xref: /illumos-gate/usr/src/uts/common/io/hxge/hxge_main.c (revision 08f1bbed5edd2a2e9c8be7b7424c32e67c2f3f2c)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  * Copyright 2012 Milan Jurik. All rights reserved.
25  * Copyright 2016 OmniTI Computer Consulting, Inc. All rights reserved.
26  */
27 
28 /*
29  * SunOs MT STREAMS Hydra 10Gb Ethernet Device Driver.
30  */
31 #include <hxge_impl.h>
32 #include <hxge_pfc.h>
33 
34 /*
35  * PSARC/2007/453 MSI-X interrupt limit override
36  * (This PSARC case is limited to MSI-X vectors
37  *  and SPARC platforms only).
38  */
39 uint32_t hxge_msi_enable = 2;
40 
41 /*
42  * Globals: tunable parameters (/etc/system or adb)
43  *
44  */
45 uint32_t hxge_rbr_size = HXGE_RBR_RBB_DEFAULT;
46 uint32_t hxge_rbr_spare_size = 0;
47 uint32_t hxge_rcr_size = HXGE_RCR_DEFAULT;
48 uint32_t hxge_tx_ring_size = HXGE_TX_RING_DEFAULT;
49 uint32_t hxge_bcopy_thresh = TX_BCOPY_MAX;
50 uint32_t hxge_dvma_thresh = TX_FASTDVMA_MIN;
51 uint32_t hxge_dma_stream_thresh = TX_STREAM_MIN;
52 uint32_t hxge_jumbo_frame_size = MAX_FRAME_SIZE;
53 
54 static hxge_os_mutex_t hxgedebuglock;
55 static int hxge_debug_init = 0;
56 
57 /*
58  * Debugging flags:
59  *		hxge_no_tx_lb : transmit load balancing
60  *		hxge_tx_lb_policy: 0 - TCP/UDP port (default)
61  *				   1 - From the Stack
62  *				   2 - Destination IP Address
63  */
64 uint32_t hxge_no_tx_lb = 0;
65 uint32_t hxge_tx_lb_policy = HXGE_TX_LB_TCPUDP;
66 
67 /*
68  * Tunables to manage the receive buffer blocks.
69  *
70  * hxge_rx_threshold_hi: copy all buffers.
71  * hxge_rx_bcopy_size_type: receive buffer block size type.
72  * hxge_rx_threshold_lo: copy only up to tunable block size type.
73  */
74 #if defined(__sparc)
75 hxge_rxbuf_threshold_t hxge_rx_threshold_hi = HXGE_RX_COPY_6;
76 hxge_rxbuf_threshold_t hxge_rx_threshold_lo = HXGE_RX_COPY_4;
77 #else
78 hxge_rxbuf_threshold_t hxge_rx_threshold_hi = HXGE_RX_COPY_NONE;
79 hxge_rxbuf_threshold_t hxge_rx_threshold_lo = HXGE_RX_COPY_NONE;
80 #endif
81 hxge_rxbuf_type_t hxge_rx_buf_size_type = RCR_PKTBUFSZ_0;
82 
83 rtrace_t hpi_rtracebuf;
84 
85 /*
86  * Function Prototypes
87  */
88 static int hxge_attach(dev_info_t *, ddi_attach_cmd_t);
89 static int hxge_detach(dev_info_t *, ddi_detach_cmd_t);
90 static void hxge_unattach(p_hxge_t);
91 
92 static hxge_status_t hxge_setup_system_dma_pages(p_hxge_t);
93 
94 static hxge_status_t hxge_setup_mutexes(p_hxge_t);
95 static void hxge_destroy_mutexes(p_hxge_t);
96 
97 static hxge_status_t hxge_map_regs(p_hxge_t hxgep);
98 static void hxge_unmap_regs(p_hxge_t hxgep);
99 
100 static hxge_status_t hxge_add_intrs(p_hxge_t hxgep);
101 static void hxge_remove_intrs(p_hxge_t hxgep);
102 static hxge_status_t hxge_add_intrs_adv(p_hxge_t hxgep);
103 static hxge_status_t hxge_add_intrs_adv_type(p_hxge_t, uint32_t);
104 static hxge_status_t hxge_add_intrs_adv_type_fix(p_hxge_t, uint32_t);
105 static void hxge_intrs_enable(p_hxge_t hxgep);
106 static void hxge_intrs_disable(p_hxge_t hxgep);
107 static void hxge_suspend(p_hxge_t);
108 static hxge_status_t hxge_resume(p_hxge_t);
109 static hxge_status_t hxge_setup_dev(p_hxge_t);
110 static void hxge_destroy_dev(p_hxge_t);
111 static hxge_status_t hxge_alloc_mem_pool(p_hxge_t);
112 static void hxge_free_mem_pool(p_hxge_t);
113 static hxge_status_t hxge_alloc_rx_mem_pool(p_hxge_t);
114 static void hxge_free_rx_mem_pool(p_hxge_t);
115 static hxge_status_t hxge_alloc_tx_mem_pool(p_hxge_t);
116 static void hxge_free_tx_mem_pool(p_hxge_t);
117 static hxge_status_t hxge_dma_mem_alloc(p_hxge_t, dma_method_t,
118     struct ddi_dma_attr *, size_t, ddi_device_acc_attr_t *, uint_t,
119     p_hxge_dma_common_t);
120 static void hxge_dma_mem_free(p_hxge_dma_common_t);
121 static hxge_status_t hxge_alloc_rx_buf_dma(p_hxge_t, uint16_t,
122     p_hxge_dma_common_t *, size_t, size_t, uint32_t *);
123 static void hxge_free_rx_buf_dma(p_hxge_t, p_hxge_dma_common_t, uint32_t);
124 static hxge_status_t hxge_alloc_rx_cntl_dma(p_hxge_t, uint16_t,
125     p_hxge_dma_common_t *, struct ddi_dma_attr *, size_t);
126 static void hxge_free_rx_cntl_dma(p_hxge_t, p_hxge_dma_common_t);
127 static hxge_status_t hxge_alloc_tx_buf_dma(p_hxge_t, uint16_t,
128     p_hxge_dma_common_t *, size_t, size_t, uint32_t *);
129 static void hxge_free_tx_buf_dma(p_hxge_t, p_hxge_dma_common_t, uint32_t);
130 static hxge_status_t hxge_alloc_tx_cntl_dma(p_hxge_t, uint16_t,
131     p_hxge_dma_common_t *, size_t);
132 static void hxge_free_tx_cntl_dma(p_hxge_t, p_hxge_dma_common_t);
133 static int hxge_init_common_dev(p_hxge_t);
134 static void hxge_uninit_common_dev(p_hxge_t);
135 
136 /*
137  * The next declarations are for the GLDv3 interface.
138  */
139 static int hxge_m_start(void *);
140 static void hxge_m_stop(void *);
141 static int hxge_m_multicst(void *, boolean_t, const uint8_t *);
142 static int hxge_m_promisc(void *, boolean_t);
143 static void hxge_m_ioctl(void *, queue_t *, mblk_t *);
144 static hxge_status_t hxge_mac_register(p_hxge_t hxgep);
145 
146 static boolean_t hxge_m_getcapab(void *, mac_capab_t, void *);
147 static boolean_t hxge_param_locked(mac_prop_id_t pr_num);
148 static int hxge_m_setprop(void *barg, const char *pr_name, mac_prop_id_t pr_num,
149     uint_t pr_valsize, const void *pr_val);
150 static int hxge_m_getprop(void *barg, const char *pr_name, mac_prop_id_t pr_num,
151     uint_t pr_valsize, void *pr_val);
152 static void hxge_m_propinfo(void *barg, const char *pr_name,
153     mac_prop_id_t pr_num, mac_prop_info_handle_t mph);
154 static int hxge_set_priv_prop(p_hxge_t hxgep, const char *pr_name,
155     uint_t pr_valsize, const void *pr_val);
156 static int hxge_get_priv_prop(p_hxge_t hxgep, const char *pr_name,
157     uint_t pr_valsize, void *pr_val);
158 static void hxge_link_poll(void *arg);
159 static void hxge_link_update(p_hxge_t hxge, link_state_t state);
160 static void hxge_msix_init(p_hxge_t hxgep);
161 
162 char *hxge_priv_props[] = {
163 	"_rxdma_intr_time",
164 	"_rxdma_intr_pkts",
165 	"_class_opt_ipv4_tcp",
166 	"_class_opt_ipv4_udp",
167 	"_class_opt_ipv4_ah",
168 	"_class_opt_ipv4_sctp",
169 	"_class_opt_ipv6_tcp",
170 	"_class_opt_ipv6_udp",
171 	"_class_opt_ipv6_ah",
172 	"_class_opt_ipv6_sctp",
173 	NULL
174 };
175 
176 #define	HXGE_MAX_PRIV_PROPS	\
177 	(sizeof (hxge_priv_props)/sizeof (mac_priv_prop_t))
178 
179 #define	HXGE_MAGIC	0x4E584745UL
180 #define	MAX_DUMP_SZ 256
181 
182 #define	HXGE_M_CALLBACK_FLAGS	\
183 	(MC_IOCTL | MC_GETCAPAB | MC_SETPROP | MC_GETPROP | MC_PROPINFO)
184 
185 extern hxge_status_t hxge_pfc_set_default_mac_addr(p_hxge_t hxgep);
186 
187 static mac_callbacks_t hxge_m_callbacks = {
188 	HXGE_M_CALLBACK_FLAGS,
189 	hxge_m_stat,
190 	hxge_m_start,
191 	hxge_m_stop,
192 	hxge_m_promisc,
193 	hxge_m_multicst,
194 	NULL,
195 	NULL,
196 	NULL,
197 	hxge_m_ioctl,
198 	hxge_m_getcapab,
199 	NULL,
200 	NULL,
201 	hxge_m_setprop,
202 	hxge_m_getprop,
203 	hxge_m_propinfo
204 };
205 
206 /* PSARC/2007/453 MSI-X interrupt limit override. */
207 #define	HXGE_MSIX_REQUEST_10G	8
208 static int hxge_create_msi_property(p_hxge_t);
209 
210 /* Enable debug messages as necessary. */
211 uint64_t hxge_debug_level = 0;
212 
213 /*
214  * This list contains the instance structures for the Hydra
215  * devices present in the system. The lock exists to guarantee
216  * mutually exclusive access to the list.
217  */
218 void *hxge_list = NULL;
219 void *hxge_hw_list = NULL;
220 hxge_os_mutex_t hxge_common_lock;
221 
222 extern uint64_t hpi_debug_level;
223 
224 extern hxge_status_t hxge_ldgv_init(p_hxge_t, int *, int *);
225 extern hxge_status_t hxge_ldgv_uninit(p_hxge_t);
226 extern hxge_status_t hxge_intr_ldgv_init(p_hxge_t);
227 extern void hxge_fm_init(p_hxge_t hxgep, ddi_device_acc_attr_t *reg_attr,
228     ddi_device_acc_attr_t *desc_attr, ddi_dma_attr_t *dma_attr);
229 extern void hxge_fm_fini(p_hxge_t hxgep);
230 
231 /*
232  * Count used to maintain the number of buffers being used
233  * by Hydra instances and loaned up to the upper layers.
234  */
235 uint32_t hxge_mblks_pending = 0;
236 
237 /*
238  * Device register access attributes for PIO.
239  */
240 static ddi_device_acc_attr_t hxge_dev_reg_acc_attr = {
241 	DDI_DEVICE_ATTR_V0,
242 	DDI_STRUCTURE_LE_ACC,
243 	DDI_STRICTORDER_ACC,
244 };
245 
246 /*
247  * Device descriptor access attributes for DMA.
248  */
249 static ddi_device_acc_attr_t hxge_dev_desc_dma_acc_attr = {
250 	DDI_DEVICE_ATTR_V0,
251 	DDI_STRUCTURE_LE_ACC,
252 	DDI_STRICTORDER_ACC
253 };
254 
255 /*
256  * Device buffer access attributes for DMA.
257  */
258 static ddi_device_acc_attr_t hxge_dev_buf_dma_acc_attr = {
259 	DDI_DEVICE_ATTR_V0,
260 	DDI_STRUCTURE_BE_ACC,
261 	DDI_STRICTORDER_ACC
262 };
263 
264 ddi_dma_attr_t hxge_rx_rcr_desc_dma_attr = {
265 	DMA_ATTR_V0,		/* version number. */
266 	0,			/* low address */
267 	0xffffffffffffffff,	/* high address */
268 	0xffffffffffffffff,	/* address counter max */
269 	0x80000,		/* alignment */
270 	0xfc00fc,		/* dlim_burstsizes */
271 	0x1,			/* minimum transfer size */
272 	0xffffffffffffffff,	/* maximum transfer size */
273 	0xffffffffffffffff,	/* maximum segment size */
274 	1,			/* scatter/gather list length */
275 	(unsigned int)1,	/* granularity */
276 	0			/* attribute flags */
277 };
278 
279 ddi_dma_attr_t hxge_tx_desc_dma_attr = {
280 	DMA_ATTR_V0,		/* version number. */
281 	0,			/* low address */
282 	0xffffffffffffffff,	/* high address */
283 	0xffffffffffffffff,	/* address counter max */
284 	0x100000,		/* alignment */
285 	0xfc00fc,		/* dlim_burstsizes */
286 	0x1,			/* minimum transfer size */
287 	0xffffffffffffffff,	/* maximum transfer size */
288 	0xffffffffffffffff,	/* maximum segment size */
289 	1,			/* scatter/gather list length */
290 	(unsigned int)1,	/* granularity */
291 	0			/* attribute flags */
292 };
293 
294 ddi_dma_attr_t hxge_rx_rbr_desc_dma_attr = {
295 	DMA_ATTR_V0,		/* version number. */
296 	0,			/* low address */
297 	0xffffffffffffffff,	/* high address */
298 	0xffffffffffffffff,	/* address counter max */
299 	0x40000,		/* alignment */
300 	0xfc00fc,		/* dlim_burstsizes */
301 	0x1,			/* minimum transfer size */
302 	0xffffffffffffffff,	/* maximum transfer size */
303 	0xffffffffffffffff,	/* maximum segment size */
304 	1,			/* scatter/gather list length */
305 	(unsigned int)1,	/* granularity */
306 	0			/* attribute flags */
307 };
308 
309 ddi_dma_attr_t hxge_rx_mbox_dma_attr = {
310 	DMA_ATTR_V0,		/* version number. */
311 	0,			/* low address */
312 	0xffffffffffffffff,	/* high address */
313 	0xffffffffffffffff,	/* address counter max */
314 #if defined(_BIG_ENDIAN)
315 	0x2000,			/* alignment */
316 #else
317 	0x1000,			/* alignment */
318 #endif
319 	0xfc00fc,		/* dlim_burstsizes */
320 	0x1,			/* minimum transfer size */
321 	0xffffffffffffffff,	/* maximum transfer size */
322 	0xffffffffffffffff,	/* maximum segment size */
323 	5,			/* scatter/gather list length */
324 	(unsigned int)1,	/* granularity */
325 	0			/* attribute flags */
326 };
327 
328 ddi_dma_attr_t hxge_tx_dma_attr = {
329 	DMA_ATTR_V0,		/* version number. */
330 	0,			/* low address */
331 	0xffffffffffffffff,	/* high address */
332 	0xffffffffffffffff,	/* address counter max */
333 #if defined(_BIG_ENDIAN)
334 	0x2000,			/* alignment */
335 #else
336 	0x1000,			/* alignment */
337 #endif
338 	0xfc00fc,		/* dlim_burstsizes */
339 	0x1,			/* minimum transfer size */
340 	0xffffffffffffffff,	/* maximum transfer size */
341 	0xffffffffffffffff,	/* maximum segment size */
342 	5,			/* scatter/gather list length */
343 	(unsigned int)1,	/* granularity */
344 	0			/* attribute flags */
345 };
346 
347 ddi_dma_attr_t hxge_rx_dma_attr = {
348 	DMA_ATTR_V0,		/* version number. */
349 	0,			/* low address */
350 	0xffffffffffffffff,	/* high address */
351 	0xffffffffffffffff,	/* address counter max */
352 	0x10000,		/* alignment */
353 	0xfc00fc,		/* dlim_burstsizes */
354 	0x1,			/* minimum transfer size */
355 	0xffffffffffffffff,	/* maximum transfer size */
356 	0xffffffffffffffff,	/* maximum segment size */
357 	1,			/* scatter/gather list length */
358 	(unsigned int)1,	/* granularity */
359 	DDI_DMA_RELAXED_ORDERING /* attribute flags */
360 };
361 
362 ddi_dma_lim_t hxge_dma_limits = {
363 	(uint_t)0,		/* dlim_addr_lo */
364 	(uint_t)0xffffffff,	/* dlim_addr_hi */
365 	(uint_t)0xffffffff,	/* dlim_cntr_max */
366 	(uint_t)0xfc00fc,	/* dlim_burstsizes for 32 and 64 bit xfers */
367 	0x1,			/* dlim_minxfer */
368 	1024			/* dlim_speed */
369 };
370 
371 dma_method_t hxge_force_dma = DVMA;
372 
373 /*
374  * dma chunk sizes.
375  *
376  * Try to allocate the largest possible size
377  * so that fewer number of dma chunks would be managed
378  */
379 size_t alloc_sizes[] = {
380     0x1000, 0x2000, 0x4000, 0x8000,
381     0x10000, 0x20000, 0x40000, 0x80000,
382     0x100000, 0x200000, 0x400000, 0x800000, 0x1000000
383 };
384 
385 /*
386  * Translate "dev_t" to a pointer to the associated "dev_info_t".
387  */
388 static int
389 hxge_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
390 {
391 	p_hxge_t	hxgep = NULL;
392 	int		instance;
393 	int		status = DDI_SUCCESS;
394 	int		i;
395 
396 	HXGE_DEBUG_MSG((hxgep, DDI_CTL, "==> hxge_attach"));
397 
398 	/*
399 	 * Get the device instance since we'll need to setup or retrieve a soft
400 	 * state for this instance.
401 	 */
402 	instance = ddi_get_instance(dip);
403 
404 	switch (cmd) {
405 	case DDI_ATTACH:
406 		HXGE_DEBUG_MSG((hxgep, DDI_CTL, "doing DDI_ATTACH"));
407 		break;
408 
409 	case DDI_RESUME:
410 		HXGE_DEBUG_MSG((hxgep, DDI_CTL, "doing DDI_RESUME"));
411 		hxgep = (p_hxge_t)ddi_get_soft_state(hxge_list, instance);
412 		if (hxgep == NULL) {
413 			status = DDI_FAILURE;
414 			break;
415 		}
416 		if (hxgep->dip != dip) {
417 			status = DDI_FAILURE;
418 			break;
419 		}
420 		if (hxgep->suspended == DDI_PM_SUSPEND) {
421 			status = ddi_dev_is_needed(hxgep->dip, 0, 1);
422 		} else {
423 			(void) hxge_resume(hxgep);
424 		}
425 		goto hxge_attach_exit;
426 
427 	case DDI_PM_RESUME:
428 		HXGE_DEBUG_MSG((hxgep, DDI_CTL, "doing DDI_PM_RESUME"));
429 		hxgep = (p_hxge_t)ddi_get_soft_state(hxge_list, instance);
430 		if (hxgep == NULL) {
431 			status = DDI_FAILURE;
432 			break;
433 		}
434 		if (hxgep->dip != dip) {
435 			status = DDI_FAILURE;
436 			break;
437 		}
438 		(void) hxge_resume(hxgep);
439 		goto hxge_attach_exit;
440 
441 	default:
442 		HXGE_DEBUG_MSG((hxgep, DDI_CTL, "doing unknown"));
443 		status = DDI_FAILURE;
444 		goto hxge_attach_exit;
445 	}
446 
447 	if (ddi_soft_state_zalloc(hxge_list, instance) == DDI_FAILURE) {
448 		status = DDI_FAILURE;
449 		HXGE_ERROR_MSG((hxgep, DDI_CTL,
450 		    "ddi_soft_state_zalloc failed"));
451 		goto hxge_attach_exit;
452 	}
453 
454 	hxgep = ddi_get_soft_state(hxge_list, instance);
455 	if (hxgep == NULL) {
456 		status = HXGE_ERROR;
457 		HXGE_ERROR_MSG((hxgep, DDI_CTL,
458 		    "ddi_get_soft_state failed"));
459 		goto hxge_attach_fail2;
460 	}
461 
462 	hxgep->drv_state = 0;
463 	hxgep->dip = dip;
464 	hxgep->instance = instance;
465 	hxgep->p_dip = ddi_get_parent(dip);
466 	hxgep->hxge_debug_level = hxge_debug_level;
467 	hpi_debug_level = hxge_debug_level;
468 
469 	/*
470 	 * Initialize MMAC struture.
471 	 */
472 	(void) hxge_pfc_num_macs_get(hxgep, &hxgep->mmac.total);
473 	hxgep->mmac.available = hxgep->mmac.total;
474 	for (i = 0; i < hxgep->mmac.total; i++) {
475 		hxgep->mmac.addrs[i].set = B_FALSE;
476 		hxgep->mmac.addrs[i].primary = B_FALSE;
477 	}
478 
479 	hxge_fm_init(hxgep, &hxge_dev_reg_acc_attr, &hxge_dev_desc_dma_acc_attr,
480 	    &hxge_rx_dma_attr);
481 
482 	status = hxge_map_regs(hxgep);
483 	if (status != HXGE_OK) {
484 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL, "hxge_map_regs failed"));
485 		goto hxge_attach_fail3;
486 	}
487 
488 	status = hxge_init_common_dev(hxgep);
489 	if (status != HXGE_OK) {
490 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
491 		    "hxge_init_common_dev failed"));
492 		goto hxge_attach_fail4;
493 	}
494 
495 	/*
496 	 * Setup the Ndd parameters for this instance.
497 	 */
498 	hxge_init_param(hxgep);
499 
500 	/*
501 	 * Setup Register Tracing Buffer.
502 	 */
503 	hpi_rtrace_buf_init((rtrace_t *)&hpi_rtracebuf);
504 
505 	/* init stats ptr */
506 	hxge_init_statsp(hxgep);
507 
508 	status = hxge_setup_mutexes(hxgep);
509 	if (status != HXGE_OK) {
510 		HXGE_DEBUG_MSG((hxgep, DDI_CTL, "set mutex failed"));
511 		goto hxge_attach_fail;
512 	}
513 
514 	/* Scrub the MSI-X memory */
515 	hxge_msix_init(hxgep);
516 
517 	status = hxge_get_config_properties(hxgep);
518 	if (status != HXGE_OK) {
519 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL, "get_hw create failed"));
520 		goto hxge_attach_fail;
521 	}
522 
523 	/*
524 	 * Setup the Kstats for the driver.
525 	 */
526 	hxge_setup_kstats(hxgep);
527 	hxge_setup_param(hxgep);
528 
529 	status = hxge_setup_system_dma_pages(hxgep);
530 	if (status != HXGE_OK) {
531 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL, "set dma page failed"));
532 		goto hxge_attach_fail;
533 	}
534 
535 	hxge_hw_id_init(hxgep);
536 	hxge_hw_init_niu_common(hxgep);
537 
538 	status = hxge_setup_dev(hxgep);
539 	if (status != DDI_SUCCESS) {
540 		HXGE_DEBUG_MSG((hxgep, DDI_CTL, "set dev failed"));
541 		goto hxge_attach_fail;
542 	}
543 
544 	status = hxge_add_intrs(hxgep);
545 	if (status != DDI_SUCCESS) {
546 		HXGE_DEBUG_MSG((hxgep, DDI_CTL, "add_intr failed"));
547 		goto hxge_attach_fail;
548 	}
549 
550 	/*
551 	 * Enable interrupts.
552 	 */
553 	hxge_intrs_enable(hxgep);
554 
555 	if ((status = hxge_mac_register(hxgep)) != HXGE_OK) {
556 		HXGE_DEBUG_MSG((hxgep, DDI_CTL,
557 		    "unable to register to mac layer (%d)", status));
558 		goto hxge_attach_fail;
559 	}
560 	mac_link_update(hxgep->mach, LINK_STATE_UNKNOWN);
561 
562 	HXGE_DEBUG_MSG((hxgep, DDI_CTL, "registered to mac (instance %d)",
563 	    instance));
564 
565 	goto hxge_attach_exit;
566 
567 hxge_attach_fail:
568 	hxge_unattach(hxgep);
569 	goto hxge_attach_fail1;
570 
571 hxge_attach_fail5:
572 	/*
573 	 * Tear down the ndd parameters setup.
574 	 */
575 	hxge_destroy_param(hxgep);
576 
577 	/*
578 	 * Tear down the kstat setup.
579 	 */
580 	hxge_destroy_kstats(hxgep);
581 
582 hxge_attach_fail4:
583 	if (hxgep->hxge_hw_p) {
584 		hxge_uninit_common_dev(hxgep);
585 		hxgep->hxge_hw_p = NULL;
586 	}
587 hxge_attach_fail3:
588 	/*
589 	 * Unmap the register setup.
590 	 */
591 	hxge_unmap_regs(hxgep);
592 
593 	hxge_fm_fini(hxgep);
594 
595 hxge_attach_fail2:
596 	ddi_soft_state_free(hxge_list, hxgep->instance);
597 
598 hxge_attach_fail1:
599 	if (status != HXGE_OK)
600 		status = (HXGE_ERROR | HXGE_DDI_FAILED);
601 	hxgep = NULL;
602 
603 hxge_attach_exit:
604 	HXGE_DEBUG_MSG((hxgep, DDI_CTL, "<== hxge_attach status = 0x%08x",
605 	    status));
606 
607 	return (status);
608 }
609 
610 static int
611 hxge_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
612 {
613 	int		status = DDI_SUCCESS;
614 	int		instance;
615 	p_hxge_t	hxgep = NULL;
616 
617 	HXGE_DEBUG_MSG((hxgep, DDI_CTL, "==> hxge_detach"));
618 	instance = ddi_get_instance(dip);
619 	hxgep = ddi_get_soft_state(hxge_list, instance);
620 	if (hxgep == NULL) {
621 		status = DDI_FAILURE;
622 		goto hxge_detach_exit;
623 	}
624 
625 	switch (cmd) {
626 	case DDI_DETACH:
627 		HXGE_DEBUG_MSG((hxgep, DDI_CTL, "doing DDI_DETACH"));
628 		break;
629 
630 	case DDI_PM_SUSPEND:
631 		HXGE_DEBUG_MSG((hxgep, DDI_CTL, "doing DDI_PM_SUSPEND"));
632 		hxgep->suspended = DDI_PM_SUSPEND;
633 		hxge_suspend(hxgep);
634 		break;
635 
636 	case DDI_SUSPEND:
637 		HXGE_DEBUG_MSG((hxgep, DDI_CTL, "doing DDI_SUSPEND"));
638 		if (hxgep->suspended != DDI_PM_SUSPEND) {
639 			hxgep->suspended = DDI_SUSPEND;
640 			hxge_suspend(hxgep);
641 		}
642 		break;
643 
644 	default:
645 		status = DDI_FAILURE;
646 		break;
647 	}
648 
649 	if (cmd != DDI_DETACH)
650 		goto hxge_detach_exit;
651 
652 	/*
653 	 * Stop the xcvr polling.
654 	 */
655 	hxgep->suspended = cmd;
656 
657 	if (hxgep->mach && (status = mac_unregister(hxgep->mach)) != 0) {
658 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
659 		    "<== hxge_detach status = 0x%08X", status));
660 		return (DDI_FAILURE);
661 	}
662 	HXGE_DEBUG_MSG((hxgep, DDI_CTL,
663 	    "<== hxge_detach (mac_unregister) status = 0x%08X", status));
664 
665 	hxge_unattach(hxgep);
666 	hxgep = NULL;
667 
668 hxge_detach_exit:
669 	HXGE_DEBUG_MSG((hxgep, DDI_CTL, "<== hxge_detach status = 0x%08X",
670 	    status));
671 
672 	return (status);
673 }
674 
675 static void
676 hxge_unattach(p_hxge_t hxgep)
677 {
678 	HXGE_DEBUG_MSG((hxgep, DDI_CTL, "==> hxge_unattach"));
679 
680 	if (hxgep == NULL || hxgep->dev_regs == NULL) {
681 		return;
682 	}
683 
684 	if (hxgep->hxge_hw_p) {
685 		hxge_uninit_common_dev(hxgep);
686 		hxgep->hxge_hw_p = NULL;
687 	}
688 
689 	if (hxgep->hxge_timerid) {
690 		hxge_stop_timer(hxgep, hxgep->hxge_timerid);
691 		hxgep->hxge_timerid = 0;
692 	}
693 
694 	/* Stop interrupts. */
695 	hxge_intrs_disable(hxgep);
696 
697 	/* Stop any further interrupts. */
698 	hxge_remove_intrs(hxgep);
699 
700 	/* Stop the device and free resources. */
701 	hxge_destroy_dev(hxgep);
702 
703 	/* Tear down the ndd parameters setup. */
704 	hxge_destroy_param(hxgep);
705 
706 	/* Tear down the kstat setup. */
707 	hxge_destroy_kstats(hxgep);
708 
709 	/*
710 	 * Remove the list of ndd parameters which were setup during attach.
711 	 */
712 	if (hxgep->dip) {
713 		HXGE_DEBUG_MSG((hxgep, OBP_CTL,
714 		    " hxge_unattach: remove all properties"));
715 		(void) ddi_prop_remove_all(hxgep->dip);
716 	}
717 
718 	/*
719 	 * Reset RDC, TDC, PFC, and VMAC blocks from PEU to clear any
720 	 * previous state before unmapping the registers.
721 	 */
722 	HXGE_REG_WR32(hxgep->hpi_handle, BLOCK_RESET, 0x0000001E);
723 	HXGE_DELAY(1000);
724 
725 	/*
726 	 * Unmap the register setup.
727 	 */
728 	hxge_unmap_regs(hxgep);
729 
730 	hxge_fm_fini(hxgep);
731 
732 	/* Destroy all mutexes.  */
733 	hxge_destroy_mutexes(hxgep);
734 
735 	/*
736 	 * Free the soft state data structures allocated with this instance.
737 	 */
738 	ddi_soft_state_free(hxge_list, hxgep->instance);
739 
740 	HXGE_DEBUG_MSG((NULL, DDI_CTL, "<== hxge_unattach"));
741 }
742 
743 static hxge_status_t
744 hxge_map_regs(p_hxge_t hxgep)
745 {
746 	int		ddi_status = DDI_SUCCESS;
747 	p_dev_regs_t	dev_regs;
748 
749 #ifdef	HXGE_DEBUG
750 	char		*sysname;
751 #endif
752 
753 	off_t		regsize;
754 	hxge_status_t	status = HXGE_OK;
755 	int		nregs;
756 
757 	HXGE_DEBUG_MSG((hxgep, DDI_CTL, "==> hxge_map_regs"));
758 
759 	if (ddi_dev_nregs(hxgep->dip, &nregs) != DDI_SUCCESS)
760 		return (HXGE_ERROR);
761 
762 	HXGE_DEBUG_MSG((hxgep, DDI_CTL, "hxge_map_regs: nregs: %d", nregs));
763 
764 	hxgep->dev_regs = NULL;
765 	dev_regs = KMEM_ZALLOC(sizeof (dev_regs_t), KM_SLEEP);
766 	dev_regs->hxge_regh = NULL;
767 	dev_regs->hxge_pciregh = NULL;
768 	dev_regs->hxge_msix_regh = NULL;
769 
770 	(void) ddi_dev_regsize(hxgep->dip, 0, &regsize);
771 	HXGE_DEBUG_MSG((hxgep, DDI_CTL,
772 	    "hxge_map_regs: pci config size 0x%x", regsize));
773 
774 	ddi_status = ddi_regs_map_setup(hxgep->dip, 0,
775 	    (caddr_t *)&(dev_regs->hxge_pciregp), 0, 0,
776 	    &hxge_dev_reg_acc_attr, &dev_regs->hxge_pciregh);
777 	if (ddi_status != DDI_SUCCESS) {
778 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
779 		    "ddi_map_regs, hxge bus config regs failed"));
780 		goto hxge_map_regs_fail0;
781 	}
782 
783 	HXGE_DEBUG_MSG((hxgep, DDI_CTL,
784 	    "hxge_map_reg: PCI config addr 0x%0llx handle 0x%0llx",
785 	    dev_regs->hxge_pciregp,
786 	    dev_regs->hxge_pciregh));
787 
788 	(void) ddi_dev_regsize(hxgep->dip, 1, &regsize);
789 	HXGE_DEBUG_MSG((hxgep, DDI_CTL,
790 	    "hxge_map_regs: pio size 0x%x", regsize));
791 
792 	/* set up the device mapped register */
793 	ddi_status = ddi_regs_map_setup(hxgep->dip, 1,
794 	    (caddr_t *)&(dev_regs->hxge_regp), 0, 0,
795 	    &hxge_dev_reg_acc_attr, &dev_regs->hxge_regh);
796 
797 	if (ddi_status != DDI_SUCCESS) {
798 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
799 		    "ddi_map_regs for Hydra global reg failed"));
800 		goto hxge_map_regs_fail1;
801 	}
802 
803 	/* set up the msi/msi-x mapped register */
804 	(void) ddi_dev_regsize(hxgep->dip, 2, &regsize);
805 	HXGE_DEBUG_MSG((hxgep, DDI_CTL,
806 	    "hxge_map_regs: msix size 0x%x", regsize));
807 
808 	ddi_status = ddi_regs_map_setup(hxgep->dip, 2,
809 	    (caddr_t *)&(dev_regs->hxge_msix_regp), 0, 0,
810 	    &hxge_dev_reg_acc_attr, &dev_regs->hxge_msix_regh);
811 
812 	if (ddi_status != DDI_SUCCESS) {
813 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
814 		    "ddi_map_regs for msi reg failed"));
815 		goto hxge_map_regs_fail2;
816 	}
817 
818 	hxgep->dev_regs = dev_regs;
819 
820 	HPI_PCI_ACC_HANDLE_SET(hxgep, dev_regs->hxge_pciregh);
821 	HPI_PCI_ADD_HANDLE_SET(hxgep, (hpi_reg_ptr_t)dev_regs->hxge_pciregp);
822 	HPI_MSI_ACC_HANDLE_SET(hxgep, dev_regs->hxge_msix_regh);
823 	HPI_MSI_ADD_HANDLE_SET(hxgep, (hpi_reg_ptr_t)dev_regs->hxge_msix_regp);
824 
825 	HPI_ACC_HANDLE_SET(hxgep, dev_regs->hxge_regh);
826 	HPI_ADD_HANDLE_SET(hxgep, (hpi_reg_ptr_t)dev_regs->hxge_regp);
827 
828 	HPI_REG_ACC_HANDLE_SET(hxgep, dev_regs->hxge_regh);
829 	HPI_REG_ADD_HANDLE_SET(hxgep, (hpi_reg_ptr_t)dev_regs->hxge_regp);
830 
831 	HXGE_DEBUG_MSG((hxgep, DDI_CTL, "hxge_map_reg: hardware addr 0x%0llx "
832 	    " handle 0x%0llx", dev_regs->hxge_regp, dev_regs->hxge_regh));
833 
834 	goto hxge_map_regs_exit;
835 
836 hxge_map_regs_fail3:
837 	if (dev_regs->hxge_msix_regh) {
838 		ddi_regs_map_free(&dev_regs->hxge_msix_regh);
839 	}
840 
841 hxge_map_regs_fail2:
842 	if (dev_regs->hxge_regh) {
843 		ddi_regs_map_free(&dev_regs->hxge_regh);
844 	}
845 
846 hxge_map_regs_fail1:
847 	if (dev_regs->hxge_pciregh) {
848 		ddi_regs_map_free(&dev_regs->hxge_pciregh);
849 	}
850 
851 hxge_map_regs_fail0:
852 	HXGE_DEBUG_MSG((hxgep, DDI_CTL, "Freeing register set memory"));
853 	kmem_free(dev_regs, sizeof (dev_regs_t));
854 
855 hxge_map_regs_exit:
856 	if (ddi_status != DDI_SUCCESS)
857 		status |= (HXGE_ERROR | HXGE_DDI_FAILED);
858 	HXGE_DEBUG_MSG((hxgep, DDI_CTL, "<== hxge_map_regs"));
859 	return (status);
860 }
861 
862 static void
863 hxge_unmap_regs(p_hxge_t hxgep)
864 {
865 	HXGE_DEBUG_MSG((hxgep, DDI_CTL, "==> hxge_unmap_regs"));
866 	if (hxgep->dev_regs) {
867 		if (hxgep->dev_regs->hxge_pciregh) {
868 			HXGE_DEBUG_MSG((hxgep, DDI_CTL,
869 			    "==> hxge_unmap_regs: bus"));
870 			ddi_regs_map_free(&hxgep->dev_regs->hxge_pciregh);
871 			hxgep->dev_regs->hxge_pciregh = NULL;
872 		}
873 
874 		if (hxgep->dev_regs->hxge_regh) {
875 			HXGE_DEBUG_MSG((hxgep, DDI_CTL,
876 			    "==> hxge_unmap_regs: device registers"));
877 			ddi_regs_map_free(&hxgep->dev_regs->hxge_regh);
878 			hxgep->dev_regs->hxge_regh = NULL;
879 		}
880 
881 		if (hxgep->dev_regs->hxge_msix_regh) {
882 			HXGE_DEBUG_MSG((hxgep, DDI_CTL,
883 			    "==> hxge_unmap_regs: device interrupts"));
884 			ddi_regs_map_free(&hxgep->dev_regs->hxge_msix_regh);
885 			hxgep->dev_regs->hxge_msix_regh = NULL;
886 		}
887 		kmem_free(hxgep->dev_regs, sizeof (dev_regs_t));
888 		hxgep->dev_regs = NULL;
889 	}
890 	HXGE_DEBUG_MSG((hxgep, DDI_CTL, "<== hxge_unmap_regs"));
891 }
892 
893 static hxge_status_t
894 hxge_setup_mutexes(p_hxge_t hxgep)
895 {
896 	int		ddi_status = DDI_SUCCESS;
897 	hxge_status_t	status = HXGE_OK;
898 
899 	HXGE_DEBUG_MSG((hxgep, DDI_CTL, "==> hxge_setup_mutexes"));
900 
901 	/*
902 	 * Get the interrupt cookie so the mutexes can be Initialised.
903 	 */
904 	ddi_status = ddi_get_iblock_cookie(hxgep->dip, 0,
905 	    &hxgep->interrupt_cookie);
906 
907 	if (ddi_status != DDI_SUCCESS) {
908 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
909 		    "<== hxge_setup_mutexes: failed 0x%x", ddi_status));
910 		goto hxge_setup_mutexes_exit;
911 	}
912 
913 	/*
914 	 * Initialize mutex's for this device.
915 	 */
916 	MUTEX_INIT(hxgep->genlock, NULL,
917 	    MUTEX_DRIVER, (void *) hxgep->interrupt_cookie);
918 	MUTEX_INIT(&hxgep->vmac_lock, NULL,
919 	    MUTEX_DRIVER, (void *) hxgep->interrupt_cookie);
920 	MUTEX_INIT(&hxgep->ouraddr_lock, NULL,
921 	    MUTEX_DRIVER, (void *) hxgep->interrupt_cookie);
922 	RW_INIT(&hxgep->filter_lock, NULL,
923 	    RW_DRIVER, (void *) hxgep->interrupt_cookie);
924 	MUTEX_INIT(&hxgep->pio_lock, NULL,
925 	    MUTEX_DRIVER, (void *) hxgep->interrupt_cookie);
926 	MUTEX_INIT(&hxgep->timeout.lock, NULL,
927 	    MUTEX_DRIVER, (void *) hxgep->interrupt_cookie);
928 
929 hxge_setup_mutexes_exit:
930 	HXGE_DEBUG_MSG((hxgep, DDI_CTL,
931 	    "<== hxge_setup_mutexes status = %x", status));
932 
933 	if (ddi_status != DDI_SUCCESS)
934 		status |= (HXGE_ERROR | HXGE_DDI_FAILED);
935 
936 	return (status);
937 }
938 
939 static void
940 hxge_destroy_mutexes(p_hxge_t hxgep)
941 {
942 	HXGE_DEBUG_MSG((hxgep, DDI_CTL, "==> hxge_destroy_mutexes"));
943 	RW_DESTROY(&hxgep->filter_lock);
944 	MUTEX_DESTROY(&hxgep->vmac_lock);
945 	MUTEX_DESTROY(&hxgep->ouraddr_lock);
946 	MUTEX_DESTROY(hxgep->genlock);
947 	MUTEX_DESTROY(&hxgep->pio_lock);
948 	MUTEX_DESTROY(&hxgep->timeout.lock);
949 
950 	if (hxge_debug_init == 1) {
951 		MUTEX_DESTROY(&hxgedebuglock);
952 		hxge_debug_init = 0;
953 	}
954 
955 	HXGE_DEBUG_MSG((hxgep, DDI_CTL, "<== hxge_destroy_mutexes"));
956 }
957 
958 hxge_status_t
959 hxge_init(p_hxge_t hxgep)
960 {
961 	hxge_status_t status = HXGE_OK;
962 
963 	HXGE_DEBUG_MSG((hxgep, STR_CTL, "==> hxge_init"));
964 
965 	if (hxgep->drv_state & STATE_HW_INITIALIZED) {
966 		return (status);
967 	}
968 
969 	/*
970 	 * Allocate system memory for the receive/transmit buffer blocks and
971 	 * receive/transmit descriptor rings.
972 	 */
973 	status = hxge_alloc_mem_pool(hxgep);
974 	if (status != HXGE_OK) {
975 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL, "alloc mem failed\n"));
976 		goto hxge_init_fail1;
977 	}
978 
979 	/*
980 	 * Initialize and enable TXDMA channels.
981 	 */
982 	status = hxge_init_txdma_channels(hxgep);
983 	if (status != HXGE_OK) {
984 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL, "init txdma failed\n"));
985 		goto hxge_init_fail3;
986 	}
987 
988 	/*
989 	 * Initialize and enable RXDMA channels.
990 	 */
991 	status = hxge_init_rxdma_channels(hxgep);
992 	if (status != HXGE_OK) {
993 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL, "init rxdma failed\n"));
994 		goto hxge_init_fail4;
995 	}
996 
997 	/*
998 	 * Initialize TCAM
999 	 */
1000 	status = hxge_classify_init(hxgep);
1001 	if (status != HXGE_OK) {
1002 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL, "init classify failed\n"));
1003 		goto hxge_init_fail5;
1004 	}
1005 
1006 	/*
1007 	 * Initialize the VMAC block.
1008 	 */
1009 	status = hxge_vmac_init(hxgep);
1010 	if (status != HXGE_OK) {
1011 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL, "init MAC failed\n"));
1012 		goto hxge_init_fail5;
1013 	}
1014 
1015 	/* Bringup - this may be unnecessary when PXE and FCODE available */
1016 	status = hxge_pfc_set_default_mac_addr(hxgep);
1017 	if (status != HXGE_OK) {
1018 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
1019 		    "Default Address Failure\n"));
1020 		goto hxge_init_fail5;
1021 	}
1022 
1023 	/*
1024 	 * Enable hardware interrupts.
1025 	 */
1026 	hxge_intr_hw_enable(hxgep);
1027 	hxgep->drv_state |= STATE_HW_INITIALIZED;
1028 
1029 	goto hxge_init_exit;
1030 
1031 hxge_init_fail5:
1032 	hxge_uninit_rxdma_channels(hxgep);
1033 hxge_init_fail4:
1034 	hxge_uninit_txdma_channels(hxgep);
1035 hxge_init_fail3:
1036 	hxge_free_mem_pool(hxgep);
1037 hxge_init_fail1:
1038 	HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
1039 	    "<== hxge_init status (failed) = 0x%08x", status));
1040 	return (status);
1041 
1042 hxge_init_exit:
1043 
1044 	HXGE_DEBUG_MSG((hxgep, DDI_CTL, "<== hxge_init status = 0x%08x",
1045 	    status));
1046 
1047 	return (status);
1048 }
1049 
1050 timeout_id_t
1051 hxge_start_timer(p_hxge_t hxgep, fptrv_t func, int msec)
1052 {
1053 	if ((hxgep->suspended == 0) || (hxgep->suspended == DDI_RESUME)) {
1054 		return (timeout(func, (caddr_t)hxgep,
1055 		    drv_usectohz(1000 * msec)));
1056 	}
1057 	return (NULL);
1058 }
1059 
1060 /*ARGSUSED*/
1061 void
1062 hxge_stop_timer(p_hxge_t hxgep, timeout_id_t timerid)
1063 {
1064 	if (timerid) {
1065 		(void) untimeout(timerid);
1066 	}
1067 }
1068 
1069 void
1070 hxge_uninit(p_hxge_t hxgep)
1071 {
1072 	HXGE_DEBUG_MSG((hxgep, DDI_CTL, "==> hxge_uninit"));
1073 
1074 	if (!(hxgep->drv_state & STATE_HW_INITIALIZED)) {
1075 		HXGE_DEBUG_MSG((hxgep, DDI_CTL,
1076 		    "==> hxge_uninit: not initialized"));
1077 		HXGE_DEBUG_MSG((hxgep, DDI_CTL, "<== hxge_uninit"));
1078 		return;
1079 	}
1080 
1081 	/* Stop timer */
1082 	if (hxgep->hxge_timerid) {
1083 		hxge_stop_timer(hxgep, hxgep->hxge_timerid);
1084 		hxgep->hxge_timerid = 0;
1085 	}
1086 
1087 	(void) hxge_intr_hw_disable(hxgep);
1088 
1089 	/* Reset the receive VMAC side.  */
1090 	(void) hxge_rx_vmac_disable(hxgep);
1091 
1092 	/* Free classification resources */
1093 	(void) hxge_classify_uninit(hxgep);
1094 
1095 	/* Reset the transmit/receive DMA side.  */
1096 	(void) hxge_txdma_hw_mode(hxgep, HXGE_DMA_STOP);
1097 	(void) hxge_rxdma_hw_mode(hxgep, HXGE_DMA_STOP);
1098 
1099 	hxge_uninit_txdma_channels(hxgep);
1100 	hxge_uninit_rxdma_channels(hxgep);
1101 
1102 	/* Reset the transmit VMAC side.  */
1103 	(void) hxge_tx_vmac_disable(hxgep);
1104 
1105 	hxge_free_mem_pool(hxgep);
1106 
1107 	hxgep->drv_state &= ~STATE_HW_INITIALIZED;
1108 
1109 	HXGE_DEBUG_MSG((hxgep, DDI_CTL, "<== hxge_uninit"));
1110 }
1111 
1112 /*ARGSUSED*/
1113 /*VARARGS*/
1114 void
1115 hxge_debug_msg(p_hxge_t hxgep, uint64_t level, char *fmt, ...)
1116 {
1117 	char		msg_buffer[1048];
1118 	char		prefix_buffer[32];
1119 	int		instance;
1120 	uint64_t	debug_level;
1121 	int		cmn_level = CE_CONT;
1122 	va_list		ap;
1123 
1124 	debug_level = (hxgep == NULL) ? hxge_debug_level :
1125 	    hxgep->hxge_debug_level;
1126 
1127 	if ((level & debug_level) || (level == HXGE_NOTE) ||
1128 	    (level == HXGE_ERR_CTL)) {
1129 		/* do the msg processing */
1130 		if (hxge_debug_init == 0) {
1131 			MUTEX_INIT(&hxgedebuglock, NULL, MUTEX_DRIVER, NULL);
1132 			hxge_debug_init = 1;
1133 		}
1134 
1135 		MUTEX_ENTER(&hxgedebuglock);
1136 
1137 		if ((level & HXGE_NOTE)) {
1138 			cmn_level = CE_NOTE;
1139 		}
1140 
1141 		if (level & HXGE_ERR_CTL) {
1142 			cmn_level = CE_WARN;
1143 		}
1144 
1145 		va_start(ap, fmt);
1146 		(void) vsprintf(msg_buffer, fmt, ap);
1147 		va_end(ap);
1148 
1149 		if (hxgep == NULL) {
1150 			instance = -1;
1151 			(void) sprintf(prefix_buffer, "%s :", "hxge");
1152 		} else {
1153 			instance = hxgep->instance;
1154 			(void) sprintf(prefix_buffer,
1155 			    "%s%d :", "hxge", instance);
1156 		}
1157 
1158 		MUTEX_EXIT(&hxgedebuglock);
1159 		cmn_err(cmn_level, "%s %s\n", prefix_buffer, msg_buffer);
1160 	}
1161 }
1162 
1163 char *
1164 hxge_dump_packet(char *addr, int size)
1165 {
1166 	uchar_t		*ap = (uchar_t *)addr;
1167 	int		i;
1168 	static char	etherbuf[1024];
1169 	char		*cp = etherbuf;
1170 	char		digits[] = "0123456789abcdef";
1171 
1172 	if (!size)
1173 		size = 60;
1174 
1175 	if (size > MAX_DUMP_SZ) {
1176 		/* Dump the leading bytes */
1177 		for (i = 0; i < MAX_DUMP_SZ / 2; i++) {
1178 			if (*ap > 0x0f)
1179 				*cp++ = digits[*ap >> 4];
1180 			*cp++ = digits[*ap++ & 0xf];
1181 			*cp++ = ':';
1182 		}
1183 		for (i = 0; i < 20; i++)
1184 			*cp++ = '.';
1185 		/* Dump the last MAX_DUMP_SZ/2 bytes */
1186 		ap = (uchar_t *)(addr + (size - MAX_DUMP_SZ / 2));
1187 		for (i = 0; i < MAX_DUMP_SZ / 2; i++) {
1188 			if (*ap > 0x0f)
1189 				*cp++ = digits[*ap >> 4];
1190 			*cp++ = digits[*ap++ & 0xf];
1191 			*cp++ = ':';
1192 		}
1193 	} else {
1194 		for (i = 0; i < size; i++) {
1195 			if (*ap > 0x0f)
1196 				*cp++ = digits[*ap >> 4];
1197 			*cp++ = digits[*ap++ & 0xf];
1198 			*cp++ = ':';
1199 		}
1200 	}
1201 	*--cp = 0;
1202 	return (etherbuf);
1203 }
1204 
1205 static void
1206 hxge_suspend(p_hxge_t hxgep)
1207 {
1208 	HXGE_DEBUG_MSG((hxgep, DDI_CTL, "==> hxge_suspend"));
1209 
1210 	/*
1211 	 * Stop the link status timer before hxge_intrs_disable() to avoid
1212 	 * accessing the the MSIX table simultaneously. Note that the timer
1213 	 * routine polls for MSIX parity errors.
1214 	 */
1215 	MUTEX_ENTER(&hxgep->timeout.lock);
1216 	if (hxgep->timeout.id)
1217 		(void) untimeout(hxgep->timeout.id);
1218 	MUTEX_EXIT(&hxgep->timeout.lock);
1219 
1220 	hxge_intrs_disable(hxgep);
1221 	hxge_destroy_dev(hxgep);
1222 
1223 	HXGE_DEBUG_MSG((hxgep, DDI_CTL, "<== hxge_suspend"));
1224 }
1225 
1226 static hxge_status_t
1227 hxge_resume(p_hxge_t hxgep)
1228 {
1229 	hxge_status_t status = HXGE_OK;
1230 
1231 	HXGE_DEBUG_MSG((hxgep, DDI_CTL, "==> hxge_resume"));
1232 	hxgep->suspended = DDI_RESUME;
1233 
1234 	(void) hxge_rxdma_hw_mode(hxgep, HXGE_DMA_START);
1235 	(void) hxge_txdma_hw_mode(hxgep, HXGE_DMA_START);
1236 
1237 	(void) hxge_rx_vmac_enable(hxgep);
1238 	(void) hxge_tx_vmac_enable(hxgep);
1239 
1240 	hxge_intrs_enable(hxgep);
1241 
1242 	hxgep->suspended = 0;
1243 
1244 	/*
1245 	 * Resume the link status timer after hxge_intrs_enable to avoid
1246 	 * accessing MSIX table simultaneously.
1247 	 */
1248 	MUTEX_ENTER(&hxgep->timeout.lock);
1249 	hxgep->timeout.id = timeout(hxge_link_poll, (void *)hxgep,
1250 	    hxgep->timeout.ticks);
1251 	MUTEX_EXIT(&hxgep->timeout.lock);
1252 
1253 	HXGE_DEBUG_MSG((hxgep, DDI_CTL,
1254 	    "<== hxge_resume status = 0x%x", status));
1255 
1256 	return (status);
1257 }
1258 
1259 static hxge_status_t
1260 hxge_setup_dev(p_hxge_t hxgep)
1261 {
1262 	hxge_status_t status = HXGE_OK;
1263 
1264 	HXGE_DEBUG_MSG((hxgep, DDI_CTL, "==> hxge_setup_dev"));
1265 
1266 	status = hxge_link_init(hxgep);
1267 	if (fm_check_acc_handle(hxgep->dev_regs->hxge_regh) != DDI_FM_OK) {
1268 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
1269 		    "Bad register acc handle"));
1270 		status = HXGE_ERROR;
1271 	}
1272 
1273 	if (status != HXGE_OK) {
1274 		HXGE_DEBUG_MSG((hxgep, MAC_CTL,
1275 		    " hxge_setup_dev status (link init 0x%08x)", status));
1276 		goto hxge_setup_dev_exit;
1277 	}
1278 
1279 hxge_setup_dev_exit:
1280 	HXGE_DEBUG_MSG((hxgep, DDI_CTL,
1281 	    "<== hxge_setup_dev status = 0x%08x", status));
1282 
1283 	return (status);
1284 }
1285 
1286 static void
1287 hxge_destroy_dev(p_hxge_t hxgep)
1288 {
1289 	HXGE_DEBUG_MSG((hxgep, DDI_CTL, "==> hxge_destroy_dev"));
1290 
1291 	(void) hxge_hw_stop(hxgep);
1292 
1293 	HXGE_DEBUG_MSG((hxgep, DDI_CTL, "<== hxge_destroy_dev"));
1294 }
1295 
1296 static hxge_status_t
1297 hxge_setup_system_dma_pages(p_hxge_t hxgep)
1298 {
1299 	int			ddi_status = DDI_SUCCESS;
1300 	uint_t			count;
1301 	ddi_dma_cookie_t	cookie;
1302 	uint_t			iommu_pagesize;
1303 	hxge_status_t		status = HXGE_OK;
1304 
1305 	HXGE_DEBUG_MSG((hxgep, DDI_CTL, "==> hxge_setup_system_dma_pages"));
1306 
1307 	hxgep->sys_page_sz = ddi_ptob(hxgep->dip, (ulong_t)1);
1308 	iommu_pagesize = dvma_pagesize(hxgep->dip);
1309 
1310 	HXGE_DEBUG_MSG((hxgep, DDI_CTL,
1311 	    " hxge_setup_system_dma_pages: page %d (ddi_ptob %d) "
1312 	    " default_block_size %d iommu_pagesize %d",
1313 	    hxgep->sys_page_sz, ddi_ptob(hxgep->dip, (ulong_t)1),
1314 	    hxgep->rx_default_block_size, iommu_pagesize));
1315 
1316 	if (iommu_pagesize != 0) {
1317 		if (hxgep->sys_page_sz == iommu_pagesize) {
1318 			/* Hydra support up to 8K pages */
1319 			if (iommu_pagesize > 0x2000)
1320 				hxgep->sys_page_sz = 0x2000;
1321 		} else {
1322 			if (hxgep->sys_page_sz > iommu_pagesize)
1323 				hxgep->sys_page_sz = iommu_pagesize;
1324 		}
1325 	}
1326 
1327 	hxgep->sys_page_mask = ~(hxgep->sys_page_sz - 1);
1328 
1329 	HXGE_DEBUG_MSG((hxgep, DDI_CTL,
1330 	    "==> hxge_setup_system_dma_pages: page %d (ddi_ptob %d) "
1331 	    "default_block_size %d page mask %d",
1332 	    hxgep->sys_page_sz, ddi_ptob(hxgep->dip, (ulong_t)1),
1333 	    hxgep->rx_default_block_size, hxgep->sys_page_mask));
1334 
1335 	switch (hxgep->sys_page_sz) {
1336 	default:
1337 		hxgep->sys_page_sz = 0x1000;
1338 		hxgep->sys_page_mask = ~(hxgep->sys_page_sz - 1);
1339 		hxgep->rx_default_block_size = 0x1000;
1340 		hxgep->rx_bksize_code = RBR_BKSIZE_4K;
1341 		break;
1342 	case 0x1000:
1343 		hxgep->rx_default_block_size = 0x1000;
1344 		hxgep->rx_bksize_code = RBR_BKSIZE_4K;
1345 		break;
1346 	case 0x2000:
1347 		hxgep->rx_default_block_size = 0x2000;
1348 		hxgep->rx_bksize_code = RBR_BKSIZE_8K;
1349 		break;
1350 	}
1351 
1352 	hxge_rx_dma_attr.dma_attr_align = hxgep->sys_page_sz;
1353 	hxge_tx_dma_attr.dma_attr_align = hxgep->sys_page_sz;
1354 
1355 	/*
1356 	 * Get the system DMA burst size.
1357 	 */
1358 	ddi_status = ddi_dma_alloc_handle(hxgep->dip, &hxge_tx_dma_attr,
1359 	    DDI_DMA_DONTWAIT, 0, &hxgep->dmasparehandle);
1360 	if (ddi_status != DDI_SUCCESS) {
1361 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
1362 		    "ddi_dma_alloc_handle: failed status 0x%x", ddi_status));
1363 		goto hxge_get_soft_properties_exit;
1364 	}
1365 
1366 	ddi_status = ddi_dma_addr_bind_handle(hxgep->dmasparehandle, NULL,
1367 	    (caddr_t)hxgep->dmasparehandle, sizeof (hxgep->dmasparehandle),
1368 	    DDI_DMA_RDWR | DDI_DMA_CONSISTENT, DDI_DMA_DONTWAIT, 0,
1369 	    &cookie, &count);
1370 	if (ddi_status != DDI_DMA_MAPPED) {
1371 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
1372 		    "Binding spare handle to find system burstsize failed."));
1373 		ddi_status = DDI_FAILURE;
1374 		goto hxge_get_soft_properties_fail1;
1375 	}
1376 
1377 	hxgep->sys_burst_sz = ddi_dma_burstsizes(hxgep->dmasparehandle);
1378 	(void) ddi_dma_unbind_handle(hxgep->dmasparehandle);
1379 
1380 hxge_get_soft_properties_fail1:
1381 	ddi_dma_free_handle(&hxgep->dmasparehandle);
1382 
1383 hxge_get_soft_properties_exit:
1384 
1385 	if (ddi_status != DDI_SUCCESS)
1386 		status |= (HXGE_ERROR | HXGE_DDI_FAILED);
1387 
1388 	HXGE_DEBUG_MSG((hxgep, DDI_CTL,
1389 	    "<== hxge_setup_system_dma_pages status = 0x%08x", status));
1390 
1391 	return (status);
1392 }
1393 
1394 static hxge_status_t
1395 hxge_alloc_mem_pool(p_hxge_t hxgep)
1396 {
1397 	hxge_status_t status = HXGE_OK;
1398 
1399 	HXGE_DEBUG_MSG((hxgep, DDI_CTL, "==> hxge_alloc_mem_pool"));
1400 
1401 	status = hxge_alloc_rx_mem_pool(hxgep);
1402 	if (status != HXGE_OK) {
1403 		return (HXGE_ERROR);
1404 	}
1405 
1406 	status = hxge_alloc_tx_mem_pool(hxgep);
1407 	if (status != HXGE_OK) {
1408 		hxge_free_rx_mem_pool(hxgep);
1409 		return (HXGE_ERROR);
1410 	}
1411 
1412 	HXGE_DEBUG_MSG((hxgep, DDI_CTL, "<== hxge_alloc_mem_pool"));
1413 	return (HXGE_OK);
1414 }
1415 
1416 static void
1417 hxge_free_mem_pool(p_hxge_t hxgep)
1418 {
1419 	HXGE_DEBUG_MSG((hxgep, MEM_CTL, "==> hxge_free_mem_pool"));
1420 
1421 	hxge_free_rx_mem_pool(hxgep);
1422 	hxge_free_tx_mem_pool(hxgep);
1423 
1424 	HXGE_DEBUG_MSG((hxgep, MEM_CTL, "<== hxge_free_mem_pool"));
1425 }
1426 
1427 static hxge_status_t
1428 hxge_alloc_rx_mem_pool(p_hxge_t hxgep)
1429 {
1430 	int			i, j;
1431 	uint32_t		ndmas, st_rdc;
1432 	p_hxge_dma_pt_cfg_t	p_all_cfgp;
1433 	p_hxge_hw_pt_cfg_t	p_cfgp;
1434 	p_hxge_dma_pool_t	dma_poolp;
1435 	p_hxge_dma_common_t	*dma_buf_p;
1436 	p_hxge_dma_pool_t	dma_rbr_cntl_poolp;
1437 	p_hxge_dma_common_t	*dma_rbr_cntl_p;
1438 	p_hxge_dma_pool_t	dma_rcr_cntl_poolp;
1439 	p_hxge_dma_common_t	*dma_rcr_cntl_p;
1440 	p_hxge_dma_pool_t	dma_mbox_cntl_poolp;
1441 	p_hxge_dma_common_t	*dma_mbox_cntl_p;
1442 	size_t			rx_buf_alloc_size;
1443 	size_t			rx_rbr_cntl_alloc_size;
1444 	size_t			rx_rcr_cntl_alloc_size;
1445 	size_t			rx_mbox_cntl_alloc_size;
1446 	uint32_t		*num_chunks;	/* per dma */
1447 	hxge_status_t		status = HXGE_OK;
1448 
1449 	uint32_t		hxge_port_rbr_size;
1450 	uint32_t		hxge_port_rbr_spare_size;
1451 	uint32_t		hxge_port_rcr_size;
1452 
1453 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "==> hxge_alloc_rx_mem_pool"));
1454 
1455 	p_all_cfgp = (p_hxge_dma_pt_cfg_t)&hxgep->pt_config;
1456 	p_cfgp = (p_hxge_hw_pt_cfg_t)&p_all_cfgp->hw_config;
1457 	st_rdc = p_cfgp->start_rdc;
1458 	ndmas = p_cfgp->max_rdcs;
1459 
1460 	HXGE_DEBUG_MSG((hxgep, DMA_CTL,
1461 	    " hxge_alloc_rx_mem_pool st_rdc %d ndmas %d", st_rdc, ndmas));
1462 
1463 	/*
1464 	 * Allocate memory for each receive DMA channel.
1465 	 */
1466 	dma_poolp = (p_hxge_dma_pool_t)KMEM_ZALLOC(sizeof (hxge_dma_pool_t),
1467 	    KM_SLEEP);
1468 	dma_buf_p = (p_hxge_dma_common_t *)KMEM_ZALLOC(
1469 	    sizeof (p_hxge_dma_common_t) * ndmas, KM_SLEEP);
1470 
1471 	dma_rbr_cntl_poolp = (p_hxge_dma_pool_t)
1472 	    KMEM_ZALLOC(sizeof (hxge_dma_pool_t), KM_SLEEP);
1473 	dma_rbr_cntl_p = (p_hxge_dma_common_t *)KMEM_ZALLOC(
1474 	    sizeof (p_hxge_dma_common_t) * ndmas, KM_SLEEP);
1475 	dma_rcr_cntl_poolp = (p_hxge_dma_pool_t)
1476 	    KMEM_ZALLOC(sizeof (hxge_dma_pool_t), KM_SLEEP);
1477 	dma_rcr_cntl_p = (p_hxge_dma_common_t *)KMEM_ZALLOC(
1478 	    sizeof (p_hxge_dma_common_t) * ndmas, KM_SLEEP);
1479 	dma_mbox_cntl_poolp = (p_hxge_dma_pool_t)
1480 	    KMEM_ZALLOC(sizeof (hxge_dma_pool_t), KM_SLEEP);
1481 	dma_mbox_cntl_p = (p_hxge_dma_common_t *)KMEM_ZALLOC(
1482 	    sizeof (p_hxge_dma_common_t) * ndmas, KM_SLEEP);
1483 
1484 	num_chunks = (uint32_t *)KMEM_ZALLOC(sizeof (uint32_t) * ndmas,
1485 	    KM_SLEEP);
1486 
1487 	/*
1488 	 * Assume that each DMA channel will be configured with default block
1489 	 * size. rbr block counts are mod of batch count (16).
1490 	 */
1491 	hxge_port_rbr_size = p_all_cfgp->rbr_size;
1492 	hxge_port_rcr_size = p_all_cfgp->rcr_size;
1493 
1494 	if (!hxge_port_rbr_size) {
1495 		hxge_port_rbr_size = HXGE_RBR_RBB_DEFAULT;
1496 	}
1497 
1498 	if (hxge_port_rbr_size % HXGE_RXDMA_POST_BATCH) {
1499 		hxge_port_rbr_size = (HXGE_RXDMA_POST_BATCH *
1500 		    (hxge_port_rbr_size / HXGE_RXDMA_POST_BATCH + 1));
1501 	}
1502 
1503 	p_all_cfgp->rbr_size = hxge_port_rbr_size;
1504 	hxge_port_rbr_spare_size = hxge_rbr_spare_size;
1505 
1506 	if (hxge_port_rbr_spare_size % HXGE_RXDMA_POST_BATCH) {
1507 		hxge_port_rbr_spare_size = (HXGE_RXDMA_POST_BATCH *
1508 		    (hxge_port_rbr_spare_size / HXGE_RXDMA_POST_BATCH + 1));
1509 	}
1510 
1511 	rx_buf_alloc_size = (hxgep->rx_default_block_size *
1512 	    (hxge_port_rbr_size + hxge_port_rbr_spare_size));
1513 
1514 	/*
1515 	 * Addresses of receive block ring, receive completion ring and the
1516 	 * mailbox must be all cache-aligned (64 bytes).
1517 	 */
1518 	rx_rbr_cntl_alloc_size = hxge_port_rbr_size + hxge_port_rbr_spare_size;
1519 	rx_rbr_cntl_alloc_size *= sizeof (rx_desc_t);
1520 	rx_rcr_cntl_alloc_size = sizeof (rcr_entry_t) * hxge_port_rcr_size;
1521 	rx_mbox_cntl_alloc_size = sizeof (rxdma_mailbox_t);
1522 
1523 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_alloc_rx_mem_pool: "
1524 	    "hxge_port_rbr_size = %d hxge_port_rbr_spare_size = %d "
1525 	    "hxge_port_rcr_size = %d rx_cntl_alloc_size = %d",
1526 	    hxge_port_rbr_size, hxge_port_rbr_spare_size,
1527 	    hxge_port_rcr_size, rx_cntl_alloc_size));
1528 
1529 	hxgep->hxge_port_rbr_size = hxge_port_rbr_size;
1530 	hxgep->hxge_port_rcr_size = hxge_port_rcr_size;
1531 
1532 	/*
1533 	 * Allocate memory for receive buffers and descriptor rings. Replace
1534 	 * allocation functions with interface functions provided by the
1535 	 * partition manager when it is available.
1536 	 */
1537 	/*
1538 	 * Allocate memory for the receive buffer blocks.
1539 	 */
1540 	for (i = 0; i < ndmas; i++) {
1541 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
1542 		    " hxge_alloc_rx_mem_pool to alloc mem: "
1543 		    " dma %d dma_buf_p %llx &dma_buf_p %llx",
1544 		    i, dma_buf_p[i], &dma_buf_p[i]));
1545 
1546 		num_chunks[i] = 0;
1547 
1548 		status = hxge_alloc_rx_buf_dma(hxgep, st_rdc, &dma_buf_p[i],
1549 		    rx_buf_alloc_size, hxgep->rx_default_block_size,
1550 		    &num_chunks[i]);
1551 		if (status != HXGE_OK) {
1552 			break;
1553 		}
1554 
1555 		st_rdc++;
1556 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
1557 		    " hxge_alloc_rx_mem_pool DONE  alloc mem: "
1558 		    "dma %d dma_buf_p %llx &dma_buf_p %llx", i,
1559 		    dma_buf_p[i], &dma_buf_p[i]));
1560 	}
1561 
1562 	if (i < ndmas) {
1563 		goto hxge_alloc_rx_mem_fail1;
1564 	}
1565 
1566 	/*
1567 	 * Allocate memory for descriptor rings and mailbox.
1568 	 */
1569 	st_rdc = p_cfgp->start_rdc;
1570 	for (j = 0; j < ndmas; j++) {
1571 		if ((status = hxge_alloc_rx_cntl_dma(hxgep, st_rdc,
1572 		    &dma_rbr_cntl_p[j], &hxge_rx_rbr_desc_dma_attr,
1573 		    rx_rbr_cntl_alloc_size)) != HXGE_OK) {
1574 			break;
1575 		}
1576 
1577 		if ((status = hxge_alloc_rx_cntl_dma(hxgep, st_rdc,
1578 		    &dma_rcr_cntl_p[j], &hxge_rx_rcr_desc_dma_attr,
1579 		    rx_rcr_cntl_alloc_size)) != HXGE_OK) {
1580 			break;
1581 		}
1582 
1583 		if ((status = hxge_alloc_rx_cntl_dma(hxgep, st_rdc,
1584 		    &dma_mbox_cntl_p[j], &hxge_rx_mbox_dma_attr,
1585 		    rx_mbox_cntl_alloc_size)) != HXGE_OK) {
1586 			break;
1587 		}
1588 		st_rdc++;
1589 	}
1590 
1591 	if (j < ndmas) {
1592 		goto hxge_alloc_rx_mem_fail2;
1593 	}
1594 
1595 	dma_poolp->ndmas = ndmas;
1596 	dma_poolp->num_chunks = num_chunks;
1597 	dma_poolp->buf_allocated = B_TRUE;
1598 	hxgep->rx_buf_pool_p = dma_poolp;
1599 	dma_poolp->dma_buf_pool_p = dma_buf_p;
1600 
1601 	dma_rbr_cntl_poolp->ndmas = ndmas;
1602 	dma_rbr_cntl_poolp->buf_allocated = B_TRUE;
1603 	hxgep->rx_rbr_cntl_pool_p = dma_rbr_cntl_poolp;
1604 	dma_rbr_cntl_poolp->dma_buf_pool_p = dma_rbr_cntl_p;
1605 
1606 	dma_rcr_cntl_poolp->ndmas = ndmas;
1607 	dma_rcr_cntl_poolp->buf_allocated = B_TRUE;
1608 	hxgep->rx_rcr_cntl_pool_p = dma_rcr_cntl_poolp;
1609 	dma_rcr_cntl_poolp->dma_buf_pool_p = dma_rcr_cntl_p;
1610 
1611 	dma_mbox_cntl_poolp->ndmas = ndmas;
1612 	dma_mbox_cntl_poolp->buf_allocated = B_TRUE;
1613 	hxgep->rx_mbox_cntl_pool_p = dma_mbox_cntl_poolp;
1614 	dma_mbox_cntl_poolp->dma_buf_pool_p = dma_mbox_cntl_p;
1615 
1616 	goto hxge_alloc_rx_mem_pool_exit;
1617 
1618 hxge_alloc_rx_mem_fail2:
1619 	/* Free control buffers */
1620 	HXGE_DEBUG_MSG((hxgep, DMA_CTL,
1621 	    "==> hxge_alloc_rx_mem_pool: freeing control bufs (%d)", j));
1622 	for (; j >= 0; j--) {
1623 		hxge_free_rx_cntl_dma(hxgep,
1624 		    (p_hxge_dma_common_t)dma_rbr_cntl_p[j]);
1625 		hxge_free_rx_cntl_dma(hxgep,
1626 		    (p_hxge_dma_common_t)dma_rcr_cntl_p[j]);
1627 		hxge_free_rx_cntl_dma(hxgep,
1628 		    (p_hxge_dma_common_t)dma_mbox_cntl_p[j]);
1629 		HXGE_DEBUG_MSG((hxgep, DMA_CTL,
1630 		    "==> hxge_alloc_rx_mem_pool: control bufs freed (%d)", j));
1631 	}
1632 	HXGE_DEBUG_MSG((hxgep, DMA_CTL,
1633 	    "==> hxge_alloc_rx_mem_pool: control bufs freed (%d)", j));
1634 
1635 hxge_alloc_rx_mem_fail1:
1636 	/* Free data buffers */
1637 	i--;
1638 	HXGE_DEBUG_MSG((hxgep, DMA_CTL,
1639 	    "==> hxge_alloc_rx_mem_pool: freeing data bufs (%d)", i));
1640 	for (; i >= 0; i--) {
1641 		hxge_free_rx_buf_dma(hxgep, (p_hxge_dma_common_t)dma_buf_p[i],
1642 		    num_chunks[i]);
1643 	}
1644 	HXGE_DEBUG_MSG((hxgep, DMA_CTL,
1645 	    "==> hxge_alloc_rx_mem_pool: data bufs freed (%d)", i));
1646 
1647 	KMEM_FREE(num_chunks, sizeof (uint32_t) * ndmas);
1648 	KMEM_FREE(dma_poolp, sizeof (hxge_dma_pool_t));
1649 	KMEM_FREE(dma_buf_p, ndmas * sizeof (p_hxge_dma_common_t));
1650 	KMEM_FREE(dma_rbr_cntl_poolp, sizeof (hxge_dma_pool_t));
1651 	KMEM_FREE(dma_rbr_cntl_p, ndmas * sizeof (p_hxge_dma_common_t));
1652 	KMEM_FREE(dma_rcr_cntl_poolp, sizeof (hxge_dma_pool_t));
1653 	KMEM_FREE(dma_rcr_cntl_p, ndmas * sizeof (p_hxge_dma_common_t));
1654 	KMEM_FREE(dma_mbox_cntl_poolp, sizeof (hxge_dma_pool_t));
1655 	KMEM_FREE(dma_mbox_cntl_p, ndmas * sizeof (p_hxge_dma_common_t));
1656 
1657 hxge_alloc_rx_mem_pool_exit:
1658 	HXGE_DEBUG_MSG((hxgep, DMA_CTL,
1659 	    "<== hxge_alloc_rx_mem_pool:status 0x%08x", status));
1660 
1661 	return (status);
1662 }
1663 
1664 static void
1665 hxge_free_rx_mem_pool(p_hxge_t hxgep)
1666 {
1667 	uint32_t		i, ndmas;
1668 	p_hxge_dma_pool_t	dma_poolp;
1669 	p_hxge_dma_common_t	*dma_buf_p;
1670 	p_hxge_dma_pool_t	dma_rbr_cntl_poolp;
1671 	p_hxge_dma_common_t	*dma_rbr_cntl_p;
1672 	p_hxge_dma_pool_t	dma_rcr_cntl_poolp;
1673 	p_hxge_dma_common_t	*dma_rcr_cntl_p;
1674 	p_hxge_dma_pool_t	dma_mbox_cntl_poolp;
1675 	p_hxge_dma_common_t	*dma_mbox_cntl_p;
1676 	uint32_t		*num_chunks;
1677 
1678 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "==> hxge_free_rx_mem_pool"));
1679 
1680 	dma_poolp = hxgep->rx_buf_pool_p;
1681 	if (dma_poolp == NULL || (!dma_poolp->buf_allocated)) {
1682 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "<== hxge_free_rx_mem_pool "
1683 		    "(null rx buf pool or buf not allocated"));
1684 		return;
1685 	}
1686 
1687 	dma_rbr_cntl_poolp = hxgep->rx_rbr_cntl_pool_p;
1688 	if (dma_rbr_cntl_poolp == NULL ||
1689 	    (!dma_rbr_cntl_poolp->buf_allocated)) {
1690 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
1691 		    "<== hxge_free_rx_mem_pool "
1692 		    "(null rbr cntl buf pool or rbr cntl buf not allocated"));
1693 		return;
1694 	}
1695 
1696 	dma_rcr_cntl_poolp = hxgep->rx_rcr_cntl_pool_p;
1697 	if (dma_rcr_cntl_poolp == NULL ||
1698 	    (!dma_rcr_cntl_poolp->buf_allocated)) {
1699 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
1700 		    "<== hxge_free_rx_mem_pool "
1701 		    "(null rcr cntl buf pool or rcr cntl buf not allocated"));
1702 		return;
1703 	}
1704 
1705 	dma_mbox_cntl_poolp = hxgep->rx_mbox_cntl_pool_p;
1706 	if (dma_mbox_cntl_poolp == NULL ||
1707 	    (!dma_mbox_cntl_poolp->buf_allocated)) {
1708 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
1709 		    "<== hxge_free_rx_mem_pool "
1710 		    "(null mbox cntl buf pool or mbox cntl buf not allocated"));
1711 		return;
1712 	}
1713 
1714 	dma_buf_p = dma_poolp->dma_buf_pool_p;
1715 	num_chunks = dma_poolp->num_chunks;
1716 
1717 	dma_rbr_cntl_p = dma_rbr_cntl_poolp->dma_buf_pool_p;
1718 	dma_rcr_cntl_p = dma_rcr_cntl_poolp->dma_buf_pool_p;
1719 	dma_mbox_cntl_p = dma_mbox_cntl_poolp->dma_buf_pool_p;
1720 	ndmas = dma_rbr_cntl_poolp->ndmas;
1721 
1722 	for (i = 0; i < ndmas; i++) {
1723 		hxge_free_rx_buf_dma(hxgep, dma_buf_p[i], num_chunks[i]);
1724 	}
1725 
1726 	for (i = 0; i < ndmas; i++) {
1727 		hxge_free_rx_cntl_dma(hxgep, dma_rbr_cntl_p[i]);
1728 		hxge_free_rx_cntl_dma(hxgep, dma_rcr_cntl_p[i]);
1729 		hxge_free_rx_cntl_dma(hxgep, dma_mbox_cntl_p[i]);
1730 	}
1731 
1732 	for (i = 0; i < ndmas; i++) {
1733 		KMEM_FREE(dma_buf_p[i],
1734 		    sizeof (hxge_dma_common_t) * HXGE_DMA_BLOCK);
1735 		KMEM_FREE(dma_rbr_cntl_p[i], sizeof (hxge_dma_common_t));
1736 		KMEM_FREE(dma_rcr_cntl_p[i], sizeof (hxge_dma_common_t));
1737 		KMEM_FREE(dma_mbox_cntl_p[i], sizeof (hxge_dma_common_t));
1738 	}
1739 
1740 	KMEM_FREE(num_chunks, sizeof (uint32_t) * ndmas);
1741 	KMEM_FREE(dma_rbr_cntl_p, ndmas * sizeof (p_hxge_dma_common_t));
1742 	KMEM_FREE(dma_rbr_cntl_poolp, sizeof (hxge_dma_pool_t));
1743 	KMEM_FREE(dma_rcr_cntl_p, ndmas * sizeof (p_hxge_dma_common_t));
1744 	KMEM_FREE(dma_rcr_cntl_poolp, sizeof (hxge_dma_pool_t));
1745 	KMEM_FREE(dma_mbox_cntl_p, ndmas * sizeof (p_hxge_dma_common_t));
1746 	KMEM_FREE(dma_mbox_cntl_poolp, sizeof (hxge_dma_pool_t));
1747 	KMEM_FREE(dma_buf_p, ndmas * sizeof (p_hxge_dma_common_t));
1748 	KMEM_FREE(dma_poolp, sizeof (hxge_dma_pool_t));
1749 
1750 	hxgep->rx_buf_pool_p = NULL;
1751 	hxgep->rx_rbr_cntl_pool_p = NULL;
1752 	hxgep->rx_rcr_cntl_pool_p = NULL;
1753 	hxgep->rx_mbox_cntl_pool_p = NULL;
1754 
1755 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "<== hxge_free_rx_mem_pool"));
1756 }
1757 
1758 static hxge_status_t
1759 hxge_alloc_rx_buf_dma(p_hxge_t hxgep, uint16_t dma_channel,
1760     p_hxge_dma_common_t *dmap,
1761     size_t alloc_size, size_t block_size, uint32_t *num_chunks)
1762 {
1763 	p_hxge_dma_common_t	rx_dmap;
1764 	hxge_status_t		status = HXGE_OK;
1765 	size_t			total_alloc_size;
1766 	size_t			allocated = 0;
1767 	int			i, size_index, array_size;
1768 
1769 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "==> hxge_alloc_rx_buf_dma"));
1770 
1771 	rx_dmap = (p_hxge_dma_common_t)
1772 	    KMEM_ZALLOC(sizeof (hxge_dma_common_t) * HXGE_DMA_BLOCK, KM_SLEEP);
1773 
1774 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
1775 	    " alloc_rx_buf_dma rdc %d asize %x bsize %x bbuf %llx ",
1776 	    dma_channel, alloc_size, block_size, dmap));
1777 
1778 	total_alloc_size = alloc_size;
1779 
1780 	i = 0;
1781 	size_index = 0;
1782 	array_size = sizeof (alloc_sizes) / sizeof (size_t);
1783 	while ((size_index < array_size) &&
1784 	    (alloc_sizes[size_index] < alloc_size))
1785 		size_index++;
1786 	if (size_index >= array_size) {
1787 		size_index = array_size - 1;
1788 	}
1789 
1790 	while ((allocated < total_alloc_size) &&
1791 	    (size_index >= 0) && (i < HXGE_DMA_BLOCK)) {
1792 		rx_dmap[i].dma_chunk_index = i;
1793 		rx_dmap[i].block_size = block_size;
1794 		rx_dmap[i].alength = alloc_sizes[size_index];
1795 		rx_dmap[i].orig_alength = rx_dmap[i].alength;
1796 		rx_dmap[i].nblocks = alloc_sizes[size_index] / block_size;
1797 		rx_dmap[i].dma_channel = dma_channel;
1798 		rx_dmap[i].contig_alloc_type = B_FALSE;
1799 
1800 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
1801 		    "alloc_rx_buf_dma rdc %d chunk %d bufp %llx size %x "
1802 		    "i %d nblocks %d alength %d",
1803 		    dma_channel, i, &rx_dmap[i], block_size,
1804 		    i, rx_dmap[i].nblocks, rx_dmap[i].alength));
1805 		status = hxge_dma_mem_alloc(hxgep, hxge_force_dma,
1806 		    &hxge_rx_dma_attr, rx_dmap[i].alength,
1807 		    &hxge_dev_buf_dma_acc_attr,
1808 		    DDI_DMA_READ | DDI_DMA_STREAMING,
1809 		    (p_hxge_dma_common_t)(&rx_dmap[i]));
1810 		if (status != HXGE_OK) {
1811 			HXGE_DEBUG_MSG((hxgep, DMA_CTL,
1812 			    " hxge_alloc_rx_buf_dma: Alloc Failed: "
1813 			    " for size: %d", alloc_sizes[size_index]));
1814 			size_index--;
1815 		} else {
1816 			HXGE_DEBUG_MSG((hxgep, DMA_CTL,
1817 			    " alloc_rx_buf_dma allocated rdc %d "
1818 			    "chunk %d size %x dvma %x bufp %llx ",
1819 			    dma_channel, i, rx_dmap[i].alength,
1820 			    rx_dmap[i].ioaddr_pp, &rx_dmap[i]));
1821 			i++;
1822 			allocated += alloc_sizes[size_index];
1823 		}
1824 	}
1825 
1826 	if (allocated < total_alloc_size) {
1827 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
1828 		    " hxge_alloc_rx_buf_dma failed due to"
1829 		    " allocated(%d) < required(%d)",
1830 		    allocated, total_alloc_size));
1831 		goto hxge_alloc_rx_mem_fail1;
1832 	}
1833 
1834 	HXGE_DEBUG_MSG((hxgep, DMA_CTL,
1835 	    " alloc_rx_buf_dma rdc %d allocated %d chunks", dma_channel, i));
1836 
1837 	*num_chunks = i;
1838 	*dmap = rx_dmap;
1839 
1840 	goto hxge_alloc_rx_mem_exit;
1841 
1842 hxge_alloc_rx_mem_fail1:
1843 	KMEM_FREE(rx_dmap, sizeof (hxge_dma_common_t) * HXGE_DMA_BLOCK);
1844 
1845 hxge_alloc_rx_mem_exit:
1846 	HXGE_DEBUG_MSG((hxgep, DMA_CTL,
1847 	    "<== hxge_alloc_rx_buf_dma status 0x%08x", status));
1848 
1849 	return (status);
1850 }
1851 
1852 /*ARGSUSED*/
1853 static void
1854 hxge_free_rx_buf_dma(p_hxge_t hxgep, p_hxge_dma_common_t dmap,
1855     uint32_t num_chunks)
1856 {
1857 	int i;
1858 
1859 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
1860 	    "==> hxge_free_rx_buf_dma: # of chunks %d", num_chunks));
1861 
1862 	for (i = 0; i < num_chunks; i++) {
1863 		HXGE_DEBUG_MSG((hxgep, MEM2_CTL,
1864 		    "==> hxge_free_rx_buf_dma: chunk %d dmap 0x%llx", i, dmap));
1865 		hxge_dma_mem_free(dmap++);
1866 	}
1867 
1868 	HXGE_DEBUG_MSG((hxgep, MEM2_CTL, "<== hxge_free_rx_buf_dma"));
1869 }
1870 
1871 /*ARGSUSED*/
1872 static hxge_status_t
1873 hxge_alloc_rx_cntl_dma(p_hxge_t hxgep, uint16_t dma_channel,
1874     p_hxge_dma_common_t *dmap, struct ddi_dma_attr *attr, size_t size)
1875 {
1876 	p_hxge_dma_common_t	rx_dmap;
1877 	hxge_status_t		status = HXGE_OK;
1878 
1879 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "==> hxge_alloc_rx_cntl_dma"));
1880 
1881 	rx_dmap = (p_hxge_dma_common_t)
1882 	    KMEM_ZALLOC(sizeof (hxge_dma_common_t), KM_SLEEP);
1883 
1884 	rx_dmap->contig_alloc_type = B_FALSE;
1885 
1886 	status = hxge_dma_mem_alloc(hxgep, hxge_force_dma,
1887 	    attr, size, &hxge_dev_desc_dma_acc_attr,
1888 	    DDI_DMA_RDWR | DDI_DMA_CONSISTENT, rx_dmap);
1889 	if (status != HXGE_OK) {
1890 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
1891 		    " hxge_alloc_rx_cntl_dma: Alloc Failed: "
1892 		    " for size: %d", size));
1893 		goto hxge_alloc_rx_cntl_dma_fail1;
1894 	}
1895 
1896 	*dmap = rx_dmap;
1897 
1898 	goto hxge_alloc_rx_cntl_dma_exit;
1899 
1900 hxge_alloc_rx_cntl_dma_fail1:
1901 	KMEM_FREE(rx_dmap, sizeof (hxge_dma_common_t));
1902 
1903 hxge_alloc_rx_cntl_dma_exit:
1904 	HXGE_DEBUG_MSG((hxgep, DMA_CTL,
1905 	    "<== hxge_alloc_rx_cntl_dma status 0x%08x", status));
1906 
1907 	return (status);
1908 }
1909 
1910 /*ARGSUSED*/
1911 static void
1912 hxge_free_rx_cntl_dma(p_hxge_t hxgep, p_hxge_dma_common_t dmap)
1913 {
1914 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "==> hxge_free_rx_cntl_dma"));
1915 
1916 	hxge_dma_mem_free(dmap);
1917 
1918 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "<== hxge_free_rx_cntl_dma"));
1919 }
1920 
1921 static hxge_status_t
1922 hxge_alloc_tx_mem_pool(p_hxge_t hxgep)
1923 {
1924 	hxge_status_t		status = HXGE_OK;
1925 	int			i, j;
1926 	uint32_t		ndmas, st_tdc;
1927 	p_hxge_dma_pt_cfg_t	p_all_cfgp;
1928 	p_hxge_hw_pt_cfg_t	p_cfgp;
1929 	p_hxge_dma_pool_t	dma_poolp;
1930 	p_hxge_dma_common_t	*dma_buf_p;
1931 	p_hxge_dma_pool_t	dma_cntl_poolp;
1932 	p_hxge_dma_common_t	*dma_cntl_p;
1933 	size_t			tx_buf_alloc_size;
1934 	size_t			tx_cntl_alloc_size;
1935 	uint32_t		*num_chunks;	/* per dma */
1936 
1937 	HXGE_DEBUG_MSG((hxgep, MEM_CTL, "==> hxge_alloc_tx_mem_pool"));
1938 
1939 	p_all_cfgp = (p_hxge_dma_pt_cfg_t)&hxgep->pt_config;
1940 	p_cfgp = (p_hxge_hw_pt_cfg_t)&p_all_cfgp->hw_config;
1941 	st_tdc = p_cfgp->start_tdc;
1942 	ndmas = p_cfgp->max_tdcs;
1943 
1944 	HXGE_DEBUG_MSG((hxgep, MEM_CTL, "==> hxge_alloc_tx_mem_pool: "
1945 	    "p_cfgp 0x%016llx start_tdc %d ndmas %d hxgep->max_tdcs %d",
1946 	    p_cfgp, p_cfgp->start_tdc, p_cfgp->max_tdcs, hxgep->max_tdcs));
1947 	/*
1948 	 * Allocate memory for each transmit DMA channel.
1949 	 */
1950 	dma_poolp = (p_hxge_dma_pool_t)KMEM_ZALLOC(sizeof (hxge_dma_pool_t),
1951 	    KM_SLEEP);
1952 	dma_buf_p = (p_hxge_dma_common_t *)KMEM_ZALLOC(
1953 	    sizeof (p_hxge_dma_common_t) * ndmas, KM_SLEEP);
1954 
1955 	dma_cntl_poolp = (p_hxge_dma_pool_t)
1956 	    KMEM_ZALLOC(sizeof (hxge_dma_pool_t), KM_SLEEP);
1957 	dma_cntl_p = (p_hxge_dma_common_t *)KMEM_ZALLOC(
1958 	    sizeof (p_hxge_dma_common_t) * ndmas, KM_SLEEP);
1959 
1960 	hxgep->hxge_port_tx_ring_size = hxge_tx_ring_size;
1961 
1962 	/*
1963 	 * Assume that each DMA channel will be configured with default
1964 	 * transmit bufer size for copying transmit data. (For packet payload
1965 	 * over this limit, packets will not be copied.)
1966 	 */
1967 	tx_buf_alloc_size = (hxge_bcopy_thresh * hxge_tx_ring_size);
1968 
1969 	/*
1970 	 * Addresses of transmit descriptor ring and the mailbox must be all
1971 	 * cache-aligned (64 bytes).
1972 	 */
1973 	tx_cntl_alloc_size = hxge_tx_ring_size;
1974 	tx_cntl_alloc_size *= (sizeof (tx_desc_t));
1975 	tx_cntl_alloc_size += sizeof (txdma_mailbox_t);
1976 
1977 	num_chunks = (uint32_t *)KMEM_ZALLOC(sizeof (uint32_t) * ndmas,
1978 	    KM_SLEEP);
1979 
1980 	/*
1981 	 * Allocate memory for transmit buffers and descriptor rings. Replace
1982 	 * allocation functions with interface functions provided by the
1983 	 * partition manager when it is available.
1984 	 *
1985 	 * Allocate memory for the transmit buffer pool.
1986 	 */
1987 	for (i = 0; i < ndmas; i++) {
1988 		num_chunks[i] = 0;
1989 		status = hxge_alloc_tx_buf_dma(hxgep, st_tdc, &dma_buf_p[i],
1990 		    tx_buf_alloc_size, hxge_bcopy_thresh, &num_chunks[i]);
1991 		if (status != HXGE_OK) {
1992 			break;
1993 		}
1994 		st_tdc++;
1995 	}
1996 
1997 	if (i < ndmas) {
1998 		goto hxge_alloc_tx_mem_pool_fail1;
1999 	}
2000 
2001 	st_tdc = p_cfgp->start_tdc;
2002 
2003 	/*
2004 	 * Allocate memory for descriptor rings and mailbox.
2005 	 */
2006 	for (j = 0; j < ndmas; j++) {
2007 		status = hxge_alloc_tx_cntl_dma(hxgep, st_tdc, &dma_cntl_p[j],
2008 		    tx_cntl_alloc_size);
2009 		if (status != HXGE_OK) {
2010 			break;
2011 		}
2012 		st_tdc++;
2013 	}
2014 
2015 	if (j < ndmas) {
2016 		goto hxge_alloc_tx_mem_pool_fail2;
2017 	}
2018 
2019 	dma_poolp->ndmas = ndmas;
2020 	dma_poolp->num_chunks = num_chunks;
2021 	dma_poolp->buf_allocated = B_TRUE;
2022 	dma_poolp->dma_buf_pool_p = dma_buf_p;
2023 	hxgep->tx_buf_pool_p = dma_poolp;
2024 
2025 	dma_cntl_poolp->ndmas = ndmas;
2026 	dma_cntl_poolp->buf_allocated = B_TRUE;
2027 	dma_cntl_poolp->dma_buf_pool_p = dma_cntl_p;
2028 	hxgep->tx_cntl_pool_p = dma_cntl_poolp;
2029 
2030 	HXGE_DEBUG_MSG((hxgep, MEM_CTL,
2031 	    "==> hxge_alloc_tx_mem_pool: start_tdc %d "
2032 	    "ndmas %d poolp->ndmas %d", st_tdc, ndmas, dma_poolp->ndmas));
2033 
2034 	goto hxge_alloc_tx_mem_pool_exit;
2035 
2036 hxge_alloc_tx_mem_pool_fail2:
2037 	/* Free control buffers */
2038 	j--;
2039 	for (; j >= 0; j--) {
2040 		hxge_free_tx_cntl_dma(hxgep,
2041 		    (p_hxge_dma_common_t)dma_cntl_p[j]);
2042 	}
2043 
2044 hxge_alloc_tx_mem_pool_fail1:
2045 	/* Free data buffers */
2046 	i--;
2047 	for (; i >= 0; i--) {
2048 		hxge_free_tx_buf_dma(hxgep, (p_hxge_dma_common_t)dma_buf_p[i],
2049 		    num_chunks[i]);
2050 	}
2051 
2052 	KMEM_FREE(dma_poolp, sizeof (hxge_dma_pool_t));
2053 	KMEM_FREE(dma_buf_p, ndmas * sizeof (p_hxge_dma_common_t));
2054 	KMEM_FREE(dma_cntl_poolp, sizeof (hxge_dma_pool_t));
2055 	KMEM_FREE(dma_cntl_p, ndmas * sizeof (p_hxge_dma_common_t));
2056 	KMEM_FREE(num_chunks, sizeof (uint32_t) * ndmas);
2057 
2058 hxge_alloc_tx_mem_pool_exit:
2059 	HXGE_DEBUG_MSG((hxgep, MEM_CTL,
2060 	    "<== hxge_alloc_tx_mem_pool:status 0x%08x", status));
2061 
2062 	return (status);
2063 }
2064 
2065 static hxge_status_t
2066 hxge_alloc_tx_buf_dma(p_hxge_t hxgep, uint16_t dma_channel,
2067     p_hxge_dma_common_t *dmap, size_t alloc_size,
2068     size_t block_size, uint32_t *num_chunks)
2069 {
2070 	p_hxge_dma_common_t	tx_dmap;
2071 	hxge_status_t		status = HXGE_OK;
2072 	size_t			total_alloc_size;
2073 	size_t			allocated = 0;
2074 	int			i, size_index, array_size;
2075 
2076 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "==> hxge_alloc_tx_buf_dma"));
2077 
2078 	tx_dmap = (p_hxge_dma_common_t)
2079 	    KMEM_ZALLOC(sizeof (hxge_dma_common_t) * HXGE_DMA_BLOCK, KM_SLEEP);
2080 
2081 	total_alloc_size = alloc_size;
2082 	i = 0;
2083 	size_index = 0;
2084 	array_size = sizeof (alloc_sizes) / sizeof (size_t);
2085 	while ((size_index < array_size) &&
2086 	    (alloc_sizes[size_index] < alloc_size))
2087 		size_index++;
2088 	if (size_index >= array_size) {
2089 		size_index = array_size - 1;
2090 	}
2091 
2092 	while ((allocated < total_alloc_size) &&
2093 	    (size_index >= 0) && (i < HXGE_DMA_BLOCK)) {
2094 		tx_dmap[i].dma_chunk_index = i;
2095 		tx_dmap[i].block_size = block_size;
2096 		tx_dmap[i].alength = alloc_sizes[size_index];
2097 		tx_dmap[i].orig_alength = tx_dmap[i].alength;
2098 		tx_dmap[i].nblocks = alloc_sizes[size_index] / block_size;
2099 		tx_dmap[i].dma_channel = dma_channel;
2100 		tx_dmap[i].contig_alloc_type = B_FALSE;
2101 
2102 		status = hxge_dma_mem_alloc(hxgep, hxge_force_dma,
2103 		    &hxge_tx_dma_attr, tx_dmap[i].alength,
2104 		    &hxge_dev_buf_dma_acc_attr,
2105 		    DDI_DMA_WRITE | DDI_DMA_STREAMING,
2106 		    (p_hxge_dma_common_t)(&tx_dmap[i]));
2107 		if (status != HXGE_OK) {
2108 			HXGE_DEBUG_MSG((hxgep, DMA_CTL,
2109 			    " hxge_alloc_tx_buf_dma: Alloc Failed: "
2110 			    " for size: %d", alloc_sizes[size_index]));
2111 			size_index--;
2112 		} else {
2113 			i++;
2114 			allocated += alloc_sizes[size_index];
2115 		}
2116 	}
2117 
2118 	if (allocated < total_alloc_size) {
2119 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2120 		    " hxge_alloc_tx_buf_dma: failed due to"
2121 		    " allocated(%d) < required(%d)",
2122 		    allocated, total_alloc_size));
2123 		goto hxge_alloc_tx_mem_fail1;
2124 	}
2125 
2126 	*num_chunks = i;
2127 	*dmap = tx_dmap;
2128 	HXGE_DEBUG_MSG((hxgep, DMA_CTL,
2129 	    "==> hxge_alloc_tx_buf_dma dmap 0x%016llx num chunks %d",
2130 	    *dmap, i));
2131 	goto hxge_alloc_tx_mem_exit;
2132 
2133 hxge_alloc_tx_mem_fail1:
2134 	KMEM_FREE(tx_dmap, sizeof (hxge_dma_common_t) * HXGE_DMA_BLOCK);
2135 
2136 hxge_alloc_tx_mem_exit:
2137 	HXGE_DEBUG_MSG((hxgep, DMA_CTL,
2138 	    "<== hxge_alloc_tx_buf_dma status 0x%08x", status));
2139 
2140 	return (status);
2141 }
2142 
2143 /*ARGSUSED*/
2144 static void
2145 hxge_free_tx_buf_dma(p_hxge_t hxgep, p_hxge_dma_common_t dmap,
2146     uint32_t num_chunks)
2147 {
2148 	int i;
2149 
2150 	HXGE_DEBUG_MSG((hxgep, MEM_CTL, "==> hxge_free_tx_buf_dma"));
2151 
2152 	for (i = 0; i < num_chunks; i++) {
2153 		hxge_dma_mem_free(dmap++);
2154 	}
2155 
2156 	HXGE_DEBUG_MSG((hxgep, MEM_CTL, "<== hxge_free_tx_buf_dma"));
2157 }
2158 
2159 /*ARGSUSED*/
2160 static hxge_status_t
2161 hxge_alloc_tx_cntl_dma(p_hxge_t hxgep, uint16_t dma_channel,
2162     p_hxge_dma_common_t *dmap, size_t size)
2163 {
2164 	p_hxge_dma_common_t	tx_dmap;
2165 	hxge_status_t		status = HXGE_OK;
2166 
2167 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "==> hxge_alloc_tx_cntl_dma"));
2168 
2169 	tx_dmap = (p_hxge_dma_common_t)KMEM_ZALLOC(sizeof (hxge_dma_common_t),
2170 	    KM_SLEEP);
2171 
2172 	tx_dmap->contig_alloc_type = B_FALSE;
2173 
2174 	status = hxge_dma_mem_alloc(hxgep, hxge_force_dma,
2175 	    &hxge_tx_desc_dma_attr, size, &hxge_dev_desc_dma_acc_attr,
2176 	    DDI_DMA_RDWR | DDI_DMA_CONSISTENT, tx_dmap);
2177 	if (status != HXGE_OK) {
2178 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2179 		    " hxge_alloc_tx_cntl_dma: Alloc Failed: "
2180 		    " for size: %d", size));
2181 		goto hxge_alloc_tx_cntl_dma_fail1;
2182 	}
2183 
2184 	*dmap = tx_dmap;
2185 
2186 	goto hxge_alloc_tx_cntl_dma_exit;
2187 
2188 hxge_alloc_tx_cntl_dma_fail1:
2189 	KMEM_FREE(tx_dmap, sizeof (hxge_dma_common_t));
2190 
2191 hxge_alloc_tx_cntl_dma_exit:
2192 	HXGE_DEBUG_MSG((hxgep, DMA_CTL,
2193 	    "<== hxge_alloc_tx_cntl_dma status 0x%08x", status));
2194 
2195 	return (status);
2196 }
2197 
2198 /*ARGSUSED*/
2199 static void
2200 hxge_free_tx_cntl_dma(p_hxge_t hxgep, p_hxge_dma_common_t dmap)
2201 {
2202 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "==> hxge_free_tx_cntl_dma"));
2203 
2204 	hxge_dma_mem_free(dmap);
2205 
2206 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "<== hxge_free_tx_cntl_dma"));
2207 }
2208 
2209 static void
2210 hxge_free_tx_mem_pool(p_hxge_t hxgep)
2211 {
2212 	uint32_t		i, ndmas;
2213 	p_hxge_dma_pool_t	dma_poolp;
2214 	p_hxge_dma_common_t	*dma_buf_p;
2215 	p_hxge_dma_pool_t	dma_cntl_poolp;
2216 	p_hxge_dma_common_t	*dma_cntl_p;
2217 	uint32_t		*num_chunks;
2218 
2219 	HXGE_DEBUG_MSG((hxgep, MEM3_CTL, "==> hxge_free_tx_mem_pool"));
2220 
2221 	dma_poolp = hxgep->tx_buf_pool_p;
2222 	if (dma_poolp == NULL || (!dma_poolp->buf_allocated)) {
2223 		HXGE_DEBUG_MSG((hxgep, MEM3_CTL,
2224 		    "<== hxge_free_tx_mem_pool "
2225 		    "(null rx buf pool or buf not allocated"));
2226 		return;
2227 	}
2228 
2229 	dma_cntl_poolp = hxgep->tx_cntl_pool_p;
2230 	if (dma_cntl_poolp == NULL || (!dma_cntl_poolp->buf_allocated)) {
2231 		HXGE_DEBUG_MSG((hxgep, MEM3_CTL,
2232 		    "<== hxge_free_tx_mem_pool "
2233 		    "(null tx cntl buf pool or cntl buf not allocated"));
2234 		return;
2235 	}
2236 
2237 	dma_buf_p = dma_poolp->dma_buf_pool_p;
2238 	num_chunks = dma_poolp->num_chunks;
2239 
2240 	dma_cntl_p = dma_cntl_poolp->dma_buf_pool_p;
2241 	ndmas = dma_cntl_poolp->ndmas;
2242 
2243 	for (i = 0; i < ndmas; i++) {
2244 		hxge_free_tx_buf_dma(hxgep, dma_buf_p[i], num_chunks[i]);
2245 	}
2246 
2247 	for (i = 0; i < ndmas; i++) {
2248 		hxge_free_tx_cntl_dma(hxgep, dma_cntl_p[i]);
2249 	}
2250 
2251 	for (i = 0; i < ndmas; i++) {
2252 		KMEM_FREE(dma_buf_p[i],
2253 		    sizeof (hxge_dma_common_t) * HXGE_DMA_BLOCK);
2254 		KMEM_FREE(dma_cntl_p[i], sizeof (hxge_dma_common_t));
2255 	}
2256 
2257 	KMEM_FREE(num_chunks, sizeof (uint32_t) * ndmas);
2258 	KMEM_FREE(dma_cntl_p, ndmas * sizeof (p_hxge_dma_common_t));
2259 	KMEM_FREE(dma_cntl_poolp, sizeof (hxge_dma_pool_t));
2260 	KMEM_FREE(dma_buf_p, ndmas * sizeof (p_hxge_dma_common_t));
2261 	KMEM_FREE(dma_poolp, sizeof (hxge_dma_pool_t));
2262 
2263 	hxgep->tx_buf_pool_p = NULL;
2264 	hxgep->tx_cntl_pool_p = NULL;
2265 
2266 	HXGE_DEBUG_MSG((hxgep, MEM3_CTL, "<== hxge_free_tx_mem_pool"));
2267 }
2268 
2269 /*ARGSUSED*/
2270 static hxge_status_t
2271 hxge_dma_mem_alloc(p_hxge_t hxgep, dma_method_t method,
2272     struct ddi_dma_attr *dma_attrp,
2273     size_t length, ddi_device_acc_attr_t *acc_attr_p, uint_t xfer_flags,
2274     p_hxge_dma_common_t dma_p)
2275 {
2276 	caddr_t		kaddrp;
2277 	int		ddi_status = DDI_SUCCESS;
2278 
2279 	dma_p->dma_handle = NULL;
2280 	dma_p->acc_handle = NULL;
2281 	dma_p->kaddrp = NULL;
2282 
2283 	ddi_status = ddi_dma_alloc_handle(hxgep->dip, dma_attrp,
2284 	    DDI_DMA_DONTWAIT, NULL, &dma_p->dma_handle);
2285 	if (ddi_status != DDI_SUCCESS) {
2286 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2287 		    "hxge_dma_mem_alloc:ddi_dma_alloc_handle failed."));
2288 		return (HXGE_ERROR | HXGE_DDI_FAILED);
2289 	}
2290 
2291 	ddi_status = ddi_dma_mem_alloc(dma_p->dma_handle, length, acc_attr_p,
2292 	    xfer_flags, DDI_DMA_DONTWAIT, 0, &kaddrp, &dma_p->alength,
2293 	    &dma_p->acc_handle);
2294 	if (ddi_status != DDI_SUCCESS) {
2295 		/* The caller will decide whether it is fatal */
2296 		HXGE_DEBUG_MSG((hxgep, DMA_CTL,
2297 		    "hxge_dma_mem_alloc:ddi_dma_mem_alloc failed"));
2298 		ddi_dma_free_handle(&dma_p->dma_handle);
2299 		dma_p->dma_handle = NULL;
2300 		return (HXGE_ERROR | HXGE_DDI_FAILED);
2301 	}
2302 
2303 	if (dma_p->alength < length) {
2304 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2305 		    "hxge_dma_mem_alloc:ddi_dma_mem_alloc < length."));
2306 		ddi_dma_mem_free(&dma_p->acc_handle);
2307 		ddi_dma_free_handle(&dma_p->dma_handle);
2308 		dma_p->acc_handle = NULL;
2309 		dma_p->dma_handle = NULL;
2310 		return (HXGE_ERROR);
2311 	}
2312 
2313 	ddi_status = ddi_dma_addr_bind_handle(dma_p->dma_handle, NULL,
2314 	    kaddrp, dma_p->alength, xfer_flags, DDI_DMA_DONTWAIT, 0,
2315 	    &dma_p->dma_cookie, &dma_p->ncookies);
2316 	if (ddi_status != DDI_DMA_MAPPED) {
2317 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2318 		    "hxge_dma_mem_alloc:di_dma_addr_bind failed "
2319 		    "(staus 0x%x ncookies %d.)", ddi_status, dma_p->ncookies));
2320 		if (dma_p->acc_handle) {
2321 			ddi_dma_mem_free(&dma_p->acc_handle);
2322 			dma_p->acc_handle = NULL;
2323 		}
2324 		ddi_dma_free_handle(&dma_p->dma_handle);
2325 		dma_p->dma_handle = NULL;
2326 		return (HXGE_ERROR | HXGE_DDI_FAILED);
2327 	}
2328 
2329 	if (dma_p->ncookies != 1) {
2330 		HXGE_DEBUG_MSG((hxgep, DMA_CTL,
2331 		    "hxge_dma_mem_alloc:ddi_dma_addr_bind > 1 cookie"
2332 		    "(staus 0x%x ncookies %d.)", ddi_status, dma_p->ncookies));
2333 		if (dma_p->acc_handle) {
2334 			ddi_dma_mem_free(&dma_p->acc_handle);
2335 			dma_p->acc_handle = NULL;
2336 		}
2337 		(void) ddi_dma_unbind_handle(dma_p->dma_handle);
2338 		ddi_dma_free_handle(&dma_p->dma_handle);
2339 		dma_p->dma_handle = NULL;
2340 		return (HXGE_ERROR);
2341 	}
2342 
2343 	dma_p->kaddrp = kaddrp;
2344 #if defined(__i386)
2345 	dma_p->ioaddr_pp =
2346 	    (unsigned char *)(uint32_t)dma_p->dma_cookie.dmac_laddress;
2347 #else
2348 	dma_p->ioaddr_pp = (unsigned char *) dma_p->dma_cookie.dmac_laddress;
2349 #endif
2350 
2351 	HPI_DMA_ACC_HANDLE_SET(dma_p, dma_p->acc_handle);
2352 
2353 	HXGE_DEBUG_MSG((hxgep, DMA_CTL, "<== hxge_dma_mem_alloc: "
2354 	    "dma buffer allocated: dma_p $%p "
2355 	    "return dmac_ladress from cookie $%p dmac_size %d "
2356 	    "dma_p->ioaddr_p $%p "
2357 	    "dma_p->orig_ioaddr_p $%p "
2358 	    "orig_vatopa $%p "
2359 	    "alength %d (0x%x) "
2360 	    "kaddrp $%p "
2361 	    "length %d (0x%x)",
2362 	    dma_p,
2363 	    dma_p->dma_cookie.dmac_laddress,
2364 	    dma_p->dma_cookie.dmac_size,
2365 	    dma_p->ioaddr_pp,
2366 	    dma_p->orig_ioaddr_pp,
2367 	    dma_p->orig_vatopa,
2368 	    dma_p->alength, dma_p->alength,
2369 	    kaddrp,
2370 	    length, length));
2371 
2372 	return (HXGE_OK);
2373 }
2374 
2375 static void
2376 hxge_dma_mem_free(p_hxge_dma_common_t dma_p)
2377 {
2378 	if (dma_p == NULL)
2379 		return;
2380 
2381 	if (dma_p->dma_handle != NULL) {
2382 		if (dma_p->ncookies) {
2383 			(void) ddi_dma_unbind_handle(dma_p->dma_handle);
2384 			dma_p->ncookies = 0;
2385 		}
2386 		ddi_dma_free_handle(&dma_p->dma_handle);
2387 		dma_p->dma_handle = NULL;
2388 	}
2389 
2390 	if (dma_p->acc_handle != NULL) {
2391 		ddi_dma_mem_free(&dma_p->acc_handle);
2392 		dma_p->acc_handle = NULL;
2393 		HPI_DMA_ACC_HANDLE_SET(dma_p, NULL);
2394 	}
2395 
2396 	dma_p->kaddrp = NULL;
2397 	dma_p->alength = 0;
2398 }
2399 
2400 /*
2401  *	hxge_m_start() -- start transmitting and receiving.
2402  *
2403  *	This function is called by the MAC layer when the first
2404  *	stream is open to prepare the hardware ready for sending
2405  *	and transmitting packets.
2406  */
2407 static int
2408 hxge_m_start(void *arg)
2409 {
2410 	p_hxge_t hxgep = (p_hxge_t)arg;
2411 
2412 	HXGE_DEBUG_MSG((hxgep, NEMO_CTL, "==> hxge_m_start"));
2413 
2414 	MUTEX_ENTER(hxgep->genlock);
2415 
2416 	if (hxge_init(hxgep) != DDI_SUCCESS) {
2417 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2418 		    "<== hxge_m_start: initialization failed"));
2419 		MUTEX_EXIT(hxgep->genlock);
2420 		return (EIO);
2421 	}
2422 
2423 	if (hxgep->hxge_mac_state != HXGE_MAC_STARTED) {
2424 		/*
2425 		 * Start timer to check the system error and tx hangs
2426 		 */
2427 		hxgep->hxge_timerid = hxge_start_timer(hxgep,
2428 		    hxge_check_hw_state, HXGE_CHECK_TIMER);
2429 
2430 		hxgep->hxge_mac_state = HXGE_MAC_STARTED;
2431 
2432 		hxgep->timeout.link_status = 0;
2433 		hxgep->timeout.report_link_status = B_TRUE;
2434 		hxgep->timeout.ticks = drv_usectohz(2 * 1000000);
2435 
2436 		/* Start the link status timer to check the link status */
2437 		MUTEX_ENTER(&hxgep->timeout.lock);
2438 		hxgep->timeout.id = timeout(hxge_link_poll, (void *)hxgep,
2439 		    hxgep->timeout.ticks);
2440 		MUTEX_EXIT(&hxgep->timeout.lock);
2441 	}
2442 
2443 	MUTEX_EXIT(hxgep->genlock);
2444 
2445 	HXGE_DEBUG_MSG((hxgep, NEMO_CTL, "<== hxge_m_start"));
2446 
2447 	return (0);
2448 }
2449 
2450 /*
2451  * hxge_m_stop(): stop transmitting and receiving.
2452  */
2453 static void
2454 hxge_m_stop(void *arg)
2455 {
2456 	p_hxge_t hxgep = (p_hxge_t)arg;
2457 
2458 	HXGE_DEBUG_MSG((hxgep, NEMO_CTL, "==> hxge_m_stop"));
2459 
2460 	if (hxgep->hxge_timerid) {
2461 		hxge_stop_timer(hxgep, hxgep->hxge_timerid);
2462 		hxgep->hxge_timerid = 0;
2463 	}
2464 
2465 	/* Stop the link status timer before unregistering */
2466 	MUTEX_ENTER(&hxgep->timeout.lock);
2467 	if (hxgep->timeout.id) {
2468 		(void) untimeout(hxgep->timeout.id);
2469 		hxgep->timeout.id = 0;
2470 	}
2471 	hxge_link_update(hxgep, LINK_STATE_DOWN);
2472 	MUTEX_EXIT(&hxgep->timeout.lock);
2473 
2474 	MUTEX_ENTER(hxgep->genlock);
2475 
2476 	hxge_uninit(hxgep);
2477 
2478 	hxgep->hxge_mac_state = HXGE_MAC_STOPPED;
2479 
2480 	MUTEX_EXIT(hxgep->genlock);
2481 
2482 	HXGE_DEBUG_MSG((hxgep, NEMO_CTL, "<== hxge_m_stop"));
2483 }
2484 
2485 static int
2486 hxge_m_multicst(void *arg, boolean_t add, const uint8_t *mca)
2487 {
2488 	p_hxge_t		hxgep = (p_hxge_t)arg;
2489 	struct ether_addr	addrp;
2490 
2491 	HXGE_DEBUG_MSG((hxgep, MAC_CTL, "==> hxge_m_multicst: add %d", add));
2492 
2493 	bcopy(mca, (uint8_t *)&addrp, ETHERADDRL);
2494 
2495 	if (add) {
2496 		if (hxge_add_mcast_addr(hxgep, &addrp)) {
2497 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2498 			    "<== hxge_m_multicst: add multicast failed"));
2499 			return (EINVAL);
2500 		}
2501 	} else {
2502 		if (hxge_del_mcast_addr(hxgep, &addrp)) {
2503 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2504 			    "<== hxge_m_multicst: del multicast failed"));
2505 			return (EINVAL);
2506 		}
2507 	}
2508 
2509 	HXGE_DEBUG_MSG((hxgep, MAC_CTL, "<== hxge_m_multicst"));
2510 
2511 	return (0);
2512 }
2513 
2514 static int
2515 hxge_m_promisc(void *arg, boolean_t on)
2516 {
2517 	p_hxge_t hxgep = (p_hxge_t)arg;
2518 
2519 	HXGE_DEBUG_MSG((hxgep, MAC_CTL, "==> hxge_m_promisc: on %d", on));
2520 
2521 	if (hxge_set_promisc(hxgep, on)) {
2522 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2523 		    "<== hxge_m_promisc: set promisc failed"));
2524 		return (EINVAL);
2525 	}
2526 
2527 	HXGE_DEBUG_MSG((hxgep, MAC_CTL, "<== hxge_m_promisc: on %d", on));
2528 
2529 	return (0);
2530 }
2531 
2532 static void
2533 hxge_m_ioctl(void *arg, queue_t *wq, mblk_t *mp)
2534 {
2535 	p_hxge_t	hxgep = (p_hxge_t)arg;
2536 	struct iocblk	*iocp = (struct iocblk *)mp->b_rptr;
2537 	boolean_t	need_privilege;
2538 	int		err;
2539 	int		cmd;
2540 
2541 	HXGE_DEBUG_MSG((hxgep, NEMO_CTL, "==> hxge_m_ioctl"));
2542 
2543 	iocp = (struct iocblk *)mp->b_rptr;
2544 	iocp->ioc_error = 0;
2545 	need_privilege = B_TRUE;
2546 	cmd = iocp->ioc_cmd;
2547 
2548 	HXGE_DEBUG_MSG((hxgep, NEMO_CTL, "==> hxge_m_ioctl: cmd 0x%08x", cmd));
2549 	switch (cmd) {
2550 	default:
2551 		miocnak(wq, mp, 0, EINVAL);
2552 		HXGE_DEBUG_MSG((hxgep, NEMO_CTL, "<== hxge_m_ioctl: invalid"));
2553 		return;
2554 
2555 	case LB_GET_INFO_SIZE:
2556 	case LB_GET_INFO:
2557 	case LB_GET_MODE:
2558 		need_privilege = B_FALSE;
2559 		break;
2560 
2561 	case LB_SET_MODE:
2562 		break;
2563 
2564 	case ND_GET:
2565 		need_privilege = B_FALSE;
2566 		break;
2567 	case ND_SET:
2568 		break;
2569 
2570 	case HXGE_GET_TX_RING_SZ:
2571 	case HXGE_GET_TX_DESC:
2572 	case HXGE_TX_SIDE_RESET:
2573 	case HXGE_RX_SIDE_RESET:
2574 	case HXGE_GLOBAL_RESET:
2575 	case HXGE_RESET_MAC:
2576 	case HXGE_PUT_TCAM:
2577 	case HXGE_GET_TCAM:
2578 	case HXGE_RTRACE:
2579 
2580 		need_privilege = B_FALSE;
2581 		break;
2582 	}
2583 
2584 	if (need_privilege) {
2585 		err = secpolicy_net_config(iocp->ioc_cr, B_FALSE);
2586 		if (err != 0) {
2587 			miocnak(wq, mp, 0, err);
2588 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
2589 			    "<== hxge_m_ioctl: no priv"));
2590 			return;
2591 		}
2592 	}
2593 
2594 	switch (cmd) {
2595 	case ND_GET:
2596 		HXGE_DEBUG_MSG((hxgep, NEMO_CTL, "ND_GET command"));
2597 	case ND_SET:
2598 		HXGE_DEBUG_MSG((hxgep, NEMO_CTL, "ND_SET command"));
2599 		hxge_param_ioctl(hxgep, wq, mp, iocp);
2600 		break;
2601 
2602 	case LB_GET_MODE:
2603 	case LB_SET_MODE:
2604 	case LB_GET_INFO_SIZE:
2605 	case LB_GET_INFO:
2606 		hxge_loopback_ioctl(hxgep, wq, mp, iocp);
2607 		break;
2608 
2609 	case HXGE_PUT_TCAM:
2610 	case HXGE_GET_TCAM:
2611 	case HXGE_GET_TX_RING_SZ:
2612 	case HXGE_GET_TX_DESC:
2613 	case HXGE_TX_SIDE_RESET:
2614 	case HXGE_RX_SIDE_RESET:
2615 	case HXGE_GLOBAL_RESET:
2616 	case HXGE_RESET_MAC:
2617 		HXGE_DEBUG_MSG((hxgep, NEMO_CTL,
2618 		    "==> hxge_m_ioctl: cmd 0x%x", cmd));
2619 		hxge_hw_ioctl(hxgep, wq, mp, iocp);
2620 		break;
2621 	}
2622 
2623 	HXGE_DEBUG_MSG((hxgep, NEMO_CTL, "<== hxge_m_ioctl"));
2624 }
2625 
2626 /*ARGSUSED*/
2627 static int
2628 hxge_tx_ring_start(mac_ring_driver_t rdriver, uint64_t mr_gen_num)
2629 {
2630 	p_hxge_ring_handle_t	rhp = (p_hxge_ring_handle_t)rdriver;
2631 	p_hxge_t		hxgep;
2632 	p_tx_ring_t		ring;
2633 
2634 	ASSERT(rhp != NULL);
2635 	ASSERT((rhp->index >= 0) && (rhp->index < HXGE_MAX_TDCS));
2636 
2637 	hxgep = rhp->hxgep;
2638 
2639 	/*
2640 	 * Get the ring pointer.
2641 	 */
2642 	ring = hxgep->tx_rings->rings[rhp->index];
2643 
2644 	/*
2645 	 * Fill in the handle for the transmit.
2646 	 */
2647 	MUTEX_ENTER(&ring->lock);
2648 	rhp->started = B_TRUE;
2649 	ring->ring_handle = rhp->ring_handle;
2650 	MUTEX_EXIT(&ring->lock);
2651 
2652 	return (0);
2653 }
2654 
2655 static void
2656 hxge_tx_ring_stop(mac_ring_driver_t rdriver)
2657 {
2658 	p_hxge_ring_handle_t    rhp = (p_hxge_ring_handle_t)rdriver;
2659 	p_hxge_t		hxgep;
2660 	p_tx_ring_t		ring;
2661 
2662 	ASSERT(rhp != NULL);
2663 	ASSERT((rhp->index >= 0) && (rhp->index < HXGE_MAX_TDCS));
2664 
2665 	hxgep = rhp->hxgep;
2666 	ring = hxgep->tx_rings->rings[rhp->index];
2667 
2668 	MUTEX_ENTER(&ring->lock);
2669 	ring->ring_handle = (mac_ring_handle_t)NULL;
2670 	rhp->started = B_FALSE;
2671 	MUTEX_EXIT(&ring->lock);
2672 }
2673 
2674 static int
2675 hxge_rx_ring_start(mac_ring_driver_t rdriver, uint64_t mr_gen_num)
2676 {
2677 	p_hxge_ring_handle_t	rhp = (p_hxge_ring_handle_t)rdriver;
2678 	p_hxge_t		hxgep;
2679 	p_rx_rcr_ring_t		ring;
2680 	int			i;
2681 
2682 	ASSERT(rhp != NULL);
2683 	ASSERT((rhp->index >= 0) && (rhp->index < HXGE_MAX_TDCS));
2684 
2685 	hxgep = rhp->hxgep;
2686 
2687 	/*
2688 	 * Get pointer to ring.
2689 	 */
2690 	ring = hxgep->rx_rcr_rings->rcr_rings[rhp->index];
2691 
2692 	MUTEX_ENTER(&ring->lock);
2693 
2694 	if (rhp->started) {
2695 		MUTEX_EXIT(&ring->lock);
2696 		return (0);
2697 	}
2698 
2699 	/*
2700 	 * Set the ldvp and ldgp pointers to enable/disable
2701 	 * polling.
2702 	 */
2703 	for (i = 0; i < hxgep->ldgvp->maxldvs; i++) {
2704 		if ((hxgep->ldgvp->ldvp[i].is_rxdma == 1) &&
2705 		    (hxgep->ldgvp->ldvp[i].channel == rhp->index)) {
2706 			ring->ldvp = &hxgep->ldgvp->ldvp[i];
2707 			ring->ldgp = hxgep->ldgvp->ldvp[i].ldgp;
2708 			break;
2709 		}
2710 	}
2711 
2712 	rhp->started = B_TRUE;
2713 	ring->rcr_mac_handle = rhp->ring_handle;
2714 	ring->rcr_gen_num = mr_gen_num;
2715 	MUTEX_EXIT(&ring->lock);
2716 
2717 	return (0);
2718 }
2719 
2720 static void
2721 hxge_rx_ring_stop(mac_ring_driver_t rdriver)
2722 {
2723 	p_hxge_ring_handle_t	rhp = (p_hxge_ring_handle_t)rdriver;
2724 	p_hxge_t		hxgep;
2725 	p_rx_rcr_ring_t		ring;
2726 
2727 	ASSERT(rhp != NULL);
2728 	ASSERT((rhp->index >= 0) && (rhp->index < HXGE_MAX_TDCS));
2729 
2730 	hxgep = rhp->hxgep;
2731 	ring =  hxgep->rx_rcr_rings->rcr_rings[rhp->index];
2732 
2733 	MUTEX_ENTER(&ring->lock);
2734 	rhp->started = B_TRUE;
2735 	ring->rcr_mac_handle = NULL;
2736 	ring->ldvp = NULL;
2737 	ring->ldgp = NULL;
2738 	MUTEX_EXIT(&ring->lock);
2739 }
2740 
2741 static int
2742 hxge_rx_group_start(mac_group_driver_t gdriver)
2743 {
2744 	hxge_ring_group_t	*group = (hxge_ring_group_t *)gdriver;
2745 
2746 	ASSERT(group->hxgep != NULL);
2747 	ASSERT(group->hxgep->hxge_mac_state == HXGE_MAC_STARTED);
2748 
2749 	MUTEX_ENTER(group->hxgep->genlock);
2750 	group->started = B_TRUE;
2751 	MUTEX_EXIT(group->hxgep->genlock);
2752 
2753 	return (0);
2754 }
2755 
2756 static void
2757 hxge_rx_group_stop(mac_group_driver_t gdriver)
2758 {
2759 	hxge_ring_group_t	*group = (hxge_ring_group_t *)gdriver;
2760 
2761 	ASSERT(group->hxgep != NULL);
2762 	ASSERT(group->hxgep->hxge_mac_state == HXGE_MAC_STARTED);
2763 	ASSERT(group->started == B_TRUE);
2764 
2765 	MUTEX_ENTER(group->hxgep->genlock);
2766 	group->started = B_FALSE;
2767 	MUTEX_EXIT(group->hxgep->genlock);
2768 }
2769 
2770 static int
2771 hxge_mmac_get_slot(p_hxge_t hxgep, int *slot)
2772 {
2773 	int	i;
2774 
2775 	/*
2776 	 * Find an open slot.
2777 	 */
2778 	for (i = 0; i < hxgep->mmac.total; i++) {
2779 		if (!hxgep->mmac.addrs[i].set) {
2780 			*slot = i;
2781 			return (0);
2782 		}
2783 	}
2784 
2785 	return (ENXIO);
2786 }
2787 
2788 static int
2789 hxge_mmac_set_addr(p_hxge_t hxgep, int slot, const uint8_t *addr)
2790 {
2791 	struct ether_addr	eaddr;
2792 	hxge_status_t		status = HXGE_OK;
2793 
2794 	bcopy(addr, (uint8_t *)&eaddr, ETHERADDRL);
2795 
2796 	/*
2797 	 * Set new interface local address and re-init device.
2798 	 * This is destructive to any other streams attached
2799 	 * to this device.
2800 	 */
2801 	RW_ENTER_WRITER(&hxgep->filter_lock);
2802 	status = hxge_pfc_set_mac_address(hxgep, slot, &eaddr);
2803 	RW_EXIT(&hxgep->filter_lock);
2804 	if (status != HXGE_OK)
2805 		return (status);
2806 
2807 	hxgep->mmac.addrs[slot].set = B_TRUE;
2808 	bcopy(addr, hxgep->mmac.addrs[slot].addr, ETHERADDRL);
2809 	hxgep->mmac.available--;
2810 	if (slot == HXGE_MAC_DEFAULT_ADDR_SLOT)
2811 		hxgep->mmac.addrs[slot].primary = B_TRUE;
2812 
2813 	return (0);
2814 }
2815 
2816 static int
2817 hxge_mmac_find_addr(p_hxge_t hxgep, const uint8_t *addr, int *slot)
2818 {
2819 	int	i, result;
2820 
2821 	for (i = 0; i < hxgep->mmac.total; i++) {
2822 		if (hxgep->mmac.addrs[i].set) {
2823 			result = memcmp(hxgep->mmac.addrs[i].addr,
2824 			    addr, ETHERADDRL);
2825 			if (result == 0) {
2826 				*slot = i;
2827 				return (0);
2828 			}
2829 		}
2830 	}
2831 
2832 	return (EINVAL);
2833 }
2834 
2835 static int
2836 hxge_mmac_unset_addr(p_hxge_t hxgep, int slot)
2837 {
2838 	hxge_status_t	status;
2839 	int		i;
2840 
2841 	status = hxge_pfc_clear_mac_address(hxgep, slot);
2842 	if (status != HXGE_OK)
2843 		return (status);
2844 
2845 	for (i = 0; i < ETHERADDRL; i++)
2846 		hxgep->mmac.addrs[slot].addr[i] = 0;
2847 
2848 	hxgep->mmac.addrs[slot].set = B_FALSE;
2849 	if (slot == HXGE_MAC_DEFAULT_ADDR_SLOT)
2850 		hxgep->mmac.addrs[slot].primary = B_FALSE;
2851 	hxgep->mmac.available++;
2852 
2853 	return (0);
2854 }
2855 
2856 static int
2857 hxge_rx_group_add_mac(void *arg, const uint8_t *mac_addr)
2858 {
2859 	hxge_ring_group_t	*group = arg;
2860 	p_hxge_t		hxgep = group->hxgep;
2861 	int			slot = 0;
2862 
2863 	ASSERT(group->type == MAC_RING_TYPE_RX);
2864 
2865 	MUTEX_ENTER(hxgep->genlock);
2866 
2867 	/*
2868 	 * Find a slot for the address.
2869 	 */
2870 	if (hxge_mmac_get_slot(hxgep, &slot) != 0) {
2871 		MUTEX_EXIT(hxgep->genlock);
2872 		return (ENOSPC);
2873 	}
2874 
2875 	/*
2876 	 * Program the MAC address.
2877 	 */
2878 	if (hxge_mmac_set_addr(hxgep, slot, mac_addr) != 0) {
2879 		MUTEX_EXIT(hxgep->genlock);
2880 		return (ENOSPC);
2881 	}
2882 
2883 	MUTEX_EXIT(hxgep->genlock);
2884 	return (0);
2885 }
2886 
2887 static int
2888 hxge_rx_group_rem_mac(void *arg, const uint8_t *mac_addr)
2889 {
2890 	hxge_ring_group_t	*group = arg;
2891 	p_hxge_t		hxgep = group->hxgep;
2892 	int			rv, slot;
2893 
2894 	ASSERT(group->type == MAC_RING_TYPE_RX);
2895 
2896 	MUTEX_ENTER(hxgep->genlock);
2897 
2898 	if ((rv = hxge_mmac_find_addr(hxgep, mac_addr, &slot)) != 0) {
2899 		MUTEX_EXIT(hxgep->genlock);
2900 		return (rv);
2901 	}
2902 
2903 	if ((rv = hxge_mmac_unset_addr(hxgep, slot)) != 0) {
2904 		MUTEX_EXIT(hxgep->genlock);
2905 		return (rv);
2906 	}
2907 
2908 	MUTEX_EXIT(hxgep->genlock);
2909 	return (0);
2910 }
2911 
2912 static void
2913 hxge_group_get(void *arg, mac_ring_type_t type, int groupid,
2914     mac_group_info_t *infop, mac_group_handle_t gh)
2915 {
2916 	p_hxge_t		hxgep = arg;
2917 	hxge_ring_group_t	*group;
2918 
2919 	ASSERT(type == MAC_RING_TYPE_RX);
2920 
2921 	switch (type) {
2922 	case MAC_RING_TYPE_RX:
2923 		group = &hxgep->rx_groups[groupid];
2924 		group->hxgep = hxgep;
2925 		group->ghandle = gh;
2926 		group->index = groupid;
2927 		group->type = type;
2928 
2929 		infop->mgi_driver = (mac_group_driver_t)group;
2930 		infop->mgi_start = hxge_rx_group_start;
2931 		infop->mgi_stop = hxge_rx_group_stop;
2932 		infop->mgi_addmac = hxge_rx_group_add_mac;
2933 		infop->mgi_remmac = hxge_rx_group_rem_mac;
2934 		infop->mgi_count = HXGE_MAX_RDCS;
2935 		break;
2936 
2937 	case MAC_RING_TYPE_TX:
2938 	default:
2939 		break;
2940 	}
2941 }
2942 
2943 static int
2944 hxge_ring_get_htable_idx(p_hxge_t hxgep, mac_ring_type_t type, uint32_t channel)
2945 {
2946 	int i;
2947 
2948 	ASSERT(hxgep->ldgvp != NULL);
2949 
2950 	switch (type) {
2951 	case MAC_RING_TYPE_RX:
2952 		for (i = 0; i < hxgep->ldgvp->maxldvs; i++) {
2953 			if ((hxgep->ldgvp->ldvp[i].is_rxdma) &&
2954 			    (hxgep->ldgvp->ldvp[i].channel == channel)) {
2955 				return ((int)
2956 				    hxgep->ldgvp->ldvp[i].ldgp->htable_idx);
2957 			}
2958 		}
2959 		break;
2960 
2961 	case MAC_RING_TYPE_TX:
2962 		for (i = 0; i < hxgep->ldgvp->maxldvs; i++) {
2963 			if ((hxgep->ldgvp->ldvp[i].is_txdma) &&
2964 			    (hxgep->ldgvp->ldvp[i].channel == channel)) {
2965 				return ((int)
2966 				    hxgep->ldgvp->ldvp[i].ldgp->htable_idx);
2967 			}
2968 		}
2969 		break;
2970 
2971 	default:
2972 		break;
2973 	}
2974 
2975 	return (-1);
2976 }
2977 
2978 /*
2979  * Callback function for the GLDv3 layer to register all rings.
2980  */
2981 /*ARGSUSED*/
2982 static void
2983 hxge_fill_ring(void *arg, mac_ring_type_t type, const int rg_index,
2984     const int index, mac_ring_info_t *infop, mac_ring_handle_t rh)
2985 {
2986 	p_hxge_t	hxgep = arg;
2987 
2988 	ASSERT(hxgep != NULL);
2989 	ASSERT(infop != NULL);
2990 
2991 	switch (type) {
2992 	case MAC_RING_TYPE_TX: {
2993 		p_hxge_ring_handle_t	rhp;
2994 		mac_intr_t		*mintr = &infop->mri_intr;
2995 		p_hxge_intr_t		intrp;
2996 		int			htable_idx;
2997 
2998 		ASSERT((index >= 0) && (index < HXGE_MAX_TDCS));
2999 		rhp = &hxgep->tx_ring_handles[index];
3000 		rhp->hxgep = hxgep;
3001 		rhp->index = index;
3002 		rhp->ring_handle = rh;
3003 		infop->mri_driver = (mac_ring_driver_t)rhp;
3004 		infop->mri_start = hxge_tx_ring_start;
3005 		infop->mri_stop = hxge_tx_ring_stop;
3006 		infop->mri_tx = hxge_tx_ring_send;
3007 		infop->mri_stat = hxge_tx_ring_stat;
3008 
3009 		intrp = (p_hxge_intr_t)&hxgep->hxge_intr_type;
3010 		htable_idx = hxge_ring_get_htable_idx(hxgep, type, index);
3011 		if (htable_idx >= 0)
3012 			mintr->mi_ddi_handle = intrp->htable[htable_idx];
3013 		else
3014 			mintr->mi_ddi_handle = NULL;
3015 		break;
3016 	}
3017 
3018 	case MAC_RING_TYPE_RX: {
3019 		p_hxge_ring_handle_t    rhp;
3020 		mac_intr_t		hxge_mac_intr;
3021 		p_hxge_intr_t		intrp;
3022 		int			htable_idx;
3023 
3024 		ASSERT((index >= 0) && (index < HXGE_MAX_RDCS));
3025 		rhp = &hxgep->rx_ring_handles[index];
3026 		rhp->hxgep = hxgep;
3027 		rhp->index = index;
3028 		rhp->ring_handle = rh;
3029 
3030 		/*
3031 		 * Entrypoint to enable interrupt (disable poll) and
3032 		 * disable interrupt (enable poll).
3033 		 */
3034 		hxge_mac_intr.mi_handle = (mac_intr_handle_t)rhp;
3035 		hxge_mac_intr.mi_enable = (mac_intr_enable_t)hxge_disable_poll;
3036 		hxge_mac_intr.mi_disable = (mac_intr_disable_t)hxge_enable_poll;
3037 
3038 		intrp = (p_hxge_intr_t)&hxgep->hxge_intr_type;
3039 		htable_idx = hxge_ring_get_htable_idx(hxgep, type, index);
3040 		if (htable_idx >= 0)
3041 			hxge_mac_intr.mi_ddi_handle = intrp->htable[htable_idx];
3042 		else
3043 			hxge_mac_intr.mi_ddi_handle = NULL;
3044 
3045 		infop->mri_driver = (mac_ring_driver_t)rhp;
3046 		infop->mri_start = hxge_rx_ring_start;
3047 		infop->mri_stop = hxge_rx_ring_stop;
3048 		infop->mri_intr = hxge_mac_intr;
3049 		infop->mri_poll = hxge_rx_poll;
3050 		infop->mri_stat = hxge_rx_ring_stat;
3051 		break;
3052 	}
3053 
3054 	default:
3055 		break;
3056 	}
3057 }
3058 
3059 /*ARGSUSED*/
3060 boolean_t
3061 hxge_m_getcapab(void *arg, mac_capab_t cap, void *cap_data)
3062 {
3063 	p_hxge_t	hxgep = arg;
3064 
3065 	switch (cap) {
3066 	case MAC_CAPAB_HCKSUM: {
3067 		uint32_t	*txflags = cap_data;
3068 
3069 		*txflags = HCKSUM_INET_PARTIAL;
3070 		break;
3071 	}
3072 
3073 	case MAC_CAPAB_RINGS: {
3074 		mac_capab_rings_t	*cap_rings = cap_data;
3075 
3076 		MUTEX_ENTER(hxgep->genlock);
3077 		if (cap_rings->mr_type == MAC_RING_TYPE_RX) {
3078 			cap_rings->mr_group_type = MAC_GROUP_TYPE_STATIC;
3079 			cap_rings->mr_rnum = HXGE_MAX_RDCS;
3080 			cap_rings->mr_rget = hxge_fill_ring;
3081 			cap_rings->mr_gnum = HXGE_MAX_RX_GROUPS;
3082 			cap_rings->mr_gget = hxge_group_get;
3083 			cap_rings->mr_gaddring = NULL;
3084 			cap_rings->mr_gremring = NULL;
3085 		} else {
3086 			cap_rings->mr_group_type = MAC_GROUP_TYPE_STATIC;
3087 			cap_rings->mr_rnum = HXGE_MAX_TDCS;
3088 			cap_rings->mr_rget = hxge_fill_ring;
3089 			cap_rings->mr_gnum = 0;
3090 			cap_rings->mr_gget = NULL;
3091 			cap_rings->mr_gaddring = NULL;
3092 			cap_rings->mr_gremring = NULL;
3093 		}
3094 		MUTEX_EXIT(hxgep->genlock);
3095 		break;
3096 	}
3097 
3098 	default:
3099 		return (B_FALSE);
3100 	}
3101 	return (B_TRUE);
3102 }
3103 
3104 static boolean_t
3105 hxge_param_locked(mac_prop_id_t pr_num)
3106 {
3107 	/*
3108 	 * All adv_* parameters are locked (read-only) while
3109 	 * the device is in any sort of loopback mode ...
3110 	 */
3111 	switch (pr_num) {
3112 		case MAC_PROP_ADV_1000FDX_CAP:
3113 		case MAC_PROP_EN_1000FDX_CAP:
3114 		case MAC_PROP_ADV_1000HDX_CAP:
3115 		case MAC_PROP_EN_1000HDX_CAP:
3116 		case MAC_PROP_ADV_100FDX_CAP:
3117 		case MAC_PROP_EN_100FDX_CAP:
3118 		case MAC_PROP_ADV_100HDX_CAP:
3119 		case MAC_PROP_EN_100HDX_CAP:
3120 		case MAC_PROP_ADV_10FDX_CAP:
3121 		case MAC_PROP_EN_10FDX_CAP:
3122 		case MAC_PROP_ADV_10HDX_CAP:
3123 		case MAC_PROP_EN_10HDX_CAP:
3124 		case MAC_PROP_AUTONEG:
3125 		case MAC_PROP_FLOWCTRL:
3126 			return (B_TRUE);
3127 	}
3128 	return (B_FALSE);
3129 }
3130 
3131 /*
3132  * callback functions for set/get of properties
3133  */
3134 static int
3135 hxge_m_setprop(void *barg, const char *pr_name, mac_prop_id_t pr_num,
3136     uint_t pr_valsize, const void *pr_val)
3137 {
3138 	hxge_t		*hxgep = barg;
3139 	p_hxge_stats_t	statsp;
3140 	int		err = 0;
3141 	uint32_t	new_mtu, old_framesize, new_framesize;
3142 
3143 	HXGE_DEBUG_MSG((hxgep, DLADM_CTL, "==> hxge_m_setprop"));
3144 
3145 	statsp = hxgep->statsp;
3146 	MUTEX_ENTER(hxgep->genlock);
3147 	if (statsp->port_stats.lb_mode != hxge_lb_normal &&
3148 	    hxge_param_locked(pr_num)) {
3149 		/*
3150 		 * All adv_* parameters are locked (read-only)
3151 		 * while the device is in any sort of loopback mode.
3152 		 */
3153 		HXGE_DEBUG_MSG((hxgep, DLADM_CTL,
3154 		    "==> hxge_m_setprop: loopback mode: read only"));
3155 		MUTEX_EXIT(hxgep->genlock);
3156 		return (EBUSY);
3157 	}
3158 
3159 	switch (pr_num) {
3160 		/*
3161 		 * These properties are either not exist or read only
3162 		 */
3163 		case MAC_PROP_EN_1000FDX_CAP:
3164 		case MAC_PROP_EN_100FDX_CAP:
3165 		case MAC_PROP_EN_10FDX_CAP:
3166 		case MAC_PROP_EN_1000HDX_CAP:
3167 		case MAC_PROP_EN_100HDX_CAP:
3168 		case MAC_PROP_EN_10HDX_CAP:
3169 		case MAC_PROP_ADV_1000FDX_CAP:
3170 		case MAC_PROP_ADV_1000HDX_CAP:
3171 		case MAC_PROP_ADV_100FDX_CAP:
3172 		case MAC_PROP_ADV_100HDX_CAP:
3173 		case MAC_PROP_ADV_10FDX_CAP:
3174 		case MAC_PROP_ADV_10HDX_CAP:
3175 		case MAC_PROP_STATUS:
3176 		case MAC_PROP_SPEED:
3177 		case MAC_PROP_DUPLEX:
3178 		case MAC_PROP_AUTONEG:
3179 		/*
3180 		 * Flow control is handled in the shared domain and
3181 		 * it is readonly here.
3182 		 */
3183 		case MAC_PROP_FLOWCTRL:
3184 			err = EINVAL;
3185 			HXGE_DEBUG_MSG((hxgep, DLADM_CTL,
3186 			    "==> hxge_m_setprop:  read only property %d",
3187 			    pr_num));
3188 			break;
3189 
3190 		case MAC_PROP_MTU:
3191 			bcopy(pr_val, &new_mtu, sizeof (new_mtu));
3192 			HXGE_DEBUG_MSG((hxgep, DLADM_CTL,
3193 			    "==> hxge_m_setprop: set MTU: %d", new_mtu));
3194 
3195 			new_framesize = new_mtu + MTU_TO_FRAME_SIZE;
3196 			if (new_framesize == hxgep->vmac.maxframesize) {
3197 				err = 0;
3198 				break;
3199 			}
3200 
3201 			if (hxgep->hxge_mac_state == HXGE_MAC_STARTED) {
3202 				err = EBUSY;
3203 				break;
3204 			}
3205 
3206 			if (new_framesize < MIN_FRAME_SIZE ||
3207 			    new_framesize > MAX_FRAME_SIZE) {
3208 				err = EINVAL;
3209 				break;
3210 			}
3211 
3212 			old_framesize = hxgep->vmac.maxframesize;
3213 			hxgep->vmac.maxframesize = (uint16_t)new_framesize;
3214 
3215 			if (hxge_vmac_set_framesize(hxgep)) {
3216 				hxgep->vmac.maxframesize =
3217 				    (uint16_t)old_framesize;
3218 				err = EINVAL;
3219 				break;
3220 			}
3221 
3222 			err = mac_maxsdu_update(hxgep->mach, new_mtu);
3223 			if (err) {
3224 				hxgep->vmac.maxframesize =
3225 				    (uint16_t)old_framesize;
3226 				(void) hxge_vmac_set_framesize(hxgep);
3227 			}
3228 
3229 			HXGE_DEBUG_MSG((hxgep, DLADM_CTL,
3230 			    "==> hxge_m_setprop: set MTU: %d maxframe %d",
3231 			    new_mtu, hxgep->vmac.maxframesize));
3232 			break;
3233 
3234 		case MAC_PROP_PRIVATE:
3235 			HXGE_DEBUG_MSG((hxgep, DLADM_CTL,
3236 			    "==> hxge_m_setprop: private property"));
3237 			err = hxge_set_priv_prop(hxgep, pr_name, pr_valsize,
3238 			    pr_val);
3239 			break;
3240 
3241 		default:
3242 			err = ENOTSUP;
3243 			break;
3244 	}
3245 
3246 	MUTEX_EXIT(hxgep->genlock);
3247 
3248 	HXGE_DEBUG_MSG((hxgep, DLADM_CTL,
3249 	    "<== hxge_m_setprop (return %d)", err));
3250 
3251 	return (err);
3252 }
3253 
3254 static int
3255 hxge_m_getprop(void *barg, const char *pr_name, mac_prop_id_t pr_num,
3256     uint_t pr_valsize, void *pr_val)
3257 {
3258 	hxge_t		*hxgep = barg;
3259 	p_hxge_stats_t	statsp = hxgep->statsp;
3260 	int		err = 0;
3261 	link_flowctrl_t fl;
3262 	uint64_t	tmp = 0;
3263 	link_state_t	ls;
3264 
3265 	HXGE_DEBUG_MSG((hxgep, DLADM_CTL,
3266 	    "==> hxge_m_getprop: pr_num %d", pr_num));
3267 
3268 	switch (pr_num) {
3269 		case MAC_PROP_DUPLEX:
3270 			*(uint8_t *)pr_val = statsp->mac_stats.link_duplex;
3271 			HXGE_DEBUG_MSG((hxgep, DLADM_CTL,
3272 			    "==> hxge_m_getprop: duplex mode %d",
3273 			    *(uint8_t *)pr_val));
3274 			break;
3275 
3276 		case MAC_PROP_SPEED:
3277 			ASSERT(pr_valsize >= sizeof (uint64_t));
3278 			tmp = statsp->mac_stats.link_speed * 1000000ull;
3279 			bcopy(&tmp, pr_val, sizeof (tmp));
3280 			break;
3281 
3282 		case MAC_PROP_STATUS:
3283 			ASSERT(pr_valsize >= sizeof (link_state_t));
3284 			if (!statsp->mac_stats.link_up)
3285 				ls = LINK_STATE_DOWN;
3286 			else
3287 				ls = LINK_STATE_UP;
3288 			bcopy(&ls, pr_val, sizeof (ls));
3289 			break;
3290 
3291 		case MAC_PROP_FLOWCTRL:
3292 			/*
3293 			 * Flow control is supported by the shared domain and
3294 			 * it is currently transmit only
3295 			 */
3296 			ASSERT(pr_valsize < sizeof (link_flowctrl_t));
3297 			fl = LINK_FLOWCTRL_TX;
3298 			bcopy(&fl, pr_val, sizeof (fl));
3299 			break;
3300 		case MAC_PROP_AUTONEG:
3301 			/* 10G link only and it is not negotiable */
3302 			*(uint8_t *)pr_val = 0;
3303 			break;
3304 		case MAC_PROP_ADV_1000FDX_CAP:
3305 		case MAC_PROP_ADV_100FDX_CAP:
3306 		case MAC_PROP_ADV_10FDX_CAP:
3307 		case MAC_PROP_ADV_1000HDX_CAP:
3308 		case MAC_PROP_ADV_100HDX_CAP:
3309 		case MAC_PROP_ADV_10HDX_CAP:
3310 		case MAC_PROP_EN_1000FDX_CAP:
3311 		case MAC_PROP_EN_100FDX_CAP:
3312 		case MAC_PROP_EN_10FDX_CAP:
3313 		case MAC_PROP_EN_1000HDX_CAP:
3314 		case MAC_PROP_EN_100HDX_CAP:
3315 		case MAC_PROP_EN_10HDX_CAP:
3316 			err = ENOTSUP;
3317 			break;
3318 
3319 		case MAC_PROP_PRIVATE:
3320 			err = hxge_get_priv_prop(hxgep, pr_name, pr_valsize,
3321 			    pr_val);
3322 			break;
3323 
3324 		default:
3325 			err = ENOTSUP;
3326 			break;
3327 	}
3328 
3329 	HXGE_DEBUG_MSG((hxgep, DLADM_CTL, "<== hxge_m_getprop"));
3330 
3331 	return (err);
3332 }
3333 
3334 static void
3335 hxge_m_propinfo(void *arg, const char *pr_name,
3336     mac_prop_id_t pr_num, mac_prop_info_handle_t prh)
3337 {
3338 	_NOTE(ARGUNUSED(arg));
3339 	switch (pr_num) {
3340 	case MAC_PROP_DUPLEX:
3341 	case MAC_PROP_SPEED:
3342 	case MAC_PROP_STATUS:
3343 	case MAC_PROP_AUTONEG:
3344 	case MAC_PROP_FLOWCTRL:
3345 		mac_prop_info_set_perm(prh, MAC_PROP_PERM_READ);
3346 		break;
3347 
3348 	case MAC_PROP_MTU:
3349 		mac_prop_info_set_range_uint32(prh,
3350 		    MIN_FRAME_SIZE - MTU_TO_FRAME_SIZE,
3351 		    MAX_FRAME_SIZE - MTU_TO_FRAME_SIZE);
3352 		break;
3353 
3354 	case MAC_PROP_PRIVATE: {
3355 		char valstr[MAXNAMELEN];
3356 
3357 		bzero(valstr, sizeof (valstr));
3358 
3359 		/* Receive Interrupt Blanking Parameters */
3360 		if (strcmp(pr_name, "_rxdma_intr_time") == 0) {
3361 			(void) snprintf(valstr, sizeof (valstr), "%d",
3362 			    RXDMA_RCR_TO_DEFAULT);
3363 		} else if (strcmp(pr_name, "_rxdma_intr_pkts") == 0) {
3364 			(void) snprintf(valstr, sizeof (valstr), "%d",
3365 			    RXDMA_RCR_PTHRES_DEFAULT);
3366 
3367 		/* Classification and Load Distribution Configuration */
3368 		} else if (strcmp(pr_name, "_class_opt_ipv4_tcp") == 0 ||
3369 		    strcmp(pr_name, "_class_opt_ipv4_udp") == 0 ||
3370 		    strcmp(pr_name, "_class_opt_ipv4_ah") == 0 ||
3371 		    strcmp(pr_name, "_class_opt_ipv4_sctp") == 0 ||
3372 		    strcmp(pr_name, "_class_opt_ipv6_tcp") == 0 ||
3373 		    strcmp(pr_name, "_class_opt_ipv6_udp") == 0 ||
3374 		    strcmp(pr_name, "_class_opt_ipv6_ah") == 0 ||
3375 		    strcmp(pr_name, "_class_opt_ipv6_sctp") == 0) {
3376 			(void) snprintf(valstr, sizeof (valstr), "%d",
3377 			    HXGE_CLASS_TCAM_LOOKUP);
3378 		}
3379 
3380 		if (strlen(valstr) > 0)
3381 			mac_prop_info_set_default_str(prh, valstr);
3382 		break;
3383 	}
3384 	}
3385 }
3386 
3387 
3388 /* ARGSUSED */
3389 static int
3390 hxge_set_priv_prop(p_hxge_t hxgep, const char *pr_name, uint_t pr_valsize,
3391     const void *pr_val)
3392 {
3393 	p_hxge_param_t	param_arr = hxgep->param_arr;
3394 	int		err = 0;
3395 
3396 	HXGE_DEBUG_MSG((hxgep, DLADM_CTL,
3397 	    "==> hxge_set_priv_prop: name %s (value %s)", pr_name, pr_val));
3398 
3399 	if (pr_val == NULL) {
3400 		return (EINVAL);
3401 	}
3402 
3403 	/* Blanking */
3404 	if (strcmp(pr_name, "_rxdma_intr_time") == 0) {
3405 		err = hxge_param_rx_intr_time(hxgep, NULL, NULL,
3406 		    (char *)pr_val, (caddr_t)&param_arr[param_rxdma_intr_time]);
3407 	} else if (strcmp(pr_name, "_rxdma_intr_pkts") == 0) {
3408 		err = hxge_param_rx_intr_pkts(hxgep, NULL, NULL,
3409 		    (char *)pr_val, (caddr_t)&param_arr[param_rxdma_intr_pkts]);
3410 
3411 	/* Classification */
3412 	} else if (strcmp(pr_name, "_class_opt_ipv4_tcp") == 0) {
3413 		err = hxge_param_set_ip_opt(hxgep, NULL, NULL, (char *)pr_val,
3414 		    (caddr_t)&param_arr[param_class_opt_ipv4_tcp]);
3415 	} else if (strcmp(pr_name, "_class_opt_ipv4_udp") == 0) {
3416 		err = hxge_param_set_ip_opt(hxgep, NULL, NULL, (char *)pr_val,
3417 		    (caddr_t)&param_arr[param_class_opt_ipv4_udp]);
3418 	} else if (strcmp(pr_name, "_class_opt_ipv4_ah") == 0) {
3419 		err = hxge_param_set_ip_opt(hxgep, NULL, NULL, (char *)pr_val,
3420 		    (caddr_t)&param_arr[param_class_opt_ipv4_ah]);
3421 	} else if (strcmp(pr_name, "_class_opt_ipv4_sctp") == 0) {
3422 		err = hxge_param_set_ip_opt(hxgep, NULL, NULL, (char *)pr_val,
3423 		    (caddr_t)&param_arr[param_class_opt_ipv4_sctp]);
3424 	} else if (strcmp(pr_name, "_class_opt_ipv6_tcp") == 0) {
3425 		err = hxge_param_set_ip_opt(hxgep, NULL, NULL, (char *)pr_val,
3426 		    (caddr_t)&param_arr[param_class_opt_ipv6_tcp]);
3427 	} else if (strcmp(pr_name, "_class_opt_ipv6_udp") == 0) {
3428 		err = hxge_param_set_ip_opt(hxgep, NULL, NULL, (char *)pr_val,
3429 		    (caddr_t)&param_arr[param_class_opt_ipv6_udp]);
3430 	} else if (strcmp(pr_name, "_class_opt_ipv6_ah") == 0) {
3431 		err = hxge_param_set_ip_opt(hxgep, NULL, NULL, (char *)pr_val,
3432 		    (caddr_t)&param_arr[param_class_opt_ipv6_ah]);
3433 	} else if (strcmp(pr_name, "_class_opt_ipv6_sctp") == 0) {
3434 		err = hxge_param_set_ip_opt(hxgep, NULL, NULL, (char *)pr_val,
3435 		    (caddr_t)&param_arr[param_class_opt_ipv6_sctp]);
3436 	} else {
3437 		err = ENOTSUP;
3438 	}
3439 
3440 	HXGE_DEBUG_MSG((hxgep, DLADM_CTL,
3441 	    "<== hxge_set_priv_prop: err %d", err));
3442 
3443 	return (err);
3444 }
3445 
3446 static int
3447 hxge_get_priv_prop(p_hxge_t hxgep, const char *pr_name, uint_t pr_valsize,
3448     void *pr_val)
3449 {
3450 	p_hxge_param_t	param_arr = hxgep->param_arr;
3451 	char		valstr[MAXNAMELEN];
3452 	int		err = 0;
3453 	uint_t		strsize;
3454 	int		value = 0;
3455 
3456 	HXGE_DEBUG_MSG((hxgep, DLADM_CTL,
3457 	    "==> hxge_get_priv_prop: property %s", pr_name));
3458 
3459 	/* Receive Interrupt Blanking Parameters */
3460 	if (strcmp(pr_name, "_rxdma_intr_time") == 0) {
3461 		value = hxgep->intr_timeout;
3462 	} else if (strcmp(pr_name, "_rxdma_intr_pkts") == 0) {
3463 		value = hxgep->intr_threshold;
3464 
3465 	/* Classification and Load Distribution Configuration */
3466 	} else if (strcmp(pr_name, "_class_opt_ipv4_tcp") == 0) {
3467 		err = hxge_param_get_ip_opt(hxgep, NULL, NULL,
3468 		    (caddr_t)&param_arr[param_class_opt_ipv4_tcp]);
3469 
3470 		value = (int)param_arr[param_class_opt_ipv4_tcp].value;
3471 	} else if (strcmp(pr_name, "_class_opt_ipv4_udp") == 0) {
3472 		err = hxge_param_get_ip_opt(hxgep, NULL, NULL,
3473 		    (caddr_t)&param_arr[param_class_opt_ipv4_udp]);
3474 
3475 		value = (int)param_arr[param_class_opt_ipv4_udp].value;
3476 	} else if (strcmp(pr_name, "_class_opt_ipv4_ah") == 0) {
3477 		err = hxge_param_get_ip_opt(hxgep, NULL, NULL,
3478 		    (caddr_t)&param_arr[param_class_opt_ipv4_ah]);
3479 
3480 		value = (int)param_arr[param_class_opt_ipv4_ah].value;
3481 	} else if (strcmp(pr_name, "_class_opt_ipv4_sctp") == 0) {
3482 		err = hxge_param_get_ip_opt(hxgep, NULL, NULL,
3483 		    (caddr_t)&param_arr[param_class_opt_ipv4_sctp]);
3484 
3485 		value = (int)param_arr[param_class_opt_ipv4_sctp].value;
3486 	} else if (strcmp(pr_name, "_class_opt_ipv6_tcp") == 0) {
3487 		err = hxge_param_get_ip_opt(hxgep, NULL, NULL,
3488 		    (caddr_t)&param_arr[param_class_opt_ipv6_tcp]);
3489 
3490 		value = (int)param_arr[param_class_opt_ipv6_tcp].value;
3491 	} else if (strcmp(pr_name, "_class_opt_ipv6_udp") == 0) {
3492 		err = hxge_param_get_ip_opt(hxgep, NULL, NULL,
3493 		    (caddr_t)&param_arr[param_class_opt_ipv6_udp]);
3494 
3495 		value = (int)param_arr[param_class_opt_ipv6_udp].value;
3496 	} else if (strcmp(pr_name, "_class_opt_ipv6_ah") == 0) {
3497 		err = hxge_param_get_ip_opt(hxgep, NULL, NULL,
3498 		    (caddr_t)&param_arr[param_class_opt_ipv6_ah]);
3499 
3500 		value = (int)param_arr[param_class_opt_ipv6_ah].value;
3501 	} else if (strcmp(pr_name, "_class_opt_ipv6_sctp") == 0) {
3502 		err = hxge_param_get_ip_opt(hxgep, NULL, NULL,
3503 		    (caddr_t)&param_arr[param_class_opt_ipv6_sctp]);
3504 
3505 		value = (int)param_arr[param_class_opt_ipv6_sctp].value;
3506 	} else {
3507 		err = ENOTSUP;
3508 	}
3509 
3510 	if (err == 0) {
3511 		(void) snprintf(valstr, sizeof (valstr), "0x%x", value);
3512 
3513 		strsize = (uint_t)strlen(valstr);
3514 		if (pr_valsize < strsize) {
3515 			err = ENOBUFS;
3516 		} else {
3517 			(void) strlcpy(pr_val, valstr, pr_valsize);
3518 		}
3519 	}
3520 
3521 	HXGE_DEBUG_MSG((hxgep, DLADM_CTL,
3522 	    "<== hxge_get_priv_prop: return %d", err));
3523 
3524 	return (err);
3525 }
3526 /*
3527  * Module loading and removing entry points.
3528  */
3529 DDI_DEFINE_STREAM_OPS(hxge_dev_ops, nulldev, nulldev, hxge_attach, hxge_detach,
3530     nodev, NULL, D_MP, NULL, NULL);
3531 
3532 extern struct mod_ops mod_driverops;
3533 
3534 #define	HXGE_DESC_VER	"HXGE 10Gb Ethernet Driver"
3535 
3536 /*
3537  * Module linkage information for the kernel.
3538  */
3539 static struct modldrv hxge_modldrv = {
3540 	&mod_driverops,
3541 	HXGE_DESC_VER,
3542 	&hxge_dev_ops
3543 };
3544 
3545 static struct modlinkage modlinkage = {
3546 	MODREV_1, (void *) &hxge_modldrv, NULL
3547 };
3548 
3549 int
3550 _init(void)
3551 {
3552 	int status;
3553 
3554 	HXGE_DEBUG_MSG((NULL, MOD_CTL, "==> _init"));
3555 	mac_init_ops(&hxge_dev_ops, "hxge");
3556 	status = ddi_soft_state_init(&hxge_list, sizeof (hxge_t), 0);
3557 	if (status != 0) {
3558 		HXGE_ERROR_MSG((NULL, HXGE_ERR_CTL,
3559 		    "failed to init device soft state"));
3560 		mac_fini_ops(&hxge_dev_ops);
3561 		goto _init_exit;
3562 	}
3563 
3564 	status = mod_install(&modlinkage);
3565 	if (status != 0) {
3566 		ddi_soft_state_fini(&hxge_list);
3567 		HXGE_ERROR_MSG((NULL, HXGE_ERR_CTL, "Mod install failed"));
3568 		goto _init_exit;
3569 	}
3570 
3571 	MUTEX_INIT(&hxge_common_lock, NULL, MUTEX_DRIVER, NULL);
3572 
3573 _init_exit:
3574 	HXGE_DEBUG_MSG((NULL, MOD_CTL, "_init status = 0x%X", status));
3575 
3576 	return (status);
3577 }
3578 
3579 int
3580 _fini(void)
3581 {
3582 	int status;
3583 
3584 	HXGE_DEBUG_MSG((NULL, MOD_CTL, "==> _fini"));
3585 
3586 	HXGE_DEBUG_MSG((NULL, MOD_CTL, "==> _fini: mod_remove"));
3587 
3588 	if (hxge_mblks_pending)
3589 		return (EBUSY);
3590 
3591 	status = mod_remove(&modlinkage);
3592 	if (status != DDI_SUCCESS) {
3593 		HXGE_DEBUG_MSG((NULL, MOD_CTL,
3594 		    "Module removal failed 0x%08x", status));
3595 		goto _fini_exit;
3596 	}
3597 
3598 	mac_fini_ops(&hxge_dev_ops);
3599 
3600 	ddi_soft_state_fini(&hxge_list);
3601 
3602 	MUTEX_DESTROY(&hxge_common_lock);
3603 
3604 _fini_exit:
3605 	HXGE_DEBUG_MSG((NULL, MOD_CTL, "_fini status = 0x%08x", status));
3606 
3607 	return (status);
3608 }
3609 
3610 int
3611 _info(struct modinfo *modinfop)
3612 {
3613 	int status;
3614 
3615 	HXGE_DEBUG_MSG((NULL, MOD_CTL, "==> _info"));
3616 	status = mod_info(&modlinkage, modinfop);
3617 	HXGE_DEBUG_MSG((NULL, MOD_CTL, " _info status = 0x%X", status));
3618 
3619 	return (status);
3620 }
3621 
3622 /*ARGSUSED*/
3623 static hxge_status_t
3624 hxge_add_intrs(p_hxge_t hxgep)
3625 {
3626 	int		intr_types;
3627 	int		type = 0;
3628 	int		ddi_status = DDI_SUCCESS;
3629 	hxge_status_t	status = HXGE_OK;
3630 
3631 	HXGE_DEBUG_MSG((hxgep, INT_CTL, "==> hxge_add_intrs"));
3632 
3633 	hxgep->hxge_intr_type.intr_registered = B_FALSE;
3634 	hxgep->hxge_intr_type.intr_enabled = B_FALSE;
3635 	hxgep->hxge_intr_type.msi_intx_cnt = 0;
3636 	hxgep->hxge_intr_type.intr_added = 0;
3637 	hxgep->hxge_intr_type.niu_msi_enable = B_FALSE;
3638 	hxgep->hxge_intr_type.intr_type = 0;
3639 
3640 	if (hxge_msi_enable) {
3641 		hxgep->hxge_intr_type.niu_msi_enable = B_TRUE;
3642 	}
3643 
3644 	/* Get the supported interrupt types */
3645 	if ((ddi_status = ddi_intr_get_supported_types(hxgep->dip, &intr_types))
3646 	    != DDI_SUCCESS) {
3647 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL, "<== hxge_add_intrs: "
3648 		    "ddi_intr_get_supported_types failed: status 0x%08x",
3649 		    ddi_status));
3650 		return (HXGE_ERROR | HXGE_DDI_FAILED);
3651 	}
3652 
3653 	hxgep->hxge_intr_type.intr_types = intr_types;
3654 
3655 	HXGE_DEBUG_MSG((hxgep, INT_CTL, "==> hxge_add_intrs: "
3656 	    "ddi_intr_get_supported_types: 0x%08x", intr_types));
3657 
3658 	/*
3659 	 * Pick the interrupt type to use MSIX, MSI, INTX hxge_msi_enable:
3660 	 *	(1): 1 - MSI
3661 	 *	(2): 2 - MSI-X
3662 	 *	others - FIXED
3663 	 */
3664 	switch (hxge_msi_enable) {
3665 	default:
3666 		type = DDI_INTR_TYPE_FIXED;
3667 		HXGE_DEBUG_MSG((hxgep, INT_CTL, "==> hxge_add_intrs: "
3668 		    "use fixed (intx emulation) type %08x", type));
3669 		break;
3670 
3671 	case 2:
3672 		HXGE_DEBUG_MSG((hxgep, INT_CTL, "==> hxge_add_intrs: "
3673 		    "ddi_intr_get_supported_types: 0x%08x", intr_types));
3674 		if (intr_types & DDI_INTR_TYPE_MSIX) {
3675 			type = DDI_INTR_TYPE_MSIX;
3676 			HXGE_DEBUG_MSG((hxgep, INT_CTL,
3677 			    "==> hxge_add_intrs: "
3678 			    "ddi_intr_get_supported_types: MSIX 0x%08x", type));
3679 		} else if (intr_types & DDI_INTR_TYPE_MSI) {
3680 			type = DDI_INTR_TYPE_MSI;
3681 			HXGE_DEBUG_MSG((hxgep, INT_CTL,
3682 			    "==> hxge_add_intrs: "
3683 			    "ddi_intr_get_supported_types: MSI 0x%08x", type));
3684 		} else if (intr_types & DDI_INTR_TYPE_FIXED) {
3685 			type = DDI_INTR_TYPE_FIXED;
3686 			HXGE_DEBUG_MSG((hxgep, INT_CTL, "==> hxge_add_intrs: "
3687 			    "ddi_intr_get_supported_types: MSXED0x%08x", type));
3688 		}
3689 		break;
3690 
3691 	case 1:
3692 		if (intr_types & DDI_INTR_TYPE_MSI) {
3693 			type = DDI_INTR_TYPE_MSI;
3694 			HXGE_DEBUG_MSG((hxgep, INT_CTL,
3695 			    "==> hxge_add_intrs: "
3696 			    "ddi_intr_get_supported_types: MSI 0x%08x", type));
3697 		} else if (intr_types & DDI_INTR_TYPE_MSIX) {
3698 			type = DDI_INTR_TYPE_MSIX;
3699 			HXGE_DEBUG_MSG((hxgep, INT_CTL,
3700 			    "==> hxge_add_intrs: "
3701 			    "ddi_intr_get_supported_types: MSIX 0x%08x", type));
3702 		} else if (intr_types & DDI_INTR_TYPE_FIXED) {
3703 			type = DDI_INTR_TYPE_FIXED;
3704 			HXGE_DEBUG_MSG((hxgep, INT_CTL,
3705 			    "==> hxge_add_intrs: "
3706 			    "ddi_intr_get_supported_types: MSXED0x%08x", type));
3707 		}
3708 	}
3709 
3710 	hxgep->hxge_intr_type.intr_type = type;
3711 	if ((type == DDI_INTR_TYPE_MSIX || type == DDI_INTR_TYPE_MSI ||
3712 	    type == DDI_INTR_TYPE_FIXED) &&
3713 	    hxgep->hxge_intr_type.niu_msi_enable) {
3714 		if ((status = hxge_add_intrs_adv(hxgep)) != DDI_SUCCESS) {
3715 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3716 			    " hxge_add_intrs: "
3717 			    " hxge_add_intrs_adv failed: status 0x%08x",
3718 			    status));
3719 			return (status);
3720 		} else {
3721 			HXGE_DEBUG_MSG((hxgep, DDI_CTL, "==> hxge_add_intrs: "
3722 			    "interrupts registered : type %d", type));
3723 			hxgep->hxge_intr_type.intr_registered = B_TRUE;
3724 
3725 			HXGE_DEBUG_MSG((hxgep, DDI_CTL,
3726 			    "\nAdded advanced hxge add_intr_adv "
3727 			    "intr type 0x%x\n", type));
3728 
3729 			return (status);
3730 		}
3731 	}
3732 
3733 	if (!hxgep->hxge_intr_type.intr_registered) {
3734 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3735 		    "==> hxge_add_intrs: failed to register interrupts"));
3736 		return (HXGE_ERROR | HXGE_DDI_FAILED);
3737 	}
3738 
3739 	HXGE_DEBUG_MSG((hxgep, INT_CTL, "<== hxge_add_intrs"));
3740 
3741 	return (status);
3742 }
3743 
3744 /*ARGSUSED*/
3745 static hxge_status_t
3746 hxge_add_intrs_adv(p_hxge_t hxgep)
3747 {
3748 	int		intr_type;
3749 	p_hxge_intr_t	intrp;
3750 	hxge_status_t	status;
3751 
3752 	HXGE_DEBUG_MSG((hxgep, INT_CTL, "==> hxge_add_intrs_adv"));
3753 
3754 	intrp = (p_hxge_intr_t)&hxgep->hxge_intr_type;
3755 	intr_type = intrp->intr_type;
3756 
3757 	HXGE_DEBUG_MSG((hxgep, INT_CTL, "==> hxge_add_intrs_adv: type 0x%x",
3758 	    intr_type));
3759 
3760 	switch (intr_type) {
3761 	case DDI_INTR_TYPE_MSI:		/* 0x2 */
3762 	case DDI_INTR_TYPE_MSIX:	/* 0x4 */
3763 		status = hxge_add_intrs_adv_type(hxgep, intr_type);
3764 		break;
3765 
3766 	case DDI_INTR_TYPE_FIXED:	/* 0x1 */
3767 		status = hxge_add_intrs_adv_type_fix(hxgep, intr_type);
3768 		break;
3769 
3770 	default:
3771 		status = HXGE_ERROR;
3772 		break;
3773 	}
3774 
3775 	HXGE_DEBUG_MSG((hxgep, INT_CTL, "<== hxge_add_intrs_adv"));
3776 
3777 	return (status);
3778 }
3779 
3780 /*ARGSUSED*/
3781 static hxge_status_t
3782 hxge_add_intrs_adv_type(p_hxge_t hxgep, uint32_t int_type)
3783 {
3784 	dev_info_t	*dip = hxgep->dip;
3785 	p_hxge_ldg_t	ldgp;
3786 	p_hxge_intr_t	intrp;
3787 	uint_t		*inthandler;
3788 	void		*arg1, *arg2;
3789 	int		behavior;
3790 	int		nintrs, navail;
3791 	int		nactual, nrequired, nrequest;
3792 	int		inum = 0;
3793 	int		loop = 0;
3794 	int		x, y;
3795 	int		ddi_status = DDI_SUCCESS;
3796 	hxge_status_t	status = HXGE_OK;
3797 
3798 	HXGE_DEBUG_MSG((hxgep, INT_CTL, "==> hxge_add_intrs_adv_type"));
3799 
3800 	intrp = (p_hxge_intr_t)&hxgep->hxge_intr_type;
3801 
3802 	ddi_status = ddi_intr_get_nintrs(dip, int_type, &nintrs);
3803 	if ((ddi_status != DDI_SUCCESS) || (nintrs == 0)) {
3804 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3805 		    "ddi_intr_get_nintrs() failed, status: 0x%x%, "
3806 		    "nintrs: %d", ddi_status, nintrs));
3807 		return (HXGE_ERROR | HXGE_DDI_FAILED);
3808 	}
3809 
3810 	ddi_status = ddi_intr_get_navail(dip, int_type, &navail);
3811 	if ((ddi_status != DDI_SUCCESS) || (navail == 0)) {
3812 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3813 		    "ddi_intr_get_navail() failed, status: 0x%x%, "
3814 		    "nintrs: %d", ddi_status, navail));
3815 		return (HXGE_ERROR | HXGE_DDI_FAILED);
3816 	}
3817 
3818 	HXGE_DEBUG_MSG((hxgep, INT_CTL,
3819 	    "ddi_intr_get_navail() returned: intr type %d nintrs %d, navail %d",
3820 	    int_type, nintrs, navail));
3821 
3822 	/* PSARC/2007/453 MSI-X interrupt limit override */
3823 	if (int_type == DDI_INTR_TYPE_MSIX) {
3824 		nrequest = hxge_create_msi_property(hxgep);
3825 		if (nrequest < navail) {
3826 			navail = nrequest;
3827 			HXGE_DEBUG_MSG((hxgep, INT_CTL,
3828 			    "hxge_add_intrs_adv_type: nintrs %d "
3829 			    "navail %d (nrequest %d)",
3830 			    nintrs, navail, nrequest));
3831 		}
3832 	}
3833 
3834 	if (int_type == DDI_INTR_TYPE_MSI && !ISP2(navail)) {
3835 		/* MSI must be power of 2 */
3836 		if ((navail & 16) == 16) {
3837 			navail = 16;
3838 		} else if ((navail & 8) == 8) {
3839 			navail = 8;
3840 		} else if ((navail & 4) == 4) {
3841 			navail = 4;
3842 		} else if ((navail & 2) == 2) {
3843 			navail = 2;
3844 		} else {
3845 			navail = 1;
3846 		}
3847 		HXGE_DEBUG_MSG((hxgep, INT_CTL,
3848 		    "ddi_intr_get_navail(): (msi power of 2) nintrs %d, "
3849 		    "navail %d", nintrs, navail));
3850 	}
3851 
3852 	HXGE_DEBUG_MSG((hxgep, INT_CTL,
3853 	    "requesting: intr type %d nintrs %d, navail %d",
3854 	    int_type, nintrs, navail));
3855 
3856 	behavior = ((int_type == DDI_INTR_TYPE_FIXED) ? DDI_INTR_ALLOC_STRICT :
3857 	    DDI_INTR_ALLOC_NORMAL);
3858 	intrp->intr_size = navail * sizeof (ddi_intr_handle_t);
3859 	intrp->htable = kmem_zalloc(intrp->intr_size, KM_SLEEP);
3860 
3861 	ddi_status = ddi_intr_alloc(dip, intrp->htable, int_type, inum,
3862 	    navail, &nactual, behavior);
3863 	if (ddi_status != DDI_SUCCESS || nactual == 0) {
3864 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3865 		    " ddi_intr_alloc() failed: %d", ddi_status));
3866 		kmem_free(intrp->htable, intrp->intr_size);
3867 		return (HXGE_ERROR | HXGE_DDI_FAILED);
3868 	}
3869 
3870 	HXGE_DEBUG_MSG((hxgep, INT_CTL,
3871 	    "ddi_intr_alloc() returned: navail %d nactual %d",
3872 	    navail, nactual));
3873 
3874 	if ((ddi_status = ddi_intr_get_pri(intrp->htable[0],
3875 	    (uint_t *)&intrp->pri)) != DDI_SUCCESS) {
3876 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3877 		    " ddi_intr_get_pri() failed: %d", ddi_status));
3878 		/* Free already allocated interrupts */
3879 		for (y = 0; y < nactual; y++) {
3880 			(void) ddi_intr_free(intrp->htable[y]);
3881 		}
3882 
3883 		kmem_free(intrp->htable, intrp->intr_size);
3884 		return (HXGE_ERROR | HXGE_DDI_FAILED);
3885 	}
3886 
3887 	nrequired = 0;
3888 	status = hxge_ldgv_init(hxgep, &nactual, &nrequired);
3889 	if (status != HXGE_OK) {
3890 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3891 		    "hxge_add_intrs_adv_typ:hxge_ldgv_init "
3892 		    "failed: 0x%x", status));
3893 		/* Free already allocated interrupts */
3894 		for (y = 0; y < nactual; y++) {
3895 			(void) ddi_intr_free(intrp->htable[y]);
3896 		}
3897 
3898 		kmem_free(intrp->htable, intrp->intr_size);
3899 		return (status);
3900 	}
3901 
3902 	ldgp = hxgep->ldgvp->ldgp;
3903 	HXGE_DEBUG_MSG((hxgep, INT_CTL,
3904 	    "After hxge_ldgv_init(): nreq %d nactual %d", nrequired, nactual));
3905 
3906 	if (nactual < nrequired)
3907 		loop = nactual;
3908 	else
3909 		loop = nrequired;
3910 
3911 	for (x = 0; x < loop; x++, ldgp++) {
3912 		ldgp->vector = (uint8_t)x;
3913 		arg1 = ldgp->ldvp;
3914 		arg2 = hxgep;
3915 		if (ldgp->nldvs == 1) {
3916 			inthandler = (uint_t *)ldgp->ldvp->ldv_intr_handler;
3917 			HXGE_DEBUG_MSG((hxgep, INT_CTL,
3918 			    "hxge_add_intrs_adv_type: arg1 0x%x arg2 0x%x: "
3919 			    "1-1 int handler (entry %d)\n",
3920 			    arg1, arg2, x));
3921 		} else if (ldgp->nldvs > 1) {
3922 			inthandler = (uint_t *)ldgp->sys_intr_handler;
3923 			HXGE_DEBUG_MSG((hxgep, INT_CTL,
3924 			    "hxge_add_intrs_adv_type: arg1 0x%x arg2 0x%x: "
3925 			    "nldevs %d int handler (entry %d)\n",
3926 			    arg1, arg2, ldgp->nldvs, x));
3927 		}
3928 		HXGE_DEBUG_MSG((hxgep, INT_CTL,
3929 		    "==> hxge_add_intrs_adv_type: ddi_add_intr(inum) #%d "
3930 		    "htable 0x%llx", x, intrp->htable[x]));
3931 
3932 		if ((ddi_status = ddi_intr_add_handler(intrp->htable[x],
3933 		    (ddi_intr_handler_t *)inthandler, arg1, arg2)) !=
3934 		    DDI_SUCCESS) {
3935 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
3936 			    "==> hxge_add_intrs_adv_type: failed #%d "
3937 			    "status 0x%x", x, ddi_status));
3938 			for (y = 0; y < intrp->intr_added; y++) {
3939 				(void) ddi_intr_remove_handler(
3940 				    intrp->htable[y]);
3941 			}
3942 
3943 			/* Free already allocated intr */
3944 			for (y = 0; y < nactual; y++) {
3945 				(void) ddi_intr_free(intrp->htable[y]);
3946 			}
3947 			kmem_free(intrp->htable, intrp->intr_size);
3948 
3949 			(void) hxge_ldgv_uninit(hxgep);
3950 
3951 			return (HXGE_ERROR | HXGE_DDI_FAILED);
3952 		}
3953 
3954 		ldgp->htable_idx = x;
3955 		intrp->intr_added++;
3956 	}
3957 	intrp->msi_intx_cnt = nactual;
3958 
3959 	HXGE_DEBUG_MSG((hxgep, INT_CTL,
3960 	    "Requested: %d, Allowed: %d msi_intx_cnt %d intr_added %d",
3961 	    navail, nactual, intrp->msi_intx_cnt, intrp->intr_added));
3962 
3963 	(void) ddi_intr_get_cap(intrp->htable[0], &intrp->intr_cap);
3964 	(void) hxge_intr_ldgv_init(hxgep);
3965 
3966 	HXGE_DEBUG_MSG((hxgep, INT_CTL, "<== hxge_add_intrs_adv_type"));
3967 
3968 	return (status);
3969 }
3970 
3971 /*ARGSUSED*/
3972 static hxge_status_t
3973 hxge_add_intrs_adv_type_fix(p_hxge_t hxgep, uint32_t int_type)
3974 {
3975 	dev_info_t	*dip = hxgep->dip;
3976 	p_hxge_ldg_t	ldgp;
3977 	p_hxge_intr_t	intrp;
3978 	uint_t		*inthandler;
3979 	void		*arg1, *arg2;
3980 	int		behavior;
3981 	int		nintrs, navail;
3982 	int		nactual, nrequired;
3983 	int		inum = 0;
3984 	int		x, y;
3985 	int		ddi_status = DDI_SUCCESS;
3986 	hxge_status_t	status = HXGE_OK;
3987 
3988 	HXGE_DEBUG_MSG((hxgep, INT_CTL, "==> hxge_add_intrs_adv_type_fix"));
3989 	intrp = (p_hxge_intr_t)&hxgep->hxge_intr_type;
3990 
3991 	ddi_status = ddi_intr_get_nintrs(dip, int_type, &nintrs);
3992 	if ((ddi_status != DDI_SUCCESS) || (nintrs == 0)) {
3993 		HXGE_DEBUG_MSG((hxgep, INT_CTL,
3994 		    "ddi_intr_get_nintrs() failed, status: 0x%x%, "
3995 		    "nintrs: %d", status, nintrs));
3996 		return (HXGE_ERROR | HXGE_DDI_FAILED);
3997 	}
3998 
3999 	ddi_status = ddi_intr_get_navail(dip, int_type, &navail);
4000 	if ((ddi_status != DDI_SUCCESS) || (navail == 0)) {
4001 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
4002 		    "ddi_intr_get_navail() failed, status: 0x%x%, "
4003 		    "nintrs: %d", ddi_status, navail));
4004 		return (HXGE_ERROR | HXGE_DDI_FAILED);
4005 	}
4006 
4007 	HXGE_DEBUG_MSG((hxgep, INT_CTL,
4008 	    "ddi_intr_get_navail() returned: nintrs %d, naavail %d",
4009 	    nintrs, navail));
4010 
4011 	behavior = ((int_type == DDI_INTR_TYPE_FIXED) ? DDI_INTR_ALLOC_STRICT :
4012 	    DDI_INTR_ALLOC_NORMAL);
4013 	intrp->intr_size = navail * sizeof (ddi_intr_handle_t);
4014 	intrp->htable = kmem_alloc(intrp->intr_size, KM_SLEEP);
4015 	ddi_status = ddi_intr_alloc(dip, intrp->htable, int_type, inum,
4016 	    navail, &nactual, behavior);
4017 	if (ddi_status != DDI_SUCCESS || nactual == 0) {
4018 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
4019 		    " ddi_intr_alloc() failed: %d", ddi_status));
4020 		kmem_free(intrp->htable, intrp->intr_size);
4021 		return (HXGE_ERROR | HXGE_DDI_FAILED);
4022 	}
4023 
4024 	if ((ddi_status = ddi_intr_get_pri(intrp->htable[0],
4025 	    (uint_t *)&intrp->pri)) != DDI_SUCCESS) {
4026 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
4027 		    " ddi_intr_get_pri() failed: %d", ddi_status));
4028 		/* Free already allocated interrupts */
4029 		for (y = 0; y < nactual; y++) {
4030 			(void) ddi_intr_free(intrp->htable[y]);
4031 		}
4032 
4033 		kmem_free(intrp->htable, intrp->intr_size);
4034 		return (HXGE_ERROR | HXGE_DDI_FAILED);
4035 	}
4036 
4037 	nrequired = 0;
4038 	status = hxge_ldgv_init(hxgep, &nactual, &nrequired);
4039 	if (status != HXGE_OK) {
4040 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
4041 		    "hxge_add_intrs_adv_type_fix:hxge_ldgv_init "
4042 		    "failed: 0x%x", status));
4043 		/* Free already allocated interrupts */
4044 		for (y = 0; y < nactual; y++) {
4045 			(void) ddi_intr_free(intrp->htable[y]);
4046 		}
4047 
4048 		kmem_free(intrp->htable, intrp->intr_size);
4049 		return (status);
4050 	}
4051 
4052 	ldgp = hxgep->ldgvp->ldgp;
4053 	for (x = 0; x < nrequired; x++, ldgp++) {
4054 		ldgp->vector = (uint8_t)x;
4055 		arg1 = ldgp->ldvp;
4056 		arg2 = hxgep;
4057 		if (ldgp->nldvs == 1) {
4058 			inthandler = (uint_t *)ldgp->ldvp->ldv_intr_handler;
4059 			HXGE_DEBUG_MSG((hxgep, INT_CTL,
4060 			    "hxge_add_intrs_adv_type_fix: "
4061 			    "1-1 int handler(%d) ldg %d ldv %d "
4062 			    "arg1 $%p arg2 $%p\n",
4063 			    x, ldgp->ldg, ldgp->ldvp->ldv, arg1, arg2));
4064 		} else if (ldgp->nldvs > 1) {
4065 			inthandler = (uint_t *)ldgp->sys_intr_handler;
4066 			HXGE_DEBUG_MSG((hxgep, INT_CTL,
4067 			    "hxge_add_intrs_adv_type_fix: "
4068 			    "shared ldv %d int handler(%d) ldv %d ldg %d"
4069 			    "arg1 0x%016llx arg2 0x%016llx\n",
4070 			    x, ldgp->nldvs, ldgp->ldg, ldgp->ldvp->ldv,
4071 			    arg1, arg2));
4072 		}
4073 
4074 		if ((ddi_status = ddi_intr_add_handler(intrp->htable[x],
4075 		    (ddi_intr_handler_t *)inthandler, arg1, arg2)) !=
4076 		    DDI_SUCCESS) {
4077 			HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL,
4078 			    "==> hxge_add_intrs_adv_type_fix: failed #%d "
4079 			    "status 0x%x", x, ddi_status));
4080 			for (y = 0; y < intrp->intr_added; y++) {
4081 				(void) ddi_intr_remove_handler(
4082 				    intrp->htable[y]);
4083 			}
4084 			for (y = 0; y < nactual; y++) {
4085 				(void) ddi_intr_free(intrp->htable[y]);
4086 			}
4087 			/* Free already allocated intr */
4088 			kmem_free(intrp->htable, intrp->intr_size);
4089 
4090 			(void) hxge_ldgv_uninit(hxgep);
4091 
4092 			return (HXGE_ERROR | HXGE_DDI_FAILED);
4093 		}
4094 		intrp->intr_added++;
4095 	}
4096 
4097 	intrp->msi_intx_cnt = nactual;
4098 
4099 	(void) ddi_intr_get_cap(intrp->htable[0], &intrp->intr_cap);
4100 
4101 	status = hxge_intr_ldgv_init(hxgep);
4102 
4103 	HXGE_DEBUG_MSG((hxgep, INT_CTL, "<== hxge_add_intrs_adv_type_fix"));
4104 
4105 	return (status);
4106 }
4107 
4108 /*ARGSUSED*/
4109 static void
4110 hxge_remove_intrs(p_hxge_t hxgep)
4111 {
4112 	int		i, inum;
4113 	p_hxge_intr_t	intrp;
4114 
4115 	HXGE_DEBUG_MSG((hxgep, INT_CTL, "==> hxge_remove_intrs"));
4116 	intrp = (p_hxge_intr_t)&hxgep->hxge_intr_type;
4117 	if (!intrp->intr_registered) {
4118 		HXGE_DEBUG_MSG((hxgep, INT_CTL,
4119 		    "<== hxge_remove_intrs: interrupts not registered"));
4120 		return;
4121 	}
4122 
4123 	HXGE_DEBUG_MSG((hxgep, INT_CTL, "==> hxge_remove_intrs:advanced"));
4124 
4125 	if (intrp->intr_cap & DDI_INTR_FLAG_BLOCK) {
4126 		(void) ddi_intr_block_disable(intrp->htable,
4127 		    intrp->intr_added);
4128 	} else {
4129 		for (i = 0; i < intrp->intr_added; i++) {
4130 			(void) ddi_intr_disable(intrp->htable[i]);
4131 		}
4132 	}
4133 
4134 	for (inum = 0; inum < intrp->intr_added; inum++) {
4135 		if (intrp->htable[inum]) {
4136 			(void) ddi_intr_remove_handler(intrp->htable[inum]);
4137 		}
4138 	}
4139 
4140 	for (inum = 0; inum < intrp->msi_intx_cnt; inum++) {
4141 		if (intrp->htable[inum]) {
4142 			HXGE_DEBUG_MSG((hxgep, DDI_CTL,
4143 			    "hxge_remove_intrs: ddi_intr_free inum %d "
4144 			    "msi_intx_cnt %d intr_added %d",
4145 			    inum, intrp->msi_intx_cnt, intrp->intr_added));
4146 
4147 			(void) ddi_intr_free(intrp->htable[inum]);
4148 		}
4149 	}
4150 
4151 	kmem_free(intrp->htable, intrp->intr_size);
4152 	intrp->intr_registered = B_FALSE;
4153 	intrp->intr_enabled = B_FALSE;
4154 	intrp->msi_intx_cnt = 0;
4155 	intrp->intr_added = 0;
4156 
4157 	(void) hxge_ldgv_uninit(hxgep);
4158 
4159 	HXGE_DEBUG_MSG((hxgep, INT_CTL, "<== hxge_remove_intrs"));
4160 }
4161 
4162 /*ARGSUSED*/
4163 static void
4164 hxge_intrs_enable(p_hxge_t hxgep)
4165 {
4166 	p_hxge_intr_t	intrp;
4167 	int		i;
4168 	int		status;
4169 
4170 	HXGE_DEBUG_MSG((hxgep, INT_CTL, "==> hxge_intrs_enable"));
4171 
4172 	intrp = (p_hxge_intr_t)&hxgep->hxge_intr_type;
4173 
4174 	if (!intrp->intr_registered) {
4175 		HXGE_ERROR_MSG((hxgep, HXGE_ERR_CTL, "<== hxge_intrs_enable: "
4176 		    "interrupts are not registered"));
4177 		return;
4178 	}
4179 
4180 	if (intrp->intr_enabled) {
4181 		HXGE_DEBUG_MSG((hxgep, INT_CTL,
4182 		    "<== hxge_intrs_enable: already enabled"));
4183 		return;
4184 	}
4185 
4186 	if (intrp->intr_cap & DDI_INTR_FLAG_BLOCK) {
4187 		status = ddi_intr_block_enable(intrp->htable,
4188 		    intrp->intr_added);
4189 		HXGE_DEBUG_MSG((hxgep, INT_CTL, "==> hxge_intrs_enable "
4190 		    "block enable - status 0x%x total inums #%d\n",
4191 		    status, intrp->intr_added));
4192 	} else {
4193 		for (i = 0; i < intrp->intr_added; i++) {
4194 			status = ddi_intr_enable(intrp->htable[i]);
4195 			HXGE_DEBUG_MSG((hxgep, INT_CTL, "==> hxge_intrs_enable "
4196 			    "ddi_intr_enable:enable - status 0x%x "
4197 			    "total inums %d enable inum #%d\n",
4198 			    status, intrp->intr_added, i));
4199 			if (status == DDI_SUCCESS) {
4200 				intrp->intr_enabled = B_TRUE;
4201 			}
4202 		}
4203 	}
4204 
4205 	HXGE_DEBUG_MSG((hxgep, INT_CTL, "<== hxge_intrs_enable"));
4206 }
4207 
4208 /*ARGSUSED*/
4209 static void
4210 hxge_intrs_disable(p_hxge_t hxgep)
4211 {
4212 	p_hxge_intr_t	intrp;
4213 	int		i;
4214 
4215 	HXGE_DEBUG_MSG((hxgep, INT_CTL, "==> hxge_intrs_disable"));
4216 
4217 	intrp = (p_hxge_intr_t)&hxgep->hxge_intr_type;
4218 
4219 	if (!intrp->intr_registered) {
4220 		HXGE_DEBUG_MSG((hxgep, INT_CTL, "<== hxge_intrs_disable: "
4221 		    "interrupts are not registered"));
4222 		return;
4223 	}
4224 
4225 	if (intrp->intr_cap & DDI_INTR_FLAG_BLOCK) {
4226 		(void) ddi_intr_block_disable(intrp->htable,
4227 		    intrp->intr_added);
4228 	} else {
4229 		for (i = 0; i < intrp->intr_added; i++) {
4230 			(void) ddi_intr_disable(intrp->htable[i]);
4231 		}
4232 	}
4233 
4234 	intrp->intr_enabled = B_FALSE;
4235 	HXGE_DEBUG_MSG((hxgep, INT_CTL, "<== hxge_intrs_disable"));
4236 }
4237 
4238 static hxge_status_t
4239 hxge_mac_register(p_hxge_t hxgep)
4240 {
4241 	mac_register_t	*macp;
4242 	int		status;
4243 
4244 	HXGE_DEBUG_MSG((hxgep, DDI_CTL, "==> hxge_mac_register"));
4245 
4246 	if ((macp = mac_alloc(MAC_VERSION)) == NULL)
4247 		return (HXGE_ERROR);
4248 
4249 	macp->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
4250 	macp->m_driver = hxgep;
4251 	macp->m_dip = hxgep->dip;
4252 	macp->m_src_addr = hxgep->ouraddr.ether_addr_octet;
4253 	macp->m_callbacks = &hxge_m_callbacks;
4254 	macp->m_min_sdu = 0;
4255 	macp->m_max_sdu = hxgep->vmac.maxframesize - MTU_TO_FRAME_SIZE;
4256 	macp->m_margin = VLAN_TAGSZ;
4257 	macp->m_priv_props = hxge_priv_props;
4258 	macp->m_v12n = MAC_VIRT_LEVEL1;
4259 
4260 	HXGE_DEBUG_MSG((hxgep, DDI_CTL,
4261 	    "hxge_mac_register: ether addr is %x:%x:%x:%x:%x:%x",
4262 	    macp->m_src_addr[0],
4263 	    macp->m_src_addr[1],
4264 	    macp->m_src_addr[2],
4265 	    macp->m_src_addr[3],
4266 	    macp->m_src_addr[4],
4267 	    macp->m_src_addr[5]));
4268 
4269 	status = mac_register(macp, &hxgep->mach);
4270 	mac_free(macp);
4271 
4272 	if (status != 0) {
4273 		cmn_err(CE_WARN,
4274 		    "hxge_mac_register failed (status %d instance %d)",
4275 		    status, hxgep->instance);
4276 		return (HXGE_ERROR);
4277 	}
4278 
4279 	HXGE_DEBUG_MSG((hxgep, DDI_CTL, "<== hxge_mac_register success "
4280 	    "(instance %d)", hxgep->instance));
4281 
4282 	return (HXGE_OK);
4283 }
4284 
4285 static int
4286 hxge_init_common_dev(p_hxge_t hxgep)
4287 {
4288 	p_hxge_hw_list_t	hw_p;
4289 	dev_info_t		*p_dip;
4290 
4291 	HXGE_DEBUG_MSG((hxgep, MOD_CTL, "==> hxge_init_common_dev"));
4292 
4293 	p_dip = hxgep->p_dip;
4294 	MUTEX_ENTER(&hxge_common_lock);
4295 
4296 	/*
4297 	 * Loop through existing per Hydra hardware list.
4298 	 */
4299 	for (hw_p = hxge_hw_list; hw_p; hw_p = hw_p->next) {
4300 		HXGE_DEBUG_MSG((hxgep, MOD_CTL,
4301 		    "==> hxge_init_common_dev: hw_p $%p parent dip $%p",
4302 		    hw_p, p_dip));
4303 		if (hw_p->parent_devp == p_dip) {
4304 			hxgep->hxge_hw_p = hw_p;
4305 			hw_p->ndevs++;
4306 			hw_p->hxge_p = hxgep;
4307 			HXGE_DEBUG_MSG((hxgep, MOD_CTL,
4308 			    "==> hxge_init_common_device: "
4309 			    "hw_p $%p parent dip $%p ndevs %d (found)",
4310 			    hw_p, p_dip, hw_p->ndevs));
4311 			break;
4312 		}
4313 	}
4314 
4315 	if (hw_p == NULL) {
4316 		HXGE_DEBUG_MSG((hxgep, MOD_CTL,
4317 		    "==> hxge_init_common_dev: parent dip $%p (new)", p_dip));
4318 		hw_p = kmem_zalloc(sizeof (hxge_hw_list_t), KM_SLEEP);
4319 		hw_p->parent_devp = p_dip;
4320 		hw_p->magic = HXGE_MAGIC;
4321 		hxgep->hxge_hw_p = hw_p;
4322 		hw_p->ndevs++;
4323 		hw_p->hxge_p = hxgep;
4324 		hw_p->next = hxge_hw_list;
4325 
4326 		MUTEX_INIT(&hw_p->hxge_cfg_lock, NULL, MUTEX_DRIVER, NULL);
4327 		MUTEX_INIT(&hw_p->hxge_tcam_lock, NULL, MUTEX_DRIVER, NULL);
4328 		MUTEX_INIT(&hw_p->hxge_vlan_lock, NULL, MUTEX_DRIVER, NULL);
4329 
4330 		hxge_hw_list = hw_p;
4331 	}
4332 	MUTEX_EXIT(&hxge_common_lock);
4333 	HXGE_DEBUG_MSG((hxgep, MOD_CTL,
4334 	    "==> hxge_init_common_dev (hxge_hw_list) $%p", hxge_hw_list));
4335 	HXGE_DEBUG_MSG((hxgep, MOD_CTL, "<== hxge_init_common_dev"));
4336 
4337 	return (HXGE_OK);
4338 }
4339 
4340 static void
4341 hxge_uninit_common_dev(p_hxge_t hxgep)
4342 {
4343 	p_hxge_hw_list_t	hw_p, h_hw_p;
4344 	dev_info_t		*p_dip;
4345 
4346 	HXGE_DEBUG_MSG((hxgep, MOD_CTL, "==> hxge_uninit_common_dev"));
4347 	if (hxgep->hxge_hw_p == NULL) {
4348 		HXGE_DEBUG_MSG((hxgep, MOD_CTL,
4349 		    "<== hxge_uninit_common_dev (no common)"));
4350 		return;
4351 	}
4352 
4353 	MUTEX_ENTER(&hxge_common_lock);
4354 	h_hw_p = hxge_hw_list;
4355 	for (hw_p = hxge_hw_list; hw_p; hw_p = hw_p->next) {
4356 		p_dip = hw_p->parent_devp;
4357 		if (hxgep->hxge_hw_p == hw_p && p_dip == hxgep->p_dip &&
4358 		    hxgep->hxge_hw_p->magic == HXGE_MAGIC &&
4359 		    hw_p->magic == HXGE_MAGIC) {
4360 			HXGE_DEBUG_MSG((hxgep, MOD_CTL,
4361 			    "==> hxge_uninit_common_dev: "
4362 			    "hw_p $%p parent dip $%p ndevs %d (found)",
4363 			    hw_p, p_dip, hw_p->ndevs));
4364 
4365 			hxgep->hxge_hw_p = NULL;
4366 			if (hw_p->ndevs) {
4367 				hw_p->ndevs--;
4368 			}
4369 			hw_p->hxge_p = NULL;
4370 			if (!hw_p->ndevs) {
4371 				MUTEX_DESTROY(&hw_p->hxge_vlan_lock);
4372 				MUTEX_DESTROY(&hw_p->hxge_tcam_lock);
4373 				MUTEX_DESTROY(&hw_p->hxge_cfg_lock);
4374 				HXGE_DEBUG_MSG((hxgep, MOD_CTL,
4375 				    "==> hxge_uninit_common_dev: "
4376 				    "hw_p $%p parent dip $%p ndevs %d (last)",
4377 				    hw_p, p_dip, hw_p->ndevs));
4378 
4379 				if (hw_p == hxge_hw_list) {
4380 					HXGE_DEBUG_MSG((hxgep, MOD_CTL,
4381 					    "==> hxge_uninit_common_dev:"
4382 					    "remove head "
4383 					    "hw_p $%p parent dip $%p "
4384 					    "ndevs %d (head)",
4385 					    hw_p, p_dip, hw_p->ndevs));
4386 					hxge_hw_list = hw_p->next;
4387 				} else {
4388 					HXGE_DEBUG_MSG((hxgep, MOD_CTL,
4389 					    "==> hxge_uninit_common_dev:"
4390 					    "remove middle "
4391 					    "hw_p $%p parent dip $%p "
4392 					    "ndevs %d (middle)",
4393 					    hw_p, p_dip, hw_p->ndevs));
4394 					h_hw_p->next = hw_p->next;
4395 				}
4396 
4397 				KMEM_FREE(hw_p, sizeof (hxge_hw_list_t));
4398 			}
4399 			break;
4400 		} else {
4401 			h_hw_p = hw_p;
4402 		}
4403 	}
4404 
4405 	MUTEX_EXIT(&hxge_common_lock);
4406 	HXGE_DEBUG_MSG((hxgep, MOD_CTL,
4407 	    "==> hxge_uninit_common_dev (hxge_hw_list) $%p", hxge_hw_list));
4408 
4409 	HXGE_DEBUG_MSG((hxgep, MOD_CTL, "<= hxge_uninit_common_dev"));
4410 }
4411 
4412 #define	HXGE_MSIX_ENTRIES		32
4413 #define	HXGE_MSIX_WAIT_COUNT		10
4414 #define	HXGE_MSIX_PARITY_CHECK_COUNT	30
4415 
4416 static void
4417 hxge_link_poll(void *arg)
4418 {
4419 	p_hxge_t		hxgep = (p_hxge_t)arg;
4420 	hpi_handle_t		handle;
4421 	cip_link_stat_t		link_stat;
4422 	hxge_timeout		*to = &hxgep->timeout;
4423 
4424 	handle = HXGE_DEV_HPI_HANDLE(hxgep);
4425 	HXGE_REG_RD32(handle, CIP_LINK_STAT, &link_stat.value);
4426 
4427 	if (to->report_link_status ||
4428 	    (to->link_status != link_stat.bits.xpcs0_link_up)) {
4429 		to->link_status = link_stat.bits.xpcs0_link_up;
4430 		to->report_link_status = B_FALSE;
4431 
4432 		if (link_stat.bits.xpcs0_link_up) {
4433 			hxge_link_update(hxgep, LINK_STATE_UP);
4434 		} else {
4435 			hxge_link_update(hxgep, LINK_STATE_DOWN);
4436 		}
4437 	}
4438 
4439 	/* Restart the link status timer to check the link status */
4440 	MUTEX_ENTER(&to->lock);
4441 	to->id = timeout(hxge_link_poll, arg, to->ticks);
4442 	MUTEX_EXIT(&to->lock);
4443 }
4444 
4445 static void
4446 hxge_link_update(p_hxge_t hxgep, link_state_t state)
4447 {
4448 	p_hxge_stats_t		statsp = (p_hxge_stats_t)hxgep->statsp;
4449 
4450 	mac_link_update(hxgep->mach, state);
4451 	if (state == LINK_STATE_UP) {
4452 		statsp->mac_stats.link_speed = 10000;
4453 		statsp->mac_stats.link_duplex = 2;
4454 		statsp->mac_stats.link_up = 1;
4455 	} else {
4456 		statsp->mac_stats.link_speed = 0;
4457 		statsp->mac_stats.link_duplex = 0;
4458 		statsp->mac_stats.link_up = 0;
4459 	}
4460 }
4461 
4462 static void
4463 hxge_msix_init(p_hxge_t hxgep)
4464 {
4465 	uint32_t		data0;
4466 	uint32_t		data1;
4467 	uint32_t		data2;
4468 	int			i;
4469 	uint32_t		msix_entry0;
4470 	uint32_t		msix_entry1;
4471 	uint32_t		msix_entry2;
4472 	uint32_t		msix_entry3;
4473 
4474 	/* Change to use MSIx bar instead of indirect access */
4475 	for (i = 0; i < HXGE_MSIX_ENTRIES; i++) {
4476 		data0 = 0xffffffff - i;
4477 		data1 = 0xffffffff - i - 1;
4478 		data2 = 0xffffffff - i - 2;
4479 
4480 		HXGE_REG_WR32(hxgep->hpi_msi_handle, i * 16, data0);
4481 		HXGE_REG_WR32(hxgep->hpi_msi_handle, i * 16 + 4, data1);
4482 		HXGE_REG_WR32(hxgep->hpi_msi_handle, i * 16 + 8, data2);
4483 		HXGE_REG_WR32(hxgep->hpi_msi_handle, i * 16 + 12, 0);
4484 	}
4485 
4486 	/* Initialize ram data out buffer. */
4487 	for (i = 0; i < HXGE_MSIX_ENTRIES; i++) {
4488 		HXGE_REG_RD32(hxgep->hpi_msi_handle, i * 16, &msix_entry0);
4489 		HXGE_REG_RD32(hxgep->hpi_msi_handle, i * 16 + 4, &msix_entry1);
4490 		HXGE_REG_RD32(hxgep->hpi_msi_handle, i * 16 + 8, &msix_entry2);
4491 		HXGE_REG_RD32(hxgep->hpi_msi_handle, i * 16 + 12, &msix_entry3);
4492 	}
4493 }
4494 
4495 /*
4496  * The following function is to support
4497  * PSARC/2007/453 MSI-X interrupt limit override.
4498  */
4499 static int
4500 hxge_create_msi_property(p_hxge_t hxgep)
4501 {
4502 	int	nmsi;
4503 	extern	int ncpus;
4504 
4505 	HXGE_DEBUG_MSG((hxgep, MOD_CTL, "==>hxge_create_msi_property"));
4506 
4507 	(void) ddi_prop_create(DDI_DEV_T_NONE, hxgep->dip,
4508 	    DDI_PROP_CANSLEEP, "#msix-request", NULL, 0);
4509 	/*
4510 	 * The maximum MSI-X requested will be 8.
4511 	 * If the # of CPUs is less than 8, we will reqeust
4512 	 * # MSI-X based on the # of CPUs.
4513 	 */
4514 	if (ncpus >= HXGE_MSIX_REQUEST_10G) {
4515 		nmsi = HXGE_MSIX_REQUEST_10G;
4516 	} else {
4517 		nmsi = ncpus;
4518 	}
4519 
4520 	HXGE_DEBUG_MSG((hxgep, MOD_CTL,
4521 	    "==>hxge_create_msi_property(10G): exists 0x%x (nmsi %d)",
4522 	    ddi_prop_exists(DDI_DEV_T_NONE, hxgep->dip,
4523 	    DDI_PROP_CANSLEEP, "#msix-request"), nmsi));
4524 
4525 	HXGE_DEBUG_MSG((hxgep, MOD_CTL, "<==hxge_create_msi_property"));
4526 	return (nmsi);
4527 }
4528