xref: /titanic_52/usr/src/uts/i86pc/io/apix/apix.c (revision d828baa435f064b9a001576f57c5f19a0a32755c)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
24  */
25 /*
26  * Copyright (c) 2010, Intel Corporation.
27  * All rights reserved.
28  */
29 /*
30  * Copyright (c) 2017, Joyent, Inc.  All rights reserved.
31  */
32 
33 /*
34  * To understand how the apix module interacts with the interrupt subsystem read
35  * the theory statement in uts/i86pc/os/intr.c.
36  */
37 
38 /*
39  * PSMI 1.1 extensions are supported only in 2.6 and later versions.
40  * PSMI 1.2 extensions are supported only in 2.7 and later versions.
41  * PSMI 1.3 and 1.4 extensions are supported in Solaris 10.
42  * PSMI 1.5 extensions are supported in Solaris Nevada.
43  * PSMI 1.6 extensions are supported in Solaris Nevada.
44  * PSMI 1.7 extensions are supported in Solaris Nevada.
45  */
46 #define	PSMI_1_7
47 
48 #include <sys/processor.h>
49 #include <sys/time.h>
50 #include <sys/psm.h>
51 #include <sys/smp_impldefs.h>
52 #include <sys/cram.h>
53 #include <sys/acpi/acpi.h>
54 #include <sys/acpica.h>
55 #include <sys/psm_common.h>
56 #include <sys/pit.h>
57 #include <sys/ddi.h>
58 #include <sys/sunddi.h>
59 #include <sys/ddi_impldefs.h>
60 #include <sys/pci.h>
61 #include <sys/promif.h>
62 #include <sys/x86_archext.h>
63 #include <sys/cpc_impl.h>
64 #include <sys/uadmin.h>
65 #include <sys/panic.h>
66 #include <sys/debug.h>
67 #include <sys/archsystm.h>
68 #include <sys/trap.h>
69 #include <sys/machsystm.h>
70 #include <sys/sysmacros.h>
71 #include <sys/cpuvar.h>
72 #include <sys/rm_platter.h>
73 #include <sys/privregs.h>
74 #include <sys/note.h>
75 #include <sys/pci_intr_lib.h>
76 #include <sys/spl.h>
77 #include <sys/clock.h>
78 #include <sys/cyclic.h>
79 #include <sys/dditypes.h>
80 #include <sys/sunddi.h>
81 #include <sys/x_call.h>
82 #include <sys/reboot.h>
83 #include <sys/mach_intr.h>
84 #include <sys/apix.h>
85 #include <sys/apix_irm_impl.h>
86 
87 static int apix_probe();
88 static void apix_init();
89 static void apix_picinit(void);
90 static int apix_intr_enter(int, int *);
91 static void apix_intr_exit(int, int);
92 static void apix_setspl(int);
93 static int apix_disable_intr(processorid_t);
94 static void apix_enable_intr(processorid_t);
95 static int apix_get_clkvect(int);
96 static int apix_get_ipivect(int, int);
97 static void apix_post_cyclic_setup(void *);
98 static int apix_post_cpu_start();
99 static int apix_intr_ops(dev_info_t *, ddi_intr_handle_impl_t *,
100     psm_intr_op_t, int *);
101 
102 /*
103  * Helper functions for apix_intr_ops()
104  */
105 static void apix_redistribute_compute(void);
106 static int apix_get_pending(apix_vector_t *);
107 static apix_vector_t *apix_get_req_vector(ddi_intr_handle_impl_t *, ushort_t);
108 static int apix_get_intr_info(ddi_intr_handle_impl_t *, apic_get_intr_t *);
109 static char *apix_get_apic_type(void);
110 static int apix_intx_get_pending(int);
111 static void apix_intx_set_mask(int irqno);
112 static void apix_intx_clear_mask(int irqno);
113 static int apix_intx_get_shared(int irqno);
114 static void apix_intx_set_shared(int irqno, int delta);
115 static apix_vector_t *apix_intx_xlate_vector(dev_info_t *, int,
116     struct intrspec *);
117 static int apix_intx_alloc_vector(dev_info_t *, int, struct intrspec *);
118 
119 extern int apic_clkinit(int);
120 
121 /* IRM initialization for APIX PSM module */
122 extern void apix_irm_init(void);
123 
124 extern int irm_enable;
125 
126 /*
127  *	Local static data
128  */
129 static struct	psm_ops apix_ops = {
130 	apix_probe,
131 
132 	apix_init,
133 	apix_picinit,
134 	apix_intr_enter,
135 	apix_intr_exit,
136 	apix_setspl,
137 	apix_addspl,
138 	apix_delspl,
139 	apix_disable_intr,
140 	apix_enable_intr,
141 	NULL,			/* psm_softlvl_to_irq */
142 	NULL,			/* psm_set_softintr */
143 
144 	apic_set_idlecpu,
145 	apic_unset_idlecpu,
146 
147 	apic_clkinit,
148 	apix_get_clkvect,
149 	NULL,			/* psm_hrtimeinit */
150 	apic_gethrtime,
151 
152 	apic_get_next_processorid,
153 	apic_cpu_start,
154 	apix_post_cpu_start,
155 	apic_shutdown,
156 	apix_get_ipivect,
157 	apic_send_ipi,
158 
159 	NULL,			/* psm_translate_irq */
160 	NULL,			/* psm_notify_error */
161 	NULL,			/* psm_notify_func */
162 	apic_timer_reprogram,
163 	apic_timer_enable,
164 	apic_timer_disable,
165 	apix_post_cyclic_setup,
166 	apic_preshutdown,
167 	apix_intr_ops,		/* Advanced DDI Interrupt framework */
168 	apic_state,		/* save, restore apic state for S3 */
169 	apic_cpu_ops,		/* CPU control interface. */
170 };
171 
172 struct psm_ops *psmops = &apix_ops;
173 
174 static struct	psm_info apix_psm_info = {
175 	PSM_INFO_VER01_7,			/* version */
176 	PSM_OWN_EXCLUSIVE,			/* ownership */
177 	&apix_ops,				/* operation */
178 	APIX_NAME,				/* machine name */
179 	"apix MPv1.4 compatible",
180 };
181 
182 static void *apix_hdlp;
183 
184 static int apix_is_enabled = 0;
185 
186 /*
187  * Flag to indicate if APIX is to be enabled only for platforms
188  * with specific hw feature(s).
189  */
190 int apix_hw_chk_enable = 1;
191 
192 /*
193  * Hw features that are checked for enabling APIX support.
194  */
195 #define	APIX_SUPPORT_X2APIC	0x00000001
196 uint_t apix_supported_hw = APIX_SUPPORT_X2APIC;
197 
198 /*
199  * apix_lock is used for cpu selection and vector re-binding
200  */
201 lock_t apix_lock;
202 apix_impl_t *apixs[NCPU];
203 /*
204  * Mapping between device interrupt and the allocated vector. Indexed
205  * by major number.
206  */
207 apix_dev_vector_t **apix_dev_vector;
208 /*
209  * Mapping between device major number and cpu id. It gets used
210  * when interrupt binding policy round robin with affinity is
211  * applied. With that policy, devices with the same major number
212  * will be bound to the same CPU.
213  */
214 processorid_t *apix_major_to_cpu;	/* major to cpu mapping */
215 kmutex_t apix_mutex;	/* for apix_dev_vector & apix_major_to_cpu */
216 
217 int apix_nipis = 16;	/* Maximum number of IPIs */
218 /*
219  * Maximum number of vectors in a CPU that can be used for interrupt
220  * allocation (including IPIs and the reserved vectors).
221  */
222 int apix_cpu_nvectors = APIX_NVECTOR;
223 
224 /* gcpu.h */
225 
226 extern void apic_do_interrupt(struct regs *rp, trap_trace_rec_t *ttp);
227 extern void apic_change_eoi();
228 
229 /*
230  *	This is the loadable module wrapper
231  */
232 
233 int
234 _init(void)
235 {
236 	if (apic_coarse_hrtime)
237 		apix_ops.psm_gethrtime = &apic_gettime;
238 	return (psm_mod_init(&apix_hdlp, &apix_psm_info));
239 }
240 
241 int
242 _fini(void)
243 {
244 	return (psm_mod_fini(&apix_hdlp, &apix_psm_info));
245 }
246 
247 int
248 _info(struct modinfo *modinfop)
249 {
250 	return (psm_mod_info(&apix_hdlp, &apix_psm_info, modinfop));
251 }
252 
253 static int
254 apix_probe()
255 {
256 	int rval;
257 
258 	if (apix_enable == 0)
259 		return (PSM_FAILURE);
260 
261 	/* check for hw features if specified  */
262 	if (apix_hw_chk_enable) {
263 		/* check if x2APIC mode is supported */
264 		if ((apix_supported_hw & APIX_SUPPORT_X2APIC) ==
265 		    APIX_SUPPORT_X2APIC) {
266 			if (apic_local_mode() == LOCAL_X2APIC) {
267 				/* x2APIC mode activated by BIOS, switch ops */
268 				apic_mode = LOCAL_X2APIC;
269 				apic_change_ops();
270 			} else if (!apic_detect_x2apic()) {
271 				/* x2APIC mode is not supported in the hw */
272 				apix_enable = 0;
273 			}
274 		}
275 		if (apix_enable == 0)
276 			return (PSM_FAILURE);
277 	}
278 
279 	rval = apic_probe_common(apix_psm_info.p_mach_idstring);
280 	if (rval == PSM_SUCCESS)
281 		apix_is_enabled = 1;
282 	else
283 		apix_is_enabled = 0;
284 	return (rval);
285 }
286 
287 /*
288  * Initialize the data structures needed by pcplusmpx module.
289  * Specifically, the data structures used by addspl() and delspl()
290  * routines.
291  */
292 static void
293 apix_softinit()
294 {
295 	int i, *iptr;
296 	apix_impl_t *hdlp;
297 	int nproc;
298 
299 	nproc = max(apic_nproc, apic_max_nproc);
300 
301 	hdlp = kmem_zalloc(nproc * sizeof (apix_impl_t), KM_SLEEP);
302 	for (i = 0; i < nproc; i++) {
303 		apixs[i] = &hdlp[i];
304 		apixs[i]->x_cpuid = i;
305 		LOCK_INIT_CLEAR(&apixs[i]->x_lock);
306 	}
307 
308 	/* cpu 0 is always up (for now) */
309 	apic_cpus[0].aci_status = APIC_CPU_ONLINE | APIC_CPU_INTR_ENABLE;
310 
311 	iptr = (int *)&apic_irq_table[0];
312 	for (i = 0; i <= APIC_MAX_VECTOR; i++) {
313 		apic_level_intr[i] = 0;
314 		*iptr++ = NULL;
315 	}
316 	mutex_init(&airq_mutex, NULL, MUTEX_DEFAULT, NULL);
317 
318 	apix_dev_vector = kmem_zalloc(sizeof (apix_dev_vector_t *) * devcnt,
319 	    KM_SLEEP);
320 
321 	if (apic_intr_policy == INTR_ROUND_ROBIN_WITH_AFFINITY) {
322 		apix_major_to_cpu = kmem_zalloc(sizeof (int) * devcnt,
323 		    KM_SLEEP);
324 		for (i = 0; i < devcnt; i++)
325 			apix_major_to_cpu[i] = IRQ_UNINIT;
326 	}
327 
328 	mutex_init(&apix_mutex, NULL, MUTEX_DEFAULT, NULL);
329 }
330 
331 static int
332 apix_get_pending_spl(void)
333 {
334 	int cpuid = CPU->cpu_id;
335 
336 	return (bsrw_insn(apixs[cpuid]->x_intr_pending));
337 }
338 
339 static uintptr_t
340 apix_get_intr_handler(int cpu, short vec)
341 {
342 	apix_vector_t *apix_vector;
343 
344 	ASSERT(cpu < apic_nproc && vec < APIX_NVECTOR);
345 	if (cpu >= apic_nproc)
346 		return (NULL);
347 
348 	apix_vector = apixs[cpu]->x_vectbl[vec];
349 
350 	return ((uintptr_t)(apix_vector->v_autovect));
351 }
352 
353 static void
354 apix_init()
355 {
356 	extern void (*do_interrupt_common)(struct regs *, trap_trace_rec_t *);
357 
358 	APIC_VERBOSE(INIT, (CE_CONT, "apix: psm_softinit\n"));
359 
360 	do_interrupt_common = apix_do_interrupt;
361 	addintr = apix_add_avintr;
362 	remintr = apix_rem_avintr;
363 	get_pending_spl = apix_get_pending_spl;
364 	get_intr_handler = apix_get_intr_handler;
365 	psm_get_localapicid = apic_get_localapicid;
366 	psm_get_ioapicid = apic_get_ioapicid;
367 
368 	apix_softinit();
369 
370 #if !defined(__amd64)
371 	if (cpuid_have_cr8access(CPU))
372 		apic_have_32bit_cr8 = 1;
373 #endif
374 
375 	/*
376 	 * Initialize IRM pool parameters
377 	 */
378 	if (irm_enable) {
379 		int	i;
380 		int	lowest_irq;
381 		int	highest_irq;
382 
383 		/* number of CPUs present */
384 		apix_irminfo.apix_ncpus = apic_nproc;
385 		/* total number of entries in all of the IOAPICs present */
386 		lowest_irq = apic_io_vectbase[0];
387 		highest_irq = apic_io_vectend[0];
388 		for (i = 1; i < apic_io_max; i++) {
389 			if (apic_io_vectbase[i] < lowest_irq)
390 				lowest_irq = apic_io_vectbase[i];
391 			if (apic_io_vectend[i] > highest_irq)
392 				highest_irq = apic_io_vectend[i];
393 		}
394 		apix_irminfo.apix_ioapic_max_vectors =
395 		    highest_irq - lowest_irq + 1;
396 		/*
397 		 * Number of available per-CPU vectors excluding
398 		 * reserved vectors for Dtrace, int80, system-call,
399 		 * fast-trap, etc.
400 		 */
401 		apix_irminfo.apix_per_cpu_vectors = APIX_NAVINTR -
402 		    APIX_SW_RESERVED_VECTORS;
403 
404 		/* Number of vectors (pre) allocated (SCI and HPET) */
405 		apix_irminfo.apix_vectors_allocated = 0;
406 		if (apic_hpet_vect != -1)
407 			apix_irminfo.apix_vectors_allocated++;
408 		if (apic_sci_vect != -1)
409 			apix_irminfo.apix_vectors_allocated++;
410 	}
411 }
412 
413 static void
414 apix_init_intr()
415 {
416 	processorid_t	cpun = psm_get_cpu_id();
417 	uint_t nlvt;
418 	uint32_t svr = AV_UNIT_ENABLE | APIC_SPUR_INTR;
419 	extern void cmi_cmci_trap(void);
420 
421 	apic_reg_ops->apic_write_task_reg(APIC_MASK_ALL);
422 
423 	if (apic_mode == LOCAL_APIC) {
424 		/*
425 		 * We are running APIC in MMIO mode.
426 		 */
427 		if (apic_flat_model) {
428 			apic_reg_ops->apic_write(APIC_FORMAT_REG,
429 			    APIC_FLAT_MODEL);
430 		} else {
431 			apic_reg_ops->apic_write(APIC_FORMAT_REG,
432 			    APIC_CLUSTER_MODEL);
433 		}
434 
435 		apic_reg_ops->apic_write(APIC_DEST_REG,
436 		    AV_HIGH_ORDER >> cpun);
437 	}
438 
439 	if (apic_directed_EOI_supported()) {
440 		/*
441 		 * Setting the 12th bit in the Spurious Interrupt Vector
442 		 * Register suppresses broadcast EOIs generated by the local
443 		 * APIC. The suppression of broadcast EOIs happens only when
444 		 * interrupts are level-triggered.
445 		 */
446 		svr |= APIC_SVR_SUPPRESS_BROADCAST_EOI;
447 	}
448 
449 	/* need to enable APIC before unmasking NMI */
450 	apic_reg_ops->apic_write(APIC_SPUR_INT_REG, svr);
451 
452 	/*
453 	 * Presence of an invalid vector with delivery mode AV_FIXED can
454 	 * cause an error interrupt, even if the entry is masked...so
455 	 * write a valid vector to LVT entries along with the mask bit
456 	 */
457 
458 	/* All APICs have timer and LINT0/1 */
459 	apic_reg_ops->apic_write(APIC_LOCAL_TIMER, AV_MASK|APIC_RESV_IRQ);
460 	apic_reg_ops->apic_write(APIC_INT_VECT0, AV_MASK|APIC_RESV_IRQ);
461 	apic_reg_ops->apic_write(APIC_INT_VECT1, AV_NMI);	/* enable NMI */
462 
463 	/*
464 	 * On integrated APICs, the number of LVT entries is
465 	 * 'Max LVT entry' + 1; on 82489DX's (non-integrated
466 	 * APICs), nlvt is "3" (LINT0, LINT1, and timer)
467 	 */
468 
469 	if (apic_cpus[cpun].aci_local_ver < APIC_INTEGRATED_VERS) {
470 		nlvt = 3;
471 	} else {
472 		nlvt = ((apic_reg_ops->apic_read(APIC_VERS_REG) >> 16) &
473 		    0xFF) + 1;
474 	}
475 
476 	if (nlvt >= 5) {
477 		/* Enable performance counter overflow interrupt */
478 
479 		if (!is_x86_feature(x86_featureset, X86FSET_MSR))
480 			apic_enable_cpcovf_intr = 0;
481 		if (apic_enable_cpcovf_intr) {
482 			if (apic_cpcovf_vect == 0) {
483 				int ipl = APIC_PCINT_IPL;
484 
485 				apic_cpcovf_vect = apix_get_ipivect(ipl, -1);
486 				ASSERT(apic_cpcovf_vect);
487 
488 				(void) add_avintr(NULL, ipl,
489 				    (avfunc)kcpc_hw_overflow_intr,
490 				    "apic pcint", apic_cpcovf_vect,
491 				    NULL, NULL, NULL, NULL);
492 				kcpc_hw_overflow_intr_installed = 1;
493 				kcpc_hw_enable_cpc_intr =
494 				    apic_cpcovf_mask_clear;
495 			}
496 			apic_reg_ops->apic_write(APIC_PCINT_VECT,
497 			    apic_cpcovf_vect);
498 		}
499 	}
500 
501 	if (nlvt >= 6) {
502 		/* Only mask TM intr if the BIOS apparently doesn't use it */
503 
504 		uint32_t lvtval;
505 
506 		lvtval = apic_reg_ops->apic_read(APIC_THERM_VECT);
507 		if (((lvtval & AV_MASK) == AV_MASK) ||
508 		    ((lvtval & AV_DELIV_MODE) != AV_SMI)) {
509 			apic_reg_ops->apic_write(APIC_THERM_VECT,
510 			    AV_MASK|APIC_RESV_IRQ);
511 		}
512 	}
513 
514 	/* Enable error interrupt */
515 
516 	if (nlvt >= 4 && apic_enable_error_intr) {
517 		if (apic_errvect == 0) {
518 			int ipl = 0xf;	/* get highest priority intr */
519 			apic_errvect = apix_get_ipivect(ipl, -1);
520 			ASSERT(apic_errvect);
521 			/*
522 			 * Not PSMI compliant, but we are going to merge
523 			 * with ON anyway
524 			 */
525 			(void) add_avintr(NULL, ipl,
526 			    (avfunc)apic_error_intr, "apic error intr",
527 			    apic_errvect, NULL, NULL, NULL, NULL);
528 		}
529 		apic_reg_ops->apic_write(APIC_ERR_VECT, apic_errvect);
530 		apic_reg_ops->apic_write(APIC_ERROR_STATUS, 0);
531 		apic_reg_ops->apic_write(APIC_ERROR_STATUS, 0);
532 	}
533 
534 	/* Enable CMCI interrupt */
535 	if (cmi_enable_cmci) {
536 		mutex_enter(&cmci_cpu_setup_lock);
537 		if (cmci_cpu_setup_registered == 0) {
538 			mutex_enter(&cpu_lock);
539 			register_cpu_setup_func(cmci_cpu_setup, NULL);
540 			mutex_exit(&cpu_lock);
541 			cmci_cpu_setup_registered = 1;
542 		}
543 		mutex_exit(&cmci_cpu_setup_lock);
544 
545 		if (apic_cmci_vect == 0) {
546 			int ipl = 0x2;
547 			apic_cmci_vect = apix_get_ipivect(ipl, -1);
548 			ASSERT(apic_cmci_vect);
549 
550 			(void) add_avintr(NULL, ipl,
551 			    (avfunc)cmi_cmci_trap, "apic cmci intr",
552 			    apic_cmci_vect, NULL, NULL, NULL, NULL);
553 		}
554 		apic_reg_ops->apic_write(APIC_CMCI_VECT, apic_cmci_vect);
555 	}
556 
557 	apic_reg_ops->apic_write_task_reg(0);
558 }
559 
560 static void
561 apix_picinit(void)
562 {
563 	int i, j;
564 	uint_t isr;
565 
566 	APIC_VERBOSE(INIT, (CE_CONT, "apix: psm_picinit\n"));
567 
568 	/*
569 	 * initialize interrupt remapping before apic
570 	 * hardware initialization
571 	 */
572 	apic_intrmap_init(apic_mode);
573 	if (apic_vt_ops == psm_vt_ops)
574 		apix_mul_ioapic_method = APIC_MUL_IOAPIC_IIR;
575 
576 	/*
577 	 * On UniSys Model 6520, the BIOS leaves vector 0x20 isr
578 	 * bit on without clearing it with EOI.  Since softint
579 	 * uses vector 0x20 to interrupt itself, so softint will
580 	 * not work on this machine.  In order to fix this problem
581 	 * a check is made to verify all the isr bits are clear.
582 	 * If not, EOIs are issued to clear the bits.
583 	 */
584 	for (i = 7; i >= 1; i--) {
585 		isr = apic_reg_ops->apic_read(APIC_ISR_REG + (i * 4));
586 		if (isr != 0)
587 			for (j = 0; ((j < 32) && (isr != 0)); j++)
588 				if (isr & (1 << j)) {
589 					apic_reg_ops->apic_write(
590 					    APIC_EOI_REG, 0);
591 					isr &= ~(1 << j);
592 					apic_error |= APIC_ERR_BOOT_EOI;
593 				}
594 	}
595 
596 	/* set a flag so we know we have run apic_picinit() */
597 	apic_picinit_called = 1;
598 	LOCK_INIT_CLEAR(&apic_gethrtime_lock);
599 	LOCK_INIT_CLEAR(&apic_ioapic_lock);
600 	LOCK_INIT_CLEAR(&apic_error_lock);
601 	LOCK_INIT_CLEAR(&apic_mode_switch_lock);
602 
603 	picsetup();	 /* initialise the 8259 */
604 
605 	/* add nmi handler - least priority nmi handler */
606 	LOCK_INIT_CLEAR(&apic_nmi_lock);
607 
608 	if (!psm_add_nmintr(0, (avfunc) apic_nmi_intr,
609 	    "apix NMI handler", (caddr_t)NULL))
610 		cmn_err(CE_WARN, "apix: Unable to add nmi handler");
611 
612 	apix_init_intr();
613 
614 	/* enable apic mode if imcr present */
615 	if (apic_imcrp) {
616 		outb(APIC_IMCR_P1, (uchar_t)APIC_IMCR_SELECT);
617 		outb(APIC_IMCR_P2, (uchar_t)APIC_IMCR_APIC);
618 	}
619 
620 	ioapix_init_intr(IOAPIC_MASK);
621 
622 	/* setup global IRM pool if applicable */
623 	if (irm_enable)
624 		apix_irm_init();
625 }
626 
627 static __inline__ void
628 apix_send_eoi(void)
629 {
630 	if (apic_mode == LOCAL_APIC)
631 		LOCAL_APIC_WRITE_REG(APIC_EOI_REG, 0);
632 	else
633 		X2APIC_WRITE(APIC_EOI_REG, 0);
634 }
635 
636 /*
637  * platform_intr_enter
638  *
639  *	Called at the beginning of the interrupt service routine, but unlike
640  *	pcplusmp, does not mask interrupts. An EOI is given to the interrupt
641  *	controller to enable other HW interrupts but interrupts are still
642  * 	masked by the IF flag.
643  *
644  *	Return -1 for spurious interrupts
645  *
646  */
647 static int
648 apix_intr_enter(int ipl, int *vectorp)
649 {
650 	struct cpu *cpu = CPU;
651 	uint32_t cpuid = CPU->cpu_id;
652 	apic_cpus_info_t *cpu_infop;
653 	uchar_t vector;
654 	apix_vector_t *vecp;
655 	int nipl = -1;
656 
657 	/*
658 	 * The real vector delivered is (*vectorp + 0x20), but our caller
659 	 * subtracts 0x20 from the vector before passing it to us.
660 	 * (That's why APIC_BASE_VECT is 0x20.)
661 	 */
662 	vector = *vectorp = (uchar_t)*vectorp + APIC_BASE_VECT;
663 
664 	cpu_infop = &apic_cpus[cpuid];
665 	if (vector == APIC_SPUR_INTR) {
666 		cpu_infop->aci_spur_cnt++;
667 		return (APIC_INT_SPURIOUS);
668 	}
669 
670 	vecp = xv_vector(cpuid, vector);
671 	if (vecp == NULL) {
672 		if (APIX_IS_FAKE_INTR(vector))
673 			nipl = apix_rebindinfo.i_pri;
674 		apix_send_eoi();
675 		return (nipl);
676 	}
677 	nipl = vecp->v_pri;
678 
679 	/* if interrupted by the clock, increment apic_nsec_since_boot */
680 	if (vector == (apic_clkvect + APIC_BASE_VECT)) {
681 		if (!apic_oneshot) {
682 			/* NOTE: this is not MT aware */
683 			apic_hrtime_stamp++;
684 			apic_nsec_since_boot += apic_nsec_per_intr;
685 			apic_hrtime_stamp++;
686 			last_count_read = apic_hertz_count;
687 			apix_redistribute_compute();
688 		}
689 
690 		apix_send_eoi();
691 
692 		return (nipl);
693 	}
694 
695 	ASSERT(vecp->v_state != APIX_STATE_OBSOLETED);
696 
697 	/* pre-EOI handling for level-triggered interrupts */
698 	if (!APIX_IS_DIRECTED_EOI(apix_mul_ioapic_method) &&
699 	    (vecp->v_type & APIX_TYPE_FIXED) && apic_level_intr[vecp->v_inum])
700 		apix_level_intr_pre_eoi(vecp->v_inum);
701 
702 	/* send back EOI */
703 	apix_send_eoi();
704 
705 	cpu_infop->aci_current[nipl] = vector;
706 	if ((nipl > ipl) && (nipl > cpu->cpu_base_spl)) {
707 		cpu_infop->aci_curipl = (uchar_t)nipl;
708 		cpu_infop->aci_ISR_in_progress |= 1 << nipl;
709 	}
710 
711 #ifdef	DEBUG
712 	if (vector >= APIX_IPI_MIN)
713 		return (nipl);	/* skip IPI */
714 
715 	APIC_DEBUG_BUF_PUT(vector);
716 	APIC_DEBUG_BUF_PUT(vecp->v_inum);
717 	APIC_DEBUG_BUF_PUT(nipl);
718 	APIC_DEBUG_BUF_PUT(psm_get_cpu_id());
719 	if ((apic_stretch_interrupts) && (apic_stretch_ISR & (1 << nipl)))
720 		drv_usecwait(apic_stretch_interrupts);
721 #endif /* DEBUG */
722 
723 	return (nipl);
724 }
725 
726 /*
727  * Any changes made to this function must also change X2APIC
728  * version of intr_exit.
729  */
730 static void
731 apix_intr_exit(int prev_ipl, int arg2)
732 {
733 	int cpuid = psm_get_cpu_id();
734 	apic_cpus_info_t *cpu_infop = &apic_cpus[cpuid];
735 	apix_impl_t *apixp = apixs[cpuid];
736 
737 	UNREFERENCED_1PARAMETER(arg2);
738 
739 	cpu_infop->aci_curipl = (uchar_t)prev_ipl;
740 	/* ISR above current pri could not be in progress */
741 	cpu_infop->aci_ISR_in_progress &= (2 << prev_ipl) - 1;
742 
743 	if (apixp->x_obsoletes != NULL) {
744 		if (APIX_CPU_LOCK_HELD(cpuid))
745 			return;
746 
747 		APIX_ENTER_CPU_LOCK(cpuid);
748 		(void) apix_obsolete_vector(apixp->x_obsoletes);
749 		APIX_LEAVE_CPU_LOCK(cpuid);
750 	}
751 }
752 
753 /*
754  * The pcplusmp setspl code uses the TPR to mask all interrupts at or below the
755  * given ipl, but apix never uses the TPR and we never mask a subset of the
756  * interrupts. They are either all blocked by the IF flag or all can come in.
757  *
758  * For setspl, we mask all interrupts for XC_HI_PIL (15), otherwise, interrupts
759  * can come in if currently enabled by the IF flag. This table shows the state
760  * of the IF flag when we leave this function.
761  *
762  *    curr IF |	ipl == 15	ipl != 15
763  *    --------+---------------------------
764  *       0    |    0		    0
765  *       1    |    0		    1
766  */
767 static void
768 apix_setspl(int ipl)
769 {
770 	/*
771 	 * Interrupts at ipl above this cannot be in progress, so the following
772 	 * mask is ok.
773 	 */
774 	apic_cpus[psm_get_cpu_id()].aci_ISR_in_progress &= (2 << ipl) - 1;
775 
776 	if (ipl == XC_HI_PIL)
777 		cli();
778 }
779 
780 int
781 apix_addspl(int virtvec, int ipl, int min_ipl, int max_ipl)
782 {
783 	uint32_t cpuid = APIX_VIRTVEC_CPU(virtvec);
784 	uchar_t vector = (uchar_t)APIX_VIRTVEC_VECTOR(virtvec);
785 	apix_vector_t *vecp = xv_vector(cpuid, vector);
786 
787 	UNREFERENCED_3PARAMETER(ipl, min_ipl, max_ipl);
788 	ASSERT(vecp != NULL && LOCK_HELD(&apix_lock));
789 
790 	if (vecp->v_type == APIX_TYPE_FIXED)
791 		apix_intx_set_shared(vecp->v_inum, 1);
792 
793 	/* There are more interrupts, so it's already been enabled */
794 	if (vecp->v_share > 1)
795 		return (PSM_SUCCESS);
796 
797 	/* return if it is not hardware interrupt */
798 	if (vecp->v_type == APIX_TYPE_IPI)
799 		return (PSM_SUCCESS);
800 
801 	/*
802 	 * if apix_picinit() has not been called yet, just return.
803 	 * At the end of apic_picinit(), we will call setup_io_intr().
804 	 */
805 	if (!apic_picinit_called)
806 		return (PSM_SUCCESS);
807 
808 	(void) apix_setup_io_intr(vecp);
809 
810 	return (PSM_SUCCESS);
811 }
812 
813 int
814 apix_delspl(int virtvec, int ipl, int min_ipl, int max_ipl)
815 {
816 	uint32_t cpuid = APIX_VIRTVEC_CPU(virtvec);
817 	uchar_t vector = (uchar_t)APIX_VIRTVEC_VECTOR(virtvec);
818 	apix_vector_t *vecp = xv_vector(cpuid, vector);
819 
820 	UNREFERENCED_3PARAMETER(ipl, min_ipl, max_ipl);
821 	ASSERT(vecp != NULL && LOCK_HELD(&apix_lock));
822 
823 	if (vecp->v_type == APIX_TYPE_FIXED)
824 		apix_intx_set_shared(vecp->v_inum, -1);
825 
826 	/* There are more interrupts */
827 	if (vecp->v_share > 1)
828 		return (PSM_SUCCESS);
829 
830 	/* return if it is not hardware interrupt */
831 	if (vecp->v_type == APIX_TYPE_IPI)
832 		return (PSM_SUCCESS);
833 
834 	if (!apic_picinit_called) {
835 		cmn_err(CE_WARN, "apix: delete 0x%x before apic init",
836 		    virtvec);
837 		return (PSM_SUCCESS);
838 	}
839 
840 	apix_disable_vector(vecp);
841 
842 	return (PSM_SUCCESS);
843 }
844 
845 /*
846  * Try and disable all interrupts. We just assign interrupts to other
847  * processors based on policy. If any were bound by user request, we
848  * let them continue and return failure. We do not bother to check
849  * for cache affinity while rebinding.
850  */
851 static int
852 apix_disable_intr(processorid_t cpun)
853 {
854 	apix_impl_t *apixp = apixs[cpun];
855 	apix_vector_t *vecp, *newp;
856 	int bindcpu, i, hardbound = 0, errbound = 0, ret, loop, type;
857 
858 	lock_set(&apix_lock);
859 
860 	apic_cpus[cpun].aci_status &= ~APIC_CPU_INTR_ENABLE;
861 	apic_cpus[cpun].aci_curipl = 0;
862 
863 	/* if this is for SUSPEND operation, skip rebinding */
864 	if (apic_cpus[cpun].aci_status & APIC_CPU_SUSPEND) {
865 		for (i = APIX_AVINTR_MIN; i <= APIX_AVINTR_MAX; i++) {
866 			vecp = apixp->x_vectbl[i];
867 			if (!IS_VECT_ENABLED(vecp))
868 				continue;
869 
870 			apix_disable_vector(vecp);
871 		}
872 		lock_clear(&apix_lock);
873 		return (PSM_SUCCESS);
874 	}
875 
876 	for (i = APIX_AVINTR_MIN; i <= APIX_AVINTR_MAX; i++) {
877 		vecp = apixp->x_vectbl[i];
878 		if (!IS_VECT_ENABLED(vecp))
879 			continue;
880 
881 		if (vecp->v_flags & APIX_VECT_USER_BOUND) {
882 			hardbound++;
883 			continue;
884 		}
885 		type = vecp->v_type;
886 
887 		/*
888 		 * If there are bound interrupts on this cpu, then
889 		 * rebind them to other processors.
890 		 */
891 		loop = 0;
892 		do {
893 			bindcpu = apic_find_cpu(APIC_CPU_INTR_ENABLE);
894 
895 			if (type != APIX_TYPE_MSI)
896 				newp = apix_set_cpu(vecp, bindcpu, &ret);
897 			else
898 				newp = apix_grp_set_cpu(vecp, bindcpu, &ret);
899 		} while ((newp == NULL) && (loop++ < apic_nproc));
900 
901 		if (loop >= apic_nproc) {
902 			errbound++;
903 			cmn_err(CE_WARN, "apix: failed to rebind vector %x/%x",
904 			    vecp->v_cpuid, vecp->v_vector);
905 		}
906 	}
907 
908 	lock_clear(&apix_lock);
909 
910 	if (hardbound || errbound) {
911 		cmn_err(CE_WARN, "Could not disable interrupts on %d"
912 		    "due to user bound interrupts or failed operation",
913 		    cpun);
914 		return (PSM_FAILURE);
915 	}
916 
917 	return (PSM_SUCCESS);
918 }
919 
920 /*
921  * Bind interrupts to specified CPU
922  */
923 static void
924 apix_enable_intr(processorid_t cpun)
925 {
926 	apix_vector_t *vecp;
927 	int i, ret;
928 	processorid_t n;
929 
930 	lock_set(&apix_lock);
931 
932 	apic_cpus[cpun].aci_status |= APIC_CPU_INTR_ENABLE;
933 
934 	/* interrupt enabling for system resume */
935 	if (apic_cpus[cpun].aci_status & APIC_CPU_SUSPEND) {
936 		for (i = APIX_AVINTR_MIN; i <= APIX_AVINTR_MAX; i++) {
937 			vecp = xv_vector(cpun, i);
938 			if (!IS_VECT_ENABLED(vecp))
939 				continue;
940 
941 			apix_enable_vector(vecp);
942 		}
943 		apic_cpus[cpun].aci_status &= ~APIC_CPU_SUSPEND;
944 	}
945 
946 	for (n = 0; n < apic_nproc; n++) {
947 		if (!apic_cpu_in_range(n) || n == cpun ||
948 		    (apic_cpus[n].aci_status & APIC_CPU_INTR_ENABLE) == 0)
949 			continue;
950 
951 		for (i = APIX_AVINTR_MIN; i <= APIX_AVINTR_MAX; i++) {
952 			vecp = xv_vector(n, i);
953 			if (!IS_VECT_ENABLED(vecp) ||
954 			    vecp->v_bound_cpuid != cpun)
955 				continue;
956 
957 			if (vecp->v_type != APIX_TYPE_MSI)
958 				(void) apix_set_cpu(vecp, cpun, &ret);
959 			else
960 				(void) apix_grp_set_cpu(vecp, cpun, &ret);
961 		}
962 	}
963 
964 	lock_clear(&apix_lock);
965 }
966 
967 /*
968  * Allocate vector for IPI
969  * type == -1 indicates it is an internal request. Do not change
970  * resv_vector for these requests.
971  */
972 static int
973 apix_get_ipivect(int ipl, int type)
974 {
975 	uchar_t vector;
976 
977 	if ((vector = apix_alloc_ipi(ipl)) > 0) {
978 		if (type != -1)
979 			apic_resv_vector[ipl] = vector;
980 		return (vector);
981 	}
982 	apic_error |= APIC_ERR_GET_IPIVECT_FAIL;
983 	return (-1);	/* shouldn't happen */
984 }
985 
986 static int
987 apix_get_clkvect(int ipl)
988 {
989 	int vector;
990 
991 	if ((vector = apix_get_ipivect(ipl, -1)) == -1)
992 		return (-1);
993 
994 	apic_clkvect = vector - APIC_BASE_VECT;
995 	APIC_VERBOSE(IPI, (CE_CONT, "apix: clock vector = %x\n",
996 	    apic_clkvect));
997 	return (vector);
998 }
999 
1000 static int
1001 apix_post_cpu_start()
1002 {
1003 	int cpun;
1004 	static int cpus_started = 1;
1005 
1006 	/* We know this CPU + BSP  started successfully. */
1007 	cpus_started++;
1008 
1009 	/*
1010 	 * On BSP we would have enabled X2APIC, if supported by processor,
1011 	 * in acpi_probe(), but on AP we do it here.
1012 	 *
1013 	 * We enable X2APIC mode only if BSP is running in X2APIC & the
1014 	 * local APIC mode of the current CPU is MMIO (xAPIC).
1015 	 */
1016 	if (apic_mode == LOCAL_X2APIC && apic_detect_x2apic() &&
1017 	    apic_local_mode() == LOCAL_APIC) {
1018 		apic_enable_x2apic();
1019 	}
1020 
1021 	/*
1022 	 * Switch back to x2apic IPI sending method for performance when target
1023 	 * CPU has entered x2apic mode.
1024 	 */
1025 	if (apic_mode == LOCAL_X2APIC) {
1026 		apic_switch_ipi_callback(B_FALSE);
1027 	}
1028 
1029 	splx(ipltospl(LOCK_LEVEL));
1030 	apix_init_intr();
1031 
1032 	/*
1033 	 * since some systems don't enable the internal cache on the non-boot
1034 	 * cpus, so we have to enable them here
1035 	 */
1036 	setcr0(getcr0() & ~(CR0_CD | CR0_NW));
1037 
1038 #ifdef	DEBUG
1039 	APIC_AV_PENDING_SET();
1040 #else
1041 	if (apic_mode == LOCAL_APIC)
1042 		APIC_AV_PENDING_SET();
1043 #endif	/* DEBUG */
1044 
1045 	/*
1046 	 * We may be booting, or resuming from suspend; aci_status will
1047 	 * be APIC_CPU_INTR_ENABLE if coming from suspend, so we add the
1048 	 * APIC_CPU_ONLINE flag here rather than setting aci_status completely.
1049 	 */
1050 	cpun = psm_get_cpu_id();
1051 	apic_cpus[cpun].aci_status |= APIC_CPU_ONLINE;
1052 
1053 	apic_reg_ops->apic_write(APIC_DIVIDE_REG, apic_divide_reg_init);
1054 
1055 	return (PSM_SUCCESS);
1056 }
1057 
1058 /*
1059  * If this module needs a periodic handler for the interrupt distribution, it
1060  * can be added here. The argument to the periodic handler is not currently
1061  * used, but is reserved for future.
1062  */
1063 static void
1064 apix_post_cyclic_setup(void *arg)
1065 {
1066 	UNREFERENCED_1PARAMETER(arg);
1067 
1068 	cyc_handler_t cyh;
1069 	cyc_time_t cyt;
1070 
1071 	/* cpu_lock is held */
1072 	/* set up a periodic handler for intr redistribution */
1073 
1074 	/*
1075 	 * In peridoc mode intr redistribution processing is done in
1076 	 * apic_intr_enter during clk intr processing
1077 	 */
1078 	if (!apic_oneshot)
1079 		return;
1080 
1081 	/*
1082 	 * Register a periodical handler for the redistribution processing.
1083 	 * Though we would generally prefer to use the DDI interface for
1084 	 * periodic handler invocation, ddi_periodic_add(9F), we are
1085 	 * unfortunately already holding cpu_lock, which ddi_periodic_add will
1086 	 * attempt to take for us.  Thus, we add our own cyclic directly:
1087 	 */
1088 	cyh.cyh_func = (void (*)(void *))apix_redistribute_compute;
1089 	cyh.cyh_arg = NULL;
1090 	cyh.cyh_level = CY_LOW_LEVEL;
1091 
1092 	cyt.cyt_when = 0;
1093 	cyt.cyt_interval = apic_redistribute_sample_interval;
1094 
1095 	apic_cyclic_id = cyclic_add(&cyh, &cyt);
1096 }
1097 
1098 /*
1099  * Called the first time we enable x2apic mode on this cpu.
1100  * Update some of the function pointers to use x2apic routines.
1101  */
1102 void
1103 x2apic_update_psm()
1104 {
1105 	struct psm_ops *pops = &apix_ops;
1106 
1107 	ASSERT(pops != NULL);
1108 
1109 	/*
1110 	 * The pcplusmp module's version of x2apic_update_psm makes additional
1111 	 * changes that we do not have to make here. It needs to make those
1112 	 * changes because pcplusmp relies on the TPR register and the means of
1113 	 * addressing that changes when using the local apic versus the x2apic.
1114 	 * It's also worth noting that the apix driver specific function end up
1115 	 * being apix_foo as opposed to apic_foo and x2apic_foo.
1116 	 */
1117 	pops->psm_send_ipi = x2apic_send_ipi;
1118 
1119 	send_dirintf = pops->psm_send_ipi;
1120 
1121 	apic_mode = LOCAL_X2APIC;
1122 	apic_change_ops();
1123 }
1124 
1125 /*
1126  * This function provides external interface to the nexus for all
1127  * functionalities related to the new DDI interrupt framework.
1128  *
1129  * Input:
1130  * dip     - pointer to the dev_info structure of the requested device
1131  * hdlp    - pointer to the internal interrupt handle structure for the
1132  *	     requested interrupt
1133  * intr_op - opcode for this call
1134  * result  - pointer to the integer that will hold the result to be
1135  *	     passed back if return value is PSM_SUCCESS
1136  *
1137  * Output:
1138  * return value is either PSM_SUCCESS or PSM_FAILURE
1139  */
1140 static int
1141 apix_intr_ops(dev_info_t *dip, ddi_intr_handle_impl_t *hdlp,
1142     psm_intr_op_t intr_op, int *result)
1143 {
1144 	int		cap;
1145 	apix_vector_t	*vecp, *newvecp;
1146 	struct intrspec *ispec, intr_spec;
1147 	processorid_t target;
1148 
1149 	ispec = &intr_spec;
1150 	ispec->intrspec_pri = hdlp->ih_pri;
1151 	ispec->intrspec_vec = hdlp->ih_inum;
1152 	ispec->intrspec_func = hdlp->ih_cb_func;
1153 
1154 	switch (intr_op) {
1155 	case PSM_INTR_OP_ALLOC_VECTORS:
1156 		switch (hdlp->ih_type) {
1157 		case DDI_INTR_TYPE_MSI:
1158 			/* allocate MSI vectors */
1159 			*result = apix_alloc_msi(dip, hdlp->ih_inum,
1160 			    hdlp->ih_scratch1,
1161 			    (int)(uintptr_t)hdlp->ih_scratch2);
1162 			break;
1163 		case DDI_INTR_TYPE_MSIX:
1164 			/* allocate MSI-X vectors */
1165 			*result = apix_alloc_msix(dip, hdlp->ih_inum,
1166 			    hdlp->ih_scratch1,
1167 			    (int)(uintptr_t)hdlp->ih_scratch2);
1168 			break;
1169 		case DDI_INTR_TYPE_FIXED:
1170 			/* allocate or share vector for fixed */
1171 			if ((ihdl_plat_t *)hdlp->ih_private == NULL) {
1172 				return (PSM_FAILURE);
1173 			}
1174 			ispec = ((ihdl_plat_t *)hdlp->ih_private)->ip_ispecp;
1175 			*result = apix_intx_alloc_vector(dip, hdlp->ih_inum,
1176 			    ispec);
1177 			break;
1178 		default:
1179 			return (PSM_FAILURE);
1180 		}
1181 		break;
1182 	case PSM_INTR_OP_FREE_VECTORS:
1183 		apix_free_vectors(dip, hdlp->ih_inum, hdlp->ih_scratch1,
1184 		    hdlp->ih_type);
1185 		break;
1186 	case PSM_INTR_OP_XLATE_VECTOR:
1187 		/*
1188 		 * Vectors are allocated by ALLOC and freed by FREE.
1189 		 * XLATE finds and returns APIX_VIRTVEC_VECTOR(cpu, vector).
1190 		 */
1191 		*result = APIX_INVALID_VECT;
1192 		vecp = apix_get_dev_map(dip, hdlp->ih_inum, hdlp->ih_type);
1193 		if (vecp != NULL) {
1194 			*result = APIX_VIRTVECTOR(vecp->v_cpuid,
1195 			    vecp->v_vector);
1196 			break;
1197 		}
1198 
1199 		/*
1200 		 * No vector to device mapping exists. If this is FIXED type
1201 		 * then check if this IRQ is already mapped for another device
1202 		 * then return the vector number for it (i.e. shared IRQ case).
1203 		 * Otherwise, return PSM_FAILURE.
1204 		 */
1205 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED) {
1206 			vecp = apix_intx_xlate_vector(dip, hdlp->ih_inum,
1207 			    ispec);
1208 			*result = (vecp == NULL) ? APIX_INVALID_VECT :
1209 			    APIX_VIRTVECTOR(vecp->v_cpuid, vecp->v_vector);
1210 		}
1211 		if (*result == APIX_INVALID_VECT)
1212 			return (PSM_FAILURE);
1213 		break;
1214 	case PSM_INTR_OP_GET_PENDING:
1215 		vecp = apix_get_dev_map(dip, hdlp->ih_inum, hdlp->ih_type);
1216 		if (vecp == NULL)
1217 			return (PSM_FAILURE);
1218 
1219 		*result = apix_get_pending(vecp);
1220 		break;
1221 	case PSM_INTR_OP_CLEAR_MASK:
1222 		if (hdlp->ih_type != DDI_INTR_TYPE_FIXED)
1223 			return (PSM_FAILURE);
1224 
1225 		vecp = apix_get_dev_map(dip, hdlp->ih_inum, hdlp->ih_type);
1226 		if (vecp == NULL)
1227 			return (PSM_FAILURE);
1228 
1229 		apix_intx_clear_mask(vecp->v_inum);
1230 		break;
1231 	case PSM_INTR_OP_SET_MASK:
1232 		if (hdlp->ih_type != DDI_INTR_TYPE_FIXED)
1233 			return (PSM_FAILURE);
1234 
1235 		vecp = apix_get_dev_map(dip, hdlp->ih_inum, hdlp->ih_type);
1236 		if (vecp == NULL)
1237 			return (PSM_FAILURE);
1238 
1239 		apix_intx_set_mask(vecp->v_inum);
1240 		break;
1241 	case PSM_INTR_OP_GET_SHARED:
1242 		if (hdlp->ih_type != DDI_INTR_TYPE_FIXED)
1243 			return (PSM_FAILURE);
1244 
1245 		vecp = apix_get_dev_map(dip, hdlp->ih_inum, hdlp->ih_type);
1246 		if (vecp == NULL)
1247 			return (PSM_FAILURE);
1248 
1249 		*result = apix_intx_get_shared(vecp->v_inum);
1250 		break;
1251 	case PSM_INTR_OP_SET_PRI:
1252 		/*
1253 		 * Called prior to adding the interrupt handler or when
1254 		 * an interrupt handler is unassigned.
1255 		 */
1256 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED)
1257 			return (PSM_SUCCESS);
1258 
1259 		if (apix_get_dev_map(dip, hdlp->ih_inum, hdlp->ih_type) == NULL)
1260 			return (PSM_FAILURE);
1261 
1262 		break;
1263 	case PSM_INTR_OP_SET_CPU:
1264 	case PSM_INTR_OP_GRP_SET_CPU:
1265 		/*
1266 		 * The interrupt handle given here has been allocated
1267 		 * specifically for this command, and ih_private carries
1268 		 * a CPU value.
1269 		 */
1270 		*result = EINVAL;
1271 		target = (int)(intptr_t)hdlp->ih_private;
1272 		if (!apic_cpu_in_range(target)) {
1273 			DDI_INTR_IMPLDBG((CE_WARN,
1274 			    "[grp_]set_cpu: cpu out of range: %d\n", target));
1275 			return (PSM_FAILURE);
1276 		}
1277 
1278 		lock_set(&apix_lock);
1279 
1280 		vecp = apix_get_req_vector(hdlp, hdlp->ih_flags);
1281 		if (!IS_VECT_ENABLED(vecp)) {
1282 			DDI_INTR_IMPLDBG((CE_WARN,
1283 			    "[grp]_set_cpu: invalid vector 0x%x\n",
1284 			    hdlp->ih_vector));
1285 			lock_clear(&apix_lock);
1286 			return (PSM_FAILURE);
1287 		}
1288 
1289 		*result = 0;
1290 
1291 		if (intr_op == PSM_INTR_OP_SET_CPU)
1292 			newvecp = apix_set_cpu(vecp, target, result);
1293 		else
1294 			newvecp = apix_grp_set_cpu(vecp, target, result);
1295 
1296 		lock_clear(&apix_lock);
1297 
1298 		if (newvecp == NULL) {
1299 			*result = EIO;
1300 			return (PSM_FAILURE);
1301 		}
1302 		newvecp->v_bound_cpuid = target;
1303 		hdlp->ih_vector = APIX_VIRTVECTOR(newvecp->v_cpuid,
1304 		    newvecp->v_vector);
1305 		break;
1306 
1307 	case PSM_INTR_OP_GET_INTR:
1308 		/*
1309 		 * The interrupt handle given here has been allocated
1310 		 * specifically for this command, and ih_private carries
1311 		 * a pointer to a apic_get_intr_t.
1312 		 */
1313 		if (apix_get_intr_info(hdlp, hdlp->ih_private) != PSM_SUCCESS)
1314 			return (PSM_FAILURE);
1315 		break;
1316 
1317 	case PSM_INTR_OP_CHECK_MSI:
1318 		/*
1319 		 * Check MSI/X is supported or not at APIC level and
1320 		 * masked off the MSI/X bits in hdlp->ih_type if not
1321 		 * supported before return.  If MSI/X is supported,
1322 		 * leave the ih_type unchanged and return.
1323 		 *
1324 		 * hdlp->ih_type passed in from the nexus has all the
1325 		 * interrupt types supported by the device.
1326 		 */
1327 		if (apic_support_msi == 0) {	/* uninitialized */
1328 			/*
1329 			 * if apic_support_msi is not set, call
1330 			 * apic_check_msi_support() to check whether msi
1331 			 * is supported first
1332 			 */
1333 			if (apic_check_msi_support() == PSM_SUCCESS)
1334 				apic_support_msi = 1;	/* supported */
1335 			else
1336 				apic_support_msi = -1;	/* not-supported */
1337 		}
1338 		if (apic_support_msi == 1) {
1339 			if (apic_msix_enable)
1340 				*result = hdlp->ih_type;
1341 			else
1342 				*result = hdlp->ih_type & ~DDI_INTR_TYPE_MSIX;
1343 		} else
1344 			*result = hdlp->ih_type & ~(DDI_INTR_TYPE_MSI |
1345 			    DDI_INTR_TYPE_MSIX);
1346 		break;
1347 	case PSM_INTR_OP_GET_CAP:
1348 		cap = DDI_INTR_FLAG_PENDING;
1349 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED)
1350 			cap |= DDI_INTR_FLAG_MASKABLE;
1351 		*result = cap;
1352 		break;
1353 	case PSM_INTR_OP_APIC_TYPE:
1354 		((apic_get_type_t *)(hdlp->ih_private))->avgi_type =
1355 		    apix_get_apic_type();
1356 		((apic_get_type_t *)(hdlp->ih_private))->avgi_num_intr =
1357 		    APIX_IPI_MIN;
1358 		((apic_get_type_t *)(hdlp->ih_private))->avgi_num_cpu =
1359 		    apic_nproc;
1360 		hdlp->ih_ver = apic_get_apic_version();
1361 		break;
1362 	case PSM_INTR_OP_SET_CAP:
1363 	default:
1364 		return (PSM_FAILURE);
1365 	}
1366 
1367 	return (PSM_SUCCESS);
1368 }
1369 
1370 static void
1371 apix_cleanup_busy(void)
1372 {
1373 	int i, j;
1374 	apix_vector_t *vecp;
1375 
1376 	for (i = 0; i < apic_nproc; i++) {
1377 		if (!apic_cpu_in_range(i))
1378 			continue;
1379 		apic_cpus[i].aci_busy = 0;
1380 		for (j = APIX_AVINTR_MIN; j < APIX_AVINTR_MAX; j++) {
1381 			if ((vecp = xv_vector(i, j)) != NULL)
1382 				vecp->v_busy = 0;
1383 		}
1384 	}
1385 }
1386 
1387 static void
1388 apix_redistribute_compute(void)
1389 {
1390 	int	i, j, max_busy;
1391 
1392 	if (!apic_enable_dynamic_migration)
1393 		return;
1394 
1395 	if (++apic_nticks == apic_sample_factor_redistribution) {
1396 		/*
1397 		 * Time to call apic_intr_redistribute().
1398 		 * reset apic_nticks. This will cause max_busy
1399 		 * to be calculated below and if it is more than
1400 		 * apic_int_busy, we will do the whole thing
1401 		 */
1402 		apic_nticks = 0;
1403 	}
1404 	max_busy = 0;
1405 	for (i = 0; i < apic_nproc; i++) {
1406 		if (!apic_cpu_in_range(i))
1407 			continue;
1408 		/*
1409 		 * Check if curipl is non zero & if ISR is in
1410 		 * progress
1411 		 */
1412 		if (((j = apic_cpus[i].aci_curipl) != 0) &&
1413 		    (apic_cpus[i].aci_ISR_in_progress & (1 << j))) {
1414 
1415 			int	vect;
1416 			apic_cpus[i].aci_busy++;
1417 			vect = apic_cpus[i].aci_current[j];
1418 			apixs[i]->x_vectbl[vect]->v_busy++;
1419 		}
1420 
1421 		if (!apic_nticks &&
1422 		    (apic_cpus[i].aci_busy > max_busy))
1423 			max_busy = apic_cpus[i].aci_busy;
1424 	}
1425 	if (!apic_nticks) {
1426 		if (max_busy > apic_int_busy_mark) {
1427 		/*
1428 		 * We could make the following check be
1429 		 * skipped > 1 in which case, we get a
1430 		 * redistribution at half the busy mark (due to
1431 		 * double interval). Need to be able to collect
1432 		 * more empirical data to decide if that is a
1433 		 * good strategy. Punt for now.
1434 		 */
1435 			apix_cleanup_busy();
1436 			apic_skipped_redistribute = 0;
1437 		} else
1438 			apic_skipped_redistribute++;
1439 	}
1440 }
1441 
1442 /*
1443  * intr_ops() service routines
1444  */
1445 
1446 static int
1447 apix_get_pending(apix_vector_t *vecp)
1448 {
1449 	int bit, index, irr, pending;
1450 
1451 	/* need to get on the bound cpu */
1452 	mutex_enter(&cpu_lock);
1453 	affinity_set(vecp->v_cpuid);
1454 
1455 	index = vecp->v_vector / 32;
1456 	bit = vecp->v_vector % 32;
1457 	irr = apic_reg_ops->apic_read(APIC_IRR_REG + index);
1458 
1459 	affinity_clear();
1460 	mutex_exit(&cpu_lock);
1461 
1462 	pending = (irr & (1 << bit)) ? 1 : 0;
1463 	if (!pending && vecp->v_type == APIX_TYPE_FIXED)
1464 		pending = apix_intx_get_pending(vecp->v_inum);
1465 
1466 	return (pending);
1467 }
1468 
1469 static apix_vector_t *
1470 apix_get_req_vector(ddi_intr_handle_impl_t *hdlp, ushort_t flags)
1471 {
1472 	apix_vector_t *vecp;
1473 	processorid_t cpuid;
1474 	int32_t virt_vec = 0;
1475 
1476 	switch (flags & PSMGI_INTRBY_FLAGS) {
1477 	case PSMGI_INTRBY_IRQ:
1478 		return (apix_intx_get_vector(hdlp->ih_vector));
1479 	case PSMGI_INTRBY_VEC:
1480 		virt_vec = (virt_vec == 0) ? hdlp->ih_vector : virt_vec;
1481 
1482 		cpuid = APIX_VIRTVEC_CPU(virt_vec);
1483 		if (!apic_cpu_in_range(cpuid))
1484 			return (NULL);
1485 
1486 		vecp = xv_vector(cpuid, APIX_VIRTVEC_VECTOR(virt_vec));
1487 		break;
1488 	case PSMGI_INTRBY_DEFAULT:
1489 		vecp = apix_get_dev_map(hdlp->ih_dip, hdlp->ih_inum,
1490 		    hdlp->ih_type);
1491 		break;
1492 	default:
1493 		return (NULL);
1494 	}
1495 
1496 	return (vecp);
1497 }
1498 
1499 static int
1500 apix_get_intr_info(ddi_intr_handle_impl_t *hdlp,
1501     apic_get_intr_t *intr_params_p)
1502 {
1503 	apix_vector_t *vecp;
1504 	struct autovec *av_dev;
1505 	int i;
1506 
1507 	vecp = apix_get_req_vector(hdlp, intr_params_p->avgi_req_flags);
1508 	if (IS_VECT_FREE(vecp)) {
1509 		intr_params_p->avgi_num_devs = 0;
1510 		intr_params_p->avgi_cpu_id = 0;
1511 		intr_params_p->avgi_req_flags = 0;
1512 		return (PSM_SUCCESS);
1513 	}
1514 
1515 	if (intr_params_p->avgi_req_flags & PSMGI_REQ_CPUID) {
1516 		intr_params_p->avgi_cpu_id = vecp->v_cpuid;
1517 
1518 		/* Return user bound info for intrd. */
1519 		if (intr_params_p->avgi_cpu_id & IRQ_USER_BOUND) {
1520 			intr_params_p->avgi_cpu_id &= ~IRQ_USER_BOUND;
1521 			intr_params_p->avgi_cpu_id |= PSMGI_CPU_USER_BOUND;
1522 		}
1523 	}
1524 
1525 	if (intr_params_p->avgi_req_flags & PSMGI_REQ_VECTOR)
1526 		intr_params_p->avgi_vector = vecp->v_vector;
1527 
1528 	if (intr_params_p->avgi_req_flags &
1529 	    (PSMGI_REQ_NUM_DEVS | PSMGI_REQ_GET_DEVS))
1530 		/* Get number of devices from apic_irq table shared field. */
1531 		intr_params_p->avgi_num_devs = vecp->v_share;
1532 
1533 	if (intr_params_p->avgi_req_flags &  PSMGI_REQ_GET_DEVS) {
1534 
1535 		intr_params_p->avgi_req_flags  |= PSMGI_REQ_NUM_DEVS;
1536 
1537 		/* Some devices have NULL dip.  Don't count these. */
1538 		if (intr_params_p->avgi_num_devs > 0) {
1539 			for (i = 0, av_dev = vecp->v_autovect; av_dev;
1540 			    av_dev = av_dev->av_link) {
1541 				if (av_dev->av_vector && av_dev->av_dip)
1542 					i++;
1543 			}
1544 			intr_params_p->avgi_num_devs =
1545 			    (uint8_t)MIN(intr_params_p->avgi_num_devs, i);
1546 		}
1547 
1548 		/* There are no viable dips to return. */
1549 		if (intr_params_p->avgi_num_devs == 0) {
1550 			intr_params_p->avgi_dip_list = NULL;
1551 
1552 		} else {	/* Return list of dips */
1553 
1554 			/* Allocate space in array for that number of devs. */
1555 			intr_params_p->avgi_dip_list = kmem_zalloc(
1556 			    intr_params_p->avgi_num_devs *
1557 			    sizeof (dev_info_t *),
1558 			    KM_NOSLEEP);
1559 			if (intr_params_p->avgi_dip_list == NULL) {
1560 				DDI_INTR_IMPLDBG((CE_WARN,
1561 				    "apix_get_vector_intr_info: no memory"));
1562 				return (PSM_FAILURE);
1563 			}
1564 
1565 			/*
1566 			 * Loop through the device list of the autovec table
1567 			 * filling in the dip array.
1568 			 *
1569 			 * Note that the autovect table may have some special
1570 			 * entries which contain NULL dips.  These will be
1571 			 * ignored.
1572 			 */
1573 			for (i = 0, av_dev = vecp->v_autovect; av_dev;
1574 			    av_dev = av_dev->av_link) {
1575 				if (av_dev->av_vector && av_dev->av_dip)
1576 					intr_params_p->avgi_dip_list[i++] =
1577 					    av_dev->av_dip;
1578 			}
1579 		}
1580 	}
1581 
1582 	return (PSM_SUCCESS);
1583 }
1584 
1585 static char *
1586 apix_get_apic_type(void)
1587 {
1588 	return (apix_psm_info.p_mach_idstring);
1589 }
1590 
1591 apix_vector_t *
1592 apix_set_cpu(apix_vector_t *vecp, int new_cpu, int *result)
1593 {
1594 	apix_vector_t *newp = NULL;
1595 	dev_info_t *dip;
1596 	int inum, cap_ptr;
1597 	ddi_acc_handle_t handle;
1598 	ddi_intr_msix_t *msix_p = NULL;
1599 	ushort_t msix_ctrl;
1600 	uintptr_t off;
1601 	uint32_t mask;
1602 
1603 	ASSERT(LOCK_HELD(&apix_lock));
1604 	*result = ENXIO;
1605 
1606 	/* Fail if this is an MSI intr and is part of a group. */
1607 	if (vecp->v_type == APIX_TYPE_MSI) {
1608 		if (i_ddi_intr_get_current_nintrs(APIX_GET_DIP(vecp)) > 1)
1609 			return (NULL);
1610 		else
1611 			return (apix_grp_set_cpu(vecp, new_cpu, result));
1612 	}
1613 
1614 	/*
1615 	 * Mask MSI-X. It's unmasked when MSI-X gets enabled.
1616 	 */
1617 	if (vecp->v_type == APIX_TYPE_MSIX && IS_VECT_ENABLED(vecp)) {
1618 		if ((dip = APIX_GET_DIP(vecp)) == NULL)
1619 			return (NULL);
1620 		inum = vecp->v_devp->dv_inum;
1621 
1622 		handle = i_ddi_get_pci_config_handle(dip);
1623 		cap_ptr = i_ddi_get_msi_msix_cap_ptr(dip);
1624 		msix_ctrl = pci_config_get16(handle, cap_ptr + PCI_MSIX_CTRL);
1625 		if ((msix_ctrl & PCI_MSIX_FUNCTION_MASK) == 0) {
1626 			/*
1627 			 * Function is not masked, then mask "inum"th
1628 			 * entry in the MSI-X table
1629 			 */
1630 			msix_p = i_ddi_get_msix(dip);
1631 			off = (uintptr_t)msix_p->msix_tbl_addr + (inum *
1632 			    PCI_MSIX_VECTOR_SIZE) + PCI_MSIX_VECTOR_CTRL_OFFSET;
1633 			mask = ddi_get32(msix_p->msix_tbl_hdl, (uint32_t *)off);
1634 			ddi_put32(msix_p->msix_tbl_hdl, (uint32_t *)off,
1635 			    mask | 1);
1636 		}
1637 	}
1638 
1639 	*result = 0;
1640 	if ((newp = apix_rebind(vecp, new_cpu, 1)) == NULL)
1641 		*result = EIO;
1642 
1643 	/* Restore mask bit */
1644 	if (msix_p != NULL)
1645 		ddi_put32(msix_p->msix_tbl_hdl, (uint32_t *)off, mask);
1646 
1647 	return (newp);
1648 }
1649 
1650 /*
1651  * Set cpu for MSIs
1652  */
1653 apix_vector_t *
1654 apix_grp_set_cpu(apix_vector_t *vecp, int new_cpu, int *result)
1655 {
1656 	apix_vector_t *newp, *vp;
1657 	uint32_t orig_cpu = vecp->v_cpuid;
1658 	int orig_vect = vecp->v_vector;
1659 	int i, num_vectors, cap_ptr, msi_mask_off;
1660 	uint32_t msi_pvm;
1661 	ushort_t msi_ctrl;
1662 	ddi_acc_handle_t handle;
1663 	dev_info_t *dip;
1664 
1665 	APIC_VERBOSE(INTR, (CE_CONT, "apix_grp_set_cpu: oldcpu: %x, vector: %x,"
1666 	    " newcpu:%x\n", vecp->v_cpuid, vecp->v_vector, new_cpu));
1667 
1668 	ASSERT(LOCK_HELD(&apix_lock));
1669 
1670 	*result = ENXIO;
1671 
1672 	if (vecp->v_type != APIX_TYPE_MSI) {
1673 		DDI_INTR_IMPLDBG((CE_WARN, "set_grp: intr not MSI\n"));
1674 		return (NULL);
1675 	}
1676 
1677 	if ((dip = APIX_GET_DIP(vecp)) == NULL)
1678 		return (NULL);
1679 
1680 	num_vectors = i_ddi_intr_get_current_nintrs(dip);
1681 	if ((num_vectors < 1) || ((num_vectors - 1) & orig_vect)) {
1682 		APIC_VERBOSE(INTR, (CE_WARN,
1683 		    "set_grp: base vec not part of a grp or not aligned: "
1684 		    "vec:0x%x, num_vec:0x%x\n", orig_vect, num_vectors));
1685 		return (NULL);
1686 	}
1687 
1688 	if (vecp->v_inum != apix_get_min_dev_inum(dip, vecp->v_type))
1689 		return (NULL);
1690 
1691 	*result = EIO;
1692 	for (i = 1; i < num_vectors; i++) {
1693 		if ((vp = xv_vector(orig_cpu, orig_vect + i)) == NULL)
1694 			return (NULL);
1695 #ifdef DEBUG
1696 		/*
1697 		 * Sanity check: CPU and dip is the same for all entries.
1698 		 * May be called when first msi to be enabled, at this time
1699 		 * add_avintr() is not called for other msi
1700 		 */
1701 		if ((vp->v_share != 0) &&
1702 		    ((APIX_GET_DIP(vp) != dip) ||
1703 		    (vp->v_cpuid != vecp->v_cpuid))) {
1704 			APIC_VERBOSE(INTR, (CE_WARN,
1705 			    "set_grp: cpu or dip for vec 0x%x difft than for "
1706 			    "vec 0x%x\n", orig_vect, orig_vect + i));
1707 			APIC_VERBOSE(INTR, (CE_WARN,
1708 			    "  cpu: %d vs %d, dip: 0x%p vs 0x%p\n", orig_cpu,
1709 			    vp->v_cpuid, (void *)dip,
1710 			    (void *)APIX_GET_DIP(vp)));
1711 			return (NULL);
1712 		}
1713 #endif /* DEBUG */
1714 	}
1715 
1716 	cap_ptr = i_ddi_get_msi_msix_cap_ptr(dip);
1717 	handle = i_ddi_get_pci_config_handle(dip);
1718 	msi_ctrl = pci_config_get16(handle, cap_ptr + PCI_MSI_CTRL);
1719 
1720 	/* MSI Per vector masking is supported. */
1721 	if (msi_ctrl & PCI_MSI_PVM_MASK) {
1722 		if (msi_ctrl &  PCI_MSI_64BIT_MASK)
1723 			msi_mask_off = cap_ptr + PCI_MSI_64BIT_MASKBITS;
1724 		else
1725 			msi_mask_off = cap_ptr + PCI_MSI_32BIT_MASK;
1726 		msi_pvm = pci_config_get32(handle, msi_mask_off);
1727 		pci_config_put32(handle, msi_mask_off, (uint32_t)-1);
1728 		APIC_VERBOSE(INTR, (CE_CONT,
1729 		    "set_grp: pvm supported.  Mask set to 0x%x\n",
1730 		    pci_config_get32(handle, msi_mask_off)));
1731 	}
1732 
1733 	if ((newp = apix_rebind(vecp, new_cpu, num_vectors)) != NULL)
1734 		*result = 0;
1735 
1736 	/* Reenable vectors if per vector masking is supported. */
1737 	if (msi_ctrl & PCI_MSI_PVM_MASK) {
1738 		pci_config_put32(handle, msi_mask_off, msi_pvm);
1739 		APIC_VERBOSE(INTR, (CE_CONT,
1740 		    "set_grp: pvm supported.  Mask restored to 0x%x\n",
1741 		    pci_config_get32(handle, msi_mask_off)));
1742 	}
1743 
1744 	return (newp);
1745 }
1746 
1747 void
1748 apix_intx_set_vector(int irqno, uint32_t cpuid, uchar_t vector)
1749 {
1750 	apic_irq_t *irqp;
1751 
1752 	mutex_enter(&airq_mutex);
1753 	irqp = apic_irq_table[irqno];
1754 	irqp->airq_cpu = cpuid;
1755 	irqp->airq_vector = vector;
1756 	apic_record_rdt_entry(irqp, irqno);
1757 	mutex_exit(&airq_mutex);
1758 }
1759 
1760 apix_vector_t *
1761 apix_intx_get_vector(int irqno)
1762 {
1763 	apic_irq_t *irqp;
1764 	uint32_t cpuid;
1765 	uchar_t vector;
1766 
1767 	mutex_enter(&airq_mutex);
1768 	irqp = apic_irq_table[irqno & 0xff];
1769 	if (IS_IRQ_FREE(irqp) || (irqp->airq_cpu == IRQ_UNINIT)) {
1770 		mutex_exit(&airq_mutex);
1771 		return (NULL);
1772 	}
1773 	cpuid = irqp->airq_cpu;
1774 	vector = irqp->airq_vector;
1775 	mutex_exit(&airq_mutex);
1776 
1777 	return (xv_vector(cpuid, vector));
1778 }
1779 
1780 /*
1781  * Must called with interrupts disabled and apic_ioapic_lock held
1782  */
1783 void
1784 apix_intx_enable(int irqno)
1785 {
1786 	uchar_t ioapicindex, intin;
1787 	apic_irq_t *irqp = apic_irq_table[irqno];
1788 	ioapic_rdt_t irdt;
1789 	apic_cpus_info_t *cpu_infop;
1790 	apix_vector_t *vecp = xv_vector(irqp->airq_cpu, irqp->airq_vector);
1791 
1792 	ASSERT(LOCK_HELD(&apic_ioapic_lock) && !IS_IRQ_FREE(irqp));
1793 
1794 	ioapicindex = irqp->airq_ioapicindex;
1795 	intin = irqp->airq_intin_no;
1796 	cpu_infop =  &apic_cpus[irqp->airq_cpu];
1797 
1798 	irdt.ir_lo = AV_PDEST | AV_FIXED | irqp->airq_rdt_entry;
1799 	irdt.ir_hi = cpu_infop->aci_local_id;
1800 
1801 	apic_vt_ops->apic_intrmap_alloc_entry(&vecp->v_intrmap_private, NULL,
1802 	    vecp->v_type, 1, ioapicindex);
1803 	apic_vt_ops->apic_intrmap_map_entry(vecp->v_intrmap_private,
1804 	    (void *)&irdt, vecp->v_type, 1);
1805 	apic_vt_ops->apic_intrmap_record_rdt(vecp->v_intrmap_private, &irdt);
1806 
1807 	/* write RDT entry high dword - destination */
1808 	WRITE_IOAPIC_RDT_ENTRY_HIGH_DWORD(ioapicindex, intin,
1809 	    irdt.ir_hi);
1810 
1811 	/* Write the vector, trigger, and polarity portion of the RDT */
1812 	WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapicindex, intin, irdt.ir_lo);
1813 
1814 	vecp->v_state = APIX_STATE_ENABLED;
1815 
1816 	APIC_VERBOSE_IOAPIC((CE_CONT, "apix_intx_enable: ioapic 0x%x"
1817 	    " intin 0x%x rdt_low 0x%x rdt_high 0x%x\n",
1818 	    ioapicindex, intin, irdt.ir_lo, irdt.ir_hi));
1819 }
1820 
1821 /*
1822  * Must called with interrupts disabled and apic_ioapic_lock held
1823  */
1824 void
1825 apix_intx_disable(int irqno)
1826 {
1827 	apic_irq_t *irqp = apic_irq_table[irqno];
1828 	int ioapicindex, intin;
1829 
1830 	ASSERT(LOCK_HELD(&apic_ioapic_lock) && !IS_IRQ_FREE(irqp));
1831 	/*
1832 	 * The assumption here is that this is safe, even for
1833 	 * systems with IOAPICs that suffer from the hardware
1834 	 * erratum because all devices have been quiesced before
1835 	 * they unregister their interrupt handlers.  If that
1836 	 * assumption turns out to be false, this mask operation
1837 	 * can induce the same erratum result we're trying to
1838 	 * avoid.
1839 	 */
1840 	ioapicindex = irqp->airq_ioapicindex;
1841 	intin = irqp->airq_intin_no;
1842 	ioapic_write(ioapicindex, APIC_RDT_CMD + 2 * intin, AV_MASK);
1843 
1844 	APIC_VERBOSE_IOAPIC((CE_CONT, "apix_intx_disable: ioapic 0x%x"
1845 	    " intin 0x%x\n", ioapicindex, intin));
1846 }
1847 
1848 void
1849 apix_intx_free(int irqno)
1850 {
1851 	apic_irq_t *irqp;
1852 
1853 	mutex_enter(&airq_mutex);
1854 	irqp = apic_irq_table[irqno];
1855 
1856 	if (IS_IRQ_FREE(irqp)) {
1857 		mutex_exit(&airq_mutex);
1858 		return;
1859 	}
1860 
1861 	irqp->airq_mps_intr_index = FREE_INDEX;
1862 	irqp->airq_cpu = IRQ_UNINIT;
1863 	irqp->airq_vector = APIX_INVALID_VECT;
1864 	mutex_exit(&airq_mutex);
1865 }
1866 
1867 #ifdef DEBUG
1868 int apix_intr_deliver_timeouts = 0;
1869 int apix_intr_rirr_timeouts = 0;
1870 int apix_intr_rirr_reset_failure = 0;
1871 #endif
1872 int apix_max_reps_irr_pending = 10;
1873 
1874 #define	GET_RDT_BITS(ioapic, intin, bits)	\
1875 	(READ_IOAPIC_RDT_ENTRY_LOW_DWORD((ioapic), (intin)) & (bits))
1876 #define	APIX_CHECK_IRR_DELAY	drv_usectohz(5000)
1877 
1878 int
1879 apix_intx_rebind(int irqno, processorid_t cpuid, uchar_t vector)
1880 {
1881 	apic_irq_t *irqp = apic_irq_table[irqno];
1882 	ulong_t iflag;
1883 	int waited, ioapic_ix, intin_no, level, repeats, rdt_entry, masked;
1884 
1885 	ASSERT(irqp != NULL);
1886 
1887 	iflag = intr_clear();
1888 	lock_set(&apic_ioapic_lock);
1889 
1890 	ioapic_ix = irqp->airq_ioapicindex;
1891 	intin_no = irqp->airq_intin_no;
1892 	level = apic_level_intr[irqno];
1893 
1894 	/*
1895 	 * Wait for the delivery status bit to be cleared. This should
1896 	 * be a very small amount of time.
1897 	 */
1898 	repeats = 0;
1899 	do {
1900 		repeats++;
1901 
1902 		for (waited = 0; waited < apic_max_reps_clear_pending;
1903 		    waited++) {
1904 			if (GET_RDT_BITS(ioapic_ix, intin_no, AV_PENDING) == 0)
1905 				break;
1906 		}
1907 		if (!level)
1908 			break;
1909 
1910 		/*
1911 		 * Mask the RDT entry for level-triggered interrupts.
1912 		 */
1913 		irqp->airq_rdt_entry |= AV_MASK;
1914 		rdt_entry = READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
1915 		    intin_no);
1916 		if ((masked = (rdt_entry & AV_MASK)) == 0) {
1917 			/* Mask it */
1918 			WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix, intin_no,
1919 			    AV_MASK | rdt_entry);
1920 		}
1921 
1922 		/*
1923 		 * If there was a race and an interrupt was injected
1924 		 * just before we masked, check for that case here.
1925 		 * Then, unmask the RDT entry and try again.  If we're
1926 		 * on our last try, don't unmask (because we want the
1927 		 * RDT entry to remain masked for the rest of the
1928 		 * function).
1929 		 */
1930 		rdt_entry = READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
1931 		    intin_no);
1932 		if ((masked == 0) && ((rdt_entry & AV_PENDING) != 0) &&
1933 		    (repeats < apic_max_reps_clear_pending)) {
1934 			/* Unmask it */
1935 			WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
1936 			    intin_no, rdt_entry & ~AV_MASK);
1937 			irqp->airq_rdt_entry &= ~AV_MASK;
1938 		}
1939 	} while ((rdt_entry & AV_PENDING) &&
1940 	    (repeats < apic_max_reps_clear_pending));
1941 
1942 #ifdef DEBUG
1943 	if (GET_RDT_BITS(ioapic_ix, intin_no, AV_PENDING) != 0)
1944 		apix_intr_deliver_timeouts++;
1945 #endif
1946 
1947 	if (!level || !APIX_IS_MASK_RDT(apix_mul_ioapic_method))
1948 		goto done;
1949 
1950 	/*
1951 	 * wait for remote IRR to be cleared for level-triggered
1952 	 * interrupts
1953 	 */
1954 	repeats = 0;
1955 	do {
1956 		repeats++;
1957 
1958 		for (waited = 0; waited < apic_max_reps_clear_pending;
1959 		    waited++) {
1960 			if (GET_RDT_BITS(ioapic_ix, intin_no, AV_REMOTE_IRR)
1961 			    == 0)
1962 				break;
1963 		}
1964 
1965 		if (GET_RDT_BITS(ioapic_ix, intin_no, AV_REMOTE_IRR) != 0) {
1966 			lock_clear(&apic_ioapic_lock);
1967 			intr_restore(iflag);
1968 
1969 			delay(APIX_CHECK_IRR_DELAY);
1970 
1971 			iflag = intr_clear();
1972 			lock_set(&apic_ioapic_lock);
1973 		}
1974 	} while (repeats < apix_max_reps_irr_pending);
1975 
1976 	if (repeats >= apix_max_reps_irr_pending) {
1977 #ifdef DEBUG
1978 		apix_intr_rirr_timeouts++;
1979 #endif
1980 
1981 		/*
1982 		 * If we waited and the Remote IRR bit is still not cleared,
1983 		 * AND if we've invoked the timeout APIC_REPROGRAM_MAX_TIMEOUTS
1984 		 * times for this interrupt, try the last-ditch workaround:
1985 		 */
1986 		if (GET_RDT_BITS(ioapic_ix, intin_no, AV_REMOTE_IRR) != 0) {
1987 			/*
1988 			 * Trying to clear the bit through normal
1989 			 * channels has failed.  So as a last-ditch
1990 			 * effort, try to set the trigger mode to
1991 			 * edge, then to level.  This has been
1992 			 * observed to work on many systems.
1993 			 */
1994 			WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
1995 			    intin_no,
1996 			    READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
1997 			    intin_no) & ~AV_LEVEL);
1998 			WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
1999 			    intin_no,
2000 			    READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
2001 			    intin_no) | AV_LEVEL);
2002 		}
2003 
2004 		if (GET_RDT_BITS(ioapic_ix, intin_no, AV_REMOTE_IRR) != 0) {
2005 #ifdef DEBUG
2006 			apix_intr_rirr_reset_failure++;
2007 #endif
2008 			lock_clear(&apic_ioapic_lock);
2009 			intr_restore(iflag);
2010 			prom_printf("apix: Remote IRR still "
2011 			    "not clear for IOAPIC %d intin %d.\n"
2012 			    "\tInterrupts to this pin may cease "
2013 			    "functioning.\n", ioapic_ix, intin_no);
2014 			return (1);	/* return failure */
2015 		}
2016 	}
2017 
2018 done:
2019 	/* change apic_irq_table */
2020 	lock_clear(&apic_ioapic_lock);
2021 	intr_restore(iflag);
2022 	apix_intx_set_vector(irqno, cpuid, vector);
2023 	iflag = intr_clear();
2024 	lock_set(&apic_ioapic_lock);
2025 
2026 	/* reprogramme IO-APIC RDT entry */
2027 	apix_intx_enable(irqno);
2028 
2029 	lock_clear(&apic_ioapic_lock);
2030 	intr_restore(iflag);
2031 
2032 	return (0);
2033 }
2034 
2035 static int
2036 apix_intx_get_pending(int irqno)
2037 {
2038 	apic_irq_t *irqp;
2039 	int intin, ioapicindex, pending;
2040 	ulong_t iflag;
2041 
2042 	mutex_enter(&airq_mutex);
2043 	irqp = apic_irq_table[irqno];
2044 	if (IS_IRQ_FREE(irqp)) {
2045 		mutex_exit(&airq_mutex);
2046 		return (0);
2047 	}
2048 
2049 	/* check IO-APIC delivery status */
2050 	intin = irqp->airq_intin_no;
2051 	ioapicindex = irqp->airq_ioapicindex;
2052 	mutex_exit(&airq_mutex);
2053 
2054 	iflag = intr_clear();
2055 	lock_set(&apic_ioapic_lock);
2056 
2057 	pending = (READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapicindex, intin) &
2058 	    AV_PENDING) ? 1 : 0;
2059 
2060 	lock_clear(&apic_ioapic_lock);
2061 	intr_restore(iflag);
2062 
2063 	return (pending);
2064 }
2065 
2066 /*
2067  * This function will mask the interrupt on the I/O APIC
2068  */
2069 static void
2070 apix_intx_set_mask(int irqno)
2071 {
2072 	int intin, ioapixindex, rdt_entry;
2073 	ulong_t iflag;
2074 	apic_irq_t *irqp;
2075 
2076 	mutex_enter(&airq_mutex);
2077 	irqp = apic_irq_table[irqno];
2078 
2079 	ASSERT(irqp->airq_mps_intr_index != FREE_INDEX);
2080 
2081 	intin = irqp->airq_intin_no;
2082 	ioapixindex = irqp->airq_ioapicindex;
2083 	mutex_exit(&airq_mutex);
2084 
2085 	iflag = intr_clear();
2086 	lock_set(&apic_ioapic_lock);
2087 
2088 	rdt_entry = READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapixindex, intin);
2089 
2090 	/* clear mask */
2091 	WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapixindex, intin,
2092 	    (AV_MASK | rdt_entry));
2093 
2094 	lock_clear(&apic_ioapic_lock);
2095 	intr_restore(iflag);
2096 }
2097 
2098 /*
2099  * This function will clear the mask for the interrupt on the I/O APIC
2100  */
2101 static void
2102 apix_intx_clear_mask(int irqno)
2103 {
2104 	int intin, ioapixindex, rdt_entry;
2105 	ulong_t iflag;
2106 	apic_irq_t *irqp;
2107 
2108 	mutex_enter(&airq_mutex);
2109 	irqp = apic_irq_table[irqno];
2110 
2111 	ASSERT(irqp->airq_mps_intr_index != FREE_INDEX);
2112 
2113 	intin = irqp->airq_intin_no;
2114 	ioapixindex = irqp->airq_ioapicindex;
2115 	mutex_exit(&airq_mutex);
2116 
2117 	iflag = intr_clear();
2118 	lock_set(&apic_ioapic_lock);
2119 
2120 	rdt_entry = READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapixindex, intin);
2121 
2122 	/* clear mask */
2123 	WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapixindex, intin,
2124 	    ((~AV_MASK) & rdt_entry));
2125 
2126 	lock_clear(&apic_ioapic_lock);
2127 	intr_restore(iflag);
2128 }
2129 
2130 /*
2131  * For level-triggered interrupt, mask the IRQ line. Mask means
2132  * new interrupts will not be delivered. The interrupt already
2133  * accepted by a local APIC is not affected
2134  */
2135 void
2136 apix_level_intr_pre_eoi(int irq)
2137 {
2138 	apic_irq_t *irqp = apic_irq_table[irq];
2139 	int apic_ix, intin_ix;
2140 
2141 	if (irqp == NULL)
2142 		return;
2143 
2144 	ASSERT(apic_level_intr[irq] == TRIGGER_MODE_LEVEL);
2145 
2146 	lock_set(&apic_ioapic_lock);
2147 
2148 	intin_ix = irqp->airq_intin_no;
2149 	apic_ix = irqp->airq_ioapicindex;
2150 
2151 	if (irqp->airq_cpu != CPU->cpu_id) {
2152 		if (!APIX_IS_MASK_RDT(apix_mul_ioapic_method))
2153 			ioapic_write_eoi(apic_ix, irqp->airq_vector);
2154 		lock_clear(&apic_ioapic_lock);
2155 		return;
2156 	}
2157 
2158 	if (apix_mul_ioapic_method == APIC_MUL_IOAPIC_IOXAPIC) {
2159 		/*
2160 		 * This is a IOxAPIC and there is EOI register:
2161 		 * 	Change the vector to reserved unused vector, so that
2162 		 * 	the EOI	from Local APIC won't clear the Remote IRR for
2163 		 * 	this level trigger interrupt. Instead, we'll manually
2164 		 * 	clear it in apix_post_hardint() after ISR handling.
2165 		 */
2166 		WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(apic_ix, intin_ix,
2167 		    (irqp->airq_rdt_entry & (~0xff)) | APIX_RESV_VECTOR);
2168 	} else {
2169 		WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(apic_ix, intin_ix,
2170 		    AV_MASK | irqp->airq_rdt_entry);
2171 	}
2172 
2173 	lock_clear(&apic_ioapic_lock);
2174 }
2175 
2176 /*
2177  * For level-triggered interrupt, unmask the IRQ line
2178  * or restore the original vector number.
2179  */
2180 void
2181 apix_level_intr_post_dispatch(int irq)
2182 {
2183 	apic_irq_t *irqp = apic_irq_table[irq];
2184 	int apic_ix, intin_ix;
2185 
2186 	if (irqp == NULL)
2187 		return;
2188 
2189 	lock_set(&apic_ioapic_lock);
2190 
2191 	intin_ix = irqp->airq_intin_no;
2192 	apic_ix = irqp->airq_ioapicindex;
2193 
2194 	if (APIX_IS_DIRECTED_EOI(apix_mul_ioapic_method)) {
2195 		/*
2196 		 * Already sent EOI back to Local APIC.
2197 		 * Send EOI to IO-APIC
2198 		 */
2199 		ioapic_write_eoi(apic_ix, irqp->airq_vector);
2200 	} else {
2201 		/* clear the mask or restore the vector */
2202 		WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(apic_ix, intin_ix,
2203 		    irqp->airq_rdt_entry);
2204 
2205 		/* send EOI to IOxAPIC */
2206 		if (apix_mul_ioapic_method == APIC_MUL_IOAPIC_IOXAPIC)
2207 			ioapic_write_eoi(apic_ix, irqp->airq_vector);
2208 	}
2209 
2210 	lock_clear(&apic_ioapic_lock);
2211 }
2212 
2213 static int
2214 apix_intx_get_shared(int irqno)
2215 {
2216 	apic_irq_t *irqp;
2217 	int share;
2218 
2219 	mutex_enter(&airq_mutex);
2220 	irqp = apic_irq_table[irqno];
2221 	if (IS_IRQ_FREE(irqp) || (irqp->airq_cpu == IRQ_UNINIT)) {
2222 		mutex_exit(&airq_mutex);
2223 		return (0);
2224 	}
2225 	share = irqp->airq_share;
2226 	mutex_exit(&airq_mutex);
2227 
2228 	return (share);
2229 }
2230 
2231 static void
2232 apix_intx_set_shared(int irqno, int delta)
2233 {
2234 	apic_irq_t *irqp;
2235 
2236 	mutex_enter(&airq_mutex);
2237 	irqp = apic_irq_table[irqno];
2238 	if (IS_IRQ_FREE(irqp)) {
2239 		mutex_exit(&airq_mutex);
2240 		return;
2241 	}
2242 	irqp->airq_share += delta;
2243 	mutex_exit(&airq_mutex);
2244 }
2245 
2246 /*
2247  * Setup IRQ table. Return IRQ no or -1 on failure
2248  */
2249 static int
2250 apix_intx_setup(dev_info_t *dip, int inum, int irqno,
2251     struct apic_io_intr *intrp, struct intrspec *ispec, iflag_t *iflagp)
2252 {
2253 	int origirq = ispec->intrspec_vec;
2254 	int newirq;
2255 	short intr_index;
2256 	uchar_t ipin, ioapic, ioapicindex;
2257 	apic_irq_t *irqp;
2258 
2259 	UNREFERENCED_1PARAMETER(inum);
2260 
2261 	if (intrp != NULL) {
2262 		intr_index = (short)(intrp - apic_io_intrp);
2263 		ioapic = intrp->intr_destid;
2264 		ipin = intrp->intr_destintin;
2265 
2266 		/* Find ioapicindex. If destid was ALL, we will exit with 0. */
2267 		for (ioapicindex = apic_io_max - 1; ioapicindex; ioapicindex--)
2268 			if (apic_io_id[ioapicindex] == ioapic)
2269 				break;
2270 		ASSERT((ioapic == apic_io_id[ioapicindex]) ||
2271 		    (ioapic == INTR_ALL_APIC));
2272 
2273 		/* check whether this intin# has been used by another irqno */
2274 		if ((newirq = apic_find_intin(ioapicindex, ipin)) != -1)
2275 			return (newirq);
2276 
2277 	} else if (iflagp != NULL) {	/* ACPI */
2278 		intr_index = ACPI_INDEX;
2279 		ioapicindex = acpi_find_ioapic(irqno);
2280 		ASSERT(ioapicindex != 0xFF);
2281 		ioapic = apic_io_id[ioapicindex];
2282 		ipin = irqno - apic_io_vectbase[ioapicindex];
2283 
2284 		if (apic_irq_table[irqno] &&
2285 		    apic_irq_table[irqno]->airq_mps_intr_index == ACPI_INDEX) {
2286 			ASSERT(apic_irq_table[irqno]->airq_intin_no == ipin &&
2287 			    apic_irq_table[irqno]->airq_ioapicindex ==
2288 			    ioapicindex);
2289 			return (irqno);
2290 		}
2291 
2292 	} else {	/* default configuration */
2293 		intr_index = DEFAULT_INDEX;
2294 		ioapicindex = 0;
2295 		ioapic = apic_io_id[ioapicindex];
2296 		ipin = (uchar_t)irqno;
2297 	}
2298 
2299 	/* allocate a new IRQ no */
2300 	if ((irqp = apic_irq_table[irqno]) == NULL) {
2301 		irqp = kmem_zalloc(sizeof (apic_irq_t), KM_SLEEP);
2302 		apic_irq_table[irqno] = irqp;
2303 	} else {
2304 		if (irqp->airq_mps_intr_index != FREE_INDEX) {
2305 			newirq = apic_allocate_irq(apic_first_avail_irq);
2306 			if (newirq == -1) {
2307 				return (-1);
2308 			}
2309 			irqno = newirq;
2310 			irqp = apic_irq_table[irqno];
2311 			ASSERT(irqp != NULL);
2312 		}
2313 	}
2314 	apic_max_device_irq = max(irqno, apic_max_device_irq);
2315 	apic_min_device_irq = min(irqno, apic_min_device_irq);
2316 
2317 	irqp->airq_mps_intr_index = intr_index;
2318 	irqp->airq_ioapicindex = ioapicindex;
2319 	irqp->airq_intin_no = ipin;
2320 	irqp->airq_dip = dip;
2321 	irqp->airq_origirq = (uchar_t)origirq;
2322 	if (iflagp != NULL)
2323 		irqp->airq_iflag = *iflagp;
2324 	irqp->airq_cpu = IRQ_UNINIT;
2325 	irqp->airq_vector = 0;
2326 
2327 	return (irqno);
2328 }
2329 
2330 /*
2331  * Setup IRQ table for non-pci devices. Return IRQ no or -1 on error
2332  */
2333 static int
2334 apix_intx_setup_nonpci(dev_info_t *dip, int inum, int bustype,
2335     struct intrspec *ispec)
2336 {
2337 	int irqno = ispec->intrspec_vec;
2338 	int newirq, i;
2339 	iflag_t intr_flag;
2340 	ACPI_SUBTABLE_HEADER	*hp;
2341 	ACPI_MADT_INTERRUPT_OVERRIDE *isop;
2342 	struct apic_io_intr *intrp;
2343 
2344 	if (!apic_enable_acpi || apic_use_acpi_madt_only) {
2345 		int busid;
2346 
2347 		if (bustype == 0)
2348 			bustype = eisa_level_intr_mask ? BUS_EISA : BUS_ISA;
2349 
2350 		/* loop checking BUS_ISA/BUS_EISA */
2351 		for (i = 0; i < 2; i++) {
2352 			if (((busid = apic_find_bus_id(bustype)) != -1) &&
2353 			    ((intrp = apic_find_io_intr_w_busid(irqno, busid))
2354 			    != NULL)) {
2355 				return (apix_intx_setup(dip, inum, irqno,
2356 				    intrp, ispec, NULL));
2357 			}
2358 			bustype = (bustype == BUS_EISA) ? BUS_ISA : BUS_EISA;
2359 		}
2360 
2361 		/* fall back to default configuration */
2362 		return (-1);
2363 	}
2364 
2365 	/* search iso entries first */
2366 	if (acpi_iso_cnt != 0) {
2367 		hp = (ACPI_SUBTABLE_HEADER *)acpi_isop;
2368 		i = 0;
2369 		while (i < acpi_iso_cnt) {
2370 			if (hp->Type == ACPI_MADT_TYPE_INTERRUPT_OVERRIDE) {
2371 				isop = (ACPI_MADT_INTERRUPT_OVERRIDE *) hp;
2372 				if (isop->Bus == 0 &&
2373 				    isop->SourceIrq == irqno) {
2374 					newirq = isop->GlobalIrq;
2375 					intr_flag.intr_po = isop->IntiFlags &
2376 					    ACPI_MADT_POLARITY_MASK;
2377 					intr_flag.intr_el = (isop->IntiFlags &
2378 					    ACPI_MADT_TRIGGER_MASK) >> 2;
2379 					intr_flag.bustype = BUS_ISA;
2380 
2381 					return (apix_intx_setup(dip, inum,
2382 					    newirq, NULL, ispec, &intr_flag));
2383 				}
2384 				i++;
2385 			}
2386 			hp = (ACPI_SUBTABLE_HEADER *)(((char *)hp) +
2387 			    hp->Length);
2388 		}
2389 	}
2390 	intr_flag.intr_po = INTR_PO_ACTIVE_HIGH;
2391 	intr_flag.intr_el = INTR_EL_EDGE;
2392 	intr_flag.bustype = BUS_ISA;
2393 	return (apix_intx_setup(dip, inum, irqno, NULL, ispec, &intr_flag));
2394 }
2395 
2396 
2397 /*
2398  * Setup IRQ table for pci devices. Return IRQ no or -1 on error
2399  */
2400 static int
2401 apix_intx_setup_pci(dev_info_t *dip, int inum, int bustype,
2402     struct intrspec *ispec)
2403 {
2404 	int busid, devid, pci_irq;
2405 	ddi_acc_handle_t cfg_handle;
2406 	uchar_t ipin;
2407 	iflag_t intr_flag;
2408 	struct apic_io_intr *intrp;
2409 
2410 	if (acpica_get_bdf(dip, &busid, &devid, NULL) != 0)
2411 		return (-1);
2412 
2413 	if (busid == 0 && apic_pci_bus_total == 1)
2414 		busid = (int)apic_single_pci_busid;
2415 
2416 	if (pci_config_setup(dip, &cfg_handle) != DDI_SUCCESS)
2417 		return (-1);
2418 	ipin = pci_config_get8(cfg_handle, PCI_CONF_IPIN) - PCI_INTA;
2419 	pci_config_teardown(&cfg_handle);
2420 
2421 	if (apic_enable_acpi && !apic_use_acpi_madt_only) {	/* ACPI */
2422 		if (apic_acpi_translate_pci_irq(dip, busid, devid,
2423 		    ipin, &pci_irq, &intr_flag) != ACPI_PSM_SUCCESS)
2424 			return (-1);
2425 
2426 		intr_flag.bustype = (uchar_t)bustype;
2427 		return (apix_intx_setup(dip, inum, pci_irq, NULL, ispec,
2428 		    &intr_flag));
2429 	}
2430 
2431 	/* MP configuration table */
2432 	pci_irq = ((devid & 0x1f) << 2) | (ipin & 0x3);
2433 	if ((intrp = apic_find_io_intr_w_busid(pci_irq, busid)) == NULL) {
2434 		pci_irq = apic_handle_pci_pci_bridge(dip, devid, ipin, &intrp);
2435 		if (pci_irq == -1)
2436 			return (-1);
2437 	}
2438 
2439 	return (apix_intx_setup(dip, inum, pci_irq, intrp, ispec, NULL));
2440 }
2441 
2442 /*
2443  * Translate and return IRQ no
2444  */
2445 static int
2446 apix_intx_xlate_irq(dev_info_t *dip, int inum, struct intrspec *ispec)
2447 {
2448 	int newirq, irqno = ispec->intrspec_vec;
2449 	int parent_is_pci_or_pciex = 0, child_is_pciex = 0;
2450 	int bustype = 0, dev_len;
2451 	char dev_type[16];
2452 
2453 	if (apic_defconf) {
2454 		mutex_enter(&airq_mutex);
2455 		goto defconf;
2456 	}
2457 
2458 	if ((dip == NULL) || (!apic_irq_translate && !apic_enable_acpi)) {
2459 		mutex_enter(&airq_mutex);
2460 		goto nonpci;
2461 	}
2462 
2463 	/*
2464 	 * use ddi_getlongprop_buf() instead of ddi_prop_lookup_string()
2465 	 * to avoid extra buffer allocation.
2466 	 */
2467 	dev_len = sizeof (dev_type);
2468 	if (ddi_getlongprop_buf(DDI_DEV_T_ANY, ddi_get_parent(dip),
2469 	    DDI_PROP_DONTPASS, "device_type", (caddr_t)dev_type,
2470 	    &dev_len) == DDI_PROP_SUCCESS) {
2471 		if ((strcmp(dev_type, "pci") == 0) ||
2472 		    (strcmp(dev_type, "pciex") == 0))
2473 			parent_is_pci_or_pciex = 1;
2474 	}
2475 
2476 	if (ddi_getlongprop_buf(DDI_DEV_T_ANY, dip,
2477 	    DDI_PROP_DONTPASS, "compatible", (caddr_t)dev_type,
2478 	    &dev_len) == DDI_PROP_SUCCESS) {
2479 		if (strstr(dev_type, "pciex"))
2480 			child_is_pciex = 1;
2481 	}
2482 
2483 	mutex_enter(&airq_mutex);
2484 
2485 	if (parent_is_pci_or_pciex) {
2486 		bustype = child_is_pciex ? BUS_PCIE : BUS_PCI;
2487 		newirq = apix_intx_setup_pci(dip, inum, bustype, ispec);
2488 		if (newirq != -1)
2489 			goto done;
2490 		bustype = 0;
2491 	} else if (strcmp(dev_type, "isa") == 0)
2492 		bustype = BUS_ISA;
2493 	else if (strcmp(dev_type, "eisa") == 0)
2494 		bustype = BUS_EISA;
2495 
2496 nonpci:
2497 	newirq = apix_intx_setup_nonpci(dip, inum, bustype, ispec);
2498 	if (newirq != -1)
2499 		goto done;
2500 
2501 defconf:
2502 	newirq = apix_intx_setup(dip, inum, irqno, NULL, ispec, NULL);
2503 	if (newirq == -1) {
2504 		mutex_exit(&airq_mutex);
2505 		return (-1);
2506 	}
2507 done:
2508 	ASSERT(apic_irq_table[newirq]);
2509 	mutex_exit(&airq_mutex);
2510 	return (newirq);
2511 }
2512 
2513 static int
2514 apix_intx_alloc_vector(dev_info_t *dip, int inum, struct intrspec *ispec)
2515 {
2516 	int irqno;
2517 	apix_vector_t *vecp;
2518 
2519 	if ((irqno = apix_intx_xlate_irq(dip, inum, ispec)) == -1)
2520 		return (0);
2521 
2522 	if ((vecp = apix_alloc_intx(dip, inum, irqno)) == NULL)
2523 		return (0);
2524 
2525 	DDI_INTR_IMPLDBG((CE_CONT, "apix_intx_alloc_vector: dip=0x%p name=%s "
2526 	    "irqno=0x%x cpuid=%d vector=0x%x\n",
2527 	    (void *)dip, ddi_driver_name(dip), irqno,
2528 	    vecp->v_cpuid, vecp->v_vector));
2529 
2530 	return (1);
2531 }
2532 
2533 /*
2534  * Return the vector number if the translated IRQ for this device
2535  * has a vector mapping setup. If no IRQ setup exists or no vector is
2536  * allocated to it then return 0.
2537  */
2538 static apix_vector_t *
2539 apix_intx_xlate_vector(dev_info_t *dip, int inum, struct intrspec *ispec)
2540 {
2541 	int irqno;
2542 	apix_vector_t *vecp;
2543 
2544 	/* get the IRQ number */
2545 	if ((irqno = apix_intx_xlate_irq(dip, inum, ispec)) == -1)
2546 		return (NULL);
2547 
2548 	/* get the vector number if a vector is allocated to this irqno */
2549 	vecp = apix_intx_get_vector(irqno);
2550 
2551 	return (vecp);
2552 }
2553 
2554 /* stub function */
2555 int
2556 apix_loaded(void)
2557 {
2558 	return (apix_is_enabled);
2559 }
2560