xref: /illumos-gate/usr/src/uts/i86pc/io/apix/apix.c (revision be5b6b318451fd76f08b0716a69a138a9a97e87d)
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) 2013, 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 			    apic_detect_x2apic())) {
268 				/* x2APIC mode is not supported in the hw */
269 				apix_enable = 0;
270 			}
271 		}
272 		if (apix_enable == 0)
273 			return (PSM_FAILURE);
274 	}
275 
276 	rval = apic_probe_common(apix_psm_info.p_mach_idstring);
277 	if (rval == PSM_SUCCESS)
278 		apix_is_enabled = 1;
279 	else
280 		apix_is_enabled = 0;
281 	return (rval);
282 }
283 
284 /*
285  * Initialize the data structures needed by pcplusmpx module.
286  * Specifically, the data structures used by addspl() and delspl()
287  * routines.
288  */
289 static void
290 apix_softinit()
291 {
292 	int i, *iptr;
293 	apix_impl_t *hdlp;
294 	int nproc;
295 
296 	nproc = max(apic_nproc, apic_max_nproc);
297 
298 	hdlp = kmem_zalloc(nproc * sizeof (apix_impl_t), KM_SLEEP);
299 	for (i = 0; i < nproc; i++) {
300 		apixs[i] = &hdlp[i];
301 		apixs[i]->x_cpuid = i;
302 		LOCK_INIT_CLEAR(&apixs[i]->x_lock);
303 	}
304 
305 	/* cpu 0 is always up (for now) */
306 	apic_cpus[0].aci_status = APIC_CPU_ONLINE | APIC_CPU_INTR_ENABLE;
307 
308 	iptr = (int *)&apic_irq_table[0];
309 	for (i = 0; i <= APIC_MAX_VECTOR; i++) {
310 		apic_level_intr[i] = 0;
311 		*iptr++ = NULL;
312 	}
313 	mutex_init(&airq_mutex, NULL, MUTEX_DEFAULT, NULL);
314 
315 	apix_dev_vector = kmem_zalloc(sizeof (apix_dev_vector_t *) * devcnt,
316 	    KM_SLEEP);
317 
318 	if (apic_intr_policy == INTR_ROUND_ROBIN_WITH_AFFINITY) {
319 		apix_major_to_cpu = kmem_zalloc(sizeof (int) * devcnt,
320 		    KM_SLEEP);
321 		for (i = 0; i < devcnt; i++)
322 			apix_major_to_cpu[i] = IRQ_UNINIT;
323 	}
324 
325 	mutex_init(&apix_mutex, NULL, MUTEX_DEFAULT, NULL);
326 }
327 
328 static int
329 apix_get_pending_spl(void)
330 {
331 	int cpuid = CPU->cpu_id;
332 
333 	return (bsrw_insn(apixs[cpuid]->x_intr_pending));
334 }
335 
336 static uintptr_t
337 apix_get_intr_handler(int cpu, short vec)
338 {
339 	apix_vector_t *apix_vector;
340 
341 	ASSERT(cpu < apic_nproc && vec < APIX_NVECTOR);
342 	if (cpu >= apic_nproc)
343 		return (NULL);
344 
345 	apix_vector = apixs[cpu]->x_vectbl[vec];
346 
347 	return ((uintptr_t)(apix_vector->v_autovect));
348 }
349 
350 #if defined(__amd64)
351 static unsigned char dummy_cpu_pri[MAXIPL + 1] = {
352 	0, 0, 0, 0, 0, 0, 0, 0,
353 	0, 0, 0, 0, 0, 0, 0, 0, 0
354 };
355 #endif
356 
357 static void
358 apix_init()
359 {
360 	extern void (*do_interrupt_common)(struct regs *, trap_trace_rec_t *);
361 
362 	APIC_VERBOSE(INIT, (CE_CONT, "apix: psm_softinit\n"));
363 
364 	do_interrupt_common = apix_do_interrupt;
365 	addintr = apix_add_avintr;
366 	remintr = apix_rem_avintr;
367 	get_pending_spl = apix_get_pending_spl;
368 	get_intr_handler = apix_get_intr_handler;
369 	psm_get_localapicid = apic_get_localapicid;
370 	psm_get_ioapicid = apic_get_ioapicid;
371 
372 	apix_softinit();
373 #if defined(__amd64)
374 	CPU->cpu_pri_data = dummy_cpu_pri;
375 #else
376 	if (cpuid_have_cr8access(CPU))
377 		apic_have_32bit_cr8 = 1;
378 #endif	/* __amd64 */
379 
380 	/*
381 	 * Initialize IRM pool parameters
382 	 */
383 	if (irm_enable) {
384 		int	i;
385 		int	lowest_irq;
386 		int	highest_irq;
387 
388 		/* number of CPUs present */
389 		apix_irminfo.apix_ncpus = apic_nproc;
390 		/* total number of entries in all of the IOAPICs present */
391 		lowest_irq = apic_io_vectbase[0];
392 		highest_irq = apic_io_vectend[0];
393 		for (i = 1; i < apic_io_max; i++) {
394 			if (apic_io_vectbase[i] < lowest_irq)
395 				lowest_irq = apic_io_vectbase[i];
396 			if (apic_io_vectend[i] > highest_irq)
397 				highest_irq = apic_io_vectend[i];
398 		}
399 		apix_irminfo.apix_ioapic_max_vectors =
400 		    highest_irq - lowest_irq + 1;
401 		/*
402 		 * Number of available per-CPU vectors excluding
403 		 * reserved vectors for Dtrace, int80, system-call,
404 		 * fast-trap, etc.
405 		 */
406 		apix_irminfo.apix_per_cpu_vectors = APIX_NAVINTR -
407 		    APIX_SW_RESERVED_VECTORS;
408 
409 		/* Number of vectors (pre) allocated (SCI and HPET) */
410 		apix_irminfo.apix_vectors_allocated = 0;
411 		if (apic_hpet_vect != -1)
412 			apix_irminfo.apix_vectors_allocated++;
413 		if (apic_sci_vect != -1)
414 			apix_irminfo.apix_vectors_allocated++;
415 	}
416 }
417 
418 static void
419 apix_init_intr()
420 {
421 	processorid_t	cpun = psm_get_cpu_id();
422 	uint_t nlvt;
423 	uint32_t svr = AV_UNIT_ENABLE | APIC_SPUR_INTR;
424 	extern void cmi_cmci_trap(void);
425 
426 	apic_reg_ops->apic_write_task_reg(APIC_MASK_ALL);
427 
428 	if (apic_mode == LOCAL_APIC) {
429 		/*
430 		 * We are running APIC in MMIO mode.
431 		 */
432 		if (apic_flat_model) {
433 			apic_reg_ops->apic_write(APIC_FORMAT_REG,
434 			    APIC_FLAT_MODEL);
435 		} else {
436 			apic_reg_ops->apic_write(APIC_FORMAT_REG,
437 			    APIC_CLUSTER_MODEL);
438 		}
439 
440 		apic_reg_ops->apic_write(APIC_DEST_REG,
441 		    AV_HIGH_ORDER >> cpun);
442 	}
443 
444 	if (apic_directed_EOI_supported()) {
445 		/*
446 		 * Setting the 12th bit in the Spurious Interrupt Vector
447 		 * Register suppresses broadcast EOIs generated by the local
448 		 * APIC. The suppression of broadcast EOIs happens only when
449 		 * interrupts are level-triggered.
450 		 */
451 		svr |= APIC_SVR_SUPPRESS_BROADCAST_EOI;
452 	}
453 
454 	/* need to enable APIC before unmasking NMI */
455 	apic_reg_ops->apic_write(APIC_SPUR_INT_REG, svr);
456 
457 	/*
458 	 * Presence of an invalid vector with delivery mode AV_FIXED can
459 	 * cause an error interrupt, even if the entry is masked...so
460 	 * write a valid vector to LVT entries along with the mask bit
461 	 */
462 
463 	/* All APICs have timer and LINT0/1 */
464 	apic_reg_ops->apic_write(APIC_LOCAL_TIMER, AV_MASK|APIC_RESV_IRQ);
465 	apic_reg_ops->apic_write(APIC_INT_VECT0, AV_MASK|APIC_RESV_IRQ);
466 	apic_reg_ops->apic_write(APIC_INT_VECT1, AV_NMI);	/* enable NMI */
467 
468 	/*
469 	 * On integrated APICs, the number of LVT entries is
470 	 * 'Max LVT entry' + 1; on 82489DX's (non-integrated
471 	 * APICs), nlvt is "3" (LINT0, LINT1, and timer)
472 	 */
473 
474 	if (apic_cpus[cpun].aci_local_ver < APIC_INTEGRATED_VERS) {
475 		nlvt = 3;
476 	} else {
477 		nlvt = ((apic_reg_ops->apic_read(APIC_VERS_REG) >> 16) &
478 		    0xFF) + 1;
479 	}
480 
481 	if (nlvt >= 5) {
482 		/* Enable performance counter overflow interrupt */
483 
484 		if (!is_x86_feature(x86_featureset, X86FSET_MSR))
485 			apic_enable_cpcovf_intr = 0;
486 		if (apic_enable_cpcovf_intr) {
487 			if (apic_cpcovf_vect == 0) {
488 				int ipl = APIC_PCINT_IPL;
489 
490 				apic_cpcovf_vect = apix_get_ipivect(ipl, -1);
491 				ASSERT(apic_cpcovf_vect);
492 
493 				(void) add_avintr(NULL, ipl,
494 				    (avfunc)kcpc_hw_overflow_intr,
495 				    "apic pcint", apic_cpcovf_vect,
496 				    NULL, NULL, NULL, NULL);
497 				kcpc_hw_overflow_intr_installed = 1;
498 				kcpc_hw_enable_cpc_intr =
499 				    apic_cpcovf_mask_clear;
500 			}
501 			apic_reg_ops->apic_write(APIC_PCINT_VECT,
502 			    apic_cpcovf_vect);
503 		}
504 	}
505 
506 	if (nlvt >= 6) {
507 		/* Only mask TM intr if the BIOS apparently doesn't use it */
508 
509 		uint32_t lvtval;
510 
511 		lvtval = apic_reg_ops->apic_read(APIC_THERM_VECT);
512 		if (((lvtval & AV_MASK) == AV_MASK) ||
513 		    ((lvtval & AV_DELIV_MODE) != AV_SMI)) {
514 			apic_reg_ops->apic_write(APIC_THERM_VECT,
515 			    AV_MASK|APIC_RESV_IRQ);
516 		}
517 	}
518 
519 	/* Enable error interrupt */
520 
521 	if (nlvt >= 4 && apic_enable_error_intr) {
522 		if (apic_errvect == 0) {
523 			int ipl = 0xf;	/* get highest priority intr */
524 			apic_errvect = apix_get_ipivect(ipl, -1);
525 			ASSERT(apic_errvect);
526 			/*
527 			 * Not PSMI compliant, but we are going to merge
528 			 * with ON anyway
529 			 */
530 			(void) add_avintr(NULL, ipl,
531 			    (avfunc)apic_error_intr, "apic error intr",
532 			    apic_errvect, NULL, NULL, NULL, NULL);
533 		}
534 		apic_reg_ops->apic_write(APIC_ERR_VECT, apic_errvect);
535 		apic_reg_ops->apic_write(APIC_ERROR_STATUS, 0);
536 		apic_reg_ops->apic_write(APIC_ERROR_STATUS, 0);
537 	}
538 
539 	/* Enable CMCI interrupt */
540 	if (cmi_enable_cmci) {
541 		mutex_enter(&cmci_cpu_setup_lock);
542 		if (cmci_cpu_setup_registered == 0) {
543 			mutex_enter(&cpu_lock);
544 			register_cpu_setup_func(cmci_cpu_setup, NULL);
545 			mutex_exit(&cpu_lock);
546 			cmci_cpu_setup_registered = 1;
547 		}
548 		mutex_exit(&cmci_cpu_setup_lock);
549 
550 		if (apic_cmci_vect == 0) {
551 			int ipl = 0x2;
552 			apic_cmci_vect = apix_get_ipivect(ipl, -1);
553 			ASSERT(apic_cmci_vect);
554 
555 			(void) add_avintr(NULL, ipl,
556 			    (avfunc)cmi_cmci_trap, "apic cmci intr",
557 			    apic_cmci_vect, NULL, NULL, NULL, NULL);
558 		}
559 		apic_reg_ops->apic_write(APIC_CMCI_VECT, apic_cmci_vect);
560 	}
561 
562 	apic_reg_ops->apic_write_task_reg(0);
563 }
564 
565 static void
566 apix_picinit(void)
567 {
568 	int i, j;
569 	uint_t isr;
570 
571 	APIC_VERBOSE(INIT, (CE_CONT, "apix: psm_picinit\n"));
572 
573 	/*
574 	 * initialize interrupt remapping before apic
575 	 * hardware initialization
576 	 */
577 	apic_intrmap_init(apic_mode);
578 	if (apic_vt_ops == psm_vt_ops)
579 		apix_mul_ioapic_method = APIC_MUL_IOAPIC_IIR;
580 
581 	/*
582 	 * On UniSys Model 6520, the BIOS leaves vector 0x20 isr
583 	 * bit on without clearing it with EOI.  Since softint
584 	 * uses vector 0x20 to interrupt itself, so softint will
585 	 * not work on this machine.  In order to fix this problem
586 	 * a check is made to verify all the isr bits are clear.
587 	 * If not, EOIs are issued to clear the bits.
588 	 */
589 	for (i = 7; i >= 1; i--) {
590 		isr = apic_reg_ops->apic_read(APIC_ISR_REG + (i * 4));
591 		if (isr != 0)
592 			for (j = 0; ((j < 32) && (isr != 0)); j++)
593 				if (isr & (1 << j)) {
594 					apic_reg_ops->apic_write(
595 					    APIC_EOI_REG, 0);
596 					isr &= ~(1 << j);
597 					apic_error |= APIC_ERR_BOOT_EOI;
598 				}
599 	}
600 
601 	/* set a flag so we know we have run apic_picinit() */
602 	apic_picinit_called = 1;
603 	LOCK_INIT_CLEAR(&apic_gethrtime_lock);
604 	LOCK_INIT_CLEAR(&apic_ioapic_lock);
605 	LOCK_INIT_CLEAR(&apic_error_lock);
606 	LOCK_INIT_CLEAR(&apic_mode_switch_lock);
607 
608 	picsetup();	 /* initialise the 8259 */
609 
610 	/* add nmi handler - least priority nmi handler */
611 	LOCK_INIT_CLEAR(&apic_nmi_lock);
612 
613 	if (!psm_add_nmintr(0, (avfunc) apic_nmi_intr,
614 	    "apix NMI handler", (caddr_t)NULL))
615 		cmn_err(CE_WARN, "apix: Unable to add nmi handler");
616 
617 	apix_init_intr();
618 
619 	/* enable apic mode if imcr present */
620 	if (apic_imcrp) {
621 		outb(APIC_IMCR_P1, (uchar_t)APIC_IMCR_SELECT);
622 		outb(APIC_IMCR_P2, (uchar_t)APIC_IMCR_APIC);
623 	}
624 
625 	ioapix_init_intr(IOAPIC_MASK);
626 
627 	/* setup global IRM pool if applicable */
628 	if (irm_enable)
629 		apix_irm_init();
630 }
631 
632 static __inline__ void
633 apix_send_eoi(void)
634 {
635 	if (apic_mode == LOCAL_APIC)
636 		LOCAL_APIC_WRITE_REG(APIC_EOI_REG, 0);
637 	else
638 		X2APIC_WRITE(APIC_EOI_REG, 0);
639 }
640 
641 /*
642  * platform_intr_enter
643  *
644  *	Called at the beginning of the interrupt service routine, but unlike
645  *	pcplusmp, does not mask interrupts. An EOI is given to the interrupt
646  *	controller to enable other HW interrupts but interrupts are still
647  * 	masked by the IF flag.
648  *
649  *	Return -1 for spurious interrupts
650  *
651  */
652 static int
653 apix_intr_enter(int ipl, int *vectorp)
654 {
655 	struct cpu *cpu = CPU;
656 	uint32_t cpuid = CPU->cpu_id;
657 	apic_cpus_info_t *cpu_infop;
658 	uchar_t vector;
659 	apix_vector_t *vecp;
660 	int nipl = -1;
661 
662 	/*
663 	 * The real vector delivered is (*vectorp + 0x20), but our caller
664 	 * subtracts 0x20 from the vector before passing it to us.
665 	 * (That's why APIC_BASE_VECT is 0x20.)
666 	 */
667 	vector = *vectorp = (uchar_t)*vectorp + APIC_BASE_VECT;
668 
669 	cpu_infop = &apic_cpus[cpuid];
670 	if (vector == APIC_SPUR_INTR) {
671 		cpu_infop->aci_spur_cnt++;
672 		return (APIC_INT_SPURIOUS);
673 	}
674 
675 	vecp = xv_vector(cpuid, vector);
676 	if (vecp == NULL) {
677 		if (APIX_IS_FAKE_INTR(vector))
678 			nipl = apix_rebindinfo.i_pri;
679 		apix_send_eoi();
680 		return (nipl);
681 	}
682 	nipl = vecp->v_pri;
683 
684 	/* if interrupted by the clock, increment apic_nsec_since_boot */
685 	if (vector == (apic_clkvect + APIC_BASE_VECT)) {
686 		if (!apic_oneshot) {
687 			/* NOTE: this is not MT aware */
688 			apic_hrtime_stamp++;
689 			apic_nsec_since_boot += apic_nsec_per_intr;
690 			apic_hrtime_stamp++;
691 			last_count_read = apic_hertz_count;
692 			apix_redistribute_compute();
693 		}
694 
695 		apix_send_eoi();
696 
697 		return (nipl);
698 	}
699 
700 	ASSERT(vecp->v_state != APIX_STATE_OBSOLETED);
701 
702 	/* pre-EOI handling for level-triggered interrupts */
703 	if (!APIX_IS_DIRECTED_EOI(apix_mul_ioapic_method) &&
704 	    (vecp->v_type & APIX_TYPE_FIXED) && apic_level_intr[vecp->v_inum])
705 		apix_level_intr_pre_eoi(vecp->v_inum);
706 
707 	/* send back EOI */
708 	apix_send_eoi();
709 
710 	cpu_infop->aci_current[nipl] = vector;
711 	if ((nipl > ipl) && (nipl > cpu->cpu_base_spl)) {
712 		cpu_infop->aci_curipl = (uchar_t)nipl;
713 		cpu_infop->aci_ISR_in_progress |= 1 << nipl;
714 	}
715 
716 #ifdef	DEBUG
717 	if (vector >= APIX_IPI_MIN)
718 		return (nipl);	/* skip IPI */
719 
720 	APIC_DEBUG_BUF_PUT(vector);
721 	APIC_DEBUG_BUF_PUT(vecp->v_inum);
722 	APIC_DEBUG_BUF_PUT(nipl);
723 	APIC_DEBUG_BUF_PUT(psm_get_cpu_id());
724 	if ((apic_stretch_interrupts) && (apic_stretch_ISR & (1 << nipl)))
725 		drv_usecwait(apic_stretch_interrupts);
726 #endif /* DEBUG */
727 
728 	return (nipl);
729 }
730 
731 /*
732  * Any changes made to this function must also change X2APIC
733  * version of intr_exit.
734  */
735 static void
736 apix_intr_exit(int prev_ipl, int arg2)
737 {
738 	int cpuid = psm_get_cpu_id();
739 	apic_cpus_info_t *cpu_infop = &apic_cpus[cpuid];
740 	apix_impl_t *apixp = apixs[cpuid];
741 
742 	UNREFERENCED_1PARAMETER(arg2);
743 
744 	cpu_infop->aci_curipl = (uchar_t)prev_ipl;
745 	/* ISR above current pri could not be in progress */
746 	cpu_infop->aci_ISR_in_progress &= (2 << prev_ipl) - 1;
747 
748 	if (apixp->x_obsoletes != NULL) {
749 		if (APIX_CPU_LOCK_HELD(cpuid))
750 			return;
751 
752 		APIX_ENTER_CPU_LOCK(cpuid);
753 		(void) apix_obsolete_vector(apixp->x_obsoletes);
754 		APIX_LEAVE_CPU_LOCK(cpuid);
755 	}
756 }
757 
758 /*
759  * The pcplusmp setspl code uses the TPR to mask all interrupts at or below the
760  * given ipl, but apix never uses the TPR and we never mask a subset of the
761  * interrupts. They are either all blocked by the IF flag or all can come in.
762  *
763  * For setspl, we mask all interrupts for XC_HI_PIL (15), otherwise, interrupts
764  * can come in if currently enabled by the IF flag. This table shows the state
765  * of the IF flag when we leave this function.
766  *
767  *    curr IF |	ipl == 15	ipl != 15
768  *    --------+---------------------------
769  *       0    |    0		    0
770  *       1    |    0		    1
771  */
772 static void
773 apix_setspl(int ipl)
774 {
775 	/*
776 	 * Interrupts at ipl above this cannot be in progress, so the following
777 	 * mask is ok.
778 	 */
779 	apic_cpus[psm_get_cpu_id()].aci_ISR_in_progress &= (2 << ipl) - 1;
780 
781 	if (ipl == XC_HI_PIL)
782 		cli();
783 }
784 
785 int
786 apix_addspl(int virtvec, int ipl, int min_ipl, int max_ipl)
787 {
788 	uint32_t cpuid = APIX_VIRTVEC_CPU(virtvec);
789 	uchar_t vector = (uchar_t)APIX_VIRTVEC_VECTOR(virtvec);
790 	apix_vector_t *vecp = xv_vector(cpuid, vector);
791 
792 	UNREFERENCED_3PARAMETER(ipl, min_ipl, max_ipl);
793 	ASSERT(vecp != NULL && LOCK_HELD(&apix_lock));
794 
795 	if (vecp->v_type == APIX_TYPE_FIXED)
796 		apix_intx_set_shared(vecp->v_inum, 1);
797 
798 	/* There are more interrupts, so it's already been enabled */
799 	if (vecp->v_share > 1)
800 		return (PSM_SUCCESS);
801 
802 	/* return if it is not hardware interrupt */
803 	if (vecp->v_type == APIX_TYPE_IPI)
804 		return (PSM_SUCCESS);
805 
806 	/*
807 	 * if apix_picinit() has not been called yet, just return.
808 	 * At the end of apic_picinit(), we will call setup_io_intr().
809 	 */
810 	if (!apic_picinit_called)
811 		return (PSM_SUCCESS);
812 
813 	(void) apix_setup_io_intr(vecp);
814 
815 	return (PSM_SUCCESS);
816 }
817 
818 int
819 apix_delspl(int virtvec, int ipl, int min_ipl, int max_ipl)
820 {
821 	uint32_t cpuid = APIX_VIRTVEC_CPU(virtvec);
822 	uchar_t vector = (uchar_t)APIX_VIRTVEC_VECTOR(virtvec);
823 	apix_vector_t *vecp = xv_vector(cpuid, vector);
824 
825 	UNREFERENCED_3PARAMETER(ipl, min_ipl, max_ipl);
826 	ASSERT(vecp != NULL && LOCK_HELD(&apix_lock));
827 
828 	if (vecp->v_type == APIX_TYPE_FIXED)
829 		apix_intx_set_shared(vecp->v_inum, -1);
830 
831 	/* There are more interrupts */
832 	if (vecp->v_share > 1)
833 		return (PSM_SUCCESS);
834 
835 	/* return if it is not hardware interrupt */
836 	if (vecp->v_type == APIX_TYPE_IPI)
837 		return (PSM_SUCCESS);
838 
839 	if (!apic_picinit_called) {
840 		cmn_err(CE_WARN, "apix: delete 0x%x before apic init",
841 		    virtvec);
842 		return (PSM_SUCCESS);
843 	}
844 
845 	apix_disable_vector(vecp);
846 
847 	return (PSM_SUCCESS);
848 }
849 
850 /*
851  * Try and disable all interrupts. We just assign interrupts to other
852  * processors based on policy. If any were bound by user request, we
853  * let them continue and return failure. We do not bother to check
854  * for cache affinity while rebinding.
855  */
856 static int
857 apix_disable_intr(processorid_t cpun)
858 {
859 	apix_impl_t *apixp = apixs[cpun];
860 	apix_vector_t *vecp, *newp;
861 	int bindcpu, i, hardbound = 0, errbound = 0, ret, loop, type;
862 
863 	lock_set(&apix_lock);
864 
865 	apic_cpus[cpun].aci_status &= ~APIC_CPU_INTR_ENABLE;
866 	apic_cpus[cpun].aci_curipl = 0;
867 
868 	/* if this is for SUSPEND operation, skip rebinding */
869 	if (apic_cpus[cpun].aci_status & APIC_CPU_SUSPEND) {
870 		for (i = APIX_AVINTR_MIN; i <= APIX_AVINTR_MAX; i++) {
871 			vecp = apixp->x_vectbl[i];
872 			if (!IS_VECT_ENABLED(vecp))
873 				continue;
874 
875 			apix_disable_vector(vecp);
876 		}
877 		lock_clear(&apix_lock);
878 		return (PSM_SUCCESS);
879 	}
880 
881 	for (i = APIX_AVINTR_MIN; i <= APIX_AVINTR_MAX; i++) {
882 		vecp = apixp->x_vectbl[i];
883 		if (!IS_VECT_ENABLED(vecp))
884 			continue;
885 
886 		if (vecp->v_flags & APIX_VECT_USER_BOUND) {
887 			hardbound++;
888 			continue;
889 		}
890 		type = vecp->v_type;
891 
892 		/*
893 		 * If there are bound interrupts on this cpu, then
894 		 * rebind them to other processors.
895 		 */
896 		loop = 0;
897 		do {
898 			bindcpu = apic_find_cpu(APIC_CPU_INTR_ENABLE);
899 
900 			if (type != APIX_TYPE_MSI)
901 				newp = apix_set_cpu(vecp, bindcpu, &ret);
902 			else
903 				newp = apix_grp_set_cpu(vecp, bindcpu, &ret);
904 		} while ((newp == NULL) && (loop++ < apic_nproc));
905 
906 		if (loop >= apic_nproc) {
907 			errbound++;
908 			cmn_err(CE_WARN, "apix: failed to rebind vector %x/%x",
909 			    vecp->v_cpuid, vecp->v_vector);
910 		}
911 	}
912 
913 	lock_clear(&apix_lock);
914 
915 	if (hardbound || errbound) {
916 		cmn_err(CE_WARN, "Could not disable interrupts on %d"
917 		    "due to user bound interrupts or failed operation",
918 		    cpun);
919 		return (PSM_FAILURE);
920 	}
921 
922 	return (PSM_SUCCESS);
923 }
924 
925 /*
926  * Bind interrupts to specified CPU
927  */
928 static void
929 apix_enable_intr(processorid_t cpun)
930 {
931 	apix_vector_t *vecp;
932 	int i, ret;
933 	processorid_t n;
934 
935 	lock_set(&apix_lock);
936 
937 	apic_cpus[cpun].aci_status |= APIC_CPU_INTR_ENABLE;
938 
939 	/* interrupt enabling for system resume */
940 	if (apic_cpus[cpun].aci_status & APIC_CPU_SUSPEND) {
941 		for (i = APIX_AVINTR_MIN; i <= APIX_AVINTR_MAX; i++) {
942 			vecp = xv_vector(cpun, i);
943 			if (!IS_VECT_ENABLED(vecp))
944 				continue;
945 
946 			apix_enable_vector(vecp);
947 		}
948 		apic_cpus[cpun].aci_status &= ~APIC_CPU_SUSPEND;
949 	}
950 
951 	for (n = 0; n < apic_nproc; n++) {
952 		if (!apic_cpu_in_range(n) || n == cpun ||
953 		    (apic_cpus[n].aci_status & APIC_CPU_INTR_ENABLE) == 0)
954 			continue;
955 
956 		for (i = APIX_AVINTR_MIN; i <= APIX_AVINTR_MAX; i++) {
957 			vecp = xv_vector(n, i);
958 			if (!IS_VECT_ENABLED(vecp) ||
959 			    vecp->v_bound_cpuid != cpun)
960 				continue;
961 
962 			if (vecp->v_type != APIX_TYPE_MSI)
963 				(void) apix_set_cpu(vecp, cpun, &ret);
964 			else
965 				(void) apix_grp_set_cpu(vecp, cpun, &ret);
966 		}
967 	}
968 
969 	lock_clear(&apix_lock);
970 }
971 
972 /*
973  * Allocate vector for IPI
974  * type == -1 indicates it is an internal request. Do not change
975  * resv_vector for these requests.
976  */
977 static int
978 apix_get_ipivect(int ipl, int type)
979 {
980 	uchar_t vector;
981 
982 	if ((vector = apix_alloc_ipi(ipl)) > 0) {
983 		if (type != -1)
984 			apic_resv_vector[ipl] = vector;
985 		return (vector);
986 	}
987 	apic_error |= APIC_ERR_GET_IPIVECT_FAIL;
988 	return (-1);	/* shouldn't happen */
989 }
990 
991 static int
992 apix_get_clkvect(int ipl)
993 {
994 	int vector;
995 
996 	if ((vector = apix_get_ipivect(ipl, -1)) == -1)
997 		return (-1);
998 
999 	apic_clkvect = vector - APIC_BASE_VECT;
1000 	APIC_VERBOSE(IPI, (CE_CONT, "apix: clock vector = %x\n",
1001 	    apic_clkvect));
1002 	return (vector);
1003 }
1004 
1005 static int
1006 apix_post_cpu_start()
1007 {
1008 	int cpun;
1009 	static int cpus_started = 1;
1010 
1011 	/* We know this CPU + BSP  started successfully. */
1012 	cpus_started++;
1013 
1014 	/*
1015 	 * On BSP we would have enabled X2APIC, if supported by processor,
1016 	 * in acpi_probe(), but on AP we do it here.
1017 	 *
1018 	 * We enable X2APIC mode only if BSP is running in X2APIC & the
1019 	 * local APIC mode of the current CPU is MMIO (xAPIC).
1020 	 */
1021 	if (apic_mode == LOCAL_X2APIC && apic_detect_x2apic() &&
1022 	    apic_local_mode() == LOCAL_APIC) {
1023 		apic_enable_x2apic();
1024 	}
1025 
1026 	/*
1027 	 * Switch back to x2apic IPI sending method for performance when target
1028 	 * CPU has entered x2apic mode.
1029 	 */
1030 	if (apic_mode == LOCAL_X2APIC) {
1031 		apic_switch_ipi_callback(B_FALSE);
1032 	}
1033 
1034 	splx(ipltospl(LOCK_LEVEL));
1035 	apix_init_intr();
1036 
1037 	/*
1038 	 * since some systems don't enable the internal cache on the non-boot
1039 	 * cpus, so we have to enable them here
1040 	 */
1041 	setcr0(getcr0() & ~(CR0_CD | CR0_NW));
1042 
1043 #ifdef	DEBUG
1044 	APIC_AV_PENDING_SET();
1045 #else
1046 	if (apic_mode == LOCAL_APIC)
1047 		APIC_AV_PENDING_SET();
1048 #endif	/* DEBUG */
1049 
1050 	/*
1051 	 * We may be booting, or resuming from suspend; aci_status will
1052 	 * be APIC_CPU_INTR_ENABLE if coming from suspend, so we add the
1053 	 * APIC_CPU_ONLINE flag here rather than setting aci_status completely.
1054 	 */
1055 	cpun = psm_get_cpu_id();
1056 	apic_cpus[cpun].aci_status |= APIC_CPU_ONLINE;
1057 
1058 	apic_reg_ops->apic_write(APIC_DIVIDE_REG, apic_divide_reg_init);
1059 
1060 	return (PSM_SUCCESS);
1061 }
1062 
1063 /*
1064  * If this module needs a periodic handler for the interrupt distribution, it
1065  * can be added here. The argument to the periodic handler is not currently
1066  * used, but is reserved for future.
1067  */
1068 static void
1069 apix_post_cyclic_setup(void *arg)
1070 {
1071 	UNREFERENCED_1PARAMETER(arg);
1072 
1073 	cyc_handler_t cyh;
1074 	cyc_time_t cyt;
1075 
1076 	/* cpu_lock is held */
1077 	/* set up a periodic handler for intr redistribution */
1078 
1079 	/*
1080 	 * In peridoc mode intr redistribution processing is done in
1081 	 * apic_intr_enter during clk intr processing
1082 	 */
1083 	if (!apic_oneshot)
1084 		return;
1085 
1086 	/*
1087 	 * Register a periodical handler for the redistribution processing.
1088 	 * Though we would generally prefer to use the DDI interface for
1089 	 * periodic handler invocation, ddi_periodic_add(9F), we are
1090 	 * unfortunately already holding cpu_lock, which ddi_periodic_add will
1091 	 * attempt to take for us.  Thus, we add our own cyclic directly:
1092 	 */
1093 	cyh.cyh_func = (void (*)(void *))apix_redistribute_compute;
1094 	cyh.cyh_arg = NULL;
1095 	cyh.cyh_level = CY_LOW_LEVEL;
1096 
1097 	cyt.cyt_when = 0;
1098 	cyt.cyt_interval = apic_redistribute_sample_interval;
1099 
1100 	apic_cyclic_id = cyclic_add(&cyh, &cyt);
1101 }
1102 
1103 /*
1104  * Called the first time we enable x2apic mode on this cpu.
1105  * Update some of the function pointers to use x2apic routines.
1106  */
1107 void
1108 x2apic_update_psm()
1109 {
1110 	struct psm_ops *pops = &apix_ops;
1111 
1112 	ASSERT(pops != NULL);
1113 
1114 	/*
1115 	 * The pcplusmp module's version of x2apic_update_psm makes additional
1116 	 * changes that we do not have to make here. It needs to make those
1117 	 * changes because pcplusmp relies on the TPR register and the means of
1118 	 * addressing that changes when using the local apic versus the x2apic.
1119 	 * It's also worth noting that the apix driver specific function end up
1120 	 * being apix_foo as opposed to apic_foo and x2apic_foo.
1121 	 */
1122 	pops->psm_send_ipi = x2apic_send_ipi;
1123 
1124 	send_dirintf = pops->psm_send_ipi;
1125 
1126 	apic_mode = LOCAL_X2APIC;
1127 	apic_change_ops();
1128 }
1129 
1130 /*
1131  * This function provides external interface to the nexus for all
1132  * functionalities related to the new DDI interrupt framework.
1133  *
1134  * Input:
1135  * dip     - pointer to the dev_info structure of the requested device
1136  * hdlp    - pointer to the internal interrupt handle structure for the
1137  *	     requested interrupt
1138  * intr_op - opcode for this call
1139  * result  - pointer to the integer that will hold the result to be
1140  *	     passed back if return value is PSM_SUCCESS
1141  *
1142  * Output:
1143  * return value is either PSM_SUCCESS or PSM_FAILURE
1144  */
1145 static int
1146 apix_intr_ops(dev_info_t *dip, ddi_intr_handle_impl_t *hdlp,
1147     psm_intr_op_t intr_op, int *result)
1148 {
1149 	int		cap;
1150 	apix_vector_t	*vecp, *newvecp;
1151 	struct intrspec *ispec, intr_spec;
1152 	processorid_t target;
1153 
1154 	ispec = &intr_spec;
1155 	ispec->intrspec_pri = hdlp->ih_pri;
1156 	ispec->intrspec_vec = hdlp->ih_inum;
1157 	ispec->intrspec_func = hdlp->ih_cb_func;
1158 
1159 	switch (intr_op) {
1160 	case PSM_INTR_OP_ALLOC_VECTORS:
1161 		switch (hdlp->ih_type) {
1162 		case DDI_INTR_TYPE_MSI:
1163 			/* allocate MSI vectors */
1164 			*result = apix_alloc_msi(dip, hdlp->ih_inum,
1165 			    hdlp->ih_scratch1,
1166 			    (int)(uintptr_t)hdlp->ih_scratch2);
1167 			break;
1168 		case DDI_INTR_TYPE_MSIX:
1169 			/* allocate MSI-X vectors */
1170 			*result = apix_alloc_msix(dip, hdlp->ih_inum,
1171 			    hdlp->ih_scratch1,
1172 			    (int)(uintptr_t)hdlp->ih_scratch2);
1173 			break;
1174 		case DDI_INTR_TYPE_FIXED:
1175 			/* allocate or share vector for fixed */
1176 			if ((ihdl_plat_t *)hdlp->ih_private == NULL) {
1177 				return (PSM_FAILURE);
1178 			}
1179 			ispec = ((ihdl_plat_t *)hdlp->ih_private)->ip_ispecp;
1180 			*result = apix_intx_alloc_vector(dip, hdlp->ih_inum,
1181 			    ispec);
1182 			break;
1183 		default:
1184 			return (PSM_FAILURE);
1185 		}
1186 		break;
1187 	case PSM_INTR_OP_FREE_VECTORS:
1188 		apix_free_vectors(dip, hdlp->ih_inum, hdlp->ih_scratch1,
1189 		    hdlp->ih_type);
1190 		break;
1191 	case PSM_INTR_OP_XLATE_VECTOR:
1192 		/*
1193 		 * Vectors are allocated by ALLOC and freed by FREE.
1194 		 * XLATE finds and returns APIX_VIRTVEC_VECTOR(cpu, vector).
1195 		 */
1196 		*result = APIX_INVALID_VECT;
1197 		vecp = apix_get_dev_map(dip, hdlp->ih_inum, hdlp->ih_type);
1198 		if (vecp != NULL) {
1199 			*result = APIX_VIRTVECTOR(vecp->v_cpuid,
1200 			    vecp->v_vector);
1201 			break;
1202 		}
1203 
1204 		/*
1205 		 * No vector to device mapping exists. If this is FIXED type
1206 		 * then check if this IRQ is already mapped for another device
1207 		 * then return the vector number for it (i.e. shared IRQ case).
1208 		 * Otherwise, return PSM_FAILURE.
1209 		 */
1210 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED) {
1211 			vecp = apix_intx_xlate_vector(dip, hdlp->ih_inum,
1212 			    ispec);
1213 			*result = (vecp == NULL) ? APIX_INVALID_VECT :
1214 			    APIX_VIRTVECTOR(vecp->v_cpuid, vecp->v_vector);
1215 		}
1216 		if (*result == APIX_INVALID_VECT)
1217 			return (PSM_FAILURE);
1218 		break;
1219 	case PSM_INTR_OP_GET_PENDING:
1220 		vecp = apix_get_dev_map(dip, hdlp->ih_inum, hdlp->ih_type);
1221 		if (vecp == NULL)
1222 			return (PSM_FAILURE);
1223 
1224 		*result = apix_get_pending(vecp);
1225 		break;
1226 	case PSM_INTR_OP_CLEAR_MASK:
1227 		if (hdlp->ih_type != DDI_INTR_TYPE_FIXED)
1228 			return (PSM_FAILURE);
1229 
1230 		vecp = apix_get_dev_map(dip, hdlp->ih_inum, hdlp->ih_type);
1231 		if (vecp == NULL)
1232 			return (PSM_FAILURE);
1233 
1234 		apix_intx_clear_mask(vecp->v_inum);
1235 		break;
1236 	case PSM_INTR_OP_SET_MASK:
1237 		if (hdlp->ih_type != DDI_INTR_TYPE_FIXED)
1238 			return (PSM_FAILURE);
1239 
1240 		vecp = apix_get_dev_map(dip, hdlp->ih_inum, hdlp->ih_type);
1241 		if (vecp == NULL)
1242 			return (PSM_FAILURE);
1243 
1244 		apix_intx_set_mask(vecp->v_inum);
1245 		break;
1246 	case PSM_INTR_OP_GET_SHARED:
1247 		if (hdlp->ih_type != DDI_INTR_TYPE_FIXED)
1248 			return (PSM_FAILURE);
1249 
1250 		vecp = apix_get_dev_map(dip, hdlp->ih_inum, hdlp->ih_type);
1251 		if (vecp == NULL)
1252 			return (PSM_FAILURE);
1253 
1254 		*result = apix_intx_get_shared(vecp->v_inum);
1255 		break;
1256 	case PSM_INTR_OP_SET_PRI:
1257 		/*
1258 		 * Called prior to adding the interrupt handler or when
1259 		 * an interrupt handler is unassigned.
1260 		 */
1261 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED)
1262 			return (PSM_SUCCESS);
1263 
1264 		if (apix_get_dev_map(dip, hdlp->ih_inum, hdlp->ih_type) == NULL)
1265 			return (PSM_FAILURE);
1266 
1267 		break;
1268 	case PSM_INTR_OP_SET_CPU:
1269 	case PSM_INTR_OP_GRP_SET_CPU:
1270 		/*
1271 		 * The interrupt handle given here has been allocated
1272 		 * specifically for this command, and ih_private carries
1273 		 * a CPU value.
1274 		 */
1275 		*result = EINVAL;
1276 		target = (int)(intptr_t)hdlp->ih_private;
1277 		if (!apic_cpu_in_range(target)) {
1278 			DDI_INTR_IMPLDBG((CE_WARN,
1279 			    "[grp_]set_cpu: cpu out of range: %d\n", target));
1280 			return (PSM_FAILURE);
1281 		}
1282 
1283 		lock_set(&apix_lock);
1284 
1285 		vecp = apix_get_req_vector(hdlp, hdlp->ih_flags);
1286 		if (!IS_VECT_ENABLED(vecp)) {
1287 			DDI_INTR_IMPLDBG((CE_WARN,
1288 			    "[grp]_set_cpu: invalid vector 0x%x\n",
1289 			    hdlp->ih_vector));
1290 			lock_clear(&apix_lock);
1291 			return (PSM_FAILURE);
1292 		}
1293 
1294 		*result = 0;
1295 
1296 		if (intr_op == PSM_INTR_OP_SET_CPU)
1297 			newvecp = apix_set_cpu(vecp, target, result);
1298 		else
1299 			newvecp = apix_grp_set_cpu(vecp, target, result);
1300 
1301 		lock_clear(&apix_lock);
1302 
1303 		if (newvecp == NULL) {
1304 			*result = EIO;
1305 			return (PSM_FAILURE);
1306 		}
1307 		newvecp->v_bound_cpuid = target;
1308 		hdlp->ih_vector = APIX_VIRTVECTOR(newvecp->v_cpuid,
1309 		    newvecp->v_vector);
1310 		break;
1311 
1312 	case PSM_INTR_OP_GET_INTR:
1313 		/*
1314 		 * The interrupt handle given here has been allocated
1315 		 * specifically for this command, and ih_private carries
1316 		 * a pointer to a apic_get_intr_t.
1317 		 */
1318 		if (apix_get_intr_info(hdlp, hdlp->ih_private) != PSM_SUCCESS)
1319 			return (PSM_FAILURE);
1320 		break;
1321 
1322 	case PSM_INTR_OP_CHECK_MSI:
1323 		/*
1324 		 * Check MSI/X is supported or not at APIC level and
1325 		 * masked off the MSI/X bits in hdlp->ih_type if not
1326 		 * supported before return.  If MSI/X is supported,
1327 		 * leave the ih_type unchanged and return.
1328 		 *
1329 		 * hdlp->ih_type passed in from the nexus has all the
1330 		 * interrupt types supported by the device.
1331 		 */
1332 		if (apic_support_msi == 0) {	/* uninitialized */
1333 			/*
1334 			 * if apic_support_msi is not set, call
1335 			 * apic_check_msi_support() to check whether msi
1336 			 * is supported first
1337 			 */
1338 			if (apic_check_msi_support() == PSM_SUCCESS)
1339 				apic_support_msi = 1;	/* supported */
1340 			else
1341 				apic_support_msi = -1;	/* not-supported */
1342 		}
1343 		if (apic_support_msi == 1) {
1344 			if (apic_msix_enable)
1345 				*result = hdlp->ih_type;
1346 			else
1347 				*result = hdlp->ih_type & ~DDI_INTR_TYPE_MSIX;
1348 		} else
1349 			*result = hdlp->ih_type & ~(DDI_INTR_TYPE_MSI |
1350 			    DDI_INTR_TYPE_MSIX);
1351 		break;
1352 	case PSM_INTR_OP_GET_CAP:
1353 		cap = DDI_INTR_FLAG_PENDING;
1354 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED)
1355 			cap |= DDI_INTR_FLAG_MASKABLE;
1356 		*result = cap;
1357 		break;
1358 	case PSM_INTR_OP_APIC_TYPE:
1359 		((apic_get_type_t *)(hdlp->ih_private))->avgi_type =
1360 		    apix_get_apic_type();
1361 		((apic_get_type_t *)(hdlp->ih_private))->avgi_num_intr =
1362 		    APIX_IPI_MIN;
1363 		((apic_get_type_t *)(hdlp->ih_private))->avgi_num_cpu =
1364 		    apic_nproc;
1365 		hdlp->ih_ver = apic_get_apic_version();
1366 		break;
1367 	case PSM_INTR_OP_SET_CAP:
1368 	default:
1369 		return (PSM_FAILURE);
1370 	}
1371 
1372 	return (PSM_SUCCESS);
1373 }
1374 
1375 static void
1376 apix_cleanup_busy(void)
1377 {
1378 	int i, j;
1379 	apix_vector_t *vecp;
1380 
1381 	for (i = 0; i < apic_nproc; i++) {
1382 		if (!apic_cpu_in_range(i))
1383 			continue;
1384 		apic_cpus[i].aci_busy = 0;
1385 		for (j = APIX_AVINTR_MIN; j < APIX_AVINTR_MAX; j++) {
1386 			if ((vecp = xv_vector(i, j)) != NULL)
1387 				vecp->v_busy = 0;
1388 		}
1389 	}
1390 }
1391 
1392 static void
1393 apix_redistribute_compute(void)
1394 {
1395 	int	i, j, max_busy;
1396 
1397 	if (!apic_enable_dynamic_migration)
1398 		return;
1399 
1400 	if (++apic_nticks == apic_sample_factor_redistribution) {
1401 		/*
1402 		 * Time to call apic_intr_redistribute().
1403 		 * reset apic_nticks. This will cause max_busy
1404 		 * to be calculated below and if it is more than
1405 		 * apic_int_busy, we will do the whole thing
1406 		 */
1407 		apic_nticks = 0;
1408 	}
1409 	max_busy = 0;
1410 	for (i = 0; i < apic_nproc; i++) {
1411 		if (!apic_cpu_in_range(i))
1412 			continue;
1413 		/*
1414 		 * Check if curipl is non zero & if ISR is in
1415 		 * progress
1416 		 */
1417 		if (((j = apic_cpus[i].aci_curipl) != 0) &&
1418 		    (apic_cpus[i].aci_ISR_in_progress & (1 << j))) {
1419 
1420 			int	vect;
1421 			apic_cpus[i].aci_busy++;
1422 			vect = apic_cpus[i].aci_current[j];
1423 			apixs[i]->x_vectbl[vect]->v_busy++;
1424 		}
1425 
1426 		if (!apic_nticks &&
1427 		    (apic_cpus[i].aci_busy > max_busy))
1428 			max_busy = apic_cpus[i].aci_busy;
1429 	}
1430 	if (!apic_nticks) {
1431 		if (max_busy > apic_int_busy_mark) {
1432 		/*
1433 		 * We could make the following check be
1434 		 * skipped > 1 in which case, we get a
1435 		 * redistribution at half the busy mark (due to
1436 		 * double interval). Need to be able to collect
1437 		 * more empirical data to decide if that is a
1438 		 * good strategy. Punt for now.
1439 		 */
1440 			apix_cleanup_busy();
1441 			apic_skipped_redistribute = 0;
1442 		} else
1443 			apic_skipped_redistribute++;
1444 	}
1445 }
1446 
1447 /*
1448  * intr_ops() service routines
1449  */
1450 
1451 static int
1452 apix_get_pending(apix_vector_t *vecp)
1453 {
1454 	int bit, index, irr, pending;
1455 
1456 	/* need to get on the bound cpu */
1457 	mutex_enter(&cpu_lock);
1458 	affinity_set(vecp->v_cpuid);
1459 
1460 	index = vecp->v_vector / 32;
1461 	bit = vecp->v_vector % 32;
1462 	irr = apic_reg_ops->apic_read(APIC_IRR_REG + index);
1463 
1464 	affinity_clear();
1465 	mutex_exit(&cpu_lock);
1466 
1467 	pending = (irr & (1 << bit)) ? 1 : 0;
1468 	if (!pending && vecp->v_type == APIX_TYPE_FIXED)
1469 		pending = apix_intx_get_pending(vecp->v_inum);
1470 
1471 	return (pending);
1472 }
1473 
1474 static apix_vector_t *
1475 apix_get_req_vector(ddi_intr_handle_impl_t *hdlp, ushort_t flags)
1476 {
1477 	apix_vector_t *vecp;
1478 	processorid_t cpuid;
1479 	int32_t virt_vec = 0;
1480 
1481 	switch (flags & PSMGI_INTRBY_FLAGS) {
1482 	case PSMGI_INTRBY_IRQ:
1483 		return (apix_intx_get_vector(hdlp->ih_vector));
1484 	case PSMGI_INTRBY_VEC:
1485 		virt_vec = (virt_vec == 0) ? hdlp->ih_vector : virt_vec;
1486 
1487 		cpuid = APIX_VIRTVEC_CPU(virt_vec);
1488 		if (!apic_cpu_in_range(cpuid))
1489 			return (NULL);
1490 
1491 		vecp = xv_vector(cpuid, APIX_VIRTVEC_VECTOR(virt_vec));
1492 		break;
1493 	case PSMGI_INTRBY_DEFAULT:
1494 		vecp = apix_get_dev_map(hdlp->ih_dip, hdlp->ih_inum,
1495 		    hdlp->ih_type);
1496 		break;
1497 	default:
1498 		return (NULL);
1499 	}
1500 
1501 	return (vecp);
1502 }
1503 
1504 static int
1505 apix_get_intr_info(ddi_intr_handle_impl_t *hdlp,
1506     apic_get_intr_t *intr_params_p)
1507 {
1508 	apix_vector_t *vecp;
1509 	struct autovec *av_dev;
1510 	int i;
1511 
1512 	vecp = apix_get_req_vector(hdlp, intr_params_p->avgi_req_flags);
1513 	if (IS_VECT_FREE(vecp)) {
1514 		intr_params_p->avgi_num_devs = 0;
1515 		intr_params_p->avgi_cpu_id = 0;
1516 		intr_params_p->avgi_req_flags = 0;
1517 		return (PSM_SUCCESS);
1518 	}
1519 
1520 	if (intr_params_p->avgi_req_flags & PSMGI_REQ_CPUID) {
1521 		intr_params_p->avgi_cpu_id = vecp->v_cpuid;
1522 
1523 		/* Return user bound info for intrd. */
1524 		if (intr_params_p->avgi_cpu_id & IRQ_USER_BOUND) {
1525 			intr_params_p->avgi_cpu_id &= ~IRQ_USER_BOUND;
1526 			intr_params_p->avgi_cpu_id |= PSMGI_CPU_USER_BOUND;
1527 		}
1528 	}
1529 
1530 	if (intr_params_p->avgi_req_flags & PSMGI_REQ_VECTOR)
1531 		intr_params_p->avgi_vector = vecp->v_vector;
1532 
1533 	if (intr_params_p->avgi_req_flags &
1534 	    (PSMGI_REQ_NUM_DEVS | PSMGI_REQ_GET_DEVS))
1535 		/* Get number of devices from apic_irq table shared field. */
1536 		intr_params_p->avgi_num_devs = vecp->v_share;
1537 
1538 	if (intr_params_p->avgi_req_flags &  PSMGI_REQ_GET_DEVS) {
1539 
1540 		intr_params_p->avgi_req_flags  |= PSMGI_REQ_NUM_DEVS;
1541 
1542 		/* Some devices have NULL dip.  Don't count these. */
1543 		if (intr_params_p->avgi_num_devs > 0) {
1544 			for (i = 0, av_dev = vecp->v_autovect; av_dev;
1545 			    av_dev = av_dev->av_link) {
1546 				if (av_dev->av_vector && av_dev->av_dip)
1547 					i++;
1548 			}
1549 			intr_params_p->avgi_num_devs =
1550 			    (uint8_t)MIN(intr_params_p->avgi_num_devs, i);
1551 		}
1552 
1553 		/* There are no viable dips to return. */
1554 		if (intr_params_p->avgi_num_devs == 0) {
1555 			intr_params_p->avgi_dip_list = NULL;
1556 
1557 		} else {	/* Return list of dips */
1558 
1559 			/* Allocate space in array for that number of devs. */
1560 			intr_params_p->avgi_dip_list = kmem_zalloc(
1561 			    intr_params_p->avgi_num_devs *
1562 			    sizeof (dev_info_t *),
1563 			    KM_NOSLEEP);
1564 			if (intr_params_p->avgi_dip_list == NULL) {
1565 				DDI_INTR_IMPLDBG((CE_WARN,
1566 				    "apix_get_vector_intr_info: no memory"));
1567 				return (PSM_FAILURE);
1568 			}
1569 
1570 			/*
1571 			 * Loop through the device list of the autovec table
1572 			 * filling in the dip array.
1573 			 *
1574 			 * Note that the autovect table may have some special
1575 			 * entries which contain NULL dips.  These will be
1576 			 * ignored.
1577 			 */
1578 			for (i = 0, av_dev = vecp->v_autovect; av_dev;
1579 			    av_dev = av_dev->av_link) {
1580 				if (av_dev->av_vector && av_dev->av_dip)
1581 					intr_params_p->avgi_dip_list[i++] =
1582 					    av_dev->av_dip;
1583 			}
1584 		}
1585 	}
1586 
1587 	return (PSM_SUCCESS);
1588 }
1589 
1590 static char *
1591 apix_get_apic_type(void)
1592 {
1593 	return (apix_psm_info.p_mach_idstring);
1594 }
1595 
1596 apix_vector_t *
1597 apix_set_cpu(apix_vector_t *vecp, int new_cpu, int *result)
1598 {
1599 	apix_vector_t *newp = NULL;
1600 	dev_info_t *dip;
1601 	int inum, cap_ptr;
1602 	ddi_acc_handle_t handle;
1603 	ddi_intr_msix_t *msix_p = NULL;
1604 	ushort_t msix_ctrl;
1605 	uintptr_t off;
1606 	uint32_t mask;
1607 
1608 	ASSERT(LOCK_HELD(&apix_lock));
1609 	*result = ENXIO;
1610 
1611 	/* Fail if this is an MSI intr and is part of a group. */
1612 	if (vecp->v_type == APIX_TYPE_MSI) {
1613 		if (i_ddi_intr_get_current_nintrs(APIX_GET_DIP(vecp)) > 1)
1614 			return (NULL);
1615 		else
1616 			return (apix_grp_set_cpu(vecp, new_cpu, result));
1617 	}
1618 
1619 	/*
1620 	 * Mask MSI-X. It's unmasked when MSI-X gets enabled.
1621 	 */
1622 	if (vecp->v_type == APIX_TYPE_MSIX && IS_VECT_ENABLED(vecp)) {
1623 		if ((dip = APIX_GET_DIP(vecp)) == NULL)
1624 			return (NULL);
1625 		inum = vecp->v_devp->dv_inum;
1626 
1627 		handle = i_ddi_get_pci_config_handle(dip);
1628 		cap_ptr = i_ddi_get_msi_msix_cap_ptr(dip);
1629 		msix_ctrl = pci_config_get16(handle, cap_ptr + PCI_MSIX_CTRL);
1630 		if ((msix_ctrl & PCI_MSIX_FUNCTION_MASK) == 0) {
1631 			/*
1632 			 * Function is not masked, then mask "inum"th
1633 			 * entry in the MSI-X table
1634 			 */
1635 			msix_p = i_ddi_get_msix(dip);
1636 			off = (uintptr_t)msix_p->msix_tbl_addr + (inum *
1637 			    PCI_MSIX_VECTOR_SIZE) + PCI_MSIX_VECTOR_CTRL_OFFSET;
1638 			mask = ddi_get32(msix_p->msix_tbl_hdl, (uint32_t *)off);
1639 			ddi_put32(msix_p->msix_tbl_hdl, (uint32_t *)off,
1640 			    mask | 1);
1641 		}
1642 	}
1643 
1644 	*result = 0;
1645 	if ((newp = apix_rebind(vecp, new_cpu, 1)) == NULL)
1646 		*result = EIO;
1647 
1648 	/* Restore mask bit */
1649 	if (msix_p != NULL)
1650 		ddi_put32(msix_p->msix_tbl_hdl, (uint32_t *)off, mask);
1651 
1652 	return (newp);
1653 }
1654 
1655 /*
1656  * Set cpu for MSIs
1657  */
1658 apix_vector_t *
1659 apix_grp_set_cpu(apix_vector_t *vecp, int new_cpu, int *result)
1660 {
1661 	apix_vector_t *newp, *vp;
1662 	uint32_t orig_cpu = vecp->v_cpuid;
1663 	int orig_vect = vecp->v_vector;
1664 	int i, num_vectors, cap_ptr, msi_mask_off;
1665 	uint32_t msi_pvm;
1666 	ushort_t msi_ctrl;
1667 	ddi_acc_handle_t handle;
1668 	dev_info_t *dip;
1669 
1670 	APIC_VERBOSE(INTR, (CE_CONT, "apix_grp_set_cpu: oldcpu: %x, vector: %x,"
1671 	    " newcpu:%x\n", vecp->v_cpuid, vecp->v_vector, new_cpu));
1672 
1673 	ASSERT(LOCK_HELD(&apix_lock));
1674 
1675 	*result = ENXIO;
1676 
1677 	if (vecp->v_type != APIX_TYPE_MSI) {
1678 		DDI_INTR_IMPLDBG((CE_WARN, "set_grp: intr not MSI\n"));
1679 		return (NULL);
1680 	}
1681 
1682 	if ((dip = APIX_GET_DIP(vecp)) == NULL)
1683 		return (NULL);
1684 
1685 	num_vectors = i_ddi_intr_get_current_nintrs(dip);
1686 	if ((num_vectors < 1) || ((num_vectors - 1) & orig_vect)) {
1687 		APIC_VERBOSE(INTR, (CE_WARN,
1688 		    "set_grp: base vec not part of a grp or not aligned: "
1689 		    "vec:0x%x, num_vec:0x%x\n", orig_vect, num_vectors));
1690 		return (NULL);
1691 	}
1692 
1693 	if (vecp->v_inum != apix_get_min_dev_inum(dip, vecp->v_type))
1694 		return (NULL);
1695 
1696 	*result = EIO;
1697 	for (i = 1; i < num_vectors; i++) {
1698 		if ((vp = xv_vector(orig_cpu, orig_vect + i)) == NULL)
1699 			return (NULL);
1700 #ifdef DEBUG
1701 		/*
1702 		 * Sanity check: CPU and dip is the same for all entries.
1703 		 * May be called when first msi to be enabled, at this time
1704 		 * add_avintr() is not called for other msi
1705 		 */
1706 		if ((vp->v_share != 0) &&
1707 		    ((APIX_GET_DIP(vp) != dip) ||
1708 		    (vp->v_cpuid != vecp->v_cpuid))) {
1709 			APIC_VERBOSE(INTR, (CE_WARN,
1710 			    "set_grp: cpu or dip for vec 0x%x difft than for "
1711 			    "vec 0x%x\n", orig_vect, orig_vect + i));
1712 			APIC_VERBOSE(INTR, (CE_WARN,
1713 			    "  cpu: %d vs %d, dip: 0x%p vs 0x%p\n", orig_cpu,
1714 			    vp->v_cpuid, (void *)dip,
1715 			    (void *)APIX_GET_DIP(vp)));
1716 			return (NULL);
1717 		}
1718 #endif /* DEBUG */
1719 	}
1720 
1721 	cap_ptr = i_ddi_get_msi_msix_cap_ptr(dip);
1722 	handle = i_ddi_get_pci_config_handle(dip);
1723 	msi_ctrl = pci_config_get16(handle, cap_ptr + PCI_MSI_CTRL);
1724 
1725 	/* MSI Per vector masking is supported. */
1726 	if (msi_ctrl & PCI_MSI_PVM_MASK) {
1727 		if (msi_ctrl &  PCI_MSI_64BIT_MASK)
1728 			msi_mask_off = cap_ptr + PCI_MSI_64BIT_MASKBITS;
1729 		else
1730 			msi_mask_off = cap_ptr + PCI_MSI_32BIT_MASK;
1731 		msi_pvm = pci_config_get32(handle, msi_mask_off);
1732 		pci_config_put32(handle, msi_mask_off, (uint32_t)-1);
1733 		APIC_VERBOSE(INTR, (CE_CONT,
1734 		    "set_grp: pvm supported.  Mask set to 0x%x\n",
1735 		    pci_config_get32(handle, msi_mask_off)));
1736 	}
1737 
1738 	if ((newp = apix_rebind(vecp, new_cpu, num_vectors)) != NULL)
1739 		*result = 0;
1740 
1741 	/* Reenable vectors if per vector masking is supported. */
1742 	if (msi_ctrl & PCI_MSI_PVM_MASK) {
1743 		pci_config_put32(handle, msi_mask_off, msi_pvm);
1744 		APIC_VERBOSE(INTR, (CE_CONT,
1745 		    "set_grp: pvm supported.  Mask restored to 0x%x\n",
1746 		    pci_config_get32(handle, msi_mask_off)));
1747 	}
1748 
1749 	return (newp);
1750 }
1751 
1752 void
1753 apix_intx_set_vector(int irqno, uint32_t cpuid, uchar_t vector)
1754 {
1755 	apic_irq_t *irqp;
1756 
1757 	mutex_enter(&airq_mutex);
1758 	irqp = apic_irq_table[irqno];
1759 	irqp->airq_cpu = cpuid;
1760 	irqp->airq_vector = vector;
1761 	apic_record_rdt_entry(irqp, irqno);
1762 	mutex_exit(&airq_mutex);
1763 }
1764 
1765 apix_vector_t *
1766 apix_intx_get_vector(int irqno)
1767 {
1768 	apic_irq_t *irqp;
1769 	uint32_t cpuid;
1770 	uchar_t vector;
1771 
1772 	mutex_enter(&airq_mutex);
1773 	irqp = apic_irq_table[irqno & 0xff];
1774 	if (IS_IRQ_FREE(irqp) || (irqp->airq_cpu == IRQ_UNINIT)) {
1775 		mutex_exit(&airq_mutex);
1776 		return (NULL);
1777 	}
1778 	cpuid = irqp->airq_cpu;
1779 	vector = irqp->airq_vector;
1780 	mutex_exit(&airq_mutex);
1781 
1782 	return (xv_vector(cpuid, vector));
1783 }
1784 
1785 /*
1786  * Must called with interrupts disabled and apic_ioapic_lock held
1787  */
1788 void
1789 apix_intx_enable(int irqno)
1790 {
1791 	uchar_t ioapicindex, intin;
1792 	apic_irq_t *irqp = apic_irq_table[irqno];
1793 	ioapic_rdt_t irdt;
1794 	apic_cpus_info_t *cpu_infop;
1795 	apix_vector_t *vecp = xv_vector(irqp->airq_cpu, irqp->airq_vector);
1796 
1797 	ASSERT(LOCK_HELD(&apic_ioapic_lock) && !IS_IRQ_FREE(irqp));
1798 
1799 	ioapicindex = irqp->airq_ioapicindex;
1800 	intin = irqp->airq_intin_no;
1801 	cpu_infop =  &apic_cpus[irqp->airq_cpu];
1802 
1803 	irdt.ir_lo = AV_PDEST | AV_FIXED | irqp->airq_rdt_entry;
1804 	irdt.ir_hi = cpu_infop->aci_local_id;
1805 
1806 	apic_vt_ops->apic_intrmap_alloc_entry(&vecp->v_intrmap_private, NULL,
1807 	    vecp->v_type, 1, ioapicindex);
1808 	apic_vt_ops->apic_intrmap_map_entry(vecp->v_intrmap_private,
1809 	    (void *)&irdt, vecp->v_type, 1);
1810 	apic_vt_ops->apic_intrmap_record_rdt(vecp->v_intrmap_private, &irdt);
1811 
1812 	/* write RDT entry high dword - destination */
1813 	WRITE_IOAPIC_RDT_ENTRY_HIGH_DWORD(ioapicindex, intin,
1814 	    irdt.ir_hi);
1815 
1816 	/* Write the vector, trigger, and polarity portion of the RDT */
1817 	WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapicindex, intin, irdt.ir_lo);
1818 
1819 	vecp->v_state = APIX_STATE_ENABLED;
1820 
1821 	APIC_VERBOSE_IOAPIC((CE_CONT, "apix_intx_enable: ioapic 0x%x"
1822 	    " intin 0x%x rdt_low 0x%x rdt_high 0x%x\n",
1823 	    ioapicindex, intin, irdt.ir_lo, irdt.ir_hi));
1824 }
1825 
1826 /*
1827  * Must called with interrupts disabled and apic_ioapic_lock held
1828  */
1829 void
1830 apix_intx_disable(int irqno)
1831 {
1832 	apic_irq_t *irqp = apic_irq_table[irqno];
1833 	int ioapicindex, intin;
1834 
1835 	ASSERT(LOCK_HELD(&apic_ioapic_lock) && !IS_IRQ_FREE(irqp));
1836 	/*
1837 	 * The assumption here is that this is safe, even for
1838 	 * systems with IOAPICs that suffer from the hardware
1839 	 * erratum because all devices have been quiesced before
1840 	 * they unregister their interrupt handlers.  If that
1841 	 * assumption turns out to be false, this mask operation
1842 	 * can induce the same erratum result we're trying to
1843 	 * avoid.
1844 	 */
1845 	ioapicindex = irqp->airq_ioapicindex;
1846 	intin = irqp->airq_intin_no;
1847 	ioapic_write(ioapicindex, APIC_RDT_CMD + 2 * intin, AV_MASK);
1848 
1849 	APIC_VERBOSE_IOAPIC((CE_CONT, "apix_intx_disable: ioapic 0x%x"
1850 	    " intin 0x%x\n", ioapicindex, intin));
1851 }
1852 
1853 void
1854 apix_intx_free(int irqno)
1855 {
1856 	apic_irq_t *irqp;
1857 
1858 	mutex_enter(&airq_mutex);
1859 	irqp = apic_irq_table[irqno];
1860 
1861 	if (IS_IRQ_FREE(irqp)) {
1862 		mutex_exit(&airq_mutex);
1863 		return;
1864 	}
1865 
1866 	irqp->airq_mps_intr_index = FREE_INDEX;
1867 	irqp->airq_cpu = IRQ_UNINIT;
1868 	irqp->airq_vector = APIX_INVALID_VECT;
1869 	mutex_exit(&airq_mutex);
1870 }
1871 
1872 #ifdef DEBUG
1873 int apix_intr_deliver_timeouts = 0;
1874 int apix_intr_rirr_timeouts = 0;
1875 int apix_intr_rirr_reset_failure = 0;
1876 #endif
1877 int apix_max_reps_irr_pending = 10;
1878 
1879 #define	GET_RDT_BITS(ioapic, intin, bits)	\
1880 	(READ_IOAPIC_RDT_ENTRY_LOW_DWORD((ioapic), (intin)) & (bits))
1881 #define	APIX_CHECK_IRR_DELAY	drv_usectohz(5000)
1882 
1883 int
1884 apix_intx_rebind(int irqno, processorid_t cpuid, uchar_t vector)
1885 {
1886 	apic_irq_t *irqp = apic_irq_table[irqno];
1887 	ulong_t iflag;
1888 	int waited, ioapic_ix, intin_no, level, repeats, rdt_entry, masked;
1889 
1890 	ASSERT(irqp != NULL);
1891 
1892 	iflag = intr_clear();
1893 	lock_set(&apic_ioapic_lock);
1894 
1895 	ioapic_ix = irqp->airq_ioapicindex;
1896 	intin_no = irqp->airq_intin_no;
1897 	level = apic_level_intr[irqno];
1898 
1899 	/*
1900 	 * Wait for the delivery status bit to be cleared. This should
1901 	 * be a very small amount of time.
1902 	 */
1903 	repeats = 0;
1904 	do {
1905 		repeats++;
1906 
1907 		for (waited = 0; waited < apic_max_reps_clear_pending;
1908 		    waited++) {
1909 			if (GET_RDT_BITS(ioapic_ix, intin_no, AV_PENDING) == 0)
1910 				break;
1911 		}
1912 		if (!level)
1913 			break;
1914 
1915 		/*
1916 		 * Mask the RDT entry for level-triggered interrupts.
1917 		 */
1918 		irqp->airq_rdt_entry |= AV_MASK;
1919 		rdt_entry = READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
1920 		    intin_no);
1921 		if ((masked = (rdt_entry & AV_MASK)) == 0) {
1922 			/* Mask it */
1923 			WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix, intin_no,
1924 			    AV_MASK | rdt_entry);
1925 		}
1926 
1927 		/*
1928 		 * If there was a race and an interrupt was injected
1929 		 * just before we masked, check for that case here.
1930 		 * Then, unmask the RDT entry and try again.  If we're
1931 		 * on our last try, don't unmask (because we want the
1932 		 * RDT entry to remain masked for the rest of the
1933 		 * function).
1934 		 */
1935 		rdt_entry = READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
1936 		    intin_no);
1937 		if ((masked == 0) && ((rdt_entry & AV_PENDING) != 0) &&
1938 		    (repeats < apic_max_reps_clear_pending)) {
1939 			/* Unmask it */
1940 			WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
1941 			    intin_no, rdt_entry & ~AV_MASK);
1942 			irqp->airq_rdt_entry &= ~AV_MASK;
1943 		}
1944 	} while ((rdt_entry & AV_PENDING) &&
1945 	    (repeats < apic_max_reps_clear_pending));
1946 
1947 #ifdef DEBUG
1948 	if (GET_RDT_BITS(ioapic_ix, intin_no, AV_PENDING) != 0)
1949 		apix_intr_deliver_timeouts++;
1950 #endif
1951 
1952 	if (!level || !APIX_IS_MASK_RDT(apix_mul_ioapic_method))
1953 		goto done;
1954 
1955 	/*
1956 	 * wait for remote IRR to be cleared for level-triggered
1957 	 * interrupts
1958 	 */
1959 	repeats = 0;
1960 	do {
1961 		repeats++;
1962 
1963 		for (waited = 0; waited < apic_max_reps_clear_pending;
1964 		    waited++) {
1965 			if (GET_RDT_BITS(ioapic_ix, intin_no, AV_REMOTE_IRR)
1966 			    == 0)
1967 				break;
1968 		}
1969 
1970 		if (GET_RDT_BITS(ioapic_ix, intin_no, AV_REMOTE_IRR) != 0) {
1971 			lock_clear(&apic_ioapic_lock);
1972 			intr_restore(iflag);
1973 
1974 			delay(APIX_CHECK_IRR_DELAY);
1975 
1976 			iflag = intr_clear();
1977 			lock_set(&apic_ioapic_lock);
1978 		}
1979 	} while (repeats < apix_max_reps_irr_pending);
1980 
1981 	if (repeats >= apix_max_reps_irr_pending) {
1982 #ifdef DEBUG
1983 		apix_intr_rirr_timeouts++;
1984 #endif
1985 
1986 		/*
1987 		 * If we waited and the Remote IRR bit is still not cleared,
1988 		 * AND if we've invoked the timeout APIC_REPROGRAM_MAX_TIMEOUTS
1989 		 * times for this interrupt, try the last-ditch workaround:
1990 		 */
1991 		if (GET_RDT_BITS(ioapic_ix, intin_no, AV_REMOTE_IRR) != 0) {
1992 			/*
1993 			 * Trying to clear the bit through normal
1994 			 * channels has failed.  So as a last-ditch
1995 			 * effort, try to set the trigger mode to
1996 			 * edge, then to level.  This has been
1997 			 * observed to work on many systems.
1998 			 */
1999 			WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
2000 			    intin_no,
2001 			    READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
2002 			    intin_no) & ~AV_LEVEL);
2003 			WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
2004 			    intin_no,
2005 			    READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
2006 			    intin_no) | AV_LEVEL);
2007 		}
2008 
2009 		if (GET_RDT_BITS(ioapic_ix, intin_no, AV_REMOTE_IRR) != 0) {
2010 #ifdef DEBUG
2011 			apix_intr_rirr_reset_failure++;
2012 #endif
2013 			lock_clear(&apic_ioapic_lock);
2014 			intr_restore(iflag);
2015 			prom_printf("apix: Remote IRR still "
2016 			    "not clear for IOAPIC %d intin %d.\n"
2017 			    "\tInterrupts to this pin may cease "
2018 			    "functioning.\n", ioapic_ix, intin_no);
2019 			return (1);	/* return failure */
2020 		}
2021 	}
2022 
2023 done:
2024 	/* change apic_irq_table */
2025 	lock_clear(&apic_ioapic_lock);
2026 	intr_restore(iflag);
2027 	apix_intx_set_vector(irqno, cpuid, vector);
2028 	iflag = intr_clear();
2029 	lock_set(&apic_ioapic_lock);
2030 
2031 	/* reprogramme IO-APIC RDT entry */
2032 	apix_intx_enable(irqno);
2033 
2034 	lock_clear(&apic_ioapic_lock);
2035 	intr_restore(iflag);
2036 
2037 	return (0);
2038 }
2039 
2040 static int
2041 apix_intx_get_pending(int irqno)
2042 {
2043 	apic_irq_t *irqp;
2044 	int intin, ioapicindex, pending;
2045 	ulong_t iflag;
2046 
2047 	mutex_enter(&airq_mutex);
2048 	irqp = apic_irq_table[irqno];
2049 	if (IS_IRQ_FREE(irqp)) {
2050 		mutex_exit(&airq_mutex);
2051 		return (0);
2052 	}
2053 
2054 	/* check IO-APIC delivery status */
2055 	intin = irqp->airq_intin_no;
2056 	ioapicindex = irqp->airq_ioapicindex;
2057 	mutex_exit(&airq_mutex);
2058 
2059 	iflag = intr_clear();
2060 	lock_set(&apic_ioapic_lock);
2061 
2062 	pending = (READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapicindex, intin) &
2063 	    AV_PENDING) ? 1 : 0;
2064 
2065 	lock_clear(&apic_ioapic_lock);
2066 	intr_restore(iflag);
2067 
2068 	return (pending);
2069 }
2070 
2071 /*
2072  * This function will mask the interrupt on the I/O APIC
2073  */
2074 static void
2075 apix_intx_set_mask(int irqno)
2076 {
2077 	int intin, ioapixindex, rdt_entry;
2078 	ulong_t iflag;
2079 	apic_irq_t *irqp;
2080 
2081 	mutex_enter(&airq_mutex);
2082 	irqp = apic_irq_table[irqno];
2083 
2084 	ASSERT(irqp->airq_mps_intr_index != FREE_INDEX);
2085 
2086 	intin = irqp->airq_intin_no;
2087 	ioapixindex = irqp->airq_ioapicindex;
2088 	mutex_exit(&airq_mutex);
2089 
2090 	iflag = intr_clear();
2091 	lock_set(&apic_ioapic_lock);
2092 
2093 	rdt_entry = READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapixindex, intin);
2094 
2095 	/* clear mask */
2096 	WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapixindex, intin,
2097 	    (AV_MASK | rdt_entry));
2098 
2099 	lock_clear(&apic_ioapic_lock);
2100 	intr_restore(iflag);
2101 }
2102 
2103 /*
2104  * This function will clear the mask for the interrupt on the I/O APIC
2105  */
2106 static void
2107 apix_intx_clear_mask(int irqno)
2108 {
2109 	int intin, ioapixindex, rdt_entry;
2110 	ulong_t iflag;
2111 	apic_irq_t *irqp;
2112 
2113 	mutex_enter(&airq_mutex);
2114 	irqp = apic_irq_table[irqno];
2115 
2116 	ASSERT(irqp->airq_mps_intr_index != FREE_INDEX);
2117 
2118 	intin = irqp->airq_intin_no;
2119 	ioapixindex = irqp->airq_ioapicindex;
2120 	mutex_exit(&airq_mutex);
2121 
2122 	iflag = intr_clear();
2123 	lock_set(&apic_ioapic_lock);
2124 
2125 	rdt_entry = READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapixindex, intin);
2126 
2127 	/* clear mask */
2128 	WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapixindex, intin,
2129 	    ((~AV_MASK) & rdt_entry));
2130 
2131 	lock_clear(&apic_ioapic_lock);
2132 	intr_restore(iflag);
2133 }
2134 
2135 /*
2136  * For level-triggered interrupt, mask the IRQ line. Mask means
2137  * new interrupts will not be delivered. The interrupt already
2138  * accepted by a local APIC is not affected
2139  */
2140 void
2141 apix_level_intr_pre_eoi(int irq)
2142 {
2143 	apic_irq_t *irqp = apic_irq_table[irq];
2144 	int apic_ix, intin_ix;
2145 
2146 	if (irqp == NULL)
2147 		return;
2148 
2149 	ASSERT(apic_level_intr[irq] == TRIGGER_MODE_LEVEL);
2150 
2151 	lock_set(&apic_ioapic_lock);
2152 
2153 	intin_ix = irqp->airq_intin_no;
2154 	apic_ix = irqp->airq_ioapicindex;
2155 
2156 	if (irqp->airq_cpu != CPU->cpu_id) {
2157 		if (!APIX_IS_MASK_RDT(apix_mul_ioapic_method))
2158 			ioapic_write_eoi(apic_ix, irqp->airq_vector);
2159 		lock_clear(&apic_ioapic_lock);
2160 		return;
2161 	}
2162 
2163 	if (apix_mul_ioapic_method == APIC_MUL_IOAPIC_IOXAPIC) {
2164 		/*
2165 		 * This is a IOxAPIC and there is EOI register:
2166 		 * 	Change the vector to reserved unused vector, so that
2167 		 * 	the EOI	from Local APIC won't clear the Remote IRR for
2168 		 * 	this level trigger interrupt. Instead, we'll manually
2169 		 * 	clear it in apix_post_hardint() after ISR handling.
2170 		 */
2171 		WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(apic_ix, intin_ix,
2172 		    (irqp->airq_rdt_entry & (~0xff)) | APIX_RESV_VECTOR);
2173 	} else {
2174 		WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(apic_ix, intin_ix,
2175 		    AV_MASK | irqp->airq_rdt_entry);
2176 	}
2177 
2178 	lock_clear(&apic_ioapic_lock);
2179 }
2180 
2181 /*
2182  * For level-triggered interrupt, unmask the IRQ line
2183  * or restore the original vector number.
2184  */
2185 void
2186 apix_level_intr_post_dispatch(int irq)
2187 {
2188 	apic_irq_t *irqp = apic_irq_table[irq];
2189 	int apic_ix, intin_ix;
2190 
2191 	if (irqp == NULL)
2192 		return;
2193 
2194 	lock_set(&apic_ioapic_lock);
2195 
2196 	intin_ix = irqp->airq_intin_no;
2197 	apic_ix = irqp->airq_ioapicindex;
2198 
2199 	if (APIX_IS_DIRECTED_EOI(apix_mul_ioapic_method)) {
2200 		/*
2201 		 * Already sent EOI back to Local APIC.
2202 		 * Send EOI to IO-APIC
2203 		 */
2204 		ioapic_write_eoi(apic_ix, irqp->airq_vector);
2205 	} else {
2206 		/* clear the mask or restore the vector */
2207 		WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(apic_ix, intin_ix,
2208 		    irqp->airq_rdt_entry);
2209 
2210 		/* send EOI to IOxAPIC */
2211 		if (apix_mul_ioapic_method == APIC_MUL_IOAPIC_IOXAPIC)
2212 			ioapic_write_eoi(apic_ix, irqp->airq_vector);
2213 	}
2214 
2215 	lock_clear(&apic_ioapic_lock);
2216 }
2217 
2218 static int
2219 apix_intx_get_shared(int irqno)
2220 {
2221 	apic_irq_t *irqp;
2222 	int share;
2223 
2224 	mutex_enter(&airq_mutex);
2225 	irqp = apic_irq_table[irqno];
2226 	if (IS_IRQ_FREE(irqp) || (irqp->airq_cpu == IRQ_UNINIT)) {
2227 		mutex_exit(&airq_mutex);
2228 		return (0);
2229 	}
2230 	share = irqp->airq_share;
2231 	mutex_exit(&airq_mutex);
2232 
2233 	return (share);
2234 }
2235 
2236 static void
2237 apix_intx_set_shared(int irqno, int delta)
2238 {
2239 	apic_irq_t *irqp;
2240 
2241 	mutex_enter(&airq_mutex);
2242 	irqp = apic_irq_table[irqno];
2243 	if (IS_IRQ_FREE(irqp)) {
2244 		mutex_exit(&airq_mutex);
2245 		return;
2246 	}
2247 	irqp->airq_share += delta;
2248 	mutex_exit(&airq_mutex);
2249 }
2250 
2251 /*
2252  * Setup IRQ table. Return IRQ no or -1 on failure
2253  */
2254 static int
2255 apix_intx_setup(dev_info_t *dip, int inum, int irqno,
2256     struct apic_io_intr *intrp, struct intrspec *ispec, iflag_t *iflagp)
2257 {
2258 	int origirq = ispec->intrspec_vec;
2259 	int newirq;
2260 	short intr_index;
2261 	uchar_t ipin, ioapic, ioapicindex;
2262 	apic_irq_t *irqp;
2263 
2264 	UNREFERENCED_1PARAMETER(inum);
2265 
2266 	if (intrp != NULL) {
2267 		intr_index = (short)(intrp - apic_io_intrp);
2268 		ioapic = intrp->intr_destid;
2269 		ipin = intrp->intr_destintin;
2270 
2271 		/* Find ioapicindex. If destid was ALL, we will exit with 0. */
2272 		for (ioapicindex = apic_io_max - 1; ioapicindex; ioapicindex--)
2273 			if (apic_io_id[ioapicindex] == ioapic)
2274 				break;
2275 		ASSERT((ioapic == apic_io_id[ioapicindex]) ||
2276 		    (ioapic == INTR_ALL_APIC));
2277 
2278 		/* check whether this intin# has been used by another irqno */
2279 		if ((newirq = apic_find_intin(ioapicindex, ipin)) != -1)
2280 			return (newirq);
2281 
2282 	} else if (iflagp != NULL) {	/* ACPI */
2283 		intr_index = ACPI_INDEX;
2284 		ioapicindex = acpi_find_ioapic(irqno);
2285 		ASSERT(ioapicindex != 0xFF);
2286 		ioapic = apic_io_id[ioapicindex];
2287 		ipin = irqno - apic_io_vectbase[ioapicindex];
2288 
2289 		if (apic_irq_table[irqno] &&
2290 		    apic_irq_table[irqno]->airq_mps_intr_index == ACPI_INDEX) {
2291 			ASSERT(apic_irq_table[irqno]->airq_intin_no == ipin &&
2292 			    apic_irq_table[irqno]->airq_ioapicindex ==
2293 			    ioapicindex);
2294 			return (irqno);
2295 		}
2296 
2297 	} else {	/* default configuration */
2298 		intr_index = DEFAULT_INDEX;
2299 		ioapicindex = 0;
2300 		ioapic = apic_io_id[ioapicindex];
2301 		ipin = (uchar_t)irqno;
2302 	}
2303 
2304 	/* allocate a new IRQ no */
2305 	if ((irqp = apic_irq_table[irqno]) == NULL) {
2306 		irqp = kmem_zalloc(sizeof (apic_irq_t), KM_SLEEP);
2307 		apic_irq_table[irqno] = irqp;
2308 	} else {
2309 		if (irqp->airq_mps_intr_index != FREE_INDEX) {
2310 			newirq = apic_allocate_irq(apic_first_avail_irq);
2311 			if (newirq == -1) {
2312 				return (-1);
2313 			}
2314 			irqno = newirq;
2315 			irqp = apic_irq_table[irqno];
2316 			ASSERT(irqp != NULL);
2317 		}
2318 	}
2319 	apic_max_device_irq = max(irqno, apic_max_device_irq);
2320 	apic_min_device_irq = min(irqno, apic_min_device_irq);
2321 
2322 	irqp->airq_mps_intr_index = intr_index;
2323 	irqp->airq_ioapicindex = ioapicindex;
2324 	irqp->airq_intin_no = ipin;
2325 	irqp->airq_dip = dip;
2326 	irqp->airq_origirq = (uchar_t)origirq;
2327 	if (iflagp != NULL)
2328 		irqp->airq_iflag = *iflagp;
2329 	irqp->airq_cpu = IRQ_UNINIT;
2330 	irqp->airq_vector = 0;
2331 
2332 	return (irqno);
2333 }
2334 
2335 /*
2336  * Setup IRQ table for non-pci devices. Return IRQ no or -1 on error
2337  */
2338 static int
2339 apix_intx_setup_nonpci(dev_info_t *dip, int inum, int bustype,
2340     struct intrspec *ispec)
2341 {
2342 	int irqno = ispec->intrspec_vec;
2343 	int newirq, i;
2344 	iflag_t intr_flag;
2345 	ACPI_SUBTABLE_HEADER	*hp;
2346 	ACPI_MADT_INTERRUPT_OVERRIDE *isop;
2347 	struct apic_io_intr *intrp;
2348 
2349 	if (!apic_enable_acpi || apic_use_acpi_madt_only) {
2350 		int busid;
2351 
2352 		if (bustype == 0)
2353 			bustype = eisa_level_intr_mask ? BUS_EISA : BUS_ISA;
2354 
2355 		/* loop checking BUS_ISA/BUS_EISA */
2356 		for (i = 0; i < 2; i++) {
2357 			if (((busid = apic_find_bus_id(bustype)) != -1) &&
2358 			    ((intrp = apic_find_io_intr_w_busid(irqno, busid))
2359 			    != NULL)) {
2360 				return (apix_intx_setup(dip, inum, irqno,
2361 				    intrp, ispec, NULL));
2362 			}
2363 			bustype = (bustype == BUS_EISA) ? BUS_ISA : BUS_EISA;
2364 		}
2365 
2366 		/* fall back to default configuration */
2367 		return (-1);
2368 	}
2369 
2370 	/* search iso entries first */
2371 	if (acpi_iso_cnt != 0) {
2372 		hp = (ACPI_SUBTABLE_HEADER *)acpi_isop;
2373 		i = 0;
2374 		while (i < acpi_iso_cnt) {
2375 			if (hp->Type == ACPI_MADT_TYPE_INTERRUPT_OVERRIDE) {
2376 				isop = (ACPI_MADT_INTERRUPT_OVERRIDE *) hp;
2377 				if (isop->Bus == 0 &&
2378 				    isop->SourceIrq == irqno) {
2379 					newirq = isop->GlobalIrq;
2380 					intr_flag.intr_po = isop->IntiFlags &
2381 					    ACPI_MADT_POLARITY_MASK;
2382 					intr_flag.intr_el = (isop->IntiFlags &
2383 					    ACPI_MADT_TRIGGER_MASK) >> 2;
2384 					intr_flag.bustype = BUS_ISA;
2385 
2386 					return (apix_intx_setup(dip, inum,
2387 					    newirq, NULL, ispec, &intr_flag));
2388 				}
2389 				i++;
2390 			}
2391 			hp = (ACPI_SUBTABLE_HEADER *)(((char *)hp) +
2392 			    hp->Length);
2393 		}
2394 	}
2395 	intr_flag.intr_po = INTR_PO_ACTIVE_HIGH;
2396 	intr_flag.intr_el = INTR_EL_EDGE;
2397 	intr_flag.bustype = BUS_ISA;
2398 	return (apix_intx_setup(dip, inum, irqno, NULL, ispec, &intr_flag));
2399 }
2400 
2401 
2402 /*
2403  * Setup IRQ table for pci devices. Return IRQ no or -1 on error
2404  */
2405 static int
2406 apix_intx_setup_pci(dev_info_t *dip, int inum, int bustype,
2407     struct intrspec *ispec)
2408 {
2409 	int busid, devid, pci_irq;
2410 	ddi_acc_handle_t cfg_handle;
2411 	uchar_t ipin;
2412 	iflag_t intr_flag;
2413 	struct apic_io_intr *intrp;
2414 
2415 	if (acpica_get_bdf(dip, &busid, &devid, NULL) != 0)
2416 		return (-1);
2417 
2418 	if (busid == 0 && apic_pci_bus_total == 1)
2419 		busid = (int)apic_single_pci_busid;
2420 
2421 	if (pci_config_setup(dip, &cfg_handle) != DDI_SUCCESS)
2422 		return (-1);
2423 	ipin = pci_config_get8(cfg_handle, PCI_CONF_IPIN) - PCI_INTA;
2424 	pci_config_teardown(&cfg_handle);
2425 
2426 	if (apic_enable_acpi && !apic_use_acpi_madt_only) {	/* ACPI */
2427 		if (apic_acpi_translate_pci_irq(dip, busid, devid,
2428 		    ipin, &pci_irq, &intr_flag) != ACPI_PSM_SUCCESS)
2429 			return (-1);
2430 
2431 		intr_flag.bustype = (uchar_t)bustype;
2432 		return (apix_intx_setup(dip, inum, pci_irq, NULL, ispec,
2433 		    &intr_flag));
2434 	}
2435 
2436 	/* MP configuration table */
2437 	pci_irq = ((devid & 0x1f) << 2) | (ipin & 0x3);
2438 	if ((intrp = apic_find_io_intr_w_busid(pci_irq, busid)) == NULL) {
2439 		pci_irq = apic_handle_pci_pci_bridge(dip, devid, ipin, &intrp);
2440 		if (pci_irq == -1)
2441 			return (-1);
2442 	}
2443 
2444 	return (apix_intx_setup(dip, inum, pci_irq, intrp, ispec, NULL));
2445 }
2446 
2447 /*
2448  * Translate and return IRQ no
2449  */
2450 static int
2451 apix_intx_xlate_irq(dev_info_t *dip, int inum, struct intrspec *ispec)
2452 {
2453 	int newirq, irqno = ispec->intrspec_vec;
2454 	int parent_is_pci_or_pciex = 0, child_is_pciex = 0;
2455 	int bustype = 0, dev_len;
2456 	char dev_type[16];
2457 
2458 	if (apic_defconf) {
2459 		mutex_enter(&airq_mutex);
2460 		goto defconf;
2461 	}
2462 
2463 	if ((dip == NULL) || (!apic_irq_translate && !apic_enable_acpi)) {
2464 		mutex_enter(&airq_mutex);
2465 		goto nonpci;
2466 	}
2467 
2468 	/*
2469 	 * use ddi_getlongprop_buf() instead of ddi_prop_lookup_string()
2470 	 * to avoid extra buffer allocation.
2471 	 */
2472 	dev_len = sizeof (dev_type);
2473 	if (ddi_getlongprop_buf(DDI_DEV_T_ANY, ddi_get_parent(dip),
2474 	    DDI_PROP_DONTPASS, "device_type", (caddr_t)dev_type,
2475 	    &dev_len) == DDI_PROP_SUCCESS) {
2476 		if ((strcmp(dev_type, "pci") == 0) ||
2477 		    (strcmp(dev_type, "pciex") == 0))
2478 			parent_is_pci_or_pciex = 1;
2479 	}
2480 
2481 	if (ddi_getlongprop_buf(DDI_DEV_T_ANY, dip,
2482 	    DDI_PROP_DONTPASS, "compatible", (caddr_t)dev_type,
2483 	    &dev_len) == DDI_PROP_SUCCESS) {
2484 		if (strstr(dev_type, "pciex"))
2485 			child_is_pciex = 1;
2486 	}
2487 
2488 	mutex_enter(&airq_mutex);
2489 
2490 	if (parent_is_pci_or_pciex) {
2491 		bustype = child_is_pciex ? BUS_PCIE : BUS_PCI;
2492 		newirq = apix_intx_setup_pci(dip, inum, bustype, ispec);
2493 		if (newirq != -1)
2494 			goto done;
2495 		bustype = 0;
2496 	} else if (strcmp(dev_type, "isa") == 0)
2497 		bustype = BUS_ISA;
2498 	else if (strcmp(dev_type, "eisa") == 0)
2499 		bustype = BUS_EISA;
2500 
2501 nonpci:
2502 	newirq = apix_intx_setup_nonpci(dip, inum, bustype, ispec);
2503 	if (newirq != -1)
2504 		goto done;
2505 
2506 defconf:
2507 	newirq = apix_intx_setup(dip, inum, irqno, NULL, ispec, NULL);
2508 	if (newirq == -1) {
2509 		mutex_exit(&airq_mutex);
2510 		return (-1);
2511 	}
2512 done:
2513 	ASSERT(apic_irq_table[newirq]);
2514 	mutex_exit(&airq_mutex);
2515 	return (newirq);
2516 }
2517 
2518 static int
2519 apix_intx_alloc_vector(dev_info_t *dip, int inum, struct intrspec *ispec)
2520 {
2521 	int irqno;
2522 	apix_vector_t *vecp;
2523 
2524 	if ((irqno = apix_intx_xlate_irq(dip, inum, ispec)) == -1)
2525 		return (0);
2526 
2527 	if ((vecp = apix_alloc_intx(dip, inum, irqno)) == NULL)
2528 		return (0);
2529 
2530 	DDI_INTR_IMPLDBG((CE_CONT, "apix_intx_alloc_vector: dip=0x%p name=%s "
2531 	    "irqno=0x%x cpuid=%d vector=0x%x\n",
2532 	    (void *)dip, ddi_driver_name(dip), irqno,
2533 	    vecp->v_cpuid, vecp->v_vector));
2534 
2535 	return (1);
2536 }
2537 
2538 /*
2539  * Return the vector number if the translated IRQ for this device
2540  * has a vector mapping setup. If no IRQ setup exists or no vector is
2541  * allocated to it then return 0.
2542  */
2543 static apix_vector_t *
2544 apix_intx_xlate_vector(dev_info_t *dip, int inum, struct intrspec *ispec)
2545 {
2546 	int irqno;
2547 	apix_vector_t *vecp;
2548 
2549 	/* get the IRQ number */
2550 	if ((irqno = apix_intx_xlate_irq(dip, inum, ispec)) == -1)
2551 		return (NULL);
2552 
2553 	/* get the vector number if a vector is allocated to this irqno */
2554 	vecp = apix_intx_get_vector(irqno);
2555 
2556 	return (vecp);
2557 }
2558 
2559 /* stub function */
2560 int
2561 apix_loaded(void)
2562 {
2563 	return (apix_is_enabled);
2564 }
2565