xref: /illumos-gate/usr/src/uts/i86pc/io/pcplusmp/apic.c (revision 4de2612967d06c4fdbf524a62556a1e8118a006f)
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, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 /*
30  * PSMI 1.1 extensions are supported only in 2.6 and later versions.
31  * PSMI 1.2 extensions are supported only in 2.7 and later versions.
32  * PSMI 1.3 and 1.4 extensions are supported in Solaris 10.
33  * PSMI 1.5 extensions are supported in Solaris Nevada.
34  */
35 #define	PSMI_1_5
36 
37 #include <sys/processor.h>
38 #include <sys/time.h>
39 #include <sys/psm.h>
40 #include <sys/smp_impldefs.h>
41 #include <sys/cram.h>
42 #include <sys/acpi/acpi.h>
43 #include <sys/acpica.h>
44 #include <sys/psm_common.h>
45 #include "apic.h"
46 #include <sys/pit.h>
47 #include <sys/ddi.h>
48 #include <sys/sunddi.h>
49 #include <sys/ddi_impldefs.h>
50 #include <sys/pci.h>
51 #include <sys/promif.h>
52 #include <sys/x86_archext.h>
53 #include <sys/cpc_impl.h>
54 #include <sys/uadmin.h>
55 #include <sys/panic.h>
56 #include <sys/debug.h>
57 #include <sys/archsystm.h>
58 #include <sys/trap.h>
59 #include <sys/machsystm.h>
60 #include <sys/cpuvar.h>
61 #include <sys/rm_platter.h>
62 #include <sys/privregs.h>
63 #include <sys/cyclic.h>
64 #include <sys/note.h>
65 #include <sys/pci_intr_lib.h>
66 
67 /*
68  *	Local Function Prototypes
69  */
70 static void apic_init_intr();
71 static void apic_ret();
72 static int apic_handle_defconf();
73 static int apic_parse_mpct(caddr_t mpct, int bypass);
74 static struct apic_mpfps_hdr *apic_find_fps_sig(caddr_t fptr, int size);
75 static int apic_checksum(caddr_t bptr, int len);
76 static int get_apic_cmd1();
77 static int get_apic_pri();
78 static int apic_find_bus_type(char *bus);
79 static int apic_find_bus(int busid);
80 static int apic_find_bus_id(int bustype);
81 static struct apic_io_intr *apic_find_io_intr(int irqno);
82 int apic_allocate_irq(int irq);
83 static int apic_find_free_irq(int start, int end);
84 static uchar_t apic_allocate_vector(int ipl, int irq, int pri);
85 static void apic_modify_vector(uchar_t vector, int irq);
86 static void apic_mark_vector(uchar_t oldvector, uchar_t newvector);
87 static uchar_t apic_xlate_vector(uchar_t oldvector);
88 static void apic_xlate_vector_free_timeout_handler(void *arg);
89 static void apic_free_vector(uchar_t vector);
90 static void apic_reprogram_timeout_handler(void *arg);
91 static int apic_check_stuck_interrupt(apic_irq_t *irq_ptr, int old_bind_cpu,
92     int new_bind_cpu, volatile int32_t *ioapic, int intin_no, int which_irq);
93 static int apic_setup_io_intr(apic_irq_t *irqptr, int irq);
94 static int apic_setup_io_intr_deferred(apic_irq_t *irqptr, int irq);
95 static void apic_record_rdt_entry(apic_irq_t *irqptr, int irq);
96 static struct apic_io_intr *apic_find_io_intr_w_busid(int irqno, int busid);
97 static int apic_find_intin(uchar_t ioapic, uchar_t intin);
98 static int apic_handle_pci_pci_bridge(dev_info_t *idip, int child_devno,
99     int child_ipin, struct apic_io_intr **intrp);
100 static int apic_setup_irq_table(dev_info_t *dip, int irqno,
101     struct apic_io_intr *intrp, struct intrspec *ispec, iflag_t *intr_flagp,
102     int type);
103 static int apic_setup_sci_irq_table(int irqno, uchar_t ipl,
104     iflag_t *intr_flagp);
105 static void apic_nmi_intr(caddr_t arg);
106 uchar_t apic_bind_intr(dev_info_t *dip, int irq, uchar_t ioapicid,
107     uchar_t intin);
108 static int apic_rebind(apic_irq_t *irq_ptr, int bind_cpu, int acquire_lock,
109     int when);
110 static int apic_rebind_all(apic_irq_t *irq_ptr, int bind_cpu, int safe);
111 static void apic_intr_redistribute();
112 static void apic_cleanup_busy();
113 static void apic_set_pwroff_method_from_mpcnfhdr(struct apic_mp_cnf_hdr *hdrp);
114 int apic_introp_xlate(dev_info_t *dip, struct intrspec *ispec, int type);
115 
116 /* ACPI support routines */
117 static int acpi_probe(void);
118 static int apic_acpi_irq_configure(acpi_psm_lnk_t *acpipsmlnkp, dev_info_t *dip,
119     int *pci_irqp, iflag_t *intr_flagp);
120 
121 static int apic_acpi_translate_pci_irq(dev_info_t *dip, int busid, int devid,
122     int ipin, int *pci_irqp, iflag_t *intr_flagp);
123 static uchar_t acpi_find_ioapic(int irq);
124 static int acpi_intr_compatible(iflag_t iflag1, iflag_t iflag2);
125 
126 /*
127  *	standard MP entries
128  */
129 static int	apic_probe();
130 static int	apic_clkinit();
131 static int	apic_getclkirq(int ipl);
132 static uint_t	apic_calibrate(volatile uint32_t *addr,
133     uint16_t *pit_ticks_adj);
134 static hrtime_t apic_gettime();
135 static hrtime_t apic_gethrtime();
136 static void	apic_init();
137 static void	apic_picinit(void);
138 static void	apic_cpu_start(processorid_t cpun, caddr_t rm_code);
139 static int	apic_post_cpu_start(void);
140 static void	apic_send_ipi(int cpun, int ipl);
141 static void	apic_set_softintr(int softintr);
142 static void	apic_set_idlecpu(processorid_t cpun);
143 static void	apic_unset_idlecpu(processorid_t cpun);
144 static int	apic_softlvl_to_irq(int ipl);
145 static int	apic_intr_enter(int ipl, int *vect);
146 static void	apic_intr_exit(int ipl, int vect);
147 static void	apic_setspl(int ipl);
148 static int	apic_addspl(int ipl, int vector, int min_ipl, int max_ipl);
149 static int	apic_delspl(int ipl, int vector, int min_ipl, int max_ipl);
150 static void	apic_shutdown(int cmd, int fcn);
151 static void	apic_preshutdown(int cmd, int fcn);
152 static int	apic_disable_intr(processorid_t cpun);
153 static void	apic_enable_intr(processorid_t cpun);
154 static processorid_t	apic_get_next_processorid(processorid_t cpun);
155 static int		apic_get_ipivect(int ipl, int type);
156 static void	apic_timer_reprogram(hrtime_t time);
157 static void	apic_timer_enable(void);
158 static void	apic_timer_disable(void);
159 static void	apic_post_cyclic_setup(void *arg);
160 extern int	apic_intr_ops(dev_info_t *, ddi_intr_handle_impl_t *,
161 		    psm_intr_op_t, int *);
162 
163 static int	apic_oneshot = 0;
164 int	apic_oneshot_enable = 1; /* to allow disabling one-shot capability */
165 
166 /*
167  * These variables are frequently accessed in apic_intr_enter(),
168  * apic_intr_exit and apic_setspl, so group them together
169  */
170 volatile uint32_t *apicadr =  NULL;	/* virtual addr of local APIC	*/
171 int apic_setspl_delay = 1;		/* apic_setspl - delay enable	*/
172 int apic_clkvect;
173 
174 /* ACPI SCI interrupt configuration; -1 if SCI not used */
175 int apic_sci_vect = -1;
176 iflag_t apic_sci_flags;
177 
178 /* vector at which error interrupts come in */
179 int apic_errvect;
180 int apic_enable_error_intr = 1;
181 int apic_error_display_delay = 100;
182 
183 /* vector at which performance counter overflow interrupts come in */
184 int apic_cpcovf_vect;
185 int apic_enable_cpcovf_intr = 1;
186 
187 /* Max wait time (in microsecs) for flags to clear in an RDT entry. */
188 static int apic_max_usecs_clear_pending = 1000;
189 
190 /* Amt of usecs to wait before checking if RDT flags have reset. */
191 #define	APIC_USECS_PER_WAIT_INTERVAL 100
192 
193 /* Maximum number of times to retry reprogramming via the timeout */
194 #define	APIC_REPROGRAM_MAX_TIMEOUTS 10
195 
196 /* timeout delay for IOAPIC delayed reprogramming */
197 #define	APIC_REPROGRAM_TIMEOUT_DELAY 5 /* microseconds */
198 
199 /* Parameter to apic_rebind(): Should reprogramming be done now or later? */
200 #define	DEFERRED 1
201 #define	IMMEDIATE 0
202 
203 /*
204  * number of bits per byte, from <sys/param.h>
205  */
206 #define	UCHAR_MAX	((1 << NBBY) - 1)
207 
208 uchar_t	apic_reserved_irqlist[MAX_ISA_IRQ];
209 
210 /*
211  * The following vector assignments influence the value of ipltopri and
212  * vectortoipl. Note that vectors 0 - 0x1f are not used. We can program
213  * idle to 0 and IPL 0 to 0x10 to differentiate idle in case
214  * we care to do so in future. Note some IPLs which are rarely used
215  * will share the vector ranges and heavily used IPLs (5 and 6) have
216  * a wide range.
217  *	IPL		Vector range.		as passed to intr_enter
218  *	0		none.
219  *	1,2,3		0x20-0x2f		0x0-0xf
220  *	4		0x30-0x3f		0x10-0x1f
221  *	5		0x40-0x5f		0x20-0x3f
222  *	6		0x60-0x7f		0x40-0x5f
223  *	7,8,9		0x80-0x8f		0x60-0x6f
224  *	10		0x90-0x9f		0x70-0x7f
225  *	11		0xa0-0xaf		0x80-0x8f
226  *	...		...
227  *	16		0xf0-0xff		0xd0-0xdf
228  */
229 uchar_t apic_vectortoipl[APIC_AVAIL_VECTOR / APIC_VECTOR_PER_IPL] = {
230 	3, 4, 5, 5, 6, 6, 9, 10, 11, 12, 13, 14, 15, 16
231 };
232 	/*
233 	 * The ipl of an ISR at vector X is apic_vectortoipl[X<<4]
234 	 * NOTE that this is vector as passed into intr_enter which is
235 	 * programmed vector - 0x20 (APIC_BASE_VECT)
236 	 */
237 
238 uchar_t	apic_ipltopri[MAXIPL + 1];	/* unix ipl to apic pri	*/
239 	/* The taskpri to be programmed into apic to mask given ipl */
240 
241 #if defined(__amd64)
242 uchar_t	apic_cr8pri[MAXIPL + 1];	/* unix ipl to cr8 pri	*/
243 #endif
244 
245 /*
246  * Patchable global variables.
247  */
248 int	apic_forceload = 0;
249 
250 #define	INTR_ROUND_ROBIN_WITH_AFFINITY	0
251 #define	INTR_ROUND_ROBIN		1
252 #define	INTR_LOWEST_PRIORITY		2
253 
254 int	apic_intr_policy = INTR_ROUND_ROBIN_WITH_AFFINITY;
255 
256 static int	apic_next_bind_cpu = 2; /* For round robin assignment */
257 					/* start with cpu 1 */
258 
259 int	apic_coarse_hrtime = 1;		/* 0 - use accurate slow gethrtime() */
260 					/* 1 - use gettime() for performance */
261 int	apic_flat_model = 0;		/* 0 - clustered. 1 - flat */
262 int	apic_enable_hwsoftint = 0;	/* 0 - disable, 1 - enable	*/
263 int	apic_enable_bind_log = 1;	/* 1 - display interrupt binding log */
264 int	apic_panic_on_nmi = 0;
265 int	apic_panic_on_apic_error = 0;
266 
267 int	apic_verbose = 0;
268 
269 /* Flag definitions for apic_verbose */
270 #define	APIC_VERBOSE_IOAPIC_FLAG		0x00000001
271 #define	APIC_VERBOSE_IRQ_FLAG			0x00000002
272 #define	APIC_VERBOSE_POWEROFF_FLAG		0x00000004
273 #define	APIC_VERBOSE_POWEROFF_PAUSE_FLAG	0x00000008
274 
275 
276 #define	APIC_VERBOSE_IOAPIC(fmt) \
277 	if (apic_verbose & APIC_VERBOSE_IOAPIC_FLAG) \
278 		cmn_err fmt;
279 
280 #define	APIC_VERBOSE_IRQ(fmt) \
281 	if (apic_verbose & APIC_VERBOSE_IRQ_FLAG) \
282 		cmn_err fmt;
283 
284 #define	APIC_VERBOSE_POWEROFF(fmt) \
285 	if (apic_verbose & APIC_VERBOSE_POWEROFF_FLAG) \
286 		prom_printf fmt;
287 
288 
289 /* Now the ones for Dynamic Interrupt distribution */
290 int	apic_enable_dynamic_migration = 1;
291 
292 /*
293  * If enabled, the distribution works as follows:
294  * On every interrupt entry, the current ipl for the CPU is set in cpu_info
295  * and the irq corresponding to the ipl is also set in the aci_current array.
296  * interrupt exit and setspl (due to soft interrupts) will cause the current
297  * ipl to be be changed. This is cache friendly as these frequently used
298  * paths write into a per cpu structure.
299  *
300  * Sampling is done by checking the structures for all CPUs and incrementing
301  * the busy field of the irq (if any) executing on each CPU and the busy field
302  * of the corresponding CPU.
303  * In periodic mode this is done on every clock interrupt.
304  * In one-shot mode, this is done thru a cyclic with an interval of
305  * apic_redistribute_sample_interval (default 10 milli sec).
306  *
307  * Every apic_sample_factor_redistribution times we sample, we do computations
308  * to decide which interrupt needs to be migrated (see comments
309  * before apic_intr_redistribute().
310  */
311 
312 /*
313  * Following 3 variables start as % and can be patched or set using an
314  * API to be defined in future. They will be scaled to
315  * sample_factor_redistribution which is in turn set to hertz+1 (in periodic
316  * mode), or 101 in one-shot mode to stagger it away from one sec processing
317  */
318 
319 int	apic_int_busy_mark = 60;
320 int	apic_int_free_mark = 20;
321 int	apic_diff_for_redistribution = 10;
322 
323 /* sampling interval for interrupt redistribution for dynamic migration */
324 int	apic_redistribute_sample_interval = NANOSEC / 100; /* 10 millisec */
325 
326 /*
327  * number of times we sample before deciding to redistribute interrupts
328  * for dynamic migration
329  */
330 int	apic_sample_factor_redistribution = 101;
331 
332 /* timeout for xlate_vector, mark_vector */
333 int	apic_revector_timeout = 16 * 10000; /* 160 millisec */
334 
335 int	apic_redist_cpu_skip = 0;
336 int	apic_num_imbalance = 0;
337 int	apic_num_rebind = 0;
338 
339 int	apic_nproc = 0;
340 int	apic_defconf = 0;
341 int	apic_irq_translate = 0;
342 int	apic_spec_rev = 0;
343 int	apic_imcrp = 0;
344 
345 int	apic_use_acpi = 1;	/* 1 = use ACPI, 0 = don't use ACPI */
346 int	apic_use_acpi_madt_only = 0;	/* 1=ONLY use MADT from ACPI */
347 
348 /*
349  * For interrupt link devices, if apic_unconditional_srs is set, an irq resource
350  * will be assigned (via _SRS). If it is not set, use the current
351  * irq setting (via _CRS), but only if that irq is in the set of possible
352  * irqs (returned by _PRS) for the device.
353  */
354 int	apic_unconditional_srs = 1;
355 
356 /*
357  * For interrupt link devices, if apic_prefer_crs is set when we are
358  * assigning an IRQ resource to a device, prefer the current IRQ setting
359  * over other possible irq settings under same conditions.
360  */
361 
362 int	apic_prefer_crs = 1;
363 
364 
365 /* minimum number of timer ticks to program to */
366 int apic_min_timer_ticks = 1;
367 /*
368  *	Local static data
369  */
370 static struct	psm_ops apic_ops = {
371 	apic_probe,
372 
373 	apic_init,
374 	apic_picinit,
375 	apic_intr_enter,
376 	apic_intr_exit,
377 	apic_setspl,
378 	apic_addspl,
379 	apic_delspl,
380 	apic_disable_intr,
381 	apic_enable_intr,
382 	apic_softlvl_to_irq,
383 	apic_set_softintr,
384 
385 	apic_set_idlecpu,
386 	apic_unset_idlecpu,
387 
388 	apic_clkinit,
389 	apic_getclkirq,
390 	(void (*)(void))NULL,		/* psm_hrtimeinit */
391 	apic_gethrtime,
392 
393 	apic_get_next_processorid,
394 	apic_cpu_start,
395 	apic_post_cpu_start,
396 	apic_shutdown,
397 	apic_get_ipivect,
398 	apic_send_ipi,
399 
400 	(int (*)(dev_info_t *, int))NULL,	/* psm_translate_irq */
401 	(int (*)(todinfo_t *))NULL,	/* psm_tod_get */
402 	(int (*)(todinfo_t *))NULL,	/* psm_tod_set */
403 	(void (*)(int, char *))NULL,	/* psm_notify_error */
404 	(void (*)(int))NULL,		/* psm_notify_func */
405 	apic_timer_reprogram,
406 	apic_timer_enable,
407 	apic_timer_disable,
408 	apic_post_cyclic_setup,
409 	apic_preshutdown,
410 	apic_intr_ops			/* Advanced DDI Interrupt framework */
411 };
412 
413 
414 static struct	psm_info apic_psm_info = {
415 	PSM_INFO_VER01_5,			/* version */
416 	PSM_OWN_EXCLUSIVE,			/* ownership */
417 	(struct psm_ops *)&apic_ops,		/* operation */
418 	"pcplusmp",				/* machine name */
419 	"pcplusmp v1.4 compatible %I%",
420 };
421 
422 static void *apic_hdlp;
423 
424 #ifdef DEBUG
425 #define	DENT		0x0001
426 int	apic_debug = 0;
427 /*
428  * set apic_restrict_vector to the # of vectors we want to allow per range
429  * useful in testing shared interrupt logic by setting it to 2 or 3
430  */
431 int	apic_restrict_vector = 0;
432 
433 #define	APIC_DEBUG_MSGBUFSIZE	2048
434 int	apic_debug_msgbuf[APIC_DEBUG_MSGBUFSIZE];
435 int	apic_debug_msgbufindex = 0;
436 
437 /*
438  * Put "int" info into debug buffer. No MP consistency, but light weight.
439  * Good enough for most debugging.
440  */
441 #define	APIC_DEBUG_BUF_PUT(x) \
442 	apic_debug_msgbuf[apic_debug_msgbufindex++] = x; \
443 	if (apic_debug_msgbufindex >= (APIC_DEBUG_MSGBUFSIZE - NCPU)) \
444 		apic_debug_msgbufindex = 0;
445 
446 #endif /* DEBUG */
447 
448 apic_cpus_info_t	*apic_cpus;
449 
450 static uint_t	apic_cpumask = 0;
451 static uint_t	apic_flag;
452 
453 /* Flag to indicate that we need to shut down all processors */
454 static uint_t	apic_shutdown_processors;
455 
456 uint_t apic_nsec_per_intr = 0;
457 
458 /*
459  * apic_let_idle_redistribute can have the following values:
460  * 0 - If clock decremented it from 1 to 0, clock has to call redistribute.
461  * apic_redistribute_lock prevents multiple idle cpus from redistributing
462  */
463 int	apic_num_idle_redistributions = 0;
464 static	int apic_let_idle_redistribute = 0;
465 static	uint_t apic_nticks = 0;
466 static	uint_t apic_skipped_redistribute = 0;
467 
468 /* to gather intr data and redistribute */
469 static void apic_redistribute_compute(void);
470 
471 static	uint_t last_count_read = 0;
472 static	lock_t	apic_gethrtime_lock;
473 volatile int	apic_hrtime_stamp = 0;
474 volatile hrtime_t apic_nsec_since_boot = 0;
475 static uint_t apic_hertz_count, apic_nsec_per_tick;
476 static hrtime_t apic_nsec_max;
477 
478 static	hrtime_t	apic_last_hrtime = 0;
479 int		apic_hrtime_error = 0;
480 int		apic_remote_hrterr = 0;
481 int		apic_num_nmis = 0;
482 int		apic_apic_error = 0;
483 int		apic_num_apic_errors = 0;
484 int		apic_num_cksum_errors = 0;
485 
486 static	uchar_t	apic_io_id[MAX_IO_APIC];
487 static	uchar_t	apic_io_ver[MAX_IO_APIC];
488 static	uchar_t	apic_io_vectbase[MAX_IO_APIC];
489 static	uchar_t	apic_io_vectend[MAX_IO_APIC];
490 volatile int32_t *apicioadr[MAX_IO_APIC];
491 
492 /*
493  * First available slot to be used as IRQ index into the apic_irq_table
494  * for those interrupts (like MSI/X) that don't have a physical IRQ.
495  */
496 int apic_first_avail_irq  = APIC_FIRST_FREE_IRQ;
497 
498 /*
499  * apic_ioapic_lock protects the ioapics (reg select), the status, temp_bound
500  * and bound elements of cpus_info and the temp_cpu element of irq_struct
501  */
502 lock_t	apic_ioapic_lock;
503 
504 /*
505  * apic_ioapic_reprogram_lock prevents a CPU from exiting
506  * apic_intr_exit before IOAPIC reprogramming information
507  * is collected.
508  */
509 static	lock_t	apic_ioapic_reprogram_lock;
510 static	int	apic_io_max = 0;	/* no. of i/o apics enabled */
511 
512 static	struct apic_io_intr *apic_io_intrp = 0;
513 static	struct apic_bus	*apic_busp;
514 
515 uchar_t	apic_vector_to_irq[APIC_MAX_VECTOR+1];
516 static	uchar_t	apic_resv_vector[MAXIPL+1];
517 
518 static	char	apic_level_intr[APIC_MAX_VECTOR+1];
519 static	int	apic_error = 0;
520 /* values which apic_error can take. Not catastrophic, but may help debug */
521 #define	APIC_ERR_BOOT_EOI		0x1
522 #define	APIC_ERR_GET_IPIVECT_FAIL	0x2
523 #define	APIC_ERR_INVALID_INDEX		0x4
524 #define	APIC_ERR_MARK_VECTOR_FAIL	0x8
525 #define	APIC_ERR_APIC_ERROR		0x40000000
526 #define	APIC_ERR_NMI			0x80000000
527 
528 static	int	apic_cmos_ssb_set = 0;
529 
530 static	uint32_t	eisa_level_intr_mask = 0;
531 	/* At least MSB will be set if EISA bus */
532 
533 static	int	apic_pci_bus_total = 0;
534 static	uchar_t	apic_single_pci_busid = 0;
535 
536 
537 /*
538  * airq_mutex protects additions to the apic_irq_table - the first
539  * pointer and any airq_nexts off of that one. It also protects
540  * apic_max_device_irq & apic_min_device_irq. It also guarantees
541  * that share_id is unique as new ids are generated only when new
542  * irq_t structs are linked in. Once linked in the structs are never
543  * deleted. temp_cpu & mps_intr_index field indicate if it is programmed
544  * or allocated. Note that there is a slight gap between allocating in
545  * apic_introp_xlate and programming in addspl.
546  */
547 kmutex_t	airq_mutex;
548 apic_irq_t	*apic_irq_table[APIC_MAX_VECTOR+1];
549 int		apic_max_device_irq = 0;
550 int		apic_min_device_irq = APIC_MAX_VECTOR;
551 
552 /* use to make sure only one cpu handles the nmi */
553 static	lock_t	apic_nmi_lock;
554 /* use to make sure only one cpu handles the error interrupt */
555 static	lock_t	apic_error_lock;
556 
557 /*
558  * Following declarations are for revectoring; used when ISRs at different
559  * IPLs share an irq.
560  */
561 static	lock_t	apic_revector_lock;
562 static	int	apic_revector_pending = 0;
563 static	uchar_t	*apic_oldvec_to_newvec;
564 static	uchar_t	*apic_newvec_to_oldvec;
565 
566 /* Ensures that the IOAPIC-reprogramming timeout is not reentrant */
567 static	kmutex_t	apic_reprogram_timeout_mutex;
568 
569 static	struct	ioapic_reprogram_data {
570 	int		valid;	 /* This entry is valid */
571 	int		bindcpu; /* The CPU to which the int will be bound */
572 	unsigned	timeouts; /* # times the reprogram timeout was called */
573 } apic_reprogram_info[APIC_MAX_VECTOR+1];
574 /*
575  * APIC_MAX_VECTOR + 1 is the maximum # of IRQs as well. apic_reprogram_info
576  * is indexed by IRQ number, NOT by vector number.
577  */
578 
579 
580 /*
581  * The following added to identify a software poweroff method if available.
582  */
583 
584 static struct {
585 	int	poweroff_method;
586 	char	oem_id[APIC_MPS_OEM_ID_LEN + 1];	/* MAX + 1 for NULL */
587 	char	prod_id[APIC_MPS_PROD_ID_LEN + 1];	/* MAX + 1 for NULL */
588 } apic_mps_ids[] = {
589 	{ APIC_POWEROFF_VIA_RTC,	"INTEL",	"ALDER" },   /* 4300 */
590 	{ APIC_POWEROFF_VIA_RTC,	"NCR",		"AMC" },    /* 4300 */
591 	{ APIC_POWEROFF_VIA_ASPEN_BMC,	"INTEL",	"A450NX" },  /* 4400? */
592 	{ APIC_POWEROFF_VIA_ASPEN_BMC,	"INTEL",	"AD450NX" }, /* 4400 */
593 	{ APIC_POWEROFF_VIA_ASPEN_BMC,	"INTEL",	"AC450NX" }, /* 4400R */
594 	{ APIC_POWEROFF_VIA_SITKA_BMC,	"INTEL",	"S450NX" },  /* S50  */
595 	{ APIC_POWEROFF_VIA_SITKA_BMC,	"INTEL",	"SC450NX" }  /* S50? */
596 };
597 
598 int	apic_poweroff_method = APIC_POWEROFF_NONE;
599 
600 static	struct {
601 	uchar_t	cntl;
602 	uchar_t	data;
603 } aspen_bmc[] = {
604 	{ CC_SMS_WR_START,	0x18 },		/* NetFn/LUN */
605 	{ CC_SMS_WR_NEXT,	0x24 },		/* Cmd SET_WATCHDOG_TIMER */
606 	{ CC_SMS_WR_NEXT,	0x84 },		/* DataByte 1: SMS/OS no log */
607 	{ CC_SMS_WR_NEXT,	0x2 },		/* DataByte 2: Power Down */
608 	{ CC_SMS_WR_NEXT,	0x0 },		/* DataByte 3: no pre-timeout */
609 	{ CC_SMS_WR_NEXT,	0x0 },		/* DataByte 4: timer expir. */
610 	{ CC_SMS_WR_NEXT,	0xa },		/* DataByte 5: init countdown */
611 	{ CC_SMS_WR_END,	0x0 },		/* DataByte 6: init countdown */
612 
613 	{ CC_SMS_WR_START,	0x18 },		/* NetFn/LUN */
614 	{ CC_SMS_WR_END,	0x22 }		/* Cmd RESET_WATCHDOG_TIMER */
615 };
616 
617 static	struct {
618 	int	port;
619 	uchar_t	data;
620 } sitka_bmc[] = {
621 	{ SMS_COMMAND_REGISTER,	SMS_WRITE_START },
622 	{ SMS_DATA_REGISTER,	0x18 },		/* NetFn/LUN */
623 	{ SMS_DATA_REGISTER,	0x24 },		/* Cmd SET_WATCHDOG_TIMER */
624 	{ SMS_DATA_REGISTER,	0x84 },		/* DataByte 1: SMS/OS no log */
625 	{ SMS_DATA_REGISTER,	0x2 },		/* DataByte 2: Power Down */
626 	{ SMS_DATA_REGISTER,	0x0 },		/* DataByte 3: no pre-timeout */
627 	{ SMS_DATA_REGISTER,	0x0 },		/* DataByte 4: timer expir. */
628 	{ SMS_DATA_REGISTER,	0xa },		/* DataByte 5: init countdown */
629 	{ SMS_COMMAND_REGISTER,	SMS_WRITE_END },
630 	{ SMS_DATA_REGISTER,	0x0 },		/* DataByte 6: init countdown */
631 
632 	{ SMS_COMMAND_REGISTER,	SMS_WRITE_START },
633 	{ SMS_DATA_REGISTER,	0x18 },		/* NetFn/LUN */
634 	{ SMS_COMMAND_REGISTER,	SMS_WRITE_END },
635 	{ SMS_DATA_REGISTER,	0x22 }		/* Cmd RESET_WATCHDOG_TIMER */
636 };
637 
638 
639 /* Patchable global variables. */
640 int		apic_kmdb_on_nmi = 0;		/* 0 - no, 1 - yes enter kmdb */
641 int		apic_debug_mps_id = 0;		/* 1 - print MPS ID strings */
642 
643 /*
644  * ACPI definitions
645  */
646 /* _PIC method arguments */
647 #define	ACPI_PIC_MODE	0
648 #define	ACPI_APIC_MODE	1
649 
650 /* APIC error flags we care about */
651 #define	APIC_SEND_CS_ERROR	0x01
652 #define	APIC_RECV_CS_ERROR	0x02
653 #define	APIC_CS_ERRORS		(APIC_SEND_CS_ERROR|APIC_RECV_CS_ERROR)
654 
655 /*
656  * ACPI variables
657  */
658 /* 1 = acpi is enabled & working, 0 = acpi is not enabled or not there */
659 static	int apic_enable_acpi = 0;
660 
661 /* ACPI Multiple APIC Description Table ptr */
662 static	MULTIPLE_APIC_TABLE *acpi_mapic_dtp = NULL;
663 
664 /* ACPI Interrupt Source Override Structure ptr */
665 static	MADT_INTERRUPT_OVERRIDE *acpi_isop = NULL;
666 static	int acpi_iso_cnt = 0;
667 
668 /* ACPI Non-maskable Interrupt Sources ptr */
669 static	MADT_NMI_SOURCE *acpi_nmi_sp = NULL;
670 static	int acpi_nmi_scnt = 0;
671 static	MADT_LOCAL_APIC_NMI *acpi_nmi_cp = NULL;
672 static	int acpi_nmi_ccnt = 0;
673 
674 /*
675  * extern declarations
676  */
677 extern	int	intr_clear(void);
678 extern	void	intr_restore(uint_t);
679 #if defined(__amd64)
680 extern	int	intpri_use_cr8;
681 #endif	/* __amd64 */
682 
683 extern int	apic_pci_msi_enable_vector(dev_info_t *, int, int,
684 		    int, int, int);
685 extern apic_irq_t *apic_find_irq(dev_info_t *, struct intrspec *, int);
686 
687 /*
688  *	This is the loadable module wrapper
689  */
690 
691 int
692 _init(void)
693 {
694 	if (apic_coarse_hrtime)
695 		apic_ops.psm_gethrtime = &apic_gettime;
696 	return (psm_mod_init(&apic_hdlp, &apic_psm_info));
697 }
698 
699 int
700 _fini(void)
701 {
702 	return (psm_mod_fini(&apic_hdlp, &apic_psm_info));
703 }
704 
705 int
706 _info(struct modinfo *modinfop)
707 {
708 	return (psm_mod_info(&apic_hdlp, &apic_psm_info, modinfop));
709 }
710 
711 /*
712  * Auto-configuration routines
713  */
714 
715 /*
716  * Look at MPSpec 1.4 (Intel Order # 242016-005) for details of what we do here
717  * May work with 1.1 - but not guaranteed.
718  * According to the MP Spec, the MP floating pointer structure
719  * will be searched in the order described below:
720  * 1. In the first kilobyte of Extended BIOS Data Area (EBDA)
721  * 2. Within the last kilobyte of system base memory
722  * 3. In the BIOS ROM address space between 0F0000h and 0FFFFh
723  * Once we find the right signature with proper checksum, we call
724  * either handle_defconf or parse_mpct to get all info necessary for
725  * subsequent operations.
726  */
727 static int
728 apic_probe()
729 {
730 	uint32_t mpct_addr, ebda_start = 0, base_mem_end;
731 	caddr_t	biosdatap;
732 	caddr_t	mpct;
733 	caddr_t	fptr;
734 	int	i, mpct_size, mapsize, retval = PSM_FAILURE;
735 	ushort_t	ebda_seg, base_mem_size;
736 	struct	apic_mpfps_hdr	*fpsp;
737 	struct	apic_mp_cnf_hdr	*hdrp;
738 	int bypass_cpu_and_ioapics_in_mptables;
739 	int acpi_user_options;
740 
741 	if (apic_forceload < 0)
742 		return (retval);
743 
744 	/* Allow override for MADT-only mode */
745 	acpi_user_options = ddi_prop_get_int(DDI_DEV_T_ANY, ddi_root_node(), 0,
746 	    "acpi-user-options", 0);
747 	apic_use_acpi_madt_only = ((acpi_user_options & ACPI_OUSER_MADT) != 0);
748 
749 	/* Allow apic_use_acpi to override MADT-only mode */
750 	if (!apic_use_acpi)
751 		apic_use_acpi_madt_only = 0;
752 
753 	retval = acpi_probe();
754 
755 	/*
756 	 * mapin the bios data area 40:0
757 	 * 40:13h - two-byte location reports the base memory size
758 	 * 40:0Eh - two-byte location for the exact starting address of
759 	 *	    the EBDA segment for EISA
760 	 */
761 	biosdatap = psm_map_phys(0x400, 0x20, PROT_READ);
762 	if (!biosdatap)
763 		return (retval);
764 	fpsp = (struct apic_mpfps_hdr *)NULL;
765 	mapsize = MPFPS_RAM_WIN_LEN;
766 	/*LINTED: pointer cast may result in improper alignment */
767 	ebda_seg = *((ushort_t *)(biosdatap+0xe));
768 	/* check the 1k of EBDA */
769 	if (ebda_seg) {
770 		ebda_start = ((uint32_t)ebda_seg) << 4;
771 		fptr = psm_map_phys(ebda_start, MPFPS_RAM_WIN_LEN, PROT_READ);
772 		if (fptr) {
773 			if (!(fpsp =
774 			    apic_find_fps_sig(fptr, MPFPS_RAM_WIN_LEN)))
775 				psm_unmap_phys(fptr, MPFPS_RAM_WIN_LEN);
776 		}
777 	}
778 	/* If not in EBDA, check the last k of system base memory */
779 	if (!fpsp) {
780 		/*LINTED: pointer cast may result in improper alignment */
781 		base_mem_size = *((ushort_t *)(biosdatap + 0x13));
782 
783 		if (base_mem_size > 512)
784 			base_mem_end = 639 * 1024;
785 		else
786 			base_mem_end = 511 * 1024;
787 		/* if ebda == last k of base mem, skip to check BIOS ROM */
788 		if (base_mem_end != ebda_start) {
789 
790 			fptr = psm_map_phys(base_mem_end, MPFPS_RAM_WIN_LEN,
791 			    PROT_READ);
792 
793 			if (fptr) {
794 				if (!(fpsp = apic_find_fps_sig(fptr,
795 				    MPFPS_RAM_WIN_LEN)))
796 					psm_unmap_phys(fptr, MPFPS_RAM_WIN_LEN);
797 			}
798 		}
799 	}
800 	psm_unmap_phys(biosdatap, 0x20);
801 
802 	/* If still cannot find it, check the BIOS ROM space */
803 	if (!fpsp) {
804 		mapsize = MPFPS_ROM_WIN_LEN;
805 		fptr = psm_map_phys(MPFPS_ROM_WIN_START,
806 		    MPFPS_ROM_WIN_LEN, PROT_READ);
807 		if (fptr) {
808 			if (!(fpsp =
809 			    apic_find_fps_sig(fptr, MPFPS_ROM_WIN_LEN))) {
810 				psm_unmap_phys(fptr, MPFPS_ROM_WIN_LEN);
811 				return (retval);
812 			}
813 		}
814 	}
815 
816 	if (apic_checksum((caddr_t)fpsp, fpsp->mpfps_length * 16) != 0) {
817 		psm_unmap_phys(fptr, MPFPS_ROM_WIN_LEN);
818 		return (retval);
819 	}
820 
821 	apic_spec_rev = fpsp->mpfps_spec_rev;
822 	if ((apic_spec_rev != 04) && (apic_spec_rev != 01)) {
823 		psm_unmap_phys(fptr, MPFPS_ROM_WIN_LEN);
824 		return (retval);
825 	}
826 
827 	/* check IMCR is present or not */
828 	apic_imcrp = fpsp->mpfps_featinfo2 & MPFPS_FEATINFO2_IMCRP;
829 
830 	/* check default configuration (dual CPUs) */
831 	if ((apic_defconf = fpsp->mpfps_featinfo1) != 0) {
832 		psm_unmap_phys(fptr, mapsize);
833 		return (apic_handle_defconf());
834 	}
835 
836 	/* MP Configuration Table */
837 	mpct_addr = (uint32_t)(fpsp->mpfps_mpct_paddr);
838 
839 	psm_unmap_phys(fptr, mapsize); /* unmap floating ptr struct */
840 
841 	/*
842 	 * Map in enough memory for the MP Configuration Table Header.
843 	 * Use this table to read the total length of the BIOS data and
844 	 * map in all the info
845 	 */
846 	/*LINTED: pointer cast may result in improper alignment */
847 	hdrp = (struct apic_mp_cnf_hdr *)psm_map_phys(mpct_addr,
848 	    sizeof (struct apic_mp_cnf_hdr), PROT_READ);
849 	if (!hdrp)
850 		return (retval);
851 
852 	/* check mp configuration table signature PCMP */
853 	if (hdrp->mpcnf_sig != 0x504d4350) {
854 		psm_unmap_phys((caddr_t)hdrp, sizeof (struct apic_mp_cnf_hdr));
855 		return (retval);
856 	}
857 	mpct_size = (int)hdrp->mpcnf_tbl_length;
858 
859 	apic_set_pwroff_method_from_mpcnfhdr(hdrp);
860 
861 	psm_unmap_phys((caddr_t)hdrp, sizeof (struct apic_mp_cnf_hdr));
862 
863 	if ((retval == PSM_SUCCESS) && !apic_use_acpi_madt_only) {
864 		/* This is an ACPI machine No need for further checks */
865 		return (retval);
866 	}
867 
868 	/*
869 	 * Map in the entries for this machine, ie. Processor
870 	 * Entry Tables, Bus Entry Tables, etc.
871 	 * They are in fixed order following one another
872 	 */
873 	mpct = psm_map_phys(mpct_addr, mpct_size, PROT_READ);
874 	if (!mpct)
875 		return (retval);
876 
877 	if (apic_checksum(mpct, mpct_size) != 0)
878 		goto apic_fail1;
879 
880 
881 	/*LINTED: pointer cast may result in improper alignment */
882 	hdrp = (struct apic_mp_cnf_hdr *)mpct;
883 	/*LINTED: pointer cast may result in improper alignment */
884 	apicadr = (uint32_t *)psm_map_phys((uint32_t)hdrp->mpcnf_local_apic,
885 	    APIC_LOCAL_MEMLEN, PROT_READ | PROT_WRITE);
886 	if (!apicadr)
887 		goto apic_fail1;
888 
889 	/* Parse all information in the tables */
890 	bypass_cpu_and_ioapics_in_mptables = (retval == PSM_SUCCESS);
891 	if (apic_parse_mpct(mpct, bypass_cpu_and_ioapics_in_mptables) ==
892 	    PSM_SUCCESS)
893 		return (PSM_SUCCESS);
894 
895 	for (i = 0; i < apic_io_max; i++)
896 		psm_unmap_phys((caddr_t)apicioadr[i], APIC_IO_MEMLEN);
897 	if (apic_cpus)
898 		kmem_free(apic_cpus, sizeof (*apic_cpus) * apic_nproc);
899 	if (apicadr)
900 		psm_unmap_phys((caddr_t)apicadr, APIC_LOCAL_MEMLEN);
901 apic_fail1:
902 	psm_unmap_phys(mpct, mpct_size);
903 	return (retval);
904 }
905 
906 static void
907 apic_set_pwroff_method_from_mpcnfhdr(struct apic_mp_cnf_hdr *hdrp)
908 {
909 	int	i;
910 
911 	for (i = 0; i < (sizeof (apic_mps_ids) / sizeof (apic_mps_ids[0]));
912 	    i++) {
913 		if ((strncmp(hdrp->mpcnf_oem_str, apic_mps_ids[i].oem_id,
914 		    strlen(apic_mps_ids[i].oem_id)) == 0) &&
915 		    (strncmp(hdrp->mpcnf_prod_str, apic_mps_ids[i].prod_id,
916 		    strlen(apic_mps_ids[i].prod_id)) == 0)) {
917 
918 			apic_poweroff_method = apic_mps_ids[i].poweroff_method;
919 			break;
920 		}
921 	}
922 
923 	if (apic_debug_mps_id != 0) {
924 		cmn_err(CE_CONT, "pcplusmp: MPS OEM ID = '%c%c%c%c%c%c%c%c'"
925 		    "Product ID = '%c%c%c%c%c%c%c%c%c%c%c%c'\n",
926 		    hdrp->mpcnf_oem_str[0],
927 		    hdrp->mpcnf_oem_str[1],
928 		    hdrp->mpcnf_oem_str[2],
929 		    hdrp->mpcnf_oem_str[3],
930 		    hdrp->mpcnf_oem_str[4],
931 		    hdrp->mpcnf_oem_str[5],
932 		    hdrp->mpcnf_oem_str[6],
933 		    hdrp->mpcnf_oem_str[7],
934 		    hdrp->mpcnf_prod_str[0],
935 		    hdrp->mpcnf_prod_str[1],
936 		    hdrp->mpcnf_prod_str[2],
937 		    hdrp->mpcnf_prod_str[3],
938 		    hdrp->mpcnf_prod_str[4],
939 		    hdrp->mpcnf_prod_str[5],
940 		    hdrp->mpcnf_prod_str[6],
941 		    hdrp->mpcnf_prod_str[7],
942 		    hdrp->mpcnf_prod_str[8],
943 		    hdrp->mpcnf_prod_str[9],
944 		    hdrp->mpcnf_prod_str[10],
945 		    hdrp->mpcnf_prod_str[11]);
946 	}
947 }
948 
949 static int
950 acpi_probe(void)
951 {
952 	int			i, id, intmax, ver, index, rv;
953 	int			acpi_verboseflags = 0;
954 	int			madt_seen, madt_size;
955 	APIC_HEADER		*ap;
956 	MADT_PROCESSOR_APIC	*mpa;
957 	MADT_IO_APIC		*mia;
958 	MADT_IO_SAPIC		*misa;
959 	MADT_INTERRUPT_OVERRIDE	*mio;
960 	MADT_NMI_SOURCE		*mns;
961 	MADT_INTERRUPT_SOURCE	*mis;
962 	MADT_LOCAL_APIC_NMI	*mlan;
963 	MADT_ADDRESS_OVERRIDE	*mao;
964 	ACPI_OBJECT_LIST 	arglist;
965 	ACPI_OBJECT		arg;
966 	int			sci;
967 	iflag_t			sci_flags;
968 	volatile int32_t	*ioapic;
969 	char			local_ids[NCPU];
970 	char			proc_ids[NCPU];
971 	uchar_t			hid;
972 
973 	if (!apic_use_acpi)
974 		return (PSM_FAILURE);
975 
976 	if (AcpiGetFirmwareTable(APIC_SIG, 1, ACPI_LOGICAL_ADDRESSING,
977 	    (ACPI_TABLE_HEADER **) &acpi_mapic_dtp) != AE_OK)
978 		return (PSM_FAILURE);
979 
980 	apicadr = (uint32_t *)psm_map_phys(
981 	    (uint32_t)acpi_mapic_dtp->LocalApicAddress,
982 	    APIC_LOCAL_MEMLEN, PROT_READ | PROT_WRITE);
983 	if (!apicadr)
984 		return (PSM_FAILURE);
985 
986 	id = apicadr[APIC_LID_REG];
987 	local_ids[0] = (uchar_t)(((uint_t)id) >> 24);
988 	apic_nproc = index = 1;
989 	apic_io_max = 0;
990 
991 	ap = (APIC_HEADER *) (acpi_mapic_dtp + 1);
992 	madt_size = acpi_mapic_dtp->Length;
993 	madt_seen = sizeof (*acpi_mapic_dtp);
994 
995 	while (madt_seen < madt_size) {
996 		switch (ap->Type) {
997 		case APIC_PROCESSOR:
998 			mpa = (MADT_PROCESSOR_APIC *) ap;
999 			if (mpa->ProcessorEnabled) {
1000 				if (mpa->LocalApicId == local_ids[0])
1001 					proc_ids[0] = mpa->ProcessorId;
1002 				else if (apic_nproc < NCPU) {
1003 					local_ids[index] = mpa->LocalApicId;
1004 					proc_ids[index] = mpa->ProcessorId;
1005 					index++;
1006 					apic_nproc++;
1007 				} else
1008 					cmn_err(CE_WARN, "pcplusmp: exceeded "
1009 					    "maximum no. of CPUs (= %d)", NCPU);
1010 			}
1011 			break;
1012 
1013 		case APIC_IO:
1014 			mia = (MADT_IO_APIC *) ap;
1015 			if (apic_io_max < MAX_IO_APIC) {
1016 				apic_io_id[apic_io_max] = mia->IoApicId;
1017 				apic_io_vectbase[apic_io_max] =
1018 				    mia->Interrupt;
1019 				ioapic = apicioadr[apic_io_max] =
1020 				    (int32_t *)psm_map_phys(
1021 				    (uint32_t)mia->Address,
1022 				    APIC_IO_MEMLEN, PROT_READ | PROT_WRITE);
1023 				if (!ioapic)
1024 					goto cleanup;
1025 				apic_io_max++;
1026 			}
1027 			break;
1028 
1029 		case APIC_XRUPT_OVERRIDE:
1030 			mio = (MADT_INTERRUPT_OVERRIDE *) ap;
1031 			if (acpi_isop == NULL)
1032 				acpi_isop = mio;
1033 			acpi_iso_cnt++;
1034 			break;
1035 
1036 		case APIC_NMI:
1037 			/* UNIMPLEMENTED */
1038 			mns = (MADT_NMI_SOURCE *) ap;
1039 			if (acpi_nmi_sp == NULL)
1040 				acpi_nmi_sp = mns;
1041 			acpi_nmi_scnt++;
1042 
1043 			cmn_err(CE_NOTE, "!apic: nmi source: %d %d %d\n",
1044 				mns->Interrupt, mns->Polarity,
1045 				mns->TriggerMode);
1046 			break;
1047 
1048 		case APIC_LOCAL_NMI:
1049 			/* UNIMPLEMENTED */
1050 			mlan = (MADT_LOCAL_APIC_NMI *) ap;
1051 			if (acpi_nmi_cp == NULL)
1052 				acpi_nmi_cp = mlan;
1053 			acpi_nmi_ccnt++;
1054 
1055 			cmn_err(CE_NOTE, "!apic: local nmi: %d %d %d %d\n",
1056 				mlan->ProcessorId, mlan->Polarity,
1057 				mlan->TriggerMode, mlan->Lint);
1058 			break;
1059 
1060 		case APIC_ADDRESS_OVERRIDE:
1061 			/* UNIMPLEMENTED */
1062 			mao = (MADT_ADDRESS_OVERRIDE *) ap;
1063 			cmn_err(CE_NOTE, "!apic: address override: %lx\n",
1064 				(long)mao->Address);
1065 			break;
1066 
1067 		case APIC_IO_SAPIC:
1068 			/* UNIMPLEMENTED */
1069 			misa = (MADT_IO_SAPIC *) ap;
1070 
1071 			cmn_err(CE_NOTE, "!apic: io sapic: %d %d %lx\n",
1072 				misa->IoSapicId, misa->InterruptBase,
1073 				(long)misa->Address);
1074 			break;
1075 
1076 		case APIC_XRUPT_SOURCE:
1077 			/* UNIMPLEMENTED */
1078 			mis = (MADT_INTERRUPT_SOURCE *) ap;
1079 
1080 			cmn_err(CE_NOTE,
1081 				"!apic: irq source: %d %d %d %d %d %d %d\n",
1082 				mis->ProcessorId, mis->ProcessorEid,
1083 				mis->Interrupt, mis->Polarity,
1084 				mis->TriggerMode, mis->InterruptType,
1085 				mis->IoSapicVector);
1086 			break;
1087 		case APIC_RESERVED:
1088 		default:
1089 			goto cleanup;
1090 		}
1091 
1092 		/* advance to next entry */
1093 		madt_seen += ap->Length;
1094 		ap = (APIC_HEADER *)(((char *)ap) + ap->Length);
1095 	}
1096 
1097 	if ((apic_cpus = kmem_zalloc(sizeof (*apic_cpus) * apic_nproc,
1098 	    KM_NOSLEEP)) == NULL)
1099 		goto cleanup;
1100 
1101 	apic_cpumask = (1 << apic_nproc) - 1;
1102 
1103 	/*
1104 	 * ACPI doesn't provide the local apic ver, get it directly from the
1105 	 * local apic
1106 	 */
1107 	ver = apicadr[APIC_VERS_REG];
1108 	for (i = 0; i < apic_nproc; i++) {
1109 		apic_cpus[i].aci_local_id = local_ids[i];
1110 		apic_cpus[i].aci_local_ver = (uchar_t)(ver & 0xFF);
1111 	}
1112 	for (i = 0; i < apic_io_max; i++) {
1113 		ioapic = apicioadr[i];
1114 
1115 		/*
1116 		 * need to check Sitka on the following acpi problem
1117 		 * On the Sitka, the ioapic's apic_id field isn't reporting
1118 		 * the actual io apic id. We have reported this problem
1119 		 * to Intel. Until they fix the problem, we will get the
1120 		 * actual id directly from the ioapic.
1121 		 */
1122 		ioapic[APIC_IO_REG] = APIC_ID_CMD;
1123 		id = ioapic[APIC_IO_DATA];
1124 		hid = (uchar_t)(((uint_t)id) >> 24);
1125 
1126 		if (hid != apic_io_id[i]) {
1127 			if (apic_io_id[i] == 0)
1128 				apic_io_id[i] = hid;
1129 			else { /* set ioapic id to whatever reported by ACPI */
1130 				id = ((int32_t)apic_io_id[i]) << 24;
1131 				ioapic[APIC_IO_REG] = APIC_ID_CMD;
1132 				ioapic[APIC_IO_DATA] = id;
1133 			}
1134 		}
1135 		ioapic[APIC_IO_REG] = APIC_VERS_CMD;
1136 		ver = ioapic[APIC_IO_DATA];
1137 		apic_io_ver[i] = (uchar_t)(ver & 0xff);
1138 		intmax = (ver >> 16) & 0xff;
1139 		apic_io_vectend[i] = apic_io_vectbase[i] + intmax;
1140 		if (apic_first_avail_irq <= apic_io_vectend[i])
1141 			apic_first_avail_irq = apic_io_vectend[i] + 1;
1142 	}
1143 
1144 
1145 	/*
1146 	 * Process SCI configuration here
1147 	 * An error may be returned here if
1148 	 * acpi-user-options specifies legacy mode
1149 	 * (no SCI, no ACPI mode)
1150 	 */
1151 	if (acpica_get_sci(&sci, &sci_flags) != AE_OK)
1152 		sci = -1;
1153 
1154 	/*
1155 	 * Now call acpi_init() to generate namespaces
1156 	 * If this fails, we don't attempt to use ACPI
1157 	 * even if we were able to get a MADT above
1158 	 */
1159 	if (acpica_init() != AE_OK)
1160 		goto cleanup;
1161 
1162 	/*
1163 	 * Squirrel away the SCI and flags for later on
1164 	 * in apic_picinit() when we're ready
1165 	 */
1166 	apic_sci_vect = sci;
1167 	apic_sci_flags = sci_flags;
1168 
1169 	if (apic_verbose & APIC_VERBOSE_IRQ_FLAG)
1170 		acpi_verboseflags |= PSM_VERBOSE_IRQ_FLAG;
1171 
1172 	if (apic_verbose & APIC_VERBOSE_POWEROFF_FLAG)
1173 		acpi_verboseflags |= PSM_VERBOSE_POWEROFF_FLAG;
1174 
1175 	if (apic_verbose & APIC_VERBOSE_POWEROFF_PAUSE_FLAG)
1176 		acpi_verboseflags |= PSM_VERBOSE_POWEROFF_PAUSE_FLAG;
1177 
1178 	if (acpi_psm_init(apic_psm_info.p_mach_idstring, acpi_verboseflags) ==
1179 	    ACPI_PSM_FAILURE)
1180 		goto cleanup;
1181 
1182 	/* Enable ACPI APIC interrupt routing */
1183 	arglist.Count = 1;
1184 	arglist.Pointer = &arg;
1185 	arg.Type = ACPI_TYPE_INTEGER;
1186 	arg.Integer.Value = ACPI_APIC_MODE;	/* 1 */
1187 	rv = AcpiEvaluateObject(NULL, "\\_PIC", &arglist, NULL);
1188 	if (rv == AE_OK) {
1189 		build_reserved_irqlist((uchar_t *)apic_reserved_irqlist);
1190 		apic_enable_acpi = 1;
1191 		if (apic_use_acpi_madt_only) {
1192 			cmn_err(CE_CONT,
1193 			    "?Using ACPI for CPU/IOAPIC information ONLY\n");
1194 		}
1195 		return (PSM_SUCCESS);
1196 	}
1197 	/* if setting APIC mode failed above, we fall through to cleanup */
1198 
1199 cleanup:
1200 	if (apicadr != NULL) {
1201 		psm_unmap_phys((caddr_t)apicadr, APIC_LOCAL_MEMLEN);
1202 		apicadr = NULL;
1203 	}
1204 	apic_nproc = 0;
1205 	for (i = 0; i < apic_io_max; i++) {
1206 		psm_unmap_phys((caddr_t)apicioadr[i], APIC_IO_MEMLEN);
1207 		apicioadr[i] = NULL;
1208 	}
1209 	apic_io_max = 0;
1210 	acpi_isop = NULL;
1211 	acpi_iso_cnt = 0;
1212 	acpi_nmi_sp = NULL;
1213 	acpi_nmi_scnt = 0;
1214 	acpi_nmi_cp = NULL;
1215 	acpi_nmi_ccnt = 0;
1216 	return (PSM_FAILURE);
1217 }
1218 
1219 /*
1220  * Handle default configuration. Fill in reqd global variables & tables
1221  * Fill all details as MP table does not give any more info
1222  */
1223 static int
1224 apic_handle_defconf()
1225 {
1226 	uint_t	lid;
1227 
1228 	/*LINTED: pointer cast may result in improper alignment */
1229 	apicioadr[0] = (int32_t *)psm_map_phys(APIC_IO_ADDR,
1230 	    APIC_IO_MEMLEN, PROT_READ | PROT_WRITE);
1231 	/*LINTED: pointer cast may result in improper alignment */
1232 	apicadr = (uint32_t *)psm_map_phys(APIC_LOCAL_ADDR,
1233 	    APIC_LOCAL_MEMLEN, PROT_READ | PROT_WRITE);
1234 	apic_cpus = (apic_cpus_info_t *)
1235 	    kmem_zalloc(sizeof (*apic_cpus) * 2, KM_NOSLEEP);
1236 	if ((!apicadr) || (!apicioadr[0]) || (!apic_cpus))
1237 		goto apic_handle_defconf_fail;
1238 	apic_cpumask = 3;
1239 	apic_nproc = 2;
1240 	lid = apicadr[APIC_LID_REG];
1241 	apic_cpus[0].aci_local_id = (uchar_t)(lid >> APIC_ID_BIT_OFFSET);
1242 	/*
1243 	 * According to the PC+MP spec 1.1, the local ids
1244 	 * for the default configuration has to be 0 or 1
1245 	 */
1246 	if (apic_cpus[0].aci_local_id == 1)
1247 		apic_cpus[1].aci_local_id = 0;
1248 	else if (apic_cpus[0].aci_local_id == 0)
1249 		apic_cpus[1].aci_local_id = 1;
1250 	else
1251 		goto apic_handle_defconf_fail;
1252 
1253 	apic_io_id[0] = 2;
1254 	apic_io_max = 1;
1255 	if (apic_defconf >= 5) {
1256 		apic_cpus[0].aci_local_ver = APIC_INTEGRATED_VERS;
1257 		apic_cpus[1].aci_local_ver = APIC_INTEGRATED_VERS;
1258 		apic_io_ver[0] = APIC_INTEGRATED_VERS;
1259 	} else {
1260 		apic_cpus[0].aci_local_ver = 0;		/* 82489 DX */
1261 		apic_cpus[1].aci_local_ver = 0;
1262 		apic_io_ver[0] = 0;
1263 	}
1264 	if (apic_defconf == 2 || apic_defconf == 3 || apic_defconf == 6)
1265 		eisa_level_intr_mask = (inb(EISA_LEVEL_CNTL + 1) << 8) |
1266 		    inb(EISA_LEVEL_CNTL) | ((uint_t)INT32_MAX + 1);
1267 	return (PSM_SUCCESS);
1268 
1269 apic_handle_defconf_fail:
1270 	if (apic_cpus)
1271 		kmem_free(apic_cpus, sizeof (*apic_cpus) * 2);
1272 	if (apicadr)
1273 		psm_unmap_phys((caddr_t)apicadr, APIC_LOCAL_MEMLEN);
1274 	if (apicioadr[0])
1275 		psm_unmap_phys((caddr_t)apicioadr[0], APIC_IO_MEMLEN);
1276 	return (PSM_FAILURE);
1277 }
1278 
1279 /* Parse the entries in MP configuration table and collect info that we need */
1280 static int
1281 apic_parse_mpct(caddr_t mpct, int bypass_cpus_and_ioapics)
1282 {
1283 	struct	apic_procent	*procp;
1284 	struct	apic_bus	*busp;
1285 	struct	apic_io_entry	*ioapicp;
1286 	struct	apic_io_intr	*intrp;
1287 	volatile int32_t	*ioapic;
1288 	uint_t	lid;
1289 	int	id;
1290 	uchar_t hid;
1291 
1292 	/*LINTED: pointer cast may result in improper alignment */
1293 	procp = (struct apic_procent *)(mpct + sizeof (struct apic_mp_cnf_hdr));
1294 
1295 	/* No need to count cpu entries if we won't use them */
1296 	if (!bypass_cpus_and_ioapics) {
1297 
1298 		/* Find max # of CPUS and allocate structure accordingly */
1299 		apic_nproc = 0;
1300 		while (procp->proc_entry == APIC_CPU_ENTRY) {
1301 			if (procp->proc_cpuflags & CPUFLAGS_EN) {
1302 				apic_nproc++;
1303 			}
1304 			procp++;
1305 		}
1306 		if (apic_nproc > NCPU)
1307 			cmn_err(CE_WARN, "pcplusmp: exceeded "
1308 			    "maximum no. of CPUs (= %d)", NCPU);
1309 		if (!apic_nproc || !(apic_cpus = (apic_cpus_info_t *)
1310 		    kmem_zalloc(sizeof (*apic_cpus)*apic_nproc, KM_NOSLEEP)))
1311 			return (PSM_FAILURE);
1312 	}
1313 
1314 	/*LINTED: pointer cast may result in improper alignment */
1315 	procp = (struct apic_procent *)(mpct + sizeof (struct apic_mp_cnf_hdr));
1316 
1317 	/*
1318 	 * start with index 1 as 0 needs to be filled in with Boot CPU, but
1319 	 * if we're bypassing this information, it has already been filled
1320 	 * in by acpi_probe(), so don't overwrite it.
1321 	 */
1322 	if (!bypass_cpus_and_ioapics)
1323 		apic_nproc = 1;
1324 
1325 	while (procp->proc_entry == APIC_CPU_ENTRY) {
1326 		/* check whether the cpu exists or not */
1327 		if (!bypass_cpus_and_ioapics &&
1328 		    procp->proc_cpuflags & CPUFLAGS_EN) {
1329 			if (procp->proc_cpuflags & CPUFLAGS_BP) { /* Boot CPU */
1330 				lid = apicadr[APIC_LID_REG];
1331 				apic_cpus[0].aci_local_id = procp->proc_apicid;
1332 				if (apic_cpus[0].aci_local_id !=
1333 				    (uchar_t)(lid >> APIC_ID_BIT_OFFSET)) {
1334 					return (PSM_FAILURE);
1335 				}
1336 				apic_cpus[0].aci_local_ver =
1337 				    procp->proc_version;
1338 			} else {
1339 
1340 				apic_cpus[apic_nproc].aci_local_id =
1341 				    procp->proc_apicid;
1342 				apic_cpus[apic_nproc].aci_local_ver =
1343 				    procp->proc_version;
1344 				apic_nproc++;
1345 
1346 			}
1347 		}
1348 		procp++;
1349 	}
1350 
1351 	if (!bypass_cpus_and_ioapics) {
1352 		/* convert the number of processors into a cpumask */
1353 		apic_cpumask = (1 << apic_nproc) - 1;
1354 	}
1355 
1356 	/*
1357 	 * Save start of bus entries for later use.
1358 	 * Get EISA level cntrl if EISA bus is present.
1359 	 * Also get the CPI bus id for single CPI bus case
1360 	 */
1361 	apic_busp = busp = (struct apic_bus *)procp;
1362 	while (busp->bus_entry == APIC_BUS_ENTRY) {
1363 		lid = apic_find_bus_type((char *)&busp->bus_str1);
1364 		if (lid	== BUS_EISA) {
1365 			eisa_level_intr_mask = (inb(EISA_LEVEL_CNTL + 1) << 8) |
1366 			    inb(EISA_LEVEL_CNTL) | ((uint_t)INT32_MAX + 1);
1367 		} else if (lid == BUS_PCI) {
1368 			/*
1369 			 * apic_single_pci_busid will be used only if
1370 			 * apic_pic_bus_total is equal to 1
1371 			 */
1372 			apic_pci_bus_total++;
1373 			apic_single_pci_busid = busp->bus_id;
1374 		}
1375 		busp++;
1376 	}
1377 
1378 	ioapicp = (struct apic_io_entry *)busp;
1379 
1380 	if (!bypass_cpus_and_ioapics)
1381 		apic_io_max = 0;
1382 	do {
1383 		if (!bypass_cpus_and_ioapics && apic_io_max < MAX_IO_APIC) {
1384 			if (ioapicp->io_flags & IOAPIC_FLAGS_EN) {
1385 				apic_io_id[apic_io_max] = ioapicp->io_apicid;
1386 				apic_io_ver[apic_io_max] = ioapicp->io_version;
1387 		/*LINTED: pointer cast may result in improper alignment */
1388 				apicioadr[apic_io_max] =
1389 				    (int32_t *)psm_map_phys(
1390 				    (uint32_t)ioapicp->io_apic_addr,
1391 				    APIC_IO_MEMLEN, PROT_READ | PROT_WRITE);
1392 
1393 				if (!apicioadr[apic_io_max])
1394 					return (PSM_FAILURE);
1395 
1396 				ioapic = apicioadr[apic_io_max];
1397 				ioapic[APIC_IO_REG] = APIC_ID_CMD;
1398 				id = ioapic[APIC_IO_DATA];
1399 				hid = (uchar_t)(((uint_t)id) >> 24);
1400 
1401 				if (hid != apic_io_id[apic_io_max]) {
1402 					if (apic_io_id[apic_io_max] == 0)
1403 						apic_io_id[apic_io_max] = hid;
1404 					else {
1405 						/*
1406 						 * set ioapic id to whatever
1407 						 * reported by MPS
1408 						 *
1409 						 * may not need to set index
1410 						 * again ???
1411 						 * take it out and try
1412 						 */
1413 
1414 						id = ((int32_t)
1415 						    apic_io_id[apic_io_max]) <<
1416 						    24;
1417 
1418 						ioapic[APIC_IO_REG] =
1419 						    APIC_ID_CMD;
1420 
1421 						ioapic[APIC_IO_DATA] = id;
1422 
1423 					}
1424 				}
1425 				apic_io_max++;
1426 			}
1427 		}
1428 		ioapicp++;
1429 	} while (ioapicp->io_entry == APIC_IO_ENTRY);
1430 
1431 	apic_io_intrp = (struct apic_io_intr *)ioapicp;
1432 
1433 	intrp = apic_io_intrp;
1434 	while (intrp->intr_entry == APIC_IO_INTR_ENTRY) {
1435 		if ((intrp->intr_irq > APIC_MAX_ISA_IRQ) ||
1436 		    (apic_find_bus(intrp->intr_busid) == BUS_PCI)) {
1437 			apic_irq_translate = 1;
1438 			break;
1439 		}
1440 		intrp++;
1441 	}
1442 
1443 	return (PSM_SUCCESS);
1444 }
1445 
1446 static struct apic_mpfps_hdr *
1447 apic_find_fps_sig(caddr_t cptr, int len)
1448 {
1449 	int	i;
1450 
1451 	/* Look for the pattern "_MP_" */
1452 	for (i = 0; i < len; i += 16) {
1453 		if ((*(cptr+i) == '_') &&
1454 		    (*(cptr+i+1) == 'M') &&
1455 		    (*(cptr+i+2) == 'P') &&
1456 		    (*(cptr+i+3) == '_'))
1457 		    /*LINTED: pointer cast may result in improper alignment */
1458 			return ((struct apic_mpfps_hdr *)(cptr + i));
1459 	}
1460 	return (NULL);
1461 }
1462 
1463 static int
1464 apic_checksum(caddr_t bptr, int len)
1465 {
1466 	int	i;
1467 	uchar_t	cksum;
1468 
1469 	cksum = 0;
1470 	for (i = 0; i < len; i++)
1471 		cksum += *bptr++;
1472 	return ((int)cksum);
1473 }
1474 
1475 
1476 /*
1477  * Initialise vector->ipl and ipl->pri arrays. level_intr and irqtable
1478  * are also set to NULL. vector->irq is set to a value which cannot map
1479  * to a real irq to show that it is free.
1480  */
1481 void
1482 apic_init()
1483 {
1484 	int	i;
1485 	int	*iptr;
1486 
1487 	int	j = 1;
1488 	apic_ipltopri[0] = APIC_VECTOR_PER_IPL; /* leave 0 for idle */
1489 	for (i = 0; i < (APIC_AVAIL_VECTOR / APIC_VECTOR_PER_IPL); i++) {
1490 		if ((i < ((APIC_AVAIL_VECTOR / APIC_VECTOR_PER_IPL) - 1)) &&
1491 		    (apic_vectortoipl[i + 1] == apic_vectortoipl[i]))
1492 			/* get to highest vector at the same ipl */
1493 			continue;
1494 		for (; j <= apic_vectortoipl[i]; j++) {
1495 			apic_ipltopri[j] = (i << APIC_IPL_SHIFT) +
1496 			    APIC_BASE_VECT;
1497 		}
1498 	}
1499 	for (; j < MAXIPL + 1; j++)
1500 		/* fill up any empty ipltopri slots */
1501 		apic_ipltopri[j] = (i << APIC_IPL_SHIFT) + APIC_BASE_VECT;
1502 
1503 	/* cpu 0 is always up */
1504 	apic_cpus[0].aci_status = APIC_CPU_ONLINE | APIC_CPU_INTR_ENABLE;
1505 
1506 	iptr = (int *)&apic_irq_table[0];
1507 	for (i = 0; i <= APIC_MAX_VECTOR; i++) {
1508 		apic_level_intr[i] = 0;
1509 		*iptr++ = NULL;
1510 		apic_vector_to_irq[i] = APIC_RESV_IRQ;
1511 		apic_reprogram_info[i].valid = 0;
1512 		apic_reprogram_info[i].bindcpu = 0;
1513 		apic_reprogram_info[i].timeouts = 0;
1514 	}
1515 
1516 	/*
1517 	 * Allocate a dummy irq table entry for the reserved entry.
1518 	 * This takes care of the race between removing an irq and
1519 	 * clock detecting a CPU in that irq during interrupt load
1520 	 * sampling.
1521 	 */
1522 	apic_irq_table[APIC_RESV_IRQ] =
1523 	    kmem_zalloc(sizeof (apic_irq_t), KM_NOSLEEP);
1524 
1525 	mutex_init(&airq_mutex, NULL, MUTEX_DEFAULT, NULL);
1526 	mutex_init(&apic_reprogram_timeout_mutex, NULL, MUTEX_DEFAULT, NULL);
1527 #if defined(__amd64)
1528 	/*
1529 	 * Make cpu-specific interrupt info point to cr8pri vector
1530 	 */
1531 	for (i = 0; i <= MAXIPL; i++)
1532 		apic_cr8pri[i] = apic_ipltopri[i] >> APIC_IPL_SHIFT;
1533 	CPU->cpu_pri_data = apic_cr8pri;
1534 	intpri_use_cr8 = 1;
1535 #endif	/* __amd64 */
1536 }
1537 
1538 /*
1539  * handler for APIC Error interrupt. Just print a warning and continue
1540  */
1541 static int
1542 apic_error_intr()
1543 {
1544 	uint_t	error0, error1, error;
1545 	uint_t	i;
1546 
1547 	/*
1548 	 * We need to write before read as per 7.4.17 of system prog manual.
1549 	 * We do both and or the results to be safe
1550 	 */
1551 	error0 = apicadr[APIC_ERROR_STATUS];
1552 	apicadr[APIC_ERROR_STATUS] = 0;
1553 	error1 = apicadr[APIC_ERROR_STATUS];
1554 	error = error0 | error1;
1555 
1556 	/*
1557 	 * Clear the APIC error status (do this on all cpus that enter here)
1558 	 * (two writes are required due to the semantics of accessing the
1559 	 * error status register.)
1560 	 */
1561 	apicadr[APIC_ERROR_STATUS] = 0;
1562 	apicadr[APIC_ERROR_STATUS] = 0;
1563 
1564 	/*
1565 	 * Prevent more than 1 CPU from handling error interrupt causing
1566 	 * double printing (interleave of characters from multiple
1567 	 * CPU's when using prom_printf)
1568 	 */
1569 	if (lock_try(&apic_error_lock) == 0)
1570 		return (error ? DDI_INTR_CLAIMED : DDI_INTR_UNCLAIMED);
1571 	if (error) {
1572 #if	DEBUG
1573 		if (apic_debug)
1574 			debug_enter("pcplusmp: APIC Error interrupt received");
1575 #endif /* DEBUG */
1576 		if (apic_panic_on_apic_error)
1577 			cmn_err(CE_PANIC,
1578 			    "APIC Error interrupt on CPU %d. Status = %x\n",
1579 			    psm_get_cpu_id(), error);
1580 		else {
1581 			if ((error & ~APIC_CS_ERRORS) == 0) {
1582 				/* cksum error only */
1583 				apic_error |= APIC_ERR_APIC_ERROR;
1584 				apic_apic_error |= error;
1585 				apic_num_apic_errors++;
1586 				apic_num_cksum_errors++;
1587 			} else {
1588 				/*
1589 				 * prom_printf is the best shot we have of
1590 				 * something which is problem free from
1591 				 * high level/NMI type of interrupts
1592 				 */
1593 				prom_printf("APIC Error interrupt on CPU %d. "
1594 				    "Status 0 = %x, Status 1 = %x\n",
1595 				    psm_get_cpu_id(), error0, error1);
1596 				apic_error |= APIC_ERR_APIC_ERROR;
1597 				apic_apic_error |= error;
1598 				apic_num_apic_errors++;
1599 				for (i = 0; i < apic_error_display_delay; i++) {
1600 					tenmicrosec();
1601 				}
1602 				/*
1603 				 * provide more delay next time limited to
1604 				 * roughly 1 clock tick time
1605 				 */
1606 				if (apic_error_display_delay < 500)
1607 					apic_error_display_delay *= 2;
1608 			}
1609 		}
1610 		lock_clear(&apic_error_lock);
1611 		return (DDI_INTR_CLAIMED);
1612 	} else {
1613 		lock_clear(&apic_error_lock);
1614 		return (DDI_INTR_UNCLAIMED);
1615 	}
1616 	/* NOTREACHED */
1617 }
1618 
1619 /*
1620  * Turn off the mask bit in the performance counter Local Vector Table entry.
1621  */
1622 static void
1623 apic_cpcovf_mask_clear(void)
1624 {
1625 	apicadr[APIC_PCINT_VECT] &= ~APIC_LVT_MASK;
1626 }
1627 
1628 static void
1629 apic_init_intr()
1630 {
1631 	processorid_t	cpun = psm_get_cpu_id();
1632 
1633 #if defined(__amd64)
1634 	setcr8((ulong_t)(APIC_MASK_ALL >> APIC_IPL_SHIFT));
1635 #else
1636 	apicadr[APIC_TASK_REG] = APIC_MASK_ALL;
1637 #endif
1638 
1639 	if (apic_flat_model)
1640 		apicadr[APIC_FORMAT_REG] = APIC_FLAT_MODEL;
1641 	else
1642 		apicadr[APIC_FORMAT_REG] = APIC_CLUSTER_MODEL;
1643 	apicadr[APIC_DEST_REG] = AV_HIGH_ORDER >> cpun;
1644 
1645 	/* need to enable APIC before unmasking NMI */
1646 	apicadr[APIC_SPUR_INT_REG] = AV_UNIT_ENABLE | APIC_SPUR_INTR;
1647 
1648 	apicadr[APIC_LOCAL_TIMER] = AV_MASK;
1649 	apicadr[APIC_INT_VECT0]	= AV_MASK;	/* local intr reg 0 */
1650 	apicadr[APIC_INT_VECT1] = AV_NMI;	/* enable NMI */
1651 
1652 	if (apic_cpus[cpun].aci_local_ver < APIC_INTEGRATED_VERS)
1653 		return;
1654 
1655 	/* Enable performance counter overflow interrupt */
1656 
1657 	if ((x86_feature & X86_MSR) != X86_MSR)
1658 		apic_enable_cpcovf_intr = 0;
1659 	if (apic_enable_cpcovf_intr) {
1660 		if (apic_cpcovf_vect == 0) {
1661 			int ipl = APIC_PCINT_IPL;
1662 			int irq = apic_get_ipivect(ipl, -1);
1663 
1664 			ASSERT(irq != -1);
1665 			apic_cpcovf_vect = apic_irq_table[irq]->airq_vector;
1666 			ASSERT(apic_cpcovf_vect);
1667 			(void) add_avintr(NULL, ipl,
1668 			    (avfunc)kcpc_hw_overflow_intr,
1669 			    "apic pcint", irq, NULL, NULL, NULL);
1670 			kcpc_hw_overflow_intr_installed = 1;
1671 			kcpc_hw_enable_cpc_intr = apic_cpcovf_mask_clear;
1672 		}
1673 		apicadr[APIC_PCINT_VECT] = apic_cpcovf_vect;
1674 	}
1675 
1676 	/* Enable error interrupt */
1677 
1678 	if (apic_enable_error_intr) {
1679 		if (apic_errvect == 0) {
1680 			int ipl = 0xf;	/* get highest priority intr */
1681 			int irq = apic_get_ipivect(ipl, -1);
1682 
1683 			ASSERT(irq != -1);
1684 			apic_errvect = apic_irq_table[irq]->airq_vector;
1685 			ASSERT(apic_errvect);
1686 			/*
1687 			 * Not PSMI compliant, but we are going to merge
1688 			 * with ON anyway
1689 			 */
1690 			(void) add_avintr((void *)NULL, ipl,
1691 			    (avfunc)apic_error_intr, "apic error intr",
1692 			    irq, NULL, NULL, NULL);
1693 		}
1694 		apicadr[APIC_ERR_VECT] = apic_errvect;
1695 		apicadr[APIC_ERROR_STATUS] = 0;
1696 		apicadr[APIC_ERROR_STATUS] = 0;
1697 	}
1698 }
1699 
1700 static void
1701 apic_disable_local_apic()
1702 {
1703 	apicadr[APIC_TASK_REG] = APIC_MASK_ALL;
1704 	apicadr[APIC_LOCAL_TIMER] = AV_MASK;
1705 	apicadr[APIC_INT_VECT0] = AV_MASK;	/* local intr reg 0 */
1706 	apicadr[APIC_INT_VECT1] = AV_MASK;	/* disable NMI */
1707 	apicadr[APIC_ERR_VECT] = AV_MASK;	/* and error interrupt */
1708 	apicadr[APIC_PCINT_VECT] = AV_MASK;	/* and perf counter intr */
1709 	apicadr[APIC_SPUR_INT_REG] = APIC_SPUR_INTR;
1710 }
1711 
1712 static void
1713 apic_picinit(void)
1714 {
1715 	int i, j;
1716 	uint_t isr;
1717 	volatile int32_t *ioapic;
1718 	apic_irq_t	*irqptr;
1719 	struct intrspec ispec;
1720 
1721 	/*
1722 	 * On UniSys Model 6520, the BIOS leaves vector 0x20 isr
1723 	 * bit on without clearing it with EOI.  Since softint
1724 	 * uses vector 0x20 to interrupt itself, so softint will
1725 	 * not work on this machine.  In order to fix this problem
1726 	 * a check is made to verify all the isr bits are clear.
1727 	 * If not, EOIs are issued to clear the bits.
1728 	 */
1729 	for (i = 7; i >= 1; i--) {
1730 		if ((isr = apicadr[APIC_ISR_REG + (i * 4)]) != 0)
1731 			for (j = 0; ((j < 32) && (isr != 0)); j++)
1732 				if (isr & (1 << j)) {
1733 					apicadr[APIC_EOI_REG] = 0;
1734 					isr &= ~(1 << j);
1735 					apic_error |= APIC_ERR_BOOT_EOI;
1736 				}
1737 	}
1738 
1739 	/* set a flag so we know we have run apic_picinit() */
1740 	apic_flag = 1;
1741 	LOCK_INIT_CLEAR(&apic_gethrtime_lock);
1742 	LOCK_INIT_CLEAR(&apic_ioapic_lock);
1743 	LOCK_INIT_CLEAR(&apic_revector_lock);
1744 	LOCK_INIT_CLEAR(&apic_ioapic_reprogram_lock);
1745 	LOCK_INIT_CLEAR(&apic_error_lock);
1746 
1747 	picsetup();	 /* initialise the 8259 */
1748 
1749 	/* add nmi handler - least priority nmi handler */
1750 	LOCK_INIT_CLEAR(&apic_nmi_lock);
1751 
1752 	if (!psm_add_nmintr(0, (avfunc) apic_nmi_intr,
1753 	    "pcplusmp NMI handler", (caddr_t)NULL))
1754 		cmn_err(CE_WARN, "pcplusmp: Unable to add nmi handler");
1755 
1756 	apic_init_intr();
1757 
1758 	/* enable apic mode if imcr present */
1759 	if (apic_imcrp) {
1760 		outb(APIC_IMCR_P1, (uchar_t)APIC_IMCR_SELECT);
1761 		outb(APIC_IMCR_P2, (uchar_t)APIC_IMCR_APIC);
1762 	}
1763 
1764 	/* mask interrupt vectors					*/
1765 	for (j = 0; j < apic_io_max; j++) {
1766 		int intin_max;
1767 		ioapic = apicioadr[j];
1768 		ioapic[APIC_IO_REG] = APIC_VERS_CMD;
1769 		/* Bits 23-16 define the maximum redirection entries */
1770 		intin_max = (ioapic[APIC_IO_DATA] >> 16) & 0xff;
1771 		for (i = 0; i < intin_max; i++) {
1772 			ioapic[APIC_IO_REG] = APIC_RDT_CMD + 2 * i;
1773 			ioapic[APIC_IO_DATA] = AV_MASK;
1774 		}
1775 	}
1776 
1777 	/*
1778 	 * Hack alert: deal with ACPI SCI interrupt chicken/egg here
1779 	 */
1780 	if (apic_sci_vect > 0) {
1781 		/*
1782 		 * acpica has already done add_avintr(); we just
1783 		 * to finish the job by mimicing translate_irq()
1784 		 *
1785 		 * Fake up an intrspec and setup the tables
1786 		 */
1787 		ispec.intrspec_vec = apic_sci_vect;
1788 		ispec.intrspec_pri = SCI_IPL;
1789 
1790 		if (apic_setup_irq_table(NULL, apic_sci_vect, NULL,
1791 		    &ispec, &apic_sci_flags, DDI_INTR_TYPE_FIXED) < 0) {
1792 			cmn_err(CE_WARN, "!apic: SCI setup failed");
1793 			return;
1794 		}
1795 		irqptr = apic_irq_table[apic_sci_vect];
1796 
1797 		/* Program I/O APIC */
1798 		(void) apic_setup_io_intr(irqptr, apic_sci_vect);
1799 	}
1800 }
1801 
1802 
1803 static void
1804 apic_cpu_start(processorid_t cpun, caddr_t rm_code)
1805 {
1806 	int		loop_count;
1807 	uint32_t	vector;
1808 	uint_t		cpu_id, iflag;
1809 
1810 	cpu_id = apic_cpus[cpun].aci_local_id;
1811 
1812 	apic_cmos_ssb_set = 1;
1813 
1814 	/*
1815 	 * Interrupts on BSP cpu will be disabled during these startup
1816 	 * steps in order to avoid unwanted side effects from
1817 	 * executing interrupt handlers on a problematic BIOS.
1818 	 */
1819 
1820 	iflag = intr_clear();
1821 	outb(CMOS_ADDR, SSB);
1822 	outb(CMOS_DATA, BIOS_SHUTDOWN);
1823 
1824 	while (get_apic_cmd1() & AV_PENDING)
1825 		apic_ret();
1826 
1827 	/* for integrated - make sure there is one INIT IPI in buffer */
1828 	/* for external - it will wake up the cpu */
1829 	apicadr[APIC_INT_CMD2] = cpu_id << APIC_ICR_ID_BIT_OFFSET;
1830 	apicadr[APIC_INT_CMD1] = AV_ASSERT | AV_RESET;
1831 
1832 	/* If only 1 CPU is installed, PENDING bit will not go low */
1833 	for (loop_count = 0x1000; loop_count; loop_count--)
1834 		if (get_apic_cmd1() & AV_PENDING)
1835 			apic_ret();
1836 		else
1837 			break;
1838 
1839 	apicadr[APIC_INT_CMD2] = cpu_id << APIC_ICR_ID_BIT_OFFSET;
1840 	apicadr[APIC_INT_CMD1] = AV_DEASSERT | AV_RESET;
1841 
1842 	drv_usecwait(20000);		/* 20 milli sec */
1843 
1844 	if (apic_cpus[cpun].aci_local_ver >= APIC_INTEGRATED_VERS) {
1845 		/* integrated apic */
1846 
1847 		rm_code = (caddr_t)(uintptr_t)rm_platter_pa;
1848 		vector = (rm_platter_pa >> MMU_PAGESHIFT) &
1849 		    (APIC_VECTOR_MASK | APIC_IPL_MASK);
1850 
1851 		/* to offset the INIT IPI queue up in the buffer */
1852 		apicadr[APIC_INT_CMD2] = cpu_id << APIC_ICR_ID_BIT_OFFSET;
1853 		apicadr[APIC_INT_CMD1] = vector | AV_STARTUP;
1854 
1855 		drv_usecwait(200);		/* 20 micro sec */
1856 
1857 		apicadr[APIC_INT_CMD2] = cpu_id << APIC_ICR_ID_BIT_OFFSET;
1858 		apicadr[APIC_INT_CMD1] = vector | AV_STARTUP;
1859 
1860 		drv_usecwait(200);		/* 20 micro sec */
1861 	}
1862 	intr_restore(iflag);
1863 }
1864 
1865 
1866 #ifdef	DEBUG
1867 int	apic_break_on_cpu = 9;
1868 int	apic_stretch_interrupts = 0;
1869 int	apic_stretch_ISR = 1 << 3;	/* IPL of 3 matches nothing now */
1870 
1871 void
1872 apic_break()
1873 {
1874 }
1875 #endif /* DEBUG */
1876 
1877 /*
1878  * platform_intr_enter
1879  *
1880  *	Called at the beginning of the interrupt service routine to
1881  *	mask all level equal to and below the interrupt priority
1882  *	of the interrupting vector.  An EOI should be given to
1883  *	the interrupt controller to enable other HW interrupts.
1884  *
1885  *	Return -1 for spurious interrupts
1886  *
1887  */
1888 /*ARGSUSED*/
1889 static int
1890 apic_intr_enter(int ipl, int *vectorp)
1891 {
1892 	uchar_t vector;
1893 	int nipl;
1894 	int irq, iflag;
1895 	apic_cpus_info_t *cpu_infop;
1896 
1897 	/*
1898 	 * The real vector programmed in APIC is *vectorp + 0x20
1899 	 * But, cmnint code subtracts 0x20 before pushing it.
1900 	 * Hence APIC_BASE_VECT is 0x20.
1901 	 */
1902 
1903 	vector = (uchar_t)*vectorp;
1904 
1905 	/* if interrupted by the clock, increment apic_nsec_since_boot */
1906 	if (vector == apic_clkvect) {
1907 		if (!apic_oneshot) {
1908 			/* NOTE: this is not MT aware */
1909 			apic_hrtime_stamp++;
1910 			apic_nsec_since_boot += apic_nsec_per_intr;
1911 			apic_hrtime_stamp++;
1912 			last_count_read = apic_hertz_count;
1913 			apic_redistribute_compute();
1914 		}
1915 
1916 		/* We will avoid all the book keeping overhead for clock */
1917 		nipl = apic_vectortoipl[vector >> APIC_IPL_SHIFT];
1918 #if defined(__amd64)
1919 		setcr8((ulong_t)apic_cr8pri[nipl]);
1920 #else
1921 		apicadr[APIC_TASK_REG] = apic_ipltopri[nipl];
1922 #endif
1923 		*vectorp = apic_vector_to_irq[vector + APIC_BASE_VECT];
1924 		apicadr[APIC_EOI_REG] = 0;
1925 		return (nipl);
1926 	}
1927 
1928 	cpu_infop = &apic_cpus[psm_get_cpu_id()];
1929 
1930 	if (vector == (APIC_SPUR_INTR - APIC_BASE_VECT)) {
1931 		cpu_infop->aci_spur_cnt++;
1932 		return (APIC_INT_SPURIOUS);
1933 	}
1934 
1935 	/* Check if the vector we got is really what we need */
1936 	if (apic_revector_pending) {
1937 		/*
1938 		 * Disable interrupts for the duration of
1939 		 * the vector translation to prevent a self-race for
1940 		 * the apic_revector_lock.  This cannot be done
1941 		 * in apic_xlate_vector because it is recursive and
1942 		 * we want the vector translation to be atomic with
1943 		 * respect to other (higher-priority) interrupts.
1944 		 */
1945 		iflag = intr_clear();
1946 		vector = apic_xlate_vector(vector + APIC_BASE_VECT) -
1947 		    APIC_BASE_VECT;
1948 		intr_restore(iflag);
1949 	}
1950 
1951 	nipl = apic_vectortoipl[vector >> APIC_IPL_SHIFT];
1952 	*vectorp = irq = apic_vector_to_irq[vector + APIC_BASE_VECT];
1953 
1954 #if defined(__amd64)
1955 	setcr8((ulong_t)apic_cr8pri[nipl]);
1956 #else
1957 	apicadr[APIC_TASK_REG] = apic_ipltopri[nipl];
1958 #endif
1959 
1960 	cpu_infop->aci_current[nipl] = (uchar_t)irq;
1961 	cpu_infop->aci_curipl = (uchar_t)nipl;
1962 	cpu_infop->aci_ISR_in_progress |= 1 << nipl;
1963 
1964 	/*
1965 	 * apic_level_intr could have been assimilated into the irq struct.
1966 	 * but, having it as a character array is more efficient in terms of
1967 	 * cache usage. So, we leave it as is.
1968 	 */
1969 	if (!apic_level_intr[irq])
1970 		apicadr[APIC_EOI_REG] = 0;
1971 
1972 #ifdef	DEBUG
1973 	APIC_DEBUG_BUF_PUT(vector);
1974 	APIC_DEBUG_BUF_PUT(irq);
1975 	APIC_DEBUG_BUF_PUT(nipl);
1976 	APIC_DEBUG_BUF_PUT(psm_get_cpu_id());
1977 	if ((apic_stretch_interrupts) && (apic_stretch_ISR & (1 << nipl)))
1978 		drv_usecwait(apic_stretch_interrupts);
1979 
1980 	if (apic_break_on_cpu == psm_get_cpu_id())
1981 		apic_break();
1982 #endif /* DEBUG */
1983 	return (nipl);
1984 }
1985 
1986 static void
1987 apic_intr_exit(int prev_ipl, int irq)
1988 {
1989 	apic_cpus_info_t *cpu_infop;
1990 
1991 #if defined(__amd64)
1992 	setcr8((ulong_t)apic_cr8pri[prev_ipl]);
1993 #else
1994 	apicadr[APIC_TASK_REG] = apic_ipltopri[prev_ipl];
1995 #endif
1996 
1997 	cpu_infop = &apic_cpus[psm_get_cpu_id()];
1998 	if (apic_level_intr[irq])
1999 		apicadr[APIC_EOI_REG] = 0;
2000 
2001 	cpu_infop->aci_curipl = (uchar_t)prev_ipl;
2002 	/* ISR above current pri could not be in progress */
2003 	cpu_infop->aci_ISR_in_progress &= (2 << prev_ipl) - 1;
2004 }
2005 
2006 /*
2007  * Mask all interrupts below or equal to the given IPL
2008  */
2009 static void
2010 apic_setspl(int ipl)
2011 {
2012 
2013 #if defined(__amd64)
2014 	setcr8((ulong_t)apic_cr8pri[ipl]);
2015 #else
2016 	apicadr[APIC_TASK_REG] = apic_ipltopri[ipl];
2017 #endif
2018 
2019 	/* interrupts at ipl above this cannot be in progress */
2020 	apic_cpus[psm_get_cpu_id()].aci_ISR_in_progress &= (2 << ipl) - 1;
2021 	/*
2022 	 * this is a patch fix for the ALR QSMP P5 machine, so that interrupts
2023 	 * have enough time to come in before the priority is raised again
2024 	 * during the idle() loop.
2025 	 */
2026 	if (apic_setspl_delay)
2027 		(void) get_apic_pri();
2028 }
2029 
2030 /*
2031  * trigger a software interrupt at the given IPL
2032  */
2033 static void
2034 apic_set_softintr(int ipl)
2035 {
2036 	int vector;
2037 	uint_t flag;
2038 
2039 	vector = apic_resv_vector[ipl];
2040 
2041 	flag = intr_clear();
2042 
2043 	while (get_apic_cmd1() & AV_PENDING)
2044 		apic_ret();
2045 
2046 	/* generate interrupt at vector on itself only */
2047 	apicadr[APIC_INT_CMD1] = AV_SH_SELF | vector;
2048 
2049 	intr_restore(flag);
2050 }
2051 
2052 /*
2053  * generates an interprocessor interrupt to another CPU
2054  */
2055 static void
2056 apic_send_ipi(int cpun, int ipl)
2057 {
2058 	int vector;
2059 	uint_t flag;
2060 
2061 	vector = apic_resv_vector[ipl];
2062 
2063 	flag = intr_clear();
2064 
2065 	while (get_apic_cmd1() & AV_PENDING)
2066 		apic_ret();
2067 
2068 	apicadr[APIC_INT_CMD2] =
2069 	    apic_cpus[cpun].aci_local_id << APIC_ICR_ID_BIT_OFFSET;
2070 	apicadr[APIC_INT_CMD1] = vector;
2071 
2072 	intr_restore(flag);
2073 }
2074 
2075 
2076 /*ARGSUSED*/
2077 static void
2078 apic_set_idlecpu(processorid_t cpun)
2079 {
2080 }
2081 
2082 /*ARGSUSED*/
2083 static void
2084 apic_unset_idlecpu(processorid_t cpun)
2085 {
2086 }
2087 
2088 
2089 static void
2090 apic_ret()
2091 {
2092 }
2093 
2094 static int
2095 get_apic_cmd1()
2096 {
2097 	return (apicadr[APIC_INT_CMD1]);
2098 }
2099 
2100 static int
2101 get_apic_pri()
2102 {
2103 #if defined(__amd64)
2104 	return ((int)getcr8());
2105 #else
2106 	return (apicadr[APIC_TASK_REG]);
2107 #endif
2108 }
2109 
2110 /*
2111  * If apic_coarse_time == 1, then apic_gettime() is used instead of
2112  * apic_gethrtime().  This is used for performance instead of accuracy.
2113  */
2114 
2115 static hrtime_t
2116 apic_gettime()
2117 {
2118 	int old_hrtime_stamp;
2119 	hrtime_t temp;
2120 
2121 	/*
2122 	 * In one-shot mode, we do not keep time, so if anyone
2123 	 * calls psm_gettime() directly, we vector over to
2124 	 * gethrtime().
2125 	 * one-shot mode MUST NOT be enabled if this psm is the source of
2126 	 * hrtime.
2127 	 */
2128 
2129 	if (apic_oneshot)
2130 		return (gethrtime());
2131 
2132 
2133 gettime_again:
2134 	while ((old_hrtime_stamp = apic_hrtime_stamp) & 1)
2135 		apic_ret();
2136 
2137 	temp = apic_nsec_since_boot;
2138 
2139 	if (apic_hrtime_stamp != old_hrtime_stamp) {	/* got an interrupt */
2140 		goto gettime_again;
2141 	}
2142 	return (temp);
2143 }
2144 
2145 /*
2146  * Here we return the number of nanoseconds since booting.  Note every
2147  * clock interrupt increments apic_nsec_since_boot by the appropriate
2148  * amount.
2149  */
2150 static hrtime_t
2151 apic_gethrtime()
2152 {
2153 	int curr_timeval, countval, elapsed_ticks, oflags;
2154 	int old_hrtime_stamp, status;
2155 	hrtime_t temp;
2156 	uchar_t	cpun;
2157 
2158 
2159 	/*
2160 	 * In one-shot mode, we do not keep time, so if anyone
2161 	 * calls psm_gethrtime() directly, we vector over to
2162 	 * gethrtime().
2163 	 * one-shot mode MUST NOT be enabled if this psm is the source of
2164 	 * hrtime.
2165 	 */
2166 
2167 	if (apic_oneshot)
2168 		return (gethrtime());
2169 
2170 	oflags = intr_clear();	/* prevent migration */
2171 
2172 	cpun = (uchar_t)((uint_t)apicadr[APIC_LID_REG] >> APIC_ID_BIT_OFFSET);
2173 
2174 	lock_set(&apic_gethrtime_lock);
2175 
2176 gethrtime_again:
2177 	while ((old_hrtime_stamp = apic_hrtime_stamp) & 1)
2178 		apic_ret();
2179 
2180 	/*
2181 	 * Check to see which CPU we are on.  Note the time is kept on
2182 	 * the local APIC of CPU 0.  If on CPU 0, simply read the current
2183 	 * counter.  If on another CPU, issue a remote read command to CPU 0.
2184 	 */
2185 	if (cpun == apic_cpus[0].aci_local_id) {
2186 		countval = apicadr[APIC_CURR_COUNT];
2187 	} else {
2188 		while (get_apic_cmd1() & AV_PENDING)
2189 			apic_ret();
2190 
2191 		apicadr[APIC_INT_CMD2] =
2192 		    apic_cpus[0].aci_local_id << APIC_ICR_ID_BIT_OFFSET;
2193 		apicadr[APIC_INT_CMD1] = APIC_CURR_ADD|AV_REMOTE;
2194 
2195 		while ((status = get_apic_cmd1()) & AV_READ_PENDING)
2196 			apic_ret();
2197 
2198 		if (status & AV_REMOTE_STATUS)	/* 1 = valid */
2199 			countval = apicadr[APIC_REMOTE_READ];
2200 		else {	/* 0 = invalid */
2201 			apic_remote_hrterr++;
2202 			/*
2203 			 * return last hrtime right now, will need more
2204 			 * testing if change to retry
2205 			 */
2206 			temp = apic_last_hrtime;
2207 
2208 			lock_clear(&apic_gethrtime_lock);
2209 
2210 			intr_restore(oflags);
2211 
2212 			return (temp);
2213 		}
2214 	}
2215 	if (countval > last_count_read)
2216 		countval = 0;
2217 	else
2218 		last_count_read = countval;
2219 
2220 	elapsed_ticks = apic_hertz_count - countval;
2221 
2222 	curr_timeval = elapsed_ticks * apic_nsec_per_tick;
2223 	temp = apic_nsec_since_boot + curr_timeval;
2224 
2225 	if (apic_hrtime_stamp != old_hrtime_stamp) {	/* got an interrupt */
2226 		/* we might have clobbered last_count_read. Restore it */
2227 		last_count_read = apic_hertz_count;
2228 		goto gethrtime_again;
2229 	}
2230 
2231 	if (temp < apic_last_hrtime) {
2232 		/* return last hrtime if error occurs */
2233 		apic_hrtime_error++;
2234 		temp = apic_last_hrtime;
2235 	}
2236 	else
2237 		apic_last_hrtime = temp;
2238 
2239 	lock_clear(&apic_gethrtime_lock);
2240 	intr_restore(oflags);
2241 
2242 	return (temp);
2243 }
2244 
2245 /* apic NMI handler */
2246 /*ARGSUSED*/
2247 static void
2248 apic_nmi_intr(caddr_t arg)
2249 {
2250 	if (apic_shutdown_processors) {
2251 		apic_disable_local_apic();
2252 		return;
2253 	}
2254 
2255 	if (lock_try(&apic_nmi_lock)) {
2256 		if (apic_kmdb_on_nmi) {
2257 			if (psm_debugger() == 0) {
2258 				cmn_err(CE_PANIC,
2259 				    "NMI detected, kmdb is not available.");
2260 			} else {
2261 				debug_enter("\nNMI detected, entering kmdb.\n");
2262 			}
2263 		} else {
2264 			if (apic_panic_on_nmi) {
2265 				/* Keep panic from entering kmdb. */
2266 				nopanicdebug = 1;
2267 				cmn_err(CE_PANIC, "pcplusmp: NMI received");
2268 			} else {
2269 				/*
2270 				 * prom_printf is the best shot we have
2271 				 * of something which is problem free from
2272 				 * high level/NMI type of interrupts
2273 				 */
2274 				prom_printf("pcplusmp: NMI received\n");
2275 				apic_error |= APIC_ERR_NMI;
2276 				apic_num_nmis++;
2277 			}
2278 		}
2279 		lock_clear(&apic_nmi_lock);
2280 	}
2281 }
2282 
2283 /*
2284  * Add mask bits to disable interrupt vector from happening
2285  * at or above IPL. In addition, it should remove mask bits
2286  * to enable interrupt vectors below the given IPL.
2287  *
2288  * Both add and delspl are complicated by the fact that different interrupts
2289  * may share IRQs. This can happen in two ways.
2290  * 1. The same H/W line is shared by more than 1 device
2291  * 1a. with interrupts at different IPLs
2292  * 1b. with interrupts at same IPL
2293  * 2. We ran out of vectors at a given IPL and started sharing vectors.
2294  * 1b and 2 should be handled gracefully, except for the fact some ISRs
2295  * will get called often when no interrupt is pending for the device.
2296  * For 1a, we just hope that the machine blows up with the person who
2297  * set it up that way!. In the meantime, we handle it at the higher IPL.
2298  */
2299 /*ARGSUSED*/
2300 static int
2301 apic_addspl(int irqno, int ipl, int min_ipl, int max_ipl)
2302 {
2303 	uchar_t vector;
2304 	int iflag;
2305 	apic_irq_t *irqptr, *irqheadptr;
2306 	int irqindex;
2307 
2308 	ASSERT(max_ipl <= UCHAR_MAX);
2309 	irqindex = IRQINDEX(irqno);
2310 
2311 	if ((irqindex == -1) || (!apic_irq_table[irqindex]))
2312 		return (PSM_FAILURE);
2313 
2314 	irqptr = irqheadptr = apic_irq_table[irqindex];
2315 
2316 	DDI_INTR_IMPLDBG((CE_CONT, "apic_addspl: dip=0x%p type=%d irqno=0x%x "
2317 	    "vector=0x%x\n", (void *)irqptr->airq_dip,
2318 	    irqptr->airq_mps_intr_index, irqno, irqptr->airq_vector));
2319 
2320 	while (irqptr) {
2321 		if (VIRTIRQ(irqindex, irqptr->airq_share_id) == irqno)
2322 			break;
2323 		irqptr = irqptr->airq_next;
2324 	}
2325 	irqptr->airq_share++;
2326 
2327 	/* return if it is not hardware interrupt */
2328 	if (irqptr->airq_mps_intr_index == RESERVE_INDEX)
2329 		return (PSM_SUCCESS);
2330 
2331 	/* Or if there are more interupts at a higher IPL */
2332 	if (ipl != max_ipl)
2333 		return (PSM_SUCCESS);
2334 
2335 	/*
2336 	 * if apic_picinit() has not been called yet, just return.
2337 	 * At the end of apic_picinit(), we will call setup_io_intr().
2338 	 */
2339 
2340 	if (!apic_flag)
2341 		return (PSM_SUCCESS);
2342 
2343 	iflag = intr_clear();
2344 
2345 	/*
2346 	 * Upgrade vector if max_ipl is not earlier ipl. If we cannot allocate,
2347 	 * return failure. Not very elegant, but then we hope the
2348 	 * machine will blow up with ...
2349 	 */
2350 	if (irqptr->airq_ipl != max_ipl) {
2351 		vector = apic_allocate_vector(max_ipl, irqindex, 1);
2352 		if (vector == 0) {
2353 			intr_restore(iflag);
2354 			irqptr->airq_share--;
2355 			return (PSM_FAILURE);
2356 		}
2357 		irqptr = irqheadptr;
2358 		apic_mark_vector(irqptr->airq_vector, vector);
2359 		while (irqptr) {
2360 			irqptr->airq_vector = vector;
2361 			irqptr->airq_ipl = (uchar_t)max_ipl;
2362 			/*
2363 			 * reprogram irq being added and every one else
2364 			 * who is not in the UNINIT state
2365 			 */
2366 			if ((VIRTIRQ(irqindex, irqptr->airq_share_id) ==
2367 			    irqno) || (irqptr->airq_temp_cpu != IRQ_UNINIT)) {
2368 				apic_record_rdt_entry(irqptr, irqindex);
2369 				(void) apic_setup_io_intr(irqptr, irqindex);
2370 			}
2371 			irqptr = irqptr->airq_next;
2372 		}
2373 		intr_restore(iflag);
2374 		return (PSM_SUCCESS);
2375 	}
2376 
2377 	ASSERT(irqptr);
2378 	(void) apic_setup_io_intr(irqptr, irqindex);
2379 	intr_restore(iflag);
2380 	return (PSM_SUCCESS);
2381 }
2382 
2383 /*
2384  * Recompute mask bits for the given interrupt vector.
2385  * If there is no interrupt servicing routine for this
2386  * vector, this function should disable interrupt vector
2387  * from happening at all IPLs. If there are still
2388  * handlers using the given vector, this function should
2389  * disable the given vector from happening below the lowest
2390  * IPL of the remaining hadlers.
2391  */
2392 /*ARGSUSED*/
2393 static int
2394 apic_delspl(int irqno, int ipl, int min_ipl, int max_ipl)
2395 {
2396 	uchar_t vector, bind_cpu;
2397 	int	iflag, intin, irqindex;
2398 	volatile int32_t *ioapic;
2399 	apic_irq_t	*irqptr, *irqheadptr;
2400 
2401 	irqindex = IRQINDEX(irqno);
2402 	irqptr = irqheadptr = apic_irq_table[irqindex];
2403 
2404 	DDI_INTR_IMPLDBG((CE_CONT, "apic_delspl: dip=0x%p type=%d irqno=0x%x "
2405 	    "vector=0x%x\n", (void *)irqptr->airq_dip,
2406 	    irqptr->airq_mps_intr_index, irqno, irqptr->airq_vector));
2407 
2408 	while (irqptr) {
2409 		if (VIRTIRQ(irqindex, irqptr->airq_share_id) == irqno)
2410 			break;
2411 		irqptr = irqptr->airq_next;
2412 	}
2413 	ASSERT(irqptr);
2414 
2415 	irqptr->airq_share--;
2416 
2417 	if (ipl < max_ipl)
2418 		return (PSM_SUCCESS);
2419 
2420 	/* return if it is not hardware interrupt */
2421 	if (irqptr->airq_mps_intr_index == RESERVE_INDEX)
2422 		return (PSM_SUCCESS);
2423 
2424 	if (!apic_flag) {
2425 		/*
2426 		 * Clear irq_struct. If two devices shared an intpt
2427 		 * line & 1 unloaded before picinit, we are hosed. But, then
2428 		 * we hope the machine will ...
2429 		 */
2430 		irqptr->airq_mps_intr_index = FREE_INDEX;
2431 		irqptr->airq_temp_cpu = IRQ_UNINIT;
2432 		apic_free_vector(irqptr->airq_vector);
2433 		return (PSM_SUCCESS);
2434 	}
2435 	/*
2436 	 * Downgrade vector to new max_ipl if needed.If we cannot allocate,
2437 	 * use old IPL. Not very elegant, but then we hope ...
2438 	 */
2439 	if ((irqptr->airq_ipl != max_ipl) && (max_ipl != PSM_INVALID_IPL)) {
2440 		apic_irq_t	*irqp;
2441 		if (vector = apic_allocate_vector(max_ipl, irqno, 1)) {
2442 			apic_mark_vector(irqheadptr->airq_vector, vector);
2443 			irqp = irqheadptr;
2444 			while (irqp) {
2445 				irqp->airq_vector = vector;
2446 				irqp->airq_ipl = (uchar_t)max_ipl;
2447 				if (irqp->airq_temp_cpu != IRQ_UNINIT) {
2448 					apic_record_rdt_entry(irqp, irqindex);
2449 					(void) apic_setup_io_intr(irqp,
2450 					    irqindex);
2451 				}
2452 				irqp = irqp->airq_next;
2453 			}
2454 		}
2455 	}
2456 
2457 	if (irqptr->airq_share)
2458 		return (PSM_SUCCESS);
2459 
2460 	ioapic = apicioadr[irqptr->airq_ioapicindex];
2461 	intin = irqptr->airq_intin_no;
2462 	iflag = intr_clear();
2463 	lock_set(&apic_ioapic_lock);
2464 	ioapic[APIC_IO_REG] = APIC_RDT_CMD + 2 * intin;
2465 	ioapic[APIC_IO_DATA] = AV_MASK;
2466 
2467 	/* Disable the MSI/X vector */
2468 	if (APIC_IS_MSI_OR_MSIX_INDEX(irqptr->airq_mps_intr_index)) {
2469 		int type = (irqptr->airq_mps_intr_index == MSI_INDEX) ?
2470 		    DDI_INTR_TYPE_MSI : DDI_INTR_TYPE_MSIX;
2471 
2472 		/*
2473 		 * Make sure we only disable on the last
2474 		 * of the multi-MSI support
2475 		 */
2476 		if (i_ddi_intr_get_current_nintrs(irqptr->airq_dip) == 1) {
2477 			(void) pci_msi_unconfigure(irqptr->airq_dip, type,
2478 			    irqptr->airq_ioapicindex);
2479 
2480 			(void) pci_msi_disable_mode(irqptr->airq_dip, type,
2481 			    irqptr->airq_ioapicindex);
2482 		}
2483 	}
2484 
2485 	if (max_ipl == PSM_INVALID_IPL) {
2486 		ASSERT(irqheadptr == irqptr);
2487 		bind_cpu = irqptr->airq_temp_cpu;
2488 		if (((uchar_t)bind_cpu != IRQ_UNBOUND) &&
2489 		    ((uchar_t)bind_cpu != IRQ_UNINIT)) {
2490 			ASSERT((bind_cpu & ~IRQ_USER_BOUND) < apic_nproc);
2491 			if (bind_cpu & IRQ_USER_BOUND) {
2492 				/* If hardbound, temp_cpu == cpu */
2493 				bind_cpu &= ~IRQ_USER_BOUND;
2494 				apic_cpus[bind_cpu].aci_bound--;
2495 			} else
2496 				apic_cpus[bind_cpu].aci_temp_bound--;
2497 		}
2498 		lock_clear(&apic_ioapic_lock);
2499 		intr_restore(iflag);
2500 		irqptr->airq_temp_cpu = IRQ_UNINIT;
2501 		irqptr->airq_mps_intr_index = FREE_INDEX;
2502 		apic_free_vector(irqptr->airq_vector);
2503 		return (PSM_SUCCESS);
2504 	}
2505 	lock_clear(&apic_ioapic_lock);
2506 	intr_restore(iflag);
2507 
2508 	mutex_enter(&airq_mutex);
2509 	if ((irqptr == apic_irq_table[irqindex])) {
2510 		apic_irq_t	*oldirqptr;
2511 		/* Move valid irq entry to the head */
2512 		irqheadptr = oldirqptr = irqptr;
2513 		irqptr = irqptr->airq_next;
2514 		ASSERT(irqptr);
2515 		while (irqptr) {
2516 			if (irqptr->airq_mps_intr_index != FREE_INDEX)
2517 				break;
2518 			oldirqptr = irqptr;
2519 			irqptr = irqptr->airq_next;
2520 		}
2521 		/* remove all invalid ones from the beginning */
2522 		apic_irq_table[irqindex] = irqptr;
2523 		/*
2524 		 * and link them back after the head. The invalid ones
2525 		 * begin with irqheadptr and end at oldirqptr
2526 		 */
2527 		oldirqptr->airq_next = irqptr->airq_next;
2528 		irqptr->airq_next = irqheadptr;
2529 	}
2530 	mutex_exit(&airq_mutex);
2531 
2532 	irqptr->airq_temp_cpu = IRQ_UNINIT;
2533 	irqptr->airq_mps_intr_index = FREE_INDEX;
2534 	return (PSM_SUCCESS);
2535 }
2536 
2537 /*
2538  * Return HW interrupt number corresponding to the given IPL
2539  */
2540 /*ARGSUSED*/
2541 static int
2542 apic_softlvl_to_irq(int ipl)
2543 {
2544 	/*
2545 	 * Do not use apic to trigger soft interrupt.
2546 	 * It will cause the system to hang when 2 hardware interrupts
2547 	 * at the same priority with the softint are already accepted
2548 	 * by the apic.  Cause the AV_PENDING bit will not be cleared
2549 	 * until one of the hardware interrupt is eoi'ed.  If we need
2550 	 * to send an ipi at this time, we will end up looping forever
2551 	 * to wait for the AV_PENDING bit to clear.
2552 	 */
2553 	return (PSM_SV_SOFTWARE);
2554 }
2555 
2556 static int
2557 apic_post_cpu_start()
2558 {
2559 	int i, cpun;
2560 	apic_irq_t *irq_ptr;
2561 
2562 	apic_init_intr();
2563 
2564 	/*
2565 	 * since some systems don't enable the internal cache on the non-boot
2566 	 * cpus, so we have to enable them here
2567 	 */
2568 	setcr0(getcr0() & ~(0x60000000));
2569 
2570 	while (get_apic_cmd1() & AV_PENDING)
2571 		apic_ret();
2572 
2573 	cpun = psm_get_cpu_id();
2574 	apic_cpus[cpun].aci_status = APIC_CPU_ONLINE | APIC_CPU_INTR_ENABLE;
2575 
2576 	for (i = apic_min_device_irq; i <= apic_max_device_irq; i++) {
2577 		irq_ptr = apic_irq_table[i];
2578 		if ((irq_ptr == NULL) ||
2579 		    ((irq_ptr->airq_cpu & ~IRQ_USER_BOUND) != cpun))
2580 			continue;
2581 
2582 		while (irq_ptr) {
2583 			if (irq_ptr->airq_temp_cpu != IRQ_UNINIT)
2584 				(void) apic_rebind(irq_ptr, cpun, 1, IMMEDIATE);
2585 			irq_ptr = irq_ptr->airq_next;
2586 		}
2587 	}
2588 
2589 	return (PSM_SUCCESS);
2590 }
2591 
2592 processorid_t
2593 apic_get_next_processorid(processorid_t cpu_id)
2594 {
2595 
2596 	int i;
2597 
2598 	if (cpu_id == -1)
2599 		return ((processorid_t)0);
2600 
2601 	for (i = cpu_id + 1; i < NCPU; i++) {
2602 		if (apic_cpumask & (1 << i))
2603 			return (i);
2604 	}
2605 
2606 	return ((processorid_t)-1);
2607 }
2608 
2609 
2610 /*
2611  * type == -1 indicates it is an internal request. Do not change
2612  * resv_vector for these requests
2613  */
2614 static int
2615 apic_get_ipivect(int ipl, int type)
2616 {
2617 	uchar_t vector;
2618 	int irq;
2619 
2620 	if (irq = apic_allocate_irq(APIC_VECTOR(ipl))) {
2621 		if (vector = apic_allocate_vector(ipl, irq, 1)) {
2622 			apic_irq_table[irq]->airq_mps_intr_index =
2623 			    RESERVE_INDEX;
2624 			apic_irq_table[irq]->airq_vector = vector;
2625 			if (type != -1) {
2626 				apic_resv_vector[ipl] = vector;
2627 			}
2628 			return (irq);
2629 		}
2630 	}
2631 	apic_error |= APIC_ERR_GET_IPIVECT_FAIL;
2632 	return (-1);	/* shouldn't happen */
2633 }
2634 
2635 static int
2636 apic_getclkirq(int ipl)
2637 {
2638 	int	irq;
2639 
2640 	if ((irq = apic_get_ipivect(ipl, -1)) == -1)
2641 		return (-1);
2642 	/*
2643 	 * Note the vector in apic_clkvect for per clock handling.
2644 	 */
2645 	apic_clkvect = apic_irq_table[irq]->airq_vector - APIC_BASE_VECT;
2646 	APIC_VERBOSE_IOAPIC((CE_NOTE, "get_clkirq: vector = %x\n",
2647 	    apic_clkvect));
2648 	return (irq);
2649 }
2650 
2651 /*
2652  * Return the number of APIC clock ticks elapsed for 8245 to decrement
2653  * (APIC_TIME_COUNT + pit_ticks_adj) ticks.
2654  */
2655 static uint_t
2656 apic_calibrate(volatile uint32_t *addr, uint16_t *pit_ticks_adj)
2657 {
2658 	uint8_t		pit_tick_lo;
2659 	uint16_t	pit_tick, target_pit_tick;
2660 	uint32_t	start_apic_tick, end_apic_tick;
2661 	int		iflag;
2662 
2663 	addr += APIC_CURR_COUNT;
2664 
2665 	iflag = intr_clear();
2666 
2667 	do {
2668 		pit_tick_lo = inb(PITCTR0_PORT);
2669 		pit_tick = (inb(PITCTR0_PORT) << 8) | pit_tick_lo;
2670 	} while (pit_tick < APIC_TIME_MIN ||
2671 	    pit_tick_lo <= APIC_LB_MIN || pit_tick_lo >= APIC_LB_MAX);
2672 
2673 	/*
2674 	 * Wait for the 8254 to decrement by 5 ticks to ensure
2675 	 * we didn't start in the middle of a tick.
2676 	 * Compare with 0x10 for the wrap around case.
2677 	 */
2678 	target_pit_tick = pit_tick - 5;
2679 	do {
2680 		pit_tick_lo = inb(PITCTR0_PORT);
2681 		pit_tick = (inb(PITCTR0_PORT) << 8) | pit_tick_lo;
2682 	} while (pit_tick > target_pit_tick || pit_tick_lo < 0x10);
2683 
2684 	start_apic_tick = *addr;
2685 
2686 	/*
2687 	 * Wait for the 8254 to decrement by
2688 	 * (APIC_TIME_COUNT + pit_ticks_adj) ticks
2689 	 */
2690 	target_pit_tick = pit_tick - APIC_TIME_COUNT;
2691 	do {
2692 		pit_tick_lo = inb(PITCTR0_PORT);
2693 		pit_tick = (inb(PITCTR0_PORT) << 8) | pit_tick_lo;
2694 	} while (pit_tick > target_pit_tick || pit_tick_lo < 0x10);
2695 
2696 	end_apic_tick = *addr;
2697 
2698 	*pit_ticks_adj = target_pit_tick - pit_tick;
2699 
2700 	intr_restore(iflag);
2701 
2702 	return (start_apic_tick - end_apic_tick);
2703 }
2704 
2705 /*
2706  * Initialise the APIC timer on the local APIC of CPU 0 to the desired
2707  * frequency.  Note at this stage in the boot sequence, the boot processor
2708  * is the only active processor.
2709  * hertz value of 0 indicates a one-shot mode request.  In this case
2710  * the function returns the resolution (in nanoseconds) for the hardware
2711  * timer interrupt.  If one-shot mode capability is not available,
2712  * the return value will be 0. apic_enable_oneshot is a global switch
2713  * for disabling the functionality.
2714  * A non-zero positive value for hertz indicates a periodic mode request.
2715  * In this case the hardware will be programmed to generate clock interrupts
2716  * at hertz frequency and returns the resolution of interrupts in
2717  * nanosecond.
2718  */
2719 
2720 static int
2721 apic_clkinit(int hertz)
2722 {
2723 
2724 	uint_t		apic_ticks = 0;
2725 	uint_t		pit_time;
2726 	int		ret;
2727 	uint16_t	pit_ticks_adj;
2728 	static int	firsttime = 1;
2729 
2730 	if (firsttime) {
2731 		/* first time calibrate */
2732 
2733 		apicadr[APIC_DIVIDE_REG] = 0x0;
2734 		apicadr[APIC_INIT_COUNT] = APIC_MAXVAL;
2735 
2736 		/* set periodic interrupt based on CLKIN */
2737 		apicadr[APIC_LOCAL_TIMER] =
2738 		    (apic_clkvect + APIC_BASE_VECT) | AV_TIME;
2739 		tenmicrosec();
2740 
2741 		apic_ticks = apic_calibrate(apicadr, &pit_ticks_adj);
2742 
2743 		apicadr[APIC_LOCAL_TIMER] =
2744 		    (apic_clkvect + APIC_BASE_VECT) | AV_MASK;
2745 		/*
2746 		 * pit time is the amount of real time (in nanoseconds ) it took
2747 		 * the 8254 to decrement (APIC_TIME_COUNT + pit_ticks_adj) ticks
2748 		 */
2749 		pit_time = ((longlong_t)(APIC_TIME_COUNT +
2750 		    pit_ticks_adj) * NANOSEC) / PIT_HZ;
2751 
2752 		/*
2753 		 * Determine the number of nanoseconds per APIC clock tick
2754 		 * and then determine how many APIC ticks to interrupt at the
2755 		 * desired frequency
2756 		 */
2757 		apic_nsec_per_tick = pit_time / apic_ticks;
2758 		if (apic_nsec_per_tick == 0)
2759 			apic_nsec_per_tick = 1;
2760 
2761 		/* the interval timer initial count is 32 bit max */
2762 		apic_nsec_max = (hrtime_t)apic_nsec_per_tick * APIC_MAXVAL;
2763 		firsttime = 0;
2764 	}
2765 
2766 	if (hertz != 0) {
2767 		/* periodic */
2768 		apic_nsec_per_intr = NANOSEC / hertz;
2769 		apic_hertz_count = (longlong_t)apic_nsec_per_intr /
2770 		    apic_nsec_per_tick;
2771 		apic_sample_factor_redistribution = hertz + 1;
2772 	}
2773 
2774 	apic_int_busy_mark = (apic_int_busy_mark *
2775 	    apic_sample_factor_redistribution) / 100;
2776 	apic_int_free_mark = (apic_int_free_mark *
2777 	    apic_sample_factor_redistribution) / 100;
2778 	apic_diff_for_redistribution = (apic_diff_for_redistribution *
2779 	    apic_sample_factor_redistribution) / 100;
2780 
2781 	if (hertz == 0) {
2782 		/* requested one_shot */
2783 		if (!apic_oneshot_enable)
2784 			return (0);
2785 		apic_oneshot = 1;
2786 		ret = (int)apic_nsec_per_tick;
2787 	} else {
2788 		/* program the local APIC to interrupt at the given frequency */
2789 		apicadr[APIC_INIT_COUNT] = apic_hertz_count;
2790 		apicadr[APIC_LOCAL_TIMER] =
2791 		    (apic_clkvect + APIC_BASE_VECT) | AV_TIME;
2792 		apic_oneshot = 0;
2793 		ret = NANOSEC / hertz;
2794 	}
2795 
2796 	return (ret);
2797 
2798 }
2799 
2800 /*
2801  * apic_preshutdown:
2802  * Called early in shutdown whilst we can still access filesystems to do
2803  * things like loading modules which will be required to complete shutdown
2804  * after filesystems are all unmounted.
2805  */
2806 static void
2807 apic_preshutdown(int cmd, int fcn)
2808 {
2809 	APIC_VERBOSE_POWEROFF(("apic_preshutdown(%d,%d); m=%d a=%d\n",
2810 	    cmd, fcn, apic_poweroff_method, apic_enable_acpi));
2811 
2812 	if ((cmd != A_SHUTDOWN) || (fcn != AD_POWEROFF)) {
2813 		return;
2814 	}
2815 }
2816 
2817 static void
2818 apic_shutdown(int cmd, int fcn)
2819 {
2820 	int iflag, restarts, attempts;
2821 	int i, j;
2822 	volatile int32_t *ioapic;
2823 	uchar_t	byte;
2824 
2825 	/* Send NMI to all CPUs except self to do per processor shutdown */
2826 	iflag = intr_clear();
2827 	while (get_apic_cmd1() & AV_PENDING)
2828 		apic_ret();
2829 	apic_shutdown_processors = 1;
2830 	apicadr[APIC_INT_CMD1] = AV_NMI | AV_LEVEL | AV_SH_ALL_EXCSELF;
2831 
2832 	/* restore cmos shutdown byte before reboot */
2833 	if (apic_cmos_ssb_set) {
2834 		outb(CMOS_ADDR, SSB);
2835 		outb(CMOS_DATA, 0);
2836 	}
2837 	/* Disable the I/O APIC redirection entries */
2838 	for (j = 0; j < apic_io_max; j++) {
2839 		int intin_max;
2840 		ioapic = apicioadr[j];
2841 		ioapic[APIC_IO_REG] = APIC_VERS_CMD;
2842 		/* Bits 23-16 define the maximum redirection entries */
2843 		intin_max = (ioapic[APIC_IO_DATA] >> 16) & 0xff;
2844 		for (i = 0; i < intin_max; i++) {
2845 			ioapic[APIC_IO_REG] = APIC_RDT_CMD + 2 * i;
2846 			ioapic[APIC_IO_DATA] = AV_MASK;
2847 		}
2848 	}
2849 
2850 	/*	disable apic mode if imcr present	*/
2851 	if (apic_imcrp) {
2852 		outb(APIC_IMCR_P1, (uchar_t)APIC_IMCR_SELECT);
2853 		outb(APIC_IMCR_P2, (uchar_t)APIC_IMCR_PIC);
2854 	}
2855 
2856 	apic_disable_local_apic();
2857 
2858 	intr_restore(iflag);
2859 
2860 	if ((cmd != A_SHUTDOWN) || (fcn != AD_POWEROFF)) {
2861 		return;
2862 	}
2863 
2864 	switch (apic_poweroff_method) {
2865 		case APIC_POWEROFF_VIA_RTC:
2866 
2867 			/* select the extended NVRAM bank in the RTC */
2868 			outb(CMOS_ADDR, RTC_REGA);
2869 			byte = inb(CMOS_DATA);
2870 			outb(CMOS_DATA, (byte | EXT_BANK));
2871 
2872 			outb(CMOS_ADDR, PFR_REG);
2873 
2874 			/* for Predator must toggle the PAB bit */
2875 			byte = inb(CMOS_DATA);
2876 
2877 			/*
2878 			 * clear power active bar, wakeup alarm and
2879 			 * kickstart
2880 			 */
2881 			byte &= ~(PAB_CBIT | WF_FLAG | KS_FLAG);
2882 			outb(CMOS_DATA, byte);
2883 
2884 			/* delay before next write */
2885 			drv_usecwait(1000);
2886 
2887 			/* for S40 the following would suffice */
2888 			byte = inb(CMOS_DATA);
2889 
2890 			/* power active bar control bit */
2891 			byte |= PAB_CBIT;
2892 			outb(CMOS_DATA, byte);
2893 
2894 			break;
2895 
2896 		case APIC_POWEROFF_VIA_ASPEN_BMC:
2897 			restarts = 0;
2898 restart_aspen_bmc:
2899 			if (++restarts == 3)
2900 				break;
2901 			attempts = 0;
2902 			do {
2903 				byte = inb(MISMIC_FLAG_REGISTER);
2904 				byte &= MISMIC_BUSY_MASK;
2905 				if (byte != 0) {
2906 					drv_usecwait(1000);
2907 					if (attempts >= 3)
2908 						goto restart_aspen_bmc;
2909 					++attempts;
2910 				}
2911 			} while (byte != 0);
2912 			outb(MISMIC_CNTL_REGISTER, CC_SMS_GET_STATUS);
2913 			byte = inb(MISMIC_FLAG_REGISTER);
2914 			byte |= 0x1;
2915 			outb(MISMIC_FLAG_REGISTER, byte);
2916 			i = 0;
2917 			for (; i < (sizeof (aspen_bmc)/sizeof (aspen_bmc[0]));
2918 			    i++) {
2919 				attempts = 0;
2920 				do {
2921 					byte = inb(MISMIC_FLAG_REGISTER);
2922 					byte &= MISMIC_BUSY_MASK;
2923 					if (byte != 0) {
2924 						drv_usecwait(1000);
2925 						if (attempts >= 3)
2926 							goto restart_aspen_bmc;
2927 						++attempts;
2928 					}
2929 				} while (byte != 0);
2930 				outb(MISMIC_CNTL_REGISTER, aspen_bmc[i].cntl);
2931 				outb(MISMIC_DATA_REGISTER, aspen_bmc[i].data);
2932 				byte = inb(MISMIC_FLAG_REGISTER);
2933 				byte |= 0x1;
2934 				outb(MISMIC_FLAG_REGISTER, byte);
2935 			}
2936 			break;
2937 
2938 		case APIC_POWEROFF_VIA_SITKA_BMC:
2939 			restarts = 0;
2940 restart_sitka_bmc:
2941 			if (++restarts == 3)
2942 				break;
2943 			attempts = 0;
2944 			do {
2945 				byte = inb(SMS_STATUS_REGISTER);
2946 				byte &= SMS_STATE_MASK;
2947 				if ((byte == SMS_READ_STATE) ||
2948 				    (byte == SMS_WRITE_STATE)) {
2949 					drv_usecwait(1000);
2950 					if (attempts >= 3)
2951 						goto restart_sitka_bmc;
2952 					++attempts;
2953 				}
2954 			} while ((byte == SMS_READ_STATE) ||
2955 			    (byte == SMS_WRITE_STATE));
2956 			outb(SMS_COMMAND_REGISTER, SMS_GET_STATUS);
2957 			i = 0;
2958 			for (; i < (sizeof (sitka_bmc)/sizeof (sitka_bmc[0]));
2959 			    i++) {
2960 				attempts = 0;
2961 				do {
2962 					byte = inb(SMS_STATUS_REGISTER);
2963 					byte &= SMS_IBF_MASK;
2964 					if (byte != 0) {
2965 						drv_usecwait(1000);
2966 						if (attempts >= 3)
2967 							goto restart_sitka_bmc;
2968 						++attempts;
2969 					}
2970 				} while (byte != 0);
2971 				outb(sitka_bmc[i].port, sitka_bmc[i].data);
2972 			}
2973 			break;
2974 
2975 		case APIC_POWEROFF_NONE:
2976 
2977 			/* If no APIC direct method, we will try using ACPI */
2978 			if (apic_enable_acpi) {
2979 				if (acpi_poweroff() == 1)
2980 					return;
2981 			} else
2982 				return;
2983 
2984 			break;
2985 	}
2986 	/*
2987 	 * Wait a limited time here for power to go off.
2988 	 * If the power does not go off, then there was a
2989 	 * problem and we should continue to the halt which
2990 	 * prints a message for the user to press a key to
2991 	 * reboot.
2992 	 */
2993 	drv_usecwait(7000000); /* wait seven seconds */
2994 
2995 }
2996 
2997 /*
2998  * Try and disable all interrupts. We just assign interrupts to other
2999  * processors based on policy. If any were bound by user request, we
3000  * let them continue and return failure. We do not bother to check
3001  * for cache affinity while rebinding.
3002  */
3003 
3004 static int
3005 apic_disable_intr(processorid_t cpun)
3006 {
3007 	int bind_cpu = 0, i, hardbound = 0, iflag;
3008 	apic_irq_t *irq_ptr;
3009 
3010 	if (cpun == 0)
3011 		return (PSM_FAILURE);
3012 
3013 	iflag = intr_clear();
3014 	lock_set(&apic_ioapic_lock);
3015 	apic_cpus[cpun].aci_status &= ~APIC_CPU_INTR_ENABLE;
3016 	lock_clear(&apic_ioapic_lock);
3017 	intr_restore(iflag);
3018 	apic_cpus[cpun].aci_curipl = 0;
3019 	i = apic_min_device_irq;
3020 	for (; i <= apic_max_device_irq; i++) {
3021 		/*
3022 		 * If there are bound interrupts on this cpu, then
3023 		 * rebind them to other processors.
3024 		 */
3025 		if ((irq_ptr = apic_irq_table[i]) != NULL) {
3026 			ASSERT((irq_ptr->airq_temp_cpu == IRQ_UNBOUND) ||
3027 			    (irq_ptr->airq_temp_cpu == IRQ_UNINIT) ||
3028 			    ((irq_ptr->airq_temp_cpu & ~IRQ_USER_BOUND) <
3029 			    apic_nproc));
3030 
3031 			if (irq_ptr->airq_temp_cpu == (cpun | IRQ_USER_BOUND)) {
3032 				hardbound = 1;
3033 				continue;
3034 			}
3035 
3036 			if (irq_ptr->airq_temp_cpu == cpun) {
3037 				do {
3038 					apic_next_bind_cpu += 2;
3039 					bind_cpu = apic_next_bind_cpu / 2;
3040 					if (bind_cpu >= apic_nproc) {
3041 						apic_next_bind_cpu = 1;
3042 						bind_cpu = 0;
3043 
3044 					}
3045 				} while (apic_rebind_all(irq_ptr, bind_cpu, 1));
3046 			}
3047 		}
3048 	}
3049 	if (hardbound) {
3050 		cmn_err(CE_WARN, "Could not disable interrupts on %d"
3051 		    "due to user bound interrupts", cpun);
3052 		return (PSM_FAILURE);
3053 	}
3054 	else
3055 		return (PSM_SUCCESS);
3056 }
3057 
3058 static void
3059 apic_enable_intr(processorid_t cpun)
3060 {
3061 	int	i, iflag;
3062 	apic_irq_t *irq_ptr;
3063 
3064 	iflag = intr_clear();
3065 	lock_set(&apic_ioapic_lock);
3066 	apic_cpus[cpun].aci_status |= APIC_CPU_INTR_ENABLE;
3067 	lock_clear(&apic_ioapic_lock);
3068 	intr_restore(iflag);
3069 
3070 	i = apic_min_device_irq;
3071 	for (i = apic_min_device_irq; i <= apic_max_device_irq; i++) {
3072 		if ((irq_ptr = apic_irq_table[i]) != NULL) {
3073 			if ((irq_ptr->airq_cpu & ~IRQ_USER_BOUND) == cpun) {
3074 				(void) apic_rebind_all(irq_ptr,
3075 				    irq_ptr->airq_cpu, 1);
3076 			}
3077 		}
3078 	}
3079 }
3080 
3081 /*
3082  * apic_introp_xlate() replaces apic_translate_irq() and is
3083  * called only from apic_intr_ops().  With the new ADII framework,
3084  * the priority can no longer be retrived through i_ddi_get_intrspec().
3085  * It has to be passed in from the caller.
3086  */
3087 int
3088 apic_introp_xlate(dev_info_t *dip, struct intrspec *ispec, int type)
3089 {
3090 	char dev_type[16];
3091 	int dev_len, pci_irq, newirq, bustype, devid, busid, i;
3092 	int irqno = ispec->intrspec_vec;
3093 	ddi_acc_handle_t cfg_handle;
3094 	uchar_t ipin;
3095 	struct apic_io_intr *intrp;
3096 	iflag_t intr_flag;
3097 	APIC_HEADER	*hp;
3098 	MADT_INTERRUPT_OVERRIDE	*isop;
3099 	apic_irq_t *airqp;
3100 
3101 	DDI_INTR_IMPLDBG((CE_CONT, "apic_introp_xlate: dip=0x%p name=%s "
3102 	    "type=%d irqno=0x%x\n", (void *)dip, ddi_get_name(dip), type,
3103 	    irqno));
3104 
3105 	if (DDI_INTR_IS_MSI_OR_MSIX(type)) {
3106 		if ((airqp = apic_find_irq(dip, ispec, type)) != NULL)
3107 			return (apic_vector_to_irq[airqp->airq_vector]);
3108 		return (apic_setup_irq_table(dip, irqno, NULL, ispec,
3109 		    NULL, type));
3110 	}
3111 
3112 	bustype = 0;
3113 
3114 	/* check if we have already translated this irq */
3115 	mutex_enter(&airq_mutex);
3116 	newirq = apic_min_device_irq;
3117 	for (; newirq <= apic_max_device_irq; newirq++) {
3118 		airqp = apic_irq_table[newirq];
3119 		while (airqp) {
3120 			if ((airqp->airq_dip == dip) &&
3121 			    (airqp->airq_origirq == irqno) &&
3122 			    (airqp->airq_mps_intr_index != FREE_INDEX)) {
3123 
3124 				mutex_exit(&airq_mutex);
3125 				return (VIRTIRQ(newirq, airqp->airq_share_id));
3126 			}
3127 			airqp = airqp->airq_next;
3128 		}
3129 	}
3130 	mutex_exit(&airq_mutex);
3131 
3132 	if (apic_defconf)
3133 		goto defconf;
3134 
3135 	if ((dip == NULL) || (!apic_irq_translate && !apic_enable_acpi))
3136 		goto nonpci;
3137 
3138 	dev_len = sizeof (dev_type);
3139 	if (ddi_getlongprop_buf(DDI_DEV_T_ANY, ddi_get_parent(dip),
3140 	    DDI_PROP_DONTPASS, "device_type", (caddr_t)dev_type,
3141 	    &dev_len) != DDI_PROP_SUCCESS) {
3142 		goto nonpci;
3143 	}
3144 
3145 	if ((strcmp(dev_type, "pci") == 0) ||
3146 	    (strcmp(dev_type, "pciex") == 0)) {
3147 		/* pci device */
3148 		if (acpica_get_bdf(dip, &busid, &devid, NULL) != 0)
3149 			goto nonpci;
3150 		if (busid == 0 && apic_pci_bus_total == 1)
3151 			busid = (int)apic_single_pci_busid;
3152 
3153 		if (pci_config_setup(dip, &cfg_handle) != DDI_SUCCESS)
3154 			goto nonpci;
3155 		ipin = pci_config_get8(cfg_handle, PCI_CONF_IPIN) - PCI_INTA;
3156 		pci_config_teardown(&cfg_handle);
3157 		if (apic_enable_acpi && !apic_use_acpi_madt_only) {
3158 			if (apic_acpi_translate_pci_irq(dip, busid, devid,
3159 			    ipin, &pci_irq, &intr_flag) != ACPI_PSM_SUCCESS)
3160 				goto nonpci;
3161 
3162 			intr_flag.bustype = BUS_PCI;
3163 			if ((newirq = apic_setup_irq_table(dip, pci_irq, NULL,
3164 			    ispec, &intr_flag, type)) == -1)
3165 				goto nonpci;
3166 			return (newirq);
3167 		} else {
3168 			pci_irq = ((devid & 0x1f) << 2) | (ipin & 0x3);
3169 			if ((intrp = apic_find_io_intr_w_busid(pci_irq, busid))
3170 			    == NULL) {
3171 				if ((pci_irq = apic_handle_pci_pci_bridge(dip,
3172 				    devid, ipin, &intrp)) == -1)
3173 					goto nonpci;
3174 			}
3175 			if ((newirq = apic_setup_irq_table(dip, pci_irq, intrp,
3176 			    ispec, NULL, type)) == -1)
3177 				goto nonpci;
3178 			return (newirq);
3179 		}
3180 	} else if (strcmp(dev_type, "isa") == 0)
3181 		bustype = BUS_ISA;
3182 	else if (strcmp(dev_type, "eisa") == 0)
3183 		bustype = BUS_EISA;
3184 
3185 nonpci:
3186 	if (apic_enable_acpi && !apic_use_acpi_madt_only) {
3187 		/* search iso entries first */
3188 		if (acpi_iso_cnt != 0) {
3189 			hp = (APIC_HEADER *)acpi_isop;
3190 			i = 0;
3191 			while (i < acpi_iso_cnt) {
3192 				if (hp->Type == APIC_XRUPT_OVERRIDE) {
3193 					isop = (MADT_INTERRUPT_OVERRIDE *)hp;
3194 					if (isop->Bus == 0 &&
3195 					    isop->Source == irqno) {
3196 						newirq = isop->Interrupt;
3197 						intr_flag.intr_po =
3198 						    isop->Polarity;
3199 						intr_flag.intr_el =
3200 						    isop->TriggerMode;
3201 						intr_flag.bustype = BUS_ISA;
3202 
3203 						return (apic_setup_irq_table(
3204 						    dip, newirq, NULL, ispec,
3205 						    &intr_flag, type));
3206 
3207 					}
3208 					i++;
3209 				}
3210 				hp = (APIC_HEADER *)(((char *)hp) +
3211 				    hp->Length);
3212 			}
3213 		}
3214 		intr_flag.intr_po = INTR_PO_ACTIVE_HIGH;
3215 		intr_flag.intr_el = INTR_EL_EDGE;
3216 		intr_flag.bustype = BUS_ISA;
3217 		return (apic_setup_irq_table(dip, irqno, NULL, ispec,
3218 		    &intr_flag, type));
3219 	} else {
3220 		if (bustype == 0)
3221 			bustype = eisa_level_intr_mask ? BUS_EISA : BUS_ISA;
3222 		for (i = 0; i < 2; i++) {
3223 			if (((busid = apic_find_bus_id(bustype)) != -1) &&
3224 			    ((intrp = apic_find_io_intr_w_busid(irqno, busid))
3225 			    != NULL)) {
3226 				if ((newirq = apic_setup_irq_table(dip, irqno,
3227 				    intrp, ispec, NULL, type)) != -1) {
3228 					return (newirq);
3229 				}
3230 				goto defconf;
3231 			}
3232 			bustype = (bustype == BUS_EISA) ? BUS_ISA : BUS_EISA;
3233 		}
3234 	}
3235 
3236 /* MPS default configuration */
3237 defconf:
3238 	newirq = apic_setup_irq_table(dip, irqno, NULL, ispec, NULL, type);
3239 	if (newirq == -1)
3240 		return (newirq);
3241 	ASSERT(IRQINDEX(newirq) == irqno);
3242 	ASSERT(apic_irq_table[irqno]);
3243 	return (newirq);
3244 }
3245 
3246 
3247 
3248 
3249 
3250 
3251 /*
3252  * On machines with PCI-PCI bridges, a device behind a PCI-PCI bridge
3253  * needs special handling.  We may need to chase up the device tree,
3254  * using the PCI-PCI Bridge specification's "rotating IPIN assumptions",
3255  * to find the IPIN at the root bus that relates to the IPIN on the
3256  * subsidiary bus (for ACPI or MP).  We may, however, have an entry
3257  * in the MP table or the ACPI namespace for this device itself.
3258  * We handle both cases in the search below.
3259  */
3260 /* this is the non-acpi version */
3261 static int
3262 apic_handle_pci_pci_bridge(dev_info_t *idip, int child_devno, int child_ipin,
3263 			struct apic_io_intr **intrp)
3264 {
3265 	dev_info_t *dipp, *dip;
3266 	int pci_irq;
3267 	ddi_acc_handle_t cfg_handle;
3268 	int bridge_devno, bridge_bus;
3269 	int ipin;
3270 
3271 	dip = idip;
3272 
3273 	/*CONSTCOND*/
3274 	while (1) {
3275 		if ((dipp = ddi_get_parent(dip)) == (dev_info_t *)NULL)
3276 			return (-1);
3277 		if ((pci_config_setup(dipp, &cfg_handle) == DDI_SUCCESS) &&
3278 		    (pci_config_get8(cfg_handle, PCI_CONF_BASCLASS) ==
3279 		    PCI_CLASS_BRIDGE) && (pci_config_get8(cfg_handle,
3280 		    PCI_CONF_SUBCLASS) == PCI_BRIDGE_PCI)) {
3281 			pci_config_teardown(&cfg_handle);
3282 			if (acpica_get_bdf(dipp, &bridge_bus, &bridge_devno,
3283 			    NULL) != 0)
3284 				return (-1);
3285 			/*
3286 			 * This is the rotating scheme that Compaq is using
3287 			 * and documented in the pci to pci spec.  Also, if
3288 			 * the pci to pci bridge is behind another pci to
3289 			 * pci bridge, then it need to keep transversing
3290 			 * up until an interrupt entry is found or reach
3291 			 * the top of the tree
3292 			 */
3293 			ipin = (child_devno + child_ipin) % PCI_INTD;
3294 				if (bridge_bus == 0 && apic_pci_bus_total == 1)
3295 					bridge_bus = (int)apic_single_pci_busid;
3296 				pci_irq = ((bridge_devno & 0x1f) << 2) |
3297 				    (ipin & 0x3);
3298 				if ((*intrp = apic_find_io_intr_w_busid(pci_irq,
3299 				    bridge_bus)) != NULL) {
3300 					return (pci_irq);
3301 				}
3302 			dip = dipp;
3303 			child_devno = bridge_devno;
3304 			child_ipin = ipin;
3305 		} else
3306 			return (-1);
3307 	}
3308 	/*LINTED: function will not fall off the bottom */
3309 }
3310 
3311 
3312 
3313 
3314 static uchar_t
3315 acpi_find_ioapic(int irq)
3316 {
3317 	int i;
3318 
3319 	for (i = 0; i < apic_io_max; i++) {
3320 		if (irq >= apic_io_vectbase[i] && irq <= apic_io_vectend[i])
3321 			return (i);
3322 	}
3323 	return (0xFF);	/* shouldn't happen */
3324 }
3325 
3326 /*
3327  * See if two irqs are compatible for sharing a vector.
3328  * Currently we only support sharing of PCI devices.
3329  */
3330 static int
3331 acpi_intr_compatible(iflag_t iflag1, iflag_t iflag2)
3332 {
3333 	uint_t	level1, po1;
3334 	uint_t	level2, po2;
3335 
3336 	/* Assume active high by default */
3337 	po1 = 0;
3338 	po2 = 0;
3339 
3340 	if (iflag1.bustype != iflag2.bustype || iflag1.bustype != BUS_PCI)
3341 		return (0);
3342 
3343 	if (iflag1.intr_el == INTR_EL_CONFORM)
3344 		level1 = AV_LEVEL;
3345 	else
3346 		level1 = (iflag1.intr_el == INTR_EL_LEVEL) ? AV_LEVEL : 0;
3347 
3348 	if (level1 && ((iflag1.intr_po == INTR_PO_ACTIVE_LOW) ||
3349 	    (iflag1.intr_po == INTR_PO_CONFORM)))
3350 		po1 = AV_ACTIVE_LOW;
3351 
3352 	if (iflag2.intr_el == INTR_EL_CONFORM)
3353 		level2 = AV_LEVEL;
3354 	else
3355 		level2 = (iflag2.intr_el == INTR_EL_LEVEL) ? AV_LEVEL : 0;
3356 
3357 	if (level2 && ((iflag2.intr_po == INTR_PO_ACTIVE_LOW) ||
3358 	    (iflag2.intr_po == INTR_PO_CONFORM)))
3359 		po2 = AV_ACTIVE_LOW;
3360 
3361 	if ((level1 == level2) && (po1 == po2))
3362 		return (1);
3363 
3364 	return (0);
3365 }
3366 
3367 /*
3368  * Attempt to share vector with someone else
3369  */
3370 static int
3371 apic_share_vector(int irqno, iflag_t *intr_flagp, short intr_index, int ipl,
3372 	uchar_t ioapicindex, uchar_t ipin, apic_irq_t **irqptrp)
3373 {
3374 #ifdef DEBUG
3375 	apic_irq_t *tmpirqp = NULL;
3376 #endif /* DEBUG */
3377 	apic_irq_t *irqptr, dummyirq;
3378 	int	newirq, chosen_irq = -1, share = 127;
3379 	int	lowest, highest, i;
3380 	uchar_t	share_id;
3381 
3382 	DDI_INTR_IMPLDBG((CE_CONT, "apic_share_vector: irqno=0x%x "
3383 	    "intr_index=0x%x ipl=0x%x\n", irqno, intr_index, ipl));
3384 
3385 	highest = apic_ipltopri[ipl] + APIC_VECTOR_MASK;
3386 	lowest = apic_ipltopri[ipl-1] + APIC_VECTOR_PER_IPL;
3387 
3388 	if (highest < lowest) /* Both ipl and ipl-1 map to same pri */
3389 		lowest -= APIC_VECTOR_PER_IPL;
3390 	dummyirq.airq_mps_intr_index = intr_index;
3391 	dummyirq.airq_ioapicindex = ioapicindex;
3392 	dummyirq.airq_intin_no = ipin;
3393 	if (intr_flagp)
3394 		dummyirq.airq_iflag = *intr_flagp;
3395 	apic_record_rdt_entry(&dummyirq, irqno);
3396 	for (i = lowest; i <= highest; i++) {
3397 		newirq = apic_vector_to_irq[i];
3398 		if (newirq == APIC_RESV_IRQ)
3399 			continue;
3400 		irqptr = apic_irq_table[newirq];
3401 
3402 		if ((dummyirq.airq_rdt_entry & 0xFF00) !=
3403 		    (irqptr->airq_rdt_entry & 0xFF00))
3404 			/* not compatible */
3405 			continue;
3406 
3407 		if (irqptr->airq_share < share) {
3408 			share = irqptr->airq_share;
3409 			chosen_irq = newirq;
3410 		}
3411 	}
3412 	if (chosen_irq != -1) {
3413 		/*
3414 		 * Assign a share id which is free or which is larger
3415 		 * than the largest one.
3416 		 */
3417 		share_id = 1;
3418 		mutex_enter(&airq_mutex);
3419 		irqptr = apic_irq_table[chosen_irq];
3420 		while (irqptr) {
3421 			if (irqptr->airq_mps_intr_index == FREE_INDEX) {
3422 				share_id = irqptr->airq_share_id;
3423 				break;
3424 			}
3425 			if (share_id <= irqptr->airq_share_id)
3426 				share_id = irqptr->airq_share_id + 1;
3427 #ifdef DEBUG
3428 			tmpirqp = irqptr;
3429 #endif /* DEBUG */
3430 			irqptr = irqptr->airq_next;
3431 		}
3432 		if (!irqptr) {
3433 			irqptr = kmem_zalloc(sizeof (apic_irq_t), KM_SLEEP);
3434 			irqptr->airq_temp_cpu = IRQ_UNINIT;
3435 			irqptr->airq_next =
3436 			    apic_irq_table[chosen_irq]->airq_next;
3437 			apic_irq_table[chosen_irq]->airq_next = irqptr;
3438 #ifdef	DEBUG
3439 			tmpirqp = apic_irq_table[chosen_irq];
3440 #endif /* DEBUG */
3441 		}
3442 		irqptr->airq_mps_intr_index = intr_index;
3443 		irqptr->airq_ioapicindex = ioapicindex;
3444 		irqptr->airq_intin_no = ipin;
3445 		if (intr_flagp)
3446 			irqptr->airq_iflag = *intr_flagp;
3447 		irqptr->airq_vector = apic_irq_table[chosen_irq]->airq_vector;
3448 		irqptr->airq_share_id = share_id;
3449 		apic_record_rdt_entry(irqptr, irqno);
3450 		*irqptrp = irqptr;
3451 #ifdef	DEBUG
3452 		/* shuffle the pointers to test apic_delspl path */
3453 		if (tmpirqp) {
3454 			tmpirqp->airq_next = irqptr->airq_next;
3455 			irqptr->airq_next = apic_irq_table[chosen_irq];
3456 			apic_irq_table[chosen_irq] = irqptr;
3457 		}
3458 #endif /* DEBUG */
3459 		mutex_exit(&airq_mutex);
3460 		return (VIRTIRQ(chosen_irq, share_id));
3461 	}
3462 	return (-1);
3463 }
3464 
3465 /*
3466  *
3467  */
3468 static int
3469 apic_setup_irq_table(dev_info_t *dip, int irqno, struct apic_io_intr *intrp,
3470     struct intrspec *ispec, iflag_t *intr_flagp, int type)
3471 {
3472 	int origirq = ispec->intrspec_vec;
3473 	uchar_t ipl = ispec->intrspec_pri;
3474 	int	newirq, intr_index;
3475 	uchar_t	ipin, ioapic, ioapicindex, vector;
3476 	apic_irq_t *irqptr;
3477 	major_t	major;
3478 	dev_info_t	*sdip;
3479 
3480 	DDI_INTR_IMPLDBG((CE_CONT, "apic_setup_irq_table: dip=0x%p type=%d "
3481 	    "irqno=0x%x origirq=0x%x\n", (void *)dip, type, irqno, origirq));
3482 
3483 	ASSERT(ispec != NULL);
3484 
3485 	major =  (dip != NULL) ? ddi_name_to_major(ddi_get_name(dip)) : 0;
3486 
3487 	if (DDI_INTR_IS_MSI_OR_MSIX(type)) {
3488 		/* MSI/X doesn't need to setup ioapic stuffs */
3489 		ioapicindex = 0xff;
3490 		ioapic = 0xff;
3491 		ipin = (uchar_t)0xff;
3492 		intr_index = (type == DDI_INTR_TYPE_MSI) ? MSI_INDEX :
3493 		    MSIX_INDEX;
3494 		mutex_enter(&airq_mutex);
3495 		if ((irqno = apic_allocate_irq(apic_first_avail_irq)) == -1) {
3496 			mutex_exit(&airq_mutex);
3497 			/* need an irq for MSI/X to index into autovect[] */
3498 			cmn_err(CE_WARN, "No interrupt irq: %s instance %d",
3499 			    ddi_get_name(dip), ddi_get_instance(dip));
3500 			return (-1);
3501 		}
3502 		mutex_exit(&airq_mutex);
3503 
3504 	} else if (intrp != NULL) {
3505 		intr_index = (int)(intrp - apic_io_intrp);
3506 		ioapic = intrp->intr_destid;
3507 		ipin = intrp->intr_destintin;
3508 		/* Find ioapicindex. If destid was ALL, we will exit with 0. */
3509 		for (ioapicindex = apic_io_max - 1; ioapicindex; ioapicindex--)
3510 			if (apic_io_id[ioapicindex] == ioapic)
3511 				break;
3512 		ASSERT((ioapic == apic_io_id[ioapicindex]) ||
3513 		    (ioapic == INTR_ALL_APIC));
3514 
3515 		/* check whether this intin# has been used by another irqno */
3516 		if ((newirq = apic_find_intin(ioapicindex, ipin)) != -1) {
3517 			return (newirq);
3518 		}
3519 
3520 	} else if (intr_flagp != NULL) {
3521 		/* ACPI case */
3522 		intr_index = ACPI_INDEX;
3523 		ioapicindex = acpi_find_ioapic(irqno);
3524 		ASSERT(ioapicindex != 0xFF);
3525 		ioapic = apic_io_id[ioapicindex];
3526 		ipin = irqno - apic_io_vectbase[ioapicindex];
3527 		if (apic_irq_table[irqno] &&
3528 		    apic_irq_table[irqno]->airq_mps_intr_index == ACPI_INDEX) {
3529 			ASSERT(apic_irq_table[irqno]->airq_intin_no == ipin &&
3530 			    apic_irq_table[irqno]->airq_ioapicindex ==
3531 			    ioapicindex);
3532 			return (irqno);
3533 		}
3534 
3535 	} else {
3536 		/* default configuration */
3537 		ioapicindex = 0;
3538 		ioapic = apic_io_id[ioapicindex];
3539 		ipin = (uchar_t)irqno;
3540 		intr_index = DEFAULT_INDEX;
3541 	}
3542 
3543 	if (ispec == NULL) {
3544 		APIC_VERBOSE_IOAPIC((CE_WARN, "No intrspec for irqno = %x\n",
3545 		    irqno));
3546 	} else if ((vector = apic_allocate_vector(ipl, irqno, 0)) == 0) {
3547 		if ((newirq = apic_share_vector(irqno, intr_flagp, intr_index,
3548 		    ipl, ioapicindex, ipin, &irqptr)) != -1) {
3549 			irqptr->airq_ipl = ipl;
3550 			irqptr->airq_origirq = (uchar_t)origirq;
3551 			irqptr->airq_dip = dip;
3552 			irqptr->airq_major = major;
3553 			sdip = apic_irq_table[IRQINDEX(newirq)]->airq_dip;
3554 			/* This is OK to do really */
3555 			if (sdip == NULL) {
3556 				cmn_err(CE_WARN, "Sharing vectors: %s"
3557 				    " instance %d and SCI",
3558 				    ddi_get_name(dip), ddi_get_instance(dip));
3559 			} else {
3560 				cmn_err(CE_WARN, "Sharing vectors: %s"
3561 				    " instance %d and %s instance %d",
3562 				    ddi_get_name(sdip), ddi_get_instance(sdip),
3563 				    ddi_get_name(dip), ddi_get_instance(dip));
3564 			}
3565 			return (newirq);
3566 		}
3567 		/* try high priority allocation now  that share has failed */
3568 		if ((vector = apic_allocate_vector(ipl, irqno, 1)) == 0) {
3569 			cmn_err(CE_WARN, "No interrupt vector: %s instance %d",
3570 			    ddi_get_name(dip), ddi_get_instance(dip));
3571 			return (-1);
3572 		}
3573 	}
3574 
3575 	mutex_enter(&airq_mutex);
3576 	if (apic_irq_table[irqno] == NULL) {
3577 		irqptr = kmem_zalloc(sizeof (apic_irq_t), KM_SLEEP);
3578 		irqptr->airq_temp_cpu = IRQ_UNINIT;
3579 		apic_irq_table[irqno] = irqptr;
3580 	} else {
3581 		irqptr = apic_irq_table[irqno];
3582 		if (irqptr->airq_mps_intr_index != FREE_INDEX) {
3583 			/*
3584 			 * The slot is used by another irqno, so allocate
3585 			 * a free irqno for this interrupt
3586 			 */
3587 			newirq = apic_allocate_irq(apic_first_avail_irq);
3588 			if (newirq == -1) {
3589 				mutex_exit(&airq_mutex);
3590 				return (-1);
3591 			}
3592 			irqno = newirq;
3593 			irqptr = apic_irq_table[irqno];
3594 			if (irqptr == NULL) {
3595 				irqptr = kmem_zalloc(sizeof (apic_irq_t),
3596 				    KM_SLEEP);
3597 				irqptr->airq_temp_cpu = IRQ_UNINIT;
3598 				apic_irq_table[irqno] = irqptr;
3599 			}
3600 			apic_modify_vector(vector, newirq);
3601 		}
3602 	}
3603 	apic_max_device_irq = max(irqno, apic_max_device_irq);
3604 	apic_min_device_irq = min(irqno, apic_min_device_irq);
3605 	mutex_exit(&airq_mutex);
3606 	irqptr->airq_ioapicindex = ioapicindex;
3607 	irqptr->airq_intin_no = ipin;
3608 	irqptr->airq_ipl = ipl;
3609 	irqptr->airq_vector = vector;
3610 	irqptr->airq_origirq = (uchar_t)origirq;
3611 	irqptr->airq_share_id = 0;
3612 	irqptr->airq_mps_intr_index = (short)intr_index;
3613 	irqptr->airq_dip = dip;
3614 	irqptr->airq_major = major;
3615 	irqptr->airq_cpu = apic_bind_intr(dip, irqno, ioapic, ipin);
3616 	if (intr_flagp)
3617 		irqptr->airq_iflag = *intr_flagp;
3618 
3619 	if (!DDI_INTR_IS_MSI_OR_MSIX(type)) {
3620 		/* setup I/O APIC entry for non-MSI/X interrupts */
3621 		apic_record_rdt_entry(irqptr, irqno);
3622 	}
3623 	return (irqno);
3624 }
3625 
3626 /*
3627  * return the cpu to which this intr should be bound.
3628  * Check properties or any other mechanism to see if user wants it
3629  * bound to a specific CPU. If so, return the cpu id with high bit set.
3630  * If not, use the policy to choose a cpu and return the id.
3631  */
3632 uchar_t
3633 apic_bind_intr(dev_info_t *dip, int irq, uchar_t ioapicid, uchar_t intin)
3634 {
3635 	int	instance, instno, prop_len, bind_cpu, count;
3636 	uint_t	i, rc;
3637 	uchar_t	cpu;
3638 	major_t	major;
3639 	char	*name, *drv_name, *prop_val, *cptr;
3640 	char	prop_name[32];
3641 
3642 
3643 	if (apic_intr_policy == INTR_LOWEST_PRIORITY)
3644 		return (IRQ_UNBOUND);
3645 
3646 	drv_name = NULL;
3647 	rc = DDI_PROP_NOT_FOUND;
3648 	major = (major_t)-1;
3649 	if (dip != NULL) {
3650 		name = ddi_get_name(dip);
3651 		major = ddi_name_to_major(name);
3652 		drv_name = ddi_major_to_name(major);
3653 		instance = ddi_get_instance(dip);
3654 		if (apic_intr_policy == INTR_ROUND_ROBIN_WITH_AFFINITY) {
3655 			i = apic_min_device_irq;
3656 			for (; i <= apic_max_device_irq; i++) {
3657 
3658 				if ((i == irq) || (apic_irq_table[i] == NULL) ||
3659 				    (apic_irq_table[i]->airq_mps_intr_index
3660 				    == FREE_INDEX))
3661 					continue;
3662 
3663 				if ((apic_irq_table[i]->airq_major == major) &&
3664 				    (!(apic_irq_table[i]->airq_cpu &
3665 				    IRQ_USER_BOUND))) {
3666 
3667 					cpu = apic_irq_table[i]->airq_cpu;
3668 
3669 					cmn_err(CE_CONT,
3670 					    "!pcplusmp: %s (%s) instance #%d "
3671 					    "vector 0x%x ioapic 0x%x "
3672 					    "intin 0x%x is bound to cpu %d\n",
3673 					    name, drv_name, instance, irq,
3674 					    ioapicid, intin, cpu);
3675 					return (cpu);
3676 				}
3677 			}
3678 		}
3679 		/*
3680 		 * search for "drvname"_intpt_bind_cpus property first, the
3681 		 * syntax of the property should be "a[,b,c,...]" where
3682 		 * instance 0 binds to cpu a, instance 1 binds to cpu b,
3683 		 * instance 3 binds to cpu c...
3684 		 * ddi_getlongprop() will search /option first, then /
3685 		 * if "drvname"_intpt_bind_cpus doesn't exist, then find
3686 		 * intpt_bind_cpus property.  The syntax is the same, and
3687 		 * it applies to all the devices if its "drvname" specific
3688 		 * property doesn't exist
3689 		 */
3690 		(void) strcpy(prop_name, drv_name);
3691 		(void) strcat(prop_name, "_intpt_bind_cpus");
3692 		rc = ddi_getlongprop(DDI_DEV_T_ANY, dip, 0, prop_name,
3693 		    (caddr_t)&prop_val, &prop_len);
3694 		if (rc != DDI_PROP_SUCCESS) {
3695 			rc = ddi_getlongprop(DDI_DEV_T_ANY, dip, 0,
3696 			    "intpt_bind_cpus", (caddr_t)&prop_val, &prop_len);
3697 		}
3698 	}
3699 	if (rc == DDI_PROP_SUCCESS) {
3700 		for (i = count = 0; i < (prop_len - 1); i++)
3701 			if (prop_val[i] == ',')
3702 				count++;
3703 		if (prop_val[i-1] != ',')
3704 			count++;
3705 		/*
3706 		 * if somehow the binding instances defined in the
3707 		 * property are not enough for this instno., then
3708 		 * reuse the pattern for the next instance until
3709 		 * it reaches the requested instno
3710 		 */
3711 		instno = instance % count;
3712 		i = 0;
3713 		cptr = prop_val;
3714 		while (i < instno)
3715 			if (*cptr++ == ',')
3716 				i++;
3717 		bind_cpu = stoi(&cptr);
3718 		kmem_free(prop_val, prop_len);
3719 		/* if specific cpu is bogus, then default to cpu 0 */
3720 		if (bind_cpu >= apic_nproc) {
3721 			cmn_err(CE_WARN, "pcplusmp: %s=%s: CPU %d not present",
3722 			    prop_name, prop_val, bind_cpu);
3723 			bind_cpu = 0;
3724 		} else {
3725 			/* indicate that we are bound at user request */
3726 			bind_cpu |= IRQ_USER_BOUND;
3727 		}
3728 		/*
3729 		 * no need to check apic_cpus[].aci_status, if specific cpu is
3730 		 * not up, then post_cpu_start will handle it.
3731 		 */
3732 	} else {
3733 		/*
3734 		 * We change bind_cpu only for every two calls
3735 		 * as most drivers still do 2 add_intrs for every
3736 		 * interrupt
3737 		 */
3738 		bind_cpu = (apic_next_bind_cpu++) / 2;
3739 		if (bind_cpu >= apic_nproc) {
3740 			apic_next_bind_cpu = 1;
3741 			bind_cpu = 0;
3742 		}
3743 	}
3744 	if (drv_name != NULL)
3745 		cmn_err(CE_CONT, "!pcplusmp: %s (%s) instance %d "
3746 		    "vector 0x%x ioapic 0x%x intin 0x%x is bound to cpu %d\n",
3747 		    name, drv_name, instance,
3748 		    irq, ioapicid, intin, bind_cpu & ~IRQ_USER_BOUND);
3749 	else
3750 		cmn_err(CE_CONT, "!pcplusmp: "
3751 		    "vector 0x%x ioapic 0x%x intin 0x%x is bound to cpu %d\n",
3752 		    irq, ioapicid, intin, bind_cpu & ~IRQ_USER_BOUND);
3753 
3754 	return ((uchar_t)bind_cpu);
3755 }
3756 
3757 static struct apic_io_intr *
3758 apic_find_io_intr_w_busid(int irqno, int busid)
3759 {
3760 	struct	apic_io_intr	*intrp;
3761 
3762 	/*
3763 	 * It can have more than 1 entry with same source bus IRQ,
3764 	 * but unique with the source bus id
3765 	 */
3766 	intrp = apic_io_intrp;
3767 	if (intrp != NULL) {
3768 		while (intrp->intr_entry == APIC_IO_INTR_ENTRY) {
3769 			if (intrp->intr_irq == irqno &&
3770 			    intrp->intr_busid == busid &&
3771 			    intrp->intr_type == IO_INTR_INT)
3772 				return (intrp);
3773 			intrp++;
3774 		}
3775 	}
3776 	APIC_VERBOSE_IOAPIC((CE_NOTE, "Did not find io intr for irqno:"
3777 	    "busid %x:%x\n", irqno, busid));
3778 	return ((struct apic_io_intr *)NULL);
3779 }
3780 
3781 
3782 struct mps_bus_info {
3783 	char	*bus_name;
3784 	int	bus_id;
3785 } bus_info_array[] = {
3786 	"ISA ", BUS_ISA,
3787 	"PCI ", BUS_PCI,
3788 	"EISA ", BUS_EISA,
3789 	"XPRESS", BUS_XPRESS,
3790 	"PCMCIA", BUS_PCMCIA,
3791 	"VL ", BUS_VL,
3792 	"CBUS ", BUS_CBUS,
3793 	"CBUSII", BUS_CBUSII,
3794 	"FUTURE", BUS_FUTURE,
3795 	"INTERN", BUS_INTERN,
3796 	"MBI ", BUS_MBI,
3797 	"MBII ", BUS_MBII,
3798 	"MPI ", BUS_MPI,
3799 	"MPSA ", BUS_MPSA,
3800 	"NUBUS ", BUS_NUBUS,
3801 	"TC ", BUS_TC,
3802 	"VME ", BUS_VME
3803 };
3804 
3805 static int
3806 apic_find_bus_type(char *bus)
3807 {
3808 	int	i = 0;
3809 
3810 	for (; i < sizeof (bus_info_array)/sizeof (struct mps_bus_info); i++)
3811 		if (strncmp(bus, bus_info_array[i].bus_name,
3812 		    strlen(bus_info_array[i].bus_name)) == 0)
3813 			return (bus_info_array[i].bus_id);
3814 	APIC_VERBOSE_IOAPIC((CE_WARN, "Did not find bus type for bus %s", bus));
3815 	return (0);
3816 }
3817 
3818 static int
3819 apic_find_bus(int busid)
3820 {
3821 	struct	apic_bus	*busp;
3822 
3823 	busp = apic_busp;
3824 	while (busp->bus_entry == APIC_BUS_ENTRY) {
3825 		if (busp->bus_id == busid)
3826 			return (apic_find_bus_type((char *)&busp->bus_str1));
3827 		busp++;
3828 	}
3829 	APIC_VERBOSE_IOAPIC((CE_WARN, "Did not find bus for bus id %x", busid));
3830 	return (0);
3831 }
3832 
3833 static int
3834 apic_find_bus_id(int bustype)
3835 {
3836 	struct	apic_bus	*busp;
3837 
3838 	busp = apic_busp;
3839 	while (busp->bus_entry == APIC_BUS_ENTRY) {
3840 		if (apic_find_bus_type((char *)&busp->bus_str1) == bustype)
3841 			return (busp->bus_id);
3842 		busp++;
3843 	}
3844 	APIC_VERBOSE_IOAPIC((CE_WARN, "Did not find bus id for bustype %x",
3845 	    bustype));
3846 	return (-1);
3847 }
3848 
3849 /*
3850  * Check if a particular irq need to be reserved for any io_intr
3851  */
3852 static struct apic_io_intr *
3853 apic_find_io_intr(int irqno)
3854 {
3855 	struct	apic_io_intr	*intrp;
3856 
3857 	intrp = apic_io_intrp;
3858 	if (intrp != NULL) {
3859 		while (intrp->intr_entry == APIC_IO_INTR_ENTRY) {
3860 			if (intrp->intr_irq == irqno &&
3861 			    intrp->intr_type == IO_INTR_INT)
3862 				return (intrp);
3863 			intrp++;
3864 		}
3865 	}
3866 	return ((struct apic_io_intr *)NULL);
3867 }
3868 
3869 /*
3870  * Check if the given ioapicindex intin combination has already been assigned
3871  * an irq. If so return irqno. Else -1
3872  */
3873 static int
3874 apic_find_intin(uchar_t ioapic, uchar_t intin)
3875 {
3876 	apic_irq_t *irqptr;
3877 	int	i;
3878 
3879 	/* find ioapic and intin in the apic_irq_table[] and return the index */
3880 	for (i = apic_min_device_irq; i <= apic_max_device_irq; i++) {
3881 		irqptr = apic_irq_table[i];
3882 		while (irqptr) {
3883 			if ((irqptr->airq_mps_intr_index >= 0) &&
3884 			    (irqptr->airq_intin_no == intin) &&
3885 			    (irqptr->airq_ioapicindex == ioapic)) {
3886 				APIC_VERBOSE_IOAPIC((CE_NOTE, "!Found irq "
3887 				    "entry for ioapic:intin %x:%x "
3888 				    "shared interrupts ?", ioapic, intin));
3889 				return (i);
3890 			}
3891 			irqptr = irqptr->airq_next;
3892 		}
3893 	}
3894 	return (-1);
3895 }
3896 
3897 int
3898 apic_allocate_irq(int irq)
3899 {
3900 	int	freeirq, i;
3901 
3902 	if ((freeirq = apic_find_free_irq(irq, (APIC_RESV_IRQ - 1))) == -1)
3903 		if ((freeirq = apic_find_free_irq(APIC_FIRST_FREE_IRQ,
3904 		    (irq - 1))) == -1) {
3905 			/*
3906 			 * if BIOS really defines every single irq in the mps
3907 			 * table, then don't worry about conflicting with
3908 			 * them, just use any free slot in apic_irq_table
3909 			 */
3910 			for (i = APIC_FIRST_FREE_IRQ; i < APIC_RESV_IRQ; i++) {
3911 				if ((apic_irq_table[i] == NULL) ||
3912 				    apic_irq_table[i]->airq_mps_intr_index ==
3913 				    FREE_INDEX) {
3914 				freeirq = i;
3915 				break;
3916 			}
3917 		}
3918 		if (freeirq == -1) {
3919 			/* This shouldn't happen, but just in case */
3920 			cmn_err(CE_WARN, "pcplusmp: NO available IRQ");
3921 			return (-1);
3922 		}
3923 	}
3924 	if (apic_irq_table[freeirq] == NULL) {
3925 		apic_irq_table[freeirq] =
3926 		    kmem_zalloc(sizeof (apic_irq_t), KM_NOSLEEP);
3927 		if (apic_irq_table[freeirq] == NULL) {
3928 			cmn_err(CE_WARN, "pcplusmp: NO memory to allocate IRQ");
3929 			return (-1);
3930 		}
3931 		apic_irq_table[freeirq]->airq_mps_intr_index = FREE_INDEX;
3932 	}
3933 	return (freeirq);
3934 }
3935 
3936 static int
3937 apic_find_free_irq(int start, int end)
3938 {
3939 	int	i;
3940 
3941 	for (i = start; i <= end; i++)
3942 		/* Check if any I/O entry needs this IRQ */
3943 		if (apic_find_io_intr(i) == NULL) {
3944 			/* Then see if it is free */
3945 			if ((apic_irq_table[i] == NULL) ||
3946 			    (apic_irq_table[i]->airq_mps_intr_index ==
3947 			    FREE_INDEX)) {
3948 				return (i);
3949 			}
3950 		}
3951 	return (-1);
3952 }
3953 
3954 /*
3955  * Allocate a free vector for irq at ipl. Takes care of merging of multiple
3956  * IPLs into a single APIC level as well as stretching some IPLs onto multiple
3957  * levels. APIC_HI_PRI_VECTS interrupts are reserved for high priority
3958  * requests and allocated only when pri is set.
3959  */
3960 static uchar_t
3961 apic_allocate_vector(int ipl, int irq, int pri)
3962 {
3963 	int	lowest, highest, i;
3964 
3965 	highest = apic_ipltopri[ipl] + APIC_VECTOR_MASK;
3966 	lowest = apic_ipltopri[ipl - 1] + APIC_VECTOR_PER_IPL;
3967 
3968 	if (highest < lowest) /* Both ipl and ipl - 1 map to same pri */
3969 		lowest -= APIC_VECTOR_PER_IPL;
3970 
3971 #ifdef	DEBUG
3972 	if (apic_restrict_vector)	/* for testing shared interrupt logic */
3973 		highest = lowest + apic_restrict_vector + APIC_HI_PRI_VECTS;
3974 #endif /* DEBUG */
3975 	if (pri == 0)
3976 		highest -= APIC_HI_PRI_VECTS;
3977 
3978 	for (i = lowest; i < highest; i++) {
3979 		if ((i == T_FASTTRAP) || (i == APIC_SPUR_INTR) ||
3980 			(i == T_SYSCALLINT) || (i == T_DTRACE_PROBE) ||
3981 			(i == T_DTRACE_RET))
3982 			continue;
3983 		if (apic_vector_to_irq[i] == APIC_RESV_IRQ) {
3984 			apic_vector_to_irq[i] = (uchar_t)irq;
3985 			return (i);
3986 		}
3987 	}
3988 
3989 	return (0);
3990 }
3991 
3992 static void
3993 apic_modify_vector(uchar_t vector, int irq)
3994 {
3995 	apic_vector_to_irq[vector] = (uchar_t)irq;
3996 }
3997 
3998 /*
3999  * Mark vector as being in the process of being deleted. Interrupts
4000  * may still come in on some CPU. The moment an interrupt comes with
4001  * the new vector, we know we can free the old one. Called only from
4002  * addspl and delspl with interrupts disabled. Because an interrupt
4003  * can be shared, but no interrupt from either device may come in,
4004  * we also use a timeout mechanism, which we arbitrarily set to
4005  * apic_revector_timeout microseconds.
4006  */
4007 static void
4008 apic_mark_vector(uchar_t oldvector, uchar_t newvector)
4009 {
4010 	int iflag = intr_clear();
4011 	lock_set(&apic_revector_lock);
4012 	if (!apic_oldvec_to_newvec) {
4013 		apic_oldvec_to_newvec =
4014 		    kmem_zalloc(sizeof (newvector) * APIC_MAX_VECTOR * 2,
4015 		    KM_NOSLEEP);
4016 
4017 		if (!apic_oldvec_to_newvec) {
4018 			/*
4019 			 * This failure is not catastrophic.
4020 			 * But, the oldvec will never be freed.
4021 			 */
4022 			apic_error |= APIC_ERR_MARK_VECTOR_FAIL;
4023 			lock_clear(&apic_revector_lock);
4024 			intr_restore(iflag);
4025 			return;
4026 		}
4027 		apic_newvec_to_oldvec = &apic_oldvec_to_newvec[APIC_MAX_VECTOR];
4028 	}
4029 
4030 	/* See if we already did this for drivers which do double addintrs */
4031 	if (apic_oldvec_to_newvec[oldvector] != newvector) {
4032 		apic_oldvec_to_newvec[oldvector] = newvector;
4033 		apic_newvec_to_oldvec[newvector] = oldvector;
4034 		apic_revector_pending++;
4035 	}
4036 	lock_clear(&apic_revector_lock);
4037 	intr_restore(iflag);
4038 	(void) timeout(apic_xlate_vector_free_timeout_handler,
4039 	    (void *)(uintptr_t)oldvector, drv_usectohz(apic_revector_timeout));
4040 }
4041 
4042 /*
4043  * xlate_vector is called from intr_enter if revector_pending is set.
4044  * It will xlate it if needed and mark the old vector as free.
4045  */
4046 static uchar_t
4047 apic_xlate_vector(uchar_t vector)
4048 {
4049 	uchar_t	newvector, oldvector = 0;
4050 
4051 	lock_set(&apic_revector_lock);
4052 	/* Do we really need to do this ? */
4053 	if (!apic_revector_pending) {
4054 		lock_clear(&apic_revector_lock);
4055 		return (vector);
4056 	}
4057 	if ((newvector = apic_oldvec_to_newvec[vector]) != 0)
4058 		oldvector = vector;
4059 	else {
4060 		/*
4061 		 * The incoming vector is new . See if a stale entry is
4062 		 * remaining
4063 		 */
4064 		if ((oldvector = apic_newvec_to_oldvec[vector]) != 0)
4065 			newvector = vector;
4066 	}
4067 
4068 	if (oldvector) {
4069 		apic_revector_pending--;
4070 		apic_oldvec_to_newvec[oldvector] = 0;
4071 		apic_newvec_to_oldvec[newvector] = 0;
4072 		apic_free_vector(oldvector);
4073 		lock_clear(&apic_revector_lock);
4074 		/* There could have been more than one reprogramming! */
4075 		return (apic_xlate_vector(newvector));
4076 	}
4077 	lock_clear(&apic_revector_lock);
4078 	return (vector);
4079 }
4080 
4081 void
4082 apic_xlate_vector_free_timeout_handler(void *arg)
4083 {
4084 	int iflag;
4085 	uchar_t oldvector, newvector;
4086 
4087 	oldvector = (uchar_t)(uintptr_t)arg;
4088 	iflag = intr_clear();
4089 	lock_set(&apic_revector_lock);
4090 	if ((newvector = apic_oldvec_to_newvec[oldvector]) != 0) {
4091 		apic_free_vector(oldvector);
4092 		apic_oldvec_to_newvec[oldvector] = 0;
4093 		apic_newvec_to_oldvec[newvector] = 0;
4094 		apic_revector_pending--;
4095 	}
4096 
4097 	lock_clear(&apic_revector_lock);
4098 	intr_restore(iflag);
4099 }
4100 
4101 
4102 /* Mark vector as not being used by any irq */
4103 static void
4104 apic_free_vector(uchar_t vector)
4105 {
4106 	apic_vector_to_irq[vector] = APIC_RESV_IRQ;
4107 }
4108 
4109 /*
4110  * compute the polarity, trigger mode and vector for programming into
4111  * the I/O apic and record in airq_rdt_entry.
4112  */
4113 static void
4114 apic_record_rdt_entry(apic_irq_t *irqptr, int irq)
4115 {
4116 	int	ioapicindex, bus_type, vector;
4117 	short	intr_index;
4118 	uint_t	level, po, io_po;
4119 	struct apic_io_intr *iointrp;
4120 
4121 	intr_index = irqptr->airq_mps_intr_index;
4122 	DDI_INTR_IMPLDBG((CE_CONT, "apic_record_rdt_entry: intr_index=%d "
4123 	    "irq = 0x%x dip = 0x%p vector = 0x%x\n", intr_index, irq,
4124 	    (void *)irqptr->airq_dip, irqptr->airq_vector));
4125 
4126 	if (intr_index == RESERVE_INDEX) {
4127 		apic_error |= APIC_ERR_INVALID_INDEX;
4128 		return;
4129 	} else if (APIC_IS_MSI_OR_MSIX_INDEX(intr_index)) {
4130 		return;
4131 	}
4132 
4133 	vector = irqptr->airq_vector;
4134 	ioapicindex = irqptr->airq_ioapicindex;
4135 	/* Assume edge triggered by default */
4136 	level = 0;
4137 	/* Assume active high by default */
4138 	po = 0;
4139 
4140 	if (intr_index == DEFAULT_INDEX || intr_index == FREE_INDEX) {
4141 		ASSERT(irq < 16);
4142 		if (eisa_level_intr_mask & (1 << irq))
4143 			level = AV_LEVEL;
4144 		if (intr_index == FREE_INDEX && apic_defconf == 0)
4145 			apic_error |= APIC_ERR_INVALID_INDEX;
4146 	} else if (intr_index == ACPI_INDEX) {
4147 		bus_type = irqptr->airq_iflag.bustype;
4148 		if (irqptr->airq_iflag.intr_el == INTR_EL_CONFORM) {
4149 			if (bus_type == BUS_PCI)
4150 				level = AV_LEVEL;
4151 		} else
4152 			level = (irqptr->airq_iflag.intr_el == INTR_EL_LEVEL) ?
4153 			    AV_LEVEL : 0;
4154 		if (level &&
4155 		    ((irqptr->airq_iflag.intr_po == INTR_PO_ACTIVE_LOW) ||
4156 		    (irqptr->airq_iflag.intr_po == INTR_PO_CONFORM &&
4157 		    bus_type == BUS_PCI)))
4158 			po = AV_ACTIVE_LOW;
4159 	} else {
4160 		iointrp = apic_io_intrp + intr_index;
4161 		bus_type = apic_find_bus(iointrp->intr_busid);
4162 		if (iointrp->intr_el == INTR_EL_CONFORM) {
4163 			if ((irq < 16) && (eisa_level_intr_mask & (1 << irq)))
4164 				level = AV_LEVEL;
4165 			else if (bus_type == BUS_PCI)
4166 				level = AV_LEVEL;
4167 		} else
4168 			level = (iointrp->intr_el == INTR_EL_LEVEL) ?
4169 			    AV_LEVEL : 0;
4170 		if (level && ((iointrp->intr_po == INTR_PO_ACTIVE_LOW) ||
4171 		    (iointrp->intr_po == INTR_PO_CONFORM &&
4172 		    bus_type == BUS_PCI)))
4173 			po = AV_ACTIVE_LOW;
4174 	}
4175 	if (level)
4176 		apic_level_intr[irq] = 1;
4177 	/*
4178 	 * The 82489DX External APIC cannot do active low polarity interrupts.
4179 	 */
4180 	if (po && (apic_io_ver[ioapicindex] != IOAPIC_VER_82489DX))
4181 		io_po = po;
4182 	else
4183 		io_po = 0;
4184 
4185 	if (apic_verbose & APIC_VERBOSE_IOAPIC_FLAG)
4186 		printf("setio: ioapic=%x intin=%x level=%x po=%x vector=%x\n",
4187 		    ioapicindex, irqptr->airq_intin_no, level, io_po, vector);
4188 
4189 	irqptr->airq_rdt_entry = level|io_po|vector;
4190 }
4191 
4192 /*
4193  * Call rebind to do the actual programming.
4194  */
4195 static int
4196 apic_setup_io_intr(apic_irq_t *irqptr, int irq)
4197 {
4198 	int rv;
4199 
4200 	if (rv = apic_rebind(irqptr, apic_irq_table[irq]->airq_cpu, 1,
4201 	    IMMEDIATE))
4202 		/* CPU is not up or interrupt is disabled. Fall back to 0 */
4203 		rv = apic_rebind(irqptr, 0, 1, IMMEDIATE);
4204 
4205 	return (rv);
4206 }
4207 
4208 /*
4209  * Deferred reprogramming: Call apic_rebind to do the real work.
4210  */
4211 static int
4212 apic_setup_io_intr_deferred(apic_irq_t *irqptr, int irq)
4213 {
4214 	int rv;
4215 
4216 	if (rv = apic_rebind(irqptr, apic_irq_table[irq]->airq_cpu, 1,
4217 	    DEFERRED))
4218 		/* CPU is not up or interrupt is disabled. Fall back to 0 */
4219 		rv = apic_rebind(irqptr, 0, 1, DEFERRED);
4220 
4221 	return (rv);
4222 }
4223 
4224 /*
4225  * Bind interrupt corresponding to irq_ptr to bind_cpu. acquire_lock
4226  * if false (0) means lock is already held (e.g: in rebind_all).
4227  */
4228 static int
4229 apic_rebind(apic_irq_t *irq_ptr, int bind_cpu, int acquire_lock, int when)
4230 {
4231 	int			intin_no;
4232 	volatile int32_t	*ioapic;
4233 	uchar_t			airq_temp_cpu;
4234 	apic_cpus_info_t	*cpu_infop;
4235 	int			iflag;
4236 	int		which_irq = apic_vector_to_irq[irq_ptr->airq_vector];
4237 
4238 	intin_no = irq_ptr->airq_intin_no;
4239 	ioapic = apicioadr[irq_ptr->airq_ioapicindex];
4240 	airq_temp_cpu = irq_ptr->airq_temp_cpu;
4241 	if (airq_temp_cpu != IRQ_UNINIT && airq_temp_cpu != IRQ_UNBOUND) {
4242 		if (airq_temp_cpu & IRQ_USER_BOUND)
4243 			/* Mask off high bit so it can be used as array index */
4244 			airq_temp_cpu &= ~IRQ_USER_BOUND;
4245 
4246 		ASSERT(airq_temp_cpu < apic_nproc);
4247 	}
4248 
4249 	iflag = intr_clear();
4250 
4251 	if (acquire_lock)
4252 		lock_set(&apic_ioapic_lock);
4253 
4254 	/*
4255 	 * Can't bind to a CPU that's not online:
4256 	 */
4257 	cpu_infop = &apic_cpus[bind_cpu & ~IRQ_USER_BOUND];
4258 	if (!(cpu_infop->aci_status & APIC_CPU_INTR_ENABLE)) {
4259 
4260 		if (acquire_lock)
4261 			lock_clear(&apic_ioapic_lock);
4262 
4263 		intr_restore(iflag);
4264 		return (1);
4265 	}
4266 
4267 	/*
4268 	 * If this is a deferred reprogramming attempt, ensure we have
4269 	 * not been passed stale data:
4270 	 */
4271 	if ((when == DEFERRED) &&
4272 	    (apic_reprogram_info[which_irq].valid == 0)) {
4273 		/* stale info, so just return */
4274 		if (acquire_lock)
4275 			lock_clear(&apic_ioapic_lock);
4276 
4277 		intr_restore(iflag);
4278 		return (0);
4279 	}
4280 
4281 	/*
4282 	 * If this interrupt has been delivered to a CPU and that CPU
4283 	 * has not handled it yet, we cannot reprogram the IOAPIC now:
4284 	 */
4285 	if (!APIC_IS_MSI_OR_MSIX_INDEX(irq_ptr->airq_mps_intr_index) &&
4286 	    apic_check_stuck_interrupt(irq_ptr, airq_temp_cpu, bind_cpu,
4287 	    ioapic, intin_no, which_irq) != 0) {
4288 
4289 		if (acquire_lock)
4290 			lock_clear(&apic_ioapic_lock);
4291 
4292 		intr_restore(iflag);
4293 		return (0);
4294 	}
4295 
4296 	/*
4297 	 * NOTE: We do not unmask the RDT here, as an interrupt MAY still
4298 	 * come in before we have a chance to reprogram it below.  The
4299 	 * reprogramming below will simultaneously change and unmask the
4300 	 * RDT entry.
4301 	 */
4302 
4303 	if ((uchar_t)bind_cpu == IRQ_UNBOUND) {
4304 		/* Write the RDT entry -- no specific CPU binding */
4305 		WRITE_IOAPIC_RDT_ENTRY_HIGH_DWORD(ioapic, intin_no, AV_TOALL);
4306 
4307 		if (airq_temp_cpu != IRQ_UNINIT && airq_temp_cpu != IRQ_UNBOUND)
4308 			apic_cpus[airq_temp_cpu].aci_temp_bound--;
4309 
4310 		/* Write the vector, trigger, and polarity portion of the RDT */
4311 		WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no,
4312 		    AV_LDEST | AV_LOPRI | irq_ptr->airq_rdt_entry);
4313 		if (acquire_lock)
4314 			lock_clear(&apic_ioapic_lock);
4315 		irq_ptr->airq_temp_cpu = IRQ_UNBOUND;
4316 		intr_restore(iflag);
4317 		return (0);
4318 	}
4319 
4320 	if (bind_cpu & IRQ_USER_BOUND) {
4321 		cpu_infop->aci_bound++;
4322 	} else {
4323 		cpu_infop->aci_temp_bound++;
4324 	}
4325 	ASSERT((bind_cpu & ~IRQ_USER_BOUND) < apic_nproc);
4326 	if (!APIC_IS_MSI_OR_MSIX_INDEX(irq_ptr->airq_mps_intr_index)) {
4327 		/* Write the RDT entry -- bind to a specific CPU: */
4328 		WRITE_IOAPIC_RDT_ENTRY_HIGH_DWORD(ioapic, intin_no,
4329 		    cpu_infop->aci_local_id << APIC_ID_BIT_OFFSET);
4330 	}
4331 	if ((airq_temp_cpu != IRQ_UNBOUND) && (airq_temp_cpu != IRQ_UNINIT)) {
4332 		apic_cpus[airq_temp_cpu].aci_temp_bound--;
4333 	}
4334 	if (!APIC_IS_MSI_OR_MSIX_INDEX(irq_ptr->airq_mps_intr_index)) {
4335 		/* Write the vector, trigger, and polarity portion of the RDT */
4336 		WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no,
4337 		    AV_PDEST | AV_FIXED | irq_ptr->airq_rdt_entry);
4338 	} else {
4339 		if (irq_ptr->airq_ioapicindex == irq_ptr->airq_origirq) {
4340 			/* first one */
4341 			DDI_INTR_IMPLDBG((CE_CONT, "apic_rebind: call "
4342 			    "apic_pci_msi_enable_vector\n"));
4343 			if (apic_pci_msi_enable_vector(irq_ptr->airq_dip,
4344 			    (irq_ptr->airq_mps_intr_index == MSI_INDEX) ?
4345 			    DDI_INTR_TYPE_MSI : DDI_INTR_TYPE_MSIX, which_irq,
4346 			    irq_ptr->airq_vector, irq_ptr->airq_intin_no,
4347 			    cpu_infop->aci_local_id) != PSM_SUCCESS) {
4348 				cmn_err(CE_WARN, "pcplusmp: "
4349 					"apic_pci_msi_enable_vector "
4350 					"returned PSM_FAILURE");
4351 			}
4352 		}
4353 		if ((irq_ptr->airq_ioapicindex + irq_ptr->airq_intin_no - 1) ==
4354 		    irq_ptr->airq_origirq) { /* last one */
4355 			DDI_INTR_IMPLDBG((CE_CONT, "apic_rebind: call "
4356 			    "pci_msi_enable_mode\n"));
4357 			if (pci_msi_enable_mode(irq_ptr->airq_dip,
4358 			    (irq_ptr->airq_mps_intr_index == MSI_INDEX) ?
4359 			    DDI_INTR_TYPE_MSI : DDI_INTR_TYPE_MSIX,
4360 			    which_irq) != DDI_SUCCESS) {
4361 				DDI_INTR_IMPLDBG((CE_CONT, "pcplusmp: "
4362 				    "pci_msi_enable failed\n"));
4363 				(void) pci_msi_unconfigure(irq_ptr->airq_dip,
4364 				(irq_ptr->airq_mps_intr_index == MSI_INDEX) ?
4365 				DDI_INTR_TYPE_MSI : DDI_INTR_TYPE_MSIX,
4366 				which_irq);
4367 			}
4368 		}
4369 	}
4370 	if (acquire_lock)
4371 		lock_clear(&apic_ioapic_lock);
4372 	irq_ptr->airq_temp_cpu = (uchar_t)bind_cpu;
4373 	apic_redist_cpu_skip &= ~(1 << (bind_cpu & ~IRQ_USER_BOUND));
4374 	intr_restore(iflag);
4375 	return (0);
4376 }
4377 
4378 /*
4379  * Checks to see if the IOAPIC interrupt entry specified has its Remote IRR
4380  * bit set.  Sets up a timeout to perform the reprogramming at a later time
4381  * if it cannot wait for the Remote IRR bit to clear (or if waiting did not
4382  * result in the bit's clearing).
4383  *
4384  * This function will mask the RDT entry if the Remote IRR bit is set.
4385  *
4386  * Returns non-zero if the caller should defer IOAPIC reprogramming.
4387  */
4388 static int
4389 apic_check_stuck_interrupt(apic_irq_t *irq_ptr, int old_bind_cpu,
4390 	int new_bind_cpu, volatile int32_t *ioapic, int intin_no, int which_irq)
4391 {
4392 	int32_t			rdt_entry;
4393 	int			waited;
4394 
4395 	/* Mask the RDT entry, but only if it's a level-triggered interrupt */
4396 	rdt_entry = READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no);
4397 	if ((rdt_entry & (AV_LEVEL|AV_MASK)) == AV_LEVEL) {
4398 
4399 		/* Mask it */
4400 		WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no,
4401 		    AV_MASK | rdt_entry);
4402 	}
4403 
4404 	/*
4405 	 * Wait for the delivery pending bit to clear.
4406 	 */
4407 	if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no) &
4408 	    (AV_LEVEL|AV_PENDING)) == (AV_LEVEL|AV_PENDING)) {
4409 
4410 		/*
4411 		 * If we're still waiting on the delivery of this interrupt,
4412 		 * continue to wait here until it is delivered (this should be
4413 		 * a very small amount of time, but include a timeout just in
4414 		 * case).
4415 		 */
4416 		for (waited = 0; waited < apic_max_usecs_clear_pending;
4417 		    waited += APIC_USECS_PER_WAIT_INTERVAL) {
4418 			if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no)
4419 			    & AV_PENDING) == 0) {
4420 				break;
4421 			}
4422 			drv_usecwait(APIC_USECS_PER_WAIT_INTERVAL);
4423 		}
4424 
4425 		if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no) &
4426 		    AV_PENDING) != 0) {
4427 			cmn_err(CE_WARN, "!IOAPIC %d intin %d: Could not "
4428 			    "deliver interrupt to local APIC within "
4429 			    "%d usecs.", irq_ptr->airq_ioapicindex,
4430 			    irq_ptr->airq_intin_no,
4431 			    apic_max_usecs_clear_pending);
4432 		}
4433 	}
4434 
4435 	/*
4436 	 * If the remote IRR bit is set, then the interrupt has been sent
4437 	 * to a CPU for processing.  We have no choice but to wait for
4438 	 * that CPU to process the interrupt, at which point the remote IRR
4439 	 * bit will be cleared.
4440 	 */
4441 	if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no) &
4442 	    (AV_LEVEL|AV_REMOTE_IRR)) == (AV_LEVEL|AV_REMOTE_IRR)) {
4443 
4444 		/*
4445 		 * If the CPU that this RDT is bound to is NOT the current
4446 		 * CPU, wait until that CPU handles the interrupt and ACKs
4447 		 * it.  If this interrupt is not bound to any CPU (that is,
4448 		 * if it's bound to the logical destination of "anyone"), it
4449 		 * may have been delivered to the current CPU so handle that
4450 		 * case by deferring the reprogramming (below).
4451 		 */
4452 		kpreempt_disable();
4453 		if ((old_bind_cpu != IRQ_UNBOUND) &&
4454 		    (old_bind_cpu != IRQ_UNINIT) &&
4455 		    (old_bind_cpu != psm_get_cpu_id())) {
4456 			for (waited = 0; waited < apic_max_usecs_clear_pending;
4457 			    waited += APIC_USECS_PER_WAIT_INTERVAL) {
4458 				if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic,
4459 				    intin_no) & AV_REMOTE_IRR) == 0) {
4460 
4461 					/* Clear the reprogramming state: */
4462 					lock_set(&apic_ioapic_reprogram_lock);
4463 
4464 					apic_reprogram_info[which_irq].valid
4465 					    = 0;
4466 					apic_reprogram_info[which_irq].bindcpu
4467 					    = 0;
4468 					apic_reprogram_info[which_irq].timeouts
4469 					    = 0;
4470 
4471 					lock_clear(&apic_ioapic_reprogram_lock);
4472 
4473 					/* Remote IRR has cleared! */
4474 					kpreempt_enable();
4475 					return (0);
4476 				}
4477 				drv_usecwait(APIC_USECS_PER_WAIT_INTERVAL);
4478 			}
4479 		}
4480 		kpreempt_enable();
4481 
4482 		/*
4483 		 * If we waited and the Remote IRR bit is still not cleared,
4484 		 * AND if we've invoked the timeout APIC_REPROGRAM_MAX_TIMEOUTS
4485 		 * times for this interrupt, try the last-ditch workarounds:
4486 		 */
4487 		if (apic_reprogram_info[which_irq].timeouts >=
4488 		    APIC_REPROGRAM_MAX_TIMEOUTS) {
4489 
4490 			if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no)
4491 			    & AV_REMOTE_IRR) != 0) {
4492 				/*
4493 				 * Trying to clear the bit through normal
4494 				 * channels has failed.  So as a last-ditch
4495 				 * effort, try to set the trigger mode to
4496 				 * edge, then to level.  This has been
4497 				 * observed to work on many systems.
4498 				 */
4499 				WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic,
4500 				    intin_no,
4501 				    READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic,
4502 				    intin_no) & ~AV_LEVEL);
4503 
4504 				WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic,
4505 				    intin_no,
4506 				    READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic,
4507 				    intin_no) | AV_LEVEL);
4508 
4509 				/*
4510 				 * If the bit's STILL set, declare total and
4511 				 * utter failure
4512 				 */
4513 				if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic,
4514 				    intin_no) & AV_REMOTE_IRR) != 0) {
4515 					cmn_err(CE_WARN, "!IOAPIC %d intin %d: "
4516 					    "Remote IRR failed to reset "
4517 					    "within %d usecs.  Interrupts to "
4518 					    "this pin may cease to function.",
4519 					    irq_ptr->airq_ioapicindex,
4520 					    irq_ptr->airq_intin_no,
4521 					    apic_max_usecs_clear_pending);
4522 				}
4523 			}
4524 			/* Clear the reprogramming state: */
4525 			lock_set(&apic_ioapic_reprogram_lock);
4526 
4527 			apic_reprogram_info[which_irq].valid = 0;
4528 			apic_reprogram_info[which_irq].bindcpu = 0;
4529 			apic_reprogram_info[which_irq].timeouts = 0;
4530 
4531 			lock_clear(&apic_ioapic_reprogram_lock);
4532 		} else {
4533 #ifdef DEBUG
4534 			cmn_err(CE_WARN, "Deferring reprogramming of irq %d",
4535 			    which_irq);
4536 #endif	/* DEBUG */
4537 			/*
4538 			 * If waiting for the Remote IRR bit (above) didn't
4539 			 * allow it to clear, defer the reprogramming:
4540 			 */
4541 			lock_set(&apic_ioapic_reprogram_lock);
4542 
4543 			apic_reprogram_info[which_irq].valid = 1;
4544 			apic_reprogram_info[which_irq].bindcpu = new_bind_cpu;
4545 			apic_reprogram_info[which_irq].timeouts++;
4546 
4547 			lock_clear(&apic_ioapic_reprogram_lock);
4548 
4549 			/* Fire up a timeout to handle this later */
4550 			(void) timeout(apic_reprogram_timeout_handler,
4551 			    (void *) 0,
4552 			    drv_usectohz(APIC_REPROGRAM_TIMEOUT_DELAY));
4553 
4554 			/* Inform caller to defer IOAPIC programming: */
4555 			return (1);
4556 		}
4557 	}
4558 	return (0);
4559 }
4560 
4561 /*
4562  * Timeout handler that performs the APIC reprogramming
4563  */
4564 /*ARGSUSED*/
4565 static void
4566 apic_reprogram_timeout_handler(void *arg)
4567 {
4568 	/*LINTED: set but not used in function*/
4569 	int i, result;
4570 
4571 	/* Serialize access to this function */
4572 	mutex_enter(&apic_reprogram_timeout_mutex);
4573 
4574 	/*
4575 	 * For each entry in the reprogramming state that's valid,
4576 	 * try the reprogramming again:
4577 	 */
4578 	for (i = 0; i < APIC_MAX_VECTOR; i++) {
4579 		if (apic_reprogram_info[i].valid == 0)
4580 			continue;
4581 		/*
4582 		 * Though we can't really do anything about errors
4583 		 * at this point, keep track of them for reporting.
4584 		 * Note that it is very possible for apic_setup_io_intr
4585 		 * to re-register this very timeout if the Remote IRR bit
4586 		 * has not yet cleared.
4587 		 */
4588 		result = apic_setup_io_intr_deferred(apic_irq_table[i], i);
4589 
4590 #ifdef DEBUG
4591 		if (result)
4592 			cmn_err(CE_WARN, "apic_reprogram_timeout: "
4593 			    "apic_setup_io_intr returned nonzero for "
4594 			    "irq=%d!", i);
4595 #endif	/* DEBUG */
4596 	}
4597 
4598 	mutex_exit(&apic_reprogram_timeout_mutex);
4599 }
4600 
4601 
4602 /*
4603  * Called to migrate all interrupts at an irq to another cpu. safe
4604  * if true means we are not being called from an interrupt
4605  * context and hence it is safe to do a lock_set. If false
4606  * do only a lock_try and return failure ( non 0 ) if we cannot get it
4607  */
4608 static int
4609 apic_rebind_all(apic_irq_t *irq_ptr, int bind_cpu, int safe)
4610 {
4611 	apic_irq_t	*irqptr = irq_ptr;
4612 	int		retval = 0;
4613 	int		iflag;
4614 
4615 	iflag = intr_clear();
4616 	if (!safe) {
4617 		if (lock_try(&apic_ioapic_lock) == 0) {
4618 			intr_restore(iflag);
4619 			return (1);
4620 		}
4621 	} else
4622 		lock_set(&apic_ioapic_lock);
4623 
4624 	while (irqptr) {
4625 		if (irqptr->airq_temp_cpu != IRQ_UNINIT)
4626 			retval |= apic_rebind(irqptr, bind_cpu, 0, IMMEDIATE);
4627 		irqptr = irqptr->airq_next;
4628 	}
4629 	lock_clear(&apic_ioapic_lock);
4630 	intr_restore(iflag);
4631 	return (retval);
4632 }
4633 
4634 /*
4635  * apic_intr_redistribute does all the messy computations for identifying
4636  * which interrupt to move to which CPU. Currently we do just one interrupt
4637  * at a time. This reduces the time we spent doing all this within clock
4638  * interrupt. When it is done in idle, we could do more than 1.
4639  * First we find the most busy and the most free CPU (time in ISR only)
4640  * skipping those CPUs that has been identified as being ineligible (cpu_skip)
4641  * Then we look for IRQs which are closest to the difference between the
4642  * most busy CPU and the average ISR load. We try to find one whose load
4643  * is less than difference.If none exists, then we chose one larger than the
4644  * difference, provided it does not make the most idle CPU worse than the
4645  * most busy one. In the end, we clear all the busy fields for CPUs. For
4646  * IRQs, they are cleared as they are scanned.
4647  */
4648 static void
4649 apic_intr_redistribute()
4650 {
4651 	int busiest_cpu, most_free_cpu;
4652 	int cpu_free, cpu_busy, max_busy, min_busy;
4653 	int min_free, diff;
4654 	int	average_busy, cpus_online;
4655 	int i, busy;
4656 	apic_cpus_info_t *cpu_infop;
4657 	apic_irq_t *min_busy_irq = NULL;
4658 	apic_irq_t *max_busy_irq = NULL;
4659 
4660 	busiest_cpu = most_free_cpu = -1;
4661 	cpu_free = cpu_busy = max_busy = average_busy = 0;
4662 	min_free = apic_sample_factor_redistribution;
4663 	cpus_online = 0;
4664 	/*
4665 	 * Below we will check for CPU_INTR_ENABLE, bound, temp_bound, temp_cpu
4666 	 * without ioapic_lock. That is OK as we are just doing statistical
4667 	 * sampling anyway and any inaccuracy now will get corrected next time
4668 	 * The call to rebind which actually changes things will make sure
4669 	 * we are consistent.
4670 	 */
4671 	for (i = 0; i < apic_nproc; i++) {
4672 		if (!(apic_redist_cpu_skip & (1 << i)) &&
4673 		    (apic_cpus[i].aci_status & APIC_CPU_INTR_ENABLE)) {
4674 
4675 			cpu_infop = &apic_cpus[i];
4676 			/*
4677 			 * If no unbound interrupts or only 1 total on this
4678 			 * CPU, skip
4679 			 */
4680 			if (!cpu_infop->aci_temp_bound ||
4681 			    (cpu_infop->aci_bound + cpu_infop->aci_temp_bound)
4682 			    == 1) {
4683 				apic_redist_cpu_skip |= 1 << i;
4684 				continue;
4685 			}
4686 
4687 			busy = cpu_infop->aci_busy;
4688 			average_busy += busy;
4689 			cpus_online++;
4690 			if (max_busy < busy) {
4691 				max_busy = busy;
4692 				busiest_cpu = i;
4693 			}
4694 			if (min_free > busy) {
4695 				min_free = busy;
4696 				most_free_cpu = i;
4697 			}
4698 			if (busy > apic_int_busy_mark) {
4699 				cpu_busy |= 1 << i;
4700 			} else {
4701 				if (busy < apic_int_free_mark)
4702 					cpu_free |= 1 << i;
4703 			}
4704 		}
4705 	}
4706 	if ((cpu_busy && cpu_free) ||
4707 	    (max_busy >= (min_free + apic_diff_for_redistribution))) {
4708 
4709 		apic_num_imbalance++;
4710 #ifdef	DEBUG
4711 		if (apic_verbose & APIC_VERBOSE_IOAPIC_FLAG) {
4712 			prom_printf(
4713 			    "redistribute busy=%x free=%x max=%x min=%x",
4714 			    cpu_busy, cpu_free, max_busy, min_free);
4715 		}
4716 #endif /* DEBUG */
4717 
4718 
4719 		average_busy /= cpus_online;
4720 
4721 		diff = max_busy - average_busy;
4722 		min_busy = max_busy; /* start with the max possible value */
4723 		max_busy = 0;
4724 		min_busy_irq = max_busy_irq = NULL;
4725 		i = apic_min_device_irq;
4726 		for (; i < apic_max_device_irq; i++) {
4727 			apic_irq_t *irq_ptr;
4728 			/* Change to linked list per CPU ? */
4729 			if ((irq_ptr = apic_irq_table[i]) == NULL)
4730 				continue;
4731 			/* Check for irq_busy & decide which one to move */
4732 			/* Also zero them for next round */
4733 			if ((irq_ptr->airq_temp_cpu == busiest_cpu) &&
4734 			    irq_ptr->airq_busy) {
4735 				if (irq_ptr->airq_busy < diff) {
4736 					/*
4737 					 * Check for least busy CPU,
4738 					 * best fit or what ?
4739 					 */
4740 					if (max_busy < irq_ptr->airq_busy) {
4741 						/*
4742 						 * Most busy within the
4743 						 * required differential
4744 						 */
4745 						max_busy = irq_ptr->airq_busy;
4746 						max_busy_irq = irq_ptr;
4747 					}
4748 				} else {
4749 					if (min_busy > irq_ptr->airq_busy) {
4750 						/*
4751 						 * least busy, but more than
4752 						 * the reqd diff
4753 						 */
4754 						if (min_busy <
4755 						    (diff + average_busy -
4756 						    min_free)) {
4757 							/*
4758 							 * Making sure new cpu
4759 							 * will not end up
4760 							 * worse
4761 							 */
4762 							min_busy =
4763 							    irq_ptr->airq_busy;
4764 
4765 							min_busy_irq = irq_ptr;
4766 						}
4767 					}
4768 				}
4769 			}
4770 			irq_ptr->airq_busy = 0;
4771 		}
4772 
4773 		if (max_busy_irq != NULL) {
4774 #ifdef	DEBUG
4775 			if (apic_verbose & APIC_VERBOSE_IOAPIC_FLAG) {
4776 				prom_printf("rebinding %x to %x",
4777 				    max_busy_irq->airq_vector, most_free_cpu);
4778 			}
4779 #endif /* DEBUG */
4780 			if (apic_rebind_all(max_busy_irq, most_free_cpu, 0)
4781 			    == 0)
4782 				/* Make change permenant */
4783 				max_busy_irq->airq_cpu = (uchar_t)most_free_cpu;
4784 		} else if (min_busy_irq != NULL) {
4785 #ifdef	DEBUG
4786 			if (apic_verbose & APIC_VERBOSE_IOAPIC_FLAG) {
4787 				prom_printf("rebinding %x to %x",
4788 				    min_busy_irq->airq_vector, most_free_cpu);
4789 			}
4790 #endif /* DEBUG */
4791 
4792 			if (apic_rebind_all(min_busy_irq, most_free_cpu, 0) ==
4793 			    0)
4794 				/* Make change permenant */
4795 				min_busy_irq->airq_cpu = (uchar_t)most_free_cpu;
4796 		} else {
4797 			if (cpu_busy != (1 << busiest_cpu)) {
4798 				apic_redist_cpu_skip |= 1 << busiest_cpu;
4799 				/*
4800 				 * We leave cpu_skip set so that next time we
4801 				 * can choose another cpu
4802 				 */
4803 			}
4804 		}
4805 		apic_num_rebind++;
4806 	} else {
4807 		/*
4808 		 * found nothing. Could be that we skipped over valid CPUs
4809 		 * or we have balanced everything. If we had a variable
4810 		 * ticks_for_redistribution, it could be increased here.
4811 		 * apic_int_busy, int_free etc would also need to be
4812 		 * changed.
4813 		 */
4814 		if (apic_redist_cpu_skip)
4815 			apic_redist_cpu_skip = 0;
4816 	}
4817 	for (i = 0; i < apic_nproc; i++) {
4818 		apic_cpus[i].aci_busy = 0;
4819 	}
4820 }
4821 
4822 static void
4823 apic_cleanup_busy()
4824 {
4825 	int i;
4826 	apic_irq_t *irq_ptr;
4827 
4828 	for (i = 0; i < apic_nproc; i++) {
4829 		apic_cpus[i].aci_busy = 0;
4830 	}
4831 
4832 	for (i = apic_min_device_irq; i < apic_max_device_irq; i++) {
4833 		if ((irq_ptr = apic_irq_table[i]) != NULL)
4834 			irq_ptr->airq_busy = 0;
4835 	}
4836 	apic_skipped_redistribute = 0;
4837 }
4838 
4839 
4840 /*
4841  * This function will reprogram the timer.
4842  *
4843  * When in oneshot mode the argument is the absolute time in future to
4844  * generate the interrupt at.
4845  *
4846  * When in periodic mode, the argument is the interval at which the
4847  * interrupts should be generated. There is no need to support the periodic
4848  * mode timer change at this time.
4849  */
4850 static void
4851 apic_timer_reprogram(hrtime_t time)
4852 {
4853 	hrtime_t now;
4854 	uint_t ticks;
4855 
4856 	/*
4857 	 * We should be called from high PIL context (CBE_HIGH_PIL),
4858 	 * so kpreempt is disabled.
4859 	 */
4860 
4861 	if (!apic_oneshot) {
4862 		/* time is the interval for periodic mode */
4863 		ticks = (uint_t)((time) / apic_nsec_per_tick);
4864 	} else {
4865 		/* one shot mode */
4866 
4867 		now = gethrtime();
4868 
4869 		if (time <= now) {
4870 			/*
4871 			 * requested to generate an interrupt in the past
4872 			 * generate an interrupt as soon as possible
4873 			 */
4874 			ticks = apic_min_timer_ticks;
4875 		} else if ((time - now) > apic_nsec_max) {
4876 			/*
4877 			 * requested to generate an interrupt at a time
4878 			 * further than what we are capable of. Set to max
4879 			 * the hardware can handle
4880 			 */
4881 
4882 			ticks = APIC_MAXVAL;
4883 #ifdef DEBUG
4884 			cmn_err(CE_CONT, "apic_timer_reprogram, request at"
4885 			    "  %lld  too far in future, current time"
4886 			    "  %lld \n", time, now);
4887 #endif	/* DEBUG */
4888 		} else
4889 			ticks = (uint_t)((time - now) / apic_nsec_per_tick);
4890 	}
4891 
4892 	if (ticks < apic_min_timer_ticks)
4893 		ticks = apic_min_timer_ticks;
4894 
4895 	apicadr[APIC_INIT_COUNT] = ticks;
4896 
4897 }
4898 
4899 /*
4900  * This function will enable timer interrupts.
4901  */
4902 static void
4903 apic_timer_enable(void)
4904 {
4905 	/*
4906 	 * We should be Called from high PIL context (CBE_HIGH_PIL),
4907 	 * so kpreempt is disabled.
4908 	 */
4909 
4910 	if (!apic_oneshot)
4911 		apicadr[APIC_LOCAL_TIMER] =
4912 		    (apic_clkvect + APIC_BASE_VECT) | AV_TIME;
4913 	else {
4914 		/* one shot */
4915 		apicadr[APIC_LOCAL_TIMER] = (apic_clkvect + APIC_BASE_VECT);
4916 	}
4917 }
4918 
4919 /*
4920  * This function will disable timer interrupts.
4921  */
4922 static void
4923 apic_timer_disable(void)
4924 {
4925 	/*
4926 	 * We should be Called from high PIL context (CBE_HIGH_PIL),
4927 	 * so kpreempt is disabled.
4928 	 */
4929 
4930 	apicadr[APIC_LOCAL_TIMER] = (apic_clkvect + APIC_BASE_VECT) | AV_MASK;
4931 }
4932 
4933 
4934 cyclic_id_t apic_cyclic_id;
4935 
4936 /*
4937  * If this module needs to be a consumer of cyclic subsystem, they
4938  * can be added here, since at this time kernel cyclic subsystem is initialized
4939  * argument is not currently used, and is reserved for future.
4940  */
4941 static void
4942 apic_post_cyclic_setup(void *arg)
4943 {
4944 _NOTE(ARGUNUSED(arg))
4945 	cyc_handler_t hdlr;
4946 	cyc_time_t when;
4947 
4948 	/* cpu_lock is held */
4949 
4950 	/* set up cyclics for intr redistribution */
4951 
4952 	/*
4953 	 * In peridoc mode intr redistribution processing is done in
4954 	 * apic_intr_enter during clk intr processing
4955 	 */
4956 	if (!apic_oneshot)
4957 		return;
4958 
4959 	hdlr.cyh_level = CY_LOW_LEVEL;
4960 	hdlr.cyh_func = (cyc_func_t)apic_redistribute_compute;
4961 	hdlr.cyh_arg = NULL;
4962 
4963 	when.cyt_when = 0;
4964 	when.cyt_interval = apic_redistribute_sample_interval;
4965 	apic_cyclic_id = cyclic_add(&hdlr, &when);
4966 
4967 
4968 }
4969 
4970 static void
4971 apic_redistribute_compute(void)
4972 {
4973 	int	i, j, max_busy;
4974 
4975 	if (apic_enable_dynamic_migration) {
4976 		if (++apic_nticks == apic_sample_factor_redistribution) {
4977 			/*
4978 			 * Time to call apic_intr_redistribute().
4979 			 * reset apic_nticks. This will cause max_busy
4980 			 * to be calculated below and if it is more than
4981 			 * apic_int_busy, we will do the whole thing
4982 			 */
4983 			apic_nticks = 0;
4984 		}
4985 		max_busy = 0;
4986 		for (i = 0; i < apic_nproc; i++) {
4987 
4988 			/*
4989 			 * Check if curipl is non zero & if ISR is in
4990 			 * progress
4991 			 */
4992 			if (((j = apic_cpus[i].aci_curipl) != 0) &&
4993 			    (apic_cpus[i].aci_ISR_in_progress & (1 << j))) {
4994 
4995 				int	irq;
4996 				apic_cpus[i].aci_busy++;
4997 				irq = apic_cpus[i].aci_current[j];
4998 				apic_irq_table[irq]->airq_busy++;
4999 			}
5000 
5001 			if (!apic_nticks &&
5002 			    (apic_cpus[i].aci_busy > max_busy))
5003 				max_busy = apic_cpus[i].aci_busy;
5004 		}
5005 		if (!apic_nticks) {
5006 			if (max_busy > apic_int_busy_mark) {
5007 			/*
5008 			 * We could make the following check be
5009 			 * skipped > 1 in which case, we get a
5010 			 * redistribution at half the busy mark (due to
5011 			 * double interval). Need to be able to collect
5012 			 * more empirical data to decide if that is a
5013 			 * good strategy. Punt for now.
5014 			 */
5015 				if (apic_skipped_redistribute)
5016 					apic_cleanup_busy();
5017 				else
5018 					apic_intr_redistribute();
5019 			} else
5020 				apic_skipped_redistribute++;
5021 		}
5022 	}
5023 }
5024 
5025 
5026 static int
5027 apic_acpi_translate_pci_irq(dev_info_t *dip, int busid, int devid,
5028     int ipin, int *pci_irqp, iflag_t *intr_flagp)
5029 {
5030 
5031 	int status;
5032 	acpi_psm_lnk_t acpipsmlnk;
5033 
5034 	if ((status = acpi_get_irq_cache_ent(busid, devid, ipin, pci_irqp,
5035 	    intr_flagp)) == ACPI_PSM_SUCCESS) {
5036 		APIC_VERBOSE_IRQ((CE_CONT, "!pcplusmp: Found irqno %d "
5037 		    "from cache for device %s, instance #%d\n", *pci_irqp,
5038 		    ddi_get_name(dip), ddi_get_instance(dip)));
5039 		return (status);
5040 	}
5041 
5042 	bzero(&acpipsmlnk, sizeof (acpi_psm_lnk_t));
5043 
5044 	if ((status = acpi_translate_pci_irq(dip, ipin, pci_irqp, intr_flagp,
5045 	    &acpipsmlnk)) == ACPI_PSM_FAILURE) {
5046 		APIC_VERBOSE_IRQ((CE_WARN, "pcplusmp: "
5047 		    " acpi_translate_pci_irq failed for device %s, instance"
5048 		    " #%d", ddi_get_name(dip), ddi_get_instance(dip)));
5049 		return (status);
5050 	}
5051 
5052 	if (status == ACPI_PSM_PARTIAL && acpipsmlnk.lnkobj != NULL) {
5053 		status = apic_acpi_irq_configure(&acpipsmlnk, dip, pci_irqp,
5054 		    intr_flagp);
5055 		if (status != ACPI_PSM_SUCCESS) {
5056 			status = acpi_get_current_irq_resource(&acpipsmlnk,
5057 			    pci_irqp, intr_flagp);
5058 		}
5059 	}
5060 
5061 	if (status == ACPI_PSM_SUCCESS) {
5062 		acpi_new_irq_cache_ent(busid, devid, ipin, *pci_irqp,
5063 		    intr_flagp, &acpipsmlnk);
5064 
5065 		APIC_VERBOSE_IRQ((CE_CONT, "pcplusmp: [ACPI] "
5066 		    "new irq %d for device %s, instance #%d\n",
5067 		    *pci_irqp, ddi_get_name(dip), ddi_get_instance(dip)));
5068 	}
5069 
5070 	return (status);
5071 }
5072 
5073 /*
5074  * Configures the irq for the interrupt link device identified by
5075  * acpipsmlnkp.
5076  *
5077  * Gets the current and the list of possible irq settings for the
5078  * device. If apic_unconditional_srs is not set, and the current
5079  * resource setting is in the list of possible irq settings,
5080  * current irq resource setting is passed to the caller.
5081  *
5082  * Otherwise, picks an irq number from the list of possible irq
5083  * settings, and sets the irq of the device to this value.
5084  * If prefer_crs is set, among a set of irq numbers in the list that have
5085  * the least number of devices sharing the interrupt, we pick current irq
5086  * resource setting if it is a member of this set.
5087  *
5088  * Passes the irq number in the value pointed to by pci_irqp, and
5089  * polarity and sensitivity in the structure pointed to by dipintrflagp
5090  * to the caller.
5091  *
5092  * Note that if setting the irq resource failed, but successfuly obtained
5093  * the current irq resource settings, passes the current irq resources
5094  * and considers it a success.
5095  *
5096  * Returns:
5097  * ACPI_PSM_SUCCESS on success.
5098  *
5099  * ACPI_PSM_FAILURE if an error occured during the configuration or
5100  * if a suitable irq was not found for this device, or if setting the
5101  * irq resource and obtaining the current resource fails.
5102  *
5103  */
5104 static int
5105 apic_acpi_irq_configure(acpi_psm_lnk_t *acpipsmlnkp, dev_info_t *dip,
5106     int *pci_irqp, iflag_t *dipintr_flagp)
5107 {
5108 
5109 	int i, min_share, foundnow, done = 0;
5110 	int32_t irq;
5111 	int32_t share_irq = -1;
5112 	int32_t chosen_irq = -1;
5113 	int cur_irq = -1;
5114 	acpi_irqlist_t *irqlistp;
5115 	acpi_irqlist_t *irqlistent;
5116 
5117 	if ((acpi_get_possible_irq_resources(acpipsmlnkp, &irqlistp))
5118 	    == ACPI_PSM_FAILURE) {
5119 		APIC_VERBOSE_IRQ((CE_WARN, "!pcplusmp: Unable to determine "
5120 		    "or assign IRQ for device %s, instance #%d: The system was "
5121 		    "unable to get the list of potential IRQs from ACPI.",
5122 		    ddi_get_name(dip), ddi_get_instance(dip)));
5123 
5124 		return (ACPI_PSM_FAILURE);
5125 	}
5126 
5127 	if ((acpi_get_current_irq_resource(acpipsmlnkp, &cur_irq,
5128 	    dipintr_flagp) == ACPI_PSM_SUCCESS) && (!apic_unconditional_srs) &&
5129 	    (cur_irq > 0)) {
5130 		/*
5131 		 * If an IRQ is set in CRS and that IRQ exists in the set
5132 		 * returned from _PRS, return that IRQ, otherwise print
5133 		 * a warning
5134 		 */
5135 
5136 		if (acpi_irqlist_find_irq(irqlistp, cur_irq, NULL)
5137 		    == ACPI_PSM_SUCCESS) {
5138 
5139 			acpi_free_irqlist(irqlistp);
5140 			ASSERT(pci_irqp != NULL);
5141 			*pci_irqp = cur_irq;
5142 			return (ACPI_PSM_SUCCESS);
5143 		}
5144 
5145 		APIC_VERBOSE_IRQ((CE_WARN, "!pcplusmp: Could not find the "
5146 		    "current irq %d for device %s, instance #%d in ACPI's "
5147 		    "list of possible irqs for this device. Picking one from "
5148 		    " the latter list.", cur_irq, ddi_get_name(dip),
5149 		    ddi_get_instance(dip)));
5150 	}
5151 
5152 	irqlistent = irqlistp;
5153 	min_share = 255;
5154 
5155 	while (irqlistent != NULL) {
5156 		irqlistent->intr_flags.bustype = BUS_PCI;
5157 
5158 		for (foundnow = 0, i = 0; i < irqlistent->num_irqs; i++) {
5159 
5160 			irq = irqlistent->irqs[i];
5161 
5162 			if ((irq < 16) && (apic_reserved_irqlist[irq]))
5163 				continue;
5164 
5165 			if (irq == 0) {
5166 				/* invalid irq number */
5167 				continue;
5168 			}
5169 
5170 			if ((apic_irq_table[irq] == NULL) ||
5171 			    (apic_irq_table[irq]->airq_dip == dip)) {
5172 				chosen_irq = irq;
5173 				foundnow = 1;
5174 				/*
5175 				 * If we do not prefer current irq from crs
5176 				 * or if we do and this irq is the same as
5177 				 * current irq from crs, this is the one
5178 				 * to pick.
5179 				 */
5180 				if (!(apic_prefer_crs) || (irq == cur_irq)) {
5181 					done = 1;
5182 					break;
5183 				}
5184 				continue;
5185 			}
5186 
5187 			if (irqlistent->intr_flags.intr_el == INTR_EL_EDGE)
5188 				continue;
5189 
5190 			if (!acpi_intr_compatible(irqlistent->intr_flags,
5191 			    apic_irq_table[irq]->airq_iflag))
5192 				continue;
5193 
5194 			if ((apic_irq_table[irq]->airq_share < min_share) ||
5195 			    ((apic_irq_table[irq]->airq_share == min_share) &&
5196 			    (cur_irq == irq) && (apic_prefer_crs))) {
5197 				min_share = apic_irq_table[irq]->airq_share;
5198 				share_irq = irq;
5199 				foundnow = 1;
5200 			}
5201 		}
5202 
5203 		/*
5204 		 * If we found an IRQ in the inner loop this time, save the
5205 		 * details from the irqlist for later use.
5206 		 */
5207 		if (foundnow && ((chosen_irq != -1) || (share_irq != -1))) {
5208 			/*
5209 			 * Copy the acpi_prs_private_t and flags from this
5210 			 * irq list entry, since we found an irq from this
5211 			 * entry.
5212 			 */
5213 			acpipsmlnkp->acpi_prs_prv = irqlistent->acpi_prs_prv;
5214 			*dipintr_flagp = irqlistent->intr_flags;
5215 		}
5216 
5217 		if (done)
5218 			break;
5219 
5220 		/* Go to the next irqlist entry */
5221 		irqlistent = irqlistent->next;
5222 	}
5223 
5224 
5225 	acpi_free_irqlist(irqlistp);
5226 	if (chosen_irq != -1)
5227 		irq = chosen_irq;
5228 	else if (share_irq != -1)
5229 		irq = share_irq;
5230 	else {
5231 		APIC_VERBOSE_IRQ((CE_WARN, "!pcplusmp: Could not find a "
5232 		    "suitable irq from the list of possible irqs for device "
5233 		    "%s, instance #%d in ACPI's list of possible irqs",
5234 		    ddi_get_name(dip), ddi_get_instance(dip)));
5235 		return (ACPI_PSM_FAILURE);
5236 	}
5237 
5238 	APIC_VERBOSE_IRQ((CE_CONT, "!pcplusmp: Setting irq %d for device %s "
5239 	    "instance #%d\n", irq, ddi_get_name(dip), ddi_get_instance(dip)));
5240 
5241 	if ((acpi_set_irq_resource(acpipsmlnkp, irq)) == ACPI_PSM_SUCCESS) {
5242 		/*
5243 		 * setting irq was successful, check to make sure CRS
5244 		 * reflects that. If CRS does not agree with what we
5245 		 * set, return the irq that was set.
5246 		 */
5247 
5248 		if (acpi_get_current_irq_resource(acpipsmlnkp, &cur_irq,
5249 		    dipintr_flagp) == ACPI_PSM_SUCCESS) {
5250 
5251 			if (cur_irq != irq)
5252 				APIC_VERBOSE_IRQ((CE_WARN, "!pcplusmp: "
5253 				    "IRQ resource set (irqno %d) for device %s "
5254 				    "instance #%d, differs from current "
5255 				    "setting irqno %d",
5256 				    irq, ddi_get_name(dip),
5257 				    ddi_get_instance(dip), cur_irq));
5258 		}
5259 
5260 		/*
5261 		 * return the irq that was set, and not what CRS reports,
5262 		 * since CRS has been seen to be bogus on some systems
5263 		 */
5264 		cur_irq = irq;
5265 	} else {
5266 		APIC_VERBOSE_IRQ((CE_WARN, "!pcplusmp: set resource irq %d "
5267 		    "failed for device %s instance #%d",
5268 		    irq, ddi_get_name(dip), ddi_get_instance(dip)));
5269 
5270 		if (cur_irq == -1)
5271 			return (ACPI_PSM_FAILURE);
5272 	}
5273 
5274 	ASSERT(pci_irqp != NULL);
5275 	*pci_irqp = cur_irq;
5276 	return (ACPI_PSM_SUCCESS);
5277 }
5278