xref: /illumos-gate/usr/src/uts/i86pc/io/mp_platform_common.c (revision 6a074c93c5dee390d8ca2377f42e55418f0a9eb3)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 /*
29  * PSMI 1.1 extensions are supported only in 2.6 and later versions.
30  * PSMI 1.2 extensions are supported only in 2.7 and later versions.
31  * PSMI 1.3 and 1.4 extensions are supported in Solaris 10.
32  * PSMI 1.5 extensions are supported in Solaris Nevada.
33  */
34 #define	PSMI_1_5
35 
36 #include <sys/processor.h>
37 #include <sys/time.h>
38 #include <sys/psm.h>
39 #include <sys/smp_impldefs.h>
40 #include <sys/cram.h>
41 #include <sys/acpi/acpi.h>
42 #include <sys/acpica.h>
43 #include <sys/psm_common.h>
44 #include <sys/apic.h>
45 #include <sys/pit.h>
46 #include <sys/ddi.h>
47 #include <sys/sunddi.h>
48 #include <sys/ddi_impldefs.h>
49 #include <sys/pci.h>
50 #include <sys/promif.h>
51 #include <sys/x86_archext.h>
52 #include <sys/cpc_impl.h>
53 #include <sys/uadmin.h>
54 #include <sys/panic.h>
55 #include <sys/debug.h>
56 #include <sys/archsystm.h>
57 #include <sys/trap.h>
58 #include <sys/machsystm.h>
59 #include <sys/cpuvar.h>
60 #include <sys/rm_platter.h>
61 #include <sys/privregs.h>
62 #include <sys/cyclic.h>
63 #include <sys/note.h>
64 #include <sys/pci_intr_lib.h>
65 #include <sys/sunndi.h>
66 
67 
68 /*
69  *	Local Function Prototypes
70  */
71 static int apic_handle_defconf();
72 static int apic_parse_mpct(caddr_t mpct, int bypass);
73 static struct apic_mpfps_hdr *apic_find_fps_sig(caddr_t fptr, int size);
74 static int apic_checksum(caddr_t bptr, int len);
75 static int apic_find_bus_type(char *bus);
76 static int apic_find_bus(int busid);
77 static int apic_find_bus_id(int bustype);
78 static struct apic_io_intr *apic_find_io_intr(int irqno);
79 static int apic_find_free_irq(int start, int end);
80 static void apic_mark_vector(uchar_t oldvector, uchar_t newvector);
81 static void apic_xlate_vector_free_timeout_handler(void *arg);
82 static int apic_check_stuck_interrupt(apic_irq_t *irq_ptr, int old_bind_cpu,
83     int new_bind_cpu, int apicindex, int intin_no, int which_irq,
84     struct ioapic_reprogram_data *drep);
85 static void apic_record_rdt_entry(apic_irq_t *irqptr, int irq);
86 static struct apic_io_intr *apic_find_io_intr_w_busid(int irqno, int busid);
87 static int apic_find_intin(uchar_t ioapic, uchar_t intin);
88 static int apic_handle_pci_pci_bridge(dev_info_t *idip, int child_devno,
89     int child_ipin, struct apic_io_intr **intrp);
90 static int apic_setup_irq_table(dev_info_t *dip, int irqno,
91     struct apic_io_intr *intrp, struct intrspec *ispec, iflag_t *intr_flagp,
92     int type);
93 static void apic_set_pwroff_method_from_mpcnfhdr(struct apic_mp_cnf_hdr *hdrp);
94 static void apic_try_deferred_reprogram(int ipl, int vect);
95 static void delete_defer_repro_ent(int which_irq);
96 static void apic_ioapic_wait_pending_clear(int ioapicindex,
97     int intin_no);
98 static boolean_t apic_is_ioapic_AMD_813x(uint32_t physaddr);
99 
100 int apic_debug_mps_id = 0;	/* 1 - print MPS ID strings */
101 
102 /* ACPI SCI interrupt configuration; -1 if SCI not used */
103 int apic_sci_vect = -1;
104 iflag_t apic_sci_flags;
105 
106 /*
107  * psm name pointer
108  */
109 static char *psm_name;
110 
111 /* ACPI support routines */
112 static int acpi_probe(char *);
113 static int apic_acpi_irq_configure(acpi_psm_lnk_t *acpipsmlnkp, dev_info_t *dip,
114     int *pci_irqp, iflag_t *intr_flagp);
115 
116 static int apic_acpi_translate_pci_irq(dev_info_t *dip, int busid, int devid,
117     int ipin, int *pci_irqp, iflag_t *intr_flagp);
118 static uchar_t acpi_find_ioapic(int irq);
119 static int acpi_intr_compatible(iflag_t iflag1, iflag_t iflag2);
120 
121 
122 /*
123  * number of bits per byte, from <sys/param.h>
124  */
125 #define	UCHAR_MAX	((1 << NBBY) - 1)
126 
127 /* Max wait time (in repetitions) for flags to clear in an RDT entry. */
128 int apic_max_reps_clear_pending = 1000;
129 
130 /* The irq # is implicit in the array index: */
131 struct ioapic_reprogram_data apic_reprogram_info[APIC_MAX_VECTOR+1];
132 /*
133  * APIC_MAX_VECTOR + 1 is the maximum # of IRQs as well. ioapic_reprogram_info
134  * is indexed by IRQ number, NOT by vector number.
135  */
136 
137 int	apic_intr_policy = INTR_ROUND_ROBIN_WITH_AFFINITY;
138 
139 int	apic_next_bind_cpu = 1; /* For round robin assignment */
140 				/* start with cpu 1 */
141 
142 /*
143  * If enabled, the distribution works as follows:
144  * On every interrupt entry, the current ipl for the CPU is set in cpu_info
145  * and the irq corresponding to the ipl is also set in the aci_current array.
146  * interrupt exit and setspl (due to soft interrupts) will cause the current
147  * ipl to be be changed. This is cache friendly as these frequently used
148  * paths write into a per cpu structure.
149  *
150  * Sampling is done by checking the structures for all CPUs and incrementing
151  * the busy field of the irq (if any) executing on each CPU and the busy field
152  * of the corresponding CPU.
153  * In periodic mode this is done on every clock interrupt.
154  * In one-shot mode, this is done thru a cyclic with an interval of
155  * apic_redistribute_sample_interval (default 10 milli sec).
156  *
157  * Every apic_sample_factor_redistribution times we sample, we do computations
158  * to decide which interrupt needs to be migrated (see comments
159  * before apic_intr_redistribute().
160  */
161 
162 /*
163  * Following 3 variables start as % and can be patched or set using an
164  * API to be defined in future. They will be scaled to
165  * sample_factor_redistribution which is in turn set to hertz+1 (in periodic
166  * mode), or 101 in one-shot mode to stagger it away from one sec processing
167  */
168 
169 int	apic_int_busy_mark = 60;
170 int	apic_int_free_mark = 20;
171 int	apic_diff_for_redistribution = 10;
172 
173 /* sampling interval for interrupt redistribution for dynamic migration */
174 int	apic_redistribute_sample_interval = NANOSEC / 100; /* 10 millisec */
175 
176 /*
177  * number of times we sample before deciding to redistribute interrupts
178  * for dynamic migration
179  */
180 int	apic_sample_factor_redistribution = 101;
181 
182 /* timeout for xlate_vector, mark_vector */
183 int	apic_revector_timeout = 16 * 10000; /* 160 millisec */
184 
185 int	apic_redist_cpu_skip = 0;
186 int	apic_num_imbalance = 0;
187 int	apic_num_rebind = 0;
188 
189 int	apic_nproc = 0;
190 size_t	apic_cpus_size = 0;
191 int	apic_defconf = 0;
192 int	apic_irq_translate = 0;
193 int	apic_spec_rev = 0;
194 int	apic_imcrp = 0;
195 
196 int	apic_use_acpi = 1;	/* 1 = use ACPI, 0 = don't use ACPI */
197 int	apic_use_acpi_madt_only = 0;	/* 1=ONLY use MADT from ACPI */
198 
199 /*
200  * For interrupt link devices, if apic_unconditional_srs is set, an irq resource
201  * will be assigned (via _SRS). If it is not set, use the current
202  * irq setting (via _CRS), but only if that irq is in the set of possible
203  * irqs (returned by _PRS) for the device.
204  */
205 int	apic_unconditional_srs = 1;
206 
207 /*
208  * For interrupt link devices, if apic_prefer_crs is set when we are
209  * assigning an IRQ resource to a device, prefer the current IRQ setting
210  * over other possible irq settings under same conditions.
211  */
212 
213 int	apic_prefer_crs = 1;
214 
215 uchar_t	apic_io_id[MAX_IO_APIC];
216 volatile uint32_t *apicioadr[MAX_IO_APIC];
217 static	uchar_t	apic_io_ver[MAX_IO_APIC];
218 static	uchar_t	apic_io_vectbase[MAX_IO_APIC];
219 static	uchar_t	apic_io_vectend[MAX_IO_APIC];
220 uchar_t apic_reserved_irqlist[MAX_ISA_IRQ + 1];
221 uint32_t apic_physaddr[MAX_IO_APIC];
222 
223 static	boolean_t ioapic_mask_workaround[MAX_IO_APIC];
224 
225 /*
226  * First available slot to be used as IRQ index into the apic_irq_table
227  * for those interrupts (like MSI/X) that don't have a physical IRQ.
228  */
229 int apic_first_avail_irq  = APIC_FIRST_FREE_IRQ;
230 
231 /*
232  * apic_ioapic_lock protects the ioapics (reg select), the status, temp_bound
233  * and bound elements of cpus_info and the temp_cpu element of irq_struct
234  */
235 lock_t	apic_ioapic_lock;
236 
237 /*
238  * apic_defer_reprogram_lock ensures that only one processor is handling
239  * deferred interrupt programming at *_intr_exit time.
240  */
241 static	lock_t	apic_defer_reprogram_lock;
242 
243 /*
244  * The current number of deferred reprogrammings outstanding
245  */
246 uint_t	apic_reprogram_outstanding = 0;
247 
248 #ifdef DEBUG
249 /*
250  * Counters that keep track of deferred reprogramming stats
251  */
252 uint_t	apic_intr_deferrals = 0;
253 uint_t	apic_intr_deliver_timeouts = 0;
254 uint_t	apic_last_ditch_reprogram_failures = 0;
255 uint_t	apic_deferred_setup_failures = 0;
256 uint_t	apic_defer_repro_total_retries = 0;
257 uint_t	apic_defer_repro_successes = 0;
258 uint_t	apic_deferred_spurious_enters = 0;
259 #endif
260 
261 static	int	apic_io_max = 0;	/* no. of i/o apics enabled */
262 
263 static	struct apic_io_intr *apic_io_intrp = 0;
264 static	struct apic_bus	*apic_busp;
265 
266 uchar_t	apic_vector_to_irq[APIC_MAX_VECTOR+1];
267 uchar_t	apic_resv_vector[MAXIPL+1];
268 
269 char	apic_level_intr[APIC_MAX_VECTOR+1];
270 
271 static	uint32_t	eisa_level_intr_mask = 0;
272 	/* At least MSB will be set if EISA bus */
273 
274 static	int	apic_pci_bus_total = 0;
275 static	uchar_t	apic_single_pci_busid = 0;
276 
277 /*
278  * airq_mutex protects additions to the apic_irq_table - the first
279  * pointer and any airq_nexts off of that one. It also protects
280  * apic_max_device_irq & apic_min_device_irq. It also guarantees
281  * that share_id is unique as new ids are generated only when new
282  * irq_t structs are linked in. Once linked in the structs are never
283  * deleted. temp_cpu & mps_intr_index field indicate if it is programmed
284  * or allocated. Note that there is a slight gap between allocating in
285  * apic_introp_xlate and programming in addspl.
286  */
287 kmutex_t	airq_mutex;
288 apic_irq_t	*apic_irq_table[APIC_MAX_VECTOR+1];
289 int		apic_max_device_irq = 0;
290 int		apic_min_device_irq = APIC_MAX_VECTOR;
291 
292 /*
293  * Following declarations are for revectoring; used when ISRs at different
294  * IPLs share an irq.
295  */
296 static	lock_t	apic_revector_lock;
297 int	apic_revector_pending = 0;
298 static	uchar_t	*apic_oldvec_to_newvec;
299 static	uchar_t	*apic_newvec_to_oldvec;
300 
301 typedef struct prs_irq_list_ent {
302 	int			list_prio;
303 	int32_t			irq;
304 	iflag_t			intrflags;
305 	acpi_prs_private_t	prsprv;
306 	struct prs_irq_list_ent	*next;
307 } prs_irq_list_t;
308 
309 
310 /*
311  * ACPI variables
312  */
313 /* 1 = acpi is enabled & working, 0 = acpi is not enabled or not there */
314 int apic_enable_acpi = 0;
315 
316 /* ACPI Multiple APIC Description Table ptr */
317 static	MULTIPLE_APIC_TABLE *acpi_mapic_dtp = NULL;
318 
319 /* ACPI Interrupt Source Override Structure ptr */
320 static	MADT_INTERRUPT_OVERRIDE *acpi_isop = NULL;
321 static	int acpi_iso_cnt = 0;
322 
323 /* ACPI Non-maskable Interrupt Sources ptr */
324 static	MADT_NMI_SOURCE *acpi_nmi_sp = NULL;
325 static	int acpi_nmi_scnt = 0;
326 static	MADT_LOCAL_APIC_NMI *acpi_nmi_cp = NULL;
327 static	int acpi_nmi_ccnt = 0;
328 
329 /*
330  * The following added to identify a software poweroff method if available.
331  */
332 
333 static struct {
334 	int	poweroff_method;
335 	char	oem_id[APIC_MPS_OEM_ID_LEN + 1];	/* MAX + 1 for NULL */
336 	char	prod_id[APIC_MPS_PROD_ID_LEN + 1];	/* MAX + 1 for NULL */
337 } apic_mps_ids[] = {
338 	{ APIC_POWEROFF_VIA_RTC,	"INTEL",	"ALDER" },   /* 4300 */
339 	{ APIC_POWEROFF_VIA_RTC,	"NCR",		"AMC" },    /* 4300 */
340 	{ APIC_POWEROFF_VIA_ASPEN_BMC,	"INTEL",	"A450NX" },  /* 4400? */
341 	{ APIC_POWEROFF_VIA_ASPEN_BMC,	"INTEL",	"AD450NX" }, /* 4400 */
342 	{ APIC_POWEROFF_VIA_ASPEN_BMC,	"INTEL",	"AC450NX" }, /* 4400R */
343 	{ APIC_POWEROFF_VIA_SITKA_BMC,	"INTEL",	"S450NX" },  /* S50  */
344 	{ APIC_POWEROFF_VIA_SITKA_BMC,	"INTEL",	"SC450NX" }  /* S50? */
345 };
346 
347 int	apic_poweroff_method = APIC_POWEROFF_NONE;
348 
349 /*
350  * Auto-configuration routines
351  */
352 
353 /*
354  * Look at MPSpec 1.4 (Intel Order # 242016-005) for details of what we do here
355  * May work with 1.1 - but not guaranteed.
356  * According to the MP Spec, the MP floating pointer structure
357  * will be searched in the order described below:
358  * 1. In the first kilobyte of Extended BIOS Data Area (EBDA)
359  * 2. Within the last kilobyte of system base memory
360  * 3. In the BIOS ROM address space between 0F0000h and 0FFFFh
361  * Once we find the right signature with proper checksum, we call
362  * either handle_defconf or parse_mpct to get all info necessary for
363  * subsequent operations.
364  */
365 int
366 apic_probe_common(char *modname)
367 {
368 	uint32_t mpct_addr, ebda_start = 0, base_mem_end;
369 	caddr_t	biosdatap;
370 	caddr_t	mpct;
371 	caddr_t	fptr;
372 	int	i, mpct_size, mapsize, retval = PSM_FAILURE;
373 	ushort_t	ebda_seg, base_mem_size;
374 	struct	apic_mpfps_hdr	*fpsp;
375 	struct	apic_mp_cnf_hdr	*hdrp;
376 	int bypass_cpu_and_ioapics_in_mptables;
377 	int acpi_user_options;
378 
379 	if (apic_forceload < 0)
380 		return (retval);
381 
382 	/*
383 	 * Remember who we are
384 	 */
385 	psm_name = modname;
386 
387 	/* Allow override for MADT-only mode */
388 	acpi_user_options = ddi_prop_get_int(DDI_DEV_T_ANY, ddi_root_node(), 0,
389 	    "acpi-user-options", 0);
390 	apic_use_acpi_madt_only = ((acpi_user_options & ACPI_OUSER_MADT) != 0);
391 
392 	/* Allow apic_use_acpi to override MADT-only mode */
393 	if (!apic_use_acpi)
394 		apic_use_acpi_madt_only = 0;
395 
396 	retval = acpi_probe(modname);
397 
398 	/*
399 	 * mapin the bios data area 40:0
400 	 * 40:13h - two-byte location reports the base memory size
401 	 * 40:0Eh - two-byte location for the exact starting address of
402 	 *	    the EBDA segment for EISA
403 	 */
404 	biosdatap = psm_map_phys(0x400, 0x20, PROT_READ);
405 	if (!biosdatap)
406 		return (retval);
407 	fpsp = (struct apic_mpfps_hdr *)NULL;
408 	mapsize = MPFPS_RAM_WIN_LEN;
409 	/*LINTED: pointer cast may result in improper alignment */
410 	ebda_seg = *((ushort_t *)(biosdatap+0xe));
411 	/* check the 1k of EBDA */
412 	if (ebda_seg) {
413 		ebda_start = ((uint32_t)ebda_seg) << 4;
414 		fptr = psm_map_phys(ebda_start, MPFPS_RAM_WIN_LEN, PROT_READ);
415 		if (fptr) {
416 			if (!(fpsp =
417 			    apic_find_fps_sig(fptr, MPFPS_RAM_WIN_LEN)))
418 				psm_unmap_phys(fptr, MPFPS_RAM_WIN_LEN);
419 		}
420 	}
421 	/* If not in EBDA, check the last k of system base memory */
422 	if (!fpsp) {
423 		/*LINTED: pointer cast may result in improper alignment */
424 		base_mem_size = *((ushort_t *)(biosdatap + 0x13));
425 
426 		if (base_mem_size > 512)
427 			base_mem_end = 639 * 1024;
428 		else
429 			base_mem_end = 511 * 1024;
430 		/* if ebda == last k of base mem, skip to check BIOS ROM */
431 		if (base_mem_end != ebda_start) {
432 
433 			fptr = psm_map_phys(base_mem_end, MPFPS_RAM_WIN_LEN,
434 			    PROT_READ);
435 
436 			if (fptr) {
437 				if (!(fpsp = apic_find_fps_sig(fptr,
438 				    MPFPS_RAM_WIN_LEN)))
439 					psm_unmap_phys(fptr, MPFPS_RAM_WIN_LEN);
440 			}
441 		}
442 	}
443 	psm_unmap_phys(biosdatap, 0x20);
444 
445 	/* If still cannot find it, check the BIOS ROM space */
446 	if (!fpsp) {
447 		mapsize = MPFPS_ROM_WIN_LEN;
448 		fptr = psm_map_phys(MPFPS_ROM_WIN_START,
449 		    MPFPS_ROM_WIN_LEN, PROT_READ);
450 		if (fptr) {
451 			if (!(fpsp =
452 			    apic_find_fps_sig(fptr, MPFPS_ROM_WIN_LEN))) {
453 				psm_unmap_phys(fptr, MPFPS_ROM_WIN_LEN);
454 				return (retval);
455 			}
456 		}
457 	}
458 
459 	if (apic_checksum((caddr_t)fpsp, fpsp->mpfps_length * 16) != 0) {
460 		psm_unmap_phys(fptr, MPFPS_ROM_WIN_LEN);
461 		return (retval);
462 	}
463 
464 	apic_spec_rev = fpsp->mpfps_spec_rev;
465 	if ((apic_spec_rev != 04) && (apic_spec_rev != 01)) {
466 		psm_unmap_phys(fptr, MPFPS_ROM_WIN_LEN);
467 		return (retval);
468 	}
469 
470 	/* check IMCR is present or not */
471 	apic_imcrp = fpsp->mpfps_featinfo2 & MPFPS_FEATINFO2_IMCRP;
472 
473 	/* check default configuration (dual CPUs) */
474 	if ((apic_defconf = fpsp->mpfps_featinfo1) != 0) {
475 		psm_unmap_phys(fptr, mapsize);
476 		return (apic_handle_defconf());
477 	}
478 
479 	/* MP Configuration Table */
480 	mpct_addr = (uint32_t)(fpsp->mpfps_mpct_paddr);
481 
482 	psm_unmap_phys(fptr, mapsize); /* unmap floating ptr struct */
483 
484 	/*
485 	 * Map in enough memory for the MP Configuration Table Header.
486 	 * Use this table to read the total length of the BIOS data and
487 	 * map in all the info
488 	 */
489 	/*LINTED: pointer cast may result in improper alignment */
490 	hdrp = (struct apic_mp_cnf_hdr *)psm_map_phys(mpct_addr,
491 	    sizeof (struct apic_mp_cnf_hdr), PROT_READ);
492 	if (!hdrp)
493 		return (retval);
494 
495 	/* check mp configuration table signature PCMP */
496 	if (hdrp->mpcnf_sig != 0x504d4350) {
497 		psm_unmap_phys((caddr_t)hdrp, sizeof (struct apic_mp_cnf_hdr));
498 		return (retval);
499 	}
500 	mpct_size = (int)hdrp->mpcnf_tbl_length;
501 
502 	apic_set_pwroff_method_from_mpcnfhdr(hdrp);
503 
504 	psm_unmap_phys((caddr_t)hdrp, sizeof (struct apic_mp_cnf_hdr));
505 
506 	if ((retval == PSM_SUCCESS) && !apic_use_acpi_madt_only) {
507 		/* This is an ACPI machine No need for further checks */
508 		return (retval);
509 	}
510 
511 	/*
512 	 * Map in the entries for this machine, ie. Processor
513 	 * Entry Tables, Bus Entry Tables, etc.
514 	 * They are in fixed order following one another
515 	 */
516 	mpct = psm_map_phys(mpct_addr, mpct_size, PROT_READ);
517 	if (!mpct)
518 		return (retval);
519 
520 	if (apic_checksum(mpct, mpct_size) != 0)
521 		goto apic_fail1;
522 
523 
524 	/*LINTED: pointer cast may result in improper alignment */
525 	hdrp = (struct apic_mp_cnf_hdr *)mpct;
526 	apicadr = (uint32_t *)mapin_apic((uint32_t)hdrp->mpcnf_local_apic,
527 	    APIC_LOCAL_MEMLEN, PROT_READ | PROT_WRITE);
528 	if (!apicadr)
529 		goto apic_fail1;
530 
531 	/* Parse all information in the tables */
532 	bypass_cpu_and_ioapics_in_mptables = (retval == PSM_SUCCESS);
533 	if (apic_parse_mpct(mpct, bypass_cpu_and_ioapics_in_mptables) ==
534 	    PSM_SUCCESS)
535 		return (PSM_SUCCESS);
536 
537 	for (i = 0; i < apic_io_max; i++)
538 		mapout_ioapic((caddr_t)apicioadr[i], APIC_IO_MEMLEN);
539 	if (apic_cpus)
540 		kmem_free(apic_cpus, apic_cpus_size);
541 	if (apicadr)
542 		mapout_apic((caddr_t)apicadr, APIC_LOCAL_MEMLEN);
543 apic_fail1:
544 	psm_unmap_phys(mpct, mpct_size);
545 	return (retval);
546 }
547 
548 static void
549 apic_set_pwroff_method_from_mpcnfhdr(struct apic_mp_cnf_hdr *hdrp)
550 {
551 	int	i;
552 
553 	for (i = 0; i < (sizeof (apic_mps_ids) / sizeof (apic_mps_ids[0]));
554 	    i++) {
555 		if ((strncmp(hdrp->mpcnf_oem_str, apic_mps_ids[i].oem_id,
556 		    strlen(apic_mps_ids[i].oem_id)) == 0) &&
557 		    (strncmp(hdrp->mpcnf_prod_str, apic_mps_ids[i].prod_id,
558 		    strlen(apic_mps_ids[i].prod_id)) == 0)) {
559 
560 			apic_poweroff_method = apic_mps_ids[i].poweroff_method;
561 			break;
562 		}
563 	}
564 
565 	if (apic_debug_mps_id != 0) {
566 		cmn_err(CE_CONT, "%s: MPS OEM ID = '%c%c%c%c%c%c%c%c'"
567 		    "Product ID = '%c%c%c%c%c%c%c%c%c%c%c%c'\n",
568 		    psm_name,
569 		    hdrp->mpcnf_oem_str[0],
570 		    hdrp->mpcnf_oem_str[1],
571 		    hdrp->mpcnf_oem_str[2],
572 		    hdrp->mpcnf_oem_str[3],
573 		    hdrp->mpcnf_oem_str[4],
574 		    hdrp->mpcnf_oem_str[5],
575 		    hdrp->mpcnf_oem_str[6],
576 		    hdrp->mpcnf_oem_str[7],
577 		    hdrp->mpcnf_prod_str[0],
578 		    hdrp->mpcnf_prod_str[1],
579 		    hdrp->mpcnf_prod_str[2],
580 		    hdrp->mpcnf_prod_str[3],
581 		    hdrp->mpcnf_prod_str[4],
582 		    hdrp->mpcnf_prod_str[5],
583 		    hdrp->mpcnf_prod_str[6],
584 		    hdrp->mpcnf_prod_str[7],
585 		    hdrp->mpcnf_prod_str[8],
586 		    hdrp->mpcnf_prod_str[9],
587 		    hdrp->mpcnf_prod_str[10],
588 		    hdrp->mpcnf_prod_str[11]);
589 	}
590 }
591 
592 static int
593 acpi_probe(char *modname)
594 {
595 	int			i, intmax, index, rv;
596 	uint32_t		id, ver;
597 	int			acpi_verboseflags = 0;
598 	int			madt_seen, madt_size;
599 	APIC_HEADER		*ap;
600 	MADT_PROCESSOR_APIC	*mpa;
601 	MADT_IO_APIC		*mia;
602 	MADT_IO_SAPIC		*misa;
603 	MADT_INTERRUPT_OVERRIDE	*mio;
604 	MADT_NMI_SOURCE		*mns;
605 	MADT_INTERRUPT_SOURCE	*mis;
606 	MADT_LOCAL_APIC_NMI	*mlan;
607 	MADT_ADDRESS_OVERRIDE	*mao;
608 	ACPI_OBJECT_LIST 	arglist;
609 	ACPI_OBJECT		arg;
610 	int			sci;
611 	iflag_t			sci_flags;
612 	volatile uint32_t	*ioapic;
613 	int			apic_ix;
614 	char			local_ids[NCPU];
615 	char			proc_ids[NCPU];
616 	uchar_t			hid;
617 
618 	if (!apic_use_acpi)
619 		return (PSM_FAILURE);
620 
621 	if (AcpiGetFirmwareTable(APIC_SIG, 1, ACPI_LOGICAL_ADDRESSING,
622 	    (ACPI_TABLE_HEADER **) &acpi_mapic_dtp) != AE_OK)
623 		return (PSM_FAILURE);
624 
625 	apicadr = mapin_apic((uint32_t)acpi_mapic_dtp->LocalApicAddress,
626 	    APIC_LOCAL_MEMLEN, PROT_READ | PROT_WRITE);
627 	if (!apicadr)
628 		return (PSM_FAILURE);
629 
630 	id = apicadr[APIC_LID_REG];
631 	local_ids[0] = (uchar_t)(id >> 24);
632 	apic_nproc = index = 1;
633 	CPUSET_ONLY(apic_cpumask, 0);
634 	apic_io_max = 0;
635 
636 	ap = (APIC_HEADER *) (acpi_mapic_dtp + 1);
637 	madt_size = acpi_mapic_dtp->Length;
638 	madt_seen = sizeof (*acpi_mapic_dtp);
639 
640 	while (madt_seen < madt_size) {
641 		switch (ap->Type) {
642 		case APIC_PROCESSOR:
643 			mpa = (MADT_PROCESSOR_APIC *) ap;
644 			if (mpa->ProcessorEnabled) {
645 				if (mpa->LocalApicId == local_ids[0]) {
646 					proc_ids[0] = mpa->ProcessorId;
647 					acpica_map_cpu(0, mpa);
648 				} else if (apic_nproc < NCPU) {
649 					local_ids[index] = mpa->LocalApicId;
650 					proc_ids[index] = mpa->ProcessorId;
651 					CPUSET_ADD(apic_cpumask, index);
652 					acpica_map_cpu(index, mpa);
653 					index++;
654 					apic_nproc++;
655 				} else
656 					cmn_err(CE_WARN, "%s: exceeded "
657 					    "maximum no. of CPUs (= %d)",
658 					    psm_name,  NCPU);
659 			}
660 			break;
661 
662 		case APIC_IO:
663 			mia = (MADT_IO_APIC *) ap;
664 			if (apic_io_max < MAX_IO_APIC) {
665 				apic_ix = apic_io_max;
666 				apic_io_id[apic_io_max] = mia->IoApicId;
667 				apic_io_vectbase[apic_io_max] =
668 				    mia->Interrupt;
669 				apic_physaddr[apic_io_max] =
670 				    (uint32_t)mia->Address;
671 				ioapic = apicioadr[apic_io_max] =
672 				    mapin_ioapic((uint32_t)mia->Address,
673 				    APIC_IO_MEMLEN, PROT_READ | PROT_WRITE);
674 				if (!ioapic)
675 					goto cleanup;
676 				ioapic_mask_workaround[apic_io_max] =
677 				    apic_is_ioapic_AMD_813x(mia->Address);
678 				apic_io_max++;
679 			}
680 			break;
681 
682 		case APIC_XRUPT_OVERRIDE:
683 			mio = (MADT_INTERRUPT_OVERRIDE *) ap;
684 			if (acpi_isop == NULL)
685 				acpi_isop = mio;
686 			acpi_iso_cnt++;
687 			break;
688 
689 		case APIC_NMI:
690 			/* UNIMPLEMENTED */
691 			mns = (MADT_NMI_SOURCE *) ap;
692 			if (acpi_nmi_sp == NULL)
693 				acpi_nmi_sp = mns;
694 			acpi_nmi_scnt++;
695 
696 			cmn_err(CE_NOTE, "!apic: nmi source: %d %d %d\n",
697 			    mns->Interrupt, mns->Polarity,
698 			    mns->TriggerMode);
699 			break;
700 
701 		case APIC_LOCAL_NMI:
702 			/* UNIMPLEMENTED */
703 			mlan = (MADT_LOCAL_APIC_NMI *) ap;
704 			if (acpi_nmi_cp == NULL)
705 				acpi_nmi_cp = mlan;
706 			acpi_nmi_ccnt++;
707 
708 			cmn_err(CE_NOTE, "!apic: local nmi: %d %d %d %d\n",
709 			    mlan->ProcessorId, mlan->Polarity,
710 			    mlan->TriggerMode, mlan->Lint);
711 			break;
712 
713 		case APIC_ADDRESS_OVERRIDE:
714 			/* UNIMPLEMENTED */
715 			mao = (MADT_ADDRESS_OVERRIDE *) ap;
716 			cmn_err(CE_NOTE, "!apic: address override: %lx\n",
717 			    (long)mao->Address);
718 			break;
719 
720 		case APIC_IO_SAPIC:
721 			/* UNIMPLEMENTED */
722 			misa = (MADT_IO_SAPIC *) ap;
723 
724 			cmn_err(CE_NOTE, "!apic: io sapic: %d %d %lx\n",
725 			    misa->IoSapicId, misa->InterruptBase,
726 			    (long)misa->Address);
727 			break;
728 
729 		case APIC_XRUPT_SOURCE:
730 			/* UNIMPLEMENTED */
731 			mis = (MADT_INTERRUPT_SOURCE *) ap;
732 
733 			cmn_err(CE_NOTE,
734 			    "!apic: irq source: %d %d %d %d %d %d %d\n",
735 			    mis->ProcessorId, mis->ProcessorEid,
736 			    mis->Interrupt, mis->Polarity,
737 			    mis->TriggerMode, mis->InterruptType,
738 			    mis->IoSapicVector);
739 			break;
740 		default:
741 			break;
742 		}
743 
744 		/* advance to next entry */
745 		madt_seen += ap->Length;
746 		ap = (APIC_HEADER *)(((char *)ap) + ap->Length);
747 	}
748 
749 	apic_cpus_size = apic_nproc * sizeof (*apic_cpus);
750 	if ((apic_cpus = kmem_zalloc(apic_cpus_size, KM_NOSLEEP)) == NULL)
751 		goto cleanup;
752 
753 	/*
754 	 * ACPI doesn't provide the local apic ver, get it directly from the
755 	 * local apic
756 	 */
757 	ver = apicadr[APIC_VERS_REG];
758 	for (i = 0; i < apic_nproc; i++) {
759 		apic_cpus[i].aci_local_id = local_ids[i];
760 		apic_cpus[i].aci_local_ver = (uchar_t)(ver & 0xFF);
761 	}
762 	for (i = 0; i < apic_io_max; i++) {
763 		apic_ix = i;
764 
765 		/*
766 		 * need to check Sitka on the following acpi problem
767 		 * On the Sitka, the ioapic's apic_id field isn't reporting
768 		 * the actual io apic id. We have reported this problem
769 		 * to Intel. Until they fix the problem, we will get the
770 		 * actual id directly from the ioapic.
771 		 */
772 		id = ioapic_read(apic_ix, APIC_ID_CMD);
773 		hid = (uchar_t)(id >> 24);
774 
775 		if (hid != apic_io_id[i]) {
776 			if (apic_io_id[i] == 0)
777 				apic_io_id[i] = hid;
778 			else { /* set ioapic id to whatever reported by ACPI */
779 				id = ((uint32_t)apic_io_id[i]) << 24;
780 				ioapic_write(apic_ix, APIC_ID_CMD, id);
781 			}
782 		}
783 		ver = ioapic_read(apic_ix, APIC_VERS_CMD);
784 		apic_io_ver[i] = (uchar_t)(ver & 0xff);
785 		intmax = (ver >> 16) & 0xff;
786 		apic_io_vectend[i] = apic_io_vectbase[i] + intmax;
787 		if (apic_first_avail_irq <= apic_io_vectend[i])
788 			apic_first_avail_irq = apic_io_vectend[i] + 1;
789 	}
790 
791 
792 	/*
793 	 * Process SCI configuration here
794 	 * An error may be returned here if
795 	 * acpi-user-options specifies legacy mode
796 	 * (no SCI, no ACPI mode)
797 	 */
798 	if (acpica_get_sci(&sci, &sci_flags) != AE_OK)
799 		sci = -1;
800 
801 	/*
802 	 * Now call acpi_init() to generate namespaces
803 	 * If this fails, we don't attempt to use ACPI
804 	 * even if we were able to get a MADT above
805 	 */
806 	if (acpica_init() != AE_OK)
807 		goto cleanup;
808 
809 	/*
810 	 * Call acpica_build_processor_map() now that we have
811 	 * ACPI namesspace access
812 	 */
813 	acpica_build_processor_map();
814 
815 	/*
816 	 * Squirrel away the SCI and flags for later on
817 	 * in apic_picinit() when we're ready
818 	 */
819 	apic_sci_vect = sci;
820 	apic_sci_flags = sci_flags;
821 
822 	if (apic_verbose & APIC_VERBOSE_IRQ_FLAG)
823 		acpi_verboseflags |= PSM_VERBOSE_IRQ_FLAG;
824 
825 	if (apic_verbose & APIC_VERBOSE_POWEROFF_FLAG)
826 		acpi_verboseflags |= PSM_VERBOSE_POWEROFF_FLAG;
827 
828 	if (apic_verbose & APIC_VERBOSE_POWEROFF_PAUSE_FLAG)
829 		acpi_verboseflags |= PSM_VERBOSE_POWEROFF_PAUSE_FLAG;
830 
831 	if (acpi_psm_init(modname, acpi_verboseflags) == ACPI_PSM_FAILURE)
832 		goto cleanup;
833 
834 	/* Enable ACPI APIC interrupt routing */
835 	arglist.Count = 1;
836 	arglist.Pointer = &arg;
837 	arg.Type = ACPI_TYPE_INTEGER;
838 	arg.Integer.Value = ACPI_APIC_MODE;	/* 1 */
839 	rv = AcpiEvaluateObject(NULL, "\\_PIC", &arglist, NULL);
840 	if (rv == AE_OK) {
841 		build_reserved_irqlist((uchar_t *)apic_reserved_irqlist);
842 		apic_enable_acpi = 1;
843 		if (apic_use_acpi_madt_only) {
844 			cmn_err(CE_CONT,
845 			    "?Using ACPI for CPU/IOAPIC information ONLY\n");
846 		}
847 		return (PSM_SUCCESS);
848 	}
849 	/* if setting APIC mode failed above, we fall through to cleanup */
850 
851 cleanup:
852 	if (apicadr != NULL) {
853 		mapout_apic((caddr_t)apicadr, APIC_LOCAL_MEMLEN);
854 		apicadr = NULL;
855 	}
856 	apic_nproc = 0;
857 	for (i = 0; i < apic_io_max; i++) {
858 		mapout_ioapic((caddr_t)apicioadr[i], APIC_IO_MEMLEN);
859 		apicioadr[i] = NULL;
860 	}
861 	apic_io_max = 0;
862 	acpi_isop = NULL;
863 	acpi_iso_cnt = 0;
864 	acpi_nmi_sp = NULL;
865 	acpi_nmi_scnt = 0;
866 	acpi_nmi_cp = NULL;
867 	acpi_nmi_ccnt = 0;
868 	return (PSM_FAILURE);
869 }
870 
871 /*
872  * Handle default configuration. Fill in reqd global variables & tables
873  * Fill all details as MP table does not give any more info
874  */
875 static int
876 apic_handle_defconf()
877 {
878 	uint_t	lid;
879 
880 	/*LINTED: pointer cast may result in improper alignment */
881 	apicioadr[0] = mapin_ioapic(APIC_IO_ADDR,
882 	    APIC_IO_MEMLEN, PROT_READ | PROT_WRITE);
883 	/*LINTED: pointer cast may result in improper alignment */
884 	apicadr = (uint32_t *)psm_map_phys(APIC_LOCAL_ADDR,
885 	    APIC_LOCAL_MEMLEN, PROT_READ);
886 	apic_cpus_size = 2 * sizeof (*apic_cpus);
887 	apic_cpus = (apic_cpus_info_t *)
888 	    kmem_zalloc(apic_cpus_size, KM_NOSLEEP);
889 	if ((!apicadr) || (!apicioadr[0]) || (!apic_cpus))
890 		goto apic_handle_defconf_fail;
891 	CPUSET_ONLY(apic_cpumask, 0);
892 	CPUSET_ADD(apic_cpumask, 1);
893 	apic_nproc = 2;
894 	lid = apicadr[APIC_LID_REG];
895 	apic_cpus[0].aci_local_id = (uchar_t)(lid >> APIC_ID_BIT_OFFSET);
896 	/*
897 	 * According to the PC+MP spec 1.1, the local ids
898 	 * for the default configuration has to be 0 or 1
899 	 */
900 	if (apic_cpus[0].aci_local_id == 1)
901 		apic_cpus[1].aci_local_id = 0;
902 	else if (apic_cpus[0].aci_local_id == 0)
903 		apic_cpus[1].aci_local_id = 1;
904 	else
905 		goto apic_handle_defconf_fail;
906 
907 	apic_io_id[0] = 2;
908 	apic_io_max = 1;
909 	if (apic_defconf >= 5) {
910 		apic_cpus[0].aci_local_ver = APIC_INTEGRATED_VERS;
911 		apic_cpus[1].aci_local_ver = APIC_INTEGRATED_VERS;
912 		apic_io_ver[0] = APIC_INTEGRATED_VERS;
913 	} else {
914 		apic_cpus[0].aci_local_ver = 0;		/* 82489 DX */
915 		apic_cpus[1].aci_local_ver = 0;
916 		apic_io_ver[0] = 0;
917 	}
918 	if (apic_defconf == 2 || apic_defconf == 3 || apic_defconf == 6)
919 		eisa_level_intr_mask = (inb(EISA_LEVEL_CNTL + 1) << 8) |
920 		    inb(EISA_LEVEL_CNTL) | ((uint_t)INT32_MAX + 1);
921 	return (PSM_SUCCESS);
922 
923 apic_handle_defconf_fail:
924 	if (apic_cpus)
925 		kmem_free(apic_cpus, apic_cpus_size);
926 	if (apicadr)
927 		mapout_apic((caddr_t)apicadr, APIC_LOCAL_MEMLEN);
928 	if (apicioadr[0])
929 		mapout_ioapic((caddr_t)apicioadr[0], APIC_IO_MEMLEN);
930 	return (PSM_FAILURE);
931 }
932 
933 /* Parse the entries in MP configuration table and collect info that we need */
934 static int
935 apic_parse_mpct(caddr_t mpct, int bypass_cpus_and_ioapics)
936 {
937 	struct	apic_procent	*procp;
938 	struct	apic_bus	*busp;
939 	struct	apic_io_entry	*ioapicp;
940 	struct	apic_io_intr	*intrp;
941 	int			apic_ix;
942 	uint_t	lid;
943 	uint32_t	id;
944 	uchar_t hid;
945 
946 	/*LINTED: pointer cast may result in improper alignment */
947 	procp = (struct apic_procent *)(mpct + sizeof (struct apic_mp_cnf_hdr));
948 
949 	/* No need to count cpu entries if we won't use them */
950 	if (!bypass_cpus_and_ioapics) {
951 
952 		/* Find max # of CPUS and allocate structure accordingly */
953 		apic_nproc = 0;
954 		CPUSET_ZERO(apic_cpumask);
955 		while (procp->proc_entry == APIC_CPU_ENTRY) {
956 			if (procp->proc_cpuflags & CPUFLAGS_EN) {
957 				if (apic_nproc < NCPU)
958 					CPUSET_ADD(apic_cpumask, apic_nproc);
959 				apic_nproc++;
960 			}
961 			procp++;
962 		}
963 		if (apic_nproc > NCPU)
964 			cmn_err(CE_WARN, "%s: exceeded "
965 			    "maximum no. of CPUs (= %d)", psm_name, NCPU);
966 		apic_cpus_size = apic_nproc * sizeof (*apic_cpus);
967 		if (!apic_nproc || !(apic_cpus = (apic_cpus_info_t *)
968 		    kmem_zalloc(apic_cpus_size, KM_NOSLEEP)))
969 			return (PSM_FAILURE);
970 	}
971 
972 	/*LINTED: pointer cast may result in improper alignment */
973 	procp = (struct apic_procent *)(mpct + sizeof (struct apic_mp_cnf_hdr));
974 
975 	/*
976 	 * start with index 1 as 0 needs to be filled in with Boot CPU, but
977 	 * if we're bypassing this information, it has already been filled
978 	 * in by acpi_probe(), so don't overwrite it.
979 	 */
980 	if (!bypass_cpus_and_ioapics)
981 		apic_nproc = 1;
982 
983 	while (procp->proc_entry == APIC_CPU_ENTRY) {
984 		/* check whether the cpu exists or not */
985 		if (!bypass_cpus_and_ioapics &&
986 		    procp->proc_cpuflags & CPUFLAGS_EN) {
987 			if (procp->proc_cpuflags & CPUFLAGS_BP) { /* Boot CPU */
988 				lid = apicadr[APIC_LID_REG];
989 				apic_cpus[0].aci_local_id = procp->proc_apicid;
990 				if (apic_cpus[0].aci_local_id !=
991 				    (uchar_t)(lid >> APIC_ID_BIT_OFFSET)) {
992 					return (PSM_FAILURE);
993 				}
994 				apic_cpus[0].aci_local_ver =
995 				    procp->proc_version;
996 			} else {
997 
998 				apic_cpus[apic_nproc].aci_local_id =
999 				    procp->proc_apicid;
1000 				apic_cpus[apic_nproc].aci_local_ver =
1001 				    procp->proc_version;
1002 				apic_nproc++;
1003 
1004 			}
1005 		}
1006 		procp++;
1007 	}
1008 
1009 	/*
1010 	 * Save start of bus entries for later use.
1011 	 * Get EISA level cntrl if EISA bus is present.
1012 	 * Also get the CPI bus id for single CPI bus case
1013 	 */
1014 	apic_busp = busp = (struct apic_bus *)procp;
1015 	while (busp->bus_entry == APIC_BUS_ENTRY) {
1016 		lid = apic_find_bus_type((char *)&busp->bus_str1);
1017 		if (lid	== BUS_EISA) {
1018 			eisa_level_intr_mask = (inb(EISA_LEVEL_CNTL + 1) << 8) |
1019 			    inb(EISA_LEVEL_CNTL) | ((uint_t)INT32_MAX + 1);
1020 		} else if (lid == BUS_PCI) {
1021 			/*
1022 			 * apic_single_pci_busid will be used only if
1023 			 * apic_pic_bus_total is equal to 1
1024 			 */
1025 			apic_pci_bus_total++;
1026 			apic_single_pci_busid = busp->bus_id;
1027 		}
1028 		busp++;
1029 	}
1030 
1031 	ioapicp = (struct apic_io_entry *)busp;
1032 
1033 	if (!bypass_cpus_and_ioapics)
1034 		apic_io_max = 0;
1035 	do {
1036 		if (!bypass_cpus_and_ioapics && apic_io_max < MAX_IO_APIC) {
1037 			if (ioapicp->io_flags & IOAPIC_FLAGS_EN) {
1038 				apic_io_id[apic_io_max] = ioapicp->io_apicid;
1039 				apic_io_ver[apic_io_max] = ioapicp->io_version;
1040 		/*LINTED: pointer cast may result in improper alignment */
1041 				apicioadr[apic_io_max] =
1042 				    mapin_ioapic(
1043 				    (uint32_t)ioapicp->io_apic_addr,
1044 				    APIC_IO_MEMLEN, PROT_READ | PROT_WRITE);
1045 
1046 				if (!apicioadr[apic_io_max])
1047 					return (PSM_FAILURE);
1048 
1049 				ioapic_mask_workaround[apic_io_max] =
1050 				    apic_is_ioapic_AMD_813x(
1051 				    ioapicp->io_apic_addr);
1052 
1053 				apic_ix = apic_io_max;
1054 				id = ioapic_read(apic_ix, APIC_ID_CMD);
1055 				hid = (uchar_t)(id >> 24);
1056 
1057 				if (hid != apic_io_id[apic_io_max]) {
1058 					if (apic_io_id[apic_io_max] == 0)
1059 						apic_io_id[apic_io_max] = hid;
1060 					else {
1061 						/*
1062 						 * set ioapic id to whatever
1063 						 * reported by MPS
1064 						 *
1065 						 * may not need to set index
1066 						 * again ???
1067 						 * take it out and try
1068 						 */
1069 
1070 						id = ((uint32_t)
1071 						    apic_io_id[apic_io_max]) <<
1072 						    24;
1073 
1074 						ioapic_write(apic_ix,
1075 						    APIC_ID_CMD, id);
1076 					}
1077 				}
1078 				apic_io_max++;
1079 			}
1080 		}
1081 		ioapicp++;
1082 	} while (ioapicp->io_entry == APIC_IO_ENTRY);
1083 
1084 	apic_io_intrp = (struct apic_io_intr *)ioapicp;
1085 
1086 	intrp = apic_io_intrp;
1087 	while (intrp->intr_entry == APIC_IO_INTR_ENTRY) {
1088 		if ((intrp->intr_irq > APIC_MAX_ISA_IRQ) ||
1089 		    (apic_find_bus(intrp->intr_busid) == BUS_PCI)) {
1090 			apic_irq_translate = 1;
1091 			break;
1092 		}
1093 		intrp++;
1094 	}
1095 
1096 	return (PSM_SUCCESS);
1097 }
1098 
1099 boolean_t
1100 apic_cpu_in_range(int cpu)
1101 {
1102 	return ((cpu & ~IRQ_USER_BOUND) < apic_nproc);
1103 }
1104 
1105 uint16_t
1106 apic_get_apic_version()
1107 {
1108 	int i;
1109 	uchar_t min_io_apic_ver = 0;
1110 	static uint16_t version;		/* Cache as value is constant */
1111 	static boolean_t found = B_FALSE;	/* Accomodate zero version */
1112 
1113 	if (found == B_FALSE) {
1114 		found = B_TRUE;
1115 
1116 		/*
1117 		 * Don't assume all IO APICs in the system are the same.
1118 		 *
1119 		 * Set to the minimum version.
1120 		 */
1121 		for (i = 0; i < apic_io_max; i++) {
1122 			if ((apic_io_ver[i] != 0) &&
1123 			    ((min_io_apic_ver == 0) ||
1124 			    (min_io_apic_ver >= apic_io_ver[i])))
1125 				min_io_apic_ver = apic_io_ver[i];
1126 		}
1127 
1128 		/* Assume all local APICs are of the same version. */
1129 		version = (min_io_apic_ver << 8) | apic_cpus[0].aci_local_ver;
1130 	}
1131 	return (version);
1132 }
1133 
1134 static struct apic_mpfps_hdr *
1135 apic_find_fps_sig(caddr_t cptr, int len)
1136 {
1137 	int	i;
1138 
1139 	/* Look for the pattern "_MP_" */
1140 	for (i = 0; i < len; i += 16) {
1141 		if ((*(cptr+i) == '_') &&
1142 		    (*(cptr+i+1) == 'M') &&
1143 		    (*(cptr+i+2) == 'P') &&
1144 		    (*(cptr+i+3) == '_'))
1145 		    /*LINTED: pointer cast may result in improper alignment */
1146 			return ((struct apic_mpfps_hdr *)(cptr + i));
1147 	}
1148 	return (NULL);
1149 }
1150 
1151 static int
1152 apic_checksum(caddr_t bptr, int len)
1153 {
1154 	int	i;
1155 	uchar_t	cksum;
1156 
1157 	cksum = 0;
1158 	for (i = 0; i < len; i++)
1159 		cksum += *bptr++;
1160 	return ((int)cksum);
1161 }
1162 
1163 
1164 /*
1165  * Initialise vector->ipl and ipl->pri arrays. level_intr and irqtable
1166  * are also set to NULL. vector->irq is set to a value which cannot map
1167  * to a real irq to show that it is free.
1168  */
1169 void
1170 apic_init_common()
1171 {
1172 	int	i, j, indx;
1173 	int	*iptr;
1174 
1175 	/*
1176 	 * Initialize apic_ipls from apic_vectortoipl.  This array is
1177 	 * used in apic_intr_enter to determine the IPL to use for the
1178 	 * corresponding vector.  On some systems, due to hardware errata
1179 	 * and interrupt sharing, the IPL may not correspond to the IPL listed
1180 	 * in apic_vectortoipl (see apic_addspl and apic_delspl).
1181 	 */
1182 	for (i = 0; i < (APIC_AVAIL_VECTOR / APIC_VECTOR_PER_IPL); i++) {
1183 		indx = i * APIC_VECTOR_PER_IPL;
1184 
1185 		for (j = 0; j < APIC_VECTOR_PER_IPL; j++, indx++)
1186 			apic_ipls[indx] = apic_vectortoipl[i];
1187 	}
1188 
1189 	/* cpu 0 is always up (for now) */
1190 	apic_cpus[0].aci_status = APIC_CPU_ONLINE | APIC_CPU_INTR_ENABLE;
1191 
1192 	iptr = (int *)&apic_irq_table[0];
1193 	for (i = 0; i <= APIC_MAX_VECTOR; i++) {
1194 		apic_level_intr[i] = 0;
1195 		*iptr++ = NULL;
1196 		apic_vector_to_irq[i] = APIC_RESV_IRQ;
1197 
1198 		/* These *must* be initted to B_TRUE! */
1199 		apic_reprogram_info[i].done = B_TRUE;
1200 		apic_reprogram_info[i].irqp = NULL;
1201 		apic_reprogram_info[i].tries = 0;
1202 		apic_reprogram_info[i].bindcpu = 0;
1203 	}
1204 
1205 	/*
1206 	 * Allocate a dummy irq table entry for the reserved entry.
1207 	 * This takes care of the race between removing an irq and
1208 	 * clock detecting a CPU in that irq during interrupt load
1209 	 * sampling.
1210 	 */
1211 	apic_irq_table[APIC_RESV_IRQ] =
1212 	    kmem_zalloc(sizeof (apic_irq_t), KM_NOSLEEP);
1213 
1214 	mutex_init(&airq_mutex, NULL, MUTEX_DEFAULT, NULL);
1215 }
1216 
1217 void
1218 ioapic_init_intr(int mask_apic)
1219 {
1220 	int apic_ix;
1221 	struct intrspec ispec;
1222 	apic_irq_t *irqptr;
1223 	int i, j;
1224 	ulong_t iflag;
1225 
1226 	LOCK_INIT_CLEAR(&apic_revector_lock);
1227 	LOCK_INIT_CLEAR(&apic_defer_reprogram_lock);
1228 
1229 	/* mask interrupt vectors */
1230 	for (j = 0; j < apic_io_max && mask_apic; j++) {
1231 		int intin_max;
1232 
1233 		apic_ix = j;
1234 		/* Bits 23-16 define the maximum redirection entries */
1235 		intin_max = (ioapic_read(apic_ix, APIC_VERS_CMD) >> 16) & 0xff;
1236 		for (i = 0; i < intin_max; i++)
1237 			ioapic_write(apic_ix, APIC_RDT_CMD + 2 * i, AV_MASK);
1238 	}
1239 
1240 	/*
1241 	 * Hack alert: deal with ACPI SCI interrupt chicken/egg here
1242 	 */
1243 	if (apic_sci_vect > 0) {
1244 		/*
1245 		 * acpica has already done add_avintr(); we just
1246 		 * to finish the job by mimicing translate_irq()
1247 		 *
1248 		 * Fake up an intrspec and setup the tables
1249 		 */
1250 		ispec.intrspec_vec = apic_sci_vect;
1251 		ispec.intrspec_pri = SCI_IPL;
1252 
1253 		if (apic_setup_irq_table(NULL, apic_sci_vect, NULL,
1254 		    &ispec, &apic_sci_flags, DDI_INTR_TYPE_FIXED) < 0) {
1255 			cmn_err(CE_WARN, "!apic: SCI setup failed");
1256 			return;
1257 		}
1258 		irqptr = apic_irq_table[apic_sci_vect];
1259 
1260 		iflag = intr_clear();
1261 		lock_set(&apic_ioapic_lock);
1262 
1263 		/* Program I/O APIC */
1264 		(void) apic_setup_io_intr(irqptr, apic_sci_vect, B_FALSE);
1265 
1266 		lock_clear(&apic_ioapic_lock);
1267 		intr_restore(iflag);
1268 
1269 		irqptr->airq_share++;
1270 	}
1271 }
1272 
1273 /*
1274  * Add mask bits to disable interrupt vector from happening
1275  * at or above IPL. In addition, it should remove mask bits
1276  * to enable interrupt vectors below the given IPL.
1277  *
1278  * Both add and delspl are complicated by the fact that different interrupts
1279  * may share IRQs. This can happen in two ways.
1280  * 1. The same H/W line is shared by more than 1 device
1281  * 1a. with interrupts at different IPLs
1282  * 1b. with interrupts at same IPL
1283  * 2. We ran out of vectors at a given IPL and started sharing vectors.
1284  * 1b and 2 should be handled gracefully, except for the fact some ISRs
1285  * will get called often when no interrupt is pending for the device.
1286  * For 1a, we just hope that the machine blows up with the person who
1287  * set it up that way!. In the meantime, we handle it at the higher IPL.
1288  */
1289 /*ARGSUSED*/
1290 int
1291 apic_addspl_common(int irqno, int ipl, int min_ipl, int max_ipl)
1292 {
1293 	uchar_t vector;
1294 	ulong_t iflag;
1295 	apic_irq_t *irqptr, *irqheadptr;
1296 	int irqindex;
1297 
1298 	ASSERT(max_ipl <= UCHAR_MAX);
1299 	irqindex = IRQINDEX(irqno);
1300 
1301 	if ((irqindex == -1) || (!apic_irq_table[irqindex]))
1302 		return (PSM_FAILURE);
1303 
1304 	mutex_enter(&airq_mutex);
1305 	irqptr = irqheadptr = apic_irq_table[irqindex];
1306 
1307 	DDI_INTR_IMPLDBG((CE_CONT, "apic_addspl: dip=0x%p type=%d irqno=0x%x "
1308 	    "vector=0x%x\n", (void *)irqptr->airq_dip,
1309 	    irqptr->airq_mps_intr_index, irqno, irqptr->airq_vector));
1310 
1311 	while (irqptr) {
1312 		if (VIRTIRQ(irqindex, irqptr->airq_share_id) == irqno)
1313 			break;
1314 		irqptr = irqptr->airq_next;
1315 	}
1316 	irqptr->airq_share++;
1317 
1318 	mutex_exit(&airq_mutex);
1319 
1320 	/* return if it is not hardware interrupt */
1321 	if (irqptr->airq_mps_intr_index == RESERVE_INDEX)
1322 		return (PSM_SUCCESS);
1323 
1324 	/* Or if there are more interupts at a higher IPL */
1325 	if (ipl != max_ipl)
1326 		return (PSM_SUCCESS);
1327 
1328 	/*
1329 	 * if apic_picinit() has not been called yet, just return.
1330 	 * At the end of apic_picinit(), we will call setup_io_intr().
1331 	 */
1332 
1333 	if (!apic_picinit_called)
1334 		return (PSM_SUCCESS);
1335 
1336 	/*
1337 	 * Upgrade vector if max_ipl is not earlier ipl. If we cannot allocate,
1338 	 * return failure. Not very elegant, but then we hope the
1339 	 * machine will blow up with ...
1340 	 */
1341 	if (irqptr->airq_ipl != max_ipl &&
1342 	    !ioapic_mask_workaround[irqptr->airq_ioapicindex]) {
1343 
1344 		vector = apic_allocate_vector(max_ipl, irqindex, 1);
1345 		if (vector == 0) {
1346 			irqptr->airq_share--;
1347 			return (PSM_FAILURE);
1348 		}
1349 		irqptr = irqheadptr;
1350 		apic_mark_vector(irqptr->airq_vector, vector);
1351 		while (irqptr) {
1352 			irqptr->airq_vector = vector;
1353 			irqptr->airq_ipl = (uchar_t)max_ipl;
1354 			/*
1355 			 * reprogram irq being added and every one else
1356 			 * who is not in the UNINIT state
1357 			 */
1358 			if ((VIRTIRQ(irqindex, irqptr->airq_share_id) ==
1359 			    irqno) || (irqptr->airq_temp_cpu != IRQ_UNINIT)) {
1360 				apic_record_rdt_entry(irqptr, irqindex);
1361 
1362 				iflag = intr_clear();
1363 				lock_set(&apic_ioapic_lock);
1364 
1365 				(void) apic_setup_io_intr(irqptr, irqindex,
1366 				    B_FALSE);
1367 
1368 				lock_clear(&apic_ioapic_lock);
1369 				intr_restore(iflag);
1370 			}
1371 			irqptr = irqptr->airq_next;
1372 		}
1373 		return (PSM_SUCCESS);
1374 
1375 	} else if (irqptr->airq_ipl != max_ipl &&
1376 	    ioapic_mask_workaround[irqptr->airq_ioapicindex]) {
1377 		/*
1378 		 * We cannot upgrade the vector, but we can change
1379 		 * the IPL that this vector induces.
1380 		 *
1381 		 * Note that we subtract APIC_BASE_VECT from the vector
1382 		 * here because this array is used in apic_intr_enter
1383 		 * (no need to add APIC_BASE_VECT in that hot code
1384 		 * path since we can do it in the rarely-executed path
1385 		 * here).
1386 		 */
1387 		apic_ipls[irqptr->airq_vector - APIC_BASE_VECT] =
1388 		    (uchar_t)max_ipl;
1389 
1390 		irqptr = irqheadptr;
1391 		while (irqptr) {
1392 			irqptr->airq_ipl = (uchar_t)max_ipl;
1393 			irqptr = irqptr->airq_next;
1394 		}
1395 
1396 		return (PSM_SUCCESS);
1397 	}
1398 
1399 	ASSERT(irqptr);
1400 
1401 	iflag = intr_clear();
1402 	lock_set(&apic_ioapic_lock);
1403 
1404 	(void) apic_setup_io_intr(irqptr, irqindex, B_FALSE);
1405 
1406 	lock_clear(&apic_ioapic_lock);
1407 	intr_restore(iflag);
1408 
1409 	return (PSM_SUCCESS);
1410 }
1411 
1412 /*
1413  * Recompute mask bits for the given interrupt vector.
1414  * If there is no interrupt servicing routine for this
1415  * vector, this function should disable interrupt vector
1416  * from happening at all IPLs. If there are still
1417  * handlers using the given vector, this function should
1418  * disable the given vector from happening below the lowest
1419  * IPL of the remaining hadlers.
1420  */
1421 /*ARGSUSED*/
1422 int
1423 apic_delspl_common(int irqno, int ipl, int min_ipl, int max_ipl)
1424 {
1425 	uchar_t vector, bind_cpu;
1426 	int intin, irqindex;
1427 	int apic_ix;
1428 	apic_irq_t	*irqptr, *irqheadptr, *irqp;
1429 	ulong_t iflag;
1430 
1431 	mutex_enter(&airq_mutex);
1432 	irqindex = IRQINDEX(irqno);
1433 	irqptr = irqheadptr = apic_irq_table[irqindex];
1434 
1435 	DDI_INTR_IMPLDBG((CE_CONT, "apic_delspl: dip=0x%p type=%d irqno=0x%x "
1436 	    "vector=0x%x\n", (void *)irqptr->airq_dip,
1437 	    irqptr->airq_mps_intr_index, irqno, irqptr->airq_vector));
1438 
1439 	while (irqptr) {
1440 		if (VIRTIRQ(irqindex, irqptr->airq_share_id) == irqno)
1441 			break;
1442 		irqptr = irqptr->airq_next;
1443 	}
1444 	ASSERT(irqptr);
1445 
1446 	irqptr->airq_share--;
1447 
1448 	mutex_exit(&airq_mutex);
1449 
1450 	if (ipl < max_ipl)
1451 		return (PSM_SUCCESS);
1452 
1453 	/* return if it is not hardware interrupt */
1454 	if (irqptr->airq_mps_intr_index == RESERVE_INDEX)
1455 		return (PSM_SUCCESS);
1456 
1457 	if (!apic_picinit_called) {
1458 		/*
1459 		 * Clear irq_struct. If two devices shared an intpt
1460 		 * line & 1 unloaded before picinit, we are hosed. But, then
1461 		 * we hope the machine will ...
1462 		 */
1463 		irqptr->airq_mps_intr_index = FREE_INDEX;
1464 		irqptr->airq_temp_cpu = IRQ_UNINIT;
1465 		apic_free_vector(irqptr->airq_vector);
1466 		return (PSM_SUCCESS);
1467 	}
1468 	/*
1469 	 * Downgrade vector to new max_ipl if needed.If we cannot allocate,
1470 	 * use old IPL. Not very elegant, but then we hope ...
1471 	 */
1472 	if ((irqptr->airq_ipl != max_ipl) && (max_ipl != PSM_INVALID_IPL) &&
1473 	    !ioapic_mask_workaround[irqptr->airq_ioapicindex]) {
1474 		apic_irq_t	*irqp;
1475 		if (vector = apic_allocate_vector(max_ipl, irqno, 1)) {
1476 			apic_mark_vector(irqheadptr->airq_vector, vector);
1477 			irqp = irqheadptr;
1478 			while (irqp) {
1479 				irqp->airq_vector = vector;
1480 				irqp->airq_ipl = (uchar_t)max_ipl;
1481 				if (irqp->airq_temp_cpu != IRQ_UNINIT) {
1482 					apic_record_rdt_entry(irqp, irqindex);
1483 
1484 					iflag = intr_clear();
1485 					lock_set(&apic_ioapic_lock);
1486 
1487 					(void) apic_setup_io_intr(irqp,
1488 					    irqindex, B_FALSE);
1489 
1490 					lock_clear(&apic_ioapic_lock);
1491 					intr_restore(iflag);
1492 				}
1493 				irqp = irqp->airq_next;
1494 			}
1495 		}
1496 
1497 	} else if (irqptr->airq_ipl != max_ipl &&
1498 	    max_ipl != PSM_INVALID_IPL &&
1499 	    ioapic_mask_workaround[irqptr->airq_ioapicindex]) {
1500 
1501 	/*
1502 	 * We cannot downgrade the IPL of the vector below the vector's
1503 	 * hardware priority. If we did, it would be possible for a
1504 	 * higher-priority hardware vector to interrupt a CPU running at an IPL
1505 	 * lower than the hardware priority of the interrupting vector (but
1506 	 * higher than the soft IPL of this IRQ). When this happens, we would
1507 	 * then try to drop the IPL BELOW what it was (effectively dropping
1508 	 * below base_spl) which would be potentially catastrophic.
1509 	 *
1510 	 * (e.g. Suppose the hardware vector associated with this IRQ is 0x40
1511 	 * (hardware IPL of 4).  Further assume that the old IPL of this IRQ
1512 	 * was 4, but the new IPL is 1.  If we forced vector 0x40 to result in
1513 	 * an IPL of 1, it would be possible for the processor to be executing
1514 	 * at IPL 3 and for an interrupt to come in on vector 0x40, interrupting
1515 	 * the currently-executing ISR.  When apic_intr_enter consults
1516 	 * apic_irqs[], it will return 1, bringing the IPL of the CPU down to 1
1517 	 * so even though the processor was running at IPL 4, an IPL 1
1518 	 * interrupt will have interrupted it, which must not happen)).
1519 	 *
1520 	 * Effectively, this means that the hardware priority corresponding to
1521 	 * the IRQ's IPL (in apic_ipls[]) cannot be lower than the vector's
1522 	 * hardware priority.
1523 	 *
1524 	 * (In the above example, then, after removal of the IPL 4 device's
1525 	 * interrupt handler, the new IPL will continue to be 4 because the
1526 	 * hardware priority that IPL 1 implies is lower than the hardware
1527 	 * priority of the vector used.)
1528 	 */
1529 		/* apic_ipls is indexed by vector, starting at APIC_BASE_VECT */
1530 		const int apic_ipls_index = irqptr->airq_vector -
1531 		    APIC_BASE_VECT;
1532 		const int vect_inherent_hwpri = irqptr->airq_vector >>
1533 		    APIC_IPL_SHIFT;
1534 
1535 		/*
1536 		 * If there are still devices using this IRQ, determine the
1537 		 * new ipl to use.
1538 		 */
1539 		if (irqptr->airq_share) {
1540 			int vect_desired_hwpri, hwpri;
1541 
1542 			ASSERT(max_ipl < MAXIPL);
1543 			vect_desired_hwpri = apic_ipltopri[max_ipl] >>
1544 			    APIC_IPL_SHIFT;
1545 
1546 			/*
1547 			 * If the desired IPL's hardware priority is lower
1548 			 * than that of the vector, use the hardware priority
1549 			 * of the vector to determine the new IPL.
1550 			 */
1551 			hwpri = (vect_desired_hwpri < vect_inherent_hwpri) ?
1552 			    vect_inherent_hwpri : vect_desired_hwpri;
1553 
1554 			/*
1555 			 * Now, to get the right index for apic_vectortoipl,
1556 			 * we need to subtract APIC_BASE_VECT from the
1557 			 * hardware-vector-equivalent (in hwpri).  Since hwpri
1558 			 * is already shifted, we shift APIC_BASE_VECT before
1559 			 * doing the subtraction.
1560 			 */
1561 			hwpri -= (APIC_BASE_VECT >> APIC_IPL_SHIFT);
1562 
1563 			ASSERT(hwpri >= 0);
1564 			ASSERT(hwpri < MAXIPL);
1565 			max_ipl = apic_vectortoipl[hwpri];
1566 			apic_ipls[apic_ipls_index] = max_ipl;
1567 
1568 			irqp = irqheadptr;
1569 			while (irqp) {
1570 				irqp->airq_ipl = (uchar_t)max_ipl;
1571 				irqp = irqp->airq_next;
1572 			}
1573 		} else {
1574 			/*
1575 			 * No more devices on this IRQ, so reset this vector's
1576 			 * element in apic_ipls to the original IPL for this
1577 			 * vector
1578 			 */
1579 			apic_ipls[apic_ipls_index] =
1580 			    apic_vectortoipl[vect_inherent_hwpri];
1581 		}
1582 	}
1583 
1584 	if (irqptr->airq_share)
1585 		return (PSM_SUCCESS);
1586 
1587 	iflag = intr_clear();
1588 	lock_set(&apic_ioapic_lock);
1589 
1590 	if (irqptr->airq_mps_intr_index == MSI_INDEX) {
1591 		/*
1592 		 * Disable the MSI vector
1593 		 * Make sure we only disable on the last
1594 		 * of the multi-MSI support
1595 		 */
1596 		if (i_ddi_intr_get_current_nintrs(irqptr->airq_dip) == 1) {
1597 			apic_pci_msi_unconfigure(irqptr->airq_dip,
1598 			    DDI_INTR_TYPE_MSI, irqptr->airq_ioapicindex);
1599 
1600 			apic_pci_msi_disable_mode(irqptr->airq_dip,
1601 			    DDI_INTR_TYPE_MSI);
1602 		}
1603 	} else if (irqptr->airq_mps_intr_index == MSIX_INDEX) {
1604 		/*
1605 		 * Disable the MSI-X vector
1606 		 * needs to clear its mask and addr/data for each MSI-X
1607 		 */
1608 		apic_pci_msi_unconfigure(irqptr->airq_dip, DDI_INTR_TYPE_MSIX,
1609 		    irqptr->airq_origirq);
1610 		/*
1611 		 * Make sure we only disable on the last MSI-X
1612 		 */
1613 		if (i_ddi_intr_get_current_nintrs(irqptr->airq_dip) == 1) {
1614 			apic_pci_msi_disable_mode(irqptr->airq_dip,
1615 			    DDI_INTR_TYPE_MSIX);
1616 		}
1617 	} else {
1618 		/*
1619 		 * The assumption here is that this is safe, even for
1620 		 * systems with IOAPICs that suffer from the hardware
1621 		 * erratum because all devices have been quiesced before
1622 		 * they unregister their interrupt handlers.  If that
1623 		 * assumption turns out to be false, this mask operation
1624 		 * can induce the same erratum result we're trying to
1625 		 * avoid.
1626 		 */
1627 		apic_ix = irqptr->airq_ioapicindex;
1628 		intin = irqptr->airq_intin_no;
1629 		ioapic_write(apic_ix, APIC_RDT_CMD + 2 * intin, AV_MASK);
1630 	}
1631 
1632 	if (max_ipl == PSM_INVALID_IPL) {
1633 		ASSERT(irqheadptr == irqptr);
1634 		bind_cpu = irqptr->airq_temp_cpu;
1635 		if (((uchar_t)bind_cpu != IRQ_UNBOUND) &&
1636 		    ((uchar_t)bind_cpu != IRQ_UNINIT)) {
1637 			ASSERT((bind_cpu & ~IRQ_USER_BOUND) < apic_nproc);
1638 			if (bind_cpu & IRQ_USER_BOUND) {
1639 				/* If hardbound, temp_cpu == cpu */
1640 				bind_cpu &= ~IRQ_USER_BOUND;
1641 				apic_cpus[bind_cpu].aci_bound--;
1642 			} else
1643 				apic_cpus[bind_cpu].aci_temp_bound--;
1644 		}
1645 		irqptr->airq_temp_cpu = IRQ_UNINIT;
1646 		irqptr->airq_mps_intr_index = FREE_INDEX;
1647 		lock_clear(&apic_ioapic_lock);
1648 		intr_restore(iflag);
1649 		apic_free_vector(irqptr->airq_vector);
1650 		return (PSM_SUCCESS);
1651 	}
1652 	lock_clear(&apic_ioapic_lock);
1653 	intr_restore(iflag);
1654 
1655 	mutex_enter(&airq_mutex);
1656 	if ((irqptr == apic_irq_table[irqindex])) {
1657 		apic_irq_t	*oldirqptr;
1658 		/* Move valid irq entry to the head */
1659 		irqheadptr = oldirqptr = irqptr;
1660 		irqptr = irqptr->airq_next;
1661 		ASSERT(irqptr);
1662 		while (irqptr) {
1663 			if (irqptr->airq_mps_intr_index != FREE_INDEX)
1664 				break;
1665 			oldirqptr = irqptr;
1666 			irqptr = irqptr->airq_next;
1667 		}
1668 		/* remove all invalid ones from the beginning */
1669 		apic_irq_table[irqindex] = irqptr;
1670 		/*
1671 		 * and link them back after the head. The invalid ones
1672 		 * begin with irqheadptr and end at oldirqptr
1673 		 */
1674 		oldirqptr->airq_next = irqptr->airq_next;
1675 		irqptr->airq_next = irqheadptr;
1676 	}
1677 	mutex_exit(&airq_mutex);
1678 
1679 	irqptr->airq_temp_cpu = IRQ_UNINIT;
1680 	irqptr->airq_mps_intr_index = FREE_INDEX;
1681 
1682 	return (PSM_SUCCESS);
1683 }
1684 
1685 /*
1686  * apic_introp_xlate() replaces apic_translate_irq() and is
1687  * called only from apic_intr_ops().  With the new ADII framework,
1688  * the priority can no longer be retrieved through i_ddi_get_intrspec().
1689  * It has to be passed in from the caller.
1690  */
1691 int
1692 apic_introp_xlate(dev_info_t *dip, struct intrspec *ispec, int type)
1693 {
1694 	char dev_type[16];
1695 	int dev_len, pci_irq, newirq, bustype, devid, busid, i;
1696 	int irqno = ispec->intrspec_vec;
1697 	ddi_acc_handle_t cfg_handle;
1698 	uchar_t ipin;
1699 	struct apic_io_intr *intrp;
1700 	iflag_t intr_flag;
1701 	APIC_HEADER	*hp;
1702 	MADT_INTERRUPT_OVERRIDE	*isop;
1703 	apic_irq_t *airqp;
1704 	int parent_is_pci_or_pciex = 0;
1705 	int child_is_pciex = 0;
1706 
1707 	DDI_INTR_IMPLDBG((CE_CONT, "apic_introp_xlate: dip=0x%p name=%s "
1708 	    "type=%d irqno=0x%x\n", (void *)dip, ddi_get_name(dip), type,
1709 	    irqno));
1710 
1711 	dev_len = sizeof (dev_type);
1712 	if (ddi_getlongprop_buf(DDI_DEV_T_ANY, ddi_get_parent(dip),
1713 	    DDI_PROP_DONTPASS, "device_type", (caddr_t)dev_type,
1714 	    &dev_len) == DDI_PROP_SUCCESS) {
1715 		if ((strcmp(dev_type, "pci") == 0) ||
1716 		    (strcmp(dev_type, "pciex") == 0))
1717 			parent_is_pci_or_pciex = 1;
1718 	}
1719 
1720 	if (parent_is_pci_or_pciex && ddi_prop_get_int(DDI_DEV_T_ANY, dip,
1721 	    DDI_PROP_DONTPASS, "pcie-capid-pointer", PCI_CAP_NEXT_PTR_NULL) !=
1722 	    PCI_CAP_NEXT_PTR_NULL) {
1723 		child_is_pciex = 1;
1724 	}
1725 
1726 	if (DDI_INTR_IS_MSI_OR_MSIX(type)) {
1727 		if ((airqp = apic_find_irq(dip, ispec, type)) != NULL) {
1728 			airqp->airq_iflag.bustype =
1729 			    child_is_pciex ? BUS_PCIE : BUS_PCI;
1730 			return (apic_vector_to_irq[airqp->airq_vector]);
1731 		}
1732 		return (apic_setup_irq_table(dip, irqno, NULL, ispec,
1733 		    NULL, type));
1734 	}
1735 
1736 	bustype = 0;
1737 
1738 	/* check if we have already translated this irq */
1739 	mutex_enter(&airq_mutex);
1740 	newirq = apic_min_device_irq;
1741 	for (; newirq <= apic_max_device_irq; newirq++) {
1742 		airqp = apic_irq_table[newirq];
1743 		while (airqp) {
1744 			if ((airqp->airq_dip == dip) &&
1745 			    (airqp->airq_origirq == irqno) &&
1746 			    (airqp->airq_mps_intr_index != FREE_INDEX)) {
1747 
1748 				mutex_exit(&airq_mutex);
1749 				return (VIRTIRQ(newirq, airqp->airq_share_id));
1750 			}
1751 			airqp = airqp->airq_next;
1752 		}
1753 	}
1754 	mutex_exit(&airq_mutex);
1755 
1756 	if (apic_defconf)
1757 		goto defconf;
1758 
1759 	if ((dip == NULL) || (!apic_irq_translate && !apic_enable_acpi))
1760 		goto nonpci;
1761 
1762 	if (parent_is_pci_or_pciex) {
1763 		/* pci device */
1764 		if (acpica_get_bdf(dip, &busid, &devid, NULL) != 0)
1765 			goto nonpci;
1766 		if (busid == 0 && apic_pci_bus_total == 1)
1767 			busid = (int)apic_single_pci_busid;
1768 
1769 		if (pci_config_setup(dip, &cfg_handle) != DDI_SUCCESS)
1770 			goto nonpci;
1771 		ipin = pci_config_get8(cfg_handle, PCI_CONF_IPIN) - PCI_INTA;
1772 		pci_config_teardown(&cfg_handle);
1773 		if (apic_enable_acpi && !apic_use_acpi_madt_only) {
1774 			if (apic_acpi_translate_pci_irq(dip, busid, devid,
1775 			    ipin, &pci_irq, &intr_flag) != ACPI_PSM_SUCCESS)
1776 				goto nonpci;
1777 
1778 			intr_flag.bustype = child_is_pciex ? BUS_PCIE : BUS_PCI;
1779 			if ((newirq = apic_setup_irq_table(dip, pci_irq, NULL,
1780 			    ispec, &intr_flag, type)) == -1)
1781 				goto nonpci;
1782 			return (newirq);
1783 		} else {
1784 			pci_irq = ((devid & 0x1f) << 2) | (ipin & 0x3);
1785 			if ((intrp = apic_find_io_intr_w_busid(pci_irq, busid))
1786 			    == NULL) {
1787 				if ((pci_irq = apic_handle_pci_pci_bridge(dip,
1788 				    devid, ipin, &intrp)) == -1)
1789 					goto nonpci;
1790 			}
1791 			if ((newirq = apic_setup_irq_table(dip, pci_irq, intrp,
1792 			    ispec, NULL, type)) == -1)
1793 				goto nonpci;
1794 			return (newirq);
1795 		}
1796 	} else if (strcmp(dev_type, "isa") == 0)
1797 		bustype = BUS_ISA;
1798 	else if (strcmp(dev_type, "eisa") == 0)
1799 		bustype = BUS_EISA;
1800 
1801 nonpci:
1802 	if (apic_enable_acpi && !apic_use_acpi_madt_only) {
1803 		/* search iso entries first */
1804 		if (acpi_iso_cnt != 0) {
1805 			hp = (APIC_HEADER *)acpi_isop;
1806 			i = 0;
1807 			while (i < acpi_iso_cnt) {
1808 				if (hp->Type == APIC_XRUPT_OVERRIDE) {
1809 					isop = (MADT_INTERRUPT_OVERRIDE *)hp;
1810 					if (isop->Bus == 0 &&
1811 					    isop->Source == irqno) {
1812 						newirq = isop->Interrupt;
1813 						intr_flag.intr_po =
1814 						    isop->Polarity;
1815 						intr_flag.intr_el =
1816 						    isop->TriggerMode;
1817 						intr_flag.bustype = BUS_ISA;
1818 
1819 						return (apic_setup_irq_table(
1820 						    dip, newirq, NULL, ispec,
1821 						    &intr_flag, type));
1822 
1823 					}
1824 					i++;
1825 				}
1826 				hp = (APIC_HEADER *)(((char *)hp) +
1827 				    hp->Length);
1828 			}
1829 		}
1830 		intr_flag.intr_po = INTR_PO_ACTIVE_HIGH;
1831 		intr_flag.intr_el = INTR_EL_EDGE;
1832 		intr_flag.bustype = BUS_ISA;
1833 		return (apic_setup_irq_table(dip, irqno, NULL, ispec,
1834 		    &intr_flag, type));
1835 	} else {
1836 		if (bustype == 0)
1837 			bustype = eisa_level_intr_mask ? BUS_EISA : BUS_ISA;
1838 		for (i = 0; i < 2; i++) {
1839 			if (((busid = apic_find_bus_id(bustype)) != -1) &&
1840 			    ((intrp = apic_find_io_intr_w_busid(irqno, busid))
1841 			    != NULL)) {
1842 				if ((newirq = apic_setup_irq_table(dip, irqno,
1843 				    intrp, ispec, NULL, type)) != -1) {
1844 					return (newirq);
1845 				}
1846 				goto defconf;
1847 			}
1848 			bustype = (bustype == BUS_EISA) ? BUS_ISA : BUS_EISA;
1849 		}
1850 	}
1851 
1852 /* MPS default configuration */
1853 defconf:
1854 	newirq = apic_setup_irq_table(dip, irqno, NULL, ispec, NULL, type);
1855 	if (newirq == -1)
1856 		return (newirq);
1857 	ASSERT(IRQINDEX(newirq) == irqno);
1858 	ASSERT(apic_irq_table[irqno]);
1859 	return (newirq);
1860 }
1861 
1862 
1863 
1864 
1865 
1866 
1867 /*
1868  * On machines with PCI-PCI bridges, a device behind a PCI-PCI bridge
1869  * needs special handling.  We may need to chase up the device tree,
1870  * using the PCI-PCI Bridge specification's "rotating IPIN assumptions",
1871  * to find the IPIN at the root bus that relates to the IPIN on the
1872  * subsidiary bus (for ACPI or MP).  We may, however, have an entry
1873  * in the MP table or the ACPI namespace for this device itself.
1874  * We handle both cases in the search below.
1875  */
1876 /* this is the non-acpi version */
1877 static int
1878 apic_handle_pci_pci_bridge(dev_info_t *idip, int child_devno, int child_ipin,
1879 			struct apic_io_intr **intrp)
1880 {
1881 	dev_info_t *dipp, *dip;
1882 	int pci_irq;
1883 	ddi_acc_handle_t cfg_handle;
1884 	int bridge_devno, bridge_bus;
1885 	int ipin;
1886 
1887 	dip = idip;
1888 
1889 	/*CONSTCOND*/
1890 	while (1) {
1891 		if (((dipp = ddi_get_parent(dip)) == (dev_info_t *)NULL) ||
1892 		    (pci_config_setup(dipp, &cfg_handle) != DDI_SUCCESS))
1893 			return (-1);
1894 		if ((pci_config_get8(cfg_handle, PCI_CONF_BASCLASS) ==
1895 		    PCI_CLASS_BRIDGE) && (pci_config_get8(cfg_handle,
1896 		    PCI_CONF_SUBCLASS) == PCI_BRIDGE_PCI)) {
1897 			pci_config_teardown(&cfg_handle);
1898 			if (acpica_get_bdf(dipp, &bridge_bus, &bridge_devno,
1899 			    NULL) != 0)
1900 				return (-1);
1901 			/*
1902 			 * This is the rotating scheme documented in the
1903 			 * PCI-to-PCI spec.  If the PCI-to-PCI bridge is
1904 			 * behind another PCI-to-PCI bridge, then it needs
1905 			 * to keep ascending until an interrupt entry is
1906 			 * found or the root is reached.
1907 			 */
1908 			ipin = (child_devno + child_ipin) % PCI_INTD;
1909 				if (bridge_bus == 0 && apic_pci_bus_total == 1)
1910 					bridge_bus = (int)apic_single_pci_busid;
1911 				pci_irq = ((bridge_devno & 0x1f) << 2) |
1912 				    (ipin & 0x3);
1913 				if ((*intrp = apic_find_io_intr_w_busid(pci_irq,
1914 				    bridge_bus)) != NULL) {
1915 					return (pci_irq);
1916 				}
1917 			dip = dipp;
1918 			child_devno = bridge_devno;
1919 			child_ipin = ipin;
1920 		} else {
1921 			pci_config_teardown(&cfg_handle);
1922 			return (-1);
1923 		}
1924 	}
1925 	/*LINTED: function will not fall off the bottom */
1926 }
1927 
1928 
1929 
1930 
1931 static uchar_t
1932 acpi_find_ioapic(int irq)
1933 {
1934 	int i;
1935 
1936 	for (i = 0; i < apic_io_max; i++) {
1937 		if (irq >= apic_io_vectbase[i] && irq <= apic_io_vectend[i])
1938 			return (i);
1939 	}
1940 	return (0xFF);	/* shouldn't happen */
1941 }
1942 
1943 /*
1944  * See if two irqs are compatible for sharing a vector.
1945  * Currently we only support sharing of PCI devices.
1946  */
1947 static int
1948 acpi_intr_compatible(iflag_t iflag1, iflag_t iflag2)
1949 {
1950 	uint_t	level1, po1;
1951 	uint_t	level2, po2;
1952 
1953 	/* Assume active high by default */
1954 	po1 = 0;
1955 	po2 = 0;
1956 
1957 	if (iflag1.bustype != iflag2.bustype || iflag1.bustype != BUS_PCI)
1958 		return (0);
1959 
1960 	if (iflag1.intr_el == INTR_EL_CONFORM)
1961 		level1 = AV_LEVEL;
1962 	else
1963 		level1 = (iflag1.intr_el == INTR_EL_LEVEL) ? AV_LEVEL : 0;
1964 
1965 	if (level1 && ((iflag1.intr_po == INTR_PO_ACTIVE_LOW) ||
1966 	    (iflag1.intr_po == INTR_PO_CONFORM)))
1967 		po1 = AV_ACTIVE_LOW;
1968 
1969 	if (iflag2.intr_el == INTR_EL_CONFORM)
1970 		level2 = AV_LEVEL;
1971 	else
1972 		level2 = (iflag2.intr_el == INTR_EL_LEVEL) ? AV_LEVEL : 0;
1973 
1974 	if (level2 && ((iflag2.intr_po == INTR_PO_ACTIVE_LOW) ||
1975 	    (iflag2.intr_po == INTR_PO_CONFORM)))
1976 		po2 = AV_ACTIVE_LOW;
1977 
1978 	if ((level1 == level2) && (po1 == po2))
1979 		return (1);
1980 
1981 	return (0);
1982 }
1983 
1984 /*
1985  * Attempt to share vector with someone else
1986  */
1987 static int
1988 apic_share_vector(int irqno, iflag_t *intr_flagp, short intr_index, int ipl,
1989 	uchar_t ioapicindex, uchar_t ipin, apic_irq_t **irqptrp)
1990 {
1991 #ifdef DEBUG
1992 	apic_irq_t *tmpirqp = NULL;
1993 #endif /* DEBUG */
1994 	apic_irq_t *irqptr, dummyirq;
1995 	int	newirq, chosen_irq = -1, share = 127;
1996 	int	lowest, highest, i;
1997 	uchar_t	share_id;
1998 
1999 	DDI_INTR_IMPLDBG((CE_CONT, "apic_share_vector: irqno=0x%x "
2000 	    "intr_index=0x%x ipl=0x%x\n", irqno, intr_index, ipl));
2001 
2002 	highest = apic_ipltopri[ipl] + APIC_VECTOR_MASK;
2003 	lowest = apic_ipltopri[ipl-1] + APIC_VECTOR_PER_IPL;
2004 
2005 	if (highest < lowest) /* Both ipl and ipl-1 map to same pri */
2006 		lowest -= APIC_VECTOR_PER_IPL;
2007 	dummyirq.airq_mps_intr_index = intr_index;
2008 	dummyirq.airq_ioapicindex = ioapicindex;
2009 	dummyirq.airq_intin_no = ipin;
2010 	if (intr_flagp)
2011 		dummyirq.airq_iflag = *intr_flagp;
2012 	apic_record_rdt_entry(&dummyirq, irqno);
2013 	for (i = lowest; i <= highest; i++) {
2014 		newirq = apic_vector_to_irq[i];
2015 		if (newirq == APIC_RESV_IRQ)
2016 			continue;
2017 		irqptr = apic_irq_table[newirq];
2018 
2019 		if ((dummyirq.airq_rdt_entry & 0xFF00) !=
2020 		    (irqptr->airq_rdt_entry & 0xFF00))
2021 			/* not compatible */
2022 			continue;
2023 
2024 		if (irqptr->airq_share < share) {
2025 			share = irqptr->airq_share;
2026 			chosen_irq = newirq;
2027 		}
2028 	}
2029 	if (chosen_irq != -1) {
2030 		/*
2031 		 * Assign a share id which is free or which is larger
2032 		 * than the largest one.
2033 		 */
2034 		share_id = 1;
2035 		mutex_enter(&airq_mutex);
2036 		irqptr = apic_irq_table[chosen_irq];
2037 		while (irqptr) {
2038 			if (irqptr->airq_mps_intr_index == FREE_INDEX) {
2039 				share_id = irqptr->airq_share_id;
2040 				break;
2041 			}
2042 			if (share_id <= irqptr->airq_share_id)
2043 				share_id = irqptr->airq_share_id + 1;
2044 #ifdef DEBUG
2045 			tmpirqp = irqptr;
2046 #endif /* DEBUG */
2047 			irqptr = irqptr->airq_next;
2048 		}
2049 		if (!irqptr) {
2050 			irqptr = kmem_zalloc(sizeof (apic_irq_t), KM_SLEEP);
2051 			irqptr->airq_temp_cpu = IRQ_UNINIT;
2052 			irqptr->airq_next =
2053 			    apic_irq_table[chosen_irq]->airq_next;
2054 			apic_irq_table[chosen_irq]->airq_next = irqptr;
2055 #ifdef	DEBUG
2056 			tmpirqp = apic_irq_table[chosen_irq];
2057 #endif /* DEBUG */
2058 		}
2059 		irqptr->airq_mps_intr_index = intr_index;
2060 		irqptr->airq_ioapicindex = ioapicindex;
2061 		irqptr->airq_intin_no = ipin;
2062 		if (intr_flagp)
2063 			irqptr->airq_iflag = *intr_flagp;
2064 		irqptr->airq_vector = apic_irq_table[chosen_irq]->airq_vector;
2065 		irqptr->airq_share_id = share_id;
2066 		apic_record_rdt_entry(irqptr, irqno);
2067 		*irqptrp = irqptr;
2068 #ifdef	DEBUG
2069 		/* shuffle the pointers to test apic_delspl path */
2070 		if (tmpirqp) {
2071 			tmpirqp->airq_next = irqptr->airq_next;
2072 			irqptr->airq_next = apic_irq_table[chosen_irq];
2073 			apic_irq_table[chosen_irq] = irqptr;
2074 		}
2075 #endif /* DEBUG */
2076 		mutex_exit(&airq_mutex);
2077 		return (VIRTIRQ(chosen_irq, share_id));
2078 	}
2079 	return (-1);
2080 }
2081 
2082 /*
2083  *
2084  */
2085 static int
2086 apic_setup_irq_table(dev_info_t *dip, int irqno, struct apic_io_intr *intrp,
2087     struct intrspec *ispec, iflag_t *intr_flagp, int type)
2088 {
2089 	int origirq = ispec->intrspec_vec;
2090 	uchar_t ipl = ispec->intrspec_pri;
2091 	int	newirq, intr_index;
2092 	uchar_t	ipin, ioapic, ioapicindex, vector;
2093 	apic_irq_t *irqptr;
2094 	major_t	major;
2095 	dev_info_t	*sdip;
2096 
2097 	DDI_INTR_IMPLDBG((CE_CONT, "apic_setup_irq_table: dip=0x%p type=%d "
2098 	    "irqno=0x%x origirq=0x%x\n", (void *)dip, type, irqno, origirq));
2099 
2100 	ASSERT(ispec != NULL);
2101 
2102 	major =  (dip != NULL) ? ddi_name_to_major(ddi_get_name(dip)) : 0;
2103 
2104 	if (DDI_INTR_IS_MSI_OR_MSIX(type)) {
2105 		/* MSI/X doesn't need to setup ioapic stuffs */
2106 		ioapicindex = 0xff;
2107 		ioapic = 0xff;
2108 		ipin = (uchar_t)0xff;
2109 		intr_index = (type == DDI_INTR_TYPE_MSI) ? MSI_INDEX :
2110 		    MSIX_INDEX;
2111 		mutex_enter(&airq_mutex);
2112 		if ((irqno = apic_allocate_irq(apic_first_avail_irq)) == -1) {
2113 			mutex_exit(&airq_mutex);
2114 			/* need an irq for MSI/X to index into autovect[] */
2115 			cmn_err(CE_WARN, "No interrupt irq: %s instance %d",
2116 			    ddi_get_name(dip), ddi_get_instance(dip));
2117 			return (-1);
2118 		}
2119 		mutex_exit(&airq_mutex);
2120 
2121 	} else if (intrp != NULL) {
2122 		intr_index = (int)(intrp - apic_io_intrp);
2123 		ioapic = intrp->intr_destid;
2124 		ipin = intrp->intr_destintin;
2125 		/* Find ioapicindex. If destid was ALL, we will exit with 0. */
2126 		for (ioapicindex = apic_io_max - 1; ioapicindex; ioapicindex--)
2127 			if (apic_io_id[ioapicindex] == ioapic)
2128 				break;
2129 		ASSERT((ioapic == apic_io_id[ioapicindex]) ||
2130 		    (ioapic == INTR_ALL_APIC));
2131 
2132 		/* check whether this intin# has been used by another irqno */
2133 		if ((newirq = apic_find_intin(ioapicindex, ipin)) != -1) {
2134 			return (newirq);
2135 		}
2136 
2137 	} else if (intr_flagp != NULL) {
2138 		/* ACPI case */
2139 		intr_index = ACPI_INDEX;
2140 		ioapicindex = acpi_find_ioapic(irqno);
2141 		ASSERT(ioapicindex != 0xFF);
2142 		ioapic = apic_io_id[ioapicindex];
2143 		ipin = irqno - apic_io_vectbase[ioapicindex];
2144 		if (apic_irq_table[irqno] &&
2145 		    apic_irq_table[irqno]->airq_mps_intr_index == ACPI_INDEX) {
2146 			ASSERT(apic_irq_table[irqno]->airq_intin_no == ipin &&
2147 			    apic_irq_table[irqno]->airq_ioapicindex ==
2148 			    ioapicindex);
2149 			return (irqno);
2150 		}
2151 
2152 	} else {
2153 		/* default configuration */
2154 		ioapicindex = 0;
2155 		ioapic = apic_io_id[ioapicindex];
2156 		ipin = (uchar_t)irqno;
2157 		intr_index = DEFAULT_INDEX;
2158 	}
2159 
2160 	if (ispec == NULL) {
2161 		APIC_VERBOSE_IOAPIC((CE_WARN, "No intrspec for irqno = %x\n",
2162 		    irqno));
2163 	} else if ((vector = apic_allocate_vector(ipl, irqno, 0)) == 0) {
2164 		if ((newirq = apic_share_vector(irqno, intr_flagp, intr_index,
2165 		    ipl, ioapicindex, ipin, &irqptr)) != -1) {
2166 			irqptr->airq_ipl = ipl;
2167 			irqptr->airq_origirq = (uchar_t)origirq;
2168 			irqptr->airq_dip = dip;
2169 			irqptr->airq_major = major;
2170 			sdip = apic_irq_table[IRQINDEX(newirq)]->airq_dip;
2171 			/* This is OK to do really */
2172 			if (sdip == NULL) {
2173 				cmn_err(CE_WARN, "Sharing vectors: %s"
2174 				    " instance %d and SCI",
2175 				    ddi_get_name(dip), ddi_get_instance(dip));
2176 			} else {
2177 				cmn_err(CE_WARN, "Sharing vectors: %s"
2178 				    " instance %d and %s instance %d",
2179 				    ddi_get_name(sdip), ddi_get_instance(sdip),
2180 				    ddi_get_name(dip), ddi_get_instance(dip));
2181 			}
2182 			return (newirq);
2183 		}
2184 		/* try high priority allocation now  that share has failed */
2185 		if ((vector = apic_allocate_vector(ipl, irqno, 1)) == 0) {
2186 			cmn_err(CE_WARN, "No interrupt vector: %s instance %d",
2187 			    ddi_get_name(dip), ddi_get_instance(dip));
2188 			return (-1);
2189 		}
2190 	}
2191 
2192 	mutex_enter(&airq_mutex);
2193 	if (apic_irq_table[irqno] == NULL) {
2194 		irqptr = kmem_zalloc(sizeof (apic_irq_t), KM_SLEEP);
2195 		irqptr->airq_temp_cpu = IRQ_UNINIT;
2196 		apic_irq_table[irqno] = irqptr;
2197 	} else {
2198 		irqptr = apic_irq_table[irqno];
2199 		if (irqptr->airq_mps_intr_index != FREE_INDEX) {
2200 			/*
2201 			 * The slot is used by another irqno, so allocate
2202 			 * a free irqno for this interrupt
2203 			 */
2204 			newirq = apic_allocate_irq(apic_first_avail_irq);
2205 			if (newirq == -1) {
2206 				mutex_exit(&airq_mutex);
2207 				return (-1);
2208 			}
2209 			irqno = newirq;
2210 			irqptr = apic_irq_table[irqno];
2211 			if (irqptr == NULL) {
2212 				irqptr = kmem_zalloc(sizeof (apic_irq_t),
2213 				    KM_SLEEP);
2214 				irqptr->airq_temp_cpu = IRQ_UNINIT;
2215 				apic_irq_table[irqno] = irqptr;
2216 			}
2217 			vector = apic_modify_vector(vector, newirq);
2218 		}
2219 	}
2220 	apic_max_device_irq = max(irqno, apic_max_device_irq);
2221 	apic_min_device_irq = min(irqno, apic_min_device_irq);
2222 	mutex_exit(&airq_mutex);
2223 	irqptr->airq_ioapicindex = ioapicindex;
2224 	irqptr->airq_intin_no = ipin;
2225 	irqptr->airq_ipl = ipl;
2226 	irqptr->airq_vector = vector;
2227 	irqptr->airq_origirq = (uchar_t)origirq;
2228 	irqptr->airq_share_id = 0;
2229 	irqptr->airq_mps_intr_index = (short)intr_index;
2230 	irqptr->airq_dip = dip;
2231 	irqptr->airq_major = major;
2232 	irqptr->airq_cpu = apic_bind_intr(dip, irqno, ioapic, ipin);
2233 	if (intr_flagp)
2234 		irqptr->airq_iflag = *intr_flagp;
2235 
2236 	if (!DDI_INTR_IS_MSI_OR_MSIX(type)) {
2237 		/* setup I/O APIC entry for non-MSI/X interrupts */
2238 		apic_record_rdt_entry(irqptr, irqno);
2239 	}
2240 	return (irqno);
2241 }
2242 
2243 /*
2244  * return the cpu to which this intr should be bound.
2245  * Check properties or any other mechanism to see if user wants it
2246  * bound to a specific CPU. If so, return the cpu id with high bit set.
2247  * If not, use the policy to choose a cpu and return the id.
2248  */
2249 uchar_t
2250 apic_bind_intr(dev_info_t *dip, int irq, uchar_t ioapicid, uchar_t intin)
2251 {
2252 	int	instance, instno, prop_len, bind_cpu, count;
2253 	uint_t	i, rc;
2254 	uchar_t	cpu;
2255 	major_t	major;
2256 	char	*name, *drv_name, *prop_val, *cptr;
2257 	char	prop_name[32];
2258 
2259 
2260 	if (apic_intr_policy == INTR_LOWEST_PRIORITY)
2261 		return (IRQ_UNBOUND);
2262 
2263 	drv_name = NULL;
2264 	rc = DDI_PROP_NOT_FOUND;
2265 	major = (major_t)-1;
2266 	if (dip != NULL) {
2267 		name = ddi_get_name(dip);
2268 		major = ddi_name_to_major(name);
2269 		drv_name = ddi_major_to_name(major);
2270 		instance = ddi_get_instance(dip);
2271 		if (apic_intr_policy == INTR_ROUND_ROBIN_WITH_AFFINITY) {
2272 			i = apic_min_device_irq;
2273 			for (; i <= apic_max_device_irq; i++) {
2274 
2275 				if ((i == irq) || (apic_irq_table[i] == NULL) ||
2276 				    (apic_irq_table[i]->airq_mps_intr_index
2277 				    == FREE_INDEX))
2278 					continue;
2279 
2280 				if ((apic_irq_table[i]->airq_major == major) &&
2281 				    (!(apic_irq_table[i]->airq_cpu &
2282 				    IRQ_USER_BOUND))) {
2283 
2284 					cpu = apic_irq_table[i]->airq_cpu;
2285 
2286 					cmn_err(CE_CONT,
2287 					    "!%s: %s (%s) instance #%d "
2288 					    "vector 0x%x ioapic 0x%x "
2289 					    "intin 0x%x is bound to cpu %d\n",
2290 					    psm_name,
2291 					    name, drv_name, instance, irq,
2292 					    ioapicid, intin, cpu);
2293 					return (cpu);
2294 				}
2295 			}
2296 		}
2297 		/*
2298 		 * search for "drvname"_intpt_bind_cpus property first, the
2299 		 * syntax of the property should be "a[,b,c,...]" where
2300 		 * instance 0 binds to cpu a, instance 1 binds to cpu b,
2301 		 * instance 3 binds to cpu c...
2302 		 * ddi_getlongprop() will search /option first, then /
2303 		 * if "drvname"_intpt_bind_cpus doesn't exist, then find
2304 		 * intpt_bind_cpus property.  The syntax is the same, and
2305 		 * it applies to all the devices if its "drvname" specific
2306 		 * property doesn't exist
2307 		 */
2308 		(void) strcpy(prop_name, drv_name);
2309 		(void) strcat(prop_name, "_intpt_bind_cpus");
2310 		rc = ddi_getlongprop(DDI_DEV_T_ANY, dip, 0, prop_name,
2311 		    (caddr_t)&prop_val, &prop_len);
2312 		if (rc != DDI_PROP_SUCCESS) {
2313 			rc = ddi_getlongprop(DDI_DEV_T_ANY, dip, 0,
2314 			    "intpt_bind_cpus", (caddr_t)&prop_val, &prop_len);
2315 		}
2316 	}
2317 	if (rc == DDI_PROP_SUCCESS) {
2318 		for (i = count = 0; i < (prop_len - 1); i++)
2319 			if (prop_val[i] == ',')
2320 				count++;
2321 		if (prop_val[i-1] != ',')
2322 			count++;
2323 		/*
2324 		 * if somehow the binding instances defined in the
2325 		 * property are not enough for this instno., then
2326 		 * reuse the pattern for the next instance until
2327 		 * it reaches the requested instno
2328 		 */
2329 		instno = instance % count;
2330 		i = 0;
2331 		cptr = prop_val;
2332 		while (i < instno)
2333 			if (*cptr++ == ',')
2334 				i++;
2335 		bind_cpu = stoi(&cptr);
2336 		kmem_free(prop_val, prop_len);
2337 		/* if specific cpu is bogus, then default to cpu 0 */
2338 		if (bind_cpu >= apic_nproc) {
2339 			cmn_err(CE_WARN, "%s: %s=%s: CPU %d not present",
2340 			    psm_name, prop_name, prop_val, bind_cpu);
2341 			bind_cpu = 0;
2342 		} else {
2343 			/* indicate that we are bound at user request */
2344 			bind_cpu |= IRQ_USER_BOUND;
2345 		}
2346 		/*
2347 		 * no need to check apic_cpus[].aci_status, if specific cpu is
2348 		 * not up, then post_cpu_start will handle it.
2349 		 */
2350 	} else {
2351 		bind_cpu = apic_next_bind_cpu++;
2352 		if (bind_cpu >= apic_nproc) {
2353 			apic_next_bind_cpu = 1;
2354 			bind_cpu = 0;
2355 		}
2356 	}
2357 	if (drv_name != NULL)
2358 		cmn_err(CE_CONT, "!%s: %s (%s) instance %d "
2359 		    "vector 0x%x ioapic 0x%x intin 0x%x is bound to cpu %d\n",
2360 		    psm_name, name, drv_name, instance,
2361 		    irq, ioapicid, intin, bind_cpu & ~IRQ_USER_BOUND);
2362 	else
2363 		cmn_err(CE_CONT, "!%s: "
2364 		    "vector 0x%x ioapic 0x%x intin 0x%x is bound to cpu %d\n",
2365 		    psm_name, irq, ioapicid, intin, bind_cpu & ~IRQ_USER_BOUND);
2366 
2367 	return ((uchar_t)bind_cpu);
2368 }
2369 
2370 static struct apic_io_intr *
2371 apic_find_io_intr_w_busid(int irqno, int busid)
2372 {
2373 	struct	apic_io_intr	*intrp;
2374 
2375 	/*
2376 	 * It can have more than 1 entry with same source bus IRQ,
2377 	 * but unique with the source bus id
2378 	 */
2379 	intrp = apic_io_intrp;
2380 	if (intrp != NULL) {
2381 		while (intrp->intr_entry == APIC_IO_INTR_ENTRY) {
2382 			if (intrp->intr_irq == irqno &&
2383 			    intrp->intr_busid == busid &&
2384 			    intrp->intr_type == IO_INTR_INT)
2385 				return (intrp);
2386 			intrp++;
2387 		}
2388 	}
2389 	APIC_VERBOSE_IOAPIC((CE_NOTE, "Did not find io intr for irqno:"
2390 	    "busid %x:%x\n", irqno, busid));
2391 	return ((struct apic_io_intr *)NULL);
2392 }
2393 
2394 
2395 struct mps_bus_info {
2396 	char	*bus_name;
2397 	int	bus_id;
2398 } bus_info_array[] = {
2399 	"ISA ", BUS_ISA,
2400 	"PCI ", BUS_PCI,
2401 	"EISA ", BUS_EISA,
2402 	"XPRESS", BUS_XPRESS,
2403 	"PCMCIA", BUS_PCMCIA,
2404 	"VL ", BUS_VL,
2405 	"CBUS ", BUS_CBUS,
2406 	"CBUSII", BUS_CBUSII,
2407 	"FUTURE", BUS_FUTURE,
2408 	"INTERN", BUS_INTERN,
2409 	"MBI ", BUS_MBI,
2410 	"MBII ", BUS_MBII,
2411 	"MPI ", BUS_MPI,
2412 	"MPSA ", BUS_MPSA,
2413 	"NUBUS ", BUS_NUBUS,
2414 	"TC ", BUS_TC,
2415 	"VME ", BUS_VME,
2416 	"PCI-E ", BUS_PCIE
2417 };
2418 
2419 static int
2420 apic_find_bus_type(char *bus)
2421 {
2422 	int	i = 0;
2423 
2424 	for (; i < sizeof (bus_info_array)/sizeof (struct mps_bus_info); i++)
2425 		if (strncmp(bus, bus_info_array[i].bus_name,
2426 		    strlen(bus_info_array[i].bus_name)) == 0)
2427 			return (bus_info_array[i].bus_id);
2428 	APIC_VERBOSE_IOAPIC((CE_WARN, "Did not find bus type for bus %s", bus));
2429 	return (0);
2430 }
2431 
2432 static int
2433 apic_find_bus(int busid)
2434 {
2435 	struct	apic_bus	*busp;
2436 
2437 	busp = apic_busp;
2438 	while (busp->bus_entry == APIC_BUS_ENTRY) {
2439 		if (busp->bus_id == busid)
2440 			return (apic_find_bus_type((char *)&busp->bus_str1));
2441 		busp++;
2442 	}
2443 	APIC_VERBOSE_IOAPIC((CE_WARN, "Did not find bus for bus id %x", busid));
2444 	return (0);
2445 }
2446 
2447 static int
2448 apic_find_bus_id(int bustype)
2449 {
2450 	struct	apic_bus	*busp;
2451 
2452 	busp = apic_busp;
2453 	while (busp->bus_entry == APIC_BUS_ENTRY) {
2454 		if (apic_find_bus_type((char *)&busp->bus_str1) == bustype)
2455 			return (busp->bus_id);
2456 		busp++;
2457 	}
2458 	APIC_VERBOSE_IOAPIC((CE_WARN, "Did not find bus id for bustype %x",
2459 	    bustype));
2460 	return (-1);
2461 }
2462 
2463 /*
2464  * Check if a particular irq need to be reserved for any io_intr
2465  */
2466 static struct apic_io_intr *
2467 apic_find_io_intr(int irqno)
2468 {
2469 	struct	apic_io_intr	*intrp;
2470 
2471 	intrp = apic_io_intrp;
2472 	if (intrp != NULL) {
2473 		while (intrp->intr_entry == APIC_IO_INTR_ENTRY) {
2474 			if (intrp->intr_irq == irqno &&
2475 			    intrp->intr_type == IO_INTR_INT)
2476 				return (intrp);
2477 			intrp++;
2478 		}
2479 	}
2480 	return ((struct apic_io_intr *)NULL);
2481 }
2482 
2483 /*
2484  * Check if the given ioapicindex intin combination has already been assigned
2485  * an irq. If so return irqno. Else -1
2486  */
2487 static int
2488 apic_find_intin(uchar_t ioapic, uchar_t intin)
2489 {
2490 	apic_irq_t *irqptr;
2491 	int	i;
2492 
2493 	/* find ioapic and intin in the apic_irq_table[] and return the index */
2494 	for (i = apic_min_device_irq; i <= apic_max_device_irq; i++) {
2495 		irqptr = apic_irq_table[i];
2496 		while (irqptr) {
2497 			if ((irqptr->airq_mps_intr_index >= 0) &&
2498 			    (irqptr->airq_intin_no == intin) &&
2499 			    (irqptr->airq_ioapicindex == ioapic)) {
2500 				APIC_VERBOSE_IOAPIC((CE_NOTE, "!Found irq "
2501 				    "entry for ioapic:intin %x:%x "
2502 				    "shared interrupts ?", ioapic, intin));
2503 				return (i);
2504 			}
2505 			irqptr = irqptr->airq_next;
2506 		}
2507 	}
2508 	return (-1);
2509 }
2510 
2511 int
2512 apic_allocate_irq(int irq)
2513 {
2514 	int	freeirq, i;
2515 
2516 	if ((freeirq = apic_find_free_irq(irq, (APIC_RESV_IRQ - 1))) == -1)
2517 		if ((freeirq = apic_find_free_irq(APIC_FIRST_FREE_IRQ,
2518 		    (irq - 1))) == -1) {
2519 			/*
2520 			 * if BIOS really defines every single irq in the mps
2521 			 * table, then don't worry about conflicting with
2522 			 * them, just use any free slot in apic_irq_table
2523 			 */
2524 			for (i = APIC_FIRST_FREE_IRQ; i < APIC_RESV_IRQ; i++) {
2525 				if ((apic_irq_table[i] == NULL) ||
2526 				    apic_irq_table[i]->airq_mps_intr_index ==
2527 				    FREE_INDEX) {
2528 				freeirq = i;
2529 				break;
2530 			}
2531 		}
2532 		if (freeirq == -1) {
2533 			/* This shouldn't happen, but just in case */
2534 			cmn_err(CE_WARN, "%s: NO available IRQ", psm_name);
2535 			return (-1);
2536 		}
2537 	}
2538 	if (apic_irq_table[freeirq] == NULL) {
2539 		apic_irq_table[freeirq] =
2540 		    kmem_zalloc(sizeof (apic_irq_t), KM_NOSLEEP);
2541 		if (apic_irq_table[freeirq] == NULL) {
2542 			cmn_err(CE_WARN, "%s: NO memory to allocate IRQ",
2543 			    psm_name);
2544 			return (-1);
2545 		}
2546 		apic_irq_table[freeirq]->airq_mps_intr_index = FREE_INDEX;
2547 	}
2548 	return (freeirq);
2549 }
2550 
2551 static int
2552 apic_find_free_irq(int start, int end)
2553 {
2554 	int	i;
2555 
2556 	for (i = start; i <= end; i++)
2557 		/* Check if any I/O entry needs this IRQ */
2558 		if (apic_find_io_intr(i) == NULL) {
2559 			/* Then see if it is free */
2560 			if ((apic_irq_table[i] == NULL) ||
2561 			    (apic_irq_table[i]->airq_mps_intr_index ==
2562 			    FREE_INDEX)) {
2563 				return (i);
2564 			}
2565 		}
2566 	return (-1);
2567 }
2568 
2569 
2570 /*
2571  * Mark vector as being in the process of being deleted. Interrupts
2572  * may still come in on some CPU. The moment an interrupt comes with
2573  * the new vector, we know we can free the old one. Called only from
2574  * addspl and delspl with interrupts disabled. Because an interrupt
2575  * can be shared, but no interrupt from either device may come in,
2576  * we also use a timeout mechanism, which we arbitrarily set to
2577  * apic_revector_timeout microseconds.
2578  */
2579 static void
2580 apic_mark_vector(uchar_t oldvector, uchar_t newvector)
2581 {
2582 	ulong_t iflag;
2583 
2584 	iflag = intr_clear();
2585 	lock_set(&apic_revector_lock);
2586 	if (!apic_oldvec_to_newvec) {
2587 		apic_oldvec_to_newvec =
2588 		    kmem_zalloc(sizeof (newvector) * APIC_MAX_VECTOR * 2,
2589 		    KM_NOSLEEP);
2590 
2591 		if (!apic_oldvec_to_newvec) {
2592 			/*
2593 			 * This failure is not catastrophic.
2594 			 * But, the oldvec will never be freed.
2595 			 */
2596 			apic_error |= APIC_ERR_MARK_VECTOR_FAIL;
2597 			lock_clear(&apic_revector_lock);
2598 			intr_restore(iflag);
2599 			return;
2600 		}
2601 		apic_newvec_to_oldvec = &apic_oldvec_to_newvec[APIC_MAX_VECTOR];
2602 	}
2603 
2604 	/* See if we already did this for drivers which do double addintrs */
2605 	if (apic_oldvec_to_newvec[oldvector] != newvector) {
2606 		apic_oldvec_to_newvec[oldvector] = newvector;
2607 		apic_newvec_to_oldvec[newvector] = oldvector;
2608 		apic_revector_pending++;
2609 	}
2610 	lock_clear(&apic_revector_lock);
2611 	intr_restore(iflag);
2612 	(void) timeout(apic_xlate_vector_free_timeout_handler,
2613 	    (void *)(uintptr_t)oldvector, drv_usectohz(apic_revector_timeout));
2614 }
2615 
2616 /*
2617  * xlate_vector is called from intr_enter if revector_pending is set.
2618  * It will xlate it if needed and mark the old vector as free.
2619  */
2620 uchar_t
2621 apic_xlate_vector(uchar_t vector)
2622 {
2623 	uchar_t	newvector, oldvector = 0;
2624 
2625 	lock_set(&apic_revector_lock);
2626 	/* Do we really need to do this ? */
2627 	if (!apic_revector_pending) {
2628 		lock_clear(&apic_revector_lock);
2629 		return (vector);
2630 	}
2631 	if ((newvector = apic_oldvec_to_newvec[vector]) != 0)
2632 		oldvector = vector;
2633 	else {
2634 		/*
2635 		 * The incoming vector is new . See if a stale entry is
2636 		 * remaining
2637 		 */
2638 		if ((oldvector = apic_newvec_to_oldvec[vector]) != 0)
2639 			newvector = vector;
2640 	}
2641 
2642 	if (oldvector) {
2643 		apic_revector_pending--;
2644 		apic_oldvec_to_newvec[oldvector] = 0;
2645 		apic_newvec_to_oldvec[newvector] = 0;
2646 		apic_free_vector(oldvector);
2647 		lock_clear(&apic_revector_lock);
2648 		/* There could have been more than one reprogramming! */
2649 		return (apic_xlate_vector(newvector));
2650 	}
2651 	lock_clear(&apic_revector_lock);
2652 	return (vector);
2653 }
2654 
2655 void
2656 apic_xlate_vector_free_timeout_handler(void *arg)
2657 {
2658 	ulong_t iflag;
2659 	uchar_t oldvector, newvector;
2660 
2661 	oldvector = (uchar_t)(uintptr_t)arg;
2662 	iflag = intr_clear();
2663 	lock_set(&apic_revector_lock);
2664 	if ((newvector = apic_oldvec_to_newvec[oldvector]) != 0) {
2665 		apic_free_vector(oldvector);
2666 		apic_oldvec_to_newvec[oldvector] = 0;
2667 		apic_newvec_to_oldvec[newvector] = 0;
2668 		apic_revector_pending--;
2669 	}
2670 
2671 	lock_clear(&apic_revector_lock);
2672 	intr_restore(iflag);
2673 }
2674 
2675 
2676 /*
2677  * compute the polarity, trigger mode and vector for programming into
2678  * the I/O apic and record in airq_rdt_entry.
2679  */
2680 static void
2681 apic_record_rdt_entry(apic_irq_t *irqptr, int irq)
2682 {
2683 	int	ioapicindex, bus_type, vector;
2684 	short	intr_index;
2685 	uint_t	level, po, io_po;
2686 	struct apic_io_intr *iointrp;
2687 
2688 	intr_index = irqptr->airq_mps_intr_index;
2689 	DDI_INTR_IMPLDBG((CE_CONT, "apic_record_rdt_entry: intr_index=%d "
2690 	    "irq = 0x%x dip = 0x%p vector = 0x%x\n", intr_index, irq,
2691 	    (void *)irqptr->airq_dip, irqptr->airq_vector));
2692 
2693 	if (intr_index == RESERVE_INDEX) {
2694 		apic_error |= APIC_ERR_INVALID_INDEX;
2695 		return;
2696 	} else if (APIC_IS_MSI_OR_MSIX_INDEX(intr_index)) {
2697 		return;
2698 	}
2699 
2700 	vector = irqptr->airq_vector;
2701 	ioapicindex = irqptr->airq_ioapicindex;
2702 	/* Assume edge triggered by default */
2703 	level = 0;
2704 	/* Assume active high by default */
2705 	po = 0;
2706 
2707 	if (intr_index == DEFAULT_INDEX || intr_index == FREE_INDEX) {
2708 		ASSERT(irq < 16);
2709 		if (eisa_level_intr_mask & (1 << irq))
2710 			level = AV_LEVEL;
2711 		if (intr_index == FREE_INDEX && apic_defconf == 0)
2712 			apic_error |= APIC_ERR_INVALID_INDEX;
2713 	} else if (intr_index == ACPI_INDEX) {
2714 		bus_type = irqptr->airq_iflag.bustype;
2715 		if (irqptr->airq_iflag.intr_el == INTR_EL_CONFORM) {
2716 			if (bus_type == BUS_PCI)
2717 				level = AV_LEVEL;
2718 		} else
2719 			level = (irqptr->airq_iflag.intr_el == INTR_EL_LEVEL) ?
2720 			    AV_LEVEL : 0;
2721 		if (level &&
2722 		    ((irqptr->airq_iflag.intr_po == INTR_PO_ACTIVE_LOW) ||
2723 		    (irqptr->airq_iflag.intr_po == INTR_PO_CONFORM &&
2724 		    bus_type == BUS_PCI)))
2725 			po = AV_ACTIVE_LOW;
2726 	} else {
2727 		iointrp = apic_io_intrp + intr_index;
2728 		bus_type = apic_find_bus(iointrp->intr_busid);
2729 		if (iointrp->intr_el == INTR_EL_CONFORM) {
2730 			if ((irq < 16) && (eisa_level_intr_mask & (1 << irq)))
2731 				level = AV_LEVEL;
2732 			else if (bus_type == BUS_PCI)
2733 				level = AV_LEVEL;
2734 		} else
2735 			level = (iointrp->intr_el == INTR_EL_LEVEL) ?
2736 			    AV_LEVEL : 0;
2737 		if (level && ((iointrp->intr_po == INTR_PO_ACTIVE_LOW) ||
2738 		    (iointrp->intr_po == INTR_PO_CONFORM &&
2739 		    bus_type == BUS_PCI)))
2740 			po = AV_ACTIVE_LOW;
2741 	}
2742 	if (level)
2743 		apic_level_intr[irq] = 1;
2744 	/*
2745 	 * The 82489DX External APIC cannot do active low polarity interrupts.
2746 	 */
2747 	if (po && (apic_io_ver[ioapicindex] != IOAPIC_VER_82489DX))
2748 		io_po = po;
2749 	else
2750 		io_po = 0;
2751 
2752 	if (apic_verbose & APIC_VERBOSE_IOAPIC_FLAG)
2753 		printf("setio: ioapic=%x intin=%x level=%x po=%x vector=%x\n",
2754 		    ioapicindex, irqptr->airq_intin_no, level, io_po, vector);
2755 
2756 	irqptr->airq_rdt_entry = level|io_po|vector;
2757 }
2758 
2759 /*
2760  * Bind interrupt corresponding to irq_ptr to bind_cpu.
2761  * Must be called with interrupts disabled and apic_ioapic_lock held
2762  */
2763 int
2764 apic_rebind(apic_irq_t *irq_ptr, int bind_cpu,
2765     struct ioapic_reprogram_data *drep)
2766 {
2767 	int			ioapicindex, intin_no;
2768 	uchar_t			airq_temp_cpu;
2769 	apic_cpus_info_t	*cpu_infop;
2770 	uint32_t		rdt_entry;
2771 	int			which_irq;
2772 
2773 	which_irq = apic_vector_to_irq[irq_ptr->airq_vector];
2774 
2775 	intin_no = irq_ptr->airq_intin_no;
2776 	ioapicindex = irq_ptr->airq_ioapicindex;
2777 	airq_temp_cpu = irq_ptr->airq_temp_cpu;
2778 	if (airq_temp_cpu != IRQ_UNINIT && airq_temp_cpu != IRQ_UNBOUND) {
2779 		if (airq_temp_cpu & IRQ_USER_BOUND)
2780 			/* Mask off high bit so it can be used as array index */
2781 			airq_temp_cpu &= ~IRQ_USER_BOUND;
2782 
2783 		ASSERT(airq_temp_cpu < apic_nproc);
2784 	}
2785 
2786 	/*
2787 	 * Can't bind to a CPU that's not accepting interrupts:
2788 	 */
2789 	cpu_infop = &apic_cpus[bind_cpu & ~IRQ_USER_BOUND];
2790 	if (!(cpu_infop->aci_status & APIC_CPU_INTR_ENABLE))
2791 		return (1);
2792 
2793 	/*
2794 	 * If we are about to change the interrupt vector for this interrupt,
2795 	 * and this interrupt is level-triggered, attached to an IOAPIC,
2796 	 * has been delivered to a CPU and that CPU has not handled it
2797 	 * yet, we cannot reprogram the IOAPIC now.
2798 	 */
2799 	if (!APIC_IS_MSI_OR_MSIX_INDEX(irq_ptr->airq_mps_intr_index)) {
2800 
2801 		rdt_entry = READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapicindex,
2802 		    intin_no);
2803 
2804 		if ((irq_ptr->airq_vector != RDT_VECTOR(rdt_entry)) &&
2805 		    apic_check_stuck_interrupt(irq_ptr, airq_temp_cpu,
2806 		    bind_cpu, ioapicindex, intin_no, which_irq, drep) != 0) {
2807 
2808 			return (0);
2809 		}
2810 
2811 		/*
2812 		 * NOTE: We do not unmask the RDT here, as an interrupt MAY
2813 		 * still come in before we have a chance to reprogram it below.
2814 		 * The reprogramming below will simultaneously change and
2815 		 * unmask the RDT entry.
2816 		 */
2817 
2818 		if ((uchar_t)bind_cpu == IRQ_UNBOUND) {
2819 			rdt_entry = AV_LDEST | AV_LOPRI |
2820 			    irq_ptr->airq_rdt_entry;
2821 
2822 			/* Write the RDT entry -- no specific CPU binding */
2823 			WRITE_IOAPIC_RDT_ENTRY_HIGH_DWORD(ioapicindex, intin_no,
2824 			    AV_TOALL);
2825 
2826 			if (airq_temp_cpu != IRQ_UNINIT && airq_temp_cpu !=
2827 			    IRQ_UNBOUND)
2828 				apic_cpus[airq_temp_cpu].aci_temp_bound--;
2829 
2830 			/*
2831 			 * Write the vector, trigger, and polarity portion of
2832 			 * the RDT
2833 			 */
2834 			WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapicindex, intin_no,
2835 			    rdt_entry);
2836 
2837 			irq_ptr->airq_temp_cpu = IRQ_UNBOUND;
2838 			return (0);
2839 		}
2840 	}
2841 
2842 	if (bind_cpu & IRQ_USER_BOUND) {
2843 		cpu_infop->aci_bound++;
2844 	} else {
2845 		cpu_infop->aci_temp_bound++;
2846 	}
2847 	ASSERT((bind_cpu & ~IRQ_USER_BOUND) < apic_nproc);
2848 	if (!APIC_IS_MSI_OR_MSIX_INDEX(irq_ptr->airq_mps_intr_index)) {
2849 		/* Write the RDT entry -- bind to a specific CPU: */
2850 		WRITE_IOAPIC_RDT_ENTRY_HIGH_DWORD(ioapicindex, intin_no,
2851 		    cpu_infop->aci_local_id << APIC_ID_BIT_OFFSET);
2852 	}
2853 	if ((airq_temp_cpu != IRQ_UNBOUND) && (airq_temp_cpu != IRQ_UNINIT)) {
2854 		apic_cpus[airq_temp_cpu].aci_temp_bound--;
2855 	}
2856 	if (!APIC_IS_MSI_OR_MSIX_INDEX(irq_ptr->airq_mps_intr_index)) {
2857 
2858 		rdt_entry = AV_PDEST | AV_FIXED | irq_ptr->airq_rdt_entry;
2859 
2860 		/* Write the vector, trigger, and polarity portion of the RDT */
2861 		WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapicindex, intin_no,
2862 		    rdt_entry);
2863 
2864 	} else {
2865 		int type = (irq_ptr->airq_mps_intr_index == MSI_INDEX) ?
2866 		    DDI_INTR_TYPE_MSI : DDI_INTR_TYPE_MSIX;
2867 		if (type == DDI_INTR_TYPE_MSI) {
2868 			if (irq_ptr->airq_ioapicindex ==
2869 			    irq_ptr->airq_origirq) {
2870 				/* first one */
2871 				DDI_INTR_IMPLDBG((CE_CONT, "apic_rebind: call "
2872 				    "apic_pci_msi_enable_vector\n"));
2873 				apic_pci_msi_enable_vector(irq_ptr->airq_dip,
2874 				    type, which_irq, irq_ptr->airq_vector,
2875 				    irq_ptr->airq_intin_no,
2876 				    cpu_infop->aci_local_id);
2877 			}
2878 			if ((irq_ptr->airq_ioapicindex +
2879 			    irq_ptr->airq_intin_no - 1) ==
2880 			    irq_ptr->airq_origirq) { /* last one */
2881 				DDI_INTR_IMPLDBG((CE_CONT, "apic_rebind: call "
2882 				    "apic_pci_msi_enable_mode\n"));
2883 				apic_pci_msi_enable_mode(irq_ptr->airq_dip,
2884 				    type, which_irq);
2885 			}
2886 		} else { /* MSI-X */
2887 			apic_pci_msi_enable_vector(irq_ptr->airq_dip, type,
2888 			    irq_ptr->airq_origirq, irq_ptr->airq_vector, 1,
2889 			    cpu_infop->aci_local_id);
2890 			apic_pci_msi_enable_mode(irq_ptr->airq_dip, type,
2891 			    irq_ptr->airq_origirq);
2892 		}
2893 	}
2894 	irq_ptr->airq_temp_cpu = (uchar_t)bind_cpu;
2895 	apic_redist_cpu_skip &= ~(1 << (bind_cpu & ~IRQ_USER_BOUND));
2896 	return (0);
2897 }
2898 
2899 static void
2900 apic_last_ditch_clear_remote_irr(int ioapic_ix, int intin_no)
2901 {
2902 	if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix, intin_no)
2903 	    & AV_REMOTE_IRR) != 0) {
2904 		/*
2905 		 * Trying to clear the bit through normal
2906 		 * channels has failed.  So as a last-ditch
2907 		 * effort, try to set the trigger mode to
2908 		 * edge, then to level.  This has been
2909 		 * observed to work on many systems.
2910 		 */
2911 		WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
2912 		    intin_no,
2913 		    READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
2914 		    intin_no) & ~AV_LEVEL);
2915 
2916 		WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
2917 		    intin_no,
2918 		    READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
2919 		    intin_no) | AV_LEVEL);
2920 
2921 		/*
2922 		 * If the bit's STILL set, this interrupt may
2923 		 * be hosed.
2924 		 */
2925 		if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
2926 		    intin_no) & AV_REMOTE_IRR) != 0) {
2927 
2928 			prom_printf("%s: Remote IRR still "
2929 			    "not clear for IOAPIC %d intin %d.\n"
2930 			    "\tInterrupts to this pin may cease "
2931 			    "functioning.\n", psm_name, ioapic_ix,
2932 			    intin_no);
2933 #ifdef DEBUG
2934 			apic_last_ditch_reprogram_failures++;
2935 #endif
2936 		}
2937 	}
2938 }
2939 
2940 /*
2941  * This function is protected by apic_ioapic_lock coupled with the
2942  * fact that interrupts are disabled.
2943  */
2944 static void
2945 delete_defer_repro_ent(int which_irq)
2946 {
2947 	ASSERT(which_irq >= 0);
2948 	ASSERT(which_irq <= 255);
2949 
2950 	if (apic_reprogram_info[which_irq].done)
2951 		return;
2952 
2953 	apic_reprogram_info[which_irq].done = B_TRUE;
2954 
2955 #ifdef DEBUG
2956 	apic_defer_repro_total_retries +=
2957 	    apic_reprogram_info[which_irq].tries;
2958 
2959 	apic_defer_repro_successes++;
2960 #endif
2961 
2962 	if (--apic_reprogram_outstanding == 0) {
2963 
2964 		setlvlx = psm_intr_exit_fn();
2965 	}
2966 }
2967 
2968 
2969 /*
2970  * Interrupts must be disabled during this function to prevent
2971  * self-deadlock.  Interrupts are disabled because this function
2972  * is called from apic_check_stuck_interrupt(), which is called
2973  * from apic_rebind(), which requires its caller to disable interrupts.
2974  */
2975 static void
2976 add_defer_repro_ent(apic_irq_t *irq_ptr, int which_irq, int new_bind_cpu)
2977 {
2978 	ASSERT(which_irq >= 0);
2979 	ASSERT(which_irq <= 255);
2980 
2981 	/*
2982 	 * On the off-chance that there's already a deferred
2983 	 * reprogramming on this irq, check, and if so, just update the
2984 	 * CPU and irq pointer to which the interrupt is targeted, then return.
2985 	 */
2986 	if (!apic_reprogram_info[which_irq].done) {
2987 		apic_reprogram_info[which_irq].bindcpu = new_bind_cpu;
2988 		apic_reprogram_info[which_irq].irqp = irq_ptr;
2989 		return;
2990 	}
2991 
2992 	apic_reprogram_info[which_irq].irqp = irq_ptr;
2993 	apic_reprogram_info[which_irq].bindcpu = new_bind_cpu;
2994 	apic_reprogram_info[which_irq].tries = 0;
2995 	/*
2996 	 * This must be the last thing set, since we're not
2997 	 * grabbing any locks, apic_try_deferred_reprogram() will
2998 	 * make its decision about using this entry iff done
2999 	 * is false.
3000 	 */
3001 	apic_reprogram_info[which_irq].done = B_FALSE;
3002 
3003 	/*
3004 	 * If there were previously no deferred reprogrammings, change
3005 	 * setlvlx to call apic_try_deferred_reprogram()
3006 	 */
3007 	if (++apic_reprogram_outstanding == 1) {
3008 
3009 		setlvlx = apic_try_deferred_reprogram;
3010 	}
3011 }
3012 
3013 static void
3014 apic_try_deferred_reprogram(int prev_ipl, int irq)
3015 {
3016 	int reproirq, iflag;
3017 	struct ioapic_reprogram_data *drep;
3018 
3019 	(*psm_intr_exit_fn())(prev_ipl, irq);
3020 
3021 	if (!lock_try(&apic_defer_reprogram_lock)) {
3022 		return;
3023 	}
3024 
3025 	/*
3026 	 * Acquire the apic_ioapic_lock so that any other operations that
3027 	 * may affect the apic_reprogram_info state are serialized.
3028 	 * It's still possible for the last deferred reprogramming to clear
3029 	 * between the time we entered this function and the time we get to
3030 	 * the for loop below.  In that case, *setlvlx will have been set
3031 	 * back to *_intr_exit and drep will be NULL. (There's no way to
3032 	 * stop that from happening -- we would need to grab a lock before
3033 	 * calling *setlvlx, which is neither realistic nor prudent).
3034 	 */
3035 	iflag = intr_clear();
3036 	lock_set(&apic_ioapic_lock);
3037 
3038 	/*
3039 	 * For each deferred RDT entry, try to reprogram it now.  Note that
3040 	 * there is no lock acquisition to read apic_reprogram_info because
3041 	 * '.done' is set only after the other fields in the structure are set.
3042 	 */
3043 
3044 	drep = NULL;
3045 	for (reproirq = 0; reproirq <= APIC_MAX_VECTOR; reproirq++) {
3046 		if (apic_reprogram_info[reproirq].done == B_FALSE) {
3047 			drep = &apic_reprogram_info[reproirq];
3048 			break;
3049 		}
3050 	}
3051 
3052 	/*
3053 	 * Either we found a deferred action to perform, or
3054 	 * we entered this function spuriously, after *setlvlx
3055 	 * was restored to point to *_intr_exit.  Any other
3056 	 * permutation is invalid.
3057 	 */
3058 	ASSERT(drep != NULL || *setlvlx == psm_intr_exit_fn());
3059 
3060 	/*
3061 	 * Though we can't really do anything about errors
3062 	 * at this point, keep track of them for reporting.
3063 	 * Note that it is very possible for apic_setup_io_intr
3064 	 * to re-register this very timeout if the Remote IRR bit
3065 	 * has not yet cleared.
3066 	 */
3067 
3068 #ifdef DEBUG
3069 	if (drep != NULL) {
3070 		if (apic_setup_io_intr(drep, reproirq, B_TRUE) != 0) {
3071 			apic_deferred_setup_failures++;
3072 		}
3073 	} else {
3074 		apic_deferred_spurious_enters++;
3075 	}
3076 #else
3077 	if (drep != NULL)
3078 		(void) apic_setup_io_intr(drep, reproirq, B_TRUE);
3079 #endif
3080 
3081 	lock_clear(&apic_ioapic_lock);
3082 	intr_restore(iflag);
3083 
3084 	lock_clear(&apic_defer_reprogram_lock);
3085 }
3086 
3087 static void
3088 apic_ioapic_wait_pending_clear(int ioapic_ix, int intin_no)
3089 {
3090 	int waited;
3091 
3092 	/*
3093 	 * Wait for the delivery pending bit to clear.
3094 	 */
3095 	if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix, intin_no) &
3096 	    (AV_LEVEL|AV_PENDING)) == (AV_LEVEL|AV_PENDING)) {
3097 
3098 		/*
3099 		 * If we're still waiting on the delivery of this interrupt,
3100 		 * continue to wait here until it is delivered (this should be
3101 		 * a very small amount of time, but include a timeout just in
3102 		 * case).
3103 		 */
3104 		for (waited = 0; waited < apic_max_reps_clear_pending;
3105 		    waited++) {
3106 			if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
3107 			    intin_no) & AV_PENDING) == 0) {
3108 				break;
3109 			}
3110 		}
3111 	}
3112 }
3113 
3114 
3115 /*
3116  * Checks to see if the IOAPIC interrupt entry specified has its Remote IRR
3117  * bit set.  Calls functions that modify the function that setlvlx points to,
3118  * so that the reprogramming can be retried very shortly.
3119  *
3120  * This function will mask the RDT entry if the interrupt is level-triggered.
3121  * (The caller is responsible for unmasking the RDT entry.)
3122  *
3123  * Returns non-zero if the caller should defer IOAPIC reprogramming.
3124  */
3125 static int
3126 apic_check_stuck_interrupt(apic_irq_t *irq_ptr, int old_bind_cpu,
3127     int new_bind_cpu, int ioapic_ix, int intin_no, int which_irq,
3128     struct ioapic_reprogram_data *drep)
3129 {
3130 	int32_t			rdt_entry;
3131 	int			waited;
3132 	int			reps = 0;
3133 
3134 	/*
3135 	 * Wait for the delivery pending bit to clear.
3136 	 */
3137 	do {
3138 		++reps;
3139 
3140 		apic_ioapic_wait_pending_clear(ioapic_ix, intin_no);
3141 
3142 		/*
3143 		 * Mask the RDT entry, but only if it's a level-triggered
3144 		 * interrupt
3145 		 */
3146 		rdt_entry = READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
3147 		    intin_no);
3148 		if ((rdt_entry & (AV_LEVEL|AV_MASK)) == AV_LEVEL) {
3149 
3150 			/* Mask it */
3151 			WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix, intin_no,
3152 			    AV_MASK | rdt_entry);
3153 		}
3154 
3155 		if ((rdt_entry & AV_LEVEL) == AV_LEVEL) {
3156 			/*
3157 			 * If there was a race and an interrupt was injected
3158 			 * just before we masked, check for that case here.
3159 			 * Then, unmask the RDT entry and try again.  If we're
3160 			 * on our last try, don't unmask (because we want the
3161 			 * RDT entry to remain masked for the rest of the
3162 			 * function).
3163 			 */
3164 			rdt_entry = READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
3165 			    intin_no);
3166 			if ((rdt_entry & AV_PENDING) &&
3167 			    (reps < apic_max_reps_clear_pending)) {
3168 				/* Unmask it */
3169 				WRITE_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
3170 				    intin_no, rdt_entry & ~AV_MASK);
3171 			}
3172 		}
3173 
3174 	} while ((rdt_entry & AV_PENDING) &&
3175 	    (reps < apic_max_reps_clear_pending));
3176 
3177 #ifdef DEBUG
3178 		if (rdt_entry & AV_PENDING)
3179 			apic_intr_deliver_timeouts++;
3180 #endif
3181 
3182 	/*
3183 	 * If the remote IRR bit is set, then the interrupt has been sent
3184 	 * to a CPU for processing.  We have no choice but to wait for
3185 	 * that CPU to process the interrupt, at which point the remote IRR
3186 	 * bit will be cleared.
3187 	 */
3188 	if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix, intin_no) &
3189 	    (AV_LEVEL|AV_REMOTE_IRR)) == (AV_LEVEL|AV_REMOTE_IRR)) {
3190 
3191 		/*
3192 		 * If the CPU that this RDT is bound to is NOT the current
3193 		 * CPU, wait until that CPU handles the interrupt and ACKs
3194 		 * it.  If this interrupt is not bound to any CPU (that is,
3195 		 * if it's bound to the logical destination of "anyone"), it
3196 		 * may have been delivered to the current CPU so handle that
3197 		 * case by deferring the reprogramming (below).
3198 		 */
3199 		if ((old_bind_cpu != IRQ_UNBOUND) &&
3200 		    (old_bind_cpu != IRQ_UNINIT) &&
3201 		    (old_bind_cpu != psm_get_cpu_id())) {
3202 			for (waited = 0; waited < apic_max_reps_clear_pending;
3203 			    waited++) {
3204 				if ((READ_IOAPIC_RDT_ENTRY_LOW_DWORD(ioapic_ix,
3205 				    intin_no) & AV_REMOTE_IRR) == 0) {
3206 
3207 					delete_defer_repro_ent(which_irq);
3208 
3209 					/* Remote IRR has cleared! */
3210 					return (0);
3211 				}
3212 			}
3213 		}
3214 
3215 		/*
3216 		 * If we waited and the Remote IRR bit is still not cleared,
3217 		 * AND if we've invoked the timeout APIC_REPROGRAM_MAX_TIMEOUTS
3218 		 * times for this interrupt, try the last-ditch workaround:
3219 		 */
3220 		if (drep && drep->tries >= APIC_REPROGRAM_MAX_TRIES) {
3221 
3222 			apic_last_ditch_clear_remote_irr(ioapic_ix, intin_no);
3223 
3224 			/* Mark this one as reprogrammed: */
3225 			delete_defer_repro_ent(which_irq);
3226 
3227 			return (0);
3228 		} else {
3229 #ifdef DEBUG
3230 			apic_intr_deferrals++;
3231 #endif
3232 
3233 			/*
3234 			 * If waiting for the Remote IRR bit (above) didn't
3235 			 * allow it to clear, defer the reprogramming.
3236 			 * Add a new deferred-programming entry if the
3237 			 * caller passed a NULL one (and update the existing one
3238 			 * in case anything changed).
3239 			 */
3240 			add_defer_repro_ent(irq_ptr, which_irq, new_bind_cpu);
3241 			if (drep)
3242 				drep->tries++;
3243 
3244 			/* Inform caller to defer IOAPIC programming: */
3245 			return (1);
3246 		}
3247 
3248 	}
3249 
3250 	/* Remote IRR is clear */
3251 	delete_defer_repro_ent(which_irq);
3252 
3253 	return (0);
3254 }
3255 
3256 /*
3257  * Called to migrate all interrupts at an irq to another cpu.
3258  * Must be called with interrupts disabled and apic_ioapic_lock held
3259  */
3260 int
3261 apic_rebind_all(apic_irq_t *irq_ptr, int bind_cpu)
3262 {
3263 	apic_irq_t	*irqptr = irq_ptr;
3264 	int		retval = 0;
3265 
3266 	while (irqptr) {
3267 		if (irqptr->airq_temp_cpu != IRQ_UNINIT)
3268 			retval |= apic_rebind(irqptr, bind_cpu, NULL);
3269 		irqptr = irqptr->airq_next;
3270 	}
3271 
3272 	return (retval);
3273 }
3274 
3275 /*
3276  * apic_intr_redistribute does all the messy computations for identifying
3277  * which interrupt to move to which CPU. Currently we do just one interrupt
3278  * at a time. This reduces the time we spent doing all this within clock
3279  * interrupt. When it is done in idle, we could do more than 1.
3280  * First we find the most busy and the most free CPU (time in ISR only)
3281  * skipping those CPUs that has been identified as being ineligible (cpu_skip)
3282  * Then we look for IRQs which are closest to the difference between the
3283  * most busy CPU and the average ISR load. We try to find one whose load
3284  * is less than difference.If none exists, then we chose one larger than the
3285  * difference, provided it does not make the most idle CPU worse than the
3286  * most busy one. In the end, we clear all the busy fields for CPUs. For
3287  * IRQs, they are cleared as they are scanned.
3288  */
3289 void
3290 apic_intr_redistribute()
3291 {
3292 	int busiest_cpu, most_free_cpu;
3293 	int cpu_free, cpu_busy, max_busy, min_busy;
3294 	int min_free, diff;
3295 	int average_busy, cpus_online;
3296 	int i, busy, iflag;
3297 	apic_cpus_info_t *cpu_infop;
3298 	apic_irq_t *min_busy_irq = NULL;
3299 	apic_irq_t *max_busy_irq = NULL;
3300 
3301 	busiest_cpu = most_free_cpu = -1;
3302 	cpu_free = cpu_busy = max_busy = average_busy = 0;
3303 	min_free = apic_sample_factor_redistribution;
3304 	cpus_online = 0;
3305 	/*
3306 	 * Below we will check for CPU_INTR_ENABLE, bound, temp_bound, temp_cpu
3307 	 * without ioapic_lock. That is OK as we are just doing statistical
3308 	 * sampling anyway and any inaccuracy now will get corrected next time
3309 	 * The call to rebind which actually changes things will make sure
3310 	 * we are consistent.
3311 	 */
3312 	for (i = 0; i < apic_nproc; i++) {
3313 		if (!(apic_redist_cpu_skip & (1 << i)) &&
3314 		    (apic_cpus[i].aci_status & APIC_CPU_INTR_ENABLE)) {
3315 
3316 			cpu_infop = &apic_cpus[i];
3317 			/*
3318 			 * If no unbound interrupts or only 1 total on this
3319 			 * CPU, skip
3320 			 */
3321 			if (!cpu_infop->aci_temp_bound ||
3322 			    (cpu_infop->aci_bound + cpu_infop->aci_temp_bound)
3323 			    == 1) {
3324 				apic_redist_cpu_skip |= 1 << i;
3325 				continue;
3326 			}
3327 
3328 			busy = cpu_infop->aci_busy;
3329 			average_busy += busy;
3330 			cpus_online++;
3331 			if (max_busy < busy) {
3332 				max_busy = busy;
3333 				busiest_cpu = i;
3334 			}
3335 			if (min_free > busy) {
3336 				min_free = busy;
3337 				most_free_cpu = i;
3338 			}
3339 			if (busy > apic_int_busy_mark) {
3340 				cpu_busy |= 1 << i;
3341 			} else {
3342 				if (busy < apic_int_free_mark)
3343 					cpu_free |= 1 << i;
3344 			}
3345 		}
3346 	}
3347 	if ((cpu_busy && cpu_free) ||
3348 	    (max_busy >= (min_free + apic_diff_for_redistribution))) {
3349 
3350 		apic_num_imbalance++;
3351 #ifdef	DEBUG
3352 		if (apic_verbose & APIC_VERBOSE_IOAPIC_FLAG) {
3353 			prom_printf(
3354 			    "redistribute busy=%x free=%x max=%x min=%x",
3355 			    cpu_busy, cpu_free, max_busy, min_free);
3356 		}
3357 #endif /* DEBUG */
3358 
3359 
3360 		average_busy /= cpus_online;
3361 
3362 		diff = max_busy - average_busy;
3363 		min_busy = max_busy; /* start with the max possible value */
3364 		max_busy = 0;
3365 		min_busy_irq = max_busy_irq = NULL;
3366 		i = apic_min_device_irq;
3367 		for (; i < apic_max_device_irq; i++) {
3368 			apic_irq_t *irq_ptr;
3369 			/* Change to linked list per CPU ? */
3370 			if ((irq_ptr = apic_irq_table[i]) == NULL)
3371 				continue;
3372 			/* Check for irq_busy & decide which one to move */
3373 			/* Also zero them for next round */
3374 			if ((irq_ptr->airq_temp_cpu == busiest_cpu) &&
3375 			    irq_ptr->airq_busy) {
3376 				if (irq_ptr->airq_busy < diff) {
3377 					/*
3378 					 * Check for least busy CPU,
3379 					 * best fit or what ?
3380 					 */
3381 					if (max_busy < irq_ptr->airq_busy) {
3382 						/*
3383 						 * Most busy within the
3384 						 * required differential
3385 						 */
3386 						max_busy = irq_ptr->airq_busy;
3387 						max_busy_irq = irq_ptr;
3388 					}
3389 				} else {
3390 					if (min_busy > irq_ptr->airq_busy) {
3391 						/*
3392 						 * least busy, but more than
3393 						 * the reqd diff
3394 						 */
3395 						if (min_busy <
3396 						    (diff + average_busy -
3397 						    min_free)) {
3398 							/*
3399 							 * Making sure new cpu
3400 							 * will not end up
3401 							 * worse
3402 							 */
3403 							min_busy =
3404 							    irq_ptr->airq_busy;
3405 
3406 							min_busy_irq = irq_ptr;
3407 						}
3408 					}
3409 				}
3410 			}
3411 			irq_ptr->airq_busy = 0;
3412 		}
3413 
3414 		if (max_busy_irq != NULL) {
3415 #ifdef	DEBUG
3416 			if (apic_verbose & APIC_VERBOSE_IOAPIC_FLAG) {
3417 				prom_printf("rebinding %x to %x",
3418 				    max_busy_irq->airq_vector, most_free_cpu);
3419 			}
3420 #endif /* DEBUG */
3421 			iflag = intr_clear();
3422 			if (lock_try(&apic_ioapic_lock)) {
3423 				if (apic_rebind_all(max_busy_irq,
3424 				    most_free_cpu) == 0) {
3425 					/* Make change permenant */
3426 					max_busy_irq->airq_cpu =
3427 					    (uchar_t)most_free_cpu;
3428 				}
3429 				lock_clear(&apic_ioapic_lock);
3430 			}
3431 			intr_restore(iflag);
3432 
3433 		} else if (min_busy_irq != NULL) {
3434 #ifdef	DEBUG
3435 			if (apic_verbose & APIC_VERBOSE_IOAPIC_FLAG) {
3436 				prom_printf("rebinding %x to %x",
3437 				    min_busy_irq->airq_vector, most_free_cpu);
3438 			}
3439 #endif /* DEBUG */
3440 
3441 			iflag = intr_clear();
3442 			if (lock_try(&apic_ioapic_lock)) {
3443 				if (apic_rebind_all(min_busy_irq,
3444 				    most_free_cpu) == 0) {
3445 					/* Make change permenant */
3446 					min_busy_irq->airq_cpu =
3447 					    (uchar_t)most_free_cpu;
3448 				}
3449 				lock_clear(&apic_ioapic_lock);
3450 			}
3451 			intr_restore(iflag);
3452 
3453 		} else {
3454 			if (cpu_busy != (1 << busiest_cpu)) {
3455 				apic_redist_cpu_skip |= 1 << busiest_cpu;
3456 				/*
3457 				 * We leave cpu_skip set so that next time we
3458 				 * can choose another cpu
3459 				 */
3460 			}
3461 		}
3462 		apic_num_rebind++;
3463 	} else {
3464 		/*
3465 		 * found nothing. Could be that we skipped over valid CPUs
3466 		 * or we have balanced everything. If we had a variable
3467 		 * ticks_for_redistribution, it could be increased here.
3468 		 * apic_int_busy, int_free etc would also need to be
3469 		 * changed.
3470 		 */
3471 		if (apic_redist_cpu_skip)
3472 			apic_redist_cpu_skip = 0;
3473 	}
3474 	for (i = 0; i < apic_nproc; i++) {
3475 		apic_cpus[i].aci_busy = 0;
3476 	}
3477 }
3478 
3479 void
3480 apic_cleanup_busy()
3481 {
3482 	int i;
3483 	apic_irq_t *irq_ptr;
3484 
3485 	for (i = 0; i < apic_nproc; i++) {
3486 		apic_cpus[i].aci_busy = 0;
3487 	}
3488 
3489 	for (i = apic_min_device_irq; i < apic_max_device_irq; i++) {
3490 		if ((irq_ptr = apic_irq_table[i]) != NULL)
3491 			irq_ptr->airq_busy = 0;
3492 	}
3493 }
3494 
3495 
3496 static int
3497 apic_acpi_translate_pci_irq(dev_info_t *dip, int busid, int devid,
3498     int ipin, int *pci_irqp, iflag_t *intr_flagp)
3499 {
3500 
3501 	int status;
3502 	acpi_psm_lnk_t acpipsmlnk;
3503 
3504 	if ((status = acpi_get_irq_cache_ent(busid, devid, ipin, pci_irqp,
3505 	    intr_flagp)) == ACPI_PSM_SUCCESS) {
3506 		APIC_VERBOSE_IRQ((CE_CONT, "!%s: Found irqno %d "
3507 		    "from cache for device %s, instance #%d\n", psm_name,
3508 		    *pci_irqp, ddi_get_name(dip), ddi_get_instance(dip)));
3509 		return (status);
3510 	}
3511 
3512 	bzero(&acpipsmlnk, sizeof (acpi_psm_lnk_t));
3513 
3514 	if ((status = acpi_translate_pci_irq(dip, ipin, pci_irqp, intr_flagp,
3515 	    &acpipsmlnk)) == ACPI_PSM_FAILURE) {
3516 		APIC_VERBOSE_IRQ((CE_WARN, "%s: "
3517 		    " acpi_translate_pci_irq failed for device %s, instance"
3518 		    " #%d", psm_name, ddi_get_name(dip),
3519 		    ddi_get_instance(dip)));
3520 		return (status);
3521 	}
3522 
3523 	if (status == ACPI_PSM_PARTIAL && acpipsmlnk.lnkobj != NULL) {
3524 		status = apic_acpi_irq_configure(&acpipsmlnk, dip, pci_irqp,
3525 		    intr_flagp);
3526 		if (status != ACPI_PSM_SUCCESS) {
3527 			status = acpi_get_current_irq_resource(&acpipsmlnk,
3528 			    pci_irqp, intr_flagp);
3529 		}
3530 	}
3531 
3532 	if (status == ACPI_PSM_SUCCESS) {
3533 		acpi_new_irq_cache_ent(busid, devid, ipin, *pci_irqp,
3534 		    intr_flagp, &acpipsmlnk);
3535 
3536 		APIC_VERBOSE_IRQ((CE_CONT, "%s: [ACPI] "
3537 		    "new irq %d for device %s, instance #%d\n", psm_name,
3538 		    *pci_irqp, ddi_get_name(dip), ddi_get_instance(dip)));
3539 	}
3540 
3541 	return (status);
3542 }
3543 
3544 /*
3545  * Adds an entry to the irq list passed in, and returns the new list.
3546  * Entries are added in priority order (lower numerical priorities are
3547  * placed closer to the head of the list)
3548  */
3549 static prs_irq_list_t *
3550 acpi_insert_prs_irq_ent(prs_irq_list_t *listp, int priority, int irq,
3551     iflag_t *iflagp, acpi_prs_private_t *prsprvp)
3552 {
3553 	struct prs_irq_list_ent *newent, *prevp = NULL, *origlistp;
3554 
3555 	newent = kmem_zalloc(sizeof (struct prs_irq_list_ent), KM_SLEEP);
3556 
3557 	newent->list_prio = priority;
3558 	newent->irq = irq;
3559 	newent->intrflags = *iflagp;
3560 	newent->prsprv = *prsprvp;
3561 	/* ->next is NULL from kmem_zalloc */
3562 
3563 	/*
3564 	 * New list -- return the new entry as the list.
3565 	 */
3566 	if (listp == NULL)
3567 		return (newent);
3568 
3569 	/*
3570 	 * Save original list pointer for return (since we're not modifying
3571 	 * the head)
3572 	 */
3573 	origlistp = listp;
3574 
3575 	/*
3576 	 * Insertion sort, with entries with identical keys stored AFTER
3577 	 * existing entries (the less-than-or-equal test of priority does
3578 	 * this for us).
3579 	 */
3580 	while (listp != NULL && listp->list_prio <= priority) {
3581 		prevp = listp;
3582 		listp = listp->next;
3583 	}
3584 
3585 	newent->next = listp;
3586 
3587 	if (prevp == NULL) { /* Add at head of list (newent is the new head) */
3588 		return (newent);
3589 	} else {
3590 		prevp->next = newent;
3591 		return (origlistp);
3592 	}
3593 }
3594 
3595 /*
3596  * Frees the list passed in, deallocating all memory and leaving *listpp
3597  * set to NULL.
3598  */
3599 static void
3600 acpi_destroy_prs_irq_list(prs_irq_list_t **listpp)
3601 {
3602 	struct prs_irq_list_ent *nextp;
3603 
3604 	ASSERT(listpp != NULL);
3605 
3606 	while (*listpp != NULL) {
3607 		nextp = (*listpp)->next;
3608 		kmem_free(*listpp, sizeof (struct prs_irq_list_ent));
3609 		*listpp = nextp;
3610 	}
3611 }
3612 
3613 /*
3614  * apic_choose_irqs_from_prs returns a list of irqs selected from the list of
3615  * irqs returned by the link device's _PRS method.  The irqs are chosen
3616  * to minimize contention in situations where the interrupt link device
3617  * can be programmed to steer interrupts to different interrupt controller
3618  * inputs (some of which may already be in use).  The list is sorted in order
3619  * of irqs to use, with the highest priority given to interrupt controller
3620  * inputs that are not shared.   When an interrupt controller input
3621  * must be shared, apic_choose_irqs_from_prs adds the possible irqs to the
3622  * returned list in the order that minimizes sharing (thereby ensuring lowest
3623  * possible latency from interrupt trigger time to ISR execution time).
3624  */
3625 static prs_irq_list_t *
3626 apic_choose_irqs_from_prs(acpi_irqlist_t *irqlistent, dev_info_t *dip,
3627     int crs_irq)
3628 {
3629 	int32_t irq;
3630 	int i;
3631 	prs_irq_list_t *prsirqlistp = NULL;
3632 	iflag_t iflags;
3633 
3634 	while (irqlistent != NULL) {
3635 		irqlistent->intr_flags.bustype = BUS_PCI;
3636 
3637 		for (i = 0; i < irqlistent->num_irqs; i++) {
3638 
3639 			irq = irqlistent->irqs[i];
3640 
3641 			if (irq <= 0) {
3642 				/* invalid irq number */
3643 				continue;
3644 			}
3645 
3646 			if ((irq < 16) && (apic_reserved_irqlist[irq]))
3647 				continue;
3648 
3649 			if ((apic_irq_table[irq] == NULL) ||
3650 			    (apic_irq_table[irq]->airq_dip == dip)) {
3651 
3652 				prsirqlistp = acpi_insert_prs_irq_ent(
3653 				    prsirqlistp, 0 /* Highest priority */, irq,
3654 				    &irqlistent->intr_flags,
3655 				    &irqlistent->acpi_prs_prv);
3656 
3657 				/*
3658 				 * If we do not prefer the current irq from _CRS
3659 				 * or if we do and this irq is the same as the
3660 				 * current irq from _CRS, this is the one
3661 				 * to pick.
3662 				 */
3663 				if (!(apic_prefer_crs) || (irq == crs_irq)) {
3664 					return (prsirqlistp);
3665 				}
3666 				continue;
3667 			}
3668 
3669 			/*
3670 			 * Edge-triggered interrupts cannot be shared
3671 			 */
3672 			if (irqlistent->intr_flags.intr_el == INTR_EL_EDGE)
3673 				continue;
3674 
3675 			/*
3676 			 * To work around BIOSes that contain incorrect
3677 			 * interrupt polarity information in interrupt
3678 			 * descriptors returned by _PRS, we assume that
3679 			 * the polarity of the other device sharing this
3680 			 * interrupt controller input is compatible.
3681 			 * If it's not, the caller will catch it when
3682 			 * the caller invokes the link device's _CRS method
3683 			 * (after invoking its _SRS method).
3684 			 */
3685 			iflags = irqlistent->intr_flags;
3686 			iflags.intr_po =
3687 			    apic_irq_table[irq]->airq_iflag.intr_po;
3688 
3689 			if (!acpi_intr_compatible(iflags,
3690 			    apic_irq_table[irq]->airq_iflag)) {
3691 				APIC_VERBOSE_IRQ((CE_CONT, "!%s: irq %d "
3692 				    "not compatible [%x:%x:%x !~ %x:%x:%x]",
3693 				    psm_name, irq,
3694 				    iflags.intr_po,
3695 				    iflags.intr_el,
3696 				    iflags.bustype,
3697 				    apic_irq_table[irq]->airq_iflag.intr_po,
3698 				    apic_irq_table[irq]->airq_iflag.intr_el,
3699 				    apic_irq_table[irq]->airq_iflag.bustype));
3700 				continue;
3701 			}
3702 
3703 			/*
3704 			 * If we prefer the irq from _CRS, no need
3705 			 * to search any further (and make sure
3706 			 * to add this irq with the highest priority
3707 			 * so it's tried first).
3708 			 */
3709 			if (crs_irq == irq && apic_prefer_crs) {
3710 
3711 				return (acpi_insert_prs_irq_ent(
3712 				    prsirqlistp,
3713 				    0 /* Highest priority */,
3714 				    irq, &iflags,
3715 				    &irqlistent->acpi_prs_prv));
3716 			}
3717 
3718 			/*
3719 			 * Priority is equal to the share count (lower
3720 			 * share count is higher priority). Note that
3721 			 * the intr flags passed in here are the ones we
3722 			 * changed above -- if incorrect, it will be
3723 			 * caught by the caller's _CRS flags comparison.
3724 			 */
3725 			prsirqlistp = acpi_insert_prs_irq_ent(
3726 			    prsirqlistp,
3727 			    apic_irq_table[irq]->airq_share, irq,
3728 			    &iflags, &irqlistent->acpi_prs_prv);
3729 		}
3730 
3731 		/* Go to the next irqlist entry */
3732 		irqlistent = irqlistent->next;
3733 	}
3734 
3735 	return (prsirqlistp);
3736 }
3737 
3738 /*
3739  * Configures the irq for the interrupt link device identified by
3740  * acpipsmlnkp.
3741  *
3742  * Gets the current and the list of possible irq settings for the
3743  * device. If apic_unconditional_srs is not set, and the current
3744  * resource setting is in the list of possible irq settings,
3745  * current irq resource setting is passed to the caller.
3746  *
3747  * Otherwise, picks an irq number from the list of possible irq
3748  * settings, and sets the irq of the device to this value.
3749  * If prefer_crs is set, among a set of irq numbers in the list that have
3750  * the least number of devices sharing the interrupt, we pick current irq
3751  * resource setting if it is a member of this set.
3752  *
3753  * Passes the irq number in the value pointed to by pci_irqp, and
3754  * polarity and sensitivity in the structure pointed to by dipintrflagp
3755  * to the caller.
3756  *
3757  * Note that if setting the irq resource failed, but successfuly obtained
3758  * the current irq resource settings, passes the current irq resources
3759  * and considers it a success.
3760  *
3761  * Returns:
3762  * ACPI_PSM_SUCCESS on success.
3763  *
3764  * ACPI_PSM_FAILURE if an error occured during the configuration or
3765  * if a suitable irq was not found for this device, or if setting the
3766  * irq resource and obtaining the current resource fails.
3767  *
3768  */
3769 static int
3770 apic_acpi_irq_configure(acpi_psm_lnk_t *acpipsmlnkp, dev_info_t *dip,
3771     int *pci_irqp, iflag_t *dipintr_flagp)
3772 {
3773 	int32_t irq;
3774 	int cur_irq = -1;
3775 	acpi_irqlist_t *irqlistp;
3776 	prs_irq_list_t *prs_irq_listp, *prs_irq_entp;
3777 	boolean_t found_irq = B_FALSE;
3778 
3779 	dipintr_flagp->bustype = BUS_PCI;
3780 
3781 	if ((acpi_get_possible_irq_resources(acpipsmlnkp, &irqlistp))
3782 	    == ACPI_PSM_FAILURE) {
3783 		APIC_VERBOSE_IRQ((CE_WARN, "!%s: Unable to determine "
3784 		    "or assign IRQ for device %s, instance #%d: The system was "
3785 		    "unable to get the list of potential IRQs from ACPI.",
3786 		    psm_name, ddi_get_name(dip), ddi_get_instance(dip)));
3787 
3788 		return (ACPI_PSM_FAILURE);
3789 	}
3790 
3791 	if ((acpi_get_current_irq_resource(acpipsmlnkp, &cur_irq,
3792 	    dipintr_flagp) == ACPI_PSM_SUCCESS) && (!apic_unconditional_srs) &&
3793 	    (cur_irq > 0)) {
3794 		/*
3795 		 * If an IRQ is set in CRS and that IRQ exists in the set
3796 		 * returned from _PRS, return that IRQ, otherwise print
3797 		 * a warning
3798 		 */
3799 
3800 		if (acpi_irqlist_find_irq(irqlistp, cur_irq, NULL)
3801 		    == ACPI_PSM_SUCCESS) {
3802 
3803 			ASSERT(pci_irqp != NULL);
3804 			*pci_irqp = cur_irq;
3805 			acpi_free_irqlist(irqlistp);
3806 			return (ACPI_PSM_SUCCESS);
3807 		}
3808 
3809 		APIC_VERBOSE_IRQ((CE_WARN, "!%s: Could not find the "
3810 		    "current irq %d for device %s, instance #%d in ACPI's "
3811 		    "list of possible irqs for this device. Picking one from "
3812 		    " the latter list.", psm_name, cur_irq, ddi_get_name(dip),
3813 		    ddi_get_instance(dip)));
3814 	}
3815 
3816 	if ((prs_irq_listp = apic_choose_irqs_from_prs(irqlistp, dip,
3817 	    cur_irq)) == NULL) {
3818 
3819 		APIC_VERBOSE_IRQ((CE_WARN, "!%s: Could not find a "
3820 		    "suitable irq from the list of possible irqs for device "
3821 		    "%s, instance #%d in ACPI's list of possible irqs",
3822 		    psm_name, ddi_get_name(dip), ddi_get_instance(dip)));
3823 
3824 		acpi_free_irqlist(irqlistp);
3825 		return (ACPI_PSM_FAILURE);
3826 	}
3827 
3828 	acpi_free_irqlist(irqlistp);
3829 
3830 	for (prs_irq_entp = prs_irq_listp;
3831 	    prs_irq_entp != NULL && found_irq == B_FALSE;
3832 	    prs_irq_entp = prs_irq_entp->next) {
3833 
3834 		acpipsmlnkp->acpi_prs_prv = prs_irq_entp->prsprv;
3835 		irq = prs_irq_entp->irq;
3836 
3837 		APIC_VERBOSE_IRQ((CE_CONT, "!%s: Setting irq %d for "
3838 		    "device %s instance #%d\n", psm_name, irq,
3839 		    ddi_get_name(dip), ddi_get_instance(dip)));
3840 
3841 		if ((acpi_set_irq_resource(acpipsmlnkp, irq))
3842 		    == ACPI_PSM_SUCCESS) {
3843 			/*
3844 			 * setting irq was successful, check to make sure CRS
3845 			 * reflects that. If CRS does not agree with what we
3846 			 * set, return the irq that was set.
3847 			 */
3848 
3849 			if (acpi_get_current_irq_resource(acpipsmlnkp, &cur_irq,
3850 			    dipintr_flagp) == ACPI_PSM_SUCCESS) {
3851 
3852 				if (cur_irq != irq)
3853 					APIC_VERBOSE_IRQ((CE_WARN,
3854 					    "!%s: IRQ resource set "
3855 					    "(irqno %d) for device %s "
3856 					    "instance #%d, differs from "
3857 					    "current setting irqno %d",
3858 					    psm_name, irq, ddi_get_name(dip),
3859 					    ddi_get_instance(dip), cur_irq));
3860 			} else {
3861 				/*
3862 				 * On at least one system, there was a bug in
3863 				 * a DSDT method called by _STA, causing _STA to
3864 				 * indicate that the link device was disabled
3865 				 * (when, in fact, it was enabled).  Since _SRS
3866 				 * succeeded, assume that _CRS is lying and use
3867 				 * the iflags from this _PRS interrupt choice.
3868 				 * If we're wrong about the flags, the polarity
3869 				 * will be incorrect and we may get an interrupt
3870 				 * storm, but there's not much else we can do
3871 				 * at this point.
3872 				 */
3873 				*dipintr_flagp = prs_irq_entp->intrflags;
3874 			}
3875 
3876 			/*
3877 			 * Return the irq that was set, and not what _CRS
3878 			 * reports, since _CRS has been seen to return
3879 			 * different IRQs than what was passed to _SRS on some
3880 			 * systems (and just not return successfully on others).
3881 			 */
3882 			cur_irq = irq;
3883 			found_irq = B_TRUE;
3884 		} else {
3885 			APIC_VERBOSE_IRQ((CE_WARN, "!%s: set resource "
3886 			    "irq %d failed for device %s instance #%d",
3887 			    psm_name, irq, ddi_get_name(dip),
3888 			    ddi_get_instance(dip)));
3889 
3890 			if (cur_irq == -1) {
3891 				acpi_destroy_prs_irq_list(&prs_irq_listp);
3892 				return (ACPI_PSM_FAILURE);
3893 			}
3894 		}
3895 	}
3896 
3897 	acpi_destroy_prs_irq_list(&prs_irq_listp);
3898 
3899 	if (!found_irq)
3900 		return (ACPI_PSM_FAILURE);
3901 
3902 	ASSERT(pci_irqp != NULL);
3903 	*pci_irqp = cur_irq;
3904 	return (ACPI_PSM_SUCCESS);
3905 }
3906 
3907 void
3908 ioapic_disable_redirection()
3909 {
3910 	int ioapic_ix;
3911 	int intin_max;
3912 	int intin_ix;
3913 
3914 	/* Disable the I/O APIC redirection entries */
3915 	for (ioapic_ix = 0; ioapic_ix < apic_io_max; ioapic_ix++) {
3916 
3917 		/* Bits 23-16 define the maximum redirection entries */
3918 		intin_max = (ioapic_read(ioapic_ix, APIC_VERS_CMD) >> 16)
3919 		    & 0xff;
3920 
3921 		for (intin_ix = 0; intin_ix < intin_max; intin_ix++) {
3922 			/*
3923 			 * The assumption here is that this is safe, even for
3924 			 * systems with IOAPICs that suffer from the hardware
3925 			 * erratum because all devices have been quiesced before
3926 			 * this function is called from apic_shutdown()
3927 			 * (or equivalent). If that assumption turns out to be
3928 			 * false, this mask operation can induce the same
3929 			 * erratum result we're trying to avoid.
3930 			 */
3931 			ioapic_write(ioapic_ix, APIC_RDT_CMD + 2 * intin_ix,
3932 			    AV_MASK);
3933 		}
3934 	}
3935 }
3936 
3937 /*
3938  * Looks for an IOAPIC with the specified physical address in the /ioapics
3939  * node in the device tree (created by the PCI enumerator).
3940  */
3941 static boolean_t
3942 apic_is_ioapic_AMD_813x(uint32_t physaddr)
3943 {
3944 	/*
3945 	 * Look in /ioapics, for the ioapic with
3946 	 * the physical address given
3947 	 */
3948 	dev_info_t *ioapicsnode = ddi_find_devinfo(IOAPICS_NODE_NAME, -1, 0);
3949 	dev_info_t *ioapic_child;
3950 	boolean_t rv = B_FALSE;
3951 	int vid, did;
3952 	uint64_t ioapic_paddr;
3953 	boolean_t done = B_FALSE;
3954 
3955 	if (ioapicsnode == NULL)
3956 		return (B_FALSE);
3957 
3958 	/* Load first child: */
3959 	ioapic_child = ddi_get_child(ioapicsnode);
3960 	while (!done && ioapic_child != 0) { /* Iterate over children */
3961 
3962 		if ((ioapic_paddr = (uint64_t)ddi_prop_get_int64(DDI_DEV_T_ANY,
3963 		    ioapic_child, DDI_PROP_DONTPASS, "reg", 0))
3964 		    != 0 && physaddr == ioapic_paddr) {
3965 
3966 			vid = ddi_prop_get_int(DDI_DEV_T_ANY, ioapic_child,
3967 			    DDI_PROP_DONTPASS, IOAPICS_PROP_VENID, 0);
3968 
3969 			if (vid == VENID_AMD) {
3970 
3971 				did = ddi_prop_get_int(DDI_DEV_T_ANY,
3972 				    ioapic_child, DDI_PROP_DONTPASS,
3973 				    IOAPICS_PROP_DEVID, 0);
3974 
3975 				if (did == DEVID_8131_IOAPIC ||
3976 				    did == DEVID_8132_IOAPIC) {
3977 
3978 					rv = B_TRUE;
3979 					done = B_TRUE;
3980 				}
3981 			}
3982 		}
3983 
3984 		if (!done)
3985 			ioapic_child = ddi_get_next_sibling(ioapic_child);
3986 	}
3987 
3988 	/* The ioapics node was held by ddi_find_devinfo, so release it */
3989 	ndi_rele_devi(ioapicsnode);
3990 	return (rv);
3991 }
3992