xref: /titanic_41/usr/src/uts/common/io/e1000g/e1000g_rx.c (revision ba2be53024c0b999e74ba9adcd7d80fec5df8c57)
1 /*
2  * This file is provided under a CDDLv1 license.  When using or
3  * redistributing this file, you may do so under this license.
4  * In redistributing this file this license must be included
5  * and no other modification of this header file is permitted.
6  *
7  * CDDL LICENSE SUMMARY
8  *
9  * Copyright(c) 1999 - 2008 Intel Corporation. All rights reserved.
10  *
11  * The contents of this file are subject to the terms of Version
12  * 1.0 of the Common Development and Distribution License (the "License").
13  *
14  * You should have received a copy of the License with this software.
15  * You can obtain a copy of the License at
16  *	http://www.opensolaris.org/os/licensing.
17  * See the License for the specific language governing permissions
18  * and limitations under the License.
19  */
20 
21 /*
22  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms of the CDDLv1.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 /*
29  * **********************************************************************
30  *									*
31  * Module Name:								*
32  *   e1000g_rx.c							*
33  *									*
34  * Abstract:								*
35  *   This file contains some routines that take care of Receive		*
36  *   interrupt and also for the received packets it sends up to		*
37  *   upper layer.							*
38  *   It tries to do a zero copy if free buffers are available in	*
39  *   the pool.								*
40  *									*
41  * **********************************************************************
42  */
43 
44 #include "e1000g_sw.h"
45 #include "e1000g_debug.h"
46 
47 static p_rx_sw_packet_t e1000g_get_buf(e1000g_rx_ring_t *rx_ring);
48 #pragma	inline(e1000g_get_buf)
49 static void e1000g_priv_devi_list_clean();
50 
51 /*
52  * e1000g_rxfree_func - the call-back function to reclaim rx buffer
53  *
54  * This function is called when an mp is freed by the user thru
55  * freeb call (Only for mp constructed through desballoc call)
56  * It returns back the freed buffer to the freelist
57  */
58 void
59 e1000g_rxfree_func(p_rx_sw_packet_t packet)
60 {
61 	struct e1000g *Adapter;
62 	e1000g_rx_ring_t *rx_ring;
63 
64 	rx_ring = (e1000g_rx_ring_t *)packet->rx_ring;
65 	Adapter = rx_ring->adapter;
66 
67 	/*
68 	 * Here the rx recycling processes different rx packets in different
69 	 * threads, so we protect it with RW_READER to ensure it won't block
70 	 * other rx recycling threads.
71 	 */
72 	rw_enter(&e1000g_rx_detach_lock, RW_READER);
73 
74 	if (packet->flag == E1000G_RX_SW_FREE) {
75 		rw_exit(&e1000g_rx_detach_lock);
76 		return;
77 	}
78 
79 	if (packet->flag == E1000G_RX_SW_STOP) {
80 		packet->flag = E1000G_RX_SW_FREE;
81 		rw_exit(&e1000g_rx_detach_lock);
82 
83 		rw_enter(&e1000g_rx_detach_lock, RW_WRITER);
84 		rx_ring->pending_count--;
85 		e1000g_mblks_pending--;
86 
87 		if (rx_ring->pending_count == 0) {
88 			while (rx_ring->pending_list != NULL) {
89 				packet = rx_ring->pending_list;
90 				rx_ring->pending_list =
91 				    rx_ring->pending_list->next;
92 
93 				ASSERT(packet->mp == NULL);
94 				e1000g_free_rx_sw_packet(packet);
95 			}
96 		}
97 
98 		/*
99 		 * If e1000g_force_detach is enabled, we need to clean up
100 		 * the idle priv_dip entries in the private dip list while
101 		 * e1000g_mblks_pending is zero.
102 		 */
103 		if (e1000g_force_detach && (e1000g_mblks_pending == 0))
104 			e1000g_priv_devi_list_clean();
105 		rw_exit(&e1000g_rx_detach_lock);
106 		return;
107 	}
108 
109 	if (packet->flag == E1000G_RX_SW_DETACH) {
110 		packet->flag = E1000G_RX_SW_FREE;
111 		rw_exit(&e1000g_rx_detach_lock);
112 
113 		ASSERT(packet->mp == NULL);
114 		e1000g_free_rx_sw_packet(packet);
115 
116 		/*
117 		 * Here the e1000g_mblks_pending may be modified by different
118 		 * rx recycling threads simultaneously, so we need to protect
119 		 * it with RW_WRITER.
120 		 */
121 		rw_enter(&e1000g_rx_detach_lock, RW_WRITER);
122 		e1000g_mblks_pending--;
123 
124 		/*
125 		 * If e1000g_force_detach is enabled, we need to clean up
126 		 * the idle priv_dip entries in the private dip list while
127 		 * e1000g_mblks_pending is zero.
128 		 */
129 		if (e1000g_force_detach && (e1000g_mblks_pending == 0))
130 			e1000g_priv_devi_list_clean();
131 		rw_exit(&e1000g_rx_detach_lock);
132 		return;
133 	}
134 
135 	packet->flag = E1000G_RX_SW_FREE;
136 
137 	if (packet->mp == NULL) {
138 		/*
139 		 * Allocate a mblk that binds to the data buffer
140 		 */
141 		packet->mp = desballoc((unsigned char *)
142 		    packet->rx_buf->address - E1000G_IPALIGNROOM,
143 		    packet->rx_buf->size + E1000G_IPALIGNROOM,
144 		    BPRI_MED, &packet->free_rtn);
145 
146 		if (packet->mp != NULL) {
147 			packet->mp->b_rptr += E1000G_IPALIGNROOM;
148 			packet->mp->b_wptr += E1000G_IPALIGNROOM;
149 		} else {
150 			E1000G_STAT(rx_ring->stat_esballoc_fail);
151 		}
152 	}
153 
154 	mutex_enter(&rx_ring->freelist_lock);
155 	QUEUE_PUSH_TAIL(&rx_ring->free_list, &packet->Link);
156 	rx_ring->avail_freepkt++;
157 	mutex_exit(&rx_ring->freelist_lock);
158 
159 	rw_exit(&e1000g_rx_detach_lock);
160 }
161 
162 /*
163  * e1000g_priv_devi_list_clean - clean up e1000g_private_devi_list
164  *
165  * We will walk the e1000g_private_devi_list to free the entry marked
166  * with the E1000G_PRIV_DEVI_DETACH flag.
167  */
168 static void
169 e1000g_priv_devi_list_clean()
170 {
171 	private_devi_list_t *devi_node, *devi_del;
172 
173 	if (e1000g_private_devi_list == NULL)
174 		return;
175 
176 	devi_node = e1000g_private_devi_list;
177 	while ((devi_node != NULL) &&
178 	    (devi_node->flag == E1000G_PRIV_DEVI_DETACH)) {
179 		e1000g_private_devi_list = devi_node->next;
180 		kmem_free(devi_node->priv_dip,
181 		    sizeof (struct dev_info));
182 		kmem_free(devi_node,
183 		    sizeof (private_devi_list_t));
184 		devi_node = e1000g_private_devi_list;
185 	}
186 	if (e1000g_private_devi_list == NULL)
187 		return;
188 	while (devi_node->next != NULL) {
189 		if (devi_node->next->flag == E1000G_PRIV_DEVI_DETACH) {
190 			devi_del = devi_node->next;
191 			devi_node->next = devi_del->next;
192 			kmem_free(devi_del->priv_dip,
193 			    sizeof (struct dev_info));
194 			kmem_free(devi_del,
195 			    sizeof (private_devi_list_t));
196 		} else {
197 			devi_node = devi_node->next;
198 		}
199 	}
200 }
201 
202 /*
203  * e1000g_rx_setup - setup rx data structures
204  *
205  * This routine initializes all of the receive related
206  * structures. This includes the receive descriptors, the
207  * actual receive buffers, and the rx_sw_packet software
208  * structures.
209  */
210 void
211 e1000g_rx_setup(struct e1000g *Adapter)
212 {
213 	struct e1000_hw *hw;
214 	p_rx_sw_packet_t packet;
215 	struct e1000_rx_desc *descriptor;
216 	uint32_t buf_low;
217 	uint32_t buf_high;
218 	uint32_t reg_val;
219 	int i;
220 	int size;
221 	e1000g_rx_ring_t *rx_ring;
222 
223 	hw = &Adapter->shared;
224 	rx_ring = Adapter->rx_ring;
225 
226 	/*
227 	 * zero out all of the receive buffer descriptor memory
228 	 * assures any previous data or status is erased
229 	 */
230 	bzero(rx_ring->rbd_area,
231 	    sizeof (struct e1000_rx_desc) * Adapter->rx_desc_num);
232 
233 	if (!Adapter->rx_buffer_setup) {
234 		/* Init the list of "Receive Buffer" */
235 		QUEUE_INIT_LIST(&rx_ring->recv_list);
236 
237 		/* Init the list of "Free Receive Buffer" */
238 		QUEUE_INIT_LIST(&rx_ring->free_list);
239 
240 		/*
241 		 * Setup Receive list and the Free list. Note that
242 		 * the both were allocated in one packet area.
243 		 */
244 		packet = rx_ring->packet_area;
245 		descriptor = rx_ring->rbd_first;
246 
247 		for (i = 0; i < Adapter->rx_desc_num;
248 		    i++, packet = packet->next, descriptor++) {
249 			ASSERT(packet != NULL);
250 			ASSERT(descriptor != NULL);
251 			descriptor->buffer_addr =
252 			    packet->rx_buf->dma_address;
253 
254 			/* Add this rx_sw_packet to the receive list */
255 			QUEUE_PUSH_TAIL(&rx_ring->recv_list,
256 			    &packet->Link);
257 		}
258 
259 		for (i = 0; i < Adapter->rx_freelist_num;
260 		    i++, packet = packet->next) {
261 			ASSERT(packet != NULL);
262 			/* Add this rx_sw_packet to the free list */
263 			QUEUE_PUSH_TAIL(&rx_ring->free_list,
264 			    &packet->Link);
265 		}
266 		rx_ring->avail_freepkt = Adapter->rx_freelist_num;
267 
268 		Adapter->rx_buffer_setup = B_TRUE;
269 	} else {
270 		/* Setup the initial pointer to the first rx descriptor */
271 		packet = (p_rx_sw_packet_t)
272 		    QUEUE_GET_HEAD(&rx_ring->recv_list);
273 		descriptor = rx_ring->rbd_first;
274 
275 		for (i = 0; i < Adapter->rx_desc_num; i++) {
276 			ASSERT(packet != NULL);
277 			ASSERT(descriptor != NULL);
278 			descriptor->buffer_addr =
279 			    packet->rx_buf->dma_address;
280 
281 			/* Get next rx_sw_packet */
282 			packet = (p_rx_sw_packet_t)
283 			    QUEUE_GET_NEXT(&rx_ring->recv_list, &packet->Link);
284 			descriptor++;
285 		}
286 	}
287 
288 	E1000_WRITE_REG(&Adapter->shared, E1000_RDTR, Adapter->rx_intr_delay);
289 	E1000G_DEBUGLOG_1(Adapter, E1000G_INFO_LEVEL,
290 	    "E1000_RDTR: 0x%x\n", Adapter->rx_intr_delay);
291 	if (hw->mac.type >= e1000_82540) {
292 		E1000_WRITE_REG(&Adapter->shared, E1000_RADV,
293 		    Adapter->rx_intr_abs_delay);
294 		E1000G_DEBUGLOG_1(Adapter, E1000G_INFO_LEVEL,
295 		    "E1000_RADV: 0x%x\n", Adapter->rx_intr_abs_delay);
296 	}
297 
298 	/*
299 	 * Setup our descriptor pointers
300 	 */
301 	rx_ring->rbd_next = rx_ring->rbd_first;
302 
303 	size = Adapter->rx_desc_num * sizeof (struct e1000_rx_desc);
304 	E1000_WRITE_REG(hw, E1000_RDLEN, size);
305 	size = E1000_READ_REG(hw, E1000_RDLEN);
306 
307 	/* To get lower order bits */
308 	buf_low = (uint32_t)rx_ring->rbd_dma_addr;
309 	/* To get the higher order bits */
310 	buf_high = (uint32_t)(rx_ring->rbd_dma_addr >> 32);
311 
312 	E1000_WRITE_REG(hw, E1000_RDBAH, buf_high);
313 	E1000_WRITE_REG(hw, E1000_RDBAL, buf_low);
314 
315 	/*
316 	 * Setup our HW Rx Head & Tail descriptor pointers
317 	 */
318 	E1000_WRITE_REG(hw, E1000_RDT,
319 	    (uint32_t)(rx_ring->rbd_last - rx_ring->rbd_first));
320 	E1000_WRITE_REG(hw, E1000_RDH, 0);
321 
322 	/*
323 	 * Setup the Receive Control Register (RCTL), and ENABLE the
324 	 * receiver. The initial configuration is to: Enable the receiver,
325 	 * accept broadcasts, discard bad packets (and long packets),
326 	 * disable VLAN filter checking, set the receive descriptor
327 	 * minimum threshold size to 1/2, and the receive buffer size to
328 	 * 2k.
329 	 */
330 	reg_val = E1000_RCTL_EN |	/* Enable Receive Unit */
331 	    E1000_RCTL_BAM |		/* Accept Broadcast Packets */
332 	    E1000_RCTL_LPE |		/* Large Packet Enable bit */
333 	    (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT) |
334 	    E1000_RCTL_RDMTS_HALF |
335 	    E1000_RCTL_LBM_NO;		/* Loopback Mode = none */
336 
337 	if (Adapter->strip_crc)
338 		reg_val |= E1000_RCTL_SECRC;	/* Strip Ethernet CRC */
339 
340 	switch (hw->mac.max_frame_size) {
341 	case ETHERMAX:
342 		reg_val |= E1000_RCTL_SZ_2048;
343 		break;
344 	case FRAME_SIZE_UPTO_4K:
345 		reg_val |= E1000_RCTL_SZ_4096 | E1000_RCTL_BSEX;
346 		break;
347 	case FRAME_SIZE_UPTO_8K:
348 		reg_val |= E1000_RCTL_SZ_8192 | E1000_RCTL_BSEX;
349 		break;
350 	case FRAME_SIZE_UPTO_9K:
351 	case FRAME_SIZE_UPTO_16K:
352 		reg_val |= E1000_RCTL_SZ_16384 | E1000_RCTL_BSEX;
353 		break;
354 	default:
355 		reg_val |= E1000_RCTL_SZ_2048;
356 		break;
357 	}
358 
359 	if (e1000_tbi_sbp_enabled_82543(hw))
360 		reg_val |= E1000_RCTL_SBP;
361 
362 	/*
363 	 * Enable early receives on supported devices, only takes effect when
364 	 * packet size is equal or larger than the specified value (in 8 byte
365 	 * units), e.g. using jumbo frames when setting to E1000_ERT_2048
366 	 */
367 	if ((hw->mac.type == e1000_82573) || (hw->mac.type == e1000_ich9lan))
368 		E1000_WRITE_REG(hw, E1000_ERT, E1000_ERT_2048);
369 
370 	E1000_WRITE_REG(hw, E1000_RCTL, reg_val);
371 
372 	reg_val =
373 	    E1000_RXCSUM_TUOFL |	/* TCP/UDP checksum offload Enable */
374 	    E1000_RXCSUM_IPOFL;		/* IP checksum offload Enable */
375 
376 	E1000_WRITE_REG(hw, E1000_RXCSUM, reg_val);
377 }
378 
379 /*
380  * e1000g_get_buf - get an rx sw packet from the free_list
381  */
382 static p_rx_sw_packet_t
383 e1000g_get_buf(e1000g_rx_ring_t *rx_ring)
384 {
385 	struct e1000g *Adapter;
386 	p_rx_sw_packet_t packet;
387 
388 	Adapter = rx_ring->adapter;
389 
390 	mutex_enter(&rx_ring->freelist_lock);
391 	packet = (p_rx_sw_packet_t)
392 	    QUEUE_POP_HEAD(&rx_ring->free_list);
393 	if (packet != NULL)
394 		rx_ring->avail_freepkt--;
395 	mutex_exit(&rx_ring->freelist_lock);
396 
397 	return (packet);
398 }
399 
400 /*
401  * e1000g_receive - main receive routine
402  *
403  * This routine will process packets received in an interrupt
404  */
405 mblk_t *
406 e1000g_receive(struct e1000g *Adapter)
407 {
408 	struct e1000_hw *hw;
409 	mblk_t *nmp;
410 	mblk_t *ret_mp;
411 	mblk_t *ret_nmp;
412 	struct e1000_rx_desc *current_desc;
413 	struct e1000_rx_desc *last_desc;
414 	p_rx_sw_packet_t packet;
415 	p_rx_sw_packet_t newpkt;
416 	USHORT length;
417 	uint32_t pkt_count;
418 	uint32_t desc_count;
419 	boolean_t accept_frame;
420 	boolean_t end_of_packet;
421 	boolean_t need_copy;
422 	e1000g_rx_ring_t *rx_ring;
423 	dma_buffer_t *rx_buf;
424 	uint16_t cksumflags;
425 
426 	ret_mp = NULL;
427 	ret_nmp = NULL;
428 	pkt_count = 0;
429 	desc_count = 0;
430 	cksumflags = 0;
431 
432 	hw = &Adapter->shared;
433 	rx_ring = Adapter->rx_ring;
434 
435 	/* Sync the Rx descriptor DMA buffers */
436 	(void) ddi_dma_sync(rx_ring->rbd_dma_handle,
437 	    0, 0, DDI_DMA_SYNC_FORKERNEL);
438 
439 	if (e1000g_check_dma_handle(rx_ring->rbd_dma_handle) != DDI_FM_OK) {
440 		ddi_fm_service_impact(Adapter->dip, DDI_SERVICE_DEGRADED);
441 		Adapter->chip_state = E1000G_ERROR;
442 	}
443 
444 	current_desc = rx_ring->rbd_next;
445 	if (!(current_desc->status & E1000_RXD_STAT_DD)) {
446 		/*
447 		 * don't send anything up. just clear the RFD
448 		 */
449 		E1000G_DEBUG_STAT(rx_ring->stat_none);
450 		return (ret_mp);
451 	}
452 
453 	/*
454 	 * Loop through the receive descriptors starting at the last known
455 	 * descriptor owned by the hardware that begins a packet.
456 	 */
457 	while ((current_desc->status & E1000_RXD_STAT_DD) &&
458 	    (pkt_count < Adapter->rx_limit_onintr)) {
459 
460 		desc_count++;
461 		/*
462 		 * Now this can happen in Jumbo frame situation.
463 		 */
464 		if (current_desc->status & E1000_RXD_STAT_EOP) {
465 			/* packet has EOP set */
466 			end_of_packet = B_TRUE;
467 		} else {
468 			/*
469 			 * If this received buffer does not have the
470 			 * End-Of-Packet bit set, the received packet
471 			 * will consume multiple buffers. We won't send this
472 			 * packet upstack till we get all the related buffers.
473 			 */
474 			end_of_packet = B_FALSE;
475 		}
476 
477 		/*
478 		 * Get a pointer to the actual receive buffer
479 		 * The mp->b_rptr is mapped to The CurrentDescriptor
480 		 * Buffer Address.
481 		 */
482 		packet =
483 		    (p_rx_sw_packet_t)QUEUE_GET_HEAD(&rx_ring->recv_list);
484 		ASSERT(packet != NULL);
485 
486 		rx_buf = packet->rx_buf;
487 
488 		length = current_desc->length;
489 
490 #ifdef __sparc
491 		if (packet->dma_type == USE_DVMA)
492 			dvma_sync(rx_buf->dma_handle, 0,
493 			    DDI_DMA_SYNC_FORKERNEL);
494 		else
495 			(void) ddi_dma_sync(rx_buf->dma_handle,
496 			    E1000G_IPALIGNROOM, length,
497 			    DDI_DMA_SYNC_FORKERNEL);
498 #else
499 		(void) ddi_dma_sync(rx_buf->dma_handle,
500 		    E1000G_IPALIGNROOM, length,
501 		    DDI_DMA_SYNC_FORKERNEL);
502 #endif
503 
504 		if (e1000g_check_dma_handle(
505 		    rx_buf->dma_handle) != DDI_FM_OK) {
506 			ddi_fm_service_impact(Adapter->dip,
507 			    DDI_SERVICE_DEGRADED);
508 			Adapter->chip_state = E1000G_ERROR;
509 		}
510 
511 		accept_frame = (current_desc->errors == 0) ||
512 		    ((current_desc->errors &
513 		    (E1000_RXD_ERR_TCPE | E1000_RXD_ERR_IPE)) != 0);
514 
515 		if (hw->mac.type == e1000_82543) {
516 			unsigned char last_byte;
517 
518 			last_byte =
519 			    *((unsigned char *)rx_buf->address + length - 1);
520 
521 			if (TBI_ACCEPT(hw,
522 			    current_desc->status, current_desc->errors,
523 			    current_desc->length, last_byte)) {
524 
525 				e1000_tbi_adjust_stats(Adapter,
526 				    length, hw->mac.addr);
527 
528 				length--;
529 				accept_frame = B_TRUE;
530 			} else if (e1000_tbi_sbp_enabled_82543(hw) &&
531 			    (current_desc->errors == E1000_RXD_ERR_CE)) {
532 				accept_frame = B_TRUE;
533 			}
534 		}
535 
536 		/*
537 		 * Indicate the packet to the NOS if it was good.
538 		 * Normally, hardware will discard bad packets for us.
539 		 * Check for the packet to be a valid Ethernet packet
540 		 */
541 		if (!accept_frame) {
542 			/*
543 			 * error in incoming packet, either the packet is not a
544 			 * ethernet size packet, or the packet has an error. In
545 			 * either case, the packet will simply be discarded.
546 			 */
547 			E1000G_DEBUGLOG_0(Adapter, E1000G_INFO_LEVEL,
548 			    "Process Receive Interrupts: Error in Packet\n");
549 
550 			E1000G_STAT(rx_ring->stat_error);
551 			/*
552 			 * Returning here as we are done here. There is
553 			 * no point in waiting for while loop to elapse
554 			 * and the things which were done. More efficient
555 			 * and less error prone...
556 			 */
557 			goto rx_drop;
558 		}
559 
560 		/*
561 		 * If the Ethernet CRC is not stripped by the hardware,
562 		 * we need to strip it before sending it up to the stack.
563 		 */
564 		if (end_of_packet && !Adapter->strip_crc) {
565 			if (length > CRC_LENGTH) {
566 				length -= CRC_LENGTH;
567 			} else {
568 				/*
569 				 * If the fragment is smaller than the CRC,
570 				 * drop this fragment, do the processing of
571 				 * the end of the packet.
572 				 */
573 				ASSERT(rx_ring->rx_mblk_tail != NULL);
574 				rx_ring->rx_mblk_tail->b_wptr -=
575 				    CRC_LENGTH - length;
576 				rx_ring->rx_mblk_len -=
577 				    CRC_LENGTH - length;
578 
579 				QUEUE_POP_HEAD(&rx_ring->recv_list);
580 
581 				goto rx_end_of_packet;
582 			}
583 		}
584 
585 		need_copy = B_TRUE;
586 
587 		if (length <= Adapter->rx_bcopy_thresh)
588 			goto rx_copy;
589 
590 		/*
591 		 * Get the pre-constructed mblk that was associated
592 		 * to the receive data buffer.
593 		 */
594 		if (packet->mp == NULL) {
595 			packet->mp = desballoc((unsigned char *)
596 			    rx_buf->address - E1000G_IPALIGNROOM,
597 			    length + E1000G_IPALIGNROOM,
598 			    BPRI_MED, &packet->free_rtn);
599 
600 			if (packet->mp != NULL) {
601 				packet->mp->b_rptr += E1000G_IPALIGNROOM;
602 				packet->mp->b_wptr += E1000G_IPALIGNROOM;
603 			} else {
604 				E1000G_STAT(rx_ring->stat_esballoc_fail);
605 			}
606 		}
607 
608 		if (packet->mp != NULL) {
609 			/*
610 			 * We have two sets of buffer pool. One associated with
611 			 * the Rxdescriptors and other a freelist buffer pool.
612 			 * Each time we get a good packet, Try to get a buffer
613 			 * from the freelist pool using e1000g_get_buf. If we
614 			 * get free buffer, then replace the descriptor buffer
615 			 * address with the free buffer we just got, and pass
616 			 * the pre-constructed mblk upstack. (note no copying)
617 			 *
618 			 * If we failed to get a free buffer, then try to
619 			 * allocate a new buffer(mp) and copy the recv buffer
620 			 * content to our newly allocated buffer(mp). Don't
621 			 * disturb the desriptor buffer address. (note copying)
622 			 */
623 			newpkt = e1000g_get_buf(rx_ring);
624 
625 			if (newpkt != NULL) {
626 				/*
627 				 * Get the mblk associated to the data,
628 				 * and strip it off the sw packet.
629 				 */
630 				nmp = packet->mp;
631 				packet->mp = NULL;
632 				packet->flag = E1000G_RX_SW_SENDUP;
633 
634 				/*
635 				 * Now replace old buffer with the new
636 				 * one we got from free list
637 				 * Both the RxSwPacket as well as the
638 				 * Receive Buffer Descriptor will now
639 				 * point to this new packet.
640 				 */
641 				packet = newpkt;
642 
643 				current_desc->buffer_addr =
644 				    newpkt->rx_buf->dma_address;
645 
646 				need_copy = B_FALSE;
647 			} else {
648 				E1000G_DEBUG_STAT(rx_ring->stat_no_freepkt);
649 			}
650 		}
651 
652 rx_copy:
653 		if (need_copy) {
654 			/*
655 			 * No buffers available on free list,
656 			 * bcopy the data from the buffer and
657 			 * keep the original buffer. Dont want to
658 			 * do this.. Yack but no other way
659 			 */
660 			if ((nmp = allocb(length + E1000G_IPALIGNROOM,
661 			    BPRI_MED)) == NULL) {
662 				/*
663 				 * The system has no buffers available
664 				 * to send up the incoming packet, hence
665 				 * the packet will have to be processed
666 				 * when there're more buffers available.
667 				 */
668 				E1000G_STAT(rx_ring->stat_allocb_fail);
669 				goto rx_drop;
670 			}
671 			nmp->b_rptr += E1000G_IPALIGNROOM;
672 			nmp->b_wptr += E1000G_IPALIGNROOM;
673 			/*
674 			 * The free list did not have any buffers
675 			 * available, so, the received packet will
676 			 * have to be copied into a mp and the original
677 			 * buffer will have to be retained for future
678 			 * packet reception.
679 			 */
680 			bcopy(rx_buf->address, nmp->b_wptr, length);
681 		}
682 
683 		/*
684 		 * The rx_sw_packet MUST be popped off the
685 		 * RxSwPacketList before either a putnext or freemsg
686 		 * is done on the mp that has now been created by the
687 		 * desballoc. If not, it is possible that the free
688 		 * routine will get called from the interrupt context
689 		 * and try to put this packet on the free list
690 		 */
691 		(p_rx_sw_packet_t)QUEUE_POP_HEAD(&rx_ring->recv_list);
692 
693 		ASSERT(nmp != NULL);
694 		nmp->b_wptr += length;
695 
696 		if (rx_ring->rx_mblk == NULL) {
697 			/*
698 			 *  TCP/UDP checksum offload and
699 			 *  IP checksum offload
700 			 */
701 			if (!(current_desc->status & E1000_RXD_STAT_IXSM)) {
702 				/*
703 				 * Check TCP/UDP checksum
704 				 */
705 				if ((current_desc->status &
706 				    E1000_RXD_STAT_TCPCS) &&
707 				    !(current_desc->errors &
708 				    E1000_RXD_ERR_TCPE))
709 					cksumflags |= HCK_FULLCKSUM |
710 					    HCK_FULLCKSUM_OK;
711 				/*
712 				 * Check IP Checksum
713 				 */
714 				if ((current_desc->status &
715 				    E1000_RXD_STAT_IPCS) &&
716 				    !(current_desc->errors &
717 				    E1000_RXD_ERR_IPE))
718 					cksumflags |= HCK_IPV4_HDRCKSUM;
719 			}
720 		}
721 
722 		/*
723 		 * We need to maintain our packet chain in the global
724 		 * Adapter structure, for the Rx processing can end
725 		 * with a fragment that has no EOP set.
726 		 */
727 		if (rx_ring->rx_mblk == NULL) {
728 			/* Get the head of the message chain */
729 			rx_ring->rx_mblk = nmp;
730 			rx_ring->rx_mblk_tail = nmp;
731 			rx_ring->rx_mblk_len = length;
732 		} else {	/* Not the first packet */
733 			/* Continue adding buffers */
734 			rx_ring->rx_mblk_tail->b_cont = nmp;
735 			rx_ring->rx_mblk_tail = nmp;
736 			rx_ring->rx_mblk_len += length;
737 		}
738 		ASSERT(rx_ring->rx_mblk != NULL);
739 		ASSERT(rx_ring->rx_mblk_tail != NULL);
740 		ASSERT(rx_ring->rx_mblk_tail->b_cont == NULL);
741 
742 		/*
743 		 * Now this MP is ready to travel upwards but some more
744 		 * fragments are coming.
745 		 * We will send packet upwards as soon as we get EOP
746 		 * set on the packet.
747 		 */
748 		if (!end_of_packet) {
749 			/*
750 			 * continue to get the next descriptor,
751 			 * Tail would be advanced at the end
752 			 */
753 			goto rx_next_desc;
754 		}
755 
756 rx_end_of_packet:
757 		/*
758 		 * Found packet with EOP
759 		 * Process the last fragment.
760 		 */
761 		if (cksumflags != 0) {
762 			(void) hcksum_assoc(rx_ring->rx_mblk,
763 			    NULL, NULL, 0, 0, 0, 0, cksumflags, 0);
764 			cksumflags = 0;
765 		}
766 
767 		/*
768 		 * Count packets that span multi-descriptors
769 		 */
770 		E1000G_DEBUG_STAT_COND(rx_ring->stat_multi_desc,
771 		    (rx_ring->rx_mblk->b_cont != NULL));
772 
773 		/*
774 		 * Append to list to send upstream
775 		 */
776 		if (ret_mp == NULL) {
777 			ret_mp = ret_nmp = rx_ring->rx_mblk;
778 		} else {
779 			ret_nmp->b_next = rx_ring->rx_mblk;
780 			ret_nmp = rx_ring->rx_mblk;
781 		}
782 		ret_nmp->b_next = NULL;
783 
784 		rx_ring->rx_mblk = NULL;
785 		rx_ring->rx_mblk_tail = NULL;
786 		rx_ring->rx_mblk_len = 0;
787 
788 		pkt_count++;
789 
790 rx_next_desc:
791 		/*
792 		 * Zero out the receive descriptors status
793 		 */
794 		current_desc->status = 0;
795 
796 		if (current_desc == rx_ring->rbd_last)
797 			rx_ring->rbd_next = rx_ring->rbd_first;
798 		else
799 			rx_ring->rbd_next++;
800 
801 		last_desc = current_desc;
802 		current_desc = rx_ring->rbd_next;
803 
804 		/*
805 		 * Put the buffer that we just indicated back
806 		 * at the end of our list
807 		 */
808 		QUEUE_PUSH_TAIL(&rx_ring->recv_list,
809 		    &packet->Link);
810 	}	/* while loop */
811 
812 	/* Sync the Rx descriptor DMA buffers */
813 	(void) ddi_dma_sync(rx_ring->rbd_dma_handle,
814 	    0, 0, DDI_DMA_SYNC_FORDEV);
815 
816 	/*
817 	 * Advance the E1000's Receive Queue #0 "Tail Pointer".
818 	 */
819 	E1000_WRITE_REG(hw, E1000_RDT,
820 	    (uint32_t)(last_desc - rx_ring->rbd_first));
821 
822 	if (e1000g_check_acc_handle(Adapter->osdep.reg_handle) != DDI_FM_OK) {
823 		ddi_fm_service_impact(Adapter->dip, DDI_SERVICE_DEGRADED);
824 		Adapter->chip_state = E1000G_ERROR;
825 	}
826 
827 	Adapter->rx_pkt_cnt = pkt_count;
828 
829 	return (ret_mp);
830 
831 rx_drop:
832 	/*
833 	 * Zero out the receive descriptors status
834 	 */
835 	current_desc->status = 0;
836 
837 	/* Sync the Rx descriptor DMA buffers */
838 	(void) ddi_dma_sync(rx_ring->rbd_dma_handle,
839 	    0, 0, DDI_DMA_SYNC_FORDEV);
840 
841 	if (current_desc == rx_ring->rbd_last)
842 		rx_ring->rbd_next = rx_ring->rbd_first;
843 	else
844 		rx_ring->rbd_next++;
845 
846 	last_desc = current_desc;
847 
848 	(p_rx_sw_packet_t)QUEUE_POP_HEAD(&rx_ring->recv_list);
849 
850 	QUEUE_PUSH_TAIL(&rx_ring->recv_list, &packet->Link);
851 	/*
852 	 * Reclaim all old buffers already allocated during
853 	 * Jumbo receives.....for incomplete reception
854 	 */
855 	if (rx_ring->rx_mblk != NULL) {
856 		freemsg(rx_ring->rx_mblk);
857 		rx_ring->rx_mblk = NULL;
858 		rx_ring->rx_mblk_tail = NULL;
859 		rx_ring->rx_mblk_len = 0;
860 	}
861 	/*
862 	 * Advance the E1000's Receive Queue #0 "Tail Pointer".
863 	 */
864 	E1000_WRITE_REG(hw, E1000_RDT,
865 	    (uint32_t)(last_desc - rx_ring->rbd_first));
866 
867 	if (e1000g_check_acc_handle(Adapter->osdep.reg_handle) != DDI_FM_OK) {
868 		ddi_fm_service_impact(Adapter->dip, DDI_SERVICE_DEGRADED);
869 		Adapter->chip_state = E1000G_ERROR;
870 	}
871 
872 	return (ret_mp);
873 }
874