xref: /titanic_50/usr/src/uts/i86pc/io/pcplusmp/apic.c (revision e27732d8ead160782f0326b20fe46c0e9126df9b)
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) 1993, 2010, Oracle and/or its affiliates. All rights reserved.
24  */
25 /*
26  * Copyright (c) 2010, Intel Corporation.
27  * All rights reserved.
28  */
29 
30 
31 /*
32  * PSMI 1.1 extensions are supported only in 2.6 and later versions.
33  * PSMI 1.2 extensions are supported only in 2.7 and later versions.
34  * PSMI 1.3 and 1.4 extensions are supported in Solaris 10.
35  * PSMI 1.5 extensions are supported in Solaris Nevada.
36  * PSMI 1.6 extensions are supported in Solaris Nevada.
37  * PSMI 1.7 extensions are supported in Solaris Nevada.
38  */
39 #define	PSMI_1_7
40 
41 #include <sys/processor.h>
42 #include <sys/time.h>
43 #include <sys/psm.h>
44 #include <sys/smp_impldefs.h>
45 #include <sys/cram.h>
46 #include <sys/acpi/acpi.h>
47 #include <sys/acpica.h>
48 #include <sys/psm_common.h>
49 #include <sys/apic.h>
50 #include <sys/pit.h>
51 #include <sys/ddi.h>
52 #include <sys/sunddi.h>
53 #include <sys/ddi_impldefs.h>
54 #include <sys/pci.h>
55 #include <sys/promif.h>
56 #include <sys/x86_archext.h>
57 #include <sys/cpc_impl.h>
58 #include <sys/uadmin.h>
59 #include <sys/panic.h>
60 #include <sys/debug.h>
61 #include <sys/archsystm.h>
62 #include <sys/trap.h>
63 #include <sys/machsystm.h>
64 #include <sys/sysmacros.h>
65 #include <sys/cpuvar.h>
66 #include <sys/rm_platter.h>
67 #include <sys/privregs.h>
68 #include <sys/note.h>
69 #include <sys/pci_intr_lib.h>
70 #include <sys/spl.h>
71 #include <sys/clock.h>
72 #include <sys/dditypes.h>
73 #include <sys/sunddi.h>
74 #include <sys/x_call.h>
75 #include <sys/reboot.h>
76 #include <sys/hpet.h>
77 #include <sys/apic_common.h>
78 
79 /*
80  *	Local Function Prototypes
81  */
82 static void apic_init_intr(void);
83 
84 /*
85  *	standard MP entries
86  */
87 static int	apic_probe(void);
88 static int	apic_getclkirq(int ipl);
89 static uint_t	apic_calibrate(volatile uint32_t *addr,
90     uint16_t *pit_ticks_adj);
91 static void	apic_init(void);
92 static void	apic_picinit(void);
93 static int	apic_post_cpu_start(void);
94 static int	apic_intr_enter(int ipl, int *vect);
95 static void	apic_setspl(int ipl);
96 static void	x2apic_setspl(int ipl);
97 static int	apic_addspl(int ipl, int vector, int min_ipl, int max_ipl);
98 static int	apic_delspl(int ipl, int vector, int min_ipl, int max_ipl);
99 static int	apic_disable_intr(processorid_t cpun);
100 static void	apic_enable_intr(processorid_t cpun);
101 static int		apic_get_ipivect(int ipl, int type);
102 static void	apic_post_cyclic_setup(void *arg);
103 
104 /*
105  * The following vector assignments influence the value of ipltopri and
106  * vectortoipl. Note that vectors 0 - 0x1f are not used. We can program
107  * idle to 0 and IPL 0 to 0xf to differentiate idle in case
108  * we care to do so in future. Note some IPLs which are rarely used
109  * will share the vector ranges and heavily used IPLs (5 and 6) have
110  * a wide range.
111  *
112  * This array is used to initialize apic_ipls[] (in apic_init()).
113  *
114  *	IPL		Vector range.		as passed to intr_enter
115  *	0		none.
116  *	1,2,3		0x20-0x2f		0x0-0xf
117  *	4		0x30-0x3f		0x10-0x1f
118  *	5		0x40-0x5f		0x20-0x3f
119  *	6		0x60-0x7f		0x40-0x5f
120  *	7,8,9		0x80-0x8f		0x60-0x6f
121  *	10		0x90-0x9f		0x70-0x7f
122  *	11		0xa0-0xaf		0x80-0x8f
123  *	...		...
124  *	15		0xe0-0xef		0xc0-0xcf
125  *	15		0xf0-0xff		0xd0-0xdf
126  */
127 uchar_t apic_vectortoipl[APIC_AVAIL_VECTOR / APIC_VECTOR_PER_IPL] = {
128 	3, 4, 5, 5, 6, 6, 9, 10, 11, 12, 13, 14, 15, 15
129 };
130 	/*
131 	 * The ipl of an ISR at vector X is apic_vectortoipl[X>>4]
132 	 * NOTE that this is vector as passed into intr_enter which is
133 	 * programmed vector - 0x20 (APIC_BASE_VECT)
134 	 */
135 
136 uchar_t	apic_ipltopri[MAXIPL + 1];	/* unix ipl to apic pri	*/
137 	/* The taskpri to be programmed into apic to mask given ipl */
138 
139 #if defined(__amd64)
140 uchar_t	apic_cr8pri[MAXIPL + 1];	/* unix ipl to cr8 pri	*/
141 #endif
142 
143 /*
144  * Correlation of the hardware vector to the IPL in use, initialized
145  * from apic_vectortoipl[] in apic_init().  The final IPLs may not correlate
146  * to the IPLs in apic_vectortoipl on some systems that share interrupt lines
147  * connected to errata-stricken IOAPICs
148  */
149 uchar_t apic_ipls[APIC_AVAIL_VECTOR];
150 
151 /*
152  * Patchable global variables.
153  */
154 int	apic_enable_hwsoftint = 0;	/* 0 - disable, 1 - enable	*/
155 int	apic_enable_bind_log = 1;	/* 1 - display interrupt binding log */
156 
157 /*
158  *	Local static data
159  */
160 static struct	psm_ops apic_ops = {
161 	apic_probe,
162 
163 	apic_init,
164 	apic_picinit,
165 	apic_intr_enter,
166 	apic_intr_exit,
167 	apic_setspl,
168 	apic_addspl,
169 	apic_delspl,
170 	apic_disable_intr,
171 	apic_enable_intr,
172 	(int (*)(int))NULL,		/* psm_softlvl_to_irq */
173 	(void (*)(int))NULL,		/* psm_set_softintr */
174 
175 	apic_set_idlecpu,
176 	apic_unset_idlecpu,
177 
178 	apic_clkinit,
179 	apic_getclkirq,
180 	(void (*)(void))NULL,		/* psm_hrtimeinit */
181 	apic_gethrtime,
182 
183 	apic_get_next_processorid,
184 	apic_cpu_start,
185 	apic_post_cpu_start,
186 	apic_shutdown,
187 	apic_get_ipivect,
188 	apic_send_ipi,
189 
190 	(int (*)(dev_info_t *, int))NULL,	/* psm_translate_irq */
191 	(void (*)(int, char *))NULL,	/* psm_notify_error */
192 	(void (*)(int))NULL,		/* psm_notify_func */
193 	apic_timer_reprogram,
194 	apic_timer_enable,
195 	apic_timer_disable,
196 	apic_post_cyclic_setup,
197 	apic_preshutdown,
198 	apic_intr_ops,			/* Advanced DDI Interrupt framework */
199 	apic_state,			/* save, restore apic state for S3 */
200 	apic_cpu_ops,			/* CPU control interface. */
201 };
202 
203 struct psm_ops *psmops = &apic_ops;
204 
205 static struct	psm_info apic_psm_info = {
206 	PSM_INFO_VER01_7,			/* version */
207 	PSM_OWN_EXCLUSIVE,			/* ownership */
208 	(struct psm_ops *)&apic_ops,		/* operation */
209 	APIC_PCPLUSMP_NAME,			/* machine name */
210 	"pcplusmp v1.4 compatible",
211 };
212 
213 static void *apic_hdlp;
214 
215 /*
216  * apic_let_idle_redistribute can have the following values:
217  * 0 - If clock decremented it from 1 to 0, clock has to call redistribute.
218  * apic_redistribute_lock prevents multiple idle cpus from redistributing
219  */
220 int	apic_num_idle_redistributions = 0;
221 static	int apic_let_idle_redistribute = 0;
222 
223 /* to gather intr data and redistribute */
224 static void apic_redistribute_compute(void);
225 
226 /*
227  *	This is the loadable module wrapper
228  */
229 
230 int
231 _init(void)
232 {
233 	if (apic_coarse_hrtime)
234 		apic_ops.psm_gethrtime = &apic_gettime;
235 	return (psm_mod_init(&apic_hdlp, &apic_psm_info));
236 }
237 
238 int
239 _fini(void)
240 {
241 	return (psm_mod_fini(&apic_hdlp, &apic_psm_info));
242 }
243 
244 int
245 _info(struct modinfo *modinfop)
246 {
247 	return (psm_mod_info(&apic_hdlp, &apic_psm_info, modinfop));
248 }
249 
250 static int
251 apic_probe(void)
252 {
253 	/* check if apix is initialized */
254 	if (apix_enable && apix_loaded())
255 		return (PSM_FAILURE);
256 	else
257 		apix_enable = 0; /* continue using pcplusmp PSM */
258 
259 	return (apic_probe_common(apic_psm_info.p_mach_idstring));
260 }
261 
262 static uchar_t
263 apic_xlate_vector_by_irq(uchar_t irq)
264 {
265 	if (apic_irq_table[irq] == NULL)
266 		return (0);
267 
268 	return (apic_irq_table[irq]->airq_vector);
269 }
270 
271 void
272 apic_init(void)
273 {
274 	int i;
275 	int	j = 1;
276 
277 	psm_get_ioapicid = apic_get_ioapicid;
278 	psm_get_localapicid = apic_get_localapicid;
279 	psm_xlate_vector_by_irq = apic_xlate_vector_by_irq;
280 
281 	apic_ipltopri[0] = APIC_VECTOR_PER_IPL; /* leave 0 for idle */
282 	for (i = 0; i < (APIC_AVAIL_VECTOR / APIC_VECTOR_PER_IPL); i++) {
283 		if ((i < ((APIC_AVAIL_VECTOR / APIC_VECTOR_PER_IPL) - 1)) &&
284 		    (apic_vectortoipl[i + 1] == apic_vectortoipl[i]))
285 			/* get to highest vector at the same ipl */
286 			continue;
287 		for (; j <= apic_vectortoipl[i]; j++) {
288 			apic_ipltopri[j] = (i << APIC_IPL_SHIFT) +
289 			    APIC_BASE_VECT;
290 		}
291 	}
292 	for (; j < MAXIPL + 1; j++)
293 		/* fill up any empty ipltopri slots */
294 		apic_ipltopri[j] = (i << APIC_IPL_SHIFT) + APIC_BASE_VECT;
295 	apic_init_common();
296 #if defined(__amd64)
297 	/*
298 	 * Make cpu-specific interrupt info point to cr8pri vector
299 	 */
300 	for (i = 0; i <= MAXIPL; i++)
301 		apic_cr8pri[i] = apic_ipltopri[i] >> APIC_IPL_SHIFT;
302 	CPU->cpu_pri_data = apic_cr8pri;
303 #else
304 	if (cpuid_have_cr8access(CPU))
305 		apic_have_32bit_cr8 = 1;
306 #endif	/* __amd64 */
307 }
308 
309 static void
310 apic_init_intr(void)
311 {
312 	processorid_t	cpun = psm_get_cpu_id();
313 	uint_t nlvt;
314 	uint32_t svr = AV_UNIT_ENABLE | APIC_SPUR_INTR;
315 
316 	apic_reg_ops->apic_write_task_reg(APIC_MASK_ALL);
317 
318 	if (apic_mode == LOCAL_APIC) {
319 		/*
320 		 * We are running APIC in MMIO mode.
321 		 */
322 		if (apic_flat_model) {
323 			apic_reg_ops->apic_write(APIC_FORMAT_REG,
324 			    APIC_FLAT_MODEL);
325 		} else {
326 			apic_reg_ops->apic_write(APIC_FORMAT_REG,
327 			    APIC_CLUSTER_MODEL);
328 		}
329 
330 		apic_reg_ops->apic_write(APIC_DEST_REG,
331 		    AV_HIGH_ORDER >> cpun);
332 	}
333 
334 	if (apic_directed_EOI_supported()) {
335 		/*
336 		 * Setting the 12th bit in the Spurious Interrupt Vector
337 		 * Register suppresses broadcast EOIs generated by the local
338 		 * APIC. The suppression of broadcast EOIs happens only when
339 		 * interrupts are level-triggered.
340 		 */
341 		svr |= APIC_SVR_SUPPRESS_BROADCAST_EOI;
342 	}
343 
344 	/* need to enable APIC before unmasking NMI */
345 	apic_reg_ops->apic_write(APIC_SPUR_INT_REG, svr);
346 
347 	/*
348 	 * Presence of an invalid vector with delivery mode AV_FIXED can
349 	 * cause an error interrupt, even if the entry is masked...so
350 	 * write a valid vector to LVT entries along with the mask bit
351 	 */
352 
353 	/* All APICs have timer and LINT0/1 */
354 	apic_reg_ops->apic_write(APIC_LOCAL_TIMER, AV_MASK|APIC_RESV_IRQ);
355 	apic_reg_ops->apic_write(APIC_INT_VECT0, AV_MASK|APIC_RESV_IRQ);
356 	apic_reg_ops->apic_write(APIC_INT_VECT1, AV_NMI);	/* enable NMI */
357 
358 	/*
359 	 * On integrated APICs, the number of LVT entries is
360 	 * 'Max LVT entry' + 1; on 82489DX's (non-integrated
361 	 * APICs), nlvt is "3" (LINT0, LINT1, and timer)
362 	 */
363 
364 	if (apic_cpus[cpun].aci_local_ver < APIC_INTEGRATED_VERS) {
365 		nlvt = 3;
366 	} else {
367 		nlvt = ((apic_reg_ops->apic_read(APIC_VERS_REG) >> 16) &
368 		    0xFF) + 1;
369 	}
370 
371 	if (nlvt >= 5) {
372 		/* Enable performance counter overflow interrupt */
373 
374 		if ((x86_feature & X86_MSR) != X86_MSR)
375 			apic_enable_cpcovf_intr = 0;
376 		if (apic_enable_cpcovf_intr) {
377 			if (apic_cpcovf_vect == 0) {
378 				int ipl = APIC_PCINT_IPL;
379 				int irq = apic_get_ipivect(ipl, -1);
380 
381 				ASSERT(irq != -1);
382 				apic_cpcovf_vect =
383 				    apic_irq_table[irq]->airq_vector;
384 				ASSERT(apic_cpcovf_vect);
385 				(void) add_avintr(NULL, ipl,
386 				    (avfunc)kcpc_hw_overflow_intr,
387 				    "apic pcint", irq, NULL, NULL, NULL, NULL);
388 				kcpc_hw_overflow_intr_installed = 1;
389 				kcpc_hw_enable_cpc_intr =
390 				    apic_cpcovf_mask_clear;
391 			}
392 			apic_reg_ops->apic_write(APIC_PCINT_VECT,
393 			    apic_cpcovf_vect);
394 		}
395 	}
396 
397 	if (nlvt >= 6) {
398 		/* Only mask TM intr if the BIOS apparently doesn't use it */
399 
400 		uint32_t lvtval;
401 
402 		lvtval = apic_reg_ops->apic_read(APIC_THERM_VECT);
403 		if (((lvtval & AV_MASK) == AV_MASK) ||
404 		    ((lvtval & AV_DELIV_MODE) != AV_SMI)) {
405 			apic_reg_ops->apic_write(APIC_THERM_VECT,
406 			    AV_MASK|APIC_RESV_IRQ);
407 		}
408 	}
409 
410 	/* Enable error interrupt */
411 
412 	if (nlvt >= 4 && apic_enable_error_intr) {
413 		if (apic_errvect == 0) {
414 			int ipl = 0xf;	/* get highest priority intr */
415 			int irq = apic_get_ipivect(ipl, -1);
416 
417 			ASSERT(irq != -1);
418 			apic_errvect = apic_irq_table[irq]->airq_vector;
419 			ASSERT(apic_errvect);
420 			/*
421 			 * Not PSMI compliant, but we are going to merge
422 			 * with ON anyway
423 			 */
424 			(void) add_avintr((void *)NULL, ipl,
425 			    (avfunc)apic_error_intr, "apic error intr",
426 			    irq, NULL, NULL, NULL, NULL);
427 		}
428 		apic_reg_ops->apic_write(APIC_ERR_VECT, apic_errvect);
429 		apic_reg_ops->apic_write(APIC_ERROR_STATUS, 0);
430 		apic_reg_ops->apic_write(APIC_ERROR_STATUS, 0);
431 	}
432 
433 	/* Enable CMCI interrupt */
434 	if (cmi_enable_cmci) {
435 
436 		mutex_enter(&cmci_cpu_setup_lock);
437 		if (cmci_cpu_setup_registered == 0) {
438 			mutex_enter(&cpu_lock);
439 			register_cpu_setup_func(cmci_cpu_setup, NULL);
440 			mutex_exit(&cpu_lock);
441 			cmci_cpu_setup_registered = 1;
442 		}
443 		mutex_exit(&cmci_cpu_setup_lock);
444 
445 		if (apic_cmci_vect == 0) {
446 			int ipl = 0x2;
447 			int irq = apic_get_ipivect(ipl, -1);
448 
449 			ASSERT(irq != -1);
450 			apic_cmci_vect = apic_irq_table[irq]->airq_vector;
451 			ASSERT(apic_cmci_vect);
452 
453 			(void) add_avintr(NULL, ipl,
454 			    (avfunc)cmi_cmci_trap,
455 			    "apic cmci intr", irq, NULL, NULL, NULL, NULL);
456 		}
457 		apic_reg_ops->apic_write(APIC_CMCI_VECT, apic_cmci_vect);
458 	}
459 }
460 
461 static void
462 apic_picinit(void)
463 {
464 	int i, j;
465 	uint_t isr;
466 
467 	/*
468 	 * Initialize and enable interrupt remapping before apic
469 	 * hardware initialization
470 	 */
471 	apic_intrmap_init(apic_mode);
472 
473 	/*
474 	 * On UniSys Model 6520, the BIOS leaves vector 0x20 isr
475 	 * bit on without clearing it with EOI.  Since softint
476 	 * uses vector 0x20 to interrupt itself, so softint will
477 	 * not work on this machine.  In order to fix this problem
478 	 * a check is made to verify all the isr bits are clear.
479 	 * If not, EOIs are issued to clear the bits.
480 	 */
481 	for (i = 7; i >= 1; i--) {
482 		isr = apic_reg_ops->apic_read(APIC_ISR_REG + (i * 4));
483 		if (isr != 0)
484 			for (j = 0; ((j < 32) && (isr != 0)); j++)
485 				if (isr & (1 << j)) {
486 					apic_reg_ops->apic_write(
487 					    APIC_EOI_REG, 0);
488 					isr &= ~(1 << j);
489 					apic_error |= APIC_ERR_BOOT_EOI;
490 				}
491 	}
492 
493 	/* set a flag so we know we have run apic_picinit() */
494 	apic_picinit_called = 1;
495 	LOCK_INIT_CLEAR(&apic_gethrtime_lock);
496 	LOCK_INIT_CLEAR(&apic_ioapic_lock);
497 	LOCK_INIT_CLEAR(&apic_error_lock);
498 	LOCK_INIT_CLEAR(&apic_mode_switch_lock);
499 
500 	picsetup();	 /* initialise the 8259 */
501 
502 	/* add nmi handler - least priority nmi handler */
503 	LOCK_INIT_CLEAR(&apic_nmi_lock);
504 
505 	if (!psm_add_nmintr(0, (avfunc) apic_nmi_intr,
506 	    "pcplusmp NMI handler", (caddr_t)NULL))
507 		cmn_err(CE_WARN, "pcplusmp: Unable to add nmi handler");
508 
509 	/*
510 	 * Check for directed-EOI capability in the local APIC.
511 	 */
512 	if (apic_directed_EOI_supported() == 1) {
513 		apic_set_directed_EOI_handler();
514 	}
515 
516 	apic_init_intr();
517 
518 	/* enable apic mode if imcr present */
519 	if (apic_imcrp) {
520 		outb(APIC_IMCR_P1, (uchar_t)APIC_IMCR_SELECT);
521 		outb(APIC_IMCR_P2, (uchar_t)APIC_IMCR_APIC);
522 	}
523 
524 	ioapic_init_intr(IOAPIC_MASK);
525 }
526 
527 #ifdef	DEBUG
528 void
529 apic_break(void)
530 {
531 }
532 #endif /* DEBUG */
533 
534 /*
535  * platform_intr_enter
536  *
537  *	Called at the beginning of the interrupt service routine to
538  *	mask all level equal to and below the interrupt priority
539  *	of the interrupting vector.  An EOI should be given to
540  *	the interrupt controller to enable other HW interrupts.
541  *
542  *	Return -1 for spurious interrupts
543  *
544  */
545 /*ARGSUSED*/
546 static int
547 apic_intr_enter(int ipl, int *vectorp)
548 {
549 	uchar_t vector;
550 	int nipl;
551 	int irq;
552 	ulong_t iflag;
553 	apic_cpus_info_t *cpu_infop;
554 
555 	/*
556 	 * The real vector delivered is (*vectorp + 0x20), but our caller
557 	 * subtracts 0x20 from the vector before passing it to us.
558 	 * (That's why APIC_BASE_VECT is 0x20.)
559 	 */
560 	vector = (uchar_t)*vectorp;
561 
562 	/* if interrupted by the clock, increment apic_nsec_since_boot */
563 	if (vector == apic_clkvect) {
564 		if (!apic_oneshot) {
565 			/* NOTE: this is not MT aware */
566 			apic_hrtime_stamp++;
567 			apic_nsec_since_boot += apic_nsec_per_intr;
568 			apic_hrtime_stamp++;
569 			last_count_read = apic_hertz_count;
570 			apic_redistribute_compute();
571 		}
572 
573 		/* We will avoid all the book keeping overhead for clock */
574 		nipl = apic_ipls[vector];
575 
576 		*vectorp = apic_vector_to_irq[vector + APIC_BASE_VECT];
577 		if (apic_mode == LOCAL_APIC) {
578 #if defined(__amd64)
579 			setcr8((ulong_t)(apic_ipltopri[nipl] >>
580 			    APIC_IPL_SHIFT));
581 #else
582 			if (apic_have_32bit_cr8)
583 				setcr8((ulong_t)(apic_ipltopri[nipl] >>
584 				    APIC_IPL_SHIFT));
585 			else
586 				LOCAL_APIC_WRITE_REG(APIC_TASK_REG,
587 				    (uint32_t)apic_ipltopri[nipl]);
588 #endif
589 			LOCAL_APIC_WRITE_REG(APIC_EOI_REG, 0);
590 		} else {
591 			X2APIC_WRITE(APIC_TASK_REG, apic_ipltopri[nipl]);
592 			X2APIC_WRITE(APIC_EOI_REG, 0);
593 		}
594 
595 		return (nipl);
596 	}
597 
598 	cpu_infop = &apic_cpus[psm_get_cpu_id()];
599 
600 	if (vector == (APIC_SPUR_INTR - APIC_BASE_VECT)) {
601 		cpu_infop->aci_spur_cnt++;
602 		return (APIC_INT_SPURIOUS);
603 	}
604 
605 	/* Check if the vector we got is really what we need */
606 	if (apic_revector_pending) {
607 		/*
608 		 * Disable interrupts for the duration of
609 		 * the vector translation to prevent a self-race for
610 		 * the apic_revector_lock.  This cannot be done
611 		 * in apic_xlate_vector because it is recursive and
612 		 * we want the vector translation to be atomic with
613 		 * respect to other (higher-priority) interrupts.
614 		 */
615 		iflag = intr_clear();
616 		vector = apic_xlate_vector(vector + APIC_BASE_VECT) -
617 		    APIC_BASE_VECT;
618 		intr_restore(iflag);
619 	}
620 
621 	nipl = apic_ipls[vector];
622 	*vectorp = irq = apic_vector_to_irq[vector + APIC_BASE_VECT];
623 
624 	if (apic_mode == LOCAL_APIC) {
625 #if defined(__amd64)
626 		setcr8((ulong_t)(apic_ipltopri[nipl] >> APIC_IPL_SHIFT));
627 #else
628 		if (apic_have_32bit_cr8)
629 			setcr8((ulong_t)(apic_ipltopri[nipl] >>
630 			    APIC_IPL_SHIFT));
631 		else
632 			LOCAL_APIC_WRITE_REG(APIC_TASK_REG,
633 			    (uint32_t)apic_ipltopri[nipl]);
634 #endif
635 	} else {
636 		X2APIC_WRITE(APIC_TASK_REG, apic_ipltopri[nipl]);
637 	}
638 
639 	cpu_infop->aci_current[nipl] = (uchar_t)irq;
640 	cpu_infop->aci_curipl = (uchar_t)nipl;
641 	cpu_infop->aci_ISR_in_progress |= 1 << nipl;
642 
643 	/*
644 	 * apic_level_intr could have been assimilated into the irq struct.
645 	 * but, having it as a character array is more efficient in terms of
646 	 * cache usage. So, we leave it as is.
647 	 */
648 	if (!apic_level_intr[irq]) {
649 		if (apic_mode == LOCAL_APIC) {
650 			LOCAL_APIC_WRITE_REG(APIC_EOI_REG, 0);
651 		} else {
652 			X2APIC_WRITE(APIC_EOI_REG, 0);
653 		}
654 	}
655 
656 #ifdef	DEBUG
657 	APIC_DEBUG_BUF_PUT(vector);
658 	APIC_DEBUG_BUF_PUT(irq);
659 	APIC_DEBUG_BUF_PUT(nipl);
660 	APIC_DEBUG_BUF_PUT(psm_get_cpu_id());
661 	if ((apic_stretch_interrupts) && (apic_stretch_ISR & (1 << nipl)))
662 		drv_usecwait(apic_stretch_interrupts);
663 
664 	if (apic_break_on_cpu == psm_get_cpu_id())
665 		apic_break();
666 #endif /* DEBUG */
667 	return (nipl);
668 }
669 
670 /*
671  * This macro is a common code used by MMIO local apic and X2APIC
672  * local apic.
673  */
674 #define	APIC_INTR_EXIT() \
675 { \
676 	cpu_infop = &apic_cpus[psm_get_cpu_id()]; \
677 	if (apic_level_intr[irq]) \
678 		apic_reg_ops->apic_send_eoi(irq); \
679 	cpu_infop->aci_curipl = (uchar_t)prev_ipl; \
680 	/* ISR above current pri could not be in progress */ \
681 	cpu_infop->aci_ISR_in_progress &= (2 << prev_ipl) - 1; \
682 }
683 
684 /*
685  * Any changes made to this function must also change X2APIC
686  * version of intr_exit.
687  */
688 void
689 apic_intr_exit(int prev_ipl, int irq)
690 {
691 	apic_cpus_info_t *cpu_infop;
692 
693 #if defined(__amd64)
694 	setcr8((ulong_t)apic_cr8pri[prev_ipl]);
695 #else
696 	if (apic_have_32bit_cr8)
697 		setcr8((ulong_t)(apic_ipltopri[prev_ipl] >> APIC_IPL_SHIFT));
698 	else
699 		apicadr[APIC_TASK_REG] = apic_ipltopri[prev_ipl];
700 #endif
701 
702 	APIC_INTR_EXIT();
703 }
704 
705 /*
706  * Same as apic_intr_exit() except it uses MSR rather than MMIO
707  * to access local apic registers.
708  */
709 void
710 x2apic_intr_exit(int prev_ipl, int irq)
711 {
712 	apic_cpus_info_t *cpu_infop;
713 
714 	X2APIC_WRITE(APIC_TASK_REG, apic_ipltopri[prev_ipl]);
715 	APIC_INTR_EXIT();
716 }
717 
718 intr_exit_fn_t
719 psm_intr_exit_fn(void)
720 {
721 	if (apic_mode == LOCAL_X2APIC)
722 		return (x2apic_intr_exit);
723 
724 	return (apic_intr_exit);
725 }
726 
727 /*
728  * Mask all interrupts below or equal to the given IPL.
729  * Any changes made to this function must also change X2APIC
730  * version of setspl.
731  */
732 static void
733 apic_setspl(int ipl)
734 {
735 #if defined(__amd64)
736 	setcr8((ulong_t)apic_cr8pri[ipl]);
737 #else
738 	if (apic_have_32bit_cr8)
739 		setcr8((ulong_t)(apic_ipltopri[ipl] >> APIC_IPL_SHIFT));
740 	else
741 		apicadr[APIC_TASK_REG] = apic_ipltopri[ipl];
742 #endif
743 
744 	/* interrupts at ipl above this cannot be in progress */
745 	apic_cpus[psm_get_cpu_id()].aci_ISR_in_progress &= (2 << ipl) - 1;
746 	/*
747 	 * this is a patch fix for the ALR QSMP P5 machine, so that interrupts
748 	 * have enough time to come in before the priority is raised again
749 	 * during the idle() loop.
750 	 */
751 	if (apic_setspl_delay)
752 		(void) apic_reg_ops->apic_get_pri();
753 }
754 
755 /*
756  * X2APIC version of setspl.
757  * Mask all interrupts below or equal to the given IPL
758  */
759 static void
760 x2apic_setspl(int ipl)
761 {
762 	X2APIC_WRITE(APIC_TASK_REG, apic_ipltopri[ipl]);
763 
764 	/* interrupts at ipl above this cannot be in progress */
765 	apic_cpus[psm_get_cpu_id()].aci_ISR_in_progress &= (2 << ipl) - 1;
766 }
767 
768 /*ARGSUSED*/
769 static int
770 apic_addspl(int irqno, int ipl, int min_ipl, int max_ipl)
771 {
772 	return (apic_addspl_common(irqno, ipl, min_ipl, max_ipl));
773 }
774 
775 static int
776 apic_delspl(int irqno, int ipl, int min_ipl, int max_ipl)
777 {
778 	return (apic_delspl_common(irqno, ipl, min_ipl,  max_ipl));
779 }
780 
781 static int
782 apic_post_cpu_start(void)
783 {
784 	int cpun;
785 	static int cpus_started = 1;
786 
787 	/* We know this CPU + BSP  started successfully. */
788 	cpus_started++;
789 
790 	/*
791 	 * On BSP we would have enabled X2APIC, if supported by processor,
792 	 * in acpi_probe(), but on AP we do it here.
793 	 *
794 	 * We enable X2APIC mode only if BSP is running in X2APIC & the
795 	 * local APIC mode of the current CPU is MMIO (xAPIC).
796 	 */
797 	if (apic_mode == LOCAL_X2APIC && apic_detect_x2apic() &&
798 	    apic_local_mode() == LOCAL_APIC) {
799 		apic_enable_x2apic();
800 	}
801 
802 	/*
803 	 * Switch back to x2apic IPI sending method for performance when target
804 	 * CPU has entered x2apic mode.
805 	 */
806 	if (apic_mode == LOCAL_X2APIC) {
807 		apic_switch_ipi_callback(B_FALSE);
808 	}
809 
810 	splx(ipltospl(LOCK_LEVEL));
811 	apic_init_intr();
812 
813 	/*
814 	 * since some systems don't enable the internal cache on the non-boot
815 	 * cpus, so we have to enable them here
816 	 */
817 	setcr0(getcr0() & ~(CR0_CD | CR0_NW));
818 
819 #ifdef	DEBUG
820 	APIC_AV_PENDING_SET();
821 #else
822 	if (apic_mode == LOCAL_APIC)
823 		APIC_AV_PENDING_SET();
824 #endif	/* DEBUG */
825 
826 	/*
827 	 * We may be booting, or resuming from suspend; aci_status will
828 	 * be APIC_CPU_INTR_ENABLE if coming from suspend, so we add the
829 	 * APIC_CPU_ONLINE flag here rather than setting aci_status completely.
830 	 */
831 	cpun = psm_get_cpu_id();
832 	apic_cpus[cpun].aci_status |= APIC_CPU_ONLINE;
833 
834 	apic_reg_ops->apic_write(APIC_DIVIDE_REG, apic_divide_reg_init);
835 	return (PSM_SUCCESS);
836 }
837 
838 /*
839  * type == -1 indicates it is an internal request. Do not change
840  * resv_vector for these requests
841  */
842 static int
843 apic_get_ipivect(int ipl, int type)
844 {
845 	uchar_t vector;
846 	int irq;
847 
848 	if ((irq = apic_allocate_irq(APIC_VECTOR(ipl))) != -1) {
849 		if (vector = apic_allocate_vector(ipl, irq, 1)) {
850 			apic_irq_table[irq]->airq_mps_intr_index =
851 			    RESERVE_INDEX;
852 			apic_irq_table[irq]->airq_vector = vector;
853 			if (type != -1) {
854 				apic_resv_vector[ipl] = vector;
855 			}
856 			return (irq);
857 		}
858 	}
859 	apic_error |= APIC_ERR_GET_IPIVECT_FAIL;
860 	return (-1);	/* shouldn't happen */
861 }
862 
863 static int
864 apic_getclkirq(int ipl)
865 {
866 	int	irq;
867 
868 	if ((irq = apic_get_ipivect(ipl, -1)) == -1)
869 		return (-1);
870 	/*
871 	 * Note the vector in apic_clkvect for per clock handling.
872 	 */
873 	apic_clkvect = apic_irq_table[irq]->airq_vector - APIC_BASE_VECT;
874 	APIC_VERBOSE_IOAPIC((CE_NOTE, "get_clkirq: vector = %x\n",
875 	    apic_clkvect));
876 	return (irq);
877 }
878 
879 /*
880  * Try and disable all interrupts. We just assign interrupts to other
881  * processors based on policy. If any were bound by user request, we
882  * let them continue and return failure. We do not bother to check
883  * for cache affinity while rebinding.
884  */
885 
886 static int
887 apic_disable_intr(processorid_t cpun)
888 {
889 	int bind_cpu = 0, i, hardbound = 0;
890 	apic_irq_t *irq_ptr;
891 	ulong_t iflag;
892 
893 	iflag = intr_clear();
894 	lock_set(&apic_ioapic_lock);
895 
896 	for (i = 0; i <= APIC_MAX_VECTOR; i++) {
897 		if (apic_reprogram_info[i].done == B_FALSE) {
898 			if (apic_reprogram_info[i].bindcpu == cpun) {
899 				/*
900 				 * CPU is busy -- it's the target of
901 				 * a pending reprogramming attempt
902 				 */
903 				lock_clear(&apic_ioapic_lock);
904 				intr_restore(iflag);
905 				return (PSM_FAILURE);
906 			}
907 		}
908 	}
909 
910 	apic_cpus[cpun].aci_status &= ~APIC_CPU_INTR_ENABLE;
911 
912 	apic_cpus[cpun].aci_curipl = 0;
913 
914 	i = apic_min_device_irq;
915 	for (; i <= apic_max_device_irq; i++) {
916 		/*
917 		 * If there are bound interrupts on this cpu, then
918 		 * rebind them to other processors.
919 		 */
920 		if ((irq_ptr = apic_irq_table[i]) != NULL) {
921 			ASSERT((irq_ptr->airq_temp_cpu == IRQ_UNBOUND) ||
922 			    (irq_ptr->airq_temp_cpu == IRQ_UNINIT) ||
923 			    (apic_cpu_in_range(irq_ptr->airq_temp_cpu)));
924 
925 			if (irq_ptr->airq_temp_cpu == (cpun | IRQ_USER_BOUND)) {
926 				hardbound = 1;
927 				continue;
928 			}
929 
930 			if (irq_ptr->airq_temp_cpu == cpun) {
931 				do {
932 					bind_cpu =
933 					    apic_find_cpu(APIC_CPU_INTR_ENABLE);
934 				} while (apic_rebind_all(irq_ptr, bind_cpu));
935 			}
936 		}
937 	}
938 
939 	lock_clear(&apic_ioapic_lock);
940 	intr_restore(iflag);
941 
942 	if (hardbound) {
943 		cmn_err(CE_WARN, "Could not disable interrupts on %d"
944 		    "due to user bound interrupts", cpun);
945 		return (PSM_FAILURE);
946 	}
947 	else
948 		return (PSM_SUCCESS);
949 }
950 
951 /*
952  * Bind interrupts to the CPU's local APIC.
953  * Interrupts should not be bound to a CPU's local APIC until the CPU
954  * is ready to receive interrupts.
955  */
956 static void
957 apic_enable_intr(processorid_t cpun)
958 {
959 	int	i;
960 	apic_irq_t *irq_ptr;
961 	ulong_t iflag;
962 
963 	iflag = intr_clear();
964 	lock_set(&apic_ioapic_lock);
965 
966 	apic_cpus[cpun].aci_status |= APIC_CPU_INTR_ENABLE;
967 
968 	i = apic_min_device_irq;
969 	for (i = apic_min_device_irq; i <= apic_max_device_irq; i++) {
970 		if ((irq_ptr = apic_irq_table[i]) != NULL) {
971 			if ((irq_ptr->airq_cpu & ~IRQ_USER_BOUND) == cpun) {
972 				(void) apic_rebind_all(irq_ptr,
973 				    irq_ptr->airq_cpu);
974 			}
975 		}
976 	}
977 
978 	if (apic_cpus[cpun].aci_status & APIC_CPU_SUSPEND)
979 		apic_cpus[cpun].aci_status &= ~APIC_CPU_SUSPEND;
980 
981 	lock_clear(&apic_ioapic_lock);
982 	intr_restore(iflag);
983 }
984 
985 /*
986  * If this module needs a periodic handler for the interrupt distribution, it
987  * can be added here. The argument to the periodic handler is not currently
988  * used, but is reserved for future.
989  */
990 static void
991 apic_post_cyclic_setup(void *arg)
992 {
993 _NOTE(ARGUNUSED(arg))
994 	/* cpu_lock is held */
995 	/* set up a periodic handler for intr redistribution */
996 
997 	/*
998 	 * In peridoc mode intr redistribution processing is done in
999 	 * apic_intr_enter during clk intr processing
1000 	 */
1001 	if (!apic_oneshot)
1002 		return;
1003 	/*
1004 	 * Register a periodical handler for the redistribution processing.
1005 	 * On X86, CY_LOW_LEVEL is mapped to the level 2 interrupt, so
1006 	 * DDI_IPL_2 should be passed to ddi_periodic_add() here.
1007 	 */
1008 	apic_periodic_id = ddi_periodic_add(
1009 	    (void (*)(void *))apic_redistribute_compute, NULL,
1010 	    apic_redistribute_sample_interval, DDI_IPL_2);
1011 }
1012 
1013 static void
1014 apic_redistribute_compute(void)
1015 {
1016 	int	i, j, max_busy;
1017 
1018 	if (apic_enable_dynamic_migration) {
1019 		if (++apic_nticks == apic_sample_factor_redistribution) {
1020 			/*
1021 			 * Time to call apic_intr_redistribute().
1022 			 * reset apic_nticks. This will cause max_busy
1023 			 * to be calculated below and if it is more than
1024 			 * apic_int_busy, we will do the whole thing
1025 			 */
1026 			apic_nticks = 0;
1027 		}
1028 		max_busy = 0;
1029 		for (i = 0; i < apic_nproc; i++) {
1030 			if (!apic_cpu_in_range(i))
1031 				continue;
1032 
1033 			/*
1034 			 * Check if curipl is non zero & if ISR is in
1035 			 * progress
1036 			 */
1037 			if (((j = apic_cpus[i].aci_curipl) != 0) &&
1038 			    (apic_cpus[i].aci_ISR_in_progress & (1 << j))) {
1039 
1040 				int	irq;
1041 				apic_cpus[i].aci_busy++;
1042 				irq = apic_cpus[i].aci_current[j];
1043 				apic_irq_table[irq]->airq_busy++;
1044 			}
1045 
1046 			if (!apic_nticks &&
1047 			    (apic_cpus[i].aci_busy > max_busy))
1048 				max_busy = apic_cpus[i].aci_busy;
1049 		}
1050 		if (!apic_nticks) {
1051 			if (max_busy > apic_int_busy_mark) {
1052 			/*
1053 			 * We could make the following check be
1054 			 * skipped > 1 in which case, we get a
1055 			 * redistribution at half the busy mark (due to
1056 			 * double interval). Need to be able to collect
1057 			 * more empirical data to decide if that is a
1058 			 * good strategy. Punt for now.
1059 			 */
1060 				if (apic_skipped_redistribute) {
1061 					apic_cleanup_busy();
1062 					apic_skipped_redistribute = 0;
1063 				} else {
1064 					apic_intr_redistribute();
1065 				}
1066 			} else
1067 				apic_skipped_redistribute++;
1068 		}
1069 	}
1070 }
1071 
1072 
1073 /*
1074  * The following functions are in the platform specific file so that they
1075  * can be different functions depending on whether we are running on
1076  * bare metal or a hypervisor.
1077  */
1078 
1079 /*
1080  * Check to make sure there are enough irq slots
1081  */
1082 int
1083 apic_check_free_irqs(int count)
1084 {
1085 	int i, avail;
1086 
1087 	avail = 0;
1088 	for (i = APIC_FIRST_FREE_IRQ; i < APIC_RESV_IRQ; i++) {
1089 		if ((apic_irq_table[i] == NULL) ||
1090 		    apic_irq_table[i]->airq_mps_intr_index == FREE_INDEX) {
1091 			if (++avail >= count)
1092 				return (PSM_SUCCESS);
1093 		}
1094 	}
1095 	return (PSM_FAILURE);
1096 }
1097 
1098 /*
1099  * This function allocates "count" MSI vector(s) for the given "dip/pri/type"
1100  */
1101 int
1102 apic_alloc_msi_vectors(dev_info_t *dip, int inum, int count, int pri,
1103     int behavior)
1104 {
1105 	int	rcount, i;
1106 	uchar_t	start, irqno;
1107 	uint32_t cpu;
1108 	major_t	major;
1109 	apic_irq_t	*irqptr;
1110 
1111 	DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_msi_vectors: dip=0x%p "
1112 	    "inum=0x%x  pri=0x%x count=0x%x behavior=%d\n",
1113 	    (void *)dip, inum, pri, count, behavior));
1114 
1115 	if (count > 1) {
1116 		if (behavior == DDI_INTR_ALLOC_STRICT &&
1117 		    apic_multi_msi_enable == 0)
1118 			return (0);
1119 		if (apic_multi_msi_enable == 0)
1120 			count = 1;
1121 	}
1122 
1123 	if ((rcount = apic_navail_vector(dip, pri)) > count)
1124 		rcount = count;
1125 	else if (rcount == 0 || (rcount < count &&
1126 	    behavior == DDI_INTR_ALLOC_STRICT))
1127 		return (0);
1128 
1129 	/* if not ISP2, then round it down */
1130 	if (!ISP2(rcount))
1131 		rcount = 1 << (highbit(rcount) - 1);
1132 
1133 	mutex_enter(&airq_mutex);
1134 
1135 	for (start = 0; rcount > 0; rcount >>= 1) {
1136 		if ((start = apic_find_multi_vectors(pri, rcount)) != 0 ||
1137 		    behavior == DDI_INTR_ALLOC_STRICT)
1138 			break;
1139 	}
1140 
1141 	if (start == 0) {
1142 		/* no vector available */
1143 		mutex_exit(&airq_mutex);
1144 		return (0);
1145 	}
1146 
1147 	if (apic_check_free_irqs(rcount) == PSM_FAILURE) {
1148 		/* not enough free irq slots available */
1149 		mutex_exit(&airq_mutex);
1150 		return (0);
1151 	}
1152 
1153 	major = (dip != NULL) ? ddi_driver_major(dip) : 0;
1154 	for (i = 0; i < rcount; i++) {
1155 		if ((irqno = apic_allocate_irq(apic_first_avail_irq)) ==
1156 		    (uchar_t)-1) {
1157 			/*
1158 			 * shouldn't happen because of the
1159 			 * apic_check_free_irqs() check earlier
1160 			 */
1161 			mutex_exit(&airq_mutex);
1162 			DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_msi_vectors: "
1163 			    "apic_allocate_irq failed\n"));
1164 			return (i);
1165 		}
1166 		apic_max_device_irq = max(irqno, apic_max_device_irq);
1167 		apic_min_device_irq = min(irqno, apic_min_device_irq);
1168 		irqptr = apic_irq_table[irqno];
1169 #ifdef	DEBUG
1170 		if (apic_vector_to_irq[start + i] != APIC_RESV_IRQ)
1171 			DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_msi_vectors: "
1172 			    "apic_vector_to_irq is not APIC_RESV_IRQ\n"));
1173 #endif
1174 		apic_vector_to_irq[start + i] = (uchar_t)irqno;
1175 
1176 		irqptr->airq_vector = (uchar_t)(start + i);
1177 		irqptr->airq_ioapicindex = (uchar_t)inum;	/* start */
1178 		irqptr->airq_intin_no = (uchar_t)rcount;
1179 		irqptr->airq_ipl = pri;
1180 		irqptr->airq_vector = start + i;
1181 		irqptr->airq_origirq = (uchar_t)(inum + i);
1182 		irqptr->airq_share_id = 0;
1183 		irqptr->airq_mps_intr_index = MSI_INDEX;
1184 		irqptr->airq_dip = dip;
1185 		irqptr->airq_major = major;
1186 		if (i == 0) /* they all bound to the same cpu */
1187 			cpu = irqptr->airq_cpu = apic_bind_intr(dip, irqno,
1188 			    0xff, 0xff);
1189 		else
1190 			irqptr->airq_cpu = cpu;
1191 		DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_msi_vectors: irq=0x%x "
1192 		    "dip=0x%p vector=0x%x origirq=0x%x pri=0x%x\n", irqno,
1193 		    (void *)irqptr->airq_dip, irqptr->airq_vector,
1194 		    irqptr->airq_origirq, pri));
1195 	}
1196 	mutex_exit(&airq_mutex);
1197 	return (rcount);
1198 }
1199 
1200 /*
1201  * This function allocates "count" MSI-X vector(s) for the given "dip/pri/type"
1202  */
1203 int
1204 apic_alloc_msix_vectors(dev_info_t *dip, int inum, int count, int pri,
1205     int behavior)
1206 {
1207 	int	rcount, i;
1208 	major_t	major;
1209 
1210 	mutex_enter(&airq_mutex);
1211 
1212 	if ((rcount = apic_navail_vector(dip, pri)) > count)
1213 		rcount = count;
1214 	else if (rcount == 0 || (rcount < count &&
1215 	    behavior == DDI_INTR_ALLOC_STRICT)) {
1216 		rcount = 0;
1217 		goto out;
1218 	}
1219 
1220 	if (apic_check_free_irqs(rcount) == PSM_FAILURE) {
1221 		/* not enough free irq slots available */
1222 		rcount = 0;
1223 		goto out;
1224 	}
1225 
1226 	major = (dip != NULL) ? ddi_driver_major(dip) : 0;
1227 	for (i = 0; i < rcount; i++) {
1228 		uchar_t	vector, irqno;
1229 		apic_irq_t	*irqptr;
1230 
1231 		if ((irqno = apic_allocate_irq(apic_first_avail_irq)) ==
1232 		    (uchar_t)-1) {
1233 			/*
1234 			 * shouldn't happen because of the
1235 			 * apic_check_free_irqs() check earlier
1236 			 */
1237 			DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_msix_vectors: "
1238 			    "apic_allocate_irq failed\n"));
1239 			rcount = i;
1240 			goto out;
1241 		}
1242 		if ((vector = apic_allocate_vector(pri, irqno, 1)) == 0) {
1243 			/*
1244 			 * shouldn't happen because of the
1245 			 * apic_navail_vector() call earlier
1246 			 */
1247 			DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_msix_vectors: "
1248 			    "apic_allocate_vector failed\n"));
1249 			rcount = i;
1250 			goto out;
1251 		}
1252 		apic_max_device_irq = max(irqno, apic_max_device_irq);
1253 		apic_min_device_irq = min(irqno, apic_min_device_irq);
1254 		irqptr = apic_irq_table[irqno];
1255 		irqptr->airq_vector = (uchar_t)vector;
1256 		irqptr->airq_ipl = pri;
1257 		irqptr->airq_origirq = (uchar_t)(inum + i);
1258 		irqptr->airq_share_id = 0;
1259 		irqptr->airq_mps_intr_index = MSIX_INDEX;
1260 		irqptr->airq_dip = dip;
1261 		irqptr->airq_major = major;
1262 		irqptr->airq_cpu = apic_bind_intr(dip, irqno, 0xff, 0xff);
1263 	}
1264 out:
1265 	mutex_exit(&airq_mutex);
1266 	return (rcount);
1267 }
1268 
1269 /*
1270  * Allocate a free vector for irq at ipl. Takes care of merging of multiple
1271  * IPLs into a single APIC level as well as stretching some IPLs onto multiple
1272  * levels. APIC_HI_PRI_VECTS interrupts are reserved for high priority
1273  * requests and allocated only when pri is set.
1274  */
1275 uchar_t
1276 apic_allocate_vector(int ipl, int irq, int pri)
1277 {
1278 	int	lowest, highest, i;
1279 
1280 	highest = apic_ipltopri[ipl] + APIC_VECTOR_MASK;
1281 	lowest = apic_ipltopri[ipl - 1] + APIC_VECTOR_PER_IPL;
1282 
1283 	if (highest < lowest) /* Both ipl and ipl - 1 map to same pri */
1284 		lowest -= APIC_VECTOR_PER_IPL;
1285 
1286 #ifdef	DEBUG
1287 	if (apic_restrict_vector)	/* for testing shared interrupt logic */
1288 		highest = lowest + apic_restrict_vector + APIC_HI_PRI_VECTS;
1289 #endif /* DEBUG */
1290 	if (pri == 0)
1291 		highest -= APIC_HI_PRI_VECTS;
1292 
1293 	for (i = lowest; i <= highest; i++) {
1294 		if (APIC_CHECK_RESERVE_VECTORS(i))
1295 			continue;
1296 		if (apic_vector_to_irq[i] == APIC_RESV_IRQ) {
1297 			apic_vector_to_irq[i] = (uchar_t)irq;
1298 			return (i);
1299 		}
1300 	}
1301 
1302 	return (0);
1303 }
1304 
1305 /* Mark vector as not being used by any irq */
1306 void
1307 apic_free_vector(uchar_t vector)
1308 {
1309 	apic_vector_to_irq[vector] = APIC_RESV_IRQ;
1310 }
1311 
1312 /*
1313  * Call rebind to do the actual programming.
1314  * Must be called with interrupts disabled and apic_ioapic_lock held
1315  * 'p' is polymorphic -- if this function is called to process a deferred
1316  * reprogramming, p is of type 'struct ioapic_reprogram_data *', from which
1317  * the irq pointer is retrieved.  If not doing deferred reprogramming,
1318  * p is of the type 'apic_irq_t *'.
1319  *
1320  * apic_ioapic_lock must be held across this call, as it protects apic_rebind
1321  * and it protects apic_get_next_bind_cpu() from a race in which a CPU can be
1322  * taken offline after a cpu is selected, but before apic_rebind is called to
1323  * bind interrupts to it.
1324  */
1325 int
1326 apic_setup_io_intr(void *p, int irq, boolean_t deferred)
1327 {
1328 	apic_irq_t *irqptr;
1329 	struct ioapic_reprogram_data *drep = NULL;
1330 	int rv;
1331 
1332 	if (deferred) {
1333 		drep = (struct ioapic_reprogram_data *)p;
1334 		ASSERT(drep != NULL);
1335 		irqptr = drep->irqp;
1336 	} else
1337 		irqptr = (apic_irq_t *)p;
1338 
1339 	ASSERT(irqptr != NULL);
1340 
1341 	rv = apic_rebind(irqptr, apic_irq_table[irq]->airq_cpu, drep);
1342 	if (rv) {
1343 		/*
1344 		 * CPU is not up or interrupts are disabled. Fall back to
1345 		 * the first available CPU
1346 		 */
1347 		rv = apic_rebind(irqptr, apic_find_cpu(APIC_CPU_INTR_ENABLE),
1348 		    drep);
1349 	}
1350 
1351 	return (rv);
1352 }
1353 
1354 
1355 uchar_t
1356 apic_modify_vector(uchar_t vector, int irq)
1357 {
1358 	apic_vector_to_irq[vector] = (uchar_t)irq;
1359 	return (vector);
1360 }
1361 
1362 char *
1363 apic_get_apic_type(void)
1364 {
1365 	return (apic_psm_info.p_mach_idstring);
1366 }
1367 
1368 void
1369 x2apic_update_psm(void)
1370 {
1371 	struct psm_ops *pops = &apic_ops;
1372 
1373 	ASSERT(pops != NULL);
1374 
1375 	pops->psm_intr_exit = x2apic_intr_exit;
1376 	pops->psm_setspl = x2apic_setspl;
1377 
1378 	pops->psm_send_ipi =  x2apic_send_ipi;
1379 	send_dirintf = pops->psm_send_ipi;
1380 
1381 	apic_mode = LOCAL_X2APIC;
1382 	apic_change_ops();
1383 }
1384