xref: /titanic_41/usr/src/uts/i86pc/io/pcplusmp/apic.c (revision 749f21d359d8fbd020c974a1a5227316221bfc9c)
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 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 = 0;
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 boolean_t
1447 apic_cpu_in_range(int cpu)
1448 {
1449 	return ((cpu & ~IRQ_USER_BOUND) < apic_nproc);
1450 }
1451 
1452 static struct apic_mpfps_hdr *
1453 apic_find_fps_sig(caddr_t cptr, int len)
1454 {
1455 	int	i;
1456 
1457 	/* Look for the pattern "_MP_" */
1458 	for (i = 0; i < len; i += 16) {
1459 		if ((*(cptr+i) == '_') &&
1460 		    (*(cptr+i+1) == 'M') &&
1461 		    (*(cptr+i+2) == 'P') &&
1462 		    (*(cptr+i+3) == '_'))
1463 		    /*LINTED: pointer cast may result in improper alignment */
1464 			return ((struct apic_mpfps_hdr *)(cptr + i));
1465 	}
1466 	return (NULL);
1467 }
1468 
1469 static int
1470 apic_checksum(caddr_t bptr, int len)
1471 {
1472 	int	i;
1473 	uchar_t	cksum;
1474 
1475 	cksum = 0;
1476 	for (i = 0; i < len; i++)
1477 		cksum += *bptr++;
1478 	return ((int)cksum);
1479 }
1480 
1481 
1482 /*
1483  * Initialise vector->ipl and ipl->pri arrays. level_intr and irqtable
1484  * are also set to NULL. vector->irq is set to a value which cannot map
1485  * to a real irq to show that it is free.
1486  */
1487 void
1488 apic_init()
1489 {
1490 	int	i;
1491 	int	*iptr;
1492 
1493 	int	j = 1;
1494 	apic_ipltopri[0] = APIC_VECTOR_PER_IPL; /* leave 0 for idle */
1495 	for (i = 0; i < (APIC_AVAIL_VECTOR / APIC_VECTOR_PER_IPL); i++) {
1496 		if ((i < ((APIC_AVAIL_VECTOR / APIC_VECTOR_PER_IPL) - 1)) &&
1497 		    (apic_vectortoipl[i + 1] == apic_vectortoipl[i]))
1498 			/* get to highest vector at the same ipl */
1499 			continue;
1500 		for (; j <= apic_vectortoipl[i]; j++) {
1501 			apic_ipltopri[j] = (i << APIC_IPL_SHIFT) +
1502 			    APIC_BASE_VECT;
1503 		}
1504 	}
1505 	for (; j < MAXIPL + 1; j++)
1506 		/* fill up any empty ipltopri slots */
1507 		apic_ipltopri[j] = (i << APIC_IPL_SHIFT) + APIC_BASE_VECT;
1508 
1509 	/* cpu 0 is always up */
1510 	apic_cpus[0].aci_status = APIC_CPU_ONLINE | APIC_CPU_INTR_ENABLE;
1511 
1512 	iptr = (int *)&apic_irq_table[0];
1513 	for (i = 0; i <= APIC_MAX_VECTOR; i++) {
1514 		apic_level_intr[i] = 0;
1515 		*iptr++ = NULL;
1516 		apic_vector_to_irq[i] = APIC_RESV_IRQ;
1517 		apic_reprogram_info[i].valid = 0;
1518 		apic_reprogram_info[i].bindcpu = 0;
1519 		apic_reprogram_info[i].timeouts = 0;
1520 	}
1521 
1522 	/*
1523 	 * Allocate a dummy irq table entry for the reserved entry.
1524 	 * This takes care of the race between removing an irq and
1525 	 * clock detecting a CPU in that irq during interrupt load
1526 	 * sampling.
1527 	 */
1528 	apic_irq_table[APIC_RESV_IRQ] =
1529 	    kmem_zalloc(sizeof (apic_irq_t), KM_NOSLEEP);
1530 
1531 	mutex_init(&airq_mutex, NULL, MUTEX_DEFAULT, NULL);
1532 	mutex_init(&apic_reprogram_timeout_mutex, NULL, MUTEX_DEFAULT, NULL);
1533 #if defined(__amd64)
1534 	/*
1535 	 * Make cpu-specific interrupt info point to cr8pri vector
1536 	 */
1537 	for (i = 0; i <= MAXIPL; i++)
1538 		apic_cr8pri[i] = apic_ipltopri[i] >> APIC_IPL_SHIFT;
1539 	CPU->cpu_pri_data = apic_cr8pri;
1540 	intpri_use_cr8 = 1;
1541 #endif	/* __amd64 */
1542 }
1543 
1544 /*
1545  * handler for APIC Error interrupt. Just print a warning and continue
1546  */
1547 static int
1548 apic_error_intr()
1549 {
1550 	uint_t	error0, error1, error;
1551 	uint_t	i;
1552 
1553 	/*
1554 	 * We need to write before read as per 7.4.17 of system prog manual.
1555 	 * We do both and or the results to be safe
1556 	 */
1557 	error0 = apicadr[APIC_ERROR_STATUS];
1558 	apicadr[APIC_ERROR_STATUS] = 0;
1559 	error1 = apicadr[APIC_ERROR_STATUS];
1560 	error = error0 | error1;
1561 
1562 	/*
1563 	 * Clear the APIC error status (do this on all cpus that enter here)
1564 	 * (two writes are required due to the semantics of accessing the
1565 	 * error status register.)
1566 	 */
1567 	apicadr[APIC_ERROR_STATUS] = 0;
1568 	apicadr[APIC_ERROR_STATUS] = 0;
1569 
1570 	/*
1571 	 * Prevent more than 1 CPU from handling error interrupt causing
1572 	 * double printing (interleave of characters from multiple
1573 	 * CPU's when using prom_printf)
1574 	 */
1575 	if (lock_try(&apic_error_lock) == 0)
1576 		return (error ? DDI_INTR_CLAIMED : DDI_INTR_UNCLAIMED);
1577 	if (error) {
1578 #if	DEBUG
1579 		if (apic_debug)
1580 			debug_enter("pcplusmp: APIC Error interrupt received");
1581 #endif /* DEBUG */
1582 		if (apic_panic_on_apic_error)
1583 			cmn_err(CE_PANIC,
1584 			    "APIC Error interrupt on CPU %d. Status = %x\n",
1585 			    psm_get_cpu_id(), error);
1586 		else {
1587 			if ((error & ~APIC_CS_ERRORS) == 0) {
1588 				/* cksum error only */
1589 				apic_error |= APIC_ERR_APIC_ERROR;
1590 				apic_apic_error |= error;
1591 				apic_num_apic_errors++;
1592 				apic_num_cksum_errors++;
1593 			} else {
1594 				/*
1595 				 * prom_printf is the best shot we have of
1596 				 * something which is problem free from
1597 				 * high level/NMI type of interrupts
1598 				 */
1599 				prom_printf("APIC Error interrupt on CPU %d. "
1600 				    "Status 0 = %x, Status 1 = %x\n",
1601 				    psm_get_cpu_id(), error0, error1);
1602 				apic_error |= APIC_ERR_APIC_ERROR;
1603 				apic_apic_error |= error;
1604 				apic_num_apic_errors++;
1605 				for (i = 0; i < apic_error_display_delay; i++) {
1606 					tenmicrosec();
1607 				}
1608 				/*
1609 				 * provide more delay next time limited to
1610 				 * roughly 1 clock tick time
1611 				 */
1612 				if (apic_error_display_delay < 500)
1613 					apic_error_display_delay *= 2;
1614 			}
1615 		}
1616 		lock_clear(&apic_error_lock);
1617 		return (DDI_INTR_CLAIMED);
1618 	} else {
1619 		lock_clear(&apic_error_lock);
1620 		return (DDI_INTR_UNCLAIMED);
1621 	}
1622 	/* NOTREACHED */
1623 }
1624 
1625 /*
1626  * Turn off the mask bit in the performance counter Local Vector Table entry.
1627  */
1628 static void
1629 apic_cpcovf_mask_clear(void)
1630 {
1631 	apicadr[APIC_PCINT_VECT] &= ~APIC_LVT_MASK;
1632 }
1633 
1634 static void
1635 apic_init_intr()
1636 {
1637 	processorid_t	cpun = psm_get_cpu_id();
1638 
1639 #if defined(__amd64)
1640 	setcr8((ulong_t)(APIC_MASK_ALL >> APIC_IPL_SHIFT));
1641 #else
1642 	apicadr[APIC_TASK_REG] = APIC_MASK_ALL;
1643 #endif
1644 
1645 	if (apic_flat_model)
1646 		apicadr[APIC_FORMAT_REG] = APIC_FLAT_MODEL;
1647 	else
1648 		apicadr[APIC_FORMAT_REG] = APIC_CLUSTER_MODEL;
1649 	apicadr[APIC_DEST_REG] = AV_HIGH_ORDER >> cpun;
1650 
1651 	/* need to enable APIC before unmasking NMI */
1652 	apicadr[APIC_SPUR_INT_REG] = AV_UNIT_ENABLE | APIC_SPUR_INTR;
1653 
1654 	apicadr[APIC_LOCAL_TIMER] = AV_MASK;
1655 	apicadr[APIC_INT_VECT0]	= AV_MASK;	/* local intr reg 0 */
1656 	apicadr[APIC_INT_VECT1] = AV_NMI;	/* enable NMI */
1657 
1658 	if (apic_cpus[cpun].aci_local_ver < APIC_INTEGRATED_VERS)
1659 		return;
1660 
1661 	/* Enable performance counter overflow interrupt */
1662 
1663 	if ((x86_feature & X86_MSR) != X86_MSR)
1664 		apic_enable_cpcovf_intr = 0;
1665 	if (apic_enable_cpcovf_intr) {
1666 		if (apic_cpcovf_vect == 0) {
1667 			int ipl = APIC_PCINT_IPL;
1668 			int irq = apic_get_ipivect(ipl, -1);
1669 
1670 			ASSERT(irq != -1);
1671 			apic_cpcovf_vect = apic_irq_table[irq]->airq_vector;
1672 			ASSERT(apic_cpcovf_vect);
1673 			(void) add_avintr(NULL, ipl,
1674 			    (avfunc)kcpc_hw_overflow_intr,
1675 			    "apic pcint", irq, NULL, NULL, NULL, NULL);
1676 			kcpc_hw_overflow_intr_installed = 1;
1677 			kcpc_hw_enable_cpc_intr = apic_cpcovf_mask_clear;
1678 		}
1679 		apicadr[APIC_PCINT_VECT] = apic_cpcovf_vect;
1680 	}
1681 
1682 	/* Enable error interrupt */
1683 
1684 	if (apic_enable_error_intr) {
1685 		if (apic_errvect == 0) {
1686 			int ipl = 0xf;	/* get highest priority intr */
1687 			int irq = apic_get_ipivect(ipl, -1);
1688 
1689 			ASSERT(irq != -1);
1690 			apic_errvect = apic_irq_table[irq]->airq_vector;
1691 			ASSERT(apic_errvect);
1692 			/*
1693 			 * Not PSMI compliant, but we are going to merge
1694 			 * with ON anyway
1695 			 */
1696 			(void) add_avintr((void *)NULL, ipl,
1697 			    (avfunc)apic_error_intr, "apic error intr",
1698 			    irq, NULL, NULL, NULL, NULL);
1699 		}
1700 		apicadr[APIC_ERR_VECT] = apic_errvect;
1701 		apicadr[APIC_ERROR_STATUS] = 0;
1702 		apicadr[APIC_ERROR_STATUS] = 0;
1703 	}
1704 }
1705 
1706 static void
1707 apic_disable_local_apic()
1708 {
1709 	apicadr[APIC_TASK_REG] = APIC_MASK_ALL;
1710 	apicadr[APIC_LOCAL_TIMER] = AV_MASK;
1711 	apicadr[APIC_INT_VECT0] = AV_MASK;	/* local intr reg 0 */
1712 	apicadr[APIC_INT_VECT1] = AV_MASK;	/* disable NMI */
1713 	apicadr[APIC_ERR_VECT] = AV_MASK;	/* and error interrupt */
1714 	apicadr[APIC_PCINT_VECT] = AV_MASK;	/* and perf counter intr */
1715 	apicadr[APIC_SPUR_INT_REG] = APIC_SPUR_INTR;
1716 }
1717 
1718 static void
1719 apic_picinit(void)
1720 {
1721 	int i, j;
1722 	uint_t isr;
1723 	volatile int32_t *ioapic;
1724 	apic_irq_t	*irqptr;
1725 	struct intrspec ispec;
1726 
1727 	/*
1728 	 * On UniSys Model 6520, the BIOS leaves vector 0x20 isr
1729 	 * bit on without clearing it with EOI.  Since softint
1730 	 * uses vector 0x20 to interrupt itself, so softint will
1731 	 * not work on this machine.  In order to fix this problem
1732 	 * a check is made to verify all the isr bits are clear.
1733 	 * If not, EOIs are issued to clear the bits.
1734 	 */
1735 	for (i = 7; i >= 1; i--) {
1736 		if ((isr = apicadr[APIC_ISR_REG + (i * 4)]) != 0)
1737 			for (j = 0; ((j < 32) && (isr != 0)); j++)
1738 				if (isr & (1 << j)) {
1739 					apicadr[APIC_EOI_REG] = 0;
1740 					isr &= ~(1 << j);
1741 					apic_error |= APIC_ERR_BOOT_EOI;
1742 				}
1743 	}
1744 
1745 	/* set a flag so we know we have run apic_picinit() */
1746 	apic_flag = 1;
1747 	LOCK_INIT_CLEAR(&apic_gethrtime_lock);
1748 	LOCK_INIT_CLEAR(&apic_ioapic_lock);
1749 	LOCK_INIT_CLEAR(&apic_revector_lock);
1750 	LOCK_INIT_CLEAR(&apic_ioapic_reprogram_lock);
1751 	LOCK_INIT_CLEAR(&apic_error_lock);
1752 
1753 	picsetup();	 /* initialise the 8259 */
1754 
1755 	/* add nmi handler - least priority nmi handler */
1756 	LOCK_INIT_CLEAR(&apic_nmi_lock);
1757 
1758 	if (!psm_add_nmintr(0, (avfunc) apic_nmi_intr,
1759 	    "pcplusmp NMI handler", (caddr_t)NULL))
1760 		cmn_err(CE_WARN, "pcplusmp: Unable to add nmi handler");
1761 
1762 	apic_init_intr();
1763 
1764 	/* enable apic mode if imcr present */
1765 	if (apic_imcrp) {
1766 		outb(APIC_IMCR_P1, (uchar_t)APIC_IMCR_SELECT);
1767 		outb(APIC_IMCR_P2, (uchar_t)APIC_IMCR_APIC);
1768 	}
1769 
1770 	/* mask interrupt vectors					*/
1771 	for (j = 0; j < apic_io_max; j++) {
1772 		int intin_max;
1773 		ioapic = apicioadr[j];
1774 		ioapic[APIC_IO_REG] = APIC_VERS_CMD;
1775 		/* Bits 23-16 define the maximum redirection entries */
1776 		intin_max = (ioapic[APIC_IO_DATA] >> 16) & 0xff;
1777 		for (i = 0; i < intin_max; i++) {
1778 			ioapic[APIC_IO_REG] = APIC_RDT_CMD + 2 * i;
1779 			ioapic[APIC_IO_DATA] = AV_MASK;
1780 		}
1781 	}
1782 
1783 	/*
1784 	 * Hack alert: deal with ACPI SCI interrupt chicken/egg here
1785 	 */
1786 	if (apic_sci_vect > 0) {
1787 		/*
1788 		 * acpica has already done add_avintr(); we just
1789 		 * to finish the job by mimicing translate_irq()
1790 		 *
1791 		 * Fake up an intrspec and setup the tables
1792 		 */
1793 		ispec.intrspec_vec = apic_sci_vect;
1794 		ispec.intrspec_pri = SCI_IPL;
1795 
1796 		if (apic_setup_irq_table(NULL, apic_sci_vect, NULL,
1797 		    &ispec, &apic_sci_flags, DDI_INTR_TYPE_FIXED) < 0) {
1798 			cmn_err(CE_WARN, "!apic: SCI setup failed");
1799 			return;
1800 		}
1801 		irqptr = apic_irq_table[apic_sci_vect];
1802 
1803 		/* Program I/O APIC */
1804 		(void) apic_setup_io_intr(irqptr, apic_sci_vect);
1805 
1806 		irqptr->airq_share++;
1807 	}
1808 }
1809 
1810 
1811 static void
1812 apic_cpu_start(processorid_t cpun, caddr_t rm_code)
1813 {
1814 	int		loop_count;
1815 	uint32_t	vector;
1816 	uint_t		cpu_id, iflag;
1817 
1818 	cpu_id = apic_cpus[cpun].aci_local_id;
1819 
1820 	apic_cmos_ssb_set = 1;
1821 
1822 	/*
1823 	 * Interrupts on BSP cpu will be disabled during these startup
1824 	 * steps in order to avoid unwanted side effects from
1825 	 * executing interrupt handlers on a problematic BIOS.
1826 	 */
1827 
1828 	iflag = intr_clear();
1829 	outb(CMOS_ADDR, SSB);
1830 	outb(CMOS_DATA, BIOS_SHUTDOWN);
1831 
1832 	while (get_apic_cmd1() & AV_PENDING)
1833 		apic_ret();
1834 
1835 	/* for integrated - make sure there is one INIT IPI in buffer */
1836 	/* for external - it will wake up the cpu */
1837 	apicadr[APIC_INT_CMD2] = cpu_id << APIC_ICR_ID_BIT_OFFSET;
1838 	apicadr[APIC_INT_CMD1] = AV_ASSERT | AV_RESET;
1839 
1840 	/* If only 1 CPU is installed, PENDING bit will not go low */
1841 	for (loop_count = 0x1000; loop_count; loop_count--)
1842 		if (get_apic_cmd1() & AV_PENDING)
1843 			apic_ret();
1844 		else
1845 			break;
1846 
1847 	apicadr[APIC_INT_CMD2] = cpu_id << APIC_ICR_ID_BIT_OFFSET;
1848 	apicadr[APIC_INT_CMD1] = AV_DEASSERT | AV_RESET;
1849 
1850 	drv_usecwait(20000);		/* 20 milli sec */
1851 
1852 	if (apic_cpus[cpun].aci_local_ver >= APIC_INTEGRATED_VERS) {
1853 		/* integrated apic */
1854 
1855 		rm_code = (caddr_t)(uintptr_t)rm_platter_pa;
1856 		vector = (rm_platter_pa >> MMU_PAGESHIFT) &
1857 		    (APIC_VECTOR_MASK | APIC_IPL_MASK);
1858 
1859 		/* to offset the INIT IPI queue up in the buffer */
1860 		apicadr[APIC_INT_CMD2] = cpu_id << APIC_ICR_ID_BIT_OFFSET;
1861 		apicadr[APIC_INT_CMD1] = vector | AV_STARTUP;
1862 
1863 		drv_usecwait(200);		/* 20 micro sec */
1864 
1865 		apicadr[APIC_INT_CMD2] = cpu_id << APIC_ICR_ID_BIT_OFFSET;
1866 		apicadr[APIC_INT_CMD1] = vector | AV_STARTUP;
1867 
1868 		drv_usecwait(200);		/* 20 micro sec */
1869 	}
1870 	intr_restore(iflag);
1871 }
1872 
1873 
1874 #ifdef	DEBUG
1875 int	apic_break_on_cpu = 9;
1876 int	apic_stretch_interrupts = 0;
1877 int	apic_stretch_ISR = 1 << 3;	/* IPL of 3 matches nothing now */
1878 
1879 void
1880 apic_break()
1881 {
1882 }
1883 #endif /* DEBUG */
1884 
1885 /*
1886  * platform_intr_enter
1887  *
1888  *	Called at the beginning of the interrupt service routine to
1889  *	mask all level equal to and below the interrupt priority
1890  *	of the interrupting vector.  An EOI should be given to
1891  *	the interrupt controller to enable other HW interrupts.
1892  *
1893  *	Return -1 for spurious interrupts
1894  *
1895  */
1896 /*ARGSUSED*/
1897 static int
1898 apic_intr_enter(int ipl, int *vectorp)
1899 {
1900 	uchar_t vector;
1901 	int nipl;
1902 	int irq, iflag;
1903 	apic_cpus_info_t *cpu_infop;
1904 
1905 	/*
1906 	 * The real vector programmed in APIC is *vectorp + 0x20
1907 	 * But, cmnint code subtracts 0x20 before pushing it.
1908 	 * Hence APIC_BASE_VECT is 0x20.
1909 	 */
1910 
1911 	vector = (uchar_t)*vectorp;
1912 
1913 	/* if interrupted by the clock, increment apic_nsec_since_boot */
1914 	if (vector == apic_clkvect) {
1915 		if (!apic_oneshot) {
1916 			/* NOTE: this is not MT aware */
1917 			apic_hrtime_stamp++;
1918 			apic_nsec_since_boot += apic_nsec_per_intr;
1919 			apic_hrtime_stamp++;
1920 			last_count_read = apic_hertz_count;
1921 			apic_redistribute_compute();
1922 		}
1923 
1924 		/* We will avoid all the book keeping overhead for clock */
1925 		nipl = apic_vectortoipl[vector >> APIC_IPL_SHIFT];
1926 #if defined(__amd64)
1927 		setcr8((ulong_t)apic_cr8pri[nipl]);
1928 #else
1929 		apicadr[APIC_TASK_REG] = apic_ipltopri[nipl];
1930 #endif
1931 		*vectorp = apic_vector_to_irq[vector + APIC_BASE_VECT];
1932 		apicadr[APIC_EOI_REG] = 0;
1933 		return (nipl);
1934 	}
1935 
1936 	cpu_infop = &apic_cpus[psm_get_cpu_id()];
1937 
1938 	if (vector == (APIC_SPUR_INTR - APIC_BASE_VECT)) {
1939 		cpu_infop->aci_spur_cnt++;
1940 		return (APIC_INT_SPURIOUS);
1941 	}
1942 
1943 	/* Check if the vector we got is really what we need */
1944 	if (apic_revector_pending) {
1945 		/*
1946 		 * Disable interrupts for the duration of
1947 		 * the vector translation to prevent a self-race for
1948 		 * the apic_revector_lock.  This cannot be done
1949 		 * in apic_xlate_vector because it is recursive and
1950 		 * we want the vector translation to be atomic with
1951 		 * respect to other (higher-priority) interrupts.
1952 		 */
1953 		iflag = intr_clear();
1954 		vector = apic_xlate_vector(vector + APIC_BASE_VECT) -
1955 		    APIC_BASE_VECT;
1956 		intr_restore(iflag);
1957 	}
1958 
1959 	nipl = apic_vectortoipl[vector >> APIC_IPL_SHIFT];
1960 	*vectorp = irq = apic_vector_to_irq[vector + APIC_BASE_VECT];
1961 
1962 #if defined(__amd64)
1963 	setcr8((ulong_t)apic_cr8pri[nipl]);
1964 #else
1965 	apicadr[APIC_TASK_REG] = apic_ipltopri[nipl];
1966 #endif
1967 
1968 	cpu_infop->aci_current[nipl] = (uchar_t)irq;
1969 	cpu_infop->aci_curipl = (uchar_t)nipl;
1970 	cpu_infop->aci_ISR_in_progress |= 1 << nipl;
1971 
1972 	/*
1973 	 * apic_level_intr could have been assimilated into the irq struct.
1974 	 * but, having it as a character array is more efficient in terms of
1975 	 * cache usage. So, we leave it as is.
1976 	 */
1977 	if (!apic_level_intr[irq])
1978 		apicadr[APIC_EOI_REG] = 0;
1979 
1980 #ifdef	DEBUG
1981 	APIC_DEBUG_BUF_PUT(vector);
1982 	APIC_DEBUG_BUF_PUT(irq);
1983 	APIC_DEBUG_BUF_PUT(nipl);
1984 	APIC_DEBUG_BUF_PUT(psm_get_cpu_id());
1985 	if ((apic_stretch_interrupts) && (apic_stretch_ISR & (1 << nipl)))
1986 		drv_usecwait(apic_stretch_interrupts);
1987 
1988 	if (apic_break_on_cpu == psm_get_cpu_id())
1989 		apic_break();
1990 #endif /* DEBUG */
1991 	return (nipl);
1992 }
1993 
1994 static void
1995 apic_intr_exit(int prev_ipl, int irq)
1996 {
1997 	apic_cpus_info_t *cpu_infop;
1998 
1999 #if defined(__amd64)
2000 	setcr8((ulong_t)apic_cr8pri[prev_ipl]);
2001 #else
2002 	apicadr[APIC_TASK_REG] = apic_ipltopri[prev_ipl];
2003 #endif
2004 
2005 	cpu_infop = &apic_cpus[psm_get_cpu_id()];
2006 	if (apic_level_intr[irq])
2007 		apicadr[APIC_EOI_REG] = 0;
2008 
2009 	cpu_infop->aci_curipl = (uchar_t)prev_ipl;
2010 	/* ISR above current pri could not be in progress */
2011 	cpu_infop->aci_ISR_in_progress &= (2 << prev_ipl) - 1;
2012 }
2013 
2014 /*
2015  * Mask all interrupts below or equal to the given IPL
2016  */
2017 static void
2018 apic_setspl(int ipl)
2019 {
2020 
2021 #if defined(__amd64)
2022 	setcr8((ulong_t)apic_cr8pri[ipl]);
2023 #else
2024 	apicadr[APIC_TASK_REG] = apic_ipltopri[ipl];
2025 #endif
2026 
2027 	/* interrupts at ipl above this cannot be in progress */
2028 	apic_cpus[psm_get_cpu_id()].aci_ISR_in_progress &= (2 << ipl) - 1;
2029 	/*
2030 	 * this is a patch fix for the ALR QSMP P5 machine, so that interrupts
2031 	 * have enough time to come in before the priority is raised again
2032 	 * during the idle() loop.
2033 	 */
2034 	if (apic_setspl_delay)
2035 		(void) get_apic_pri();
2036 }
2037 
2038 /*
2039  * trigger a software interrupt at the given IPL
2040  */
2041 static void
2042 apic_set_softintr(int ipl)
2043 {
2044 	int vector;
2045 	uint_t flag;
2046 
2047 	vector = apic_resv_vector[ipl];
2048 
2049 	flag = intr_clear();
2050 
2051 	while (get_apic_cmd1() & AV_PENDING)
2052 		apic_ret();
2053 
2054 	/* generate interrupt at vector on itself only */
2055 	apicadr[APIC_INT_CMD1] = AV_SH_SELF | vector;
2056 
2057 	intr_restore(flag);
2058 }
2059 
2060 /*
2061  * generates an interprocessor interrupt to another CPU
2062  */
2063 static void
2064 apic_send_ipi(int cpun, int ipl)
2065 {
2066 	int vector;
2067 	uint_t flag;
2068 
2069 	vector = apic_resv_vector[ipl];
2070 
2071 	flag = intr_clear();
2072 
2073 	while (get_apic_cmd1() & AV_PENDING)
2074 		apic_ret();
2075 
2076 	apicadr[APIC_INT_CMD2] =
2077 	    apic_cpus[cpun].aci_local_id << APIC_ICR_ID_BIT_OFFSET;
2078 	apicadr[APIC_INT_CMD1] = vector;
2079 
2080 	intr_restore(flag);
2081 }
2082 
2083 
2084 /*ARGSUSED*/
2085 static void
2086 apic_set_idlecpu(processorid_t cpun)
2087 {
2088 }
2089 
2090 /*ARGSUSED*/
2091 static void
2092 apic_unset_idlecpu(processorid_t cpun)
2093 {
2094 }
2095 
2096 
2097 static void
2098 apic_ret()
2099 {
2100 }
2101 
2102 static int
2103 get_apic_cmd1()
2104 {
2105 	return (apicadr[APIC_INT_CMD1]);
2106 }
2107 
2108 static int
2109 get_apic_pri()
2110 {
2111 #if defined(__amd64)
2112 	return ((int)getcr8());
2113 #else
2114 	return (apicadr[APIC_TASK_REG]);
2115 #endif
2116 }
2117 
2118 /*
2119  * If apic_coarse_time == 1, then apic_gettime() is used instead of
2120  * apic_gethrtime().  This is used for performance instead of accuracy.
2121  */
2122 
2123 static hrtime_t
2124 apic_gettime()
2125 {
2126 	int old_hrtime_stamp;
2127 	hrtime_t temp;
2128 
2129 	/*
2130 	 * In one-shot mode, we do not keep time, so if anyone
2131 	 * calls psm_gettime() directly, we vector over to
2132 	 * gethrtime().
2133 	 * one-shot mode MUST NOT be enabled if this psm is the source of
2134 	 * hrtime.
2135 	 */
2136 
2137 	if (apic_oneshot)
2138 		return (gethrtime());
2139 
2140 
2141 gettime_again:
2142 	while ((old_hrtime_stamp = apic_hrtime_stamp) & 1)
2143 		apic_ret();
2144 
2145 	temp = apic_nsec_since_boot;
2146 
2147 	if (apic_hrtime_stamp != old_hrtime_stamp) {	/* got an interrupt */
2148 		goto gettime_again;
2149 	}
2150 	return (temp);
2151 }
2152 
2153 /*
2154  * Here we return the number of nanoseconds since booting.  Note every
2155  * clock interrupt increments apic_nsec_since_boot by the appropriate
2156  * amount.
2157  */
2158 static hrtime_t
2159 apic_gethrtime()
2160 {
2161 	int curr_timeval, countval, elapsed_ticks, oflags;
2162 	int old_hrtime_stamp, status;
2163 	hrtime_t temp;
2164 	uchar_t	cpun;
2165 
2166 
2167 	/*
2168 	 * In one-shot mode, we do not keep time, so if anyone
2169 	 * calls psm_gethrtime() directly, we vector over to
2170 	 * gethrtime().
2171 	 * one-shot mode MUST NOT be enabled if this psm is the source of
2172 	 * hrtime.
2173 	 */
2174 
2175 	if (apic_oneshot)
2176 		return (gethrtime());
2177 
2178 	oflags = intr_clear();	/* prevent migration */
2179 
2180 	cpun = (uchar_t)((uint_t)apicadr[APIC_LID_REG] >> APIC_ID_BIT_OFFSET);
2181 
2182 	lock_set(&apic_gethrtime_lock);
2183 
2184 gethrtime_again:
2185 	while ((old_hrtime_stamp = apic_hrtime_stamp) & 1)
2186 		apic_ret();
2187 
2188 	/*
2189 	 * Check to see which CPU we are on.  Note the time is kept on
2190 	 * the local APIC of CPU 0.  If on CPU 0, simply read the current
2191 	 * counter.  If on another CPU, issue a remote read command to CPU 0.
2192 	 */
2193 	if (cpun == apic_cpus[0].aci_local_id) {
2194 		countval = apicadr[APIC_CURR_COUNT];
2195 	} else {
2196 		while (get_apic_cmd1() & AV_PENDING)
2197 			apic_ret();
2198 
2199 		apicadr[APIC_INT_CMD2] =
2200 		    apic_cpus[0].aci_local_id << APIC_ICR_ID_BIT_OFFSET;
2201 		apicadr[APIC_INT_CMD1] = APIC_CURR_ADD|AV_REMOTE;
2202 
2203 		while ((status = get_apic_cmd1()) & AV_READ_PENDING)
2204 			apic_ret();
2205 
2206 		if (status & AV_REMOTE_STATUS)	/* 1 = valid */
2207 			countval = apicadr[APIC_REMOTE_READ];
2208 		else {	/* 0 = invalid */
2209 			apic_remote_hrterr++;
2210 			/*
2211 			 * return last hrtime right now, will need more
2212 			 * testing if change to retry
2213 			 */
2214 			temp = apic_last_hrtime;
2215 
2216 			lock_clear(&apic_gethrtime_lock);
2217 
2218 			intr_restore(oflags);
2219 
2220 			return (temp);
2221 		}
2222 	}
2223 	if (countval > last_count_read)
2224 		countval = 0;
2225 	else
2226 		last_count_read = countval;
2227 
2228 	elapsed_ticks = apic_hertz_count - countval;
2229 
2230 	curr_timeval = elapsed_ticks * apic_nsec_per_tick;
2231 	temp = apic_nsec_since_boot + curr_timeval;
2232 
2233 	if (apic_hrtime_stamp != old_hrtime_stamp) {	/* got an interrupt */
2234 		/* we might have clobbered last_count_read. Restore it */
2235 		last_count_read = apic_hertz_count;
2236 		goto gethrtime_again;
2237 	}
2238 
2239 	if (temp < apic_last_hrtime) {
2240 		/* return last hrtime if error occurs */
2241 		apic_hrtime_error++;
2242 		temp = apic_last_hrtime;
2243 	}
2244 	else
2245 		apic_last_hrtime = temp;
2246 
2247 	lock_clear(&apic_gethrtime_lock);
2248 	intr_restore(oflags);
2249 
2250 	return (temp);
2251 }
2252 
2253 /* apic NMI handler */
2254 /*ARGSUSED*/
2255 static void
2256 apic_nmi_intr(caddr_t arg)
2257 {
2258 	if (apic_shutdown_processors) {
2259 		apic_disable_local_apic();
2260 		return;
2261 	}
2262 
2263 	if (lock_try(&apic_nmi_lock)) {
2264 		if (apic_kmdb_on_nmi) {
2265 			if (psm_debugger() == 0) {
2266 				cmn_err(CE_PANIC,
2267 				    "NMI detected, kmdb is not available.");
2268 			} else {
2269 				debug_enter("\nNMI detected, entering kmdb.\n");
2270 			}
2271 		} else {
2272 			if (apic_panic_on_nmi) {
2273 				/* Keep panic from entering kmdb. */
2274 				nopanicdebug = 1;
2275 				cmn_err(CE_PANIC, "pcplusmp: NMI received");
2276 			} else {
2277 				/*
2278 				 * prom_printf is the best shot we have
2279 				 * of something which is problem free from
2280 				 * high level/NMI type of interrupts
2281 				 */
2282 				prom_printf("pcplusmp: NMI received\n");
2283 				apic_error |= APIC_ERR_NMI;
2284 				apic_num_nmis++;
2285 			}
2286 		}
2287 		lock_clear(&apic_nmi_lock);
2288 	}
2289 }
2290 
2291 /*
2292  * Add mask bits to disable interrupt vector from happening
2293  * at or above IPL. In addition, it should remove mask bits
2294  * to enable interrupt vectors below the given IPL.
2295  *
2296  * Both add and delspl are complicated by the fact that different interrupts
2297  * may share IRQs. This can happen in two ways.
2298  * 1. The same H/W line is shared by more than 1 device
2299  * 1a. with interrupts at different IPLs
2300  * 1b. with interrupts at same IPL
2301  * 2. We ran out of vectors at a given IPL and started sharing vectors.
2302  * 1b and 2 should be handled gracefully, except for the fact some ISRs
2303  * will get called often when no interrupt is pending for the device.
2304  * For 1a, we just hope that the machine blows up with the person who
2305  * set it up that way!. In the meantime, we handle it at the higher IPL.
2306  */
2307 /*ARGSUSED*/
2308 static int
2309 apic_addspl(int irqno, int ipl, int min_ipl, int max_ipl)
2310 {
2311 	uchar_t vector;
2312 	int iflag;
2313 	apic_irq_t *irqptr, *irqheadptr;
2314 	int irqindex;
2315 
2316 	ASSERT(max_ipl <= UCHAR_MAX);
2317 	irqindex = IRQINDEX(irqno);
2318 
2319 	if ((irqindex == -1) || (!apic_irq_table[irqindex]))
2320 		return (PSM_FAILURE);
2321 
2322 	irqptr = irqheadptr = apic_irq_table[irqindex];
2323 
2324 	DDI_INTR_IMPLDBG((CE_CONT, "apic_addspl: dip=0x%p type=%d irqno=0x%x "
2325 	    "vector=0x%x\n", (void *)irqptr->airq_dip,
2326 	    irqptr->airq_mps_intr_index, irqno, irqptr->airq_vector));
2327 
2328 	while (irqptr) {
2329 		if (VIRTIRQ(irqindex, irqptr->airq_share_id) == irqno)
2330 			break;
2331 		irqptr = irqptr->airq_next;
2332 	}
2333 	irqptr->airq_share++;
2334 
2335 	/* return if it is not hardware interrupt */
2336 	if (irqptr->airq_mps_intr_index == RESERVE_INDEX)
2337 		return (PSM_SUCCESS);
2338 
2339 	/* Or if there are more interupts at a higher IPL */
2340 	if (ipl != max_ipl)
2341 		return (PSM_SUCCESS);
2342 
2343 	/*
2344 	 * if apic_picinit() has not been called yet, just return.
2345 	 * At the end of apic_picinit(), we will call setup_io_intr().
2346 	 */
2347 
2348 	if (!apic_flag)
2349 		return (PSM_SUCCESS);
2350 
2351 	iflag = intr_clear();
2352 
2353 	/*
2354 	 * Upgrade vector if max_ipl is not earlier ipl. If we cannot allocate,
2355 	 * return failure. Not very elegant, but then we hope the
2356 	 * machine will blow up with ...
2357 	 */
2358 	if (irqptr->airq_ipl != max_ipl) {
2359 		vector = apic_allocate_vector(max_ipl, irqindex, 1);
2360 		if (vector == 0) {
2361 			intr_restore(iflag);
2362 			irqptr->airq_share--;
2363 			return (PSM_FAILURE);
2364 		}
2365 		irqptr = irqheadptr;
2366 		apic_mark_vector(irqptr->airq_vector, vector);
2367 		while (irqptr) {
2368 			irqptr->airq_vector = vector;
2369 			irqptr->airq_ipl = (uchar_t)max_ipl;
2370 			/*
2371 			 * reprogram irq being added and every one else
2372 			 * who is not in the UNINIT state
2373 			 */
2374 			if ((VIRTIRQ(irqindex, irqptr->airq_share_id) ==
2375 			    irqno) || (irqptr->airq_temp_cpu != IRQ_UNINIT)) {
2376 				apic_record_rdt_entry(irqptr, irqindex);
2377 				(void) apic_setup_io_intr(irqptr, irqindex);
2378 			}
2379 			irqptr = irqptr->airq_next;
2380 		}
2381 		intr_restore(iflag);
2382 		return (PSM_SUCCESS);
2383 	}
2384 
2385 	ASSERT(irqptr);
2386 	(void) apic_setup_io_intr(irqptr, irqindex);
2387 	intr_restore(iflag);
2388 	return (PSM_SUCCESS);
2389 }
2390 
2391 /*
2392  * Recompute mask bits for the given interrupt vector.
2393  * If there is no interrupt servicing routine for this
2394  * vector, this function should disable interrupt vector
2395  * from happening at all IPLs. If there are still
2396  * handlers using the given vector, this function should
2397  * disable the given vector from happening below the lowest
2398  * IPL of the remaining hadlers.
2399  */
2400 /*ARGSUSED*/
2401 static int
2402 apic_delspl(int irqno, int ipl, int min_ipl, int max_ipl)
2403 {
2404 	uchar_t vector, bind_cpu;
2405 	int	iflag, intin, irqindex;
2406 	volatile int32_t *ioapic;
2407 	apic_irq_t	*irqptr, *irqheadptr;
2408 
2409 	irqindex = IRQINDEX(irqno);
2410 	irqptr = irqheadptr = apic_irq_table[irqindex];
2411 
2412 	DDI_INTR_IMPLDBG((CE_CONT, "apic_delspl: dip=0x%p type=%d irqno=0x%x "
2413 	    "vector=0x%x\n", (void *)irqptr->airq_dip,
2414 	    irqptr->airq_mps_intr_index, irqno, irqptr->airq_vector));
2415 
2416 	while (irqptr) {
2417 		if (VIRTIRQ(irqindex, irqptr->airq_share_id) == irqno)
2418 			break;
2419 		irqptr = irqptr->airq_next;
2420 	}
2421 	ASSERT(irqptr);
2422 
2423 	irqptr->airq_share--;
2424 
2425 	if (ipl < max_ipl)
2426 		return (PSM_SUCCESS);
2427 
2428 	/* return if it is not hardware interrupt */
2429 	if (irqptr->airq_mps_intr_index == RESERVE_INDEX)
2430 		return (PSM_SUCCESS);
2431 
2432 	if (!apic_flag) {
2433 		/*
2434 		 * Clear irq_struct. If two devices shared an intpt
2435 		 * line & 1 unloaded before picinit, we are hosed. But, then
2436 		 * we hope the machine will ...
2437 		 */
2438 		irqptr->airq_mps_intr_index = FREE_INDEX;
2439 		irqptr->airq_temp_cpu = IRQ_UNINIT;
2440 		apic_free_vector(irqptr->airq_vector);
2441 		return (PSM_SUCCESS);
2442 	}
2443 	/*
2444 	 * Downgrade vector to new max_ipl if needed.If we cannot allocate,
2445 	 * use old IPL. Not very elegant, but then we hope ...
2446 	 */
2447 	if ((irqptr->airq_ipl != max_ipl) && (max_ipl != PSM_INVALID_IPL)) {
2448 		apic_irq_t	*irqp;
2449 		if (vector = apic_allocate_vector(max_ipl, irqno, 1)) {
2450 			apic_mark_vector(irqheadptr->airq_vector, vector);
2451 			irqp = irqheadptr;
2452 			while (irqp) {
2453 				irqp->airq_vector = vector;
2454 				irqp->airq_ipl = (uchar_t)max_ipl;
2455 				if (irqp->airq_temp_cpu != IRQ_UNINIT) {
2456 					apic_record_rdt_entry(irqp, irqindex);
2457 					(void) apic_setup_io_intr(irqp,
2458 					    irqindex);
2459 				}
2460 				irqp = irqp->airq_next;
2461 			}
2462 		}
2463 	}
2464 
2465 	if (irqptr->airq_share)
2466 		return (PSM_SUCCESS);
2467 
2468 	ioapic = apicioadr[irqptr->airq_ioapicindex];
2469 	intin = irqptr->airq_intin_no;
2470 	iflag = intr_clear();
2471 	lock_set(&apic_ioapic_lock);
2472 	ioapic[APIC_IO_REG] = APIC_RDT_CMD + 2 * intin;
2473 	ioapic[APIC_IO_DATA] = AV_MASK;
2474 
2475 	/* Disable the MSI/X vector */
2476 	if (APIC_IS_MSI_OR_MSIX_INDEX(irqptr->airq_mps_intr_index)) {
2477 		int type = (irqptr->airq_mps_intr_index == MSI_INDEX) ?
2478 		    DDI_INTR_TYPE_MSI : DDI_INTR_TYPE_MSIX;
2479 
2480 		/*
2481 		 * Make sure we only disable on the last
2482 		 * of the multi-MSI support
2483 		 */
2484 		if (i_ddi_intr_get_current_nintrs(irqptr->airq_dip) == 1) {
2485 			(void) pci_msi_unconfigure(irqptr->airq_dip, type,
2486 			    irqptr->airq_ioapicindex);
2487 
2488 			(void) pci_msi_disable_mode(irqptr->airq_dip, type,
2489 			    irqptr->airq_ioapicindex);
2490 		}
2491 	}
2492 
2493 	if (max_ipl == PSM_INVALID_IPL) {
2494 		ASSERT(irqheadptr == irqptr);
2495 		bind_cpu = irqptr->airq_temp_cpu;
2496 		if (((uchar_t)bind_cpu != IRQ_UNBOUND) &&
2497 		    ((uchar_t)bind_cpu != IRQ_UNINIT)) {
2498 			ASSERT((bind_cpu & ~IRQ_USER_BOUND) < apic_nproc);
2499 			if (bind_cpu & IRQ_USER_BOUND) {
2500 				/* If hardbound, temp_cpu == cpu */
2501 				bind_cpu &= ~IRQ_USER_BOUND;
2502 				apic_cpus[bind_cpu].aci_bound--;
2503 			} else
2504 				apic_cpus[bind_cpu].aci_temp_bound--;
2505 		}
2506 		lock_clear(&apic_ioapic_lock);
2507 		intr_restore(iflag);
2508 		irqptr->airq_temp_cpu = IRQ_UNINIT;
2509 		irqptr->airq_mps_intr_index = FREE_INDEX;
2510 		apic_free_vector(irqptr->airq_vector);
2511 		return (PSM_SUCCESS);
2512 	}
2513 	lock_clear(&apic_ioapic_lock);
2514 	intr_restore(iflag);
2515 
2516 	mutex_enter(&airq_mutex);
2517 	if ((irqptr == apic_irq_table[irqindex])) {
2518 		apic_irq_t	*oldirqptr;
2519 		/* Move valid irq entry to the head */
2520 		irqheadptr = oldirqptr = irqptr;
2521 		irqptr = irqptr->airq_next;
2522 		ASSERT(irqptr);
2523 		while (irqptr) {
2524 			if (irqptr->airq_mps_intr_index != FREE_INDEX)
2525 				break;
2526 			oldirqptr = irqptr;
2527 			irqptr = irqptr->airq_next;
2528 		}
2529 		/* remove all invalid ones from the beginning */
2530 		apic_irq_table[irqindex] = irqptr;
2531 		/*
2532 		 * and link them back after the head. The invalid ones
2533 		 * begin with irqheadptr and end at oldirqptr
2534 		 */
2535 		oldirqptr->airq_next = irqptr->airq_next;
2536 		irqptr->airq_next = irqheadptr;
2537 	}
2538 	mutex_exit(&airq_mutex);
2539 
2540 	irqptr->airq_temp_cpu = IRQ_UNINIT;
2541 	irqptr->airq_mps_intr_index = FREE_INDEX;
2542 	return (PSM_SUCCESS);
2543 }
2544 
2545 /*
2546  * Return HW interrupt number corresponding to the given IPL
2547  */
2548 /*ARGSUSED*/
2549 static int
2550 apic_softlvl_to_irq(int ipl)
2551 {
2552 	/*
2553 	 * Do not use apic to trigger soft interrupt.
2554 	 * It will cause the system to hang when 2 hardware interrupts
2555 	 * at the same priority with the softint are already accepted
2556 	 * by the apic.  Cause the AV_PENDING bit will not be cleared
2557 	 * until one of the hardware interrupt is eoi'ed.  If we need
2558 	 * to send an ipi at this time, we will end up looping forever
2559 	 * to wait for the AV_PENDING bit to clear.
2560 	 */
2561 	return (PSM_SV_SOFTWARE);
2562 }
2563 
2564 static int
2565 apic_post_cpu_start()
2566 {
2567 	int i, cpun;
2568 	apic_irq_t *irq_ptr;
2569 
2570 	apic_init_intr();
2571 
2572 	/*
2573 	 * since some systems don't enable the internal cache on the non-boot
2574 	 * cpus, so we have to enable them here
2575 	 */
2576 	setcr0(getcr0() & ~(0x60000000));
2577 
2578 	while (get_apic_cmd1() & AV_PENDING)
2579 		apic_ret();
2580 
2581 	cpun = psm_get_cpu_id();
2582 	apic_cpus[cpun].aci_status = APIC_CPU_ONLINE | APIC_CPU_INTR_ENABLE;
2583 
2584 	for (i = apic_min_device_irq; i <= apic_max_device_irq; i++) {
2585 		irq_ptr = apic_irq_table[i];
2586 		if ((irq_ptr == NULL) ||
2587 		    ((irq_ptr->airq_cpu & ~IRQ_USER_BOUND) != cpun))
2588 			continue;
2589 
2590 		while (irq_ptr) {
2591 			if (irq_ptr->airq_temp_cpu != IRQ_UNINIT)
2592 				(void) apic_rebind(irq_ptr, cpun, 1, IMMEDIATE);
2593 			irq_ptr = irq_ptr->airq_next;
2594 		}
2595 	}
2596 
2597 	return (PSM_SUCCESS);
2598 }
2599 
2600 processorid_t
2601 apic_get_next_processorid(processorid_t cpu_id)
2602 {
2603 
2604 	int i;
2605 
2606 	if (cpu_id == -1)
2607 		return ((processorid_t)0);
2608 
2609 	for (i = cpu_id + 1; i < NCPU; i++) {
2610 		if (apic_cpumask & (1 << i))
2611 			return (i);
2612 	}
2613 
2614 	return ((processorid_t)-1);
2615 }
2616 
2617 
2618 /*
2619  * type == -1 indicates it is an internal request. Do not change
2620  * resv_vector for these requests
2621  */
2622 static int
2623 apic_get_ipivect(int ipl, int type)
2624 {
2625 	uchar_t vector;
2626 	int irq;
2627 
2628 	if (irq = apic_allocate_irq(APIC_VECTOR(ipl))) {
2629 		if (vector = apic_allocate_vector(ipl, irq, 1)) {
2630 			apic_irq_table[irq]->airq_mps_intr_index =
2631 			    RESERVE_INDEX;
2632 			apic_irq_table[irq]->airq_vector = vector;
2633 			if (type != -1) {
2634 				apic_resv_vector[ipl] = vector;
2635 			}
2636 			return (irq);
2637 		}
2638 	}
2639 	apic_error |= APIC_ERR_GET_IPIVECT_FAIL;
2640 	return (-1);	/* shouldn't happen */
2641 }
2642 
2643 static int
2644 apic_getclkirq(int ipl)
2645 {
2646 	int	irq;
2647 
2648 	if ((irq = apic_get_ipivect(ipl, -1)) == -1)
2649 		return (-1);
2650 	/*
2651 	 * Note the vector in apic_clkvect for per clock handling.
2652 	 */
2653 	apic_clkvect = apic_irq_table[irq]->airq_vector - APIC_BASE_VECT;
2654 	APIC_VERBOSE_IOAPIC((CE_NOTE, "get_clkirq: vector = %x\n",
2655 	    apic_clkvect));
2656 	return (irq);
2657 }
2658 
2659 /*
2660  * Return the number of APIC clock ticks elapsed for 8245 to decrement
2661  * (APIC_TIME_COUNT + pit_ticks_adj) ticks.
2662  */
2663 static uint_t
2664 apic_calibrate(volatile uint32_t *addr, uint16_t *pit_ticks_adj)
2665 {
2666 	uint8_t		pit_tick_lo;
2667 	uint16_t	pit_tick, target_pit_tick;
2668 	uint32_t	start_apic_tick, end_apic_tick;
2669 	int		iflag;
2670 
2671 	addr += APIC_CURR_COUNT;
2672 
2673 	iflag = intr_clear();
2674 
2675 	do {
2676 		pit_tick_lo = inb(PITCTR0_PORT);
2677 		pit_tick = (inb(PITCTR0_PORT) << 8) | pit_tick_lo;
2678 	} while (pit_tick < APIC_TIME_MIN ||
2679 	    pit_tick_lo <= APIC_LB_MIN || pit_tick_lo >= APIC_LB_MAX);
2680 
2681 	/*
2682 	 * Wait for the 8254 to decrement by 5 ticks to ensure
2683 	 * we didn't start in the middle of a tick.
2684 	 * Compare with 0x10 for the wrap around case.
2685 	 */
2686 	target_pit_tick = pit_tick - 5;
2687 	do {
2688 		pit_tick_lo = inb(PITCTR0_PORT);
2689 		pit_tick = (inb(PITCTR0_PORT) << 8) | pit_tick_lo;
2690 	} while (pit_tick > target_pit_tick || pit_tick_lo < 0x10);
2691 
2692 	start_apic_tick = *addr;
2693 
2694 	/*
2695 	 * Wait for the 8254 to decrement by
2696 	 * (APIC_TIME_COUNT + pit_ticks_adj) ticks
2697 	 */
2698 	target_pit_tick = pit_tick - APIC_TIME_COUNT;
2699 	do {
2700 		pit_tick_lo = inb(PITCTR0_PORT);
2701 		pit_tick = (inb(PITCTR0_PORT) << 8) | pit_tick_lo;
2702 	} while (pit_tick > target_pit_tick || pit_tick_lo < 0x10);
2703 
2704 	end_apic_tick = *addr;
2705 
2706 	*pit_ticks_adj = target_pit_tick - pit_tick;
2707 
2708 	intr_restore(iflag);
2709 
2710 	return (start_apic_tick - end_apic_tick);
2711 }
2712 
2713 /*
2714  * Initialise the APIC timer on the local APIC of CPU 0 to the desired
2715  * frequency.  Note at this stage in the boot sequence, the boot processor
2716  * is the only active processor.
2717  * hertz value of 0 indicates a one-shot mode request.  In this case
2718  * the function returns the resolution (in nanoseconds) for the hardware
2719  * timer interrupt.  If one-shot mode capability is not available,
2720  * the return value will be 0. apic_enable_oneshot is a global switch
2721  * for disabling the functionality.
2722  * A non-zero positive value for hertz indicates a periodic mode request.
2723  * In this case the hardware will be programmed to generate clock interrupts
2724  * at hertz frequency and returns the resolution of interrupts in
2725  * nanosecond.
2726  */
2727 
2728 static int
2729 apic_clkinit(int hertz)
2730 {
2731 
2732 	uint_t		apic_ticks = 0;
2733 	uint_t		pit_time;
2734 	int		ret;
2735 	uint16_t	pit_ticks_adj;
2736 	static int	firsttime = 1;
2737 
2738 	if (firsttime) {
2739 		/* first time calibrate */
2740 
2741 		apicadr[APIC_DIVIDE_REG] = 0x0;
2742 		apicadr[APIC_INIT_COUNT] = APIC_MAXVAL;
2743 
2744 		/* set periodic interrupt based on CLKIN */
2745 		apicadr[APIC_LOCAL_TIMER] =
2746 		    (apic_clkvect + APIC_BASE_VECT) | AV_TIME;
2747 		tenmicrosec();
2748 
2749 		apic_ticks = apic_calibrate(apicadr, &pit_ticks_adj);
2750 
2751 		apicadr[APIC_LOCAL_TIMER] =
2752 		    (apic_clkvect + APIC_BASE_VECT) | AV_MASK;
2753 		/*
2754 		 * pit time is the amount of real time (in nanoseconds ) it took
2755 		 * the 8254 to decrement (APIC_TIME_COUNT + pit_ticks_adj) ticks
2756 		 */
2757 		pit_time = ((longlong_t)(APIC_TIME_COUNT +
2758 		    pit_ticks_adj) * NANOSEC) / PIT_HZ;
2759 
2760 		/*
2761 		 * Determine the number of nanoseconds per APIC clock tick
2762 		 * and then determine how many APIC ticks to interrupt at the
2763 		 * desired frequency
2764 		 */
2765 		apic_nsec_per_tick = pit_time / apic_ticks;
2766 		if (apic_nsec_per_tick == 0)
2767 			apic_nsec_per_tick = 1;
2768 
2769 		/* the interval timer initial count is 32 bit max */
2770 		apic_nsec_max = (hrtime_t)apic_nsec_per_tick * APIC_MAXVAL;
2771 		firsttime = 0;
2772 	}
2773 
2774 	if (hertz != 0) {
2775 		/* periodic */
2776 		apic_nsec_per_intr = NANOSEC / hertz;
2777 		apic_hertz_count = (longlong_t)apic_nsec_per_intr /
2778 		    apic_nsec_per_tick;
2779 		apic_sample_factor_redistribution = hertz + 1;
2780 	}
2781 
2782 	apic_int_busy_mark = (apic_int_busy_mark *
2783 	    apic_sample_factor_redistribution) / 100;
2784 	apic_int_free_mark = (apic_int_free_mark *
2785 	    apic_sample_factor_redistribution) / 100;
2786 	apic_diff_for_redistribution = (apic_diff_for_redistribution *
2787 	    apic_sample_factor_redistribution) / 100;
2788 
2789 	if (hertz == 0) {
2790 		/* requested one_shot */
2791 		if (!apic_oneshot_enable)
2792 			return (0);
2793 		apic_oneshot = 1;
2794 		ret = (int)apic_nsec_per_tick;
2795 	} else {
2796 		/* program the local APIC to interrupt at the given frequency */
2797 		apicadr[APIC_INIT_COUNT] = apic_hertz_count;
2798 		apicadr[APIC_LOCAL_TIMER] =
2799 		    (apic_clkvect + APIC_BASE_VECT) | AV_TIME;
2800 		apic_oneshot = 0;
2801 		ret = NANOSEC / hertz;
2802 	}
2803 
2804 	return (ret);
2805 
2806 }
2807 
2808 /*
2809  * apic_preshutdown:
2810  * Called early in shutdown whilst we can still access filesystems to do
2811  * things like loading modules which will be required to complete shutdown
2812  * after filesystems are all unmounted.
2813  */
2814 static void
2815 apic_preshutdown(int cmd, int fcn)
2816 {
2817 	APIC_VERBOSE_POWEROFF(("apic_preshutdown(%d,%d); m=%d a=%d\n",
2818 	    cmd, fcn, apic_poweroff_method, apic_enable_acpi));
2819 
2820 	if ((cmd != A_SHUTDOWN) || (fcn != AD_POWEROFF)) {
2821 		return;
2822 	}
2823 }
2824 
2825 static void
2826 apic_shutdown(int cmd, int fcn)
2827 {
2828 	int iflag, restarts, attempts;
2829 	int i, j;
2830 	volatile int32_t *ioapic;
2831 	uchar_t	byte;
2832 
2833 	/* Send NMI to all CPUs except self to do per processor shutdown */
2834 	iflag = intr_clear();
2835 	while (get_apic_cmd1() & AV_PENDING)
2836 		apic_ret();
2837 	apic_shutdown_processors = 1;
2838 	apicadr[APIC_INT_CMD1] = AV_NMI | AV_LEVEL | AV_SH_ALL_EXCSELF;
2839 
2840 	/* restore cmos shutdown byte before reboot */
2841 	if (apic_cmos_ssb_set) {
2842 		outb(CMOS_ADDR, SSB);
2843 		outb(CMOS_DATA, 0);
2844 	}
2845 	/* Disable the I/O APIC redirection entries */
2846 	for (j = 0; j < apic_io_max; j++) {
2847 		int intin_max;
2848 		ioapic = apicioadr[j];
2849 		ioapic[APIC_IO_REG] = APIC_VERS_CMD;
2850 		/* Bits 23-16 define the maximum redirection entries */
2851 		intin_max = (ioapic[APIC_IO_DATA] >> 16) & 0xff;
2852 		for (i = 0; i < intin_max; i++) {
2853 			ioapic[APIC_IO_REG] = APIC_RDT_CMD + 2 * i;
2854 			ioapic[APIC_IO_DATA] = AV_MASK;
2855 		}
2856 	}
2857 
2858 	/*	disable apic mode if imcr present	*/
2859 	if (apic_imcrp) {
2860 		outb(APIC_IMCR_P1, (uchar_t)APIC_IMCR_SELECT);
2861 		outb(APIC_IMCR_P2, (uchar_t)APIC_IMCR_PIC);
2862 	}
2863 
2864 	apic_disable_local_apic();
2865 
2866 	intr_restore(iflag);
2867 
2868 	if ((cmd != A_SHUTDOWN) || (fcn != AD_POWEROFF)) {
2869 		return;
2870 	}
2871 
2872 	switch (apic_poweroff_method) {
2873 		case APIC_POWEROFF_VIA_RTC:
2874 
2875 			/* select the extended NVRAM bank in the RTC */
2876 			outb(CMOS_ADDR, RTC_REGA);
2877 			byte = inb(CMOS_DATA);
2878 			outb(CMOS_DATA, (byte | EXT_BANK));
2879 
2880 			outb(CMOS_ADDR, PFR_REG);
2881 
2882 			/* for Predator must toggle the PAB bit */
2883 			byte = inb(CMOS_DATA);
2884 
2885 			/*
2886 			 * clear power active bar, wakeup alarm and
2887 			 * kickstart
2888 			 */
2889 			byte &= ~(PAB_CBIT | WF_FLAG | KS_FLAG);
2890 			outb(CMOS_DATA, byte);
2891 
2892 			/* delay before next write */
2893 			drv_usecwait(1000);
2894 
2895 			/* for S40 the following would suffice */
2896 			byte = inb(CMOS_DATA);
2897 
2898 			/* power active bar control bit */
2899 			byte |= PAB_CBIT;
2900 			outb(CMOS_DATA, byte);
2901 
2902 			break;
2903 
2904 		case APIC_POWEROFF_VIA_ASPEN_BMC:
2905 			restarts = 0;
2906 restart_aspen_bmc:
2907 			if (++restarts == 3)
2908 				break;
2909 			attempts = 0;
2910 			do {
2911 				byte = inb(MISMIC_FLAG_REGISTER);
2912 				byte &= MISMIC_BUSY_MASK;
2913 				if (byte != 0) {
2914 					drv_usecwait(1000);
2915 					if (attempts >= 3)
2916 						goto restart_aspen_bmc;
2917 					++attempts;
2918 				}
2919 			} while (byte != 0);
2920 			outb(MISMIC_CNTL_REGISTER, CC_SMS_GET_STATUS);
2921 			byte = inb(MISMIC_FLAG_REGISTER);
2922 			byte |= 0x1;
2923 			outb(MISMIC_FLAG_REGISTER, byte);
2924 			i = 0;
2925 			for (; i < (sizeof (aspen_bmc)/sizeof (aspen_bmc[0]));
2926 			    i++) {
2927 				attempts = 0;
2928 				do {
2929 					byte = inb(MISMIC_FLAG_REGISTER);
2930 					byte &= MISMIC_BUSY_MASK;
2931 					if (byte != 0) {
2932 						drv_usecwait(1000);
2933 						if (attempts >= 3)
2934 							goto restart_aspen_bmc;
2935 						++attempts;
2936 					}
2937 				} while (byte != 0);
2938 				outb(MISMIC_CNTL_REGISTER, aspen_bmc[i].cntl);
2939 				outb(MISMIC_DATA_REGISTER, aspen_bmc[i].data);
2940 				byte = inb(MISMIC_FLAG_REGISTER);
2941 				byte |= 0x1;
2942 				outb(MISMIC_FLAG_REGISTER, byte);
2943 			}
2944 			break;
2945 
2946 		case APIC_POWEROFF_VIA_SITKA_BMC:
2947 			restarts = 0;
2948 restart_sitka_bmc:
2949 			if (++restarts == 3)
2950 				break;
2951 			attempts = 0;
2952 			do {
2953 				byte = inb(SMS_STATUS_REGISTER);
2954 				byte &= SMS_STATE_MASK;
2955 				if ((byte == SMS_READ_STATE) ||
2956 				    (byte == SMS_WRITE_STATE)) {
2957 					drv_usecwait(1000);
2958 					if (attempts >= 3)
2959 						goto restart_sitka_bmc;
2960 					++attempts;
2961 				}
2962 			} while ((byte == SMS_READ_STATE) ||
2963 			    (byte == SMS_WRITE_STATE));
2964 			outb(SMS_COMMAND_REGISTER, SMS_GET_STATUS);
2965 			i = 0;
2966 			for (; i < (sizeof (sitka_bmc)/sizeof (sitka_bmc[0]));
2967 			    i++) {
2968 				attempts = 0;
2969 				do {
2970 					byte = inb(SMS_STATUS_REGISTER);
2971 					byte &= SMS_IBF_MASK;
2972 					if (byte != 0) {
2973 						drv_usecwait(1000);
2974 						if (attempts >= 3)
2975 							goto restart_sitka_bmc;
2976 						++attempts;
2977 					}
2978 				} while (byte != 0);
2979 				outb(sitka_bmc[i].port, sitka_bmc[i].data);
2980 			}
2981 			break;
2982 
2983 		case APIC_POWEROFF_NONE:
2984 
2985 			/* If no APIC direct method, we will try using ACPI */
2986 			if (apic_enable_acpi) {
2987 				if (acpi_poweroff() == 1)
2988 					return;
2989 			} else
2990 				return;
2991 
2992 			break;
2993 	}
2994 	/*
2995 	 * Wait a limited time here for power to go off.
2996 	 * If the power does not go off, then there was a
2997 	 * problem and we should continue to the halt which
2998 	 * prints a message for the user to press a key to
2999 	 * reboot.
3000 	 */
3001 	drv_usecwait(7000000); /* wait seven seconds */
3002 
3003 }
3004 
3005 /*
3006  * Try and disable all interrupts. We just assign interrupts to other
3007  * processors based on policy. If any were bound by user request, we
3008  * let them continue and return failure. We do not bother to check
3009  * for cache affinity while rebinding.
3010  */
3011 
3012 static int
3013 apic_disable_intr(processorid_t cpun)
3014 {
3015 	int bind_cpu = 0, i, hardbound = 0, iflag;
3016 	apic_irq_t *irq_ptr;
3017 
3018 	if (cpun == 0)
3019 		return (PSM_FAILURE);
3020 
3021 	iflag = intr_clear();
3022 	lock_set(&apic_ioapic_lock);
3023 	apic_cpus[cpun].aci_status &= ~APIC_CPU_INTR_ENABLE;
3024 	lock_clear(&apic_ioapic_lock);
3025 	intr_restore(iflag);
3026 	apic_cpus[cpun].aci_curipl = 0;
3027 	i = apic_min_device_irq;
3028 	for (; i <= apic_max_device_irq; i++) {
3029 		/*
3030 		 * If there are bound interrupts on this cpu, then
3031 		 * rebind them to other processors.
3032 		 */
3033 		if ((irq_ptr = apic_irq_table[i]) != NULL) {
3034 			ASSERT((irq_ptr->airq_temp_cpu == IRQ_UNBOUND) ||
3035 			    (irq_ptr->airq_temp_cpu == IRQ_UNINIT) ||
3036 			    ((irq_ptr->airq_temp_cpu & ~IRQ_USER_BOUND) <
3037 			    apic_nproc));
3038 
3039 			if (irq_ptr->airq_temp_cpu == (cpun | IRQ_USER_BOUND)) {
3040 				hardbound = 1;
3041 				continue;
3042 			}
3043 
3044 			if (irq_ptr->airq_temp_cpu == cpun) {
3045 				do {
3046 					apic_next_bind_cpu += 2;
3047 					bind_cpu = apic_next_bind_cpu / 2;
3048 					if (bind_cpu >= apic_nproc) {
3049 						apic_next_bind_cpu = 1;
3050 						bind_cpu = 0;
3051 
3052 					}
3053 				} while (apic_rebind_all(irq_ptr, bind_cpu, 1));
3054 			}
3055 		}
3056 	}
3057 	if (hardbound) {
3058 		cmn_err(CE_WARN, "Could not disable interrupts on %d"
3059 		    "due to user bound interrupts", cpun);
3060 		return (PSM_FAILURE);
3061 	}
3062 	else
3063 		return (PSM_SUCCESS);
3064 }
3065 
3066 static void
3067 apic_enable_intr(processorid_t cpun)
3068 {
3069 	int	i, iflag;
3070 	apic_irq_t *irq_ptr;
3071 
3072 	iflag = intr_clear();
3073 	lock_set(&apic_ioapic_lock);
3074 	apic_cpus[cpun].aci_status |= APIC_CPU_INTR_ENABLE;
3075 	lock_clear(&apic_ioapic_lock);
3076 	intr_restore(iflag);
3077 
3078 	i = apic_min_device_irq;
3079 	for (i = apic_min_device_irq; i <= apic_max_device_irq; i++) {
3080 		if ((irq_ptr = apic_irq_table[i]) != NULL) {
3081 			if ((irq_ptr->airq_cpu & ~IRQ_USER_BOUND) == cpun) {
3082 				(void) apic_rebind_all(irq_ptr,
3083 				    irq_ptr->airq_cpu, 1);
3084 			}
3085 		}
3086 	}
3087 }
3088 
3089 /*
3090  * apic_introp_xlate() replaces apic_translate_irq() and is
3091  * called only from apic_intr_ops().  With the new ADII framework,
3092  * the priority can no longer be retrived through i_ddi_get_intrspec().
3093  * It has to be passed in from the caller.
3094  */
3095 int
3096 apic_introp_xlate(dev_info_t *dip, struct intrspec *ispec, int type)
3097 {
3098 	char dev_type[16];
3099 	int dev_len, pci_irq, newirq, bustype, devid, busid, i;
3100 	int irqno = ispec->intrspec_vec;
3101 	ddi_acc_handle_t cfg_handle;
3102 	uchar_t ipin;
3103 	struct apic_io_intr *intrp;
3104 	iflag_t intr_flag;
3105 	APIC_HEADER	*hp;
3106 	MADT_INTERRUPT_OVERRIDE	*isop;
3107 	apic_irq_t *airqp;
3108 
3109 	DDI_INTR_IMPLDBG((CE_CONT, "apic_introp_xlate: dip=0x%p name=%s "
3110 	    "type=%d irqno=0x%x\n", (void *)dip, ddi_get_name(dip), type,
3111 	    irqno));
3112 
3113 	if (DDI_INTR_IS_MSI_OR_MSIX(type)) {
3114 		if ((airqp = apic_find_irq(dip, ispec, type)) != NULL)
3115 			return (apic_vector_to_irq[airqp->airq_vector]);
3116 		return (apic_setup_irq_table(dip, irqno, NULL, ispec,
3117 		    NULL, type));
3118 	}
3119 
3120 	bustype = 0;
3121 
3122 	/* check if we have already translated this irq */
3123 	mutex_enter(&airq_mutex);
3124 	newirq = apic_min_device_irq;
3125 	for (; newirq <= apic_max_device_irq; newirq++) {
3126 		airqp = apic_irq_table[newirq];
3127 		while (airqp) {
3128 			if ((airqp->airq_dip == dip) &&
3129 			    (airqp->airq_origirq == irqno) &&
3130 			    (airqp->airq_mps_intr_index != FREE_INDEX)) {
3131 
3132 				mutex_exit(&airq_mutex);
3133 				return (VIRTIRQ(newirq, airqp->airq_share_id));
3134 			}
3135 			airqp = airqp->airq_next;
3136 		}
3137 	}
3138 	mutex_exit(&airq_mutex);
3139 
3140 	if (apic_defconf)
3141 		goto defconf;
3142 
3143 	if ((dip == NULL) || (!apic_irq_translate && !apic_enable_acpi))
3144 		goto nonpci;
3145 
3146 	dev_len = sizeof (dev_type);
3147 	if (ddi_getlongprop_buf(DDI_DEV_T_ANY, ddi_get_parent(dip),
3148 	    DDI_PROP_DONTPASS, "device_type", (caddr_t)dev_type,
3149 	    &dev_len) != DDI_PROP_SUCCESS) {
3150 		goto nonpci;
3151 	}
3152 
3153 	if ((strcmp(dev_type, "pci") == 0) ||
3154 	    (strcmp(dev_type, "pciex") == 0)) {
3155 		/* pci device */
3156 		if (acpica_get_bdf(dip, &busid, &devid, NULL) != 0)
3157 			goto nonpci;
3158 		if (busid == 0 && apic_pci_bus_total == 1)
3159 			busid = (int)apic_single_pci_busid;
3160 
3161 		if (pci_config_setup(dip, &cfg_handle) != DDI_SUCCESS)
3162 			goto nonpci;
3163 		ipin = pci_config_get8(cfg_handle, PCI_CONF_IPIN) - PCI_INTA;
3164 		pci_config_teardown(&cfg_handle);
3165 		if (apic_enable_acpi && !apic_use_acpi_madt_only) {
3166 			if (apic_acpi_translate_pci_irq(dip, busid, devid,
3167 			    ipin, &pci_irq, &intr_flag) != ACPI_PSM_SUCCESS)
3168 				goto nonpci;
3169 
3170 			intr_flag.bustype = BUS_PCI;
3171 			if ((newirq = apic_setup_irq_table(dip, pci_irq, NULL,
3172 			    ispec, &intr_flag, type)) == -1)
3173 				goto nonpci;
3174 			return (newirq);
3175 		} else {
3176 			pci_irq = ((devid & 0x1f) << 2) | (ipin & 0x3);
3177 			if ((intrp = apic_find_io_intr_w_busid(pci_irq, busid))
3178 			    == NULL) {
3179 				if ((pci_irq = apic_handle_pci_pci_bridge(dip,
3180 				    devid, ipin, &intrp)) == -1)
3181 					goto nonpci;
3182 			}
3183 			if ((newirq = apic_setup_irq_table(dip, pci_irq, intrp,
3184 			    ispec, NULL, type)) == -1)
3185 				goto nonpci;
3186 			return (newirq);
3187 		}
3188 	} else if (strcmp(dev_type, "isa") == 0)
3189 		bustype = BUS_ISA;
3190 	else if (strcmp(dev_type, "eisa") == 0)
3191 		bustype = BUS_EISA;
3192 
3193 nonpci:
3194 	if (apic_enable_acpi && !apic_use_acpi_madt_only) {
3195 		/* search iso entries first */
3196 		if (acpi_iso_cnt != 0) {
3197 			hp = (APIC_HEADER *)acpi_isop;
3198 			i = 0;
3199 			while (i < acpi_iso_cnt) {
3200 				if (hp->Type == APIC_XRUPT_OVERRIDE) {
3201 					isop = (MADT_INTERRUPT_OVERRIDE *)hp;
3202 					if (isop->Bus == 0 &&
3203 					    isop->Source == irqno) {
3204 						newirq = isop->Interrupt;
3205 						intr_flag.intr_po =
3206 						    isop->Polarity;
3207 						intr_flag.intr_el =
3208 						    isop->TriggerMode;
3209 						intr_flag.bustype = BUS_ISA;
3210 
3211 						return (apic_setup_irq_table(
3212 						    dip, newirq, NULL, ispec,
3213 						    &intr_flag, type));
3214 
3215 					}
3216 					i++;
3217 				}
3218 				hp = (APIC_HEADER *)(((char *)hp) +
3219 				    hp->Length);
3220 			}
3221 		}
3222 		intr_flag.intr_po = INTR_PO_ACTIVE_HIGH;
3223 		intr_flag.intr_el = INTR_EL_EDGE;
3224 		intr_flag.bustype = BUS_ISA;
3225 		return (apic_setup_irq_table(dip, irqno, NULL, ispec,
3226 		    &intr_flag, type));
3227 	} else {
3228 		if (bustype == 0)
3229 			bustype = eisa_level_intr_mask ? BUS_EISA : BUS_ISA;
3230 		for (i = 0; i < 2; i++) {
3231 			if (((busid = apic_find_bus_id(bustype)) != -1) &&
3232 			    ((intrp = apic_find_io_intr_w_busid(irqno, busid))
3233 			    != NULL)) {
3234 				if ((newirq = apic_setup_irq_table(dip, irqno,
3235 				    intrp, ispec, NULL, type)) != -1) {
3236 					return (newirq);
3237 				}
3238 				goto defconf;
3239 			}
3240 			bustype = (bustype == BUS_EISA) ? BUS_ISA : BUS_EISA;
3241 		}
3242 	}
3243 
3244 /* MPS default configuration */
3245 defconf:
3246 	newirq = apic_setup_irq_table(dip, irqno, NULL, ispec, NULL, type);
3247 	if (newirq == -1)
3248 		return (newirq);
3249 	ASSERT(IRQINDEX(newirq) == irqno);
3250 	ASSERT(apic_irq_table[irqno]);
3251 	return (newirq);
3252 }
3253 
3254 
3255 
3256 
3257 
3258 
3259 /*
3260  * On machines with PCI-PCI bridges, a device behind a PCI-PCI bridge
3261  * needs special handling.  We may need to chase up the device tree,
3262  * using the PCI-PCI Bridge specification's "rotating IPIN assumptions",
3263  * to find the IPIN at the root bus that relates to the IPIN on the
3264  * subsidiary bus (for ACPI or MP).  We may, however, have an entry
3265  * in the MP table or the ACPI namespace for this device itself.
3266  * We handle both cases in the search below.
3267  */
3268 /* this is the non-acpi version */
3269 static int
3270 apic_handle_pci_pci_bridge(dev_info_t *idip, int child_devno, int child_ipin,
3271 			struct apic_io_intr **intrp)
3272 {
3273 	dev_info_t *dipp, *dip;
3274 	int pci_irq;
3275 	ddi_acc_handle_t cfg_handle;
3276 	int bridge_devno, bridge_bus;
3277 	int ipin;
3278 
3279 	dip = idip;
3280 
3281 	/*CONSTCOND*/
3282 	while (1) {
3283 		if ((dipp = ddi_get_parent(dip)) == (dev_info_t *)NULL)
3284 			return (-1);
3285 		if ((pci_config_setup(dipp, &cfg_handle) == DDI_SUCCESS) &&
3286 		    (pci_config_get8(cfg_handle, PCI_CONF_BASCLASS) ==
3287 		    PCI_CLASS_BRIDGE) && (pci_config_get8(cfg_handle,
3288 		    PCI_CONF_SUBCLASS) == PCI_BRIDGE_PCI)) {
3289 			pci_config_teardown(&cfg_handle);
3290 			if (acpica_get_bdf(dipp, &bridge_bus, &bridge_devno,
3291 			    NULL) != 0)
3292 				return (-1);
3293 			/*
3294 			 * This is the rotating scheme that Compaq is using
3295 			 * and documented in the pci to pci spec.  Also, if
3296 			 * the pci to pci bridge is behind another pci to
3297 			 * pci bridge, then it need to keep transversing
3298 			 * up until an interrupt entry is found or reach
3299 			 * the top of the tree
3300 			 */
3301 			ipin = (child_devno + child_ipin) % PCI_INTD;
3302 				if (bridge_bus == 0 && apic_pci_bus_total == 1)
3303 					bridge_bus = (int)apic_single_pci_busid;
3304 				pci_irq = ((bridge_devno & 0x1f) << 2) |
3305 				    (ipin & 0x3);
3306 				if ((*intrp = apic_find_io_intr_w_busid(pci_irq,
3307 				    bridge_bus)) != NULL) {
3308 					return (pci_irq);
3309 				}
3310 			dip = dipp;
3311 			child_devno = bridge_devno;
3312 			child_ipin = ipin;
3313 		} else
3314 			return (-1);
3315 	}
3316 	/*LINTED: function will not fall off the bottom */
3317 }
3318 
3319 
3320 
3321 
3322 static uchar_t
3323 acpi_find_ioapic(int irq)
3324 {
3325 	int i;
3326 
3327 	for (i = 0; i < apic_io_max; i++) {
3328 		if (irq >= apic_io_vectbase[i] && irq <= apic_io_vectend[i])
3329 			return (i);
3330 	}
3331 	return (0xFF);	/* shouldn't happen */
3332 }
3333 
3334 /*
3335  * See if two irqs are compatible for sharing a vector.
3336  * Currently we only support sharing of PCI devices.
3337  */
3338 static int
3339 acpi_intr_compatible(iflag_t iflag1, iflag_t iflag2)
3340 {
3341 	uint_t	level1, po1;
3342 	uint_t	level2, po2;
3343 
3344 	/* Assume active high by default */
3345 	po1 = 0;
3346 	po2 = 0;
3347 
3348 	if (iflag1.bustype != iflag2.bustype || iflag1.bustype != BUS_PCI)
3349 		return (0);
3350 
3351 	if (iflag1.intr_el == INTR_EL_CONFORM)
3352 		level1 = AV_LEVEL;
3353 	else
3354 		level1 = (iflag1.intr_el == INTR_EL_LEVEL) ? AV_LEVEL : 0;
3355 
3356 	if (level1 && ((iflag1.intr_po == INTR_PO_ACTIVE_LOW) ||
3357 	    (iflag1.intr_po == INTR_PO_CONFORM)))
3358 		po1 = AV_ACTIVE_LOW;
3359 
3360 	if (iflag2.intr_el == INTR_EL_CONFORM)
3361 		level2 = AV_LEVEL;
3362 	else
3363 		level2 = (iflag2.intr_el == INTR_EL_LEVEL) ? AV_LEVEL : 0;
3364 
3365 	if (level2 && ((iflag2.intr_po == INTR_PO_ACTIVE_LOW) ||
3366 	    (iflag2.intr_po == INTR_PO_CONFORM)))
3367 		po2 = AV_ACTIVE_LOW;
3368 
3369 	if ((level1 == level2) && (po1 == po2))
3370 		return (1);
3371 
3372 	return (0);
3373 }
3374 
3375 /*
3376  * Attempt to share vector with someone else
3377  */
3378 static int
3379 apic_share_vector(int irqno, iflag_t *intr_flagp, short intr_index, int ipl,
3380 	uchar_t ioapicindex, uchar_t ipin, apic_irq_t **irqptrp)
3381 {
3382 #ifdef DEBUG
3383 	apic_irq_t *tmpirqp = NULL;
3384 #endif /* DEBUG */
3385 	apic_irq_t *irqptr, dummyirq;
3386 	int	newirq, chosen_irq = -1, share = 127;
3387 	int	lowest, highest, i;
3388 	uchar_t	share_id;
3389 
3390 	DDI_INTR_IMPLDBG((CE_CONT, "apic_share_vector: irqno=0x%x "
3391 	    "intr_index=0x%x ipl=0x%x\n", irqno, intr_index, ipl));
3392 
3393 	highest = apic_ipltopri[ipl] + APIC_VECTOR_MASK;
3394 	lowest = apic_ipltopri[ipl-1] + APIC_VECTOR_PER_IPL;
3395 
3396 	if (highest < lowest) /* Both ipl and ipl-1 map to same pri */
3397 		lowest -= APIC_VECTOR_PER_IPL;
3398 	dummyirq.airq_mps_intr_index = intr_index;
3399 	dummyirq.airq_ioapicindex = ioapicindex;
3400 	dummyirq.airq_intin_no = ipin;
3401 	if (intr_flagp)
3402 		dummyirq.airq_iflag = *intr_flagp;
3403 	apic_record_rdt_entry(&dummyirq, irqno);
3404 	for (i = lowest; i <= highest; i++) {
3405 		newirq = apic_vector_to_irq[i];
3406 		if (newirq == APIC_RESV_IRQ)
3407 			continue;
3408 		irqptr = apic_irq_table[newirq];
3409 
3410 		if ((dummyirq.airq_rdt_entry & 0xFF00) !=
3411 		    (irqptr->airq_rdt_entry & 0xFF00))
3412 			/* not compatible */
3413 			continue;
3414 
3415 		if (irqptr->airq_share < share) {
3416 			share = irqptr->airq_share;
3417 			chosen_irq = newirq;
3418 		}
3419 	}
3420 	if (chosen_irq != -1) {
3421 		/*
3422 		 * Assign a share id which is free or which is larger
3423 		 * than the largest one.
3424 		 */
3425 		share_id = 1;
3426 		mutex_enter(&airq_mutex);
3427 		irqptr = apic_irq_table[chosen_irq];
3428 		while (irqptr) {
3429 			if (irqptr->airq_mps_intr_index == FREE_INDEX) {
3430 				share_id = irqptr->airq_share_id;
3431 				break;
3432 			}
3433 			if (share_id <= irqptr->airq_share_id)
3434 				share_id = irqptr->airq_share_id + 1;
3435 #ifdef DEBUG
3436 			tmpirqp = irqptr;
3437 #endif /* DEBUG */
3438 			irqptr = irqptr->airq_next;
3439 		}
3440 		if (!irqptr) {
3441 			irqptr = kmem_zalloc(sizeof (apic_irq_t), KM_SLEEP);
3442 			irqptr->airq_temp_cpu = IRQ_UNINIT;
3443 			irqptr->airq_next =
3444 			    apic_irq_table[chosen_irq]->airq_next;
3445 			apic_irq_table[chosen_irq]->airq_next = irqptr;
3446 #ifdef	DEBUG
3447 			tmpirqp = apic_irq_table[chosen_irq];
3448 #endif /* DEBUG */
3449 		}
3450 		irqptr->airq_mps_intr_index = intr_index;
3451 		irqptr->airq_ioapicindex = ioapicindex;
3452 		irqptr->airq_intin_no = ipin;
3453 		if (intr_flagp)
3454 			irqptr->airq_iflag = *intr_flagp;
3455 		irqptr->airq_vector = apic_irq_table[chosen_irq]->airq_vector;
3456 		irqptr->airq_share_id = share_id;
3457 		apic_record_rdt_entry(irqptr, irqno);
3458 		*irqptrp = irqptr;
3459 #ifdef	DEBUG
3460 		/* shuffle the pointers to test apic_delspl path */
3461 		if (tmpirqp) {
3462 			tmpirqp->airq_next = irqptr->airq_next;
3463 			irqptr->airq_next = apic_irq_table[chosen_irq];
3464 			apic_irq_table[chosen_irq] = irqptr;
3465 		}
3466 #endif /* DEBUG */
3467 		mutex_exit(&airq_mutex);
3468 		return (VIRTIRQ(chosen_irq, share_id));
3469 	}
3470 	return (-1);
3471 }
3472 
3473 /*
3474  *
3475  */
3476 static int
3477 apic_setup_irq_table(dev_info_t *dip, int irqno, struct apic_io_intr *intrp,
3478     struct intrspec *ispec, iflag_t *intr_flagp, int type)
3479 {
3480 	int origirq = ispec->intrspec_vec;
3481 	uchar_t ipl = ispec->intrspec_pri;
3482 	int	newirq, intr_index;
3483 	uchar_t	ipin, ioapic, ioapicindex, vector;
3484 	apic_irq_t *irqptr;
3485 	major_t	major;
3486 	dev_info_t	*sdip;
3487 
3488 	DDI_INTR_IMPLDBG((CE_CONT, "apic_setup_irq_table: dip=0x%p type=%d "
3489 	    "irqno=0x%x origirq=0x%x\n", (void *)dip, type, irqno, origirq));
3490 
3491 	ASSERT(ispec != NULL);
3492 
3493 	major =  (dip != NULL) ? ddi_name_to_major(ddi_get_name(dip)) : 0;
3494 
3495 	if (DDI_INTR_IS_MSI_OR_MSIX(type)) {
3496 		/* MSI/X doesn't need to setup ioapic stuffs */
3497 		ioapicindex = 0xff;
3498 		ioapic = 0xff;
3499 		ipin = (uchar_t)0xff;
3500 		intr_index = (type == DDI_INTR_TYPE_MSI) ? MSI_INDEX :
3501 		    MSIX_INDEX;
3502 		mutex_enter(&airq_mutex);
3503 		if ((irqno = apic_allocate_irq(apic_first_avail_irq)) == -1) {
3504 			mutex_exit(&airq_mutex);
3505 			/* need an irq for MSI/X to index into autovect[] */
3506 			cmn_err(CE_WARN, "No interrupt irq: %s instance %d",
3507 			    ddi_get_name(dip), ddi_get_instance(dip));
3508 			return (-1);
3509 		}
3510 		mutex_exit(&airq_mutex);
3511 
3512 	} else if (intrp != NULL) {
3513 		intr_index = (int)(intrp - apic_io_intrp);
3514 		ioapic = intrp->intr_destid;
3515 		ipin = intrp->intr_destintin;
3516 		/* Find ioapicindex. If destid was ALL, we will exit with 0. */
3517 		for (ioapicindex = apic_io_max - 1; ioapicindex; ioapicindex--)
3518 			if (apic_io_id[ioapicindex] == ioapic)
3519 				break;
3520 		ASSERT((ioapic == apic_io_id[ioapicindex]) ||
3521 		    (ioapic == INTR_ALL_APIC));
3522 
3523 		/* check whether this intin# has been used by another irqno */
3524 		if ((newirq = apic_find_intin(ioapicindex, ipin)) != -1) {
3525 			return (newirq);
3526 		}
3527 
3528 	} else if (intr_flagp != NULL) {
3529 		/* ACPI case */
3530 		intr_index = ACPI_INDEX;
3531 		ioapicindex = acpi_find_ioapic(irqno);
3532 		ASSERT(ioapicindex != 0xFF);
3533 		ioapic = apic_io_id[ioapicindex];
3534 		ipin = irqno - apic_io_vectbase[ioapicindex];
3535 		if (apic_irq_table[irqno] &&
3536 		    apic_irq_table[irqno]->airq_mps_intr_index == ACPI_INDEX) {
3537 			ASSERT(apic_irq_table[irqno]->airq_intin_no == ipin &&
3538 			    apic_irq_table[irqno]->airq_ioapicindex ==
3539 			    ioapicindex);
3540 			return (irqno);
3541 		}
3542 
3543 	} else {
3544 		/* default configuration */
3545 		ioapicindex = 0;
3546 		ioapic = apic_io_id[ioapicindex];
3547 		ipin = (uchar_t)irqno;
3548 		intr_index = DEFAULT_INDEX;
3549 	}
3550 
3551 	if (ispec == NULL) {
3552 		APIC_VERBOSE_IOAPIC((CE_WARN, "No intrspec for irqno = %x\n",
3553 		    irqno));
3554 	} else if ((vector = apic_allocate_vector(ipl, irqno, 0)) == 0) {
3555 		if ((newirq = apic_share_vector(irqno, intr_flagp, intr_index,
3556 		    ipl, ioapicindex, ipin, &irqptr)) != -1) {
3557 			irqptr->airq_ipl = ipl;
3558 			irqptr->airq_origirq = (uchar_t)origirq;
3559 			irqptr->airq_dip = dip;
3560 			irqptr->airq_major = major;
3561 			sdip = apic_irq_table[IRQINDEX(newirq)]->airq_dip;
3562 			/* This is OK to do really */
3563 			if (sdip == NULL) {
3564 				cmn_err(CE_WARN, "Sharing vectors: %s"
3565 				    " instance %d and SCI",
3566 				    ddi_get_name(dip), ddi_get_instance(dip));
3567 			} else {
3568 				cmn_err(CE_WARN, "Sharing vectors: %s"
3569 				    " instance %d and %s instance %d",
3570 				    ddi_get_name(sdip), ddi_get_instance(sdip),
3571 				    ddi_get_name(dip), ddi_get_instance(dip));
3572 			}
3573 			return (newirq);
3574 		}
3575 		/* try high priority allocation now  that share has failed */
3576 		if ((vector = apic_allocate_vector(ipl, irqno, 1)) == 0) {
3577 			cmn_err(CE_WARN, "No interrupt vector: %s instance %d",
3578 			    ddi_get_name(dip), ddi_get_instance(dip));
3579 			return (-1);
3580 		}
3581 	}
3582 
3583 	mutex_enter(&airq_mutex);
3584 	if (apic_irq_table[irqno] == NULL) {
3585 		irqptr = kmem_zalloc(sizeof (apic_irq_t), KM_SLEEP);
3586 		irqptr->airq_temp_cpu = IRQ_UNINIT;
3587 		apic_irq_table[irqno] = irqptr;
3588 	} else {
3589 		irqptr = apic_irq_table[irqno];
3590 		if (irqptr->airq_mps_intr_index != FREE_INDEX) {
3591 			/*
3592 			 * The slot is used by another irqno, so allocate
3593 			 * a free irqno for this interrupt
3594 			 */
3595 			newirq = apic_allocate_irq(apic_first_avail_irq);
3596 			if (newirq == -1) {
3597 				mutex_exit(&airq_mutex);
3598 				return (-1);
3599 			}
3600 			irqno = newirq;
3601 			irqptr = apic_irq_table[irqno];
3602 			if (irqptr == NULL) {
3603 				irqptr = kmem_zalloc(sizeof (apic_irq_t),
3604 				    KM_SLEEP);
3605 				irqptr->airq_temp_cpu = IRQ_UNINIT;
3606 				apic_irq_table[irqno] = irqptr;
3607 			}
3608 			apic_modify_vector(vector, newirq);
3609 		}
3610 	}
3611 	apic_max_device_irq = max(irqno, apic_max_device_irq);
3612 	apic_min_device_irq = min(irqno, apic_min_device_irq);
3613 	mutex_exit(&airq_mutex);
3614 	irqptr->airq_ioapicindex = ioapicindex;
3615 	irqptr->airq_intin_no = ipin;
3616 	irqptr->airq_ipl = ipl;
3617 	irqptr->airq_vector = vector;
3618 	irqptr->airq_origirq = (uchar_t)origirq;
3619 	irqptr->airq_share_id = 0;
3620 	irqptr->airq_mps_intr_index = (short)intr_index;
3621 	irqptr->airq_dip = dip;
3622 	irqptr->airq_major = major;
3623 	irqptr->airq_cpu = apic_bind_intr(dip, irqno, ioapic, ipin);
3624 	if (intr_flagp)
3625 		irqptr->airq_iflag = *intr_flagp;
3626 
3627 	if (!DDI_INTR_IS_MSI_OR_MSIX(type)) {
3628 		/* setup I/O APIC entry for non-MSI/X interrupts */
3629 		apic_record_rdt_entry(irqptr, irqno);
3630 	}
3631 	return (irqno);
3632 }
3633 
3634 /*
3635  * return the cpu to which this intr should be bound.
3636  * Check properties or any other mechanism to see if user wants it
3637  * bound to a specific CPU. If so, return the cpu id with high bit set.
3638  * If not, use the policy to choose a cpu and return the id.
3639  */
3640 uchar_t
3641 apic_bind_intr(dev_info_t *dip, int irq, uchar_t ioapicid, uchar_t intin)
3642 {
3643 	int	instance, instno, prop_len, bind_cpu, count;
3644 	uint_t	i, rc;
3645 	uchar_t	cpu;
3646 	major_t	major;
3647 	char	*name, *drv_name, *prop_val, *cptr;
3648 	char	prop_name[32];
3649 
3650 
3651 	if (apic_intr_policy == INTR_LOWEST_PRIORITY)
3652 		return (IRQ_UNBOUND);
3653 
3654 	drv_name = NULL;
3655 	rc = DDI_PROP_NOT_FOUND;
3656 	major = (major_t)-1;
3657 	if (dip != NULL) {
3658 		name = ddi_get_name(dip);
3659 		major = ddi_name_to_major(name);
3660 		drv_name = ddi_major_to_name(major);
3661 		instance = ddi_get_instance(dip);
3662 		if (apic_intr_policy == INTR_ROUND_ROBIN_WITH_AFFINITY) {
3663 			i = apic_min_device_irq;
3664 			for (; i <= apic_max_device_irq; i++) {
3665 
3666 				if ((i == irq) || (apic_irq_table[i] == NULL) ||
3667 				    (apic_irq_table[i]->airq_mps_intr_index
3668 				    == FREE_INDEX))
3669 					continue;
3670 
3671 				if ((apic_irq_table[i]->airq_major == major) &&
3672 				    (!(apic_irq_table[i]->airq_cpu &
3673 				    IRQ_USER_BOUND))) {
3674 
3675 					cpu = apic_irq_table[i]->airq_cpu;
3676 
3677 					cmn_err(CE_CONT,
3678 					    "!pcplusmp: %s (%s) instance #%d "
3679 					    "vector 0x%x ioapic 0x%x "
3680 					    "intin 0x%x is bound to cpu %d\n",
3681 					    name, drv_name, instance, irq,
3682 					    ioapicid, intin, cpu);
3683 					return (cpu);
3684 				}
3685 			}
3686 		}
3687 		/*
3688 		 * search for "drvname"_intpt_bind_cpus property first, the
3689 		 * syntax of the property should be "a[,b,c,...]" where
3690 		 * instance 0 binds to cpu a, instance 1 binds to cpu b,
3691 		 * instance 3 binds to cpu c...
3692 		 * ddi_getlongprop() will search /option first, then /
3693 		 * if "drvname"_intpt_bind_cpus doesn't exist, then find
3694 		 * intpt_bind_cpus property.  The syntax is the same, and
3695 		 * it applies to all the devices if its "drvname" specific
3696 		 * property doesn't exist
3697 		 */
3698 		(void) strcpy(prop_name, drv_name);
3699 		(void) strcat(prop_name, "_intpt_bind_cpus");
3700 		rc = ddi_getlongprop(DDI_DEV_T_ANY, dip, 0, prop_name,
3701 		    (caddr_t)&prop_val, &prop_len);
3702 		if (rc != DDI_PROP_SUCCESS) {
3703 			rc = ddi_getlongprop(DDI_DEV_T_ANY, dip, 0,
3704 			    "intpt_bind_cpus", (caddr_t)&prop_val, &prop_len);
3705 		}
3706 	}
3707 	if (rc == DDI_PROP_SUCCESS) {
3708 		for (i = count = 0; i < (prop_len - 1); i++)
3709 			if (prop_val[i] == ',')
3710 				count++;
3711 		if (prop_val[i-1] != ',')
3712 			count++;
3713 		/*
3714 		 * if somehow the binding instances defined in the
3715 		 * property are not enough for this instno., then
3716 		 * reuse the pattern for the next instance until
3717 		 * it reaches the requested instno
3718 		 */
3719 		instno = instance % count;
3720 		i = 0;
3721 		cptr = prop_val;
3722 		while (i < instno)
3723 			if (*cptr++ == ',')
3724 				i++;
3725 		bind_cpu = stoi(&cptr);
3726 		kmem_free(prop_val, prop_len);
3727 		/* if specific cpu is bogus, then default to cpu 0 */
3728 		if (bind_cpu >= apic_nproc) {
3729 			cmn_err(CE_WARN, "pcplusmp: %s=%s: CPU %d not present",
3730 			    prop_name, prop_val, bind_cpu);
3731 			bind_cpu = 0;
3732 		} else {
3733 			/* indicate that we are bound at user request */
3734 			bind_cpu |= IRQ_USER_BOUND;
3735 		}
3736 		/*
3737 		 * no need to check apic_cpus[].aci_status, if specific cpu is
3738 		 * not up, then post_cpu_start will handle it.
3739 		 */
3740 	} else {
3741 		/*
3742 		 * We change bind_cpu only for every two calls
3743 		 * as most drivers still do 2 add_intrs for every
3744 		 * interrupt
3745 		 */
3746 		bind_cpu = (apic_next_bind_cpu++) / 2;
3747 		if (bind_cpu >= apic_nproc) {
3748 			apic_next_bind_cpu = 1;
3749 			bind_cpu = 0;
3750 		}
3751 	}
3752 	if (drv_name != NULL)
3753 		cmn_err(CE_CONT, "!pcplusmp: %s (%s) instance %d "
3754 		    "vector 0x%x ioapic 0x%x intin 0x%x is bound to cpu %d\n",
3755 		    name, drv_name, instance,
3756 		    irq, ioapicid, intin, bind_cpu & ~IRQ_USER_BOUND);
3757 	else
3758 		cmn_err(CE_CONT, "!pcplusmp: "
3759 		    "vector 0x%x ioapic 0x%x intin 0x%x is bound to cpu %d\n",
3760 		    irq, ioapicid, intin, bind_cpu & ~IRQ_USER_BOUND);
3761 
3762 	return ((uchar_t)bind_cpu);
3763 }
3764 
3765 static struct apic_io_intr *
3766 apic_find_io_intr_w_busid(int irqno, int busid)
3767 {
3768 	struct	apic_io_intr	*intrp;
3769 
3770 	/*
3771 	 * It can have more than 1 entry with same source bus IRQ,
3772 	 * but unique with the source bus id
3773 	 */
3774 	intrp = apic_io_intrp;
3775 	if (intrp != NULL) {
3776 		while (intrp->intr_entry == APIC_IO_INTR_ENTRY) {
3777 			if (intrp->intr_irq == irqno &&
3778 			    intrp->intr_busid == busid &&
3779 			    intrp->intr_type == IO_INTR_INT)
3780 				return (intrp);
3781 			intrp++;
3782 		}
3783 	}
3784 	APIC_VERBOSE_IOAPIC((CE_NOTE, "Did not find io intr for irqno:"
3785 	    "busid %x:%x\n", irqno, busid));
3786 	return ((struct apic_io_intr *)NULL);
3787 }
3788 
3789 
3790 struct mps_bus_info {
3791 	char	*bus_name;
3792 	int	bus_id;
3793 } bus_info_array[] = {
3794 	"ISA ", BUS_ISA,
3795 	"PCI ", BUS_PCI,
3796 	"EISA ", BUS_EISA,
3797 	"XPRESS", BUS_XPRESS,
3798 	"PCMCIA", BUS_PCMCIA,
3799 	"VL ", BUS_VL,
3800 	"CBUS ", BUS_CBUS,
3801 	"CBUSII", BUS_CBUSII,
3802 	"FUTURE", BUS_FUTURE,
3803 	"INTERN", BUS_INTERN,
3804 	"MBI ", BUS_MBI,
3805 	"MBII ", BUS_MBII,
3806 	"MPI ", BUS_MPI,
3807 	"MPSA ", BUS_MPSA,
3808 	"NUBUS ", BUS_NUBUS,
3809 	"TC ", BUS_TC,
3810 	"VME ", BUS_VME
3811 };
3812 
3813 static int
3814 apic_find_bus_type(char *bus)
3815 {
3816 	int	i = 0;
3817 
3818 	for (; i < sizeof (bus_info_array)/sizeof (struct mps_bus_info); i++)
3819 		if (strncmp(bus, bus_info_array[i].bus_name,
3820 		    strlen(bus_info_array[i].bus_name)) == 0)
3821 			return (bus_info_array[i].bus_id);
3822 	APIC_VERBOSE_IOAPIC((CE_WARN, "Did not find bus type for bus %s", bus));
3823 	return (0);
3824 }
3825 
3826 static int
3827 apic_find_bus(int busid)
3828 {
3829 	struct	apic_bus	*busp;
3830 
3831 	busp = apic_busp;
3832 	while (busp->bus_entry == APIC_BUS_ENTRY) {
3833 		if (busp->bus_id == busid)
3834 			return (apic_find_bus_type((char *)&busp->bus_str1));
3835 		busp++;
3836 	}
3837 	APIC_VERBOSE_IOAPIC((CE_WARN, "Did not find bus for bus id %x", busid));
3838 	return (0);
3839 }
3840 
3841 static int
3842 apic_find_bus_id(int bustype)
3843 {
3844 	struct	apic_bus	*busp;
3845 
3846 	busp = apic_busp;
3847 	while (busp->bus_entry == APIC_BUS_ENTRY) {
3848 		if (apic_find_bus_type((char *)&busp->bus_str1) == bustype)
3849 			return (busp->bus_id);
3850 		busp++;
3851 	}
3852 	APIC_VERBOSE_IOAPIC((CE_WARN, "Did not find bus id for bustype %x",
3853 	    bustype));
3854 	return (-1);
3855 }
3856 
3857 /*
3858  * Check if a particular irq need to be reserved for any io_intr
3859  */
3860 static struct apic_io_intr *
3861 apic_find_io_intr(int irqno)
3862 {
3863 	struct	apic_io_intr	*intrp;
3864 
3865 	intrp = apic_io_intrp;
3866 	if (intrp != NULL) {
3867 		while (intrp->intr_entry == APIC_IO_INTR_ENTRY) {
3868 			if (intrp->intr_irq == irqno &&
3869 			    intrp->intr_type == IO_INTR_INT)
3870 				return (intrp);
3871 			intrp++;
3872 		}
3873 	}
3874 	return ((struct apic_io_intr *)NULL);
3875 }
3876 
3877 /*
3878  * Check if the given ioapicindex intin combination has already been assigned
3879  * an irq. If so return irqno. Else -1
3880  */
3881 static int
3882 apic_find_intin(uchar_t ioapic, uchar_t intin)
3883 {
3884 	apic_irq_t *irqptr;
3885 	int	i;
3886 
3887 	/* find ioapic and intin in the apic_irq_table[] and return the index */
3888 	for (i = apic_min_device_irq; i <= apic_max_device_irq; i++) {
3889 		irqptr = apic_irq_table[i];
3890 		while (irqptr) {
3891 			if ((irqptr->airq_mps_intr_index >= 0) &&
3892 			    (irqptr->airq_intin_no == intin) &&
3893 			    (irqptr->airq_ioapicindex == ioapic)) {
3894 				APIC_VERBOSE_IOAPIC((CE_NOTE, "!Found irq "
3895 				    "entry for ioapic:intin %x:%x "
3896 				    "shared interrupts ?", ioapic, intin));
3897 				return (i);
3898 			}
3899 			irqptr = irqptr->airq_next;
3900 		}
3901 	}
3902 	return (-1);
3903 }
3904 
3905 int
3906 apic_allocate_irq(int irq)
3907 {
3908 	int	freeirq, i;
3909 
3910 	if ((freeirq = apic_find_free_irq(irq, (APIC_RESV_IRQ - 1))) == -1)
3911 		if ((freeirq = apic_find_free_irq(APIC_FIRST_FREE_IRQ,
3912 		    (irq - 1))) == -1) {
3913 			/*
3914 			 * if BIOS really defines every single irq in the mps
3915 			 * table, then don't worry about conflicting with
3916 			 * them, just use any free slot in apic_irq_table
3917 			 */
3918 			for (i = APIC_FIRST_FREE_IRQ; i < APIC_RESV_IRQ; i++) {
3919 				if ((apic_irq_table[i] == NULL) ||
3920 				    apic_irq_table[i]->airq_mps_intr_index ==
3921 				    FREE_INDEX) {
3922 				freeirq = i;
3923 				break;
3924 			}
3925 		}
3926 		if (freeirq == -1) {
3927 			/* This shouldn't happen, but just in case */
3928 			cmn_err(CE_WARN, "pcplusmp: NO available IRQ");
3929 			return (-1);
3930 		}
3931 	}
3932 	if (apic_irq_table[freeirq] == NULL) {
3933 		apic_irq_table[freeirq] =
3934 		    kmem_zalloc(sizeof (apic_irq_t), KM_NOSLEEP);
3935 		if (apic_irq_table[freeirq] == NULL) {
3936 			cmn_err(CE_WARN, "pcplusmp: NO memory to allocate IRQ");
3937 			return (-1);
3938 		}
3939 		apic_irq_table[freeirq]->airq_mps_intr_index = FREE_INDEX;
3940 	}
3941 	return (freeirq);
3942 }
3943 
3944 static int
3945 apic_find_free_irq(int start, int end)
3946 {
3947 	int	i;
3948 
3949 	for (i = start; i <= end; i++)
3950 		/* Check if any I/O entry needs this IRQ */
3951 		if (apic_find_io_intr(i) == NULL) {
3952 			/* Then see if it is free */
3953 			if ((apic_irq_table[i] == NULL) ||
3954 			    (apic_irq_table[i]->airq_mps_intr_index ==
3955 			    FREE_INDEX)) {
3956 				return (i);
3957 			}
3958 		}
3959 	return (-1);
3960 }
3961 
3962 /*
3963  * Allocate a free vector for irq at ipl. Takes care of merging of multiple
3964  * IPLs into a single APIC level as well as stretching some IPLs onto multiple
3965  * levels. APIC_HI_PRI_VECTS interrupts are reserved for high priority
3966  * requests and allocated only when pri is set.
3967  */
3968 static uchar_t
3969 apic_allocate_vector(int ipl, int irq, int pri)
3970 {
3971 	int	lowest, highest, i;
3972 
3973 	highest = apic_ipltopri[ipl] + APIC_VECTOR_MASK;
3974 	lowest = apic_ipltopri[ipl - 1] + APIC_VECTOR_PER_IPL;
3975 
3976 	if (highest < lowest) /* Both ipl and ipl - 1 map to same pri */
3977 		lowest -= APIC_VECTOR_PER_IPL;
3978 
3979 #ifdef	DEBUG
3980 	if (apic_restrict_vector)	/* for testing shared interrupt logic */
3981 		highest = lowest + apic_restrict_vector + APIC_HI_PRI_VECTS;
3982 #endif /* DEBUG */
3983 	if (pri == 0)
3984 		highest -= APIC_HI_PRI_VECTS;
3985 
3986 	for (i = lowest; i < highest; i++) {
3987 		if ((i == T_FASTTRAP) || (i == APIC_SPUR_INTR) ||
3988 			(i == T_SYSCALLINT) || (i == T_DTRACE_PROBE) ||
3989 			(i == T_DTRACE_RET))
3990 			continue;
3991 		if (apic_vector_to_irq[i] == APIC_RESV_IRQ) {
3992 			apic_vector_to_irq[i] = (uchar_t)irq;
3993 			return (i);
3994 		}
3995 	}
3996 
3997 	return (0);
3998 }
3999 
4000 static void
4001 apic_modify_vector(uchar_t vector, int irq)
4002 {
4003 	apic_vector_to_irq[vector] = (uchar_t)irq;
4004 }
4005 
4006 /*
4007  * Mark vector as being in the process of being deleted. Interrupts
4008  * may still come in on some CPU. The moment an interrupt comes with
4009  * the new vector, we know we can free the old one. Called only from
4010  * addspl and delspl with interrupts disabled. Because an interrupt
4011  * can be shared, but no interrupt from either device may come in,
4012  * we also use a timeout mechanism, which we arbitrarily set to
4013  * apic_revector_timeout microseconds.
4014  */
4015 static void
4016 apic_mark_vector(uchar_t oldvector, uchar_t newvector)
4017 {
4018 	int iflag = intr_clear();
4019 	lock_set(&apic_revector_lock);
4020 	if (!apic_oldvec_to_newvec) {
4021 		apic_oldvec_to_newvec =
4022 		    kmem_zalloc(sizeof (newvector) * APIC_MAX_VECTOR * 2,
4023 		    KM_NOSLEEP);
4024 
4025 		if (!apic_oldvec_to_newvec) {
4026 			/*
4027 			 * This failure is not catastrophic.
4028 			 * But, the oldvec will never be freed.
4029 			 */
4030 			apic_error |= APIC_ERR_MARK_VECTOR_FAIL;
4031 			lock_clear(&apic_revector_lock);
4032 			intr_restore(iflag);
4033 			return;
4034 		}
4035 		apic_newvec_to_oldvec = &apic_oldvec_to_newvec[APIC_MAX_VECTOR];
4036 	}
4037 
4038 	/* See if we already did this for drivers which do double addintrs */
4039 	if (apic_oldvec_to_newvec[oldvector] != newvector) {
4040 		apic_oldvec_to_newvec[oldvector] = newvector;
4041 		apic_newvec_to_oldvec[newvector] = oldvector;
4042 		apic_revector_pending++;
4043 	}
4044 	lock_clear(&apic_revector_lock);
4045 	intr_restore(iflag);
4046 	(void) timeout(apic_xlate_vector_free_timeout_handler,
4047 	    (void *)(uintptr_t)oldvector, drv_usectohz(apic_revector_timeout));
4048 }
4049 
4050 /*
4051  * xlate_vector is called from intr_enter if revector_pending is set.
4052  * It will xlate it if needed and mark the old vector as free.
4053  */
4054 static uchar_t
4055 apic_xlate_vector(uchar_t vector)
4056 {
4057 	uchar_t	newvector, oldvector = 0;
4058 
4059 	lock_set(&apic_revector_lock);
4060 	/* Do we really need to do this ? */
4061 	if (!apic_revector_pending) {
4062 		lock_clear(&apic_revector_lock);
4063 		return (vector);
4064 	}
4065 	if ((newvector = apic_oldvec_to_newvec[vector]) != 0)
4066 		oldvector = vector;
4067 	else {
4068 		/*
4069 		 * The incoming vector is new . See if a stale entry is
4070 		 * remaining
4071 		 */
4072 		if ((oldvector = apic_newvec_to_oldvec[vector]) != 0)
4073 			newvector = vector;
4074 	}
4075 
4076 	if (oldvector) {
4077 		apic_revector_pending--;
4078 		apic_oldvec_to_newvec[oldvector] = 0;
4079 		apic_newvec_to_oldvec[newvector] = 0;
4080 		apic_free_vector(oldvector);
4081 		lock_clear(&apic_revector_lock);
4082 		/* There could have been more than one reprogramming! */
4083 		return (apic_xlate_vector(newvector));
4084 	}
4085 	lock_clear(&apic_revector_lock);
4086 	return (vector);
4087 }
4088 
4089 void
4090 apic_xlate_vector_free_timeout_handler(void *arg)
4091 {
4092 	int iflag;
4093 	uchar_t oldvector, newvector;
4094 
4095 	oldvector = (uchar_t)(uintptr_t)arg;
4096 	iflag = intr_clear();
4097 	lock_set(&apic_revector_lock);
4098 	if ((newvector = apic_oldvec_to_newvec[oldvector]) != 0) {
4099 		apic_free_vector(oldvector);
4100 		apic_oldvec_to_newvec[oldvector] = 0;
4101 		apic_newvec_to_oldvec[newvector] = 0;
4102 		apic_revector_pending--;
4103 	}
4104 
4105 	lock_clear(&apic_revector_lock);
4106 	intr_restore(iflag);
4107 }
4108 
4109 
4110 /* Mark vector as not being used by any irq */
4111 static void
4112 apic_free_vector(uchar_t vector)
4113 {
4114 	apic_vector_to_irq[vector] = APIC_RESV_IRQ;
4115 }
4116 
4117 /*
4118  * compute the polarity, trigger mode and vector for programming into
4119  * the I/O apic and record in airq_rdt_entry.
4120  */
4121 static void
4122 apic_record_rdt_entry(apic_irq_t *irqptr, int irq)
4123 {
4124 	int	ioapicindex, bus_type, vector;
4125 	short	intr_index;
4126 	uint_t	level, po, io_po;
4127 	struct apic_io_intr *iointrp;
4128 
4129 	intr_index = irqptr->airq_mps_intr_index;
4130 	DDI_INTR_IMPLDBG((CE_CONT, "apic_record_rdt_entry: intr_index=%d "
4131 	    "irq = 0x%x dip = 0x%p vector = 0x%x\n", intr_index, irq,
4132 	    (void *)irqptr->airq_dip, irqptr->airq_vector));
4133 
4134 	if (intr_index == RESERVE_INDEX) {
4135 		apic_error |= APIC_ERR_INVALID_INDEX;
4136 		return;
4137 	} else if (APIC_IS_MSI_OR_MSIX_INDEX(intr_index)) {
4138 		return;
4139 	}
4140 
4141 	vector = irqptr->airq_vector;
4142 	ioapicindex = irqptr->airq_ioapicindex;
4143 	/* Assume edge triggered by default */
4144 	level = 0;
4145 	/* Assume active high by default */
4146 	po = 0;
4147 
4148 	if (intr_index == DEFAULT_INDEX || intr_index == FREE_INDEX) {
4149 		ASSERT(irq < 16);
4150 		if (eisa_level_intr_mask & (1 << irq))
4151 			level = AV_LEVEL;
4152 		if (intr_index == FREE_INDEX && apic_defconf == 0)
4153 			apic_error |= APIC_ERR_INVALID_INDEX;
4154 	} else if (intr_index == ACPI_INDEX) {
4155 		bus_type = irqptr->airq_iflag.bustype;
4156 		if (irqptr->airq_iflag.intr_el == INTR_EL_CONFORM) {
4157 			if (bus_type == BUS_PCI)
4158 				level = AV_LEVEL;
4159 		} else
4160 			level = (irqptr->airq_iflag.intr_el == INTR_EL_LEVEL) ?
4161 			    AV_LEVEL : 0;
4162 		if (level &&
4163 		    ((irqptr->airq_iflag.intr_po == INTR_PO_ACTIVE_LOW) ||
4164 		    (irqptr->airq_iflag.intr_po == INTR_PO_CONFORM &&
4165 		    bus_type == BUS_PCI)))
4166 			po = AV_ACTIVE_LOW;
4167 	} else {
4168 		iointrp = apic_io_intrp + intr_index;
4169 		bus_type = apic_find_bus(iointrp->intr_busid);
4170 		if (iointrp->intr_el == INTR_EL_CONFORM) {
4171 			if ((irq < 16) && (eisa_level_intr_mask & (1 << irq)))
4172 				level = AV_LEVEL;
4173 			else if (bus_type == BUS_PCI)
4174 				level = AV_LEVEL;
4175 		} else
4176 			level = (iointrp->intr_el == INTR_EL_LEVEL) ?
4177 			    AV_LEVEL : 0;
4178 		if (level && ((iointrp->intr_po == INTR_PO_ACTIVE_LOW) ||
4179 		    (iointrp->intr_po == INTR_PO_CONFORM &&
4180 		    bus_type == BUS_PCI)))
4181 			po = AV_ACTIVE_LOW;
4182 	}
4183 	if (level)
4184 		apic_level_intr[irq] = 1;
4185 	/*
4186 	 * The 82489DX External APIC cannot do active low polarity interrupts.
4187 	 */
4188 	if (po && (apic_io_ver[ioapicindex] != IOAPIC_VER_82489DX))
4189 		io_po = po;
4190 	else
4191 		io_po = 0;
4192 
4193 	if (apic_verbose & APIC_VERBOSE_IOAPIC_FLAG)
4194 		printf("setio: ioapic=%x intin=%x level=%x po=%x vector=%x\n",
4195 		    ioapicindex, irqptr->airq_intin_no, level, io_po, vector);
4196 
4197 	irqptr->airq_rdt_entry = level|io_po|vector;
4198 }
4199 
4200 /*
4201  * Call rebind to do the actual programming.
4202  */
4203 static int
4204 apic_setup_io_intr(apic_irq_t *irqptr, int irq)
4205 {
4206 	int rv;
4207 
4208 	if (rv = apic_rebind(irqptr, apic_irq_table[irq]->airq_cpu, 1,
4209 	    IMMEDIATE))
4210 		/* CPU is not up or interrupt is disabled. Fall back to 0 */
4211 		rv = apic_rebind(irqptr, 0, 1, IMMEDIATE);
4212 
4213 	return (rv);
4214 }
4215 
4216 /*
4217  * Deferred reprogramming: Call apic_rebind to do the real work.
4218  */
4219 static int
4220 apic_setup_io_intr_deferred(apic_irq_t *irqptr, int irq)
4221 {
4222 	int rv;
4223 
4224 	if (rv = apic_rebind(irqptr, apic_irq_table[irq]->airq_cpu, 1,
4225 	    DEFERRED))
4226 		/* CPU is not up or interrupt is disabled. Fall back to 0 */
4227 		rv = apic_rebind(irqptr, 0, 1, DEFERRED);
4228 
4229 	return (rv);
4230 }
4231 
4232 /*
4233  * Bind interrupt corresponding to irq_ptr to bind_cpu. acquire_lock
4234  * if false (0) means lock is already held (e.g: in rebind_all).
4235  */
4236 static int
4237 apic_rebind(apic_irq_t *irq_ptr, int bind_cpu, int acquire_lock, int when)
4238 {
4239 	int			intin_no;
4240 	volatile int32_t	*ioapic;
4241 	uchar_t			airq_temp_cpu;
4242 	apic_cpus_info_t	*cpu_infop;
4243 	int			iflag;
4244 	int		which_irq = apic_vector_to_irq[irq_ptr->airq_vector];
4245 
4246 	intin_no = irq_ptr->airq_intin_no;
4247 	ioapic = apicioadr[irq_ptr->airq_ioapicindex];
4248 	airq_temp_cpu = irq_ptr->airq_temp_cpu;
4249 	if (airq_temp_cpu != IRQ_UNINIT && airq_temp_cpu != IRQ_UNBOUND) {
4250 		if (airq_temp_cpu & IRQ_USER_BOUND)
4251 			/* Mask off high bit so it can be used as array index */
4252 			airq_temp_cpu &= ~IRQ_USER_BOUND;
4253 
4254 		ASSERT(airq_temp_cpu < apic_nproc);
4255 	}
4256 
4257 	iflag = intr_clear();
4258 
4259 	if (acquire_lock)
4260 		lock_set(&apic_ioapic_lock);
4261 
4262 	/*
4263 	 * Can't bind to a CPU that's not online:
4264 	 */
4265 	cpu_infop = &apic_cpus[bind_cpu & ~IRQ_USER_BOUND];
4266 	if (!(cpu_infop->aci_status & APIC_CPU_INTR_ENABLE)) {
4267 
4268 		if (acquire_lock)
4269 			lock_clear(&apic_ioapic_lock);
4270 
4271 		intr_restore(iflag);
4272 		return (1);
4273 	}
4274 
4275 	/*
4276 	 * If this is a deferred reprogramming attempt, ensure we have
4277 	 * not been passed stale data:
4278 	 */
4279 	if ((when == DEFERRED) &&
4280 	    (apic_reprogram_info[which_irq].valid == 0)) {
4281 		/* stale info, so just return */
4282 		if (acquire_lock)
4283 			lock_clear(&apic_ioapic_lock);
4284 
4285 		intr_restore(iflag);
4286 		return (0);
4287 	}
4288 
4289 	/*
4290 	 * If this interrupt has been delivered to a CPU and that CPU
4291 	 * has not handled it yet, we cannot reprogram the IOAPIC now:
4292 	 */
4293 	if (!APIC_IS_MSI_OR_MSIX_INDEX(irq_ptr->airq_mps_intr_index) &&
4294 	    apic_check_stuck_interrupt(irq_ptr, airq_temp_cpu, bind_cpu,
4295 	    ioapic, intin_no, which_irq) != 0) {
4296 
4297 		if (acquire_lock)
4298 			lock_clear(&apic_ioapic_lock);
4299 
4300 		intr_restore(iflag);
4301 		return (0);
4302 	}
4303 
4304 	/*
4305 	 * NOTE: We do not unmask the RDT here, as an interrupt MAY still
4306 	 * come in before we have a chance to reprogram it below.  The
4307 	 * reprogramming below will simultaneously change and unmask the
4308 	 * RDT entry.
4309 	 */
4310 
4311 	if ((uchar_t)bind_cpu == IRQ_UNBOUND) {
4312 		/* Write the RDT entry -- no specific CPU binding */
4313 		WRITE_IOAPIC_RDT_ENTRY_HIGH_DWORD(ioapic, intin_no, AV_TOALL);
4314 
4315 		if (airq_temp_cpu != IRQ_UNINIT && airq_temp_cpu != IRQ_UNBOUND)
4316 			apic_cpus[airq_temp_cpu].aci_temp_bound--;
4317 
4318 		/* Write the vector, trigger, and polarity portion of the RDT */
4319 		WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no,
4320 		    AV_LDEST | AV_LOPRI | irq_ptr->airq_rdt_entry);
4321 		if (acquire_lock)
4322 			lock_clear(&apic_ioapic_lock);
4323 		irq_ptr->airq_temp_cpu = IRQ_UNBOUND;
4324 		intr_restore(iflag);
4325 		return (0);
4326 	}
4327 
4328 	if (bind_cpu & IRQ_USER_BOUND) {
4329 		cpu_infop->aci_bound++;
4330 	} else {
4331 		cpu_infop->aci_temp_bound++;
4332 	}
4333 	ASSERT((bind_cpu & ~IRQ_USER_BOUND) < apic_nproc);
4334 	if (!APIC_IS_MSI_OR_MSIX_INDEX(irq_ptr->airq_mps_intr_index)) {
4335 		/* Write the RDT entry -- bind to a specific CPU: */
4336 		WRITE_IOAPIC_RDT_ENTRY_HIGH_DWORD(ioapic, intin_no,
4337 		    cpu_infop->aci_local_id << APIC_ID_BIT_OFFSET);
4338 	}
4339 	if ((airq_temp_cpu != IRQ_UNBOUND) && (airq_temp_cpu != IRQ_UNINIT)) {
4340 		apic_cpus[airq_temp_cpu].aci_temp_bound--;
4341 	}
4342 	if (!APIC_IS_MSI_OR_MSIX_INDEX(irq_ptr->airq_mps_intr_index)) {
4343 		/* Write the vector, trigger, and polarity portion of the RDT */
4344 		WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no,
4345 		    AV_PDEST | AV_FIXED | irq_ptr->airq_rdt_entry);
4346 	} else {
4347 		int type = (irq_ptr->airq_mps_intr_index == MSI_INDEX) ?
4348 		    DDI_INTR_TYPE_MSI : DDI_INTR_TYPE_MSIX;
4349 		(void) pci_msi_disable_mode(irq_ptr->airq_dip, type,
4350 		    irq_ptr->airq_ioapicindex);
4351 		if (irq_ptr->airq_ioapicindex == irq_ptr->airq_origirq) {
4352 			/* first one */
4353 			DDI_INTR_IMPLDBG((CE_CONT, "apic_rebind: call "
4354 			    "apic_pci_msi_enable_vector\n"));
4355 			if (apic_pci_msi_enable_vector(irq_ptr->airq_dip, type,
4356 			    which_irq, irq_ptr->airq_vector,
4357 			    irq_ptr->airq_intin_no,
4358 			    cpu_infop->aci_local_id) != PSM_SUCCESS) {
4359 				cmn_err(CE_WARN, "pcplusmp: "
4360 					"apic_pci_msi_enable_vector "
4361 					"returned PSM_FAILURE");
4362 			}
4363 		}
4364 		if ((irq_ptr->airq_ioapicindex + irq_ptr->airq_intin_no - 1) ==
4365 		    irq_ptr->airq_origirq) { /* last one */
4366 			DDI_INTR_IMPLDBG((CE_CONT, "apic_rebind: call "
4367 			    "pci_msi_enable_mode\n"));
4368 			if (pci_msi_enable_mode(irq_ptr->airq_dip, type,
4369 			    which_irq) != DDI_SUCCESS) {
4370 				DDI_INTR_IMPLDBG((CE_CONT, "pcplusmp: "
4371 				    "pci_msi_enable failed\n"));
4372 				(void) pci_msi_unconfigure(irq_ptr->airq_dip,
4373 				(irq_ptr->airq_mps_intr_index == MSI_INDEX) ?
4374 				DDI_INTR_TYPE_MSI : DDI_INTR_TYPE_MSIX,
4375 				which_irq);
4376 			}
4377 		}
4378 	}
4379 	if (acquire_lock)
4380 		lock_clear(&apic_ioapic_lock);
4381 	irq_ptr->airq_temp_cpu = (uchar_t)bind_cpu;
4382 	apic_redist_cpu_skip &= ~(1 << (bind_cpu & ~IRQ_USER_BOUND));
4383 	intr_restore(iflag);
4384 	return (0);
4385 }
4386 
4387 /*
4388  * Checks to see if the IOAPIC interrupt entry specified has its Remote IRR
4389  * bit set.  Sets up a timeout to perform the reprogramming at a later time
4390  * if it cannot wait for the Remote IRR bit to clear (or if waiting did not
4391  * result in the bit's clearing).
4392  *
4393  * This function will mask the RDT entry if the Remote IRR bit is set.
4394  *
4395  * Returns non-zero if the caller should defer IOAPIC reprogramming.
4396  */
4397 static int
4398 apic_check_stuck_interrupt(apic_irq_t *irq_ptr, int old_bind_cpu,
4399 	int new_bind_cpu, volatile int32_t *ioapic, int intin_no, int which_irq)
4400 {
4401 	int32_t			rdt_entry;
4402 	int			waited;
4403 
4404 	/* Mask the RDT entry, but only if it's a level-triggered interrupt */
4405 	rdt_entry = READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no);
4406 	if ((rdt_entry & (AV_LEVEL|AV_MASK)) == AV_LEVEL) {
4407 
4408 		/* Mask it */
4409 		WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no,
4410 		    AV_MASK | rdt_entry);
4411 	}
4412 
4413 	/*
4414 	 * Wait for the delivery pending bit to clear.
4415 	 */
4416 	if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no) &
4417 	    (AV_LEVEL|AV_PENDING)) == (AV_LEVEL|AV_PENDING)) {
4418 
4419 		/*
4420 		 * If we're still waiting on the delivery of this interrupt,
4421 		 * continue to wait here until it is delivered (this should be
4422 		 * a very small amount of time, but include a timeout just in
4423 		 * case).
4424 		 */
4425 		for (waited = 0; waited < apic_max_usecs_clear_pending;
4426 		    waited += APIC_USECS_PER_WAIT_INTERVAL) {
4427 			if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no)
4428 			    & AV_PENDING) == 0) {
4429 				break;
4430 			}
4431 			drv_usecwait(APIC_USECS_PER_WAIT_INTERVAL);
4432 		}
4433 
4434 		if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no) &
4435 		    AV_PENDING) != 0) {
4436 			cmn_err(CE_WARN, "!IOAPIC %d intin %d: Could not "
4437 			    "deliver interrupt to local APIC within "
4438 			    "%d usecs.", irq_ptr->airq_ioapicindex,
4439 			    irq_ptr->airq_intin_no,
4440 			    apic_max_usecs_clear_pending);
4441 		}
4442 	}
4443 
4444 	/*
4445 	 * If the remote IRR bit is set, then the interrupt has been sent
4446 	 * to a CPU for processing.  We have no choice but to wait for
4447 	 * that CPU to process the interrupt, at which point the remote IRR
4448 	 * bit will be cleared.
4449 	 */
4450 	if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no) &
4451 	    (AV_LEVEL|AV_REMOTE_IRR)) == (AV_LEVEL|AV_REMOTE_IRR)) {
4452 
4453 		/*
4454 		 * If the CPU that this RDT is bound to is NOT the current
4455 		 * CPU, wait until that CPU handles the interrupt and ACKs
4456 		 * it.  If this interrupt is not bound to any CPU (that is,
4457 		 * if it's bound to the logical destination of "anyone"), it
4458 		 * may have been delivered to the current CPU so handle that
4459 		 * case by deferring the reprogramming (below).
4460 		 */
4461 		kpreempt_disable();
4462 		if ((old_bind_cpu != IRQ_UNBOUND) &&
4463 		    (old_bind_cpu != IRQ_UNINIT) &&
4464 		    (old_bind_cpu != psm_get_cpu_id())) {
4465 			for (waited = 0; waited < apic_max_usecs_clear_pending;
4466 			    waited += APIC_USECS_PER_WAIT_INTERVAL) {
4467 				if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic,
4468 				    intin_no) & AV_REMOTE_IRR) == 0) {
4469 
4470 					/* Clear the reprogramming state: */
4471 					lock_set(&apic_ioapic_reprogram_lock);
4472 
4473 					apic_reprogram_info[which_irq].valid
4474 					    = 0;
4475 					apic_reprogram_info[which_irq].bindcpu
4476 					    = 0;
4477 					apic_reprogram_info[which_irq].timeouts
4478 					    = 0;
4479 
4480 					lock_clear(&apic_ioapic_reprogram_lock);
4481 
4482 					/* Remote IRR has cleared! */
4483 					kpreempt_enable();
4484 					return (0);
4485 				}
4486 				drv_usecwait(APIC_USECS_PER_WAIT_INTERVAL);
4487 			}
4488 		}
4489 		kpreempt_enable();
4490 
4491 		/*
4492 		 * If we waited and the Remote IRR bit is still not cleared,
4493 		 * AND if we've invoked the timeout APIC_REPROGRAM_MAX_TIMEOUTS
4494 		 * times for this interrupt, try the last-ditch workarounds:
4495 		 */
4496 		if (apic_reprogram_info[which_irq].timeouts >=
4497 		    APIC_REPROGRAM_MAX_TIMEOUTS) {
4498 
4499 			if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no)
4500 			    & AV_REMOTE_IRR) != 0) {
4501 				/*
4502 				 * Trying to clear the bit through normal
4503 				 * channels has failed.  So as a last-ditch
4504 				 * effort, try to set the trigger mode to
4505 				 * edge, then to level.  This has been
4506 				 * observed to work on many systems.
4507 				 */
4508 				WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic,
4509 				    intin_no,
4510 				    READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic,
4511 				    intin_no) & ~AV_LEVEL);
4512 
4513 				WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic,
4514 				    intin_no,
4515 				    READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic,
4516 				    intin_no) | AV_LEVEL);
4517 
4518 				/*
4519 				 * If the bit's STILL set, declare total and
4520 				 * utter failure
4521 				 */
4522 				if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic,
4523 				    intin_no) & AV_REMOTE_IRR) != 0) {
4524 					cmn_err(CE_WARN, "!IOAPIC %d intin %d: "
4525 					    "Remote IRR failed to reset "
4526 					    "within %d usecs.  Interrupts to "
4527 					    "this pin may cease to function.",
4528 					    irq_ptr->airq_ioapicindex,
4529 					    irq_ptr->airq_intin_no,
4530 					    apic_max_usecs_clear_pending);
4531 				}
4532 			}
4533 			/* Clear the reprogramming state: */
4534 			lock_set(&apic_ioapic_reprogram_lock);
4535 
4536 			apic_reprogram_info[which_irq].valid = 0;
4537 			apic_reprogram_info[which_irq].bindcpu = 0;
4538 			apic_reprogram_info[which_irq].timeouts = 0;
4539 
4540 			lock_clear(&apic_ioapic_reprogram_lock);
4541 		} else {
4542 #ifdef DEBUG
4543 			cmn_err(CE_WARN, "Deferring reprogramming of irq %d",
4544 			    which_irq);
4545 #endif	/* DEBUG */
4546 			/*
4547 			 * If waiting for the Remote IRR bit (above) didn't
4548 			 * allow it to clear, defer the reprogramming:
4549 			 */
4550 			lock_set(&apic_ioapic_reprogram_lock);
4551 
4552 			apic_reprogram_info[which_irq].valid = 1;
4553 			apic_reprogram_info[which_irq].bindcpu = new_bind_cpu;
4554 			apic_reprogram_info[which_irq].timeouts++;
4555 
4556 			lock_clear(&apic_ioapic_reprogram_lock);
4557 
4558 			/* Fire up a timeout to handle this later */
4559 			(void) timeout(apic_reprogram_timeout_handler,
4560 			    (void *) 0,
4561 			    drv_usectohz(APIC_REPROGRAM_TIMEOUT_DELAY));
4562 
4563 			/* Inform caller to defer IOAPIC programming: */
4564 			return (1);
4565 		}
4566 	}
4567 	return (0);
4568 }
4569 
4570 /*
4571  * Timeout handler that performs the APIC reprogramming
4572  */
4573 /*ARGSUSED*/
4574 static void
4575 apic_reprogram_timeout_handler(void *arg)
4576 {
4577 	/*LINTED: set but not used in function*/
4578 	int i, result;
4579 
4580 	/* Serialize access to this function */
4581 	mutex_enter(&apic_reprogram_timeout_mutex);
4582 
4583 	/*
4584 	 * For each entry in the reprogramming state that's valid,
4585 	 * try the reprogramming again:
4586 	 */
4587 	for (i = 0; i < APIC_MAX_VECTOR; i++) {
4588 		if (apic_reprogram_info[i].valid == 0)
4589 			continue;
4590 		/*
4591 		 * Though we can't really do anything about errors
4592 		 * at this point, keep track of them for reporting.
4593 		 * Note that it is very possible for apic_setup_io_intr
4594 		 * to re-register this very timeout if the Remote IRR bit
4595 		 * has not yet cleared.
4596 		 */
4597 		result = apic_setup_io_intr_deferred(apic_irq_table[i], i);
4598 
4599 #ifdef DEBUG
4600 		if (result)
4601 			cmn_err(CE_WARN, "apic_reprogram_timeout: "
4602 			    "apic_setup_io_intr returned nonzero for "
4603 			    "irq=%d!", i);
4604 #endif	/* DEBUG */
4605 	}
4606 
4607 	mutex_exit(&apic_reprogram_timeout_mutex);
4608 }
4609 
4610 
4611 /*
4612  * Called to migrate all interrupts at an irq to another cpu. safe
4613  * if true means we are not being called from an interrupt
4614  * context and hence it is safe to do a lock_set. If false
4615  * do only a lock_try and return failure ( non 0 ) if we cannot get it
4616  */
4617 int
4618 apic_rebind_all(apic_irq_t *irq_ptr, int bind_cpu, int safe)
4619 {
4620 	apic_irq_t	*irqptr = irq_ptr;
4621 	int		retval = 0;
4622 	int		iflag;
4623 
4624 	iflag = intr_clear();
4625 	if (!safe) {
4626 		if (lock_try(&apic_ioapic_lock) == 0) {
4627 			intr_restore(iflag);
4628 			return (1);
4629 		}
4630 	} else
4631 		lock_set(&apic_ioapic_lock);
4632 
4633 	while (irqptr) {
4634 		if (irqptr->airq_temp_cpu != IRQ_UNINIT)
4635 			retval |= apic_rebind(irqptr, bind_cpu, 0, IMMEDIATE);
4636 		irqptr = irqptr->airq_next;
4637 	}
4638 	lock_clear(&apic_ioapic_lock);
4639 	intr_restore(iflag);
4640 	return (retval);
4641 }
4642 
4643 /*
4644  * apic_intr_redistribute does all the messy computations for identifying
4645  * which interrupt to move to which CPU. Currently we do just one interrupt
4646  * at a time. This reduces the time we spent doing all this within clock
4647  * interrupt. When it is done in idle, we could do more than 1.
4648  * First we find the most busy and the most free CPU (time in ISR only)
4649  * skipping those CPUs that has been identified as being ineligible (cpu_skip)
4650  * Then we look for IRQs which are closest to the difference between the
4651  * most busy CPU and the average ISR load. We try to find one whose load
4652  * is less than difference.If none exists, then we chose one larger than the
4653  * difference, provided it does not make the most idle CPU worse than the
4654  * most busy one. In the end, we clear all the busy fields for CPUs. For
4655  * IRQs, they are cleared as they are scanned.
4656  */
4657 static void
4658 apic_intr_redistribute()
4659 {
4660 	int busiest_cpu, most_free_cpu;
4661 	int cpu_free, cpu_busy, max_busy, min_busy;
4662 	int min_free, diff;
4663 	int	average_busy, cpus_online;
4664 	int i, busy;
4665 	apic_cpus_info_t *cpu_infop;
4666 	apic_irq_t *min_busy_irq = NULL;
4667 	apic_irq_t *max_busy_irq = NULL;
4668 
4669 	busiest_cpu = most_free_cpu = -1;
4670 	cpu_free = cpu_busy = max_busy = average_busy = 0;
4671 	min_free = apic_sample_factor_redistribution;
4672 	cpus_online = 0;
4673 	/*
4674 	 * Below we will check for CPU_INTR_ENABLE, bound, temp_bound, temp_cpu
4675 	 * without ioapic_lock. That is OK as we are just doing statistical
4676 	 * sampling anyway and any inaccuracy now will get corrected next time
4677 	 * The call to rebind which actually changes things will make sure
4678 	 * we are consistent.
4679 	 */
4680 	for (i = 0; i < apic_nproc; i++) {
4681 		if (!(apic_redist_cpu_skip & (1 << i)) &&
4682 		    (apic_cpus[i].aci_status & APIC_CPU_INTR_ENABLE)) {
4683 
4684 			cpu_infop = &apic_cpus[i];
4685 			/*
4686 			 * If no unbound interrupts or only 1 total on this
4687 			 * CPU, skip
4688 			 */
4689 			if (!cpu_infop->aci_temp_bound ||
4690 			    (cpu_infop->aci_bound + cpu_infop->aci_temp_bound)
4691 			    == 1) {
4692 				apic_redist_cpu_skip |= 1 << i;
4693 				continue;
4694 			}
4695 
4696 			busy = cpu_infop->aci_busy;
4697 			average_busy += busy;
4698 			cpus_online++;
4699 			if (max_busy < busy) {
4700 				max_busy = busy;
4701 				busiest_cpu = i;
4702 			}
4703 			if (min_free > busy) {
4704 				min_free = busy;
4705 				most_free_cpu = i;
4706 			}
4707 			if (busy > apic_int_busy_mark) {
4708 				cpu_busy |= 1 << i;
4709 			} else {
4710 				if (busy < apic_int_free_mark)
4711 					cpu_free |= 1 << i;
4712 			}
4713 		}
4714 	}
4715 	if ((cpu_busy && cpu_free) ||
4716 	    (max_busy >= (min_free + apic_diff_for_redistribution))) {
4717 
4718 		apic_num_imbalance++;
4719 #ifdef	DEBUG
4720 		if (apic_verbose & APIC_VERBOSE_IOAPIC_FLAG) {
4721 			prom_printf(
4722 			    "redistribute busy=%x free=%x max=%x min=%x",
4723 			    cpu_busy, cpu_free, max_busy, min_free);
4724 		}
4725 #endif /* DEBUG */
4726 
4727 
4728 		average_busy /= cpus_online;
4729 
4730 		diff = max_busy - average_busy;
4731 		min_busy = max_busy; /* start with the max possible value */
4732 		max_busy = 0;
4733 		min_busy_irq = max_busy_irq = NULL;
4734 		i = apic_min_device_irq;
4735 		for (; i < apic_max_device_irq; i++) {
4736 			apic_irq_t *irq_ptr;
4737 			/* Change to linked list per CPU ? */
4738 			if ((irq_ptr = apic_irq_table[i]) == NULL)
4739 				continue;
4740 			/* Check for irq_busy & decide which one to move */
4741 			/* Also zero them for next round */
4742 			if ((irq_ptr->airq_temp_cpu == busiest_cpu) &&
4743 			    irq_ptr->airq_busy) {
4744 				if (irq_ptr->airq_busy < diff) {
4745 					/*
4746 					 * Check for least busy CPU,
4747 					 * best fit or what ?
4748 					 */
4749 					if (max_busy < irq_ptr->airq_busy) {
4750 						/*
4751 						 * Most busy within the
4752 						 * required differential
4753 						 */
4754 						max_busy = irq_ptr->airq_busy;
4755 						max_busy_irq = irq_ptr;
4756 					}
4757 				} else {
4758 					if (min_busy > irq_ptr->airq_busy) {
4759 						/*
4760 						 * least busy, but more than
4761 						 * the reqd diff
4762 						 */
4763 						if (min_busy <
4764 						    (diff + average_busy -
4765 						    min_free)) {
4766 							/*
4767 							 * Making sure new cpu
4768 							 * will not end up
4769 							 * worse
4770 							 */
4771 							min_busy =
4772 							    irq_ptr->airq_busy;
4773 
4774 							min_busy_irq = irq_ptr;
4775 						}
4776 					}
4777 				}
4778 			}
4779 			irq_ptr->airq_busy = 0;
4780 		}
4781 
4782 		if (max_busy_irq != NULL) {
4783 #ifdef	DEBUG
4784 			if (apic_verbose & APIC_VERBOSE_IOAPIC_FLAG) {
4785 				prom_printf("rebinding %x to %x",
4786 				    max_busy_irq->airq_vector, most_free_cpu);
4787 			}
4788 #endif /* DEBUG */
4789 			if (apic_rebind_all(max_busy_irq, most_free_cpu, 0)
4790 			    == 0)
4791 				/* Make change permenant */
4792 				max_busy_irq->airq_cpu = (uchar_t)most_free_cpu;
4793 		} else if (min_busy_irq != NULL) {
4794 #ifdef	DEBUG
4795 			if (apic_verbose & APIC_VERBOSE_IOAPIC_FLAG) {
4796 				prom_printf("rebinding %x to %x",
4797 				    min_busy_irq->airq_vector, most_free_cpu);
4798 			}
4799 #endif /* DEBUG */
4800 
4801 			if (apic_rebind_all(min_busy_irq, most_free_cpu, 0) ==
4802 			    0)
4803 				/* Make change permenant */
4804 				min_busy_irq->airq_cpu = (uchar_t)most_free_cpu;
4805 		} else {
4806 			if (cpu_busy != (1 << busiest_cpu)) {
4807 				apic_redist_cpu_skip |= 1 << busiest_cpu;
4808 				/*
4809 				 * We leave cpu_skip set so that next time we
4810 				 * can choose another cpu
4811 				 */
4812 			}
4813 		}
4814 		apic_num_rebind++;
4815 	} else {
4816 		/*
4817 		 * found nothing. Could be that we skipped over valid CPUs
4818 		 * or we have balanced everything. If we had a variable
4819 		 * ticks_for_redistribution, it could be increased here.
4820 		 * apic_int_busy, int_free etc would also need to be
4821 		 * changed.
4822 		 */
4823 		if (apic_redist_cpu_skip)
4824 			apic_redist_cpu_skip = 0;
4825 	}
4826 	for (i = 0; i < apic_nproc; i++) {
4827 		apic_cpus[i].aci_busy = 0;
4828 	}
4829 }
4830 
4831 static void
4832 apic_cleanup_busy()
4833 {
4834 	int i;
4835 	apic_irq_t *irq_ptr;
4836 
4837 	for (i = 0; i < apic_nproc; i++) {
4838 		apic_cpus[i].aci_busy = 0;
4839 	}
4840 
4841 	for (i = apic_min_device_irq; i < apic_max_device_irq; i++) {
4842 		if ((irq_ptr = apic_irq_table[i]) != NULL)
4843 			irq_ptr->airq_busy = 0;
4844 	}
4845 	apic_skipped_redistribute = 0;
4846 }
4847 
4848 
4849 /*
4850  * This function will reprogram the timer.
4851  *
4852  * When in oneshot mode the argument is the absolute time in future to
4853  * generate the interrupt at.
4854  *
4855  * When in periodic mode, the argument is the interval at which the
4856  * interrupts should be generated. There is no need to support the periodic
4857  * mode timer change at this time.
4858  */
4859 static void
4860 apic_timer_reprogram(hrtime_t time)
4861 {
4862 	hrtime_t now;
4863 	uint_t ticks;
4864 
4865 	/*
4866 	 * We should be called from high PIL context (CBE_HIGH_PIL),
4867 	 * so kpreempt is disabled.
4868 	 */
4869 
4870 	if (!apic_oneshot) {
4871 		/* time is the interval for periodic mode */
4872 		ticks = (uint_t)((time) / apic_nsec_per_tick);
4873 	} else {
4874 		/* one shot mode */
4875 
4876 		now = gethrtime();
4877 
4878 		if (time <= now) {
4879 			/*
4880 			 * requested to generate an interrupt in the past
4881 			 * generate an interrupt as soon as possible
4882 			 */
4883 			ticks = apic_min_timer_ticks;
4884 		} else if ((time - now) > apic_nsec_max) {
4885 			/*
4886 			 * requested to generate an interrupt at a time
4887 			 * further than what we are capable of. Set to max
4888 			 * the hardware can handle
4889 			 */
4890 
4891 			ticks = APIC_MAXVAL;
4892 #ifdef DEBUG
4893 			cmn_err(CE_CONT, "apic_timer_reprogram, request at"
4894 			    "  %lld  too far in future, current time"
4895 			    "  %lld \n", time, now);
4896 #endif	/* DEBUG */
4897 		} else
4898 			ticks = (uint_t)((time - now) / apic_nsec_per_tick);
4899 	}
4900 
4901 	if (ticks < apic_min_timer_ticks)
4902 		ticks = apic_min_timer_ticks;
4903 
4904 	apicadr[APIC_INIT_COUNT] = ticks;
4905 
4906 }
4907 
4908 /*
4909  * This function will enable timer interrupts.
4910  */
4911 static void
4912 apic_timer_enable(void)
4913 {
4914 	/*
4915 	 * We should be Called from high PIL context (CBE_HIGH_PIL),
4916 	 * so kpreempt is disabled.
4917 	 */
4918 
4919 	if (!apic_oneshot)
4920 		apicadr[APIC_LOCAL_TIMER] =
4921 		    (apic_clkvect + APIC_BASE_VECT) | AV_TIME;
4922 	else {
4923 		/* one shot */
4924 		apicadr[APIC_LOCAL_TIMER] = (apic_clkvect + APIC_BASE_VECT);
4925 	}
4926 }
4927 
4928 /*
4929  * This function will disable timer interrupts.
4930  */
4931 static void
4932 apic_timer_disable(void)
4933 {
4934 	/*
4935 	 * We should be Called from high PIL context (CBE_HIGH_PIL),
4936 	 * so kpreempt is disabled.
4937 	 */
4938 
4939 	apicadr[APIC_LOCAL_TIMER] = (apic_clkvect + APIC_BASE_VECT) | AV_MASK;
4940 }
4941 
4942 
4943 cyclic_id_t apic_cyclic_id;
4944 
4945 /*
4946  * If this module needs to be a consumer of cyclic subsystem, they
4947  * can be added here, since at this time kernel cyclic subsystem is initialized
4948  * argument is not currently used, and is reserved for future.
4949  */
4950 static void
4951 apic_post_cyclic_setup(void *arg)
4952 {
4953 _NOTE(ARGUNUSED(arg))
4954 	cyc_handler_t hdlr;
4955 	cyc_time_t when;
4956 
4957 	/* cpu_lock is held */
4958 
4959 	/* set up cyclics for intr redistribution */
4960 
4961 	/*
4962 	 * In peridoc mode intr redistribution processing is done in
4963 	 * apic_intr_enter during clk intr processing
4964 	 */
4965 	if (!apic_oneshot)
4966 		return;
4967 
4968 	hdlr.cyh_level = CY_LOW_LEVEL;
4969 	hdlr.cyh_func = (cyc_func_t)apic_redistribute_compute;
4970 	hdlr.cyh_arg = NULL;
4971 
4972 	when.cyt_when = 0;
4973 	when.cyt_interval = apic_redistribute_sample_interval;
4974 	apic_cyclic_id = cyclic_add(&hdlr, &when);
4975 
4976 
4977 }
4978 
4979 static void
4980 apic_redistribute_compute(void)
4981 {
4982 	int	i, j, max_busy;
4983 
4984 	if (apic_enable_dynamic_migration) {
4985 		if (++apic_nticks == apic_sample_factor_redistribution) {
4986 			/*
4987 			 * Time to call apic_intr_redistribute().
4988 			 * reset apic_nticks. This will cause max_busy
4989 			 * to be calculated below and if it is more than
4990 			 * apic_int_busy, we will do the whole thing
4991 			 */
4992 			apic_nticks = 0;
4993 		}
4994 		max_busy = 0;
4995 		for (i = 0; i < apic_nproc; i++) {
4996 
4997 			/*
4998 			 * Check if curipl is non zero & if ISR is in
4999 			 * progress
5000 			 */
5001 			if (((j = apic_cpus[i].aci_curipl) != 0) &&
5002 			    (apic_cpus[i].aci_ISR_in_progress & (1 << j))) {
5003 
5004 				int	irq;
5005 				apic_cpus[i].aci_busy++;
5006 				irq = apic_cpus[i].aci_current[j];
5007 				apic_irq_table[irq]->airq_busy++;
5008 			}
5009 
5010 			if (!apic_nticks &&
5011 			    (apic_cpus[i].aci_busy > max_busy))
5012 				max_busy = apic_cpus[i].aci_busy;
5013 		}
5014 		if (!apic_nticks) {
5015 			if (max_busy > apic_int_busy_mark) {
5016 			/*
5017 			 * We could make the following check be
5018 			 * skipped > 1 in which case, we get a
5019 			 * redistribution at half the busy mark (due to
5020 			 * double interval). Need to be able to collect
5021 			 * more empirical data to decide if that is a
5022 			 * good strategy. Punt for now.
5023 			 */
5024 				if (apic_skipped_redistribute)
5025 					apic_cleanup_busy();
5026 				else
5027 					apic_intr_redistribute();
5028 			} else
5029 				apic_skipped_redistribute++;
5030 		}
5031 	}
5032 }
5033 
5034 
5035 static int
5036 apic_acpi_translate_pci_irq(dev_info_t *dip, int busid, int devid,
5037     int ipin, int *pci_irqp, iflag_t *intr_flagp)
5038 {
5039 
5040 	int status;
5041 	acpi_psm_lnk_t acpipsmlnk;
5042 
5043 	if ((status = acpi_get_irq_cache_ent(busid, devid, ipin, pci_irqp,
5044 	    intr_flagp)) == ACPI_PSM_SUCCESS) {
5045 		APIC_VERBOSE_IRQ((CE_CONT, "!pcplusmp: Found irqno %d "
5046 		    "from cache for device %s, instance #%d\n", *pci_irqp,
5047 		    ddi_get_name(dip), ddi_get_instance(dip)));
5048 		return (status);
5049 	}
5050 
5051 	bzero(&acpipsmlnk, sizeof (acpi_psm_lnk_t));
5052 
5053 	if ((status = acpi_translate_pci_irq(dip, ipin, pci_irqp, intr_flagp,
5054 	    &acpipsmlnk)) == ACPI_PSM_FAILURE) {
5055 		APIC_VERBOSE_IRQ((CE_WARN, "pcplusmp: "
5056 		    " acpi_translate_pci_irq failed for device %s, instance"
5057 		    " #%d", ddi_get_name(dip), ddi_get_instance(dip)));
5058 		return (status);
5059 	}
5060 
5061 	if (status == ACPI_PSM_PARTIAL && acpipsmlnk.lnkobj != NULL) {
5062 		status = apic_acpi_irq_configure(&acpipsmlnk, dip, pci_irqp,
5063 		    intr_flagp);
5064 		if (status != ACPI_PSM_SUCCESS) {
5065 			status = acpi_get_current_irq_resource(&acpipsmlnk,
5066 			    pci_irqp, intr_flagp);
5067 		}
5068 	}
5069 
5070 	if (status == ACPI_PSM_SUCCESS) {
5071 		acpi_new_irq_cache_ent(busid, devid, ipin, *pci_irqp,
5072 		    intr_flagp, &acpipsmlnk);
5073 
5074 		APIC_VERBOSE_IRQ((CE_CONT, "pcplusmp: [ACPI] "
5075 		    "new irq %d for device %s, instance #%d\n",
5076 		    *pci_irqp, ddi_get_name(dip), ddi_get_instance(dip)));
5077 	}
5078 
5079 	return (status);
5080 }
5081 
5082 /*
5083  * Configures the irq for the interrupt link device identified by
5084  * acpipsmlnkp.
5085  *
5086  * Gets the current and the list of possible irq settings for the
5087  * device. If apic_unconditional_srs is not set, and the current
5088  * resource setting is in the list of possible irq settings,
5089  * current irq resource setting is passed to the caller.
5090  *
5091  * Otherwise, picks an irq number from the list of possible irq
5092  * settings, and sets the irq of the device to this value.
5093  * If prefer_crs is set, among a set of irq numbers in the list that have
5094  * the least number of devices sharing the interrupt, we pick current irq
5095  * resource setting if it is a member of this set.
5096  *
5097  * Passes the irq number in the value pointed to by pci_irqp, and
5098  * polarity and sensitivity in the structure pointed to by dipintrflagp
5099  * to the caller.
5100  *
5101  * Note that if setting the irq resource failed, but successfuly obtained
5102  * the current irq resource settings, passes the current irq resources
5103  * and considers it a success.
5104  *
5105  * Returns:
5106  * ACPI_PSM_SUCCESS on success.
5107  *
5108  * ACPI_PSM_FAILURE if an error occured during the configuration or
5109  * if a suitable irq was not found for this device, or if setting the
5110  * irq resource and obtaining the current resource fails.
5111  *
5112  */
5113 static int
5114 apic_acpi_irq_configure(acpi_psm_lnk_t *acpipsmlnkp, dev_info_t *dip,
5115     int *pci_irqp, iflag_t *dipintr_flagp)
5116 {
5117 
5118 	int i, min_share, foundnow, done = 0;
5119 	int32_t irq;
5120 	int32_t share_irq = -1;
5121 	int32_t chosen_irq = -1;
5122 	int cur_irq = -1;
5123 	acpi_irqlist_t *irqlistp;
5124 	acpi_irqlist_t *irqlistent;
5125 
5126 	if ((acpi_get_possible_irq_resources(acpipsmlnkp, &irqlistp))
5127 	    == ACPI_PSM_FAILURE) {
5128 		APIC_VERBOSE_IRQ((CE_WARN, "!pcplusmp: Unable to determine "
5129 		    "or assign IRQ for device %s, instance #%d: The system was "
5130 		    "unable to get the list of potential IRQs from ACPI.",
5131 		    ddi_get_name(dip), ddi_get_instance(dip)));
5132 
5133 		return (ACPI_PSM_FAILURE);
5134 	}
5135 
5136 	if ((acpi_get_current_irq_resource(acpipsmlnkp, &cur_irq,
5137 	    dipintr_flagp) == ACPI_PSM_SUCCESS) && (!apic_unconditional_srs) &&
5138 	    (cur_irq > 0)) {
5139 		/*
5140 		 * If an IRQ is set in CRS and that IRQ exists in the set
5141 		 * returned from _PRS, return that IRQ, otherwise print
5142 		 * a warning
5143 		 */
5144 
5145 		if (acpi_irqlist_find_irq(irqlistp, cur_irq, NULL)
5146 		    == ACPI_PSM_SUCCESS) {
5147 
5148 			acpi_free_irqlist(irqlistp);
5149 			ASSERT(pci_irqp != NULL);
5150 			*pci_irqp = cur_irq;
5151 			return (ACPI_PSM_SUCCESS);
5152 		}
5153 
5154 		APIC_VERBOSE_IRQ((CE_WARN, "!pcplusmp: Could not find the "
5155 		    "current irq %d for device %s, instance #%d in ACPI's "
5156 		    "list of possible irqs for this device. Picking one from "
5157 		    " the latter list.", cur_irq, ddi_get_name(dip),
5158 		    ddi_get_instance(dip)));
5159 	}
5160 
5161 	irqlistent = irqlistp;
5162 	min_share = 255;
5163 
5164 	while (irqlistent != NULL) {
5165 		irqlistent->intr_flags.bustype = BUS_PCI;
5166 
5167 		for (foundnow = 0, i = 0; i < irqlistent->num_irqs; i++) {
5168 
5169 			irq = irqlistent->irqs[i];
5170 
5171 			if ((irq < 16) && (apic_reserved_irqlist[irq]))
5172 				continue;
5173 
5174 			if (irq == 0) {
5175 				/* invalid irq number */
5176 				continue;
5177 			}
5178 
5179 			if ((apic_irq_table[irq] == NULL) ||
5180 			    (apic_irq_table[irq]->airq_dip == dip)) {
5181 				chosen_irq = irq;
5182 				foundnow = 1;
5183 				/*
5184 				 * If we do not prefer current irq from crs
5185 				 * or if we do and this irq is the same as
5186 				 * current irq from crs, this is the one
5187 				 * to pick.
5188 				 */
5189 				if (!(apic_prefer_crs) || (irq == cur_irq)) {
5190 					done = 1;
5191 					break;
5192 				}
5193 				continue;
5194 			}
5195 
5196 			if (irqlistent->intr_flags.intr_el == INTR_EL_EDGE)
5197 				continue;
5198 
5199 			if (!acpi_intr_compatible(irqlistent->intr_flags,
5200 			    apic_irq_table[irq]->airq_iflag))
5201 				continue;
5202 
5203 			if ((apic_irq_table[irq]->airq_share < min_share) ||
5204 			    ((apic_irq_table[irq]->airq_share == min_share) &&
5205 			    (cur_irq == irq) && (apic_prefer_crs))) {
5206 				min_share = apic_irq_table[irq]->airq_share;
5207 				share_irq = irq;
5208 				foundnow = 1;
5209 			}
5210 		}
5211 
5212 		/*
5213 		 * If we found an IRQ in the inner loop this time, save the
5214 		 * details from the irqlist for later use.
5215 		 */
5216 		if (foundnow && ((chosen_irq != -1) || (share_irq != -1))) {
5217 			/*
5218 			 * Copy the acpi_prs_private_t and flags from this
5219 			 * irq list entry, since we found an irq from this
5220 			 * entry.
5221 			 */
5222 			acpipsmlnkp->acpi_prs_prv = irqlistent->acpi_prs_prv;
5223 			*dipintr_flagp = irqlistent->intr_flags;
5224 		}
5225 
5226 		if (done)
5227 			break;
5228 
5229 		/* Go to the next irqlist entry */
5230 		irqlistent = irqlistent->next;
5231 	}
5232 
5233 
5234 	acpi_free_irqlist(irqlistp);
5235 	if (chosen_irq != -1)
5236 		irq = chosen_irq;
5237 	else if (share_irq != -1)
5238 		irq = share_irq;
5239 	else {
5240 		APIC_VERBOSE_IRQ((CE_WARN, "!pcplusmp: Could not find a "
5241 		    "suitable irq from the list of possible irqs for device "
5242 		    "%s, instance #%d in ACPI's list of possible irqs",
5243 		    ddi_get_name(dip), ddi_get_instance(dip)));
5244 		return (ACPI_PSM_FAILURE);
5245 	}
5246 
5247 	APIC_VERBOSE_IRQ((CE_CONT, "!pcplusmp: Setting irq %d for device %s "
5248 	    "instance #%d\n", irq, ddi_get_name(dip), ddi_get_instance(dip)));
5249 
5250 	if ((acpi_set_irq_resource(acpipsmlnkp, irq)) == ACPI_PSM_SUCCESS) {
5251 		/*
5252 		 * setting irq was successful, check to make sure CRS
5253 		 * reflects that. If CRS does not agree with what we
5254 		 * set, return the irq that was set.
5255 		 */
5256 
5257 		if (acpi_get_current_irq_resource(acpipsmlnkp, &cur_irq,
5258 		    dipintr_flagp) == ACPI_PSM_SUCCESS) {
5259 
5260 			if (cur_irq != irq)
5261 				APIC_VERBOSE_IRQ((CE_WARN, "!pcplusmp: "
5262 				    "IRQ resource set (irqno %d) for device %s "
5263 				    "instance #%d, differs from current "
5264 				    "setting irqno %d",
5265 				    irq, ddi_get_name(dip),
5266 				    ddi_get_instance(dip), cur_irq));
5267 		}
5268 
5269 		/*
5270 		 * return the irq that was set, and not what CRS reports,
5271 		 * since CRS has been seen to be bogus on some systems
5272 		 */
5273 		cur_irq = irq;
5274 	} else {
5275 		APIC_VERBOSE_IRQ((CE_WARN, "!pcplusmp: set resource irq %d "
5276 		    "failed for device %s instance #%d",
5277 		    irq, ddi_get_name(dip), ddi_get_instance(dip)));
5278 
5279 		if (cur_irq == -1)
5280 			return (ACPI_PSM_FAILURE);
5281 	}
5282 
5283 	ASSERT(pci_irqp != NULL);
5284 	*pci_irqp = cur_irq;
5285 	return (ACPI_PSM_SUCCESS);
5286 }
5287