xref: /titanic_52/usr/src/uts/common/io/nxge/nxge_main.c (revision 68c47f65208790c466e5e484f2293d3baed71c6a)
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 /*
23  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 /*
28  * SunOs MT STREAMS NIU/Neptune 10Gb Ethernet Device Driver.
29  */
30 #include	<sys/nxge/nxge_impl.h>
31 #include	<sys/nxge/nxge_hio.h>
32 #include	<sys/nxge/nxge_rxdma.h>
33 #include	<sys/pcie.h>
34 
35 uint32_t 	nxge_use_partition = 0;		/* debug partition flag */
36 uint32_t 	nxge_dma_obp_props_only = 1;	/* use obp published props */
37 uint32_t 	nxge_use_rdc_intr = 1;		/* debug to assign rdc intr */
38 /*
39  * PSARC/2007/453 MSI-X interrupt limit override
40  */
41 uint32_t	nxge_msi_enable = 2;
42 
43 /*
44  * Software workaround for a Neptune (PCI-E)
45  * hardware interrupt bug which the hardware
46  * may generate spurious interrupts after the
47  * device interrupt handler was removed. If this flag
48  * is enabled, the driver will reset the
49  * hardware when devices are being detached.
50  */
51 uint32_t	nxge_peu_reset_enable = 0;
52 
53 /*
54  * Software workaround for the hardware
55  * checksum bugs that affect packet transmission
56  * and receive:
57  *
58  * Usage of nxge_cksum_offload:
59  *
60  *  (1) nxge_cksum_offload = 0 (default):
61  *	- transmits packets:
62  *	  TCP: uses the hardware checksum feature.
63  *	  UDP: driver will compute the software checksum
64  *	       based on the partial checksum computed
65  *	       by the IP layer.
66  *	- receives packets
67  *	  TCP: marks packets checksum flags based on hardware result.
68  *	  UDP: will not mark checksum flags.
69  *
70  *  (2) nxge_cksum_offload = 1:
71  *	- transmit packets:
72  *	  TCP/UDP: uses the hardware checksum feature.
73  *	- receives packets
74  *	  TCP/UDP: marks packet checksum flags based on hardware result.
75  *
76  *  (3) nxge_cksum_offload = 2:
77  *	- The driver will not register its checksum capability.
78  *	  Checksum for both TCP and UDP will be computed
79  *	  by the stack.
80  *	- The software LSO is not allowed in this case.
81  *
82  *  (4) nxge_cksum_offload > 2:
83  *	- Will be treated as it is set to 2
84  *	  (stack will compute the checksum).
85  *
86  *  (5) If the hardware bug is fixed, this workaround
87  *	needs to be updated accordingly to reflect
88  *	the new hardware revision.
89  */
90 uint32_t	nxge_cksum_offload = 0;
91 
92 /*
93  * Globals: tunable parameters (/etc/system or adb)
94  *
95  */
96 uint32_t 	nxge_rbr_size = NXGE_RBR_RBB_DEFAULT;
97 uint32_t 	nxge_rbr_spare_size = 0;
98 uint32_t 	nxge_rcr_size = NXGE_RCR_DEFAULT;
99 uint16_t	nxge_rdc_buf_offset = SW_OFFSET_NO_OFFSET;
100 uint32_t 	nxge_tx_ring_size = NXGE_TX_RING_DEFAULT;
101 boolean_t 	nxge_no_msg = B_TRUE;		/* control message display */
102 uint32_t 	nxge_no_link_notify = 0;	/* control DL_NOTIFY */
103 uint32_t 	nxge_bcopy_thresh = TX_BCOPY_MAX;
104 uint32_t 	nxge_dvma_thresh = TX_FASTDVMA_MIN;
105 uint32_t 	nxge_dma_stream_thresh = TX_STREAM_MIN;
106 uint32_t	nxge_jumbo_mtu	= TX_JUMBO_MTU;
107 nxge_tx_mode_t	nxge_tx_scheme = NXGE_USE_SERIAL;
108 
109 /* MAX LSO size */
110 #define		NXGE_LSO_MAXLEN	65535
111 uint32_t	nxge_lso_max = NXGE_LSO_MAXLEN;
112 
113 
114 /*
115  * Add tunable to reduce the amount of time spent in the
116  * ISR doing Rx Processing.
117  */
118 uint32_t nxge_max_rx_pkts = 1024;
119 
120 /*
121  * Tunables to manage the receive buffer blocks.
122  *
123  * nxge_rx_threshold_hi: copy all buffers.
124  * nxge_rx_bcopy_size_type: receive buffer block size type.
125  * nxge_rx_threshold_lo: copy only up to tunable block size type.
126  */
127 nxge_rxbuf_threshold_t nxge_rx_threshold_hi = NXGE_RX_COPY_6;
128 nxge_rxbuf_type_t nxge_rx_buf_size_type = RCR_PKTBUFSZ_0;
129 nxge_rxbuf_threshold_t nxge_rx_threshold_lo = NXGE_RX_COPY_3;
130 
131 /* Use kmem_alloc() to allocate data buffers. */
132 #if defined(__sparc)
133 uint32_t	nxge_use_kmem_alloc = 1;
134 #elif defined(__i386)
135 uint32_t	nxge_use_kmem_alloc = 0;
136 #else
137 uint32_t	nxge_use_kmem_alloc = 1;
138 #endif
139 
140 rtrace_t npi_rtracebuf;
141 
142 /*
143  * The hardware sometimes fails to allow enough time for the link partner
144  * to send an acknowledgement for packets that the hardware sent to it. The
145  * hardware resends the packets earlier than it should be in those instances.
146  * This behavior caused some switches to acknowledge the wrong packets
147  * and it triggered the fatal error.
148  * This software workaround is to set the replay timer to a value
149  * suggested by the hardware team.
150  *
151  * PCI config space replay timer register:
152  *     The following replay timeout value is 0xc
153  *     for bit 14:18.
154  */
155 #define	PCI_REPLAY_TIMEOUT_CFG_OFFSET	0xb8
156 #define	PCI_REPLAY_TIMEOUT_SHIFT	14
157 
158 uint32_t	nxge_set_replay_timer = 1;
159 uint32_t	nxge_replay_timeout = 0xc;
160 
161 /*
162  * The transmit serialization sometimes causes
163  * longer sleep before calling the driver transmit
164  * function as it sleeps longer than it should.
165  * The performace group suggests that a time wait tunable
166  * can be used to set the maximum wait time when needed
167  * and the default is set to 1 tick.
168  */
169 uint32_t	nxge_tx_serial_maxsleep = 1;
170 
171 #if	defined(sun4v)
172 /*
173  * Hypervisor N2/NIU services information.
174  */
175 /*
176  * The following is the default API supported:
177  * major 1 and minor 1.
178  *
179  * Please update the MAX_NIU_MAJORS,
180  * MAX_NIU_MINORS, and minor number supported
181  * when the newer Hypervior API interfaces
182  * are added. Also, please update nxge_hsvc_register()
183  * if needed.
184  */
185 static hsvc_info_t niu_hsvc = {
186 	HSVC_REV_1, NULL, HSVC_GROUP_NIU, NIU_MAJOR_VER,
187 	NIU_MINOR_VER, "nxge"
188 };
189 
190 static int nxge_hsvc_register(p_nxge_t);
191 #endif
192 
193 /*
194  * Function Prototypes
195  */
196 static int nxge_attach(dev_info_t *, ddi_attach_cmd_t);
197 static int nxge_detach(dev_info_t *, ddi_detach_cmd_t);
198 static void nxge_unattach(p_nxge_t);
199 static int nxge_quiesce(dev_info_t *);
200 
201 #if NXGE_PROPERTY
202 static void nxge_remove_hard_properties(p_nxge_t);
203 #endif
204 
205 /*
206  * These two functions are required by nxge_hio.c
207  */
208 extern int nxge_m_mmac_remove(void *arg, int slot);
209 extern void nxge_grp_cleanup(p_nxge_t nxge);
210 
211 static nxge_status_t nxge_setup_system_dma_pages(p_nxge_t);
212 
213 static nxge_status_t nxge_setup_mutexes(p_nxge_t);
214 static void nxge_destroy_mutexes(p_nxge_t);
215 
216 static nxge_status_t nxge_map_regs(p_nxge_t nxgep);
217 static void nxge_unmap_regs(p_nxge_t nxgep);
218 #ifdef	NXGE_DEBUG
219 static void nxge_test_map_regs(p_nxge_t nxgep);
220 #endif
221 
222 static nxge_status_t nxge_add_intrs(p_nxge_t nxgep);
223 static void nxge_remove_intrs(p_nxge_t nxgep);
224 
225 static nxge_status_t nxge_add_intrs_adv(p_nxge_t nxgep);
226 static nxge_status_t nxge_add_intrs_adv_type(p_nxge_t, uint32_t);
227 static nxge_status_t nxge_add_intrs_adv_type_fix(p_nxge_t, uint32_t);
228 static void nxge_intrs_enable(p_nxge_t nxgep);
229 static void nxge_intrs_disable(p_nxge_t nxgep);
230 
231 static void nxge_suspend(p_nxge_t);
232 static nxge_status_t nxge_resume(p_nxge_t);
233 
234 static nxge_status_t nxge_setup_dev(p_nxge_t);
235 static void nxge_destroy_dev(p_nxge_t);
236 
237 static nxge_status_t nxge_alloc_mem_pool(p_nxge_t);
238 static void nxge_free_mem_pool(p_nxge_t);
239 
240 nxge_status_t nxge_alloc_rx_mem_pool(p_nxge_t);
241 static void nxge_free_rx_mem_pool(p_nxge_t);
242 
243 nxge_status_t nxge_alloc_tx_mem_pool(p_nxge_t);
244 static void nxge_free_tx_mem_pool(p_nxge_t);
245 
246 static nxge_status_t nxge_dma_mem_alloc(p_nxge_t, dma_method_t,
247 	struct ddi_dma_attr *,
248 	size_t, ddi_device_acc_attr_t *, uint_t,
249 	p_nxge_dma_common_t);
250 
251 static void nxge_dma_mem_free(p_nxge_dma_common_t);
252 static void nxge_dma_free_rx_data_buf(p_nxge_dma_common_t);
253 
254 static nxge_status_t nxge_alloc_rx_buf_dma(p_nxge_t, uint16_t,
255 	p_nxge_dma_common_t *, size_t, size_t, uint32_t *);
256 static void nxge_free_rx_buf_dma(p_nxge_t, p_nxge_dma_common_t, uint32_t);
257 
258 static nxge_status_t nxge_alloc_rx_cntl_dma(p_nxge_t, uint16_t,
259 	p_nxge_dma_common_t *, size_t);
260 static void nxge_free_rx_cntl_dma(p_nxge_t, p_nxge_dma_common_t);
261 
262 extern nxge_status_t nxge_alloc_tx_buf_dma(p_nxge_t, uint16_t,
263 	p_nxge_dma_common_t *, size_t, size_t, uint32_t *);
264 static void nxge_free_tx_buf_dma(p_nxge_t, p_nxge_dma_common_t, uint32_t);
265 
266 extern nxge_status_t nxge_alloc_tx_cntl_dma(p_nxge_t, uint16_t,
267 	p_nxge_dma_common_t *,
268 	size_t);
269 static void nxge_free_tx_cntl_dma(p_nxge_t, p_nxge_dma_common_t);
270 
271 static int nxge_init_common_dev(p_nxge_t);
272 static void nxge_uninit_common_dev(p_nxge_t);
273 extern int nxge_param_set_mac(p_nxge_t, queue_t *, mblk_t *,
274     char *, caddr_t);
275 #if defined(sun4v)
276 extern nxge_status_t nxge_hio_rdc_enable(p_nxge_t nxgep);
277 extern nxge_status_t nxge_hio_rdc_intr_arm(p_nxge_t nxge, boolean_t arm);
278 #endif
279 
280 /*
281  * The next declarations are for the GLDv3 interface.
282  */
283 static int nxge_m_start(void *);
284 static void nxge_m_stop(void *);
285 static int nxge_m_multicst(void *, boolean_t, const uint8_t *);
286 static int nxge_m_promisc(void *, boolean_t);
287 static void nxge_m_ioctl(void *, queue_t *, mblk_t *);
288 nxge_status_t nxge_mac_register(p_nxge_t);
289 static int nxge_altmac_set(p_nxge_t nxgep, uint8_t *mac_addr,
290 	int slot, int rdctbl, boolean_t usetbl);
291 void nxge_mmac_kstat_update(p_nxge_t nxgep, int slot,
292 	boolean_t factory);
293 
294 static void nxge_m_getfactaddr(void *, uint_t, uint8_t *);
295 static	boolean_t nxge_m_getcapab(void *, mac_capab_t, void *);
296 static int nxge_m_setprop(void *, const char *, mac_prop_id_t,
297     uint_t, const void *);
298 static int nxge_m_getprop(void *, const char *, mac_prop_id_t,
299     uint_t, void *);
300 static void nxge_m_propinfo(void *, const char *, mac_prop_id_t,
301     mac_prop_info_handle_t);
302 static void nxge_priv_propinfo(const char *, mac_prop_info_handle_t);
303 static int nxge_set_priv_prop(nxge_t *, const char *, uint_t,
304     const void *);
305 static int nxge_get_priv_prop(nxge_t *, const char *, uint_t, void *);
306 static void nxge_fill_ring(void *, mac_ring_type_t, const int, const int,
307     mac_ring_info_t *, mac_ring_handle_t);
308 static void nxge_group_add_ring(mac_group_driver_t, mac_ring_driver_t,
309     mac_ring_type_t);
310 static void nxge_group_rem_ring(mac_group_driver_t, mac_ring_driver_t,
311     mac_ring_type_t);
312 
313 static void nxge_niu_peu_reset(p_nxge_t nxgep);
314 static void nxge_set_pci_replay_timeout(nxge_t *);
315 
316 char *nxge_priv_props[] = {
317 	"_adv_10gfdx_cap",
318 	"_adv_pause_cap",
319 	"_function_number",
320 	"_fw_version",
321 	"_port_mode",
322 	"_hot_swap_phy",
323 	"_rxdma_intr_time",
324 	"_rxdma_intr_pkts",
325 	"_class_opt_ipv4_tcp",
326 	"_class_opt_ipv4_udp",
327 	"_class_opt_ipv4_ah",
328 	"_class_opt_ipv4_sctp",
329 	"_class_opt_ipv6_tcp",
330 	"_class_opt_ipv6_udp",
331 	"_class_opt_ipv6_ah",
332 	"_class_opt_ipv6_sctp",
333 	"_soft_lso_enable",
334 	NULL
335 };
336 
337 #define	NXGE_NEPTUNE_MAGIC	0x4E584745UL
338 #define	MAX_DUMP_SZ 256
339 
340 #define	NXGE_M_CALLBACK_FLAGS	\
341 	(MC_IOCTL | MC_GETCAPAB | MC_SETPROP | MC_GETPROP | MC_PROPINFO)
342 
343 mac_callbacks_t nxge_m_callbacks = {
344 	NXGE_M_CALLBACK_FLAGS,
345 	nxge_m_stat,
346 	nxge_m_start,
347 	nxge_m_stop,
348 	nxge_m_promisc,
349 	nxge_m_multicst,
350 	NULL,
351 	NULL,
352 	NULL,
353 	nxge_m_ioctl,
354 	nxge_m_getcapab,
355 	NULL,
356 	NULL,
357 	nxge_m_setprop,
358 	nxge_m_getprop,
359 	nxge_m_propinfo
360 };
361 
362 void
363 nxge_err_inject(p_nxge_t, queue_t *, mblk_t *);
364 
365 /* PSARC/2007/453 MSI-X interrupt limit override. */
366 #define	NXGE_MSIX_REQUEST_10G	8
367 #define	NXGE_MSIX_REQUEST_1G	2
368 static int nxge_create_msi_property(p_nxge_t);
369 /*
370  * For applications that care about the
371  * latency, it was requested by PAE and the
372  * customers that the driver has tunables that
373  * allow the user to tune it to a higher number
374  * interrupts to spread the interrupts among
375  * multiple channels. The DDI framework limits
376  * the maximum number of MSI-X resources to allocate
377  * to 8 (ddi_msix_alloc_limit). If more than 8
378  * is set, ddi_msix_alloc_limit must be set accordingly.
379  * The default number of MSI interrupts are set to
380  * 8 for 10G and 2 for 1G link.
381  */
382 #define	NXGE_MSIX_MAX_ALLOWED	32
383 uint32_t nxge_msix_10g_intrs = NXGE_MSIX_REQUEST_10G;
384 uint32_t nxge_msix_1g_intrs = NXGE_MSIX_REQUEST_1G;
385 
386 /*
387  * These global variables control the message
388  * output.
389  */
390 out_dbgmsg_t nxge_dbgmsg_out = DBG_CONSOLE | STR_LOG;
391 uint64_t nxge_debug_level;
392 
393 /*
394  * This list contains the instance structures for the Neptune
395  * devices present in the system. The lock exists to guarantee
396  * mutually exclusive access to the list.
397  */
398 void 			*nxge_list = NULL;
399 void			*nxge_hw_list = NULL;
400 nxge_os_mutex_t 	nxge_common_lock;
401 nxge_os_mutex_t 	nxgedebuglock;
402 
403 extern uint64_t 	npi_debug_level;
404 
405 extern nxge_status_t	nxge_ldgv_init(p_nxge_t, int *, int *);
406 extern nxge_status_t	nxge_ldgv_init_n2(p_nxge_t, int *, int *);
407 extern nxge_status_t	nxge_ldgv_uninit(p_nxge_t);
408 extern nxge_status_t	nxge_intr_ldgv_init(p_nxge_t);
409 extern void		nxge_fm_init(p_nxge_t,
410 					ddi_device_acc_attr_t *,
411 					ddi_dma_attr_t *);
412 extern void		nxge_fm_fini(p_nxge_t);
413 extern npi_status_t	npi_mac_altaddr_disable(npi_handle_t, uint8_t, uint8_t);
414 
415 /*
416  * Count used to maintain the number of buffers being used
417  * by Neptune instances and loaned up to the upper layers.
418  */
419 uint32_t nxge_mblks_pending = 0;
420 
421 /*
422  * Device register access attributes for PIO.
423  */
424 static ddi_device_acc_attr_t nxge_dev_reg_acc_attr = {
425 	DDI_DEVICE_ATTR_V1,
426 	DDI_STRUCTURE_LE_ACC,
427 	DDI_STRICTORDER_ACC,
428 	DDI_DEFAULT_ACC
429 };
430 
431 /*
432  * Device descriptor access attributes for DMA.
433  */
434 static ddi_device_acc_attr_t nxge_dev_desc_dma_acc_attr = {
435 	DDI_DEVICE_ATTR_V0,
436 	DDI_STRUCTURE_LE_ACC,
437 	DDI_STRICTORDER_ACC
438 };
439 
440 /*
441  * Device buffer access attributes for DMA.
442  */
443 static ddi_device_acc_attr_t nxge_dev_buf_dma_acc_attr = {
444 	DDI_DEVICE_ATTR_V0,
445 	DDI_STRUCTURE_BE_ACC,
446 	DDI_STRICTORDER_ACC
447 };
448 
449 ddi_dma_attr_t nxge_desc_dma_attr = {
450 	DMA_ATTR_V0,		/* version number. */
451 	0,			/* low address */
452 	0xffffffffffffffff,	/* high address */
453 	0xffffffffffffffff,	/* address counter max */
454 #ifndef NIU_PA_WORKAROUND
455 	0x100000,		/* alignment */
456 #else
457 	0x2000,
458 #endif
459 	0xfc00fc,		/* dlim_burstsizes */
460 	0x1,			/* minimum transfer size */
461 	0xffffffffffffffff,	/* maximum transfer size */
462 	0xffffffffffffffff,	/* maximum segment size */
463 	1,			/* scatter/gather list length */
464 	(unsigned int) 1,	/* granularity */
465 	0			/* attribute flags */
466 };
467 
468 ddi_dma_attr_t nxge_tx_dma_attr = {
469 	DMA_ATTR_V0,		/* version number. */
470 	0,			/* low address */
471 	0xffffffffffffffff,	/* high address */
472 	0xffffffffffffffff,	/* address counter max */
473 #if defined(_BIG_ENDIAN)
474 	0x2000,			/* alignment */
475 #else
476 	0x1000,			/* alignment */
477 #endif
478 	0xfc00fc,		/* dlim_burstsizes */
479 	0x1,			/* minimum transfer size */
480 	0xffffffffffffffff,	/* maximum transfer size */
481 	0xffffffffffffffff,	/* maximum segment size */
482 	5,			/* scatter/gather list length */
483 	(unsigned int) 1,	/* granularity */
484 	0			/* attribute flags */
485 };
486 
487 ddi_dma_attr_t nxge_rx_dma_attr = {
488 	DMA_ATTR_V0,		/* version number. */
489 	0,			/* low address */
490 	0xffffffffffffffff,	/* high address */
491 	0xffffffffffffffff,	/* address counter max */
492 	0x2000,			/* alignment */
493 	0xfc00fc,		/* dlim_burstsizes */
494 	0x1,			/* minimum transfer size */
495 	0xffffffffffffffff,	/* maximum transfer size */
496 	0xffffffffffffffff,	/* maximum segment size */
497 	1,			/* scatter/gather list length */
498 	(unsigned int) 1,	/* granularity */
499 	DDI_DMA_RELAXED_ORDERING /* attribute flags */
500 };
501 
502 ddi_dma_lim_t nxge_dma_limits = {
503 	(uint_t)0,		/* dlim_addr_lo */
504 	(uint_t)0xffffffff,	/* dlim_addr_hi */
505 	(uint_t)0xffffffff,	/* dlim_cntr_max */
506 	(uint_t)0xfc00fc,	/* dlim_burstsizes for 32 and 64 bit xfers */
507 	0x1,			/* dlim_minxfer */
508 	1024			/* dlim_speed */
509 };
510 
511 dma_method_t nxge_force_dma = DVMA;
512 
513 /*
514  * dma chunk sizes.
515  *
516  * Try to allocate the largest possible size
517  * so that fewer number of dma chunks would be managed
518  */
519 #ifdef NIU_PA_WORKAROUND
520 size_t alloc_sizes [] = {0x2000};
521 #else
522 size_t alloc_sizes [] = {0x1000, 0x2000, 0x4000, 0x8000,
523 		0x10000, 0x20000, 0x40000, 0x80000,
524 		0x100000, 0x200000, 0x400000, 0x800000,
525 		0x1000000, 0x2000000, 0x4000000};
526 #endif
527 
528 /*
529  * Translate "dev_t" to a pointer to the associated "dev_info_t".
530  */
531 
532 extern void nxge_get_environs(nxge_t *);
533 
534 static int
535 nxge_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
536 {
537 	p_nxge_t	nxgep = NULL;
538 	int		instance;
539 	int		status = DDI_SUCCESS;
540 	uint8_t		portn;
541 	nxge_mmac_t	*mmac_info;
542 
543 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "==> nxge_attach"));
544 
545 	/*
546 	 * Get the device instance since we'll need to setup
547 	 * or retrieve a soft state for this instance.
548 	 */
549 	instance = ddi_get_instance(dip);
550 
551 	switch (cmd) {
552 	case DDI_ATTACH:
553 		NXGE_DEBUG_MSG((nxgep, DDI_CTL, "doing DDI_ATTACH"));
554 		break;
555 
556 	case DDI_RESUME:
557 		NXGE_DEBUG_MSG((nxgep, DDI_CTL, "doing DDI_RESUME"));
558 		nxgep = (p_nxge_t)ddi_get_soft_state(nxge_list, instance);
559 		if (nxgep == NULL) {
560 			status = DDI_FAILURE;
561 			break;
562 		}
563 		if (nxgep->dip != dip) {
564 			status = DDI_FAILURE;
565 			break;
566 		}
567 		if (nxgep->suspended == DDI_PM_SUSPEND) {
568 			status = ddi_dev_is_needed(nxgep->dip, 0, 1);
569 		} else {
570 			status = nxge_resume(nxgep);
571 		}
572 		goto nxge_attach_exit;
573 
574 	case DDI_PM_RESUME:
575 		NXGE_DEBUG_MSG((nxgep, DDI_CTL, "doing DDI_PM_RESUME"));
576 		nxgep = (p_nxge_t)ddi_get_soft_state(nxge_list, instance);
577 		if (nxgep == NULL) {
578 			status = DDI_FAILURE;
579 			break;
580 		}
581 		if (nxgep->dip != dip) {
582 			status = DDI_FAILURE;
583 			break;
584 		}
585 		status = nxge_resume(nxgep);
586 		goto nxge_attach_exit;
587 
588 	default:
589 		NXGE_DEBUG_MSG((nxgep, DDI_CTL, "doing unknown"));
590 		status = DDI_FAILURE;
591 		goto nxge_attach_exit;
592 	}
593 
594 
595 	if (ddi_soft_state_zalloc(nxge_list, instance) == DDI_FAILURE) {
596 		status = DDI_FAILURE;
597 		goto nxge_attach_exit;
598 	}
599 
600 	nxgep = ddi_get_soft_state(nxge_list, instance);
601 	if (nxgep == NULL) {
602 		status = NXGE_ERROR;
603 		goto nxge_attach_fail2;
604 	}
605 
606 	nxgep->nxge_magic = NXGE_MAGIC;
607 
608 	nxgep->drv_state = 0;
609 	nxgep->dip = dip;
610 	nxgep->instance = instance;
611 	nxgep->p_dip = ddi_get_parent(dip);
612 	nxgep->nxge_debug_level = nxge_debug_level;
613 	npi_debug_level = nxge_debug_level;
614 
615 	/* Are we a guest running in a Hybrid I/O environment? */
616 	nxge_get_environs(nxgep);
617 
618 	status = nxge_map_regs(nxgep);
619 
620 	if (status != NXGE_OK) {
621 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL, "nxge_map_regs failed"));
622 		goto nxge_attach_fail3;
623 	}
624 
625 	nxge_fm_init(nxgep, &nxge_dev_reg_acc_attr, &nxge_rx_dma_attr);
626 
627 	/* Create & initialize the per-Neptune data structure */
628 	/* (even if we're a guest). */
629 	status = nxge_init_common_dev(nxgep);
630 	if (status != NXGE_OK) {
631 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
632 		    "nxge_init_common_dev failed"));
633 		goto nxge_attach_fail4;
634 	}
635 
636 	/*
637 	 * Software workaround: set the replay timer.
638 	 */
639 	if (nxgep->niu_type != N2_NIU) {
640 		nxge_set_pci_replay_timeout(nxgep);
641 	}
642 
643 #if defined(sun4v)
644 	/* This is required by nxge_hio_init(), which follows. */
645 	if ((status = nxge_hsvc_register(nxgep)) != DDI_SUCCESS)
646 		goto nxge_attach_fail4;
647 #endif
648 
649 	if ((status = nxge_hio_init(nxgep)) != NXGE_OK) {
650 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
651 		    "nxge_hio_init failed"));
652 		goto nxge_attach_fail4;
653 	}
654 
655 	if (nxgep->niu_type == NEPTUNE_2_10GF) {
656 		if (nxgep->function_num > 1) {
657 			NXGE_DEBUG_MSG((nxgep, DDI_CTL, "Unsupported"
658 			    " function %d. Only functions 0 and 1 are "
659 			    "supported for this card.", nxgep->function_num));
660 			status = NXGE_ERROR;
661 			goto nxge_attach_fail4;
662 		}
663 	}
664 
665 	if (isLDOMguest(nxgep)) {
666 		/*
667 		 * Use the function number here.
668 		 */
669 		nxgep->mac.portnum = nxgep->function_num;
670 		nxgep->mac.porttype = PORT_TYPE_LOGICAL;
671 
672 		/* XXX We'll set the MAC address counts to 1 for now. */
673 		mmac_info = &nxgep->nxge_mmac_info;
674 		mmac_info->num_mmac = 1;
675 		mmac_info->naddrfree = 1;
676 	} else {
677 		portn = NXGE_GET_PORT_NUM(nxgep->function_num);
678 		nxgep->mac.portnum = portn;
679 		if ((portn == 0) || (portn == 1))
680 			nxgep->mac.porttype = PORT_TYPE_XMAC;
681 		else
682 			nxgep->mac.porttype = PORT_TYPE_BMAC;
683 		/*
684 		 * Neptune has 4 ports, the first 2 ports use XMAC (10G MAC)
685 		 * internally, the rest 2 ports use BMAC (1G "Big" MAC).
686 		 * The two types of MACs have different characterizations.
687 		 */
688 		mmac_info = &nxgep->nxge_mmac_info;
689 		if (nxgep->function_num < 2) {
690 			mmac_info->num_mmac = XMAC_MAX_ALT_ADDR_ENTRY;
691 			mmac_info->naddrfree = XMAC_MAX_ALT_ADDR_ENTRY;
692 		} else {
693 			mmac_info->num_mmac = BMAC_MAX_ALT_ADDR_ENTRY;
694 			mmac_info->naddrfree = BMAC_MAX_ALT_ADDR_ENTRY;
695 		}
696 	}
697 	/*
698 	 * Setup the Ndd parameters for the this instance.
699 	 */
700 	nxge_init_param(nxgep);
701 
702 	/*
703 	 * Setup Register Tracing Buffer.
704 	 */
705 	npi_rtrace_buf_init((rtrace_t *)&npi_rtracebuf);
706 
707 	/* init stats ptr */
708 	nxge_init_statsp(nxgep);
709 
710 	/*
711 	 * Copy the vpd info from eeprom to a local data
712 	 * structure, and then check its validity.
713 	 */
714 	if (!isLDOMguest(nxgep)) {
715 		int *regp;
716 		uint_t reglen;
717 		int rv;
718 
719 		nxge_vpd_info_get(nxgep);
720 
721 		/* Find the NIU config handle. */
722 		rv = ddi_prop_lookup_int_array(DDI_DEV_T_ANY,
723 		    ddi_get_parent(nxgep->dip), DDI_PROP_DONTPASS,
724 		    "reg", &regp, &reglen);
725 
726 		if (rv != DDI_PROP_SUCCESS) {
727 			goto nxge_attach_fail5;
728 		}
729 		/*
730 		 * The address_hi, that is the first int, in the reg
731 		 * property consists of config handle, but need to remove
732 		 * the bits 28-31 which are OBP specific info.
733 		 */
734 		nxgep->niu_cfg_hdl = (*regp) & 0xFFFFFFF;
735 		ddi_prop_free(regp);
736 	}
737 
738 	/*
739 	 * Set the defaults for the MTU size.
740 	 */
741 	nxge_hw_id_init(nxgep);
742 
743 	if (isLDOMguest(nxgep)) {
744 		uchar_t *prop_val;
745 		uint_t prop_len;
746 		uint32_t max_frame_size;
747 
748 		extern void nxge_get_logical_props(p_nxge_t);
749 
750 		nxgep->statsp->mac_stats.xcvr_inuse = LOGICAL_XCVR;
751 		nxgep->mac.portmode = PORT_LOGICAL;
752 		(void) ddi_prop_update_string(DDI_DEV_T_NONE, nxgep->dip,
753 		    "phy-type", "virtual transceiver");
754 
755 		nxgep->nports = 1;
756 		nxgep->board_ver = 0;	/* XXX What? */
757 
758 		/*
759 		 * local-mac-address property gives us info on which
760 		 * specific MAC address the Hybrid resource is associated
761 		 * with.
762 		 */
763 		if (ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, nxgep->dip, 0,
764 		    "local-mac-address", &prop_val,
765 		    &prop_len) != DDI_PROP_SUCCESS) {
766 			goto nxge_attach_fail5;
767 		}
768 		if (prop_len !=  ETHERADDRL) {
769 			ddi_prop_free(prop_val);
770 			goto nxge_attach_fail5;
771 		}
772 		ether_copy(prop_val, nxgep->hio_mac_addr);
773 		ddi_prop_free(prop_val);
774 		nxge_get_logical_props(nxgep);
775 
776 		/*
777 		 * Enable Jumbo property based on the "max-frame-size"
778 		 * property value.
779 		 */
780 		max_frame_size = ddi_prop_get_int(DDI_DEV_T_ANY,
781 		    nxgep->dip, DDI_PROP_DONTPASS | DDI_PROP_NOTPROM,
782 		    "max-frame-size", NXGE_MTU_DEFAULT_MAX);
783 		if ((max_frame_size > NXGE_MTU_DEFAULT_MAX) &&
784 		    (max_frame_size <= TX_JUMBO_MTU)) {
785 			nxgep->mac.is_jumbo = B_TRUE;
786 			nxgep->mac.maxframesize = (uint16_t)max_frame_size;
787 			nxgep->mac.default_mtu = nxgep->mac.maxframesize -
788 			    NXGE_EHEADER_VLAN_CRC;
789 		}
790 	} else {
791 		status = nxge_xcvr_find(nxgep);
792 
793 		if (status != NXGE_OK) {
794 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL, "nxge_attach: "
795 			    " Couldn't determine card type"
796 			    " .... exit "));
797 			goto nxge_attach_fail5;
798 		}
799 
800 		status = nxge_get_config_properties(nxgep);
801 
802 		if (status != NXGE_OK) {
803 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
804 			    "get_hw create failed"));
805 			goto nxge_attach_fail;
806 		}
807 	}
808 
809 	/*
810 	 * Setup the Kstats for the driver.
811 	 */
812 	nxge_setup_kstats(nxgep);
813 
814 	if (!isLDOMguest(nxgep))
815 		nxge_setup_param(nxgep);
816 
817 	status = nxge_setup_system_dma_pages(nxgep);
818 	if (status != NXGE_OK) {
819 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL, "set dma page failed"));
820 		goto nxge_attach_fail;
821 	}
822 
823 
824 	if (!isLDOMguest(nxgep))
825 		nxge_hw_init_niu_common(nxgep);
826 
827 	status = nxge_setup_mutexes(nxgep);
828 	if (status != NXGE_OK) {
829 		NXGE_DEBUG_MSG((nxgep, DDI_CTL, "set mutex failed"));
830 		goto nxge_attach_fail;
831 	}
832 
833 #if defined(sun4v)
834 	if (isLDOMguest(nxgep)) {
835 		/* Find our VR & channel sets. */
836 		status = nxge_hio_vr_add(nxgep);
837 		if (status != DDI_SUCCESS) {
838 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
839 			    "nxge_hio_vr_add failed"));
840 			(void) hsvc_unregister(&nxgep->niu_hsvc);
841 			nxgep->niu_hsvc_available = B_FALSE;
842 			goto nxge_attach_fail;
843 		}
844 		goto nxge_attach_exit;
845 	}
846 #endif
847 
848 	status = nxge_setup_dev(nxgep);
849 	if (status != DDI_SUCCESS) {
850 		NXGE_DEBUG_MSG((nxgep, DDI_CTL, "set dev failed"));
851 		goto nxge_attach_fail;
852 	}
853 
854 	status = nxge_add_intrs(nxgep);
855 	if (status != DDI_SUCCESS) {
856 		NXGE_DEBUG_MSG((nxgep, DDI_CTL, "add_intr failed"));
857 		goto nxge_attach_fail;
858 	}
859 
860 	/* If a guest, register with vio_net instead. */
861 	if ((status = nxge_mac_register(nxgep)) != NXGE_OK) {
862 		NXGE_DEBUG_MSG((nxgep, DDI_CTL,
863 		    "unable to register to mac layer (%d)", status));
864 		goto nxge_attach_fail;
865 	}
866 
867 	mac_link_update(nxgep->mach, LINK_STATE_UNKNOWN);
868 
869 	NXGE_DEBUG_MSG((nxgep, DDI_CTL,
870 	    "registered to mac (instance %d)", instance));
871 
872 	/* nxge_link_monitor calls xcvr.check_link recursively */
873 	(void) nxge_link_monitor(nxgep, LINK_MONITOR_START);
874 
875 	goto nxge_attach_exit;
876 
877 nxge_attach_fail:
878 	nxge_unattach(nxgep);
879 	goto nxge_attach_fail1;
880 
881 nxge_attach_fail5:
882 	/*
883 	 * Tear down the ndd parameters setup.
884 	 */
885 	nxge_destroy_param(nxgep);
886 
887 	/*
888 	 * Tear down the kstat setup.
889 	 */
890 	nxge_destroy_kstats(nxgep);
891 
892 nxge_attach_fail4:
893 	if (nxgep->nxge_hw_p) {
894 		nxge_uninit_common_dev(nxgep);
895 		nxgep->nxge_hw_p = NULL;
896 	}
897 
898 nxge_attach_fail3:
899 	/*
900 	 * Unmap the register setup.
901 	 */
902 	nxge_unmap_regs(nxgep);
903 
904 	nxge_fm_fini(nxgep);
905 
906 nxge_attach_fail2:
907 	ddi_soft_state_free(nxge_list, nxgep->instance);
908 
909 nxge_attach_fail1:
910 	if (status != NXGE_OK)
911 		status = (NXGE_ERROR | NXGE_DDI_FAILED);
912 	nxgep = NULL;
913 
914 nxge_attach_exit:
915 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "<== nxge_attach status = 0x%08x",
916 	    status));
917 
918 	return (status);
919 }
920 
921 static int
922 nxge_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
923 {
924 	int 		status = DDI_SUCCESS;
925 	int 		instance;
926 	p_nxge_t 	nxgep = NULL;
927 
928 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "==> nxge_detach"));
929 	instance = ddi_get_instance(dip);
930 	nxgep = ddi_get_soft_state(nxge_list, instance);
931 	if (nxgep == NULL) {
932 		status = DDI_FAILURE;
933 		goto nxge_detach_exit;
934 	}
935 
936 	switch (cmd) {
937 	case DDI_DETACH:
938 		NXGE_DEBUG_MSG((nxgep, DDI_CTL, "doing DDI_DETACH"));
939 		break;
940 
941 	case DDI_PM_SUSPEND:
942 		NXGE_DEBUG_MSG((nxgep, DDI_CTL, "doing DDI_PM_SUSPEND"));
943 		nxgep->suspended = DDI_PM_SUSPEND;
944 		nxge_suspend(nxgep);
945 		break;
946 
947 	case DDI_SUSPEND:
948 		NXGE_DEBUG_MSG((nxgep, DDI_CTL, "doing DDI_SUSPEND"));
949 		if (nxgep->suspended != DDI_PM_SUSPEND) {
950 			nxgep->suspended = DDI_SUSPEND;
951 			nxge_suspend(nxgep);
952 		}
953 		break;
954 
955 	default:
956 		status = DDI_FAILURE;
957 	}
958 
959 	if (cmd != DDI_DETACH)
960 		goto nxge_detach_exit;
961 
962 	/*
963 	 * Stop the xcvr polling.
964 	 */
965 	nxgep->suspended = cmd;
966 
967 	(void) nxge_link_monitor(nxgep, LINK_MONITOR_STOP);
968 
969 	if (nxgep->mach && (status = mac_unregister(nxgep->mach)) != 0) {
970 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
971 		    "<== nxge_detach status = 0x%08X", status));
972 		return (DDI_FAILURE);
973 	}
974 
975 	NXGE_DEBUG_MSG((nxgep, DDI_CTL,
976 	    "<== nxge_detach (mac_unregister) status = 0x%08X", status));
977 
978 	nxge_unattach(nxgep);
979 	nxgep = NULL;
980 
981 nxge_detach_exit:
982 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "<== nxge_detach status = 0x%08X",
983 	    status));
984 
985 	return (status);
986 }
987 
988 static void
989 nxge_unattach(p_nxge_t nxgep)
990 {
991 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "==> nxge_unattach"));
992 
993 	if (nxgep == NULL || nxgep->dev_regs == NULL) {
994 		return;
995 	}
996 
997 	nxgep->nxge_magic = 0;
998 
999 	if (nxgep->nxge_timerid) {
1000 		nxge_stop_timer(nxgep, nxgep->nxge_timerid);
1001 		nxgep->nxge_timerid = 0;
1002 	}
1003 
1004 	/*
1005 	 * If this flag is set, it will affect the Neptune
1006 	 * only.
1007 	 */
1008 	if ((nxgep->niu_type != N2_NIU) && nxge_peu_reset_enable) {
1009 		nxge_niu_peu_reset(nxgep);
1010 	}
1011 
1012 #if	defined(sun4v)
1013 	if (isLDOMguest(nxgep)) {
1014 		(void) nxge_hio_vr_release(nxgep);
1015 	}
1016 #endif
1017 
1018 	if (nxgep->nxge_hw_p) {
1019 		nxge_uninit_common_dev(nxgep);
1020 		nxgep->nxge_hw_p = NULL;
1021 	}
1022 
1023 #if	defined(sun4v)
1024 	if (nxgep->niu_type == N2_NIU && nxgep->niu_hsvc_available == B_TRUE) {
1025 		(void) hsvc_unregister(&nxgep->niu_hsvc);
1026 		nxgep->niu_hsvc_available = B_FALSE;
1027 	}
1028 #endif
1029 	/*
1030 	 * Stop any further interrupts.
1031 	 */
1032 	nxge_remove_intrs(nxgep);
1033 
1034 	/*
1035 	 * Stop the device and free resources.
1036 	 */
1037 	if (!isLDOMguest(nxgep)) {
1038 		nxge_destroy_dev(nxgep);
1039 	}
1040 
1041 	/*
1042 	 * Tear down the ndd parameters setup.
1043 	 */
1044 	nxge_destroy_param(nxgep);
1045 
1046 	/*
1047 	 * Tear down the kstat setup.
1048 	 */
1049 	nxge_destroy_kstats(nxgep);
1050 
1051 	/*
1052 	 * Destroy all mutexes.
1053 	 */
1054 	nxge_destroy_mutexes(nxgep);
1055 
1056 	/*
1057 	 * Remove the list of ndd parameters which
1058 	 * were setup during attach.
1059 	 */
1060 	if (nxgep->dip) {
1061 		NXGE_DEBUG_MSG((nxgep, OBP_CTL,
1062 		    " nxge_unattach: remove all properties"));
1063 
1064 		(void) ddi_prop_remove_all(nxgep->dip);
1065 	}
1066 
1067 #if NXGE_PROPERTY
1068 	nxge_remove_hard_properties(nxgep);
1069 #endif
1070 
1071 	/*
1072 	 * Unmap the register setup.
1073 	 */
1074 	nxge_unmap_regs(nxgep);
1075 
1076 	nxge_fm_fini(nxgep);
1077 
1078 	ddi_soft_state_free(nxge_list, nxgep->instance);
1079 
1080 	NXGE_DEBUG_MSG((NULL, DDI_CTL, "<== nxge_unattach"));
1081 }
1082 
1083 #if defined(sun4v)
1084 int
1085 nxge_hsvc_register(nxge_t *nxgep)
1086 {
1087 	nxge_status_t status;
1088 	int i, j;
1089 
1090 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "==> nxge_hsvc_register"));
1091 	if (nxgep->niu_type != N2_NIU) {
1092 		NXGE_DEBUG_MSG((nxgep, DDI_CTL, "<== nxge_hsvc_register"));
1093 		return (DDI_SUCCESS);
1094 	}
1095 
1096 	/*
1097 	 * Currently, the NIU Hypervisor API supports two major versions:
1098 	 * version 1 and 2.
1099 	 * If Hypervisor introduces a higher major or minor version,
1100 	 * please update NIU_MAJOR_HI and NIU_MINOR_HI accordingly.
1101 	 */
1102 	nxgep->niu_hsvc_available = B_FALSE;
1103 	bcopy(&niu_hsvc, &nxgep->niu_hsvc,
1104 	    sizeof (hsvc_info_t));
1105 
1106 	for (i = NIU_MAJOR_HI; i > 0; i--) {
1107 		nxgep->niu_hsvc.hsvc_major = i;
1108 		for (j = NIU_MINOR_HI; j >= 0; j--) {
1109 			nxgep->niu_hsvc.hsvc_minor = j;
1110 			NXGE_DEBUG_MSG((nxgep, DDI_CTL,
1111 			    "nxge_hsvc_register: %s: negotiating "
1112 			    "hypervisor services revision %d "
1113 			    "group: 0x%lx major: 0x%lx "
1114 			    "minor: 0x%lx",
1115 			    nxgep->niu_hsvc.hsvc_modname,
1116 			    nxgep->niu_hsvc.hsvc_rev,
1117 			    nxgep->niu_hsvc.hsvc_group,
1118 			    nxgep->niu_hsvc.hsvc_major,
1119 			    nxgep->niu_hsvc.hsvc_minor,
1120 			    nxgep->niu_min_ver));
1121 
1122 			if ((status = hsvc_register(&nxgep->niu_hsvc,
1123 			    &nxgep->niu_min_ver)) == 0) {
1124 				/* Use the supported minor */
1125 				nxgep->niu_hsvc.hsvc_minor = nxgep->niu_min_ver;
1126 				NXGE_DEBUG_MSG((nxgep, DDI_CTL,
1127 				    "nxge_hsvc_register: %s: negotiated "
1128 				    "hypervisor services revision %d "
1129 				    "group: 0x%lx major: 0x%lx "
1130 				    "minor: 0x%lx (niu_min_ver 0x%lx)",
1131 				    nxgep->niu_hsvc.hsvc_modname,
1132 				    nxgep->niu_hsvc.hsvc_rev,
1133 				    nxgep->niu_hsvc.hsvc_group,
1134 				    nxgep->niu_hsvc.hsvc_major,
1135 				    nxgep->niu_hsvc.hsvc_minor,
1136 				    nxgep->niu_min_ver));
1137 
1138 				nxgep->niu_hsvc_available = B_TRUE;
1139 				NXGE_DEBUG_MSG((nxgep, DDI_CTL,
1140 				    "<== nxge_hsvc_register: "
1141 				    "NIU Hypervisor service enabled"));
1142 				return (DDI_SUCCESS);
1143 			}
1144 
1145 			NXGE_DEBUG_MSG((nxgep, DDI_CTL,
1146 			    "nxge_hsvc_register: %s: negotiated failed - "
1147 			    "try lower major number "
1148 			    "hypervisor services revision %d "
1149 			    "group: 0x%lx major: 0x%lx minor: 0x%lx "
1150 			    "errno: %d",
1151 			    nxgep->niu_hsvc.hsvc_modname,
1152 			    nxgep->niu_hsvc.hsvc_rev,
1153 			    nxgep->niu_hsvc.hsvc_group,
1154 			    nxgep->niu_hsvc.hsvc_major,
1155 			    nxgep->niu_hsvc.hsvc_minor, status));
1156 		}
1157 	}
1158 
1159 	NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
1160 	    "nxge_hsvc_register: %s: cannot negotiate "
1161 	    "hypervisor services revision %d group: 0x%lx "
1162 	    "major: 0x%lx minor: 0x%lx errno: %d",
1163 	    niu_hsvc.hsvc_modname, niu_hsvc.hsvc_rev,
1164 	    niu_hsvc.hsvc_group, niu_hsvc.hsvc_major,
1165 	    niu_hsvc.hsvc_minor, status));
1166 
1167 	NXGE_DEBUG_MSG((nxgep, DDI_CTL,
1168 	    "<== nxge_hsvc_register: Register to NIU Hypervisor failed"));
1169 
1170 	return (DDI_FAILURE);
1171 }
1172 #endif
1173 
1174 static char n2_siu_name[] = "niu";
1175 
1176 static nxge_status_t
1177 nxge_map_regs(p_nxge_t nxgep)
1178 {
1179 	int		ddi_status = DDI_SUCCESS;
1180 	p_dev_regs_t 	dev_regs;
1181 	char		buf[MAXPATHLEN + 1];
1182 	char 		*devname;
1183 #ifdef	NXGE_DEBUG
1184 	char 		*sysname;
1185 #endif
1186 	off_t		regsize;
1187 	nxge_status_t	status = NXGE_OK;
1188 #if !defined(_BIG_ENDIAN)
1189 	off_t pci_offset;
1190 	uint16_t pcie_devctl;
1191 #endif
1192 
1193 	if (isLDOMguest(nxgep)) {
1194 		return (nxge_guest_regs_map(nxgep));
1195 	}
1196 
1197 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "==> nxge_map_regs"));
1198 	nxgep->dev_regs = NULL;
1199 	dev_regs = KMEM_ZALLOC(sizeof (dev_regs_t), KM_SLEEP);
1200 	dev_regs->nxge_regh = NULL;
1201 	dev_regs->nxge_pciregh = NULL;
1202 	dev_regs->nxge_msix_regh = NULL;
1203 	dev_regs->nxge_vir_regh = NULL;
1204 	dev_regs->nxge_vir2_regh = NULL;
1205 	nxgep->niu_type = NIU_TYPE_NONE;
1206 
1207 	devname = ddi_pathname(nxgep->dip, buf);
1208 	ASSERT(strlen(devname) > 0);
1209 	NXGE_DEBUG_MSG((nxgep, DDI_CTL,
1210 	    "nxge_map_regs: pathname devname %s", devname));
1211 
1212 	/*
1213 	 * The driver is running on a N2-NIU system if devname is something
1214 	 * like "/niu@80/network@0"
1215 	 */
1216 	if (strstr(devname, n2_siu_name)) {
1217 		/* N2/NIU */
1218 		nxgep->niu_type = N2_NIU;
1219 		NXGE_DEBUG_MSG((nxgep, DDI_CTL,
1220 		    "nxge_map_regs: N2/NIU devname %s", devname));
1221 		/*
1222 		 * Get function number:
1223 		 *  - N2/NIU: "/niu@80/network@0" and "/niu@80/network@1"
1224 		 */
1225 		nxgep->function_num =
1226 		    (devname[strlen(devname) -1] == '1' ? 1 : 0);
1227 		NXGE_DEBUG_MSG((nxgep, DDI_CTL,
1228 		    "nxge_map_regs: N2/NIU function number %d",
1229 		    nxgep->function_num));
1230 	} else {
1231 		int		*prop_val;
1232 		uint_t 		prop_len;
1233 		uint8_t 	func_num;
1234 
1235 		if (ddi_prop_lookup_int_array(DDI_DEV_T_ANY, nxgep->dip,
1236 		    0, "reg",
1237 		    &prop_val, &prop_len) != DDI_PROP_SUCCESS) {
1238 			NXGE_DEBUG_MSG((nxgep, VPD_CTL,
1239 			    "Reg property not found"));
1240 			ddi_status = DDI_FAILURE;
1241 			goto nxge_map_regs_fail0;
1242 
1243 		} else {
1244 			func_num = (prop_val[0] >> 8) & 0x7;
1245 			NXGE_DEBUG_MSG((nxgep, DDI_CTL,
1246 			    "Reg property found: fun # %d",
1247 			    func_num));
1248 			nxgep->function_num = func_num;
1249 			if (isLDOMguest(nxgep)) {
1250 				nxgep->function_num /= 2;
1251 				return (NXGE_OK);
1252 			}
1253 			ddi_prop_free(prop_val);
1254 		}
1255 	}
1256 
1257 	switch (nxgep->niu_type) {
1258 	default:
1259 		(void) ddi_dev_regsize(nxgep->dip, 0, &regsize);
1260 		NXGE_DEBUG_MSG((nxgep, DDI_CTL,
1261 		    "nxge_map_regs: pci config size 0x%x", regsize));
1262 
1263 		ddi_status = ddi_regs_map_setup(nxgep->dip, 0,
1264 		    (caddr_t *)&(dev_regs->nxge_pciregp), 0, 0,
1265 		    &nxge_dev_reg_acc_attr, &dev_regs->nxge_pciregh);
1266 		if (ddi_status != DDI_SUCCESS) {
1267 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
1268 			    "ddi_map_regs, nxge bus config regs failed"));
1269 			goto nxge_map_regs_fail0;
1270 		}
1271 		NXGE_DEBUG_MSG((nxgep, DDI_CTL,
1272 		    "nxge_map_reg: PCI config addr 0x%0llx "
1273 		    " handle 0x%0llx", dev_regs->nxge_pciregp,
1274 		    dev_regs->nxge_pciregh));
1275 			/*
1276 			 * IMP IMP
1277 			 * workaround  for bit swapping bug in HW
1278 			 * which ends up in no-snoop = yes
1279 			 * resulting, in DMA not synched properly
1280 			 */
1281 #if !defined(_BIG_ENDIAN)
1282 		/* workarounds for x86 systems */
1283 		pci_offset = 0x80 + PCIE_DEVCTL;
1284 		pcie_devctl = pci_config_get16(dev_regs->nxge_pciregh,
1285 		    pci_offset);
1286 		pcie_devctl &= ~PCIE_DEVCTL_ENABLE_NO_SNOOP;
1287 		pcie_devctl |= PCIE_DEVCTL_RO_EN;
1288 		pci_config_put16(dev_regs->nxge_pciregh, pci_offset,
1289 		    pcie_devctl);
1290 #endif
1291 
1292 		(void) ddi_dev_regsize(nxgep->dip, 1, &regsize);
1293 		NXGE_DEBUG_MSG((nxgep, DDI_CTL,
1294 		    "nxge_map_regs: pio size 0x%x", regsize));
1295 		/* set up the device mapped register */
1296 		ddi_status = ddi_regs_map_setup(nxgep->dip, 1,
1297 		    (caddr_t *)&(dev_regs->nxge_regp), 0, 0,
1298 		    &nxge_dev_reg_acc_attr, &dev_regs->nxge_regh);
1299 		if (ddi_status != DDI_SUCCESS) {
1300 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
1301 			    "ddi_map_regs for Neptune global reg failed"));
1302 			goto nxge_map_regs_fail1;
1303 		}
1304 
1305 		/* set up the msi/msi-x mapped register */
1306 		(void) ddi_dev_regsize(nxgep->dip, 2, &regsize);
1307 		NXGE_DEBUG_MSG((nxgep, DDI_CTL,
1308 		    "nxge_map_regs: msix size 0x%x", regsize));
1309 		ddi_status = ddi_regs_map_setup(nxgep->dip, 2,
1310 		    (caddr_t *)&(dev_regs->nxge_msix_regp), 0, 0,
1311 		    &nxge_dev_reg_acc_attr, &dev_regs->nxge_msix_regh);
1312 		if (ddi_status != DDI_SUCCESS) {
1313 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
1314 			    "ddi_map_regs for msi reg failed"));
1315 			goto nxge_map_regs_fail2;
1316 		}
1317 
1318 		/* set up the vio region mapped register */
1319 		(void) ddi_dev_regsize(nxgep->dip, 3, &regsize);
1320 		NXGE_DEBUG_MSG((nxgep, DDI_CTL,
1321 		    "nxge_map_regs: vio size 0x%x", regsize));
1322 		ddi_status = ddi_regs_map_setup(nxgep->dip, 3,
1323 		    (caddr_t *)&(dev_regs->nxge_vir_regp), 0, 0,
1324 		    &nxge_dev_reg_acc_attr, &dev_regs->nxge_vir_regh);
1325 
1326 		if (ddi_status != DDI_SUCCESS) {
1327 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
1328 			    "ddi_map_regs for nxge vio reg failed"));
1329 			goto nxge_map_regs_fail3;
1330 		}
1331 		nxgep->dev_regs = dev_regs;
1332 
1333 		NPI_PCI_ACC_HANDLE_SET(nxgep, dev_regs->nxge_pciregh);
1334 		NPI_PCI_ADD_HANDLE_SET(nxgep,
1335 		    (npi_reg_ptr_t)dev_regs->nxge_pciregp);
1336 		NPI_MSI_ACC_HANDLE_SET(nxgep, dev_regs->nxge_msix_regh);
1337 		NPI_MSI_ADD_HANDLE_SET(nxgep,
1338 		    (npi_reg_ptr_t)dev_regs->nxge_msix_regp);
1339 
1340 		NPI_ACC_HANDLE_SET(nxgep, dev_regs->nxge_regh);
1341 		NPI_ADD_HANDLE_SET(nxgep, (npi_reg_ptr_t)dev_regs->nxge_regp);
1342 
1343 		NPI_REG_ACC_HANDLE_SET(nxgep, dev_regs->nxge_regh);
1344 		NPI_REG_ADD_HANDLE_SET(nxgep,
1345 		    (npi_reg_ptr_t)dev_regs->nxge_regp);
1346 
1347 		NPI_VREG_ACC_HANDLE_SET(nxgep, dev_regs->nxge_vir_regh);
1348 		NPI_VREG_ADD_HANDLE_SET(nxgep,
1349 		    (npi_reg_ptr_t)dev_regs->nxge_vir_regp);
1350 
1351 		break;
1352 
1353 	case N2_NIU:
1354 		NXGE_DEBUG_MSG((nxgep, DDI_CTL, "ddi_map_regs, NIU"));
1355 		/*
1356 		 * Set up the device mapped register (FWARC 2006/556)
1357 		 * (changed back to 1: reg starts at 1!)
1358 		 */
1359 		(void) ddi_dev_regsize(nxgep->dip, 1, &regsize);
1360 		NXGE_DEBUG_MSG((nxgep, DDI_CTL,
1361 		    "nxge_map_regs: dev size 0x%x", regsize));
1362 		ddi_status = ddi_regs_map_setup(nxgep->dip, 1,
1363 		    (caddr_t *)&(dev_regs->nxge_regp), 0, 0,
1364 		    &nxge_dev_reg_acc_attr, &dev_regs->nxge_regh);
1365 
1366 		if (ddi_status != DDI_SUCCESS) {
1367 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
1368 			    "ddi_map_regs for N2/NIU, global reg failed "));
1369 			goto nxge_map_regs_fail1;
1370 		}
1371 
1372 		/* set up the first vio region mapped register */
1373 		(void) ddi_dev_regsize(nxgep->dip, 2, &regsize);
1374 		NXGE_DEBUG_MSG((nxgep, DDI_CTL,
1375 		    "nxge_map_regs: vio (1) size 0x%x", regsize));
1376 		ddi_status = ddi_regs_map_setup(nxgep->dip, 2,
1377 		    (caddr_t *)&(dev_regs->nxge_vir_regp), 0, 0,
1378 		    &nxge_dev_reg_acc_attr, &dev_regs->nxge_vir_regh);
1379 
1380 		if (ddi_status != DDI_SUCCESS) {
1381 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
1382 			    "ddi_map_regs for nxge vio reg failed"));
1383 			goto nxge_map_regs_fail2;
1384 		}
1385 		/* set up the second vio region mapped register */
1386 		(void) ddi_dev_regsize(nxgep->dip, 3, &regsize);
1387 		NXGE_DEBUG_MSG((nxgep, DDI_CTL,
1388 		    "nxge_map_regs: vio (3) size 0x%x", regsize));
1389 		ddi_status = ddi_regs_map_setup(nxgep->dip, 3,
1390 		    (caddr_t *)&(dev_regs->nxge_vir2_regp), 0, 0,
1391 		    &nxge_dev_reg_acc_attr, &dev_regs->nxge_vir2_regh);
1392 
1393 		if (ddi_status != DDI_SUCCESS) {
1394 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
1395 			    "ddi_map_regs for nxge vio2 reg failed"));
1396 			goto nxge_map_regs_fail3;
1397 		}
1398 		nxgep->dev_regs = dev_regs;
1399 
1400 		NPI_ACC_HANDLE_SET(nxgep, dev_regs->nxge_regh);
1401 		NPI_ADD_HANDLE_SET(nxgep, (npi_reg_ptr_t)dev_regs->nxge_regp);
1402 
1403 		NPI_REG_ACC_HANDLE_SET(nxgep, dev_regs->nxge_regh);
1404 		NPI_REG_ADD_HANDLE_SET(nxgep,
1405 		    (npi_reg_ptr_t)dev_regs->nxge_regp);
1406 
1407 		NPI_VREG_ACC_HANDLE_SET(nxgep, dev_regs->nxge_vir_regh);
1408 		NPI_VREG_ADD_HANDLE_SET(nxgep,
1409 		    (npi_reg_ptr_t)dev_regs->nxge_vir_regp);
1410 
1411 		NPI_V2REG_ACC_HANDLE_SET(nxgep, dev_regs->nxge_vir2_regh);
1412 		NPI_V2REG_ADD_HANDLE_SET(nxgep,
1413 		    (npi_reg_ptr_t)dev_regs->nxge_vir2_regp);
1414 
1415 		break;
1416 	}
1417 
1418 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "nxge_map_reg: hardware addr 0x%0llx "
1419 	    " handle 0x%0llx", dev_regs->nxge_regp, dev_regs->nxge_regh));
1420 
1421 	goto nxge_map_regs_exit;
1422 nxge_map_regs_fail3:
1423 	if (dev_regs->nxge_msix_regh) {
1424 		ddi_regs_map_free(&dev_regs->nxge_msix_regh);
1425 	}
1426 	if (dev_regs->nxge_vir_regh) {
1427 		ddi_regs_map_free(&dev_regs->nxge_regh);
1428 	}
1429 nxge_map_regs_fail2:
1430 	if (dev_regs->nxge_regh) {
1431 		ddi_regs_map_free(&dev_regs->nxge_regh);
1432 	}
1433 nxge_map_regs_fail1:
1434 	if (dev_regs->nxge_pciregh) {
1435 		ddi_regs_map_free(&dev_regs->nxge_pciregh);
1436 	}
1437 nxge_map_regs_fail0:
1438 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "Freeing register set memory"));
1439 	kmem_free(dev_regs, sizeof (dev_regs_t));
1440 
1441 nxge_map_regs_exit:
1442 	if (ddi_status != DDI_SUCCESS)
1443 		status |= (NXGE_ERROR | NXGE_DDI_FAILED);
1444 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "<== nxge_map_regs"));
1445 	return (status);
1446 }
1447 
1448 static void
1449 nxge_unmap_regs(p_nxge_t nxgep)
1450 {
1451 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "==> nxge_unmap_regs"));
1452 
1453 	if (isLDOMguest(nxgep)) {
1454 		nxge_guest_regs_map_free(nxgep);
1455 		return;
1456 	}
1457 
1458 	if (nxgep->dev_regs) {
1459 		if (nxgep->dev_regs->nxge_pciregh) {
1460 			NXGE_DEBUG_MSG((nxgep, DDI_CTL,
1461 			    "==> nxge_unmap_regs: bus"));
1462 			ddi_regs_map_free(&nxgep->dev_regs->nxge_pciregh);
1463 			nxgep->dev_regs->nxge_pciregh = NULL;
1464 		}
1465 		if (nxgep->dev_regs->nxge_regh) {
1466 			NXGE_DEBUG_MSG((nxgep, DDI_CTL,
1467 			    "==> nxge_unmap_regs: device registers"));
1468 			ddi_regs_map_free(&nxgep->dev_regs->nxge_regh);
1469 			nxgep->dev_regs->nxge_regh = NULL;
1470 		}
1471 		if (nxgep->dev_regs->nxge_msix_regh) {
1472 			NXGE_DEBUG_MSG((nxgep, DDI_CTL,
1473 			    "==> nxge_unmap_regs: device interrupts"));
1474 			ddi_regs_map_free(&nxgep->dev_regs->nxge_msix_regh);
1475 			nxgep->dev_regs->nxge_msix_regh = NULL;
1476 		}
1477 		if (nxgep->dev_regs->nxge_vir_regh) {
1478 			NXGE_DEBUG_MSG((nxgep, DDI_CTL,
1479 			    "==> nxge_unmap_regs: vio region"));
1480 			ddi_regs_map_free(&nxgep->dev_regs->nxge_vir_regh);
1481 			nxgep->dev_regs->nxge_vir_regh = NULL;
1482 		}
1483 		if (nxgep->dev_regs->nxge_vir2_regh) {
1484 			NXGE_DEBUG_MSG((nxgep, DDI_CTL,
1485 			    "==> nxge_unmap_regs: vio2 region"));
1486 			ddi_regs_map_free(&nxgep->dev_regs->nxge_vir2_regh);
1487 			nxgep->dev_regs->nxge_vir2_regh = NULL;
1488 		}
1489 
1490 		kmem_free(nxgep->dev_regs, sizeof (dev_regs_t));
1491 		nxgep->dev_regs = NULL;
1492 	}
1493 
1494 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "<== nxge_unmap_regs"));
1495 }
1496 
1497 static nxge_status_t
1498 nxge_setup_mutexes(p_nxge_t nxgep)
1499 {
1500 	int ddi_status = DDI_SUCCESS;
1501 	nxge_status_t status = NXGE_OK;
1502 	nxge_classify_t *classify_ptr;
1503 	int partition;
1504 
1505 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "==> nxge_setup_mutexes"));
1506 
1507 	/*
1508 	 * Get the interrupt cookie so the mutexes can be
1509 	 * Initialized.
1510 	 */
1511 	if (isLDOMguest(nxgep)) {
1512 		nxgep->interrupt_cookie = 0;
1513 	} else {
1514 		ddi_status = ddi_get_iblock_cookie(nxgep->dip, 0,
1515 		    &nxgep->interrupt_cookie);
1516 
1517 		if (ddi_status != DDI_SUCCESS) {
1518 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
1519 			    "<== nxge_setup_mutexes: failed 0x%x",
1520 			    ddi_status));
1521 			goto nxge_setup_mutexes_exit;
1522 		}
1523 	}
1524 
1525 	cv_init(&nxgep->poll_cv, NULL, CV_DRIVER, NULL);
1526 	MUTEX_INIT(&nxgep->poll_lock, NULL,
1527 	    MUTEX_DRIVER, (void *)nxgep->interrupt_cookie);
1528 
1529 	/*
1530 	 * Initialize mutexes for this device.
1531 	 */
1532 	MUTEX_INIT(nxgep->genlock, NULL,
1533 	    MUTEX_DRIVER, (void *)nxgep->interrupt_cookie);
1534 	MUTEX_INIT(&nxgep->ouraddr_lock, NULL,
1535 	    MUTEX_DRIVER, (void *)nxgep->interrupt_cookie);
1536 	MUTEX_INIT(&nxgep->mif_lock, NULL,
1537 	    MUTEX_DRIVER, (void *)nxgep->interrupt_cookie);
1538 	MUTEX_INIT(&nxgep->group_lock, NULL,
1539 	    MUTEX_DRIVER, (void *)nxgep->interrupt_cookie);
1540 	RW_INIT(&nxgep->filter_lock, NULL,
1541 	    RW_DRIVER, (void *)nxgep->interrupt_cookie);
1542 
1543 	classify_ptr = &nxgep->classifier;
1544 		/*
1545 		 * FFLP Mutexes are never used in interrupt context
1546 		 * as fflp operation can take very long time to
1547 		 * complete and hence not suitable to invoke from interrupt
1548 		 * handlers.
1549 		 */
1550 	MUTEX_INIT(&classify_ptr->tcam_lock, NULL,
1551 	    NXGE_MUTEX_DRIVER, (void *)nxgep->interrupt_cookie);
1552 	if (NXGE_IS_VALID_NEPTUNE_TYPE(nxgep)) {
1553 		MUTEX_INIT(&classify_ptr->fcram_lock, NULL,
1554 		    NXGE_MUTEX_DRIVER, (void *)nxgep->interrupt_cookie);
1555 		for (partition = 0; partition < MAX_PARTITION; partition++) {
1556 			MUTEX_INIT(&classify_ptr->hash_lock[partition], NULL,
1557 			    NXGE_MUTEX_DRIVER, (void *)nxgep->interrupt_cookie);
1558 		}
1559 	}
1560 
1561 nxge_setup_mutexes_exit:
1562 	NXGE_DEBUG_MSG((nxgep, DDI_CTL,
1563 	    "<== nxge_setup_mutexes status = %x", status));
1564 
1565 	if (ddi_status != DDI_SUCCESS)
1566 		status |= (NXGE_ERROR | NXGE_DDI_FAILED);
1567 
1568 	return (status);
1569 }
1570 
1571 static void
1572 nxge_destroy_mutexes(p_nxge_t nxgep)
1573 {
1574 	int partition;
1575 	nxge_classify_t *classify_ptr;
1576 
1577 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "==> nxge_destroy_mutexes"));
1578 	RW_DESTROY(&nxgep->filter_lock);
1579 	MUTEX_DESTROY(&nxgep->group_lock);
1580 	MUTEX_DESTROY(&nxgep->mif_lock);
1581 	MUTEX_DESTROY(&nxgep->ouraddr_lock);
1582 	MUTEX_DESTROY(nxgep->genlock);
1583 
1584 	classify_ptr = &nxgep->classifier;
1585 	MUTEX_DESTROY(&classify_ptr->tcam_lock);
1586 
1587 	/* Destroy all polling resources. */
1588 	MUTEX_DESTROY(&nxgep->poll_lock);
1589 	cv_destroy(&nxgep->poll_cv);
1590 
1591 	/* free data structures, based on HW type */
1592 	if (NXGE_IS_VALID_NEPTUNE_TYPE(nxgep)) {
1593 		MUTEX_DESTROY(&classify_ptr->fcram_lock);
1594 		for (partition = 0; partition < MAX_PARTITION; partition++) {
1595 			MUTEX_DESTROY(&classify_ptr->hash_lock[partition]);
1596 		}
1597 	}
1598 
1599 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "<== nxge_destroy_mutexes"));
1600 }
1601 
1602 nxge_status_t
1603 nxge_init(p_nxge_t nxgep)
1604 {
1605 	nxge_status_t status = NXGE_OK;
1606 
1607 	NXGE_DEBUG_MSG((nxgep, STR_CTL, "==> nxge_init"));
1608 
1609 	if (nxgep->drv_state & STATE_HW_INITIALIZED) {
1610 		return (status);
1611 	}
1612 
1613 	/*
1614 	 * Allocate system memory for the receive/transmit buffer blocks
1615 	 * and receive/transmit descriptor rings.
1616 	 */
1617 	status = nxge_alloc_mem_pool(nxgep);
1618 	if (status != NXGE_OK) {
1619 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL, "alloc mem failed\n"));
1620 		goto nxge_init_fail1;
1621 	}
1622 
1623 	if (!isLDOMguest(nxgep)) {
1624 		/*
1625 		 * Initialize and enable the TXC registers.
1626 		 * (Globally enable the Tx controller,
1627 		 *  enable the port, configure the dma channel bitmap,
1628 		 *  configure the max burst size).
1629 		 */
1630 		status = nxge_txc_init(nxgep);
1631 		if (status != NXGE_OK) {
1632 			NXGE_ERROR_MSG((nxgep,
1633 			    NXGE_ERR_CTL, "init txc failed\n"));
1634 			goto nxge_init_fail2;
1635 		}
1636 	}
1637 
1638 	/*
1639 	 * Initialize and enable TXDMA channels.
1640 	 */
1641 	status = nxge_init_txdma_channels(nxgep);
1642 	if (status != NXGE_OK) {
1643 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL, "init txdma failed\n"));
1644 		goto nxge_init_fail3;
1645 	}
1646 
1647 	/*
1648 	 * Initialize and enable RXDMA channels.
1649 	 */
1650 	status = nxge_init_rxdma_channels(nxgep);
1651 	if (status != NXGE_OK) {
1652 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL, "init rxdma failed\n"));
1653 		goto nxge_init_fail4;
1654 	}
1655 
1656 	/*
1657 	 * The guest domain is now done.
1658 	 */
1659 	if (isLDOMguest(nxgep)) {
1660 		nxgep->drv_state |= STATE_HW_INITIALIZED;
1661 		goto nxge_init_exit;
1662 	}
1663 
1664 	/*
1665 	 * Initialize TCAM and FCRAM (Neptune).
1666 	 */
1667 	status = nxge_classify_init(nxgep);
1668 	if (status != NXGE_OK) {
1669 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL, "init classify failed\n"));
1670 		goto nxge_init_fail5;
1671 	}
1672 
1673 	/*
1674 	 * Initialize ZCP
1675 	 */
1676 	status = nxge_zcp_init(nxgep);
1677 	if (status != NXGE_OK) {
1678 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL, "init ZCP failed\n"));
1679 		goto nxge_init_fail5;
1680 	}
1681 
1682 	/*
1683 	 * Initialize IPP.
1684 	 */
1685 	status = nxge_ipp_init(nxgep);
1686 	if (status != NXGE_OK) {
1687 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL, "init IPP failed\n"));
1688 		goto nxge_init_fail5;
1689 	}
1690 
1691 	/*
1692 	 * Initialize the MAC block.
1693 	 */
1694 	status = nxge_mac_init(nxgep);
1695 	if (status != NXGE_OK) {
1696 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL, "init MAC failed\n"));
1697 		goto nxge_init_fail5;
1698 	}
1699 
1700 	/*
1701 	 * Enable the interrrupts for DDI.
1702 	 */
1703 	nxge_intrs_enable(nxgep);
1704 
1705 	nxgep->drv_state |= STATE_HW_INITIALIZED;
1706 
1707 	goto nxge_init_exit;
1708 
1709 nxge_init_fail5:
1710 	nxge_uninit_rxdma_channels(nxgep);
1711 nxge_init_fail4:
1712 	nxge_uninit_txdma_channels(nxgep);
1713 nxge_init_fail3:
1714 	if (!isLDOMguest(nxgep)) {
1715 		(void) nxge_txc_uninit(nxgep);
1716 	}
1717 nxge_init_fail2:
1718 	nxge_free_mem_pool(nxgep);
1719 nxge_init_fail1:
1720 	NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
1721 	    "<== nxge_init status (failed) = 0x%08x", status));
1722 	return (status);
1723 
1724 nxge_init_exit:
1725 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "<== nxge_init status = 0x%08x",
1726 	    status));
1727 	return (status);
1728 }
1729 
1730 
1731 timeout_id_t
1732 nxge_start_timer(p_nxge_t nxgep, fptrv_t func, int msec)
1733 {
1734 	if ((nxgep->suspended == 0) || (nxgep->suspended == DDI_RESUME)) {
1735 		return (timeout(func, (caddr_t)nxgep,
1736 		    drv_usectohz(1000 * msec)));
1737 	}
1738 	return (NULL);
1739 }
1740 
1741 /*ARGSUSED*/
1742 void
1743 nxge_stop_timer(p_nxge_t nxgep, timeout_id_t timerid)
1744 {
1745 	if (timerid) {
1746 		(void) untimeout(timerid);
1747 	}
1748 }
1749 
1750 void
1751 nxge_uninit(p_nxge_t nxgep)
1752 {
1753 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "==> nxge_uninit"));
1754 
1755 	if (!(nxgep->drv_state & STATE_HW_INITIALIZED)) {
1756 		NXGE_DEBUG_MSG((nxgep, DDI_CTL,
1757 		    "==> nxge_uninit: not initialized"));
1758 		NXGE_DEBUG_MSG((nxgep, DDI_CTL,
1759 		    "<== nxge_uninit"));
1760 		return;
1761 	}
1762 
1763 	if (!isLDOMguest(nxgep)) {
1764 		/*
1765 		 * Reset the receive MAC side.
1766 		 */
1767 		(void) nxge_rx_mac_disable(nxgep);
1768 
1769 		/*
1770 		 * Drain the IPP.
1771 		 */
1772 		(void) nxge_ipp_drain(nxgep);
1773 	}
1774 
1775 	/* stop timer */
1776 	if (nxgep->nxge_timerid) {
1777 		nxge_stop_timer(nxgep, nxgep->nxge_timerid);
1778 		nxgep->nxge_timerid = 0;
1779 	}
1780 
1781 	(void) nxge_link_monitor(nxgep, LINK_MONITOR_STOP);
1782 	(void) nxge_intr_hw_disable(nxgep);
1783 
1784 
1785 	/* Disable and soft reset the IPP */
1786 	if (!isLDOMguest(nxgep))
1787 		(void) nxge_ipp_disable(nxgep);
1788 
1789 	/* Free classification resources */
1790 	(void) nxge_classify_uninit(nxgep);
1791 
1792 	/*
1793 	 * Reset the transmit/receive DMA side.
1794 	 */
1795 	(void) nxge_txdma_hw_mode(nxgep, NXGE_DMA_STOP);
1796 	(void) nxge_rxdma_hw_mode(nxgep, NXGE_DMA_STOP);
1797 
1798 	nxge_uninit_txdma_channels(nxgep);
1799 	nxge_uninit_rxdma_channels(nxgep);
1800 
1801 	/*
1802 	 * Reset the transmit MAC side.
1803 	 */
1804 	(void) nxge_tx_mac_disable(nxgep);
1805 
1806 	nxge_free_mem_pool(nxgep);
1807 
1808 	/*
1809 	 * Start the timer if the reset flag is not set.
1810 	 * If this reset flag is set, the link monitor
1811 	 * will not be started in order to stop furthur bus
1812 	 * activities coming from this interface.
1813 	 * The driver will start the monitor function
1814 	 * if the interface was initialized again later.
1815 	 */
1816 	if (!nxge_peu_reset_enable) {
1817 		(void) nxge_link_monitor(nxgep, LINK_MONITOR_START);
1818 	}
1819 
1820 	nxgep->drv_state &= ~STATE_HW_INITIALIZED;
1821 
1822 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "<== nxge_uninit: "
1823 	    "nxge_mblks_pending %d", nxge_mblks_pending));
1824 }
1825 
1826 void
1827 nxge_get64(p_nxge_t nxgep, p_mblk_t mp)
1828 {
1829 	uint64_t	reg;
1830 	uint64_t	regdata;
1831 	int		i, retry;
1832 
1833 	bcopy((char *)mp->b_rptr, (char *)&reg, sizeof (uint64_t));
1834 	regdata = 0;
1835 	retry = 1;
1836 
1837 	for (i = 0; i < retry; i++) {
1838 		NXGE_REG_RD64(nxgep->npi_handle, reg, &regdata);
1839 	}
1840 	bcopy((char *)&regdata, (char *)mp->b_rptr, sizeof (uint64_t));
1841 }
1842 
1843 void
1844 nxge_put64(p_nxge_t nxgep, p_mblk_t mp)
1845 {
1846 	uint64_t	reg;
1847 	uint64_t	buf[2];
1848 
1849 	bcopy((char *)mp->b_rptr, (char *)&buf[0], 2 * sizeof (uint64_t));
1850 	reg = buf[0];
1851 
1852 	NXGE_NPI_PIO_WRITE64(nxgep->npi_handle, reg, buf[1]);
1853 }
1854 
1855 /*ARGSUSED*/
1856 /*VARARGS*/
1857 void
1858 nxge_debug_msg(p_nxge_t nxgep, uint64_t level, char *fmt, ...)
1859 {
1860 	char msg_buffer[1048];
1861 	char prefix_buffer[32];
1862 	int instance;
1863 	uint64_t debug_level;
1864 	int cmn_level = CE_CONT;
1865 	va_list ap;
1866 
1867 	if (nxgep && nxgep->nxge_debug_level != nxge_debug_level) {
1868 		/* In case a developer has changed nxge_debug_level. */
1869 		if (nxgep->nxge_debug_level != nxge_debug_level)
1870 			nxgep->nxge_debug_level = nxge_debug_level;
1871 	}
1872 
1873 	debug_level = (nxgep == NULL) ? nxge_debug_level :
1874 	    nxgep->nxge_debug_level;
1875 
1876 	if ((level & debug_level) ||
1877 	    (level == NXGE_NOTE) ||
1878 	    (level == NXGE_ERR_CTL)) {
1879 		/* do the msg processing */
1880 		MUTEX_ENTER(&nxgedebuglock);
1881 
1882 		if ((level & NXGE_NOTE)) {
1883 			cmn_level = CE_NOTE;
1884 		}
1885 
1886 		if (level & NXGE_ERR_CTL) {
1887 			cmn_level = CE_WARN;
1888 		}
1889 
1890 		va_start(ap, fmt);
1891 		(void) vsprintf(msg_buffer, fmt, ap);
1892 		va_end(ap);
1893 		if (nxgep == NULL) {
1894 			instance = -1;
1895 			(void) sprintf(prefix_buffer, "%s :", "nxge");
1896 		} else {
1897 			instance = nxgep->instance;
1898 			(void) sprintf(prefix_buffer,
1899 			    "%s%d :", "nxge", instance);
1900 		}
1901 
1902 		MUTEX_EXIT(&nxgedebuglock);
1903 		cmn_err(cmn_level, "!%s %s\n",
1904 		    prefix_buffer, msg_buffer);
1905 
1906 	}
1907 }
1908 
1909 char *
1910 nxge_dump_packet(char *addr, int size)
1911 {
1912 	uchar_t *ap = (uchar_t *)addr;
1913 	int i;
1914 	static char etherbuf[1024];
1915 	char *cp = etherbuf;
1916 	char digits[] = "0123456789abcdef";
1917 
1918 	if (!size)
1919 		size = 60;
1920 
1921 	if (size > MAX_DUMP_SZ) {
1922 		/* Dump the leading bytes */
1923 		for (i = 0; i < MAX_DUMP_SZ/2; i++) {
1924 			if (*ap > 0x0f)
1925 				*cp++ = digits[*ap >> 4];
1926 			*cp++ = digits[*ap++ & 0xf];
1927 			*cp++ = ':';
1928 		}
1929 		for (i = 0; i < 20; i++)
1930 			*cp++ = '.';
1931 		/* Dump the last MAX_DUMP_SZ/2 bytes */
1932 		ap = (uchar_t *)(addr + (size - MAX_DUMP_SZ/2));
1933 		for (i = 0; i < MAX_DUMP_SZ/2; i++) {
1934 			if (*ap > 0x0f)
1935 				*cp++ = digits[*ap >> 4];
1936 			*cp++ = digits[*ap++ & 0xf];
1937 			*cp++ = ':';
1938 		}
1939 	} else {
1940 		for (i = 0; i < size; i++) {
1941 			if (*ap > 0x0f)
1942 				*cp++ = digits[*ap >> 4];
1943 			*cp++ = digits[*ap++ & 0xf];
1944 			*cp++ = ':';
1945 		}
1946 	}
1947 	*--cp = 0;
1948 	return (etherbuf);
1949 }
1950 
1951 #ifdef	NXGE_DEBUG
1952 static void
1953 nxge_test_map_regs(p_nxge_t nxgep)
1954 {
1955 	ddi_acc_handle_t cfg_handle;
1956 	p_pci_cfg_t	cfg_ptr;
1957 	ddi_acc_handle_t dev_handle;
1958 	char		*dev_ptr;
1959 	ddi_acc_handle_t pci_config_handle;
1960 	uint32_t	regval;
1961 	int		i;
1962 
1963 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "==> nxge_test_map_regs"));
1964 
1965 	dev_handle = nxgep->dev_regs->nxge_regh;
1966 	dev_ptr = (char *)nxgep->dev_regs->nxge_regp;
1967 
1968 	if (NXGE_IS_VALID_NEPTUNE_TYPE(nxgep)) {
1969 		cfg_handle = nxgep->dev_regs->nxge_pciregh;
1970 		cfg_ptr = (void *)nxgep->dev_regs->nxge_pciregp;
1971 
1972 		NXGE_DEBUG_MSG((nxgep, DDI_CTL,
1973 		    "Neptune PCI regp cfg_ptr 0x%llx", (char *)cfg_ptr));
1974 		NXGE_DEBUG_MSG((nxgep, DDI_CTL,
1975 		    "Neptune PCI cfg_ptr vendor id ptr 0x%llx",
1976 		    &cfg_ptr->vendorid));
1977 		NXGE_DEBUG_MSG((nxgep, DDI_CTL,
1978 		    "\tvendorid 0x%x devid 0x%x",
1979 		    NXGE_PIO_READ16(cfg_handle, &cfg_ptr->vendorid, 0),
1980 		    NXGE_PIO_READ16(cfg_handle, &cfg_ptr->devid,    0)));
1981 		NXGE_DEBUG_MSG((nxgep, DDI_CTL,
1982 		    "PCI BAR: base 0x%x base14 0x%x base 18 0x%x "
1983 		    "bar1c 0x%x",
1984 		    NXGE_PIO_READ32(cfg_handle, &cfg_ptr->base,   0),
1985 		    NXGE_PIO_READ32(cfg_handle, &cfg_ptr->base14, 0),
1986 		    NXGE_PIO_READ32(cfg_handle, &cfg_ptr->base18, 0),
1987 		    NXGE_PIO_READ32(cfg_handle, &cfg_ptr->base1c, 0)));
1988 		NXGE_DEBUG_MSG((nxgep, DDI_CTL,
1989 		    "\nNeptune PCI BAR: base20 0x%x base24 0x%x "
1990 		    "base 28 0x%x bar2c 0x%x\n",
1991 		    NXGE_PIO_READ32(cfg_handle, &cfg_ptr->base20, 0),
1992 		    NXGE_PIO_READ32(cfg_handle, &cfg_ptr->base24, 0),
1993 		    NXGE_PIO_READ32(cfg_handle, &cfg_ptr->base28, 0),
1994 		    NXGE_PIO_READ32(cfg_handle, &cfg_ptr->base2c, 0)));
1995 		NXGE_DEBUG_MSG((nxgep, DDI_CTL,
1996 		    "\nNeptune PCI BAR: base30 0x%x\n",
1997 		    NXGE_PIO_READ32(cfg_handle, &cfg_ptr->base30, 0)));
1998 
1999 		cfg_handle = nxgep->dev_regs->nxge_pciregh;
2000 		cfg_ptr = (void *)nxgep->dev_regs->nxge_pciregp;
2001 		NXGE_DEBUG_MSG((nxgep, DDI_CTL,
2002 		    "first  0x%llx second 0x%llx third 0x%llx "
2003 		    "last 0x%llx ",
2004 		    NXGE_PIO_READ64(dev_handle,
2005 		    (uint64_t *)(dev_ptr + 0),  0),
2006 		    NXGE_PIO_READ64(dev_handle,
2007 		    (uint64_t *)(dev_ptr + 8),  0),
2008 		    NXGE_PIO_READ64(dev_handle,
2009 		    (uint64_t *)(dev_ptr + 16), 0),
2010 		    NXGE_PIO_READ64(cfg_handle,
2011 		    (uint64_t *)(dev_ptr + 24), 0)));
2012 	}
2013 }
2014 
2015 #endif
2016 
2017 static void
2018 nxge_suspend(p_nxge_t nxgep)
2019 {
2020 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "==> nxge_suspend"));
2021 
2022 	nxge_intrs_disable(nxgep);
2023 	nxge_destroy_dev(nxgep);
2024 
2025 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "<== nxge_suspend"));
2026 }
2027 
2028 static nxge_status_t
2029 nxge_resume(p_nxge_t nxgep)
2030 {
2031 	nxge_status_t status = NXGE_OK;
2032 
2033 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "==> nxge_resume"));
2034 
2035 	nxgep->suspended = DDI_RESUME;
2036 	(void) nxge_link_monitor(nxgep, LINK_MONITOR_START);
2037 	(void) nxge_rxdma_hw_mode(nxgep, NXGE_DMA_START);
2038 	(void) nxge_txdma_hw_mode(nxgep, NXGE_DMA_START);
2039 	(void) nxge_rx_mac_enable(nxgep);
2040 	(void) nxge_tx_mac_enable(nxgep);
2041 	nxge_intrs_enable(nxgep);
2042 	nxgep->suspended = 0;
2043 
2044 	NXGE_DEBUG_MSG((nxgep, DDI_CTL,
2045 	    "<== nxge_resume status = 0x%x", status));
2046 	return (status);
2047 }
2048 
2049 static nxge_status_t
2050 nxge_setup_dev(p_nxge_t nxgep)
2051 {
2052 	nxge_status_t	status = NXGE_OK;
2053 
2054 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "==> nxge_setup_dev port %d",
2055 	    nxgep->mac.portnum));
2056 
2057 	status = nxge_link_init(nxgep);
2058 
2059 	if (fm_check_acc_handle(nxgep->dev_regs->nxge_regh) != DDI_FM_OK) {
2060 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
2061 		    "port%d Bad register acc handle", nxgep->mac.portnum));
2062 		status = NXGE_ERROR;
2063 	}
2064 
2065 	if (status != NXGE_OK) {
2066 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
2067 		    " nxge_setup_dev status "
2068 		    "(xcvr init 0x%08x)", status));
2069 		goto nxge_setup_dev_exit;
2070 	}
2071 
2072 nxge_setup_dev_exit:
2073 	NXGE_DEBUG_MSG((nxgep, DDI_CTL,
2074 	    "<== nxge_setup_dev port %d status = 0x%08x",
2075 	    nxgep->mac.portnum, status));
2076 
2077 	return (status);
2078 }
2079 
2080 static void
2081 nxge_destroy_dev(p_nxge_t nxgep)
2082 {
2083 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "==> nxge_destroy_dev"));
2084 
2085 	(void) nxge_link_monitor(nxgep, LINK_MONITOR_STOP);
2086 
2087 	(void) nxge_hw_stop(nxgep);
2088 
2089 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "<== nxge_destroy_dev"));
2090 }
2091 
2092 static nxge_status_t
2093 nxge_setup_system_dma_pages(p_nxge_t nxgep)
2094 {
2095 	int 			ddi_status = DDI_SUCCESS;
2096 	uint_t 			count;
2097 	ddi_dma_cookie_t 	cookie;
2098 	uint_t 			iommu_pagesize;
2099 	nxge_status_t		status = NXGE_OK;
2100 
2101 	NXGE_ERROR_MSG((nxgep, DDI_CTL, "==> nxge_setup_system_dma_pages"));
2102 	nxgep->sys_page_sz = ddi_ptob(nxgep->dip, (ulong_t)1);
2103 	if (nxgep->niu_type != N2_NIU) {
2104 		iommu_pagesize = dvma_pagesize(nxgep->dip);
2105 		NXGE_DEBUG_MSG((nxgep, DDI_CTL,
2106 		    " nxge_setup_system_dma_pages: page %d (ddi_ptob %d) "
2107 		    " default_block_size %d iommu_pagesize %d",
2108 		    nxgep->sys_page_sz,
2109 		    ddi_ptob(nxgep->dip, (ulong_t)1),
2110 		    nxgep->rx_default_block_size,
2111 		    iommu_pagesize));
2112 
2113 		if (iommu_pagesize != 0) {
2114 			if (nxgep->sys_page_sz == iommu_pagesize) {
2115 				if (iommu_pagesize > 0x4000)
2116 					nxgep->sys_page_sz = 0x4000;
2117 			} else {
2118 				if (nxgep->sys_page_sz > iommu_pagesize)
2119 					nxgep->sys_page_sz = iommu_pagesize;
2120 			}
2121 		}
2122 	}
2123 	nxgep->sys_page_mask = ~(nxgep->sys_page_sz - 1);
2124 	NXGE_DEBUG_MSG((nxgep, DDI_CTL,
2125 	    "==> nxge_setup_system_dma_pages: page %d (ddi_ptob %d) "
2126 	    "default_block_size %d page mask %d",
2127 	    nxgep->sys_page_sz,
2128 	    ddi_ptob(nxgep->dip, (ulong_t)1),
2129 	    nxgep->rx_default_block_size,
2130 	    nxgep->sys_page_mask));
2131 
2132 
2133 	switch (nxgep->sys_page_sz) {
2134 	default:
2135 		nxgep->sys_page_sz = 0x1000;
2136 		nxgep->sys_page_mask = ~(nxgep->sys_page_sz - 1);
2137 		nxgep->rx_default_block_size = 0x1000;
2138 		nxgep->rx_bksize_code = RBR_BKSIZE_4K;
2139 		break;
2140 	case 0x1000:
2141 		nxgep->rx_default_block_size = 0x1000;
2142 		nxgep->rx_bksize_code = RBR_BKSIZE_4K;
2143 		break;
2144 	case 0x2000:
2145 		nxgep->rx_default_block_size = 0x2000;
2146 		nxgep->rx_bksize_code = RBR_BKSIZE_8K;
2147 		break;
2148 	case 0x4000:
2149 		nxgep->rx_default_block_size = 0x4000;
2150 		nxgep->rx_bksize_code = RBR_BKSIZE_16K;
2151 		break;
2152 	case 0x8000:
2153 		nxgep->rx_default_block_size = 0x8000;
2154 		nxgep->rx_bksize_code = RBR_BKSIZE_32K;
2155 		break;
2156 	}
2157 
2158 #ifndef USE_RX_BIG_BUF
2159 	nxge_rx_dma_attr.dma_attr_align = nxgep->sys_page_sz;
2160 #else
2161 		nxgep->rx_default_block_size = 0x2000;
2162 		nxgep->rx_bksize_code = RBR_BKSIZE_8K;
2163 #endif
2164 	/*
2165 	 * Get the system DMA burst size.
2166 	 */
2167 	ddi_status = ddi_dma_alloc_handle(nxgep->dip, &nxge_tx_dma_attr,
2168 	    DDI_DMA_DONTWAIT, 0,
2169 	    &nxgep->dmasparehandle);
2170 	if (ddi_status != DDI_SUCCESS) {
2171 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
2172 		    "ddi_dma_alloc_handle: failed "
2173 		    " status 0x%x", ddi_status));
2174 		goto nxge_get_soft_properties_exit;
2175 	}
2176 
2177 	ddi_status = ddi_dma_addr_bind_handle(nxgep->dmasparehandle, NULL,
2178 	    (caddr_t)nxgep->dmasparehandle,
2179 	    sizeof (nxgep->dmasparehandle),
2180 	    DDI_DMA_RDWR | DDI_DMA_CONSISTENT,
2181 	    DDI_DMA_DONTWAIT, 0,
2182 	    &cookie, &count);
2183 	if (ddi_status != DDI_DMA_MAPPED) {
2184 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
2185 		    "Binding spare handle to find system"
2186 		    " burstsize failed."));
2187 		ddi_status = DDI_FAILURE;
2188 		goto nxge_get_soft_properties_fail1;
2189 	}
2190 
2191 	nxgep->sys_burst_sz = ddi_dma_burstsizes(nxgep->dmasparehandle);
2192 	(void) ddi_dma_unbind_handle(nxgep->dmasparehandle);
2193 
2194 nxge_get_soft_properties_fail1:
2195 	ddi_dma_free_handle(&nxgep->dmasparehandle);
2196 
2197 nxge_get_soft_properties_exit:
2198 
2199 	if (ddi_status != DDI_SUCCESS)
2200 		status |= (NXGE_ERROR | NXGE_DDI_FAILED);
2201 
2202 	NXGE_DEBUG_MSG((nxgep, DDI_CTL,
2203 	    "<== nxge_setup_system_dma_pages status = 0x%08x", status));
2204 	return (status);
2205 }
2206 
2207 static nxge_status_t
2208 nxge_alloc_mem_pool(p_nxge_t nxgep)
2209 {
2210 	nxge_status_t	status = NXGE_OK;
2211 
2212 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "==> nxge_alloc_mem_pool"));
2213 
2214 	status = nxge_alloc_rx_mem_pool(nxgep);
2215 	if (status != NXGE_OK) {
2216 		return (NXGE_ERROR);
2217 	}
2218 
2219 	status = nxge_alloc_tx_mem_pool(nxgep);
2220 	if (status != NXGE_OK) {
2221 		nxge_free_rx_mem_pool(nxgep);
2222 		return (NXGE_ERROR);
2223 	}
2224 
2225 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "<== nxge_alloc_mem_pool"));
2226 	return (NXGE_OK);
2227 }
2228 
2229 static void
2230 nxge_free_mem_pool(p_nxge_t nxgep)
2231 {
2232 	NXGE_DEBUG_MSG((nxgep, MEM_CTL, "==> nxge_free_mem_pool"));
2233 
2234 	nxge_free_rx_mem_pool(nxgep);
2235 	nxge_free_tx_mem_pool(nxgep);
2236 
2237 	NXGE_DEBUG_MSG((nxgep, MEM_CTL, "<== nxge_free_mem_pool"));
2238 }
2239 
2240 nxge_status_t
2241 nxge_alloc_rx_mem_pool(p_nxge_t nxgep)
2242 {
2243 	uint32_t		rdc_max;
2244 	p_nxge_dma_pt_cfg_t	p_all_cfgp;
2245 	p_nxge_hw_pt_cfg_t	p_cfgp;
2246 	p_nxge_dma_pool_t	dma_poolp;
2247 	p_nxge_dma_common_t	*dma_buf_p;
2248 	p_nxge_dma_pool_t	dma_cntl_poolp;
2249 	p_nxge_dma_common_t	*dma_cntl_p;
2250 	uint32_t 		*num_chunks; /* per dma */
2251 	nxge_status_t		status = NXGE_OK;
2252 
2253 	uint32_t		nxge_port_rbr_size;
2254 	uint32_t		nxge_port_rbr_spare_size;
2255 	uint32_t		nxge_port_rcr_size;
2256 	uint32_t		rx_cntl_alloc_size;
2257 
2258 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "==> nxge_alloc_rx_mem_pool"));
2259 
2260 	p_all_cfgp = (p_nxge_dma_pt_cfg_t)&nxgep->pt_config;
2261 	p_cfgp = (p_nxge_hw_pt_cfg_t)&p_all_cfgp->hw_config;
2262 	rdc_max = NXGE_MAX_RDCS;
2263 
2264 	/*
2265 	 * Allocate memory for the common DMA data structures.
2266 	 */
2267 	dma_poolp = (p_nxge_dma_pool_t)KMEM_ZALLOC(sizeof (nxge_dma_pool_t),
2268 	    KM_SLEEP);
2269 	dma_buf_p = (p_nxge_dma_common_t *)KMEM_ZALLOC(
2270 	    sizeof (p_nxge_dma_common_t) * rdc_max, KM_SLEEP);
2271 
2272 	dma_cntl_poolp = (p_nxge_dma_pool_t)
2273 	    KMEM_ZALLOC(sizeof (nxge_dma_pool_t), KM_SLEEP);
2274 	dma_cntl_p = (p_nxge_dma_common_t *)KMEM_ZALLOC(
2275 	    sizeof (p_nxge_dma_common_t) * rdc_max, KM_SLEEP);
2276 
2277 	num_chunks = (uint32_t *)KMEM_ZALLOC(
2278 	    sizeof (uint32_t) * rdc_max, KM_SLEEP);
2279 
2280 	/*
2281 	 * Assume that each DMA channel will be configured with
2282 	 * the default block size.
2283 	 * rbr block counts are modulo the batch count (16).
2284 	 */
2285 	nxge_port_rbr_size = p_all_cfgp->rbr_size;
2286 	nxge_port_rcr_size = p_all_cfgp->rcr_size;
2287 
2288 	if (!nxge_port_rbr_size) {
2289 		nxge_port_rbr_size = NXGE_RBR_RBB_DEFAULT;
2290 	}
2291 	if (nxge_port_rbr_size % NXGE_RXDMA_POST_BATCH) {
2292 		nxge_port_rbr_size = (NXGE_RXDMA_POST_BATCH *
2293 		    (nxge_port_rbr_size / NXGE_RXDMA_POST_BATCH + 1));
2294 	}
2295 
2296 	p_all_cfgp->rbr_size = nxge_port_rbr_size;
2297 	nxge_port_rbr_spare_size = nxge_rbr_spare_size;
2298 
2299 	if (nxge_port_rbr_spare_size % NXGE_RXDMA_POST_BATCH) {
2300 		nxge_port_rbr_spare_size = (NXGE_RXDMA_POST_BATCH *
2301 		    (nxge_port_rbr_spare_size / NXGE_RXDMA_POST_BATCH + 1));
2302 	}
2303 	if (nxge_port_rbr_size > RBR_DEFAULT_MAX_BLKS) {
2304 		NXGE_DEBUG_MSG((nxgep, MEM_CTL,
2305 		    "nxge_alloc_rx_mem_pool: RBR size too high %d, "
2306 		    "set to default %d",
2307 		    nxge_port_rbr_size, RBR_DEFAULT_MAX_BLKS));
2308 		nxge_port_rbr_size = RBR_DEFAULT_MAX_BLKS;
2309 	}
2310 	if (nxge_port_rcr_size > RCR_DEFAULT_MAX) {
2311 		NXGE_DEBUG_MSG((nxgep, MEM_CTL,
2312 		    "nxge_alloc_rx_mem_pool: RCR too high %d, "
2313 		    "set to default %d",
2314 		    nxge_port_rcr_size, RCR_DEFAULT_MAX));
2315 		nxge_port_rcr_size = RCR_DEFAULT_MAX;
2316 	}
2317 
2318 	/*
2319 	 * N2/NIU has limitation on the descriptor sizes (contiguous
2320 	 * memory allocation on data buffers to 4M (contig_mem_alloc)
2321 	 * and little endian for control buffers (must use the ddi/dki mem alloc
2322 	 * function).
2323 	 */
2324 #if	defined(sun4v) && defined(NIU_LP_WORKAROUND)
2325 	if (nxgep->niu_type == N2_NIU) {
2326 		nxge_port_rbr_spare_size = 0;
2327 		if ((nxge_port_rbr_size > NXGE_NIU_CONTIG_RBR_MAX) ||
2328 		    (!ISP2(nxge_port_rbr_size))) {
2329 			nxge_port_rbr_size = NXGE_NIU_CONTIG_RBR_MAX;
2330 		}
2331 		if ((nxge_port_rcr_size > NXGE_NIU_CONTIG_RCR_MAX) ||
2332 		    (!ISP2(nxge_port_rcr_size))) {
2333 			nxge_port_rcr_size = NXGE_NIU_CONTIG_RCR_MAX;
2334 		}
2335 	}
2336 #endif
2337 
2338 	/*
2339 	 * Addresses of receive block ring, receive completion ring and the
2340 	 * mailbox must be all cache-aligned (64 bytes).
2341 	 */
2342 	rx_cntl_alloc_size = nxge_port_rbr_size + nxge_port_rbr_spare_size;
2343 	rx_cntl_alloc_size *= (sizeof (rx_desc_t));
2344 	rx_cntl_alloc_size += (sizeof (rcr_entry_t) * nxge_port_rcr_size);
2345 	rx_cntl_alloc_size += sizeof (rxdma_mailbox_t);
2346 
2347 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_alloc_rx_mem_pool: "
2348 	    "nxge_port_rbr_size = %d nxge_port_rbr_spare_size = %d "
2349 	    "nxge_port_rcr_size = %d "
2350 	    "rx_cntl_alloc_size = %d",
2351 	    nxge_port_rbr_size, nxge_port_rbr_spare_size,
2352 	    nxge_port_rcr_size,
2353 	    rx_cntl_alloc_size));
2354 
2355 #if	defined(sun4v) && defined(NIU_LP_WORKAROUND)
2356 	if (nxgep->niu_type == N2_NIU) {
2357 		uint32_t rx_buf_alloc_size = (nxgep->rx_default_block_size *
2358 		    (nxge_port_rbr_size + nxge_port_rbr_spare_size));
2359 
2360 		if (!ISP2(rx_buf_alloc_size)) {
2361 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
2362 			    "==> nxge_alloc_rx_mem_pool: "
2363 			    " must be power of 2"));
2364 			status |= (NXGE_ERROR | NXGE_DDI_FAILED);
2365 			goto nxge_alloc_rx_mem_pool_exit;
2366 		}
2367 
2368 		if (rx_buf_alloc_size > (1 << 22)) {
2369 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
2370 			    "==> nxge_alloc_rx_mem_pool: "
2371 			    " limit size to 4M"));
2372 			status |= (NXGE_ERROR | NXGE_DDI_FAILED);
2373 			goto nxge_alloc_rx_mem_pool_exit;
2374 		}
2375 
2376 		if (rx_cntl_alloc_size < 0x2000) {
2377 			rx_cntl_alloc_size = 0x2000;
2378 		}
2379 	}
2380 #endif
2381 	nxgep->nxge_port_rbr_size = nxge_port_rbr_size;
2382 	nxgep->nxge_port_rcr_size = nxge_port_rcr_size;
2383 	nxgep->nxge_port_rbr_spare_size = nxge_port_rbr_spare_size;
2384 	nxgep->nxge_port_rx_cntl_alloc_size = rx_cntl_alloc_size;
2385 
2386 	dma_poolp->ndmas = p_cfgp->max_rdcs;
2387 	dma_poolp->num_chunks = num_chunks;
2388 	dma_poolp->buf_allocated = B_TRUE;
2389 	nxgep->rx_buf_pool_p = dma_poolp;
2390 	dma_poolp->dma_buf_pool_p = dma_buf_p;
2391 
2392 	dma_cntl_poolp->ndmas = p_cfgp->max_rdcs;
2393 	dma_cntl_poolp->buf_allocated = B_TRUE;
2394 	nxgep->rx_cntl_pool_p = dma_cntl_poolp;
2395 	dma_cntl_poolp->dma_buf_pool_p = dma_cntl_p;
2396 
2397 	/* Allocate the receive rings, too. */
2398 	nxgep->rx_rbr_rings =
2399 	    KMEM_ZALLOC(sizeof (rx_rbr_rings_t), KM_SLEEP);
2400 	nxgep->rx_rbr_rings->rbr_rings =
2401 	    KMEM_ZALLOC(sizeof (p_rx_rbr_ring_t) * rdc_max, KM_SLEEP);
2402 	nxgep->rx_rcr_rings =
2403 	    KMEM_ZALLOC(sizeof (rx_rcr_rings_t), KM_SLEEP);
2404 	nxgep->rx_rcr_rings->rcr_rings =
2405 	    KMEM_ZALLOC(sizeof (p_rx_rcr_ring_t) * rdc_max, KM_SLEEP);
2406 	nxgep->rx_mbox_areas_p =
2407 	    KMEM_ZALLOC(sizeof (rx_mbox_areas_t), KM_SLEEP);
2408 	nxgep->rx_mbox_areas_p->rxmbox_areas =
2409 	    KMEM_ZALLOC(sizeof (p_rx_mbox_t) * rdc_max, KM_SLEEP);
2410 
2411 	nxgep->rx_rbr_rings->ndmas = nxgep->rx_rcr_rings->ndmas =
2412 	    p_cfgp->max_rdcs;
2413 
2414 	NXGE_DEBUG_MSG((nxgep, DMA_CTL,
2415 	    "<== nxge_alloc_rx_mem_pool:status 0x%08x", status));
2416 
2417 nxge_alloc_rx_mem_pool_exit:
2418 	return (status);
2419 }
2420 
2421 /*
2422  * nxge_alloc_rxb
2423  *
2424  *	Allocate buffers for an RDC.
2425  *
2426  * Arguments:
2427  * 	nxgep
2428  * 	channel	The channel to map into our kernel space.
2429  *
2430  * Notes:
2431  *
2432  * NPI function calls:
2433  *
2434  * NXGE function calls:
2435  *
2436  * Registers accessed:
2437  *
2438  * Context:
2439  *
2440  * Taking apart:
2441  *
2442  * Open questions:
2443  *
2444  */
2445 nxge_status_t
2446 nxge_alloc_rxb(
2447 	p_nxge_t nxgep,
2448 	int channel)
2449 {
2450 	size_t			rx_buf_alloc_size;
2451 	nxge_status_t		status = NXGE_OK;
2452 
2453 	nxge_dma_common_t	**data;
2454 	nxge_dma_common_t	**control;
2455 	uint32_t 		*num_chunks;
2456 
2457 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "==> nxge_alloc_rbb"));
2458 
2459 	/*
2460 	 * Allocate memory for the receive buffers and descriptor rings.
2461 	 * Replace these allocation functions with the interface functions
2462 	 * provided by the partition manager if/when they are available.
2463 	 */
2464 
2465 	/*
2466 	 * Allocate memory for the receive buffer blocks.
2467 	 */
2468 	rx_buf_alloc_size = (nxgep->rx_default_block_size *
2469 	    (nxgep->nxge_port_rbr_size + nxgep->nxge_port_rbr_spare_size));
2470 
2471 	data = &nxgep->rx_buf_pool_p->dma_buf_pool_p[channel];
2472 	num_chunks = &nxgep->rx_buf_pool_p->num_chunks[channel];
2473 
2474 	if ((status = nxge_alloc_rx_buf_dma(
2475 	    nxgep, channel, data, rx_buf_alloc_size,
2476 	    nxgep->rx_default_block_size, num_chunks)) != NXGE_OK) {
2477 		return (status);
2478 	}
2479 
2480 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "<== nxge_alloc_rxb(): "
2481 	    "dma %d dma_buf_p %llx &dma_buf_p %llx", channel, *data, data));
2482 
2483 	/*
2484 	 * Allocate memory for descriptor rings and mailbox.
2485 	 */
2486 	control = &nxgep->rx_cntl_pool_p->dma_buf_pool_p[channel];
2487 
2488 	if ((status = nxge_alloc_rx_cntl_dma(
2489 	    nxgep, channel, control, nxgep->nxge_port_rx_cntl_alloc_size))
2490 	    != NXGE_OK) {
2491 		nxge_free_rx_cntl_dma(nxgep, *control);
2492 		(*data)->buf_alloc_state |= BUF_ALLOCATED_WAIT_FREE;
2493 		nxge_free_rx_buf_dma(nxgep, *data, *num_chunks);
2494 		return (status);
2495 	}
2496 
2497 	NXGE_DEBUG_MSG((nxgep, DMA_CTL,
2498 	    "<== nxge_alloc_rx_mem_pool:status 0x%08x", status));
2499 
2500 	return (status);
2501 }
2502 
2503 void
2504 nxge_free_rxb(
2505 	p_nxge_t nxgep,
2506 	int channel)
2507 {
2508 	nxge_dma_common_t	*data;
2509 	nxge_dma_common_t	*control;
2510 	uint32_t 		num_chunks;
2511 
2512 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "==> nxge_alloc_rbb"));
2513 
2514 	data = nxgep->rx_buf_pool_p->dma_buf_pool_p[channel];
2515 	num_chunks = nxgep->rx_buf_pool_p->num_chunks[channel];
2516 	nxge_free_rx_buf_dma(nxgep, data, num_chunks);
2517 
2518 	nxgep->rx_buf_pool_p->dma_buf_pool_p[channel] = 0;
2519 	nxgep->rx_buf_pool_p->num_chunks[channel] = 0;
2520 
2521 	control = nxgep->rx_cntl_pool_p->dma_buf_pool_p[channel];
2522 	nxge_free_rx_cntl_dma(nxgep, control);
2523 
2524 	nxgep->rx_cntl_pool_p->dma_buf_pool_p[channel] = 0;
2525 
2526 	KMEM_FREE(data, sizeof (nxge_dma_common_t) * NXGE_DMA_BLOCK);
2527 	KMEM_FREE(control, sizeof (nxge_dma_common_t));
2528 
2529 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "<== nxge_alloc_rbb"));
2530 }
2531 
2532 static void
2533 nxge_free_rx_mem_pool(p_nxge_t nxgep)
2534 {
2535 	int rdc_max = NXGE_MAX_RDCS;
2536 
2537 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_free_rx_mem_pool"));
2538 
2539 	if (!nxgep->rx_buf_pool_p || !nxgep->rx_buf_pool_p->buf_allocated) {
2540 		NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
2541 		    "<== nxge_free_rx_mem_pool "
2542 		    "(null rx buf pool or buf not allocated"));
2543 		return;
2544 	}
2545 	if (!nxgep->rx_cntl_pool_p || !nxgep->rx_cntl_pool_p->buf_allocated) {
2546 		NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
2547 		    "<== nxge_free_rx_mem_pool "
2548 		    "(null rx cntl buf pool or cntl buf not allocated"));
2549 		return;
2550 	}
2551 
2552 	KMEM_FREE(nxgep->rx_cntl_pool_p->dma_buf_pool_p,
2553 	    sizeof (p_nxge_dma_common_t) * rdc_max);
2554 	KMEM_FREE(nxgep->rx_cntl_pool_p, sizeof (nxge_dma_pool_t));
2555 
2556 	KMEM_FREE(nxgep->rx_buf_pool_p->num_chunks,
2557 	    sizeof (uint32_t) * rdc_max);
2558 	KMEM_FREE(nxgep->rx_buf_pool_p->dma_buf_pool_p,
2559 	    sizeof (p_nxge_dma_common_t) * rdc_max);
2560 	KMEM_FREE(nxgep->rx_buf_pool_p, sizeof (nxge_dma_pool_t));
2561 
2562 	nxgep->rx_buf_pool_p = 0;
2563 	nxgep->rx_cntl_pool_p = 0;
2564 
2565 	KMEM_FREE(nxgep->rx_rbr_rings->rbr_rings,
2566 	    sizeof (p_rx_rbr_ring_t) * rdc_max);
2567 	KMEM_FREE(nxgep->rx_rbr_rings, sizeof (rx_rbr_rings_t));
2568 	KMEM_FREE(nxgep->rx_rcr_rings->rcr_rings,
2569 	    sizeof (p_rx_rcr_ring_t) * rdc_max);
2570 	KMEM_FREE(nxgep->rx_rcr_rings, sizeof (rx_rcr_rings_t));
2571 	KMEM_FREE(nxgep->rx_mbox_areas_p->rxmbox_areas,
2572 	    sizeof (p_rx_mbox_t) * rdc_max);
2573 	KMEM_FREE(nxgep->rx_mbox_areas_p, sizeof (rx_mbox_areas_t));
2574 
2575 	nxgep->rx_rbr_rings = 0;
2576 	nxgep->rx_rcr_rings = 0;
2577 	nxgep->rx_mbox_areas_p = 0;
2578 
2579 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "<== nxge_free_rx_mem_pool"));
2580 }
2581 
2582 
2583 static nxge_status_t
2584 nxge_alloc_rx_buf_dma(p_nxge_t nxgep, uint16_t dma_channel,
2585 	p_nxge_dma_common_t *dmap,
2586 	size_t alloc_size, size_t block_size, uint32_t *num_chunks)
2587 {
2588 	p_nxge_dma_common_t 	rx_dmap;
2589 	nxge_status_t		status = NXGE_OK;
2590 	size_t			total_alloc_size;
2591 	size_t			allocated = 0;
2592 	int			i, size_index, array_size;
2593 	boolean_t		use_kmem_alloc = B_FALSE;
2594 
2595 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "==> nxge_alloc_rx_buf_dma"));
2596 
2597 	rx_dmap = (p_nxge_dma_common_t)
2598 	    KMEM_ZALLOC(sizeof (nxge_dma_common_t) * NXGE_DMA_BLOCK,
2599 	    KM_SLEEP);
2600 
2601 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
2602 	    " alloc_rx_buf_dma rdc %d asize %x bsize %x bbuf %llx ",
2603 	    dma_channel, alloc_size, block_size, dmap));
2604 
2605 	total_alloc_size = alloc_size;
2606 
2607 #if defined(RX_USE_RECLAIM_POST)
2608 	total_alloc_size = alloc_size + alloc_size/4;
2609 #endif
2610 
2611 	i = 0;
2612 	size_index = 0;
2613 	array_size =  sizeof (alloc_sizes)/sizeof (size_t);
2614 	while ((size_index < array_size) &&
2615 	    (alloc_sizes[size_index] < alloc_size))
2616 		size_index++;
2617 	if (size_index >= array_size) {
2618 		size_index = array_size - 1;
2619 	}
2620 
2621 	/* For Neptune, use kmem_alloc if the kmem flag is set. */
2622 	if (nxgep->niu_type != N2_NIU && nxge_use_kmem_alloc) {
2623 		use_kmem_alloc = B_TRUE;
2624 #if defined(__i386) || defined(__amd64)
2625 		size_index = 0;
2626 #endif
2627 		NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
2628 		    "==> nxge_alloc_rx_buf_dma: "
2629 		    "Neptune use kmem_alloc() - size_index %d",
2630 		    size_index));
2631 	}
2632 
2633 	while ((allocated < total_alloc_size) &&
2634 	    (size_index >= 0) && (i < NXGE_DMA_BLOCK)) {
2635 		rx_dmap[i].dma_chunk_index = i;
2636 		rx_dmap[i].block_size = block_size;
2637 		rx_dmap[i].alength = alloc_sizes[size_index];
2638 		rx_dmap[i].orig_alength = rx_dmap[i].alength;
2639 		rx_dmap[i].nblocks = alloc_sizes[size_index] / block_size;
2640 		rx_dmap[i].dma_channel = dma_channel;
2641 		rx_dmap[i].contig_alloc_type = B_FALSE;
2642 		rx_dmap[i].kmem_alloc_type = B_FALSE;
2643 		rx_dmap[i].buf_alloc_type = DDI_MEM_ALLOC;
2644 
2645 		/*
2646 		 * N2/NIU: data buffers must be contiguous as the driver
2647 		 *	   needs to call Hypervisor api to set up
2648 		 *	   logical pages.
2649 		 */
2650 		if ((nxgep->niu_type == N2_NIU) && (NXGE_DMA_BLOCK == 1)) {
2651 			rx_dmap[i].contig_alloc_type = B_TRUE;
2652 			rx_dmap[i].buf_alloc_type = CONTIG_MEM_ALLOC;
2653 		} else if (use_kmem_alloc) {
2654 			/* For Neptune, use kmem_alloc */
2655 			NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
2656 			    "==> nxge_alloc_rx_buf_dma: "
2657 			    "Neptune use kmem_alloc()"));
2658 			rx_dmap[i].kmem_alloc_type = B_TRUE;
2659 			rx_dmap[i].buf_alloc_type = KMEM_ALLOC;
2660 		}
2661 
2662 		NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
2663 		    "alloc_rx_buf_dma rdc %d chunk %d bufp %llx size %x "
2664 		    "i %d nblocks %d alength %d",
2665 		    dma_channel, i, &rx_dmap[i], block_size,
2666 		    i, rx_dmap[i].nblocks,
2667 		    rx_dmap[i].alength));
2668 		status = nxge_dma_mem_alloc(nxgep, nxge_force_dma,
2669 		    &nxge_rx_dma_attr,
2670 		    rx_dmap[i].alength,
2671 		    &nxge_dev_buf_dma_acc_attr,
2672 		    DDI_DMA_READ | DDI_DMA_STREAMING,
2673 		    (p_nxge_dma_common_t)(&rx_dmap[i]));
2674 		if (status != NXGE_OK) {
2675 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
2676 			    "nxge_alloc_rx_buf_dma: Alloc Failed: "
2677 			    "dma %d size_index %d size requested %d",
2678 			    dma_channel,
2679 			    size_index,
2680 			    rx_dmap[i].alength));
2681 			size_index--;
2682 		} else {
2683 			rx_dmap[i].buf_alloc_state = BUF_ALLOCATED;
2684 			NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
2685 			    " nxge_alloc_rx_buf_dma DONE  alloc mem: "
2686 			    "dma %d dma_buf_p $%p kaddrp $%p alength %d "
2687 			    "buf_alloc_state %d alloc_type %d",
2688 			    dma_channel,
2689 			    &rx_dmap[i],
2690 			    rx_dmap[i].kaddrp,
2691 			    rx_dmap[i].alength,
2692 			    rx_dmap[i].buf_alloc_state,
2693 			    rx_dmap[i].buf_alloc_type));
2694 			NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
2695 			    " alloc_rx_buf_dma allocated rdc %d "
2696 			    "chunk %d size %x dvma %x bufp %llx kaddrp $%p",
2697 			    dma_channel, i, rx_dmap[i].alength,
2698 			    rx_dmap[i].ioaddr_pp, &rx_dmap[i],
2699 			    rx_dmap[i].kaddrp));
2700 			i++;
2701 			allocated += alloc_sizes[size_index];
2702 		}
2703 	}
2704 
2705 	if (allocated < total_alloc_size) {
2706 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
2707 		    "==> nxge_alloc_rx_buf_dma: not enough for channel %d "
2708 		    "allocated 0x%x requested 0x%x",
2709 		    dma_channel,
2710 		    allocated, total_alloc_size));
2711 		status = NXGE_ERROR;
2712 		goto nxge_alloc_rx_mem_fail1;
2713 	}
2714 
2715 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
2716 	    "==> nxge_alloc_rx_buf_dma: Allocated for channel %d "
2717 	    "allocated 0x%x requested 0x%x",
2718 	    dma_channel,
2719 	    allocated, total_alloc_size));
2720 
2721 	NXGE_DEBUG_MSG((nxgep, DMA_CTL,
2722 	    " alloc_rx_buf_dma rdc %d allocated %d chunks",
2723 	    dma_channel, i));
2724 	*num_chunks = i;
2725 	*dmap = rx_dmap;
2726 
2727 	goto nxge_alloc_rx_mem_exit;
2728 
2729 nxge_alloc_rx_mem_fail1:
2730 	KMEM_FREE(rx_dmap, sizeof (nxge_dma_common_t) * NXGE_DMA_BLOCK);
2731 
2732 nxge_alloc_rx_mem_exit:
2733 	NXGE_DEBUG_MSG((nxgep, DMA_CTL,
2734 	    "<== nxge_alloc_rx_buf_dma status 0x%08x", status));
2735 
2736 	return (status);
2737 }
2738 
2739 /*ARGSUSED*/
2740 static void
2741 nxge_free_rx_buf_dma(p_nxge_t nxgep, p_nxge_dma_common_t dmap,
2742     uint32_t num_chunks)
2743 {
2744 	int		i;
2745 
2746 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
2747 	    "==> nxge_free_rx_buf_dma: # of chunks %d", num_chunks));
2748 
2749 	if (dmap == 0)
2750 		return;
2751 
2752 	for (i = 0; i < num_chunks; i++) {
2753 		NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
2754 		    "==> nxge_free_rx_buf_dma: chunk %d dmap 0x%llx",
2755 		    i, dmap));
2756 		nxge_dma_free_rx_data_buf(dmap++);
2757 	}
2758 
2759 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_free_rx_buf_dma"));
2760 }
2761 
2762 /*ARGSUSED*/
2763 static nxge_status_t
2764 nxge_alloc_rx_cntl_dma(p_nxge_t nxgep, uint16_t dma_channel,
2765     p_nxge_dma_common_t *dmap, size_t size)
2766 {
2767 	p_nxge_dma_common_t 	rx_dmap;
2768 	nxge_status_t		status = NXGE_OK;
2769 
2770 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "==> nxge_alloc_rx_cntl_dma"));
2771 
2772 	rx_dmap = (p_nxge_dma_common_t)
2773 	    KMEM_ZALLOC(sizeof (nxge_dma_common_t), KM_SLEEP);
2774 
2775 	rx_dmap->contig_alloc_type = B_FALSE;
2776 	rx_dmap->kmem_alloc_type = B_FALSE;
2777 
2778 	status = nxge_dma_mem_alloc(nxgep, nxge_force_dma,
2779 	    &nxge_desc_dma_attr,
2780 	    size,
2781 	    &nxge_dev_desc_dma_acc_attr,
2782 	    DDI_DMA_RDWR | DDI_DMA_CONSISTENT,
2783 	    rx_dmap);
2784 	if (status != NXGE_OK) {
2785 		goto nxge_alloc_rx_cntl_dma_fail1;
2786 	}
2787 
2788 	*dmap = rx_dmap;
2789 	goto nxge_alloc_rx_cntl_dma_exit;
2790 
2791 nxge_alloc_rx_cntl_dma_fail1:
2792 	KMEM_FREE(rx_dmap, sizeof (nxge_dma_common_t));
2793 
2794 nxge_alloc_rx_cntl_dma_exit:
2795 	NXGE_DEBUG_MSG((nxgep, DMA_CTL,
2796 	    "<== nxge_alloc_rx_cntl_dma status 0x%08x", status));
2797 
2798 	return (status);
2799 }
2800 
2801 /*ARGSUSED*/
2802 static void
2803 nxge_free_rx_cntl_dma(p_nxge_t nxgep, p_nxge_dma_common_t dmap)
2804 {
2805 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "==> nxge_free_rx_cntl_dma"));
2806 
2807 	if (dmap == 0)
2808 		return;
2809 
2810 	nxge_dma_mem_free(dmap);
2811 
2812 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "<== nxge_free_rx_cntl_dma"));
2813 }
2814 
2815 typedef struct {
2816 	size_t	tx_size;
2817 	size_t	cr_size;
2818 	size_t	threshhold;
2819 } nxge_tdc_sizes_t;
2820 
2821 static
2822 nxge_status_t
2823 nxge_tdc_sizes(
2824 	nxge_t *nxgep,
2825 	nxge_tdc_sizes_t *sizes)
2826 {
2827 	uint32_t threshhold;	/* The bcopy() threshhold */
2828 	size_t tx_size;		/* Transmit buffer size */
2829 	size_t cr_size;		/* Completion ring size */
2830 
2831 	/*
2832 	 * Assume that each DMA channel will be configured with the
2833 	 * default transmit buffer size for copying transmit data.
2834 	 * (If a packet is bigger than this, it will not be copied.)
2835 	 */
2836 	if (nxgep->niu_type == N2_NIU) {
2837 		threshhold = TX_BCOPY_SIZE;
2838 	} else {
2839 		threshhold = nxge_bcopy_thresh;
2840 	}
2841 	tx_size = nxge_tx_ring_size * threshhold;
2842 
2843 	cr_size = nxge_tx_ring_size * sizeof (tx_desc_t);
2844 	cr_size += sizeof (txdma_mailbox_t);
2845 
2846 #if	defined(sun4v) && defined(NIU_LP_WORKAROUND)
2847 	if (nxgep->niu_type == N2_NIU) {
2848 		if (!ISP2(tx_size)) {
2849 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
2850 			    "==> nxge_tdc_sizes: Tx size"
2851 			    " must be power of 2"));
2852 			return (NXGE_ERROR);
2853 		}
2854 
2855 		if (tx_size > (1 << 22)) {
2856 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
2857 			    "==> nxge_tdc_sizes: Tx size"
2858 			    " limited to 4M"));
2859 			return (NXGE_ERROR);
2860 		}
2861 
2862 		if (cr_size < 0x2000)
2863 			cr_size = 0x2000;
2864 	}
2865 #endif
2866 
2867 	sizes->threshhold = threshhold;
2868 	sizes->tx_size = tx_size;
2869 	sizes->cr_size = cr_size;
2870 
2871 	return (NXGE_OK);
2872 }
2873 /*
2874  * nxge_alloc_txb
2875  *
2876  *	Allocate buffers for an TDC.
2877  *
2878  * Arguments:
2879  * 	nxgep
2880  * 	channel	The channel to map into our kernel space.
2881  *
2882  * Notes:
2883  *
2884  * NPI function calls:
2885  *
2886  * NXGE function calls:
2887  *
2888  * Registers accessed:
2889  *
2890  * Context:
2891  *
2892  * Taking apart:
2893  *
2894  * Open questions:
2895  *
2896  */
2897 nxge_status_t
2898 nxge_alloc_txb(
2899 	p_nxge_t nxgep,
2900 	int channel)
2901 {
2902 	nxge_dma_common_t	**dma_buf_p;
2903 	nxge_dma_common_t	**dma_cntl_p;
2904 	uint32_t 		*num_chunks;
2905 	nxge_status_t		status = NXGE_OK;
2906 
2907 	nxge_tdc_sizes_t	sizes;
2908 
2909 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "==> nxge_alloc_tbb"));
2910 
2911 	if (nxge_tdc_sizes(nxgep, &sizes) != NXGE_OK)
2912 		return (NXGE_ERROR);
2913 
2914 	/*
2915 	 * Allocate memory for transmit buffers and descriptor rings.
2916 	 * Replace these allocation functions with the interface functions
2917 	 * provided by the partition manager Real Soon Now.
2918 	 */
2919 	dma_buf_p = &nxgep->tx_buf_pool_p->dma_buf_pool_p[channel];
2920 	num_chunks = &nxgep->tx_buf_pool_p->num_chunks[channel];
2921 
2922 	dma_cntl_p = &nxgep->tx_cntl_pool_p->dma_buf_pool_p[channel];
2923 
2924 	/*
2925 	 * Allocate memory for transmit buffers and descriptor rings.
2926 	 * Replace allocation functions with interface functions provided
2927 	 * by the partition manager when it is available.
2928 	 *
2929 	 * Allocate memory for the transmit buffer pool.
2930 	 */
2931 	NXGE_DEBUG_MSG((nxgep, DMA_CTL,
2932 	    "sizes: tx: %ld, cr:%ld, th:%ld",
2933 	    sizes.tx_size, sizes.cr_size, sizes.threshhold));
2934 
2935 	*num_chunks = 0;
2936 	status = nxge_alloc_tx_buf_dma(nxgep, channel, dma_buf_p,
2937 	    sizes.tx_size, sizes.threshhold, num_chunks);
2938 	if (status != NXGE_OK) {
2939 		cmn_err(CE_NOTE, "nxge_alloc_tx_buf_dma failed!");
2940 		return (status);
2941 	}
2942 
2943 	/*
2944 	 * Allocate memory for descriptor rings and mailbox.
2945 	 */
2946 	status = nxge_alloc_tx_cntl_dma(nxgep, channel, dma_cntl_p,
2947 	    sizes.cr_size);
2948 	if (status != NXGE_OK) {
2949 		nxge_free_tx_buf_dma(nxgep, *dma_buf_p, *num_chunks);
2950 		cmn_err(CE_NOTE, "nxge_alloc_tx_cntl_dma failed!");
2951 		return (status);
2952 	}
2953 
2954 	return (NXGE_OK);
2955 }
2956 
2957 void
2958 nxge_free_txb(
2959 	p_nxge_t nxgep,
2960 	int channel)
2961 {
2962 	nxge_dma_common_t	*data;
2963 	nxge_dma_common_t	*control;
2964 	uint32_t 		num_chunks;
2965 
2966 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "==> nxge_free_txb"));
2967 
2968 	data = nxgep->tx_buf_pool_p->dma_buf_pool_p[channel];
2969 	num_chunks = nxgep->tx_buf_pool_p->num_chunks[channel];
2970 	nxge_free_tx_buf_dma(nxgep, data, num_chunks);
2971 
2972 	nxgep->tx_buf_pool_p->dma_buf_pool_p[channel] = 0;
2973 	nxgep->tx_buf_pool_p->num_chunks[channel] = 0;
2974 
2975 	control = nxgep->tx_cntl_pool_p->dma_buf_pool_p[channel];
2976 	nxge_free_tx_cntl_dma(nxgep, control);
2977 
2978 	nxgep->tx_cntl_pool_p->dma_buf_pool_p[channel] = 0;
2979 
2980 	KMEM_FREE(data, sizeof (nxge_dma_common_t) * NXGE_DMA_BLOCK);
2981 	KMEM_FREE(control, sizeof (nxge_dma_common_t));
2982 
2983 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "<== nxge_free_txb"));
2984 }
2985 
2986 /*
2987  * nxge_alloc_tx_mem_pool
2988  *
2989  *	This function allocates all of the per-port TDC control data structures.
2990  *	The per-channel (TDC) data structures are allocated when needed.
2991  *
2992  * Arguments:
2993  * 	nxgep
2994  *
2995  * Notes:
2996  *
2997  * Context:
2998  *	Any domain
2999  */
3000 nxge_status_t
3001 nxge_alloc_tx_mem_pool(p_nxge_t nxgep)
3002 {
3003 	nxge_hw_pt_cfg_t	*p_cfgp;
3004 	nxge_dma_pool_t		*dma_poolp;
3005 	nxge_dma_common_t	**dma_buf_p;
3006 	nxge_dma_pool_t		*dma_cntl_poolp;
3007 	nxge_dma_common_t	**dma_cntl_p;
3008 	uint32_t		*num_chunks; /* per dma */
3009 	int			tdc_max;
3010 
3011 	NXGE_DEBUG_MSG((nxgep, MEM_CTL, "==> nxge_alloc_tx_mem_pool"));
3012 
3013 	p_cfgp = &nxgep->pt_config.hw_config;
3014 	tdc_max = NXGE_MAX_TDCS;
3015 
3016 	/*
3017 	 * Allocate memory for each transmit DMA channel.
3018 	 */
3019 	dma_poolp = (p_nxge_dma_pool_t)KMEM_ZALLOC(sizeof (nxge_dma_pool_t),
3020 	    KM_SLEEP);
3021 	dma_buf_p = (p_nxge_dma_common_t *)KMEM_ZALLOC(
3022 	    sizeof (p_nxge_dma_common_t) * tdc_max, KM_SLEEP);
3023 
3024 	dma_cntl_poolp = (p_nxge_dma_pool_t)
3025 	    KMEM_ZALLOC(sizeof (nxge_dma_pool_t), KM_SLEEP);
3026 	dma_cntl_p = (p_nxge_dma_common_t *)KMEM_ZALLOC(
3027 	    sizeof (p_nxge_dma_common_t) * tdc_max, KM_SLEEP);
3028 
3029 	if (nxge_tx_ring_size > TDC_DEFAULT_MAX) {
3030 		NXGE_DEBUG_MSG((nxgep, MEM_CTL,
3031 		    "nxge_alloc_tx_mem_pool: TDC too high %d, "
3032 		    "set to default %d",
3033 		    nxge_tx_ring_size, TDC_DEFAULT_MAX));
3034 		nxge_tx_ring_size = TDC_DEFAULT_MAX;
3035 	}
3036 
3037 #if	defined(sun4v) && defined(NIU_LP_WORKAROUND)
3038 	/*
3039 	 * N2/NIU has limitation on the descriptor sizes (contiguous
3040 	 * memory allocation on data buffers to 4M (contig_mem_alloc)
3041 	 * and little endian for control buffers (must use the ddi/dki mem alloc
3042 	 * function). The transmit ring is limited to 8K (includes the
3043 	 * mailbox).
3044 	 */
3045 	if (nxgep->niu_type == N2_NIU) {
3046 		if ((nxge_tx_ring_size > NXGE_NIU_CONTIG_TX_MAX) ||
3047 		    (!ISP2(nxge_tx_ring_size))) {
3048 			nxge_tx_ring_size = NXGE_NIU_CONTIG_TX_MAX;
3049 		}
3050 	}
3051 #endif
3052 
3053 	nxgep->nxge_port_tx_ring_size = nxge_tx_ring_size;
3054 
3055 	num_chunks = (uint32_t *)KMEM_ZALLOC(
3056 	    sizeof (uint32_t) * tdc_max, KM_SLEEP);
3057 
3058 	dma_poolp->ndmas = p_cfgp->tdc.owned;
3059 	dma_poolp->num_chunks = num_chunks;
3060 	dma_poolp->dma_buf_pool_p = dma_buf_p;
3061 	nxgep->tx_buf_pool_p = dma_poolp;
3062 
3063 	dma_poolp->buf_allocated = B_TRUE;
3064 
3065 	dma_cntl_poolp->ndmas = p_cfgp->tdc.owned;
3066 	dma_cntl_poolp->dma_buf_pool_p = dma_cntl_p;
3067 	nxgep->tx_cntl_pool_p = dma_cntl_poolp;
3068 
3069 	dma_cntl_poolp->buf_allocated = B_TRUE;
3070 
3071 	nxgep->tx_rings =
3072 	    KMEM_ZALLOC(sizeof (tx_rings_t), KM_SLEEP);
3073 	nxgep->tx_rings->rings =
3074 	    KMEM_ZALLOC(sizeof (p_tx_ring_t) * tdc_max, KM_SLEEP);
3075 	nxgep->tx_mbox_areas_p =
3076 	    KMEM_ZALLOC(sizeof (tx_mbox_areas_t), KM_SLEEP);
3077 	nxgep->tx_mbox_areas_p->txmbox_areas_p =
3078 	    KMEM_ZALLOC(sizeof (p_tx_mbox_t) * tdc_max, KM_SLEEP);
3079 
3080 	nxgep->tx_rings->ndmas = p_cfgp->tdc.owned;
3081 
3082 	NXGE_DEBUG_MSG((nxgep, MEM_CTL,
3083 	    "==> nxge_alloc_tx_mem_pool: ndmas %d poolp->ndmas %d",
3084 	    tdc_max, dma_poolp->ndmas));
3085 
3086 	return (NXGE_OK);
3087 }
3088 
3089 nxge_status_t
3090 nxge_alloc_tx_buf_dma(p_nxge_t nxgep, uint16_t dma_channel,
3091     p_nxge_dma_common_t *dmap, size_t alloc_size,
3092     size_t block_size, uint32_t *num_chunks)
3093 {
3094 	p_nxge_dma_common_t 	tx_dmap;
3095 	nxge_status_t		status = NXGE_OK;
3096 	size_t			total_alloc_size;
3097 	size_t			allocated = 0;
3098 	int			i, size_index, array_size;
3099 
3100 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "==> nxge_alloc_tx_buf_dma"));
3101 
3102 	tx_dmap = (p_nxge_dma_common_t)
3103 	    KMEM_ZALLOC(sizeof (nxge_dma_common_t) * NXGE_DMA_BLOCK,
3104 	    KM_SLEEP);
3105 
3106 	total_alloc_size = alloc_size;
3107 	i = 0;
3108 	size_index = 0;
3109 	array_size =  sizeof (alloc_sizes) /  sizeof (size_t);
3110 	while ((size_index < array_size) &&
3111 	    (alloc_sizes[size_index] < alloc_size))
3112 		size_index++;
3113 	if (size_index >= array_size) {
3114 		size_index = array_size - 1;
3115 	}
3116 
3117 	while ((allocated < total_alloc_size) &&
3118 	    (size_index >= 0) && (i < NXGE_DMA_BLOCK)) {
3119 
3120 		tx_dmap[i].dma_chunk_index = i;
3121 		tx_dmap[i].block_size = block_size;
3122 		tx_dmap[i].alength = alloc_sizes[size_index];
3123 		tx_dmap[i].orig_alength = tx_dmap[i].alength;
3124 		tx_dmap[i].nblocks = alloc_sizes[size_index] / block_size;
3125 		tx_dmap[i].dma_channel = dma_channel;
3126 		tx_dmap[i].contig_alloc_type = B_FALSE;
3127 		tx_dmap[i].kmem_alloc_type = B_FALSE;
3128 
3129 		/*
3130 		 * N2/NIU: data buffers must be contiguous as the driver
3131 		 *	   needs to call Hypervisor api to set up
3132 		 *	   logical pages.
3133 		 */
3134 		if ((nxgep->niu_type == N2_NIU) && (NXGE_DMA_BLOCK == 1)) {
3135 			tx_dmap[i].contig_alloc_type = B_TRUE;
3136 		}
3137 
3138 		status = nxge_dma_mem_alloc(nxgep, nxge_force_dma,
3139 		    &nxge_tx_dma_attr,
3140 		    tx_dmap[i].alength,
3141 		    &nxge_dev_buf_dma_acc_attr,
3142 		    DDI_DMA_WRITE | DDI_DMA_STREAMING,
3143 		    (p_nxge_dma_common_t)(&tx_dmap[i]));
3144 		if (status != NXGE_OK) {
3145 			size_index--;
3146 		} else {
3147 			i++;
3148 			allocated += alloc_sizes[size_index];
3149 		}
3150 	}
3151 
3152 	if (allocated < total_alloc_size) {
3153 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3154 		    "==> nxge_alloc_tx_buf_dma: not enough channel %d: "
3155 		    "allocated 0x%x requested 0x%x",
3156 		    dma_channel,
3157 		    allocated, total_alloc_size));
3158 		status = NXGE_ERROR;
3159 		goto nxge_alloc_tx_mem_fail1;
3160 	}
3161 
3162 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3163 	    "==> nxge_alloc_tx_buf_dma: Allocated for channel %d: "
3164 	    "allocated 0x%x requested 0x%x",
3165 	    dma_channel,
3166 	    allocated, total_alloc_size));
3167 
3168 	*num_chunks = i;
3169 	*dmap = tx_dmap;
3170 	NXGE_DEBUG_MSG((nxgep, DMA_CTL,
3171 	    "==> nxge_alloc_tx_buf_dma dmap 0x%016llx num chunks %d",
3172 	    *dmap, i));
3173 	goto nxge_alloc_tx_mem_exit;
3174 
3175 nxge_alloc_tx_mem_fail1:
3176 	KMEM_FREE(tx_dmap, sizeof (nxge_dma_common_t) * NXGE_DMA_BLOCK);
3177 
3178 nxge_alloc_tx_mem_exit:
3179 	NXGE_DEBUG_MSG((nxgep, DMA_CTL,
3180 	    "<== nxge_alloc_tx_buf_dma status 0x%08x", status));
3181 
3182 	return (status);
3183 }
3184 
3185 /*ARGSUSED*/
3186 static void
3187 nxge_free_tx_buf_dma(p_nxge_t nxgep, p_nxge_dma_common_t dmap,
3188     uint32_t num_chunks)
3189 {
3190 	int		i;
3191 
3192 	NXGE_DEBUG_MSG((nxgep, MEM_CTL, "==> nxge_free_tx_buf_dma"));
3193 
3194 	if (dmap == 0)
3195 		return;
3196 
3197 	for (i = 0; i < num_chunks; i++) {
3198 		nxge_dma_mem_free(dmap++);
3199 	}
3200 
3201 	NXGE_DEBUG_MSG((nxgep, MEM_CTL, "<== nxge_free_tx_buf_dma"));
3202 }
3203 
3204 /*ARGSUSED*/
3205 nxge_status_t
3206 nxge_alloc_tx_cntl_dma(p_nxge_t nxgep, uint16_t dma_channel,
3207     p_nxge_dma_common_t *dmap, size_t size)
3208 {
3209 	p_nxge_dma_common_t 	tx_dmap;
3210 	nxge_status_t		status = NXGE_OK;
3211 
3212 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "==> nxge_alloc_tx_cntl_dma"));
3213 	tx_dmap = (p_nxge_dma_common_t)
3214 	    KMEM_ZALLOC(sizeof (nxge_dma_common_t), KM_SLEEP);
3215 
3216 	tx_dmap->contig_alloc_type = B_FALSE;
3217 	tx_dmap->kmem_alloc_type = B_FALSE;
3218 
3219 	status = nxge_dma_mem_alloc(nxgep, nxge_force_dma,
3220 	    &nxge_desc_dma_attr,
3221 	    size,
3222 	    &nxge_dev_desc_dma_acc_attr,
3223 	    DDI_DMA_RDWR | DDI_DMA_CONSISTENT,
3224 	    tx_dmap);
3225 	if (status != NXGE_OK) {
3226 		goto nxge_alloc_tx_cntl_dma_fail1;
3227 	}
3228 
3229 	*dmap = tx_dmap;
3230 	goto nxge_alloc_tx_cntl_dma_exit;
3231 
3232 nxge_alloc_tx_cntl_dma_fail1:
3233 	KMEM_FREE(tx_dmap, sizeof (nxge_dma_common_t));
3234 
3235 nxge_alloc_tx_cntl_dma_exit:
3236 	NXGE_DEBUG_MSG((nxgep, DMA_CTL,
3237 	    "<== nxge_alloc_tx_cntl_dma status 0x%08x", status));
3238 
3239 	return (status);
3240 }
3241 
3242 /*ARGSUSED*/
3243 static void
3244 nxge_free_tx_cntl_dma(p_nxge_t nxgep, p_nxge_dma_common_t dmap)
3245 {
3246 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "==> nxge_free_tx_cntl_dma"));
3247 
3248 	if (dmap == 0)
3249 		return;
3250 
3251 	nxge_dma_mem_free(dmap);
3252 
3253 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "<== nxge_free_tx_cntl_dma"));
3254 }
3255 
3256 /*
3257  * nxge_free_tx_mem_pool
3258  *
3259  *	This function frees all of the per-port TDC control data structures.
3260  *	The per-channel (TDC) data structures are freed when the channel
3261  *	is stopped.
3262  *
3263  * Arguments:
3264  * 	nxgep
3265  *
3266  * Notes:
3267  *
3268  * Context:
3269  *	Any domain
3270  */
3271 static void
3272 nxge_free_tx_mem_pool(p_nxge_t nxgep)
3273 {
3274 	int tdc_max = NXGE_MAX_TDCS;
3275 
3276 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "==> nxge_free_tx_mem_pool"));
3277 
3278 	if (!nxgep->tx_buf_pool_p || !nxgep->tx_buf_pool_p->buf_allocated) {
3279 		NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3280 		    "<== nxge_free_tx_mem_pool "
3281 		    "(null tx buf pool or buf not allocated"));
3282 		return;
3283 	}
3284 	if (!nxgep->tx_cntl_pool_p || !nxgep->tx_cntl_pool_p->buf_allocated) {
3285 		NXGE_DEBUG_MSG((nxgep, MEM2_CTL,
3286 		    "<== nxge_free_tx_mem_pool "
3287 		    "(null tx cntl buf pool or cntl buf not allocated"));
3288 		return;
3289 	}
3290 
3291 	/* 1. Free the mailboxes. */
3292 	KMEM_FREE(nxgep->tx_mbox_areas_p->txmbox_areas_p,
3293 	    sizeof (p_tx_mbox_t) * tdc_max);
3294 	KMEM_FREE(nxgep->tx_mbox_areas_p, sizeof (tx_mbox_areas_t));
3295 
3296 	nxgep->tx_mbox_areas_p = 0;
3297 
3298 	/* 2. Free the transmit ring arrays. */
3299 	KMEM_FREE(nxgep->tx_rings->rings,
3300 	    sizeof (p_tx_ring_t) * tdc_max);
3301 	KMEM_FREE(nxgep->tx_rings, sizeof (tx_rings_t));
3302 
3303 	nxgep->tx_rings = 0;
3304 
3305 	/* 3. Free the completion ring data structures. */
3306 	KMEM_FREE(nxgep->tx_cntl_pool_p->dma_buf_pool_p,
3307 	    sizeof (p_nxge_dma_common_t) * tdc_max);
3308 	KMEM_FREE(nxgep->tx_cntl_pool_p, sizeof (nxge_dma_pool_t));
3309 
3310 	nxgep->tx_cntl_pool_p = 0;
3311 
3312 	/* 4. Free the data ring data structures. */
3313 	KMEM_FREE(nxgep->tx_buf_pool_p->num_chunks,
3314 	    sizeof (uint32_t) * tdc_max);
3315 	KMEM_FREE(nxgep->tx_buf_pool_p->dma_buf_pool_p,
3316 	    sizeof (p_nxge_dma_common_t) * tdc_max);
3317 	KMEM_FREE(nxgep->tx_buf_pool_p, sizeof (nxge_dma_pool_t));
3318 
3319 	nxgep->tx_buf_pool_p = 0;
3320 
3321 	NXGE_DEBUG_MSG((nxgep, MEM2_CTL, "<== nxge_free_tx_mem_pool"));
3322 }
3323 
3324 /*ARGSUSED*/
3325 static nxge_status_t
3326 nxge_dma_mem_alloc(p_nxge_t nxgep, dma_method_t method,
3327 	struct ddi_dma_attr *dma_attrp,
3328 	size_t length, ddi_device_acc_attr_t *acc_attr_p, uint_t xfer_flags,
3329 	p_nxge_dma_common_t dma_p)
3330 {
3331 	caddr_t 		kaddrp;
3332 	int			ddi_status = DDI_SUCCESS;
3333 	boolean_t		contig_alloc_type;
3334 	boolean_t		kmem_alloc_type;
3335 
3336 	contig_alloc_type = dma_p->contig_alloc_type;
3337 
3338 	if (contig_alloc_type && (nxgep->niu_type != N2_NIU)) {
3339 		/*
3340 		 * contig_alloc_type for contiguous memory only allowed
3341 		 * for N2/NIU.
3342 		 */
3343 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3344 		    "nxge_dma_mem_alloc: alloc type not allowed (%d)",
3345 		    dma_p->contig_alloc_type));
3346 		return (NXGE_ERROR | NXGE_DDI_FAILED);
3347 	}
3348 
3349 	dma_p->dma_handle = NULL;
3350 	dma_p->acc_handle = NULL;
3351 	dma_p->kaddrp = dma_p->last_kaddrp = NULL;
3352 	dma_p->first_ioaddr_pp = dma_p->last_ioaddr_pp = NULL;
3353 	ddi_status = ddi_dma_alloc_handle(nxgep->dip, dma_attrp,
3354 	    DDI_DMA_DONTWAIT, NULL, &dma_p->dma_handle);
3355 	if (ddi_status != DDI_SUCCESS) {
3356 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3357 		    "nxge_dma_mem_alloc:ddi_dma_alloc_handle failed."));
3358 		return (NXGE_ERROR | NXGE_DDI_FAILED);
3359 	}
3360 
3361 	kmem_alloc_type = dma_p->kmem_alloc_type;
3362 
3363 	switch (contig_alloc_type) {
3364 	case B_FALSE:
3365 		switch (kmem_alloc_type) {
3366 		case B_FALSE:
3367 			ddi_status = ddi_dma_mem_alloc(dma_p->dma_handle,
3368 			    length,
3369 			    acc_attr_p,
3370 			    xfer_flags,
3371 			    DDI_DMA_DONTWAIT, 0, &kaddrp, &dma_p->alength,
3372 			    &dma_p->acc_handle);
3373 			if (ddi_status != DDI_SUCCESS) {
3374 				NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3375 				    "nxge_dma_mem_alloc: "
3376 				    "ddi_dma_mem_alloc failed"));
3377 				ddi_dma_free_handle(&dma_p->dma_handle);
3378 				dma_p->dma_handle = NULL;
3379 				return (NXGE_ERROR | NXGE_DDI_FAILED);
3380 			}
3381 			if (dma_p->alength < length) {
3382 				NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3383 				    "nxge_dma_mem_alloc:di_dma_mem_alloc "
3384 				    "< length."));
3385 				ddi_dma_mem_free(&dma_p->acc_handle);
3386 				ddi_dma_free_handle(&dma_p->dma_handle);
3387 				dma_p->acc_handle = NULL;
3388 				dma_p->dma_handle = NULL;
3389 				return (NXGE_ERROR);
3390 			}
3391 
3392 			ddi_status = ddi_dma_addr_bind_handle(dma_p->dma_handle,
3393 			    NULL,
3394 			    kaddrp, dma_p->alength, xfer_flags,
3395 			    DDI_DMA_DONTWAIT,
3396 			    0, &dma_p->dma_cookie, &dma_p->ncookies);
3397 			if (ddi_status != DDI_DMA_MAPPED) {
3398 				NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3399 				    "nxge_dma_mem_alloc: ddi_dma_addr_bind "
3400 				    "failed "
3401 				    "(staus 0x%x ncookies %d.)", ddi_status,
3402 				    dma_p->ncookies));
3403 				if (dma_p->acc_handle) {
3404 					ddi_dma_mem_free(&dma_p->acc_handle);
3405 					dma_p->acc_handle = NULL;
3406 				}
3407 				ddi_dma_free_handle(&dma_p->dma_handle);
3408 				dma_p->dma_handle = NULL;
3409 				return (NXGE_ERROR | NXGE_DDI_FAILED);
3410 			}
3411 
3412 			if (dma_p->ncookies != 1) {
3413 				NXGE_DEBUG_MSG((nxgep, DMA_CTL,
3414 				    "nxge_dma_mem_alloc:ddi_dma_addr_bind "
3415 				    "> 1 cookie"
3416 				    "(staus 0x%x ncookies %d.)", ddi_status,
3417 				    dma_p->ncookies));
3418 				(void) ddi_dma_unbind_handle(dma_p->dma_handle);
3419 				if (dma_p->acc_handle) {
3420 					ddi_dma_mem_free(&dma_p->acc_handle);
3421 					dma_p->acc_handle = NULL;
3422 				}
3423 				ddi_dma_free_handle(&dma_p->dma_handle);
3424 				dma_p->dma_handle = NULL;
3425 				dma_p->acc_handle = NULL;
3426 				return (NXGE_ERROR);
3427 			}
3428 			break;
3429 
3430 		case B_TRUE:
3431 			kaddrp = KMEM_ALLOC(length, KM_NOSLEEP);
3432 			if (kaddrp == NULL) {
3433 				NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3434 				    "nxge_dma_mem_alloc:ddi_dma_mem_alloc "
3435 				    "kmem alloc failed"));
3436 				return (NXGE_ERROR);
3437 			}
3438 
3439 			dma_p->alength = length;
3440 			ddi_status = ddi_dma_addr_bind_handle(dma_p->dma_handle,
3441 			    NULL, kaddrp, dma_p->alength, xfer_flags,
3442 			    DDI_DMA_DONTWAIT, 0,
3443 			    &dma_p->dma_cookie, &dma_p->ncookies);
3444 			if (ddi_status != DDI_DMA_MAPPED) {
3445 				NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3446 				    "nxge_dma_mem_alloc:ddi_dma_addr_bind: "
3447 				    "(kmem_alloc) failed kaddrp $%p length %d "
3448 				    "(staus 0x%x (%d) ncookies %d.)",
3449 				    kaddrp, length,
3450 				    ddi_status, ddi_status, dma_p->ncookies));
3451 				KMEM_FREE(kaddrp, length);
3452 				dma_p->acc_handle = NULL;
3453 				ddi_dma_free_handle(&dma_p->dma_handle);
3454 				dma_p->dma_handle = NULL;
3455 				dma_p->kaddrp = NULL;
3456 				return (NXGE_ERROR | NXGE_DDI_FAILED);
3457 			}
3458 
3459 			if (dma_p->ncookies != 1) {
3460 				NXGE_DEBUG_MSG((nxgep, DMA_CTL,
3461 				    "nxge_dma_mem_alloc:ddi_dma_addr_bind "
3462 				    "(kmem_alloc) > 1 cookie"
3463 				    "(staus 0x%x ncookies %d.)", ddi_status,
3464 				    dma_p->ncookies));
3465 				(void) ddi_dma_unbind_handle(dma_p->dma_handle);
3466 				KMEM_FREE(kaddrp, length);
3467 				ddi_dma_free_handle(&dma_p->dma_handle);
3468 				dma_p->dma_handle = NULL;
3469 				dma_p->acc_handle = NULL;
3470 				dma_p->kaddrp = NULL;
3471 				return (NXGE_ERROR);
3472 			}
3473 
3474 			dma_p->kaddrp = kaddrp;
3475 
3476 			NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
3477 			    "nxge_dma_mem_alloc: kmem_alloc dmap $%p "
3478 			    "kaddr $%p alength %d",
3479 			    dma_p,
3480 			    kaddrp,
3481 			    dma_p->alength));
3482 			break;
3483 		}
3484 		break;
3485 
3486 #if	defined(sun4v) && defined(NIU_LP_WORKAROUND)
3487 	case B_TRUE:
3488 		kaddrp = (caddr_t)contig_mem_alloc(length);
3489 		if (kaddrp == NULL) {
3490 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3491 			    "nxge_dma_mem_alloc:contig_mem_alloc failed."));
3492 			ddi_dma_free_handle(&dma_p->dma_handle);
3493 			return (NXGE_ERROR | NXGE_DDI_FAILED);
3494 		}
3495 
3496 		dma_p->alength = length;
3497 		ddi_status = ddi_dma_addr_bind_handle(dma_p->dma_handle, NULL,
3498 		    kaddrp, dma_p->alength, xfer_flags, DDI_DMA_DONTWAIT, 0,
3499 		    &dma_p->dma_cookie, &dma_p->ncookies);
3500 		if (ddi_status != DDI_DMA_MAPPED) {
3501 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3502 			    "nxge_dma_mem_alloc:di_dma_addr_bind failed "
3503 			    "(status 0x%x ncookies %d.)", ddi_status,
3504 			    dma_p->ncookies));
3505 
3506 			NXGE_DEBUG_MSG((nxgep, DMA_CTL,
3507 			    "==> nxge_dma_mem_alloc: (not mapped)"
3508 			    "length %lu (0x%x) "
3509 			    "free contig kaddrp $%p "
3510 			    "va_to_pa $%p",
3511 			    length, length,
3512 			    kaddrp,
3513 			    va_to_pa(kaddrp)));
3514 
3515 
3516 			contig_mem_free((void *)kaddrp, length);
3517 			ddi_dma_free_handle(&dma_p->dma_handle);
3518 
3519 			dma_p->dma_handle = NULL;
3520 			dma_p->acc_handle = NULL;
3521 			dma_p->alength = NULL;
3522 			dma_p->kaddrp = NULL;
3523 
3524 			return (NXGE_ERROR | NXGE_DDI_FAILED);
3525 		}
3526 
3527 		if (dma_p->ncookies != 1 ||
3528 		    (dma_p->dma_cookie.dmac_laddress == NULL)) {
3529 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3530 			    "nxge_dma_mem_alloc:di_dma_addr_bind > 1 "
3531 			    "cookie or "
3532 			    "dmac_laddress is NULL $%p size %d "
3533 			    " (status 0x%x ncookies %d.)",
3534 			    ddi_status,
3535 			    dma_p->dma_cookie.dmac_laddress,
3536 			    dma_p->dma_cookie.dmac_size,
3537 			    dma_p->ncookies));
3538 
3539 			contig_mem_free((void *)kaddrp, length);
3540 			(void) ddi_dma_unbind_handle(dma_p->dma_handle);
3541 			ddi_dma_free_handle(&dma_p->dma_handle);
3542 
3543 			dma_p->alength = 0;
3544 			dma_p->dma_handle = NULL;
3545 			dma_p->acc_handle = NULL;
3546 			dma_p->kaddrp = NULL;
3547 
3548 			return (NXGE_ERROR | NXGE_DDI_FAILED);
3549 		}
3550 		break;
3551 
3552 #else
3553 	case B_TRUE:
3554 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3555 		    "nxge_dma_mem_alloc: invalid alloc type for !sun4v"));
3556 		return (NXGE_ERROR | NXGE_DDI_FAILED);
3557 #endif
3558 	}
3559 
3560 	dma_p->kaddrp = kaddrp;
3561 	dma_p->last_kaddrp = (unsigned char *)kaddrp +
3562 	    dma_p->alength - RXBUF_64B_ALIGNED;
3563 #if defined(__i386)
3564 	dma_p->ioaddr_pp =
3565 	    (unsigned char *)(uint32_t)dma_p->dma_cookie.dmac_laddress;
3566 #else
3567 	dma_p->ioaddr_pp = (unsigned char *)dma_p->dma_cookie.dmac_laddress;
3568 #endif
3569 	dma_p->last_ioaddr_pp =
3570 #if defined(__i386)
3571 	    (unsigned char *)(uint32_t)dma_p->dma_cookie.dmac_laddress +
3572 #else
3573 	    (unsigned char *)dma_p->dma_cookie.dmac_laddress +
3574 #endif
3575 	    dma_p->alength - RXBUF_64B_ALIGNED;
3576 
3577 	NPI_DMA_ACC_HANDLE_SET(dma_p, dma_p->acc_handle);
3578 
3579 #if	defined(sun4v) && defined(NIU_LP_WORKAROUND)
3580 	dma_p->orig_ioaddr_pp =
3581 	    (unsigned char *)dma_p->dma_cookie.dmac_laddress;
3582 	dma_p->orig_alength = length;
3583 	dma_p->orig_kaddrp = kaddrp;
3584 	dma_p->orig_vatopa = (uint64_t)va_to_pa(kaddrp);
3585 #endif
3586 
3587 	NXGE_DEBUG_MSG((nxgep, DMA_CTL, "<== nxge_dma_mem_alloc: "
3588 	    "dma buffer allocated: dma_p $%p "
3589 	    "return dmac_ladress from cookie $%p cookie dmac_size %d "
3590 	    "dma_p->ioaddr_p $%p "
3591 	    "dma_p->orig_ioaddr_p $%p "
3592 	    "orig_vatopa $%p "
3593 	    "alength %d (0x%x) "
3594 	    "kaddrp $%p "
3595 	    "length %d (0x%x)",
3596 	    dma_p,
3597 	    dma_p->dma_cookie.dmac_laddress, dma_p->dma_cookie.dmac_size,
3598 	    dma_p->ioaddr_pp,
3599 	    dma_p->orig_ioaddr_pp,
3600 	    dma_p->orig_vatopa,
3601 	    dma_p->alength, dma_p->alength,
3602 	    kaddrp,
3603 	    length, length));
3604 
3605 	return (NXGE_OK);
3606 }
3607 
3608 static void
3609 nxge_dma_mem_free(p_nxge_dma_common_t dma_p)
3610 {
3611 	if (dma_p->dma_handle != NULL) {
3612 		if (dma_p->ncookies) {
3613 			(void) ddi_dma_unbind_handle(dma_p->dma_handle);
3614 			dma_p->ncookies = 0;
3615 		}
3616 		ddi_dma_free_handle(&dma_p->dma_handle);
3617 		dma_p->dma_handle = NULL;
3618 	}
3619 
3620 	if (dma_p->acc_handle != NULL) {
3621 		ddi_dma_mem_free(&dma_p->acc_handle);
3622 		dma_p->acc_handle = NULL;
3623 		NPI_DMA_ACC_HANDLE_SET(dma_p, NULL);
3624 	}
3625 
3626 #if	defined(sun4v) && defined(NIU_LP_WORKAROUND)
3627 	if (dma_p->contig_alloc_type &&
3628 	    dma_p->orig_kaddrp && dma_p->orig_alength) {
3629 		NXGE_DEBUG_MSG((NULL, DMA_CTL, "nxge_dma_mem_free: "
3630 		    "kaddrp $%p (orig_kaddrp $%p)"
3631 		    "mem type %d ",
3632 		    "orig_alength %d "
3633 		    "alength 0x%x (%d)",
3634 		    dma_p->kaddrp,
3635 		    dma_p->orig_kaddrp,
3636 		    dma_p->contig_alloc_type,
3637 		    dma_p->orig_alength,
3638 		    dma_p->alength, dma_p->alength));
3639 
3640 		contig_mem_free(dma_p->orig_kaddrp, dma_p->orig_alength);
3641 		dma_p->orig_alength = NULL;
3642 		dma_p->orig_kaddrp = NULL;
3643 		dma_p->contig_alloc_type = B_FALSE;
3644 	}
3645 #endif
3646 	dma_p->kaddrp = NULL;
3647 	dma_p->alength = NULL;
3648 }
3649 
3650 static void
3651 nxge_dma_free_rx_data_buf(p_nxge_dma_common_t dma_p)
3652 {
3653 	uint64_t kaddr;
3654 	uint32_t buf_size;
3655 
3656 	NXGE_DEBUG_MSG((NULL, DMA_CTL, "==> nxge_dma_free_rx_data_buf"));
3657 
3658 	if (dma_p->dma_handle != NULL) {
3659 		if (dma_p->ncookies) {
3660 			(void) ddi_dma_unbind_handle(dma_p->dma_handle);
3661 			dma_p->ncookies = 0;
3662 		}
3663 		ddi_dma_free_handle(&dma_p->dma_handle);
3664 		dma_p->dma_handle = NULL;
3665 	}
3666 
3667 	if (dma_p->acc_handle != NULL) {
3668 		ddi_dma_mem_free(&dma_p->acc_handle);
3669 		dma_p->acc_handle = NULL;
3670 		NPI_DMA_ACC_HANDLE_SET(dma_p, NULL);
3671 	}
3672 
3673 	NXGE_DEBUG_MSG((NULL, DMA_CTL,
3674 	    "==> nxge_dma_free_rx_data_buf: dmap $%p buf_alloc_state %d",
3675 	    dma_p,
3676 	    dma_p->buf_alloc_state));
3677 
3678 	if (!(dma_p->buf_alloc_state & BUF_ALLOCATED_WAIT_FREE)) {
3679 		NXGE_DEBUG_MSG((NULL, DMA_CTL,
3680 		    "<== nxge_dma_free_rx_data_buf: "
3681 		    "outstanding data buffers"));
3682 		return;
3683 	}
3684 
3685 #if	defined(sun4v) && defined(NIU_LP_WORKAROUND)
3686 	if (dma_p->contig_alloc_type &&
3687 	    dma_p->orig_kaddrp && dma_p->orig_alength) {
3688 		NXGE_DEBUG_MSG((NULL, DMA_CTL, "nxge_dma_free_rx_data_buf: "
3689 		    "kaddrp $%p (orig_kaddrp $%p)"
3690 		    "mem type %d ",
3691 		    "orig_alength %d "
3692 		    "alength 0x%x (%d)",
3693 		    dma_p->kaddrp,
3694 		    dma_p->orig_kaddrp,
3695 		    dma_p->contig_alloc_type,
3696 		    dma_p->orig_alength,
3697 		    dma_p->alength, dma_p->alength));
3698 
3699 		kaddr = (uint64_t)dma_p->orig_kaddrp;
3700 		buf_size = dma_p->orig_alength;
3701 		nxge_free_buf(CONTIG_MEM_ALLOC, kaddr, buf_size);
3702 		dma_p->orig_alength = NULL;
3703 		dma_p->orig_kaddrp = NULL;
3704 		dma_p->contig_alloc_type = B_FALSE;
3705 		dma_p->kaddrp = NULL;
3706 		dma_p->alength = NULL;
3707 		return;
3708 	}
3709 #endif
3710 
3711 	if (dma_p->kmem_alloc_type) {
3712 		NXGE_DEBUG_MSG((NULL, DMA_CTL,
3713 		    "nxge_dma_free_rx_data_buf: free kmem "
3714 		    "kaddrp $%p (orig_kaddrp $%p)"
3715 		    "alloc type %d "
3716 		    "orig_alength %d "
3717 		    "alength 0x%x (%d)",
3718 		    dma_p->kaddrp,
3719 		    dma_p->orig_kaddrp,
3720 		    dma_p->kmem_alloc_type,
3721 		    dma_p->orig_alength,
3722 		    dma_p->alength, dma_p->alength));
3723 #if defined(__i386)
3724 		kaddr = (uint64_t)(uint32_t)dma_p->kaddrp;
3725 #else
3726 		kaddr = (uint64_t)dma_p->kaddrp;
3727 #endif
3728 		buf_size = dma_p->orig_alength;
3729 		NXGE_DEBUG_MSG((NULL, DMA_CTL,
3730 		    "nxge_dma_free_rx_data_buf: free dmap $%p "
3731 		    "kaddr $%p buf_size %d",
3732 		    dma_p,
3733 		    kaddr, buf_size));
3734 		nxge_free_buf(KMEM_ALLOC, kaddr, buf_size);
3735 		dma_p->alength = 0;
3736 		dma_p->orig_alength = 0;
3737 		dma_p->kaddrp = NULL;
3738 		dma_p->kmem_alloc_type = B_FALSE;
3739 	}
3740 
3741 	NXGE_DEBUG_MSG((NULL, DMA_CTL, "<== nxge_dma_free_rx_data_buf"));
3742 }
3743 
3744 /*
3745  *	nxge_m_start() -- start transmitting and receiving.
3746  *
3747  *	This function is called by the MAC layer when the first
3748  *	stream is open to prepare the hardware ready for sending
3749  *	and transmitting packets.
3750  */
3751 static int
3752 nxge_m_start(void *arg)
3753 {
3754 	p_nxge_t 	nxgep = (p_nxge_t)arg;
3755 
3756 	NXGE_DEBUG_MSG((nxgep, NXGE_CTL, "==> nxge_m_start"));
3757 
3758 	/*
3759 	 * Are we already started?
3760 	 */
3761 	if (nxgep->nxge_mac_state == NXGE_MAC_STARTED) {
3762 		return (0);
3763 	}
3764 
3765 	if (nxge_peu_reset_enable && !nxgep->nxge_link_poll_timerid) {
3766 		(void) nxge_link_monitor(nxgep, LINK_MONITOR_START);
3767 	}
3768 
3769 	/*
3770 	 * Make sure RX MAC is disabled while we initialize.
3771 	 */
3772 	if (!isLDOMguest(nxgep)) {
3773 		(void) nxge_rx_mac_disable(nxgep);
3774 	}
3775 
3776 	/*
3777 	 * Grab the global lock.
3778 	 */
3779 	MUTEX_ENTER(nxgep->genlock);
3780 
3781 	/*
3782 	 * Initialize the driver and hardware.
3783 	 */
3784 	if (nxge_init(nxgep) != NXGE_OK) {
3785 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3786 		    "<== nxge_m_start: initialization failed"));
3787 		MUTEX_EXIT(nxgep->genlock);
3788 		return (EIO);
3789 	}
3790 
3791 	/*
3792 	 * Start timer to check the system error and tx hangs
3793 	 */
3794 	if (!isLDOMguest(nxgep))
3795 		nxgep->nxge_timerid = nxge_start_timer(nxgep,
3796 		    nxge_check_hw_state, NXGE_CHECK_TIMER);
3797 #if defined(sun4v)
3798 	else
3799 		nxge_hio_start_timer(nxgep);
3800 #endif
3801 
3802 	nxgep->link_notify = B_TRUE;
3803 	nxgep->link_check_count = 0;
3804 	nxgep->nxge_mac_state = NXGE_MAC_STARTED;
3805 
3806 	/*
3807 	 * Let the global lock go, since we are intialized.
3808 	 */
3809 	MUTEX_EXIT(nxgep->genlock);
3810 
3811 	/*
3812 	 * Let the MAC start receiving packets, now that
3813 	 * we are initialized.
3814 	 */
3815 	if (!isLDOMguest(nxgep)) {
3816 		if (nxge_rx_mac_enable(nxgep) != NXGE_OK) {
3817 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3818 			    "<== nxge_m_start: enable of RX mac failed"));
3819 			return (EIO);
3820 		}
3821 
3822 		/*
3823 		 * Enable hardware interrupts.
3824 		 */
3825 		nxge_intr_hw_enable(nxgep);
3826 	}
3827 #if defined(sun4v)
3828 	else {
3829 		/*
3830 		 * In guest domain we enable RDCs and their interrupts as
3831 		 * the last step.
3832 		 */
3833 		if (nxge_hio_rdc_enable(nxgep) != NXGE_OK) {
3834 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3835 			    "<== nxge_m_start: enable of RDCs failed"));
3836 			return (EIO);
3837 		}
3838 
3839 		if (nxge_hio_rdc_intr_arm(nxgep, B_TRUE) != NXGE_OK) {
3840 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3841 			    "<== nxge_m_start: intrs enable for RDCs failed"));
3842 			return (EIO);
3843 		}
3844 	}
3845 #endif
3846 	NXGE_DEBUG_MSG((nxgep, NXGE_CTL, "<== nxge_m_start"));
3847 	return (0);
3848 }
3849 
3850 static boolean_t
3851 nxge_check_groups_stopped(p_nxge_t nxgep)
3852 {
3853 	int	i;
3854 
3855 	for (i = 0; i < NXGE_MAX_RDC_GROUPS; i++) {
3856 		if (nxgep->rx_hio_groups[i].started)
3857 			return (B_FALSE);
3858 	}
3859 
3860 	return (B_TRUE);
3861 }
3862 
3863 /*
3864  *	nxge_m_stop(): stop transmitting and receiving.
3865  */
3866 static void
3867 nxge_m_stop(void *arg)
3868 {
3869 	p_nxge_t 	nxgep = (p_nxge_t)arg;
3870 	boolean_t	groups_stopped;
3871 
3872 	NXGE_DEBUG_MSG((nxgep, NXGE_CTL, "==> nxge_m_stop"));
3873 
3874 	/*
3875 	 * Are the groups stopped?
3876 	 */
3877 	groups_stopped = nxge_check_groups_stopped(nxgep);
3878 	ASSERT(groups_stopped == B_TRUE);
3879 	if (!groups_stopped) {
3880 		cmn_err(CE_WARN, "nxge(%d): groups are not stopped!\n",
3881 		    nxgep->instance);
3882 		return;
3883 	}
3884 
3885 	if (!isLDOMguest(nxgep)) {
3886 		/*
3887 		 * Disable the RX mac.
3888 		 */
3889 		(void) nxge_rx_mac_disable(nxgep);
3890 
3891 		/*
3892 		 * Wait for the IPP to drain.
3893 		 */
3894 		(void) nxge_ipp_drain(nxgep);
3895 
3896 		/*
3897 		 * Disable hardware interrupts.
3898 		 */
3899 		nxge_intr_hw_disable(nxgep);
3900 	}
3901 #if defined(sun4v)
3902 	else {
3903 		(void) nxge_hio_rdc_intr_arm(nxgep, B_FALSE);
3904 	}
3905 #endif
3906 
3907 	/*
3908 	 * Grab the global lock.
3909 	 */
3910 	MUTEX_ENTER(nxgep->genlock);
3911 
3912 	nxgep->nxge_mac_state = NXGE_MAC_STOPPING;
3913 	if (nxgep->nxge_timerid) {
3914 		nxge_stop_timer(nxgep, nxgep->nxge_timerid);
3915 		nxgep->nxge_timerid = 0;
3916 	}
3917 
3918 	/*
3919 	 * Clean up.
3920 	 */
3921 	nxge_uninit(nxgep);
3922 
3923 	nxgep->nxge_mac_state = NXGE_MAC_STOPPED;
3924 
3925 	/*
3926 	 * Let go of the global lock.
3927 	 */
3928 	MUTEX_EXIT(nxgep->genlock);
3929 	NXGE_DEBUG_MSG((nxgep, NXGE_CTL, "<== nxge_m_stop"));
3930 }
3931 
3932 static int
3933 nxge_m_multicst(void *arg, boolean_t add, const uint8_t *mca)
3934 {
3935 	p_nxge_t 	nxgep = (p_nxge_t)arg;
3936 	struct 		ether_addr addrp;
3937 
3938 	NXGE_DEBUG_MSG((nxgep, MAC_CTL,
3939 	    "==> nxge_m_multicst: add %d", add));
3940 
3941 	bcopy(mca, (uint8_t *)&addrp, ETHERADDRL);
3942 	if (add) {
3943 		if (nxge_add_mcast_addr(nxgep, &addrp)) {
3944 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3945 			    "<== nxge_m_multicst: add multicast failed"));
3946 			return (EINVAL);
3947 		}
3948 	} else {
3949 		if (nxge_del_mcast_addr(nxgep, &addrp)) {
3950 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3951 			    "<== nxge_m_multicst: del multicast failed"));
3952 			return (EINVAL);
3953 		}
3954 	}
3955 
3956 	NXGE_DEBUG_MSG((nxgep, MAC_CTL, "<== nxge_m_multicst"));
3957 
3958 	return (0);
3959 }
3960 
3961 static int
3962 nxge_m_promisc(void *arg, boolean_t on)
3963 {
3964 	p_nxge_t 	nxgep = (p_nxge_t)arg;
3965 
3966 	NXGE_DEBUG_MSG((nxgep, MAC_CTL,
3967 	    "==> nxge_m_promisc: on %d", on));
3968 
3969 	if (nxge_set_promisc(nxgep, on)) {
3970 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
3971 		    "<== nxge_m_promisc: set promisc failed"));
3972 		return (EINVAL);
3973 	}
3974 
3975 	NXGE_DEBUG_MSG((nxgep, MAC_CTL,
3976 	    "<== nxge_m_promisc: on %d", on));
3977 
3978 	return (0);
3979 }
3980 
3981 static void
3982 nxge_m_ioctl(void *arg,  queue_t *wq, mblk_t *mp)
3983 {
3984 	p_nxge_t 	nxgep = (p_nxge_t)arg;
3985 	struct 		iocblk *iocp;
3986 	boolean_t 	need_privilege;
3987 	int 		err;
3988 	int 		cmd;
3989 
3990 	NXGE_DEBUG_MSG((nxgep, NXGE_CTL, "==> nxge_m_ioctl"));
3991 
3992 	iocp = (struct iocblk *)mp->b_rptr;
3993 	iocp->ioc_error = 0;
3994 	need_privilege = B_TRUE;
3995 	cmd = iocp->ioc_cmd;
3996 	NXGE_DEBUG_MSG((nxgep, NXGE_CTL, "==> nxge_m_ioctl: cmd 0x%08x", cmd));
3997 	switch (cmd) {
3998 	default:
3999 		miocnak(wq, mp, 0, EINVAL);
4000 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL, "<== nxge_m_ioctl: invalid"));
4001 		return;
4002 
4003 	case LB_GET_INFO_SIZE:
4004 	case LB_GET_INFO:
4005 	case LB_GET_MODE:
4006 		need_privilege = B_FALSE;
4007 		break;
4008 	case LB_SET_MODE:
4009 		break;
4010 
4011 
4012 	case NXGE_GET_MII:
4013 	case NXGE_PUT_MII:
4014 	case NXGE_GET64:
4015 	case NXGE_PUT64:
4016 	case NXGE_GET_TX_RING_SZ:
4017 	case NXGE_GET_TX_DESC:
4018 	case NXGE_TX_SIDE_RESET:
4019 	case NXGE_RX_SIDE_RESET:
4020 	case NXGE_GLOBAL_RESET:
4021 	case NXGE_RESET_MAC:
4022 	case NXGE_TX_REGS_DUMP:
4023 	case NXGE_RX_REGS_DUMP:
4024 	case NXGE_INT_REGS_DUMP:
4025 	case NXGE_VIR_INT_REGS_DUMP:
4026 	case NXGE_PUT_TCAM:
4027 	case NXGE_GET_TCAM:
4028 	case NXGE_RTRACE:
4029 	case NXGE_RDUMP:
4030 	case NXGE_RX_CLASS:
4031 	case NXGE_RX_HASH:
4032 
4033 		need_privilege = B_FALSE;
4034 		break;
4035 	case NXGE_INJECT_ERR:
4036 		cmn_err(CE_NOTE, "!nxge_m_ioctl: Inject error\n");
4037 		nxge_err_inject(nxgep, wq, mp);
4038 		break;
4039 	}
4040 
4041 	if (need_privilege) {
4042 		err = secpolicy_net_config(iocp->ioc_cr, B_FALSE);
4043 		if (err != 0) {
4044 			miocnak(wq, mp, 0, err);
4045 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
4046 			    "<== nxge_m_ioctl: no priv"));
4047 			return;
4048 		}
4049 	}
4050 
4051 	switch (cmd) {
4052 
4053 	case LB_GET_MODE:
4054 	case LB_SET_MODE:
4055 	case LB_GET_INFO_SIZE:
4056 	case LB_GET_INFO:
4057 		nxge_loopback_ioctl(nxgep, wq, mp, iocp);
4058 		break;
4059 
4060 	case NXGE_GET_MII:
4061 	case NXGE_PUT_MII:
4062 	case NXGE_PUT_TCAM:
4063 	case NXGE_GET_TCAM:
4064 	case NXGE_GET64:
4065 	case NXGE_PUT64:
4066 	case NXGE_GET_TX_RING_SZ:
4067 	case NXGE_GET_TX_DESC:
4068 	case NXGE_TX_SIDE_RESET:
4069 	case NXGE_RX_SIDE_RESET:
4070 	case NXGE_GLOBAL_RESET:
4071 	case NXGE_RESET_MAC:
4072 	case NXGE_TX_REGS_DUMP:
4073 	case NXGE_RX_REGS_DUMP:
4074 	case NXGE_INT_REGS_DUMP:
4075 	case NXGE_VIR_INT_REGS_DUMP:
4076 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
4077 		    "==> nxge_m_ioctl: cmd 0x%x", cmd));
4078 		nxge_hw_ioctl(nxgep, wq, mp, iocp);
4079 		break;
4080 	case NXGE_RX_CLASS:
4081 		if (nxge_rxclass_ioctl(nxgep, wq, mp->b_cont) < 0)
4082 			miocnak(wq, mp, 0, EINVAL);
4083 		else
4084 			miocack(wq, mp, sizeof (rx_class_cfg_t), 0);
4085 		break;
4086 	case NXGE_RX_HASH:
4087 
4088 		if (nxge_rxhash_ioctl(nxgep, wq, mp->b_cont) < 0)
4089 			miocnak(wq, mp, 0, EINVAL);
4090 		else
4091 			miocack(wq, mp, sizeof (cfg_cmd_t), 0);
4092 		break;
4093 	}
4094 
4095 	NXGE_DEBUG_MSG((nxgep, NXGE_CTL, "<== nxge_m_ioctl"));
4096 }
4097 
4098 extern void nxge_rx_hw_blank(void *arg, time_t ticks, uint_t count);
4099 
4100 void
4101 nxge_mmac_kstat_update(p_nxge_t nxgep, int slot, boolean_t factory)
4102 {
4103 	p_nxge_mmac_stats_t mmac_stats;
4104 	int i;
4105 	nxge_mmac_t *mmac_info;
4106 
4107 	mmac_info = &nxgep->nxge_mmac_info;
4108 
4109 	mmac_stats = &nxgep->statsp->mmac_stats;
4110 	mmac_stats->mmac_max_cnt = mmac_info->num_mmac;
4111 	mmac_stats->mmac_avail_cnt = mmac_info->naddrfree;
4112 
4113 	for (i = 0; i < ETHERADDRL; i++) {
4114 		if (factory) {
4115 			mmac_stats->mmac_avail_pool[slot-1].ether_addr_octet[i]
4116 			    = mmac_info->factory_mac_pool[slot][
4117 			    (ETHERADDRL-1) - i];
4118 		} else {
4119 			mmac_stats->mmac_avail_pool[slot-1].ether_addr_octet[i]
4120 			    = mmac_info->mac_pool[slot].addr[
4121 			    (ETHERADDRL - 1) - i];
4122 		}
4123 	}
4124 }
4125 
4126 /*
4127  * nxge_altmac_set() -- Set an alternate MAC address
4128  */
4129 static int
4130 nxge_altmac_set(p_nxge_t nxgep, uint8_t *maddr, int slot,
4131 	int rdctbl, boolean_t usetbl)
4132 {
4133 	uint8_t addrn;
4134 	uint8_t portn;
4135 	npi_mac_addr_t altmac;
4136 	hostinfo_t mac_rdc;
4137 	p_nxge_class_pt_cfg_t clscfgp;
4138 
4139 
4140 	altmac.w2 = ((uint16_t)maddr[0] << 8) | ((uint16_t)maddr[1] & 0x0ff);
4141 	altmac.w1 = ((uint16_t)maddr[2] << 8) | ((uint16_t)maddr[3] & 0x0ff);
4142 	altmac.w0 = ((uint16_t)maddr[4] << 8) | ((uint16_t)maddr[5] & 0x0ff);
4143 
4144 	portn = nxgep->mac.portnum;
4145 	addrn = (uint8_t)slot - 1;
4146 
4147 	if (npi_mac_altaddr_entry(nxgep->npi_handle, OP_SET,
4148 	    nxgep->function_num, addrn, &altmac) != NPI_SUCCESS)
4149 		return (EIO);
4150 
4151 	/*
4152 	 * Set the rdc table number for the host info entry
4153 	 * for this mac address slot.
4154 	 */
4155 	clscfgp = (p_nxge_class_pt_cfg_t)&nxgep->class_config;
4156 	mac_rdc.value = 0;
4157 	if (usetbl)
4158 		mac_rdc.bits.w0.rdc_tbl_num = rdctbl;
4159 	else
4160 		mac_rdc.bits.w0.rdc_tbl_num =
4161 		    clscfgp->mac_host_info[addrn].rdctbl;
4162 	mac_rdc.bits.w0.mac_pref = clscfgp->mac_host_info[addrn].mpr_npr;
4163 
4164 	if (npi_mac_hostinfo_entry(nxgep->npi_handle, OP_SET,
4165 	    nxgep->function_num, addrn, &mac_rdc) != NPI_SUCCESS) {
4166 		return (EIO);
4167 	}
4168 
4169 	/*
4170 	 * Enable comparison with the alternate MAC address.
4171 	 * While the first alternate addr is enabled by bit 1 of register
4172 	 * BMAC_ALTAD_CMPEN, it is enabled by bit 0 of register
4173 	 * XMAC_ADDR_CMPEN, so slot needs to be converted to addrn
4174 	 * accordingly before calling npi_mac_altaddr_entry.
4175 	 */
4176 	if (portn == XMAC_PORT_0 || portn == XMAC_PORT_1)
4177 		addrn = (uint8_t)slot - 1;
4178 	else
4179 		addrn = (uint8_t)slot;
4180 
4181 	if (npi_mac_altaddr_enable(nxgep->npi_handle,
4182 	    nxgep->function_num, addrn) != NPI_SUCCESS) {
4183 		return (EIO);
4184 	}
4185 
4186 	return (0);
4187 }
4188 
4189 /*
4190  * nxeg_m_mmac_add_g() - find an unused address slot, set the address
4191  * value to the one specified, enable the port to start filtering on
4192  * the new MAC address.  Returns 0 on success.
4193  */
4194 int
4195 nxge_m_mmac_add_g(void *arg, const uint8_t *maddr, int rdctbl,
4196 	boolean_t usetbl)
4197 {
4198 	p_nxge_t nxgep = arg;
4199 	int slot;
4200 	nxge_mmac_t *mmac_info;
4201 	int err;
4202 	nxge_status_t status;
4203 
4204 	mutex_enter(nxgep->genlock);
4205 
4206 	/*
4207 	 * Make sure that nxge is initialized, if _start() has
4208 	 * not been called.
4209 	 */
4210 	if (!(nxgep->drv_state & STATE_HW_INITIALIZED)) {
4211 		status = nxge_init(nxgep);
4212 		if (status != NXGE_OK) {
4213 			mutex_exit(nxgep->genlock);
4214 			return (ENXIO);
4215 		}
4216 	}
4217 
4218 	mmac_info = &nxgep->nxge_mmac_info;
4219 	if (mmac_info->naddrfree == 0) {
4220 		mutex_exit(nxgep->genlock);
4221 		return (ENOSPC);
4222 	}
4223 
4224 	/*
4225 	 * 	Search for the first available slot. Because naddrfree
4226 	 * is not zero, we are guaranteed to find one.
4227 	 *	Each of the first two ports of Neptune has 16 alternate
4228 	 * MAC slots but only the first 7 (of 15) slots have assigned factory
4229 	 * MAC addresses. We first search among the slots without bundled
4230 	 * factory MACs. If we fail to find one in that range, then we
4231 	 * search the slots with bundled factory MACs.  A factory MAC
4232 	 * will be wasted while the slot is used with a user MAC address.
4233 	 * But the slot could be used by factory MAC again after calling
4234 	 * nxge_m_mmac_remove and nxge_m_mmac_reserve.
4235 	 */
4236 	for (slot = 0; slot <= mmac_info->num_mmac; slot++) {
4237 		if (!(mmac_info->mac_pool[slot].flags & MMAC_SLOT_USED))
4238 			break;
4239 	}
4240 
4241 	ASSERT(slot <= mmac_info->num_mmac);
4242 
4243 	if ((err = nxge_altmac_set(nxgep, (uint8_t *)maddr, slot, rdctbl,
4244 	    usetbl)) != 0) {
4245 		mutex_exit(nxgep->genlock);
4246 		return (err);
4247 	}
4248 
4249 	bcopy(maddr, mmac_info->mac_pool[slot].addr, ETHERADDRL);
4250 	mmac_info->mac_pool[slot].flags |= MMAC_SLOT_USED;
4251 	mmac_info->mac_pool[slot].flags &= ~MMAC_VENDOR_ADDR;
4252 	mmac_info->naddrfree--;
4253 	nxge_mmac_kstat_update(nxgep, slot, B_FALSE);
4254 
4255 	mutex_exit(nxgep->genlock);
4256 	return (0);
4257 }
4258 
4259 /*
4260  * Remove the specified mac address and update the HW not to filter
4261  * the mac address anymore.
4262  */
4263 int
4264 nxge_m_mmac_remove(void *arg, int slot)
4265 {
4266 	p_nxge_t nxgep = arg;
4267 	nxge_mmac_t *mmac_info;
4268 	uint8_t addrn;
4269 	uint8_t portn;
4270 	int err = 0;
4271 	nxge_status_t status;
4272 
4273 	mutex_enter(nxgep->genlock);
4274 
4275 	/*
4276 	 * Make sure that nxge is initialized, if _start() has
4277 	 * not been called.
4278 	 */
4279 	if (!(nxgep->drv_state & STATE_HW_INITIALIZED)) {
4280 		status = nxge_init(nxgep);
4281 		if (status != NXGE_OK) {
4282 			mutex_exit(nxgep->genlock);
4283 			return (ENXIO);
4284 		}
4285 	}
4286 
4287 	mmac_info = &nxgep->nxge_mmac_info;
4288 	if (slot < 1 || slot > mmac_info->num_mmac) {
4289 		mutex_exit(nxgep->genlock);
4290 		return (EINVAL);
4291 	}
4292 
4293 	portn = nxgep->mac.portnum;
4294 	if (portn == XMAC_PORT_0 || portn == XMAC_PORT_1)
4295 		addrn = (uint8_t)slot - 1;
4296 	else
4297 		addrn = (uint8_t)slot;
4298 
4299 	if (mmac_info->mac_pool[slot].flags & MMAC_SLOT_USED) {
4300 		if (npi_mac_altaddr_disable(nxgep->npi_handle, portn, addrn)
4301 		    == NPI_SUCCESS) {
4302 			mmac_info->naddrfree++;
4303 			mmac_info->mac_pool[slot].flags &= ~MMAC_SLOT_USED;
4304 			/*
4305 			 * Regardless if the MAC we just stopped filtering
4306 			 * is a user addr or a facory addr, we must set
4307 			 * the MMAC_VENDOR_ADDR flag if this slot has an
4308 			 * associated factory MAC to indicate that a factory
4309 			 * MAC is available.
4310 			 */
4311 			if (slot <= mmac_info->num_factory_mmac) {
4312 				mmac_info->mac_pool[slot].flags
4313 				    |= MMAC_VENDOR_ADDR;
4314 			}
4315 			/*
4316 			 * Clear mac_pool[slot].addr so that kstat shows 0
4317 			 * alternate MAC address if the slot is not used.
4318 			 * (But nxge_m_mmac_get returns the factory MAC even
4319 			 * when the slot is not used!)
4320 			 */
4321 			bzero(mmac_info->mac_pool[slot].addr, ETHERADDRL);
4322 			nxge_mmac_kstat_update(nxgep, slot, B_FALSE);
4323 		} else {
4324 			err = EIO;
4325 		}
4326 	} else {
4327 		err = EINVAL;
4328 	}
4329 
4330 	mutex_exit(nxgep->genlock);
4331 	return (err);
4332 }
4333 
4334 /*
4335  * The callback to query all the factory addresses. naddr must be the same as
4336  * the number of factory addresses (returned by MAC_CAPAB_MULTIFACTADDR), and
4337  * mcm_addr is the space allocated for keep all the addresses, whose size is
4338  * naddr * MAXMACADDRLEN.
4339  */
4340 static void
4341 nxge_m_getfactaddr(void *arg, uint_t naddr, uint8_t *addr)
4342 {
4343 	nxge_t		*nxgep = arg;
4344 	nxge_mmac_t	*mmac_info;
4345 	int		i;
4346 
4347 	mutex_enter(nxgep->genlock);
4348 
4349 	mmac_info = &nxgep->nxge_mmac_info;
4350 	ASSERT(naddr == mmac_info->num_factory_mmac);
4351 
4352 	for (i = 0; i < naddr; i++) {
4353 		bcopy(mmac_info->factory_mac_pool[i + 1],
4354 		    addr + i * MAXMACADDRLEN, ETHERADDRL);
4355 	}
4356 
4357 	mutex_exit(nxgep->genlock);
4358 }
4359 
4360 
4361 static boolean_t
4362 nxge_m_getcapab(void *arg, mac_capab_t cap, void *cap_data)
4363 {
4364 	nxge_t *nxgep = arg;
4365 	uint32_t *txflags = cap_data;
4366 
4367 	switch (cap) {
4368 	case MAC_CAPAB_HCKSUM:
4369 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
4370 		    "==> nxge_m_getcapab: checksum %d", nxge_cksum_offload));
4371 		if (nxge_cksum_offload <= 1) {
4372 			*txflags = HCKSUM_INET_PARTIAL;
4373 		}
4374 		break;
4375 
4376 	case MAC_CAPAB_MULTIFACTADDR: {
4377 		mac_capab_multifactaddr_t	*mfacp = cap_data;
4378 
4379 		if (!isLDOMguest(nxgep)) {
4380 			mutex_enter(nxgep->genlock);
4381 			mfacp->mcm_naddr =
4382 			    nxgep->nxge_mmac_info.num_factory_mmac;
4383 			mfacp->mcm_getaddr = nxge_m_getfactaddr;
4384 			mutex_exit(nxgep->genlock);
4385 		}
4386 		break;
4387 	}
4388 
4389 	case MAC_CAPAB_LSO: {
4390 		mac_capab_lso_t *cap_lso = cap_data;
4391 
4392 		if (nxgep->soft_lso_enable) {
4393 			if (nxge_cksum_offload <= 1) {
4394 				cap_lso->lso_flags = LSO_TX_BASIC_TCP_IPV4;
4395 				if (nxge_lso_max > NXGE_LSO_MAXLEN) {
4396 					nxge_lso_max = NXGE_LSO_MAXLEN;
4397 				}
4398 				cap_lso->lso_basic_tcp_ipv4.lso_max =
4399 				    nxge_lso_max;
4400 			}
4401 			break;
4402 		} else {
4403 			return (B_FALSE);
4404 		}
4405 	}
4406 
4407 	case MAC_CAPAB_RINGS: {
4408 		mac_capab_rings_t	*cap_rings = cap_data;
4409 		p_nxge_hw_pt_cfg_t	p_cfgp = &nxgep->pt_config.hw_config;
4410 
4411 		mutex_enter(nxgep->genlock);
4412 		if (cap_rings->mr_type == MAC_RING_TYPE_RX) {
4413 			if (isLDOMguest(nxgep))  {
4414 				cap_rings->mr_group_type =
4415 				    MAC_GROUP_TYPE_STATIC;
4416 				cap_rings->mr_rnum =
4417 				    NXGE_HIO_SHARE_MAX_CHANNELS;
4418 				cap_rings->mr_rget = nxge_fill_ring;
4419 				cap_rings->mr_gnum = 1;
4420 				cap_rings->mr_gget = nxge_hio_group_get;
4421 				cap_rings->mr_gaddring = NULL;
4422 				cap_rings->mr_gremring = NULL;
4423 			} else {
4424 				/*
4425 				 * Service Domain.
4426 				 */
4427 				cap_rings->mr_group_type =
4428 				    MAC_GROUP_TYPE_DYNAMIC;
4429 				cap_rings->mr_rnum = p_cfgp->max_rdcs;
4430 				cap_rings->mr_rget = nxge_fill_ring;
4431 				cap_rings->mr_gnum = p_cfgp->max_rdc_grpids;
4432 				cap_rings->mr_gget = nxge_hio_group_get;
4433 				cap_rings->mr_gaddring = nxge_group_add_ring;
4434 				cap_rings->mr_gremring = nxge_group_rem_ring;
4435 			}
4436 
4437 			NXGE_DEBUG_MSG((nxgep, RX_CTL,
4438 			    "==> nxge_m_getcapab: rx nrings[%d] ngroups[%d]",
4439 			    p_cfgp->max_rdcs, p_cfgp->max_rdc_grpids));
4440 		} else {
4441 			/*
4442 			 * TX Rings.
4443 			 */
4444 			if (isLDOMguest(nxgep)) {
4445 				cap_rings->mr_group_type =
4446 				    MAC_GROUP_TYPE_STATIC;
4447 				cap_rings->mr_rnum =
4448 				    NXGE_HIO_SHARE_MAX_CHANNELS;
4449 				cap_rings->mr_rget = nxge_fill_ring;
4450 				cap_rings->mr_gnum = 0;
4451 				cap_rings->mr_gget = NULL;
4452 				cap_rings->mr_gaddring = NULL;
4453 				cap_rings->mr_gremring = NULL;
4454 			} else {
4455 				/*
4456 				 * Service Domain.
4457 				 */
4458 				cap_rings->mr_group_type =
4459 				    MAC_GROUP_TYPE_DYNAMIC;
4460 				cap_rings->mr_rnum = p_cfgp->tdc.count;
4461 				cap_rings->mr_rget = nxge_fill_ring;
4462 
4463 				/*
4464 				 * Share capable.
4465 				 *
4466 				 * Do not report the default group: hence -1
4467 				 */
4468 				cap_rings->mr_gnum =
4469 				    NXGE_MAX_TDC_GROUPS / nxgep->nports - 1;
4470 				cap_rings->mr_gget = nxge_hio_group_get;
4471 				cap_rings->mr_gaddring = nxge_group_add_ring;
4472 				cap_rings->mr_gremring = nxge_group_rem_ring;
4473 			}
4474 
4475 			NXGE_DEBUG_MSG((nxgep, TX_CTL,
4476 			    "==> nxge_m_getcapab: tx rings # of rings %d",
4477 			    p_cfgp->tdc.count));
4478 		}
4479 		mutex_exit(nxgep->genlock);
4480 		break;
4481 	}
4482 
4483 #if defined(sun4v)
4484 	case MAC_CAPAB_SHARES: {
4485 		mac_capab_share_t *mshares = (mac_capab_share_t *)cap_data;
4486 
4487 		/*
4488 		 * Only the service domain driver responds to
4489 		 * this capability request.
4490 		 */
4491 		mutex_enter(nxgep->genlock);
4492 		if (isLDOMservice(nxgep)) {
4493 			mshares->ms_snum = 3;
4494 			mshares->ms_handle = (void *)nxgep;
4495 			mshares->ms_salloc = nxge_hio_share_alloc;
4496 			mshares->ms_sfree = nxge_hio_share_free;
4497 			mshares->ms_sadd = nxge_hio_share_add_group;
4498 			mshares->ms_sremove = nxge_hio_share_rem_group;
4499 			mshares->ms_squery = nxge_hio_share_query;
4500 			mshares->ms_sbind = nxge_hio_share_bind;
4501 			mshares->ms_sunbind = nxge_hio_share_unbind;
4502 			mutex_exit(nxgep->genlock);
4503 		} else {
4504 			mutex_exit(nxgep->genlock);
4505 			return (B_FALSE);
4506 		}
4507 		break;
4508 	}
4509 #endif
4510 	default:
4511 		return (B_FALSE);
4512 	}
4513 	return (B_TRUE);
4514 }
4515 
4516 static boolean_t
4517 nxge_param_locked(mac_prop_id_t pr_num)
4518 {
4519 	/*
4520 	 * All adv_* parameters are locked (read-only) while
4521 	 * the device is in any sort of loopback mode ...
4522 	 */
4523 	switch (pr_num) {
4524 		case MAC_PROP_ADV_1000FDX_CAP:
4525 		case MAC_PROP_EN_1000FDX_CAP:
4526 		case MAC_PROP_ADV_1000HDX_CAP:
4527 		case MAC_PROP_EN_1000HDX_CAP:
4528 		case MAC_PROP_ADV_100FDX_CAP:
4529 		case MAC_PROP_EN_100FDX_CAP:
4530 		case MAC_PROP_ADV_100HDX_CAP:
4531 		case MAC_PROP_EN_100HDX_CAP:
4532 		case MAC_PROP_ADV_10FDX_CAP:
4533 		case MAC_PROP_EN_10FDX_CAP:
4534 		case MAC_PROP_ADV_10HDX_CAP:
4535 		case MAC_PROP_EN_10HDX_CAP:
4536 		case MAC_PROP_AUTONEG:
4537 		case MAC_PROP_FLOWCTRL:
4538 			return (B_TRUE);
4539 	}
4540 	return (B_FALSE);
4541 }
4542 
4543 /*
4544  * callback functions for set/get of properties
4545  */
4546 static int
4547 nxge_m_setprop(void *barg, const char *pr_name, mac_prop_id_t pr_num,
4548     uint_t pr_valsize, const void *pr_val)
4549 {
4550 	nxge_t		*nxgep = barg;
4551 	p_nxge_param_t	param_arr = nxgep->param_arr;
4552 	p_nxge_stats_t	statsp = nxgep->statsp;
4553 	int		err = 0;
4554 
4555 	NXGE_DEBUG_MSG((nxgep, NXGE_CTL, "==> nxge_m_setprop"));
4556 
4557 	mutex_enter(nxgep->genlock);
4558 	if (statsp->port_stats.lb_mode != nxge_lb_normal &&
4559 	    nxge_param_locked(pr_num)) {
4560 		/*
4561 		 * All adv_* parameters are locked (read-only)
4562 		 * while the device is in any sort of loopback mode.
4563 		 */
4564 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
4565 		    "==> nxge_m_setprop: loopback mode: read only"));
4566 		mutex_exit(nxgep->genlock);
4567 		return (EBUSY);
4568 	}
4569 
4570 	switch (pr_num) {
4571 	case MAC_PROP_EN_1000FDX_CAP:
4572 		nxgep->param_en_1000fdx =
4573 		    param_arr[param_anar_1000fdx].value = *(uint8_t *)pr_val;
4574 		goto reprogram;
4575 
4576 	case MAC_PROP_EN_100FDX_CAP:
4577 		nxgep->param_en_100fdx =
4578 		    param_arr[param_anar_100fdx].value = *(uint8_t *)pr_val;
4579 		goto reprogram;
4580 
4581 	case MAC_PROP_EN_10FDX_CAP:
4582 		nxgep->param_en_10fdx =
4583 		    param_arr[param_anar_10fdx].value = *(uint8_t *)pr_val;
4584 		goto reprogram;
4585 
4586 	case MAC_PROP_AUTONEG:
4587 		param_arr[param_autoneg].value = *(uint8_t *)pr_val;
4588 		goto reprogram;
4589 
4590 	case MAC_PROP_MTU: {
4591 		uint32_t cur_mtu, new_mtu, old_framesize;
4592 
4593 		cur_mtu = nxgep->mac.default_mtu;
4594 		ASSERT(pr_valsize >= sizeof (new_mtu));
4595 		bcopy(pr_val, &new_mtu, sizeof (new_mtu));
4596 
4597 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
4598 		    "==> nxge_m_setprop: set MTU: %d is_jumbo %d",
4599 		    new_mtu, nxgep->mac.is_jumbo));
4600 
4601 		if (new_mtu == cur_mtu) {
4602 			err = 0;
4603 			break;
4604 		}
4605 
4606 		if (nxgep->nxge_mac_state == NXGE_MAC_STARTED) {
4607 			err = EBUSY;
4608 			break;
4609 		}
4610 
4611 		if ((new_mtu < NXGE_DEFAULT_MTU) ||
4612 		    (new_mtu > NXGE_MAXIMUM_MTU)) {
4613 			err = EINVAL;
4614 			break;
4615 		}
4616 
4617 		old_framesize = (uint32_t)nxgep->mac.maxframesize;
4618 		nxgep->mac.maxframesize = (uint16_t)
4619 		    (new_mtu + NXGE_EHEADER_VLAN_CRC);
4620 		if (nxge_mac_set_framesize(nxgep)) {
4621 			nxgep->mac.maxframesize =
4622 			    (uint16_t)old_framesize;
4623 			err = EINVAL;
4624 			break;
4625 		}
4626 
4627 		nxgep->mac.default_mtu = new_mtu;
4628 		nxgep->mac.is_jumbo = (new_mtu > NXGE_DEFAULT_MTU);
4629 
4630 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
4631 		    "==> nxge_m_setprop: set MTU: %d maxframe %d",
4632 		    new_mtu, nxgep->mac.maxframesize));
4633 		break;
4634 	}
4635 
4636 	case MAC_PROP_FLOWCTRL: {
4637 		link_flowctrl_t	fl;
4638 
4639 		ASSERT(pr_valsize >= sizeof (fl));
4640 		bcopy(pr_val, &fl, sizeof (fl));
4641 
4642 		switch (fl) {
4643 		case LINK_FLOWCTRL_NONE:
4644 			param_arr[param_anar_pause].value = 0;
4645 			break;
4646 
4647 		case LINK_FLOWCTRL_RX:
4648 			param_arr[param_anar_pause].value = 1;
4649 			break;
4650 
4651 		case LINK_FLOWCTRL_TX:
4652 		case LINK_FLOWCTRL_BI:
4653 			err = EINVAL;
4654 			break;
4655 		default:
4656 			err = EINVAL;
4657 			break;
4658 		}
4659 reprogram:
4660 		if ((err == 0) && !isLDOMguest(nxgep)) {
4661 			if (!nxge_param_link_update(nxgep)) {
4662 				err = EINVAL;
4663 			}
4664 		} else {
4665 			err = EINVAL;
4666 		}
4667 		break;
4668 	}
4669 
4670 	case MAC_PROP_PRIVATE:
4671 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
4672 		    "==> nxge_m_setprop: private property"));
4673 		err = nxge_set_priv_prop(nxgep, pr_name, pr_valsize, pr_val);
4674 		break;
4675 
4676 	default:
4677 		err = ENOTSUP;
4678 		break;
4679 	}
4680 
4681 	mutex_exit(nxgep->genlock);
4682 
4683 	NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
4684 	    "<== nxge_m_setprop (return %d)", err));
4685 	return (err);
4686 }
4687 
4688 static int
4689 nxge_m_getprop(void *barg, const char *pr_name, mac_prop_id_t pr_num,
4690     uint_t pr_valsize, void *pr_val)
4691 {
4692 	nxge_t 		*nxgep = barg;
4693 	p_nxge_param_t	param_arr = nxgep->param_arr;
4694 	p_nxge_stats_t	statsp = nxgep->statsp;
4695 
4696 	NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
4697 	    "==> nxge_m_getprop: pr_num %d", pr_num));
4698 
4699 	switch (pr_num) {
4700 	case MAC_PROP_DUPLEX:
4701 		*(uint8_t *)pr_val = statsp->mac_stats.link_duplex;
4702 		break;
4703 
4704 	case MAC_PROP_SPEED: {
4705 		uint64_t val = statsp->mac_stats.link_speed * 1000000ull;
4706 
4707 		ASSERT(pr_valsize >= sizeof (val));
4708 		bcopy(&val, pr_val, sizeof (val));
4709 		break;
4710 	}
4711 
4712 	case MAC_PROP_STATUS: {
4713 		link_state_t state = statsp->mac_stats.link_up ?
4714 		    LINK_STATE_UP : LINK_STATE_DOWN;
4715 
4716 		ASSERT(pr_valsize >= sizeof (state));
4717 		bcopy(&state, pr_val, sizeof (state));
4718 		break;
4719 	}
4720 
4721 	case MAC_PROP_AUTONEG:
4722 		*(uint8_t *)pr_val = param_arr[param_autoneg].value;
4723 		break;
4724 
4725 	case MAC_PROP_FLOWCTRL: {
4726 		link_flowctrl_t fl = param_arr[param_anar_pause].value != 0 ?
4727 		    LINK_FLOWCTRL_RX : LINK_FLOWCTRL_NONE;
4728 
4729 		ASSERT(pr_valsize >= sizeof (fl));
4730 		bcopy(&fl, pr_val, sizeof (fl));
4731 		break;
4732 	}
4733 
4734 	case MAC_PROP_ADV_1000FDX_CAP:
4735 		*(uint8_t *)pr_val = param_arr[param_anar_1000fdx].value;
4736 		break;
4737 
4738 	case MAC_PROP_EN_1000FDX_CAP:
4739 		*(uint8_t *)pr_val = nxgep->param_en_1000fdx;
4740 		break;
4741 
4742 	case MAC_PROP_ADV_100FDX_CAP:
4743 		*(uint8_t *)pr_val = param_arr[param_anar_100fdx].value;
4744 		break;
4745 
4746 	case MAC_PROP_EN_100FDX_CAP:
4747 		*(uint8_t *)pr_val = nxgep->param_en_100fdx;
4748 		break;
4749 
4750 	case MAC_PROP_ADV_10FDX_CAP:
4751 		*(uint8_t *)pr_val = param_arr[param_anar_10fdx].value;
4752 		break;
4753 
4754 	case MAC_PROP_EN_10FDX_CAP:
4755 		*(uint8_t *)pr_val = nxgep->param_en_10fdx;
4756 		break;
4757 
4758 	case MAC_PROP_PRIVATE:
4759 		return (nxge_get_priv_prop(nxgep, pr_name, pr_valsize,
4760 		    pr_val));
4761 
4762 	default:
4763 		return (ENOTSUP);
4764 	}
4765 
4766 	return (0);
4767 }
4768 
4769 static void
4770 nxge_m_propinfo(void *barg, const char *pr_name, mac_prop_id_t pr_num,
4771     mac_prop_info_handle_t prh)
4772 {
4773 	nxge_t		*nxgep = barg;
4774 	p_nxge_stats_t	statsp = nxgep->statsp;
4775 
4776 	/*
4777 	 * By default permissions are read/write unless specified
4778 	 * otherwise by the driver.
4779 	 */
4780 
4781 	switch (pr_num) {
4782 	case MAC_PROP_DUPLEX:
4783 	case MAC_PROP_SPEED:
4784 	case MAC_PROP_STATUS:
4785 	case MAC_PROP_EN_1000HDX_CAP:
4786 	case MAC_PROP_EN_100HDX_CAP:
4787 	case MAC_PROP_EN_10HDX_CAP:
4788 	case MAC_PROP_ADV_1000FDX_CAP:
4789 	case MAC_PROP_ADV_1000HDX_CAP:
4790 	case MAC_PROP_ADV_100FDX_CAP:
4791 	case MAC_PROP_ADV_100HDX_CAP:
4792 	case MAC_PROP_ADV_10FDX_CAP:
4793 	case MAC_PROP_ADV_10HDX_CAP:
4794 		/*
4795 		 * Note that read-only properties don't need to
4796 		 * provide default values since they cannot be
4797 		 * changed by the administrator.
4798 		 */
4799 		mac_prop_info_set_perm(prh, MAC_PROP_PERM_READ);
4800 		break;
4801 
4802 	case MAC_PROP_EN_1000FDX_CAP:
4803 	case MAC_PROP_EN_100FDX_CAP:
4804 	case MAC_PROP_EN_10FDX_CAP:
4805 		mac_prop_info_set_default_uint8(prh, 1);
4806 		break;
4807 
4808 	case MAC_PROP_AUTONEG:
4809 		mac_prop_info_set_default_uint8(prh, 1);
4810 		break;
4811 
4812 	case MAC_PROP_FLOWCTRL:
4813 		mac_prop_info_set_default_link_flowctrl(prh, LINK_FLOWCTRL_RX);
4814 		break;
4815 
4816 	case MAC_PROP_MTU:
4817 		mac_prop_info_set_range_uint32(prh,
4818 		    NXGE_DEFAULT_MTU, NXGE_MAXIMUM_MTU);
4819 		break;
4820 
4821 	case MAC_PROP_PRIVATE:
4822 		nxge_priv_propinfo(pr_name, prh);
4823 		break;
4824 	}
4825 
4826 	mutex_enter(nxgep->genlock);
4827 	if (statsp->port_stats.lb_mode != nxge_lb_normal &&
4828 	    nxge_param_locked(pr_num)) {
4829 		/*
4830 		 * Some properties are locked (read-only) while the
4831 		 * device is in any sort of loopback mode.
4832 		 */
4833 		mac_prop_info_set_perm(prh, MAC_PROP_PERM_READ);
4834 	}
4835 	mutex_exit(nxgep->genlock);
4836 }
4837 
4838 static void
4839 nxge_priv_propinfo(const char *pr_name, mac_prop_info_handle_t prh)
4840 {
4841 	char valstr[64];
4842 
4843 	bzero(valstr, sizeof (valstr));
4844 
4845 	if (strcmp(pr_name, "_function_number") == 0 ||
4846 	    strcmp(pr_name, "_fw_version") == 0 ||
4847 	    strcmp(pr_name, "_port_mode") == 0 ||
4848 	    strcmp(pr_name, "_hot_swap_phy") == 0) {
4849 		mac_prop_info_set_perm(prh, MAC_PROP_PERM_READ);
4850 
4851 	} else if (strcmp(pr_name, "_rxdma_intr_time") == 0) {
4852 		(void) snprintf(valstr, sizeof (valstr),
4853 		    "%d", RXDMA_RCR_TO_DEFAULT);
4854 
4855 	} else if (strcmp(pr_name, "_rxdma_intr_pkts") == 0) {
4856 		(void) snprintf(valstr, sizeof (valstr),
4857 		    "%d", RXDMA_RCR_PTHRES_DEFAULT);
4858 
4859 	} else 	if (strcmp(pr_name, "_class_opt_ipv4_tcp") == 0 ||
4860 	    strcmp(pr_name, "_class_opt_ipv4_udp") == 0 ||
4861 	    strcmp(pr_name, "_class_opt_ipv4_ah") == 0 ||
4862 	    strcmp(pr_name, "_class_opt_ipv4_sctp") == 0 ||
4863 	    strcmp(pr_name, "_class_opt_ipv6_tcp") == 0 ||
4864 	    strcmp(pr_name, "_class_opt_ipv6_udp") == 0 ||
4865 	    strcmp(pr_name, "_class_opt_ipv6_ah") == 0 ||
4866 	    strcmp(pr_name, "_class_opt_ipv6_sctp") == 0) {
4867 		(void) snprintf(valstr, sizeof (valstr), "%x",
4868 		    NXGE_CLASS_FLOW_GEN_SERVER);
4869 
4870 	} else if (strcmp(pr_name, "_soft_lso_enable") == 0) {
4871 		(void) snprintf(valstr, sizeof (valstr), "%d", 0);
4872 
4873 	} else 	if (strcmp(pr_name, "_adv_10gfdx_cap") == 0) {
4874 		(void) snprintf(valstr, sizeof (valstr), "%d", 1);
4875 
4876 	} else if (strcmp(pr_name, "_adv_pause_cap") == 0) {
4877 		(void) snprintf(valstr, sizeof (valstr), "%d", 1);
4878 	}
4879 
4880 	if (strlen(valstr) > 0)
4881 		mac_prop_info_set_default_str(prh, valstr);
4882 }
4883 
4884 /* ARGSUSED */
4885 static int
4886 nxge_set_priv_prop(p_nxge_t nxgep, const char *pr_name, uint_t pr_valsize,
4887     const void *pr_val)
4888 {
4889 	p_nxge_param_t	param_arr = nxgep->param_arr;
4890 	int		err = 0;
4891 	long		result;
4892 
4893 	NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
4894 	    "==> nxge_set_priv_prop: name %s", pr_name));
4895 
4896 	/* Blanking */
4897 	if (strcmp(pr_name, "_rxdma_intr_time") == 0) {
4898 		err = nxge_param_rx_intr_time(nxgep, NULL, NULL,
4899 		    (char *)pr_val,
4900 		    (caddr_t)&param_arr[param_rxdma_intr_time]);
4901 		if (err) {
4902 			NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
4903 			    "<== nxge_set_priv_prop: "
4904 			    "unable to set (%s)", pr_name));
4905 			err = EINVAL;
4906 		} else {
4907 			err = 0;
4908 			NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
4909 			    "<== nxge_set_priv_prop: "
4910 			    "set (%s)", pr_name));
4911 		}
4912 
4913 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
4914 		    "<== nxge_set_priv_prop: name %s (value %d)",
4915 		    pr_name, result));
4916 
4917 		return (err);
4918 	}
4919 
4920 	if (strcmp(pr_name, "_rxdma_intr_pkts") == 0) {
4921 		err = nxge_param_rx_intr_pkts(nxgep, NULL, NULL,
4922 		    (char *)pr_val,
4923 		    (caddr_t)&param_arr[param_rxdma_intr_pkts]);
4924 		if (err) {
4925 			NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
4926 			    "<== nxge_set_priv_prop: "
4927 			    "unable to set (%s)", pr_name));
4928 			err = EINVAL;
4929 		} else {
4930 			err = 0;
4931 			NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
4932 			    "<== nxge_set_priv_prop: "
4933 			    "set (%s)", pr_name));
4934 		}
4935 
4936 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
4937 		    "<== nxge_set_priv_prop: name %s (value %d)",
4938 		    pr_name, result));
4939 
4940 		return (err);
4941 	}
4942 
4943 	/* Classification */
4944 	if (strcmp(pr_name, "_class_opt_ipv4_tcp") == 0) {
4945 		if (pr_val == NULL) {
4946 			err = EINVAL;
4947 			return (err);
4948 		}
4949 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
4950 
4951 		err = nxge_param_set_ip_opt(nxgep, NULL,
4952 		    NULL, (char *)pr_val,
4953 		    (caddr_t)&param_arr[param_class_opt_ipv4_tcp]);
4954 
4955 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
4956 		    "<== nxge_set_priv_prop: name %s (value 0x%x)",
4957 		    pr_name, result));
4958 
4959 		return (err);
4960 	}
4961 
4962 	if (strcmp(pr_name, "_class_opt_ipv4_udp") == 0) {
4963 		if (pr_val == NULL) {
4964 			err = EINVAL;
4965 			return (err);
4966 		}
4967 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
4968 
4969 		err = nxge_param_set_ip_opt(nxgep, NULL,
4970 		    NULL, (char *)pr_val,
4971 		    (caddr_t)&param_arr[param_class_opt_ipv4_udp]);
4972 
4973 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
4974 		    "<== nxge_set_priv_prop: name %s (value 0x%x)",
4975 		    pr_name, result));
4976 
4977 		return (err);
4978 	}
4979 	if (strcmp(pr_name, "_class_opt_ipv4_ah") == 0) {
4980 		if (pr_val == NULL) {
4981 			err = EINVAL;
4982 			return (err);
4983 		}
4984 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
4985 
4986 		err = nxge_param_set_ip_opt(nxgep, NULL,
4987 		    NULL, (char *)pr_val,
4988 		    (caddr_t)&param_arr[param_class_opt_ipv4_ah]);
4989 
4990 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
4991 		    "<== nxge_set_priv_prop: name %s (value 0x%x)",
4992 		    pr_name, result));
4993 
4994 		return (err);
4995 	}
4996 	if (strcmp(pr_name, "_class_opt_ipv4_sctp") == 0) {
4997 		if (pr_val == NULL) {
4998 			err = EINVAL;
4999 			return (err);
5000 		}
5001 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
5002 
5003 		err = nxge_param_set_ip_opt(nxgep, NULL,
5004 		    NULL, (char *)pr_val,
5005 		    (caddr_t)&param_arr[param_class_opt_ipv4_sctp]);
5006 
5007 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
5008 		    "<== nxge_set_priv_prop: name %s (value 0x%x)",
5009 		    pr_name, result));
5010 
5011 		return (err);
5012 	}
5013 
5014 	if (strcmp(pr_name, "_class_opt_ipv6_tcp") == 0) {
5015 		if (pr_val == NULL) {
5016 			err = EINVAL;
5017 			return (err);
5018 		}
5019 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
5020 
5021 		err = nxge_param_set_ip_opt(nxgep, NULL,
5022 		    NULL, (char *)pr_val,
5023 		    (caddr_t)&param_arr[param_class_opt_ipv6_tcp]);
5024 
5025 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
5026 		    "<== nxge_set_priv_prop: name %s (value 0x%x)",
5027 		    pr_name, result));
5028 
5029 		return (err);
5030 	}
5031 
5032 	if (strcmp(pr_name, "_class_opt_ipv6_udp") == 0) {
5033 		if (pr_val == NULL) {
5034 			err = EINVAL;
5035 			return (err);
5036 		}
5037 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
5038 
5039 		err = nxge_param_set_ip_opt(nxgep, NULL,
5040 		    NULL, (char *)pr_val,
5041 		    (caddr_t)&param_arr[param_class_opt_ipv6_udp]);
5042 
5043 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
5044 		    "<== nxge_set_priv_prop: name %s (value 0x%x)",
5045 		    pr_name, result));
5046 
5047 		return (err);
5048 	}
5049 	if (strcmp(pr_name, "_class_opt_ipv6_ah") == 0) {
5050 		if (pr_val == NULL) {
5051 			err = EINVAL;
5052 			return (err);
5053 		}
5054 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
5055 
5056 		err = nxge_param_set_ip_opt(nxgep, NULL,
5057 		    NULL, (char *)pr_val,
5058 		    (caddr_t)&param_arr[param_class_opt_ipv6_ah]);
5059 
5060 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
5061 		    "<== nxge_set_priv_prop: name %s (value 0x%x)",
5062 		    pr_name, result));
5063 
5064 		return (err);
5065 	}
5066 	if (strcmp(pr_name, "_class_opt_ipv6_sctp") == 0) {
5067 		if (pr_val == NULL) {
5068 			err = EINVAL;
5069 			return (err);
5070 		}
5071 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
5072 
5073 		err = nxge_param_set_ip_opt(nxgep, NULL,
5074 		    NULL, (char *)pr_val,
5075 		    (caddr_t)&param_arr[param_class_opt_ipv6_sctp]);
5076 
5077 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
5078 		    "<== nxge_set_priv_prop: name %s (value 0x%x)",
5079 		    pr_name, result));
5080 
5081 		return (err);
5082 	}
5083 
5084 	if (strcmp(pr_name, "_soft_lso_enable") == 0) {
5085 		if (pr_val == NULL) {
5086 			NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
5087 			    "==> nxge_set_priv_prop: name %s (null)", pr_name));
5088 			err = EINVAL;
5089 			return (err);
5090 		}
5091 
5092 		(void) ddi_strtol(pr_val, (char **)NULL, 0, &result);
5093 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
5094 		    "<== nxge_set_priv_prop: name %s "
5095 		    "(lso %d pr_val %s value %d)",
5096 		    pr_name, nxgep->soft_lso_enable, pr_val, result));
5097 
5098 		if (result > 1 || result < 0) {
5099 			err = EINVAL;
5100 		} else {
5101 			if (nxgep->soft_lso_enable == (uint32_t)result) {
5102 				NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
5103 				    "no change (%d %d)",
5104 				    nxgep->soft_lso_enable, result));
5105 				return (0);
5106 			}
5107 		}
5108 
5109 		nxgep->soft_lso_enable = (int)result;
5110 
5111 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
5112 		    "<== nxge_set_priv_prop: name %s (value %d)",
5113 		    pr_name, result));
5114 
5115 		return (err);
5116 	}
5117 	/*
5118 	 * Commands like "ndd -set /dev/nxge0 adv_10gfdx_cap 1" cause the
5119 	 * following code to be executed.
5120 	 */
5121 	if (strcmp(pr_name, "_adv_10gfdx_cap") == 0) {
5122 		err = nxge_param_set_mac(nxgep, NULL, NULL, (char *)pr_val,
5123 		    (caddr_t)&param_arr[param_anar_10gfdx]);
5124 		return (err);
5125 	}
5126 	if (strcmp(pr_name, "_adv_pause_cap") == 0) {
5127 		err = nxge_param_set_mac(nxgep, NULL, NULL, (char *)pr_val,
5128 		    (caddr_t)&param_arr[param_anar_pause]);
5129 		return (err);
5130 	}
5131 
5132 	return (EINVAL);
5133 }
5134 
5135 static int
5136 nxge_get_priv_prop(p_nxge_t nxgep, const char *pr_name, uint_t pr_valsize,
5137     void *pr_val)
5138 {
5139 	p_nxge_param_t	param_arr = nxgep->param_arr;
5140 	char		valstr[MAXNAMELEN];
5141 	int		err = EINVAL;
5142 	uint_t		strsize;
5143 
5144 	NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
5145 	    "==> nxge_get_priv_prop: property %s", pr_name));
5146 
5147 	/* function number */
5148 	if (strcmp(pr_name, "_function_number") == 0) {
5149 		(void) snprintf(valstr, sizeof (valstr), "%d",
5150 		    nxgep->function_num);
5151 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
5152 		    "==> nxge_get_priv_prop: name %s "
5153 		    "(value %d valstr %s)",
5154 		    pr_name, nxgep->function_num, valstr));
5155 
5156 		err = 0;
5157 		goto done;
5158 	}
5159 
5160 	/* Neptune firmware version */
5161 	if (strcmp(pr_name, "_fw_version") == 0) {
5162 		(void) snprintf(valstr, sizeof (valstr), "%s",
5163 		    nxgep->vpd_info.ver);
5164 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
5165 		    "==> nxge_get_priv_prop: name %s "
5166 		    "(value %d valstr %s)",
5167 		    pr_name, nxgep->vpd_info.ver, valstr));
5168 
5169 		err = 0;
5170 		goto done;
5171 	}
5172 
5173 	/* port PHY mode */
5174 	if (strcmp(pr_name, "_port_mode") == 0) {
5175 		switch (nxgep->mac.portmode) {
5176 		case PORT_1G_COPPER:
5177 			(void) snprintf(valstr, sizeof (valstr), "1G copper %s",
5178 			    nxgep->hot_swappable_phy ?
5179 			    "[Hot Swappable]" : "");
5180 			break;
5181 		case PORT_1G_FIBER:
5182 			(void) snprintf(valstr, sizeof (valstr), "1G fiber %s",
5183 			    nxgep->hot_swappable_phy ?
5184 			    "[hot swappable]" : "");
5185 			break;
5186 		case PORT_10G_COPPER:
5187 			(void) snprintf(valstr, sizeof (valstr),
5188 			    "10G copper %s",
5189 			    nxgep->hot_swappable_phy ?
5190 			    "[hot swappable]" : "");
5191 			break;
5192 		case PORT_10G_FIBER:
5193 			(void) snprintf(valstr, sizeof (valstr), "10G fiber %s",
5194 			    nxgep->hot_swappable_phy ?
5195 			    "[hot swappable]" : "");
5196 			break;
5197 		case PORT_10G_SERDES:
5198 			(void) snprintf(valstr, sizeof (valstr),
5199 			    "10G serdes %s", nxgep->hot_swappable_phy ?
5200 			    "[hot swappable]" : "");
5201 			break;
5202 		case PORT_1G_SERDES:
5203 			(void) snprintf(valstr, sizeof (valstr), "1G serdes %s",
5204 			    nxgep->hot_swappable_phy ?
5205 			    "[hot swappable]" : "");
5206 			break;
5207 		case PORT_1G_TN1010:
5208 			(void) snprintf(valstr, sizeof (valstr),
5209 			    "1G TN1010 copper %s", nxgep->hot_swappable_phy ?
5210 			    "[hot swappable]" : "");
5211 			break;
5212 		case PORT_10G_TN1010:
5213 			(void) snprintf(valstr, sizeof (valstr),
5214 			    "10G TN1010 copper %s", nxgep->hot_swappable_phy ?
5215 			    "[hot swappable]" : "");
5216 			break;
5217 		case PORT_1G_RGMII_FIBER:
5218 			(void) snprintf(valstr, sizeof (valstr),
5219 			    "1G rgmii fiber %s", nxgep->hot_swappable_phy ?
5220 			    "[hot swappable]" : "");
5221 			break;
5222 		case PORT_HSP_MODE:
5223 			(void) snprintf(valstr, sizeof (valstr),
5224 			    "phy not present[hot swappable]");
5225 			break;
5226 		default:
5227 			(void) snprintf(valstr, sizeof (valstr), "unknown %s",
5228 			    nxgep->hot_swappable_phy ?
5229 			    "[hot swappable]" : "");
5230 			break;
5231 		}
5232 
5233 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
5234 		    "==> nxge_get_priv_prop: name %s (value %s)",
5235 		    pr_name, valstr));
5236 
5237 		err = 0;
5238 		goto done;
5239 	}
5240 
5241 	/* Hot swappable PHY */
5242 	if (strcmp(pr_name, "_hot_swap_phy") == 0) {
5243 		(void) snprintf(valstr, sizeof (valstr), "%s",
5244 		    nxgep->hot_swappable_phy ?
5245 		    "yes" : "no");
5246 
5247 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
5248 		    "==> nxge_get_priv_prop: name %s "
5249 		    "(value %d valstr %s)",
5250 		    pr_name, nxgep->hot_swappable_phy, valstr));
5251 
5252 		err = 0;
5253 		goto done;
5254 	}
5255 
5256 
5257 	/* Receive Interrupt Blanking Parameters */
5258 	if (strcmp(pr_name, "_rxdma_intr_time") == 0) {
5259 		err = 0;
5260 		(void) snprintf(valstr, sizeof (valstr), "%d",
5261 		    nxgep->intr_timeout);
5262 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
5263 		    "==> nxge_get_priv_prop: name %s (value %d)",
5264 		    pr_name,
5265 		    (uint32_t)nxgep->intr_timeout));
5266 		goto done;
5267 	}
5268 
5269 	if (strcmp(pr_name, "_rxdma_intr_pkts") == 0) {
5270 		err = 0;
5271 		(void) snprintf(valstr, sizeof (valstr), "%d",
5272 		    nxgep->intr_threshold);
5273 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
5274 		    "==> nxge_get_priv_prop: name %s (value %d)",
5275 		    pr_name, (uint32_t)nxgep->intr_threshold));
5276 
5277 		goto done;
5278 	}
5279 
5280 	/* Classification and Load Distribution Configuration */
5281 	if (strcmp(pr_name, "_class_opt_ipv4_tcp") == 0) {
5282 		err = nxge_dld_get_ip_opt(nxgep,
5283 		    (caddr_t)&param_arr[param_class_opt_ipv4_tcp]);
5284 
5285 		(void) snprintf(valstr, sizeof (valstr), "%x",
5286 		    (int)param_arr[param_class_opt_ipv4_tcp].value);
5287 
5288 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
5289 		    "==> nxge_get_priv_prop: %s", valstr));
5290 		goto done;
5291 	}
5292 
5293 	if (strcmp(pr_name, "_class_opt_ipv4_udp") == 0) {
5294 		err = nxge_dld_get_ip_opt(nxgep,
5295 		    (caddr_t)&param_arr[param_class_opt_ipv4_udp]);
5296 
5297 		(void) snprintf(valstr, sizeof (valstr), "%x",
5298 		    (int)param_arr[param_class_opt_ipv4_udp].value);
5299 
5300 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
5301 		    "==> nxge_get_priv_prop: %s", valstr));
5302 		goto done;
5303 	}
5304 	if (strcmp(pr_name, "_class_opt_ipv4_ah") == 0) {
5305 		err = nxge_dld_get_ip_opt(nxgep,
5306 		    (caddr_t)&param_arr[param_class_opt_ipv4_ah]);
5307 
5308 		(void) snprintf(valstr, sizeof (valstr), "%x",
5309 		    (int)param_arr[param_class_opt_ipv4_ah].value);
5310 
5311 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
5312 		    "==> nxge_get_priv_prop: %s", valstr));
5313 		goto done;
5314 	}
5315 
5316 	if (strcmp(pr_name, "_class_opt_ipv4_sctp") == 0) {
5317 		err = nxge_dld_get_ip_opt(nxgep,
5318 		    (caddr_t)&param_arr[param_class_opt_ipv4_sctp]);
5319 
5320 		(void) snprintf(valstr, sizeof (valstr), "%x",
5321 		    (int)param_arr[param_class_opt_ipv4_sctp].value);
5322 
5323 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
5324 		    "==> nxge_get_priv_prop: %s", valstr));
5325 		goto done;
5326 	}
5327 
5328 	if (strcmp(pr_name, "_class_opt_ipv6_tcp") == 0) {
5329 		err = nxge_dld_get_ip_opt(nxgep,
5330 		    (caddr_t)&param_arr[param_class_opt_ipv6_tcp]);
5331 
5332 		(void) snprintf(valstr, sizeof (valstr), "%x",
5333 		    (int)param_arr[param_class_opt_ipv6_tcp].value);
5334 
5335 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
5336 		    "==> nxge_get_priv_prop: %s", valstr));
5337 		goto done;
5338 	}
5339 
5340 	if (strcmp(pr_name, "_class_opt_ipv6_udp") == 0) {
5341 		err = nxge_dld_get_ip_opt(nxgep,
5342 		    (caddr_t)&param_arr[param_class_opt_ipv6_udp]);
5343 
5344 		(void) snprintf(valstr, sizeof (valstr), "%x",
5345 		    (int)param_arr[param_class_opt_ipv6_udp].value);
5346 
5347 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
5348 		    "==> nxge_get_priv_prop: %s", valstr));
5349 		goto done;
5350 	}
5351 
5352 	if (strcmp(pr_name, "_class_opt_ipv6_ah") == 0) {
5353 		err = nxge_dld_get_ip_opt(nxgep,
5354 		    (caddr_t)&param_arr[param_class_opt_ipv6_ah]);
5355 
5356 		(void) snprintf(valstr, sizeof (valstr), "%x",
5357 		    (int)param_arr[param_class_opt_ipv6_ah].value);
5358 
5359 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
5360 		    "==> nxge_get_priv_prop: %s", valstr));
5361 		goto done;
5362 	}
5363 
5364 	if (strcmp(pr_name, "_class_opt_ipv6_sctp") == 0) {
5365 		err = nxge_dld_get_ip_opt(nxgep,
5366 		    (caddr_t)&param_arr[param_class_opt_ipv6_sctp]);
5367 
5368 		(void) snprintf(valstr, sizeof (valstr), "%x",
5369 		    (int)param_arr[param_class_opt_ipv6_sctp].value);
5370 
5371 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
5372 		    "==> nxge_get_priv_prop: %s", valstr));
5373 		goto done;
5374 	}
5375 
5376 	/* Software LSO */
5377 	if (strcmp(pr_name, "_soft_lso_enable") == 0) {
5378 		(void) snprintf(valstr, sizeof (valstr),
5379 		    "%d", nxgep->soft_lso_enable);
5380 		err = 0;
5381 		NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
5382 		    "==> nxge_get_priv_prop: name %s (value %d)",
5383 		    pr_name, nxgep->soft_lso_enable));
5384 
5385 		goto done;
5386 	}
5387 	if (strcmp(pr_name, "_adv_10gfdx_cap") == 0) {
5388 		err = 0;
5389 		if (nxgep->param_arr[param_anar_10gfdx].value != 0) {
5390 			(void) snprintf(valstr, sizeof (valstr), "%d", 1);
5391 			goto done;
5392 		} else {
5393 			(void) snprintf(valstr, sizeof (valstr), "%d", 0);
5394 			goto done;
5395 		}
5396 	}
5397 	if (strcmp(pr_name, "_adv_pause_cap") == 0) {
5398 		err = 0;
5399 		if (nxgep->param_arr[param_anar_pause].value != 0) {
5400 			(void) snprintf(valstr, sizeof (valstr), "%d", 1);
5401 			goto done;
5402 		} else {
5403 			(void) snprintf(valstr, sizeof (valstr), "%d", 0);
5404 			goto done;
5405 		}
5406 	}
5407 
5408 done:
5409 	if (err == 0) {
5410 		strsize = (uint_t)strlen(valstr);
5411 		if (pr_valsize < strsize) {
5412 			err = ENOBUFS;
5413 		} else {
5414 			(void) strlcpy(pr_val, valstr, pr_valsize);
5415 		}
5416 	}
5417 
5418 	NXGE_DEBUG_MSG((nxgep, NXGE_CTL,
5419 	    "<== nxge_get_priv_prop: return %d", err));
5420 	return (err);
5421 }
5422 
5423 /*
5424  * Module loading and removing entry points.
5425  */
5426 
5427 DDI_DEFINE_STREAM_OPS(nxge_dev_ops, nulldev, nulldev, nxge_attach, nxge_detach,
5428     nodev, NULL, D_MP, NULL, nxge_quiesce);
5429 
5430 #define	NXGE_DESC_VER		"Sun NIU 10Gb Ethernet"
5431 
5432 /*
5433  * Module linkage information for the kernel.
5434  */
5435 static struct modldrv 	nxge_modldrv = {
5436 	&mod_driverops,
5437 	NXGE_DESC_VER,
5438 	&nxge_dev_ops
5439 };
5440 
5441 static struct modlinkage modlinkage = {
5442 	MODREV_1, (void *) &nxge_modldrv, NULL
5443 };
5444 
5445 int
5446 _init(void)
5447 {
5448 	int		status;
5449 
5450 	MUTEX_INIT(&nxgedebuglock, NULL, MUTEX_DRIVER, NULL);
5451 
5452 	NXGE_DEBUG_MSG((NULL, MOD_CTL, "==> _init"));
5453 
5454 	mac_init_ops(&nxge_dev_ops, "nxge");
5455 
5456 	status = ddi_soft_state_init(&nxge_list, sizeof (nxge_t), 0);
5457 	if (status != 0) {
5458 		NXGE_ERROR_MSG((NULL, NXGE_ERR_CTL,
5459 		    "failed to init device soft state"));
5460 		goto _init_exit;
5461 	}
5462 
5463 	status = mod_install(&modlinkage);
5464 	if (status != 0) {
5465 		ddi_soft_state_fini(&nxge_list);
5466 		NXGE_ERROR_MSG((NULL, NXGE_ERR_CTL, "Mod install failed"));
5467 		goto _init_exit;
5468 	}
5469 
5470 	MUTEX_INIT(&nxge_common_lock, NULL, MUTEX_DRIVER, NULL);
5471 
5472 	NXGE_DEBUG_MSG((NULL, MOD_CTL, "<== _init status = 0x%X", status));
5473 	return (status);
5474 
5475 _init_exit:
5476 	NXGE_DEBUG_MSG((NULL, MOD_CTL, "<== _init status = 0x%X", status));
5477 	MUTEX_DESTROY(&nxgedebuglock);
5478 	return (status);
5479 }
5480 
5481 int
5482 _fini(void)
5483 {
5484 	int		status;
5485 
5486 	NXGE_DEBUG_MSG((NULL, MOD_CTL, "==> _fini"));
5487 	NXGE_DEBUG_MSG((NULL, MOD_CTL, "==> _fini: mod_remove"));
5488 
5489 	if (nxge_mblks_pending)
5490 		return (EBUSY);
5491 
5492 	status = mod_remove(&modlinkage);
5493 	if (status != DDI_SUCCESS) {
5494 		NXGE_DEBUG_MSG((NULL, MOD_CTL,
5495 		    "Module removal failed 0x%08x",
5496 		    status));
5497 		goto _fini_exit;
5498 	}
5499 
5500 	mac_fini_ops(&nxge_dev_ops);
5501 
5502 	ddi_soft_state_fini(&nxge_list);
5503 
5504 	NXGE_DEBUG_MSG((NULL, MOD_CTL, "<== _fini status = 0x%08x", status));
5505 
5506 	MUTEX_DESTROY(&nxge_common_lock);
5507 	MUTEX_DESTROY(&nxgedebuglock);
5508 	return (status);
5509 
5510 _fini_exit:
5511 	NXGE_DEBUG_MSG((NULL, MOD_CTL, "<== _fini status = 0x%08x", status));
5512 	return (status);
5513 }
5514 
5515 int
5516 _info(struct modinfo *modinfop)
5517 {
5518 	int		status;
5519 
5520 	NXGE_DEBUG_MSG((NULL, MOD_CTL, "==> _info"));
5521 	status = mod_info(&modlinkage, modinfop);
5522 	NXGE_DEBUG_MSG((NULL, MOD_CTL, " _info status = 0x%X", status));
5523 
5524 	return (status);
5525 }
5526 
5527 /*ARGSUSED*/
5528 static int
5529 nxge_tx_ring_start(mac_ring_driver_t rdriver, uint64_t mr_gen_num)
5530 {
5531 	p_nxge_ring_handle_t	rhp = (p_nxge_ring_handle_t)rdriver;
5532 	p_nxge_t		nxgep = rhp->nxgep;
5533 	uint32_t		channel;
5534 	p_tx_ring_t		ring;
5535 
5536 	channel = nxgep->pt_config.hw_config.tdc.start + rhp->index;
5537 	ring = nxgep->tx_rings->rings[channel];
5538 
5539 	MUTEX_ENTER(&ring->lock);
5540 	ASSERT(ring->tx_ring_handle == NULL);
5541 	ring->tx_ring_handle = rhp->ring_handle;
5542 	MUTEX_EXIT(&ring->lock);
5543 
5544 	return (0);
5545 }
5546 
5547 static void
5548 nxge_tx_ring_stop(mac_ring_driver_t rdriver)
5549 {
5550 	p_nxge_ring_handle_t	rhp = (p_nxge_ring_handle_t)rdriver;
5551 	p_nxge_t		nxgep = rhp->nxgep;
5552 	uint32_t		channel;
5553 	p_tx_ring_t		ring;
5554 
5555 	channel = nxgep->pt_config.hw_config.tdc.start + rhp->index;
5556 	ring = nxgep->tx_rings->rings[channel];
5557 
5558 	MUTEX_ENTER(&ring->lock);
5559 	ASSERT(ring->tx_ring_handle != NULL);
5560 	ring->tx_ring_handle = (mac_ring_handle_t)NULL;
5561 	MUTEX_EXIT(&ring->lock);
5562 }
5563 
5564 int
5565 nxge_rx_ring_start(mac_ring_driver_t rdriver, uint64_t mr_gen_num)
5566 {
5567 	p_nxge_ring_handle_t	rhp = (p_nxge_ring_handle_t)rdriver;
5568 	p_nxge_t		nxgep = rhp->nxgep;
5569 	uint32_t		channel;
5570 	p_rx_rcr_ring_t		ring;
5571 	int			i;
5572 
5573 	channel = nxgep->pt_config.hw_config.start_rdc + rhp->index;
5574 	ring =  nxgep->rx_rcr_rings->rcr_rings[channel];
5575 
5576 	MUTEX_ENTER(&ring->lock);
5577 
5578 	if (ring->started) {
5579 		ASSERT(ring->started == B_FALSE);
5580 		MUTEX_EXIT(&ring->lock);
5581 		return (0);
5582 	}
5583 
5584 	/* set rcr_ring */
5585 	for (i = 0; i < nxgep->ldgvp->maxldvs; i++) {
5586 		if ((nxgep->ldgvp->ldvp[i].is_rxdma) &&
5587 		    (nxgep->ldgvp->ldvp[i].channel == channel)) {
5588 			ring->ldvp = &nxgep->ldgvp->ldvp[i];
5589 			ring->ldgp = nxgep->ldgvp->ldvp[i].ldgp;
5590 		}
5591 	}
5592 
5593 	ring->rcr_mac_handle = rhp->ring_handle;
5594 	ring->rcr_gen_num = mr_gen_num;
5595 	ring->started = B_TRUE;
5596 	rhp->ring_gen_num = mr_gen_num;
5597 	MUTEX_EXIT(&ring->lock);
5598 
5599 	return (0);
5600 }
5601 
5602 static void
5603 nxge_rx_ring_stop(mac_ring_driver_t rdriver)
5604 {
5605 	p_nxge_ring_handle_t	rhp = (p_nxge_ring_handle_t)rdriver;
5606 	p_nxge_t		nxgep = rhp->nxgep;
5607 	uint32_t		channel;
5608 	p_rx_rcr_ring_t		ring;
5609 
5610 	channel = nxgep->pt_config.hw_config.start_rdc + rhp->index;
5611 	ring =  nxgep->rx_rcr_rings->rcr_rings[channel];
5612 
5613 	MUTEX_ENTER(&ring->lock);
5614 	ASSERT(ring->started == B_TRUE);
5615 	ring->rcr_mac_handle = NULL;
5616 	ring->ldvp = NULL;
5617 	ring->ldgp = NULL;
5618 	ring->started = B_FALSE;
5619 	MUTEX_EXIT(&ring->lock);
5620 }
5621 
5622 static int
5623 nxge_ring_get_htable_idx(p_nxge_t nxgep, mac_ring_type_t type, uint32_t channel)
5624 {
5625 	int	i;
5626 
5627 #if defined(sun4v)
5628 	if (isLDOMguest(nxgep)) {
5629 		return (nxge_hio_get_dc_htable_idx(nxgep,
5630 		    (type == MAC_RING_TYPE_TX) ? VP_BOUND_TX : VP_BOUND_RX,
5631 		    channel));
5632 	}
5633 #endif
5634 
5635 	ASSERT(nxgep->ldgvp != NULL);
5636 
5637 	switch (type) {
5638 	case MAC_RING_TYPE_TX:
5639 		for (i = 0; i < nxgep->ldgvp->maxldvs; i++) {
5640 			if ((nxgep->ldgvp->ldvp[i].is_txdma) &&
5641 			    (nxgep->ldgvp->ldvp[i].channel == channel)) {
5642 				return ((int)
5643 				    nxgep->ldgvp->ldvp[i].ldgp->htable_idx);
5644 			}
5645 		}
5646 		break;
5647 
5648 	case MAC_RING_TYPE_RX:
5649 		for (i = 0; i < nxgep->ldgvp->maxldvs; i++) {
5650 			if ((nxgep->ldgvp->ldvp[i].is_rxdma) &&
5651 			    (nxgep->ldgvp->ldvp[i].channel == channel)) {
5652 				return ((int)
5653 				    nxgep->ldgvp->ldvp[i].ldgp->htable_idx);
5654 			}
5655 		}
5656 	}
5657 
5658 	return (-1);
5659 }
5660 
5661 /*
5662  * Callback funtion for MAC layer to register all rings.
5663  */
5664 static void
5665 nxge_fill_ring(void *arg, mac_ring_type_t rtype, const int rg_index,
5666     const int index, mac_ring_info_t *infop, mac_ring_handle_t rh)
5667 {
5668 	p_nxge_t		nxgep = (p_nxge_t)arg;
5669 	p_nxge_hw_pt_cfg_t	p_cfgp = &nxgep->pt_config.hw_config;
5670 	p_nxge_intr_t		intrp;
5671 	uint32_t		channel;
5672 	int			htable_idx;
5673 	p_nxge_ring_handle_t	rhandlep;
5674 
5675 	ASSERT(nxgep != NULL);
5676 	ASSERT(p_cfgp != NULL);
5677 	ASSERT(infop != NULL);
5678 
5679 	NXGE_DEBUG_MSG((nxgep, DDI_CTL,
5680 	    "==> nxge_fill_ring 0x%x index %d", rtype, index));
5681 
5682 
5683 	switch (rtype) {
5684 	case MAC_RING_TYPE_TX: {
5685 		mac_intr_t	*mintr = &infop->mri_intr;
5686 
5687 		NXGE_DEBUG_MSG((nxgep, TX_CTL,
5688 		    "==> nxge_fill_ring (TX) 0x%x index %d ntdcs %d",
5689 		    rtype, index, p_cfgp->tdc.count));
5690 
5691 		ASSERT((index >= 0) && (index < p_cfgp->tdc.count));
5692 		rhandlep = &nxgep->tx_ring_handles[index];
5693 		rhandlep->nxgep = nxgep;
5694 		rhandlep->index = index;
5695 		rhandlep->ring_handle = rh;
5696 
5697 		channel = nxgep->pt_config.hw_config.tdc.start + index;
5698 		rhandlep->channel = channel;
5699 		intrp = (p_nxge_intr_t)&nxgep->nxge_intr_type;
5700 		htable_idx = nxge_ring_get_htable_idx(nxgep, rtype,
5701 		    channel);
5702 		if (htable_idx >= 0)
5703 			mintr->mi_ddi_handle = intrp->htable[htable_idx];
5704 		else
5705 			mintr->mi_ddi_handle = NULL;
5706 
5707 		infop->mri_driver = (mac_ring_driver_t)rhandlep;
5708 		infop->mri_start = nxge_tx_ring_start;
5709 		infop->mri_stop = nxge_tx_ring_stop;
5710 		infop->mri_tx = nxge_tx_ring_send;
5711 		infop->mri_stat = nxge_tx_ring_stat;
5712 		infop->mri_flags = MAC_RING_TX_SERIALIZE;
5713 		break;
5714 	}
5715 
5716 	case MAC_RING_TYPE_RX: {
5717 		mac_intr_t		nxge_mac_intr;
5718 		int			nxge_rindex;
5719 		p_nxge_intr_t		intrp;
5720 
5721 		intrp = (p_nxge_intr_t)&nxgep->nxge_intr_type;
5722 
5723 		NXGE_DEBUG_MSG((nxgep, RX_CTL,
5724 		    "==> nxge_fill_ring (RX) 0x%x index %d nrdcs %d",
5725 		    rtype, index, p_cfgp->max_rdcs));
5726 
5727 		/*
5728 		 * 'index' is the ring index within the group.
5729 		 * Find the ring index in the nxge instance.
5730 		 */
5731 		nxge_rindex = nxge_get_rxring_index(nxgep, rg_index, index);
5732 		channel = nxgep->pt_config.hw_config.start_rdc + index;
5733 		intrp = (p_nxge_intr_t)&nxgep->nxge_intr_type;
5734 
5735 		ASSERT((nxge_rindex >= 0) && (nxge_rindex < p_cfgp->max_rdcs));
5736 		rhandlep = &nxgep->rx_ring_handles[nxge_rindex];
5737 		rhandlep->nxgep = nxgep;
5738 		rhandlep->index = nxge_rindex;
5739 		rhandlep->ring_handle = rh;
5740 		rhandlep->channel = channel;
5741 
5742 		/*
5743 		 * Entrypoint to enable interrupt (disable poll) and
5744 		 * disable interrupt (enable poll).
5745 		 */
5746 		bzero(&nxge_mac_intr, sizeof (nxge_mac_intr));
5747 		nxge_mac_intr.mi_handle = (mac_intr_handle_t)rhandlep;
5748 		nxge_mac_intr.mi_enable = (mac_intr_enable_t)nxge_disable_poll;
5749 		nxge_mac_intr.mi_disable = (mac_intr_disable_t)nxge_enable_poll;
5750 
5751 		htable_idx =  nxge_ring_get_htable_idx(nxgep, rtype,
5752 		    channel);
5753 		if (htable_idx >= 0)
5754 			nxge_mac_intr.mi_ddi_handle = intrp->htable[htable_idx];
5755 		else
5756 			nxge_mac_intr.mi_ddi_handle = NULL;
5757 
5758 		infop->mri_driver = (mac_ring_driver_t)rhandlep;
5759 		infop->mri_start = nxge_rx_ring_start;
5760 		infop->mri_stop = nxge_rx_ring_stop;
5761 		infop->mri_intr = nxge_mac_intr;
5762 		infop->mri_poll = nxge_rx_poll;
5763 		infop->mri_stat = nxge_rx_ring_stat;
5764 		infop->mri_flags = MAC_RING_RX_ENQUEUE;
5765 		break;
5766 	}
5767 
5768 	default:
5769 		break;
5770 	}
5771 
5772 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "<== nxge_fill_ring 0x%x", rtype));
5773 }
5774 
5775 static void
5776 nxge_group_add_ring(mac_group_driver_t gh, mac_ring_driver_t rh,
5777     mac_ring_type_t type)
5778 {
5779 	nxge_ring_group_t	*rgroup = (nxge_ring_group_t *)gh;
5780 	nxge_ring_handle_t	*rhandle = (nxge_ring_handle_t *)rh;
5781 	nxge_t			*nxge;
5782 	nxge_grp_t		*grp;
5783 	nxge_rdc_grp_t		*rdc_grp;
5784 	uint16_t		channel;	/* device-wise ring id */
5785 	int			dev_gindex;
5786 	int			rv;
5787 
5788 	nxge = rgroup->nxgep;
5789 
5790 	switch (type) {
5791 	case MAC_RING_TYPE_TX:
5792 		/*
5793 		 * nxge_grp_dc_add takes a channel number which is a
5794 		 * "devise" ring ID.
5795 		 */
5796 		channel = nxge->pt_config.hw_config.tdc.start + rhandle->index;
5797 
5798 		/*
5799 		 * Remove the ring from the default group
5800 		 */
5801 		if (rgroup->gindex != 0) {
5802 			(void) nxge_grp_dc_remove(nxge, VP_BOUND_TX, channel);
5803 		}
5804 
5805 		/*
5806 		 * nxge->tx_set.group[] is an array of groups indexed by
5807 		 * a "port" group ID.
5808 		 */
5809 		grp = nxge->tx_set.group[rgroup->gindex];
5810 		rv = nxge_grp_dc_add(nxge, grp, VP_BOUND_TX, channel);
5811 		if (rv != 0) {
5812 			NXGE_ERROR_MSG((nxge, NXGE_ERR_CTL,
5813 			    "nxge_group_add_ring: nxge_grp_dc_add failed"));
5814 		}
5815 		break;
5816 
5817 	case MAC_RING_TYPE_RX:
5818 		/*
5819 		 * nxge->rx_set.group[] is an array of groups indexed by
5820 		 * a "port" group ID.
5821 		 */
5822 		grp = nxge->rx_set.group[rgroup->gindex];
5823 
5824 		dev_gindex = nxge->pt_config.hw_config.def_mac_rxdma_grpid +
5825 		    rgroup->gindex;
5826 		rdc_grp = &nxge->pt_config.rdc_grps[dev_gindex];
5827 
5828 		/*
5829 		 * nxge_grp_dc_add takes a channel number which is a
5830 		 * "devise" ring ID.
5831 		 */
5832 		channel = nxge->pt_config.hw_config.start_rdc + rhandle->index;
5833 		rv = nxge_grp_dc_add(nxge, grp, VP_BOUND_RX, channel);
5834 		if (rv != 0) {
5835 			NXGE_ERROR_MSG((nxge, NXGE_ERR_CTL,
5836 			    "nxge_group_add_ring: nxge_grp_dc_add failed"));
5837 		}
5838 
5839 		rdc_grp->map |= (1 << channel);
5840 		rdc_grp->max_rdcs++;
5841 
5842 		(void) nxge_init_fzc_rdc_tbl(nxge, rdc_grp, rgroup->rdctbl);
5843 		break;
5844 	}
5845 }
5846 
5847 static void
5848 nxge_group_rem_ring(mac_group_driver_t gh, mac_ring_driver_t rh,
5849     mac_ring_type_t type)
5850 {
5851 	nxge_ring_group_t	*rgroup = (nxge_ring_group_t *)gh;
5852 	nxge_ring_handle_t	*rhandle = (nxge_ring_handle_t *)rh;
5853 	nxge_t			*nxge;
5854 	uint16_t		channel;	/* device-wise ring id */
5855 	nxge_rdc_grp_t		*rdc_grp;
5856 	int			dev_gindex;
5857 
5858 	nxge = rgroup->nxgep;
5859 
5860 	switch (type) {
5861 	case MAC_RING_TYPE_TX:
5862 		dev_gindex = nxge->pt_config.hw_config.def_mac_txdma_grpid +
5863 		    rgroup->gindex;
5864 		channel = nxge->pt_config.hw_config.tdc.start + rhandle->index;
5865 		nxge_grp_dc_remove(nxge, VP_BOUND_TX, channel);
5866 
5867 		/*
5868 		 * Add the ring back to the default group
5869 		 */
5870 		if (rgroup->gindex != 0) {
5871 			nxge_grp_t *grp;
5872 			grp = nxge->tx_set.group[0];
5873 			(void) nxge_grp_dc_add(nxge, grp, VP_BOUND_TX, channel);
5874 		}
5875 		break;
5876 
5877 	case MAC_RING_TYPE_RX:
5878 		dev_gindex = nxge->pt_config.hw_config.def_mac_rxdma_grpid +
5879 		    rgroup->gindex;
5880 		rdc_grp = &nxge->pt_config.rdc_grps[dev_gindex];
5881 		channel = rdc_grp->start_rdc + rhandle->index;
5882 		nxge_grp_dc_remove(nxge, VP_BOUND_RX, channel);
5883 
5884 		rdc_grp->map &= ~(1 << channel);
5885 		rdc_grp->max_rdcs--;
5886 
5887 		(void) nxge_init_fzc_rdc_tbl(nxge, rdc_grp, rgroup->rdctbl);
5888 		break;
5889 	}
5890 }
5891 
5892 
5893 /*ARGSUSED*/
5894 static nxge_status_t
5895 nxge_add_intrs(p_nxge_t nxgep)
5896 {
5897 
5898 	int		intr_types;
5899 	int		type = 0;
5900 	int		ddi_status = DDI_SUCCESS;
5901 	nxge_status_t	status = NXGE_OK;
5902 
5903 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "==> nxge_add_intrs"));
5904 
5905 	nxgep->nxge_intr_type.intr_registered = B_FALSE;
5906 	nxgep->nxge_intr_type.intr_enabled = B_FALSE;
5907 	nxgep->nxge_intr_type.msi_intx_cnt = 0;
5908 	nxgep->nxge_intr_type.intr_added = 0;
5909 	nxgep->nxge_intr_type.niu_msi_enable = B_FALSE;
5910 	nxgep->nxge_intr_type.intr_type = 0;
5911 
5912 	if (nxgep->niu_type == N2_NIU) {
5913 		nxgep->nxge_intr_type.niu_msi_enable = B_TRUE;
5914 	} else if (nxge_msi_enable) {
5915 		nxgep->nxge_intr_type.niu_msi_enable = B_TRUE;
5916 	}
5917 
5918 	/* Get the supported interrupt types */
5919 	if ((ddi_status = ddi_intr_get_supported_types(nxgep->dip, &intr_types))
5920 	    != DDI_SUCCESS) {
5921 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL, "<== nxge_add_intrs: "
5922 		    "ddi_intr_get_supported_types failed: status 0x%08x",
5923 		    ddi_status));
5924 		return (NXGE_ERROR | NXGE_DDI_FAILED);
5925 	}
5926 	nxgep->nxge_intr_type.intr_types = intr_types;
5927 
5928 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "==> nxge_add_intrs: "
5929 	    "ddi_intr_get_supported_types: 0x%08x", intr_types));
5930 
5931 	/*
5932 	 * Solaris MSIX is not supported yet. use MSI for now.
5933 	 * nxge_msi_enable (1):
5934 	 *	1 - MSI		2 - MSI-X	others - FIXED
5935 	 */
5936 	switch (nxge_msi_enable) {
5937 	default:
5938 		type = DDI_INTR_TYPE_FIXED;
5939 		NXGE_DEBUG_MSG((nxgep, INT_CTL, "==> nxge_add_intrs: "
5940 		    "use fixed (intx emulation) type %08x",
5941 		    type));
5942 		break;
5943 
5944 	case 2:
5945 		NXGE_DEBUG_MSG((nxgep, INT_CTL, "==> nxge_add_intrs: "
5946 		    "ddi_intr_get_supported_types: 0x%08x", intr_types));
5947 		if (intr_types & DDI_INTR_TYPE_MSIX) {
5948 			type = DDI_INTR_TYPE_MSIX;
5949 			NXGE_DEBUG_MSG((nxgep, DDI_CTL, "==> nxge_add_intrs: "
5950 			    "ddi_intr_get_supported_types: MSIX 0x%08x",
5951 			    type));
5952 		} else if (intr_types & DDI_INTR_TYPE_MSI) {
5953 			type = DDI_INTR_TYPE_MSI;
5954 			NXGE_DEBUG_MSG((nxgep, DDI_CTL, "==> nxge_add_intrs: "
5955 			    "ddi_intr_get_supported_types: MSI 0x%08x",
5956 			    type));
5957 		} else if (intr_types & DDI_INTR_TYPE_FIXED) {
5958 			type = DDI_INTR_TYPE_FIXED;
5959 			NXGE_DEBUG_MSG((nxgep, INT_CTL, "==> nxge_add_intrs: "
5960 			    "ddi_intr_get_supported_types: MSXED0x%08x",
5961 			    type));
5962 		}
5963 		break;
5964 
5965 	case 1:
5966 		if (intr_types & DDI_INTR_TYPE_MSI) {
5967 			type = DDI_INTR_TYPE_MSI;
5968 			NXGE_DEBUG_MSG((nxgep, INT_CTL, "==> nxge_add_intrs: "
5969 			    "ddi_intr_get_supported_types: MSI 0x%08x",
5970 			    type));
5971 		} else if (intr_types & DDI_INTR_TYPE_MSIX) {
5972 			type = DDI_INTR_TYPE_MSIX;
5973 			NXGE_DEBUG_MSG((nxgep, DDI_CTL, "==> nxge_add_intrs: "
5974 			    "ddi_intr_get_supported_types: MSIX 0x%08x",
5975 			    type));
5976 		} else if (intr_types & DDI_INTR_TYPE_FIXED) {
5977 			type = DDI_INTR_TYPE_FIXED;
5978 			NXGE_DEBUG_MSG((nxgep, DDI_CTL, "==> nxge_add_intrs: "
5979 			    "ddi_intr_get_supported_types: MSXED0x%08x",
5980 			    type));
5981 		}
5982 	}
5983 
5984 	nxgep->nxge_intr_type.intr_type = type;
5985 	if ((type == DDI_INTR_TYPE_MSIX || type == DDI_INTR_TYPE_MSI ||
5986 	    type == DDI_INTR_TYPE_FIXED) &&
5987 	    nxgep->nxge_intr_type.niu_msi_enable) {
5988 		if ((status = nxge_add_intrs_adv(nxgep)) != DDI_SUCCESS) {
5989 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
5990 			    " nxge_add_intrs: "
5991 			    " nxge_add_intrs_adv failed: status 0x%08x",
5992 			    status));
5993 			return (status);
5994 		} else {
5995 			NXGE_DEBUG_MSG((nxgep, DDI_CTL, "==> nxge_add_intrs: "
5996 			    "interrupts registered : type %d", type));
5997 			nxgep->nxge_intr_type.intr_registered = B_TRUE;
5998 
5999 			NXGE_DEBUG_MSG((nxgep, DDI_CTL,
6000 			    "\nAdded advanced nxge add_intr_adv "
6001 			    "intr type 0x%x\n", type));
6002 
6003 			return (status);
6004 		}
6005 	}
6006 
6007 	if (!nxgep->nxge_intr_type.intr_registered) {
6008 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL, "==> nxge_add_intrs: "
6009 		    "failed to register interrupts"));
6010 		return (NXGE_ERROR | NXGE_DDI_FAILED);
6011 	}
6012 
6013 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "<== nxge_add_intrs"));
6014 	return (status);
6015 }
6016 
6017 static nxge_status_t
6018 nxge_add_intrs_adv(p_nxge_t nxgep)
6019 {
6020 	int		intr_type;
6021 	p_nxge_intr_t	intrp;
6022 
6023 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "==> nxge_add_intrs_adv"));
6024 
6025 	intrp = (p_nxge_intr_t)&nxgep->nxge_intr_type;
6026 	intr_type = intrp->intr_type;
6027 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "==> nxge_add_intrs_adv: type 0x%x",
6028 	    intr_type));
6029 
6030 	switch (intr_type) {
6031 	case DDI_INTR_TYPE_MSI: /* 0x2 */
6032 	case DDI_INTR_TYPE_MSIX: /* 0x4 */
6033 		return (nxge_add_intrs_adv_type(nxgep, intr_type));
6034 
6035 	case DDI_INTR_TYPE_FIXED: /* 0x1 */
6036 		return (nxge_add_intrs_adv_type_fix(nxgep, intr_type));
6037 
6038 	default:
6039 		return (NXGE_ERROR);
6040 	}
6041 }
6042 
6043 
6044 /*ARGSUSED*/
6045 static nxge_status_t
6046 nxge_add_intrs_adv_type(p_nxge_t nxgep, uint32_t int_type)
6047 {
6048 	dev_info_t		*dip = nxgep->dip;
6049 	p_nxge_ldg_t		ldgp;
6050 	p_nxge_intr_t		intrp;
6051 	uint_t			*inthandler;
6052 	void			*arg1, *arg2;
6053 	int			behavior;
6054 	int			nintrs, navail, nrequest;
6055 	int			nactual, nrequired;
6056 	int			inum = 0;
6057 	int			x, y;
6058 	int			ddi_status = DDI_SUCCESS;
6059 	nxge_status_t		status = NXGE_OK;
6060 
6061 	NXGE_DEBUG_MSG((nxgep, INT_CTL, "==> nxge_add_intrs_adv_type"));
6062 	intrp = (p_nxge_intr_t)&nxgep->nxge_intr_type;
6063 	intrp->start_inum = 0;
6064 
6065 	ddi_status = ddi_intr_get_nintrs(dip, int_type, &nintrs);
6066 	if ((ddi_status != DDI_SUCCESS) || (nintrs == 0)) {
6067 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
6068 		    "ddi_intr_get_nintrs() failed, status: 0x%x%, "
6069 		    "nintrs: %d", ddi_status, nintrs));
6070 		return (NXGE_ERROR | NXGE_DDI_FAILED);
6071 	}
6072 
6073 	ddi_status = ddi_intr_get_navail(dip, int_type, &navail);
6074 	if ((ddi_status != DDI_SUCCESS) || (navail == 0)) {
6075 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
6076 		    "ddi_intr_get_navail() failed, status: 0x%x%, "
6077 		    "nintrs: %d", ddi_status, navail));
6078 		return (NXGE_ERROR | NXGE_DDI_FAILED);
6079 	}
6080 
6081 	NXGE_DEBUG_MSG((nxgep, INT_CTL,
6082 	    "ddi_intr_get_navail() returned: nintrs %d, navail %d",
6083 	    nintrs, navail));
6084 
6085 	/* PSARC/2007/453 MSI-X interrupt limit override */
6086 	if (int_type == DDI_INTR_TYPE_MSIX) {
6087 		nrequest = nxge_create_msi_property(nxgep);
6088 		if (nrequest < navail) {
6089 			navail = nrequest;
6090 			NXGE_DEBUG_MSG((nxgep, INT_CTL,
6091 			    "nxge_add_intrs_adv_type: nintrs %d "
6092 			    "navail %d (nrequest %d)",
6093 			    nintrs, navail, nrequest));
6094 		}
6095 	}
6096 
6097 	if (int_type == DDI_INTR_TYPE_MSI && !ISP2(navail)) {
6098 		/* MSI must be power of 2 */
6099 		if ((navail & 16) == 16) {
6100 			navail = 16;
6101 		} else if ((navail & 8) == 8) {
6102 			navail = 8;
6103 		} else if ((navail & 4) == 4) {
6104 			navail = 4;
6105 		} else if ((navail & 2) == 2) {
6106 			navail = 2;
6107 		} else {
6108 			navail = 1;
6109 		}
6110 		NXGE_DEBUG_MSG((nxgep, INT_CTL,
6111 		    "ddi_intr_get_navail(): (msi power of 2) nintrs %d, "
6112 		    "navail %d", nintrs, navail));
6113 	}
6114 
6115 	behavior = ((int_type == DDI_INTR_TYPE_FIXED) ? DDI_INTR_ALLOC_STRICT :
6116 	    DDI_INTR_ALLOC_NORMAL);
6117 	intrp->intr_size = navail * sizeof (ddi_intr_handle_t);
6118 	intrp->htable = kmem_alloc(intrp->intr_size, KM_SLEEP);
6119 	ddi_status = ddi_intr_alloc(dip, intrp->htable, int_type, inum,
6120 	    navail, &nactual, behavior);
6121 	if (ddi_status != DDI_SUCCESS || nactual == 0) {
6122 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
6123 		    " ddi_intr_alloc() failed: %d",
6124 		    ddi_status));
6125 		kmem_free(intrp->htable, intrp->intr_size);
6126 		return (NXGE_ERROR | NXGE_DDI_FAILED);
6127 	}
6128 
6129 	if ((ddi_status = ddi_intr_get_pri(intrp->htable[0],
6130 	    (uint_t *)&intrp->pri)) != DDI_SUCCESS) {
6131 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
6132 		    " ddi_intr_get_pri() failed: %d",
6133 		    ddi_status));
6134 		/* Free already allocated interrupts */
6135 		for (y = 0; y < nactual; y++) {
6136 			(void) ddi_intr_free(intrp->htable[y]);
6137 		}
6138 
6139 		kmem_free(intrp->htable, intrp->intr_size);
6140 		return (NXGE_ERROR | NXGE_DDI_FAILED);
6141 	}
6142 
6143 	nrequired = 0;
6144 	switch (nxgep->niu_type) {
6145 	default:
6146 		status = nxge_ldgv_init(nxgep, &nactual, &nrequired);
6147 		break;
6148 
6149 	case N2_NIU:
6150 		status = nxge_ldgv_init_n2(nxgep, &nactual, &nrequired);
6151 		break;
6152 	}
6153 
6154 	if (status != NXGE_OK) {
6155 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
6156 		    "nxge_add_intrs_adv_typ:nxge_ldgv_init "
6157 		    "failed: 0x%x", status));
6158 		/* Free already allocated interrupts */
6159 		for (y = 0; y < nactual; y++) {
6160 			(void) ddi_intr_free(intrp->htable[y]);
6161 		}
6162 
6163 		kmem_free(intrp->htable, intrp->intr_size);
6164 		return (status);
6165 	}
6166 
6167 	ldgp = nxgep->ldgvp->ldgp;
6168 	for (x = 0; x < nrequired; x++, ldgp++) {
6169 		ldgp->vector = (uint8_t)x;
6170 		ldgp->intdata = SID_DATA(ldgp->func, x);
6171 		arg1 = ldgp->ldvp;
6172 		arg2 = nxgep;
6173 		if (ldgp->nldvs == 1) {
6174 			inthandler = (uint_t *)ldgp->ldvp->ldv_intr_handler;
6175 			NXGE_DEBUG_MSG((nxgep, INT_CTL,
6176 			    "nxge_add_intrs_adv_type: "
6177 			    "arg1 0x%x arg2 0x%x: "
6178 			    "1-1 int handler (entry %d intdata 0x%x)\n",
6179 			    arg1, arg2,
6180 			    x, ldgp->intdata));
6181 		} else if (ldgp->nldvs > 1) {
6182 			inthandler = (uint_t *)ldgp->sys_intr_handler;
6183 			NXGE_DEBUG_MSG((nxgep, INT_CTL,
6184 			    "nxge_add_intrs_adv_type: "
6185 			    "arg1 0x%x arg2 0x%x: "
6186 			    "nldevs %d int handler "
6187 			    "(entry %d intdata 0x%x)\n",
6188 			    arg1, arg2,
6189 			    ldgp->nldvs, x, ldgp->intdata));
6190 		}
6191 
6192 		NXGE_DEBUG_MSG((nxgep, INT_CTL,
6193 		    "==> nxge_add_intrs_adv_type: ddi_add_intr(inum) #%d "
6194 		    "htable 0x%llx", x, intrp->htable[x]));
6195 
6196 		if ((ddi_status = ddi_intr_add_handler(intrp->htable[x],
6197 		    (ddi_intr_handler_t *)inthandler, arg1, arg2))
6198 		    != DDI_SUCCESS) {
6199 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
6200 			    "==> nxge_add_intrs_adv_type: failed #%d "
6201 			    "status 0x%x", x, ddi_status));
6202 			for (y = 0; y < intrp->intr_added; y++) {
6203 				(void) ddi_intr_remove_handler(
6204 				    intrp->htable[y]);
6205 			}
6206 			/* Free already allocated intr */
6207 			for (y = 0; y < nactual; y++) {
6208 				(void) ddi_intr_free(intrp->htable[y]);
6209 			}
6210 			kmem_free(intrp->htable, intrp->intr_size);
6211 
6212 			(void) nxge_ldgv_uninit(nxgep);
6213 
6214 			return (NXGE_ERROR | NXGE_DDI_FAILED);
6215 		}
6216 
6217 		ldgp->htable_idx = x;
6218 		intrp->intr_added++;
6219 	}
6220 
6221 	intrp->msi_intx_cnt = nactual;
6222 
6223 	NXGE_DEBUG_MSG((nxgep, DDI_CTL,
6224 	    "Requested: %d, Allowed: %d msi_intx_cnt %d intr_added %d",
6225 	    navail, nactual,
6226 	    intrp->msi_intx_cnt,
6227 	    intrp->intr_added));
6228 
6229 	(void) ddi_intr_get_cap(intrp->htable[0], &intrp->intr_cap);
6230 
6231 	(void) nxge_intr_ldgv_init(nxgep);
6232 
6233 	NXGE_DEBUG_MSG((nxgep, INT_CTL, "<== nxge_add_intrs_adv_type"));
6234 
6235 	return (status);
6236 }
6237 
6238 /*ARGSUSED*/
6239 static nxge_status_t
6240 nxge_add_intrs_adv_type_fix(p_nxge_t nxgep, uint32_t int_type)
6241 {
6242 	dev_info_t		*dip = nxgep->dip;
6243 	p_nxge_ldg_t		ldgp;
6244 	p_nxge_intr_t		intrp;
6245 	uint_t			*inthandler;
6246 	void			*arg1, *arg2;
6247 	int			behavior;
6248 	int			nintrs, navail;
6249 	int			nactual, nrequired;
6250 	int			inum = 0;
6251 	int			x, y;
6252 	int			ddi_status = DDI_SUCCESS;
6253 	nxge_status_t		status = NXGE_OK;
6254 
6255 	NXGE_DEBUG_MSG((nxgep, INT_CTL, "==> nxge_add_intrs_adv_type_fix"));
6256 	intrp = (p_nxge_intr_t)&nxgep->nxge_intr_type;
6257 	intrp->start_inum = 0;
6258 
6259 	ddi_status = ddi_intr_get_nintrs(dip, int_type, &nintrs);
6260 	if ((ddi_status != DDI_SUCCESS) || (nintrs == 0)) {
6261 		NXGE_DEBUG_MSG((nxgep, INT_CTL,
6262 		    "ddi_intr_get_nintrs() failed, status: 0x%x%, "
6263 		    "nintrs: %d", status, nintrs));
6264 		return (NXGE_ERROR | NXGE_DDI_FAILED);
6265 	}
6266 
6267 	ddi_status = ddi_intr_get_navail(dip, int_type, &navail);
6268 	if ((ddi_status != DDI_SUCCESS) || (navail == 0)) {
6269 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
6270 		    "ddi_intr_get_navail() failed, status: 0x%x%, "
6271 		    "nintrs: %d", ddi_status, navail));
6272 		return (NXGE_ERROR | NXGE_DDI_FAILED);
6273 	}
6274 
6275 	NXGE_DEBUG_MSG((nxgep, INT_CTL,
6276 	    "ddi_intr_get_navail() returned: nintrs %d, naavail %d",
6277 	    nintrs, navail));
6278 
6279 	behavior = ((int_type == DDI_INTR_TYPE_FIXED) ? DDI_INTR_ALLOC_STRICT :
6280 	    DDI_INTR_ALLOC_NORMAL);
6281 	intrp->intr_size = navail * sizeof (ddi_intr_handle_t);
6282 	intrp->htable = kmem_alloc(intrp->intr_size, KM_SLEEP);
6283 	ddi_status = ddi_intr_alloc(dip, intrp->htable, int_type, inum,
6284 	    navail, &nactual, behavior);
6285 	if (ddi_status != DDI_SUCCESS || nactual == 0) {
6286 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
6287 		    " ddi_intr_alloc() failed: %d",
6288 		    ddi_status));
6289 		kmem_free(intrp->htable, intrp->intr_size);
6290 		return (NXGE_ERROR | NXGE_DDI_FAILED);
6291 	}
6292 
6293 	if ((ddi_status = ddi_intr_get_pri(intrp->htable[0],
6294 	    (uint_t *)&intrp->pri)) != DDI_SUCCESS) {
6295 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
6296 		    " ddi_intr_get_pri() failed: %d",
6297 		    ddi_status));
6298 		/* Free already allocated interrupts */
6299 		for (y = 0; y < nactual; y++) {
6300 			(void) ddi_intr_free(intrp->htable[y]);
6301 		}
6302 
6303 		kmem_free(intrp->htable, intrp->intr_size);
6304 		return (NXGE_ERROR | NXGE_DDI_FAILED);
6305 	}
6306 
6307 	nrequired = 0;
6308 	switch (nxgep->niu_type) {
6309 	default:
6310 		status = nxge_ldgv_init(nxgep, &nactual, &nrequired);
6311 		break;
6312 
6313 	case N2_NIU:
6314 		status = nxge_ldgv_init_n2(nxgep, &nactual, &nrequired);
6315 		break;
6316 	}
6317 
6318 	if (status != NXGE_OK) {
6319 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
6320 		    "nxge_add_intrs_adv_type_fix:nxge_ldgv_init "
6321 		    "failed: 0x%x", status));
6322 		/* Free already allocated interrupts */
6323 		for (y = 0; y < nactual; y++) {
6324 			(void) ddi_intr_free(intrp->htable[y]);
6325 		}
6326 
6327 		kmem_free(intrp->htable, intrp->intr_size);
6328 		return (status);
6329 	}
6330 
6331 	ldgp = nxgep->ldgvp->ldgp;
6332 	for (x = 0; x < nrequired; x++, ldgp++) {
6333 		ldgp->vector = (uint8_t)x;
6334 		if (nxgep->niu_type != N2_NIU) {
6335 			ldgp->intdata = SID_DATA(ldgp->func, x);
6336 		}
6337 
6338 		arg1 = ldgp->ldvp;
6339 		arg2 = nxgep;
6340 		if (ldgp->nldvs == 1) {
6341 			inthandler = (uint_t *)ldgp->ldvp->ldv_intr_handler;
6342 			NXGE_DEBUG_MSG((nxgep, INT_CTL,
6343 			    "nxge_add_intrs_adv_type_fix: "
6344 			    "1-1 int handler(%d) ldg %d ldv %d "
6345 			    "arg1 $%p arg2 $%p\n",
6346 			    x, ldgp->ldg, ldgp->ldvp->ldv,
6347 			    arg1, arg2));
6348 		} else if (ldgp->nldvs > 1) {
6349 			inthandler = (uint_t *)ldgp->sys_intr_handler;
6350 			NXGE_DEBUG_MSG((nxgep, INT_CTL,
6351 			    "nxge_add_intrs_adv_type_fix: "
6352 			    "shared ldv %d int handler(%d) ldv %d ldg %d"
6353 			    "arg1 0x%016llx arg2 0x%016llx\n",
6354 			    x, ldgp->nldvs, ldgp->ldg, ldgp->ldvp->ldv,
6355 			    arg1, arg2));
6356 		}
6357 
6358 		if ((ddi_status = ddi_intr_add_handler(intrp->htable[x],
6359 		    (ddi_intr_handler_t *)inthandler, arg1, arg2))
6360 		    != DDI_SUCCESS) {
6361 			NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
6362 			    "==> nxge_add_intrs_adv_type_fix: failed #%d "
6363 			    "status 0x%x", x, ddi_status));
6364 			for (y = 0; y < intrp->intr_added; y++) {
6365 				(void) ddi_intr_remove_handler(
6366 				    intrp->htable[y]);
6367 			}
6368 			for (y = 0; y < nactual; y++) {
6369 				(void) ddi_intr_free(intrp->htable[y]);
6370 			}
6371 			/* Free already allocated intr */
6372 			kmem_free(intrp->htable, intrp->intr_size);
6373 
6374 			(void) nxge_ldgv_uninit(nxgep);
6375 
6376 			return (NXGE_ERROR | NXGE_DDI_FAILED);
6377 		}
6378 
6379 		ldgp->htable_idx = x;
6380 		intrp->intr_added++;
6381 	}
6382 
6383 	intrp->msi_intx_cnt = nactual;
6384 
6385 	(void) ddi_intr_get_cap(intrp->htable[0], &intrp->intr_cap);
6386 
6387 	status = nxge_intr_ldgv_init(nxgep);
6388 	NXGE_DEBUG_MSG((nxgep, INT_CTL, "<== nxge_add_intrs_adv_type_fix"));
6389 
6390 	return (status);
6391 }
6392 
6393 static void
6394 nxge_remove_intrs(p_nxge_t nxgep)
6395 {
6396 	int		i, inum;
6397 	p_nxge_intr_t	intrp;
6398 
6399 	NXGE_DEBUG_MSG((nxgep, INT_CTL, "==> nxge_remove_intrs"));
6400 	intrp = (p_nxge_intr_t)&nxgep->nxge_intr_type;
6401 	if (!intrp->intr_registered) {
6402 		NXGE_DEBUG_MSG((nxgep, INT_CTL,
6403 		    "<== nxge_remove_intrs: interrupts not registered"));
6404 		return;
6405 	}
6406 
6407 	NXGE_DEBUG_MSG((nxgep, INT_CTL, "==> nxge_remove_intrs:advanced"));
6408 
6409 	if (intrp->intr_cap & DDI_INTR_FLAG_BLOCK) {
6410 		(void) ddi_intr_block_disable(intrp->htable,
6411 		    intrp->intr_added);
6412 	} else {
6413 		for (i = 0; i < intrp->intr_added; i++) {
6414 			(void) ddi_intr_disable(intrp->htable[i]);
6415 		}
6416 	}
6417 
6418 	for (inum = 0; inum < intrp->intr_added; inum++) {
6419 		if (intrp->htable[inum]) {
6420 			(void) ddi_intr_remove_handler(intrp->htable[inum]);
6421 		}
6422 	}
6423 
6424 	for (inum = 0; inum < intrp->msi_intx_cnt; inum++) {
6425 		if (intrp->htable[inum]) {
6426 			NXGE_DEBUG_MSG((nxgep, DDI_CTL,
6427 			    "nxge_remove_intrs: ddi_intr_free inum %d "
6428 			    "msi_intx_cnt %d intr_added %d",
6429 			    inum,
6430 			    intrp->msi_intx_cnt,
6431 			    intrp->intr_added));
6432 
6433 			(void) ddi_intr_free(intrp->htable[inum]);
6434 		}
6435 	}
6436 
6437 	kmem_free(intrp->htable, intrp->intr_size);
6438 	intrp->intr_registered = B_FALSE;
6439 	intrp->intr_enabled = B_FALSE;
6440 	intrp->msi_intx_cnt = 0;
6441 	intrp->intr_added = 0;
6442 
6443 	(void) nxge_ldgv_uninit(nxgep);
6444 
6445 	(void) ddi_prop_remove(DDI_DEV_T_NONE, nxgep->dip,
6446 	    "#msix-request");
6447 
6448 	NXGE_DEBUG_MSG((nxgep, INT_CTL, "<== nxge_remove_intrs"));
6449 }
6450 
6451 /*ARGSUSED*/
6452 static void
6453 nxge_intrs_enable(p_nxge_t nxgep)
6454 {
6455 	p_nxge_intr_t	intrp;
6456 	int		i;
6457 	int		status;
6458 
6459 	NXGE_DEBUG_MSG((nxgep, INT_CTL, "==> nxge_intrs_enable"));
6460 
6461 	intrp = (p_nxge_intr_t)&nxgep->nxge_intr_type;
6462 
6463 	if (!intrp->intr_registered) {
6464 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL, "<== nxge_intrs_enable: "
6465 		    "interrupts are not registered"));
6466 		return;
6467 	}
6468 
6469 	if (intrp->intr_enabled) {
6470 		NXGE_DEBUG_MSG((nxgep, INT_CTL,
6471 		    "<== nxge_intrs_enable: already enabled"));
6472 		return;
6473 	}
6474 
6475 	if (intrp->intr_cap & DDI_INTR_FLAG_BLOCK) {
6476 		status = ddi_intr_block_enable(intrp->htable,
6477 		    intrp->intr_added);
6478 		NXGE_DEBUG_MSG((nxgep, INT_CTL, "==> nxge_intrs_enable "
6479 		    "block enable - status 0x%x total inums #%d\n",
6480 		    status, intrp->intr_added));
6481 	} else {
6482 		for (i = 0; i < intrp->intr_added; i++) {
6483 			status = ddi_intr_enable(intrp->htable[i]);
6484 			NXGE_DEBUG_MSG((nxgep, INT_CTL, "==> nxge_intrs_enable "
6485 			    "ddi_intr_enable:enable - status 0x%x "
6486 			    "total inums %d enable inum #%d\n",
6487 			    status, intrp->intr_added, i));
6488 			if (status == DDI_SUCCESS) {
6489 				intrp->intr_enabled = B_TRUE;
6490 			}
6491 		}
6492 	}
6493 
6494 	NXGE_DEBUG_MSG((nxgep, INT_CTL, "<== nxge_intrs_enable"));
6495 }
6496 
6497 /*ARGSUSED*/
6498 static void
6499 nxge_intrs_disable(p_nxge_t nxgep)
6500 {
6501 	p_nxge_intr_t	intrp;
6502 	int		i;
6503 
6504 	NXGE_DEBUG_MSG((nxgep, INT_CTL, "==> nxge_intrs_disable"));
6505 
6506 	intrp = (p_nxge_intr_t)&nxgep->nxge_intr_type;
6507 
6508 	if (!intrp->intr_registered) {
6509 		NXGE_DEBUG_MSG((nxgep, INT_CTL, "<== nxge_intrs_disable: "
6510 		    "interrupts are not registered"));
6511 		return;
6512 	}
6513 
6514 	if (intrp->intr_cap & DDI_INTR_FLAG_BLOCK) {
6515 		(void) ddi_intr_block_disable(intrp->htable,
6516 		    intrp->intr_added);
6517 	} else {
6518 		for (i = 0; i < intrp->intr_added; i++) {
6519 			(void) ddi_intr_disable(intrp->htable[i]);
6520 		}
6521 	}
6522 
6523 	intrp->intr_enabled = B_FALSE;
6524 	NXGE_DEBUG_MSG((nxgep, INT_CTL, "<== nxge_intrs_disable"));
6525 }
6526 
6527 nxge_status_t
6528 nxge_mac_register(p_nxge_t nxgep)
6529 {
6530 	mac_register_t *macp;
6531 	int		status;
6532 
6533 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "==> nxge_mac_register"));
6534 
6535 	if ((macp = mac_alloc(MAC_VERSION)) == NULL)
6536 		return (NXGE_ERROR);
6537 
6538 	macp->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
6539 	macp->m_driver = nxgep;
6540 	macp->m_dip = nxgep->dip;
6541 	if (!isLDOMguest(nxgep)) {
6542 		macp->m_src_addr = nxgep->ouraddr.ether_addr_octet;
6543 	} else {
6544 		macp->m_src_addr = KMEM_ZALLOC(MAXMACADDRLEN, KM_SLEEP);
6545 		macp->m_dst_addr = KMEM_ZALLOC(MAXMACADDRLEN, KM_SLEEP);
6546 		(void) memset(macp->m_src_addr, 0xff, sizeof (MAXMACADDRLEN));
6547 	}
6548 	macp->m_callbacks = &nxge_m_callbacks;
6549 	macp->m_min_sdu = 0;
6550 	nxgep->mac.default_mtu = nxgep->mac.maxframesize -
6551 	    NXGE_EHEADER_VLAN_CRC;
6552 	macp->m_max_sdu = nxgep->mac.default_mtu;
6553 	macp->m_margin = VLAN_TAGSZ;
6554 	macp->m_priv_props = nxge_priv_props;
6555 	if (isLDOMguest(nxgep))
6556 		macp->m_v12n = MAC_VIRT_LEVEL1;
6557 	else
6558 		macp->m_v12n = MAC_VIRT_HIO | MAC_VIRT_LEVEL1;
6559 
6560 	NXGE_DEBUG_MSG((nxgep, MAC_CTL,
6561 	    "==> nxge_mac_register: instance %d "
6562 	    "max_sdu %d margin %d maxframe %d (header %d)",
6563 	    nxgep->instance,
6564 	    macp->m_max_sdu, macp->m_margin,
6565 	    nxgep->mac.maxframesize,
6566 	    NXGE_EHEADER_VLAN_CRC));
6567 
6568 	status = mac_register(macp, &nxgep->mach);
6569 	if (isLDOMguest(nxgep)) {
6570 		KMEM_FREE(macp->m_src_addr, MAXMACADDRLEN);
6571 		KMEM_FREE(macp->m_dst_addr, MAXMACADDRLEN);
6572 	}
6573 	mac_free(macp);
6574 
6575 	if (status != 0) {
6576 		cmn_err(CE_WARN,
6577 		    "!nxge_mac_register failed (status %d instance %d)",
6578 		    status, nxgep->instance);
6579 		return (NXGE_ERROR);
6580 	}
6581 
6582 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "<== nxge_mac_register success "
6583 	    "(instance %d)", nxgep->instance));
6584 
6585 	return (NXGE_OK);
6586 }
6587 
6588 void
6589 nxge_err_inject(p_nxge_t nxgep, queue_t *wq, mblk_t *mp)
6590 {
6591 	ssize_t		size;
6592 	mblk_t		*nmp;
6593 	uint8_t		blk_id;
6594 	uint8_t		chan;
6595 	uint32_t	err_id;
6596 	err_inject_t	*eip;
6597 
6598 	NXGE_DEBUG_MSG((nxgep, STR_CTL, "==> nxge_err_inject"));
6599 
6600 	size = 1024;
6601 	nmp = mp->b_cont;
6602 	eip = (err_inject_t *)nmp->b_rptr;
6603 	blk_id = eip->blk_id;
6604 	err_id = eip->err_id;
6605 	chan = eip->chan;
6606 	cmn_err(CE_NOTE, "!blk_id = 0x%x\n", blk_id);
6607 	cmn_err(CE_NOTE, "!err_id = 0x%x\n", err_id);
6608 	cmn_err(CE_NOTE, "!chan = 0x%x\n", chan);
6609 	switch (blk_id) {
6610 	case MAC_BLK_ID:
6611 		break;
6612 	case TXMAC_BLK_ID:
6613 		break;
6614 	case RXMAC_BLK_ID:
6615 		break;
6616 	case MIF_BLK_ID:
6617 		break;
6618 	case IPP_BLK_ID:
6619 		nxge_ipp_inject_err(nxgep, err_id);
6620 		break;
6621 	case TXC_BLK_ID:
6622 		nxge_txc_inject_err(nxgep, err_id);
6623 		break;
6624 	case TXDMA_BLK_ID:
6625 		nxge_txdma_inject_err(nxgep, err_id, chan);
6626 		break;
6627 	case RXDMA_BLK_ID:
6628 		nxge_rxdma_inject_err(nxgep, err_id, chan);
6629 		break;
6630 	case ZCP_BLK_ID:
6631 		nxge_zcp_inject_err(nxgep, err_id);
6632 		break;
6633 	case ESPC_BLK_ID:
6634 		break;
6635 	case FFLP_BLK_ID:
6636 		break;
6637 	case PHY_BLK_ID:
6638 		break;
6639 	case ETHER_SERDES_BLK_ID:
6640 		break;
6641 	case PCIE_SERDES_BLK_ID:
6642 		break;
6643 	case VIR_BLK_ID:
6644 		break;
6645 	}
6646 
6647 	nmp->b_wptr = nmp->b_rptr + size;
6648 	NXGE_DEBUG_MSG((nxgep, STR_CTL, "<== nxge_err_inject"));
6649 
6650 	miocack(wq, mp, (int)size, 0);
6651 }
6652 
6653 static int
6654 nxge_init_common_dev(p_nxge_t nxgep)
6655 {
6656 	p_nxge_hw_list_t	hw_p;
6657 	dev_info_t 		*p_dip;
6658 
6659 	ASSERT(nxgep != NULL);
6660 
6661 	NXGE_DEBUG_MSG((nxgep, MOD_CTL, "==> nxge_init_common_device"));
6662 
6663 	p_dip = nxgep->p_dip;
6664 	MUTEX_ENTER(&nxge_common_lock);
6665 	NXGE_DEBUG_MSG((nxgep, MOD_CTL,
6666 	    "==> nxge_init_common_dev:func # %d",
6667 	    nxgep->function_num));
6668 	/*
6669 	 * Loop through existing per neptune hardware list.
6670 	 */
6671 	for (hw_p = nxge_hw_list; hw_p; hw_p = hw_p->next) {
6672 		NXGE_DEBUG_MSG((nxgep, MOD_CTL,
6673 		    "==> nxge_init_common_device:func # %d "
6674 		    "hw_p $%p parent dip $%p",
6675 		    nxgep->function_num,
6676 		    hw_p,
6677 		    p_dip));
6678 		if (hw_p->parent_devp == p_dip) {
6679 			nxgep->nxge_hw_p = hw_p;
6680 			hw_p->ndevs++;
6681 			hw_p->nxge_p[nxgep->function_num] = nxgep;
6682 			NXGE_DEBUG_MSG((nxgep, MOD_CTL,
6683 			    "==> nxge_init_common_device:func # %d "
6684 			    "hw_p $%p parent dip $%p "
6685 			    "ndevs %d (found)",
6686 			    nxgep->function_num,
6687 			    hw_p,
6688 			    p_dip,
6689 			    hw_p->ndevs));
6690 			break;
6691 		}
6692 	}
6693 
6694 	if (hw_p == NULL) {
6695 
6696 		char **prop_val;
6697 		uint_t prop_len;
6698 		int i;
6699 
6700 		NXGE_DEBUG_MSG((nxgep, MOD_CTL,
6701 		    "==> nxge_init_common_device:func # %d "
6702 		    "parent dip $%p (new)",
6703 		    nxgep->function_num,
6704 		    p_dip));
6705 		hw_p = kmem_zalloc(sizeof (nxge_hw_list_t), KM_SLEEP);
6706 		hw_p->parent_devp = p_dip;
6707 		hw_p->magic = NXGE_NEPTUNE_MAGIC;
6708 		nxgep->nxge_hw_p = hw_p;
6709 		hw_p->ndevs++;
6710 		hw_p->nxge_p[nxgep->function_num] = nxgep;
6711 		hw_p->next = nxge_hw_list;
6712 		if (nxgep->niu_type == N2_NIU) {
6713 			hw_p->niu_type = N2_NIU;
6714 			hw_p->platform_type = P_NEPTUNE_NIU;
6715 			hw_p->tcam_size = TCAM_NIU_TCAM_MAX_ENTRY;
6716 		} else {
6717 			hw_p->niu_type = NIU_TYPE_NONE;
6718 			hw_p->platform_type = P_NEPTUNE_NONE;
6719 			hw_p->tcam_size = TCAM_NXGE_TCAM_MAX_ENTRY;
6720 		}
6721 
6722 		hw_p->tcam = KMEM_ZALLOC(sizeof (tcam_flow_spec_t) *
6723 		    hw_p->tcam_size, KM_SLEEP);
6724 
6725 		MUTEX_INIT(&hw_p->nxge_cfg_lock, NULL, MUTEX_DRIVER, NULL);
6726 		MUTEX_INIT(&hw_p->nxge_tcam_lock, NULL, MUTEX_DRIVER, NULL);
6727 		MUTEX_INIT(&hw_p->nxge_vlan_lock, NULL, MUTEX_DRIVER, NULL);
6728 		MUTEX_INIT(&hw_p->nxge_mdio_lock, NULL, MUTEX_DRIVER, NULL);
6729 
6730 		nxge_hw_list = hw_p;
6731 
6732 		if (ddi_prop_lookup_string_array(DDI_DEV_T_ANY, nxgep->dip, 0,
6733 		    "compatible", &prop_val, &prop_len) == DDI_PROP_SUCCESS) {
6734 			for (i = 0; i < prop_len; i++) {
6735 				if ((strcmp((caddr_t)prop_val[i],
6736 				    NXGE_ROCK_COMPATIBLE) == 0)) {
6737 					hw_p->platform_type = P_NEPTUNE_ROCK;
6738 					NXGE_DEBUG_MSG((nxgep, MOD_CTL,
6739 					    "ROCK hw_p->platform_type %d",
6740 					    hw_p->platform_type));
6741 					break;
6742 				}
6743 				NXGE_DEBUG_MSG((nxgep, MOD_CTL,
6744 				    "nxge_init_common_dev: read compatible"
6745 				    " property[%d] val[%s]",
6746 				    i, (caddr_t)prop_val[i]));
6747 			}
6748 		}
6749 
6750 		ddi_prop_free(prop_val);
6751 
6752 		(void) nxge_scan_ports_phy(nxgep, nxge_hw_list);
6753 	}
6754 
6755 	MUTEX_EXIT(&nxge_common_lock);
6756 
6757 	nxgep->platform_type = hw_p->platform_type;
6758 	NXGE_DEBUG_MSG((nxgep, MOD_CTL, "nxgep->platform_type %d",
6759 	    nxgep->platform_type));
6760 	if (nxgep->niu_type != N2_NIU) {
6761 		nxgep->niu_type = hw_p->niu_type;
6762 	}
6763 
6764 	NXGE_DEBUG_MSG((nxgep, MOD_CTL,
6765 	    "==> nxge_init_common_device (nxge_hw_list) $%p",
6766 	    nxge_hw_list));
6767 	NXGE_DEBUG_MSG((nxgep, MOD_CTL, "<== nxge_init_common_device"));
6768 
6769 	return (NXGE_OK);
6770 }
6771 
6772 static void
6773 nxge_uninit_common_dev(p_nxge_t nxgep)
6774 {
6775 	p_nxge_hw_list_t	hw_p, h_hw_p;
6776 	p_nxge_dma_pt_cfg_t	p_dma_cfgp;
6777 	p_nxge_hw_pt_cfg_t	p_cfgp;
6778 	dev_info_t 		*p_dip;
6779 
6780 	ASSERT(nxgep != NULL);
6781 
6782 	NXGE_DEBUG_MSG((nxgep, MOD_CTL, "==> nxge_uninit_common_device"));
6783 	if (nxgep->nxge_hw_p == NULL) {
6784 		NXGE_DEBUG_MSG((nxgep, MOD_CTL,
6785 		    "<== nxge_uninit_common_device (no common)"));
6786 		return;
6787 	}
6788 
6789 	MUTEX_ENTER(&nxge_common_lock);
6790 	h_hw_p = nxge_hw_list;
6791 	for (hw_p = nxge_hw_list; hw_p; hw_p = hw_p->next) {
6792 		p_dip = hw_p->parent_devp;
6793 		if (nxgep->nxge_hw_p == hw_p &&
6794 		    p_dip == nxgep->p_dip &&
6795 		    nxgep->nxge_hw_p->magic == NXGE_NEPTUNE_MAGIC &&
6796 		    hw_p->magic == NXGE_NEPTUNE_MAGIC) {
6797 
6798 			NXGE_DEBUG_MSG((nxgep, MOD_CTL,
6799 			    "==> nxge_uninit_common_device:func # %d "
6800 			    "hw_p $%p parent dip $%p "
6801 			    "ndevs %d (found)",
6802 			    nxgep->function_num,
6803 			    hw_p,
6804 			    p_dip,
6805 			    hw_p->ndevs));
6806 
6807 			/*
6808 			 * Release the RDC table, a shared resoruce
6809 			 * of the nxge hardware.  The RDC table was
6810 			 * assigned to this instance of nxge in
6811 			 * nxge_use_cfg_dma_config().
6812 			 */
6813 			if (!isLDOMguest(nxgep)) {
6814 				p_dma_cfgp =
6815 				    (p_nxge_dma_pt_cfg_t)&nxgep->pt_config;
6816 				p_cfgp =
6817 				    (p_nxge_hw_pt_cfg_t)&p_dma_cfgp->hw_config;
6818 				(void) nxge_fzc_rdc_tbl_unbind(nxgep,
6819 				    p_cfgp->def_mac_rxdma_grpid);
6820 
6821 				/* Cleanup any outstanding groups.  */
6822 				nxge_grp_cleanup(nxgep);
6823 			}
6824 
6825 			if (hw_p->ndevs) {
6826 				hw_p->ndevs--;
6827 			}
6828 			hw_p->nxge_p[nxgep->function_num] = NULL;
6829 			if (!hw_p->ndevs) {
6830 				KMEM_FREE(hw_p->tcam,
6831 				    sizeof (tcam_flow_spec_t) *
6832 				    hw_p->tcam_size);
6833 				MUTEX_DESTROY(&hw_p->nxge_vlan_lock);
6834 				MUTEX_DESTROY(&hw_p->nxge_tcam_lock);
6835 				MUTEX_DESTROY(&hw_p->nxge_cfg_lock);
6836 				MUTEX_DESTROY(&hw_p->nxge_mdio_lock);
6837 				NXGE_DEBUG_MSG((nxgep, MOD_CTL,
6838 				    "==> nxge_uninit_common_device: "
6839 				    "func # %d "
6840 				    "hw_p $%p parent dip $%p "
6841 				    "ndevs %d (last)",
6842 				    nxgep->function_num,
6843 				    hw_p,
6844 				    p_dip,
6845 				    hw_p->ndevs));
6846 
6847 				nxge_hio_uninit(nxgep);
6848 
6849 				if (hw_p == nxge_hw_list) {
6850 					NXGE_DEBUG_MSG((nxgep, MOD_CTL,
6851 					    "==> nxge_uninit_common_device:"
6852 					    "remove head func # %d "
6853 					    "hw_p $%p parent dip $%p "
6854 					    "ndevs %d (head)",
6855 					    nxgep->function_num,
6856 					    hw_p,
6857 					    p_dip,
6858 					    hw_p->ndevs));
6859 					nxge_hw_list = hw_p->next;
6860 				} else {
6861 					NXGE_DEBUG_MSG((nxgep, MOD_CTL,
6862 					    "==> nxge_uninit_common_device:"
6863 					    "remove middle func # %d "
6864 					    "hw_p $%p parent dip $%p "
6865 					    "ndevs %d (middle)",
6866 					    nxgep->function_num,
6867 					    hw_p,
6868 					    p_dip,
6869 					    hw_p->ndevs));
6870 					h_hw_p->next = hw_p->next;
6871 				}
6872 
6873 				nxgep->nxge_hw_p = NULL;
6874 				KMEM_FREE(hw_p, sizeof (nxge_hw_list_t));
6875 			}
6876 			break;
6877 		} else {
6878 			h_hw_p = hw_p;
6879 		}
6880 	}
6881 
6882 	MUTEX_EXIT(&nxge_common_lock);
6883 	NXGE_DEBUG_MSG((nxgep, MOD_CTL,
6884 	    "==> nxge_uninit_common_device (nxge_hw_list) $%p",
6885 	    nxge_hw_list));
6886 
6887 	NXGE_DEBUG_MSG((nxgep, MOD_CTL, "<= nxge_uninit_common_device"));
6888 }
6889 
6890 /*
6891  * Determines the number of ports from the niu_type or the platform type.
6892  * Returns the number of ports, or returns zero on failure.
6893  */
6894 
6895 int
6896 nxge_get_nports(p_nxge_t nxgep)
6897 {
6898 	int	nports = 0;
6899 
6900 	switch (nxgep->niu_type) {
6901 	case N2_NIU:
6902 	case NEPTUNE_2_10GF:
6903 		nports = 2;
6904 		break;
6905 	case NEPTUNE_4_1GC:
6906 	case NEPTUNE_2_10GF_2_1GC:
6907 	case NEPTUNE_1_10GF_3_1GC:
6908 	case NEPTUNE_1_1GC_1_10GF_2_1GC:
6909 	case NEPTUNE_2_10GF_2_1GRF:
6910 		nports = 4;
6911 		break;
6912 	default:
6913 		switch (nxgep->platform_type) {
6914 		case P_NEPTUNE_NIU:
6915 		case P_NEPTUNE_ATLAS_2PORT:
6916 			nports = 2;
6917 			break;
6918 		case P_NEPTUNE_ATLAS_4PORT:
6919 		case P_NEPTUNE_MARAMBA_P0:
6920 		case P_NEPTUNE_MARAMBA_P1:
6921 		case P_NEPTUNE_ROCK:
6922 		case P_NEPTUNE_ALONSO:
6923 			nports = 4;
6924 			break;
6925 		default:
6926 			break;
6927 		}
6928 		break;
6929 	}
6930 
6931 	return (nports);
6932 }
6933 
6934 /*
6935  * The following two functions are to support
6936  * PSARC/2007/453 MSI-X interrupt limit override.
6937  */
6938 static int
6939 nxge_create_msi_property(p_nxge_t nxgep)
6940 {
6941 	int	nmsi;
6942 	extern	int ncpus;
6943 
6944 	NXGE_DEBUG_MSG((nxgep, MOD_CTL, "==>nxge_create_msi_property"));
6945 
6946 	switch (nxgep->mac.portmode) {
6947 	case PORT_10G_COPPER:
6948 	case PORT_10G_FIBER:
6949 	case PORT_10G_TN1010:
6950 		(void) ddi_prop_create(DDI_DEV_T_NONE, nxgep->dip,
6951 		    DDI_PROP_CANSLEEP, "#msix-request", NULL, 0);
6952 		/*
6953 		 * The maximum MSI-X requested will be 8.
6954 		 * If the # of CPUs is less than 8, we will request
6955 		 * # MSI-X based on the # of CPUs (default).
6956 		 */
6957 		NXGE_DEBUG_MSG((nxgep, MOD_CTL,
6958 		    "==>nxge_create_msi_property (10G): nxge_msix_10g_intrs %d",
6959 		    nxge_msix_10g_intrs));
6960 		if ((nxge_msix_10g_intrs == 0) ||
6961 		    (nxge_msix_10g_intrs > NXGE_MSIX_MAX_ALLOWED)) {
6962 			nmsi = NXGE_MSIX_REQUEST_10G;
6963 			NXGE_DEBUG_MSG((nxgep, MOD_CTL,
6964 			    "==>nxge_create_msi_property (10G): reset to 8"));
6965 		} else {
6966 			nmsi = nxge_msix_10g_intrs;
6967 		}
6968 
6969 		/*
6970 		 * If # of interrupts requested is 8 (default),
6971 		 * the checking of the number of cpus will be
6972 		 * be maintained.
6973 		 */
6974 		if ((nmsi == NXGE_MSIX_REQUEST_10G) &&
6975 		    (ncpus < nmsi)) {
6976 			NXGE_DEBUG_MSG((nxgep, MOD_CTL,
6977 			    "==>nxge_create_msi_property (10G): reset to 8"));
6978 			nmsi = ncpus;
6979 		}
6980 		NXGE_DEBUG_MSG((nxgep, MOD_CTL,
6981 		    "==>nxge_create_msi_property(10G): exists 0x%x (nmsi %d)",
6982 		    ddi_prop_exists(DDI_DEV_T_NONE, nxgep->dip,
6983 		    DDI_PROP_CANSLEEP, "#msix-request"), nmsi));
6984 		break;
6985 
6986 	default:
6987 		(void) ddi_prop_create(DDI_DEV_T_NONE, nxgep->dip,
6988 		    DDI_PROP_CANSLEEP, "#msix-request", NULL, 0);
6989 		NXGE_DEBUG_MSG((nxgep, MOD_CTL,
6990 		    "==>nxge_create_msi_property (1G): nxge_msix_1g_intrs %d",
6991 		    nxge_msix_1g_intrs));
6992 		if ((nxge_msix_1g_intrs == 0) ||
6993 		    (nxge_msix_1g_intrs > NXGE_MSIX_MAX_ALLOWED)) {
6994 			nmsi = NXGE_MSIX_REQUEST_1G;
6995 			NXGE_DEBUG_MSG((nxgep, MOD_CTL,
6996 			    "==>nxge_create_msi_property (1G): reset to 2"));
6997 		} else {
6998 			nmsi = nxge_msix_1g_intrs;
6999 		}
7000 		NXGE_DEBUG_MSG((nxgep, MOD_CTL,
7001 		    "==>nxge_create_msi_property(1G): exists 0x%x (nmsi %d)",
7002 		    ddi_prop_exists(DDI_DEV_T_NONE, nxgep->dip,
7003 		    DDI_PROP_CANSLEEP, "#msix-request"), nmsi));
7004 		break;
7005 	}
7006 
7007 	NXGE_DEBUG_MSG((nxgep, MOD_CTL, "<==nxge_create_msi_property"));
7008 	return (nmsi);
7009 }
7010 
7011 /*
7012  * The following is a software around for the Neptune hardware's
7013  * interrupt bugs; The Neptune hardware may generate spurious interrupts when
7014  * an interrupr handler is removed.
7015  */
7016 #define	NXGE_PCI_PORT_LOGIC_OFFSET	0x98
7017 #define	NXGE_PIM_RESET			(1ULL << 29)
7018 #define	NXGE_GLU_RESET			(1ULL << 30)
7019 #define	NXGE_NIU_RESET			(1ULL << 31)
7020 #define	NXGE_PCI_RESET_ALL		(NXGE_PIM_RESET |	\
7021 					NXGE_GLU_RESET |	\
7022 					NXGE_NIU_RESET)
7023 
7024 #define	NXGE_WAIT_QUITE_TIME		200000
7025 #define	NXGE_WAIT_QUITE_RETRY		40
7026 #define	NXGE_PCI_RESET_WAIT		1000000 /* one second */
7027 
7028 static void
7029 nxge_niu_peu_reset(p_nxge_t nxgep)
7030 {
7031 	uint32_t	rvalue;
7032 	p_nxge_hw_list_t hw_p;
7033 	p_nxge_t	fnxgep;
7034 	int		i, j;
7035 
7036 	NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL, "==> nxge_niu_peu_reset"));
7037 	if ((hw_p = nxgep->nxge_hw_p) == NULL) {
7038 		NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
7039 		    "==> nxge_niu_peu_reset: NULL hardware pointer"));
7040 		return;
7041 	}
7042 
7043 	NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
7044 	    "==> nxge_niu_peu_reset: flags 0x%x link timer id %d timer id %d",
7045 	    hw_p->flags, nxgep->nxge_link_poll_timerid,
7046 	    nxgep->nxge_timerid));
7047 
7048 	MUTEX_ENTER(&hw_p->nxge_cfg_lock);
7049 	/*
7050 	 * Make sure other instances from the same hardware
7051 	 * stop sending PIO and in quiescent state.
7052 	 */
7053 	for (i = 0; i < NXGE_MAX_PORTS; i++) {
7054 		fnxgep = hw_p->nxge_p[i];
7055 		NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
7056 		    "==> nxge_niu_peu_reset: checking entry %d "
7057 		    "nxgep $%p", i, fnxgep));
7058 #ifdef	NXGE_DEBUG
7059 		if (fnxgep) {
7060 			NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
7061 			    "==> nxge_niu_peu_reset: entry %d (function %d) "
7062 			    "link timer id %d hw timer id %d",
7063 			    i, fnxgep->function_num,
7064 			    fnxgep->nxge_link_poll_timerid,
7065 			    fnxgep->nxge_timerid));
7066 		}
7067 #endif
7068 		if (fnxgep && fnxgep != nxgep &&
7069 		    (fnxgep->nxge_timerid || fnxgep->nxge_link_poll_timerid)) {
7070 			NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
7071 			    "==> nxge_niu_peu_reset: checking $%p "
7072 			    "(function %d) timer ids",
7073 			    fnxgep, fnxgep->function_num));
7074 			for (j = 0; j < NXGE_WAIT_QUITE_RETRY; j++) {
7075 				NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
7076 				    "==> nxge_niu_peu_reset: waiting"));
7077 				NXGE_DELAY(NXGE_WAIT_QUITE_TIME);
7078 				if (!fnxgep->nxge_timerid &&
7079 				    !fnxgep->nxge_link_poll_timerid) {
7080 					break;
7081 				}
7082 			}
7083 			NXGE_DELAY(NXGE_WAIT_QUITE_TIME);
7084 			if (fnxgep->nxge_timerid ||
7085 			    fnxgep->nxge_link_poll_timerid) {
7086 				MUTEX_EXIT(&hw_p->nxge_cfg_lock);
7087 				NXGE_ERROR_MSG((nxgep, NXGE_ERR_CTL,
7088 				    "<== nxge_niu_peu_reset: cannot reset "
7089 				    "hardware (devices are still in use)"));
7090 				return;
7091 			}
7092 		}
7093 	}
7094 
7095 	if ((hw_p->flags & COMMON_RESET_NIU_PCI) != COMMON_RESET_NIU_PCI) {
7096 		hw_p->flags |= COMMON_RESET_NIU_PCI;
7097 		rvalue = pci_config_get32(nxgep->dev_regs->nxge_pciregh,
7098 		    NXGE_PCI_PORT_LOGIC_OFFSET);
7099 		NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
7100 		    "nxge_niu_peu_reset: read offset 0x%x (%d) "
7101 		    "(data 0x%x)",
7102 		    NXGE_PCI_PORT_LOGIC_OFFSET,
7103 		    NXGE_PCI_PORT_LOGIC_OFFSET,
7104 		    rvalue));
7105 
7106 		rvalue |= NXGE_PCI_RESET_ALL;
7107 		pci_config_put32(nxgep->dev_regs->nxge_pciregh,
7108 		    NXGE_PCI_PORT_LOGIC_OFFSET, rvalue);
7109 		NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL,
7110 		    "nxge_niu_peu_reset: RESETTING NIU: write NIU reset 0x%x",
7111 		    rvalue));
7112 
7113 		NXGE_DELAY(NXGE_PCI_RESET_WAIT);
7114 	}
7115 
7116 	MUTEX_EXIT(&hw_p->nxge_cfg_lock);
7117 	NXGE_DEBUG_MSG((nxgep, NXGE_ERR_CTL, "<== nxge_niu_peu_reset"));
7118 }
7119 
7120 static void
7121 nxge_set_pci_replay_timeout(p_nxge_t nxgep)
7122 {
7123 	p_dev_regs_t	dev_regs;
7124 	uint32_t	value;
7125 
7126 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "==> nxge_set_pci_replay_timeout"));
7127 
7128 	if (!nxge_set_replay_timer) {
7129 		NXGE_DEBUG_MSG((nxgep, DDI_CTL,
7130 		    "==> nxge_set_pci_replay_timeout: will not change "
7131 		    "the timeout"));
7132 		return;
7133 	}
7134 
7135 	dev_regs = nxgep->dev_regs;
7136 	NXGE_DEBUG_MSG((nxgep, DDI_CTL,
7137 	    "==> nxge_set_pci_replay_timeout: dev_regs 0x%p pcireg 0x%p",
7138 	    dev_regs, dev_regs->nxge_pciregh));
7139 
7140 	if (dev_regs == NULL || (dev_regs->nxge_pciregh == NULL)) {
7141 		NXGE_DEBUG_MSG((nxgep, DDI_CTL,
7142 		    "==> nxge_set_pci_replay_timeout: NULL dev_regs $%p or "
7143 		    "no PCI handle",
7144 		    dev_regs));
7145 		return;
7146 	}
7147 	value = (pci_config_get32(dev_regs->nxge_pciregh,
7148 	    PCI_REPLAY_TIMEOUT_CFG_OFFSET) |
7149 	    (nxge_replay_timeout << PCI_REPLAY_TIMEOUT_SHIFT));
7150 
7151 	NXGE_DEBUG_MSG((nxgep, DDI_CTL,
7152 	    "nxge_set_pci_replay_timeout: replay timeout value before set 0x%x "
7153 	    "(timeout value to set 0x%x at offset 0x%x) value 0x%x",
7154 	    pci_config_get32(dev_regs->nxge_pciregh,
7155 	    PCI_REPLAY_TIMEOUT_CFG_OFFSET), nxge_replay_timeout,
7156 	    PCI_REPLAY_TIMEOUT_CFG_OFFSET, value));
7157 
7158 	pci_config_put32(dev_regs->nxge_pciregh, PCI_REPLAY_TIMEOUT_CFG_OFFSET,
7159 	    value);
7160 
7161 	NXGE_DEBUG_MSG((nxgep, DDI_CTL,
7162 	    "nxge_set_pci_replay_timeout: replay timeout value after set 0x%x",
7163 	    pci_config_get32(dev_regs->nxge_pciregh,
7164 	    PCI_REPLAY_TIMEOUT_CFG_OFFSET)));
7165 
7166 	NXGE_DEBUG_MSG((nxgep, DDI_CTL, "<== nxge_set_pci_replay_timeout"));
7167 }
7168 
7169 /*
7170  * quiesce(9E) entry point.
7171  *
7172  * This function is called when the system is single-threaded at high
7173  * PIL with preemption disabled. Therefore, this function must not be
7174  * blocked.
7175  *
7176  * This function returns DDI_SUCCESS on success, or DDI_FAILURE on failure.
7177  * DDI_FAILURE indicates an error condition and should almost never happen.
7178  */
7179 static int
7180 nxge_quiesce(dev_info_t *dip)
7181 {
7182 	int instance = ddi_get_instance(dip);
7183 	p_nxge_t nxgep = (p_nxge_t)ddi_get_soft_state(nxge_list, instance);
7184 
7185 	if (nxgep == NULL)
7186 		return (DDI_FAILURE);
7187 
7188 	/* Turn off debugging */
7189 	nxge_debug_level = NO_DEBUG;
7190 	nxgep->nxge_debug_level = NO_DEBUG;
7191 	npi_debug_level = NO_DEBUG;
7192 
7193 	/*
7194 	 * Stop link monitor only when linkchkmod is interrupt based
7195 	 */
7196 	if (nxgep->mac.linkchkmode == LINKCHK_INTR) {
7197 		(void) nxge_link_monitor(nxgep, LINK_MONITOR_STOP);
7198 	}
7199 
7200 	(void) nxge_intr_hw_disable(nxgep);
7201 
7202 	/*
7203 	 * Reset the receive MAC side.
7204 	 */
7205 	(void) nxge_rx_mac_disable(nxgep);
7206 
7207 	/* Disable and soft reset the IPP */
7208 	if (!isLDOMguest(nxgep))
7209 		(void) nxge_ipp_disable(nxgep);
7210 
7211 	/*
7212 	 * Reset the transmit/receive DMA side.
7213 	 */
7214 	(void) nxge_txdma_hw_mode(nxgep, NXGE_DMA_STOP);
7215 	(void) nxge_rxdma_hw_mode(nxgep, NXGE_DMA_STOP);
7216 
7217 	/*
7218 	 * Reset the transmit MAC side.
7219 	 */
7220 	(void) nxge_tx_mac_disable(nxgep);
7221 
7222 	return (DDI_SUCCESS);
7223 }
7224