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 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #include <sys/types.h>
27 #include <sys/systm.h>
28 #include <sys/ddi.h>
29 #include <sys/sysmacros.h>
30 #include <sys/archsystm.h>
31 #include <sys/vmsystm.h>
32 #include <sys/machparam.h>
33 #include <sys/machsystm.h>
34 #include <sys/machthread.h>
35 #include <sys/cpu.h>
36 #include <sys/cmp.h>
37 #include <sys/elf_SPARC.h>
38 #include <vm/hat_sfmmu.h>
39 #include <vm/seg_kmem.h>
40 #include <sys/cpuvar.h>
41 #include <sys/cheetahregs.h>
42 #include <sys/us3_module.h>
43 #include <sys/async.h>
44 #include <sys/cmn_err.h>
45 #include <sys/debug.h>
46 #include <sys/dditypes.h>
47 #include <sys/prom_debug.h>
48 #include <sys/prom_plat.h>
49 #include <sys/cpu_module.h>
50 #include <sys/sysmacros.h>
51 #include <sys/intreg.h>
52 #include <sys/clock.h>
53 #include <sys/platform_module.h>
54 #include <sys/machtrap.h>
55 #include <sys/ontrap.h>
56 #include <sys/panic.h>
57 #include <sys/memlist.h>
58 #include <sys/bootconf.h>
59 #include <sys/ivintr.h>
60 #include <sys/atomic.h>
61 #include <sys/fm/protocol.h>
62 #include <sys/fm/cpu/UltraSPARC-III.h>
63 #include <sys/errclassify.h>
64 
65 #ifdef	CHEETAHPLUS_ERRATUM_25
66 #include <sys/cyclic.h>
67 #endif	/* CHEETAHPLUS_ERRATUM_25 */
68 
69 /* cpu estar private data */
70 typedef struct {
71 	uint8_t state : 7;
72 	uint8_t valid : 1;
73 } mcu_fsm_def_t;
74 mcu_fsm_def_t mcu_fsm_init_state[NCPU];
75 
76 #if defined(JALAPENO) && defined(JALAPENO_ERRATA_85)
77 /*
78  * jp_errata_85_enable can be set to 0 in /etc/system to disable
79  * JP Errata 85 workaround.
80  *
81  * jp_errata_85_allow_slow_scrub is usually set to !jp_errata_85_enable,
82  * but can be overridden in /etc/system.  If set, it allows the scrubber
83  * to run in 1/2 or 1/32 mode.  If a cpu is vulnerable to errata 85,
84  * this value should be zero.
85  *
86  * jp_errata_85_active is an internal variable and must not be
87  * set/changed via /etc/system or in any other way.
88  */
89 extern int	jp_errata_85_enable;	/* for /etc/system use */
90 extern int	jp_errata_85_allow_slow_scrub;	/* for /etc/system use */
91 
92 int	jp_errata_85_active = -1;	/* warn: modified in code ONLY */
93 uint64_t	jp_estar_tl0_data[8];
94 uint64_t	jp_estar_tl1_data[8];
95 #endif	/* JALAPENO && JALAPENO_ERRATA_85 */
96 
97 /*
98  * Setup trap handlers.
99  */
100 void
cpu_init_trap(void)101 cpu_init_trap(void)
102 {
103 	CH_SET_TRAP(pil15_epilogue, ch_pil15_interrupt_instr);
104 
105 	CH_SET_TRAP(tt0_fecc, fecc_err_instr);
106 	CH_SET_TRAP(tt1_fecc, fecc_err_tl1_instr);
107 	CH_SET_TRAP(tt1_swtrap0, fecc_err_tl1_cont_instr);
108 
109 	CH_SET_TRAP(tt0_dperr, dcache_parity_instr);
110 	CH_SET_TRAP(tt1_dperr, dcache_parity_tl1_instr);
111 	CH_SET_TRAP(tt1_swtrap1, dcache_parity_tl1_cont_instr);
112 
113 	CH_SET_TRAP(tt0_iperr, icache_parity_instr);
114 	CH_SET_TRAP(tt1_iperr, icache_parity_tl1_instr);
115 	CH_SET_TRAP(tt1_swtrap2, icache_parity_tl1_cont_instr);
116 }
117 
118 
119 static int
getintprop(pnode_t node,char * name,int deflt)120 getintprop(pnode_t node, char *name, int deflt)
121 {
122 	int	value;
123 
124 	switch (prom_getproplen(node, name)) {
125 	case sizeof (int):
126 		(void) prom_getprop(node, name, (caddr_t)&value);
127 		break;
128 
129 	default:
130 		value = deflt;
131 		break;
132 	}
133 
134 	return (value);
135 }
136 
137 /*
138  * Set the magic constants of the implementation.
139  */
140 /*ARGSUSED*/
141 void
cpu_fiximp(pnode_t dnode)142 cpu_fiximp(pnode_t dnode)
143 {
144 	int i, a;
145 	extern int vac_size, vac_shift;
146 	extern uint_t vac_mask;
147 
148 	static struct {
149 		char	*name;
150 		int	*var;
151 		int	defval;
152 	} prop[] = {
153 		"dcache-size", &dcache_size, CH_DCACHE_SIZE,
154 		"dcache-line-size", &dcache_linesize, CH_DCACHE_LSIZE,
155 		"icache-size", &icache_size, CH_ICACHE_SIZE,
156 		"icache-line-size", &icache_linesize, CH_ICACHE_LSIZE,
157 		"ecache-size", &ecache_size, JP_ECACHE_MAX_SIZE,
158 		"ecache-line-size", &ecache_alignsize, JP_ECACHE_MAX_LSIZE,
159 		"ecache-associativity", &ecache_associativity, JP_ECACHE_NWAY
160 	};
161 
162 	for (i = 0; i < sizeof (prop) / sizeof (prop[0]); i++)
163 		*prop[i].var = getintprop(dnode, prop[i].name, prop[i].defval);
164 
165 	ecache_setsize = ecache_size / ecache_associativity;
166 
167 	vac_size = CH_VAC_SIZE;
168 	vac_mask = MMU_PAGEMASK & (vac_size - 1);
169 	i = 0; a = vac_size;
170 	while (a >>= 1)
171 		++i;
172 	vac_shift = i;
173 	shm_alignment = vac_size;
174 	vac = 1;
175 }
176 
177 void
send_mondo_set(cpuset_t set)178 send_mondo_set(cpuset_t set)
179 {
180 	int lo, busy, nack, shipped = 0;
181 	uint16_t i, cpuids[IDSR_BN_SETS];
182 	uint64_t idsr, nackmask = 0, busymask, curnack, curbusy;
183 	uint64_t starttick, endtick, tick, lasttick;
184 #ifdef	CHEETAHPLUS_ERRATUM_25
185 	int recovered = 0;
186 	int cpuid;
187 #endif
188 
189 	ASSERT(!CPUSET_ISNULL(set));
190 	starttick = lasttick = gettick();
191 
192 	/*
193 	 * Lower 2 bits of the agent ID determine which BUSY/NACK pair
194 	 * will be used for dispatching interrupt. For now, assume
195 	 * there are no more than IDSR_BN_SETS CPUs, hence no aliasing
196 	 * issues with respect to BUSY/NACK pair usage.
197 	 */
198 	for (i = 0; i < NCPU; i++)
199 		if (CPU_IN_SET(set, i)) {
200 			shipit(i, shipped /* ignored */);
201 			nackmask |= IDSR_NACK_BIT(CPUID_TO_BN_PAIR(i));
202 			cpuids[CPUID_TO_BN_PAIR(i)] = i;
203 			shipped++;
204 			CPUSET_DEL(set, i);
205 			if (CPUSET_ISNULL(set))
206 				break;
207 		}
208 	CPU_STATS_ADDQ(CPU, sys, xcalls, shipped);
209 
210 	busymask = IDSR_NACK_TO_BUSY(nackmask);
211 	busy = nack = 0;
212 	endtick = starttick + xc_tick_limit;
213 	for (;;) {
214 		idsr = getidsr();
215 		if (idsr == 0)
216 			break;
217 		tick = gettick();
218 		/*
219 		 * If there is a big jump between the current tick
220 		 * count and lasttick, we have probably hit a break
221 		 * point.  Adjust endtick accordingly to avoid panic.
222 		 */
223 		if (tick > (lasttick + xc_tick_jump_limit))
224 			endtick += (tick - lasttick);
225 		lasttick = tick;
226 		if (tick > endtick) {
227 			if (panic_quiesce)
228 				return;
229 #ifdef	CHEETAHPLUS_ERRATUM_25
230 			cpuid = -1;
231 			for (i = 0; i < IDSR_BN_SETS; i++) {
232 				if (idsr & (IDSR_NACK_BIT(i) |
233 				    IDSR_BUSY_BIT(i))) {
234 					cpuid = cpuids[i];
235 					break;
236 				}
237 			}
238 			if (cheetah_sendmondo_recover && cpuid != -1 &&
239 			    recovered == 0) {
240 				if (mondo_recover(cpuid, i)) {
241 					/*
242 					 * We claimed the whole memory or
243 					 * full scan is disabled.
244 					 */
245 					recovered++;
246 				}
247 				tick = gettick();
248 				endtick = tick + xc_tick_limit;
249 				lasttick = tick;
250 				/*
251 				 * Recheck idsr
252 				 */
253 				continue;
254 			} else
255 #endif	/* CHEETAHPLUS_ERRATUM_25 */
256 			{
257 				cmn_err(CE_CONT, "send mondo timeout "
258 				    "[%d NACK %d BUSY]\nIDSR 0x%"
259 				    "" PRIx64 "  cpuids:", nack, busy, idsr);
260 				for (i = 0; i < IDSR_BN_SETS; i++) {
261 					if (idsr & (IDSR_NACK_BIT(i) |
262 					    IDSR_BUSY_BIT(i))) {
263 						cmn_err(CE_CONT, " 0x%x",
264 						    cpuids[i]);
265 					}
266 				}
267 				cmn_err(CE_CONT, "\n");
268 				cmn_err(CE_PANIC, "send_mondo_set: timeout");
269 			}
270 		}
271 		curnack = idsr & nackmask;
272 		curbusy = idsr & busymask;
273 		if (curbusy) {
274 			busy++;
275 			continue;
276 		}
277 
278 #ifdef SEND_MONDO_STATS
279 		{
280 			int n = gettick() - starttick;
281 			if (n < 8192)
282 				x_nack_stimes[n >> 7]++;
283 		}
284 #endif
285 		while (gettick() < (tick + sys_clock_mhz))
286 			;
287 		do {
288 			lo = lowbit(curnack) - 1;
289 			i = IDSR_NACK_IDX(lo);
290 			shipit(cpuids[i], i);
291 			curnack &= ~(1ull << lo);
292 		} while (curnack);
293 		nack++;
294 		busy = 0;
295 	}
296 #ifdef SEND_MONDO_STATS
297 	{
298 		int n = gettick() - starttick;
299 		if (n < 8192)
300 			x_set_stimes[n >> 7]++;
301 		else
302 			x_set_ltimes[(n >> 13) & 0xf]++;
303 	}
304 	x_set_cpus[shipped]++;
305 #endif
306 }
307 
308 /*
309  * Handles error logging for implementation specific error types
310  */
311 int
cpu_impl_async_log_err(void * flt,errorq_elem_t * eqep)312 cpu_impl_async_log_err(void *flt, errorq_elem_t *eqep)
313 {
314 	ch_async_flt_t *ch_flt = (ch_async_flt_t *)flt;
315 	struct async_flt *aflt = (struct async_flt *)flt;
316 	uint64_t errors;
317 
318 	switch (ch_flt->flt_type) {
319 
320 	case CPU_IC_PARITY:
321 		cpu_async_log_ic_parity_err(flt);
322 		return (CH_ASYNC_LOG_DONE);
323 
324 	case CPU_DC_PARITY:
325 		cpu_async_log_dc_parity_err(flt);
326 		return (CH_ASYNC_LOG_DONE);
327 
328 	case CPU_RCE:
329 		if (page_retire_check(aflt->flt_addr, &errors) == EINVAL) {
330 			CE_XDIAG_SETSKIPCODE(aflt->flt_disp,
331 			    CE_XDIAG_SKIP_NOPP);
332 		} else if (errors != PR_OK) {
333 			CE_XDIAG_SETSKIPCODE(aflt->flt_disp,
334 			    CE_XDIAG_SKIP_PAGEDET);
335 		} else if (ce_scrub_xdiag_recirc(aflt, ce_queue, eqep,
336 		    offsetof(ch_async_flt_t, cmn_asyncflt))) {
337 			return (CH_ASYNC_LOG_RECIRC);
338 		}
339 		/*FALLTHRU*/
340 	/*
341 	 * cases where we just want to report the error and continue.
342 	 */
343 	case CPU_BPAR:
344 	case CPU_UMS:
345 	case CPU_FRC:
346 	case CPU_FRU:
347 		cpu_log_err(aflt);
348 		return (CH_ASYNC_LOG_DONE);
349 
350 	/*
351 	 * Cases where we want to fall through to handle panicking.
352 	 */
353 	case CPU_RUE:
354 		cpu_log_err(aflt);
355 		return (CH_ASYNC_LOG_CONTINUE);
356 
357 	default:
358 		return (CH_ASYNC_LOG_UNKNOWN);
359 	}
360 }
361 
362 /*
363  * Figure out if Ecache is direct-mapped (Cheetah or Cheetah+ with Ecache
364  * control ECCR_ASSOC bit off or 2-way (Cheetah+ with ECCR_ASSOC on).
365  * We need to do this on the fly because we may have mixed Cheetah+'s with
366  * both direct and 2-way Ecaches.
367  */
368 int
cpu_ecache_nway(void)369 cpu_ecache_nway(void)
370 {
371 	return (JP_ECACHE_NWAY);
372 }
373 
374 /*
375  * Note that these are entered into the table in the order:
376  * Fatal Errors first, orphaned UCU/UCC, AFAR Overwrite policy,
377  * FRC/FRU, and finally IVPE.
378  *
379  * Afar overwrite policy is:
380  * Jalapeno:
381  *   UCU,UCC > RUE,UE,EDU,WDU,CPU,WBP,BP > RCE,CE,EDC,WDC,CPC >
382  *   TO,BERR > UMS,OM
383  * Serrano:
384  *   UCU,UCC > RUE,UE,EDU,WDU,CPU,WBP,BP > RCE,CE,EDC,WDC,CPC,ETI,ETC >
385  *   TO,BERR > UMS,OM
386  */
387 ecc_type_to_info_t ecc_type_to_info[] = {
388 
389 	/* Fatal Errors */
390 	C_AFSR_JETO,	"JETO ",	ECC_ALL_TRAPS,	CPU_FATAL,
391 		"JETO Fatal",
392 		FM_EREPORT_PAYLOAD_SYSTEM1,
393 		FM_EREPORT_CPU_USIII_JETO,
394 	C_AFSR_SCE,	"SCE ",		ECC_ALL_TRAPS,	CPU_FATAL,
395 		"SCE Fatal",
396 		FM_EREPORT_PAYLOAD_SYSTEM1,
397 		FM_EREPORT_CPU_USIII_SCE,
398 	C_AFSR_JEIC,	"JEIC ",	ECC_ALL_TRAPS,	CPU_FATAL,
399 		"JEIC Fatal",
400 		FM_EREPORT_PAYLOAD_SYSTEM1,
401 		FM_EREPORT_CPU_USIII_JEIC,
402 	C_AFSR_JEIT,	"JEIT ",	ECC_ALL_TRAPS,	CPU_FATAL,
403 		"JEIT Fatal",
404 		FM_EREPORT_PAYLOAD_SYSTEM1,
405 		FM_EREPORT_CPU_USIII_JEIT,
406 	C_AFSR_JEIS,	"JEIS ",	ECC_ALL_TRAPS,	CPU_FATAL,
407 		"JEIS Fatal",
408 		FM_EREPORT_PAYLOAD_SYSTEM1,
409 		FM_EREPORT_CPU_USIII_JEIS,
410 #if defined(JALAPENO)
411 	C_AFSR_ETP,	"ETP ",		ECC_ALL_TRAPS,	CPU_FATAL,
412 		"ETP Fatal",
413 		FM_EREPORT_PAYLOAD_L2_TAG_PE,
414 		FM_EREPORT_CPU_USIII_ETP,
415 #elif defined(SERRANO)
416 	C_AFSR_ETS,	"ETS ",		ECC_ASYNC_TRAPS, CPU_FATAL,
417 		"ETS Fatal",
418 		FM_EREPORT_PAYLOAD_L2_TAG_ECC,
419 		FM_EREPORT_CPU_USIII_ETS,
420 	C_AFSR_ETU,	"ETU ",		ECC_ASYNC_TRAPS, CPU_FATAL,
421 		"ETU Fatal",
422 		FM_EREPORT_PAYLOAD_L2_TAG_ECC,
423 		FM_EREPORT_CPU_USIII_ETU,
424 #endif	/* SERRANO */
425 	C_AFSR_IERR,	"IERR ", 	ECC_ALL_TRAPS,	CPU_FATAL,
426 		"IERR Fatal",
427 		FM_EREPORT_PAYLOAD_SYSTEM2,
428 		FM_EREPORT_CPU_USIII_IERR,
429 	C_AFSR_ISAP,	"ISAP ",	ECC_ALL_TRAPS,	CPU_FATAL,
430 		"ISAP Fatal",
431 		FM_EREPORT_PAYLOAD_SYSTEM1,
432 		FM_EREPORT_CPU_USIII_ISAP,
433 
434 	/* Orphaned UCU/UCC Errors */
435 	C_AFSR_UCU,	"OUCU ",	ECC_ORPH_TRAPS, CPU_ORPH,
436 		"Orphaned UCU",
437 		FM_EREPORT_PAYLOAD_L2_DATA,
438 		FM_EREPORT_CPU_USIII_UCU,
439 	C_AFSR_UCC,	"OUCC ",	ECC_ORPH_TRAPS, CPU_ORPH,
440 		"Orphaned UCC",
441 		FM_EREPORT_PAYLOAD_L2_DATA,
442 		FM_EREPORT_CPU_USIII_UCC,
443 
444 	/* UCU, UCC */
445 	C_AFSR_UCU,	"UCU ",		ECC_F_TRAP,	CPU_UE_ECACHE,
446 		"UCU",
447 		FM_EREPORT_PAYLOAD_L2_DATA,
448 		FM_EREPORT_CPU_USIII_UCU,
449 	C_AFSR_UCC,	"UCC ",		ECC_F_TRAP,	CPU_CE_ECACHE,
450 		"UCC",
451 		FM_EREPORT_PAYLOAD_L2_DATA,
452 		FM_EREPORT_CPU_USIII_UCC,
453 
454 
455 	/* RUE, UE, EDU:ST, EDU:BLD, WDU, CPU, BP, WBP */
456 	C_AFSR_RUE,	"RUE ",		ECC_ASYNC_TRAPS, CPU_RUE,
457 		"Uncorrectable remote memory/cache (RUE)",
458 		FM_EREPORT_PAYLOAD_MEMORY,
459 		FM_EREPORT_CPU_USIII_RUE,
460 	C_AFSR_UE,	"UE ",		ECC_ASYNC_TRAPS, CPU_UE,
461 		"Uncorrectable memory (UE)",
462 		FM_EREPORT_PAYLOAD_MEMORY,
463 		FM_EREPORT_CPU_USIII_UE,
464 	C_AFSR_EDU,	"EDU ",		ECC_C_TRAP,	CPU_UE_ECACHE_RETIRE,
465 		"EDU:ST",
466 		FM_EREPORT_PAYLOAD_L2_DATA,
467 		FM_EREPORT_CPU_USIII_EDUST,
468 	C_AFSR_EDU,	"EDU ",		ECC_D_TRAP,	CPU_UE_ECACHE_RETIRE,
469 		"EDU:BLD",
470 		FM_EREPORT_PAYLOAD_L2_DATA,
471 		FM_EREPORT_CPU_USIII_EDUBL,
472 	C_AFSR_WDU,	"WDU ",		ECC_C_TRAP,	CPU_UE_ECACHE_RETIRE,
473 		"WDU",
474 		FM_EREPORT_PAYLOAD_L2_DATA,
475 		FM_EREPORT_CPU_USIII_WDU,
476 	C_AFSR_CPU,	"CPU ",		ECC_C_TRAP,	CPU_UE_ECACHE,
477 		"CPU",
478 		FM_EREPORT_PAYLOAD_L2_DATA,
479 		FM_EREPORT_CPU_USIII_CPU,
480 	C_AFSR_WBP,	"WBP ",		ECC_C_TRAP,	CPU_BPAR,
481 		"JBUS parity error on writeback or block store (WBP)",
482 		FM_EREPORT_PAYLOAD_SYSTEM3,
483 		FM_EREPORT_CPU_USIII_WBP,
484 	C_AFSR_BP,	"BP ",		ECC_ASYNC_TRAPS, CPU_BPAR,
485 		"JBUS parity error on returned read data (BP)",
486 		FM_EREPORT_PAYLOAD_SYSTEM3,
487 		FM_EREPORT_CPU_USIII_BP,
488 
489 	/* RCE, CE, EDC, WDC, CPC */
490 	C_AFSR_RCE,	"RCE ",		ECC_C_TRAP,	CPU_RCE,
491 		"Corrected remote memory/cache (RCE)",
492 		FM_EREPORT_PAYLOAD_MEMORY,
493 		FM_EREPORT_CPU_USIII_RCE,
494 	C_AFSR_CE,	"CE ",		ECC_C_TRAP,	CPU_CE,
495 		"Corrected memory (CE)",
496 		FM_EREPORT_PAYLOAD_MEMORY,
497 		FM_EREPORT_CPU_USIII_CE,
498 	C_AFSR_EDC,	"EDC ",		ECC_C_TRAP,	CPU_CE_ECACHE,
499 		"EDC",
500 		FM_EREPORT_PAYLOAD_L2_DATA,
501 		FM_EREPORT_CPU_USIII_EDC,
502 	C_AFSR_WDC,	"WDC ",		ECC_C_TRAP,	CPU_CE_ECACHE,
503 		"WDC",
504 		FM_EREPORT_PAYLOAD_L2_DATA,
505 		FM_EREPORT_CPU_USIII_WDC,
506 	C_AFSR_CPC,	"CPC ",		ECC_C_TRAP,	CPU_CE_ECACHE,
507 		"CPC",
508 		FM_EREPORT_PAYLOAD_L2_DATA,
509 		FM_EREPORT_CPU_USIII_CPC,
510 #if defined(SERRANO)
511 	/* ETI, ETC */
512 	C_AFSR_ETI,	"ETI",	ECC_F_TRAP | ECC_C_TRAP, CPU_CE_ECACHE,
513 		"ETI",
514 		FM_EREPORT_PAYLOAD_L2_TAG_ECC,
515 		FM_EREPORT_CPU_USIII_ETI,
516 	C_AFSR_ETC,	"ETC",	ECC_F_TRAP | ECC_C_TRAP, CPU_CE_ECACHE,
517 		"ETC",
518 		FM_EREPORT_PAYLOAD_L2_TAG_ECC,
519 		FM_EREPORT_CPU_USIII_ETC,
520 #endif	/* SERRANO */
521 
522 	/* TO, BERR */
523 	C_AFSR_TO,	"TO ",		ECC_ASYNC_TRAPS, CPU_TO,
524 		"Timeout (TO)",
525 		FM_EREPORT_PAYLOAD_IO,
526 		FM_EREPORT_CPU_USIII_TO,
527 	C_AFSR_BERR,	"BERR ",	ECC_ASYNC_TRAPS, CPU_BERR,
528 		"Bus Error (BERR)",
529 		FM_EREPORT_PAYLOAD_IO,
530 		FM_EREPORT_CPU_USIII_BERR,
531 
532 	/* UMS, OM */
533 	C_AFSR_UMS,	"UMS ",		ECC_C_TRAP,	 CPU_UMS,
534 		"Unsupported store (UMS)",
535 		FM_EREPORT_PAYLOAD_IO,
536 		FM_EREPORT_CPU_USIII_UMS,
537 	C_AFSR_OM,	"OM ",		ECC_ASYNC_TRAPS, CPU_BERR,
538 		"Out of range memory (OM)",
539 		FM_EREPORT_PAYLOAD_IO,
540 		FM_EREPORT_CPU_USIII_OM,
541 
542 	/* FRC, FRU */
543 	C_AFSR_FRC,	"FRC ",		ECC_C_TRAP,	CPU_FRC,
544 		"Corrected memory (FRC)",
545 		FM_EREPORT_PAYLOAD_MEMORY,
546 		FM_EREPORT_CPU_USIII_FRC,
547 	C_AFSR_FRU,	"FRU ",		ECC_C_TRAP,	 CPU_FRU,
548 		"Uncorrectable memory (FRU)",
549 		FM_EREPORT_PAYLOAD_MEMORY,
550 		FM_EREPORT_CPU_USIII_FRU,
551 
552 	/* IVPE */
553 	C_AFSR_IVPE,	"IVPE ",	ECC_C_TRAP,	CPU_IV,
554 		"IVPE",
555 		FM_EREPORT_PAYLOAD_SYSTEM1,
556 		FM_EREPORT_CPU_USIII_IVPE,
557 
558 	0,		NULL,		0,		0,
559 		NULL,
560 		FM_EREPORT_PAYLOAD_UNKNOWN,
561 		FM_EREPORT_CPU_USIII_UNKNOWN,
562 };
563 
564 /*
565  * J_REQ overwrite policy (see UltraSPARC-IIIi PRM)
566  *
567  *   Class 4:  RUE, BP, WBP
568  *   Class 3:  RCE
569  *   Class 2:  TO, BERR
570  *   Class 1:  UMS
571  */
572 uint64_t jreq_overwrite[] = {
573 	C_AFSR_RUE | C_AFSR_BP | C_AFSR_WBP,
574 	C_AFSR_RCE,
575 	C_AFSR_TO | C_AFSR_BERR,
576 	C_AFSR_UMS,
577 	0
578 };
579 
580 /*
581  * AGENT ID overwrite policy (see UltraSPARC-IIIi PRM)
582  *
583  *   Class 2:  CPU, FRU
584  *   Class 1:  CPC, FRC
585  */
586 uint64_t jbus_aid_overwrite[] = {
587 	C_AFSR_CPU | C_AFSR_FRU,
588 	C_AFSR_CPC | C_AFSR_FRC,
589 	0
590 };
591 
592 int
afsr_to_jaid_status(uint64_t afsr,uint64_t afsr_bit)593 afsr_to_jaid_status(uint64_t afsr, uint64_t afsr_bit)
594 {
595 	return (afsr_to_overw_status(afsr, afsr_bit, jbus_aid_overwrite));
596 }
597 
598 /*
599  * See UltraSPARC-IIIi+ PRM
600  *   Class 5:  ETS, ETU, EFES
601  *   Class 4:  UCC, UCU
602  *   Class 3:  UE, RUE, BP, WBP, EDU, WDU, CPU
603  *   Class 2:  CE, RCE, EDC, WDC, CPC, ETI, ETC
604  *   Class 1:  TO, BERR
605  *   Class 0:  UMS, OM
606  *
607  * See UltraSPARC-IIIi PRM
608  *   Class 5:  ETP
609  *   Class 4:  UCC, UCU
610  *   Class 3:  UE, RUE, BP, WBP, EDU, WDU
611  *   Class 2:  CE, RCE, EDC, WDC
612  *   Class 1:  TO, BERR
613  *   Class 0:  UMS, OM
614  */
615 uint64_t afar_overwrite[] = {
616 #if defined(JALAPENO)
617 	C_AFSR_ETP,
618 #elif defined(SERRANO)
619 	C_AFSR_ETS | C_AFSR_ETU | C_AFSR_EFES,
620 #endif /* SERRANO */
621 	C_AFSR_UCC | C_AFSR_UCU,
622 	C_AFSR_UE | C_AFSR_RUE | C_AFSR_BP | C_AFSR_WBP | C_AFSR_EDU |
623 	    C_AFSR_WDU | C_AFSR_CPU,
624 #if defined(SERRANO)
625 	C_AFSR_ETI | C_AFSR_ETC |
626 #endif /* SERRANO */
627 	C_AFSR_CE | C_AFSR_RCE | C_AFSR_EDC | C_AFSR_WDC | C_AFSR_CPC,
628 	C_AFSR_TO | C_AFSR_BERR,
629 	C_AFSR_UMS | C_AFSR_OM,
630 	0
631 };
632 
633 #if defined(SERRANO)
634 /*
635  * Serrano has a second AFAR that captures the physical address on
636  * FRC/FRU errors (which Jalapeno does not).  This register also
637  * captures the address for UE and CE errors.
638  *
639  * See UltraSPARC-IIIi+ PRM
640  *  Class 3: UE
641  *  Class 2: FRU
642  *  Class 1: CE
643  *  Class 0: FRC
644  */
645 uint64_t afar2_overwrite[] = {
646 	C_AFSR_UE,
647 	C_AFSR_FRU,
648 	C_AFSR_CE,
649 	C_AFSR_FRC,
650 	0
651 };
652 #endif  /* SERRANO */
653 
654 /*
655  * See UltraSPARC-IIIi PRM
656  *   Class 2:  UE, FRU, EDU, WDU, UCU, CPU
657  *   Class 1:  CE, FRC, EDC, WDC, UCC, CPC
658  */
659 uint64_t esynd_overwrite[] = {
660 #if defined(SERRANO)
661 	C_AFSR_ETS | C_AFSR_ETU |
662 #endif	/* SERRANO */
663 	C_AFSR_UE | C_AFSR_FRU | C_AFSR_EDU | C_AFSR_WDU | C_AFSR_UCU |
664 	    C_AFSR_CPU,
665 	C_AFSR_CE | C_AFSR_FRC | C_AFSR_EDC | C_AFSR_WDC | C_AFSR_UCC |
666 	    C_AFSR_CPC,
667 	0
668 };
669 
670 /*
671  * Prioritized list of Error bits for BSYND (referred to as
672  * MSYND to share code with CHEETAH & CHEETAH_PLUS) overwrite.
673  * See UltraSPARC-IIIi PRM
674  *   Class 3:  ISAP
675  *   Class 2:  BP
676  *   Class 1:  WBP, IVPE
677  */
678 uint64_t msynd_overwrite[] = {
679 	C_AFSR_ISAP,
680 	C_AFSR_BP,
681 	C_AFSR_WBP | C_AFSR_IVPE,
682 	0
683 };
684 
685 /*
686  * change cpu speed bits -- new speed will be normal-speed/divisor.
687  *
688  * The Jalapeno memory controllers are required to drain outstanding
689  * memory transactions within 32 JBus clocks in order to be ready
690  * to enter Estar mode.  In some corner cases however, that time
691  * fell short.
692  *
693  * A safe software solution is to force MCU to act like in Estar mode,
694  * then delay 1us (in ppm code) prior to assert J_CHNG_L signal.
695  * To reverse the effect, upon exiting Estar, software restores the
696  * MCU to its original state.
697  */
698 /* ARGSUSED1 */
699 void
cpu_change_speed(uint64_t divisor,uint64_t arg2)700 cpu_change_speed(uint64_t divisor, uint64_t arg2)
701 {
702 	bus_config_eclk_t	*bceclk;
703 	uint64_t		reg;
704 	uint64_t		oldreg;
705 	uint64_t		mreg;
706 	uint64_t		val64;
707 	int			id = (CPU)->cpu_id;
708 	processor_info_t	*pi = &(CPU->cpu_type_info);
709 
710 #if defined(JALAPENO) && defined(JALAPENO_ERRATA_85)
711 	/*
712 	 * ASI Ecache flush in 1/2 or 1/32 speed mode can result
713 	 * in CPU fatal reset (JETO or IERR/TO on MP). A workaround
714 	 * is to force the CPU to full speed mode prior to using
715 	 * ASI Ecache flush opeartion to flush E$. Since we can't
716 	 * always use cross calls at the time of flushing E$, we
717 	 * cannot change other CPU speed. Hence, this workaround
718 	 * is applicable to uniprocessor configuration only and
719 	 * can't be used in multiprocessor configuration.
720 	 *
721 	 * Note that this workaround is activated only when the CPU
722 	 * has been fully initialized and its speed is lowered by the
723 	 * ppm for the first time. It can be disabled via /etc/system
724 	 * by setting jp_errata_85_enable to 0 and rebooting the
725 	 * system.
726 	 */
727 	if ((jp_errata_85_active == -1) &&
728 	    jp_errata_85_enable &&
729 	    (divisor != JBUS_CONFIG_ECLK_1_DIV)) {
730 		if (ncpus == 1)
731 			jp_errata_85_active = 1;
732 		else
733 			jp_errata_85_active = 0;
734 	}
735 	if ((!jp_errata_85_allow_slow_scrub) && (CPU_PRIVATE(CPU) != NULL)) {
736 		int i;
737 		ch_scrub_misc_t	*chpr_scrubp =
738 		    CPU_PRIVATE_PTR(CPU, chpr_scrub_misc);
739 
740 		/* We're only allowed to run the scrubbers at full speed */
741 
742 		for (i = 0; i < CACHE_SCRUBBER_COUNT; i++) {
743 			chpr_scrubp->chsm_enable[i] =
744 			    (divisor == JBUS_CONFIG_ECLK_1_DIV);
745 		}
746 	}
747 #endif	/* JALAPENO && JALAPENO_ERRATA_85 */
748 
749 	/*
750 	 * We're only interested in mcu_ctl_reg1 bit 26 and 25, of which
751 	 * the value will be stored in the lower half of a byte.  The
752 	 * top bit of this byte is designated as a valid bit - 0 means
753 	 * invalid, 1 means valid.
754 	 */
755 	if (!mcu_fsm_init_state[id].valid) {
756 		val64 = get_mcu_ctl_reg1() & JP_MCU_FSM_MASK;
757 		mcu_fsm_init_state[id].state = val64 >> JP_MCU_FSM_SHIFT;
758 		mcu_fsm_init_state[id].valid = 1;
759 	}
760 
761 	for (bceclk = bus_config_eclk; bceclk->divisor; bceclk++) {
762 		if (bceclk->divisor != divisor)
763 			continue;
764 		reg = get_jbus_config();
765 		oldreg = reg;
766 		reg &= ~JBUS_CONFIG_ECLK_MASK;
767 		reg |= bceclk->mask;
768 		set_jbus_config(reg);
769 		(void) get_jbus_config();
770 
771 		/*
772 		 * MCU workaround, refer to Jalapeno spec, EnergyStar section
773 		 * for detail.
774 		 */
775 
776 		/* Upon entering engery star mode, turn off extra MCU FSMs */
777 		if (((oldreg & JBUS_CONFIG_ECLK_MASK) == JBUS_CONFIG_ECLK_1) &&
778 		    ((divisor == JBUS_CONFIG_ECLK_2_DIV) ||
779 		    (divisor == JBUS_CONFIG_ECLK_32_DIV))) {
780 			mreg = get_mcu_ctl_reg1();
781 			if ((mreg & JP_MCU_FSM_MASK) != 0) {
782 				mreg &= ~JP_MCU_FSM_MASK;
783 				set_mcu_ctl_reg1(mreg);
784 				(void) get_mcu_ctl_reg1();
785 			}
786 		/* Upon exiting energy star mode, restore extra MCU FSMs */
787 		} else if (divisor == JBUS_CONFIG_ECLK_1_DIV) {
788 			mreg = get_mcu_ctl_reg1();
789 			val64 = mcu_fsm_init_state[id].state;
790 			mreg |= val64 << JP_MCU_FSM_SHIFT;
791 			set_mcu_ctl_reg1(mreg);
792 			(void) get_mcu_ctl_reg1();
793 		}
794 		CPU->cpu_m.divisor = (uchar_t)divisor;
795 		cpu_set_curr_clock(((uint64_t)pi->pi_clock * 1000000) /
796 		    divisor);
797 		return;
798 	}
799 	/*
800 	 * We will reach here only if OBP and kernel don't agree on
801 	 * the speeds supported by the CPU.
802 	 */
803 	cmn_err(CE_WARN, "cpu_change_speed: bad divisor %" PRIu64, divisor);
804 }
805 
806 /*
807  * Cpu private initialization.  This includes allocating the cpu_private
808  * data structure, initializing it, and initializing the scrubber for this
809  * cpu.  This function calls cpu_init_ecache_scrub_dr to init the scrubber.
810  * We use kmem_cache_create for the cheetah private data structure because
811  * it needs to be allocated on a PAGESIZE (8192) byte boundary.
812  */
813 void
cpu_init_private(struct cpu * cp)814 cpu_init_private(struct cpu *cp)
815 {
816 	cheetah_private_t *chprp;
817 	int i;
818 
819 	ASSERT(CPU_PRIVATE(cp) == NULL);
820 
821 	/* LINTED: E_TRUE_LOGICAL_EXPR */
822 	ASSERT((offsetof(cheetah_private_t, chpr_tl1_err_data) +
823 	    sizeof (ch_err_tl1_data_t) * CH_ERR_TL1_TLMAX) <= PAGESIZE);
824 
825 #if defined(SERRANO)
826 	if (!IS_SERRANO(cpunodes[cp->cpu_id].implementation)) {
827 		cmn_err(CE_PANIC, "CPU%d: implementation 0x%x not supported"
828 		    " on UltraSPARC-IIIi+ code\n", cp->cpu_id,
829 		    cpunodes[cp->cpu_id].implementation);
830 	}
831 #else /* SERRANO */
832 	if (!IS_JALAPENO(cpunodes[cp->cpu_id].implementation)) {
833 		cmn_err(CE_PANIC, "CPU%d: implementation 0x%x not supported"
834 		    " on UltraSPARC-IIIi code\n", cp->cpu_id,
835 		    cpunodes[cp->cpu_id].implementation);
836 	}
837 #endif /* SERRANO */
838 
839 	/*
840 	 * If the ch_private_cache has not been created, create it.
841 	 */
842 	if (ch_private_cache == NULL) {
843 		ch_private_cache = kmem_cache_create("ch_private_cache",
844 		    sizeof (cheetah_private_t), PAGESIZE, NULL, NULL,
845 		    NULL, NULL, static_arena, 0);
846 	}
847 
848 	chprp = CPU_PRIVATE(cp) = kmem_cache_alloc(ch_private_cache, KM_SLEEP);
849 
850 	bzero(chprp, sizeof (cheetah_private_t));
851 	chprp->chpr_fecctl0_logout.clo_data.chd_afar = LOGOUT_INVALID;
852 	chprp->chpr_cecc_logout.clo_data.chd_afar = LOGOUT_INVALID;
853 	chprp->chpr_async_logout.clo_data.chd_afar = LOGOUT_INVALID;
854 	for (i = 0; i < CH_ERR_TL1_TLMAX; i++)
855 		chprp->chpr_tl1_err_data[i].ch_err_tl1_logout.clo_data.chd_afar
856 		    = LOGOUT_INVALID;
857 
858 	chprp->chpr_icache_size = CH_ICACHE_SIZE;
859 	chprp->chpr_icache_linesize = CH_ICACHE_LSIZE;
860 
861 	cpu_init_ecache_scrub_dr(cp);
862 
863 	chprp->chpr_ec_set_size = cpunodes[cp->cpu_id].ecache_size /
864 	    cpu_ecache_nway();
865 
866 	adjust_hw_copy_limits(cpunodes[cp->cpu_id].ecache_size);
867 	ch_err_tl1_paddrs[cp->cpu_id] = va_to_pa(chprp);
868 	ASSERT(ch_err_tl1_paddrs[cp->cpu_id] != -1);
869 }
870 
871 /*
872  * Clear the error state registers for this CPU.
873  * For Jalapeno, just clear the AFSR
874  */
875 void
set_cpu_error_state(ch_cpu_errors_t * cpu_error_regs)876 set_cpu_error_state(ch_cpu_errors_t *cpu_error_regs)
877 {
878 	set_asyncflt(cpu_error_regs->afsr & ~C_AFSR_FATAL_ERRS);
879 }
880 
881 /*
882  * Update cpu_offline_set so the scrubber knows which cpus are offline
883  */
884 /*ARGSUSED*/
885 int
cpu_scrub_cpu_setup(cpu_setup_t what,int cpuid,void * arg)886 cpu_scrub_cpu_setup(cpu_setup_t what, int cpuid, void *arg)
887 {
888 	switch (what) {
889 	case CPU_ON:
890 	case CPU_INIT:
891 		CPUSET_DEL(cpu_offline_set, cpuid);
892 		break;
893 	case CPU_OFF:
894 		CPUSET_ADD(cpu_offline_set, cpuid);
895 		break;
896 	default:
897 		break;
898 	}
899 	return (0);
900 }
901