xref: /titanic_41/usr/src/uts/sun4u/cpu/opl_olympus.c (revision 004388ebfdfe2ed7dfd2d153a876dfcc22d2c006)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 #include <sys/types.h>
29 #include <sys/systm.h>
30 #include <sys/ddi.h>
31 #include <sys/sysmacros.h>
32 #include <sys/archsystm.h>
33 #include <sys/vmsystm.h>
34 #include <sys/machparam.h>
35 #include <sys/machsystm.h>
36 #include <sys/machthread.h>
37 #include <sys/cpu.h>
38 #include <sys/cmp.h>
39 #include <sys/elf_SPARC.h>
40 #include <vm/vm_dep.h>
41 #include <vm/hat_sfmmu.h>
42 #include <vm/seg_kpm.h>
43 #include <sys/cpuvar.h>
44 #include <sys/opl_olympus_regs.h>
45 #include <sys/opl_module.h>
46 #include <sys/async.h>
47 #include <sys/cmn_err.h>
48 #include <sys/debug.h>
49 #include <sys/dditypes.h>
50 #include <sys/cpu_module.h>
51 #include <sys/sysmacros.h>
52 #include <sys/intreg.h>
53 #include <sys/clock.h>
54 #include <sys/platform_module.h>
55 #include <sys/ontrap.h>
56 #include <sys/panic.h>
57 #include <sys/memlist.h>
58 #include <sys/ndifm.h>
59 #include <sys/ddifm.h>
60 #include <sys/fm/protocol.h>
61 #include <sys/fm/util.h>
62 #include <sys/fm/cpu/SPARC64-VI.h>
63 #include <sys/dtrace.h>
64 #include <sys/watchpoint.h>
65 #include <sys/promif.h>
66 
67 /*
68  * Internal functions.
69  */
70 static int cpu_sync_log_err(void *flt);
71 static void cpu_payload_add_aflt(struct async_flt *, nvlist_t *, nvlist_t *);
72 static void opl_cpu_sync_error(struct regs *, ulong_t, ulong_t, uint_t, uint_t);
73 static int  cpu_flt_in_memory(opl_async_flt_t *, uint64_t);
74 
75 /*
76  * Error counters resetting interval.
77  */
78 static int opl_async_check_interval = 60;		/* 1 min */
79 
80 /*
81  * Maximum number of contexts for Olympus-C.
82  */
83 #define	MAX_NCTXS	(1 << 13)
84 
85 /* Will be set !NULL for SPARC64-VI and derivatives. */
86 static uchar_t ctx_pgsz_arr[MAX_NCTXS];
87 uchar_t *ctx_pgsz_array = ctx_pgsz_arr;
88 
89 /*
90  * PA[22:0] represent Displacement in Jupiter
91  * configuration space.
92  */
93 uint_t	root_phys_addr_lo_mask = 0x7fffffu;
94 
95 /*
96  * set in /etc/system to control logging of user BERR/TO's
97  */
98 int cpu_berr_to_verbose = 0;
99 
100 static int min_ecache_size;
101 static uint_t priv_hcl_1;
102 static uint_t priv_hcl_2;
103 static uint_t priv_hcl_4;
104 static uint_t priv_hcl_8;
105 
106 /*
107  * Olympus error log
108  */
109 static opl_errlog_t	*opl_err_log;
110 
111 /*
112  * UE is classified into four classes (MEM, CHANNEL, CPU, PATH).
113  * No any other ecc_type_info insertion is allowed in between the following
114  * four UE classess.
115  */
116 ecc_type_to_info_t ecc_type_to_info[] = {
117 	SFSR_UE,	"UE ",	(OPL_ECC_SYNC_TRAP), OPL_CPU_SYNC_UE,
118 	"Uncorrectable ECC",  FM_EREPORT_PAYLOAD_SYNC,
119 	FM_EREPORT_CPU_UE_MEM,
120 	SFSR_UE,	"UE ",	(OPL_ECC_SYNC_TRAP), OPL_CPU_SYNC_UE,
121 	"Uncorrectable ECC",  FM_EREPORT_PAYLOAD_SYNC,
122 	FM_EREPORT_CPU_UE_CHANNEL,
123 	SFSR_UE,	"UE ",	(OPL_ECC_SYNC_TRAP), OPL_CPU_SYNC_UE,
124 	"Uncorrectable ECC",  FM_EREPORT_PAYLOAD_SYNC,
125 	FM_EREPORT_CPU_UE_CPU,
126 	SFSR_UE,	"UE ",	(OPL_ECC_SYNC_TRAP), OPL_CPU_SYNC_UE,
127 	"Uncorrectable ECC",  FM_EREPORT_PAYLOAD_SYNC,
128 	FM_EREPORT_CPU_UE_PATH,
129 	SFSR_BERR, "BERR ", (OPL_ECC_SYNC_TRAP), OPL_CPU_SYNC_OTHERS,
130 	"Bus Error",  FM_EREPORT_PAYLOAD_SYNC,
131 	FM_EREPORT_CPU_BERR,
132 	SFSR_TO, "TO ", (OPL_ECC_SYNC_TRAP), OPL_CPU_SYNC_OTHERS,
133 	"Bus Timeout",  FM_EREPORT_PAYLOAD_SYNC,
134 	FM_EREPORT_CPU_BTO,
135 	SFSR_TLB_MUL, "TLB_MUL ", (OPL_ECC_SYNC_TRAP), OPL_CPU_SYNC_OTHERS,
136 	"TLB MultiHit",  FM_EREPORT_PAYLOAD_SYNC,
137 	FM_EREPORT_CPU_MTLB,
138 	SFSR_TLB_PRT, "TLB_PRT ", (OPL_ECC_SYNC_TRAP), OPL_CPU_SYNC_OTHERS,
139 	"TLB Parity",  FM_EREPORT_PAYLOAD_SYNC,
140 	FM_EREPORT_CPU_TLBP,
141 
142 	UGESR_IAUG_CRE, "IAUG_CRE", OPL_ECC_URGENT_TRAP, OPL_CPU_URGENT,
143 	"IAUG CRE",  FM_EREPORT_PAYLOAD_URGENT,
144 	FM_EREPORT_CPU_CRE,
145 	UGESR_IAUG_TSBCTXT, "IAUG_TSBCTXT",
146 	OPL_ECC_URGENT_TRAP, OPL_CPU_URGENT,
147 	"IAUG TSBCTXT",  FM_EREPORT_PAYLOAD_URGENT,
148 	FM_EREPORT_CPU_TSBCTX,
149 	UGESR_IUG_TSBP, "IUG_TSBP", OPL_ECC_URGENT_TRAP, OPL_CPU_URGENT,
150 	"IUG TSBP",  FM_EREPORT_PAYLOAD_URGENT,
151 	FM_EREPORT_CPU_TSBP,
152 	UGESR_IUG_PSTATE, "IUG_PSTATE", OPL_ECC_URGENT_TRAP, OPL_CPU_URGENT,
153 	"IUG PSTATE",  FM_EREPORT_PAYLOAD_URGENT,
154 	FM_EREPORT_CPU_PSTATE,
155 	UGESR_IUG_TSTATE, "IUG_TSTATE", OPL_ECC_URGENT_TRAP, OPL_CPU_URGENT,
156 	"IUG TSTATE",  FM_EREPORT_PAYLOAD_URGENT,
157 	FM_EREPORT_CPU_TSTATE,
158 	UGESR_IUG_F, "IUG_F", OPL_ECC_URGENT_TRAP, OPL_CPU_URGENT,
159 	"IUG FREG",  FM_EREPORT_PAYLOAD_URGENT,
160 	FM_EREPORT_CPU_IUG_F,
161 	UGESR_IUG_R, "IUG_R", OPL_ECC_URGENT_TRAP, OPL_CPU_URGENT,
162 	"IUG RREG",  FM_EREPORT_PAYLOAD_URGENT,
163 	FM_EREPORT_CPU_IUG_R,
164 	UGESR_AUG_SDC, "AUG_SDC", OPL_ECC_URGENT_TRAP, OPL_CPU_URGENT,
165 	"AUG SDC",  FM_EREPORT_PAYLOAD_URGENT,
166 	FM_EREPORT_CPU_SDC,
167 	UGESR_IUG_WDT, "IUG_WDT", OPL_ECC_URGENT_TRAP, OPL_CPU_URGENT,
168 	"IUG WDT",  FM_EREPORT_PAYLOAD_URGENT,
169 	FM_EREPORT_CPU_WDT,
170 	UGESR_IUG_DTLB, "IUG_DTLB", OPL_ECC_URGENT_TRAP, OPL_CPU_URGENT,
171 	"IUG DTLB",  FM_EREPORT_PAYLOAD_URGENT,
172 	FM_EREPORT_CPU_DTLB,
173 	UGESR_IUG_ITLB, "IUG_ITLB", OPL_ECC_URGENT_TRAP, OPL_CPU_URGENT,
174 	"IUG ITLB",  FM_EREPORT_PAYLOAD_URGENT,
175 	FM_EREPORT_CPU_ITLB,
176 	UGESR_IUG_COREERR, "IUG_COREERR",
177 	OPL_ECC_URGENT_TRAP, OPL_CPU_URGENT,
178 	"IUG COREERR",  FM_EREPORT_PAYLOAD_URGENT,
179 	FM_EREPORT_CPU_CORE,
180 	UGESR_MULTI_DAE, "MULTI_DAE", OPL_ECC_URGENT_TRAP, OPL_CPU_URGENT,
181 	"MULTI DAE",  FM_EREPORT_PAYLOAD_URGENT,
182 	FM_EREPORT_CPU_DAE,
183 	UGESR_MULTI_IAE, "MULTI_IAE", OPL_ECC_URGENT_TRAP, OPL_CPU_URGENT,
184 	"MULTI IAE",  FM_EREPORT_PAYLOAD_URGENT,
185 	FM_EREPORT_CPU_IAE,
186 	UGESR_MULTI_UGE, "MULTI_UGE", OPL_ECC_URGENT_TRAP, OPL_CPU_URGENT,
187 	"MULTI UGE",  FM_EREPORT_PAYLOAD_URGENT,
188 	FM_EREPORT_CPU_UGE,
189 	0,		NULL,		0,		0,
190 	NULL,  0,	   0,
191 };
192 
193 int (*p2get_mem_info)(int synd_code, uint64_t paddr,
194 		uint64_t *mem_sizep, uint64_t *seg_sizep, uint64_t *bank_sizep,
195 		int *segsp, int *banksp, int *mcidp);
196 
197 
198 /*
199  * Setup trap handlers for 0xA, 0x32, 0x40 trap types.
200  */
201 void
202 cpu_init_trap(void)
203 {
204 	OPL_SET_TRAP(tt0_iae, opl_serr_instr);
205 	OPL_SET_TRAP(tt1_iae, opl_serr_instr);
206 	OPL_SET_TRAP(tt0_dae, opl_serr_instr);
207 	OPL_SET_TRAP(tt1_dae, opl_serr_instr);
208 	OPL_SET_TRAP(tt0_asdat, opl_ugerr_instr);
209 	OPL_SET_TRAP(tt1_asdat, opl_ugerr_instr);
210 }
211 
212 static int
213 getintprop(pnode_t node, char *name, int deflt)
214 {
215 	int	value;
216 
217 	switch (prom_getproplen(node, name)) {
218 	case sizeof (int):
219 		(void) prom_getprop(node, name, (caddr_t)&value);
220 		break;
221 
222 	default:
223 		value = deflt;
224 		break;
225 	}
226 
227 	return (value);
228 }
229 
230 /*
231  * Set the magic constants of the implementation.
232  */
233 /*ARGSUSED*/
234 void
235 cpu_fiximp(pnode_t dnode)
236 {
237 	int i, a;
238 	extern int vac_size, vac_shift;
239 	extern uint_t vac_mask;
240 
241 	static struct {
242 		char	*name;
243 		int	*var;
244 		int	defval;
245 	} prop[] = {
246 		"l1-dcache-size", &dcache_size, OPL_DCACHE_SIZE,
247 		"l1-dcache-line-size", &dcache_linesize, OPL_DCACHE_LSIZE,
248 		"l1-icache-size", &icache_size, OPL_ICACHE_SIZE,
249 		"l1-icache-line-size", &icache_linesize, OPL_ICACHE_LSIZE,
250 		"l2-cache-size", &ecache_size, OPL_ECACHE_SIZE,
251 		"l2-cache-line-size", &ecache_alignsize, OPL_ECACHE_LSIZE,
252 		"l2-cache-associativity", &ecache_associativity, OPL_ECACHE_NWAY
253 	};
254 
255 	for (i = 0; i < sizeof (prop) / sizeof (prop[0]); i++)
256 		*prop[i].var = getintprop(dnode, prop[i].name, prop[i].defval);
257 
258 	ecache_setsize = ecache_size / ecache_associativity;
259 
260 	vac_size = OPL_VAC_SIZE;
261 	vac_mask = MMU_PAGEMASK & (vac_size - 1);
262 	i = 0; a = vac_size;
263 	while (a >>= 1)
264 		++i;
265 	vac_shift = i;
266 	shm_alignment = vac_size;
267 	vac = 1;
268 }
269 
270 void
271 send_mondo_set(cpuset_t set)
272 {
273 	int lo, busy, nack, shipped = 0;
274 	uint16_t i, cpuids[IDSR_BN_SETS];
275 	uint64_t idsr, nackmask = 0, busymask, curnack, curbusy;
276 	uint64_t starttick, endtick, tick, lasttick;
277 #if (NCPU > IDSR_BN_SETS)
278 	int index = 0;
279 	int ncpuids = 0;
280 #endif
281 #ifdef	OLYMPUS_C_REV_A_ERRATA_XCALL
282 	int bn_sets = IDSR_BN_SETS;
283 	uint64_t ver;
284 
285 	ASSERT(NCPU > bn_sets);
286 #endif
287 
288 	ASSERT(!CPUSET_ISNULL(set));
289 	starttick = lasttick = gettick();
290 
291 #ifdef	OLYMPUS_C_REV_A_ERRATA_XCALL
292 	ver = ultra_getver();
293 	if (((ULTRA_VER_IMPL(ver)) == OLYMPUS_C_IMPL) &&
294 		((OLYMPUS_REV_MASK(ver)) == OLYMPUS_C_A))
295 		bn_sets = 1;
296 #endif
297 
298 #if (NCPU <= IDSR_BN_SETS)
299 	for (i = 0; i < NCPU; i++)
300 		if (CPU_IN_SET(set, i)) {
301 			shipit(i, shipped);
302 			nackmask |= IDSR_NACK_BIT(shipped);
303 			cpuids[shipped++] = i;
304 			CPUSET_DEL(set, i);
305 			if (CPUSET_ISNULL(set))
306 				break;
307 		}
308 	CPU_STATS_ADDQ(CPU, sys, xcalls, shipped);
309 #else
310 	for (i = 0; i < NCPU; i++)
311 		if (CPU_IN_SET(set, i)) {
312 			ncpuids++;
313 
314 			/*
315 			 * Ship only to the first (IDSR_BN_SETS) CPUs.  If we
316 			 * find we have shipped to more than (IDSR_BN_SETS)
317 			 * CPUs, set "index" to the highest numbered CPU in
318 			 * the set so we can ship to other CPUs a bit later on.
319 			 */
320 #ifdef	OLYMPUS_C_REV_A_ERRATA_XCALL
321 			if (shipped < bn_sets) {
322 #else
323 			if (shipped < IDSR_BN_SETS) {
324 #endif
325 				shipit(i, shipped);
326 				nackmask |= IDSR_NACK_BIT(shipped);
327 				cpuids[shipped++] = i;
328 				CPUSET_DEL(set, i);
329 				if (CPUSET_ISNULL(set))
330 					break;
331 			} else
332 				index = (int)i;
333 		}
334 
335 	CPU_STATS_ADDQ(CPU, sys, xcalls, ncpuids);
336 #endif
337 
338 	busymask = IDSR_NACK_TO_BUSY(nackmask);
339 	busy = nack = 0;
340 	endtick = starttick + xc_tick_limit;
341 	for (;;) {
342 		idsr = getidsr();
343 #if (NCPU <= IDSR_BN_SETS)
344 		if (idsr == 0)
345 			break;
346 #else
347 		if (idsr == 0 && shipped == ncpuids)
348 			break;
349 #endif
350 		tick = gettick();
351 		/*
352 		 * If there is a big jump between the current tick
353 		 * count and lasttick, we have probably hit a break
354 		 * point.  Adjust endtick accordingly to avoid panic.
355 		 */
356 		if (tick > (lasttick + xc_tick_jump_limit))
357 			endtick += (tick - lasttick);
358 		lasttick = tick;
359 		if (tick > endtick) {
360 			if (panic_quiesce)
361 				return;
362 			cmn_err(CE_CONT, "send mondo timeout "
363 				"[%d NACK %d BUSY]\nIDSR 0x%"
364 				"" PRIx64 "  cpuids:", nack, busy, idsr);
365 #ifdef	OLYMPUS_C_REV_A_ERRATA_XCALL
366 			for (i = 0; i < bn_sets; i++) {
367 #else
368 			for (i = 0; i < IDSR_BN_SETS; i++) {
369 #endif
370 				if (idsr & (IDSR_NACK_BIT(i) |
371 				    IDSR_BUSY_BIT(i))) {
372 					cmn_err(CE_CONT, " 0x%x",
373 						cpuids[i]);
374 				}
375 			}
376 			cmn_err(CE_CONT, "\n");
377 			cmn_err(CE_PANIC, "send_mondo_set: timeout");
378 		}
379 		curnack = idsr & nackmask;
380 		curbusy = idsr & busymask;
381 
382 #ifdef OLYMPUS_C_REV_B_ERRATA_XCALL
383 		/*
384 		 * Only proceed to send more xcalls if all the
385 		 * cpus in the previous IDSR_BN_SETS were completed.
386 		 */
387 		if (curbusy) {
388 			busy++;
389 			continue;
390 		}
391 #endif /* OLYMPUS_C_REV_B_ERRATA_XCALL */
392 
393 #if (NCPU > IDSR_BN_SETS)
394 		if (shipped < ncpuids) {
395 			uint64_t cpus_left;
396 			uint16_t next = (uint16_t)index;
397 
398 			cpus_left = ~(IDSR_NACK_TO_BUSY(curnack) | curbusy) &
399 			    busymask;
400 
401 			if (cpus_left) {
402 				do {
403 					/*
404 					 * Sequence through and ship to the
405 					 * remainder of the CPUs in the system
406 					 * (e.g. other than the first
407 					 * (IDSR_BN_SETS)) in reverse order.
408 					 */
409 					lo = lowbit(cpus_left) - 1;
410 					i = IDSR_BUSY_IDX(lo);
411 					shipit(next, i);
412 					shipped++;
413 					cpuids[i] = next;
414 
415 					/*
416 					 * If we've processed all the CPUs,
417 					 * exit the loop now and save
418 					 * instructions.
419 					 */
420 					if (shipped == ncpuids)
421 						break;
422 
423 					for ((index = ((int)next - 1));
424 						index >= 0; index--)
425 						if (CPU_IN_SET(set, index)) {
426 							next = (uint16_t)index;
427 							break;
428 						}
429 
430 					cpus_left &= ~(1ull << lo);
431 				} while (cpus_left);
432 				continue;
433 			}
434 		}
435 #endif
436 #ifndef	OLYMPUS_C_REV_B_ERRATA_XCALL
437 		if (curbusy) {
438 			busy++;
439 			continue;
440 		}
441 #endif	/* OLYMPUS_C_REV_B_ERRATA_XCALL */
442 #ifdef SEND_MONDO_STATS
443 		{
444 			int n = gettick() - starttick;
445 			if (n < 8192)
446 				x_nack_stimes[n >> 7]++;
447 		}
448 #endif
449 		while (gettick() < (tick + sys_clock_mhz))
450 			;
451 		do {
452 			lo = lowbit(curnack) - 1;
453 			i = IDSR_NACK_IDX(lo);
454 			shipit(cpuids[i], i);
455 			curnack &= ~(1ull << lo);
456 		} while (curnack);
457 		nack++;
458 		busy = 0;
459 	}
460 #ifdef SEND_MONDO_STATS
461 	{
462 		int n = gettick() - starttick;
463 		if (n < 8192)
464 			x_set_stimes[n >> 7]++;
465 		else
466 			x_set_ltimes[(n >> 13) & 0xf]++;
467 	}
468 	x_set_cpus[shipped]++;
469 #endif
470 }
471 
472 /*
473  * Cpu private initialization.
474  */
475 void
476 cpu_init_private(struct cpu *cp)
477 {
478 	if (!(IS_OLYMPUS_C(cpunodes[cp->cpu_id].implementation))) {
479 		cmn_err(CE_PANIC, "CPU%d Impl %d: Only SPARC64-VI is supported",
480 			cp->cpu_id, cpunodes[cp->cpu_id].implementation);
481 	}
482 
483 	adjust_hw_copy_limits(cpunodes[cp->cpu_id].ecache_size);
484 }
485 
486 void
487 cpu_setup(void)
488 {
489 	extern int at_flags;
490 	extern int disable_delay_tlb_flush, delay_tlb_flush;
491 	extern int cpc_has_overflow_intr;
492 	extern int disable_text_largepages;
493 	extern int use_text_pgsz4m;
494 	uint64_t cpu0_log;
495 	extern	 uint64_t opl_cpu0_err_log;
496 
497 	/*
498 	 * Initialize Error log Scratch register for error handling.
499 	 */
500 
501 	cpu0_log = va_to_pa(&opl_cpu0_err_log);
502 	opl_error_setup(cpu0_log);
503 
504 	/*
505 	 * Enable MMU translating multiple page sizes for
506 	 * sITLB and sDTLB.
507 	 */
508 	opl_mpg_enable();
509 
510 	/*
511 	 * Setup chip-specific trap handlers.
512 	 */
513 	cpu_init_trap();
514 
515 	cache |= (CACHE_VAC | CACHE_PTAG | CACHE_IOCOHERENT);
516 
517 	at_flags = EF_SPARC_32PLUS | EF_SPARC_SUN_US1 | EF_SPARC_SUN_US3;
518 
519 	/*
520 	 * Use the maximum number of contexts available for SPARC64-VI
521 	 * unless it has been tuned for debugging.
522 	 * We are checking against 0 here since this value can be patched
523 	 * while booting.  It can not be patched via /etc/system since it
524 	 * will be patched too late and thus cause the system to panic.
525 	 */
526 	if (nctxs == 0)
527 		nctxs = MAX_NCTXS;
528 
529 	/*
530 	 * Due to the number of entries in the fully-associative tlb
531 	 * this may have to be tuned lower than in spitfire.
532 	 */
533 	pp_slots = MIN(8, MAXPP_SLOTS);
534 
535 	/*
536 	 * Block stores do not invalidate all pages of the d$, pagecopy
537 	 * et. al. need virtual translations with virtual coloring taken
538 	 * into consideration.  prefetch/ldd will pollute the d$ on the
539 	 * load side.
540 	 */
541 	pp_consistent_coloring = PPAGE_STORE_VCOLORING | PPAGE_LOADS_POLLUTE;
542 
543 	if (use_page_coloring) {
544 		do_pg_coloring = 1;
545 		if (use_virtual_coloring)
546 			do_virtual_coloring = 1;
547 	}
548 
549 	isa_list =
550 	    "sparcv9+vis2 sparcv9+vis sparcv9 "
551 	    "sparcv8plus+vis2 sparcv8plus+vis sparcv8plus "
552 	    "sparcv8 sparcv8-fsmuld sparcv7 sparc";
553 
554 	cpu_hwcap_flags = AV_SPARC_VIS | AV_SPARC_VIS2;
555 
556 	/*
557 	 * On SPARC64-VI, there's no hole in the virtual address space
558 	 */
559 	hole_start = hole_end = 0;
560 
561 	/*
562 	 * The kpm mapping window.
563 	 * kpm_size:
564 	 *	The size of a single kpm range.
565 	 *	The overall size will be: kpm_size * vac_colors.
566 	 * kpm_vbase:
567 	 *	The virtual start address of the kpm range within the kernel
568 	 *	virtual address space. kpm_vbase has to be kpm_size aligned.
569 	 */
570 	kpm_size = (size_t)(128ull * 1024 * 1024 * 1024 * 1024); /* 128TB */
571 	kpm_size_shift = 47;
572 	kpm_vbase = (caddr_t)0x8000000000000000ull; /* 8EB */
573 	kpm_smallpages = 1;
574 
575 	/*
576 	 * The traptrace code uses either %tick or %stick for
577 	 * timestamping.  We have %stick so we can use it.
578 	 */
579 	traptrace_use_stick = 1;
580 
581 	/*
582 	 * SPARC64-VI has a performance counter overflow interrupt
583 	 */
584 	cpc_has_overflow_intr = 1;
585 
586 	/*
587 	 * Use SPARC64-VI flush-all support
588 	 */
589 	if (!disable_delay_tlb_flush)
590 		delay_tlb_flush = 1;
591 
592 	/*
593 	 * Declare that this architecture/cpu combination does not support
594 	 * fpRAS.
595 	 */
596 	fpras_implemented = 0;
597 
598 	/*
599 	 * Enable 4M pages to be used for mapping user text by default.  Don't
600 	 * use large pages for initialized data segments since we may not know
601 	 * at exec() time what should be the preferred large page size for DTLB
602 	 * programming.
603 	 */
604 	use_text_pgsz4m = 1;
605 	disable_text_largepages = (1 << TTE64K) | (1 << TTE512K) |
606 	    (1 << TTE32M) | (1 << TTE256M);
607 }
608 
609 /*
610  * Called by setcpudelay
611  */
612 void
613 cpu_init_tick_freq(void)
614 {
615 	/*
616 	 * For SPARC64-VI we want to use the system clock rate as
617 	 * the basis for low level timing, due to support of mixed
618 	 * speed CPUs and power managment.
619 	 */
620 	if (system_clock_freq == 0)
621 		cmn_err(CE_PANIC, "setcpudelay: invalid system_clock_freq");
622 
623 	sys_tick_freq = system_clock_freq;
624 }
625 
626 #ifdef SEND_MONDO_STATS
627 uint32_t x_one_stimes[64];
628 uint32_t x_one_ltimes[16];
629 uint32_t x_set_stimes[64];
630 uint32_t x_set_ltimes[16];
631 uint32_t x_set_cpus[NCPU];
632 uint32_t x_nack_stimes[64];
633 #endif
634 
635 /*
636  * Note: A version of this function is used by the debugger via the KDI,
637  * and must be kept in sync with this version.  Any changes made to this
638  * function to support new chips or to accomodate errata must also be included
639  * in the KDI-specific version.  See us3_kdi.c.
640  */
641 void
642 send_one_mondo(int cpuid)
643 {
644 	int busy, nack;
645 	uint64_t idsr, starttick, endtick, tick, lasttick;
646 	uint64_t busymask;
647 
648 	CPU_STATS_ADDQ(CPU, sys, xcalls, 1);
649 	starttick = lasttick = gettick();
650 	shipit(cpuid, 0);
651 	endtick = starttick + xc_tick_limit;
652 	busy = nack = 0;
653 	busymask = IDSR_BUSY;
654 	for (;;) {
655 		idsr = getidsr();
656 		if (idsr == 0)
657 			break;
658 
659 		tick = gettick();
660 		/*
661 		 * If there is a big jump between the current tick
662 		 * count and lasttick, we have probably hit a break
663 		 * point.  Adjust endtick accordingly to avoid panic.
664 		 */
665 		if (tick > (lasttick + xc_tick_jump_limit))
666 			endtick += (tick - lasttick);
667 		lasttick = tick;
668 		if (tick > endtick) {
669 			if (panic_quiesce)
670 				return;
671 			cmn_err(CE_PANIC, "send mondo timeout "
672 				"(target 0x%x) [%d NACK %d BUSY]",
673 					cpuid, nack, busy);
674 		}
675 
676 		if (idsr & busymask) {
677 			busy++;
678 			continue;
679 		}
680 		drv_usecwait(1);
681 		shipit(cpuid, 0);
682 		nack++;
683 		busy = 0;
684 	}
685 #ifdef SEND_MONDO_STATS
686 	{
687 		int n = gettick() - starttick;
688 		if (n < 8192)
689 			x_one_stimes[n >> 7]++;
690 		else
691 			x_one_ltimes[(n >> 13) & 0xf]++;
692 	}
693 #endif
694 }
695 
696 /*
697  * init_mmu_page_sizes is set to one after the bootup time initialization
698  * via mmu_init_mmu_page_sizes, to indicate that mmu_page_sizes has a
699  * valid value.
700  *
701  * mmu_disable_ism_large_pages and mmu_disable_large_pages are the mmu-specific
702  * versions of disable_ism_large_pages and disable_large_pages, and feed back
703  * into those two hat variables at hat initialization time.
704  *
705  */
706 int init_mmu_page_sizes = 0;
707 static int mmu_disable_ism_large_pages = ((1 << TTE64K) |
708 	(1 << TTE512K) | (1 << TTE256M));
709 static int mmu_disable_large_pages = 0;
710 
711 /*
712  * Re-initialize mmu_page_sizes and friends, for SPARC64-VI mmu support.
713  * Called during very early bootup from check_cpus_set().
714  * Can be called to verify that mmu_page_sizes are set up correctly.
715  *
716  * Set Olympus defaults. We do not use the function parameter.
717  */
718 /*ARGSUSED*/
719 int
720 mmu_init_mmu_page_sizes(int32_t not_used)
721 {
722 	if (!init_mmu_page_sizes) {
723 		mmu_page_sizes = MMU_PAGE_SIZES;
724 		mmu_hashcnt = MAX_HASHCNT;
725 		mmu_ism_pagesize = MMU_PAGESIZE32M;
726 		mmu_exported_pagesize_mask = (1 << TTE8K) |
727 		    (1 << TTE64K) | (1 << TTE512K) | (1 << TTE4M) |
728 		    (1 << TTE32M) | (1 << TTE256M);
729 		init_mmu_page_sizes = 1;
730 		return (0);
731 	}
732 	return (1);
733 }
734 
735 /* SPARC64-VI worst case DTLB parameters */
736 #ifndef	LOCKED_DTLB_ENTRIES
737 #define	LOCKED_DTLB_ENTRIES	5	/* 2 user TSBs, 2 nucleus, + OBP */
738 #endif
739 #define	TOTAL_DTLB_ENTRIES	32
740 #define	AVAIL_32M_ENTRIES	0
741 #define	AVAIL_256M_ENTRIES	0
742 #define	AVAIL_DTLB_ENTRIES	(TOTAL_DTLB_ENTRIES - LOCKED_DTLB_ENTRIES)
743 static uint64_t ttecnt_threshold[MMU_PAGE_SIZES] = {
744 	AVAIL_DTLB_ENTRIES, AVAIL_DTLB_ENTRIES,
745 	AVAIL_DTLB_ENTRIES, AVAIL_DTLB_ENTRIES,
746 	AVAIL_DTLB_ENTRIES, AVAIL_DTLB_ENTRIES};
747 
748 size_t
749 mmu_map_pgsz(size_t pgsize)
750 {
751 	struct proc *p = curproc;
752 	struct as *as = p->p_as;
753 	struct hat *hat = as->a_hat;
754 	uint_t pgsz0, pgsz1;
755 	size_t size0, size1;
756 
757 	ASSERT(mmu_page_sizes == max_mmu_page_sizes);
758 	pgsz0 = hat->sfmmu_pgsz[0];
759 	pgsz1 = hat->sfmmu_pgsz[1];
760 	size0 = hw_page_array[pgsz0].hp_size;
761 	size1 = hw_page_array[pgsz1].hp_size;
762 	/* Allow use of a larger pagesize if neither TLB is reprogrammed. */
763 	if ((pgsz0 == TTE8K) && (pgsz1 == TTE8K)) {
764 		return (pgsize);
765 	/* Allow use of requested pagesize if TLB is reprogrammed to it. */
766 	} else if ((pgsize == size0) || (pgsize == size1)) {
767 		return (pgsize);
768 	/* Use larger reprogrammed TLB size if pgsize is atleast that big. */
769 	} else if (pgsz1 > pgsz0) {
770 		if (pgsize >= size1)
771 			return (size1);
772 	/* Use smaller reprogrammed TLB size if pgsize is atleast that big. */
773 	} else {
774 		if (pgsize >= size0)
775 			return (size0);
776 	}
777 	return (pgsize);
778 }
779 
780 /*
781  * The function returns the mmu-specific values for the
782  * hat's disable_large_pages and disable_ism_large_pages variables.
783  */
784 int
785 mmu_large_pages_disabled(uint_t flag)
786 {
787 	int pages_disable = 0;
788 
789 	if (flag == HAT_LOAD) {
790 		pages_disable =  mmu_disable_large_pages;
791 	} else if (flag == HAT_LOAD_SHARE) {
792 		pages_disable = mmu_disable_ism_large_pages;
793 	}
794 	return (pages_disable);
795 }
796 
797 /*
798  * mmu_init_large_pages is called with the desired ism_pagesize parameter.
799  * It may be called from set_platform_defaults, if some value other than 32M
800  * is desired.  mmu_ism_pagesize is the tunable.  If it has a bad value,
801  * then only warn, since it would be bad form to panic due to a user typo.
802  *
803  * The function re-initializes the mmu_disable_ism_large_pages variable.
804  */
805 void
806 mmu_init_large_pages(size_t ism_pagesize)
807 {
808 	switch (ism_pagesize) {
809 	case MMU_PAGESIZE4M:
810 		mmu_disable_ism_large_pages = ((1 << TTE64K) |
811 		    (1 << TTE512K) | (1 << TTE32M) | (1 << TTE256M));
812 		break;
813 	case MMU_PAGESIZE32M:
814 		mmu_disable_ism_large_pages = ((1 << TTE64K) |
815 		    (1 << TTE512K) | (1 << TTE256M));
816 		break;
817 	case MMU_PAGESIZE256M:
818 		mmu_disable_ism_large_pages = ((1 << TTE64K) |
819 		    (1 << TTE512K) | (1 << TTE32M));
820 		break;
821 	default:
822 		cmn_err(CE_WARN, "Unrecognized mmu_ism_pagesize value 0x%lx",
823 		    ism_pagesize);
824 		break;
825 	}
826 }
827 
828 /*ARGSUSED*/
829 uint_t
830 mmu_preferred_pgsz(struct hat *hat, caddr_t addr, size_t len)
831 {
832 	sfmmu_t *sfmmup = (sfmmu_t *)hat;
833 	uint_t pgsz0, pgsz1;
834 	uint_t szc, maxszc = mmu_page_sizes - 1;
835 	size_t pgsz;
836 	extern int disable_large_pages;
837 
838 	pgsz0 = (uint_t)sfmmup->sfmmu_pgsz[0];
839 	pgsz1 = (uint_t)sfmmup->sfmmu_pgsz[1];
840 
841 	/*
842 	 * If either of the TLBs are reprogrammed, choose
843 	 * the largest mapping size as the preferred size,
844 	 * if it fits the size and alignment constraints.
845 	 * Else return the largest mapping size that fits,
846 	 * if neither TLB is reprogrammed.
847 	 */
848 	if (pgsz0 > TTE8K || pgsz1 > TTE8K) {
849 		if (pgsz1 > pgsz0) {	/* First try pgsz1 */
850 			pgsz = hw_page_array[pgsz1].hp_size;
851 			if ((len >= pgsz) && IS_P2ALIGNED(addr, pgsz))
852 				return (pgsz1);
853 		}
854 		if (pgsz0 > TTE8K) {	/* Then try pgsz0, if !TTE8K */
855 			pgsz = hw_page_array[pgsz0].hp_size;
856 			if ((len >= pgsz) && IS_P2ALIGNED(addr, pgsz))
857 				return (pgsz0);
858 		}
859 	} else { /* Otherwise pick best fit if neither TLB is reprogrammed. */
860 		for (szc = maxszc; szc > TTE8K; szc--) {
861 			if (disable_large_pages & (1 << szc))
862 				continue;
863 
864 			pgsz = hw_page_array[szc].hp_size;
865 			if ((len >= pgsz) && IS_P2ALIGNED(addr, pgsz))
866 				return (szc);
867 		}
868 	}
869 	return (TTE8K);
870 }
871 
872 /*
873  * Function to reprogram the TLBs when page sizes used
874  * by a process change significantly.
875  */
876 void
877 mmu_setup_page_sizes(struct hat *hat, uint64_t *ttecnt)
878 {
879 	extern int page_szc(size_t);
880 	uint8_t pgsz0, pgsz1;
881 
882 	/*
883 	 * Don't program 2nd dtlb for kernel and ism hat
884 	 */
885 	if (hat->sfmmu_ismhat || hat == ksfmmup)
886 		return;
887 
888 	/*
889 	 * hat->sfmmu_pgsz[] is an array whose elements
890 	 * contain a sorted order of page sizes.  Element
891 	 * 0 is the most commonly used page size, followed
892 	 * by element 1, and so on.
893 	 *
894 	 * ttecnt[] is an array of per-page-size page counts
895 	 * mapped into the process.
896 	 *
897 	 * If the HAT's choice for page sizes is unsuitable,
898 	 * we can override it here.  The new values written
899 	 * to the array will be handed back to us later to
900 	 * do the actual programming of the TLB hardware.
901 	 *
902 	 */
903 	pgsz0 = (uint8_t)MIN(hat->sfmmu_pgsz[0], hat->sfmmu_pgsz[1]);
904 	pgsz1 = (uint8_t)MAX(hat->sfmmu_pgsz[0], hat->sfmmu_pgsz[1]);
905 
906 	/*
907 	 * This implements PAGESIZE programming of the sTLB
908 	 * if large TTE counts don't exceed the thresholds.
909 	 */
910 	if (ttecnt[pgsz0] < ttecnt_threshold[pgsz0])
911 		pgsz0 = page_szc(MMU_PAGESIZE);
912 	if (ttecnt[pgsz1] < ttecnt_threshold[pgsz1])
913 		pgsz1 = page_szc(MMU_PAGESIZE);
914 	hat->sfmmu_pgsz[0] = pgsz0;
915 	hat->sfmmu_pgsz[1] = pgsz1;
916 	/* otherwise, accept what the HAT chose for us */
917 }
918 
919 /*
920  * The HAT calls this function when an MMU context is allocated so that we
921  * can reprogram the large TLBs appropriately for the new process using
922  * the context.
923  *
924  * The caller must hold the HAT lock.
925  */
926 void
927 mmu_set_ctx_page_sizes(struct hat *hat)
928 {
929 	uint8_t pgsz0, pgsz1;
930 	uint8_t new_cext;
931 
932 	ASSERT(sfmmu_hat_lock_held(hat));
933 	/*
934 	 * Don't program 2nd dtlb for kernel and ism hat
935 	 */
936 	if (hat->sfmmu_ismhat || hat == ksfmmup)
937 		return;
938 
939 	/*
940 	 * If supported, reprogram the TLBs to a larger pagesize.
941 	 */
942 	pgsz0 = hat->sfmmu_pgsz[0];
943 	pgsz1 = hat->sfmmu_pgsz[1];
944 	ASSERT(pgsz0 < mmu_page_sizes);
945 	ASSERT(pgsz1 < mmu_page_sizes);
946 	new_cext = TAGACCEXT_MKSZPAIR(pgsz1, pgsz0);
947 	if (hat->sfmmu_cext != new_cext) {
948 		hat->sfmmu_cext = new_cext;
949 	}
950 	ctx_pgsz_array[hat->sfmmu_cnum] = hat->sfmmu_cext;
951 	/*
952 	 * sfmmu_setctx_sec() will take care of the
953 	 * rest of the dirty work for us.
954 	 */
955 }
956 
957 /*
958  * Return processor specific async error structure
959  * size used.
960  */
961 int
962 cpu_aflt_size(void)
963 {
964 	return (sizeof (opl_async_flt_t));
965 }
966 
967 /*
968  * The cpu_sync_log_err() function is called via the [uc]e_drain() function to
969  * post-process CPU events that are dequeued.  As such, it can be invoked
970  * from softint context, from AST processing in the trap() flow, or from the
971  * panic flow.  We decode the CPU-specific data, and take appropriate actions.
972  * Historically this entry point was used to log the actual cmn_err(9F) text;
973  * now with FMA it is used to prepare 'flt' to be converted into an ereport.
974  * With FMA this function now also returns a flag which indicates to the
975  * caller whether the ereport should be posted (1) or suppressed (0).
976  */
977 /*ARGSUSED*/
978 static int
979 cpu_sync_log_err(void *flt)
980 {
981 	opl_async_flt_t *opl_flt = (opl_async_flt_t *)flt;
982 	struct async_flt *aflt = (struct async_flt *)flt;
983 
984 	/*
985 	 * No extra processing of urgent error events.
986 	 * Always generate ereports for these events.
987 	 */
988 	if (aflt->flt_status == OPL_ECC_URGENT_TRAP)
989 		return (1);
990 
991 	/*
992 	 * Additional processing for synchronous errors.
993 	 */
994 	switch (opl_flt->flt_type) {
995 	case OPL_CPU_INV_SFSR:
996 		return (1);
997 
998 	case OPL_CPU_SYNC_UE:
999 		/*
1000 		 * The validity: SFSR_MK_UE bit has been checked
1001 		 * in opl_cpu_sync_error()
1002 		 * No more check is required.
1003 		 *
1004 		 * opl_flt->flt_eid_mod and flt_eid_sid have been set by H/W,
1005 		 * and they have been retrieved in cpu_queue_events()
1006 		 */
1007 
1008 		if (opl_flt->flt_eid_mod == OPL_ERRID_MEM) {
1009 			ASSERT(aflt->flt_in_memory);
1010 			/*
1011 			 * We want to skip logging only if ALL the following
1012 			 * conditions are true:
1013 			 *
1014 			 *	1. We are not panicing already.
1015 			 *	2. The error is a memory error.
1016 			 *	3. There is only one error.
1017 			 *	4. The error is on a retired page.
1018 			 *	5. The error occurred under on_trap
1019 			 *	protection AFLT_PROT_EC
1020 			 */
1021 			if (!panicstr && aflt->flt_prot == AFLT_PROT_EC &&
1022 			    page_retire_check(aflt->flt_addr, NULL) == 0) {
1023 				/*
1024 				 * Do not log an error from
1025 				 * the retired page
1026 				 */
1027 				softcall(ecc_page_zero, (void *)aflt->flt_addr);
1028 				return (0);
1029 			}
1030 			if (!panicstr)
1031 				cpu_page_retire(opl_flt);
1032 		}
1033 		return (1);
1034 
1035 	case OPL_CPU_SYNC_OTHERS:
1036 		/*
1037 		 * For the following error cases, the processor HW does
1038 		 * not set the flt_eid_mod/flt_eid_sid. Instead, SW will attempt
1039 		 * to assign appropriate values here to reflect what we
1040 		 * think is the most likely cause of the problem w.r.t to
1041 		 * the particular error event.  For Buserr and timeout
1042 		 * error event, we will assign OPL_ERRID_CHANNEL as the
1043 		 * most likely reason.  For TLB parity or multiple hit
1044 		 * error events, we will assign the reason as
1045 		 * OPL_ERRID_CPU (cpu related problem) and set the
1046 		 * flt_eid_sid to point to the cpuid.
1047 		 */
1048 
1049 		if (opl_flt->flt_bit & (SFSR_BERR|SFSR_TO)) {
1050 			/*
1051 			 * flt_eid_sid will not be used for this case.
1052 			 */
1053 			opl_flt->flt_eid_mod = OPL_ERRID_CHANNEL;
1054 		}
1055 		if (opl_flt->flt_bit & (SFSR_TLB_MUL|SFSR_TLB_PRT)) {
1056 			    opl_flt->flt_eid_mod = OPL_ERRID_CPU;
1057 			    opl_flt->flt_eid_sid = aflt->flt_inst;
1058 		}
1059 
1060 		/*
1061 		 * In case of no effective error bit
1062 		 */
1063 		if ((opl_flt->flt_bit & SFSR_ERRS) == 0) {
1064 			    opl_flt->flt_eid_mod = OPL_ERRID_CPU;
1065 			    opl_flt->flt_eid_sid = aflt->flt_inst;
1066 		}
1067 		break;
1068 
1069 		default:
1070 			return (1);
1071 	}
1072 	return (1);
1073 }
1074 
1075 /*
1076  * Retire the bad page that may contain the flushed error.
1077  */
1078 void
1079 cpu_page_retire(opl_async_flt_t *opl_flt)
1080 {
1081 	struct async_flt *aflt = (struct async_flt *)opl_flt;
1082 	(void) page_retire(aflt->flt_addr, PR_UE);
1083 }
1084 
1085 /*
1086  * Invoked by error_init() early in startup and therefore before
1087  * startup_errorq() is called to drain any error Q -
1088  *
1089  * startup()
1090  *   startup_end()
1091  *     error_init()
1092  *       cpu_error_init()
1093  * errorq_init()
1094  *   errorq_drain()
1095  * start_other_cpus()
1096  *
1097  * The purpose of this routine is to create error-related taskqs.  Taskqs
1098  * are used for this purpose because cpu_lock can't be grabbed from interrupt
1099  * context.
1100  *
1101  */
1102 /*ARGSUSED*/
1103 void
1104 cpu_error_init(int items)
1105 {
1106 	opl_err_log = (opl_errlog_t *)
1107 	    kmem_alloc(ERRLOG_ALLOC_SZ, KM_SLEEP);
1108 	if ((uint64_t)opl_err_log & MMU_PAGEOFFSET)
1109 		cmn_err(CE_PANIC, "The base address of the error log "
1110 		    "is not page aligned");
1111 }
1112 
1113 /*
1114  * We route all errors through a single switch statement.
1115  */
1116 void
1117 cpu_ue_log_err(struct async_flt *aflt)
1118 {
1119 	switch (aflt->flt_class) {
1120 	case CPU_FAULT:
1121 		if (cpu_sync_log_err(aflt))
1122 			cpu_ereport_post(aflt);
1123 		break;
1124 
1125 	case BUS_FAULT:
1126 		bus_async_log_err(aflt);
1127 		break;
1128 
1129 	default:
1130 		cmn_err(CE_WARN, "discarding async error %p with invalid "
1131 		    "fault class (0x%x)", (void *)aflt, aflt->flt_class);
1132 		return;
1133 	}
1134 }
1135 
1136 /*
1137  * Routine for panic hook callback from panic_idle().
1138  *
1139  * Nothing to do here.
1140  */
1141 void
1142 cpu_async_panic_callb(void)
1143 {
1144 }
1145 
1146 /*
1147  * Routine to return a string identifying the physical name
1148  * associated with a memory/cache error.
1149  */
1150 /*ARGSUSED*/
1151 int
1152 cpu_get_mem_unum(int synd_status, ushort_t flt_synd, uint64_t flt_stat,
1153     uint64_t flt_addr, int flt_bus_id, int flt_in_memory,
1154     ushort_t flt_status, char *buf, int buflen, int *lenp)
1155 {
1156 	int synd_code;
1157 	int ret;
1158 
1159 	/*
1160 	 * An AFSR of -1 defaults to a memory syndrome.
1161 	 */
1162 	synd_code = (int)flt_synd;
1163 
1164 	if (&plat_get_mem_unum) {
1165 		if ((ret = plat_get_mem_unum(synd_code, flt_addr, flt_bus_id,
1166 			flt_in_memory, flt_status, buf, buflen, lenp)) != 0) {
1167 			buf[0] = '\0';
1168 			*lenp = 0;
1169 		}
1170 		return (ret);
1171 	}
1172 	buf[0] = '\0';
1173 	*lenp = 0;
1174 	return (ENOTSUP);
1175 }
1176 
1177 /*
1178  * Wrapper for cpu_get_mem_unum() routine that takes an
1179  * async_flt struct rather than explicit arguments.
1180  */
1181 int
1182 cpu_get_mem_unum_aflt(int synd_status, struct async_flt *aflt,
1183     char *buf, int buflen, int *lenp)
1184 {
1185 	/*
1186 	 * We always pass -1 so that cpu_get_mem_unum will interpret this as a
1187 	 * memory error.
1188 	 */
1189 	return (cpu_get_mem_unum(synd_status, aflt->flt_synd,
1190 	    (uint64_t)-1,
1191 	    aflt->flt_addr, aflt->flt_bus_id, aflt->flt_in_memory,
1192 	    aflt->flt_status, buf, buflen, lenp));
1193 }
1194 
1195 /*
1196  * This routine is a more generic interface to cpu_get_mem_unum()
1197  * that may be used by other modules (e.g. mm).
1198  */
1199 /*ARGSUSED*/
1200 int
1201 cpu_get_mem_name(uint64_t synd, uint64_t *afsr, uint64_t afar,
1202     char *buf, int buflen, int *lenp)
1203 {
1204 	int synd_status, flt_in_memory, ret;
1205 	ushort_t flt_status = 0;
1206 	char unum[UNUM_NAMLEN];
1207 
1208 	/*
1209 	 * Check for an invalid address.
1210 	 */
1211 	if (afar == (uint64_t)-1)
1212 		return (ENXIO);
1213 
1214 	if (synd == (uint64_t)-1)
1215 		synd_status = AFLT_STAT_INVALID;
1216 	else
1217 		synd_status = AFLT_STAT_VALID;
1218 
1219 	flt_in_memory = (*afsr & SFSR_MEMORY) &&
1220 		pf_is_memory(afar >> MMU_PAGESHIFT);
1221 
1222 	ret = cpu_get_mem_unum(synd_status, (ushort_t)synd, *afsr, afar,
1223 		CPU->cpu_id, flt_in_memory, flt_status, unum,
1224 		UNUM_NAMLEN, lenp);
1225 	if (ret != 0)
1226 		return (ret);
1227 
1228 	if (*lenp >= buflen)
1229 		return (ENAMETOOLONG);
1230 
1231 	(void) strncpy(buf, unum, buflen);
1232 
1233 	return (0);
1234 }
1235 
1236 /*
1237  * Routine to return memory information associated
1238  * with a physical address and syndrome.
1239  */
1240 /*ARGSUSED*/
1241 int
1242 cpu_get_mem_info(uint64_t synd, uint64_t afar,
1243     uint64_t *mem_sizep, uint64_t *seg_sizep, uint64_t *bank_sizep,
1244     int *segsp, int *banksp, int *mcidp)
1245 {
1246 	int synd_code = (int)synd;
1247 
1248 	if (afar == (uint64_t)-1)
1249 		return (ENXIO);
1250 
1251 	if (p2get_mem_info != NULL)
1252 		return ((p2get_mem_info)(synd_code, afar,
1253 			mem_sizep, seg_sizep, bank_sizep,
1254 			segsp, banksp, mcidp));
1255 	else
1256 		return (ENOTSUP);
1257 }
1258 
1259 /*
1260  * Routine to return a string identifying the physical
1261  * name associated with a cpuid.
1262  */
1263 int
1264 cpu_get_cpu_unum(int cpuid, char *buf, int buflen, int *lenp)
1265 {
1266 	int ret;
1267 	char unum[UNUM_NAMLEN];
1268 
1269 	if (&plat_get_cpu_unum) {
1270 		if ((ret = plat_get_cpu_unum(cpuid, unum, UNUM_NAMLEN, lenp))
1271 			!= 0)
1272 			return (ret);
1273 	} else {
1274 		return (ENOTSUP);
1275 	}
1276 
1277 	if (*lenp >= buflen)
1278 		return (ENAMETOOLONG);
1279 
1280 	(void) strncpy(buf, unum, *lenp);
1281 
1282 	return (0);
1283 }
1284 
1285 /*
1286  * This routine exports the name buffer size.
1287  */
1288 size_t
1289 cpu_get_name_bufsize()
1290 {
1291 	return (UNUM_NAMLEN);
1292 }
1293 
1294 /*
1295  * Flush the entire ecache by ASI_L2_CNTL.U2_FLUSH
1296  */
1297 void
1298 cpu_flush_ecache(void)
1299 {
1300 	flush_ecache(ecache_flushaddr, cpunodes[CPU->cpu_id].ecache_size,
1301 	    cpunodes[CPU->cpu_id].ecache_linesize);
1302 }
1303 
1304 static uint8_t
1305 flt_to_trap_type(struct async_flt *aflt)
1306 {
1307 	if (aflt->flt_status & OPL_ECC_ISYNC_TRAP)
1308 		return (TRAP_TYPE_ECC_I);
1309 	if (aflt->flt_status & OPL_ECC_DSYNC_TRAP)
1310 		return (TRAP_TYPE_ECC_D);
1311 	if (aflt->flt_status & OPL_ECC_URGENT_TRAP)
1312 		return (TRAP_TYPE_URGENT);
1313 	return (-1);
1314 }
1315 
1316 /*
1317  * Encode the data saved in the opl_async_flt_t struct into
1318  * the FM ereport payload.
1319  */
1320 /* ARGSUSED */
1321 static void
1322 cpu_payload_add_aflt(struct async_flt *aflt, nvlist_t *payload,
1323 		nvlist_t *resource)
1324 {
1325 	opl_async_flt_t *opl_flt = (opl_async_flt_t *)aflt;
1326 	char unum[UNUM_NAMLEN];
1327 	char sbuf[21]; /* sizeof (UINT64_MAX) + '\0' */
1328 	int len;
1329 
1330 
1331 	if (aflt->flt_payload & FM_EREPORT_PAYLOAD_FLAG_SFSR) {
1332 		fm_payload_set(payload, FM_EREPORT_PAYLOAD_NAME_SFSR,
1333 			DATA_TYPE_UINT64, aflt->flt_stat, NULL);
1334 	}
1335 	if (aflt->flt_payload & FM_EREPORT_PAYLOAD_FLAG_SFAR) {
1336 		fm_payload_set(payload, FM_EREPORT_PAYLOAD_NAME_SFAR,
1337 			DATA_TYPE_UINT64, aflt->flt_addr, NULL);
1338 	}
1339 	if (aflt->flt_payload & FM_EREPORT_PAYLOAD_FLAG_UGESR) {
1340 		fm_payload_set(payload, FM_EREPORT_PAYLOAD_NAME_UGESR,
1341 			DATA_TYPE_UINT64, aflt->flt_stat, NULL);
1342 	}
1343 	if (aflt->flt_payload & FM_EREPORT_PAYLOAD_FLAG_PC) {
1344 		fm_payload_set(payload, FM_EREPORT_PAYLOAD_NAME_PC,
1345 		    DATA_TYPE_UINT64, (uint64_t)aflt->flt_pc, NULL);
1346 	}
1347 	if (aflt->flt_payload & FM_EREPORT_PAYLOAD_FLAG_TL) {
1348 		fm_payload_set(payload, FM_EREPORT_PAYLOAD_NAME_TL,
1349 		    DATA_TYPE_UINT8, (uint8_t)aflt->flt_tl, NULL);
1350 	}
1351 	if (aflt->flt_payload & FM_EREPORT_PAYLOAD_FLAG_TT) {
1352 		fm_payload_set(payload, FM_EREPORT_PAYLOAD_NAME_TT,
1353 		    DATA_TYPE_UINT8, flt_to_trap_type(aflt), NULL);
1354 	}
1355 	if (aflt->flt_payload & FM_EREPORT_PAYLOAD_FLAG_PRIV) {
1356 		fm_payload_set(payload, FM_EREPORT_PAYLOAD_NAME_PRIV,
1357 		    DATA_TYPE_BOOLEAN_VALUE,
1358 		    (aflt->flt_priv ? B_TRUE : B_FALSE), NULL);
1359 	}
1360 	if (aflt->flt_payload & FM_EREPORT_PAYLOAD_FLAG_FLT_STATUS) {
1361 		fm_payload_set(payload, FM_EREPORT_PAYLOAD_NAME_FLT_STATUS,
1362 			DATA_TYPE_UINT64, (uint64_t)aflt->flt_status, NULL);
1363 	}
1364 
1365 	switch (opl_flt->flt_eid_mod) {
1366 	case OPL_ERRID_CPU:
1367 		(void) snprintf(sbuf, sizeof (sbuf), "%llX",
1368 			(u_longlong_t)cpunodes[opl_flt->flt_eid_sid].device_id);
1369 		(void) fm_fmri_cpu_set(resource, FM_CPU_SCHEME_VERSION,
1370 			NULL, opl_flt->flt_eid_sid,
1371 			(uint8_t *)&cpunodes[opl_flt->flt_eid_sid].version,
1372 			sbuf);
1373 		fm_payload_set(payload,
1374 			FM_EREPORT_PAYLOAD_NAME_RESOURCE,
1375 			DATA_TYPE_NVLIST, resource, NULL);
1376 		break;
1377 
1378 	case OPL_ERRID_CHANNEL:
1379 		/*
1380 		 * No resource is created but the cpumem DE will find
1381 		 * the defective path by retreiving EID from SFSR which is
1382 		 * included in the payload.
1383 		 */
1384 		break;
1385 
1386 	case OPL_ERRID_MEM:
1387 		(void) cpu_get_mem_unum_aflt(0, aflt, unum, UNUM_NAMLEN, &len);
1388 		(void) fm_fmri_mem_set(resource, FM_MEM_SCHEME_VERSION,
1389 			NULL, unum, NULL, (uint64_t)-1);
1390 		fm_payload_set(payload, FM_EREPORT_PAYLOAD_NAME_RESOURCE,
1391 			DATA_TYPE_NVLIST, resource, NULL);
1392 		break;
1393 
1394 	case OPL_ERRID_PATH:
1395 		/*
1396 		 * No resource is created but the cpumem DE will find
1397 		 * the defective path by retreiving EID from SFSR which is
1398 		 * included in the payload.
1399 		 */
1400 		break;
1401 	}
1402 }
1403 
1404 /*
1405  * Returns whether fault address is valid for this error bit and
1406  * whether the address is "in memory" (i.e. pf_is_memory returns 1).
1407  */
1408 /*ARGSUSED*/
1409 static int
1410 cpu_flt_in_memory(opl_async_flt_t *opl_flt, uint64_t t_afsr_bit)
1411 {
1412 	struct async_flt *aflt = (struct async_flt *)opl_flt;
1413 
1414 	if (aflt->flt_status & (OPL_ECC_SYNC_TRAP)) {
1415 		return ((t_afsr_bit & SFSR_MEMORY) &&
1416 		    pf_is_memory(aflt->flt_addr >> MMU_PAGESHIFT));
1417 	}
1418 	return (0);
1419 }
1420 
1421 /*
1422  * In OPL SCF does the stick synchronization.
1423  */
1424 void
1425 sticksync_slave(void)
1426 {
1427 }
1428 
1429 /*
1430  * In OPL SCF does the stick synchronization.
1431  */
1432 void
1433 sticksync_master(void)
1434 {
1435 }
1436 
1437 /*
1438  * Cpu private unitialization.  OPL cpus do not use the private area.
1439  */
1440 void
1441 cpu_uninit_private(struct cpu *cp)
1442 {
1443 	cmp_delete_cpu(cp->cpu_id);
1444 }
1445 
1446 /*
1447  * Always flush an entire cache.
1448  */
1449 void
1450 cpu_error_ecache_flush(void)
1451 {
1452 	cpu_flush_ecache();
1453 }
1454 
1455 void
1456 cpu_ereport_post(struct async_flt *aflt)
1457 {
1458 	char *cpu_type, buf[FM_MAX_CLASS];
1459 	nv_alloc_t *nva = NULL;
1460 	nvlist_t *ereport, *detector, *resource;
1461 	errorq_elem_t *eqep;
1462 	char sbuf[21]; /* sizeof (UINT64_MAX) + '\0' */
1463 
1464 	if (aflt->flt_panic || panicstr) {
1465 		eqep = errorq_reserve(ereport_errorq);
1466 		if (eqep == NULL)
1467 			return;
1468 		ereport = errorq_elem_nvl(ereport_errorq, eqep);
1469 		nva = errorq_elem_nva(ereport_errorq, eqep);
1470 	} else {
1471 		ereport = fm_nvlist_create(nva);
1472 	}
1473 
1474 	/*
1475 	 * Create the scheme "cpu" FMRI.
1476 	 */
1477 	detector = fm_nvlist_create(nva);
1478 	resource = fm_nvlist_create(nva);
1479 	switch (cpunodes[aflt->flt_inst].implementation) {
1480 	case OLYMPUS_C_IMPL:
1481 		cpu_type = FM_EREPORT_CPU_SPARC64_VI;
1482 		break;
1483 	default:
1484 		cpu_type = FM_EREPORT_CPU_UNSUPPORTED;
1485 		break;
1486 	}
1487 	(void) snprintf(sbuf, sizeof (sbuf), "%llX",
1488 	    (u_longlong_t)cpunodes[aflt->flt_inst].device_id);
1489 	(void) fm_fmri_cpu_set(detector, FM_CPU_SCHEME_VERSION, NULL,
1490 	    aflt->flt_inst, (uint8_t *)&cpunodes[aflt->flt_inst].version,
1491 	    sbuf);
1492 
1493 	/*
1494 	 * Encode all the common data into the ereport.
1495 	 */
1496 	(void) snprintf(buf, FM_MAX_CLASS, "%s.%s.%s",
1497 	    FM_ERROR_CPU, cpu_type, aflt->flt_erpt_class);
1498 
1499 	fm_ereport_set(ereport, FM_EREPORT_VERSION, buf,
1500 	    fm_ena_generate(aflt->flt_id, FM_ENA_FMT1), detector, NULL);
1501 
1502 	/*
1503 	 * Encode the error specific data that was saved in
1504 	 * the async_flt structure into the ereport.
1505 	 */
1506 	cpu_payload_add_aflt(aflt, ereport, resource);
1507 
1508 	if (aflt->flt_panic || panicstr) {
1509 		errorq_commit(ereport_errorq, eqep, ERRORQ_SYNC);
1510 	} else {
1511 		(void) fm_ereport_post(ereport, EVCH_TRYHARD);
1512 		fm_nvlist_destroy(ereport, FM_NVA_FREE);
1513 		fm_nvlist_destroy(detector, FM_NVA_FREE);
1514 		fm_nvlist_destroy(resource, FM_NVA_FREE);
1515 	}
1516 }
1517 
1518 void
1519 cpu_run_bus_error_handlers(struct async_flt *aflt, int expected)
1520 {
1521 	int status;
1522 	ddi_fm_error_t de;
1523 
1524 	bzero(&de, sizeof (ddi_fm_error_t));
1525 
1526 	de.fme_version = DDI_FME_VERSION;
1527 	de.fme_ena = fm_ena_generate(aflt->flt_id, FM_ENA_FMT1);
1528 	de.fme_flag = expected;
1529 	de.fme_bus_specific = (void *)aflt->flt_addr;
1530 	status = ndi_fm_handler_dispatch(ddi_root_node(), NULL, &de);
1531 	if ((aflt->flt_prot == AFLT_PROT_NONE) && (status == DDI_FM_FATAL))
1532 		aflt->flt_panic = 1;
1533 }
1534 
1535 void
1536 cpu_errorq_dispatch(char *error_class, void *payload, size_t payload_sz,
1537     errorq_t *eqp, uint_t flag)
1538 {
1539 	struct async_flt *aflt = (struct async_flt *)payload;
1540 
1541 	aflt->flt_erpt_class = error_class;
1542 	errorq_dispatch(eqp, payload, payload_sz, flag);
1543 }
1544 
1545 void
1546 adjust_hw_copy_limits(int ecache_size)
1547 {
1548 	/*
1549 	 * Set hw copy limits.
1550 	 *
1551 	 * /etc/system will be parsed later and can override one or more
1552 	 * of these settings.
1553 	 *
1554 	 * At this time, ecache size seems only mildly relevant.
1555 	 * We seem to run into issues with the d-cache and stalls
1556 	 * we see on misses.
1557 	 *
1558 	 * Cycle measurement indicates that 2 byte aligned copies fare
1559 	 * little better than doing things with VIS at around 512 bytes.
1560 	 * 4 byte aligned shows promise until around 1024 bytes. 8 Byte
1561 	 * aligned is faster whenever the source and destination data
1562 	 * in cache and the total size is less than 2 Kbytes.  The 2K
1563 	 * limit seems to be driven by the 2K write cache.
1564 	 * When more than 2K of copies are done in non-VIS mode, stores
1565 	 * backup in the write cache.  In VIS mode, the write cache is
1566 	 * bypassed, allowing faster cache-line writes aligned on cache
1567 	 * boundaries.
1568 	 *
1569 	 * In addition, in non-VIS mode, there is no prefetching, so
1570 	 * for larger copies, the advantage of prefetching to avoid even
1571 	 * occasional cache misses is enough to justify using the VIS code.
1572 	 *
1573 	 * During testing, it was discovered that netbench ran 3% slower
1574 	 * when hw_copy_limit_8 was 2K or larger.  Apparently for server
1575 	 * applications, data is only used once (copied to the output
1576 	 * buffer, then copied by the network device off the system).  Using
1577 	 * the VIS copy saves more L2 cache state.  Network copies are
1578 	 * around 1.3K to 1.5K in size for historical reasons.
1579 	 *
1580 	 * Therefore, a limit of 1K bytes will be used for the 8 byte
1581 	 * aligned copy even for large caches and 8 MB ecache.  The
1582 	 * infrastructure to allow different limits for different sized
1583 	 * caches is kept to allow further tuning in later releases.
1584 	 */
1585 
1586 	if (min_ecache_size == 0 && use_hw_bcopy) {
1587 		/*
1588 		 * First time through - should be before /etc/system
1589 		 * is read.
1590 		 * Could skip the checks for zero but this lets us
1591 		 * preserve any debugger rewrites.
1592 		 */
1593 		if (hw_copy_limit_1 == 0) {
1594 			hw_copy_limit_1 = VIS_COPY_THRESHOLD;
1595 			priv_hcl_1 = hw_copy_limit_1;
1596 		}
1597 		if (hw_copy_limit_2 == 0) {
1598 			hw_copy_limit_2 = 2 * VIS_COPY_THRESHOLD;
1599 			priv_hcl_2 = hw_copy_limit_2;
1600 		}
1601 		if (hw_copy_limit_4 == 0) {
1602 			hw_copy_limit_4 = 4 * VIS_COPY_THRESHOLD;
1603 			priv_hcl_4 = hw_copy_limit_4;
1604 		}
1605 		if (hw_copy_limit_8 == 0) {
1606 			hw_copy_limit_8 = 4 * VIS_COPY_THRESHOLD;
1607 			priv_hcl_8 = hw_copy_limit_8;
1608 		}
1609 		min_ecache_size = ecache_size;
1610 	} else {
1611 		/*
1612 		 * MP initialization. Called *after* /etc/system has
1613 		 * been parsed. One CPU has already been initialized.
1614 		 * Need to cater for /etc/system having scragged one
1615 		 * of our values.
1616 		 */
1617 		if (ecache_size == min_ecache_size) {
1618 			/*
1619 			 * Same size ecache. We do nothing unless we
1620 			 * have a pessimistic ecache setting. In that
1621 			 * case we become more optimistic (if the cache is
1622 			 * large enough).
1623 			 */
1624 			if (hw_copy_limit_8 == 4 * VIS_COPY_THRESHOLD) {
1625 				/*
1626 				 * Need to adjust hw_copy_limit* from our
1627 				 * pessimistic uniprocessor value to a more
1628 				 * optimistic UP value *iff* it hasn't been
1629 				 * reset.
1630 				 */
1631 				if ((ecache_size > 1048576) &&
1632 				    (priv_hcl_8 == hw_copy_limit_8)) {
1633 					if (ecache_size <= 2097152)
1634 						hw_copy_limit_8 = 4 *
1635 						    VIS_COPY_THRESHOLD;
1636 					else if (ecache_size <= 4194304)
1637 						hw_copy_limit_8 = 4 *
1638 						    VIS_COPY_THRESHOLD;
1639 					else
1640 						hw_copy_limit_8 = 4 *
1641 						    VIS_COPY_THRESHOLD;
1642 					priv_hcl_8 = hw_copy_limit_8;
1643 				}
1644 			}
1645 		} else if (ecache_size < min_ecache_size) {
1646 			/*
1647 			 * A different ecache size. Can this even happen?
1648 			 */
1649 			if (priv_hcl_8 == hw_copy_limit_8) {
1650 				/*
1651 				 * The previous value that we set
1652 				 * is unchanged (i.e., it hasn't been
1653 				 * scragged by /etc/system). Rewrite it.
1654 				 */
1655 				if (ecache_size <= 1048576)
1656 					hw_copy_limit_8 = 8 *
1657 					    VIS_COPY_THRESHOLD;
1658 				else if (ecache_size <= 2097152)
1659 					hw_copy_limit_8 = 8 *
1660 					    VIS_COPY_THRESHOLD;
1661 				else if (ecache_size <= 4194304)
1662 					hw_copy_limit_8 = 8 *
1663 					    VIS_COPY_THRESHOLD;
1664 				else
1665 					hw_copy_limit_8 = 10 *
1666 					    VIS_COPY_THRESHOLD;
1667 				priv_hcl_8 = hw_copy_limit_8;
1668 				min_ecache_size = ecache_size;
1669 			}
1670 		}
1671 	}
1672 }
1673 
1674 #define	VIS_BLOCKSIZE		64
1675 
1676 int
1677 dtrace_blksuword32_err(uintptr_t addr, uint32_t *data)
1678 {
1679 	int ret, watched;
1680 
1681 	watched = watch_disable_addr((void *)addr, VIS_BLOCKSIZE, S_WRITE);
1682 	ret = dtrace_blksuword32(addr, data, 0);
1683 	if (watched)
1684 		watch_enable_addr((void *)addr, VIS_BLOCKSIZE, S_WRITE);
1685 
1686 	return (ret);
1687 }
1688 
1689 void
1690 opl_cpu_reg_init()
1691 {
1692 	uint64_t	this_cpu_log;
1693 
1694 	/*
1695 	 * We do not need to re-initialize cpu0 registers.
1696 	 */
1697 	if (cpu[getprocessorid()] == &cpu0)
1698 		return;
1699 
1700 	/*
1701 	 * Initialize Error log Scratch register for error handling.
1702 	 */
1703 
1704 	this_cpu_log = va_to_pa((void*)(((uint64_t)opl_err_log) +
1705 		ERRLOG_BUFSZ * (getprocessorid())));
1706 	opl_error_setup(this_cpu_log);
1707 
1708 	/*
1709 	 * Enable MMU translating multiple page sizes for
1710 	 * sITLB and sDTLB.
1711 	 */
1712 	opl_mpg_enable();
1713 }
1714 
1715 /*
1716  * Queue one event in ue_queue based on ecc_type_to_info entry.
1717  */
1718 static void
1719 cpu_queue_one_event(opl_async_flt_t *opl_flt, char *reason,
1720     ecc_type_to_info_t *eccp)
1721 {
1722 	struct async_flt *aflt = (struct async_flt *)opl_flt;
1723 
1724 	if (reason &&
1725 	    strlen(reason) + strlen(eccp->ec_reason) < MAX_REASON_STRING) {
1726 		(void) strcat(reason, eccp->ec_reason);
1727 	}
1728 
1729 	opl_flt->flt_bit = eccp->ec_afsr_bit;
1730 	opl_flt->flt_type = eccp->ec_flt_type;
1731 	aflt->flt_in_memory = cpu_flt_in_memory(opl_flt, opl_flt->flt_bit);
1732 	aflt->flt_payload = eccp->ec_err_payload;
1733 
1734 	ASSERT(aflt->flt_status & (OPL_ECC_SYNC_TRAP|OPL_ECC_URGENT_TRAP));
1735 	cpu_errorq_dispatch(eccp->ec_err_class,
1736 		(void *)opl_flt, sizeof (opl_async_flt_t),
1737 		ue_queue,
1738 		aflt->flt_panic);
1739 }
1740 
1741 /*
1742  * Queue events on async event queue one event per error bit.
1743  * Return number of events queued.
1744  */
1745 int
1746 cpu_queue_events(opl_async_flt_t *opl_flt, char *reason, uint64_t t_afsr_errs)
1747 {
1748 	struct async_flt *aflt = (struct async_flt *)opl_flt;
1749 	ecc_type_to_info_t *eccp;
1750 	int nevents = 0;
1751 
1752 	/*
1753 	 * Queue expected errors, error bit and fault type must must match
1754 	 * in the ecc_type_to_info table.
1755 	 */
1756 	for (eccp = ecc_type_to_info; t_afsr_errs != 0 && eccp->ec_desc != NULL;
1757 		eccp++) {
1758 		if ((eccp->ec_afsr_bit & t_afsr_errs) != 0 &&
1759 		    (eccp->ec_flags & aflt->flt_status) != 0) {
1760 			/*
1761 			 * UE error event can be further
1762 			 * classified/breakdown into finer granularity
1763 			 * based on the flt_eid_mod value set by HW.  We do
1764 			 * special handling here so that we can report UE
1765 			 * error in finer granularity as ue_mem,
1766 			 * ue_channel, ue_cpu or ue_path.
1767 			 */
1768 			if (eccp->ec_flt_type == OPL_CPU_SYNC_UE) {
1769 				opl_flt->flt_eid_mod =
1770 					(aflt->flt_stat & SFSR_EID_MOD)
1771 					>> SFSR_EID_MOD_SHIFT;
1772 				opl_flt->flt_eid_sid =
1773 					(aflt->flt_stat & SFSR_EID_SID)
1774 					>> SFSR_EID_SID_SHIFT;
1775 				/*
1776 				 * Need to advance eccp pointer by flt_eid_mod
1777 				 * so that we get an appropriate ecc pointer
1778 				 *
1779 				 * EID			# of advances
1780 				 * ----------------------------------
1781 				 * OPL_ERRID_MEM	0
1782 				 * OPL_ERRID_CHANNEL	1
1783 				 * OPL_ERRID_CPU	2
1784 				 * OPL_ERRID_PATH	3
1785 				 */
1786 				eccp += opl_flt->flt_eid_mod;
1787 			}
1788 			cpu_queue_one_event(opl_flt, reason, eccp);
1789 			t_afsr_errs &= ~eccp->ec_afsr_bit;
1790 			nevents++;
1791 		}
1792 	}
1793 
1794 	return (nevents);
1795 }
1796 
1797 /*
1798  * Sync. error wrapper functions.
1799  * We use these functions in order to transfer here from the
1800  * nucleus trap handler information about trap type (data or
1801  * instruction) and trap level (0 or above 0). This way we
1802  * get rid of using SFSR's reserved bits.
1803  */
1804 
1805 #define	OPL_SYNC_TL0	0
1806 #define	OPL_SYNC_TL1	1
1807 #define	OPL_ISYNC_ERR	0
1808 #define	OPL_DSYNC_ERR	1
1809 
1810 void
1811 opl_cpu_isync_tl0_error(struct regs *rp, ulong_t p_sfar, ulong_t p_sfsr)
1812 {
1813 	uint64_t t_sfar = p_sfar;
1814 	uint64_t t_sfsr = p_sfsr;
1815 
1816 	opl_cpu_sync_error(rp, t_sfar, t_sfsr,
1817 	    OPL_SYNC_TL0, OPL_ISYNC_ERR);
1818 }
1819 
1820 void
1821 opl_cpu_isync_tl1_error(struct regs *rp, ulong_t p_sfar, ulong_t p_sfsr)
1822 {
1823 	uint64_t t_sfar = p_sfar;
1824 	uint64_t t_sfsr = p_sfsr;
1825 
1826 	opl_cpu_sync_error(rp, t_sfar, t_sfsr,
1827 	    OPL_SYNC_TL1, OPL_ISYNC_ERR);
1828 }
1829 
1830 void
1831 opl_cpu_dsync_tl0_error(struct regs *rp, ulong_t p_sfar, ulong_t p_sfsr)
1832 {
1833 	uint64_t t_sfar = p_sfar;
1834 	uint64_t t_sfsr = p_sfsr;
1835 
1836 	opl_cpu_sync_error(rp, t_sfar, t_sfsr,
1837 	    OPL_SYNC_TL0, OPL_DSYNC_ERR);
1838 }
1839 
1840 void
1841 opl_cpu_dsync_tl1_error(struct regs *rp, ulong_t p_sfar, ulong_t p_sfsr)
1842 {
1843 	uint64_t t_sfar = p_sfar;
1844 	uint64_t t_sfsr = p_sfsr;
1845 
1846 	opl_cpu_sync_error(rp, t_sfar, t_sfsr,
1847 	    OPL_SYNC_TL1, OPL_DSYNC_ERR);
1848 }
1849 
1850 /*
1851  * The fj sync err handler transfers control here for UE, BERR, TO, TLB_MUL
1852  * and TLB_PRT.
1853  * This function is designed based on cpu_deferred_error().
1854  */
1855 
1856 static void
1857 opl_cpu_sync_error(struct regs *rp, ulong_t t_sfar, ulong_t t_sfsr,
1858     uint_t tl, uint_t derr)
1859 {
1860 	opl_async_flt_t opl_flt;
1861 	struct async_flt *aflt;
1862 	int trampolined = 0;
1863 	char pr_reason[MAX_REASON_STRING];
1864 	uint64_t log_sfsr;
1865 	int expected = DDI_FM_ERR_UNEXPECTED;
1866 	ddi_acc_hdl_t *hp;
1867 
1868 	/*
1869 	 * We need to look at p_flag to determine if the thread detected an
1870 	 * error while dumping core.  We can't grab p_lock here, but it's ok
1871 	 * because we just need a consistent snapshot and we know that everyone
1872 	 * else will store a consistent set of bits while holding p_lock.  We
1873 	 * don't have to worry about a race because SDOCORE is set once prior
1874 	 * to doing i/o from the process's address space and is never cleared.
1875 	 */
1876 	uint_t pflag = ttoproc(curthread)->p_flag;
1877 
1878 	pr_reason[0] = '\0';
1879 
1880 	/*
1881 	 * handle the specific error
1882 	 */
1883 	bzero(&opl_flt, sizeof (opl_async_flt_t));
1884 	aflt = (struct async_flt *)&opl_flt;
1885 	aflt->flt_id = gethrtime_waitfree();
1886 	aflt->flt_bus_id = getprocessorid();
1887 	aflt->flt_inst = CPU->cpu_id;
1888 	aflt->flt_stat = t_sfsr;
1889 	aflt->flt_addr = t_sfar;
1890 	aflt->flt_pc = (caddr_t)rp->r_pc;
1891 	aflt->flt_prot = (uchar_t)AFLT_PROT_NONE;
1892 	aflt->flt_class = (uchar_t)CPU_FAULT;
1893 	aflt->flt_priv = (uchar_t)
1894 		(tl == 1 ? 1 : ((rp->r_tstate & TSTATE_PRIV) ?  1 : 0));
1895 	aflt->flt_tl = (uchar_t)tl;
1896 	aflt->flt_panic = (uchar_t)(tl != 0 || aft_testfatal != 0 ||
1897 	    (t_sfsr & (SFSR_TLB_MUL|SFSR_TLB_PRT)) != 0);
1898 	aflt->flt_core = (pflag & SDOCORE) ? 1 : 0;
1899 	aflt->flt_status = (derr) ? OPL_ECC_DSYNC_TRAP : OPL_ECC_ISYNC_TRAP;
1900 
1901 	/*
1902 	 * If SFSR.FV is not set, both SFSR and SFAR/SFPAR values are uncertain.
1903 	 * So, clear all error bits to avoid mis-handling and force the system
1904 	 * panicked.
1905 	 * We skip all the procedures below down to the panic message call.
1906 	 */
1907 	if (!(t_sfsr & SFSR_FV)) {
1908 		opl_flt.flt_type = OPL_CPU_INV_SFSR;
1909 		aflt->flt_panic = 1;
1910 		aflt->flt_payload = FM_EREPORT_PAYLOAD_SYNC;
1911 		cpu_errorq_dispatch(FM_EREPORT_CPU_INV_SFSR,
1912 			(void *)&opl_flt, sizeof (opl_async_flt_t), ue_queue,
1913 			aflt->flt_panic);
1914 		fm_panic("%sErrors(s)", "invalid SFSR");
1915 	}
1916 
1917 	/*
1918 	 * If either UE and MK bit is off, this is not valid UE error.
1919 	 * If it is not valid UE error, clear UE & MK_UE bits to prevent
1920 	 * mis-handling below.
1921 	 * aflt->flt_stat keeps the original bits as a reference.
1922 	 */
1923 	if ((t_sfsr & (SFSR_MK_UE|SFSR_UE)) !=
1924 	    (SFSR_MK_UE|SFSR_UE)) {
1925 		t_sfsr &= ~(SFSR_MK_UE|SFSR_UE);
1926 	}
1927 
1928 	/*
1929 	 * If the trap occurred in privileged mode at TL=0, we need to check to
1930 	 * see if we were executing in the kernel under on_trap() or t_lofault
1931 	 * protection.  If so, modify the saved registers so that we return
1932 	 * from the trap to the appropriate trampoline routine.
1933 	 */
1934 	if (!aflt->flt_panic && aflt->flt_priv && tl == 0) {
1935 		if (curthread->t_ontrap != NULL) {
1936 			on_trap_data_t *otp = curthread->t_ontrap;
1937 
1938 			if (otp->ot_prot & OT_DATA_EC) {
1939 				aflt->flt_prot = (uchar_t)AFLT_PROT_EC;
1940 				otp->ot_trap |= (ushort_t)OT_DATA_EC;
1941 				rp->r_pc = otp->ot_trampoline;
1942 				rp->r_npc = rp->r_pc + 4;
1943 				trampolined = 1;
1944 			}
1945 
1946 			if ((t_sfsr & (SFSR_TO | SFSR_BERR)) &&
1947 			    (otp->ot_prot & OT_DATA_ACCESS)) {
1948 				aflt->flt_prot = (uchar_t)AFLT_PROT_ACCESS;
1949 				otp->ot_trap |= (ushort_t)OT_DATA_ACCESS;
1950 				rp->r_pc = otp->ot_trampoline;
1951 				rp->r_npc = rp->r_pc + 4;
1952 				trampolined = 1;
1953 				/*
1954 				 * for peeks and caut_gets errors are expected
1955 				 */
1956 				hp = (ddi_acc_hdl_t *)otp->ot_handle;
1957 				if (!hp)
1958 					expected = DDI_FM_ERR_PEEK;
1959 				else if (hp->ah_acc.devacc_attr_access ==
1960 				    DDI_CAUTIOUS_ACC)
1961 					expected = DDI_FM_ERR_EXPECTED;
1962 			}
1963 
1964 		} else if (curthread->t_lofault) {
1965 			aflt->flt_prot = AFLT_PROT_COPY;
1966 			rp->r_g1 = EFAULT;
1967 			rp->r_pc = curthread->t_lofault;
1968 			rp->r_npc = rp->r_pc + 4;
1969 			trampolined = 1;
1970 		}
1971 	}
1972 
1973 	/*
1974 	 * If we're in user mode or we're doing a protected copy, we either
1975 	 * want the ASTON code below to send a signal to the user process
1976 	 * or we want to panic if aft_panic is set.
1977 	 *
1978 	 * If we're in privileged mode and we're not doing a copy, then we
1979 	 * need to check if we've trampolined.  If we haven't trampolined,
1980 	 * we should panic.
1981 	 */
1982 	if (!aflt->flt_priv || aflt->flt_prot == AFLT_PROT_COPY) {
1983 		if (t_sfsr & (SFSR_ERRS & ~(SFSR_BERR | SFSR_TO)))
1984 			aflt->flt_panic |= aft_panic;
1985 	} else if (!trampolined) {
1986 		aflt->flt_panic = 1;
1987 	}
1988 
1989 	/*
1990 	 * If we've trampolined due to a privileged TO or BERR, or if an
1991 	 * unprivileged TO or BERR occurred, we don't want to enqueue an
1992 	 * event for that TO or BERR.  Queue all other events (if any) besides
1993 	 * the TO/BERR.
1994 	 */
1995 	log_sfsr = t_sfsr;
1996 	if (trampolined) {
1997 		log_sfsr &= ~(SFSR_TO | SFSR_BERR);
1998 	} else if (!aflt->flt_priv) {
1999 		/*
2000 		 * User mode, suppress messages if
2001 		 * cpu_berr_to_verbose is not set.
2002 		 */
2003 		if (!cpu_berr_to_verbose)
2004 			log_sfsr &= ~(SFSR_TO | SFSR_BERR);
2005 	}
2006 
2007 	if (((log_sfsr & SFSR_ERRS) &&
2008 		(cpu_queue_events(&opl_flt, pr_reason, t_sfsr) == 0)) ||
2009 	    ((t_sfsr & SFSR_ERRS) == 0)) {
2010 		opl_flt.flt_type = OPL_CPU_INV_SFSR;
2011 		aflt->flt_payload = FM_EREPORT_PAYLOAD_SYNC;
2012 		cpu_errorq_dispatch(FM_EREPORT_CPU_INV_SFSR,
2013 			(void *)&opl_flt, sizeof (opl_async_flt_t), ue_queue,
2014 			aflt->flt_panic);
2015 	}
2016 
2017 	if (t_sfsr & (SFSR_UE|SFSR_TO|SFSR_BERR)) {
2018 		cpu_run_bus_error_handlers(aflt, expected);
2019 	}
2020 
2021 	/*
2022 	 * Panic here if aflt->flt_panic has been set.  Enqueued errors will
2023 	 * be logged as part of the panic flow.
2024 	 */
2025 	if (aflt->flt_panic) {
2026 		if (pr_reason[0] == 0)
2027 			strcpy(pr_reason, "invalid SFSR ");
2028 
2029 		fm_panic("%sErrors(s)", pr_reason);
2030 	}
2031 
2032 	/*
2033 	 * If we queued an error and we are going to return from the trap and
2034 	 * the error was in user mode or inside of a copy routine, set AST flag
2035 	 * so the queue will be drained before returning to user mode.  The
2036 	 * AST processing will also act on our failure policy.
2037 	 */
2038 	if (!aflt->flt_priv || aflt->flt_prot == AFLT_PROT_COPY) {
2039 		int pcb_flag = 0;
2040 
2041 		if (t_sfsr & (SFSR_ERRS &
2042 			~(SFSR_BERR | SFSR_TO)))
2043 			pcb_flag |= ASYNC_HWERR;
2044 
2045 		if (t_sfsr & SFSR_BERR)
2046 			pcb_flag |= ASYNC_BERR;
2047 
2048 		if (t_sfsr & SFSR_TO)
2049 			pcb_flag |= ASYNC_BTO;
2050 
2051 		ttolwp(curthread)->lwp_pcb.pcb_flags |= pcb_flag;
2052 		aston(curthread);
2053 	}
2054 }
2055 
2056 /*ARGSUSED*/
2057 void
2058 opl_cpu_urgent_error(struct regs *rp, ulong_t p_ugesr, ulong_t tl)
2059 {
2060 	opl_async_flt_t opl_flt;
2061 	struct async_flt *aflt;
2062 	char pr_reason[MAX_REASON_STRING];
2063 
2064 	/* normalize tl */
2065 	tl = (tl >= 2 ? 1 : 0);
2066 	pr_reason[0] = '\0';
2067 
2068 	bzero(&opl_flt, sizeof (opl_async_flt_t));
2069 	aflt = (struct async_flt *)&opl_flt;
2070 	aflt->flt_id = gethrtime_waitfree();
2071 	aflt->flt_bus_id = getprocessorid();
2072 	aflt->flt_inst = CPU->cpu_id;
2073 	aflt->flt_stat = p_ugesr;
2074 	aflt->flt_pc = (caddr_t)rp->r_pc;
2075 	aflt->flt_class = (uchar_t)CPU_FAULT;
2076 	aflt->flt_tl = tl;
2077 	aflt->flt_priv = (uchar_t)
2078 		(tl == 1 ? 1 : ((rp->r_tstate & TSTATE_PRIV) ?  1 : 0));
2079 	aflt->flt_status = OPL_ECC_URGENT_TRAP;
2080 	aflt->flt_panic = 1;
2081 	/*
2082 	 * HW does not set mod/sid in case of urgent error.
2083 	 * So we have to set it here.
2084 	 */
2085 	opl_flt.flt_eid_mod = OPL_ERRID_CPU;
2086 	opl_flt.flt_eid_sid = aflt->flt_inst;
2087 
2088 	if (cpu_queue_events(&opl_flt, pr_reason, p_ugesr) == 0) {
2089 		opl_flt.flt_type = OPL_CPU_INV_UGESR;
2090 		aflt->flt_payload = FM_EREPORT_PAYLOAD_URGENT;
2091 		cpu_errorq_dispatch(FM_EREPORT_CPU_INV_URG,
2092 			(void *)&opl_flt, sizeof (opl_async_flt_t),
2093 			ue_queue, aflt->flt_panic);
2094 	}
2095 
2096 	fm_panic("Urgent Error");
2097 }
2098 
2099 /*
2100  * Initialization error counters resetting.
2101  */
2102 /* ARGSUSED */
2103 static void
2104 opl_ras_online(void *arg, cpu_t *cp, cyc_handler_t *hdlr, cyc_time_t *when)
2105 {
2106 	hdlr->cyh_func = (cyc_func_t)ras_cntr_reset;
2107 	hdlr->cyh_level = CY_LOW_LEVEL;
2108 	hdlr->cyh_arg = (void *)(uintptr_t)cp->cpu_id;
2109 
2110 	when->cyt_when = cp->cpu_id * (((hrtime_t)NANOSEC * 10)/ NCPU);
2111 	when->cyt_interval = (hrtime_t)NANOSEC * opl_async_check_interval;
2112 }
2113 
2114 void
2115 cpu_mp_init(void)
2116 {
2117 	cyc_omni_handler_t hdlr;
2118 
2119 	hdlr.cyo_online = opl_ras_online;
2120 	hdlr.cyo_offline = NULL;
2121 	hdlr.cyo_arg = NULL;
2122 	mutex_enter(&cpu_lock);
2123 	(void) cyclic_add_omni(&hdlr);
2124 	mutex_exit(&cpu_lock);
2125 }
2126 
2127 /*ARGSUSED*/
2128 void
2129 mmu_init_kernel_pgsz(struct hat *hat)
2130 {
2131 }
2132 
2133 size_t
2134 mmu_get_kernel_lpsize(size_t lpsize)
2135 {
2136 	uint_t tte;
2137 
2138 	if (lpsize == 0) {
2139 		/* no setting for segkmem_lpsize in /etc/system: use default */
2140 		return (MMU_PAGESIZE4M);
2141 	}
2142 
2143 	for (tte = TTE8K; tte <= TTE4M; tte++) {
2144 		if (lpsize == TTEBYTES(tte))
2145 			return (lpsize);
2146 	}
2147 
2148 	return (TTEBYTES(TTE8K));
2149 }
2150 
2151 /*
2152  * The following are functions that are unused in
2153  * OPL cpu module. They are defined here to resolve
2154  * dependencies in the "unix" module.
2155  * Unused functions that should never be called in
2156  * OPL are coded with ASSERT(0).
2157  */
2158 
2159 void
2160 cpu_disable_errors(void)
2161 {}
2162 
2163 void
2164 cpu_enable_errors(void)
2165 { ASSERT(0); }
2166 
2167 /*ARGSUSED*/
2168 void
2169 cpu_ce_scrub_mem_err(struct async_flt *ecc, boolean_t t)
2170 { ASSERT(0); }
2171 
2172 /*ARGSUSED*/
2173 void
2174 cpu_faulted_enter(struct cpu *cp)
2175 {}
2176 
2177 /*ARGSUSED*/
2178 void
2179 cpu_faulted_exit(struct cpu *cp)
2180 {}
2181 
2182 /*ARGSUSED*/
2183 void
2184 cpu_check_allcpus(struct async_flt *aflt)
2185 {}
2186 
2187 /*ARGSUSED*/
2188 void
2189 cpu_ce_log_err(struct async_flt *aflt, errorq_elem_t *t)
2190 { ASSERT(0); }
2191 
2192 /*ARGSUSED*/
2193 void
2194 cpu_check_ce(int flag, uint64_t pa, caddr_t va, uint_t psz)
2195 { ASSERT(0); }
2196 
2197 /*ARGSUSED*/
2198 void
2199 cpu_ce_count_unum(struct async_flt *ecc, int len, char *unum)
2200 { ASSERT(0); }
2201 
2202 /*ARGSUSED*/
2203 void
2204 cpu_busy_ecache_scrub(struct cpu *cp)
2205 {}
2206 
2207 /*ARGSUSED*/
2208 void
2209 cpu_idle_ecache_scrub(struct cpu *cp)
2210 {}
2211 
2212 /* ARGSUSED */
2213 void
2214 cpu_change_speed(uint64_t divisor, uint64_t arg2)
2215 { ASSERT(0); }
2216 
2217 void
2218 cpu_init_cache_scrub(void)
2219 {}
2220 
2221 /* ARGSUSED */
2222 int
2223 cpu_get_mem_sid(char *unum, char *buf, int buflen, int *lenp)
2224 {
2225 	return (ENOTSUP);
2226 }
2227 
2228 /* ARGSUSED */
2229 int
2230 cpu_get_mem_addr(char *unum, char *sid, uint64_t offset, uint64_t *addrp)
2231 {
2232 	return (ENOTSUP);
2233 }
2234 
2235 /* ARGSUSED */
2236 int
2237 cpu_get_mem_offset(uint64_t flt_addr, uint64_t *offp)
2238 {
2239 	return (ENOTSUP);
2240 }
2241 
2242 /*ARGSUSED*/
2243 void
2244 itlb_rd_entry(uint_t entry, tte_t *tte, uint64_t *va_tag)
2245 { ASSERT(0); }
2246 
2247 /*ARGSUSED*/
2248 void
2249 dtlb_rd_entry(uint_t entry, tte_t *tte, uint64_t *va_tag)
2250 { ASSERT(0); }
2251