xref: /titanic_52/usr/src/uts/sun4u/cpu/us3_cheetah.c (revision 6185db853e024a486ff8837e6784dd290d866112)
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/hat_sfmmu.h>
41 #include <vm/seg_kmem.h>
42 #include <sys/cpuvar.h>
43 #include <sys/cheetahregs.h>
44 #include <sys/us3_module.h>
45 #include <sys/async.h>
46 #include <sys/cmn_err.h>
47 #include <sys/debug.h>
48 #include <sys/dditypes.h>
49 #include <sys/prom_debug.h>
50 #include <sys/prom_plat.h>
51 #include <sys/cpu_module.h>
52 #include <sys/sysmacros.h>
53 #include <sys/intreg.h>
54 #include <sys/clock.h>
55 #include <sys/platform_module.h>
56 #include <sys/machtrap.h>
57 #include <sys/ontrap.h>
58 #include <sys/panic.h>
59 #include <sys/memlist.h>
60 #include <sys/bootconf.h>
61 #include <sys/ivintr.h>
62 #include <sys/atomic.h>
63 #include <sys/fm/protocol.h>
64 #include <sys/fm/cpu/UltraSPARC-III.h>
65 #include <vm/vm_dep.h>
66 
67 #ifdef	CHEETAHPLUS_ERRATUM_25
68 #include <sys/cyclic.h>
69 #endif	/* CHEETAHPLUS_ERRATUM_25 */
70 
71 /*
72  * Note that 'Cheetah PRM' refers to:
73  *   SPARC V9 JPS1 Implementation Supplement: Sun UltraSPARC-III
74  */
75 
76 /*
77  * Setup trap handlers.
78  */
79 void
80 cpu_init_trap(void)
81 {
82 	CH_SET_TRAP(tt_pil15, ch_pil15_interrupt_instr);
83 
84 	CH_SET_TRAP(tt0_fecc, fecc_err_instr);
85 	CH_SET_TRAP(tt1_fecc, fecc_err_tl1_instr);
86 	CH_SET_TRAP(tt1_swtrap0, fecc_err_tl1_cont_instr);
87 }
88 
89 static int
90 getintprop(pnode_t node, char *name, int deflt)
91 {
92 	int	value;
93 
94 	switch (prom_getproplen(node, name)) {
95 	case sizeof (int):
96 		(void) prom_getprop(node, name, (caddr_t)&value);
97 		break;
98 
99 	default:
100 		value = deflt;
101 		break;
102 	}
103 
104 	return (value);
105 }
106 
107 /*
108  * Set the magic constants of the implementation.
109  */
110 /*ARGSUSED*/
111 void
112 cpu_fiximp(pnode_t dnode)
113 {
114 	int i, a;
115 
116 	static struct {
117 		char	*name;
118 		int	*var;
119 		int	defval;
120 	} prop[] = {
121 		"dcache-size", &dcache_size, CH_DCACHE_SIZE,
122 		"dcache-line-size", &dcache_linesize, CH_DCACHE_LSIZE,
123 		"icache-size", &icache_size, CH_ICACHE_SIZE,
124 		"icache-line-size", &icache_linesize, CH_ICACHE_LSIZE,
125 		"ecache-size", &ecache_size, CH_ECACHE_MAX_SIZE,
126 		"ecache-line-size", &ecache_alignsize, CH_ECACHE_MAX_LSIZE,
127 		"ecache-associativity", &ecache_associativity, CH_ECACHE_NWAY
128 	};
129 
130 	for (i = 0; i < sizeof (prop) / sizeof (prop[0]); i++)
131 		*prop[i].var = getintprop(dnode, prop[i].name, prop[i].defval);
132 
133 	ecache_setsize = ecache_size / ecache_associativity;
134 
135 	vac_size = CH_VAC_SIZE;
136 	vac_mask = MMU_PAGEMASK & (vac_size - 1);
137 	i = 0; a = vac_size;
138 	while (a >>= 1)
139 		++i;
140 	vac_shift = i;
141 	shm_alignment = vac_size;
142 	vac = 1;
143 
144 	/*
145 	 * Cheetah's large page support has problems with large numbers of
146 	 * large pages, so just disable large pages out-of-the-box.
147 	 * Note that the other defaults are set in sun4u/vm/mach_vm_dep.c.
148 	 */
149 	max_uheap_lpsize = MMU_PAGESIZE;
150 	max_ustack_lpsize = MMU_PAGESIZE;
151 	max_privmap_lpsize = MMU_PAGESIZE;
152 	max_utext_lpsize = MMU_PAGESIZE;
153 	max_shm_lpsize = MMU_PAGESIZE;
154 }
155 
156 void
157 send_mondo_set(cpuset_t set)
158 {
159 	int lo, busy, nack, shipped = 0;
160 	uint16_t i, cpuids[IDSR_BN_SETS];
161 	uint64_t idsr, nackmask = 0, busymask, curnack, curbusy;
162 	uint64_t starttick, endtick, tick, lasttick;
163 #if (NCPU > IDSR_BN_SETS)
164 	int index = 0;
165 	int ncpuids = 0;
166 #endif
167 #ifdef	CHEETAHPLUS_ERRATUM_25
168 	int recovered = 0;
169 	int cpuid;
170 #endif
171 
172 	ASSERT(!CPUSET_ISNULL(set));
173 	starttick = lasttick = gettick();
174 
175 #if (NCPU <= IDSR_BN_SETS)
176 	for (i = 0; i < NCPU; i++)
177 		if (CPU_IN_SET(set, i)) {
178 			shipit(i, shipped);
179 			nackmask |= IDSR_NACK_BIT(shipped);
180 			cpuids[shipped++] = i;
181 			CPUSET_DEL(set, i);
182 			if (CPUSET_ISNULL(set))
183 				break;
184 		}
185 	CPU_STATS_ADDQ(CPU, sys, xcalls, shipped);
186 #else
187 	for (i = 0; i < NCPU; i++)
188 		if (CPU_IN_SET(set, i)) {
189 			ncpuids++;
190 
191 			/*
192 			 * Ship only to the first (IDSR_BN_SETS) CPUs.  If we
193 			 * find we have shipped to more than (IDSR_BN_SETS)
194 			 * CPUs, set "index" to the highest numbered CPU in
195 			 * the set so we can ship to other CPUs a bit later on.
196 			 */
197 			if (shipped < IDSR_BN_SETS) {
198 				shipit(i, shipped);
199 				nackmask |= IDSR_NACK_BIT(shipped);
200 				cpuids[shipped++] = i;
201 				CPUSET_DEL(set, i);
202 				if (CPUSET_ISNULL(set))
203 					break;
204 			} else
205 				index = (int)i;
206 		}
207 
208 	CPU_STATS_ADDQ(CPU, sys, xcalls, ncpuids);
209 #endif
210 
211 	busymask = IDSR_NACK_TO_BUSY(nackmask);
212 	busy = nack = 0;
213 	endtick = starttick + xc_tick_limit;
214 	for (;;) {
215 		idsr = getidsr();
216 #if (NCPU <= IDSR_BN_SETS)
217 		if (idsr == 0)
218 			break;
219 #else
220 		if (idsr == 0 && shipped == ncpuids)
221 			break;
222 #endif
223 		tick = gettick();
224 		/*
225 		 * If there is a big jump between the current tick
226 		 * count and lasttick, we have probably hit a break
227 		 * point.  Adjust endtick accordingly to avoid panic.
228 		 */
229 		if (tick > (lasttick + xc_tick_jump_limit))
230 			endtick += (tick - lasttick);
231 		lasttick = tick;
232 		if (tick > endtick) {
233 			if (panic_quiesce)
234 				return;
235 #ifdef	CHEETAHPLUS_ERRATUM_25
236 			cpuid = -1;
237 			for (i = 0; i < IDSR_BN_SETS; i++) {
238 				if (idsr & (IDSR_NACK_BIT(i) |
239 				    IDSR_BUSY_BIT(i))) {
240 					cpuid = cpuids[i];
241 					break;
242 				}
243 			}
244 			if (cheetah_sendmondo_recover && cpuid != -1 &&
245 			    recovered == 0) {
246 				if (mondo_recover(cpuid, i)) {
247 					/*
248 					 * We claimed the whole memory or
249 					 * full scan is disabled.
250 					 */
251 					recovered++;
252 				}
253 				tick = gettick();
254 				endtick = tick + xc_tick_limit;
255 				lasttick = tick;
256 				/*
257 				 * Recheck idsr
258 				 */
259 				continue;
260 			} else
261 #endif	/* CHEETAHPLUS_ERRATUM_25 */
262 			{
263 				cmn_err(CE_CONT, "send mondo timeout "
264 				    "[%d NACK %d BUSY]\nIDSR 0x%"
265 				    "" PRIx64 "  cpuids:", nack, busy, idsr);
266 				for (i = 0; i < IDSR_BN_SETS; i++) {
267 					if (idsr & (IDSR_NACK_BIT(i) |
268 					    IDSR_BUSY_BIT(i))) {
269 						cmn_err(CE_CONT, " 0x%x",
270 						    cpuids[i]);
271 					}
272 				}
273 				cmn_err(CE_CONT, "\n");
274 				cmn_err(CE_PANIC, "send_mondo_set: timeout");
275 			}
276 		}
277 		curnack = idsr & nackmask;
278 		curbusy = idsr & busymask;
279 #if (NCPU > IDSR_BN_SETS)
280 		if (shipped < ncpuids) {
281 			uint64_t cpus_left;
282 			uint16_t next = (uint16_t)index;
283 
284 			cpus_left = ~(IDSR_NACK_TO_BUSY(curnack) | curbusy) &
285 				busymask;
286 
287 			if (cpus_left) {
288 				do {
289 					/*
290 					 * Sequence through and ship to the
291 					 * remainder of the CPUs in the system
292 					 * (e.g. other than the first
293 					 * (IDSR_BN_SETS)) in reverse order.
294 					 */
295 					lo = lowbit(cpus_left) - 1;
296 					i = IDSR_BUSY_IDX(lo);
297 					shipit(next, i);
298 					shipped++;
299 					cpuids[i] = next;
300 
301 					/*
302 					 * If we've processed all the CPUs,
303 					 * exit the loop now and save
304 					 * instructions.
305 					 */
306 					if (shipped == ncpuids)
307 						break;
308 
309 					for ((index = ((int)next - 1));
310 					    index >= 0; index--)
311 						if (CPU_IN_SET(set, index)) {
312 							next = (uint16_t)index;
313 							break;
314 						}
315 
316 					cpus_left &= ~(1ull << lo);
317 				} while (cpus_left);
318 #ifdef	CHEETAHPLUS_ERRATUM_25
319 				/*
320 				 * Clear recovered because we are sending to
321 				 * a new set of targets.
322 				 */
323 				recovered = 0;
324 #endif
325 				continue;
326 			}
327 		}
328 #endif
329 		if (curbusy) {
330 			busy++;
331 			continue;
332 		}
333 
334 #ifdef SEND_MONDO_STATS
335 		{
336 			int n = gettick() - starttick;
337 			if (n < 8192)
338 				x_nack_stimes[n >> 7]++;
339 		}
340 #endif
341 		while (gettick() < (tick + sys_clock_mhz))
342 			;
343 		do {
344 			lo = lowbit(curnack) - 1;
345 			i = IDSR_NACK_IDX(lo);
346 			shipit(cpuids[i], i);
347 			curnack &= ~(1ull << lo);
348 		} while (curnack);
349 		nack++;
350 		busy = 0;
351 	}
352 #ifdef SEND_MONDO_STATS
353 	{
354 		int n = gettick() - starttick;
355 		if (n < 8192)
356 			x_set_stimes[n >> 7]++;
357 		else
358 			x_set_ltimes[(n >> 13) & 0xf]++;
359 	}
360 	x_set_cpus[shipped]++;
361 #endif
362 }
363 
364 /*
365  * Handles error logging for implementation specific error types.
366  */
367 /*ARGSUSED*/
368 int
369 cpu_impl_async_log_err(void *flt, errorq_elem_t *eqep)
370 {
371 	/* There aren't any error types which are specific to cheetah only */
372 	return (CH_ASYNC_LOG_UNKNOWN);
373 }
374 
375 /*
376  * Figure out if Ecache is direct-mapped (Cheetah or Cheetah+ with Ecache
377  * control ECCR_ASSOC bit off or 2-way (Cheetah+ with ECCR_ASSOC on).
378  * We need to do this on the fly because we may have mixed Cheetah+'s with
379  * both direct and 2-way Ecaches.
380  */
381 int
382 cpu_ecache_nway(void)
383 {
384 	return (CH_ECACHE_NWAY);
385 }
386 
387 /*
388  * Note that these are entered into the table: Fatal Errors (PERR, IERR,
389  * ISAP, EMU) first, orphaned UCU/UCC, AFAR Overwrite policy, finally IVU, IVC.
390  * Afar overwrite policy is:
391  *   UCU,UCC > UE,EDU,WDU,CPU > CE,EDC,EMC,WDC,CPC > TO,BERR
392  */
393 ecc_type_to_info_t ecc_type_to_info[] = {
394 
395 	/* Fatal Errors */
396 	C_AFSR_PERR,	"PERR ",	ECC_ALL_TRAPS,	CPU_FATAL,
397 		"PERR Fatal",
398 		FM_EREPORT_PAYLOAD_SYSTEM2,
399 		FM_EREPORT_CPU_USIII_PERR,
400 	C_AFSR_IERR,	"IERR ", 	ECC_ALL_TRAPS,	CPU_FATAL,
401 		"IERR Fatal",
402 		FM_EREPORT_PAYLOAD_SYSTEM2,
403 		FM_EREPORT_CPU_USIII_IERR,
404 	C_AFSR_ISAP,	"ISAP ",	ECC_ALL_TRAPS,	CPU_FATAL,
405 		"ISAP Fatal",
406 		FM_EREPORT_PAYLOAD_SYSTEM1,
407 		FM_EREPORT_CPU_USIII_ISAP,
408 	C_AFSR_EMU,	"EMU ",		ECC_ASYNC_TRAPS, CPU_FATAL,
409 		"EMU Fatal",
410 		FM_EREPORT_PAYLOAD_MEMORY,
411 		FM_EREPORT_CPU_USIII_EMU,
412 
413 	/* Orphaned UCC/UCU Errors */
414 	C_AFSR_UCU,	"OUCU ",	ECC_ORPH_TRAPS, CPU_ORPH,
415 		"Orphaned UCU",
416 		FM_EREPORT_PAYLOAD_L2_DATA,
417 		FM_EREPORT_CPU_USIII_UCU,
418 	C_AFSR_UCC,	"OUCC ",	ECC_ORPH_TRAPS, CPU_ORPH,
419 		"Orphaned UCC",
420 		FM_EREPORT_PAYLOAD_L2_DATA,
421 		FM_EREPORT_CPU_USIII_UCC,
422 
423 	/* UCU, UCC */
424 	C_AFSR_UCU,	"UCU ",		ECC_F_TRAP,	CPU_UE_ECACHE,
425 		"UCU",
426 		FM_EREPORT_PAYLOAD_L2_DATA,
427 		FM_EREPORT_CPU_USIII_UCU,
428 	C_AFSR_UCC,	"UCC ",		ECC_F_TRAP,	CPU_CE_ECACHE,
429 		"UCC",
430 		FM_EREPORT_PAYLOAD_L2_DATA,
431 		FM_EREPORT_CPU_USIII_UCC,
432 
433 	/* UE, EDU:ST, EDU:BLD, WDU, CPU */
434 	C_AFSR_UE,	"UE ",		ECC_ASYNC_TRAPS, CPU_UE,
435 		"Uncorrectable system bus (UE)",
436 		FM_EREPORT_PAYLOAD_MEMORY,
437 		FM_EREPORT_CPU_USIII_UE,
438 	C_AFSR_EDU,	"EDU ",		ECC_C_TRAP,	CPU_UE_ECACHE_RETIRE,
439 		"EDU:ST",
440 		FM_EREPORT_PAYLOAD_L2_DATA,
441 		FM_EREPORT_CPU_USIII_EDUST,
442 	C_AFSR_EDU,	"EDU ",		ECC_D_TRAP,	CPU_UE_ECACHE_RETIRE,
443 		"EDU:BLD",
444 		FM_EREPORT_PAYLOAD_L2_DATA,
445 		FM_EREPORT_CPU_USIII_EDUBL,
446 	C_AFSR_WDU,	"WDU ",		ECC_C_TRAP,	CPU_UE_ECACHE_RETIRE,
447 		"WDU",
448 		FM_EREPORT_PAYLOAD_L2_DATA,
449 		FM_EREPORT_CPU_USIII_WDU,
450 	C_AFSR_CPU,	"CPU ",		ECC_C_TRAP,	CPU_UE_ECACHE,
451 		"CPU",
452 		FM_EREPORT_PAYLOAD_L2_DATA,
453 		FM_EREPORT_CPU_USIII_CPU,
454 
455 	/* CE, EDC, EMC, WDC, CPC */
456 	C_AFSR_CE,	"CE ",		ECC_C_TRAP,	CPU_CE,
457 		"Corrected system bus (CE)",
458 		FM_EREPORT_PAYLOAD_MEMORY,
459 		FM_EREPORT_CPU_USIII_CE,
460 	C_AFSR_EDC,	"EDC ",		ECC_C_TRAP,	CPU_CE_ECACHE,
461 		"EDC",
462 		FM_EREPORT_PAYLOAD_L2_DATA,
463 		FM_EREPORT_CPU_USIII_EDC,
464 	C_AFSR_EMC,	"EMC ",		ECC_C_TRAP,	CPU_EMC,
465 		"EMC",
466 		FM_EREPORT_PAYLOAD_MEMORY,
467 		FM_EREPORT_CPU_USIII_EMC,
468 	C_AFSR_WDC,	"WDC ",		ECC_C_TRAP,	CPU_CE_ECACHE,
469 		"WDC",
470 		FM_EREPORT_PAYLOAD_L2_DATA,
471 		FM_EREPORT_CPU_USIII_WDC,
472 	C_AFSR_CPC,	"CPC ",		ECC_C_TRAP,	CPU_CE_ECACHE,
473 		"CPC",
474 		FM_EREPORT_PAYLOAD_L2_DATA,
475 		FM_EREPORT_CPU_USIII_CPC,
476 
477 	/* TO, BERR */
478 	C_AFSR_TO,	"TO ",		ECC_ASYNC_TRAPS, CPU_TO,
479 		"Timeout (TO)",
480 		FM_EREPORT_PAYLOAD_IO,
481 		FM_EREPORT_CPU_USIII_TO,
482 	C_AFSR_BERR,	"BERR ",	ECC_ASYNC_TRAPS, CPU_BERR,
483 		"Bus Error (BERR)",
484 		FM_EREPORT_PAYLOAD_IO,
485 		FM_EREPORT_CPU_USIII_BERR,
486 
487 	/* IVU, IVC */
488 	C_AFSR_IVU,	"IVU ",		ECC_C_TRAP,	CPU_IV,
489 		"IVU",
490 		FM_EREPORT_PAYLOAD_SYSTEM1,
491 		FM_EREPORT_CPU_USIII_IVU,
492 	C_AFSR_IVC,	"IVC ",		ECC_C_TRAP,	CPU_IV,
493 		"IVC",
494 		FM_EREPORT_PAYLOAD_SYSTEM1,
495 		FM_EREPORT_CPU_USIII_IVC,
496 
497 	0,		NULL,		0,		0,
498 		NULL,
499 		FM_EREPORT_PAYLOAD_UNKNOWN,
500 		FM_EREPORT_CPU_USIII_UNKNOWN,
501 };
502 
503 /*
504  * Prioritized list of Error bits for AFAR overwrite.
505  * See Cheetah PRM P.6.1
506  *   Class 4:  UCC, UCU
507  *   Class 3:  UE, EDU, EMU, WDU, CPU
508  *   Class 2:  CE, EDC, EMC, WDC, CPC
509  *   Class 1:  TO, BERR
510  */
511 uint64_t afar_overwrite[] = {
512 	C_AFSR_UCC | C_AFSR_UCU,
513 	C_AFSR_UE | C_AFSR_EDU | C_AFSR_EMU | C_AFSR_WDU | C_AFSR_CPU,
514 	C_AFSR_CE | C_AFSR_EDC | C_AFSR_EMC | C_AFSR_WDC | C_AFSR_CPC,
515 	C_AFSR_TO | C_AFSR_BERR,
516 	0
517 };
518 
519 /*
520  * Prioritized list of Error bits for ESYND overwrite.
521  * See Cheetah PRM P.6.2
522  *   Class 2:  UE, IVU, EDU, WDU, UCU, CPU
523  *   Class 1:  CE, IVC, EDC, WDC, UCC, CPC
524  */
525 uint64_t esynd_overwrite[] = {
526 	C_AFSR_UE | C_AFSR_IVU | C_AFSR_EDU | C_AFSR_WDU | C_AFSR_UCU |
527 	    C_AFSR_CPU,
528 	C_AFSR_CE | C_AFSR_IVC | C_AFSR_EDC | C_AFSR_WDC | C_AFSR_UCC |
529 	    C_AFSR_CPC,
530 	0
531 };
532 
533 /*
534  * Prioritized list of Error bits for MSYND overwrite.
535  * See Cheetah PRM P.6.3
536  *   Class 2:  EMU
537  *   Class 1:  EMC
538  */
539 uint64_t msynd_overwrite[] = {
540 	C_AFSR_EMU,
541 	C_AFSR_EMC,
542 	0
543 };
544 
545 /*
546  * change cpu speed bits -- new speed will be normal-speed/divisor.
547  *
548  * The Jalapeno memory controllers are required to drain outstanding
549  * memory transactions within 32 JBus clocks in order to be ready
550  * to enter Estar mode.  In some corner cases however, that time
551  * fell short.
552  *
553  * A safe software solution is to force MCU to act like in Estar mode,
554  * then delay 1us (in ppm code) prior to assert J_CHNG_L signal.
555  * To reverse the effect, upon exiting Estar, software restores the
556  * MCU to its original state.
557  */
558 /* ARGSUSED1 */
559 void
560 cpu_change_speed(uint64_t divisor, uint64_t arg2)
561 {
562 	bus_config_eclk_t *bceclk;
563 	uint64_t		reg;
564 
565 	for (bceclk = bus_config_eclk; bceclk->divisor; bceclk++) {
566 		if (bceclk->divisor != divisor)
567 			continue;
568 		reg = get_safari_config();
569 		reg &= ~SAFARI_CONFIG_ECLK_MASK;
570 		reg |= bceclk->mask;
571 		set_safari_config(reg);
572 		CPU->cpu_m.divisor = (uchar_t)divisor;
573 		return;
574 	}
575 	/*
576 	 * We will reach here only if OBP and kernel don't agree on
577 	 * the speeds supported by the CPU.
578 	 */
579 	cmn_err(CE_WARN, "cpu_change_speed: bad divisor %" PRIu64, divisor);
580 }
581 
582 /*
583  * Cpu private initialization.  This includes allocating the cpu_private
584  * data structure, initializing it, and initializing the scrubber for this
585  * cpu.  This function calls cpu_init_ecache_scrub_dr to init the scrubber.
586  * We use kmem_cache_create for the cheetah private data structure because
587  * it needs to be allocated on a PAGESIZE (8192) byte boundary.
588  */
589 void
590 cpu_init_private(struct cpu *cp)
591 {
592 	cheetah_private_t *chprp;
593 	int i;
594 
595 	ASSERT(CPU_PRIVATE(cp) == NULL);
596 
597 	/* LINTED: E_TRUE_LOGICAL_EXPR */
598 	ASSERT((offsetof(cheetah_private_t, chpr_tl1_err_data) +
599 	    sizeof (ch_err_tl1_data_t) * CH_ERR_TL1_TLMAX) <= PAGESIZE);
600 
601 	/*
602 	 * Running with a Cheetah+, Jaguar, or Panther on a Cheetah CPU
603 	 * machine is not a supported configuration. Attempting to do so
604 	 * may result in unpredictable failures (e.g. running Cheetah+
605 	 * CPUs with Cheetah E$ disp flush) so don't allow it.
606 	 *
607 	 * This is just defensive code since this configuration mismatch
608 	 * should have been caught prior to OS execution.
609 	 */
610 	if (!IS_CHEETAH(cpunodes[cp->cpu_id].implementation)) {
611 		cmn_err(CE_PANIC, "CPU%d: UltraSPARC-III+/IV/IV+ not"
612 		    " supported on UltraSPARC-III code\n", cp->cpu_id);
613 	}
614 
615 	/*
616 	 * If the ch_private_cache has not been created, create it.
617 	 */
618 	if (ch_private_cache == NULL) {
619 		ch_private_cache = kmem_cache_create("ch_private_cache",
620 		    sizeof (cheetah_private_t), PAGESIZE, NULL, NULL,
621 		    NULL, NULL, static_arena, 0);
622 	}
623 
624 	chprp = CPU_PRIVATE(cp) = kmem_cache_alloc(ch_private_cache, KM_SLEEP);
625 
626 	bzero(chprp, sizeof (cheetah_private_t));
627 	chprp->chpr_fecctl0_logout.clo_data.chd_afar = LOGOUT_INVALID;
628 	chprp->chpr_cecc_logout.clo_data.chd_afar = LOGOUT_INVALID;
629 	chprp->chpr_async_logout.clo_data.chd_afar = LOGOUT_INVALID;
630 	for (i = 0; i < CH_ERR_TL1_TLMAX; i++)
631 		chprp->chpr_tl1_err_data[i].ch_err_tl1_logout.clo_data.chd_afar
632 		    = LOGOUT_INVALID;
633 
634 	chprp->chpr_icache_size = CH_ICACHE_SIZE;
635 	chprp->chpr_icache_linesize = CH_ICACHE_LSIZE;
636 
637 	cpu_init_ecache_scrub_dr(cp);
638 
639 	chprp->chpr_ec_set_size = cpunodes[cp->cpu_id].ecache_size /
640 	    cpu_ecache_nway();
641 
642 	adjust_hw_copy_limits(cpunodes[cp->cpu_id].ecache_size);
643 	ch_err_tl1_paddrs[cp->cpu_id] = va_to_pa(chprp);
644 	ASSERT(ch_err_tl1_paddrs[cp->cpu_id] != -1);
645 }
646 
647 /*
648  * Clear the error state registers for this CPU.
649  * For Cheetah, just clear the AFSR
650  */
651 void
652 set_cpu_error_state(ch_cpu_errors_t *cpu_error_regs)
653 {
654 	set_asyncflt(cpu_error_regs->afsr & ~C_AFSR_FATAL_ERRS);
655 }
656 
657 /*
658  * For Cheetah, the error recovery code uses an alternate flush area in the
659  * TL>0 fast ECC handler.  ecache_tl1_flushaddr is the physical address of
660  * this exclusive displacement flush area.
661  */
662 uint64_t ecache_tl1_flushaddr = (uint64_t)-1; /* physaddr for E$ flushing */
663 
664 /*
665  * Allocate and initialize the exclusive displacement flush area.
666  * Must be called before startup_bop_gone().
667  */
668 caddr_t
669 ecache_init_scrub_flush_area(caddr_t alloc_base)
670 {
671 	unsigned size = 2 * CH_ECACHE_8M_SIZE;
672 	caddr_t tmp_alloc_base = alloc_base;
673 	caddr_t flush_alloc_base =
674 	    (caddr_t)roundup((uintptr_t)alloc_base, size);
675 	caddr_t ecache_tl1_virtaddr;
676 
677 	/*
678 	 * Allocate the physical memory for the exclusive flush area
679 	 *
680 	 * Need to allocate an exclusive flush area that is twice the
681 	 * largest supported E$ size, physically contiguous, and
682 	 * aligned on twice the largest E$ size boundary.
683 	 *
684 	 * Memory allocated via BOP_ALLOC is included in the "cage"
685 	 * from the DR perspective and due to this, its physical
686 	 * address will never change and the memory will not be
687 	 * removed.
688 	 *
689 	 * BOP_ALLOC takes 4 arguments: bootops, virtual address hint,
690 	 * size of the area to allocate, and alignment of the area to
691 	 * allocate. It returns zero if the allocation fails, or the
692 	 * virtual address for a successful allocation. Memory BOP_ALLOC'd
693 	 * is physically contiguous.
694 	 */
695 	if ((ecache_tl1_virtaddr = (caddr_t)BOP_ALLOC(bootops,
696 	    flush_alloc_base, size, size)) != NULL) {
697 
698 		tmp_alloc_base =
699 		    (caddr_t)roundup((uintptr_t)(ecache_tl1_virtaddr + size),
700 		    ecache_alignsize);
701 
702 		/*
703 		 * get the physical address of the exclusive flush area
704 		 */
705 		ecache_tl1_flushaddr = va_to_pa(ecache_tl1_virtaddr);
706 
707 	} else {
708 		ecache_tl1_virtaddr = (caddr_t)-1;
709 		cmn_err(CE_NOTE, "!ecache_init_scrub_flush_area failed\n");
710 	}
711 
712 	return (tmp_alloc_base);
713 }
714 
715 /*
716  * Update cpu_offline_set so the scrubber knows which cpus are offline
717  */
718 /*ARGSUSED*/
719 int
720 cpu_scrub_cpu_setup(cpu_setup_t what, int cpuid, void *arg)
721 {
722 	switch (what) {
723 	case CPU_ON:
724 	case CPU_INIT:
725 		CPUSET_DEL(cpu_offline_set, cpuid);
726 		break;
727 	case CPU_OFF:
728 		CPUSET_ADD(cpu_offline_set, cpuid);
729 		break;
730 	default:
731 		break;
732 	}
733 	return (0);
734 }
735