xref: /illumos-gate/usr/src/uts/i86pc/io/pcplusmp/apic.c (revision 814a60b13c0ad90e5d2edfd29a7a84bbf416cc1a)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 /*
30  * PSMI 1.1 extensions are supported only in 2.6 and later versions.
31  * PSMI 1.2 extensions are supported only in 2.7 and later versions.
32  * PSMI 1.3 and 1.4 extensions are supported in Solaris 10.
33  * PSMI 1.5 extensions are supported in Solaris Nevada.
34  */
35 #define	PSMI_1_5
36 
37 #include <sys/processor.h>
38 #include <sys/time.h>
39 #include <sys/psm.h>
40 #include <sys/smp_impldefs.h>
41 #include <sys/cram.h>
42 #include <sys/acpi/acpi.h>
43 #include <sys/acpica.h>
44 #include <sys/psm_common.h>
45 #include "apic.h"
46 #include <sys/pit.h>
47 #include <sys/ddi.h>
48 #include <sys/sunddi.h>
49 #include <sys/ddi_impldefs.h>
50 #include <sys/pci.h>
51 #include <sys/promif.h>
52 #include <sys/x86_archext.h>
53 #include <sys/cpc_impl.h>
54 #include <sys/uadmin.h>
55 #include <sys/panic.h>
56 #include <sys/debug.h>
57 #include <sys/archsystm.h>
58 #include <sys/trap.h>
59 #include <sys/machsystm.h>
60 #include <sys/cpuvar.h>
61 #include <sys/rm_platter.h>
62 #include <sys/privregs.h>
63 #include <sys/cyclic.h>
64 #include <sys/note.h>
65 #include <sys/pci_intr_lib.h>
66 
67 /*
68  *	Local Function Prototypes
69  */
70 static void apic_init_intr();
71 static void apic_ret();
72 static int apic_handle_defconf();
73 static int apic_parse_mpct(caddr_t mpct, int bypass);
74 static struct apic_mpfps_hdr *apic_find_fps_sig(caddr_t fptr, int size);
75 static int apic_checksum(caddr_t bptr, int len);
76 static int get_apic_cmd1();
77 static int get_apic_pri();
78 static int apic_find_bus_type(char *bus);
79 static int apic_find_bus(int busid);
80 static int apic_find_bus_id(int bustype);
81 static struct apic_io_intr *apic_find_io_intr(int irqno);
82 int apic_allocate_irq(int irq);
83 static int apic_find_free_irq(int start, int end);
84 static uchar_t apic_allocate_vector(int ipl, int irq, int pri);
85 static void apic_modify_vector(uchar_t vector, int irq);
86 static void apic_mark_vector(uchar_t oldvector, uchar_t newvector);
87 static uchar_t apic_xlate_vector(uchar_t oldvector);
88 static void apic_xlate_vector_free_timeout_handler(void *arg);
89 static void apic_free_vector(uchar_t vector);
90 static void apic_reprogram_timeout_handler(void *arg);
91 static int apic_check_stuck_interrupt(apic_irq_t *irq_ptr, int old_bind_cpu,
92     int new_bind_cpu, volatile int32_t *ioapic, int intin_no, int which_irq);
93 static int apic_setup_io_intr(apic_irq_t *irqptr, int irq);
94 static int apic_setup_io_intr_deferred(apic_irq_t *irqptr, int irq);
95 static void apic_record_rdt_entry(apic_irq_t *irqptr, int irq);
96 static struct apic_io_intr *apic_find_io_intr_w_busid(int irqno, int busid);
97 static int apic_find_intin(uchar_t ioapic, uchar_t intin);
98 static int apic_handle_pci_pci_bridge(dev_info_t *idip, int child_devno,
99     int child_ipin, struct apic_io_intr **intrp);
100 static int apic_setup_irq_table(dev_info_t *dip, int irqno,
101     struct apic_io_intr *intrp, struct intrspec *ispec, iflag_t *intr_flagp,
102     int type);
103 static int apic_setup_sci_irq_table(int irqno, uchar_t ipl,
104     iflag_t *intr_flagp);
105 static void apic_nmi_intr(caddr_t arg);
106 uchar_t apic_bind_intr(dev_info_t *dip, int irq, uchar_t ioapicid,
107     uchar_t intin);
108 static int apic_rebind(apic_irq_t *irq_ptr, int bind_cpu, int acquire_lock,
109     int when);
110 static int apic_rebind_all(apic_irq_t *irq_ptr, int bind_cpu, int safe);
111 static void apic_intr_redistribute();
112 static void apic_cleanup_busy();
113 static void apic_set_pwroff_method_from_mpcnfhdr(struct apic_mp_cnf_hdr *hdrp);
114 int apic_introp_xlate(dev_info_t *dip, struct intrspec *ispec, int type);
115 
116 /* ACPI support routines */
117 static int acpi_probe(void);
118 static int apic_acpi_irq_configure(acpi_psm_lnk_t *acpipsmlnkp, dev_info_t *dip,
119     int *pci_irqp, iflag_t *intr_flagp);
120 
121 static int apic_acpi_translate_pci_irq(dev_info_t *dip, int busid, int devid,
122     int ipin, int *pci_irqp, iflag_t *intr_flagp);
123 static uchar_t acpi_find_ioapic(int irq);
124 static int acpi_intr_compatible(iflag_t iflag1, iflag_t iflag2);
125 
126 /*
127  *	standard MP entries
128  */
129 static int	apic_probe();
130 static int	apic_clkinit();
131 static int	apic_getclkirq(int ipl);
132 static uint_t	apic_calibrate(volatile uint32_t *addr,
133     uint16_t *pit_ticks_adj);
134 static hrtime_t apic_gettime();
135 static hrtime_t apic_gethrtime();
136 static void	apic_init();
137 static void	apic_picinit(void);
138 static void	apic_cpu_start(processorid_t cpun, caddr_t rm_code);
139 static int	apic_post_cpu_start(void);
140 static void	apic_send_ipi(int cpun, int ipl);
141 static void	apic_set_softintr(int softintr);
142 static void	apic_set_idlecpu(processorid_t cpun);
143 static void	apic_unset_idlecpu(processorid_t cpun);
144 static int	apic_softlvl_to_irq(int ipl);
145 static int	apic_intr_enter(int ipl, int *vect);
146 static void	apic_intr_exit(int ipl, int vect);
147 static void	apic_setspl(int ipl);
148 static int	apic_addspl(int ipl, int vector, int min_ipl, int max_ipl);
149 static int	apic_delspl(int ipl, int vector, int min_ipl, int max_ipl);
150 static void	apic_shutdown(int cmd, int fcn);
151 static void	apic_preshutdown(int cmd, int fcn);
152 static int	apic_disable_intr(processorid_t cpun);
153 static void	apic_enable_intr(processorid_t cpun);
154 static processorid_t	apic_get_next_processorid(processorid_t cpun);
155 static int		apic_get_ipivect(int ipl, int type);
156 static void	apic_timer_reprogram(hrtime_t time);
157 static void	apic_timer_enable(void);
158 static void	apic_timer_disable(void);
159 static void	apic_post_cyclic_setup(void *arg);
160 extern int	apic_intr_ops(dev_info_t *, ddi_intr_handle_impl_t *,
161 		    psm_intr_op_t, int *);
162 
163 static int	apic_oneshot = 0;
164 int	apic_oneshot_enable = 1; /* to allow disabling one-shot capability */
165 
166 /*
167  * These variables are frequently accessed in apic_intr_enter(),
168  * apic_intr_exit and apic_setspl, so group them together
169  */
170 volatile uint32_t *apicadr =  NULL;	/* virtual addr of local APIC	*/
171 int apic_setspl_delay = 1;		/* apic_setspl - delay enable	*/
172 int apic_clkvect;
173 
174 /* ACPI SCI interrupt configuration; -1 if SCI not used */
175 int apic_sci_vect = -1;
176 iflag_t apic_sci_flags;
177 
178 /* vector at which error interrupts come in */
179 int apic_errvect;
180 int apic_enable_error_intr = 1;
181 int apic_error_display_delay = 100;
182 
183 /* vector at which performance counter overflow interrupts come in */
184 int apic_cpcovf_vect;
185 int apic_enable_cpcovf_intr = 1;
186 
187 /* Max wait time (in microsecs) for flags to clear in an RDT entry. */
188 static int apic_max_usecs_clear_pending = 1000;
189 
190 /* Amt of usecs to wait before checking if RDT flags have reset. */
191 #define	APIC_USECS_PER_WAIT_INTERVAL 100
192 
193 /* Maximum number of times to retry reprogramming via the timeout */
194 #define	APIC_REPROGRAM_MAX_TIMEOUTS 10
195 
196 /* timeout delay for IOAPIC delayed reprogramming */
197 #define	APIC_REPROGRAM_TIMEOUT_DELAY 5 /* microseconds */
198 
199 /* Parameter to apic_rebind(): Should reprogramming be done now or later? */
200 #define	DEFERRED 1
201 #define	IMMEDIATE 0
202 
203 /*
204  * number of bits per byte, from <sys/param.h>
205  */
206 #define	UCHAR_MAX	((1 << NBBY) - 1)
207 
208 uchar_t	apic_reserved_irqlist[MAX_ISA_IRQ];
209 
210 /*
211  * The following vector assignments influence the value of ipltopri and
212  * vectortoipl. Note that vectors 0 - 0x1f are not used. We can program
213  * idle to 0 and IPL 0 to 0x10 to differentiate idle in case
214  * we care to do so in future. Note some IPLs which are rarely used
215  * will share the vector ranges and heavily used IPLs (5 and 6) have
216  * a wide range.
217  *	IPL		Vector range.		as passed to intr_enter
218  *	0		none.
219  *	1,2,3		0x20-0x2f		0x0-0xf
220  *	4		0x30-0x3f		0x10-0x1f
221  *	5		0x40-0x5f		0x20-0x3f
222  *	6		0x60-0x7f		0x40-0x5f
223  *	7,8,9		0x80-0x8f		0x60-0x6f
224  *	10		0x90-0x9f		0x70-0x7f
225  *	11		0xa0-0xaf		0x80-0x8f
226  *	...		...
227  *	16		0xf0-0xff		0xd0-0xdf
228  */
229 uchar_t apic_vectortoipl[APIC_AVAIL_VECTOR / APIC_VECTOR_PER_IPL] = {
230 	3, 4, 5, 5, 6, 6, 9, 10, 11, 12, 13, 14, 15, 16
231 };
232 	/*
233 	 * The ipl of an ISR at vector X is apic_vectortoipl[X<<4]
234 	 * NOTE that this is vector as passed into intr_enter which is
235 	 * programmed vector - 0x20 (APIC_BASE_VECT)
236 	 */
237 
238 uchar_t	apic_ipltopri[MAXIPL + 1];	/* unix ipl to apic pri	*/
239 	/* The taskpri to be programmed into apic to mask given ipl */
240 
241 #if defined(__amd64)
242 uchar_t	apic_cr8pri[MAXIPL + 1];	/* unix ipl to cr8 pri	*/
243 #endif
244 
245 /*
246  * Patchable global variables.
247  */
248 int	apic_forceload = 0;
249 
250 #define	INTR_ROUND_ROBIN_WITH_AFFINITY	0
251 #define	INTR_ROUND_ROBIN		1
252 #define	INTR_LOWEST_PRIORITY		2
253 
254 int	apic_intr_policy = INTR_ROUND_ROBIN_WITH_AFFINITY;
255 
256 static int	apic_next_bind_cpu = 2; /* For round robin assignment */
257 					/* start with cpu 1 */
258 
259 int	apic_coarse_hrtime = 1;		/* 0 - use accurate slow gethrtime() */
260 					/* 1 - use gettime() for performance */
261 int	apic_flat_model = 0;		/* 0 - clustered. 1 - flat */
262 int	apic_enable_hwsoftint = 0;	/* 0 - disable, 1 - enable	*/
263 int	apic_enable_bind_log = 1;	/* 1 - display interrupt binding log */
264 int	apic_panic_on_nmi = 0;
265 int	apic_panic_on_apic_error = 0;
266 
267 int	apic_verbose = 0;
268 
269 /* Flag definitions for apic_verbose */
270 #define	APIC_VERBOSE_IOAPIC_FLAG		0x00000001
271 #define	APIC_VERBOSE_IRQ_FLAG			0x00000002
272 #define	APIC_VERBOSE_POWEROFF_FLAG		0x00000004
273 #define	APIC_VERBOSE_POWEROFF_PAUSE_FLAG	0x00000008
274 
275 
276 #define	APIC_VERBOSE_IOAPIC(fmt) \
277 	if (apic_verbose & APIC_VERBOSE_IOAPIC_FLAG) \
278 		cmn_err fmt;
279 
280 #define	APIC_VERBOSE_IRQ(fmt) \
281 	if (apic_verbose & APIC_VERBOSE_IRQ_FLAG) \
282 		cmn_err fmt;
283 
284 #define	APIC_VERBOSE_POWEROFF(fmt) \
285 	if (apic_verbose & APIC_VERBOSE_POWEROFF_FLAG) \
286 		prom_printf fmt;
287 
288 
289 /* Now the ones for Dynamic Interrupt distribution */
290 int	apic_enable_dynamic_migration = 1;
291 
292 /*
293  * If enabled, the distribution works as follows:
294  * On every interrupt entry, the current ipl for the CPU is set in cpu_info
295  * and the irq corresponding to the ipl is also set in the aci_current array.
296  * interrupt exit and setspl (due to soft interrupts) will cause the current
297  * ipl to be be changed. This is cache friendly as these frequently used
298  * paths write into a per cpu structure.
299  *
300  * Sampling is done by checking the structures for all CPUs and incrementing
301  * the busy field of the irq (if any) executing on each CPU and the busy field
302  * of the corresponding CPU.
303  * In periodic mode this is done on every clock interrupt.
304  * In one-shot mode, this is done thru a cyclic with an interval of
305  * apic_redistribute_sample_interval (default 10 milli sec).
306  *
307  * Every apic_sample_factor_redistribution times we sample, we do computations
308  * to decide which interrupt needs to be migrated (see comments
309  * before apic_intr_redistribute().
310  */
311 
312 /*
313  * Following 3 variables start as % and can be patched or set using an
314  * API to be defined in future. They will be scaled to
315  * sample_factor_redistribution which is in turn set to hertz+1 (in periodic
316  * mode), or 101 in one-shot mode to stagger it away from one sec processing
317  */
318 
319 int	apic_int_busy_mark = 60;
320 int	apic_int_free_mark = 20;
321 int	apic_diff_for_redistribution = 10;
322 
323 /* sampling interval for interrupt redistribution for dynamic migration */
324 int	apic_redistribute_sample_interval = NANOSEC / 100; /* 10 millisec */
325 
326 /*
327  * number of times we sample before deciding to redistribute interrupts
328  * for dynamic migration
329  */
330 int	apic_sample_factor_redistribution = 101;
331 
332 /* timeout for xlate_vector, mark_vector */
333 int	apic_revector_timeout = 16 * 10000; /* 160 millisec */
334 
335 int	apic_redist_cpu_skip = 0;
336 int	apic_num_imbalance = 0;
337 int	apic_num_rebind = 0;
338 
339 int	apic_nproc = 0;
340 int	apic_defconf = 0;
341 int	apic_irq_translate = 0;
342 int	apic_spec_rev = 0;
343 int	apic_imcrp = 0;
344 
345 int	apic_use_acpi = 1;	/* 1 = use ACPI, 0 = don't use ACPI */
346 int	apic_use_acpi_madt_only = 0;	/* 1=ONLY use MADT from ACPI */
347 
348 /*
349  * For interrupt link devices, if apic_unconditional_srs is set, an irq resource
350  * will be assigned (via _SRS). If it is not set, use the current
351  * irq setting (via _CRS), but only if that irq is in the set of possible
352  * irqs (returned by _PRS) for the device.
353  */
354 int	apic_unconditional_srs = 1;
355 
356 /*
357  * For interrupt link devices, if apic_prefer_crs is set when we are
358  * assigning an IRQ resource to a device, prefer the current IRQ setting
359  * over other possible irq settings under same conditions.
360  */
361 
362 int	apic_prefer_crs = 1;
363 
364 
365 /* minimum number of timer ticks to program to */
366 int apic_min_timer_ticks = 1;
367 /*
368  *	Local static data
369  */
370 static struct	psm_ops apic_ops = {
371 	apic_probe,
372 
373 	apic_init,
374 	apic_picinit,
375 	apic_intr_enter,
376 	apic_intr_exit,
377 	apic_setspl,
378 	apic_addspl,
379 	apic_delspl,
380 	apic_disable_intr,
381 	apic_enable_intr,
382 	apic_softlvl_to_irq,
383 	apic_set_softintr,
384 
385 	apic_set_idlecpu,
386 	apic_unset_idlecpu,
387 
388 	apic_clkinit,
389 	apic_getclkirq,
390 	(void (*)(void))NULL,		/* psm_hrtimeinit */
391 	apic_gethrtime,
392 
393 	apic_get_next_processorid,
394 	apic_cpu_start,
395 	apic_post_cpu_start,
396 	apic_shutdown,
397 	apic_get_ipivect,
398 	apic_send_ipi,
399 
400 	(int (*)(dev_info_t *, int))NULL,	/* psm_translate_irq */
401 	(int (*)(todinfo_t *))NULL,	/* psm_tod_get */
402 	(int (*)(todinfo_t *))NULL,	/* psm_tod_set */
403 	(void (*)(int, char *))NULL,	/* psm_notify_error */
404 	(void (*)(int))NULL,		/* psm_notify_func */
405 	apic_timer_reprogram,
406 	apic_timer_enable,
407 	apic_timer_disable,
408 	apic_post_cyclic_setup,
409 	apic_preshutdown,
410 	apic_intr_ops			/* Advanced DDI Interrupt framework */
411 };
412 
413 
414 static struct	psm_info apic_psm_info = {
415 	PSM_INFO_VER01_5,			/* version */
416 	PSM_OWN_EXCLUSIVE,			/* ownership */
417 	(struct psm_ops *)&apic_ops,		/* operation */
418 	"pcplusmp",				/* machine name */
419 	"pcplusmp v1.4 compatible %I%",
420 };
421 
422 static void *apic_hdlp;
423 
424 #ifdef DEBUG
425 #define	DENT		0x0001
426 int	apic_debug = 0;
427 /*
428  * set apic_restrict_vector to the # of vectors we want to allow per range
429  * useful in testing shared interrupt logic by setting it to 2 or 3
430  */
431 int	apic_restrict_vector = 0;
432 
433 #define	APIC_DEBUG_MSGBUFSIZE	2048
434 int	apic_debug_msgbuf[APIC_DEBUG_MSGBUFSIZE];
435 int	apic_debug_msgbufindex = 0;
436 
437 /*
438  * Put "int" info into debug buffer. No MP consistency, but light weight.
439  * Good enough for most debugging.
440  */
441 #define	APIC_DEBUG_BUF_PUT(x) \
442 	apic_debug_msgbuf[apic_debug_msgbufindex++] = x; \
443 	if (apic_debug_msgbufindex >= (APIC_DEBUG_MSGBUFSIZE - NCPU)) \
444 		apic_debug_msgbufindex = 0;
445 
446 #endif /* DEBUG */
447 
448 apic_cpus_info_t	*apic_cpus;
449 
450 static uint_t	apic_cpumask = 0;
451 static uint_t	apic_flag;
452 
453 /* Flag to indicate that we need to shut down all processors */
454 static uint_t	apic_shutdown_processors;
455 
456 uint_t apic_nsec_per_intr = 0;
457 
458 /*
459  * apic_let_idle_redistribute can have the following values:
460  * 0 - If clock decremented it from 1 to 0, clock has to call redistribute.
461  * apic_redistribute_lock prevents multiple idle cpus from redistributing
462  */
463 int	apic_num_idle_redistributions = 0;
464 static	int apic_let_idle_redistribute = 0;
465 static	uint_t apic_nticks = 0;
466 static	uint_t apic_skipped_redistribute = 0;
467 
468 /* to gather intr data and redistribute */
469 static void apic_redistribute_compute(void);
470 
471 static	uint_t last_count_read = 0;
472 static	lock_t	apic_gethrtime_lock;
473 volatile int	apic_hrtime_stamp = 0;
474 volatile hrtime_t apic_nsec_since_boot = 0;
475 static uint_t apic_hertz_count, apic_nsec_per_tick;
476 static hrtime_t apic_nsec_max;
477 
478 static	hrtime_t	apic_last_hrtime = 0;
479 int		apic_hrtime_error = 0;
480 int		apic_remote_hrterr = 0;
481 int		apic_num_nmis = 0;
482 int		apic_apic_error = 0;
483 int		apic_num_apic_errors = 0;
484 int		apic_num_cksum_errors = 0;
485 
486 static	uchar_t	apic_io_id[MAX_IO_APIC];
487 static	uchar_t	apic_io_ver[MAX_IO_APIC];
488 static	uchar_t	apic_io_vectbase[MAX_IO_APIC];
489 static	uchar_t	apic_io_vectend[MAX_IO_APIC];
490 volatile int32_t *apicioadr[MAX_IO_APIC];
491 /*
492  * apic_ioapic_lock protects the ioapics (reg select), the status, temp_bound
493  * and bound elements of cpus_info and the temp_cpu element of irq_struct
494  */
495 lock_t	apic_ioapic_lock;
496 
497 /*
498  * apic_ioapic_reprogram_lock prevents a CPU from exiting
499  * apic_intr_exit before IOAPIC reprogramming information
500  * is collected.
501  */
502 static	lock_t	apic_ioapic_reprogram_lock;
503 static	int	apic_io_max = 0;	/* no. of i/o apics enabled */
504 
505 static	struct apic_io_intr *apic_io_intrp = 0;
506 static	struct apic_bus	*apic_busp;
507 
508 uchar_t	apic_vector_to_irq[APIC_MAX_VECTOR+1];
509 static	uchar_t	apic_resv_vector[MAXIPL+1];
510 
511 static	char	apic_level_intr[APIC_MAX_VECTOR+1];
512 static	int	apic_error = 0;
513 /* values which apic_error can take. Not catastrophic, but may help debug */
514 #define	APIC_ERR_BOOT_EOI		0x1
515 #define	APIC_ERR_GET_IPIVECT_FAIL	0x2
516 #define	APIC_ERR_INVALID_INDEX		0x4
517 #define	APIC_ERR_MARK_VECTOR_FAIL	0x8
518 #define	APIC_ERR_APIC_ERROR		0x40000000
519 #define	APIC_ERR_NMI			0x80000000
520 
521 static	int	apic_cmos_ssb_set = 0;
522 
523 static	uint32_t	eisa_level_intr_mask = 0;
524 	/* At least MSB will be set if EISA bus */
525 
526 static	int	apic_pci_bus_total = 0;
527 static	uchar_t	apic_single_pci_busid = 0;
528 
529 
530 /*
531  * airq_mutex protects additions to the apic_irq_table - the first
532  * pointer and any airq_nexts off of that one. It also protects
533  * apic_max_device_irq & apic_min_device_irq. It also guarantees
534  * that share_id is unique as new ids are generated only when new
535  * irq_t structs are linked in. Once linked in the structs are never
536  * deleted. temp_cpu & mps_intr_index field indicate if it is programmed
537  * or allocated. Note that there is a slight gap between allocating in
538  * apic_introp_xlate and programming in addspl.
539  */
540 kmutex_t	airq_mutex;
541 apic_irq_t	*apic_irq_table[APIC_MAX_VECTOR+1];
542 int		apic_max_device_irq = 0;
543 int		apic_min_device_irq = APIC_MAX_VECTOR;
544 
545 /* use to make sure only one cpu handles the nmi */
546 static	lock_t	apic_nmi_lock;
547 /* use to make sure only one cpu handles the error interrupt */
548 static	lock_t	apic_error_lock;
549 
550 /*
551  * Following declarations are for revectoring; used when ISRs at different
552  * IPLs share an irq.
553  */
554 static	lock_t	apic_revector_lock;
555 static	int	apic_revector_pending = 0;
556 static	uchar_t	*apic_oldvec_to_newvec;
557 static	uchar_t	*apic_newvec_to_oldvec;
558 
559 /* Ensures that the IOAPIC-reprogramming timeout is not reentrant */
560 static	kmutex_t	apic_reprogram_timeout_mutex;
561 
562 static	struct	ioapic_reprogram_data {
563 	int		valid;	 /* This entry is valid */
564 	int		bindcpu; /* The CPU to which the int will be bound */
565 	unsigned	timeouts; /* # times the reprogram timeout was called */
566 } apic_reprogram_info[APIC_MAX_VECTOR+1];
567 /*
568  * APIC_MAX_VECTOR + 1 is the maximum # of IRQs as well. apic_reprogram_info
569  * is indexed by IRQ number, NOT by vector number.
570  */
571 
572 
573 /*
574  * The following added to identify a software poweroff method if available.
575  */
576 
577 static struct {
578 	int	poweroff_method;
579 	char	oem_id[APIC_MPS_OEM_ID_LEN + 1];	/* MAX + 1 for NULL */
580 	char	prod_id[APIC_MPS_PROD_ID_LEN + 1];	/* MAX + 1 for NULL */
581 } apic_mps_ids[] = {
582 	{ APIC_POWEROFF_VIA_RTC,	"INTEL",	"ALDER" },   /* 4300 */
583 	{ APIC_POWEROFF_VIA_RTC,	"NCR",		"AMC" },    /* 4300 */
584 	{ APIC_POWEROFF_VIA_ASPEN_BMC,	"INTEL",	"A450NX" },  /* 4400? */
585 	{ APIC_POWEROFF_VIA_ASPEN_BMC,	"INTEL",	"AD450NX" }, /* 4400 */
586 	{ APIC_POWEROFF_VIA_ASPEN_BMC,	"INTEL",	"AC450NX" }, /* 4400R */
587 	{ APIC_POWEROFF_VIA_SITKA_BMC,	"INTEL",	"S450NX" },  /* S50  */
588 	{ APIC_POWEROFF_VIA_SITKA_BMC,	"INTEL",	"SC450NX" }  /* S50? */
589 };
590 
591 int	apic_poweroff_method = APIC_POWEROFF_NONE;
592 
593 static	struct {
594 	uchar_t	cntl;
595 	uchar_t	data;
596 } aspen_bmc[] = {
597 	{ CC_SMS_WR_START,	0x18 },		/* NetFn/LUN */
598 	{ CC_SMS_WR_NEXT,	0x24 },		/* Cmd SET_WATCHDOG_TIMER */
599 	{ CC_SMS_WR_NEXT,	0x84 },		/* DataByte 1: SMS/OS no log */
600 	{ CC_SMS_WR_NEXT,	0x2 },		/* DataByte 2: Power Down */
601 	{ CC_SMS_WR_NEXT,	0x0 },		/* DataByte 3: no pre-timeout */
602 	{ CC_SMS_WR_NEXT,	0x0 },		/* DataByte 4: timer expir. */
603 	{ CC_SMS_WR_NEXT,	0xa },		/* DataByte 5: init countdown */
604 	{ CC_SMS_WR_END,	0x0 },		/* DataByte 6: init countdown */
605 
606 	{ CC_SMS_WR_START,	0x18 },		/* NetFn/LUN */
607 	{ CC_SMS_WR_END,	0x22 }		/* Cmd RESET_WATCHDOG_TIMER */
608 };
609 
610 static	struct {
611 	int	port;
612 	uchar_t	data;
613 } sitka_bmc[] = {
614 	{ SMS_COMMAND_REGISTER,	SMS_WRITE_START },
615 	{ SMS_DATA_REGISTER,	0x18 },		/* NetFn/LUN */
616 	{ SMS_DATA_REGISTER,	0x24 },		/* Cmd SET_WATCHDOG_TIMER */
617 	{ SMS_DATA_REGISTER,	0x84 },		/* DataByte 1: SMS/OS no log */
618 	{ SMS_DATA_REGISTER,	0x2 },		/* DataByte 2: Power Down */
619 	{ SMS_DATA_REGISTER,	0x0 },		/* DataByte 3: no pre-timeout */
620 	{ SMS_DATA_REGISTER,	0x0 },		/* DataByte 4: timer expir. */
621 	{ SMS_DATA_REGISTER,	0xa },		/* DataByte 5: init countdown */
622 	{ SMS_COMMAND_REGISTER,	SMS_WRITE_END },
623 	{ SMS_DATA_REGISTER,	0x0 },		/* DataByte 6: init countdown */
624 
625 	{ SMS_COMMAND_REGISTER,	SMS_WRITE_START },
626 	{ SMS_DATA_REGISTER,	0x18 },		/* NetFn/LUN */
627 	{ SMS_COMMAND_REGISTER,	SMS_WRITE_END },
628 	{ SMS_DATA_REGISTER,	0x22 }		/* Cmd RESET_WATCHDOG_TIMER */
629 };
630 
631 
632 /* Patchable global variables. */
633 int		apic_kmdb_on_nmi = 0;		/* 0 - no, 1 - yes enter kmdb */
634 int		apic_debug_mps_id = 0;		/* 1 - print MPS ID strings */
635 
636 /*
637  * ACPI definitions
638  */
639 /* _PIC method arguments */
640 #define	ACPI_PIC_MODE	0
641 #define	ACPI_APIC_MODE	1
642 
643 /* APIC error flags we care about */
644 #define	APIC_SEND_CS_ERROR	0x01
645 #define	APIC_RECV_CS_ERROR	0x02
646 #define	APIC_CS_ERRORS		(APIC_SEND_CS_ERROR|APIC_RECV_CS_ERROR)
647 
648 /*
649  * ACPI variables
650  */
651 /* 1 = acpi is enabled & working, 0 = acpi is not enabled or not there */
652 static	int apic_enable_acpi = 0;
653 
654 /* ACPI Multiple APIC Description Table ptr */
655 static	MULTIPLE_APIC_TABLE *acpi_mapic_dtp = NULL;
656 
657 /* ACPI Interrupt Source Override Structure ptr */
658 static	MADT_INTERRUPT_OVERRIDE *acpi_isop = NULL;
659 static	int acpi_iso_cnt = 0;
660 
661 /* ACPI Non-maskable Interrupt Sources ptr */
662 static	MADT_NMI_SOURCE *acpi_nmi_sp = NULL;
663 static	int acpi_nmi_scnt = 0;
664 static	MADT_LOCAL_APIC_NMI *acpi_nmi_cp = NULL;
665 static	int acpi_nmi_ccnt = 0;
666 
667 /*
668  * extern declarations
669  */
670 extern	int	intr_clear(void);
671 extern	void	intr_restore(uint_t);
672 #if defined(__amd64)
673 extern	int	intpri_use_cr8;
674 #endif	/* __amd64 */
675 
676 extern int	apic_pci_msi_enable_vector(dev_info_t *, int, int,
677 		    int, int, int);
678 extern apic_irq_t *apic_find_irq(dev_info_t *, struct intrspec *, int);
679 
680 /*
681  *	This is the loadable module wrapper
682  */
683 
684 int
685 _init(void)
686 {
687 	if (apic_coarse_hrtime)
688 		apic_ops.psm_gethrtime = &apic_gettime;
689 	return (psm_mod_init(&apic_hdlp, &apic_psm_info));
690 }
691 
692 int
693 _fini(void)
694 {
695 	return (psm_mod_fini(&apic_hdlp, &apic_psm_info));
696 }
697 
698 int
699 _info(struct modinfo *modinfop)
700 {
701 	return (psm_mod_info(&apic_hdlp, &apic_psm_info, modinfop));
702 }
703 
704 /*
705  * Auto-configuration routines
706  */
707 
708 /*
709  * Look at MPSpec 1.4 (Intel Order # 242016-005) for details of what we do here
710  * May work with 1.1 - but not guaranteed.
711  * According to the MP Spec, the MP floating pointer structure
712  * will be searched in the order described below:
713  * 1. In the first kilobyte of Extended BIOS Data Area (EBDA)
714  * 2. Within the last kilobyte of system base memory
715  * 3. In the BIOS ROM address space between 0F0000h and 0FFFFh
716  * Once we find the right signature with proper checksum, we call
717  * either handle_defconf or parse_mpct to get all info necessary for
718  * subsequent operations.
719  */
720 static int
721 apic_probe()
722 {
723 	uint32_t mpct_addr, ebda_start = 0, base_mem_end;
724 	caddr_t	biosdatap;
725 	caddr_t	mpct;
726 	caddr_t	fptr;
727 	int	i, mpct_size, mapsize, retval = PSM_FAILURE;
728 	ushort_t	ebda_seg, base_mem_size;
729 	struct	apic_mpfps_hdr	*fpsp;
730 	struct	apic_mp_cnf_hdr	*hdrp;
731 	int bypass_cpu_and_ioapics_in_mptables;
732 	int acpi_user_options;
733 
734 	if (apic_forceload < 0)
735 		return (retval);
736 
737 	/* Allow override for MADT-only mode */
738 	acpi_user_options = ddi_prop_get_int(DDI_DEV_T_ANY, ddi_root_node(), 0,
739 	    "acpi-user-options", 0);
740 	apic_use_acpi_madt_only = ((acpi_user_options & ACPI_OUSER_MADT) != 0);
741 
742 	/* Allow apic_use_acpi to override MADT-only mode */
743 	if (!apic_use_acpi)
744 		apic_use_acpi_madt_only = 0;
745 
746 	retval = acpi_probe();
747 
748 	/*
749 	 * mapin the bios data area 40:0
750 	 * 40:13h - two-byte location reports the base memory size
751 	 * 40:0Eh - two-byte location for the exact starting address of
752 	 *	    the EBDA segment for EISA
753 	 */
754 	biosdatap = psm_map_phys(0x400, 0x20, PROT_READ);
755 	if (!biosdatap)
756 		return (retval);
757 	fpsp = (struct apic_mpfps_hdr *)NULL;
758 	mapsize = MPFPS_RAM_WIN_LEN;
759 	/*LINTED: pointer cast may result in improper alignment */
760 	ebda_seg = *((ushort_t *)(biosdatap+0xe));
761 	/* check the 1k of EBDA */
762 	if (ebda_seg) {
763 		ebda_start = ((uint32_t)ebda_seg) << 4;
764 		fptr = psm_map_phys(ebda_start, MPFPS_RAM_WIN_LEN, PROT_READ);
765 		if (fptr) {
766 			if (!(fpsp =
767 			    apic_find_fps_sig(fptr, MPFPS_RAM_WIN_LEN)))
768 				psm_unmap_phys(fptr, MPFPS_RAM_WIN_LEN);
769 		}
770 	}
771 	/* If not in EBDA, check the last k of system base memory */
772 	if (!fpsp) {
773 		/*LINTED: pointer cast may result in improper alignment */
774 		base_mem_size = *((ushort_t *)(biosdatap + 0x13));
775 
776 		if (base_mem_size > 512)
777 			base_mem_end = 639 * 1024;
778 		else
779 			base_mem_end = 511 * 1024;
780 		/* if ebda == last k of base mem, skip to check BIOS ROM */
781 		if (base_mem_end != ebda_start) {
782 
783 			fptr = psm_map_phys(base_mem_end, MPFPS_RAM_WIN_LEN,
784 			    PROT_READ);
785 
786 			if (fptr) {
787 				if (!(fpsp = apic_find_fps_sig(fptr,
788 				    MPFPS_RAM_WIN_LEN)))
789 					psm_unmap_phys(fptr, MPFPS_RAM_WIN_LEN);
790 			}
791 		}
792 	}
793 	psm_unmap_phys(biosdatap, 0x20);
794 
795 	/* If still cannot find it, check the BIOS ROM space */
796 	if (!fpsp) {
797 		mapsize = MPFPS_ROM_WIN_LEN;
798 		fptr = psm_map_phys(MPFPS_ROM_WIN_START,
799 		    MPFPS_ROM_WIN_LEN, PROT_READ);
800 		if (fptr) {
801 			if (!(fpsp =
802 			    apic_find_fps_sig(fptr, MPFPS_ROM_WIN_LEN))) {
803 				psm_unmap_phys(fptr, MPFPS_ROM_WIN_LEN);
804 				return (retval);
805 			}
806 		}
807 	}
808 
809 	if (apic_checksum((caddr_t)fpsp, fpsp->mpfps_length * 16) != 0) {
810 		psm_unmap_phys(fptr, MPFPS_ROM_WIN_LEN);
811 		return (retval);
812 	}
813 
814 	apic_spec_rev = fpsp->mpfps_spec_rev;
815 	if ((apic_spec_rev != 04) && (apic_spec_rev != 01)) {
816 		psm_unmap_phys(fptr, MPFPS_ROM_WIN_LEN);
817 		return (retval);
818 	}
819 
820 	/* check IMCR is present or not */
821 	apic_imcrp = fpsp->mpfps_featinfo2 & MPFPS_FEATINFO2_IMCRP;
822 
823 	/* check default configuration (dual CPUs) */
824 	if ((apic_defconf = fpsp->mpfps_featinfo1) != 0) {
825 		psm_unmap_phys(fptr, mapsize);
826 		return (apic_handle_defconf());
827 	}
828 
829 	/* MP Configuration Table */
830 	mpct_addr = (uint32_t)(fpsp->mpfps_mpct_paddr);
831 
832 	psm_unmap_phys(fptr, mapsize); /* unmap floating ptr struct */
833 
834 	/*
835 	 * Map in enough memory for the MP Configuration Table Header.
836 	 * Use this table to read the total length of the BIOS data and
837 	 * map in all the info
838 	 */
839 	/*LINTED: pointer cast may result in improper alignment */
840 	hdrp = (struct apic_mp_cnf_hdr *)psm_map_phys(mpct_addr,
841 	    sizeof (struct apic_mp_cnf_hdr), PROT_READ);
842 	if (!hdrp)
843 		return (retval);
844 
845 	/* check mp configuration table signature PCMP */
846 	if (hdrp->mpcnf_sig != 0x504d4350) {
847 		psm_unmap_phys((caddr_t)hdrp, sizeof (struct apic_mp_cnf_hdr));
848 		return (retval);
849 	}
850 	mpct_size = (int)hdrp->mpcnf_tbl_length;
851 
852 	apic_set_pwroff_method_from_mpcnfhdr(hdrp);
853 
854 	psm_unmap_phys((caddr_t)hdrp, sizeof (struct apic_mp_cnf_hdr));
855 
856 	if ((retval == PSM_SUCCESS) && !apic_use_acpi_madt_only) {
857 		/* This is an ACPI machine No need for further checks */
858 		return (retval);
859 	}
860 
861 	/*
862 	 * Map in the entries for this machine, ie. Processor
863 	 * Entry Tables, Bus Entry Tables, etc.
864 	 * They are in fixed order following one another
865 	 */
866 	mpct = psm_map_phys(mpct_addr, mpct_size, PROT_READ);
867 	if (!mpct)
868 		return (retval);
869 
870 	if (apic_checksum(mpct, mpct_size) != 0)
871 		goto apic_fail1;
872 
873 
874 	/*LINTED: pointer cast may result in improper alignment */
875 	hdrp = (struct apic_mp_cnf_hdr *)mpct;
876 	/*LINTED: pointer cast may result in improper alignment */
877 	apicadr = (uint32_t *)psm_map_phys((uint32_t)hdrp->mpcnf_local_apic,
878 	    APIC_LOCAL_MEMLEN, PROT_READ | PROT_WRITE);
879 	if (!apicadr)
880 		goto apic_fail1;
881 
882 	/* Parse all information in the tables */
883 	bypass_cpu_and_ioapics_in_mptables = (retval == PSM_SUCCESS);
884 	if (apic_parse_mpct(mpct, bypass_cpu_and_ioapics_in_mptables) ==
885 	    PSM_SUCCESS)
886 		return (PSM_SUCCESS);
887 
888 	for (i = 0; i < apic_io_max; i++)
889 		psm_unmap_phys((caddr_t)apicioadr[i], APIC_IO_MEMLEN);
890 	if (apic_cpus)
891 		kmem_free(apic_cpus, sizeof (*apic_cpus) * apic_nproc);
892 	if (apicadr)
893 		psm_unmap_phys((caddr_t)apicadr, APIC_LOCAL_MEMLEN);
894 apic_fail1:
895 	psm_unmap_phys(mpct, mpct_size);
896 	return (retval);
897 }
898 
899 static void
900 apic_set_pwroff_method_from_mpcnfhdr(struct apic_mp_cnf_hdr *hdrp)
901 {
902 	int	i;
903 
904 	for (i = 0; i < (sizeof (apic_mps_ids) / sizeof (apic_mps_ids[0]));
905 	    i++) {
906 		if ((strncmp(hdrp->mpcnf_oem_str, apic_mps_ids[i].oem_id,
907 		    strlen(apic_mps_ids[i].oem_id)) == 0) &&
908 		    (strncmp(hdrp->mpcnf_prod_str, apic_mps_ids[i].prod_id,
909 		    strlen(apic_mps_ids[i].prod_id)) == 0)) {
910 
911 			apic_poweroff_method = apic_mps_ids[i].poweroff_method;
912 			break;
913 		}
914 	}
915 
916 	if (apic_debug_mps_id != 0) {
917 		cmn_err(CE_CONT, "pcplusmp: MPS OEM ID = '%c%c%c%c%c%c%c%c'"
918 		    "Product ID = '%c%c%c%c%c%c%c%c%c%c%c%c'\n",
919 		    hdrp->mpcnf_oem_str[0],
920 		    hdrp->mpcnf_oem_str[1],
921 		    hdrp->mpcnf_oem_str[2],
922 		    hdrp->mpcnf_oem_str[3],
923 		    hdrp->mpcnf_oem_str[4],
924 		    hdrp->mpcnf_oem_str[5],
925 		    hdrp->mpcnf_oem_str[6],
926 		    hdrp->mpcnf_oem_str[7],
927 		    hdrp->mpcnf_prod_str[0],
928 		    hdrp->mpcnf_prod_str[1],
929 		    hdrp->mpcnf_prod_str[2],
930 		    hdrp->mpcnf_prod_str[3],
931 		    hdrp->mpcnf_prod_str[4],
932 		    hdrp->mpcnf_prod_str[5],
933 		    hdrp->mpcnf_prod_str[6],
934 		    hdrp->mpcnf_prod_str[7],
935 		    hdrp->mpcnf_prod_str[8],
936 		    hdrp->mpcnf_prod_str[9],
937 		    hdrp->mpcnf_prod_str[10],
938 		    hdrp->mpcnf_prod_str[11]);
939 	}
940 }
941 
942 static int
943 acpi_probe(void)
944 {
945 	int			i, id, intmax, ver, index, rv;
946 	int			acpi_verboseflags = 0;
947 	int			madt_seen, madt_size;
948 	APIC_HEADER		*ap;
949 	MADT_PROCESSOR_APIC	*mpa;
950 	MADT_IO_APIC		*mia;
951 	MADT_IO_SAPIC		*misa;
952 	MADT_INTERRUPT_OVERRIDE	*mio;
953 	MADT_NMI_SOURCE		*mns;
954 	MADT_INTERRUPT_SOURCE	*mis;
955 	MADT_LOCAL_APIC_NMI	*mlan;
956 	MADT_ADDRESS_OVERRIDE	*mao;
957 	ACPI_OBJECT_LIST 	arglist;
958 	ACPI_OBJECT		arg;
959 	int			sci;
960 	iflag_t			sci_flags;
961 	volatile int32_t	*ioapic;
962 	char			local_ids[NCPU];
963 	char			proc_ids[NCPU];
964 	uchar_t			hid;
965 
966 	if (!apic_use_acpi)
967 		return (PSM_FAILURE);
968 
969 	if (AcpiGetFirmwareTable(APIC_SIG, 1, ACPI_LOGICAL_ADDRESSING,
970 	    (ACPI_TABLE_HEADER **) &acpi_mapic_dtp) != AE_OK)
971 		return (PSM_FAILURE);
972 
973 	apicadr = (uint32_t *)psm_map_phys(
974 	    (uint32_t)acpi_mapic_dtp->LocalApicAddress,
975 	    APIC_LOCAL_MEMLEN, PROT_READ | PROT_WRITE);
976 	if (!apicadr)
977 		return (PSM_FAILURE);
978 
979 	id = apicadr[APIC_LID_REG];
980 	local_ids[0] = (uchar_t)(((uint_t)id) >> 24);
981 	apic_nproc = index = 1;
982 	apic_io_max = 0;
983 
984 	ap = (APIC_HEADER *) (acpi_mapic_dtp + 1);
985 	madt_size = acpi_mapic_dtp->Length;
986 	madt_seen = sizeof (*acpi_mapic_dtp);
987 
988 	while (madt_seen < madt_size) {
989 		switch (ap->Type) {
990 		case APIC_PROCESSOR:
991 			mpa = (MADT_PROCESSOR_APIC *) ap;
992 			if (mpa->ProcessorEnabled) {
993 				if (mpa->LocalApicId == local_ids[0])
994 					proc_ids[0] = mpa->ProcessorId;
995 				else if (apic_nproc < NCPU) {
996 					local_ids[index] = mpa->LocalApicId;
997 					proc_ids[index] = mpa->ProcessorId;
998 					index++;
999 					apic_nproc++;
1000 				} else
1001 					cmn_err(CE_WARN, "pcplusmp: exceeded "
1002 					    "maximum no. of CPUs (= %d)", NCPU);
1003 			}
1004 			break;
1005 
1006 		case APIC_IO:
1007 			mia = (MADT_IO_APIC *) ap;
1008 			if (apic_io_max < MAX_IO_APIC) {
1009 				apic_io_id[apic_io_max] = mia->IoApicId;
1010 				apic_io_vectbase[apic_io_max] =
1011 				    mia->Interrupt;
1012 				ioapic = apicioadr[apic_io_max] =
1013 				    (int32_t *)psm_map_phys(
1014 				    (uint32_t)mia->Address,
1015 				    APIC_IO_MEMLEN, PROT_READ | PROT_WRITE);
1016 				if (!ioapic)
1017 					goto cleanup;
1018 				apic_io_max++;
1019 			}
1020 			break;
1021 
1022 		case APIC_XRUPT_OVERRIDE:
1023 			mio = (MADT_INTERRUPT_OVERRIDE *) ap;
1024 			if (acpi_isop == NULL)
1025 				acpi_isop = mio;
1026 			acpi_iso_cnt++;
1027 			break;
1028 
1029 		case APIC_NMI:
1030 			/* UNIMPLEMENTED */
1031 			mns = (MADT_NMI_SOURCE *) ap;
1032 			if (acpi_nmi_sp == NULL)
1033 				acpi_nmi_sp = mns;
1034 			acpi_nmi_scnt++;
1035 
1036 			cmn_err(CE_NOTE, "!apic: nmi source: %d %d %d\n",
1037 				mns->Interrupt, mns->Polarity,
1038 				mns->TriggerMode);
1039 			break;
1040 
1041 		case APIC_LOCAL_NMI:
1042 			/* UNIMPLEMENTED */
1043 			mlan = (MADT_LOCAL_APIC_NMI *) ap;
1044 			if (acpi_nmi_cp == NULL)
1045 				acpi_nmi_cp = mlan;
1046 			acpi_nmi_ccnt++;
1047 
1048 			cmn_err(CE_NOTE, "!apic: local nmi: %d %d %d %d\n",
1049 				mlan->ProcessorId, mlan->Polarity,
1050 				mlan->TriggerMode, mlan->Lint);
1051 			break;
1052 
1053 		case APIC_ADDRESS_OVERRIDE:
1054 			/* UNIMPLEMENTED */
1055 			mao = (MADT_ADDRESS_OVERRIDE *) ap;
1056 			cmn_err(CE_NOTE, "!apic: address override: %lx\n",
1057 				(long)mao->Address);
1058 			break;
1059 
1060 		case APIC_IO_SAPIC:
1061 			/* UNIMPLEMENTED */
1062 			misa = (MADT_IO_SAPIC *) ap;
1063 
1064 			cmn_err(CE_NOTE, "!apic: io sapic: %d %d %lx\n",
1065 				misa->IoSapicId, misa->InterruptBase,
1066 				(long)misa->Address);
1067 			break;
1068 
1069 		case APIC_XRUPT_SOURCE:
1070 			/* UNIMPLEMENTED */
1071 			mis = (MADT_INTERRUPT_SOURCE *) ap;
1072 
1073 			cmn_err(CE_NOTE,
1074 				"!apic: irq source: %d %d %d %d %d %d %d\n",
1075 				mis->ProcessorId, mis->ProcessorEid,
1076 				mis->Interrupt, mis->Polarity,
1077 				mis->TriggerMode, mis->InterruptType,
1078 				mis->IoSapicVector);
1079 			break;
1080 		case APIC_RESERVED:
1081 		default:
1082 			goto cleanup;
1083 		}
1084 
1085 		/* advance to next entry */
1086 		madt_seen += ap->Length;
1087 		ap = (APIC_HEADER *)(((char *)ap) + ap->Length);
1088 	}
1089 
1090 	if ((apic_cpus = kmem_zalloc(sizeof (*apic_cpus) * apic_nproc,
1091 	    KM_NOSLEEP)) == NULL)
1092 		goto cleanup;
1093 
1094 	apic_cpumask = (1 << apic_nproc) - 1;
1095 
1096 	/*
1097 	 * ACPI doesn't provide the local apic ver, get it directly from the
1098 	 * local apic
1099 	 */
1100 	ver = apicadr[APIC_VERS_REG];
1101 	for (i = 0; i < apic_nproc; i++) {
1102 		apic_cpus[i].aci_local_id = local_ids[i];
1103 		apic_cpus[i].aci_local_ver = (uchar_t)(ver & 0xFF);
1104 	}
1105 	for (i = 0; i < apic_io_max; i++) {
1106 		ioapic = apicioadr[i];
1107 
1108 		/*
1109 		 * need to check Sitka on the following acpi problem
1110 		 * On the Sitka, the ioapic's apic_id field isn't reporting
1111 		 * the actual io apic id. We have reported this problem
1112 		 * to Intel. Until they fix the problem, we will get the
1113 		 * actual id directly from the ioapic.
1114 		 */
1115 		ioapic[APIC_IO_REG] = APIC_ID_CMD;
1116 		id = ioapic[APIC_IO_DATA];
1117 		hid = (uchar_t)(((uint_t)id) >> 24);
1118 
1119 		if (hid != apic_io_id[i]) {
1120 			if (apic_io_id[i] == 0)
1121 				apic_io_id[i] = hid;
1122 			else { /* set ioapic id to whatever reported by ACPI */
1123 				id = ((int32_t)apic_io_id[i]) << 24;
1124 				ioapic[APIC_IO_REG] = APIC_ID_CMD;
1125 				ioapic[APIC_IO_DATA] = id;
1126 			}
1127 		}
1128 		ioapic[APIC_IO_REG] = APIC_VERS_CMD;
1129 		ver = ioapic[APIC_IO_DATA];
1130 		apic_io_ver[i] = (uchar_t)(ver & 0xff);
1131 		intmax = (ver >> 16) & 0xff;
1132 		apic_io_vectend[i] = apic_io_vectbase[i] + intmax;
1133 	}
1134 
1135 
1136 	/*
1137 	 * Process SCI configuration here
1138 	 * An error may be returned here if
1139 	 * acpi-user-options specifies legacy mode
1140 	 * (no SCI, no ACPI mode)
1141 	 */
1142 	if (acpica_get_sci(&sci, &sci_flags) != AE_OK)
1143 		sci = -1;
1144 
1145 	/*
1146 	 * Now call acpi_init() to generate namespaces
1147 	 * If this fails, we don't attempt to use ACPI
1148 	 * even if we were able to get a MADT above
1149 	 */
1150 	if (acpica_init() != AE_OK)
1151 		goto cleanup;
1152 
1153 	/*
1154 	 * Squirrel away the SCI and flags for later on
1155 	 * in apic_picinit() when we're ready
1156 	 */
1157 	apic_sci_vect = sci;
1158 	apic_sci_flags = sci_flags;
1159 
1160 	if (apic_verbose & APIC_VERBOSE_IRQ_FLAG)
1161 		acpi_verboseflags |= PSM_VERBOSE_IRQ_FLAG;
1162 
1163 	if (apic_verbose & APIC_VERBOSE_POWEROFF_FLAG)
1164 		acpi_verboseflags |= PSM_VERBOSE_POWEROFF_FLAG;
1165 
1166 	if (apic_verbose & APIC_VERBOSE_POWEROFF_PAUSE_FLAG)
1167 		acpi_verboseflags |= PSM_VERBOSE_POWEROFF_PAUSE_FLAG;
1168 
1169 	if (acpi_psm_init(apic_psm_info.p_mach_idstring, acpi_verboseflags) ==
1170 	    ACPI_PSM_FAILURE)
1171 		goto cleanup;
1172 
1173 	/* Enable ACPI APIC interrupt routing */
1174 	arglist.Count = 1;
1175 	arglist.Pointer = &arg;
1176 	arg.Type = ACPI_TYPE_INTEGER;
1177 	arg.Integer.Value = ACPI_APIC_MODE;	/* 1 */
1178 	rv = AcpiEvaluateObject(NULL, "\\_PIC", &arglist, NULL);
1179 	if (rv == AE_OK) {
1180 		build_reserved_irqlist((uchar_t *)apic_reserved_irqlist);
1181 		apic_enable_acpi = 1;
1182 		if (apic_use_acpi_madt_only) {
1183 			cmn_err(CE_CONT,
1184 			    "?Using ACPI for CPU/IOAPIC information ONLY\n");
1185 		}
1186 		return (PSM_SUCCESS);
1187 	}
1188 	/* if setting APIC mode failed above, we fall through to cleanup */
1189 
1190 cleanup:
1191 	if (apicadr != NULL) {
1192 		psm_unmap_phys((caddr_t)apicadr, APIC_LOCAL_MEMLEN);
1193 		apicadr = NULL;
1194 	}
1195 	apic_nproc = 0;
1196 	for (i = 0; i < apic_io_max; i++) {
1197 		psm_unmap_phys((caddr_t)apicioadr[i], APIC_IO_MEMLEN);
1198 		apicioadr[i] = NULL;
1199 	}
1200 	apic_io_max = 0;
1201 	acpi_isop = NULL;
1202 	acpi_iso_cnt = 0;
1203 	acpi_nmi_sp = NULL;
1204 	acpi_nmi_scnt = 0;
1205 	acpi_nmi_cp = NULL;
1206 	acpi_nmi_ccnt = 0;
1207 	return (PSM_FAILURE);
1208 }
1209 
1210 /*
1211  * Handle default configuration. Fill in reqd global variables & tables
1212  * Fill all details as MP table does not give any more info
1213  */
1214 static int
1215 apic_handle_defconf()
1216 {
1217 	uint_t	lid;
1218 
1219 	/*LINTED: pointer cast may result in improper alignment */
1220 	apicioadr[0] = (int32_t *)psm_map_phys(APIC_IO_ADDR,
1221 	    APIC_IO_MEMLEN, PROT_READ | PROT_WRITE);
1222 	/*LINTED: pointer cast may result in improper alignment */
1223 	apicadr = (uint32_t *)psm_map_phys(APIC_LOCAL_ADDR,
1224 	    APIC_LOCAL_MEMLEN, PROT_READ | PROT_WRITE);
1225 	apic_cpus = (apic_cpus_info_t *)
1226 	    kmem_zalloc(sizeof (*apic_cpus) * 2, KM_NOSLEEP);
1227 	if ((!apicadr) || (!apicioadr[0]) || (!apic_cpus))
1228 		goto apic_handle_defconf_fail;
1229 	apic_cpumask = 3;
1230 	apic_nproc = 2;
1231 	lid = apicadr[APIC_LID_REG];
1232 	apic_cpus[0].aci_local_id = (uchar_t)(lid >> APIC_ID_BIT_OFFSET);
1233 	/*
1234 	 * According to the PC+MP spec 1.1, the local ids
1235 	 * for the default configuration has to be 0 or 1
1236 	 */
1237 	if (apic_cpus[0].aci_local_id == 1)
1238 		apic_cpus[1].aci_local_id = 0;
1239 	else if (apic_cpus[0].aci_local_id == 0)
1240 		apic_cpus[1].aci_local_id = 1;
1241 	else
1242 		goto apic_handle_defconf_fail;
1243 
1244 	apic_io_id[0] = 2;
1245 	apic_io_max = 1;
1246 	if (apic_defconf >= 5) {
1247 		apic_cpus[0].aci_local_ver = APIC_INTEGRATED_VERS;
1248 		apic_cpus[1].aci_local_ver = APIC_INTEGRATED_VERS;
1249 		apic_io_ver[0] = APIC_INTEGRATED_VERS;
1250 	} else {
1251 		apic_cpus[0].aci_local_ver = 0;		/* 82489 DX */
1252 		apic_cpus[1].aci_local_ver = 0;
1253 		apic_io_ver[0] = 0;
1254 	}
1255 	if (apic_defconf == 2 || apic_defconf == 3 || apic_defconf == 6)
1256 		eisa_level_intr_mask = (inb(EISA_LEVEL_CNTL + 1) << 8) |
1257 		    inb(EISA_LEVEL_CNTL) | ((uint_t)INT32_MAX + 1);
1258 	return (PSM_SUCCESS);
1259 
1260 apic_handle_defconf_fail:
1261 	if (apic_cpus)
1262 		kmem_free(apic_cpus, sizeof (*apic_cpus) * 2);
1263 	if (apicadr)
1264 		psm_unmap_phys((caddr_t)apicadr, APIC_LOCAL_MEMLEN);
1265 	if (apicioadr[0])
1266 		psm_unmap_phys((caddr_t)apicioadr[0], APIC_IO_MEMLEN);
1267 	return (PSM_FAILURE);
1268 }
1269 
1270 /* Parse the entries in MP configuration table and collect info that we need */
1271 static int
1272 apic_parse_mpct(caddr_t mpct, int bypass_cpus_and_ioapics)
1273 {
1274 	struct	apic_procent	*procp;
1275 	struct	apic_bus	*busp;
1276 	struct	apic_io_entry	*ioapicp;
1277 	struct	apic_io_intr	*intrp;
1278 	volatile int32_t	*ioapic;
1279 	uint_t	lid;
1280 	int	id;
1281 	uchar_t hid;
1282 
1283 	/*LINTED: pointer cast may result in improper alignment */
1284 	procp = (struct apic_procent *)(mpct + sizeof (struct apic_mp_cnf_hdr));
1285 
1286 	/* No need to count cpu entries if we won't use them */
1287 	if (!bypass_cpus_and_ioapics) {
1288 
1289 		/* Find max # of CPUS and allocate structure accordingly */
1290 		apic_nproc = 0;
1291 		while (procp->proc_entry == APIC_CPU_ENTRY) {
1292 			if (procp->proc_cpuflags & CPUFLAGS_EN) {
1293 				apic_nproc++;
1294 			}
1295 			procp++;
1296 		}
1297 		if (apic_nproc > NCPU)
1298 			cmn_err(CE_WARN, "pcplusmp: exceeded "
1299 			    "maximum no. of CPUs (= %d)", NCPU);
1300 		if (!apic_nproc || !(apic_cpus = (apic_cpus_info_t *)
1301 		    kmem_zalloc(sizeof (*apic_cpus)*apic_nproc, KM_NOSLEEP)))
1302 			return (PSM_FAILURE);
1303 	}
1304 
1305 	/*LINTED: pointer cast may result in improper alignment */
1306 	procp = (struct apic_procent *)(mpct + sizeof (struct apic_mp_cnf_hdr));
1307 
1308 	/*
1309 	 * start with index 1 as 0 needs to be filled in with Boot CPU, but
1310 	 * if we're bypassing this information, it has already been filled
1311 	 * in by acpi_probe(), so don't overwrite it.
1312 	 */
1313 	if (!bypass_cpus_and_ioapics)
1314 		apic_nproc = 1;
1315 
1316 	while (procp->proc_entry == APIC_CPU_ENTRY) {
1317 		/* check whether the cpu exists or not */
1318 		if (!bypass_cpus_and_ioapics &&
1319 		    procp->proc_cpuflags & CPUFLAGS_EN) {
1320 			if (procp->proc_cpuflags & CPUFLAGS_BP) { /* Boot CPU */
1321 				lid = apicadr[APIC_LID_REG];
1322 				apic_cpus[0].aci_local_id = procp->proc_apicid;
1323 				if (apic_cpus[0].aci_local_id !=
1324 				    (uchar_t)(lid >> APIC_ID_BIT_OFFSET)) {
1325 					return (PSM_FAILURE);
1326 				}
1327 				apic_cpus[0].aci_local_ver =
1328 				    procp->proc_version;
1329 			} else {
1330 
1331 				apic_cpus[apic_nproc].aci_local_id =
1332 				    procp->proc_apicid;
1333 				apic_cpus[apic_nproc].aci_local_ver =
1334 				    procp->proc_version;
1335 				apic_nproc++;
1336 
1337 			}
1338 		}
1339 		procp++;
1340 	}
1341 
1342 	if (!bypass_cpus_and_ioapics) {
1343 		/* convert the number of processors into a cpumask */
1344 		apic_cpumask = (1 << apic_nproc) - 1;
1345 	}
1346 
1347 	/*
1348 	 * Save start of bus entries for later use.
1349 	 * Get EISA level cntrl if EISA bus is present.
1350 	 * Also get the CPI bus id for single CPI bus case
1351 	 */
1352 	apic_busp = busp = (struct apic_bus *)procp;
1353 	while (busp->bus_entry == APIC_BUS_ENTRY) {
1354 		lid = apic_find_bus_type((char *)&busp->bus_str1);
1355 		if (lid	== BUS_EISA) {
1356 			eisa_level_intr_mask = (inb(EISA_LEVEL_CNTL + 1) << 8) |
1357 			    inb(EISA_LEVEL_CNTL) | ((uint_t)INT32_MAX + 1);
1358 		} else if (lid == BUS_PCI) {
1359 			/*
1360 			 * apic_single_pci_busid will be used only if
1361 			 * apic_pic_bus_total is equal to 1
1362 			 */
1363 			apic_pci_bus_total++;
1364 			apic_single_pci_busid = busp->bus_id;
1365 		}
1366 		busp++;
1367 	}
1368 
1369 	ioapicp = (struct apic_io_entry *)busp;
1370 
1371 	if (!bypass_cpus_and_ioapics)
1372 		apic_io_max = 0;
1373 	do {
1374 		if (!bypass_cpus_and_ioapics && apic_io_max < MAX_IO_APIC) {
1375 			if (ioapicp->io_flags & IOAPIC_FLAGS_EN) {
1376 				apic_io_id[apic_io_max] = ioapicp->io_apicid;
1377 				apic_io_ver[apic_io_max] = ioapicp->io_version;
1378 		/*LINTED: pointer cast may result in improper alignment */
1379 				apicioadr[apic_io_max] =
1380 				    (int32_t *)psm_map_phys(
1381 				    (uint32_t)ioapicp->io_apic_addr,
1382 				    APIC_IO_MEMLEN, PROT_READ | PROT_WRITE);
1383 
1384 				if (!apicioadr[apic_io_max])
1385 					return (PSM_FAILURE);
1386 
1387 				ioapic = apicioadr[apic_io_max];
1388 				ioapic[APIC_IO_REG] = APIC_ID_CMD;
1389 				id = ioapic[APIC_IO_DATA];
1390 				hid = (uchar_t)(((uint_t)id) >> 24);
1391 
1392 				if (hid != apic_io_id[apic_io_max]) {
1393 					if (apic_io_id[apic_io_max] == 0)
1394 						apic_io_id[apic_io_max] = hid;
1395 					else {
1396 						/*
1397 						 * set ioapic id to whatever
1398 						 * reported by MPS
1399 						 *
1400 						 * may not need to set index
1401 						 * again ???
1402 						 * take it out and try
1403 						 */
1404 
1405 						id = ((int32_t)
1406 						    apic_io_id[apic_io_max]) <<
1407 						    24;
1408 
1409 						ioapic[APIC_IO_REG] =
1410 						    APIC_ID_CMD;
1411 
1412 						ioapic[APIC_IO_DATA] = id;
1413 
1414 					}
1415 				}
1416 				apic_io_max++;
1417 			}
1418 		}
1419 		ioapicp++;
1420 	} while (ioapicp->io_entry == APIC_IO_ENTRY);
1421 
1422 	apic_io_intrp = (struct apic_io_intr *)ioapicp;
1423 
1424 	intrp = apic_io_intrp;
1425 	while (intrp->intr_entry == APIC_IO_INTR_ENTRY) {
1426 		if ((intrp->intr_irq > APIC_MAX_ISA_IRQ) ||
1427 		    (apic_find_bus(intrp->intr_busid) == BUS_PCI)) {
1428 			apic_irq_translate = 1;
1429 			break;
1430 		}
1431 		intrp++;
1432 	}
1433 
1434 	return (PSM_SUCCESS);
1435 }
1436 
1437 static struct apic_mpfps_hdr *
1438 apic_find_fps_sig(caddr_t cptr, int len)
1439 {
1440 	int	i;
1441 
1442 	/* Look for the pattern "_MP_" */
1443 	for (i = 0; i < len; i += 16) {
1444 		if ((*(cptr+i) == '_') &&
1445 		    (*(cptr+i+1) == 'M') &&
1446 		    (*(cptr+i+2) == 'P') &&
1447 		    (*(cptr+i+3) == '_'))
1448 		    /*LINTED: pointer cast may result in improper alignment */
1449 			return ((struct apic_mpfps_hdr *)(cptr + i));
1450 	}
1451 	return (NULL);
1452 }
1453 
1454 static int
1455 apic_checksum(caddr_t bptr, int len)
1456 {
1457 	int	i;
1458 	uchar_t	cksum;
1459 
1460 	cksum = 0;
1461 	for (i = 0; i < len; i++)
1462 		cksum += *bptr++;
1463 	return ((int)cksum);
1464 }
1465 
1466 
1467 /*
1468  * Initialise vector->ipl and ipl->pri arrays. level_intr and irqtable
1469  * are also set to NULL. vector->irq is set to a value which cannot map
1470  * to a real irq to show that it is free.
1471  */
1472 void
1473 apic_init()
1474 {
1475 	int	i;
1476 	int	*iptr;
1477 
1478 	int	j = 1;
1479 	apic_ipltopri[0] = APIC_VECTOR_PER_IPL; /* leave 0 for idle */
1480 	for (i = 0; i < (APIC_AVAIL_VECTOR / APIC_VECTOR_PER_IPL); i++) {
1481 		if ((i < ((APIC_AVAIL_VECTOR / APIC_VECTOR_PER_IPL) - 1)) &&
1482 		    (apic_vectortoipl[i + 1] == apic_vectortoipl[i]))
1483 			/* get to highest vector at the same ipl */
1484 			continue;
1485 		for (; j <= apic_vectortoipl[i]; j++) {
1486 			apic_ipltopri[j] = (i << APIC_IPL_SHIFT) +
1487 			    APIC_BASE_VECT;
1488 		}
1489 	}
1490 	for (; j < MAXIPL + 1; j++)
1491 		/* fill up any empty ipltopri slots */
1492 		apic_ipltopri[j] = (i << APIC_IPL_SHIFT) + APIC_BASE_VECT;
1493 
1494 	/* cpu 0 is always up */
1495 	apic_cpus[0].aci_status = APIC_CPU_ONLINE | APIC_CPU_INTR_ENABLE;
1496 
1497 	iptr = (int *)&apic_irq_table[0];
1498 	for (i = 0; i <= APIC_MAX_VECTOR; i++) {
1499 		apic_level_intr[i] = 0;
1500 		*iptr++ = NULL;
1501 		apic_vector_to_irq[i] = APIC_RESV_IRQ;
1502 		apic_reprogram_info[i].valid = 0;
1503 		apic_reprogram_info[i].bindcpu = 0;
1504 		apic_reprogram_info[i].timeouts = 0;
1505 	}
1506 
1507 	/*
1508 	 * Allocate a dummy irq table entry for the reserved entry.
1509 	 * This takes care of the race between removing an irq and
1510 	 * clock detecting a CPU in that irq during interrupt load
1511 	 * sampling.
1512 	 */
1513 	apic_irq_table[APIC_RESV_IRQ] =
1514 	    kmem_zalloc(sizeof (apic_irq_t), KM_NOSLEEP);
1515 
1516 	mutex_init(&airq_mutex, NULL, MUTEX_DEFAULT, NULL);
1517 	mutex_init(&apic_reprogram_timeout_mutex, NULL, MUTEX_DEFAULT, NULL);
1518 #if defined(__amd64)
1519 	/*
1520 	 * Make cpu-specific interrupt info point to cr8pri vector
1521 	 */
1522 	for (i = 0; i <= MAXIPL; i++)
1523 		apic_cr8pri[i] = apic_ipltopri[i] >> APIC_IPL_SHIFT;
1524 	CPU->cpu_pri_data = apic_cr8pri;
1525 	intpri_use_cr8 = 1;
1526 #endif	/* __amd64 */
1527 }
1528 
1529 /*
1530  * handler for APIC Error interrupt. Just print a warning and continue
1531  */
1532 static int
1533 apic_error_intr()
1534 {
1535 	uint_t	error0, error1, error;
1536 	uint_t	i;
1537 
1538 	/*
1539 	 * We need to write before read as per 7.4.17 of system prog manual.
1540 	 * We do both and or the results to be safe
1541 	 */
1542 	error0 = apicadr[APIC_ERROR_STATUS];
1543 	apicadr[APIC_ERROR_STATUS] = 0;
1544 	error1 = apicadr[APIC_ERROR_STATUS];
1545 	error = error0 | error1;
1546 
1547 	/*
1548 	 * Prevent more than 1 CPU from handling error interrupt causing
1549 	 * double printing (interleave of characters from multiple
1550 	 * CPU's when using prom_printf)
1551 	 */
1552 	if (lock_try(&apic_error_lock) == 0)
1553 		return (error ? DDI_INTR_CLAIMED : DDI_INTR_UNCLAIMED);
1554 	if (error) {
1555 #if	DEBUG
1556 		if (apic_debug)
1557 			debug_enter("pcplusmp: APIC Error interrupt received");
1558 #endif /* DEBUG */
1559 		if (apic_panic_on_apic_error)
1560 			cmn_err(CE_PANIC,
1561 			    "APIC Error interrupt on CPU %d. Status = %x\n",
1562 			    psm_get_cpu_id(), error);
1563 		else {
1564 			if ((error & ~APIC_CS_ERRORS) == 0) {
1565 				/* cksum error only */
1566 				apic_error |= APIC_ERR_APIC_ERROR;
1567 				apic_apic_error |= error;
1568 				apic_num_apic_errors++;
1569 				apic_num_cksum_errors++;
1570 			} else {
1571 				/*
1572 				 * prom_printf is the best shot we have of
1573 				 * something which is problem free from
1574 				 * high level/NMI type of interrupts
1575 				 */
1576 				prom_printf("APIC Error interrupt on CPU %d. "
1577 				    "Status 0 = %x, Status 1 = %x\n",
1578 				    psm_get_cpu_id(), error0, error1);
1579 				apic_error |= APIC_ERR_APIC_ERROR;
1580 				apic_apic_error |= error;
1581 				apic_num_apic_errors++;
1582 				for (i = 0; i < apic_error_display_delay; i++) {
1583 					tenmicrosec();
1584 				}
1585 				/*
1586 				 * provide more delay next time limited to
1587 				 * roughly 1 clock tick time
1588 				 */
1589 				if (apic_error_display_delay < 500)
1590 					apic_error_display_delay *= 2;
1591 			}
1592 		}
1593 		lock_clear(&apic_error_lock);
1594 		return (DDI_INTR_CLAIMED);
1595 	} else {
1596 		lock_clear(&apic_error_lock);
1597 		return (DDI_INTR_UNCLAIMED);
1598 	}
1599 	/* NOTREACHED */
1600 }
1601 
1602 /*
1603  * Turn off the mask bit in the performance counter Local Vector Table entry.
1604  */
1605 static void
1606 apic_cpcovf_mask_clear(void)
1607 {
1608 	apicadr[APIC_PCINT_VECT] &= ~APIC_LVT_MASK;
1609 }
1610 
1611 static void
1612 apic_init_intr()
1613 {
1614 	processorid_t	cpun = psm_get_cpu_id();
1615 
1616 #if defined(__amd64)
1617 	setcr8((ulong_t)(APIC_MASK_ALL >> APIC_IPL_SHIFT));
1618 #else
1619 	apicadr[APIC_TASK_REG] = APIC_MASK_ALL;
1620 #endif
1621 
1622 	if (apic_flat_model)
1623 		apicadr[APIC_FORMAT_REG] = APIC_FLAT_MODEL;
1624 	else
1625 		apicadr[APIC_FORMAT_REG] = APIC_CLUSTER_MODEL;
1626 	apicadr[APIC_DEST_REG] = AV_HIGH_ORDER >> cpun;
1627 
1628 	/* need to enable APIC before unmasking NMI */
1629 	apicadr[APIC_SPUR_INT_REG] = AV_UNIT_ENABLE | APIC_SPUR_INTR;
1630 
1631 	apicadr[APIC_LOCAL_TIMER] = AV_MASK;
1632 	apicadr[APIC_INT_VECT0]	= AV_MASK;	/* local intr reg 0 */
1633 	apicadr[APIC_INT_VECT1] = AV_NMI;	/* enable NMI */
1634 
1635 	if (apic_cpus[cpun].aci_local_ver < APIC_INTEGRATED_VERS)
1636 		return;
1637 
1638 	/* Enable performance counter overflow interrupt */
1639 
1640 	if ((x86_feature & X86_MSR) != X86_MSR)
1641 		apic_enable_cpcovf_intr = 0;
1642 	if (apic_enable_cpcovf_intr) {
1643 		if (apic_cpcovf_vect == 0) {
1644 			int ipl = APIC_PCINT_IPL;
1645 			int irq = apic_get_ipivect(ipl, -1);
1646 
1647 			ASSERT(irq != -1);
1648 			apic_cpcovf_vect = apic_irq_table[irq]->airq_vector;
1649 			ASSERT(apic_cpcovf_vect);
1650 			(void) add_avintr(NULL, ipl,
1651 			    (avfunc)kcpc_hw_overflow_intr,
1652 			    "apic pcint", irq, NULL, NULL, NULL);
1653 			kcpc_hw_overflow_intr_installed = 1;
1654 			kcpc_hw_enable_cpc_intr = apic_cpcovf_mask_clear;
1655 		}
1656 		apicadr[APIC_PCINT_VECT] = apic_cpcovf_vect;
1657 	}
1658 
1659 	/* Enable error interrupt */
1660 
1661 	if (apic_enable_error_intr) {
1662 		if (apic_errvect == 0) {
1663 			int ipl = 0xf;	/* get highest priority intr */
1664 			int irq = apic_get_ipivect(ipl, -1);
1665 
1666 			ASSERT(irq != -1);
1667 			apic_errvect = apic_irq_table[irq]->airq_vector;
1668 			ASSERT(apic_errvect);
1669 			/*
1670 			 * Not PSMI compliant, but we are going to merge
1671 			 * with ON anyway
1672 			 */
1673 			(void) add_avintr((void *)NULL, ipl,
1674 			    (avfunc)apic_error_intr, "apic error intr",
1675 			    irq, NULL, NULL, NULL);
1676 		}
1677 		apicadr[APIC_ERR_VECT] = apic_errvect;
1678 		apicadr[APIC_ERROR_STATUS] = 0;
1679 		apicadr[APIC_ERROR_STATUS] = 0;
1680 	}
1681 }
1682 
1683 static void
1684 apic_disable_local_apic()
1685 {
1686 	apicadr[APIC_TASK_REG] = APIC_MASK_ALL;
1687 	apicadr[APIC_LOCAL_TIMER] = AV_MASK;
1688 	apicadr[APIC_INT_VECT0] = AV_MASK;	/* local intr reg 0 */
1689 	apicadr[APIC_INT_VECT1] = AV_MASK;	/* disable NMI */
1690 	apicadr[APIC_ERR_VECT] = AV_MASK;	/* and error interrupt */
1691 	apicadr[APIC_PCINT_VECT] = AV_MASK;	/* and perf counter intr */
1692 	apicadr[APIC_SPUR_INT_REG] = APIC_SPUR_INTR;
1693 }
1694 
1695 static void
1696 apic_picinit(void)
1697 {
1698 	int i, j;
1699 	uint_t isr;
1700 	volatile int32_t *ioapic;
1701 	apic_irq_t	*irqptr;
1702 	struct intrspec ispec;
1703 
1704 	/*
1705 	 * On UniSys Model 6520, the BIOS leaves vector 0x20 isr
1706 	 * bit on without clearing it with EOI.  Since softint
1707 	 * uses vector 0x20 to interrupt itself, so softint will
1708 	 * not work on this machine.  In order to fix this problem
1709 	 * a check is made to verify all the isr bits are clear.
1710 	 * If not, EOIs are issued to clear the bits.
1711 	 */
1712 	for (i = 7; i >= 1; i--) {
1713 		if ((isr = apicadr[APIC_ISR_REG + (i * 4)]) != 0)
1714 			for (j = 0; ((j < 32) && (isr != 0)); j++)
1715 				if (isr & (1 << j)) {
1716 					apicadr[APIC_EOI_REG] = 0;
1717 					isr &= ~(1 << j);
1718 					apic_error |= APIC_ERR_BOOT_EOI;
1719 				}
1720 	}
1721 
1722 	/* set a flag so we know we have run apic_picinit() */
1723 	apic_flag = 1;
1724 	LOCK_INIT_CLEAR(&apic_gethrtime_lock);
1725 	LOCK_INIT_CLEAR(&apic_ioapic_lock);
1726 	LOCK_INIT_CLEAR(&apic_revector_lock);
1727 	LOCK_INIT_CLEAR(&apic_ioapic_reprogram_lock);
1728 	LOCK_INIT_CLEAR(&apic_error_lock);
1729 
1730 	picsetup();	 /* initialise the 8259 */
1731 
1732 	/* add nmi handler - least priority nmi handler */
1733 	LOCK_INIT_CLEAR(&apic_nmi_lock);
1734 
1735 	if (!psm_add_nmintr(0, (avfunc) apic_nmi_intr,
1736 	    "pcplusmp NMI handler", (caddr_t)NULL))
1737 		cmn_err(CE_WARN, "pcplusmp: Unable to add nmi handler");
1738 
1739 	apic_init_intr();
1740 
1741 	/* enable apic mode if imcr present */
1742 	if (apic_imcrp) {
1743 		outb(APIC_IMCR_P1, (uchar_t)APIC_IMCR_SELECT);
1744 		outb(APIC_IMCR_P2, (uchar_t)APIC_IMCR_APIC);
1745 	}
1746 
1747 	/* mask interrupt vectors					*/
1748 	for (j = 0; j < apic_io_max; j++) {
1749 		int intin_max;
1750 		ioapic = apicioadr[j];
1751 		ioapic[APIC_IO_REG] = APIC_VERS_CMD;
1752 		/* Bits 23-16 define the maximum redirection entries */
1753 		intin_max = (ioapic[APIC_IO_DATA] >> 16) & 0xff;
1754 		for (i = 0; i < intin_max; i++) {
1755 			ioapic[APIC_IO_REG] = APIC_RDT_CMD + 2 * i;
1756 			ioapic[APIC_IO_DATA] = AV_MASK;
1757 		}
1758 	}
1759 
1760 	/*
1761 	 * Hack alert: deal with ACPI SCI interrupt chicken/egg here
1762 	 */
1763 	if (apic_sci_vect > 0) {
1764 		/*
1765 		 * acpica has already done add_avintr(); we just
1766 		 * to finish the job by mimicing translate_irq()
1767 		 *
1768 		 * Fake up an intrspec and setup the tables
1769 		 */
1770 		ispec.intrspec_vec = apic_sci_vect;
1771 		ispec.intrspec_pri = SCI_IPL;
1772 
1773 		if (apic_setup_irq_table(NULL, apic_sci_vect, NULL,
1774 		    &ispec, &apic_sci_flags, DDI_INTR_TYPE_FIXED) < 0) {
1775 			cmn_err(CE_WARN, "!apic: SCI setup failed");
1776 			return;
1777 		}
1778 		irqptr = apic_irq_table[apic_sci_vect];
1779 
1780 		/* Program I/O APIC */
1781 		(void) apic_setup_io_intr(irqptr, apic_sci_vect);
1782 	}
1783 }
1784 
1785 
1786 static void
1787 apic_cpu_start(processorid_t cpun, caddr_t rm_code)
1788 {
1789 	int		loop_count;
1790 	uint32_t	vector;
1791 	uint_t		cpu_id, iflag;
1792 
1793 	cpu_id = apic_cpus[cpun].aci_local_id;
1794 
1795 	apic_cmos_ssb_set = 1;
1796 
1797 	/*
1798 	 * Interrupts on BSP cpu will be disabled during these startup
1799 	 * steps in order to avoid unwanted side effects from
1800 	 * executing interrupt handlers on a problematic BIOS.
1801 	 */
1802 
1803 	iflag = intr_clear();
1804 	outb(CMOS_ADDR, SSB);
1805 	outb(CMOS_DATA, BIOS_SHUTDOWN);
1806 
1807 	while (get_apic_cmd1() & AV_PENDING)
1808 		apic_ret();
1809 
1810 	/* for integrated - make sure there is one INIT IPI in buffer */
1811 	/* for external - it will wake up the cpu */
1812 	apicadr[APIC_INT_CMD2] = cpu_id << APIC_ICR_ID_BIT_OFFSET;
1813 	apicadr[APIC_INT_CMD1] = AV_ASSERT | AV_RESET;
1814 
1815 	/* If only 1 CPU is installed, PENDING bit will not go low */
1816 	for (loop_count = 0x1000; loop_count; loop_count--)
1817 		if (get_apic_cmd1() & AV_PENDING)
1818 			apic_ret();
1819 		else
1820 			break;
1821 
1822 	apicadr[APIC_INT_CMD2] = cpu_id << APIC_ICR_ID_BIT_OFFSET;
1823 	apicadr[APIC_INT_CMD1] = AV_DEASSERT | AV_RESET;
1824 
1825 	drv_usecwait(20000);		/* 20 milli sec */
1826 
1827 	if (apic_cpus[cpun].aci_local_ver >= APIC_INTEGRATED_VERS) {
1828 		/* integrated apic */
1829 
1830 		rm_code = (caddr_t)(uintptr_t)rm_platter_pa;
1831 		vector = (rm_platter_pa >> MMU_PAGESHIFT) &
1832 		    (APIC_VECTOR_MASK | APIC_IPL_MASK);
1833 
1834 		/* to offset the INIT IPI queue up in the buffer */
1835 		apicadr[APIC_INT_CMD2] = cpu_id << APIC_ICR_ID_BIT_OFFSET;
1836 		apicadr[APIC_INT_CMD1] = vector | AV_STARTUP;
1837 
1838 		drv_usecwait(200);		/* 20 micro sec */
1839 
1840 		apicadr[APIC_INT_CMD2] = cpu_id << APIC_ICR_ID_BIT_OFFSET;
1841 		apicadr[APIC_INT_CMD1] = vector | AV_STARTUP;
1842 
1843 		drv_usecwait(200);		/* 20 micro sec */
1844 	}
1845 	intr_restore(iflag);
1846 }
1847 
1848 
1849 #ifdef	DEBUG
1850 int	apic_break_on_cpu = 9;
1851 int	apic_stretch_interrupts = 0;
1852 int	apic_stretch_ISR = 1 << 3;	/* IPL of 3 matches nothing now */
1853 
1854 void
1855 apic_break()
1856 {
1857 }
1858 #endif /* DEBUG */
1859 
1860 /*
1861  * platform_intr_enter
1862  *
1863  *	Called at the beginning of the interrupt service routine to
1864  *	mask all level equal to and below the interrupt priority
1865  *	of the interrupting vector.  An EOI should be given to
1866  *	the interrupt controller to enable other HW interrupts.
1867  *
1868  *	Return -1 for spurious interrupts
1869  *
1870  */
1871 /*ARGSUSED*/
1872 static int
1873 apic_intr_enter(int ipl, int *vectorp)
1874 {
1875 	uchar_t vector;
1876 	int nipl;
1877 	int irq, iflag;
1878 	apic_cpus_info_t *cpu_infop;
1879 
1880 	/*
1881 	 * The real vector programmed in APIC is *vectorp + 0x20
1882 	 * But, cmnint code subtracts 0x20 before pushing it.
1883 	 * Hence APIC_BASE_VECT is 0x20.
1884 	 */
1885 
1886 	vector = (uchar_t)*vectorp;
1887 
1888 	/* if interrupted by the clock, increment apic_nsec_since_boot */
1889 	if (vector == apic_clkvect) {
1890 		if (!apic_oneshot) {
1891 			/* NOTE: this is not MT aware */
1892 			apic_hrtime_stamp++;
1893 			apic_nsec_since_boot += apic_nsec_per_intr;
1894 			apic_hrtime_stamp++;
1895 			last_count_read = apic_hertz_count;
1896 			apic_redistribute_compute();
1897 		}
1898 
1899 		/* We will avoid all the book keeping overhead for clock */
1900 		nipl = apic_vectortoipl[vector >> APIC_IPL_SHIFT];
1901 #if defined(__amd64)
1902 		setcr8((ulong_t)apic_cr8pri[nipl]);
1903 #else
1904 		apicadr[APIC_TASK_REG] = apic_ipltopri[nipl];
1905 #endif
1906 		*vectorp = apic_vector_to_irq[vector + APIC_BASE_VECT];
1907 		apicadr[APIC_EOI_REG] = 0;
1908 		return (nipl);
1909 	}
1910 
1911 	cpu_infop = &apic_cpus[psm_get_cpu_id()];
1912 
1913 	if (vector == (APIC_SPUR_INTR - APIC_BASE_VECT)) {
1914 		cpu_infop->aci_spur_cnt++;
1915 		return (APIC_INT_SPURIOUS);
1916 	}
1917 
1918 	/* Check if the vector we got is really what we need */
1919 	if (apic_revector_pending) {
1920 		/*
1921 		 * Disable interrupts for the duration of
1922 		 * the vector translation to prevent a self-race for
1923 		 * the apic_revector_lock.  This cannot be done
1924 		 * in apic_xlate_vector because it is recursive and
1925 		 * we want the vector translation to be atomic with
1926 		 * respect to other (higher-priority) interrupts.
1927 		 */
1928 		iflag = intr_clear();
1929 		vector = apic_xlate_vector(vector + APIC_BASE_VECT) -
1930 		    APIC_BASE_VECT;
1931 		intr_restore(iflag);
1932 	}
1933 
1934 	nipl = apic_vectortoipl[vector >> APIC_IPL_SHIFT];
1935 	*vectorp = irq = apic_vector_to_irq[vector + APIC_BASE_VECT];
1936 
1937 #if defined(__amd64)
1938 	setcr8((ulong_t)apic_cr8pri[nipl]);
1939 #else
1940 	apicadr[APIC_TASK_REG] = apic_ipltopri[nipl];
1941 #endif
1942 
1943 	cpu_infop->aci_current[nipl] = (uchar_t)irq;
1944 	cpu_infop->aci_curipl = (uchar_t)nipl;
1945 	cpu_infop->aci_ISR_in_progress |= 1 << nipl;
1946 
1947 	/*
1948 	 * apic_level_intr could have been assimilated into the irq struct.
1949 	 * but, having it as a character array is more efficient in terms of
1950 	 * cache usage. So, we leave it as is.
1951 	 */
1952 	if (!apic_level_intr[irq])
1953 		apicadr[APIC_EOI_REG] = 0;
1954 
1955 #ifdef	DEBUG
1956 	APIC_DEBUG_BUF_PUT(vector);
1957 	APIC_DEBUG_BUF_PUT(irq);
1958 	APIC_DEBUG_BUF_PUT(nipl);
1959 	APIC_DEBUG_BUF_PUT(psm_get_cpu_id());
1960 	if ((apic_stretch_interrupts) && (apic_stretch_ISR & (1 << nipl)))
1961 		drv_usecwait(apic_stretch_interrupts);
1962 
1963 	if (apic_break_on_cpu == psm_get_cpu_id())
1964 		apic_break();
1965 #endif /* DEBUG */
1966 	return (nipl);
1967 }
1968 
1969 static void
1970 apic_intr_exit(int prev_ipl, int irq)
1971 {
1972 	apic_cpus_info_t *cpu_infop;
1973 
1974 #if defined(__amd64)
1975 	setcr8((ulong_t)apic_cr8pri[prev_ipl]);
1976 #else
1977 	apicadr[APIC_TASK_REG] = apic_ipltopri[prev_ipl];
1978 #endif
1979 
1980 	cpu_infop = &apic_cpus[psm_get_cpu_id()];
1981 	if (apic_level_intr[irq])
1982 		apicadr[APIC_EOI_REG] = 0;
1983 
1984 	cpu_infop->aci_curipl = (uchar_t)prev_ipl;
1985 	/* ISR above current pri could not be in progress */
1986 	cpu_infop->aci_ISR_in_progress &= (2 << prev_ipl) - 1;
1987 }
1988 
1989 /*
1990  * Mask all interrupts below or equal to the given IPL
1991  */
1992 static void
1993 apic_setspl(int ipl)
1994 {
1995 
1996 #if defined(__amd64)
1997 	setcr8((ulong_t)apic_cr8pri[ipl]);
1998 #else
1999 	apicadr[APIC_TASK_REG] = apic_ipltopri[ipl];
2000 #endif
2001 
2002 	/* interrupts at ipl above this cannot be in progress */
2003 	apic_cpus[psm_get_cpu_id()].aci_ISR_in_progress &= (2 << ipl) - 1;
2004 	/*
2005 	 * this is a patch fix for the ALR QSMP P5 machine, so that interrupts
2006 	 * have enough time to come in before the priority is raised again
2007 	 * during the idle() loop.
2008 	 */
2009 	if (apic_setspl_delay)
2010 		(void) get_apic_pri();
2011 }
2012 
2013 /*
2014  * trigger a software interrupt at the given IPL
2015  */
2016 static void
2017 apic_set_softintr(int ipl)
2018 {
2019 	int vector;
2020 	uint_t flag;
2021 
2022 	vector = apic_resv_vector[ipl];
2023 
2024 	flag = intr_clear();
2025 
2026 	while (get_apic_cmd1() & AV_PENDING)
2027 		apic_ret();
2028 
2029 	/* generate interrupt at vector on itself only */
2030 	apicadr[APIC_INT_CMD1] = AV_SH_SELF | vector;
2031 
2032 	intr_restore(flag);
2033 }
2034 
2035 /*
2036  * generates an interprocessor interrupt to another CPU
2037  */
2038 static void
2039 apic_send_ipi(int cpun, int ipl)
2040 {
2041 	int vector;
2042 	uint_t flag;
2043 
2044 	vector = apic_resv_vector[ipl];
2045 
2046 	flag = intr_clear();
2047 
2048 	while (get_apic_cmd1() & AV_PENDING)
2049 		apic_ret();
2050 
2051 	apicadr[APIC_INT_CMD2] =
2052 	    apic_cpus[cpun].aci_local_id << APIC_ICR_ID_BIT_OFFSET;
2053 	apicadr[APIC_INT_CMD1] = vector;
2054 
2055 	intr_restore(flag);
2056 }
2057 
2058 
2059 /*ARGSUSED*/
2060 static void
2061 apic_set_idlecpu(processorid_t cpun)
2062 {
2063 }
2064 
2065 /*ARGSUSED*/
2066 static void
2067 apic_unset_idlecpu(processorid_t cpun)
2068 {
2069 }
2070 
2071 
2072 static void
2073 apic_ret()
2074 {
2075 }
2076 
2077 static int
2078 get_apic_cmd1()
2079 {
2080 	return (apicadr[APIC_INT_CMD1]);
2081 }
2082 
2083 static int
2084 get_apic_pri()
2085 {
2086 #if defined(__amd64)
2087 	return ((int)getcr8());
2088 #else
2089 	return (apicadr[APIC_TASK_REG]);
2090 #endif
2091 }
2092 
2093 /*
2094  * If apic_coarse_time == 1, then apic_gettime() is used instead of
2095  * apic_gethrtime().  This is used for performance instead of accuracy.
2096  */
2097 
2098 static hrtime_t
2099 apic_gettime()
2100 {
2101 	int old_hrtime_stamp;
2102 	hrtime_t temp;
2103 
2104 	/*
2105 	 * In one-shot mode, we do not keep time, so if anyone
2106 	 * calls psm_gettime() directly, we vector over to
2107 	 * gethrtime().
2108 	 * one-shot mode MUST NOT be enabled if this psm is the source of
2109 	 * hrtime.
2110 	 */
2111 
2112 	if (apic_oneshot)
2113 		return (gethrtime());
2114 
2115 
2116 gettime_again:
2117 	while ((old_hrtime_stamp = apic_hrtime_stamp) & 1)
2118 		apic_ret();
2119 
2120 	temp = apic_nsec_since_boot;
2121 
2122 	if (apic_hrtime_stamp != old_hrtime_stamp) {	/* got an interrupt */
2123 		goto gettime_again;
2124 	}
2125 	return (temp);
2126 }
2127 
2128 /*
2129  * Here we return the number of nanoseconds since booting.  Note every
2130  * clock interrupt increments apic_nsec_since_boot by the appropriate
2131  * amount.
2132  */
2133 static hrtime_t
2134 apic_gethrtime()
2135 {
2136 	int curr_timeval, countval, elapsed_ticks, oflags;
2137 	int old_hrtime_stamp, status;
2138 	hrtime_t temp;
2139 	uchar_t	cpun;
2140 
2141 
2142 	/*
2143 	 * In one-shot mode, we do not keep time, so if anyone
2144 	 * calls psm_gethrtime() directly, we vector over to
2145 	 * gethrtime().
2146 	 * one-shot mode MUST NOT be enabled if this psm is the source of
2147 	 * hrtime.
2148 	 */
2149 
2150 	if (apic_oneshot)
2151 		return (gethrtime());
2152 
2153 	oflags = intr_clear();	/* prevent migration */
2154 
2155 	cpun = (uchar_t)((uint_t)apicadr[APIC_LID_REG] >> APIC_ID_BIT_OFFSET);
2156 
2157 	lock_set(&apic_gethrtime_lock);
2158 
2159 gethrtime_again:
2160 	while ((old_hrtime_stamp = apic_hrtime_stamp) & 1)
2161 		apic_ret();
2162 
2163 	/*
2164 	 * Check to see which CPU we are on.  Note the time is kept on
2165 	 * the local APIC of CPU 0.  If on CPU 0, simply read the current
2166 	 * counter.  If on another CPU, issue a remote read command to CPU 0.
2167 	 */
2168 	if (cpun == apic_cpus[0].aci_local_id) {
2169 		countval = apicadr[APIC_CURR_COUNT];
2170 	} else {
2171 		while (get_apic_cmd1() & AV_PENDING)
2172 			apic_ret();
2173 
2174 		apicadr[APIC_INT_CMD2] =
2175 		    apic_cpus[0].aci_local_id << APIC_ICR_ID_BIT_OFFSET;
2176 		apicadr[APIC_INT_CMD1] = APIC_CURR_ADD|AV_REMOTE;
2177 
2178 		while ((status = get_apic_cmd1()) & AV_READ_PENDING)
2179 			apic_ret();
2180 
2181 		if (status & AV_REMOTE_STATUS)	/* 1 = valid */
2182 			countval = apicadr[APIC_REMOTE_READ];
2183 		else {	/* 0 = invalid */
2184 			apic_remote_hrterr++;
2185 			/*
2186 			 * return last hrtime right now, will need more
2187 			 * testing if change to retry
2188 			 */
2189 			temp = apic_last_hrtime;
2190 
2191 			lock_clear(&apic_gethrtime_lock);
2192 
2193 			intr_restore(oflags);
2194 
2195 			return (temp);
2196 		}
2197 	}
2198 	if (countval > last_count_read)
2199 		countval = 0;
2200 	else
2201 		last_count_read = countval;
2202 
2203 	elapsed_ticks = apic_hertz_count - countval;
2204 
2205 	curr_timeval = elapsed_ticks * apic_nsec_per_tick;
2206 	temp = apic_nsec_since_boot + curr_timeval;
2207 
2208 	if (apic_hrtime_stamp != old_hrtime_stamp) {	/* got an interrupt */
2209 		/* we might have clobbered last_count_read. Restore it */
2210 		last_count_read = apic_hertz_count;
2211 		goto gethrtime_again;
2212 	}
2213 
2214 	if (temp < apic_last_hrtime) {
2215 		/* return last hrtime if error occurs */
2216 		apic_hrtime_error++;
2217 		temp = apic_last_hrtime;
2218 	}
2219 	else
2220 		apic_last_hrtime = temp;
2221 
2222 	lock_clear(&apic_gethrtime_lock);
2223 	intr_restore(oflags);
2224 
2225 	return (temp);
2226 }
2227 
2228 /* apic NMI handler */
2229 /*ARGSUSED*/
2230 static void
2231 apic_nmi_intr(caddr_t arg)
2232 {
2233 	if (apic_shutdown_processors) {
2234 		apic_disable_local_apic();
2235 		return;
2236 	}
2237 
2238 	if (lock_try(&apic_nmi_lock)) {
2239 		if (apic_kmdb_on_nmi) {
2240 			if (psm_debugger() == 0) {
2241 				cmn_err(CE_PANIC,
2242 				    "NMI detected, kmdb is not available.");
2243 			} else {
2244 				debug_enter("\nNMI detected, entering kmdb.\n");
2245 			}
2246 		} else {
2247 			if (apic_panic_on_nmi) {
2248 				/* Keep panic from entering kmdb. */
2249 				nopanicdebug = 1;
2250 				cmn_err(CE_PANIC, "pcplusmp: NMI received");
2251 			} else {
2252 				/*
2253 				 * prom_printf is the best shot we have
2254 				 * of something which is problem free from
2255 				 * high level/NMI type of interrupts
2256 				 */
2257 				prom_printf("pcplusmp: NMI received\n");
2258 				apic_error |= APIC_ERR_NMI;
2259 				apic_num_nmis++;
2260 			}
2261 		}
2262 		lock_clear(&apic_nmi_lock);
2263 	}
2264 }
2265 
2266 /*
2267  * Add mask bits to disable interrupt vector from happening
2268  * at or above IPL. In addition, it should remove mask bits
2269  * to enable interrupt vectors below the given IPL.
2270  *
2271  * Both add and delspl are complicated by the fact that different interrupts
2272  * may share IRQs. This can happen in two ways.
2273  * 1. The same H/W line is shared by more than 1 device
2274  * 1a. with interrupts at different IPLs
2275  * 1b. with interrupts at same IPL
2276  * 2. We ran out of vectors at a given IPL and started sharing vectors.
2277  * 1b and 2 should be handled gracefully, except for the fact some ISRs
2278  * will get called often when no interrupt is pending for the device.
2279  * For 1a, we just hope that the machine blows up with the person who
2280  * set it up that way!. In the meantime, we handle it at the higher IPL.
2281  */
2282 /*ARGSUSED*/
2283 static int
2284 apic_addspl(int irqno, int ipl, int min_ipl, int max_ipl)
2285 {
2286 	uchar_t vector;
2287 	int iflag;
2288 	apic_irq_t *irqptr, *irqheadptr;
2289 	int irqindex;
2290 
2291 	ASSERT(max_ipl <= UCHAR_MAX);
2292 	irqindex = IRQINDEX(irqno);
2293 
2294 	if ((irqindex == -1) || (!apic_irq_table[irqindex]))
2295 		return (PSM_FAILURE);
2296 
2297 	irqptr = irqheadptr = apic_irq_table[irqindex];
2298 
2299 	DDI_INTR_IMPLDBG((CE_CONT, "apic_addspl: dip=0x%p type=%d irqno=0x%x "
2300 	    "vector=0x%x\n", (void *)irqptr->airq_dip,
2301 	    irqptr->airq_mps_intr_index, irqno, irqptr->airq_vector));
2302 
2303 	while (irqptr) {
2304 		if (VIRTIRQ(irqindex, irqptr->airq_share_id) == irqno)
2305 			break;
2306 		irqptr = irqptr->airq_next;
2307 	}
2308 	irqptr->airq_share++;
2309 
2310 	/* return if it is not hardware interrupt */
2311 	if (irqptr->airq_mps_intr_index == RESERVE_INDEX)
2312 		return (PSM_SUCCESS);
2313 
2314 	/* Or if there are more interupts at a higher IPL */
2315 	if (ipl != max_ipl)
2316 		return (PSM_SUCCESS);
2317 
2318 	/*
2319 	 * if apic_picinit() has not been called yet, just return.
2320 	 * At the end of apic_picinit(), we will call setup_io_intr().
2321 	 */
2322 
2323 	if (!apic_flag)
2324 		return (PSM_SUCCESS);
2325 
2326 	iflag = intr_clear();
2327 
2328 	/*
2329 	 * Upgrade vector if max_ipl is not earlier ipl. If we cannot allocate,
2330 	 * return failure. Not very elegant, but then we hope the
2331 	 * machine will blow up with ...
2332 	 */
2333 	if (irqptr->airq_ipl != max_ipl) {
2334 		vector = apic_allocate_vector(max_ipl, irqindex, 1);
2335 		if (vector == 0) {
2336 			intr_restore(iflag);
2337 			irqptr->airq_share--;
2338 			return (PSM_FAILURE);
2339 		}
2340 		irqptr = irqheadptr;
2341 		apic_mark_vector(irqptr->airq_vector, vector);
2342 		while (irqptr) {
2343 			irqptr->airq_vector = vector;
2344 			irqptr->airq_ipl = (uchar_t)max_ipl;
2345 			/*
2346 			 * reprogram irq being added and every one else
2347 			 * who is not in the UNINIT state
2348 			 */
2349 			if ((VIRTIRQ(irqindex, irqptr->airq_share_id) ==
2350 			    irqno) || (irqptr->airq_temp_cpu != IRQ_UNINIT)) {
2351 				apic_record_rdt_entry(irqptr, irqindex);
2352 				(void) apic_setup_io_intr(irqptr, irqindex);
2353 			}
2354 			irqptr = irqptr->airq_next;
2355 		}
2356 		intr_restore(iflag);
2357 		return (PSM_SUCCESS);
2358 	}
2359 
2360 	ASSERT(irqptr);
2361 	(void) apic_setup_io_intr(irqptr, irqindex);
2362 	intr_restore(iflag);
2363 	return (PSM_SUCCESS);
2364 }
2365 
2366 /*
2367  * Recompute mask bits for the given interrupt vector.
2368  * If there is no interrupt servicing routine for this
2369  * vector, this function should disable interrupt vector
2370  * from happening at all IPLs. If there are still
2371  * handlers using the given vector, this function should
2372  * disable the given vector from happening below the lowest
2373  * IPL of the remaining hadlers.
2374  */
2375 /*ARGSUSED*/
2376 static int
2377 apic_delspl(int irqno, int ipl, int min_ipl, int max_ipl)
2378 {
2379 	uchar_t vector, bind_cpu;
2380 	int	iflag, intin, irqindex;
2381 	volatile int32_t *ioapic;
2382 	apic_irq_t	*irqptr, *irqheadptr;
2383 
2384 	irqindex = IRQINDEX(irqno);
2385 	irqptr = irqheadptr = apic_irq_table[irqindex];
2386 
2387 	DDI_INTR_IMPLDBG((CE_CONT, "apic_delspl: dip=0x%p type=%d irqno=0x%x "
2388 	    "vector=0x%x\n", (void *)irqptr->airq_dip,
2389 	    irqptr->airq_mps_intr_index, irqno, irqptr->airq_vector));
2390 
2391 	while (irqptr) {
2392 		if (VIRTIRQ(irqindex, irqptr->airq_share_id) == irqno)
2393 			break;
2394 		irqptr = irqptr->airq_next;
2395 	}
2396 	ASSERT(irqptr);
2397 
2398 	irqptr->airq_share--;
2399 
2400 	if (ipl < max_ipl)
2401 		return (PSM_SUCCESS);
2402 
2403 	/* return if it is not hardware interrupt */
2404 	if (irqptr->airq_mps_intr_index == RESERVE_INDEX)
2405 		return (PSM_SUCCESS);
2406 
2407 	if (!apic_flag) {
2408 		/*
2409 		 * Clear irq_struct. If two devices shared an intpt
2410 		 * line & 1 unloaded before picinit, we are hosed. But, then
2411 		 * we hope the machine will ...
2412 		 */
2413 		irqptr->airq_mps_intr_index = FREE_INDEX;
2414 		irqptr->airq_temp_cpu = IRQ_UNINIT;
2415 		apic_free_vector(irqptr->airq_vector);
2416 		return (PSM_SUCCESS);
2417 	}
2418 	/*
2419 	 * Downgrade vector to new max_ipl if needed.If we cannot allocate,
2420 	 * use old IPL. Not very elegant, but then we hope ...
2421 	 */
2422 	if ((irqptr->airq_ipl != max_ipl) && (max_ipl != PSM_INVALID_IPL)) {
2423 		apic_irq_t	*irqp;
2424 		if (vector = apic_allocate_vector(max_ipl, irqno, 1)) {
2425 			apic_mark_vector(irqheadptr->airq_vector, vector);
2426 			irqp = irqheadptr;
2427 			while (irqp) {
2428 				irqp->airq_vector = vector;
2429 				irqp->airq_ipl = (uchar_t)max_ipl;
2430 				if (irqp->airq_temp_cpu != IRQ_UNINIT) {
2431 					apic_record_rdt_entry(irqp, irqindex);
2432 					(void) apic_setup_io_intr(irqp,
2433 					    irqindex);
2434 				}
2435 				irqp = irqp->airq_next;
2436 			}
2437 		}
2438 	}
2439 
2440 	if (irqptr->airq_share)
2441 		return (PSM_SUCCESS);
2442 
2443 	ioapic = apicioadr[irqptr->airq_ioapicindex];
2444 	intin = irqptr->airq_intin_no;
2445 	iflag = intr_clear();
2446 	lock_set(&apic_ioapic_lock);
2447 	ioapic[APIC_IO_REG] = APIC_RDT_CMD + 2 * intin;
2448 	ioapic[APIC_IO_DATA] = AV_MASK;
2449 
2450 	/* Disable the MSI/X vector */
2451 	if (APIC_IS_MSI_OR_MSIX_INDEX(irqptr->airq_mps_intr_index)) {
2452 		int type = (irqptr->airq_mps_intr_index == MSI_INDEX) ?
2453 		    DDI_INTR_TYPE_MSI : DDI_INTR_TYPE_MSIX;
2454 
2455 		/*
2456 		 * Make sure we only disable on the last
2457 		 * of the multi-MSI support
2458 		 */
2459 		if (i_ddi_intr_get_current_nintrs(irqptr->airq_dip) == 1) {
2460 			(void) pci_msi_unconfigure(irqptr->airq_dip, type,
2461 			    irqptr->airq_ioapicindex);
2462 
2463 			(void) pci_msi_disable_mode(irqptr->airq_dip, type,
2464 			    irqptr->airq_ioapicindex);
2465 		}
2466 	}
2467 
2468 	if (max_ipl == PSM_INVALID_IPL) {
2469 		ASSERT(irqheadptr == irqptr);
2470 		bind_cpu = irqptr->airq_temp_cpu;
2471 		if (((uchar_t)bind_cpu != IRQ_UNBOUND) &&
2472 		    ((uchar_t)bind_cpu != IRQ_UNINIT)) {
2473 			ASSERT((bind_cpu & ~IRQ_USER_BOUND) < apic_nproc);
2474 			if (bind_cpu & IRQ_USER_BOUND) {
2475 				/* If hardbound, temp_cpu == cpu */
2476 				bind_cpu &= ~IRQ_USER_BOUND;
2477 				apic_cpus[bind_cpu].aci_bound--;
2478 			} else
2479 				apic_cpus[bind_cpu].aci_temp_bound--;
2480 		}
2481 		lock_clear(&apic_ioapic_lock);
2482 		intr_restore(iflag);
2483 		irqptr->airq_temp_cpu = IRQ_UNINIT;
2484 		irqptr->airq_mps_intr_index = FREE_INDEX;
2485 		apic_free_vector(irqptr->airq_vector);
2486 		return (PSM_SUCCESS);
2487 	}
2488 	lock_clear(&apic_ioapic_lock);
2489 	intr_restore(iflag);
2490 
2491 	mutex_enter(&airq_mutex);
2492 	if ((irqptr == apic_irq_table[irqindex])) {
2493 		apic_irq_t	*oldirqptr;
2494 		/* Move valid irq entry to the head */
2495 		irqheadptr = oldirqptr = irqptr;
2496 		irqptr = irqptr->airq_next;
2497 		ASSERT(irqptr);
2498 		while (irqptr) {
2499 			if (irqptr->airq_mps_intr_index != FREE_INDEX)
2500 				break;
2501 			oldirqptr = irqptr;
2502 			irqptr = irqptr->airq_next;
2503 		}
2504 		/* remove all invalid ones from the beginning */
2505 		apic_irq_table[irqindex] = irqptr;
2506 		/*
2507 		 * and link them back after the head. The invalid ones
2508 		 * begin with irqheadptr and end at oldirqptr
2509 		 */
2510 		oldirqptr->airq_next = irqptr->airq_next;
2511 		irqptr->airq_next = irqheadptr;
2512 	}
2513 	mutex_exit(&airq_mutex);
2514 
2515 	irqptr->airq_temp_cpu = IRQ_UNINIT;
2516 	irqptr->airq_mps_intr_index = FREE_INDEX;
2517 	return (PSM_SUCCESS);
2518 }
2519 
2520 /*
2521  * Return HW interrupt number corresponding to the given IPL
2522  */
2523 /*ARGSUSED*/
2524 static int
2525 apic_softlvl_to_irq(int ipl)
2526 {
2527 	/*
2528 	 * Do not use apic to trigger soft interrupt.
2529 	 * It will cause the system to hang when 2 hardware interrupts
2530 	 * at the same priority with the softint are already accepted
2531 	 * by the apic.  Cause the AV_PENDING bit will not be cleared
2532 	 * until one of the hardware interrupt is eoi'ed.  If we need
2533 	 * to send an ipi at this time, we will end up looping forever
2534 	 * to wait for the AV_PENDING bit to clear.
2535 	 */
2536 	return (PSM_SV_SOFTWARE);
2537 }
2538 
2539 static int
2540 apic_post_cpu_start()
2541 {
2542 	int i, cpun;
2543 	apic_irq_t *irq_ptr;
2544 
2545 	apic_init_intr();
2546 
2547 	/*
2548 	 * since some systems don't enable the internal cache on the non-boot
2549 	 * cpus, so we have to enable them here
2550 	 */
2551 	setcr0(getcr0() & ~(0x60000000));
2552 
2553 	while (get_apic_cmd1() & AV_PENDING)
2554 		apic_ret();
2555 
2556 	cpun = psm_get_cpu_id();
2557 	apic_cpus[cpun].aci_status = APIC_CPU_ONLINE | APIC_CPU_INTR_ENABLE;
2558 
2559 	for (i = apic_min_device_irq; i <= apic_max_device_irq; i++) {
2560 		irq_ptr = apic_irq_table[i];
2561 		if ((irq_ptr == NULL) ||
2562 		    ((irq_ptr->airq_cpu & ~IRQ_USER_BOUND) != cpun))
2563 			continue;
2564 
2565 		while (irq_ptr) {
2566 			if (irq_ptr->airq_temp_cpu != IRQ_UNINIT)
2567 				(void) apic_rebind(irq_ptr, cpun, 1, IMMEDIATE);
2568 			irq_ptr = irq_ptr->airq_next;
2569 		}
2570 	}
2571 
2572 	return (PSM_SUCCESS);
2573 }
2574 
2575 processorid_t
2576 apic_get_next_processorid(processorid_t cpu_id)
2577 {
2578 
2579 	int i;
2580 
2581 	if (cpu_id == -1)
2582 		return ((processorid_t)0);
2583 
2584 	for (i = cpu_id + 1; i < NCPU; i++) {
2585 		if (apic_cpumask & (1 << i))
2586 			return (i);
2587 	}
2588 
2589 	return ((processorid_t)-1);
2590 }
2591 
2592 
2593 /*
2594  * type == -1 indicates it is an internal request. Do not change
2595  * resv_vector for these requests
2596  */
2597 static int
2598 apic_get_ipivect(int ipl, int type)
2599 {
2600 	uchar_t vector;
2601 	int irq;
2602 
2603 	if (irq = apic_allocate_irq(APIC_VECTOR(ipl))) {
2604 		if (vector = apic_allocate_vector(ipl, irq, 1)) {
2605 			apic_irq_table[irq]->airq_mps_intr_index =
2606 			    RESERVE_INDEX;
2607 			apic_irq_table[irq]->airq_vector = vector;
2608 			if (type != -1) {
2609 				apic_resv_vector[ipl] = vector;
2610 			}
2611 			return (irq);
2612 		}
2613 	}
2614 	apic_error |= APIC_ERR_GET_IPIVECT_FAIL;
2615 	return (-1);	/* shouldn't happen */
2616 }
2617 
2618 static int
2619 apic_getclkirq(int ipl)
2620 {
2621 	int	irq;
2622 
2623 	if ((irq = apic_get_ipivect(ipl, -1)) == -1)
2624 		return (-1);
2625 	/*
2626 	 * Note the vector in apic_clkvect for per clock handling.
2627 	 */
2628 	apic_clkvect = apic_irq_table[irq]->airq_vector - APIC_BASE_VECT;
2629 	APIC_VERBOSE_IOAPIC((CE_NOTE, "get_clkirq: vector = %x\n",
2630 	    apic_clkvect));
2631 	return (irq);
2632 }
2633 
2634 /*
2635  * Return the number of APIC clock ticks elapsed for 8245 to decrement
2636  * (APIC_TIME_COUNT + pit_ticks_adj) ticks.
2637  */
2638 static uint_t
2639 apic_calibrate(volatile uint32_t *addr, uint16_t *pit_ticks_adj)
2640 {
2641 	uint8_t		pit_tick_lo;
2642 	uint16_t	pit_tick, target_pit_tick;
2643 	uint32_t	start_apic_tick, end_apic_tick;
2644 	int		iflag;
2645 
2646 	addr += APIC_CURR_COUNT;
2647 
2648 	iflag = intr_clear();
2649 
2650 	do {
2651 		pit_tick_lo = inb(PITCTR0_PORT);
2652 		pit_tick = (inb(PITCTR0_PORT) << 8) | pit_tick_lo;
2653 	} while (pit_tick < APIC_TIME_MIN ||
2654 	    pit_tick_lo <= APIC_LB_MIN || pit_tick_lo >= APIC_LB_MAX);
2655 
2656 	/*
2657 	 * Wait for the 8254 to decrement by 5 ticks to ensure
2658 	 * we didn't start in the middle of a tick.
2659 	 * Compare with 0x10 for the wrap around case.
2660 	 */
2661 	target_pit_tick = pit_tick - 5;
2662 	do {
2663 		pit_tick_lo = inb(PITCTR0_PORT);
2664 		pit_tick = (inb(PITCTR0_PORT) << 8) | pit_tick_lo;
2665 	} while (pit_tick > target_pit_tick || pit_tick_lo < 0x10);
2666 
2667 	start_apic_tick = *addr;
2668 
2669 	/*
2670 	 * Wait for the 8254 to decrement by
2671 	 * (APIC_TIME_COUNT + pit_ticks_adj) ticks
2672 	 */
2673 	target_pit_tick = pit_tick - APIC_TIME_COUNT;
2674 	do {
2675 		pit_tick_lo = inb(PITCTR0_PORT);
2676 		pit_tick = (inb(PITCTR0_PORT) << 8) | pit_tick_lo;
2677 	} while (pit_tick > target_pit_tick || pit_tick_lo < 0x10);
2678 
2679 	end_apic_tick = *addr;
2680 
2681 	*pit_ticks_adj = target_pit_tick - pit_tick;
2682 
2683 	intr_restore(iflag);
2684 
2685 	return (start_apic_tick - end_apic_tick);
2686 }
2687 
2688 /*
2689  * Initialise the APIC timer on the local APIC of CPU 0 to the desired
2690  * frequency.  Note at this stage in the boot sequence, the boot processor
2691  * is the only active processor.
2692  * hertz value of 0 indicates a one-shot mode request.  In this case
2693  * the function returns the resolution (in nanoseconds) for the hardware
2694  * timer interrupt.  If one-shot mode capability is not available,
2695  * the return value will be 0. apic_enable_oneshot is a global switch
2696  * for disabling the functionality.
2697  * A non-zero positive value for hertz indicates a periodic mode request.
2698  * In this case the hardware will be programmed to generate clock interrupts
2699  * at hertz frequency and returns the resolution of interrupts in
2700  * nanosecond.
2701  */
2702 
2703 static int
2704 apic_clkinit(int hertz)
2705 {
2706 
2707 	uint_t		apic_ticks = 0;
2708 	uint_t		pit_time;
2709 	int		ret;
2710 	uint16_t	pit_ticks_adj;
2711 	static int	firsttime = 1;
2712 
2713 	if (firsttime) {
2714 		/* first time calibrate */
2715 
2716 		apicadr[APIC_DIVIDE_REG] = 0x0;
2717 		apicadr[APIC_INIT_COUNT] = APIC_MAXVAL;
2718 
2719 		/* set periodic interrupt based on CLKIN */
2720 		apicadr[APIC_LOCAL_TIMER] =
2721 		    (apic_clkvect + APIC_BASE_VECT) | AV_TIME;
2722 		tenmicrosec();
2723 
2724 		apic_ticks = apic_calibrate(apicadr, &pit_ticks_adj);
2725 
2726 		apicadr[APIC_LOCAL_TIMER] =
2727 		    (apic_clkvect + APIC_BASE_VECT) | AV_MASK;
2728 		/*
2729 		 * pit time is the amount of real time (in nanoseconds ) it took
2730 		 * the 8254 to decrement (APIC_TIME_COUNT + pit_ticks_adj) ticks
2731 		 */
2732 		pit_time = ((longlong_t)(APIC_TIME_COUNT +
2733 		    pit_ticks_adj) * NANOSEC) / PIT_HZ;
2734 
2735 		/*
2736 		 * Determine the number of nanoseconds per APIC clock tick
2737 		 * and then determine how many APIC ticks to interrupt at the
2738 		 * desired frequency
2739 		 */
2740 		apic_nsec_per_tick = pit_time / apic_ticks;
2741 		if (apic_nsec_per_tick == 0)
2742 			apic_nsec_per_tick = 1;
2743 
2744 		/* the interval timer initial count is 32 bit max */
2745 		apic_nsec_max = (hrtime_t)apic_nsec_per_tick * APIC_MAXVAL;
2746 		firsttime = 0;
2747 	}
2748 
2749 	if (hertz != 0) {
2750 		/* periodic */
2751 		apic_nsec_per_intr = NANOSEC / hertz;
2752 		apic_hertz_count = (longlong_t)apic_nsec_per_intr /
2753 		    apic_nsec_per_tick;
2754 		apic_sample_factor_redistribution = hertz + 1;
2755 	}
2756 
2757 	apic_int_busy_mark = (apic_int_busy_mark *
2758 	    apic_sample_factor_redistribution) / 100;
2759 	apic_int_free_mark = (apic_int_free_mark *
2760 	    apic_sample_factor_redistribution) / 100;
2761 	apic_diff_for_redistribution = (apic_diff_for_redistribution *
2762 	    apic_sample_factor_redistribution) / 100;
2763 
2764 	if (hertz == 0) {
2765 		/* requested one_shot */
2766 		if (!apic_oneshot_enable)
2767 			return (0);
2768 		apic_oneshot = 1;
2769 		ret = (int)apic_nsec_per_tick;
2770 	} else {
2771 		/* program the local APIC to interrupt at the given frequency */
2772 		apicadr[APIC_INIT_COUNT] = apic_hertz_count;
2773 		apicadr[APIC_LOCAL_TIMER] =
2774 		    (apic_clkvect + APIC_BASE_VECT) | AV_TIME;
2775 		apic_oneshot = 0;
2776 		ret = NANOSEC / hertz;
2777 	}
2778 
2779 	return (ret);
2780 
2781 }
2782 
2783 /*
2784  * apic_preshutdown:
2785  * Called early in shutdown whilst we can still access filesystems to do
2786  * things like loading modules which will be required to complete shutdown
2787  * after filesystems are all unmounted.
2788  */
2789 static void
2790 apic_preshutdown(int cmd, int fcn)
2791 {
2792 	APIC_VERBOSE_POWEROFF(("apic_preshutdown(%d,%d); m=%d a=%d\n",
2793 	    cmd, fcn, apic_poweroff_method, apic_enable_acpi));
2794 
2795 	if ((cmd != A_SHUTDOWN) || (fcn != AD_POWEROFF)) {
2796 		return;
2797 	}
2798 }
2799 
2800 static void
2801 apic_shutdown(int cmd, int fcn)
2802 {
2803 	int iflag, restarts, attempts;
2804 	int i, j;
2805 	volatile int32_t *ioapic;
2806 	uchar_t	byte;
2807 
2808 	/* Send NMI to all CPUs except self to do per processor shutdown */
2809 	iflag = intr_clear();
2810 	while (get_apic_cmd1() & AV_PENDING)
2811 		apic_ret();
2812 	apic_shutdown_processors = 1;
2813 	apicadr[APIC_INT_CMD1] = AV_NMI | AV_LEVEL | AV_SH_ALL_EXCSELF;
2814 
2815 	/* restore cmos shutdown byte before reboot */
2816 	if (apic_cmos_ssb_set) {
2817 		outb(CMOS_ADDR, SSB);
2818 		outb(CMOS_DATA, 0);
2819 	}
2820 	/* Disable the I/O APIC redirection entries */
2821 	for (j = 0; j < apic_io_max; j++) {
2822 		int intin_max;
2823 		ioapic = apicioadr[j];
2824 		ioapic[APIC_IO_REG] = APIC_VERS_CMD;
2825 		/* Bits 23-16 define the maximum redirection entries */
2826 		intin_max = (ioapic[APIC_IO_DATA] >> 16) & 0xff;
2827 		for (i = 0; i < intin_max; i++) {
2828 			ioapic[APIC_IO_REG] = APIC_RDT_CMD + 2 * i;
2829 			ioapic[APIC_IO_DATA] = AV_MASK;
2830 		}
2831 	}
2832 
2833 	/*	disable apic mode if imcr present	*/
2834 	if (apic_imcrp) {
2835 		outb(APIC_IMCR_P1, (uchar_t)APIC_IMCR_SELECT);
2836 		outb(APIC_IMCR_P2, (uchar_t)APIC_IMCR_PIC);
2837 	}
2838 
2839 	apic_disable_local_apic();
2840 
2841 	intr_restore(iflag);
2842 
2843 	if ((cmd != A_SHUTDOWN) || (fcn != AD_POWEROFF)) {
2844 		return;
2845 	}
2846 
2847 	switch (apic_poweroff_method) {
2848 		case APIC_POWEROFF_VIA_RTC:
2849 
2850 			/* select the extended NVRAM bank in the RTC */
2851 			outb(CMOS_ADDR, RTC_REGA);
2852 			byte = inb(CMOS_DATA);
2853 			outb(CMOS_DATA, (byte | EXT_BANK));
2854 
2855 			outb(CMOS_ADDR, PFR_REG);
2856 
2857 			/* for Predator must toggle the PAB bit */
2858 			byte = inb(CMOS_DATA);
2859 
2860 			/*
2861 			 * clear power active bar, wakeup alarm and
2862 			 * kickstart
2863 			 */
2864 			byte &= ~(PAB_CBIT | WF_FLAG | KS_FLAG);
2865 			outb(CMOS_DATA, byte);
2866 
2867 			/* delay before next write */
2868 			drv_usecwait(1000);
2869 
2870 			/* for S40 the following would suffice */
2871 			byte = inb(CMOS_DATA);
2872 
2873 			/* power active bar control bit */
2874 			byte |= PAB_CBIT;
2875 			outb(CMOS_DATA, byte);
2876 
2877 			break;
2878 
2879 		case APIC_POWEROFF_VIA_ASPEN_BMC:
2880 			restarts = 0;
2881 restart_aspen_bmc:
2882 			if (++restarts == 3)
2883 				break;
2884 			attempts = 0;
2885 			do {
2886 				byte = inb(MISMIC_FLAG_REGISTER);
2887 				byte &= MISMIC_BUSY_MASK;
2888 				if (byte != 0) {
2889 					drv_usecwait(1000);
2890 					if (attempts >= 3)
2891 						goto restart_aspen_bmc;
2892 					++attempts;
2893 				}
2894 			} while (byte != 0);
2895 			outb(MISMIC_CNTL_REGISTER, CC_SMS_GET_STATUS);
2896 			byte = inb(MISMIC_FLAG_REGISTER);
2897 			byte |= 0x1;
2898 			outb(MISMIC_FLAG_REGISTER, byte);
2899 			i = 0;
2900 			for (; i < (sizeof (aspen_bmc)/sizeof (aspen_bmc[0]));
2901 			    i++) {
2902 				attempts = 0;
2903 				do {
2904 					byte = inb(MISMIC_FLAG_REGISTER);
2905 					byte &= MISMIC_BUSY_MASK;
2906 					if (byte != 0) {
2907 						drv_usecwait(1000);
2908 						if (attempts >= 3)
2909 							goto restart_aspen_bmc;
2910 						++attempts;
2911 					}
2912 				} while (byte != 0);
2913 				outb(MISMIC_CNTL_REGISTER, aspen_bmc[i].cntl);
2914 				outb(MISMIC_DATA_REGISTER, aspen_bmc[i].data);
2915 				byte = inb(MISMIC_FLAG_REGISTER);
2916 				byte |= 0x1;
2917 				outb(MISMIC_FLAG_REGISTER, byte);
2918 			}
2919 			break;
2920 
2921 		case APIC_POWEROFF_VIA_SITKA_BMC:
2922 			restarts = 0;
2923 restart_sitka_bmc:
2924 			if (++restarts == 3)
2925 				break;
2926 			attempts = 0;
2927 			do {
2928 				byte = inb(SMS_STATUS_REGISTER);
2929 				byte &= SMS_STATE_MASK;
2930 				if ((byte == SMS_READ_STATE) ||
2931 				    (byte == SMS_WRITE_STATE)) {
2932 					drv_usecwait(1000);
2933 					if (attempts >= 3)
2934 						goto restart_sitka_bmc;
2935 					++attempts;
2936 				}
2937 			} while ((byte == SMS_READ_STATE) ||
2938 			    (byte == SMS_WRITE_STATE));
2939 			outb(SMS_COMMAND_REGISTER, SMS_GET_STATUS);
2940 			i = 0;
2941 			for (; i < (sizeof (sitka_bmc)/sizeof (sitka_bmc[0]));
2942 			    i++) {
2943 				attempts = 0;
2944 				do {
2945 					byte = inb(SMS_STATUS_REGISTER);
2946 					byte &= SMS_IBF_MASK;
2947 					if (byte != 0) {
2948 						drv_usecwait(1000);
2949 						if (attempts >= 3)
2950 							goto restart_sitka_bmc;
2951 						++attempts;
2952 					}
2953 				} while (byte != 0);
2954 				outb(sitka_bmc[i].port, sitka_bmc[i].data);
2955 			}
2956 			break;
2957 
2958 		case APIC_POWEROFF_NONE:
2959 
2960 			/* If no APIC direct method, we will try using ACPI */
2961 			if (apic_enable_acpi) {
2962 				if (acpi_poweroff() == 1)
2963 					return;
2964 			} else
2965 				return;
2966 
2967 			break;
2968 	}
2969 	/*
2970 	 * Wait a limited time here for power to go off.
2971 	 * If the power does not go off, then there was a
2972 	 * problem and we should continue to the halt which
2973 	 * prints a message for the user to press a key to
2974 	 * reboot.
2975 	 */
2976 	drv_usecwait(7000000); /* wait seven seconds */
2977 
2978 }
2979 
2980 /*
2981  * Try and disable all interrupts. We just assign interrupts to other
2982  * processors based on policy. If any were bound by user request, we
2983  * let them continue and return failure. We do not bother to check
2984  * for cache affinity while rebinding.
2985  */
2986 
2987 static int
2988 apic_disable_intr(processorid_t cpun)
2989 {
2990 	int bind_cpu = 0, i, hardbound = 0, iflag;
2991 	apic_irq_t *irq_ptr;
2992 
2993 	if (cpun == 0)
2994 		return (PSM_FAILURE);
2995 
2996 	iflag = intr_clear();
2997 	lock_set(&apic_ioapic_lock);
2998 	apic_cpus[cpun].aci_status &= ~APIC_CPU_INTR_ENABLE;
2999 	lock_clear(&apic_ioapic_lock);
3000 	intr_restore(iflag);
3001 	apic_cpus[cpun].aci_curipl = 0;
3002 	i = apic_min_device_irq;
3003 	for (; i <= apic_max_device_irq; i++) {
3004 		/*
3005 		 * If there are bound interrupts on this cpu, then
3006 		 * rebind them to other processors.
3007 		 */
3008 		if ((irq_ptr = apic_irq_table[i]) != NULL) {
3009 			ASSERT((irq_ptr->airq_temp_cpu == IRQ_UNBOUND) ||
3010 			    (irq_ptr->airq_temp_cpu == IRQ_UNINIT) ||
3011 			    ((irq_ptr->airq_temp_cpu & ~IRQ_USER_BOUND) <
3012 			    apic_nproc));
3013 
3014 			if (irq_ptr->airq_temp_cpu == (cpun | IRQ_USER_BOUND)) {
3015 				hardbound = 1;
3016 				continue;
3017 			}
3018 
3019 			if (irq_ptr->airq_temp_cpu == cpun) {
3020 				do {
3021 					apic_next_bind_cpu += 2;
3022 					bind_cpu = apic_next_bind_cpu / 2;
3023 					if (bind_cpu >= apic_nproc) {
3024 						apic_next_bind_cpu = 1;
3025 						bind_cpu = 0;
3026 
3027 					}
3028 				} while (apic_rebind_all(irq_ptr, bind_cpu, 1));
3029 			}
3030 		}
3031 	}
3032 	if (hardbound) {
3033 		cmn_err(CE_WARN, "Could not disable interrupts on %d"
3034 		    "due to user bound interrupts", cpun);
3035 		return (PSM_FAILURE);
3036 	}
3037 	else
3038 		return (PSM_SUCCESS);
3039 }
3040 
3041 static void
3042 apic_enable_intr(processorid_t cpun)
3043 {
3044 	int	i, iflag;
3045 	apic_irq_t *irq_ptr;
3046 
3047 	iflag = intr_clear();
3048 	lock_set(&apic_ioapic_lock);
3049 	apic_cpus[cpun].aci_status |= APIC_CPU_INTR_ENABLE;
3050 	lock_clear(&apic_ioapic_lock);
3051 	intr_restore(iflag);
3052 
3053 	i = apic_min_device_irq;
3054 	for (i = apic_min_device_irq; i <= apic_max_device_irq; i++) {
3055 		if ((irq_ptr = apic_irq_table[i]) != NULL) {
3056 			if ((irq_ptr->airq_cpu & ~IRQ_USER_BOUND) == cpun) {
3057 				(void) apic_rebind_all(irq_ptr,
3058 				    irq_ptr->airq_cpu, 1);
3059 			}
3060 		}
3061 	}
3062 }
3063 
3064 /*
3065  * apic_introp_xlate() replaces apic_translate_irq() and is
3066  * called only from apic_intr_ops().  With the new ADII framework,
3067  * the priority can no longer be retrived through i_ddi_get_intrspec().
3068  * It has to be passed in from the caller.
3069  */
3070 int
3071 apic_introp_xlate(dev_info_t *dip, struct intrspec *ispec, int type)
3072 {
3073 	char dev_type[16];
3074 	int dev_len, pci_irq, newirq, bustype, devid, busid, i;
3075 	int irqno = ispec->intrspec_vec;
3076 	ddi_acc_handle_t cfg_handle;
3077 	uchar_t ipin;
3078 	struct apic_io_intr *intrp;
3079 	iflag_t intr_flag;
3080 	APIC_HEADER	*hp;
3081 	MADT_INTERRUPT_OVERRIDE	*isop;
3082 	apic_irq_t *airqp;
3083 
3084 	DDI_INTR_IMPLDBG((CE_CONT, "apic_introp_xlate: dip=0x%p name=%s "
3085 	    "type=%d irqno=0x%x\n", (void *)dip, ddi_get_name(dip), type,
3086 	    irqno));
3087 
3088 	if (DDI_INTR_IS_MSI_OR_MSIX(type)) {
3089 		if ((airqp = apic_find_irq(dip, ispec, type)) != NULL)
3090 			return (apic_vector_to_irq[airqp->airq_vector]);
3091 		return (apic_setup_irq_table(dip, irqno, NULL, ispec,
3092 		    NULL, type));
3093 	}
3094 
3095 	bustype = 0;
3096 
3097 	/* check if we have already translated this irq */
3098 	mutex_enter(&airq_mutex);
3099 	newirq = apic_min_device_irq;
3100 	for (; newirq <= apic_max_device_irq; newirq++) {
3101 		airqp = apic_irq_table[newirq];
3102 		while (airqp) {
3103 			if ((airqp->airq_dip == dip) &&
3104 			    (airqp->airq_origirq == irqno) &&
3105 			    (airqp->airq_mps_intr_index != FREE_INDEX)) {
3106 
3107 				mutex_exit(&airq_mutex);
3108 				return (VIRTIRQ(newirq, airqp->airq_share_id));
3109 			}
3110 			airqp = airqp->airq_next;
3111 		}
3112 	}
3113 	mutex_exit(&airq_mutex);
3114 
3115 	if (apic_defconf)
3116 		goto defconf;
3117 
3118 	if ((dip == NULL) || (!apic_irq_translate && !apic_enable_acpi))
3119 		goto nonpci;
3120 
3121 	dev_len = sizeof (dev_type);
3122 	if (ddi_getlongprop_buf(DDI_DEV_T_NONE, ddi_get_parent(dip),
3123 	    DDI_PROP_DONTPASS, "device_type", (caddr_t)dev_type,
3124 	    &dev_len) != DDI_PROP_SUCCESS) {
3125 		goto nonpci;
3126 	}
3127 
3128 	if (strcmp(dev_type, "pci") == 0) {
3129 		/* pci device */
3130 		if (acpica_get_bdf(dip, &busid, &devid, NULL) != 0)
3131 			goto nonpci;
3132 		if (busid == 0 && apic_pci_bus_total == 1)
3133 			busid = (int)apic_single_pci_busid;
3134 
3135 		if (pci_config_setup(dip, &cfg_handle) != DDI_SUCCESS)
3136 			goto nonpci;
3137 		ipin = pci_config_get8(cfg_handle, PCI_CONF_IPIN) - PCI_INTA;
3138 		pci_config_teardown(&cfg_handle);
3139 		if (apic_enable_acpi && !apic_use_acpi_madt_only) {
3140 			if (apic_acpi_translate_pci_irq(dip, busid, devid,
3141 			    ipin, &pci_irq, &intr_flag) != ACPI_PSM_SUCCESS)
3142 				goto nonpci;
3143 
3144 			intr_flag.bustype = BUS_PCI;
3145 			if ((newirq = apic_setup_irq_table(dip, pci_irq, NULL,
3146 			    ispec, &intr_flag, type)) == -1)
3147 				goto nonpci;
3148 			return (newirq);
3149 		} else {
3150 			pci_irq = ((devid & 0x1f) << 2) | (ipin & 0x3);
3151 			if ((intrp = apic_find_io_intr_w_busid(pci_irq, busid))
3152 			    == NULL) {
3153 				if ((pci_irq = apic_handle_pci_pci_bridge(dip,
3154 				    devid, ipin, &intrp)) == -1)
3155 					goto nonpci;
3156 			}
3157 			if ((newirq = apic_setup_irq_table(dip, pci_irq, intrp,
3158 			    ispec, NULL, type)) == -1)
3159 				goto nonpci;
3160 			return (newirq);
3161 		}
3162 	} else if (strcmp(dev_type, "isa") == 0)
3163 		bustype = BUS_ISA;
3164 	else if (strcmp(dev_type, "eisa") == 0)
3165 		bustype = BUS_EISA;
3166 
3167 nonpci:
3168 	if (apic_enable_acpi && !apic_use_acpi_madt_only) {
3169 		/* search iso entries first */
3170 		if (acpi_iso_cnt != 0) {
3171 			hp = (APIC_HEADER *)acpi_isop;
3172 			i = 0;
3173 			while (i < acpi_iso_cnt) {
3174 				if (hp->Type == APIC_XRUPT_OVERRIDE) {
3175 					isop = (MADT_INTERRUPT_OVERRIDE *)hp;
3176 					if (isop->Bus == 0 &&
3177 					    isop->Source == irqno) {
3178 						newirq = isop->Interrupt;
3179 						intr_flag.intr_po =
3180 						    isop->Polarity;
3181 						intr_flag.intr_el =
3182 						    isop->TriggerMode;
3183 						intr_flag.bustype = BUS_ISA;
3184 
3185 						return (apic_setup_irq_table(
3186 						    dip, newirq, NULL, ispec,
3187 						    &intr_flag, type));
3188 
3189 					}
3190 					i++;
3191 				}
3192 				hp = (APIC_HEADER *)(((char *)hp) +
3193 				    hp->Length);
3194 			}
3195 		}
3196 		intr_flag.intr_po = INTR_PO_ACTIVE_HIGH;
3197 		intr_flag.intr_el = INTR_EL_EDGE;
3198 		intr_flag.bustype = BUS_ISA;
3199 		return (apic_setup_irq_table(dip, irqno, NULL, ispec,
3200 		    &intr_flag, type));
3201 	} else {
3202 		if (bustype == 0)
3203 			bustype = eisa_level_intr_mask ? BUS_EISA : BUS_ISA;
3204 		for (i = 0; i < 2; i++) {
3205 			if (((busid = apic_find_bus_id(bustype)) != -1) &&
3206 			    ((intrp = apic_find_io_intr_w_busid(irqno, busid))
3207 			    != NULL)) {
3208 				if ((newirq = apic_setup_irq_table(dip, irqno,
3209 				    intrp, ispec, NULL, type)) != -1) {
3210 					return (newirq);
3211 				}
3212 				goto defconf;
3213 			}
3214 			bustype = (bustype == BUS_EISA) ? BUS_ISA : BUS_EISA;
3215 		}
3216 	}
3217 
3218 /* MPS default configuration */
3219 defconf:
3220 	newirq = apic_setup_irq_table(dip, irqno, NULL, ispec, NULL, type);
3221 	if (newirq == -1)
3222 		return (newirq);
3223 	ASSERT(IRQINDEX(newirq) == irqno);
3224 	ASSERT(apic_irq_table[irqno]);
3225 	return (newirq);
3226 }
3227 
3228 
3229 
3230 
3231 
3232 
3233 /*
3234  * On machines with PCI-PCI bridges, a device behind a PCI-PCI bridge
3235  * needs special handling.  We may need to chase up the device tree,
3236  * using the PCI-PCI Bridge specification's "rotating IPIN assumptions",
3237  * to find the IPIN at the root bus that relates to the IPIN on the
3238  * subsidiary bus (for ACPI or MP).  We may, however, have an entry
3239  * in the MP table or the ACPI namespace for this device itself.
3240  * We handle both cases in the search below.
3241  */
3242 /* this is the non-acpi version */
3243 static int
3244 apic_handle_pci_pci_bridge(dev_info_t *idip, int child_devno, int child_ipin,
3245 			struct apic_io_intr **intrp)
3246 {
3247 	dev_info_t *dipp, *dip;
3248 	int pci_irq;
3249 	ddi_acc_handle_t cfg_handle;
3250 	int bridge_devno, bridge_bus;
3251 	int ipin;
3252 
3253 	dip = idip;
3254 
3255 	/*CONSTCOND*/
3256 	while (1) {
3257 		if ((dipp = ddi_get_parent(dip)) == (dev_info_t *)NULL)
3258 			return (-1);
3259 		if ((pci_config_setup(dipp, &cfg_handle) == DDI_SUCCESS) &&
3260 		    (pci_config_get8(cfg_handle, PCI_CONF_BASCLASS) ==
3261 		    PCI_CLASS_BRIDGE) && (pci_config_get8(cfg_handle,
3262 		    PCI_CONF_SUBCLASS) == PCI_BRIDGE_PCI)) {
3263 			pci_config_teardown(&cfg_handle);
3264 			if (acpica_get_bdf(dipp, &bridge_bus, &bridge_devno,
3265 			    NULL) != 0)
3266 				return (-1);
3267 			/*
3268 			 * This is the rotating scheme that Compaq is using
3269 			 * and documented in the pci to pci spec.  Also, if
3270 			 * the pci to pci bridge is behind another pci to
3271 			 * pci bridge, then it need to keep transversing
3272 			 * up until an interrupt entry is found or reach
3273 			 * the top of the tree
3274 			 */
3275 			ipin = (child_devno + child_ipin) % PCI_INTD;
3276 				if (bridge_bus == 0 && apic_pci_bus_total == 1)
3277 					bridge_bus = (int)apic_single_pci_busid;
3278 				pci_irq = ((bridge_devno & 0x1f) << 2) |
3279 				    (ipin & 0x3);
3280 				if ((*intrp = apic_find_io_intr_w_busid(pci_irq,
3281 				    bridge_bus)) != NULL) {
3282 					return (pci_irq);
3283 				}
3284 			dip = dipp;
3285 			child_devno = bridge_devno;
3286 			child_ipin = ipin;
3287 		} else
3288 			return (-1);
3289 	}
3290 	/*LINTED: function will not fall off the bottom */
3291 }
3292 
3293 
3294 
3295 
3296 static uchar_t
3297 acpi_find_ioapic(int irq)
3298 {
3299 	int i;
3300 
3301 	for (i = 0; i < apic_io_max; i++) {
3302 		if (irq >= apic_io_vectbase[i] && irq <= apic_io_vectend[i])
3303 			return (i);
3304 	}
3305 	return (0xFF);	/* shouldn't happen */
3306 }
3307 
3308 /*
3309  * See if two irqs are compatible for sharing a vector.
3310  * Currently we only support sharing of PCI devices.
3311  */
3312 static int
3313 acpi_intr_compatible(iflag_t iflag1, iflag_t iflag2)
3314 {
3315 	uint_t	level1, po1;
3316 	uint_t	level2, po2;
3317 
3318 	/* Assume active high by default */
3319 	po1 = 0;
3320 	po2 = 0;
3321 
3322 	if (iflag1.bustype != iflag2.bustype || iflag1.bustype != BUS_PCI)
3323 		return (0);
3324 
3325 	if (iflag1.intr_el == INTR_EL_CONFORM)
3326 		level1 = AV_LEVEL;
3327 	else
3328 		level1 = (iflag1.intr_el == INTR_EL_LEVEL) ? AV_LEVEL : 0;
3329 
3330 	if (level1 && ((iflag1.intr_po == INTR_PO_ACTIVE_LOW) ||
3331 	    (iflag1.intr_po == INTR_PO_CONFORM)))
3332 		po1 = AV_ACTIVE_LOW;
3333 
3334 	if (iflag2.intr_el == INTR_EL_CONFORM)
3335 		level2 = AV_LEVEL;
3336 	else
3337 		level2 = (iflag2.intr_el == INTR_EL_LEVEL) ? AV_LEVEL : 0;
3338 
3339 	if (level2 && ((iflag2.intr_po == INTR_PO_ACTIVE_LOW) ||
3340 	    (iflag2.intr_po == INTR_PO_CONFORM)))
3341 		po2 = AV_ACTIVE_LOW;
3342 
3343 	if ((level1 == level2) && (po1 == po2))
3344 		return (1);
3345 
3346 	return (0);
3347 }
3348 
3349 /*
3350  * Attempt to share vector with someone else
3351  */
3352 static int
3353 apic_share_vector(int irqno, iflag_t *intr_flagp, short intr_index, int ipl,
3354 	uchar_t ioapicindex, uchar_t ipin, apic_irq_t **irqptrp)
3355 {
3356 #ifdef DEBUG
3357 	apic_irq_t *tmpirqp = NULL;
3358 #endif /* DEBUG */
3359 	apic_irq_t *irqptr, dummyirq;
3360 	int	newirq, chosen_irq = -1, share = 127;
3361 	int	lowest, highest, i;
3362 	uchar_t	share_id;
3363 
3364 	DDI_INTR_IMPLDBG((CE_CONT, "apic_share_vector: irqno=0x%x "
3365 	    "intr_index=0x%x ipl=0x%x\n", irqno, intr_index, ipl));
3366 
3367 	highest = apic_ipltopri[ipl] + APIC_VECTOR_MASK;
3368 	lowest = apic_ipltopri[ipl-1] + APIC_VECTOR_PER_IPL;
3369 
3370 	if (highest < lowest) /* Both ipl and ipl-1 map to same pri */
3371 		lowest -= APIC_VECTOR_PER_IPL;
3372 	dummyirq.airq_mps_intr_index = intr_index;
3373 	dummyirq.airq_ioapicindex = ioapicindex;
3374 	dummyirq.airq_intin_no = ipin;
3375 	if (intr_flagp)
3376 		dummyirq.airq_iflag = *intr_flagp;
3377 	apic_record_rdt_entry(&dummyirq, irqno);
3378 	for (i = lowest; i <= highest; i++) {
3379 		newirq = apic_vector_to_irq[i];
3380 		if (newirq == APIC_RESV_IRQ)
3381 			continue;
3382 		irqptr = apic_irq_table[newirq];
3383 
3384 		if ((dummyirq.airq_rdt_entry & 0xFF00) !=
3385 		    (irqptr->airq_rdt_entry & 0xFF00))
3386 			/* not compatible */
3387 			continue;
3388 
3389 		if (irqptr->airq_share < share) {
3390 			share = irqptr->airq_share;
3391 			chosen_irq = newirq;
3392 		}
3393 	}
3394 	if (chosen_irq != -1) {
3395 		/*
3396 		 * Assign a share id which is free or which is larger
3397 		 * than the largest one.
3398 		 */
3399 		share_id = 1;
3400 		mutex_enter(&airq_mutex);
3401 		irqptr = apic_irq_table[chosen_irq];
3402 		while (irqptr) {
3403 			if (irqptr->airq_mps_intr_index == FREE_INDEX) {
3404 				share_id = irqptr->airq_share_id;
3405 				break;
3406 			}
3407 			if (share_id <= irqptr->airq_share_id)
3408 				share_id = irqptr->airq_share_id + 1;
3409 #ifdef DEBUG
3410 			tmpirqp = irqptr;
3411 #endif /* DEBUG */
3412 			irqptr = irqptr->airq_next;
3413 		}
3414 		if (!irqptr) {
3415 			irqptr = kmem_zalloc(sizeof (apic_irq_t), KM_SLEEP);
3416 			irqptr->airq_temp_cpu = IRQ_UNINIT;
3417 			irqptr->airq_next =
3418 			    apic_irq_table[chosen_irq]->airq_next;
3419 			apic_irq_table[chosen_irq]->airq_next = irqptr;
3420 #ifdef	DEBUG
3421 			tmpirqp = apic_irq_table[chosen_irq];
3422 #endif /* DEBUG */
3423 		}
3424 		irqptr->airq_mps_intr_index = intr_index;
3425 		irqptr->airq_ioapicindex = ioapicindex;
3426 		irqptr->airq_intin_no = ipin;
3427 		if (intr_flagp)
3428 			irqptr->airq_iflag = *intr_flagp;
3429 		irqptr->airq_vector = apic_irq_table[chosen_irq]->airq_vector;
3430 		irqptr->airq_share_id = share_id;
3431 		apic_record_rdt_entry(irqptr, irqno);
3432 		*irqptrp = irqptr;
3433 #ifdef	DEBUG
3434 		/* shuffle the pointers to test apic_delspl path */
3435 		if (tmpirqp) {
3436 			tmpirqp->airq_next = irqptr->airq_next;
3437 			irqptr->airq_next = apic_irq_table[chosen_irq];
3438 			apic_irq_table[chosen_irq] = irqptr;
3439 		}
3440 #endif /* DEBUG */
3441 		mutex_exit(&airq_mutex);
3442 		return (VIRTIRQ(chosen_irq, share_id));
3443 	}
3444 	return (-1);
3445 }
3446 
3447 /*
3448  *
3449  */
3450 static int
3451 apic_setup_irq_table(dev_info_t *dip, int irqno, struct apic_io_intr *intrp,
3452     struct intrspec *ispec, iflag_t *intr_flagp, int type)
3453 {
3454 	int origirq = ispec->intrspec_vec;
3455 	uchar_t ipl = ispec->intrspec_pri;
3456 	int	newirq, intr_index;
3457 	uchar_t	ipin, ioapic, ioapicindex, vector;
3458 	apic_irq_t *irqptr;
3459 	major_t	major;
3460 	dev_info_t	*sdip;
3461 
3462 	DDI_INTR_IMPLDBG((CE_CONT, "apic_setup_irq_table: dip=0x%p type=%d "
3463 	    "irqno=0x%x origirq=0x%x\n", (void *)dip, type, irqno, origirq));
3464 
3465 	ASSERT(ispec != NULL);
3466 
3467 	major =  (dip != NULL) ? ddi_name_to_major(ddi_get_name(dip)) : 0;
3468 
3469 	if (DDI_INTR_IS_MSI_OR_MSIX(type)) {
3470 		/* MSI/X doesn't need to setup ioapic stuffs */
3471 		ioapicindex = 0xff;
3472 		ioapic = 0xff;
3473 		ipin = (uchar_t)0xff;
3474 		intr_index = (type == DDI_INTR_TYPE_MSI) ? MSI_INDEX :
3475 		    MSIX_INDEX;
3476 		mutex_enter(&airq_mutex);
3477 		if ((irqno = apic_allocate_irq(APIC_FIRST_FREE_IRQ)) == -1) {
3478 			mutex_exit(&airq_mutex);
3479 			/* need an irq for MSI/X to index into autovect[] */
3480 			cmn_err(CE_WARN, "No interrupt irq: %s instance %d",
3481 			    ddi_get_name(dip), ddi_get_instance(dip));
3482 			return (-1);
3483 		}
3484 		mutex_exit(&airq_mutex);
3485 
3486 	} else if (intrp != NULL) {
3487 		intr_index = (int)(intrp - apic_io_intrp);
3488 		ioapic = intrp->intr_destid;
3489 		ipin = intrp->intr_destintin;
3490 		/* Find ioapicindex. If destid was ALL, we will exit with 0. */
3491 		for (ioapicindex = apic_io_max - 1; ioapicindex; ioapicindex--)
3492 			if (apic_io_id[ioapicindex] == ioapic)
3493 				break;
3494 		ASSERT((ioapic == apic_io_id[ioapicindex]) ||
3495 		    (ioapic == INTR_ALL_APIC));
3496 
3497 		/* check whether this intin# has been used by another irqno */
3498 		if ((newirq = apic_find_intin(ioapicindex, ipin)) != -1) {
3499 			return (newirq);
3500 		}
3501 
3502 	} else if (intr_flagp != NULL) {
3503 		/* ACPI case */
3504 		intr_index = ACPI_INDEX;
3505 		ioapicindex = acpi_find_ioapic(irqno);
3506 		ASSERT(ioapicindex != 0xFF);
3507 		ioapic = apic_io_id[ioapicindex];
3508 		ipin = irqno - apic_io_vectbase[ioapicindex];
3509 		if (apic_irq_table[irqno] &&
3510 		    apic_irq_table[irqno]->airq_mps_intr_index == ACPI_INDEX) {
3511 			ASSERT(apic_irq_table[irqno]->airq_intin_no == ipin &&
3512 			    apic_irq_table[irqno]->airq_ioapicindex ==
3513 			    ioapicindex);
3514 			return (irqno);
3515 		}
3516 
3517 	} else {
3518 		/* default configuration */
3519 		ioapicindex = 0;
3520 		ioapic = apic_io_id[ioapicindex];
3521 		ipin = (uchar_t)irqno;
3522 		intr_index = DEFAULT_INDEX;
3523 	}
3524 
3525 	if (ispec == NULL) {
3526 		APIC_VERBOSE_IOAPIC((CE_WARN, "No intrspec for irqno = %x\n",
3527 		    irqno));
3528 	} else if ((vector = apic_allocate_vector(ipl, irqno, 0)) == 0) {
3529 		if ((newirq = apic_share_vector(irqno, intr_flagp, intr_index,
3530 		    ipl, ioapicindex, ipin, &irqptr)) != -1) {
3531 			irqptr->airq_ipl = ipl;
3532 			irqptr->airq_origirq = (uchar_t)origirq;
3533 			irqptr->airq_dip = dip;
3534 			irqptr->airq_major = major;
3535 			sdip = apic_irq_table[IRQINDEX(newirq)]->airq_dip;
3536 			/* This is OK to do really */
3537 			if (sdip == NULL) {
3538 				cmn_err(CE_WARN, "Sharing vectors: %s"
3539 				    " instance %d and SCI",
3540 				    ddi_get_name(dip), ddi_get_instance(dip));
3541 			} else {
3542 				cmn_err(CE_WARN, "Sharing vectors: %s"
3543 				    " instance %d and %s instance %d",
3544 				    ddi_get_name(sdip), ddi_get_instance(sdip),
3545 				    ddi_get_name(dip), ddi_get_instance(dip));
3546 			}
3547 			return (newirq);
3548 		}
3549 		/* try high priority allocation now  that share has failed */
3550 		if ((vector = apic_allocate_vector(ipl, irqno, 1)) == 0) {
3551 			cmn_err(CE_WARN, "No interrupt vector: %s instance %d",
3552 			    ddi_get_name(dip), ddi_get_instance(dip));
3553 			return (-1);
3554 		}
3555 	}
3556 
3557 	mutex_enter(&airq_mutex);
3558 	if (apic_irq_table[irqno] == NULL) {
3559 		irqptr = kmem_zalloc(sizeof (apic_irq_t), KM_SLEEP);
3560 		irqptr->airq_temp_cpu = IRQ_UNINIT;
3561 		apic_irq_table[irqno] = irqptr;
3562 	} else {
3563 		irqptr = apic_irq_table[irqno];
3564 		if (irqptr->airq_mps_intr_index != FREE_INDEX) {
3565 			/*
3566 			 * The slot is used by another irqno, so allocate
3567 			 * a free irqno for this interrupt
3568 			 */
3569 			newirq = apic_allocate_irq(APIC_FIRST_FREE_IRQ);
3570 			if (newirq == -1) {
3571 				mutex_exit(&airq_mutex);
3572 				return (-1);
3573 			}
3574 			irqno = newirq;
3575 			irqptr = apic_irq_table[irqno];
3576 			if (irqptr == NULL) {
3577 				irqptr = kmem_zalloc(sizeof (apic_irq_t),
3578 				    KM_SLEEP);
3579 				irqptr->airq_temp_cpu = IRQ_UNINIT;
3580 				apic_irq_table[irqno] = irqptr;
3581 			}
3582 			apic_modify_vector(vector, newirq);
3583 		}
3584 	}
3585 	apic_max_device_irq = max(irqno, apic_max_device_irq);
3586 	apic_min_device_irq = min(irqno, apic_min_device_irq);
3587 	mutex_exit(&airq_mutex);
3588 	irqptr->airq_ioapicindex = ioapicindex;
3589 	irqptr->airq_intin_no = ipin;
3590 	irqptr->airq_ipl = ipl;
3591 	irqptr->airq_vector = vector;
3592 	irqptr->airq_origirq = (uchar_t)origirq;
3593 	irqptr->airq_share_id = 0;
3594 	irqptr->airq_mps_intr_index = (short)intr_index;
3595 	irqptr->airq_dip = dip;
3596 	irqptr->airq_major = major;
3597 	irqptr->airq_cpu = apic_bind_intr(dip, irqno, ioapic, ipin);
3598 	if (intr_flagp)
3599 		irqptr->airq_iflag = *intr_flagp;
3600 
3601 	if (!DDI_INTR_IS_MSI_OR_MSIX(type)) {
3602 		/* setup I/O APIC entry for non-MSI/X interrupts */
3603 		apic_record_rdt_entry(irqptr, irqno);
3604 	}
3605 	return (irqno);
3606 }
3607 
3608 /*
3609  * return the cpu to which this intr should be bound.
3610  * Check properties or any other mechanism to see if user wants it
3611  * bound to a specific CPU. If so, return the cpu id with high bit set.
3612  * If not, use the policy to choose a cpu and return the id.
3613  */
3614 uchar_t
3615 apic_bind_intr(dev_info_t *dip, int irq, uchar_t ioapicid, uchar_t intin)
3616 {
3617 	int	instance, instno, prop_len, bind_cpu, count;
3618 	uint_t	i, rc;
3619 	uchar_t	cpu;
3620 	major_t	major;
3621 	char	*name, *drv_name, *prop_val, *cptr;
3622 	char	prop_name[32];
3623 
3624 
3625 	if (apic_intr_policy == INTR_LOWEST_PRIORITY)
3626 		return (IRQ_UNBOUND);
3627 
3628 	drv_name = NULL;
3629 	rc = DDI_PROP_NOT_FOUND;
3630 	major = (major_t)-1;
3631 	if (dip != NULL) {
3632 		name = ddi_get_name(dip);
3633 		major = ddi_name_to_major(name);
3634 		drv_name = ddi_major_to_name(major);
3635 		instance = ddi_get_instance(dip);
3636 		if (apic_intr_policy == INTR_ROUND_ROBIN_WITH_AFFINITY) {
3637 			i = apic_min_device_irq;
3638 			for (; i <= apic_max_device_irq; i++) {
3639 
3640 				if ((i == irq) || (apic_irq_table[i] == NULL) ||
3641 				    (apic_irq_table[i]->airq_mps_intr_index
3642 				    == FREE_INDEX))
3643 					continue;
3644 
3645 				if ((apic_irq_table[i]->airq_major == major) &&
3646 				    (!(apic_irq_table[i]->airq_cpu &
3647 				    IRQ_USER_BOUND))) {
3648 
3649 					cpu = apic_irq_table[i]->airq_cpu;
3650 
3651 					cmn_err(CE_CONT,
3652 					    "!pcplusmp: %s (%s) instance #%d "
3653 					    "vector 0x%x ioapic 0x%x "
3654 					    "intin 0x%x is bound to cpu %d\n",
3655 					    name, drv_name, instance, irq,
3656 					    ioapicid, intin, cpu);
3657 					return (cpu);
3658 				}
3659 			}
3660 		}
3661 		/*
3662 		 * search for "drvname"_intpt_bind_cpus property first, the
3663 		 * syntax of the property should be "a[,b,c,...]" where
3664 		 * instance 0 binds to cpu a, instance 1 binds to cpu b,
3665 		 * instance 3 binds to cpu c...
3666 		 * ddi_getlongprop() will search /option first, then /
3667 		 * if "drvname"_intpt_bind_cpus doesn't exist, then find
3668 		 * intpt_bind_cpus property.  The syntax is the same, and
3669 		 * it applies to all the devices if its "drvname" specific
3670 		 * property doesn't exist
3671 		 */
3672 		(void) strcpy(prop_name, drv_name);
3673 		(void) strcat(prop_name, "_intpt_bind_cpus");
3674 		rc = ddi_getlongprop(DDI_DEV_T_ANY, dip, 0, prop_name,
3675 		    (caddr_t)&prop_val, &prop_len);
3676 		if (rc != DDI_PROP_SUCCESS) {
3677 			rc = ddi_getlongprop(DDI_DEV_T_ANY, dip, 0,
3678 			    "intpt_bind_cpus", (caddr_t)&prop_val, &prop_len);
3679 		}
3680 	}
3681 	if (rc == DDI_PROP_SUCCESS) {
3682 		for (i = count = 0; i < (prop_len - 1); i++)
3683 			if (prop_val[i] == ',')
3684 				count++;
3685 		if (prop_val[i-1] != ',')
3686 			count++;
3687 		/*
3688 		 * if somehow the binding instances defined in the
3689 		 * property are not enough for this instno., then
3690 		 * reuse the pattern for the next instance until
3691 		 * it reaches the requested instno
3692 		 */
3693 		instno = instance % count;
3694 		i = 0;
3695 		cptr = prop_val;
3696 		while (i < instno)
3697 			if (*cptr++ == ',')
3698 				i++;
3699 		bind_cpu = stoi(&cptr);
3700 		kmem_free(prop_val, prop_len);
3701 		/* if specific cpu is bogus, then default to cpu 0 */
3702 		if (bind_cpu >= apic_nproc) {
3703 			cmn_err(CE_WARN, "pcplusmp: %s=%s: CPU %d not present",
3704 			    prop_name, prop_val, bind_cpu);
3705 			bind_cpu = 0;
3706 		} else {
3707 			/* indicate that we are bound at user request */
3708 			bind_cpu |= IRQ_USER_BOUND;
3709 		}
3710 		/*
3711 		 * no need to check apic_cpus[].aci_status, if specific cpu is
3712 		 * not up, then post_cpu_start will handle it.
3713 		 */
3714 	} else {
3715 		/*
3716 		 * We change bind_cpu only for every two calls
3717 		 * as most drivers still do 2 add_intrs for every
3718 		 * interrupt
3719 		 */
3720 		bind_cpu = (apic_next_bind_cpu++) / 2;
3721 		if (bind_cpu >= apic_nproc) {
3722 			apic_next_bind_cpu = 1;
3723 			bind_cpu = 0;
3724 		}
3725 	}
3726 	if (drv_name != NULL)
3727 		cmn_err(CE_CONT, "!pcplusmp: %s (%s) instance %d "
3728 		    "vector 0x%x ioapic 0x%x intin 0x%x is bound to cpu %d\n",
3729 		    name, drv_name, instance,
3730 		    irq, ioapicid, intin, bind_cpu & ~IRQ_USER_BOUND);
3731 	else
3732 		cmn_err(CE_CONT, "!pcplusmp: "
3733 		    "vector 0x%x ioapic 0x%x intin 0x%x is bound to cpu %d\n",
3734 		    irq, ioapicid, intin, bind_cpu & ~IRQ_USER_BOUND);
3735 
3736 	return ((uchar_t)bind_cpu);
3737 }
3738 
3739 static struct apic_io_intr *
3740 apic_find_io_intr_w_busid(int irqno, int busid)
3741 {
3742 	struct	apic_io_intr	*intrp;
3743 
3744 	/*
3745 	 * It can have more than 1 entry with same source bus IRQ,
3746 	 * but unique with the source bus id
3747 	 */
3748 	intrp = apic_io_intrp;
3749 	if (intrp != NULL) {
3750 		while (intrp->intr_entry == APIC_IO_INTR_ENTRY) {
3751 			if (intrp->intr_irq == irqno &&
3752 			    intrp->intr_busid == busid &&
3753 			    intrp->intr_type == IO_INTR_INT)
3754 				return (intrp);
3755 			intrp++;
3756 		}
3757 	}
3758 	APIC_VERBOSE_IOAPIC((CE_NOTE, "Did not find io intr for irqno:"
3759 	    "busid %x:%x\n", irqno, busid));
3760 	return ((struct apic_io_intr *)NULL);
3761 }
3762 
3763 
3764 struct mps_bus_info {
3765 	char	*bus_name;
3766 	int	bus_id;
3767 } bus_info_array[] = {
3768 	"ISA ", BUS_ISA,
3769 	"PCI ", BUS_PCI,
3770 	"EISA ", BUS_EISA,
3771 	"XPRESS", BUS_XPRESS,
3772 	"PCMCIA", BUS_PCMCIA,
3773 	"VL ", BUS_VL,
3774 	"CBUS ", BUS_CBUS,
3775 	"CBUSII", BUS_CBUSII,
3776 	"FUTURE", BUS_FUTURE,
3777 	"INTERN", BUS_INTERN,
3778 	"MBI ", BUS_MBI,
3779 	"MBII ", BUS_MBII,
3780 	"MPI ", BUS_MPI,
3781 	"MPSA ", BUS_MPSA,
3782 	"NUBUS ", BUS_NUBUS,
3783 	"TC ", BUS_TC,
3784 	"VME ", BUS_VME
3785 };
3786 
3787 static int
3788 apic_find_bus_type(char *bus)
3789 {
3790 	int	i = 0;
3791 
3792 	for (; i < sizeof (bus_info_array)/sizeof (struct mps_bus_info); i++)
3793 		if (strncmp(bus, bus_info_array[i].bus_name,
3794 		    strlen(bus_info_array[i].bus_name)) == 0)
3795 			return (bus_info_array[i].bus_id);
3796 	APIC_VERBOSE_IOAPIC((CE_WARN, "Did not find bus type for bus %s", bus));
3797 	return (0);
3798 }
3799 
3800 static int
3801 apic_find_bus(int busid)
3802 {
3803 	struct	apic_bus	*busp;
3804 
3805 	busp = apic_busp;
3806 	while (busp->bus_entry == APIC_BUS_ENTRY) {
3807 		if (busp->bus_id == busid)
3808 			return (apic_find_bus_type((char *)&busp->bus_str1));
3809 		busp++;
3810 	}
3811 	APIC_VERBOSE_IOAPIC((CE_WARN, "Did not find bus for bus id %x", busid));
3812 	return (0);
3813 }
3814 
3815 static int
3816 apic_find_bus_id(int bustype)
3817 {
3818 	struct	apic_bus	*busp;
3819 
3820 	busp = apic_busp;
3821 	while (busp->bus_entry == APIC_BUS_ENTRY) {
3822 		if (apic_find_bus_type((char *)&busp->bus_str1) == bustype)
3823 			return (busp->bus_id);
3824 		busp++;
3825 	}
3826 	APIC_VERBOSE_IOAPIC((CE_WARN, "Did not find bus id for bustype %x",
3827 	    bustype));
3828 	return (-1);
3829 }
3830 
3831 /*
3832  * Check if a particular irq need to be reserved for any io_intr
3833  */
3834 static struct apic_io_intr *
3835 apic_find_io_intr(int irqno)
3836 {
3837 	struct	apic_io_intr	*intrp;
3838 
3839 	intrp = apic_io_intrp;
3840 	if (intrp != NULL) {
3841 		while (intrp->intr_entry == APIC_IO_INTR_ENTRY) {
3842 			if (intrp->intr_irq == irqno &&
3843 			    intrp->intr_type == IO_INTR_INT)
3844 				return (intrp);
3845 			intrp++;
3846 		}
3847 	}
3848 	return ((struct apic_io_intr *)NULL);
3849 }
3850 
3851 /*
3852  * Check if the given ioapicindex intin combination has already been assigned
3853  * an irq. If so return irqno. Else -1
3854  */
3855 static int
3856 apic_find_intin(uchar_t ioapic, uchar_t intin)
3857 {
3858 	apic_irq_t *irqptr;
3859 	int	i;
3860 
3861 	/* find ioapic and intin in the apic_irq_table[] and return the index */
3862 	for (i = apic_min_device_irq; i <= apic_max_device_irq; i++) {
3863 		irqptr = apic_irq_table[i];
3864 		while (irqptr) {
3865 			if ((irqptr->airq_mps_intr_index >= 0) &&
3866 			    (irqptr->airq_intin_no == intin) &&
3867 			    (irqptr->airq_ioapicindex == ioapic)) {
3868 				APIC_VERBOSE_IOAPIC((CE_NOTE, "!Found irq "
3869 				    "entry for ioapic:intin %x:%x "
3870 				    "shared interrupts ?", ioapic, intin));
3871 				return (i);
3872 			}
3873 			irqptr = irqptr->airq_next;
3874 		}
3875 	}
3876 	return (-1);
3877 }
3878 
3879 int
3880 apic_allocate_irq(int irq)
3881 {
3882 	int	freeirq, i;
3883 
3884 	if ((freeirq = apic_find_free_irq(irq, (APIC_RESV_IRQ - 1))) == -1)
3885 		if ((freeirq = apic_find_free_irq(APIC_FIRST_FREE_IRQ,
3886 		    (irq - 1))) == -1) {
3887 			/*
3888 			 * if BIOS really defines every single irq in the mps
3889 			 * table, then don't worry about conflicting with
3890 			 * them, just use any free slot in apic_irq_table
3891 			 */
3892 			for (i = APIC_FIRST_FREE_IRQ; i < APIC_RESV_IRQ; i++) {
3893 				if ((apic_irq_table[i] == NULL) ||
3894 				    apic_irq_table[i]->airq_mps_intr_index ==
3895 				    FREE_INDEX) {
3896 				freeirq = i;
3897 				break;
3898 			}
3899 		}
3900 		if (freeirq == -1) {
3901 			/* This shouldn't happen, but just in case */
3902 			cmn_err(CE_WARN, "pcplusmp: NO available IRQ");
3903 			return (-1);
3904 		}
3905 	}
3906 	if (apic_irq_table[freeirq] == NULL) {
3907 		apic_irq_table[freeirq] =
3908 		    kmem_zalloc(sizeof (apic_irq_t), KM_NOSLEEP);
3909 		if (apic_irq_table[freeirq] == NULL) {
3910 			cmn_err(CE_WARN, "pcplusmp: NO memory to allocate IRQ");
3911 			return (-1);
3912 		}
3913 		apic_irq_table[freeirq]->airq_mps_intr_index = FREE_INDEX;
3914 	}
3915 	return (freeirq);
3916 }
3917 
3918 static int
3919 apic_find_free_irq(int start, int end)
3920 {
3921 	int	i;
3922 
3923 	for (i = start; i <= end; i++)
3924 		/* Check if any I/O entry needs this IRQ */
3925 		if (apic_find_io_intr(i) == NULL) {
3926 			/* Then see if it is free */
3927 			if ((apic_irq_table[i] == NULL) ||
3928 			    (apic_irq_table[i]->airq_mps_intr_index ==
3929 			    FREE_INDEX)) {
3930 				return (i);
3931 			}
3932 		}
3933 	return (-1);
3934 }
3935 
3936 /*
3937  * Allocate a free vector for irq at ipl. Takes care of merging of multiple
3938  * IPLs into a single APIC level as well as stretching some IPLs onto multiple
3939  * levels. APIC_HI_PRI_VECTS interrupts are reserved for high priority
3940  * requests and allocated only when pri is set.
3941  */
3942 static uchar_t
3943 apic_allocate_vector(int ipl, int irq, int pri)
3944 {
3945 	int	lowest, highest, i;
3946 
3947 	highest = apic_ipltopri[ipl] + APIC_VECTOR_MASK;
3948 	lowest = apic_ipltopri[ipl - 1] + APIC_VECTOR_PER_IPL;
3949 
3950 	if (highest < lowest) /* Both ipl and ipl - 1 map to same pri */
3951 		lowest -= APIC_VECTOR_PER_IPL;
3952 
3953 #ifdef	DEBUG
3954 	if (apic_restrict_vector)	/* for testing shared interrupt logic */
3955 		highest = lowest + apic_restrict_vector + APIC_HI_PRI_VECTS;
3956 #endif /* DEBUG */
3957 	if (pri == 0)
3958 		highest -= APIC_HI_PRI_VECTS;
3959 
3960 	for (i = lowest; i < highest; i++) {
3961 		if ((i == T_FASTTRAP) || (i == APIC_SPUR_INTR) ||
3962 			(i == T_SYSCALLINT) || (i == T_DTRACE_PROBE) ||
3963 			(i == T_DTRACE_RET))
3964 			continue;
3965 		if (apic_vector_to_irq[i] == APIC_RESV_IRQ) {
3966 			apic_vector_to_irq[i] = (uchar_t)irq;
3967 			return (i);
3968 		}
3969 	}
3970 
3971 	return (0);
3972 }
3973 
3974 static void
3975 apic_modify_vector(uchar_t vector, int irq)
3976 {
3977 	apic_vector_to_irq[vector] = (uchar_t)irq;
3978 }
3979 
3980 /*
3981  * Mark vector as being in the process of being deleted. Interrupts
3982  * may still come in on some CPU. The moment an interrupt comes with
3983  * the new vector, we know we can free the old one. Called only from
3984  * addspl and delspl with interrupts disabled. Because an interrupt
3985  * can be shared, but no interrupt from either device may come in,
3986  * we also use a timeout mechanism, which we arbitrarily set to
3987  * apic_revector_timeout microseconds.
3988  */
3989 static void
3990 apic_mark_vector(uchar_t oldvector, uchar_t newvector)
3991 {
3992 	int iflag = intr_clear();
3993 	lock_set(&apic_revector_lock);
3994 	if (!apic_oldvec_to_newvec) {
3995 		apic_oldvec_to_newvec =
3996 		    kmem_zalloc(sizeof (newvector) * APIC_MAX_VECTOR * 2,
3997 		    KM_NOSLEEP);
3998 
3999 		if (!apic_oldvec_to_newvec) {
4000 			/*
4001 			 * This failure is not catastrophic.
4002 			 * But, the oldvec will never be freed.
4003 			 */
4004 			apic_error |= APIC_ERR_MARK_VECTOR_FAIL;
4005 			lock_clear(&apic_revector_lock);
4006 			intr_restore(iflag);
4007 			return;
4008 		}
4009 		apic_newvec_to_oldvec = &apic_oldvec_to_newvec[APIC_MAX_VECTOR];
4010 	}
4011 
4012 	/* See if we already did this for drivers which do double addintrs */
4013 	if (apic_oldvec_to_newvec[oldvector] != newvector) {
4014 		apic_oldvec_to_newvec[oldvector] = newvector;
4015 		apic_newvec_to_oldvec[newvector] = oldvector;
4016 		apic_revector_pending++;
4017 	}
4018 	lock_clear(&apic_revector_lock);
4019 	intr_restore(iflag);
4020 	(void) timeout(apic_xlate_vector_free_timeout_handler,
4021 	    (void *)(uintptr_t)oldvector, drv_usectohz(apic_revector_timeout));
4022 }
4023 
4024 /*
4025  * xlate_vector is called from intr_enter if revector_pending is set.
4026  * It will xlate it if needed and mark the old vector as free.
4027  */
4028 static uchar_t
4029 apic_xlate_vector(uchar_t vector)
4030 {
4031 	uchar_t	newvector, oldvector = 0;
4032 
4033 	lock_set(&apic_revector_lock);
4034 	/* Do we really need to do this ? */
4035 	if (!apic_revector_pending) {
4036 		lock_clear(&apic_revector_lock);
4037 		return (vector);
4038 	}
4039 	if ((newvector = apic_oldvec_to_newvec[vector]) != 0)
4040 		oldvector = vector;
4041 	else {
4042 		/*
4043 		 * The incoming vector is new . See if a stale entry is
4044 		 * remaining
4045 		 */
4046 		if ((oldvector = apic_newvec_to_oldvec[vector]) != 0)
4047 			newvector = vector;
4048 	}
4049 
4050 	if (oldvector) {
4051 		apic_revector_pending--;
4052 		apic_oldvec_to_newvec[oldvector] = 0;
4053 		apic_newvec_to_oldvec[newvector] = 0;
4054 		apic_free_vector(oldvector);
4055 		lock_clear(&apic_revector_lock);
4056 		/* There could have been more than one reprogramming! */
4057 		return (apic_xlate_vector(newvector));
4058 	}
4059 	lock_clear(&apic_revector_lock);
4060 	return (vector);
4061 }
4062 
4063 void
4064 apic_xlate_vector_free_timeout_handler(void *arg)
4065 {
4066 	int iflag;
4067 	uchar_t oldvector, newvector;
4068 
4069 	oldvector = (uchar_t)(uintptr_t)arg;
4070 	iflag = intr_clear();
4071 	lock_set(&apic_revector_lock);
4072 	if ((newvector = apic_oldvec_to_newvec[oldvector]) != 0) {
4073 		apic_free_vector(oldvector);
4074 		apic_oldvec_to_newvec[oldvector] = 0;
4075 		apic_newvec_to_oldvec[newvector] = 0;
4076 		apic_revector_pending--;
4077 	}
4078 
4079 	lock_clear(&apic_revector_lock);
4080 	intr_restore(iflag);
4081 }
4082 
4083 
4084 /* Mark vector as not being used by any irq */
4085 static void
4086 apic_free_vector(uchar_t vector)
4087 {
4088 	apic_vector_to_irq[vector] = APIC_RESV_IRQ;
4089 }
4090 
4091 /*
4092  * compute the polarity, trigger mode and vector for programming into
4093  * the I/O apic and record in airq_rdt_entry.
4094  */
4095 static void
4096 apic_record_rdt_entry(apic_irq_t *irqptr, int irq)
4097 {
4098 	int	ioapicindex, bus_type, vector;
4099 	short	intr_index;
4100 	uint_t	level, po, io_po;
4101 	struct apic_io_intr *iointrp;
4102 
4103 	intr_index = irqptr->airq_mps_intr_index;
4104 	DDI_INTR_IMPLDBG((CE_CONT, "apic_record_rdt_entry: intr_index=%d "
4105 	    "irq = 0x%x dip = 0x%p vector = 0x%x\n", intr_index, irq,
4106 	    (void *)irqptr->airq_dip, irqptr->airq_vector));
4107 
4108 	if (intr_index == RESERVE_INDEX) {
4109 		apic_error |= APIC_ERR_INVALID_INDEX;
4110 		return;
4111 	} else if (APIC_IS_MSI_OR_MSIX_INDEX(intr_index)) {
4112 		return;
4113 	}
4114 
4115 	vector = irqptr->airq_vector;
4116 	ioapicindex = irqptr->airq_ioapicindex;
4117 	/* Assume edge triggered by default */
4118 	level = 0;
4119 	/* Assume active high by default */
4120 	po = 0;
4121 
4122 	if (intr_index == DEFAULT_INDEX || intr_index == FREE_INDEX) {
4123 		ASSERT(irq < 16);
4124 		if (eisa_level_intr_mask & (1 << irq))
4125 			level = AV_LEVEL;
4126 		if (intr_index == FREE_INDEX && apic_defconf == 0)
4127 			apic_error |= APIC_ERR_INVALID_INDEX;
4128 	} else if (intr_index == ACPI_INDEX) {
4129 		bus_type = irqptr->airq_iflag.bustype;
4130 		if (irqptr->airq_iflag.intr_el == INTR_EL_CONFORM) {
4131 			if (bus_type == BUS_PCI)
4132 				level = AV_LEVEL;
4133 		} else
4134 			level = (irqptr->airq_iflag.intr_el == INTR_EL_LEVEL) ?
4135 			    AV_LEVEL : 0;
4136 		if (level &&
4137 		    ((irqptr->airq_iflag.intr_po == INTR_PO_ACTIVE_LOW) ||
4138 		    (irqptr->airq_iflag.intr_po == INTR_PO_CONFORM &&
4139 		    bus_type == BUS_PCI)))
4140 			po = AV_ACTIVE_LOW;
4141 	} else {
4142 		iointrp = apic_io_intrp + intr_index;
4143 		bus_type = apic_find_bus(iointrp->intr_busid);
4144 		if (iointrp->intr_el == INTR_EL_CONFORM) {
4145 			if ((irq < 16) && (eisa_level_intr_mask & (1 << irq)))
4146 				level = AV_LEVEL;
4147 			else if (bus_type == BUS_PCI)
4148 				level = AV_LEVEL;
4149 		} else
4150 			level = (iointrp->intr_el == INTR_EL_LEVEL) ?
4151 			    AV_LEVEL : 0;
4152 		if (level && ((iointrp->intr_po == INTR_PO_ACTIVE_LOW) ||
4153 		    (iointrp->intr_po == INTR_PO_CONFORM &&
4154 		    bus_type == BUS_PCI)))
4155 			po = AV_ACTIVE_LOW;
4156 	}
4157 	if (level)
4158 		apic_level_intr[irq] = 1;
4159 	/*
4160 	 * The 82489DX External APIC cannot do active low polarity interrupts.
4161 	 */
4162 	if (po && (apic_io_ver[ioapicindex] != IOAPIC_VER_82489DX))
4163 		io_po = po;
4164 	else
4165 		io_po = 0;
4166 
4167 	if (apic_verbose & APIC_VERBOSE_IOAPIC_FLAG)
4168 		printf("setio: ioapic=%x intin=%x level=%x po=%x vector=%x\n",
4169 		    ioapicindex, irqptr->airq_intin_no, level, io_po, vector);
4170 
4171 	irqptr->airq_rdt_entry = level|io_po|vector;
4172 }
4173 
4174 /*
4175  * Call rebind to do the actual programming.
4176  */
4177 static int
4178 apic_setup_io_intr(apic_irq_t *irqptr, int irq)
4179 {
4180 	int rv;
4181 
4182 	if (rv = apic_rebind(irqptr, apic_irq_table[irq]->airq_cpu, 1,
4183 	    IMMEDIATE))
4184 		/* CPU is not up or interrupt is disabled. Fall back to 0 */
4185 		rv = apic_rebind(irqptr, 0, 1, IMMEDIATE);
4186 
4187 	return (rv);
4188 }
4189 
4190 /*
4191  * Deferred reprogramming: Call apic_rebind to do the real work.
4192  */
4193 static int
4194 apic_setup_io_intr_deferred(apic_irq_t *irqptr, int irq)
4195 {
4196 	int rv;
4197 
4198 	if (rv = apic_rebind(irqptr, apic_irq_table[irq]->airq_cpu, 1,
4199 	    DEFERRED))
4200 		/* CPU is not up or interrupt is disabled. Fall back to 0 */
4201 		rv = apic_rebind(irqptr, 0, 1, DEFERRED);
4202 
4203 	return (rv);
4204 }
4205 
4206 /*
4207  * Bind interrupt corresponding to irq_ptr to bind_cpu. acquire_lock
4208  * if false (0) means lock is already held (e.g: in rebind_all).
4209  */
4210 static int
4211 apic_rebind(apic_irq_t *irq_ptr, int bind_cpu, int acquire_lock, int when)
4212 {
4213 	int			intin_no;
4214 	volatile int32_t	*ioapic;
4215 	uchar_t			airq_temp_cpu;
4216 	apic_cpus_info_t	*cpu_infop;
4217 	int			iflag;
4218 	int		which_irq = apic_vector_to_irq[irq_ptr->airq_vector];
4219 
4220 	intin_no = irq_ptr->airq_intin_no;
4221 	ioapic = apicioadr[irq_ptr->airq_ioapicindex];
4222 	airq_temp_cpu = irq_ptr->airq_temp_cpu;
4223 	if (airq_temp_cpu != IRQ_UNINIT && airq_temp_cpu != IRQ_UNBOUND) {
4224 		if (airq_temp_cpu & IRQ_USER_BOUND)
4225 			/* Mask off high bit so it can be used as array index */
4226 			airq_temp_cpu &= ~IRQ_USER_BOUND;
4227 
4228 		ASSERT(airq_temp_cpu < apic_nproc);
4229 	}
4230 
4231 	iflag = intr_clear();
4232 
4233 	if (acquire_lock)
4234 		lock_set(&apic_ioapic_lock);
4235 
4236 	/*
4237 	 * Can't bind to a CPU that's not online:
4238 	 */
4239 	cpu_infop = &apic_cpus[bind_cpu & ~IRQ_USER_BOUND];
4240 	if (!(cpu_infop->aci_status & APIC_CPU_INTR_ENABLE)) {
4241 
4242 		if (acquire_lock)
4243 			lock_clear(&apic_ioapic_lock);
4244 
4245 		intr_restore(iflag);
4246 		return (1);
4247 	}
4248 
4249 	/*
4250 	 * If this is a deferred reprogramming attempt, ensure we have
4251 	 * not been passed stale data:
4252 	 */
4253 	if ((when == DEFERRED) &&
4254 	    (apic_reprogram_info[which_irq].valid == 0)) {
4255 		/* stale info, so just return */
4256 		if (acquire_lock)
4257 			lock_clear(&apic_ioapic_lock);
4258 
4259 		intr_restore(iflag);
4260 		return (0);
4261 	}
4262 
4263 	/*
4264 	 * If this interrupt has been delivered to a CPU and that CPU
4265 	 * has not handled it yet, we cannot reprogram the IOAPIC now:
4266 	 */
4267 	if (!APIC_IS_MSI_OR_MSIX_INDEX(irq_ptr->airq_mps_intr_index) &&
4268 	    apic_check_stuck_interrupt(irq_ptr, airq_temp_cpu, bind_cpu,
4269 	    ioapic, intin_no, which_irq) != 0) {
4270 
4271 		if (acquire_lock)
4272 			lock_clear(&apic_ioapic_lock);
4273 
4274 		intr_restore(iflag);
4275 		return (0);
4276 	}
4277 
4278 	/*
4279 	 * NOTE: We do not unmask the RDT here, as an interrupt MAY still
4280 	 * come in before we have a chance to reprogram it below.  The
4281 	 * reprogramming below will simultaneously change and unmask the
4282 	 * RDT entry.
4283 	 */
4284 
4285 	if ((uchar_t)bind_cpu == IRQ_UNBOUND) {
4286 		/* Write the RDT entry -- no specific CPU binding */
4287 		WRITE_IOAPIC_RDT_ENTRY_HIGH_DWORD(ioapic, intin_no, AV_TOALL);
4288 
4289 		if (airq_temp_cpu != IRQ_UNINIT && airq_temp_cpu != IRQ_UNBOUND)
4290 			apic_cpus[airq_temp_cpu].aci_temp_bound--;
4291 
4292 		/* Write the vector, trigger, and polarity portion of the RDT */
4293 		WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no,
4294 		    AV_LDEST | AV_LOPRI | irq_ptr->airq_rdt_entry);
4295 		if (acquire_lock)
4296 			lock_clear(&apic_ioapic_lock);
4297 		irq_ptr->airq_temp_cpu = IRQ_UNBOUND;
4298 		intr_restore(iflag);
4299 		return (0);
4300 	}
4301 
4302 	if (bind_cpu & IRQ_USER_BOUND) {
4303 		cpu_infop->aci_bound++;
4304 	} else {
4305 		cpu_infop->aci_temp_bound++;
4306 	}
4307 	ASSERT((bind_cpu & ~IRQ_USER_BOUND) < apic_nproc);
4308 	if (!APIC_IS_MSI_OR_MSIX_INDEX(irq_ptr->airq_mps_intr_index)) {
4309 		/* Write the RDT entry -- bind to a specific CPU: */
4310 		WRITE_IOAPIC_RDT_ENTRY_HIGH_DWORD(ioapic, intin_no,
4311 		    cpu_infop->aci_local_id << APIC_ID_BIT_OFFSET);
4312 	}
4313 	if ((airq_temp_cpu != IRQ_UNBOUND) && (airq_temp_cpu != IRQ_UNINIT)) {
4314 		apic_cpus[airq_temp_cpu].aci_temp_bound--;
4315 	}
4316 	if (!APIC_IS_MSI_OR_MSIX_INDEX(irq_ptr->airq_mps_intr_index)) {
4317 		/* Write the vector, trigger, and polarity portion of the RDT */
4318 		WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no,
4319 		    AV_PDEST | AV_FIXED | irq_ptr->airq_rdt_entry);
4320 	} else {
4321 		if (irq_ptr->airq_ioapicindex == irq_ptr->airq_origirq) {
4322 			/* first one */
4323 			DDI_INTR_IMPLDBG((CE_CONT, "apic_rebind: call "
4324 			    "apic_pci_msi_enable_vector\n"));
4325 			if (apic_pci_msi_enable_vector(irq_ptr->airq_dip,
4326 			    (irq_ptr->airq_mps_intr_index == MSI_INDEX) ?
4327 			    DDI_INTR_TYPE_MSI : DDI_INTR_TYPE_MSIX, which_irq,
4328 			    irq_ptr->airq_vector, irq_ptr->airq_intin_no,
4329 			    cpu_infop->aci_local_id) != PSM_SUCCESS) {
4330 				cmn_err(CE_WARN, "pcplusmp: "
4331 					"apic_pci_msi_enable_vector "
4332 					"returned PSM_FAILURE");
4333 			}
4334 		}
4335 		if ((irq_ptr->airq_ioapicindex + irq_ptr->airq_intin_no - 1) ==
4336 		    irq_ptr->airq_origirq) { /* last one */
4337 			DDI_INTR_IMPLDBG((CE_CONT, "apic_rebind: call "
4338 			    "pci_msi_enable_mode\n"));
4339 			if (pci_msi_enable_mode(irq_ptr->airq_dip,
4340 			    (irq_ptr->airq_mps_intr_index == MSI_INDEX) ?
4341 			    DDI_INTR_TYPE_MSI : DDI_INTR_TYPE_MSIX,
4342 			    which_irq) != DDI_SUCCESS) {
4343 				DDI_INTR_IMPLDBG((CE_CONT, "pcplusmp: "
4344 				    "pci_msi_enable failed\n"));
4345 				(void) pci_msi_unconfigure(irq_ptr->airq_dip,
4346 				(irq_ptr->airq_mps_intr_index == MSI_INDEX) ?
4347 				DDI_INTR_TYPE_MSI : DDI_INTR_TYPE_MSIX,
4348 				which_irq);
4349 			}
4350 		}
4351 	}
4352 	if (acquire_lock)
4353 		lock_clear(&apic_ioapic_lock);
4354 	irq_ptr->airq_temp_cpu = (uchar_t)bind_cpu;
4355 	apic_redist_cpu_skip &= ~(1 << (bind_cpu & ~IRQ_USER_BOUND));
4356 	intr_restore(iflag);
4357 	return (0);
4358 }
4359 
4360 /*
4361  * Checks to see if the IOAPIC interrupt entry specified has its Remote IRR
4362  * bit set.  Sets up a timeout to perform the reprogramming at a later time
4363  * if it cannot wait for the Remote IRR bit to clear (or if waiting did not
4364  * result in the bit's clearing).
4365  *
4366  * This function will mask the RDT entry if the Remote IRR bit is set.
4367  *
4368  * Returns non-zero if the caller should defer IOAPIC reprogramming.
4369  */
4370 static int
4371 apic_check_stuck_interrupt(apic_irq_t *irq_ptr, int old_bind_cpu,
4372 	int new_bind_cpu, volatile int32_t *ioapic, int intin_no, int which_irq)
4373 {
4374 	int32_t			rdt_entry;
4375 	int			waited;
4376 
4377 	/* Mask the RDT entry, but only if it's a level-triggered interrupt */
4378 	rdt_entry = READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no);
4379 	if ((rdt_entry & (AV_LEVEL|AV_MASK)) == AV_LEVEL) {
4380 
4381 		/* Mask it */
4382 		WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no,
4383 		    AV_MASK | rdt_entry);
4384 	}
4385 
4386 	/*
4387 	 * Wait for the delivery pending bit to clear.
4388 	 */
4389 	if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no) &
4390 	    (AV_LEVEL|AV_PENDING)) == (AV_LEVEL|AV_PENDING)) {
4391 
4392 		/*
4393 		 * If we're still waiting on the delivery of this interrupt,
4394 		 * continue to wait here until it is delivered (this should be
4395 		 * a very small amount of time, but include a timeout just in
4396 		 * case).
4397 		 */
4398 		for (waited = 0; waited < apic_max_usecs_clear_pending;
4399 		    waited += APIC_USECS_PER_WAIT_INTERVAL) {
4400 			if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no)
4401 			    & AV_PENDING) == 0) {
4402 				break;
4403 			}
4404 			drv_usecwait(APIC_USECS_PER_WAIT_INTERVAL);
4405 		}
4406 
4407 		if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no) &
4408 		    AV_PENDING) != 0) {
4409 			cmn_err(CE_WARN, "!IOAPIC %d intin %d: Could not "
4410 			    "deliver interrupt to local APIC within "
4411 			    "%d usecs.", irq_ptr->airq_ioapicindex,
4412 			    irq_ptr->airq_intin_no,
4413 			    apic_max_usecs_clear_pending);
4414 		}
4415 	}
4416 
4417 	/*
4418 	 * If the remote IRR bit is set, then the interrupt has been sent
4419 	 * to a CPU for processing.  We have no choice but to wait for
4420 	 * that CPU to process the interrupt, at which point the remote IRR
4421 	 * bit will be cleared.
4422 	 */
4423 	if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no) &
4424 	    (AV_LEVEL|AV_REMOTE_IRR)) == (AV_LEVEL|AV_REMOTE_IRR)) {
4425 
4426 		/*
4427 		 * If the CPU that this RDT is bound to is NOT the current
4428 		 * CPU, wait until that CPU handles the interrupt and ACKs
4429 		 * it.  If this interrupt is not bound to any CPU (that is,
4430 		 * if it's bound to the logical destination of "anyone"), it
4431 		 * may have been delivered to the current CPU so handle that
4432 		 * case by deferring the reprogramming (below).
4433 		 */
4434 		kpreempt_disable();
4435 		if ((old_bind_cpu != IRQ_UNBOUND) &&
4436 		    (old_bind_cpu != IRQ_UNINIT) &&
4437 		    (old_bind_cpu != psm_get_cpu_id())) {
4438 			for (waited = 0; waited < apic_max_usecs_clear_pending;
4439 			    waited += APIC_USECS_PER_WAIT_INTERVAL) {
4440 				if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic,
4441 				    intin_no) & AV_REMOTE_IRR) == 0) {
4442 
4443 					/* Clear the reprogramming state: */
4444 					lock_set(&apic_ioapic_reprogram_lock);
4445 
4446 					apic_reprogram_info[which_irq].valid
4447 					    = 0;
4448 					apic_reprogram_info[which_irq].bindcpu
4449 					    = 0;
4450 					apic_reprogram_info[which_irq].timeouts
4451 					    = 0;
4452 
4453 					lock_clear(&apic_ioapic_reprogram_lock);
4454 
4455 					/* Remote IRR has cleared! */
4456 					kpreempt_enable();
4457 					return (0);
4458 				}
4459 				drv_usecwait(APIC_USECS_PER_WAIT_INTERVAL);
4460 			}
4461 		}
4462 		kpreempt_enable();
4463 
4464 		/*
4465 		 * If we waited and the Remote IRR bit is still not cleared,
4466 		 * AND if we've invoked the timeout APIC_REPROGRAM_MAX_TIMEOUTS
4467 		 * times for this interrupt, try the last-ditch workarounds:
4468 		 */
4469 		if (apic_reprogram_info[which_irq].timeouts >=
4470 		    APIC_REPROGRAM_MAX_TIMEOUTS) {
4471 
4472 			if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic, intin_no)
4473 			    & AV_REMOTE_IRR) != 0) {
4474 				/*
4475 				 * Trying to clear the bit through normal
4476 				 * channels has failed.  So as a last-ditch
4477 				 * effort, try to set the trigger mode to
4478 				 * edge, then to level.  This has been
4479 				 * observed to work on many systems.
4480 				 */
4481 				WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic,
4482 				    intin_no,
4483 				    READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic,
4484 				    intin_no) & ~AV_LEVEL);
4485 
4486 				WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic,
4487 				    intin_no,
4488 				    READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic,
4489 				    intin_no) | AV_LEVEL);
4490 
4491 				/*
4492 				 * If the bit's STILL set, declare total and
4493 				 * utter failure
4494 				 */
4495 				if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic,
4496 				    intin_no) & AV_REMOTE_IRR) != 0) {
4497 					cmn_err(CE_WARN, "!IOAPIC %d intin %d: "
4498 					    "Remote IRR failed to reset "
4499 					    "within %d usecs.  Interrupts to "
4500 					    "this pin may cease to function.",
4501 					    irq_ptr->airq_ioapicindex,
4502 					    irq_ptr->airq_intin_no,
4503 					    apic_max_usecs_clear_pending);
4504 				}
4505 			}
4506 			/* Clear the reprogramming state: */
4507 			lock_set(&apic_ioapic_reprogram_lock);
4508 
4509 			apic_reprogram_info[which_irq].valid = 0;
4510 			apic_reprogram_info[which_irq].bindcpu = 0;
4511 			apic_reprogram_info[which_irq].timeouts = 0;
4512 
4513 			lock_clear(&apic_ioapic_reprogram_lock);
4514 		} else {
4515 #ifdef DEBUG
4516 			cmn_err(CE_WARN, "Deferring reprogramming of irq %d",
4517 			    which_irq);
4518 #endif	/* DEBUG */
4519 			/*
4520 			 * If waiting for the Remote IRR bit (above) didn't
4521 			 * allow it to clear, defer the reprogramming:
4522 			 */
4523 			lock_set(&apic_ioapic_reprogram_lock);
4524 
4525 			apic_reprogram_info[which_irq].valid = 1;
4526 			apic_reprogram_info[which_irq].bindcpu = new_bind_cpu;
4527 			apic_reprogram_info[which_irq].timeouts++;
4528 
4529 			lock_clear(&apic_ioapic_reprogram_lock);
4530 
4531 			/* Fire up a timeout to handle this later */
4532 			(void) timeout(apic_reprogram_timeout_handler,
4533 			    (void *) 0,
4534 			    drv_usectohz(APIC_REPROGRAM_TIMEOUT_DELAY));
4535 
4536 			/* Inform caller to defer IOAPIC programming: */
4537 			return (1);
4538 		}
4539 	}
4540 	return (0);
4541 }
4542 
4543 /*
4544  * Timeout handler that performs the APIC reprogramming
4545  */
4546 /*ARGSUSED*/
4547 static void
4548 apic_reprogram_timeout_handler(void *arg)
4549 {
4550 	/*LINTED: set but not used in function*/
4551 	int i, result;
4552 
4553 	/* Serialize access to this function */
4554 	mutex_enter(&apic_reprogram_timeout_mutex);
4555 
4556 	/*
4557 	 * For each entry in the reprogramming state that's valid,
4558 	 * try the reprogramming again:
4559 	 */
4560 	for (i = 0; i < APIC_MAX_VECTOR; i++) {
4561 		if (apic_reprogram_info[i].valid == 0)
4562 			continue;
4563 		/*
4564 		 * Though we can't really do anything about errors
4565 		 * at this point, keep track of them for reporting.
4566 		 * Note that it is very possible for apic_setup_io_intr
4567 		 * to re-register this very timeout if the Remote IRR bit
4568 		 * has not yet cleared.
4569 		 */
4570 		result = apic_setup_io_intr_deferred(apic_irq_table[i], i);
4571 
4572 #ifdef DEBUG
4573 		if (result)
4574 			cmn_err(CE_WARN, "apic_reprogram_timeout: "
4575 			    "apic_setup_io_intr returned nonzero for "
4576 			    "irq=%d!", i);
4577 #endif	/* DEBUG */
4578 	}
4579 
4580 	mutex_exit(&apic_reprogram_timeout_mutex);
4581 }
4582 
4583 
4584 /*
4585  * Called to migrate all interrupts at an irq to another cpu. safe
4586  * if true means we are not being called from an interrupt
4587  * context and hence it is safe to do a lock_set. If false
4588  * do only a lock_try and return failure ( non 0 ) if we cannot get it
4589  */
4590 static int
4591 apic_rebind_all(apic_irq_t *irq_ptr, int bind_cpu, int safe)
4592 {
4593 	apic_irq_t	*irqptr = irq_ptr;
4594 	int		retval = 0;
4595 	int		iflag;
4596 
4597 	iflag = intr_clear();
4598 	if (!safe) {
4599 		if (lock_try(&apic_ioapic_lock) == 0) {
4600 			intr_restore(iflag);
4601 			return (1);
4602 		}
4603 	} else
4604 		lock_set(&apic_ioapic_lock);
4605 
4606 	while (irqptr) {
4607 		if (irqptr->airq_temp_cpu != IRQ_UNINIT)
4608 			retval |= apic_rebind(irqptr, bind_cpu, 0, IMMEDIATE);
4609 		irqptr = irqptr->airq_next;
4610 	}
4611 	lock_clear(&apic_ioapic_lock);
4612 	intr_restore(iflag);
4613 	return (retval);
4614 }
4615 
4616 /*
4617  * apic_intr_redistribute does all the messy computations for identifying
4618  * which interrupt to move to which CPU. Currently we do just one interrupt
4619  * at a time. This reduces the time we spent doing all this within clock
4620  * interrupt. When it is done in idle, we could do more than 1.
4621  * First we find the most busy and the most free CPU (time in ISR only)
4622  * skipping those CPUs that has been identified as being ineligible (cpu_skip)
4623  * Then we look for IRQs which are closest to the difference between the
4624  * most busy CPU and the average ISR load. We try to find one whose load
4625  * is less than difference.If none exists, then we chose one larger than the
4626  * difference, provided it does not make the most idle CPU worse than the
4627  * most busy one. In the end, we clear all the busy fields for CPUs. For
4628  * IRQs, they are cleared as they are scanned.
4629  */
4630 static void
4631 apic_intr_redistribute()
4632 {
4633 	int busiest_cpu, most_free_cpu;
4634 	int cpu_free, cpu_busy, max_busy, min_busy;
4635 	int min_free, diff;
4636 	int	average_busy, cpus_online;
4637 	int i, busy;
4638 	apic_cpus_info_t *cpu_infop;
4639 	apic_irq_t *min_busy_irq = NULL;
4640 	apic_irq_t *max_busy_irq = NULL;
4641 
4642 	busiest_cpu = most_free_cpu = -1;
4643 	cpu_free = cpu_busy = max_busy = average_busy = 0;
4644 	min_free = apic_sample_factor_redistribution;
4645 	cpus_online = 0;
4646 	/*
4647 	 * Below we will check for CPU_INTR_ENABLE, bound, temp_bound, temp_cpu
4648 	 * without ioapic_lock. That is OK as we are just doing statistical
4649 	 * sampling anyway and any inaccuracy now will get corrected next time
4650 	 * The call to rebind which actually changes things will make sure
4651 	 * we are consistent.
4652 	 */
4653 	for (i = 0; i < apic_nproc; i++) {
4654 		if (!(apic_redist_cpu_skip & (1 << i)) &&
4655 		    (apic_cpus[i].aci_status & APIC_CPU_INTR_ENABLE)) {
4656 
4657 			cpu_infop = &apic_cpus[i];
4658 			/*
4659 			 * If no unbound interrupts or only 1 total on this
4660 			 * CPU, skip
4661 			 */
4662 			if (!cpu_infop->aci_temp_bound ||
4663 			    (cpu_infop->aci_bound + cpu_infop->aci_temp_bound)
4664 			    == 1) {
4665 				apic_redist_cpu_skip |= 1 << i;
4666 				continue;
4667 			}
4668 
4669 			busy = cpu_infop->aci_busy;
4670 			average_busy += busy;
4671 			cpus_online++;
4672 			if (max_busy < busy) {
4673 				max_busy = busy;
4674 				busiest_cpu = i;
4675 			}
4676 			if (min_free > busy) {
4677 				min_free = busy;
4678 				most_free_cpu = i;
4679 			}
4680 			if (busy > apic_int_busy_mark) {
4681 				cpu_busy |= 1 << i;
4682 			} else {
4683 				if (busy < apic_int_free_mark)
4684 					cpu_free |= 1 << i;
4685 			}
4686 		}
4687 	}
4688 	if ((cpu_busy && cpu_free) ||
4689 	    (max_busy >= (min_free + apic_diff_for_redistribution))) {
4690 
4691 		apic_num_imbalance++;
4692 #ifdef	DEBUG
4693 		if (apic_verbose & APIC_VERBOSE_IOAPIC_FLAG) {
4694 			prom_printf(
4695 			    "redistribute busy=%x free=%x max=%x min=%x",
4696 			    cpu_busy, cpu_free, max_busy, min_free);
4697 		}
4698 #endif /* DEBUG */
4699 
4700 
4701 		average_busy /= cpus_online;
4702 
4703 		diff = max_busy - average_busy;
4704 		min_busy = max_busy; /* start with the max possible value */
4705 		max_busy = 0;
4706 		min_busy_irq = max_busy_irq = NULL;
4707 		i = apic_min_device_irq;
4708 		for (; i < apic_max_device_irq; i++) {
4709 			apic_irq_t *irq_ptr;
4710 			/* Change to linked list per CPU ? */
4711 			if ((irq_ptr = apic_irq_table[i]) == NULL)
4712 				continue;
4713 			/* Check for irq_busy & decide which one to move */
4714 			/* Also zero them for next round */
4715 			if ((irq_ptr->airq_temp_cpu == busiest_cpu) &&
4716 			    irq_ptr->airq_busy) {
4717 				if (irq_ptr->airq_busy < diff) {
4718 					/*
4719 					 * Check for least busy CPU,
4720 					 * best fit or what ?
4721 					 */
4722 					if (max_busy < irq_ptr->airq_busy) {
4723 						/*
4724 						 * Most busy within the
4725 						 * required differential
4726 						 */
4727 						max_busy = irq_ptr->airq_busy;
4728 						max_busy_irq = irq_ptr;
4729 					}
4730 				} else {
4731 					if (min_busy > irq_ptr->airq_busy) {
4732 						/*
4733 						 * least busy, but more than
4734 						 * the reqd diff
4735 						 */
4736 						if (min_busy <
4737 						    (diff + average_busy -
4738 						    min_free)) {
4739 							/*
4740 							 * Making sure new cpu
4741 							 * will not end up
4742 							 * worse
4743 							 */
4744 							min_busy =
4745 							    irq_ptr->airq_busy;
4746 
4747 							min_busy_irq = irq_ptr;
4748 						}
4749 					}
4750 				}
4751 			}
4752 			irq_ptr->airq_busy = 0;
4753 		}
4754 
4755 		if (max_busy_irq != NULL) {
4756 #ifdef	DEBUG
4757 			if (apic_verbose & APIC_VERBOSE_IOAPIC_FLAG) {
4758 				prom_printf("rebinding %x to %x",
4759 				    max_busy_irq->airq_vector, most_free_cpu);
4760 			}
4761 #endif /* DEBUG */
4762 			if (apic_rebind_all(max_busy_irq, most_free_cpu, 0)
4763 			    == 0)
4764 				/* Make change permenant */
4765 				max_busy_irq->airq_cpu = (uchar_t)most_free_cpu;
4766 		} else if (min_busy_irq != NULL) {
4767 #ifdef	DEBUG
4768 			if (apic_verbose & APIC_VERBOSE_IOAPIC_FLAG) {
4769 				prom_printf("rebinding %x to %x",
4770 				    min_busy_irq->airq_vector, most_free_cpu);
4771 			}
4772 #endif /* DEBUG */
4773 
4774 			if (apic_rebind_all(min_busy_irq, most_free_cpu, 0) ==
4775 			    0)
4776 				/* Make change permenant */
4777 				min_busy_irq->airq_cpu = (uchar_t)most_free_cpu;
4778 		} else {
4779 			if (cpu_busy != (1 << busiest_cpu)) {
4780 				apic_redist_cpu_skip |= 1 << busiest_cpu;
4781 				/*
4782 				 * We leave cpu_skip set so that next time we
4783 				 * can choose another cpu
4784 				 */
4785 			}
4786 		}
4787 		apic_num_rebind++;
4788 	} else {
4789 		/*
4790 		 * found nothing. Could be that we skipped over valid CPUs
4791 		 * or we have balanced everything. If we had a variable
4792 		 * ticks_for_redistribution, it could be increased here.
4793 		 * apic_int_busy, int_free etc would also need to be
4794 		 * changed.
4795 		 */
4796 		if (apic_redist_cpu_skip)
4797 			apic_redist_cpu_skip = 0;
4798 	}
4799 	for (i = 0; i < apic_nproc; i++) {
4800 		apic_cpus[i].aci_busy = 0;
4801 	}
4802 }
4803 
4804 static void
4805 apic_cleanup_busy()
4806 {
4807 	int i;
4808 	apic_irq_t *irq_ptr;
4809 
4810 	for (i = 0; i < apic_nproc; i++) {
4811 		apic_cpus[i].aci_busy = 0;
4812 	}
4813 
4814 	for (i = apic_min_device_irq; i < apic_max_device_irq; i++) {
4815 		if ((irq_ptr = apic_irq_table[i]) != NULL)
4816 			irq_ptr->airq_busy = 0;
4817 	}
4818 	apic_skipped_redistribute = 0;
4819 }
4820 
4821 
4822 /*
4823  * This function will reprogram the timer.
4824  *
4825  * When in oneshot mode the argument is the absolute time in future to
4826  * generate the interrupt at.
4827  *
4828  * When in periodic mode, the argument is the interval at which the
4829  * interrupts should be generated. There is no need to support the periodic
4830  * mode timer change at this time.
4831  */
4832 static void
4833 apic_timer_reprogram(hrtime_t time)
4834 {
4835 	hrtime_t now;
4836 	uint_t ticks;
4837 
4838 	/*
4839 	 * We should be called from high PIL context (CBE_HIGH_PIL),
4840 	 * so kpreempt is disabled.
4841 	 */
4842 
4843 	if (!apic_oneshot) {
4844 		/* time is the interval for periodic mode */
4845 		ticks = (uint_t)((time) / apic_nsec_per_tick);
4846 	} else {
4847 		/* one shot mode */
4848 
4849 		now = gethrtime();
4850 
4851 		if (time <= now) {
4852 			/*
4853 			 * requested to generate an interrupt in the past
4854 			 * generate an interrupt as soon as possible
4855 			 */
4856 			ticks = apic_min_timer_ticks;
4857 		} else if ((time - now) > apic_nsec_max) {
4858 			/*
4859 			 * requested to generate an interrupt at a time
4860 			 * further than what we are capable of. Set to max
4861 			 * the hardware can handle
4862 			 */
4863 
4864 			ticks = APIC_MAXVAL;
4865 #ifdef DEBUG
4866 			cmn_err(CE_CONT, "apic_timer_reprogram, request at"
4867 			    "  %lld  too far in future, current time"
4868 			    "  %lld \n", time, now);
4869 #endif	/* DEBUG */
4870 		} else
4871 			ticks = (uint_t)((time - now) / apic_nsec_per_tick);
4872 	}
4873 
4874 	if (ticks < apic_min_timer_ticks)
4875 		ticks = apic_min_timer_ticks;
4876 
4877 	apicadr[APIC_INIT_COUNT] = ticks;
4878 
4879 }
4880 
4881 /*
4882  * This function will enable timer interrupts.
4883  */
4884 static void
4885 apic_timer_enable(void)
4886 {
4887 	/*
4888 	 * We should be Called from high PIL context (CBE_HIGH_PIL),
4889 	 * so kpreempt is disabled.
4890 	 */
4891 
4892 	if (!apic_oneshot)
4893 		apicadr[APIC_LOCAL_TIMER] =
4894 		    (apic_clkvect + APIC_BASE_VECT) | AV_TIME;
4895 	else {
4896 		/* one shot */
4897 		apicadr[APIC_LOCAL_TIMER] = (apic_clkvect + APIC_BASE_VECT);
4898 	}
4899 }
4900 
4901 /*
4902  * This function will disable timer interrupts.
4903  */
4904 static void
4905 apic_timer_disable(void)
4906 {
4907 	/*
4908 	 * We should be Called from high PIL context (CBE_HIGH_PIL),
4909 	 * so kpreempt is disabled.
4910 	 */
4911 
4912 	apicadr[APIC_LOCAL_TIMER] = (apic_clkvect + APIC_BASE_VECT) | AV_MASK;
4913 }
4914 
4915 
4916 cyclic_id_t apic_cyclic_id;
4917 
4918 /*
4919  * If this module needs to be a consumer of cyclic subsystem, they
4920  * can be added here, since at this time kernel cyclic subsystem is initialized
4921  * argument is not currently used, and is reserved for future.
4922  */
4923 static void
4924 apic_post_cyclic_setup(void *arg)
4925 {
4926 _NOTE(ARGUNUSED(arg))
4927 	cyc_handler_t hdlr;
4928 	cyc_time_t when;
4929 
4930 	/* cpu_lock is held */
4931 
4932 	/* set up cyclics for intr redistribution */
4933 
4934 	/*
4935 	 * In peridoc mode intr redistribution processing is done in
4936 	 * apic_intr_enter during clk intr processing
4937 	 */
4938 	if (!apic_oneshot)
4939 		return;
4940 
4941 	hdlr.cyh_level = CY_LOW_LEVEL;
4942 	hdlr.cyh_func = (cyc_func_t)apic_redistribute_compute;
4943 	hdlr.cyh_arg = NULL;
4944 
4945 	when.cyt_when = 0;
4946 	when.cyt_interval = apic_redistribute_sample_interval;
4947 	apic_cyclic_id = cyclic_add(&hdlr, &when);
4948 
4949 
4950 }
4951 
4952 static void
4953 apic_redistribute_compute(void)
4954 {
4955 	int	i, j, max_busy;
4956 
4957 	if (apic_enable_dynamic_migration) {
4958 		if (++apic_nticks == apic_sample_factor_redistribution) {
4959 			/*
4960 			 * Time to call apic_intr_redistribute().
4961 			 * reset apic_nticks. This will cause max_busy
4962 			 * to be calculated below and if it is more than
4963 			 * apic_int_busy, we will do the whole thing
4964 			 */
4965 			apic_nticks = 0;
4966 		}
4967 		max_busy = 0;
4968 		for (i = 0; i < apic_nproc; i++) {
4969 
4970 			/*
4971 			 * Check if curipl is non zero & if ISR is in
4972 			 * progress
4973 			 */
4974 			if (((j = apic_cpus[i].aci_curipl) != 0) &&
4975 			    (apic_cpus[i].aci_ISR_in_progress & (1 << j))) {
4976 
4977 				int	irq;
4978 				apic_cpus[i].aci_busy++;
4979 				irq = apic_cpus[i].aci_current[j];
4980 				apic_irq_table[irq]->airq_busy++;
4981 			}
4982 
4983 			if (!apic_nticks &&
4984 			    (apic_cpus[i].aci_busy > max_busy))
4985 				max_busy = apic_cpus[i].aci_busy;
4986 		}
4987 		if (!apic_nticks) {
4988 			if (max_busy > apic_int_busy_mark) {
4989 			/*
4990 			 * We could make the following check be
4991 			 * skipped > 1 in which case, we get a
4992 			 * redistribution at half the busy mark (due to
4993 			 * double interval). Need to be able to collect
4994 			 * more empirical data to decide if that is a
4995 			 * good strategy. Punt for now.
4996 			 */
4997 				if (apic_skipped_redistribute)
4998 					apic_cleanup_busy();
4999 				else
5000 					apic_intr_redistribute();
5001 			} else
5002 				apic_skipped_redistribute++;
5003 		}
5004 	}
5005 }
5006 
5007 
5008 static int
5009 apic_acpi_translate_pci_irq(dev_info_t *dip, int busid, int devid,
5010     int ipin, int *pci_irqp, iflag_t *intr_flagp)
5011 {
5012 
5013 	int status;
5014 	acpi_psm_lnk_t acpipsmlnk;
5015 
5016 	if ((status = acpi_get_irq_cache_ent(busid, devid, ipin, pci_irqp,
5017 	    intr_flagp)) == ACPI_PSM_SUCCESS) {
5018 		APIC_VERBOSE_IRQ((CE_CONT, "!pcplusmp: Found irqno %d "
5019 		    "from cache for device %s, instance #%d\n", *pci_irqp,
5020 		    ddi_get_name(dip), ddi_get_instance(dip)));
5021 		return (status);
5022 	}
5023 
5024 	bzero(&acpipsmlnk, sizeof (acpi_psm_lnk_t));
5025 
5026 	if ((status = acpi_translate_pci_irq(dip, ipin, pci_irqp, intr_flagp,
5027 	    &acpipsmlnk)) == ACPI_PSM_FAILURE) {
5028 		APIC_VERBOSE_IRQ((CE_WARN, "pcplusmp: "
5029 		    " acpi_translate_pci_irq failed for device %s, instance"
5030 		    " #%d", ddi_get_name(dip), ddi_get_instance(dip)));
5031 		return (status);
5032 	}
5033 
5034 	if (status == ACPI_PSM_PARTIAL && acpipsmlnk.lnkobj != NULL) {
5035 		status = apic_acpi_irq_configure(&acpipsmlnk, dip, pci_irqp,
5036 		    intr_flagp);
5037 		if (status != ACPI_PSM_SUCCESS) {
5038 			status = acpi_get_current_irq_resource(&acpipsmlnk,
5039 			    pci_irqp, intr_flagp);
5040 		}
5041 	}
5042 
5043 	if (status == ACPI_PSM_SUCCESS) {
5044 		acpi_new_irq_cache_ent(busid, devid, ipin, *pci_irqp,
5045 		    intr_flagp, &acpipsmlnk);
5046 
5047 		APIC_VERBOSE_IRQ((CE_CONT, "pcplusmp: [ACPI] "
5048 		    "new irq %d for device %s, instance #%d\n",
5049 		    *pci_irqp, ddi_get_name(dip), ddi_get_instance(dip)));
5050 	}
5051 
5052 	return (status);
5053 }
5054 
5055 /*
5056  * Configures the irq for the interrupt link device identified by
5057  * acpipsmlnkp.
5058  *
5059  * Gets the current and the list of possible irq settings for the
5060  * device. If apic_unconditional_srs is not set, and the current
5061  * resource setting is in the list of possible irq settings,
5062  * current irq resource setting is passed to the caller.
5063  *
5064  * Otherwise, picks an irq number from the list of possible irq
5065  * settings, and sets the irq of the device to this value.
5066  * If prefer_crs is set, among a set of irq numbers in the list that have
5067  * the least number of devices sharing the interrupt, we pick current irq
5068  * resource setting if it is a member of this set.
5069  *
5070  * Passes the irq number in the value pointed to by pci_irqp, and
5071  * polarity and sensitivity in the structure pointed to by dipintrflagp
5072  * to the caller.
5073  *
5074  * Note that if setting the irq resource failed, but successfuly obtained
5075  * the current irq resource settings, passes the current irq resources
5076  * and considers it a success.
5077  *
5078  * Returns:
5079  * ACPI_PSM_SUCCESS on success.
5080  *
5081  * ACPI_PSM_FAILURE if an error occured during the configuration or
5082  * if a suitable irq was not found for this device, or if setting the
5083  * irq resource and obtaining the current resource fails.
5084  *
5085  */
5086 static int
5087 apic_acpi_irq_configure(acpi_psm_lnk_t *acpipsmlnkp, dev_info_t *dip,
5088     int *pci_irqp, iflag_t *dipintr_flagp)
5089 {
5090 
5091 	int i, min_share, foundnow, done = 0;
5092 	int32_t irq;
5093 	int32_t share_irq = -1;
5094 	int32_t chosen_irq = -1;
5095 	int cur_irq = -1;
5096 	acpi_irqlist_t *irqlistp;
5097 	acpi_irqlist_t *irqlistent;
5098 
5099 	if ((acpi_get_possible_irq_resources(acpipsmlnkp, &irqlistp))
5100 	    == ACPI_PSM_FAILURE) {
5101 		APIC_VERBOSE_IRQ((CE_WARN, "!pcplusmp: Unable to determine "
5102 		    "or assign IRQ for device %s, instance #%d: The system was "
5103 		    "unable to get the list of potential IRQs from ACPI.",
5104 		    ddi_get_name(dip), ddi_get_instance(dip)));
5105 
5106 		return (ACPI_PSM_FAILURE);
5107 	}
5108 
5109 	if ((acpi_get_current_irq_resource(acpipsmlnkp, &cur_irq,
5110 	    dipintr_flagp) == ACPI_PSM_SUCCESS) && (!apic_unconditional_srs) &&
5111 	    (cur_irq > 0)) {
5112 		/*
5113 		 * If an IRQ is set in CRS and that IRQ exists in the set
5114 		 * returned from _PRS, return that IRQ, otherwise print
5115 		 * a warning
5116 		 */
5117 
5118 		if (acpi_irqlist_find_irq(irqlistp, cur_irq, NULL)
5119 		    == ACPI_PSM_SUCCESS) {
5120 
5121 			acpi_free_irqlist(irqlistp);
5122 			ASSERT(pci_irqp != NULL);
5123 			*pci_irqp = cur_irq;
5124 			return (ACPI_PSM_SUCCESS);
5125 		}
5126 
5127 		APIC_VERBOSE_IRQ((CE_WARN, "!pcplusmp: Could not find the "
5128 		    "current irq %d for device %s, instance #%d in ACPI's "
5129 		    "list of possible irqs for this device. Picking one from "
5130 		    " the latter list.", cur_irq, ddi_get_name(dip),
5131 		    ddi_get_instance(dip)));
5132 	}
5133 
5134 	irqlistent = irqlistp;
5135 	min_share = 255;
5136 
5137 	while (irqlistent != NULL) {
5138 		irqlistent->intr_flags.bustype = BUS_PCI;
5139 
5140 		for (foundnow = 0, i = 0; i < irqlistent->num_irqs; i++) {
5141 
5142 			irq = irqlistent->irqs[i];
5143 
5144 			if ((irq < 16) && (apic_reserved_irqlist[irq]))
5145 				continue;
5146 
5147 			if (irq == 0) {
5148 				/* invalid irq number */
5149 				continue;
5150 			}
5151 
5152 			if ((apic_irq_table[irq] == NULL) ||
5153 			    (apic_irq_table[irq]->airq_dip == dip)) {
5154 				chosen_irq = irq;
5155 				foundnow = 1;
5156 				/*
5157 				 * If we do not prefer current irq from crs
5158 				 * or if we do and this irq is the same as
5159 				 * current irq from crs, this is the one
5160 				 * to pick.
5161 				 */
5162 				if (!(apic_prefer_crs) || (irq == cur_irq)) {
5163 					done = 1;
5164 					break;
5165 				}
5166 				continue;
5167 			}
5168 
5169 			if (irqlistent->intr_flags.intr_el == INTR_EL_EDGE)
5170 				continue;
5171 
5172 			if (!acpi_intr_compatible(irqlistent->intr_flags,
5173 			    apic_irq_table[irq]->airq_iflag))
5174 				continue;
5175 
5176 			if ((apic_irq_table[irq]->airq_share < min_share) ||
5177 			    ((apic_irq_table[irq]->airq_share == min_share) &&
5178 			    (cur_irq == irq) && (apic_prefer_crs))) {
5179 				min_share = apic_irq_table[irq]->airq_share;
5180 				share_irq = irq;
5181 				foundnow = 1;
5182 			}
5183 		}
5184 
5185 		/*
5186 		 * If we found an IRQ in the inner loop this time, save the
5187 		 * details from the irqlist for later use.
5188 		 */
5189 		if (foundnow && ((chosen_irq != -1) || (share_irq != -1))) {
5190 			/*
5191 			 * Copy the acpi_prs_private_t and flags from this
5192 			 * irq list entry, since we found an irq from this
5193 			 * entry.
5194 			 */
5195 			acpipsmlnkp->acpi_prs_prv = irqlistent->acpi_prs_prv;
5196 			*dipintr_flagp = irqlistent->intr_flags;
5197 		}
5198 
5199 		if (done)
5200 			break;
5201 
5202 		/* Go to the next irqlist entry */
5203 		irqlistent = irqlistent->next;
5204 	}
5205 
5206 
5207 	acpi_free_irqlist(irqlistp);
5208 	if (chosen_irq != -1)
5209 		irq = chosen_irq;
5210 	else if (share_irq != -1)
5211 		irq = share_irq;
5212 	else {
5213 		APIC_VERBOSE_IRQ((CE_WARN, "!pcplusmp: Could not find a "
5214 		    "suitable irq from the list of possible irqs for device "
5215 		    "%s, instance #%d in ACPI's list of possible irqs",
5216 		    ddi_get_name(dip), ddi_get_instance(dip)));
5217 		return (ACPI_PSM_FAILURE);
5218 	}
5219 
5220 	APIC_VERBOSE_IRQ((CE_CONT, "!pcplusmp: Setting irq %d for device %s "
5221 	    "instance #%d\n", irq, ddi_get_name(dip), ddi_get_instance(dip)));
5222 
5223 	if ((acpi_set_irq_resource(acpipsmlnkp, irq)) == ACPI_PSM_SUCCESS) {
5224 		/*
5225 		 * setting irq was successful, check to make sure CRS
5226 		 * reflects that. If CRS does not agree with what we
5227 		 * set, return the irq that was set.
5228 		 */
5229 
5230 		if (acpi_get_current_irq_resource(acpipsmlnkp, &cur_irq,
5231 		    dipintr_flagp) == ACPI_PSM_SUCCESS) {
5232 
5233 			if (cur_irq != irq)
5234 				APIC_VERBOSE_IRQ((CE_WARN, "!pcplusmp: "
5235 				    "IRQ resource set (irqno %d) for device %s "
5236 				    "instance #%d, differs from current "
5237 				    "setting irqno %d",
5238 				    irq, ddi_get_name(dip),
5239 				    ddi_get_instance(dip), cur_irq));
5240 		}
5241 
5242 		/*
5243 		 * return the irq that was set, and not what CRS reports,
5244 		 * since CRS has been seen to be bogus on some systems
5245 		 */
5246 		cur_irq = irq;
5247 	} else {
5248 		APIC_VERBOSE_IRQ((CE_WARN, "!pcplusmp: set resource irq %d "
5249 		    "failed for device %s instance #%d",
5250 		    irq, ddi_get_name(dip), ddi_get_instance(dip)));
5251 
5252 		if (cur_irq == -1)
5253 			return (ACPI_PSM_FAILURE);
5254 	}
5255 
5256 	ASSERT(pci_irqp != NULL);
5257 	*pci_irqp = cur_irq;
5258 	return (ACPI_PSM_SUCCESS);
5259 }
5260