xref: /titanic_50/usr/src/uts/sun4/io/px/px_intr.c (revision 0aa3cd4d26810aec2570d1529242ae8c446b6a5d)
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 2007 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 /*
29  * PX nexus interrupt handling:
30  *	PX device interrupt handler wrapper
31  *	PIL lookup routine
32  *	PX device interrupt related initchild code
33  */
34 
35 #include <sys/types.h>
36 #include <sys/kmem.h>
37 #include <sys/async.h>
38 #include <sys/spl.h>
39 #include <sys/sunddi.h>
40 #include <sys/fm/protocol.h>
41 #include <sys/fm/util.h>
42 #include <sys/machsystm.h>	/* e_ddi_nodeid_to_dip() */
43 #include <sys/ddi_impldefs.h>
44 #include <sys/sdt.h>
45 #include <sys/atomic.h>
46 #include "px_obj.h"
47 #include <sys/ontrap.h>
48 #include <sys/membar.h>
49 #include <sys/clock.h>
50 
51 /*
52  * interrupt jabber:
53  *
54  * When an interrupt line is jabbering, every time the state machine for the
55  * associated ino is idled, a new mondo will be sent and the ino will go into
56  * the pending state again. The mondo will cause a new call to
57  * px_intr_wrapper() which normally idles the ino's state machine which would
58  * precipitate another trip round the loop.
59  *
60  * The loop can be broken by preventing the ino's state machine from being
61  * idled when an interrupt line is jabbering. See the comment at the
62  * beginning of px_intr_wrapper() explaining how the 'interrupt jabber
63  * protection' code does this.
64  */
65 
66 /*LINTLIBRARY*/
67 
68 /*
69  * If the unclaimed interrupt count has reached the limit set by
70  * pci_unclaimed_intr_max within the time limit, then all interrupts
71  * on this ino is blocked by not idling the interrupt state machine.
72  */
73 static int
74 px_spurintr(px_ino_pil_t *ipil_p)
75 {
76 	px_ino_t	*ino_p = ipil_p->ipil_ino_p;
77 	px_ih_t		*ih_p = ipil_p->ipil_ih_start;
78 	px_t		*px_p = ino_p->ino_ib_p->ib_px_p;
79 	char		*err_fmt_str;
80 	boolean_t	blocked = B_FALSE;
81 	int		i;
82 
83 	if (ino_p->ino_unclaimed_intrs > px_unclaimed_intr_max)
84 		return (DDI_INTR_CLAIMED);
85 
86 	if (!ino_p->ino_unclaimed_intrs)
87 		ino_p->ino_spurintr_begin = ddi_get_lbolt();
88 
89 	ino_p->ino_unclaimed_intrs++;
90 
91 	if (ino_p->ino_unclaimed_intrs <= px_unclaimed_intr_max)
92 		goto clear;
93 
94 	if (drv_hztousec(ddi_get_lbolt() - ino_p->ino_spurintr_begin)
95 	    > px_spurintr_duration) {
96 		ino_p->ino_unclaimed_intrs = 0;
97 		goto clear;
98 	}
99 	err_fmt_str = "%s%d: ino 0x%x blocked";
100 	blocked = B_TRUE;
101 	goto warn;
102 clear:
103 	err_fmt_str = "!%s%d: spurious interrupt from ino 0x%x";
104 warn:
105 	cmn_err(CE_WARN, err_fmt_str, NAMEINST(px_p->px_dip), ino_p->ino_ino);
106 	for (i = 0; i < ipil_p->ipil_ih_size; i++, ih_p = ih_p->ih_next)
107 		cmn_err(CE_CONT, "!%s-%d#%x ", NAMEINST(ih_p->ih_dip),
108 		    ih_p->ih_inum);
109 	cmn_err(CE_CONT, "!\n");
110 
111 	/* Clear the pending state */
112 	if (blocked == B_FALSE) {
113 		if (px_lib_intr_setstate(px_p->px_dip, ino_p->ino_sysino,
114 		    INTR_IDLE_STATE) != DDI_SUCCESS)
115 			return (DDI_INTR_UNCLAIMED);
116 	}
117 
118 	return (DDI_INTR_CLAIMED);
119 }
120 
121 extern uint64_t intr_get_time(void);
122 
123 /*
124  * px_intx_intr (INTx or legacy interrupt handler)
125  *
126  * This routine is used as wrapper around interrupt handlers installed by child
127  * device drivers.  This routine invokes the driver interrupt handlers and
128  * examines the return codes.
129  *
130  * There is a count of unclaimed interrupts kept on a per-ino basis. If at
131  * least one handler claims the interrupt then the counter is halved and the
132  * interrupt state machine is idled. If no handler claims the interrupt then
133  * the counter is incremented by one and the state machine is idled.
134  * If the count ever reaches the limit value set by pci_unclaimed_intr_max
135  * then the interrupt state machine is not idled thus preventing any further
136  * interrupts on that ino. The state machine will only be idled again if a
137  * handler is subsequently added or removed.
138  *
139  * return value: DDI_INTR_CLAIMED if any handlers claimed the interrupt,
140  * DDI_INTR_UNCLAIMED otherwise.
141  */
142 uint_t
143 px_intx_intr(caddr_t arg)
144 {
145 	px_ino_pil_t	*ipil_p = (px_ino_pil_t *)arg;
146 	px_ino_t	*ino_p = ipil_p->ipil_ino_p;
147 	px_t		*px_p = ino_p->ino_ib_p->ib_px_p;
148 	px_ih_t		*ih_p = ipil_p->ipil_ih_start;
149 	ushort_t	pil = ipil_p->ipil_pil;
150 	uint_t		result = 0, r = DDI_INTR_UNCLAIMED;
151 	int		i;
152 
153 	DBG(DBG_INTX_INTR, px_p->px_dip, "px_intx_intr:"
154 	    "ino=%x sysino=%llx pil=%x ih_size=%x ih_lst=%x\n",
155 	    ino_p->ino_ino, ino_p->ino_sysino, ipil_p->ipil_pil,
156 	    ipil_p->ipil_ih_size, ipil_p->ipil_ih_head);
157 
158 	for (i = 0; i < ipil_p->ipil_ih_size; i++, ih_p = ih_p->ih_next) {
159 		dev_info_t *dip = ih_p->ih_dip;
160 		uint_t (*handler)() = ih_p->ih_handler;
161 		caddr_t arg1 = ih_p->ih_handler_arg1;
162 		caddr_t arg2 = ih_p->ih_handler_arg2;
163 
164 		if (ih_p->ih_intr_state == PX_INTR_STATE_DISABLE) {
165 			DBG(DBG_INTX_INTR, px_p->px_dip,
166 			    "px_intx_intr: %s%d interrupt %d is disabled\n",
167 			    ddi_driver_name(dip), ddi_get_instance(dip),
168 			    ino_p->ino_ino);
169 
170 			continue;
171 		}
172 
173 		DBG(DBG_INTX_INTR, px_p->px_dip, "px_intx_intr:"
174 		    "ino=%x handler=%p arg1 =%p arg2 = %p\n",
175 		    ino_p->ino_ino, handler, arg1, arg2);
176 
177 		DTRACE_PROBE4(interrupt__start, dev_info_t, dip,
178 		    void *, handler, caddr_t, arg1, caddr_t, arg2);
179 
180 		r = (*handler)(arg1, arg2);
181 
182 		/*
183 		 * Account for time used by this interrupt. Protect against
184 		 * conflicting writes to ih_ticks from ib_intr_dist_all() by
185 		 * using atomic ops.
186 		 */
187 
188 		if (pil <= LOCK_LEVEL)
189 			atomic_add_64(&ih_p->ih_ticks, intr_get_time());
190 
191 		DTRACE_PROBE4(interrupt__complete, dev_info_t, dip,
192 		    void *, handler, caddr_t, arg1, int, r);
193 
194 		result += r;
195 
196 		if (px_check_all_handlers)
197 			continue;
198 		if (result)
199 			break;
200 	}
201 
202 	if (result)
203 		ino_p->ino_claimed |= (1 << pil);
204 
205 	/* Interrupt can only be cleared after all pil levels are handled */
206 	if (pil != ino_p->ino_lopil)
207 		return (DDI_INTR_CLAIMED);
208 
209 	if (!ino_p->ino_claimed) {
210 		if (px_unclaimed_intr_block)
211 			return (px_spurintr(ipil_p));
212 	}
213 
214 	ino_p->ino_unclaimed_intrs = 0;
215 	ino_p->ino_claimed = 0;
216 
217 	/* Clear the pending state */
218 	if (px_lib_intr_setstate(px_p->px_dip,
219 	    ino_p->ino_sysino, INTR_IDLE_STATE) != DDI_SUCCESS)
220 		return (DDI_INTR_UNCLAIMED);
221 
222 	return (DDI_INTR_CLAIMED);
223 }
224 
225 /*
226  * px_msiq_intr (MSI/X or PCIe MSG interrupt handler)
227  *
228  * This routine is used as wrapper around interrupt handlers installed by child
229  * device drivers.  This routine invokes the driver interrupt handlers and
230  * examines the return codes.
231  *
232  * There is a count of unclaimed interrupts kept on a per-ino basis. If at
233  * least one handler claims the interrupt then the counter is halved and the
234  * interrupt state machine is idled. If no handler claims the interrupt then
235  * the counter is incremented by one and the state machine is idled.
236  * If the count ever reaches the limit value set by pci_unclaimed_intr_max
237  * then the interrupt state machine is not idled thus preventing any further
238  * interrupts on that ino. The state machine will only be idled again if a
239  * handler is subsequently added or removed.
240  *
241  * return value: DDI_INTR_CLAIMED if any handlers claimed the interrupt,
242  * DDI_INTR_UNCLAIMED otherwise.
243  */
244 uint_t
245 px_msiq_intr(caddr_t arg)
246 {
247 	px_ino_pil_t	*ipil_p = (px_ino_pil_t *)arg;
248 	px_ino_t	*ino_p = ipil_p->ipil_ino_p;
249 	px_t		*px_p = ino_p->ino_ib_p->ib_px_p;
250 	px_msiq_state_t	*msiq_state_p = &px_p->px_ib_p->ib_msiq_state;
251 	px_msiq_t	*msiq_p = ino_p->ino_msiq_p;
252 	dev_info_t	*dip = px_p->px_dip;
253 	ushort_t	pil = ipil_p->ipil_pil;
254 	msiq_rec_t	msiq_rec, *msiq_rec_p = &msiq_rec;
255 	msiqhead_t	*curr_head_p;
256 	msiqtail_t	curr_tail_index;
257 	msgcode_t	msg_code;
258 	px_ih_t		*ih_p;
259 	uint_t		ret = DDI_INTR_UNCLAIMED;
260 	int		i, j;
261 
262 	DBG(DBG_MSIQ_INTR, dip, "px_msiq_intr: msiq_id =%x ino=%x pil=%x "
263 	    "ih_size=%x ih_lst=%x\n", msiq_p->msiq_id, ino_p->ino_ino,
264 	    ipil_p->ipil_pil, ipil_p->ipil_ih_size, ipil_p->ipil_ih_head);
265 
266 	/*
267 	 * The px_msiq_intr() handles multiple interrupt priorities and it
268 	 * will set msiq->msiq_rec2process to the number of MSIQ records to
269 	 * process while handling the highest priority interrupt. Subsequent
270 	 * lower priority interrupts will just process any unprocessed MSIQ
271 	 * records or will just return immediately.
272 	 */
273 	if (msiq_p->msiq_recs2process == 0) {
274 		/* Read current MSIQ tail index */
275 		px_lib_msiq_gettail(dip, msiq_p->msiq_id, &curr_tail_index);
276 		msiq_p->msiq_new_head_index = msiq_p->msiq_curr_head_index;
277 
278 		if (curr_tail_index < msiq_p->msiq_curr_head_index)
279 			curr_tail_index += msiq_state_p->msiq_rec_cnt;
280 
281 		msiq_p->msiq_recs2process = curr_tail_index -
282 		    msiq_p->msiq_curr_head_index;
283 	}
284 
285 	DBG(DBG_MSIQ_INTR, dip, "px_msiq_intr: curr_head %x new_head %x "
286 	    "rec2process %x\n", msiq_p->msiq_curr_head_index,
287 	    msiq_p->msiq_new_head_index, msiq_p->msiq_recs2process);
288 
289 	/* If all MSIQ records are already processed, just return immediately */
290 	if ((msiq_p->msiq_new_head_index - msiq_p->msiq_curr_head_index)
291 	    == msiq_p->msiq_recs2process)
292 		goto intr_done;
293 
294 	curr_head_p = (msiqhead_t *)((caddr_t)msiq_p->msiq_base_p +
295 	    (msiq_p->msiq_curr_head_index * sizeof (msiq_rec_t)));
296 
297 	/*
298 	 * Calculate the number of recs to process by taking the difference
299 	 * between the head and tail pointers. For all records we always
300 	 * verify that we have a valid record type before we do any processing.
301 	 * If triggered, we should always have at least one valid record.
302 	 */
303 	for (i = 0; i < msiq_p->msiq_recs2process; i++) {
304 		/* Read next MSIQ record */
305 		px_lib_get_msiq_rec(dip, curr_head_p, msiq_rec_p);
306 
307 		DBG(DBG_MSIQ_INTR, dip, "px_msiq_intr: MSIQ RECORD, "
308 		    "msiq_rec_type 0x%llx msiq_rec_rid 0x%llx\n",
309 		    msiq_rec_p->msiq_rec_type, msiq_rec_p->msiq_rec_rid);
310 
311 		if (!msiq_rec_p->msiq_rec_type)
312 			goto next_rec;
313 
314 		/* Check MSIQ record type */
315 		switch (msiq_rec_p->msiq_rec_type) {
316 		case MSG_REC:
317 			msg_code = msiq_rec_p->msiq_rec_data.msg.msg_code;
318 			DBG(DBG_MSIQ_INTR, dip, "px_msiq_intr: PCIE MSG "
319 			    "record, msg type 0x%x\n", msg_code);
320 			break;
321 		case MSI32_REC:
322 		case MSI64_REC:
323 			msg_code = msiq_rec_p->msiq_rec_data.msi.msi_data;
324 			DBG(DBG_MSIQ_INTR, dip, "px_msiq_intr: MSI record, "
325 			    "msi 0x%x\n", msg_code);
326 
327 			/* Clear MSI state */
328 			px_lib_msi_setstate(dip, (msinum_t)msg_code,
329 			    PCI_MSI_STATE_IDLE);
330 			break;
331 		default:
332 			msg_code = 0;
333 			cmn_err(CE_WARN, "%s%d: px_msiq_intr: 0x%x MSIQ "
334 			    "record type is not supported",
335 			    ddi_driver_name(dip), ddi_get_instance(dip),
336 			    msiq_rec_p->msiq_rec_type);
337 
338 			goto next_rec;
339 		}
340 
341 		/*
342 		 * Scan through px_ih_t linked list, searching for the
343 		 * right px_ih_t, matching MSIQ record data.
344 		 */
345 		for (j = 0, ih_p = ipil_p->ipil_ih_start;
346 		    ih_p && (j < ipil_p->ipil_ih_size) &&
347 		    ((ih_p->ih_msg_code != msg_code) ||
348 		    (ih_p->ih_rec_type != msiq_rec_p->msiq_rec_type));
349 		    ih_p = ih_p->ih_next, j++);
350 
351 		if ((ih_p->ih_msg_code == msg_code) &&
352 		    (ih_p->ih_rec_type == msiq_rec_p->msiq_rec_type)) {
353 			dev_info_t *dip = ih_p->ih_dip;
354 			uint_t (*handler)() = ih_p->ih_handler;
355 			caddr_t arg1 = ih_p->ih_handler_arg1;
356 			caddr_t arg2 = ih_p->ih_handler_arg2;
357 
358 			DBG(DBG_MSIQ_INTR, dip, "px_msiq_intr: ino=%x data=%x "
359 			    "handler=%p arg1 =%p arg2=%p\n", ino_p->ino_ino,
360 			    msg_code, handler, arg1, arg2);
361 
362 			DTRACE_PROBE4(interrupt__start, dev_info_t, dip,
363 			    void *, handler, caddr_t, arg1, caddr_t, arg2);
364 
365 			/*
366 			 * Special case for PCIE Error Messages.
367 			 * The current frame work doesn't fit PCIE Err Msgs
368 			 * This should be fixed when PCIE MESSAGES as a whole
369 			 * is architected correctly.
370 			 */
371 			if ((msg_code == PCIE_MSG_CODE_ERR_COR) ||
372 			    (msg_code == PCIE_MSG_CODE_ERR_NONFATAL) ||
373 			    (msg_code == PCIE_MSG_CODE_ERR_FATAL)) {
374 				ret = px_err_fabric_intr(px_p, msg_code,
375 				    msiq_rec_p->msiq_rec_rid);
376 			} else
377 				ret = (*handler)(arg1, arg2);
378 
379 			/*
380 			 * Account for time used by this interrupt. Protect
381 			 * against conflicting writes to ih_ticks from
382 			 * ib_intr_dist_all() by using atomic ops.
383 			 */
384 
385 			if (pil <= LOCK_LEVEL)
386 				atomic_add_64(&ih_p->ih_ticks, intr_get_time());
387 
388 			DTRACE_PROBE4(interrupt__complete, dev_info_t, dip,
389 			    void *, handler, caddr_t, arg1, int, ret);
390 
391 			msiq_p->msiq_new_head_index++;
392 			px_lib_clr_msiq_rec(dip, curr_head_p);
393 		} else {
394 			DBG(DBG_MSIQ_INTR, dip, "px_msiq_intr:"
395 			    "No matching MSIQ record found\n");
396 		}
397 next_rec:
398 		/* Get the pointer next EQ record */
399 		curr_head_p = (msiqhead_t *)
400 		    ((caddr_t)curr_head_p + sizeof (msiq_rec_t));
401 
402 		/* Check for overflow condition */
403 		if (curr_head_p >= (msiqhead_t *)((caddr_t)msiq_p->msiq_base_p
404 		    + (msiq_state_p->msiq_rec_cnt * sizeof (msiq_rec_t))))
405 			curr_head_p = (msiqhead_t *)msiq_p->msiq_base_p;
406 	}
407 
408 	DBG(DBG_MSIQ_INTR, dip, "px_msiq_intr: No of MSIQ recs processed %x\n",
409 	    (msiq_p->msiq_new_head_index - msiq_p->msiq_curr_head_index));
410 
411 	DBG(DBG_MSIQ_INTR, dip, "px_msiq_intr: curr_head %x new_head %x "
412 	    "rec2process %x\n", msiq_p->msiq_curr_head_index,
413 	    msiq_p->msiq_new_head_index, msiq_p->msiq_recs2process);
414 
415 	/* ino_claimed used just for debugging purpose */
416 	if (ret)
417 		ino_p->ino_claimed |= (1 << pil);
418 
419 intr_done:
420 	/* Interrupt can only be cleared after all pil levels are handled */
421 	if (pil != ino_p->ino_lopil)
422 		return (DDI_INTR_CLAIMED);
423 
424 	if (msiq_p->msiq_new_head_index <= msiq_p->msiq_curr_head_index)  {
425 		if (px_unclaimed_intr_block)
426 			return (px_spurintr(ipil_p));
427 	}
428 
429 	/*  Update MSIQ head index with no of MSIQ records processed */
430 	if (msiq_p->msiq_new_head_index >= msiq_state_p->msiq_rec_cnt)
431 		msiq_p->msiq_new_head_index -= msiq_state_p->msiq_rec_cnt;
432 
433 	msiq_p->msiq_curr_head_index = msiq_p->msiq_new_head_index;
434 	px_lib_msiq_sethead(dip, msiq_p->msiq_id, msiq_p->msiq_new_head_index);
435 
436 	msiq_p->msiq_new_head_index = 0;
437 	msiq_p->msiq_recs2process = 0;
438 	ino_p->ino_claimed = 0;
439 
440 	/* Clear the pending state */
441 	if (px_lib_intr_setstate(dip, ino_p->ino_sysino,
442 	    INTR_IDLE_STATE) != DDI_SUCCESS)
443 		return (DDI_INTR_UNCLAIMED);
444 
445 	return (DDI_INTR_CLAIMED);
446 }
447 
448 dev_info_t *
449 px_get_my_childs_dip(dev_info_t *dip, dev_info_t *rdip)
450 {
451 	dev_info_t	*cdip = rdip;
452 
453 	for (; ddi_get_parent(cdip) != dip; cdip = ddi_get_parent(cdip))
454 		;
455 
456 	return (cdip);
457 }
458 
459 /* Default class to pil value mapping */
460 px_class_val_t px_default_pil [] = {
461 	{0x000000, 0xff0000, 0x1},	/* Class code for pre-2.0 devices */
462 	{0x010000, 0xff0000, 0x4},	/* Mass Storage Controller */
463 	{0x020000, 0xff0000, 0x6},	/* Network Controller */
464 	{0x030000, 0xff0000, 0x9},	/* Display Controller */
465 	{0x040000, 0xff0000, 0x9},	/* Multimedia Controller */
466 	{0x050000, 0xff0000, 0x9},	/* Memory Controller */
467 	{0x060000, 0xff0000, 0x9},	/* Bridge Controller */
468 	{0x0c0000, 0xffff00, 0x9},	/* Serial Bus, FireWire (IEEE 1394) */
469 	{0x0c0100, 0xffff00, 0x4},	/* Serial Bus, ACCESS.bus */
470 	{0x0c0200, 0xffff00, 0x4},	/* Serial Bus, SSA */
471 	{0x0c0300, 0xffff00, 0x9},	/* Serial Bus Universal Serial Bus */
472 	{0x0c0400, 0xffff00, 0x6},	/* Serial Bus, Fibre Channel */
473 	{0x0c0600, 0xffff00, 0x6}	/* Serial Bus, Infiniband */
474 };
475 
476 /*
477  * Default class to intr_weight value mapping (% of CPU).  A driver.conf
478  * entry on or above the pci node like
479  *
480  *	pci-class-intr-weights= 0x020000, 0xff0000, 30;
481  *
482  * can be used to augment or override entries in the default table below.
483  *
484  * NB: The values below give NICs preference on redistribution, and provide
485  * NICs some isolation from other interrupt sources. We need better interfaces
486  * that allow the NIC driver to identify a specific NIC instance as high
487  * bandwidth, and thus deserving of separation from other low bandwidth
488  * NICs additional isolation from other interrupt sources.
489  *
490  * NB: We treat Infiniband like a NIC.
491  */
492 px_class_val_t px_default_intr_weight [] = {
493 	{0x020000, 0xff0000, 35},	/* Network Controller */
494 	{0x010000, 0xff0000, 10},	/* Mass Storage Controller */
495 	{0x0c0400, 0xffff00, 10},	/* Serial Bus, Fibre Channel */
496 	{0x0c0600, 0xffff00, 50}	/* Serial Bus, Infiniband */
497 };
498 
499 static uint32_t
500 px_match_class_val(uint32_t key, px_class_val_t *rec_p, int nrec,
501     uint32_t default_val)
502 {
503 	int	i;
504 
505 	for (i = 0; i < nrec; rec_p++, i++) {
506 		if ((rec_p->class_code & rec_p->class_mask) ==
507 		    (key & rec_p->class_mask))
508 			return (rec_p->class_val);
509 	}
510 
511 	return (default_val);
512 }
513 
514 /*
515  * px_class_to_val
516  *
517  * Return the configuration value, based on class code and sub class code,
518  * from the specified property based or default px_class_val_t table.
519  */
520 uint32_t
521 px_class_to_val(dev_info_t *rdip, char *property_name, px_class_val_t *rec_p,
522     int nrec, uint32_t default_val)
523 {
524 	int property_len;
525 	uint32_t class_code;
526 	px_class_val_t *conf;
527 	uint32_t val = default_val;
528 
529 	/*
530 	 * Use the "class-code" property to get the base and sub class
531 	 * codes for the requesting device.
532 	 */
533 	class_code = (uint32_t)ddi_prop_get_int(DDI_DEV_T_ANY, rdip,
534 	    DDI_PROP_DONTPASS, "class-code", -1);
535 
536 	if (class_code == -1)
537 		return (val);
538 
539 	/* look up the val from the default table */
540 	val = px_match_class_val(class_code, rec_p, nrec, val);
541 
542 	/* see if there is a more specific property specified value */
543 	if (ddi_getlongprop(DDI_DEV_T_ANY, rdip, DDI_PROP_NOTPROM,
544 	    property_name, (caddr_t)&conf, &property_len))
545 		return (val);
546 
547 	if ((property_len % sizeof (px_class_val_t)) == 0)
548 		val = px_match_class_val(class_code, conf,
549 		    property_len / sizeof (px_class_val_t), val);
550 	kmem_free(conf, property_len);
551 	return (val);
552 }
553 
554 /* px_class_to_pil: return the pil for a given device. */
555 uint32_t
556 px_class_to_pil(dev_info_t *rdip)
557 {
558 	uint32_t pil;
559 
560 	/* Default pil is 1 */
561 	pil = px_class_to_val(rdip,
562 	    "pci-class-priorities", px_default_pil,
563 	    sizeof (px_default_pil) / sizeof (px_class_val_t), 1);
564 
565 	/* Range check the result */
566 	if (pil >= 0xf)
567 		pil = 1;
568 
569 	return (pil);
570 }
571 
572 /* px_class_to_intr_weight: return the intr_weight for a given device. */
573 static int32_t
574 px_class_to_intr_weight(dev_info_t *rdip)
575 {
576 	int32_t intr_weight;
577 
578 	/* default weight is 0% */
579 	intr_weight = px_class_to_val(rdip,
580 	    "pci-class-intr-weights", px_default_intr_weight,
581 	    sizeof (px_default_intr_weight) / sizeof (px_class_val_t), 0);
582 
583 	/* range check the result */
584 	if (intr_weight < 0)
585 		intr_weight = 0;
586 	if (intr_weight > 1000)
587 		intr_weight = 1000;
588 
589 	return (intr_weight);
590 }
591 
592 /* ARGSUSED */
593 int
594 px_intx_ops(dev_info_t *dip, dev_info_t *rdip, ddi_intr_op_t intr_op,
595     ddi_intr_handle_impl_t *hdlp, void *result)
596 {
597 	px_t	*px_p = DIP_TO_STATE(dip);
598 	int	ret = DDI_SUCCESS;
599 
600 	DBG(DBG_INTROPS, dip, "px_intx_ops: dip=%x rdip=%x intr_op=%x "
601 	    "handle=%p\n", dip, rdip, intr_op, hdlp);
602 
603 	switch (intr_op) {
604 	case DDI_INTROP_GETCAP:
605 		ret = pci_intx_get_cap(rdip, (int *)result);
606 		break;
607 	case DDI_INTROP_SETCAP:
608 		DBG(DBG_INTROPS, dip, "px_intx_ops: SetCap is not supported\n");
609 		ret = DDI_ENOTSUP;
610 		break;
611 	case DDI_INTROP_ALLOC:
612 		*(int *)result = hdlp->ih_scratch1;
613 		break;
614 	case DDI_INTROP_FREE:
615 		break;
616 	case DDI_INTROP_GETPRI:
617 		*(int *)result = hdlp->ih_pri ?
618 		    hdlp->ih_pri : px_class_to_pil(rdip);
619 		break;
620 	case DDI_INTROP_SETPRI:
621 		break;
622 	case DDI_INTROP_ADDISR:
623 		ret = px_add_intx_intr(dip, rdip, hdlp);
624 		break;
625 	case DDI_INTROP_REMISR:
626 		ret = px_rem_intx_intr(dip, rdip, hdlp);
627 		break;
628 	case DDI_INTROP_ENABLE:
629 		ret = px_ib_update_intr_state(px_p, rdip, hdlp->ih_inum,
630 		    hdlp->ih_vector, hdlp->ih_pri, PX_INTR_STATE_ENABLE, 0, 0);
631 		break;
632 	case DDI_INTROP_DISABLE:
633 		ret = px_ib_update_intr_state(px_p, rdip, hdlp->ih_inum,
634 		    hdlp->ih_vector, hdlp->ih_pri, PX_INTR_STATE_DISABLE, 0, 0);
635 		break;
636 	case DDI_INTROP_SETMASK:
637 		ret = pci_intx_set_mask(rdip);
638 		break;
639 	case DDI_INTROP_CLRMASK:
640 		ret = pci_intx_clr_mask(rdip);
641 		break;
642 	case DDI_INTROP_GETPENDING:
643 		ret = pci_intx_get_pending(rdip, (int *)result);
644 		break;
645 	case DDI_INTROP_NINTRS:
646 	case DDI_INTROP_NAVAIL:
647 		*(int *)result = i_ddi_get_intx_nintrs(rdip);
648 		break;
649 	default:
650 		ret = DDI_ENOTSUP;
651 		break;
652 	}
653 
654 	return (ret);
655 }
656 
657 /* ARGSUSED */
658 int
659 px_msix_ops(dev_info_t *dip, dev_info_t *rdip, ddi_intr_op_t intr_op,
660     ddi_intr_handle_impl_t *hdlp, void *result)
661 {
662 	px_t			*px_p = DIP_TO_STATE(dip);
663 	px_msi_state_t		*msi_state_p = &px_p->px_ib_p->ib_msi_state;
664 	msiq_rec_type_t		msiq_rec_type;
665 	msi_type_t		msi_type;
666 	uint64_t		msi_addr;
667 	msinum_t		msi_num;
668 	msiqid_t		msiq_id;
669 	uint_t			nintrs;
670 	int			i, ret = DDI_SUCCESS;
671 
672 	DBG(DBG_INTROPS, dip, "px_msix_ops: dip=%x rdip=%x intr_op=%x "
673 	    "handle=%p\n", dip, rdip, intr_op, hdlp);
674 
675 	/* Check for MSI64 support */
676 	if ((hdlp->ih_cap & DDI_INTR_FLAG_MSI64) && msi_state_p->msi_addr64) {
677 		msiq_rec_type = MSI64_REC;
678 		msi_type = MSI64_TYPE;
679 		msi_addr = msi_state_p->msi_addr64;
680 	} else {
681 		msiq_rec_type = MSI32_REC;
682 		msi_type = MSI32_TYPE;
683 		msi_addr = msi_state_p->msi_addr32;
684 	}
685 
686 	switch (intr_op) {
687 	case DDI_INTROP_GETCAP:
688 		ret = pci_msi_get_cap(rdip, hdlp->ih_type, (int *)result);
689 		break;
690 	case DDI_INTROP_SETCAP:
691 		DBG(DBG_INTROPS, dip, "px_msix_ops: SetCap is not supported\n");
692 		ret = DDI_ENOTSUP;
693 		break;
694 	case DDI_INTROP_ALLOC:
695 		/*
696 		 * We need to restrict this allocation in future
697 		 * based on Resource Management policies.
698 		 */
699 		if ((ret = px_msi_alloc(px_p, rdip, hdlp->ih_inum,
700 		    hdlp->ih_scratch1, (uintptr_t)hdlp->ih_scratch2, &msi_num,
701 		    (int *)result)) != DDI_SUCCESS) {
702 			DBG(DBG_INTROPS, dip, "px_msix_ops: allocation "
703 			    "failed, rdip 0x%p type 0x%d inum 0x%x "
704 			    "count 0x%x\n", rdip, hdlp->ih_type, hdlp->ih_inum,
705 			    hdlp->ih_scratch1);
706 
707 			return (ret);
708 		}
709 
710 		if ((hdlp->ih_type == DDI_INTR_TYPE_MSIX) &&
711 		    (i_ddi_get_msix(rdip) == NULL)) {
712 			ddi_intr_msix_t		*msix_p;
713 
714 			if (msix_p = pci_msix_init(rdip)) {
715 				i_ddi_set_msix(rdip, msix_p);
716 				break;
717 			}
718 
719 			DBG(DBG_INTROPS, dip, "px_msix_ops: MSI-X allocation "
720 			    "failed, rdip 0x%p inum 0x%x\n", rdip,
721 			    hdlp->ih_inum);
722 
723 			(void) px_msi_free(px_p, rdip, hdlp->ih_inum,
724 			    hdlp->ih_scratch1);
725 
726 			return (DDI_FAILURE);
727 		}
728 
729 		break;
730 	case DDI_INTROP_FREE:
731 		(void) pci_msi_disable_mode(rdip, hdlp->ih_type, NULL);
732 		(void) pci_msi_unconfigure(rdip, hdlp->ih_type, hdlp->ih_inum);
733 
734 		if (hdlp->ih_type == DDI_INTR_TYPE_MSI)
735 			goto msi_free;
736 
737 		if (hdlp->ih_flags & DDI_INTR_MSIX_DUP)
738 			break;
739 
740 		if (((i_ddi_intr_get_current_nintrs(hdlp->ih_dip) - 1) == 0) &&
741 		    (i_ddi_get_msix(rdip))) {
742 			pci_msix_fini(i_ddi_get_msix(rdip));
743 			i_ddi_set_msix(rdip, NULL);
744 		}
745 msi_free:
746 		(void) px_msi_free(px_p, rdip, hdlp->ih_inum,
747 		    hdlp->ih_scratch1);
748 		break;
749 	case DDI_INTROP_GETPRI:
750 		*(int *)result = hdlp->ih_pri ?
751 		    hdlp->ih_pri : px_class_to_pil(rdip);
752 		break;
753 	case DDI_INTROP_SETPRI:
754 		break;
755 	case DDI_INTROP_ADDISR:
756 		if ((ret = px_msi_get_msinum(px_p, hdlp->ih_dip,
757 		    hdlp->ih_inum, &msi_num)) != DDI_SUCCESS)
758 			return (ret);
759 
760 		if ((ret = px_add_msiq_intr(dip, rdip, hdlp,
761 		    msiq_rec_type, msi_num, &msiq_id)) != DDI_SUCCESS) {
762 			DBG(DBG_INTROPS, dip, "px_msix_ops: Add MSI handler "
763 			    "failed, rdip 0x%p msi 0x%x\n", rdip, msi_num);
764 			return (ret);
765 		}
766 
767 		DBG(DBG_INTROPS, dip, "px_msix_ops: msiq used 0x%x\n", msiq_id);
768 
769 		if ((ret = px_lib_msi_setmsiq(dip, msi_num,
770 		    msiq_id, msi_type)) != DDI_SUCCESS) {
771 			(void) px_rem_msiq_intr(dip, rdip,
772 			    hdlp, msiq_rec_type, msi_num, msiq_id);
773 			return (ret);
774 		}
775 
776 		if ((ret = px_lib_msi_setstate(dip, msi_num,
777 		    PCI_MSI_STATE_IDLE)) != DDI_SUCCESS) {
778 			(void) px_rem_msiq_intr(dip, rdip,
779 			    hdlp, msiq_rec_type, msi_num, msiq_id);
780 			return (ret);
781 		}
782 
783 		hdlp->ih_vector = msi_num;
784 		break;
785 	case DDI_INTROP_DUPVEC:
786 		DBG(DBG_INTROPS, dip, "px_msix_ops: dupisr - inum: %x, "
787 		    "new_vector: %x\n", hdlp->ih_inum, hdlp->ih_scratch1);
788 
789 		ret = pci_msix_dup(hdlp->ih_dip, hdlp->ih_inum,
790 		    hdlp->ih_scratch1);
791 		break;
792 	case DDI_INTROP_REMISR:
793 		msi_num = hdlp->ih_vector;
794 
795 		if ((ret = px_lib_msi_getmsiq(dip, msi_num,
796 		    &msiq_id)) != DDI_SUCCESS)
797 			return (ret);
798 
799 		if ((ret = px_lib_msi_setstate(dip, msi_num,
800 		    PCI_MSI_STATE_IDLE)) != DDI_SUCCESS)
801 			return (ret);
802 
803 		ret = px_rem_msiq_intr(dip, rdip,
804 		    hdlp, msiq_rec_type, msi_num, msiq_id);
805 
806 		hdlp->ih_vector = 0;
807 		break;
808 	case DDI_INTROP_ENABLE:
809 		msi_num = hdlp->ih_vector;
810 
811 		if ((ret = px_lib_msi_setvalid(dip, msi_num,
812 		    PCI_MSI_VALID)) != DDI_SUCCESS)
813 			return (ret);
814 
815 		if ((pci_is_msi_enabled(rdip, hdlp->ih_type) != DDI_SUCCESS) ||
816 		    (hdlp->ih_type == DDI_INTR_TYPE_MSIX)) {
817 			nintrs = i_ddi_intr_get_current_nintrs(hdlp->ih_dip);
818 
819 			if ((ret = pci_msi_configure(rdip, hdlp->ih_type,
820 			    nintrs, hdlp->ih_inum, msi_addr,
821 			    hdlp->ih_type == DDI_INTR_TYPE_MSIX ?
822 			    msi_num : msi_num & ~(nintrs - 1))) != DDI_SUCCESS)
823 				return (ret);
824 
825 			if ((ret = pci_msi_enable_mode(rdip, hdlp->ih_type))
826 			    != DDI_SUCCESS)
827 				return (ret);
828 		}
829 
830 		if ((ret = pci_msi_clr_mask(rdip, hdlp->ih_type,
831 		    hdlp->ih_inum)) != DDI_SUCCESS)
832 			return (ret);
833 
834 		if (hdlp->ih_flags & DDI_INTR_MSIX_DUP)
835 			break;
836 
837 		if ((ret = px_lib_msi_getmsiq(dip, msi_num,
838 		    &msiq_id)) != DDI_SUCCESS)
839 			return (ret);
840 
841 		ret = px_ib_update_intr_state(px_p, rdip, hdlp->ih_inum,
842 		    px_msiqid_to_devino(px_p, msiq_id), hdlp->ih_pri,
843 		    PX_INTR_STATE_ENABLE, msiq_rec_type, msi_num);
844 
845 		break;
846 	case DDI_INTROP_DISABLE:
847 		msi_num = hdlp->ih_vector;
848 
849 		if ((ret = pci_msi_set_mask(rdip, hdlp->ih_type,
850 		    hdlp->ih_inum)) != DDI_SUCCESS)
851 			return (ret);
852 
853 		if ((ret = px_lib_msi_setvalid(dip, msi_num,
854 		    PCI_MSI_INVALID)) != DDI_SUCCESS)
855 			return (ret);
856 
857 		if (hdlp->ih_flags & DDI_INTR_MSIX_DUP)
858 			break;
859 
860 		if ((ret = px_lib_msi_getmsiq(dip, msi_num,
861 		    &msiq_id)) != DDI_SUCCESS)
862 			return (ret);
863 
864 		ret = px_ib_update_intr_state(px_p, rdip,
865 		    hdlp->ih_inum, px_msiqid_to_devino(px_p, msiq_id),
866 		    hdlp->ih_pri, PX_INTR_STATE_DISABLE, msiq_rec_type,
867 		    msi_num);
868 
869 		break;
870 	case DDI_INTROP_BLOCKENABLE:
871 		nintrs = i_ddi_intr_get_current_nintrs(hdlp->ih_dip);
872 		msi_num = hdlp->ih_vector;
873 
874 		if ((ret = pci_msi_configure(rdip, hdlp->ih_type,
875 		    nintrs, hdlp->ih_inum, msi_addr,
876 		    msi_num & ~(nintrs - 1))) != DDI_SUCCESS)
877 			return (ret);
878 
879 		for (i = 0; i < nintrs; i++, msi_num++) {
880 			if ((ret = px_lib_msi_setvalid(dip, msi_num,
881 			    PCI_MSI_VALID)) != DDI_SUCCESS)
882 				return (ret);
883 
884 			if ((ret = px_lib_msi_getmsiq(dip, msi_num,
885 			    &msiq_id)) != DDI_SUCCESS)
886 				return (ret);
887 
888 			if ((ret = px_ib_update_intr_state(px_p, rdip,
889 			    hdlp->ih_inum + i, px_msiqid_to_devino(px_p,
890 			    msiq_id), hdlp->ih_pri, PX_INTR_STATE_ENABLE,
891 			    msiq_rec_type, msi_num)) != DDI_SUCCESS)
892 				return (ret);
893 		}
894 
895 		ret = pci_msi_enable_mode(rdip, hdlp->ih_type);
896 		break;
897 	case DDI_INTROP_BLOCKDISABLE:
898 		nintrs = i_ddi_intr_get_current_nintrs(hdlp->ih_dip);
899 		msi_num = hdlp->ih_vector;
900 
901 		if ((ret = pci_msi_disable_mode(rdip, hdlp->ih_type,
902 		    hdlp->ih_cap & DDI_INTR_FLAG_BLOCK)) != DDI_SUCCESS)
903 			return (ret);
904 
905 		for (i = 0; i < nintrs; i++, msi_num++) {
906 			if ((ret = px_lib_msi_setvalid(dip, msi_num,
907 			    PCI_MSI_INVALID)) != DDI_SUCCESS)
908 				return (ret);
909 
910 			if ((ret = px_lib_msi_getmsiq(dip, msi_num,
911 			    &msiq_id)) != DDI_SUCCESS)
912 				return (ret);
913 
914 			if ((ret = px_ib_update_intr_state(px_p, rdip,
915 			    hdlp->ih_inum + i, px_msiqid_to_devino(px_p,
916 			    msiq_id), hdlp->ih_pri, PX_INTR_STATE_DISABLE,
917 			    msiq_rec_type, msi_num)) != DDI_SUCCESS)
918 				return (ret);
919 		}
920 
921 		break;
922 	case DDI_INTROP_SETMASK:
923 		ret = pci_msi_set_mask(rdip, hdlp->ih_type, hdlp->ih_inum);
924 		break;
925 	case DDI_INTROP_CLRMASK:
926 		ret = pci_msi_clr_mask(rdip, hdlp->ih_type, hdlp->ih_inum);
927 		break;
928 	case DDI_INTROP_GETPENDING:
929 		ret = pci_msi_get_pending(rdip, hdlp->ih_type,
930 		    hdlp->ih_inum, (int *)result);
931 		break;
932 	case DDI_INTROP_NINTRS:
933 		ret = pci_msi_get_nintrs(rdip, hdlp->ih_type, (int *)result);
934 		break;
935 	case DDI_INTROP_NAVAIL:
936 		/* XXX - a new interface may be needed */
937 		ret = pci_msi_get_nintrs(rdip, hdlp->ih_type, (int *)result);
938 		break;
939 	default:
940 		ret = DDI_ENOTSUP;
941 		break;
942 	}
943 
944 	return (ret);
945 }
946 
947 static struct {
948 	kstat_named_t pxintr_ks_name;
949 	kstat_named_t pxintr_ks_type;
950 	kstat_named_t pxintr_ks_cpu;
951 	kstat_named_t pxintr_ks_pil;
952 	kstat_named_t pxintr_ks_time;
953 	kstat_named_t pxintr_ks_ino;
954 	kstat_named_t pxintr_ks_cookie;
955 	kstat_named_t pxintr_ks_devpath;
956 	kstat_named_t pxintr_ks_buspath;
957 } pxintr_ks_template = {
958 	{ "name",	KSTAT_DATA_CHAR },
959 	{ "type",	KSTAT_DATA_CHAR },
960 	{ "cpu",	KSTAT_DATA_UINT64 },
961 	{ "pil",	KSTAT_DATA_UINT64 },
962 	{ "time",	KSTAT_DATA_UINT64 },
963 	{ "ino",	KSTAT_DATA_UINT64 },
964 	{ "cookie",	KSTAT_DATA_UINT64 },
965 	{ "devpath",	KSTAT_DATA_STRING },
966 	{ "buspath",	KSTAT_DATA_STRING },
967 };
968 
969 static uint32_t pxintr_ks_instance;
970 static char ih_devpath[MAXPATHLEN];
971 static char ih_buspath[MAXPATHLEN];
972 kmutex_t pxintr_ks_template_lock;
973 
974 int
975 px_ks_update(kstat_t *ksp, int rw)
976 {
977 	px_ih_t *ih_p = ksp->ks_private;
978 	int maxlen = sizeof (pxintr_ks_template.pxintr_ks_name.value.c);
979 	px_ino_pil_t *ipil_p = ih_p->ih_ipil_p;
980 	px_ino_t *ino_p = ipil_p->ipil_ino_p;
981 	px_t *px_p = ino_p->ino_ib_p->ib_px_p;
982 	devino_t ino;
983 	sysino_t sysino;
984 
985 	ino = ino_p->ino_ino;
986 	(void) px_lib_intr_devino_to_sysino(px_p->px_dip, ino, &sysino);
987 
988 	(void) snprintf(pxintr_ks_template.pxintr_ks_name.value.c, maxlen,
989 	    "%s%d", ddi_driver_name(ih_p->ih_dip),
990 	    ddi_get_instance(ih_p->ih_dip));
991 
992 	(void) ddi_pathname(ih_p->ih_dip, ih_devpath);
993 	(void) ddi_pathname(px_p->px_dip, ih_buspath);
994 	kstat_named_setstr(&pxintr_ks_template.pxintr_ks_devpath, ih_devpath);
995 	kstat_named_setstr(&pxintr_ks_template.pxintr_ks_buspath, ih_buspath);
996 
997 	if (ih_p->ih_intr_state == PX_INTR_STATE_ENABLE) {
998 
999 		(void) strcpy(pxintr_ks_template.pxintr_ks_type.value.c,
1000 		    (ih_p->ih_rec_type == 0) ? "fixed" : "msi");
1001 		pxintr_ks_template.pxintr_ks_cpu.value.ui64 = ino_p->ino_cpuid;
1002 		pxintr_ks_template.pxintr_ks_pil.value.ui64 = ipil_p->ipil_pil;
1003 		pxintr_ks_template.pxintr_ks_time.value.ui64 = ih_p->ih_nsec +
1004 		    (uint64_t)tick2ns((hrtime_t)ih_p->ih_ticks,
1005 		    ino_p->ino_cpuid);
1006 		pxintr_ks_template.pxintr_ks_ino.value.ui64 = ino;
1007 		pxintr_ks_template.pxintr_ks_cookie.value.ui64 = sysino;
1008 	} else {
1009 		(void) strcpy(pxintr_ks_template.pxintr_ks_type.value.c,
1010 		    "disabled");
1011 		pxintr_ks_template.pxintr_ks_cpu.value.ui64 = 0;
1012 		pxintr_ks_template.pxintr_ks_pil.value.ui64 = 0;
1013 		pxintr_ks_template.pxintr_ks_time.value.ui64 = 0;
1014 		pxintr_ks_template.pxintr_ks_ino.value.ui64 = 0;
1015 		pxintr_ks_template.pxintr_ks_cookie.value.ui64 = 0;
1016 	}
1017 	return (0);
1018 }
1019 
1020 void
1021 px_create_intr_kstats(px_ih_t *ih_p)
1022 {
1023 	msiq_rec_type_t rec_type = ih_p->ih_rec_type;
1024 
1025 	ASSERT(ih_p->ih_ksp == NULL);
1026 
1027 	/*
1028 	 * Create pci_intrs::: kstats for all ih types except messages,
1029 	 * which represent unusual conditions and don't need to be tracked.
1030 	 */
1031 	if (rec_type == 0 || rec_type == MSI32_REC || rec_type == MSI64_REC) {
1032 		ih_p->ih_ksp = kstat_create("pci_intrs",
1033 		    atomic_inc_32_nv(&pxintr_ks_instance), "config",
1034 		    "interrupts", KSTAT_TYPE_NAMED,
1035 		    sizeof (pxintr_ks_template) / sizeof (kstat_named_t),
1036 		    KSTAT_FLAG_VIRTUAL);
1037 	}
1038 	if (ih_p->ih_ksp != NULL) {
1039 		ih_p->ih_ksp->ks_data_size += MAXPATHLEN * 2;
1040 		ih_p->ih_ksp->ks_lock = &pxintr_ks_template_lock;
1041 		ih_p->ih_ksp->ks_data = &pxintr_ks_template;
1042 		ih_p->ih_ksp->ks_private = ih_p;
1043 		ih_p->ih_ksp->ks_update = px_ks_update;
1044 	}
1045 }
1046 
1047 /*
1048  * px_add_intx_intr:
1049  *
1050  * This function is called to register INTx and legacy hardware
1051  * interrupt pins interrupts.
1052  */
1053 int
1054 px_add_intx_intr(dev_info_t *dip, dev_info_t *rdip,
1055     ddi_intr_handle_impl_t *hdlp)
1056 {
1057 	px_t		*px_p = INST_TO_STATE(ddi_get_instance(dip));
1058 	px_ib_t		*ib_p = px_p->px_ib_p;
1059 	devino_t	ino;
1060 	px_ih_t		*ih_p;
1061 	px_ino_t	*ino_p;
1062 	px_ino_pil_t	*ipil_p, *ipil_list;
1063 	int32_t		weight;
1064 	int		ret = DDI_SUCCESS;
1065 
1066 	ino = hdlp->ih_vector;
1067 
1068 	DBG(DBG_A_INTX, dip, "px_add_intx_intr: rdip=%s%d ino=%x "
1069 	    "handler=%x arg1=%x arg2=%x\n", ddi_driver_name(rdip),
1070 	    ddi_get_instance(rdip), ino, hdlp->ih_cb_func,
1071 	    hdlp->ih_cb_arg1, hdlp->ih_cb_arg2);
1072 
1073 	ih_p = px_ib_alloc_ih(rdip, hdlp->ih_inum,
1074 	    hdlp->ih_cb_func, hdlp->ih_cb_arg1, hdlp->ih_cb_arg2, 0, 0);
1075 
1076 	mutex_enter(&ib_p->ib_ino_lst_mutex);
1077 
1078 	ino_p = px_ib_locate_ino(ib_p, ino);
1079 	ipil_list = ino_p ? ino_p->ino_ipil_p : NULL;
1080 
1081 	/* Sharing ino */
1082 	if (ino_p && (ipil_p = px_ib_ino_locate_ipil(ino_p, hdlp->ih_pri))) {
1083 		if (px_ib_intr_locate_ih(ipil_p, rdip, hdlp->ih_inum, 0, 0)) {
1084 			DBG(DBG_A_INTX, dip, "px_add_intx_intr: "
1085 			    "dup intr #%d\n", hdlp->ih_inum);
1086 
1087 			ret = DDI_FAILURE;
1088 			goto fail1;
1089 		}
1090 
1091 		/* Save mondo value in hdlp */
1092 		hdlp->ih_vector = ino_p->ino_sysino;
1093 
1094 		if ((ret = px_ib_ino_add_intr(px_p, ipil_p,
1095 		    ih_p)) != DDI_SUCCESS)
1096 			goto fail1;
1097 
1098 		goto ino_done;
1099 	}
1100 
1101 	if (hdlp->ih_pri == 0)
1102 		hdlp->ih_pri = px_class_to_pil(rdip);
1103 
1104 	ipil_p = px_ib_new_ino_pil(ib_p, ino, hdlp->ih_pri, ih_p);
1105 	ino_p = ipil_p->ipil_ino_p;
1106 
1107 	/* Save mondo value in hdlp */
1108 	hdlp->ih_vector = ino_p->ino_sysino;
1109 
1110 	DBG(DBG_A_INTX, dip, "px_add_intx_intr: pil=0x%x mondo=0x%x\n",
1111 	    hdlp->ih_pri, hdlp->ih_vector);
1112 
1113 	DDI_INTR_ASSIGN_HDLR_N_ARGS(hdlp,
1114 	    (ddi_intr_handler_t *)px_intx_intr, (caddr_t)ipil_p, NULL);
1115 
1116 	ret = i_ddi_add_ivintr(hdlp);
1117 
1118 	/*
1119 	 * Restore original interrupt handler
1120 	 * and arguments in interrupt handle.
1121 	 */
1122 	DDI_INTR_ASSIGN_HDLR_N_ARGS(hdlp, ih_p->ih_handler,
1123 	    ih_p->ih_handler_arg1, ih_p->ih_handler_arg2);
1124 
1125 	if (ret != DDI_SUCCESS)
1126 		goto fail2;
1127 
1128 	/* Save the pil for this ino */
1129 	ipil_p->ipil_pil = hdlp->ih_pri;
1130 
1131 	/* Select cpu, saving it for sharing and removal */
1132 	if (ipil_list == NULL) {
1133 		ino_p->ino_cpuid = intr_dist_cpuid();
1134 
1135 		/* Enable interrupt */
1136 		px_ib_intr_enable(px_p, ino_p->ino_cpuid, ino);
1137 	}
1138 
1139 ino_done:
1140 	/* Add weight to the cpu that we are already targeting */
1141 	weight = px_class_to_intr_weight(rdip);
1142 	intr_dist_cpuid_add_device_weight(ino_p->ino_cpuid, rdip, weight);
1143 
1144 	ih_p->ih_ipil_p = ipil_p;
1145 	px_create_intr_kstats(ih_p);
1146 	if (ih_p->ih_ksp)
1147 		kstat_install(ih_p->ih_ksp);
1148 	mutex_exit(&ib_p->ib_ino_lst_mutex);
1149 
1150 	DBG(DBG_A_INTX, dip, "px_add_intx_intr: done! Interrupt 0x%x pil=%x\n",
1151 	    ino_p->ino_sysino, hdlp->ih_pri);
1152 
1153 	return (ret);
1154 fail2:
1155 	px_ib_delete_ino_pil(ib_p, ipil_p);
1156 fail1:
1157 	if (ih_p->ih_config_handle)
1158 		pci_config_teardown(&ih_p->ih_config_handle);
1159 
1160 	mutex_exit(&ib_p->ib_ino_lst_mutex);
1161 	kmem_free(ih_p, sizeof (px_ih_t));
1162 
1163 	DBG(DBG_A_INTX, dip, "px_add_intx_intr: Failed! Interrupt 0x%x "
1164 	    "pil=%x\n", ino_p->ino_sysino, hdlp->ih_pri);
1165 
1166 	return (ret);
1167 }
1168 
1169 /*
1170  * px_rem_intx_intr:
1171  *
1172  * This function is called to unregister INTx and legacy hardware
1173  * interrupt pins interrupts.
1174  */
1175 int
1176 px_rem_intx_intr(dev_info_t *dip, dev_info_t *rdip,
1177     ddi_intr_handle_impl_t *hdlp)
1178 {
1179 	px_t		*px_p = INST_TO_STATE(ddi_get_instance(dip));
1180 	px_ib_t		*ib_p = px_p->px_ib_p;
1181 	devino_t	ino;
1182 	cpuid_t		curr_cpu;
1183 	px_ino_t	*ino_p;
1184 	px_ino_pil_t	*ipil_p;
1185 	px_ih_t		*ih_p;
1186 	int		ret = DDI_SUCCESS;
1187 
1188 	ino = hdlp->ih_vector;
1189 
1190 	DBG(DBG_R_INTX, dip, "px_rem_intx_intr: rdip=%s%d ino=%x\n",
1191 	    ddi_driver_name(rdip), ddi_get_instance(rdip), ino);
1192 
1193 	mutex_enter(&ib_p->ib_ino_lst_mutex);
1194 
1195 	ino_p = px_ib_locate_ino(ib_p, ino);
1196 	ipil_p = px_ib_ino_locate_ipil(ino_p, hdlp->ih_pri);
1197 	ih_p = px_ib_intr_locate_ih(ipil_p, rdip, hdlp->ih_inum, 0, 0);
1198 
1199 	/* Get the current cpu */
1200 	if ((ret = px_lib_intr_gettarget(px_p->px_dip, ino_p->ino_sysino,
1201 	    &curr_cpu)) != DDI_SUCCESS)
1202 		goto fail;
1203 
1204 	if ((ret = px_ib_ino_rem_intr(px_p, ipil_p, ih_p)) != DDI_SUCCESS)
1205 		goto fail;
1206 
1207 	intr_dist_cpuid_rem_device_weight(ino_p->ino_cpuid, rdip);
1208 
1209 	if (ipil_p->ipil_ih_size == 0) {
1210 		hdlp->ih_vector = ino_p->ino_sysino;
1211 		i_ddi_rem_ivintr(hdlp);
1212 
1213 		px_ib_delete_ino_pil(ib_p, ipil_p);
1214 	}
1215 
1216 	if (ino_p->ino_ipil_size == 0) {
1217 		kmem_free(ino_p, sizeof (px_ino_t));
1218 	} else {
1219 		/* Re-enable interrupt only if mapping register still shared */
1220 		PX_INTR_ENABLE(px_p->px_dip, ino_p->ino_sysino, curr_cpu);
1221 	}
1222 
1223 fail:
1224 	mutex_exit(&ib_p->ib_ino_lst_mutex);
1225 	return (ret);
1226 }
1227 
1228 /*
1229  * px_add_msiq_intr:
1230  *
1231  * This function is called to register MSI/Xs and PCIe message interrupts.
1232  */
1233 int
1234 px_add_msiq_intr(dev_info_t *dip, dev_info_t *rdip,
1235     ddi_intr_handle_impl_t *hdlp, msiq_rec_type_t rec_type,
1236     msgcode_t msg_code, msiqid_t *msiq_id_p)
1237 {
1238 	px_t		*px_p = INST_TO_STATE(ddi_get_instance(dip));
1239 	px_ib_t		*ib_p = px_p->px_ib_p;
1240 	px_msiq_state_t	*msiq_state_p = &ib_p->ib_msiq_state;
1241 	devino_t	ino;
1242 	px_ih_t		*ih_p;
1243 	px_ino_t	*ino_p;
1244 	px_ino_pil_t	*ipil_p, *ipil_list;
1245 	int32_t		weight;
1246 	int		ret = DDI_SUCCESS;
1247 
1248 	DBG(DBG_MSIQ, dip, "px_add_msiq_intr: rdip=%s%d handler=%x "
1249 	    "arg1=%x arg2=%x\n", ddi_driver_name(rdip), ddi_get_instance(rdip),
1250 	    hdlp->ih_cb_func, hdlp->ih_cb_arg1, hdlp->ih_cb_arg2);
1251 
1252 	if ((ret = px_msiq_alloc(px_p, rec_type, msiq_id_p)) != DDI_SUCCESS) {
1253 		DBG(DBG_MSIQ, dip, "px_add_msiq_intr: "
1254 		    "msiq allocation failed\n");
1255 		return (ret);
1256 	}
1257 
1258 	ino = px_msiqid_to_devino(px_p, *msiq_id_p);
1259 
1260 	ih_p = px_ib_alloc_ih(rdip, hdlp->ih_inum, hdlp->ih_cb_func,
1261 	    hdlp->ih_cb_arg1, hdlp->ih_cb_arg2, rec_type, msg_code);
1262 
1263 	mutex_enter(&ib_p->ib_ino_lst_mutex);
1264 
1265 	ino_p = px_ib_locate_ino(ib_p, ino);
1266 	ipil_list = ino_p ? ino_p->ino_ipil_p : NULL;
1267 
1268 	/* Sharing ino */
1269 	if (ino_p && (ipil_p = px_ib_ino_locate_ipil(ino_p, hdlp->ih_pri))) {
1270 		if (px_ib_intr_locate_ih(ipil_p, rdip,
1271 		    hdlp->ih_inum, rec_type, msg_code)) {
1272 			DBG(DBG_MSIQ, dip, "px_add_msiq_intr: "
1273 			    "dup intr #%d\n", hdlp->ih_inum);
1274 
1275 			ret = DDI_FAILURE;
1276 			goto fail1;
1277 		}
1278 
1279 		/* Save mondo value in hdlp */
1280 		hdlp->ih_vector = ino_p->ino_sysino;
1281 
1282 		if ((ret = px_ib_ino_add_intr(px_p, ipil_p,
1283 		    ih_p)) != DDI_SUCCESS)
1284 			goto fail1;
1285 
1286 		goto ino_done;
1287 	}
1288 
1289 	if (hdlp->ih_pri == 0)
1290 		hdlp->ih_pri = px_class_to_pil(rdip);
1291 
1292 	ipil_p = px_ib_new_ino_pil(ib_p, ino, hdlp->ih_pri, ih_p);
1293 	ino_p = ipil_p->ipil_ino_p;
1294 
1295 	ino_p->ino_msiq_p = msiq_state_p->msiq_p +
1296 	    (*msiq_id_p - msiq_state_p->msiq_1st_msiq_id);
1297 
1298 	/* Save mondo value in hdlp */
1299 	hdlp->ih_vector = ino_p->ino_sysino;
1300 
1301 	DBG(DBG_MSIQ, dip, "px_add_msiq_intr: pil=0x%x mondo=0x%x\n",
1302 	    hdlp->ih_pri, hdlp->ih_vector);
1303 
1304 	DDI_INTR_ASSIGN_HDLR_N_ARGS(hdlp,
1305 	    (ddi_intr_handler_t *)px_msiq_intr, (caddr_t)ipil_p, NULL);
1306 
1307 	ret = i_ddi_add_ivintr(hdlp);
1308 
1309 	/*
1310 	 * Restore original interrupt handler
1311 	 * and arguments in interrupt handle.
1312 	 */
1313 	DDI_INTR_ASSIGN_HDLR_N_ARGS(hdlp, ih_p->ih_handler,
1314 	    ih_p->ih_handler_arg1, ih_p->ih_handler_arg2);
1315 
1316 	if (ret != DDI_SUCCESS)
1317 		goto fail2;
1318 
1319 	/* Save the pil for this ino */
1320 	ipil_p->ipil_pil = hdlp->ih_pri;
1321 
1322 	/* Select cpu, saving it for sharing and removal */
1323 	if (ipil_list == NULL) {
1324 		ino_p->ino_cpuid = intr_dist_cpuid();
1325 
1326 		/* Enable MSIQ */
1327 		px_lib_msiq_setstate(dip, *msiq_id_p, PCI_MSIQ_STATE_IDLE);
1328 		px_lib_msiq_setvalid(dip, *msiq_id_p, PCI_MSIQ_VALID);
1329 
1330 		/* Enable interrupt */
1331 		px_ib_intr_enable(px_p, ino_p->ino_cpuid, ino);
1332 	}
1333 
1334 ino_done:
1335 	/* Add weight to the cpu that we are already targeting */
1336 	weight = px_class_to_intr_weight(rdip);
1337 	intr_dist_cpuid_add_device_weight(ino_p->ino_cpuid, rdip, weight);
1338 
1339 	ih_p->ih_ipil_p = ipil_p;
1340 	px_create_intr_kstats(ih_p);
1341 	if (ih_p->ih_ksp)
1342 		kstat_install(ih_p->ih_ksp);
1343 	mutex_exit(&ib_p->ib_ino_lst_mutex);
1344 
1345 	DBG(DBG_MSIQ, dip, "px_add_msiq_intr: done! Interrupt 0x%x pil=%x\n",
1346 	    ino_p->ino_sysino, hdlp->ih_pri);
1347 
1348 	return (ret);
1349 fail2:
1350 	px_ib_delete_ino_pil(ib_p, ipil_p);
1351 fail1:
1352 	if (ih_p->ih_config_handle)
1353 		pci_config_teardown(&ih_p->ih_config_handle);
1354 
1355 	mutex_exit(&ib_p->ib_ino_lst_mutex);
1356 	kmem_free(ih_p, sizeof (px_ih_t));
1357 
1358 	DBG(DBG_MSIQ, dip, "px_add_msiq_intr: Failed! Interrupt 0x%x pil=%x\n",
1359 	    ino_p->ino_sysino, hdlp->ih_pri);
1360 
1361 	return (ret);
1362 }
1363 
1364 /*
1365  * px_rem_msiq_intr:
1366  *
1367  * This function is called to unregister MSI/Xs and PCIe message interrupts.
1368  */
1369 int
1370 px_rem_msiq_intr(dev_info_t *dip, dev_info_t *rdip,
1371     ddi_intr_handle_impl_t *hdlp, msiq_rec_type_t rec_type,
1372     msgcode_t msg_code, msiqid_t msiq_id)
1373 {
1374 	px_t		*px_p = INST_TO_STATE(ddi_get_instance(dip));
1375 	px_ib_t		*ib_p = px_p->px_ib_p;
1376 	devino_t	ino = px_msiqid_to_devino(px_p, msiq_id);
1377 	cpuid_t		curr_cpu;
1378 	px_ino_t	*ino_p;
1379 	px_ino_pil_t	*ipil_p;
1380 	px_ih_t		*ih_p;
1381 	int		ret = DDI_SUCCESS;
1382 
1383 	DBG(DBG_MSIQ, dip, "px_rem_msiq_intr: rdip=%s%d msiq_id=%x ino=%x\n",
1384 	    ddi_driver_name(rdip), ddi_get_instance(rdip), msiq_id, ino);
1385 
1386 	mutex_enter(&ib_p->ib_ino_lst_mutex);
1387 
1388 	ino_p = px_ib_locate_ino(ib_p, ino);
1389 	ipil_p = px_ib_ino_locate_ipil(ino_p, hdlp->ih_pri);
1390 	ih_p = px_ib_intr_locate_ih(ipil_p, rdip, hdlp->ih_inum, rec_type,
1391 	    msg_code);
1392 
1393 	/* Get the current cpu */
1394 	if ((ret = px_lib_intr_gettarget(px_p->px_dip, ino_p->ino_sysino,
1395 	    &curr_cpu)) != DDI_SUCCESS)
1396 		goto fail;
1397 
1398 	if ((ret = px_ib_ino_rem_intr(px_p, ipil_p, ih_p)) != DDI_SUCCESS)
1399 		goto fail;
1400 
1401 	intr_dist_cpuid_rem_device_weight(ino_p->ino_cpuid, rdip);
1402 
1403 	if (ipil_p->ipil_ih_size == 0) {
1404 		hdlp->ih_vector = ino_p->ino_sysino;
1405 		i_ddi_rem_ivintr(hdlp);
1406 
1407 		px_ib_delete_ino_pil(ib_p, ipil_p);
1408 
1409 		if (ino_p->ino_ipil_size == 0)
1410 			px_lib_msiq_setvalid(dip,
1411 			    px_devino_to_msiqid(px_p, ino), PCI_MSIQ_INVALID);
1412 
1413 		(void) px_msiq_free(px_p, msiq_id);
1414 	}
1415 
1416 	if (ino_p->ino_ipil_size == 0) {
1417 		kmem_free(ino_p, sizeof (px_ino_t));
1418 	} else {
1419 		/* Re-enable interrupt only if mapping register still shared */
1420 		PX_INTR_ENABLE(px_p->px_dip, ino_p->ino_sysino, curr_cpu);
1421 	}
1422 
1423 fail:
1424 	mutex_exit(&ib_p->ib_ino_lst_mutex);
1425 	return (ret);
1426 }
1427