xref: /titanic_50/usr/src/uts/sun4/io/px/px_intr.c (revision d6b882a974afe4a3ab38793d5de10bae2e078d39)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 /*
27  * PX nexus interrupt handling:
28  *	PX device interrupt handler wrapper
29  *	PIL lookup routine
30  *	PX device interrupt related initchild code
31  */
32 
33 #include <sys/types.h>
34 #include <sys/kmem.h>
35 #include <sys/async.h>
36 #include <sys/spl.h>
37 #include <sys/sunddi.h>
38 #include <sys/fm/protocol.h>
39 #include <sys/fm/util.h>
40 #include <sys/machsystm.h>	/* e_ddi_nodeid_to_dip() */
41 #include <sys/ddi_impldefs.h>
42 #include <sys/sdt.h>
43 #include <sys/atomic.h>
44 #include "px_obj.h"
45 #include <sys/ontrap.h>
46 #include <sys/membar.h>
47 #include <sys/clock.h>
48 
49 /*
50  * interrupt jabber:
51  *
52  * When an interrupt line is jabbering, every time the state machine for the
53  * associated ino is idled, a new mondo will be sent and the ino will go into
54  * the pending state again. The mondo will cause a new call to
55  * px_intr_wrapper() which normally idles the ino's state machine which would
56  * precipitate another trip round the loop.
57  *
58  * The loop can be broken by preventing the ino's state machine from being
59  * idled when an interrupt line is jabbering. See the comment at the
60  * beginning of px_intr_wrapper() explaining how the 'interrupt jabber
61  * protection' code does this.
62  */
63 
64 /*LINTLIBRARY*/
65 
66 /*
67  * If the unclaimed interrupt count has reached the limit set by
68  * pci_unclaimed_intr_max within the time limit, then all interrupts
69  * on this ino is blocked by not idling the interrupt state machine.
70  */
71 static int
72 px_spurintr(px_ino_pil_t *ipil_p)
73 {
74 	px_ino_t	*ino_p = ipil_p->ipil_ino_p;
75 	px_ih_t		*ih_p = ipil_p->ipil_ih_start;
76 	px_t		*px_p = ino_p->ino_ib_p->ib_px_p;
77 	char		*err_fmt_str;
78 	boolean_t	blocked = B_FALSE;
79 	int		i;
80 
81 	if (ino_p->ino_unclaimed_intrs > px_unclaimed_intr_max)
82 		return (DDI_INTR_CLAIMED);
83 
84 	if (!ino_p->ino_unclaimed_intrs)
85 		ino_p->ino_spurintr_begin = ddi_get_lbolt();
86 
87 	ino_p->ino_unclaimed_intrs++;
88 
89 	if (ino_p->ino_unclaimed_intrs <= px_unclaimed_intr_max)
90 		goto clear;
91 
92 	if (drv_hztousec(ddi_get_lbolt() - ino_p->ino_spurintr_begin)
93 	    > px_spurintr_duration) {
94 		ino_p->ino_unclaimed_intrs = 0;
95 		goto clear;
96 	}
97 	err_fmt_str = "%s%d: ino 0x%x blocked";
98 	blocked = B_TRUE;
99 	goto warn;
100 clear:
101 	err_fmt_str = "!%s%d: spurious interrupt from ino 0x%x";
102 warn:
103 	cmn_err(CE_WARN, err_fmt_str, NAMEINST(px_p->px_dip), ino_p->ino_ino);
104 	for (i = 0; i < ipil_p->ipil_ih_size; i++, ih_p = ih_p->ih_next)
105 		cmn_err(CE_CONT, "!%s-%d#%x ", NAMEINST(ih_p->ih_dip),
106 		    ih_p->ih_inum);
107 	cmn_err(CE_CONT, "!\n");
108 
109 	/* Clear the pending state */
110 	if (blocked == B_FALSE) {
111 		if (px_lib_intr_setstate(px_p->px_dip, ino_p->ino_sysino,
112 		    INTR_IDLE_STATE) != DDI_SUCCESS)
113 			return (DDI_INTR_UNCLAIMED);
114 	}
115 
116 	return (DDI_INTR_CLAIMED);
117 }
118 
119 extern uint64_t intr_get_time(void);
120 
121 /*
122  * px_intx_intr (INTx or legacy interrupt handler)
123  *
124  * This routine is used as wrapper around interrupt handlers installed by child
125  * device drivers.  This routine invokes the driver interrupt handlers and
126  * examines the return codes.
127  *
128  * There is a count of unclaimed interrupts kept on a per-ino basis. If at
129  * least one handler claims the interrupt then the counter is halved and the
130  * interrupt state machine is idled. If no handler claims the interrupt then
131  * the counter is incremented by one and the state machine is idled.
132  * If the count ever reaches the limit value set by pci_unclaimed_intr_max
133  * then the interrupt state machine is not idled thus preventing any further
134  * interrupts on that ino. The state machine will only be idled again if a
135  * handler is subsequently added or removed.
136  *
137  * return value: DDI_INTR_CLAIMED if any handlers claimed the interrupt,
138  * DDI_INTR_UNCLAIMED otherwise.
139  */
140 uint_t
141 px_intx_intr(caddr_t arg)
142 {
143 	px_ino_pil_t	*ipil_p = (px_ino_pil_t *)arg;
144 	px_ino_t	*ino_p = ipil_p->ipil_ino_p;
145 	px_t		*px_p = ino_p->ino_ib_p->ib_px_p;
146 	px_ih_t		*ih_p = ipil_p->ipil_ih_start;
147 	ushort_t	pil = ipil_p->ipil_pil;
148 	uint_t		result = 0, r = DDI_INTR_UNCLAIMED;
149 	int		i;
150 
151 	DBG(DBG_INTX_INTR, px_p->px_dip, "px_intx_intr:"
152 	    "ino=%x sysino=%llx pil=%x ih_size=%x ih_lst=%x\n",
153 	    ino_p->ino_ino, ino_p->ino_sysino, ipil_p->ipil_pil,
154 	    ipil_p->ipil_ih_size, ipil_p->ipil_ih_head);
155 
156 	for (i = 0; i < ipil_p->ipil_ih_size; i++, ih_p = ih_p->ih_next) {
157 		dev_info_t *dip = ih_p->ih_dip;
158 		uint_t (*handler)() = ih_p->ih_handler;
159 		caddr_t arg1 = ih_p->ih_handler_arg1;
160 		caddr_t arg2 = ih_p->ih_handler_arg2;
161 
162 		if (ih_p->ih_intr_state == PX_INTR_STATE_DISABLE) {
163 			DBG(DBG_INTX_INTR, px_p->px_dip,
164 			    "px_intx_intr: %s%d interrupt %d is disabled\n",
165 			    ddi_driver_name(dip), ddi_get_instance(dip),
166 			    ino_p->ino_ino);
167 
168 			continue;
169 		}
170 
171 		DBG(DBG_INTX_INTR, px_p->px_dip, "px_intx_intr:"
172 		    "ino=%x handler=%p arg1 =%p arg2 = %p\n",
173 		    ino_p->ino_ino, handler, arg1, arg2);
174 
175 		DTRACE_PROBE4(interrupt__start, dev_info_t, dip,
176 		    void *, handler, caddr_t, arg1, caddr_t, arg2);
177 
178 		r = (*handler)(arg1, arg2);
179 
180 		/*
181 		 * Account for time used by this interrupt. Protect against
182 		 * conflicting writes to ih_ticks from ib_intr_dist_all() by
183 		 * using atomic ops.
184 		 */
185 
186 		if (pil <= LOCK_LEVEL)
187 			atomic_add_64(&ih_p->ih_ticks, intr_get_time());
188 
189 		DTRACE_PROBE4(interrupt__complete, dev_info_t, dip,
190 		    void *, handler, caddr_t, arg1, int, r);
191 
192 		result += r;
193 
194 		if (px_check_all_handlers)
195 			continue;
196 		if (result)
197 			break;
198 	}
199 
200 	if (result)
201 		ino_p->ino_claimed |= (1 << pil);
202 
203 	/* Interrupt can only be cleared after all pil levels are handled */
204 	if (pil != ino_p->ino_lopil)
205 		return (DDI_INTR_CLAIMED);
206 
207 	if (!ino_p->ino_claimed) {
208 		if (px_unclaimed_intr_block)
209 			return (px_spurintr(ipil_p));
210 	}
211 
212 	ino_p->ino_unclaimed_intrs = 0;
213 	ino_p->ino_claimed = 0;
214 
215 	/* Clear the pending state */
216 	if (px_lib_intr_setstate(px_p->px_dip,
217 	    ino_p->ino_sysino, INTR_IDLE_STATE) != DDI_SUCCESS)
218 		return (DDI_INTR_UNCLAIMED);
219 
220 	return (DDI_INTR_CLAIMED);
221 }
222 
223 /*
224  * px_msiq_intr (MSI/X or PCIe MSG interrupt handler)
225  *
226  * This routine is used as wrapper around interrupt handlers installed by child
227  * device drivers.  This routine invokes the driver interrupt handlers and
228  * examines the return codes.
229  *
230  * There is a count of unclaimed interrupts kept on a per-ino basis. If at
231  * least one handler claims the interrupt then the counter is halved and the
232  * interrupt state machine is idled. If no handler claims the interrupt then
233  * the counter is incremented by one and the state machine is idled.
234  * If the count ever reaches the limit value set by pci_unclaimed_intr_max
235  * then the interrupt state machine is not idled thus preventing any further
236  * interrupts on that ino. The state machine will only be idled again if a
237  * handler is subsequently added or removed.
238  *
239  * return value: DDI_INTR_CLAIMED if any handlers claimed the interrupt,
240  * DDI_INTR_UNCLAIMED otherwise.
241  */
242 uint_t
243 px_msiq_intr(caddr_t arg)
244 {
245 	px_ino_pil_t	*ipil_p = (px_ino_pil_t *)arg;
246 	px_ino_t	*ino_p = ipil_p->ipil_ino_p;
247 	px_t		*px_p = ino_p->ino_ib_p->ib_px_p;
248 	px_msiq_state_t	*msiq_state_p = &px_p->px_ib_p->ib_msiq_state;
249 	px_msiq_t	*msiq_p = ino_p->ino_msiq_p;
250 	dev_info_t	*dip = px_p->px_dip;
251 	ushort_t	pil = ipil_p->ipil_pil;
252 	msiq_rec_t	msiq_rec, *msiq_rec_p = &msiq_rec;
253 	msiqhead_t	*curr_head_p;
254 	msiqtail_t	curr_tail_index;
255 	msgcode_t	msg_code;
256 	px_ih_t		*ih_p;
257 	uint_t		ret = DDI_INTR_UNCLAIMED;
258 	int		i, j;
259 
260 	DBG(DBG_MSIQ_INTR, dip, "px_msiq_intr: msiq_id =%x ino=%x pil=%x "
261 	    "ih_size=%x ih_lst=%x\n", msiq_p->msiq_id, ino_p->ino_ino,
262 	    ipil_p->ipil_pil, ipil_p->ipil_ih_size, ipil_p->ipil_ih_head);
263 
264 	/*
265 	 * The px_msiq_intr() handles multiple interrupt priorities and it
266 	 * will set msiq->msiq_rec2process to the number of MSIQ records to
267 	 * process while handling the highest priority interrupt. Subsequent
268 	 * lower priority interrupts will just process any unprocessed MSIQ
269 	 * records or will just return immediately.
270 	 */
271 	if (msiq_p->msiq_recs2process == 0) {
272 		/* Read current MSIQ tail index */
273 		px_lib_msiq_gettail(dip, msiq_p->msiq_id, &curr_tail_index);
274 		msiq_p->msiq_new_head_index = msiq_p->msiq_curr_head_index;
275 
276 		if (curr_tail_index < msiq_p->msiq_curr_head_index)
277 			curr_tail_index += msiq_state_p->msiq_rec_cnt;
278 
279 		msiq_p->msiq_recs2process = curr_tail_index -
280 		    msiq_p->msiq_curr_head_index;
281 	}
282 
283 	DBG(DBG_MSIQ_INTR, dip, "px_msiq_intr: curr_head %x new_head %x "
284 	    "rec2process %x\n", msiq_p->msiq_curr_head_index,
285 	    msiq_p->msiq_new_head_index, msiq_p->msiq_recs2process);
286 
287 	/* If all MSIQ records are already processed, just return immediately */
288 	if ((msiq_p->msiq_new_head_index - msiq_p->msiq_curr_head_index)
289 	    == msiq_p->msiq_recs2process)
290 		goto intr_done;
291 
292 	curr_head_p = (msiqhead_t *)((caddr_t)msiq_p->msiq_base_p +
293 	    (msiq_p->msiq_curr_head_index * sizeof (msiq_rec_t)));
294 
295 	/*
296 	 * Calculate the number of recs to process by taking the difference
297 	 * between the head and tail pointers. For all records we always
298 	 * verify that we have a valid record type before we do any processing.
299 	 * If triggered, we should always have at least one valid record.
300 	 */
301 	for (i = 0; i < msiq_p->msiq_recs2process; i++) {
302 		/* Read next MSIQ record */
303 		px_lib_get_msiq_rec(dip, curr_head_p, msiq_rec_p);
304 
305 		DBG(DBG_MSIQ_INTR, dip, "px_msiq_intr: MSIQ RECORD, "
306 		    "msiq_rec_type 0x%llx msiq_rec_rid 0x%llx\n",
307 		    msiq_rec_p->msiq_rec_type, msiq_rec_p->msiq_rec_rid);
308 
309 		if (!msiq_rec_p->msiq_rec_type)
310 			goto next_rec;
311 
312 		/* Check MSIQ record type */
313 		switch (msiq_rec_p->msiq_rec_type) {
314 		case MSG_REC:
315 			msg_code = msiq_rec_p->msiq_rec_data.msg.msg_code;
316 			DBG(DBG_MSIQ_INTR, dip, "px_msiq_intr: PCIE MSG "
317 			    "record, msg type 0x%x\n", msg_code);
318 			break;
319 		case MSI32_REC:
320 		case MSI64_REC:
321 			msg_code = msiq_rec_p->msiq_rec_data.msi.msi_data;
322 			DBG(DBG_MSIQ_INTR, dip, "px_msiq_intr: MSI record, "
323 			    "msi 0x%x\n", msg_code);
324 
325 			/* Clear MSI state */
326 			px_lib_msi_setstate(dip, (msinum_t)msg_code,
327 			    PCI_MSI_STATE_IDLE);
328 			break;
329 		default:
330 			msg_code = 0;
331 			cmn_err(CE_WARN, "%s%d: px_msiq_intr: 0x%x MSIQ "
332 			    "record type is not supported",
333 			    ddi_driver_name(dip), ddi_get_instance(dip),
334 			    msiq_rec_p->msiq_rec_type);
335 
336 			goto next_rec;
337 		}
338 
339 		/*
340 		 * Scan through px_ih_t linked list, searching for the
341 		 * right px_ih_t, matching MSIQ record data.
342 		 */
343 		for (j = 0, ih_p = ipil_p->ipil_ih_start;
344 		    ih_p && (j < ipil_p->ipil_ih_size) &&
345 		    ((ih_p->ih_msg_code != msg_code) ||
346 		    (ih_p->ih_rec_type != msiq_rec_p->msiq_rec_type));
347 		    ih_p = ih_p->ih_next, j++)
348 			;
349 
350 		if ((ih_p->ih_msg_code == msg_code) &&
351 		    (ih_p->ih_rec_type == msiq_rec_p->msiq_rec_type)) {
352 			dev_info_t *dip = ih_p->ih_dip;
353 			uint_t (*handler)() = ih_p->ih_handler;
354 			caddr_t arg1 = ih_p->ih_handler_arg1;
355 			caddr_t arg2 = ih_p->ih_handler_arg2;
356 
357 			DBG(DBG_MSIQ_INTR, dip, "px_msiq_intr: ino=%x data=%x "
358 			    "handler=%p arg1 =%p arg2=%p\n", ino_p->ino_ino,
359 			    msg_code, handler, arg1, arg2);
360 
361 			DTRACE_PROBE4(interrupt__start, dev_info_t, dip,
362 			    void *, handler, caddr_t, arg1, caddr_t, arg2);
363 
364 			/*
365 			 * Special case for PCIE Error Messages.
366 			 * The current frame work doesn't fit PCIE Err Msgs
367 			 * This should be fixed when PCIE MESSAGES as a whole
368 			 * is architected correctly.
369 			 */
370 			if ((msg_code == PCIE_MSG_CODE_ERR_COR) ||
371 			    (msg_code == PCIE_MSG_CODE_ERR_NONFATAL) ||
372 			    (msg_code == PCIE_MSG_CODE_ERR_FATAL)) {
373 				ret = px_err_fabric_intr(px_p, msg_code,
374 				    msiq_rec_p->msiq_rec_rid);
375 			} else
376 				ret = (*handler)(arg1, arg2);
377 
378 			/*
379 			 * Account for time used by this interrupt. Protect
380 			 * against conflicting writes to ih_ticks from
381 			 * ib_intr_dist_all() by using atomic ops.
382 			 */
383 
384 			if (pil <= LOCK_LEVEL)
385 				atomic_add_64(&ih_p->ih_ticks, intr_get_time());
386 
387 			DTRACE_PROBE4(interrupt__complete, dev_info_t, dip,
388 			    void *, handler, caddr_t, arg1, int, ret);
389 
390 			msiq_p->msiq_new_head_index++;
391 			px_lib_clr_msiq_rec(dip, curr_head_p);
392 		} else {
393 			DBG(DBG_MSIQ_INTR, dip, "px_msiq_intr:"
394 			    "No matching MSIQ record found\n");
395 		}
396 next_rec:
397 		/* Get the pointer next EQ record */
398 		curr_head_p = (msiqhead_t *)
399 		    ((caddr_t)curr_head_p + sizeof (msiq_rec_t));
400 
401 		/* Check for overflow condition */
402 		if (curr_head_p >= (msiqhead_t *)((caddr_t)msiq_p->msiq_base_p
403 		    + (msiq_state_p->msiq_rec_cnt * sizeof (msiq_rec_t))))
404 			curr_head_p = (msiqhead_t *)msiq_p->msiq_base_p;
405 	}
406 
407 	DBG(DBG_MSIQ_INTR, dip, "px_msiq_intr: No of MSIQ recs processed %x\n",
408 	    (msiq_p->msiq_new_head_index - msiq_p->msiq_curr_head_index));
409 
410 	DBG(DBG_MSIQ_INTR, dip, "px_msiq_intr: curr_head %x new_head %x "
411 	    "rec2process %x\n", msiq_p->msiq_curr_head_index,
412 	    msiq_p->msiq_new_head_index, msiq_p->msiq_recs2process);
413 
414 	/* ino_claimed used just for debugging purpose */
415 	if (ret)
416 		ino_p->ino_claimed |= (1 << pil);
417 
418 intr_done:
419 	/* Interrupt can only be cleared after all pil levels are handled */
420 	if (pil != ino_p->ino_lopil)
421 		return (DDI_INTR_CLAIMED);
422 
423 	if (msiq_p->msiq_new_head_index <= msiq_p->msiq_curr_head_index)  {
424 		if (px_unclaimed_intr_block)
425 			return (px_spurintr(ipil_p));
426 	}
427 
428 	/*  Update MSIQ head index with no of MSIQ records processed */
429 	if (msiq_p->msiq_new_head_index >= msiq_state_p->msiq_rec_cnt)
430 		msiq_p->msiq_new_head_index -= msiq_state_p->msiq_rec_cnt;
431 
432 	msiq_p->msiq_curr_head_index = msiq_p->msiq_new_head_index;
433 	px_lib_msiq_sethead(dip, msiq_p->msiq_id, msiq_p->msiq_new_head_index);
434 
435 	msiq_p->msiq_new_head_index = 0;
436 	msiq_p->msiq_recs2process = 0;
437 	ino_p->ino_claimed = 0;
438 
439 	/* Clear the pending state */
440 	if (px_lib_intr_setstate(dip, ino_p->ino_sysino,
441 	    INTR_IDLE_STATE) != DDI_SUCCESS)
442 		return (DDI_INTR_UNCLAIMED);
443 
444 	return (DDI_INTR_CLAIMED);
445 }
446 
447 dev_info_t *
448 px_get_my_childs_dip(dev_info_t *dip, dev_info_t *rdip)
449 {
450 	dev_info_t	*cdip = rdip;
451 
452 	for (; ddi_get_parent(cdip) != dip; cdip = ddi_get_parent(cdip))
453 		;
454 
455 	return (cdip);
456 }
457 
458 /* ARGSUSED */
459 int
460 px_intx_ops(dev_info_t *dip, dev_info_t *rdip, ddi_intr_op_t intr_op,
461     ddi_intr_handle_impl_t *hdlp, void *result)
462 {
463 	px_t	*px_p = DIP_TO_STATE(dip);
464 	int	ret = DDI_SUCCESS;
465 
466 	DBG(DBG_INTROPS, dip, "px_intx_ops: dip=%x rdip=%x intr_op=%x "
467 	    "handle=%p\n", dip, rdip, intr_op, hdlp);
468 
469 	switch (intr_op) {
470 	case DDI_INTROP_GETCAP:
471 		ret = pci_intx_get_cap(rdip, (int *)result);
472 		break;
473 	case DDI_INTROP_SETCAP:
474 		DBG(DBG_INTROPS, dip, "px_intx_ops: SetCap is not supported\n");
475 		ret = DDI_ENOTSUP;
476 		break;
477 	case DDI_INTROP_ALLOC:
478 		*(int *)result = hdlp->ih_scratch1;
479 		break;
480 	case DDI_INTROP_FREE:
481 		break;
482 	case DDI_INTROP_GETPRI:
483 		*(int *)result = hdlp->ih_pri ?
484 		    hdlp->ih_pri : pci_class_to_pil(rdip);
485 		break;
486 	case DDI_INTROP_SETPRI:
487 		break;
488 	case DDI_INTROP_ADDISR:
489 		ret = px_add_intx_intr(dip, rdip, hdlp);
490 		break;
491 	case DDI_INTROP_REMISR:
492 		ret = px_rem_intx_intr(dip, rdip, hdlp);
493 		break;
494 	case DDI_INTROP_ENABLE:
495 		ret = px_ib_update_intr_state(px_p, rdip, hdlp->ih_inum,
496 		    hdlp->ih_vector, hdlp->ih_pri, PX_INTR_STATE_ENABLE, 0, 0);
497 		break;
498 	case DDI_INTROP_DISABLE:
499 		ret = px_ib_update_intr_state(px_p, rdip, hdlp->ih_inum,
500 		    hdlp->ih_vector, hdlp->ih_pri, PX_INTR_STATE_DISABLE, 0, 0);
501 		break;
502 	case DDI_INTROP_SETMASK:
503 		ret = pci_intx_set_mask(rdip);
504 		break;
505 	case DDI_INTROP_CLRMASK:
506 		ret = pci_intx_clr_mask(rdip);
507 		break;
508 	case DDI_INTROP_GETPENDING:
509 		ret = pci_intx_get_pending(rdip, (int *)result);
510 		break;
511 	case DDI_INTROP_NINTRS:
512 	case DDI_INTROP_NAVAIL:
513 		*(int *)result = i_ddi_get_intx_nintrs(rdip);
514 		break;
515 	default:
516 		ret = DDI_ENOTSUP;
517 		break;
518 	}
519 
520 	return (ret);
521 }
522 
523 /* ARGSUSED */
524 int
525 px_msix_ops(dev_info_t *dip, dev_info_t *rdip, ddi_intr_op_t intr_op,
526     ddi_intr_handle_impl_t *hdlp, void *result)
527 {
528 	px_t			*px_p = DIP_TO_STATE(dip);
529 	px_msi_state_t		*msi_state_p = &px_p->px_ib_p->ib_msi_state;
530 	msiq_rec_type_t		msiq_rec_type;
531 	msi_type_t		msi_type;
532 	uint64_t		msi_addr;
533 	msinum_t		msi_num;
534 	msiqid_t		msiq_id;
535 	uint_t			nintrs;
536 	int			i, ret = DDI_SUCCESS;
537 
538 	DBG(DBG_INTROPS, dip, "px_msix_ops: dip=%x rdip=%x intr_op=%x "
539 	    "handle=%p\n", dip, rdip, intr_op, hdlp);
540 
541 	/* Check for MSI64 support */
542 	if ((hdlp->ih_cap & DDI_INTR_FLAG_MSI64) && msi_state_p->msi_addr64) {
543 		msiq_rec_type = MSI64_REC;
544 		msi_type = MSI64_TYPE;
545 		msi_addr = msi_state_p->msi_addr64;
546 	} else {
547 		msiq_rec_type = MSI32_REC;
548 		msi_type = MSI32_TYPE;
549 		msi_addr = msi_state_p->msi_addr32;
550 	}
551 
552 	switch (intr_op) {
553 	case DDI_INTROP_GETCAP:
554 		ret = pci_msi_get_cap(rdip, hdlp->ih_type, (int *)result);
555 		break;
556 	case DDI_INTROP_SETCAP:
557 		DBG(DBG_INTROPS, dip, "px_msix_ops: SetCap is not supported\n");
558 		ret = DDI_ENOTSUP;
559 		break;
560 	case DDI_INTROP_ALLOC:
561 		/*
562 		 * We need to restrict this allocation in future
563 		 * based on Resource Management policies.
564 		 */
565 		if ((ret = px_msi_alloc(px_p, rdip, hdlp->ih_type,
566 		    hdlp->ih_inum, hdlp->ih_scratch1,
567 		    (uintptr_t)hdlp->ih_scratch2,
568 		    (int *)result)) != DDI_SUCCESS) {
569 			DBG(DBG_INTROPS, dip, "px_msix_ops: allocation "
570 			    "failed, rdip 0x%p type 0x%d inum 0x%x "
571 			    "count 0x%x\n", rdip, hdlp->ih_type, hdlp->ih_inum,
572 			    hdlp->ih_scratch1);
573 
574 			return (ret);
575 		}
576 
577 		if ((hdlp->ih_type == DDI_INTR_TYPE_MSIX) &&
578 		    (i_ddi_get_msix(rdip) == NULL)) {
579 			ddi_intr_msix_t		*msix_p;
580 
581 			if (msix_p = pci_msix_init(rdip)) {
582 				i_ddi_set_msix(rdip, msix_p);
583 				break;
584 			}
585 
586 			DBG(DBG_INTROPS, dip, "px_msix_ops: MSI-X allocation "
587 			    "failed, rdip 0x%p inum 0x%x\n", rdip,
588 			    hdlp->ih_inum);
589 
590 			(void) px_msi_free(px_p, rdip, hdlp->ih_inum,
591 			    hdlp->ih_scratch1);
592 
593 			return (DDI_FAILURE);
594 		}
595 
596 		break;
597 	case DDI_INTROP_FREE:
598 		(void) pci_msi_unconfigure(rdip, hdlp->ih_type, hdlp->ih_inum);
599 
600 		if (hdlp->ih_type == DDI_INTR_TYPE_MSI)
601 			goto msi_free;
602 
603 		if (hdlp->ih_flags & DDI_INTR_MSIX_DUP)
604 			break;
605 
606 		if (((i_ddi_intr_get_current_nintrs(hdlp->ih_dip) - 1) == 0) &&
607 		    (i_ddi_get_msix(rdip))) {
608 			pci_msix_fini(i_ddi_get_msix(rdip));
609 			i_ddi_set_msix(rdip, NULL);
610 		}
611 msi_free:
612 		(void) px_msi_free(px_p, rdip, hdlp->ih_inum,
613 		    hdlp->ih_scratch1);
614 		break;
615 	case DDI_INTROP_GETPRI:
616 		*(int *)result = hdlp->ih_pri ?
617 		    hdlp->ih_pri : pci_class_to_pil(rdip);
618 		break;
619 	case DDI_INTROP_SETPRI:
620 		break;
621 	case DDI_INTROP_ADDISR:
622 		if ((ret = px_msi_get_msinum(px_p, hdlp->ih_dip,
623 		    hdlp->ih_inum, &msi_num)) != DDI_SUCCESS)
624 			return (ret);
625 
626 		if ((ret = px_add_msiq_intr(dip, rdip, hdlp,
627 		    msiq_rec_type, msi_num, &msiq_id)) != DDI_SUCCESS) {
628 			DBG(DBG_INTROPS, dip, "px_msix_ops: Add MSI handler "
629 			    "failed, rdip 0x%p msi 0x%x\n", rdip, msi_num);
630 			return (ret);
631 		}
632 
633 		DBG(DBG_INTROPS, dip, "px_msix_ops: msiq used 0x%x\n", msiq_id);
634 
635 		if ((ret = px_lib_msi_setmsiq(dip, msi_num,
636 		    msiq_id, msi_type)) != DDI_SUCCESS) {
637 			(void) px_rem_msiq_intr(dip, rdip,
638 			    hdlp, msiq_rec_type, msi_num, msiq_id);
639 			return (ret);
640 		}
641 
642 		if ((ret = px_lib_msi_setstate(dip, msi_num,
643 		    PCI_MSI_STATE_IDLE)) != DDI_SUCCESS) {
644 			(void) px_rem_msiq_intr(dip, rdip,
645 			    hdlp, msiq_rec_type, msi_num, msiq_id);
646 			return (ret);
647 		}
648 
649 		hdlp->ih_vector = msi_num;
650 		break;
651 	case DDI_INTROP_DUPVEC:
652 		DBG(DBG_INTROPS, dip, "px_msix_ops: dupisr - inum: %x, "
653 		    "new_vector: %x\n", hdlp->ih_inum, hdlp->ih_scratch1);
654 
655 		ret = pci_msix_dup(hdlp->ih_dip, hdlp->ih_inum,
656 		    hdlp->ih_scratch1);
657 		break;
658 	case DDI_INTROP_REMISR:
659 		msi_num = hdlp->ih_vector;
660 
661 		if ((ret = px_lib_msi_getmsiq(dip, msi_num,
662 		    &msiq_id)) != DDI_SUCCESS)
663 			return (ret);
664 
665 		if ((ret = px_lib_msi_setstate(dip, msi_num,
666 		    PCI_MSI_STATE_IDLE)) != DDI_SUCCESS)
667 			return (ret);
668 
669 		ret = px_rem_msiq_intr(dip, rdip,
670 		    hdlp, msiq_rec_type, msi_num, msiq_id);
671 
672 		hdlp->ih_vector = 0;
673 		break;
674 	case DDI_INTROP_ENABLE:
675 		msi_num = hdlp->ih_vector;
676 
677 		if ((ret = px_lib_msi_setvalid(dip, msi_num,
678 		    PCI_MSI_VALID)) != DDI_SUCCESS)
679 			return (ret);
680 
681 		if ((pci_is_msi_enabled(rdip, hdlp->ih_type) != DDI_SUCCESS) ||
682 		    (hdlp->ih_type == DDI_INTR_TYPE_MSIX)) {
683 			nintrs = i_ddi_intr_get_current_nintrs(hdlp->ih_dip);
684 
685 			if ((ret = pci_msi_configure(rdip, hdlp->ih_type,
686 			    nintrs, hdlp->ih_inum, msi_addr,
687 			    hdlp->ih_type == DDI_INTR_TYPE_MSIX ?
688 			    msi_num : msi_num & ~(nintrs - 1))) != DDI_SUCCESS)
689 				return (ret);
690 
691 			if ((ret = pci_msi_enable_mode(rdip, hdlp->ih_type))
692 			    != DDI_SUCCESS)
693 				return (ret);
694 		}
695 
696 		if ((ret = pci_msi_clr_mask(rdip, hdlp->ih_type,
697 		    hdlp->ih_inum)) != DDI_SUCCESS)
698 			return (ret);
699 
700 		if (hdlp->ih_flags & DDI_INTR_MSIX_DUP)
701 			break;
702 
703 		if ((ret = px_lib_msi_getmsiq(dip, msi_num,
704 		    &msiq_id)) != DDI_SUCCESS)
705 			return (ret);
706 
707 		ret = px_ib_update_intr_state(px_p, rdip, hdlp->ih_inum,
708 		    px_msiqid_to_devino(px_p, msiq_id), hdlp->ih_pri,
709 		    PX_INTR_STATE_ENABLE, msiq_rec_type, msi_num);
710 
711 		break;
712 	case DDI_INTROP_DISABLE:
713 		msi_num = hdlp->ih_vector;
714 
715 		if ((ret = pci_msi_disable_mode(rdip, hdlp->ih_type,
716 		    hdlp->ih_cap & DDI_INTR_FLAG_BLOCK)) != DDI_SUCCESS)
717 			return (ret);
718 
719 		if ((ret = pci_msi_set_mask(rdip, hdlp->ih_type,
720 		    hdlp->ih_inum)) != DDI_SUCCESS)
721 			return (ret);
722 
723 		if ((ret = px_lib_msi_setvalid(dip, msi_num,
724 		    PCI_MSI_INVALID)) != DDI_SUCCESS)
725 			return (ret);
726 
727 		if (hdlp->ih_flags & DDI_INTR_MSIX_DUP)
728 			break;
729 
730 		if ((ret = px_lib_msi_getmsiq(dip, msi_num,
731 		    &msiq_id)) != DDI_SUCCESS)
732 			return (ret);
733 
734 		ret = px_ib_update_intr_state(px_p, rdip,
735 		    hdlp->ih_inum, px_msiqid_to_devino(px_p, msiq_id),
736 		    hdlp->ih_pri, PX_INTR_STATE_DISABLE, msiq_rec_type,
737 		    msi_num);
738 
739 		break;
740 	case DDI_INTROP_BLOCKENABLE:
741 		nintrs = i_ddi_intr_get_current_nintrs(hdlp->ih_dip);
742 		msi_num = hdlp->ih_vector;
743 
744 		if ((ret = pci_msi_configure(rdip, hdlp->ih_type,
745 		    nintrs, hdlp->ih_inum, msi_addr,
746 		    msi_num & ~(nintrs - 1))) != DDI_SUCCESS)
747 			return (ret);
748 
749 		for (i = 0; i < nintrs; i++, msi_num++) {
750 			if ((ret = px_lib_msi_setvalid(dip, msi_num,
751 			    PCI_MSI_VALID)) != DDI_SUCCESS)
752 				return (ret);
753 
754 			if ((ret = px_lib_msi_getmsiq(dip, msi_num,
755 			    &msiq_id)) != DDI_SUCCESS)
756 				return (ret);
757 
758 			if ((ret = px_ib_update_intr_state(px_p, rdip,
759 			    hdlp->ih_inum + i, px_msiqid_to_devino(px_p,
760 			    msiq_id), hdlp->ih_pri, PX_INTR_STATE_ENABLE,
761 			    msiq_rec_type, msi_num)) != DDI_SUCCESS)
762 				return (ret);
763 		}
764 
765 		ret = pci_msi_enable_mode(rdip, hdlp->ih_type);
766 		break;
767 	case DDI_INTROP_BLOCKDISABLE:
768 		nintrs = i_ddi_intr_get_current_nintrs(hdlp->ih_dip);
769 		msi_num = hdlp->ih_vector;
770 
771 		if ((ret = pci_msi_disable_mode(rdip, hdlp->ih_type,
772 		    hdlp->ih_cap & DDI_INTR_FLAG_BLOCK)) != DDI_SUCCESS)
773 			return (ret);
774 
775 		for (i = 0; i < nintrs; i++, msi_num++) {
776 			if ((ret = px_lib_msi_setvalid(dip, msi_num,
777 			    PCI_MSI_INVALID)) != DDI_SUCCESS)
778 				return (ret);
779 
780 			if ((ret = px_lib_msi_getmsiq(dip, msi_num,
781 			    &msiq_id)) != DDI_SUCCESS)
782 				return (ret);
783 
784 			if ((ret = px_ib_update_intr_state(px_p, rdip,
785 			    hdlp->ih_inum + i, px_msiqid_to_devino(px_p,
786 			    msiq_id), hdlp->ih_pri, PX_INTR_STATE_DISABLE,
787 			    msiq_rec_type, msi_num)) != DDI_SUCCESS)
788 				return (ret);
789 		}
790 
791 		break;
792 	case DDI_INTROP_SETMASK:
793 		ret = pci_msi_set_mask(rdip, hdlp->ih_type, hdlp->ih_inum);
794 		break;
795 	case DDI_INTROP_CLRMASK:
796 		ret = pci_msi_clr_mask(rdip, hdlp->ih_type, hdlp->ih_inum);
797 		break;
798 	case DDI_INTROP_GETPENDING:
799 		ret = pci_msi_get_pending(rdip, hdlp->ih_type,
800 		    hdlp->ih_inum, (int *)result);
801 		break;
802 	case DDI_INTROP_NINTRS:
803 		ret = pci_msi_get_nintrs(rdip, hdlp->ih_type, (int *)result);
804 		break;
805 	case DDI_INTROP_NAVAIL:
806 		/* XXX - a new interface may be needed */
807 		ret = pci_msi_get_nintrs(rdip, hdlp->ih_type, (int *)result);
808 		break;
809 	case DDI_INTROP_GETPOOL:
810 		if (msi_state_p->msi_pool_p == NULL) {
811 			*(ddi_irm_pool_t **)result = NULL;
812 			return (DDI_ENOTSUP);
813 		}
814 		*(ddi_irm_pool_t **)result = msi_state_p->msi_pool_p;
815 		ret = DDI_SUCCESS;
816 		break;
817 	default:
818 		ret = DDI_ENOTSUP;
819 		break;
820 	}
821 
822 	return (ret);
823 }
824 
825 static struct {
826 	kstat_named_t pxintr_ks_name;
827 	kstat_named_t pxintr_ks_type;
828 	kstat_named_t pxintr_ks_cpu;
829 	kstat_named_t pxintr_ks_pil;
830 	kstat_named_t pxintr_ks_time;
831 	kstat_named_t pxintr_ks_ino;
832 	kstat_named_t pxintr_ks_cookie;
833 	kstat_named_t pxintr_ks_devpath;
834 	kstat_named_t pxintr_ks_buspath;
835 } pxintr_ks_template = {
836 	{ "name",	KSTAT_DATA_CHAR },
837 	{ "type",	KSTAT_DATA_CHAR },
838 	{ "cpu",	KSTAT_DATA_UINT64 },
839 	{ "pil",	KSTAT_DATA_UINT64 },
840 	{ "time",	KSTAT_DATA_UINT64 },
841 	{ "ino",	KSTAT_DATA_UINT64 },
842 	{ "cookie",	KSTAT_DATA_UINT64 },
843 	{ "devpath",	KSTAT_DATA_STRING },
844 	{ "buspath",	KSTAT_DATA_STRING },
845 };
846 
847 static uint32_t pxintr_ks_instance;
848 static char ih_devpath[MAXPATHLEN];
849 static char ih_buspath[MAXPATHLEN];
850 kmutex_t pxintr_ks_template_lock;
851 
852 int
853 px_ks_update(kstat_t *ksp, int rw)
854 {
855 	px_ih_t *ih_p = ksp->ks_private;
856 	int maxlen = sizeof (pxintr_ks_template.pxintr_ks_name.value.c);
857 	px_ino_pil_t *ipil_p = ih_p->ih_ipil_p;
858 	px_ino_t *ino_p = ipil_p->ipil_ino_p;
859 	px_t *px_p = ino_p->ino_ib_p->ib_px_p;
860 	devino_t ino;
861 	sysino_t sysino;
862 
863 	ino = ino_p->ino_ino;
864 	if (px_lib_intr_devino_to_sysino(px_p->px_dip, ino, &sysino) !=
865 	    DDI_SUCCESS) {
866 		cmn_err(CE_WARN, "px_ks_update: px_lib_intr_devino_to_sysino "
867 		    "failed");
868 	}
869 
870 	(void) snprintf(pxintr_ks_template.pxintr_ks_name.value.c, maxlen,
871 	    "%s%d", ddi_driver_name(ih_p->ih_dip),
872 	    ddi_get_instance(ih_p->ih_dip));
873 
874 	(void) ddi_pathname(ih_p->ih_dip, ih_devpath);
875 	(void) ddi_pathname(px_p->px_dip, ih_buspath);
876 	kstat_named_setstr(&pxintr_ks_template.pxintr_ks_devpath, ih_devpath);
877 	kstat_named_setstr(&pxintr_ks_template.pxintr_ks_buspath, ih_buspath);
878 
879 	if (ih_p->ih_intr_state == PX_INTR_STATE_ENABLE) {
880 
881 		switch (i_ddi_intr_get_current_type(ih_p->ih_dip)) {
882 		case DDI_INTR_TYPE_MSI:
883 			(void) strcpy(pxintr_ks_template.pxintr_ks_type.value.c,
884 			    "msi");
885 			break;
886 		case DDI_INTR_TYPE_MSIX:
887 			(void) strcpy(pxintr_ks_template.pxintr_ks_type.value.c,
888 			    "msix");
889 			break;
890 		default:
891 			(void) strcpy(pxintr_ks_template.pxintr_ks_type.value.c,
892 			    "fixed");
893 			break;
894 		}
895 
896 		pxintr_ks_template.pxintr_ks_cpu.value.ui64 = ino_p->ino_cpuid;
897 		pxintr_ks_template.pxintr_ks_pil.value.ui64 = ipil_p->ipil_pil;
898 		pxintr_ks_template.pxintr_ks_time.value.ui64 = ih_p->ih_nsec +
899 		    (uint64_t)tick2ns((hrtime_t)ih_p->ih_ticks,
900 		    ino_p->ino_cpuid);
901 		pxintr_ks_template.pxintr_ks_ino.value.ui64 = ino;
902 		pxintr_ks_template.pxintr_ks_cookie.value.ui64 = sysino;
903 	} else {
904 		(void) strcpy(pxintr_ks_template.pxintr_ks_type.value.c,
905 		    "disabled");
906 		pxintr_ks_template.pxintr_ks_cpu.value.ui64 = 0;
907 		pxintr_ks_template.pxintr_ks_pil.value.ui64 = 0;
908 		pxintr_ks_template.pxintr_ks_time.value.ui64 = 0;
909 		pxintr_ks_template.pxintr_ks_ino.value.ui64 = 0;
910 		pxintr_ks_template.pxintr_ks_cookie.value.ui64 = 0;
911 	}
912 	return (0);
913 }
914 
915 void
916 px_create_intr_kstats(px_ih_t *ih_p)
917 {
918 	msiq_rec_type_t rec_type = ih_p->ih_rec_type;
919 
920 	ASSERT(ih_p->ih_ksp == NULL);
921 
922 	/*
923 	 * Create pci_intrs::: kstats for all ih types except messages,
924 	 * which represent unusual conditions and don't need to be tracked.
925 	 */
926 	if (rec_type == 0 || rec_type == MSI32_REC || rec_type == MSI64_REC) {
927 		ih_p->ih_ksp = kstat_create("pci_intrs",
928 		    atomic_inc_32_nv(&pxintr_ks_instance), "config",
929 		    "interrupts", KSTAT_TYPE_NAMED,
930 		    sizeof (pxintr_ks_template) / sizeof (kstat_named_t),
931 		    KSTAT_FLAG_VIRTUAL);
932 	}
933 	if (ih_p->ih_ksp != NULL) {
934 		ih_p->ih_ksp->ks_data_size += MAXPATHLEN * 2;
935 		ih_p->ih_ksp->ks_lock = &pxintr_ks_template_lock;
936 		ih_p->ih_ksp->ks_data = &pxintr_ks_template;
937 		ih_p->ih_ksp->ks_private = ih_p;
938 		ih_p->ih_ksp->ks_update = px_ks_update;
939 	}
940 }
941 
942 /*
943  * px_add_intx_intr:
944  *
945  * This function is called to register INTx and legacy hardware
946  * interrupt pins interrupts.
947  */
948 int
949 px_add_intx_intr(dev_info_t *dip, dev_info_t *rdip,
950     ddi_intr_handle_impl_t *hdlp)
951 {
952 	px_t		*px_p = INST_TO_STATE(ddi_get_instance(dip));
953 	px_ib_t		*ib_p = px_p->px_ib_p;
954 	devino_t	ino;
955 	px_ih_t		*ih_p;
956 	px_ino_t	*ino_p;
957 	px_ino_pil_t	*ipil_p, *ipil_list;
958 	int32_t		weight;
959 	int		ret = DDI_SUCCESS;
960 
961 	ino = hdlp->ih_vector;
962 
963 	DBG(DBG_A_INTX, dip, "px_add_intx_intr: rdip=%s%d ino=%x "
964 	    "handler=%x arg1=%x arg2=%x\n", ddi_driver_name(rdip),
965 	    ddi_get_instance(rdip), ino, hdlp->ih_cb_func,
966 	    hdlp->ih_cb_arg1, hdlp->ih_cb_arg2);
967 
968 	ih_p = px_ib_alloc_ih(rdip, hdlp->ih_inum,
969 	    hdlp->ih_cb_func, hdlp->ih_cb_arg1, hdlp->ih_cb_arg2, 0, 0);
970 
971 	mutex_enter(&ib_p->ib_ino_lst_mutex);
972 
973 	ino_p = px_ib_locate_ino(ib_p, ino);
974 	ipil_list = ino_p ? ino_p->ino_ipil_p : NULL;
975 
976 	/* Sharing ino */
977 	if (ino_p && (ipil_p = px_ib_ino_locate_ipil(ino_p, hdlp->ih_pri))) {
978 		if (px_ib_intr_locate_ih(ipil_p, rdip, hdlp->ih_inum, 0, 0)) {
979 			DBG(DBG_A_INTX, dip, "px_add_intx_intr: "
980 			    "dup intr #%d\n", hdlp->ih_inum);
981 
982 			ret = DDI_FAILURE;
983 			goto fail1;
984 		}
985 
986 		/* Save mondo value in hdlp */
987 		hdlp->ih_vector = ino_p->ino_sysino;
988 
989 		if ((ret = px_ib_ino_add_intr(px_p, ipil_p,
990 		    ih_p)) != DDI_SUCCESS)
991 			goto fail1;
992 
993 		goto ino_done;
994 	}
995 
996 	if (hdlp->ih_pri == 0)
997 		hdlp->ih_pri = pci_class_to_pil(rdip);
998 
999 	ipil_p = px_ib_new_ino_pil(ib_p, ino, hdlp->ih_pri, ih_p);
1000 	ino_p = ipil_p->ipil_ino_p;
1001 
1002 	/* Save mondo value in hdlp */
1003 	hdlp->ih_vector = ino_p->ino_sysino;
1004 
1005 	DBG(DBG_A_INTX, dip, "px_add_intx_intr: pil=0x%x mondo=0x%x\n",
1006 	    hdlp->ih_pri, hdlp->ih_vector);
1007 
1008 	DDI_INTR_ASSIGN_HDLR_N_ARGS(hdlp,
1009 	    (ddi_intr_handler_t *)px_intx_intr, (caddr_t)ipil_p, NULL);
1010 
1011 	ret = i_ddi_add_ivintr(hdlp);
1012 
1013 	/*
1014 	 * Restore original interrupt handler
1015 	 * and arguments in interrupt handle.
1016 	 */
1017 	DDI_INTR_ASSIGN_HDLR_N_ARGS(hdlp, ih_p->ih_handler,
1018 	    ih_p->ih_handler_arg1, ih_p->ih_handler_arg2);
1019 
1020 	if (ret != DDI_SUCCESS)
1021 		goto fail2;
1022 
1023 	/* Save the pil for this ino */
1024 	ipil_p->ipil_pil = hdlp->ih_pri;
1025 
1026 	/* Select cpu, saving it for sharing and removal */
1027 	if (ipil_list == NULL) {
1028 		ino_p->ino_cpuid = intr_dist_cpuid();
1029 
1030 		/* Enable interrupt */
1031 		px_ib_intr_enable(px_p, ino_p->ino_cpuid, ino);
1032 	}
1033 
1034 ino_done:
1035 	/* Add weight to the cpu that we are already targeting */
1036 	weight = pci_class_to_intr_weight(rdip);
1037 	intr_dist_cpuid_add_device_weight(ino_p->ino_cpuid, rdip, weight);
1038 
1039 	ih_p->ih_ipil_p = ipil_p;
1040 	px_create_intr_kstats(ih_p);
1041 	if (ih_p->ih_ksp)
1042 		kstat_install(ih_p->ih_ksp);
1043 	mutex_exit(&ib_p->ib_ino_lst_mutex);
1044 
1045 	DBG(DBG_A_INTX, dip, "px_add_intx_intr: done! Interrupt 0x%x pil=%x\n",
1046 	    ino_p->ino_sysino, hdlp->ih_pri);
1047 
1048 	return (ret);
1049 fail2:
1050 	px_ib_delete_ino_pil(ib_p, ipil_p);
1051 fail1:
1052 	if (ih_p->ih_config_handle)
1053 		pci_config_teardown(&ih_p->ih_config_handle);
1054 
1055 	mutex_exit(&ib_p->ib_ino_lst_mutex);
1056 	kmem_free(ih_p, sizeof (px_ih_t));
1057 
1058 	DBG(DBG_A_INTX, dip, "px_add_intx_intr: Failed! Interrupt 0x%x "
1059 	    "pil=%x\n", ino_p->ino_sysino, hdlp->ih_pri);
1060 
1061 	return (ret);
1062 }
1063 
1064 /*
1065  * px_rem_intx_intr:
1066  *
1067  * This function is called to unregister INTx and legacy hardware
1068  * interrupt pins interrupts.
1069  */
1070 int
1071 px_rem_intx_intr(dev_info_t *dip, dev_info_t *rdip,
1072     ddi_intr_handle_impl_t *hdlp)
1073 {
1074 	px_t		*px_p = INST_TO_STATE(ddi_get_instance(dip));
1075 	px_ib_t		*ib_p = px_p->px_ib_p;
1076 	devino_t	ino;
1077 	cpuid_t		curr_cpu;
1078 	px_ino_t	*ino_p;
1079 	px_ino_pil_t	*ipil_p;
1080 	px_ih_t		*ih_p;
1081 	int		ret = DDI_SUCCESS;
1082 
1083 	ino = hdlp->ih_vector;
1084 
1085 	DBG(DBG_R_INTX, dip, "px_rem_intx_intr: rdip=%s%d ino=%x\n",
1086 	    ddi_driver_name(rdip), ddi_get_instance(rdip), ino);
1087 
1088 	mutex_enter(&ib_p->ib_ino_lst_mutex);
1089 
1090 	ino_p = px_ib_locate_ino(ib_p, ino);
1091 	ipil_p = px_ib_ino_locate_ipil(ino_p, hdlp->ih_pri);
1092 	ih_p = px_ib_intr_locate_ih(ipil_p, rdip, hdlp->ih_inum, 0, 0);
1093 
1094 	/* Get the current cpu */
1095 	if ((ret = px_lib_intr_gettarget(px_p->px_dip, ino_p->ino_sysino,
1096 	    &curr_cpu)) != DDI_SUCCESS)
1097 		goto fail;
1098 
1099 	if ((ret = px_ib_ino_rem_intr(px_p, ipil_p, ih_p)) != DDI_SUCCESS)
1100 		goto fail;
1101 
1102 	intr_dist_cpuid_rem_device_weight(ino_p->ino_cpuid, rdip);
1103 
1104 	if (ipil_p->ipil_ih_size == 0) {
1105 		hdlp->ih_vector = ino_p->ino_sysino;
1106 		i_ddi_rem_ivintr(hdlp);
1107 
1108 		px_ib_delete_ino_pil(ib_p, ipil_p);
1109 	}
1110 
1111 	if (ino_p->ino_ipil_size == 0) {
1112 		kmem_free(ino_p, sizeof (px_ino_t));
1113 	} else {
1114 		/* Re-enable interrupt only if mapping register still shared */
1115 		PX_INTR_ENABLE(px_p->px_dip, ino_p->ino_sysino, curr_cpu);
1116 	}
1117 
1118 fail:
1119 	mutex_exit(&ib_p->ib_ino_lst_mutex);
1120 	return (ret);
1121 }
1122 
1123 /*
1124  * px_add_msiq_intr:
1125  *
1126  * This function is called to register MSI/Xs and PCIe message interrupts.
1127  */
1128 int
1129 px_add_msiq_intr(dev_info_t *dip, dev_info_t *rdip,
1130     ddi_intr_handle_impl_t *hdlp, msiq_rec_type_t rec_type,
1131     msgcode_t msg_code, msiqid_t *msiq_id_p)
1132 {
1133 	px_t		*px_p = INST_TO_STATE(ddi_get_instance(dip));
1134 	px_ib_t		*ib_p = px_p->px_ib_p;
1135 	px_msiq_state_t	*msiq_state_p = &ib_p->ib_msiq_state;
1136 	devino_t	ino;
1137 	px_ih_t		*ih_p;
1138 	px_ino_t	*ino_p;
1139 	px_ino_pil_t	*ipil_p, *ipil_list;
1140 	int32_t		weight;
1141 	int		ret = DDI_SUCCESS;
1142 
1143 	DBG(DBG_MSIQ, dip, "px_add_msiq_intr: rdip=%s%d handler=%x "
1144 	    "arg1=%x arg2=%x\n", ddi_driver_name(rdip), ddi_get_instance(rdip),
1145 	    hdlp->ih_cb_func, hdlp->ih_cb_arg1, hdlp->ih_cb_arg2);
1146 
1147 	if ((ret = px_msiq_alloc(px_p, rec_type, msiq_id_p)) != DDI_SUCCESS) {
1148 		DBG(DBG_MSIQ, dip, "px_add_msiq_intr: "
1149 		    "msiq allocation failed\n");
1150 		return (ret);
1151 	}
1152 
1153 	ino = px_msiqid_to_devino(px_p, *msiq_id_p);
1154 
1155 	ih_p = px_ib_alloc_ih(rdip, hdlp->ih_inum, hdlp->ih_cb_func,
1156 	    hdlp->ih_cb_arg1, hdlp->ih_cb_arg2, rec_type, msg_code);
1157 
1158 	mutex_enter(&ib_p->ib_ino_lst_mutex);
1159 
1160 	ino_p = px_ib_locate_ino(ib_p, ino);
1161 	ipil_list = ino_p ? ino_p->ino_ipil_p : NULL;
1162 
1163 	/* Sharing ino */
1164 	if (ino_p && (ipil_p = px_ib_ino_locate_ipil(ino_p, hdlp->ih_pri))) {
1165 		if (px_ib_intr_locate_ih(ipil_p, rdip,
1166 		    hdlp->ih_inum, rec_type, msg_code)) {
1167 			DBG(DBG_MSIQ, dip, "px_add_msiq_intr: "
1168 			    "dup intr #%d\n", hdlp->ih_inum);
1169 
1170 			ret = DDI_FAILURE;
1171 			goto fail1;
1172 		}
1173 
1174 		/* Save mondo value in hdlp */
1175 		hdlp->ih_vector = ino_p->ino_sysino;
1176 
1177 		if ((ret = px_ib_ino_add_intr(px_p, ipil_p,
1178 		    ih_p)) != DDI_SUCCESS)
1179 			goto fail1;
1180 
1181 		goto ino_done;
1182 	}
1183 
1184 	if (hdlp->ih_pri == 0)
1185 		hdlp->ih_pri = pci_class_to_pil(rdip);
1186 
1187 	ipil_p = px_ib_new_ino_pil(ib_p, ino, hdlp->ih_pri, ih_p);
1188 	ino_p = ipil_p->ipil_ino_p;
1189 
1190 	ino_p->ino_msiq_p = msiq_state_p->msiq_p +
1191 	    (*msiq_id_p - msiq_state_p->msiq_1st_msiq_id);
1192 
1193 	/* Save mondo value in hdlp */
1194 	hdlp->ih_vector = ino_p->ino_sysino;
1195 
1196 	DBG(DBG_MSIQ, dip, "px_add_msiq_intr: pil=0x%x mondo=0x%x\n",
1197 	    hdlp->ih_pri, hdlp->ih_vector);
1198 
1199 	DDI_INTR_ASSIGN_HDLR_N_ARGS(hdlp,
1200 	    (ddi_intr_handler_t *)px_msiq_intr, (caddr_t)ipil_p, NULL);
1201 
1202 	ret = i_ddi_add_ivintr(hdlp);
1203 
1204 	/*
1205 	 * Restore original interrupt handler
1206 	 * and arguments in interrupt handle.
1207 	 */
1208 	DDI_INTR_ASSIGN_HDLR_N_ARGS(hdlp, ih_p->ih_handler,
1209 	    ih_p->ih_handler_arg1, ih_p->ih_handler_arg2);
1210 
1211 	if (ret != DDI_SUCCESS)
1212 		goto fail2;
1213 
1214 	/* Save the pil for this ino */
1215 	ipil_p->ipil_pil = hdlp->ih_pri;
1216 
1217 	/* Select cpu, saving it for sharing and removal */
1218 	if (ipil_list == NULL) {
1219 		ino_p->ino_cpuid = intr_dist_cpuid();
1220 
1221 		/* Enable MSIQ */
1222 		px_lib_msiq_setstate(dip, *msiq_id_p, PCI_MSIQ_STATE_IDLE);
1223 		px_lib_msiq_setvalid(dip, *msiq_id_p, PCI_MSIQ_VALID);
1224 
1225 		/* Enable interrupt */
1226 		px_ib_intr_enable(px_p, ino_p->ino_cpuid, ino);
1227 	}
1228 
1229 ino_done:
1230 	/* Add weight to the cpu that we are already targeting */
1231 	weight = pci_class_to_intr_weight(rdip);
1232 	intr_dist_cpuid_add_device_weight(ino_p->ino_cpuid, rdip, weight);
1233 
1234 	ih_p->ih_ipil_p = ipil_p;
1235 	px_create_intr_kstats(ih_p);
1236 	if (ih_p->ih_ksp)
1237 		kstat_install(ih_p->ih_ksp);
1238 	mutex_exit(&ib_p->ib_ino_lst_mutex);
1239 
1240 	DBG(DBG_MSIQ, dip, "px_add_msiq_intr: done! Interrupt 0x%x pil=%x\n",
1241 	    ino_p->ino_sysino, hdlp->ih_pri);
1242 
1243 	return (ret);
1244 fail2:
1245 	px_ib_delete_ino_pil(ib_p, ipil_p);
1246 fail1:
1247 	if (ih_p->ih_config_handle)
1248 		pci_config_teardown(&ih_p->ih_config_handle);
1249 
1250 	mutex_exit(&ib_p->ib_ino_lst_mutex);
1251 	kmem_free(ih_p, sizeof (px_ih_t));
1252 
1253 	DBG(DBG_MSIQ, dip, "px_add_msiq_intr: Failed! Interrupt 0x%x pil=%x\n",
1254 	    ino_p->ino_sysino, hdlp->ih_pri);
1255 
1256 	return (ret);
1257 }
1258 
1259 /*
1260  * px_rem_msiq_intr:
1261  *
1262  * This function is called to unregister MSI/Xs and PCIe message interrupts.
1263  */
1264 int
1265 px_rem_msiq_intr(dev_info_t *dip, dev_info_t *rdip,
1266     ddi_intr_handle_impl_t *hdlp, msiq_rec_type_t rec_type,
1267     msgcode_t msg_code, msiqid_t msiq_id)
1268 {
1269 	px_t		*px_p = INST_TO_STATE(ddi_get_instance(dip));
1270 	px_ib_t		*ib_p = px_p->px_ib_p;
1271 	devino_t	ino = px_msiqid_to_devino(px_p, msiq_id);
1272 	cpuid_t		curr_cpu;
1273 	px_ino_t	*ino_p;
1274 	px_ino_pil_t	*ipil_p;
1275 	px_ih_t		*ih_p;
1276 	int		ret = DDI_SUCCESS;
1277 
1278 	DBG(DBG_MSIQ, dip, "px_rem_msiq_intr: rdip=%s%d msiq_id=%x ino=%x\n",
1279 	    ddi_driver_name(rdip), ddi_get_instance(rdip), msiq_id, ino);
1280 
1281 	mutex_enter(&ib_p->ib_ino_lst_mutex);
1282 
1283 	ino_p = px_ib_locate_ino(ib_p, ino);
1284 	ipil_p = px_ib_ino_locate_ipil(ino_p, hdlp->ih_pri);
1285 	ih_p = px_ib_intr_locate_ih(ipil_p, rdip, hdlp->ih_inum, rec_type,
1286 	    msg_code);
1287 
1288 	/* Get the current cpu */
1289 	if ((ret = px_lib_intr_gettarget(px_p->px_dip, ino_p->ino_sysino,
1290 	    &curr_cpu)) != DDI_SUCCESS)
1291 		goto fail;
1292 
1293 	if ((ret = px_ib_ino_rem_intr(px_p, ipil_p, ih_p)) != DDI_SUCCESS)
1294 		goto fail;
1295 
1296 	intr_dist_cpuid_rem_device_weight(ino_p->ino_cpuid, rdip);
1297 
1298 	if (ipil_p->ipil_ih_size == 0) {
1299 		hdlp->ih_vector = ino_p->ino_sysino;
1300 		i_ddi_rem_ivintr(hdlp);
1301 
1302 		px_ib_delete_ino_pil(ib_p, ipil_p);
1303 
1304 		if (ino_p->ino_ipil_size == 0)
1305 			px_lib_msiq_setvalid(dip,
1306 			    px_devino_to_msiqid(px_p, ino), PCI_MSIQ_INVALID);
1307 
1308 		(void) px_msiq_free(px_p, msiq_id);
1309 	}
1310 
1311 	if (ino_p->ino_ipil_size == 0) {
1312 		kmem_free(ino_p, sizeof (px_ino_t));
1313 	} else {
1314 		/* Re-enable interrupt only if mapping register still shared */
1315 		PX_INTR_ENABLE(px_p->px_dip, ino_p->ino_sysino, curr_cpu);
1316 	}
1317 
1318 fail:
1319 	mutex_exit(&ib_p->ib_ino_lst_mutex);
1320 	return (ret);
1321 }
1322