.bits.rbr_cpl_to) {
2165 		rdc_stats->rbr_tmout++;
2166 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2167 		    HXGE_FM_EREPORT_RDMC_RBR_CPL_TO);
2168 		rxchan_fatal = B_TRUE;
2169 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2170 		    "==> hxge_rx_err_evnts(channel %d): "
2171 		    "fatal error: rx_rbr_timeout", channel));
2172 	}
2173 
2174 	if ((cs.bits.rcr_shadow_par_err) || (cs.bits.rbr_prefetch_par_err)) {
2175 		(void) hpi_rxdma_ring_perr_stat_get(handle,
2176 		    &rdc_stats->errlog.pre_par, &rdc_stats->errlog.sha_par);
2177 	}
2178 
2179 	if (cs.bits.rcr_shadow_par_err) {
2180 		rdc_stats->rcr_sha_par++;
2181 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2182 		    HXGE_FM_EREPORT_RDMC_RCR_SHA_PAR);
2183 		rxchan_fatal = B_TRUE;
2184 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2185 		    "==> hxge_rx_err_evnts(channel %d): "
2186 		    "fatal error: rcr_shadow_par_err", channel));
2187 	}
2188 
2189 	if (cs.bits.rbr_prefetch_par_err) {
2190 		rdc_stats->rbr_pre_par++;
2191 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2192 		    HXGE_FM_EREPORT_RDMC_RBR_PRE_PAR);
2193 		rxchan_fatal = B_TRUE;
2194 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2195 		    "==> hxge_rx_err_evnts(channel %d): "
2196 		    "fatal error: rbr_prefetch_par_err", channel));
2197 	}
2198 
2199 	if (cs.bits.rbr_pre_empty) {
2200 		rdc_stats->rbr_pre_empty++;
2201 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2202 		    HXGE_FM_EREPORT_RDMC_RBR_PRE_EMPTY);
2203 		rxchan_fatal = B_TRUE;
2204 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2205 		    "==> hxge_rx_err_evnts(channel %d): "
2206 		    "fatal error: rbr_pre_empty", channel));
2207 	}
2208 
2209 	if (cs.bits.peu_resp_err) {
2210 		rdc_stats->peu_resp_err++;
2211 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2212 		    HXGE_FM_EREPORT_RDMC_PEU_RESP_ERR);
2213 		rxchan_fatal = B_TRUE;
2214 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2215 		    "==> hxge_rx_err_evnts(channel %d): "
2216 		    "fatal error: peu_resp_err", channel));
2217 	}
2218 
2219 	if (cs.bits.rcr_thres) {
2220 		rdc_stats->rcr_thres++;
2221 	}
2222 
2223 	if (cs.bits.rcr_to) {
2224 		rdc_stats->rcr_to++;
2225 	}
2226 
2227 	if (cs.bits.rcr_shadow_full) {
2228 		rdc_stats->rcr_shadow_full++;
2229 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2230 		    HXGE_FM_EREPORT_RDMC_RCR_SHA_FULL);
2231 		rxchan_fatal = B_TRUE;
2232 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2233 		    "==> hxge_rx_err_evnts(channel %d): "
2234 		    "fatal error: rcr_shadow_full", channel));
2235 	}
2236 
2237 	if (cs.bits.rcr_full) {
2238 		rdc_stats->rcrfull++;
2239 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2240 		    HXGE_FM_EREPORT_RDMC_RCRFULL);
2241 		rxchan_fatal = B_TRUE;
2242 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2243 		    "==> hxge_rx_err_evnts(channel %d): "
2244 		    "fatal error: rcrfull error", channel));
2245 	}
2246 
2247 	if (cs.bits.rbr_empty) {
2248 		rdc_stats->rbr_empty++;
2249 		hxge_rx_rbr_empty_recover(hxgep, channel);
2250 	}
2251 
2252 	if (cs.bits.rbr_full) {
2253 		rdc_stats->rbrfull++;
2254 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2255 		    HXGE_FM_EREPORT_RDMC_RBRFULL);
2256 		rxchan_fatal = B_TRUE;
2257 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2258 		    "==> hxge_rx_err_evnts(channel %d): "
2259 		    "fatal error: rbr_full error", channel));
2260 	}
2261 
2262 	if (rxchan_fatal) {
2263 		p_rx_rcr_ring_t	rcrp;
2264 		p_rx_rbr_ring_t rbrp;
2265 
2266 		rcrp = hxgep->rx_rcr_rings->rcr_rings[channel];
2267 		rbrp = rcrp->rx_rbr_p;
2268 
2269 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2270 		    " hxge_rx_err_evnts: fatal error on Channel #%d\n",
2271 		    channel));
2272 		MUTEX_ENTER(&rbrp->post_lock);
2273 		/* This function needs to be inside the post_lock */
2274 		status = hxge_rxdma_fatal_err_recover(hxgep, channel);
2275 		MUTEX_EXIT(&rbrp->post_lock);
2276 		if (status == HXGE_OK) {
2277 			FM_SERVICE_RESTORED(hxgep);
2278 		}
2279 	}
2280 	HXGE_DEBUG_MSG((hxgep, INT_CTL, "<== hxge_rx_err_evnts"));
2281 
2282 	return (status);
2283 }
2284 
2285 static hxge_status_t
2286 hxge_map_rxdma(p_hxge_t hxgep)
2287 {
2288 	int			i, ndmas;
2289 	uint16_t		channel;
2290 	p_rx_rbr_rings_t	rx_rbr_rings;
2291 	p_rx_rbr_ring_t		*rbr_rings;
2292 	p_rx_rcr_rings_t	rx_rcr_rings;
2293 	p_rx_rcr_ring_t		*rcr_rings;
2294 	p_rx_mbox_areas_t	rx_mbox_areas_p;
2295 	p_rx_mbox_t		*rx_mbox_p;
2296 	p_hxge_dma_pool_t	dma_buf_poolp;
2297 	p_hxge_dma_common_t	*dma_buf_p;
2298 	p_hxge_dma_pool_t	dma_rbr_cntl_poolp;
2299 	p_hxge_dma_common_t	*dma_rbr_cntl_p;
2300 	p_hxge_dma_pool_t	dma_rcr_cntl_poolp;
2301 	p_hxge_dma_common_t	*dma_rcr_cntl_p;
2302 	p_hxge_dma_pool_t	dma_mbox_cntl_poolp;
2303 	p_hxge_dma_common_t	*dma_mbox_cntl_p;
2304 	uint32_t		*num_chunks;
2305 	hxge_status_t		status = HXGE_OK;
2306 
2307 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_map_rxdma"));
2308 
2309 	dma_buf_poolp = hxgep->rx_buf_pool_p;
2310 	dma_rbr_cntl_poolp = hxgep->rx_rbr_cntl_pool_p;
2311 	dma_rcr_cntl_poolp = hxgep->rx_rcr_cntl_pool_p;
2312 	dma_mbox_cntl_poolp = hxgep->rx_mbox_cntl_pool_p;
2313 
2314 	if (!dma_buf_poolp->buf_allocated ||
2315 	    !dma_rbr_cntl_poolp->buf_allocated ||
2316 	    !dma_rcr_cntl_poolp->buf_allocated ||
2317 	    !dma_mbox_cntl_poolp->buf_allocated) {
2318 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2319 		    "<== hxge_map_rxdma: buf not allocated"));
2320 		return (HXGE_ERROR);
2321 	}
2322 
2323 	ndmas = dma_buf_poolp->ndmas;
2324 	if (!ndmas) {
2325 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
2326 		    "<== hxge_map_rxdma: no dma allocated"));
2327 		return (HXGE_ERROR);
2328 	}
2329 
2330 	num_chunks = dma_buf_poolp->num_chunks;
2331 	dma_buf_p = dma_buf_poolp->dma_buf_pool_p;
2332 	dma_rbr_cntl_p = dma_rbr_cntl_poolp->dma_buf_pool_p;
2333 	dma_rcr_cntl_p = dma_rcr_cntl_poolp->dma_buf_pool_p;
2334 	dma_mbox_cntl_p = dma_mbox_cntl_poolp->dma_buf_pool_p;
2335 
2336 	rx_rbr_rings = (p_rx_rbr_rings_t)
2337 	    KMEM_ZALLOC(sizeof (rx_rbr_rings_t), KM_SLEEP);
2338 	rbr_rings = (p_rx_rbr_ring_t *)KMEM_ZALLOC(
2339 	    sizeof (p_rx_rbr_ring_t) * ndmas, KM_SLEEP);
2340 
2341 	rx_rcr_rings = (p_rx_rcr_rings_t)
2342 	    KMEM_ZALLOC(sizeof (rx_rcr_rings_t), KM_SLEEP);
2343 	rcr_rings = (p_rx_rcr_ring_t *)KMEM_ZALLOC(
2344 	    sizeof (p_rx_rcr_ring_t) * ndmas, KM_SLEEP);
2345 
2346 	rx_mbox_areas_p = (p_rx_mbox_areas_t)
2347 	    KMEM_ZALLOC(sizeof (rx_mbox_areas_t), KM_SLEEP);
2348 	rx_mbox_p = (p_rx_mbox_t *)KMEM_ZALLOC(
2349 	    sizeof (p_rx_mbox_t) * ndmas, KM_SLEEP);
2350 
2351 	/*
2352 	 * Timeout should be set based on the system clock divider.
2353 	 * The following timeout value of 1 assumes that the
2354 	 * granularity (1000) is 3 microseconds running at 300MHz.
2355 	 */
2356 
2357 	hxgep->intr_threshold = RXDMA_RCR_PTHRES_DEFAULT;
2358 	hxgep->intr_timeout = RXDMA_RCR_TO_DEFAULT;
2359 
2360 	/*
2361 	 * Map descriptors from the buffer polls for each dam channel.
2362 	 */
2363 	for (i = 0; i < ndmas; i++) {
2364 		/*
2365 		 * Set up and prepare buffer blocks, descriptors and mailbox.
2366 		 */
2367 		channel = ((p_hxge_dma_common_t)dma_buf_p[i])->dma_channel;
2368 		status = hxge_map_rxdma_channel(hxgep, channel,
2369 		    (p_hxge_dma_common_t *)&dma_buf_p[i],
2370 		    (p_rx_rbr_ring_t *)&rbr_rings[i],
2371 		    num_chunks[i],
2372 		    (p_hxge_dma_common_t *)&dma_rbr_cntl_p[i],
2373 		    (p_hxge_dma_common_t *)&dma_rcr_cntl_p[i],
2374 		    (p_hxge_dma_common_t *)&dma_mbox_cntl_p[i],
2375 		    (p_rx_rcr_ring_t *)&rcr_rings[i],
2376 		    (p_rx_mbox_t *)&rx_mbox_p[i]);
2377 		if (status != HXGE_OK) {
2378 			goto hxge_map_rxdma_fail1;
2379 		}
2380 		rbr_rings[i]->index = (uint16_t)i;
2381 		rcr_rings[i]->index = (uint16_t)i;
2382 		rcr_rings[i]->rdc_stats = &hxgep->statsp->rdc_stats[i];
2383 	}
2384 
2385 	rx_rbr_rings->ndmas = rx_rcr_rings->ndmas = ndmas;
2386 	rx_rbr_rings->rbr_rings = rbr_rings;
2387 	hxgep->rx_rbr_rings = rx_rbr_rings;
2388 	rx_rcr_rings->rcr_rings = rcr_rings;
2389 	hxgep->rx_rcr_rings = rx_rcr_rings;
2390 
2391 	rx_mbox_areas_p->rxmbox_areas = rx_mbox_p;
2392 	hxgep->rx_mbox_areas_p = rx_mbox_areas_p;
2393 
2394 	goto hxge_map_rxdma_exit;
2395 
2396 hxge_map_rxdma_fail1:
2397 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2398 	    "==> hxge_map_rxdma: unmap rbr,rcr (status 0x%x channel %d i %d)",
2399 	    status, channel, i));
2400 	i--;
2401 	for (; i >= 0; i--) {
2402 		channel = ((p_hxge_dma_common_t)dma_buf_p[i])->dma_channel;
2403 		hxge_unmap_rxdma_channel(hxgep, channel,
2404 		    rbr_rings[i], rcr_rings[i], rx_mbox_p[i]);
2405 	}
2406 
2407 	KMEM_FREE(rbr_rings, sizeof (p_rx_rbr_ring_t) * ndmas);
2408 	KMEM_FREE(rx_rbr_rings, sizeof (rx_rbr_rings_t));
2409 	KMEM_FREE(rcr_rings, sizeof (p_rx_rcr_ring_t) * ndmas);
2410 	KMEM_FREE(rx_rcr_rings, sizeof (rx_rcr_rings_t));
2411 	KMEM_FREE(rx_mbox_p, sizeof (p_rx_mbox_t) * ndmas);
2412 	KMEM_FREE(rx_mbox_areas_p, sizeof (rx_mbox_areas_t));
2413 
2414 hxge_map_rxdma_exit:
2415 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2416 	    "<== hxge_map_rxdma: (status 0x%x channel %d)", status, channel));
2417 
2418 	return (status);
2419 }
2420 
2421 static void
2422 hxge_unmap_rxdma(p_hxge_t hxgep)
2423 {
2424 	int			i, ndmas;
2425 	uint16_t		channel;
2426 	p_rx_rbr_rings_t	rx_rbr_rings;
2427 	p_rx_rbr_ring_t		*rbr_rings;
2428 	p_rx_rcr_rings_t	rx_rcr_rings;
2429 	p_rx_rcr_ring_t		*rcr_rings;
2430 	p_rx_mbox_areas_t	rx_mbox_areas_p;
2431 	p_rx_mbox_t		*rx_mbox_p;
2432 	p_hxge_dma_pool_t	dma_buf_poolp;
2433 	p_hxge_dma_pool_t	dma_rbr_cntl_poolp;
2434 	p_hxge_dma_pool_t	dma_rcr_cntl_poolp;
2435 	p_hxge_dma_pool_t	dma_mbox_cntl_poolp;
2436 	p_hxge_dma_common_t	*dma_buf_p;
2437 
2438 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_unmap_rxdma"));
2439 
2440 	dma_buf_poolp = hxgep->rx_buf_pool_p;
2441 	dma_rbr_cntl_poolp = hxgep->rx_rbr_cntl_pool_p;
2442 	dma_rcr_cntl_poolp = hxgep->rx_rcr_cntl_pool_p;
2443 	dma_mbox_cntl_poolp = hxgep->rx_mbox_cntl_pool_p;
2444 
2445 	if (!dma_buf_poolp->buf_allocated ||
2446 	    !dma_rbr_cntl_poolp->buf_allocated ||
2447 	    !dma_rcr_cntl_poolp->buf_allocated ||
2448 	    !dma_mbox_cntl_poolp->buf_allocated) {
2449 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2450 		    "<== hxge_unmap_rxdma: NULL buf pointers"));
2451 		return;
2452 	}
2453 
2454 	rx_rbr_rings = hxgep->rx_rbr_rings;
2455 	rx_rcr_rings = hxgep->rx_rcr_rings;
2456 	if (rx_rbr_rings == NULL || rx_rcr_rings == NULL) {
2457 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2458 		    "<== hxge_unmap_rxdma: NULL pointers"));
2459 		return;
2460 	}
2461 
2462 	ndmas = rx_rbr_rings->ndmas;
2463 	if (!ndmas) {
2464 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2465 		    "<== hxge_unmap_rxdma: no channel"));
2466 		return;
2467 	}
2468 
2469 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2470 	    "==> hxge_unmap_rxdma (ndmas %d)", ndmas));
2471 
2472 	rbr_rings = rx_rbr_rings->rbr_rings;
2473 	rcr_rings = rx_rcr_rings->rcr_rings;
2474 	rx_mbox_areas_p = hxgep->rx_mbox_areas_p;
2475 	rx_mbox_p = rx_mbox_areas_p->rxmbox_areas;
2476 	dma_buf_p = dma_buf_poolp->dma_buf_pool_p;
2477 
2478 	for (i = 0; i < ndmas; i++) {
2479 		channel = ((p_hxge_dma_common_t)dma_buf_p[i])->dma_channel;
2480 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2481 		    "==> hxge_unmap_rxdma (ndmas %d) channel %d",
2482 		    ndmas, channel));
2483 		(void) hxge_unmap_rxdma_channel(hxgep, channel,
2484 		    (p_rx_rbr_ring_t)rbr_rings[i],
2485 		    (p_rx_rcr_ring_t)rcr_rings[i],
2486 		    (p_rx_mbox_t)rx_mbox_p[i]);
2487 	}
2488 
2489 	KMEM_FREE(rx_rbr_rings, sizeof (rx_rbr_rings_t));
2490 	KMEM_FREE(rbr_rings, sizeof (p_rx_rbr_ring_t) * ndmas);
2491 	KMEM_FREE(rx_rcr_rings, sizeof (rx_rcr_rings_t));
2492 	KMEM_FREE(rcr_rings, sizeof (p_rx_rcr_ring_t) * ndmas);
2493 	KMEM_FREE(rx_mbox_areas_p, sizeof (rx_mbox_areas_t));
2494 	KMEM_FREE(rx_mbox_p, sizeof (p_rx_mbox_t) * ndmas);
2495 
2496 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "<== hxge_unmap_rxdma"));
2497 }
2498 
2499 hxge_status_t
2500 hxge_map_rxdma_channel(p_hxge_t hxgep, uint16_t channel,
2501     p_hxge_dma_common_t *dma_buf_p, p_rx_rbr_ring_t *rbr_p,
2502     uint32_t num_chunks, p_hxge_dma_common_t *dma_rbr_cntl_p,
2503     p_hxge_dma_common_t *dma_rcr_cntl_p, p_hxge_dma_common_t *dma_mbox_cntl_p,
2504     p_rx_rcr_ring_t *rcr_p, p_rx_mbox_t *rx_mbox_p)
2505 {
2506 	int status = HXGE_OK;
2507 
2508 	/*
2509 	 * Set up and prepare buffer blocks, descriptors and mailbox.
2510 	 */
2511 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2512 	    "==> hxge_map_rxdma_channel (channel %d)", channel));
2513 
2514 	/*
2515 	 * Receive buffer blocks
2516 	 */
2517 	status = hxge_map_rxdma_channel_buf_ring(hxgep, channel,
2518 	    dma_buf_p, rbr_p, num_chunks);
2519 	if (status != HXGE_OK) {
2520 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2521 		    "==> hxge_map_rxdma_channel (channel %d): "
2522 		    "map buffer failed 0x%x", channel, status));
2523 		goto hxge_map_rxdma_channel_exit;
2524 	}
2525 
2526 	/*
2527 	 * Receive block ring, completion ring and mailbox.
2528 	 */
2529 	status = hxge_map_rxdma_channel_cfg_ring(hxgep, channel,
2530 	    dma_rbr_cntl_p, dma_rcr_cntl_p, dma_mbox_cntl_p,
2531 	    rbr_p, rcr_p, rx_mbox_p);
2532 	if (status != HXGE_OK) {
2533 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2534 		    "==> hxge_map_rxdma_channel (channel %d): "
2535 		    "map config failed 0x%x", channel, status));
2536 		goto hxge_map_rxdma_channel_fail2;
2537 	}
2538 	goto hxge_map_rxdma_channel_exit;
2539 
2540 hxge_map_rxdma_channel_fail3:
2541 	/* Free rbr, rcr */
2542 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2543 	    "==> hxge_map_rxdma_channel: free rbr/rcr (status 0x%x channel %d)",
2544 	    status, channel));
2545 	hxge_unmap_rxdma_channel_cfg_ring(hxgep, *rcr_p, *rx_mbox_p);
2546 
2547 hxge_map_rxdma_channel_fail2:
2548 	/* Free buffer blocks */
2549 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2550 	    "==> hxge_map_rxdma_channel: free rx buffers"
2551 	    "(hxgep 0x%x status 0x%x channel %d)",
2552 	    hxgep, status, channel));
2553 	hxge_unmap_rxdma_channel_buf_ring(hxgep, *rbr_p);
2554 
2555 	status = HXGE_ERROR;
2556 
2557 hxge_map_rxdma_channel_exit:
2558 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2559 	    "<== hxge_map_rxdma_channel: (hxgep 0x%x status 0x%x channel %d)",
2560 	    hxgep, status, channel));
2561 
2562 	return (status);
2563 }
2564 
2565 /*ARGSUSED*/
2566 static void
2567 hxge_unmap_rxdma_channel(p_hxge_t hxgep, uint16_t channel,
2568     p_rx_rbr_ring_t rbr_p, p_rx_rcr_ring_t rcr_p, p_rx_mbox_t rx_mbox_p)
2569 {
2570 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2571 	    "==> hxge_unmap_rxdma_channel (channel %d)", channel));
2572 
2573 	/*
2574 	 * unmap receive block ring, completion ring and mailbox.
2575 	 */
2576 	(void) hxge_unmap_rxdma_channel_cfg_ring(hxgep, rcr_p, rx_mbox_p);
2577 
2578 	/* unmap buffer blocks */
2579 	(void) hxge_unmap_rxdma_channel_buf_ring(hxgep, rbr_p);
2580 
2581 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "<== hxge_unmap_rxdma_channel"));
2582 }
2583 
2584 /*ARGSUSED*/
2585 static hxge_status_t
2586 hxge_map_rxdma_channel_cfg_ring(p_hxge_t hxgep, uint16_t dma_channel,
2587     p_hxge_dma_common_t *dma_rbr_cntl_p, p_hxge_dma_common_t *dma_rcr_cntl_p,
2588     p_hxge_dma_common_t *dma_mbox_cntl_p, p_rx_rbr_ring_t *rbr_p,
2589     p_rx_rcr_ring_t *rcr_p, p_rx_mbox_t *rx_mbox_p)
2590 {
2591 	p_rx_rbr_ring_t 	rbrp;
2592 	p_rx_rcr_ring_t 	rcrp;
2593 	p_rx_mbox_t 		mboxp;
2594 	p_hxge_dma_common_t 	cntl_dmap;
2595 	p_hxge_dma_common_t 	dmap;
2596 	p_rx_msg_t 		*rx_msg_ring;
2597 	p_rx_msg_t 		rx_msg_p;
2598 	rdc_rbr_cfg_a_t		*rcfga_p;
2599 	rdc_rbr_cfg_b_t		*rcfgb_p;
2600 	rdc_rcr_cfg_a_t		*cfga_p;
2601 	rdc_rcr_cfg_b_t		*cfgb_p;
2602 	rdc_rx_cfg1_t		*cfig1_p;
2603 	rdc_rx_cfg2_t		*cfig2_p;
2604 	rdc_rbr_kick_t		*kick_p;
2605 	uint32_t		dmaaddrp;
2606 	uint32_t		*rbr_vaddrp;
2607 	uint32_t		bkaddr;
2608 	hxge_status_t		status = HXGE_OK;
2609 	int			i;
2610 	uint32_t 		hxge_port_rcr_size;
2611 
2612 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2613 	    "==> hxge_map_rxdma_channel_cfg_ring"));
2614 
2615 	cntl_dmap = *dma_rbr_cntl_p;
2616 
2617 	/*
2618 	 * Map in the receive block ring
2619 	 */
2620 	rbrp = *rbr_p;
2621 	dmap = (p_hxge_dma_common_t)&rbrp->rbr_desc;
2622 	hxge_setup_dma_common(dmap, cntl_dmap, rbrp->rbb_max, 4);
2623 
2624 	/*
2625 	 * Zero out buffer block ring descriptors.
2626 	 */
2627 	bzero((caddr_t)dmap->kaddrp, dmap->alength);
2628 
2629 	rcfga_p = &(rbrp->rbr_cfga);
2630 	rcfgb_p = &(rbrp->rbr_cfgb);
2631 	kick_p = &(rbrp->rbr_kick);
2632 	rcfga_p->value = 0;
2633 	rcfgb_p->value = 0;
2634 	kick_p->value = 0;
2635 	rbrp->rbr_addr = dmap->dma_cookie.dmac_laddress;
2636 	rcfga_p->value = (rbrp->rbr_addr &
2637 	    (RBR_CFIG_A_STDADDR_MASK | RBR_CFIG_A_STDADDR_BASE_MASK));
2638 	rcfga_p->value |= ((uint64_t)rbrp->rbb_max << RBR_CFIG_A_LEN_SHIFT);
2639 
2640 	/* XXXX: how to choose packet buffer sizes */
2641 	rcfgb_p->bits.bufsz0 = rbrp->pkt_buf_size0;
2642 	rcfgb_p->bits.vld0 = 1;
2643 	rcfgb_p->bits.bufsz1 = rbrp->pkt_buf_size1;
2644 	rcfgb_p->bits.vld1 = 1;
2645 	rcfgb_p->bits.bufsz2 = rbrp->pkt_buf_size2;
2646 	rcfgb_p->bits.vld2 = 1;
2647 	rcfgb_p->bits.bksize = hxgep->rx_bksize_code;
2648 
2649 	/*
2650 	 * For each buffer block, enter receive block address to the ring.
2651 	 */
2652 	rbr_vaddrp = (uint32_t *)dmap->kaddrp;
2653 	rbrp->rbr_desc_vp = (uint32_t *)dmap->kaddrp;
2654 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2655 	    "==> hxge_map_rxdma_channel_cfg_ring: channel %d "
2656 	    "rbr_vaddrp $%p", dma_channel, rbr_vaddrp));
2657 
2658 	rx_msg_ring = rbrp->rx_msg_ring;
2659 	for (i = 0; i < rbrp->tnblocks; i++) {
2660 		rx_msg_p = rx_msg_ring[i];
2661 		rx_msg_p->hxgep = hxgep;
2662 		rx_msg_p->rx_rbr_p = rbrp;
2663 		bkaddr = (uint32_t)
2664 		    ((rx_msg_p->buf_dma.dma_cookie.dmac_laddress >>
2665 		    RBR_BKADDR_SHIFT));
2666 		rx_msg_p->free = B_FALSE;
2667 		rx_msg_p->max_usage_cnt = 0xbaddcafe;
2668 
2669 		*rbr_vaddrp++ = bkaddr;
2670 	}
2671 
2672 	kick_p->bits.bkadd = rbrp->rbb_max;
2673 	rbrp->rbr_wr_index = (rbrp->rbb_max - 1);
2674 
2675 	rbrp->rbr_rd_index = 0;
2676 
2677 	rbrp->rbr_consumed = 0;
2678 	rbrp->rbr_used = 0;
2679 	rbrp->rbr_use_bcopy = B_TRUE;
2680 	rbrp->rbr_bufsize_type = RCR_PKTBUFSZ_0;
2681 
2682 	/*
2683 	 * Do bcopy on packets greater than bcopy size once the lo threshold is
2684 	 * reached. This lo threshold should be less than the hi threshold.
2685 	 *
2686 	 * Do bcopy on every packet once the hi threshold is reached.
2687 	 */
2688 	if (hxge_rx_threshold_lo >= hxge_rx_threshold_hi) {
2689 		/* default it to use hi */
2690 		hxge_rx_threshold_lo = hxge_rx_threshold_hi;
2691 	}
2692 	if (hxge_rx_buf_size_type > HXGE_RBR_TYPE2) {
2693 		hxge_rx_buf_size_type = HXGE_RBR_TYPE2;
2694 	}
2695 	rbrp->rbr_bufsize_type = hxge_rx_buf_size_type;
2696 
2697 	switch (hxge_rx_threshold_hi) {
2698 	default:
2699 	case HXGE_RX_COPY_NONE:
2700 		/* Do not do bcopy at all */
2701 		rbrp->rbr_use_bcopy = B_FALSE;
2702 		rbrp->rbr_threshold_hi = rbrp->rbb_max;
2703 		break;
2704 
2705 	case HXGE_RX_COPY_1:
2706 	case HXGE_RX_COPY_2:
2707 	case HXGE_RX_COPY_3:
2708 	case HXGE_RX_COPY_4:
2709 	case HXGE_RX_COPY_5:
2710 	case HXGE_RX_COPY_6:
2711 	case HXGE_RX_COPY_7:
2712 		rbrp->rbr_threshold_hi =
2713 		    rbrp->rbb_max * (hxge_rx_threshold_hi) /
2714 		    HXGE_RX_BCOPY_SCALE;
2715 		break;
2716 
2717 	case HXGE_RX_COPY_ALL:
2718 		rbrp->rbr_threshold_hi = 0;
2719 		break;
2720 	}
2721 
2722 	switch (hxge_rx_threshold_lo) {
2723 	default:
2724 	case HXGE_RX_COPY_NONE:
2725 		/* Do not do bcopy at all */
2726 		if (rbrp->rbr_use_bcopy) {
2727 			rbrp->rbr_use_bcopy = B_FALSE;
2728 		}
2729 		rbrp->rbr_threshold_lo = rbrp->rbb_max;
2730 		break;
2731 
2732 	case HXGE_RX_COPY_1:
2733 	case HXGE_RX_COPY_2:
2734 	case HXGE_RX_COPY_3:
2735 	case HXGE_RX_COPY_4:
2736 	case HXGE_RX_COPY_5:
2737 	case HXGE_RX_COPY_6:
2738 	case HXGE_RX_COPY_7:
2739 		rbrp->rbr_threshold_lo =
2740 		    rbrp->rbb_max * (hxge_rx_threshold_lo) /
2741 		    HXGE_RX_BCOPY_SCALE;
2742 		break;
2743 
2744 	case HXGE_RX_COPY_ALL:
2745 		rbrp->rbr_threshold_lo = 0;
2746 		break;
2747 	}
2748 
2749 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
2750 	    "hxge_map_rxdma_channel_cfg_ring: channel %d rbb_max %d "
2751 	    "rbrp->rbr_bufsize_type %d rbb_threshold_hi %d "
2752 	    "rbb_threshold_lo %d",
2753 	    dma_channel, rbrp->rbb_max, rbrp->rbr_bufsize_type,
2754 	    rbrp->rbr_threshold_hi, rbrp->rbr_threshold_lo));
2755 
2756 	/* Map in the receive completion ring */
2757 	rcrp = (p_rx_rcr_ring_t)KMEM_ZALLOC(sizeof (rx_rcr_ring_t), KM_SLEEP);
2758 	rcrp->rdc = dma_channel;
2759 	rcrp->hxgep = hxgep;
2760 
2761 	hxge_port_rcr_size = hxgep->hxge_port_rcr_size;
2762 	rcrp->comp_size = hxge_port_rcr_size;
2763 	rcrp->comp_wrap_mask = hxge_port_rcr_size - 1;
2764 
2765 	cntl_dmap = *dma_rcr_cntl_p;
2766 
2767 	dmap = (p_hxge_dma_common_t)&rcrp->rcr_desc;
2768 	hxge_setup_dma_common(dmap, cntl_dmap, rcrp->comp_size,
2769 	    sizeof (rcr_entry_t));
2770 	rcrp->comp_rd_index = 0;
2771 	rcrp->comp_wt_index = 0;
2772 	rcrp->rcr_desc_rd_head_p = rcrp->rcr_desc_first_p =
2773 	    (p_rcr_entry_t)DMA_COMMON_VPTR(rcrp->rcr_desc);
2774 #if defined(__i386)
2775 	rcrp->rcr_desc_rd_head_pp = rcrp->rcr_desc_first_pp =
2776 	    (p_rcr_entry_t)(uint32_t)DMA_COMMON_IOADDR(rcrp->rcr_desc);
2777 #else
2778 	rcrp->rcr_desc_rd_head_pp = rcrp->rcr_desc_first_pp =
2779 	    (p_rcr_entry_t)DMA_COMMON_IOADDR(rcrp->rcr_desc);
2780 #endif
2781 	rcrp->rcr_desc_last_p = rcrp->rcr_desc_rd_head_p +
2782 	    (hxge_port_rcr_size - 1);
2783 	rcrp->rcr_desc_last_pp = rcrp->rcr_desc_rd_head_pp +
2784 	    (hxge_port_rcr_size - 1);
2785 
2786 	rcrp->rcr_tail_begin = DMA_COMMON_IOADDR(rcrp->rcr_desc);
2787 	rcrp->rcr_tail_begin = (rcrp->rcr_tail_begin & 0x7ffffULL) >> 3;
2788 
2789 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2790 	    "==> hxge_map_rxdma_channel_cfg_ring: channel %d "
2791 	    "rbr_vaddrp $%p rcr_desc_rd_head_p $%p "
2792 	    "rcr_desc_rd_head_pp $%p rcr_desc_rd_last_p $%p "
2793 	    "rcr_desc_rd_last_pp $%p ",
2794 	    dma_channel, rbr_vaddrp, rcrp->rcr_desc_rd_head_p,
2795 	    rcrp->rcr_desc_rd_head_pp, rcrp->rcr_desc_last_p,
2796 	    rcrp->rcr_desc_last_pp));
2797 
2798 	/*
2799 	 * Zero out buffer block ring descriptors.
2800 	 */
2801 	bzero((caddr_t)dmap->kaddrp, dmap->alength);
2802 	rcrp->intr_timeout = hxgep->intr_timeout;
2803 	rcrp->intr_threshold = hxgep->intr_threshold;
2804 	rcrp->full_hdr_flag = B_FALSE;
2805 	rcrp->sw_priv_hdr_len = 0;
2806 
2807 	cfga_p = &(rcrp->rcr_cfga);
2808 	cfgb_p = &(rcrp->rcr_cfgb);
2809 	cfga_p->value = 0;
2810 	cfgb_p->value = 0;
2811 	rcrp->rcr_addr = dmap->dma_cookie.dmac_laddress;
2812 
2813 	cfga_p->value = (rcrp->rcr_addr &
2814 	    (RCRCFIG_A_STADDR_MASK | RCRCFIG_A_STADDR_BASE_MASK));
2815 
2816 	cfga_p->value |= ((uint64_t)rcrp->comp_size << RCRCFIG_A_LEN_SHIF);
2817 
2818 	/*
2819 	 * Timeout should be set based on the system clock divider. The
2820 	 * following timeout value of 1 assumes that the granularity (1000) is
2821 	 * 3 microseconds running at 300MHz.
2822 	 */
2823 	cfgb_p->bits.pthres = rcrp->intr_threshold;
2824 	cfgb_p->bits.timeout = rcrp->intr_timeout;
2825 	cfgb_p->bits.entout = 1;
2826 
2827 	/* Map in the mailbox */
2828 	cntl_dmap = *dma_mbox_cntl_p;
2829 	mboxp = (p_rx_mbox_t)KMEM_ZALLOC(sizeof (rx_mbox_t), KM_SLEEP);
2830 	dmap = (p_hxge_dma_common_t)&mboxp->rx_mbox;
2831 	hxge_setup_dma_common(dmap, cntl_dmap, 1, sizeof (rxdma_mailbox_t));
2832 	cfig1_p = (rdc_rx_cfg1_t *)&mboxp->rx_cfg1;
2833 	cfig2_p = (rdc_rx_cfg2_t *)&mboxp->rx_cfg2;
2834 	cfig1_p->value = cfig2_p->value = 0;
2835 
2836 	mboxp->mbox_addr = dmap->dma_cookie.dmac_laddress;
2837 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2838 	    "==> hxge_map_rxdma_channel_cfg_ring: "
2839 	    "channel %d cfg1 0x%016llx cfig2 0x%016llx cookie 0x%016llx",
2840 	    dma_channel, cfig1_p->value, cfig2_p->value,
2841 	    mboxp->mbox_addr));
2842 
2843 	dmaaddrp = (uint32_t)((dmap->dma_cookie.dmac_laddress >> 32) & 0xfff);
2844 	cfig1_p->bits.mbaddr_h = dmaaddrp;
2845 
2846 	dmaaddrp = (uint32_t)(dmap->dma_cookie.dmac_laddress & 0xffffffff);
2847 	dmaaddrp = (uint32_t)(dmap->dma_cookie.dmac_laddress &
2848 	    RXDMA_CFIG2_MBADDR_L_MASK);
2849 
2850 	cfig2_p->bits.mbaddr_l = (dmaaddrp >> RXDMA_CFIG2_MBADDR_L_SHIFT);
2851 
2852 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2853 	    "==> hxge_map_rxdma_channel_cfg_ring: channel %d damaddrp $%p "
2854 	    "cfg1 0x%016llx cfig2 0x%016llx",
2855 	    dma_channel, dmaaddrp, cfig1_p->value, cfig2_p->value));
2856 
2857 	cfig2_p->bits.full_hdr = rcrp->full_hdr_flag;
2858 	cfig2_p->bits.offset = rcrp->sw_priv_hdr_len;
2859 
2860 	rbrp->rx_rcr_p = rcrp;
2861 	rcrp->rx_rbr_p = rbrp;
2862 	*rcr_p = rcrp;
2863 	*rx_mbox_p = mboxp;
2864 
2865 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2866 	    "<== hxge_map_rxdma_channel_cfg_ring status 0x%08x", status));
2867 	return (status);
2868 }
2869 
2870 /*ARGSUSED*/
2871 static void
2872 hxge_unmap_rxdma_channel_cfg_ring(p_hxge_t hxgep,
2873     p_rx_rcr_ring_t rcr_p, p_rx_mbox_t rx_mbox_p)
2874 {
2875 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2876 	    "==> hxge_unmap_rxdma_channel_cfg_ring: channel %d", rcr_p->rdc));
2877 
2878 	KMEM_FREE(rcr_p, sizeof (rx_rcr_ring_t));
2879 	KMEM_FREE(rx_mbox_p, sizeof (rx_mbox_t));
2880 
2881 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2882 	    "<== hxge_unmap_rxdma_channel_cfg_ring"));
2883 }
2884 
2885 static hxge_status_t
2886 hxge_map_rxdma_channel_buf_ring(p_hxge_t hxgep, uint16_t channel,
2887     p_hxge_dma_common_t *dma_buf_p,
2888     p_rx_rbr_ring_t *rbr_p, uint32_t num_chunks)
2889 {
2890 	p_rx_rbr_ring_t		rbrp;
2891 	p_hxge_dma_common_t	dma_bufp, tmp_bufp;
2892 	p_rx_msg_t		*rx_msg_ring;
2893 	p_rx_msg_t		rx_msg_p;
2894 	p_mblk_t		mblk_p;
2895 
2896 	rxring_info_t *ring_info;
2897 	hxge_status_t status = HXGE_OK;
2898 	int i, j, index;
2899 	uint32_t size, bsize, nblocks, nmsgs;
2900 
2901 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2902 	    "==> hxge_map_rxdma_channel_buf_ring: channel %d", channel));
2903 
2904 	dma_bufp = tmp_bufp = *dma_buf_p;
2905 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2906 	    " hxge_map_rxdma_channel_buf_ring: channel %d to map %d "
2907 	    "chunks bufp 0x%016llx", channel, num_chunks, dma_bufp));
2908 
2909 	nmsgs = 0;
2910 	for (i = 0; i < num_chunks; i++, tmp_bufp++) {
2911 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2912 		    "==> hxge_map_rxdma_channel_buf_ring: channel %d "
2913 		    "bufp 0x%016llx nblocks %d nmsgs %d",
2914 		    channel, tmp_bufp, tmp_bufp->nblocks, nmsgs));
2915 		nmsgs += tmp_bufp->nblocks;
2916 	}
2917 	if (!nmsgs) {
2918 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2919 		    "<== hxge_map_rxdma_channel_buf_ring: channel %d "
2920 		    "no msg blocks", channel));
2921 		status = HXGE_ERROR;
2922 		goto hxge_map_rxdma_channel_buf_ring_exit;
2923 	}
2924 	rbrp = (p_rx_rbr_ring_t)KMEM_ZALLOC(sizeof (rx_rbr_ring_t), KM_SLEEP);
2925 
2926 	size = nmsgs * sizeof (p_rx_msg_t);
2927 	rx_msg_ring = KMEM_ZALLOC(size, KM_SLEEP);
2928 	ring_info = (rxring_info_t *)KMEM_ZALLOC(sizeof (rxring_info_t),
2929 	    KM_SLEEP);
2930 
2931 	MUTEX_INIT(&rbrp->lock, NULL, MUTEX_DRIVER,
2932 	    (void *) hxgep->interrupt_cookie);
2933 	MUTEX_INIT(&rbrp->post_lock, NULL, MUTEX_DRIVER,
2934 	    (void *) hxgep->interrupt_cookie);
2935 
2936 	rbrp->rdc = channel;
2937 	rbrp->num_blocks = num_chunks;
2938 	rbrp->tnblocks = nmsgs;
2939 	rbrp->rbb_max = nmsgs;
2940 	rbrp->rbr_max_size = nmsgs;
2941 	rbrp->rbr_wrap_mask = (rbrp->rbb_max - 1);
2942 
2943 	/*
2944 	 * Buffer sizes suggested by NIU architect. 256, 512 and 2K.
2945 	 */
2946 
2947 	switch (hxgep->rx_bksize_code) {
2948 	case RBR_BKSIZE_4K:
2949 		rbrp->pkt_buf_size0 = RBR_BUFSZ0_256B;
2950 		rbrp->pkt_buf_size0_bytes = RBR_BUFSZ0_256_BYTES;
2951 		rbrp->hpi_pkt_buf_size0 = SIZE_256B;
2952 		break;
2953 	case RBR_BKSIZE_8K:
2954 		/* Use 512 to avoid possible rcr_full condition */
2955 		rbrp->pkt_buf_size0 = RBR_BUFSZ0_512B;
2956 		rbrp->pkt_buf_size0_bytes = RBR_BUFSZ0_512_BYTES;
2957 		rbrp->hpi_pkt_buf_size0 = SIZE_512B;
2958 		break;
2959 	}
2960 
2961 	rbrp->pkt_buf_size1 = RBR_BUFSZ1_1K;
2962 	rbrp->pkt_buf_size1_bytes = RBR_BUFSZ1_1K_BYTES;
2963 	rbrp->hpi_pkt_buf_size1 = SIZE_1KB;
2964 
2965 	rbrp->block_size = hxgep->rx_default_block_size;
2966 
2967 	if (!hxgep->param_arr[param_accept_jumbo].value) {
2968 		rbrp->pkt_buf_size2 = RBR_BUFSZ2_2K;
2969 		rbrp->pkt_buf_size2_bytes = RBR_BUFSZ2_2K_BYTES;
2970 		rbrp->hpi_pkt_buf_size2 = SIZE_2KB;
2971 	} else {
2972 		rbrp->hpi_pkt_buf_size2 = SIZE_4KB;
2973 		rbrp->pkt_buf_size2 = RBR_BUFSZ2_4K;
2974 		rbrp->pkt_buf_size2_bytes = RBR_BUFSZ2_4K_BYTES;
2975 	}
2976 
2977 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2978 	    "==> hxge_map_rxdma_channel_buf_ring: channel %d "
2979 	    "actual rbr max %d rbb_max %d nmsgs %d "
2980 	    "rbrp->block_size %d default_block_size %d "
2981 	    "(config hxge_rbr_size %d hxge_rbr_spare_size %d)",
2982 	    channel, rbrp->rbr_max_size, rbrp->rbb_max, nmsgs,
2983 	    rbrp->block_size, hxgep->rx_default_block_size,
2984 	    hxge_rbr_size, hxge_rbr_spare_size));
2985 
2986 	/*
2987 	 * Map in buffers from the buffer pool.
2988 	 * Note that num_blocks is the num_chunks. For Sparc, there is likely
2989 	 * only one chunk. For x86, there will be many chunks.
2990 	 * Loop over chunks.
2991 	 */
2992 	index = 0;
2993 	for (i = 0; i < rbrp->num_blocks; i++, dma_bufp++) {
2994 		bsize = dma_bufp->block_size;
2995 		nblocks = dma_bufp->nblocks;
2996 #if defined(__i386)
2997 		ring_info->buffer[i].dvma_addr = (uint32_t)dma_bufp->ioaddr_pp;
2998 #else
2999 		ring_info->buffer[i].dvma_addr = (uint64_t)dma_bufp->ioaddr_pp;
3000 #endif
3001 		ring_info->buffer[i].buf_index = i;
3002 		ring_info->buffer[i].buf_size = dma_bufp->alength;
3003 		ring_info->buffer[i].start_index = index;
3004 #if defined(__i386)
3005 		ring_info->buffer[i].kaddr = (uint32_t)dma_bufp->kaddrp;
3006 #else
3007 		ring_info->buffer[i].kaddr = (uint64_t)dma_bufp->kaddrp;
3008 #endif
3009 
3010 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3011 		    " hxge_map_rxdma_channel_buf_ring: map channel %d "
3012 		    "chunk %d nblocks %d chunk_size %x block_size 0x%x "
3013 		    "dma_bufp $%p dvma_addr $%p", channel, i,
3014 		    dma_bufp->nblocks,
3015 		    ring_info->buffer[i].buf_size, bsize, dma_bufp,
3016 		    ring_info->buffer[i].dvma_addr));
3017 
3018 		/* loop over blocks within a chunk */
3019 		for (j = 0; j < nblocks; j++) {
3020 			if ((rx_msg_p = hxge_allocb(bsize, BPRI_LO,
3021 			    dma_bufp)) == NULL) {
3022 				HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3023 				    "allocb failed (index %d i %d j %d)",
3024 				    index, i, j));
3025 				goto hxge_map_rxdma_channel_buf_ring_fail1;
3026 			}
3027 			rx_msg_ring[index] = rx_msg_p;
3028 			rx_msg_p->block_index = index;
3029 			rx_msg_p->shifted_addr = (uint32_t)
3030 			    ((rx_msg_p->buf_dma.dma_cookie.dmac_laddress >>
3031 			    RBR_BKADDR_SHIFT));
3032 			/*
3033 			 * Too much output
3034 			 * HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3035 			 *	"index %d j %d rx_msg_p $%p mblk %p",
3036 			 *	index, j, rx_msg_p, rx_msg_p->rx_mblk_p));
3037 			 */
3038 			mblk_p = rx_msg_p->rx_mblk_p;
3039 			mblk_p->b_wptr = mblk_p->b_rptr + bsize;
3040 
3041 			rbrp->rbr_ref_cnt++;
3042 			index++;
3043 			rx_msg_p->buf_dma.dma_channel = channel;
3044 		}
3045 	}
3046 	if (i < rbrp->num_blocks) {
3047 		goto hxge_map_rxdma_channel_buf_ring_fail1;
3048 	}
3049 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3050 	    "hxge_map_rxdma_channel_buf_ring: done buf init "
3051 	    "channel %d msg block entries %d", channel, index));
3052 	ring_info->block_size_mask = bsize - 1;
3053 	rbrp->rx_msg_ring = rx_msg_ring;
3054 	rbrp->dma_bufp = dma_buf_p;
3055 	rbrp->ring_info = ring_info;
3056 
3057 	status = hxge_rxbuf_index_info_init(hxgep, rbrp);
3058 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, " hxge_map_rxdma_channel_buf_ring: "
3059 	    "channel %d done buf info init", channel));
3060 
3061 	/*
3062 	 * Finally, permit hxge_freeb() to call hxge_post_page().
3063 	 */
3064 	rbrp->rbr_state = RBR_POSTING;
3065 
3066 	*rbr_p = rbrp;
3067 
3068 	goto hxge_map_rxdma_channel_buf_ring_exit;
3069 
3070 hxge_map_rxdma_channel_buf_ring_fail1:
3071 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3072 	    " hxge_map_rxdma_channel_buf_ring: failed channel (0x%x)",
3073 	    channel, status));
3074 
3075 	index--;
3076 	for (; index >= 0; index--) {
3077 		rx_msg_p = rx_msg_ring[index];
3078 		if (rx_msg_p != NULL) {
3079 			freeb(rx_msg_p->rx_mblk_p);
3080 			rx_msg_ring[index] = NULL;
3081 		}
3082 	}
3083 
3084 hxge_map_rxdma_channel_buf_ring_fail:
3085 	MUTEX_DESTROY(&rbrp->post_lock);
3086 	MUTEX_DESTROY(&rbrp->lock);
3087 	KMEM_FREE(ring_info, sizeof (rxring_info_t));
3088 	KMEM_FREE(rx_msg_ring, size);
3089 	KMEM_FREE(rbrp, sizeof (rx_rbr_ring_t));
3090 
3091 	status = HXGE_ERROR;
3092 
3093 hxge_map_rxdma_channel_buf_ring_exit:
3094 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3095 	    "<== hxge_map_rxdma_channel_buf_ring status 0x%08x", status));
3096 
3097 	return (status);
3098 }
3099 
3100 /*ARGSUSED*/
3101 static void
3102 hxge_unmap_rxdma_channel_buf_ring(p_hxge_t hxgep,
3103     p_rx_rbr_ring_t rbr_p)
3104 {
3105 	p_rx_msg_t	*rx_msg_ring;
3106 	p_rx_msg_t	rx_msg_p;
3107 	rxring_info_t	*ring_info;
3108 	int		i;
3109 	uint32_t	size;
3110 
3111 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3112 	    "==> hxge_unmap_rxdma_channel_buf_ring"));
3113 	if (rbr_p == NULL) {
3114 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
3115 		    "<== hxge_unmap_rxdma_channel_buf_ring: NULL rbrp"));
3116 		return;
3117 	}
3118 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3119 	    "==> hxge_unmap_rxdma_channel_buf_ring: channel %d", rbr_p->rdc));
3120 
3121 	rx_msg_ring = rbr_p->rx_msg_ring;
3122 	ring_info = rbr_p->ring_info;
3123 
3124 	if (rx_msg_ring == NULL || ring_info == NULL) {
3125 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3126 		    "<== hxge_unmap_rxdma_channel_buf_ring: "
3127 		    "rx_msg_ring $%p ring_info $%p", rx_msg_p, ring_info));
3128 		return;
3129 	}
3130 
3131 	size = rbr_p->tnblocks * sizeof (p_rx_msg_t);
3132 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3133 	    " hxge_unmap_rxdma_channel_buf_ring: channel %d chunks %d "
3134 	    "tnblocks %d (max %d) size ptrs %d ", rbr_p->rdc, rbr_p->num_blocks,
3135 	    rbr_p->tnblocks, rbr_p->rbr_max_size, size));
3136 
3137 	for (i = 0; i < rbr_p->tnblocks; i++) {
3138 		rx_msg_p = rx_msg_ring[i];
3139 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3140 		    " hxge_unmap_rxdma_channel_buf_ring: "
3141 		    "rx_msg_p $%p", rx_msg_p));
3142 		if (rx_msg_p != NULL) {
3143 			freeb(rx_msg_p->rx_mblk_p);
3144 			rx_msg_ring[i] = NULL;
3145 		}
3146 	}
3147 
3148 	/*
3149 	 * We no longer may use the mutex <post_lock>. By setting
3150 	 * <rbr_state> to anything but POSTING, we prevent
3151 	 * hxge_post_page() from accessing a dead mutex.
3152 	 */
3153 	rbr_p->rbr_state = RBR_UNMAPPING;
3154 	MUTEX_DESTROY(&rbr_p->post_lock);
3155 
3156 	MUTEX_DESTROY(&rbr_p->lock);
3157 	KMEM_FREE(ring_info, sizeof (rxring_info_t));
3158 	KMEM_FREE(rx_msg_ring, size);
3159 
3160 	if (rbr_p->rbr_ref_cnt == 0) {
3161 		/* This is the normal state of affairs. */
3162 		KMEM_FREE(rbr_p, sizeof (*rbr_p));
3163 	} else {
3164 		/*
3165 		 * Some of our buffers are still being used.
3166 		 * Therefore, tell hxge_freeb() this ring is
3167 		 * unmapped, so it may free <rbr_p> for us.
3168 		 */
3169 		rbr_p->rbr_state = RBR_UNMAPPED;
3170 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3171 		    "unmap_rxdma_buf_ring: %d %s outstanding.",
3172 		    rbr_p->rbr_ref_cnt,
3173 		    rbr_p->rbr_ref_cnt == 1 ? "msg" : "msgs"));
3174 	}
3175 
3176 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3177 	    "<== hxge_unmap_rxdma_channel_buf_ring"));
3178 }
3179 
3180 static hxge_status_t
3181 hxge_rxdma_hw_start_common(p_hxge_t hxgep)
3182 {
3183 	hxge_status_t status = HXGE_OK;
3184 
3185 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_hw_start_common"));
3186 
3187 	/*
3188 	 * Load the sharable parameters by writing to the function zero control
3189 	 * registers. These FZC registers should be initialized only once for
3190 	 * the entire chip.
3191 	 */
3192 	(void) hxge_init_fzc_rx_common(hxgep);
3193 
3194 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_hw_start_common"));
3195 
3196 	return (status);
3197 }
3198 
3199 static hxge_status_t
3200 hxge_rxdma_hw_start(p_hxge_t hxgep)
3201 {
3202 	int			i, ndmas;
3203 	uint16_t		channel;
3204 	p_rx_rbr_rings_t	rx_rbr_rings;
3205 	p_rx_rbr_ring_t		*rbr_rings;
3206 	p_rx_rcr_rings_t	rx_rcr_rings;
3207 	p_rx_rcr_ring_t		*rcr_rings;
3208 	p_rx_mbox_areas_t	rx_mbox_areas_p;
3209 	p_rx_mbox_t		*rx_mbox_p;
3210 	hxge_status_t		status = HXGE_OK;
3211 
3212 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_hw_start"));
3213 
3214 	rx_rbr_rings = hxgep->rx_rbr_rings;
3215 	rx_rcr_rings = hxgep->rx_rcr_rings;
3216 	if (rx_rbr_rings == NULL || rx_rcr_rings == NULL) {
3217 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
3218 		    "<== hxge_rxdma_hw_start: NULL ring pointers"));
3219 		return (HXGE_ERROR);
3220 	}
3221 
3222 	ndmas = rx_rbr_rings->ndmas;
3223 	if (ndmas == 0) {
3224 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
3225 		    "<== hxge_rxdma_hw_start: no dma channel allocated"));
3226 		return (HXGE_ERROR);
3227 	}
3228 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3229 	    "==> hxge_rxdma_hw_start (ndmas %d)", ndmas));
3230 
3231 	/*
3232 	 * Scrub the RDC Rx DMA Prefetch Buffer Command.
3233 	 */
3234 	for (i = 0; i < 128; i++) {
3235 		HXGE_REG_WR64(hxgep->hpi_handle, RDC_PREF_CMD, i);
3236 	}
3237 
3238 	/*
3239 	 * Scrub Rx DMA Shadow Tail Command.
3240 	 */
3241 	for (i = 0; i < 64; i++) {
3242 		HXGE_REG_WR64(hxgep->hpi_handle, RDC_SHADOW_CMD, i);
3243 	}
3244 
3245 	/*
3246 	 * Scrub Rx DMA Control Fifo Command.
3247 	 */
3248 	for (i = 0; i < 512; i++) {
3249 		HXGE_REG_WR64(hxgep->hpi_handle, RDC_CTRL_FIFO_CMD, i);
3250 	}
3251 
3252 	/*
3253 	 * Scrub Rx DMA Data Fifo Command.
3254 	 */
3255 	for (i = 0; i < 1536; i++) {
3256 		HXGE_REG_WR64(hxgep->hpi_handle, RDC_DATA_FIFO_CMD, i);
3257 	}
3258 
3259 	/*
3260 	 * Reset the FIFO Error Stat.
3261 	 */
3262 	HXGE_REG_WR64(hxgep->hpi_handle, RDC_FIFO_ERR_STAT, 0xFF);
3263 
3264 	/* Set the error mask to receive interrupts */
3265 	HXGE_REG_WR64(hxgep->hpi_handle, RDC_FIFO_ERR_INT_MASK, 0x0);
3266 
3267 	rbr_rings = rx_rbr_rings->rbr_rings;
3268 	rcr_rings = rx_rcr_rings->rcr_rings;
3269 	rx_mbox_areas_p = hxgep->rx_mbox_areas_p;
3270 	if (rx_mbox_areas_p) {
3271 		rx_mbox_p = rx_mbox_areas_p->rxmbox_areas;
3272 	}
3273 
3274 	for (i = 0; i < ndmas; i++) {
3275 		channel = rbr_rings[i]->rdc;
3276 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3277 		    "==> hxge_rxdma_hw_start (ndmas %d) channel %d",
3278 		    ndmas, channel));
3279 		status = hxge_rxdma_start_channel(hxgep, channel,
3280 		    (p_rx_rbr_ring_t)rbr_rings[i],
3281 		    (p_rx_rcr_ring_t)rcr_rings[i],
3282 		    (p_rx_mbox_t)rx_mbox_p[i], rbr_rings[i]->rbb_max);
3283 		if (status != HXGE_OK) {
3284 			goto hxge_rxdma_hw_start_fail1;
3285 		}
3286 	}
3287 
3288 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_hw_start: "
3289 	    "rx_rbr_rings 0x%016llx rings 0x%016llx",
3290 	    rx_rbr_rings, rx_rcr_rings));
3291 	goto hxge_rxdma_hw_start_exit;
3292 
3293 hxge_rxdma_hw_start_fail1:
3294 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3295 	    "==> hxge_rxdma_hw_start: disable "
3296 	    "(status 0x%x channel %d i %d)", status, channel, i));
3297 	for (; i >= 0; i--) {
3298 		channel = rbr_rings[i]->rdc;
3299 		(void) hxge_rxdma_stop_channel(hxgep, channel);
3300 	}
3301 
3302 hxge_rxdma_hw_start_exit:
3303 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3304 	    "==> hxge_rxdma_hw_start: (status 0x%x)", status));
3305 	return (status);
3306 }
3307 
3308 static void
3309 hxge_rxdma_hw_stop(p_hxge_t hxgep)
3310 {
3311 	int			i, ndmas;
3312 	uint16_t		channel;
3313 	p_rx_rbr_rings_t	rx_rbr_rings;
3314 	p_rx_rbr_ring_t		*rbr_rings;
3315 	p_rx_rcr_rings_t	rx_rcr_rings;
3316 
3317 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_hw_stop"));
3318 
3319 	rx_rbr_rings = hxgep->rx_rbr_rings;
3320 	rx_rcr_rings = hxgep->rx_rcr_rings;
3321 
3322 	if (rx_rbr_rings == NULL || rx_rcr_rings == NULL) {
3323 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
3324 		    "<== hxge_rxdma_hw_stop: NULL ring pointers"));
3325 		return;
3326 	}
3327 
3328 	ndmas = rx_rbr_rings->ndmas;
3329 	if (!ndmas) {
3330 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
3331 		    "<== hxge_rxdma_hw_stop: no dma channel allocated"));
3332 		return;
3333 	}
3334 
3335 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3336 	    "==> hxge_rxdma_hw_stop (ndmas %d)", ndmas));
3337 
3338 	rbr_rings = rx_rbr_rings->rbr_rings;
3339 	for (i = 0; i < ndmas; i++) {
3340 		channel = rbr_rings[i]->rdc;
3341 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3342 		    "==> hxge_rxdma_hw_stop (ndmas %d) channel %d",
3343 		    ndmas, channel));
3344 		(void) hxge_rxdma_stop_channel(hxgep, channel);
3345 	}
3346 
3347 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_hw_stop: "
3348 	    "rx_rbr_rings 0x%016llx rings 0x%016llx",
3349 	    rx_rbr_rings, rx_rcr_rings));
3350 
3351 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "<== hxge_rxdma_hw_stop"));
3352 }
3353 
3354 static hxge_status_t
3355 hxge_rxdma_start_channel(p_hxge_t hxgep, uint16_t channel,
3356     p_rx_rbr_ring_t rbr_p, p_rx_rcr_ring_t rcr_p, p_rx_mbox_t mbox_p,
3357     int n_init_kick)
3358 {
3359 	hpi_handle_t		handle;
3360 	hpi_status_t		rs = HPI_SUCCESS;
3361 	rdc_stat_t		cs;
3362 	rdc_int_mask_t		ent_mask;
3363 	hxge_status_t		status = HXGE_OK;
3364 
3365 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_start_channel"));
3366 
3367 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
3368 
3369 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "hxge_rxdma_start_channel: "
3370 	    "hpi handle addr $%p acc $%p",
3371 	    hxgep->hpi_handle.regp, hxgep->hpi_handle.regh));
3372 
3373 	/* Reset RXDMA channel */
3374 	rs = hpi_rxdma_cfg_rdc_reset(handle, channel);
3375 	if (rs != HPI_SUCCESS) {
3376 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3377 		    "==> hxge_rxdma_start_channel: "
3378 		    "reset rxdma failed (0x%08x channel %d)",
3379 		    status, channel));
3380 		return (HXGE_ERROR | rs);
3381 	}
3382 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3383 	    "==> hxge_rxdma_start_channel: reset done: channel %d", channel));
3384 
3385 	/*
3386 	 * Initialize the RXDMA channel specific FZC control configurations.
3387 	 * These FZC registers are pertaining to each RX channel (logical
3388 	 * pages).
3389 	 */
3390 	status = hxge_init_fzc_rxdma_channel(hxgep,
3391 	    channel, rbr_p, rcr_p, mbox_p);
3392 	if (status != HXGE_OK) {
3393 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3394 		    "==> hxge_rxdma_start_channel: "
3395 		    "init fzc rxdma failed (0x%08x channel %d)",
3396 		    status, channel));
3397 		return (status);
3398 	}
3399 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3400 	    "==> hxge_rxdma_start_channel: fzc done"));
3401 
3402 	/*
3403 	 * Zero out the shadow  and prefetch ram.
3404 	 */
3405 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3406 	    "==> hxge_rxdma_start_channel: ram done"));
3407 
3408 	/* Set up the interrupt event masks. */
3409 	ent_mask.value = 0;
3410 	rs = hpi_rxdma_event_mask(handle, OP_SET, channel, &ent_mask);
3411 	if (rs != HPI_SUCCESS) {
3412 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3413 		    "==> hxge_rxdma_start_channel: "
3414 		    "init rxdma event masks failed (0x%08x channel %d)",
3415 		    status, channel));
3416 		return (HXGE_ERROR | rs);
3417 	}
3418 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_start_channel: "
3419 	    "event done: channel %d (mask 0x%016llx)",
3420 	    channel, ent_mask.value));
3421 
3422 	/*
3423 	 * Load RXDMA descriptors, buffers, mailbox, initialise the receive DMA
3424 	 * channels and enable each DMA channel.
3425 	 */
3426 	status = hxge_enable_rxdma_channel(hxgep,
3427 	    channel, rbr_p, rcr_p, mbox_p, n_init_kick);
3428 	if (status != HXGE_OK) {
3429 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3430 		    " hxge_rxdma_start_channel: "
3431 		    " init enable rxdma failed (0x%08x channel %d)",
3432 		    status, channel));
3433 		return (status);
3434 	}
3435 
3436 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_start_channel: "
3437 	    "control done - channel %d cs 0x%016llx", channel, cs.value));
3438 
3439 	/*
3440 	 * Initialize the receive DMA control and status register
3441 	 * Note that rdc_stat HAS to be set after RBR and RCR rings are set
3442 	 */
3443 	cs.value = 0;
3444 	cs.bits.mex = 1;
3445 	cs.bits.rcr_thres = 1;
3446 	cs.bits.rcr_to = 1;
3447 	cs.bits.rbr_empty = 1;
3448 	status = hxge_init_rxdma_channel_cntl_stat(hxgep, channel, &cs);
3449 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_start_channel: "
3450 	    "channel %d rx_dma_cntl_stat 0x%0016llx", channel, cs.value));
3451 	if (status != HXGE_OK) {
3452 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3453 		    "==> hxge_rxdma_start_channel: "
3454 		    "init rxdma control register failed (0x%08x channel %d",
3455 		    status, channel));
3456 		return (status);
3457 	}
3458 
3459 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_start_channel: "
3460 	    "control done - channel %d cs 0x%016llx", channel, cs.value));
3461 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3462 	    "==> hxge_rxdma_start_channel: enable done"));
3463 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "<== hxge_rxdma_start_channel"));
3464 
3465 	return (HXGE_OK);
3466 }
3467 
3468 static hxge_status_t
3469 hxge_rxdma_stop_channel(p_hxge_t hxgep, uint16_t channel)
3470 {
3471 	hpi_handle_t		handle;
3472 	hpi_status_t		rs = HPI_SUCCESS;
3473 	rdc_stat_t		cs;
3474 	rdc_int_mask_t		ent_mask;
3475 	hxge_status_t		status = HXGE_OK;
3476 
3477 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rxdma_stop_channel"));
3478 
3479 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
3480 
3481 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "hxge_rxdma_stop_channel: "
3482 	    "hpi handle addr $%p acc $%p",
3483 	    hxgep->hpi_handle.regp, hxgep->hpi_handle.regh));
3484 
3485 	/* Reset RXDMA channel */
3486 	rs = hpi_rxdma_cfg_rdc_reset(handle, channel);
3487 	if (rs != HPI_SUCCESS) {
3488 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3489 		    " hxge_rxdma_stop_channel: "
3490 		    " reset rxdma failed (0x%08x channel %d)",
3491 		    rs, channel));
3492 		return (HXGE_ERROR | rs);
3493 	}
3494 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
3495 	    "==> hxge_rxdma_stop_channel: reset done"));
3496 
3497 	/* Set up the interrupt event masks. */
3498 	ent_mask.value = RDC_INT_MASK_ALL;
3499 	rs = hpi_rxdma_event_mask(handle, OP_SET, channel, &ent_mask);
3500 	if (rs != HPI_SUCCESS) {
3501 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3502 		    "==> hxge_rxdma_stop_channel: "
3503 		    "set rxdma event masks failed (0x%08x channel %d)",
3504 		    rs, channel));
3505 		return (HXGE_ERROR | rs);
3506 	}
3507 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
3508 	    "==> hxge_rxdma_stop_channel: event done"));
3509 
3510 	/* Initialize the receive DMA control and status register */
3511 	cs.value = 0;
3512 	status = hxge_init_rxdma_channel_cntl_stat(hxgep, channel, &cs);
3513 
3514 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rxdma_stop_channel: control "
3515 	    " to default (all 0s) 0x%08x", cs.value));
3516 
3517 	if (status != HXGE_OK) {
3518 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3519 		    " hxge_rxdma_stop_channel: init rxdma"
3520 		    " control register failed (0x%08x channel %d",
3521 		    status, channel));
3522 		return (status);
3523 	}
3524 
3525 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
3526 	    "==> hxge_rxdma_stop_channel: control done"));
3527 
3528 	/* disable dma channel */
3529 	status = hxge_disable_rxdma_channel(hxgep, channel);
3530 
3531 	if (status != HXGE_OK) {
3532 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3533 		    " hxge_rxdma_stop_channel: "
3534 		    " init enable rxdma failed (0x%08x channel %d)",
3535 		    status, channel));
3536 		return (status);
3537 	}
3538 
3539 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
3540 	    "==> hxge_rxdma_stop_channel: disable done"));
3541 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_rxdma_stop_channel"));
3542 
3543 	return (HXGE_OK);
3544 }
3545 
3546 hxge_status_t
3547 hxge_rxdma_handle_sys_errors(p_hxge_t hxgep)
3548 {
3549 	hpi_handle_t		handle;
3550 	p_hxge_rdc_sys_stats_t	statsp;
3551 	rdc_fifo_err_stat_t	stat;
3552 	hxge_status_t		status = HXGE_OK;
3553 
3554 	handle = hxgep->hpi_handle;
3555 	statsp = (p_hxge_rdc_sys_stats_t)&hxgep->statsp->rdc_sys_stats;
3556 
3557 	/* Clear the int_dbg register in case it is an injected err */
3558 	HXGE_REG_WR64(handle, RDC_FIFO_ERR_INT_DBG, 0x0);
3559 
3560 	/* Get the error status and clear the register */
3561 	HXGE_REG_RD64(handle, RDC_FIFO_ERR_STAT, &stat.value);
3562 	HXGE_REG_WR64(handle, RDC_FIFO_ERR_STAT, stat.value);
3563 
3564 	if (stat.bits.rx_ctrl_fifo_sec) {
3565 		statsp->ctrl_fifo_sec++;
3566 		if (statsp->ctrl_fifo_sec == 1)
3567 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3568 			    "==> hxge_rxdma_handle_sys_errors: "
3569 			    "rx_ctrl_fifo_sec"));
3570 	}
3571 
3572 	if (stat.bits.rx_ctrl_fifo_ded) {
3573 		/* Global fatal error encountered */
3574 		statsp->ctrl_fifo_ded++;
3575 		HXGE_FM_REPORT_ERROR(hxgep, NULL,
3576 		    HXGE_FM_EREPORT_RDMC_CTRL_FIFO_DED);
3577 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3578 		    "==> hxge_rxdma_handle_sys_errors: "
3579 		    "fatal error: rx_ctrl_fifo_ded error"));
3580 	}
3581 
3582 	if (stat.bits.rx_data_fifo_sec) {
3583 		statsp->data_fifo_sec++;
3584 		if (statsp->data_fifo_sec == 1)
3585 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3586 			    "==> hxge_rxdma_handle_sys_errors: "
3587 			    "rx_data_fifo_sec"));
3588 	}
3589 
3590 	if (stat.bits.rx_data_fifo_ded) {
3591 		/* Global fatal error encountered */
3592 		statsp->data_fifo_ded++;
3593 		HXGE_FM_REPORT_ERROR(hxgep, NULL,
3594 		    HXGE_FM_EREPORT_RDMC_DATA_FIFO_DED);
3595 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3596 		    "==> hxge_rxdma_handle_sys_errors: "
3597 		    "fatal error: rx_data_fifo_ded error"));
3598 	}
3599 
3600 	if (stat.bits.rx_ctrl_fifo_ded || stat.bits.rx_data_fifo_ded) {
3601 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3602 		    " hxge_rxdma_handle_sys_errors: fatal error\n"));
3603 		status = hxge_rx_port_fatal_err_recover(hxgep);
3604 		if (status == HXGE_OK) {
3605 			FM_SERVICE_RESTORED(hxgep);
3606 		}
3607 	}
3608 
3609 	return (HXGE_OK);
3610 }
3611 
3612 static hxge_status_t
3613 hxge_rxdma_fatal_err_recover(p_hxge_t hxgep, uint16_t channel)
3614 {
3615 	hpi_handle_t		handle;
3616 	hpi_status_t 		rs = HPI_SUCCESS;
3617 	hxge_status_t 		status = HXGE_OK;
3618 	p_rx_rbr_ring_t		rbrp;
3619 	p_rx_rcr_ring_t		rcrp;
3620 	p_rx_mbox_t		mboxp;
3621 	rdc_int_mask_t		ent_mask;
3622 	p_hxge_dma_common_t	dmap;
3623 	int			ring_idx;
3624 	p_rx_msg_t		rx_msg_p;
3625 	int			i;
3626 	uint32_t		hxge_port_rcr_size;
3627 	uint64_t		tmp;
3628 	int			n_init_kick = 0;
3629 
3630 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rxdma_fatal_err_recover"));
3631 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3632 	    "Recovering from RxDMAChannel#%d error...", channel));
3633 
3634 	/*
3635 	 * Stop the dma channel waits for the stop done. If the stop done bit
3636 	 * is not set, then create an error.
3637 	 */
3638 
3639 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
3640 
3641 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Rx DMA stop..."));
3642 
3643 	ring_idx = hxge_rxdma_get_ring_index(hxgep, channel);
3644 	rbrp = (p_rx_rbr_ring_t)hxgep->rx_rbr_rings->rbr_rings[ring_idx];
3645 	rcrp = (p_rx_rcr_ring_t)hxgep->rx_rcr_rings->rcr_rings[ring_idx];
3646 
3647 	MUTEX_ENTER(&rcrp->lock);
3648 	MUTEX_ENTER(&rbrp->lock);
3649 
3650 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Disable RxDMA channel..."));
3651 
3652 	rs = hpi_rxdma_cfg_rdc_disable(handle, channel);
3653 	if (rs != HPI_SUCCESS) {
3654 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3655 		    "hxge_disable_rxdma_channel:failed"));
3656 		goto fail;
3657 	}
3658 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Disable RxDMA interrupt..."));
3659 
3660 	/* Disable interrupt */
3661 	ent_mask.value = RDC_INT_MASK_ALL;
3662 	rs = hpi_rxdma_event_mask(handle, OP_SET, channel, &ent_mask);
3663 	if (rs != HPI_SUCCESS) {
3664 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3665 		    "Set rxdma event masks failed (channel %d)", channel));
3666 	}
3667 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "RxDMA channel reset..."));
3668 
3669 	/* Reset RXDMA channel */
3670 	rs = hpi_rxdma_cfg_rdc_reset(handle, channel);
3671 	if (rs != HPI_SUCCESS) {
3672 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3673 		    "Reset rxdma failed (channel %d)", channel));
3674 		goto fail;
3675 	}
3676 	hxge_port_rcr_size = hxgep->hxge_port_rcr_size;
3677 	mboxp = (p_rx_mbox_t)hxgep->rx_mbox_areas_p->rxmbox_areas[ring_idx];
3678 
3679 	rbrp->rbr_wr_index = (rbrp->rbb_max - 1);
3680 	rbrp->rbr_rd_index = 0;
3681 
3682 	rcrp->comp_rd_index = 0;
3683 	rcrp->comp_wt_index = 0;
3684 	rcrp->rcr_desc_rd_head_p = rcrp->rcr_desc_first_p =
3685 	    (p_rcr_entry_t)DMA_COMMON_VPTR(rcrp->rcr_desc);
3686 #if defined(__i386)
3687 	rcrp->rcr_desc_rd_head_pp = rcrp->rcr_desc_first_pp =
3688 	    (p_rcr_entry_t)(uint32_t)DMA_COMMON_IOADDR(rcrp->rcr_desc);
3689 #else
3690 	rcrp->rcr_desc_rd_head_pp = rcrp->rcr_desc_first_pp =
3691 	    (p_rcr_entry_t)DMA_COMMON_IOADDR(rcrp->rcr_desc);
3692 #endif
3693 
3694 	rcrp->rcr_desc_last_p = rcrp->rcr_desc_rd_head_p +
3695 	    (hxge_port_rcr_size - 1);
3696 	rcrp->rcr_desc_last_pp = rcrp->rcr_desc_rd_head_pp +
3697 	    (hxge_port_rcr_size - 1);
3698 
3699 	rcrp->rcr_tail_begin = DMA_COMMON_IOADDR(rcrp->rcr_desc);
3700 	rcrp->rcr_tail_begin = (rcrp->rcr_tail_begin & 0x7ffffULL) >> 3;
3701 
3702 	dmap = (p_hxge_dma_common_t)&rcrp->rcr_desc;
3703 	bzero((caddr_t)dmap->kaddrp, dmap->alength);
3704 
3705 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "rbr entries = %d\n",
3706 	    rbrp->rbr_max_size));
3707 
3708 	/* Count the number of buffers owned by the hardware at this moment */
3709 	for (i = 0; i < rbrp->rbr_max_size; i++) {
3710 		rx_msg_p = rbrp->rx_msg_ring[i];
3711 		if (rx_msg_p->ref_cnt == 1) {
3712 			n_init_kick++;
3713 		}
3714 	}
3715 
3716 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "RxDMA channel re-start..."));
3717 
3718 	/*
3719 	 * This is error recover! Some buffers are owned by the hardware and
3720 	 * the rest are owned by the apps. We should only kick in those
3721 	 * owned by the hardware initially. The apps will post theirs
3722 	 * eventually.
3723 	 */
3724 	status = hxge_rxdma_start_channel(hxgep, channel, rbrp, rcrp, mboxp,
3725 	    n_init_kick);
3726 	if (status != HXGE_OK) {
3727 		goto fail;
3728 	}
3729 
3730 	/*
3731 	 * The DMA channel may disable itself automatically.
3732 	 * The following is a work-around.
3733 	 */
3734 	HXGE_REG_RD64(handle, RDC_RX_CFG1, &tmp);
3735 	rs = hpi_rxdma_cfg_rdc_enable(handle, channel);
3736 	if (rs != HPI_SUCCESS) {
3737 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3738 		    "hpi_rxdma_cfg_rdc_enable (channel %d)", channel));
3739 	}
3740 
3741 	MUTEX_EXIT(&rbrp->lock);
3742 	MUTEX_EXIT(&rcrp->lock);
3743 
3744 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3745 	    "Recovery Successful, RxDMAChannel#%d Restored", channel));
3746 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_rxdma_fatal_err_recover"));
3747 
3748 	return (HXGE_OK);
3749 
3750 fail:
3751 	MUTEX_EXIT(&rbrp->lock);
3752 	MUTEX_EXIT(&rcrp->lock);
3753 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL, "Recovery failed"));
3754 
3755 	return (HXGE_ERROR | rs);
3756 }
3757 
3758 static hxge_status_t
3759 hxge_rx_port_fatal_err_recover(p_hxge_t hxgep)
3760 {
3761 	hxge_status_t		status = HXGE_OK;
3762 	p_hxge_dma_common_t	*dma_buf_p;
3763 	uint16_t		channel;
3764 	int			ndmas;
3765 	int			i;
3766 	block_reset_t		reset_reg;
3767 	p_rx_rcr_ring_t	rcrp;
3768 	p_rx_rbr_ring_t rbrp;
3769 
3770 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rx_port_fatal_err_recover"));
3771 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL, "Recovering from RDC error ..."));
3772 
3773 	/* Reset RDC block from PEU for this fatal error */
3774 	reset_reg.value = 0;
3775 	reset_reg.bits.rdc_rst = 1;
3776 	HXGE_REG_WR32(hxgep->hpi_handle, BLOCK_RESET, reset_reg.value);
3777 
3778 	/* Disable RxMAC */
3779 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Disable RxMAC...\n"));
3780 	if (hxge_rx_vmac_disable(hxgep) != HXGE_OK)
3781 		goto fail;
3782 
3783 	HXGE_DELAY(1000);
3784 
3785 	/* Restore any common settings after PEU reset */
3786 	if (hxge_rxdma_hw_start_common(hxgep) != HXGE_OK)
3787 		goto fail;
3788 
3789 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Stop all RxDMA channels..."));
3790 
3791 	ndmas = hxgep->rx_buf_pool_p->ndmas;
3792 	dma_buf_p = hxgep->rx_buf_pool_p->dma_buf_pool_p;
3793 
3794 	for (i = 0; i < ndmas; i++) {
3795 		channel = ((p_hxge_dma_common_t)dma_buf_p[i])->dma_channel;
3796 		rcrp = hxgep->rx_rcr_rings->rcr_rings[channel];
3797 		rbrp = rcrp->rx_rbr_p;
3798 
3799 		MUTEX_ENTER(&rbrp->post_lock);
3800 		/* This function needs to be inside the post_lock */
3801 		if (hxge_rxdma_fatal_err_recover(hxgep, channel) != HXGE_OK) {
3802 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3803 			    "Could not recover channel %d", channel));
3804 		}
3805 		MUTEX_EXIT(&rbrp->post_lock);
3806 	}
3807 
3808 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Reset RxMAC..."));
3809 
3810 	/* Reset RxMAC */
3811 	if (hxge_rx_vmac_reset(hxgep) != HXGE_OK) {
3812 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3813 		    "hxge_rx_port_fatal_err_recover: Failed to reset RxMAC"));
3814 		goto fail;
3815 	}
3816 
3817 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Re-initialize RxMAC..."));
3818 
3819 	/* Re-Initialize RxMAC */
3820 	if ((status = hxge_rx_vmac_init(hxgep)) != HXGE_OK) {
3821 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3822 		    "hxge_rx_port_fatal_err_recover: Failed to reset RxMAC"));
3823 		goto fail;
3824 	}
3825 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Re-enable RxMAC..."));
3826 
3827 	/* Re-enable RxMAC */
3828 	if ((status = hxge_rx_vmac_enable(hxgep)) != HXGE_OK) {
3829 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3830 		    "hxge_rx_port_fatal_err_recover: Failed to enable RxMAC"));
3831 		goto fail;
3832 	}
3833 
3834 	/* Reset the error mask since PEU reset cleared it */
3835 	HXGE_REG_WR64(hxgep->hpi_handle, RDC_FIFO_ERR_INT_MASK, 0x0);
3836 
3837 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3838 	    "Recovery Successful, RxPort Restored"));
3839 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_rx_port_fatal_err_recover"));
3840 
3841 	return (HXGE_OK);
3842 fail:
3843 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL, "Recovery failed"));
3844 	return (status);
3845 }
3846 
3847 static void
3848 hxge_rbr_empty_restore(p_hxge_t hxgep, p_rx_rbr_ring_t rx_rbr_p)
3849 {
3850 	hpi_status_t		hpi_status;
3851 	hxge_status_t		status;
3852 	int			i;
3853 	p_hxge_rx_ring_stats_t	rdc_stats;
3854 
3855 	rdc_stats = &hxgep->statsp->rdc_stats[rx_rbr_p->rdc];
3856 	rdc_stats->rbr_empty_restore++;
3857 	rx_rbr_p->rbr_is_empty = B_FALSE;
3858 
3859 	/*
3860 	 * Complete the processing for the RBR Empty by:
3861 	 *	0) kicking back HXGE_RBR_EMPTY_THRESHOLD
3862 	 *	   packets.
3863 	 *	1) Disable the RX vmac.
3864 	 *	2) Re-enable the affected DMA channel.
3865 	 *	3) Re-enable the RX vmac.
3866 	 */
3867 
3868 	/*
3869 	 * Disable the RX VMAC, but setting the framelength
3870 	 * to 0, since there is a hardware bug when disabling
3871 	 * the vmac.
3872 	 */
3873 	MUTEX_ENTER(hxgep->genlock);
3874 	(void) hpi_vmac_rx_set_framesize(
3875 	    HXGE_DEV_HPI_HANDLE(hxgep), (uint16_t)0);
3876 
3877 	hpi_status = hpi_rxdma_cfg_rdc_enable(
3878 	    HXGE_DEV_HPI_HANDLE(hxgep), rx_rbr_p->rdc);
3879 	if (hpi_status != HPI_SUCCESS) {
3880 		rdc_stats->rbr_empty_fail++;
3881 
3882 		/* Assume we are already inside the post_lock */
3883 		status = hxge_rxdma_fatal_err_recover(hxgep, rx_rbr_p->rdc);
3884 		if (status != HXGE_OK) {
3885 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3886 			    "hxge(%d): channel(%d) is empty.",
3887 			    hxgep->instance, rx_rbr_p->rdc));
3888 		}
3889 	}
3890 
3891 	for (i = 0; i < 1024; i++) {
3892 		uint64_t value;
3893 		RXDMA_REG_READ64(HXGE_DEV_HPI_HANDLE(hxgep),
3894 		    RDC_STAT, i & 3, &value);
3895 	}
3896 
3897 	/*
3898 	 * Re-enable the RX VMAC.
3899 	 */
3900 	(void) hpi_vmac_rx_set_framesize(HXGE_DEV_HPI_HANDLE(hxgep),
3901 	    (uint16_t)hxgep->vmac.maxframesize);
3902 	MUTEX_EXIT(hxgep->genlock);
3903 }
3904