xref: /linux/arch/powerpc/platforms/powermac/smp.c (revision c537b994505099b7197e7d3125b942ecbcc51eb6)
1 /*
2  * SMP support for power macintosh.
3  *
4  * We support both the old "powersurge" SMP architecture
5  * and the current Core99 (G4 PowerMac) machines.
6  *
7  * Note that we don't support the very first rev. of
8  * Apple/DayStar 2 CPUs board, the one with the funky
9  * watchdog. Hopefully, none of these should be there except
10  * maybe internally to Apple. I should probably still add some
11  * code to detect this card though and disable SMP. --BenH.
12  *
13  * Support Macintosh G4 SMP by Troy Benjegerdes (hozer@drgw.net)
14  * and Ben Herrenschmidt <benh@kernel.crashing.org>.
15  *
16  * Support for DayStar quad CPU cards
17  * Copyright (C) XLR8, Inc. 1994-2000
18  *
19  *  This program is free software; you can redistribute it and/or
20  *  modify it under the terms of the GNU General Public License
21  *  as published by the Free Software Foundation; either version
22  *  2 of the License, or (at your option) any later version.
23  */
24 #include <linux/kernel.h>
25 #include <linux/sched.h>
26 #include <linux/smp.h>
27 #include <linux/smp_lock.h>
28 #include <linux/interrupt.h>
29 #include <linux/kernel_stat.h>
30 #include <linux/delay.h>
31 #include <linux/init.h>
32 #include <linux/spinlock.h>
33 #include <linux/errno.h>
34 #include <linux/hardirq.h>
35 #include <linux/cpu.h>
36 #include <linux/compiler.h>
37 
38 #include <asm/ptrace.h>
39 #include <asm/atomic.h>
40 #include <asm/irq.h>
41 #include <asm/page.h>
42 #include <asm/pgtable.h>
43 #include <asm/sections.h>
44 #include <asm/io.h>
45 #include <asm/prom.h>
46 #include <asm/smp.h>
47 #include <asm/machdep.h>
48 #include <asm/pmac_feature.h>
49 #include <asm/time.h>
50 #include <asm/mpic.h>
51 #include <asm/cacheflush.h>
52 #include <asm/keylargo.h>
53 #include <asm/pmac_low_i2c.h>
54 #include <asm/pmac_pfunc.h>
55 
56 #define DEBUG
57 
58 #ifdef DEBUG
59 #define DBG(fmt...) udbg_printf(fmt)
60 #else
61 #define DBG(fmt...)
62 #endif
63 
64 extern void __secondary_start_pmac_0(void);
65 extern int pmac_pfunc_base_install(void);
66 
67 #ifdef CONFIG_PPC32
68 
69 /* Sync flag for HW tb sync */
70 static volatile int sec_tb_reset = 0;
71 
72 /*
73  * Powersurge (old powermac SMP) support.
74  */
75 
76 /* Addresses for powersurge registers */
77 #define HAMMERHEAD_BASE		0xf8000000
78 #define HHEAD_CONFIG		0x90
79 #define HHEAD_SEC_INTR		0xc0
80 
81 /* register for interrupting the primary processor on the powersurge */
82 /* N.B. this is actually the ethernet ROM! */
83 #define PSURGE_PRI_INTR		0xf3019000
84 
85 /* register for storing the start address for the secondary processor */
86 /* N.B. this is the PCI config space address register for the 1st bridge */
87 #define PSURGE_START		0xf2800000
88 
89 /* Daystar/XLR8 4-CPU card */
90 #define PSURGE_QUAD_REG_ADDR	0xf8800000
91 
92 #define PSURGE_QUAD_IRQ_SET	0
93 #define PSURGE_QUAD_IRQ_CLR	1
94 #define PSURGE_QUAD_IRQ_PRIMARY	2
95 #define PSURGE_QUAD_CKSTOP_CTL	3
96 #define PSURGE_QUAD_PRIMARY_ARB	4
97 #define PSURGE_QUAD_BOARD_ID	6
98 #define PSURGE_QUAD_WHICH_CPU	7
99 #define PSURGE_QUAD_CKSTOP_RDBK	8
100 #define PSURGE_QUAD_RESET_CTL	11
101 
102 #define PSURGE_QUAD_OUT(r, v)	(out_8(quad_base + ((r) << 4) + 4, (v)))
103 #define PSURGE_QUAD_IN(r)	(in_8(quad_base + ((r) << 4) + 4) & 0x0f)
104 #define PSURGE_QUAD_BIS(r, v)	(PSURGE_QUAD_OUT((r), PSURGE_QUAD_IN(r) | (v)))
105 #define PSURGE_QUAD_BIC(r, v)	(PSURGE_QUAD_OUT((r), PSURGE_QUAD_IN(r) & ~(v)))
106 
107 /* virtual addresses for the above */
108 static volatile u8 __iomem *hhead_base;
109 static volatile u8 __iomem *quad_base;
110 static volatile u32 __iomem *psurge_pri_intr;
111 static volatile u8 __iomem *psurge_sec_intr;
112 static volatile u32 __iomem *psurge_start;
113 
114 /* values for psurge_type */
115 #define PSURGE_NONE		-1
116 #define PSURGE_DUAL		0
117 #define PSURGE_QUAD_OKEE	1
118 #define PSURGE_QUAD_COTTON	2
119 #define PSURGE_QUAD_ICEGRASS	3
120 
121 /* what sort of powersurge board we have */
122 static int psurge_type = PSURGE_NONE;
123 
124 /*
125  * Set and clear IPIs for powersurge.
126  */
127 static inline void psurge_set_ipi(int cpu)
128 {
129 	if (psurge_type == PSURGE_NONE)
130 		return;
131 	if (cpu == 0)
132 		in_be32(psurge_pri_intr);
133 	else if (psurge_type == PSURGE_DUAL)
134 		out_8(psurge_sec_intr, 0);
135 	else
136 		PSURGE_QUAD_OUT(PSURGE_QUAD_IRQ_SET, 1 << cpu);
137 }
138 
139 static inline void psurge_clr_ipi(int cpu)
140 {
141 	if (cpu > 0) {
142 		switch(psurge_type) {
143 		case PSURGE_DUAL:
144 			out_8(psurge_sec_intr, ~0);
145 		case PSURGE_NONE:
146 			break;
147 		default:
148 			PSURGE_QUAD_OUT(PSURGE_QUAD_IRQ_CLR, 1 << cpu);
149 		}
150 	}
151 }
152 
153 /*
154  * On powersurge (old SMP powermac architecture) we don't have
155  * separate IPIs for separate messages like openpic does.  Instead
156  * we have a bitmap for each processor, where a 1 bit means that
157  * the corresponding message is pending for that processor.
158  * Ideally each cpu's entry would be in a different cache line.
159  *  -- paulus.
160  */
161 static unsigned long psurge_smp_message[NR_CPUS];
162 
163 void psurge_smp_message_recv(void)
164 {
165 	int cpu = smp_processor_id();
166 	int msg;
167 
168 	/* clear interrupt */
169 	psurge_clr_ipi(cpu);
170 
171 	if (num_online_cpus() < 2)
172 		return;
173 
174 	/* make sure there is a message there */
175 	for (msg = 0; msg < 4; msg++)
176 		if (test_and_clear_bit(msg, &psurge_smp_message[cpu]))
177 			smp_message_recv(msg);
178 }
179 
180 irqreturn_t psurge_primary_intr(int irq, void *d)
181 {
182 	psurge_smp_message_recv();
183 	return IRQ_HANDLED;
184 }
185 
186 static void smp_psurge_message_pass(int target, int msg)
187 {
188 	int i;
189 
190 	if (num_online_cpus() < 2)
191 		return;
192 
193 	for_each_online_cpu(i) {
194 		if (target == MSG_ALL
195 		    || (target == MSG_ALL_BUT_SELF && i != smp_processor_id())
196 		    || target == i) {
197 			set_bit(msg, &psurge_smp_message[i]);
198 			psurge_set_ipi(i);
199 		}
200 	}
201 }
202 
203 /*
204  * Determine a quad card presence. We read the board ID register, we
205  * force the data bus to change to something else, and we read it again.
206  * It it's stable, then the register probably exist (ugh !)
207  */
208 static int __init psurge_quad_probe(void)
209 {
210 	int type;
211 	unsigned int i;
212 
213 	type = PSURGE_QUAD_IN(PSURGE_QUAD_BOARD_ID);
214 	if (type < PSURGE_QUAD_OKEE || type > PSURGE_QUAD_ICEGRASS
215 	    || type != PSURGE_QUAD_IN(PSURGE_QUAD_BOARD_ID))
216 		return PSURGE_DUAL;
217 
218 	/* looks OK, try a slightly more rigorous test */
219 	/* bogus is not necessarily cacheline-aligned,
220 	   though I don't suppose that really matters.  -- paulus */
221 	for (i = 0; i < 100; i++) {
222 		volatile u32 bogus[8];
223 		bogus[(0+i)%8] = 0x00000000;
224 		bogus[(1+i)%8] = 0x55555555;
225 		bogus[(2+i)%8] = 0xFFFFFFFF;
226 		bogus[(3+i)%8] = 0xAAAAAAAA;
227 		bogus[(4+i)%8] = 0x33333333;
228 		bogus[(5+i)%8] = 0xCCCCCCCC;
229 		bogus[(6+i)%8] = 0xCCCCCCCC;
230 		bogus[(7+i)%8] = 0x33333333;
231 		wmb();
232 		asm volatile("dcbf 0,%0" : : "r" (bogus) : "memory");
233 		mb();
234 		if (type != PSURGE_QUAD_IN(PSURGE_QUAD_BOARD_ID))
235 			return PSURGE_DUAL;
236 	}
237 	return type;
238 }
239 
240 static void __init psurge_quad_init(void)
241 {
242 	int procbits;
243 
244 	if (ppc_md.progress) ppc_md.progress("psurge_quad_init", 0x351);
245 	procbits = ~PSURGE_QUAD_IN(PSURGE_QUAD_WHICH_CPU);
246 	if (psurge_type == PSURGE_QUAD_ICEGRASS)
247 		PSURGE_QUAD_BIS(PSURGE_QUAD_RESET_CTL, procbits);
248 	else
249 		PSURGE_QUAD_BIC(PSURGE_QUAD_CKSTOP_CTL, procbits);
250 	mdelay(33);
251 	out_8(psurge_sec_intr, ~0);
252 	PSURGE_QUAD_OUT(PSURGE_QUAD_IRQ_CLR, procbits);
253 	PSURGE_QUAD_BIS(PSURGE_QUAD_RESET_CTL, procbits);
254 	if (psurge_type != PSURGE_QUAD_ICEGRASS)
255 		PSURGE_QUAD_BIS(PSURGE_QUAD_CKSTOP_CTL, procbits);
256 	PSURGE_QUAD_BIC(PSURGE_QUAD_PRIMARY_ARB, procbits);
257 	mdelay(33);
258 	PSURGE_QUAD_BIC(PSURGE_QUAD_RESET_CTL, procbits);
259 	mdelay(33);
260 	PSURGE_QUAD_BIS(PSURGE_QUAD_PRIMARY_ARB, procbits);
261 	mdelay(33);
262 }
263 
264 static int __init smp_psurge_probe(void)
265 {
266 	int i, ncpus;
267 
268 	/* We don't do SMP on the PPC601 -- paulus */
269 	if (PVR_VER(mfspr(SPRN_PVR)) == 1)
270 		return 1;
271 
272 	/*
273 	 * The powersurge cpu board can be used in the generation
274 	 * of powermacs that have a socket for an upgradeable cpu card,
275 	 * including the 7500, 8500, 9500, 9600.
276 	 * The device tree doesn't tell you if you have 2 cpus because
277 	 * OF doesn't know anything about the 2nd processor.
278 	 * Instead we look for magic bits in magic registers,
279 	 * in the hammerhead memory controller in the case of the
280 	 * dual-cpu powersurge board.  -- paulus.
281 	 */
282 	if (find_devices("hammerhead") == NULL)
283 		return 1;
284 
285 	hhead_base = ioremap(HAMMERHEAD_BASE, 0x800);
286 	quad_base = ioremap(PSURGE_QUAD_REG_ADDR, 1024);
287 	psurge_sec_intr = hhead_base + HHEAD_SEC_INTR;
288 
289 	psurge_type = psurge_quad_probe();
290 	if (psurge_type != PSURGE_DUAL) {
291 		psurge_quad_init();
292 		/* All released cards using this HW design have 4 CPUs */
293 		ncpus = 4;
294 	} else {
295 		iounmap(quad_base);
296 		if ((in_8(hhead_base + HHEAD_CONFIG) & 0x02) == 0) {
297 			/* not a dual-cpu card */
298 			iounmap(hhead_base);
299 			psurge_type = PSURGE_NONE;
300 			return 1;
301 		}
302 		ncpus = 2;
303 	}
304 
305 	psurge_start = ioremap(PSURGE_START, 4);
306 	psurge_pri_intr = ioremap(PSURGE_PRI_INTR, 4);
307 
308 	/*
309 	 * This is necessary because OF doesn't know about the
310 	 * secondary cpu(s), and thus there aren't nodes in the
311 	 * device tree for them, and smp_setup_cpu_maps hasn't
312 	 * set their bits in cpu_possible_map and cpu_present_map.
313 	 */
314 	if (ncpus > NR_CPUS)
315 		ncpus = NR_CPUS;
316 	for (i = 1; i < ncpus ; ++i) {
317 		cpu_set(i, cpu_present_map);
318 		cpu_set(i, cpu_possible_map);
319 		set_hard_smp_processor_id(i, i);
320 	}
321 
322 	if (ppc_md.progress) ppc_md.progress("smp_psurge_probe - done", 0x352);
323 
324 	return ncpus;
325 }
326 
327 static void __init smp_psurge_kick_cpu(int nr)
328 {
329 	unsigned long start = __pa(__secondary_start_pmac_0) + nr * 8;
330 	unsigned long a;
331 	int i;
332 
333 	/* may need to flush here if secondary bats aren't setup */
334 	for (a = KERNELBASE; a < KERNELBASE + 0x800000; a += 32)
335 		asm volatile("dcbf 0,%0" : : "r" (a) : "memory");
336 	asm volatile("sync");
337 
338 	if (ppc_md.progress) ppc_md.progress("smp_psurge_kick_cpu", 0x353);
339 
340 	out_be32(psurge_start, start);
341 	mb();
342 
343 	psurge_set_ipi(nr);
344 	/*
345 	 * We can't use udelay here because the timebase is now frozen.
346 	 */
347 	for (i = 0; i < 2000; ++i)
348 		barrier();
349 	psurge_clr_ipi(nr);
350 
351 	if (ppc_md.progress) ppc_md.progress("smp_psurge_kick_cpu - done", 0x354);
352 }
353 
354 /*
355  * With the dual-cpu powersurge board, the decrementers and timebases
356  * of both cpus are frozen after the secondary cpu is started up,
357  * until we give the secondary cpu another interrupt.  This routine
358  * uses this to get the timebases synchronized.
359  *  -- paulus.
360  */
361 static void __init psurge_dual_sync_tb(int cpu_nr)
362 {
363 	int t;
364 
365 	set_dec(tb_ticks_per_jiffy);
366 	/* XXX fixme */
367 	set_tb(0, 0);
368 
369 	if (cpu_nr > 0) {
370 		mb();
371 		sec_tb_reset = 1;
372 		return;
373 	}
374 
375 	/* wait for the secondary to have reset its TB before proceeding */
376 	for (t = 10000000; t > 0 && !sec_tb_reset; --t)
377 		;
378 
379 	/* now interrupt the secondary, starting both TBs */
380 	psurge_set_ipi(1);
381 }
382 
383 static struct irqaction psurge_irqaction = {
384 	.handler = psurge_primary_intr,
385 	.flags = IRQF_DISABLED,
386 	.mask = CPU_MASK_NONE,
387 	.name = "primary IPI",
388 };
389 
390 static void __init smp_psurge_setup_cpu(int cpu_nr)
391 {
392 
393 	if (cpu_nr == 0) {
394 		/* If we failed to start the second CPU, we should still
395 		 * send it an IPI to start the timebase & DEC or we might
396 		 * have them stuck.
397 		 */
398 		if (num_online_cpus() < 2) {
399 			if (psurge_type == PSURGE_DUAL)
400 				psurge_set_ipi(1);
401 			return;
402 		}
403 		/* reset the entry point so if we get another intr we won't
404 		 * try to startup again */
405 		out_be32(psurge_start, 0x100);
406 		if (setup_irq(30, &psurge_irqaction))
407 			printk(KERN_ERR "Couldn't get primary IPI interrupt");
408 	}
409 
410 	if (psurge_type == PSURGE_DUAL)
411 		psurge_dual_sync_tb(cpu_nr);
412 }
413 
414 void __init smp_psurge_take_timebase(void)
415 {
416 	/* Dummy implementation */
417 }
418 
419 void __init smp_psurge_give_timebase(void)
420 {
421 	/* Dummy implementation */
422 }
423 
424 /* PowerSurge-style Macs */
425 struct smp_ops_t psurge_smp_ops = {
426 	.message_pass	= smp_psurge_message_pass,
427 	.probe		= smp_psurge_probe,
428 	.kick_cpu	= smp_psurge_kick_cpu,
429 	.setup_cpu	= smp_psurge_setup_cpu,
430 	.give_timebase	= smp_psurge_give_timebase,
431 	.take_timebase	= smp_psurge_take_timebase,
432 };
433 #endif /* CONFIG_PPC32 - actually powersurge support */
434 
435 /*
436  * Core 99 and later support
437  */
438 
439 static void (*pmac_tb_freeze)(int freeze);
440 static u64 timebase;
441 static int tb_req;
442 
443 static void smp_core99_give_timebase(void)
444 {
445 	unsigned long flags;
446 
447 	local_irq_save(flags);
448 
449 	while(!tb_req)
450 		barrier();
451 	tb_req = 0;
452 	(*pmac_tb_freeze)(1);
453 	mb();
454 	timebase = get_tb();
455 	mb();
456 	while (timebase)
457 		barrier();
458 	mb();
459 	(*pmac_tb_freeze)(0);
460 	mb();
461 
462 	local_irq_restore(flags);
463 }
464 
465 
466 static void __devinit smp_core99_take_timebase(void)
467 {
468 	unsigned long flags;
469 
470 	local_irq_save(flags);
471 
472 	tb_req = 1;
473 	mb();
474 	while (!timebase)
475 		barrier();
476 	mb();
477 	set_tb(timebase >> 32, timebase & 0xffffffff);
478 	timebase = 0;
479 	mb();
480 	set_dec(tb_ticks_per_jiffy/2);
481 
482 	local_irq_restore(flags);
483 }
484 
485 #ifdef CONFIG_PPC64
486 /*
487  * G5s enable/disable the timebase via an i2c-connected clock chip.
488  */
489 static struct pmac_i2c_bus *pmac_tb_clock_chip_host;
490 static u8 pmac_tb_pulsar_addr;
491 
492 static void smp_core99_cypress_tb_freeze(int freeze)
493 {
494 	u8 data;
495 	int rc;
496 
497 	/* Strangely, the device-tree says address is 0xd2, but darwin
498 	 * accesses 0xd0 ...
499 	 */
500 	pmac_i2c_setmode(pmac_tb_clock_chip_host,
501 			 pmac_i2c_mode_combined);
502 	rc = pmac_i2c_xfer(pmac_tb_clock_chip_host,
503 			   0xd0 | pmac_i2c_read,
504 			   1, 0x81, &data, 1);
505 	if (rc != 0)
506 		goto bail;
507 
508 	data = (data & 0xf3) | (freeze ? 0x00 : 0x0c);
509 
510        	pmac_i2c_setmode(pmac_tb_clock_chip_host, pmac_i2c_mode_stdsub);
511 	rc = pmac_i2c_xfer(pmac_tb_clock_chip_host,
512 			   0xd0 | pmac_i2c_write,
513 			   1, 0x81, &data, 1);
514 
515  bail:
516 	if (rc != 0) {
517 		printk("Cypress Timebase %s rc: %d\n",
518 		       freeze ? "freeze" : "unfreeze", rc);
519 		panic("Timebase freeze failed !\n");
520 	}
521 }
522 
523 
524 static void smp_core99_pulsar_tb_freeze(int freeze)
525 {
526 	u8 data;
527 	int rc;
528 
529 	pmac_i2c_setmode(pmac_tb_clock_chip_host,
530 			 pmac_i2c_mode_combined);
531 	rc = pmac_i2c_xfer(pmac_tb_clock_chip_host,
532 			   pmac_tb_pulsar_addr | pmac_i2c_read,
533 			   1, 0x2e, &data, 1);
534 	if (rc != 0)
535 		goto bail;
536 
537 	data = (data & 0x88) | (freeze ? 0x11 : 0x22);
538 
539 	pmac_i2c_setmode(pmac_tb_clock_chip_host, pmac_i2c_mode_stdsub);
540 	rc = pmac_i2c_xfer(pmac_tb_clock_chip_host,
541 			   pmac_tb_pulsar_addr | pmac_i2c_write,
542 			   1, 0x2e, &data, 1);
543  bail:
544 	if (rc != 0) {
545 		printk(KERN_ERR "Pulsar Timebase %s rc: %d\n",
546 		       freeze ? "freeze" : "unfreeze", rc);
547 		panic("Timebase freeze failed !\n");
548 	}
549 }
550 
551 static void __init smp_core99_setup_i2c_hwsync(int ncpus)
552 {
553 	struct device_node *cc = NULL;
554 	struct device_node *p;
555 	const char *name = NULL;
556 	const u32 *reg;
557 	int ok;
558 
559 	/* Look for the clock chip */
560 	while ((cc = of_find_node_by_name(cc, "i2c-hwclock")) != NULL) {
561 		p = of_get_parent(cc);
562 		ok = p && device_is_compatible(p, "uni-n-i2c");
563 		of_node_put(p);
564 		if (!ok)
565 			continue;
566 
567 		pmac_tb_clock_chip_host = pmac_i2c_find_bus(cc);
568 		if (pmac_tb_clock_chip_host == NULL)
569 			continue;
570 		reg = get_property(cc, "reg", NULL);
571 		if (reg == NULL)
572 			continue;
573 		switch (*reg) {
574 		case 0xd2:
575 			if (device_is_compatible(cc,"pulsar-legacy-slewing")) {
576 				pmac_tb_freeze = smp_core99_pulsar_tb_freeze;
577 				pmac_tb_pulsar_addr = 0xd2;
578 				name = "Pulsar";
579 			} else if (device_is_compatible(cc, "cy28508")) {
580 				pmac_tb_freeze = smp_core99_cypress_tb_freeze;
581 				name = "Cypress";
582 			}
583 			break;
584 		case 0xd4:
585 			pmac_tb_freeze = smp_core99_pulsar_tb_freeze;
586 			pmac_tb_pulsar_addr = 0xd4;
587 			name = "Pulsar";
588 			break;
589 		}
590 		if (pmac_tb_freeze != NULL)
591 			break;
592 	}
593 	if (pmac_tb_freeze != NULL) {
594 		/* Open i2c bus for synchronous access */
595 		if (pmac_i2c_open(pmac_tb_clock_chip_host, 1)) {
596 			printk(KERN_ERR "Failed top open i2c bus for clock"
597 			       " sync, fallback to software sync !\n");
598 			goto no_i2c_sync;
599 		}
600 		printk(KERN_INFO "Processor timebase sync using %s i2c clock\n",
601 		       name);
602 		return;
603 	}
604  no_i2c_sync:
605 	pmac_tb_freeze = NULL;
606 	pmac_tb_clock_chip_host = NULL;
607 }
608 
609 
610 
611 /*
612  * Newer G5s uses a platform function
613  */
614 
615 static void smp_core99_pfunc_tb_freeze(int freeze)
616 {
617 	struct device_node *cpus;
618 	struct pmf_args args;
619 
620 	cpus = of_find_node_by_path("/cpus");
621 	BUG_ON(cpus == NULL);
622 	args.count = 1;
623 	args.u[0].v = !freeze;
624 	pmf_call_function(cpus, "cpu-timebase", &args);
625 	of_node_put(cpus);
626 }
627 
628 #else /* CONFIG_PPC64 */
629 
630 /*
631  * SMP G4 use a GPIO to enable/disable the timebase.
632  */
633 
634 static unsigned int core99_tb_gpio;	/* Timebase freeze GPIO */
635 
636 static void smp_core99_gpio_tb_freeze(int freeze)
637 {
638 	if (freeze)
639 		pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, core99_tb_gpio, 4);
640 	else
641 		pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, core99_tb_gpio, 0);
642 	pmac_call_feature(PMAC_FTR_READ_GPIO, NULL, core99_tb_gpio, 0);
643 }
644 
645 
646 #endif /* !CONFIG_PPC64 */
647 
648 /* L2 and L3 cache settings to pass from CPU0 to CPU1 on G4 cpus */
649 volatile static long int core99_l2_cache;
650 volatile static long int core99_l3_cache;
651 
652 static void __devinit core99_init_caches(int cpu)
653 {
654 #ifndef CONFIG_PPC64
655 	if (!cpu_has_feature(CPU_FTR_L2CR))
656 		return;
657 
658 	if (cpu == 0) {
659 		core99_l2_cache = _get_L2CR();
660 		printk("CPU0: L2CR is %lx\n", core99_l2_cache);
661 	} else {
662 		printk("CPU%d: L2CR was %lx\n", cpu, _get_L2CR());
663 		_set_L2CR(0);
664 		_set_L2CR(core99_l2_cache);
665 		printk("CPU%d: L2CR set to %lx\n", cpu, core99_l2_cache);
666 	}
667 
668 	if (!cpu_has_feature(CPU_FTR_L3CR))
669 		return;
670 
671 	if (cpu == 0){
672 		core99_l3_cache = _get_L3CR();
673 		printk("CPU0: L3CR is %lx\n", core99_l3_cache);
674 	} else {
675 		printk("CPU%d: L3CR was %lx\n", cpu, _get_L3CR());
676 		_set_L3CR(0);
677 		_set_L3CR(core99_l3_cache);
678 		printk("CPU%d: L3CR set to %lx\n", cpu, core99_l3_cache);
679 	}
680 #endif /* !CONFIG_PPC64 */
681 }
682 
683 static void __init smp_core99_setup(int ncpus)
684 {
685 #ifdef CONFIG_PPC64
686 
687 	/* i2c based HW sync on some G5s */
688 	if (machine_is_compatible("PowerMac7,2") ||
689 	    machine_is_compatible("PowerMac7,3") ||
690 	    machine_is_compatible("RackMac3,1"))
691 		smp_core99_setup_i2c_hwsync(ncpus);
692 
693 	/* pfunc based HW sync on recent G5s */
694 	if (pmac_tb_freeze == NULL) {
695 		struct device_node *cpus =
696 			of_find_node_by_path("/cpus");
697 		if (cpus &&
698 		    get_property(cpus, "platform-cpu-timebase", NULL)) {
699 			pmac_tb_freeze = smp_core99_pfunc_tb_freeze;
700 			printk(KERN_INFO "Processor timebase sync using"
701 			       " platform function\n");
702 		}
703 	}
704 
705 #else /* CONFIG_PPC64 */
706 
707 	/* GPIO based HW sync on ppc32 Core99 */
708 	if (pmac_tb_freeze == NULL && !machine_is_compatible("MacRISC4")) {
709 		struct device_node *cpu;
710 		const u32 *tbprop = NULL;
711 
712 		core99_tb_gpio = KL_GPIO_TB_ENABLE;	/* default value */
713 		cpu = of_find_node_by_type(NULL, "cpu");
714 		if (cpu != NULL) {
715 			tbprop = get_property(cpu, "timebase-enable", NULL);
716 			if (tbprop)
717 				core99_tb_gpio = *tbprop;
718 			of_node_put(cpu);
719 		}
720 		pmac_tb_freeze = smp_core99_gpio_tb_freeze;
721 		printk(KERN_INFO "Processor timebase sync using"
722 		       " GPIO 0x%02x\n", core99_tb_gpio);
723 	}
724 
725 #endif /* CONFIG_PPC64 */
726 
727 	/* No timebase sync, fallback to software */
728 	if (pmac_tb_freeze == NULL) {
729 		smp_ops->give_timebase = smp_generic_give_timebase;
730 		smp_ops->take_timebase = smp_generic_take_timebase;
731 		printk(KERN_INFO "Processor timebase sync using software\n");
732 	}
733 
734 #ifndef CONFIG_PPC64
735 	{
736 		int i;
737 
738 		/* XXX should get this from reg properties */
739 		for (i = 1; i < ncpus; ++i)
740 			smp_hw_index[i] = i;
741 	}
742 #endif
743 
744 	/* 32 bits SMP can't NAP */
745 	if (!machine_is_compatible("MacRISC4"))
746 		powersave_nap = 0;
747 }
748 
749 static int __init smp_core99_probe(void)
750 {
751 	struct device_node *cpus;
752 	int ncpus = 0;
753 
754 	if (ppc_md.progress) ppc_md.progress("smp_core99_probe", 0x345);
755 
756 	/* Count CPUs in the device-tree */
757        	for (cpus = NULL; (cpus = of_find_node_by_type(cpus, "cpu")) != NULL;)
758 	       	++ncpus;
759 
760 	printk(KERN_INFO "PowerMac SMP probe found %d cpus\n", ncpus);
761 
762 	/* Nothing more to do if less than 2 of them */
763 	if (ncpus <= 1)
764 		return 1;
765 
766 	/* We need to perform some early initialisations before we can start
767 	 * setting up SMP as we are running before initcalls
768 	 */
769 	pmac_pfunc_base_install();
770 	pmac_i2c_init();
771 
772 	/* Setup various bits like timebase sync method, ability to nap, ... */
773 	smp_core99_setup(ncpus);
774 
775 	/* Install IPIs */
776 	mpic_request_ipis();
777 
778 	/* Collect l2cr and l3cr values from CPU 0 */
779 	core99_init_caches(0);
780 
781 	return ncpus;
782 }
783 
784 static void __devinit smp_core99_kick_cpu(int nr)
785 {
786 	unsigned int save_vector;
787 	unsigned long target, flags;
788 	volatile unsigned int *vector
789 		 = ((volatile unsigned int *)(KERNELBASE+0x100));
790 
791 	if (nr < 0 || nr > 3)
792 		return;
793 
794 	if (ppc_md.progress)
795 		ppc_md.progress("smp_core99_kick_cpu", 0x346);
796 
797 	local_irq_save(flags);
798 
799 	/* Save reset vector */
800 	save_vector = *vector;
801 
802 	/* Setup fake reset vector that does
803 	 *   b __secondary_start_pmac_0 + nr*8 - KERNELBASE
804 	 */
805 	target = (unsigned long) __secondary_start_pmac_0 + nr * 8;
806 	create_branch((unsigned long)vector, target, BRANCH_SET_LINK);
807 
808 	/* Put some life in our friend */
809 	pmac_call_feature(PMAC_FTR_RESET_CPU, NULL, nr, 0);
810 
811 	/* FIXME: We wait a bit for the CPU to take the exception, I should
812 	 * instead wait for the entry code to set something for me. Well,
813 	 * ideally, all that crap will be done in prom.c and the CPU left
814 	 * in a RAM-based wait loop like CHRP.
815 	 */
816 	mdelay(1);
817 
818 	/* Restore our exception vector */
819 	*vector = save_vector;
820 	flush_icache_range((unsigned long) vector, (unsigned long) vector + 4);
821 
822 	local_irq_restore(flags);
823 	if (ppc_md.progress) ppc_md.progress("smp_core99_kick_cpu done", 0x347);
824 }
825 
826 static void __devinit smp_core99_setup_cpu(int cpu_nr)
827 {
828 	/* Setup L2/L3 */
829 	if (cpu_nr != 0)
830 		core99_init_caches(cpu_nr);
831 
832 	/* Setup openpic */
833 	mpic_setup_this_cpu();
834 
835 	if (cpu_nr == 0) {
836 #ifdef CONFIG_PPC64
837 		extern void g5_phy_disable_cpu1(void);
838 
839 		/* Close i2c bus if it was used for tb sync */
840 		if (pmac_tb_clock_chip_host) {
841 			pmac_i2c_close(pmac_tb_clock_chip_host);
842 			pmac_tb_clock_chip_host	= NULL;
843 		}
844 
845 		/* If we didn't start the second CPU, we must take
846 		 * it off the bus
847 		 */
848 		if (machine_is_compatible("MacRISC4") &&
849 		    num_online_cpus() < 2)
850 			g5_phy_disable_cpu1();
851 #endif /* CONFIG_PPC64 */
852 
853 		if (ppc_md.progress)
854 			ppc_md.progress("core99_setup_cpu 0 done", 0x349);
855 	}
856 }
857 
858 
859 #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PPC32)
860 
861 int smp_core99_cpu_disable(void)
862 {
863 	cpu_clear(smp_processor_id(), cpu_online_map);
864 
865 	/* XXX reset cpu affinity here */
866 	mpic_cpu_set_priority(0xf);
867 	asm volatile("mtdec %0" : : "r" (0x7fffffff));
868 	mb();
869 	udelay(20);
870 	asm volatile("mtdec %0" : : "r" (0x7fffffff));
871 	return 0;
872 }
873 
874 extern void low_cpu_die(void) __attribute__((noreturn)); /* in sleep.S */
875 static int cpu_dead[NR_CPUS];
876 
877 void cpu_die(void)
878 {
879 	local_irq_disable();
880 	cpu_dead[smp_processor_id()] = 1;
881 	mb();
882 	low_cpu_die();
883 }
884 
885 void smp_core99_cpu_die(unsigned int cpu)
886 {
887 	int timeout;
888 
889 	timeout = 1000;
890 	while (!cpu_dead[cpu]) {
891 		if (--timeout == 0) {
892 			printk("CPU %u refused to die!\n", cpu);
893 			break;
894 		}
895 		msleep(1);
896 	}
897 	cpu_dead[cpu] = 0;
898 }
899 
900 #endif
901 
902 /* Core99 Macs (dual G4s and G5s) */
903 struct smp_ops_t core99_smp_ops = {
904 	.message_pass	= smp_mpic_message_pass,
905 	.probe		= smp_core99_probe,
906 	.kick_cpu	= smp_core99_kick_cpu,
907 	.setup_cpu	= smp_core99_setup_cpu,
908 	.give_timebase	= smp_core99_give_timebase,
909 	.take_timebase	= smp_core99_take_timebase,
910 #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PPC32)
911 	.cpu_disable	= smp_core99_cpu_disable,
912 	.cpu_die	= smp_core99_cpu_die,
913 #endif
914 };
915