xref: /illumos-gate/usr/src/uts/i86pc/io/apix/apix.c (revision 1b09309c5ebed5c0bf14e2b396bf626c5aa30034)
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  * Copyright 2018 Joyent, Inc.
29  */
30 
31 /*
32  * To understand how the apix module interacts with the interrupt subsystem read
33  * the theory statement in uts/i86pc/os/intr.c.
34  */
35 
36 /*
37  * PSMI 1.1 extensions are supported only in 2.6 and later versions.
38  * PSMI 1.2 extensions are supported only in 2.7 and later versions.
39  * PSMI 1.3 and 1.4 extensions are supported in Solaris 10.
40  * PSMI 1.5 extensions are supported in Solaris Nevada.
41  * PSMI 1.6 extensions are supported in Solaris Nevada.
42  * PSMI 1.7 extensions are supported in Solaris Nevada.
43  */
44 #define	PSMI_1_7
45 
46 #include <sys/processor.h>
47 #include <sys/time.h>
48 #include <sys/psm.h>
49 #include <sys/smp_impldefs.h>
50 #include <sys/cram.h>
51 #include <sys/acpi/acpi.h>
52 #include <sys/acpica.h>
53 #include <sys/psm_common.h>
54 #include <sys/pit.h>
55 #include <sys/ddi.h>
56 #include <sys/sunddi.h>
57 #include <sys/ddi_impldefs.h>
58 #include <sys/pci.h>
59 #include <sys/promif.h>
60 #include <sys/x86_archext.h>
61 #include <sys/cpc_impl.h>
62 #include <sys/uadmin.h>
63 #include <sys/panic.h>
64 #include <sys/debug.h>
65 #include <sys/archsystm.h>
66 #include <sys/trap.h>
67 #include <sys/machsystm.h>
68 #include <sys/sysmacros.h>
69 #include <sys/cpuvar.h>
70 #include <sys/rm_platter.h>
71 #include <sys/privregs.h>
72 #include <sys/note.h>
73 #include <sys/pci_intr_lib.h>
74 #include <sys/spl.h>
75 #include <sys/clock.h>
76 #include <sys/cyclic.h>
77 #include <sys/dditypes.h>
78 #include <sys/sunddi.h>
79 #include <sys/x_call.h>
80 #include <sys/reboot.h>
81 #include <sys/mach_intr.h>
82 #include <sys/apix.h>
83 #include <sys/apix_irm_impl.h>
84 
85 static int apix_probe();
86 static void apix_init();
87 static void apix_picinit(void);
88 static int apix_intr_enter(int, int *);
89 static void apix_intr_exit(int, int);
90 static void apix_setspl(int);
91 static int apix_disable_intr(processorid_t);
92 static void apix_enable_intr(processorid_t);
93 static int apix_get_clkvect(int);
94 static int apix_get_ipivect(int, int);
95 static void apix_post_cyclic_setup(void *);
96 static int apix_post_cpu_start();
97 static int apix_intr_ops(dev_info_t *, ddi_intr_handle_impl_t *,
98     psm_intr_op_t, int *);
99 
100 /*
101  * Helper functions for apix_intr_ops()
102  */
103 static void apix_redistribute_compute(void);
104 static int apix_get_pending(apix_vector_t *);
105 static apix_vector_t *apix_get_req_vector(ddi_intr_handle_impl_t *, ushort_t);
106 static int apix_get_intr_info(ddi_intr_handle_impl_t *, apic_get_intr_t *);
107 static char *apix_get_apic_type(void);
108 static int apix_intx_get_pending(int);
109 static void apix_intx_set_mask(int irqno);
110 static void apix_intx_clear_mask(int irqno);
111 static int apix_intx_get_shared(int irqno);
112 static void apix_intx_set_shared(int irqno, int delta);
113 static apix_vector_t *apix_intx_xlate_vector(dev_info_t *, int,
114     struct intrspec *);
115 static int apix_intx_alloc_vector(dev_info_t *, int, struct intrspec *);
116 
117 extern int apic_clkinit(int);
118 
119 /* IRM initialization for APIX PSM module */
120 extern void apix_irm_init(void);
121 
122 extern int irm_enable;
123 
124 /*
125  *	Local static data
126  */
127 static struct	psm_ops apix_ops = {
128 	apix_probe,
129 
130 	apix_init,
131 	apix_picinit,
132 	apix_intr_enter,
133 	apix_intr_exit,
134 	apix_setspl,
135 	apix_addspl,
136 	apix_delspl,
137 	apix_disable_intr,
138 	apix_enable_intr,
139 	NULL,			/* psm_softlvl_to_irq */
140 	NULL,			/* psm_set_softintr */
141 
142 	apic_set_idlecpu,
143 	apic_unset_idlecpu,
144 
145 	apic_clkinit,
146 	apix_get_clkvect,
147 	NULL,			/* psm_hrtimeinit */
148 	apic_gethrtime,
149 
150 	apic_get_next_processorid,
151 	apic_cpu_start,
152 	apix_post_cpu_start,
153 	apic_shutdown,
154 	apix_get_ipivect,
155 	apic_send_ipi,
156 
157 	NULL,			/* psm_translate_irq */
158 	NULL,			/* psm_notify_error */
159 	NULL,			/* psm_notify_func */
160 	apic_timer_reprogram,
161 	apic_timer_enable,
162 	apic_timer_disable,
163 	apix_post_cyclic_setup,
164 	apic_preshutdown,
165 	apix_intr_ops,		/* Advanced DDI Interrupt framework */
166 	apic_state,		/* save, restore apic state for S3 */
167 	apic_cpu_ops,		/* CPU control interface. */
168 
169 	apic_get_pir_ipivect,
170 	apic_send_pir_ipi,
171 	apic_cmci_setup
172 };
173 
174 struct psm_ops *psmops = &apix_ops;
175 
176 static struct	psm_info apix_psm_info = {
177 	PSM_INFO_VER01_7,			/* version */
178 	PSM_OWN_EXCLUSIVE,			/* ownership */
179 	&apix_ops,				/* operation */
180 	APIX_NAME,				/* machine name */
181 	"apix MPv1.4 compatible",
182 };
183 
184 static void *apix_hdlp;
185 
186 static int apix_is_enabled = 0;
187 
188 /*
189  * apix_lock is used for cpu selection and vector re-binding
190  */
191 lock_t apix_lock;
192 apix_impl_t *apixs[NCPU];
193 /*
194  * Mapping between device interrupt and the allocated vector. Indexed
195  * by major number.
196  */
197 apix_dev_vector_t **apix_dev_vector;
198 /*
199  * Mapping between device major number and cpu id. It gets used
200  * when interrupt binding policy round robin with affinity is
201  * applied. With that policy, devices with the same major number
202  * will be bound to the same CPU.
203  */
204 processorid_t *apix_major_to_cpu;	/* major to cpu mapping */
205 kmutex_t apix_mutex;	/* for apix_dev_vector & apix_major_to_cpu */
206 
207 int apix_nipis = 16;	/* Maximum number of IPIs */
208 /*
209  * Maximum number of vectors in a CPU that can be used for interrupt
210  * allocation (including IPIs and the reserved vectors).
211  */
212 int apix_cpu_nvectors = APIX_NVECTOR;
213 
214 /* number of CPUs in power-on transition state */
215 static int apic_poweron_cnt = 0;
216 
217 /* gcpu.h */
218 
219 extern void apic_do_interrupt(struct regs *rp, trap_trace_rec_t *ttp);
220 extern void apic_change_eoi();
221 
222 /*
223  *	This is the loadable module wrapper
224  */
225 
226 int
227 _init(void)
228 {
229 	if (apic_coarse_hrtime)
230 		apix_ops.psm_gethrtime = &apic_gettime;
231 	return (psm_mod_init(&apix_hdlp, &apix_psm_info));
232 }
233 
234 int
235 _fini(void)
236 {
237 	return (psm_mod_fini(&apix_hdlp, &apix_psm_info));
238 }
239 
240 int
241 _info(struct modinfo *modinfop)
242 {
243 	return (psm_mod_info(&apix_hdlp, &apix_psm_info, modinfop));
244 }
245 
246 static int
247 apix_probe()
248 {
249 	int rval;
250 
251 	if (apix_enable == 0)
252 		return (PSM_FAILURE);
253 
254 	/*
255 	 * FIXME Temporarily disable apix module on Xen HVM platform due to
256 	 * known hang during boot (see #3605).
257 	 *
258 	 * Please remove when/if the issue is resolved.
259 	 */
260 	if (get_hwenv() & HW_XEN_HVM)
261 		return (PSM_FAILURE);
262 
263 	if (apic_local_mode() == LOCAL_X2APIC) {
264 		/* x2APIC mode activated by BIOS, switch ops */
265 		apic_mode = LOCAL_X2APIC;
266 		apic_change_ops();
267 	}
268 
269 	rval = apic_probe_common(apix_psm_info.p_mach_idstring);
270 	if (rval == PSM_SUCCESS)
271 		apix_is_enabled = 1;
272 	else
273 		apix_is_enabled = 0;
274 	return (rval);
275 }
276 
277 /*
278  * Initialize the data structures needed by pcplusmpx module.
279  * Specifically, the data structures used by addspl() and delspl()
280  * routines.
281  */
282 static void
283 apix_softinit()
284 {
285 	int i, *iptr;
286 	apix_impl_t *hdlp;
287 	int nproc;
288 
289 	nproc = max(apic_nproc, apic_max_nproc);
290 
291 	hdlp = kmem_zalloc(nproc * sizeof (apix_impl_t), KM_SLEEP);
292 	for (i = 0; i < nproc; i++) {
293 		apixs[i] = &hdlp[i];
294 		apixs[i]->x_cpuid = i;
295 		LOCK_INIT_CLEAR(&apixs[i]->x_lock);
296 	}
297 
298 	/* cpu 0 is always up (for now) */
299 	apic_cpus[0].aci_status = APIC_CPU_ONLINE | APIC_CPU_INTR_ENABLE;
300 
301 	iptr = (int *)&apic_irq_table[0];
302 	for (i = 0; i <= APIC_MAX_VECTOR; i++) {
303 		apic_level_intr[i] = 0;
304 		*iptr++ = 0;
305 	}
306 	mutex_init(&airq_mutex, NULL, MUTEX_DEFAULT, NULL);
307 
308 	apix_dev_vector = kmem_zalloc(sizeof (apix_dev_vector_t *) * devcnt,
309 	    KM_SLEEP);
310 
311 	if (apic_intr_policy == INTR_ROUND_ROBIN_WITH_AFFINITY) {
312 		apix_major_to_cpu = kmem_zalloc(sizeof (int) * devcnt,
313 		    KM_SLEEP);
314 		for (i = 0; i < devcnt; i++)
315 			apix_major_to_cpu[i] = IRQ_UNINIT;
316 	}
317 
318 	mutex_init(&apix_mutex, NULL, MUTEX_DEFAULT, NULL);
319 }
320 
321 static int
322 apix_get_pending_spl(void)
323 {
324 	int cpuid = CPU->cpu_id;
325 
326 	return (bsrw_insn(apixs[cpuid]->x_intr_pending));
327 }
328 
329 static uintptr_t
330 apix_get_intr_handler(int cpu, short vec)
331 {
332 	apix_vector_t *apix_vector;
333 
334 	ASSERT(cpu < apic_nproc && vec < APIX_NVECTOR);
335 	if (cpu >= apic_nproc || vec >= APIX_NVECTOR)
336 		return (0);
337 
338 	apix_vector = apixs[cpu]->x_vectbl[vec];
339 
340 	return ((uintptr_t)(apix_vector->v_autovect));
341 }
342 
343 static void
344 apix_init()
345 {
346 	extern void (*do_interrupt_common)(struct regs *, trap_trace_rec_t *);
347 
348 	APIC_VERBOSE(INIT, (CE_CONT, "apix: psm_softinit\n"));
349 
350 	do_interrupt_common = apix_do_interrupt;
351 	addintr = apix_add_avintr;
352 	remintr = apix_rem_avintr;
353 	get_pending_spl = apix_get_pending_spl;
354 	get_intr_handler = apix_get_intr_handler;
355 	psm_get_localapicid = apic_get_localapicid;
356 	psm_get_ioapicid = apic_get_ioapicid;
357 
358 	apix_softinit();
359 
360 	apic_pir_vect = apix_get_ipivect(XC_CPUPOKE_PIL, -1);
361 
362 	/*
363 	 * Initialize IRM pool parameters
364 	 */
365 	if (irm_enable) {
366 		int	i;
367 		int	lowest_irq;
368 		int	highest_irq;
369 
370 		/* number of CPUs present */
371 		apix_irminfo.apix_ncpus = apic_nproc;
372 		/* total number of entries in all of the IOAPICs present */
373 		lowest_irq = apic_io_vectbase[0];
374 		highest_irq = apic_io_vectend[0];
375 		for (i = 1; i < apic_io_max; i++) {
376 			if (apic_io_vectbase[i] < lowest_irq)
377 				lowest_irq = apic_io_vectbase[i];
378 			if (apic_io_vectend[i] > highest_irq)
379 				highest_irq = apic_io_vectend[i];
380 		}
381 		apix_irminfo.apix_ioapic_max_vectors =
382 		    highest_irq - lowest_irq + 1;
383 		/*
384 		 * Number of available per-CPU vectors excluding
385 		 * reserved vectors for Dtrace, int80, system-call,
386 		 * fast-trap, etc.
387 		 */
388 		apix_irminfo.apix_per_cpu_vectors = APIX_NAVINTR -
389 		    APIX_SW_RESERVED_VECTORS;
390 
391 		/* Number of vectors (pre) allocated (SCI and HPET) */
392 		apix_irminfo.apix_vectors_allocated = 0;
393 		if (apic_hpet_vect != -1)
394 			apix_irminfo.apix_vectors_allocated++;
395 		if (apic_sci_vect != -1)
396 			apix_irminfo.apix_vectors_allocated++;
397 	}
398 }
399 
400 static void
401 apix_init_intr()
402 {
403 	processorid_t	cpun = psm_get_cpu_id();
404 	uint_t nlvt;
405 	uint32_t svr = AV_UNIT_ENABLE | APIC_SPUR_INTR;
406 	extern void cmi_cmci_trap(void);
407 
408 	apic_reg_ops->apic_write_task_reg(APIC_MASK_ALL);
409 
410 	if (apic_mode == LOCAL_APIC) {
411 		/*
412 		 * We are running APIC in MMIO mode.
413 		 */
414 		if (apic_flat_model) {
415 			apic_reg_ops->apic_write(APIC_FORMAT_REG,
416 			    APIC_FLAT_MODEL);
417 		} else {
418 			apic_reg_ops->apic_write(APIC_FORMAT_REG,
419 			    APIC_CLUSTER_MODEL);
420 		}
421 
422 		apic_reg_ops->apic_write(APIC_DEST_REG,
423 		    AV_HIGH_ORDER >> cpun);
424 	}
425 
426 	if (apic_directed_EOI_supported()) {
427 		/*
428 		 * Setting the 12th bit in the Spurious Interrupt Vector
429 		 * Register suppresses broadcast EOIs generated by the local
430 		 * APIC. The suppression of broadcast EOIs happens only when
431 		 * interrupts are level-triggered.
432 		 */
433 		svr |= APIC_SVR_SUPPRESS_BROADCAST_EOI;
434 	}
435 
436 	/* need to enable APIC before unmasking NMI */
437 	apic_reg_ops->apic_write(APIC_SPUR_INT_REG, svr);
438 
439 	/*
440 	 * Presence of an invalid vector with delivery mode AV_FIXED can
441 	 * cause an error interrupt, even if the entry is masked...so
442 	 * write a valid vector to LVT entries along with the mask bit
443 	 */
444 
445 	/* All APICs have timer and LINT0/1 */
446 	apic_reg_ops->apic_write(APIC_LOCAL_TIMER, AV_MASK|APIC_RESV_IRQ);
447 	apic_reg_ops->apic_write(APIC_INT_VECT0, AV_MASK|APIC_RESV_IRQ);
448 	apic_reg_ops->apic_write(APIC_INT_VECT1, AV_NMI);	/* enable NMI */
449 
450 	/*
451 	 * On integrated APICs, the number of LVT entries is
452 	 * 'Max LVT entry' + 1; on 82489DX's (non-integrated
453 	 * APICs), nlvt is "3" (LINT0, LINT1, and timer)
454 	 */
455 
456 	if (apic_cpus[cpun].aci_local_ver < APIC_INTEGRATED_VERS) {
457 		nlvt = 3;
458 	} else {
459 		nlvt = ((apic_reg_ops->apic_read(APIC_VERS_REG) >> 16) &
460 		    0xFF) + 1;
461 	}
462 
463 	if (nlvt >= 5) {
464 		/* Enable performance counter overflow interrupt */
465 
466 		if (!is_x86_feature(x86_featureset, X86FSET_MSR))
467 			apic_enable_cpcovf_intr = 0;
468 		if (apic_enable_cpcovf_intr) {
469 			if (apic_cpcovf_vect == 0) {
470 				int ipl = APIC_PCINT_IPL;
471 
472 				apic_cpcovf_vect = apix_get_ipivect(ipl, -1);
473 				ASSERT(apic_cpcovf_vect);
474 
475 				(void) add_avintr(NULL, ipl,
476 				    (avfunc)kcpc_hw_overflow_intr,
477 				    "apic pcint", apic_cpcovf_vect,
478 				    NULL, NULL, NULL, NULL);
479 				kcpc_hw_overflow_intr_installed = 1;
480 				kcpc_hw_enable_cpc_intr =
481 				    apic_cpcovf_mask_clear;
482 			}
483 			apic_reg_ops->apic_write(APIC_PCINT_VECT,
484 			    apic_cpcovf_vect);
485 		}
486 	}
487 
488 	if (nlvt >= 6) {
489 		/* Only mask TM intr if the BIOS apparently doesn't use it */
490 
491 		uint32_t lvtval;
492 
493 		lvtval = apic_reg_ops->apic_read(APIC_THERM_VECT);
494 		if (((lvtval & AV_MASK) == AV_MASK) ||
495 		    ((lvtval & AV_DELIV_MODE) != AV_SMI)) {
496 			apic_reg_ops->apic_write(APIC_THERM_VECT,
497 			    AV_MASK|APIC_RESV_IRQ);
498 		}
499 	}
500 
501 	/* Enable error interrupt */
502 
503 	if (nlvt >= 4 && apic_enable_error_intr) {
504 		if (apic_errvect == 0) {
505 			int ipl = 0xf;	/* get highest priority intr */
506 			apic_errvect = apix_get_ipivect(ipl, -1);
507 			ASSERT(apic_errvect);
508 			/*
509 			 * Not PSMI compliant, but we are going to merge
510 			 * with ON anyway
511 			 */
512 			(void) add_avintr(NULL, ipl,
513 			    (avfunc)apic_error_intr, "apic error intr",
514 			    apic_errvect, NULL, NULL, NULL, NULL);
515 		}
516 		apic_reg_ops->apic_write(APIC_ERR_VECT, apic_errvect);
517 		apic_reg_ops->apic_write(APIC_ERROR_STATUS, 0);
518 		apic_reg_ops->apic_write(APIC_ERROR_STATUS, 0);
519 	}
520 
521 	/*
522 	 * Ensure a CMCI interrupt is allocated, regardless of whether it is
523 	 * enabled or not.
524 	 */
525 	if (apic_cmci_vect == 0) {
526 		const int ipl = 0x2;
527 		apic_cmci_vect = apix_get_ipivect(ipl, -1);
528 		ASSERT(apic_cmci_vect);
529 
530 		(void) add_avintr(NULL, ipl,
531 		    (avfunc)cmi_cmci_trap, "apic cmci intr",
532 		    apic_cmci_vect, NULL, NULL, NULL, NULL);
533 	}
534 
535 	apic_reg_ops->apic_write_task_reg(0);
536 }
537 
538 static void
539 apix_picinit(void)
540 {
541 	int i, j;
542 	uint_t isr;
543 
544 	APIC_VERBOSE(INIT, (CE_CONT, "apix: psm_picinit\n"));
545 
546 	/*
547 	 * initialize interrupt remapping before apic
548 	 * hardware initialization
549 	 */
550 	apic_intrmap_init(apic_mode);
551 	if (apic_vt_ops == psm_vt_ops)
552 		apix_mul_ioapic_method = APIC_MUL_IOAPIC_IIR;
553 
554 	/*
555 	 * On UniSys Model 6520, the BIOS leaves vector 0x20 isr
556 	 * bit on without clearing it with EOI.  Since softint
557 	 * uses vector 0x20 to interrupt itself, so softint will
558 	 * not work on this machine.  In order to fix this problem
559 	 * a check is made to verify all the isr bits are clear.
560 	 * If not, EOIs are issued to clear the bits.
561 	 */
562 	for (i = 7; i >= 1; i--) {
563 		isr = apic_reg_ops->apic_read(APIC_ISR_REG + (i * 4));
564 		if (isr != 0)
565 			for (j = 0; ((j < 32) && (isr != 0)); j++)
566 				if (isr & (1 << j)) {
567 					apic_reg_ops->apic_write(
568 					    APIC_EOI_REG, 0);
569 					isr &= ~(1 << j);
570 					apic_error |= APIC_ERR_BOOT_EOI;
571 				}
572 	}
573 
574 	/* set a flag so we know we have run apic_picinit() */
575 	apic_picinit_called = 1;
576 	LOCK_INIT_CLEAR(&apic_gethrtime_lock);
577 	LOCK_INIT_CLEAR(&apic_ioapic_lock);
578 	LOCK_INIT_CLEAR(&apic_error_lock);
579 	LOCK_INIT_CLEAR(&apic_mode_switch_lock);
580 
581 	picsetup();	 /* initialise the 8259 */
582 
583 	/* add nmi handler - least priority nmi handler */
584 	LOCK_INIT_CLEAR(&apic_nmi_lock);
585 
586 	if (!psm_add_nmintr(0, apic_nmi_intr,
587 	    "apix NMI handler", (caddr_t)NULL))
588 		cmn_err(CE_WARN, "apix: Unable to add nmi handler");
589 
590 	apix_init_intr();
591 
592 	/* enable apic mode if imcr present */
593 	if (apic_imcrp) {
594 		outb(APIC_IMCR_P1, (uchar_t)APIC_IMCR_SELECT);
595 		outb(APIC_IMCR_P2, (uchar_t)APIC_IMCR_APIC);
596 	}
597 
598 	ioapix_init_intr(IOAPIC_MASK);
599 
600 	/* setup global IRM pool if applicable */
601 	if (irm_enable)
602 		apix_irm_init();
603 }
604 
605 static __inline__ void
606 apix_send_eoi(void)
607 {
608 	if (apic_mode == LOCAL_APIC)
609 		LOCAL_APIC_WRITE_REG(APIC_EOI_REG, 0);
610 	else
611 		X2APIC_WRITE(APIC_EOI_REG, 0);
612 }
613 
614 /*
615  * platform_intr_enter
616  *
617  *	Called at the beginning of the interrupt service routine, but unlike
618  *	pcplusmp, does not mask interrupts. An EOI is given to the interrupt
619  *	controller to enable other HW interrupts but interrupts are still
620  *	masked by the IF flag.
621  *
622  *	Return -1 for spurious interrupts
623  *
624  */
625 static int
626 apix_intr_enter(int ipl, int *vectorp)
627 {
628 	struct cpu *cpu = CPU;
629 	uint32_t cpuid = CPU->cpu_id;
630 	apic_cpus_info_t *cpu_infop;
631 	uchar_t vector;
632 	apix_vector_t *vecp;
633 	int nipl = -1;
634 
635 	/*
636 	 * The real vector delivered is (*vectorp + 0x20), but our caller
637 	 * subtracts 0x20 from the vector before passing it to us.
638 	 * (That's why APIC_BASE_VECT is 0x20.)
639 	 */
640 	vector = *vectorp = (uchar_t)*vectorp + APIC_BASE_VECT;
641 
642 	cpu_infop = &apic_cpus[cpuid];
643 	if (vector == APIC_SPUR_INTR) {
644 		cpu_infop->aci_spur_cnt++;
645 		return (APIC_INT_SPURIOUS);
646 	}
647 
648 	vecp = xv_vector(cpuid, vector);
649 	if (vecp == NULL) {
650 		if (APIX_IS_FAKE_INTR(vector))
651 			nipl = apix_rebindinfo.i_pri;
652 		apix_send_eoi();
653 		return (nipl);
654 	}
655 	nipl = vecp->v_pri;
656 
657 	/* if interrupted by the clock, increment apic_nsec_since_boot */
658 	if (vector == (apic_clkvect + APIC_BASE_VECT)) {
659 		if (!apic_oneshot) {
660 			/* NOTE: this is not MT aware */
661 			apic_hrtime_stamp++;
662 			apic_nsec_since_boot += apic_nsec_per_intr;
663 			apic_hrtime_stamp++;
664 			last_count_read = apic_hertz_count;
665 			apix_redistribute_compute();
666 		}
667 
668 		apix_send_eoi();
669 
670 		return (nipl);
671 	}
672 
673 	ASSERT(vecp->v_state != APIX_STATE_OBSOLETED);
674 
675 	/* pre-EOI handling for level-triggered interrupts */
676 	if (!APIX_IS_DIRECTED_EOI(apix_mul_ioapic_method) &&
677 	    (vecp->v_type & APIX_TYPE_FIXED) && apic_level_intr[vecp->v_inum])
678 		apix_level_intr_pre_eoi(vecp->v_inum);
679 
680 	/* send back EOI */
681 	apix_send_eoi();
682 
683 	cpu_infop->aci_current[nipl] = vector;
684 	if ((nipl > ipl) && (nipl > cpu->cpu_base_spl)) {
685 		cpu_infop->aci_curipl = (uchar_t)nipl;
686 		cpu_infop->aci_ISR_in_progress |= 1 << nipl;
687 	}
688 
689 #ifdef	DEBUG
690 	if (vector >= APIX_IPI_MIN)
691 		return (nipl);	/* skip IPI */
692 
693 	APIC_DEBUG_BUF_PUT(vector);
694 	APIC_DEBUG_BUF_PUT(vecp->v_inum);
695 	APIC_DEBUG_BUF_PUT(nipl);
696 	APIC_DEBUG_BUF_PUT(psm_get_cpu_id());
697 	if ((apic_stretch_interrupts) && (apic_stretch_ISR & (1 << nipl)))
698 		drv_usecwait(apic_stretch_interrupts);
699 #endif /* DEBUG */
700 
701 	return (nipl);
702 }
703 
704 /*
705  * Any changes made to this function must also change X2APIC
706  * version of intr_exit.
707  */
708 static void
709 apix_intr_exit(int prev_ipl, int arg2)
710 {
711 	int cpuid = psm_get_cpu_id();
712 	apic_cpus_info_t *cpu_infop = &apic_cpus[cpuid];
713 	apix_impl_t *apixp = apixs[cpuid];
714 
715 	UNREFERENCED_1PARAMETER(arg2);
716 
717 	cpu_infop->aci_curipl = (uchar_t)prev_ipl;
718 	/* ISR above current pri could not be in progress */
719 	cpu_infop->aci_ISR_in_progress &= (2 << prev_ipl) - 1;
720 
721 	if (apixp->x_obsoletes != NULL) {
722 		if (APIX_CPU_LOCK_HELD(cpuid))
723 			return;
724 
725 		APIX_ENTER_CPU_LOCK(cpuid);
726 		(void) apix_obsolete_vector(apixp->x_obsoletes);
727 		APIX_LEAVE_CPU_LOCK(cpuid);
728 	}
729 }
730 
731 /*
732  * The pcplusmp setspl code uses the TPR to mask all interrupts at or below the
733  * given ipl, but apix never uses the TPR and we never mask a subset of the
734  * interrupts. They are either all blocked by the IF flag or all can come in.
735  *
736  * For setspl, we mask all interrupts for XC_HI_PIL (15), otherwise, interrupts
737  * can come in if currently enabled by the IF flag. This table shows the state
738  * of the IF flag when we leave this function.
739  *
740  *    curr IF |	ipl == 15	ipl != 15
741  *    --------+---------------------------
742  *       0    |    0		    0
743  *       1    |    0		    1
744  */
745 static void
746 apix_setspl(int ipl)
747 {
748 	/*
749 	 * Interrupts at ipl above this cannot be in progress, so the following
750 	 * mask is ok.
751 	 */
752 	apic_cpus[psm_get_cpu_id()].aci_ISR_in_progress &= (2 << ipl) - 1;
753 
754 	if (ipl == XC_HI_PIL)
755 		cli();
756 }
757 
758 int
759 apix_addspl(int virtvec, int ipl, int min_ipl, int max_ipl)
760 {
761 	uint32_t cpuid = APIX_VIRTVEC_CPU(virtvec);
762 	uchar_t vector = (uchar_t)APIX_VIRTVEC_VECTOR(virtvec);
763 	apix_vector_t *vecp = xv_vector(cpuid, vector);
764 
765 	UNREFERENCED_3PARAMETER(ipl, min_ipl, max_ipl);
766 	ASSERT(vecp != NULL && LOCK_HELD(&apix_lock));
767 
768 	if (vecp->v_type == APIX_TYPE_FIXED)
769 		apix_intx_set_shared(vecp->v_inum, 1);
770 
771 	/* There are more interrupts, so it's already been enabled */
772 	if (vecp->v_share > 1)
773 		return (PSM_SUCCESS);
774 
775 	/* return if it is not hardware interrupt */
776 	if (vecp->v_type == APIX_TYPE_IPI)
777 		return (PSM_SUCCESS);
778 
779 	/*
780 	 * if apix_picinit() has not been called yet, just return.
781 	 * At the end of apic_picinit(), we will call setup_io_intr().
782 	 */
783 	if (!apic_picinit_called)
784 		return (PSM_SUCCESS);
785 
786 	(void) apix_setup_io_intr(vecp);
787 
788 	return (PSM_SUCCESS);
789 }
790 
791 int
792 apix_delspl(int virtvec, int ipl, int min_ipl, int max_ipl)
793 {
794 	uint32_t cpuid = APIX_VIRTVEC_CPU(virtvec);
795 	uchar_t vector = (uchar_t)APIX_VIRTVEC_VECTOR(virtvec);
796 	apix_vector_t *vecp = xv_vector(cpuid, vector);
797 
798 	UNREFERENCED_3PARAMETER(ipl, min_ipl, max_ipl);
799 	ASSERT(vecp != NULL && LOCK_HELD(&apix_lock));
800 
801 	if (vecp->v_type == APIX_TYPE_FIXED)
802 		apix_intx_set_shared(vecp->v_inum, -1);
803 
804 	/* There are more interrupts */
805 	if (vecp->v_share > 1)
806 		return (PSM_SUCCESS);
807 
808 	/* return if it is not hardware interrupt */
809 	if (vecp->v_type == APIX_TYPE_IPI)
810 		return (PSM_SUCCESS);
811 
812 	if (!apic_picinit_called) {
813 		cmn_err(CE_WARN, "apix: delete 0x%x before apic init",
814 		    virtvec);
815 		return (PSM_SUCCESS);
816 	}
817 
818 	apix_disable_vector(vecp);
819 
820 	return (PSM_SUCCESS);
821 }
822 
823 /*
824  * Try and disable all interrupts. We just assign interrupts to other
825  * processors based on policy. If any were bound by user request, we
826  * let them continue and return failure. We do not bother to check
827  * for cache affinity while rebinding.
828  */
829 static int
830 apix_disable_intr(processorid_t cpun)
831 {
832 	apix_impl_t *apixp = apixs[cpun];
833 	apix_vector_t *vecp, *newp;
834 	int bindcpu, i, hardbound = 0, errbound = 0, ret, loop, type;
835 
836 	lock_set(&apix_lock);
837 
838 	apic_cpus[cpun].aci_status &= ~APIC_CPU_INTR_ENABLE;
839 	apic_cpus[cpun].aci_curipl = 0;
840 
841 	/* if this is for SUSPEND operation, skip rebinding */
842 	if (apic_cpus[cpun].aci_status & APIC_CPU_SUSPEND) {
843 		for (i = APIX_AVINTR_MIN; i <= APIX_AVINTR_MAX; i++) {
844 			vecp = apixp->x_vectbl[i];
845 			if (!IS_VECT_ENABLED(vecp))
846 				continue;
847 
848 			apix_disable_vector(vecp);
849 		}
850 		lock_clear(&apix_lock);
851 		return (PSM_SUCCESS);
852 	}
853 
854 	for (i = APIX_AVINTR_MIN; i <= APIX_AVINTR_MAX; i++) {
855 		vecp = apixp->x_vectbl[i];
856 		if (!IS_VECT_ENABLED(vecp))
857 			continue;
858 
859 		if (vecp->v_flags & APIX_VECT_USER_BOUND) {
860 			hardbound++;
861 			continue;
862 		}
863 		type = vecp->v_type;
864 
865 		/*
866 		 * If there are bound interrupts on this cpu, then
867 		 * rebind them to other processors.
868 		 */
869 		loop = 0;
870 		do {
871 			bindcpu = apic_find_cpu(APIC_CPU_INTR_ENABLE);
872 
873 			if (type != APIX_TYPE_MSI)
874 				newp = apix_set_cpu(vecp, bindcpu, &ret);
875 			else
876 				newp = apix_grp_set_cpu(vecp, bindcpu, &ret);
877 		} while ((newp == NULL) && (loop++ < apic_nproc));
878 
879 		if (loop >= apic_nproc) {
880 			errbound++;
881 			cmn_err(CE_WARN, "apix: failed to rebind vector %x/%x",
882 			    vecp->v_cpuid, vecp->v_vector);
883 		}
884 	}
885 
886 	lock_clear(&apix_lock);
887 
888 	if (hardbound || errbound) {
889 		cmn_err(CE_WARN, "Could not disable interrupts on %d"
890 		    "due to user bound interrupts or failed operation",
891 		    cpun);
892 		return (PSM_FAILURE);
893 	}
894 
895 	return (PSM_SUCCESS);
896 }
897 
898 /*
899  * Bind interrupts to specified CPU
900  */
901 static void
902 apix_enable_intr(processorid_t cpun)
903 {
904 	apix_vector_t *vecp;
905 	int i, ret;
906 	processorid_t n;
907 
908 	lock_set(&apix_lock);
909 
910 	apic_cpus[cpun].aci_status |= APIC_CPU_INTR_ENABLE;
911 
912 	/* interrupt enabling for system resume */
913 	if (apic_cpus[cpun].aci_status & APIC_CPU_SUSPEND) {
914 		for (i = APIX_AVINTR_MIN; i <= APIX_AVINTR_MAX; i++) {
915 			vecp = xv_vector(cpun, i);
916 			if (!IS_VECT_ENABLED(vecp))
917 				continue;
918 
919 			apix_enable_vector(vecp);
920 		}
921 		apic_cpus[cpun].aci_status &= ~APIC_CPU_SUSPEND;
922 	}
923 
924 	for (n = 0; n < apic_nproc; n++) {
925 		if (!apic_cpu_in_range(n) || n == cpun ||
926 		    (apic_cpus[n].aci_status & APIC_CPU_INTR_ENABLE) == 0)
927 			continue;
928 
929 		for (i = APIX_AVINTR_MIN; i <= APIX_AVINTR_MAX; i++) {
930 			vecp = xv_vector(n, i);
931 			if (!IS_VECT_ENABLED(vecp) ||
932 			    vecp->v_bound_cpuid != cpun)
933 				continue;
934 
935 			if (vecp->v_type != APIX_TYPE_MSI)
936 				(void) apix_set_cpu(vecp, cpun, &ret);
937 			else
938 				(void) apix_grp_set_cpu(vecp, cpun, &ret);
939 		}
940 	}
941 
942 	lock_clear(&apix_lock);
943 }
944 
945 /*
946  * Allocate vector for IPI
947  * type == -1 indicates it is an internal request. Do not change
948  * resv_vector for these requests.
949  */
950 static int
951 apix_get_ipivect(int ipl, int type)
952 {
953 	uchar_t vector;
954 
955 	if ((vector = apix_alloc_ipi(ipl)) > 0) {
956 		if (type != -1)
957 			apic_resv_vector[ipl] = vector;
958 		return (vector);
959 	}
960 	apic_error |= APIC_ERR_GET_IPIVECT_FAIL;
961 	return (-1);	/* shouldn't happen */
962 }
963 
964 static int
965 apix_get_clkvect(int ipl)
966 {
967 	int vector;
968 
969 	if ((vector = apix_get_ipivect(ipl, -1)) == -1)
970 		return (-1);
971 
972 	apic_clkvect = vector - APIC_BASE_VECT;
973 	APIC_VERBOSE(IPI, (CE_CONT, "apix: clock vector = %x\n",
974 	    apic_clkvect));
975 	return (vector);
976 }
977 
978 static int
979 apix_post_cpu_start()
980 {
981 	int cpun;
982 	static int cpus_started = 1;
983 
984 	/* We know this CPU + BSP  started successfully. */
985 	cpus_started++;
986 
987 	/*
988 	 * On BSP we would have enabled X2APIC, if supported by processor,
989 	 * in acpi_probe(), but on AP we do it here.
990 	 *
991 	 * We enable X2APIC mode only if BSP is running in X2APIC & the
992 	 * local APIC mode of the current CPU is MMIO (xAPIC).
993 	 */
994 	if (apic_mode == LOCAL_X2APIC && apic_detect_x2apic() &&
995 	    apic_local_mode() == LOCAL_APIC) {
996 		apic_enable_x2apic();
997 	}
998 
999 	/*
1000 	 * Switch back to x2apic IPI sending method for performance when target
1001 	 * CPU has entered x2apic mode.
1002 	 */
1003 	if (apic_mode == LOCAL_X2APIC) {
1004 		apic_switch_ipi_callback(B_FALSE);
1005 	}
1006 
1007 	splx(ipltospl(LOCK_LEVEL));
1008 	apix_init_intr();
1009 
1010 	/*
1011 	 * since some systems don't enable the internal cache on the non-boot
1012 	 * cpus, so we have to enable them here
1013 	 */
1014 	setcr0(getcr0() & ~(CR0_CD | CR0_NW));
1015 
1016 #ifdef	DEBUG
1017 	APIC_AV_PENDING_SET();
1018 #else
1019 	if (apic_mode == LOCAL_APIC)
1020 		APIC_AV_PENDING_SET();
1021 #endif	/* DEBUG */
1022 
1023 	/*
1024 	 * We may be booting, or resuming from suspend; aci_status will
1025 	 * be APIC_CPU_INTR_ENABLE if coming from suspend, so we add the
1026 	 * APIC_CPU_ONLINE flag here rather than setting aci_status completely.
1027 	 */
1028 	cpun = psm_get_cpu_id();
1029 	apic_cpus[cpun].aci_status |= APIC_CPU_ONLINE;
1030 
1031 	apic_reg_ops->apic_write(APIC_DIVIDE_REG, apic_divide_reg_init);
1032 
1033 	return (PSM_SUCCESS);
1034 }
1035 
1036 /*
1037  * If this module needs a periodic handler for the interrupt distribution, it
1038  * can be added here. The argument to the periodic handler is not currently
1039  * used, but is reserved for future.
1040  */
1041 static void
1042 apix_post_cyclic_setup(void *arg)
1043 {
1044 	UNREFERENCED_1PARAMETER(arg);
1045 
1046 	cyc_handler_t cyh;
1047 	cyc_time_t cyt;
1048 
1049 	/* cpu_lock is held */
1050 	/* set up a periodic handler for intr redistribution */
1051 
1052 	/*
1053 	 * In peridoc mode intr redistribution processing is done in
1054 	 * apic_intr_enter during clk intr processing
1055 	 */
1056 	if (!apic_oneshot)
1057 		return;
1058 
1059 	/*
1060 	 * Register a periodical handler for the redistribution processing.
1061 	 * Though we would generally prefer to use the DDI interface for
1062 	 * periodic handler invocation, ddi_periodic_add(9F), we are
1063 	 * unfortunately already holding cpu_lock, which ddi_periodic_add will
1064 	 * attempt to take for us.  Thus, we add our own cyclic directly:
1065 	 */
1066 	cyh.cyh_func = (void (*)(void *))apix_redistribute_compute;
1067 	cyh.cyh_arg = NULL;
1068 	cyh.cyh_level = CY_LOW_LEVEL;
1069 
1070 	cyt.cyt_when = 0;
1071 	cyt.cyt_interval = apic_redistribute_sample_interval;
1072 
1073 	apic_cyclic_id = cyclic_add(&cyh, &cyt);
1074 }
1075 
1076 /*
1077  * Called the first time we enable x2apic mode on this cpu.
1078  * Update some of the function pointers to use x2apic routines.
1079  */
1080 void
1081 x2apic_update_psm()
1082 {
1083 	struct psm_ops *pops = &apix_ops;
1084 
1085 	ASSERT(pops != NULL);
1086 
1087 	/*
1088 	 * The pcplusmp module's version of x2apic_update_psm makes additional
1089 	 * changes that we do not have to make here. It needs to make those
1090 	 * changes because pcplusmp relies on the TPR register and the means of
1091 	 * addressing that changes when using the local apic versus the x2apic.
1092 	 * It's also worth noting that the apix driver specific function end up
1093 	 * being apix_foo as opposed to apic_foo and x2apic_foo.
1094 	 */
1095 	pops->psm_send_ipi = x2apic_send_ipi;
1096 	send_dirintf = pops->psm_send_ipi;
1097 
1098 	pops->psm_send_pir_ipi = x2apic_send_pir_ipi;
1099 	psm_send_pir_ipi = pops->psm_send_pir_ipi;
1100 
1101 	apic_mode = LOCAL_X2APIC;
1102 	apic_change_ops();
1103 }
1104 
1105 /*
1106  * This function provides external interface to the nexus for all
1107  * functionalities related to the new DDI interrupt framework.
1108  *
1109  * Input:
1110  * dip     - pointer to the dev_info structure of the requested device
1111  * hdlp    - pointer to the internal interrupt handle structure for the
1112  *	     requested interrupt
1113  * intr_op - opcode for this call
1114  * result  - pointer to the integer that will hold the result to be
1115  *	     passed back if return value is PSM_SUCCESS
1116  *
1117  * Output:
1118  * return value is either PSM_SUCCESS or PSM_FAILURE
1119  */
1120 static int
1121 apix_intr_ops(dev_info_t *dip, ddi_intr_handle_impl_t *hdlp,
1122     psm_intr_op_t intr_op, int *result)
1123 {
1124 	int		cap;
1125 	apix_vector_t	*vecp, *newvecp;
1126 	struct intrspec *ispec, intr_spec;
1127 	processorid_t target;
1128 
1129 	ispec = &intr_spec;
1130 	ispec->intrspec_pri = hdlp->ih_pri;
1131 	ispec->intrspec_vec = hdlp->ih_inum;
1132 	ispec->intrspec_func = hdlp->ih_cb_func;
1133 
1134 	switch (intr_op) {
1135 	case PSM_INTR_OP_ALLOC_VECTORS:
1136 		switch (hdlp->ih_type) {
1137 		case DDI_INTR_TYPE_MSI:
1138 			/* allocate MSI vectors */
1139 			*result = apix_alloc_msi(dip, hdlp->ih_inum,
1140 			    hdlp->ih_scratch1,
1141 			    (int)(uintptr_t)hdlp->ih_scratch2);
1142 			break;
1143 		case DDI_INTR_TYPE_MSIX:
1144 			/* allocate MSI-X vectors */
1145 			*result = apix_alloc_msix(dip, hdlp->ih_inum,
1146 			    hdlp->ih_scratch1,
1147 			    (int)(uintptr_t)hdlp->ih_scratch2);
1148 			break;
1149 		case DDI_INTR_TYPE_FIXED:
1150 			/* allocate or share vector for fixed */
1151 			if ((ihdl_plat_t *)hdlp->ih_private == NULL) {
1152 				return (PSM_FAILURE);
1153 			}
1154 			ispec = ((ihdl_plat_t *)hdlp->ih_private)->ip_ispecp;
1155 			*result = apix_intx_alloc_vector(dip, hdlp->ih_inum,
1156 			    ispec);
1157 			break;
1158 		default:
1159 			return (PSM_FAILURE);
1160 		}
1161 		break;
1162 	case PSM_INTR_OP_FREE_VECTORS:
1163 		apix_free_vectors(dip, hdlp->ih_inum, hdlp->ih_scratch1,
1164 		    hdlp->ih_type);
1165 		break;
1166 	case PSM_INTR_OP_XLATE_VECTOR:
1167 		/*
1168 		 * Vectors are allocated by ALLOC and freed by FREE.
1169 		 * XLATE finds and returns APIX_VIRTVEC_VECTOR(cpu, vector).
1170 		 */
1171 		*result = APIX_INVALID_VECT;
1172 		vecp = apix_get_dev_map(dip, hdlp->ih_inum, hdlp->ih_type);
1173 		if (vecp != NULL) {
1174 			*result = APIX_VIRTVECTOR(vecp->v_cpuid,
1175 			    vecp->v_vector);
1176 			break;
1177 		}
1178 
1179 		/*
1180 		 * No vector to device mapping exists. If this is FIXED type
1181 		 * then check if this IRQ is already mapped for another device
1182 		 * then return the vector number for it (i.e. shared IRQ case).
1183 		 * Otherwise, return PSM_FAILURE.
1184 		 */
1185 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED) {
1186 			vecp = apix_intx_xlate_vector(dip, hdlp->ih_inum,
1187 			    ispec);
1188 			*result = (vecp == NULL) ? APIX_INVALID_VECT :
1189 			    APIX_VIRTVECTOR(vecp->v_cpuid, vecp->v_vector);
1190 		}
1191 		if (*result == APIX_INVALID_VECT)
1192 			return (PSM_FAILURE);
1193 		break;
1194 	case PSM_INTR_OP_GET_PENDING:
1195 		vecp = apix_get_dev_map(dip, hdlp->ih_inum, hdlp->ih_type);
1196 		if (vecp == NULL)
1197 			return (PSM_FAILURE);
1198 
1199 		*result = apix_get_pending(vecp);
1200 		break;
1201 	case PSM_INTR_OP_CLEAR_MASK:
1202 		if (hdlp->ih_type != DDI_INTR_TYPE_FIXED)
1203 			return (PSM_FAILURE);
1204 
1205 		vecp = apix_get_dev_map(dip, hdlp->ih_inum, hdlp->ih_type);
1206 		if (vecp == NULL)
1207 			return (PSM_FAILURE);
1208 
1209 		apix_intx_clear_mask(vecp->v_inum);
1210 		break;
1211 	case PSM_INTR_OP_SET_MASK:
1212 		if (hdlp->ih_type != DDI_INTR_TYPE_FIXED)
1213 			return (PSM_FAILURE);
1214 
1215 		vecp = apix_get_dev_map(dip, hdlp->ih_inum, hdlp->ih_type);
1216 		if (vecp == NULL)
1217 			return (PSM_FAILURE);
1218 
1219 		apix_intx_set_mask(vecp->v_inum);
1220 		break;
1221 	case PSM_INTR_OP_GET_SHARED:
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 		*result = apix_intx_get_shared(vecp->v_inum);
1230 		break;
1231 	case PSM_INTR_OP_SET_PRI:
1232 		/*
1233 		 * Called prior to adding the interrupt handler or when
1234 		 * an interrupt handler is unassigned.
1235 		 */
1236 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED)
1237 			return (PSM_SUCCESS);
1238 
1239 		if (apix_get_dev_map(dip, hdlp->ih_inum, hdlp->ih_type) == NULL)
1240 			return (PSM_FAILURE);
1241 
1242 		break;
1243 	case PSM_INTR_OP_SET_CPU:
1244 	case PSM_INTR_OP_GRP_SET_CPU:
1245 		/*
1246 		 * The interrupt handle given here has been allocated
1247 		 * specifically for this command, and ih_private carries
1248 		 * a CPU value.
1249 		 */
1250 		*result = EINVAL;
1251 		target = (int)(intptr_t)hdlp->ih_private;
1252 		if (!apic_cpu_in_range(target)) {
1253 			DDI_INTR_IMPLDBG((CE_WARN,
1254 			    "[grp_]set_cpu: cpu out of range: %d\n", target));
1255 			return (PSM_FAILURE);
1256 		}
1257 
1258 		lock_set(&apix_lock);
1259 
1260 		vecp = apix_get_req_vector(hdlp, hdlp->ih_flags);
1261 		if (!IS_VECT_ENABLED(vecp)) {
1262 			DDI_INTR_IMPLDBG((CE_WARN,
1263 			    "[grp]_set_cpu: invalid vector 0x%x\n",
1264 			    hdlp->ih_vector));
1265 			lock_clear(&apix_lock);
1266 			return (PSM_FAILURE);
1267 		}
1268 
1269 		*result = 0;
1270 
1271 		if (intr_op == PSM_INTR_OP_SET_CPU)
1272 			newvecp = apix_set_cpu(vecp, target, result);
1273 		else
1274 			newvecp = apix_grp_set_cpu(vecp, target, result);
1275 
1276 		lock_clear(&apix_lock);
1277 
1278 		if (newvecp == NULL) {
1279 			*result = EIO;
1280 			return (PSM_FAILURE);
1281 		}
1282 		newvecp->v_bound_cpuid = target;
1283 		hdlp->ih_vector = APIX_VIRTVECTOR(newvecp->v_cpuid,
1284 		    newvecp->v_vector);
1285 		break;
1286 
1287 	case PSM_INTR_OP_GET_INTR:
1288 		/*
1289 		 * The interrupt handle given here has been allocated
1290 		 * specifically for this command, and ih_private carries
1291 		 * a pointer to a apic_get_intr_t.
1292 		 */
1293 		if (apix_get_intr_info(hdlp, hdlp->ih_private) != PSM_SUCCESS)
1294 			return (PSM_FAILURE);
1295 		break;
1296 
1297 	case PSM_INTR_OP_CHECK_MSI:
1298 		/*
1299 		 * Check MSI/X is supported or not at APIC level and
1300 		 * masked off the MSI/X bits in hdlp->ih_type if not
1301 		 * supported before return.  If MSI/X is supported,
1302 		 * leave the ih_type unchanged and return.
1303 		 *
1304 		 * hdlp->ih_type passed in from the nexus has all the
1305 		 * interrupt types supported by the device.
1306 		 */
1307 		if (apic_support_msi == 0) {	/* uninitialized */
1308 			/*
1309 			 * if apic_support_msi is not set, call
1310 			 * apic_check_msi_support() to check whether msi
1311 			 * is supported first
1312 			 */
1313 			if (apic_check_msi_support() == PSM_SUCCESS)
1314 				apic_support_msi = 1;	/* supported */
1315 			else
1316 				apic_support_msi = -1;	/* not-supported */
1317 		}
1318 		if (apic_support_msi == 1) {
1319 			if (apic_msix_enable)
1320 				*result = hdlp->ih_type;
1321 			else
1322 				*result = hdlp->ih_type & ~DDI_INTR_TYPE_MSIX;
1323 		} else
1324 			*result = hdlp->ih_type & ~(DDI_INTR_TYPE_MSI |
1325 			    DDI_INTR_TYPE_MSIX);
1326 		break;
1327 	case PSM_INTR_OP_GET_CAP:
1328 		cap = DDI_INTR_FLAG_PENDING;
1329 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED)
1330 			cap |= DDI_INTR_FLAG_MASKABLE;
1331 		*result = cap;
1332 		break;
1333 	case PSM_INTR_OP_APIC_TYPE:
1334 		((apic_get_type_t *)(hdlp->ih_private))->avgi_type =
1335 		    apix_get_apic_type();
1336 		((apic_get_type_t *)(hdlp->ih_private))->avgi_num_intr =
1337 		    APIX_IPI_MIN;
1338 		((apic_get_type_t *)(hdlp->ih_private))->avgi_num_cpu =
1339 		    apic_nproc;
1340 		hdlp->ih_ver = apic_get_apic_version();
1341 		break;
1342 	case PSM_INTR_OP_SET_CAP:
1343 	default:
1344 		return (PSM_FAILURE);
1345 	}
1346 
1347 	return (PSM_SUCCESS);
1348 }
1349 
1350 static void
1351 apix_cleanup_busy(void)
1352 {
1353 	int i, j;
1354 	apix_vector_t *vecp;
1355 
1356 	for (i = 0; i < apic_nproc; i++) {
1357 		if (!apic_cpu_in_range(i))
1358 			continue;
1359 		apic_cpus[i].aci_busy = 0;
1360 		for (j = APIX_AVINTR_MIN; j < APIX_AVINTR_MAX; j++) {
1361 			if ((vecp = xv_vector(i, j)) != NULL)
1362 				vecp->v_busy = 0;
1363 		}
1364 	}
1365 }
1366 
1367 static void
1368 apix_redistribute_compute(void)
1369 {
1370 	int	i, j, max_busy;
1371 
1372 	if (!apic_enable_dynamic_migration)
1373 		return;
1374 
1375 	if (++apic_nticks == apic_sample_factor_redistribution) {
1376 		/*
1377 		 * Time to call apic_intr_redistribute().
1378 		 * reset apic_nticks. This will cause max_busy
1379 		 * to be calculated below and if it is more than
1380 		 * apic_int_busy, we will do the whole thing
1381 		 */
1382 		apic_nticks = 0;
1383 	}
1384 	max_busy = 0;
1385 	for (i = 0; i < apic_nproc; i++) {
1386 		if (!apic_cpu_in_range(i))
1387 			continue;
1388 		/*
1389 		 * Check if curipl is non zero & if ISR is in
1390 		 * progress
1391 		 */
1392 		if (((j = apic_cpus[i].aci_curipl) != 0) &&
1393 		    (apic_cpus[i].aci_ISR_in_progress & (1 << j))) {
1394 
1395 			int	vect;
1396 			apic_cpus[i].aci_busy++;
1397 			vect = apic_cpus[i].aci_current[j];
1398 			apixs[i]->x_vectbl[vect]->v_busy++;
1399 		}
1400 
1401 		if (!apic_nticks &&
1402 		    (apic_cpus[i].aci_busy > max_busy))
1403 			max_busy = apic_cpus[i].aci_busy;
1404 	}
1405 	if (!apic_nticks) {
1406 		if (max_busy > apic_int_busy_mark) {
1407 		/*
1408 		 * We could make the following check be
1409 		 * skipped > 1 in which case, we get a
1410 		 * redistribution at half the busy mark (due to
1411 		 * double interval). Need to be able to collect
1412 		 * more empirical data to decide if that is a
1413 		 * good strategy. Punt for now.
1414 		 */
1415 			apix_cleanup_busy();
1416 			apic_skipped_redistribute = 0;
1417 		} else
1418 			apic_skipped_redistribute++;
1419 	}
1420 }
1421 
1422 /*
1423  * intr_ops() service routines
1424  */
1425 
1426 static int
1427 apix_get_pending(apix_vector_t *vecp)
1428 {
1429 	int bit, index, irr, pending;
1430 
1431 	/* need to get on the bound cpu */
1432 	mutex_enter(&cpu_lock);
1433 	affinity_set(vecp->v_cpuid);
1434 
1435 	index = vecp->v_vector / 32;
1436 	bit = vecp->v_vector % 32;
1437 	irr = apic_reg_ops->apic_read(APIC_IRR_REG + index);
1438 
1439 	affinity_clear();
1440 	mutex_exit(&cpu_lock);
1441 
1442 	pending = (irr & (1 << bit)) ? 1 : 0;
1443 	if (!pending && vecp->v_type == APIX_TYPE_FIXED)
1444 		pending = apix_intx_get_pending(vecp->v_inum);
1445 
1446 	return (pending);
1447 }
1448 
1449 static apix_vector_t *
1450 apix_get_req_vector(ddi_intr_handle_impl_t *hdlp, ushort_t flags)
1451 {
1452 	apix_vector_t *vecp;
1453 	processorid_t cpuid;
1454 	int32_t virt_vec = 0;
1455 
1456 	switch (flags & PSMGI_INTRBY_FLAGS) {
1457 	case PSMGI_INTRBY_IRQ:
1458 		return (apix_intx_get_vector(hdlp->ih_vector));
1459 	case PSMGI_INTRBY_VEC:
1460 		virt_vec = (virt_vec == 0) ? hdlp->ih_vector : virt_vec;
1461 
1462 		cpuid = APIX_VIRTVEC_CPU(virt_vec);
1463 		if (!apic_cpu_in_range(cpuid))
1464 			return (NULL);
1465 
1466 		vecp = xv_vector(cpuid, APIX_VIRTVEC_VECTOR(virt_vec));
1467 		break;
1468 	case PSMGI_INTRBY_DEFAULT:
1469 		vecp = apix_get_dev_map(hdlp->ih_dip, hdlp->ih_inum,
1470 		    hdlp->ih_type);
1471 		break;
1472 	default:
1473 		return (NULL);
1474 	}
1475 
1476 	return (vecp);
1477 }
1478 
1479 static int
1480 apix_get_intr_info(ddi_intr_handle_impl_t *hdlp,
1481     apic_get_intr_t *intr_params_p)
1482 {
1483 	apix_vector_t *vecp;
1484 	struct autovec *av_dev;
1485 	int i;
1486 
1487 	vecp = apix_get_req_vector(hdlp, intr_params_p->avgi_req_flags);
1488 	if (IS_VECT_FREE(vecp)) {
1489 		intr_params_p->avgi_num_devs = 0;
1490 		intr_params_p->avgi_cpu_id = 0;
1491 		intr_params_p->avgi_req_flags = 0;
1492 		return (PSM_SUCCESS);
1493 	}
1494 
1495 	if (intr_params_p->avgi_req_flags & PSMGI_REQ_CPUID) {
1496 		intr_params_p->avgi_cpu_id = vecp->v_cpuid;
1497 
1498 		/* Return user bound info for intrd. */
1499 		if (intr_params_p->avgi_cpu_id & IRQ_USER_BOUND) {
1500 			intr_params_p->avgi_cpu_id &= ~IRQ_USER_BOUND;
1501 			intr_params_p->avgi_cpu_id |= PSMGI_CPU_USER_BOUND;
1502 		}
1503 	}
1504 
1505 	if (intr_params_p->avgi_req_flags & PSMGI_REQ_VECTOR)
1506 		intr_params_p->avgi_vector = vecp->v_vector;
1507 
1508 	if (intr_params_p->avgi_req_flags &
1509 	    (PSMGI_REQ_NUM_DEVS | PSMGI_REQ_GET_DEVS))
1510 		/* Get number of devices from apic_irq table shared field. */
1511 		intr_params_p->avgi_num_devs = vecp->v_share;
1512 
1513 	if (intr_params_p->avgi_req_flags &  PSMGI_REQ_GET_DEVS) {
1514 
1515 		intr_params_p->avgi_req_flags  |= PSMGI_REQ_NUM_DEVS;
1516 
1517 		/* Some devices have NULL dip.  Don't count these. */
1518 		if (intr_params_p->avgi_num_devs > 0) {
1519 			for (i = 0, av_dev = vecp->v_autovect; av_dev;
1520 			    av_dev = av_dev->av_link) {
1521 				if (av_dev->av_vector && av_dev->av_dip)
1522 					i++;
1523 			}
1524 			intr_params_p->avgi_num_devs =
1525 			    (uint8_t)MIN(intr_params_p->avgi_num_devs, i);
1526 		}
1527 
1528 		/* There are no viable dips to return. */
1529 		if (intr_params_p->avgi_num_devs == 0) {
1530 			intr_params_p->avgi_dip_list = NULL;
1531 
1532 		} else {	/* Return list of dips */
1533 
1534 			/* Allocate space in array for that number of devs. */
1535 			intr_params_p->avgi_dip_list = kmem_zalloc(
1536 			    intr_params_p->avgi_num_devs *
1537 			    sizeof (dev_info_t *),
1538 			    KM_NOSLEEP);
1539 			if (intr_params_p->avgi_dip_list == NULL) {
1540 				DDI_INTR_IMPLDBG((CE_WARN,
1541 				    "apix_get_vector_intr_info: no memory"));
1542 				return (PSM_FAILURE);
1543 			}
1544 
1545 			/*
1546 			 * Loop through the device list of the autovec table
1547 			 * filling in the dip array.
1548 			 *
1549 			 * Note that the autovect table may have some special
1550 			 * entries which contain NULL dips.  These will be
1551 			 * ignored.
1552 			 */
1553 			for (i = 0, av_dev = vecp->v_autovect; av_dev;
1554 			    av_dev = av_dev->av_link) {
1555 				if (av_dev->av_vector && av_dev->av_dip)
1556 					intr_params_p->avgi_dip_list[i++] =
1557 					    av_dev->av_dip;
1558 			}
1559 		}
1560 	}
1561 
1562 	return (PSM_SUCCESS);
1563 }
1564 
1565 static char *
1566 apix_get_apic_type(void)
1567 {
1568 	return (apix_psm_info.p_mach_idstring);
1569 }
1570 
1571 apix_vector_t *
1572 apix_set_cpu(apix_vector_t *vecp, int new_cpu, int *result)
1573 {
1574 	apix_vector_t *newp = NULL;
1575 	dev_info_t *dip;
1576 	int inum, cap_ptr;
1577 	ddi_acc_handle_t handle;
1578 	ddi_intr_msix_t *msix_p = NULL;
1579 	ushort_t msix_ctrl;
1580 	uintptr_t off = 0;
1581 	uint32_t mask = 0;
1582 
1583 	ASSERT(LOCK_HELD(&apix_lock));
1584 	*result = ENXIO;
1585 
1586 	/* Fail if this is an MSI intr and is part of a group. */
1587 	if (vecp->v_type == APIX_TYPE_MSI) {
1588 		if (i_ddi_intr_get_current_nintrs(APIX_GET_DIP(vecp)) > 1)
1589 			return (NULL);
1590 		else
1591 			return (apix_grp_set_cpu(vecp, new_cpu, result));
1592 	}
1593 
1594 	/*
1595 	 * Mask MSI-X. It's unmasked when MSI-X gets enabled.
1596 	 */
1597 	if (vecp->v_type == APIX_TYPE_MSIX && IS_VECT_ENABLED(vecp)) {
1598 		if ((dip = APIX_GET_DIP(vecp)) == NULL)
1599 			return (NULL);
1600 		inum = vecp->v_devp->dv_inum;
1601 
1602 		handle = i_ddi_get_pci_config_handle(dip);
1603 		cap_ptr = i_ddi_get_msi_msix_cap_ptr(dip);
1604 		msix_ctrl = pci_config_get16(handle, cap_ptr + PCI_MSIX_CTRL);
1605 		if ((msix_ctrl & PCI_MSIX_FUNCTION_MASK) == 0) {
1606 			/*
1607 			 * Function is not masked, then mask "inum"th
1608 			 * entry in the MSI-X table
1609 			 */
1610 			msix_p = i_ddi_get_msix(dip);
1611 			off = (uintptr_t)msix_p->msix_tbl_addr + (inum *
1612 			    PCI_MSIX_VECTOR_SIZE) + PCI_MSIX_VECTOR_CTRL_OFFSET;
1613 			mask = ddi_get32(msix_p->msix_tbl_hdl, (uint32_t *)off);
1614 			ddi_put32(msix_p->msix_tbl_hdl, (uint32_t *)off,
1615 			    mask | 1);
1616 		}
1617 	}
1618 
1619 	*result = 0;
1620 	if ((newp = apix_rebind(vecp, new_cpu, 1)) == NULL)
1621 		*result = EIO;
1622 
1623 	/* Restore mask bit */
1624 	if (msix_p != NULL)
1625 		ddi_put32(msix_p->msix_tbl_hdl, (uint32_t *)off, mask);
1626 
1627 	return (newp);
1628 }
1629 
1630 /*
1631  * Set cpu for MSIs
1632  */
1633 apix_vector_t *
1634 apix_grp_set_cpu(apix_vector_t *vecp, int new_cpu, int *result)
1635 {
1636 	apix_vector_t *newp, *vp;
1637 	uint32_t orig_cpu = vecp->v_cpuid;
1638 	int orig_vect = vecp->v_vector;
1639 	int i, num_vectors, cap_ptr, msi_mask_off = 0;
1640 	uint32_t msi_pvm = 0;
1641 	ushort_t msi_ctrl;
1642 	ddi_acc_handle_t handle;
1643 	dev_info_t *dip;
1644 
1645 	APIC_VERBOSE(INTR, (CE_CONT, "apix_grp_set_cpu: oldcpu: %x, vector: %x,"
1646 	    " newcpu:%x\n", vecp->v_cpuid, vecp->v_vector, new_cpu));
1647 
1648 	ASSERT(LOCK_HELD(&apix_lock));
1649 
1650 	*result = ENXIO;
1651 
1652 	if (vecp->v_type != APIX_TYPE_MSI) {
1653 		DDI_INTR_IMPLDBG((CE_WARN, "set_grp: intr not MSI\n"));
1654 		return (NULL);
1655 	}
1656 
1657 	if ((dip = APIX_GET_DIP(vecp)) == NULL)
1658 		return (NULL);
1659 
1660 	num_vectors = i_ddi_intr_get_current_nintrs(dip);
1661 	if ((num_vectors < 1) || ((num_vectors - 1) & orig_vect)) {
1662 		APIC_VERBOSE(INTR, (CE_WARN,
1663 		    "set_grp: base vec not part of a grp or not aligned: "
1664 		    "vec:0x%x, num_vec:0x%x\n", orig_vect, num_vectors));
1665 		return (NULL);
1666 	}
1667 
1668 	if (vecp->v_inum != apix_get_min_dev_inum(dip, vecp->v_type))
1669 		return (NULL);
1670 
1671 	*result = EIO;
1672 	for (i = 1; i < num_vectors; i++) {
1673 		if ((vp = xv_vector(orig_cpu, orig_vect + i)) == NULL)
1674 			return (NULL);
1675 #ifdef DEBUG
1676 		/*
1677 		 * Sanity check: CPU and dip is the same for all entries.
1678 		 * May be called when first msi to be enabled, at this time
1679 		 * add_avintr() is not called for other msi
1680 		 */
1681 		if ((vp->v_share != 0) &&
1682 		    ((APIX_GET_DIP(vp) != dip) ||
1683 		    (vp->v_cpuid != vecp->v_cpuid))) {
1684 			APIC_VERBOSE(INTR, (CE_WARN,
1685 			    "set_grp: cpu or dip for vec 0x%x difft than for "
1686 			    "vec 0x%x\n", orig_vect, orig_vect + i));
1687 			APIC_VERBOSE(INTR, (CE_WARN,
1688 			    "  cpu: %d vs %d, dip: 0x%p vs 0x%p\n", orig_cpu,
1689 			    vp->v_cpuid, (void *)dip,
1690 			    (void *)APIX_GET_DIP(vp)));
1691 			return (NULL);
1692 		}
1693 #endif /* DEBUG */
1694 	}
1695 
1696 	cap_ptr = i_ddi_get_msi_msix_cap_ptr(dip);
1697 	handle = i_ddi_get_pci_config_handle(dip);
1698 	msi_ctrl = pci_config_get16(handle, cap_ptr + PCI_MSI_CTRL);
1699 
1700 	/* MSI Per vector masking is supported. */
1701 	if (msi_ctrl & PCI_MSI_PVM_MASK) {
1702 		if (msi_ctrl &  PCI_MSI_64BIT_MASK)
1703 			msi_mask_off = cap_ptr + PCI_MSI_64BIT_MASKBITS;
1704 		else
1705 			msi_mask_off = cap_ptr + PCI_MSI_32BIT_MASK;
1706 		msi_pvm = pci_config_get32(handle, msi_mask_off);
1707 		pci_config_put32(handle, msi_mask_off, (uint32_t)-1);
1708 		APIC_VERBOSE(INTR, (CE_CONT,
1709 		    "set_grp: pvm supported.  Mask set to 0x%x\n",
1710 		    pci_config_get32(handle, msi_mask_off)));
1711 	}
1712 
1713 	if ((newp = apix_rebind(vecp, new_cpu, num_vectors)) != NULL)
1714 		*result = 0;
1715 
1716 	/* Reenable vectors if per vector masking is supported. */
1717 	if (msi_ctrl & PCI_MSI_PVM_MASK) {
1718 		pci_config_put32(handle, msi_mask_off, msi_pvm);
1719 		APIC_VERBOSE(INTR, (CE_CONT,
1720 		    "set_grp: pvm supported.  Mask restored to 0x%x\n",
1721 		    pci_config_get32(handle, msi_mask_off)));
1722 	}
1723 
1724 	return (newp);
1725 }
1726 
1727 void
1728 apix_intx_set_vector(int irqno, uint32_t cpuid, uchar_t vector)
1729 {
1730 	apic_irq_t *irqp;
1731 
1732 	mutex_enter(&airq_mutex);
1733 	irqp = apic_irq_table[irqno];
1734 	irqp->airq_cpu = cpuid;
1735 	irqp->airq_vector = vector;
1736 	apic_record_rdt_entry(irqp, irqno);
1737 	mutex_exit(&airq_mutex);
1738 }
1739 
1740 apix_vector_t *
1741 apix_intx_get_vector(int irqno)
1742 {
1743 	apic_irq_t *irqp;
1744 	uint32_t cpuid;
1745 	uchar_t vector;
1746 
1747 	mutex_enter(&airq_mutex);
1748 	irqp = apic_irq_table[irqno & 0xff];
1749 	if (IS_IRQ_FREE(irqp) || (irqp->airq_cpu == IRQ_UNINIT)) {
1750 		mutex_exit(&airq_mutex);
1751 		return (NULL);
1752 	}
1753 	cpuid = irqp->airq_cpu;
1754 	vector = irqp->airq_vector;
1755 	mutex_exit(&airq_mutex);
1756 
1757 	return (xv_vector(cpuid, vector));
1758 }
1759 
1760 /*
1761  * Must called with interrupts disabled and apic_ioapic_lock held
1762  */
1763 void
1764 apix_intx_enable(int irqno)
1765 {
1766 	uchar_t ioapicindex, intin;
1767 	apic_irq_t *irqp = apic_irq_table[irqno];
1768 	ioapic_rdt_t irdt;
1769 	apic_cpus_info_t *cpu_infop;
1770 	apix_vector_t *vecp = xv_vector(irqp->airq_cpu, irqp->airq_vector);
1771 
1772 	ASSERT(LOCK_HELD(&apic_ioapic_lock) && !IS_IRQ_FREE(irqp));
1773 
1774 	ioapicindex = irqp->airq_ioapicindex;
1775 	intin = irqp->airq_intin_no;
1776 	cpu_infop =  &apic_cpus[irqp->airq_cpu];
1777 
1778 	irdt.ir_lo = AV_PDEST | AV_FIXED | irqp->airq_rdt_entry;
1779 	irdt.ir_hi = cpu_infop->aci_local_id;
1780 
1781 	apic_vt_ops->apic_intrmap_alloc_entry(&vecp->v_intrmap_private, NULL,
1782 	    vecp->v_type, 1, ioapicindex);
1783 	apic_vt_ops->apic_intrmap_map_entry(vecp->v_intrmap_private,
1784 	    (void *)&irdt, vecp->v_type, 1);
1785 	apic_vt_ops->apic_intrmap_record_rdt(vecp->v_intrmap_private, &irdt);
1786 
1787 	/* write RDT entry high dword - destination */
1788 	WRITE_IOAPIC_RDT_ENTRY_HIGH_DWORD(ioapicindex, intin,
1789 	    irdt.ir_hi);
1790 
1791 	/* Write the vector, trigger, and polarity portion of the RDT */
1792 	WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapicindex, intin, irdt.ir_lo);
1793 
1794 	vecp->v_state = APIX_STATE_ENABLED;
1795 
1796 	APIC_VERBOSE_IOAPIC((CE_CONT, "apix_intx_enable: ioapic 0x%x"
1797 	    " intin 0x%x rdt_low 0x%x rdt_high 0x%x\n",
1798 	    ioapicindex, intin, irdt.ir_lo, irdt.ir_hi));
1799 }
1800 
1801 /*
1802  * Must called with interrupts disabled and apic_ioapic_lock held
1803  */
1804 void
1805 apix_intx_disable(int irqno)
1806 {
1807 	apic_irq_t *irqp = apic_irq_table[irqno];
1808 	int ioapicindex, intin;
1809 
1810 	ASSERT(LOCK_HELD(&apic_ioapic_lock) && !IS_IRQ_FREE(irqp));
1811 	/*
1812 	 * The assumption here is that this is safe, even for
1813 	 * systems with IOAPICs that suffer from the hardware
1814 	 * erratum because all devices have been quiesced before
1815 	 * they unregister their interrupt handlers.  If that
1816 	 * assumption turns out to be false, this mask operation
1817 	 * can induce the same erratum result we're trying to
1818 	 * avoid.
1819 	 */
1820 	ioapicindex = irqp->airq_ioapicindex;
1821 	intin = irqp->airq_intin_no;
1822 	ioapic_write(ioapicindex, APIC_RDT_CMD + 2 * intin, AV_MASK);
1823 
1824 	APIC_VERBOSE_IOAPIC((CE_CONT, "apix_intx_disable: ioapic 0x%x"
1825 	    " intin 0x%x\n", ioapicindex, intin));
1826 }
1827 
1828 void
1829 apix_intx_free(int irqno)
1830 {
1831 	apic_irq_t *irqp;
1832 
1833 	mutex_enter(&airq_mutex);
1834 	irqp = apic_irq_table[irqno];
1835 
1836 	if (IS_IRQ_FREE(irqp)) {
1837 		mutex_exit(&airq_mutex);
1838 		return;
1839 	}
1840 
1841 	irqp->airq_mps_intr_index = FREE_INDEX;
1842 	irqp->airq_cpu = IRQ_UNINIT;
1843 	irqp->airq_vector = APIX_INVALID_VECT;
1844 	mutex_exit(&airq_mutex);
1845 }
1846 
1847 #ifdef DEBUG
1848 int apix_intr_deliver_timeouts = 0;
1849 int apix_intr_rirr_timeouts = 0;
1850 int apix_intr_rirr_reset_failure = 0;
1851 #endif
1852 int apix_max_reps_irr_pending = 10;
1853 
1854 #define	GET_RDT_BITS(ioapic, intin, bits)	\
1855 	(READ_IOAPIC_RDT_ENTRY_LOW_DWORD((ioapic), (intin)) & (bits))
1856 #define	APIX_CHECK_IRR_DELAY	drv_usectohz(5000)
1857 
1858 int
1859 apix_intx_rebind(int irqno, processorid_t cpuid, uchar_t vector)
1860 {
1861 	apic_irq_t *irqp = apic_irq_table[irqno];
1862 	ulong_t iflag;
1863 	int waited, ioapic_ix, intin_no, level, repeats, rdt_entry, masked;
1864 
1865 	ASSERT(irqp != NULL);
1866 
1867 	iflag = intr_clear();
1868 	lock_set(&apic_ioapic_lock);
1869 
1870 	ioapic_ix = irqp->airq_ioapicindex;
1871 	intin_no = irqp->airq_intin_no;
1872 	level = apic_level_intr[irqno];
1873 
1874 	/*
1875 	 * Wait for the delivery status bit to be cleared. This should
1876 	 * be a very small amount of time.
1877 	 */
1878 	repeats = 0;
1879 	do {
1880 		repeats++;
1881 
1882 		for (waited = 0; waited < apic_max_reps_clear_pending;
1883 		    waited++) {
1884 			if (GET_RDT_BITS(ioapic_ix, intin_no, AV_PENDING) == 0)
1885 				break;
1886 		}
1887 		if (!level)
1888 			break;
1889 
1890 		/*
1891 		 * Mask the RDT entry for level-triggered interrupts.
1892 		 */
1893 		irqp->airq_rdt_entry |= AV_MASK;
1894 		rdt_entry = READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
1895 		    intin_no);
1896 		if ((masked = (rdt_entry & AV_MASK)) == 0) {
1897 			/* Mask it */
1898 			WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix, intin_no,
1899 			    AV_MASK | rdt_entry);
1900 		}
1901 
1902 		/*
1903 		 * If there was a race and an interrupt was injected
1904 		 * just before we masked, check for that case here.
1905 		 * Then, unmask the RDT entry and try again.  If we're
1906 		 * on our last try, don't unmask (because we want the
1907 		 * RDT entry to remain masked for the rest of the
1908 		 * function).
1909 		 */
1910 		rdt_entry = READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
1911 		    intin_no);
1912 		if ((masked == 0) && ((rdt_entry & AV_PENDING) != 0) &&
1913 		    (repeats < apic_max_reps_clear_pending)) {
1914 			/* Unmask it */
1915 			WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
1916 			    intin_no, rdt_entry & ~AV_MASK);
1917 			irqp->airq_rdt_entry &= ~AV_MASK;
1918 		}
1919 	} while ((rdt_entry & AV_PENDING) &&
1920 	    (repeats < apic_max_reps_clear_pending));
1921 
1922 #ifdef DEBUG
1923 	if (GET_RDT_BITS(ioapic_ix, intin_no, AV_PENDING) != 0)
1924 		apix_intr_deliver_timeouts++;
1925 #endif
1926 
1927 	if (!level || !APIX_IS_MASK_RDT(apix_mul_ioapic_method))
1928 		goto done;
1929 
1930 	/*
1931 	 * wait for remote IRR to be cleared for level-triggered
1932 	 * interrupts
1933 	 */
1934 	repeats = 0;
1935 	do {
1936 		repeats++;
1937 
1938 		for (waited = 0; waited < apic_max_reps_clear_pending;
1939 		    waited++) {
1940 			if (GET_RDT_BITS(ioapic_ix, intin_no, AV_REMOTE_IRR)
1941 			    == 0)
1942 				break;
1943 		}
1944 
1945 		if (GET_RDT_BITS(ioapic_ix, intin_no, AV_REMOTE_IRR) != 0) {
1946 			lock_clear(&apic_ioapic_lock);
1947 			intr_restore(iflag);
1948 
1949 			delay(APIX_CHECK_IRR_DELAY);
1950 
1951 			iflag = intr_clear();
1952 			lock_set(&apic_ioapic_lock);
1953 		}
1954 	} while (repeats < apix_max_reps_irr_pending);
1955 
1956 	if (repeats >= apix_max_reps_irr_pending) {
1957 #ifdef DEBUG
1958 		apix_intr_rirr_timeouts++;
1959 #endif
1960 
1961 		/*
1962 		 * If we waited and the Remote IRR bit is still not cleared,
1963 		 * AND if we've invoked the timeout APIC_REPROGRAM_MAX_TIMEOUTS
1964 		 * times for this interrupt, try the last-ditch workaround:
1965 		 */
1966 		if (GET_RDT_BITS(ioapic_ix, intin_no, AV_REMOTE_IRR) != 0) {
1967 			/*
1968 			 * Trying to clear the bit through normal
1969 			 * channels has failed.  So as a last-ditch
1970 			 * effort, try to set the trigger mode to
1971 			 * edge, then to level.  This has been
1972 			 * observed to work on many systems.
1973 			 */
1974 			WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
1975 			    intin_no,
1976 			    READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
1977 			    intin_no) & ~AV_LEVEL);
1978 			WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
1979 			    intin_no,
1980 			    READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
1981 			    intin_no) | AV_LEVEL);
1982 		}
1983 
1984 		if (GET_RDT_BITS(ioapic_ix, intin_no, AV_REMOTE_IRR) != 0) {
1985 #ifdef DEBUG
1986 			apix_intr_rirr_reset_failure++;
1987 #endif
1988 			lock_clear(&apic_ioapic_lock);
1989 			intr_restore(iflag);
1990 			prom_printf("apix: Remote IRR still "
1991 			    "not clear for IOAPIC %d intin %d.\n"
1992 			    "\tInterrupts to this pin may cease "
1993 			    "functioning.\n", ioapic_ix, intin_no);
1994 			return (1);	/* return failure */
1995 		}
1996 	}
1997 
1998 done:
1999 	/* change apic_irq_table */
2000 	lock_clear(&apic_ioapic_lock);
2001 	intr_restore(iflag);
2002 	apix_intx_set_vector(irqno, cpuid, vector);
2003 	iflag = intr_clear();
2004 	lock_set(&apic_ioapic_lock);
2005 
2006 	/* reprogramme IO-APIC RDT entry */
2007 	apix_intx_enable(irqno);
2008 
2009 	lock_clear(&apic_ioapic_lock);
2010 	intr_restore(iflag);
2011 
2012 	return (0);
2013 }
2014 
2015 static int
2016 apix_intx_get_pending(int irqno)
2017 {
2018 	apic_irq_t *irqp;
2019 	int intin, ioapicindex, pending;
2020 	ulong_t iflag;
2021 
2022 	mutex_enter(&airq_mutex);
2023 	irqp = apic_irq_table[irqno];
2024 	if (IS_IRQ_FREE(irqp)) {
2025 		mutex_exit(&airq_mutex);
2026 		return (0);
2027 	}
2028 
2029 	/* check IO-APIC delivery status */
2030 	intin = irqp->airq_intin_no;
2031 	ioapicindex = irqp->airq_ioapicindex;
2032 	mutex_exit(&airq_mutex);
2033 
2034 	iflag = intr_clear();
2035 	lock_set(&apic_ioapic_lock);
2036 
2037 	pending = (READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapicindex, intin) &
2038 	    AV_PENDING) ? 1 : 0;
2039 
2040 	lock_clear(&apic_ioapic_lock);
2041 	intr_restore(iflag);
2042 
2043 	return (pending);
2044 }
2045 
2046 /*
2047  * This function will mask the interrupt on the I/O APIC
2048  */
2049 static void
2050 apix_intx_set_mask(int irqno)
2051 {
2052 	int intin, ioapixindex, rdt_entry;
2053 	ulong_t iflag;
2054 	apic_irq_t *irqp;
2055 
2056 	mutex_enter(&airq_mutex);
2057 	irqp = apic_irq_table[irqno];
2058 
2059 	ASSERT(irqp->airq_mps_intr_index != FREE_INDEX);
2060 
2061 	intin = irqp->airq_intin_no;
2062 	ioapixindex = irqp->airq_ioapicindex;
2063 	mutex_exit(&airq_mutex);
2064 
2065 	iflag = intr_clear();
2066 	lock_set(&apic_ioapic_lock);
2067 
2068 	rdt_entry = READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapixindex, intin);
2069 
2070 	/* clear mask */
2071 	WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapixindex, intin,
2072 	    (AV_MASK | rdt_entry));
2073 
2074 	lock_clear(&apic_ioapic_lock);
2075 	intr_restore(iflag);
2076 }
2077 
2078 /*
2079  * This function will clear the mask for the interrupt on the I/O APIC
2080  */
2081 static void
2082 apix_intx_clear_mask(int irqno)
2083 {
2084 	int intin, ioapixindex, rdt_entry;
2085 	ulong_t iflag;
2086 	apic_irq_t *irqp;
2087 
2088 	mutex_enter(&airq_mutex);
2089 	irqp = apic_irq_table[irqno];
2090 
2091 	ASSERT(irqp->airq_mps_intr_index != FREE_INDEX);
2092 
2093 	intin = irqp->airq_intin_no;
2094 	ioapixindex = irqp->airq_ioapicindex;
2095 	mutex_exit(&airq_mutex);
2096 
2097 	iflag = intr_clear();
2098 	lock_set(&apic_ioapic_lock);
2099 
2100 	rdt_entry = READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapixindex, intin);
2101 
2102 	/* clear mask */
2103 	WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapixindex, intin,
2104 	    ((~AV_MASK) & rdt_entry));
2105 
2106 	lock_clear(&apic_ioapic_lock);
2107 	intr_restore(iflag);
2108 }
2109 
2110 /*
2111  * For level-triggered interrupt, mask the IRQ line. Mask means
2112  * new interrupts will not be delivered. The interrupt already
2113  * accepted by a local APIC is not affected
2114  */
2115 void
2116 apix_level_intr_pre_eoi(int irq)
2117 {
2118 	apic_irq_t *irqp = apic_irq_table[irq];
2119 	int apic_ix, intin_ix;
2120 
2121 	if (irqp == NULL)
2122 		return;
2123 
2124 	ASSERT(apic_level_intr[irq] == TRIGGER_MODE_LEVEL);
2125 
2126 	lock_set(&apic_ioapic_lock);
2127 
2128 	intin_ix = irqp->airq_intin_no;
2129 	apic_ix = irqp->airq_ioapicindex;
2130 
2131 	if (irqp->airq_cpu != CPU->cpu_id) {
2132 		if (!APIX_IS_MASK_RDT(apix_mul_ioapic_method))
2133 			ioapic_write_eoi(apic_ix, irqp->airq_vector);
2134 		lock_clear(&apic_ioapic_lock);
2135 		return;
2136 	}
2137 
2138 	if (apix_mul_ioapic_method == APIC_MUL_IOAPIC_IOXAPIC) {
2139 		/*
2140 		 * This is a IOxAPIC and there is EOI register:
2141 		 *	Change the vector to reserved unused vector, so that
2142 		 *	the EOI	from Local APIC won't clear the Remote IRR for
2143 		 *	this level trigger interrupt. Instead, we'll manually
2144 		 *	clear it in apix_post_hardint() after ISR handling.
2145 		 */
2146 		WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(apic_ix, intin_ix,
2147 		    (irqp->airq_rdt_entry & (~0xff)) | APIX_RESV_VECTOR);
2148 	} else {
2149 		WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(apic_ix, intin_ix,
2150 		    AV_MASK | irqp->airq_rdt_entry);
2151 	}
2152 
2153 	lock_clear(&apic_ioapic_lock);
2154 }
2155 
2156 /*
2157  * For level-triggered interrupt, unmask the IRQ line
2158  * or restore the original vector number.
2159  */
2160 void
2161 apix_level_intr_post_dispatch(int irq)
2162 {
2163 	apic_irq_t *irqp = apic_irq_table[irq];
2164 	int apic_ix, intin_ix;
2165 
2166 	if (irqp == NULL)
2167 		return;
2168 
2169 	lock_set(&apic_ioapic_lock);
2170 
2171 	intin_ix = irqp->airq_intin_no;
2172 	apic_ix = irqp->airq_ioapicindex;
2173 
2174 	if (APIX_IS_DIRECTED_EOI(apix_mul_ioapic_method)) {
2175 		/*
2176 		 * Already sent EOI back to Local APIC.
2177 		 * Send EOI to IO-APIC
2178 		 */
2179 		ioapic_write_eoi(apic_ix, irqp->airq_vector);
2180 	} else {
2181 		/* clear the mask or restore the vector */
2182 		WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(apic_ix, intin_ix,
2183 		    irqp->airq_rdt_entry);
2184 
2185 		/* send EOI to IOxAPIC */
2186 		if (apix_mul_ioapic_method == APIC_MUL_IOAPIC_IOXAPIC)
2187 			ioapic_write_eoi(apic_ix, irqp->airq_vector);
2188 	}
2189 
2190 	lock_clear(&apic_ioapic_lock);
2191 }
2192 
2193 static int
2194 apix_intx_get_shared(int irqno)
2195 {
2196 	apic_irq_t *irqp;
2197 	int share;
2198 
2199 	mutex_enter(&airq_mutex);
2200 	irqp = apic_irq_table[irqno];
2201 	if (IS_IRQ_FREE(irqp) || (irqp->airq_cpu == IRQ_UNINIT)) {
2202 		mutex_exit(&airq_mutex);
2203 		return (0);
2204 	}
2205 	share = irqp->airq_share;
2206 	mutex_exit(&airq_mutex);
2207 
2208 	return (share);
2209 }
2210 
2211 static void
2212 apix_intx_set_shared(int irqno, int delta)
2213 {
2214 	apic_irq_t *irqp;
2215 
2216 	mutex_enter(&airq_mutex);
2217 	irqp = apic_irq_table[irqno];
2218 	if (IS_IRQ_FREE(irqp)) {
2219 		mutex_exit(&airq_mutex);
2220 		return;
2221 	}
2222 	irqp->airq_share += delta;
2223 	mutex_exit(&airq_mutex);
2224 }
2225 
2226 /*
2227  * Setup IRQ table. Return IRQ no or -1 on failure
2228  */
2229 static int
2230 apix_intx_setup(dev_info_t *dip, int inum, int irqno,
2231     struct apic_io_intr *intrp, struct intrspec *ispec, iflag_t *iflagp)
2232 {
2233 	int origirq = ispec->intrspec_vec;
2234 	int newirq;
2235 	short intr_index;
2236 	uchar_t ipin, ioapic, ioapicindex;
2237 	apic_irq_t *irqp;
2238 
2239 	UNREFERENCED_1PARAMETER(inum);
2240 
2241 	if (intrp != NULL) {
2242 		intr_index = (short)(intrp - apic_io_intrp);
2243 		ioapic = intrp->intr_destid;
2244 		ipin = intrp->intr_destintin;
2245 
2246 		/* Find ioapicindex. If destid was ALL, we will exit with 0. */
2247 		for (ioapicindex = apic_io_max - 1; ioapicindex; ioapicindex--)
2248 			if (apic_io_id[ioapicindex] == ioapic)
2249 				break;
2250 		ASSERT((ioapic == apic_io_id[ioapicindex]) ||
2251 		    (ioapic == INTR_ALL_APIC));
2252 
2253 		/* check whether this intin# has been used by another irqno */
2254 		if ((newirq = apic_find_intin(ioapicindex, ipin)) != -1)
2255 			return (newirq);
2256 
2257 	} else if (iflagp != NULL) {	/* ACPI */
2258 		intr_index = ACPI_INDEX;
2259 		ioapicindex = acpi_find_ioapic(irqno);
2260 		ASSERT(ioapicindex != 0xFF);
2261 		ioapic = apic_io_id[ioapicindex];
2262 		ipin = irqno - apic_io_vectbase[ioapicindex];
2263 
2264 		if (apic_irq_table[irqno] &&
2265 		    apic_irq_table[irqno]->airq_mps_intr_index == ACPI_INDEX) {
2266 			ASSERT(apic_irq_table[irqno]->airq_intin_no == ipin &&
2267 			    apic_irq_table[irqno]->airq_ioapicindex ==
2268 			    ioapicindex);
2269 			return (irqno);
2270 		}
2271 
2272 	} else {	/* default configuration */
2273 		intr_index = DEFAULT_INDEX;
2274 		ioapicindex = 0;
2275 		ioapic = apic_io_id[ioapicindex];
2276 		ipin = (uchar_t)irqno;
2277 	}
2278 
2279 	/* allocate a new IRQ no */
2280 	if ((irqp = apic_irq_table[irqno]) == NULL) {
2281 		irqp = kmem_zalloc(sizeof (apic_irq_t), KM_SLEEP);
2282 		apic_irq_table[irqno] = irqp;
2283 	} else {
2284 		if (irqp->airq_mps_intr_index != FREE_INDEX) {
2285 			newirq = apic_allocate_irq(apic_first_avail_irq);
2286 			if (newirq == -1) {
2287 				return (-1);
2288 			}
2289 			irqno = newirq;
2290 			irqp = apic_irq_table[irqno];
2291 			ASSERT(irqp != NULL);
2292 		}
2293 	}
2294 	apic_max_device_irq = max(irqno, apic_max_device_irq);
2295 	apic_min_device_irq = min(irqno, apic_min_device_irq);
2296 
2297 	irqp->airq_mps_intr_index = intr_index;
2298 	irqp->airq_ioapicindex = ioapicindex;
2299 	irqp->airq_intin_no = ipin;
2300 	irqp->airq_dip = dip;
2301 	irqp->airq_origirq = (uchar_t)origirq;
2302 	if (iflagp != NULL)
2303 		irqp->airq_iflag = *iflagp;
2304 	irqp->airq_cpu = IRQ_UNINIT;
2305 	irqp->airq_vector = 0;
2306 
2307 	return (irqno);
2308 }
2309 
2310 /*
2311  * Setup IRQ table for non-pci devices. Return IRQ no or -1 on error
2312  */
2313 static int
2314 apix_intx_setup_nonpci(dev_info_t *dip, int inum, int bustype,
2315     struct intrspec *ispec)
2316 {
2317 	int irqno = ispec->intrspec_vec;
2318 	int newirq, i;
2319 	iflag_t intr_flag;
2320 	ACPI_SUBTABLE_HEADER	*hp;
2321 	ACPI_MADT_INTERRUPT_OVERRIDE *isop;
2322 	struct apic_io_intr *intrp;
2323 
2324 	if (!apic_enable_acpi || apic_use_acpi_madt_only) {
2325 		int busid;
2326 
2327 		if (bustype == 0)
2328 			bustype = eisa_level_intr_mask ? BUS_EISA : BUS_ISA;
2329 
2330 		/* loop checking BUS_ISA/BUS_EISA */
2331 		for (i = 0; i < 2; i++) {
2332 			if (((busid = apic_find_bus_id(bustype)) != -1) &&
2333 			    ((intrp = apic_find_io_intr_w_busid(irqno, busid))
2334 			    != NULL)) {
2335 				return (apix_intx_setup(dip, inum, irqno,
2336 				    intrp, ispec, NULL));
2337 			}
2338 			bustype = (bustype == BUS_EISA) ? BUS_ISA : BUS_EISA;
2339 		}
2340 
2341 		/* fall back to default configuration */
2342 		return (-1);
2343 	}
2344 
2345 	/* search iso entries first */
2346 	if (acpi_iso_cnt != 0) {
2347 		hp = (ACPI_SUBTABLE_HEADER *)acpi_isop;
2348 		i = 0;
2349 		while (i < acpi_iso_cnt) {
2350 			if (hp->Type == ACPI_MADT_TYPE_INTERRUPT_OVERRIDE) {
2351 				isop = (ACPI_MADT_INTERRUPT_OVERRIDE *) hp;
2352 				if (isop->Bus == 0 &&
2353 				    isop->SourceIrq == irqno) {
2354 					newirq = isop->GlobalIrq;
2355 					intr_flag.intr_po = isop->IntiFlags &
2356 					    ACPI_MADT_POLARITY_MASK;
2357 					intr_flag.intr_el = (isop->IntiFlags &
2358 					    ACPI_MADT_TRIGGER_MASK) >> 2;
2359 					intr_flag.bustype = BUS_ISA;
2360 
2361 					return (apix_intx_setup(dip, inum,
2362 					    newirq, NULL, ispec, &intr_flag));
2363 				}
2364 				i++;
2365 			}
2366 			hp = (ACPI_SUBTABLE_HEADER *)(((char *)hp) +
2367 			    hp->Length);
2368 		}
2369 	}
2370 	intr_flag.intr_po = INTR_PO_ACTIVE_HIGH;
2371 	intr_flag.intr_el = INTR_EL_EDGE;
2372 	intr_flag.bustype = BUS_ISA;
2373 	return (apix_intx_setup(dip, inum, irqno, NULL, ispec, &intr_flag));
2374 }
2375 
2376 
2377 /*
2378  * Setup IRQ table for pci devices. Return IRQ no or -1 on error
2379  */
2380 static int
2381 apix_intx_setup_pci(dev_info_t *dip, int inum, int bustype,
2382     struct intrspec *ispec)
2383 {
2384 	int busid, devid, pci_irq;
2385 	ddi_acc_handle_t cfg_handle;
2386 	uchar_t ipin;
2387 	iflag_t intr_flag;
2388 	struct apic_io_intr *intrp;
2389 
2390 	if (acpica_get_bdf(dip, &busid, &devid, NULL) != 0)
2391 		return (-1);
2392 
2393 	if (busid == 0 && apic_pci_bus_total == 1)
2394 		busid = (int)apic_single_pci_busid;
2395 
2396 	if (pci_config_setup(dip, &cfg_handle) != DDI_SUCCESS)
2397 		return (-1);
2398 	ipin = pci_config_get8(cfg_handle, PCI_CONF_IPIN) - PCI_INTA;
2399 	pci_config_teardown(&cfg_handle);
2400 
2401 	if (apic_enable_acpi && !apic_use_acpi_madt_only) {	/* ACPI */
2402 		if (apic_acpi_translate_pci_irq(dip, busid, devid,
2403 		    ipin, &pci_irq, &intr_flag) != ACPI_PSM_SUCCESS)
2404 			return (-1);
2405 
2406 		intr_flag.bustype = (uchar_t)bustype;
2407 		return (apix_intx_setup(dip, inum, pci_irq, NULL, ispec,
2408 		    &intr_flag));
2409 	}
2410 
2411 	/* MP configuration table */
2412 	pci_irq = ((devid & 0x1f) << 2) | (ipin & 0x3);
2413 	if ((intrp = apic_find_io_intr_w_busid(pci_irq, busid)) == NULL) {
2414 		pci_irq = apic_handle_pci_pci_bridge(dip, devid, ipin, &intrp);
2415 		if (pci_irq == -1)
2416 			return (-1);
2417 	}
2418 
2419 	return (apix_intx_setup(dip, inum, pci_irq, intrp, ispec, NULL));
2420 }
2421 
2422 /*
2423  * Translate and return IRQ no
2424  */
2425 static int
2426 apix_intx_xlate_irq(dev_info_t *dip, int inum, struct intrspec *ispec)
2427 {
2428 	int newirq, irqno = ispec->intrspec_vec;
2429 	int parent_is_pci_or_pciex = 0, child_is_pciex = 0;
2430 	int bustype = 0, dev_len;
2431 	char dev_type[16];
2432 
2433 	if (apic_defconf) {
2434 		mutex_enter(&airq_mutex);
2435 		goto defconf;
2436 	}
2437 
2438 	if ((dip == NULL) || (!apic_irq_translate && !apic_enable_acpi)) {
2439 		mutex_enter(&airq_mutex);
2440 		goto nonpci;
2441 	}
2442 
2443 	/*
2444 	 * use ddi_getlongprop_buf() instead of ddi_prop_lookup_string()
2445 	 * to avoid extra buffer allocation.
2446 	 */
2447 	dev_len = sizeof (dev_type);
2448 	if (ddi_getlongprop_buf(DDI_DEV_T_ANY, ddi_get_parent(dip),
2449 	    DDI_PROP_DONTPASS, "device_type", (caddr_t)dev_type,
2450 	    &dev_len) == DDI_PROP_SUCCESS) {
2451 		if ((strcmp(dev_type, "pci") == 0) ||
2452 		    (strcmp(dev_type, "pciex") == 0))
2453 			parent_is_pci_or_pciex = 1;
2454 	}
2455 
2456 	if (ddi_getlongprop_buf(DDI_DEV_T_ANY, dip,
2457 	    DDI_PROP_DONTPASS, "compatible", (caddr_t)dev_type,
2458 	    &dev_len) == DDI_PROP_SUCCESS) {
2459 		if (strstr(dev_type, "pciex"))
2460 			child_is_pciex = 1;
2461 	}
2462 
2463 	mutex_enter(&airq_mutex);
2464 
2465 	if (parent_is_pci_or_pciex) {
2466 		bustype = child_is_pciex ? BUS_PCIE : BUS_PCI;
2467 		newirq = apix_intx_setup_pci(dip, inum, bustype, ispec);
2468 		if (newirq != -1)
2469 			goto done;
2470 		bustype = 0;
2471 	} else if (strcmp(dev_type, "isa") == 0)
2472 		bustype = BUS_ISA;
2473 	else if (strcmp(dev_type, "eisa") == 0)
2474 		bustype = BUS_EISA;
2475 
2476 nonpci:
2477 	newirq = apix_intx_setup_nonpci(dip, inum, bustype, ispec);
2478 	if (newirq != -1)
2479 		goto done;
2480 
2481 defconf:
2482 	newirq = apix_intx_setup(dip, inum, irqno, NULL, ispec, NULL);
2483 	if (newirq == -1) {
2484 		mutex_exit(&airq_mutex);
2485 		return (-1);
2486 	}
2487 done:
2488 	ASSERT(apic_irq_table[newirq]);
2489 	mutex_exit(&airq_mutex);
2490 	return (newirq);
2491 }
2492 
2493 static int
2494 apix_intx_alloc_vector(dev_info_t *dip, int inum, struct intrspec *ispec)
2495 {
2496 	int irqno;
2497 	apix_vector_t *vecp;
2498 
2499 	if ((irqno = apix_intx_xlate_irq(dip, inum, ispec)) == -1)
2500 		return (0);
2501 
2502 	if ((vecp = apix_alloc_intx(dip, inum, irqno)) == NULL)
2503 		return (0);
2504 
2505 	DDI_INTR_IMPLDBG((CE_CONT, "apix_intx_alloc_vector: dip=0x%p name=%s "
2506 	    "irqno=0x%x cpuid=%d vector=0x%x\n",
2507 	    (void *)dip, ddi_driver_name(dip), irqno,
2508 	    vecp->v_cpuid, vecp->v_vector));
2509 
2510 	return (1);
2511 }
2512 
2513 /*
2514  * Return the vector number if the translated IRQ for this device
2515  * has a vector mapping setup. If no IRQ setup exists or no vector is
2516  * allocated to it then return 0.
2517  */
2518 static apix_vector_t *
2519 apix_intx_xlate_vector(dev_info_t *dip, int inum, struct intrspec *ispec)
2520 {
2521 	int irqno;
2522 	apix_vector_t *vecp;
2523 
2524 	/* get the IRQ number */
2525 	if ((irqno = apix_intx_xlate_irq(dip, inum, ispec)) == -1)
2526 		return (NULL);
2527 
2528 	/* get the vector number if a vector is allocated to this irqno */
2529 	vecp = apix_intx_get_vector(irqno);
2530 
2531 	return (vecp);
2532 }
2533 
2534 /*
2535  * Switch between safe and x2APIC IPI sending method.
2536  * The CPU may power on in xapic mode or x2apic mode. If the CPU needs to send
2537  * an IPI to other CPUs before entering x2APIC mode, it still needs to use the
2538  * xAPIC method. Before sending a StartIPI to the target CPU, psm_send_ipi will
2539  * be changed to apic_common_send_ipi, which detects current local APIC mode and
2540  * use the right method to send an IPI. If some CPUs fail to start up,
2541  * apic_poweron_cnt won't return to zero, so apic_common_send_ipi will always be
2542  * used. psm_send_ipi can't be simply changed back to x2apic_send_ipi if some
2543  * CPUs failed to start up because those failed CPUs may recover itself later at
2544  * unpredictable time.
2545  */
2546 void
2547 apic_switch_ipi_callback(boolean_t enter)
2548 {
2549 	ulong_t iflag;
2550 	struct psm_ops *pops = psmops;
2551 
2552 	iflag = intr_clear();
2553 	lock_set(&apic_mode_switch_lock);
2554 	if (enter) {
2555 		ASSERT(apic_poweron_cnt >= 0);
2556 		if (apic_poweron_cnt == 0) {
2557 			pops->psm_send_ipi = apic_common_send_ipi;
2558 			send_dirintf = pops->psm_send_ipi;
2559 			pops->psm_send_pir_ipi = apic_common_send_pir_ipi;
2560 			psm_send_pir_ipi = pops->psm_send_pir_ipi;
2561 		}
2562 		apic_poweron_cnt++;
2563 	} else {
2564 		ASSERT(apic_poweron_cnt > 0);
2565 		apic_poweron_cnt--;
2566 		if (apic_poweron_cnt == 0) {
2567 			pops->psm_send_ipi = x2apic_send_ipi;
2568 			send_dirintf = pops->psm_send_ipi;
2569 			pops->psm_send_pir_ipi = x2apic_send_pir_ipi;
2570 			psm_send_pir_ipi = pops->psm_send_pir_ipi;
2571 		}
2572 	}
2573 	lock_clear(&apic_mode_switch_lock);
2574 	intr_restore(iflag);
2575 }
2576 
2577 /* stub function */
2578 int
2579 apix_loaded(void)
2580 {
2581 	return (apix_is_enabled);
2582 }
2583