xref: /titanic_51/usr/src/uts/sun4u/cpu/us3_cheetah.c (revision 193974072f41a843678abf5f61979c748687e66b)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 #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 	max_bootlp_tteszc = TTE8K;
155 }
156 
157 void
158 send_mondo_set(cpuset_t set)
159 {
160 	int lo, busy, nack, shipped = 0;
161 	uint16_t i, cpuids[IDSR_BN_SETS];
162 	uint64_t idsr, nackmask = 0, busymask, curnack, curbusy;
163 	uint64_t starttick, endtick, tick, lasttick;
164 #if (NCPU > IDSR_BN_SETS)
165 	int index = 0;
166 	int ncpuids = 0;
167 #endif
168 #ifdef	CHEETAHPLUS_ERRATUM_25
169 	int recovered = 0;
170 	int cpuid;
171 #endif
172 
173 	ASSERT(!CPUSET_ISNULL(set));
174 	starttick = lasttick = gettick();
175 
176 #if (NCPU <= IDSR_BN_SETS)
177 	for (i = 0; i < NCPU; i++)
178 		if (CPU_IN_SET(set, i)) {
179 			shipit(i, shipped);
180 			nackmask |= IDSR_NACK_BIT(shipped);
181 			cpuids[shipped++] = i;
182 			CPUSET_DEL(set, i);
183 			if (CPUSET_ISNULL(set))
184 				break;
185 		}
186 	CPU_STATS_ADDQ(CPU, sys, xcalls, shipped);
187 #else
188 	for (i = 0; i < NCPU; i++)
189 		if (CPU_IN_SET(set, i)) {
190 			ncpuids++;
191 
192 			/*
193 			 * Ship only to the first (IDSR_BN_SETS) CPUs.  If we
194 			 * find we have shipped to more than (IDSR_BN_SETS)
195 			 * CPUs, set "index" to the highest numbered CPU in
196 			 * the set so we can ship to other CPUs a bit later on.
197 			 */
198 			if (shipped < IDSR_BN_SETS) {
199 				shipit(i, shipped);
200 				nackmask |= IDSR_NACK_BIT(shipped);
201 				cpuids[shipped++] = i;
202 				CPUSET_DEL(set, i);
203 				if (CPUSET_ISNULL(set))
204 					break;
205 			} else
206 				index = (int)i;
207 		}
208 
209 	CPU_STATS_ADDQ(CPU, sys, xcalls, ncpuids);
210 #endif
211 
212 	busymask = IDSR_NACK_TO_BUSY(nackmask);
213 	busy = nack = 0;
214 	endtick = starttick + xc_tick_limit;
215 	for (;;) {
216 		idsr = getidsr();
217 #if (NCPU <= IDSR_BN_SETS)
218 		if (idsr == 0)
219 			break;
220 #else
221 		if (idsr == 0 && shipped == ncpuids)
222 			break;
223 #endif
224 		tick = gettick();
225 		/*
226 		 * If there is a big jump between the current tick
227 		 * count and lasttick, we have probably hit a break
228 		 * point.  Adjust endtick accordingly to avoid panic.
229 		 */
230 		if (tick > (lasttick + xc_tick_jump_limit))
231 			endtick += (tick - lasttick);
232 		lasttick = tick;
233 		if (tick > endtick) {
234 			if (panic_quiesce)
235 				return;
236 #ifdef	CHEETAHPLUS_ERRATUM_25
237 			cpuid = -1;
238 			for (i = 0; i < IDSR_BN_SETS; i++) {
239 				if (idsr & (IDSR_NACK_BIT(i) |
240 				    IDSR_BUSY_BIT(i))) {
241 					cpuid = cpuids[i];
242 					break;
243 				}
244 			}
245 			if (cheetah_sendmondo_recover && cpuid != -1 &&
246 			    recovered == 0) {
247 				if (mondo_recover(cpuid, i)) {
248 					/*
249 					 * We claimed the whole memory or
250 					 * full scan is disabled.
251 					 */
252 					recovered++;
253 				}
254 				tick = gettick();
255 				endtick = tick + xc_tick_limit;
256 				lasttick = tick;
257 				/*
258 				 * Recheck idsr
259 				 */
260 				continue;
261 			} else
262 #endif	/* CHEETAHPLUS_ERRATUM_25 */
263 			{
264 				cmn_err(CE_CONT, "send mondo timeout "
265 				    "[%d NACK %d BUSY]\nIDSR 0x%"
266 				    "" PRIx64 "  cpuids:", nack, busy, idsr);
267 				for (i = 0; i < IDSR_BN_SETS; i++) {
268 					if (idsr & (IDSR_NACK_BIT(i) |
269 					    IDSR_BUSY_BIT(i))) {
270 						cmn_err(CE_CONT, " 0x%x",
271 						    cpuids[i]);
272 					}
273 				}
274 				cmn_err(CE_CONT, "\n");
275 				cmn_err(CE_PANIC, "send_mondo_set: timeout");
276 			}
277 		}
278 		curnack = idsr & nackmask;
279 		curbusy = idsr & busymask;
280 #if (NCPU > IDSR_BN_SETS)
281 		if (shipped < ncpuids) {
282 			uint64_t cpus_left;
283 			uint16_t next = (uint16_t)index;
284 
285 			cpus_left = ~(IDSR_NACK_TO_BUSY(curnack) | curbusy) &
286 			    busymask;
287 
288 			if (cpus_left) {
289 				do {
290 					/*
291 					 * Sequence through and ship to the
292 					 * remainder of the CPUs in the system
293 					 * (e.g. other than the first
294 					 * (IDSR_BN_SETS)) in reverse order.
295 					 */
296 					lo = lowbit(cpus_left) - 1;
297 					i = IDSR_BUSY_IDX(lo);
298 					shipit(next, i);
299 					shipped++;
300 					cpuids[i] = next;
301 
302 					/*
303 					 * If we've processed all the CPUs,
304 					 * exit the loop now and save
305 					 * instructions.
306 					 */
307 					if (shipped == ncpuids)
308 						break;
309 
310 					for ((index = ((int)next - 1));
311 					    index >= 0; index--)
312 						if (CPU_IN_SET(set, index)) {
313 							next = (uint16_t)index;
314 							break;
315 						}
316 
317 					cpus_left &= ~(1ull << lo);
318 				} while (cpus_left);
319 #ifdef	CHEETAHPLUS_ERRATUM_25
320 				/*
321 				 * Clear recovered because we are sending to
322 				 * a new set of targets.
323 				 */
324 				recovered = 0;
325 #endif
326 				continue;
327 			}
328 		}
329 #endif
330 		if (curbusy) {
331 			busy++;
332 			continue;
333 		}
334 
335 #ifdef SEND_MONDO_STATS
336 		{
337 			int n = gettick() - starttick;
338 			if (n < 8192)
339 				x_nack_stimes[n >> 7]++;
340 		}
341 #endif
342 		while (gettick() < (tick + sys_clock_mhz))
343 			;
344 		do {
345 			lo = lowbit(curnack) - 1;
346 			i = IDSR_NACK_IDX(lo);
347 			shipit(cpuids[i], i);
348 			curnack &= ~(1ull << lo);
349 		} while (curnack);
350 		nack++;
351 		busy = 0;
352 	}
353 #ifdef SEND_MONDO_STATS
354 	{
355 		int n = gettick() - starttick;
356 		if (n < 8192)
357 			x_set_stimes[n >> 7]++;
358 		else
359 			x_set_ltimes[(n >> 13) & 0xf]++;
360 	}
361 	x_set_cpus[shipped]++;
362 #endif
363 }
364 
365 /*
366  * Handles error logging for implementation specific error types.
367  */
368 /*ARGSUSED*/
369 int
370 cpu_impl_async_log_err(void *flt, errorq_elem_t *eqep)
371 {
372 	/* There aren't any error types which are specific to cheetah only */
373 	return (CH_ASYNC_LOG_UNKNOWN);
374 }
375 
376 /*
377  * Figure out if Ecache is direct-mapped (Cheetah or Cheetah+ with Ecache
378  * control ECCR_ASSOC bit off or 2-way (Cheetah+ with ECCR_ASSOC on).
379  * We need to do this on the fly because we may have mixed Cheetah+'s with
380  * both direct and 2-way Ecaches.
381  */
382 int
383 cpu_ecache_nway(void)
384 {
385 	return (CH_ECACHE_NWAY);
386 }
387 
388 /*
389  * Note that these are entered into the table: Fatal Errors (PERR, IERR,
390  * ISAP, EMU) first, orphaned UCU/UCC, AFAR Overwrite policy, finally IVU, IVC.
391  * Afar overwrite policy is:
392  *   UCU,UCC > UE,EDU,WDU,CPU > CE,EDC,EMC,WDC,CPC > TO,BERR
393  */
394 ecc_type_to_info_t ecc_type_to_info[] = {
395 
396 	/* Fatal Errors */
397 	C_AFSR_PERR,	"PERR ",	ECC_ALL_TRAPS,	CPU_FATAL,
398 		"PERR Fatal",
399 		FM_EREPORT_PAYLOAD_SYSTEM2,
400 		FM_EREPORT_CPU_USIII_PERR,
401 	C_AFSR_IERR,	"IERR ", 	ECC_ALL_TRAPS,	CPU_FATAL,
402 		"IERR Fatal",
403 		FM_EREPORT_PAYLOAD_SYSTEM2,
404 		FM_EREPORT_CPU_USIII_IERR,
405 	C_AFSR_ISAP,	"ISAP ",	ECC_ALL_TRAPS,	CPU_FATAL,
406 		"ISAP Fatal",
407 		FM_EREPORT_PAYLOAD_SYSTEM1,
408 		FM_EREPORT_CPU_USIII_ISAP,
409 	C_AFSR_EMU,	"EMU ",		ECC_ASYNC_TRAPS, CPU_FATAL,
410 		"EMU Fatal",
411 		FM_EREPORT_PAYLOAD_MEMORY,
412 		FM_EREPORT_CPU_USIII_EMU,
413 
414 	/* Orphaned UCC/UCU Errors */
415 	C_AFSR_UCU,	"OUCU ",	ECC_ORPH_TRAPS, CPU_ORPH,
416 		"Orphaned UCU",
417 		FM_EREPORT_PAYLOAD_L2_DATA,
418 		FM_EREPORT_CPU_USIII_UCU,
419 	C_AFSR_UCC,	"OUCC ",	ECC_ORPH_TRAPS, CPU_ORPH,
420 		"Orphaned UCC",
421 		FM_EREPORT_PAYLOAD_L2_DATA,
422 		FM_EREPORT_CPU_USIII_UCC,
423 
424 	/* UCU, UCC */
425 	C_AFSR_UCU,	"UCU ",		ECC_F_TRAP,	CPU_UE_ECACHE,
426 		"UCU",
427 		FM_EREPORT_PAYLOAD_L2_DATA,
428 		FM_EREPORT_CPU_USIII_UCU,
429 	C_AFSR_UCC,	"UCC ",		ECC_F_TRAP,	CPU_CE_ECACHE,
430 		"UCC",
431 		FM_EREPORT_PAYLOAD_L2_DATA,
432 		FM_EREPORT_CPU_USIII_UCC,
433 
434 	/* UE, EDU:ST, EDU:BLD, WDU, CPU */
435 	C_AFSR_UE,	"UE ",		ECC_ASYNC_TRAPS, CPU_UE,
436 		"Uncorrectable system bus (UE)",
437 		FM_EREPORT_PAYLOAD_MEMORY,
438 		FM_EREPORT_CPU_USIII_UE,
439 	C_AFSR_EDU,	"EDU ",		ECC_C_TRAP,	CPU_UE_ECACHE_RETIRE,
440 		"EDU:ST",
441 		FM_EREPORT_PAYLOAD_L2_DATA,
442 		FM_EREPORT_CPU_USIII_EDUST,
443 	C_AFSR_EDU,	"EDU ",		ECC_D_TRAP,	CPU_UE_ECACHE_RETIRE,
444 		"EDU:BLD",
445 		FM_EREPORT_PAYLOAD_L2_DATA,
446 		FM_EREPORT_CPU_USIII_EDUBL,
447 	C_AFSR_WDU,	"WDU ",		ECC_C_TRAP,	CPU_UE_ECACHE_RETIRE,
448 		"WDU",
449 		FM_EREPORT_PAYLOAD_L2_DATA,
450 		FM_EREPORT_CPU_USIII_WDU,
451 	C_AFSR_CPU,	"CPU ",		ECC_C_TRAP,	CPU_UE_ECACHE,
452 		"CPU",
453 		FM_EREPORT_PAYLOAD_L2_DATA,
454 		FM_EREPORT_CPU_USIII_CPU,
455 
456 	/* CE, EDC, EMC, WDC, CPC */
457 	C_AFSR_CE,	"CE ",		ECC_C_TRAP,	CPU_CE,
458 		"Corrected system bus (CE)",
459 		FM_EREPORT_PAYLOAD_MEMORY,
460 		FM_EREPORT_CPU_USIII_CE,
461 	C_AFSR_EDC,	"EDC ",		ECC_C_TRAP,	CPU_CE_ECACHE,
462 		"EDC",
463 		FM_EREPORT_PAYLOAD_L2_DATA,
464 		FM_EREPORT_CPU_USIII_EDC,
465 	C_AFSR_EMC,	"EMC ",		ECC_C_TRAP,	CPU_EMC,
466 		"EMC",
467 		FM_EREPORT_PAYLOAD_MEMORY,
468 		FM_EREPORT_CPU_USIII_EMC,
469 	C_AFSR_WDC,	"WDC ",		ECC_C_TRAP,	CPU_CE_ECACHE,
470 		"WDC",
471 		FM_EREPORT_PAYLOAD_L2_DATA,
472 		FM_EREPORT_CPU_USIII_WDC,
473 	C_AFSR_CPC,	"CPC ",		ECC_C_TRAP,	CPU_CE_ECACHE,
474 		"CPC",
475 		FM_EREPORT_PAYLOAD_L2_DATA,
476 		FM_EREPORT_CPU_USIII_CPC,
477 
478 	/* TO, BERR */
479 	C_AFSR_TO,	"TO ",		ECC_ASYNC_TRAPS, CPU_TO,
480 		"Timeout (TO)",
481 		FM_EREPORT_PAYLOAD_IO,
482 		FM_EREPORT_CPU_USIII_TO,
483 	C_AFSR_BERR,	"BERR ",	ECC_ASYNC_TRAPS, CPU_BERR,
484 		"Bus Error (BERR)",
485 		FM_EREPORT_PAYLOAD_IO,
486 		FM_EREPORT_CPU_USIII_BERR,
487 
488 	/* IVU, IVC */
489 	C_AFSR_IVU,	"IVU ",		ECC_C_TRAP,	CPU_IV,
490 		"IVU",
491 		FM_EREPORT_PAYLOAD_SYSTEM1,
492 		FM_EREPORT_CPU_USIII_IVU,
493 	C_AFSR_IVC,	"IVC ",		ECC_C_TRAP,	CPU_IV,
494 		"IVC",
495 		FM_EREPORT_PAYLOAD_SYSTEM1,
496 		FM_EREPORT_CPU_USIII_IVC,
497 
498 	0,		NULL,		0,		0,
499 		NULL,
500 		FM_EREPORT_PAYLOAD_UNKNOWN,
501 		FM_EREPORT_CPU_USIII_UNKNOWN,
502 };
503 
504 /*
505  * Prioritized list of Error bits for AFAR overwrite.
506  * See Cheetah PRM P.6.1
507  *   Class 4:  UCC, UCU
508  *   Class 3:  UE, EDU, EMU, WDU, CPU
509  *   Class 2:  CE, EDC, EMC, WDC, CPC
510  *   Class 1:  TO, BERR
511  */
512 uint64_t afar_overwrite[] = {
513 	C_AFSR_UCC | C_AFSR_UCU,
514 	C_AFSR_UE | C_AFSR_EDU | C_AFSR_EMU | C_AFSR_WDU | C_AFSR_CPU,
515 	C_AFSR_CE | C_AFSR_EDC | C_AFSR_EMC | C_AFSR_WDC | C_AFSR_CPC,
516 	C_AFSR_TO | C_AFSR_BERR,
517 	0
518 };
519 
520 /*
521  * Prioritized list of Error bits for ESYND overwrite.
522  * See Cheetah PRM P.6.2
523  *   Class 2:  UE, IVU, EDU, WDU, UCU, CPU
524  *   Class 1:  CE, IVC, EDC, WDC, UCC, CPC
525  */
526 uint64_t esynd_overwrite[] = {
527 	C_AFSR_UE | C_AFSR_IVU | C_AFSR_EDU | C_AFSR_WDU | C_AFSR_UCU |
528 	    C_AFSR_CPU,
529 	C_AFSR_CE | C_AFSR_IVC | C_AFSR_EDC | C_AFSR_WDC | C_AFSR_UCC |
530 	    C_AFSR_CPC,
531 	0
532 };
533 
534 /*
535  * Prioritized list of Error bits for MSYND overwrite.
536  * See Cheetah PRM P.6.3
537  *   Class 2:  EMU
538  *   Class 1:  EMC
539  */
540 uint64_t msynd_overwrite[] = {
541 	C_AFSR_EMU,
542 	C_AFSR_EMC,
543 	0
544 };
545 
546 /*
547  * change cpu speed bits -- new speed will be normal-speed/divisor.
548  *
549  * The Jalapeno memory controllers are required to drain outstanding
550  * memory transactions within 32 JBus clocks in order to be ready
551  * to enter Estar mode.  In some corner cases however, that time
552  * fell short.
553  *
554  * A safe software solution is to force MCU to act like in Estar mode,
555  * then delay 1us (in ppm code) prior to assert J_CHNG_L signal.
556  * To reverse the effect, upon exiting Estar, software restores the
557  * MCU to its original state.
558  */
559 /* ARGSUSED1 */
560 void
561 cpu_change_speed(uint64_t divisor, uint64_t arg2)
562 {
563 	bus_config_eclk_t	*bceclk;
564 	uint64_t		reg;
565 	processor_info_t	*pi = &(CPU->cpu_type_info);
566 
567 	for (bceclk = bus_config_eclk; bceclk->divisor; bceclk++) {
568 		if (bceclk->divisor != divisor)
569 			continue;
570 		reg = get_safari_config();
571 		reg &= ~SAFARI_CONFIG_ECLK_MASK;
572 		reg |= bceclk->mask;
573 		set_safari_config(reg);
574 		CPU->cpu_m.divisor = (uchar_t)divisor;
575 		CPU->cpu_curr_clock =
576 		    (((uint64_t)pi->pi_clock * 1000000) / divisor);
577 		return;
578 	}
579 	/*
580 	 * We will reach here only if OBP and kernel don't agree on
581 	 * the speeds supported by the CPU.
582 	 */
583 	cmn_err(CE_WARN, "cpu_change_speed: bad divisor %" PRIu64, divisor);
584 }
585 
586 /*
587  * Cpu private initialization.  This includes allocating the cpu_private
588  * data structure, initializing it, and initializing the scrubber for this
589  * cpu.  This function calls cpu_init_ecache_scrub_dr to init the scrubber.
590  * We use kmem_cache_create for the cheetah private data structure because
591  * it needs to be allocated on a PAGESIZE (8192) byte boundary.
592  */
593 void
594 cpu_init_private(struct cpu *cp)
595 {
596 	cheetah_private_t *chprp;
597 	int i;
598 
599 	ASSERT(CPU_PRIVATE(cp) == NULL);
600 
601 	/* LINTED: E_TRUE_LOGICAL_EXPR */
602 	ASSERT((offsetof(cheetah_private_t, chpr_tl1_err_data) +
603 	    sizeof (ch_err_tl1_data_t) * CH_ERR_TL1_TLMAX) <= PAGESIZE);
604 
605 	/*
606 	 * Running with a Cheetah+, Jaguar, or Panther on a Cheetah CPU
607 	 * machine is not a supported configuration. Attempting to do so
608 	 * may result in unpredictable failures (e.g. running Cheetah+
609 	 * CPUs with Cheetah E$ disp flush) so don't allow it.
610 	 *
611 	 * This is just defensive code since this configuration mismatch
612 	 * should have been caught prior to OS execution.
613 	 */
614 	if (!IS_CHEETAH(cpunodes[cp->cpu_id].implementation)) {
615 		cmn_err(CE_PANIC, "CPU%d: UltraSPARC-III+/IV/IV+ not"
616 		    " supported on UltraSPARC-III code\n", cp->cpu_id);
617 	}
618 
619 	/*
620 	 * If the ch_private_cache has not been created, create it.
621 	 */
622 	if (ch_private_cache == NULL) {
623 		ch_private_cache = kmem_cache_create("ch_private_cache",
624 		    sizeof (cheetah_private_t), PAGESIZE, NULL, NULL,
625 		    NULL, NULL, static_arena, 0);
626 	}
627 
628 	chprp = CPU_PRIVATE(cp) = kmem_cache_alloc(ch_private_cache, KM_SLEEP);
629 
630 	bzero(chprp, sizeof (cheetah_private_t));
631 	chprp->chpr_fecctl0_logout.clo_data.chd_afar = LOGOUT_INVALID;
632 	chprp->chpr_cecc_logout.clo_data.chd_afar = LOGOUT_INVALID;
633 	chprp->chpr_async_logout.clo_data.chd_afar = LOGOUT_INVALID;
634 	for (i = 0; i < CH_ERR_TL1_TLMAX; i++)
635 		chprp->chpr_tl1_err_data[i].ch_err_tl1_logout.clo_data.chd_afar
636 		    = LOGOUT_INVALID;
637 
638 	chprp->chpr_icache_size = CH_ICACHE_SIZE;
639 	chprp->chpr_icache_linesize = CH_ICACHE_LSIZE;
640 
641 	cpu_init_ecache_scrub_dr(cp);
642 
643 	chprp->chpr_ec_set_size = cpunodes[cp->cpu_id].ecache_size /
644 	    cpu_ecache_nway();
645 
646 	adjust_hw_copy_limits(cpunodes[cp->cpu_id].ecache_size);
647 	ch_err_tl1_paddrs[cp->cpu_id] = va_to_pa(chprp);
648 	ASSERT(ch_err_tl1_paddrs[cp->cpu_id] != -1);
649 }
650 
651 /*
652  * Clear the error state registers for this CPU.
653  * For Cheetah, just clear the AFSR
654  */
655 void
656 set_cpu_error_state(ch_cpu_errors_t *cpu_error_regs)
657 {
658 	set_asyncflt(cpu_error_regs->afsr & ~C_AFSR_FATAL_ERRS);
659 }
660 
661 /*
662  * For Cheetah, the error recovery code uses an alternate flush area in the
663  * TL>0 fast ECC handler.  ecache_tl1_flushaddr is the physical address of
664  * this exclusive displacement flush area.
665  */
666 uint64_t ecache_tl1_flushaddr = (uint64_t)-1; /* physaddr for E$ flushing */
667 
668 /*
669  * Allocate and initialize the exclusive displacement flush area.
670  */
671 caddr_t
672 ecache_init_scrub_flush_area(caddr_t alloc_base)
673 {
674 	unsigned size = 2 * CH_ECACHE_8M_SIZE;
675 	caddr_t tmp_alloc_base = alloc_base;
676 	caddr_t flush_alloc_base =
677 	    (caddr_t)roundup((uintptr_t)alloc_base, size);
678 	caddr_t ecache_tl1_virtaddr;
679 
680 	/*
681 	 * Allocate the physical memory for the exclusive flush area
682 	 *
683 	 * Need to allocate an exclusive flush area that is twice the
684 	 * largest supported E$ size, physically contiguous, and
685 	 * aligned on twice the largest E$ size boundary.
686 	 *
687 	 * Memory allocated via prom_alloc is included in the "cage"
688 	 * from the DR perspective and due to this, its physical
689 	 * address will never change and the memory will not be
690 	 * removed.
691 	 *
692 	 * prom_alloc takes 3 arguments: bootops, virtual address hint,
693 	 * size of the area to allocate, and alignment of the area to
694 	 * allocate. It returns zero if the allocation fails, or the
695 	 * virtual address for a successful allocation. Memory prom_alloc'd
696 	 * is physically contiguous.
697 	 */
698 	if ((ecache_tl1_virtaddr =
699 	    prom_alloc(flush_alloc_base, size, size)) != NULL) {
700 
701 		tmp_alloc_base =
702 		    (caddr_t)roundup((uintptr_t)(ecache_tl1_virtaddr + size),
703 		    ecache_alignsize);
704 
705 		/*
706 		 * get the physical address of the exclusive flush area
707 		 */
708 		ecache_tl1_flushaddr = va_to_pa(ecache_tl1_virtaddr);
709 
710 	} else {
711 		ecache_tl1_virtaddr = (caddr_t)-1;
712 		cmn_err(CE_NOTE, "!ecache_init_scrub_flush_area failed\n");
713 	}
714 
715 	return (tmp_alloc_base);
716 }
717 
718 /*
719  * Update cpu_offline_set so the scrubber knows which cpus are offline
720  */
721 /*ARGSUSED*/
722 int
723 cpu_scrub_cpu_setup(cpu_setup_t what, int cpuid, void *arg)
724 {
725 	switch (what) {
726 	case CPU_ON:
727 	case CPU_INIT:
728 		CPUSET_DEL(cpu_offline_set, cpuid);
729 		break;
730 	case CPU_OFF:
731 		CPUSET_ADD(cpu_offline_set, cpuid);
732 		break;
733 	default:
734 		break;
735 	}
736 	return (0);
737 }
738