xref: /titanic_51/usr/src/uts/i86pc/io/pcplusmp/apic.c (revision 82beb6028da8d7d7f8562908ca027bd4a1cc7d37)
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 #include <sys/apic_timer.h>
79 
80 /*
81  *	Local Function Prototypes
82  */
83 static void apic_init_intr(void);
84 
85 /*
86  *	standard MP entries
87  */
88 static int	apic_probe(void);
89 static int	apic_getclkirq(int ipl);
90 static void	apic_init(void);
91 static void	apic_picinit(void);
92 static int	apic_post_cpu_start(void);
93 static int	apic_intr_enter(int ipl, int *vect);
94 static void	apic_setspl(int ipl);
95 static void	x2apic_setspl(int ipl);
96 static int	apic_addspl(int ipl, int vector, int min_ipl, int max_ipl);
97 static int	apic_delspl(int ipl, int vector, int min_ipl, int max_ipl);
98 static int	apic_disable_intr(processorid_t cpun);
99 static void	apic_enable_intr(processorid_t cpun);
100 static int		apic_get_ipivect(int ipl, int type);
101 static void	apic_post_cyclic_setup(void *arg);
102 
103 /*
104  * The following vector assignments influence the value of ipltopri and
105  * vectortoipl. Note that vectors 0 - 0x1f are not used. We can program
106  * idle to 0 and IPL 0 to 0xf to differentiate idle in case
107  * we care to do so in future. Note some IPLs which are rarely used
108  * will share the vector ranges and heavily used IPLs (5 and 6) have
109  * a wide range.
110  *
111  * This array is used to initialize apic_ipls[] (in apic_init()).
112  *
113  *	IPL		Vector range.		as passed to intr_enter
114  *	0		none.
115  *	1,2,3		0x20-0x2f		0x0-0xf
116  *	4		0x30-0x3f		0x10-0x1f
117  *	5		0x40-0x5f		0x20-0x3f
118  *	6		0x60-0x7f		0x40-0x5f
119  *	7,8,9		0x80-0x8f		0x60-0x6f
120  *	10		0x90-0x9f		0x70-0x7f
121  *	11		0xa0-0xaf		0x80-0x8f
122  *	...		...
123  *	15		0xe0-0xef		0xc0-0xcf
124  *	15		0xf0-0xff		0xd0-0xdf
125  */
126 uchar_t apic_vectortoipl[APIC_AVAIL_VECTOR / APIC_VECTOR_PER_IPL] = {
127 	3, 4, 5, 5, 6, 6, 9, 10, 11, 12, 13, 14, 15, 15
128 };
129 	/*
130 	 * The ipl of an ISR at vector X is apic_vectortoipl[X>>4]
131 	 * NOTE that this is vector as passed into intr_enter which is
132 	 * programmed vector - 0x20 (APIC_BASE_VECT)
133 	 */
134 
135 uchar_t	apic_ipltopri[MAXIPL + 1];	/* unix ipl to apic pri	*/
136 	/* The taskpri to be programmed into apic to mask given ipl */
137 
138 #if defined(__amd64)
139 uchar_t	apic_cr8pri[MAXIPL + 1];	/* unix ipl to cr8 pri	*/
140 #endif
141 
142 /*
143  * Correlation of the hardware vector to the IPL in use, initialized
144  * from apic_vectortoipl[] in apic_init().  The final IPLs may not correlate
145  * to the IPLs in apic_vectortoipl on some systems that share interrupt lines
146  * connected to errata-stricken IOAPICs
147  */
148 uchar_t apic_ipls[APIC_AVAIL_VECTOR];
149 
150 /*
151  * Patchable global variables.
152  */
153 int	apic_enable_hwsoftint = 0;	/* 0 - disable, 1 - enable	*/
154 int	apic_enable_bind_log = 1;	/* 1 - display interrupt binding log */
155 
156 /*
157  *	Local static data
158  */
159 static struct	psm_ops apic_ops = {
160 	apic_probe,
161 
162 	apic_init,
163 	apic_picinit,
164 	apic_intr_enter,
165 	apic_intr_exit,
166 	apic_setspl,
167 	apic_addspl,
168 	apic_delspl,
169 	apic_disable_intr,
170 	apic_enable_intr,
171 	(int (*)(int))NULL,		/* psm_softlvl_to_irq */
172 	(void (*)(int))NULL,		/* psm_set_softintr */
173 
174 	apic_set_idlecpu,
175 	apic_unset_idlecpu,
176 
177 	apic_clkinit,
178 	apic_getclkirq,
179 	(void (*)(void))NULL,		/* psm_hrtimeinit */
180 	apic_gethrtime,
181 
182 	apic_get_next_processorid,
183 	apic_cpu_start,
184 	apic_post_cpu_start,
185 	apic_shutdown,
186 	apic_get_ipivect,
187 	apic_send_ipi,
188 
189 	(int (*)(dev_info_t *, int))NULL,	/* psm_translate_irq */
190 	(void (*)(int, char *))NULL,	/* psm_notify_error */
191 	(void (*)(int))NULL,		/* psm_notify_func */
192 	apic_timer_reprogram,
193 	apic_timer_enable,
194 	apic_timer_disable,
195 	apic_post_cyclic_setup,
196 	apic_preshutdown,
197 	apic_intr_ops,			/* Advanced DDI Interrupt framework */
198 	apic_state,			/* save, restore apic state for S3 */
199 	apic_cpu_ops,			/* CPU control interface. */
200 };
201 
202 struct psm_ops *psmops = &apic_ops;
203 
204 static struct	psm_info apic_psm_info = {
205 	PSM_INFO_VER01_7,			/* version */
206 	PSM_OWN_EXCLUSIVE,			/* ownership */
207 	(struct psm_ops *)&apic_ops,		/* operation */
208 	APIC_PCPLUSMP_NAME,			/* machine name */
209 	"pcplusmp v1.4 compatible",
210 };
211 
212 static void *apic_hdlp;
213 
214 /*
215  * apic_let_idle_redistribute can have the following values:
216  * 0 - If clock decremented it from 1 to 0, clock has to call redistribute.
217  * apic_redistribute_lock prevents multiple idle cpus from redistributing
218  */
219 int	apic_num_idle_redistributions = 0;
220 static	int apic_let_idle_redistribute = 0;
221 
222 /* to gather intr data and redistribute */
223 static void apic_redistribute_compute(void);
224 
225 /*
226  *	This is the loadable module wrapper
227  */
228 
229 int
230 _init(void)
231 {
232 	if (apic_coarse_hrtime)
233 		apic_ops.psm_gethrtime = &apic_gettime;
234 	return (psm_mod_init(&apic_hdlp, &apic_psm_info));
235 }
236 
237 int
238 _fini(void)
239 {
240 	return (psm_mod_fini(&apic_hdlp, &apic_psm_info));
241 }
242 
243 int
244 _info(struct modinfo *modinfop)
245 {
246 	return (psm_mod_info(&apic_hdlp, &apic_psm_info, modinfop));
247 }
248 
249 static int
250 apic_probe(void)
251 {
252 	/* check if apix is initialized */
253 	if (apix_enable && apix_loaded())
254 		return (PSM_FAILURE);
255 	else
256 		apix_enable = 0; /* continue using pcplusmp PSM */
257 
258 	return (apic_probe_common(apic_psm_info.p_mach_idstring));
259 }
260 
261 static uchar_t
262 apic_xlate_vector_by_irq(uchar_t irq)
263 {
264 	if (apic_irq_table[irq] == NULL)
265 		return (0);
266 
267 	return (apic_irq_table[irq]->airq_vector);
268 }
269 
270 void
271 apic_init(void)
272 {
273 	int i;
274 	int	j = 1;
275 
276 	psm_get_ioapicid = apic_get_ioapicid;
277 	psm_get_localapicid = apic_get_localapicid;
278 	psm_xlate_vector_by_irq = apic_xlate_vector_by_irq;
279 
280 	apic_ipltopri[0] = APIC_VECTOR_PER_IPL; /* leave 0 for idle */
281 	for (i = 0; i < (APIC_AVAIL_VECTOR / APIC_VECTOR_PER_IPL); i++) {
282 		if ((i < ((APIC_AVAIL_VECTOR / APIC_VECTOR_PER_IPL) - 1)) &&
283 		    (apic_vectortoipl[i + 1] == apic_vectortoipl[i]))
284 			/* get to highest vector at the same ipl */
285 			continue;
286 		for (; j <= apic_vectortoipl[i]; j++) {
287 			apic_ipltopri[j] = (i << APIC_IPL_SHIFT) +
288 			    APIC_BASE_VECT;
289 		}
290 	}
291 	for (; j < MAXIPL + 1; j++)
292 		/* fill up any empty ipltopri slots */
293 		apic_ipltopri[j] = (i << APIC_IPL_SHIFT) + APIC_BASE_VECT;
294 	apic_init_common();
295 #if defined(__amd64)
296 	/*
297 	 * Make cpu-specific interrupt info point to cr8pri vector
298 	 */
299 	for (i = 0; i <= MAXIPL; i++)
300 		apic_cr8pri[i] = apic_ipltopri[i] >> APIC_IPL_SHIFT;
301 	CPU->cpu_pri_data = apic_cr8pri;
302 #else
303 	if (cpuid_have_cr8access(CPU))
304 		apic_have_32bit_cr8 = 1;
305 #endif	/* __amd64 */
306 }
307 
308 static void
309 apic_init_intr(void)
310 {
311 	processorid_t	cpun = psm_get_cpu_id();
312 	uint_t nlvt;
313 	uint32_t svr = AV_UNIT_ENABLE | APIC_SPUR_INTR;
314 
315 	apic_reg_ops->apic_write_task_reg(APIC_MASK_ALL);
316 
317 	if (apic_mode == LOCAL_APIC) {
318 		/*
319 		 * We are running APIC in MMIO mode.
320 		 */
321 		if (apic_flat_model) {
322 			apic_reg_ops->apic_write(APIC_FORMAT_REG,
323 			    APIC_FLAT_MODEL);
324 		} else {
325 			apic_reg_ops->apic_write(APIC_FORMAT_REG,
326 			    APIC_CLUSTER_MODEL);
327 		}
328 
329 		apic_reg_ops->apic_write(APIC_DEST_REG,
330 		    AV_HIGH_ORDER >> cpun);
331 	}
332 
333 	if (apic_directed_EOI_supported()) {
334 		/*
335 		 * Setting the 12th bit in the Spurious Interrupt Vector
336 		 * Register suppresses broadcast EOIs generated by the local
337 		 * APIC. The suppression of broadcast EOIs happens only when
338 		 * interrupts are level-triggered.
339 		 */
340 		svr |= APIC_SVR_SUPPRESS_BROADCAST_EOI;
341 	}
342 
343 	/* need to enable APIC before unmasking NMI */
344 	apic_reg_ops->apic_write(APIC_SPUR_INT_REG, svr);
345 
346 	/*
347 	 * Presence of an invalid vector with delivery mode AV_FIXED can
348 	 * cause an error interrupt, even if the entry is masked...so
349 	 * write a valid vector to LVT entries along with the mask bit
350 	 */
351 
352 	/* All APICs have timer and LINT0/1 */
353 	apic_reg_ops->apic_write(APIC_LOCAL_TIMER, AV_MASK|APIC_RESV_IRQ);
354 	apic_reg_ops->apic_write(APIC_INT_VECT0, AV_MASK|APIC_RESV_IRQ);
355 	apic_reg_ops->apic_write(APIC_INT_VECT1, AV_NMI);	/* enable NMI */
356 
357 	/*
358 	 * On integrated APICs, the number of LVT entries is
359 	 * 'Max LVT entry' + 1; on 82489DX's (non-integrated
360 	 * APICs), nlvt is "3" (LINT0, LINT1, and timer)
361 	 */
362 
363 	if (apic_cpus[cpun].aci_local_ver < APIC_INTEGRATED_VERS) {
364 		nlvt = 3;
365 	} else {
366 		nlvt = ((apic_reg_ops->apic_read(APIC_VERS_REG) >> 16) &
367 		    0xFF) + 1;
368 	}
369 
370 	if (nlvt >= 5) {
371 		/* Enable performance counter overflow interrupt */
372 
373 		if (!is_x86_feature(x86_featureset, X86FSET_MSR))
374 			apic_enable_cpcovf_intr = 0;
375 		if (apic_enable_cpcovf_intr) {
376 			if (apic_cpcovf_vect == 0) {
377 				int ipl = APIC_PCINT_IPL;
378 				int irq = apic_get_ipivect(ipl, -1);
379 
380 				ASSERT(irq != -1);
381 				apic_cpcovf_vect =
382 				    apic_irq_table[irq]->airq_vector;
383 				ASSERT(apic_cpcovf_vect);
384 				(void) add_avintr(NULL, ipl,
385 				    (avfunc)kcpc_hw_overflow_intr,
386 				    "apic pcint", irq, NULL, NULL, NULL, NULL);
387 				kcpc_hw_overflow_intr_installed = 1;
388 				kcpc_hw_enable_cpc_intr =
389 				    apic_cpcovf_mask_clear;
390 			}
391 			apic_reg_ops->apic_write(APIC_PCINT_VECT,
392 			    apic_cpcovf_vect);
393 		}
394 	}
395 
396 	if (nlvt >= 6) {
397 		/* Only mask TM intr if the BIOS apparently doesn't use it */
398 
399 		uint32_t lvtval;
400 
401 		lvtval = apic_reg_ops->apic_read(APIC_THERM_VECT);
402 		if (((lvtval & AV_MASK) == AV_MASK) ||
403 		    ((lvtval & AV_DELIV_MODE) != AV_SMI)) {
404 			apic_reg_ops->apic_write(APIC_THERM_VECT,
405 			    AV_MASK|APIC_RESV_IRQ);
406 		}
407 	}
408 
409 	/* Enable error interrupt */
410 
411 	if (nlvt >= 4 && apic_enable_error_intr) {
412 		if (apic_errvect == 0) {
413 			int ipl = 0xf;	/* get highest priority intr */
414 			int irq = apic_get_ipivect(ipl, -1);
415 
416 			ASSERT(irq != -1);
417 			apic_errvect = apic_irq_table[irq]->airq_vector;
418 			ASSERT(apic_errvect);
419 			/*
420 			 * Not PSMI compliant, but we are going to merge
421 			 * with ON anyway
422 			 */
423 			(void) add_avintr((void *)NULL, ipl,
424 			    (avfunc)apic_error_intr, "apic error intr",
425 			    irq, NULL, NULL, NULL, NULL);
426 		}
427 		apic_reg_ops->apic_write(APIC_ERR_VECT, apic_errvect);
428 		apic_reg_ops->apic_write(APIC_ERROR_STATUS, 0);
429 		apic_reg_ops->apic_write(APIC_ERROR_STATUS, 0);
430 	}
431 
432 	/* Enable CMCI interrupt */
433 	if (cmi_enable_cmci) {
434 
435 		mutex_enter(&cmci_cpu_setup_lock);
436 		if (cmci_cpu_setup_registered == 0) {
437 			mutex_enter(&cpu_lock);
438 			register_cpu_setup_func(cmci_cpu_setup, NULL);
439 			mutex_exit(&cpu_lock);
440 			cmci_cpu_setup_registered = 1;
441 		}
442 		mutex_exit(&cmci_cpu_setup_lock);
443 
444 		if (apic_cmci_vect == 0) {
445 			int ipl = 0x2;
446 			int irq = apic_get_ipivect(ipl, -1);
447 
448 			ASSERT(irq != -1);
449 			apic_cmci_vect = apic_irq_table[irq]->airq_vector;
450 			ASSERT(apic_cmci_vect);
451 
452 			(void) add_avintr(NULL, ipl,
453 			    (avfunc)cmi_cmci_trap,
454 			    "apic cmci intr", irq, NULL, NULL, NULL, NULL);
455 		}
456 		apic_reg_ops->apic_write(APIC_CMCI_VECT, apic_cmci_vect);
457 	}
458 }
459 
460 static void
461 apic_picinit(void)
462 {
463 	int i, j;
464 	uint_t isr;
465 
466 	/*
467 	 * Initialize and enable interrupt remapping before apic
468 	 * hardware initialization
469 	 */
470 	apic_intrmap_init(apic_mode);
471 
472 	/*
473 	 * On UniSys Model 6520, the BIOS leaves vector 0x20 isr
474 	 * bit on without clearing it with EOI.  Since softint
475 	 * uses vector 0x20 to interrupt itself, so softint will
476 	 * not work on this machine.  In order to fix this problem
477 	 * a check is made to verify all the isr bits are clear.
478 	 * If not, EOIs are issued to clear the bits.
479 	 */
480 	for (i = 7; i >= 1; i--) {
481 		isr = apic_reg_ops->apic_read(APIC_ISR_REG + (i * 4));
482 		if (isr != 0)
483 			for (j = 0; ((j < 32) && (isr != 0)); j++)
484 				if (isr & (1 << j)) {
485 					apic_reg_ops->apic_write(
486 					    APIC_EOI_REG, 0);
487 					isr &= ~(1 << j);
488 					apic_error |= APIC_ERR_BOOT_EOI;
489 				}
490 	}
491 
492 	/* set a flag so we know we have run apic_picinit() */
493 	apic_picinit_called = 1;
494 	LOCK_INIT_CLEAR(&apic_gethrtime_lock);
495 	LOCK_INIT_CLEAR(&apic_ioapic_lock);
496 	LOCK_INIT_CLEAR(&apic_error_lock);
497 	LOCK_INIT_CLEAR(&apic_mode_switch_lock);
498 
499 	picsetup();	 /* initialise the 8259 */
500 
501 	/* add nmi handler - least priority nmi handler */
502 	LOCK_INIT_CLEAR(&apic_nmi_lock);
503 
504 	if (!psm_add_nmintr(0, (avfunc) apic_nmi_intr,
505 	    "pcplusmp NMI handler", (caddr_t)NULL))
506 		cmn_err(CE_WARN, "pcplusmp: Unable to add nmi handler");
507 
508 	/*
509 	 * Check for directed-EOI capability in the local APIC.
510 	 */
511 	if (apic_directed_EOI_supported() == 1) {
512 		apic_set_directed_EOI_handler();
513 	}
514 
515 	apic_init_intr();
516 
517 	/* enable apic mode if imcr present */
518 	if (apic_imcrp) {
519 		outb(APIC_IMCR_P1, (uchar_t)APIC_IMCR_SELECT);
520 		outb(APIC_IMCR_P2, (uchar_t)APIC_IMCR_APIC);
521 	}
522 
523 	ioapic_init_intr(IOAPIC_MASK);
524 }
525 
526 #ifdef	DEBUG
527 void
528 apic_break(void)
529 {
530 }
531 #endif /* DEBUG */
532 
533 /*
534  * platform_intr_enter
535  *
536  *	Called at the beginning of the interrupt service routine to
537  *	mask all level equal to and below the interrupt priority
538  *	of the interrupting vector.  An EOI should be given to
539  *	the interrupt controller to enable other HW interrupts.
540  *
541  *	Return -1 for spurious interrupts
542  *
543  */
544 /*ARGSUSED*/
545 static int
546 apic_intr_enter(int ipl, int *vectorp)
547 {
548 	uchar_t vector;
549 	int nipl;
550 	int irq;
551 	ulong_t iflag;
552 	apic_cpus_info_t *cpu_infop;
553 
554 	/*
555 	 * The real vector delivered is (*vectorp + 0x20), but our caller
556 	 * subtracts 0x20 from the vector before passing it to us.
557 	 * (That's why APIC_BASE_VECT is 0x20.)
558 	 */
559 	vector = (uchar_t)*vectorp;
560 
561 	/* if interrupted by the clock, increment apic_nsec_since_boot */
562 	if (vector == apic_clkvect) {
563 		if (!apic_oneshot) {
564 			/* NOTE: this is not MT aware */
565 			apic_hrtime_stamp++;
566 			apic_nsec_since_boot += apic_nsec_per_intr;
567 			apic_hrtime_stamp++;
568 			last_count_read = apic_hertz_count;
569 			apic_redistribute_compute();
570 		}
571 
572 		/* We will avoid all the book keeping overhead for clock */
573 		nipl = apic_ipls[vector];
574 
575 		*vectorp = apic_vector_to_irq[vector + APIC_BASE_VECT];
576 		if (apic_mode == LOCAL_APIC) {
577 #if defined(__amd64)
578 			setcr8((ulong_t)(apic_ipltopri[nipl] >>
579 			    APIC_IPL_SHIFT));
580 #else
581 			if (apic_have_32bit_cr8)
582 				setcr8((ulong_t)(apic_ipltopri[nipl] >>
583 				    APIC_IPL_SHIFT));
584 			else
585 				LOCAL_APIC_WRITE_REG(APIC_TASK_REG,
586 				    (uint32_t)apic_ipltopri[nipl]);
587 #endif
588 			LOCAL_APIC_WRITE_REG(APIC_EOI_REG, 0);
589 		} else {
590 			X2APIC_WRITE(APIC_TASK_REG, apic_ipltopri[nipl]);
591 			X2APIC_WRITE(APIC_EOI_REG, 0);
592 		}
593 
594 		return (nipl);
595 	}
596 
597 	cpu_infop = &apic_cpus[psm_get_cpu_id()];
598 
599 	if (vector == (APIC_SPUR_INTR - APIC_BASE_VECT)) {
600 		cpu_infop->aci_spur_cnt++;
601 		return (APIC_INT_SPURIOUS);
602 	}
603 
604 	/* Check if the vector we got is really what we need */
605 	if (apic_revector_pending) {
606 		/*
607 		 * Disable interrupts for the duration of
608 		 * the vector translation to prevent a self-race for
609 		 * the apic_revector_lock.  This cannot be done
610 		 * in apic_xlate_vector because it is recursive and
611 		 * we want the vector translation to be atomic with
612 		 * respect to other (higher-priority) interrupts.
613 		 */
614 		iflag = intr_clear();
615 		vector = apic_xlate_vector(vector + APIC_BASE_VECT) -
616 		    APIC_BASE_VECT;
617 		intr_restore(iflag);
618 	}
619 
620 	nipl = apic_ipls[vector];
621 	*vectorp = irq = apic_vector_to_irq[vector + APIC_BASE_VECT];
622 
623 	if (apic_mode == LOCAL_APIC) {
624 #if defined(__amd64)
625 		setcr8((ulong_t)(apic_ipltopri[nipl] >> APIC_IPL_SHIFT));
626 #else
627 		if (apic_have_32bit_cr8)
628 			setcr8((ulong_t)(apic_ipltopri[nipl] >>
629 			    APIC_IPL_SHIFT));
630 		else
631 			LOCAL_APIC_WRITE_REG(APIC_TASK_REG,
632 			    (uint32_t)apic_ipltopri[nipl]);
633 #endif
634 	} else {
635 		X2APIC_WRITE(APIC_TASK_REG, apic_ipltopri[nipl]);
636 	}
637 
638 	cpu_infop->aci_current[nipl] = (uchar_t)irq;
639 	cpu_infop->aci_curipl = (uchar_t)nipl;
640 	cpu_infop->aci_ISR_in_progress |= 1 << nipl;
641 
642 	/*
643 	 * apic_level_intr could have been assimilated into the irq struct.
644 	 * but, having it as a character array is more efficient in terms of
645 	 * cache usage. So, we leave it as is.
646 	 */
647 	if (!apic_level_intr[irq]) {
648 		if (apic_mode == LOCAL_APIC) {
649 			LOCAL_APIC_WRITE_REG(APIC_EOI_REG, 0);
650 		} else {
651 			X2APIC_WRITE(APIC_EOI_REG, 0);
652 		}
653 	}
654 
655 #ifdef	DEBUG
656 	APIC_DEBUG_BUF_PUT(vector);
657 	APIC_DEBUG_BUF_PUT(irq);
658 	APIC_DEBUG_BUF_PUT(nipl);
659 	APIC_DEBUG_BUF_PUT(psm_get_cpu_id());
660 	if ((apic_stretch_interrupts) && (apic_stretch_ISR & (1 << nipl)))
661 		drv_usecwait(apic_stretch_interrupts);
662 
663 	if (apic_break_on_cpu == psm_get_cpu_id())
664 		apic_break();
665 #endif /* DEBUG */
666 	return (nipl);
667 }
668 
669 /*
670  * This macro is a common code used by MMIO local apic and X2APIC
671  * local apic.
672  */
673 #define	APIC_INTR_EXIT() \
674 { \
675 	cpu_infop = &apic_cpus[psm_get_cpu_id()]; \
676 	if (apic_level_intr[irq]) \
677 		apic_reg_ops->apic_send_eoi(irq); \
678 	cpu_infop->aci_curipl = (uchar_t)prev_ipl; \
679 	/* ISR above current pri could not be in progress */ \
680 	cpu_infop->aci_ISR_in_progress &= (2 << prev_ipl) - 1; \
681 }
682 
683 /*
684  * Any changes made to this function must also change X2APIC
685  * version of intr_exit.
686  */
687 void
688 apic_intr_exit(int prev_ipl, int irq)
689 {
690 	apic_cpus_info_t *cpu_infop;
691 
692 #if defined(__amd64)
693 	setcr8((ulong_t)apic_cr8pri[prev_ipl]);
694 #else
695 	if (apic_have_32bit_cr8)
696 		setcr8((ulong_t)(apic_ipltopri[prev_ipl] >> APIC_IPL_SHIFT));
697 	else
698 		apicadr[APIC_TASK_REG] = apic_ipltopri[prev_ipl];
699 #endif
700 
701 	APIC_INTR_EXIT();
702 }
703 
704 /*
705  * Same as apic_intr_exit() except it uses MSR rather than MMIO
706  * to access local apic registers.
707  */
708 void
709 x2apic_intr_exit(int prev_ipl, int irq)
710 {
711 	apic_cpus_info_t *cpu_infop;
712 
713 	X2APIC_WRITE(APIC_TASK_REG, apic_ipltopri[prev_ipl]);
714 	APIC_INTR_EXIT();
715 }
716 
717 intr_exit_fn_t
718 psm_intr_exit_fn(void)
719 {
720 	if (apic_mode == LOCAL_X2APIC)
721 		return (x2apic_intr_exit);
722 
723 	return (apic_intr_exit);
724 }
725 
726 /*
727  * Mask all interrupts below or equal to the given IPL.
728  * Any changes made to this function must also change X2APIC
729  * version of setspl.
730  */
731 static void
732 apic_setspl(int ipl)
733 {
734 #if defined(__amd64)
735 	setcr8((ulong_t)apic_cr8pri[ipl]);
736 #else
737 	if (apic_have_32bit_cr8)
738 		setcr8((ulong_t)(apic_ipltopri[ipl] >> APIC_IPL_SHIFT));
739 	else
740 		apicadr[APIC_TASK_REG] = apic_ipltopri[ipl];
741 #endif
742 
743 	/* interrupts at ipl above this cannot be in progress */
744 	apic_cpus[psm_get_cpu_id()].aci_ISR_in_progress &= (2 << ipl) - 1;
745 	/*
746 	 * this is a patch fix for the ALR QSMP P5 machine, so that interrupts
747 	 * have enough time to come in before the priority is raised again
748 	 * during the idle() loop.
749 	 */
750 	if (apic_setspl_delay)
751 		(void) apic_reg_ops->apic_get_pri();
752 }
753 
754 /*
755  * X2APIC version of setspl.
756  * Mask all interrupts below or equal to the given IPL
757  */
758 static void
759 x2apic_setspl(int ipl)
760 {
761 	X2APIC_WRITE(APIC_TASK_REG, apic_ipltopri[ipl]);
762 
763 	/* interrupts at ipl above this cannot be in progress */
764 	apic_cpus[psm_get_cpu_id()].aci_ISR_in_progress &= (2 << ipl) - 1;
765 }
766 
767 /*ARGSUSED*/
768 static int
769 apic_addspl(int irqno, int ipl, int min_ipl, int max_ipl)
770 {
771 	return (apic_addspl_common(irqno, ipl, min_ipl, max_ipl));
772 }
773 
774 static int
775 apic_delspl(int irqno, int ipl, int min_ipl, int max_ipl)
776 {
777 	return (apic_delspl_common(irqno, ipl, min_ipl,  max_ipl));
778 }
779 
780 static int
781 apic_post_cpu_start(void)
782 {
783 	int cpun;
784 	static int cpus_started = 1;
785 
786 	/* We know this CPU + BSP  started successfully. */
787 	cpus_started++;
788 
789 	/*
790 	 * On BSP we would have enabled X2APIC, if supported by processor,
791 	 * in acpi_probe(), but on AP we do it here.
792 	 *
793 	 * We enable X2APIC mode only if BSP is running in X2APIC & the
794 	 * local APIC mode of the current CPU is MMIO (xAPIC).
795 	 */
796 	if (apic_mode == LOCAL_X2APIC && apic_detect_x2apic() &&
797 	    apic_local_mode() == LOCAL_APIC) {
798 		apic_enable_x2apic();
799 	}
800 
801 	/*
802 	 * Switch back to x2apic IPI sending method for performance when target
803 	 * CPU has entered x2apic mode.
804 	 */
805 	if (apic_mode == LOCAL_X2APIC) {
806 		apic_switch_ipi_callback(B_FALSE);
807 	}
808 
809 	splx(ipltospl(LOCK_LEVEL));
810 	apic_init_intr();
811 
812 	/*
813 	 * since some systems don't enable the internal cache on the non-boot
814 	 * cpus, so we have to enable them here
815 	 */
816 	setcr0(getcr0() & ~(CR0_CD | CR0_NW));
817 
818 #ifdef	DEBUG
819 	APIC_AV_PENDING_SET();
820 #else
821 	if (apic_mode == LOCAL_APIC)
822 		APIC_AV_PENDING_SET();
823 #endif	/* DEBUG */
824 
825 	/*
826 	 * We may be booting, or resuming from suspend; aci_status will
827 	 * be APIC_CPU_INTR_ENABLE if coming from suspend, so we add the
828 	 * APIC_CPU_ONLINE flag here rather than setting aci_status completely.
829 	 */
830 	cpun = psm_get_cpu_id();
831 	apic_cpus[cpun].aci_status |= APIC_CPU_ONLINE;
832 
833 	apic_reg_ops->apic_write(APIC_DIVIDE_REG, apic_divide_reg_init);
834 	return (PSM_SUCCESS);
835 }
836 
837 /*
838  * type == -1 indicates it is an internal request. Do not change
839  * resv_vector for these requests
840  */
841 static int
842 apic_get_ipivect(int ipl, int type)
843 {
844 	uchar_t vector;
845 	int irq;
846 
847 	if ((irq = apic_allocate_irq(APIC_VECTOR(ipl))) != -1) {
848 		if (vector = apic_allocate_vector(ipl, irq, 1)) {
849 			apic_irq_table[irq]->airq_mps_intr_index =
850 			    RESERVE_INDEX;
851 			apic_irq_table[irq]->airq_vector = vector;
852 			if (type != -1) {
853 				apic_resv_vector[ipl] = vector;
854 			}
855 			return (irq);
856 		}
857 	}
858 	apic_error |= APIC_ERR_GET_IPIVECT_FAIL;
859 	return (-1);	/* shouldn't happen */
860 }
861 
862 static int
863 apic_getclkirq(int ipl)
864 {
865 	int	irq;
866 
867 	if ((irq = apic_get_ipivect(ipl, -1)) == -1)
868 		return (-1);
869 	/*
870 	 * Note the vector in apic_clkvect for per clock handling.
871 	 */
872 	apic_clkvect = apic_irq_table[irq]->airq_vector - APIC_BASE_VECT;
873 	APIC_VERBOSE_IOAPIC((CE_NOTE, "get_clkirq: vector = %x\n",
874 	    apic_clkvect));
875 	return (irq);
876 }
877 
878 /*
879  * Try and disable all interrupts. We just assign interrupts to other
880  * processors based on policy. If any were bound by user request, we
881  * let them continue and return failure. We do not bother to check
882  * for cache affinity while rebinding.
883  */
884 
885 static int
886 apic_disable_intr(processorid_t cpun)
887 {
888 	int bind_cpu = 0, i, hardbound = 0;
889 	apic_irq_t *irq_ptr;
890 	ulong_t iflag;
891 
892 	iflag = intr_clear();
893 	lock_set(&apic_ioapic_lock);
894 
895 	for (i = 0; i <= APIC_MAX_VECTOR; i++) {
896 		if (apic_reprogram_info[i].done == B_FALSE) {
897 			if (apic_reprogram_info[i].bindcpu == cpun) {
898 				/*
899 				 * CPU is busy -- it's the target of
900 				 * a pending reprogramming attempt
901 				 */
902 				lock_clear(&apic_ioapic_lock);
903 				intr_restore(iflag);
904 				return (PSM_FAILURE);
905 			}
906 		}
907 	}
908 
909 	apic_cpus[cpun].aci_status &= ~APIC_CPU_INTR_ENABLE;
910 
911 	apic_cpus[cpun].aci_curipl = 0;
912 
913 	i = apic_min_device_irq;
914 	for (; i <= apic_max_device_irq; i++) {
915 		/*
916 		 * If there are bound interrupts on this cpu, then
917 		 * rebind them to other processors.
918 		 */
919 		if ((irq_ptr = apic_irq_table[i]) != NULL) {
920 			ASSERT((irq_ptr->airq_temp_cpu == IRQ_UNBOUND) ||
921 			    (irq_ptr->airq_temp_cpu == IRQ_UNINIT) ||
922 			    (apic_cpu_in_range(irq_ptr->airq_temp_cpu)));
923 
924 			if (irq_ptr->airq_temp_cpu == (cpun | IRQ_USER_BOUND)) {
925 				hardbound = 1;
926 				continue;
927 			}
928 
929 			if (irq_ptr->airq_temp_cpu == cpun) {
930 				do {
931 					bind_cpu =
932 					    apic_find_cpu(APIC_CPU_INTR_ENABLE);
933 				} while (apic_rebind_all(irq_ptr, bind_cpu));
934 			}
935 		}
936 	}
937 
938 	lock_clear(&apic_ioapic_lock);
939 	intr_restore(iflag);
940 
941 	if (hardbound) {
942 		cmn_err(CE_WARN, "Could not disable interrupts on %d"
943 		    "due to user bound interrupts", cpun);
944 		return (PSM_FAILURE);
945 	}
946 	else
947 		return (PSM_SUCCESS);
948 }
949 
950 /*
951  * Bind interrupts to the CPU's local APIC.
952  * Interrupts should not be bound to a CPU's local APIC until the CPU
953  * is ready to receive interrupts.
954  */
955 static void
956 apic_enable_intr(processorid_t cpun)
957 {
958 	int	i;
959 	apic_irq_t *irq_ptr;
960 	ulong_t iflag;
961 
962 	iflag = intr_clear();
963 	lock_set(&apic_ioapic_lock);
964 
965 	apic_cpus[cpun].aci_status |= APIC_CPU_INTR_ENABLE;
966 
967 	i = apic_min_device_irq;
968 	for (i = apic_min_device_irq; i <= apic_max_device_irq; i++) {
969 		if ((irq_ptr = apic_irq_table[i]) != NULL) {
970 			if ((irq_ptr->airq_cpu & ~IRQ_USER_BOUND) == cpun) {
971 				(void) apic_rebind_all(irq_ptr,
972 				    irq_ptr->airq_cpu);
973 			}
974 		}
975 	}
976 
977 	if (apic_cpus[cpun].aci_status & APIC_CPU_SUSPEND)
978 		apic_cpus[cpun].aci_status &= ~APIC_CPU_SUSPEND;
979 
980 	lock_clear(&apic_ioapic_lock);
981 	intr_restore(iflag);
982 }
983 
984 /*
985  * If this module needs a periodic handler for the interrupt distribution, it
986  * can be added here. The argument to the periodic handler is not currently
987  * used, but is reserved for future.
988  */
989 static void
990 apic_post_cyclic_setup(void *arg)
991 {
992 _NOTE(ARGUNUSED(arg))
993 	/* cpu_lock is held */
994 	/* set up a periodic handler for intr redistribution */
995 
996 	/*
997 	 * In peridoc mode intr redistribution processing is done in
998 	 * apic_intr_enter during clk intr processing
999 	 */
1000 	if (!apic_oneshot)
1001 		return;
1002 	/*
1003 	 * Register a periodical handler for the redistribution processing.
1004 	 * On X86, CY_LOW_LEVEL is mapped to the level 2 interrupt, so
1005 	 * DDI_IPL_2 should be passed to ddi_periodic_add() here.
1006 	 */
1007 	apic_periodic_id = ddi_periodic_add(
1008 	    (void (*)(void *))apic_redistribute_compute, NULL,
1009 	    apic_redistribute_sample_interval, DDI_IPL_2);
1010 }
1011 
1012 static void
1013 apic_redistribute_compute(void)
1014 {
1015 	int	i, j, max_busy;
1016 
1017 	if (apic_enable_dynamic_migration) {
1018 		if (++apic_nticks == apic_sample_factor_redistribution) {
1019 			/*
1020 			 * Time to call apic_intr_redistribute().
1021 			 * reset apic_nticks. This will cause max_busy
1022 			 * to be calculated below and if it is more than
1023 			 * apic_int_busy, we will do the whole thing
1024 			 */
1025 			apic_nticks = 0;
1026 		}
1027 		max_busy = 0;
1028 		for (i = 0; i < apic_nproc; i++) {
1029 			if (!apic_cpu_in_range(i))
1030 				continue;
1031 
1032 			/*
1033 			 * Check if curipl is non zero & if ISR is in
1034 			 * progress
1035 			 */
1036 			if (((j = apic_cpus[i].aci_curipl) != 0) &&
1037 			    (apic_cpus[i].aci_ISR_in_progress & (1 << j))) {
1038 
1039 				int	irq;
1040 				apic_cpus[i].aci_busy++;
1041 				irq = apic_cpus[i].aci_current[j];
1042 				apic_irq_table[irq]->airq_busy++;
1043 			}
1044 
1045 			if (!apic_nticks &&
1046 			    (apic_cpus[i].aci_busy > max_busy))
1047 				max_busy = apic_cpus[i].aci_busy;
1048 		}
1049 		if (!apic_nticks) {
1050 			if (max_busy > apic_int_busy_mark) {
1051 			/*
1052 			 * We could make the following check be
1053 			 * skipped > 1 in which case, we get a
1054 			 * redistribution at half the busy mark (due to
1055 			 * double interval). Need to be able to collect
1056 			 * more empirical data to decide if that is a
1057 			 * good strategy. Punt for now.
1058 			 */
1059 				if (apic_skipped_redistribute) {
1060 					apic_cleanup_busy();
1061 					apic_skipped_redistribute = 0;
1062 				} else {
1063 					apic_intr_redistribute();
1064 				}
1065 			} else
1066 				apic_skipped_redistribute++;
1067 		}
1068 	}
1069 }
1070 
1071 
1072 /*
1073  * The following functions are in the platform specific file so that they
1074  * can be different functions depending on whether we are running on
1075  * bare metal or a hypervisor.
1076  */
1077 
1078 /*
1079  * Check to make sure there are enough irq slots
1080  */
1081 int
1082 apic_check_free_irqs(int count)
1083 {
1084 	int i, avail;
1085 
1086 	avail = 0;
1087 	for (i = APIC_FIRST_FREE_IRQ; i < APIC_RESV_IRQ; i++) {
1088 		if ((apic_irq_table[i] == NULL) ||
1089 		    apic_irq_table[i]->airq_mps_intr_index == FREE_INDEX) {
1090 			if (++avail >= count)
1091 				return (PSM_SUCCESS);
1092 		}
1093 	}
1094 	return (PSM_FAILURE);
1095 }
1096 
1097 /*
1098  * This function allocates "count" MSI vector(s) for the given "dip/pri/type"
1099  */
1100 int
1101 apic_alloc_msi_vectors(dev_info_t *dip, int inum, int count, int pri,
1102     int behavior)
1103 {
1104 	int	rcount, i;
1105 	uchar_t	start, irqno;
1106 	uint32_t cpu;
1107 	major_t	major;
1108 	apic_irq_t	*irqptr;
1109 
1110 	DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_msi_vectors: dip=0x%p "
1111 	    "inum=0x%x  pri=0x%x count=0x%x behavior=%d\n",
1112 	    (void *)dip, inum, pri, count, behavior));
1113 
1114 	if (count > 1) {
1115 		if (behavior == DDI_INTR_ALLOC_STRICT &&
1116 		    apic_multi_msi_enable == 0)
1117 			return (0);
1118 		if (apic_multi_msi_enable == 0)
1119 			count = 1;
1120 	}
1121 
1122 	if ((rcount = apic_navail_vector(dip, pri)) > count)
1123 		rcount = count;
1124 	else if (rcount == 0 || (rcount < count &&
1125 	    behavior == DDI_INTR_ALLOC_STRICT))
1126 		return (0);
1127 
1128 	/* if not ISP2, then round it down */
1129 	if (!ISP2(rcount))
1130 		rcount = 1 << (highbit(rcount) - 1);
1131 
1132 	mutex_enter(&airq_mutex);
1133 
1134 	for (start = 0; rcount > 0; rcount >>= 1) {
1135 		if ((start = apic_find_multi_vectors(pri, rcount)) != 0 ||
1136 		    behavior == DDI_INTR_ALLOC_STRICT)
1137 			break;
1138 	}
1139 
1140 	if (start == 0) {
1141 		/* no vector available */
1142 		mutex_exit(&airq_mutex);
1143 		return (0);
1144 	}
1145 
1146 	if (apic_check_free_irqs(rcount) == PSM_FAILURE) {
1147 		/* not enough free irq slots available */
1148 		mutex_exit(&airq_mutex);
1149 		return (0);
1150 	}
1151 
1152 	major = (dip != NULL) ? ddi_driver_major(dip) : 0;
1153 	for (i = 0; i < rcount; i++) {
1154 		if ((irqno = apic_allocate_irq(apic_first_avail_irq)) ==
1155 		    (uchar_t)-1) {
1156 			/*
1157 			 * shouldn't happen because of the
1158 			 * apic_check_free_irqs() check earlier
1159 			 */
1160 			mutex_exit(&airq_mutex);
1161 			DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_msi_vectors: "
1162 			    "apic_allocate_irq failed\n"));
1163 			return (i);
1164 		}
1165 		apic_max_device_irq = max(irqno, apic_max_device_irq);
1166 		apic_min_device_irq = min(irqno, apic_min_device_irq);
1167 		irqptr = apic_irq_table[irqno];
1168 #ifdef	DEBUG
1169 		if (apic_vector_to_irq[start + i] != APIC_RESV_IRQ)
1170 			DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_msi_vectors: "
1171 			    "apic_vector_to_irq is not APIC_RESV_IRQ\n"));
1172 #endif
1173 		apic_vector_to_irq[start + i] = (uchar_t)irqno;
1174 
1175 		irqptr->airq_vector = (uchar_t)(start + i);
1176 		irqptr->airq_ioapicindex = (uchar_t)inum;	/* start */
1177 		irqptr->airq_intin_no = (uchar_t)rcount;
1178 		irqptr->airq_ipl = pri;
1179 		irqptr->airq_vector = start + i;
1180 		irqptr->airq_origirq = (uchar_t)(inum + i);
1181 		irqptr->airq_share_id = 0;
1182 		irqptr->airq_mps_intr_index = MSI_INDEX;
1183 		irqptr->airq_dip = dip;
1184 		irqptr->airq_major = major;
1185 		if (i == 0) /* they all bound to the same cpu */
1186 			cpu = irqptr->airq_cpu = apic_bind_intr(dip, irqno,
1187 			    0xff, 0xff);
1188 		else
1189 			irqptr->airq_cpu = cpu;
1190 		DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_msi_vectors: irq=0x%x "
1191 		    "dip=0x%p vector=0x%x origirq=0x%x pri=0x%x\n", irqno,
1192 		    (void *)irqptr->airq_dip, irqptr->airq_vector,
1193 		    irqptr->airq_origirq, pri));
1194 	}
1195 	mutex_exit(&airq_mutex);
1196 	return (rcount);
1197 }
1198 
1199 /*
1200  * This function allocates "count" MSI-X vector(s) for the given "dip/pri/type"
1201  */
1202 int
1203 apic_alloc_msix_vectors(dev_info_t *dip, int inum, int count, int pri,
1204     int behavior)
1205 {
1206 	int	rcount, i;
1207 	major_t	major;
1208 
1209 	mutex_enter(&airq_mutex);
1210 
1211 	if ((rcount = apic_navail_vector(dip, pri)) > count)
1212 		rcount = count;
1213 	else if (rcount == 0 || (rcount < count &&
1214 	    behavior == DDI_INTR_ALLOC_STRICT)) {
1215 		rcount = 0;
1216 		goto out;
1217 	}
1218 
1219 	if (apic_check_free_irqs(rcount) == PSM_FAILURE) {
1220 		/* not enough free irq slots available */
1221 		rcount = 0;
1222 		goto out;
1223 	}
1224 
1225 	major = (dip != NULL) ? ddi_driver_major(dip) : 0;
1226 	for (i = 0; i < rcount; i++) {
1227 		uchar_t	vector, irqno;
1228 		apic_irq_t	*irqptr;
1229 
1230 		if ((irqno = apic_allocate_irq(apic_first_avail_irq)) ==
1231 		    (uchar_t)-1) {
1232 			/*
1233 			 * shouldn't happen because of the
1234 			 * apic_check_free_irqs() check earlier
1235 			 */
1236 			DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_msix_vectors: "
1237 			    "apic_allocate_irq failed\n"));
1238 			rcount = i;
1239 			goto out;
1240 		}
1241 		if ((vector = apic_allocate_vector(pri, irqno, 1)) == 0) {
1242 			/*
1243 			 * shouldn't happen because of the
1244 			 * apic_navail_vector() call earlier
1245 			 */
1246 			DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_msix_vectors: "
1247 			    "apic_allocate_vector failed\n"));
1248 			rcount = i;
1249 			goto out;
1250 		}
1251 		apic_max_device_irq = max(irqno, apic_max_device_irq);
1252 		apic_min_device_irq = min(irqno, apic_min_device_irq);
1253 		irqptr = apic_irq_table[irqno];
1254 		irqptr->airq_vector = (uchar_t)vector;
1255 		irqptr->airq_ipl = pri;
1256 		irqptr->airq_origirq = (uchar_t)(inum + i);
1257 		irqptr->airq_share_id = 0;
1258 		irqptr->airq_mps_intr_index = MSIX_INDEX;
1259 		irqptr->airq_dip = dip;
1260 		irqptr->airq_major = major;
1261 		irqptr->airq_cpu = apic_bind_intr(dip, irqno, 0xff, 0xff);
1262 	}
1263 out:
1264 	mutex_exit(&airq_mutex);
1265 	return (rcount);
1266 }
1267 
1268 /*
1269  * Allocate a free vector for irq at ipl. Takes care of merging of multiple
1270  * IPLs into a single APIC level as well as stretching some IPLs onto multiple
1271  * levels. APIC_HI_PRI_VECTS interrupts are reserved for high priority
1272  * requests and allocated only when pri is set.
1273  */
1274 uchar_t
1275 apic_allocate_vector(int ipl, int irq, int pri)
1276 {
1277 	int	lowest, highest, i;
1278 
1279 	highest = apic_ipltopri[ipl] + APIC_VECTOR_MASK;
1280 	lowest = apic_ipltopri[ipl - 1] + APIC_VECTOR_PER_IPL;
1281 
1282 	if (highest < lowest) /* Both ipl and ipl - 1 map to same pri */
1283 		lowest -= APIC_VECTOR_PER_IPL;
1284 
1285 #ifdef	DEBUG
1286 	if (apic_restrict_vector)	/* for testing shared interrupt logic */
1287 		highest = lowest + apic_restrict_vector + APIC_HI_PRI_VECTS;
1288 #endif /* DEBUG */
1289 	if (pri == 0)
1290 		highest -= APIC_HI_PRI_VECTS;
1291 
1292 	for (i = lowest; i <= highest; i++) {
1293 		if (APIC_CHECK_RESERVE_VECTORS(i))
1294 			continue;
1295 		if (apic_vector_to_irq[i] == APIC_RESV_IRQ) {
1296 			apic_vector_to_irq[i] = (uchar_t)irq;
1297 			return (i);
1298 		}
1299 	}
1300 
1301 	return (0);
1302 }
1303 
1304 /* Mark vector as not being used by any irq */
1305 void
1306 apic_free_vector(uchar_t vector)
1307 {
1308 	apic_vector_to_irq[vector] = APIC_RESV_IRQ;
1309 }
1310 
1311 /*
1312  * Call rebind to do the actual programming.
1313  * Must be called with interrupts disabled and apic_ioapic_lock held
1314  * 'p' is polymorphic -- if this function is called to process a deferred
1315  * reprogramming, p is of type 'struct ioapic_reprogram_data *', from which
1316  * the irq pointer is retrieved.  If not doing deferred reprogramming,
1317  * p is of the type 'apic_irq_t *'.
1318  *
1319  * apic_ioapic_lock must be held across this call, as it protects apic_rebind
1320  * and it protects apic_get_next_bind_cpu() from a race in which a CPU can be
1321  * taken offline after a cpu is selected, but before apic_rebind is called to
1322  * bind interrupts to it.
1323  */
1324 int
1325 apic_setup_io_intr(void *p, int irq, boolean_t deferred)
1326 {
1327 	apic_irq_t *irqptr;
1328 	struct ioapic_reprogram_data *drep = NULL;
1329 	int rv;
1330 
1331 	if (deferred) {
1332 		drep = (struct ioapic_reprogram_data *)p;
1333 		ASSERT(drep != NULL);
1334 		irqptr = drep->irqp;
1335 	} else
1336 		irqptr = (apic_irq_t *)p;
1337 
1338 	ASSERT(irqptr != NULL);
1339 
1340 	rv = apic_rebind(irqptr, apic_irq_table[irq]->airq_cpu, drep);
1341 	if (rv) {
1342 		/*
1343 		 * CPU is not up or interrupts are disabled. Fall back to
1344 		 * the first available CPU
1345 		 */
1346 		rv = apic_rebind(irqptr, apic_find_cpu(APIC_CPU_INTR_ENABLE),
1347 		    drep);
1348 	}
1349 
1350 	return (rv);
1351 }
1352 
1353 
1354 uchar_t
1355 apic_modify_vector(uchar_t vector, int irq)
1356 {
1357 	apic_vector_to_irq[vector] = (uchar_t)irq;
1358 	return (vector);
1359 }
1360 
1361 char *
1362 apic_get_apic_type(void)
1363 {
1364 	return (apic_psm_info.p_mach_idstring);
1365 }
1366 
1367 void
1368 x2apic_update_psm(void)
1369 {
1370 	struct psm_ops *pops = &apic_ops;
1371 
1372 	ASSERT(pops != NULL);
1373 
1374 	pops->psm_intr_exit = x2apic_intr_exit;
1375 	pops->psm_setspl = x2apic_setspl;
1376 
1377 	pops->psm_send_ipi =  x2apic_send_ipi;
1378 	send_dirintf = pops->psm_send_ipi;
1379 
1380 	apic_mode = LOCAL_X2APIC;
1381 	apic_change_ops();
1382 }
1383