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 2006 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  * apic_introp.c:
30  *	Has code for Advanced DDI interrupt framework support.
31  */
32 
33 #include <sys/cpuvar.h>
34 #include <sys/psm.h>
35 #include "apic.h"
36 #include <sys/sunddi.h>
37 #include <sys/ddi_impldefs.h>
38 #include <sys/mach_intr.h>
39 #include <sys/sysmacros.h>
40 #include <sys/trap.h>
41 #include <sys/pci.h>
42 #include <sys/pci_intr_lib.h>
43 
44 extern struct av_head autovect[];
45 
46 /*
47  *	Local Function Prototypes
48  */
49 int		apic_pci_msi_enable_vector(dev_info_t *, int, int,
50 		    int, int, int);
51 apic_irq_t	*apic_find_irq(dev_info_t *, struct intrspec *, int);
52 static int	apic_get_pending(apic_irq_t *, int);
53 static void	apic_clear_mask(apic_irq_t *);
54 static void	apic_set_mask(apic_irq_t *);
55 static uchar_t	apic_find_multi_vectors(int, int);
56 int		apic_navail_vector(dev_info_t *, int);
57 int		apic_alloc_vectors(dev_info_t *, int, int, int, int);
58 void		apic_free_vectors(dev_info_t *, int, int, int, int);
59 int		apic_intr_ops(dev_info_t *, ddi_intr_handle_impl_t *,
60 		    psm_intr_op_t, int *);
61 
62 extern int	intr_clear(void);
63 extern void	intr_restore(uint_t);
64 extern uchar_t	apic_bind_intr(dev_info_t *, int, uchar_t, uchar_t);
65 extern int	apic_allocate_irq(int);
66 extern int	apic_introp_xlate(dev_info_t *, struct intrspec *, int);
67 extern int	apic_rebind_all(apic_irq_t *irq_ptr, int bind_cpu, int safe);
68 extern boolean_t apic_cpu_in_range(int cpu);
69 
70 /*
71  * MSI support flag:
72  * reflects whether MSI is supported at APIC level
73  * it can also be patched through /etc/system
74  *
75  *  0 = default value - don't know and need to call apic_check_msi_support()
76  *      to find out then set it accordingly
77  *  1 = supported
78  * -1 = not supported
79  */
80 int	apic_support_msi = 0;
81 
82 /* Multiple vector support for MSI */
83 int	apic_multi_msi_enable = 1;
84 int	apic_multi_msi_max = 2;
85 
86 extern uchar_t		apic_ipltopri[MAXIPL+1];
87 extern uchar_t		apic_vector_to_irq[APIC_MAX_VECTOR+1];
88 extern int		apic_max_device_irq;
89 extern int		apic_min_device_irq;
90 extern apic_irq_t	*apic_irq_table[APIC_MAX_VECTOR+1];
91 extern volatile uint32_t *apicadr; /* virtual addr of local APIC */
92 extern volatile int32_t	*apicioadr[MAX_IO_APIC];
93 extern lock_t		apic_ioapic_lock;
94 extern kmutex_t		airq_mutex;
95 extern apic_cpus_info_t	*apic_cpus;
96 extern int apic_first_avail_irq;
97 
98 
99 /*
100  * apic_pci_msi_enable_vector:
101  *	Set the address/data fields in the MSI/X capability structure
102  *	XXX: MSI-X support
103  */
104 /* ARGSUSED */
105 int
106 apic_pci_msi_enable_vector(dev_info_t *dip, int type, int inum, int vector,
107     int count, int target_apic_id)
108 {
109 	uint64_t		msi_addr, msi_data;
110 	ushort_t		msi_ctrl;
111 	int			cap_ptr = i_ddi_get_msi_msix_cap_ptr(dip);
112 	ddi_acc_handle_t	handle = i_ddi_get_pci_config_handle(dip);
113 
114 	DDI_INTR_IMPLDBG((CE_CONT, "apic_pci_msi_enable_vector: dip=0x%p\n"
115 	    "\tdriver = %s, inum=0x%x vector=0x%x apicid=0x%x\n", (void *)dip,
116 	    ddi_driver_name(dip), inum, vector, target_apic_id));
117 
118 	if (handle == NULL)
119 		return (PSM_FAILURE);
120 
121 	/* MSI Address */
122 	msi_addr = (MSI_ADDR_HDR | (target_apic_id << MSI_ADDR_DEST_SHIFT));
123 	msi_addr |= ((MSI_ADDR_RH_FIXED << MSI_ADDR_RH_SHIFT) |
124 		    (MSI_ADDR_DM_PHYSICAL << MSI_ADDR_DM_SHIFT));
125 
126 	/* MSI Data: MSI is edge triggered according to spec */
127 	msi_data = ((MSI_DATA_TM_EDGE << MSI_DATA_TM_SHIFT) | vector);
128 
129 	DDI_INTR_IMPLDBG((CE_CONT, "apic_pci_msi_enable_vector: addr=0x%lx "
130 	    "data=0x%lx\n", (long)msi_addr, (long)msi_data));
131 
132 	if (type == DDI_INTR_TYPE_MSI) {
133 		msi_ctrl = pci_config_get16(handle, cap_ptr + PCI_MSI_CTRL);
134 
135 		/* Set the bits to inform how many MSIs are enabled */
136 		msi_ctrl |= ((highbit(count) -1) << PCI_MSI_MME_SHIFT);
137 		pci_config_put16(handle, cap_ptr + PCI_MSI_CTRL, msi_ctrl);
138 
139 		pci_config_put32(handle,
140 		    cap_ptr + PCI_MSI_ADDR_OFFSET, msi_addr);
141 
142 		if (msi_ctrl &  PCI_MSI_64BIT_MASK) {
143 			pci_config_put32(handle,
144 			    cap_ptr + PCI_MSI_ADDR_OFFSET + 4, msi_addr >> 32);
145 			pci_config_put16(handle,
146 			    cap_ptr + PCI_MSI_64BIT_DATA, msi_data);
147 		} else {
148 			pci_config_put16(handle,
149 			    cap_ptr + PCI_MSI_32BIT_DATA, msi_data);
150 		}
151 
152 	} else if (type == DDI_INTR_TYPE_MSIX) {
153 		uintptr_t	off;
154 		ddi_intr_msix_t	*msix_p = i_ddi_get_msix(dip);
155 
156 		/* Offset into the "inum"th entry in the MSI-X table */
157 		off = (uintptr_t)msix_p->msix_tbl_addr +
158 		    (inum  * PCI_MSIX_VECTOR_SIZE);
159 
160 		ddi_put32(msix_p->msix_tbl_hdl,
161 		    (uint32_t *)(off + PCI_MSIX_DATA_OFFSET), msi_data);
162 		ddi_put64(msix_p->msix_tbl_hdl,
163 		    (uint64_t *)(off + PCI_MSIX_LOWER_ADDR_OFFSET), msi_addr);
164 	}
165 
166 	return (PSM_SUCCESS);
167 }
168 
169 
170 /*
171  * This function returns the no. of vectors available for the pri.
172  * dip is not used at this moment.  If we really don't need that,
173  * it will be removed.
174  */
175 /*ARGSUSED*/
176 int
177 apic_navail_vector(dev_info_t *dip, int pri)
178 {
179 	int	lowest, highest, i, navail, count;
180 
181 	DDI_INTR_IMPLDBG((CE_CONT, "apic_navail_vector: dip: %p, pri: %x\n",
182 	    (void *)dip, pri));
183 
184 	highest = apic_ipltopri[pri] + APIC_VECTOR_MASK;
185 	lowest = apic_ipltopri[pri - 1] + APIC_VECTOR_PER_IPL;
186 	navail = count = 0;
187 
188 	/* It has to be contiguous */
189 	for (i = lowest; i < highest; i++) {
190 		count = 0;
191 		while ((apic_vector_to_irq[i] == APIC_RESV_IRQ) &&
192 			(i < highest)) {
193 			if ((i == T_FASTTRAP) || (i == APIC_SPUR_INTR))
194 				break;
195 			count++;
196 			i++;
197 		}
198 		if (count > navail)
199 			navail = count;
200 	}
201 	return (navail);
202 }
203 
204 static uchar_t
205 apic_find_multi_vectors(int pri, int count)
206 {
207 	int	lowest, highest, i, navail, start;
208 
209 	DDI_INTR_IMPLDBG((CE_CONT, "apic_find_mult: pri: %x, count: %x\n",
210 	    pri, count));
211 
212 	highest = apic_ipltopri[pri] + APIC_VECTOR_MASK;
213 	lowest = apic_ipltopri[pri - 1] + APIC_VECTOR_PER_IPL;
214 	navail = 0;
215 
216 	/* It has to be contiguous */
217 	for (i = lowest; i < highest; i++) {
218 		navail = 0;
219 		start = i;
220 		while ((apic_vector_to_irq[i] == APIC_RESV_IRQ) &&
221 			(i < highest)) {
222 			if ((i == T_FASTTRAP) || (i == APIC_SPUR_INTR))
223 				break;
224 			navail++;
225 			if (navail >= count)
226 				return (start);
227 			i++;
228 		}
229 	}
230 	return (0);
231 }
232 
233 
234 /*
235  * It finds the apic_irq_t associates with the dip, ispec and type.
236  */
237 apic_irq_t *
238 apic_find_irq(dev_info_t *dip, struct intrspec *ispec, int type)
239 {
240 	apic_irq_t	*irqp;
241 	int i;
242 
243 	DDI_INTR_IMPLDBG((CE_CONT, "apic_find_irq: dip=0x%p vec=0x%x "
244 	    "ipl=0x%x type=0x%x\n", (void *)dip, ispec->intrspec_vec,
245 	    ispec->intrspec_pri, type));
246 
247 	for (i = apic_min_device_irq; i <= apic_max_device_irq; i++) {
248 		if ((irqp = apic_irq_table[i]) == NULL)
249 			continue;
250 		if ((irqp->airq_dip == dip) &&
251 		    (irqp->airq_origirq == ispec->intrspec_vec) &&
252 		    (irqp->airq_ipl == ispec->intrspec_pri)) {
253 			if (DDI_INTR_IS_MSI_OR_MSIX(type)) {
254 				if (APIC_IS_MSI_OR_MSIX_INDEX(irqp->
255 				    airq_mps_intr_index))
256 					return (irqp);
257 			} else
258 				return (irqp);
259 		}
260 	}
261 	DDI_INTR_IMPLDBG((CE_CONT, "apic_find_irq: return NULL\n"));
262 	return (NULL);
263 }
264 
265 
266 /*
267  * This function will return the pending bit of the irqp.
268  * It either comes from the IRR register of the APIC or the RDT
269  * entry of the I/O APIC.
270  * For the IRR to work, it needs to be to its binding CPU
271  */
272 static int
273 apic_get_pending(apic_irq_t *irqp, int type)
274 {
275 	int			bit, index, irr, pending;
276 	int			intin_no;
277 	volatile int32_t 	*ioapic;
278 
279 	DDI_INTR_IMPLDBG((CE_CONT, "apic_get_pending: irqp: %p, cpuid: %x "
280 	    "type: %x\n", (void *)irqp, irqp->airq_cpu & ~IRQ_USER_BOUND,
281 	    type));
282 
283 	/* need to get on the bound cpu */
284 	mutex_enter(&cpu_lock);
285 	affinity_set(irqp->airq_cpu & ~IRQ_USER_BOUND);
286 
287 	index = irqp->airq_vector / 32;
288 	bit = irqp->airq_vector % 32;
289 	irr = apicadr[APIC_IRR_REG + index];
290 
291 	affinity_clear();
292 	mutex_exit(&cpu_lock);
293 
294 	pending = (irr & (1 << bit)) ? 1 : 0;
295 	if (!pending && (type == DDI_INTR_TYPE_FIXED)) {
296 		/* check I/O APIC for fixed interrupt */
297 		intin_no = irqp->airq_intin_no;
298 		ioapic = apicioadr[irqp->airq_ioapicindex];
299 		pending = (READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no) &
300 		    AV_PENDING) ? 1 : 0;
301 	}
302 	return (pending);
303 }
304 
305 
306 /*
307  * This function will clear the mask for the interrupt on the I/O APIC
308  */
309 static void
310 apic_clear_mask(apic_irq_t *irqp)
311 {
312 	int			intin_no;
313 	int			iflag;
314 	int32_t			rdt_entry;
315 	volatile int32_t 	*ioapic;
316 
317 	DDI_INTR_IMPLDBG((CE_CONT, "apic_clear_mask: irqp: %p\n",
318 	    (void *)irqp));
319 
320 	intin_no = irqp->airq_intin_no;
321 	ioapic = apicioadr[irqp->airq_ioapicindex];
322 
323 	iflag = intr_clear();
324 	lock_set(&apic_ioapic_lock);
325 
326 	rdt_entry = READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no);
327 
328 	/* clear mask */
329 	WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no,
330 	    ((~AV_MASK) & rdt_entry));
331 
332 	lock_clear(&apic_ioapic_lock);
333 	intr_restore(iflag);
334 }
335 
336 
337 /*
338  * This function will mask the interrupt on the I/O APIC
339  */
340 static void
341 apic_set_mask(apic_irq_t *irqp)
342 {
343 	int			intin_no;
344 	volatile int32_t 	*ioapic;
345 	int			iflag;
346 	int32_t			rdt_entry;
347 
348 	DDI_INTR_IMPLDBG((CE_CONT, "apic_set_mask: irqp: %p\n", (void *)irqp));
349 
350 	intin_no = irqp->airq_intin_no;
351 	ioapic = apicioadr[irqp->airq_ioapicindex];
352 
353 	iflag = intr_clear();
354 
355 	lock_set(&apic_ioapic_lock);
356 
357 	rdt_entry = READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no);
358 
359 	/* mask it */
360 	WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no,
361 	    (AV_MASK | rdt_entry));
362 
363 	lock_clear(&apic_ioapic_lock);
364 	intr_restore(iflag);
365 }
366 
367 
368 /*
369  * This function allocate "count" vector(s) for the given "dip/pri/type"
370  */
371 int
372 apic_alloc_vectors(dev_info_t *dip, int inum, int count, int pri, int type)
373 {
374 	int	rcount, i;
375 	uchar_t	start, irqno, cpu;
376 	short	idx;
377 	major_t	major;
378 	apic_irq_t	*irqptr;
379 
380 	/* for MSI/X only */
381 	if (!DDI_INTR_IS_MSI_OR_MSIX(type))
382 		return (0);
383 
384 	DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_vectors: dip=0x%p type=%d "
385 	    "inum=0x%x  pri=0x%x count=0x%x\n",
386 	    (void *)dip, type, inum, pri, count));
387 
388 	if (count > 1) {
389 		if (apic_multi_msi_enable == 0)
390 			count = 1;
391 		else if (count > apic_multi_msi_max)
392 			count = apic_multi_msi_max;
393 	}
394 
395 	if ((rcount = apic_navail_vector(dip, pri)) > count)
396 		rcount = count;
397 
398 	mutex_enter(&airq_mutex);
399 
400 	for (start = 0; rcount > 0; rcount--) {
401 		if ((start = apic_find_multi_vectors(pri, rcount)) != 0)
402 			break;
403 	}
404 
405 	if (start == 0) {
406 		/* no vector available */
407 		mutex_exit(&airq_mutex);
408 		return (0);
409 	}
410 
411 	idx = (short)((type == DDI_INTR_TYPE_MSI) ? MSI_INDEX : MSIX_INDEX);
412 	major = (dip != NULL) ? ddi_name_to_major(ddi_get_name(dip)) : 0;
413 	for (i = 0; i < rcount; i++) {
414 		if ((irqno = apic_allocate_irq(apic_first_avail_irq)) ==
415 		    (uchar_t)-1) {
416 			mutex_exit(&airq_mutex);
417 			DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_vectors: "
418 			    "apic_allocate_irq failed\n"));
419 			return (i);
420 		}
421 		apic_max_device_irq = max(irqno, apic_max_device_irq);
422 		apic_min_device_irq = min(irqno, apic_min_device_irq);
423 		irqptr = apic_irq_table[irqno];
424 #ifdef	DEBUG
425 		if (apic_vector_to_irq[start + i] != APIC_RESV_IRQ)
426 			DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_vectors: "
427 			    "apic_vector_to_irq is not APIC_RESV_IRQ\n"));
428 #endif
429 		apic_vector_to_irq[start + i] = (uchar_t)irqno;
430 
431 		irqptr->airq_vector = (uchar_t)(start + i);
432 		irqptr->airq_ioapicindex = (uchar_t)inum;	/* start */
433 		irqptr->airq_intin_no = (uchar_t)rcount;
434 		irqptr->airq_ipl = pri;
435 		irqptr->airq_vector = start + i;
436 		irqptr->airq_origirq = (uchar_t)(inum + i);
437 		irqptr->airq_share_id = 0;
438 		irqptr->airq_mps_intr_index = idx;
439 		irqptr->airq_dip = dip;
440 		irqptr->airq_major = major;
441 		if (i == 0) /* they all bound to the same cpu */
442 			cpu = irqptr->airq_cpu = apic_bind_intr(dip, irqno,
443 				0xff, 0xff);
444 		else
445 			irqptr->airq_cpu = cpu;
446 		DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_vectors: irq=0x%x "
447 		    "dip=0x%p vector=0x%x origirq=0x%x pri=0x%x\n", irqno,
448 		    (void *)irqptr->airq_dip, irqptr->airq_vector,
449 		    irqptr->airq_origirq, pri));
450 	}
451 	mutex_exit(&airq_mutex);
452 	return (rcount);
453 }
454 
455 
456 void
457 apic_free_vectors(dev_info_t *dip, int inum, int count, int pri, int type)
458 {
459 	int i;
460 	apic_irq_t *irqptr;
461 	struct intrspec ispec;
462 
463 	DDI_INTR_IMPLDBG((CE_CONT, "apic_free_vectors: dip: %p inum: %x "
464 	    "count: %x pri: %x type: %x\n",
465 	    (void *)dip, inum, count, pri, type));
466 
467 	/* for MSI/X only */
468 	if (!DDI_INTR_IS_MSI_OR_MSIX(type))
469 		return;
470 
471 	for (i = 0; i < count; i++) {
472 		DDI_INTR_IMPLDBG((CE_CONT, "apic_free_vectors: inum=0x%x "
473 		    "pri=0x%x count=0x%x\n", inum, pri, count));
474 		ispec.intrspec_vec = inum + i;
475 		ispec.intrspec_pri = pri;
476 		if ((irqptr = apic_find_irq(dip, &ispec, type)) == NULL) {
477 			DDI_INTR_IMPLDBG((CE_CONT, "apic_free_vectors: "
478 			    "dip=0x%p inum=0x%x pri=0x%x apic_find_irq() "
479 			    "failed\n", (void *)dip, inum, pri));
480 			continue;
481 		}
482 		irqptr->airq_mps_intr_index = FREE_INDEX;
483 		apic_vector_to_irq[irqptr->airq_vector] = APIC_RESV_IRQ;
484 	}
485 }
486 
487 
488 /*
489  * check whether the system supports MSI
490  *
491  * If PCI-E capability is found, then this must be a PCI-E system.
492  * Since MSI is required for PCI-E system, it returns PSM_SUCCESS
493  * to indicate this system supports MSI.
494  */
495 int
496 apic_check_msi_support()
497 {
498 	dev_info_t *cdip;
499 	char dev_type[16];
500 	int dev_len;
501 
502 	DDI_INTR_IMPLDBG((CE_CONT, "apic_check_msi_support:\n"));
503 
504 	/*
505 	 * check whether the first level children of root_node have
506 	 * PCI-E capability
507 	 */
508 	for (cdip = ddi_get_child(ddi_root_node()); cdip != NULL;
509 	    cdip = ddi_get_next_sibling(cdip)) {
510 
511 		DDI_INTR_IMPLDBG((CE_CONT, "apic_check_msi_support: cdip: 0x%p,"
512 		    " driver: %s, binding: %s, nodename: %s\n", (void *)cdip,
513 		    ddi_driver_name(cdip), ddi_binding_name(cdip),
514 		    ddi_node_name(cdip)));
515 		dev_len = sizeof (dev_type);
516 		if (ddi_getlongprop_buf(DDI_DEV_T_ANY, cdip, DDI_PROP_DONTPASS,
517 		    "device_type", (caddr_t)dev_type, &dev_len)
518 		    != DDI_PROP_SUCCESS)
519 			continue;
520 		if (strcmp(dev_type, "pciex") == 0)
521 			return (PSM_SUCCESS);
522 	}
523 
524 	/* MSI is not supported on this system */
525 	DDI_INTR_IMPLDBG((CE_CONT, "apic_check_msi_support: no 'pciex' "
526 	    "device_type found\n"));
527 	return (PSM_FAILURE);
528 }
529 
530 int
531 apic_get_vector_intr_info(int vecirq, apic_get_intr_t *intr_params_p)
532 {
533 	struct autovec *av_dev;
534 	uchar_t irqno;
535 	int i;
536 	apic_irq_t *irq_p;
537 
538 	/* Sanity check the vector/irq argument. */
539 	ASSERT((vecirq >= 0) || (vecirq <= APIC_MAX_VECTOR));
540 
541 	mutex_enter(&airq_mutex);
542 
543 	/*
544 	 * Convert the vecirq arg to an irq using vector_to_irq table
545 	 * if the arg is a vector.  Pass thru if already an irq.
546 	 */
547 	if ((intr_params_p->avgi_req_flags & PSMGI_INTRBY_FLAGS) ==
548 	    PSMGI_INTRBY_VEC)
549 		irqno = apic_vector_to_irq[vecirq];
550 	else
551 		irqno = vecirq;
552 
553 	irq_p = apic_irq_table[irqno];
554 
555 	if ((irq_p == NULL) ||
556 	    (irq_p->airq_temp_cpu == IRQ_UNBOUND) ||
557 	    (irq_p->airq_temp_cpu == IRQ_UNINIT)) {
558 		mutex_exit(&airq_mutex);
559 		return (PSM_FAILURE);
560 	}
561 
562 	if (intr_params_p->avgi_req_flags & PSMGI_REQ_CPUID) {
563 
564 		/* Get the (temp) cpu from apic_irq table, indexed by irq. */
565 		intr_params_p->avgi_cpu_id = irq_p->airq_temp_cpu;
566 
567 		/* Return user bound info for intrd. */
568 		if (intr_params_p->avgi_cpu_id & IRQ_USER_BOUND) {
569 			intr_params_p->avgi_cpu_id &= ~IRQ_USER_BOUND;
570 			intr_params_p->avgi_cpu_id |= PSMGI_CPU_USER_BOUND;
571 		}
572 	}
573 
574 	if (intr_params_p->avgi_req_flags & PSMGI_REQ_VECTOR) {
575 		intr_params_p->avgi_vector = irq_p->airq_vector;
576 	}
577 
578 	if (intr_params_p->avgi_req_flags &
579 	    (PSMGI_REQ_NUM_DEVS | PSMGI_REQ_GET_DEVS)) {
580 		/* Get number of devices from apic_irq table shared field. */
581 		intr_params_p->avgi_num_devs = irq_p->airq_share;
582 	}
583 
584 	if (intr_params_p->avgi_req_flags &  PSMGI_REQ_GET_DEVS) {
585 
586 		intr_params_p->avgi_req_flags  |= PSMGI_REQ_NUM_DEVS;
587 
588 		/* Some devices have NULL dip.  Don't count these. */
589 		if (intr_params_p->avgi_num_devs > 0) {
590 			for (i = 0, av_dev = autovect[irqno].avh_link;
591 			    av_dev; av_dev = av_dev->av_link)
592 				if (av_dev->av_vector && av_dev->av_dip)
593 					i++;
594 			intr_params_p->avgi_num_devs =
595 			    MIN(intr_params_p->avgi_num_devs, i);
596 		}
597 
598 		/* There are no viable dips to return. */
599 		if (intr_params_p->avgi_num_devs == 0)
600 			intr_params_p->avgi_dip_list = NULL;
601 
602 		else {	/* Return list of dips */
603 
604 			/* Allocate space in array for that number of devs. */
605 			intr_params_p->avgi_dip_list = kmem_zalloc(
606 			    intr_params_p->avgi_num_devs *
607 			    sizeof (dev_info_t *),
608 			    KM_SLEEP);
609 
610 			/*
611 			 * Loop through the device list of the autovec table
612 			 * filling in the dip array.
613 			 *
614 			 * Note that the autovect table may have some special
615 			 * entries which contain NULL dips.  These will be
616 			 * ignored.
617 			 */
618 			for (i = 0, av_dev = autovect[irqno].avh_link;
619 			    av_dev; av_dev = av_dev->av_link)
620 				if (av_dev->av_vector && av_dev->av_dip)
621 					intr_params_p->avgi_dip_list[i++] =
622 					    av_dev->av_dip;
623 		}
624 	}
625 
626 	mutex_exit(&airq_mutex);
627 
628 	return (PSM_SUCCESS);
629 }
630 
631 /*
632  * apic_pci_msi_unconfigure:
633  *
634  * This and next two interfaces are copied from pci_intr_lib.c
635  * Do ensure that these two files stay in sync.
636  * These needed to be copied over here to avoid a deadlock situation on
637  * certain mp systems that use MSI interrupts.
638  *
639  * IMPORTANT regards next three interfaces:
640  * i) are called only for MSI/X interrupts.
641  * ii) called with interrupts disabled, and must not block
642  */
643 int
644 apic_pci_msi_unconfigure(dev_info_t *rdip, int type, int inum)
645 {
646 	ushort_t		msi_ctrl;
647 	int			cap_ptr = i_ddi_get_msi_msix_cap_ptr(rdip);
648 	ddi_acc_handle_t	handle = i_ddi_get_pci_config_handle(rdip);
649 
650 	if (handle == NULL)
651 		return (PSM_FAILURE);
652 
653 	if (type == DDI_INTR_TYPE_MSI) {
654 		msi_ctrl = pci_config_get16(handle, cap_ptr + PCI_MSI_CTRL);
655 		msi_ctrl &= (~PCI_MSI_MME_MASK);
656 		pci_config_put16(handle, cap_ptr + PCI_MSI_CTRL, msi_ctrl);
657 		pci_config_put32(handle, cap_ptr + PCI_MSI_ADDR_OFFSET, 0);
658 
659 		if (msi_ctrl &  PCI_MSI_64BIT_MASK) {
660 			pci_config_put16(handle,
661 			    cap_ptr + PCI_MSI_64BIT_DATA, 0);
662 			pci_config_put32(handle,
663 			    cap_ptr + PCI_MSI_ADDR_OFFSET + 4, 0);
664 		} else {
665 			pci_config_put16(handle,
666 			    cap_ptr + PCI_MSI_32BIT_DATA, 0);
667 		}
668 
669 	} else if (type == DDI_INTR_TYPE_MSIX) {
670 		uintptr_t	off;
671 		ddi_intr_msix_t	*msix_p = i_ddi_get_msix(rdip);
672 
673 		/* Offset into the "inum"th entry in the MSI-X table */
674 		off = (uintptr_t)msix_p->msix_tbl_addr +
675 		    (inum * PCI_MSIX_VECTOR_SIZE);
676 
677 		/* Reset the "data" and "addr" bits */
678 		ddi_put32(msix_p->msix_tbl_hdl,
679 		    (uint32_t *)(off + PCI_MSIX_DATA_OFFSET), 0);
680 		ddi_put64(msix_p->msix_tbl_hdl, (uint64_t *)off, 0);
681 	}
682 
683 	return (PSM_SUCCESS);
684 }
685 
686 
687 /*
688  * apic_pci_msi_enable_mode:
689  */
690 int
691 apic_pci_msi_enable_mode(dev_info_t *rdip, int type, int inum)
692 {
693 	ushort_t		msi_ctrl;
694 	int			cap_ptr = i_ddi_get_msi_msix_cap_ptr(rdip);
695 	ddi_acc_handle_t	handle = i_ddi_get_pci_config_handle(rdip);
696 
697 	if (handle == NULL)
698 		return (PSM_FAILURE);
699 
700 	if (type == DDI_INTR_TYPE_MSI) {
701 		msi_ctrl = pci_config_get16(handle, cap_ptr + PCI_MSI_CTRL);
702 		if ((msi_ctrl & PCI_MSI_ENABLE_BIT))
703 			return (PSM_SUCCESS);
704 
705 		msi_ctrl |= PCI_MSI_ENABLE_BIT;
706 		pci_config_put16(handle, cap_ptr + PCI_MSI_CTRL, msi_ctrl);
707 
708 	} else if (type == DDI_INTR_TYPE_MSIX) {
709 		uintptr_t	off;
710 		ddi_intr_msix_t	*msix_p;
711 
712 		msi_ctrl = pci_config_get16(handle, cap_ptr + PCI_MSIX_CTRL);
713 
714 		if (msi_ctrl & PCI_MSIX_ENABLE_BIT)
715 			return (PSM_SUCCESS);
716 
717 		msi_ctrl |= PCI_MSIX_ENABLE_BIT;
718 		pci_config_put16(handle, cap_ptr + PCI_MSIX_CTRL, msi_ctrl);
719 
720 		msix_p = i_ddi_get_msix(rdip);
721 
722 		/* Offset into "inum"th entry in the MSI-X table & clear mask */
723 		off = (uintptr_t)msix_p->msix_tbl_addr + (inum *
724 		    PCI_MSIX_VECTOR_SIZE) + PCI_MSIX_VECTOR_CTRL_OFFSET;
725 		ddi_put32(msix_p->msix_tbl_hdl, (uint32_t *)off, 0);
726 	}
727 
728 	return (PSM_SUCCESS);
729 }
730 
731 /*
732  * apic_pci_msi_disable_mode:
733  */
734 int
735 apic_pci_msi_disable_mode(dev_info_t *rdip, int type, int inum)
736 {
737 	ushort_t		msi_ctrl;
738 	int			cap_ptr = i_ddi_get_msi_msix_cap_ptr(rdip);
739 	ddi_acc_handle_t	handle = i_ddi_get_pci_config_handle(rdip);
740 
741 	if (handle == NULL)
742 		return (PSM_FAILURE);
743 
744 	if (type == DDI_INTR_TYPE_MSI) {
745 		msi_ctrl = pci_config_get16(handle, cap_ptr + PCI_MSI_CTRL);
746 		if (!(msi_ctrl & PCI_MSI_ENABLE_BIT))
747 			return (PSM_SUCCESS);
748 
749 		msi_ctrl &= ~PCI_MSI_ENABLE_BIT;	/* MSI disable */
750 		pci_config_put16(handle, cap_ptr + PCI_MSI_CTRL, msi_ctrl);
751 
752 	} else if (type == DDI_INTR_TYPE_MSIX) {
753 		uintptr_t	off;
754 		ddi_intr_msix_t	*msix_p;
755 
756 		msi_ctrl = pci_config_get16(handle, cap_ptr + PCI_MSIX_CTRL);
757 
758 		if (!(msi_ctrl & PCI_MSIX_ENABLE_BIT))
759 			return (PSM_SUCCESS);
760 
761 		msix_p = i_ddi_get_msix(rdip);
762 
763 		/* Offset into "inum"th entry in the MSI-X table & mask it */
764 		off = (uintptr_t)msix_p->msix_tbl_addr + (inum *
765 		    PCI_MSIX_VECTOR_SIZE) + PCI_MSIX_VECTOR_CTRL_OFFSET;
766 		ddi_put32(msix_p->msix_tbl_hdl, (uint32_t *)off, 0x1);
767 	}
768 
769 	return (PSM_SUCCESS);
770 }
771 
772 /*
773  * This function provides external interface to the nexus for all
774  * functionalities related to the new DDI interrupt framework.
775  *
776  * Input:
777  * dip     - pointer to the dev_info structure of the requested device
778  * hdlp    - pointer to the internal interrupt handle structure for the
779  *	     requested interrupt
780  * intr_op - opcode for this call
781  * result  - pointer to the integer that will hold the result to be
782  *	     passed back if return value is PSM_SUCCESS
783  *
784  * Output:
785  * return value is either PSM_SUCCESS or PSM_FAILURE
786  */
787 int
788 apic_intr_ops(dev_info_t *dip, ddi_intr_handle_impl_t *hdlp,
789     psm_intr_op_t intr_op, int *result)
790 {
791 	int		cap, ret;
792 	int		count_vec;
793 	int		cpu;
794 	int		old_priority;
795 	int		new_priority;
796 	apic_irq_t	*irqp;
797 	struct intrspec *ispec, intr_spec;
798 
799 	DDI_INTR_IMPLDBG((CE_CONT, "apic_intr_ops: dip: %p hdlp: %p "
800 	    "intr_op: %x\n", (void *)dip, (void *)hdlp, intr_op));
801 
802 	ispec = &intr_spec;
803 	ispec->intrspec_pri = hdlp->ih_pri;
804 	ispec->intrspec_vec = hdlp->ih_inum;
805 	ispec->intrspec_func = hdlp->ih_cb_func;
806 
807 	switch (intr_op) {
808 	case PSM_INTR_OP_CHECK_MSI:
809 		/*
810 		 * Check MSI/X is supported or not at APIC level and
811 		 * masked off the MSI/X bits in hdlp->ih_type if not
812 		 * supported before return.  If MSI/X is supported,
813 		 * leave the ih_type unchanged and return.
814 		 *
815 		 * hdlp->ih_type passed in from the nexus has all the
816 		 * interrupt types supported by the device.
817 		 */
818 		if (apic_support_msi == 0) {
819 			/*
820 			 * if apic_support_msi is not set, call
821 			 * apic_check_msi_support() to check whether msi
822 			 * is supported first
823 			 */
824 			if (apic_check_msi_support() == PSM_SUCCESS)
825 				apic_support_msi = 1;
826 			else
827 				apic_support_msi = -1;
828 		}
829 		if (apic_support_msi == 1)
830 			*result = hdlp->ih_type;
831 		else
832 			*result = hdlp->ih_type & ~(DDI_INTR_TYPE_MSI |
833 			    DDI_INTR_TYPE_MSIX);
834 		break;
835 	case PSM_INTR_OP_ALLOC_VECTORS:
836 		*result = apic_alloc_vectors(dip, hdlp->ih_inum,
837 		    hdlp->ih_scratch1, hdlp->ih_pri, hdlp->ih_type);
838 		break;
839 	case PSM_INTR_OP_FREE_VECTORS:
840 		apic_free_vectors(dip, hdlp->ih_inum, hdlp->ih_scratch1,
841 		    hdlp->ih_pri, hdlp->ih_type);
842 		break;
843 	case PSM_INTR_OP_NAVAIL_VECTORS:
844 		*result = apic_navail_vector(dip, hdlp->ih_pri);
845 		break;
846 	case PSM_INTR_OP_XLATE_VECTOR:
847 		ispec = ((ihdl_plat_t *)hdlp->ih_private)->ip_ispecp;
848 		*result = apic_introp_xlate(dip, ispec, hdlp->ih_type);
849 		break;
850 	case PSM_INTR_OP_GET_PENDING:
851 		if ((irqp = apic_find_irq(dip, ispec, hdlp->ih_type)) == NULL)
852 			return (PSM_FAILURE);
853 		*result = apic_get_pending(irqp, hdlp->ih_type);
854 		break;
855 	case PSM_INTR_OP_CLEAR_MASK:
856 		if (hdlp->ih_type != DDI_INTR_TYPE_FIXED)
857 			return (PSM_FAILURE);
858 		irqp = apic_find_irq(dip, ispec, hdlp->ih_type);
859 		if (irqp == NULL)
860 			return (PSM_FAILURE);
861 		apic_clear_mask(irqp);
862 		break;
863 	case PSM_INTR_OP_SET_MASK:
864 		if (hdlp->ih_type != DDI_INTR_TYPE_FIXED)
865 			return (PSM_FAILURE);
866 		if ((irqp = apic_find_irq(dip, ispec, hdlp->ih_type)) == NULL)
867 			return (PSM_FAILURE);
868 		apic_set_mask(irqp);
869 		break;
870 	case PSM_INTR_OP_GET_CAP:
871 		cap = DDI_INTR_FLAG_PENDING;
872 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED)
873 			cap |= DDI_INTR_FLAG_MASKABLE;
874 		*result = cap;
875 		break;
876 	case PSM_INTR_OP_GET_SHARED:
877 		if (hdlp->ih_type != DDI_INTR_TYPE_FIXED)
878 			return (PSM_FAILURE);
879 		if ((irqp = apic_find_irq(dip, ispec, hdlp->ih_type)) == NULL)
880 			return (PSM_FAILURE);
881 		*result = irqp->airq_share ? 1: 0;
882 		break;
883 	case PSM_INTR_OP_SET_PRI:
884 		old_priority = hdlp->ih_pri;	/* save old value */
885 		new_priority = *(int *)result;	/* try the new value */
886 
887 		/* First, check if "hdlp->ih_scratch1" vectors exist? */
888 		if (apic_navail_vector(dip, new_priority) < hdlp->ih_scratch1)
889 			return (PSM_FAILURE);
890 
891 		/* Now allocate the vectors */
892 		count_vec = apic_alloc_vectors(dip, hdlp->ih_inum,
893 		    hdlp->ih_scratch1, new_priority, hdlp->ih_type);
894 
895 		/* Did we get fewer vectors? */
896 		if (count_vec != hdlp->ih_scratch1) {
897 			apic_free_vectors(dip, hdlp->ih_inum, count_vec,
898 			    new_priority, hdlp->ih_type);
899 			return (PSM_FAILURE);
900 		}
901 
902 		/* Finally, free the previously allocated vectors */
903 		apic_free_vectors(dip, hdlp->ih_inum, count_vec,
904 		    old_priority, hdlp->ih_type);
905 		hdlp->ih_pri = new_priority; /* set the new value */
906 		break;
907 	case PSM_INTR_OP_SET_CPU:
908 		/*
909 		 * The interrupt handle given here has been allocated
910 		 * specifically for this command, and ih_private carries
911 		 * a CPU value.
912 		 */
913 		cpu = (int)(intptr_t)hdlp->ih_private;
914 
915 		if (!apic_cpu_in_range(cpu)) {
916 			*result = EINVAL;
917 			return (PSM_FAILURE);
918 		}
919 
920 		mutex_enter(&airq_mutex);
921 
922 		/* Convert the vector to the irq using vector_to_irq table. */
923 		irqp = apic_irq_table[apic_vector_to_irq[hdlp->ih_vector]];
924 		if (irqp == NULL) {
925 			mutex_exit(&airq_mutex);
926 			*result = ENXIO;
927 			return (PSM_FAILURE);
928 		}
929 		ret = apic_rebind_all(irqp, cpu, 1);
930 		mutex_exit(&airq_mutex);
931 		if (ret) {
932 			*result = EIO;
933 			return (PSM_FAILURE);
934 		}
935 		*result = 0;
936 		break;
937 	case PSM_INTR_OP_GET_INTR:
938 		/*
939 		 * The interrupt handle given here has been allocated
940 		 * specifically for this command, and ih_private carries
941 		 * a pointer to a apic_get_intr_t.
942 		 */
943 		if (apic_get_vector_intr_info(
944 		    hdlp->ih_vector, hdlp->ih_private) != PSM_SUCCESS)
945 			return (PSM_FAILURE);
946 		break;
947 	case PSM_INTR_OP_SET_CAP:
948 	default:
949 		return (PSM_FAILURE);
950 	}
951 	return (PSM_SUCCESS);
952 }
953