xref: /illumos-gate/usr/src/uts/common/io/hxge/hxge_rxdma.c (revision ed093b41a93e8563e6e1e5dae0768dda2a7bcc27)
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 2010 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 #include <hpi.h>
29 #include <hpi_vir.h>
30 
31 /*
32  * Number of blocks to accumulate before re-enabling DMA
33  * when we get RBR empty.
34  */
35 #define	HXGE_RBR_EMPTY_THRESHOLD	64
36 
37 /*
38  * Globals: tunable parameters (/etc/system or adb)
39  *
40  */
41 extern uint32_t hxge_rbr_size;
42 extern uint32_t hxge_rcr_size;
43 extern uint32_t hxge_rbr_spare_size;
44 extern uint32_t hxge_mblks_pending;
45 
46 /*
47  * Tunables to manage the receive buffer blocks.
48  *
49  * hxge_rx_threshold_hi: copy all buffers.
50  * hxge_rx_bcopy_size_type: receive buffer block size type.
51  * hxge_rx_threshold_lo: copy only up to tunable block size type.
52  */
53 extern hxge_rxbuf_threshold_t hxge_rx_threshold_hi;
54 extern hxge_rxbuf_type_t hxge_rx_buf_size_type;
55 extern hxge_rxbuf_threshold_t hxge_rx_threshold_lo;
56 
57 /*
58  * Static local functions.
59  */
60 static hxge_status_t hxge_map_rxdma(p_hxge_t hxgep);
61 static void hxge_unmap_rxdma(p_hxge_t hxgep);
62 static hxge_status_t hxge_rxdma_hw_start_common(p_hxge_t hxgep);
63 static hxge_status_t hxge_rxdma_hw_start(p_hxge_t hxgep);
64 static void hxge_rxdma_hw_stop(p_hxge_t hxgep);
65 static hxge_status_t hxge_map_rxdma_channel(p_hxge_t hxgep, uint16_t channel,
66     p_hxge_dma_common_t *dma_buf_p, p_rx_rbr_ring_t *rbr_p,
67     uint32_t num_chunks, p_hxge_dma_common_t *dma_rbr_cntl_p,
68     p_hxge_dma_common_t *dma_rcr_cntl_p, p_hxge_dma_common_t *dma_mbox_cntl_p,
69     p_rx_rcr_ring_t *rcr_p, p_rx_mbox_t *rx_mbox_p);
70 static void hxge_unmap_rxdma_channel(p_hxge_t hxgep, uint16_t channel,
71 	p_rx_rbr_ring_t rbr_p, p_rx_rcr_ring_t rcr_p, p_rx_mbox_t rx_mbox_p);
72 static hxge_status_t hxge_map_rxdma_channel_cfg_ring(p_hxge_t hxgep,
73     uint16_t dma_channel, p_hxge_dma_common_t *dma_rbr_cntl_p,
74     p_hxge_dma_common_t *dma_rcr_cntl_p, p_hxge_dma_common_t *dma_mbox_cntl_p,
75     p_rx_rbr_ring_t *rbr_p, p_rx_rcr_ring_t *rcr_p, p_rx_mbox_t *rx_mbox_p);
76 static void hxge_unmap_rxdma_channel_cfg_ring(p_hxge_t hxgep,
77 	p_rx_rcr_ring_t rcr_p, p_rx_mbox_t rx_mbox_p);
78 static hxge_status_t hxge_map_rxdma_channel_buf_ring(p_hxge_t hxgep,
79 	uint16_t channel, p_hxge_dma_common_t *dma_buf_p,
80 	p_rx_rbr_ring_t *rbr_p, uint32_t num_chunks);
81 static void hxge_unmap_rxdma_channel_buf_ring(p_hxge_t hxgep,
82 	p_rx_rbr_ring_t rbr_p);
83 static hxge_status_t hxge_rxdma_start_channel(p_hxge_t hxgep, uint16_t channel,
84 	p_rx_rbr_ring_t rbr_p, p_rx_rcr_ring_t rcr_p, p_rx_mbox_t mbox_p,
85 	int n_init_kick);
86 static hxge_status_t hxge_rxdma_stop_channel(p_hxge_t hxgep, uint16_t channel);
87 static mblk_t *hxge_rx_pkts(p_hxge_t hxgep, uint_t vindex, p_hxge_ldv_t ldvp,
88 	p_rx_rcr_ring_t	rcr_p, rdc_stat_t cs, int bytes_to_read);
89 static uint32_t hxge_scan_for_last_eop(p_rx_rcr_ring_t rcr_p,
90     p_rcr_entry_t rcr_desc_rd_head_p, uint32_t num_rcrs);
91 static void hxge_receive_packet(p_hxge_t hxgep, p_rx_rcr_ring_t rcr_p,
92 	p_rcr_entry_t rcr_desc_rd_head_p, boolean_t *multi_p,
93 	mblk_t ** mp, mblk_t ** mp_cont, uint32_t *invalid_rcr_entry);
94 static hxge_status_t hxge_disable_rxdma_channel(p_hxge_t hxgep,
95 	uint16_t channel);
96 static p_rx_msg_t hxge_allocb(size_t, uint32_t, p_hxge_dma_common_t);
97 static void hxge_freeb(p_rx_msg_t);
98 static hxge_status_t hxge_rx_err_evnts(p_hxge_t hxgep, uint_t index,
99 	p_hxge_ldv_t ldvp, rdc_stat_t cs);
100 static hxge_status_t hxge_rxbuf_index_info_init(p_hxge_t hxgep,
101 	p_rx_rbr_ring_t rx_dmap);
102 static hxge_status_t hxge_rxdma_fatal_err_recover(p_hxge_t hxgep,
103 	uint16_t channel);
104 static hxge_status_t hxge_rx_port_fatal_err_recover(p_hxge_t hxgep);
105 static void hxge_rbr_empty_restore(p_hxge_t hxgep,
106 	p_rx_rbr_ring_t rx_rbr_p);
107 
108 hxge_status_t
109 hxge_init_rxdma_channels(p_hxge_t hxgep)
110 {
111 	hxge_status_t		status = HXGE_OK;
112 	block_reset_t		reset_reg;
113 	int			i;
114 
115 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_init_rxdma_channels"));
116 
117 	for (i = 0; i < HXGE_MAX_RDCS; i++)
118 		hxgep->rdc_first_intr[i] = B_TRUE;
119 
120 	/* Reset RDC block from PEU to clear any previous state */
121 	reset_reg.value = 0;
122 	reset_reg.bits.rdc_rst = 1;
123 	HXGE_REG_WR32(hxgep->hpi_handle, BLOCK_RESET, reset_reg.value);
124 	HXGE_DELAY(1000);
125 
126 	status = hxge_map_rxdma(hxgep);
127 	if (status != HXGE_OK) {
128 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
129 		    "<== hxge_init_rxdma: status 0x%x", status));
130 		return (status);
131 	}
132 
133 	status = hxge_rxdma_hw_start_common(hxgep);
134 	if (status != HXGE_OK) {
135 		hxge_unmap_rxdma(hxgep);
136 	}
137 
138 	status = hxge_rxdma_hw_start(hxgep);
139 	if (status != HXGE_OK) {
140 		hxge_unmap_rxdma(hxgep);
141 	}
142 
143 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
144 	    "<== hxge_init_rxdma_channels: status 0x%x", status));
145 	return (status);
146 }
147 
148 void
149 hxge_uninit_rxdma_channels(p_hxge_t hxgep)
150 {
151 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_uninit_rxdma_channels"));
152 
153 	hxge_rxdma_hw_stop(hxgep);
154 	hxge_unmap_rxdma(hxgep);
155 
156 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "<== hxge_uinit_rxdma_channels"));
157 }
158 
159 hxge_status_t
160 hxge_init_rxdma_channel_cntl_stat(p_hxge_t hxgep, uint16_t channel,
161     rdc_stat_t *cs_p)
162 {
163 	hpi_handle_t	handle;
164 	hpi_status_t	rs = HPI_SUCCESS;
165 	hxge_status_t	status = HXGE_OK;
166 
167 	HXGE_DEBUG_MSG((hxgep, DMA_CTL,
168 	    "<== hxge_init_rxdma_channel_cntl_stat"));
169 
170 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
171 	rs = hpi_rxdma_control_status(handle, OP_SET, channel, cs_p);
172 
173 	if (rs != HPI_SUCCESS) {
174 		status = HXGE_ERROR | rs;
175 	}
176 	return (status);
177 }
178 
179 
180 hxge_status_t
181 hxge_enable_rxdma_channel(p_hxge_t hxgep, uint16_t channel,
182     p_rx_rbr_ring_t rbr_p, p_rx_rcr_ring_t rcr_p, p_rx_mbox_t mbox_p,
183     int n_init_kick)
184 {
185 	hpi_handle_t		handle;
186 	rdc_desc_cfg_t		rdc_desc;
187 	rdc_rcr_cfg_b_t		*cfgb_p;
188 	hpi_status_t		rs = HPI_SUCCESS;
189 
190 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "==> hxge_enable_rxdma_channel"));
191 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
192 
193 	/*
194 	 * Use configuration data composed at init time. Write to hardware the
195 	 * receive ring configurations.
196 	 */
197 	rdc_desc.mbox_enable = 1;
198 	rdc_desc.mbox_addr = mbox_p->mbox_addr;
199 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
200 	    "==> hxge_enable_rxdma_channel: mboxp $%p($%p)",
201 	    mbox_p->mbox_addr, rdc_desc.mbox_addr));
202 
203 	rdc_desc.rbr_len = rbr_p->rbb_max;
204 	rdc_desc.rbr_addr = rbr_p->rbr_addr;
205 
206 	switch (hxgep->rx_bksize_code) {
207 	case RBR_BKSIZE_4K:
208 		rdc_desc.page_size = SIZE_4KB;
209 		break;
210 	case RBR_BKSIZE_8K:
211 		rdc_desc.page_size = SIZE_8KB;
212 		break;
213 	}
214 
215 	rdc_desc.size0 = rbr_p->hpi_pkt_buf_size0;
216 	rdc_desc.valid0 = 1;
217 
218 	rdc_desc.size1 = rbr_p->hpi_pkt_buf_size1;
219 	rdc_desc.valid1 = 1;
220 
221 	rdc_desc.size2 = rbr_p->hpi_pkt_buf_size2;
222 	rdc_desc.valid2 = 1;
223 
224 	rdc_desc.full_hdr = rcr_p->full_hdr_flag;
225 	rdc_desc.offset = rcr_p->sw_priv_hdr_len;
226 
227 	rdc_desc.rcr_len = rcr_p->comp_size;
228 	rdc_desc.rcr_addr = rcr_p->rcr_addr;
229 
230 	cfgb_p = &(rcr_p->rcr_cfgb);
231 	rdc_desc.rcr_threshold = cfgb_p->bits.pthres;
232 	rdc_desc.rcr_timeout = cfgb_p->bits.timeout;
233 	rdc_desc.rcr_timeout_enable = cfgb_p->bits.entout;
234 
235 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "==> hxge_enable_rxdma_channel: "
236 	    "rbr_len qlen %d pagesize code %d rcr_len %d",
237 	    rdc_desc.rbr_len, rdc_desc.page_size, rdc_desc.rcr_len));
238 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "==> hxge_enable_rxdma_channel: "
239 	    "size 0 %d size 1 %d size 2 %d",
240 	    rbr_p->hpi_pkt_buf_size0, rbr_p->hpi_pkt_buf_size1,
241 	    rbr_p->hpi_pkt_buf_size2));
242 
243 	rs = hpi_rxdma_cfg_rdc_ring(handle, rbr_p->rdc, &rdc_desc);
244 	if (rs != HPI_SUCCESS) {
245 		return (HXGE_ERROR | rs);
246 	}
247 
248 	/*
249 	 * Enable the timeout and threshold.
250 	 */
251 	rs = hpi_rxdma_cfg_rdc_rcr_threshold(handle, channel,
252 	    rdc_desc.rcr_threshold);
253 	if (rs != HPI_SUCCESS) {
254 		return (HXGE_ERROR | rs);
255 	}
256 
257 	rs = hpi_rxdma_cfg_rdc_rcr_timeout(handle, channel,
258 	    rdc_desc.rcr_timeout);
259 	if (rs != HPI_SUCCESS) {
260 		return (HXGE_ERROR | rs);
261 	}
262 
263 	/* Kick the DMA engine */
264 	hpi_rxdma_rdc_rbr_kick(handle, channel, n_init_kick);
265 
266 	/* Clear the rbr empty bit */
267 	(void) hpi_rxdma_channel_rbr_empty_clear(handle, channel);
268 
269 	/*
270 	 * Enable the DMA
271 	 */
272 	rs = hpi_rxdma_cfg_rdc_enable(handle, channel);
273 	if (rs != HPI_SUCCESS) {
274 		return (HXGE_ERROR | rs);
275 	}
276 
277 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "<== hxge_enable_rxdma_channel"));
278 
279 	return (HXGE_OK);
280 }
281 
282 static hxge_status_t
283 hxge_disable_rxdma_channel(p_hxge_t hxgep, uint16_t channel)
284 {
285 	hpi_handle_t handle;
286 	hpi_status_t rs = HPI_SUCCESS;
287 
288 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "==> hxge_disable_rxdma_channel"));
289 
290 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
291 
292 	/* disable the DMA */
293 	rs = hpi_rxdma_cfg_rdc_disable(handle, channel);
294 	if (rs != HPI_SUCCESS) {
295 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
296 		    "<== hxge_disable_rxdma_channel:failed (0x%x)", rs));
297 		return (HXGE_ERROR | rs);
298 	}
299 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "<== hxge_disable_rxdma_channel"));
300 	return (HXGE_OK);
301 }
302 
303 hxge_status_t
304 hxge_rxdma_channel_rcrflush(p_hxge_t hxgep, uint8_t channel)
305 {
306 	hpi_handle_t	handle;
307 	hxge_status_t	status = HXGE_OK;
308 
309 	HXGE_DEBUG_MSG((hxgep, DMA_CTL,
310 	    "==> hxge_rxdma_channel_rcrflush"));
311 
312 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
313 	hpi_rxdma_rdc_rcr_flush(handle, channel);
314 
315 	HXGE_DEBUG_MSG((hxgep, DMA_CTL,
316 	    "<== hxge_rxdma_channel_rcrflush"));
317 	return (status);
318 
319 }
320 
321 #define	MID_INDEX(l, r) ((r + l + 1) >> 1)
322 
323 #define	TO_LEFT -1
324 #define	TO_RIGHT 1
325 #define	BOTH_RIGHT (TO_RIGHT + TO_RIGHT)
326 #define	BOTH_LEFT (TO_LEFT + TO_LEFT)
327 #define	IN_MIDDLE (TO_RIGHT + TO_LEFT)
328 #define	NO_HINT 0xffffffff
329 
330 /*ARGSUSED*/
331 hxge_status_t
332 hxge_rxbuf_pp_to_vp(p_hxge_t hxgep, p_rx_rbr_ring_t rbr_p,
333     uint8_t pktbufsz_type, uint64_t *pkt_buf_addr_pp,
334     uint64_t **pkt_buf_addr_p, uint32_t *bufoffset, uint32_t *msg_index)
335 {
336 	int			bufsize;
337 	uint64_t		pktbuf_pp;
338 	uint64_t		dvma_addr;
339 	rxring_info_t		*ring_info;
340 	int			base_side, end_side;
341 	int			r_index, l_index, anchor_index;
342 	int			found, search_done;
343 	uint32_t		offset, chunk_size, block_size, page_size_mask;
344 	uint32_t		chunk_index, block_index, total_index;
345 	int			max_iterations, iteration;
346 	rxbuf_index_info_t	*bufinfo;
347 
348 	HXGE_DEBUG_MSG((hxgep, RX2_CTL, "==> hxge_rxbuf_pp_to_vp"));
349 
350 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
351 	    "==> hxge_rxbuf_pp_to_vp: buf_pp $%p btype %d",
352 	    pkt_buf_addr_pp, pktbufsz_type));
353 
354 	pktbuf_pp = (uint64_t)pkt_buf_addr_pp;
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 	*pkt_buf_addr_p = (uint64_t *)((uint64_t)bufinfo[anchor_index].kaddr +
548 	    offset);
549 
550 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
551 	    "==> hxge_rxbuf_pp_to_vp: "
552 	    "total_index %d dvma_addr $%p "
553 	    "offset %d block_size %d "
554 	    "block_index %d "
555 	    "*pkt_buf_addr_p $%p",
556 	    total_index, dvma_addr, offset, block_size,
557 	    block_index, *pkt_buf_addr_p));
558 
559 	*msg_index = total_index;
560 	*bufoffset = (offset & page_size_mask);
561 
562 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
563 	    "==> hxge_rxbuf_pp_to_vp: get msg index: "
564 	    "msg_index %d bufoffset_index %d",
565 	    *msg_index, *bufoffset));
566 	HXGE_DEBUG_MSG((hxgep, RX2_CTL, "<== hxge_rxbuf_pp_to_vp"));
567 
568 	return (HXGE_OK);
569 }
570 
571 
572 /*
573  * used by quick sort (qsort) function
574  * to perform comparison
575  */
576 static int
577 hxge_sort_compare(const void *p1, const void *p2)
578 {
579 
580 	rxbuf_index_info_t *a, *b;
581 
582 	a = (rxbuf_index_info_t *)p1;
583 	b = (rxbuf_index_info_t *)p2;
584 
585 	if (a->dvma_addr > b->dvma_addr)
586 		return (1);
587 	if (a->dvma_addr < b->dvma_addr)
588 		return (-1);
589 	return (0);
590 }
591 
592 /*
593  * Grabbed this sort implementation from common/syscall/avl.c
594  *
595  * Generic shellsort, from K&R (1st ed, p 58.), somewhat modified.
596  * v = Ptr to array/vector of objs
597  * n = # objs in the array
598  * s = size of each obj (must be multiples of a word size)
599  * f = ptr to function to compare two objs
600  *	returns (-1 = less than, 0 = equal, 1 = greater than
601  */
602 void
603 hxge_ksort(caddr_t v, int n, int s, int (*f) ())
604 {
605 	int		g, i, j, ii;
606 	unsigned int	*p1, *p2;
607 	unsigned int	tmp;
608 
609 	/* No work to do */
610 	if (v == NULL || n <= 1)
611 		return;
612 	/* Sanity check on arguments */
613 	ASSERT(((uintptr_t)v & 0x3) == 0 && (s & 0x3) == 0);
614 	ASSERT(s > 0);
615 
616 	for (g = n / 2; g > 0; g /= 2) {
617 		for (i = g; i < n; i++) {
618 			for (j = i - g; j >= 0 &&
619 			    (*f) (v + j * s, v + (j + g) * s) == 1; j -= g) {
620 				p1 = (unsigned *)(v + j * s);
621 				p2 = (unsigned *)(v + (j + g) * s);
622 				for (ii = 0; ii < s / 4; ii++) {
623 					tmp = *p1;
624 					*p1++ = *p2;
625 					*p2++ = tmp;
626 				}
627 			}
628 		}
629 	}
630 }
631 
632 /*
633  * Initialize data structures required for rxdma
634  * buffer dvma->vmem address lookup
635  */
636 /*ARGSUSED*/
637 static hxge_status_t
638 hxge_rxbuf_index_info_init(p_hxge_t hxgep, p_rx_rbr_ring_t rbrp)
639 {
640 	int		index;
641 	rxring_info_t	*ring_info;
642 	int		max_iteration = 0, max_index = 0;
643 
644 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "==> hxge_rxbuf_index_info_init"));
645 
646 	ring_info = rbrp->ring_info;
647 	ring_info->hint[0] = NO_HINT;
648 	ring_info->hint[1] = NO_HINT;
649 	ring_info->hint[2] = NO_HINT;
650 	ring_info->hint[3] = NO_HINT;
651 	max_index = rbrp->num_blocks;
652 
653 	/* read the DVMA address information and sort it */
654 	/* do init of the information array */
655 
656 	HXGE_DEBUG_MSG((hxgep, DMA2_CTL,
657 	    " hxge_rxbuf_index_info_init Sort ptrs"));
658 
659 	/* sort the array */
660 	hxge_ksort((void *) ring_info->buffer, max_index,
661 	    sizeof (rxbuf_index_info_t), hxge_sort_compare);
662 
663 	for (index = 0; index < max_index; index++) {
664 		HXGE_DEBUG_MSG((hxgep, DMA2_CTL,
665 		    " hxge_rxbuf_index_info_init: sorted chunk %d "
666 		    " ioaddr $%p kaddr $%p size %x",
667 		    index, ring_info->buffer[index].dvma_addr,
668 		    ring_info->buffer[index].kaddr,
669 		    ring_info->buffer[index].buf_size));
670 	}
671 
672 	max_iteration = 0;
673 	while (max_index >= (1ULL << max_iteration))
674 		max_iteration++;
675 	ring_info->max_iterations = max_iteration + 1;
676 
677 	HXGE_DEBUG_MSG((hxgep, DMA2_CTL,
678 	    " hxge_rxbuf_index_info_init Find max iter %d",
679 	    ring_info->max_iterations));
680 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "<== hxge_rxbuf_index_info_init"));
681 
682 	return (HXGE_OK);
683 }
684 
685 /*ARGSUSED*/
686 void
687 hxge_dump_rcr_entry(p_hxge_t hxgep, p_rcr_entry_t entry_p)
688 {
689 #ifdef	HXGE_DEBUG
690 
691 	uint32_t bptr;
692 	uint64_t pp;
693 
694 	bptr = entry_p->bits.pkt_buf_addr;
695 
696 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
697 	    "\trcr entry $%p "
698 	    "\trcr entry 0x%0llx "
699 	    "\trcr entry 0x%08x "
700 	    "\trcr entry 0x%08x "
701 	    "\tvalue 0x%0llx\n"
702 	    "\tmulti = %d\n"
703 	    "\tpkt_type = 0x%x\n"
704 	    "\terror = 0x%04x\n"
705 	    "\tl2_len = %d\n"
706 	    "\tpktbufsize = %d\n"
707 	    "\tpkt_buf_addr = $%p\n"
708 	    "\tpkt_buf_addr (<< 6) = $%p\n",
709 	    entry_p,
710 	    *(int64_t *)entry_p,
711 	    *(int32_t *)entry_p,
712 	    *(int32_t *)((char *)entry_p + 32),
713 	    entry_p->value,
714 	    entry_p->bits.multi,
715 	    entry_p->bits.pkt_type,
716 	    entry_p->bits.error,
717 	    entry_p->bits.l2_len,
718 	    entry_p->bits.pktbufsz,
719 	    bptr,
720 	    entry_p->bits.pkt_buf_addr_l));
721 
722 	pp = (entry_p->value & RCR_PKT_BUF_ADDR_MASK) <<
723 	    RCR_PKT_BUF_ADDR_SHIFT;
724 
725 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "rcr pp 0x%llx l2 len %d",
726 	    pp, (*(int64_t *)entry_p >> 40) & 0x3fff));
727 #endif
728 }
729 
730 /*ARGSUSED*/
731 void
732 hxge_rxdma_stop(p_hxge_t hxgep)
733 {
734 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rxdma_stop"));
735 
736 	MUTEX_ENTER(&hxgep->vmac_lock);
737 	(void) hxge_rx_vmac_disable(hxgep);
738 	(void) hxge_rxdma_hw_mode(hxgep, HXGE_DMA_STOP);
739 	MUTEX_EXIT(&hxgep->vmac_lock);
740 
741 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_rxdma_stop"));
742 }
743 
744 void
745 hxge_rxdma_stop_reinit(p_hxge_t hxgep)
746 {
747 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rxdma_stop_reinit"));
748 
749 	(void) hxge_rxdma_stop(hxgep);
750 	(void) hxge_uninit_rxdma_channels(hxgep);
751 	(void) hxge_init_rxdma_channels(hxgep);
752 
753 	MUTEX_ENTER(&hxgep->vmac_lock);
754 	(void) hxge_rx_vmac_enable(hxgep);
755 	MUTEX_EXIT(&hxgep->vmac_lock);
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 /*
833  * Static functions start here.
834  */
835 static p_rx_msg_t
836 hxge_allocb(size_t size, uint32_t pri, p_hxge_dma_common_t dmabuf_p)
837 {
838 	p_rx_msg_t		hxge_mp = NULL;
839 	p_hxge_dma_common_t	dmamsg_p;
840 	uchar_t			*buffer;
841 
842 	hxge_mp = KMEM_ZALLOC(sizeof (rx_msg_t), KM_NOSLEEP);
843 	if (hxge_mp == NULL) {
844 		HXGE_ERROR_MSG((NULL, HXGE_ERR_CTL,
845 		    "Allocation of a rx msg failed."));
846 		goto hxge_allocb_exit;
847 	}
848 
849 	hxge_mp->use_buf_pool = B_FALSE;
850 	if (dmabuf_p) {
851 		hxge_mp->use_buf_pool = B_TRUE;
852 
853 		dmamsg_p = (p_hxge_dma_common_t)&hxge_mp->buf_dma;
854 		*dmamsg_p = *dmabuf_p;
855 		dmamsg_p->nblocks = 1;
856 		dmamsg_p->block_size = size;
857 		dmamsg_p->alength = size;
858 		buffer = (uchar_t *)dmabuf_p->kaddrp;
859 
860 		dmabuf_p->kaddrp = (void *)((char *)dmabuf_p->kaddrp + size);
861 		dmabuf_p->ioaddr_pp = (void *)
862 		    ((char *)dmabuf_p->ioaddr_pp + size);
863 
864 		dmabuf_p->alength -= size;
865 		dmabuf_p->offset += size;
866 		dmabuf_p->dma_cookie.dmac_laddress += size;
867 		dmabuf_p->dma_cookie.dmac_size -= size;
868 	} else {
869 		buffer = KMEM_ALLOC(size, KM_NOSLEEP);
870 		if (buffer == NULL) {
871 			HXGE_ERROR_MSG((NULL, HXGE_ERR_CTL,
872 			    "Allocation of a receive page failed."));
873 			goto hxge_allocb_fail1;
874 		}
875 	}
876 
877 	hxge_mp->rx_mblk_p = desballoc(buffer, size, pri, &hxge_mp->freeb);
878 	if (hxge_mp->rx_mblk_p == NULL) {
879 		HXGE_ERROR_MSG((NULL, HXGE_ERR_CTL, "desballoc failed."));
880 		goto hxge_allocb_fail2;
881 	}
882 	hxge_mp->buffer = buffer;
883 	hxge_mp->block_size = size;
884 	hxge_mp->freeb.free_func = (void (*) ()) hxge_freeb;
885 	hxge_mp->freeb.free_arg = (caddr_t)hxge_mp;
886 	hxge_mp->ref_cnt = 1;
887 	hxge_mp->free = B_TRUE;
888 	hxge_mp->rx_use_bcopy = B_FALSE;
889 
890 	atomic_inc_32(&hxge_mblks_pending);
891 
892 	goto hxge_allocb_exit;
893 
894 hxge_allocb_fail2:
895 	if (!hxge_mp->use_buf_pool) {
896 		KMEM_FREE(buffer, size);
897 	}
898 hxge_allocb_fail1:
899 	KMEM_FREE(hxge_mp, sizeof (rx_msg_t));
900 	hxge_mp = NULL;
901 
902 hxge_allocb_exit:
903 	return (hxge_mp);
904 }
905 
906 p_mblk_t
907 hxge_dupb(p_rx_msg_t hxge_mp, uint_t offset, size_t size)
908 {
909 	p_mblk_t mp;
910 
911 	HXGE_DEBUG_MSG((NULL, MEM_CTL, "==> hxge_dupb"));
912 	HXGE_DEBUG_MSG((NULL, MEM_CTL, "hxge_mp = $%p "
913 	    "offset = 0x%08X " "size = 0x%08X", hxge_mp, offset, size));
914 
915 	mp = desballoc(&hxge_mp->buffer[offset], size, 0, &hxge_mp->freeb);
916 	if (mp == NULL) {
917 		HXGE_DEBUG_MSG((NULL, RX_CTL, "desballoc failed"));
918 		goto hxge_dupb_exit;
919 	}
920 
921 	atomic_inc_32(&hxge_mp->ref_cnt);
922 
923 hxge_dupb_exit:
924 	HXGE_DEBUG_MSG((NULL, MEM_CTL, "<== hxge_dupb mp = $%p", hxge_mp));
925 	return (mp);
926 }
927 
928 p_mblk_t
929 hxge_dupb_bcopy(p_rx_msg_t hxge_mp, uint_t offset, size_t size)
930 {
931 	p_mblk_t	mp;
932 	uchar_t		*dp;
933 
934 	mp = allocb(size + HXGE_RXBUF_EXTRA, 0);
935 	if (mp == NULL) {
936 		HXGE_DEBUG_MSG((NULL, RX_CTL, "desballoc failed"));
937 		goto hxge_dupb_bcopy_exit;
938 	}
939 	dp = mp->b_rptr = mp->b_rptr + HXGE_RXBUF_EXTRA;
940 	bcopy((void *) &hxge_mp->buffer[offset], dp, size);
941 	mp->b_wptr = dp + size;
942 
943 hxge_dupb_bcopy_exit:
944 
945 	HXGE_DEBUG_MSG((NULL, MEM_CTL, "<== hxge_dupb mp = $%p", hxge_mp));
946 
947 	return (mp);
948 }
949 
950 void hxge_post_page(p_hxge_t hxgep, p_rx_rbr_ring_t rx_rbr_p,
951     p_rx_msg_t rx_msg_p);
952 
953 void
954 hxge_post_page(p_hxge_t hxgep, p_rx_rbr_ring_t rx_rbr_p, p_rx_msg_t rx_msg_p)
955 {
956 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_post_page"));
957 
958 	/* Reuse this buffer */
959 	rx_msg_p->free = B_FALSE;
960 	rx_msg_p->cur_usage_cnt = 0;
961 	rx_msg_p->max_usage_cnt = 0;
962 	rx_msg_p->pkt_buf_size = 0;
963 
964 	if (rx_rbr_p->rbr_use_bcopy) {
965 		rx_msg_p->rx_use_bcopy = B_FALSE;
966 		atomic_dec_32(&rx_rbr_p->rbr_consumed);
967 	}
968 	atomic_dec_32(&rx_rbr_p->rbr_used);
969 
970 	/*
971 	 * Get the rbr header pointer and its offset index.
972 	 */
973 	rx_rbr_p->rbr_wr_index = ((rx_rbr_p->rbr_wr_index + 1) &
974 	    rx_rbr_p->rbr_wrap_mask);
975 	rx_rbr_p->rbr_desc_vp[rx_rbr_p->rbr_wr_index] = rx_msg_p->shifted_addr;
976 
977 	/*
978 	 * Accumulate some buffers in the ring before re-enabling the
979 	 * DMA channel, if rbr empty was signaled.
980 	 */
981 	hpi_rxdma_rdc_rbr_kick(HXGE_DEV_HPI_HANDLE(hxgep), rx_rbr_p->rdc, 1);
982 	if (rx_rbr_p->rbr_is_empty && (rx_rbr_p->rbb_max -
983 	    rx_rbr_p->rbr_used) >= HXGE_RBR_EMPTY_THRESHOLD) {
984 		hxge_rbr_empty_restore(hxgep, rx_rbr_p);
985 	}
986 
987 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
988 	    "<== hxge_post_page (channel %d post_next_index %d)",
989 	    rx_rbr_p->rdc, rx_rbr_p->rbr_wr_index));
990 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_post_page"));
991 }
992 
993 void
994 hxge_freeb(p_rx_msg_t rx_msg_p)
995 {
996 	size_t		size;
997 	uchar_t		*buffer = NULL;
998 	int		ref_cnt;
999 	boolean_t	free_state = B_FALSE;
1000 	rx_rbr_ring_t	*ring = rx_msg_p->rx_rbr_p;
1001 
1002 	HXGE_DEBUG_MSG((NULL, MEM2_CTL, "==> hxge_freeb"));
1003 	HXGE_DEBUG_MSG((NULL, MEM2_CTL,
1004 	    "hxge_freeb:rx_msg_p = $%p (block pending %d)",
1005 	    rx_msg_p, hxge_mblks_pending));
1006 
1007 	if (ring == NULL)
1008 		return;
1009 
1010 	/*
1011 	 * This is to prevent posting activities while we are recovering
1012 	 * from fatal errors. This should not be a performance drag since
1013 	 * ref_cnt != 0 most times.
1014 	 */
1015 	if (ring->rbr_state == RBR_POSTING)
1016 		MUTEX_ENTER(&ring->post_lock);
1017 
1018 	/*
1019 	 * First we need to get the free state, then
1020 	 * atomic decrement the reference count to prevent
1021 	 * the race condition with the interrupt thread that
1022 	 * is processing a loaned up buffer block.
1023 	 */
1024 	free_state = rx_msg_p->free;
1025 	ref_cnt = atomic_dec_32_nv(&rx_msg_p->ref_cnt);
1026 	if (!ref_cnt) {
1027 		atomic_dec_32(&hxge_mblks_pending);
1028 
1029 		buffer = rx_msg_p->buffer;
1030 		size = rx_msg_p->block_size;
1031 
1032 		HXGE_DEBUG_MSG((NULL, MEM2_CTL, "hxge_freeb: "
1033 		    "will free: rx_msg_p = $%p (block pending %d)",
1034 		    rx_msg_p, hxge_mblks_pending));
1035 
1036 		if (!rx_msg_p->use_buf_pool) {
1037 			KMEM_FREE(buffer, size);
1038 		}
1039 
1040 		KMEM_FREE(rx_msg_p, sizeof (rx_msg_t));
1041 		/*
1042 		 * Decrement the receive buffer ring's reference
1043 		 * count, too.
1044 		 */
1045 		atomic_dec_32(&ring->rbr_ref_cnt);
1046 
1047 		/*
1048 		 * Free the receive buffer ring, iff
1049 		 * 1. all the receive buffers have been freed
1050 		 * 2. and we are in the proper state (that is,
1051 		 *    we are not UNMAPPING).
1052 		 */
1053 		if (ring->rbr_ref_cnt == 0 &&
1054 		    ring->rbr_state == RBR_UNMAPPED) {
1055 			KMEM_FREE(ring, sizeof (*ring));
1056 			/* post_lock has been destroyed already */
1057 			return;
1058 		}
1059 	}
1060 
1061 	/*
1062 	 * Repost buffer.
1063 	 */
1064 	if (free_state && (ref_cnt == 1)) {
1065 		HXGE_DEBUG_MSG((NULL, RX_CTL,
1066 		    "hxge_freeb: post page $%p:", rx_msg_p));
1067 		if (ring->rbr_state == RBR_POSTING)
1068 			hxge_post_page(rx_msg_p->hxgep, ring, rx_msg_p);
1069 	}
1070 
1071 	if (ring->rbr_state == RBR_POSTING)
1072 		MUTEX_EXIT(&ring->post_lock);
1073 
1074 	HXGE_DEBUG_MSG((NULL, MEM2_CTL, "<== hxge_freeb"));
1075 }
1076 
1077 uint_t
1078 hxge_rx_intr(caddr_t arg1, caddr_t arg2)
1079 {
1080 	p_hxge_ring_handle_t	rhp;
1081 	p_hxge_ldv_t		ldvp = (p_hxge_ldv_t)arg1;
1082 	p_hxge_t		hxgep = (p_hxge_t)arg2;
1083 	p_hxge_ldg_t		ldgp;
1084 	uint8_t			channel;
1085 	hpi_handle_t		handle;
1086 	rdc_stat_t		cs;
1087 	p_rx_rcr_ring_t		ring;
1088 	p_rx_rbr_ring_t		rbrp;
1089 	mblk_t			*mp = NULL;
1090 
1091 	if (ldvp == NULL) {
1092 		HXGE_DEBUG_MSG((NULL, RX_INT_CTL,
1093 		    "<== hxge_rx_intr: arg2 $%p arg1 $%p", hxgep, ldvp));
1094 		return (DDI_INTR_UNCLAIMED);
1095 	}
1096 
1097 	if (arg2 == NULL || (void *) ldvp->hxgep != arg2) {
1098 		hxgep = ldvp->hxgep;
1099 	}
1100 
1101 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1102 	    "==> hxge_rx_intr: arg2 $%p arg1 $%p", hxgep, ldvp));
1103 
1104 	/*
1105 	 * This interrupt handler is for a specific receive dma channel.
1106 	 */
1107 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
1108 
1109 	/*
1110 	 * Get the control and status for this channel.
1111 	 */
1112 	channel = ldvp->vdma_index;
1113 	ring = hxgep->rx_rcr_rings->rcr_rings[channel];
1114 	rhp = &hxgep->rx_ring_handles[channel];
1115 	ldgp = ldvp->ldgp;
1116 
1117 	ASSERT(ring != NULL);
1118 #if defined(DEBUG)
1119 	if (rhp->started) {
1120 		ASSERT(ring->ldgp == ldgp);
1121 		ASSERT(ring->ldvp == ldvp);
1122 	}
1123 #endif
1124 
1125 	MUTEX_ENTER(&ring->lock);
1126 
1127 	if (!ring->poll_flag) {
1128 		RXDMA_REG_READ64(handle, RDC_STAT, channel, &cs.value);
1129 		cs.bits.ptrread = 0;
1130 		cs.bits.pktread = 0;
1131 		RXDMA_REG_WRITE64(handle, RDC_STAT, channel, cs.value);
1132 
1133 		/*
1134 		 * Process packets, if we are not in polling mode, the ring is
1135 		 * started and the interface is started. The MAC layer under
1136 		 * load will be operating in polling mode for RX traffic.
1137 		 */
1138 		if ((rhp->started) &&
1139 		    (hxgep->hxge_mac_state == HXGE_MAC_STARTED)) {
1140 			mp = hxge_rx_pkts(hxgep, ldvp->vdma_index,
1141 			    ldvp, ring, cs, -1);
1142 		}
1143 
1144 		/* Process error events. */
1145 		if (cs.value & RDC_STAT_ERROR) {
1146 			MUTEX_EXIT(&ring->lock);
1147 			(void) hxge_rx_err_evnts(hxgep, channel, ldvp, cs);
1148 			MUTEX_ENTER(&ring->lock);
1149 		}
1150 
1151 		/*
1152 		 * Enable the mailbox update interrupt if we want to use
1153 		 * mailbox. We probably don't need to use mailbox as it only
1154 		 * saves us one pio read.  Also write 1 to rcrthres and
1155 		 * rcrto to clear these two edge triggered bits.
1156 		 */
1157 		rbrp = hxgep->rx_rbr_rings->rbr_rings[channel];
1158 		MUTEX_ENTER(&rbrp->post_lock);
1159 		if (!rbrp->rbr_is_empty) {
1160 			cs.value = 0;
1161 			cs.bits.mex = 1;
1162 			cs.bits.ptrread = 0;
1163 			cs.bits.pktread = 0;
1164 			RXDMA_REG_WRITE64(handle, RDC_STAT, channel, cs.value);
1165 		}
1166 		MUTEX_EXIT(&rbrp->post_lock);
1167 
1168 		if (ldgp->nldvs == 1) {
1169 			/*
1170 			 * Re-arm the group.
1171 			 */
1172 			(void) hpi_intr_ldg_mgmt_set(handle, ldgp->ldg, B_TRUE,
1173 			    ldgp->ldg_timer);
1174 		}
1175 	} else if ((ldgp->nldvs == 1) && (ring->poll_flag)) {
1176 		/*
1177 		 * Disarm the group, if we are not a shared interrupt.
1178 		 */
1179 		(void) hpi_intr_ldg_mgmt_set(handle, ldgp->ldg, B_FALSE, 0);
1180 	} else if (ring->poll_flag) {
1181 		/*
1182 		 * Mask-off this device from the group.
1183 		 */
1184 		(void) hpi_intr_mask_set(handle, ldvp->ldv, 1);
1185 	}
1186 
1187 	MUTEX_EXIT(&ring->lock);
1188 
1189 	/*
1190 	 * Send the packets up the stack.
1191 	 */
1192 	if (mp != NULL) {
1193 		mac_rx_ring(hxgep->mach, ring->rcr_mac_handle, mp,
1194 		    ring->rcr_gen_num);
1195 	}
1196 
1197 	HXGE_DEBUG_MSG((NULL, RX_INT_CTL, "<== hxge_rx_intr"));
1198 	return (DDI_INTR_CLAIMED);
1199 }
1200 
1201 /*
1202  * Enable polling for a ring. Interrupt for the ring is disabled when
1203  * the hxge interrupt comes (see hxge_rx_intr).
1204  */
1205 int
1206 hxge_enable_poll(void *arg)
1207 {
1208 	p_hxge_ring_handle_t	ring_handle = (p_hxge_ring_handle_t)arg;
1209 	p_rx_rcr_ring_t		ringp;
1210 	p_hxge_t		hxgep;
1211 	p_hxge_ldg_t		ldgp;
1212 
1213 	if (ring_handle == NULL) {
1214 		ASSERT(ring_handle != NULL);
1215 		return (1);
1216 	}
1217 
1218 
1219 	hxgep = ring_handle->hxgep;
1220 	ringp = hxgep->rx_rcr_rings->rcr_rings[ring_handle->index];
1221 
1222 	MUTEX_ENTER(&ringp->lock);
1223 
1224 	/*
1225 	 * Are we already polling ?
1226 	 */
1227 	if (ringp->poll_flag) {
1228 		MUTEX_EXIT(&ringp->lock);
1229 		return (1);
1230 	}
1231 
1232 	ldgp = ringp->ldgp;
1233 	if (ldgp == NULL) {
1234 		MUTEX_EXIT(&ringp->lock);
1235 		return (1);
1236 	}
1237 
1238 	/*
1239 	 * Enable polling
1240 	 */
1241 	ringp->poll_flag = B_TRUE;
1242 
1243 	MUTEX_EXIT(&ringp->lock);
1244 	return (0);
1245 }
1246 
1247 /*
1248  * Disable polling for a ring and enable its interrupt.
1249  */
1250 int
1251 hxge_disable_poll(void *arg)
1252 {
1253 	p_hxge_ring_handle_t	ring_handle = (p_hxge_ring_handle_t)arg;
1254 	p_rx_rcr_ring_t		ringp;
1255 	p_hxge_t		hxgep;
1256 
1257 	if (ring_handle == NULL) {
1258 		ASSERT(ring_handle != NULL);
1259 		return (0);
1260 	}
1261 
1262 	hxgep = ring_handle->hxgep;
1263 	ringp = hxgep->rx_rcr_rings->rcr_rings[ring_handle->index];
1264 
1265 	MUTEX_ENTER(&ringp->lock);
1266 
1267 	/*
1268 	 * Disable polling: enable interrupt
1269 	 */
1270 	if (ringp->poll_flag) {
1271 		hpi_handle_t		handle;
1272 		rdc_stat_t		cs;
1273 		p_hxge_ldg_t		ldgp;
1274 
1275 		/*
1276 		 * Get the control and status for this channel.
1277 		 */
1278 		handle = HXGE_DEV_HPI_HANDLE(hxgep);
1279 
1280 		/*
1281 		 * Rearm this logical group if this is a single device
1282 		 * group.
1283 		 */
1284 		ldgp = ringp->ldgp;
1285 		if (ldgp == NULL) {
1286 			MUTEX_EXIT(&ringp->lock);
1287 			return (1);
1288 		}
1289 
1290 		ringp->poll_flag = B_FALSE;
1291 
1292 		/*
1293 		 * Enable mailbox update, to start interrupts again.
1294 		 */
1295 		cs.value = 0ULL;
1296 		cs.bits.mex = 1;
1297 		cs.bits.pktread = 0;
1298 		cs.bits.ptrread = 0;
1299 		RXDMA_REG_WRITE64(handle, RDC_STAT, ringp->rdc, cs.value);
1300 
1301 		if (ldgp->nldvs == 1) {
1302 			/*
1303 			 * Re-arm the group, since it is the only member
1304 			 * of the group.
1305 			 */
1306 			(void) hpi_intr_ldg_mgmt_set(handle, ldgp->ldg, B_TRUE,
1307 			    ldgp->ldg_timer);
1308 		} else {
1309 			/*
1310 			 * Mask-on interrupts for the device and re-arm
1311 			 * the group.
1312 			 */
1313 			(void) hpi_intr_mask_set(handle, ringp->ldvp->ldv, 0);
1314 			(void) hpi_intr_ldg_mgmt_set(handle, ldgp->ldg, B_TRUE,
1315 			    ldgp->ldg_timer);
1316 		}
1317 	}
1318 	MUTEX_EXIT(&ringp->lock);
1319 	return (0);
1320 }
1321 
1322 /*
1323  * Poll 'bytes_to_pickup' bytes of message from the rx ring.
1324  */
1325 mblk_t *
1326 hxge_rx_poll(void *arg, int bytes_to_pickup)
1327 {
1328 	p_hxge_ring_handle_t	rhp = (p_hxge_ring_handle_t)arg;
1329 	p_rx_rcr_ring_t		ring;
1330 	p_hxge_t		hxgep;
1331 	hpi_handle_t		handle;
1332 	rdc_stat_t		cs;
1333 	mblk_t			*mblk;
1334 	p_hxge_ldv_t		ldvp;
1335 
1336 	hxgep = rhp->hxgep;
1337 
1338 	/*
1339 	 * Get the control and status for this channel.
1340 	 */
1341 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
1342 	ring = hxgep->rx_rcr_rings->rcr_rings[rhp->index];
1343 
1344 	MUTEX_ENTER(&ring->lock);
1345 	ASSERT(ring->poll_flag == B_TRUE);
1346 	ASSERT(rhp->started);
1347 
1348 	if (!ring->poll_flag) {
1349 		MUTEX_EXIT(&ring->lock);
1350 		return ((mblk_t *)NULL);
1351 	}
1352 
1353 	/*
1354 	 * Get the control and status bits for the ring.
1355 	 */
1356 	RXDMA_REG_READ64(handle, RDC_STAT, rhp->index, &cs.value);
1357 	cs.bits.ptrread = 0;
1358 	cs.bits.pktread = 0;
1359 	RXDMA_REG_WRITE64(handle, RDC_STAT, rhp->index, cs.value);
1360 
1361 	/*
1362 	 * Process packets.
1363 	 */
1364 	mblk = hxge_rx_pkts(hxgep, ring->ldvp->vdma_index,
1365 	    ring->ldvp, ring, cs, bytes_to_pickup);
1366 	ldvp = ring->ldvp;
1367 
1368 	/*
1369 	 * Process Error Events.
1370 	 */
1371 	if (ldvp && (cs.value & RDC_STAT_ERROR)) {
1372 		/*
1373 		 * Recovery routines will grab the RCR ring lock.
1374 		 */
1375 		MUTEX_EXIT(&ring->lock);
1376 		(void) hxge_rx_err_evnts(hxgep, ldvp->vdma_index, ldvp, cs);
1377 		MUTEX_ENTER(&ring->lock);
1378 	}
1379 
1380 	MUTEX_EXIT(&ring->lock);
1381 	return (mblk);
1382 }
1383 
1384 /*ARGSUSED*/
1385 mblk_t *
1386 hxge_rx_pkts(p_hxge_t hxgep, uint_t vindex, p_hxge_ldv_t ldvp,
1387     p_rx_rcr_ring_t rcrp, rdc_stat_t cs, int bytes_to_read)
1388 {
1389 	hpi_handle_t		handle;
1390 	uint8_t			channel;
1391 	uint32_t		comp_rd_index;
1392 	p_rcr_entry_t		rcr_desc_rd_head_p;
1393 	p_rcr_entry_t		rcr_desc_rd_head_pp;
1394 	p_mblk_t		nmp, mp_cont, head_mp, *tail_mp;
1395 	uint16_t		qlen, nrcr_read, npkt_read;
1396 	uint32_t		qlen_hw, npkts, num_rcrs;
1397 	uint32_t		invalid_rcr_entry;
1398 	boolean_t		multi;
1399 	rdc_stat_t		pktcs;
1400 	rdc_rcr_cfg_b_t		rcr_cfg_b;
1401 	uint64_t		rcr_head_index, rcr_tail_index;
1402 	uint64_t		rcr_tail;
1403 	rdc_rcr_tail_t		rcr_tail_reg;
1404 	p_hxge_rx_ring_stats_t	rdc_stats;
1405 	int			totallen = 0;
1406 
1407 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL, "==> hxge_rx_pkts:vindex %d "
1408 	    "channel %d", vindex, ldvp->channel));
1409 
1410 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
1411 	channel = rcrp->rdc;
1412 	if (channel != ldvp->channel) {
1413 		HXGE_DEBUG_MSG((hxgep, RX_INT_CTL, "==> hxge_rx_pkts:index %d "
1414 		    "channel %d, and rcr channel %d not matched.",
1415 		    vindex, ldvp->channel, channel));
1416 		return (NULL);
1417 	}
1418 
1419 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1420 	    "==> hxge_rx_pkts: START: rcr channel %d "
1421 	    "head_p $%p head_pp $%p  index %d ",
1422 	    channel, rcrp->rcr_desc_rd_head_p,
1423 	    rcrp->rcr_desc_rd_head_pp, rcrp->comp_rd_index));
1424 
1425 	(void) hpi_rxdma_rdc_rcr_qlen_get(handle, channel, &qlen);
1426 	RXDMA_REG_READ64(handle, RDC_RCR_TAIL, channel, &rcr_tail_reg.value);
1427 	rcr_tail = rcr_tail_reg.bits.tail;
1428 
1429 	if (!qlen) {
1430 		HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1431 		    "<== hxge_rx_pkts:rcr channel %d qlen %d (no pkts)",
1432 		    channel, qlen));
1433 		return (NULL);
1434 	}
1435 
1436 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rx_pkts:rcr channel %d "
1437 	    "qlen %d", channel, qlen));
1438 
1439 	comp_rd_index = rcrp->comp_rd_index;
1440 
1441 	rcr_desc_rd_head_p = rcrp->rcr_desc_rd_head_p;
1442 	rcr_desc_rd_head_pp = rcrp->rcr_desc_rd_head_pp;
1443 	nrcr_read = npkt_read = 0;
1444 
1445 	if (hxgep->rdc_first_intr[channel])
1446 		qlen_hw = qlen;
1447 	else
1448 		qlen_hw = qlen - 1;
1449 
1450 	head_mp = NULL;
1451 	tail_mp = &head_mp;
1452 	nmp = mp_cont = NULL;
1453 	multi = B_FALSE;
1454 
1455 	rcr_head_index = rcrp->rcr_desc_rd_head_p - rcrp->rcr_desc_first_p;
1456 	rcr_tail_index = rcr_tail - rcrp->rcr_tail_begin;
1457 
1458 	if (rcr_tail_index >= rcr_head_index) {
1459 		num_rcrs = rcr_tail_index - rcr_head_index;
1460 	} else {
1461 		/* rcr_tail has wrapped around */
1462 		num_rcrs = (rcrp->comp_size - rcr_head_index) + rcr_tail_index;
1463 	}
1464 
1465 	npkts = hxge_scan_for_last_eop(rcrp, rcr_desc_rd_head_p, num_rcrs);
1466 	if (!npkts)
1467 		return (NULL);
1468 
1469 	if (qlen_hw > npkts) {
1470 		HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1471 		    "Channel %d, rcr_qlen from reg %d and from rcr_tail %d\n",
1472 		    channel, qlen_hw, qlen_sw));
1473 		qlen_hw = npkts;
1474 	}
1475 
1476 	while (qlen_hw) {
1477 #ifdef HXGE_DEBUG
1478 		hxge_dump_rcr_entry(hxgep, rcr_desc_rd_head_p);
1479 #endif
1480 		/*
1481 		 * Process one completion ring entry.
1482 		 */
1483 		invalid_rcr_entry = 0;
1484 		hxge_receive_packet(hxgep,
1485 		    rcrp, rcr_desc_rd_head_p, &multi, &nmp, &mp_cont,
1486 		    &invalid_rcr_entry);
1487 		if (invalid_rcr_entry != 0) {
1488 			rdc_stats = rcrp->rdc_stats;
1489 			rdc_stats->rcr_invalids++;
1490 			HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1491 			    "Channel %d could only read 0x%x packets, "
1492 			    "but 0x%x pending\n", channel, npkt_read, qlen_hw));
1493 			break;
1494 		}
1495 
1496 		/*
1497 		 * message chaining modes (nemo msg chaining)
1498 		 */
1499 		if (nmp) {
1500 			nmp->b_next = NULL;
1501 			if (!multi && !mp_cont) { /* frame fits a partition */
1502 				*tail_mp = nmp;
1503 				tail_mp = &nmp->b_next;
1504 				nmp = NULL;
1505 			} else if (multi && !mp_cont) { /* first segment */
1506 				*tail_mp = nmp;
1507 				tail_mp = &nmp->b_cont;
1508 			} else if (multi && mp_cont) {	/* mid of multi segs */
1509 				*tail_mp = mp_cont;
1510 				tail_mp = &mp_cont->b_cont;
1511 			} else if (!multi && mp_cont) { /* last segment */
1512 				*tail_mp = mp_cont;
1513 				tail_mp = &nmp->b_next;
1514 				totallen += MBLKL(mp_cont);
1515 				nmp = NULL;
1516 			}
1517 		}
1518 
1519 		HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1520 		    "==> hxge_rx_pkts: loop: rcr channel %d "
1521 		    "before updating: multi %d "
1522 		    "nrcr_read %d "
1523 		    "npk read %d "
1524 		    "head_pp $%p  index %d ",
1525 		    channel, multi,
1526 		    nrcr_read, npkt_read, rcr_desc_rd_head_pp, comp_rd_index));
1527 
1528 		if (!multi) {
1529 			qlen_hw--;
1530 			npkt_read++;
1531 		}
1532 
1533 		/*
1534 		 * Update the next read entry.
1535 		 */
1536 		comp_rd_index = NEXT_ENTRY(comp_rd_index,
1537 		    rcrp->comp_wrap_mask);
1538 
1539 		rcr_desc_rd_head_p = NEXT_ENTRY_PTR(rcr_desc_rd_head_p,
1540 		    rcrp->rcr_desc_first_p, rcrp->rcr_desc_last_p);
1541 
1542 		nrcr_read++;
1543 
1544 		HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1545 		    "<== hxge_rx_pkts: (SAM, process one packet) "
1546 		    "nrcr_read %d", nrcr_read));
1547 		HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1548 		    "==> hxge_rx_pkts: loop: rcr channel %d "
1549 		    "multi %d nrcr_read %d npk read %d head_pp $%p  index %d ",
1550 		    channel, multi, nrcr_read, npkt_read, rcr_desc_rd_head_pp,
1551 		    comp_rd_index));
1552 
1553 		if ((bytes_to_read != -1) &&
1554 		    (totallen >= bytes_to_read)) {
1555 			break;
1556 		}
1557 	}
1558 
1559 	rcrp->rcr_desc_rd_head_pp = rcr_desc_rd_head_pp;
1560 	rcrp->comp_rd_index = comp_rd_index;
1561 	rcrp->rcr_desc_rd_head_p = rcr_desc_rd_head_p;
1562 
1563 	if ((hxgep->intr_timeout != rcrp->intr_timeout) ||
1564 	    (hxgep->intr_threshold != rcrp->intr_threshold)) {
1565 		rcrp->intr_timeout = hxgep->intr_timeout;
1566 		rcrp->intr_threshold = hxgep->intr_threshold;
1567 		rcr_cfg_b.value = 0x0ULL;
1568 		if (rcrp->intr_timeout)
1569 			rcr_cfg_b.bits.entout = 1;
1570 		rcr_cfg_b.bits.timeout = rcrp->intr_timeout;
1571 		rcr_cfg_b.bits.pthres = rcrp->intr_threshold;
1572 		RXDMA_REG_WRITE64(handle, RDC_RCR_CFG_B,
1573 		    channel, rcr_cfg_b.value);
1574 	}
1575 
1576 	pktcs.value = 0;
1577 	if (hxgep->rdc_first_intr[channel] && (npkt_read > 0)) {
1578 		hxgep->rdc_first_intr[channel] = B_FALSE;
1579 		pktcs.bits.pktread = npkt_read - 1;
1580 	} else
1581 		pktcs.bits.pktread = npkt_read;
1582 	pktcs.bits.ptrread = nrcr_read;
1583 	RXDMA_REG_WRITE64(handle, RDC_STAT, channel, pktcs.value);
1584 
1585 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1586 	    "==> hxge_rx_pkts: EXIT: rcr channel %d "
1587 	    "head_pp $%p  index %016llx ",
1588 	    channel, rcrp->rcr_desc_rd_head_pp, rcrp->comp_rd_index));
1589 
1590 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL, "<== hxge_rx_pkts"));
1591 	return (head_mp);
1592 }
1593 
1594 #define	RCR_ENTRY_PATTERN	0x5a5a6b6b7c7c8d8dULL
1595 #define	NO_PORT_BIT		0x20
1596 #define	L4_CS_EQ_BIT		0x40
1597 
1598 static uint32_t hxge_scan_for_last_eop(p_rx_rcr_ring_t rcrp,
1599     p_rcr_entry_t rcr_desc_rd_head_p, uint32_t num_rcrs)
1600 {
1601 	uint64_t	rcr_entry;
1602 	uint32_t	rcrs = 0;
1603 	uint32_t	pkts = 0;
1604 
1605 	while (rcrs < num_rcrs) {
1606 		rcr_entry = *((uint64_t *)rcr_desc_rd_head_p);
1607 
1608 		if ((rcr_entry == 0x0) || (rcr_entry == RCR_ENTRY_PATTERN))
1609 			break;
1610 
1611 		if (!(rcr_entry & RCR_MULTI_MASK))
1612 			pkts++;
1613 
1614 		rcr_desc_rd_head_p = NEXT_ENTRY_PTR(rcr_desc_rd_head_p,
1615 		    rcrp->rcr_desc_first_p, rcrp->rcr_desc_last_p);
1616 
1617 		rcrs++;
1618 	}
1619 
1620 	return (pkts);
1621 }
1622 
1623 /*ARGSUSED*/
1624 void
1625 hxge_receive_packet(p_hxge_t hxgep, p_rx_rcr_ring_t rcr_p,
1626     p_rcr_entry_t rcr_desc_rd_head_p, boolean_t *multi_p, mblk_t **mp,
1627     mblk_t **mp_cont, uint32_t *invalid_rcr_entry)
1628 {
1629 	p_mblk_t nmp = NULL;
1630 	uint64_t multi;
1631 	uint8_t channel;
1632 	boolean_t first_entry = B_TRUE;
1633 	boolean_t is_tcp_udp = B_FALSE;
1634 	boolean_t buffer_free = B_FALSE;
1635 	boolean_t error_send_up = B_FALSE;
1636 	uint8_t error_type;
1637 	uint16_t l2_len;
1638 	uint16_t skip_len;
1639 	uint8_t pktbufsz_type;
1640 	uint64_t rcr_entry;
1641 	uint64_t *pkt_buf_addr_pp;
1642 	uint64_t *pkt_buf_addr_p;
1643 	uint32_t buf_offset;
1644 	uint32_t bsize;
1645 	uint32_t msg_index;
1646 	p_rx_rbr_ring_t rx_rbr_p;
1647 	p_rx_msg_t *rx_msg_ring_p;
1648 	p_rx_msg_t rx_msg_p;
1649 	uint16_t sw_offset_bytes = 0, hdr_size = 0;
1650 	hxge_status_t status = HXGE_OK;
1651 	boolean_t is_valid = B_FALSE;
1652 	p_hxge_rx_ring_stats_t rdc_stats;
1653 	uint32_t bytes_read;
1654 	uint8_t header0 = 0;
1655 	uint8_t header1 = 0;
1656 	uint64_t pkt_type;
1657 	uint8_t no_port_bit = 0;
1658 	uint8_t l4_cs_eq_bit = 0;
1659 
1660 	channel = rcr_p->rdc;
1661 
1662 	HXGE_DEBUG_MSG((hxgep, RX2_CTL, "==> hxge_receive_packet"));
1663 
1664 	first_entry = (*mp == NULL) ? B_TRUE : B_FALSE;
1665 	rcr_entry = *((uint64_t *)rcr_desc_rd_head_p);
1666 
1667 	/* Verify the content of the rcr_entry for a hardware bug workaround */
1668 	if ((rcr_entry == 0x0) || (rcr_entry == RCR_ENTRY_PATTERN)) {
1669 		*invalid_rcr_entry = 1;
1670 		HXGE_DEBUG_MSG((hxgep, RX2_CTL, "hxge_receive_packet "
1671 		    "Channel %d invalid RCR entry 0x%llx found, returning\n",
1672 		    channel, (long long) rcr_entry));
1673 		return;
1674 	}
1675 	*((uint64_t *)rcr_desc_rd_head_p) = RCR_ENTRY_PATTERN;
1676 
1677 	multi = (rcr_entry & RCR_MULTI_MASK);
1678 	pkt_type = (rcr_entry & RCR_PKT_TYPE_MASK);
1679 
1680 	error_type = ((rcr_entry & RCR_ERROR_MASK) >> RCR_ERROR_SHIFT);
1681 	l2_len = ((rcr_entry & RCR_L2_LEN_MASK) >> RCR_L2_LEN_SHIFT);
1682 
1683 	/*
1684 	 * Hardware does not strip the CRC due bug ID 11451 where
1685 	 * the hardware mis handles minimum size packets.
1686 	 */
1687 	l2_len -= ETHERFCSL;
1688 
1689 	pktbufsz_type = ((rcr_entry & RCR_PKTBUFSZ_MASK) >>
1690 	    RCR_PKTBUFSZ_SHIFT);
1691 	pkt_buf_addr_pp = (uint64_t *)((rcr_entry & RCR_PKT_BUF_ADDR_MASK) <<
1692 	    RCR_PKT_BUF_ADDR_SHIFT);
1693 
1694 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1695 	    "==> hxge_receive_packet: entryp $%p entry 0x%0llx "
1696 	    "pkt_buf_addr_pp $%p l2_len %d multi %d "
1697 	    "error_type 0x%x pktbufsz_type %d ",
1698 	    rcr_desc_rd_head_p, rcr_entry, pkt_buf_addr_pp, l2_len,
1699 	    multi, error_type, pktbufsz_type));
1700 
1701 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1702 	    "==> hxge_receive_packet: entryp $%p entry 0x%0llx "
1703 	    "pkt_buf_addr_pp $%p l2_len %d multi %d "
1704 	    "error_type 0x%x ", rcr_desc_rd_head_p,
1705 	    rcr_entry, pkt_buf_addr_pp, l2_len, multi, error_type));
1706 
1707 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1708 	    "==> (rbr) hxge_receive_packet: entry 0x%0llx "
1709 	    "full pkt_buf_addr_pp $%p l2_len %d",
1710 	    rcr_entry, pkt_buf_addr_pp, l2_len));
1711 
1712 	/* get the stats ptr */
1713 	rdc_stats = rcr_p->rdc_stats;
1714 
1715 	if (!l2_len) {
1716 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
1717 		    "<== hxge_receive_packet: failed: l2 length is 0."));
1718 		return;
1719 	}
1720 
1721 	/* shift 6 bits to get the full io address */
1722 	pkt_buf_addr_pp = (uint64_t *)((uint64_t)pkt_buf_addr_pp <<
1723 	    RCR_PKT_BUF_ADDR_SHIFT_FULL);
1724 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1725 	    "==> (rbr) hxge_receive_packet: entry 0x%0llx "
1726 	    "full pkt_buf_addr_pp $%p l2_len %d",
1727 	    rcr_entry, pkt_buf_addr_pp, l2_len));
1728 
1729 	rx_rbr_p = rcr_p->rx_rbr_p;
1730 	rx_msg_ring_p = rx_rbr_p->rx_msg_ring;
1731 
1732 	if (first_entry) {
1733 		hdr_size = (rcr_p->full_hdr_flag ? RXDMA_HDR_SIZE_FULL :
1734 		    RXDMA_HDR_SIZE_DEFAULT);
1735 
1736 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
1737 		    "==> hxge_receive_packet: first entry 0x%016llx "
1738 		    "pkt_buf_addr_pp $%p l2_len %d hdr %d",
1739 		    rcr_entry, pkt_buf_addr_pp, l2_len, hdr_size));
1740 	}
1741 
1742 	MUTEX_ENTER(&rx_rbr_p->lock);
1743 
1744 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
1745 	    "==> (rbr 1) hxge_receive_packet: entry 0x%0llx "
1746 	    "full pkt_buf_addr_pp $%p l2_len %d",
1747 	    rcr_entry, pkt_buf_addr_pp, l2_len));
1748 
1749 	/*
1750 	 * Packet buffer address in the completion entry points to the starting
1751 	 * buffer address (offset 0). Use the starting buffer address to locate
1752 	 * the corresponding kernel address.
1753 	 */
1754 	status = hxge_rxbuf_pp_to_vp(hxgep, rx_rbr_p,
1755 	    pktbufsz_type, pkt_buf_addr_pp, &pkt_buf_addr_p,
1756 	    &buf_offset, &msg_index);
1757 
1758 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
1759 	    "==> (rbr 2) hxge_receive_packet: entry 0x%0llx "
1760 	    "full pkt_buf_addr_pp $%p l2_len %d",
1761 	    rcr_entry, pkt_buf_addr_pp, l2_len));
1762 
1763 	if (status != HXGE_OK) {
1764 		MUTEX_EXIT(&rx_rbr_p->lock);
1765 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
1766 		    "<== hxge_receive_packet: found vaddr failed %d", status));
1767 		return;
1768 	}
1769 
1770 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1771 	    "==> (rbr 3) hxge_receive_packet: entry 0x%0llx "
1772 	    "full pkt_buf_addr_pp $%p l2_len %d",
1773 	    rcr_entry, pkt_buf_addr_pp, l2_len));
1774 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1775 	    "==> (rbr 4 msgindex %d) hxge_receive_packet: entry 0x%0llx "
1776 	    "full pkt_buf_addr_pp $%p l2_len %d",
1777 	    msg_index, rcr_entry, pkt_buf_addr_pp, l2_len));
1778 
1779 	if (msg_index >= rx_rbr_p->tnblocks) {
1780 		MUTEX_EXIT(&rx_rbr_p->lock);
1781 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1782 		    "==> hxge_receive_packet: FATAL msg_index (%d) "
1783 		    "should be smaller than tnblocks (%d)\n",
1784 		    msg_index, rx_rbr_p->tnblocks));
1785 		return;
1786 	}
1787 
1788 	rx_msg_p = rx_msg_ring_p[msg_index];
1789 
1790 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1791 	    "==> (rbr 4 msgindex %d) hxge_receive_packet: entry 0x%0llx "
1792 	    "full pkt_buf_addr_pp $%p l2_len %d",
1793 	    msg_index, rcr_entry, pkt_buf_addr_pp, l2_len));
1794 
1795 	switch (pktbufsz_type) {
1796 	case RCR_PKTBUFSZ_0:
1797 		bsize = rx_rbr_p->pkt_buf_size0_bytes;
1798 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1799 		    "==> hxge_receive_packet: 0 buf %d", bsize));
1800 		break;
1801 	case RCR_PKTBUFSZ_1:
1802 		bsize = rx_rbr_p->pkt_buf_size1_bytes;
1803 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1804 		    "==> hxge_receive_packet: 1 buf %d", bsize));
1805 		break;
1806 	case RCR_PKTBUFSZ_2:
1807 		bsize = rx_rbr_p->pkt_buf_size2_bytes;
1808 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
1809 		    "==> hxge_receive_packet: 2 buf %d", bsize));
1810 		break;
1811 	case RCR_SINGLE_BLOCK:
1812 		bsize = rx_msg_p->block_size;
1813 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1814 		    "==> hxge_receive_packet: single %d", bsize));
1815 
1816 		break;
1817 	default:
1818 		MUTEX_EXIT(&rx_rbr_p->lock);
1819 		return;
1820 	}
1821 
1822 	DMA_COMMON_SYNC_OFFSET(rx_msg_p->buf_dma,
1823 	    (buf_offset + sw_offset_bytes), (hdr_size + l2_len),
1824 	    DDI_DMA_SYNC_FORCPU);
1825 
1826 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1827 	    "==> hxge_receive_packet: after first dump:usage count"));
1828 
1829 	if (rx_msg_p->cur_usage_cnt == 0) {
1830 		atomic_inc_32(&rx_rbr_p->rbr_used);
1831 		if (rx_rbr_p->rbr_use_bcopy) {
1832 			atomic_inc_32(&rx_rbr_p->rbr_consumed);
1833 			if (rx_rbr_p->rbr_consumed <
1834 			    rx_rbr_p->rbr_threshold_hi) {
1835 				if (rx_rbr_p->rbr_threshold_lo == 0 ||
1836 				    ((rx_rbr_p->rbr_consumed >=
1837 				    rx_rbr_p->rbr_threshold_lo) &&
1838 				    (rx_rbr_p->rbr_bufsize_type >=
1839 				    pktbufsz_type))) {
1840 					rx_msg_p->rx_use_bcopy = B_TRUE;
1841 				}
1842 			} else {
1843 				rx_msg_p->rx_use_bcopy = B_TRUE;
1844 			}
1845 		}
1846 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1847 		    "==> hxge_receive_packet: buf %d (new block) ", bsize));
1848 
1849 		rx_msg_p->pkt_buf_size_code = pktbufsz_type;
1850 		rx_msg_p->pkt_buf_size = bsize;
1851 		rx_msg_p->cur_usage_cnt = 1;
1852 		if (pktbufsz_type == RCR_SINGLE_BLOCK) {
1853 			HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1854 			    "==> hxge_receive_packet: buf %d (single block) ",
1855 			    bsize));
1856 			/*
1857 			 * Buffer can be reused once the free function is
1858 			 * called.
1859 			 */
1860 			rx_msg_p->max_usage_cnt = 1;
1861 			buffer_free = B_TRUE;
1862 		} else {
1863 			rx_msg_p->max_usage_cnt = rx_msg_p->block_size / bsize;
1864 			if (rx_msg_p->max_usage_cnt == 1) {
1865 				buffer_free = B_TRUE;
1866 			}
1867 		}
1868 	} else {
1869 		rx_msg_p->cur_usage_cnt++;
1870 		if (rx_msg_p->cur_usage_cnt == rx_msg_p->max_usage_cnt) {
1871 			buffer_free = B_TRUE;
1872 		}
1873 	}
1874 
1875 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
1876 	    "msgbuf index = %d l2len %d bytes usage %d max_usage %d ",
1877 	    msg_index, l2_len,
1878 	    rx_msg_p->cur_usage_cnt, rx_msg_p->max_usage_cnt));
1879 
1880 	if (error_type) {
1881 		rdc_stats->ierrors++;
1882 		/* Update error stats */
1883 		rdc_stats->errlog.compl_err_type = error_type;
1884 		HXGE_FM_REPORT_ERROR(hxgep, 0, HXGE_FM_EREPORT_RDMC_RCR_ERR);
1885 
1886 		if (error_type & RCR_CTRL_FIFO_DED) {
1887 			rdc_stats->ctrl_fifo_ecc_err++;
1888 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
1889 			    " hxge_receive_packet: "
1890 			    " channel %d RCR ctrl_fifo_ded error", channel));
1891 		} else if (error_type & RCR_DATA_FIFO_DED) {
1892 			rdc_stats->data_fifo_ecc_err++;
1893 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
1894 			    " hxge_receive_packet: channel %d"
1895 			    " RCR data_fifo_ded error", channel));
1896 		}
1897 
1898 		/*
1899 		 * Update and repost buffer block if max usage count is
1900 		 * reached.
1901 		 */
1902 		if (error_send_up == B_FALSE) {
1903 			atomic_inc_32(&rx_msg_p->ref_cnt);
1904 			if (buffer_free == B_TRUE) {
1905 				rx_msg_p->free = B_TRUE;
1906 			}
1907 
1908 			MUTEX_EXIT(&rx_rbr_p->lock);
1909 			hxge_freeb(rx_msg_p);
1910 			return;
1911 		}
1912 	}
1913 
1914 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1915 	    "==> hxge_receive_packet: DMA sync second "));
1916 
1917 	bytes_read = rcr_p->rcvd_pkt_bytes;
1918 	skip_len = sw_offset_bytes + hdr_size;
1919 
1920 	if (first_entry) {
1921 		header0 = rx_msg_p->buffer[buf_offset];
1922 		no_port_bit = header0 & NO_PORT_BIT;
1923 		header1 = rx_msg_p->buffer[buf_offset + 1];
1924 		l4_cs_eq_bit = header1 & L4_CS_EQ_BIT;
1925 	}
1926 
1927 	if (!rx_msg_p->rx_use_bcopy) {
1928 		/*
1929 		 * For loaned up buffers, the driver reference count
1930 		 * will be incremented first and then the free state.
1931 		 */
1932 		if ((nmp = hxge_dupb(rx_msg_p, buf_offset, bsize)) != NULL) {
1933 			if (first_entry) {
1934 				nmp->b_rptr = &nmp->b_rptr[skip_len];
1935 				if (l2_len < bsize - skip_len) {
1936 					nmp->b_wptr = &nmp->b_rptr[l2_len];
1937 				} else {
1938 					nmp->b_wptr = &nmp->b_rptr[bsize
1939 					    - skip_len];
1940 				}
1941 			} else {
1942 				if (l2_len - bytes_read < bsize) {
1943 					nmp->b_wptr =
1944 					    &nmp->b_rptr[l2_len - bytes_read];
1945 				} else {
1946 					nmp->b_wptr = &nmp->b_rptr[bsize];
1947 				}
1948 			}
1949 		}
1950 	} else {
1951 		if (first_entry) {
1952 			nmp = hxge_dupb_bcopy(rx_msg_p, buf_offset + skip_len,
1953 			    l2_len < bsize - skip_len ?
1954 			    l2_len : bsize - skip_len);
1955 		} else {
1956 			nmp = hxge_dupb_bcopy(rx_msg_p, buf_offset,
1957 			    l2_len - bytes_read < bsize ?
1958 			    l2_len - bytes_read : bsize);
1959 		}
1960 	}
1961 
1962 	if (nmp != NULL) {
1963 		if (first_entry)
1964 			bytes_read  = nmp->b_wptr - nmp->b_rptr;
1965 		else
1966 			bytes_read += nmp->b_wptr - nmp->b_rptr;
1967 
1968 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
1969 		    "==> hxge_receive_packet after dupb: "
1970 		    "rbr consumed %d "
1971 		    "pktbufsz_type %d "
1972 		    "nmp $%p rptr $%p wptr $%p "
1973 		    "buf_offset %d bzise %d l2_len %d skip_len %d",
1974 		    rx_rbr_p->rbr_consumed,
1975 		    pktbufsz_type,
1976 		    nmp, nmp->b_rptr, nmp->b_wptr,
1977 		    buf_offset, bsize, l2_len, skip_len));
1978 	} else {
1979 		cmn_err(CE_WARN, "!hxge_receive_packet: update stats (error)");
1980 
1981 		atomic_inc_32(&rx_msg_p->ref_cnt);
1982 		if (buffer_free == B_TRUE) {
1983 			rx_msg_p->free = B_TRUE;
1984 		}
1985 
1986 		MUTEX_EXIT(&rx_rbr_p->lock);
1987 		hxge_freeb(rx_msg_p);
1988 		return;
1989 	}
1990 
1991 	if (buffer_free == B_TRUE) {
1992 		rx_msg_p->free = B_TRUE;
1993 	}
1994 
1995 	/*
1996 	 * ERROR, FRAG and PKT_TYPE are only reported in the first entry. If a
1997 	 * packet is not fragmented and no error bit is set, then L4 checksum
1998 	 * is OK.
1999 	 */
2000 	is_valid = (nmp != NULL);
2001 	if (first_entry) {
2002 		rdc_stats->ipackets++; /* count only 1st seg for jumbo */
2003 		if (l2_len > (STD_FRAME_SIZE - ETHERFCSL))
2004 			rdc_stats->jumbo_pkts++;
2005 		rdc_stats->ibytes += skip_len + l2_len < bsize ?
2006 		    l2_len : bsize;
2007 	} else {
2008 		/*
2009 		 * Add the current portion of the packet to the kstats.
2010 		 * The current portion of the packet is calculated by using
2011 		 * length of the packet and the previously received portion.
2012 		 */
2013 		rdc_stats->ibytes += l2_len - rcr_p->rcvd_pkt_bytes < bsize ?
2014 		    l2_len - rcr_p->rcvd_pkt_bytes : bsize;
2015 	}
2016 
2017 	rcr_p->rcvd_pkt_bytes = bytes_read;
2018 
2019 	if (rx_msg_p->free && rx_msg_p->rx_use_bcopy) {
2020 		atomic_inc_32(&rx_msg_p->ref_cnt);
2021 		MUTEX_EXIT(&rx_rbr_p->lock);
2022 		hxge_freeb(rx_msg_p);
2023 	} else
2024 		MUTEX_EXIT(&rx_rbr_p->lock);
2025 
2026 	if (is_valid) {
2027 		nmp->b_cont = NULL;
2028 		if (first_entry) {
2029 			*mp = nmp;
2030 			*mp_cont = NULL;
2031 		} else {
2032 			*mp_cont = nmp;
2033 		}
2034 	}
2035 
2036 	/*
2037 	 * Update stats and hardware checksuming.
2038 	 */
2039 	if (is_valid && !multi) {
2040 		is_tcp_udp = ((pkt_type == RCR_PKT_IS_TCP ||
2041 		    pkt_type == RCR_PKT_IS_UDP) ? B_TRUE : B_FALSE);
2042 
2043 		if (!no_port_bit && l4_cs_eq_bit && is_tcp_udp && !error_type) {
2044 			mac_hcksum_set(nmp, 0, 0, 0, 0, HCK_FULLCKSUM_OK);
2045 
2046 			HXGE_DEBUG_MSG((hxgep, RX_CTL,
2047 			    "==> hxge_receive_packet: Full tcp/udp cksum "
2048 			    "is_valid 0x%x multi %d error %d",
2049 			    is_valid, multi, error_type));
2050 		}
2051 	}
2052 
2053 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
2054 	    "==> hxge_receive_packet: *mp 0x%016llx", *mp));
2055 
2056 	*multi_p = (multi == RCR_MULTI_MASK);
2057 
2058 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_receive_packet: "
2059 	    "multi %d nmp 0x%016llx *mp 0x%016llx *mp_cont 0x%016llx",
2060 	    *multi_p, nmp, *mp, *mp_cont));
2061 }
2062 
2063 static void
2064 hxge_rx_rbr_empty_recover(p_hxge_t hxgep, uint8_t channel)
2065 {
2066 	hpi_handle_t	handle;
2067 	p_rx_rcr_ring_t	rcrp;
2068 	p_rx_rbr_ring_t	rbrp;
2069 
2070 	rcrp = hxgep->rx_rcr_rings->rcr_rings[channel];
2071 	rbrp = rcrp->rx_rbr_p;
2072 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
2073 
2074 	/*
2075 	 * Wait for the channel to be quiet
2076 	 */
2077 	(void) hpi_rxdma_cfg_rdc_wait_for_qst(handle, channel);
2078 
2079 	/*
2080 	 * Post page will accumulate some buffers before re-enabling
2081 	 * the DMA channel.
2082 	 */
2083 
2084 	MUTEX_ENTER(&rbrp->post_lock);
2085 	if ((rbrp->rbb_max - rbrp->rbr_used) >= HXGE_RBR_EMPTY_THRESHOLD) {
2086 		hxge_rbr_empty_restore(hxgep, rbrp);
2087 	} else {
2088 		rbrp->rbr_is_empty = B_TRUE;
2089 	}
2090 	MUTEX_EXIT(&rbrp->post_lock);
2091 }
2092 
2093 
2094 /*ARGSUSED*/
2095 static hxge_status_t
2096 hxge_rx_err_evnts(p_hxge_t hxgep, uint_t index, p_hxge_ldv_t ldvp,
2097     rdc_stat_t cs)
2098 {
2099 	p_hxge_rx_ring_stats_t	rdc_stats;
2100 	hpi_handle_t		handle;
2101 	boolean_t		rxchan_fatal = B_FALSE;
2102 	uint8_t			channel;
2103 	hxge_status_t		status = HXGE_OK;
2104 
2105 	HXGE_DEBUG_MSG((hxgep, INT_CTL, "==> hxge_rx_err_evnts"));
2106 
2107 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
2108 	channel = ldvp->channel;
2109 
2110 	rdc_stats = &hxgep->statsp->rdc_stats[ldvp->vdma_index];
2111 
2112 	if (cs.bits.rbr_cpl_to) {
2113 		rdc_stats->rbr_tmout++;
2114 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2115 		    HXGE_FM_EREPORT_RDMC_RBR_CPL_TO);
2116 		rxchan_fatal = B_TRUE;
2117 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2118 		    "==> hxge_rx_err_evnts(channel %d): "
2119 		    "fatal error: rx_rbr_timeout", channel));
2120 	}
2121 
2122 	if ((cs.bits.rcr_shadow_par_err) || (cs.bits.rbr_prefetch_par_err)) {
2123 		(void) hpi_rxdma_ring_perr_stat_get(handle,
2124 		    &rdc_stats->errlog.pre_par, &rdc_stats->errlog.sha_par);
2125 	}
2126 
2127 	if (cs.bits.rcr_shadow_par_err) {
2128 		rdc_stats->rcr_sha_par++;
2129 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2130 		    HXGE_FM_EREPORT_RDMC_RCR_SHA_PAR);
2131 		rxchan_fatal = B_TRUE;
2132 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2133 		    "==> hxge_rx_err_evnts(channel %d): "
2134 		    "fatal error: rcr_shadow_par_err", channel));
2135 	}
2136 
2137 	if (cs.bits.rbr_prefetch_par_err) {
2138 		rdc_stats->rbr_pre_par++;
2139 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2140 		    HXGE_FM_EREPORT_RDMC_RBR_PRE_PAR);
2141 		rxchan_fatal = B_TRUE;
2142 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2143 		    "==> hxge_rx_err_evnts(channel %d): "
2144 		    "fatal error: rbr_prefetch_par_err", channel));
2145 	}
2146 
2147 	if (cs.bits.rbr_pre_empty) {
2148 		rdc_stats->rbr_pre_empty++;
2149 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2150 		    HXGE_FM_EREPORT_RDMC_RBR_PRE_EMPTY);
2151 		rxchan_fatal = B_TRUE;
2152 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2153 		    "==> hxge_rx_err_evnts(channel %d): "
2154 		    "fatal error: rbr_pre_empty", channel));
2155 	}
2156 
2157 	if (cs.bits.peu_resp_err) {
2158 		rdc_stats->peu_resp_err++;
2159 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2160 		    HXGE_FM_EREPORT_RDMC_PEU_RESP_ERR);
2161 		rxchan_fatal = B_TRUE;
2162 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2163 		    "==> hxge_rx_err_evnts(channel %d): "
2164 		    "fatal error: peu_resp_err", channel));
2165 	}
2166 
2167 	if (cs.bits.rcr_thres) {
2168 		rdc_stats->rcr_thres++;
2169 	}
2170 
2171 	if (cs.bits.rcr_to) {
2172 		rdc_stats->rcr_to++;
2173 	}
2174 
2175 	if (cs.bits.rcr_shadow_full) {
2176 		rdc_stats->rcr_shadow_full++;
2177 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2178 		    HXGE_FM_EREPORT_RDMC_RCR_SHA_FULL);
2179 		rxchan_fatal = B_TRUE;
2180 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2181 		    "==> hxge_rx_err_evnts(channel %d): "
2182 		    "fatal error: rcr_shadow_full", channel));
2183 	}
2184 
2185 	if (cs.bits.rcr_full) {
2186 		rdc_stats->rcrfull++;
2187 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2188 		    HXGE_FM_EREPORT_RDMC_RCRFULL);
2189 		rxchan_fatal = B_TRUE;
2190 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2191 		    "==> hxge_rx_err_evnts(channel %d): "
2192 		    "fatal error: rcrfull error", channel));
2193 	}
2194 
2195 	if (cs.bits.rbr_empty) {
2196 		rdc_stats->rbr_empty++;
2197 		hxge_rx_rbr_empty_recover(hxgep, channel);
2198 	}
2199 
2200 	if (cs.bits.rbr_full) {
2201 		rdc_stats->rbrfull++;
2202 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2203 		    HXGE_FM_EREPORT_RDMC_RBRFULL);
2204 		rxchan_fatal = B_TRUE;
2205 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2206 		    "==> hxge_rx_err_evnts(channel %d): "
2207 		    "fatal error: rbr_full error", channel));
2208 	}
2209 
2210 	if (rxchan_fatal) {
2211 		p_rx_rcr_ring_t	rcrp;
2212 		p_rx_rbr_ring_t rbrp;
2213 
2214 		rcrp = hxgep->rx_rcr_rings->rcr_rings[channel];
2215 		rbrp = rcrp->rx_rbr_p;
2216 
2217 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2218 		    " hxge_rx_err_evnts: fatal error on Channel #%d\n",
2219 		    channel));
2220 
2221 		MUTEX_ENTER(&rbrp->post_lock);
2222 		/* This function needs to be inside the post_lock */
2223 		status = hxge_rxdma_fatal_err_recover(hxgep, channel);
2224 		MUTEX_EXIT(&rbrp->post_lock);
2225 		if (status == HXGE_OK) {
2226 			FM_SERVICE_RESTORED(hxgep);
2227 		}
2228 	}
2229 
2230 	HXGE_DEBUG_MSG((hxgep, INT_CTL, "<== hxge_rx_err_evnts"));
2231 	return (status);
2232 }
2233 
2234 static hxge_status_t
2235 hxge_map_rxdma(p_hxge_t hxgep)
2236 {
2237 	int			i, ndmas;
2238 	uint16_t		channel;
2239 	p_rx_rbr_rings_t	rx_rbr_rings;
2240 	p_rx_rbr_ring_t		*rbr_rings;
2241 	p_rx_rcr_rings_t	rx_rcr_rings;
2242 	p_rx_rcr_ring_t		*rcr_rings;
2243 	p_rx_mbox_areas_t	rx_mbox_areas_p;
2244 	p_rx_mbox_t		*rx_mbox_p;
2245 	p_hxge_dma_pool_t	dma_buf_poolp;
2246 	p_hxge_dma_common_t	*dma_buf_p;
2247 	p_hxge_dma_pool_t	dma_rbr_cntl_poolp;
2248 	p_hxge_dma_common_t	*dma_rbr_cntl_p;
2249 	p_hxge_dma_pool_t	dma_rcr_cntl_poolp;
2250 	p_hxge_dma_common_t	*dma_rcr_cntl_p;
2251 	p_hxge_dma_pool_t	dma_mbox_cntl_poolp;
2252 	p_hxge_dma_common_t	*dma_mbox_cntl_p;
2253 	uint32_t		*num_chunks;
2254 	hxge_status_t		status = HXGE_OK;
2255 
2256 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_map_rxdma"));
2257 
2258 	dma_buf_poolp = hxgep->rx_buf_pool_p;
2259 	dma_rbr_cntl_poolp = hxgep->rx_rbr_cntl_pool_p;
2260 	dma_rcr_cntl_poolp = hxgep->rx_rcr_cntl_pool_p;
2261 	dma_mbox_cntl_poolp = hxgep->rx_mbox_cntl_pool_p;
2262 
2263 	if (!dma_buf_poolp->buf_allocated ||
2264 	    !dma_rbr_cntl_poolp->buf_allocated ||
2265 	    !dma_rcr_cntl_poolp->buf_allocated ||
2266 	    !dma_mbox_cntl_poolp->buf_allocated) {
2267 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2268 		    "<== hxge_map_rxdma: buf not allocated"));
2269 		return (HXGE_ERROR);
2270 	}
2271 
2272 	ndmas = dma_buf_poolp->ndmas;
2273 	if (!ndmas) {
2274 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
2275 		    "<== hxge_map_rxdma: no dma allocated"));
2276 		return (HXGE_ERROR);
2277 	}
2278 
2279 	num_chunks = dma_buf_poolp->num_chunks;
2280 	dma_buf_p = dma_buf_poolp->dma_buf_pool_p;
2281 	dma_rbr_cntl_p = dma_rbr_cntl_poolp->dma_buf_pool_p;
2282 	dma_rcr_cntl_p = dma_rcr_cntl_poolp->dma_buf_pool_p;
2283 	dma_mbox_cntl_p = dma_mbox_cntl_poolp->dma_buf_pool_p;
2284 
2285 	rx_rbr_rings = (p_rx_rbr_rings_t)
2286 	    KMEM_ZALLOC(sizeof (rx_rbr_rings_t), KM_SLEEP);
2287 	rbr_rings = (p_rx_rbr_ring_t *)KMEM_ZALLOC(
2288 	    sizeof (p_rx_rbr_ring_t) * ndmas, KM_SLEEP);
2289 
2290 	rx_rcr_rings = (p_rx_rcr_rings_t)
2291 	    KMEM_ZALLOC(sizeof (rx_rcr_rings_t), KM_SLEEP);
2292 	rcr_rings = (p_rx_rcr_ring_t *)KMEM_ZALLOC(
2293 	    sizeof (p_rx_rcr_ring_t) * ndmas, KM_SLEEP);
2294 
2295 	rx_mbox_areas_p = (p_rx_mbox_areas_t)
2296 	    KMEM_ZALLOC(sizeof (rx_mbox_areas_t), KM_SLEEP);
2297 	rx_mbox_p = (p_rx_mbox_t *)KMEM_ZALLOC(
2298 	    sizeof (p_rx_mbox_t) * ndmas, KM_SLEEP);
2299 
2300 	/*
2301 	 * Timeout should be set based on the system clock divider.
2302 	 * The following timeout value of 1 assumes that the
2303 	 * granularity (1000) is 3 microseconds running at 300MHz.
2304 	 */
2305 
2306 	hxgep->intr_threshold = RXDMA_RCR_PTHRES_DEFAULT;
2307 	hxgep->intr_timeout = RXDMA_RCR_TO_DEFAULT;
2308 
2309 	/*
2310 	 * Map descriptors from the buffer polls for each dam channel.
2311 	 */
2312 	for (i = 0; i < ndmas; i++) {
2313 		if (((p_hxge_dma_common_t)dma_buf_p[i]) == NULL) {
2314 			status = HXGE_ERROR;
2315 			goto hxge_map_rxdma_fail1;
2316 		}
2317 
2318 		/*
2319 		 * Set up and prepare buffer blocks, descriptors and mailbox.
2320 		 */
2321 		channel = ((p_hxge_dma_common_t)dma_buf_p[i])->dma_channel;
2322 		status = hxge_map_rxdma_channel(hxgep, channel,
2323 		    (p_hxge_dma_common_t *)&dma_buf_p[i],
2324 		    (p_rx_rbr_ring_t *)&rbr_rings[i],
2325 		    num_chunks[i],
2326 		    (p_hxge_dma_common_t *)&dma_rbr_cntl_p[i],
2327 		    (p_hxge_dma_common_t *)&dma_rcr_cntl_p[i],
2328 		    (p_hxge_dma_common_t *)&dma_mbox_cntl_p[i],
2329 		    (p_rx_rcr_ring_t *)&rcr_rings[i],
2330 		    (p_rx_mbox_t *)&rx_mbox_p[i]);
2331 		if (status != HXGE_OK) {
2332 			goto hxge_map_rxdma_fail1;
2333 		}
2334 		rbr_rings[i]->index = (uint16_t)i;
2335 		rcr_rings[i]->index = (uint16_t)i;
2336 		rcr_rings[i]->rdc_stats = &hxgep->statsp->rdc_stats[i];
2337 	}
2338 
2339 	rx_rbr_rings->ndmas = rx_rcr_rings->ndmas = ndmas;
2340 	rx_rbr_rings->rbr_rings = rbr_rings;
2341 	hxgep->rx_rbr_rings = rx_rbr_rings;
2342 	rx_rcr_rings->rcr_rings = rcr_rings;
2343 	hxgep->rx_rcr_rings = rx_rcr_rings;
2344 
2345 	rx_mbox_areas_p->rxmbox_areas = rx_mbox_p;
2346 	hxgep->rx_mbox_areas_p = rx_mbox_areas_p;
2347 
2348 	goto hxge_map_rxdma_exit;
2349 
2350 hxge_map_rxdma_fail1:
2351 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2352 	    "==> hxge_map_rxdma: unmap rbr,rcr (status 0x%x channel %d i %d)",
2353 	    status, channel, i));
2354 	i--;
2355 	for (; i >= 0; i--) {
2356 		channel = ((p_hxge_dma_common_t)dma_buf_p[i])->dma_channel;
2357 		hxge_unmap_rxdma_channel(hxgep, channel,
2358 		    rbr_rings[i], rcr_rings[i], rx_mbox_p[i]);
2359 	}
2360 
2361 	KMEM_FREE(rbr_rings, sizeof (p_rx_rbr_ring_t) * ndmas);
2362 	KMEM_FREE(rx_rbr_rings, sizeof (rx_rbr_rings_t));
2363 	KMEM_FREE(rcr_rings, sizeof (p_rx_rcr_ring_t) * ndmas);
2364 	KMEM_FREE(rx_rcr_rings, sizeof (rx_rcr_rings_t));
2365 	KMEM_FREE(rx_mbox_p, sizeof (p_rx_mbox_t) * ndmas);
2366 	KMEM_FREE(rx_mbox_areas_p, sizeof (rx_mbox_areas_t));
2367 
2368 hxge_map_rxdma_exit:
2369 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2370 	    "<== hxge_map_rxdma: (status 0x%x channel %d)", status, channel));
2371 
2372 	return (status);
2373 }
2374 
2375 static void
2376 hxge_unmap_rxdma(p_hxge_t hxgep)
2377 {
2378 	int			i, ndmas;
2379 	uint16_t		channel;
2380 	p_rx_rbr_rings_t	rx_rbr_rings;
2381 	p_rx_rbr_ring_t		*rbr_rings;
2382 	p_rx_rcr_rings_t	rx_rcr_rings;
2383 	p_rx_rcr_ring_t		*rcr_rings;
2384 	p_rx_mbox_areas_t	rx_mbox_areas_p;
2385 	p_rx_mbox_t		*rx_mbox_p;
2386 	p_hxge_dma_pool_t	dma_buf_poolp;
2387 	p_hxge_dma_pool_t	dma_rbr_cntl_poolp;
2388 	p_hxge_dma_pool_t	dma_rcr_cntl_poolp;
2389 	p_hxge_dma_pool_t	dma_mbox_cntl_poolp;
2390 	p_hxge_dma_common_t	*dma_buf_p;
2391 
2392 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_unmap_rxdma"));
2393 
2394 	dma_buf_poolp = hxgep->rx_buf_pool_p;
2395 	dma_rbr_cntl_poolp = hxgep->rx_rbr_cntl_pool_p;
2396 	dma_rcr_cntl_poolp = hxgep->rx_rcr_cntl_pool_p;
2397 	dma_mbox_cntl_poolp = hxgep->rx_mbox_cntl_pool_p;
2398 
2399 	if (!dma_buf_poolp->buf_allocated ||
2400 	    !dma_rbr_cntl_poolp->buf_allocated ||
2401 	    !dma_rcr_cntl_poolp->buf_allocated ||
2402 	    !dma_mbox_cntl_poolp->buf_allocated) {
2403 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2404 		    "<== hxge_unmap_rxdma: NULL buf pointers"));
2405 		return;
2406 	}
2407 
2408 	rx_rbr_rings = hxgep->rx_rbr_rings;
2409 	rx_rcr_rings = hxgep->rx_rcr_rings;
2410 	if (rx_rbr_rings == NULL || rx_rcr_rings == NULL) {
2411 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2412 		    "<== hxge_unmap_rxdma: NULL pointers"));
2413 		return;
2414 	}
2415 
2416 	ndmas = rx_rbr_rings->ndmas;
2417 	if (!ndmas) {
2418 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2419 		    "<== hxge_unmap_rxdma: no channel"));
2420 		return;
2421 	}
2422 
2423 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2424 	    "==> hxge_unmap_rxdma (ndmas %d)", ndmas));
2425 
2426 	rbr_rings = rx_rbr_rings->rbr_rings;
2427 	rcr_rings = rx_rcr_rings->rcr_rings;
2428 	rx_mbox_areas_p = hxgep->rx_mbox_areas_p;
2429 	rx_mbox_p = rx_mbox_areas_p->rxmbox_areas;
2430 	dma_buf_p = dma_buf_poolp->dma_buf_pool_p;
2431 
2432 	for (i = 0; i < ndmas; i++) {
2433 		channel = ((p_hxge_dma_common_t)dma_buf_p[i])->dma_channel;
2434 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2435 		    "==> hxge_unmap_rxdma (ndmas %d) channel %d",
2436 		    ndmas, channel));
2437 		(void) hxge_unmap_rxdma_channel(hxgep, channel,
2438 		    (p_rx_rbr_ring_t)rbr_rings[i],
2439 		    (p_rx_rcr_ring_t)rcr_rings[i],
2440 		    (p_rx_mbox_t)rx_mbox_p[i]);
2441 	}
2442 
2443 	KMEM_FREE(rx_rbr_rings, sizeof (rx_rbr_rings_t));
2444 	KMEM_FREE(rbr_rings, sizeof (p_rx_rbr_ring_t) * ndmas);
2445 	KMEM_FREE(rx_rcr_rings, sizeof (rx_rcr_rings_t));
2446 	KMEM_FREE(rcr_rings, sizeof (p_rx_rcr_ring_t) * ndmas);
2447 	KMEM_FREE(rx_mbox_areas_p, sizeof (rx_mbox_areas_t));
2448 	KMEM_FREE(rx_mbox_p, sizeof (p_rx_mbox_t) * ndmas);
2449 
2450 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "<== hxge_unmap_rxdma"));
2451 }
2452 
2453 hxge_status_t
2454 hxge_map_rxdma_channel(p_hxge_t hxgep, uint16_t channel,
2455     p_hxge_dma_common_t *dma_buf_p, p_rx_rbr_ring_t *rbr_p,
2456     uint32_t num_chunks, p_hxge_dma_common_t *dma_rbr_cntl_p,
2457     p_hxge_dma_common_t *dma_rcr_cntl_p, p_hxge_dma_common_t *dma_mbox_cntl_p,
2458     p_rx_rcr_ring_t *rcr_p, p_rx_mbox_t *rx_mbox_p)
2459 {
2460 	int status = HXGE_OK;
2461 
2462 	/*
2463 	 * Set up and prepare buffer blocks, descriptors and mailbox.
2464 	 */
2465 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2466 	    "==> hxge_map_rxdma_channel (channel %d)", channel));
2467 
2468 	/*
2469 	 * Receive buffer blocks
2470 	 */
2471 	status = hxge_map_rxdma_channel_buf_ring(hxgep, channel,
2472 	    dma_buf_p, rbr_p, num_chunks);
2473 	if (status != HXGE_OK) {
2474 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2475 		    "==> hxge_map_rxdma_channel (channel %d): "
2476 		    "map buffer failed 0x%x", channel, status));
2477 		goto hxge_map_rxdma_channel_exit;
2478 	}
2479 
2480 	/*
2481 	 * Receive block ring, completion ring and mailbox.
2482 	 */
2483 	status = hxge_map_rxdma_channel_cfg_ring(hxgep, channel,
2484 	    dma_rbr_cntl_p, dma_rcr_cntl_p, dma_mbox_cntl_p,
2485 	    rbr_p, rcr_p, rx_mbox_p);
2486 	if (status != HXGE_OK) {
2487 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2488 		    "==> hxge_map_rxdma_channel (channel %d): "
2489 		    "map config failed 0x%x", channel, status));
2490 		goto hxge_map_rxdma_channel_fail2;
2491 	}
2492 	goto hxge_map_rxdma_channel_exit;
2493 
2494 hxge_map_rxdma_channel_fail3:
2495 	/* Free rbr, rcr */
2496 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2497 	    "==> hxge_map_rxdma_channel: free rbr/rcr (status 0x%x channel %d)",
2498 	    status, channel));
2499 	hxge_unmap_rxdma_channel_cfg_ring(hxgep, *rcr_p, *rx_mbox_p);
2500 
2501 hxge_map_rxdma_channel_fail2:
2502 	/* Free buffer blocks */
2503 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2504 	    "==> hxge_map_rxdma_channel: free rx buffers"
2505 	    "(hxgep 0x%x status 0x%x channel %d)",
2506 	    hxgep, status, channel));
2507 	hxge_unmap_rxdma_channel_buf_ring(hxgep, *rbr_p);
2508 
2509 	status = HXGE_ERROR;
2510 
2511 hxge_map_rxdma_channel_exit:
2512 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2513 	    "<== hxge_map_rxdma_channel: (hxgep 0x%x status 0x%x channel %d)",
2514 	    hxgep, status, channel));
2515 
2516 	return (status);
2517 }
2518 
2519 /*ARGSUSED*/
2520 static void
2521 hxge_unmap_rxdma_channel(p_hxge_t hxgep, uint16_t channel,
2522     p_rx_rbr_ring_t rbr_p, p_rx_rcr_ring_t rcr_p, p_rx_mbox_t rx_mbox_p)
2523 {
2524 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2525 	    "==> hxge_unmap_rxdma_channel (channel %d)", channel));
2526 
2527 	/*
2528 	 * unmap receive block ring, completion ring and mailbox.
2529 	 */
2530 	(void) hxge_unmap_rxdma_channel_cfg_ring(hxgep, rcr_p, rx_mbox_p);
2531 
2532 	/* unmap buffer blocks */
2533 	(void) hxge_unmap_rxdma_channel_buf_ring(hxgep, rbr_p);
2534 
2535 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "<== hxge_unmap_rxdma_channel"));
2536 }
2537 
2538 /*ARGSUSED*/
2539 static hxge_status_t
2540 hxge_map_rxdma_channel_cfg_ring(p_hxge_t hxgep, uint16_t dma_channel,
2541     p_hxge_dma_common_t *dma_rbr_cntl_p, p_hxge_dma_common_t *dma_rcr_cntl_p,
2542     p_hxge_dma_common_t *dma_mbox_cntl_p, p_rx_rbr_ring_t *rbr_p,
2543     p_rx_rcr_ring_t *rcr_p, p_rx_mbox_t *rx_mbox_p)
2544 {
2545 	p_rx_rbr_ring_t		rbrp;
2546 	p_rx_rcr_ring_t		rcrp;
2547 	p_rx_mbox_t		mboxp;
2548 	p_hxge_dma_common_t	cntl_dmap;
2549 	p_hxge_dma_common_t	dmap;
2550 	p_rx_msg_t		*rx_msg_ring;
2551 	p_rx_msg_t		rx_msg_p;
2552 	rdc_rbr_cfg_a_t		*rcfga_p;
2553 	rdc_rbr_cfg_b_t		*rcfgb_p;
2554 	rdc_rcr_cfg_a_t		*cfga_p;
2555 	rdc_rcr_cfg_b_t		*cfgb_p;
2556 	rdc_rx_cfg1_t		*cfig1_p;
2557 	rdc_rx_cfg2_t		*cfig2_p;
2558 	rdc_rbr_kick_t		*kick_p;
2559 	uint32_t		dmaaddrp;
2560 	uint32_t		*rbr_vaddrp;
2561 	uint32_t		bkaddr;
2562 	hxge_status_t		status = HXGE_OK;
2563 	int			i;
2564 	uint32_t		hxge_port_rcr_size;
2565 
2566 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2567 	    "==> hxge_map_rxdma_channel_cfg_ring"));
2568 
2569 	cntl_dmap = *dma_rbr_cntl_p;
2570 
2571 	/*
2572 	 * Map in the receive block ring
2573 	 */
2574 	rbrp = *rbr_p;
2575 	dmap = (p_hxge_dma_common_t)&rbrp->rbr_desc;
2576 	hxge_setup_dma_common(dmap, cntl_dmap, rbrp->rbb_max, 4);
2577 
2578 	/*
2579 	 * Zero out buffer block ring descriptors.
2580 	 */
2581 	bzero((caddr_t)dmap->kaddrp, dmap->alength);
2582 
2583 	rcfga_p = &(rbrp->rbr_cfga);
2584 	rcfgb_p = &(rbrp->rbr_cfgb);
2585 	kick_p = &(rbrp->rbr_kick);
2586 	rcfga_p->value = 0;
2587 	rcfgb_p->value = 0;
2588 	kick_p->value = 0;
2589 	rbrp->rbr_addr = dmap->dma_cookie.dmac_laddress;
2590 	rcfga_p->value = (rbrp->rbr_addr &
2591 	    (RBR_CFIG_A_STDADDR_MASK | RBR_CFIG_A_STDADDR_BASE_MASK));
2592 	rcfga_p->value |= ((uint64_t)rbrp->rbb_max << RBR_CFIG_A_LEN_SHIFT);
2593 
2594 	/* XXXX: how to choose packet buffer sizes */
2595 	rcfgb_p->bits.bufsz0 = rbrp->pkt_buf_size0;
2596 	rcfgb_p->bits.vld0 = 1;
2597 	rcfgb_p->bits.bufsz1 = rbrp->pkt_buf_size1;
2598 	rcfgb_p->bits.vld1 = 1;
2599 	rcfgb_p->bits.bufsz2 = rbrp->pkt_buf_size2;
2600 	rcfgb_p->bits.vld2 = 1;
2601 	rcfgb_p->bits.bksize = hxgep->rx_bksize_code;
2602 
2603 	/*
2604 	 * For each buffer block, enter receive block address to the ring.
2605 	 */
2606 	rbr_vaddrp = (uint32_t *)dmap->kaddrp;
2607 	rbrp->rbr_desc_vp = (uint32_t *)dmap->kaddrp;
2608 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2609 	    "==> hxge_map_rxdma_channel_cfg_ring: channel %d "
2610 	    "rbr_vaddrp $%p", dma_channel, rbr_vaddrp));
2611 
2612 	rx_msg_ring = rbrp->rx_msg_ring;
2613 	for (i = 0; i < rbrp->tnblocks; i++) {
2614 		rx_msg_p = rx_msg_ring[i];
2615 		rx_msg_p->hxgep = hxgep;
2616 		rx_msg_p->rx_rbr_p = rbrp;
2617 		bkaddr = (uint32_t)
2618 		    ((rx_msg_p->buf_dma.dma_cookie.dmac_laddress >>
2619 		    RBR_BKADDR_SHIFT));
2620 		rx_msg_p->free = B_FALSE;
2621 		rx_msg_p->max_usage_cnt = 0xbaddcafe;
2622 
2623 		*rbr_vaddrp++ = bkaddr;
2624 	}
2625 
2626 	kick_p->bits.bkadd = rbrp->rbb_max;
2627 	rbrp->rbr_wr_index = (rbrp->rbb_max - 1);
2628 
2629 	rbrp->rbr_rd_index = 0;
2630 
2631 	rbrp->rbr_consumed = 0;
2632 	rbrp->rbr_used = 0;
2633 	rbrp->rbr_use_bcopy = B_TRUE;
2634 	rbrp->rbr_bufsize_type = RCR_PKTBUFSZ_0;
2635 
2636 	/*
2637 	 * Do bcopy on packets greater than bcopy size once the lo threshold is
2638 	 * reached. This lo threshold should be less than the hi threshold.
2639 	 *
2640 	 * Do bcopy on every packet once the hi threshold is reached.
2641 	 */
2642 	if (hxge_rx_threshold_lo >= hxge_rx_threshold_hi) {
2643 		/* default it to use hi */
2644 		hxge_rx_threshold_lo = hxge_rx_threshold_hi;
2645 	}
2646 	if (hxge_rx_buf_size_type > HXGE_RBR_TYPE2) {
2647 		hxge_rx_buf_size_type = HXGE_RBR_TYPE2;
2648 	}
2649 	rbrp->rbr_bufsize_type = hxge_rx_buf_size_type;
2650 
2651 	switch (hxge_rx_threshold_hi) {
2652 	default:
2653 	case HXGE_RX_COPY_NONE:
2654 		/* Do not do bcopy at all */
2655 		rbrp->rbr_use_bcopy = B_FALSE;
2656 		rbrp->rbr_threshold_hi = rbrp->rbb_max;
2657 		break;
2658 
2659 	case HXGE_RX_COPY_1:
2660 	case HXGE_RX_COPY_2:
2661 	case HXGE_RX_COPY_3:
2662 	case HXGE_RX_COPY_4:
2663 	case HXGE_RX_COPY_5:
2664 	case HXGE_RX_COPY_6:
2665 	case HXGE_RX_COPY_7:
2666 		rbrp->rbr_threshold_hi =
2667 		    rbrp->rbb_max * (hxge_rx_threshold_hi) /
2668 		    HXGE_RX_BCOPY_SCALE;
2669 		break;
2670 
2671 	case HXGE_RX_COPY_ALL:
2672 		rbrp->rbr_threshold_hi = 0;
2673 		break;
2674 	}
2675 
2676 	switch (hxge_rx_threshold_lo) {
2677 	default:
2678 	case HXGE_RX_COPY_NONE:
2679 		/* Do not do bcopy at all */
2680 		if (rbrp->rbr_use_bcopy) {
2681 			rbrp->rbr_use_bcopy = B_FALSE;
2682 		}
2683 		rbrp->rbr_threshold_lo = rbrp->rbb_max;
2684 		break;
2685 
2686 	case HXGE_RX_COPY_1:
2687 	case HXGE_RX_COPY_2:
2688 	case HXGE_RX_COPY_3:
2689 	case HXGE_RX_COPY_4:
2690 	case HXGE_RX_COPY_5:
2691 	case HXGE_RX_COPY_6:
2692 	case HXGE_RX_COPY_7:
2693 		rbrp->rbr_threshold_lo =
2694 		    rbrp->rbb_max * (hxge_rx_threshold_lo) /
2695 		    HXGE_RX_BCOPY_SCALE;
2696 		break;
2697 
2698 	case HXGE_RX_COPY_ALL:
2699 		rbrp->rbr_threshold_lo = 0;
2700 		break;
2701 	}
2702 
2703 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
2704 	    "hxge_map_rxdma_channel_cfg_ring: channel %d rbb_max %d "
2705 	    "rbrp->rbr_bufsize_type %d rbb_threshold_hi %d "
2706 	    "rbb_threshold_lo %d",
2707 	    dma_channel, rbrp->rbb_max, rbrp->rbr_bufsize_type,
2708 	    rbrp->rbr_threshold_hi, rbrp->rbr_threshold_lo));
2709 
2710 	/* Map in the receive completion ring */
2711 	rcrp = (p_rx_rcr_ring_t)KMEM_ZALLOC(sizeof (rx_rcr_ring_t), KM_SLEEP);
2712 	MUTEX_INIT(&rcrp->lock, NULL, MUTEX_DRIVER,
2713 	    (void *) hxgep->interrupt_cookie);
2714 	rcrp->rdc = dma_channel;
2715 	rcrp->hxgep = hxgep;
2716 
2717 	hxge_port_rcr_size = hxgep->hxge_port_rcr_size;
2718 	rcrp->comp_size = hxge_port_rcr_size;
2719 	rcrp->comp_wrap_mask = hxge_port_rcr_size - 1;
2720 
2721 	cntl_dmap = *dma_rcr_cntl_p;
2722 
2723 	dmap = (p_hxge_dma_common_t)&rcrp->rcr_desc;
2724 	hxge_setup_dma_common(dmap, cntl_dmap, rcrp->comp_size,
2725 	    sizeof (rcr_entry_t));
2726 	rcrp->comp_rd_index = 0;
2727 	rcrp->comp_wt_index = 0;
2728 	rcrp->rcr_desc_rd_head_p = rcrp->rcr_desc_first_p =
2729 	    (p_rcr_entry_t)DMA_COMMON_VPTR(rcrp->rcr_desc);
2730 	rcrp->rcr_desc_rd_head_pp = rcrp->rcr_desc_first_pp =
2731 	    (p_rcr_entry_t)DMA_COMMON_IOADDR(rcrp->rcr_desc);
2732 	rcrp->rcr_desc_last_p = rcrp->rcr_desc_rd_head_p +
2733 	    (hxge_port_rcr_size - 1);
2734 	rcrp->rcr_desc_last_pp = rcrp->rcr_desc_rd_head_pp +
2735 	    (hxge_port_rcr_size - 1);
2736 
2737 	rcrp->rcr_tail_begin = DMA_COMMON_IOADDR(rcrp->rcr_desc);
2738 	rcrp->rcr_tail_begin = (rcrp->rcr_tail_begin & 0x7ffffULL) >> 3;
2739 
2740 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2741 	    "==> hxge_map_rxdma_channel_cfg_ring: channel %d "
2742 	    "rbr_vaddrp $%p rcr_desc_rd_head_p $%p "
2743 	    "rcr_desc_rd_head_pp $%p rcr_desc_rd_last_p $%p "
2744 	    "rcr_desc_rd_last_pp $%p ",
2745 	    dma_channel, rbr_vaddrp, rcrp->rcr_desc_rd_head_p,
2746 	    rcrp->rcr_desc_rd_head_pp, rcrp->rcr_desc_last_p,
2747 	    rcrp->rcr_desc_last_pp));
2748 
2749 	/*
2750 	 * Zero out buffer block ring descriptors.
2751 	 */
2752 	bzero((caddr_t)dmap->kaddrp, dmap->alength);
2753 	rcrp->intr_timeout = hxgep->intr_timeout;
2754 	rcrp->intr_threshold = hxgep->intr_threshold;
2755 	rcrp->full_hdr_flag = B_FALSE;
2756 	rcrp->sw_priv_hdr_len = 0;
2757 
2758 	cfga_p = &(rcrp->rcr_cfga);
2759 	cfgb_p = &(rcrp->rcr_cfgb);
2760 	cfga_p->value = 0;
2761 	cfgb_p->value = 0;
2762 	rcrp->rcr_addr = dmap->dma_cookie.dmac_laddress;
2763 
2764 	cfga_p->value = (rcrp->rcr_addr &
2765 	    (RCRCFIG_A_STADDR_MASK | RCRCFIG_A_STADDR_BASE_MASK));
2766 
2767 	cfga_p->value |= ((uint64_t)rcrp->comp_size << RCRCFIG_A_LEN_SHIF);
2768 
2769 	/*
2770 	 * Timeout should be set based on the system clock divider. The
2771 	 * following timeout value of 1 assumes that the granularity (1000) is
2772 	 * 3 microseconds running at 300MHz.
2773 	 */
2774 	cfgb_p->bits.pthres = rcrp->intr_threshold;
2775 	cfgb_p->bits.timeout = rcrp->intr_timeout;
2776 	cfgb_p->bits.entout = 1;
2777 
2778 	/* Map in the mailbox */
2779 	cntl_dmap = *dma_mbox_cntl_p;
2780 	mboxp = (p_rx_mbox_t)KMEM_ZALLOC(sizeof (rx_mbox_t), KM_SLEEP);
2781 	dmap = (p_hxge_dma_common_t)&mboxp->rx_mbox;
2782 	hxge_setup_dma_common(dmap, cntl_dmap, 1, sizeof (rxdma_mailbox_t));
2783 	cfig1_p = (rdc_rx_cfg1_t *)&mboxp->rx_cfg1;
2784 	cfig2_p = (rdc_rx_cfg2_t *)&mboxp->rx_cfg2;
2785 	cfig1_p->value = cfig2_p->value = 0;
2786 
2787 	mboxp->mbox_addr = dmap->dma_cookie.dmac_laddress;
2788 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2789 	    "==> hxge_map_rxdma_channel_cfg_ring: "
2790 	    "channel %d cfg1 0x%016llx cfig2 0x%016llx cookie 0x%016llx",
2791 	    dma_channel, cfig1_p->value, cfig2_p->value,
2792 	    mboxp->mbox_addr));
2793 
2794 	dmaaddrp = (uint32_t)((dmap->dma_cookie.dmac_laddress >> 32) & 0xfff);
2795 	cfig1_p->bits.mbaddr_h = dmaaddrp;
2796 
2797 	dmaaddrp = (uint32_t)(dmap->dma_cookie.dmac_laddress & 0xffffffff);
2798 	dmaaddrp = (uint32_t)(dmap->dma_cookie.dmac_laddress &
2799 	    RXDMA_CFIG2_MBADDR_L_MASK);
2800 
2801 	cfig2_p->bits.mbaddr_l = (dmaaddrp >> RXDMA_CFIG2_MBADDR_L_SHIFT);
2802 
2803 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2804 	    "==> hxge_map_rxdma_channel_cfg_ring: channel %d damaddrp $%p "
2805 	    "cfg1 0x%016llx cfig2 0x%016llx",
2806 	    dma_channel, dmaaddrp, cfig1_p->value, cfig2_p->value));
2807 
2808 	cfig2_p->bits.full_hdr = rcrp->full_hdr_flag;
2809 	cfig2_p->bits.offset = rcrp->sw_priv_hdr_len;
2810 
2811 	rbrp->rx_rcr_p = rcrp;
2812 	rcrp->rx_rbr_p = rbrp;
2813 	*rcr_p = rcrp;
2814 	*rx_mbox_p = mboxp;
2815 
2816 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2817 	    "<== hxge_map_rxdma_channel_cfg_ring status 0x%08x", status));
2818 	return (status);
2819 }
2820 
2821 /*ARGSUSED*/
2822 static void
2823 hxge_unmap_rxdma_channel_cfg_ring(p_hxge_t hxgep,
2824     p_rx_rcr_ring_t rcr_p, p_rx_mbox_t rx_mbox_p)
2825 {
2826 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2827 	    "==> hxge_unmap_rxdma_channel_cfg_ring: channel %d", rcr_p->rdc));
2828 
2829 	MUTEX_DESTROY(&rcr_p->lock);
2830 	KMEM_FREE(rcr_p, sizeof (rx_rcr_ring_t));
2831 	KMEM_FREE(rx_mbox_p, sizeof (rx_mbox_t));
2832 
2833 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2834 	    "<== hxge_unmap_rxdma_channel_cfg_ring"));
2835 }
2836 
2837 static hxge_status_t
2838 hxge_map_rxdma_channel_buf_ring(p_hxge_t hxgep, uint16_t channel,
2839     p_hxge_dma_common_t *dma_buf_p,
2840     p_rx_rbr_ring_t *rbr_p, uint32_t num_chunks)
2841 {
2842 	p_rx_rbr_ring_t		rbrp;
2843 	p_hxge_dma_common_t	dma_bufp, tmp_bufp;
2844 	p_rx_msg_t		*rx_msg_ring;
2845 	p_rx_msg_t		rx_msg_p;
2846 	p_mblk_t		mblk_p;
2847 
2848 	rxring_info_t *ring_info;
2849 	hxge_status_t status = HXGE_OK;
2850 	int i, j, index;
2851 	uint32_t size, bsize, nblocks, nmsgs;
2852 
2853 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2854 	    "==> hxge_map_rxdma_channel_buf_ring: channel %d", channel));
2855 
2856 	dma_bufp = tmp_bufp = *dma_buf_p;
2857 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2858 	    " hxge_map_rxdma_channel_buf_ring: channel %d to map %d "
2859 	    "chunks bufp 0x%016llx", channel, num_chunks, dma_bufp));
2860 
2861 	nmsgs = 0;
2862 	for (i = 0; i < num_chunks; i++, tmp_bufp++) {
2863 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2864 		    "==> hxge_map_rxdma_channel_buf_ring: channel %d "
2865 		    "bufp 0x%016llx nblocks %d nmsgs %d",
2866 		    channel, tmp_bufp, tmp_bufp->nblocks, nmsgs));
2867 		nmsgs += tmp_bufp->nblocks;
2868 	}
2869 	if (!nmsgs) {
2870 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2871 		    "<== hxge_map_rxdma_channel_buf_ring: channel %d "
2872 		    "no msg blocks", channel));
2873 		status = HXGE_ERROR;
2874 		goto hxge_map_rxdma_channel_buf_ring_exit;
2875 	}
2876 	rbrp = (p_rx_rbr_ring_t)KMEM_ZALLOC(sizeof (rx_rbr_ring_t), KM_SLEEP);
2877 
2878 	size = nmsgs * sizeof (p_rx_msg_t);
2879 	rx_msg_ring = KMEM_ZALLOC(size, KM_SLEEP);
2880 	ring_info = (rxring_info_t *)KMEM_ZALLOC(sizeof (rxring_info_t),
2881 	    KM_SLEEP);
2882 
2883 	MUTEX_INIT(&rbrp->lock, NULL, MUTEX_DRIVER,
2884 	    (void *) hxgep->interrupt_cookie);
2885 	MUTEX_INIT(&rbrp->post_lock, NULL, MUTEX_DRIVER,
2886 	    (void *) hxgep->interrupt_cookie);
2887 
2888 	rbrp->rdc = channel;
2889 	rbrp->num_blocks = num_chunks;
2890 	rbrp->tnblocks = nmsgs;
2891 	rbrp->rbb_max = nmsgs;
2892 	rbrp->rbr_max_size = nmsgs;
2893 	rbrp->rbr_wrap_mask = (rbrp->rbb_max - 1);
2894 
2895 	/*
2896 	 * Buffer sizes: 256, 1K, and 2K.
2897 	 *
2898 	 * Blk 0 size.
2899 	 */
2900 	rbrp->pkt_buf_size0 = RBR_BUFSZ0_256B;
2901 	rbrp->pkt_buf_size0_bytes = RBR_BUFSZ0_256_BYTES;
2902 	rbrp->hpi_pkt_buf_size0 = SIZE_256B;
2903 
2904 	/*
2905 	 * Blk 1 size.
2906 	 */
2907 	rbrp->pkt_buf_size1 = RBR_BUFSZ1_1K;
2908 	rbrp->pkt_buf_size1_bytes = RBR_BUFSZ1_1K_BYTES;
2909 	rbrp->hpi_pkt_buf_size1 = SIZE_1KB;
2910 
2911 	/*
2912 	 * Blk 2 size.
2913 	 */
2914 	rbrp->pkt_buf_size2 = RBR_BUFSZ2_2K;
2915 	rbrp->pkt_buf_size2_bytes = RBR_BUFSZ2_2K_BYTES;
2916 	rbrp->hpi_pkt_buf_size2 = SIZE_2KB;
2917 
2918 	rbrp->block_size = hxgep->rx_default_block_size;
2919 
2920 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2921 	    "==> hxge_map_rxdma_channel_buf_ring: channel %d "
2922 	    "actual rbr max %d rbb_max %d nmsgs %d "
2923 	    "rbrp->block_size %d default_block_size %d "
2924 	    "(config hxge_rbr_size %d hxge_rbr_spare_size %d)",
2925 	    channel, rbrp->rbr_max_size, rbrp->rbb_max, nmsgs,
2926 	    rbrp->block_size, hxgep->rx_default_block_size,
2927 	    hxge_rbr_size, hxge_rbr_spare_size));
2928 
2929 	/*
2930 	 * Map in buffers from the buffer pool.
2931 	 * Note that num_blocks is the num_chunks. For Sparc, there is likely
2932 	 * only one chunk. For x86, there will be many chunks.
2933 	 * Loop over chunks.
2934 	 */
2935 	index = 0;
2936 	for (i = 0; i < rbrp->num_blocks; i++, dma_bufp++) {
2937 		bsize = dma_bufp->block_size;
2938 		nblocks = dma_bufp->nblocks;
2939 		ring_info->buffer[i].dvma_addr = (uint64_t)dma_bufp->ioaddr_pp;
2940 		ring_info->buffer[i].buf_index = i;
2941 		ring_info->buffer[i].buf_size = dma_bufp->alength;
2942 		ring_info->buffer[i].start_index = index;
2943 		ring_info->buffer[i].kaddr = (uint64_t)dma_bufp->kaddrp;
2944 
2945 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2946 		    " hxge_map_rxdma_channel_buf_ring: map channel %d "
2947 		    "chunk %d nblocks %d chunk_size %x block_size 0x%x "
2948 		    "dma_bufp $%p dvma_addr $%p", channel, i,
2949 		    dma_bufp->nblocks,
2950 		    ring_info->buffer[i].buf_size, bsize, dma_bufp,
2951 		    ring_info->buffer[i].dvma_addr));
2952 
2953 		/* loop over blocks within a chunk */
2954 		for (j = 0; j < nblocks; j++) {
2955 			if ((rx_msg_p = hxge_allocb(bsize, BPRI_LO,
2956 			    dma_bufp)) == NULL) {
2957 				HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2958 				    "allocb failed (index %d i %d j %d)",
2959 				    index, i, j));
2960 				goto hxge_map_rxdma_channel_buf_ring_fail1;
2961 			}
2962 			rx_msg_ring[index] = rx_msg_p;
2963 			rx_msg_p->block_index = index;
2964 			rx_msg_p->shifted_addr = (uint32_t)
2965 			    ((rx_msg_p->buf_dma.dma_cookie.dmac_laddress >>
2966 			    RBR_BKADDR_SHIFT));
2967 			/*
2968 			 * Too much output
2969 			 * HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2970 			 *	"index %d j %d rx_msg_p $%p mblk %p",
2971 			 *	index, j, rx_msg_p, rx_msg_p->rx_mblk_p));
2972 			 */
2973 			mblk_p = rx_msg_p->rx_mblk_p;
2974 			mblk_p->b_wptr = mblk_p->b_rptr + bsize;
2975 
2976 			rbrp->rbr_ref_cnt++;
2977 			index++;
2978 			rx_msg_p->buf_dma.dma_channel = channel;
2979 		}
2980 	}
2981 	if (i < rbrp->num_blocks) {
2982 		goto hxge_map_rxdma_channel_buf_ring_fail1;
2983 	}
2984 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2985 	    "hxge_map_rxdma_channel_buf_ring: done buf init "
2986 	    "channel %d msg block entries %d", channel, index));
2987 	ring_info->block_size_mask = bsize - 1;
2988 	rbrp->rx_msg_ring = rx_msg_ring;
2989 	rbrp->dma_bufp = dma_buf_p;
2990 	rbrp->ring_info = ring_info;
2991 
2992 	status = hxge_rxbuf_index_info_init(hxgep, rbrp);
2993 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, " hxge_map_rxdma_channel_buf_ring: "
2994 	    "channel %d done buf info init", channel));
2995 
2996 	/*
2997 	 * Finally, permit hxge_freeb() to call hxge_post_page().
2998 	 */
2999 	rbrp->rbr_state = RBR_POSTING;
3000 
3001 	*rbr_p = rbrp;
3002 
3003 	goto hxge_map_rxdma_channel_buf_ring_exit;
3004 
3005 hxge_map_rxdma_channel_buf_ring_fail1:
3006 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3007 	    " hxge_map_rxdma_channel_buf_ring: failed channel (0x%x)",
3008 	    channel, status));
3009 
3010 	index--;
3011 	for (; index >= 0; index--) {
3012 		rx_msg_p = rx_msg_ring[index];
3013 		if (rx_msg_p != NULL) {
3014 			freeb(rx_msg_p->rx_mblk_p);
3015 			rx_msg_ring[index] = NULL;
3016 		}
3017 	}
3018 
3019 hxge_map_rxdma_channel_buf_ring_fail:
3020 	MUTEX_DESTROY(&rbrp->post_lock);
3021 	MUTEX_DESTROY(&rbrp->lock);
3022 	KMEM_FREE(ring_info, sizeof (rxring_info_t));
3023 	KMEM_FREE(rx_msg_ring, size);
3024 	KMEM_FREE(rbrp, sizeof (rx_rbr_ring_t));
3025 
3026 	status = HXGE_ERROR;
3027 
3028 hxge_map_rxdma_channel_buf_ring_exit:
3029 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3030 	    "<== hxge_map_rxdma_channel_buf_ring status 0x%08x", status));
3031 
3032 	return (status);
3033 }
3034 
3035 /*ARGSUSED*/
3036 static void
3037 hxge_unmap_rxdma_channel_buf_ring(p_hxge_t hxgep,
3038     p_rx_rbr_ring_t rbr_p)
3039 {
3040 	p_rx_msg_t	*rx_msg_ring;
3041 	p_rx_msg_t	rx_msg_p;
3042 	rxring_info_t	*ring_info;
3043 	int		i;
3044 	uint32_t	size;
3045 
3046 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3047 	    "==> hxge_unmap_rxdma_channel_buf_ring"));
3048 	if (rbr_p == NULL) {
3049 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
3050 		    "<== hxge_unmap_rxdma_channel_buf_ring: NULL rbrp"));
3051 		return;
3052 	}
3053 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3054 	    "==> hxge_unmap_rxdma_channel_buf_ring: channel %d", rbr_p->rdc));
3055 
3056 	rx_msg_ring = rbr_p->rx_msg_ring;
3057 	ring_info = rbr_p->ring_info;
3058 
3059 	if (rx_msg_ring == NULL || ring_info == NULL) {
3060 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3061 		    "<== hxge_unmap_rxdma_channel_buf_ring: "
3062 		    "rx_msg_ring $%p ring_info $%p", rx_msg_p, ring_info));
3063 		return;
3064 	}
3065 
3066 	size = rbr_p->tnblocks * sizeof (p_rx_msg_t);
3067 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3068 	    " hxge_unmap_rxdma_channel_buf_ring: channel %d chunks %d "
3069 	    "tnblocks %d (max %d) size ptrs %d ", rbr_p->rdc, rbr_p->num_blocks,
3070 	    rbr_p->tnblocks, rbr_p->rbr_max_size, size));
3071 
3072 	for (i = 0; i < rbr_p->tnblocks; i++) {
3073 		rx_msg_p = rx_msg_ring[i];
3074 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3075 		    " hxge_unmap_rxdma_channel_buf_ring: "
3076 		    "rx_msg_p $%p", rx_msg_p));
3077 		if (rx_msg_p != NULL) {
3078 			freeb(rx_msg_p->rx_mblk_p);
3079 			rx_msg_ring[i] = NULL;
3080 		}
3081 	}
3082 
3083 	/*
3084 	 * We no longer may use the mutex <post_lock>. By setting
3085 	 * <rbr_state> to anything but POSTING, we prevent
3086 	 * hxge_post_page() from accessing a dead mutex.
3087 	 */
3088 	rbr_p->rbr_state = RBR_UNMAPPING;
3089 	MUTEX_DESTROY(&rbr_p->post_lock);
3090 
3091 	MUTEX_DESTROY(&rbr_p->lock);
3092 	KMEM_FREE(ring_info, sizeof (rxring_info_t));
3093 	KMEM_FREE(rx_msg_ring, size);
3094 
3095 	if (rbr_p->rbr_ref_cnt == 0) {
3096 		/* This is the normal state of affairs. */
3097 		KMEM_FREE(rbr_p, sizeof (*rbr_p));
3098 	} else {
3099 		/*
3100 		 * Some of our buffers are still being used.
3101 		 * Therefore, tell hxge_freeb() this ring is
3102 		 * unmapped, so it may free <rbr_p> for us.
3103 		 */
3104 		rbr_p->rbr_state = RBR_UNMAPPED;
3105 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3106 		    "unmap_rxdma_buf_ring: %d %s outstanding.",
3107 		    rbr_p->rbr_ref_cnt,
3108 		    rbr_p->rbr_ref_cnt == 1 ? "msg" : "msgs"));
3109 	}
3110 
3111 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3112 	    "<== hxge_unmap_rxdma_channel_buf_ring"));
3113 }
3114 
3115 static hxge_status_t
3116 hxge_rxdma_hw_start_common(p_hxge_t hxgep)
3117 {
3118 	hxge_status_t status = HXGE_OK;
3119 
3120 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_hw_start_common"));
3121 
3122 	/*
3123 	 * Load the sharable parameters by writing to the function zero control
3124 	 * registers. These FZC registers should be initialized only once for
3125 	 * the entire chip.
3126 	 */
3127 	(void) hxge_init_fzc_rx_common(hxgep);
3128 
3129 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_hw_start_common"));
3130 
3131 	return (status);
3132 }
3133 
3134 static hxge_status_t
3135 hxge_rxdma_hw_start(p_hxge_t hxgep)
3136 {
3137 	int			i, ndmas;
3138 	uint16_t		channel;
3139 	p_rx_rbr_rings_t	rx_rbr_rings;
3140 	p_rx_rbr_ring_t		*rbr_rings;
3141 	p_rx_rcr_rings_t	rx_rcr_rings;
3142 	p_rx_rcr_ring_t		*rcr_rings;
3143 	p_rx_mbox_areas_t	rx_mbox_areas_p;
3144 	p_rx_mbox_t		*rx_mbox_p;
3145 	hxge_status_t		status = HXGE_OK;
3146 
3147 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_hw_start"));
3148 
3149 	rx_rbr_rings = hxgep->rx_rbr_rings;
3150 	rx_rcr_rings = hxgep->rx_rcr_rings;
3151 	if (rx_rbr_rings == NULL || rx_rcr_rings == NULL) {
3152 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
3153 		    "<== hxge_rxdma_hw_start: NULL ring pointers"));
3154 		return (HXGE_ERROR);
3155 	}
3156 
3157 	ndmas = rx_rbr_rings->ndmas;
3158 	if (ndmas == 0) {
3159 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
3160 		    "<== hxge_rxdma_hw_start: no dma channel allocated"));
3161 		return (HXGE_ERROR);
3162 	}
3163 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3164 	    "==> hxge_rxdma_hw_start (ndmas %d)", ndmas));
3165 
3166 	/*
3167 	 * Scrub the RDC Rx DMA Prefetch Buffer Command.
3168 	 */
3169 	for (i = 0; i < 128; i++) {
3170 		HXGE_REG_WR64(hxgep->hpi_handle, RDC_PREF_CMD, i);
3171 	}
3172 
3173 	/*
3174 	 * Scrub Rx DMA Shadow Tail Command.
3175 	 */
3176 	for (i = 0; i < 64; i++) {
3177 		HXGE_REG_WR64(hxgep->hpi_handle, RDC_SHADOW_CMD, i);
3178 	}
3179 
3180 	/*
3181 	 * Scrub Rx DMA Control Fifo Command.
3182 	 */
3183 	for (i = 0; i < 512; i++) {
3184 		HXGE_REG_WR64(hxgep->hpi_handle, RDC_CTRL_FIFO_CMD, i);
3185 	}
3186 
3187 	/*
3188 	 * Scrub Rx DMA Data Fifo Command.
3189 	 */
3190 	for (i = 0; i < 1536; i++) {
3191 		HXGE_REG_WR64(hxgep->hpi_handle, RDC_DATA_FIFO_CMD, i);
3192 	}
3193 
3194 	/*
3195 	 * Reset the FIFO Error Stat.
3196 	 */
3197 	HXGE_REG_WR64(hxgep->hpi_handle, RDC_FIFO_ERR_STAT, 0xFF);
3198 
3199 	/* Set the error mask to receive interrupts */
3200 	HXGE_REG_WR64(hxgep->hpi_handle, RDC_FIFO_ERR_INT_MASK, 0x0);
3201 
3202 	rbr_rings = rx_rbr_rings->rbr_rings;
3203 	rcr_rings = rx_rcr_rings->rcr_rings;
3204 	rx_mbox_areas_p = hxgep->rx_mbox_areas_p;
3205 	if (rx_mbox_areas_p) {
3206 		rx_mbox_p = rx_mbox_areas_p->rxmbox_areas;
3207 	}
3208 
3209 	for (i = 0; i < ndmas; i++) {
3210 		channel = rbr_rings[i]->rdc;
3211 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3212 		    "==> hxge_rxdma_hw_start (ndmas %d) channel %d",
3213 		    ndmas, channel));
3214 		status = hxge_rxdma_start_channel(hxgep, channel,
3215 		    (p_rx_rbr_ring_t)rbr_rings[i],
3216 		    (p_rx_rcr_ring_t)rcr_rings[i],
3217 		    (p_rx_mbox_t)rx_mbox_p[i], rbr_rings[i]->rbb_max);
3218 		if (status != HXGE_OK) {
3219 			goto hxge_rxdma_hw_start_fail1;
3220 		}
3221 	}
3222 
3223 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_hw_start: "
3224 	    "rx_rbr_rings 0x%016llx rings 0x%016llx",
3225 	    rx_rbr_rings, rx_rcr_rings));
3226 	goto hxge_rxdma_hw_start_exit;
3227 
3228 hxge_rxdma_hw_start_fail1:
3229 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3230 	    "==> hxge_rxdma_hw_start: disable "
3231 	    "(status 0x%x channel %d i %d)", status, channel, i));
3232 	for (; i >= 0; i--) {
3233 		channel = rbr_rings[i]->rdc;
3234 		(void) hxge_rxdma_stop_channel(hxgep, channel);
3235 	}
3236 
3237 hxge_rxdma_hw_start_exit:
3238 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3239 	    "==> hxge_rxdma_hw_start: (status 0x%x)", status));
3240 	return (status);
3241 }
3242 
3243 static void
3244 hxge_rxdma_hw_stop(p_hxge_t hxgep)
3245 {
3246 	int			i, ndmas;
3247 	uint16_t		channel;
3248 	p_rx_rbr_rings_t	rx_rbr_rings;
3249 	p_rx_rbr_ring_t		*rbr_rings;
3250 	p_rx_rcr_rings_t	rx_rcr_rings;
3251 
3252 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_hw_stop"));
3253 
3254 	rx_rbr_rings = hxgep->rx_rbr_rings;
3255 	rx_rcr_rings = hxgep->rx_rcr_rings;
3256 
3257 	if (rx_rbr_rings == NULL || rx_rcr_rings == NULL) {
3258 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
3259 		    "<== hxge_rxdma_hw_stop: NULL ring pointers"));
3260 		return;
3261 	}
3262 
3263 	ndmas = rx_rbr_rings->ndmas;
3264 	if (!ndmas) {
3265 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
3266 		    "<== hxge_rxdma_hw_stop: no dma channel allocated"));
3267 		return;
3268 	}
3269 
3270 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3271 	    "==> hxge_rxdma_hw_stop (ndmas %d)", ndmas));
3272 
3273 	rbr_rings = rx_rbr_rings->rbr_rings;
3274 	for (i = 0; i < ndmas; i++) {
3275 		channel = rbr_rings[i]->rdc;
3276 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3277 		    "==> hxge_rxdma_hw_stop (ndmas %d) channel %d",
3278 		    ndmas, channel));
3279 		(void) hxge_rxdma_stop_channel(hxgep, channel);
3280 	}
3281 
3282 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_hw_stop: "
3283 	    "rx_rbr_rings 0x%016llx rings 0x%016llx",
3284 	    rx_rbr_rings, rx_rcr_rings));
3285 
3286 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "<== hxge_rxdma_hw_stop"));
3287 }
3288 
3289 static hxge_status_t
3290 hxge_rxdma_start_channel(p_hxge_t hxgep, uint16_t channel,
3291     p_rx_rbr_ring_t rbr_p, p_rx_rcr_ring_t rcr_p, p_rx_mbox_t mbox_p,
3292     int n_init_kick)
3293 {
3294 	hpi_handle_t		handle;
3295 	hpi_status_t		rs = HPI_SUCCESS;
3296 	rdc_stat_t		cs;
3297 	rdc_int_mask_t		ent_mask;
3298 	hxge_status_t		status = HXGE_OK;
3299 
3300 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_start_channel"));
3301 
3302 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
3303 
3304 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "hxge_rxdma_start_channel: "
3305 	    "hpi handle addr $%p acc $%p",
3306 	    hxgep->hpi_handle.regp, hxgep->hpi_handle.regh));
3307 
3308 	/* Reset RXDMA channel */
3309 	rs = hpi_rxdma_cfg_rdc_reset(handle, channel);
3310 	if (rs != HPI_SUCCESS) {
3311 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3312 		    "==> hxge_rxdma_start_channel: "
3313 		    "reset rxdma failed (0x%08x channel %d)",
3314 		    status, channel));
3315 		return (HXGE_ERROR | rs);
3316 	}
3317 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3318 	    "==> hxge_rxdma_start_channel: reset done: channel %d", channel));
3319 
3320 	/*
3321 	 * Initialize the RXDMA channel specific FZC control configurations.
3322 	 * These FZC registers are pertaining to each RX channel (logical
3323 	 * pages).
3324 	 */
3325 	status = hxge_init_fzc_rxdma_channel(hxgep,
3326 	    channel, rbr_p, rcr_p, mbox_p);
3327 	if (status != HXGE_OK) {
3328 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3329 		    "==> hxge_rxdma_start_channel: "
3330 		    "init fzc rxdma failed (0x%08x channel %d)",
3331 		    status, channel));
3332 		return (status);
3333 	}
3334 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3335 	    "==> hxge_rxdma_start_channel: fzc done"));
3336 
3337 	/*
3338 	 * Zero out the shadow  and prefetch ram.
3339 	 */
3340 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3341 	    "==> hxge_rxdma_start_channel: ram done"));
3342 
3343 	/* Set up the interrupt event masks. */
3344 	ent_mask.value = 0;
3345 	rs = hpi_rxdma_event_mask(handle, OP_SET, channel, &ent_mask);
3346 	if (rs != HPI_SUCCESS) {
3347 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3348 		    "==> hxge_rxdma_start_channel: "
3349 		    "init rxdma event masks failed (0x%08x channel %d)",
3350 		    status, channel));
3351 		return (HXGE_ERROR | rs);
3352 	}
3353 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_start_channel: "
3354 	    "event done: channel %d (mask 0x%016llx)",
3355 	    channel, ent_mask.value));
3356 
3357 	/*
3358 	 * Load RXDMA descriptors, buffers, mailbox, initialise the receive DMA
3359 	 * channels and enable each DMA channel.
3360 	 */
3361 	status = hxge_enable_rxdma_channel(hxgep,
3362 	    channel, rbr_p, rcr_p, mbox_p, n_init_kick);
3363 	if (status != HXGE_OK) {
3364 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3365 		    " hxge_rxdma_start_channel: "
3366 		    " init enable rxdma failed (0x%08x channel %d)",
3367 		    status, channel));
3368 		return (status);
3369 	}
3370 
3371 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_start_channel: "
3372 	    "control done - channel %d cs 0x%016llx", channel, cs.value));
3373 
3374 	/*
3375 	 * Initialize the receive DMA control and status register
3376 	 * Note that rdc_stat HAS to be set after RBR and RCR rings are set
3377 	 */
3378 	cs.value = 0;
3379 	cs.bits.mex = 1;
3380 	cs.bits.rcr_thres = 1;
3381 	cs.bits.rcr_to = 1;
3382 	cs.bits.rbr_empty = 1;
3383 	status = hxge_init_rxdma_channel_cntl_stat(hxgep, channel, &cs);
3384 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_start_channel: "
3385 	    "channel %d rx_dma_cntl_stat 0x%0016llx", channel, cs.value));
3386 	if (status != HXGE_OK) {
3387 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3388 		    "==> hxge_rxdma_start_channel: "
3389 		    "init rxdma control register failed (0x%08x channel %d",
3390 		    status, channel));
3391 		return (status);
3392 	}
3393 
3394 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_start_channel: "
3395 	    "control done - channel %d cs 0x%016llx", channel, cs.value));
3396 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3397 	    "==> hxge_rxdma_start_channel: enable done"));
3398 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "<== hxge_rxdma_start_channel"));
3399 	return (HXGE_OK);
3400 }
3401 
3402 static hxge_status_t
3403 hxge_rxdma_stop_channel(p_hxge_t hxgep, uint16_t channel)
3404 {
3405 	hpi_handle_t		handle;
3406 	hpi_status_t		rs = HPI_SUCCESS;
3407 	rdc_stat_t		cs;
3408 	rdc_int_mask_t		ent_mask;
3409 	hxge_status_t		status = HXGE_OK;
3410 
3411 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rxdma_stop_channel"));
3412 
3413 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
3414 
3415 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "hxge_rxdma_stop_channel: "
3416 	    "hpi handle addr $%p acc $%p",
3417 	    hxgep->hpi_handle.regp, hxgep->hpi_handle.regh));
3418 
3419 	/* Reset RXDMA channel */
3420 	rs = hpi_rxdma_cfg_rdc_reset(handle, channel);
3421 	if (rs != HPI_SUCCESS) {
3422 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3423 		    " hxge_rxdma_stop_channel: "
3424 		    " reset rxdma failed (0x%08x channel %d)",
3425 		    rs, channel));
3426 		return (HXGE_ERROR | rs);
3427 	}
3428 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
3429 	    "==> hxge_rxdma_stop_channel: reset done"));
3430 
3431 	/* Set up the interrupt event masks. */
3432 	ent_mask.value = RDC_INT_MASK_ALL;
3433 	rs = hpi_rxdma_event_mask(handle, OP_SET, channel, &ent_mask);
3434 	if (rs != HPI_SUCCESS) {
3435 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3436 		    "==> hxge_rxdma_stop_channel: "
3437 		    "set rxdma event masks failed (0x%08x channel %d)",
3438 		    rs, channel));
3439 		return (HXGE_ERROR | rs);
3440 	}
3441 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
3442 	    "==> hxge_rxdma_stop_channel: event done"));
3443 
3444 	/* Initialize the receive DMA control and status register */
3445 	cs.value = 0;
3446 	status = hxge_init_rxdma_channel_cntl_stat(hxgep, channel, &cs);
3447 
3448 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rxdma_stop_channel: control "
3449 	    " to default (all 0s) 0x%08x", cs.value));
3450 
3451 	if (status != HXGE_OK) {
3452 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3453 		    " hxge_rxdma_stop_channel: init rxdma"
3454 		    " control register failed (0x%08x channel %d",
3455 		    status, channel));
3456 		return (status);
3457 	}
3458 
3459 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
3460 	    "==> hxge_rxdma_stop_channel: control done"));
3461 
3462 	/* disable dma channel */
3463 	status = hxge_disable_rxdma_channel(hxgep, channel);
3464 
3465 	if (status != HXGE_OK) {
3466 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3467 		    " hxge_rxdma_stop_channel: "
3468 		    " init enable rxdma failed (0x%08x channel %d)",
3469 		    status, channel));
3470 		return (status);
3471 	}
3472 
3473 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
3474 	    "==> hxge_rxdma_stop_channel: disable done"));
3475 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_rxdma_stop_channel"));
3476 
3477 	return (HXGE_OK);
3478 }
3479 
3480 hxge_status_t
3481 hxge_rxdma_handle_sys_errors(p_hxge_t hxgep)
3482 {
3483 	hpi_handle_t		handle;
3484 	p_hxge_rdc_sys_stats_t	statsp;
3485 	rdc_fifo_err_stat_t	stat;
3486 	hxge_status_t		status = HXGE_OK;
3487 
3488 	handle = hxgep->hpi_handle;
3489 	statsp = (p_hxge_rdc_sys_stats_t)&hxgep->statsp->rdc_sys_stats;
3490 
3491 	/* Get the error status and clear the register */
3492 	HXGE_REG_RD64(handle, RDC_FIFO_ERR_STAT, &stat.value);
3493 	HXGE_REG_WR64(handle, RDC_FIFO_ERR_STAT, stat.value);
3494 
3495 	if (stat.bits.rx_ctrl_fifo_sec) {
3496 		statsp->ctrl_fifo_sec++;
3497 		if (statsp->ctrl_fifo_sec == 1)
3498 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3499 			    "==> hxge_rxdma_handle_sys_errors: "
3500 			    "rx_ctrl_fifo_sec"));
3501 	}
3502 
3503 	if (stat.bits.rx_ctrl_fifo_ded) {
3504 		/* Global fatal error encountered */
3505 		statsp->ctrl_fifo_ded++;
3506 		HXGE_FM_REPORT_ERROR(hxgep, 0,
3507 		    HXGE_FM_EREPORT_RDMC_CTRL_FIFO_DED);
3508 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3509 		    "==> hxge_rxdma_handle_sys_errors: "
3510 		    "fatal error: rx_ctrl_fifo_ded error"));
3511 	}
3512 
3513 	if (stat.bits.rx_data_fifo_sec) {
3514 		statsp->data_fifo_sec++;
3515 		if (statsp->data_fifo_sec == 1)
3516 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3517 			    "==> hxge_rxdma_handle_sys_errors: "
3518 			    "rx_data_fifo_sec"));
3519 	}
3520 
3521 	if (stat.bits.rx_data_fifo_ded) {
3522 		/* Global fatal error encountered */
3523 		statsp->data_fifo_ded++;
3524 		HXGE_FM_REPORT_ERROR(hxgep, 0,
3525 		    HXGE_FM_EREPORT_RDMC_DATA_FIFO_DED);
3526 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3527 		    "==> hxge_rxdma_handle_sys_errors: "
3528 		    "fatal error: rx_data_fifo_ded error"));
3529 	}
3530 
3531 	if (stat.bits.rx_ctrl_fifo_ded || stat.bits.rx_data_fifo_ded) {
3532 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3533 		    " hxge_rxdma_handle_sys_errors: fatal error\n"));
3534 		status = hxge_rx_port_fatal_err_recover(hxgep);
3535 		if (status == HXGE_OK) {
3536 			FM_SERVICE_RESTORED(hxgep);
3537 		}
3538 	}
3539 
3540 	return (HXGE_OK);
3541 }
3542 
3543 static hxge_status_t
3544 hxge_rxdma_fatal_err_recover(p_hxge_t hxgep, uint16_t channel)
3545 {
3546 	hpi_handle_t		handle;
3547 	hpi_status_t		rs = HPI_SUCCESS;
3548 	p_rx_rbr_ring_t		rbrp;
3549 	p_rx_rcr_ring_t		rcrp;
3550 	p_rx_mbox_t		mboxp;
3551 	rdc_int_mask_t		ent_mask;
3552 	p_hxge_dma_common_t	dmap;
3553 	p_rx_msg_t		rx_msg_p;
3554 	int			i;
3555 	uint32_t		hxge_port_rcr_size;
3556 	uint64_t		tmp;
3557 	int			n_init_kick = 0;
3558 
3559 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rxdma_fatal_err_recover"));
3560 
3561 	/*
3562 	 * Stop the dma channel waits for the stop done. If the stop done bit
3563 	 * is not set, then create an error.
3564 	 */
3565 
3566 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
3567 
3568 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Rx DMA stop..."));
3569 
3570 	rbrp = (p_rx_rbr_ring_t)hxgep->rx_rbr_rings->rbr_rings[channel];
3571 	rcrp = (p_rx_rcr_ring_t)hxgep->rx_rcr_rings->rcr_rings[channel];
3572 
3573 	MUTEX_ENTER(&rcrp->lock);
3574 	MUTEX_ENTER(&rbrp->lock);
3575 
3576 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Disable RxDMA channel..."));
3577 
3578 	rs = hpi_rxdma_cfg_rdc_disable(handle, channel);
3579 	if (rs != HPI_SUCCESS) {
3580 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3581 		    "hxge_disable_rxdma_channel:failed"));
3582 		goto fail;
3583 	}
3584 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Disable RxDMA interrupt..."));
3585 
3586 	/* Disable interrupt */
3587 	ent_mask.value = RDC_INT_MASK_ALL;
3588 	rs = hpi_rxdma_event_mask(handle, OP_SET, channel, &ent_mask);
3589 	if (rs != HPI_SUCCESS) {
3590 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3591 		    "Set rxdma event masks failed (channel %d)", channel));
3592 	}
3593 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "RxDMA channel reset..."));
3594 
3595 	/* Reset RXDMA channel */
3596 	rs = hpi_rxdma_cfg_rdc_reset(handle, channel);
3597 	if (rs != HPI_SUCCESS) {
3598 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3599 		    "Reset rxdma failed (channel %d)", channel));
3600 		goto fail;
3601 	}
3602 	hxge_port_rcr_size = hxgep->hxge_port_rcr_size;
3603 	mboxp = (p_rx_mbox_t)hxgep->rx_mbox_areas_p->rxmbox_areas[channel];
3604 
3605 	rbrp->rbr_wr_index = (rbrp->rbb_max - 1);
3606 	rbrp->rbr_rd_index = 0;
3607 
3608 	rcrp->comp_rd_index = 0;
3609 	rcrp->comp_wt_index = 0;
3610 	rcrp->rcr_desc_rd_head_p = rcrp->rcr_desc_first_p =
3611 	    (p_rcr_entry_t)DMA_COMMON_VPTR(rcrp->rcr_desc);
3612 	rcrp->rcr_desc_rd_head_pp = rcrp->rcr_desc_first_pp =
3613 	    (p_rcr_entry_t)DMA_COMMON_IOADDR(rcrp->rcr_desc);
3614 
3615 	rcrp->rcr_desc_last_p = rcrp->rcr_desc_rd_head_p +
3616 	    (hxge_port_rcr_size - 1);
3617 	rcrp->rcr_desc_last_pp = rcrp->rcr_desc_rd_head_pp +
3618 	    (hxge_port_rcr_size - 1);
3619 
3620 	rcrp->rcr_tail_begin = DMA_COMMON_IOADDR(rcrp->rcr_desc);
3621 	rcrp->rcr_tail_begin = (rcrp->rcr_tail_begin & 0x7ffffULL) >> 3;
3622 
3623 	dmap = (p_hxge_dma_common_t)&rcrp->rcr_desc;
3624 	bzero((caddr_t)dmap->kaddrp, dmap->alength);
3625 
3626 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "rbr entries = %d\n",
3627 	    rbrp->rbr_max_size));
3628 
3629 	/* Count the number of buffers owned by the hardware at this moment */
3630 	for (i = 0; i < rbrp->rbr_max_size; i++) {
3631 		rx_msg_p = rbrp->rx_msg_ring[i];
3632 		if (rx_msg_p->ref_cnt == 1) {
3633 			n_init_kick++;
3634 		}
3635 	}
3636 
3637 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "RxDMA channel re-start..."));
3638 
3639 	/*
3640 	 * This is error recover! Some buffers are owned by the hardware and
3641 	 * the rest are owned by the apps. We should only kick in those
3642 	 * owned by the hardware initially. The apps will post theirs
3643 	 * eventually.
3644 	 */
3645 	(void) hxge_rxdma_start_channel(hxgep, channel, rbrp, rcrp, mboxp,
3646 	    n_init_kick);
3647 
3648 	/*
3649 	 * The DMA channel may disable itself automatically.
3650 	 * The following is a work-around.
3651 	 */
3652 	HXGE_REG_RD64(handle, RDC_RX_CFG1, &tmp);
3653 	rs = hpi_rxdma_cfg_rdc_enable(handle, channel);
3654 	if (rs != HPI_SUCCESS) {
3655 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3656 		    "hpi_rxdma_cfg_rdc_enable (channel %d)", channel));
3657 	}
3658 
3659 	/*
3660 	 * Delay a bit of time by doing reads.
3661 	 */
3662 	for (i = 0; i < 1024; i++) {
3663 		uint64_t value;
3664 		RXDMA_REG_READ64(HXGE_DEV_HPI_HANDLE(hxgep),
3665 		    RDC_INT_MASK, i & 3, &value);
3666 	}
3667 
3668 	MUTEX_EXIT(&rbrp->lock);
3669 	MUTEX_EXIT(&rcrp->lock);
3670 
3671 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_rxdma_fatal_err_recover"));
3672 	return (HXGE_OK);
3673 
3674 fail:
3675 	MUTEX_EXIT(&rbrp->lock);
3676 	MUTEX_EXIT(&rcrp->lock);
3677 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3678 	    "Error Recovery failed for channel(%d)", channel));
3679 	return (HXGE_ERROR | rs);
3680 }
3681 
3682 static hxge_status_t
3683 hxge_rx_port_fatal_err_recover(p_hxge_t hxgep)
3684 {
3685 	hxge_status_t		status = HXGE_OK;
3686 	p_hxge_dma_common_t	*dma_buf_p;
3687 	uint16_t		channel;
3688 	int			ndmas;
3689 	int			i;
3690 	block_reset_t		reset_reg;
3691 	p_rx_rcr_ring_t	rcrp;
3692 	p_rx_rbr_ring_t rbrp;
3693 
3694 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rx_port_fatal_err_recover"));
3695 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL, "Recovering from RDC error ..."));
3696 
3697 	/* Disable RxMAC */
3698 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Disable RxMAC...\n"));
3699 	MUTEX_ENTER(&hxgep->vmac_lock);
3700 	if (hxge_rx_vmac_disable(hxgep) != HXGE_OK)
3701 		goto fail;
3702 
3703 	HXGE_DELAY(1000);
3704 
3705 	/*
3706 	 * Reset RDC block from PEU for this fatal error
3707 	 */
3708 	reset_reg.value = 0;
3709 	reset_reg.bits.rdc_rst = 1;
3710 	HXGE_REG_WR32(hxgep->hpi_handle, BLOCK_RESET, reset_reg.value);
3711 
3712 	HXGE_DELAY(1000);
3713 
3714 	/* Restore any common settings after PEU reset */
3715 	if (hxge_rxdma_hw_start_common(hxgep) != HXGE_OK)
3716 		goto fail;
3717 
3718 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Stop all RxDMA channels..."));
3719 
3720 	ndmas = hxgep->rx_buf_pool_p->ndmas;
3721 	dma_buf_p = hxgep->rx_buf_pool_p->dma_buf_pool_p;
3722 
3723 	for (i = 0; i < ndmas; i++) {
3724 		channel = ((p_hxge_dma_common_t)dma_buf_p[i])->dma_channel;
3725 		rcrp = hxgep->rx_rcr_rings->rcr_rings[channel];
3726 		rbrp = rcrp->rx_rbr_p;
3727 
3728 		MUTEX_ENTER(&rbrp->post_lock);
3729 
3730 		/*
3731 		 * This function needs to be inside the post_lock
3732 		 */
3733 		if (hxge_rxdma_fatal_err_recover(hxgep, channel) != HXGE_OK) {
3734 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3735 			    "Could not recover channel %d", channel));
3736 		}
3737 		MUTEX_EXIT(&rbrp->post_lock);
3738 	}
3739 
3740 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Reset RxMAC..."));
3741 
3742 	/* Reset RxMAC */
3743 	if (hxge_rx_vmac_reset(hxgep) != HXGE_OK) {
3744 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3745 		    "hxge_rx_port_fatal_err_recover: Failed to reset RxMAC"));
3746 		goto fail;
3747 	}
3748 
3749 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Re-initialize RxMAC..."));
3750 
3751 	/* Re-Initialize RxMAC */
3752 	if ((status = hxge_rx_vmac_init(hxgep)) != HXGE_OK) {
3753 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3754 		    "hxge_rx_port_fatal_err_recover: Failed to reset RxMAC"));
3755 		goto fail;
3756 	}
3757 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Re-enable RxMAC..."));
3758 
3759 	/* Re-enable RxMAC */
3760 	if ((status = hxge_rx_vmac_enable(hxgep)) != HXGE_OK) {
3761 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3762 		    "hxge_rx_port_fatal_err_recover: Failed to enable RxMAC"));
3763 		goto fail;
3764 	}
3765 	MUTEX_EXIT(&hxgep->vmac_lock);
3766 
3767 	/* Reset the error mask since PEU reset cleared it */
3768 	HXGE_REG_WR64(hxgep->hpi_handle, RDC_FIFO_ERR_INT_MASK, 0x0);
3769 
3770 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3771 	    "Recovery Successful, RxPort Restored"));
3772 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_rx_port_fatal_err_recover"));
3773 	return (HXGE_OK);
3774 
3775 fail:
3776 	MUTEX_EXIT(&hxgep->vmac_lock);
3777 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3778 	    "Error Recovery failed for hxge(%d)", hxgep->instance));
3779 	return (status);
3780 }
3781 
3782 static void
3783 hxge_rbr_empty_restore(p_hxge_t hxgep, p_rx_rbr_ring_t rx_rbr_p)
3784 {
3785 	hpi_status_t		hpi_status;
3786 	hxge_status_t		status;
3787 	rdc_stat_t		cs;
3788 	p_hxge_rx_ring_stats_t	rdc_stats;
3789 
3790 	rdc_stats = &hxgep->statsp->rdc_stats[rx_rbr_p->rdc];
3791 
3792 	/*
3793 	 * Complete the processing for the RBR Empty by:
3794 	 *	0) kicking back HXGE_RBR_EMPTY_THRESHOLD
3795 	 *	   packets.
3796 	 *	1) Disable the RX vmac.
3797 	 *	2) Re-enable the affected DMA channel.
3798 	 *	3) Re-enable the RX vmac.
3799 	 */
3800 
3801 	/*
3802 	 * Disable the RX VMAC, but setting the framelength
3803 	 * to 0, since there is a hardware bug when disabling
3804 	 * the vmac.
3805 	 */
3806 	MUTEX_ENTER(&hxgep->vmac_lock);
3807 	(void) hxge_rx_vmac_disable(hxgep);
3808 
3809 	/*
3810 	 * Re-arm the mex bit for interrupts to be enabled.
3811 	 */
3812 	cs.value = 0;
3813 	cs.bits.mex = 1;
3814 	RXDMA_REG_WRITE64(HXGE_DEV_HPI_HANDLE(hxgep), RDC_STAT,
3815 	    rx_rbr_p->rdc, cs.value);
3816 
3817 	hpi_status = hpi_rxdma_cfg_rdc_enable(
3818 	    HXGE_DEV_HPI_HANDLE(hxgep), rx_rbr_p->rdc);
3819 	if (hpi_status != HPI_SUCCESS) {
3820 		rdc_stats->rbr_empty_fail++;
3821 
3822 		/* Assume we are already inside the post_lock */
3823 		status = hxge_rxdma_fatal_err_recover(hxgep, rx_rbr_p->rdc);
3824 		if (status != HXGE_OK) {
3825 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3826 			    "hxge(%d): channel(%d) is empty.",
3827 			    hxgep->instance, rx_rbr_p->rdc));
3828 		}
3829 	}
3830 
3831 	/*
3832 	 * Re-enable the RX VMAC.
3833 	 */
3834 	(void) hxge_rx_vmac_enable(hxgep);
3835 	MUTEX_EXIT(&hxgep->vmac_lock);
3836 
3837 	rdc_stats->rbr_empty_restore++;
3838 	rx_rbr_p->rbr_is_empty = B_FALSE;
3839 }
3840