xref: /titanic_52/usr/src/uts/i86pc/io/pcplusmp/apic_introp.c (revision 8181b438236881e6d31f7155101c85c011a9b6bb)
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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright 2013 Pluribus Networks, Inc.
24  */
25 
26 /*
27  * apic_introp.c:
28  *	Has code for Advanced DDI interrupt framework support.
29  */
30 
31 #include <sys/cpuvar.h>
32 #include <sys/psm.h>
33 #include <sys/archsystm.h>
34 #include <sys/apic.h>
35 #include <sys/sunddi.h>
36 #include <sys/ddi_impldefs.h>
37 #include <sys/mach_intr.h>
38 #include <sys/sysmacros.h>
39 #include <sys/trap.h>
40 #include <sys/pci.h>
41 #include <sys/pci_intr_lib.h>
42 #include <sys/apic_common.h>
43 
44 extern struct av_head autovect[];
45 
46 /*
47  *	Local Function Prototypes
48  */
49 apic_irq_t	*apic_find_irq(dev_info_t *, struct intrspec *, int);
50 
51 /*
52  * apic_pci_msi_enable_vector:
53  *	Set the address/data fields in the MSI/X capability structure
54  *	XXX: MSI-X support
55  */
56 /* ARGSUSED */
57 void
58 apic_pci_msi_enable_vector(apic_irq_t *irq_ptr, int type, int inum, int vector,
59     int count, int target_apic_id)
60 {
61 	uint64_t		msi_addr, msi_data;
62 	ushort_t		msi_ctrl;
63 	dev_info_t		*dip = irq_ptr->airq_dip;
64 	int			cap_ptr = i_ddi_get_msi_msix_cap_ptr(dip);
65 	ddi_acc_handle_t	handle = i_ddi_get_pci_config_handle(dip);
66 	msi_regs_t		msi_regs;
67 	int			irqno, i;
68 	void			*intrmap_tbl[PCI_MSI_MAX_INTRS];
69 
70 	DDI_INTR_IMPLDBG((CE_CONT, "apic_pci_msi_enable_vector: dip=0x%p\n"
71 	    "\tdriver = %s, inum=0x%x vector=0x%x apicid=0x%x\n", (void *)dip,
72 	    ddi_driver_name(dip), inum, vector, target_apic_id));
73 
74 	ASSERT((handle != NULL) && (cap_ptr != 0));
75 
76 	msi_regs.mr_data = vector;
77 	msi_regs.mr_addr = target_apic_id;
78 
79 	for (i = 0; i < count; i++) {
80 		irqno = apic_vector_to_irq[vector + i];
81 		intrmap_tbl[i] = apic_irq_table[irqno]->airq_intrmap_private;
82 	}
83 	apic_vt_ops->apic_intrmap_alloc_entry(intrmap_tbl, dip, type,
84 	    count, 0xff);
85 	for (i = 0; i < count; i++) {
86 		irqno = apic_vector_to_irq[vector + i];
87 		apic_irq_table[irqno]->airq_intrmap_private =
88 		    intrmap_tbl[i];
89 	}
90 
91 	apic_vt_ops->apic_intrmap_map_entry(irq_ptr->airq_intrmap_private,
92 	    (void *)&msi_regs, type, count);
93 	apic_vt_ops->apic_intrmap_record_msi(irq_ptr->airq_intrmap_private,
94 	    &msi_regs);
95 
96 	/* MSI Address */
97 	msi_addr = msi_regs.mr_addr;
98 
99 	/* MSI Data: MSI is edge triggered according to spec */
100 	msi_data = msi_regs.mr_data;
101 
102 	DDI_INTR_IMPLDBG((CE_CONT, "apic_pci_msi_enable_vector: addr=0x%lx "
103 	    "data=0x%lx\n", (long)msi_addr, (long)msi_data));
104 
105 	if (type == DDI_INTR_TYPE_MSI) {
106 		msi_ctrl = pci_config_get16(handle, cap_ptr + PCI_MSI_CTRL);
107 
108 		/* Set the bits to inform how many MSIs are enabled */
109 		msi_ctrl |= ((highbit(count) -1) << PCI_MSI_MME_SHIFT);
110 		pci_config_put16(handle, cap_ptr + PCI_MSI_CTRL, msi_ctrl);
111 
112 		/*
113 		 * Only set vector if not on hypervisor
114 		 */
115 		pci_config_put32(handle,
116 		    cap_ptr + PCI_MSI_ADDR_OFFSET, msi_addr);
117 
118 		if (msi_ctrl &  PCI_MSI_64BIT_MASK) {
119 			pci_config_put32(handle,
120 			    cap_ptr + PCI_MSI_ADDR_OFFSET + 4, msi_addr >> 32);
121 			pci_config_put16(handle,
122 			    cap_ptr + PCI_MSI_64BIT_DATA, msi_data);
123 		} else {
124 			pci_config_put16(handle,
125 			    cap_ptr + PCI_MSI_32BIT_DATA, msi_data);
126 		}
127 
128 	} else if (type == DDI_INTR_TYPE_MSIX) {
129 		uintptr_t	off;
130 		ddi_intr_msix_t	*msix_p = i_ddi_get_msix(dip);
131 
132 		ASSERT(msix_p != NULL);
133 
134 		/* Offset into the "inum"th entry in the MSI-X table */
135 		off = (uintptr_t)msix_p->msix_tbl_addr +
136 		    (inum  * PCI_MSIX_VECTOR_SIZE);
137 
138 		ddi_put32(msix_p->msix_tbl_hdl,
139 		    (uint32_t *)(off + PCI_MSIX_DATA_OFFSET), msi_data);
140 		ddi_put32(msix_p->msix_tbl_hdl,
141 		    (uint32_t *)(off + PCI_MSIX_LOWER_ADDR_OFFSET), msi_addr);
142 		ddi_put32(msix_p->msix_tbl_hdl,
143 		    (uint32_t *)(off + PCI_MSIX_UPPER_ADDR_OFFSET),
144 		    msi_addr >> 32);
145 	}
146 }
147 
148 /*
149  * This function returns the no. of vectors available for the pri.
150  * dip is not used at this moment.  If we really don't need that,
151  * it will be removed.
152  */
153 /*ARGSUSED*/
154 int
155 apic_navail_vector(dev_info_t *dip, int pri)
156 {
157 	int	lowest, highest, i, navail, count;
158 
159 	DDI_INTR_IMPLDBG((CE_CONT, "apic_navail_vector: dip: %p, pri: %x\n",
160 	    (void *)dip, pri));
161 
162 	highest = apic_ipltopri[pri] + APIC_VECTOR_MASK;
163 	lowest = apic_ipltopri[pri - 1] + APIC_VECTOR_PER_IPL;
164 	navail = count = 0;
165 
166 	if (highest < lowest) /* Both ipl and ipl - 1 map to same pri */
167 		lowest -= APIC_VECTOR_PER_IPL;
168 
169 	/* It has to be contiguous */
170 	for (i = lowest; i <= highest; i++) {
171 		count = 0;
172 		while ((apic_vector_to_irq[i] == APIC_RESV_IRQ) &&
173 		    (i <= highest)) {
174 			if (APIC_CHECK_RESERVE_VECTORS(i))
175 				break;
176 			count++;
177 			i++;
178 		}
179 		if (count > navail)
180 			navail = count;
181 	}
182 	return (navail);
183 }
184 
185 /*
186  * Finds "count" contiguous MSI vectors starting at the proper alignment
187  * at "pri".
188  * Caller needs to make sure that count has to be power of 2 and should not
189  * be < 1.
190  */
191 uchar_t
192 apic_find_multi_vectors(int pri, int count)
193 {
194 	int	lowest, highest, i, navail, start, msibits;
195 
196 	DDI_INTR_IMPLDBG((CE_CONT, "apic_find_mult: pri: %x, count: %x\n",
197 	    pri, count));
198 
199 	highest = apic_ipltopri[pri] + APIC_VECTOR_MASK;
200 	lowest = apic_ipltopri[pri - 1] + APIC_VECTOR_PER_IPL;
201 	navail = 0;
202 
203 	if (highest < lowest) /* Both ipl and ipl - 1 map to same pri */
204 		lowest -= APIC_VECTOR_PER_IPL;
205 
206 	/*
207 	 * msibits is the no. of lower order message data bits for the
208 	 * allocated MSI vectors and is used to calculate the aligned
209 	 * starting vector
210 	 */
211 	msibits = count - 1;
212 
213 	/* It has to be contiguous */
214 	for (i = lowest; i <= highest; i++) {
215 		navail = 0;
216 
217 		/*
218 		 * starting vector has to be aligned accordingly for
219 		 * multiple MSIs
220 		 */
221 		if (msibits)
222 			i = (i + msibits) & ~msibits;
223 		start = i;
224 		while ((apic_vector_to_irq[i] == APIC_RESV_IRQ) &&
225 		    (i <= highest)) {
226 			if (APIC_CHECK_RESERVE_VECTORS(i))
227 				break;
228 			navail++;
229 			if (navail >= count)
230 				return (start);
231 			i++;
232 		}
233 	}
234 	return (0);
235 }
236 
237 
238 /*
239  * It finds the apic_irq_t associates with the dip, ispec and type.
240  */
241 apic_irq_t *
242 apic_find_irq(dev_info_t *dip, struct intrspec *ispec, int type)
243 {
244 	apic_irq_t	*irqp;
245 	int i;
246 
247 	DDI_INTR_IMPLDBG((CE_CONT, "apic_find_irq: dip=0x%p vec=0x%x "
248 	    "ipl=0x%x type=0x%x\n", (void *)dip, ispec->intrspec_vec,
249 	    ispec->intrspec_pri, type));
250 
251 	for (i = apic_min_device_irq; i <= apic_max_device_irq; i++) {
252 		for (irqp = apic_irq_table[i]; irqp; irqp = irqp->airq_next) {
253 			if ((irqp->airq_dip == dip) &&
254 			    (irqp->airq_origirq == ispec->intrspec_vec) &&
255 			    (irqp->airq_ipl == ispec->intrspec_pri)) {
256 				if (type == DDI_INTR_TYPE_MSI) {
257 					if (irqp->airq_mps_intr_index ==
258 					    MSI_INDEX)
259 						return (irqp);
260 				} else if (type == DDI_INTR_TYPE_MSIX) {
261 					if (irqp->airq_mps_intr_index ==
262 					    MSIX_INDEX)
263 						return (irqp);
264 				} else
265 					return (irqp);
266 			}
267 		}
268 	}
269 	DDI_INTR_IMPLDBG((CE_CONT, "apic_find_irq: return NULL\n"));
270 	return (NULL);
271 }
272 
273 /*
274  * This function will return the pending bit of the irqp.
275  * It either comes from the IRR register of the APIC or the RDT
276  * entry of the I/O APIC.
277  * For the IRR to work, it needs to be to its binding CPU
278  */
279 static int
280 apic_get_pending(apic_irq_t *irqp, int type)
281 {
282 	int			bit, index, irr, pending;
283 	int			intin_no;
284 	int			apic_ix;
285 
286 	DDI_INTR_IMPLDBG((CE_CONT, "apic_get_pending: irqp: %p, cpuid: %x "
287 	    "type: %x\n", (void *)irqp, irqp->airq_cpu & ~IRQ_USER_BOUND,
288 	    type));
289 
290 	/* need to get on the bound cpu */
291 	mutex_enter(&cpu_lock);
292 	affinity_set(irqp->airq_cpu & ~IRQ_USER_BOUND);
293 
294 	index = irqp->airq_vector / 32;
295 	bit = irqp->airq_vector % 32;
296 	irr = apic_reg_ops->apic_read(APIC_IRR_REG + index);
297 
298 	affinity_clear();
299 	mutex_exit(&cpu_lock);
300 
301 	pending = (irr & (1 << bit)) ? 1 : 0;
302 	if (!pending && (type == DDI_INTR_TYPE_FIXED)) {
303 		/* check I/O APIC for fixed interrupt */
304 		intin_no = irqp->airq_intin_no;
305 		apic_ix = irqp->airq_ioapicindex;
306 		pending = (READ_IOAPIC_RDT_ENTRY_LOW_DWORD(apic_ix, intin_no) &
307 		    AV_PENDING) ? 1 : 0;
308 	}
309 	return (pending);
310 }
311 
312 
313 /*
314  * This function will clear the mask for the interrupt on the I/O APIC
315  */
316 static void
317 apic_clear_mask(apic_irq_t *irqp)
318 {
319 	int			intin_no;
320 	ulong_t			iflag;
321 	int32_t			rdt_entry;
322 	int 			apic_ix;
323 
324 	DDI_INTR_IMPLDBG((CE_CONT, "apic_clear_mask: irqp: %p\n",
325 	    (void *)irqp));
326 
327 	intin_no = irqp->airq_intin_no;
328 	apic_ix = irqp->airq_ioapicindex;
329 
330 	iflag = intr_clear();
331 	lock_set(&apic_ioapic_lock);
332 
333 	rdt_entry = READ_IOAPIC_RDT_ENTRY_LOW_DWORD(apic_ix, intin_no);
334 
335 	/* clear mask */
336 	WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(apic_ix, intin_no,
337 	    ((~AV_MASK) & rdt_entry));
338 
339 	lock_clear(&apic_ioapic_lock);
340 	intr_restore(iflag);
341 }
342 
343 
344 /*
345  * This function will mask the interrupt on the I/O APIC
346  */
347 static void
348 apic_set_mask(apic_irq_t *irqp)
349 {
350 	int			intin_no;
351 	int 			apic_ix;
352 	ulong_t			iflag;
353 	int32_t			rdt_entry;
354 
355 	DDI_INTR_IMPLDBG((CE_CONT, "apic_set_mask: irqp: %p\n", (void *)irqp));
356 
357 	intin_no = irqp->airq_intin_no;
358 	apic_ix = irqp->airq_ioapicindex;
359 
360 	iflag = intr_clear();
361 
362 	lock_set(&apic_ioapic_lock);
363 
364 	rdt_entry = READ_IOAPIC_RDT_ENTRY_LOW_DWORD(apic_ix, intin_no);
365 
366 	/* mask it */
367 	WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(apic_ix, intin_no,
368 	    (AV_MASK | rdt_entry));
369 
370 	lock_clear(&apic_ioapic_lock);
371 	intr_restore(iflag);
372 }
373 
374 
375 void
376 apic_free_vectors(dev_info_t *dip, int inum, int count, int pri, int type)
377 {
378 	int i;
379 	apic_irq_t *irqptr;
380 	struct intrspec ispec;
381 
382 	DDI_INTR_IMPLDBG((CE_CONT, "apic_free_vectors: dip: %p inum: %x "
383 	    "count: %x pri: %x type: %x\n",
384 	    (void *)dip, inum, count, pri, type));
385 
386 	/* for MSI/X only */
387 	if (!DDI_INTR_IS_MSI_OR_MSIX(type))
388 		return;
389 
390 	for (i = 0; i < count; i++) {
391 		DDI_INTR_IMPLDBG((CE_CONT, "apic_free_vectors: inum=0x%x "
392 		    "pri=0x%x count=0x%x\n", inum, pri, count));
393 		ispec.intrspec_vec = inum + i;
394 		ispec.intrspec_pri = pri;
395 		if ((irqptr = apic_find_irq(dip, &ispec, type)) == NULL) {
396 			DDI_INTR_IMPLDBG((CE_CONT, "apic_free_vectors: "
397 			    "dip=0x%p inum=0x%x pri=0x%x apic_find_irq() "
398 			    "failed\n", (void *)dip, inum, pri));
399 			continue;
400 		}
401 		irqptr->airq_mps_intr_index = FREE_INDEX;
402 		apic_vector_to_irq[irqptr->airq_vector] = APIC_RESV_IRQ;
403 	}
404 }
405 
406 /*
407  * apic_pci_msi_enable_mode:
408  */
409 void
410 apic_pci_msi_enable_mode(dev_info_t *rdip, int type, int inum)
411 {
412 	ushort_t		msi_ctrl;
413 	int			cap_ptr = i_ddi_get_msi_msix_cap_ptr(rdip);
414 	ddi_acc_handle_t	handle = i_ddi_get_pci_config_handle(rdip);
415 
416 	ASSERT((handle != NULL) && (cap_ptr != 0));
417 
418 	if (type == DDI_INTR_TYPE_MSI) {
419 		msi_ctrl = pci_config_get16(handle, cap_ptr + PCI_MSI_CTRL);
420 		if ((msi_ctrl & PCI_MSI_ENABLE_BIT))
421 			return;
422 
423 		msi_ctrl |= PCI_MSI_ENABLE_BIT;
424 		pci_config_put16(handle, cap_ptr + PCI_MSI_CTRL, msi_ctrl);
425 
426 	} else if (type == DDI_INTR_TYPE_MSIX) {
427 		uintptr_t	off;
428 		uint32_t	mask;
429 		ddi_intr_msix_t	*msix_p;
430 
431 		msix_p = i_ddi_get_msix(rdip);
432 
433 		ASSERT(msix_p != NULL);
434 
435 		/* Offset into "inum"th entry in the MSI-X table & clear mask */
436 		off = (uintptr_t)msix_p->msix_tbl_addr + (inum *
437 		    PCI_MSIX_VECTOR_SIZE) + PCI_MSIX_VECTOR_CTRL_OFFSET;
438 
439 		mask = ddi_get32(msix_p->msix_tbl_hdl, (uint32_t *)off);
440 
441 		ddi_put32(msix_p->msix_tbl_hdl, (uint32_t *)off, (mask & ~1));
442 
443 		msi_ctrl = pci_config_get16(handle, cap_ptr + PCI_MSIX_CTRL);
444 
445 		if (!(msi_ctrl & PCI_MSIX_ENABLE_BIT)) {
446 			msi_ctrl |= PCI_MSIX_ENABLE_BIT;
447 			pci_config_put16(handle, cap_ptr + PCI_MSIX_CTRL,
448 			    msi_ctrl);
449 		}
450 	}
451 }
452 
453 static int
454 apic_set_cpu(int irqno, int cpu, int *result)
455 {
456 	apic_irq_t *irqp;
457 	ulong_t iflag;
458 	int ret;
459 
460 	DDI_INTR_IMPLDBG((CE_CONT, "APIC_SET_CPU\n"));
461 
462 	mutex_enter(&airq_mutex);
463 	irqp = apic_irq_table[irqno];
464 	mutex_exit(&airq_mutex);
465 
466 	if (irqp == NULL) {
467 		*result = ENXIO;
468 		return (PSM_FAILURE);
469 	}
470 
471 	/* Fail if this is an MSI intr and is part of a group. */
472 	if ((irqp->airq_mps_intr_index == MSI_INDEX) &&
473 	    (irqp->airq_intin_no > 1)) {
474 		*result = ENXIO;
475 		return (PSM_FAILURE);
476 	}
477 
478 	iflag = intr_clear();
479 	lock_set(&apic_ioapic_lock);
480 
481 	ret = apic_rebind_all(irqp, cpu);
482 
483 	lock_clear(&apic_ioapic_lock);
484 	intr_restore(iflag);
485 
486 	if (ret) {
487 		*result = EIO;
488 		return (PSM_FAILURE);
489 	}
490 	/*
491 	 * keep tracking the default interrupt cpu binding
492 	 */
493 	irqp->airq_cpu = cpu;
494 
495 	*result = 0;
496 	return (PSM_SUCCESS);
497 }
498 
499 static int
500 apic_grp_set_cpu(int irqno, int new_cpu, int *result)
501 {
502 	dev_info_t *orig_dip;
503 	uint32_t orig_cpu;
504 	ulong_t iflag;
505 	apic_irq_t *irqps[PCI_MSI_MAX_INTRS];
506 	int i;
507 	int cap_ptr;
508 	int msi_mask_off;
509 	ushort_t msi_ctrl;
510 	uint32_t msi_pvm;
511 	ddi_acc_handle_t handle;
512 	int num_vectors = 0;
513 	uint32_t vector;
514 
515 	DDI_INTR_IMPLDBG((CE_CONT, "APIC_GRP_SET_CPU\n"));
516 
517 	/*
518 	 * Take mutex to insure that table doesn't change out from underneath
519 	 * us while we're playing with it.
520 	 */
521 	mutex_enter(&airq_mutex);
522 	irqps[0] = apic_irq_table[irqno];
523 	orig_cpu = irqps[0]->airq_temp_cpu;
524 	orig_dip = irqps[0]->airq_dip;
525 	num_vectors = irqps[0]->airq_intin_no;
526 	vector = irqps[0]->airq_vector;
527 
528 	/* A "group" of 1 */
529 	if (num_vectors == 1) {
530 		mutex_exit(&airq_mutex);
531 		return (apic_set_cpu(irqno, new_cpu, result));
532 	}
533 
534 	*result = ENXIO;
535 
536 	if (irqps[0]->airq_mps_intr_index != MSI_INDEX) {
537 		mutex_exit(&airq_mutex);
538 		DDI_INTR_IMPLDBG((CE_CONT, "set_grp: intr not MSI\n"));
539 		goto set_grp_intr_done;
540 	}
541 	if ((num_vectors < 1) || ((num_vectors - 1) & vector)) {
542 		mutex_exit(&airq_mutex);
543 		DDI_INTR_IMPLDBG((CE_CONT,
544 		    "set_grp: base vec not part of a grp or not aligned: "
545 		    "vec:0x%x, num_vec:0x%x\n", vector, num_vectors));
546 		goto set_grp_intr_done;
547 	}
548 	DDI_INTR_IMPLDBG((CE_CONT, "set_grp: num intrs in grp: %d\n",
549 	    num_vectors));
550 
551 	ASSERT((num_vectors + vector) < APIC_MAX_VECTOR);
552 
553 	*result = EIO;
554 
555 	/*
556 	 * All IRQ entries in the table for the given device will be not
557 	 * shared.  Since they are not shared, the dip in the table will
558 	 * be true to the device of interest.
559 	 */
560 	for (i = 1; i < num_vectors; i++) {
561 		irqps[i] = apic_irq_table[apic_vector_to_irq[vector + i]];
562 		if (irqps[i] == NULL) {
563 			mutex_exit(&airq_mutex);
564 			goto set_grp_intr_done;
565 		}
566 #ifdef DEBUG
567 		/* Sanity check: CPU and dip is the same for all entries. */
568 		if ((irqps[i]->airq_dip != orig_dip) ||
569 		    (irqps[i]->airq_temp_cpu != orig_cpu)) {
570 			mutex_exit(&airq_mutex);
571 			DDI_INTR_IMPLDBG((CE_CONT,
572 			    "set_grp: cpu or dip for vec 0x%x difft than for "
573 			    "vec 0x%x\n", vector, vector + i));
574 			DDI_INTR_IMPLDBG((CE_CONT,
575 			    "  cpu: %d vs %d, dip: 0x%p vs 0x%p\n", orig_cpu,
576 			    irqps[i]->airq_temp_cpu, (void *)orig_dip,
577 			    (void *)irqps[i]->airq_dip));
578 			goto set_grp_intr_done;
579 		}
580 #endif /* DEBUG */
581 	}
582 	mutex_exit(&airq_mutex);
583 
584 	cap_ptr = i_ddi_get_msi_msix_cap_ptr(orig_dip);
585 	handle = i_ddi_get_pci_config_handle(orig_dip);
586 	msi_ctrl = pci_config_get16(handle, cap_ptr + PCI_MSI_CTRL);
587 
588 	/* MSI Per vector masking is supported. */
589 	if (msi_ctrl & PCI_MSI_PVM_MASK) {
590 		if (msi_ctrl &  PCI_MSI_64BIT_MASK)
591 			msi_mask_off = cap_ptr + PCI_MSI_64BIT_MASKBITS;
592 		else
593 			msi_mask_off = cap_ptr + PCI_MSI_32BIT_MASK;
594 		msi_pvm = pci_config_get32(handle, msi_mask_off);
595 		pci_config_put32(handle, msi_mask_off, (uint32_t)-1);
596 		DDI_INTR_IMPLDBG((CE_CONT,
597 		    "set_grp: pvm supported.  Mask set to 0x%x\n",
598 		    pci_config_get32(handle, msi_mask_off)));
599 	}
600 
601 	iflag = intr_clear();
602 	lock_set(&apic_ioapic_lock);
603 
604 	/*
605 	 * Do the first rebind and check for errors.  Apic_rebind_all returns
606 	 * an error if the CPU is not accepting interrupts.  If the first one
607 	 * succeeds they all will.
608 	 */
609 	if (apic_rebind_all(irqps[0], new_cpu))
610 		(void) apic_rebind_all(irqps[0], orig_cpu);
611 	else {
612 		irqps[0]->airq_cpu = new_cpu;
613 
614 		for (i = 1; i < num_vectors; i++) {
615 			(void) apic_rebind_all(irqps[i], new_cpu);
616 			irqps[i]->airq_cpu = new_cpu;
617 		}
618 		*result = 0;	/* SUCCESS */
619 	}
620 
621 	lock_clear(&apic_ioapic_lock);
622 	intr_restore(iflag);
623 
624 	/* Reenable vectors if per vector masking is supported. */
625 	if (msi_ctrl & PCI_MSI_PVM_MASK) {
626 		pci_config_put32(handle, msi_mask_off, msi_pvm);
627 		DDI_INTR_IMPLDBG((CE_CONT,
628 		    "set_grp: pvm supported.  Mask restored to 0x%x\n",
629 		    pci_config_get32(handle, msi_mask_off)));
630 	}
631 
632 set_grp_intr_done:
633 	if (*result != 0)
634 		return (PSM_FAILURE);
635 
636 	return (PSM_SUCCESS);
637 }
638 
639 int
640 apic_get_vector_intr_info(int vecirq, apic_get_intr_t *intr_params_p)
641 {
642 	struct autovec *av_dev;
643 	uchar_t irqno;
644 	int i;
645 	apic_irq_t *irq_p;
646 
647 	/* Sanity check the vector/irq argument. */
648 	ASSERT((vecirq >= 0) || (vecirq <= APIC_MAX_VECTOR));
649 
650 	mutex_enter(&airq_mutex);
651 
652 	/*
653 	 * Convert the vecirq arg to an irq using vector_to_irq table
654 	 * if the arg is a vector.  Pass thru if already an irq.
655 	 */
656 	if ((intr_params_p->avgi_req_flags & PSMGI_INTRBY_FLAGS) ==
657 	    PSMGI_INTRBY_VEC)
658 		irqno = apic_vector_to_irq[vecirq];
659 	else
660 		irqno = vecirq;
661 
662 	irq_p = apic_irq_table[irqno];
663 
664 	if ((irq_p == NULL) ||
665 	    ((irq_p->airq_mps_intr_index != RESERVE_INDEX) &&
666 	    ((irq_p->airq_temp_cpu == IRQ_UNBOUND) ||
667 	    (irq_p->airq_temp_cpu == IRQ_UNINIT)))) {
668 		mutex_exit(&airq_mutex);
669 		return (PSM_FAILURE);
670 	}
671 
672 	if (intr_params_p->avgi_req_flags & PSMGI_REQ_CPUID) {
673 
674 		/* Get the (temp) cpu from apic_irq table, indexed by irq. */
675 		intr_params_p->avgi_cpu_id = irq_p->airq_temp_cpu;
676 
677 		/* Return user bound info for intrd. */
678 		if (intr_params_p->avgi_cpu_id & IRQ_USER_BOUND) {
679 			intr_params_p->avgi_cpu_id &= ~IRQ_USER_BOUND;
680 			intr_params_p->avgi_cpu_id |= PSMGI_CPU_USER_BOUND;
681 		}
682 	}
683 
684 	if (intr_params_p->avgi_req_flags & PSMGI_REQ_VECTOR)
685 		intr_params_p->avgi_vector = irq_p->airq_vector;
686 
687 	if (intr_params_p->avgi_req_flags &
688 	    (PSMGI_REQ_NUM_DEVS | PSMGI_REQ_GET_DEVS))
689 		/* Get number of devices from apic_irq table shared field. */
690 		intr_params_p->avgi_num_devs = irq_p->airq_share;
691 
692 	if (intr_params_p->avgi_req_flags &  PSMGI_REQ_GET_DEVS) {
693 
694 		intr_params_p->avgi_req_flags  |= PSMGI_REQ_NUM_DEVS;
695 
696 		/* Some devices have NULL dip.  Don't count these. */
697 		if (intr_params_p->avgi_num_devs > 0) {
698 			for (i = 0, av_dev = autovect[irqno].avh_link;
699 			    av_dev; av_dev = av_dev->av_link)
700 				if (av_dev->av_vector && av_dev->av_dip)
701 					i++;
702 			intr_params_p->avgi_num_devs =
703 			    MIN(intr_params_p->avgi_num_devs, i);
704 		}
705 
706 		/* There are no viable dips to return. */
707 		if (intr_params_p->avgi_num_devs == 0)
708 			intr_params_p->avgi_dip_list = NULL;
709 
710 		else {	/* Return list of dips */
711 
712 			/* Allocate space in array for that number of devs. */
713 			intr_params_p->avgi_dip_list = kmem_zalloc(
714 			    intr_params_p->avgi_num_devs *
715 			    sizeof (dev_info_t *),
716 			    KM_SLEEP);
717 
718 			/*
719 			 * Loop through the device list of the autovec table
720 			 * filling in the dip array.
721 			 *
722 			 * Note that the autovect table may have some special
723 			 * entries which contain NULL dips.  These will be
724 			 * ignored.
725 			 */
726 			for (i = 0, av_dev = autovect[irqno].avh_link;
727 			    av_dev; av_dev = av_dev->av_link)
728 				if (av_dev->av_vector && av_dev->av_dip)
729 					intr_params_p->avgi_dip_list[i++] =
730 					    av_dev->av_dip;
731 		}
732 	}
733 
734 	mutex_exit(&airq_mutex);
735 
736 	return (PSM_SUCCESS);
737 }
738 
739 /*
740  * This function provides external interface to the nexus for all
741  * functionalities related to the new DDI interrupt framework.
742  *
743  * Input:
744  * dip     - pointer to the dev_info structure of the requested device
745  * hdlp    - pointer to the internal interrupt handle structure for the
746  *	     requested interrupt
747  * intr_op - opcode for this call
748  * result  - pointer to the integer that will hold the result to be
749  *	     passed back if return value is PSM_SUCCESS
750  *
751  * Output:
752  * return value is either PSM_SUCCESS or PSM_FAILURE
753  */
754 int
755 apic_intr_ops(dev_info_t *dip, ddi_intr_handle_impl_t *hdlp,
756     psm_intr_op_t intr_op, int *result)
757 {
758 	int		cap;
759 	int		count_vec;
760 	int		old_priority;
761 	int		new_priority;
762 	int		new_cpu;
763 	apic_irq_t	*irqp;
764 	struct intrspec *ispec, intr_spec;
765 
766 	DDI_INTR_IMPLDBG((CE_CONT, "apic_intr_ops: dip: %p hdlp: %p "
767 	    "intr_op: %x\n", (void *)dip, (void *)hdlp, intr_op));
768 
769 	ispec = &intr_spec;
770 	ispec->intrspec_pri = hdlp->ih_pri;
771 	ispec->intrspec_vec = hdlp->ih_inum;
772 	ispec->intrspec_func = hdlp->ih_cb_func;
773 
774 	switch (intr_op) {
775 	case PSM_INTR_OP_CHECK_MSI:
776 		/*
777 		 * Check MSI/X is supported or not at APIC level and
778 		 * masked off the MSI/X bits in hdlp->ih_type if not
779 		 * supported before return.  If MSI/X is supported,
780 		 * leave the ih_type unchanged and return.
781 		 *
782 		 * hdlp->ih_type passed in from the nexus has all the
783 		 * interrupt types supported by the device.
784 		 */
785 		if (apic_support_msi == 0) {
786 			/*
787 			 * if apic_support_msi is not set, call
788 			 * apic_check_msi_support() to check whether msi
789 			 * is supported first
790 			 */
791 			if (apic_check_msi_support() == PSM_SUCCESS)
792 				apic_support_msi = 1;
793 			else
794 				apic_support_msi = -1;
795 		}
796 		if (apic_support_msi == 1) {
797 			if (apic_msix_enable)
798 				*result = hdlp->ih_type;
799 			else
800 				*result = hdlp->ih_type & ~DDI_INTR_TYPE_MSIX;
801 		} else
802 			*result = hdlp->ih_type & ~(DDI_INTR_TYPE_MSI |
803 			    DDI_INTR_TYPE_MSIX);
804 		break;
805 	case PSM_INTR_OP_ALLOC_VECTORS:
806 		if (hdlp->ih_type == DDI_INTR_TYPE_MSI)
807 			*result = apic_alloc_msi_vectors(dip, hdlp->ih_inum,
808 			    hdlp->ih_scratch1, hdlp->ih_pri,
809 			    (int)(uintptr_t)hdlp->ih_scratch2);
810 		else
811 			*result = apic_alloc_msix_vectors(dip, hdlp->ih_inum,
812 			    hdlp->ih_scratch1, hdlp->ih_pri,
813 			    (int)(uintptr_t)hdlp->ih_scratch2);
814 		break;
815 	case PSM_INTR_OP_FREE_VECTORS:
816 		apic_free_vectors(dip, hdlp->ih_inum, hdlp->ih_scratch1,
817 		    hdlp->ih_pri, hdlp->ih_type);
818 		break;
819 	case PSM_INTR_OP_NAVAIL_VECTORS:
820 		*result = apic_navail_vector(dip, hdlp->ih_pri);
821 		break;
822 	case PSM_INTR_OP_XLATE_VECTOR:
823 		ispec = ((ihdl_plat_t *)hdlp->ih_private)->ip_ispecp;
824 		*result = apic_introp_xlate(dip, ispec, hdlp->ih_type);
825 		if (*result == -1)
826 			return (PSM_FAILURE);
827 		break;
828 	case PSM_INTR_OP_GET_PENDING:
829 		if ((irqp = apic_find_irq(dip, ispec, hdlp->ih_type)) == NULL)
830 			return (PSM_FAILURE);
831 		*result = apic_get_pending(irqp, hdlp->ih_type);
832 		break;
833 	case PSM_INTR_OP_CLEAR_MASK:
834 		if (hdlp->ih_type != DDI_INTR_TYPE_FIXED)
835 			return (PSM_FAILURE);
836 		irqp = apic_find_irq(dip, ispec, hdlp->ih_type);
837 		if (irqp == NULL)
838 			return (PSM_FAILURE);
839 		apic_clear_mask(irqp);
840 		break;
841 	case PSM_INTR_OP_SET_MASK:
842 		if (hdlp->ih_type != DDI_INTR_TYPE_FIXED)
843 			return (PSM_FAILURE);
844 		if ((irqp = apic_find_irq(dip, ispec, hdlp->ih_type)) == NULL)
845 			return (PSM_FAILURE);
846 		apic_set_mask(irqp);
847 		break;
848 	case PSM_INTR_OP_GET_CAP:
849 		cap = DDI_INTR_FLAG_PENDING;
850 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED)
851 			cap |= DDI_INTR_FLAG_MASKABLE;
852 		*result = cap;
853 		break;
854 	case PSM_INTR_OP_GET_SHARED:
855 		if (hdlp->ih_type != DDI_INTR_TYPE_FIXED)
856 			return (PSM_FAILURE);
857 		ispec = ((ihdl_plat_t *)hdlp->ih_private)->ip_ispecp;
858 		if ((irqp = apic_find_irq(dip, ispec, hdlp->ih_type)) == NULL)
859 			return (PSM_FAILURE);
860 		*result = (irqp->airq_share > 1) ? 1: 0;
861 		break;
862 	case PSM_INTR_OP_SET_PRI:
863 		old_priority = hdlp->ih_pri;	/* save old value */
864 		new_priority = *(int *)result;	/* try the new value */
865 
866 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED) {
867 			return (PSM_SUCCESS);
868 		}
869 
870 		/* Now allocate the vectors */
871 		if (hdlp->ih_type == DDI_INTR_TYPE_MSI) {
872 			/* SET_PRI does not support the case of multiple MSI */
873 			if (i_ddi_intr_get_current_nintrs(hdlp->ih_dip) > 1)
874 				return (PSM_FAILURE);
875 
876 			count_vec = apic_alloc_msi_vectors(dip, hdlp->ih_inum,
877 			    1, new_priority,
878 			    DDI_INTR_ALLOC_STRICT);
879 		} else {
880 			count_vec = apic_alloc_msix_vectors(dip, hdlp->ih_inum,
881 			    1, new_priority,
882 			    DDI_INTR_ALLOC_STRICT);
883 		}
884 
885 		/* Did we get new vectors? */
886 		if (!count_vec)
887 			return (PSM_FAILURE);
888 
889 		/* Finally, free the previously allocated vectors */
890 		apic_free_vectors(dip, hdlp->ih_inum, count_vec,
891 		    old_priority, hdlp->ih_type);
892 		break;
893 	case PSM_INTR_OP_SET_CPU:
894 	case PSM_INTR_OP_GRP_SET_CPU:
895 		/*
896 		 * The interrupt handle given here has been allocated
897 		 * specifically for this command, and ih_private carries
898 		 * a CPU value.
899 		 */
900 		new_cpu = (int)(intptr_t)hdlp->ih_private;
901 		if (!apic_cpu_in_range(new_cpu)) {
902 			DDI_INTR_IMPLDBG((CE_CONT,
903 			    "[grp_]set_cpu: cpu out of range: %d\n", new_cpu));
904 			*result = EINVAL;
905 			return (PSM_FAILURE);
906 		}
907 		if (hdlp->ih_vector > APIC_MAX_VECTOR) {
908 			DDI_INTR_IMPLDBG((CE_CONT,
909 			    "[grp_]set_cpu: vector out of range: %d\n",
910 			    hdlp->ih_vector));
911 			*result = EINVAL;
912 			return (PSM_FAILURE);
913 		}
914 		if ((hdlp->ih_flags & PSMGI_INTRBY_FLAGS) == PSMGI_INTRBY_VEC)
915 			hdlp->ih_vector = apic_vector_to_irq[hdlp->ih_vector];
916 		if (intr_op == PSM_INTR_OP_SET_CPU) {
917 			if (apic_set_cpu(hdlp->ih_vector, new_cpu, result) !=
918 			    PSM_SUCCESS)
919 				return (PSM_FAILURE);
920 		} else {
921 			if (apic_grp_set_cpu(hdlp->ih_vector, new_cpu,
922 			    result) != PSM_SUCCESS)
923 				return (PSM_FAILURE);
924 		}
925 		break;
926 	case PSM_INTR_OP_GET_INTR:
927 		/*
928 		 * The interrupt handle given here has been allocated
929 		 * specifically for this command, and ih_private carries
930 		 * a pointer to a apic_get_intr_t.
931 		 */
932 		if (apic_get_vector_intr_info(
933 		    hdlp->ih_vector, hdlp->ih_private) != PSM_SUCCESS)
934 			return (PSM_FAILURE);
935 		break;
936 	case PSM_INTR_OP_APIC_TYPE:
937 		((apic_get_type_t *)(hdlp->ih_private))->avgi_type =
938 		    apic_get_apic_type();
939 		((apic_get_type_t *)(hdlp->ih_private))->avgi_num_intr =
940 		    APIC_MAX_VECTOR;
941 		((apic_get_type_t *)(hdlp->ih_private))->avgi_num_cpu =
942 		    boot_ncpus;
943 		hdlp->ih_ver = apic_get_apic_version();
944 		break;
945 	case PSM_INTR_OP_SET_CAP:
946 	default:
947 		return (PSM_FAILURE);
948 	}
949 	return (PSM_SUCCESS);
950 }
951