xref: /titanic_52/usr/src/uts/i86pc/io/apix/apix.c (revision b0f673c4626e4cb1db7785287eaeed2731dfefe8)
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 	/*
375 	 * Make cpu-specific interrupt info point to cr8pri vector
376 	 */
377 	CPU->cpu_pri_data = dummy_cpu_pri;
378 #else
379 	if (cpuid_have_cr8access(CPU))
380 		apic_have_32bit_cr8 = 1;
381 #endif	/* __amd64 */
382 
383 	/*
384 	 * Initialize IRM pool parameters
385 	 */
386 	if (irm_enable) {
387 		int	i;
388 		int	lowest_irq;
389 		int	highest_irq;
390 
391 		/* number of CPUs present */
392 		apix_irminfo.apix_ncpus = apic_nproc;
393 		/* total number of entries in all of the IOAPICs present */
394 		lowest_irq = apic_io_vectbase[0];
395 		highest_irq = apic_io_vectend[0];
396 		for (i = 1; i < apic_io_max; i++) {
397 			if (apic_io_vectbase[i] < lowest_irq)
398 				lowest_irq = apic_io_vectbase[i];
399 			if (apic_io_vectend[i] > highest_irq)
400 				highest_irq = apic_io_vectend[i];
401 		}
402 		apix_irminfo.apix_ioapic_max_vectors =
403 		    highest_irq - lowest_irq + 1;
404 		/*
405 		 * Number of available per-CPU vectors excluding
406 		 * reserved vectors for Dtrace, int80, system-call,
407 		 * fast-trap, etc.
408 		 */
409 		apix_irminfo.apix_per_cpu_vectors = APIX_NAVINTR -
410 		    APIX_SW_RESERVED_VECTORS;
411 
412 		/* Number of vectors (pre) allocated (SCI and HPET) */
413 		apix_irminfo.apix_vectors_allocated = 0;
414 		if (apic_hpet_vect != -1)
415 			apix_irminfo.apix_vectors_allocated++;
416 		if (apic_sci_vect != -1)
417 			apix_irminfo.apix_vectors_allocated++;
418 	}
419 }
420 
421 static void
422 apix_init_intr()
423 {
424 	processorid_t	cpun = psm_get_cpu_id();
425 	uint_t nlvt;
426 	uint32_t svr = AV_UNIT_ENABLE | APIC_SPUR_INTR;
427 	extern void cmi_cmci_trap(void);
428 
429 	apic_reg_ops->apic_write_task_reg(APIC_MASK_ALL);
430 
431 	if (apic_mode == LOCAL_APIC) {
432 		/*
433 		 * We are running APIC in MMIO mode.
434 		 */
435 		if (apic_flat_model) {
436 			apic_reg_ops->apic_write(APIC_FORMAT_REG,
437 			    APIC_FLAT_MODEL);
438 		} else {
439 			apic_reg_ops->apic_write(APIC_FORMAT_REG,
440 			    APIC_CLUSTER_MODEL);
441 		}
442 
443 		apic_reg_ops->apic_write(APIC_DEST_REG,
444 		    AV_HIGH_ORDER >> cpun);
445 	}
446 
447 	if (apic_directed_EOI_supported()) {
448 		/*
449 		 * Setting the 12th bit in the Spurious Interrupt Vector
450 		 * Register suppresses broadcast EOIs generated by the local
451 		 * APIC. The suppression of broadcast EOIs happens only when
452 		 * interrupts are level-triggered.
453 		 */
454 		svr |= APIC_SVR_SUPPRESS_BROADCAST_EOI;
455 	}
456 
457 	/* need to enable APIC before unmasking NMI */
458 	apic_reg_ops->apic_write(APIC_SPUR_INT_REG, svr);
459 
460 	/*
461 	 * Presence of an invalid vector with delivery mode AV_FIXED can
462 	 * cause an error interrupt, even if the entry is masked...so
463 	 * write a valid vector to LVT entries along with the mask bit
464 	 */
465 
466 	/* All APICs have timer and LINT0/1 */
467 	apic_reg_ops->apic_write(APIC_LOCAL_TIMER, AV_MASK|APIC_RESV_IRQ);
468 	apic_reg_ops->apic_write(APIC_INT_VECT0, AV_MASK|APIC_RESV_IRQ);
469 	apic_reg_ops->apic_write(APIC_INT_VECT1, AV_NMI);	/* enable NMI */
470 
471 	/*
472 	 * On integrated APICs, the number of LVT entries is
473 	 * 'Max LVT entry' + 1; on 82489DX's (non-integrated
474 	 * APICs), nlvt is "3" (LINT0, LINT1, and timer)
475 	 */
476 
477 	if (apic_cpus[cpun].aci_local_ver < APIC_INTEGRATED_VERS) {
478 		nlvt = 3;
479 	} else {
480 		nlvt = ((apic_reg_ops->apic_read(APIC_VERS_REG) >> 16) &
481 		    0xFF) + 1;
482 	}
483 
484 	if (nlvt >= 5) {
485 		/* Enable performance counter overflow interrupt */
486 
487 		if (!is_x86_feature(x86_featureset, X86FSET_MSR))
488 			apic_enable_cpcovf_intr = 0;
489 		if (apic_enable_cpcovf_intr) {
490 			if (apic_cpcovf_vect == 0) {
491 				int ipl = APIC_PCINT_IPL;
492 
493 				apic_cpcovf_vect = apix_get_ipivect(ipl, -1);
494 				ASSERT(apic_cpcovf_vect);
495 
496 				(void) add_avintr(NULL, ipl,
497 				    (avfunc)kcpc_hw_overflow_intr,
498 				    "apic pcint", apic_cpcovf_vect,
499 				    NULL, NULL, NULL, NULL);
500 				kcpc_hw_overflow_intr_installed = 1;
501 				kcpc_hw_enable_cpc_intr =
502 				    apic_cpcovf_mask_clear;
503 			}
504 			apic_reg_ops->apic_write(APIC_PCINT_VECT,
505 			    apic_cpcovf_vect);
506 		}
507 	}
508 
509 	if (nlvt >= 6) {
510 		/* Only mask TM intr if the BIOS apparently doesn't use it */
511 
512 		uint32_t lvtval;
513 
514 		lvtval = apic_reg_ops->apic_read(APIC_THERM_VECT);
515 		if (((lvtval & AV_MASK) == AV_MASK) ||
516 		    ((lvtval & AV_DELIV_MODE) != AV_SMI)) {
517 			apic_reg_ops->apic_write(APIC_THERM_VECT,
518 			    AV_MASK|APIC_RESV_IRQ);
519 		}
520 	}
521 
522 	/* Enable error interrupt */
523 
524 	if (nlvt >= 4 && apic_enable_error_intr) {
525 		if (apic_errvect == 0) {
526 			int ipl = 0xf;	/* get highest priority intr */
527 			apic_errvect = apix_get_ipivect(ipl, -1);
528 			ASSERT(apic_errvect);
529 			/*
530 			 * Not PSMI compliant, but we are going to merge
531 			 * with ON anyway
532 			 */
533 			(void) add_avintr(NULL, ipl,
534 			    (avfunc)apic_error_intr, "apic error intr",
535 			    apic_errvect, NULL, NULL, NULL, NULL);
536 		}
537 		apic_reg_ops->apic_write(APIC_ERR_VECT, apic_errvect);
538 		apic_reg_ops->apic_write(APIC_ERROR_STATUS, 0);
539 		apic_reg_ops->apic_write(APIC_ERROR_STATUS, 0);
540 	}
541 
542 	/* Enable CMCI interrupt */
543 	if (cmi_enable_cmci) {
544 		mutex_enter(&cmci_cpu_setup_lock);
545 		if (cmci_cpu_setup_registered == 0) {
546 			mutex_enter(&cpu_lock);
547 			register_cpu_setup_func(cmci_cpu_setup, NULL);
548 			mutex_exit(&cpu_lock);
549 			cmci_cpu_setup_registered = 1;
550 		}
551 		mutex_exit(&cmci_cpu_setup_lock);
552 
553 		if (apic_cmci_vect == 0) {
554 			int ipl = 0x2;
555 			apic_cmci_vect = apix_get_ipivect(ipl, -1);
556 			ASSERT(apic_cmci_vect);
557 
558 			(void) add_avintr(NULL, ipl,
559 			    (avfunc)cmi_cmci_trap, "apic cmci intr",
560 			    apic_cmci_vect, NULL, NULL, NULL, NULL);
561 		}
562 		apic_reg_ops->apic_write(APIC_CMCI_VECT, apic_cmci_vect);
563 	}
564 
565 	apic_reg_ops->apic_write_task_reg(0);
566 }
567 
568 static void
569 apix_picinit(void)
570 {
571 	int i, j;
572 	uint_t isr;
573 
574 	APIC_VERBOSE(INIT, (CE_CONT, "apix: psm_picinit\n"));
575 
576 	/*
577 	 * initialize interrupt remapping before apic
578 	 * hardware initialization
579 	 */
580 	apic_intrmap_init(apic_mode);
581 	if (apic_vt_ops == psm_vt_ops)
582 		apix_mul_ioapic_method = APIC_MUL_IOAPIC_IIR;
583 
584 	/*
585 	 * On UniSys Model 6520, the BIOS leaves vector 0x20 isr
586 	 * bit on without clearing it with EOI.  Since softint
587 	 * uses vector 0x20 to interrupt itself, so softint will
588 	 * not work on this machine.  In order to fix this problem
589 	 * a check is made to verify all the isr bits are clear.
590 	 * If not, EOIs are issued to clear the bits.
591 	 */
592 	for (i = 7; i >= 1; i--) {
593 		isr = apic_reg_ops->apic_read(APIC_ISR_REG + (i * 4));
594 		if (isr != 0)
595 			for (j = 0; ((j < 32) && (isr != 0)); j++)
596 				if (isr & (1 << j)) {
597 					apic_reg_ops->apic_write(
598 					    APIC_EOI_REG, 0);
599 					isr &= ~(1 << j);
600 					apic_error |= APIC_ERR_BOOT_EOI;
601 				}
602 	}
603 
604 	/* set a flag so we know we have run apic_picinit() */
605 	apic_picinit_called = 1;
606 	LOCK_INIT_CLEAR(&apic_gethrtime_lock);
607 	LOCK_INIT_CLEAR(&apic_ioapic_lock);
608 	LOCK_INIT_CLEAR(&apic_error_lock);
609 	LOCK_INIT_CLEAR(&apic_mode_switch_lock);
610 
611 	picsetup();	 /* initialise the 8259 */
612 
613 	/* add nmi handler - least priority nmi handler */
614 	LOCK_INIT_CLEAR(&apic_nmi_lock);
615 
616 	if (!psm_add_nmintr(0, (avfunc) apic_nmi_intr,
617 	    "apix NMI handler", (caddr_t)NULL))
618 		cmn_err(CE_WARN, "apix: Unable to add nmi handler");
619 
620 	apix_init_intr();
621 
622 	/* enable apic mode if imcr present */
623 	if (apic_imcrp) {
624 		outb(APIC_IMCR_P1, (uchar_t)APIC_IMCR_SELECT);
625 		outb(APIC_IMCR_P2, (uchar_t)APIC_IMCR_APIC);
626 	}
627 
628 	ioapix_init_intr(IOAPIC_MASK);
629 
630 	/* setup global IRM pool if applicable */
631 	if (irm_enable)
632 		apix_irm_init();
633 }
634 
635 static __inline__ void
636 apix_send_eoi(void)
637 {
638 	if (apic_mode == LOCAL_APIC)
639 		LOCAL_APIC_WRITE_REG(APIC_EOI_REG, 0);
640 	else
641 		X2APIC_WRITE(APIC_EOI_REG, 0);
642 }
643 
644 /*
645  * platform_intr_enter
646  *
647  *	Called at the beginning of the interrupt service routine, but unlike
648  *	pcplusmp, does not mask interrupts. An EOI is given to the interrupt
649  *	controller to enable other HW interrupts but interrupts are still
650  * 	masked by the IF flag.
651  *
652  *	Return -1 for spurious interrupts
653  *
654  */
655 static int
656 apix_intr_enter(int ipl, int *vectorp)
657 {
658 	struct cpu *cpu = CPU;
659 	uint32_t cpuid = CPU->cpu_id;
660 	apic_cpus_info_t *cpu_infop;
661 	uchar_t vector;
662 	apix_vector_t *vecp;
663 	int nipl = -1;
664 
665 	/*
666 	 * The real vector delivered is (*vectorp + 0x20), but our caller
667 	 * subtracts 0x20 from the vector before passing it to us.
668 	 * (That's why APIC_BASE_VECT is 0x20.)
669 	 */
670 	vector = *vectorp = (uchar_t)*vectorp + APIC_BASE_VECT;
671 
672 	cpu_infop = &apic_cpus[cpuid];
673 	if (vector == APIC_SPUR_INTR) {
674 		cpu_infop->aci_spur_cnt++;
675 		return (APIC_INT_SPURIOUS);
676 	}
677 
678 	vecp = xv_vector(cpuid, vector);
679 	if (vecp == NULL) {
680 		if (APIX_IS_FAKE_INTR(vector))
681 			nipl = apix_rebindinfo.i_pri;
682 		apix_send_eoi();
683 		return (nipl);
684 	}
685 	nipl = vecp->v_pri;
686 
687 	/* if interrupted by the clock, increment apic_nsec_since_boot */
688 	if (vector == (apic_clkvect + APIC_BASE_VECT)) {
689 		if (!apic_oneshot) {
690 			/* NOTE: this is not MT aware */
691 			apic_hrtime_stamp++;
692 			apic_nsec_since_boot += apic_nsec_per_intr;
693 			apic_hrtime_stamp++;
694 			last_count_read = apic_hertz_count;
695 			apix_redistribute_compute();
696 		}
697 
698 		apix_send_eoi();
699 
700 		return (nipl);
701 	}
702 
703 	ASSERT(vecp->v_state != APIX_STATE_OBSOLETED);
704 
705 	/* pre-EOI handling for level-triggered interrupts */
706 	if (!APIX_IS_DIRECTED_EOI(apix_mul_ioapic_method) &&
707 	    (vecp->v_type & APIX_TYPE_FIXED) && apic_level_intr[vecp->v_inum])
708 		apix_level_intr_pre_eoi(vecp->v_inum);
709 
710 	/* send back EOI */
711 	apix_send_eoi();
712 
713 	cpu_infop->aci_current[nipl] = vector;
714 	if ((nipl > ipl) && (nipl > cpu->cpu_base_spl)) {
715 		cpu_infop->aci_curipl = (uchar_t)nipl;
716 		cpu_infop->aci_ISR_in_progress |= 1 << nipl;
717 	}
718 
719 #ifdef	DEBUG
720 	if (vector >= APIX_IPI_MIN)
721 		return (nipl);	/* skip IPI */
722 
723 	APIC_DEBUG_BUF_PUT(vector);
724 	APIC_DEBUG_BUF_PUT(vecp->v_inum);
725 	APIC_DEBUG_BUF_PUT(nipl);
726 	APIC_DEBUG_BUF_PUT(psm_get_cpu_id());
727 	if ((apic_stretch_interrupts) && (apic_stretch_ISR & (1 << nipl)))
728 		drv_usecwait(apic_stretch_interrupts);
729 #endif /* DEBUG */
730 
731 	return (nipl);
732 }
733 
734 /*
735  * Any changes made to this function must also change X2APIC
736  * version of intr_exit.
737  */
738 static void
739 apix_intr_exit(int prev_ipl, int arg2)
740 {
741 	int cpuid = psm_get_cpu_id();
742 	apic_cpus_info_t *cpu_infop = &apic_cpus[cpuid];
743 	apix_impl_t *apixp = apixs[cpuid];
744 
745 	UNREFERENCED_1PARAMETER(arg2);
746 
747 	cpu_infop->aci_curipl = (uchar_t)prev_ipl;
748 	/* ISR above current pri could not be in progress */
749 	cpu_infop->aci_ISR_in_progress &= (2 << prev_ipl) - 1;
750 
751 	if (apixp->x_obsoletes != NULL) {
752 		if (APIX_CPU_LOCK_HELD(cpuid))
753 			return;
754 
755 		APIX_ENTER_CPU_LOCK(cpuid);
756 		(void) apix_obsolete_vector(apixp->x_obsoletes);
757 		APIX_LEAVE_CPU_LOCK(cpuid);
758 	}
759 }
760 
761 /*
762  * The pcplusmp setspl code uses the TPR to mask all interrupts at or below the
763  * given ipl, but apix never uses the TPR and we never mask a subset of the
764  * interrupts. They are either all blocked by the IF flag or all can come in.
765  *
766  * For setspl, we mask all interrupts for XC_HI_PIL (15), otherwise, interrupts
767  * can come in if currently enabled by the IF flag. This table shows the state
768  * of the IF flag when we leave this function.
769  *
770  *    curr IF |	ipl == 15	ipl != 15
771  *    --------+---------------------------
772  *       0    |    0		    0
773  *       1    |    0		    1
774  */
775 static void
776 apix_setspl(int ipl)
777 {
778 	/*
779 	 * Interrupts at ipl above this cannot be in progress, so the following
780 	 * mask is ok.
781 	 */
782 	apic_cpus[psm_get_cpu_id()].aci_ISR_in_progress &= (2 << ipl) - 1;
783 
784 	if (ipl == XC_HI_PIL)
785 		cli();
786 }
787 
788 int
789 apix_addspl(int virtvec, int ipl, int min_ipl, int max_ipl)
790 {
791 	uint32_t cpuid = APIX_VIRTVEC_CPU(virtvec);
792 	uchar_t vector = (uchar_t)APIX_VIRTVEC_VECTOR(virtvec);
793 	apix_vector_t *vecp = xv_vector(cpuid, vector);
794 
795 	UNREFERENCED_3PARAMETER(ipl, min_ipl, max_ipl);
796 	ASSERT(vecp != NULL && LOCK_HELD(&apix_lock));
797 
798 	if (vecp->v_type == APIX_TYPE_FIXED)
799 		apix_intx_set_shared(vecp->v_inum, 1);
800 
801 	/* There are more interrupts, so it's already been enabled */
802 	if (vecp->v_share > 1)
803 		return (PSM_SUCCESS);
804 
805 	/* return if it is not hardware interrupt */
806 	if (vecp->v_type == APIX_TYPE_IPI)
807 		return (PSM_SUCCESS);
808 
809 	/*
810 	 * if apix_picinit() has not been called yet, just return.
811 	 * At the end of apic_picinit(), we will call setup_io_intr().
812 	 */
813 	if (!apic_picinit_called)
814 		return (PSM_SUCCESS);
815 
816 	(void) apix_setup_io_intr(vecp);
817 
818 	return (PSM_SUCCESS);
819 }
820 
821 int
822 apix_delspl(int virtvec, int ipl, int min_ipl, int max_ipl)
823 {
824 	uint32_t cpuid = APIX_VIRTVEC_CPU(virtvec);
825 	uchar_t vector = (uchar_t)APIX_VIRTVEC_VECTOR(virtvec);
826 	apix_vector_t *vecp = xv_vector(cpuid, vector);
827 
828 	UNREFERENCED_3PARAMETER(ipl, min_ipl, max_ipl);
829 	ASSERT(vecp != NULL && LOCK_HELD(&apix_lock));
830 
831 	if (vecp->v_type == APIX_TYPE_FIXED)
832 		apix_intx_set_shared(vecp->v_inum, -1);
833 
834 	/* There are more interrupts */
835 	if (vecp->v_share > 1)
836 		return (PSM_SUCCESS);
837 
838 	/* return if it is not hardware interrupt */
839 	if (vecp->v_type == APIX_TYPE_IPI)
840 		return (PSM_SUCCESS);
841 
842 	if (!apic_picinit_called) {
843 		cmn_err(CE_WARN, "apix: delete 0x%x before apic init",
844 		    virtvec);
845 		return (PSM_SUCCESS);
846 	}
847 
848 	apix_disable_vector(vecp);
849 
850 	return (PSM_SUCCESS);
851 }
852 
853 /*
854  * Try and disable all interrupts. We just assign interrupts to other
855  * processors based on policy. If any were bound by user request, we
856  * let them continue and return failure. We do not bother to check
857  * for cache affinity while rebinding.
858  */
859 static int
860 apix_disable_intr(processorid_t cpun)
861 {
862 	apix_impl_t *apixp = apixs[cpun];
863 	apix_vector_t *vecp, *newp;
864 	int bindcpu, i, hardbound = 0, errbound = 0, ret, loop, type;
865 
866 	lock_set(&apix_lock);
867 
868 	apic_cpus[cpun].aci_status &= ~APIC_CPU_INTR_ENABLE;
869 	apic_cpus[cpun].aci_curipl = 0;
870 
871 	/* if this is for SUSPEND operation, skip rebinding */
872 	if (apic_cpus[cpun].aci_status & APIC_CPU_SUSPEND) {
873 		for (i = APIX_AVINTR_MIN; i <= APIX_AVINTR_MAX; i++) {
874 			vecp = apixp->x_vectbl[i];
875 			if (!IS_VECT_ENABLED(vecp))
876 				continue;
877 
878 			apix_disable_vector(vecp);
879 		}
880 		lock_clear(&apix_lock);
881 		return (PSM_SUCCESS);
882 	}
883 
884 	for (i = APIX_AVINTR_MIN; i <= APIX_AVINTR_MAX; i++) {
885 		vecp = apixp->x_vectbl[i];
886 		if (!IS_VECT_ENABLED(vecp))
887 			continue;
888 
889 		if (vecp->v_flags & APIX_VECT_USER_BOUND) {
890 			hardbound++;
891 			continue;
892 		}
893 		type = vecp->v_type;
894 
895 		/*
896 		 * If there are bound interrupts on this cpu, then
897 		 * rebind them to other processors.
898 		 */
899 		loop = 0;
900 		do {
901 			bindcpu = apic_find_cpu(APIC_CPU_INTR_ENABLE);
902 
903 			if (type != APIX_TYPE_MSI)
904 				newp = apix_set_cpu(vecp, bindcpu, &ret);
905 			else
906 				newp = apix_grp_set_cpu(vecp, bindcpu, &ret);
907 		} while ((newp == NULL) && (loop++ < apic_nproc));
908 
909 		if (loop >= apic_nproc) {
910 			errbound++;
911 			cmn_err(CE_WARN, "apix: failed to rebind vector %x/%x",
912 			    vecp->v_cpuid, vecp->v_vector);
913 		}
914 	}
915 
916 	lock_clear(&apix_lock);
917 
918 	if (hardbound || errbound) {
919 		cmn_err(CE_WARN, "Could not disable interrupts on %d"
920 		    "due to user bound interrupts or failed operation",
921 		    cpun);
922 		return (PSM_FAILURE);
923 	}
924 
925 	return (PSM_SUCCESS);
926 }
927 
928 /*
929  * Bind interrupts to specified CPU
930  */
931 static void
932 apix_enable_intr(processorid_t cpun)
933 {
934 	apix_vector_t *vecp;
935 	int i, ret;
936 	processorid_t n;
937 
938 	lock_set(&apix_lock);
939 
940 	apic_cpus[cpun].aci_status |= APIC_CPU_INTR_ENABLE;
941 
942 	/* interrupt enabling for system resume */
943 	if (apic_cpus[cpun].aci_status & APIC_CPU_SUSPEND) {
944 		for (i = APIX_AVINTR_MIN; i <= APIX_AVINTR_MAX; i++) {
945 			vecp = xv_vector(cpun, i);
946 			if (!IS_VECT_ENABLED(vecp))
947 				continue;
948 
949 			apix_enable_vector(vecp);
950 		}
951 		apic_cpus[cpun].aci_status &= ~APIC_CPU_SUSPEND;
952 	}
953 
954 	for (n = 0; n < apic_nproc; n++) {
955 		if (!apic_cpu_in_range(n) || n == cpun ||
956 		    (apic_cpus[n].aci_status & APIC_CPU_INTR_ENABLE) == 0)
957 			continue;
958 
959 		for (i = APIX_AVINTR_MIN; i <= APIX_AVINTR_MAX; i++) {
960 			vecp = xv_vector(n, i);
961 			if (!IS_VECT_ENABLED(vecp) ||
962 			    vecp->v_bound_cpuid != cpun)
963 				continue;
964 
965 			if (vecp->v_type != APIX_TYPE_MSI)
966 				(void) apix_set_cpu(vecp, cpun, &ret);
967 			else
968 				(void) apix_grp_set_cpu(vecp, cpun, &ret);
969 		}
970 	}
971 
972 	lock_clear(&apix_lock);
973 }
974 
975 /*
976  * Allocate vector for IPI
977  * type == -1 indicates it is an internal request. Do not change
978  * resv_vector for these requests.
979  */
980 static int
981 apix_get_ipivect(int ipl, int type)
982 {
983 	uchar_t vector;
984 
985 	if ((vector = apix_alloc_ipi(ipl)) > 0) {
986 		if (type != -1)
987 			apic_resv_vector[ipl] = vector;
988 		return (vector);
989 	}
990 	apic_error |= APIC_ERR_GET_IPIVECT_FAIL;
991 	return (-1);	/* shouldn't happen */
992 }
993 
994 static int
995 apix_get_clkvect(int ipl)
996 {
997 	int vector;
998 
999 	if ((vector = apix_get_ipivect(ipl, -1)) == -1)
1000 		return (-1);
1001 
1002 	apic_clkvect = vector - APIC_BASE_VECT;
1003 	APIC_VERBOSE(IPI, (CE_CONT, "apix: clock vector = %x\n",
1004 	    apic_clkvect));
1005 	return (vector);
1006 }
1007 
1008 static int
1009 apix_post_cpu_start()
1010 {
1011 	int cpun;
1012 	static int cpus_started = 1;
1013 
1014 	/* We know this CPU + BSP  started successfully. */
1015 	cpus_started++;
1016 
1017 	/*
1018 	 * On BSP we would have enabled X2APIC, if supported by processor,
1019 	 * in acpi_probe(), but on AP we do it here.
1020 	 *
1021 	 * We enable X2APIC mode only if BSP is running in X2APIC & the
1022 	 * local APIC mode of the current CPU is MMIO (xAPIC).
1023 	 */
1024 	if (apic_mode == LOCAL_X2APIC && apic_detect_x2apic() &&
1025 	    apic_local_mode() == LOCAL_APIC) {
1026 		apic_enable_x2apic();
1027 	}
1028 
1029 	/*
1030 	 * Switch back to x2apic IPI sending method for performance when target
1031 	 * CPU has entered x2apic mode.
1032 	 */
1033 	if (apic_mode == LOCAL_X2APIC) {
1034 		apic_switch_ipi_callback(B_FALSE);
1035 	}
1036 
1037 	splx(ipltospl(LOCK_LEVEL));
1038 	apix_init_intr();
1039 
1040 	/*
1041 	 * since some systems don't enable the internal cache on the non-boot
1042 	 * cpus, so we have to enable them here
1043 	 */
1044 	setcr0(getcr0() & ~(CR0_CD | CR0_NW));
1045 
1046 #ifdef	DEBUG
1047 	APIC_AV_PENDING_SET();
1048 #else
1049 	if (apic_mode == LOCAL_APIC)
1050 		APIC_AV_PENDING_SET();
1051 #endif	/* DEBUG */
1052 
1053 	/*
1054 	 * We may be booting, or resuming from suspend; aci_status will
1055 	 * be APIC_CPU_INTR_ENABLE if coming from suspend, so we add the
1056 	 * APIC_CPU_ONLINE flag here rather than setting aci_status completely.
1057 	 */
1058 	cpun = psm_get_cpu_id();
1059 	apic_cpus[cpun].aci_status |= APIC_CPU_ONLINE;
1060 
1061 	apic_reg_ops->apic_write(APIC_DIVIDE_REG, apic_divide_reg_init);
1062 
1063 	return (PSM_SUCCESS);
1064 }
1065 
1066 /*
1067  * If this module needs a periodic handler for the interrupt distribution, it
1068  * can be added here. The argument to the periodic handler is not currently
1069  * used, but is reserved for future.
1070  */
1071 static void
1072 apix_post_cyclic_setup(void *arg)
1073 {
1074 	UNREFERENCED_1PARAMETER(arg);
1075 
1076 	cyc_handler_t cyh;
1077 	cyc_time_t cyt;
1078 
1079 	/* cpu_lock is held */
1080 	/* set up a periodic handler for intr redistribution */
1081 
1082 	/*
1083 	 * In peridoc mode intr redistribution processing is done in
1084 	 * apic_intr_enter during clk intr processing
1085 	 */
1086 	if (!apic_oneshot)
1087 		return;
1088 
1089 	/*
1090 	 * Register a periodical handler for the redistribution processing.
1091 	 * Though we would generally prefer to use the DDI interface for
1092 	 * periodic handler invocation, ddi_periodic_add(9F), we are
1093 	 * unfortunately already holding cpu_lock, which ddi_periodic_add will
1094 	 * attempt to take for us.  Thus, we add our own cyclic directly:
1095 	 */
1096 	cyh.cyh_func = (void (*)(void *))apix_redistribute_compute;
1097 	cyh.cyh_arg = NULL;
1098 	cyh.cyh_level = CY_LOW_LEVEL;
1099 
1100 	cyt.cyt_when = 0;
1101 	cyt.cyt_interval = apic_redistribute_sample_interval;
1102 
1103 	apic_cyclic_id = cyclic_add(&cyh, &cyt);
1104 }
1105 
1106 /*
1107  * Called the first time we enable x2apic mode on this cpu.
1108  * Update some of the function pointers to use x2apic routines.
1109  */
1110 void
1111 x2apic_update_psm()
1112 {
1113 	struct psm_ops *pops = &apix_ops;
1114 
1115 	ASSERT(pops != NULL);
1116 
1117 	/*
1118 	 * The pcplusmp module's version of x2apic_update_psm makes additional
1119 	 * changes that we do not have to make here. It needs to make those
1120 	 * changes because pcplusmp relies on the TPR register and the means of
1121 	 * addressing that changes when using the local apic versus the x2apic.
1122 	 * It's also worth noting that the apix driver specific function end up
1123 	 * being apix_foo as opposed to apic_foo and x2apic_foo.
1124 	 */
1125 	pops->psm_send_ipi = x2apic_send_ipi;
1126 
1127 	send_dirintf = pops->psm_send_ipi;
1128 
1129 	apic_mode = LOCAL_X2APIC;
1130 	apic_change_ops();
1131 }
1132 
1133 /*
1134  * This function provides external interface to the nexus for all
1135  * functionalities related to the new DDI interrupt framework.
1136  *
1137  * Input:
1138  * dip     - pointer to the dev_info structure of the requested device
1139  * hdlp    - pointer to the internal interrupt handle structure for the
1140  *	     requested interrupt
1141  * intr_op - opcode for this call
1142  * result  - pointer to the integer that will hold the result to be
1143  *	     passed back if return value is PSM_SUCCESS
1144  *
1145  * Output:
1146  * return value is either PSM_SUCCESS or PSM_FAILURE
1147  */
1148 static int
1149 apix_intr_ops(dev_info_t *dip, ddi_intr_handle_impl_t *hdlp,
1150     psm_intr_op_t intr_op, int *result)
1151 {
1152 	int		cap;
1153 	apix_vector_t	*vecp, *newvecp;
1154 	struct intrspec *ispec, intr_spec;
1155 	processorid_t target;
1156 
1157 	ispec = &intr_spec;
1158 	ispec->intrspec_pri = hdlp->ih_pri;
1159 	ispec->intrspec_vec = hdlp->ih_inum;
1160 	ispec->intrspec_func = hdlp->ih_cb_func;
1161 
1162 	switch (intr_op) {
1163 	case PSM_INTR_OP_ALLOC_VECTORS:
1164 		switch (hdlp->ih_type) {
1165 		case DDI_INTR_TYPE_MSI:
1166 			/* allocate MSI vectors */
1167 			*result = apix_alloc_msi(dip, hdlp->ih_inum,
1168 			    hdlp->ih_scratch1,
1169 			    (int)(uintptr_t)hdlp->ih_scratch2);
1170 			break;
1171 		case DDI_INTR_TYPE_MSIX:
1172 			/* allocate MSI-X vectors */
1173 			*result = apix_alloc_msix(dip, hdlp->ih_inum,
1174 			    hdlp->ih_scratch1,
1175 			    (int)(uintptr_t)hdlp->ih_scratch2);
1176 			break;
1177 		case DDI_INTR_TYPE_FIXED:
1178 			/* allocate or share vector for fixed */
1179 			if ((ihdl_plat_t *)hdlp->ih_private == NULL) {
1180 				return (PSM_FAILURE);
1181 			}
1182 			ispec = ((ihdl_plat_t *)hdlp->ih_private)->ip_ispecp;
1183 			*result = apix_intx_alloc_vector(dip, hdlp->ih_inum,
1184 			    ispec);
1185 			break;
1186 		default:
1187 			return (PSM_FAILURE);
1188 		}
1189 		break;
1190 	case PSM_INTR_OP_FREE_VECTORS:
1191 		apix_free_vectors(dip, hdlp->ih_inum, hdlp->ih_scratch1,
1192 		    hdlp->ih_type);
1193 		break;
1194 	case PSM_INTR_OP_XLATE_VECTOR:
1195 		/*
1196 		 * Vectors are allocated by ALLOC and freed by FREE.
1197 		 * XLATE finds and returns APIX_VIRTVEC_VECTOR(cpu, vector).
1198 		 */
1199 		*result = APIX_INVALID_VECT;
1200 		vecp = apix_get_dev_map(dip, hdlp->ih_inum, hdlp->ih_type);
1201 		if (vecp != NULL) {
1202 			*result = APIX_VIRTVECTOR(vecp->v_cpuid,
1203 			    vecp->v_vector);
1204 			break;
1205 		}
1206 
1207 		/*
1208 		 * No vector to device mapping exists. If this is FIXED type
1209 		 * then check if this IRQ is already mapped for another device
1210 		 * then return the vector number for it (i.e. shared IRQ case).
1211 		 * Otherwise, return PSM_FAILURE.
1212 		 */
1213 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED) {
1214 			vecp = apix_intx_xlate_vector(dip, hdlp->ih_inum,
1215 			    ispec);
1216 			*result = (vecp == NULL) ? APIX_INVALID_VECT :
1217 			    APIX_VIRTVECTOR(vecp->v_cpuid, vecp->v_vector);
1218 		}
1219 		if (*result == APIX_INVALID_VECT)
1220 			return (PSM_FAILURE);
1221 		break;
1222 	case PSM_INTR_OP_GET_PENDING:
1223 		vecp = apix_get_dev_map(dip, hdlp->ih_inum, hdlp->ih_type);
1224 		if (vecp == NULL)
1225 			return (PSM_FAILURE);
1226 
1227 		*result = apix_get_pending(vecp);
1228 		break;
1229 	case PSM_INTR_OP_CLEAR_MASK:
1230 		if (hdlp->ih_type != DDI_INTR_TYPE_FIXED)
1231 			return (PSM_FAILURE);
1232 
1233 		vecp = apix_get_dev_map(dip, hdlp->ih_inum, hdlp->ih_type);
1234 		if (vecp == NULL)
1235 			return (PSM_FAILURE);
1236 
1237 		apix_intx_clear_mask(vecp->v_inum);
1238 		break;
1239 	case PSM_INTR_OP_SET_MASK:
1240 		if (hdlp->ih_type != DDI_INTR_TYPE_FIXED)
1241 			return (PSM_FAILURE);
1242 
1243 		vecp = apix_get_dev_map(dip, hdlp->ih_inum, hdlp->ih_type);
1244 		if (vecp == NULL)
1245 			return (PSM_FAILURE);
1246 
1247 		apix_intx_set_mask(vecp->v_inum);
1248 		break;
1249 	case PSM_INTR_OP_GET_SHARED:
1250 		if (hdlp->ih_type != DDI_INTR_TYPE_FIXED)
1251 			return (PSM_FAILURE);
1252 
1253 		vecp = apix_get_dev_map(dip, hdlp->ih_inum, hdlp->ih_type);
1254 		if (vecp == NULL)
1255 			return (PSM_FAILURE);
1256 
1257 		*result = apix_intx_get_shared(vecp->v_inum);
1258 		break;
1259 	case PSM_INTR_OP_SET_PRI:
1260 		/*
1261 		 * Called prior to adding the interrupt handler or when
1262 		 * an interrupt handler is unassigned.
1263 		 */
1264 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED)
1265 			return (PSM_SUCCESS);
1266 
1267 		if (apix_get_dev_map(dip, hdlp->ih_inum, hdlp->ih_type) == NULL)
1268 			return (PSM_FAILURE);
1269 
1270 		break;
1271 	case PSM_INTR_OP_SET_CPU:
1272 	case PSM_INTR_OP_GRP_SET_CPU:
1273 		/*
1274 		 * The interrupt handle given here has been allocated
1275 		 * specifically for this command, and ih_private carries
1276 		 * a CPU value.
1277 		 */
1278 		*result = EINVAL;
1279 		target = (int)(intptr_t)hdlp->ih_private;
1280 		if (!apic_cpu_in_range(target)) {
1281 			DDI_INTR_IMPLDBG((CE_WARN,
1282 			    "[grp_]set_cpu: cpu out of range: %d\n", target));
1283 			return (PSM_FAILURE);
1284 		}
1285 
1286 		lock_set(&apix_lock);
1287 
1288 		vecp = apix_get_req_vector(hdlp, hdlp->ih_flags);
1289 		if (!IS_VECT_ENABLED(vecp)) {
1290 			DDI_INTR_IMPLDBG((CE_WARN,
1291 			    "[grp]_set_cpu: invalid vector 0x%x\n",
1292 			    hdlp->ih_vector));
1293 			lock_clear(&apix_lock);
1294 			return (PSM_FAILURE);
1295 		}
1296 
1297 		*result = 0;
1298 
1299 		if (intr_op == PSM_INTR_OP_SET_CPU)
1300 			newvecp = apix_set_cpu(vecp, target, result);
1301 		else
1302 			newvecp = apix_grp_set_cpu(vecp, target, result);
1303 
1304 		lock_clear(&apix_lock);
1305 
1306 		if (newvecp == NULL) {
1307 			*result = EIO;
1308 			return (PSM_FAILURE);
1309 		}
1310 		newvecp->v_bound_cpuid = target;
1311 		hdlp->ih_vector = APIX_VIRTVECTOR(newvecp->v_cpuid,
1312 		    newvecp->v_vector);
1313 		break;
1314 
1315 	case PSM_INTR_OP_GET_INTR:
1316 		/*
1317 		 * The interrupt handle given here has been allocated
1318 		 * specifically for this command, and ih_private carries
1319 		 * a pointer to a apic_get_intr_t.
1320 		 */
1321 		if (apix_get_intr_info(hdlp, hdlp->ih_private) != PSM_SUCCESS)
1322 			return (PSM_FAILURE);
1323 		break;
1324 
1325 	case PSM_INTR_OP_CHECK_MSI:
1326 		/*
1327 		 * Check MSI/X is supported or not at APIC level and
1328 		 * masked off the MSI/X bits in hdlp->ih_type if not
1329 		 * supported before return.  If MSI/X is supported,
1330 		 * leave the ih_type unchanged and return.
1331 		 *
1332 		 * hdlp->ih_type passed in from the nexus has all the
1333 		 * interrupt types supported by the device.
1334 		 */
1335 		if (apic_support_msi == 0) {	/* uninitialized */
1336 			/*
1337 			 * if apic_support_msi is not set, call
1338 			 * apic_check_msi_support() to check whether msi
1339 			 * is supported first
1340 			 */
1341 			if (apic_check_msi_support() == PSM_SUCCESS)
1342 				apic_support_msi = 1;	/* supported */
1343 			else
1344 				apic_support_msi = -1;	/* not-supported */
1345 		}
1346 		if (apic_support_msi == 1) {
1347 			if (apic_msix_enable)
1348 				*result = hdlp->ih_type;
1349 			else
1350 				*result = hdlp->ih_type & ~DDI_INTR_TYPE_MSIX;
1351 		} else
1352 			*result = hdlp->ih_type & ~(DDI_INTR_TYPE_MSI |
1353 			    DDI_INTR_TYPE_MSIX);
1354 		break;
1355 	case PSM_INTR_OP_GET_CAP:
1356 		cap = DDI_INTR_FLAG_PENDING;
1357 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED)
1358 			cap |= DDI_INTR_FLAG_MASKABLE;
1359 		*result = cap;
1360 		break;
1361 	case PSM_INTR_OP_APIC_TYPE:
1362 		((apic_get_type_t *)(hdlp->ih_private))->avgi_type =
1363 		    apix_get_apic_type();
1364 		((apic_get_type_t *)(hdlp->ih_private))->avgi_num_intr =
1365 		    APIX_IPI_MIN;
1366 		((apic_get_type_t *)(hdlp->ih_private))->avgi_num_cpu =
1367 		    apic_nproc;
1368 		hdlp->ih_ver = apic_get_apic_version();
1369 		break;
1370 	case PSM_INTR_OP_SET_CAP:
1371 	default:
1372 		return (PSM_FAILURE);
1373 	}
1374 
1375 	return (PSM_SUCCESS);
1376 }
1377 
1378 static void
1379 apix_cleanup_busy(void)
1380 {
1381 	int i, j;
1382 	apix_vector_t *vecp;
1383 
1384 	for (i = 0; i < apic_nproc; i++) {
1385 		if (!apic_cpu_in_range(i))
1386 			continue;
1387 		apic_cpus[i].aci_busy = 0;
1388 		for (j = APIX_AVINTR_MIN; j < APIX_AVINTR_MAX; j++) {
1389 			if ((vecp = xv_vector(i, j)) != NULL)
1390 				vecp->v_busy = 0;
1391 		}
1392 	}
1393 }
1394 
1395 static void
1396 apix_redistribute_compute(void)
1397 {
1398 	int	i, j, max_busy;
1399 
1400 	if (!apic_enable_dynamic_migration)
1401 		return;
1402 
1403 	if (++apic_nticks == apic_sample_factor_redistribution) {
1404 		/*
1405 		 * Time to call apic_intr_redistribute().
1406 		 * reset apic_nticks. This will cause max_busy
1407 		 * to be calculated below and if it is more than
1408 		 * apic_int_busy, we will do the whole thing
1409 		 */
1410 		apic_nticks = 0;
1411 	}
1412 	max_busy = 0;
1413 	for (i = 0; i < apic_nproc; i++) {
1414 		if (!apic_cpu_in_range(i))
1415 			continue;
1416 		/*
1417 		 * Check if curipl is non zero & if ISR is in
1418 		 * progress
1419 		 */
1420 		if (((j = apic_cpus[i].aci_curipl) != 0) &&
1421 		    (apic_cpus[i].aci_ISR_in_progress & (1 << j))) {
1422 
1423 			int	vect;
1424 			apic_cpus[i].aci_busy++;
1425 			vect = apic_cpus[i].aci_current[j];
1426 			apixs[i]->x_vectbl[vect]->v_busy++;
1427 		}
1428 
1429 		if (!apic_nticks &&
1430 		    (apic_cpus[i].aci_busy > max_busy))
1431 			max_busy = apic_cpus[i].aci_busy;
1432 	}
1433 	if (!apic_nticks) {
1434 		if (max_busy > apic_int_busy_mark) {
1435 		/*
1436 		 * We could make the following check be
1437 		 * skipped > 1 in which case, we get a
1438 		 * redistribution at half the busy mark (due to
1439 		 * double interval). Need to be able to collect
1440 		 * more empirical data to decide if that is a
1441 		 * good strategy. Punt for now.
1442 		 */
1443 			apix_cleanup_busy();
1444 			apic_skipped_redistribute = 0;
1445 		} else
1446 			apic_skipped_redistribute++;
1447 	}
1448 }
1449 
1450 /*
1451  * intr_ops() service routines
1452  */
1453 
1454 static int
1455 apix_get_pending(apix_vector_t *vecp)
1456 {
1457 	int bit, index, irr, pending;
1458 
1459 	/* need to get on the bound cpu */
1460 	mutex_enter(&cpu_lock);
1461 	affinity_set(vecp->v_cpuid);
1462 
1463 	index = vecp->v_vector / 32;
1464 	bit = vecp->v_vector % 32;
1465 	irr = apic_reg_ops->apic_read(APIC_IRR_REG + index);
1466 
1467 	affinity_clear();
1468 	mutex_exit(&cpu_lock);
1469 
1470 	pending = (irr & (1 << bit)) ? 1 : 0;
1471 	if (!pending && vecp->v_type == APIX_TYPE_FIXED)
1472 		pending = apix_intx_get_pending(vecp->v_inum);
1473 
1474 	return (pending);
1475 }
1476 
1477 static apix_vector_t *
1478 apix_get_req_vector(ddi_intr_handle_impl_t *hdlp, ushort_t flags)
1479 {
1480 	apix_vector_t *vecp;
1481 	processorid_t cpuid;
1482 	int32_t virt_vec = 0;
1483 
1484 	switch (flags & PSMGI_INTRBY_FLAGS) {
1485 	case PSMGI_INTRBY_IRQ:
1486 		return (apix_intx_get_vector(hdlp->ih_vector));
1487 	case PSMGI_INTRBY_VEC:
1488 		virt_vec = (virt_vec == 0) ? hdlp->ih_vector : virt_vec;
1489 
1490 		cpuid = APIX_VIRTVEC_CPU(virt_vec);
1491 		if (!apic_cpu_in_range(cpuid))
1492 			return (NULL);
1493 
1494 		vecp = xv_vector(cpuid, APIX_VIRTVEC_VECTOR(virt_vec));
1495 		break;
1496 	case PSMGI_INTRBY_DEFAULT:
1497 		vecp = apix_get_dev_map(hdlp->ih_dip, hdlp->ih_inum,
1498 		    hdlp->ih_type);
1499 		break;
1500 	default:
1501 		return (NULL);
1502 	}
1503 
1504 	return (vecp);
1505 }
1506 
1507 static int
1508 apix_get_intr_info(ddi_intr_handle_impl_t *hdlp,
1509     apic_get_intr_t *intr_params_p)
1510 {
1511 	apix_vector_t *vecp;
1512 	struct autovec *av_dev;
1513 	int i;
1514 
1515 	vecp = apix_get_req_vector(hdlp, intr_params_p->avgi_req_flags);
1516 	if (IS_VECT_FREE(vecp)) {
1517 		intr_params_p->avgi_num_devs = 0;
1518 		intr_params_p->avgi_cpu_id = 0;
1519 		intr_params_p->avgi_req_flags = 0;
1520 		return (PSM_SUCCESS);
1521 	}
1522 
1523 	if (intr_params_p->avgi_req_flags & PSMGI_REQ_CPUID) {
1524 		intr_params_p->avgi_cpu_id = vecp->v_cpuid;
1525 
1526 		/* Return user bound info for intrd. */
1527 		if (intr_params_p->avgi_cpu_id & IRQ_USER_BOUND) {
1528 			intr_params_p->avgi_cpu_id &= ~IRQ_USER_BOUND;
1529 			intr_params_p->avgi_cpu_id |= PSMGI_CPU_USER_BOUND;
1530 		}
1531 	}
1532 
1533 	if (intr_params_p->avgi_req_flags & PSMGI_REQ_VECTOR)
1534 		intr_params_p->avgi_vector = vecp->v_vector;
1535 
1536 	if (intr_params_p->avgi_req_flags &
1537 	    (PSMGI_REQ_NUM_DEVS | PSMGI_REQ_GET_DEVS))
1538 		/* Get number of devices from apic_irq table shared field. */
1539 		intr_params_p->avgi_num_devs = vecp->v_share;
1540 
1541 	if (intr_params_p->avgi_req_flags &  PSMGI_REQ_GET_DEVS) {
1542 
1543 		intr_params_p->avgi_req_flags  |= PSMGI_REQ_NUM_DEVS;
1544 
1545 		/* Some devices have NULL dip.  Don't count these. */
1546 		if (intr_params_p->avgi_num_devs > 0) {
1547 			for (i = 0, av_dev = vecp->v_autovect; av_dev;
1548 			    av_dev = av_dev->av_link) {
1549 				if (av_dev->av_vector && av_dev->av_dip)
1550 					i++;
1551 			}
1552 			intr_params_p->avgi_num_devs =
1553 			    (uint8_t)MIN(intr_params_p->avgi_num_devs, i);
1554 		}
1555 
1556 		/* There are no viable dips to return. */
1557 		if (intr_params_p->avgi_num_devs == 0) {
1558 			intr_params_p->avgi_dip_list = NULL;
1559 
1560 		} else {	/* Return list of dips */
1561 
1562 			/* Allocate space in array for that number of devs. */
1563 			intr_params_p->avgi_dip_list = kmem_zalloc(
1564 			    intr_params_p->avgi_num_devs *
1565 			    sizeof (dev_info_t *),
1566 			    KM_NOSLEEP);
1567 			if (intr_params_p->avgi_dip_list == NULL) {
1568 				DDI_INTR_IMPLDBG((CE_WARN,
1569 				    "apix_get_vector_intr_info: no memory"));
1570 				return (PSM_FAILURE);
1571 			}
1572 
1573 			/*
1574 			 * Loop through the device list of the autovec table
1575 			 * filling in the dip array.
1576 			 *
1577 			 * Note that the autovect table may have some special
1578 			 * entries which contain NULL dips.  These will be
1579 			 * ignored.
1580 			 */
1581 			for (i = 0, av_dev = vecp->v_autovect; av_dev;
1582 			    av_dev = av_dev->av_link) {
1583 				if (av_dev->av_vector && av_dev->av_dip)
1584 					intr_params_p->avgi_dip_list[i++] =
1585 					    av_dev->av_dip;
1586 			}
1587 		}
1588 	}
1589 
1590 	return (PSM_SUCCESS);
1591 }
1592 
1593 static char *
1594 apix_get_apic_type(void)
1595 {
1596 	return (apix_psm_info.p_mach_idstring);
1597 }
1598 
1599 apix_vector_t *
1600 apix_set_cpu(apix_vector_t *vecp, int new_cpu, int *result)
1601 {
1602 	apix_vector_t *newp = NULL;
1603 	dev_info_t *dip;
1604 	int inum, cap_ptr;
1605 	ddi_acc_handle_t handle;
1606 	ddi_intr_msix_t *msix_p = NULL;
1607 	ushort_t msix_ctrl;
1608 	uintptr_t off;
1609 	uint32_t mask;
1610 
1611 	ASSERT(LOCK_HELD(&apix_lock));
1612 	*result = ENXIO;
1613 
1614 	/* Fail if this is an MSI intr and is part of a group. */
1615 	if (vecp->v_type == APIX_TYPE_MSI) {
1616 		if (i_ddi_intr_get_current_nintrs(APIX_GET_DIP(vecp)) > 1)
1617 			return (NULL);
1618 		else
1619 			return (apix_grp_set_cpu(vecp, new_cpu, result));
1620 	}
1621 
1622 	/*
1623 	 * Mask MSI-X. It's unmasked when MSI-X gets enabled.
1624 	 */
1625 	if (vecp->v_type == APIX_TYPE_MSIX && IS_VECT_ENABLED(vecp)) {
1626 		if ((dip = APIX_GET_DIP(vecp)) == NULL)
1627 			return (NULL);
1628 		inum = vecp->v_devp->dv_inum;
1629 
1630 		handle = i_ddi_get_pci_config_handle(dip);
1631 		cap_ptr = i_ddi_get_msi_msix_cap_ptr(dip);
1632 		msix_ctrl = pci_config_get16(handle, cap_ptr + PCI_MSIX_CTRL);
1633 		if ((msix_ctrl & PCI_MSIX_FUNCTION_MASK) == 0) {
1634 			/*
1635 			 * Function is not masked, then mask "inum"th
1636 			 * entry in the MSI-X table
1637 			 */
1638 			msix_p = i_ddi_get_msix(dip);
1639 			off = (uintptr_t)msix_p->msix_tbl_addr + (inum *
1640 			    PCI_MSIX_VECTOR_SIZE) + PCI_MSIX_VECTOR_CTRL_OFFSET;
1641 			mask = ddi_get32(msix_p->msix_tbl_hdl, (uint32_t *)off);
1642 			ddi_put32(msix_p->msix_tbl_hdl, (uint32_t *)off,
1643 			    mask | 1);
1644 		}
1645 	}
1646 
1647 	*result = 0;
1648 	if ((newp = apix_rebind(vecp, new_cpu, 1)) == NULL)
1649 		*result = EIO;
1650 
1651 	/* Restore mask bit */
1652 	if (msix_p != NULL)
1653 		ddi_put32(msix_p->msix_tbl_hdl, (uint32_t *)off, mask);
1654 
1655 	return (newp);
1656 }
1657 
1658 /*
1659  * Set cpu for MSIs
1660  */
1661 apix_vector_t *
1662 apix_grp_set_cpu(apix_vector_t *vecp, int new_cpu, int *result)
1663 {
1664 	apix_vector_t *newp, *vp;
1665 	uint32_t orig_cpu = vecp->v_cpuid;
1666 	int orig_vect = vecp->v_vector;
1667 	int i, num_vectors, cap_ptr, msi_mask_off;
1668 	uint32_t msi_pvm;
1669 	ushort_t msi_ctrl;
1670 	ddi_acc_handle_t handle;
1671 	dev_info_t *dip;
1672 
1673 	APIC_VERBOSE(INTR, (CE_CONT, "apix_grp_set_cpu: oldcpu: %x, vector: %x,"
1674 	    " newcpu:%x\n", vecp->v_cpuid, vecp->v_vector, new_cpu));
1675 
1676 	ASSERT(LOCK_HELD(&apix_lock));
1677 
1678 	*result = ENXIO;
1679 
1680 	if (vecp->v_type != APIX_TYPE_MSI) {
1681 		DDI_INTR_IMPLDBG((CE_WARN, "set_grp: intr not MSI\n"));
1682 		return (NULL);
1683 	}
1684 
1685 	if ((dip = APIX_GET_DIP(vecp)) == NULL)
1686 		return (NULL);
1687 
1688 	num_vectors = i_ddi_intr_get_current_nintrs(dip);
1689 	if ((num_vectors < 1) || ((num_vectors - 1) & orig_vect)) {
1690 		APIC_VERBOSE(INTR, (CE_WARN,
1691 		    "set_grp: base vec not part of a grp or not aligned: "
1692 		    "vec:0x%x, num_vec:0x%x\n", orig_vect, num_vectors));
1693 		return (NULL);
1694 	}
1695 
1696 	if (vecp->v_inum != apix_get_min_dev_inum(dip, vecp->v_type))
1697 		return (NULL);
1698 
1699 	*result = EIO;
1700 	for (i = 1; i < num_vectors; i++) {
1701 		if ((vp = xv_vector(orig_cpu, orig_vect + i)) == NULL)
1702 			return (NULL);
1703 #ifdef DEBUG
1704 		/*
1705 		 * Sanity check: CPU and dip is the same for all entries.
1706 		 * May be called when first msi to be enabled, at this time
1707 		 * add_avintr() is not called for other msi
1708 		 */
1709 		if ((vp->v_share != 0) &&
1710 		    ((APIX_GET_DIP(vp) != dip) ||
1711 		    (vp->v_cpuid != vecp->v_cpuid))) {
1712 			APIC_VERBOSE(INTR, (CE_WARN,
1713 			    "set_grp: cpu or dip for vec 0x%x difft than for "
1714 			    "vec 0x%x\n", orig_vect, orig_vect + i));
1715 			APIC_VERBOSE(INTR, (CE_WARN,
1716 			    "  cpu: %d vs %d, dip: 0x%p vs 0x%p\n", orig_cpu,
1717 			    vp->v_cpuid, (void *)dip,
1718 			    (void *)APIX_GET_DIP(vp)));
1719 			return (NULL);
1720 		}
1721 #endif /* DEBUG */
1722 	}
1723 
1724 	cap_ptr = i_ddi_get_msi_msix_cap_ptr(dip);
1725 	handle = i_ddi_get_pci_config_handle(dip);
1726 	msi_ctrl = pci_config_get16(handle, cap_ptr + PCI_MSI_CTRL);
1727 
1728 	/* MSI Per vector masking is supported. */
1729 	if (msi_ctrl & PCI_MSI_PVM_MASK) {
1730 		if (msi_ctrl &  PCI_MSI_64BIT_MASK)
1731 			msi_mask_off = cap_ptr + PCI_MSI_64BIT_MASKBITS;
1732 		else
1733 			msi_mask_off = cap_ptr + PCI_MSI_32BIT_MASK;
1734 		msi_pvm = pci_config_get32(handle, msi_mask_off);
1735 		pci_config_put32(handle, msi_mask_off, (uint32_t)-1);
1736 		APIC_VERBOSE(INTR, (CE_CONT,
1737 		    "set_grp: pvm supported.  Mask set to 0x%x\n",
1738 		    pci_config_get32(handle, msi_mask_off)));
1739 	}
1740 
1741 	if ((newp = apix_rebind(vecp, new_cpu, num_vectors)) != NULL)
1742 		*result = 0;
1743 
1744 	/* Reenable vectors if per vector masking is supported. */
1745 	if (msi_ctrl & PCI_MSI_PVM_MASK) {
1746 		pci_config_put32(handle, msi_mask_off, msi_pvm);
1747 		APIC_VERBOSE(INTR, (CE_CONT,
1748 		    "set_grp: pvm supported.  Mask restored to 0x%x\n",
1749 		    pci_config_get32(handle, msi_mask_off)));
1750 	}
1751 
1752 	return (newp);
1753 }
1754 
1755 void
1756 apix_intx_set_vector(int irqno, uint32_t cpuid, uchar_t vector)
1757 {
1758 	apic_irq_t *irqp;
1759 
1760 	mutex_enter(&airq_mutex);
1761 	irqp = apic_irq_table[irqno];
1762 	irqp->airq_cpu = cpuid;
1763 	irqp->airq_vector = vector;
1764 	apic_record_rdt_entry(irqp, irqno);
1765 	mutex_exit(&airq_mutex);
1766 }
1767 
1768 apix_vector_t *
1769 apix_intx_get_vector(int irqno)
1770 {
1771 	apic_irq_t *irqp;
1772 	uint32_t cpuid;
1773 	uchar_t vector;
1774 
1775 	mutex_enter(&airq_mutex);
1776 	irqp = apic_irq_table[irqno & 0xff];
1777 	if (IS_IRQ_FREE(irqp) || (irqp->airq_cpu == IRQ_UNINIT)) {
1778 		mutex_exit(&airq_mutex);
1779 		return (NULL);
1780 	}
1781 	cpuid = irqp->airq_cpu;
1782 	vector = irqp->airq_vector;
1783 	mutex_exit(&airq_mutex);
1784 
1785 	return (xv_vector(cpuid, vector));
1786 }
1787 
1788 /*
1789  * Must called with interrupts disabled and apic_ioapic_lock held
1790  */
1791 void
1792 apix_intx_enable(int irqno)
1793 {
1794 	uchar_t ioapicindex, intin;
1795 	apic_irq_t *irqp = apic_irq_table[irqno];
1796 	ioapic_rdt_t irdt;
1797 	apic_cpus_info_t *cpu_infop;
1798 	apix_vector_t *vecp = xv_vector(irqp->airq_cpu, irqp->airq_vector);
1799 
1800 	ASSERT(LOCK_HELD(&apic_ioapic_lock) && !IS_IRQ_FREE(irqp));
1801 
1802 	ioapicindex = irqp->airq_ioapicindex;
1803 	intin = irqp->airq_intin_no;
1804 	cpu_infop =  &apic_cpus[irqp->airq_cpu];
1805 
1806 	irdt.ir_lo = AV_PDEST | AV_FIXED | irqp->airq_rdt_entry;
1807 	irdt.ir_hi = cpu_infop->aci_local_id;
1808 
1809 	apic_vt_ops->apic_intrmap_alloc_entry(&vecp->v_intrmap_private, NULL,
1810 	    vecp->v_type, 1, ioapicindex);
1811 	apic_vt_ops->apic_intrmap_map_entry(vecp->v_intrmap_private,
1812 	    (void *)&irdt, vecp->v_type, 1);
1813 	apic_vt_ops->apic_intrmap_record_rdt(vecp->v_intrmap_private, &irdt);
1814 
1815 	/* write RDT entry high dword - destination */
1816 	WRITE_IOAPIC_RDT_ENTRY_HIGH_DWORD(ioapicindex, intin,
1817 	    irdt.ir_hi);
1818 
1819 	/* Write the vector, trigger, and polarity portion of the RDT */
1820 	WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapicindex, intin, irdt.ir_lo);
1821 
1822 	vecp->v_state = APIX_STATE_ENABLED;
1823 
1824 	APIC_VERBOSE_IOAPIC((CE_CONT, "apix_intx_enable: ioapic 0x%x"
1825 	    " intin 0x%x rdt_low 0x%x rdt_high 0x%x\n",
1826 	    ioapicindex, intin, irdt.ir_lo, irdt.ir_hi));
1827 }
1828 
1829 /*
1830  * Must called with interrupts disabled and apic_ioapic_lock held
1831  */
1832 void
1833 apix_intx_disable(int irqno)
1834 {
1835 	apic_irq_t *irqp = apic_irq_table[irqno];
1836 	int ioapicindex, intin;
1837 
1838 	ASSERT(LOCK_HELD(&apic_ioapic_lock) && !IS_IRQ_FREE(irqp));
1839 	/*
1840 	 * The assumption here is that this is safe, even for
1841 	 * systems with IOAPICs that suffer from the hardware
1842 	 * erratum because all devices have been quiesced before
1843 	 * they unregister their interrupt handlers.  If that
1844 	 * assumption turns out to be false, this mask operation
1845 	 * can induce the same erratum result we're trying to
1846 	 * avoid.
1847 	 */
1848 	ioapicindex = irqp->airq_ioapicindex;
1849 	intin = irqp->airq_intin_no;
1850 	ioapic_write(ioapicindex, APIC_RDT_CMD + 2 * intin, AV_MASK);
1851 
1852 	APIC_VERBOSE_IOAPIC((CE_CONT, "apix_intx_disable: ioapic 0x%x"
1853 	    " intin 0x%x\n", ioapicindex, intin));
1854 }
1855 
1856 void
1857 apix_intx_free(int irqno)
1858 {
1859 	apic_irq_t *irqp;
1860 
1861 	mutex_enter(&airq_mutex);
1862 	irqp = apic_irq_table[irqno];
1863 
1864 	if (IS_IRQ_FREE(irqp)) {
1865 		mutex_exit(&airq_mutex);
1866 		return;
1867 	}
1868 
1869 	irqp->airq_mps_intr_index = FREE_INDEX;
1870 	irqp->airq_cpu = IRQ_UNINIT;
1871 	irqp->airq_vector = APIX_INVALID_VECT;
1872 	mutex_exit(&airq_mutex);
1873 }
1874 
1875 #ifdef DEBUG
1876 int apix_intr_deliver_timeouts = 0;
1877 int apix_intr_rirr_timeouts = 0;
1878 int apix_intr_rirr_reset_failure = 0;
1879 #endif
1880 int apix_max_reps_irr_pending = 10;
1881 
1882 #define	GET_RDT_BITS(ioapic, intin, bits)	\
1883 	(READ_IOAPIC_RDT_ENTRY_LOW_DWORD((ioapic), (intin)) & (bits))
1884 #define	APIX_CHECK_IRR_DELAY	drv_usectohz(5000)
1885 
1886 int
1887 apix_intx_rebind(int irqno, processorid_t cpuid, uchar_t vector)
1888 {
1889 	apic_irq_t *irqp = apic_irq_table[irqno];
1890 	ulong_t iflag;
1891 	int waited, ioapic_ix, intin_no, level, repeats, rdt_entry, masked;
1892 
1893 	ASSERT(irqp != NULL);
1894 
1895 	iflag = intr_clear();
1896 	lock_set(&apic_ioapic_lock);
1897 
1898 	ioapic_ix = irqp->airq_ioapicindex;
1899 	intin_no = irqp->airq_intin_no;
1900 	level = apic_level_intr[irqno];
1901 
1902 	/*
1903 	 * Wait for the delivery status bit to be cleared. This should
1904 	 * be a very small amount of time.
1905 	 */
1906 	repeats = 0;
1907 	do {
1908 		repeats++;
1909 
1910 		for (waited = 0; waited < apic_max_reps_clear_pending;
1911 		    waited++) {
1912 			if (GET_RDT_BITS(ioapic_ix, intin_no, AV_PENDING) == 0)
1913 				break;
1914 		}
1915 		if (!level)
1916 			break;
1917 
1918 		/*
1919 		 * Mask the RDT entry for level-triggered interrupts.
1920 		 */
1921 		irqp->airq_rdt_entry |= AV_MASK;
1922 		rdt_entry = READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
1923 		    intin_no);
1924 		if ((masked = (rdt_entry & AV_MASK)) == 0) {
1925 			/* Mask it */
1926 			WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix, intin_no,
1927 			    AV_MASK | rdt_entry);
1928 		}
1929 
1930 		/*
1931 		 * If there was a race and an interrupt was injected
1932 		 * just before we masked, check for that case here.
1933 		 * Then, unmask the RDT entry and try again.  If we're
1934 		 * on our last try, don't unmask (because we want the
1935 		 * RDT entry to remain masked for the rest of the
1936 		 * function).
1937 		 */
1938 		rdt_entry = READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
1939 		    intin_no);
1940 		if ((masked == 0) && ((rdt_entry & AV_PENDING) != 0) &&
1941 		    (repeats < apic_max_reps_clear_pending)) {
1942 			/* Unmask it */
1943 			WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
1944 			    intin_no, rdt_entry & ~AV_MASK);
1945 			irqp->airq_rdt_entry &= ~AV_MASK;
1946 		}
1947 	} while ((rdt_entry & AV_PENDING) &&
1948 	    (repeats < apic_max_reps_clear_pending));
1949 
1950 #ifdef DEBUG
1951 	if (GET_RDT_BITS(ioapic_ix, intin_no, AV_PENDING) != 0)
1952 		apix_intr_deliver_timeouts++;
1953 #endif
1954 
1955 	if (!level || !APIX_IS_MASK_RDT(apix_mul_ioapic_method))
1956 		goto done;
1957 
1958 	/*
1959 	 * wait for remote IRR to be cleared for level-triggered
1960 	 * interrupts
1961 	 */
1962 	repeats = 0;
1963 	do {
1964 		repeats++;
1965 
1966 		for (waited = 0; waited < apic_max_reps_clear_pending;
1967 		    waited++) {
1968 			if (GET_RDT_BITS(ioapic_ix, intin_no, AV_REMOTE_IRR)
1969 			    == 0)
1970 				break;
1971 		}
1972 
1973 		if (GET_RDT_BITS(ioapic_ix, intin_no, AV_REMOTE_IRR) != 0) {
1974 			lock_clear(&apic_ioapic_lock);
1975 			intr_restore(iflag);
1976 
1977 			delay(APIX_CHECK_IRR_DELAY);
1978 
1979 			iflag = intr_clear();
1980 			lock_set(&apic_ioapic_lock);
1981 		}
1982 	} while (repeats < apix_max_reps_irr_pending);
1983 
1984 	if (repeats >= apix_max_reps_irr_pending) {
1985 #ifdef DEBUG
1986 		apix_intr_rirr_timeouts++;
1987 #endif
1988 
1989 		/*
1990 		 * If we waited and the Remote IRR bit is still not cleared,
1991 		 * AND if we've invoked the timeout APIC_REPROGRAM_MAX_TIMEOUTS
1992 		 * times for this interrupt, try the last-ditch workaround:
1993 		 */
1994 		if (GET_RDT_BITS(ioapic_ix, intin_no, AV_REMOTE_IRR) != 0) {
1995 			/*
1996 			 * Trying to clear the bit through normal
1997 			 * channels has failed.  So as a last-ditch
1998 			 * effort, try to set the trigger mode to
1999 			 * edge, then to level.  This has been
2000 			 * observed to work on many systems.
2001 			 */
2002 			WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
2003 			    intin_no,
2004 			    READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
2005 			    intin_no) & ~AV_LEVEL);
2006 			WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
2007 			    intin_no,
2008 			    READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
2009 			    intin_no) | AV_LEVEL);
2010 		}
2011 
2012 		if (GET_RDT_BITS(ioapic_ix, intin_no, AV_REMOTE_IRR) != 0) {
2013 #ifdef DEBUG
2014 			apix_intr_rirr_reset_failure++;
2015 #endif
2016 			lock_clear(&apic_ioapic_lock);
2017 			intr_restore(iflag);
2018 			prom_printf("apix: Remote IRR still "
2019 			    "not clear for IOAPIC %d intin %d.\n"
2020 			    "\tInterrupts to this pin may cease "
2021 			    "functioning.\n", ioapic_ix, intin_no);
2022 			return (1);	/* return failure */
2023 		}
2024 	}
2025 
2026 done:
2027 	/* change apic_irq_table */
2028 	lock_clear(&apic_ioapic_lock);
2029 	intr_restore(iflag);
2030 	apix_intx_set_vector(irqno, cpuid, vector);
2031 	iflag = intr_clear();
2032 	lock_set(&apic_ioapic_lock);
2033 
2034 	/* reprogramme IO-APIC RDT entry */
2035 	apix_intx_enable(irqno);
2036 
2037 	lock_clear(&apic_ioapic_lock);
2038 	intr_restore(iflag);
2039 
2040 	return (0);
2041 }
2042 
2043 static int
2044 apix_intx_get_pending(int irqno)
2045 {
2046 	apic_irq_t *irqp;
2047 	int intin, ioapicindex, pending;
2048 	ulong_t iflag;
2049 
2050 	mutex_enter(&airq_mutex);
2051 	irqp = apic_irq_table[irqno];
2052 	if (IS_IRQ_FREE(irqp)) {
2053 		mutex_exit(&airq_mutex);
2054 		return (0);
2055 	}
2056 
2057 	/* check IO-APIC delivery status */
2058 	intin = irqp->airq_intin_no;
2059 	ioapicindex = irqp->airq_ioapicindex;
2060 	mutex_exit(&airq_mutex);
2061 
2062 	iflag = intr_clear();
2063 	lock_set(&apic_ioapic_lock);
2064 
2065 	pending = (READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapicindex, intin) &
2066 	    AV_PENDING) ? 1 : 0;
2067 
2068 	lock_clear(&apic_ioapic_lock);
2069 	intr_restore(iflag);
2070 
2071 	return (pending);
2072 }
2073 
2074 /*
2075  * This function will mask the interrupt on the I/O APIC
2076  */
2077 static void
2078 apix_intx_set_mask(int irqno)
2079 {
2080 	int intin, ioapixindex, rdt_entry;
2081 	ulong_t iflag;
2082 	apic_irq_t *irqp;
2083 
2084 	mutex_enter(&airq_mutex);
2085 	irqp = apic_irq_table[irqno];
2086 
2087 	ASSERT(irqp->airq_mps_intr_index != FREE_INDEX);
2088 
2089 	intin = irqp->airq_intin_no;
2090 	ioapixindex = irqp->airq_ioapicindex;
2091 	mutex_exit(&airq_mutex);
2092 
2093 	iflag = intr_clear();
2094 	lock_set(&apic_ioapic_lock);
2095 
2096 	rdt_entry = READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapixindex, intin);
2097 
2098 	/* clear mask */
2099 	WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapixindex, intin,
2100 	    (AV_MASK | rdt_entry));
2101 
2102 	lock_clear(&apic_ioapic_lock);
2103 	intr_restore(iflag);
2104 }
2105 
2106 /*
2107  * This function will clear the mask for the interrupt on the I/O APIC
2108  */
2109 static void
2110 apix_intx_clear_mask(int irqno)
2111 {
2112 	int intin, ioapixindex, rdt_entry;
2113 	ulong_t iflag;
2114 	apic_irq_t *irqp;
2115 
2116 	mutex_enter(&airq_mutex);
2117 	irqp = apic_irq_table[irqno];
2118 
2119 	ASSERT(irqp->airq_mps_intr_index != FREE_INDEX);
2120 
2121 	intin = irqp->airq_intin_no;
2122 	ioapixindex = irqp->airq_ioapicindex;
2123 	mutex_exit(&airq_mutex);
2124 
2125 	iflag = intr_clear();
2126 	lock_set(&apic_ioapic_lock);
2127 
2128 	rdt_entry = READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapixindex, intin);
2129 
2130 	/* clear mask */
2131 	WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapixindex, intin,
2132 	    ((~AV_MASK) & rdt_entry));
2133 
2134 	lock_clear(&apic_ioapic_lock);
2135 	intr_restore(iflag);
2136 }
2137 
2138 /*
2139  * For level-triggered interrupt, mask the IRQ line. Mask means
2140  * new interrupts will not be delivered. The interrupt already
2141  * accepted by a local APIC is not affected
2142  */
2143 void
2144 apix_level_intr_pre_eoi(int irq)
2145 {
2146 	apic_irq_t *irqp = apic_irq_table[irq];
2147 	int apic_ix, intin_ix;
2148 
2149 	if (irqp == NULL)
2150 		return;
2151 
2152 	ASSERT(apic_level_intr[irq] == TRIGGER_MODE_LEVEL);
2153 
2154 	lock_set(&apic_ioapic_lock);
2155 
2156 	intin_ix = irqp->airq_intin_no;
2157 	apic_ix = irqp->airq_ioapicindex;
2158 
2159 	if (irqp->airq_cpu != CPU->cpu_id) {
2160 		if (!APIX_IS_MASK_RDT(apix_mul_ioapic_method))
2161 			ioapic_write_eoi(apic_ix, irqp->airq_vector);
2162 		lock_clear(&apic_ioapic_lock);
2163 		return;
2164 	}
2165 
2166 	if (apix_mul_ioapic_method == APIC_MUL_IOAPIC_IOXAPIC) {
2167 		/*
2168 		 * This is a IOxAPIC and there is EOI register:
2169 		 * 	Change the vector to reserved unused vector, so that
2170 		 * 	the EOI	from Local APIC won't clear the Remote IRR for
2171 		 * 	this level trigger interrupt. Instead, we'll manually
2172 		 * 	clear it in apix_post_hardint() after ISR handling.
2173 		 */
2174 		WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(apic_ix, intin_ix,
2175 		    (irqp->airq_rdt_entry & (~0xff)) | APIX_RESV_VECTOR);
2176 	} else {
2177 		WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(apic_ix, intin_ix,
2178 		    AV_MASK | irqp->airq_rdt_entry);
2179 	}
2180 
2181 	lock_clear(&apic_ioapic_lock);
2182 }
2183 
2184 /*
2185  * For level-triggered interrupt, unmask the IRQ line
2186  * or restore the original vector number.
2187  */
2188 void
2189 apix_level_intr_post_dispatch(int irq)
2190 {
2191 	apic_irq_t *irqp = apic_irq_table[irq];
2192 	int apic_ix, intin_ix;
2193 
2194 	if (irqp == NULL)
2195 		return;
2196 
2197 	lock_set(&apic_ioapic_lock);
2198 
2199 	intin_ix = irqp->airq_intin_no;
2200 	apic_ix = irqp->airq_ioapicindex;
2201 
2202 	if (APIX_IS_DIRECTED_EOI(apix_mul_ioapic_method)) {
2203 		/*
2204 		 * Already sent EOI back to Local APIC.
2205 		 * Send EOI to IO-APIC
2206 		 */
2207 		ioapic_write_eoi(apic_ix, irqp->airq_vector);
2208 	} else {
2209 		/* clear the mask or restore the vector */
2210 		WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(apic_ix, intin_ix,
2211 		    irqp->airq_rdt_entry);
2212 
2213 		/* send EOI to IOxAPIC */
2214 		if (apix_mul_ioapic_method == APIC_MUL_IOAPIC_IOXAPIC)
2215 			ioapic_write_eoi(apic_ix, irqp->airq_vector);
2216 	}
2217 
2218 	lock_clear(&apic_ioapic_lock);
2219 }
2220 
2221 static int
2222 apix_intx_get_shared(int irqno)
2223 {
2224 	apic_irq_t *irqp;
2225 	int share;
2226 
2227 	mutex_enter(&airq_mutex);
2228 	irqp = apic_irq_table[irqno];
2229 	if (IS_IRQ_FREE(irqp) || (irqp->airq_cpu == IRQ_UNINIT)) {
2230 		mutex_exit(&airq_mutex);
2231 		return (0);
2232 	}
2233 	share = irqp->airq_share;
2234 	mutex_exit(&airq_mutex);
2235 
2236 	return (share);
2237 }
2238 
2239 static void
2240 apix_intx_set_shared(int irqno, int delta)
2241 {
2242 	apic_irq_t *irqp;
2243 
2244 	mutex_enter(&airq_mutex);
2245 	irqp = apic_irq_table[irqno];
2246 	if (IS_IRQ_FREE(irqp)) {
2247 		mutex_exit(&airq_mutex);
2248 		return;
2249 	}
2250 	irqp->airq_share += delta;
2251 	mutex_exit(&airq_mutex);
2252 }
2253 
2254 /*
2255  * Setup IRQ table. Return IRQ no or -1 on failure
2256  */
2257 static int
2258 apix_intx_setup(dev_info_t *dip, int inum, int irqno,
2259     struct apic_io_intr *intrp, struct intrspec *ispec, iflag_t *iflagp)
2260 {
2261 	int origirq = ispec->intrspec_vec;
2262 	int newirq;
2263 	short intr_index;
2264 	uchar_t ipin, ioapic, ioapicindex;
2265 	apic_irq_t *irqp;
2266 
2267 	UNREFERENCED_1PARAMETER(inum);
2268 
2269 	if (intrp != NULL) {
2270 		intr_index = (short)(intrp - apic_io_intrp);
2271 		ioapic = intrp->intr_destid;
2272 		ipin = intrp->intr_destintin;
2273 
2274 		/* Find ioapicindex. If destid was ALL, we will exit with 0. */
2275 		for (ioapicindex = apic_io_max - 1; ioapicindex; ioapicindex--)
2276 			if (apic_io_id[ioapicindex] == ioapic)
2277 				break;
2278 		ASSERT((ioapic == apic_io_id[ioapicindex]) ||
2279 		    (ioapic == INTR_ALL_APIC));
2280 
2281 		/* check whether this intin# has been used by another irqno */
2282 		if ((newirq = apic_find_intin(ioapicindex, ipin)) != -1)
2283 			return (newirq);
2284 
2285 	} else if (iflagp != NULL) {	/* ACPI */
2286 		intr_index = ACPI_INDEX;
2287 		ioapicindex = acpi_find_ioapic(irqno);
2288 		ASSERT(ioapicindex != 0xFF);
2289 		ioapic = apic_io_id[ioapicindex];
2290 		ipin = irqno - apic_io_vectbase[ioapicindex];
2291 
2292 		if (apic_irq_table[irqno] &&
2293 		    apic_irq_table[irqno]->airq_mps_intr_index == ACPI_INDEX) {
2294 			ASSERT(apic_irq_table[irqno]->airq_intin_no == ipin &&
2295 			    apic_irq_table[irqno]->airq_ioapicindex ==
2296 			    ioapicindex);
2297 			return (irqno);
2298 		}
2299 
2300 	} else {	/* default configuration */
2301 		intr_index = DEFAULT_INDEX;
2302 		ioapicindex = 0;
2303 		ioapic = apic_io_id[ioapicindex];
2304 		ipin = (uchar_t)irqno;
2305 	}
2306 
2307 	/* allocate a new IRQ no */
2308 	if ((irqp = apic_irq_table[irqno]) == NULL) {
2309 		irqp = kmem_zalloc(sizeof (apic_irq_t), KM_SLEEP);
2310 		apic_irq_table[irqno] = irqp;
2311 	} else {
2312 		if (irqp->airq_mps_intr_index != FREE_INDEX) {
2313 			newirq = apic_allocate_irq(apic_first_avail_irq);
2314 			if (newirq == -1) {
2315 				return (-1);
2316 			}
2317 			irqno = newirq;
2318 			irqp = apic_irq_table[irqno];
2319 			ASSERT(irqp != NULL);
2320 		}
2321 	}
2322 	apic_max_device_irq = max(irqno, apic_max_device_irq);
2323 	apic_min_device_irq = min(irqno, apic_min_device_irq);
2324 
2325 	irqp->airq_mps_intr_index = intr_index;
2326 	irqp->airq_ioapicindex = ioapicindex;
2327 	irqp->airq_intin_no = ipin;
2328 	irqp->airq_dip = dip;
2329 	irqp->airq_origirq = (uchar_t)origirq;
2330 	if (iflagp != NULL)
2331 		irqp->airq_iflag = *iflagp;
2332 	irqp->airq_cpu = IRQ_UNINIT;
2333 	irqp->airq_vector = 0;
2334 
2335 	return (irqno);
2336 }
2337 
2338 /*
2339  * Setup IRQ table for non-pci devices. Return IRQ no or -1 on error
2340  */
2341 static int
2342 apix_intx_setup_nonpci(dev_info_t *dip, int inum, int bustype,
2343     struct intrspec *ispec)
2344 {
2345 	int irqno = ispec->intrspec_vec;
2346 	int newirq, i;
2347 	iflag_t intr_flag;
2348 	ACPI_SUBTABLE_HEADER	*hp;
2349 	ACPI_MADT_INTERRUPT_OVERRIDE *isop;
2350 	struct apic_io_intr *intrp;
2351 
2352 	if (!apic_enable_acpi || apic_use_acpi_madt_only) {
2353 		int busid;
2354 
2355 		if (bustype == 0)
2356 			bustype = eisa_level_intr_mask ? BUS_EISA : BUS_ISA;
2357 
2358 		/* loop checking BUS_ISA/BUS_EISA */
2359 		for (i = 0; i < 2; i++) {
2360 			if (((busid = apic_find_bus_id(bustype)) != -1) &&
2361 			    ((intrp = apic_find_io_intr_w_busid(irqno, busid))
2362 			    != NULL)) {
2363 				return (apix_intx_setup(dip, inum, irqno,
2364 				    intrp, ispec, NULL));
2365 			}
2366 			bustype = (bustype == BUS_EISA) ? BUS_ISA : BUS_EISA;
2367 		}
2368 
2369 		/* fall back to default configuration */
2370 		return (-1);
2371 	}
2372 
2373 	/* search iso entries first */
2374 	if (acpi_iso_cnt != 0) {
2375 		hp = (ACPI_SUBTABLE_HEADER *)acpi_isop;
2376 		i = 0;
2377 		while (i < acpi_iso_cnt) {
2378 			if (hp->Type == ACPI_MADT_TYPE_INTERRUPT_OVERRIDE) {
2379 				isop = (ACPI_MADT_INTERRUPT_OVERRIDE *) hp;
2380 				if (isop->Bus == 0 &&
2381 				    isop->SourceIrq == irqno) {
2382 					newirq = isop->GlobalIrq;
2383 					intr_flag.intr_po = isop->IntiFlags &
2384 					    ACPI_MADT_POLARITY_MASK;
2385 					intr_flag.intr_el = (isop->IntiFlags &
2386 					    ACPI_MADT_TRIGGER_MASK) >> 2;
2387 					intr_flag.bustype = BUS_ISA;
2388 
2389 					return (apix_intx_setup(dip, inum,
2390 					    newirq, NULL, ispec, &intr_flag));
2391 				}
2392 				i++;
2393 			}
2394 			hp = (ACPI_SUBTABLE_HEADER *)(((char *)hp) +
2395 			    hp->Length);
2396 		}
2397 	}
2398 	intr_flag.intr_po = INTR_PO_ACTIVE_HIGH;
2399 	intr_flag.intr_el = INTR_EL_EDGE;
2400 	intr_flag.bustype = BUS_ISA;
2401 	return (apix_intx_setup(dip, inum, irqno, NULL, ispec, &intr_flag));
2402 }
2403 
2404 
2405 /*
2406  * Setup IRQ table for pci devices. Return IRQ no or -1 on error
2407  */
2408 static int
2409 apix_intx_setup_pci(dev_info_t *dip, int inum, int bustype,
2410     struct intrspec *ispec)
2411 {
2412 	int busid, devid, pci_irq;
2413 	ddi_acc_handle_t cfg_handle;
2414 	uchar_t ipin;
2415 	iflag_t intr_flag;
2416 	struct apic_io_intr *intrp;
2417 
2418 	if (acpica_get_bdf(dip, &busid, &devid, NULL) != 0)
2419 		return (-1);
2420 
2421 	if (busid == 0 && apic_pci_bus_total == 1)
2422 		busid = (int)apic_single_pci_busid;
2423 
2424 	if (pci_config_setup(dip, &cfg_handle) != DDI_SUCCESS)
2425 		return (-1);
2426 	ipin = pci_config_get8(cfg_handle, PCI_CONF_IPIN) - PCI_INTA;
2427 	pci_config_teardown(&cfg_handle);
2428 
2429 	if (apic_enable_acpi && !apic_use_acpi_madt_only) {	/* ACPI */
2430 		if (apic_acpi_translate_pci_irq(dip, busid, devid,
2431 		    ipin, &pci_irq, &intr_flag) != ACPI_PSM_SUCCESS)
2432 			return (-1);
2433 
2434 		intr_flag.bustype = (uchar_t)bustype;
2435 		return (apix_intx_setup(dip, inum, pci_irq, NULL, ispec,
2436 		    &intr_flag));
2437 	}
2438 
2439 	/* MP configuration table */
2440 	pci_irq = ((devid & 0x1f) << 2) | (ipin & 0x3);
2441 	if ((intrp = apic_find_io_intr_w_busid(pci_irq, busid)) == NULL) {
2442 		pci_irq = apic_handle_pci_pci_bridge(dip, devid, ipin, &intrp);
2443 		if (pci_irq == -1)
2444 			return (-1);
2445 	}
2446 
2447 	return (apix_intx_setup(dip, inum, pci_irq, intrp, ispec, NULL));
2448 }
2449 
2450 /*
2451  * Translate and return IRQ no
2452  */
2453 static int
2454 apix_intx_xlate_irq(dev_info_t *dip, int inum, struct intrspec *ispec)
2455 {
2456 	int newirq, irqno = ispec->intrspec_vec;
2457 	int parent_is_pci_or_pciex = 0, child_is_pciex = 0;
2458 	int bustype = 0, dev_len;
2459 	char dev_type[16];
2460 
2461 	if (apic_defconf) {
2462 		mutex_enter(&airq_mutex);
2463 		goto defconf;
2464 	}
2465 
2466 	if ((dip == NULL) || (!apic_irq_translate && !apic_enable_acpi)) {
2467 		mutex_enter(&airq_mutex);
2468 		goto nonpci;
2469 	}
2470 
2471 	/*
2472 	 * use ddi_getlongprop_buf() instead of ddi_prop_lookup_string()
2473 	 * to avoid extra buffer allocation.
2474 	 */
2475 	dev_len = sizeof (dev_type);
2476 	if (ddi_getlongprop_buf(DDI_DEV_T_ANY, ddi_get_parent(dip),
2477 	    DDI_PROP_DONTPASS, "device_type", (caddr_t)dev_type,
2478 	    &dev_len) == DDI_PROP_SUCCESS) {
2479 		if ((strcmp(dev_type, "pci") == 0) ||
2480 		    (strcmp(dev_type, "pciex") == 0))
2481 			parent_is_pci_or_pciex = 1;
2482 	}
2483 
2484 	if (ddi_getlongprop_buf(DDI_DEV_T_ANY, dip,
2485 	    DDI_PROP_DONTPASS, "compatible", (caddr_t)dev_type,
2486 	    &dev_len) == DDI_PROP_SUCCESS) {
2487 		if (strstr(dev_type, "pciex"))
2488 			child_is_pciex = 1;
2489 	}
2490 
2491 	mutex_enter(&airq_mutex);
2492 
2493 	if (parent_is_pci_or_pciex) {
2494 		bustype = child_is_pciex ? BUS_PCIE : BUS_PCI;
2495 		newirq = apix_intx_setup_pci(dip, inum, bustype, ispec);
2496 		if (newirq != -1)
2497 			goto done;
2498 		bustype = 0;
2499 	} else if (strcmp(dev_type, "isa") == 0)
2500 		bustype = BUS_ISA;
2501 	else if (strcmp(dev_type, "eisa") == 0)
2502 		bustype = BUS_EISA;
2503 
2504 nonpci:
2505 	newirq = apix_intx_setup_nonpci(dip, inum, bustype, ispec);
2506 	if (newirq != -1)
2507 		goto done;
2508 
2509 defconf:
2510 	newirq = apix_intx_setup(dip, inum, irqno, NULL, ispec, NULL);
2511 	if (newirq == -1) {
2512 		mutex_exit(&airq_mutex);
2513 		return (-1);
2514 	}
2515 done:
2516 	ASSERT(apic_irq_table[newirq]);
2517 	mutex_exit(&airq_mutex);
2518 	return (newirq);
2519 }
2520 
2521 static int
2522 apix_intx_alloc_vector(dev_info_t *dip, int inum, struct intrspec *ispec)
2523 {
2524 	int irqno;
2525 	apix_vector_t *vecp;
2526 
2527 	if ((irqno = apix_intx_xlate_irq(dip, inum, ispec)) == -1)
2528 		return (0);
2529 
2530 	if ((vecp = apix_alloc_intx(dip, inum, irqno)) == NULL)
2531 		return (0);
2532 
2533 	DDI_INTR_IMPLDBG((CE_CONT, "apix_intx_alloc_vector: dip=0x%p name=%s "
2534 	    "irqno=0x%x cpuid=%d vector=0x%x\n",
2535 	    (void *)dip, ddi_driver_name(dip), irqno,
2536 	    vecp->v_cpuid, vecp->v_vector));
2537 
2538 	return (1);
2539 }
2540 
2541 /*
2542  * Return the vector number if the translated IRQ for this device
2543  * has a vector mapping setup. If no IRQ setup exists or no vector is
2544  * allocated to it then return 0.
2545  */
2546 static apix_vector_t *
2547 apix_intx_xlate_vector(dev_info_t *dip, int inum, struct intrspec *ispec)
2548 {
2549 	int irqno;
2550 	apix_vector_t *vecp;
2551 
2552 	/* get the IRQ number */
2553 	if ((irqno = apix_intx_xlate_irq(dip, inum, ispec)) == -1)
2554 		return (NULL);
2555 
2556 	/* get the vector number if a vector is allocated to this irqno */
2557 	vecp = apix_intx_get_vector(irqno);
2558 
2559 	return (vecp);
2560 }
2561 
2562 /* stub function */
2563 int
2564 apix_loaded(void)
2565 {
2566 	return (apix_is_enabled);
2567 }
2568