xref: /illumos-gate/usr/src/uts/i86pc/io/apix/apix.c (revision e153cda9f9660e385e8f468253f80e59f5d454d7)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
24  */
25 /*
26  * Copyright (c) 2010, Intel Corporation.
27  * All rights reserved.
28  */
29 /*
30  * Copyright (c) 2017, Joyent, Inc.  All rights reserved.
31  */
32 
33 /*
34  * To understand how the apix module interacts with the interrupt subsystem read
35  * the theory statement in uts/i86pc/os/intr.c.
36  */
37 
38 /*
39  * PSMI 1.1 extensions are supported only in 2.6 and later versions.
40  * PSMI 1.2 extensions are supported only in 2.7 and later versions.
41  * PSMI 1.3 and 1.4 extensions are supported in Solaris 10.
42  * PSMI 1.5 extensions are supported in Solaris Nevada.
43  * PSMI 1.6 extensions are supported in Solaris Nevada.
44  * PSMI 1.7 extensions are supported in Solaris Nevada.
45  */
46 #define	PSMI_1_7
47 
48 #include <sys/processor.h>
49 #include <sys/time.h>
50 #include <sys/psm.h>
51 #include <sys/smp_impldefs.h>
52 #include <sys/cram.h>
53 #include <sys/acpi/acpi.h>
54 #include <sys/acpica.h>
55 #include <sys/psm_common.h>
56 #include <sys/pit.h>
57 #include <sys/ddi.h>
58 #include <sys/sunddi.h>
59 #include <sys/ddi_impldefs.h>
60 #include <sys/pci.h>
61 #include <sys/promif.h>
62 #include <sys/x86_archext.h>
63 #include <sys/cpc_impl.h>
64 #include <sys/uadmin.h>
65 #include <sys/panic.h>
66 #include <sys/debug.h>
67 #include <sys/archsystm.h>
68 #include <sys/trap.h>
69 #include <sys/machsystm.h>
70 #include <sys/sysmacros.h>
71 #include <sys/cpuvar.h>
72 #include <sys/rm_platter.h>
73 #include <sys/privregs.h>
74 #include <sys/note.h>
75 #include <sys/pci_intr_lib.h>
76 #include <sys/spl.h>
77 #include <sys/clock.h>
78 #include <sys/cyclic.h>
79 #include <sys/dditypes.h>
80 #include <sys/sunddi.h>
81 #include <sys/x_call.h>
82 #include <sys/reboot.h>
83 #include <sys/mach_intr.h>
84 #include <sys/apix.h>
85 #include <sys/apix_irm_impl.h>
86 
87 static int apix_probe();
88 static void apix_init();
89 static void apix_picinit(void);
90 static int apix_intr_enter(int, int *);
91 static void apix_intr_exit(int, int);
92 static void apix_setspl(int);
93 static int apix_disable_intr(processorid_t);
94 static void apix_enable_intr(processorid_t);
95 static int apix_get_clkvect(int);
96 static int apix_get_ipivect(int, int);
97 static void apix_post_cyclic_setup(void *);
98 static int apix_post_cpu_start();
99 static int apix_intr_ops(dev_info_t *, ddi_intr_handle_impl_t *,
100     psm_intr_op_t, int *);
101 
102 /*
103  * Helper functions for apix_intr_ops()
104  */
105 static void apix_redistribute_compute(void);
106 static int apix_get_pending(apix_vector_t *);
107 static apix_vector_t *apix_get_req_vector(ddi_intr_handle_impl_t *, ushort_t);
108 static int apix_get_intr_info(ddi_intr_handle_impl_t *, apic_get_intr_t *);
109 static char *apix_get_apic_type(void);
110 static int apix_intx_get_pending(int);
111 static void apix_intx_set_mask(int irqno);
112 static void apix_intx_clear_mask(int irqno);
113 static int apix_intx_get_shared(int irqno);
114 static void apix_intx_set_shared(int irqno, int delta);
115 static apix_vector_t *apix_intx_xlate_vector(dev_info_t *, int,
116     struct intrspec *);
117 static int apix_intx_alloc_vector(dev_info_t *, int, struct intrspec *);
118 
119 extern int apic_clkinit(int);
120 
121 /* IRM initialization for APIX PSM module */
122 extern void apix_irm_init(void);
123 
124 extern int irm_enable;
125 
126 /*
127  *	Local static data
128  */
129 static struct	psm_ops apix_ops = {
130 	apix_probe,
131 
132 	apix_init,
133 	apix_picinit,
134 	apix_intr_enter,
135 	apix_intr_exit,
136 	apix_setspl,
137 	apix_addspl,
138 	apix_delspl,
139 	apix_disable_intr,
140 	apix_enable_intr,
141 	NULL,			/* psm_softlvl_to_irq */
142 	NULL,			/* psm_set_softintr */
143 
144 	apic_set_idlecpu,
145 	apic_unset_idlecpu,
146 
147 	apic_clkinit,
148 	apix_get_clkvect,
149 	NULL,			/* psm_hrtimeinit */
150 	apic_gethrtime,
151 
152 	apic_get_next_processorid,
153 	apic_cpu_start,
154 	apix_post_cpu_start,
155 	apic_shutdown,
156 	apix_get_ipivect,
157 	apic_send_ipi,
158 
159 	NULL,			/* psm_translate_irq */
160 	NULL,			/* psm_notify_error */
161 	NULL,			/* psm_notify_func */
162 	apic_timer_reprogram,
163 	apic_timer_enable,
164 	apic_timer_disable,
165 	apix_post_cyclic_setup,
166 	apic_preshutdown,
167 	apix_intr_ops,		/* Advanced DDI Interrupt framework */
168 	apic_state,		/* save, restore apic state for S3 */
169 	apic_cpu_ops,		/* CPU control interface. */
170 };
171 
172 struct psm_ops *psmops = &apix_ops;
173 
174 static struct	psm_info apix_psm_info = {
175 	PSM_INFO_VER01_7,			/* version */
176 	PSM_OWN_EXCLUSIVE,			/* ownership */
177 	&apix_ops,				/* operation */
178 	APIX_NAME,				/* machine name */
179 	"apix MPv1.4 compatible",
180 };
181 
182 static void *apix_hdlp;
183 
184 static int apix_is_enabled = 0;
185 
186 /*
187  * Flag to indicate if APIX is to be enabled only for platforms
188  * with specific hw feature(s).
189  */
190 int apix_hw_chk_enable = 1;
191 
192 /*
193  * Hw features that are checked for enabling APIX support.
194  */
195 #define	APIX_SUPPORT_X2APIC	0x00000001
196 uint_t apix_supported_hw = APIX_SUPPORT_X2APIC;
197 
198 /*
199  * apix_lock is used for cpu selection and vector re-binding
200  */
201 lock_t apix_lock;
202 apix_impl_t *apixs[NCPU];
203 /*
204  * Mapping between device interrupt and the allocated vector. Indexed
205  * by major number.
206  */
207 apix_dev_vector_t **apix_dev_vector;
208 /*
209  * Mapping between device major number and cpu id. It gets used
210  * when interrupt binding policy round robin with affinity is
211  * applied. With that policy, devices with the same major number
212  * will be bound to the same CPU.
213  */
214 processorid_t *apix_major_to_cpu;	/* major to cpu mapping */
215 kmutex_t apix_mutex;	/* for apix_dev_vector & apix_major_to_cpu */
216 
217 int apix_nipis = 16;	/* Maximum number of IPIs */
218 /*
219  * Maximum number of vectors in a CPU that can be used for interrupt
220  * allocation (including IPIs and the reserved vectors).
221  */
222 int apix_cpu_nvectors = APIX_NVECTOR;
223 
224 /* number of CPUs in power-on transition state */
225 static int apic_poweron_cnt = 0;
226 
227 /* gcpu.h */
228 
229 extern void apic_do_interrupt(struct regs *rp, trap_trace_rec_t *ttp);
230 extern void apic_change_eoi();
231 
232 /*
233  *	This is the loadable module wrapper
234  */
235 
236 int
237 _init(void)
238 {
239 	if (apic_coarse_hrtime)
240 		apix_ops.psm_gethrtime = &apic_gettime;
241 	return (psm_mod_init(&apix_hdlp, &apix_psm_info));
242 }
243 
244 int
245 _fini(void)
246 {
247 	return (psm_mod_fini(&apix_hdlp, &apix_psm_info));
248 }
249 
250 int
251 _info(struct modinfo *modinfop)
252 {
253 	return (psm_mod_info(&apix_hdlp, &apix_psm_info, modinfop));
254 }
255 
256 static int
257 apix_probe()
258 {
259 	int rval;
260 
261 	if (apix_enable == 0)
262 		return (PSM_FAILURE);
263 
264 	/*
265 	 * FIXME Temporarily disable apix module on Xen HVM platform due to
266 	 * known hang during boot (see #3605).
267 	 *
268 	 * Please remove when/if the issue is resolved.
269 	 */
270 	if (get_hwenv() & HW_XEN_HVM)
271 		return (PSM_FAILURE);
272 
273 	/* check for hw features if specified  */
274 	if (apix_hw_chk_enable) {
275 		/* check if x2APIC mode is supported */
276 		if ((apix_supported_hw & APIX_SUPPORT_X2APIC) ==
277 		    APIX_SUPPORT_X2APIC) {
278 			if (apic_local_mode() == LOCAL_X2APIC) {
279 				/* x2APIC mode activated by BIOS, switch ops */
280 				apic_mode = LOCAL_X2APIC;
281 				apic_change_ops();
282 			} else if (!apic_detect_x2apic()) {
283 				/* x2APIC mode is not supported in the hw */
284 				apix_enable = 0;
285 			}
286 		}
287 		if (apix_enable == 0)
288 			return (PSM_FAILURE);
289 	}
290 
291 	rval = apic_probe_common(apix_psm_info.p_mach_idstring);
292 	if (rval == PSM_SUCCESS)
293 		apix_is_enabled = 1;
294 	else
295 		apix_is_enabled = 0;
296 	return (rval);
297 }
298 
299 /*
300  * Initialize the data structures needed by pcplusmpx module.
301  * Specifically, the data structures used by addspl() and delspl()
302  * routines.
303  */
304 static void
305 apix_softinit()
306 {
307 	int i, *iptr;
308 	apix_impl_t *hdlp;
309 	int nproc;
310 
311 	nproc = max(apic_nproc, apic_max_nproc);
312 
313 	hdlp = kmem_zalloc(nproc * sizeof (apix_impl_t), KM_SLEEP);
314 	for (i = 0; i < nproc; i++) {
315 		apixs[i] = &hdlp[i];
316 		apixs[i]->x_cpuid = i;
317 		LOCK_INIT_CLEAR(&apixs[i]->x_lock);
318 	}
319 
320 	/* cpu 0 is always up (for now) */
321 	apic_cpus[0].aci_status = APIC_CPU_ONLINE | APIC_CPU_INTR_ENABLE;
322 
323 	iptr = (int *)&apic_irq_table[0];
324 	for (i = 0; i <= APIC_MAX_VECTOR; i++) {
325 		apic_level_intr[i] = 0;
326 		*iptr++ = NULL;
327 	}
328 	mutex_init(&airq_mutex, NULL, MUTEX_DEFAULT, NULL);
329 
330 	apix_dev_vector = kmem_zalloc(sizeof (apix_dev_vector_t *) * devcnt,
331 	    KM_SLEEP);
332 
333 	if (apic_intr_policy == INTR_ROUND_ROBIN_WITH_AFFINITY) {
334 		apix_major_to_cpu = kmem_zalloc(sizeof (int) * devcnt,
335 		    KM_SLEEP);
336 		for (i = 0; i < devcnt; i++)
337 			apix_major_to_cpu[i] = IRQ_UNINIT;
338 	}
339 
340 	mutex_init(&apix_mutex, NULL, MUTEX_DEFAULT, NULL);
341 }
342 
343 static int
344 apix_get_pending_spl(void)
345 {
346 	int cpuid = CPU->cpu_id;
347 
348 	return (bsrw_insn(apixs[cpuid]->x_intr_pending));
349 }
350 
351 static uintptr_t
352 apix_get_intr_handler(int cpu, short vec)
353 {
354 	apix_vector_t *apix_vector;
355 
356 	ASSERT(cpu < apic_nproc && vec < APIX_NVECTOR);
357 	if (cpu >= apic_nproc)
358 		return (NULL);
359 
360 	apix_vector = apixs[cpu]->x_vectbl[vec];
361 
362 	return ((uintptr_t)(apix_vector->v_autovect));
363 }
364 
365 static void
366 apix_init()
367 {
368 	extern void (*do_interrupt_common)(struct regs *, trap_trace_rec_t *);
369 
370 	APIC_VERBOSE(INIT, (CE_CONT, "apix: psm_softinit\n"));
371 
372 	do_interrupt_common = apix_do_interrupt;
373 	addintr = apix_add_avintr;
374 	remintr = apix_rem_avintr;
375 	get_pending_spl = apix_get_pending_spl;
376 	get_intr_handler = apix_get_intr_handler;
377 	psm_get_localapicid = apic_get_localapicid;
378 	psm_get_ioapicid = apic_get_ioapicid;
379 
380 	apix_softinit();
381 
382 #if !defined(__amd64)
383 	if (cpuid_have_cr8access(CPU))
384 		apic_have_32bit_cr8 = 1;
385 #endif
386 
387 	/*
388 	 * Initialize IRM pool parameters
389 	 */
390 	if (irm_enable) {
391 		int	i;
392 		int	lowest_irq;
393 		int	highest_irq;
394 
395 		/* number of CPUs present */
396 		apix_irminfo.apix_ncpus = apic_nproc;
397 		/* total number of entries in all of the IOAPICs present */
398 		lowest_irq = apic_io_vectbase[0];
399 		highest_irq = apic_io_vectend[0];
400 		for (i = 1; i < apic_io_max; i++) {
401 			if (apic_io_vectbase[i] < lowest_irq)
402 				lowest_irq = apic_io_vectbase[i];
403 			if (apic_io_vectend[i] > highest_irq)
404 				highest_irq = apic_io_vectend[i];
405 		}
406 		apix_irminfo.apix_ioapic_max_vectors =
407 		    highest_irq - lowest_irq + 1;
408 		/*
409 		 * Number of available per-CPU vectors excluding
410 		 * reserved vectors for Dtrace, int80, system-call,
411 		 * fast-trap, etc.
412 		 */
413 		apix_irminfo.apix_per_cpu_vectors = APIX_NAVINTR -
414 		    APIX_SW_RESERVED_VECTORS;
415 
416 		/* Number of vectors (pre) allocated (SCI and HPET) */
417 		apix_irminfo.apix_vectors_allocated = 0;
418 		if (apic_hpet_vect != -1)
419 			apix_irminfo.apix_vectors_allocated++;
420 		if (apic_sci_vect != -1)
421 			apix_irminfo.apix_vectors_allocated++;
422 	}
423 }
424 
425 static void
426 apix_init_intr()
427 {
428 	processorid_t	cpun = psm_get_cpu_id();
429 	uint_t nlvt;
430 	uint32_t svr = AV_UNIT_ENABLE | APIC_SPUR_INTR;
431 	extern void cmi_cmci_trap(void);
432 
433 	apic_reg_ops->apic_write_task_reg(APIC_MASK_ALL);
434 
435 	if (apic_mode == LOCAL_APIC) {
436 		/*
437 		 * We are running APIC in MMIO mode.
438 		 */
439 		if (apic_flat_model) {
440 			apic_reg_ops->apic_write(APIC_FORMAT_REG,
441 			    APIC_FLAT_MODEL);
442 		} else {
443 			apic_reg_ops->apic_write(APIC_FORMAT_REG,
444 			    APIC_CLUSTER_MODEL);
445 		}
446 
447 		apic_reg_ops->apic_write(APIC_DEST_REG,
448 		    AV_HIGH_ORDER >> cpun);
449 	}
450 
451 	if (apic_directed_EOI_supported()) {
452 		/*
453 		 * Setting the 12th bit in the Spurious Interrupt Vector
454 		 * Register suppresses broadcast EOIs generated by the local
455 		 * APIC. The suppression of broadcast EOIs happens only when
456 		 * interrupts are level-triggered.
457 		 */
458 		svr |= APIC_SVR_SUPPRESS_BROADCAST_EOI;
459 	}
460 
461 	/* need to enable APIC before unmasking NMI */
462 	apic_reg_ops->apic_write(APIC_SPUR_INT_REG, svr);
463 
464 	/*
465 	 * Presence of an invalid vector with delivery mode AV_FIXED can
466 	 * cause an error interrupt, even if the entry is masked...so
467 	 * write a valid vector to LVT entries along with the mask bit
468 	 */
469 
470 	/* All APICs have timer and LINT0/1 */
471 	apic_reg_ops->apic_write(APIC_LOCAL_TIMER, AV_MASK|APIC_RESV_IRQ);
472 	apic_reg_ops->apic_write(APIC_INT_VECT0, AV_MASK|APIC_RESV_IRQ);
473 	apic_reg_ops->apic_write(APIC_INT_VECT1, AV_NMI);	/* enable NMI */
474 
475 	/*
476 	 * On integrated APICs, the number of LVT entries is
477 	 * 'Max LVT entry' + 1; on 82489DX's (non-integrated
478 	 * APICs), nlvt is "3" (LINT0, LINT1, and timer)
479 	 */
480 
481 	if (apic_cpus[cpun].aci_local_ver < APIC_INTEGRATED_VERS) {
482 		nlvt = 3;
483 	} else {
484 		nlvt = ((apic_reg_ops->apic_read(APIC_VERS_REG) >> 16) &
485 		    0xFF) + 1;
486 	}
487 
488 	if (nlvt >= 5) {
489 		/* Enable performance counter overflow interrupt */
490 
491 		if (!is_x86_feature(x86_featureset, X86FSET_MSR))
492 			apic_enable_cpcovf_intr = 0;
493 		if (apic_enable_cpcovf_intr) {
494 			if (apic_cpcovf_vect == 0) {
495 				int ipl = APIC_PCINT_IPL;
496 
497 				apic_cpcovf_vect = apix_get_ipivect(ipl, -1);
498 				ASSERT(apic_cpcovf_vect);
499 
500 				(void) add_avintr(NULL, ipl,
501 				    (avfunc)kcpc_hw_overflow_intr,
502 				    "apic pcint", apic_cpcovf_vect,
503 				    NULL, NULL, NULL, NULL);
504 				kcpc_hw_overflow_intr_installed = 1;
505 				kcpc_hw_enable_cpc_intr =
506 				    apic_cpcovf_mask_clear;
507 			}
508 			apic_reg_ops->apic_write(APIC_PCINT_VECT,
509 			    apic_cpcovf_vect);
510 		}
511 	}
512 
513 	if (nlvt >= 6) {
514 		/* Only mask TM intr if the BIOS apparently doesn't use it */
515 
516 		uint32_t lvtval;
517 
518 		lvtval = apic_reg_ops->apic_read(APIC_THERM_VECT);
519 		if (((lvtval & AV_MASK) == AV_MASK) ||
520 		    ((lvtval & AV_DELIV_MODE) != AV_SMI)) {
521 			apic_reg_ops->apic_write(APIC_THERM_VECT,
522 			    AV_MASK|APIC_RESV_IRQ);
523 		}
524 	}
525 
526 	/* Enable error interrupt */
527 
528 	if (nlvt >= 4 && apic_enable_error_intr) {
529 		if (apic_errvect == 0) {
530 			int ipl = 0xf;	/* get highest priority intr */
531 			apic_errvect = apix_get_ipivect(ipl, -1);
532 			ASSERT(apic_errvect);
533 			/*
534 			 * Not PSMI compliant, but we are going to merge
535 			 * with ON anyway
536 			 */
537 			(void) add_avintr(NULL, ipl,
538 			    (avfunc)apic_error_intr, "apic error intr",
539 			    apic_errvect, NULL, NULL, NULL, NULL);
540 		}
541 		apic_reg_ops->apic_write(APIC_ERR_VECT, apic_errvect);
542 		apic_reg_ops->apic_write(APIC_ERROR_STATUS, 0);
543 		apic_reg_ops->apic_write(APIC_ERROR_STATUS, 0);
544 	}
545 
546 	/* Enable CMCI interrupt */
547 	if (cmi_enable_cmci) {
548 		mutex_enter(&cmci_cpu_setup_lock);
549 		if (cmci_cpu_setup_registered == 0) {
550 			mutex_enter(&cpu_lock);
551 			register_cpu_setup_func(cmci_cpu_setup, NULL);
552 			mutex_exit(&cpu_lock);
553 			cmci_cpu_setup_registered = 1;
554 		}
555 		mutex_exit(&cmci_cpu_setup_lock);
556 
557 		if (apic_cmci_vect == 0) {
558 			int ipl = 0x2;
559 			apic_cmci_vect = apix_get_ipivect(ipl, -1);
560 			ASSERT(apic_cmci_vect);
561 
562 			(void) add_avintr(NULL, ipl,
563 			    (avfunc)cmi_cmci_trap, "apic cmci intr",
564 			    apic_cmci_vect, NULL, NULL, NULL, NULL);
565 		}
566 		apic_reg_ops->apic_write(APIC_CMCI_VECT, apic_cmci_vect);
567 	}
568 
569 	apic_reg_ops->apic_write_task_reg(0);
570 }
571 
572 static void
573 apix_picinit(void)
574 {
575 	int i, j;
576 	uint_t isr;
577 
578 	APIC_VERBOSE(INIT, (CE_CONT, "apix: psm_picinit\n"));
579 
580 	/*
581 	 * initialize interrupt remapping before apic
582 	 * hardware initialization
583 	 */
584 	apic_intrmap_init(apic_mode);
585 	if (apic_vt_ops == psm_vt_ops)
586 		apix_mul_ioapic_method = APIC_MUL_IOAPIC_IIR;
587 
588 	/*
589 	 * On UniSys Model 6520, the BIOS leaves vector 0x20 isr
590 	 * bit on without clearing it with EOI.  Since softint
591 	 * uses vector 0x20 to interrupt itself, so softint will
592 	 * not work on this machine.  In order to fix this problem
593 	 * a check is made to verify all the isr bits are clear.
594 	 * If not, EOIs are issued to clear the bits.
595 	 */
596 	for (i = 7; i >= 1; i--) {
597 		isr = apic_reg_ops->apic_read(APIC_ISR_REG + (i * 4));
598 		if (isr != 0)
599 			for (j = 0; ((j < 32) && (isr != 0)); j++)
600 				if (isr & (1 << j)) {
601 					apic_reg_ops->apic_write(
602 					    APIC_EOI_REG, 0);
603 					isr &= ~(1 << j);
604 					apic_error |= APIC_ERR_BOOT_EOI;
605 				}
606 	}
607 
608 	/* set a flag so we know we have run apic_picinit() */
609 	apic_picinit_called = 1;
610 	LOCK_INIT_CLEAR(&apic_gethrtime_lock);
611 	LOCK_INIT_CLEAR(&apic_ioapic_lock);
612 	LOCK_INIT_CLEAR(&apic_error_lock);
613 	LOCK_INIT_CLEAR(&apic_mode_switch_lock);
614 
615 	picsetup();	 /* initialise the 8259 */
616 
617 	/* add nmi handler - least priority nmi handler */
618 	LOCK_INIT_CLEAR(&apic_nmi_lock);
619 
620 	if (!psm_add_nmintr(0, (avfunc) apic_nmi_intr,
621 	    "apix NMI handler", (caddr_t)NULL))
622 		cmn_err(CE_WARN, "apix: Unable to add nmi handler");
623 
624 	apix_init_intr();
625 
626 	/* enable apic mode if imcr present */
627 	if (apic_imcrp) {
628 		outb(APIC_IMCR_P1, (uchar_t)APIC_IMCR_SELECT);
629 		outb(APIC_IMCR_P2, (uchar_t)APIC_IMCR_APIC);
630 	}
631 
632 	ioapix_init_intr(IOAPIC_MASK);
633 
634 	/* setup global IRM pool if applicable */
635 	if (irm_enable)
636 		apix_irm_init();
637 }
638 
639 static __inline__ void
640 apix_send_eoi(void)
641 {
642 	if (apic_mode == LOCAL_APIC)
643 		LOCAL_APIC_WRITE_REG(APIC_EOI_REG, 0);
644 	else
645 		X2APIC_WRITE(APIC_EOI_REG, 0);
646 }
647 
648 /*
649  * platform_intr_enter
650  *
651  *	Called at the beginning of the interrupt service routine, but unlike
652  *	pcplusmp, does not mask interrupts. An EOI is given to the interrupt
653  *	controller to enable other HW interrupts but interrupts are still
654  * 	masked by the IF flag.
655  *
656  *	Return -1 for spurious interrupts
657  *
658  */
659 static int
660 apix_intr_enter(int ipl, int *vectorp)
661 {
662 	struct cpu *cpu = CPU;
663 	uint32_t cpuid = CPU->cpu_id;
664 	apic_cpus_info_t *cpu_infop;
665 	uchar_t vector;
666 	apix_vector_t *vecp;
667 	int nipl = -1;
668 
669 	/*
670 	 * The real vector delivered is (*vectorp + 0x20), but our caller
671 	 * subtracts 0x20 from the vector before passing it to us.
672 	 * (That's why APIC_BASE_VECT is 0x20.)
673 	 */
674 	vector = *vectorp = (uchar_t)*vectorp + APIC_BASE_VECT;
675 
676 	cpu_infop = &apic_cpus[cpuid];
677 	if (vector == APIC_SPUR_INTR) {
678 		cpu_infop->aci_spur_cnt++;
679 		return (APIC_INT_SPURIOUS);
680 	}
681 
682 	vecp = xv_vector(cpuid, vector);
683 	if (vecp == NULL) {
684 		if (APIX_IS_FAKE_INTR(vector))
685 			nipl = apix_rebindinfo.i_pri;
686 		apix_send_eoi();
687 		return (nipl);
688 	}
689 	nipl = vecp->v_pri;
690 
691 	/* if interrupted by the clock, increment apic_nsec_since_boot */
692 	if (vector == (apic_clkvect + APIC_BASE_VECT)) {
693 		if (!apic_oneshot) {
694 			/* NOTE: this is not MT aware */
695 			apic_hrtime_stamp++;
696 			apic_nsec_since_boot += apic_nsec_per_intr;
697 			apic_hrtime_stamp++;
698 			last_count_read = apic_hertz_count;
699 			apix_redistribute_compute();
700 		}
701 
702 		apix_send_eoi();
703 
704 		return (nipl);
705 	}
706 
707 	ASSERT(vecp->v_state != APIX_STATE_OBSOLETED);
708 
709 	/* pre-EOI handling for level-triggered interrupts */
710 	if (!APIX_IS_DIRECTED_EOI(apix_mul_ioapic_method) &&
711 	    (vecp->v_type & APIX_TYPE_FIXED) && apic_level_intr[vecp->v_inum])
712 		apix_level_intr_pre_eoi(vecp->v_inum);
713 
714 	/* send back EOI */
715 	apix_send_eoi();
716 
717 	cpu_infop->aci_current[nipl] = vector;
718 	if ((nipl > ipl) && (nipl > cpu->cpu_base_spl)) {
719 		cpu_infop->aci_curipl = (uchar_t)nipl;
720 		cpu_infop->aci_ISR_in_progress |= 1 << nipl;
721 	}
722 
723 #ifdef	DEBUG
724 	if (vector >= APIX_IPI_MIN)
725 		return (nipl);	/* skip IPI */
726 
727 	APIC_DEBUG_BUF_PUT(vector);
728 	APIC_DEBUG_BUF_PUT(vecp->v_inum);
729 	APIC_DEBUG_BUF_PUT(nipl);
730 	APIC_DEBUG_BUF_PUT(psm_get_cpu_id());
731 	if ((apic_stretch_interrupts) && (apic_stretch_ISR & (1 << nipl)))
732 		drv_usecwait(apic_stretch_interrupts);
733 #endif /* DEBUG */
734 
735 	return (nipl);
736 }
737 
738 /*
739  * Any changes made to this function must also change X2APIC
740  * version of intr_exit.
741  */
742 static void
743 apix_intr_exit(int prev_ipl, int arg2)
744 {
745 	int cpuid = psm_get_cpu_id();
746 	apic_cpus_info_t *cpu_infop = &apic_cpus[cpuid];
747 	apix_impl_t *apixp = apixs[cpuid];
748 
749 	UNREFERENCED_1PARAMETER(arg2);
750 
751 	cpu_infop->aci_curipl = (uchar_t)prev_ipl;
752 	/* ISR above current pri could not be in progress */
753 	cpu_infop->aci_ISR_in_progress &= (2 << prev_ipl) - 1;
754 
755 	if (apixp->x_obsoletes != NULL) {
756 		if (APIX_CPU_LOCK_HELD(cpuid))
757 			return;
758 
759 		APIX_ENTER_CPU_LOCK(cpuid);
760 		(void) apix_obsolete_vector(apixp->x_obsoletes);
761 		APIX_LEAVE_CPU_LOCK(cpuid);
762 	}
763 }
764 
765 /*
766  * The pcplusmp setspl code uses the TPR to mask all interrupts at or below the
767  * given ipl, but apix never uses the TPR and we never mask a subset of the
768  * interrupts. They are either all blocked by the IF flag or all can come in.
769  *
770  * For setspl, we mask all interrupts for XC_HI_PIL (15), otherwise, interrupts
771  * can come in if currently enabled by the IF flag. This table shows the state
772  * of the IF flag when we leave this function.
773  *
774  *    curr IF |	ipl == 15	ipl != 15
775  *    --------+---------------------------
776  *       0    |    0		    0
777  *       1    |    0		    1
778  */
779 static void
780 apix_setspl(int ipl)
781 {
782 	/*
783 	 * Interrupts at ipl above this cannot be in progress, so the following
784 	 * mask is ok.
785 	 */
786 	apic_cpus[psm_get_cpu_id()].aci_ISR_in_progress &= (2 << ipl) - 1;
787 
788 	if (ipl == XC_HI_PIL)
789 		cli();
790 }
791 
792 int
793 apix_addspl(int virtvec, int ipl, int min_ipl, int max_ipl)
794 {
795 	uint32_t cpuid = APIX_VIRTVEC_CPU(virtvec);
796 	uchar_t vector = (uchar_t)APIX_VIRTVEC_VECTOR(virtvec);
797 	apix_vector_t *vecp = xv_vector(cpuid, vector);
798 
799 	UNREFERENCED_3PARAMETER(ipl, min_ipl, max_ipl);
800 	ASSERT(vecp != NULL && LOCK_HELD(&apix_lock));
801 
802 	if (vecp->v_type == APIX_TYPE_FIXED)
803 		apix_intx_set_shared(vecp->v_inum, 1);
804 
805 	/* There are more interrupts, so it's already been enabled */
806 	if (vecp->v_share > 1)
807 		return (PSM_SUCCESS);
808 
809 	/* return if it is not hardware interrupt */
810 	if (vecp->v_type == APIX_TYPE_IPI)
811 		return (PSM_SUCCESS);
812 
813 	/*
814 	 * if apix_picinit() has not been called yet, just return.
815 	 * At the end of apic_picinit(), we will call setup_io_intr().
816 	 */
817 	if (!apic_picinit_called)
818 		return (PSM_SUCCESS);
819 
820 	(void) apix_setup_io_intr(vecp);
821 
822 	return (PSM_SUCCESS);
823 }
824 
825 int
826 apix_delspl(int virtvec, int ipl, int min_ipl, int max_ipl)
827 {
828 	uint32_t cpuid = APIX_VIRTVEC_CPU(virtvec);
829 	uchar_t vector = (uchar_t)APIX_VIRTVEC_VECTOR(virtvec);
830 	apix_vector_t *vecp = xv_vector(cpuid, vector);
831 
832 	UNREFERENCED_3PARAMETER(ipl, min_ipl, max_ipl);
833 	ASSERT(vecp != NULL && LOCK_HELD(&apix_lock));
834 
835 	if (vecp->v_type == APIX_TYPE_FIXED)
836 		apix_intx_set_shared(vecp->v_inum, -1);
837 
838 	/* There are more interrupts */
839 	if (vecp->v_share > 1)
840 		return (PSM_SUCCESS);
841 
842 	/* return if it is not hardware interrupt */
843 	if (vecp->v_type == APIX_TYPE_IPI)
844 		return (PSM_SUCCESS);
845 
846 	if (!apic_picinit_called) {
847 		cmn_err(CE_WARN, "apix: delete 0x%x before apic init",
848 		    virtvec);
849 		return (PSM_SUCCESS);
850 	}
851 
852 	apix_disable_vector(vecp);
853 
854 	return (PSM_SUCCESS);
855 }
856 
857 /*
858  * Try and disable all interrupts. We just assign interrupts to other
859  * processors based on policy. If any were bound by user request, we
860  * let them continue and return failure. We do not bother to check
861  * for cache affinity while rebinding.
862  */
863 static int
864 apix_disable_intr(processorid_t cpun)
865 {
866 	apix_impl_t *apixp = apixs[cpun];
867 	apix_vector_t *vecp, *newp;
868 	int bindcpu, i, hardbound = 0, errbound = 0, ret, loop, type;
869 
870 	lock_set(&apix_lock);
871 
872 	apic_cpus[cpun].aci_status &= ~APIC_CPU_INTR_ENABLE;
873 	apic_cpus[cpun].aci_curipl = 0;
874 
875 	/* if this is for SUSPEND operation, skip rebinding */
876 	if (apic_cpus[cpun].aci_status & APIC_CPU_SUSPEND) {
877 		for (i = APIX_AVINTR_MIN; i <= APIX_AVINTR_MAX; i++) {
878 			vecp = apixp->x_vectbl[i];
879 			if (!IS_VECT_ENABLED(vecp))
880 				continue;
881 
882 			apix_disable_vector(vecp);
883 		}
884 		lock_clear(&apix_lock);
885 		return (PSM_SUCCESS);
886 	}
887 
888 	for (i = APIX_AVINTR_MIN; i <= APIX_AVINTR_MAX; i++) {
889 		vecp = apixp->x_vectbl[i];
890 		if (!IS_VECT_ENABLED(vecp))
891 			continue;
892 
893 		if (vecp->v_flags & APIX_VECT_USER_BOUND) {
894 			hardbound++;
895 			continue;
896 		}
897 		type = vecp->v_type;
898 
899 		/*
900 		 * If there are bound interrupts on this cpu, then
901 		 * rebind them to other processors.
902 		 */
903 		loop = 0;
904 		do {
905 			bindcpu = apic_find_cpu(APIC_CPU_INTR_ENABLE);
906 
907 			if (type != APIX_TYPE_MSI)
908 				newp = apix_set_cpu(vecp, bindcpu, &ret);
909 			else
910 				newp = apix_grp_set_cpu(vecp, bindcpu, &ret);
911 		} while ((newp == NULL) && (loop++ < apic_nproc));
912 
913 		if (loop >= apic_nproc) {
914 			errbound++;
915 			cmn_err(CE_WARN, "apix: failed to rebind vector %x/%x",
916 			    vecp->v_cpuid, vecp->v_vector);
917 		}
918 	}
919 
920 	lock_clear(&apix_lock);
921 
922 	if (hardbound || errbound) {
923 		cmn_err(CE_WARN, "Could not disable interrupts on %d"
924 		    "due to user bound interrupts or failed operation",
925 		    cpun);
926 		return (PSM_FAILURE);
927 	}
928 
929 	return (PSM_SUCCESS);
930 }
931 
932 /*
933  * Bind interrupts to specified CPU
934  */
935 static void
936 apix_enable_intr(processorid_t cpun)
937 {
938 	apix_vector_t *vecp;
939 	int i, ret;
940 	processorid_t n;
941 
942 	lock_set(&apix_lock);
943 
944 	apic_cpus[cpun].aci_status |= APIC_CPU_INTR_ENABLE;
945 
946 	/* interrupt enabling for system resume */
947 	if (apic_cpus[cpun].aci_status & APIC_CPU_SUSPEND) {
948 		for (i = APIX_AVINTR_MIN; i <= APIX_AVINTR_MAX; i++) {
949 			vecp = xv_vector(cpun, i);
950 			if (!IS_VECT_ENABLED(vecp))
951 				continue;
952 
953 			apix_enable_vector(vecp);
954 		}
955 		apic_cpus[cpun].aci_status &= ~APIC_CPU_SUSPEND;
956 	}
957 
958 	for (n = 0; n < apic_nproc; n++) {
959 		if (!apic_cpu_in_range(n) || n == cpun ||
960 		    (apic_cpus[n].aci_status & APIC_CPU_INTR_ENABLE) == 0)
961 			continue;
962 
963 		for (i = APIX_AVINTR_MIN; i <= APIX_AVINTR_MAX; i++) {
964 			vecp = xv_vector(n, i);
965 			if (!IS_VECT_ENABLED(vecp) ||
966 			    vecp->v_bound_cpuid != cpun)
967 				continue;
968 
969 			if (vecp->v_type != APIX_TYPE_MSI)
970 				(void) apix_set_cpu(vecp, cpun, &ret);
971 			else
972 				(void) apix_grp_set_cpu(vecp, cpun, &ret);
973 		}
974 	}
975 
976 	lock_clear(&apix_lock);
977 }
978 
979 /*
980  * Allocate vector for IPI
981  * type == -1 indicates it is an internal request. Do not change
982  * resv_vector for these requests.
983  */
984 static int
985 apix_get_ipivect(int ipl, int type)
986 {
987 	uchar_t vector;
988 
989 	if ((vector = apix_alloc_ipi(ipl)) > 0) {
990 		if (type != -1)
991 			apic_resv_vector[ipl] = vector;
992 		return (vector);
993 	}
994 	apic_error |= APIC_ERR_GET_IPIVECT_FAIL;
995 	return (-1);	/* shouldn't happen */
996 }
997 
998 static int
999 apix_get_clkvect(int ipl)
1000 {
1001 	int vector;
1002 
1003 	if ((vector = apix_get_ipivect(ipl, -1)) == -1)
1004 		return (-1);
1005 
1006 	apic_clkvect = vector - APIC_BASE_VECT;
1007 	APIC_VERBOSE(IPI, (CE_CONT, "apix: clock vector = %x\n",
1008 	    apic_clkvect));
1009 	return (vector);
1010 }
1011 
1012 static int
1013 apix_post_cpu_start()
1014 {
1015 	int cpun;
1016 	static int cpus_started = 1;
1017 
1018 	/* We know this CPU + BSP  started successfully. */
1019 	cpus_started++;
1020 
1021 	/*
1022 	 * On BSP we would have enabled X2APIC, if supported by processor,
1023 	 * in acpi_probe(), but on AP we do it here.
1024 	 *
1025 	 * We enable X2APIC mode only if BSP is running in X2APIC & the
1026 	 * local APIC mode of the current CPU is MMIO (xAPIC).
1027 	 */
1028 	if (apic_mode == LOCAL_X2APIC && apic_detect_x2apic() &&
1029 	    apic_local_mode() == LOCAL_APIC) {
1030 		apic_enable_x2apic();
1031 	}
1032 
1033 	/*
1034 	 * Switch back to x2apic IPI sending method for performance when target
1035 	 * CPU has entered x2apic mode.
1036 	 */
1037 	if (apic_mode == LOCAL_X2APIC) {
1038 		apic_switch_ipi_callback(B_FALSE);
1039 	}
1040 
1041 	splx(ipltospl(LOCK_LEVEL));
1042 	apix_init_intr();
1043 
1044 	/*
1045 	 * since some systems don't enable the internal cache on the non-boot
1046 	 * cpus, so we have to enable them here
1047 	 */
1048 	setcr0(getcr0() & ~(CR0_CD | CR0_NW));
1049 
1050 #ifdef	DEBUG
1051 	APIC_AV_PENDING_SET();
1052 #else
1053 	if (apic_mode == LOCAL_APIC)
1054 		APIC_AV_PENDING_SET();
1055 #endif	/* DEBUG */
1056 
1057 	/*
1058 	 * We may be booting, or resuming from suspend; aci_status will
1059 	 * be APIC_CPU_INTR_ENABLE if coming from suspend, so we add the
1060 	 * APIC_CPU_ONLINE flag here rather than setting aci_status completely.
1061 	 */
1062 	cpun = psm_get_cpu_id();
1063 	apic_cpus[cpun].aci_status |= APIC_CPU_ONLINE;
1064 
1065 	apic_reg_ops->apic_write(APIC_DIVIDE_REG, apic_divide_reg_init);
1066 
1067 	return (PSM_SUCCESS);
1068 }
1069 
1070 /*
1071  * If this module needs a periodic handler for the interrupt distribution, it
1072  * can be added here. The argument to the periodic handler is not currently
1073  * used, but is reserved for future.
1074  */
1075 static void
1076 apix_post_cyclic_setup(void *arg)
1077 {
1078 	UNREFERENCED_1PARAMETER(arg);
1079 
1080 	cyc_handler_t cyh;
1081 	cyc_time_t cyt;
1082 
1083 	/* cpu_lock is held */
1084 	/* set up a periodic handler for intr redistribution */
1085 
1086 	/*
1087 	 * In peridoc mode intr redistribution processing is done in
1088 	 * apic_intr_enter during clk intr processing
1089 	 */
1090 	if (!apic_oneshot)
1091 		return;
1092 
1093 	/*
1094 	 * Register a periodical handler for the redistribution processing.
1095 	 * Though we would generally prefer to use the DDI interface for
1096 	 * periodic handler invocation, ddi_periodic_add(9F), we are
1097 	 * unfortunately already holding cpu_lock, which ddi_periodic_add will
1098 	 * attempt to take for us.  Thus, we add our own cyclic directly:
1099 	 */
1100 	cyh.cyh_func = (void (*)(void *))apix_redistribute_compute;
1101 	cyh.cyh_arg = NULL;
1102 	cyh.cyh_level = CY_LOW_LEVEL;
1103 
1104 	cyt.cyt_when = 0;
1105 	cyt.cyt_interval = apic_redistribute_sample_interval;
1106 
1107 	apic_cyclic_id = cyclic_add(&cyh, &cyt);
1108 }
1109 
1110 /*
1111  * Called the first time we enable x2apic mode on this cpu.
1112  * Update some of the function pointers to use x2apic routines.
1113  */
1114 void
1115 x2apic_update_psm()
1116 {
1117 	struct psm_ops *pops = &apix_ops;
1118 
1119 	ASSERT(pops != NULL);
1120 
1121 	/*
1122 	 * The pcplusmp module's version of x2apic_update_psm makes additional
1123 	 * changes that we do not have to make here. It needs to make those
1124 	 * changes because pcplusmp relies on the TPR register and the means of
1125 	 * addressing that changes when using the local apic versus the x2apic.
1126 	 * It's also worth noting that the apix driver specific function end up
1127 	 * being apix_foo as opposed to apic_foo and x2apic_foo.
1128 	 */
1129 	pops->psm_send_ipi = x2apic_send_ipi;
1130 
1131 	send_dirintf = pops->psm_send_ipi;
1132 
1133 	apic_mode = LOCAL_X2APIC;
1134 	apic_change_ops();
1135 }
1136 
1137 /*
1138  * This function provides external interface to the nexus for all
1139  * functionalities related to the new DDI interrupt framework.
1140  *
1141  * Input:
1142  * dip     - pointer to the dev_info structure of the requested device
1143  * hdlp    - pointer to the internal interrupt handle structure for the
1144  *	     requested interrupt
1145  * intr_op - opcode for this call
1146  * result  - pointer to the integer that will hold the result to be
1147  *	     passed back if return value is PSM_SUCCESS
1148  *
1149  * Output:
1150  * return value is either PSM_SUCCESS or PSM_FAILURE
1151  */
1152 static int
1153 apix_intr_ops(dev_info_t *dip, ddi_intr_handle_impl_t *hdlp,
1154     psm_intr_op_t intr_op, int *result)
1155 {
1156 	int		cap;
1157 	apix_vector_t	*vecp, *newvecp;
1158 	struct intrspec *ispec, intr_spec;
1159 	processorid_t target;
1160 
1161 	ispec = &intr_spec;
1162 	ispec->intrspec_pri = hdlp->ih_pri;
1163 	ispec->intrspec_vec = hdlp->ih_inum;
1164 	ispec->intrspec_func = hdlp->ih_cb_func;
1165 
1166 	switch (intr_op) {
1167 	case PSM_INTR_OP_ALLOC_VECTORS:
1168 		switch (hdlp->ih_type) {
1169 		case DDI_INTR_TYPE_MSI:
1170 			/* allocate MSI vectors */
1171 			*result = apix_alloc_msi(dip, hdlp->ih_inum,
1172 			    hdlp->ih_scratch1,
1173 			    (int)(uintptr_t)hdlp->ih_scratch2);
1174 			break;
1175 		case DDI_INTR_TYPE_MSIX:
1176 			/* allocate MSI-X vectors */
1177 			*result = apix_alloc_msix(dip, hdlp->ih_inum,
1178 			    hdlp->ih_scratch1,
1179 			    (int)(uintptr_t)hdlp->ih_scratch2);
1180 			break;
1181 		case DDI_INTR_TYPE_FIXED:
1182 			/* allocate or share vector for fixed */
1183 			if ((ihdl_plat_t *)hdlp->ih_private == NULL) {
1184 				return (PSM_FAILURE);
1185 			}
1186 			ispec = ((ihdl_plat_t *)hdlp->ih_private)->ip_ispecp;
1187 			*result = apix_intx_alloc_vector(dip, hdlp->ih_inum,
1188 			    ispec);
1189 			break;
1190 		default:
1191 			return (PSM_FAILURE);
1192 		}
1193 		break;
1194 	case PSM_INTR_OP_FREE_VECTORS:
1195 		apix_free_vectors(dip, hdlp->ih_inum, hdlp->ih_scratch1,
1196 		    hdlp->ih_type);
1197 		break;
1198 	case PSM_INTR_OP_XLATE_VECTOR:
1199 		/*
1200 		 * Vectors are allocated by ALLOC and freed by FREE.
1201 		 * XLATE finds and returns APIX_VIRTVEC_VECTOR(cpu, vector).
1202 		 */
1203 		*result = APIX_INVALID_VECT;
1204 		vecp = apix_get_dev_map(dip, hdlp->ih_inum, hdlp->ih_type);
1205 		if (vecp != NULL) {
1206 			*result = APIX_VIRTVECTOR(vecp->v_cpuid,
1207 			    vecp->v_vector);
1208 			break;
1209 		}
1210 
1211 		/*
1212 		 * No vector to device mapping exists. If this is FIXED type
1213 		 * then check if this IRQ is already mapped for another device
1214 		 * then return the vector number for it (i.e. shared IRQ case).
1215 		 * Otherwise, return PSM_FAILURE.
1216 		 */
1217 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED) {
1218 			vecp = apix_intx_xlate_vector(dip, hdlp->ih_inum,
1219 			    ispec);
1220 			*result = (vecp == NULL) ? APIX_INVALID_VECT :
1221 			    APIX_VIRTVECTOR(vecp->v_cpuid, vecp->v_vector);
1222 		}
1223 		if (*result == APIX_INVALID_VECT)
1224 			return (PSM_FAILURE);
1225 		break;
1226 	case PSM_INTR_OP_GET_PENDING:
1227 		vecp = apix_get_dev_map(dip, hdlp->ih_inum, hdlp->ih_type);
1228 		if (vecp == NULL)
1229 			return (PSM_FAILURE);
1230 
1231 		*result = apix_get_pending(vecp);
1232 		break;
1233 	case PSM_INTR_OP_CLEAR_MASK:
1234 		if (hdlp->ih_type != DDI_INTR_TYPE_FIXED)
1235 			return (PSM_FAILURE);
1236 
1237 		vecp = apix_get_dev_map(dip, hdlp->ih_inum, hdlp->ih_type);
1238 		if (vecp == NULL)
1239 			return (PSM_FAILURE);
1240 
1241 		apix_intx_clear_mask(vecp->v_inum);
1242 		break;
1243 	case PSM_INTR_OP_SET_MASK:
1244 		if (hdlp->ih_type != DDI_INTR_TYPE_FIXED)
1245 			return (PSM_FAILURE);
1246 
1247 		vecp = apix_get_dev_map(dip, hdlp->ih_inum, hdlp->ih_type);
1248 		if (vecp == NULL)
1249 			return (PSM_FAILURE);
1250 
1251 		apix_intx_set_mask(vecp->v_inum);
1252 		break;
1253 	case PSM_INTR_OP_GET_SHARED:
1254 		if (hdlp->ih_type != DDI_INTR_TYPE_FIXED)
1255 			return (PSM_FAILURE);
1256 
1257 		vecp = apix_get_dev_map(dip, hdlp->ih_inum, hdlp->ih_type);
1258 		if (vecp == NULL)
1259 			return (PSM_FAILURE);
1260 
1261 		*result = apix_intx_get_shared(vecp->v_inum);
1262 		break;
1263 	case PSM_INTR_OP_SET_PRI:
1264 		/*
1265 		 * Called prior to adding the interrupt handler or when
1266 		 * an interrupt handler is unassigned.
1267 		 */
1268 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED)
1269 			return (PSM_SUCCESS);
1270 
1271 		if (apix_get_dev_map(dip, hdlp->ih_inum, hdlp->ih_type) == NULL)
1272 			return (PSM_FAILURE);
1273 
1274 		break;
1275 	case PSM_INTR_OP_SET_CPU:
1276 	case PSM_INTR_OP_GRP_SET_CPU:
1277 		/*
1278 		 * The interrupt handle given here has been allocated
1279 		 * specifically for this command, and ih_private carries
1280 		 * a CPU value.
1281 		 */
1282 		*result = EINVAL;
1283 		target = (int)(intptr_t)hdlp->ih_private;
1284 		if (!apic_cpu_in_range(target)) {
1285 			DDI_INTR_IMPLDBG((CE_WARN,
1286 			    "[grp_]set_cpu: cpu out of range: %d\n", target));
1287 			return (PSM_FAILURE);
1288 		}
1289 
1290 		lock_set(&apix_lock);
1291 
1292 		vecp = apix_get_req_vector(hdlp, hdlp->ih_flags);
1293 		if (!IS_VECT_ENABLED(vecp)) {
1294 			DDI_INTR_IMPLDBG((CE_WARN,
1295 			    "[grp]_set_cpu: invalid vector 0x%x\n",
1296 			    hdlp->ih_vector));
1297 			lock_clear(&apix_lock);
1298 			return (PSM_FAILURE);
1299 		}
1300 
1301 		*result = 0;
1302 
1303 		if (intr_op == PSM_INTR_OP_SET_CPU)
1304 			newvecp = apix_set_cpu(vecp, target, result);
1305 		else
1306 			newvecp = apix_grp_set_cpu(vecp, target, result);
1307 
1308 		lock_clear(&apix_lock);
1309 
1310 		if (newvecp == NULL) {
1311 			*result = EIO;
1312 			return (PSM_FAILURE);
1313 		}
1314 		newvecp->v_bound_cpuid = target;
1315 		hdlp->ih_vector = APIX_VIRTVECTOR(newvecp->v_cpuid,
1316 		    newvecp->v_vector);
1317 		break;
1318 
1319 	case PSM_INTR_OP_GET_INTR:
1320 		/*
1321 		 * The interrupt handle given here has been allocated
1322 		 * specifically for this command, and ih_private carries
1323 		 * a pointer to a apic_get_intr_t.
1324 		 */
1325 		if (apix_get_intr_info(hdlp, hdlp->ih_private) != PSM_SUCCESS)
1326 			return (PSM_FAILURE);
1327 		break;
1328 
1329 	case PSM_INTR_OP_CHECK_MSI:
1330 		/*
1331 		 * Check MSI/X is supported or not at APIC level and
1332 		 * masked off the MSI/X bits in hdlp->ih_type if not
1333 		 * supported before return.  If MSI/X is supported,
1334 		 * leave the ih_type unchanged and return.
1335 		 *
1336 		 * hdlp->ih_type passed in from the nexus has all the
1337 		 * interrupt types supported by the device.
1338 		 */
1339 		if (apic_support_msi == 0) {	/* uninitialized */
1340 			/*
1341 			 * if apic_support_msi is not set, call
1342 			 * apic_check_msi_support() to check whether msi
1343 			 * is supported first
1344 			 */
1345 			if (apic_check_msi_support() == PSM_SUCCESS)
1346 				apic_support_msi = 1;	/* supported */
1347 			else
1348 				apic_support_msi = -1;	/* not-supported */
1349 		}
1350 		if (apic_support_msi == 1) {
1351 			if (apic_msix_enable)
1352 				*result = hdlp->ih_type;
1353 			else
1354 				*result = hdlp->ih_type & ~DDI_INTR_TYPE_MSIX;
1355 		} else
1356 			*result = hdlp->ih_type & ~(DDI_INTR_TYPE_MSI |
1357 			    DDI_INTR_TYPE_MSIX);
1358 		break;
1359 	case PSM_INTR_OP_GET_CAP:
1360 		cap = DDI_INTR_FLAG_PENDING;
1361 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED)
1362 			cap |= DDI_INTR_FLAG_MASKABLE;
1363 		*result = cap;
1364 		break;
1365 	case PSM_INTR_OP_APIC_TYPE:
1366 		((apic_get_type_t *)(hdlp->ih_private))->avgi_type =
1367 		    apix_get_apic_type();
1368 		((apic_get_type_t *)(hdlp->ih_private))->avgi_num_intr =
1369 		    APIX_IPI_MIN;
1370 		((apic_get_type_t *)(hdlp->ih_private))->avgi_num_cpu =
1371 		    apic_nproc;
1372 		hdlp->ih_ver = apic_get_apic_version();
1373 		break;
1374 	case PSM_INTR_OP_SET_CAP:
1375 	default:
1376 		return (PSM_FAILURE);
1377 	}
1378 
1379 	return (PSM_SUCCESS);
1380 }
1381 
1382 static void
1383 apix_cleanup_busy(void)
1384 {
1385 	int i, j;
1386 	apix_vector_t *vecp;
1387 
1388 	for (i = 0; i < apic_nproc; i++) {
1389 		if (!apic_cpu_in_range(i))
1390 			continue;
1391 		apic_cpus[i].aci_busy = 0;
1392 		for (j = APIX_AVINTR_MIN; j < APIX_AVINTR_MAX; j++) {
1393 			if ((vecp = xv_vector(i, j)) != NULL)
1394 				vecp->v_busy = 0;
1395 		}
1396 	}
1397 }
1398 
1399 static void
1400 apix_redistribute_compute(void)
1401 {
1402 	int	i, j, max_busy;
1403 
1404 	if (!apic_enable_dynamic_migration)
1405 		return;
1406 
1407 	if (++apic_nticks == apic_sample_factor_redistribution) {
1408 		/*
1409 		 * Time to call apic_intr_redistribute().
1410 		 * reset apic_nticks. This will cause max_busy
1411 		 * to be calculated below and if it is more than
1412 		 * apic_int_busy, we will do the whole thing
1413 		 */
1414 		apic_nticks = 0;
1415 	}
1416 	max_busy = 0;
1417 	for (i = 0; i < apic_nproc; i++) {
1418 		if (!apic_cpu_in_range(i))
1419 			continue;
1420 		/*
1421 		 * Check if curipl is non zero & if ISR is in
1422 		 * progress
1423 		 */
1424 		if (((j = apic_cpus[i].aci_curipl) != 0) &&
1425 		    (apic_cpus[i].aci_ISR_in_progress & (1 << j))) {
1426 
1427 			int	vect;
1428 			apic_cpus[i].aci_busy++;
1429 			vect = apic_cpus[i].aci_current[j];
1430 			apixs[i]->x_vectbl[vect]->v_busy++;
1431 		}
1432 
1433 		if (!apic_nticks &&
1434 		    (apic_cpus[i].aci_busy > max_busy))
1435 			max_busy = apic_cpus[i].aci_busy;
1436 	}
1437 	if (!apic_nticks) {
1438 		if (max_busy > apic_int_busy_mark) {
1439 		/*
1440 		 * We could make the following check be
1441 		 * skipped > 1 in which case, we get a
1442 		 * redistribution at half the busy mark (due to
1443 		 * double interval). Need to be able to collect
1444 		 * more empirical data to decide if that is a
1445 		 * good strategy. Punt for now.
1446 		 */
1447 			apix_cleanup_busy();
1448 			apic_skipped_redistribute = 0;
1449 		} else
1450 			apic_skipped_redistribute++;
1451 	}
1452 }
1453 
1454 /*
1455  * intr_ops() service routines
1456  */
1457 
1458 static int
1459 apix_get_pending(apix_vector_t *vecp)
1460 {
1461 	int bit, index, irr, pending;
1462 
1463 	/* need to get on the bound cpu */
1464 	mutex_enter(&cpu_lock);
1465 	affinity_set(vecp->v_cpuid);
1466 
1467 	index = vecp->v_vector / 32;
1468 	bit = vecp->v_vector % 32;
1469 	irr = apic_reg_ops->apic_read(APIC_IRR_REG + index);
1470 
1471 	affinity_clear();
1472 	mutex_exit(&cpu_lock);
1473 
1474 	pending = (irr & (1 << bit)) ? 1 : 0;
1475 	if (!pending && vecp->v_type == APIX_TYPE_FIXED)
1476 		pending = apix_intx_get_pending(vecp->v_inum);
1477 
1478 	return (pending);
1479 }
1480 
1481 static apix_vector_t *
1482 apix_get_req_vector(ddi_intr_handle_impl_t *hdlp, ushort_t flags)
1483 {
1484 	apix_vector_t *vecp;
1485 	processorid_t cpuid;
1486 	int32_t virt_vec = 0;
1487 
1488 	switch (flags & PSMGI_INTRBY_FLAGS) {
1489 	case PSMGI_INTRBY_IRQ:
1490 		return (apix_intx_get_vector(hdlp->ih_vector));
1491 	case PSMGI_INTRBY_VEC:
1492 		virt_vec = (virt_vec == 0) ? hdlp->ih_vector : virt_vec;
1493 
1494 		cpuid = APIX_VIRTVEC_CPU(virt_vec);
1495 		if (!apic_cpu_in_range(cpuid))
1496 			return (NULL);
1497 
1498 		vecp = xv_vector(cpuid, APIX_VIRTVEC_VECTOR(virt_vec));
1499 		break;
1500 	case PSMGI_INTRBY_DEFAULT:
1501 		vecp = apix_get_dev_map(hdlp->ih_dip, hdlp->ih_inum,
1502 		    hdlp->ih_type);
1503 		break;
1504 	default:
1505 		return (NULL);
1506 	}
1507 
1508 	return (vecp);
1509 }
1510 
1511 static int
1512 apix_get_intr_info(ddi_intr_handle_impl_t *hdlp,
1513     apic_get_intr_t *intr_params_p)
1514 {
1515 	apix_vector_t *vecp;
1516 	struct autovec *av_dev;
1517 	int i;
1518 
1519 	vecp = apix_get_req_vector(hdlp, intr_params_p->avgi_req_flags);
1520 	if (IS_VECT_FREE(vecp)) {
1521 		intr_params_p->avgi_num_devs = 0;
1522 		intr_params_p->avgi_cpu_id = 0;
1523 		intr_params_p->avgi_req_flags = 0;
1524 		return (PSM_SUCCESS);
1525 	}
1526 
1527 	if (intr_params_p->avgi_req_flags & PSMGI_REQ_CPUID) {
1528 		intr_params_p->avgi_cpu_id = vecp->v_cpuid;
1529 
1530 		/* Return user bound info for intrd. */
1531 		if (intr_params_p->avgi_cpu_id & IRQ_USER_BOUND) {
1532 			intr_params_p->avgi_cpu_id &= ~IRQ_USER_BOUND;
1533 			intr_params_p->avgi_cpu_id |= PSMGI_CPU_USER_BOUND;
1534 		}
1535 	}
1536 
1537 	if (intr_params_p->avgi_req_flags & PSMGI_REQ_VECTOR)
1538 		intr_params_p->avgi_vector = vecp->v_vector;
1539 
1540 	if (intr_params_p->avgi_req_flags &
1541 	    (PSMGI_REQ_NUM_DEVS | PSMGI_REQ_GET_DEVS))
1542 		/* Get number of devices from apic_irq table shared field. */
1543 		intr_params_p->avgi_num_devs = vecp->v_share;
1544 
1545 	if (intr_params_p->avgi_req_flags &  PSMGI_REQ_GET_DEVS) {
1546 
1547 		intr_params_p->avgi_req_flags  |= PSMGI_REQ_NUM_DEVS;
1548 
1549 		/* Some devices have NULL dip.  Don't count these. */
1550 		if (intr_params_p->avgi_num_devs > 0) {
1551 			for (i = 0, av_dev = vecp->v_autovect; av_dev;
1552 			    av_dev = av_dev->av_link) {
1553 				if (av_dev->av_vector && av_dev->av_dip)
1554 					i++;
1555 			}
1556 			intr_params_p->avgi_num_devs =
1557 			    (uint8_t)MIN(intr_params_p->avgi_num_devs, i);
1558 		}
1559 
1560 		/* There are no viable dips to return. */
1561 		if (intr_params_p->avgi_num_devs == 0) {
1562 			intr_params_p->avgi_dip_list = NULL;
1563 
1564 		} else {	/* Return list of dips */
1565 
1566 			/* Allocate space in array for that number of devs. */
1567 			intr_params_p->avgi_dip_list = kmem_zalloc(
1568 			    intr_params_p->avgi_num_devs *
1569 			    sizeof (dev_info_t *),
1570 			    KM_NOSLEEP);
1571 			if (intr_params_p->avgi_dip_list == NULL) {
1572 				DDI_INTR_IMPLDBG((CE_WARN,
1573 				    "apix_get_vector_intr_info: no memory"));
1574 				return (PSM_FAILURE);
1575 			}
1576 
1577 			/*
1578 			 * Loop through the device list of the autovec table
1579 			 * filling in the dip array.
1580 			 *
1581 			 * Note that the autovect table may have some special
1582 			 * entries which contain NULL dips.  These will be
1583 			 * ignored.
1584 			 */
1585 			for (i = 0, av_dev = vecp->v_autovect; av_dev;
1586 			    av_dev = av_dev->av_link) {
1587 				if (av_dev->av_vector && av_dev->av_dip)
1588 					intr_params_p->avgi_dip_list[i++] =
1589 					    av_dev->av_dip;
1590 			}
1591 		}
1592 	}
1593 
1594 	return (PSM_SUCCESS);
1595 }
1596 
1597 static char *
1598 apix_get_apic_type(void)
1599 {
1600 	return (apix_psm_info.p_mach_idstring);
1601 }
1602 
1603 apix_vector_t *
1604 apix_set_cpu(apix_vector_t *vecp, int new_cpu, int *result)
1605 {
1606 	apix_vector_t *newp = NULL;
1607 	dev_info_t *dip;
1608 	int inum, cap_ptr;
1609 	ddi_acc_handle_t handle;
1610 	ddi_intr_msix_t *msix_p = NULL;
1611 	ushort_t msix_ctrl;
1612 	uintptr_t off = 0;
1613 	uint32_t mask = 0;
1614 
1615 	ASSERT(LOCK_HELD(&apix_lock));
1616 	*result = ENXIO;
1617 
1618 	/* Fail if this is an MSI intr and is part of a group. */
1619 	if (vecp->v_type == APIX_TYPE_MSI) {
1620 		if (i_ddi_intr_get_current_nintrs(APIX_GET_DIP(vecp)) > 1)
1621 			return (NULL);
1622 		else
1623 			return (apix_grp_set_cpu(vecp, new_cpu, result));
1624 	}
1625 
1626 	/*
1627 	 * Mask MSI-X. It's unmasked when MSI-X gets enabled.
1628 	 */
1629 	if (vecp->v_type == APIX_TYPE_MSIX && IS_VECT_ENABLED(vecp)) {
1630 		if ((dip = APIX_GET_DIP(vecp)) == NULL)
1631 			return (NULL);
1632 		inum = vecp->v_devp->dv_inum;
1633 
1634 		handle = i_ddi_get_pci_config_handle(dip);
1635 		cap_ptr = i_ddi_get_msi_msix_cap_ptr(dip);
1636 		msix_ctrl = pci_config_get16(handle, cap_ptr + PCI_MSIX_CTRL);
1637 		if ((msix_ctrl & PCI_MSIX_FUNCTION_MASK) == 0) {
1638 			/*
1639 			 * Function is not masked, then mask "inum"th
1640 			 * entry in the MSI-X table
1641 			 */
1642 			msix_p = i_ddi_get_msix(dip);
1643 			off = (uintptr_t)msix_p->msix_tbl_addr + (inum *
1644 			    PCI_MSIX_VECTOR_SIZE) + PCI_MSIX_VECTOR_CTRL_OFFSET;
1645 			mask = ddi_get32(msix_p->msix_tbl_hdl, (uint32_t *)off);
1646 			ddi_put32(msix_p->msix_tbl_hdl, (uint32_t *)off,
1647 			    mask | 1);
1648 		}
1649 	}
1650 
1651 	*result = 0;
1652 	if ((newp = apix_rebind(vecp, new_cpu, 1)) == NULL)
1653 		*result = EIO;
1654 
1655 	/* Restore mask bit */
1656 	if (msix_p != NULL)
1657 		ddi_put32(msix_p->msix_tbl_hdl, (uint32_t *)off, mask);
1658 
1659 	return (newp);
1660 }
1661 
1662 /*
1663  * Set cpu for MSIs
1664  */
1665 apix_vector_t *
1666 apix_grp_set_cpu(apix_vector_t *vecp, int new_cpu, int *result)
1667 {
1668 	apix_vector_t *newp, *vp;
1669 	uint32_t orig_cpu = vecp->v_cpuid;
1670 	int orig_vect = vecp->v_vector;
1671 	int i, num_vectors, cap_ptr, msi_mask_off = 0;
1672 	uint32_t msi_pvm = 0;
1673 	ushort_t msi_ctrl;
1674 	ddi_acc_handle_t handle;
1675 	dev_info_t *dip;
1676 
1677 	APIC_VERBOSE(INTR, (CE_CONT, "apix_grp_set_cpu: oldcpu: %x, vector: %x,"
1678 	    " newcpu:%x\n", vecp->v_cpuid, vecp->v_vector, new_cpu));
1679 
1680 	ASSERT(LOCK_HELD(&apix_lock));
1681 
1682 	*result = ENXIO;
1683 
1684 	if (vecp->v_type != APIX_TYPE_MSI) {
1685 		DDI_INTR_IMPLDBG((CE_WARN, "set_grp: intr not MSI\n"));
1686 		return (NULL);
1687 	}
1688 
1689 	if ((dip = APIX_GET_DIP(vecp)) == NULL)
1690 		return (NULL);
1691 
1692 	num_vectors = i_ddi_intr_get_current_nintrs(dip);
1693 	if ((num_vectors < 1) || ((num_vectors - 1) & orig_vect)) {
1694 		APIC_VERBOSE(INTR, (CE_WARN,
1695 		    "set_grp: base vec not part of a grp or not aligned: "
1696 		    "vec:0x%x, num_vec:0x%x\n", orig_vect, num_vectors));
1697 		return (NULL);
1698 	}
1699 
1700 	if (vecp->v_inum != apix_get_min_dev_inum(dip, vecp->v_type))
1701 		return (NULL);
1702 
1703 	*result = EIO;
1704 	for (i = 1; i < num_vectors; i++) {
1705 		if ((vp = xv_vector(orig_cpu, orig_vect + i)) == NULL)
1706 			return (NULL);
1707 #ifdef DEBUG
1708 		/*
1709 		 * Sanity check: CPU and dip is the same for all entries.
1710 		 * May be called when first msi to be enabled, at this time
1711 		 * add_avintr() is not called for other msi
1712 		 */
1713 		if ((vp->v_share != 0) &&
1714 		    ((APIX_GET_DIP(vp) != dip) ||
1715 		    (vp->v_cpuid != vecp->v_cpuid))) {
1716 			APIC_VERBOSE(INTR, (CE_WARN,
1717 			    "set_grp: cpu or dip for vec 0x%x difft than for "
1718 			    "vec 0x%x\n", orig_vect, orig_vect + i));
1719 			APIC_VERBOSE(INTR, (CE_WARN,
1720 			    "  cpu: %d vs %d, dip: 0x%p vs 0x%p\n", orig_cpu,
1721 			    vp->v_cpuid, (void *)dip,
1722 			    (void *)APIX_GET_DIP(vp)));
1723 			return (NULL);
1724 		}
1725 #endif /* DEBUG */
1726 	}
1727 
1728 	cap_ptr = i_ddi_get_msi_msix_cap_ptr(dip);
1729 	handle = i_ddi_get_pci_config_handle(dip);
1730 	msi_ctrl = pci_config_get16(handle, cap_ptr + PCI_MSI_CTRL);
1731 
1732 	/* MSI Per vector masking is supported. */
1733 	if (msi_ctrl & PCI_MSI_PVM_MASK) {
1734 		if (msi_ctrl &  PCI_MSI_64BIT_MASK)
1735 			msi_mask_off = cap_ptr + PCI_MSI_64BIT_MASKBITS;
1736 		else
1737 			msi_mask_off = cap_ptr + PCI_MSI_32BIT_MASK;
1738 		msi_pvm = pci_config_get32(handle, msi_mask_off);
1739 		pci_config_put32(handle, msi_mask_off, (uint32_t)-1);
1740 		APIC_VERBOSE(INTR, (CE_CONT,
1741 		    "set_grp: pvm supported.  Mask set to 0x%x\n",
1742 		    pci_config_get32(handle, msi_mask_off)));
1743 	}
1744 
1745 	if ((newp = apix_rebind(vecp, new_cpu, num_vectors)) != NULL)
1746 		*result = 0;
1747 
1748 	/* Reenable vectors if per vector masking is supported. */
1749 	if (msi_ctrl & PCI_MSI_PVM_MASK) {
1750 		pci_config_put32(handle, msi_mask_off, msi_pvm);
1751 		APIC_VERBOSE(INTR, (CE_CONT,
1752 		    "set_grp: pvm supported.  Mask restored to 0x%x\n",
1753 		    pci_config_get32(handle, msi_mask_off)));
1754 	}
1755 
1756 	return (newp);
1757 }
1758 
1759 void
1760 apix_intx_set_vector(int irqno, uint32_t cpuid, uchar_t vector)
1761 {
1762 	apic_irq_t *irqp;
1763 
1764 	mutex_enter(&airq_mutex);
1765 	irqp = apic_irq_table[irqno];
1766 	irqp->airq_cpu = cpuid;
1767 	irqp->airq_vector = vector;
1768 	apic_record_rdt_entry(irqp, irqno);
1769 	mutex_exit(&airq_mutex);
1770 }
1771 
1772 apix_vector_t *
1773 apix_intx_get_vector(int irqno)
1774 {
1775 	apic_irq_t *irqp;
1776 	uint32_t cpuid;
1777 	uchar_t vector;
1778 
1779 	mutex_enter(&airq_mutex);
1780 	irqp = apic_irq_table[irqno & 0xff];
1781 	if (IS_IRQ_FREE(irqp) || (irqp->airq_cpu == IRQ_UNINIT)) {
1782 		mutex_exit(&airq_mutex);
1783 		return (NULL);
1784 	}
1785 	cpuid = irqp->airq_cpu;
1786 	vector = irqp->airq_vector;
1787 	mutex_exit(&airq_mutex);
1788 
1789 	return (xv_vector(cpuid, vector));
1790 }
1791 
1792 /*
1793  * Must called with interrupts disabled and apic_ioapic_lock held
1794  */
1795 void
1796 apix_intx_enable(int irqno)
1797 {
1798 	uchar_t ioapicindex, intin;
1799 	apic_irq_t *irqp = apic_irq_table[irqno];
1800 	ioapic_rdt_t irdt;
1801 	apic_cpus_info_t *cpu_infop;
1802 	apix_vector_t *vecp = xv_vector(irqp->airq_cpu, irqp->airq_vector);
1803 
1804 	ASSERT(LOCK_HELD(&apic_ioapic_lock) && !IS_IRQ_FREE(irqp));
1805 
1806 	ioapicindex = irqp->airq_ioapicindex;
1807 	intin = irqp->airq_intin_no;
1808 	cpu_infop =  &apic_cpus[irqp->airq_cpu];
1809 
1810 	irdt.ir_lo = AV_PDEST | AV_FIXED | irqp->airq_rdt_entry;
1811 	irdt.ir_hi = cpu_infop->aci_local_id;
1812 
1813 	apic_vt_ops->apic_intrmap_alloc_entry(&vecp->v_intrmap_private, NULL,
1814 	    vecp->v_type, 1, ioapicindex);
1815 	apic_vt_ops->apic_intrmap_map_entry(vecp->v_intrmap_private,
1816 	    (void *)&irdt, vecp->v_type, 1);
1817 	apic_vt_ops->apic_intrmap_record_rdt(vecp->v_intrmap_private, &irdt);
1818 
1819 	/* write RDT entry high dword - destination */
1820 	WRITE_IOAPIC_RDT_ENTRY_HIGH_DWORD(ioapicindex, intin,
1821 	    irdt.ir_hi);
1822 
1823 	/* Write the vector, trigger, and polarity portion of the RDT */
1824 	WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapicindex, intin, irdt.ir_lo);
1825 
1826 	vecp->v_state = APIX_STATE_ENABLED;
1827 
1828 	APIC_VERBOSE_IOAPIC((CE_CONT, "apix_intx_enable: ioapic 0x%x"
1829 	    " intin 0x%x rdt_low 0x%x rdt_high 0x%x\n",
1830 	    ioapicindex, intin, irdt.ir_lo, irdt.ir_hi));
1831 }
1832 
1833 /*
1834  * Must called with interrupts disabled and apic_ioapic_lock held
1835  */
1836 void
1837 apix_intx_disable(int irqno)
1838 {
1839 	apic_irq_t *irqp = apic_irq_table[irqno];
1840 	int ioapicindex, intin;
1841 
1842 	ASSERT(LOCK_HELD(&apic_ioapic_lock) && !IS_IRQ_FREE(irqp));
1843 	/*
1844 	 * The assumption here is that this is safe, even for
1845 	 * systems with IOAPICs that suffer from the hardware
1846 	 * erratum because all devices have been quiesced before
1847 	 * they unregister their interrupt handlers.  If that
1848 	 * assumption turns out to be false, this mask operation
1849 	 * can induce the same erratum result we're trying to
1850 	 * avoid.
1851 	 */
1852 	ioapicindex = irqp->airq_ioapicindex;
1853 	intin = irqp->airq_intin_no;
1854 	ioapic_write(ioapicindex, APIC_RDT_CMD + 2 * intin, AV_MASK);
1855 
1856 	APIC_VERBOSE_IOAPIC((CE_CONT, "apix_intx_disable: ioapic 0x%x"
1857 	    " intin 0x%x\n", ioapicindex, intin));
1858 }
1859 
1860 void
1861 apix_intx_free(int irqno)
1862 {
1863 	apic_irq_t *irqp;
1864 
1865 	mutex_enter(&airq_mutex);
1866 	irqp = apic_irq_table[irqno];
1867 
1868 	if (IS_IRQ_FREE(irqp)) {
1869 		mutex_exit(&airq_mutex);
1870 		return;
1871 	}
1872 
1873 	irqp->airq_mps_intr_index = FREE_INDEX;
1874 	irqp->airq_cpu = IRQ_UNINIT;
1875 	irqp->airq_vector = APIX_INVALID_VECT;
1876 	mutex_exit(&airq_mutex);
1877 }
1878 
1879 #ifdef DEBUG
1880 int apix_intr_deliver_timeouts = 0;
1881 int apix_intr_rirr_timeouts = 0;
1882 int apix_intr_rirr_reset_failure = 0;
1883 #endif
1884 int apix_max_reps_irr_pending = 10;
1885 
1886 #define	GET_RDT_BITS(ioapic, intin, bits)	\
1887 	(READ_IOAPIC_RDT_ENTRY_LOW_DWORD((ioapic), (intin)) & (bits))
1888 #define	APIX_CHECK_IRR_DELAY	drv_usectohz(5000)
1889 
1890 int
1891 apix_intx_rebind(int irqno, processorid_t cpuid, uchar_t vector)
1892 {
1893 	apic_irq_t *irqp = apic_irq_table[irqno];
1894 	ulong_t iflag;
1895 	int waited, ioapic_ix, intin_no, level, repeats, rdt_entry, masked;
1896 
1897 	ASSERT(irqp != NULL);
1898 
1899 	iflag = intr_clear();
1900 	lock_set(&apic_ioapic_lock);
1901 
1902 	ioapic_ix = irqp->airq_ioapicindex;
1903 	intin_no = irqp->airq_intin_no;
1904 	level = apic_level_intr[irqno];
1905 
1906 	/*
1907 	 * Wait for the delivery status bit to be cleared. This should
1908 	 * be a very small amount of time.
1909 	 */
1910 	repeats = 0;
1911 	do {
1912 		repeats++;
1913 
1914 		for (waited = 0; waited < apic_max_reps_clear_pending;
1915 		    waited++) {
1916 			if (GET_RDT_BITS(ioapic_ix, intin_no, AV_PENDING) == 0)
1917 				break;
1918 		}
1919 		if (!level)
1920 			break;
1921 
1922 		/*
1923 		 * Mask the RDT entry for level-triggered interrupts.
1924 		 */
1925 		irqp->airq_rdt_entry |= AV_MASK;
1926 		rdt_entry = READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
1927 		    intin_no);
1928 		if ((masked = (rdt_entry & AV_MASK)) == 0) {
1929 			/* Mask it */
1930 			WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix, intin_no,
1931 			    AV_MASK | rdt_entry);
1932 		}
1933 
1934 		/*
1935 		 * If there was a race and an interrupt was injected
1936 		 * just before we masked, check for that case here.
1937 		 * Then, unmask the RDT entry and try again.  If we're
1938 		 * on our last try, don't unmask (because we want the
1939 		 * RDT entry to remain masked for the rest of the
1940 		 * function).
1941 		 */
1942 		rdt_entry = READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
1943 		    intin_no);
1944 		if ((masked == 0) && ((rdt_entry & AV_PENDING) != 0) &&
1945 		    (repeats < apic_max_reps_clear_pending)) {
1946 			/* Unmask it */
1947 			WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
1948 			    intin_no, rdt_entry & ~AV_MASK);
1949 			irqp->airq_rdt_entry &= ~AV_MASK;
1950 		}
1951 	} while ((rdt_entry & AV_PENDING) &&
1952 	    (repeats < apic_max_reps_clear_pending));
1953 
1954 #ifdef DEBUG
1955 	if (GET_RDT_BITS(ioapic_ix, intin_no, AV_PENDING) != 0)
1956 		apix_intr_deliver_timeouts++;
1957 #endif
1958 
1959 	if (!level || !APIX_IS_MASK_RDT(apix_mul_ioapic_method))
1960 		goto done;
1961 
1962 	/*
1963 	 * wait for remote IRR to be cleared for level-triggered
1964 	 * interrupts
1965 	 */
1966 	repeats = 0;
1967 	do {
1968 		repeats++;
1969 
1970 		for (waited = 0; waited < apic_max_reps_clear_pending;
1971 		    waited++) {
1972 			if (GET_RDT_BITS(ioapic_ix, intin_no, AV_REMOTE_IRR)
1973 			    == 0)
1974 				break;
1975 		}
1976 
1977 		if (GET_RDT_BITS(ioapic_ix, intin_no, AV_REMOTE_IRR) != 0) {
1978 			lock_clear(&apic_ioapic_lock);
1979 			intr_restore(iflag);
1980 
1981 			delay(APIX_CHECK_IRR_DELAY);
1982 
1983 			iflag = intr_clear();
1984 			lock_set(&apic_ioapic_lock);
1985 		}
1986 	} while (repeats < apix_max_reps_irr_pending);
1987 
1988 	if (repeats >= apix_max_reps_irr_pending) {
1989 #ifdef DEBUG
1990 		apix_intr_rirr_timeouts++;
1991 #endif
1992 
1993 		/*
1994 		 * If we waited and the Remote IRR bit is still not cleared,
1995 		 * AND if we've invoked the timeout APIC_REPROGRAM_MAX_TIMEOUTS
1996 		 * times for this interrupt, try the last-ditch workaround:
1997 		 */
1998 		if (GET_RDT_BITS(ioapic_ix, intin_no, AV_REMOTE_IRR) != 0) {
1999 			/*
2000 			 * Trying to clear the bit through normal
2001 			 * channels has failed.  So as a last-ditch
2002 			 * effort, try to set the trigger mode to
2003 			 * edge, then to level.  This has been
2004 			 * observed to work on many systems.
2005 			 */
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 			WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
2011 			    intin_no,
2012 			    READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
2013 			    intin_no) | AV_LEVEL);
2014 		}
2015 
2016 		if (GET_RDT_BITS(ioapic_ix, intin_no, AV_REMOTE_IRR) != 0) {
2017 #ifdef DEBUG
2018 			apix_intr_rirr_reset_failure++;
2019 #endif
2020 			lock_clear(&apic_ioapic_lock);
2021 			intr_restore(iflag);
2022 			prom_printf("apix: Remote IRR still "
2023 			    "not clear for IOAPIC %d intin %d.\n"
2024 			    "\tInterrupts to this pin may cease "
2025 			    "functioning.\n", ioapic_ix, intin_no);
2026 			return (1);	/* return failure */
2027 		}
2028 	}
2029 
2030 done:
2031 	/* change apic_irq_table */
2032 	lock_clear(&apic_ioapic_lock);
2033 	intr_restore(iflag);
2034 	apix_intx_set_vector(irqno, cpuid, vector);
2035 	iflag = intr_clear();
2036 	lock_set(&apic_ioapic_lock);
2037 
2038 	/* reprogramme IO-APIC RDT entry */
2039 	apix_intx_enable(irqno);
2040 
2041 	lock_clear(&apic_ioapic_lock);
2042 	intr_restore(iflag);
2043 
2044 	return (0);
2045 }
2046 
2047 static int
2048 apix_intx_get_pending(int irqno)
2049 {
2050 	apic_irq_t *irqp;
2051 	int intin, ioapicindex, pending;
2052 	ulong_t iflag;
2053 
2054 	mutex_enter(&airq_mutex);
2055 	irqp = apic_irq_table[irqno];
2056 	if (IS_IRQ_FREE(irqp)) {
2057 		mutex_exit(&airq_mutex);
2058 		return (0);
2059 	}
2060 
2061 	/* check IO-APIC delivery status */
2062 	intin = irqp->airq_intin_no;
2063 	ioapicindex = irqp->airq_ioapicindex;
2064 	mutex_exit(&airq_mutex);
2065 
2066 	iflag = intr_clear();
2067 	lock_set(&apic_ioapic_lock);
2068 
2069 	pending = (READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapicindex, intin) &
2070 	    AV_PENDING) ? 1 : 0;
2071 
2072 	lock_clear(&apic_ioapic_lock);
2073 	intr_restore(iflag);
2074 
2075 	return (pending);
2076 }
2077 
2078 /*
2079  * This function will mask the interrupt on the I/O APIC
2080  */
2081 static void
2082 apix_intx_set_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  * This function will clear the mask for the interrupt on the I/O APIC
2112  */
2113 static void
2114 apix_intx_clear_mask(int irqno)
2115 {
2116 	int intin, ioapixindex, rdt_entry;
2117 	ulong_t iflag;
2118 	apic_irq_t *irqp;
2119 
2120 	mutex_enter(&airq_mutex);
2121 	irqp = apic_irq_table[irqno];
2122 
2123 	ASSERT(irqp->airq_mps_intr_index != FREE_INDEX);
2124 
2125 	intin = irqp->airq_intin_no;
2126 	ioapixindex = irqp->airq_ioapicindex;
2127 	mutex_exit(&airq_mutex);
2128 
2129 	iflag = intr_clear();
2130 	lock_set(&apic_ioapic_lock);
2131 
2132 	rdt_entry = READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapixindex, intin);
2133 
2134 	/* clear mask */
2135 	WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapixindex, intin,
2136 	    ((~AV_MASK) & rdt_entry));
2137 
2138 	lock_clear(&apic_ioapic_lock);
2139 	intr_restore(iflag);
2140 }
2141 
2142 /*
2143  * For level-triggered interrupt, mask the IRQ line. Mask means
2144  * new interrupts will not be delivered. The interrupt already
2145  * accepted by a local APIC is not affected
2146  */
2147 void
2148 apix_level_intr_pre_eoi(int irq)
2149 {
2150 	apic_irq_t *irqp = apic_irq_table[irq];
2151 	int apic_ix, intin_ix;
2152 
2153 	if (irqp == NULL)
2154 		return;
2155 
2156 	ASSERT(apic_level_intr[irq] == TRIGGER_MODE_LEVEL);
2157 
2158 	lock_set(&apic_ioapic_lock);
2159 
2160 	intin_ix = irqp->airq_intin_no;
2161 	apic_ix = irqp->airq_ioapicindex;
2162 
2163 	if (irqp->airq_cpu != CPU->cpu_id) {
2164 		if (!APIX_IS_MASK_RDT(apix_mul_ioapic_method))
2165 			ioapic_write_eoi(apic_ix, irqp->airq_vector);
2166 		lock_clear(&apic_ioapic_lock);
2167 		return;
2168 	}
2169 
2170 	if (apix_mul_ioapic_method == APIC_MUL_IOAPIC_IOXAPIC) {
2171 		/*
2172 		 * This is a IOxAPIC and there is EOI register:
2173 		 * 	Change the vector to reserved unused vector, so that
2174 		 * 	the EOI	from Local APIC won't clear the Remote IRR for
2175 		 * 	this level trigger interrupt. Instead, we'll manually
2176 		 * 	clear it in apix_post_hardint() after ISR handling.
2177 		 */
2178 		WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(apic_ix, intin_ix,
2179 		    (irqp->airq_rdt_entry & (~0xff)) | APIX_RESV_VECTOR);
2180 	} else {
2181 		WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(apic_ix, intin_ix,
2182 		    AV_MASK | irqp->airq_rdt_entry);
2183 	}
2184 
2185 	lock_clear(&apic_ioapic_lock);
2186 }
2187 
2188 /*
2189  * For level-triggered interrupt, unmask the IRQ line
2190  * or restore the original vector number.
2191  */
2192 void
2193 apix_level_intr_post_dispatch(int irq)
2194 {
2195 	apic_irq_t *irqp = apic_irq_table[irq];
2196 	int apic_ix, intin_ix;
2197 
2198 	if (irqp == NULL)
2199 		return;
2200 
2201 	lock_set(&apic_ioapic_lock);
2202 
2203 	intin_ix = irqp->airq_intin_no;
2204 	apic_ix = irqp->airq_ioapicindex;
2205 
2206 	if (APIX_IS_DIRECTED_EOI(apix_mul_ioapic_method)) {
2207 		/*
2208 		 * Already sent EOI back to Local APIC.
2209 		 * Send EOI to IO-APIC
2210 		 */
2211 		ioapic_write_eoi(apic_ix, irqp->airq_vector);
2212 	} else {
2213 		/* clear the mask or restore the vector */
2214 		WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(apic_ix, intin_ix,
2215 		    irqp->airq_rdt_entry);
2216 
2217 		/* send EOI to IOxAPIC */
2218 		if (apix_mul_ioapic_method == APIC_MUL_IOAPIC_IOXAPIC)
2219 			ioapic_write_eoi(apic_ix, irqp->airq_vector);
2220 	}
2221 
2222 	lock_clear(&apic_ioapic_lock);
2223 }
2224 
2225 static int
2226 apix_intx_get_shared(int irqno)
2227 {
2228 	apic_irq_t *irqp;
2229 	int share;
2230 
2231 	mutex_enter(&airq_mutex);
2232 	irqp = apic_irq_table[irqno];
2233 	if (IS_IRQ_FREE(irqp) || (irqp->airq_cpu == IRQ_UNINIT)) {
2234 		mutex_exit(&airq_mutex);
2235 		return (0);
2236 	}
2237 	share = irqp->airq_share;
2238 	mutex_exit(&airq_mutex);
2239 
2240 	return (share);
2241 }
2242 
2243 static void
2244 apix_intx_set_shared(int irqno, int delta)
2245 {
2246 	apic_irq_t *irqp;
2247 
2248 	mutex_enter(&airq_mutex);
2249 	irqp = apic_irq_table[irqno];
2250 	if (IS_IRQ_FREE(irqp)) {
2251 		mutex_exit(&airq_mutex);
2252 		return;
2253 	}
2254 	irqp->airq_share += delta;
2255 	mutex_exit(&airq_mutex);
2256 }
2257 
2258 /*
2259  * Setup IRQ table. Return IRQ no or -1 on failure
2260  */
2261 static int
2262 apix_intx_setup(dev_info_t *dip, int inum, int irqno,
2263     struct apic_io_intr *intrp, struct intrspec *ispec, iflag_t *iflagp)
2264 {
2265 	int origirq = ispec->intrspec_vec;
2266 	int newirq;
2267 	short intr_index;
2268 	uchar_t ipin, ioapic, ioapicindex;
2269 	apic_irq_t *irqp;
2270 
2271 	UNREFERENCED_1PARAMETER(inum);
2272 
2273 	if (intrp != NULL) {
2274 		intr_index = (short)(intrp - apic_io_intrp);
2275 		ioapic = intrp->intr_destid;
2276 		ipin = intrp->intr_destintin;
2277 
2278 		/* Find ioapicindex. If destid was ALL, we will exit with 0. */
2279 		for (ioapicindex = apic_io_max - 1; ioapicindex; ioapicindex--)
2280 			if (apic_io_id[ioapicindex] == ioapic)
2281 				break;
2282 		ASSERT((ioapic == apic_io_id[ioapicindex]) ||
2283 		    (ioapic == INTR_ALL_APIC));
2284 
2285 		/* check whether this intin# has been used by another irqno */
2286 		if ((newirq = apic_find_intin(ioapicindex, ipin)) != -1)
2287 			return (newirq);
2288 
2289 	} else if (iflagp != NULL) {	/* ACPI */
2290 		intr_index = ACPI_INDEX;
2291 		ioapicindex = acpi_find_ioapic(irqno);
2292 		ASSERT(ioapicindex != 0xFF);
2293 		ioapic = apic_io_id[ioapicindex];
2294 		ipin = irqno - apic_io_vectbase[ioapicindex];
2295 
2296 		if (apic_irq_table[irqno] &&
2297 		    apic_irq_table[irqno]->airq_mps_intr_index == ACPI_INDEX) {
2298 			ASSERT(apic_irq_table[irqno]->airq_intin_no == ipin &&
2299 			    apic_irq_table[irqno]->airq_ioapicindex ==
2300 			    ioapicindex);
2301 			return (irqno);
2302 		}
2303 
2304 	} else {	/* default configuration */
2305 		intr_index = DEFAULT_INDEX;
2306 		ioapicindex = 0;
2307 		ioapic = apic_io_id[ioapicindex];
2308 		ipin = (uchar_t)irqno;
2309 	}
2310 
2311 	/* allocate a new IRQ no */
2312 	if ((irqp = apic_irq_table[irqno]) == NULL) {
2313 		irqp = kmem_zalloc(sizeof (apic_irq_t), KM_SLEEP);
2314 		apic_irq_table[irqno] = irqp;
2315 	} else {
2316 		if (irqp->airq_mps_intr_index != FREE_INDEX) {
2317 			newirq = apic_allocate_irq(apic_first_avail_irq);
2318 			if (newirq == -1) {
2319 				return (-1);
2320 			}
2321 			irqno = newirq;
2322 			irqp = apic_irq_table[irqno];
2323 			ASSERT(irqp != NULL);
2324 		}
2325 	}
2326 	apic_max_device_irq = max(irqno, apic_max_device_irq);
2327 	apic_min_device_irq = min(irqno, apic_min_device_irq);
2328 
2329 	irqp->airq_mps_intr_index = intr_index;
2330 	irqp->airq_ioapicindex = ioapicindex;
2331 	irqp->airq_intin_no = ipin;
2332 	irqp->airq_dip = dip;
2333 	irqp->airq_origirq = (uchar_t)origirq;
2334 	if (iflagp != NULL)
2335 		irqp->airq_iflag = *iflagp;
2336 	irqp->airq_cpu = IRQ_UNINIT;
2337 	irqp->airq_vector = 0;
2338 
2339 	return (irqno);
2340 }
2341 
2342 /*
2343  * Setup IRQ table for non-pci devices. Return IRQ no or -1 on error
2344  */
2345 static int
2346 apix_intx_setup_nonpci(dev_info_t *dip, int inum, int bustype,
2347     struct intrspec *ispec)
2348 {
2349 	int irqno = ispec->intrspec_vec;
2350 	int newirq, i;
2351 	iflag_t intr_flag;
2352 	ACPI_SUBTABLE_HEADER	*hp;
2353 	ACPI_MADT_INTERRUPT_OVERRIDE *isop;
2354 	struct apic_io_intr *intrp;
2355 
2356 	if (!apic_enable_acpi || apic_use_acpi_madt_only) {
2357 		int busid;
2358 
2359 		if (bustype == 0)
2360 			bustype = eisa_level_intr_mask ? BUS_EISA : BUS_ISA;
2361 
2362 		/* loop checking BUS_ISA/BUS_EISA */
2363 		for (i = 0; i < 2; i++) {
2364 			if (((busid = apic_find_bus_id(bustype)) != -1) &&
2365 			    ((intrp = apic_find_io_intr_w_busid(irqno, busid))
2366 			    != NULL)) {
2367 				return (apix_intx_setup(dip, inum, irqno,
2368 				    intrp, ispec, NULL));
2369 			}
2370 			bustype = (bustype == BUS_EISA) ? BUS_ISA : BUS_EISA;
2371 		}
2372 
2373 		/* fall back to default configuration */
2374 		return (-1);
2375 	}
2376 
2377 	/* search iso entries first */
2378 	if (acpi_iso_cnt != 0) {
2379 		hp = (ACPI_SUBTABLE_HEADER *)acpi_isop;
2380 		i = 0;
2381 		while (i < acpi_iso_cnt) {
2382 			if (hp->Type == ACPI_MADT_TYPE_INTERRUPT_OVERRIDE) {
2383 				isop = (ACPI_MADT_INTERRUPT_OVERRIDE *) hp;
2384 				if (isop->Bus == 0 &&
2385 				    isop->SourceIrq == irqno) {
2386 					newirq = isop->GlobalIrq;
2387 					intr_flag.intr_po = isop->IntiFlags &
2388 					    ACPI_MADT_POLARITY_MASK;
2389 					intr_flag.intr_el = (isop->IntiFlags &
2390 					    ACPI_MADT_TRIGGER_MASK) >> 2;
2391 					intr_flag.bustype = BUS_ISA;
2392 
2393 					return (apix_intx_setup(dip, inum,
2394 					    newirq, NULL, ispec, &intr_flag));
2395 				}
2396 				i++;
2397 			}
2398 			hp = (ACPI_SUBTABLE_HEADER *)(((char *)hp) +
2399 			    hp->Length);
2400 		}
2401 	}
2402 	intr_flag.intr_po = INTR_PO_ACTIVE_HIGH;
2403 	intr_flag.intr_el = INTR_EL_EDGE;
2404 	intr_flag.bustype = BUS_ISA;
2405 	return (apix_intx_setup(dip, inum, irqno, NULL, ispec, &intr_flag));
2406 }
2407 
2408 
2409 /*
2410  * Setup IRQ table for pci devices. Return IRQ no or -1 on error
2411  */
2412 static int
2413 apix_intx_setup_pci(dev_info_t *dip, int inum, int bustype,
2414     struct intrspec *ispec)
2415 {
2416 	int busid, devid, pci_irq;
2417 	ddi_acc_handle_t cfg_handle;
2418 	uchar_t ipin;
2419 	iflag_t intr_flag;
2420 	struct apic_io_intr *intrp;
2421 
2422 	if (acpica_get_bdf(dip, &busid, &devid, NULL) != 0)
2423 		return (-1);
2424 
2425 	if (busid == 0 && apic_pci_bus_total == 1)
2426 		busid = (int)apic_single_pci_busid;
2427 
2428 	if (pci_config_setup(dip, &cfg_handle) != DDI_SUCCESS)
2429 		return (-1);
2430 	ipin = pci_config_get8(cfg_handle, PCI_CONF_IPIN) - PCI_INTA;
2431 	pci_config_teardown(&cfg_handle);
2432 
2433 	if (apic_enable_acpi && !apic_use_acpi_madt_only) {	/* ACPI */
2434 		if (apic_acpi_translate_pci_irq(dip, busid, devid,
2435 		    ipin, &pci_irq, &intr_flag) != ACPI_PSM_SUCCESS)
2436 			return (-1);
2437 
2438 		intr_flag.bustype = (uchar_t)bustype;
2439 		return (apix_intx_setup(dip, inum, pci_irq, NULL, ispec,
2440 		    &intr_flag));
2441 	}
2442 
2443 	/* MP configuration table */
2444 	pci_irq = ((devid & 0x1f) << 2) | (ipin & 0x3);
2445 	if ((intrp = apic_find_io_intr_w_busid(pci_irq, busid)) == NULL) {
2446 		pci_irq = apic_handle_pci_pci_bridge(dip, devid, ipin, &intrp);
2447 		if (pci_irq == -1)
2448 			return (-1);
2449 	}
2450 
2451 	return (apix_intx_setup(dip, inum, pci_irq, intrp, ispec, NULL));
2452 }
2453 
2454 /*
2455  * Translate and return IRQ no
2456  */
2457 static int
2458 apix_intx_xlate_irq(dev_info_t *dip, int inum, struct intrspec *ispec)
2459 {
2460 	int newirq, irqno = ispec->intrspec_vec;
2461 	int parent_is_pci_or_pciex = 0, child_is_pciex = 0;
2462 	int bustype = 0, dev_len;
2463 	char dev_type[16];
2464 
2465 	if (apic_defconf) {
2466 		mutex_enter(&airq_mutex);
2467 		goto defconf;
2468 	}
2469 
2470 	if ((dip == NULL) || (!apic_irq_translate && !apic_enable_acpi)) {
2471 		mutex_enter(&airq_mutex);
2472 		goto nonpci;
2473 	}
2474 
2475 	/*
2476 	 * use ddi_getlongprop_buf() instead of ddi_prop_lookup_string()
2477 	 * to avoid extra buffer allocation.
2478 	 */
2479 	dev_len = sizeof (dev_type);
2480 	if (ddi_getlongprop_buf(DDI_DEV_T_ANY, ddi_get_parent(dip),
2481 	    DDI_PROP_DONTPASS, "device_type", (caddr_t)dev_type,
2482 	    &dev_len) == DDI_PROP_SUCCESS) {
2483 		if ((strcmp(dev_type, "pci") == 0) ||
2484 		    (strcmp(dev_type, "pciex") == 0))
2485 			parent_is_pci_or_pciex = 1;
2486 	}
2487 
2488 	if (ddi_getlongprop_buf(DDI_DEV_T_ANY, dip,
2489 	    DDI_PROP_DONTPASS, "compatible", (caddr_t)dev_type,
2490 	    &dev_len) == DDI_PROP_SUCCESS) {
2491 		if (strstr(dev_type, "pciex"))
2492 			child_is_pciex = 1;
2493 	}
2494 
2495 	mutex_enter(&airq_mutex);
2496 
2497 	if (parent_is_pci_or_pciex) {
2498 		bustype = child_is_pciex ? BUS_PCIE : BUS_PCI;
2499 		newirq = apix_intx_setup_pci(dip, inum, bustype, ispec);
2500 		if (newirq != -1)
2501 			goto done;
2502 		bustype = 0;
2503 	} else if (strcmp(dev_type, "isa") == 0)
2504 		bustype = BUS_ISA;
2505 	else if (strcmp(dev_type, "eisa") == 0)
2506 		bustype = BUS_EISA;
2507 
2508 nonpci:
2509 	newirq = apix_intx_setup_nonpci(dip, inum, bustype, ispec);
2510 	if (newirq != -1)
2511 		goto done;
2512 
2513 defconf:
2514 	newirq = apix_intx_setup(dip, inum, irqno, NULL, ispec, NULL);
2515 	if (newirq == -1) {
2516 		mutex_exit(&airq_mutex);
2517 		return (-1);
2518 	}
2519 done:
2520 	ASSERT(apic_irq_table[newirq]);
2521 	mutex_exit(&airq_mutex);
2522 	return (newirq);
2523 }
2524 
2525 static int
2526 apix_intx_alloc_vector(dev_info_t *dip, int inum, struct intrspec *ispec)
2527 {
2528 	int irqno;
2529 	apix_vector_t *vecp;
2530 
2531 	if ((irqno = apix_intx_xlate_irq(dip, inum, ispec)) == -1)
2532 		return (0);
2533 
2534 	if ((vecp = apix_alloc_intx(dip, inum, irqno)) == NULL)
2535 		return (0);
2536 
2537 	DDI_INTR_IMPLDBG((CE_CONT, "apix_intx_alloc_vector: dip=0x%p name=%s "
2538 	    "irqno=0x%x cpuid=%d vector=0x%x\n",
2539 	    (void *)dip, ddi_driver_name(dip), irqno,
2540 	    vecp->v_cpuid, vecp->v_vector));
2541 
2542 	return (1);
2543 }
2544 
2545 /*
2546  * Return the vector number if the translated IRQ for this device
2547  * has a vector mapping setup. If no IRQ setup exists or no vector is
2548  * allocated to it then return 0.
2549  */
2550 static apix_vector_t *
2551 apix_intx_xlate_vector(dev_info_t *dip, int inum, struct intrspec *ispec)
2552 {
2553 	int irqno;
2554 	apix_vector_t *vecp;
2555 
2556 	/* get the IRQ number */
2557 	if ((irqno = apix_intx_xlate_irq(dip, inum, ispec)) == -1)
2558 		return (NULL);
2559 
2560 	/* get the vector number if a vector is allocated to this irqno */
2561 	vecp = apix_intx_get_vector(irqno);
2562 
2563 	return (vecp);
2564 }
2565 
2566 /*
2567  * Switch between safe and x2APIC IPI sending method.
2568  * The CPU may power on in xapic mode or x2apic mode. If the CPU needs to send
2569  * an IPI to other CPUs before entering x2APIC mode, it still needs to use the
2570  * xAPIC method. Before sending a StartIPI to the target CPU, psm_send_ipi will
2571  * be changed to apic_common_send_ipi, which detects current local APIC mode and
2572  * use the right method to send an IPI. If some CPUs fail to start up,
2573  * apic_poweron_cnt won't return to zero, so apic_common_send_ipi will always be
2574  * used. psm_send_ipi can't be simply changed back to x2apic_send_ipi if some
2575  * CPUs failed to start up because those failed CPUs may recover itself later at
2576  * unpredictable time.
2577  */
2578 void
2579 apic_switch_ipi_callback(boolean_t enter)
2580 {
2581 	ulong_t iflag;
2582 	struct psm_ops *pops = psmops;
2583 
2584 	iflag = intr_clear();
2585 	lock_set(&apic_mode_switch_lock);
2586 	if (enter) {
2587 		ASSERT(apic_poweron_cnt >= 0);
2588 		if (apic_poweron_cnt == 0) {
2589 			pops->psm_send_ipi = apic_common_send_ipi;
2590 			send_dirintf = pops->psm_send_ipi;
2591 		}
2592 		apic_poweron_cnt++;
2593 	} else {
2594 		ASSERT(apic_poweron_cnt > 0);
2595 		apic_poweron_cnt--;
2596 		if (apic_poweron_cnt == 0) {
2597 			pops->psm_send_ipi = x2apic_send_ipi;
2598 			send_dirintf = pops->psm_send_ipi;
2599 		}
2600 	}
2601 	lock_clear(&apic_mode_switch_lock);
2602 	intr_restore(iflag);
2603 }
2604 
2605 /* stub function */
2606 int
2607 apix_loaded(void)
2608 {
2609 	return (apix_is_enabled);
2610 }
2611