xref: /titanic_41/usr/src/uts/i86pc/io/pcplusmp/apic.c (revision 74e20cfe817b82802b16fac8690dadcda76f54f5)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 /*
29  * PSMI 1.1 extensions are supported only in 2.6 and later versions.
30  * PSMI 1.2 extensions are supported only in 2.7 and later versions.
31  * PSMI 1.3 and 1.4 extensions are supported in Solaris 10.
32  * PSMI 1.5 extensions are supported in Solaris Nevada.
33  */
34 #define	PSMI_1_5
35 
36 #include <sys/processor.h>
37 #include <sys/time.h>
38 #include <sys/psm.h>
39 #include <sys/smp_impldefs.h>
40 #include <sys/cram.h>
41 #include <sys/acpi/acpi.h>
42 #include <sys/acpica.h>
43 #include <sys/psm_common.h>
44 #include "apic.h"
45 #include <sys/pit.h>
46 #include <sys/ddi.h>
47 #include <sys/sunddi.h>
48 #include <sys/ddi_impldefs.h>
49 #include <sys/pci.h>
50 #include <sys/promif.h>
51 #include <sys/x86_archext.h>
52 #include <sys/cpc_impl.h>
53 #include <sys/uadmin.h>
54 #include <sys/panic.h>
55 #include <sys/debug.h>
56 #include <sys/archsystm.h>
57 #include <sys/trap.h>
58 #include <sys/machsystm.h>
59 #include <sys/cpuvar.h>
60 #include <sys/rm_platter.h>
61 #include <sys/privregs.h>
62 #include <sys/cyclic.h>
63 #include <sys/note.h>
64 #include <sys/pci_intr_lib.h>
65 #include <sys/sunndi.h>
66 
67 struct ioapic_reprogram_data;
68 
69 /*
70  *	Local Function Prototypes
71  */
72 static void apic_init_intr();
73 static void apic_ret();
74 static int apic_handle_defconf();
75 static int apic_parse_mpct(caddr_t mpct, int bypass);
76 static struct apic_mpfps_hdr *apic_find_fps_sig(caddr_t fptr, int size);
77 static int apic_checksum(caddr_t bptr, int len);
78 static int get_apic_cmd1();
79 static int get_apic_pri();
80 static int apic_find_bus_type(char *bus);
81 static int apic_find_bus(int busid);
82 static int apic_find_bus_id(int bustype);
83 static struct apic_io_intr *apic_find_io_intr(int irqno);
84 int apic_allocate_irq(int irq);
85 static int apic_find_free_irq(int start, int end);
86 static uchar_t apic_allocate_vector(int ipl, int irq, int pri);
87 static void apic_modify_vector(uchar_t vector, int irq);
88 static void apic_mark_vector(uchar_t oldvector, uchar_t newvector);
89 static uchar_t apic_xlate_vector(uchar_t oldvector);
90 static void apic_xlate_vector_free_timeout_handler(void *arg);
91 static void apic_free_vector(uchar_t vector);
92 static void apic_reprogram_timeout_handler(void *arg);
93 static int apic_check_stuck_interrupt(apic_irq_t *irq_ptr, int old_bind_cpu,
94     int new_bind_cpu, volatile int32_t *ioapic, int intin_no, int which_irq,
95     struct ioapic_reprogram_data *drep);
96 static int apic_setup_io_intr(void *p, int irq, boolean_t deferred);
97 static void apic_record_rdt_entry(apic_irq_t *irqptr, int irq);
98 static struct apic_io_intr *apic_find_io_intr_w_busid(int irqno, int busid);
99 static int apic_find_intin(uchar_t ioapic, uchar_t intin);
100 static int apic_handle_pci_pci_bridge(dev_info_t *idip, int child_devno,
101     int child_ipin, struct apic_io_intr **intrp);
102 static int apic_setup_irq_table(dev_info_t *dip, int irqno,
103     struct apic_io_intr *intrp, struct intrspec *ispec, iflag_t *intr_flagp,
104     int type);
105 static int apic_setup_sci_irq_table(int irqno, uchar_t ipl,
106     iflag_t *intr_flagp);
107 static void apic_nmi_intr(caddr_t arg);
108 uchar_t apic_bind_intr(dev_info_t *dip, int irq, uchar_t ioapicid,
109     uchar_t intin);
110 static int apic_rebind(apic_irq_t *irq_ptr, int bind_cpu,
111     struct ioapic_reprogram_data *drep);
112 int apic_rebind_all(apic_irq_t *irq_ptr, int bind_cpu);
113 static void apic_intr_redistribute();
114 static void apic_cleanup_busy();
115 static void apic_set_pwroff_method_from_mpcnfhdr(struct apic_mp_cnf_hdr *hdrp);
116 int apic_introp_xlate(dev_info_t *dip, struct intrspec *ispec, int type);
117 static void apic_try_deferred_reprogram(int ipl, int vect);
118 static void delete_defer_repro_ent(int which_irq);
119 static void apic_ioapic_wait_pending_clear(volatile int32_t *ioapic,
120     int intin_no);
121 
122 /* ACPI support routines */
123 static int acpi_probe(void);
124 static int apic_acpi_irq_configure(acpi_psm_lnk_t *acpipsmlnkp, dev_info_t *dip,
125     int *pci_irqp, iflag_t *intr_flagp);
126 
127 static int apic_acpi_translate_pci_irq(dev_info_t *dip, int busid, int devid,
128     int ipin, int *pci_irqp, iflag_t *intr_flagp);
129 static uchar_t acpi_find_ioapic(int irq);
130 static int acpi_intr_compatible(iflag_t iflag1, iflag_t iflag2);
131 
132 /*
133  *	standard MP entries
134  */
135 static int	apic_probe();
136 static int	apic_clkinit();
137 static int	apic_getclkirq(int ipl);
138 static uint_t	apic_calibrate(volatile uint32_t *addr,
139     uint16_t *pit_ticks_adj);
140 static hrtime_t apic_gettime();
141 static hrtime_t apic_gethrtime();
142 static void	apic_init();
143 static void	apic_picinit(void);
144 static void	apic_cpu_start(processorid_t cpun, caddr_t rm_code);
145 static int	apic_post_cpu_start(void);
146 static void	apic_send_ipi(int cpun, int ipl);
147 static void	apic_set_softintr(int softintr);
148 static void	apic_set_idlecpu(processorid_t cpun);
149 static void	apic_unset_idlecpu(processorid_t cpun);
150 static int	apic_softlvl_to_irq(int ipl);
151 static int	apic_intr_enter(int ipl, int *vect);
152 static void	apic_intr_exit(int ipl, int vect);
153 static void	apic_setspl(int ipl);
154 static int	apic_addspl(int ipl, int vector, int min_ipl, int max_ipl);
155 static int	apic_delspl(int ipl, int vector, int min_ipl, int max_ipl);
156 static void	apic_shutdown(int cmd, int fcn);
157 static void	apic_preshutdown(int cmd, int fcn);
158 static int	apic_disable_intr(processorid_t cpun);
159 static void	apic_enable_intr(processorid_t cpun);
160 static processorid_t	apic_get_next_processorid(processorid_t cpun);
161 static int		apic_get_ipivect(int ipl, int type);
162 static void	apic_timer_reprogram(hrtime_t time);
163 static void	apic_timer_enable(void);
164 static void	apic_timer_disable(void);
165 static void	apic_post_cyclic_setup(void *arg);
166 extern int	apic_intr_ops(dev_info_t *, ddi_intr_handle_impl_t *,
167 		    psm_intr_op_t, int *);
168 
169 static int	apic_oneshot = 0;
170 int	apic_oneshot_enable = 1; /* to allow disabling one-shot capability */
171 
172 /*
173  * These variables are frequently accessed in apic_intr_enter(),
174  * apic_intr_exit and apic_setspl, so group them together
175  */
176 volatile uint32_t *apicadr =  NULL;	/* virtual addr of local APIC	*/
177 int apic_setspl_delay = 1;		/* apic_setspl - delay enable	*/
178 int apic_clkvect;
179 
180 /* ACPI SCI interrupt configuration; -1 if SCI not used */
181 int apic_sci_vect = -1;
182 iflag_t apic_sci_flags;
183 
184 /* vector at which error interrupts come in */
185 int apic_errvect;
186 int apic_enable_error_intr = 1;
187 int apic_error_display_delay = 100;
188 
189 /* vector at which performance counter overflow interrupts come in */
190 int apic_cpcovf_vect;
191 int apic_enable_cpcovf_intr = 1;
192 
193 /* Max wait time (in repetitions) for flags to clear in an RDT entry. */
194 static int apic_max_reps_clear_pending = 1000;
195 
196 /* Maximum number of times to retry reprogramming at apic_intr_exit time */
197 #define	APIC_REPROGRAM_MAX_TRIES 10000
198 
199 /*
200  * number of bits per byte, from <sys/param.h>
201  */
202 #define	UCHAR_MAX	((1 << NBBY) - 1)
203 
204 uchar_t	apic_reserved_irqlist[MAX_ISA_IRQ + 1];
205 
206 /*
207  * The following vector assignments influence the value of ipltopri and
208  * vectortoipl. Note that vectors 0 - 0x1f are not used. We can program
209  * idle to 0 and IPL 0 to 0x10 to differentiate idle in case
210  * we care to do so in future. Note some IPLs which are rarely used
211  * will share the vector ranges and heavily used IPLs (5 and 6) have
212  * a wide range.
213  *	IPL		Vector range.		as passed to intr_enter
214  *	0		none.
215  *	1,2,3		0x20-0x2f		0x0-0xf
216  *	4		0x30-0x3f		0x10-0x1f
217  *	5		0x40-0x5f		0x20-0x3f
218  *	6		0x60-0x7f		0x40-0x5f
219  *	7,8,9		0x80-0x8f		0x60-0x6f
220  *	10		0x90-0x9f		0x70-0x7f
221  *	11		0xa0-0xaf		0x80-0x8f
222  *	...		...
223  *	16		0xf0-0xff		0xd0-0xdf
224  */
225 uchar_t apic_vectortoipl[APIC_AVAIL_VECTOR / APIC_VECTOR_PER_IPL] = {
226 	3, 4, 5, 5, 6, 6, 9, 10, 11, 12, 13, 14, 15, 16
227 };
228 	/*
229 	 * The ipl of an ISR at vector X is apic_vectortoipl[X<<4]
230 	 * NOTE that this is vector as passed into intr_enter which is
231 	 * programmed vector - 0x20 (APIC_BASE_VECT)
232 	 */
233 
234 uchar_t	apic_ipltopri[MAXIPL + 1];	/* unix ipl to apic pri	*/
235 	/* The taskpri to be programmed into apic to mask given ipl */
236 
237 #if defined(__amd64)
238 uchar_t	apic_cr8pri[MAXIPL + 1];	/* unix ipl to cr8 pri	*/
239 #endif
240 
241 /*
242  * Patchable global variables.
243  */
244 int	apic_forceload = 0;
245 
246 #define	INTR_ROUND_ROBIN_WITH_AFFINITY	0
247 #define	INTR_ROUND_ROBIN		1
248 #define	INTR_LOWEST_PRIORITY		2
249 
250 int	apic_intr_policy = INTR_ROUND_ROBIN_WITH_AFFINITY;
251 
252 static int	apic_next_bind_cpu = 1; /* For round robin assignment */
253 					/* start with cpu 1 */
254 
255 int	apic_coarse_hrtime = 1;		/* 0 - use accurate slow gethrtime() */
256 					/* 1 - use gettime() for performance */
257 int	apic_flat_model = 0;		/* 0 - clustered. 1 - flat */
258 int	apic_enable_hwsoftint = 0;	/* 0 - disable, 1 - enable	*/
259 int	apic_enable_bind_log = 1;	/* 1 - display interrupt binding log */
260 int	apic_panic_on_nmi = 0;
261 int	apic_panic_on_apic_error = 0;
262 
263 int	apic_verbose = 0;
264 
265 /* Flag definitions for apic_verbose */
266 #define	APIC_VERBOSE_IOAPIC_FLAG		0x00000001
267 #define	APIC_VERBOSE_IRQ_FLAG			0x00000002
268 #define	APIC_VERBOSE_POWEROFF_FLAG		0x00000004
269 #define	APIC_VERBOSE_POWEROFF_PAUSE_FLAG	0x00000008
270 
271 
272 #define	APIC_VERBOSE_IOAPIC(fmt) \
273 	if (apic_verbose & APIC_VERBOSE_IOAPIC_FLAG) \
274 		cmn_err fmt;
275 
276 #define	APIC_VERBOSE_IRQ(fmt) \
277 	if (apic_verbose & APIC_VERBOSE_IRQ_FLAG) \
278 		cmn_err fmt;
279 
280 #define	APIC_VERBOSE_POWEROFF(fmt) \
281 	if (apic_verbose & APIC_VERBOSE_POWEROFF_FLAG) \
282 		prom_printf fmt;
283 
284 
285 /* Now the ones for Dynamic Interrupt distribution */
286 int	apic_enable_dynamic_migration = 0;
287 
288 /*
289  * If enabled, the distribution works as follows:
290  * On every interrupt entry, the current ipl for the CPU is set in cpu_info
291  * and the irq corresponding to the ipl is also set in the aci_current array.
292  * interrupt exit and setspl (due to soft interrupts) will cause the current
293  * ipl to be be changed. This is cache friendly as these frequently used
294  * paths write into a per cpu structure.
295  *
296  * Sampling is done by checking the structures for all CPUs and incrementing
297  * the busy field of the irq (if any) executing on each CPU and the busy field
298  * of the corresponding CPU.
299  * In periodic mode this is done on every clock interrupt.
300  * In one-shot mode, this is done thru a cyclic with an interval of
301  * apic_redistribute_sample_interval (default 10 milli sec).
302  *
303  * Every apic_sample_factor_redistribution times we sample, we do computations
304  * to decide which interrupt needs to be migrated (see comments
305  * before apic_intr_redistribute().
306  */
307 
308 /*
309  * Following 3 variables start as % and can be patched or set using an
310  * API to be defined in future. They will be scaled to
311  * sample_factor_redistribution which is in turn set to hertz+1 (in periodic
312  * mode), or 101 in one-shot mode to stagger it away from one sec processing
313  */
314 
315 int	apic_int_busy_mark = 60;
316 int	apic_int_free_mark = 20;
317 int	apic_diff_for_redistribution = 10;
318 
319 /* sampling interval for interrupt redistribution for dynamic migration */
320 int	apic_redistribute_sample_interval = NANOSEC / 100; /* 10 millisec */
321 
322 /*
323  * number of times we sample before deciding to redistribute interrupts
324  * for dynamic migration
325  */
326 int	apic_sample_factor_redistribution = 101;
327 
328 /* timeout for xlate_vector, mark_vector */
329 int	apic_revector_timeout = 16 * 10000; /* 160 millisec */
330 
331 int	apic_redist_cpu_skip = 0;
332 int	apic_num_imbalance = 0;
333 int	apic_num_rebind = 0;
334 
335 int	apic_nproc = 0;
336 int	apic_defconf = 0;
337 int	apic_irq_translate = 0;
338 int	apic_spec_rev = 0;
339 int	apic_imcrp = 0;
340 
341 int	apic_use_acpi = 1;	/* 1 = use ACPI, 0 = don't use ACPI */
342 int	apic_use_acpi_madt_only = 0;	/* 1=ONLY use MADT from ACPI */
343 
344 /*
345  * For interrupt link devices, if apic_unconditional_srs is set, an irq resource
346  * will be assigned (via _SRS). If it is not set, use the current
347  * irq setting (via _CRS), but only if that irq is in the set of possible
348  * irqs (returned by _PRS) for the device.
349  */
350 int	apic_unconditional_srs = 1;
351 
352 /*
353  * For interrupt link devices, if apic_prefer_crs is set when we are
354  * assigning an IRQ resource to a device, prefer the current IRQ setting
355  * over other possible irq settings under same conditions.
356  */
357 
358 int	apic_prefer_crs = 1;
359 
360 
361 /* minimum number of timer ticks to program to */
362 int apic_min_timer_ticks = 1;
363 /*
364  *	Local static data
365  */
366 static struct	psm_ops apic_ops = {
367 	apic_probe,
368 
369 	apic_init,
370 	apic_picinit,
371 	apic_intr_enter,
372 	apic_intr_exit,
373 	apic_setspl,
374 	apic_addspl,
375 	apic_delspl,
376 	apic_disable_intr,
377 	apic_enable_intr,
378 	apic_softlvl_to_irq,
379 	apic_set_softintr,
380 
381 	apic_set_idlecpu,
382 	apic_unset_idlecpu,
383 
384 	apic_clkinit,
385 	apic_getclkirq,
386 	(void (*)(void))NULL,		/* psm_hrtimeinit */
387 	apic_gethrtime,
388 
389 	apic_get_next_processorid,
390 	apic_cpu_start,
391 	apic_post_cpu_start,
392 	apic_shutdown,
393 	apic_get_ipivect,
394 	apic_send_ipi,
395 
396 	(int (*)(dev_info_t *, int))NULL,	/* psm_translate_irq */
397 	(int (*)(todinfo_t *))NULL,	/* psm_tod_get */
398 	(int (*)(todinfo_t *))NULL,	/* psm_tod_set */
399 	(void (*)(int, char *))NULL,	/* psm_notify_error */
400 	(void (*)(int))NULL,		/* psm_notify_func */
401 	apic_timer_reprogram,
402 	apic_timer_enable,
403 	apic_timer_disable,
404 	apic_post_cyclic_setup,
405 	apic_preshutdown,
406 	apic_intr_ops			/* Advanced DDI Interrupt framework */
407 };
408 
409 
410 static struct	psm_info apic_psm_info = {
411 	PSM_INFO_VER01_5,			/* version */
412 	PSM_OWN_EXCLUSIVE,			/* ownership */
413 	(struct psm_ops *)&apic_ops,		/* operation */
414 	"pcplusmp",				/* machine name */
415 	"pcplusmp v1.4 compatible %I%",
416 };
417 
418 static void *apic_hdlp;
419 
420 #ifdef DEBUG
421 #define	DENT		0x0001
422 int	apic_debug = 0;
423 /*
424  * set apic_restrict_vector to the # of vectors we want to allow per range
425  * useful in testing shared interrupt logic by setting it to 2 or 3
426  */
427 int	apic_restrict_vector = 0;
428 
429 #define	APIC_DEBUG_MSGBUFSIZE	2048
430 int	apic_debug_msgbuf[APIC_DEBUG_MSGBUFSIZE];
431 int	apic_debug_msgbufindex = 0;
432 
433 /*
434  * Put "int" info into debug buffer. No MP consistency, but light weight.
435  * Good enough for most debugging.
436  */
437 #define	APIC_DEBUG_BUF_PUT(x) \
438 	apic_debug_msgbuf[apic_debug_msgbufindex++] = x; \
439 	if (apic_debug_msgbufindex >= (APIC_DEBUG_MSGBUFSIZE - NCPU)) \
440 		apic_debug_msgbufindex = 0;
441 
442 #endif /* DEBUG */
443 
444 apic_cpus_info_t	*apic_cpus;
445 
446 static cpuset_t	apic_cpumask;
447 static uint_t	apic_flag;
448 
449 /* Flag to indicate that we need to shut down all processors */
450 static uint_t	apic_shutdown_processors;
451 
452 uint_t apic_nsec_per_intr = 0;
453 
454 /*
455  * apic_let_idle_redistribute can have the following values:
456  * 0 - If clock decremented it from 1 to 0, clock has to call redistribute.
457  * apic_redistribute_lock prevents multiple idle cpus from redistributing
458  */
459 int	apic_num_idle_redistributions = 0;
460 static	int apic_let_idle_redistribute = 0;
461 static	uint_t apic_nticks = 0;
462 static	uint_t apic_skipped_redistribute = 0;
463 
464 /* to gather intr data and redistribute */
465 static void apic_redistribute_compute(void);
466 
467 static	uint_t last_count_read = 0;
468 static	lock_t	apic_gethrtime_lock;
469 volatile int	apic_hrtime_stamp = 0;
470 volatile hrtime_t apic_nsec_since_boot = 0;
471 static uint_t apic_hertz_count;
472 
473 uint64_t apic_ticks_per_SFnsecs;	/* # of ticks in SF nsecs */
474 
475 static hrtime_t apic_nsec_max;
476 
477 static	hrtime_t	apic_last_hrtime = 0;
478 int		apic_hrtime_error = 0;
479 int		apic_remote_hrterr = 0;
480 int		apic_num_nmis = 0;
481 int		apic_apic_error = 0;
482 int		apic_num_apic_errors = 0;
483 int		apic_num_cksum_errors = 0;
484 
485 static	uchar_t	apic_io_id[MAX_IO_APIC];
486 static	uchar_t	apic_io_ver[MAX_IO_APIC];
487 static	uchar_t	apic_io_vectbase[MAX_IO_APIC];
488 static	uchar_t	apic_io_vectend[MAX_IO_APIC];
489 volatile int32_t *apicioadr[MAX_IO_APIC];
490 
491 /*
492  * First available slot to be used as IRQ index into the apic_irq_table
493  * for those interrupts (like MSI/X) that don't have a physical IRQ.
494  */
495 int apic_first_avail_irq  = APIC_FIRST_FREE_IRQ;
496 
497 /*
498  * apic_ioapic_lock protects the ioapics (reg select), the status, temp_bound
499  * and bound elements of cpus_info and the temp_cpu element of irq_struct
500  */
501 lock_t	apic_ioapic_lock;
502 
503 /*
504  * apic_defer_reprogram_lock ensures that only one processor is handling
505  * deferred interrupt programming at apic_intr_exit time.
506  */
507 static	lock_t	apic_defer_reprogram_lock;
508 
509 /*
510  * The current number of deferred reprogrammings outstanding
511  */
512 uint_t	apic_reprogram_outstanding = 0;
513 
514 #ifdef DEBUG
515 /*
516  * Counters that keep track of deferred reprogramming stats
517  */
518 uint_t	apic_intr_deferrals = 0;
519 uint_t	apic_intr_deliver_timeouts = 0;
520 uint_t	apic_last_ditch_reprogram_failures = 0;
521 uint_t	apic_deferred_setup_failures = 0;
522 uint_t	apic_defer_repro_total_retries = 0;
523 uint_t	apic_defer_repro_successes = 0;
524 uint_t	apic_deferred_spurious_enters = 0;
525 #endif
526 
527 static	int	apic_io_max = 0;	/* no. of i/o apics enabled */
528 
529 static	struct apic_io_intr *apic_io_intrp = 0;
530 static	struct apic_bus	*apic_busp;
531 
532 uchar_t	apic_vector_to_irq[APIC_MAX_VECTOR+1];
533 static	uchar_t	apic_resv_vector[MAXIPL+1];
534 
535 static	char	apic_level_intr[APIC_MAX_VECTOR+1];
536 static	int	apic_error = 0;
537 /* values which apic_error can take. Not catastrophic, but may help debug */
538 #define	APIC_ERR_BOOT_EOI		0x1
539 #define	APIC_ERR_GET_IPIVECT_FAIL	0x2
540 #define	APIC_ERR_INVALID_INDEX		0x4
541 #define	APIC_ERR_MARK_VECTOR_FAIL	0x8
542 #define	APIC_ERR_APIC_ERROR		0x40000000
543 #define	APIC_ERR_NMI			0x80000000
544 
545 static	int	apic_cmos_ssb_set = 0;
546 
547 static	uint32_t	eisa_level_intr_mask = 0;
548 	/* At least MSB will be set if EISA bus */
549 
550 static	int	apic_pci_bus_total = 0;
551 static	uchar_t	apic_single_pci_busid = 0;
552 
553 
554 /*
555  * airq_mutex protects additions to the apic_irq_table - the first
556  * pointer and any airq_nexts off of that one. It also protects
557  * apic_max_device_irq & apic_min_device_irq. It also guarantees
558  * that share_id is unique as new ids are generated only when new
559  * irq_t structs are linked in. Once linked in the structs are never
560  * deleted. temp_cpu & mps_intr_index field indicate if it is programmed
561  * or allocated. Note that there is a slight gap between allocating in
562  * apic_introp_xlate and programming in addspl.
563  */
564 kmutex_t	airq_mutex;
565 apic_irq_t	*apic_irq_table[APIC_MAX_VECTOR+1];
566 int		apic_max_device_irq = 0;
567 int		apic_min_device_irq = APIC_MAX_VECTOR;
568 
569 /* use to make sure only one cpu handles the nmi */
570 static	lock_t	apic_nmi_lock;
571 /* use to make sure only one cpu handles the error interrupt */
572 static	lock_t	apic_error_lock;
573 
574 /*
575  * Following declarations are for revectoring; used when ISRs at different
576  * IPLs share an irq.
577  */
578 static	lock_t	apic_revector_lock;
579 static	int	apic_revector_pending = 0;
580 static	uchar_t	*apic_oldvec_to_newvec;
581 static	uchar_t	*apic_newvec_to_oldvec;
582 
583 static	struct	ioapic_reprogram_data {
584 	boolean_t			done;
585 	apic_irq_t			*irqp;
586 	/* The CPU to which the int will be bound */
587 	int				bindcpu;
588 	/* # times the reprogram timeout was called */
589 	unsigned			tries;
590 
591 /* The irq # is implicit in the array index: */
592 } apic_reprogram_info[APIC_MAX_VECTOR + 1];
593 
594 /*
595  * APIC_MAX_VECTOR + 1 is the maximum # of IRQs as well. apic_reprogram_info
596  * is indexed by IRQ number, NOT by vector number.
597  */
598 
599 typedef struct prs_irq_list_ent {
600 	int			list_prio;
601 	int32_t			irq;
602 	iflag_t			intrflags;
603 	acpi_prs_private_t	prsprv;
604 	struct prs_irq_list_ent	*next;
605 } prs_irq_list_t;
606 
607 /*
608  * The following added to identify a software poweroff method if available.
609  */
610 
611 static struct {
612 	int	poweroff_method;
613 	char	oem_id[APIC_MPS_OEM_ID_LEN + 1];	/* MAX + 1 for NULL */
614 	char	prod_id[APIC_MPS_PROD_ID_LEN + 1];	/* MAX + 1 for NULL */
615 } apic_mps_ids[] = {
616 	{ APIC_POWEROFF_VIA_RTC,	"INTEL",	"ALDER" },   /* 4300 */
617 	{ APIC_POWEROFF_VIA_RTC,	"NCR",		"AMC" },    /* 4300 */
618 	{ APIC_POWEROFF_VIA_ASPEN_BMC,	"INTEL",	"A450NX" },  /* 4400? */
619 	{ APIC_POWEROFF_VIA_ASPEN_BMC,	"INTEL",	"AD450NX" }, /* 4400 */
620 	{ APIC_POWEROFF_VIA_ASPEN_BMC,	"INTEL",	"AC450NX" }, /* 4400R */
621 	{ APIC_POWEROFF_VIA_SITKA_BMC,	"INTEL",	"S450NX" },  /* S50  */
622 	{ APIC_POWEROFF_VIA_SITKA_BMC,	"INTEL",	"SC450NX" }  /* S50? */
623 };
624 
625 int	apic_poweroff_method = APIC_POWEROFF_NONE;
626 
627 static	struct {
628 	uchar_t	cntl;
629 	uchar_t	data;
630 } aspen_bmc[] = {
631 	{ CC_SMS_WR_START,	0x18 },		/* NetFn/LUN */
632 	{ CC_SMS_WR_NEXT,	0x24 },		/* Cmd SET_WATCHDOG_TIMER */
633 	{ CC_SMS_WR_NEXT,	0x84 },		/* DataByte 1: SMS/OS no log */
634 	{ CC_SMS_WR_NEXT,	0x2 },		/* DataByte 2: Power Down */
635 	{ CC_SMS_WR_NEXT,	0x0 },		/* DataByte 3: no pre-timeout */
636 	{ CC_SMS_WR_NEXT,	0x0 },		/* DataByte 4: timer expir. */
637 	{ CC_SMS_WR_NEXT,	0xa },		/* DataByte 5: init countdown */
638 	{ CC_SMS_WR_END,	0x0 },		/* DataByte 6: init countdown */
639 
640 	{ CC_SMS_WR_START,	0x18 },		/* NetFn/LUN */
641 	{ CC_SMS_WR_END,	0x22 }		/* Cmd RESET_WATCHDOG_TIMER */
642 };
643 
644 static	struct {
645 	int	port;
646 	uchar_t	data;
647 } sitka_bmc[] = {
648 	{ SMS_COMMAND_REGISTER,	SMS_WRITE_START },
649 	{ SMS_DATA_REGISTER,	0x18 },		/* NetFn/LUN */
650 	{ SMS_DATA_REGISTER,	0x24 },		/* Cmd SET_WATCHDOG_TIMER */
651 	{ SMS_DATA_REGISTER,	0x84 },		/* DataByte 1: SMS/OS no log */
652 	{ SMS_DATA_REGISTER,	0x2 },		/* DataByte 2: Power Down */
653 	{ SMS_DATA_REGISTER,	0x0 },		/* DataByte 3: no pre-timeout */
654 	{ SMS_DATA_REGISTER,	0x0 },		/* DataByte 4: timer expir. */
655 	{ SMS_DATA_REGISTER,	0xa },		/* DataByte 5: init countdown */
656 	{ SMS_COMMAND_REGISTER,	SMS_WRITE_END },
657 	{ SMS_DATA_REGISTER,	0x0 },		/* DataByte 6: init countdown */
658 
659 	{ SMS_COMMAND_REGISTER,	SMS_WRITE_START },
660 	{ SMS_DATA_REGISTER,	0x18 },		/* NetFn/LUN */
661 	{ SMS_COMMAND_REGISTER,	SMS_WRITE_END },
662 	{ SMS_DATA_REGISTER,	0x22 }		/* Cmd RESET_WATCHDOG_TIMER */
663 };
664 
665 
666 /* Patchable global variables. */
667 int		apic_kmdb_on_nmi = 0;		/* 0 - no, 1 - yes enter kmdb */
668 int		apic_debug_mps_id = 0;		/* 1 - print MPS ID strings */
669 uint32_t	apic_divide_reg_init = 0;	/* 0 - divide by 2 */
670 
671 /*
672  * ACPI definitions
673  */
674 /* _PIC method arguments */
675 #define	ACPI_PIC_MODE	0
676 #define	ACPI_APIC_MODE	1
677 
678 /* APIC error flags we care about */
679 #define	APIC_SEND_CS_ERROR	0x01
680 #define	APIC_RECV_CS_ERROR	0x02
681 #define	APIC_CS_ERRORS		(APIC_SEND_CS_ERROR|APIC_RECV_CS_ERROR)
682 
683 /*
684  * ACPI variables
685  */
686 /* 1 = acpi is enabled & working, 0 = acpi is not enabled or not there */
687 static	int apic_enable_acpi = 0;
688 
689 /* ACPI Multiple APIC Description Table ptr */
690 static	MULTIPLE_APIC_TABLE *acpi_mapic_dtp = NULL;
691 
692 /* ACPI Interrupt Source Override Structure ptr */
693 static	MADT_INTERRUPT_OVERRIDE *acpi_isop = NULL;
694 static	int acpi_iso_cnt = 0;
695 
696 /* ACPI Non-maskable Interrupt Sources ptr */
697 static	MADT_NMI_SOURCE *acpi_nmi_sp = NULL;
698 static	int acpi_nmi_scnt = 0;
699 static	MADT_LOCAL_APIC_NMI *acpi_nmi_cp = NULL;
700 static	int acpi_nmi_ccnt = 0;
701 
702 /*
703  * extern declarations
704  */
705 extern	int	intr_clear(void);
706 extern	void	intr_restore(uint_t);
707 #if defined(__amd64)
708 extern	int	intpri_use_cr8;
709 #endif	/* __amd64 */
710 
711 extern int	apic_pci_msi_enable_vector(dev_info_t *, int, int,
712 		    int, int, int);
713 extern apic_irq_t *apic_find_irq(dev_info_t *, struct intrspec *, int);
714 extern int	apic_pci_msi_unconfigure(dev_info_t *, int, int);
715 extern int	apic_pci_msi_disable_mode(dev_info_t *, int, int);
716 extern int	apic_pci_msi_enable_mode(dev_info_t *, int, int);
717 
718 /*
719  *	This is the loadable module wrapper
720  */
721 
722 int
723 _init(void)
724 {
725 	if (apic_coarse_hrtime)
726 		apic_ops.psm_gethrtime = &apic_gettime;
727 	return (psm_mod_init(&apic_hdlp, &apic_psm_info));
728 }
729 
730 int
731 _fini(void)
732 {
733 	return (psm_mod_fini(&apic_hdlp, &apic_psm_info));
734 }
735 
736 int
737 _info(struct modinfo *modinfop)
738 {
739 	return (psm_mod_info(&apic_hdlp, &apic_psm_info, modinfop));
740 }
741 
742 /*
743  * Auto-configuration routines
744  */
745 
746 /*
747  * Look at MPSpec 1.4 (Intel Order # 242016-005) for details of what we do here
748  * May work with 1.1 - but not guaranteed.
749  * According to the MP Spec, the MP floating pointer structure
750  * will be searched in the order described below:
751  * 1. In the first kilobyte of Extended BIOS Data Area (EBDA)
752  * 2. Within the last kilobyte of system base memory
753  * 3. In the BIOS ROM address space between 0F0000h and 0FFFFh
754  * Once we find the right signature with proper checksum, we call
755  * either handle_defconf or parse_mpct to get all info necessary for
756  * subsequent operations.
757  */
758 static int
759 apic_probe()
760 {
761 	uint32_t mpct_addr, ebda_start = 0, base_mem_end;
762 	caddr_t	biosdatap;
763 	caddr_t	mpct;
764 	caddr_t	fptr;
765 	int	i, mpct_size, mapsize, retval = PSM_FAILURE;
766 	ushort_t	ebda_seg, base_mem_size;
767 	struct	apic_mpfps_hdr	*fpsp;
768 	struct	apic_mp_cnf_hdr	*hdrp;
769 	int bypass_cpu_and_ioapics_in_mptables;
770 	int acpi_user_options;
771 
772 	if (apic_forceload < 0)
773 		return (retval);
774 
775 	/* Allow override for MADT-only mode */
776 	acpi_user_options = ddi_prop_get_int(DDI_DEV_T_ANY, ddi_root_node(), 0,
777 	    "acpi-user-options", 0);
778 	apic_use_acpi_madt_only = ((acpi_user_options & ACPI_OUSER_MADT) != 0);
779 
780 	/* Allow apic_use_acpi to override MADT-only mode */
781 	if (!apic_use_acpi)
782 		apic_use_acpi_madt_only = 0;
783 
784 	retval = acpi_probe();
785 
786 	/*
787 	 * mapin the bios data area 40:0
788 	 * 40:13h - two-byte location reports the base memory size
789 	 * 40:0Eh - two-byte location for the exact starting address of
790 	 *	    the EBDA segment for EISA
791 	 */
792 	biosdatap = psm_map_phys(0x400, 0x20, PROT_READ);
793 	if (!biosdatap)
794 		return (retval);
795 	fpsp = (struct apic_mpfps_hdr *)NULL;
796 	mapsize = MPFPS_RAM_WIN_LEN;
797 	/*LINTED: pointer cast may result in improper alignment */
798 	ebda_seg = *((ushort_t *)(biosdatap+0xe));
799 	/* check the 1k of EBDA */
800 	if (ebda_seg) {
801 		ebda_start = ((uint32_t)ebda_seg) << 4;
802 		fptr = psm_map_phys(ebda_start, MPFPS_RAM_WIN_LEN, PROT_READ);
803 		if (fptr) {
804 			if (!(fpsp =
805 			    apic_find_fps_sig(fptr, MPFPS_RAM_WIN_LEN)))
806 				psm_unmap_phys(fptr, MPFPS_RAM_WIN_LEN);
807 		}
808 	}
809 	/* If not in EBDA, check the last k of system base memory */
810 	if (!fpsp) {
811 		/*LINTED: pointer cast may result in improper alignment */
812 		base_mem_size = *((ushort_t *)(biosdatap + 0x13));
813 
814 		if (base_mem_size > 512)
815 			base_mem_end = 639 * 1024;
816 		else
817 			base_mem_end = 511 * 1024;
818 		/* if ebda == last k of base mem, skip to check BIOS ROM */
819 		if (base_mem_end != ebda_start) {
820 
821 			fptr = psm_map_phys(base_mem_end, MPFPS_RAM_WIN_LEN,
822 			    PROT_READ);
823 
824 			if (fptr) {
825 				if (!(fpsp = apic_find_fps_sig(fptr,
826 				    MPFPS_RAM_WIN_LEN)))
827 					psm_unmap_phys(fptr, MPFPS_RAM_WIN_LEN);
828 			}
829 		}
830 	}
831 	psm_unmap_phys(biosdatap, 0x20);
832 
833 	/* If still cannot find it, check the BIOS ROM space */
834 	if (!fpsp) {
835 		mapsize = MPFPS_ROM_WIN_LEN;
836 		fptr = psm_map_phys(MPFPS_ROM_WIN_START,
837 		    MPFPS_ROM_WIN_LEN, PROT_READ);
838 		if (fptr) {
839 			if (!(fpsp =
840 			    apic_find_fps_sig(fptr, MPFPS_ROM_WIN_LEN))) {
841 				psm_unmap_phys(fptr, MPFPS_ROM_WIN_LEN);
842 				return (retval);
843 			}
844 		}
845 	}
846 
847 	if (apic_checksum((caddr_t)fpsp, fpsp->mpfps_length * 16) != 0) {
848 		psm_unmap_phys(fptr, MPFPS_ROM_WIN_LEN);
849 		return (retval);
850 	}
851 
852 	apic_spec_rev = fpsp->mpfps_spec_rev;
853 	if ((apic_spec_rev != 04) && (apic_spec_rev != 01)) {
854 		psm_unmap_phys(fptr, MPFPS_ROM_WIN_LEN);
855 		return (retval);
856 	}
857 
858 	/* check IMCR is present or not */
859 	apic_imcrp = fpsp->mpfps_featinfo2 & MPFPS_FEATINFO2_IMCRP;
860 
861 	/* check default configuration (dual CPUs) */
862 	if ((apic_defconf = fpsp->mpfps_featinfo1) != 0) {
863 		psm_unmap_phys(fptr, mapsize);
864 		return (apic_handle_defconf());
865 	}
866 
867 	/* MP Configuration Table */
868 	mpct_addr = (uint32_t)(fpsp->mpfps_mpct_paddr);
869 
870 	psm_unmap_phys(fptr, mapsize); /* unmap floating ptr struct */
871 
872 	/*
873 	 * Map in enough memory for the MP Configuration Table Header.
874 	 * Use this table to read the total length of the BIOS data and
875 	 * map in all the info
876 	 */
877 	/*LINTED: pointer cast may result in improper alignment */
878 	hdrp = (struct apic_mp_cnf_hdr *)psm_map_phys(mpct_addr,
879 	    sizeof (struct apic_mp_cnf_hdr), PROT_READ);
880 	if (!hdrp)
881 		return (retval);
882 
883 	/* check mp configuration table signature PCMP */
884 	if (hdrp->mpcnf_sig != 0x504d4350) {
885 		psm_unmap_phys((caddr_t)hdrp, sizeof (struct apic_mp_cnf_hdr));
886 		return (retval);
887 	}
888 	mpct_size = (int)hdrp->mpcnf_tbl_length;
889 
890 	apic_set_pwroff_method_from_mpcnfhdr(hdrp);
891 
892 	psm_unmap_phys((caddr_t)hdrp, sizeof (struct apic_mp_cnf_hdr));
893 
894 	if ((retval == PSM_SUCCESS) && !apic_use_acpi_madt_only) {
895 		/* This is an ACPI machine No need for further checks */
896 		return (retval);
897 	}
898 
899 	/*
900 	 * Map in the entries for this machine, ie. Processor
901 	 * Entry Tables, Bus Entry Tables, etc.
902 	 * They are in fixed order following one another
903 	 */
904 	mpct = psm_map_phys(mpct_addr, mpct_size, PROT_READ);
905 	if (!mpct)
906 		return (retval);
907 
908 	if (apic_checksum(mpct, mpct_size) != 0)
909 		goto apic_fail1;
910 
911 
912 	/*LINTED: pointer cast may result in improper alignment */
913 	hdrp = (struct apic_mp_cnf_hdr *)mpct;
914 	/*LINTED: pointer cast may result in improper alignment */
915 	apicadr = (uint32_t *)psm_map_phys((uint32_t)hdrp->mpcnf_local_apic,
916 	    APIC_LOCAL_MEMLEN, PROT_READ | PROT_WRITE);
917 	if (!apicadr)
918 		goto apic_fail1;
919 
920 	/* Parse all information in the tables */
921 	bypass_cpu_and_ioapics_in_mptables = (retval == PSM_SUCCESS);
922 	if (apic_parse_mpct(mpct, bypass_cpu_and_ioapics_in_mptables) ==
923 	    PSM_SUCCESS)
924 		return (PSM_SUCCESS);
925 
926 	for (i = 0; i < apic_io_max; i++)
927 		psm_unmap_phys((caddr_t)apicioadr[i], APIC_IO_MEMLEN);
928 	if (apic_cpus)
929 		kmem_free(apic_cpus, sizeof (*apic_cpus) * apic_nproc);
930 	if (apicadr)
931 		psm_unmap_phys((caddr_t)apicadr, APIC_LOCAL_MEMLEN);
932 apic_fail1:
933 	psm_unmap_phys(mpct, mpct_size);
934 	return (retval);
935 }
936 
937 static void
938 apic_set_pwroff_method_from_mpcnfhdr(struct apic_mp_cnf_hdr *hdrp)
939 {
940 	int	i;
941 
942 	for (i = 0; i < (sizeof (apic_mps_ids) / sizeof (apic_mps_ids[0]));
943 	    i++) {
944 		if ((strncmp(hdrp->mpcnf_oem_str, apic_mps_ids[i].oem_id,
945 		    strlen(apic_mps_ids[i].oem_id)) == 0) &&
946 		    (strncmp(hdrp->mpcnf_prod_str, apic_mps_ids[i].prod_id,
947 		    strlen(apic_mps_ids[i].prod_id)) == 0)) {
948 
949 			apic_poweroff_method = apic_mps_ids[i].poweroff_method;
950 			break;
951 		}
952 	}
953 
954 	if (apic_debug_mps_id != 0) {
955 		cmn_err(CE_CONT, "pcplusmp: MPS OEM ID = '%c%c%c%c%c%c%c%c'"
956 		    "Product ID = '%c%c%c%c%c%c%c%c%c%c%c%c'\n",
957 		    hdrp->mpcnf_oem_str[0],
958 		    hdrp->mpcnf_oem_str[1],
959 		    hdrp->mpcnf_oem_str[2],
960 		    hdrp->mpcnf_oem_str[3],
961 		    hdrp->mpcnf_oem_str[4],
962 		    hdrp->mpcnf_oem_str[5],
963 		    hdrp->mpcnf_oem_str[6],
964 		    hdrp->mpcnf_oem_str[7],
965 		    hdrp->mpcnf_prod_str[0],
966 		    hdrp->mpcnf_prod_str[1],
967 		    hdrp->mpcnf_prod_str[2],
968 		    hdrp->mpcnf_prod_str[3],
969 		    hdrp->mpcnf_prod_str[4],
970 		    hdrp->mpcnf_prod_str[5],
971 		    hdrp->mpcnf_prod_str[6],
972 		    hdrp->mpcnf_prod_str[7],
973 		    hdrp->mpcnf_prod_str[8],
974 		    hdrp->mpcnf_prod_str[9],
975 		    hdrp->mpcnf_prod_str[10],
976 		    hdrp->mpcnf_prod_str[11]);
977 	}
978 }
979 
980 static int
981 acpi_probe(void)
982 {
983 	int			i, id, intmax, ver, index, rv;
984 	int			acpi_verboseflags = 0;
985 	int			madt_seen, madt_size;
986 	APIC_HEADER		*ap;
987 	MADT_PROCESSOR_APIC	*mpa;
988 	MADT_IO_APIC		*mia;
989 	MADT_IO_SAPIC		*misa;
990 	MADT_INTERRUPT_OVERRIDE	*mio;
991 	MADT_NMI_SOURCE		*mns;
992 	MADT_INTERRUPT_SOURCE	*mis;
993 	MADT_LOCAL_APIC_NMI	*mlan;
994 	MADT_ADDRESS_OVERRIDE	*mao;
995 	ACPI_OBJECT_LIST 	arglist;
996 	ACPI_OBJECT		arg;
997 	int			sci;
998 	iflag_t			sci_flags;
999 	volatile int32_t	*ioapic;
1000 	char			local_ids[NCPU];
1001 	char			proc_ids[NCPU];
1002 	uchar_t			hid;
1003 
1004 	if (!apic_use_acpi)
1005 		return (PSM_FAILURE);
1006 
1007 	if (AcpiGetFirmwareTable(APIC_SIG, 1, ACPI_LOGICAL_ADDRESSING,
1008 	    (ACPI_TABLE_HEADER **) &acpi_mapic_dtp) != AE_OK)
1009 		return (PSM_FAILURE);
1010 
1011 	apicadr = (uint32_t *)psm_map_phys(
1012 	    (uint32_t)acpi_mapic_dtp->LocalApicAddress,
1013 	    APIC_LOCAL_MEMLEN, PROT_READ | PROT_WRITE);
1014 	if (!apicadr)
1015 		return (PSM_FAILURE);
1016 
1017 	id = apicadr[APIC_LID_REG];
1018 	local_ids[0] = (uchar_t)(((uint_t)id) >> 24);
1019 	apic_nproc = index = 1;
1020 	CPUSET_ONLY(apic_cpumask, 0);
1021 	apic_io_max = 0;
1022 
1023 	ap = (APIC_HEADER *) (acpi_mapic_dtp + 1);
1024 	madt_size = acpi_mapic_dtp->Length;
1025 	madt_seen = sizeof (*acpi_mapic_dtp);
1026 
1027 	while (madt_seen < madt_size) {
1028 		switch (ap->Type) {
1029 		case APIC_PROCESSOR:
1030 			mpa = (MADT_PROCESSOR_APIC *) ap;
1031 			if (mpa->ProcessorEnabled) {
1032 				if (mpa->LocalApicId == local_ids[0])
1033 					proc_ids[0] = mpa->ProcessorId;
1034 				else if (apic_nproc < NCPU) {
1035 					local_ids[index] = mpa->LocalApicId;
1036 					proc_ids[index] = mpa->ProcessorId;
1037 					CPUSET_ADD(apic_cpumask, index);
1038 					index++;
1039 					apic_nproc++;
1040 				} else
1041 					cmn_err(CE_WARN, "pcplusmp: exceeded "
1042 					    "maximum no. of CPUs (= %d)", NCPU);
1043 			}
1044 			break;
1045 
1046 		case APIC_IO:
1047 			mia = (MADT_IO_APIC *) ap;
1048 			if (apic_io_max < MAX_IO_APIC) {
1049 				apic_io_id[apic_io_max] = mia->IoApicId;
1050 				apic_io_vectbase[apic_io_max] =
1051 				    mia->Interrupt;
1052 				ioapic = apicioadr[apic_io_max] =
1053 				    (int32_t *)psm_map_phys(
1054 				    (uint32_t)mia->Address,
1055 				    APIC_IO_MEMLEN, PROT_READ | PROT_WRITE);
1056 				if (!ioapic)
1057 					goto cleanup;
1058 				apic_io_max++;
1059 			}
1060 			break;
1061 
1062 		case APIC_XRUPT_OVERRIDE:
1063 			mio = (MADT_INTERRUPT_OVERRIDE *) ap;
1064 			if (acpi_isop == NULL)
1065 				acpi_isop = mio;
1066 			acpi_iso_cnt++;
1067 			break;
1068 
1069 		case APIC_NMI:
1070 			/* UNIMPLEMENTED */
1071 			mns = (MADT_NMI_SOURCE *) ap;
1072 			if (acpi_nmi_sp == NULL)
1073 				acpi_nmi_sp = mns;
1074 			acpi_nmi_scnt++;
1075 
1076 			cmn_err(CE_NOTE, "!apic: nmi source: %d %d %d\n",
1077 				mns->Interrupt, mns->Polarity,
1078 				mns->TriggerMode);
1079 			break;
1080 
1081 		case APIC_LOCAL_NMI:
1082 			/* UNIMPLEMENTED */
1083 			mlan = (MADT_LOCAL_APIC_NMI *) ap;
1084 			if (acpi_nmi_cp == NULL)
1085 				acpi_nmi_cp = mlan;
1086 			acpi_nmi_ccnt++;
1087 
1088 			cmn_err(CE_NOTE, "!apic: local nmi: %d %d %d %d\n",
1089 				mlan->ProcessorId, mlan->Polarity,
1090 				mlan->TriggerMode, mlan->Lint);
1091 			break;
1092 
1093 		case APIC_ADDRESS_OVERRIDE:
1094 			/* UNIMPLEMENTED */
1095 			mao = (MADT_ADDRESS_OVERRIDE *) ap;
1096 			cmn_err(CE_NOTE, "!apic: address override: %lx\n",
1097 				(long)mao->Address);
1098 			break;
1099 
1100 		case APIC_IO_SAPIC:
1101 			/* UNIMPLEMENTED */
1102 			misa = (MADT_IO_SAPIC *) ap;
1103 
1104 			cmn_err(CE_NOTE, "!apic: io sapic: %d %d %lx\n",
1105 				misa->IoSapicId, misa->InterruptBase,
1106 				(long)misa->Address);
1107 			break;
1108 
1109 		case APIC_XRUPT_SOURCE:
1110 			/* UNIMPLEMENTED */
1111 			mis = (MADT_INTERRUPT_SOURCE *) ap;
1112 
1113 			cmn_err(CE_NOTE,
1114 				"!apic: irq source: %d %d %d %d %d %d %d\n",
1115 				mis->ProcessorId, mis->ProcessorEid,
1116 				mis->Interrupt, mis->Polarity,
1117 				mis->TriggerMode, mis->InterruptType,
1118 				mis->IoSapicVector);
1119 			break;
1120 		case APIC_RESERVED:
1121 		default:
1122 			break;	/* ignore unknown items as per ACPI spec */
1123 		}
1124 
1125 		/* advance to next entry */
1126 		madt_seen += ap->Length;
1127 		ap = (APIC_HEADER *)(((char *)ap) + ap->Length);
1128 	}
1129 
1130 	if ((apic_cpus = kmem_zalloc(sizeof (*apic_cpus) * apic_nproc,
1131 	    KM_NOSLEEP)) == NULL)
1132 		goto cleanup;
1133 
1134 	/*
1135 	 * ACPI doesn't provide the local apic ver, get it directly from the
1136 	 * local apic
1137 	 */
1138 	ver = apicadr[APIC_VERS_REG];
1139 	for (i = 0; i < apic_nproc; i++) {
1140 		apic_cpus[i].aci_local_id = local_ids[i];
1141 		apic_cpus[i].aci_local_ver = (uchar_t)(ver & 0xFF);
1142 	}
1143 	for (i = 0; i < apic_io_max; i++) {
1144 		ioapic = apicioadr[i];
1145 
1146 		/*
1147 		 * need to check Sitka on the following acpi problem
1148 		 * On the Sitka, the ioapic's apic_id field isn't reporting
1149 		 * the actual io apic id. We have reported this problem
1150 		 * to Intel. Until they fix the problem, we will get the
1151 		 * actual id directly from the ioapic.
1152 		 */
1153 		ioapic[APIC_IO_REG] = APIC_ID_CMD;
1154 		id = ioapic[APIC_IO_DATA];
1155 		hid = (uchar_t)(((uint_t)id) >> 24);
1156 
1157 		if (hid != apic_io_id[i]) {
1158 			if (apic_io_id[i] == 0)
1159 				apic_io_id[i] = hid;
1160 			else { /* set ioapic id to whatever reported by ACPI */
1161 				id = ((int32_t)apic_io_id[i]) << 24;
1162 				ioapic[APIC_IO_REG] = APIC_ID_CMD;
1163 				ioapic[APIC_IO_DATA] = id;
1164 			}
1165 		}
1166 		ioapic[APIC_IO_REG] = APIC_VERS_CMD;
1167 		ver = ioapic[APIC_IO_DATA];
1168 		apic_io_ver[i] = (uchar_t)(ver & 0xff);
1169 		intmax = (ver >> 16) & 0xff;
1170 		apic_io_vectend[i] = apic_io_vectbase[i] + intmax;
1171 		if (apic_first_avail_irq <= apic_io_vectend[i])
1172 			apic_first_avail_irq = apic_io_vectend[i] + 1;
1173 	}
1174 
1175 
1176 	/*
1177 	 * Process SCI configuration here
1178 	 * An error may be returned here if
1179 	 * acpi-user-options specifies legacy mode
1180 	 * (no SCI, no ACPI mode)
1181 	 */
1182 	if (acpica_get_sci(&sci, &sci_flags) != AE_OK)
1183 		sci = -1;
1184 
1185 	/*
1186 	 * Now call acpi_init() to generate namespaces
1187 	 * If this fails, we don't attempt to use ACPI
1188 	 * even if we were able to get a MADT above
1189 	 */
1190 	if (acpica_init() != AE_OK)
1191 		goto cleanup;
1192 
1193 	/*
1194 	 * Squirrel away the SCI and flags for later on
1195 	 * in apic_picinit() when we're ready
1196 	 */
1197 	apic_sci_vect = sci;
1198 	apic_sci_flags = sci_flags;
1199 
1200 	if (apic_verbose & APIC_VERBOSE_IRQ_FLAG)
1201 		acpi_verboseflags |= PSM_VERBOSE_IRQ_FLAG;
1202 
1203 	if (apic_verbose & APIC_VERBOSE_POWEROFF_FLAG)
1204 		acpi_verboseflags |= PSM_VERBOSE_POWEROFF_FLAG;
1205 
1206 	if (apic_verbose & APIC_VERBOSE_POWEROFF_PAUSE_FLAG)
1207 		acpi_verboseflags |= PSM_VERBOSE_POWEROFF_PAUSE_FLAG;
1208 
1209 	if (acpi_psm_init(apic_psm_info.p_mach_idstring, acpi_verboseflags) ==
1210 	    ACPI_PSM_FAILURE)
1211 		goto cleanup;
1212 
1213 	/* Enable ACPI APIC interrupt routing */
1214 	arglist.Count = 1;
1215 	arglist.Pointer = &arg;
1216 	arg.Type = ACPI_TYPE_INTEGER;
1217 	arg.Integer.Value = ACPI_APIC_MODE;	/* 1 */
1218 	rv = AcpiEvaluateObject(NULL, "\\_PIC", &arglist, NULL);
1219 	if (rv == AE_OK) {
1220 		build_reserved_irqlist((uchar_t *)apic_reserved_irqlist);
1221 		apic_enable_acpi = 1;
1222 		if (apic_use_acpi_madt_only) {
1223 			cmn_err(CE_CONT,
1224 			    "?Using ACPI for CPU/IOAPIC information ONLY\n");
1225 		}
1226 		return (PSM_SUCCESS);
1227 	}
1228 	/* if setting APIC mode failed above, we fall through to cleanup */
1229 
1230 cleanup:
1231 	if (apicadr != NULL) {
1232 		psm_unmap_phys((caddr_t)apicadr, APIC_LOCAL_MEMLEN);
1233 		apicadr = NULL;
1234 	}
1235 	apic_nproc = 0;
1236 	for (i = 0; i < apic_io_max; i++) {
1237 		psm_unmap_phys((caddr_t)apicioadr[i], APIC_IO_MEMLEN);
1238 		apicioadr[i] = NULL;
1239 	}
1240 	apic_io_max = 0;
1241 	acpi_isop = NULL;
1242 	acpi_iso_cnt = 0;
1243 	acpi_nmi_sp = NULL;
1244 	acpi_nmi_scnt = 0;
1245 	acpi_nmi_cp = NULL;
1246 	acpi_nmi_ccnt = 0;
1247 	return (PSM_FAILURE);
1248 }
1249 
1250 /*
1251  * Handle default configuration. Fill in reqd global variables & tables
1252  * Fill all details as MP table does not give any more info
1253  */
1254 static int
1255 apic_handle_defconf()
1256 {
1257 	uint_t	lid;
1258 
1259 	/*LINTED: pointer cast may result in improper alignment */
1260 	apicioadr[0] = (int32_t *)psm_map_phys(APIC_IO_ADDR,
1261 	    APIC_IO_MEMLEN, PROT_READ | PROT_WRITE);
1262 	/*LINTED: pointer cast may result in improper alignment */
1263 	apicadr = (uint32_t *)psm_map_phys(APIC_LOCAL_ADDR,
1264 	    APIC_LOCAL_MEMLEN, PROT_READ | PROT_WRITE);
1265 	apic_cpus = (apic_cpus_info_t *)
1266 	    kmem_zalloc(sizeof (*apic_cpus) * 2, KM_NOSLEEP);
1267 	if ((!apicadr) || (!apicioadr[0]) || (!apic_cpus))
1268 		goto apic_handle_defconf_fail;
1269 	CPUSET_ONLY(apic_cpumask, 0);
1270 	CPUSET_ADD(apic_cpumask, 1);
1271 	apic_nproc = 2;
1272 	lid = apicadr[APIC_LID_REG];
1273 	apic_cpus[0].aci_local_id = (uchar_t)(lid >> APIC_ID_BIT_OFFSET);
1274 	/*
1275 	 * According to the PC+MP spec 1.1, the local ids
1276 	 * for the default configuration has to be 0 or 1
1277 	 */
1278 	if (apic_cpus[0].aci_local_id == 1)
1279 		apic_cpus[1].aci_local_id = 0;
1280 	else if (apic_cpus[0].aci_local_id == 0)
1281 		apic_cpus[1].aci_local_id = 1;
1282 	else
1283 		goto apic_handle_defconf_fail;
1284 
1285 	apic_io_id[0] = 2;
1286 	apic_io_max = 1;
1287 	if (apic_defconf >= 5) {
1288 		apic_cpus[0].aci_local_ver = APIC_INTEGRATED_VERS;
1289 		apic_cpus[1].aci_local_ver = APIC_INTEGRATED_VERS;
1290 		apic_io_ver[0] = APIC_INTEGRATED_VERS;
1291 	} else {
1292 		apic_cpus[0].aci_local_ver = 0;		/* 82489 DX */
1293 		apic_cpus[1].aci_local_ver = 0;
1294 		apic_io_ver[0] = 0;
1295 	}
1296 	if (apic_defconf == 2 || apic_defconf == 3 || apic_defconf == 6)
1297 		eisa_level_intr_mask = (inb(EISA_LEVEL_CNTL + 1) << 8) |
1298 		    inb(EISA_LEVEL_CNTL) | ((uint_t)INT32_MAX + 1);
1299 	return (PSM_SUCCESS);
1300 
1301 apic_handle_defconf_fail:
1302 	if (apic_cpus)
1303 		kmem_free(apic_cpus, sizeof (*apic_cpus) * 2);
1304 	if (apicadr)
1305 		psm_unmap_phys((caddr_t)apicadr, APIC_LOCAL_MEMLEN);
1306 	if (apicioadr[0])
1307 		psm_unmap_phys((caddr_t)apicioadr[0], APIC_IO_MEMLEN);
1308 	return (PSM_FAILURE);
1309 }
1310 
1311 /* Parse the entries in MP configuration table and collect info that we need */
1312 static int
1313 apic_parse_mpct(caddr_t mpct, int bypass_cpus_and_ioapics)
1314 {
1315 	struct	apic_procent	*procp;
1316 	struct	apic_bus	*busp;
1317 	struct	apic_io_entry	*ioapicp;
1318 	struct	apic_io_intr	*intrp;
1319 	volatile int32_t	*ioapic;
1320 	uint_t	lid;
1321 	int	id;
1322 	uchar_t hid;
1323 
1324 	/*LINTED: pointer cast may result in improper alignment */
1325 	procp = (struct apic_procent *)(mpct + sizeof (struct apic_mp_cnf_hdr));
1326 
1327 	/* No need to count cpu entries if we won't use them */
1328 	if (!bypass_cpus_and_ioapics) {
1329 
1330 		/* Find max # of CPUS and allocate structure accordingly */
1331 		apic_nproc = 0;
1332 		CPUSET_ZERO(apic_cpumask);
1333 		while (procp->proc_entry == APIC_CPU_ENTRY) {
1334 			if (procp->proc_cpuflags & CPUFLAGS_EN) {
1335 				if (apic_nproc < NCPU)
1336 					CPUSET_ADD(apic_cpumask, apic_nproc);
1337 				apic_nproc++;
1338 			}
1339 			procp++;
1340 		}
1341 		if (apic_nproc > NCPU)
1342 			cmn_err(CE_WARN, "pcplusmp: exceeded "
1343 			    "maximum no. of CPUs (= %d)", NCPU);
1344 		if (!apic_nproc || !(apic_cpus = (apic_cpus_info_t *)
1345 		    kmem_zalloc(sizeof (*apic_cpus)*apic_nproc, KM_NOSLEEP)))
1346 			return (PSM_FAILURE);
1347 	}
1348 
1349 	/*LINTED: pointer cast may result in improper alignment */
1350 	procp = (struct apic_procent *)(mpct + sizeof (struct apic_mp_cnf_hdr));
1351 
1352 	/*
1353 	 * start with index 1 as 0 needs to be filled in with Boot CPU, but
1354 	 * if we're bypassing this information, it has already been filled
1355 	 * in by acpi_probe(), so don't overwrite it.
1356 	 */
1357 	if (!bypass_cpus_and_ioapics)
1358 		apic_nproc = 1;
1359 
1360 	while (procp->proc_entry == APIC_CPU_ENTRY) {
1361 		/* check whether the cpu exists or not */
1362 		if (!bypass_cpus_and_ioapics &&
1363 		    procp->proc_cpuflags & CPUFLAGS_EN) {
1364 			if (procp->proc_cpuflags & CPUFLAGS_BP) { /* Boot CPU */
1365 				lid = apicadr[APIC_LID_REG];
1366 				apic_cpus[0].aci_local_id = procp->proc_apicid;
1367 				if (apic_cpus[0].aci_local_id !=
1368 				    (uchar_t)(lid >> APIC_ID_BIT_OFFSET)) {
1369 					return (PSM_FAILURE);
1370 				}
1371 				apic_cpus[0].aci_local_ver =
1372 				    procp->proc_version;
1373 			} else {
1374 
1375 				apic_cpus[apic_nproc].aci_local_id =
1376 				    procp->proc_apicid;
1377 				apic_cpus[apic_nproc].aci_local_ver =
1378 				    procp->proc_version;
1379 				apic_nproc++;
1380 
1381 			}
1382 		}
1383 		procp++;
1384 	}
1385 
1386 	/*
1387 	 * Save start of bus entries for later use.
1388 	 * Get EISA level cntrl if EISA bus is present.
1389 	 * Also get the CPI bus id for single CPI bus case
1390 	 */
1391 	apic_busp = busp = (struct apic_bus *)procp;
1392 	while (busp->bus_entry == APIC_BUS_ENTRY) {
1393 		lid = apic_find_bus_type((char *)&busp->bus_str1);
1394 		if (lid	== BUS_EISA) {
1395 			eisa_level_intr_mask = (inb(EISA_LEVEL_CNTL + 1) << 8) |
1396 			    inb(EISA_LEVEL_CNTL) | ((uint_t)INT32_MAX + 1);
1397 		} else if (lid == BUS_PCI) {
1398 			/*
1399 			 * apic_single_pci_busid will be used only if
1400 			 * apic_pic_bus_total is equal to 1
1401 			 */
1402 			apic_pci_bus_total++;
1403 			apic_single_pci_busid = busp->bus_id;
1404 		}
1405 		busp++;
1406 	}
1407 
1408 	ioapicp = (struct apic_io_entry *)busp;
1409 
1410 	if (!bypass_cpus_and_ioapics)
1411 		apic_io_max = 0;
1412 	do {
1413 		if (!bypass_cpus_and_ioapics && apic_io_max < MAX_IO_APIC) {
1414 			if (ioapicp->io_flags & IOAPIC_FLAGS_EN) {
1415 				apic_io_id[apic_io_max] = ioapicp->io_apicid;
1416 				apic_io_ver[apic_io_max] = ioapicp->io_version;
1417 		/*LINTED: pointer cast may result in improper alignment */
1418 				apicioadr[apic_io_max] =
1419 				    (int32_t *)psm_map_phys(
1420 				    (uint32_t)ioapicp->io_apic_addr,
1421 				    APIC_IO_MEMLEN, PROT_READ | PROT_WRITE);
1422 
1423 				if (!apicioadr[apic_io_max])
1424 					return (PSM_FAILURE);
1425 
1426 				ioapic = apicioadr[apic_io_max];
1427 				ioapic[APIC_IO_REG] = APIC_ID_CMD;
1428 				id = ioapic[APIC_IO_DATA];
1429 				hid = (uchar_t)(((uint_t)id) >> 24);
1430 
1431 				if (hid != apic_io_id[apic_io_max]) {
1432 					if (apic_io_id[apic_io_max] == 0)
1433 						apic_io_id[apic_io_max] = hid;
1434 					else {
1435 						/*
1436 						 * set ioapic id to whatever
1437 						 * reported by MPS
1438 						 *
1439 						 * may not need to set index
1440 						 * again ???
1441 						 * take it out and try
1442 						 */
1443 
1444 						id = ((int32_t)
1445 						    apic_io_id[apic_io_max]) <<
1446 						    24;
1447 
1448 						ioapic[APIC_IO_REG] =
1449 						    APIC_ID_CMD;
1450 
1451 						ioapic[APIC_IO_DATA] = id;
1452 
1453 					}
1454 				}
1455 				apic_io_max++;
1456 			}
1457 		}
1458 		ioapicp++;
1459 	} while (ioapicp->io_entry == APIC_IO_ENTRY);
1460 
1461 	apic_io_intrp = (struct apic_io_intr *)ioapicp;
1462 
1463 	intrp = apic_io_intrp;
1464 	while (intrp->intr_entry == APIC_IO_INTR_ENTRY) {
1465 		if ((intrp->intr_irq > APIC_MAX_ISA_IRQ) ||
1466 		    (apic_find_bus(intrp->intr_busid) == BUS_PCI)) {
1467 			apic_irq_translate = 1;
1468 			break;
1469 		}
1470 		intrp++;
1471 	}
1472 
1473 	return (PSM_SUCCESS);
1474 }
1475 
1476 boolean_t
1477 apic_cpu_in_range(int cpu)
1478 {
1479 	return ((cpu & ~IRQ_USER_BOUND) < apic_nproc);
1480 }
1481 
1482 static struct apic_mpfps_hdr *
1483 apic_find_fps_sig(caddr_t cptr, int len)
1484 {
1485 	int	i;
1486 
1487 	/* Look for the pattern "_MP_" */
1488 	for (i = 0; i < len; i += 16) {
1489 		if ((*(cptr+i) == '_') &&
1490 		    (*(cptr+i+1) == 'M') &&
1491 		    (*(cptr+i+2) == 'P') &&
1492 		    (*(cptr+i+3) == '_'))
1493 		    /*LINTED: pointer cast may result in improper alignment */
1494 			return ((struct apic_mpfps_hdr *)(cptr + i));
1495 	}
1496 	return (NULL);
1497 }
1498 
1499 static int
1500 apic_checksum(caddr_t bptr, int len)
1501 {
1502 	int	i;
1503 	uchar_t	cksum;
1504 
1505 	cksum = 0;
1506 	for (i = 0; i < len; i++)
1507 		cksum += *bptr++;
1508 	return ((int)cksum);
1509 }
1510 
1511 
1512 /*
1513  * Initialise vector->ipl and ipl->pri arrays. level_intr and irqtable
1514  * are also set to NULL. vector->irq is set to a value which cannot map
1515  * to a real irq to show that it is free.
1516  */
1517 void
1518 apic_init()
1519 {
1520 	int	i;
1521 	int	*iptr;
1522 
1523 	int	j = 1;
1524 	apic_ipltopri[0] = APIC_VECTOR_PER_IPL; /* leave 0 for idle */
1525 	for (i = 0; i < (APIC_AVAIL_VECTOR / APIC_VECTOR_PER_IPL); i++) {
1526 		if ((i < ((APIC_AVAIL_VECTOR / APIC_VECTOR_PER_IPL) - 1)) &&
1527 		    (apic_vectortoipl[i + 1] == apic_vectortoipl[i]))
1528 			/* get to highest vector at the same ipl */
1529 			continue;
1530 		for (; j <= apic_vectortoipl[i]; j++) {
1531 			apic_ipltopri[j] = (i << APIC_IPL_SHIFT) +
1532 			    APIC_BASE_VECT;
1533 		}
1534 	}
1535 	for (; j < MAXIPL + 1; j++)
1536 		/* fill up any empty ipltopri slots */
1537 		apic_ipltopri[j] = (i << APIC_IPL_SHIFT) + APIC_BASE_VECT;
1538 
1539 	/* cpu 0 is always up */
1540 	apic_cpus[0].aci_status = APIC_CPU_ONLINE | APIC_CPU_INTR_ENABLE;
1541 
1542 	iptr = (int *)&apic_irq_table[0];
1543 	for (i = 0; i <= APIC_MAX_VECTOR; i++) {
1544 		apic_level_intr[i] = 0;
1545 		*iptr++ = NULL;
1546 		apic_vector_to_irq[i] = APIC_RESV_IRQ;
1547 
1548 		/* These *must* be initted to B_TRUE! */
1549 		apic_reprogram_info[i].done = B_TRUE;
1550 		apic_reprogram_info[i].irqp = NULL;
1551 		apic_reprogram_info[i].tries = 0;
1552 		apic_reprogram_info[i].bindcpu = 0;
1553 	}
1554 
1555 	/*
1556 	 * Allocate a dummy irq table entry for the reserved entry.
1557 	 * This takes care of the race between removing an irq and
1558 	 * clock detecting a CPU in that irq during interrupt load
1559 	 * sampling.
1560 	 */
1561 	apic_irq_table[APIC_RESV_IRQ] =
1562 	    kmem_zalloc(sizeof (apic_irq_t), KM_NOSLEEP);
1563 
1564 	mutex_init(&airq_mutex, NULL, MUTEX_DEFAULT, NULL);
1565 #if defined(__amd64)
1566 	/*
1567 	 * Make cpu-specific interrupt info point to cr8pri vector
1568 	 */
1569 	for (i = 0; i <= MAXIPL; i++)
1570 		apic_cr8pri[i] = apic_ipltopri[i] >> APIC_IPL_SHIFT;
1571 	CPU->cpu_pri_data = apic_cr8pri;
1572 	intpri_use_cr8 = 1;
1573 #endif	/* __amd64 */
1574 }
1575 
1576 /*
1577  * handler for APIC Error interrupt. Just print a warning and continue
1578  */
1579 static int
1580 apic_error_intr()
1581 {
1582 	uint_t	error0, error1, error;
1583 	uint_t	i;
1584 
1585 	/*
1586 	 * We need to write before read as per 7.4.17 of system prog manual.
1587 	 * We do both and or the results to be safe
1588 	 */
1589 	error0 = apicadr[APIC_ERROR_STATUS];
1590 	apicadr[APIC_ERROR_STATUS] = 0;
1591 	error1 = apicadr[APIC_ERROR_STATUS];
1592 	error = error0 | error1;
1593 
1594 	/*
1595 	 * Clear the APIC error status (do this on all cpus that enter here)
1596 	 * (two writes are required due to the semantics of accessing the
1597 	 * error status register.)
1598 	 */
1599 	apicadr[APIC_ERROR_STATUS] = 0;
1600 	apicadr[APIC_ERROR_STATUS] = 0;
1601 
1602 	/*
1603 	 * Prevent more than 1 CPU from handling error interrupt causing
1604 	 * double printing (interleave of characters from multiple
1605 	 * CPU's when using prom_printf)
1606 	 */
1607 	if (lock_try(&apic_error_lock) == 0)
1608 		return (error ? DDI_INTR_CLAIMED : DDI_INTR_UNCLAIMED);
1609 	if (error) {
1610 #if	DEBUG
1611 		if (apic_debug)
1612 			debug_enter("pcplusmp: APIC Error interrupt received");
1613 #endif /* DEBUG */
1614 		if (apic_panic_on_apic_error)
1615 			cmn_err(CE_PANIC,
1616 			    "APIC Error interrupt on CPU %d. Status = %x\n",
1617 			    psm_get_cpu_id(), error);
1618 		else {
1619 			if ((error & ~APIC_CS_ERRORS) == 0) {
1620 				/* cksum error only */
1621 				apic_error |= APIC_ERR_APIC_ERROR;
1622 				apic_apic_error |= error;
1623 				apic_num_apic_errors++;
1624 				apic_num_cksum_errors++;
1625 			} else {
1626 				/*
1627 				 * prom_printf is the best shot we have of
1628 				 * something which is problem free from
1629 				 * high level/NMI type of interrupts
1630 				 */
1631 				prom_printf("APIC Error interrupt on CPU %d. "
1632 				    "Status 0 = %x, Status 1 = %x\n",
1633 				    psm_get_cpu_id(), error0, error1);
1634 				apic_error |= APIC_ERR_APIC_ERROR;
1635 				apic_apic_error |= error;
1636 				apic_num_apic_errors++;
1637 				for (i = 0; i < apic_error_display_delay; i++) {
1638 					tenmicrosec();
1639 				}
1640 				/*
1641 				 * provide more delay next time limited to
1642 				 * roughly 1 clock tick time
1643 				 */
1644 				if (apic_error_display_delay < 500)
1645 					apic_error_display_delay *= 2;
1646 			}
1647 		}
1648 		lock_clear(&apic_error_lock);
1649 		return (DDI_INTR_CLAIMED);
1650 	} else {
1651 		lock_clear(&apic_error_lock);
1652 		return (DDI_INTR_UNCLAIMED);
1653 	}
1654 	/* NOTREACHED */
1655 }
1656 
1657 /*
1658  * Turn off the mask bit in the performance counter Local Vector Table entry.
1659  */
1660 static void
1661 apic_cpcovf_mask_clear(void)
1662 {
1663 	apicadr[APIC_PCINT_VECT] &= ~APIC_LVT_MASK;
1664 }
1665 
1666 static void
1667 apic_init_intr()
1668 {
1669 	processorid_t	cpun = psm_get_cpu_id();
1670 
1671 #if defined(__amd64)
1672 	setcr8((ulong_t)(APIC_MASK_ALL >> APIC_IPL_SHIFT));
1673 #else
1674 	apicadr[APIC_TASK_REG] = APIC_MASK_ALL;
1675 #endif
1676 
1677 	if (apic_flat_model)
1678 		apicadr[APIC_FORMAT_REG] = APIC_FLAT_MODEL;
1679 	else
1680 		apicadr[APIC_FORMAT_REG] = APIC_CLUSTER_MODEL;
1681 	apicadr[APIC_DEST_REG] = AV_HIGH_ORDER >> cpun;
1682 
1683 	/* need to enable APIC before unmasking NMI */
1684 	apicadr[APIC_SPUR_INT_REG] = AV_UNIT_ENABLE | APIC_SPUR_INTR;
1685 
1686 	apicadr[APIC_LOCAL_TIMER] = AV_MASK;
1687 	apicadr[APIC_INT_VECT0]	= AV_MASK;	/* local intr reg 0 */
1688 	apicadr[APIC_INT_VECT1] = AV_NMI;	/* enable NMI */
1689 
1690 	if (apic_cpus[cpun].aci_local_ver < APIC_INTEGRATED_VERS)
1691 		return;
1692 
1693 	/* Enable performance counter overflow interrupt */
1694 
1695 	if ((x86_feature & X86_MSR) != X86_MSR)
1696 		apic_enable_cpcovf_intr = 0;
1697 	if (apic_enable_cpcovf_intr) {
1698 		if (apic_cpcovf_vect == 0) {
1699 			int ipl = APIC_PCINT_IPL;
1700 			int irq = apic_get_ipivect(ipl, -1);
1701 
1702 			ASSERT(irq != -1);
1703 			apic_cpcovf_vect = apic_irq_table[irq]->airq_vector;
1704 			ASSERT(apic_cpcovf_vect);
1705 			(void) add_avintr(NULL, ipl,
1706 			    (avfunc)kcpc_hw_overflow_intr,
1707 			    "apic pcint", irq, NULL, NULL, NULL, NULL);
1708 			kcpc_hw_overflow_intr_installed = 1;
1709 			kcpc_hw_enable_cpc_intr = apic_cpcovf_mask_clear;
1710 		}
1711 		apicadr[APIC_PCINT_VECT] = apic_cpcovf_vect;
1712 	}
1713 
1714 	/* Enable error interrupt */
1715 
1716 	if (apic_enable_error_intr) {
1717 		if (apic_errvect == 0) {
1718 			int ipl = 0xf;	/* get highest priority intr */
1719 			int irq = apic_get_ipivect(ipl, -1);
1720 
1721 			ASSERT(irq != -1);
1722 			apic_errvect = apic_irq_table[irq]->airq_vector;
1723 			ASSERT(apic_errvect);
1724 			/*
1725 			 * Not PSMI compliant, but we are going to merge
1726 			 * with ON anyway
1727 			 */
1728 			(void) add_avintr((void *)NULL, ipl,
1729 			    (avfunc)apic_error_intr, "apic error intr",
1730 			    irq, NULL, NULL, NULL, NULL);
1731 		}
1732 		apicadr[APIC_ERR_VECT] = apic_errvect;
1733 		apicadr[APIC_ERROR_STATUS] = 0;
1734 		apicadr[APIC_ERROR_STATUS] = 0;
1735 	}
1736 }
1737 
1738 static void
1739 apic_disable_local_apic()
1740 {
1741 	apicadr[APIC_TASK_REG] = APIC_MASK_ALL;
1742 	apicadr[APIC_LOCAL_TIMER] = AV_MASK;
1743 	apicadr[APIC_INT_VECT0] = AV_MASK;	/* local intr reg 0 */
1744 	apicadr[APIC_INT_VECT1] = AV_MASK;	/* disable NMI */
1745 	apicadr[APIC_ERR_VECT] = AV_MASK;	/* and error interrupt */
1746 	apicadr[APIC_PCINT_VECT] = AV_MASK;	/* and perf counter intr */
1747 	apicadr[APIC_SPUR_INT_REG] = APIC_SPUR_INTR;
1748 }
1749 
1750 static void
1751 apic_picinit(void)
1752 {
1753 	int i, j, iflag;
1754 	uint_t isr;
1755 	volatile int32_t *ioapic;
1756 	apic_irq_t	*irqptr;
1757 	struct intrspec ispec;
1758 
1759 	/*
1760 	 * On UniSys Model 6520, the BIOS leaves vector 0x20 isr
1761 	 * bit on without clearing it with EOI.  Since softint
1762 	 * uses vector 0x20 to interrupt itself, so softint will
1763 	 * not work on this machine.  In order to fix this problem
1764 	 * a check is made to verify all the isr bits are clear.
1765 	 * If not, EOIs are issued to clear the bits.
1766 	 */
1767 	for (i = 7; i >= 1; i--) {
1768 		if ((isr = apicadr[APIC_ISR_REG + (i * 4)]) != 0)
1769 			for (j = 0; ((j < 32) && (isr != 0)); j++)
1770 				if (isr & (1 << j)) {
1771 					apicadr[APIC_EOI_REG] = 0;
1772 					isr &= ~(1 << j);
1773 					apic_error |= APIC_ERR_BOOT_EOI;
1774 				}
1775 	}
1776 
1777 	/* set a flag so we know we have run apic_picinit() */
1778 	apic_flag = 1;
1779 	LOCK_INIT_CLEAR(&apic_gethrtime_lock);
1780 	LOCK_INIT_CLEAR(&apic_ioapic_lock);
1781 	LOCK_INIT_CLEAR(&apic_revector_lock);
1782 	LOCK_INIT_CLEAR(&apic_defer_reprogram_lock);
1783 	LOCK_INIT_CLEAR(&apic_error_lock);
1784 
1785 	picsetup();	 /* initialise the 8259 */
1786 
1787 	/* add nmi handler - least priority nmi handler */
1788 	LOCK_INIT_CLEAR(&apic_nmi_lock);
1789 
1790 	if (!psm_add_nmintr(0, (avfunc) apic_nmi_intr,
1791 	    "pcplusmp NMI handler", (caddr_t)NULL))
1792 		cmn_err(CE_WARN, "pcplusmp: Unable to add nmi handler");
1793 
1794 	apic_init_intr();
1795 
1796 	/* enable apic mode if imcr present */
1797 	if (apic_imcrp) {
1798 		outb(APIC_IMCR_P1, (uchar_t)APIC_IMCR_SELECT);
1799 		outb(APIC_IMCR_P2, (uchar_t)APIC_IMCR_APIC);
1800 	}
1801 
1802 	/* mask interrupt vectors					*/
1803 	for (j = 0; j < apic_io_max; j++) {
1804 		int intin_max;
1805 		ioapic = apicioadr[j];
1806 		ioapic[APIC_IO_REG] = APIC_VERS_CMD;
1807 		/* Bits 23-16 define the maximum redirection entries */
1808 		intin_max = (ioapic[APIC_IO_DATA] >> 16) & 0xff;
1809 		for (i = 0; i < intin_max; i++) {
1810 			ioapic[APIC_IO_REG] = APIC_RDT_CMD + 2 * i;
1811 			ioapic[APIC_IO_DATA] = AV_MASK;
1812 		}
1813 	}
1814 
1815 	/*
1816 	 * Hack alert: deal with ACPI SCI interrupt chicken/egg here
1817 	 */
1818 	if (apic_sci_vect > 0) {
1819 		/*
1820 		 * acpica has already done add_avintr(); we just
1821 		 * to finish the job by mimicing translate_irq()
1822 		 *
1823 		 * Fake up an intrspec and setup the tables
1824 		 */
1825 		ispec.intrspec_vec = apic_sci_vect;
1826 		ispec.intrspec_pri = SCI_IPL;
1827 
1828 		if (apic_setup_irq_table(NULL, apic_sci_vect, NULL,
1829 		    &ispec, &apic_sci_flags, DDI_INTR_TYPE_FIXED) < 0) {
1830 			cmn_err(CE_WARN, "!apic: SCI setup failed");
1831 			return;
1832 		}
1833 		irqptr = apic_irq_table[apic_sci_vect];
1834 
1835 		iflag = intr_clear();
1836 		lock_set(&apic_ioapic_lock);
1837 
1838 		/* Program I/O APIC */
1839 		(void) apic_setup_io_intr(irqptr, apic_sci_vect, B_FALSE);
1840 
1841 		lock_clear(&apic_ioapic_lock);
1842 		intr_restore(iflag);
1843 
1844 		irqptr->airq_share++;
1845 	}
1846 }
1847 
1848 
1849 static void
1850 apic_cpu_start(processorid_t cpun, caddr_t rm_code)
1851 {
1852 	int		loop_count;
1853 	uint32_t	vector;
1854 	uint_t		cpu_id, iflag;
1855 
1856 	cpu_id = apic_cpus[cpun].aci_local_id;
1857 
1858 	apic_cmos_ssb_set = 1;
1859 
1860 	/*
1861 	 * Interrupts on BSP cpu will be disabled during these startup
1862 	 * steps in order to avoid unwanted side effects from
1863 	 * executing interrupt handlers on a problematic BIOS.
1864 	 */
1865 
1866 	iflag = intr_clear();
1867 	outb(CMOS_ADDR, SSB);
1868 	outb(CMOS_DATA, BIOS_SHUTDOWN);
1869 
1870 	while (get_apic_cmd1() & AV_PENDING)
1871 		apic_ret();
1872 
1873 	/* for integrated - make sure there is one INIT IPI in buffer */
1874 	/* for external - it will wake up the cpu */
1875 	apicadr[APIC_INT_CMD2] = cpu_id << APIC_ICR_ID_BIT_OFFSET;
1876 	apicadr[APIC_INT_CMD1] = AV_ASSERT | AV_RESET;
1877 
1878 	/* If only 1 CPU is installed, PENDING bit will not go low */
1879 	for (loop_count = 0x1000; loop_count; loop_count--)
1880 		if (get_apic_cmd1() & AV_PENDING)
1881 			apic_ret();
1882 		else
1883 			break;
1884 
1885 	apicadr[APIC_INT_CMD2] = cpu_id << APIC_ICR_ID_BIT_OFFSET;
1886 	apicadr[APIC_INT_CMD1] = AV_DEASSERT | AV_RESET;
1887 
1888 	drv_usecwait(20000);		/* 20 milli sec */
1889 
1890 	if (apic_cpus[cpun].aci_local_ver >= APIC_INTEGRATED_VERS) {
1891 		/* integrated apic */
1892 
1893 		rm_code = (caddr_t)(uintptr_t)rm_platter_pa;
1894 		vector = (rm_platter_pa >> MMU_PAGESHIFT) &
1895 		    (APIC_VECTOR_MASK | APIC_IPL_MASK);
1896 
1897 		/* to offset the INIT IPI queue up in the buffer */
1898 		apicadr[APIC_INT_CMD2] = cpu_id << APIC_ICR_ID_BIT_OFFSET;
1899 		apicadr[APIC_INT_CMD1] = vector | AV_STARTUP;
1900 
1901 		drv_usecwait(200);		/* 20 micro sec */
1902 
1903 		apicadr[APIC_INT_CMD2] = cpu_id << APIC_ICR_ID_BIT_OFFSET;
1904 		apicadr[APIC_INT_CMD1] = vector | AV_STARTUP;
1905 
1906 		drv_usecwait(200);		/* 20 micro sec */
1907 	}
1908 	intr_restore(iflag);
1909 }
1910 
1911 
1912 #ifdef	DEBUG
1913 int	apic_break_on_cpu = 9;
1914 int	apic_stretch_interrupts = 0;
1915 int	apic_stretch_ISR = 1 << 3;	/* IPL of 3 matches nothing now */
1916 
1917 void
1918 apic_break()
1919 {
1920 }
1921 #endif /* DEBUG */
1922 
1923 /*
1924  * platform_intr_enter
1925  *
1926  *	Called at the beginning of the interrupt service routine to
1927  *	mask all level equal to and below the interrupt priority
1928  *	of the interrupting vector.  An EOI should be given to
1929  *	the interrupt controller to enable other HW interrupts.
1930  *
1931  *	Return -1 for spurious interrupts
1932  *
1933  */
1934 /*ARGSUSED*/
1935 static int
1936 apic_intr_enter(int ipl, int *vectorp)
1937 {
1938 	uchar_t vector;
1939 	int nipl;
1940 	int irq, iflag;
1941 	apic_cpus_info_t *cpu_infop;
1942 
1943 	/*
1944 	 * The real vector programmed in APIC is *vectorp + 0x20
1945 	 * But, cmnint code subtracts 0x20 before pushing it.
1946 	 * Hence APIC_BASE_VECT is 0x20.
1947 	 */
1948 
1949 	vector = (uchar_t)*vectorp;
1950 
1951 	/* if interrupted by the clock, increment apic_nsec_since_boot */
1952 	if (vector == apic_clkvect) {
1953 		if (!apic_oneshot) {
1954 			/* NOTE: this is not MT aware */
1955 			apic_hrtime_stamp++;
1956 			apic_nsec_since_boot += apic_nsec_per_intr;
1957 			apic_hrtime_stamp++;
1958 			last_count_read = apic_hertz_count;
1959 			apic_redistribute_compute();
1960 		}
1961 
1962 		/* We will avoid all the book keeping overhead for clock */
1963 		nipl = apic_vectortoipl[vector >> APIC_IPL_SHIFT];
1964 #if defined(__amd64)
1965 		setcr8((ulong_t)apic_cr8pri[nipl]);
1966 #else
1967 		apicadr[APIC_TASK_REG] = apic_ipltopri[nipl];
1968 #endif
1969 		*vectorp = apic_vector_to_irq[vector + APIC_BASE_VECT];
1970 		apicadr[APIC_EOI_REG] = 0;
1971 		return (nipl);
1972 	}
1973 
1974 	cpu_infop = &apic_cpus[psm_get_cpu_id()];
1975 
1976 	if (vector == (APIC_SPUR_INTR - APIC_BASE_VECT)) {
1977 		cpu_infop->aci_spur_cnt++;
1978 		return (APIC_INT_SPURIOUS);
1979 	}
1980 
1981 	/* Check if the vector we got is really what we need */
1982 	if (apic_revector_pending) {
1983 		/*
1984 		 * Disable interrupts for the duration of
1985 		 * the vector translation to prevent a self-race for
1986 		 * the apic_revector_lock.  This cannot be done
1987 		 * in apic_xlate_vector because it is recursive and
1988 		 * we want the vector translation to be atomic with
1989 		 * respect to other (higher-priority) interrupts.
1990 		 */
1991 		iflag = intr_clear();
1992 		vector = apic_xlate_vector(vector + APIC_BASE_VECT) -
1993 		    APIC_BASE_VECT;
1994 		intr_restore(iflag);
1995 	}
1996 
1997 	nipl = apic_vectortoipl[vector >> APIC_IPL_SHIFT];
1998 	*vectorp = irq = apic_vector_to_irq[vector + APIC_BASE_VECT];
1999 
2000 #if defined(__amd64)
2001 	setcr8((ulong_t)apic_cr8pri[nipl]);
2002 #else
2003 	apicadr[APIC_TASK_REG] = apic_ipltopri[nipl];
2004 #endif
2005 
2006 	cpu_infop->aci_current[nipl] = (uchar_t)irq;
2007 	cpu_infop->aci_curipl = (uchar_t)nipl;
2008 	cpu_infop->aci_ISR_in_progress |= 1 << nipl;
2009 
2010 	/*
2011 	 * apic_level_intr could have been assimilated into the irq struct.
2012 	 * but, having it as a character array is more efficient in terms of
2013 	 * cache usage. So, we leave it as is.
2014 	 */
2015 	if (!apic_level_intr[irq])
2016 		apicadr[APIC_EOI_REG] = 0;
2017 
2018 #ifdef	DEBUG
2019 	APIC_DEBUG_BUF_PUT(vector);
2020 	APIC_DEBUG_BUF_PUT(irq);
2021 	APIC_DEBUG_BUF_PUT(nipl);
2022 	APIC_DEBUG_BUF_PUT(psm_get_cpu_id());
2023 	if ((apic_stretch_interrupts) && (apic_stretch_ISR & (1 << nipl)))
2024 		drv_usecwait(apic_stretch_interrupts);
2025 
2026 	if (apic_break_on_cpu == psm_get_cpu_id())
2027 		apic_break();
2028 #endif /* DEBUG */
2029 	return (nipl);
2030 }
2031 
2032 static void
2033 apic_intr_exit(int prev_ipl, int irq)
2034 {
2035 	apic_cpus_info_t *cpu_infop;
2036 
2037 #if defined(__amd64)
2038 	setcr8((ulong_t)apic_cr8pri[prev_ipl]);
2039 #else
2040 	apicadr[APIC_TASK_REG] = apic_ipltopri[prev_ipl];
2041 #endif
2042 
2043 	cpu_infop = &apic_cpus[psm_get_cpu_id()];
2044 	if (apic_level_intr[irq])
2045 		apicadr[APIC_EOI_REG] = 0;
2046 
2047 	cpu_infop->aci_curipl = (uchar_t)prev_ipl;
2048 	/* ISR above current pri could not be in progress */
2049 	cpu_infop->aci_ISR_in_progress &= (2 << prev_ipl) - 1;
2050 }
2051 
2052 /*
2053  * Mask all interrupts below or equal to the given IPL
2054  */
2055 static void
2056 apic_setspl(int ipl)
2057 {
2058 
2059 #if defined(__amd64)
2060 	setcr8((ulong_t)apic_cr8pri[ipl]);
2061 #else
2062 	apicadr[APIC_TASK_REG] = apic_ipltopri[ipl];
2063 #endif
2064 
2065 	/* interrupts at ipl above this cannot be in progress */
2066 	apic_cpus[psm_get_cpu_id()].aci_ISR_in_progress &= (2 << ipl) - 1;
2067 	/*
2068 	 * this is a patch fix for the ALR QSMP P5 machine, so that interrupts
2069 	 * have enough time to come in before the priority is raised again
2070 	 * during the idle() loop.
2071 	 */
2072 	if (apic_setspl_delay)
2073 		(void) get_apic_pri();
2074 }
2075 
2076 /*
2077  * trigger a software interrupt at the given IPL
2078  */
2079 static void
2080 apic_set_softintr(int ipl)
2081 {
2082 	int vector;
2083 	uint_t flag;
2084 
2085 	vector = apic_resv_vector[ipl];
2086 
2087 	flag = intr_clear();
2088 
2089 	while (get_apic_cmd1() & AV_PENDING)
2090 		apic_ret();
2091 
2092 	/* generate interrupt at vector on itself only */
2093 	apicadr[APIC_INT_CMD1] = AV_SH_SELF | vector;
2094 
2095 	intr_restore(flag);
2096 }
2097 
2098 /*
2099  * generates an interprocessor interrupt to another CPU
2100  */
2101 static void
2102 apic_send_ipi(int cpun, int ipl)
2103 {
2104 	int vector;
2105 	uint_t flag;
2106 
2107 	vector = apic_resv_vector[ipl];
2108 
2109 	flag = intr_clear();
2110 
2111 	while (get_apic_cmd1() & AV_PENDING)
2112 		apic_ret();
2113 
2114 	apicadr[APIC_INT_CMD2] =
2115 	    apic_cpus[cpun].aci_local_id << APIC_ICR_ID_BIT_OFFSET;
2116 	apicadr[APIC_INT_CMD1] = vector;
2117 
2118 	intr_restore(flag);
2119 }
2120 
2121 
2122 /*ARGSUSED*/
2123 static void
2124 apic_set_idlecpu(processorid_t cpun)
2125 {
2126 }
2127 
2128 /*ARGSUSED*/
2129 static void
2130 apic_unset_idlecpu(processorid_t cpun)
2131 {
2132 }
2133 
2134 
2135 static void
2136 apic_ret()
2137 {
2138 }
2139 
2140 static int
2141 get_apic_cmd1()
2142 {
2143 	return (apicadr[APIC_INT_CMD1]);
2144 }
2145 
2146 static int
2147 get_apic_pri()
2148 {
2149 #if defined(__amd64)
2150 	return ((int)getcr8());
2151 #else
2152 	return (apicadr[APIC_TASK_REG]);
2153 #endif
2154 }
2155 
2156 /*
2157  * If apic_coarse_time == 1, then apic_gettime() is used instead of
2158  * apic_gethrtime().  This is used for performance instead of accuracy.
2159  */
2160 
2161 static hrtime_t
2162 apic_gettime()
2163 {
2164 	int old_hrtime_stamp;
2165 	hrtime_t temp;
2166 
2167 	/*
2168 	 * In one-shot mode, we do not keep time, so if anyone
2169 	 * calls psm_gettime() 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 
2179 gettime_again:
2180 	while ((old_hrtime_stamp = apic_hrtime_stamp) & 1)
2181 		apic_ret();
2182 
2183 	temp = apic_nsec_since_boot;
2184 
2185 	if (apic_hrtime_stamp != old_hrtime_stamp) {	/* got an interrupt */
2186 		goto gettime_again;
2187 	}
2188 	return (temp);
2189 }
2190 
2191 /*
2192  * Here we return the number of nanoseconds since booting.  Note every
2193  * clock interrupt increments apic_nsec_since_boot by the appropriate
2194  * amount.
2195  */
2196 static hrtime_t
2197 apic_gethrtime()
2198 {
2199 	int curr_timeval, countval, elapsed_ticks, oflags;
2200 	int old_hrtime_stamp, status;
2201 	hrtime_t temp;
2202 	uchar_t	cpun;
2203 
2204 
2205 	/*
2206 	 * In one-shot mode, we do not keep time, so if anyone
2207 	 * calls psm_gethrtime() directly, we vector over to
2208 	 * gethrtime().
2209 	 * one-shot mode MUST NOT be enabled if this psm is the source of
2210 	 * hrtime.
2211 	 */
2212 
2213 	if (apic_oneshot)
2214 		return (gethrtime());
2215 
2216 	oflags = intr_clear();	/* prevent migration */
2217 
2218 	cpun = (uchar_t)((uint_t)apicadr[APIC_LID_REG] >> APIC_ID_BIT_OFFSET);
2219 
2220 	lock_set(&apic_gethrtime_lock);
2221 
2222 gethrtime_again:
2223 	while ((old_hrtime_stamp = apic_hrtime_stamp) & 1)
2224 		apic_ret();
2225 
2226 	/*
2227 	 * Check to see which CPU we are on.  Note the time is kept on
2228 	 * the local APIC of CPU 0.  If on CPU 0, simply read the current
2229 	 * counter.  If on another CPU, issue a remote read command to CPU 0.
2230 	 */
2231 	if (cpun == apic_cpus[0].aci_local_id) {
2232 		countval = apicadr[APIC_CURR_COUNT];
2233 	} else {
2234 		while (get_apic_cmd1() & AV_PENDING)
2235 			apic_ret();
2236 
2237 		apicadr[APIC_INT_CMD2] =
2238 		    apic_cpus[0].aci_local_id << APIC_ICR_ID_BIT_OFFSET;
2239 		apicadr[APIC_INT_CMD1] = APIC_CURR_ADD|AV_REMOTE;
2240 
2241 		while ((status = get_apic_cmd1()) & AV_READ_PENDING)
2242 			apic_ret();
2243 
2244 		if (status & AV_REMOTE_STATUS)	/* 1 = valid */
2245 			countval = apicadr[APIC_REMOTE_READ];
2246 		else {	/* 0 = invalid */
2247 			apic_remote_hrterr++;
2248 			/*
2249 			 * return last hrtime right now, will need more
2250 			 * testing if change to retry
2251 			 */
2252 			temp = apic_last_hrtime;
2253 
2254 			lock_clear(&apic_gethrtime_lock);
2255 
2256 			intr_restore(oflags);
2257 
2258 			return (temp);
2259 		}
2260 	}
2261 	if (countval > last_count_read)
2262 		countval = 0;
2263 	else
2264 		last_count_read = countval;
2265 
2266 	elapsed_ticks = apic_hertz_count - countval;
2267 
2268 	curr_timeval = APIC_TICKS_TO_NSECS(elapsed_ticks);
2269 	temp = apic_nsec_since_boot + curr_timeval;
2270 
2271 	if (apic_hrtime_stamp != old_hrtime_stamp) {	/* got an interrupt */
2272 		/* we might have clobbered last_count_read. Restore it */
2273 		last_count_read = apic_hertz_count;
2274 		goto gethrtime_again;
2275 	}
2276 
2277 	if (temp < apic_last_hrtime) {
2278 		/* return last hrtime if error occurs */
2279 		apic_hrtime_error++;
2280 		temp = apic_last_hrtime;
2281 	}
2282 	else
2283 		apic_last_hrtime = temp;
2284 
2285 	lock_clear(&apic_gethrtime_lock);
2286 	intr_restore(oflags);
2287 
2288 	return (temp);
2289 }
2290 
2291 /* apic NMI handler */
2292 /*ARGSUSED*/
2293 static void
2294 apic_nmi_intr(caddr_t arg)
2295 {
2296 	if (apic_shutdown_processors) {
2297 		apic_disable_local_apic();
2298 		return;
2299 	}
2300 
2301 	if (lock_try(&apic_nmi_lock)) {
2302 		if (apic_kmdb_on_nmi) {
2303 			if (psm_debugger() == 0) {
2304 				cmn_err(CE_PANIC,
2305 				    "NMI detected, kmdb is not available.");
2306 			} else {
2307 				debug_enter("\nNMI detected, entering kmdb.\n");
2308 			}
2309 		} else {
2310 			if (apic_panic_on_nmi) {
2311 				/* Keep panic from entering kmdb. */
2312 				nopanicdebug = 1;
2313 				cmn_err(CE_PANIC, "pcplusmp: NMI received");
2314 			} else {
2315 				/*
2316 				 * prom_printf is the best shot we have
2317 				 * of something which is problem free from
2318 				 * high level/NMI type of interrupts
2319 				 */
2320 				prom_printf("pcplusmp: NMI received\n");
2321 				apic_error |= APIC_ERR_NMI;
2322 				apic_num_nmis++;
2323 			}
2324 		}
2325 		lock_clear(&apic_nmi_lock);
2326 	}
2327 }
2328 
2329 /*
2330  * Add mask bits to disable interrupt vector from happening
2331  * at or above IPL. In addition, it should remove mask bits
2332  * to enable interrupt vectors below the given IPL.
2333  *
2334  * Both add and delspl are complicated by the fact that different interrupts
2335  * may share IRQs. This can happen in two ways.
2336  * 1. The same H/W line is shared by more than 1 device
2337  * 1a. with interrupts at different IPLs
2338  * 1b. with interrupts at same IPL
2339  * 2. We ran out of vectors at a given IPL and started sharing vectors.
2340  * 1b and 2 should be handled gracefully, except for the fact some ISRs
2341  * will get called often when no interrupt is pending for the device.
2342  * For 1a, we just hope that the machine blows up with the person who
2343  * set it up that way!. In the meantime, we handle it at the higher IPL.
2344  */
2345 /*ARGSUSED*/
2346 static int
2347 apic_addspl(int irqno, int ipl, int min_ipl, int max_ipl)
2348 {
2349 	uchar_t vector;
2350 	int iflag;
2351 	apic_irq_t *irqptr, *irqheadptr;
2352 	int irqindex;
2353 
2354 	ASSERT(max_ipl <= UCHAR_MAX);
2355 	irqindex = IRQINDEX(irqno);
2356 
2357 	if ((irqindex == -1) || (!apic_irq_table[irqindex]))
2358 		return (PSM_FAILURE);
2359 
2360 	mutex_enter(&airq_mutex);
2361 	irqptr = irqheadptr = apic_irq_table[irqindex];
2362 
2363 	DDI_INTR_IMPLDBG((CE_CONT, "apic_addspl: dip=0x%p type=%d irqno=0x%x "
2364 	    "vector=0x%x\n", (void *)irqptr->airq_dip,
2365 	    irqptr->airq_mps_intr_index, irqno, irqptr->airq_vector));
2366 
2367 	while (irqptr) {
2368 		if (VIRTIRQ(irqindex, irqptr->airq_share_id) == irqno)
2369 			break;
2370 		irqptr = irqptr->airq_next;
2371 	}
2372 	irqptr->airq_share++;
2373 
2374 	mutex_exit(&airq_mutex);
2375 
2376 	/* return if it is not hardware interrupt */
2377 	if (irqptr->airq_mps_intr_index == RESERVE_INDEX)
2378 		return (PSM_SUCCESS);
2379 
2380 	/* Or if there are more interupts at a higher IPL */
2381 	if (ipl != max_ipl)
2382 		return (PSM_SUCCESS);
2383 
2384 	/*
2385 	 * if apic_picinit() has not been called yet, just return.
2386 	 * At the end of apic_picinit(), we will call setup_io_intr().
2387 	 */
2388 
2389 	if (!apic_flag)
2390 		return (PSM_SUCCESS);
2391 
2392 	/*
2393 	 * Upgrade vector if max_ipl is not earlier ipl. If we cannot allocate,
2394 	 * return failure. Not very elegant, but then we hope the
2395 	 * machine will blow up with ...
2396 	 */
2397 	if (irqptr->airq_ipl != max_ipl) {
2398 		vector = apic_allocate_vector(max_ipl, irqindex, 1);
2399 		if (vector == 0) {
2400 			irqptr->airq_share--;
2401 			return (PSM_FAILURE);
2402 		}
2403 		irqptr = irqheadptr;
2404 		apic_mark_vector(irqptr->airq_vector, vector);
2405 		while (irqptr) {
2406 			irqptr->airq_vector = vector;
2407 			irqptr->airq_ipl = (uchar_t)max_ipl;
2408 			/*
2409 			 * reprogram irq being added and every one else
2410 			 * who is not in the UNINIT state
2411 			 */
2412 			if ((VIRTIRQ(irqindex, irqptr->airq_share_id) ==
2413 			    irqno) || (irqptr->airq_temp_cpu != IRQ_UNINIT)) {
2414 				apic_record_rdt_entry(irqptr, irqindex);
2415 
2416 				iflag = intr_clear();
2417 				lock_set(&apic_ioapic_lock);
2418 
2419 				(void) apic_setup_io_intr(irqptr, irqindex,
2420 				    B_FALSE);
2421 
2422 				lock_clear(&apic_ioapic_lock);
2423 				intr_restore(iflag);
2424 			}
2425 			irqptr = irqptr->airq_next;
2426 		}
2427 		return (PSM_SUCCESS);
2428 	}
2429 
2430 	ASSERT(irqptr);
2431 
2432 	iflag = intr_clear();
2433 	lock_set(&apic_ioapic_lock);
2434 
2435 	(void) apic_setup_io_intr(irqptr, irqindex, B_FALSE);
2436 
2437 	lock_clear(&apic_ioapic_lock);
2438 	intr_restore(iflag);
2439 
2440 	return (PSM_SUCCESS);
2441 }
2442 
2443 /*
2444  * Recompute mask bits for the given interrupt vector.
2445  * If there is no interrupt servicing routine for this
2446  * vector, this function should disable interrupt vector
2447  * from happening at all IPLs. If there are still
2448  * handlers using the given vector, this function should
2449  * disable the given vector from happening below the lowest
2450  * IPL of the remaining hadlers.
2451  */
2452 /*ARGSUSED*/
2453 static int
2454 apic_delspl(int irqno, int ipl, int min_ipl, int max_ipl)
2455 {
2456 	uchar_t vector, bind_cpu;
2457 	int	iflag, intin, irqindex;
2458 	volatile int32_t *ioapic;
2459 	apic_irq_t	*irqptr, *irqheadptr;
2460 
2461 	mutex_enter(&airq_mutex);
2462 	irqindex = IRQINDEX(irqno);
2463 	irqptr = irqheadptr = apic_irq_table[irqindex];
2464 
2465 	DDI_INTR_IMPLDBG((CE_CONT, "apic_delspl: dip=0x%p type=%d irqno=0x%x "
2466 	    "vector=0x%x\n", (void *)irqptr->airq_dip,
2467 	    irqptr->airq_mps_intr_index, irqno, irqptr->airq_vector));
2468 
2469 	while (irqptr) {
2470 		if (VIRTIRQ(irqindex, irqptr->airq_share_id) == irqno)
2471 			break;
2472 		irqptr = irqptr->airq_next;
2473 	}
2474 	ASSERT(irqptr);
2475 
2476 	irqptr->airq_share--;
2477 
2478 	mutex_exit(&airq_mutex);
2479 
2480 	if (ipl < max_ipl)
2481 		return (PSM_SUCCESS);
2482 
2483 	/* return if it is not hardware interrupt */
2484 	if (irqptr->airq_mps_intr_index == RESERVE_INDEX)
2485 		return (PSM_SUCCESS);
2486 
2487 	if (!apic_flag) {
2488 		/*
2489 		 * Clear irq_struct. If two devices shared an intpt
2490 		 * line & 1 unloaded before picinit, we are hosed. But, then
2491 		 * we hope the machine will ...
2492 		 */
2493 		irqptr->airq_mps_intr_index = FREE_INDEX;
2494 		irqptr->airq_temp_cpu = IRQ_UNINIT;
2495 		apic_free_vector(irqptr->airq_vector);
2496 		return (PSM_SUCCESS);
2497 	}
2498 	/*
2499 	 * Downgrade vector to new max_ipl if needed.If we cannot allocate,
2500 	 * use old IPL. Not very elegant, but then we hope ...
2501 	 */
2502 	if ((irqptr->airq_ipl != max_ipl) && (max_ipl != PSM_INVALID_IPL)) {
2503 		apic_irq_t	*irqp;
2504 		if (vector = apic_allocate_vector(max_ipl, irqno, 1)) {
2505 			apic_mark_vector(irqheadptr->airq_vector, vector);
2506 			irqp = irqheadptr;
2507 			while (irqp) {
2508 				irqp->airq_vector = vector;
2509 				irqp->airq_ipl = (uchar_t)max_ipl;
2510 				if (irqp->airq_temp_cpu != IRQ_UNINIT) {
2511 					apic_record_rdt_entry(irqp, irqindex);
2512 
2513 					iflag = intr_clear();
2514 					lock_set(&apic_ioapic_lock);
2515 
2516 					(void) apic_setup_io_intr(irqp,
2517 					    irqindex, B_FALSE);
2518 
2519 					lock_clear(&apic_ioapic_lock);
2520 					intr_restore(iflag);
2521 				}
2522 				irqp = irqp->airq_next;
2523 			}
2524 		}
2525 	}
2526 
2527 	if (irqptr->airq_share)
2528 		return (PSM_SUCCESS);
2529 
2530 	iflag = intr_clear();
2531 	lock_set(&apic_ioapic_lock);
2532 
2533 	/* Disable the MSI/X vector */
2534 	if (APIC_IS_MSI_OR_MSIX_INDEX(irqptr->airq_mps_intr_index)) {
2535 		int type = (irqptr->airq_mps_intr_index == MSI_INDEX) ?
2536 		    DDI_INTR_TYPE_MSI : DDI_INTR_TYPE_MSIX;
2537 
2538 		/*
2539 		 * Make sure we only disable on the last
2540 		 * of the multi-MSI support
2541 		 */
2542 		if (i_ddi_intr_get_current_nintrs(irqptr->airq_dip) == 1) {
2543 			(void) apic_pci_msi_unconfigure(irqptr->airq_dip,
2544 			    type, irqptr->airq_ioapicindex);
2545 			(void) apic_pci_msi_disable_mode(irqptr->airq_dip,
2546 			    type, irqptr->airq_ioapicindex);
2547 		}
2548 	} else {
2549 		ioapic = apicioadr[irqptr->airq_ioapicindex];
2550 		intin = irqptr->airq_intin_no;
2551 		ioapic[APIC_IO_REG] = APIC_RDT_CMD + 2 * intin;
2552 		ioapic[APIC_IO_DATA] = AV_MASK;
2553 	}
2554 
2555 	if (max_ipl == PSM_INVALID_IPL) {
2556 		ASSERT(irqheadptr == irqptr);
2557 		bind_cpu = irqptr->airq_temp_cpu;
2558 		if (((uchar_t)bind_cpu != IRQ_UNBOUND) &&
2559 		    ((uchar_t)bind_cpu != IRQ_UNINIT)) {
2560 			ASSERT((bind_cpu & ~IRQ_USER_BOUND) < apic_nproc);
2561 			if (bind_cpu & IRQ_USER_BOUND) {
2562 				/* If hardbound, temp_cpu == cpu */
2563 				bind_cpu &= ~IRQ_USER_BOUND;
2564 				apic_cpus[bind_cpu].aci_bound--;
2565 			} else
2566 				apic_cpus[bind_cpu].aci_temp_bound--;
2567 		}
2568 		irqptr->airq_temp_cpu = IRQ_UNINIT;
2569 		irqptr->airq_mps_intr_index = FREE_INDEX;
2570 		lock_clear(&apic_ioapic_lock);
2571 		intr_restore(iflag);
2572 		apic_free_vector(irqptr->airq_vector);
2573 		return (PSM_SUCCESS);
2574 	}
2575 	lock_clear(&apic_ioapic_lock);
2576 	intr_restore(iflag);
2577 
2578 	mutex_enter(&airq_mutex);
2579 	if ((irqptr == apic_irq_table[irqindex])) {
2580 		apic_irq_t	*oldirqptr;
2581 		/* Move valid irq entry to the head */
2582 		irqheadptr = oldirqptr = irqptr;
2583 		irqptr = irqptr->airq_next;
2584 		ASSERT(irqptr);
2585 		while (irqptr) {
2586 			if (irqptr->airq_mps_intr_index != FREE_INDEX)
2587 				break;
2588 			oldirqptr = irqptr;
2589 			irqptr = irqptr->airq_next;
2590 		}
2591 		/* remove all invalid ones from the beginning */
2592 		apic_irq_table[irqindex] = irqptr;
2593 		/*
2594 		 * and link them back after the head. The invalid ones
2595 		 * begin with irqheadptr and end at oldirqptr
2596 		 */
2597 		oldirqptr->airq_next = irqptr->airq_next;
2598 		irqptr->airq_next = irqheadptr;
2599 	}
2600 	mutex_exit(&airq_mutex);
2601 
2602 	irqptr->airq_temp_cpu = IRQ_UNINIT;
2603 	irqptr->airq_mps_intr_index = FREE_INDEX;
2604 
2605 	return (PSM_SUCCESS);
2606 }
2607 
2608 /*
2609  * Return HW interrupt number corresponding to the given IPL
2610  */
2611 /*ARGSUSED*/
2612 static int
2613 apic_softlvl_to_irq(int ipl)
2614 {
2615 	/*
2616 	 * Do not use apic to trigger soft interrupt.
2617 	 * It will cause the system to hang when 2 hardware interrupts
2618 	 * at the same priority with the softint are already accepted
2619 	 * by the apic.  Cause the AV_PENDING bit will not be cleared
2620 	 * until one of the hardware interrupt is eoi'ed.  If we need
2621 	 * to send an ipi at this time, we will end up looping forever
2622 	 * to wait for the AV_PENDING bit to clear.
2623 	 */
2624 	return (PSM_SV_SOFTWARE);
2625 }
2626 
2627 static int
2628 apic_post_cpu_start()
2629 {
2630 	int i, cpun, iflag;
2631 	apic_irq_t *irq_ptr;
2632 
2633 	apic_init_intr();
2634 
2635 	/*
2636 	 * since some systems don't enable the internal cache on the non-boot
2637 	 * cpus, so we have to enable them here
2638 	 */
2639 	setcr0(getcr0() & ~(0x60000000));
2640 
2641 	while (get_apic_cmd1() & AV_PENDING)
2642 		apic_ret();
2643 
2644 	cpun = psm_get_cpu_id();
2645 
2646 	apic_cpus[cpun].aci_status = APIC_CPU_ONLINE | APIC_CPU_INTR_ENABLE;
2647 
2648 	for (i = apic_min_device_irq; i <= apic_max_device_irq; i++) {
2649 		irq_ptr = apic_irq_table[i];
2650 		if ((irq_ptr == NULL) ||
2651 		    ((irq_ptr->airq_cpu & ~IRQ_USER_BOUND) != cpun))
2652 			continue;
2653 
2654 		while (irq_ptr) {
2655 			if (irq_ptr->airq_temp_cpu != IRQ_UNINIT) {
2656 				iflag = intr_clear();
2657 				lock_set(&apic_ioapic_lock);
2658 
2659 				(void) apic_rebind(irq_ptr, cpun, NULL);
2660 
2661 				lock_clear(&apic_ioapic_lock);
2662 				intr_restore(iflag);
2663 			}
2664 			irq_ptr = irq_ptr->airq_next;
2665 		}
2666 	}
2667 
2668 
2669 	apicadr[APIC_DIVIDE_REG] = apic_divide_reg_init;
2670 	return (PSM_SUCCESS);
2671 }
2672 
2673 processorid_t
2674 apic_get_next_processorid(processorid_t cpu_id)
2675 {
2676 
2677 	int i;
2678 
2679 	if (cpu_id == -1)
2680 		return ((processorid_t)0);
2681 
2682 	for (i = cpu_id + 1; i < NCPU; i++) {
2683 		if (CPU_IN_SET(apic_cpumask, i))
2684 			return (i);
2685 	}
2686 
2687 	return ((processorid_t)-1);
2688 }
2689 
2690 
2691 /*
2692  * type == -1 indicates it is an internal request. Do not change
2693  * resv_vector for these requests
2694  */
2695 static int
2696 apic_get_ipivect(int ipl, int type)
2697 {
2698 	uchar_t vector;
2699 	int irq;
2700 
2701 	if (irq = apic_allocate_irq(APIC_VECTOR(ipl))) {
2702 		if (vector = apic_allocate_vector(ipl, irq, 1)) {
2703 			apic_irq_table[irq]->airq_mps_intr_index =
2704 			    RESERVE_INDEX;
2705 			apic_irq_table[irq]->airq_vector = vector;
2706 			if (type != -1) {
2707 				apic_resv_vector[ipl] = vector;
2708 			}
2709 			return (irq);
2710 		}
2711 	}
2712 	apic_error |= APIC_ERR_GET_IPIVECT_FAIL;
2713 	return (-1);	/* shouldn't happen */
2714 }
2715 
2716 static int
2717 apic_getclkirq(int ipl)
2718 {
2719 	int	irq;
2720 
2721 	if ((irq = apic_get_ipivect(ipl, -1)) == -1)
2722 		return (-1);
2723 	/*
2724 	 * Note the vector in apic_clkvect for per clock handling.
2725 	 */
2726 	apic_clkvect = apic_irq_table[irq]->airq_vector - APIC_BASE_VECT;
2727 	APIC_VERBOSE_IOAPIC((CE_NOTE, "get_clkirq: vector = %x\n",
2728 	    apic_clkvect));
2729 	return (irq);
2730 }
2731 
2732 
2733 /*
2734  * Return the number of APIC clock ticks elapsed for 8245 to decrement
2735  * (APIC_TIME_COUNT + pit_ticks_adj) ticks.
2736  */
2737 static uint_t
2738 apic_calibrate(volatile uint32_t *addr, uint16_t *pit_ticks_adj)
2739 {
2740 	uint8_t		pit_tick_lo;
2741 	uint16_t	pit_tick, target_pit_tick;
2742 	uint32_t	start_apic_tick, end_apic_tick;
2743 	int		iflag;
2744 
2745 	addr += APIC_CURR_COUNT;
2746 
2747 	iflag = intr_clear();
2748 
2749 	do {
2750 		pit_tick_lo = inb(PITCTR0_PORT);
2751 		pit_tick = (inb(PITCTR0_PORT) << 8) | pit_tick_lo;
2752 	} while (pit_tick < APIC_TIME_MIN ||
2753 	    pit_tick_lo <= APIC_LB_MIN || pit_tick_lo >= APIC_LB_MAX);
2754 
2755 	/*
2756 	 * Wait for the 8254 to decrement by 5 ticks to ensure
2757 	 * we didn't start in the middle of a tick.
2758 	 * Compare with 0x10 for the wrap around case.
2759 	 */
2760 	target_pit_tick = pit_tick - 5;
2761 	do {
2762 		pit_tick_lo = inb(PITCTR0_PORT);
2763 		pit_tick = (inb(PITCTR0_PORT) << 8) | pit_tick_lo;
2764 	} while (pit_tick > target_pit_tick || pit_tick_lo < 0x10);
2765 
2766 	start_apic_tick = *addr;
2767 
2768 	/*
2769 	 * Wait for the 8254 to decrement by
2770 	 * (APIC_TIME_COUNT + pit_ticks_adj) ticks
2771 	 */
2772 	target_pit_tick = pit_tick - APIC_TIME_COUNT;
2773 	do {
2774 		pit_tick_lo = inb(PITCTR0_PORT);
2775 		pit_tick = (inb(PITCTR0_PORT) << 8) | pit_tick_lo;
2776 	} while (pit_tick > target_pit_tick || pit_tick_lo < 0x10);
2777 
2778 	end_apic_tick = *addr;
2779 
2780 	*pit_ticks_adj = target_pit_tick - pit_tick;
2781 
2782 	intr_restore(iflag);
2783 
2784 	return (start_apic_tick - end_apic_tick);
2785 }
2786 
2787 /*
2788  * Initialise the APIC timer on the local APIC of CPU 0 to the desired
2789  * frequency.  Note at this stage in the boot sequence, the boot processor
2790  * is the only active processor.
2791  * hertz value of 0 indicates a one-shot mode request.  In this case
2792  * the function returns the resolution (in nanoseconds) for the hardware
2793  * timer interrupt.  If one-shot mode capability is not available,
2794  * the return value will be 0. apic_enable_oneshot is a global switch
2795  * for disabling the functionality.
2796  * A non-zero positive value for hertz indicates a periodic mode request.
2797  * In this case the hardware will be programmed to generate clock interrupts
2798  * at hertz frequency and returns the resolution of interrupts in
2799  * nanosecond.
2800  */
2801 
2802 static int
2803 apic_clkinit(int hertz)
2804 {
2805 
2806 	uint_t		apic_ticks = 0;
2807 	uint_t		pit_ticks;
2808 	int		ret;
2809 	uint16_t	pit_ticks_adj;
2810 	static int	firsttime = 1;
2811 
2812 	if (firsttime) {
2813 		/* first time calibrate on CPU0 only */
2814 
2815 		apicadr[APIC_DIVIDE_REG] = apic_divide_reg_init;
2816 		apicadr[APIC_INIT_COUNT] = APIC_MAXVAL;	/* start counting */
2817 		apic_ticks = apic_calibrate(apicadr, &pit_ticks_adj);
2818 
2819 		/* total number of PIT ticks corresponding to apic_ticks */
2820 		pit_ticks = APIC_TIME_COUNT + pit_ticks_adj;
2821 
2822 		/*
2823 		 * Determine the number of nanoseconds per APIC clock tick
2824 		 * and then determine how many APIC ticks to interrupt at the
2825 		 * desired frequency
2826 		 * apic_ticks / (pitticks / PIT_HZ) = apic_ticks_per_s
2827 		 * (apic_ticks * PIT_HZ) / pitticks = apic_ticks_per_s
2828 		 * apic_ticks_per_ns = (apic_ticks * PIT_HZ) / (pitticks * 10^9)
2829 		 * apic_ticks_per_SFns =
2830 		 *   (SF * apic_ticks * PIT_HZ) / (pitticks * 10^9)
2831 		 */
2832 		apic_ticks_per_SFnsecs =
2833 		    ((SF * apic_ticks * PIT_HZ) /
2834 		    ((uint64_t)pit_ticks * NANOSEC));
2835 
2836 		/* the interval timer initial count is 32 bit max */
2837 		apic_nsec_max = APIC_TICKS_TO_NSECS(APIC_MAXVAL);
2838 		firsttime = 0;
2839 	}
2840 
2841 	if (hertz != 0) {
2842 		/* periodic */
2843 		apic_nsec_per_intr = NANOSEC / hertz;
2844 		apic_hertz_count = APIC_NSECS_TO_TICKS(apic_nsec_per_intr);
2845 	}
2846 
2847 	apic_int_busy_mark = (apic_int_busy_mark *
2848 	    apic_sample_factor_redistribution) / 100;
2849 	apic_int_free_mark = (apic_int_free_mark *
2850 	    apic_sample_factor_redistribution) / 100;
2851 	apic_diff_for_redistribution = (apic_diff_for_redistribution *
2852 	    apic_sample_factor_redistribution) / 100;
2853 
2854 	if (hertz == 0) {
2855 		/* requested one_shot */
2856 		if (!apic_oneshot_enable)
2857 			return (0);
2858 		apic_oneshot = 1;
2859 		ret = (int)APIC_TICKS_TO_NSECS(1);
2860 	} else {
2861 		/* program the local APIC to interrupt at the given frequency */
2862 		apicadr[APIC_INIT_COUNT] = apic_hertz_count;
2863 		apicadr[APIC_LOCAL_TIMER] =
2864 		    (apic_clkvect + APIC_BASE_VECT) | AV_TIME;
2865 		apic_oneshot = 0;
2866 		ret = NANOSEC / hertz;
2867 	}
2868 
2869 	return (ret);
2870 
2871 }
2872 
2873 /*
2874  * apic_preshutdown:
2875  * Called early in shutdown whilst we can still access filesystems to do
2876  * things like loading modules which will be required to complete shutdown
2877  * after filesystems are all unmounted.
2878  */
2879 static void
2880 apic_preshutdown(int cmd, int fcn)
2881 {
2882 	APIC_VERBOSE_POWEROFF(("apic_preshutdown(%d,%d); m=%d a=%d\n",
2883 	    cmd, fcn, apic_poweroff_method, apic_enable_acpi));
2884 
2885 	if ((cmd != A_SHUTDOWN) || (fcn != AD_POWEROFF)) {
2886 		return;
2887 	}
2888 }
2889 
2890 static void
2891 apic_shutdown(int cmd, int fcn)
2892 {
2893 	int iflag, restarts, attempts;
2894 	int i, j;
2895 	volatile int32_t *ioapic;
2896 	uchar_t	byte;
2897 
2898 	/* Send NMI to all CPUs except self to do per processor shutdown */
2899 	iflag = intr_clear();
2900 	while (get_apic_cmd1() & AV_PENDING)
2901 		apic_ret();
2902 	apic_shutdown_processors = 1;
2903 	apicadr[APIC_INT_CMD1] = AV_NMI | AV_LEVEL | AV_SH_ALL_EXCSELF;
2904 
2905 	/* restore cmos shutdown byte before reboot */
2906 	if (apic_cmos_ssb_set) {
2907 		outb(CMOS_ADDR, SSB);
2908 		outb(CMOS_DATA, 0);
2909 	}
2910 	/* Disable the I/O APIC redirection entries */
2911 	for (j = 0; j < apic_io_max; j++) {
2912 		int intin_max;
2913 		ioapic = apicioadr[j];
2914 		ioapic[APIC_IO_REG] = APIC_VERS_CMD;
2915 		/* Bits 23-16 define the maximum redirection entries */
2916 		intin_max = (ioapic[APIC_IO_DATA] >> 16) & 0xff;
2917 		for (i = 0; i < intin_max; i++) {
2918 			ioapic[APIC_IO_REG] = APIC_RDT_CMD + 2 * i;
2919 			ioapic[APIC_IO_DATA] = AV_MASK;
2920 		}
2921 	}
2922 
2923 	/*	disable apic mode if imcr present	*/
2924 	if (apic_imcrp) {
2925 		outb(APIC_IMCR_P1, (uchar_t)APIC_IMCR_SELECT);
2926 		outb(APIC_IMCR_P2, (uchar_t)APIC_IMCR_PIC);
2927 	}
2928 
2929 	apic_disable_local_apic();
2930 
2931 	intr_restore(iflag);
2932 
2933 	if ((cmd != A_SHUTDOWN) || (fcn != AD_POWEROFF)) {
2934 		return;
2935 	}
2936 
2937 	switch (apic_poweroff_method) {
2938 		case APIC_POWEROFF_VIA_RTC:
2939 
2940 			/* select the extended NVRAM bank in the RTC */
2941 			outb(CMOS_ADDR, RTC_REGA);
2942 			byte = inb(CMOS_DATA);
2943 			outb(CMOS_DATA, (byte | EXT_BANK));
2944 
2945 			outb(CMOS_ADDR, PFR_REG);
2946 
2947 			/* for Predator must toggle the PAB bit */
2948 			byte = inb(CMOS_DATA);
2949 
2950 			/*
2951 			 * clear power active bar, wakeup alarm and
2952 			 * kickstart
2953 			 */
2954 			byte &= ~(PAB_CBIT | WF_FLAG | KS_FLAG);
2955 			outb(CMOS_DATA, byte);
2956 
2957 			/* delay before next write */
2958 			drv_usecwait(1000);
2959 
2960 			/* for S40 the following would suffice */
2961 			byte = inb(CMOS_DATA);
2962 
2963 			/* power active bar control bit */
2964 			byte |= PAB_CBIT;
2965 			outb(CMOS_DATA, byte);
2966 
2967 			break;
2968 
2969 		case APIC_POWEROFF_VIA_ASPEN_BMC:
2970 			restarts = 0;
2971 restart_aspen_bmc:
2972 			if (++restarts == 3)
2973 				break;
2974 			attempts = 0;
2975 			do {
2976 				byte = inb(MISMIC_FLAG_REGISTER);
2977 				byte &= MISMIC_BUSY_MASK;
2978 				if (byte != 0) {
2979 					drv_usecwait(1000);
2980 					if (attempts >= 3)
2981 						goto restart_aspen_bmc;
2982 					++attempts;
2983 				}
2984 			} while (byte != 0);
2985 			outb(MISMIC_CNTL_REGISTER, CC_SMS_GET_STATUS);
2986 			byte = inb(MISMIC_FLAG_REGISTER);
2987 			byte |= 0x1;
2988 			outb(MISMIC_FLAG_REGISTER, byte);
2989 			i = 0;
2990 			for (; i < (sizeof (aspen_bmc)/sizeof (aspen_bmc[0]));
2991 			    i++) {
2992 				attempts = 0;
2993 				do {
2994 					byte = inb(MISMIC_FLAG_REGISTER);
2995 					byte &= MISMIC_BUSY_MASK;
2996 					if (byte != 0) {
2997 						drv_usecwait(1000);
2998 						if (attempts >= 3)
2999 							goto restart_aspen_bmc;
3000 						++attempts;
3001 					}
3002 				} while (byte != 0);
3003 				outb(MISMIC_CNTL_REGISTER, aspen_bmc[i].cntl);
3004 				outb(MISMIC_DATA_REGISTER, aspen_bmc[i].data);
3005 				byte = inb(MISMIC_FLAG_REGISTER);
3006 				byte |= 0x1;
3007 				outb(MISMIC_FLAG_REGISTER, byte);
3008 			}
3009 			break;
3010 
3011 		case APIC_POWEROFF_VIA_SITKA_BMC:
3012 			restarts = 0;
3013 restart_sitka_bmc:
3014 			if (++restarts == 3)
3015 				break;
3016 			attempts = 0;
3017 			do {
3018 				byte = inb(SMS_STATUS_REGISTER);
3019 				byte &= SMS_STATE_MASK;
3020 				if ((byte == SMS_READ_STATE) ||
3021 				    (byte == SMS_WRITE_STATE)) {
3022 					drv_usecwait(1000);
3023 					if (attempts >= 3)
3024 						goto restart_sitka_bmc;
3025 					++attempts;
3026 				}
3027 			} while ((byte == SMS_READ_STATE) ||
3028 			    (byte == SMS_WRITE_STATE));
3029 			outb(SMS_COMMAND_REGISTER, SMS_GET_STATUS);
3030 			i = 0;
3031 			for (; i < (sizeof (sitka_bmc)/sizeof (sitka_bmc[0]));
3032 			    i++) {
3033 				attempts = 0;
3034 				do {
3035 					byte = inb(SMS_STATUS_REGISTER);
3036 					byte &= SMS_IBF_MASK;
3037 					if (byte != 0) {
3038 						drv_usecwait(1000);
3039 						if (attempts >= 3)
3040 							goto restart_sitka_bmc;
3041 						++attempts;
3042 					}
3043 				} while (byte != 0);
3044 				outb(sitka_bmc[i].port, sitka_bmc[i].data);
3045 			}
3046 			break;
3047 
3048 		case APIC_POWEROFF_NONE:
3049 
3050 			/* If no APIC direct method, we will try using ACPI */
3051 			if (apic_enable_acpi) {
3052 				if (acpi_poweroff() == 1)
3053 					return;
3054 			} else
3055 				return;
3056 
3057 			break;
3058 	}
3059 	/*
3060 	 * Wait a limited time here for power to go off.
3061 	 * If the power does not go off, then there was a
3062 	 * problem and we should continue to the halt which
3063 	 * prints a message for the user to press a key to
3064 	 * reboot.
3065 	 */
3066 	drv_usecwait(7000000); /* wait seven seconds */
3067 
3068 }
3069 
3070 /*
3071  * Try and disable all interrupts. We just assign interrupts to other
3072  * processors based on policy. If any were bound by user request, we
3073  * let them continue and return failure. We do not bother to check
3074  * for cache affinity while rebinding.
3075  */
3076 
3077 static int
3078 apic_disable_intr(processorid_t cpun)
3079 {
3080 	int bind_cpu = 0, i, hardbound = 0, iflag;
3081 	apic_irq_t *irq_ptr;
3082 
3083 	iflag = intr_clear();
3084 	lock_set(&apic_ioapic_lock);
3085 
3086 	for (i = 0; i <= APIC_MAX_VECTOR; i++) {
3087 		if (apic_reprogram_info[i].done == B_FALSE) {
3088 			if (apic_reprogram_info[i].bindcpu == cpun) {
3089 				/*
3090 				 * CPU is busy -- it's the target of
3091 				 * a pending reprogramming attempt
3092 				 */
3093 				lock_clear(&apic_ioapic_lock);
3094 				intr_restore(iflag);
3095 				return (PSM_FAILURE);
3096 			}
3097 		}
3098 	}
3099 
3100 	apic_cpus[cpun].aci_status &= ~APIC_CPU_INTR_ENABLE;
3101 
3102 	apic_cpus[cpun].aci_curipl = 0;
3103 
3104 	i = apic_min_device_irq;
3105 	for (; i <= apic_max_device_irq; i++) {
3106 		/*
3107 		 * If there are bound interrupts on this cpu, then
3108 		 * rebind them to other processors.
3109 		 */
3110 		if ((irq_ptr = apic_irq_table[i]) != NULL) {
3111 			ASSERT((irq_ptr->airq_temp_cpu == IRQ_UNBOUND) ||
3112 			    (irq_ptr->airq_temp_cpu == IRQ_UNINIT) ||
3113 			    ((irq_ptr->airq_temp_cpu & ~IRQ_USER_BOUND) <
3114 			    apic_nproc));
3115 
3116 			if (irq_ptr->airq_temp_cpu == (cpun | IRQ_USER_BOUND)) {
3117 				hardbound = 1;
3118 				continue;
3119 			}
3120 
3121 			if (irq_ptr->airq_temp_cpu == cpun) {
3122 				do {
3123 					apic_next_bind_cpu += 2;
3124 					bind_cpu = apic_next_bind_cpu / 2;
3125 					if (bind_cpu >= apic_nproc) {
3126 						apic_next_bind_cpu = 1;
3127 						bind_cpu = 0;
3128 
3129 					}
3130 				} while (apic_rebind_all(irq_ptr, bind_cpu));
3131 			}
3132 		}
3133 	}
3134 
3135 	lock_clear(&apic_ioapic_lock);
3136 	intr_restore(iflag);
3137 
3138 	if (hardbound) {
3139 		cmn_err(CE_WARN, "Could not disable interrupts on %d"
3140 		    "due to user bound interrupts", cpun);
3141 		return (PSM_FAILURE);
3142 	}
3143 	else
3144 		return (PSM_SUCCESS);
3145 }
3146 
3147 static void
3148 apic_enable_intr(processorid_t cpun)
3149 {
3150 	int	i, iflag;
3151 	apic_irq_t *irq_ptr;
3152 
3153 	iflag = intr_clear();
3154 	lock_set(&apic_ioapic_lock);
3155 
3156 	apic_cpus[cpun].aci_status |= APIC_CPU_INTR_ENABLE;
3157 
3158 	i = apic_min_device_irq;
3159 	for (i = apic_min_device_irq; i <= apic_max_device_irq; i++) {
3160 		if ((irq_ptr = apic_irq_table[i]) != NULL) {
3161 			if ((irq_ptr->airq_cpu & ~IRQ_USER_BOUND) == cpun) {
3162 				(void) apic_rebind_all(irq_ptr,
3163 				    irq_ptr->airq_cpu);
3164 			}
3165 		}
3166 	}
3167 
3168 	lock_clear(&apic_ioapic_lock);
3169 	intr_restore(iflag);
3170 }
3171 
3172 /*
3173  * apic_introp_xlate() replaces apic_translate_irq() and is
3174  * called only from apic_intr_ops().  With the new ADII framework,
3175  * the priority can no longer be retrieved through i_ddi_get_intrspec().
3176  * It has to be passed in from the caller.
3177  */
3178 int
3179 apic_introp_xlate(dev_info_t *dip, struct intrspec *ispec, int type)
3180 {
3181 	char dev_type[16];
3182 	int dev_len, pci_irq, newirq, bustype, devid, busid, i;
3183 	int irqno = ispec->intrspec_vec;
3184 	ddi_acc_handle_t cfg_handle;
3185 	uchar_t ipin;
3186 	struct apic_io_intr *intrp;
3187 	iflag_t intr_flag;
3188 	APIC_HEADER	*hp;
3189 	MADT_INTERRUPT_OVERRIDE	*isop;
3190 	apic_irq_t *airqp;
3191 	int parent_is_pci_or_pciex = 0;
3192 	int child_is_pciex = 0;
3193 
3194 	DDI_INTR_IMPLDBG((CE_CONT, "apic_introp_xlate: dip=0x%p name=%s "
3195 	    "type=%d irqno=0x%x\n", (void *)dip, ddi_get_name(dip), type,
3196 	    irqno));
3197 
3198 	dev_len = sizeof (dev_type);
3199 	if (ddi_getlongprop_buf(DDI_DEV_T_ANY, ddi_get_parent(dip),
3200 	    DDI_PROP_DONTPASS, "device_type", (caddr_t)dev_type,
3201 	    &dev_len) == DDI_PROP_SUCCESS) {
3202 		if ((strcmp(dev_type, "pci") == 0) ||
3203 		    (strcmp(dev_type, "pciex") == 0))
3204 			parent_is_pci_or_pciex = 1;
3205 	}
3206 
3207 	if (parent_is_pci_or_pciex && ddi_prop_get_int(DDI_DEV_T_ANY, dip,
3208 	    DDI_PROP_DONTPASS, "pcie-capid-pointer", PCI_CAP_NEXT_PTR_NULL) !=
3209 	    PCI_CAP_NEXT_PTR_NULL) {
3210 		child_is_pciex = 1;
3211 	}
3212 
3213 	if (DDI_INTR_IS_MSI_OR_MSIX(type)) {
3214 		if ((airqp = apic_find_irq(dip, ispec, type)) != NULL) {
3215 			airqp->airq_iflag.bustype =
3216 			    child_is_pciex ? BUS_PCIE : BUS_PCI;
3217 			return (apic_vector_to_irq[airqp->airq_vector]);
3218 		}
3219 		return (apic_setup_irq_table(dip, irqno, NULL, ispec,
3220 		    NULL, type));
3221 	}
3222 
3223 	bustype = 0;
3224 
3225 	/* check if we have already translated this irq */
3226 	mutex_enter(&airq_mutex);
3227 	newirq = apic_min_device_irq;
3228 	for (; newirq <= apic_max_device_irq; newirq++) {
3229 		airqp = apic_irq_table[newirq];
3230 		while (airqp) {
3231 			if ((airqp->airq_dip == dip) &&
3232 			    (airqp->airq_origirq == irqno) &&
3233 			    (airqp->airq_mps_intr_index != FREE_INDEX)) {
3234 
3235 				mutex_exit(&airq_mutex);
3236 				return (VIRTIRQ(newirq, airqp->airq_share_id));
3237 			}
3238 			airqp = airqp->airq_next;
3239 		}
3240 	}
3241 	mutex_exit(&airq_mutex);
3242 
3243 	if (apic_defconf)
3244 		goto defconf;
3245 
3246 	if ((dip == NULL) || (!apic_irq_translate && !apic_enable_acpi))
3247 		goto nonpci;
3248 
3249 	if (parent_is_pci_or_pciex) {
3250 		/* pci device */
3251 		if (acpica_get_bdf(dip, &busid, &devid, NULL) != 0)
3252 			goto nonpci;
3253 		if (busid == 0 && apic_pci_bus_total == 1)
3254 			busid = (int)apic_single_pci_busid;
3255 
3256 		if (pci_config_setup(dip, &cfg_handle) != DDI_SUCCESS)
3257 			goto nonpci;
3258 		ipin = pci_config_get8(cfg_handle, PCI_CONF_IPIN) - PCI_INTA;
3259 		pci_config_teardown(&cfg_handle);
3260 		if (apic_enable_acpi && !apic_use_acpi_madt_only) {
3261 			if (apic_acpi_translate_pci_irq(dip, busid, devid,
3262 			    ipin, &pci_irq, &intr_flag) != ACPI_PSM_SUCCESS)
3263 				goto nonpci;
3264 
3265 			intr_flag.bustype = child_is_pciex ? BUS_PCIE : BUS_PCI;
3266 			if ((newirq = apic_setup_irq_table(dip, pci_irq, NULL,
3267 			    ispec, &intr_flag, type)) == -1)
3268 				goto nonpci;
3269 			return (newirq);
3270 		} else {
3271 			pci_irq = ((devid & 0x1f) << 2) | (ipin & 0x3);
3272 			if ((intrp = apic_find_io_intr_w_busid(pci_irq, busid))
3273 			    == NULL) {
3274 				if ((pci_irq = apic_handle_pci_pci_bridge(dip,
3275 				    devid, ipin, &intrp)) == -1)
3276 					goto nonpci;
3277 			}
3278 			if ((newirq = apic_setup_irq_table(dip, pci_irq, intrp,
3279 			    ispec, NULL, type)) == -1)
3280 				goto nonpci;
3281 			return (newirq);
3282 		}
3283 	} else if (strcmp(dev_type, "isa") == 0)
3284 		bustype = BUS_ISA;
3285 	else if (strcmp(dev_type, "eisa") == 0)
3286 		bustype = BUS_EISA;
3287 
3288 nonpci:
3289 	if (apic_enable_acpi && !apic_use_acpi_madt_only) {
3290 		/* search iso entries first */
3291 		if (acpi_iso_cnt != 0) {
3292 			hp = (APIC_HEADER *)acpi_isop;
3293 			i = 0;
3294 			while (i < acpi_iso_cnt) {
3295 				if (hp->Type == APIC_XRUPT_OVERRIDE) {
3296 					isop = (MADT_INTERRUPT_OVERRIDE *)hp;
3297 					if (isop->Bus == 0 &&
3298 					    isop->Source == irqno) {
3299 						newirq = isop->Interrupt;
3300 						intr_flag.intr_po =
3301 						    isop->Polarity;
3302 						intr_flag.intr_el =
3303 						    isop->TriggerMode;
3304 						intr_flag.bustype = BUS_ISA;
3305 
3306 						return (apic_setup_irq_table(
3307 						    dip, newirq, NULL, ispec,
3308 						    &intr_flag, type));
3309 
3310 					}
3311 					i++;
3312 				}
3313 				hp = (APIC_HEADER *)(((char *)hp) +
3314 				    hp->Length);
3315 			}
3316 		}
3317 		intr_flag.intr_po = INTR_PO_ACTIVE_HIGH;
3318 		intr_flag.intr_el = INTR_EL_EDGE;
3319 		intr_flag.bustype = BUS_ISA;
3320 		return (apic_setup_irq_table(dip, irqno, NULL, ispec,
3321 		    &intr_flag, type));
3322 	} else {
3323 		if (bustype == 0)
3324 			bustype = eisa_level_intr_mask ? BUS_EISA : BUS_ISA;
3325 		for (i = 0; i < 2; i++) {
3326 			if (((busid = apic_find_bus_id(bustype)) != -1) &&
3327 			    ((intrp = apic_find_io_intr_w_busid(irqno, busid))
3328 			    != NULL)) {
3329 				if ((newirq = apic_setup_irq_table(dip, irqno,
3330 				    intrp, ispec, NULL, type)) != -1) {
3331 					return (newirq);
3332 				}
3333 				goto defconf;
3334 			}
3335 			bustype = (bustype == BUS_EISA) ? BUS_ISA : BUS_EISA;
3336 		}
3337 	}
3338 
3339 /* MPS default configuration */
3340 defconf:
3341 	newirq = apic_setup_irq_table(dip, irqno, NULL, ispec, NULL, type);
3342 	if (newirq == -1)
3343 		return (newirq);
3344 	ASSERT(IRQINDEX(newirq) == irqno);
3345 	ASSERT(apic_irq_table[irqno]);
3346 	return (newirq);
3347 }
3348 
3349 
3350 
3351 
3352 
3353 
3354 /*
3355  * On machines with PCI-PCI bridges, a device behind a PCI-PCI bridge
3356  * needs special handling.  We may need to chase up the device tree,
3357  * using the PCI-PCI Bridge specification's "rotating IPIN assumptions",
3358  * to find the IPIN at the root bus that relates to the IPIN on the
3359  * subsidiary bus (for ACPI or MP).  We may, however, have an entry
3360  * in the MP table or the ACPI namespace for this device itself.
3361  * We handle both cases in the search below.
3362  */
3363 /* this is the non-acpi version */
3364 static int
3365 apic_handle_pci_pci_bridge(dev_info_t *idip, int child_devno, int child_ipin,
3366 			struct apic_io_intr **intrp)
3367 {
3368 	dev_info_t *dipp, *dip;
3369 	int pci_irq;
3370 	ddi_acc_handle_t cfg_handle;
3371 	int bridge_devno, bridge_bus;
3372 	int ipin;
3373 
3374 	dip = idip;
3375 
3376 	/*CONSTCOND*/
3377 	while (1) {
3378 		if ((dipp = ddi_get_parent(dip)) == (dev_info_t *)NULL)
3379 			return (-1);
3380 		if ((pci_config_setup(dipp, &cfg_handle) == DDI_SUCCESS) &&
3381 		    (pci_config_get8(cfg_handle, PCI_CONF_BASCLASS) ==
3382 		    PCI_CLASS_BRIDGE) && (pci_config_get8(cfg_handle,
3383 		    PCI_CONF_SUBCLASS) == PCI_BRIDGE_PCI)) {
3384 			pci_config_teardown(&cfg_handle);
3385 			if (acpica_get_bdf(dipp, &bridge_bus, &bridge_devno,
3386 			    NULL) != 0)
3387 				return (-1);
3388 			/*
3389 			 * This is the rotating scheme that Compaq is using
3390 			 * and documented in the pci to pci spec.  Also, if
3391 			 * the pci to pci bridge is behind another pci to
3392 			 * pci bridge, then it need to keep transversing
3393 			 * up until an interrupt entry is found or reach
3394 			 * the top of the tree
3395 			 */
3396 			ipin = (child_devno + child_ipin) % PCI_INTD;
3397 				if (bridge_bus == 0 && apic_pci_bus_total == 1)
3398 					bridge_bus = (int)apic_single_pci_busid;
3399 				pci_irq = ((bridge_devno & 0x1f) << 2) |
3400 				    (ipin & 0x3);
3401 				if ((*intrp = apic_find_io_intr_w_busid(pci_irq,
3402 				    bridge_bus)) != NULL) {
3403 					return (pci_irq);
3404 				}
3405 			dip = dipp;
3406 			child_devno = bridge_devno;
3407 			child_ipin = ipin;
3408 		} else
3409 			return (-1);
3410 	}
3411 	/*LINTED: function will not fall off the bottom */
3412 }
3413 
3414 
3415 
3416 
3417 static uchar_t
3418 acpi_find_ioapic(int irq)
3419 {
3420 	int i;
3421 
3422 	for (i = 0; i < apic_io_max; i++) {
3423 		if (irq >= apic_io_vectbase[i] && irq <= apic_io_vectend[i])
3424 			return (i);
3425 	}
3426 	return (0xFF);	/* shouldn't happen */
3427 }
3428 
3429 /*
3430  * See if two irqs are compatible for sharing a vector.
3431  * Currently we only support sharing of PCI devices.
3432  */
3433 static int
3434 acpi_intr_compatible(iflag_t iflag1, iflag_t iflag2)
3435 {
3436 	uint_t	level1, po1;
3437 	uint_t	level2, po2;
3438 
3439 	/* Assume active high by default */
3440 	po1 = 0;
3441 	po2 = 0;
3442 
3443 	if (iflag1.bustype != iflag2.bustype || iflag1.bustype != BUS_PCI)
3444 		return (0);
3445 
3446 	if (iflag1.intr_el == INTR_EL_CONFORM)
3447 		level1 = AV_LEVEL;
3448 	else
3449 		level1 = (iflag1.intr_el == INTR_EL_LEVEL) ? AV_LEVEL : 0;
3450 
3451 	if (level1 && ((iflag1.intr_po == INTR_PO_ACTIVE_LOW) ||
3452 	    (iflag1.intr_po == INTR_PO_CONFORM)))
3453 		po1 = AV_ACTIVE_LOW;
3454 
3455 	if (iflag2.intr_el == INTR_EL_CONFORM)
3456 		level2 = AV_LEVEL;
3457 	else
3458 		level2 = (iflag2.intr_el == INTR_EL_LEVEL) ? AV_LEVEL : 0;
3459 
3460 	if (level2 && ((iflag2.intr_po == INTR_PO_ACTIVE_LOW) ||
3461 	    (iflag2.intr_po == INTR_PO_CONFORM)))
3462 		po2 = AV_ACTIVE_LOW;
3463 
3464 	if ((level1 == level2) && (po1 == po2))
3465 		return (1);
3466 
3467 	return (0);
3468 }
3469 
3470 /*
3471  * Attempt to share vector with someone else
3472  */
3473 static int
3474 apic_share_vector(int irqno, iflag_t *intr_flagp, short intr_index, int ipl,
3475 	uchar_t ioapicindex, uchar_t ipin, apic_irq_t **irqptrp)
3476 {
3477 #ifdef DEBUG
3478 	apic_irq_t *tmpirqp = NULL;
3479 #endif /* DEBUG */
3480 	apic_irq_t *irqptr, dummyirq;
3481 	int	newirq, chosen_irq = -1, share = 127;
3482 	int	lowest, highest, i;
3483 	uchar_t	share_id;
3484 
3485 	DDI_INTR_IMPLDBG((CE_CONT, "apic_share_vector: irqno=0x%x "
3486 	    "intr_index=0x%x ipl=0x%x\n", irqno, intr_index, ipl));
3487 
3488 	highest = apic_ipltopri[ipl] + APIC_VECTOR_MASK;
3489 	lowest = apic_ipltopri[ipl-1] + APIC_VECTOR_PER_IPL;
3490 
3491 	if (highest < lowest) /* Both ipl and ipl-1 map to same pri */
3492 		lowest -= APIC_VECTOR_PER_IPL;
3493 	dummyirq.airq_mps_intr_index = intr_index;
3494 	dummyirq.airq_ioapicindex = ioapicindex;
3495 	dummyirq.airq_intin_no = ipin;
3496 	if (intr_flagp)
3497 		dummyirq.airq_iflag = *intr_flagp;
3498 	apic_record_rdt_entry(&dummyirq, irqno);
3499 	for (i = lowest; i <= highest; i++) {
3500 		newirq = apic_vector_to_irq[i];
3501 		if (newirq == APIC_RESV_IRQ)
3502 			continue;
3503 		irqptr = apic_irq_table[newirq];
3504 
3505 		if ((dummyirq.airq_rdt_entry & 0xFF00) !=
3506 		    (irqptr->airq_rdt_entry & 0xFF00))
3507 			/* not compatible */
3508 			continue;
3509 
3510 		if (irqptr->airq_share < share) {
3511 			share = irqptr->airq_share;
3512 			chosen_irq = newirq;
3513 		}
3514 	}
3515 	if (chosen_irq != -1) {
3516 		/*
3517 		 * Assign a share id which is free or which is larger
3518 		 * than the largest one.
3519 		 */
3520 		share_id = 1;
3521 		mutex_enter(&airq_mutex);
3522 		irqptr = apic_irq_table[chosen_irq];
3523 		while (irqptr) {
3524 			if (irqptr->airq_mps_intr_index == FREE_INDEX) {
3525 				share_id = irqptr->airq_share_id;
3526 				break;
3527 			}
3528 			if (share_id <= irqptr->airq_share_id)
3529 				share_id = irqptr->airq_share_id + 1;
3530 #ifdef DEBUG
3531 			tmpirqp = irqptr;
3532 #endif /* DEBUG */
3533 			irqptr = irqptr->airq_next;
3534 		}
3535 		if (!irqptr) {
3536 			irqptr = kmem_zalloc(sizeof (apic_irq_t), KM_SLEEP);
3537 			irqptr->airq_temp_cpu = IRQ_UNINIT;
3538 			irqptr->airq_next =
3539 			    apic_irq_table[chosen_irq]->airq_next;
3540 			apic_irq_table[chosen_irq]->airq_next = irqptr;
3541 #ifdef	DEBUG
3542 			tmpirqp = apic_irq_table[chosen_irq];
3543 #endif /* DEBUG */
3544 		}
3545 		irqptr->airq_mps_intr_index = intr_index;
3546 		irqptr->airq_ioapicindex = ioapicindex;
3547 		irqptr->airq_intin_no = ipin;
3548 		if (intr_flagp)
3549 			irqptr->airq_iflag = *intr_flagp;
3550 		irqptr->airq_vector = apic_irq_table[chosen_irq]->airq_vector;
3551 		irqptr->airq_share_id = share_id;
3552 		apic_record_rdt_entry(irqptr, irqno);
3553 		*irqptrp = irqptr;
3554 #ifdef	DEBUG
3555 		/* shuffle the pointers to test apic_delspl path */
3556 		if (tmpirqp) {
3557 			tmpirqp->airq_next = irqptr->airq_next;
3558 			irqptr->airq_next = apic_irq_table[chosen_irq];
3559 			apic_irq_table[chosen_irq] = irqptr;
3560 		}
3561 #endif /* DEBUG */
3562 		mutex_exit(&airq_mutex);
3563 		return (VIRTIRQ(chosen_irq, share_id));
3564 	}
3565 	return (-1);
3566 }
3567 
3568 /*
3569  *
3570  */
3571 static int
3572 apic_setup_irq_table(dev_info_t *dip, int irqno, struct apic_io_intr *intrp,
3573     struct intrspec *ispec, iflag_t *intr_flagp, int type)
3574 {
3575 	int origirq = ispec->intrspec_vec;
3576 	uchar_t ipl = ispec->intrspec_pri;
3577 	int	newirq, intr_index;
3578 	uchar_t	ipin, ioapic, ioapicindex, vector;
3579 	apic_irq_t *irqptr;
3580 	major_t	major;
3581 	dev_info_t	*sdip;
3582 
3583 	DDI_INTR_IMPLDBG((CE_CONT, "apic_setup_irq_table: dip=0x%p type=%d "
3584 	    "irqno=0x%x origirq=0x%x\n", (void *)dip, type, irqno, origirq));
3585 
3586 	ASSERT(ispec != NULL);
3587 
3588 	major =  (dip != NULL) ? ddi_name_to_major(ddi_get_name(dip)) : 0;
3589 
3590 	if (DDI_INTR_IS_MSI_OR_MSIX(type)) {
3591 		/* MSI/X doesn't need to setup ioapic stuffs */
3592 		ioapicindex = 0xff;
3593 		ioapic = 0xff;
3594 		ipin = (uchar_t)0xff;
3595 		intr_index = (type == DDI_INTR_TYPE_MSI) ? MSI_INDEX :
3596 		    MSIX_INDEX;
3597 		mutex_enter(&airq_mutex);
3598 		if ((irqno = apic_allocate_irq(apic_first_avail_irq)) == -1) {
3599 			mutex_exit(&airq_mutex);
3600 			/* need an irq for MSI/X to index into autovect[] */
3601 			cmn_err(CE_WARN, "No interrupt irq: %s instance %d",
3602 			    ddi_get_name(dip), ddi_get_instance(dip));
3603 			return (-1);
3604 		}
3605 		mutex_exit(&airq_mutex);
3606 
3607 	} else if (intrp != NULL) {
3608 		intr_index = (int)(intrp - apic_io_intrp);
3609 		ioapic = intrp->intr_destid;
3610 		ipin = intrp->intr_destintin;
3611 		/* Find ioapicindex. If destid was ALL, we will exit with 0. */
3612 		for (ioapicindex = apic_io_max - 1; ioapicindex; ioapicindex--)
3613 			if (apic_io_id[ioapicindex] == ioapic)
3614 				break;
3615 		ASSERT((ioapic == apic_io_id[ioapicindex]) ||
3616 		    (ioapic == INTR_ALL_APIC));
3617 
3618 		/* check whether this intin# has been used by another irqno */
3619 		if ((newirq = apic_find_intin(ioapicindex, ipin)) != -1) {
3620 			return (newirq);
3621 		}
3622 
3623 	} else if (intr_flagp != NULL) {
3624 		/* ACPI case */
3625 		intr_index = ACPI_INDEX;
3626 		ioapicindex = acpi_find_ioapic(irqno);
3627 		ASSERT(ioapicindex != 0xFF);
3628 		ioapic = apic_io_id[ioapicindex];
3629 		ipin = irqno - apic_io_vectbase[ioapicindex];
3630 		if (apic_irq_table[irqno] &&
3631 		    apic_irq_table[irqno]->airq_mps_intr_index == ACPI_INDEX) {
3632 			ASSERT(apic_irq_table[irqno]->airq_intin_no == ipin &&
3633 			    apic_irq_table[irqno]->airq_ioapicindex ==
3634 			    ioapicindex);
3635 			return (irqno);
3636 		}
3637 
3638 	} else {
3639 		/* default configuration */
3640 		ioapicindex = 0;
3641 		ioapic = apic_io_id[ioapicindex];
3642 		ipin = (uchar_t)irqno;
3643 		intr_index = DEFAULT_INDEX;
3644 	}
3645 
3646 	if (ispec == NULL) {
3647 		APIC_VERBOSE_IOAPIC((CE_WARN, "No intrspec for irqno = %x\n",
3648 		    irqno));
3649 	} else if ((vector = apic_allocate_vector(ipl, irqno, 0)) == 0) {
3650 		if ((newirq = apic_share_vector(irqno, intr_flagp, intr_index,
3651 		    ipl, ioapicindex, ipin, &irqptr)) != -1) {
3652 			irqptr->airq_ipl = ipl;
3653 			irqptr->airq_origirq = (uchar_t)origirq;
3654 			irqptr->airq_dip = dip;
3655 			irqptr->airq_major = major;
3656 			sdip = apic_irq_table[IRQINDEX(newirq)]->airq_dip;
3657 			/* This is OK to do really */
3658 			if (sdip == NULL) {
3659 				cmn_err(CE_WARN, "Sharing vectors: %s"
3660 				    " instance %d and SCI",
3661 				    ddi_get_name(dip), ddi_get_instance(dip));
3662 			} else {
3663 				cmn_err(CE_WARN, "Sharing vectors: %s"
3664 				    " instance %d and %s instance %d",
3665 				    ddi_get_name(sdip), ddi_get_instance(sdip),
3666 				    ddi_get_name(dip), ddi_get_instance(dip));
3667 			}
3668 			return (newirq);
3669 		}
3670 		/* try high priority allocation now  that share has failed */
3671 		if ((vector = apic_allocate_vector(ipl, irqno, 1)) == 0) {
3672 			cmn_err(CE_WARN, "No interrupt vector: %s instance %d",
3673 			    ddi_get_name(dip), ddi_get_instance(dip));
3674 			return (-1);
3675 		}
3676 	}
3677 
3678 	mutex_enter(&airq_mutex);
3679 	if (apic_irq_table[irqno] == NULL) {
3680 		irqptr = kmem_zalloc(sizeof (apic_irq_t), KM_SLEEP);
3681 		irqptr->airq_temp_cpu = IRQ_UNINIT;
3682 		apic_irq_table[irqno] = irqptr;
3683 	} else {
3684 		irqptr = apic_irq_table[irqno];
3685 		if (irqptr->airq_mps_intr_index != FREE_INDEX) {
3686 			/*
3687 			 * The slot is used by another irqno, so allocate
3688 			 * a free irqno for this interrupt
3689 			 */
3690 			newirq = apic_allocate_irq(apic_first_avail_irq);
3691 			if (newirq == -1) {
3692 				mutex_exit(&airq_mutex);
3693 				return (-1);
3694 			}
3695 			irqno = newirq;
3696 			irqptr = apic_irq_table[irqno];
3697 			if (irqptr == NULL) {
3698 				irqptr = kmem_zalloc(sizeof (apic_irq_t),
3699 				    KM_SLEEP);
3700 				irqptr->airq_temp_cpu = IRQ_UNINIT;
3701 				apic_irq_table[irqno] = irqptr;
3702 			}
3703 			apic_modify_vector(vector, newirq);
3704 		}
3705 	}
3706 	apic_max_device_irq = max(irqno, apic_max_device_irq);
3707 	apic_min_device_irq = min(irqno, apic_min_device_irq);
3708 	mutex_exit(&airq_mutex);
3709 	irqptr->airq_ioapicindex = ioapicindex;
3710 	irqptr->airq_intin_no = ipin;
3711 	irqptr->airq_ipl = ipl;
3712 	irqptr->airq_vector = vector;
3713 	irqptr->airq_origirq = (uchar_t)origirq;
3714 	irqptr->airq_share_id = 0;
3715 	irqptr->airq_mps_intr_index = (short)intr_index;
3716 	irqptr->airq_dip = dip;
3717 	irqptr->airq_major = major;
3718 	irqptr->airq_cpu = apic_bind_intr(dip, irqno, ioapic, ipin);
3719 	if (intr_flagp)
3720 		irqptr->airq_iflag = *intr_flagp;
3721 
3722 	if (!DDI_INTR_IS_MSI_OR_MSIX(type)) {
3723 		/* setup I/O APIC entry for non-MSI/X interrupts */
3724 		apic_record_rdt_entry(irqptr, irqno);
3725 	}
3726 	return (irqno);
3727 }
3728 
3729 /*
3730  * return the cpu to which this intr should be bound.
3731  * Check properties or any other mechanism to see if user wants it
3732  * bound to a specific CPU. If so, return the cpu id with high bit set.
3733  * If not, use the policy to choose a cpu and return the id.
3734  */
3735 uchar_t
3736 apic_bind_intr(dev_info_t *dip, int irq, uchar_t ioapicid, uchar_t intin)
3737 {
3738 	int	instance, instno, prop_len, bind_cpu, count;
3739 	uint_t	i, rc;
3740 	uchar_t	cpu;
3741 	major_t	major;
3742 	char	*name, *drv_name, *prop_val, *cptr;
3743 	char	prop_name[32];
3744 
3745 
3746 	if (apic_intr_policy == INTR_LOWEST_PRIORITY)
3747 		return (IRQ_UNBOUND);
3748 
3749 	drv_name = NULL;
3750 	rc = DDI_PROP_NOT_FOUND;
3751 	major = (major_t)-1;
3752 	if (dip != NULL) {
3753 		name = ddi_get_name(dip);
3754 		major = ddi_name_to_major(name);
3755 		drv_name = ddi_major_to_name(major);
3756 		instance = ddi_get_instance(dip);
3757 		if (apic_intr_policy == INTR_ROUND_ROBIN_WITH_AFFINITY) {
3758 			i = apic_min_device_irq;
3759 			for (; i <= apic_max_device_irq; i++) {
3760 
3761 				if ((i == irq) || (apic_irq_table[i] == NULL) ||
3762 				    (apic_irq_table[i]->airq_mps_intr_index
3763 				    == FREE_INDEX))
3764 					continue;
3765 
3766 				if ((apic_irq_table[i]->airq_major == major) &&
3767 				    (!(apic_irq_table[i]->airq_cpu &
3768 				    IRQ_USER_BOUND))) {
3769 
3770 					cpu = apic_irq_table[i]->airq_cpu;
3771 
3772 					cmn_err(CE_CONT,
3773 					    "!pcplusmp: %s (%s) instance #%d "
3774 					    "vector 0x%x ioapic 0x%x "
3775 					    "intin 0x%x is bound to cpu %d\n",
3776 					    name, drv_name, instance, irq,
3777 					    ioapicid, intin, cpu);
3778 					return (cpu);
3779 				}
3780 			}
3781 		}
3782 		/*
3783 		 * search for "drvname"_intpt_bind_cpus property first, the
3784 		 * syntax of the property should be "a[,b,c,...]" where
3785 		 * instance 0 binds to cpu a, instance 1 binds to cpu b,
3786 		 * instance 3 binds to cpu c...
3787 		 * ddi_getlongprop() will search /option first, then /
3788 		 * if "drvname"_intpt_bind_cpus doesn't exist, then find
3789 		 * intpt_bind_cpus property.  The syntax is the same, and
3790 		 * it applies to all the devices if its "drvname" specific
3791 		 * property doesn't exist
3792 		 */
3793 		(void) strcpy(prop_name, drv_name);
3794 		(void) strcat(prop_name, "_intpt_bind_cpus");
3795 		rc = ddi_getlongprop(DDI_DEV_T_ANY, dip, 0, prop_name,
3796 		    (caddr_t)&prop_val, &prop_len);
3797 		if (rc != DDI_PROP_SUCCESS) {
3798 			rc = ddi_getlongprop(DDI_DEV_T_ANY, dip, 0,
3799 			    "intpt_bind_cpus", (caddr_t)&prop_val, &prop_len);
3800 		}
3801 	}
3802 	if (rc == DDI_PROP_SUCCESS) {
3803 		for (i = count = 0; i < (prop_len - 1); i++)
3804 			if (prop_val[i] == ',')
3805 				count++;
3806 		if (prop_val[i-1] != ',')
3807 			count++;
3808 		/*
3809 		 * if somehow the binding instances defined in the
3810 		 * property are not enough for this instno., then
3811 		 * reuse the pattern for the next instance until
3812 		 * it reaches the requested instno
3813 		 */
3814 		instno = instance % count;
3815 		i = 0;
3816 		cptr = prop_val;
3817 		while (i < instno)
3818 			if (*cptr++ == ',')
3819 				i++;
3820 		bind_cpu = stoi(&cptr);
3821 		kmem_free(prop_val, prop_len);
3822 		/* if specific cpu is bogus, then default to cpu 0 */
3823 		if (bind_cpu >= apic_nproc) {
3824 			cmn_err(CE_WARN, "pcplusmp: %s=%s: CPU %d not present",
3825 			    prop_name, prop_val, bind_cpu);
3826 			bind_cpu = 0;
3827 		} else {
3828 			/* indicate that we are bound at user request */
3829 			bind_cpu |= IRQ_USER_BOUND;
3830 		}
3831 		/*
3832 		 * no need to check apic_cpus[].aci_status, if specific cpu is
3833 		 * not up, then post_cpu_start will handle it.
3834 		 */
3835 	} else {
3836 		bind_cpu = apic_next_bind_cpu++;
3837 		if (bind_cpu >= apic_nproc) {
3838 			apic_next_bind_cpu = 1;
3839 			bind_cpu = 0;
3840 		}
3841 	}
3842 	if (drv_name != NULL)
3843 		cmn_err(CE_CONT, "!pcplusmp: %s (%s) instance %d "
3844 		    "vector 0x%x ioapic 0x%x intin 0x%x is bound to cpu %d\n",
3845 		    name, drv_name, instance,
3846 		    irq, ioapicid, intin, bind_cpu & ~IRQ_USER_BOUND);
3847 	else
3848 		cmn_err(CE_CONT, "!pcplusmp: "
3849 		    "vector 0x%x ioapic 0x%x intin 0x%x is bound to cpu %d\n",
3850 		    irq, ioapicid, intin, bind_cpu & ~IRQ_USER_BOUND);
3851 
3852 	return ((uchar_t)bind_cpu);
3853 }
3854 
3855 static struct apic_io_intr *
3856 apic_find_io_intr_w_busid(int irqno, int busid)
3857 {
3858 	struct	apic_io_intr	*intrp;
3859 
3860 	/*
3861 	 * It can have more than 1 entry with same source bus IRQ,
3862 	 * but unique with the source bus id
3863 	 */
3864 	intrp = apic_io_intrp;
3865 	if (intrp != NULL) {
3866 		while (intrp->intr_entry == APIC_IO_INTR_ENTRY) {
3867 			if (intrp->intr_irq == irqno &&
3868 			    intrp->intr_busid == busid &&
3869 			    intrp->intr_type == IO_INTR_INT)
3870 				return (intrp);
3871 			intrp++;
3872 		}
3873 	}
3874 	APIC_VERBOSE_IOAPIC((CE_NOTE, "Did not find io intr for irqno:"
3875 	    "busid %x:%x\n", irqno, busid));
3876 	return ((struct apic_io_intr *)NULL);
3877 }
3878 
3879 
3880 struct mps_bus_info {
3881 	char	*bus_name;
3882 	int	bus_id;
3883 } bus_info_array[] = {
3884 	"ISA ", BUS_ISA,
3885 	"PCI ", BUS_PCI,
3886 	"EISA ", BUS_EISA,
3887 	"XPRESS", BUS_XPRESS,
3888 	"PCMCIA", BUS_PCMCIA,
3889 	"VL ", BUS_VL,
3890 	"CBUS ", BUS_CBUS,
3891 	"CBUSII", BUS_CBUSII,
3892 	"FUTURE", BUS_FUTURE,
3893 	"INTERN", BUS_INTERN,
3894 	"MBI ", BUS_MBI,
3895 	"MBII ", BUS_MBII,
3896 	"MPI ", BUS_MPI,
3897 	"MPSA ", BUS_MPSA,
3898 	"NUBUS ", BUS_NUBUS,
3899 	"TC ", BUS_TC,
3900 	"VME ", BUS_VME,
3901 	"PCI-E ", BUS_PCIE
3902 };
3903 
3904 static int
3905 apic_find_bus_type(char *bus)
3906 {
3907 	int	i = 0;
3908 
3909 	for (; i < sizeof (bus_info_array)/sizeof (struct mps_bus_info); i++)
3910 		if (strncmp(bus, bus_info_array[i].bus_name,
3911 		    strlen(bus_info_array[i].bus_name)) == 0)
3912 			return (bus_info_array[i].bus_id);
3913 	APIC_VERBOSE_IOAPIC((CE_WARN, "Did not find bus type for bus %s", bus));
3914 	return (0);
3915 }
3916 
3917 static int
3918 apic_find_bus(int busid)
3919 {
3920 	struct	apic_bus	*busp;
3921 
3922 	busp = apic_busp;
3923 	while (busp->bus_entry == APIC_BUS_ENTRY) {
3924 		if (busp->bus_id == busid)
3925 			return (apic_find_bus_type((char *)&busp->bus_str1));
3926 		busp++;
3927 	}
3928 	APIC_VERBOSE_IOAPIC((CE_WARN, "Did not find bus for bus id %x", busid));
3929 	return (0);
3930 }
3931 
3932 static int
3933 apic_find_bus_id(int bustype)
3934 {
3935 	struct	apic_bus	*busp;
3936 
3937 	busp = apic_busp;
3938 	while (busp->bus_entry == APIC_BUS_ENTRY) {
3939 		if (apic_find_bus_type((char *)&busp->bus_str1) == bustype)
3940 			return (busp->bus_id);
3941 		busp++;
3942 	}
3943 	APIC_VERBOSE_IOAPIC((CE_WARN, "Did not find bus id for bustype %x",
3944 	    bustype));
3945 	return (-1);
3946 }
3947 
3948 /*
3949  * Check if a particular irq need to be reserved for any io_intr
3950  */
3951 static struct apic_io_intr *
3952 apic_find_io_intr(int irqno)
3953 {
3954 	struct	apic_io_intr	*intrp;
3955 
3956 	intrp = apic_io_intrp;
3957 	if (intrp != NULL) {
3958 		while (intrp->intr_entry == APIC_IO_INTR_ENTRY) {
3959 			if (intrp->intr_irq == irqno &&
3960 			    intrp->intr_type == IO_INTR_INT)
3961 				return (intrp);
3962 			intrp++;
3963 		}
3964 	}
3965 	return ((struct apic_io_intr *)NULL);
3966 }
3967 
3968 /*
3969  * Check if the given ioapicindex intin combination has already been assigned
3970  * an irq. If so return irqno. Else -1
3971  */
3972 static int
3973 apic_find_intin(uchar_t ioapic, uchar_t intin)
3974 {
3975 	apic_irq_t *irqptr;
3976 	int	i;
3977 
3978 	/* find ioapic and intin in the apic_irq_table[] and return the index */
3979 	for (i = apic_min_device_irq; i <= apic_max_device_irq; i++) {
3980 		irqptr = apic_irq_table[i];
3981 		while (irqptr) {
3982 			if ((irqptr->airq_mps_intr_index >= 0) &&
3983 			    (irqptr->airq_intin_no == intin) &&
3984 			    (irqptr->airq_ioapicindex == ioapic)) {
3985 				APIC_VERBOSE_IOAPIC((CE_NOTE, "!Found irq "
3986 				    "entry for ioapic:intin %x:%x "
3987 				    "shared interrupts ?", ioapic, intin));
3988 				return (i);
3989 			}
3990 			irqptr = irqptr->airq_next;
3991 		}
3992 	}
3993 	return (-1);
3994 }
3995 
3996 int
3997 apic_allocate_irq(int irq)
3998 {
3999 	int	freeirq, i;
4000 
4001 	if ((freeirq = apic_find_free_irq(irq, (APIC_RESV_IRQ - 1))) == -1)
4002 		if ((freeirq = apic_find_free_irq(APIC_FIRST_FREE_IRQ,
4003 		    (irq - 1))) == -1) {
4004 			/*
4005 			 * if BIOS really defines every single irq in the mps
4006 			 * table, then don't worry about conflicting with
4007 			 * them, just use any free slot in apic_irq_table
4008 			 */
4009 			for (i = APIC_FIRST_FREE_IRQ; i < APIC_RESV_IRQ; i++) {
4010 				if ((apic_irq_table[i] == NULL) ||
4011 				    apic_irq_table[i]->airq_mps_intr_index ==
4012 				    FREE_INDEX) {
4013 				freeirq = i;
4014 				break;
4015 			}
4016 		}
4017 		if (freeirq == -1) {
4018 			/* This shouldn't happen, but just in case */
4019 			cmn_err(CE_WARN, "pcplusmp: NO available IRQ");
4020 			return (-1);
4021 		}
4022 	}
4023 	if (apic_irq_table[freeirq] == NULL) {
4024 		apic_irq_table[freeirq] =
4025 		    kmem_zalloc(sizeof (apic_irq_t), KM_NOSLEEP);
4026 		if (apic_irq_table[freeirq] == NULL) {
4027 			cmn_err(CE_WARN, "pcplusmp: NO memory to allocate IRQ");
4028 			return (-1);
4029 		}
4030 		apic_irq_table[freeirq]->airq_mps_intr_index = FREE_INDEX;
4031 	}
4032 	return (freeirq);
4033 }
4034 
4035 static int
4036 apic_find_free_irq(int start, int end)
4037 {
4038 	int	i;
4039 
4040 	for (i = start; i <= end; i++)
4041 		/* Check if any I/O entry needs this IRQ */
4042 		if (apic_find_io_intr(i) == NULL) {
4043 			/* Then see if it is free */
4044 			if ((apic_irq_table[i] == NULL) ||
4045 			    (apic_irq_table[i]->airq_mps_intr_index ==
4046 			    FREE_INDEX)) {
4047 				return (i);
4048 			}
4049 		}
4050 	return (-1);
4051 }
4052 
4053 /*
4054  * Allocate a free vector for irq at ipl. Takes care of merging of multiple
4055  * IPLs into a single APIC level as well as stretching some IPLs onto multiple
4056  * levels. APIC_HI_PRI_VECTS interrupts are reserved for high priority
4057  * requests and allocated only when pri is set.
4058  */
4059 static uchar_t
4060 apic_allocate_vector(int ipl, int irq, int pri)
4061 {
4062 	int	lowest, highest, i;
4063 
4064 	highest = apic_ipltopri[ipl] + APIC_VECTOR_MASK;
4065 	lowest = apic_ipltopri[ipl - 1] + APIC_VECTOR_PER_IPL;
4066 
4067 	if (highest < lowest) /* Both ipl and ipl - 1 map to same pri */
4068 		lowest -= APIC_VECTOR_PER_IPL;
4069 
4070 #ifdef	DEBUG
4071 	if (apic_restrict_vector)	/* for testing shared interrupt logic */
4072 		highest = lowest + apic_restrict_vector + APIC_HI_PRI_VECTS;
4073 #endif /* DEBUG */
4074 	if (pri == 0)
4075 		highest -= APIC_HI_PRI_VECTS;
4076 
4077 	for (i = lowest; i < highest; i++) {
4078 		if (APIC_CHECK_RESERVE_VECTORS(i))
4079 			continue;
4080 		if (apic_vector_to_irq[i] == APIC_RESV_IRQ) {
4081 			apic_vector_to_irq[i] = (uchar_t)irq;
4082 			return (i);
4083 		}
4084 	}
4085 
4086 	return (0);
4087 }
4088 
4089 static void
4090 apic_modify_vector(uchar_t vector, int irq)
4091 {
4092 	apic_vector_to_irq[vector] = (uchar_t)irq;
4093 }
4094 
4095 /*
4096  * Mark vector as being in the process of being deleted. Interrupts
4097  * may still come in on some CPU. The moment an interrupt comes with
4098  * the new vector, we know we can free the old one. Called only from
4099  * addspl and delspl with interrupts disabled. Because an interrupt
4100  * can be shared, but no interrupt from either device may come in,
4101  * we also use a timeout mechanism, which we arbitrarily set to
4102  * apic_revector_timeout microseconds.
4103  */
4104 static void
4105 apic_mark_vector(uchar_t oldvector, uchar_t newvector)
4106 {
4107 	int iflag = intr_clear();
4108 	lock_set(&apic_revector_lock);
4109 	if (!apic_oldvec_to_newvec) {
4110 		apic_oldvec_to_newvec =
4111 		    kmem_zalloc(sizeof (newvector) * APIC_MAX_VECTOR * 2,
4112 		    KM_NOSLEEP);
4113 
4114 		if (!apic_oldvec_to_newvec) {
4115 			/*
4116 			 * This failure is not catastrophic.
4117 			 * But, the oldvec will never be freed.
4118 			 */
4119 			apic_error |= APIC_ERR_MARK_VECTOR_FAIL;
4120 			lock_clear(&apic_revector_lock);
4121 			intr_restore(iflag);
4122 			return;
4123 		}
4124 		apic_newvec_to_oldvec = &apic_oldvec_to_newvec[APIC_MAX_VECTOR];
4125 	}
4126 
4127 	/* See if we already did this for drivers which do double addintrs */
4128 	if (apic_oldvec_to_newvec[oldvector] != newvector) {
4129 		apic_oldvec_to_newvec[oldvector] = newvector;
4130 		apic_newvec_to_oldvec[newvector] = oldvector;
4131 		apic_revector_pending++;
4132 	}
4133 	lock_clear(&apic_revector_lock);
4134 	intr_restore(iflag);
4135 	(void) timeout(apic_xlate_vector_free_timeout_handler,
4136 	    (void *)(uintptr_t)oldvector, drv_usectohz(apic_revector_timeout));
4137 }
4138 
4139 /*
4140  * xlate_vector is called from intr_enter if revector_pending is set.
4141  * It will xlate it if needed and mark the old vector as free.
4142  */
4143 static uchar_t
4144 apic_xlate_vector(uchar_t vector)
4145 {
4146 	uchar_t	newvector, oldvector = 0;
4147 
4148 	lock_set(&apic_revector_lock);
4149 	/* Do we really need to do this ? */
4150 	if (!apic_revector_pending) {
4151 		lock_clear(&apic_revector_lock);
4152 		return (vector);
4153 	}
4154 	if ((newvector = apic_oldvec_to_newvec[vector]) != 0)
4155 		oldvector = vector;
4156 	else {
4157 		/*
4158 		 * The incoming vector is new . See if a stale entry is
4159 		 * remaining
4160 		 */
4161 		if ((oldvector = apic_newvec_to_oldvec[vector]) != 0)
4162 			newvector = vector;
4163 	}
4164 
4165 	if (oldvector) {
4166 		apic_revector_pending--;
4167 		apic_oldvec_to_newvec[oldvector] = 0;
4168 		apic_newvec_to_oldvec[newvector] = 0;
4169 		apic_free_vector(oldvector);
4170 		lock_clear(&apic_revector_lock);
4171 		/* There could have been more than one reprogramming! */
4172 		return (apic_xlate_vector(newvector));
4173 	}
4174 	lock_clear(&apic_revector_lock);
4175 	return (vector);
4176 }
4177 
4178 void
4179 apic_xlate_vector_free_timeout_handler(void *arg)
4180 {
4181 	int iflag;
4182 	uchar_t oldvector, newvector;
4183 
4184 	oldvector = (uchar_t)(uintptr_t)arg;
4185 	iflag = intr_clear();
4186 	lock_set(&apic_revector_lock);
4187 	if ((newvector = apic_oldvec_to_newvec[oldvector]) != 0) {
4188 		apic_free_vector(oldvector);
4189 		apic_oldvec_to_newvec[oldvector] = 0;
4190 		apic_newvec_to_oldvec[newvector] = 0;
4191 		apic_revector_pending--;
4192 	}
4193 
4194 	lock_clear(&apic_revector_lock);
4195 	intr_restore(iflag);
4196 }
4197 
4198 
4199 /* Mark vector as not being used by any irq */
4200 static void
4201 apic_free_vector(uchar_t vector)
4202 {
4203 	apic_vector_to_irq[vector] = APIC_RESV_IRQ;
4204 }
4205 
4206 /*
4207  * compute the polarity, trigger mode and vector for programming into
4208  * the I/O apic and record in airq_rdt_entry.
4209  */
4210 static void
4211 apic_record_rdt_entry(apic_irq_t *irqptr, int irq)
4212 {
4213 	int	ioapicindex, bus_type, vector;
4214 	short	intr_index;
4215 	uint_t	level, po, io_po;
4216 	struct apic_io_intr *iointrp;
4217 
4218 	intr_index = irqptr->airq_mps_intr_index;
4219 	DDI_INTR_IMPLDBG((CE_CONT, "apic_record_rdt_entry: intr_index=%d "
4220 	    "irq = 0x%x dip = 0x%p vector = 0x%x\n", intr_index, irq,
4221 	    (void *)irqptr->airq_dip, irqptr->airq_vector));
4222 
4223 	if (intr_index == RESERVE_INDEX) {
4224 		apic_error |= APIC_ERR_INVALID_INDEX;
4225 		return;
4226 	} else if (APIC_IS_MSI_OR_MSIX_INDEX(intr_index)) {
4227 		return;
4228 	}
4229 
4230 	vector = irqptr->airq_vector;
4231 	ioapicindex = irqptr->airq_ioapicindex;
4232 	/* Assume edge triggered by default */
4233 	level = 0;
4234 	/* Assume active high by default */
4235 	po = 0;
4236 
4237 	if (intr_index == DEFAULT_INDEX || intr_index == FREE_INDEX) {
4238 		ASSERT(irq < 16);
4239 		if (eisa_level_intr_mask & (1 << irq))
4240 			level = AV_LEVEL;
4241 		if (intr_index == FREE_INDEX && apic_defconf == 0)
4242 			apic_error |= APIC_ERR_INVALID_INDEX;
4243 	} else if (intr_index == ACPI_INDEX) {
4244 		bus_type = irqptr->airq_iflag.bustype;
4245 		if (irqptr->airq_iflag.intr_el == INTR_EL_CONFORM) {
4246 			if (bus_type == BUS_PCI)
4247 				level = AV_LEVEL;
4248 		} else
4249 			level = (irqptr->airq_iflag.intr_el == INTR_EL_LEVEL) ?
4250 			    AV_LEVEL : 0;
4251 		if (level &&
4252 		    ((irqptr->airq_iflag.intr_po == INTR_PO_ACTIVE_LOW) ||
4253 		    (irqptr->airq_iflag.intr_po == INTR_PO_CONFORM &&
4254 		    bus_type == BUS_PCI)))
4255 			po = AV_ACTIVE_LOW;
4256 	} else {
4257 		iointrp = apic_io_intrp + intr_index;
4258 		bus_type = apic_find_bus(iointrp->intr_busid);
4259 		if (iointrp->intr_el == INTR_EL_CONFORM) {
4260 			if ((irq < 16) && (eisa_level_intr_mask & (1 << irq)))
4261 				level = AV_LEVEL;
4262 			else if (bus_type == BUS_PCI)
4263 				level = AV_LEVEL;
4264 		} else
4265 			level = (iointrp->intr_el == INTR_EL_LEVEL) ?
4266 			    AV_LEVEL : 0;
4267 		if (level && ((iointrp->intr_po == INTR_PO_ACTIVE_LOW) ||
4268 		    (iointrp->intr_po == INTR_PO_CONFORM &&
4269 		    bus_type == BUS_PCI)))
4270 			po = AV_ACTIVE_LOW;
4271 	}
4272 	if (level)
4273 		apic_level_intr[irq] = 1;
4274 	/*
4275 	 * The 82489DX External APIC cannot do active low polarity interrupts.
4276 	 */
4277 	if (po && (apic_io_ver[ioapicindex] != IOAPIC_VER_82489DX))
4278 		io_po = po;
4279 	else
4280 		io_po = 0;
4281 
4282 	if (apic_verbose & APIC_VERBOSE_IOAPIC_FLAG)
4283 		printf("setio: ioapic=%x intin=%x level=%x po=%x vector=%x\n",
4284 		    ioapicindex, irqptr->airq_intin_no, level, io_po, vector);
4285 
4286 	irqptr->airq_rdt_entry = level|io_po|vector;
4287 }
4288 
4289 static processorid_t
4290 apic_find_cpu(int flag)
4291 {
4292 	processorid_t acid = 0;
4293 	int i;
4294 
4295 	/* Find the first CPU with the passed-in flag set */
4296 	for (i = 0; i < apic_nproc; i++) {
4297 		if (apic_cpus[i].aci_status & flag) {
4298 			acid = i;
4299 			break;
4300 		}
4301 	}
4302 
4303 	ASSERT((apic_cpus[acid].aci_status & flag) != 0);
4304 	return (acid);
4305 }
4306 
4307 /*
4308  * Call rebind to do the actual programming.
4309  * Must be called with interrupts disabled and apic_ioapic_lock held
4310  * 'p' is polymorphic -- if this function is called to process a deferred
4311  * reprogramming, p is of type 'struct ioapic_reprogram_data *', from which
4312  * the irq pointer is retrieved.  If not doing deferred reprogramming,
4313  * p is of the type 'apic_irq_t *'.
4314  *
4315  * apic_ioapic_lock must be held across this call, as it protects apic_rebind
4316  * and it protects apic_find_cpu() from a race in which a CPU can be taken
4317  * offline after a cpu is selected, but before apic_rebind is called to
4318  * bind interrupts to it.
4319  */
4320 static int
4321 apic_setup_io_intr(void *p, int irq, boolean_t deferred)
4322 {
4323 	apic_irq_t *irqptr;
4324 	struct ioapic_reprogram_data *drep = NULL;
4325 	int rv;
4326 
4327 	if (deferred) {
4328 		drep = (struct ioapic_reprogram_data *)p;
4329 		ASSERT(drep != NULL);
4330 		irqptr = drep->irqp;
4331 	} else
4332 		irqptr = (apic_irq_t *)p;
4333 
4334 	ASSERT(irqptr != NULL);
4335 
4336 	rv = apic_rebind(irqptr, apic_irq_table[irq]->airq_cpu, drep);
4337 	if (rv) {
4338 		/*
4339 		 * CPU is not up or interrupts are disabled. Fall back to
4340 		 * the first available CPU
4341 		 */
4342 		rv = apic_rebind(irqptr, apic_find_cpu(APIC_CPU_INTR_ENABLE),
4343 		    drep);
4344 	}
4345 
4346 	return (rv);
4347 }
4348 
4349 /*
4350  * Bind interrupt corresponding to irq_ptr to bind_cpu.
4351  * Must be called with interrupts disabled and apic_ioapic_lock held
4352  */
4353 static int
4354 apic_rebind(apic_irq_t *irq_ptr, int bind_cpu,
4355     struct ioapic_reprogram_data *drep)
4356 {
4357 	int			ioapicindex, intin_no;
4358 	volatile int32_t	*ioapic;
4359 	uchar_t			airq_temp_cpu;
4360 	apic_cpus_info_t	*cpu_infop;
4361 	uint32_t		rdt_entry;
4362 	int			which_irq;
4363 
4364 	which_irq = apic_vector_to_irq[irq_ptr->airq_vector];
4365 
4366 	intin_no = irq_ptr->airq_intin_no;
4367 	ioapicindex = irq_ptr->airq_ioapicindex;
4368 	ioapic = apicioadr[ioapicindex];
4369 	airq_temp_cpu = irq_ptr->airq_temp_cpu;
4370 	if (airq_temp_cpu != IRQ_UNINIT && airq_temp_cpu != IRQ_UNBOUND) {
4371 		if (airq_temp_cpu & IRQ_USER_BOUND)
4372 			/* Mask off high bit so it can be used as array index */
4373 			airq_temp_cpu &= ~IRQ_USER_BOUND;
4374 
4375 		ASSERT(airq_temp_cpu < apic_nproc);
4376 	}
4377 
4378 	/*
4379 	 * Can't bind to a CPU that's not accepting interrupts:
4380 	 */
4381 	cpu_infop = &apic_cpus[bind_cpu & ~IRQ_USER_BOUND];
4382 	if (!(cpu_infop->aci_status & APIC_CPU_INTR_ENABLE))
4383 		return (1);
4384 
4385 	/*
4386 	 * If we are about to change the interrupt vector for this interrupt,
4387 	 * and this interrupt is level-triggered, attached to an IOAPIC,
4388 	 * has been delivered to a CPU and that CPU has not handled it
4389 	 * yet, we cannot reprogram the IOAPIC now.
4390 	 */
4391 	if (!APIC_IS_MSI_OR_MSIX_INDEX(irq_ptr->airq_mps_intr_index)) {
4392 
4393 		rdt_entry = READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no);
4394 
4395 		if ((irq_ptr->airq_vector != RDT_VECTOR(rdt_entry)) &&
4396 		    apic_check_stuck_interrupt(irq_ptr, airq_temp_cpu,
4397 		    bind_cpu, ioapic, intin_no, which_irq, drep) != 0) {
4398 
4399 			return (0);
4400 		}
4401 	}
4402 
4403 	/*
4404 	 * NOTE: We do not unmask the RDT here, as an interrupt MAY still
4405 	 * come in before we have a chance to reprogram it below.  The
4406 	 * reprogramming below will simultaneously change and unmask the
4407 	 * RDT entry.
4408 	 */
4409 
4410 	if ((uchar_t)bind_cpu == IRQ_UNBOUND) {
4411 
4412 		rdt_entry = AV_LDEST | AV_LOPRI | irq_ptr->airq_rdt_entry;
4413 
4414 		/* Write the RDT entry -- no specific CPU binding */
4415 		WRITE_IOAPIC_RDT_ENTRY_HIGH_DWORD(ioapic, intin_no, AV_TOALL);
4416 
4417 		if (airq_temp_cpu != IRQ_UNINIT && airq_temp_cpu != IRQ_UNBOUND)
4418 			apic_cpus[airq_temp_cpu].aci_temp_bound--;
4419 
4420 		/* Write the vector, trigger, and polarity portion of the RDT */
4421 		WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no, rdt_entry);
4422 
4423 		irq_ptr->airq_temp_cpu = IRQ_UNBOUND;
4424 		return (0);
4425 	}
4426 
4427 	if (bind_cpu & IRQ_USER_BOUND) {
4428 		cpu_infop->aci_bound++;
4429 	} else {
4430 		cpu_infop->aci_temp_bound++;
4431 	}
4432 	ASSERT((bind_cpu & ~IRQ_USER_BOUND) < apic_nproc);
4433 	if (!APIC_IS_MSI_OR_MSIX_INDEX(irq_ptr->airq_mps_intr_index)) {
4434 		/* Write the RDT entry -- bind to a specific CPU: */
4435 		WRITE_IOAPIC_RDT_ENTRY_HIGH_DWORD(ioapic, intin_no,
4436 		    cpu_infop->aci_local_id << APIC_ID_BIT_OFFSET);
4437 	}
4438 	if ((airq_temp_cpu != IRQ_UNBOUND) && (airq_temp_cpu != IRQ_UNINIT)) {
4439 		apic_cpus[airq_temp_cpu].aci_temp_bound--;
4440 	}
4441 	if (!APIC_IS_MSI_OR_MSIX_INDEX(irq_ptr->airq_mps_intr_index)) {
4442 
4443 		rdt_entry = AV_PDEST | AV_FIXED | irq_ptr->airq_rdt_entry;
4444 
4445 		/* Write the vector, trigger, and polarity portion of the RDT */
4446 		WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no, rdt_entry);
4447 
4448 	} else {
4449 		int type = (irq_ptr->airq_mps_intr_index == MSI_INDEX) ?
4450 		    DDI_INTR_TYPE_MSI : DDI_INTR_TYPE_MSIX;
4451 		(void) apic_pci_msi_disable_mode(irq_ptr->airq_dip, type,
4452 		    ioapicindex);
4453 		if (ioapicindex == irq_ptr->airq_origirq) {
4454 			/* first one */
4455 			DDI_INTR_IMPLDBG((CE_CONT, "apic_rebind: call "
4456 			    "apic_pci_msi_enable_vector\n"));
4457 			if (apic_pci_msi_enable_vector(irq_ptr->airq_dip, type,
4458 			    which_irq, irq_ptr->airq_vector,
4459 			    irq_ptr->airq_intin_no,
4460 			    cpu_infop->aci_local_id) != PSM_SUCCESS) {
4461 				cmn_err(CE_WARN, "pcplusmp: "
4462 					"apic_pci_msi_enable_vector "
4463 					"returned PSM_FAILURE");
4464 			}
4465 		}
4466 		if ((ioapicindex + irq_ptr->airq_intin_no - 1) ==
4467 		    irq_ptr->airq_origirq) { /* last one */
4468 			DDI_INTR_IMPLDBG((CE_CONT, "apic_rebind: call "
4469 			    "pci_msi_enable_mode\n"));
4470 			if (apic_pci_msi_enable_mode(irq_ptr->airq_dip,
4471 			    type, which_irq) != PSM_SUCCESS) {
4472 				DDI_INTR_IMPLDBG((CE_CONT, "pcplusmp: "
4473 				    "pci_msi_enable failed\n"));
4474 				(void) apic_pci_msi_unconfigure(
4475 				    irq_ptr->airq_dip, type, which_irq);
4476 			}
4477 		}
4478 	}
4479 	irq_ptr->airq_temp_cpu = (uchar_t)bind_cpu;
4480 	apic_redist_cpu_skip &= ~(1 << (bind_cpu & ~IRQ_USER_BOUND));
4481 	return (0);
4482 }
4483 
4484 static void
4485 apic_last_ditch_clear_remote_irr(volatile int32_t *ioapic, int intin_no)
4486 {
4487 	if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no)
4488 	    & AV_REMOTE_IRR) != 0) {
4489 		/*
4490 		 * Trying to clear the bit through normal
4491 		 * channels has failed.  So as a last-ditch
4492 		 * effort, try to set the trigger mode to
4493 		 * edge, then to level.  This has been
4494 		 * observed to work on many systems.
4495 		 */
4496 		WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic,
4497 		    intin_no,
4498 		    READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic,
4499 		    intin_no) & ~AV_LEVEL);
4500 
4501 		WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic,
4502 		    intin_no,
4503 		    READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic,
4504 		    intin_no) | AV_LEVEL);
4505 
4506 		/*
4507 		 * If the bit's STILL set, this interrupt may
4508 		 * be hosed.
4509 		 */
4510 		if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic,
4511 		    intin_no) & AV_REMOTE_IRR) != 0) {
4512 
4513 			prom_printf("pcplusmp: Remote IRR still "
4514 			    "not clear for IOAPIC %p intin %d.\n"
4515 			    "\tInterrupts to this pin may cease "
4516 			    "functioning.\n", ioapic, intin_no);
4517 #ifdef DEBUG
4518 			apic_last_ditch_reprogram_failures++;
4519 #endif
4520 		}
4521 	}
4522 }
4523 
4524 /*
4525  * This function is protected by apic_ioapic_lock coupled with the
4526  * fact that interrupts are disabled.
4527  */
4528 static void
4529 delete_defer_repro_ent(int which_irq)
4530 {
4531 	ASSERT(which_irq >= 0);
4532 	ASSERT(which_irq <= 255);
4533 
4534 	if (apic_reprogram_info[which_irq].done)
4535 		return;
4536 
4537 	apic_reprogram_info[which_irq].done = B_TRUE;
4538 
4539 #ifdef DEBUG
4540 	apic_defer_repro_total_retries +=
4541 	    apic_reprogram_info[which_irq].tries;
4542 
4543 	apic_defer_repro_successes++;
4544 #endif
4545 
4546 	if (--apic_reprogram_outstanding == 0) {
4547 
4548 		setlvlx = apic_intr_exit;
4549 	}
4550 }
4551 
4552 
4553 /*
4554  * Interrupts must be disabled during this function to prevent
4555  * self-deadlock.  Interrupts are disabled because this function
4556  * is called from apic_check_stuck_interrupt(), which is called
4557  * from apic_rebind(), which requires its caller to disable interrupts.
4558  */
4559 static void
4560 add_defer_repro_ent(apic_irq_t *irq_ptr, int which_irq, int new_bind_cpu)
4561 {
4562 	ASSERT(which_irq >= 0);
4563 	ASSERT(which_irq <= 255);
4564 
4565 	/*
4566 	 * On the off-chance that there's already a deferred
4567 	 * reprogramming on this irq, check, and if so, just update the
4568 	 * CPU and irq pointer to which the interrupt is targeted, then return.
4569 	 */
4570 	if (!apic_reprogram_info[which_irq].done) {
4571 		apic_reprogram_info[which_irq].bindcpu = new_bind_cpu;
4572 		apic_reprogram_info[which_irq].irqp = irq_ptr;
4573 		return;
4574 	}
4575 
4576 	apic_reprogram_info[which_irq].irqp = irq_ptr;
4577 	apic_reprogram_info[which_irq].bindcpu = new_bind_cpu;
4578 	apic_reprogram_info[which_irq].tries = 0;
4579 	/*
4580 	 * This must be the last thing set, since we're not
4581 	 * grabbing any locks, apic_try_deferred_reprogram() will
4582 	 * make its decision about using this entry iff done
4583 	 * is false.
4584 	 */
4585 	apic_reprogram_info[which_irq].done = B_FALSE;
4586 
4587 	/*
4588 	 * If there were previously no deferred reprogrammings, change
4589 	 * setlvlx to call apic_try_deferred_reprogram()
4590 	 */
4591 	if (++apic_reprogram_outstanding == 1) {
4592 
4593 		setlvlx = apic_try_deferred_reprogram;
4594 	}
4595 }
4596 
4597 static void
4598 apic_try_deferred_reprogram(int prev_ipl, int irq)
4599 {
4600 	int reproirq, iflag;
4601 	struct ioapic_reprogram_data *drep;
4602 
4603 	apic_intr_exit(prev_ipl, irq);
4604 
4605 	if (!lock_try(&apic_defer_reprogram_lock)) {
4606 		return;
4607 	}
4608 
4609 	/*
4610 	 * Acquire the apic_ioapic_lock so that any other operations that
4611 	 * may affect the apic_reprogram_info state are serialized.
4612 	 * It's still possible for the last deferred reprogramming to clear
4613 	 * between the time we entered this function and the time we get to
4614 	 * the for loop below.  In that case, *setlvlx will have been set
4615 	 * back to apic_intr_exit and drep will be NULL. (There's no way to
4616 	 * stop that from happening -- we would need to grab a lock before
4617 	 * calling *setlvlx, which is neither realistic nor prudent).
4618 	 */
4619 	iflag = intr_clear();
4620 	lock_set(&apic_ioapic_lock);
4621 
4622 	/*
4623 	 * For each deferred RDT entry, try to reprogram it now.  Note that
4624 	 * there is no lock acquisition to read apic_reprogram_info because
4625 	 * '.done' is set only after the other fields in the structure are set.
4626 	 */
4627 
4628 	drep = NULL;
4629 	for (reproirq = 0; reproirq <= APIC_MAX_VECTOR; reproirq++) {
4630 		if (apic_reprogram_info[reproirq].done == B_FALSE) {
4631 			drep = &apic_reprogram_info[reproirq];
4632 			break;
4633 		}
4634 	}
4635 
4636 	/*
4637 	 * Either we found a deferred action to perform, or
4638 	 * we entered this function spuriously, after *setlvlx
4639 	 * was restored to point to apic_intr_enter.  Any other
4640 	 * permutation is invalid.
4641 	 */
4642 	ASSERT(drep != NULL || *setlvlx == apic_intr_exit);
4643 
4644 	/*
4645 	 * Though we can't really do anything about errors
4646 	 * at this point, keep track of them for reporting.
4647 	 * Note that it is very possible for apic_setup_io_intr
4648 	 * to re-register this very timeout if the Remote IRR bit
4649 	 * has not yet cleared.
4650 	 */
4651 
4652 #ifdef DEBUG
4653 	if (drep != NULL) {
4654 		if (apic_setup_io_intr(drep, reproirq, B_TRUE) != 0) {
4655 			apic_deferred_setup_failures++;
4656 		}
4657 	} else {
4658 		apic_deferred_spurious_enters++;
4659 	}
4660 #else
4661 	if (drep != NULL)
4662 		(void) apic_setup_io_intr(drep, reproirq, B_TRUE);
4663 #endif
4664 
4665 	lock_clear(&apic_ioapic_lock);
4666 	intr_restore(iflag);
4667 
4668 	lock_clear(&apic_defer_reprogram_lock);
4669 }
4670 
4671 static void
4672 apic_ioapic_wait_pending_clear(volatile int32_t *ioapic, int intin_no)
4673 {
4674 	int waited;
4675 
4676 	/*
4677 	 * Wait for the delivery pending bit to clear.
4678 	 */
4679 	if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no) &
4680 	    (AV_LEVEL|AV_PENDING)) == (AV_LEVEL|AV_PENDING)) {
4681 
4682 		/*
4683 		 * If we're still waiting on the delivery of this interrupt,
4684 		 * continue to wait here until it is delivered (this should be
4685 		 * a very small amount of time, but include a timeout just in
4686 		 * case).
4687 		 */
4688 		for (waited = 0; waited < apic_max_reps_clear_pending;
4689 		    waited++) {
4690 			if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no)
4691 			    & AV_PENDING) == 0) {
4692 				break;
4693 			}
4694 		}
4695 	}
4696 }
4697 
4698 /*
4699  * Checks to see if the IOAPIC interrupt entry specified has its Remote IRR
4700  * bit set.  Calls functions that modify the function that setlvlx points to,
4701  * so that the reprogramming can be retried very shortly.
4702  *
4703  * This function will mask the RDT entry if the interrupt is level-triggered.
4704  * (The caller is responsible for unmasking the RDT entry.)
4705  *
4706  * Returns non-zero if the caller should defer IOAPIC reprogramming.
4707  */
4708 static int
4709 apic_check_stuck_interrupt(apic_irq_t *irq_ptr, int old_bind_cpu,
4710     int new_bind_cpu, volatile int32_t *ioapic, int intin_no, int which_irq,
4711     struct ioapic_reprogram_data *drep)
4712 {
4713 	int32_t			rdt_entry;
4714 	int			waited;
4715 	int			reps = 0;
4716 
4717 	/*
4718 	 * Wait for the delivery pending bit to clear.
4719 	 */
4720 	do {
4721 		++reps;
4722 
4723 		apic_ioapic_wait_pending_clear(ioapic, intin_no);
4724 
4725 		/*
4726 		 * Mask the RDT entry, but only if it's a level-triggered
4727 		 * interrupt
4728 		 */
4729 		rdt_entry = READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no);
4730 		if ((rdt_entry & (AV_LEVEL|AV_MASK)) == AV_LEVEL) {
4731 
4732 			/* Mask it */
4733 			WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no,
4734 			    AV_MASK | rdt_entry);
4735 		}
4736 
4737 		if ((rdt_entry & AV_LEVEL) == AV_LEVEL) {
4738 			/*
4739 			 * If there was a race and an interrupt was injected
4740 			 * just before we masked, check for that case here.
4741 			 * Then, unmask the RDT entry and try again.  If we're
4742 			 * on our last try, don't unmask (because we want the
4743 			 * RDT entry to remain masked for the rest of the
4744 			 * function).
4745 			 */
4746 			rdt_entry = READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic,
4747 			    intin_no);
4748 			if ((rdt_entry & AV_PENDING) &&
4749 			    (reps < apic_max_reps_clear_pending)) {
4750 				/* Unmask it */
4751 				WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic,
4752 				    intin_no, rdt_entry & ~AV_MASK);
4753 			}
4754 		}
4755 
4756 	} while ((rdt_entry & AV_PENDING) &&
4757 	    (reps < apic_max_reps_clear_pending));
4758 
4759 #ifdef DEBUG
4760 		if (rdt_entry & AV_PENDING)
4761 			apic_intr_deliver_timeouts++;
4762 #endif
4763 
4764 	/*
4765 	 * If the remote IRR bit is set, then the interrupt has been sent
4766 	 * to a CPU for processing.  We have no choice but to wait for
4767 	 * that CPU to process the interrupt, at which point the remote IRR
4768 	 * bit will be cleared.
4769 	 */
4770 	if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no) &
4771 	    (AV_LEVEL|AV_REMOTE_IRR)) == (AV_LEVEL|AV_REMOTE_IRR)) {
4772 
4773 		/*
4774 		 * If the CPU that this RDT is bound to is NOT the current
4775 		 * CPU, wait until that CPU handles the interrupt and ACKs
4776 		 * it.  If this interrupt is not bound to any CPU (that is,
4777 		 * if it's bound to the logical destination of "anyone"), it
4778 		 * may have been delivered to the current CPU so handle that
4779 		 * case by deferring the reprogramming (below).
4780 		 */
4781 		if ((old_bind_cpu != IRQ_UNBOUND) &&
4782 		    (old_bind_cpu != IRQ_UNINIT) &&
4783 		    (old_bind_cpu != psm_get_cpu_id())) {
4784 			for (waited = 0; waited < apic_max_reps_clear_pending;
4785 			    waited++) {
4786 				if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic,
4787 				    intin_no) & AV_REMOTE_IRR) == 0) {
4788 
4789 					delete_defer_repro_ent(which_irq);
4790 
4791 					/* Remote IRR has cleared! */
4792 					return (0);
4793 				}
4794 			}
4795 		}
4796 
4797 		/*
4798 		 * If we waited and the Remote IRR bit is still not cleared,
4799 		 * AND if we've invoked the timeout APIC_REPROGRAM_MAX_TIMEOUTS
4800 		 * times for this interrupt, try the last-ditch workaround:
4801 		 */
4802 		if (drep && drep->tries >= APIC_REPROGRAM_MAX_TRIES) {
4803 
4804 			apic_last_ditch_clear_remote_irr(ioapic, intin_no);
4805 
4806 			/* Mark this one as reprogrammed: */
4807 			delete_defer_repro_ent(which_irq);
4808 
4809 			return (0);
4810 		} else {
4811 #ifdef DEBUG
4812 			apic_intr_deferrals++;
4813 #endif
4814 
4815 			/*
4816 			 * If waiting for the Remote IRR bit (above) didn't
4817 			 * allow it to clear, defer the reprogramming.
4818 			 * Add a new deferred-programming entry if the
4819 			 * caller passed a NULL one (and update the existing one
4820 			 * in case anything changed).
4821 			 */
4822 			add_defer_repro_ent(irq_ptr, which_irq, new_bind_cpu);
4823 			if (drep)
4824 				drep->tries++;
4825 
4826 			/* Inform caller to defer IOAPIC programming: */
4827 			return (1);
4828 		}
4829 
4830 	}
4831 
4832 	/* Remote IRR is clear */
4833 	delete_defer_repro_ent(which_irq);
4834 
4835 	return (0);
4836 }
4837 
4838 /*
4839  * Called to migrate all interrupts at an irq to another cpu.
4840  * Must be called with interrupts disabled and apic_ioapic_lock held
4841  */
4842 int
4843 apic_rebind_all(apic_irq_t *irq_ptr, int bind_cpu)
4844 {
4845 	apic_irq_t	*irqptr = irq_ptr;
4846 	int		retval = 0;
4847 
4848 	while (irqptr) {
4849 		if (irqptr->airq_temp_cpu != IRQ_UNINIT)
4850 			retval |= apic_rebind(irqptr, bind_cpu, NULL);
4851 		irqptr = irqptr->airq_next;
4852 	}
4853 
4854 	return (retval);
4855 }
4856 
4857 /*
4858  * apic_intr_redistribute does all the messy computations for identifying
4859  * which interrupt to move to which CPU. Currently we do just one interrupt
4860  * at a time. This reduces the time we spent doing all this within clock
4861  * interrupt. When it is done in idle, we could do more than 1.
4862  * First we find the most busy and the most free CPU (time in ISR only)
4863  * skipping those CPUs that has been identified as being ineligible (cpu_skip)
4864  * Then we look for IRQs which are closest to the difference between the
4865  * most busy CPU and the average ISR load. We try to find one whose load
4866  * is less than difference.If none exists, then we chose one larger than the
4867  * difference, provided it does not make the most idle CPU worse than the
4868  * most busy one. In the end, we clear all the busy fields for CPUs. For
4869  * IRQs, they are cleared as they are scanned.
4870  */
4871 static void
4872 apic_intr_redistribute()
4873 {
4874 	int busiest_cpu, most_free_cpu;
4875 	int cpu_free, cpu_busy, max_busy, min_busy;
4876 	int min_free, diff;
4877 	int average_busy, cpus_online;
4878 	int i, busy, iflag;
4879 	apic_cpus_info_t *cpu_infop;
4880 	apic_irq_t *min_busy_irq = NULL;
4881 	apic_irq_t *max_busy_irq = NULL;
4882 
4883 	busiest_cpu = most_free_cpu = -1;
4884 	cpu_free = cpu_busy = max_busy = average_busy = 0;
4885 	min_free = apic_sample_factor_redistribution;
4886 	cpus_online = 0;
4887 	/*
4888 	 * Below we will check for CPU_INTR_ENABLE, bound, temp_bound, temp_cpu
4889 	 * without ioapic_lock. That is OK as we are just doing statistical
4890 	 * sampling anyway and any inaccuracy now will get corrected next time
4891 	 * The call to rebind which actually changes things will make sure
4892 	 * we are consistent.
4893 	 */
4894 	for (i = 0; i < apic_nproc; i++) {
4895 		if (!(apic_redist_cpu_skip & (1 << i)) &&
4896 		    (apic_cpus[i].aci_status & APIC_CPU_INTR_ENABLE)) {
4897 
4898 			cpu_infop = &apic_cpus[i];
4899 			/*
4900 			 * If no unbound interrupts or only 1 total on this
4901 			 * CPU, skip
4902 			 */
4903 			if (!cpu_infop->aci_temp_bound ||
4904 			    (cpu_infop->aci_bound + cpu_infop->aci_temp_bound)
4905 			    == 1) {
4906 				apic_redist_cpu_skip |= 1 << i;
4907 				continue;
4908 			}
4909 
4910 			busy = cpu_infop->aci_busy;
4911 			average_busy += busy;
4912 			cpus_online++;
4913 			if (max_busy < busy) {
4914 				max_busy = busy;
4915 				busiest_cpu = i;
4916 			}
4917 			if (min_free > busy) {
4918 				min_free = busy;
4919 				most_free_cpu = i;
4920 			}
4921 			if (busy > apic_int_busy_mark) {
4922 				cpu_busy |= 1 << i;
4923 			} else {
4924 				if (busy < apic_int_free_mark)
4925 					cpu_free |= 1 << i;
4926 			}
4927 		}
4928 	}
4929 	if ((cpu_busy && cpu_free) ||
4930 	    (max_busy >= (min_free + apic_diff_for_redistribution))) {
4931 
4932 		apic_num_imbalance++;
4933 #ifdef	DEBUG
4934 		if (apic_verbose & APIC_VERBOSE_IOAPIC_FLAG) {
4935 			prom_printf(
4936 			    "redistribute busy=%x free=%x max=%x min=%x",
4937 			    cpu_busy, cpu_free, max_busy, min_free);
4938 		}
4939 #endif /* DEBUG */
4940 
4941 
4942 		average_busy /= cpus_online;
4943 
4944 		diff = max_busy - average_busy;
4945 		min_busy = max_busy; /* start with the max possible value */
4946 		max_busy = 0;
4947 		min_busy_irq = max_busy_irq = NULL;
4948 		i = apic_min_device_irq;
4949 		for (; i < apic_max_device_irq; i++) {
4950 			apic_irq_t *irq_ptr;
4951 			/* Change to linked list per CPU ? */
4952 			if ((irq_ptr = apic_irq_table[i]) == NULL)
4953 				continue;
4954 			/* Check for irq_busy & decide which one to move */
4955 			/* Also zero them for next round */
4956 			if ((irq_ptr->airq_temp_cpu == busiest_cpu) &&
4957 			    irq_ptr->airq_busy) {
4958 				if (irq_ptr->airq_busy < diff) {
4959 					/*
4960 					 * Check for least busy CPU,
4961 					 * best fit or what ?
4962 					 */
4963 					if (max_busy < irq_ptr->airq_busy) {
4964 						/*
4965 						 * Most busy within the
4966 						 * required differential
4967 						 */
4968 						max_busy = irq_ptr->airq_busy;
4969 						max_busy_irq = irq_ptr;
4970 					}
4971 				} else {
4972 					if (min_busy > irq_ptr->airq_busy) {
4973 						/*
4974 						 * least busy, but more than
4975 						 * the reqd diff
4976 						 */
4977 						if (min_busy <
4978 						    (diff + average_busy -
4979 						    min_free)) {
4980 							/*
4981 							 * Making sure new cpu
4982 							 * will not end up
4983 							 * worse
4984 							 */
4985 							min_busy =
4986 							    irq_ptr->airq_busy;
4987 
4988 							min_busy_irq = irq_ptr;
4989 						}
4990 					}
4991 				}
4992 			}
4993 			irq_ptr->airq_busy = 0;
4994 		}
4995 
4996 		if (max_busy_irq != NULL) {
4997 #ifdef	DEBUG
4998 			if (apic_verbose & APIC_VERBOSE_IOAPIC_FLAG) {
4999 				prom_printf("rebinding %x to %x",
5000 				    max_busy_irq->airq_vector, most_free_cpu);
5001 			}
5002 #endif /* DEBUG */
5003 			iflag = intr_clear();
5004 			if (lock_try(&apic_ioapic_lock)) {
5005 				if (apic_rebind_all(max_busy_irq,
5006 				    most_free_cpu) == 0) {
5007 					/* Make change permenant */
5008 					max_busy_irq->airq_cpu =
5009 					    (uchar_t)most_free_cpu;
5010 				}
5011 				lock_clear(&apic_ioapic_lock);
5012 			}
5013 			intr_restore(iflag);
5014 
5015 		} else if (min_busy_irq != NULL) {
5016 #ifdef	DEBUG
5017 			if (apic_verbose & APIC_VERBOSE_IOAPIC_FLAG) {
5018 				prom_printf("rebinding %x to %x",
5019 				    min_busy_irq->airq_vector, most_free_cpu);
5020 			}
5021 #endif /* DEBUG */
5022 
5023 			iflag = intr_clear();
5024 			if (lock_try(&apic_ioapic_lock)) {
5025 				if (apic_rebind_all(min_busy_irq,
5026 				    most_free_cpu) == 0) {
5027 					/* Make change permenant */
5028 					min_busy_irq->airq_cpu =
5029 					    (uchar_t)most_free_cpu;
5030 				}
5031 				lock_clear(&apic_ioapic_lock);
5032 			}
5033 			intr_restore(iflag);
5034 
5035 		} else {
5036 			if (cpu_busy != (1 << busiest_cpu)) {
5037 				apic_redist_cpu_skip |= 1 << busiest_cpu;
5038 				/*
5039 				 * We leave cpu_skip set so that next time we
5040 				 * can choose another cpu
5041 				 */
5042 			}
5043 		}
5044 		apic_num_rebind++;
5045 	} else {
5046 		/*
5047 		 * found nothing. Could be that we skipped over valid CPUs
5048 		 * or we have balanced everything. If we had a variable
5049 		 * ticks_for_redistribution, it could be increased here.
5050 		 * apic_int_busy, int_free etc would also need to be
5051 		 * changed.
5052 		 */
5053 		if (apic_redist_cpu_skip)
5054 			apic_redist_cpu_skip = 0;
5055 	}
5056 	for (i = 0; i < apic_nproc; i++) {
5057 		apic_cpus[i].aci_busy = 0;
5058 	}
5059 }
5060 
5061 static void
5062 apic_cleanup_busy()
5063 {
5064 	int i;
5065 	apic_irq_t *irq_ptr;
5066 
5067 	for (i = 0; i < apic_nproc; i++) {
5068 		apic_cpus[i].aci_busy = 0;
5069 	}
5070 
5071 	for (i = apic_min_device_irq; i < apic_max_device_irq; i++) {
5072 		if ((irq_ptr = apic_irq_table[i]) != NULL)
5073 			irq_ptr->airq_busy = 0;
5074 	}
5075 	apic_skipped_redistribute = 0;
5076 }
5077 
5078 
5079 /*
5080  * This function will reprogram the timer.
5081  *
5082  * When in oneshot mode the argument is the absolute time in future to
5083  * generate the interrupt at.
5084  *
5085  * When in periodic mode, the argument is the interval at which the
5086  * interrupts should be generated. There is no need to support the periodic
5087  * mode timer change at this time.
5088  */
5089 static void
5090 apic_timer_reprogram(hrtime_t time)
5091 {
5092 	hrtime_t now;
5093 	uint_t ticks;
5094 	int64_t	delta;
5095 
5096 	/*
5097 	 * We should be called from high PIL context (CBE_HIGH_PIL),
5098 	 * so kpreempt is disabled.
5099 	 */
5100 
5101 	if (!apic_oneshot) {
5102 		/* time is the interval for periodic mode */
5103 		ticks = APIC_NSECS_TO_TICKS(time);
5104 	} else {
5105 		/* one shot mode */
5106 
5107 		now = gethrtime();
5108 		delta = time - now;
5109 
5110 		if (delta <= 0) {
5111 			/*
5112 			 * requested to generate an interrupt in the past
5113 			 * generate an interrupt as soon as possible
5114 			 */
5115 			ticks = apic_min_timer_ticks;
5116 		} else if (delta > apic_nsec_max) {
5117 			/*
5118 			 * requested to generate an interrupt at a time
5119 			 * further than what we are capable of. Set to max
5120 			 * the hardware can handle
5121 			 */
5122 
5123 			ticks = APIC_MAXVAL;
5124 #ifdef DEBUG
5125 			cmn_err(CE_CONT, "apic_timer_reprogram, request at"
5126 			    "  %lld  too far in future, current time"
5127 			    "  %lld \n", time, now);
5128 #endif
5129 		} else
5130 			ticks = APIC_NSECS_TO_TICKS(delta);
5131 	}
5132 
5133 	if (ticks < apic_min_timer_ticks)
5134 		ticks = apic_min_timer_ticks;
5135 
5136 	apicadr[APIC_INIT_COUNT] = ticks;
5137 
5138 }
5139 
5140 /*
5141  * This function will enable timer interrupts.
5142  */
5143 static void
5144 apic_timer_enable(void)
5145 {
5146 	/*
5147 	 * We should be Called from high PIL context (CBE_HIGH_PIL),
5148 	 * so kpreempt is disabled.
5149 	 */
5150 
5151 	if (!apic_oneshot)
5152 		apicadr[APIC_LOCAL_TIMER] =
5153 		    (apic_clkvect + APIC_BASE_VECT) | AV_TIME;
5154 	else {
5155 		/* one shot */
5156 		apicadr[APIC_LOCAL_TIMER] = (apic_clkvect + APIC_BASE_VECT);
5157 	}
5158 }
5159 
5160 /*
5161  * This function will disable timer interrupts.
5162  */
5163 static void
5164 apic_timer_disable(void)
5165 {
5166 	/*
5167 	 * We should be Called from high PIL context (CBE_HIGH_PIL),
5168 	 * so kpreempt is disabled.
5169 	 */
5170 
5171 	apicadr[APIC_LOCAL_TIMER] = (apic_clkvect + APIC_BASE_VECT) | AV_MASK;
5172 }
5173 
5174 
5175 cyclic_id_t apic_cyclic_id;
5176 
5177 /*
5178  * If this module needs to be a consumer of cyclic subsystem, they
5179  * can be added here, since at this time kernel cyclic subsystem is initialized
5180  * argument is not currently used, and is reserved for future.
5181  */
5182 static void
5183 apic_post_cyclic_setup(void *arg)
5184 {
5185 _NOTE(ARGUNUSED(arg))
5186 	cyc_handler_t hdlr;
5187 	cyc_time_t when;
5188 
5189 	/* cpu_lock is held */
5190 
5191 	/* set up cyclics for intr redistribution */
5192 
5193 	/*
5194 	 * In peridoc mode intr redistribution processing is done in
5195 	 * apic_intr_enter during clk intr processing
5196 	 */
5197 	if (!apic_oneshot)
5198 		return;
5199 
5200 	hdlr.cyh_level = CY_LOW_LEVEL;
5201 	hdlr.cyh_func = (cyc_func_t)apic_redistribute_compute;
5202 	hdlr.cyh_arg = NULL;
5203 
5204 	when.cyt_when = 0;
5205 	when.cyt_interval = apic_redistribute_sample_interval;
5206 	apic_cyclic_id = cyclic_add(&hdlr, &when);
5207 
5208 
5209 }
5210 
5211 static void
5212 apic_redistribute_compute(void)
5213 {
5214 	int	i, j, max_busy;
5215 
5216 	if (apic_enable_dynamic_migration) {
5217 		if (++apic_nticks == apic_sample_factor_redistribution) {
5218 			/*
5219 			 * Time to call apic_intr_redistribute().
5220 			 * reset apic_nticks. This will cause max_busy
5221 			 * to be calculated below and if it is more than
5222 			 * apic_int_busy, we will do the whole thing
5223 			 */
5224 			apic_nticks = 0;
5225 		}
5226 		max_busy = 0;
5227 		for (i = 0; i < apic_nproc; i++) {
5228 
5229 			/*
5230 			 * Check if curipl is non zero & if ISR is in
5231 			 * progress
5232 			 */
5233 			if (((j = apic_cpus[i].aci_curipl) != 0) &&
5234 			    (apic_cpus[i].aci_ISR_in_progress & (1 << j))) {
5235 
5236 				int	irq;
5237 				apic_cpus[i].aci_busy++;
5238 				irq = apic_cpus[i].aci_current[j];
5239 				apic_irq_table[irq]->airq_busy++;
5240 			}
5241 
5242 			if (!apic_nticks &&
5243 			    (apic_cpus[i].aci_busy > max_busy))
5244 				max_busy = apic_cpus[i].aci_busy;
5245 		}
5246 		if (!apic_nticks) {
5247 			if (max_busy > apic_int_busy_mark) {
5248 			/*
5249 			 * We could make the following check be
5250 			 * skipped > 1 in which case, we get a
5251 			 * redistribution at half the busy mark (due to
5252 			 * double interval). Need to be able to collect
5253 			 * more empirical data to decide if that is a
5254 			 * good strategy. Punt for now.
5255 			 */
5256 				if (apic_skipped_redistribute)
5257 					apic_cleanup_busy();
5258 				else
5259 					apic_intr_redistribute();
5260 			} else
5261 				apic_skipped_redistribute++;
5262 		}
5263 	}
5264 }
5265 
5266 
5267 static int
5268 apic_acpi_translate_pci_irq(dev_info_t *dip, int busid, int devid,
5269     int ipin, int *pci_irqp, iflag_t *intr_flagp)
5270 {
5271 
5272 	int status;
5273 	acpi_psm_lnk_t acpipsmlnk;
5274 
5275 	if ((status = acpi_get_irq_cache_ent(busid, devid, ipin, pci_irqp,
5276 	    intr_flagp)) == ACPI_PSM_SUCCESS) {
5277 		APIC_VERBOSE_IRQ((CE_CONT, "!pcplusmp: Found irqno %d "
5278 		    "from cache for device %s, instance #%d\n", *pci_irqp,
5279 		    ddi_get_name(dip), ddi_get_instance(dip)));
5280 		return (status);
5281 	}
5282 
5283 	bzero(&acpipsmlnk, sizeof (acpi_psm_lnk_t));
5284 
5285 	if ((status = acpi_translate_pci_irq(dip, ipin, pci_irqp, intr_flagp,
5286 	    &acpipsmlnk)) == ACPI_PSM_FAILURE) {
5287 		APIC_VERBOSE_IRQ((CE_WARN, "pcplusmp: "
5288 		    " acpi_translate_pci_irq failed for device %s, instance"
5289 		    " #%d", ddi_get_name(dip), ddi_get_instance(dip)));
5290 		return (status);
5291 	}
5292 
5293 	if (status == ACPI_PSM_PARTIAL && acpipsmlnk.lnkobj != NULL) {
5294 		status = apic_acpi_irq_configure(&acpipsmlnk, dip, pci_irqp,
5295 		    intr_flagp);
5296 		if (status != ACPI_PSM_SUCCESS) {
5297 			status = acpi_get_current_irq_resource(&acpipsmlnk,
5298 			    pci_irqp, intr_flagp);
5299 		}
5300 	}
5301 
5302 	if (status == ACPI_PSM_SUCCESS) {
5303 		acpi_new_irq_cache_ent(busid, devid, ipin, *pci_irqp,
5304 		    intr_flagp, &acpipsmlnk);
5305 
5306 		APIC_VERBOSE_IRQ((CE_CONT, "pcplusmp: [ACPI] "
5307 		    "new irq %d for device %s, instance #%d\n",
5308 		    *pci_irqp, ddi_get_name(dip), ddi_get_instance(dip)));
5309 	}
5310 
5311 	return (status);
5312 }
5313 
5314 /*
5315  * Adds an entry to the irq list passed in, and returns the new list.
5316  * Entries are added in priority order (lower numerical priorities are
5317  * placed closer to the head of the list)
5318  */
5319 static prs_irq_list_t *
5320 acpi_insert_prs_irq_ent(prs_irq_list_t *listp, int priority, int irq,
5321     iflag_t *iflagp, acpi_prs_private_t *prsprvp)
5322 {
5323 	struct prs_irq_list_ent *newent, *prevp = NULL, *origlistp;
5324 
5325 	newent = kmem_zalloc(sizeof (struct prs_irq_list_ent), KM_SLEEP);
5326 
5327 	newent->list_prio = priority;
5328 	newent->irq = irq;
5329 	newent->intrflags = *iflagp;
5330 	newent->prsprv = *prsprvp;
5331 	/* ->next is NULL from kmem_zalloc */
5332 
5333 	/*
5334 	 * New list -- return the new entry as the list.
5335 	 */
5336 	if (listp == NULL)
5337 		return (newent);
5338 
5339 	/*
5340 	 * Save original list pointer for return (since we're not modifying
5341 	 * the head)
5342 	 */
5343 	origlistp = listp;
5344 
5345 	/*
5346 	 * Insertion sort, with entries with identical keys stored AFTER
5347 	 * existing entries (the less-than-or-equal test of priority does
5348 	 * this for us).
5349 	 */
5350 	while (listp != NULL && listp->list_prio <= priority) {
5351 		prevp = listp;
5352 		listp = listp->next;
5353 	}
5354 
5355 	newent->next = listp;
5356 
5357 	if (prevp == NULL) { /* Add at head of list (newent is the new head) */
5358 		return (newent);
5359 	} else {
5360 		prevp->next = newent;
5361 		return (origlistp);
5362 	}
5363 }
5364 
5365 /*
5366  * Frees the list passed in, deallocating all memory and leaving *listpp
5367  * set to NULL.
5368  */
5369 static void
5370 acpi_destroy_prs_irq_list(prs_irq_list_t **listpp)
5371 {
5372 	struct prs_irq_list_ent *nextp;
5373 
5374 	ASSERT(listpp != NULL);
5375 
5376 	while (*listpp != NULL) {
5377 		nextp = (*listpp)->next;
5378 		kmem_free(*listpp, sizeof (struct prs_irq_list_ent));
5379 		*listpp = nextp;
5380 	}
5381 }
5382 
5383 /*
5384  * apic_choose_irqs_from_prs returns a list of irqs selected from the list of
5385  * irqs returned by the link device's _PRS method.  The irqs are chosen
5386  * to minimize contention in situations where the interrupt link device
5387  * can be programmed to steer interrupts to different interrupt controller
5388  * inputs (some of which may already be in use).  The list is sorted in order
5389  * of irqs to use, with the highest priority given to interrupt controller
5390  * inputs that are not shared.   When an interrupt controller input
5391  * must be shared, apic_choose_irqs_from_prs adds the possible irqs to the
5392  * returned list in the order that minimizes sharing (thereby ensuring lowest
5393  * possible latency from interrupt trigger time to ISR execution time).
5394  */
5395 static prs_irq_list_t *
5396 apic_choose_irqs_from_prs(acpi_irqlist_t *irqlistent, dev_info_t *dip,
5397     int crs_irq)
5398 {
5399 	int32_t irq;
5400 	int i;
5401 	prs_irq_list_t *prsirqlistp = NULL;
5402 	iflag_t iflags;
5403 
5404 	while (irqlistent != NULL) {
5405 		irqlistent->intr_flags.bustype = BUS_PCI;
5406 
5407 		for (i = 0; i < irqlistent->num_irqs; i++) {
5408 
5409 			irq = irqlistent->irqs[i];
5410 
5411 			if (irq <= 0) {
5412 				/* invalid irq number */
5413 				continue;
5414 			}
5415 
5416 			if ((irq < 16) && (apic_reserved_irqlist[irq]))
5417 				continue;
5418 
5419 			if ((apic_irq_table[irq] == NULL) ||
5420 			    (apic_irq_table[irq]->airq_dip == dip)) {
5421 
5422 				prsirqlistp = acpi_insert_prs_irq_ent(
5423 				    prsirqlistp, 0 /* Highest priority */, irq,
5424 				    &irqlistent->intr_flags,
5425 				    &irqlistent->acpi_prs_prv);
5426 
5427 				/*
5428 				 * If we do not prefer the current irq from _CRS
5429 				 * or if we do and this irq is the same as the
5430 				 * current irq from _CRS, this is the one
5431 				 * to pick.
5432 				 */
5433 				if (!(apic_prefer_crs) || (irq == crs_irq)) {
5434 					return (prsirqlistp);
5435 				}
5436 				continue;
5437 			}
5438 
5439 			/*
5440 			 * Edge-triggered interrupts cannot be shared
5441 			 */
5442 			if (irqlistent->intr_flags.intr_el == INTR_EL_EDGE)
5443 				continue;
5444 
5445 			/*
5446 			 * To work around BIOSes that contain incorrect
5447 			 * interrupt polarity information in interrupt
5448 			 * descriptors returned by _PRS, we assume that
5449 			 * the polarity of the other device sharing this
5450 			 * interrupt controller input is compatible.
5451 			 * If it's not, the caller will catch it when
5452 			 * the caller invokes the link device's _CRS method
5453 			 * (after invoking its _SRS method).
5454 			 */
5455 			iflags = irqlistent->intr_flags;
5456 			iflags.intr_po =
5457 			    apic_irq_table[irq]->airq_iflag.intr_po;
5458 
5459 			if (!acpi_intr_compatible(iflags,
5460 			    apic_irq_table[irq]->airq_iflag))
5461 				continue;
5462 
5463 			/*
5464 			 * If we prefer the irq from _CRS, no need
5465 			 * to search any further (and make sure
5466 			 * to add this irq with the highest priority
5467 			 * so it's tried first).
5468 			 */
5469 			if (crs_irq == irq && apic_prefer_crs) {
5470 
5471 				return (acpi_insert_prs_irq_ent(
5472 				    prsirqlistp,
5473 				    0 /* Highest priority */,
5474 				    irq, &iflags,
5475 				    &irqlistent->acpi_prs_prv));
5476 			}
5477 
5478 			/*
5479 			 * Priority is equal to the share count (lower
5480 			 * share count is higher priority). Note that
5481 			 * the intr flags passed in here are the ones we
5482 			 * changed above -- if incorrect, it will be
5483 			 * caught by the caller's _CRS flags comparison.
5484 			 */
5485 			prsirqlistp = acpi_insert_prs_irq_ent(
5486 			    prsirqlistp,
5487 			    apic_irq_table[irq]->airq_share, irq,
5488 			    &iflags, &irqlistent->acpi_prs_prv);
5489 		}
5490 
5491 		/* Go to the next irqlist entry */
5492 		irqlistent = irqlistent->next;
5493 	}
5494 
5495 	return (prsirqlistp);
5496 }
5497 
5498 /*
5499  * Configures the irq for the interrupt link device identified by
5500  * acpipsmlnkp.
5501  *
5502  * Gets the current and the list of possible irq settings for the
5503  * device. If apic_unconditional_srs is not set, and the current
5504  * resource setting is in the list of possible irq settings,
5505  * current irq resource setting is passed to the caller.
5506  *
5507  * Otherwise, picks an irq number from the list of possible irq
5508  * settings, and sets the irq of the device to this value.
5509  * If prefer_crs is set, among a set of irq numbers in the list that have
5510  * the least number of devices sharing the interrupt, we pick current irq
5511  * resource setting if it is a member of this set.
5512  *
5513  * Passes the irq number in the value pointed to by pci_irqp, and
5514  * polarity and sensitivity in the structure pointed to by dipintrflagp
5515  * to the caller.
5516  *
5517  * Note that if setting the irq resource failed, but successfuly obtained
5518  * the current irq resource settings, passes the current irq resources
5519  * and considers it a success.
5520  *
5521  * Returns:
5522  * ACPI_PSM_SUCCESS on success.
5523  *
5524  * ACPI_PSM_FAILURE if an error occured during the configuration or
5525  * if a suitable irq was not found for this device, or if setting the
5526  * irq resource and obtaining the current resource fails.
5527  *
5528  */
5529 static int
5530 apic_acpi_irq_configure(acpi_psm_lnk_t *acpipsmlnkp, dev_info_t *dip,
5531     int *pci_irqp, iflag_t *dipintr_flagp)
5532 {
5533 
5534 	int32_t irq;
5535 	int cur_irq = -1;
5536 	acpi_irqlist_t *irqlistp;
5537 	prs_irq_list_t *prs_irq_listp, *prs_irq_entp;
5538 	boolean_t found_irq = B_FALSE;
5539 
5540 	dipintr_flagp->bustype = BUS_PCI;
5541 
5542 	if ((acpi_get_possible_irq_resources(acpipsmlnkp, &irqlistp))
5543 	    == ACPI_PSM_FAILURE) {
5544 		APIC_VERBOSE_IRQ((CE_WARN, "!pcplusmp: Unable to determine "
5545 		    "or assign IRQ for device %s, instance #%d: The system was "
5546 		    "unable to get the list of potential IRQs from ACPI.",
5547 		    ddi_get_name(dip), ddi_get_instance(dip)));
5548 
5549 		return (ACPI_PSM_FAILURE);
5550 	}
5551 
5552 	if ((acpi_get_current_irq_resource(acpipsmlnkp, &cur_irq,
5553 	    dipintr_flagp) == ACPI_PSM_SUCCESS) && (!apic_unconditional_srs) &&
5554 	    (cur_irq > 0)) {
5555 		/*
5556 		 * If an IRQ is set in CRS and that IRQ exists in the set
5557 		 * returned from _PRS, return that IRQ, otherwise print
5558 		 * a warning
5559 		 */
5560 
5561 		if (acpi_irqlist_find_irq(irqlistp, cur_irq, NULL)
5562 		    == ACPI_PSM_SUCCESS) {
5563 
5564 			ASSERT(pci_irqp != NULL);
5565 			*pci_irqp = cur_irq;
5566 			acpi_free_irqlist(irqlistp);
5567 			return (ACPI_PSM_SUCCESS);
5568 		}
5569 
5570 		APIC_VERBOSE_IRQ((CE_WARN, "!pcplusmp: Could not find the "
5571 		    "current irq %d for device %s, instance #%d in ACPI's "
5572 		    "list of possible irqs for this device. Picking one from "
5573 		    " the latter list.", cur_irq, ddi_get_name(dip),
5574 		    ddi_get_instance(dip)));
5575 	}
5576 
5577 	if ((prs_irq_listp = apic_choose_irqs_from_prs(irqlistp, dip,
5578 	    cur_irq)) == NULL) {
5579 
5580 		APIC_VERBOSE_IRQ((CE_WARN, "!pcplusmp: Could not find a "
5581 		    "suitable irq from the list of possible irqs for device "
5582 		    "%s, instance #%d in ACPI's list of possible irqs",
5583 		    ddi_get_name(dip), ddi_get_instance(dip)));
5584 
5585 		acpi_free_irqlist(irqlistp);
5586 		return (ACPI_PSM_FAILURE);
5587 	}
5588 
5589 	acpi_free_irqlist(irqlistp);
5590 
5591 	for (prs_irq_entp = prs_irq_listp;
5592 	    prs_irq_entp != NULL && found_irq == B_FALSE;
5593 	    prs_irq_entp = prs_irq_entp->next) {
5594 
5595 		acpipsmlnkp->acpi_prs_prv = prs_irq_entp->prsprv;
5596 		irq = prs_irq_entp->irq;
5597 
5598 		APIC_VERBOSE_IRQ((CE_CONT, "!pcplusmp: Setting irq %d for "
5599 		    "device %s instance #%d\n", irq, ddi_get_name(dip),
5600 		    ddi_get_instance(dip)));
5601 
5602 		if ((acpi_set_irq_resource(acpipsmlnkp, irq))
5603 		    == ACPI_PSM_SUCCESS) {
5604 			/*
5605 			 * setting irq was successful, check to make sure CRS
5606 			 * reflects that. If CRS does not agree with what we
5607 			 * set, return the irq that was set.
5608 			 */
5609 
5610 			if (acpi_get_current_irq_resource(acpipsmlnkp, &cur_irq,
5611 			    dipintr_flagp) == ACPI_PSM_SUCCESS) {
5612 
5613 				if (cur_irq != irq)
5614 					APIC_VERBOSE_IRQ((CE_WARN,
5615 					    "!pcplusmp: IRQ resource set "
5616 					    "(irqno %d) for device %s "
5617 					    "instance #%d, differs from "
5618 					    "current setting irqno %d",
5619 					    irq, ddi_get_name(dip),
5620 					    ddi_get_instance(dip), cur_irq));
5621 			} else {
5622 				/*
5623 				 * On at least one system, there was a bug in
5624 				 * a DSDT method called by _STA, causing _STA to
5625 				 * indicate that the link device was disabled
5626 				 * (when, in fact, it was enabled).  Since _SRS
5627 				 * succeeded, assume that _CRS is lying and use
5628 				 * the iflags from this _PRS interrupt choice.
5629 				 * If we're wrong about the flags, the polarity
5630 				 * will be incorrect and we may get an interrupt
5631 				 * storm, but there's not much else we can do
5632 				 * at this point.
5633 				 */
5634 				*dipintr_flagp = prs_irq_entp->intrflags;
5635 			}
5636 
5637 			/*
5638 			 * Return the irq that was set, and not what _CRS
5639 			 * reports, since _CRS has been seen to return
5640 			 * different IRQs than what was passed to _SRS on some
5641 			 * systems (and just not return successfully on others).
5642 			 */
5643 			cur_irq = irq;
5644 			found_irq = B_TRUE;
5645 		} else {
5646 			APIC_VERBOSE_IRQ((CE_WARN, "!pcplusmp: set resource "
5647 			    "irq %d failed for device %s instance #%d",
5648 			    irq, ddi_get_name(dip), ddi_get_instance(dip)));
5649 
5650 			if (cur_irq == -1) {
5651 				acpi_destroy_prs_irq_list(&prs_irq_listp);
5652 				return (ACPI_PSM_FAILURE);
5653 			}
5654 		}
5655 	}
5656 
5657 	acpi_destroy_prs_irq_list(&prs_irq_listp);
5658 
5659 	if (!found_irq)
5660 		return (ACPI_PSM_FAILURE);
5661 
5662 	ASSERT(pci_irqp != NULL);
5663 	*pci_irqp = cur_irq;
5664 	return (ACPI_PSM_SUCCESS);
5665 }
5666