xref: /titanic_51/usr/src/uts/common/io/hxge/hxge_rxdma.c (revision 5c066ec28ea93f3a7c93082611a61747f255290a)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #include <hxge_impl.h>
27 #include <hxge_rxdma.h>
28 
29 /*
30  * Number of blocks to accumulate before re-enabling DMA
31  * when we get RBR empty.
32  */
33 #define	HXGE_RBR_EMPTY_THRESHOLD	64
34 
35 /*
36  * Globals: tunable parameters (/etc/system or adb)
37  *
38  */
39 extern uint32_t hxge_rbr_size;
40 extern uint32_t hxge_rcr_size;
41 extern uint32_t hxge_rbr_spare_size;
42 extern uint32_t hxge_mblks_pending;
43 
44 /*
45  * Tunable to reduce the amount of time spent in the
46  * ISR doing Rx Processing.
47  */
48 extern uint32_t hxge_max_rx_pkts;
49 
50 /*
51  * Tunables to manage the receive buffer blocks.
52  *
53  * hxge_rx_threshold_hi: copy all buffers.
54  * hxge_rx_bcopy_size_type: receive buffer block size type.
55  * hxge_rx_threshold_lo: copy only up to tunable block size type.
56  */
57 extern hxge_rxbuf_threshold_t hxge_rx_threshold_hi;
58 extern hxge_rxbuf_type_t hxge_rx_buf_size_type;
59 extern hxge_rxbuf_threshold_t hxge_rx_threshold_lo;
60 
61 /*
62  * Static local functions.
63  */
64 static hxge_status_t hxge_map_rxdma(p_hxge_t hxgep);
65 static void hxge_unmap_rxdma(p_hxge_t hxgep);
66 static hxge_status_t hxge_rxdma_hw_start_common(p_hxge_t hxgep);
67 static hxge_status_t hxge_rxdma_hw_start(p_hxge_t hxgep);
68 static void hxge_rxdma_hw_stop(p_hxge_t hxgep);
69 static hxge_status_t hxge_map_rxdma_channel(p_hxge_t hxgep, uint16_t channel,
70     p_hxge_dma_common_t *dma_buf_p, p_rx_rbr_ring_t *rbr_p,
71     uint32_t num_chunks, p_hxge_dma_common_t *dma_rbr_cntl_p,
72     p_hxge_dma_common_t *dma_rcr_cntl_p, p_hxge_dma_common_t *dma_mbox_cntl_p,
73     p_rx_rcr_ring_t *rcr_p, p_rx_mbox_t *rx_mbox_p);
74 static void hxge_unmap_rxdma_channel(p_hxge_t hxgep, uint16_t channel,
75 	p_rx_rbr_ring_t rbr_p, p_rx_rcr_ring_t rcr_p, p_rx_mbox_t rx_mbox_p);
76 static hxge_status_t hxge_map_rxdma_channel_cfg_ring(p_hxge_t hxgep,
77     uint16_t dma_channel, p_hxge_dma_common_t *dma_rbr_cntl_p,
78     p_hxge_dma_common_t *dma_rcr_cntl_p, p_hxge_dma_common_t *dma_mbox_cntl_p,
79     p_rx_rbr_ring_t *rbr_p, p_rx_rcr_ring_t *rcr_p, p_rx_mbox_t *rx_mbox_p);
80 static void hxge_unmap_rxdma_channel_cfg_ring(p_hxge_t hxgep,
81 	p_rx_rcr_ring_t rcr_p, p_rx_mbox_t rx_mbox_p);
82 static hxge_status_t hxge_map_rxdma_channel_buf_ring(p_hxge_t hxgep,
83 	uint16_t channel, p_hxge_dma_common_t *dma_buf_p,
84 	p_rx_rbr_ring_t *rbr_p, uint32_t num_chunks);
85 static void hxge_unmap_rxdma_channel_buf_ring(p_hxge_t hxgep,
86 	p_rx_rbr_ring_t rbr_p);
87 static hxge_status_t hxge_rxdma_start_channel(p_hxge_t hxgep, uint16_t channel,
88 	p_rx_rbr_ring_t rbr_p, p_rx_rcr_ring_t rcr_p, p_rx_mbox_t mbox_p,
89 	int n_init_kick);
90 static hxge_status_t hxge_rxdma_stop_channel(p_hxge_t hxgep, uint16_t channel);
91 static mblk_t *hxge_rx_pkts(p_hxge_t hxgep, uint_t vindex, p_hxge_ldv_t ldvp,
92 	p_rx_rcr_ring_t	*rcr_p, rdc_stat_t cs);
93 static void hxge_receive_packet(p_hxge_t hxgep, p_rx_rcr_ring_t rcr_p,
94 	p_rcr_entry_t rcr_desc_rd_head_p, boolean_t *multi_p,
95 	mblk_t ** mp, mblk_t ** mp_cont, uint32_t *invalid_rcr_entry);
96 static hxge_status_t hxge_disable_rxdma_channel(p_hxge_t hxgep,
97 	uint16_t channel);
98 static p_rx_msg_t hxge_allocb(size_t, uint32_t, p_hxge_dma_common_t);
99 static void hxge_freeb(p_rx_msg_t);
100 static void hxge_rx_pkts_vring(p_hxge_t hxgep, uint_t vindex,
101 	p_hxge_ldv_t ldvp, rdc_stat_t cs);
102 static hxge_status_t hxge_rx_err_evnts(p_hxge_t hxgep, uint_t index,
103 	p_hxge_ldv_t ldvp, rdc_stat_t cs);
104 static hxge_status_t hxge_rxbuf_index_info_init(p_hxge_t hxgep,
105 	p_rx_rbr_ring_t rx_dmap);
106 static hxge_status_t hxge_rxdma_fatal_err_recover(p_hxge_t hxgep,
107 	uint16_t channel);
108 static hxge_status_t hxge_rx_port_fatal_err_recover(p_hxge_t hxgep);
109 static void hxge_rbr_empty_restore(p_hxge_t hxgep,
110 	p_rx_rbr_ring_t rx_rbr_p);
111 
112 hxge_status_t
113 hxge_init_rxdma_channels(p_hxge_t hxgep)
114 {
115 	hxge_status_t		status = HXGE_OK;
116 	block_reset_t		reset_reg;
117 	int			i;
118 
119 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_init_rxdma_channels"));
120 
121 	for (i = 0; i < HXGE_MAX_RDCS; i++)
122 		hxgep->rdc_first_intr[i] = B_TRUE;
123 
124 	/* Reset RDC block from PEU to clear any previous state */
125 	reset_reg.value = 0;
126 	reset_reg.bits.rdc_rst = 1;
127 	HXGE_REG_WR32(hxgep->hpi_handle, BLOCK_RESET, reset_reg.value);
128 	HXGE_DELAY(1000);
129 
130 	status = hxge_map_rxdma(hxgep);
131 	if (status != HXGE_OK) {
132 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
133 		    "<== hxge_init_rxdma: status 0x%x", status));
134 		return (status);
135 	}
136 
137 	status = hxge_rxdma_hw_start_common(hxgep);
138 	if (status != HXGE_OK) {
139 		hxge_unmap_rxdma(hxgep);
140 	}
141 
142 	status = hxge_rxdma_hw_start(hxgep);
143 	if (status != HXGE_OK) {
144 		hxge_unmap_rxdma(hxgep);
145 	}
146 
147 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
148 	    "<== hxge_init_rxdma_channels: status 0x%x", status));
149 	return (status);
150 }
151 
152 void
153 hxge_uninit_rxdma_channels(p_hxge_t hxgep)
154 {
155 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_uninit_rxdma_channels"));
156 
157 	hxge_rxdma_hw_stop(hxgep);
158 	hxge_unmap_rxdma(hxgep);
159 
160 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "<== hxge_uinit_rxdma_channels"));
161 }
162 
163 hxge_status_t
164 hxge_init_rxdma_channel_cntl_stat(p_hxge_t hxgep, uint16_t channel,
165     rdc_stat_t *cs_p)
166 {
167 	hpi_handle_t	handle;
168 	hpi_status_t	rs = HPI_SUCCESS;
169 	hxge_status_t	status = HXGE_OK;
170 
171 	HXGE_DEBUG_MSG((hxgep, DMA_CTL,
172 	    "<== hxge_init_rxdma_channel_cntl_stat"));
173 
174 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
175 	rs = hpi_rxdma_control_status(handle, OP_SET, channel, cs_p);
176 
177 	if (rs != HPI_SUCCESS) {
178 		status = HXGE_ERROR | rs;
179 	}
180 	return (status);
181 }
182 
183 
184 hxge_status_t
185 hxge_enable_rxdma_channel(p_hxge_t hxgep, uint16_t channel,
186     p_rx_rbr_ring_t rbr_p, p_rx_rcr_ring_t rcr_p, p_rx_mbox_t mbox_p,
187     int n_init_kick)
188 {
189 	hpi_handle_t		handle;
190 	rdc_desc_cfg_t 		rdc_desc;
191 	rdc_rcr_cfg_b_t		*cfgb_p;
192 	hpi_status_t		rs = HPI_SUCCESS;
193 
194 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "==> hxge_enable_rxdma_channel"));
195 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
196 
197 	/*
198 	 * Use configuration data composed at init time. Write to hardware the
199 	 * receive ring configurations.
200 	 */
201 	rdc_desc.mbox_enable = 1;
202 	rdc_desc.mbox_addr = mbox_p->mbox_addr;
203 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
204 	    "==> hxge_enable_rxdma_channel: mboxp $%p($%p)",
205 	    mbox_p->mbox_addr, rdc_desc.mbox_addr));
206 
207 	rdc_desc.rbr_len = rbr_p->rbb_max;
208 	rdc_desc.rbr_addr = rbr_p->rbr_addr;
209 
210 	switch (hxgep->rx_bksize_code) {
211 	case RBR_BKSIZE_4K:
212 		rdc_desc.page_size = SIZE_4KB;
213 		break;
214 	case RBR_BKSIZE_8K:
215 		rdc_desc.page_size = SIZE_8KB;
216 		break;
217 	}
218 
219 	rdc_desc.size0 = rbr_p->hpi_pkt_buf_size0;
220 	rdc_desc.valid0 = 1;
221 
222 	rdc_desc.size1 = rbr_p->hpi_pkt_buf_size1;
223 	rdc_desc.valid1 = 1;
224 
225 	rdc_desc.size2 = rbr_p->hpi_pkt_buf_size2;
226 	rdc_desc.valid2 = 1;
227 
228 	rdc_desc.full_hdr = rcr_p->full_hdr_flag;
229 	rdc_desc.offset = rcr_p->sw_priv_hdr_len;
230 
231 	rdc_desc.rcr_len = rcr_p->comp_size;
232 	rdc_desc.rcr_addr = rcr_p->rcr_addr;
233 
234 	cfgb_p = &(rcr_p->rcr_cfgb);
235 	rdc_desc.rcr_threshold = cfgb_p->bits.pthres;
236 	rdc_desc.rcr_timeout = cfgb_p->bits.timeout;
237 	rdc_desc.rcr_timeout_enable = cfgb_p->bits.entout;
238 
239 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "==> hxge_enable_rxdma_channel: "
240 	    "rbr_len qlen %d pagesize code %d rcr_len %d",
241 	    rdc_desc.rbr_len, rdc_desc.page_size, rdc_desc.rcr_len));
242 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "==> hxge_enable_rxdma_channel: "
243 	    "size 0 %d size 1 %d size 2 %d",
244 	    rbr_p->hpi_pkt_buf_size0, rbr_p->hpi_pkt_buf_size1,
245 	    rbr_p->hpi_pkt_buf_size2));
246 
247 	rs = hpi_rxdma_cfg_rdc_ring(handle, rbr_p->rdc, &rdc_desc);
248 	if (rs != HPI_SUCCESS) {
249 		return (HXGE_ERROR | rs);
250 	}
251 
252 	/*
253 	 * Enable the timeout and threshold.
254 	 */
255 	rs = hpi_rxdma_cfg_rdc_rcr_threshold(handle, channel,
256 	    rdc_desc.rcr_threshold);
257 	if (rs != HPI_SUCCESS) {
258 		return (HXGE_ERROR | rs);
259 	}
260 
261 	rs = hpi_rxdma_cfg_rdc_rcr_timeout(handle, channel,
262 	    rdc_desc.rcr_timeout);
263 	if (rs != HPI_SUCCESS) {
264 		return (HXGE_ERROR | rs);
265 	}
266 
267 	/* Enable the DMA */
268 	rs = hpi_rxdma_cfg_rdc_enable(handle, channel);
269 	if (rs != HPI_SUCCESS) {
270 		return (HXGE_ERROR | rs);
271 	}
272 
273 	/* Kick the DMA engine */
274 	hpi_rxdma_rdc_rbr_kick(handle, channel, n_init_kick);
275 
276 	/* Clear the rbr empty bit */
277 	(void) hpi_rxdma_channel_rbr_empty_clear(handle, channel);
278 
279 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "<== hxge_enable_rxdma_channel"));
280 
281 	return (HXGE_OK);
282 }
283 
284 static hxge_status_t
285 hxge_disable_rxdma_channel(p_hxge_t hxgep, uint16_t channel)
286 {
287 	hpi_handle_t handle;
288 	hpi_status_t rs = HPI_SUCCESS;
289 
290 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "==> hxge_disable_rxdma_channel"));
291 
292 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
293 
294 	/* disable the DMA */
295 	rs = hpi_rxdma_cfg_rdc_disable(handle, channel);
296 	if (rs != HPI_SUCCESS) {
297 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
298 		    "<== hxge_disable_rxdma_channel:failed (0x%x)", rs));
299 		return (HXGE_ERROR | rs);
300 	}
301 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "<== hxge_disable_rxdma_channel"));
302 	return (HXGE_OK);
303 }
304 
305 hxge_status_t
306 hxge_rxdma_channel_rcrflush(p_hxge_t hxgep, uint8_t channel)
307 {
308 	hpi_handle_t	handle;
309 	hxge_status_t	status = HXGE_OK;
310 
311 	HXGE_DEBUG_MSG((hxgep, DMA_CTL,
312 	    "==> hxge_rxdma_channel_rcrflush"));
313 
314 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
315 	hpi_rxdma_rdc_rcr_flush(handle, channel);
316 
317 	HXGE_DEBUG_MSG((hxgep, DMA_CTL,
318 	    "<== hxge_rxdma_channel_rcrflush"));
319 	return (status);
320 
321 }
322 
323 #define	MID_INDEX(l, r) ((r + l + 1) >> 1)
324 
325 #define	TO_LEFT -1
326 #define	TO_RIGHT 1
327 #define	BOTH_RIGHT (TO_RIGHT + TO_RIGHT)
328 #define	BOTH_LEFT (TO_LEFT + TO_LEFT)
329 #define	IN_MIDDLE (TO_RIGHT + TO_LEFT)
330 #define	NO_HINT 0xffffffff
331 
332 /*ARGSUSED*/
333 hxge_status_t
334 hxge_rxbuf_pp_to_vp(p_hxge_t hxgep, p_rx_rbr_ring_t rbr_p,
335     uint8_t pktbufsz_type, uint64_t *pkt_buf_addr_pp,
336     uint64_t **pkt_buf_addr_p, uint32_t *bufoffset, uint32_t *msg_index)
337 {
338 	int			bufsize;
339 	uint64_t		pktbuf_pp;
340 	uint64_t		dvma_addr;
341 	rxring_info_t		*ring_info;
342 	int			base_side, end_side;
343 	int			r_index, l_index, anchor_index;
344 	int			found, search_done;
345 	uint32_t		offset, chunk_size, block_size, page_size_mask;
346 	uint32_t		chunk_index, block_index, total_index;
347 	int			max_iterations, iteration;
348 	rxbuf_index_info_t	*bufinfo;
349 
350 	HXGE_DEBUG_MSG((hxgep, RX2_CTL, "==> hxge_rxbuf_pp_to_vp"));
351 
352 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
353 	    "==> hxge_rxbuf_pp_to_vp: buf_pp $%p btype %d",
354 	    pkt_buf_addr_pp, pktbufsz_type));
355 
356 #if defined(__i386)
357 	pktbuf_pp = (uint64_t)(uint32_t)pkt_buf_addr_pp;
358 #else
359 	pktbuf_pp = (uint64_t)pkt_buf_addr_pp;
360 #endif
361 
362 	switch (pktbufsz_type) {
363 	case 0:
364 		bufsize = rbr_p->pkt_buf_size0;
365 		break;
366 	case 1:
367 		bufsize = rbr_p->pkt_buf_size1;
368 		break;
369 	case 2:
370 		bufsize = rbr_p->pkt_buf_size2;
371 		break;
372 	case RCR_SINGLE_BLOCK:
373 		bufsize = 0;
374 		anchor_index = 0;
375 		break;
376 	default:
377 		return (HXGE_ERROR);
378 	}
379 
380 	if (rbr_p->num_blocks == 1) {
381 		anchor_index = 0;
382 		ring_info = rbr_p->ring_info;
383 		bufinfo = (rxbuf_index_info_t *)ring_info->buffer;
384 
385 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
386 		    "==> hxge_rxbuf_pp_to_vp: (found, 1 block) "
387 		    "buf_pp $%p btype %d anchor_index %d bufinfo $%p",
388 		    pkt_buf_addr_pp, pktbufsz_type, anchor_index, bufinfo));
389 
390 		goto found_index;
391 	}
392 
393 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
394 	    "==> hxge_rxbuf_pp_to_vp: buf_pp $%p btype %d anchor_index %d",
395 	    pkt_buf_addr_pp, pktbufsz_type, anchor_index));
396 
397 	ring_info = rbr_p->ring_info;
398 	found = B_FALSE;
399 	bufinfo = (rxbuf_index_info_t *)ring_info->buffer;
400 	iteration = 0;
401 	max_iterations = ring_info->max_iterations;
402 
403 	/*
404 	 * First check if this block have been seen recently. This is indicated
405 	 * by a hint which is initialized when the first buffer of the block is
406 	 * seen. The hint is reset when the last buffer of the block has been
407 	 * processed. As three block sizes are supported, three hints are kept.
408 	 * The idea behind the hints is that once the hardware  uses a block
409 	 * for a buffer  of that size, it will use it exclusively for that size
410 	 * and will use it until it is exhausted. It is assumed that there
411 	 * would a single block being used for the same buffer sizes at any
412 	 * given time.
413 	 */
414 	if (ring_info->hint[pktbufsz_type] != NO_HINT) {
415 		anchor_index = ring_info->hint[pktbufsz_type];
416 		dvma_addr = bufinfo[anchor_index].dvma_addr;
417 		chunk_size = bufinfo[anchor_index].buf_size;
418 		if ((pktbuf_pp >= dvma_addr) &&
419 		    (pktbuf_pp < (dvma_addr + chunk_size))) {
420 			found = B_TRUE;
421 			/*
422 			 * check if this is the last buffer in the block If so,
423 			 * then reset the hint for the size;
424 			 */
425 
426 			if ((pktbuf_pp + bufsize) >= (dvma_addr + chunk_size))
427 				ring_info->hint[pktbufsz_type] = NO_HINT;
428 		}
429 	}
430 
431 	if (found == B_FALSE) {
432 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
433 		    "==> hxge_rxbuf_pp_to_vp: (!found)"
434 		    "buf_pp $%p btype %d anchor_index %d",
435 		    pkt_buf_addr_pp, pktbufsz_type, anchor_index));
436 
437 		/*
438 		 * This is the first buffer of the block of this size. Need to
439 		 * search the whole information array. the search algorithm
440 		 * uses a binary tree search algorithm. It assumes that the
441 		 * information is already sorted with increasing order info[0]
442 		 * < info[1] < info[2]  .... < info[n-1] where n is the size of
443 		 * the information array
444 		 */
445 		r_index = rbr_p->num_blocks - 1;
446 		l_index = 0;
447 		search_done = B_FALSE;
448 		anchor_index = MID_INDEX(r_index, l_index);
449 		while (search_done == B_FALSE) {
450 			if ((r_index == l_index) ||
451 			    (iteration >= max_iterations))
452 				search_done = B_TRUE;
453 
454 			end_side = TO_RIGHT;	/* to the right */
455 			base_side = TO_LEFT;	/* to the left */
456 			/* read the DVMA address information and sort it */
457 			dvma_addr = bufinfo[anchor_index].dvma_addr;
458 			chunk_size = bufinfo[anchor_index].buf_size;
459 
460 			HXGE_DEBUG_MSG((hxgep, RX2_CTL,
461 			    "==> hxge_rxbuf_pp_to_vp: (searching)"
462 			    "buf_pp $%p btype %d "
463 			    "anchor_index %d chunk_size %d dvmaaddr $%p",
464 			    pkt_buf_addr_pp, pktbufsz_type, anchor_index,
465 			    chunk_size, dvma_addr));
466 
467 			if (pktbuf_pp >= dvma_addr)
468 				base_side = TO_RIGHT;	/* to the right */
469 			if (pktbuf_pp < (dvma_addr + chunk_size))
470 				end_side = TO_LEFT;	/* to the left */
471 
472 			switch (base_side + end_side) {
473 			case IN_MIDDLE:
474 				/* found */
475 				found = B_TRUE;
476 				search_done = B_TRUE;
477 				if ((pktbuf_pp + bufsize) <
478 				    (dvma_addr + chunk_size))
479 					ring_info->hint[pktbufsz_type] =
480 					    bufinfo[anchor_index].buf_index;
481 				break;
482 			case BOTH_RIGHT:
483 				/* not found: go to the right */
484 				l_index = anchor_index + 1;
485 				anchor_index = MID_INDEX(r_index, l_index);
486 				break;
487 
488 			case BOTH_LEFT:
489 				/* not found: go to the left */
490 				r_index = anchor_index - 1;
491 				anchor_index = MID_INDEX(r_index, l_index);
492 				break;
493 			default:	/* should not come here */
494 				return (HXGE_ERROR);
495 			}
496 			iteration++;
497 		}
498 
499 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
500 		    "==> hxge_rxbuf_pp_to_vp: (search done)"
501 		    "buf_pp $%p btype %d anchor_index %d",
502 		    pkt_buf_addr_pp, pktbufsz_type, anchor_index));
503 	}
504 
505 	if (found == B_FALSE) {
506 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
507 		    "==> hxge_rxbuf_pp_to_vp: (search failed)"
508 		    "buf_pp $%p btype %d anchor_index %d",
509 		    pkt_buf_addr_pp, pktbufsz_type, anchor_index));
510 		return (HXGE_ERROR);
511 	}
512 
513 found_index:
514 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
515 	    "==> hxge_rxbuf_pp_to_vp: (FOUND1)"
516 	    "buf_pp $%p btype %d bufsize %d anchor_index %d",
517 	    pkt_buf_addr_pp, pktbufsz_type, bufsize, anchor_index));
518 
519 	/* index of the first block in this chunk */
520 	chunk_index = bufinfo[anchor_index].start_index;
521 	dvma_addr = bufinfo[anchor_index].dvma_addr;
522 	page_size_mask = ring_info->block_size_mask;
523 
524 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
525 	    "==> hxge_rxbuf_pp_to_vp: (FOUND3), get chunk)"
526 	    "buf_pp $%p btype %d bufsize %d "
527 	    "anchor_index %d chunk_index %d dvma $%p",
528 	    pkt_buf_addr_pp, pktbufsz_type, bufsize,
529 	    anchor_index, chunk_index, dvma_addr));
530 
531 	offset = pktbuf_pp - dvma_addr;	/* offset within the chunk */
532 	block_size = rbr_p->block_size;	/* System  block(page) size */
533 
534 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
535 	    "==> hxge_rxbuf_pp_to_vp: (FOUND4), get chunk)"
536 	    "buf_pp $%p btype %d bufsize %d "
537 	    "anchor_index %d chunk_index %d dvma $%p "
538 	    "offset %d block_size %d",
539 	    pkt_buf_addr_pp, pktbufsz_type, bufsize, anchor_index,
540 	    chunk_index, dvma_addr, offset, block_size));
541 	HXGE_DEBUG_MSG((hxgep, RX2_CTL, "==> getting total index"));
542 
543 	block_index = (offset / block_size);	/* index within chunk */
544 	total_index = chunk_index + block_index;
545 
546 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
547 	    "==> hxge_rxbuf_pp_to_vp: "
548 	    "total_index %d dvma_addr $%p "
549 	    "offset %d block_size %d "
550 	    "block_index %d ",
551 	    total_index, dvma_addr, offset, block_size, block_index));
552 
553 #if defined(__i386)
554 	*pkt_buf_addr_p = (uint64_t *)((uint32_t)bufinfo[anchor_index].kaddr +
555 	    (uint32_t)offset);
556 #else
557 	*pkt_buf_addr_p = (uint64_t *)((uint64_t)bufinfo[anchor_index].kaddr +
558 	    offset);
559 #endif
560 
561 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
562 	    "==> hxge_rxbuf_pp_to_vp: "
563 	    "total_index %d dvma_addr $%p "
564 	    "offset %d block_size %d "
565 	    "block_index %d "
566 	    "*pkt_buf_addr_p $%p",
567 	    total_index, dvma_addr, offset, block_size,
568 	    block_index, *pkt_buf_addr_p));
569 
570 	*msg_index = total_index;
571 	*bufoffset = (offset & page_size_mask);
572 
573 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
574 	    "==> hxge_rxbuf_pp_to_vp: get msg index: "
575 	    "msg_index %d bufoffset_index %d",
576 	    *msg_index, *bufoffset));
577 	HXGE_DEBUG_MSG((hxgep, RX2_CTL, "<== hxge_rxbuf_pp_to_vp"));
578 
579 	return (HXGE_OK);
580 }
581 
582 
583 /*
584  * used by quick sort (qsort) function
585  * to perform comparison
586  */
587 static int
588 hxge_sort_compare(const void *p1, const void *p2)
589 {
590 
591 	rxbuf_index_info_t *a, *b;
592 
593 	a = (rxbuf_index_info_t *)p1;
594 	b = (rxbuf_index_info_t *)p2;
595 
596 	if (a->dvma_addr > b->dvma_addr)
597 		return (1);
598 	if (a->dvma_addr < b->dvma_addr)
599 		return (-1);
600 	return (0);
601 }
602 
603 /*
604  * Grabbed this sort implementation from common/syscall/avl.c
605  *
606  * Generic shellsort, from K&R (1st ed, p 58.), somewhat modified.
607  * v = Ptr to array/vector of objs
608  * n = # objs in the array
609  * s = size of each obj (must be multiples of a word size)
610  * f = ptr to function to compare two objs
611  *	returns (-1 = less than, 0 = equal, 1 = greater than
612  */
613 void
614 hxge_ksort(caddr_t v, int n, int s, int (*f) ())
615 {
616 	int		g, i, j, ii;
617 	unsigned int	*p1, *p2;
618 	unsigned int	tmp;
619 
620 	/* No work to do */
621 	if (v == NULL || n <= 1)
622 		return;
623 	/* Sanity check on arguments */
624 	ASSERT(((uintptr_t)v & 0x3) == 0 && (s & 0x3) == 0);
625 	ASSERT(s > 0);
626 
627 	for (g = n / 2; g > 0; g /= 2) {
628 		for (i = g; i < n; i++) {
629 			for (j = i - g; j >= 0 &&
630 			    (*f) (v + j * s, v + (j + g) * s) == 1; j -= g) {
631 				p1 = (unsigned *)(v + j * s);
632 				p2 = (unsigned *)(v + (j + g) * s);
633 				for (ii = 0; ii < s / 4; ii++) {
634 					tmp = *p1;
635 					*p1++ = *p2;
636 					*p2++ = tmp;
637 				}
638 			}
639 		}
640 	}
641 }
642 
643 /*
644  * Initialize data structures required for rxdma
645  * buffer dvma->vmem address lookup
646  */
647 /*ARGSUSED*/
648 static hxge_status_t
649 hxge_rxbuf_index_info_init(p_hxge_t hxgep, p_rx_rbr_ring_t rbrp)
650 {
651 	int		index;
652 	rxring_info_t	*ring_info;
653 	int		max_iteration = 0, max_index = 0;
654 
655 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "==> hxge_rxbuf_index_info_init"));
656 
657 	ring_info = rbrp->ring_info;
658 	ring_info->hint[0] = NO_HINT;
659 	ring_info->hint[1] = NO_HINT;
660 	ring_info->hint[2] = NO_HINT;
661 	max_index = rbrp->num_blocks;
662 
663 	/* read the DVMA address information and sort it */
664 	/* do init of the information array */
665 
666 	HXGE_DEBUG_MSG((hxgep, DMA2_CTL,
667 	    " hxge_rxbuf_index_info_init Sort ptrs"));
668 
669 	/* sort the array */
670 	hxge_ksort((void *) ring_info->buffer, max_index,
671 	    sizeof (rxbuf_index_info_t), hxge_sort_compare);
672 
673 	for (index = 0; index < max_index; index++) {
674 		HXGE_DEBUG_MSG((hxgep, DMA2_CTL,
675 		    " hxge_rxbuf_index_info_init: sorted chunk %d "
676 		    " ioaddr $%p kaddr $%p size %x",
677 		    index, ring_info->buffer[index].dvma_addr,
678 		    ring_info->buffer[index].kaddr,
679 		    ring_info->buffer[index].buf_size));
680 	}
681 
682 	max_iteration = 0;
683 	while (max_index >= (1ULL << max_iteration))
684 		max_iteration++;
685 	ring_info->max_iterations = max_iteration + 1;
686 
687 	HXGE_DEBUG_MSG((hxgep, DMA2_CTL,
688 	    " hxge_rxbuf_index_info_init Find max iter %d",
689 	    ring_info->max_iterations));
690 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "<== hxge_rxbuf_index_info_init"));
691 
692 	return (HXGE_OK);
693 }
694 
695 /*ARGSUSED*/
696 void
697 hxge_dump_rcr_entry(p_hxge_t hxgep, p_rcr_entry_t entry_p)
698 {
699 #ifdef	HXGE_DEBUG
700 
701 	uint32_t bptr;
702 	uint64_t pp;
703 
704 	bptr = entry_p->bits.pkt_buf_addr;
705 
706 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
707 	    "\trcr entry $%p "
708 	    "\trcr entry 0x%0llx "
709 	    "\trcr entry 0x%08x "
710 	    "\trcr entry 0x%08x "
711 	    "\tvalue 0x%0llx\n"
712 	    "\tmulti = %d\n"
713 	    "\tpkt_type = 0x%x\n"
714 	    "\terror = 0x%04x\n"
715 	    "\tl2_len = %d\n"
716 	    "\tpktbufsize = %d\n"
717 	    "\tpkt_buf_addr = $%p\n"
718 	    "\tpkt_buf_addr (<< 6) = $%p\n",
719 	    entry_p,
720 	    *(int64_t *)entry_p,
721 	    *(int32_t *)entry_p,
722 	    *(int32_t *)((char *)entry_p + 32),
723 	    entry_p->value,
724 	    entry_p->bits.multi,
725 	    entry_p->bits.pkt_type,
726 	    entry_p->bits.error,
727 	    entry_p->bits.l2_len,
728 	    entry_p->bits.pktbufsz,
729 	    bptr,
730 	    entry_p->bits.pkt_buf_addr_l));
731 
732 	pp = (entry_p->value & RCR_PKT_BUF_ADDR_MASK) <<
733 	    RCR_PKT_BUF_ADDR_SHIFT;
734 
735 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "rcr pp 0x%llx l2 len %d",
736 	    pp, (*(int64_t *)entry_p >> 40) & 0x3fff));
737 #endif
738 }
739 
740 /*ARGSUSED*/
741 void
742 hxge_rxdma_stop(p_hxge_t hxgep)
743 {
744 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rxdma_stop"));
745 
746 	(void) hxge_rx_vmac_disable(hxgep);
747 	(void) hxge_rxdma_hw_mode(hxgep, HXGE_DMA_STOP);
748 
749 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_rxdma_stop"));
750 }
751 
752 void
753 hxge_rxdma_stop_reinit(p_hxge_t hxgep)
754 {
755 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rxdma_stop_reinit"));
756 
757 	(void) hxge_rxdma_stop(hxgep);
758 	(void) hxge_uninit_rxdma_channels(hxgep);
759 	(void) hxge_init_rxdma_channels(hxgep);
760 
761 	(void) hxge_rx_vmac_enable(hxgep);
762 
763 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_rxdma_stop_reinit"));
764 }
765 
766 hxge_status_t
767 hxge_rxdma_hw_mode(p_hxge_t hxgep, boolean_t enable)
768 {
769 	int			i, ndmas;
770 	uint16_t		channel;
771 	p_rx_rbr_rings_t	rx_rbr_rings;
772 	p_rx_rbr_ring_t		*rbr_rings;
773 	hpi_handle_t		handle;
774 	hpi_status_t		rs = HPI_SUCCESS;
775 	hxge_status_t		status = HXGE_OK;
776 
777 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
778 	    "==> hxge_rxdma_hw_mode: mode %d", enable));
779 
780 	if (!(hxgep->drv_state & STATE_HW_INITIALIZED)) {
781 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
782 		    "<== hxge_rxdma_mode: not initialized"));
783 		return (HXGE_ERROR);
784 	}
785 
786 	rx_rbr_rings = hxgep->rx_rbr_rings;
787 	if (rx_rbr_rings == NULL) {
788 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
789 		    "<== hxge_rxdma_mode: NULL ring pointer"));
790 		return (HXGE_ERROR);
791 	}
792 
793 	if (rx_rbr_rings->rbr_rings == NULL) {
794 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
795 		    "<== hxge_rxdma_mode: NULL rbr rings pointer"));
796 		return (HXGE_ERROR);
797 	}
798 
799 	ndmas = rx_rbr_rings->ndmas;
800 	if (!ndmas) {
801 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
802 		    "<== hxge_rxdma_mode: no channel"));
803 		return (HXGE_ERROR);
804 	}
805 
806 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
807 	    "==> hxge_rxdma_mode (ndmas %d)", ndmas));
808 
809 	rbr_rings = rx_rbr_rings->rbr_rings;
810 
811 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
812 
813 	for (i = 0; i < ndmas; i++) {
814 		if (rbr_rings == NULL || rbr_rings[i] == NULL) {
815 			continue;
816 		}
817 		channel = rbr_rings[i]->rdc;
818 		if (enable) {
819 			HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
820 			    "==> hxge_rxdma_hw_mode: channel %d (enable)",
821 			    channel));
822 			rs = hpi_rxdma_cfg_rdc_enable(handle, channel);
823 		} else {
824 			HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
825 			    "==> hxge_rxdma_hw_mode: channel %d (disable)",
826 			    channel));
827 			rs = hpi_rxdma_cfg_rdc_disable(handle, channel);
828 		}
829 	}
830 
831 	status = ((rs == HPI_SUCCESS) ? HXGE_OK : HXGE_ERROR | rs);
832 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
833 	    "<== hxge_rxdma_hw_mode: status 0x%x", status));
834 
835 	return (status);
836 }
837 
838 int
839 hxge_rxdma_get_ring_index(p_hxge_t hxgep, uint16_t channel)
840 {
841 	int			i, ndmas;
842 	uint16_t		rdc;
843 	p_rx_rbr_rings_t 	rx_rbr_rings;
844 	p_rx_rbr_ring_t		*rbr_rings;
845 
846 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
847 	    "==> hxge_rxdma_get_ring_index: channel %d", channel));
848 
849 	rx_rbr_rings = hxgep->rx_rbr_rings;
850 	if (rx_rbr_rings == NULL) {
851 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
852 		    "<== hxge_rxdma_get_ring_index: NULL ring pointer"));
853 		return (-1);
854 	}
855 
856 	ndmas = rx_rbr_rings->ndmas;
857 	if (!ndmas) {
858 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
859 		    "<== hxge_rxdma_get_ring_index: no channel"));
860 		return (-1);
861 	}
862 
863 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
864 	    "==> hxge_rxdma_get_ring_index (ndmas %d)", ndmas));
865 
866 	rbr_rings = rx_rbr_rings->rbr_rings;
867 	for (i = 0; i < ndmas; i++) {
868 		rdc = rbr_rings[i]->rdc;
869 		if (channel == rdc) {
870 			HXGE_DEBUG_MSG((hxgep, RX_CTL,
871 			    "==> hxge_rxdma_get_rbr_ring: "
872 			    "channel %d (index %d) "
873 			    "ring %d", channel, i, rbr_rings[i]));
874 
875 			return (i);
876 		}
877 	}
878 
879 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
880 	    "<== hxge_rxdma_get_rbr_ring_index: not found"));
881 
882 	return (-1);
883 }
884 
885 /*
886  * Static functions start here.
887  */
888 static p_rx_msg_t
889 hxge_allocb(size_t size, uint32_t pri, p_hxge_dma_common_t dmabuf_p)
890 {
891 	p_rx_msg_t		hxge_mp = NULL;
892 	p_hxge_dma_common_t	dmamsg_p;
893 	uchar_t			*buffer;
894 
895 	hxge_mp = KMEM_ZALLOC(sizeof (rx_msg_t), KM_NOSLEEP);
896 	if (hxge_mp == NULL) {
897 		HXGE_ERROR_MSG((NULL, HXGE_ERR_CTL,
898 		    "Allocation of a rx msg failed."));
899 		goto hxge_allocb_exit;
900 	}
901 
902 	hxge_mp->use_buf_pool = B_FALSE;
903 	if (dmabuf_p) {
904 		hxge_mp->use_buf_pool = B_TRUE;
905 
906 		dmamsg_p = (p_hxge_dma_common_t)&hxge_mp->buf_dma;
907 		*dmamsg_p = *dmabuf_p;
908 		dmamsg_p->nblocks = 1;
909 		dmamsg_p->block_size = size;
910 		dmamsg_p->alength = size;
911 		buffer = (uchar_t *)dmabuf_p->kaddrp;
912 
913 		dmabuf_p->kaddrp = (void *)((char *)dmabuf_p->kaddrp + size);
914 		dmabuf_p->ioaddr_pp = (void *)
915 		    ((char *)dmabuf_p->ioaddr_pp + size);
916 
917 		dmabuf_p->alength -= size;
918 		dmabuf_p->offset += size;
919 		dmabuf_p->dma_cookie.dmac_laddress += size;
920 		dmabuf_p->dma_cookie.dmac_size -= size;
921 	} else {
922 		buffer = KMEM_ALLOC(size, KM_NOSLEEP);
923 		if (buffer == NULL) {
924 			HXGE_ERROR_MSG((NULL, HXGE_ERR_CTL,
925 			    "Allocation of a receive page failed."));
926 			goto hxge_allocb_fail1;
927 		}
928 	}
929 
930 	hxge_mp->rx_mblk_p = desballoc(buffer, size, pri, &hxge_mp->freeb);
931 	if (hxge_mp->rx_mblk_p == NULL) {
932 		HXGE_ERROR_MSG((NULL, HXGE_ERR_CTL, "desballoc failed."));
933 		goto hxge_allocb_fail2;
934 	}
935 	hxge_mp->buffer = buffer;
936 	hxge_mp->block_size = size;
937 	hxge_mp->freeb.free_func = (void (*) ()) hxge_freeb;
938 	hxge_mp->freeb.free_arg = (caddr_t)hxge_mp;
939 	hxge_mp->ref_cnt = 1;
940 	hxge_mp->free = B_TRUE;
941 	hxge_mp->rx_use_bcopy = B_FALSE;
942 
943 	atomic_inc_32(&hxge_mblks_pending);
944 
945 	goto hxge_allocb_exit;
946 
947 hxge_allocb_fail2:
948 	if (!hxge_mp->use_buf_pool) {
949 		KMEM_FREE(buffer, size);
950 	}
951 hxge_allocb_fail1:
952 	KMEM_FREE(hxge_mp, sizeof (rx_msg_t));
953 	hxge_mp = NULL;
954 
955 hxge_allocb_exit:
956 	return (hxge_mp);
957 }
958 
959 p_mblk_t
960 hxge_dupb(p_rx_msg_t hxge_mp, uint_t offset, size_t size)
961 {
962 	p_mblk_t mp;
963 
964 	HXGE_DEBUG_MSG((NULL, MEM_CTL, "==> hxge_dupb"));
965 	HXGE_DEBUG_MSG((NULL, MEM_CTL, "hxge_mp = $%p "
966 	    "offset = 0x%08X " "size = 0x%08X", hxge_mp, offset, size));
967 
968 	mp = desballoc(&hxge_mp->buffer[offset], size, 0, &hxge_mp->freeb);
969 	if (mp == NULL) {
970 		HXGE_DEBUG_MSG((NULL, RX_CTL, "desballoc failed"));
971 		goto hxge_dupb_exit;
972 	}
973 
974 	atomic_inc_32(&hxge_mp->ref_cnt);
975 
976 hxge_dupb_exit:
977 	HXGE_DEBUG_MSG((NULL, MEM_CTL, "<== hxge_dupb mp = $%p", hxge_mp));
978 	return (mp);
979 }
980 
981 p_mblk_t
982 hxge_dupb_bcopy(p_rx_msg_t hxge_mp, uint_t offset, size_t size)
983 {
984 	p_mblk_t	mp;
985 	uchar_t		*dp;
986 
987 	mp = allocb(size + HXGE_RXBUF_EXTRA, 0);
988 	if (mp == NULL) {
989 		HXGE_DEBUG_MSG((NULL, RX_CTL, "desballoc failed"));
990 		goto hxge_dupb_bcopy_exit;
991 	}
992 	dp = mp->b_rptr = mp->b_rptr + HXGE_RXBUF_EXTRA;
993 	bcopy((void *) &hxge_mp->buffer[offset], dp, size);
994 	mp->b_wptr = dp + size;
995 
996 hxge_dupb_bcopy_exit:
997 
998 	HXGE_DEBUG_MSG((NULL, MEM_CTL, "<== hxge_dupb mp = $%p", hxge_mp));
999 
1000 	return (mp);
1001 }
1002 
1003 void hxge_post_page(p_hxge_t hxgep, p_rx_rbr_ring_t rx_rbr_p,
1004     p_rx_msg_t rx_msg_p);
1005 
1006 void
1007 hxge_post_page(p_hxge_t hxgep, p_rx_rbr_ring_t rx_rbr_p, p_rx_msg_t rx_msg_p)
1008 {
1009 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_post_page"));
1010 
1011 	/* Reuse this buffer */
1012 	rx_msg_p->free = B_FALSE;
1013 	rx_msg_p->cur_usage_cnt = 0;
1014 	rx_msg_p->max_usage_cnt = 0;
1015 	rx_msg_p->pkt_buf_size = 0;
1016 
1017 	if (rx_rbr_p->rbr_use_bcopy) {
1018 		rx_msg_p->rx_use_bcopy = B_FALSE;
1019 		atomic_dec_32(&rx_rbr_p->rbr_consumed);
1020 	}
1021 	atomic_dec_32(&rx_rbr_p->rbr_used);
1022 
1023 	/*
1024 	 * Get the rbr header pointer and its offset index.
1025 	 */
1026 	rx_rbr_p->rbr_wr_index = ((rx_rbr_p->rbr_wr_index + 1) &
1027 	    rx_rbr_p->rbr_wrap_mask);
1028 	rx_rbr_p->rbr_desc_vp[rx_rbr_p->rbr_wr_index] = rx_msg_p->shifted_addr;
1029 
1030 	/*
1031 	 * Accumulate some buffers in the ring before re-enabling the
1032 	 * DMA channel, if rbr empty was signaled.
1033 	 */
1034 	hpi_rxdma_rdc_rbr_kick(HXGE_DEV_HPI_HANDLE(hxgep), rx_rbr_p->rdc, 1);
1035 	if (rx_rbr_p->rbr_is_empty &&
1036 	    (rx_rbr_p->rbb_max - rx_rbr_p->rbr_used) >=
1037 	    HXGE_RBR_EMPTY_THRESHOLD) {
1038 		hxge_rbr_empty_restore(hxgep, rx_rbr_p);
1039 	}
1040 
1041 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
1042 	    "<== hxge_post_page (channel %d post_next_index %d)",
1043 	    rx_rbr_p->rdc, rx_rbr_p->rbr_wr_index));
1044 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_post_page"));
1045 }
1046 
1047 void
1048 hxge_freeb(p_rx_msg_t rx_msg_p)
1049 {
1050 	size_t		size;
1051 	uchar_t		*buffer = NULL;
1052 	int		ref_cnt;
1053 	boolean_t	free_state = B_FALSE;
1054 	rx_rbr_ring_t	*ring = rx_msg_p->rx_rbr_p;
1055 
1056 	HXGE_DEBUG_MSG((NULL, MEM2_CTL, "==> hxge_freeb"));
1057 	HXGE_DEBUG_MSG((NULL, MEM2_CTL,
1058 	    "hxge_freeb:rx_msg_p = $%p (block pending %d)",
1059 	    rx_msg_p, hxge_mblks_pending));
1060 
1061 	if (ring == NULL)
1062 		return;
1063 
1064 	/*
1065 	 * This is to prevent posting activities while we are recovering
1066 	 * from fatal errors. This should not be a performance drag since
1067 	 * ref_cnt != 0 most times.
1068 	 */
1069 	if (ring->rbr_state == RBR_POSTING)
1070 		MUTEX_ENTER(&ring->post_lock);
1071 
1072 	/*
1073 	 * First we need to get the free state, then
1074 	 * atomic decrement the reference count to prevent
1075 	 * the race condition with the interrupt thread that
1076 	 * is processing a loaned up buffer block.
1077 	 */
1078 	free_state = rx_msg_p->free;
1079 	ref_cnt = atomic_add_32_nv(&rx_msg_p->ref_cnt, -1);
1080 	if (!ref_cnt) {
1081 		atomic_dec_32(&hxge_mblks_pending);
1082 
1083 		buffer = rx_msg_p->buffer;
1084 		size = rx_msg_p->block_size;
1085 
1086 		HXGE_DEBUG_MSG((NULL, MEM2_CTL, "hxge_freeb: "
1087 		    "will free: rx_msg_p = $%p (block pending %d)",
1088 		    rx_msg_p, hxge_mblks_pending));
1089 
1090 		if (!rx_msg_p->use_buf_pool) {
1091 			KMEM_FREE(buffer, size);
1092 		}
1093 
1094 		KMEM_FREE(rx_msg_p, sizeof (rx_msg_t));
1095 		/*
1096 		 * Decrement the receive buffer ring's reference
1097 		 * count, too.
1098 		 */
1099 		atomic_dec_32(&ring->rbr_ref_cnt);
1100 
1101 		/*
1102 		 * Free the receive buffer ring, iff
1103 		 * 1. all the receive buffers have been freed
1104 		 * 2. and we are in the proper state (that is,
1105 		 *    we are not UNMAPPING).
1106 		 */
1107 		if (ring->rbr_ref_cnt == 0 &&
1108 		    ring->rbr_state == RBR_UNMAPPED) {
1109 			KMEM_FREE(ring, sizeof (*ring));
1110 			/* post_lock has been destroyed already */
1111 			return;
1112 		}
1113 	}
1114 
1115 	/*
1116 	 * Repost buffer.
1117 	 */
1118 	if (free_state && (ref_cnt == 1)) {
1119 		HXGE_DEBUG_MSG((NULL, RX_CTL,
1120 		    "hxge_freeb: post page $%p:", rx_msg_p));
1121 		if (ring->rbr_state == RBR_POSTING)
1122 			hxge_post_page(rx_msg_p->hxgep, ring, rx_msg_p);
1123 	}
1124 
1125 	if (ring->rbr_state == RBR_POSTING)
1126 		MUTEX_EXIT(&ring->post_lock);
1127 
1128 	HXGE_DEBUG_MSG((NULL, MEM2_CTL, "<== hxge_freeb"));
1129 }
1130 
1131 uint_t
1132 hxge_rx_intr(caddr_t arg1, caddr_t arg2)
1133 {
1134 	p_hxge_ldv_t		ldvp = (p_hxge_ldv_t)arg1;
1135 	p_hxge_t		hxgep = (p_hxge_t)arg2;
1136 	p_hxge_ldg_t		ldgp;
1137 	uint8_t			channel;
1138 	hpi_handle_t		handle;
1139 	rdc_stat_t		cs;
1140 	uint_t			serviced = DDI_INTR_UNCLAIMED;
1141 
1142 	if (ldvp == NULL) {
1143 		HXGE_DEBUG_MSG((NULL, RX_INT_CTL,
1144 		    "<== hxge_rx_intr: arg2 $%p arg1 $%p", hxgep, ldvp));
1145 		return (DDI_INTR_UNCLAIMED);
1146 	}
1147 
1148 	if (arg2 == NULL || (void *) ldvp->hxgep != arg2) {
1149 		hxgep = ldvp->hxgep;
1150 	}
1151 
1152 	/*
1153 	 * If the interface is not started, just swallow the interrupt
1154 	 * for the logical device and don't rearm it.
1155 	 */
1156 	if (hxgep->hxge_mac_state != HXGE_MAC_STARTED)
1157 		return (DDI_INTR_CLAIMED);
1158 
1159 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1160 	    "==> hxge_rx_intr: arg2 $%p arg1 $%p", hxgep, ldvp));
1161 
1162 	/*
1163 	 * This interrupt handler is for a specific receive dma channel.
1164 	 */
1165 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
1166 
1167 	/*
1168 	 * Get the control and status for this channel.
1169 	 */
1170 	channel = ldvp->channel;
1171 	ldgp = ldvp->ldgp;
1172 	RXDMA_REG_READ64(handle, RDC_STAT, channel, &cs.value);
1173 	cs.bits.ptrread = 0;
1174 	cs.bits.pktread = 0;
1175 	RXDMA_REG_WRITE64(handle, RDC_STAT, channel, cs.value);
1176 
1177 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL, "==> hxge_rx_intr:channel %d "
1178 	    "cs 0x%016llx rcrto 0x%x rcrthres %x",
1179 	    channel, cs.value, cs.bits.rcr_to, cs.bits.rcr_thres));
1180 
1181 	hxge_rx_pkts_vring(hxgep, ldvp->vdma_index, ldvp, cs);
1182 	serviced = DDI_INTR_CLAIMED;
1183 
1184 	/* error events. */
1185 	if (cs.value & RDC_STAT_ERROR) {
1186 		(void) hxge_rx_err_evnts(hxgep, ldvp->vdma_index, ldvp, cs);
1187 	}
1188 
1189 hxge_intr_exit:
1190 	/*
1191 	 * Enable the mailbox update interrupt if we want to use mailbox. We
1192 	 * probably don't need to use mailbox as it only saves us one pio read.
1193 	 * Also write 1 to rcrthres and rcrto to clear these two edge triggered
1194 	 * bits.
1195 	 */
1196 	cs.value &= RDC_STAT_WR1C;
1197 	cs.bits.mex = 1;
1198 	cs.bits.ptrread = 0;
1199 	cs.bits.pktread = 0;
1200 	RXDMA_REG_WRITE64(handle, RDC_STAT, channel, cs.value);
1201 
1202 	/*
1203 	 * Rearm this logical group if this is a single device group.
1204 	 */
1205 	if (ldgp->nldvs == 1) {
1206 		ld_intr_mgmt_t mgm;
1207 
1208 		mgm.value = 0;
1209 		mgm.bits.arm = 1;
1210 		mgm.bits.timer = ldgp->ldg_timer;
1211 		HXGE_REG_WR32(handle,
1212 		    LD_INTR_MGMT + LDSV_OFFSET(ldgp->ldg), mgm.value);
1213 	}
1214 
1215 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1216 	    "<== hxge_rx_intr: serviced %d", serviced));
1217 	return (serviced);
1218 }
1219 
1220 static void
1221 hxge_rx_pkts_vring(p_hxge_t hxgep, uint_t vindex, p_hxge_ldv_t ldvp,
1222     rdc_stat_t cs)
1223 {
1224 	p_mblk_t		mp;
1225 	p_rx_rcr_ring_t		rcrp;
1226 
1227 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL, "==> hxge_rx_pkts_vring"));
1228 	if ((mp = hxge_rx_pkts(hxgep, vindex, ldvp, &rcrp, cs)) == NULL) {
1229 		HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1230 		    "<== hxge_rx_pkts_vring: no mp"));
1231 		return;
1232 	}
1233 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rx_pkts_vring: $%p", mp));
1234 
1235 #ifdef  HXGE_DEBUG
1236 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
1237 	    "==> hxge_rx_pkts_vring:calling mac_rx (NEMO) "
1238 	    "LEN %d mp $%p mp->b_next $%p rcrp $%p",
1239 	    (mp->b_wptr - mp->b_rptr), mp, mp->b_next, rcrp));
1240 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
1241 	    "==> hxge_rx_pkts_vring: dump packets "
1242 	    "(mp $%p b_rptr $%p b_wptr $%p):\n %s",
1243 	    mp, mp->b_rptr, mp->b_wptr,
1244 	    hxge_dump_packet((char *)mp->b_rptr, 64)));
1245 
1246 	if (mp->b_cont) {
1247 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
1248 		    "==> hxge_rx_pkts_vring: dump b_cont packets "
1249 		    "(mp->b_cont $%p b_rptr $%p b_wptr $%p):\n %s",
1250 		    mp->b_cont, mp->b_cont->b_rptr, mp->b_cont->b_wptr,
1251 		    hxge_dump_packet((char *)mp->b_cont->b_rptr,
1252 		    mp->b_cont->b_wptr - mp->b_cont->b_rptr)));
1253 		}
1254 	if (mp->b_next) {
1255 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
1256 		    "==> hxge_rx_pkts_vring: dump next packets "
1257 		    "(b_rptr $%p): %s",
1258 		    mp->b_next->b_rptr,
1259 		    hxge_dump_packet((char *)mp->b_next->b_rptr, 64)));
1260 	}
1261 #endif
1262 
1263 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
1264 	    "==> hxge_rx_pkts_vring: send packet to stack"));
1265 	mac_rx(hxgep->mach, NULL, mp);
1266 
1267 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_rx_pkts_vring"));
1268 }
1269 
1270 /*ARGSUSED*/
1271 mblk_t *
1272 hxge_rx_pkts(p_hxge_t hxgep, uint_t vindex, p_hxge_ldv_t ldvp,
1273     p_rx_rcr_ring_t *rcrp, rdc_stat_t cs)
1274 {
1275 	hpi_handle_t		handle;
1276 	uint8_t			channel;
1277 	p_rx_rcr_rings_t	rx_rcr_rings;
1278 	p_rx_rcr_ring_t		rcr_p;
1279 	uint32_t		comp_rd_index;
1280 	p_rcr_entry_t		rcr_desc_rd_head_p;
1281 	p_rcr_entry_t		rcr_desc_rd_head_pp;
1282 	p_mblk_t		nmp, mp_cont, head_mp, *tail_mp;
1283 	uint16_t		qlen, nrcr_read, npkt_read;
1284 	uint32_t		qlen_hw, qlen_sw;
1285 	uint32_t		invalid_rcr_entry;
1286 	boolean_t		multi;
1287 	rdc_rcr_cfg_b_t		rcr_cfg_b;
1288 	p_rx_mbox_t		rx_mboxp;
1289 	p_rxdma_mailbox_t	mboxp;
1290 	uint64_t		rcr_head_index, rcr_tail_index;
1291 	uint64_t		rcr_tail;
1292 	rdc_rcr_tail_t		rcr_tail_reg;
1293 	p_hxge_rx_ring_stats_t	rdc_stats;
1294 
1295 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL, "==> hxge_rx_pkts:vindex %d "
1296 	    "channel %d", vindex, ldvp->channel));
1297 
1298 	if (!(hxgep->drv_state & STATE_HW_INITIALIZED)) {
1299 		return (NULL);
1300 	}
1301 
1302 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
1303 	rx_rcr_rings = hxgep->rx_rcr_rings;
1304 	rcr_p = rx_rcr_rings->rcr_rings[vindex];
1305 	channel = rcr_p->rdc;
1306 	if (channel != ldvp->channel) {
1307 		HXGE_DEBUG_MSG((hxgep, RX_INT_CTL, "==> hxge_rx_pkts:index %d "
1308 		    "channel %d, and rcr channel %d not matched.",
1309 		    vindex, ldvp->channel, channel));
1310 		return (NULL);
1311 	}
1312 
1313 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1314 	    "==> hxge_rx_pkts: START: rcr channel %d "
1315 	    "head_p $%p head_pp $%p  index %d ",
1316 	    channel, rcr_p->rcr_desc_rd_head_p,
1317 	    rcr_p->rcr_desc_rd_head_pp, rcr_p->comp_rd_index));
1318 
1319 	rx_mboxp = hxgep->rx_mbox_areas_p->rxmbox_areas[channel];
1320 	mboxp = (p_rxdma_mailbox_t)rx_mboxp->rx_mbox.kaddrp;
1321 	(void) hpi_rxdma_rdc_rcr_qlen_get(handle, channel, &qlen);
1322 	RXDMA_REG_READ64(handle, RDC_RCR_TAIL, channel, &rcr_tail_reg.value);
1323 	rcr_tail = rcr_tail_reg.bits.tail;
1324 
1325 	if (!qlen) {
1326 		HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1327 		    "<== hxge_rx_pkts:rcr channel %d qlen %d (no pkts)",
1328 		    channel, qlen));
1329 		return (NULL);
1330 	}
1331 
1332 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rx_pkts:rcr channel %d "
1333 	    "qlen %d", channel, qlen));
1334 
1335 	comp_rd_index = rcr_p->comp_rd_index;
1336 
1337 	rcr_desc_rd_head_p = rcr_p->rcr_desc_rd_head_p;
1338 	rcr_desc_rd_head_pp = rcr_p->rcr_desc_rd_head_pp;
1339 	nrcr_read = npkt_read = 0;
1340 
1341 	/*
1342 	 * Number of packets queued (The jumbo or multi packet will be counted
1343 	 * as only one paccket and it may take up more than one completion
1344 	 * entry).
1345 	 */
1346 	if (cs.bits.rbr_empty)
1347 		qlen_hw = qlen;
1348 	else
1349 		qlen_hw = (qlen < hxge_max_rx_pkts) ? qlen : hxge_max_rx_pkts;
1350 
1351 	head_mp = NULL;
1352 	tail_mp = &head_mp;
1353 	nmp = mp_cont = NULL;
1354 	multi = B_FALSE;
1355 
1356 	rcr_head_index = rcr_p->rcr_desc_rd_head_p - rcr_p->rcr_desc_first_p;
1357 	rcr_tail_index = rcr_tail - rcr_p->rcr_tail_begin;
1358 
1359 	if (rcr_tail_index >= rcr_head_index) {
1360 		qlen_sw = rcr_tail_index - rcr_head_index;
1361 	} else {
1362 		/* rcr_tail has wrapped around */
1363 		qlen_sw = (rcr_p->comp_size - rcr_head_index) + rcr_tail_index;
1364 	}
1365 
1366 	if (qlen_hw > qlen_sw) {
1367 		HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1368 		    "Channel %d, rcr_qlen from reg %d and from rcr_tail %d\n",
1369 		    channel, qlen_hw, qlen_sw));
1370 		qlen_hw = qlen_sw;
1371 	}
1372 
1373 	while (qlen_hw) {
1374 #ifdef HXGE_DEBUG
1375 		hxge_dump_rcr_entry(hxgep, rcr_desc_rd_head_p);
1376 #endif
1377 		/*
1378 		 * Process one completion ring entry.
1379 		 */
1380 		invalid_rcr_entry = 0;
1381 		hxge_receive_packet(hxgep,
1382 		    rcr_p, rcr_desc_rd_head_p, &multi, &nmp, &mp_cont,
1383 		    &invalid_rcr_entry);
1384 		if (invalid_rcr_entry != 0) {
1385 			rdc_stats = rcr_p->rdc_stats;
1386 			rdc_stats->rcr_invalids++;
1387 			HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1388 			    "Channel %d could only read 0x%x packets, "
1389 			    "but 0x%x pending\n", channel, npkt_read, qlen_hw));
1390 			break;
1391 		}
1392 
1393 		/*
1394 		 * message chaining modes (nemo msg chaining)
1395 		 */
1396 		if (nmp) {
1397 			nmp->b_next = NULL;
1398 			if (!multi && !mp_cont) { /* frame fits a partition */
1399 				*tail_mp = nmp;
1400 				tail_mp = &nmp->b_next;
1401 				nmp = NULL;
1402 			} else if (multi && !mp_cont) { /* first segment */
1403 				*tail_mp = nmp;
1404 				tail_mp = &nmp->b_cont;
1405 			} else if (multi && mp_cont) {	/* mid of multi segs */
1406 				*tail_mp = mp_cont;
1407 				tail_mp = &mp_cont->b_cont;
1408 			} else if (!multi && mp_cont) { /* last segment */
1409 				*tail_mp = mp_cont;
1410 				tail_mp = &nmp->b_next;
1411 				nmp = NULL;
1412 			}
1413 		}
1414 
1415 		HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1416 		    "==> hxge_rx_pkts: loop: rcr channel %d "
1417 		    "before updating: multi %d "
1418 		    "nrcr_read %d "
1419 		    "npk read %d "
1420 		    "head_pp $%p  index %d ",
1421 		    channel, multi,
1422 		    nrcr_read, npkt_read, rcr_desc_rd_head_pp, comp_rd_index));
1423 
1424 		if (!multi) {
1425 			qlen_hw--;
1426 			npkt_read++;
1427 		}
1428 
1429 		/*
1430 		 * Update the next read entry.
1431 		 */
1432 		comp_rd_index = NEXT_ENTRY(comp_rd_index,
1433 		    rcr_p->comp_wrap_mask);
1434 
1435 		rcr_desc_rd_head_p = NEXT_ENTRY_PTR(rcr_desc_rd_head_p,
1436 		    rcr_p->rcr_desc_first_p, rcr_p->rcr_desc_last_p);
1437 
1438 		nrcr_read++;
1439 
1440 		HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1441 		    "<== hxge_rx_pkts: (SAM, process one packet) "
1442 		    "nrcr_read %d", nrcr_read));
1443 		HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1444 		    "==> hxge_rx_pkts: loop: rcr channel %d "
1445 		    "multi %d nrcr_read %d npk read %d head_pp $%p  index %d ",
1446 		    channel, multi, nrcr_read, npkt_read, rcr_desc_rd_head_pp,
1447 		    comp_rd_index));
1448 	}
1449 
1450 	rcr_p->rcr_desc_rd_head_pp = rcr_desc_rd_head_pp;
1451 	rcr_p->comp_rd_index = comp_rd_index;
1452 	rcr_p->rcr_desc_rd_head_p = rcr_desc_rd_head_p;
1453 
1454 	/* Adjust the mailbox queue length for a hardware bug workaround */
1455 	mboxp->rcrstat_a.bits.qlen -= npkt_read;
1456 
1457 	if ((hxgep->intr_timeout != rcr_p->intr_timeout) ||
1458 	    (hxgep->intr_threshold != rcr_p->intr_threshold)) {
1459 		rcr_p->intr_timeout = hxgep->intr_timeout;
1460 		rcr_p->intr_threshold = hxgep->intr_threshold;
1461 		rcr_cfg_b.value = 0x0ULL;
1462 		if (rcr_p->intr_timeout)
1463 			rcr_cfg_b.bits.entout = 1;
1464 		rcr_cfg_b.bits.timeout = rcr_p->intr_timeout;
1465 		rcr_cfg_b.bits.pthres = rcr_p->intr_threshold;
1466 		RXDMA_REG_WRITE64(handle, RDC_RCR_CFG_B,
1467 		    channel, rcr_cfg_b.value);
1468 	}
1469 
1470 	if (hxgep->rdc_first_intr[channel] && (npkt_read > 0)) {
1471 		hxgep->rdc_first_intr[channel] = B_FALSE;
1472 		cs.bits.pktread = npkt_read - 1;
1473 	} else
1474 		cs.bits.pktread = npkt_read;
1475 	cs.bits.ptrread = nrcr_read;
1476 	cs.value &= 0xffffffffULL;
1477 	RXDMA_REG_WRITE64(handle, RDC_STAT, channel, cs.value);
1478 
1479 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL,
1480 	    "==> hxge_rx_pkts: EXIT: rcr channel %d "
1481 	    "head_pp $%p  index %016llx ",
1482 	    channel, rcr_p->rcr_desc_rd_head_pp, rcr_p->comp_rd_index));
1483 
1484 	/*
1485 	 * Update RCR buffer pointer read and number of packets read.
1486 	 */
1487 	*rcrp = rcr_p;
1488 
1489 	HXGE_DEBUG_MSG((hxgep, RX_INT_CTL, "<== hxge_rx_pkts"));
1490 
1491 	return (head_mp);
1492 }
1493 
1494 #define	RCR_ENTRY_PATTERN	0x5a5a6b6b7c7c8d8dULL
1495 #define	NO_PORT_BIT		0x20
1496 #define	L4_CS_EQ_BIT		0x40
1497 
1498 /*ARGSUSED*/
1499 void
1500 hxge_receive_packet(p_hxge_t hxgep,
1501     p_rx_rcr_ring_t rcr_p, p_rcr_entry_t rcr_desc_rd_head_p,
1502     boolean_t *multi_p, mblk_t **mp, mblk_t **mp_cont,
1503     uint32_t *invalid_rcr_entry)
1504 {
1505 	p_mblk_t		nmp = NULL;
1506 	uint64_t		multi;
1507 	uint8_t			channel;
1508 
1509 	boolean_t first_entry = B_TRUE;
1510 	boolean_t is_tcp_udp = B_FALSE;
1511 	boolean_t buffer_free = B_FALSE;
1512 	boolean_t error_send_up = B_FALSE;
1513 	uint8_t error_type;
1514 	uint16_t l2_len;
1515 	uint16_t skip_len;
1516 	uint8_t pktbufsz_type;
1517 	uint64_t rcr_entry;
1518 	uint64_t *pkt_buf_addr_pp;
1519 	uint64_t *pkt_buf_addr_p;
1520 	uint32_t buf_offset;
1521 	uint32_t bsize;
1522 	uint32_t msg_index;
1523 	p_rx_rbr_ring_t rx_rbr_p;
1524 	p_rx_msg_t *rx_msg_ring_p;
1525 	p_rx_msg_t rx_msg_p;
1526 
1527 	uint16_t sw_offset_bytes = 0, hdr_size = 0;
1528 	hxge_status_t status = HXGE_OK;
1529 	boolean_t is_valid = B_FALSE;
1530 	p_hxge_rx_ring_stats_t rdc_stats;
1531 	uint32_t bytes_read;
1532 	uint8_t header0 = 0;
1533 	uint8_t header1 = 0;
1534 	uint64_t pkt_type;
1535 	uint8_t no_port_bit = 0;
1536 	uint8_t l4_cs_eq_bit = 0;
1537 
1538 	channel = rcr_p->rdc;
1539 
1540 	HXGE_DEBUG_MSG((hxgep, RX2_CTL, "==> hxge_receive_packet"));
1541 
1542 	first_entry = (*mp == NULL) ? B_TRUE : B_FALSE;
1543 	rcr_entry = *((uint64_t *)rcr_desc_rd_head_p);
1544 
1545 	/* Verify the content of the rcr_entry for a hardware bug workaround */
1546 	if ((rcr_entry == 0x0) || (rcr_entry == RCR_ENTRY_PATTERN)) {
1547 		*invalid_rcr_entry = 1;
1548 		HXGE_DEBUG_MSG((hxgep, RX2_CTL, "hxge_receive_packet "
1549 		    "Channel %d invalid RCR entry 0x%llx found, returning\n",
1550 		    channel, (long long) rcr_entry));
1551 		return;
1552 	}
1553 	*((uint64_t *)rcr_desc_rd_head_p) = RCR_ENTRY_PATTERN;
1554 
1555 	multi = (rcr_entry & RCR_MULTI_MASK);
1556 	pkt_type = (rcr_entry & RCR_PKT_TYPE_MASK);
1557 
1558 	error_type = ((rcr_entry & RCR_ERROR_MASK) >> RCR_ERROR_SHIFT);
1559 	l2_len = ((rcr_entry & RCR_L2_LEN_MASK) >> RCR_L2_LEN_SHIFT);
1560 
1561 	/*
1562 	 * Hardware does not strip the CRC due bug ID 11451 where
1563 	 * the hardware mis handles minimum size packets.
1564 	 */
1565 	l2_len -= ETHERFCSL;
1566 
1567 	pktbufsz_type = ((rcr_entry & RCR_PKTBUFSZ_MASK) >>
1568 	    RCR_PKTBUFSZ_SHIFT);
1569 #if defined(__i386)
1570 	pkt_buf_addr_pp = (uint64_t *)(uint32_t)((rcr_entry &
1571 	    RCR_PKT_BUF_ADDR_MASK) << RCR_PKT_BUF_ADDR_SHIFT);
1572 #else
1573 	pkt_buf_addr_pp = (uint64_t *)((rcr_entry & RCR_PKT_BUF_ADDR_MASK) <<
1574 	    RCR_PKT_BUF_ADDR_SHIFT);
1575 #endif
1576 
1577 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1578 	    "==> hxge_receive_packet: entryp $%p entry 0x%0llx "
1579 	    "pkt_buf_addr_pp $%p l2_len %d multi %d "
1580 	    "error_type 0x%x pktbufsz_type %d ",
1581 	    rcr_desc_rd_head_p, rcr_entry, pkt_buf_addr_pp, l2_len,
1582 	    multi, error_type, pktbufsz_type));
1583 
1584 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1585 	    "==> hxge_receive_packet: entryp $%p entry 0x%0llx "
1586 	    "pkt_buf_addr_pp $%p l2_len %d multi %d "
1587 	    "error_type 0x%x ", rcr_desc_rd_head_p,
1588 	    rcr_entry, pkt_buf_addr_pp, l2_len, multi, error_type));
1589 
1590 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1591 	    "==> (rbr) hxge_receive_packet: entry 0x%0llx "
1592 	    "full pkt_buf_addr_pp $%p l2_len %d",
1593 	    rcr_entry, pkt_buf_addr_pp, l2_len));
1594 
1595 	/* get the stats ptr */
1596 	rdc_stats = rcr_p->rdc_stats;
1597 
1598 	if (!l2_len) {
1599 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
1600 		    "<== hxge_receive_packet: failed: l2 length is 0."));
1601 		return;
1602 	}
1603 
1604 	/* shift 6 bits to get the full io address */
1605 #if defined(__i386)
1606 	pkt_buf_addr_pp = (uint64_t *)((uint32_t)pkt_buf_addr_pp <<
1607 	    RCR_PKT_BUF_ADDR_SHIFT_FULL);
1608 #else
1609 	pkt_buf_addr_pp = (uint64_t *)((uint64_t)pkt_buf_addr_pp <<
1610 	    RCR_PKT_BUF_ADDR_SHIFT_FULL);
1611 #endif
1612 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1613 	    "==> (rbr) hxge_receive_packet: entry 0x%0llx "
1614 	    "full pkt_buf_addr_pp $%p l2_len %d",
1615 	    rcr_entry, pkt_buf_addr_pp, l2_len));
1616 
1617 	rx_rbr_p = rcr_p->rx_rbr_p;
1618 	rx_msg_ring_p = rx_rbr_p->rx_msg_ring;
1619 
1620 	if (first_entry) {
1621 		hdr_size = (rcr_p->full_hdr_flag ? RXDMA_HDR_SIZE_FULL :
1622 		    RXDMA_HDR_SIZE_DEFAULT);
1623 
1624 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
1625 		    "==> hxge_receive_packet: first entry 0x%016llx "
1626 		    "pkt_buf_addr_pp $%p l2_len %d hdr %d",
1627 		    rcr_entry, pkt_buf_addr_pp, l2_len, hdr_size));
1628 	}
1629 
1630 	MUTEX_ENTER(&rcr_p->lock);
1631 	MUTEX_ENTER(&rx_rbr_p->lock);
1632 
1633 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
1634 	    "==> (rbr 1) hxge_receive_packet: entry 0x%0llx "
1635 	    "full pkt_buf_addr_pp $%p l2_len %d",
1636 	    rcr_entry, pkt_buf_addr_pp, l2_len));
1637 
1638 	/*
1639 	 * Packet buffer address in the completion entry points to the starting
1640 	 * buffer address (offset 0). Use the starting buffer address to locate
1641 	 * the corresponding kernel address.
1642 	 */
1643 	status = hxge_rxbuf_pp_to_vp(hxgep, rx_rbr_p,
1644 	    pktbufsz_type, pkt_buf_addr_pp, &pkt_buf_addr_p,
1645 	    &buf_offset, &msg_index);
1646 
1647 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
1648 	    "==> (rbr 2) hxge_receive_packet: entry 0x%0llx "
1649 	    "full pkt_buf_addr_pp $%p l2_len %d",
1650 	    rcr_entry, pkt_buf_addr_pp, l2_len));
1651 
1652 	if (status != HXGE_OK) {
1653 		MUTEX_EXIT(&rx_rbr_p->lock);
1654 		MUTEX_EXIT(&rcr_p->lock);
1655 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
1656 		    "<== hxge_receive_packet: found vaddr failed %d", status));
1657 		return;
1658 	}
1659 
1660 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1661 	    "==> (rbr 3) hxge_receive_packet: entry 0x%0llx "
1662 	    "full pkt_buf_addr_pp $%p l2_len %d",
1663 	    rcr_entry, pkt_buf_addr_pp, l2_len));
1664 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1665 	    "==> (rbr 4 msgindex %d) hxge_receive_packet: entry 0x%0llx "
1666 	    "full pkt_buf_addr_pp $%p l2_len %d",
1667 	    msg_index, rcr_entry, pkt_buf_addr_pp, l2_len));
1668 
1669 	if (msg_index >= rx_rbr_p->tnblocks) {
1670 		MUTEX_EXIT(&rx_rbr_p->lock);
1671 		MUTEX_EXIT(&rcr_p->lock);
1672 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1673 		    "==> hxge_receive_packet: FATAL msg_index (%d) "
1674 		    "should be smaller than tnblocks (%d)\n",
1675 		    msg_index, rx_rbr_p->tnblocks));
1676 		return;
1677 	}
1678 
1679 	rx_msg_p = rx_msg_ring_p[msg_index];
1680 
1681 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1682 	    "==> (rbr 4 msgindex %d) hxge_receive_packet: entry 0x%0llx "
1683 	    "full pkt_buf_addr_pp $%p l2_len %d",
1684 	    msg_index, rcr_entry, pkt_buf_addr_pp, l2_len));
1685 
1686 	switch (pktbufsz_type) {
1687 	case RCR_PKTBUFSZ_0:
1688 		bsize = rx_rbr_p->pkt_buf_size0_bytes;
1689 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1690 		    "==> hxge_receive_packet: 0 buf %d", bsize));
1691 		break;
1692 	case RCR_PKTBUFSZ_1:
1693 		bsize = rx_rbr_p->pkt_buf_size1_bytes;
1694 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1695 		    "==> hxge_receive_packet: 1 buf %d", bsize));
1696 		break;
1697 	case RCR_PKTBUFSZ_2:
1698 		bsize = rx_rbr_p->pkt_buf_size2_bytes;
1699 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
1700 		    "==> hxge_receive_packet: 2 buf %d", bsize));
1701 		break;
1702 	case RCR_SINGLE_BLOCK:
1703 		bsize = rx_msg_p->block_size;
1704 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1705 		    "==> hxge_receive_packet: single %d", bsize));
1706 
1707 		break;
1708 	default:
1709 		MUTEX_EXIT(&rx_rbr_p->lock);
1710 		MUTEX_EXIT(&rcr_p->lock);
1711 		return;
1712 	}
1713 
1714 	DMA_COMMON_SYNC_OFFSET(rx_msg_p->buf_dma,
1715 	    (buf_offset + sw_offset_bytes), (hdr_size + l2_len),
1716 	    DDI_DMA_SYNC_FORCPU);
1717 
1718 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1719 	    "==> hxge_receive_packet: after first dump:usage count"));
1720 
1721 	if (rx_msg_p->cur_usage_cnt == 0) {
1722 		atomic_inc_32(&rx_rbr_p->rbr_used);
1723 		if (rx_rbr_p->rbr_use_bcopy) {
1724 			atomic_inc_32(&rx_rbr_p->rbr_consumed);
1725 			if (rx_rbr_p->rbr_consumed <
1726 			    rx_rbr_p->rbr_threshold_hi) {
1727 				if (rx_rbr_p->rbr_threshold_lo == 0 ||
1728 				    ((rx_rbr_p->rbr_consumed >=
1729 				    rx_rbr_p->rbr_threshold_lo) &&
1730 				    (rx_rbr_p->rbr_bufsize_type >=
1731 				    pktbufsz_type))) {
1732 					rx_msg_p->rx_use_bcopy = B_TRUE;
1733 				}
1734 			} else {
1735 				rx_msg_p->rx_use_bcopy = B_TRUE;
1736 			}
1737 		}
1738 		HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1739 		    "==> hxge_receive_packet: buf %d (new block) ", bsize));
1740 
1741 		rx_msg_p->pkt_buf_size_code = pktbufsz_type;
1742 		rx_msg_p->pkt_buf_size = bsize;
1743 		rx_msg_p->cur_usage_cnt = 1;
1744 		if (pktbufsz_type == RCR_SINGLE_BLOCK) {
1745 			HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1746 			    "==> hxge_receive_packet: buf %d (single block) ",
1747 			    bsize));
1748 			/*
1749 			 * Buffer can be reused once the free function is
1750 			 * called.
1751 			 */
1752 			rx_msg_p->max_usage_cnt = 1;
1753 			buffer_free = B_TRUE;
1754 		} else {
1755 			rx_msg_p->max_usage_cnt = rx_msg_p->block_size / bsize;
1756 			if (rx_msg_p->max_usage_cnt == 1) {
1757 				buffer_free = B_TRUE;
1758 			}
1759 		}
1760 	} else {
1761 		rx_msg_p->cur_usage_cnt++;
1762 		if (rx_msg_p->cur_usage_cnt == rx_msg_p->max_usage_cnt) {
1763 			buffer_free = B_TRUE;
1764 		}
1765 	}
1766 
1767 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
1768 	    "msgbuf index = %d l2len %d bytes usage %d max_usage %d ",
1769 	    msg_index, l2_len,
1770 	    rx_msg_p->cur_usage_cnt, rx_msg_p->max_usage_cnt));
1771 
1772 	if (error_type) {
1773 		rdc_stats->ierrors++;
1774 		/* Update error stats */
1775 		rdc_stats->errlog.compl_err_type = error_type;
1776 		HXGE_FM_REPORT_ERROR(hxgep, NULL, HXGE_FM_EREPORT_RDMC_RCR_ERR);
1777 
1778 		if (error_type & RCR_CTRL_FIFO_DED) {
1779 			rdc_stats->ctrl_fifo_ecc_err++;
1780 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
1781 			    " hxge_receive_packet: "
1782 			    " channel %d RCR ctrl_fifo_ded error", channel));
1783 		} else if (error_type & RCR_DATA_FIFO_DED) {
1784 			rdc_stats->data_fifo_ecc_err++;
1785 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
1786 			    " hxge_receive_packet: channel %d"
1787 			    " RCR data_fifo_ded error", channel));
1788 		}
1789 
1790 		/*
1791 		 * Update and repost buffer block if max usage count is
1792 		 * reached.
1793 		 */
1794 		if (error_send_up == B_FALSE) {
1795 			atomic_inc_32(&rx_msg_p->ref_cnt);
1796 			if (buffer_free == B_TRUE) {
1797 				rx_msg_p->free = B_TRUE;
1798 			}
1799 
1800 			MUTEX_EXIT(&rx_rbr_p->lock);
1801 			MUTEX_EXIT(&rcr_p->lock);
1802 			hxge_freeb(rx_msg_p);
1803 			return;
1804 		}
1805 	}
1806 
1807 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1808 	    "==> hxge_receive_packet: DMA sync second "));
1809 
1810 	bytes_read = rcr_p->rcvd_pkt_bytes;
1811 	skip_len = sw_offset_bytes + hdr_size;
1812 
1813 	if (first_entry) {
1814 		header0 = rx_msg_p->buffer[buf_offset];
1815 		no_port_bit = header0 & NO_PORT_BIT;
1816 
1817 		header1 = rx_msg_p->buffer[buf_offset + 1];
1818 		l4_cs_eq_bit = header1 & L4_CS_EQ_BIT;
1819 	}
1820 
1821 	if (!rx_msg_p->rx_use_bcopy) {
1822 		/*
1823 		 * For loaned up buffers, the driver reference count
1824 		 * will be incremented first and then the free state.
1825 		 */
1826 		if ((nmp = hxge_dupb(rx_msg_p, buf_offset, bsize)) != NULL) {
1827 			if (first_entry) {
1828 				nmp->b_rptr = &nmp->b_rptr[skip_len];
1829 				if (l2_len < bsize - skip_len) {
1830 					nmp->b_wptr = &nmp->b_rptr[l2_len];
1831 				} else {
1832 					nmp->b_wptr = &nmp->b_rptr[bsize
1833 					    - skip_len];
1834 				}
1835 			} else {
1836 				if (l2_len - bytes_read < bsize) {
1837 					nmp->b_wptr =
1838 					    &nmp->b_rptr[l2_len - bytes_read];
1839 				} else {
1840 					nmp->b_wptr = &nmp->b_rptr[bsize];
1841 				}
1842 			}
1843 		}
1844 	} else {
1845 		if (first_entry) {
1846 			nmp = hxge_dupb_bcopy(rx_msg_p, buf_offset + skip_len,
1847 			    l2_len < bsize - skip_len ?
1848 			    l2_len : bsize - skip_len);
1849 		} else {
1850 			nmp = hxge_dupb_bcopy(rx_msg_p, buf_offset,
1851 			    l2_len - bytes_read < bsize ?
1852 			    l2_len - bytes_read : bsize);
1853 		}
1854 	}
1855 
1856 	if (nmp != NULL) {
1857 		if (first_entry)
1858 			bytes_read  = nmp->b_wptr - nmp->b_rptr;
1859 		else
1860 			bytes_read += nmp->b_wptr - nmp->b_rptr;
1861 
1862 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
1863 		    "==> hxge_receive_packet after dupb: "
1864 		    "rbr consumed %d "
1865 		    "pktbufsz_type %d "
1866 		    "nmp $%p rptr $%p wptr $%p "
1867 		    "buf_offset %d bzise %d l2_len %d skip_len %d",
1868 		    rx_rbr_p->rbr_consumed,
1869 		    pktbufsz_type,
1870 		    nmp, nmp->b_rptr, nmp->b_wptr,
1871 		    buf_offset, bsize, l2_len, skip_len));
1872 	} else {
1873 		cmn_err(CE_WARN, "!hxge_receive_packet: update stats (error)");
1874 
1875 		atomic_inc_32(&rx_msg_p->ref_cnt);
1876 		if (buffer_free == B_TRUE) {
1877 			rx_msg_p->free = B_TRUE;
1878 		}
1879 
1880 		MUTEX_EXIT(&rx_rbr_p->lock);
1881 		MUTEX_EXIT(&rcr_p->lock);
1882 		hxge_freeb(rx_msg_p);
1883 		return;
1884 	}
1885 
1886 	if (buffer_free == B_TRUE) {
1887 		rx_msg_p->free = B_TRUE;
1888 	}
1889 
1890 	/*
1891 	 * ERROR, FRAG and PKT_TYPE are only reported in the first entry. If a
1892 	 * packet is not fragmented and no error bit is set, then L4 checksum
1893 	 * is OK.
1894 	 */
1895 	is_valid = (nmp != NULL);
1896 	if (first_entry) {
1897 		rdc_stats->ipackets++; /* count only 1st seg for jumbo */
1898 		if (l2_len > (STD_FRAME_SIZE - ETHERFCSL))
1899 			rdc_stats->jumbo_pkts++;
1900 		rdc_stats->ibytes += skip_len + l2_len < bsize ?
1901 		    l2_len : bsize;
1902 	} else {
1903 		/*
1904 		 * Add the current portion of the packet to the kstats.
1905 		 * The current portion of the packet is calculated by using
1906 		 * length of the packet and the previously received portion.
1907 		 */
1908 		rdc_stats->ibytes += l2_len - rcr_p->rcvd_pkt_bytes < bsize ?
1909 		    l2_len - rcr_p->rcvd_pkt_bytes : bsize;
1910 	}
1911 
1912 	rcr_p->rcvd_pkt_bytes = bytes_read;
1913 
1914 	if (rx_msg_p->free && rx_msg_p->rx_use_bcopy) {
1915 		atomic_inc_32(&rx_msg_p->ref_cnt);
1916 		MUTEX_EXIT(&rx_rbr_p->lock);
1917 		MUTEX_EXIT(&rcr_p->lock);
1918 		hxge_freeb(rx_msg_p);
1919 	} else {
1920 		MUTEX_EXIT(&rx_rbr_p->lock);
1921 		MUTEX_EXIT(&rcr_p->lock);
1922 	}
1923 
1924 	if (is_valid) {
1925 		nmp->b_cont = NULL;
1926 		if (first_entry) {
1927 			*mp = nmp;
1928 			*mp_cont = NULL;
1929 		} else {
1930 			*mp_cont = nmp;
1931 		}
1932 	}
1933 
1934 	/*
1935 	 * Update stats and hardware checksuming.
1936 	 */
1937 	if (is_valid && !multi) {
1938 		is_tcp_udp = ((pkt_type == RCR_PKT_IS_TCP ||
1939 		    pkt_type == RCR_PKT_IS_UDP) ? B_TRUE : B_FALSE);
1940 
1941 		if (!no_port_bit && l4_cs_eq_bit && is_tcp_udp && !error_type) {
1942 			(void) hcksum_assoc(nmp, NULL, NULL, 0, 0, 0, 0,
1943 			    HCK_FULLCKSUM_OK | HCK_FULLCKSUM, 0);
1944 
1945 			HXGE_DEBUG_MSG((hxgep, RX_CTL,
1946 			    "==> hxge_receive_packet: Full tcp/udp cksum "
1947 			    "is_valid 0x%x multi %d error %d",
1948 			    is_valid, multi, error_type));
1949 		}
1950 	}
1951 
1952 	HXGE_DEBUG_MSG((hxgep, RX2_CTL,
1953 	    "==> hxge_receive_packet: *mp 0x%016llx", *mp));
1954 
1955 	*multi_p = (multi == RCR_MULTI_MASK);
1956 
1957 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_receive_packet: "
1958 	    "multi %d nmp 0x%016llx *mp 0x%016llx *mp_cont 0x%016llx",
1959 	    *multi_p, nmp, *mp, *mp_cont));
1960 }
1961 
1962 static void
1963 hxge_rx_rbr_empty_recover(p_hxge_t hxgep, uint8_t channel)
1964 {
1965 	hpi_handle_t	handle;
1966 	p_rx_rcr_ring_t	rcrp;
1967 	p_rx_rbr_ring_t	rbrp;
1968 
1969 	rcrp = hxgep->rx_rcr_rings->rcr_rings[channel];
1970 	rbrp = rcrp->rx_rbr_p;
1971 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
1972 
1973 	/*
1974 	 * Wait for the channel to be quiet
1975 	 */
1976 	(void) hpi_rxdma_cfg_rdc_wait_for_qst(handle, channel);
1977 
1978 	/*
1979 	 * Post page will accumulate some buffers before re-enabling
1980 	 * the DMA channel.
1981 	 */
1982 
1983 	MUTEX_ENTER(&rbrp->post_lock);
1984 	if ((rbrp->rbb_max - rbrp->rbr_used) >= HXGE_RBR_EMPTY_THRESHOLD) {
1985 		hxge_rbr_empty_restore(hxgep, rbrp);
1986 	} else {
1987 		rbrp->rbr_is_empty = B_TRUE;
1988 	}
1989 	MUTEX_EXIT(&rbrp->post_lock);
1990 }
1991 
1992 /*ARGSUSED*/
1993 static hxge_status_t
1994 hxge_rx_err_evnts(p_hxge_t hxgep, uint_t index, p_hxge_ldv_t ldvp,
1995     rdc_stat_t cs)
1996 {
1997 	p_hxge_rx_ring_stats_t	rdc_stats;
1998 	hpi_handle_t		handle;
1999 	boolean_t		rxchan_fatal = B_FALSE;
2000 	uint8_t			channel;
2001 	hxge_status_t		status = HXGE_OK;
2002 	uint64_t		cs_val;
2003 
2004 	HXGE_DEBUG_MSG((hxgep, INT_CTL, "==> hxge_rx_err_evnts"));
2005 
2006 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
2007 	channel = ldvp->channel;
2008 
2009 	/* Clear the interrupts */
2010 	cs.bits.pktread = 0;
2011 	cs.bits.ptrread = 0;
2012 	cs_val = cs.value & RDC_STAT_WR1C;
2013 	RXDMA_REG_WRITE64(handle, RDC_STAT, channel, cs_val);
2014 
2015 	rdc_stats = &hxgep->statsp->rdc_stats[ldvp->vdma_index];
2016 
2017 	if (cs.bits.rbr_cpl_to) {
2018 		rdc_stats->rbr_tmout++;
2019 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2020 		    HXGE_FM_EREPORT_RDMC_RBR_CPL_TO);
2021 		rxchan_fatal = B_TRUE;
2022 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2023 		    "==> hxge_rx_err_evnts(channel %d): "
2024 		    "fatal error: rx_rbr_timeout", channel));
2025 	}
2026 
2027 	if ((cs.bits.rcr_shadow_par_err) || (cs.bits.rbr_prefetch_par_err)) {
2028 		(void) hpi_rxdma_ring_perr_stat_get(handle,
2029 		    &rdc_stats->errlog.pre_par, &rdc_stats->errlog.sha_par);
2030 	}
2031 
2032 	if (cs.bits.rcr_shadow_par_err) {
2033 		rdc_stats->rcr_sha_par++;
2034 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2035 		    HXGE_FM_EREPORT_RDMC_RCR_SHA_PAR);
2036 		rxchan_fatal = B_TRUE;
2037 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2038 		    "==> hxge_rx_err_evnts(channel %d): "
2039 		    "fatal error: rcr_shadow_par_err", channel));
2040 	}
2041 
2042 	if (cs.bits.rbr_prefetch_par_err) {
2043 		rdc_stats->rbr_pre_par++;
2044 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2045 		    HXGE_FM_EREPORT_RDMC_RBR_PRE_PAR);
2046 		rxchan_fatal = B_TRUE;
2047 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2048 		    "==> hxge_rx_err_evnts(channel %d): "
2049 		    "fatal error: rbr_prefetch_par_err", channel));
2050 	}
2051 
2052 	if (cs.bits.rbr_pre_empty) {
2053 		rdc_stats->rbr_pre_empty++;
2054 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2055 		    HXGE_FM_EREPORT_RDMC_RBR_PRE_EMPTY);
2056 		rxchan_fatal = B_TRUE;
2057 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2058 		    "==> hxge_rx_err_evnts(channel %d): "
2059 		    "fatal error: rbr_pre_empty", channel));
2060 	}
2061 
2062 	if (cs.bits.peu_resp_err) {
2063 		rdc_stats->peu_resp_err++;
2064 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2065 		    HXGE_FM_EREPORT_RDMC_PEU_RESP_ERR);
2066 		rxchan_fatal = B_TRUE;
2067 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2068 		    "==> hxge_rx_err_evnts(channel %d): "
2069 		    "fatal error: peu_resp_err", channel));
2070 	}
2071 
2072 	if (cs.bits.rcr_thres) {
2073 		rdc_stats->rcr_thres++;
2074 	}
2075 
2076 	if (cs.bits.rcr_to) {
2077 		rdc_stats->rcr_to++;
2078 	}
2079 
2080 	if (cs.bits.rcr_shadow_full) {
2081 		rdc_stats->rcr_shadow_full++;
2082 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2083 		    HXGE_FM_EREPORT_RDMC_RCR_SHA_FULL);
2084 		rxchan_fatal = B_TRUE;
2085 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2086 		    "==> hxge_rx_err_evnts(channel %d): "
2087 		    "fatal error: rcr_shadow_full", channel));
2088 	}
2089 
2090 	if (cs.bits.rcr_full) {
2091 		rdc_stats->rcrfull++;
2092 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2093 		    HXGE_FM_EREPORT_RDMC_RCRFULL);
2094 		rxchan_fatal = B_TRUE;
2095 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2096 		    "==> hxge_rx_err_evnts(channel %d): "
2097 		    "fatal error: rcrfull error", channel));
2098 	}
2099 
2100 	if (cs.bits.rbr_empty) {
2101 		rdc_stats->rbr_empty++;
2102 		hxge_rx_rbr_empty_recover(hxgep, channel);
2103 	}
2104 
2105 	if (cs.bits.rbr_full) {
2106 		rdc_stats->rbrfull++;
2107 		HXGE_FM_REPORT_ERROR(hxgep, channel,
2108 		    HXGE_FM_EREPORT_RDMC_RBRFULL);
2109 		rxchan_fatal = B_TRUE;
2110 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2111 		    "==> hxge_rx_err_evnts(channel %d): "
2112 		    "fatal error: rbr_full error", channel));
2113 	}
2114 
2115 	if (rxchan_fatal) {
2116 		p_rx_rcr_ring_t	rcrp;
2117 		p_rx_rbr_ring_t rbrp;
2118 
2119 		rcrp = hxgep->rx_rcr_rings->rcr_rings[channel];
2120 		rbrp = rcrp->rx_rbr_p;
2121 
2122 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2123 		    " hxge_rx_err_evnts: fatal error on Channel #%d\n",
2124 		    channel));
2125 		MUTEX_ENTER(&rbrp->post_lock);
2126 		/* This function needs to be inside the post_lock */
2127 		status = hxge_rxdma_fatal_err_recover(hxgep, channel);
2128 		MUTEX_EXIT(&rbrp->post_lock);
2129 		if (status == HXGE_OK) {
2130 			FM_SERVICE_RESTORED(hxgep);
2131 		}
2132 	}
2133 	HXGE_DEBUG_MSG((hxgep, INT_CTL, "<== hxge_rx_err_evnts"));
2134 
2135 	return (status);
2136 }
2137 
2138 static hxge_status_t
2139 hxge_map_rxdma(p_hxge_t hxgep)
2140 {
2141 	int			i, ndmas;
2142 	uint16_t		channel;
2143 	p_rx_rbr_rings_t	rx_rbr_rings;
2144 	p_rx_rbr_ring_t		*rbr_rings;
2145 	p_rx_rcr_rings_t	rx_rcr_rings;
2146 	p_rx_rcr_ring_t		*rcr_rings;
2147 	p_rx_mbox_areas_t	rx_mbox_areas_p;
2148 	p_rx_mbox_t		*rx_mbox_p;
2149 	p_hxge_dma_pool_t	dma_buf_poolp;
2150 	p_hxge_dma_common_t	*dma_buf_p;
2151 	p_hxge_dma_pool_t	dma_rbr_cntl_poolp;
2152 	p_hxge_dma_common_t	*dma_rbr_cntl_p;
2153 	p_hxge_dma_pool_t	dma_rcr_cntl_poolp;
2154 	p_hxge_dma_common_t	*dma_rcr_cntl_p;
2155 	p_hxge_dma_pool_t	dma_mbox_cntl_poolp;
2156 	p_hxge_dma_common_t	*dma_mbox_cntl_p;
2157 	uint32_t		*num_chunks;
2158 	hxge_status_t		status = HXGE_OK;
2159 
2160 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_map_rxdma"));
2161 
2162 	dma_buf_poolp = hxgep->rx_buf_pool_p;
2163 	dma_rbr_cntl_poolp = hxgep->rx_rbr_cntl_pool_p;
2164 	dma_rcr_cntl_poolp = hxgep->rx_rcr_cntl_pool_p;
2165 	dma_mbox_cntl_poolp = hxgep->rx_mbox_cntl_pool_p;
2166 
2167 	if (!dma_buf_poolp->buf_allocated ||
2168 	    !dma_rbr_cntl_poolp->buf_allocated ||
2169 	    !dma_rcr_cntl_poolp->buf_allocated ||
2170 	    !dma_mbox_cntl_poolp->buf_allocated) {
2171 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2172 		    "<== hxge_map_rxdma: buf not allocated"));
2173 		return (HXGE_ERROR);
2174 	}
2175 
2176 	ndmas = dma_buf_poolp->ndmas;
2177 	if (!ndmas) {
2178 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
2179 		    "<== hxge_map_rxdma: no dma allocated"));
2180 		return (HXGE_ERROR);
2181 	}
2182 
2183 	num_chunks = dma_buf_poolp->num_chunks;
2184 	dma_buf_p = dma_buf_poolp->dma_buf_pool_p;
2185 	dma_rbr_cntl_p = dma_rbr_cntl_poolp->dma_buf_pool_p;
2186 	dma_rcr_cntl_p = dma_rcr_cntl_poolp->dma_buf_pool_p;
2187 	dma_mbox_cntl_p = dma_mbox_cntl_poolp->dma_buf_pool_p;
2188 
2189 	rx_rbr_rings = (p_rx_rbr_rings_t)
2190 	    KMEM_ZALLOC(sizeof (rx_rbr_rings_t), KM_SLEEP);
2191 	rbr_rings = (p_rx_rbr_ring_t *)KMEM_ZALLOC(
2192 	    sizeof (p_rx_rbr_ring_t) * ndmas, KM_SLEEP);
2193 
2194 	rx_rcr_rings = (p_rx_rcr_rings_t)
2195 	    KMEM_ZALLOC(sizeof (rx_rcr_rings_t), KM_SLEEP);
2196 	rcr_rings = (p_rx_rcr_ring_t *)KMEM_ZALLOC(
2197 	    sizeof (p_rx_rcr_ring_t) * ndmas, KM_SLEEP);
2198 
2199 	rx_mbox_areas_p = (p_rx_mbox_areas_t)
2200 	    KMEM_ZALLOC(sizeof (rx_mbox_areas_t), KM_SLEEP);
2201 	rx_mbox_p = (p_rx_mbox_t *)KMEM_ZALLOC(
2202 	    sizeof (p_rx_mbox_t) * ndmas, KM_SLEEP);
2203 
2204 	/*
2205 	 * Timeout should be set based on the system clock divider.
2206 	 * The following timeout value of 1 assumes that the
2207 	 * granularity (1000) is 3 microseconds running at 300MHz.
2208 	 */
2209 
2210 	hxgep->intr_threshold = RXDMA_RCR_PTHRES_DEFAULT;
2211 	hxgep->intr_timeout = RXDMA_RCR_TO_DEFAULT;
2212 
2213 	/*
2214 	 * Map descriptors from the buffer polls for each dam channel.
2215 	 */
2216 	for (i = 0; i < ndmas; i++) {
2217 		/*
2218 		 * Set up and prepare buffer blocks, descriptors and mailbox.
2219 		 */
2220 		channel = ((p_hxge_dma_common_t)dma_buf_p[i])->dma_channel;
2221 		status = hxge_map_rxdma_channel(hxgep, channel,
2222 		    (p_hxge_dma_common_t *)&dma_buf_p[i],
2223 		    (p_rx_rbr_ring_t *)&rbr_rings[i],
2224 		    num_chunks[i],
2225 		    (p_hxge_dma_common_t *)&dma_rbr_cntl_p[i],
2226 		    (p_hxge_dma_common_t *)&dma_rcr_cntl_p[i],
2227 		    (p_hxge_dma_common_t *)&dma_mbox_cntl_p[i],
2228 		    (p_rx_rcr_ring_t *)&rcr_rings[i],
2229 		    (p_rx_mbox_t *)&rx_mbox_p[i]);
2230 		if (status != HXGE_OK) {
2231 			goto hxge_map_rxdma_fail1;
2232 		}
2233 		rbr_rings[i]->index = (uint16_t)i;
2234 		rcr_rings[i]->index = (uint16_t)i;
2235 		rcr_rings[i]->rdc_stats = &hxgep->statsp->rdc_stats[i];
2236 	}
2237 
2238 	rx_rbr_rings->ndmas = rx_rcr_rings->ndmas = ndmas;
2239 	rx_rbr_rings->rbr_rings = rbr_rings;
2240 	hxgep->rx_rbr_rings = rx_rbr_rings;
2241 	rx_rcr_rings->rcr_rings = rcr_rings;
2242 	hxgep->rx_rcr_rings = rx_rcr_rings;
2243 
2244 	rx_mbox_areas_p->rxmbox_areas = rx_mbox_p;
2245 	hxgep->rx_mbox_areas_p = rx_mbox_areas_p;
2246 
2247 	goto hxge_map_rxdma_exit;
2248 
2249 hxge_map_rxdma_fail1:
2250 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2251 	    "==> hxge_map_rxdma: unmap rbr,rcr (status 0x%x channel %d i %d)",
2252 	    status, channel, i));
2253 	i--;
2254 	for (; i >= 0; i--) {
2255 		channel = ((p_hxge_dma_common_t)dma_buf_p[i])->dma_channel;
2256 		hxge_unmap_rxdma_channel(hxgep, channel,
2257 		    rbr_rings[i], rcr_rings[i], rx_mbox_p[i]);
2258 	}
2259 
2260 	KMEM_FREE(rbr_rings, sizeof (p_rx_rbr_ring_t) * ndmas);
2261 	KMEM_FREE(rx_rbr_rings, sizeof (rx_rbr_rings_t));
2262 	KMEM_FREE(rcr_rings, sizeof (p_rx_rcr_ring_t) * ndmas);
2263 	KMEM_FREE(rx_rcr_rings, sizeof (rx_rcr_rings_t));
2264 	KMEM_FREE(rx_mbox_p, sizeof (p_rx_mbox_t) * ndmas);
2265 	KMEM_FREE(rx_mbox_areas_p, sizeof (rx_mbox_areas_t));
2266 
2267 hxge_map_rxdma_exit:
2268 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2269 	    "<== hxge_map_rxdma: (status 0x%x channel %d)", status, channel));
2270 
2271 	return (status);
2272 }
2273 
2274 static void
2275 hxge_unmap_rxdma(p_hxge_t hxgep)
2276 {
2277 	int			i, ndmas;
2278 	uint16_t		channel;
2279 	p_rx_rbr_rings_t	rx_rbr_rings;
2280 	p_rx_rbr_ring_t		*rbr_rings;
2281 	p_rx_rcr_rings_t	rx_rcr_rings;
2282 	p_rx_rcr_ring_t		*rcr_rings;
2283 	p_rx_mbox_areas_t	rx_mbox_areas_p;
2284 	p_rx_mbox_t		*rx_mbox_p;
2285 	p_hxge_dma_pool_t	dma_buf_poolp;
2286 	p_hxge_dma_pool_t	dma_rbr_cntl_poolp;
2287 	p_hxge_dma_pool_t	dma_rcr_cntl_poolp;
2288 	p_hxge_dma_pool_t	dma_mbox_cntl_poolp;
2289 	p_hxge_dma_common_t	*dma_buf_p;
2290 
2291 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_unmap_rxdma"));
2292 
2293 	dma_buf_poolp = hxgep->rx_buf_pool_p;
2294 	dma_rbr_cntl_poolp = hxgep->rx_rbr_cntl_pool_p;
2295 	dma_rcr_cntl_poolp = hxgep->rx_rcr_cntl_pool_p;
2296 	dma_mbox_cntl_poolp = hxgep->rx_mbox_cntl_pool_p;
2297 
2298 	if (!dma_buf_poolp->buf_allocated ||
2299 	    !dma_rbr_cntl_poolp->buf_allocated ||
2300 	    !dma_rcr_cntl_poolp->buf_allocated ||
2301 	    !dma_mbox_cntl_poolp->buf_allocated) {
2302 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2303 		    "<== hxge_unmap_rxdma: NULL buf pointers"));
2304 		return;
2305 	}
2306 
2307 	rx_rbr_rings = hxgep->rx_rbr_rings;
2308 	rx_rcr_rings = hxgep->rx_rcr_rings;
2309 	if (rx_rbr_rings == NULL || rx_rcr_rings == NULL) {
2310 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2311 		    "<== hxge_unmap_rxdma: NULL pointers"));
2312 		return;
2313 	}
2314 
2315 	ndmas = rx_rbr_rings->ndmas;
2316 	if (!ndmas) {
2317 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2318 		    "<== hxge_unmap_rxdma: no channel"));
2319 		return;
2320 	}
2321 
2322 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2323 	    "==> hxge_unmap_rxdma (ndmas %d)", ndmas));
2324 
2325 	rbr_rings = rx_rbr_rings->rbr_rings;
2326 	rcr_rings = rx_rcr_rings->rcr_rings;
2327 	rx_mbox_areas_p = hxgep->rx_mbox_areas_p;
2328 	rx_mbox_p = rx_mbox_areas_p->rxmbox_areas;
2329 	dma_buf_p = dma_buf_poolp->dma_buf_pool_p;
2330 
2331 	for (i = 0; i < ndmas; i++) {
2332 		channel = ((p_hxge_dma_common_t)dma_buf_p[i])->dma_channel;
2333 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2334 		    "==> hxge_unmap_rxdma (ndmas %d) channel %d",
2335 		    ndmas, channel));
2336 		(void) hxge_unmap_rxdma_channel(hxgep, channel,
2337 		    (p_rx_rbr_ring_t)rbr_rings[i],
2338 		    (p_rx_rcr_ring_t)rcr_rings[i],
2339 		    (p_rx_mbox_t)rx_mbox_p[i]);
2340 	}
2341 
2342 	KMEM_FREE(rx_rbr_rings, sizeof (rx_rbr_rings_t));
2343 	KMEM_FREE(rbr_rings, sizeof (p_rx_rbr_ring_t) * ndmas);
2344 	KMEM_FREE(rx_rcr_rings, sizeof (rx_rcr_rings_t));
2345 	KMEM_FREE(rcr_rings, sizeof (p_rx_rcr_ring_t) * ndmas);
2346 	KMEM_FREE(rx_mbox_areas_p, sizeof (rx_mbox_areas_t));
2347 	KMEM_FREE(rx_mbox_p, sizeof (p_rx_mbox_t) * ndmas);
2348 
2349 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "<== hxge_unmap_rxdma"));
2350 }
2351 
2352 hxge_status_t
2353 hxge_map_rxdma_channel(p_hxge_t hxgep, uint16_t channel,
2354     p_hxge_dma_common_t *dma_buf_p, p_rx_rbr_ring_t *rbr_p,
2355     uint32_t num_chunks, p_hxge_dma_common_t *dma_rbr_cntl_p,
2356     p_hxge_dma_common_t *dma_rcr_cntl_p, p_hxge_dma_common_t *dma_mbox_cntl_p,
2357     p_rx_rcr_ring_t *rcr_p, p_rx_mbox_t *rx_mbox_p)
2358 {
2359 	int status = HXGE_OK;
2360 
2361 	/*
2362 	 * Set up and prepare buffer blocks, descriptors and mailbox.
2363 	 */
2364 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2365 	    "==> hxge_map_rxdma_channel (channel %d)", channel));
2366 
2367 	/*
2368 	 * Receive buffer blocks
2369 	 */
2370 	status = hxge_map_rxdma_channel_buf_ring(hxgep, channel,
2371 	    dma_buf_p, rbr_p, num_chunks);
2372 	if (status != HXGE_OK) {
2373 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2374 		    "==> hxge_map_rxdma_channel (channel %d): "
2375 		    "map buffer failed 0x%x", channel, status));
2376 		goto hxge_map_rxdma_channel_exit;
2377 	}
2378 
2379 	/*
2380 	 * Receive block ring, completion ring and mailbox.
2381 	 */
2382 	status = hxge_map_rxdma_channel_cfg_ring(hxgep, channel,
2383 	    dma_rbr_cntl_p, dma_rcr_cntl_p, dma_mbox_cntl_p,
2384 	    rbr_p, rcr_p, rx_mbox_p);
2385 	if (status != HXGE_OK) {
2386 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2387 		    "==> hxge_map_rxdma_channel (channel %d): "
2388 		    "map config failed 0x%x", channel, status));
2389 		goto hxge_map_rxdma_channel_fail2;
2390 	}
2391 	goto hxge_map_rxdma_channel_exit;
2392 
2393 hxge_map_rxdma_channel_fail3:
2394 	/* Free rbr, rcr */
2395 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2396 	    "==> hxge_map_rxdma_channel: free rbr/rcr (status 0x%x channel %d)",
2397 	    status, channel));
2398 	hxge_unmap_rxdma_channel_cfg_ring(hxgep, *rcr_p, *rx_mbox_p);
2399 
2400 hxge_map_rxdma_channel_fail2:
2401 	/* Free buffer blocks */
2402 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2403 	    "==> hxge_map_rxdma_channel: free rx buffers"
2404 	    "(hxgep 0x%x status 0x%x channel %d)",
2405 	    hxgep, status, channel));
2406 	hxge_unmap_rxdma_channel_buf_ring(hxgep, *rbr_p);
2407 
2408 	status = HXGE_ERROR;
2409 
2410 hxge_map_rxdma_channel_exit:
2411 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2412 	    "<== hxge_map_rxdma_channel: (hxgep 0x%x status 0x%x channel %d)",
2413 	    hxgep, status, channel));
2414 
2415 	return (status);
2416 }
2417 
2418 /*ARGSUSED*/
2419 static void
2420 hxge_unmap_rxdma_channel(p_hxge_t hxgep, uint16_t channel,
2421     p_rx_rbr_ring_t rbr_p, p_rx_rcr_ring_t rcr_p, p_rx_mbox_t rx_mbox_p)
2422 {
2423 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2424 	    "==> hxge_unmap_rxdma_channel (channel %d)", channel));
2425 
2426 	/*
2427 	 * unmap receive block ring, completion ring and mailbox.
2428 	 */
2429 	(void) hxge_unmap_rxdma_channel_cfg_ring(hxgep, rcr_p, rx_mbox_p);
2430 
2431 	/* unmap buffer blocks */
2432 	(void) hxge_unmap_rxdma_channel_buf_ring(hxgep, rbr_p);
2433 
2434 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "<== hxge_unmap_rxdma_channel"));
2435 }
2436 
2437 /*ARGSUSED*/
2438 static hxge_status_t
2439 hxge_map_rxdma_channel_cfg_ring(p_hxge_t hxgep, uint16_t dma_channel,
2440     p_hxge_dma_common_t *dma_rbr_cntl_p, p_hxge_dma_common_t *dma_rcr_cntl_p,
2441     p_hxge_dma_common_t *dma_mbox_cntl_p, p_rx_rbr_ring_t *rbr_p,
2442     p_rx_rcr_ring_t *rcr_p, p_rx_mbox_t *rx_mbox_p)
2443 {
2444 	p_rx_rbr_ring_t 	rbrp;
2445 	p_rx_rcr_ring_t 	rcrp;
2446 	p_rx_mbox_t 		mboxp;
2447 	p_hxge_dma_common_t 	cntl_dmap;
2448 	p_hxge_dma_common_t 	dmap;
2449 	p_rx_msg_t 		*rx_msg_ring;
2450 	p_rx_msg_t 		rx_msg_p;
2451 	rdc_rbr_cfg_a_t		*rcfga_p;
2452 	rdc_rbr_cfg_b_t		*rcfgb_p;
2453 	rdc_rcr_cfg_a_t		*cfga_p;
2454 	rdc_rcr_cfg_b_t		*cfgb_p;
2455 	rdc_rx_cfg1_t		*cfig1_p;
2456 	rdc_rx_cfg2_t		*cfig2_p;
2457 	rdc_rbr_kick_t		*kick_p;
2458 	uint32_t		dmaaddrp;
2459 	uint32_t		*rbr_vaddrp;
2460 	uint32_t		bkaddr;
2461 	hxge_status_t		status = HXGE_OK;
2462 	int			i;
2463 	uint32_t 		hxge_port_rcr_size;
2464 
2465 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2466 	    "==> hxge_map_rxdma_channel_cfg_ring"));
2467 
2468 	cntl_dmap = *dma_rbr_cntl_p;
2469 
2470 	/*
2471 	 * Map in the receive block ring
2472 	 */
2473 	rbrp = *rbr_p;
2474 	dmap = (p_hxge_dma_common_t)&rbrp->rbr_desc;
2475 	hxge_setup_dma_common(dmap, cntl_dmap, rbrp->rbb_max, 4);
2476 
2477 	/*
2478 	 * Zero out buffer block ring descriptors.
2479 	 */
2480 	bzero((caddr_t)dmap->kaddrp, dmap->alength);
2481 
2482 	rcfga_p = &(rbrp->rbr_cfga);
2483 	rcfgb_p = &(rbrp->rbr_cfgb);
2484 	kick_p = &(rbrp->rbr_kick);
2485 	rcfga_p->value = 0;
2486 	rcfgb_p->value = 0;
2487 	kick_p->value = 0;
2488 	rbrp->rbr_addr = dmap->dma_cookie.dmac_laddress;
2489 	rcfga_p->value = (rbrp->rbr_addr &
2490 	    (RBR_CFIG_A_STDADDR_MASK | RBR_CFIG_A_STDADDR_BASE_MASK));
2491 	rcfga_p->value |= ((uint64_t)rbrp->rbb_max << RBR_CFIG_A_LEN_SHIFT);
2492 
2493 	/* XXXX: how to choose packet buffer sizes */
2494 	rcfgb_p->bits.bufsz0 = rbrp->pkt_buf_size0;
2495 	rcfgb_p->bits.vld0 = 1;
2496 	rcfgb_p->bits.bufsz1 = rbrp->pkt_buf_size1;
2497 	rcfgb_p->bits.vld1 = 1;
2498 	rcfgb_p->bits.bufsz2 = rbrp->pkt_buf_size2;
2499 	rcfgb_p->bits.vld2 = 1;
2500 	rcfgb_p->bits.bksize = hxgep->rx_bksize_code;
2501 
2502 	/*
2503 	 * For each buffer block, enter receive block address to the ring.
2504 	 */
2505 	rbr_vaddrp = (uint32_t *)dmap->kaddrp;
2506 	rbrp->rbr_desc_vp = (uint32_t *)dmap->kaddrp;
2507 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2508 	    "==> hxge_map_rxdma_channel_cfg_ring: channel %d "
2509 	    "rbr_vaddrp $%p", dma_channel, rbr_vaddrp));
2510 
2511 	rx_msg_ring = rbrp->rx_msg_ring;
2512 	for (i = 0; i < rbrp->tnblocks; i++) {
2513 		rx_msg_p = rx_msg_ring[i];
2514 		rx_msg_p->hxgep = hxgep;
2515 		rx_msg_p->rx_rbr_p = rbrp;
2516 		bkaddr = (uint32_t)
2517 		    ((rx_msg_p->buf_dma.dma_cookie.dmac_laddress >>
2518 		    RBR_BKADDR_SHIFT));
2519 		rx_msg_p->free = B_FALSE;
2520 		rx_msg_p->max_usage_cnt = 0xbaddcafe;
2521 
2522 		*rbr_vaddrp++ = bkaddr;
2523 	}
2524 
2525 	kick_p->bits.bkadd = rbrp->rbb_max;
2526 	rbrp->rbr_wr_index = (rbrp->rbb_max - 1);
2527 
2528 	rbrp->rbr_rd_index = 0;
2529 
2530 	rbrp->rbr_consumed = 0;
2531 	rbrp->rbr_used = 0;
2532 	rbrp->rbr_use_bcopy = B_TRUE;
2533 	rbrp->rbr_bufsize_type = RCR_PKTBUFSZ_0;
2534 
2535 	/*
2536 	 * Do bcopy on packets greater than bcopy size once the lo threshold is
2537 	 * reached. This lo threshold should be less than the hi threshold.
2538 	 *
2539 	 * Do bcopy on every packet once the hi threshold is reached.
2540 	 */
2541 	if (hxge_rx_threshold_lo >= hxge_rx_threshold_hi) {
2542 		/* default it to use hi */
2543 		hxge_rx_threshold_lo = hxge_rx_threshold_hi;
2544 	}
2545 	if (hxge_rx_buf_size_type > HXGE_RBR_TYPE2) {
2546 		hxge_rx_buf_size_type = HXGE_RBR_TYPE2;
2547 	}
2548 	rbrp->rbr_bufsize_type = hxge_rx_buf_size_type;
2549 
2550 	switch (hxge_rx_threshold_hi) {
2551 	default:
2552 	case HXGE_RX_COPY_NONE:
2553 		/* Do not do bcopy at all */
2554 		rbrp->rbr_use_bcopy = B_FALSE;
2555 		rbrp->rbr_threshold_hi = rbrp->rbb_max;
2556 		break;
2557 
2558 	case HXGE_RX_COPY_1:
2559 	case HXGE_RX_COPY_2:
2560 	case HXGE_RX_COPY_3:
2561 	case HXGE_RX_COPY_4:
2562 	case HXGE_RX_COPY_5:
2563 	case HXGE_RX_COPY_6:
2564 	case HXGE_RX_COPY_7:
2565 		rbrp->rbr_threshold_hi =
2566 		    rbrp->rbb_max * (hxge_rx_threshold_hi) /
2567 		    HXGE_RX_BCOPY_SCALE;
2568 		break;
2569 
2570 	case HXGE_RX_COPY_ALL:
2571 		rbrp->rbr_threshold_hi = 0;
2572 		break;
2573 	}
2574 
2575 	switch (hxge_rx_threshold_lo) {
2576 	default:
2577 	case HXGE_RX_COPY_NONE:
2578 		/* Do not do bcopy at all */
2579 		if (rbrp->rbr_use_bcopy) {
2580 			rbrp->rbr_use_bcopy = B_FALSE;
2581 		}
2582 		rbrp->rbr_threshold_lo = rbrp->rbb_max;
2583 		break;
2584 
2585 	case HXGE_RX_COPY_1:
2586 	case HXGE_RX_COPY_2:
2587 	case HXGE_RX_COPY_3:
2588 	case HXGE_RX_COPY_4:
2589 	case HXGE_RX_COPY_5:
2590 	case HXGE_RX_COPY_6:
2591 	case HXGE_RX_COPY_7:
2592 		rbrp->rbr_threshold_lo =
2593 		    rbrp->rbb_max * (hxge_rx_threshold_lo) /
2594 		    HXGE_RX_BCOPY_SCALE;
2595 		break;
2596 
2597 	case HXGE_RX_COPY_ALL:
2598 		rbrp->rbr_threshold_lo = 0;
2599 		break;
2600 	}
2601 
2602 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
2603 	    "hxge_map_rxdma_channel_cfg_ring: channel %d rbb_max %d "
2604 	    "rbrp->rbr_bufsize_type %d rbb_threshold_hi %d "
2605 	    "rbb_threshold_lo %d",
2606 	    dma_channel, rbrp->rbb_max, rbrp->rbr_bufsize_type,
2607 	    rbrp->rbr_threshold_hi, rbrp->rbr_threshold_lo));
2608 
2609 	/* Map in the receive completion ring */
2610 	rcrp = (p_rx_rcr_ring_t)KMEM_ZALLOC(sizeof (rx_rcr_ring_t), KM_SLEEP);
2611 	rcrp->rdc = dma_channel;
2612 	rcrp->hxgep = hxgep;
2613 
2614 	hxge_port_rcr_size = hxgep->hxge_port_rcr_size;
2615 	rcrp->comp_size = hxge_port_rcr_size;
2616 	rcrp->comp_wrap_mask = hxge_port_rcr_size - 1;
2617 
2618 	rcrp->max_receive_pkts = hxge_max_rx_pkts;
2619 
2620 	cntl_dmap = *dma_rcr_cntl_p;
2621 
2622 	dmap = (p_hxge_dma_common_t)&rcrp->rcr_desc;
2623 	hxge_setup_dma_common(dmap, cntl_dmap, rcrp->comp_size,
2624 	    sizeof (rcr_entry_t));
2625 	rcrp->comp_rd_index = 0;
2626 	rcrp->comp_wt_index = 0;
2627 	rcrp->rcr_desc_rd_head_p = rcrp->rcr_desc_first_p =
2628 	    (p_rcr_entry_t)DMA_COMMON_VPTR(rcrp->rcr_desc);
2629 #if defined(__i386)
2630 	rcrp->rcr_desc_rd_head_pp = rcrp->rcr_desc_first_pp =
2631 	    (p_rcr_entry_t)(uint32_t)DMA_COMMON_IOADDR(rcrp->rcr_desc);
2632 #else
2633 	rcrp->rcr_desc_rd_head_pp = rcrp->rcr_desc_first_pp =
2634 	    (p_rcr_entry_t)DMA_COMMON_IOADDR(rcrp->rcr_desc);
2635 #endif
2636 	rcrp->rcr_desc_last_p = rcrp->rcr_desc_rd_head_p +
2637 	    (hxge_port_rcr_size - 1);
2638 	rcrp->rcr_desc_last_pp = rcrp->rcr_desc_rd_head_pp +
2639 	    (hxge_port_rcr_size - 1);
2640 
2641 	rcrp->rcr_tail_begin = DMA_COMMON_IOADDR(rcrp->rcr_desc);
2642 	rcrp->rcr_tail_begin = (rcrp->rcr_tail_begin & 0x7ffffULL) >> 3;
2643 
2644 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2645 	    "==> hxge_map_rxdma_channel_cfg_ring: channel %d "
2646 	    "rbr_vaddrp $%p rcr_desc_rd_head_p $%p "
2647 	    "rcr_desc_rd_head_pp $%p rcr_desc_rd_last_p $%p "
2648 	    "rcr_desc_rd_last_pp $%p ",
2649 	    dma_channel, rbr_vaddrp, rcrp->rcr_desc_rd_head_p,
2650 	    rcrp->rcr_desc_rd_head_pp, rcrp->rcr_desc_last_p,
2651 	    rcrp->rcr_desc_last_pp));
2652 
2653 	/*
2654 	 * Zero out buffer block ring descriptors.
2655 	 */
2656 	bzero((caddr_t)dmap->kaddrp, dmap->alength);
2657 	rcrp->intr_timeout = hxgep->intr_timeout;
2658 	rcrp->intr_threshold = hxgep->intr_threshold;
2659 	rcrp->full_hdr_flag = B_FALSE;
2660 	rcrp->sw_priv_hdr_len = 0;
2661 
2662 	cfga_p = &(rcrp->rcr_cfga);
2663 	cfgb_p = &(rcrp->rcr_cfgb);
2664 	cfga_p->value = 0;
2665 	cfgb_p->value = 0;
2666 	rcrp->rcr_addr = dmap->dma_cookie.dmac_laddress;
2667 
2668 	cfga_p->value = (rcrp->rcr_addr &
2669 	    (RCRCFIG_A_STADDR_MASK | RCRCFIG_A_STADDR_BASE_MASK));
2670 
2671 	cfga_p->value |= ((uint64_t)rcrp->comp_size << RCRCFIG_A_LEN_SHIF);
2672 
2673 	/*
2674 	 * Timeout should be set based on the system clock divider. The
2675 	 * following timeout value of 1 assumes that the granularity (1000) is
2676 	 * 3 microseconds running at 300MHz.
2677 	 */
2678 	cfgb_p->bits.pthres = rcrp->intr_threshold;
2679 	cfgb_p->bits.timeout = rcrp->intr_timeout;
2680 	cfgb_p->bits.entout = 1;
2681 
2682 	/* Map in the mailbox */
2683 	cntl_dmap = *dma_mbox_cntl_p;
2684 	mboxp = (p_rx_mbox_t)KMEM_ZALLOC(sizeof (rx_mbox_t), KM_SLEEP);
2685 	dmap = (p_hxge_dma_common_t)&mboxp->rx_mbox;
2686 	hxge_setup_dma_common(dmap, cntl_dmap, 1, sizeof (rxdma_mailbox_t));
2687 	cfig1_p = (rdc_rx_cfg1_t *)&mboxp->rx_cfg1;
2688 	cfig2_p = (rdc_rx_cfg2_t *)&mboxp->rx_cfg2;
2689 	cfig1_p->value = cfig2_p->value = 0;
2690 
2691 	mboxp->mbox_addr = dmap->dma_cookie.dmac_laddress;
2692 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2693 	    "==> hxge_map_rxdma_channel_cfg_ring: "
2694 	    "channel %d cfg1 0x%016llx cfig2 0x%016llx cookie 0x%016llx",
2695 	    dma_channel, cfig1_p->value, cfig2_p->value,
2696 	    mboxp->mbox_addr));
2697 
2698 	dmaaddrp = (uint32_t)((dmap->dma_cookie.dmac_laddress >> 32) & 0xfff);
2699 	cfig1_p->bits.mbaddr_h = dmaaddrp;
2700 
2701 	dmaaddrp = (uint32_t)(dmap->dma_cookie.dmac_laddress & 0xffffffff);
2702 	dmaaddrp = (uint32_t)(dmap->dma_cookie.dmac_laddress &
2703 	    RXDMA_CFIG2_MBADDR_L_MASK);
2704 
2705 	cfig2_p->bits.mbaddr_l = (dmaaddrp >> RXDMA_CFIG2_MBADDR_L_SHIFT);
2706 
2707 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2708 	    "==> hxge_map_rxdma_channel_cfg_ring: channel %d damaddrp $%p "
2709 	    "cfg1 0x%016llx cfig2 0x%016llx",
2710 	    dma_channel, dmaaddrp, cfig1_p->value, cfig2_p->value));
2711 
2712 	cfig2_p->bits.full_hdr = rcrp->full_hdr_flag;
2713 	cfig2_p->bits.offset = rcrp->sw_priv_hdr_len;
2714 
2715 	rbrp->rx_rcr_p = rcrp;
2716 	rcrp->rx_rbr_p = rbrp;
2717 	*rcr_p = rcrp;
2718 	*rx_mbox_p = mboxp;
2719 
2720 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2721 	    "<== hxge_map_rxdma_channel_cfg_ring status 0x%08x", status));
2722 	return (status);
2723 }
2724 
2725 /*ARGSUSED*/
2726 static void
2727 hxge_unmap_rxdma_channel_cfg_ring(p_hxge_t hxgep,
2728     p_rx_rcr_ring_t rcr_p, p_rx_mbox_t rx_mbox_p)
2729 {
2730 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2731 	    "==> hxge_unmap_rxdma_channel_cfg_ring: channel %d", rcr_p->rdc));
2732 
2733 	KMEM_FREE(rcr_p, sizeof (rx_rcr_ring_t));
2734 	KMEM_FREE(rx_mbox_p, sizeof (rx_mbox_t));
2735 
2736 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2737 	    "<== hxge_unmap_rxdma_channel_cfg_ring"));
2738 }
2739 
2740 static hxge_status_t
2741 hxge_map_rxdma_channel_buf_ring(p_hxge_t hxgep, uint16_t channel,
2742     p_hxge_dma_common_t *dma_buf_p,
2743     p_rx_rbr_ring_t *rbr_p, uint32_t num_chunks)
2744 {
2745 	p_rx_rbr_ring_t		rbrp;
2746 	p_hxge_dma_common_t	dma_bufp, tmp_bufp;
2747 	p_rx_msg_t		*rx_msg_ring;
2748 	p_rx_msg_t		rx_msg_p;
2749 	p_mblk_t		mblk_p;
2750 
2751 	rxring_info_t *ring_info;
2752 	hxge_status_t status = HXGE_OK;
2753 	int i, j, index;
2754 	uint32_t size, bsize, nblocks, nmsgs;
2755 
2756 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2757 	    "==> hxge_map_rxdma_channel_buf_ring: channel %d", channel));
2758 
2759 	dma_bufp = tmp_bufp = *dma_buf_p;
2760 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2761 	    " hxge_map_rxdma_channel_buf_ring: channel %d to map %d "
2762 	    "chunks bufp 0x%016llx", channel, num_chunks, dma_bufp));
2763 
2764 	nmsgs = 0;
2765 	for (i = 0; i < num_chunks; i++, tmp_bufp++) {
2766 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2767 		    "==> hxge_map_rxdma_channel_buf_ring: channel %d "
2768 		    "bufp 0x%016llx nblocks %d nmsgs %d",
2769 		    channel, tmp_bufp, tmp_bufp->nblocks, nmsgs));
2770 		nmsgs += tmp_bufp->nblocks;
2771 	}
2772 	if (!nmsgs) {
2773 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2774 		    "<== hxge_map_rxdma_channel_buf_ring: channel %d "
2775 		    "no msg blocks", channel));
2776 		status = HXGE_ERROR;
2777 		goto hxge_map_rxdma_channel_buf_ring_exit;
2778 	}
2779 	rbrp = (p_rx_rbr_ring_t)KMEM_ZALLOC(sizeof (rx_rbr_ring_t), KM_SLEEP);
2780 
2781 	size = nmsgs * sizeof (p_rx_msg_t);
2782 	rx_msg_ring = KMEM_ZALLOC(size, KM_SLEEP);
2783 	ring_info = (rxring_info_t *)KMEM_ZALLOC(sizeof (rxring_info_t),
2784 	    KM_SLEEP);
2785 
2786 	MUTEX_INIT(&rbrp->lock, NULL, MUTEX_DRIVER,
2787 	    (void *) hxgep->interrupt_cookie);
2788 	MUTEX_INIT(&rbrp->post_lock, NULL, MUTEX_DRIVER,
2789 	    (void *) hxgep->interrupt_cookie);
2790 
2791 	rbrp->rdc = channel;
2792 	rbrp->num_blocks = num_chunks;
2793 	rbrp->tnblocks = nmsgs;
2794 	rbrp->rbb_max = nmsgs;
2795 	rbrp->rbr_max_size = nmsgs;
2796 	rbrp->rbr_wrap_mask = (rbrp->rbb_max - 1);
2797 
2798 	/*
2799 	 * Buffer sizes suggested by NIU architect. 256, 512 and 2K.
2800 	 */
2801 
2802 	switch (hxgep->rx_bksize_code) {
2803 	case RBR_BKSIZE_4K:
2804 		rbrp->pkt_buf_size0 = RBR_BUFSZ0_256B;
2805 		rbrp->pkt_buf_size0_bytes = RBR_BUFSZ0_256_BYTES;
2806 		rbrp->hpi_pkt_buf_size0 = SIZE_256B;
2807 		break;
2808 	case RBR_BKSIZE_8K:
2809 		/* Use 512 to avoid possible rcr_full condition */
2810 		rbrp->pkt_buf_size0 = RBR_BUFSZ0_512B;
2811 		rbrp->pkt_buf_size0_bytes = RBR_BUFSZ0_512_BYTES;
2812 		rbrp->hpi_pkt_buf_size0 = SIZE_512B;
2813 		break;
2814 	}
2815 
2816 	rbrp->pkt_buf_size1 = RBR_BUFSZ1_1K;
2817 	rbrp->pkt_buf_size1_bytes = RBR_BUFSZ1_1K_BYTES;
2818 	rbrp->hpi_pkt_buf_size1 = SIZE_1KB;
2819 
2820 	rbrp->block_size = hxgep->rx_default_block_size;
2821 
2822 	if (!hxgep->param_arr[param_accept_jumbo].value) {
2823 		rbrp->pkt_buf_size2 = RBR_BUFSZ2_2K;
2824 		rbrp->pkt_buf_size2_bytes = RBR_BUFSZ2_2K_BYTES;
2825 		rbrp->hpi_pkt_buf_size2 = SIZE_2KB;
2826 	} else {
2827 		rbrp->hpi_pkt_buf_size2 = SIZE_4KB;
2828 		rbrp->pkt_buf_size2 = RBR_BUFSZ2_4K;
2829 		rbrp->pkt_buf_size2_bytes = RBR_BUFSZ2_4K_BYTES;
2830 	}
2831 
2832 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2833 	    "==> hxge_map_rxdma_channel_buf_ring: channel %d "
2834 	    "actual rbr max %d rbb_max %d nmsgs %d "
2835 	    "rbrp->block_size %d default_block_size %d "
2836 	    "(config hxge_rbr_size %d hxge_rbr_spare_size %d)",
2837 	    channel, rbrp->rbr_max_size, rbrp->rbb_max, nmsgs,
2838 	    rbrp->block_size, hxgep->rx_default_block_size,
2839 	    hxge_rbr_size, hxge_rbr_spare_size));
2840 
2841 	/*
2842 	 * Map in buffers from the buffer pool.
2843 	 * Note that num_blocks is the num_chunks. For Sparc, there is likely
2844 	 * only one chunk. For x86, there will be many chunks.
2845 	 * Loop over chunks.
2846 	 */
2847 	index = 0;
2848 	for (i = 0; i < rbrp->num_blocks; i++, dma_bufp++) {
2849 		bsize = dma_bufp->block_size;
2850 		nblocks = dma_bufp->nblocks;
2851 #if defined(__i386)
2852 		ring_info->buffer[i].dvma_addr = (uint32_t)dma_bufp->ioaddr_pp;
2853 #else
2854 		ring_info->buffer[i].dvma_addr = (uint64_t)dma_bufp->ioaddr_pp;
2855 #endif
2856 		ring_info->buffer[i].buf_index = i;
2857 		ring_info->buffer[i].buf_size = dma_bufp->alength;
2858 		ring_info->buffer[i].start_index = index;
2859 #if defined(__i386)
2860 		ring_info->buffer[i].kaddr = (uint32_t)dma_bufp->kaddrp;
2861 #else
2862 		ring_info->buffer[i].kaddr = (uint64_t)dma_bufp->kaddrp;
2863 #endif
2864 
2865 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2866 		    " hxge_map_rxdma_channel_buf_ring: map channel %d "
2867 		    "chunk %d nblocks %d chunk_size %x block_size 0x%x "
2868 		    "dma_bufp $%p dvma_addr $%p", channel, i,
2869 		    dma_bufp->nblocks,
2870 		    ring_info->buffer[i].buf_size, bsize, dma_bufp,
2871 		    ring_info->buffer[i].dvma_addr));
2872 
2873 		/* loop over blocks within a chunk */
2874 		for (j = 0; j < nblocks; j++) {
2875 			if ((rx_msg_p = hxge_allocb(bsize, BPRI_LO,
2876 			    dma_bufp)) == NULL) {
2877 				HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2878 				    "allocb failed (index %d i %d j %d)",
2879 				    index, i, j));
2880 				goto hxge_map_rxdma_channel_buf_ring_fail1;
2881 			}
2882 			rx_msg_ring[index] = rx_msg_p;
2883 			rx_msg_p->block_index = index;
2884 			rx_msg_p->shifted_addr = (uint32_t)
2885 			    ((rx_msg_p->buf_dma.dma_cookie.dmac_laddress >>
2886 			    RBR_BKADDR_SHIFT));
2887 			/*
2888 			 * Too much output
2889 			 * HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2890 			 *	"index %d j %d rx_msg_p $%p mblk %p",
2891 			 *	index, j, rx_msg_p, rx_msg_p->rx_mblk_p));
2892 			 */
2893 			mblk_p = rx_msg_p->rx_mblk_p;
2894 			mblk_p->b_wptr = mblk_p->b_rptr + bsize;
2895 
2896 			rbrp->rbr_ref_cnt++;
2897 			index++;
2898 			rx_msg_p->buf_dma.dma_channel = channel;
2899 		}
2900 	}
2901 	if (i < rbrp->num_blocks) {
2902 		goto hxge_map_rxdma_channel_buf_ring_fail1;
2903 	}
2904 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2905 	    "hxge_map_rxdma_channel_buf_ring: done buf init "
2906 	    "channel %d msg block entries %d", channel, index));
2907 	ring_info->block_size_mask = bsize - 1;
2908 	rbrp->rx_msg_ring = rx_msg_ring;
2909 	rbrp->dma_bufp = dma_buf_p;
2910 	rbrp->ring_info = ring_info;
2911 
2912 	status = hxge_rxbuf_index_info_init(hxgep, rbrp);
2913 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, " hxge_map_rxdma_channel_buf_ring: "
2914 	    "channel %d done buf info init", channel));
2915 
2916 	/*
2917 	 * Finally, permit hxge_freeb() to call hxge_post_page().
2918 	 */
2919 	rbrp->rbr_state = RBR_POSTING;
2920 
2921 	*rbr_p = rbrp;
2922 
2923 	goto hxge_map_rxdma_channel_buf_ring_exit;
2924 
2925 hxge_map_rxdma_channel_buf_ring_fail1:
2926 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2927 	    " hxge_map_rxdma_channel_buf_ring: failed channel (0x%x)",
2928 	    channel, status));
2929 
2930 	index--;
2931 	for (; index >= 0; index--) {
2932 		rx_msg_p = rx_msg_ring[index];
2933 		if (rx_msg_p != NULL) {
2934 			freeb(rx_msg_p->rx_mblk_p);
2935 			rx_msg_ring[index] = NULL;
2936 		}
2937 	}
2938 
2939 hxge_map_rxdma_channel_buf_ring_fail:
2940 	MUTEX_DESTROY(&rbrp->post_lock);
2941 	MUTEX_DESTROY(&rbrp->lock);
2942 	KMEM_FREE(ring_info, sizeof (rxring_info_t));
2943 	KMEM_FREE(rx_msg_ring, size);
2944 	KMEM_FREE(rbrp, sizeof (rx_rbr_ring_t));
2945 
2946 	status = HXGE_ERROR;
2947 
2948 hxge_map_rxdma_channel_buf_ring_exit:
2949 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2950 	    "<== hxge_map_rxdma_channel_buf_ring status 0x%08x", status));
2951 
2952 	return (status);
2953 }
2954 
2955 /*ARGSUSED*/
2956 static void
2957 hxge_unmap_rxdma_channel_buf_ring(p_hxge_t hxgep,
2958     p_rx_rbr_ring_t rbr_p)
2959 {
2960 	p_rx_msg_t	*rx_msg_ring;
2961 	p_rx_msg_t	rx_msg_p;
2962 	rxring_info_t	*ring_info;
2963 	int		i;
2964 	uint32_t	size;
2965 
2966 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2967 	    "==> hxge_unmap_rxdma_channel_buf_ring"));
2968 	if (rbr_p == NULL) {
2969 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
2970 		    "<== hxge_unmap_rxdma_channel_buf_ring: NULL rbrp"));
2971 		return;
2972 	}
2973 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2974 	    "==> hxge_unmap_rxdma_channel_buf_ring: channel %d", rbr_p->rdc));
2975 
2976 	rx_msg_ring = rbr_p->rx_msg_ring;
2977 	ring_info = rbr_p->ring_info;
2978 
2979 	if (rx_msg_ring == NULL || ring_info == NULL) {
2980 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2981 		    "<== hxge_unmap_rxdma_channel_buf_ring: "
2982 		    "rx_msg_ring $%p ring_info $%p", rx_msg_p, ring_info));
2983 		return;
2984 	}
2985 
2986 	size = rbr_p->tnblocks * sizeof (p_rx_msg_t);
2987 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2988 	    " hxge_unmap_rxdma_channel_buf_ring: channel %d chunks %d "
2989 	    "tnblocks %d (max %d) size ptrs %d ", rbr_p->rdc, rbr_p->num_blocks,
2990 	    rbr_p->tnblocks, rbr_p->rbr_max_size, size));
2991 
2992 	for (i = 0; i < rbr_p->tnblocks; i++) {
2993 		rx_msg_p = rx_msg_ring[i];
2994 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
2995 		    " hxge_unmap_rxdma_channel_buf_ring: "
2996 		    "rx_msg_p $%p", rx_msg_p));
2997 		if (rx_msg_p != NULL) {
2998 			freeb(rx_msg_p->rx_mblk_p);
2999 			rx_msg_ring[i] = NULL;
3000 		}
3001 	}
3002 
3003 	/*
3004 	 * We no longer may use the mutex <post_lock>. By setting
3005 	 * <rbr_state> to anything but POSTING, we prevent
3006 	 * hxge_post_page() from accessing a dead mutex.
3007 	 */
3008 	rbr_p->rbr_state = RBR_UNMAPPING;
3009 	MUTEX_DESTROY(&rbr_p->post_lock);
3010 
3011 	MUTEX_DESTROY(&rbr_p->lock);
3012 	KMEM_FREE(ring_info, sizeof (rxring_info_t));
3013 	KMEM_FREE(rx_msg_ring, size);
3014 
3015 	if (rbr_p->rbr_ref_cnt == 0) {
3016 		/* This is the normal state of affairs. */
3017 		KMEM_FREE(rbr_p, sizeof (*rbr_p));
3018 	} else {
3019 		/*
3020 		 * Some of our buffers are still being used.
3021 		 * Therefore, tell hxge_freeb() this ring is
3022 		 * unmapped, so it may free <rbr_p> for us.
3023 		 */
3024 		rbr_p->rbr_state = RBR_UNMAPPED;
3025 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3026 		    "unmap_rxdma_buf_ring: %d %s outstanding.",
3027 		    rbr_p->rbr_ref_cnt,
3028 		    rbr_p->rbr_ref_cnt == 1 ? "msg" : "msgs"));
3029 	}
3030 
3031 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3032 	    "<== hxge_unmap_rxdma_channel_buf_ring"));
3033 }
3034 
3035 static hxge_status_t
3036 hxge_rxdma_hw_start_common(p_hxge_t hxgep)
3037 {
3038 	hxge_status_t status = HXGE_OK;
3039 
3040 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_hw_start_common"));
3041 
3042 	/*
3043 	 * Load the sharable parameters by writing to the function zero control
3044 	 * registers. These FZC registers should be initialized only once for
3045 	 * the entire chip.
3046 	 */
3047 	(void) hxge_init_fzc_rx_common(hxgep);
3048 
3049 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_hw_start_common"));
3050 
3051 	return (status);
3052 }
3053 
3054 static hxge_status_t
3055 hxge_rxdma_hw_start(p_hxge_t hxgep)
3056 {
3057 	int			i, ndmas;
3058 	uint16_t		channel;
3059 	p_rx_rbr_rings_t	rx_rbr_rings;
3060 	p_rx_rbr_ring_t		*rbr_rings;
3061 	p_rx_rcr_rings_t	rx_rcr_rings;
3062 	p_rx_rcr_ring_t		*rcr_rings;
3063 	p_rx_mbox_areas_t	rx_mbox_areas_p;
3064 	p_rx_mbox_t		*rx_mbox_p;
3065 	hxge_status_t		status = HXGE_OK;
3066 
3067 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_hw_start"));
3068 
3069 	rx_rbr_rings = hxgep->rx_rbr_rings;
3070 	rx_rcr_rings = hxgep->rx_rcr_rings;
3071 	if (rx_rbr_rings == NULL || rx_rcr_rings == NULL) {
3072 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
3073 		    "<== hxge_rxdma_hw_start: NULL ring pointers"));
3074 		return (HXGE_ERROR);
3075 	}
3076 
3077 	ndmas = rx_rbr_rings->ndmas;
3078 	if (ndmas == 0) {
3079 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
3080 		    "<== hxge_rxdma_hw_start: no dma channel allocated"));
3081 		return (HXGE_ERROR);
3082 	}
3083 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3084 	    "==> hxge_rxdma_hw_start (ndmas %d)", ndmas));
3085 
3086 	/*
3087 	 * Scrub the RDC Rx DMA Prefetch Buffer Command.
3088 	 */
3089 	for (i = 0; i < 128; i++) {
3090 		HXGE_REG_WR64(hxgep->hpi_handle, RDC_PREF_CMD, i);
3091 	}
3092 
3093 	/*
3094 	 * Scrub Rx DMA Shadow Tail Command.
3095 	 */
3096 	for (i = 0; i < 64; i++) {
3097 		HXGE_REG_WR64(hxgep->hpi_handle, RDC_SHADOW_CMD, i);
3098 	}
3099 
3100 	/*
3101 	 * Scrub Rx DMA Control Fifo Command.
3102 	 */
3103 	for (i = 0; i < 512; i++) {
3104 		HXGE_REG_WR64(hxgep->hpi_handle, RDC_CTRL_FIFO_CMD, i);
3105 	}
3106 
3107 	/*
3108 	 * Scrub Rx DMA Data Fifo Command.
3109 	 */
3110 	for (i = 0; i < 1536; i++) {
3111 		HXGE_REG_WR64(hxgep->hpi_handle, RDC_DATA_FIFO_CMD, i);
3112 	}
3113 
3114 	/*
3115 	 * Reset the FIFO Error Stat.
3116 	 */
3117 	HXGE_REG_WR64(hxgep->hpi_handle, RDC_FIFO_ERR_STAT, 0xFF);
3118 
3119 	/* Set the error mask to receive interrupts */
3120 	HXGE_REG_WR64(hxgep->hpi_handle, RDC_FIFO_ERR_INT_MASK, 0x0);
3121 
3122 	rbr_rings = rx_rbr_rings->rbr_rings;
3123 	rcr_rings = rx_rcr_rings->rcr_rings;
3124 	rx_mbox_areas_p = hxgep->rx_mbox_areas_p;
3125 	if (rx_mbox_areas_p) {
3126 		rx_mbox_p = rx_mbox_areas_p->rxmbox_areas;
3127 	}
3128 
3129 	for (i = 0; i < ndmas; i++) {
3130 		channel = rbr_rings[i]->rdc;
3131 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3132 		    "==> hxge_rxdma_hw_start (ndmas %d) channel %d",
3133 		    ndmas, channel));
3134 		status = hxge_rxdma_start_channel(hxgep, channel,
3135 		    (p_rx_rbr_ring_t)rbr_rings[i],
3136 		    (p_rx_rcr_ring_t)rcr_rings[i],
3137 		    (p_rx_mbox_t)rx_mbox_p[i], rbr_rings[i]->rbb_max);
3138 		if (status != HXGE_OK) {
3139 			goto hxge_rxdma_hw_start_fail1;
3140 		}
3141 	}
3142 
3143 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_hw_start: "
3144 	    "rx_rbr_rings 0x%016llx rings 0x%016llx",
3145 	    rx_rbr_rings, rx_rcr_rings));
3146 	goto hxge_rxdma_hw_start_exit;
3147 
3148 hxge_rxdma_hw_start_fail1:
3149 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3150 	    "==> hxge_rxdma_hw_start: disable "
3151 	    "(status 0x%x channel %d i %d)", status, channel, i));
3152 	for (; i >= 0; i--) {
3153 		channel = rbr_rings[i]->rdc;
3154 		(void) hxge_rxdma_stop_channel(hxgep, channel);
3155 	}
3156 
3157 hxge_rxdma_hw_start_exit:
3158 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3159 	    "==> hxge_rxdma_hw_start: (status 0x%x)", status));
3160 	return (status);
3161 }
3162 
3163 static void
3164 hxge_rxdma_hw_stop(p_hxge_t hxgep)
3165 {
3166 	int			i, ndmas;
3167 	uint16_t		channel;
3168 	p_rx_rbr_rings_t	rx_rbr_rings;
3169 	p_rx_rbr_ring_t		*rbr_rings;
3170 	p_rx_rcr_rings_t	rx_rcr_rings;
3171 
3172 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_hw_stop"));
3173 
3174 	rx_rbr_rings = hxgep->rx_rbr_rings;
3175 	rx_rcr_rings = hxgep->rx_rcr_rings;
3176 
3177 	if (rx_rbr_rings == NULL || rx_rcr_rings == NULL) {
3178 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
3179 		    "<== hxge_rxdma_hw_stop: NULL ring pointers"));
3180 		return;
3181 	}
3182 
3183 	ndmas = rx_rbr_rings->ndmas;
3184 	if (!ndmas) {
3185 		HXGE_DEBUG_MSG((hxgep, RX_CTL,
3186 		    "<== hxge_rxdma_hw_stop: no dma channel allocated"));
3187 		return;
3188 	}
3189 
3190 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3191 	    "==> hxge_rxdma_hw_stop (ndmas %d)", ndmas));
3192 
3193 	rbr_rings = rx_rbr_rings->rbr_rings;
3194 	for (i = 0; i < ndmas; i++) {
3195 		channel = rbr_rings[i]->rdc;
3196 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3197 		    "==> hxge_rxdma_hw_stop (ndmas %d) channel %d",
3198 		    ndmas, channel));
3199 		(void) hxge_rxdma_stop_channel(hxgep, channel);
3200 	}
3201 
3202 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_hw_stop: "
3203 	    "rx_rbr_rings 0x%016llx rings 0x%016llx",
3204 	    rx_rbr_rings, rx_rcr_rings));
3205 
3206 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "<== hxge_rxdma_hw_stop"));
3207 }
3208 
3209 static hxge_status_t
3210 hxge_rxdma_start_channel(p_hxge_t hxgep, uint16_t channel,
3211     p_rx_rbr_ring_t rbr_p, p_rx_rcr_ring_t rcr_p, p_rx_mbox_t mbox_p,
3212     int n_init_kick)
3213 {
3214 	hpi_handle_t		handle;
3215 	hpi_status_t		rs = HPI_SUCCESS;
3216 	rdc_stat_t		cs;
3217 	rdc_int_mask_t		ent_mask;
3218 	hxge_status_t		status = HXGE_OK;
3219 
3220 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_start_channel"));
3221 
3222 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
3223 
3224 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "hxge_rxdma_start_channel: "
3225 	    "hpi handle addr $%p acc $%p",
3226 	    hxgep->hpi_handle.regp, hxgep->hpi_handle.regh));
3227 
3228 	/* Reset RXDMA channel */
3229 	rs = hpi_rxdma_cfg_rdc_reset(handle, channel);
3230 	if (rs != HPI_SUCCESS) {
3231 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3232 		    "==> hxge_rxdma_start_channel: "
3233 		    "reset rxdma failed (0x%08x channel %d)",
3234 		    status, channel));
3235 		return (HXGE_ERROR | rs);
3236 	}
3237 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3238 	    "==> hxge_rxdma_start_channel: reset done: channel %d", channel));
3239 
3240 	/*
3241 	 * Initialize the RXDMA channel specific FZC control configurations.
3242 	 * These FZC registers are pertaining to each RX channel (logical
3243 	 * pages).
3244 	 */
3245 	status = hxge_init_fzc_rxdma_channel(hxgep,
3246 	    channel, rbr_p, rcr_p, mbox_p);
3247 	if (status != HXGE_OK) {
3248 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3249 		    "==> hxge_rxdma_start_channel: "
3250 		    "init fzc rxdma failed (0x%08x channel %d)",
3251 		    status, channel));
3252 		return (status);
3253 	}
3254 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3255 	    "==> hxge_rxdma_start_channel: fzc done"));
3256 
3257 	/*
3258 	 * Zero out the shadow  and prefetch ram.
3259 	 */
3260 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3261 	    "==> hxge_rxdma_start_channel: ram done"));
3262 
3263 	/* Set up the interrupt event masks. */
3264 	ent_mask.value = 0;
3265 	rs = hpi_rxdma_event_mask(handle, OP_SET, channel, &ent_mask);
3266 	if (rs != HPI_SUCCESS) {
3267 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3268 		    "==> hxge_rxdma_start_channel: "
3269 		    "init rxdma event masks failed (0x%08x channel %d)",
3270 		    status, channel));
3271 		return (HXGE_ERROR | rs);
3272 	}
3273 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_start_channel: "
3274 	    "event done: channel %d (mask 0x%016llx)",
3275 	    channel, ent_mask.value));
3276 
3277 	/*
3278 	 * Load RXDMA descriptors, buffers, mailbox, initialise the receive DMA
3279 	 * channels and enable each DMA channel.
3280 	 */
3281 	status = hxge_enable_rxdma_channel(hxgep,
3282 	    channel, rbr_p, rcr_p, mbox_p, n_init_kick);
3283 	if (status != HXGE_OK) {
3284 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3285 		    " hxge_rxdma_start_channel: "
3286 		    " init enable rxdma failed (0x%08x channel %d)",
3287 		    status, channel));
3288 		return (status);
3289 	}
3290 
3291 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_start_channel: "
3292 	    "control done - channel %d cs 0x%016llx", channel, cs.value));
3293 
3294 	/*
3295 	 * Initialize the receive DMA control and status register
3296 	 * Note that rdc_stat HAS to be set after RBR and RCR rings are set
3297 	 */
3298 	cs.value = 0;
3299 	cs.bits.mex = 1;
3300 	cs.bits.rcr_thres = 1;
3301 	cs.bits.rcr_to = 1;
3302 	cs.bits.rbr_empty = 1;
3303 	status = hxge_init_rxdma_channel_cntl_stat(hxgep, channel, &cs);
3304 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_start_channel: "
3305 	    "channel %d rx_dma_cntl_stat 0x%0016llx", channel, cs.value));
3306 	if (status != HXGE_OK) {
3307 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3308 		    "==> hxge_rxdma_start_channel: "
3309 		    "init rxdma control register failed (0x%08x channel %d",
3310 		    status, channel));
3311 		return (status);
3312 	}
3313 
3314 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_rxdma_start_channel: "
3315 	    "control done - channel %d cs 0x%016llx", channel, cs.value));
3316 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
3317 	    "==> hxge_rxdma_start_channel: enable done"));
3318 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "<== hxge_rxdma_start_channel"));
3319 
3320 	return (HXGE_OK);
3321 }
3322 
3323 static hxge_status_t
3324 hxge_rxdma_stop_channel(p_hxge_t hxgep, uint16_t channel)
3325 {
3326 	hpi_handle_t		handle;
3327 	hpi_status_t		rs = HPI_SUCCESS;
3328 	rdc_stat_t		cs;
3329 	rdc_int_mask_t		ent_mask;
3330 	hxge_status_t		status = HXGE_OK;
3331 
3332 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rxdma_stop_channel"));
3333 
3334 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
3335 
3336 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "hxge_rxdma_stop_channel: "
3337 	    "hpi handle addr $%p acc $%p",
3338 	    hxgep->hpi_handle.regp, hxgep->hpi_handle.regh));
3339 
3340 	/* Reset RXDMA channel */
3341 	rs = hpi_rxdma_cfg_rdc_reset(handle, channel);
3342 	if (rs != HPI_SUCCESS) {
3343 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3344 		    " hxge_rxdma_stop_channel: "
3345 		    " reset rxdma failed (0x%08x channel %d)",
3346 		    rs, channel));
3347 		return (HXGE_ERROR | rs);
3348 	}
3349 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
3350 	    "==> hxge_rxdma_stop_channel: reset done"));
3351 
3352 	/* Set up the interrupt event masks. */
3353 	ent_mask.value = RDC_INT_MASK_ALL;
3354 	rs = hpi_rxdma_event_mask(handle, OP_SET, channel, &ent_mask);
3355 	if (rs != HPI_SUCCESS) {
3356 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3357 		    "==> hxge_rxdma_stop_channel: "
3358 		    "set rxdma event masks failed (0x%08x channel %d)",
3359 		    rs, channel));
3360 		return (HXGE_ERROR | rs);
3361 	}
3362 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
3363 	    "==> hxge_rxdma_stop_channel: event done"));
3364 
3365 	/* Initialize the receive DMA control and status register */
3366 	cs.value = 0;
3367 	status = hxge_init_rxdma_channel_cntl_stat(hxgep, channel, &cs);
3368 
3369 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rxdma_stop_channel: control "
3370 	    " to default (all 0s) 0x%08x", cs.value));
3371 
3372 	if (status != HXGE_OK) {
3373 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3374 		    " hxge_rxdma_stop_channel: init rxdma"
3375 		    " control register failed (0x%08x channel %d",
3376 		    status, channel));
3377 		return (status);
3378 	}
3379 
3380 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
3381 	    "==> hxge_rxdma_stop_channel: control done"));
3382 
3383 	/* disable dma channel */
3384 	status = hxge_disable_rxdma_channel(hxgep, channel);
3385 
3386 	if (status != HXGE_OK) {
3387 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3388 		    " hxge_rxdma_stop_channel: "
3389 		    " init enable rxdma failed (0x%08x channel %d)",
3390 		    status, channel));
3391 		return (status);
3392 	}
3393 
3394 	HXGE_DEBUG_MSG((hxgep, RX_CTL,
3395 	    "==> hxge_rxdma_stop_channel: disable done"));
3396 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_rxdma_stop_channel"));
3397 
3398 	return (HXGE_OK);
3399 }
3400 
3401 hxge_status_t
3402 hxge_rxdma_handle_sys_errors(p_hxge_t hxgep)
3403 {
3404 	hpi_handle_t		handle;
3405 	p_hxge_rdc_sys_stats_t	statsp;
3406 	rdc_fifo_err_stat_t	stat;
3407 	hxge_status_t		status = HXGE_OK;
3408 
3409 	handle = hxgep->hpi_handle;
3410 	statsp = (p_hxge_rdc_sys_stats_t)&hxgep->statsp->rdc_sys_stats;
3411 
3412 	/* Clear the int_dbg register in case it is an injected err */
3413 	HXGE_REG_WR64(handle, RDC_FIFO_ERR_INT_DBG, 0x0);
3414 
3415 	/* Get the error status and clear the register */
3416 	HXGE_REG_RD64(handle, RDC_FIFO_ERR_STAT, &stat.value);
3417 	HXGE_REG_WR64(handle, RDC_FIFO_ERR_STAT, stat.value);
3418 
3419 	if (stat.bits.rx_ctrl_fifo_sec) {
3420 		statsp->ctrl_fifo_sec++;
3421 		if (statsp->ctrl_fifo_sec == 1)
3422 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3423 			    "==> hxge_rxdma_handle_sys_errors: "
3424 			    "rx_ctrl_fifo_sec"));
3425 	}
3426 
3427 	if (stat.bits.rx_ctrl_fifo_ded) {
3428 		/* Global fatal error encountered */
3429 		statsp->ctrl_fifo_ded++;
3430 		HXGE_FM_REPORT_ERROR(hxgep, NULL,
3431 		    HXGE_FM_EREPORT_RDMC_CTRL_FIFO_DED);
3432 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3433 		    "==> hxge_rxdma_handle_sys_errors: "
3434 		    "fatal error: rx_ctrl_fifo_ded error"));
3435 	}
3436 
3437 	if (stat.bits.rx_data_fifo_sec) {
3438 		statsp->data_fifo_sec++;
3439 		if (statsp->data_fifo_sec == 1)
3440 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3441 			    "==> hxge_rxdma_handle_sys_errors: "
3442 			    "rx_data_fifo_sec"));
3443 	}
3444 
3445 	if (stat.bits.rx_data_fifo_ded) {
3446 		/* Global fatal error encountered */
3447 		statsp->data_fifo_ded++;
3448 		HXGE_FM_REPORT_ERROR(hxgep, NULL,
3449 		    HXGE_FM_EREPORT_RDMC_DATA_FIFO_DED);
3450 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3451 		    "==> hxge_rxdma_handle_sys_errors: "
3452 		    "fatal error: rx_data_fifo_ded error"));
3453 	}
3454 
3455 	if (stat.bits.rx_ctrl_fifo_ded || stat.bits.rx_data_fifo_ded) {
3456 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3457 		    " hxge_rxdma_handle_sys_errors: fatal error\n"));
3458 		status = hxge_rx_port_fatal_err_recover(hxgep);
3459 		if (status == HXGE_OK) {
3460 			FM_SERVICE_RESTORED(hxgep);
3461 		}
3462 	}
3463 
3464 	return (HXGE_OK);
3465 }
3466 
3467 static hxge_status_t
3468 hxge_rxdma_fatal_err_recover(p_hxge_t hxgep, uint16_t channel)
3469 {
3470 	hpi_handle_t		handle;
3471 	hpi_status_t 		rs = HPI_SUCCESS;
3472 	hxge_status_t 		status = HXGE_OK;
3473 	p_rx_rbr_ring_t		rbrp;
3474 	p_rx_rcr_ring_t		rcrp;
3475 	p_rx_mbox_t		mboxp;
3476 	rdc_int_mask_t		ent_mask;
3477 	p_hxge_dma_common_t	dmap;
3478 	int			ring_idx;
3479 	p_rx_msg_t		rx_msg_p;
3480 	int			i;
3481 	uint32_t		hxge_port_rcr_size;
3482 	uint64_t		tmp;
3483 	int			n_init_kick = 0;
3484 
3485 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rxdma_fatal_err_recover"));
3486 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3487 	    "Recovering from RxDMAChannel#%d error...", channel));
3488 
3489 	/*
3490 	 * Stop the dma channel waits for the stop done. If the stop done bit
3491 	 * is not set, then create an error.
3492 	 */
3493 
3494 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
3495 
3496 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Rx DMA stop..."));
3497 
3498 	ring_idx = hxge_rxdma_get_ring_index(hxgep, channel);
3499 	rbrp = (p_rx_rbr_ring_t)hxgep->rx_rbr_rings->rbr_rings[ring_idx];
3500 	rcrp = (p_rx_rcr_ring_t)hxgep->rx_rcr_rings->rcr_rings[ring_idx];
3501 
3502 	MUTEX_ENTER(&rcrp->lock);
3503 	MUTEX_ENTER(&rbrp->lock);
3504 
3505 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Disable RxDMA channel..."));
3506 
3507 	rs = hpi_rxdma_cfg_rdc_disable(handle, channel);
3508 	if (rs != HPI_SUCCESS) {
3509 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3510 		    "hxge_disable_rxdma_channel:failed"));
3511 		goto fail;
3512 	}
3513 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Disable RxDMA interrupt..."));
3514 
3515 	/* Disable interrupt */
3516 	ent_mask.value = RDC_INT_MASK_ALL;
3517 	rs = hpi_rxdma_event_mask(handle, OP_SET, channel, &ent_mask);
3518 	if (rs != HPI_SUCCESS) {
3519 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3520 		    "Set rxdma event masks failed (channel %d)", channel));
3521 	}
3522 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "RxDMA channel reset..."));
3523 
3524 	/* Reset RXDMA channel */
3525 	rs = hpi_rxdma_cfg_rdc_reset(handle, channel);
3526 	if (rs != HPI_SUCCESS) {
3527 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3528 		    "Reset rxdma failed (channel %d)", channel));
3529 		goto fail;
3530 	}
3531 	hxge_port_rcr_size = hxgep->hxge_port_rcr_size;
3532 	mboxp = (p_rx_mbox_t)hxgep->rx_mbox_areas_p->rxmbox_areas[ring_idx];
3533 
3534 	rbrp->rbr_wr_index = (rbrp->rbb_max - 1);
3535 	rbrp->rbr_rd_index = 0;
3536 
3537 	rcrp->comp_rd_index = 0;
3538 	rcrp->comp_wt_index = 0;
3539 	rcrp->rcr_desc_rd_head_p = rcrp->rcr_desc_first_p =
3540 	    (p_rcr_entry_t)DMA_COMMON_VPTR(rcrp->rcr_desc);
3541 #if defined(__i386)
3542 	rcrp->rcr_desc_rd_head_pp = rcrp->rcr_desc_first_pp =
3543 	    (p_rcr_entry_t)(uint32_t)DMA_COMMON_IOADDR(rcrp->rcr_desc);
3544 #else
3545 	rcrp->rcr_desc_rd_head_pp = rcrp->rcr_desc_first_pp =
3546 	    (p_rcr_entry_t)DMA_COMMON_IOADDR(rcrp->rcr_desc);
3547 #endif
3548 
3549 	rcrp->rcr_desc_last_p = rcrp->rcr_desc_rd_head_p +
3550 	    (hxge_port_rcr_size - 1);
3551 	rcrp->rcr_desc_last_pp = rcrp->rcr_desc_rd_head_pp +
3552 	    (hxge_port_rcr_size - 1);
3553 
3554 	rcrp->rcr_tail_begin = DMA_COMMON_IOADDR(rcrp->rcr_desc);
3555 	rcrp->rcr_tail_begin = (rcrp->rcr_tail_begin & 0x7ffffULL) >> 3;
3556 
3557 	dmap = (p_hxge_dma_common_t)&rcrp->rcr_desc;
3558 	bzero((caddr_t)dmap->kaddrp, dmap->alength);
3559 
3560 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "rbr entries = %d\n",
3561 	    rbrp->rbr_max_size));
3562 
3563 	/* Count the number of buffers owned by the hardware at this moment */
3564 	for (i = 0; i < rbrp->rbr_max_size; i++) {
3565 		rx_msg_p = rbrp->rx_msg_ring[i];
3566 		if (rx_msg_p->ref_cnt == 1) {
3567 			n_init_kick++;
3568 		}
3569 	}
3570 
3571 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "RxDMA channel re-start..."));
3572 
3573 	/*
3574 	 * This is error recover! Some buffers are owned by the hardware and
3575 	 * the rest are owned by the apps. We should only kick in those
3576 	 * owned by the hardware initially. The apps will post theirs
3577 	 * eventually.
3578 	 */
3579 	status = hxge_rxdma_start_channel(hxgep, channel, rbrp, rcrp, mboxp,
3580 	    n_init_kick);
3581 	if (status != HXGE_OK) {
3582 		goto fail;
3583 	}
3584 
3585 	/*
3586 	 * The DMA channel may disable itself automatically.
3587 	 * The following is a work-around.
3588 	 */
3589 	HXGE_REG_RD64(handle, RDC_RX_CFG1, &tmp);
3590 	rs = hpi_rxdma_cfg_rdc_enable(handle, channel);
3591 	if (rs != HPI_SUCCESS) {
3592 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3593 		    "hpi_rxdma_cfg_rdc_enable (channel %d)", channel));
3594 	}
3595 
3596 	MUTEX_EXIT(&rbrp->lock);
3597 	MUTEX_EXIT(&rcrp->lock);
3598 
3599 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3600 	    "Recovery Successful, RxDMAChannel#%d Restored", channel));
3601 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_rxdma_fatal_err_recover"));
3602 
3603 	return (HXGE_OK);
3604 
3605 fail:
3606 	MUTEX_EXIT(&rbrp->lock);
3607 	MUTEX_EXIT(&rcrp->lock);
3608 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL, "Recovery failed"));
3609 
3610 	return (HXGE_ERROR | rs);
3611 }
3612 
3613 static hxge_status_t
3614 hxge_rx_port_fatal_err_recover(p_hxge_t hxgep)
3615 {
3616 	hxge_status_t		status = HXGE_OK;
3617 	p_hxge_dma_common_t	*dma_buf_p;
3618 	uint16_t		channel;
3619 	int			ndmas;
3620 	int			i;
3621 	block_reset_t		reset_reg;
3622 	p_rx_rcr_ring_t	rcrp;
3623 	p_rx_rbr_ring_t rbrp;
3624 
3625 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "==> hxge_rx_port_fatal_err_recover"));
3626 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL, "Recovering from RDC error ..."));
3627 
3628 	/* Reset RDC block from PEU for this fatal error */
3629 	reset_reg.value = 0;
3630 	reset_reg.bits.rdc_rst = 1;
3631 	HXGE_REG_WR32(hxgep->hpi_handle, BLOCK_RESET, reset_reg.value);
3632 
3633 	/* Disable RxMAC */
3634 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Disable RxMAC...\n"));
3635 	if (hxge_rx_vmac_disable(hxgep) != HXGE_OK)
3636 		goto fail;
3637 
3638 	HXGE_DELAY(1000);
3639 
3640 	/* Restore any common settings after PEU reset */
3641 	if (hxge_rxdma_hw_start_common(hxgep) != HXGE_OK)
3642 		goto fail;
3643 
3644 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Stop all RxDMA channels..."));
3645 
3646 	ndmas = hxgep->rx_buf_pool_p->ndmas;
3647 	dma_buf_p = hxgep->rx_buf_pool_p->dma_buf_pool_p;
3648 
3649 	for (i = 0; i < ndmas; i++) {
3650 		channel = ((p_hxge_dma_common_t)dma_buf_p[i])->dma_channel;
3651 		rcrp = hxgep->rx_rcr_rings->rcr_rings[channel];
3652 		rbrp = rcrp->rx_rbr_p;
3653 
3654 		MUTEX_ENTER(&rbrp->post_lock);
3655 		/* This function needs to be inside the post_lock */
3656 		if (hxge_rxdma_fatal_err_recover(hxgep, channel) != HXGE_OK) {
3657 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3658 			    "Could not recover channel %d", channel));
3659 		}
3660 		MUTEX_EXIT(&rbrp->post_lock);
3661 	}
3662 
3663 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Reset RxMAC..."));
3664 
3665 	/* Reset RxMAC */
3666 	if (hxge_rx_vmac_reset(hxgep) != HXGE_OK) {
3667 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3668 		    "hxge_rx_port_fatal_err_recover: Failed to reset RxMAC"));
3669 		goto fail;
3670 	}
3671 
3672 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Re-initialize RxMAC..."));
3673 
3674 	/* Re-Initialize RxMAC */
3675 	if ((status = hxge_rx_vmac_init(hxgep)) != HXGE_OK) {
3676 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3677 		    "hxge_rx_port_fatal_err_recover: Failed to reset RxMAC"));
3678 		goto fail;
3679 	}
3680 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "Re-enable RxMAC..."));
3681 
3682 	/* Re-enable RxMAC */
3683 	if ((status = hxge_rx_vmac_enable(hxgep)) != HXGE_OK) {
3684 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3685 		    "hxge_rx_port_fatal_err_recover: Failed to enable RxMAC"));
3686 		goto fail;
3687 	}
3688 
3689 	/* Reset the error mask since PEU reset cleared it */
3690 	HXGE_REG_WR64(hxgep->hpi_handle, RDC_FIFO_ERR_INT_MASK, 0x0);
3691 
3692 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3693 	    "Recovery Successful, RxPort Restored"));
3694 	HXGE_DEBUG_MSG((hxgep, RX_CTL, "<== hxge_rx_port_fatal_err_recover"));
3695 
3696 	return (HXGE_OK);
3697 fail:
3698 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL, "Recovery failed"));
3699 	return (status);
3700 }
3701 
3702 static void
3703 hxge_rbr_empty_restore(p_hxge_t hxgep, p_rx_rbr_ring_t rx_rbr_p)
3704 {
3705 	hpi_status_t		hpi_status;
3706 	hxge_status_t		status;
3707 	int			i;
3708 	p_hxge_rx_ring_stats_t	rdc_stats;
3709 
3710 	rdc_stats = &hxgep->statsp->rdc_stats[rx_rbr_p->rdc];
3711 	rdc_stats->rbr_empty_restore++;
3712 	rx_rbr_p->rbr_is_empty = B_FALSE;
3713 
3714 	/*
3715 	 * Complete the processing for the RBR Empty by:
3716 	 *	0) kicking back HXGE_RBR_EMPTY_THRESHOLD
3717 	 *	   packets.
3718 	 *	1) Disable the RX vmac.
3719 	 *	2) Re-enable the affected DMA channel.
3720 	 *	3) Re-enable the RX vmac.
3721 	 */
3722 
3723 	/*
3724 	 * Disable the RX VMAC, but setting the framelength
3725 	 * to 0, since there is a hardware bug when disabling
3726 	 * the vmac.
3727 	 */
3728 	MUTEX_ENTER(hxgep->genlock);
3729 	(void) hpi_vmac_rx_set_framesize(
3730 	    HXGE_DEV_HPI_HANDLE(hxgep), (uint16_t)0);
3731 
3732 	hpi_status = hpi_rxdma_cfg_rdc_enable(
3733 	    HXGE_DEV_HPI_HANDLE(hxgep), rx_rbr_p->rdc);
3734 	if (hpi_status != HPI_SUCCESS) {
3735 		rdc_stats->rbr_empty_fail++;
3736 
3737 		/* Assume we are already inside the post_lock */
3738 		status = hxge_rxdma_fatal_err_recover(hxgep, rx_rbr_p->rdc);
3739 		if (status != HXGE_OK) {
3740 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3741 			    "hxge(%d): channel(%d) is empty.",
3742 			    hxgep->instance, rx_rbr_p->rdc));
3743 		}
3744 	}
3745 
3746 	for (i = 0; i < 1024; i++) {
3747 		uint64_t value;
3748 		RXDMA_REG_READ64(HXGE_DEV_HPI_HANDLE(hxgep),
3749 		    RDC_STAT, i & 3, &value);
3750 	}
3751 
3752 	/*
3753 	 * Re-enable the RX VMAC.
3754 	 */
3755 	(void) hpi_vmac_rx_set_framesize(HXGE_DEV_HPI_HANDLE(hxgep),
3756 	    (uint16_t)hxgep->vmac.maxframesize);
3757 	MUTEX_EXIT(hxgep->genlock);
3758 }
3759