xref: /illumos-gate/usr/src/uts/i86pc/io/pcplusmp/apic.c (revision 6d02032db7b674f185405d42cc8bf10a46a9ab3a)
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 2010 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 /*
27  * Copyright (c) 2010, Intel Corporation.
28  * All rights reserved.
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 
78 /*
79  *	Local Function Prototypes
80  */
81 static void apic_init_intr();
82 static void apic_nmi_intr(caddr_t arg, struct regs *rp);
83 
84 /*
85  *	standard MP entries
86  */
87 static int	apic_probe();
88 static int	apic_clkinit();
89 static int	apic_getclkirq(int ipl);
90 static uint_t	apic_calibrate(volatile uint32_t *addr,
91     uint16_t *pit_ticks_adj);
92 static hrtime_t apic_gettime();
93 static hrtime_t apic_gethrtime();
94 static void	apic_init();
95 static void	apic_picinit(void);
96 static int	apic_cpu_start(processorid_t cpuid, caddr_t ctx);
97 static int	apic_cpu_stop(processorid_t cpuid, caddr_t ctx);
98 static int	apic_cpu_add(psm_cpu_request_t *reqp);
99 static int	apic_cpu_remove(psm_cpu_request_t *reqp);
100 static int	apic_cpu_ops(psm_cpu_request_t *reqp);
101 static int	apic_post_cpu_start(void);
102 static void	apic_send_ipi(int cpun, int ipl);
103 static void	apic_set_idlecpu(processorid_t cpun);
104 static void	apic_unset_idlecpu(processorid_t cpun);
105 static int	apic_intr_enter(int ipl, int *vect);
106 static void	apic_setspl(int ipl);
107 static void	x2apic_setspl(int ipl);
108 static void	apic_switch_ipi_callback(boolean_t enter);
109 static int	apic_addspl(int ipl, int vector, int min_ipl, int max_ipl);
110 static int	apic_delspl(int ipl, int vector, int min_ipl, int max_ipl);
111 static void	apic_shutdown(int cmd, int fcn);
112 static void	apic_preshutdown(int cmd, int fcn);
113 static int	apic_disable_intr(processorid_t cpun);
114 static void	apic_enable_intr(processorid_t cpun);
115 static processorid_t	apic_get_next_processorid(processorid_t cpun);
116 static int		apic_get_ipivect(int ipl, int type);
117 static void	apic_timer_reprogram(hrtime_t time);
118 static void	apic_timer_enable(void);
119 static void	apic_timer_disable(void);
120 static void	apic_post_cyclic_setup(void *arg);
121 static void	apic_intrmap_init(int apic_mode);
122 static void	apic_record_ioapic_rdt(apic_irq_t *irq_ptr, ioapic_rdt_t *irdt);
123 static void	apic_record_msi(apic_irq_t *irq_ptr, msi_regs_t *mregs);
124 
125 static int	apic_oneshot = 0;
126 int	apic_oneshot_enable = 1; /* to allow disabling one-shot capability */
127 
128 /* Now the ones for Dynamic Interrupt distribution */
129 int	apic_enable_dynamic_migration = 0;
130 
131 extern int apic_have_32bit_cr8;
132 
133 /*
134  * These variables are frequently accessed in apic_intr_enter(),
135  * apic_intr_exit and apic_setspl, so group them together
136  */
137 volatile uint32_t *apicadr =  NULL;	/* virtual addr of local APIC	*/
138 int apic_setspl_delay = 1;		/* apic_setspl - delay enable	*/
139 int apic_clkvect;
140 
141 /* vector at which error interrupts come in */
142 int apic_errvect;
143 int apic_enable_error_intr = 1;
144 int apic_error_display_delay = 100;
145 
146 /* vector at which performance counter overflow interrupts come in */
147 int apic_cpcovf_vect;
148 int apic_enable_cpcovf_intr = 1;
149 
150 /* maximum loop count when sending Start IPIs. */
151 int apic_sipi_max_loop_count = 0x1000;
152 
153 /* vector at which CMCI interrupts come in */
154 int apic_cmci_vect;
155 extern int cmi_enable_cmci;
156 extern void cmi_cmci_trap(void);
157 
158 static kmutex_t cmci_cpu_setup_lock;	/* protects cmci_cpu_setup_registered */
159 static int cmci_cpu_setup_registered;
160 
161 /* number of CPUs in power-on transition state */
162 static int apic_poweron_cnt = 0;
163 static lock_t apic_mode_switch_lock;
164 
165 /*
166  * The following vector assignments influence the value of ipltopri and
167  * vectortoipl. Note that vectors 0 - 0x1f are not used. We can program
168  * idle to 0 and IPL 0 to 0xf to differentiate idle in case
169  * we care to do so in future. Note some IPLs which are rarely used
170  * will share the vector ranges and heavily used IPLs (5 and 6) have
171  * a wide range.
172  *
173  * This array is used to initialize apic_ipls[] (in apic_init()).
174  *
175  *	IPL		Vector range.		as passed to intr_enter
176  *	0		none.
177  *	1,2,3		0x20-0x2f		0x0-0xf
178  *	4		0x30-0x3f		0x10-0x1f
179  *	5		0x40-0x5f		0x20-0x3f
180  *	6		0x60-0x7f		0x40-0x5f
181  *	7,8,9		0x80-0x8f		0x60-0x6f
182  *	10		0x90-0x9f		0x70-0x7f
183  *	11		0xa0-0xaf		0x80-0x8f
184  *	...		...
185  *	15		0xe0-0xef		0xc0-0xcf
186  *	15		0xf0-0xff		0xd0-0xdf
187  */
188 uchar_t apic_vectortoipl[APIC_AVAIL_VECTOR / APIC_VECTOR_PER_IPL] = {
189 	3, 4, 5, 5, 6, 6, 9, 10, 11, 12, 13, 14, 15, 15
190 };
191 	/*
192 	 * The ipl of an ISR at vector X is apic_vectortoipl[X>>4]
193 	 * NOTE that this is vector as passed into intr_enter which is
194 	 * programmed vector - 0x20 (APIC_BASE_VECT)
195 	 */
196 
197 uchar_t	apic_ipltopri[MAXIPL + 1];	/* unix ipl to apic pri	*/
198 	/* The taskpri to be programmed into apic to mask given ipl */
199 
200 #if defined(__amd64)
201 uchar_t	apic_cr8pri[MAXIPL + 1];	/* unix ipl to cr8 pri	*/
202 #endif
203 
204 /*
205  * Correlation of the hardware vector to the IPL in use, initialized
206  * from apic_vectortoipl[] in apic_init().  The final IPLs may not correlate
207  * to the IPLs in apic_vectortoipl on some systems that share interrupt lines
208  * connected to errata-stricken IOAPICs
209  */
210 uchar_t apic_ipls[APIC_AVAIL_VECTOR];
211 
212 /*
213  * Patchable global variables.
214  */
215 int	apic_forceload = 0;
216 
217 int	apic_coarse_hrtime = 1;		/* 0 - use accurate slow gethrtime() */
218 					/* 1 - use gettime() for performance */
219 int	apic_flat_model = 0;		/* 0 - clustered. 1 - flat */
220 int	apic_enable_hwsoftint = 0;	/* 0 - disable, 1 - enable	*/
221 int	apic_enable_bind_log = 1;	/* 1 - display interrupt binding log */
222 int	apic_panic_on_nmi = 0;
223 int	apic_panic_on_apic_error = 0;
224 
225 int	apic_verbose = 0;
226 
227 /* minimum number of timer ticks to program to */
228 int apic_min_timer_ticks = 1;
229 
230 /*
231  *	Local static data
232  */
233 static struct	psm_ops apic_ops = {
234 	apic_probe,
235 
236 	apic_init,
237 	apic_picinit,
238 	apic_intr_enter,
239 	apic_intr_exit,
240 	apic_setspl,
241 	apic_addspl,
242 	apic_delspl,
243 	apic_disable_intr,
244 	apic_enable_intr,
245 	(int (*)(int))NULL,		/* psm_softlvl_to_irq */
246 	(void (*)(int))NULL,		/* psm_set_softintr */
247 
248 	apic_set_idlecpu,
249 	apic_unset_idlecpu,
250 
251 	apic_clkinit,
252 	apic_getclkirq,
253 	(void (*)(void))NULL,		/* psm_hrtimeinit */
254 	apic_gethrtime,
255 
256 	apic_get_next_processorid,
257 	apic_cpu_start,
258 	apic_post_cpu_start,
259 	apic_shutdown,
260 	apic_get_ipivect,
261 	apic_send_ipi,
262 
263 	(int (*)(dev_info_t *, int))NULL,	/* psm_translate_irq */
264 	(void (*)(int, char *))NULL,	/* psm_notify_error */
265 	(void (*)(int))NULL,		/* psm_notify_func */
266 	apic_timer_reprogram,
267 	apic_timer_enable,
268 	apic_timer_disable,
269 	apic_post_cyclic_setup,
270 	apic_preshutdown,
271 	apic_intr_ops,			/* Advanced DDI Interrupt framework */
272 	apic_state,			/* save, restore apic state for S3 */
273 	apic_cpu_ops,			/* CPU control interface. */
274 };
275 
276 
277 static struct	psm_info apic_psm_info = {
278 	PSM_INFO_VER01_7,			/* version */
279 	PSM_OWN_EXCLUSIVE,			/* ownership */
280 	(struct psm_ops *)&apic_ops,		/* operation */
281 	APIC_PCPLUSMP_NAME,			/* machine name */
282 	"pcplusmp v1.4 compatible",
283 };
284 
285 static void *apic_hdlp;
286 
287 #ifdef DEBUG
288 int	apic_debug = 0;
289 int	apic_restrict_vector = 0;
290 
291 int	apic_debug_msgbuf[APIC_DEBUG_MSGBUFSIZE];
292 int	apic_debug_msgbufindex = 0;
293 
294 #endif /* DEBUG */
295 
296 apic_cpus_info_t	*apic_cpus;
297 
298 cpuset_t	apic_cpumask;
299 uint_t	apic_picinit_called;
300 
301 /* Flag to indicate that we need to shut down all processors */
302 static uint_t	apic_shutdown_processors;
303 
304 uint_t apic_nsec_per_intr = 0;
305 
306 /*
307  * apic_let_idle_redistribute can have the following values:
308  * 0 - If clock decremented it from 1 to 0, clock has to call redistribute.
309  * apic_redistribute_lock prevents multiple idle cpus from redistributing
310  */
311 int	apic_num_idle_redistributions = 0;
312 static	int apic_let_idle_redistribute = 0;
313 static	uint_t apic_nticks = 0;
314 static	uint_t apic_skipped_redistribute = 0;
315 
316 /* to gather intr data and redistribute */
317 static void apic_redistribute_compute(void);
318 
319 static	uint_t last_count_read = 0;
320 static	lock_t	apic_gethrtime_lock;
321 volatile int	apic_hrtime_stamp = 0;
322 volatile hrtime_t apic_nsec_since_boot = 0;
323 static uint_t apic_hertz_count;
324 
325 uint64_t apic_ticks_per_SFnsecs;	/* # of ticks in SF nsecs */
326 
327 static hrtime_t apic_nsec_max;
328 
329 static	hrtime_t	apic_last_hrtime = 0;
330 int		apic_hrtime_error = 0;
331 int		apic_remote_hrterr = 0;
332 int		apic_num_nmis = 0;
333 int		apic_apic_error = 0;
334 int		apic_num_apic_errors = 0;
335 int		apic_num_cksum_errors = 0;
336 
337 int	apic_error = 0;
338 static	int	apic_cmos_ssb_set = 0;
339 
340 /* use to make sure only one cpu handles the nmi */
341 static	lock_t	apic_nmi_lock;
342 /* use to make sure only one cpu handles the error interrupt */
343 static	lock_t	apic_error_lock;
344 
345 static	struct {
346 	uchar_t	cntl;
347 	uchar_t	data;
348 } aspen_bmc[] = {
349 	{ CC_SMS_WR_START,	0x18 },		/* NetFn/LUN */
350 	{ CC_SMS_WR_NEXT,	0x24 },		/* Cmd SET_WATCHDOG_TIMER */
351 	{ CC_SMS_WR_NEXT,	0x84 },		/* DataByte 1: SMS/OS no log */
352 	{ CC_SMS_WR_NEXT,	0x2 },		/* DataByte 2: Power Down */
353 	{ CC_SMS_WR_NEXT,	0x0 },		/* DataByte 3: no pre-timeout */
354 	{ CC_SMS_WR_NEXT,	0x0 },		/* DataByte 4: timer expir. */
355 	{ CC_SMS_WR_NEXT,	0xa },		/* DataByte 5: init countdown */
356 	{ CC_SMS_WR_END,	0x0 },		/* DataByte 6: init countdown */
357 
358 	{ CC_SMS_WR_START,	0x18 },		/* NetFn/LUN */
359 	{ CC_SMS_WR_END,	0x22 }		/* Cmd RESET_WATCHDOG_TIMER */
360 };
361 
362 static	struct {
363 	int	port;
364 	uchar_t	data;
365 } sitka_bmc[] = {
366 	{ SMS_COMMAND_REGISTER,	SMS_WRITE_START },
367 	{ SMS_DATA_REGISTER,	0x18 },		/* NetFn/LUN */
368 	{ SMS_DATA_REGISTER,	0x24 },		/* Cmd SET_WATCHDOG_TIMER */
369 	{ SMS_DATA_REGISTER,	0x84 },		/* DataByte 1: SMS/OS no log */
370 	{ SMS_DATA_REGISTER,	0x2 },		/* DataByte 2: Power Down */
371 	{ SMS_DATA_REGISTER,	0x0 },		/* DataByte 3: no pre-timeout */
372 	{ SMS_DATA_REGISTER,	0x0 },		/* DataByte 4: timer expir. */
373 	{ SMS_DATA_REGISTER,	0xa },		/* DataByte 5: init countdown */
374 	{ SMS_COMMAND_REGISTER,	SMS_WRITE_END },
375 	{ SMS_DATA_REGISTER,	0x0 },		/* DataByte 6: init countdown */
376 
377 	{ SMS_COMMAND_REGISTER,	SMS_WRITE_START },
378 	{ SMS_DATA_REGISTER,	0x18 },		/* NetFn/LUN */
379 	{ SMS_COMMAND_REGISTER,	SMS_WRITE_END },
380 	{ SMS_DATA_REGISTER,	0x22 }		/* Cmd RESET_WATCHDOG_TIMER */
381 };
382 
383 /* Patchable global variables. */
384 int		apic_kmdb_on_nmi = 0;		/* 0 - no, 1 - yes enter kmdb */
385 uint32_t	apic_divide_reg_init = 0;	/* 0 - divide by 2 */
386 
387 /* default apic ops without interrupt remapping */
388 static apic_intrmap_ops_t apic_nointrmap_ops = {
389 	(int (*)(int))return_instr,
390 	(void (*)(int))return_instr,
391 	(void (*)(apic_irq_t *))return_instr,
392 	(void (*)(apic_irq_t *, void *))return_instr,
393 	(void (*)(apic_irq_t *))return_instr,
394 	apic_record_ioapic_rdt,
395 	apic_record_msi,
396 };
397 
398 apic_intrmap_ops_t *apic_vt_ops = &apic_nointrmap_ops;
399 
400 /*
401  *	This is the loadable module wrapper
402  */
403 
404 int
405 _init(void)
406 {
407 	if (apic_coarse_hrtime)
408 		apic_ops.psm_gethrtime = &apic_gettime;
409 	return (psm_mod_init(&apic_hdlp, &apic_psm_info));
410 }
411 
412 int
413 _fini(void)
414 {
415 	return (psm_mod_fini(&apic_hdlp, &apic_psm_info));
416 }
417 
418 int
419 _info(struct modinfo *modinfop)
420 {
421 	return (psm_mod_info(&apic_hdlp, &apic_psm_info, modinfop));
422 }
423 
424 
425 static int
426 apic_probe()
427 {
428 	return (apic_probe_common(apic_psm_info.p_mach_idstring));
429 }
430 
431 void
432 apic_init()
433 {
434 	int i;
435 	int	j = 1;
436 
437 	apic_ipltopri[0] = APIC_VECTOR_PER_IPL; /* leave 0 for idle */
438 	for (i = 0; i < (APIC_AVAIL_VECTOR / APIC_VECTOR_PER_IPL); i++) {
439 		if ((i < ((APIC_AVAIL_VECTOR / APIC_VECTOR_PER_IPL) - 1)) &&
440 		    (apic_vectortoipl[i + 1] == apic_vectortoipl[i]))
441 			/* get to highest vector at the same ipl */
442 			continue;
443 		for (; j <= apic_vectortoipl[i]; j++) {
444 			apic_ipltopri[j] = (i << APIC_IPL_SHIFT) +
445 			    APIC_BASE_VECT;
446 		}
447 	}
448 	for (; j < MAXIPL + 1; j++)
449 		/* fill up any empty ipltopri slots */
450 		apic_ipltopri[j] = (i << APIC_IPL_SHIFT) + APIC_BASE_VECT;
451 	apic_init_common();
452 #if defined(__amd64)
453 	/*
454 	 * Make cpu-specific interrupt info point to cr8pri vector
455 	 */
456 	for (i = 0; i <= MAXIPL; i++)
457 		apic_cr8pri[i] = apic_ipltopri[i] >> APIC_IPL_SHIFT;
458 	CPU->cpu_pri_data = apic_cr8pri;
459 #else
460 	if (cpuid_have_cr8access(CPU))
461 		apic_have_32bit_cr8 = 1;
462 #endif	/* __amd64 */
463 }
464 
465 /*
466  * handler for APIC Error interrupt. Just print a warning and continue
467  */
468 static int
469 apic_error_intr()
470 {
471 	uint_t	error0, error1, error;
472 	uint_t	i;
473 
474 	/*
475 	 * We need to write before read as per 7.4.17 of system prog manual.
476 	 * We do both and or the results to be safe
477 	 */
478 	error0 = apic_reg_ops->apic_read(APIC_ERROR_STATUS);
479 	apic_reg_ops->apic_write(APIC_ERROR_STATUS, 0);
480 	error1 = apic_reg_ops->apic_read(APIC_ERROR_STATUS);
481 	error = error0 | error1;
482 
483 	/*
484 	 * Clear the APIC error status (do this on all cpus that enter here)
485 	 * (two writes are required due to the semantics of accessing the
486 	 * error status register.)
487 	 */
488 	apic_reg_ops->apic_write(APIC_ERROR_STATUS, 0);
489 	apic_reg_ops->apic_write(APIC_ERROR_STATUS, 0);
490 
491 	/*
492 	 * Prevent more than 1 CPU from handling error interrupt causing
493 	 * double printing (interleave of characters from multiple
494 	 * CPU's when using prom_printf)
495 	 */
496 	if (lock_try(&apic_error_lock) == 0)
497 		return (error ? DDI_INTR_CLAIMED : DDI_INTR_UNCLAIMED);
498 	if (error) {
499 #if	DEBUG
500 		if (apic_debug)
501 			debug_enter("pcplusmp: APIC Error interrupt received");
502 #endif /* DEBUG */
503 		if (apic_panic_on_apic_error)
504 			cmn_err(CE_PANIC,
505 			    "APIC Error interrupt on CPU %d. Status = %x\n",
506 			    psm_get_cpu_id(), error);
507 		else {
508 			if ((error & ~APIC_CS_ERRORS) == 0) {
509 				/* cksum error only */
510 				apic_error |= APIC_ERR_APIC_ERROR;
511 				apic_apic_error |= error;
512 				apic_num_apic_errors++;
513 				apic_num_cksum_errors++;
514 			} else {
515 				/*
516 				 * prom_printf is the best shot we have of
517 				 * something which is problem free from
518 				 * high level/NMI type of interrupts
519 				 */
520 				prom_printf("APIC Error interrupt on CPU %d. "
521 				    "Status 0 = %x, Status 1 = %x\n",
522 				    psm_get_cpu_id(), error0, error1);
523 				apic_error |= APIC_ERR_APIC_ERROR;
524 				apic_apic_error |= error;
525 				apic_num_apic_errors++;
526 				for (i = 0; i < apic_error_display_delay; i++) {
527 					tenmicrosec();
528 				}
529 				/*
530 				 * provide more delay next time limited to
531 				 * roughly 1 clock tick time
532 				 */
533 				if (apic_error_display_delay < 500)
534 					apic_error_display_delay *= 2;
535 			}
536 		}
537 		lock_clear(&apic_error_lock);
538 		return (DDI_INTR_CLAIMED);
539 	} else {
540 		lock_clear(&apic_error_lock);
541 		return (DDI_INTR_UNCLAIMED);
542 	}
543 	/* NOTREACHED */
544 }
545 
546 /*
547  * Turn off the mask bit in the performance counter Local Vector Table entry.
548  */
549 static void
550 apic_cpcovf_mask_clear(void)
551 {
552 	apic_reg_ops->apic_write(APIC_PCINT_VECT,
553 	    (apic_reg_ops->apic_read(APIC_PCINT_VECT) & ~APIC_LVT_MASK));
554 }
555 
556 /*ARGSUSED*/
557 static int
558 apic_cmci_enable(xc_arg_t arg1, xc_arg_t arg2, xc_arg_t arg3)
559 {
560 	apic_reg_ops->apic_write(APIC_CMCI_VECT, apic_cmci_vect);
561 	return (0);
562 }
563 
564 /*ARGSUSED*/
565 static int
566 apic_cmci_disable(xc_arg_t arg1, xc_arg_t arg2, xc_arg_t arg3)
567 {
568 	apic_reg_ops->apic_write(APIC_CMCI_VECT, apic_cmci_vect | AV_MASK);
569 	return (0);
570 }
571 
572 /*ARGSUSED*/
573 static int
574 cmci_cpu_setup(cpu_setup_t what, int cpuid, void *arg)
575 {
576 	cpuset_t	cpu_set;
577 
578 	CPUSET_ONLY(cpu_set, cpuid);
579 
580 	switch (what) {
581 		case CPU_ON:
582 			xc_call(NULL, NULL, NULL, CPUSET2BV(cpu_set),
583 			    (xc_func_t)apic_cmci_enable);
584 			break;
585 
586 		case CPU_OFF:
587 			xc_call(NULL, NULL, NULL, CPUSET2BV(cpu_set),
588 			    (xc_func_t)apic_cmci_disable);
589 			break;
590 
591 		default:
592 			break;
593 	}
594 
595 	return (0);
596 }
597 
598 static void
599 apic_init_intr()
600 {
601 	processorid_t	cpun = psm_get_cpu_id();
602 	uint_t nlvt;
603 	uint32_t svr = AV_UNIT_ENABLE | APIC_SPUR_INTR;
604 
605 	apic_reg_ops->apic_write_task_reg(APIC_MASK_ALL);
606 
607 	if (apic_mode == LOCAL_APIC) {
608 		/*
609 		 * We are running APIC in MMIO mode.
610 		 */
611 		if (apic_flat_model) {
612 			apic_reg_ops->apic_write(APIC_FORMAT_REG,
613 			    APIC_FLAT_MODEL);
614 		} else {
615 			apic_reg_ops->apic_write(APIC_FORMAT_REG,
616 			    APIC_CLUSTER_MODEL);
617 		}
618 
619 		apic_reg_ops->apic_write(APIC_DEST_REG,
620 		    AV_HIGH_ORDER >> cpun);
621 	}
622 
623 	if (apic_directed_EOI_supported()) {
624 		/*
625 		 * Setting the 12th bit in the Spurious Interrupt Vector
626 		 * Register suppresses broadcast EOIs generated by the local
627 		 * APIC. The suppression of broadcast EOIs happens only when
628 		 * interrupts are level-triggered.
629 		 */
630 		svr |= APIC_SVR_SUPPRESS_BROADCAST_EOI;
631 	}
632 
633 	/* need to enable APIC before unmasking NMI */
634 	apic_reg_ops->apic_write(APIC_SPUR_INT_REG, svr);
635 
636 	/*
637 	 * Presence of an invalid vector with delivery mode AV_FIXED can
638 	 * cause an error interrupt, even if the entry is masked...so
639 	 * write a valid vector to LVT entries along with the mask bit
640 	 */
641 
642 	/* All APICs have timer and LINT0/1 */
643 	apic_reg_ops->apic_write(APIC_LOCAL_TIMER, AV_MASK|APIC_RESV_IRQ);
644 	apic_reg_ops->apic_write(APIC_INT_VECT0, AV_MASK|APIC_RESV_IRQ);
645 	apic_reg_ops->apic_write(APIC_INT_VECT1, AV_NMI);	/* enable NMI */
646 
647 	/*
648 	 * On integrated APICs, the number of LVT entries is
649 	 * 'Max LVT entry' + 1; on 82489DX's (non-integrated
650 	 * APICs), nlvt is "3" (LINT0, LINT1, and timer)
651 	 */
652 
653 	if (apic_cpus[cpun].aci_local_ver < APIC_INTEGRATED_VERS) {
654 		nlvt = 3;
655 	} else {
656 		nlvt = ((apic_reg_ops->apic_read(APIC_VERS_REG) >> 16) &
657 		    0xFF) + 1;
658 	}
659 
660 	if (nlvt >= 5) {
661 		/* Enable performance counter overflow interrupt */
662 
663 		if ((x86_feature & X86_MSR) != X86_MSR)
664 			apic_enable_cpcovf_intr = 0;
665 		if (apic_enable_cpcovf_intr) {
666 			if (apic_cpcovf_vect == 0) {
667 				int ipl = APIC_PCINT_IPL;
668 				int irq = apic_get_ipivect(ipl, -1);
669 
670 				ASSERT(irq != -1);
671 				apic_cpcovf_vect =
672 				    apic_irq_table[irq]->airq_vector;
673 				ASSERT(apic_cpcovf_vect);
674 				(void) add_avintr(NULL, ipl,
675 				    (avfunc)kcpc_hw_overflow_intr,
676 				    "apic pcint", irq, NULL, NULL, NULL, NULL);
677 				kcpc_hw_overflow_intr_installed = 1;
678 				kcpc_hw_enable_cpc_intr =
679 				    apic_cpcovf_mask_clear;
680 			}
681 			apic_reg_ops->apic_write(APIC_PCINT_VECT,
682 			    apic_cpcovf_vect);
683 		}
684 	}
685 
686 	if (nlvt >= 6) {
687 		/* Only mask TM intr if the BIOS apparently doesn't use it */
688 
689 		uint32_t lvtval;
690 
691 		lvtval = apic_reg_ops->apic_read(APIC_THERM_VECT);
692 		if (((lvtval & AV_MASK) == AV_MASK) ||
693 		    ((lvtval & AV_DELIV_MODE) != AV_SMI)) {
694 			apic_reg_ops->apic_write(APIC_THERM_VECT,
695 			    AV_MASK|APIC_RESV_IRQ);
696 		}
697 	}
698 
699 	/* Enable error interrupt */
700 
701 	if (nlvt >= 4 && apic_enable_error_intr) {
702 		if (apic_errvect == 0) {
703 			int ipl = 0xf;	/* get highest priority intr */
704 			int irq = apic_get_ipivect(ipl, -1);
705 
706 			ASSERT(irq != -1);
707 			apic_errvect = apic_irq_table[irq]->airq_vector;
708 			ASSERT(apic_errvect);
709 			/*
710 			 * Not PSMI compliant, but we are going to merge
711 			 * with ON anyway
712 			 */
713 			(void) add_avintr((void *)NULL, ipl,
714 			    (avfunc)apic_error_intr, "apic error intr",
715 			    irq, NULL, NULL, NULL, NULL);
716 		}
717 		apic_reg_ops->apic_write(APIC_ERR_VECT, apic_errvect);
718 		apic_reg_ops->apic_write(APIC_ERROR_STATUS, 0);
719 		apic_reg_ops->apic_write(APIC_ERROR_STATUS, 0);
720 	}
721 
722 	/* Enable CMCI interrupt */
723 	if (cmi_enable_cmci) {
724 
725 		mutex_enter(&cmci_cpu_setup_lock);
726 		if (cmci_cpu_setup_registered == 0) {
727 			mutex_enter(&cpu_lock);
728 			register_cpu_setup_func(cmci_cpu_setup, NULL);
729 			mutex_exit(&cpu_lock);
730 			cmci_cpu_setup_registered = 1;
731 		}
732 		mutex_exit(&cmci_cpu_setup_lock);
733 
734 		if (apic_cmci_vect == 0) {
735 			int ipl = 0x2;
736 			int irq = apic_get_ipivect(ipl, -1);
737 
738 			ASSERT(irq != -1);
739 			apic_cmci_vect = apic_irq_table[irq]->airq_vector;
740 			ASSERT(apic_cmci_vect);
741 
742 			(void) add_avintr(NULL, ipl,
743 			    (avfunc)cmi_cmci_trap,
744 			    "apic cmci intr", irq, NULL, NULL, NULL, NULL);
745 		}
746 		apic_reg_ops->apic_write(APIC_CMCI_VECT, apic_cmci_vect);
747 	}
748 }
749 
750 static void
751 apic_disable_local_apic()
752 {
753 	apic_reg_ops->apic_write_task_reg(APIC_MASK_ALL);
754 	apic_reg_ops->apic_write(APIC_LOCAL_TIMER, AV_MASK);
755 
756 	/* local intr reg 0 */
757 	apic_reg_ops->apic_write(APIC_INT_VECT0, AV_MASK);
758 
759 	/* disable NMI */
760 	apic_reg_ops->apic_write(APIC_INT_VECT1, AV_MASK);
761 
762 	/* and error interrupt */
763 	apic_reg_ops->apic_write(APIC_ERR_VECT, AV_MASK);
764 
765 	/* and perf counter intr */
766 	apic_reg_ops->apic_write(APIC_PCINT_VECT, AV_MASK);
767 
768 	apic_reg_ops->apic_write(APIC_SPUR_INT_REG, APIC_SPUR_INTR);
769 }
770 
771 static void
772 apic_picinit(void)
773 {
774 	int i, j;
775 	uint_t isr;
776 
777 	/*
778 	 * Initialize and enable interrupt remapping before apic
779 	 * hardware initialization
780 	 */
781 	apic_intrmap_init(apic_mode);
782 
783 	/*
784 	 * On UniSys Model 6520, the BIOS leaves vector 0x20 isr
785 	 * bit on without clearing it with EOI.  Since softint
786 	 * uses vector 0x20 to interrupt itself, so softint will
787 	 * not work on this machine.  In order to fix this problem
788 	 * a check is made to verify all the isr bits are clear.
789 	 * If not, EOIs are issued to clear the bits.
790 	 */
791 	for (i = 7; i >= 1; i--) {
792 		isr = apic_reg_ops->apic_read(APIC_ISR_REG + (i * 4));
793 		if (isr != 0)
794 			for (j = 0; ((j < 32) && (isr != 0)); j++)
795 				if (isr & (1 << j)) {
796 					apic_reg_ops->apic_write(
797 					    APIC_EOI_REG, 0);
798 					isr &= ~(1 << j);
799 					apic_error |= APIC_ERR_BOOT_EOI;
800 				}
801 	}
802 
803 	/* set a flag so we know we have run apic_picinit() */
804 	apic_picinit_called = 1;
805 	LOCK_INIT_CLEAR(&apic_gethrtime_lock);
806 	LOCK_INIT_CLEAR(&apic_ioapic_lock);
807 	LOCK_INIT_CLEAR(&apic_error_lock);
808 	LOCK_INIT_CLEAR(&apic_mode_switch_lock);
809 
810 	picsetup();	 /* initialise the 8259 */
811 
812 	/* add nmi handler - least priority nmi handler */
813 	LOCK_INIT_CLEAR(&apic_nmi_lock);
814 
815 	if (!psm_add_nmintr(0, (avfunc) apic_nmi_intr,
816 	    "pcplusmp NMI handler", (caddr_t)NULL))
817 		cmn_err(CE_WARN, "pcplusmp: Unable to add nmi handler");
818 
819 	/*
820 	 * Check for directed-EOI capability in the local APIC.
821 	 */
822 	if (apic_directed_EOI_supported() == 1) {
823 		apic_set_directed_EOI_handler();
824 	}
825 
826 	apic_init_intr();
827 
828 	/* enable apic mode if imcr present */
829 	if (apic_imcrp) {
830 		outb(APIC_IMCR_P1, (uchar_t)APIC_IMCR_SELECT);
831 		outb(APIC_IMCR_P2, (uchar_t)APIC_IMCR_APIC);
832 	}
833 
834 	ioapic_init_intr(IOAPIC_MASK);
835 }
836 
837 static void
838 apic_cpu_send_SIPI(processorid_t cpun, boolean_t start)
839 {
840 	int		loop_count;
841 	uint32_t	vector;
842 	uint_t		apicid;
843 	ulong_t		iflag;
844 
845 	apicid =  apic_cpus[cpun].aci_local_id;
846 
847 	/*
848 	 * Interrupts on current CPU will be disabled during the
849 	 * steps in order to avoid unwanted side effects from
850 	 * executing interrupt handlers on a problematic BIOS.
851 	 */
852 	iflag = intr_clear();
853 
854 	if (start) {
855 		outb(CMOS_ADDR, SSB);
856 		outb(CMOS_DATA, BIOS_SHUTDOWN);
857 	}
858 
859 	/*
860 	 * According to X2APIC specification in section '2.3.5.1' of
861 	 * Interrupt Command Register Semantics, the semantics of
862 	 * programming the Interrupt Command Register to dispatch an interrupt
863 	 * is simplified. A single MSR write to the 64-bit ICR is required
864 	 * for dispatching an interrupt. Specifically, with the 64-bit MSR
865 	 * interface to ICR, system software is not required to check the
866 	 * status of the delivery status bit prior to writing to the ICR
867 	 * to send an IPI. With the removal of the Delivery Status bit,
868 	 * system software no longer has a reason to read the ICR. It remains
869 	 * readable only to aid in debugging.
870 	 */
871 #ifdef	DEBUG
872 	APIC_AV_PENDING_SET();
873 #else
874 	if (apic_mode == LOCAL_APIC) {
875 		APIC_AV_PENDING_SET();
876 	}
877 #endif /* DEBUG */
878 
879 	/* for integrated - make sure there is one INIT IPI in buffer */
880 	/* for external - it will wake up the cpu */
881 	apic_reg_ops->apic_write_int_cmd(apicid, AV_ASSERT | AV_RESET);
882 
883 	/* If only 1 CPU is installed, PENDING bit will not go low */
884 	for (loop_count = apic_sipi_max_loop_count; loop_count; loop_count--) {
885 		if (apic_mode == LOCAL_APIC &&
886 		    apic_reg_ops->apic_read(APIC_INT_CMD1) & AV_PENDING)
887 			apic_ret();
888 		else
889 			break;
890 	}
891 
892 	apic_reg_ops->apic_write_int_cmd(apicid, AV_DEASSERT | AV_RESET);
893 	drv_usecwait(20000);		/* 20 milli sec */
894 
895 	if (apic_cpus[cpun].aci_local_ver >= APIC_INTEGRATED_VERS) {
896 		/* integrated apic */
897 
898 		vector = (rm_platter_pa >> MMU_PAGESHIFT) &
899 		    (APIC_VECTOR_MASK | APIC_IPL_MASK);
900 
901 		/* to offset the INIT IPI queue up in the buffer */
902 		apic_reg_ops->apic_write_int_cmd(apicid, vector | AV_STARTUP);
903 		drv_usecwait(200);		/* 20 micro sec */
904 
905 		/*
906 		 * send the second SIPI (Startup IPI) as recommended by Intel
907 		 * software development manual.
908 		 */
909 		apic_reg_ops->apic_write_int_cmd(apicid, vector | AV_STARTUP);
910 		drv_usecwait(200);	/* 20 micro sec */
911 	}
912 
913 	intr_restore(iflag);
914 }
915 
916 /*ARGSUSED1*/
917 static int
918 apic_cpu_start(processorid_t cpun, caddr_t arg)
919 {
920 	ASSERT(MUTEX_HELD(&cpu_lock));
921 
922 	if (!apic_cpu_in_range(cpun)) {
923 		return (EINVAL);
924 	}
925 
926 	/*
927 	 * Switch to apic_common_send_ipi for safety during starting other CPUs.
928 	 */
929 	if (apic_mode == LOCAL_X2APIC) {
930 		apic_switch_ipi_callback(B_TRUE);
931 	}
932 
933 	apic_cmos_ssb_set = 1;
934 	apic_cpu_send_SIPI(cpun, B_TRUE);
935 
936 	return (0);
937 }
938 
939 /*
940  * Put CPU into halted state with interrupts disabled.
941  */
942 /*ARGSUSED1*/
943 static int
944 apic_cpu_stop(processorid_t cpun, caddr_t arg)
945 {
946 	int		rc;
947 	cpu_t 		*cp;
948 	extern cpuset_t cpu_ready_set;
949 	extern void cpu_idle_intercept_cpu(cpu_t *cp);
950 
951 	ASSERT(MUTEX_HELD(&cpu_lock));
952 
953 	if (!apic_cpu_in_range(cpun)) {
954 		return (EINVAL);
955 	}
956 	if (apic_cpus[cpun].aci_local_ver < APIC_INTEGRATED_VERS) {
957 		return (ENOTSUP);
958 	}
959 
960 	cp = cpu_get(cpun);
961 	ASSERT(cp != NULL);
962 	ASSERT((cp->cpu_flags & CPU_OFFLINE) != 0);
963 	ASSERT((cp->cpu_flags & CPU_QUIESCED) != 0);
964 	ASSERT((cp->cpu_flags & CPU_ENABLE) == 0);
965 
966 	/* Clear CPU_READY flag to disable cross calls. */
967 	cp->cpu_flags &= ~CPU_READY;
968 	CPUSET_ATOMIC_DEL(cpu_ready_set, cpun);
969 	rc = xc_flush_cpu(cp);
970 	if (rc != 0) {
971 		CPUSET_ATOMIC_ADD(cpu_ready_set, cpun);
972 		cp->cpu_flags |= CPU_READY;
973 		return (rc);
974 	}
975 
976 	/* Intercept target CPU at a safe point before powering it off. */
977 	cpu_idle_intercept_cpu(cp);
978 
979 	apic_cpu_send_SIPI(cpun, B_FALSE);
980 	cp->cpu_flags &= ~CPU_RUNNING;
981 
982 	return (0);
983 }
984 
985 static int
986 apic_cpu_ops(psm_cpu_request_t *reqp)
987 {
988 	if (reqp == NULL) {
989 		return (EINVAL);
990 	}
991 
992 	switch (reqp->pcr_cmd) {
993 	case PSM_CPU_ADD:
994 		return (apic_cpu_add(reqp));
995 
996 	case PSM_CPU_REMOVE:
997 		return (apic_cpu_remove(reqp));
998 
999 	case PSM_CPU_STOP:
1000 		return (apic_cpu_stop(reqp->req.cpu_stop.cpuid,
1001 		    reqp->req.cpu_stop.ctx));
1002 
1003 	default:
1004 		return (ENOTSUP);
1005 	}
1006 }
1007 
1008 #ifdef	DEBUG
1009 int	apic_break_on_cpu = 9;
1010 int	apic_stretch_interrupts = 0;
1011 int	apic_stretch_ISR = 1 << 3;	/* IPL of 3 matches nothing now */
1012 
1013 void
1014 apic_break()
1015 {
1016 }
1017 #endif /* DEBUG */
1018 
1019 /*
1020  * platform_intr_enter
1021  *
1022  *	Called at the beginning of the interrupt service routine to
1023  *	mask all level equal to and below the interrupt priority
1024  *	of the interrupting vector.  An EOI should be given to
1025  *	the interrupt controller to enable other HW interrupts.
1026  *
1027  *	Return -1 for spurious interrupts
1028  *
1029  */
1030 /*ARGSUSED*/
1031 static int
1032 apic_intr_enter(int ipl, int *vectorp)
1033 {
1034 	uchar_t vector;
1035 	int nipl;
1036 	int irq;
1037 	ulong_t iflag;
1038 	apic_cpus_info_t *cpu_infop;
1039 
1040 	/*
1041 	 * The real vector delivered is (*vectorp + 0x20), but our caller
1042 	 * subtracts 0x20 from the vector before passing it to us.
1043 	 * (That's why APIC_BASE_VECT is 0x20.)
1044 	 */
1045 	vector = (uchar_t)*vectorp;
1046 
1047 	/* if interrupted by the clock, increment apic_nsec_since_boot */
1048 	if (vector == apic_clkvect) {
1049 		if (!apic_oneshot) {
1050 			/* NOTE: this is not MT aware */
1051 			apic_hrtime_stamp++;
1052 			apic_nsec_since_boot += apic_nsec_per_intr;
1053 			apic_hrtime_stamp++;
1054 			last_count_read = apic_hertz_count;
1055 			apic_redistribute_compute();
1056 		}
1057 
1058 		/* We will avoid all the book keeping overhead for clock */
1059 		nipl = apic_ipls[vector];
1060 
1061 		*vectorp = apic_vector_to_irq[vector + APIC_BASE_VECT];
1062 		if (apic_mode == LOCAL_APIC) {
1063 #if defined(__amd64)
1064 			setcr8((ulong_t)(apic_ipltopri[nipl] >>
1065 			    APIC_IPL_SHIFT));
1066 #else
1067 			if (apic_have_32bit_cr8)
1068 				setcr8((ulong_t)(apic_ipltopri[nipl] >>
1069 				    APIC_IPL_SHIFT));
1070 			else
1071 				LOCAL_APIC_WRITE_REG(APIC_TASK_REG,
1072 				    (uint32_t)apic_ipltopri[nipl]);
1073 #endif
1074 			LOCAL_APIC_WRITE_REG(APIC_EOI_REG, 0);
1075 		} else {
1076 			X2APIC_WRITE(APIC_TASK_REG, apic_ipltopri[nipl]);
1077 			X2APIC_WRITE(APIC_EOI_REG, 0);
1078 		}
1079 
1080 		return (nipl);
1081 	}
1082 
1083 	cpu_infop = &apic_cpus[psm_get_cpu_id()];
1084 
1085 	if (vector == (APIC_SPUR_INTR - APIC_BASE_VECT)) {
1086 		cpu_infop->aci_spur_cnt++;
1087 		return (APIC_INT_SPURIOUS);
1088 	}
1089 
1090 	/* Check if the vector we got is really what we need */
1091 	if (apic_revector_pending) {
1092 		/*
1093 		 * Disable interrupts for the duration of
1094 		 * the vector translation to prevent a self-race for
1095 		 * the apic_revector_lock.  This cannot be done
1096 		 * in apic_xlate_vector because it is recursive and
1097 		 * we want the vector translation to be atomic with
1098 		 * respect to other (higher-priority) interrupts.
1099 		 */
1100 		iflag = intr_clear();
1101 		vector = apic_xlate_vector(vector + APIC_BASE_VECT) -
1102 		    APIC_BASE_VECT;
1103 		intr_restore(iflag);
1104 	}
1105 
1106 	nipl = apic_ipls[vector];
1107 	*vectorp = irq = apic_vector_to_irq[vector + APIC_BASE_VECT];
1108 
1109 	if (apic_mode == LOCAL_APIC) {
1110 #if defined(__amd64)
1111 		setcr8((ulong_t)(apic_ipltopri[nipl] >> APIC_IPL_SHIFT));
1112 #else
1113 		if (apic_have_32bit_cr8)
1114 			setcr8((ulong_t)(apic_ipltopri[nipl] >>
1115 			    APIC_IPL_SHIFT));
1116 		else
1117 			LOCAL_APIC_WRITE_REG(APIC_TASK_REG,
1118 			    (uint32_t)apic_ipltopri[nipl]);
1119 #endif
1120 	} else {
1121 		X2APIC_WRITE(APIC_TASK_REG, apic_ipltopri[nipl]);
1122 	}
1123 
1124 	cpu_infop->aci_current[nipl] = (uchar_t)irq;
1125 	cpu_infop->aci_curipl = (uchar_t)nipl;
1126 	cpu_infop->aci_ISR_in_progress |= 1 << nipl;
1127 
1128 	/*
1129 	 * apic_level_intr could have been assimilated into the irq struct.
1130 	 * but, having it as a character array is more efficient in terms of
1131 	 * cache usage. So, we leave it as is.
1132 	 */
1133 	if (!apic_level_intr[irq]) {
1134 		if (apic_mode == LOCAL_APIC) {
1135 			LOCAL_APIC_WRITE_REG(APIC_EOI_REG, 0);
1136 		} else {
1137 			X2APIC_WRITE(APIC_EOI_REG, 0);
1138 		}
1139 	}
1140 
1141 #ifdef	DEBUG
1142 	APIC_DEBUG_BUF_PUT(vector);
1143 	APIC_DEBUG_BUF_PUT(irq);
1144 	APIC_DEBUG_BUF_PUT(nipl);
1145 	APIC_DEBUG_BUF_PUT(psm_get_cpu_id());
1146 	if ((apic_stretch_interrupts) && (apic_stretch_ISR & (1 << nipl)))
1147 		drv_usecwait(apic_stretch_interrupts);
1148 
1149 	if (apic_break_on_cpu == psm_get_cpu_id())
1150 		apic_break();
1151 #endif /* DEBUG */
1152 	return (nipl);
1153 }
1154 
1155 /*
1156  * This macro is a common code used by MMIO local apic and X2APIC
1157  * local apic.
1158  */
1159 #define	APIC_INTR_EXIT() \
1160 { \
1161 	cpu_infop = &apic_cpus[psm_get_cpu_id()]; \
1162 	if (apic_level_intr[irq]) \
1163 		apic_reg_ops->apic_send_eoi(irq); \
1164 	cpu_infop->aci_curipl = (uchar_t)prev_ipl; \
1165 	/* ISR above current pri could not be in progress */ \
1166 	cpu_infop->aci_ISR_in_progress &= (2 << prev_ipl) - 1; \
1167 }
1168 
1169 /*
1170  * Any changes made to this function must also change X2APIC
1171  * version of intr_exit.
1172  */
1173 void
1174 apic_intr_exit(int prev_ipl, int irq)
1175 {
1176 	apic_cpus_info_t *cpu_infop;
1177 
1178 #if defined(__amd64)
1179 	setcr8((ulong_t)apic_cr8pri[prev_ipl]);
1180 #else
1181 	if (apic_have_32bit_cr8)
1182 		setcr8((ulong_t)(apic_ipltopri[prev_ipl] >> APIC_IPL_SHIFT));
1183 	else
1184 		apicadr[APIC_TASK_REG] = apic_ipltopri[prev_ipl];
1185 #endif
1186 
1187 	APIC_INTR_EXIT();
1188 }
1189 
1190 /*
1191  * Same as apic_intr_exit() except it uses MSR rather than MMIO
1192  * to access local apic registers.
1193  */
1194 void
1195 x2apic_intr_exit(int prev_ipl, int irq)
1196 {
1197 	apic_cpus_info_t *cpu_infop;
1198 
1199 	X2APIC_WRITE(APIC_TASK_REG, apic_ipltopri[prev_ipl]);
1200 	APIC_INTR_EXIT();
1201 }
1202 
1203 intr_exit_fn_t
1204 psm_intr_exit_fn(void)
1205 {
1206 	if (apic_mode == LOCAL_X2APIC)
1207 		return (x2apic_intr_exit);
1208 
1209 	return (apic_intr_exit);
1210 }
1211 
1212 /*
1213  * Mask all interrupts below or equal to the given IPL.
1214  * Any changes made to this function must also change X2APIC
1215  * version of setspl.
1216  */
1217 static void
1218 apic_setspl(int ipl)
1219 {
1220 #if defined(__amd64)
1221 	setcr8((ulong_t)apic_cr8pri[ipl]);
1222 #else
1223 	if (apic_have_32bit_cr8)
1224 		setcr8((ulong_t)(apic_ipltopri[ipl] >> APIC_IPL_SHIFT));
1225 	else
1226 		apicadr[APIC_TASK_REG] = apic_ipltopri[ipl];
1227 #endif
1228 
1229 	/* interrupts at ipl above this cannot be in progress */
1230 	apic_cpus[psm_get_cpu_id()].aci_ISR_in_progress &= (2 << ipl) - 1;
1231 	/*
1232 	 * this is a patch fix for the ALR QSMP P5 machine, so that interrupts
1233 	 * have enough time to come in before the priority is raised again
1234 	 * during the idle() loop.
1235 	 */
1236 	if (apic_setspl_delay)
1237 		(void) apic_reg_ops->apic_get_pri();
1238 }
1239 
1240 /*
1241  * X2APIC version of setspl.
1242  * Mask all interrupts below or equal to the given IPL
1243  */
1244 static void
1245 x2apic_setspl(int ipl)
1246 {
1247 	X2APIC_WRITE(APIC_TASK_REG, apic_ipltopri[ipl]);
1248 
1249 	/* interrupts at ipl above this cannot be in progress */
1250 	apic_cpus[psm_get_cpu_id()].aci_ISR_in_progress &= (2 << ipl) - 1;
1251 }
1252 
1253 /*
1254  * generates an interprocessor interrupt to another CPU. Any changes made to
1255  * this routine must be accompanied by similar changes to
1256  * apic_common_send_ipi().
1257  */
1258 static void
1259 apic_send_ipi(int cpun, int ipl)
1260 {
1261 	int vector;
1262 	ulong_t flag;
1263 
1264 	vector = apic_resv_vector[ipl];
1265 
1266 	ASSERT((vector >= APIC_BASE_VECT) && (vector <= APIC_SPUR_INTR));
1267 
1268 	flag = intr_clear();
1269 
1270 	APIC_AV_PENDING_SET();
1271 
1272 	apic_reg_ops->apic_write_int_cmd(apic_cpus[cpun].aci_local_id,
1273 	    vector);
1274 
1275 	intr_restore(flag);
1276 }
1277 
1278 
1279 /*ARGSUSED*/
1280 static void
1281 apic_set_idlecpu(processorid_t cpun)
1282 {
1283 }
1284 
1285 /*ARGSUSED*/
1286 static void
1287 apic_unset_idlecpu(processorid_t cpun)
1288 {
1289 }
1290 
1291 
1292 void
1293 apic_ret()
1294 {
1295 }
1296 
1297 /*
1298  * If apic_coarse_time == 1, then apic_gettime() is used instead of
1299  * apic_gethrtime().  This is used for performance instead of accuracy.
1300  */
1301 
1302 static hrtime_t
1303 apic_gettime()
1304 {
1305 	int old_hrtime_stamp;
1306 	hrtime_t temp;
1307 
1308 	/*
1309 	 * In one-shot mode, we do not keep time, so if anyone
1310 	 * calls psm_gettime() directly, we vector over to
1311 	 * gethrtime().
1312 	 * one-shot mode MUST NOT be enabled if this psm is the source of
1313 	 * hrtime.
1314 	 */
1315 
1316 	if (apic_oneshot)
1317 		return (gethrtime());
1318 
1319 
1320 gettime_again:
1321 	while ((old_hrtime_stamp = apic_hrtime_stamp) & 1)
1322 		apic_ret();
1323 
1324 	temp = apic_nsec_since_boot;
1325 
1326 	if (apic_hrtime_stamp != old_hrtime_stamp) {	/* got an interrupt */
1327 		goto gettime_again;
1328 	}
1329 	return (temp);
1330 }
1331 
1332 /*
1333  * Here we return the number of nanoseconds since booting.  Note every
1334  * clock interrupt increments apic_nsec_since_boot by the appropriate
1335  * amount.
1336  */
1337 static hrtime_t
1338 apic_gethrtime()
1339 {
1340 	int curr_timeval, countval, elapsed_ticks;
1341 	int old_hrtime_stamp, status;
1342 	hrtime_t temp;
1343 	uint32_t cpun;
1344 	ulong_t oflags;
1345 
1346 	/*
1347 	 * In one-shot mode, we do not keep time, so if anyone
1348 	 * calls psm_gethrtime() directly, we vector over to
1349 	 * gethrtime().
1350 	 * one-shot mode MUST NOT be enabled if this psm is the source of
1351 	 * hrtime.
1352 	 */
1353 
1354 	if (apic_oneshot)
1355 		return (gethrtime());
1356 
1357 	oflags = intr_clear();	/* prevent migration */
1358 
1359 	cpun = apic_reg_ops->apic_read(APIC_LID_REG);
1360 	if (apic_mode == LOCAL_APIC)
1361 		cpun >>= APIC_ID_BIT_OFFSET;
1362 
1363 	lock_set(&apic_gethrtime_lock);
1364 
1365 gethrtime_again:
1366 	while ((old_hrtime_stamp = apic_hrtime_stamp) & 1)
1367 		apic_ret();
1368 
1369 	/*
1370 	 * Check to see which CPU we are on.  Note the time is kept on
1371 	 * the local APIC of CPU 0.  If on CPU 0, simply read the current
1372 	 * counter.  If on another CPU, issue a remote read command to CPU 0.
1373 	 */
1374 	if (cpun == apic_cpus[0].aci_local_id) {
1375 		countval = apic_reg_ops->apic_read(APIC_CURR_COUNT);
1376 	} else {
1377 #ifdef	DEBUG
1378 		APIC_AV_PENDING_SET();
1379 #else
1380 		if (apic_mode == LOCAL_APIC)
1381 			APIC_AV_PENDING_SET();
1382 #endif /* DEBUG */
1383 
1384 		apic_reg_ops->apic_write_int_cmd(
1385 		    apic_cpus[0].aci_local_id, APIC_CURR_ADD | AV_REMOTE);
1386 
1387 		while ((status = apic_reg_ops->apic_read(APIC_INT_CMD1))
1388 		    & AV_READ_PENDING) {
1389 			apic_ret();
1390 		}
1391 
1392 		if (status & AV_REMOTE_STATUS)	/* 1 = valid */
1393 			countval = apic_reg_ops->apic_read(APIC_REMOTE_READ);
1394 		else {	/* 0 = invalid */
1395 			apic_remote_hrterr++;
1396 			/*
1397 			 * return last hrtime right now, will need more
1398 			 * testing if change to retry
1399 			 */
1400 			temp = apic_last_hrtime;
1401 
1402 			lock_clear(&apic_gethrtime_lock);
1403 
1404 			intr_restore(oflags);
1405 
1406 			return (temp);
1407 		}
1408 	}
1409 	if (countval > last_count_read)
1410 		countval = 0;
1411 	else
1412 		last_count_read = countval;
1413 
1414 	elapsed_ticks = apic_hertz_count - countval;
1415 
1416 	curr_timeval = APIC_TICKS_TO_NSECS(elapsed_ticks);
1417 	temp = apic_nsec_since_boot + curr_timeval;
1418 
1419 	if (apic_hrtime_stamp != old_hrtime_stamp) {	/* got an interrupt */
1420 		/* we might have clobbered last_count_read. Restore it */
1421 		last_count_read = apic_hertz_count;
1422 		goto gethrtime_again;
1423 	}
1424 
1425 	if (temp < apic_last_hrtime) {
1426 		/* return last hrtime if error occurs */
1427 		apic_hrtime_error++;
1428 		temp = apic_last_hrtime;
1429 	}
1430 	else
1431 		apic_last_hrtime = temp;
1432 
1433 	lock_clear(&apic_gethrtime_lock);
1434 	intr_restore(oflags);
1435 
1436 	return (temp);
1437 }
1438 
1439 /* apic NMI handler */
1440 /*ARGSUSED*/
1441 static void
1442 apic_nmi_intr(caddr_t arg, struct regs *rp)
1443 {
1444 	if (apic_shutdown_processors) {
1445 		apic_disable_local_apic();
1446 		return;
1447 	}
1448 
1449 	apic_error |= APIC_ERR_NMI;
1450 
1451 	if (!lock_try(&apic_nmi_lock))
1452 		return;
1453 	apic_num_nmis++;
1454 
1455 	if (apic_kmdb_on_nmi && psm_debugger()) {
1456 		debug_enter("NMI received: entering kmdb\n");
1457 	} else if (apic_panic_on_nmi) {
1458 		/* Keep panic from entering kmdb. */
1459 		nopanicdebug = 1;
1460 		panic("NMI received\n");
1461 	} else {
1462 		/*
1463 		 * prom_printf is the best shot we have of something which is
1464 		 * problem free from high level/NMI type of interrupts
1465 		 */
1466 		prom_printf("NMI received\n");
1467 	}
1468 
1469 	lock_clear(&apic_nmi_lock);
1470 }
1471 
1472 /*ARGSUSED*/
1473 static int
1474 apic_addspl(int irqno, int ipl, int min_ipl, int max_ipl)
1475 {
1476 	return (apic_addspl_common(irqno, ipl, min_ipl, max_ipl));
1477 }
1478 
1479 static int
1480 apic_delspl(int irqno, int ipl, int min_ipl, int max_ipl)
1481 {
1482 	return (apic_delspl_common(irqno, ipl, min_ipl,  max_ipl));
1483 }
1484 
1485 static int
1486 apic_post_cpu_start()
1487 {
1488 	int cpun;
1489 	static int cpus_started = 1;
1490 
1491 	/* We know this CPU + BSP  started successfully. */
1492 	cpus_started++;
1493 
1494 	/*
1495 	 * On BSP we would have enabled X2APIC, if supported by processor,
1496 	 * in acpi_probe(), but on AP we do it here.
1497 	 *
1498 	 * We enable X2APIC mode only if BSP is running in X2APIC & the
1499 	 * local APIC mode of the current CPU is MMIO (xAPIC).
1500 	 */
1501 	if (apic_mode == LOCAL_X2APIC && apic_detect_x2apic() &&
1502 	    apic_local_mode() == LOCAL_APIC) {
1503 		apic_enable_x2apic();
1504 	}
1505 
1506 	/*
1507 	 * Switch back to x2apic IPI sending method for performance when target
1508 	 * CPU has entered x2apic mode.
1509 	 */
1510 	if (apic_mode == LOCAL_X2APIC) {
1511 		apic_switch_ipi_callback(B_FALSE);
1512 	}
1513 
1514 	splx(ipltospl(LOCK_LEVEL));
1515 	apic_init_intr();
1516 
1517 	/*
1518 	 * since some systems don't enable the internal cache on the non-boot
1519 	 * cpus, so we have to enable them here
1520 	 */
1521 	setcr0(getcr0() & ~(CR0_CD | CR0_NW));
1522 
1523 #ifdef	DEBUG
1524 	APIC_AV_PENDING_SET();
1525 #else
1526 	if (apic_mode == LOCAL_APIC)
1527 		APIC_AV_PENDING_SET();
1528 #endif	/* DEBUG */
1529 
1530 	/*
1531 	 * We may be booting, or resuming from suspend; aci_status will
1532 	 * be APIC_CPU_INTR_ENABLE if coming from suspend, so we add the
1533 	 * APIC_CPU_ONLINE flag here rather than setting aci_status completely.
1534 	 */
1535 	cpun = psm_get_cpu_id();
1536 	apic_cpus[cpun].aci_status |= APIC_CPU_ONLINE;
1537 
1538 	apic_reg_ops->apic_write(APIC_DIVIDE_REG, apic_divide_reg_init);
1539 	return (PSM_SUCCESS);
1540 }
1541 
1542 processorid_t
1543 apic_get_next_processorid(processorid_t cpu_id)
1544 {
1545 
1546 	int i;
1547 
1548 	if (cpu_id == -1)
1549 		return ((processorid_t)0);
1550 
1551 	for (i = cpu_id + 1; i < NCPU; i++) {
1552 		if (apic_cpu_in_range(i))
1553 			return (i);
1554 	}
1555 
1556 	return ((processorid_t)-1);
1557 }
1558 
1559 static int
1560 apic_cpu_add(psm_cpu_request_t *reqp)
1561 {
1562 	int i, rv = 0;
1563 	ulong_t iflag;
1564 	boolean_t first = B_TRUE;
1565 	uchar_t localver;
1566 	uint32_t localid, procid;
1567 	processorid_t cpuid = (processorid_t)-1;
1568 	mach_cpu_add_arg_t *ap;
1569 
1570 	ASSERT(reqp != NULL);
1571 	reqp->req.cpu_add.cpuid = (processorid_t)-1;
1572 
1573 	/* Check whether CPU hotplug is supported. */
1574 	if (!plat_dr_support_cpu() || apic_max_nproc == -1) {
1575 		return (ENOTSUP);
1576 	}
1577 
1578 	ap = (mach_cpu_add_arg_t *)reqp->req.cpu_add.argp;
1579 	switch (ap->type) {
1580 	case MACH_CPU_ARG_LOCAL_APIC:
1581 		localid = ap->arg.apic.apic_id;
1582 		procid = ap->arg.apic.proc_id;
1583 		if (localid >= 255 || procid > 255) {
1584 			cmn_err(CE_WARN,
1585 			    "!apic: apicid(%u) or procid(%u) is invalid.",
1586 			    localid, procid);
1587 			return (EINVAL);
1588 		}
1589 		break;
1590 
1591 	case MACH_CPU_ARG_LOCAL_X2APIC:
1592 		localid = ap->arg.apic.apic_id;
1593 		procid = ap->arg.apic.proc_id;
1594 		if (localid >= UINT32_MAX) {
1595 			cmn_err(CE_WARN,
1596 			    "!apic: x2apicid(%u) is invalid.", localid);
1597 			return (EINVAL);
1598 		} else if (localid >= 255 && apic_mode == LOCAL_APIC) {
1599 			cmn_err(CE_WARN, "!apic: system is in APIC mode, "
1600 			    "can't support x2APIC processor.");
1601 			return (ENOTSUP);
1602 		}
1603 		break;
1604 
1605 	default:
1606 		cmn_err(CE_WARN,
1607 		    "!apic: unknown argument type %d to apic_cpu_add().",
1608 		    ap->type);
1609 		return (EINVAL);
1610 	}
1611 
1612 	/* Use apic_ioapic_lock to sync with apic_find_next_cpu_intr. */
1613 	iflag = intr_clear();
1614 	lock_set(&apic_ioapic_lock);
1615 
1616 	/* Check whether local APIC id already exists. */
1617 	for (i = 0; i < apic_nproc; i++) {
1618 		if (!CPU_IN_SET(apic_cpumask, i))
1619 			continue;
1620 		if (apic_cpus[i].aci_local_id == localid) {
1621 			lock_clear(&apic_ioapic_lock);
1622 			intr_restore(iflag);
1623 			cmn_err(CE_WARN,
1624 			    "!apic: local apic id %u already exists.",
1625 			    localid);
1626 			return (EEXIST);
1627 		} else if (apic_cpus[i].aci_processor_id == procid) {
1628 			lock_clear(&apic_ioapic_lock);
1629 			intr_restore(iflag);
1630 			cmn_err(CE_WARN,
1631 			    "!apic: processor id %u already exists.",
1632 			    (int)procid);
1633 			return (EEXIST);
1634 		}
1635 
1636 		/*
1637 		 * There's no local APIC version number available in MADT table,
1638 		 * so assume that all CPUs are homogeneous and use local APIC
1639 		 * version number of the first existing CPU.
1640 		 */
1641 		if (first) {
1642 			first = B_FALSE;
1643 			localver = apic_cpus[i].aci_local_ver;
1644 		}
1645 	}
1646 	ASSERT(first == B_FALSE);
1647 
1648 	/*
1649 	 * Try to assign the same cpuid if APIC id exists in the dirty cache.
1650 	 */
1651 	for (i = 0; i < apic_max_nproc; i++) {
1652 		if (CPU_IN_SET(apic_cpumask, i)) {
1653 			ASSERT((apic_cpus[i].aci_status & APIC_CPU_FREE) == 0);
1654 			continue;
1655 		}
1656 		ASSERT(apic_cpus[i].aci_status & APIC_CPU_FREE);
1657 		if ((apic_cpus[i].aci_status & APIC_CPU_DIRTY) &&
1658 		    apic_cpus[i].aci_local_id == localid &&
1659 		    apic_cpus[i].aci_processor_id == procid) {
1660 			cpuid = i;
1661 			break;
1662 		}
1663 	}
1664 
1665 	/* Avoid the dirty cache and allocate fresh slot if possible. */
1666 	if (cpuid == (processorid_t)-1) {
1667 		for (i = 0; i < apic_max_nproc; i++) {
1668 			if ((apic_cpus[i].aci_status & APIC_CPU_FREE) &&
1669 			    (apic_cpus[i].aci_status & APIC_CPU_DIRTY) == 0) {
1670 				cpuid = i;
1671 				break;
1672 			}
1673 		}
1674 	}
1675 
1676 	/* Try to find any free slot as last resort. */
1677 	if (cpuid == (processorid_t)-1) {
1678 		for (i = 0; i < apic_max_nproc; i++) {
1679 			if (apic_cpus[i].aci_status & APIC_CPU_FREE) {
1680 				cpuid = i;
1681 				break;
1682 			}
1683 		}
1684 	}
1685 
1686 	if (cpuid == (processorid_t)-1) {
1687 		lock_clear(&apic_ioapic_lock);
1688 		intr_restore(iflag);
1689 		cmn_err(CE_NOTE,
1690 		    "!apic: failed to allocate cpu id for processor %u.",
1691 		    procid);
1692 		rv = EAGAIN;
1693 	} else if (ACPI_FAILURE(acpica_map_cpu(cpuid, procid))) {
1694 		lock_clear(&apic_ioapic_lock);
1695 		intr_restore(iflag);
1696 		cmn_err(CE_NOTE,
1697 		    "!apic: failed to build mapping for processor %u.",
1698 		    procid);
1699 		rv = EBUSY;
1700 	} else {
1701 		ASSERT(cpuid >= 0 && cpuid < NCPU);
1702 		ASSERT(cpuid < apic_max_nproc && cpuid < max_ncpus);
1703 		bzero(&apic_cpus[cpuid], sizeof (apic_cpus[0]));
1704 		apic_cpus[cpuid].aci_processor_id = procid;
1705 		apic_cpus[cpuid].aci_local_id = localid;
1706 		apic_cpus[cpuid].aci_local_ver = localver;
1707 		CPUSET_ATOMIC_ADD(apic_cpumask, cpuid);
1708 		if (cpuid >= apic_nproc) {
1709 			apic_nproc = cpuid + 1;
1710 		}
1711 		lock_clear(&apic_ioapic_lock);
1712 		intr_restore(iflag);
1713 		reqp->req.cpu_add.cpuid = cpuid;
1714 	}
1715 
1716 	return (rv);
1717 }
1718 
1719 static int
1720 apic_cpu_remove(psm_cpu_request_t *reqp)
1721 {
1722 	int i;
1723 	ulong_t iflag;
1724 	processorid_t cpuid;
1725 
1726 	/* Check whether CPU hotplug is supported. */
1727 	if (!plat_dr_support_cpu() || apic_max_nproc == -1) {
1728 		return (ENOTSUP);
1729 	}
1730 
1731 	cpuid = reqp->req.cpu_remove.cpuid;
1732 
1733 	/* Use apic_ioapic_lock to sync with apic_find_next_cpu_intr. */
1734 	iflag = intr_clear();
1735 	lock_set(&apic_ioapic_lock);
1736 
1737 	if (!apic_cpu_in_range(cpuid)) {
1738 		lock_clear(&apic_ioapic_lock);
1739 		intr_restore(iflag);
1740 		cmn_err(CE_WARN,
1741 		    "!apic: cpuid %d doesn't exist in apic_cpus array.",
1742 		    cpuid);
1743 		return (ENODEV);
1744 	}
1745 	ASSERT((apic_cpus[cpuid].aci_status & APIC_CPU_FREE) == 0);
1746 
1747 	if (ACPI_FAILURE(acpica_unmap_cpu(cpuid))) {
1748 		lock_clear(&apic_ioapic_lock);
1749 		intr_restore(iflag);
1750 		return (ENOENT);
1751 	}
1752 
1753 	if (cpuid == apic_nproc - 1) {
1754 		/*
1755 		 * We are removing the highest numbered cpuid so we need to
1756 		 * find the next highest cpuid as the new value for apic_nproc.
1757 		 */
1758 		for (i = apic_nproc; i > 0; i--) {
1759 			if (CPU_IN_SET(apic_cpumask, i - 1)) {
1760 				apic_nproc = i;
1761 				break;
1762 			}
1763 		}
1764 		/* at least one CPU left */
1765 		ASSERT(i > 0);
1766 	}
1767 	CPUSET_ATOMIC_DEL(apic_cpumask, cpuid);
1768 	/* mark slot as free and keep it in the dirty cache */
1769 	apic_cpus[cpuid].aci_status = APIC_CPU_FREE | APIC_CPU_DIRTY;
1770 
1771 	lock_clear(&apic_ioapic_lock);
1772 	intr_restore(iflag);
1773 
1774 	return (0);
1775 }
1776 
1777 /*
1778  * type == -1 indicates it is an internal request. Do not change
1779  * resv_vector for these requests
1780  */
1781 static int
1782 apic_get_ipivect(int ipl, int type)
1783 {
1784 	uchar_t vector;
1785 	int irq;
1786 
1787 	if ((irq = apic_allocate_irq(APIC_VECTOR(ipl))) != -1) {
1788 		if (vector = apic_allocate_vector(ipl, irq, 1)) {
1789 			apic_irq_table[irq]->airq_mps_intr_index =
1790 			    RESERVE_INDEX;
1791 			apic_irq_table[irq]->airq_vector = vector;
1792 			if (type != -1) {
1793 				apic_resv_vector[ipl] = vector;
1794 			}
1795 			return (irq);
1796 		}
1797 	}
1798 	apic_error |= APIC_ERR_GET_IPIVECT_FAIL;
1799 	return (-1);	/* shouldn't happen */
1800 }
1801 
1802 static int
1803 apic_getclkirq(int ipl)
1804 {
1805 	int	irq;
1806 
1807 	if ((irq = apic_get_ipivect(ipl, -1)) == -1)
1808 		return (-1);
1809 	/*
1810 	 * Note the vector in apic_clkvect for per clock handling.
1811 	 */
1812 	apic_clkvect = apic_irq_table[irq]->airq_vector - APIC_BASE_VECT;
1813 	APIC_VERBOSE_IOAPIC((CE_NOTE, "get_clkirq: vector = %x\n",
1814 	    apic_clkvect));
1815 	return (irq);
1816 }
1817 
1818 
1819 /*
1820  * Return the number of APIC clock ticks elapsed for 8245 to decrement
1821  * (APIC_TIME_COUNT + pit_ticks_adj) ticks.
1822  */
1823 static uint_t
1824 apic_calibrate(volatile uint32_t *addr, uint16_t *pit_ticks_adj)
1825 {
1826 	uint8_t		pit_tick_lo;
1827 	uint16_t	pit_tick, target_pit_tick;
1828 	uint32_t	start_apic_tick, end_apic_tick;
1829 	ulong_t		iflag;
1830 	uint32_t	reg;
1831 
1832 	reg = addr + APIC_CURR_COUNT - apicadr;
1833 
1834 	iflag = intr_clear();
1835 
1836 	do {
1837 		pit_tick_lo = inb(PITCTR0_PORT);
1838 		pit_tick = (inb(PITCTR0_PORT) << 8) | pit_tick_lo;
1839 	} while (pit_tick < APIC_TIME_MIN ||
1840 	    pit_tick_lo <= APIC_LB_MIN || pit_tick_lo >= APIC_LB_MAX);
1841 
1842 	/*
1843 	 * Wait for the 8254 to decrement by 5 ticks to ensure
1844 	 * we didn't start in the middle of a tick.
1845 	 * Compare with 0x10 for the wrap around case.
1846 	 */
1847 	target_pit_tick = pit_tick - 5;
1848 	do {
1849 		pit_tick_lo = inb(PITCTR0_PORT);
1850 		pit_tick = (inb(PITCTR0_PORT) << 8) | pit_tick_lo;
1851 	} while (pit_tick > target_pit_tick || pit_tick_lo < 0x10);
1852 
1853 	start_apic_tick = apic_reg_ops->apic_read(reg);
1854 
1855 	/*
1856 	 * Wait for the 8254 to decrement by
1857 	 * (APIC_TIME_COUNT + pit_ticks_adj) ticks
1858 	 */
1859 	target_pit_tick = pit_tick - APIC_TIME_COUNT;
1860 	do {
1861 		pit_tick_lo = inb(PITCTR0_PORT);
1862 		pit_tick = (inb(PITCTR0_PORT) << 8) | pit_tick_lo;
1863 	} while (pit_tick > target_pit_tick || pit_tick_lo < 0x10);
1864 
1865 	end_apic_tick = apic_reg_ops->apic_read(reg);
1866 
1867 	*pit_ticks_adj = target_pit_tick - pit_tick;
1868 
1869 	intr_restore(iflag);
1870 
1871 	return (start_apic_tick - end_apic_tick);
1872 }
1873 
1874 /*
1875  * Initialise the APIC timer on the local APIC of CPU 0 to the desired
1876  * frequency.  Note at this stage in the boot sequence, the boot processor
1877  * is the only active processor.
1878  * hertz value of 0 indicates a one-shot mode request.  In this case
1879  * the function returns the resolution (in nanoseconds) for the hardware
1880  * timer interrupt.  If one-shot mode capability is not available,
1881  * the return value will be 0. apic_enable_oneshot is a global switch
1882  * for disabling the functionality.
1883  * A non-zero positive value for hertz indicates a periodic mode request.
1884  * In this case the hardware will be programmed to generate clock interrupts
1885  * at hertz frequency and returns the resolution of interrupts in
1886  * nanosecond.
1887  */
1888 
1889 static int
1890 apic_clkinit(int hertz)
1891 {
1892 	uint_t		apic_ticks = 0;
1893 	uint_t		pit_ticks;
1894 	int		ret;
1895 	uint16_t	pit_ticks_adj;
1896 	static int	firsttime = 1;
1897 
1898 	if (firsttime) {
1899 		/* first time calibrate on CPU0 only */
1900 
1901 		apic_reg_ops->apic_write(APIC_DIVIDE_REG, apic_divide_reg_init);
1902 		apic_reg_ops->apic_write(APIC_INIT_COUNT, APIC_MAXVAL);
1903 		apic_ticks = apic_calibrate(apicadr, &pit_ticks_adj);
1904 
1905 		/* total number of PIT ticks corresponding to apic_ticks */
1906 		pit_ticks = APIC_TIME_COUNT + pit_ticks_adj;
1907 
1908 		/*
1909 		 * Determine the number of nanoseconds per APIC clock tick
1910 		 * and then determine how many APIC ticks to interrupt at the
1911 		 * desired frequency
1912 		 * apic_ticks / (pitticks / PIT_HZ) = apic_ticks_per_s
1913 		 * (apic_ticks * PIT_HZ) / pitticks = apic_ticks_per_s
1914 		 * apic_ticks_per_ns = (apic_ticks * PIT_HZ) / (pitticks * 10^9)
1915 		 * pic_ticks_per_SFns =
1916 		 *   (SF * apic_ticks * PIT_HZ) / (pitticks * 10^9)
1917 		 */
1918 		apic_ticks_per_SFnsecs =
1919 		    ((SF * apic_ticks * PIT_HZ) /
1920 		    ((uint64_t)pit_ticks * NANOSEC));
1921 
1922 		/* the interval timer initial count is 32 bit max */
1923 		apic_nsec_max = APIC_TICKS_TO_NSECS(APIC_MAXVAL);
1924 		firsttime = 0;
1925 	}
1926 
1927 	if (hertz != 0) {
1928 		/* periodic */
1929 		apic_nsec_per_intr = NANOSEC / hertz;
1930 		apic_hertz_count = APIC_NSECS_TO_TICKS(apic_nsec_per_intr);
1931 	}
1932 
1933 	apic_int_busy_mark = (apic_int_busy_mark *
1934 	    apic_sample_factor_redistribution) / 100;
1935 	apic_int_free_mark = (apic_int_free_mark *
1936 	    apic_sample_factor_redistribution) / 100;
1937 	apic_diff_for_redistribution = (apic_diff_for_redistribution *
1938 	    apic_sample_factor_redistribution) / 100;
1939 
1940 	if (hertz == 0) {
1941 		/* requested one_shot */
1942 		if (!tsc_gethrtime_enable || !apic_oneshot_enable)
1943 			return (0);
1944 		apic_oneshot = 1;
1945 		ret = (int)APIC_TICKS_TO_NSECS(1);
1946 	} else {
1947 		/* program the local APIC to interrupt at the given frequency */
1948 		apic_reg_ops->apic_write(APIC_INIT_COUNT, apic_hertz_count);
1949 		apic_reg_ops->apic_write(APIC_LOCAL_TIMER,
1950 		    (apic_clkvect + APIC_BASE_VECT) | AV_TIME);
1951 		apic_oneshot = 0;
1952 		ret = NANOSEC / hertz;
1953 	}
1954 
1955 	return (ret);
1956 
1957 }
1958 
1959 /*
1960  * apic_preshutdown:
1961  * Called early in shutdown whilst we can still access filesystems to do
1962  * things like loading modules which will be required to complete shutdown
1963  * after filesystems are all unmounted.
1964  */
1965 static void
1966 apic_preshutdown(int cmd, int fcn)
1967 {
1968 	APIC_VERBOSE_POWEROFF(("apic_preshutdown(%d,%d); m=%d a=%d\n",
1969 	    cmd, fcn, apic_poweroff_method, apic_enable_acpi));
1970 
1971 	if ((cmd != A_SHUTDOWN) || (fcn != AD_POWEROFF)) {
1972 		return;
1973 	}
1974 }
1975 
1976 static void
1977 apic_shutdown(int cmd, int fcn)
1978 {
1979 	int restarts, attempts;
1980 	int i;
1981 	uchar_t	byte;
1982 	ulong_t iflag;
1983 
1984 	hpet_acpi_fini();
1985 
1986 	/* Send NMI to all CPUs except self to do per processor shutdown */
1987 	iflag = intr_clear();
1988 #ifdef	DEBUG
1989 	APIC_AV_PENDING_SET();
1990 #else
1991 	if (apic_mode == LOCAL_APIC)
1992 		APIC_AV_PENDING_SET();
1993 #endif /* DEBUG */
1994 	apic_shutdown_processors = 1;
1995 	apic_reg_ops->apic_write(APIC_INT_CMD1,
1996 	    AV_NMI | AV_LEVEL | AV_SH_ALL_EXCSELF);
1997 
1998 	/* restore cmos shutdown byte before reboot */
1999 	if (apic_cmos_ssb_set) {
2000 		outb(CMOS_ADDR, SSB);
2001 		outb(CMOS_DATA, 0);
2002 	}
2003 
2004 	ioapic_disable_redirection();
2005 
2006 	/*	disable apic mode if imcr present	*/
2007 	if (apic_imcrp) {
2008 		outb(APIC_IMCR_P1, (uchar_t)APIC_IMCR_SELECT);
2009 		outb(APIC_IMCR_P2, (uchar_t)APIC_IMCR_PIC);
2010 	}
2011 
2012 	apic_disable_local_apic();
2013 
2014 	intr_restore(iflag);
2015 
2016 	/* remainder of function is for shutdown cases only */
2017 	if (cmd != A_SHUTDOWN)
2018 		return;
2019 
2020 	/*
2021 	 * Switch system back into Legacy-Mode if using ACPI and
2022 	 * not powering-off.  Some BIOSes need to remain in ACPI-mode
2023 	 * for power-off to succeed (Dell Dimension 4600)
2024 	 * Do not disable ACPI while doing fastreboot
2025 	 */
2026 	if (apic_enable_acpi && fcn != AD_POWEROFF && fcn != AD_FASTREBOOT)
2027 		(void) AcpiDisable();
2028 
2029 	if (fcn == AD_FASTREBOOT) {
2030 		apic_reg_ops->apic_write(APIC_INT_CMD1,
2031 		    AV_ASSERT | AV_RESET | AV_SH_ALL_EXCSELF);
2032 	}
2033 
2034 	/* remainder of function is for shutdown+poweroff case only */
2035 	if (fcn != AD_POWEROFF)
2036 		return;
2037 
2038 	switch (apic_poweroff_method) {
2039 		case APIC_POWEROFF_VIA_RTC:
2040 
2041 			/* select the extended NVRAM bank in the RTC */
2042 			outb(CMOS_ADDR, RTC_REGA);
2043 			byte = inb(CMOS_DATA);
2044 			outb(CMOS_DATA, (byte | EXT_BANK));
2045 
2046 			outb(CMOS_ADDR, PFR_REG);
2047 
2048 			/* for Predator must toggle the PAB bit */
2049 			byte = inb(CMOS_DATA);
2050 
2051 			/*
2052 			 * clear power active bar, wakeup alarm and
2053 			 * kickstart
2054 			 */
2055 			byte &= ~(PAB_CBIT | WF_FLAG | KS_FLAG);
2056 			outb(CMOS_DATA, byte);
2057 
2058 			/* delay before next write */
2059 			drv_usecwait(1000);
2060 
2061 			/* for S40 the following would suffice */
2062 			byte = inb(CMOS_DATA);
2063 
2064 			/* power active bar control bit */
2065 			byte |= PAB_CBIT;
2066 			outb(CMOS_DATA, byte);
2067 
2068 			break;
2069 
2070 		case APIC_POWEROFF_VIA_ASPEN_BMC:
2071 			restarts = 0;
2072 restart_aspen_bmc:
2073 			if (++restarts == 3)
2074 				break;
2075 			attempts = 0;
2076 			do {
2077 				byte = inb(MISMIC_FLAG_REGISTER);
2078 				byte &= MISMIC_BUSY_MASK;
2079 				if (byte != 0) {
2080 					drv_usecwait(1000);
2081 					if (attempts >= 3)
2082 						goto restart_aspen_bmc;
2083 					++attempts;
2084 				}
2085 			} while (byte != 0);
2086 			outb(MISMIC_CNTL_REGISTER, CC_SMS_GET_STATUS);
2087 			byte = inb(MISMIC_FLAG_REGISTER);
2088 			byte |= 0x1;
2089 			outb(MISMIC_FLAG_REGISTER, byte);
2090 			i = 0;
2091 			for (; i < (sizeof (aspen_bmc)/sizeof (aspen_bmc[0]));
2092 			    i++) {
2093 				attempts = 0;
2094 				do {
2095 					byte = inb(MISMIC_FLAG_REGISTER);
2096 					byte &= MISMIC_BUSY_MASK;
2097 					if (byte != 0) {
2098 						drv_usecwait(1000);
2099 						if (attempts >= 3)
2100 							goto restart_aspen_bmc;
2101 						++attempts;
2102 					}
2103 				} while (byte != 0);
2104 				outb(MISMIC_CNTL_REGISTER, aspen_bmc[i].cntl);
2105 				outb(MISMIC_DATA_REGISTER, aspen_bmc[i].data);
2106 				byte = inb(MISMIC_FLAG_REGISTER);
2107 				byte |= 0x1;
2108 				outb(MISMIC_FLAG_REGISTER, byte);
2109 			}
2110 			break;
2111 
2112 		case APIC_POWEROFF_VIA_SITKA_BMC:
2113 			restarts = 0;
2114 restart_sitka_bmc:
2115 			if (++restarts == 3)
2116 				break;
2117 			attempts = 0;
2118 			do {
2119 				byte = inb(SMS_STATUS_REGISTER);
2120 				byte &= SMS_STATE_MASK;
2121 				if ((byte == SMS_READ_STATE) ||
2122 				    (byte == SMS_WRITE_STATE)) {
2123 					drv_usecwait(1000);
2124 					if (attempts >= 3)
2125 						goto restart_sitka_bmc;
2126 					++attempts;
2127 				}
2128 			} while ((byte == SMS_READ_STATE) ||
2129 			    (byte == SMS_WRITE_STATE));
2130 			outb(SMS_COMMAND_REGISTER, SMS_GET_STATUS);
2131 			i = 0;
2132 			for (; i < (sizeof (sitka_bmc)/sizeof (sitka_bmc[0]));
2133 			    i++) {
2134 				attempts = 0;
2135 				do {
2136 					byte = inb(SMS_STATUS_REGISTER);
2137 					byte &= SMS_IBF_MASK;
2138 					if (byte != 0) {
2139 						drv_usecwait(1000);
2140 						if (attempts >= 3)
2141 							goto restart_sitka_bmc;
2142 						++attempts;
2143 					}
2144 				} while (byte != 0);
2145 				outb(sitka_bmc[i].port, sitka_bmc[i].data);
2146 			}
2147 			break;
2148 
2149 		case APIC_POWEROFF_NONE:
2150 
2151 			/* If no APIC direct method, we will try using ACPI */
2152 			if (apic_enable_acpi) {
2153 				if (acpi_poweroff() == 1)
2154 					return;
2155 			} else
2156 				return;
2157 
2158 			break;
2159 	}
2160 	/*
2161 	 * Wait a limited time here for power to go off.
2162 	 * If the power does not go off, then there was a
2163 	 * problem and we should continue to the halt which
2164 	 * prints a message for the user to press a key to
2165 	 * reboot.
2166 	 */
2167 	drv_usecwait(7000000); /* wait seven seconds */
2168 
2169 }
2170 
2171 /*
2172  * Try and disable all interrupts. We just assign interrupts to other
2173  * processors based on policy. If any were bound by user request, we
2174  * let them continue and return failure. We do not bother to check
2175  * for cache affinity while rebinding.
2176  */
2177 
2178 static int
2179 apic_disable_intr(processorid_t cpun)
2180 {
2181 	int bind_cpu = 0, i, hardbound = 0;
2182 	apic_irq_t *irq_ptr;
2183 	ulong_t iflag;
2184 
2185 	iflag = intr_clear();
2186 	lock_set(&apic_ioapic_lock);
2187 
2188 	for (i = 0; i <= APIC_MAX_VECTOR; i++) {
2189 		if (apic_reprogram_info[i].done == B_FALSE) {
2190 			if (apic_reprogram_info[i].bindcpu == cpun) {
2191 				/*
2192 				 * CPU is busy -- it's the target of
2193 				 * a pending reprogramming attempt
2194 				 */
2195 				lock_clear(&apic_ioapic_lock);
2196 				intr_restore(iflag);
2197 				return (PSM_FAILURE);
2198 			}
2199 		}
2200 	}
2201 
2202 	apic_cpus[cpun].aci_status &= ~APIC_CPU_INTR_ENABLE;
2203 
2204 	apic_cpus[cpun].aci_curipl = 0;
2205 
2206 	i = apic_min_device_irq;
2207 	for (; i <= apic_max_device_irq; i++) {
2208 		/*
2209 		 * If there are bound interrupts on this cpu, then
2210 		 * rebind them to other processors.
2211 		 */
2212 		if ((irq_ptr = apic_irq_table[i]) != NULL) {
2213 			ASSERT((irq_ptr->airq_temp_cpu == IRQ_UNBOUND) ||
2214 			    (irq_ptr->airq_temp_cpu == IRQ_UNINIT) ||
2215 			    (apic_cpu_in_range(irq_ptr->airq_temp_cpu)));
2216 
2217 			if (irq_ptr->airq_temp_cpu == (cpun | IRQ_USER_BOUND)) {
2218 				hardbound = 1;
2219 				continue;
2220 			}
2221 
2222 			if (irq_ptr->airq_temp_cpu == cpun) {
2223 				do {
2224 					bind_cpu = apic_find_next_cpu_intr();
2225 				} while (apic_rebind_all(irq_ptr, bind_cpu));
2226 			}
2227 		}
2228 	}
2229 
2230 	lock_clear(&apic_ioapic_lock);
2231 	intr_restore(iflag);
2232 
2233 	if (hardbound) {
2234 		cmn_err(CE_WARN, "Could not disable interrupts on %d"
2235 		    "due to user bound interrupts", cpun);
2236 		return (PSM_FAILURE);
2237 	}
2238 	else
2239 		return (PSM_SUCCESS);
2240 }
2241 
2242 /*
2243  * Bind interrupts to the CPU's local APIC.
2244  * Interrupts should not be bound to a CPU's local APIC until the CPU
2245  * is ready to receive interrupts.
2246  */
2247 static void
2248 apic_enable_intr(processorid_t cpun)
2249 {
2250 	int	i;
2251 	apic_irq_t *irq_ptr;
2252 	ulong_t iflag;
2253 
2254 	iflag = intr_clear();
2255 	lock_set(&apic_ioapic_lock);
2256 
2257 	apic_cpus[cpun].aci_status |= APIC_CPU_INTR_ENABLE;
2258 
2259 	i = apic_min_device_irq;
2260 	for (i = apic_min_device_irq; i <= apic_max_device_irq; i++) {
2261 		if ((irq_ptr = apic_irq_table[i]) != NULL) {
2262 			if ((irq_ptr->airq_cpu & ~IRQ_USER_BOUND) == cpun) {
2263 				(void) apic_rebind_all(irq_ptr,
2264 				    irq_ptr->airq_cpu);
2265 			}
2266 		}
2267 	}
2268 
2269 	lock_clear(&apic_ioapic_lock);
2270 	intr_restore(iflag);
2271 }
2272 
2273 
2274 /*
2275  * This function will reprogram the timer.
2276  *
2277  * When in oneshot mode the argument is the absolute time in future to
2278  * generate the interrupt at.
2279  *
2280  * When in periodic mode, the argument is the interval at which the
2281  * interrupts should be generated. There is no need to support the periodic
2282  * mode timer change at this time.
2283  */
2284 static void
2285 apic_timer_reprogram(hrtime_t time)
2286 {
2287 	hrtime_t now;
2288 	uint_t ticks;
2289 	int64_t delta;
2290 
2291 	/*
2292 	 * We should be called from high PIL context (CBE_HIGH_PIL),
2293 	 * so kpreempt is disabled.
2294 	 */
2295 
2296 	if (!apic_oneshot) {
2297 		/* time is the interval for periodic mode */
2298 		ticks = APIC_NSECS_TO_TICKS(time);
2299 	} else {
2300 		/* one shot mode */
2301 
2302 		now = gethrtime();
2303 		delta = time - now;
2304 
2305 		if (delta <= 0) {
2306 			/*
2307 			 * requested to generate an interrupt in the past
2308 			 * generate an interrupt as soon as possible
2309 			 */
2310 			ticks = apic_min_timer_ticks;
2311 		} else if (delta > apic_nsec_max) {
2312 			/*
2313 			 * requested to generate an interrupt at a time
2314 			 * further than what we are capable of. Set to max
2315 			 * the hardware can handle
2316 			 */
2317 
2318 			ticks = APIC_MAXVAL;
2319 #ifdef DEBUG
2320 			cmn_err(CE_CONT, "apic_timer_reprogram, request at"
2321 			    "  %lld  too far in future, current time"
2322 			    "  %lld \n", time, now);
2323 #endif
2324 		} else
2325 			ticks = APIC_NSECS_TO_TICKS(delta);
2326 	}
2327 
2328 	if (ticks < apic_min_timer_ticks)
2329 		ticks = apic_min_timer_ticks;
2330 
2331 	apic_reg_ops->apic_write(APIC_INIT_COUNT, ticks);
2332 }
2333 
2334 /*
2335  * This function will enable timer interrupts.
2336  */
2337 static void
2338 apic_timer_enable(void)
2339 {
2340 	/*
2341 	 * We should be Called from high PIL context (CBE_HIGH_PIL),
2342 	 * so kpreempt is disabled.
2343 	 */
2344 
2345 	if (!apic_oneshot) {
2346 		apic_reg_ops->apic_write(APIC_LOCAL_TIMER,
2347 		    (apic_clkvect + APIC_BASE_VECT) | AV_TIME);
2348 	} else {
2349 		/* one shot */
2350 		apic_reg_ops->apic_write(APIC_LOCAL_TIMER,
2351 		    (apic_clkvect + APIC_BASE_VECT));
2352 	}
2353 }
2354 
2355 /*
2356  * This function will disable timer interrupts.
2357  */
2358 static void
2359 apic_timer_disable(void)
2360 {
2361 	/*
2362 	 * We should be Called from high PIL context (CBE_HIGH_PIL),
2363 	 * so kpreempt is disabled.
2364 	 */
2365 	apic_reg_ops->apic_write(APIC_LOCAL_TIMER,
2366 	    (apic_clkvect + APIC_BASE_VECT) | AV_MASK);
2367 }
2368 
2369 /*
2370  * Set timer far into the future and return timer
2371  * current Count in nanoseconds.
2372  */
2373 hrtime_t
2374 apic_timer_stop_count(void)
2375 {
2376 	hrtime_t	ns_val;
2377 	int		enable_val, count_val;
2378 
2379 	/*
2380 	 * Should be called with interrupts disabled.
2381 	 */
2382 	ASSERT(!interrupts_enabled());
2383 
2384 	enable_val = apic_reg_ops->apic_read(APIC_LOCAL_TIMER);
2385 	if ((enable_val & AV_MASK) == AV_MASK)
2386 		return ((hrtime_t)-1);		/* timer is disabled */
2387 
2388 	count_val = apic_reg_ops->apic_read(APIC_CURR_COUNT);
2389 	ns_val = APIC_TICKS_TO_NSECS(count_val);
2390 
2391 	apic_reg_ops->apic_write(APIC_INIT_COUNT, APIC_MAXVAL);
2392 
2393 	return (ns_val);
2394 }
2395 
2396 /*
2397  * Reprogram timer after Deep C-State.
2398  */
2399 void
2400 apic_timer_restart(hrtime_t time)
2401 {
2402 	apic_timer_reprogram(time);
2403 }
2404 
2405 ddi_periodic_t apic_periodic_id;
2406 
2407 /*
2408  * If this module needs a periodic handler for the interrupt distribution, it
2409  * can be added here. The argument to the periodic handler is not currently
2410  * used, but is reserved for future.
2411  */
2412 static void
2413 apic_post_cyclic_setup(void *arg)
2414 {
2415 _NOTE(ARGUNUSED(arg))
2416 	/* cpu_lock is held */
2417 	/* set up a periodic handler for intr redistribution */
2418 
2419 	/*
2420 	 * In peridoc mode intr redistribution processing is done in
2421 	 * apic_intr_enter during clk intr processing
2422 	 */
2423 	if (!apic_oneshot)
2424 		return;
2425 	/*
2426 	 * Register a periodical handler for the redistribution processing.
2427 	 * On X86, CY_LOW_LEVEL is mapped to the level 2 interrupt, so
2428 	 * DDI_IPL_2 should be passed to ddi_periodic_add() here.
2429 	 */
2430 	apic_periodic_id = ddi_periodic_add(
2431 	    (void (*)(void *))apic_redistribute_compute, NULL,
2432 	    apic_redistribute_sample_interval, DDI_IPL_2);
2433 }
2434 
2435 static void
2436 apic_redistribute_compute(void)
2437 {
2438 	int	i, j, max_busy;
2439 
2440 	if (apic_enable_dynamic_migration) {
2441 		if (++apic_nticks == apic_sample_factor_redistribution) {
2442 			/*
2443 			 * Time to call apic_intr_redistribute().
2444 			 * reset apic_nticks. This will cause max_busy
2445 			 * to be calculated below and if it is more than
2446 			 * apic_int_busy, we will do the whole thing
2447 			 */
2448 			apic_nticks = 0;
2449 		}
2450 		max_busy = 0;
2451 		for (i = 0; i < apic_nproc; i++) {
2452 			if (!apic_cpu_in_range(i))
2453 				continue;
2454 
2455 			/*
2456 			 * Check if curipl is non zero & if ISR is in
2457 			 * progress
2458 			 */
2459 			if (((j = apic_cpus[i].aci_curipl) != 0) &&
2460 			    (apic_cpus[i].aci_ISR_in_progress & (1 << j))) {
2461 
2462 				int	irq;
2463 				apic_cpus[i].aci_busy++;
2464 				irq = apic_cpus[i].aci_current[j];
2465 				apic_irq_table[irq]->airq_busy++;
2466 			}
2467 
2468 			if (!apic_nticks &&
2469 			    (apic_cpus[i].aci_busy > max_busy))
2470 				max_busy = apic_cpus[i].aci_busy;
2471 		}
2472 		if (!apic_nticks) {
2473 			if (max_busy > apic_int_busy_mark) {
2474 			/*
2475 			 * We could make the following check be
2476 			 * skipped > 1 in which case, we get a
2477 			 * redistribution at half the busy mark (due to
2478 			 * double interval). Need to be able to collect
2479 			 * more empirical data to decide if that is a
2480 			 * good strategy. Punt for now.
2481 			 */
2482 				if (apic_skipped_redistribute) {
2483 					apic_cleanup_busy();
2484 					apic_skipped_redistribute = 0;
2485 				} else {
2486 					apic_intr_redistribute();
2487 				}
2488 			} else
2489 				apic_skipped_redistribute++;
2490 		}
2491 	}
2492 }
2493 
2494 
2495 /*
2496  * The following functions are in the platform specific file so that they
2497  * can be different functions depending on whether we are running on
2498  * bare metal or a hypervisor.
2499  */
2500 
2501 /*
2502  * map an apic for memory-mapped access
2503  */
2504 uint32_t *
2505 mapin_apic(uint32_t addr, size_t len, int flags)
2506 {
2507 	/*LINTED: pointer cast may result in improper alignment */
2508 	return ((uint32_t *)psm_map_phys(addr, len, flags));
2509 }
2510 
2511 uint32_t *
2512 mapin_ioapic(uint32_t addr, size_t len, int flags)
2513 {
2514 	return (mapin_apic(addr, len, flags));
2515 }
2516 
2517 /*
2518  * unmap an apic
2519  */
2520 void
2521 mapout_apic(caddr_t addr, size_t len)
2522 {
2523 	psm_unmap_phys(addr, len);
2524 }
2525 
2526 void
2527 mapout_ioapic(caddr_t addr, size_t len)
2528 {
2529 	mapout_apic(addr, len);
2530 }
2531 
2532 /*
2533  * Check to make sure there are enough irq slots
2534  */
2535 int
2536 apic_check_free_irqs(int count)
2537 {
2538 	int i, avail;
2539 
2540 	avail = 0;
2541 	for (i = APIC_FIRST_FREE_IRQ; i < APIC_RESV_IRQ; i++) {
2542 		if ((apic_irq_table[i] == NULL) ||
2543 		    apic_irq_table[i]->airq_mps_intr_index == FREE_INDEX) {
2544 			if (++avail >= count)
2545 				return (PSM_SUCCESS);
2546 		}
2547 	}
2548 	return (PSM_FAILURE);
2549 }
2550 
2551 /*
2552  * This function allocates "count" MSI vector(s) for the given "dip/pri/type"
2553  */
2554 int
2555 apic_alloc_msi_vectors(dev_info_t *dip, int inum, int count, int pri,
2556     int behavior)
2557 {
2558 	int	rcount, i;
2559 	uchar_t	start, irqno;
2560 	uint32_t cpu;
2561 	major_t	major;
2562 	apic_irq_t	*irqptr;
2563 
2564 	DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_msi_vectors: dip=0x%p "
2565 	    "inum=0x%x  pri=0x%x count=0x%x behavior=%d\n",
2566 	    (void *)dip, inum, pri, count, behavior));
2567 
2568 	if (count > 1) {
2569 		if (behavior == DDI_INTR_ALLOC_STRICT &&
2570 		    apic_multi_msi_enable == 0)
2571 			return (0);
2572 		if (apic_multi_msi_enable == 0)
2573 			count = 1;
2574 	}
2575 
2576 	if ((rcount = apic_navail_vector(dip, pri)) > count)
2577 		rcount = count;
2578 	else if (rcount == 0 || (rcount < count &&
2579 	    behavior == DDI_INTR_ALLOC_STRICT))
2580 		return (0);
2581 
2582 	/* if not ISP2, then round it down */
2583 	if (!ISP2(rcount))
2584 		rcount = 1 << (highbit(rcount) - 1);
2585 
2586 	mutex_enter(&airq_mutex);
2587 
2588 	for (start = 0; rcount > 0; rcount >>= 1) {
2589 		if ((start = apic_find_multi_vectors(pri, rcount)) != 0 ||
2590 		    behavior == DDI_INTR_ALLOC_STRICT)
2591 			break;
2592 	}
2593 
2594 	if (start == 0) {
2595 		/* no vector available */
2596 		mutex_exit(&airq_mutex);
2597 		return (0);
2598 	}
2599 
2600 	if (apic_check_free_irqs(rcount) == PSM_FAILURE) {
2601 		/* not enough free irq slots available */
2602 		mutex_exit(&airq_mutex);
2603 		return (0);
2604 	}
2605 
2606 	major = (dip != NULL) ? ddi_driver_major(dip) : 0;
2607 	for (i = 0; i < rcount; i++) {
2608 		if ((irqno = apic_allocate_irq(apic_first_avail_irq)) ==
2609 		    (uchar_t)-1) {
2610 			/*
2611 			 * shouldn't happen because of the
2612 			 * apic_check_free_irqs() check earlier
2613 			 */
2614 			mutex_exit(&airq_mutex);
2615 			DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_msi_vectors: "
2616 			    "apic_allocate_irq failed\n"));
2617 			return (i);
2618 		}
2619 		apic_max_device_irq = max(irqno, apic_max_device_irq);
2620 		apic_min_device_irq = min(irqno, apic_min_device_irq);
2621 		irqptr = apic_irq_table[irqno];
2622 #ifdef	DEBUG
2623 		if (apic_vector_to_irq[start + i] != APIC_RESV_IRQ)
2624 			DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_msi_vectors: "
2625 			    "apic_vector_to_irq is not APIC_RESV_IRQ\n"));
2626 #endif
2627 		apic_vector_to_irq[start + i] = (uchar_t)irqno;
2628 
2629 		irqptr->airq_vector = (uchar_t)(start + i);
2630 		irqptr->airq_ioapicindex = (uchar_t)inum;	/* start */
2631 		irqptr->airq_intin_no = (uchar_t)rcount;
2632 		irqptr->airq_ipl = pri;
2633 		irqptr->airq_vector = start + i;
2634 		irqptr->airq_origirq = (uchar_t)(inum + i);
2635 		irqptr->airq_share_id = 0;
2636 		irqptr->airq_mps_intr_index = MSI_INDEX;
2637 		irqptr->airq_dip = dip;
2638 		irqptr->airq_major = major;
2639 		if (i == 0) /* they all bound to the same cpu */
2640 			cpu = irqptr->airq_cpu = apic_bind_intr(dip, irqno,
2641 			    0xff, 0xff);
2642 		else
2643 			irqptr->airq_cpu = cpu;
2644 		DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_msi_vectors: irq=0x%x "
2645 		    "dip=0x%p vector=0x%x origirq=0x%x pri=0x%x\n", irqno,
2646 		    (void *)irqptr->airq_dip, irqptr->airq_vector,
2647 		    irqptr->airq_origirq, pri));
2648 	}
2649 	mutex_exit(&airq_mutex);
2650 	return (rcount);
2651 }
2652 
2653 /*
2654  * This function allocates "count" MSI-X vector(s) for the given "dip/pri/type"
2655  */
2656 int
2657 apic_alloc_msix_vectors(dev_info_t *dip, int inum, int count, int pri,
2658     int behavior)
2659 {
2660 	int	rcount, i;
2661 	major_t	major;
2662 
2663 	mutex_enter(&airq_mutex);
2664 
2665 	if ((rcount = apic_navail_vector(dip, pri)) > count)
2666 		rcount = count;
2667 	else if (rcount == 0 || (rcount < count &&
2668 	    behavior == DDI_INTR_ALLOC_STRICT)) {
2669 		rcount = 0;
2670 		goto out;
2671 	}
2672 
2673 	if (apic_check_free_irqs(rcount) == PSM_FAILURE) {
2674 		/* not enough free irq slots available */
2675 		rcount = 0;
2676 		goto out;
2677 	}
2678 
2679 	major = (dip != NULL) ? ddi_driver_major(dip) : 0;
2680 	for (i = 0; i < rcount; i++) {
2681 		uchar_t	vector, irqno;
2682 		apic_irq_t	*irqptr;
2683 
2684 		if ((irqno = apic_allocate_irq(apic_first_avail_irq)) ==
2685 		    (uchar_t)-1) {
2686 			/*
2687 			 * shouldn't happen because of the
2688 			 * apic_check_free_irqs() check earlier
2689 			 */
2690 			DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_msix_vectors: "
2691 			    "apic_allocate_irq failed\n"));
2692 			rcount = i;
2693 			goto out;
2694 		}
2695 		if ((vector = apic_allocate_vector(pri, irqno, 1)) == 0) {
2696 			/*
2697 			 * shouldn't happen because of the
2698 			 * apic_navail_vector() call earlier
2699 			 */
2700 			DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_msix_vectors: "
2701 			    "apic_allocate_vector failed\n"));
2702 			rcount = i;
2703 			goto out;
2704 		}
2705 		apic_max_device_irq = max(irqno, apic_max_device_irq);
2706 		apic_min_device_irq = min(irqno, apic_min_device_irq);
2707 		irqptr = apic_irq_table[irqno];
2708 		irqptr->airq_vector = (uchar_t)vector;
2709 		irqptr->airq_ipl = pri;
2710 		irqptr->airq_origirq = (uchar_t)(inum + i);
2711 		irqptr->airq_share_id = 0;
2712 		irqptr->airq_mps_intr_index = MSIX_INDEX;
2713 		irqptr->airq_dip = dip;
2714 		irqptr->airq_major = major;
2715 		irqptr->airq_cpu = apic_bind_intr(dip, irqno, 0xff, 0xff);
2716 	}
2717 out:
2718 	mutex_exit(&airq_mutex);
2719 	return (rcount);
2720 }
2721 
2722 /*
2723  * Allocate a free vector for irq at ipl. Takes care of merging of multiple
2724  * IPLs into a single APIC level as well as stretching some IPLs onto multiple
2725  * levels. APIC_HI_PRI_VECTS interrupts are reserved for high priority
2726  * requests and allocated only when pri is set.
2727  */
2728 uchar_t
2729 apic_allocate_vector(int ipl, int irq, int pri)
2730 {
2731 	int	lowest, highest, i;
2732 
2733 	highest = apic_ipltopri[ipl] + APIC_VECTOR_MASK;
2734 	lowest = apic_ipltopri[ipl - 1] + APIC_VECTOR_PER_IPL;
2735 
2736 	if (highest < lowest) /* Both ipl and ipl - 1 map to same pri */
2737 		lowest -= APIC_VECTOR_PER_IPL;
2738 
2739 #ifdef	DEBUG
2740 	if (apic_restrict_vector)	/* for testing shared interrupt logic */
2741 		highest = lowest + apic_restrict_vector + APIC_HI_PRI_VECTS;
2742 #endif /* DEBUG */
2743 	if (pri == 0)
2744 		highest -= APIC_HI_PRI_VECTS;
2745 
2746 	for (i = lowest; i <= highest; i++) {
2747 		if (APIC_CHECK_RESERVE_VECTORS(i))
2748 			continue;
2749 		if (apic_vector_to_irq[i] == APIC_RESV_IRQ) {
2750 			apic_vector_to_irq[i] = (uchar_t)irq;
2751 			return (i);
2752 		}
2753 	}
2754 
2755 	return (0);
2756 }
2757 
2758 /* Mark vector as not being used by any irq */
2759 void
2760 apic_free_vector(uchar_t vector)
2761 {
2762 	apic_vector_to_irq[vector] = APIC_RESV_IRQ;
2763 }
2764 
2765 uint32_t
2766 ioapic_read(int ioapic_ix, uint32_t reg)
2767 {
2768 	volatile uint32_t *ioapic;
2769 
2770 	ioapic = apicioadr[ioapic_ix];
2771 	ioapic[APIC_IO_REG] = reg;
2772 	return (ioapic[APIC_IO_DATA]);
2773 }
2774 
2775 void
2776 ioapic_write(int ioapic_ix, uint32_t reg, uint32_t value)
2777 {
2778 	volatile uint32_t *ioapic;
2779 
2780 	ioapic = apicioadr[ioapic_ix];
2781 	ioapic[APIC_IO_REG] = reg;
2782 	ioapic[APIC_IO_DATA] = value;
2783 }
2784 
2785 void
2786 ioapic_write_eoi(int ioapic_ix, uint32_t value)
2787 {
2788 	volatile uint32_t *ioapic;
2789 
2790 	ioapic = apicioadr[ioapic_ix];
2791 	ioapic[APIC_IO_EOI] = value;
2792 }
2793 
2794 /*
2795  * Call rebind to do the actual programming.
2796  * Must be called with interrupts disabled and apic_ioapic_lock held
2797  * 'p' is polymorphic -- if this function is called to process a deferred
2798  * reprogramming, p is of type 'struct ioapic_reprogram_data *', from which
2799  * the irq pointer is retrieved.  If not doing deferred reprogramming,
2800  * p is of the type 'apic_irq_t *'.
2801  *
2802  * apic_ioapic_lock must be held across this call, as it protects apic_rebind
2803  * and it protects apic_find_next_cpu_intr() from a race in which a CPU can be
2804  * taken offline after a cpu is selected, but before apic_rebind is called to
2805  * bind interrupts to it.
2806  */
2807 int
2808 apic_setup_io_intr(void *p, int irq, boolean_t deferred)
2809 {
2810 	apic_irq_t *irqptr;
2811 	struct ioapic_reprogram_data *drep = NULL;
2812 	int rv;
2813 
2814 	if (deferred) {
2815 		drep = (struct ioapic_reprogram_data *)p;
2816 		ASSERT(drep != NULL);
2817 		irqptr = drep->irqp;
2818 	} else
2819 		irqptr = (apic_irq_t *)p;
2820 
2821 	ASSERT(irqptr != NULL);
2822 
2823 	rv = apic_rebind(irqptr, apic_irq_table[irq]->airq_cpu, drep);
2824 	if (rv) {
2825 		/*
2826 		 * CPU is not up or interrupts are disabled. Fall back to
2827 		 * the first available CPU
2828 		 */
2829 		rv = apic_rebind(irqptr, apic_find_next_cpu_intr(), drep);
2830 	}
2831 
2832 	return (rv);
2833 }
2834 
2835 
2836 uchar_t
2837 apic_modify_vector(uchar_t vector, int irq)
2838 {
2839 	apic_vector_to_irq[vector] = (uchar_t)irq;
2840 	return (vector);
2841 }
2842 
2843 char *
2844 apic_get_apic_type()
2845 {
2846 	return (apic_psm_info.p_mach_idstring);
2847 }
2848 
2849 /*
2850  * Switch between safe and x2APIC IPI sending method.
2851  * CPU may power on in xapic mode or x2apic mode. If CPU needs to send IPI to
2852  * other CPUs before entering x2APIC mode, it still needs to xAPIC method.
2853  * Before sending StartIPI to target CPU, psm_send_ipi will be changed to
2854  * apic_common_send_ipi, which detects current local APIC mode and use right
2855  * method to send IPI. If some CPUs fail to start up, apic_poweron_cnt
2856  * won't return to zero, so apic_common_send_ipi will always be used.
2857  * psm_send_ipi can't be simply changed back to x2apic_send_ipi if some CPUs
2858  * failed to start up because those failed CPUs may recover itself later at
2859  * unpredictable time.
2860  */
2861 static void
2862 apic_switch_ipi_callback(boolean_t enter)
2863 {
2864 	ulong_t iflag;
2865 	struct psm_ops *pops = &apic_ops;
2866 
2867 	iflag = intr_clear();
2868 	lock_set(&apic_mode_switch_lock);
2869 	if (enter) {
2870 		ASSERT(apic_poweron_cnt >= 0);
2871 		if (apic_poweron_cnt == 0) {
2872 			pops->psm_send_ipi = apic_common_send_ipi;
2873 			send_dirintf = pops->psm_send_ipi;
2874 		}
2875 		apic_poweron_cnt++;
2876 	} else {
2877 		ASSERT(apic_poweron_cnt > 0);
2878 		apic_poweron_cnt--;
2879 		if (apic_poweron_cnt == 0) {
2880 			pops->psm_send_ipi = x2apic_send_ipi;
2881 			send_dirintf = pops->psm_send_ipi;
2882 		}
2883 	}
2884 	lock_clear(&apic_mode_switch_lock);
2885 	intr_restore(iflag);
2886 }
2887 
2888 void
2889 x2apic_update_psm()
2890 {
2891 	struct psm_ops *pops = &apic_ops;
2892 
2893 	ASSERT(pops != NULL);
2894 
2895 	pops->psm_intr_exit = x2apic_intr_exit;
2896 	pops->psm_setspl = x2apic_setspl;
2897 	pops->psm_send_ipi =  x2apic_send_ipi;
2898 	send_dirintf = pops->psm_send_ipi;
2899 
2900 	apic_mode = LOCAL_X2APIC;
2901 	apic_change_ops();
2902 }
2903 
2904 static void
2905 apic_intrmap_init(int apic_mode)
2906 {
2907 	int suppress_brdcst_eoi = 0;
2908 
2909 	if (psm_vt_ops != NULL) {
2910 		/*
2911 		 * Since X2APIC requires the use of interrupt remapping
2912 		 * (though this is not documented explicitly in the Intel
2913 		 * documentation (yet)), initialize interrupt remapping
2914 		 * support before initializing the X2APIC unit.
2915 		 */
2916 		if (((apic_intrmap_ops_t *)psm_vt_ops)->
2917 		    apic_intrmap_init(apic_mode) == DDI_SUCCESS) {
2918 
2919 			apic_vt_ops = psm_vt_ops;
2920 
2921 			/*
2922 			 * We leverage the interrupt remapping engine to
2923 			 * suppress broadcast EOI; thus we must send the
2924 			 * directed EOI with the directed-EOI handler.
2925 			 */
2926 			if (apic_directed_EOI_supported() == 0) {
2927 				suppress_brdcst_eoi = 1;
2928 			}
2929 
2930 			apic_vt_ops->apic_intrmap_enable(suppress_brdcst_eoi);
2931 
2932 			if (apic_detect_x2apic()) {
2933 				apic_enable_x2apic();
2934 			}
2935 
2936 			if (apic_directed_EOI_supported() == 0) {
2937 				apic_set_directed_EOI_handler();
2938 			}
2939 		}
2940 	}
2941 }
2942 
2943 /*ARGSUSED*/
2944 static void
2945 apic_record_ioapic_rdt(apic_irq_t *irq_ptr, ioapic_rdt_t *irdt)
2946 {
2947 	irdt->ir_hi <<= APIC_ID_BIT_OFFSET;
2948 }
2949 
2950 /*ARGSUSED*/
2951 static void
2952 apic_record_msi(apic_irq_t *irq_ptr, msi_regs_t *mregs)
2953 {
2954 	mregs->mr_addr = MSI_ADDR_HDR |
2955 	    (MSI_ADDR_RH_FIXED << MSI_ADDR_RH_SHIFT) |
2956 	    (MSI_ADDR_DM_PHYSICAL << MSI_ADDR_DM_SHIFT) |
2957 	    (mregs->mr_addr << MSI_ADDR_DEST_SHIFT);
2958 	mregs->mr_data = (MSI_DATA_TM_EDGE << MSI_DATA_TM_SHIFT) |
2959 	    mregs->mr_data;
2960 }
2961